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 btrfs_fs_info *fs_info,
25 struct extent_buffer *dst,
26 struct extent_buffer *src, int empty);
27 static int balance_node_right(struct btrfs_trans_handle *trans,
28 struct btrfs_fs_info *fs_info,
29 struct extent_buffer *dst_buf,
30 struct extent_buffer *src_buf);
31 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
34 struct btrfs_path *btrfs_alloc_path(void)
36 return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
40 * set all locked nodes in the path to blocking locks. This should
41 * be done before scheduling
43 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
46 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
47 if (!p->nodes[i] || !p->locks[i])
50 * If we currently have a spinning reader or writer lock this
51 * will bump the count of blocking holders and drop the
54 if (p->locks[i] == BTRFS_READ_LOCK) {
55 btrfs_set_lock_blocking_read(p->nodes[i]);
56 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
57 } else if (p->locks[i] == BTRFS_WRITE_LOCK) {
58 btrfs_set_lock_blocking_write(p->nodes[i]);
59 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
64 /* this also releases the path */
65 void btrfs_free_path(struct btrfs_path *p)
69 btrfs_release_path(p);
70 kmem_cache_free(btrfs_path_cachep, p);
74 * path release drops references on the extent buffers in the path
75 * and it drops any locks held by this path
77 * It is safe to call this on paths that no locks or extent buffers held.
79 noinline void btrfs_release_path(struct btrfs_path *p)
83 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
88 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
91 free_extent_buffer(p->nodes[i]);
97 * safely gets a reference on the root node of a tree. A lock
98 * is not taken, so a concurrent writer may put a different node
99 * at the root of the tree. See btrfs_lock_root_node for the
102 * The extent buffer returned by this has a reference taken, so
103 * it won't disappear. It may stop being the root of the tree
104 * at any time because there are no locks held.
106 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
108 struct extent_buffer *eb;
112 eb = rcu_dereference(root->node);
115 * RCU really hurts here, we could free up the root node because
116 * it was COWed but we may not get the new root node yet so do
117 * the inc_not_zero dance and if it doesn't work then
118 * synchronize_rcu and try again.
120 if (atomic_inc_not_zero(&eb->refs)) {
130 /* loop around taking references on and locking the root node of the
131 * tree until you end up with a lock on the root. A locked buffer
132 * is returned, with a reference held.
134 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
136 struct extent_buffer *eb;
139 eb = btrfs_root_node(root);
141 if (eb == root->node)
143 btrfs_tree_unlock(eb);
144 free_extent_buffer(eb);
149 /* loop around taking references on and locking the root node of the
150 * tree until you end up with a lock on the root. A locked buffer
151 * is returned, with a reference held.
153 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
155 struct extent_buffer *eb;
158 eb = btrfs_root_node(root);
159 btrfs_tree_read_lock(eb);
160 if (eb == root->node)
162 btrfs_tree_read_unlock(eb);
163 free_extent_buffer(eb);
168 /* cowonly root (everything not a reference counted cow subvolume), just get
169 * put onto a simple dirty list. transaction.c walks this to make sure they
170 * get properly updated on disk.
172 static void add_root_to_dirty_list(struct btrfs_root *root)
174 struct btrfs_fs_info *fs_info = root->fs_info;
176 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
177 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
180 spin_lock(&fs_info->trans_lock);
181 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
182 /* Want the extent tree to be the last on the list */
183 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
184 list_move_tail(&root->dirty_list,
185 &fs_info->dirty_cowonly_roots);
187 list_move(&root->dirty_list,
188 &fs_info->dirty_cowonly_roots);
190 spin_unlock(&fs_info->trans_lock);
194 * used by snapshot creation to make a copy of a root for a tree with
195 * a given objectid. The buffer with the new root node is returned in
196 * cow_ret, and this func returns zero on success or a negative error code.
198 int btrfs_copy_root(struct btrfs_trans_handle *trans,
199 struct btrfs_root *root,
200 struct extent_buffer *buf,
201 struct extent_buffer **cow_ret, u64 new_root_objectid)
203 struct btrfs_fs_info *fs_info = root->fs_info;
204 struct extent_buffer *cow;
207 struct btrfs_disk_key disk_key;
209 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
210 trans->transid != fs_info->running_transaction->transid);
211 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
212 trans->transid != root->last_trans);
214 level = btrfs_header_level(buf);
216 btrfs_item_key(buf, &disk_key, 0);
218 btrfs_node_key(buf, &disk_key, 0);
220 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
221 &disk_key, level, buf->start, 0);
225 copy_extent_buffer_full(cow, buf);
226 btrfs_set_header_bytenr(cow, cow->start);
227 btrfs_set_header_generation(cow, trans->transid);
228 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
229 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
230 BTRFS_HEADER_FLAG_RELOC);
231 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
232 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
234 btrfs_set_header_owner(cow, new_root_objectid);
236 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
238 WARN_ON(btrfs_header_generation(buf) > trans->transid);
239 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
240 ret = btrfs_inc_ref(trans, root, cow, 1);
242 ret = btrfs_inc_ref(trans, root, cow, 0);
247 btrfs_mark_buffer_dirty(cow);
256 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
257 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
259 MOD_LOG_ROOT_REPLACE,
262 struct tree_mod_root {
267 struct tree_mod_elem {
273 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
276 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
279 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
280 struct btrfs_disk_key key;
283 /* this is used for op == MOD_LOG_MOVE_KEYS */
289 /* this is used for op == MOD_LOG_ROOT_REPLACE */
290 struct tree_mod_root old_root;
294 * Pull a new tree mod seq number for our operation.
296 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
298 return atomic64_inc_return(&fs_info->tree_mod_seq);
302 * This adds a new blocker to the tree mod log's blocker list if the @elem
303 * passed does not already have a sequence number set. So when a caller expects
304 * to record tree modifications, it should ensure to set elem->seq to zero
305 * before calling btrfs_get_tree_mod_seq.
306 * Returns a fresh, unused tree log modification sequence number, even if no new
309 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
310 struct seq_list *elem)
312 write_lock(&fs_info->tree_mod_log_lock);
313 spin_lock(&fs_info->tree_mod_seq_lock);
315 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
316 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
318 spin_unlock(&fs_info->tree_mod_seq_lock);
319 write_unlock(&fs_info->tree_mod_log_lock);
324 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
325 struct seq_list *elem)
327 struct rb_root *tm_root;
328 struct rb_node *node;
329 struct rb_node *next;
330 struct seq_list *cur_elem;
331 struct tree_mod_elem *tm;
332 u64 min_seq = (u64)-1;
333 u64 seq_putting = elem->seq;
338 spin_lock(&fs_info->tree_mod_seq_lock);
339 list_del(&elem->list);
342 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
343 if (cur_elem->seq < min_seq) {
344 if (seq_putting > cur_elem->seq) {
346 * blocker with lower sequence number exists, we
347 * cannot remove anything from the log
349 spin_unlock(&fs_info->tree_mod_seq_lock);
352 min_seq = cur_elem->seq;
355 spin_unlock(&fs_info->tree_mod_seq_lock);
358 * anything that's lower than the lowest existing (read: blocked)
359 * sequence number can be removed from the tree.
361 write_lock(&fs_info->tree_mod_log_lock);
362 tm_root = &fs_info->tree_mod_log;
363 for (node = rb_first(tm_root); node; node = next) {
364 next = rb_next(node);
365 tm = rb_entry(node, struct tree_mod_elem, node);
366 if (tm->seq > min_seq)
368 rb_erase(node, tm_root);
371 write_unlock(&fs_info->tree_mod_log_lock);
375 * key order of the log:
376 * node/leaf start address -> sequence
378 * The 'start address' is the logical address of the *new* root node
379 * for root replace operations, or the logical address of the affected
380 * block for all other operations.
382 * Note: must be called with write lock for fs_info::tree_mod_log_lock.
385 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
387 struct rb_root *tm_root;
388 struct rb_node **new;
389 struct rb_node *parent = NULL;
390 struct tree_mod_elem *cur;
392 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
394 tm_root = &fs_info->tree_mod_log;
395 new = &tm_root->rb_node;
397 cur = rb_entry(*new, struct tree_mod_elem, node);
399 if (cur->logical < tm->logical)
400 new = &((*new)->rb_left);
401 else if (cur->logical > tm->logical)
402 new = &((*new)->rb_right);
403 else if (cur->seq < tm->seq)
404 new = &((*new)->rb_left);
405 else if (cur->seq > tm->seq)
406 new = &((*new)->rb_right);
411 rb_link_node(&tm->node, parent, new);
412 rb_insert_color(&tm->node, tm_root);
417 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
418 * returns zero with the tree_mod_log_lock acquired. The caller must hold
419 * this until all tree mod log insertions are recorded in the rb tree and then
420 * write unlock fs_info::tree_mod_log_lock.
422 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
423 struct extent_buffer *eb) {
425 if (list_empty(&(fs_info)->tree_mod_seq_list))
427 if (eb && btrfs_header_level(eb) == 0)
430 write_lock(&fs_info->tree_mod_log_lock);
431 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
432 write_unlock(&fs_info->tree_mod_log_lock);
439 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
440 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
441 struct extent_buffer *eb)
444 if (list_empty(&(fs_info)->tree_mod_seq_list))
446 if (eb && btrfs_header_level(eb) == 0)
452 static struct tree_mod_elem *
453 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
454 enum mod_log_op op, gfp_t flags)
456 struct tree_mod_elem *tm;
458 tm = kzalloc(sizeof(*tm), flags);
462 tm->logical = eb->start;
463 if (op != MOD_LOG_KEY_ADD) {
464 btrfs_node_key(eb, &tm->key, slot);
465 tm->blockptr = btrfs_node_blockptr(eb, slot);
469 tm->generation = btrfs_node_ptr_generation(eb, slot);
470 RB_CLEAR_NODE(&tm->node);
475 static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot,
476 enum mod_log_op op, gfp_t flags)
478 struct tree_mod_elem *tm;
481 if (!tree_mod_need_log(eb->fs_info, eb))
484 tm = alloc_tree_mod_elem(eb, slot, op, flags);
488 if (tree_mod_dont_log(eb->fs_info, eb)) {
493 ret = __tree_mod_log_insert(eb->fs_info, tm);
494 write_unlock(&eb->fs_info->tree_mod_log_lock);
501 static noinline int tree_mod_log_insert_move(struct extent_buffer *eb,
502 int dst_slot, int src_slot, int nr_items)
504 struct tree_mod_elem *tm = NULL;
505 struct tree_mod_elem **tm_list = NULL;
510 if (!tree_mod_need_log(eb->fs_info, eb))
513 tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
517 tm = kzalloc(sizeof(*tm), GFP_NOFS);
523 tm->logical = eb->start;
525 tm->move.dst_slot = dst_slot;
526 tm->move.nr_items = nr_items;
527 tm->op = MOD_LOG_MOVE_KEYS;
529 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
530 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
531 MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
538 if (tree_mod_dont_log(eb->fs_info, eb))
543 * When we override something during the move, we log these removals.
544 * This can only happen when we move towards the beginning of the
545 * buffer, i.e. dst_slot < src_slot.
547 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
548 ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]);
553 ret = __tree_mod_log_insert(eb->fs_info, tm);
556 write_unlock(&eb->fs_info->tree_mod_log_lock);
561 for (i = 0; i < nr_items; i++) {
562 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
563 rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log);
567 write_unlock(&eb->fs_info->tree_mod_log_lock);
575 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
576 struct tree_mod_elem **tm_list,
582 for (i = nritems - 1; i >= 0; i--) {
583 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
585 for (j = nritems - 1; j > i; j--)
586 rb_erase(&tm_list[j]->node,
587 &fs_info->tree_mod_log);
595 static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root,
596 struct extent_buffer *new_root, int log_removal)
598 struct btrfs_fs_info *fs_info = old_root->fs_info;
599 struct tree_mod_elem *tm = NULL;
600 struct tree_mod_elem **tm_list = NULL;
605 if (!tree_mod_need_log(fs_info, NULL))
608 if (log_removal && btrfs_header_level(old_root) > 0) {
609 nritems = btrfs_header_nritems(old_root);
610 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
616 for (i = 0; i < nritems; i++) {
617 tm_list[i] = alloc_tree_mod_elem(old_root, i,
618 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
626 tm = kzalloc(sizeof(*tm), GFP_NOFS);
632 tm->logical = new_root->start;
633 tm->old_root.logical = old_root->start;
634 tm->old_root.level = btrfs_header_level(old_root);
635 tm->generation = btrfs_header_generation(old_root);
636 tm->op = MOD_LOG_ROOT_REPLACE;
638 if (tree_mod_dont_log(fs_info, NULL))
642 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
644 ret = __tree_mod_log_insert(fs_info, tm);
646 write_unlock(&fs_info->tree_mod_log_lock);
655 for (i = 0; i < nritems; i++)
664 static struct tree_mod_elem *
665 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
668 struct rb_root *tm_root;
669 struct rb_node *node;
670 struct tree_mod_elem *cur = NULL;
671 struct tree_mod_elem *found = NULL;
673 read_lock(&fs_info->tree_mod_log_lock);
674 tm_root = &fs_info->tree_mod_log;
675 node = tm_root->rb_node;
677 cur = rb_entry(node, struct tree_mod_elem, node);
678 if (cur->logical < start) {
679 node = node->rb_left;
680 } else if (cur->logical > start) {
681 node = node->rb_right;
682 } else if (cur->seq < min_seq) {
683 node = node->rb_left;
684 } else if (!smallest) {
685 /* we want the node with the highest seq */
687 BUG_ON(found->seq > cur->seq);
689 node = node->rb_left;
690 } else if (cur->seq > min_seq) {
691 /* we want the node with the smallest seq */
693 BUG_ON(found->seq < cur->seq);
695 node = node->rb_right;
701 read_unlock(&fs_info->tree_mod_log_lock);
707 * this returns the element from the log with the smallest time sequence
708 * value that's in the log (the oldest log item). any element with a time
709 * sequence lower than min_seq will be ignored.
711 static struct tree_mod_elem *
712 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
715 return __tree_mod_log_search(fs_info, start, min_seq, 1);
719 * this returns the element from the log with the largest time sequence
720 * value that's in the log (the most recent log item). any element with
721 * a time sequence lower than min_seq will be ignored.
723 static struct tree_mod_elem *
724 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
726 return __tree_mod_log_search(fs_info, start, min_seq, 0);
730 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
731 struct extent_buffer *src, unsigned long dst_offset,
732 unsigned long src_offset, int nr_items)
735 struct tree_mod_elem **tm_list = NULL;
736 struct tree_mod_elem **tm_list_add, **tm_list_rem;
740 if (!tree_mod_need_log(fs_info, NULL))
743 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
746 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
751 tm_list_add = tm_list;
752 tm_list_rem = tm_list + nr_items;
753 for (i = 0; i < nr_items; i++) {
754 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
755 MOD_LOG_KEY_REMOVE, GFP_NOFS);
756 if (!tm_list_rem[i]) {
761 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
762 MOD_LOG_KEY_ADD, GFP_NOFS);
763 if (!tm_list_add[i]) {
769 if (tree_mod_dont_log(fs_info, NULL))
773 for (i = 0; i < nr_items; i++) {
774 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
777 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
782 write_unlock(&fs_info->tree_mod_log_lock);
788 for (i = 0; i < nr_items * 2; i++) {
789 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
790 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
794 write_unlock(&fs_info->tree_mod_log_lock);
800 static noinline int tree_mod_log_free_eb(struct extent_buffer *eb)
802 struct tree_mod_elem **tm_list = NULL;
807 if (btrfs_header_level(eb) == 0)
810 if (!tree_mod_need_log(eb->fs_info, NULL))
813 nritems = btrfs_header_nritems(eb);
814 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
818 for (i = 0; i < nritems; i++) {
819 tm_list[i] = alloc_tree_mod_elem(eb, i,
820 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
827 if (tree_mod_dont_log(eb->fs_info, eb))
830 ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems);
831 write_unlock(&eb->fs_info->tree_mod_log_lock);
839 for (i = 0; i < nritems; i++)
847 * check if the tree block can be shared by multiple trees
849 int btrfs_block_can_be_shared(struct btrfs_root *root,
850 struct extent_buffer *buf)
853 * Tree blocks not in reference counted trees and tree roots
854 * are never shared. If a block was allocated after the last
855 * snapshot and the block was not allocated by tree relocation,
856 * we know the block is not shared.
858 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
859 buf != root->node && buf != root->commit_root &&
860 (btrfs_header_generation(buf) <=
861 btrfs_root_last_snapshot(&root->root_item) ||
862 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
868 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
869 struct btrfs_root *root,
870 struct extent_buffer *buf,
871 struct extent_buffer *cow,
874 struct btrfs_fs_info *fs_info = root->fs_info;
882 * Backrefs update rules:
884 * Always use full backrefs for extent pointers in tree block
885 * allocated by tree relocation.
887 * If a shared tree block is no longer referenced by its owner
888 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
889 * use full backrefs for extent pointers in tree block.
891 * If a tree block is been relocating
892 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
893 * use full backrefs for extent pointers in tree block.
894 * The reason for this is some operations (such as drop tree)
895 * are only allowed for blocks use full backrefs.
898 if (btrfs_block_can_be_shared(root, buf)) {
899 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
900 btrfs_header_level(buf), 1,
906 btrfs_handle_fs_error(fs_info, ret, NULL);
911 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
912 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
913 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
918 owner = btrfs_header_owner(buf);
919 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
920 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
923 if ((owner == root->root_key.objectid ||
924 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
925 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
926 ret = btrfs_inc_ref(trans, root, buf, 1);
930 if (root->root_key.objectid ==
931 BTRFS_TREE_RELOC_OBJECTID) {
932 ret = btrfs_dec_ref(trans, root, buf, 0);
935 ret = btrfs_inc_ref(trans, root, cow, 1);
939 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
942 if (root->root_key.objectid ==
943 BTRFS_TREE_RELOC_OBJECTID)
944 ret = btrfs_inc_ref(trans, root, cow, 1);
946 ret = btrfs_inc_ref(trans, root, cow, 0);
950 if (new_flags != 0) {
951 int level = btrfs_header_level(buf);
953 ret = btrfs_set_disk_extent_flags(trans, fs_info,
956 new_flags, level, 0);
961 if (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);
969 ret = btrfs_dec_ref(trans, root, buf, 1);
973 clean_tree_block(fs_info, buf);
979 static struct extent_buffer *alloc_tree_block_no_bg_flush(
980 struct btrfs_trans_handle *trans,
981 struct btrfs_root *root,
983 const struct btrfs_disk_key *disk_key,
988 struct btrfs_fs_info *fs_info = root->fs_info;
989 struct extent_buffer *ret;
992 * If we are COWing a node/leaf from the extent, chunk, device or free
993 * space trees, make sure that we do not finish block group creation of
994 * pending block groups. We do this to avoid a deadlock.
995 * COWing can result in allocation of a new chunk, and flushing pending
996 * block groups (btrfs_create_pending_block_groups()) can be triggered
997 * when finishing allocation of a new chunk. Creation of a pending block
998 * group modifies the extent, chunk, device and free space trees,
999 * therefore we could deadlock with ourselves since we are holding a
1000 * lock on an extent buffer that btrfs_create_pending_block_groups() may
1002 * For similar reasons, we also need to delay flushing pending block
1003 * groups when splitting a leaf or node, from one of those trees, since
1004 * we are holding a write lock on it and its parent or when inserting a
1005 * new root node for one of those trees.
1007 if (root == fs_info->extent_root ||
1008 root == fs_info->chunk_root ||
1009 root == fs_info->dev_root ||
1010 root == fs_info->free_space_root)
1011 trans->can_flush_pending_bgs = false;
1013 ret = btrfs_alloc_tree_block(trans, root, parent_start,
1014 root->root_key.objectid, disk_key, level,
1016 trans->can_flush_pending_bgs = true;
1022 * does the dirty work in cow of a single block. The parent block (if
1023 * supplied) is updated to point to the new cow copy. The new buffer is marked
1024 * dirty and returned locked. If you modify the block it needs to be marked
1027 * search_start -- an allocation hint for the new block
1029 * empty_size -- a hint that you plan on doing more cow. This is the size in
1030 * bytes the allocator should try to find free next to the block it returns.
1031 * This is just a hint and may be ignored by the allocator.
1033 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1034 struct btrfs_root *root,
1035 struct extent_buffer *buf,
1036 struct extent_buffer *parent, int parent_slot,
1037 struct extent_buffer **cow_ret,
1038 u64 search_start, u64 empty_size)
1040 struct btrfs_fs_info *fs_info = root->fs_info;
1041 struct btrfs_disk_key disk_key;
1042 struct extent_buffer *cow;
1045 int unlock_orig = 0;
1046 u64 parent_start = 0;
1048 if (*cow_ret == buf)
1051 btrfs_assert_tree_locked(buf);
1053 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1054 trans->transid != fs_info->running_transaction->transid);
1055 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1056 trans->transid != root->last_trans);
1058 level = btrfs_header_level(buf);
1061 btrfs_item_key(buf, &disk_key, 0);
1063 btrfs_node_key(buf, &disk_key, 0);
1065 if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
1066 parent_start = parent->start;
1068 cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key,
1069 level, search_start, empty_size);
1071 return PTR_ERR(cow);
1073 /* cow is set to blocking by btrfs_init_new_buffer */
1075 copy_extent_buffer_full(cow, buf);
1076 btrfs_set_header_bytenr(cow, cow->start);
1077 btrfs_set_header_generation(cow, trans->transid);
1078 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1079 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1080 BTRFS_HEADER_FLAG_RELOC);
1081 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1082 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1084 btrfs_set_header_owner(cow, root->root_key.objectid);
1086 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
1088 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1090 btrfs_abort_transaction(trans, ret);
1094 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1095 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1097 btrfs_abort_transaction(trans, ret);
1102 if (buf == root->node) {
1103 WARN_ON(parent && parent != buf);
1104 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1105 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1106 parent_start = buf->start;
1108 extent_buffer_get(cow);
1109 ret = tree_mod_log_insert_root(root->node, cow, 1);
1111 rcu_assign_pointer(root->node, cow);
1113 btrfs_free_tree_block(trans, root, buf, parent_start,
1115 free_extent_buffer(buf);
1116 add_root_to_dirty_list(root);
1118 WARN_ON(trans->transid != btrfs_header_generation(parent));
1119 tree_mod_log_insert_key(parent, parent_slot,
1120 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1121 btrfs_set_node_blockptr(parent, parent_slot,
1123 btrfs_set_node_ptr_generation(parent, parent_slot,
1125 btrfs_mark_buffer_dirty(parent);
1127 ret = tree_mod_log_free_eb(buf);
1129 btrfs_abort_transaction(trans, ret);
1133 btrfs_free_tree_block(trans, root, buf, parent_start,
1137 btrfs_tree_unlock(buf);
1138 free_extent_buffer_stale(buf);
1139 btrfs_mark_buffer_dirty(cow);
1145 * returns the logical address of the oldest predecessor of the given root.
1146 * entries older than time_seq are ignored.
1148 static struct tree_mod_elem *__tree_mod_log_oldest_root(
1149 struct extent_buffer *eb_root, u64 time_seq)
1151 struct tree_mod_elem *tm;
1152 struct tree_mod_elem *found = NULL;
1153 u64 root_logical = eb_root->start;
1160 * the very last operation that's logged for a root is the
1161 * replacement operation (if it is replaced at all). this has
1162 * the logical address of the *new* root, making it the very
1163 * first operation that's logged for this root.
1166 tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
1171 * if there are no tree operation for the oldest root, we simply
1172 * return it. this should only happen if that (old) root is at
1179 * if there's an operation that's not a root replacement, we
1180 * found the oldest version of our root. normally, we'll find a
1181 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1183 if (tm->op != MOD_LOG_ROOT_REPLACE)
1187 root_logical = tm->old_root.logical;
1191 /* if there's no old root to return, return what we found instead */
1199 * tm is a pointer to the first operation to rewind within eb. then, all
1200 * previous operations will be rewound (until we reach something older than
1204 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1205 u64 time_seq, struct tree_mod_elem *first_tm)
1208 struct rb_node *next;
1209 struct tree_mod_elem *tm = first_tm;
1210 unsigned long o_dst;
1211 unsigned long o_src;
1212 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1214 n = btrfs_header_nritems(eb);
1215 read_lock(&fs_info->tree_mod_log_lock);
1216 while (tm && tm->seq >= time_seq) {
1218 * all the operations are recorded with the operator used for
1219 * the modification. as we're going backwards, we do the
1220 * opposite of each operation here.
1223 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1224 BUG_ON(tm->slot < n);
1226 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1227 case MOD_LOG_KEY_REMOVE:
1228 btrfs_set_node_key(eb, &tm->key, tm->slot);
1229 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1230 btrfs_set_node_ptr_generation(eb, tm->slot,
1234 case MOD_LOG_KEY_REPLACE:
1235 BUG_ON(tm->slot >= n);
1236 btrfs_set_node_key(eb, &tm->key, tm->slot);
1237 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1238 btrfs_set_node_ptr_generation(eb, tm->slot,
1241 case MOD_LOG_KEY_ADD:
1242 /* if a move operation is needed it's in the log */
1245 case MOD_LOG_MOVE_KEYS:
1246 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1247 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1248 memmove_extent_buffer(eb, o_dst, o_src,
1249 tm->move.nr_items * p_size);
1251 case MOD_LOG_ROOT_REPLACE:
1253 * this operation is special. for roots, this must be
1254 * handled explicitly before rewinding.
1255 * for non-roots, this operation may exist if the node
1256 * was a root: root A -> child B; then A gets empty and
1257 * B is promoted to the new root. in the mod log, we'll
1258 * have a root-replace operation for B, a tree block
1259 * that is no root. we simply ignore that operation.
1263 next = rb_next(&tm->node);
1266 tm = rb_entry(next, struct tree_mod_elem, node);
1267 if (tm->logical != first_tm->logical)
1270 read_unlock(&fs_info->tree_mod_log_lock);
1271 btrfs_set_header_nritems(eb, n);
1275 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1276 * is returned. If rewind operations happen, a fresh buffer is returned. The
1277 * returned buffer is always read-locked. If the returned buffer is not the
1278 * input buffer, the lock on the input buffer is released and the input buffer
1279 * is freed (its refcount is decremented).
1281 static struct extent_buffer *
1282 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1283 struct extent_buffer *eb, u64 time_seq)
1285 struct extent_buffer *eb_rewin;
1286 struct tree_mod_elem *tm;
1291 if (btrfs_header_level(eb) == 0)
1294 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1298 btrfs_set_path_blocking(path);
1299 btrfs_set_lock_blocking_read(eb);
1301 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1302 BUG_ON(tm->slot != 0);
1303 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1305 btrfs_tree_read_unlock_blocking(eb);
1306 free_extent_buffer(eb);
1309 btrfs_set_header_bytenr(eb_rewin, eb->start);
1310 btrfs_set_header_backref_rev(eb_rewin,
1311 btrfs_header_backref_rev(eb));
1312 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1313 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1315 eb_rewin = btrfs_clone_extent_buffer(eb);
1317 btrfs_tree_read_unlock_blocking(eb);
1318 free_extent_buffer(eb);
1323 btrfs_tree_read_unlock_blocking(eb);
1324 free_extent_buffer(eb);
1326 btrfs_tree_read_lock(eb_rewin);
1327 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1328 WARN_ON(btrfs_header_nritems(eb_rewin) >
1329 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1335 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1336 * value. If there are no changes, the current root->root_node is returned. If
1337 * anything changed in between, there's a fresh buffer allocated on which the
1338 * rewind operations are done. In any case, the returned buffer is read locked.
1339 * Returns NULL on error (with no locks held).
1341 static inline struct extent_buffer *
1342 get_old_root(struct btrfs_root *root, u64 time_seq)
1344 struct btrfs_fs_info *fs_info = root->fs_info;
1345 struct tree_mod_elem *tm;
1346 struct extent_buffer *eb = NULL;
1347 struct extent_buffer *eb_root;
1348 struct extent_buffer *old;
1349 struct tree_mod_root *old_root = NULL;
1350 u64 old_generation = 0;
1354 eb_root = btrfs_read_lock_root_node(root);
1355 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1359 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1360 old_root = &tm->old_root;
1361 old_generation = tm->generation;
1362 logical = old_root->logical;
1363 level = old_root->level;
1365 logical = eb_root->start;
1366 level = btrfs_header_level(eb_root);
1369 tm = tree_mod_log_search(fs_info, logical, time_seq);
1370 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1371 btrfs_tree_read_unlock(eb_root);
1372 free_extent_buffer(eb_root);
1373 old = read_tree_block(fs_info, logical, 0, level, NULL);
1374 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1376 free_extent_buffer(old);
1378 "failed to read tree block %llu from get_old_root",
1381 eb = btrfs_clone_extent_buffer(old);
1382 free_extent_buffer(old);
1384 } else if (old_root) {
1385 btrfs_tree_read_unlock(eb_root);
1386 free_extent_buffer(eb_root);
1387 eb = alloc_dummy_extent_buffer(fs_info, logical);
1389 btrfs_set_lock_blocking_read(eb_root);
1390 eb = btrfs_clone_extent_buffer(eb_root);
1391 btrfs_tree_read_unlock_blocking(eb_root);
1392 free_extent_buffer(eb_root);
1397 btrfs_tree_read_lock(eb);
1399 btrfs_set_header_bytenr(eb, eb->start);
1400 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1401 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1402 btrfs_set_header_level(eb, old_root->level);
1403 btrfs_set_header_generation(eb, old_generation);
1406 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1408 WARN_ON(btrfs_header_level(eb) != 0);
1409 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1414 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1416 struct tree_mod_elem *tm;
1418 struct extent_buffer *eb_root = btrfs_root_node(root);
1420 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1421 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1422 level = tm->old_root.level;
1424 level = btrfs_header_level(eb_root);
1426 free_extent_buffer(eb_root);
1431 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1432 struct btrfs_root *root,
1433 struct extent_buffer *buf)
1435 if (btrfs_is_testing(root->fs_info))
1438 /* Ensure we can see the FORCE_COW bit */
1439 smp_mb__before_atomic();
1442 * We do not need to cow a block if
1443 * 1) this block is not created or changed in this transaction;
1444 * 2) this block does not belong to TREE_RELOC tree;
1445 * 3) the root is not forced COW.
1447 * What is forced COW:
1448 * when we create snapshot during committing the transaction,
1449 * after we've finished copying src root, we must COW the shared
1450 * block to ensure the metadata consistency.
1452 if (btrfs_header_generation(buf) == trans->transid &&
1453 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1454 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1455 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1456 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1462 * cows a single block, see __btrfs_cow_block for the real work.
1463 * This version of it has extra checks so that a block isn't COWed more than
1464 * once per transaction, as long as it hasn't been written yet
1466 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1467 struct btrfs_root *root, struct extent_buffer *buf,
1468 struct extent_buffer *parent, int parent_slot,
1469 struct extent_buffer **cow_ret)
1471 struct btrfs_fs_info *fs_info = root->fs_info;
1475 if (test_bit(BTRFS_ROOT_DELETING, &root->state))
1477 "COW'ing blocks on a fs root that's being dropped");
1479 if (trans->transaction != fs_info->running_transaction)
1480 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1482 fs_info->running_transaction->transid);
1484 if (trans->transid != fs_info->generation)
1485 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1486 trans->transid, fs_info->generation);
1488 if (!should_cow_block(trans, root, buf)) {
1489 trans->dirty = true;
1494 search_start = buf->start & ~((u64)SZ_1G - 1);
1497 btrfs_set_lock_blocking_write(parent);
1498 btrfs_set_lock_blocking_write(buf);
1501 * Before CoWing this block for later modification, check if it's
1502 * the subtree root and do the delayed subtree trace if needed.
1504 * Also We don't care about the error, as it's handled internally.
1506 btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
1507 ret = __btrfs_cow_block(trans, root, buf, parent,
1508 parent_slot, cow_ret, search_start, 0);
1510 trace_btrfs_cow_block(root, buf, *cow_ret);
1516 * helper function for defrag to decide if two blocks pointed to by a
1517 * node are actually close by
1519 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1521 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1523 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1529 * compare two keys in a memcmp fashion
1531 static int comp_keys(const struct btrfs_disk_key *disk,
1532 const struct btrfs_key *k2)
1534 struct btrfs_key k1;
1536 btrfs_disk_key_to_cpu(&k1, disk);
1538 return btrfs_comp_cpu_keys(&k1, k2);
1542 * same as comp_keys only with two btrfs_key's
1544 int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1546 if (k1->objectid > k2->objectid)
1548 if (k1->objectid < k2->objectid)
1550 if (k1->type > k2->type)
1552 if (k1->type < k2->type)
1554 if (k1->offset > k2->offset)
1556 if (k1->offset < k2->offset)
1562 * this is used by the defrag code to go through all the
1563 * leaves pointed to by a node and reallocate them so that
1564 * disk order is close to key order
1566 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1567 struct btrfs_root *root, struct extent_buffer *parent,
1568 int start_slot, u64 *last_ret,
1569 struct btrfs_key *progress)
1571 struct btrfs_fs_info *fs_info = root->fs_info;
1572 struct extent_buffer *cur;
1575 u64 search_start = *last_ret;
1585 int progress_passed = 0;
1586 struct btrfs_disk_key disk_key;
1588 parent_level = btrfs_header_level(parent);
1590 WARN_ON(trans->transaction != fs_info->running_transaction);
1591 WARN_ON(trans->transid != fs_info->generation);
1593 parent_nritems = btrfs_header_nritems(parent);
1594 blocksize = fs_info->nodesize;
1595 end_slot = parent_nritems - 1;
1597 if (parent_nritems <= 1)
1600 btrfs_set_lock_blocking_write(parent);
1602 for (i = start_slot; i <= end_slot; i++) {
1603 struct btrfs_key first_key;
1606 btrfs_node_key(parent, &disk_key, i);
1607 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1610 progress_passed = 1;
1611 blocknr = btrfs_node_blockptr(parent, i);
1612 gen = btrfs_node_ptr_generation(parent, i);
1613 btrfs_node_key_to_cpu(parent, &first_key, i);
1614 if (last_block == 0)
1615 last_block = blocknr;
1618 other = btrfs_node_blockptr(parent, i - 1);
1619 close = close_blocks(blocknr, other, blocksize);
1621 if (!close && i < end_slot) {
1622 other = btrfs_node_blockptr(parent, i + 1);
1623 close = close_blocks(blocknr, other, blocksize);
1626 last_block = blocknr;
1630 cur = find_extent_buffer(fs_info, blocknr);
1632 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1635 if (!cur || !uptodate) {
1637 cur = read_tree_block(fs_info, blocknr, gen,
1641 return PTR_ERR(cur);
1642 } else if (!extent_buffer_uptodate(cur)) {
1643 free_extent_buffer(cur);
1646 } else if (!uptodate) {
1647 err = btrfs_read_buffer(cur, gen,
1648 parent_level - 1,&first_key);
1650 free_extent_buffer(cur);
1655 if (search_start == 0)
1656 search_start = last_block;
1658 btrfs_tree_lock(cur);
1659 btrfs_set_lock_blocking_write(cur);
1660 err = __btrfs_cow_block(trans, root, cur, parent, i,
1663 (end_slot - i) * blocksize));
1665 btrfs_tree_unlock(cur);
1666 free_extent_buffer(cur);
1669 search_start = cur->start;
1670 last_block = cur->start;
1671 *last_ret = search_start;
1672 btrfs_tree_unlock(cur);
1673 free_extent_buffer(cur);
1679 * search for key in the extent_buffer. The items start at offset p,
1680 * and they are item_size apart. There are 'max' items in p.
1682 * the slot in the array is returned via slot, and it points to
1683 * the place where you would insert key if it is not found in
1686 * slot may point to max if the key is bigger than all of the keys
1688 static noinline int generic_bin_search(struct extent_buffer *eb,
1689 unsigned long p, int item_size,
1690 const struct btrfs_key *key,
1697 struct btrfs_disk_key *tmp = NULL;
1698 struct btrfs_disk_key unaligned;
1699 unsigned long offset;
1701 unsigned long map_start = 0;
1702 unsigned long map_len = 0;
1706 btrfs_err(eb->fs_info,
1707 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1708 __func__, low, high, eb->start,
1709 btrfs_header_owner(eb), btrfs_header_level(eb));
1713 while (low < high) {
1714 mid = (low + high) / 2;
1715 offset = p + mid * item_size;
1717 if (!kaddr || offset < map_start ||
1718 (offset + sizeof(struct btrfs_disk_key)) >
1719 map_start + map_len) {
1721 err = map_private_extent_buffer(eb, offset,
1722 sizeof(struct btrfs_disk_key),
1723 &kaddr, &map_start, &map_len);
1726 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1728 } else if (err == 1) {
1729 read_extent_buffer(eb, &unaligned,
1730 offset, sizeof(unaligned));
1737 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1740 ret = comp_keys(tmp, key);
1756 * simple bin_search frontend that does the right thing for
1759 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1760 int level, int *slot)
1763 return generic_bin_search(eb,
1764 offsetof(struct btrfs_leaf, items),
1765 sizeof(struct btrfs_item),
1766 key, btrfs_header_nritems(eb),
1769 return generic_bin_search(eb,
1770 offsetof(struct btrfs_node, ptrs),
1771 sizeof(struct btrfs_key_ptr),
1772 key, btrfs_header_nritems(eb),
1776 static void root_add_used(struct btrfs_root *root, u32 size)
1778 spin_lock(&root->accounting_lock);
1779 btrfs_set_root_used(&root->root_item,
1780 btrfs_root_used(&root->root_item) + size);
1781 spin_unlock(&root->accounting_lock);
1784 static void root_sub_used(struct btrfs_root *root, u32 size)
1786 spin_lock(&root->accounting_lock);
1787 btrfs_set_root_used(&root->root_item,
1788 btrfs_root_used(&root->root_item) - size);
1789 spin_unlock(&root->accounting_lock);
1792 /* given a node and slot number, this reads the blocks it points to. The
1793 * extent buffer is returned with a reference taken (but unlocked).
1795 static noinline struct extent_buffer *
1796 read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent,
1799 int level = btrfs_header_level(parent);
1800 struct extent_buffer *eb;
1801 struct btrfs_key first_key;
1803 if (slot < 0 || slot >= btrfs_header_nritems(parent))
1804 return ERR_PTR(-ENOENT);
1808 btrfs_node_key_to_cpu(parent, &first_key, slot);
1809 eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot),
1810 btrfs_node_ptr_generation(parent, slot),
1811 level - 1, &first_key);
1812 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1813 free_extent_buffer(eb);
1821 * node level balancing, used to make sure nodes are in proper order for
1822 * item deletion. We balance from the top down, so we have to make sure
1823 * that a deletion won't leave an node completely empty later on.
1825 static noinline int balance_level(struct btrfs_trans_handle *trans,
1826 struct btrfs_root *root,
1827 struct btrfs_path *path, int level)
1829 struct btrfs_fs_info *fs_info = root->fs_info;
1830 struct extent_buffer *right = NULL;
1831 struct extent_buffer *mid;
1832 struct extent_buffer *left = NULL;
1833 struct extent_buffer *parent = NULL;
1837 int orig_slot = path->slots[level];
1842 mid = path->nodes[level];
1844 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1845 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1846 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1848 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1850 if (level < BTRFS_MAX_LEVEL - 1) {
1851 parent = path->nodes[level + 1];
1852 pslot = path->slots[level + 1];
1856 * deal with the case where there is only one pointer in the root
1857 * by promoting the node below to a root
1860 struct extent_buffer *child;
1862 if (btrfs_header_nritems(mid) != 1)
1865 /* promote the child to a root */
1866 child = read_node_slot(fs_info, mid, 0);
1867 if (IS_ERR(child)) {
1868 ret = PTR_ERR(child);
1869 btrfs_handle_fs_error(fs_info, ret, NULL);
1873 btrfs_tree_lock(child);
1874 btrfs_set_lock_blocking_write(child);
1875 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1877 btrfs_tree_unlock(child);
1878 free_extent_buffer(child);
1882 ret = tree_mod_log_insert_root(root->node, child, 1);
1884 rcu_assign_pointer(root->node, child);
1886 add_root_to_dirty_list(root);
1887 btrfs_tree_unlock(child);
1889 path->locks[level] = 0;
1890 path->nodes[level] = NULL;
1891 clean_tree_block(fs_info, mid);
1892 btrfs_tree_unlock(mid);
1893 /* once for the path */
1894 free_extent_buffer(mid);
1896 root_sub_used(root, mid->len);
1897 btrfs_free_tree_block(trans, root, mid, 0, 1);
1898 /* once for the root ptr */
1899 free_extent_buffer_stale(mid);
1902 if (btrfs_header_nritems(mid) >
1903 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1906 left = read_node_slot(fs_info, parent, pslot - 1);
1911 btrfs_tree_lock(left);
1912 btrfs_set_lock_blocking_write(left);
1913 wret = btrfs_cow_block(trans, root, left,
1914 parent, pslot - 1, &left);
1921 right = read_node_slot(fs_info, parent, pslot + 1);
1926 btrfs_tree_lock(right);
1927 btrfs_set_lock_blocking_write(right);
1928 wret = btrfs_cow_block(trans, root, right,
1929 parent, pslot + 1, &right);
1936 /* first, try to make some room in the middle buffer */
1938 orig_slot += btrfs_header_nritems(left);
1939 wret = push_node_left(trans, fs_info, left, mid, 1);
1945 * then try to empty the right most buffer into the middle
1948 wret = push_node_left(trans, fs_info, mid, right, 1);
1949 if (wret < 0 && wret != -ENOSPC)
1951 if (btrfs_header_nritems(right) == 0) {
1952 clean_tree_block(fs_info, right);
1953 btrfs_tree_unlock(right);
1954 del_ptr(root, path, level + 1, pslot + 1);
1955 root_sub_used(root, right->len);
1956 btrfs_free_tree_block(trans, root, right, 0, 1);
1957 free_extent_buffer_stale(right);
1960 struct btrfs_disk_key right_key;
1961 btrfs_node_key(right, &right_key, 0);
1962 ret = tree_mod_log_insert_key(parent, pslot + 1,
1963 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1965 btrfs_set_node_key(parent, &right_key, pslot + 1);
1966 btrfs_mark_buffer_dirty(parent);
1969 if (btrfs_header_nritems(mid) == 1) {
1971 * we're not allowed to leave a node with one item in the
1972 * tree during a delete. A deletion from lower in the tree
1973 * could try to delete the only pointer in this node.
1974 * So, pull some keys from the left.
1975 * There has to be a left pointer at this point because
1976 * otherwise we would have pulled some pointers from the
1981 btrfs_handle_fs_error(fs_info, ret, NULL);
1984 wret = balance_node_right(trans, fs_info, mid, left);
1990 wret = push_node_left(trans, fs_info, left, mid, 1);
1996 if (btrfs_header_nritems(mid) == 0) {
1997 clean_tree_block(fs_info, mid);
1998 btrfs_tree_unlock(mid);
1999 del_ptr(root, path, level + 1, pslot);
2000 root_sub_used(root, mid->len);
2001 btrfs_free_tree_block(trans, root, mid, 0, 1);
2002 free_extent_buffer_stale(mid);
2005 /* update the parent key to reflect our changes */
2006 struct btrfs_disk_key mid_key;
2007 btrfs_node_key(mid, &mid_key, 0);
2008 ret = tree_mod_log_insert_key(parent, pslot,
2009 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2011 btrfs_set_node_key(parent, &mid_key, pslot);
2012 btrfs_mark_buffer_dirty(parent);
2015 /* update the path */
2017 if (btrfs_header_nritems(left) > orig_slot) {
2018 extent_buffer_get(left);
2019 /* left was locked after cow */
2020 path->nodes[level] = left;
2021 path->slots[level + 1] -= 1;
2022 path->slots[level] = orig_slot;
2024 btrfs_tree_unlock(mid);
2025 free_extent_buffer(mid);
2028 orig_slot -= btrfs_header_nritems(left);
2029 path->slots[level] = orig_slot;
2032 /* double check we haven't messed things up */
2034 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2038 btrfs_tree_unlock(right);
2039 free_extent_buffer(right);
2042 if (path->nodes[level] != left)
2043 btrfs_tree_unlock(left);
2044 free_extent_buffer(left);
2049 /* Node balancing for insertion. Here we only split or push nodes around
2050 * when they are completely full. This is also done top down, so we
2051 * have to be pessimistic.
2053 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2054 struct btrfs_root *root,
2055 struct btrfs_path *path, int level)
2057 struct btrfs_fs_info *fs_info = root->fs_info;
2058 struct extent_buffer *right = NULL;
2059 struct extent_buffer *mid;
2060 struct extent_buffer *left = NULL;
2061 struct extent_buffer *parent = NULL;
2065 int orig_slot = path->slots[level];
2070 mid = path->nodes[level];
2071 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2073 if (level < BTRFS_MAX_LEVEL - 1) {
2074 parent = path->nodes[level + 1];
2075 pslot = path->slots[level + 1];
2081 left = read_node_slot(fs_info, parent, pslot - 1);
2085 /* first, try to make some room in the middle buffer */
2089 btrfs_tree_lock(left);
2090 btrfs_set_lock_blocking_write(left);
2092 left_nr = btrfs_header_nritems(left);
2093 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2096 ret = btrfs_cow_block(trans, root, left, parent,
2101 wret = push_node_left(trans, fs_info,
2108 struct btrfs_disk_key disk_key;
2109 orig_slot += left_nr;
2110 btrfs_node_key(mid, &disk_key, 0);
2111 ret = tree_mod_log_insert_key(parent, pslot,
2112 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2114 btrfs_set_node_key(parent, &disk_key, pslot);
2115 btrfs_mark_buffer_dirty(parent);
2116 if (btrfs_header_nritems(left) > orig_slot) {
2117 path->nodes[level] = left;
2118 path->slots[level + 1] -= 1;
2119 path->slots[level] = orig_slot;
2120 btrfs_tree_unlock(mid);
2121 free_extent_buffer(mid);
2124 btrfs_header_nritems(left);
2125 path->slots[level] = orig_slot;
2126 btrfs_tree_unlock(left);
2127 free_extent_buffer(left);
2131 btrfs_tree_unlock(left);
2132 free_extent_buffer(left);
2134 right = read_node_slot(fs_info, parent, pslot + 1);
2139 * then try to empty the right most buffer into the middle
2144 btrfs_tree_lock(right);
2145 btrfs_set_lock_blocking_write(right);
2147 right_nr = btrfs_header_nritems(right);
2148 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2151 ret = btrfs_cow_block(trans, root, right,
2157 wret = balance_node_right(trans, fs_info,
2164 struct btrfs_disk_key disk_key;
2166 btrfs_node_key(right, &disk_key, 0);
2167 ret = tree_mod_log_insert_key(parent, pslot + 1,
2168 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2170 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2171 btrfs_mark_buffer_dirty(parent);
2173 if (btrfs_header_nritems(mid) <= orig_slot) {
2174 path->nodes[level] = right;
2175 path->slots[level + 1] += 1;
2176 path->slots[level] = orig_slot -
2177 btrfs_header_nritems(mid);
2178 btrfs_tree_unlock(mid);
2179 free_extent_buffer(mid);
2181 btrfs_tree_unlock(right);
2182 free_extent_buffer(right);
2186 btrfs_tree_unlock(right);
2187 free_extent_buffer(right);
2193 * readahead one full node of leaves, finding things that are close
2194 * to the block in 'slot', and triggering ra on them.
2196 static void reada_for_search(struct btrfs_fs_info *fs_info,
2197 struct btrfs_path *path,
2198 int level, int slot, u64 objectid)
2200 struct extent_buffer *node;
2201 struct btrfs_disk_key disk_key;
2206 struct extent_buffer *eb;
2214 if (!path->nodes[level])
2217 node = path->nodes[level];
2219 search = btrfs_node_blockptr(node, slot);
2220 blocksize = fs_info->nodesize;
2221 eb = find_extent_buffer(fs_info, search);
2223 free_extent_buffer(eb);
2229 nritems = btrfs_header_nritems(node);
2233 if (path->reada == READA_BACK) {
2237 } else if (path->reada == READA_FORWARD) {
2242 if (path->reada == READA_BACK && objectid) {
2243 btrfs_node_key(node, &disk_key, nr);
2244 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2247 search = btrfs_node_blockptr(node, nr);
2248 if ((search <= target && target - search <= 65536) ||
2249 (search > target && search - target <= 65536)) {
2250 readahead_tree_block(fs_info, search);
2254 if ((nread > 65536 || nscan > 32))
2259 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2260 struct btrfs_path *path, int level)
2264 struct extent_buffer *parent;
2265 struct extent_buffer *eb;
2270 parent = path->nodes[level + 1];
2274 nritems = btrfs_header_nritems(parent);
2275 slot = path->slots[level + 1];
2278 block1 = btrfs_node_blockptr(parent, slot - 1);
2279 gen = btrfs_node_ptr_generation(parent, slot - 1);
2280 eb = find_extent_buffer(fs_info, block1);
2282 * if we get -eagain from btrfs_buffer_uptodate, we
2283 * don't want to return eagain here. That will loop
2286 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2288 free_extent_buffer(eb);
2290 if (slot + 1 < nritems) {
2291 block2 = btrfs_node_blockptr(parent, slot + 1);
2292 gen = btrfs_node_ptr_generation(parent, slot + 1);
2293 eb = find_extent_buffer(fs_info, block2);
2294 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2296 free_extent_buffer(eb);
2300 readahead_tree_block(fs_info, block1);
2302 readahead_tree_block(fs_info, block2);
2307 * when we walk down the tree, it is usually safe to unlock the higher layers
2308 * in the tree. The exceptions are when our path goes through slot 0, because
2309 * operations on the tree might require changing key pointers higher up in the
2312 * callers might also have set path->keep_locks, which tells this code to keep
2313 * the lock if the path points to the last slot in the block. This is part of
2314 * walking through the tree, and selecting the next slot in the higher block.
2316 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2317 * if lowest_unlock is 1, level 0 won't be unlocked
2319 static noinline void unlock_up(struct btrfs_path *path, int level,
2320 int lowest_unlock, int min_write_lock_level,
2321 int *write_lock_level)
2324 int skip_level = level;
2326 struct extent_buffer *t;
2328 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2329 if (!path->nodes[i])
2331 if (!path->locks[i])
2333 if (!no_skips && path->slots[i] == 0) {
2337 if (!no_skips && path->keep_locks) {
2340 nritems = btrfs_header_nritems(t);
2341 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2346 if (skip_level < i && i >= lowest_unlock)
2350 if (i >= lowest_unlock && i > skip_level) {
2351 btrfs_tree_unlock_rw(t, path->locks[i]);
2353 if (write_lock_level &&
2354 i > min_write_lock_level &&
2355 i <= *write_lock_level) {
2356 *write_lock_level = i - 1;
2363 * This releases any locks held in the path starting at level and
2364 * going all the way up to the root.
2366 * btrfs_search_slot will keep the lock held on higher nodes in a few
2367 * corner cases, such as COW of the block at slot zero in the node. This
2368 * ignores those rules, and it should only be called when there are no
2369 * more updates to be done higher up in the tree.
2371 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2375 if (path->keep_locks)
2378 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2379 if (!path->nodes[i])
2381 if (!path->locks[i])
2383 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2389 * helper function for btrfs_search_slot. The goal is to find a block
2390 * in cache without setting the path to blocking. If we find the block
2391 * we return zero and the path is unchanged.
2393 * If we can't find the block, we set the path blocking and do some
2394 * reada. -EAGAIN is returned and the search must be repeated.
2397 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2398 struct extent_buffer **eb_ret, int level, int slot,
2399 const struct btrfs_key *key)
2401 struct btrfs_fs_info *fs_info = root->fs_info;
2404 struct extent_buffer *b = *eb_ret;
2405 struct extent_buffer *tmp;
2406 struct btrfs_key first_key;
2410 blocknr = btrfs_node_blockptr(b, slot);
2411 gen = btrfs_node_ptr_generation(b, slot);
2412 parent_level = btrfs_header_level(b);
2413 btrfs_node_key_to_cpu(b, &first_key, slot);
2415 tmp = find_extent_buffer(fs_info, blocknr);
2417 /* first we do an atomic uptodate check */
2418 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2423 /* the pages were up to date, but we failed
2424 * the generation number check. Do a full
2425 * read for the generation number that is correct.
2426 * We must do this without dropping locks so
2427 * we can trust our generation number
2429 btrfs_set_path_blocking(p);
2431 /* now we're allowed to do a blocking uptodate check */
2432 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2437 free_extent_buffer(tmp);
2438 btrfs_release_path(p);
2443 * reduce lock contention at high levels
2444 * of the btree by dropping locks before
2445 * we read. Don't release the lock on the current
2446 * level because we need to walk this node to figure
2447 * out which blocks to read.
2449 btrfs_unlock_up_safe(p, level + 1);
2450 btrfs_set_path_blocking(p);
2452 if (p->reada != READA_NONE)
2453 reada_for_search(fs_info, p, level, slot, key->objectid);
2456 tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2460 * If the read above didn't mark this buffer up to date,
2461 * it will never end up being up to date. Set ret to EIO now
2462 * and give up so that our caller doesn't loop forever
2465 if (!extent_buffer_uptodate(tmp))
2467 free_extent_buffer(tmp);
2472 btrfs_release_path(p);
2477 * helper function for btrfs_search_slot. This does all of the checks
2478 * for node-level blocks and does any balancing required based on
2481 * If no extra work was required, zero is returned. If we had to
2482 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2486 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2487 struct btrfs_root *root, struct btrfs_path *p,
2488 struct extent_buffer *b, int level, int ins_len,
2489 int *write_lock_level)
2491 struct btrfs_fs_info *fs_info = root->fs_info;
2494 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2495 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2498 if (*write_lock_level < level + 1) {
2499 *write_lock_level = level + 1;
2500 btrfs_release_path(p);
2504 btrfs_set_path_blocking(p);
2505 reada_for_balance(fs_info, p, level);
2506 sret = split_node(trans, root, p, level);
2513 b = p->nodes[level];
2514 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2515 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2518 if (*write_lock_level < level + 1) {
2519 *write_lock_level = level + 1;
2520 btrfs_release_path(p);
2524 btrfs_set_path_blocking(p);
2525 reada_for_balance(fs_info, p, level);
2526 sret = balance_level(trans, root, p, level);
2532 b = p->nodes[level];
2534 btrfs_release_path(p);
2537 BUG_ON(btrfs_header_nritems(b) == 1);
2547 static void key_search_validate(struct extent_buffer *b,
2548 const struct btrfs_key *key,
2551 #ifdef CONFIG_BTRFS_ASSERT
2552 struct btrfs_disk_key disk_key;
2554 btrfs_cpu_key_to_disk(&disk_key, key);
2557 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2558 offsetof(struct btrfs_leaf, items[0].key),
2561 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2562 offsetof(struct btrfs_node, ptrs[0].key),
2567 static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
2568 int level, int *prev_cmp, int *slot)
2570 if (*prev_cmp != 0) {
2571 *prev_cmp = btrfs_bin_search(b, key, level, slot);
2575 key_search_validate(b, key, level);
2581 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2582 u64 iobjectid, u64 ioff, u8 key_type,
2583 struct btrfs_key *found_key)
2586 struct btrfs_key key;
2587 struct extent_buffer *eb;
2592 key.type = key_type;
2593 key.objectid = iobjectid;
2596 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2600 eb = path->nodes[0];
2601 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2602 ret = btrfs_next_leaf(fs_root, path);
2605 eb = path->nodes[0];
2608 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2609 if (found_key->type != key.type ||
2610 found_key->objectid != key.objectid)
2616 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2617 struct btrfs_path *p,
2618 int write_lock_level)
2620 struct btrfs_fs_info *fs_info = root->fs_info;
2621 struct extent_buffer *b;
2625 /* We try very hard to do read locks on the root */
2626 root_lock = BTRFS_READ_LOCK;
2628 if (p->search_commit_root) {
2630 * The commit roots are read only so we always do read locks,
2631 * and we always must hold the commit_root_sem when doing
2632 * searches on them, the only exception is send where we don't
2633 * want to block transaction commits for a long time, so
2634 * we need to clone the commit root in order to avoid races
2635 * with transaction commits that create a snapshot of one of
2636 * the roots used by a send operation.
2638 if (p->need_commit_sem) {
2639 down_read(&fs_info->commit_root_sem);
2640 b = btrfs_clone_extent_buffer(root->commit_root);
2641 up_read(&fs_info->commit_root_sem);
2643 return ERR_PTR(-ENOMEM);
2646 b = root->commit_root;
2647 extent_buffer_get(b);
2649 level = btrfs_header_level(b);
2651 * Ensure that all callers have set skip_locking when
2652 * p->search_commit_root = 1.
2654 ASSERT(p->skip_locking == 1);
2659 if (p->skip_locking) {
2660 b = btrfs_root_node(root);
2661 level = btrfs_header_level(b);
2666 * If the level is set to maximum, we can skip trying to get the read
2669 if (write_lock_level < BTRFS_MAX_LEVEL) {
2671 * We don't know the level of the root node until we actually
2672 * have it read locked
2674 b = btrfs_read_lock_root_node(root);
2675 level = btrfs_header_level(b);
2676 if (level > write_lock_level)
2679 /* Whoops, must trade for write lock */
2680 btrfs_tree_read_unlock(b);
2681 free_extent_buffer(b);
2684 b = btrfs_lock_root_node(root);
2685 root_lock = BTRFS_WRITE_LOCK;
2687 /* The level might have changed, check again */
2688 level = btrfs_header_level(b);
2691 p->nodes[level] = b;
2692 if (!p->skip_locking)
2693 p->locks[level] = root_lock;
2695 * Callers are responsible for dropping b's references.
2702 * btrfs_search_slot - look for a key in a tree and perform necessary
2703 * modifications to preserve tree invariants.
2705 * @trans: Handle of transaction, used when modifying the tree
2706 * @p: Holds all btree nodes along the search path
2707 * @root: The root node of the tree
2708 * @key: The key we are looking for
2709 * @ins_len: Indicates purpose of search, for inserts it is 1, for
2710 * deletions it's -1. 0 for plain searches
2711 * @cow: boolean should CoW operations be performed. Must always be 1
2712 * when modifying the tree.
2714 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2715 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2717 * If @key is found, 0 is returned and you can find the item in the leaf level
2718 * of the path (level 0)
2720 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2721 * points to the slot where it should be inserted
2723 * If an error is encountered while searching the tree a negative error number
2726 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2727 const struct btrfs_key *key, struct btrfs_path *p,
2728 int ins_len, int cow)
2730 struct btrfs_fs_info *fs_info = root->fs_info;
2731 struct extent_buffer *b;
2736 int lowest_unlock = 1;
2737 /* everything at write_lock_level or lower must be write locked */
2738 int write_lock_level = 0;
2739 u8 lowest_level = 0;
2740 int min_write_lock_level;
2743 lowest_level = p->lowest_level;
2744 WARN_ON(lowest_level && ins_len > 0);
2745 WARN_ON(p->nodes[0] != NULL);
2746 BUG_ON(!cow && ins_len);
2751 /* when we are removing items, we might have to go up to level
2752 * two as we update tree pointers Make sure we keep write
2753 * for those levels as well
2755 write_lock_level = 2;
2756 } else if (ins_len > 0) {
2758 * for inserting items, make sure we have a write lock on
2759 * level 1 so we can update keys
2761 write_lock_level = 1;
2765 write_lock_level = -1;
2767 if (cow && (p->keep_locks || p->lowest_level))
2768 write_lock_level = BTRFS_MAX_LEVEL;
2770 min_write_lock_level = write_lock_level;
2774 b = btrfs_search_slot_get_root(root, p, write_lock_level);
2781 level = btrfs_header_level(b);
2784 * setup the path here so we can release it under lock
2785 * contention with the cow code
2788 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2791 * if we don't really need to cow this block
2792 * then we don't want to set the path blocking,
2793 * so we test it here
2795 if (!should_cow_block(trans, root, b)) {
2796 trans->dirty = true;
2801 * must have write locks on this node and the
2804 if (level > write_lock_level ||
2805 (level + 1 > write_lock_level &&
2806 level + 1 < BTRFS_MAX_LEVEL &&
2807 p->nodes[level + 1])) {
2808 write_lock_level = level + 1;
2809 btrfs_release_path(p);
2813 btrfs_set_path_blocking(p);
2815 err = btrfs_cow_block(trans, root, b, NULL, 0,
2818 err = btrfs_cow_block(trans, root, b,
2819 p->nodes[level + 1],
2820 p->slots[level + 1], &b);
2827 p->nodes[level] = b;
2829 * Leave path with blocking locks to avoid massive
2830 * lock context switch, this is made on purpose.
2834 * we have a lock on b and as long as we aren't changing
2835 * the tree, there is no way to for the items in b to change.
2836 * It is safe to drop the lock on our parent before we
2837 * go through the expensive btree search on b.
2839 * If we're inserting or deleting (ins_len != 0), then we might
2840 * be changing slot zero, which may require changing the parent.
2841 * So, we can't drop the lock until after we know which slot
2842 * we're operating on.
2844 if (!ins_len && !p->keep_locks) {
2847 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2848 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2853 ret = key_search(b, key, level, &prev_cmp, &slot);
2859 if (ret && slot > 0) {
2863 p->slots[level] = slot;
2864 err = setup_nodes_for_search(trans, root, p, b, level,
2865 ins_len, &write_lock_level);
2872 b = p->nodes[level];
2873 slot = p->slots[level];
2876 * slot 0 is special, if we change the key
2877 * we have to update the parent pointer
2878 * which means we must have a write lock
2881 if (slot == 0 && ins_len &&
2882 write_lock_level < level + 1) {
2883 write_lock_level = level + 1;
2884 btrfs_release_path(p);
2888 unlock_up(p, level, lowest_unlock,
2889 min_write_lock_level, &write_lock_level);
2891 if (level == lowest_level) {
2897 err = read_block_for_search(root, p, &b, level,
2906 if (!p->skip_locking) {
2907 level = btrfs_header_level(b);
2908 if (level <= write_lock_level) {
2909 err = btrfs_try_tree_write_lock(b);
2911 btrfs_set_path_blocking(p);
2914 p->locks[level] = BTRFS_WRITE_LOCK;
2916 err = btrfs_tree_read_lock_atomic(b);
2918 btrfs_set_path_blocking(p);
2919 btrfs_tree_read_lock(b);
2921 p->locks[level] = BTRFS_READ_LOCK;
2923 p->nodes[level] = b;
2926 p->slots[level] = slot;
2928 btrfs_leaf_free_space(fs_info, b) < ins_len) {
2929 if (write_lock_level < 1) {
2930 write_lock_level = 1;
2931 btrfs_release_path(p);
2935 btrfs_set_path_blocking(p);
2936 err = split_leaf(trans, root, key,
2937 p, ins_len, ret == 0);
2945 if (!p->search_for_split)
2946 unlock_up(p, level, lowest_unlock,
2947 min_write_lock_level, NULL);
2954 * we don't really know what they plan on doing with the path
2955 * from here on, so for now just mark it as blocking
2957 if (!p->leave_spinning)
2958 btrfs_set_path_blocking(p);
2959 if (ret < 0 && !p->skip_release_on_error)
2960 btrfs_release_path(p);
2965 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2966 * current state of the tree together with the operations recorded in the tree
2967 * modification log to search for the key in a previous version of this tree, as
2968 * denoted by the time_seq parameter.
2970 * Naturally, there is no support for insert, delete or cow operations.
2972 * The resulting path and return value will be set up as if we called
2973 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2975 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2976 struct btrfs_path *p, u64 time_seq)
2978 struct btrfs_fs_info *fs_info = root->fs_info;
2979 struct extent_buffer *b;
2984 int lowest_unlock = 1;
2985 u8 lowest_level = 0;
2988 lowest_level = p->lowest_level;
2989 WARN_ON(p->nodes[0] != NULL);
2991 if (p->search_commit_root) {
2993 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2997 b = get_old_root(root, time_seq);
3002 level = btrfs_header_level(b);
3003 p->locks[level] = BTRFS_READ_LOCK;
3006 level = btrfs_header_level(b);
3007 p->nodes[level] = b;
3010 * we have a lock on b and as long as we aren't changing
3011 * the tree, there is no way to for the items in b to change.
3012 * It is safe to drop the lock on our parent before we
3013 * go through the expensive btree search on b.
3015 btrfs_unlock_up_safe(p, level + 1);
3018 * Since we can unwind ebs we want to do a real search every
3022 ret = key_search(b, key, level, &prev_cmp, &slot);
3026 if (ret && slot > 0) {
3030 p->slots[level] = slot;
3031 unlock_up(p, level, lowest_unlock, 0, NULL);
3033 if (level == lowest_level) {
3039 err = read_block_for_search(root, p, &b, level,
3048 level = btrfs_header_level(b);
3049 err = btrfs_tree_read_lock_atomic(b);
3051 btrfs_set_path_blocking(p);
3052 btrfs_tree_read_lock(b);
3054 b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3059 p->locks[level] = BTRFS_READ_LOCK;
3060 p->nodes[level] = b;
3062 p->slots[level] = slot;
3063 unlock_up(p, level, lowest_unlock, 0, NULL);
3069 if (!p->leave_spinning)
3070 btrfs_set_path_blocking(p);
3072 btrfs_release_path(p);
3078 * helper to use instead of search slot if no exact match is needed but
3079 * instead the next or previous item should be returned.
3080 * When find_higher is true, the next higher item is returned, the next lower
3082 * When return_any and find_higher are both true, and no higher item is found,
3083 * return the next lower instead.
3084 * When return_any is true and find_higher is false, and no lower item is found,
3085 * return the next higher instead.
3086 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3089 int btrfs_search_slot_for_read(struct btrfs_root *root,
3090 const struct btrfs_key *key,
3091 struct btrfs_path *p, int find_higher,
3095 struct extent_buffer *leaf;
3098 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3102 * a return value of 1 means the path is at the position where the
3103 * item should be inserted. Normally this is the next bigger item,
3104 * but in case the previous item is the last in a leaf, path points
3105 * to the first free slot in the previous leaf, i.e. at an invalid
3111 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3112 ret = btrfs_next_leaf(root, p);
3118 * no higher item found, return the next
3123 btrfs_release_path(p);
3127 if (p->slots[0] == 0) {
3128 ret = btrfs_prev_leaf(root, p);
3133 if (p->slots[0] == btrfs_header_nritems(leaf))
3140 * no lower item found, return the next
3145 btrfs_release_path(p);
3155 * adjust the pointers going up the tree, starting at level
3156 * making sure the right key of each node is points to 'key'.
3157 * This is used after shifting pointers to the left, so it stops
3158 * fixing up pointers when a given leaf/node is not in slot 0 of the
3162 static void fixup_low_keys(struct btrfs_path *path,
3163 struct btrfs_disk_key *key, int level)
3166 struct extent_buffer *t;
3169 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3170 int tslot = path->slots[i];
3172 if (!path->nodes[i])
3175 ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3178 btrfs_set_node_key(t, key, tslot);
3179 btrfs_mark_buffer_dirty(path->nodes[i]);
3188 * This function isn't completely safe. It's the caller's responsibility
3189 * that the new key won't break the order
3191 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3192 struct btrfs_path *path,
3193 const struct btrfs_key *new_key)
3195 struct btrfs_disk_key disk_key;
3196 struct extent_buffer *eb;
3199 eb = path->nodes[0];
3200 slot = path->slots[0];
3202 btrfs_item_key(eb, &disk_key, slot - 1);
3203 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3205 if (slot < btrfs_header_nritems(eb) - 1) {
3206 btrfs_item_key(eb, &disk_key, slot + 1);
3207 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3210 btrfs_cpu_key_to_disk(&disk_key, new_key);
3211 btrfs_set_item_key(eb, &disk_key, slot);
3212 btrfs_mark_buffer_dirty(eb);
3214 fixup_low_keys(path, &disk_key, 1);
3218 * try to push data from one node into the next node left in the
3221 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3222 * error, and > 0 if there was no room in the left hand block.
3224 static int push_node_left(struct btrfs_trans_handle *trans,
3225 struct btrfs_fs_info *fs_info,
3226 struct extent_buffer *dst,
3227 struct extent_buffer *src, int empty)
3234 src_nritems = btrfs_header_nritems(src);
3235 dst_nritems = btrfs_header_nritems(dst);
3236 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3237 WARN_ON(btrfs_header_generation(src) != trans->transid);
3238 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3240 if (!empty && src_nritems <= 8)
3243 if (push_items <= 0)
3247 push_items = min(src_nritems, push_items);
3248 if (push_items < src_nritems) {
3249 /* leave at least 8 pointers in the node if
3250 * we aren't going to empty it
3252 if (src_nritems - push_items < 8) {
3253 if (push_items <= 8)
3259 push_items = min(src_nritems - 8, push_items);
3261 ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0,
3264 btrfs_abort_transaction(trans, ret);
3267 copy_extent_buffer(dst, src,
3268 btrfs_node_key_ptr_offset(dst_nritems),
3269 btrfs_node_key_ptr_offset(0),
3270 push_items * sizeof(struct btrfs_key_ptr));
3272 if (push_items < src_nritems) {
3274 * Don't call tree_mod_log_insert_move here, key removal was
3275 * already fully logged by tree_mod_log_eb_copy above.
3277 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3278 btrfs_node_key_ptr_offset(push_items),
3279 (src_nritems - push_items) *
3280 sizeof(struct btrfs_key_ptr));
3282 btrfs_set_header_nritems(src, src_nritems - push_items);
3283 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3284 btrfs_mark_buffer_dirty(src);
3285 btrfs_mark_buffer_dirty(dst);
3291 * try to push data from one node into the next node right in the
3294 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3295 * error, and > 0 if there was no room in the right hand block.
3297 * this will only push up to 1/2 the contents of the left node over
3299 static int balance_node_right(struct btrfs_trans_handle *trans,
3300 struct btrfs_fs_info *fs_info,
3301 struct extent_buffer *dst,
3302 struct extent_buffer *src)
3310 WARN_ON(btrfs_header_generation(src) != trans->transid);
3311 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3313 src_nritems = btrfs_header_nritems(src);
3314 dst_nritems = btrfs_header_nritems(dst);
3315 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3316 if (push_items <= 0)
3319 if (src_nritems < 4)
3322 max_push = src_nritems / 2 + 1;
3323 /* don't try to empty the node */
3324 if (max_push >= src_nritems)
3327 if (max_push < push_items)
3328 push_items = max_push;
3330 ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3332 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3333 btrfs_node_key_ptr_offset(0),
3335 sizeof(struct btrfs_key_ptr));
3337 ret = tree_mod_log_eb_copy(fs_info, dst, src, 0,
3338 src_nritems - push_items, push_items);
3340 btrfs_abort_transaction(trans, ret);
3343 copy_extent_buffer(dst, src,
3344 btrfs_node_key_ptr_offset(0),
3345 btrfs_node_key_ptr_offset(src_nritems - push_items),
3346 push_items * sizeof(struct btrfs_key_ptr));
3348 btrfs_set_header_nritems(src, src_nritems - push_items);
3349 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3351 btrfs_mark_buffer_dirty(src);
3352 btrfs_mark_buffer_dirty(dst);
3358 * helper function to insert a new root level in the tree.
3359 * A new node is allocated, and a single item is inserted to
3360 * point to the existing root
3362 * returns zero on success or < 0 on failure.
3364 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3365 struct btrfs_root *root,
3366 struct btrfs_path *path, int level)
3368 struct btrfs_fs_info *fs_info = root->fs_info;
3370 struct extent_buffer *lower;
3371 struct extent_buffer *c;
3372 struct extent_buffer *old;
3373 struct btrfs_disk_key lower_key;
3376 BUG_ON(path->nodes[level]);
3377 BUG_ON(path->nodes[level-1] != root->node);
3379 lower = path->nodes[level-1];
3381 btrfs_item_key(lower, &lower_key, 0);
3383 btrfs_node_key(lower, &lower_key, 0);
3385 c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level,
3386 root->node->start, 0);
3390 root_add_used(root, fs_info->nodesize);
3392 btrfs_set_header_nritems(c, 1);
3393 btrfs_set_node_key(c, &lower_key, 0);
3394 btrfs_set_node_blockptr(c, 0, lower->start);
3395 lower_gen = btrfs_header_generation(lower);
3396 WARN_ON(lower_gen != trans->transid);
3398 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3400 btrfs_mark_buffer_dirty(c);
3403 ret = tree_mod_log_insert_root(root->node, c, 0);
3405 rcu_assign_pointer(root->node, c);
3407 /* the super has an extra ref to root->node */
3408 free_extent_buffer(old);
3410 add_root_to_dirty_list(root);
3411 extent_buffer_get(c);
3412 path->nodes[level] = c;
3413 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3414 path->slots[level] = 0;
3419 * worker function to insert a single pointer in a node.
3420 * the node should have enough room for the pointer already
3422 * slot and level indicate where you want the key to go, and
3423 * blocknr is the block the key points to.
3425 static void insert_ptr(struct btrfs_trans_handle *trans,
3426 struct btrfs_fs_info *fs_info, struct btrfs_path *path,
3427 struct btrfs_disk_key *key, u64 bytenr,
3428 int slot, int level)
3430 struct extent_buffer *lower;
3434 BUG_ON(!path->nodes[level]);
3435 btrfs_assert_tree_locked(path->nodes[level]);
3436 lower = path->nodes[level];
3437 nritems = btrfs_header_nritems(lower);
3438 BUG_ON(slot > nritems);
3439 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info));
3440 if (slot != nritems) {
3442 ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3446 memmove_extent_buffer(lower,
3447 btrfs_node_key_ptr_offset(slot + 1),
3448 btrfs_node_key_ptr_offset(slot),
3449 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3452 ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3456 btrfs_set_node_key(lower, key, slot);
3457 btrfs_set_node_blockptr(lower, slot, bytenr);
3458 WARN_ON(trans->transid == 0);
3459 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3460 btrfs_set_header_nritems(lower, nritems + 1);
3461 btrfs_mark_buffer_dirty(lower);
3465 * split the node at the specified level in path in two.
3466 * The path is corrected to point to the appropriate node after the split
3468 * Before splitting this tries to make some room in the node by pushing
3469 * left and right, if either one works, it returns right away.
3471 * returns 0 on success and < 0 on failure
3473 static noinline int split_node(struct btrfs_trans_handle *trans,
3474 struct btrfs_root *root,
3475 struct btrfs_path *path, int level)
3477 struct btrfs_fs_info *fs_info = root->fs_info;
3478 struct extent_buffer *c;
3479 struct extent_buffer *split;
3480 struct btrfs_disk_key disk_key;
3485 c = path->nodes[level];
3486 WARN_ON(btrfs_header_generation(c) != trans->transid);
3487 if (c == root->node) {
3489 * trying to split the root, lets make a new one
3491 * tree mod log: We don't log_removal old root in
3492 * insert_new_root, because that root buffer will be kept as a
3493 * normal node. We are going to log removal of half of the
3494 * elements below with tree_mod_log_eb_copy. We're holding a
3495 * tree lock on the buffer, which is why we cannot race with
3496 * other tree_mod_log users.
3498 ret = insert_new_root(trans, root, path, level + 1);
3502 ret = push_nodes_for_insert(trans, root, path, level);
3503 c = path->nodes[level];
3504 if (!ret && btrfs_header_nritems(c) <
3505 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3511 c_nritems = btrfs_header_nritems(c);
3512 mid = (c_nritems + 1) / 2;
3513 btrfs_node_key(c, &disk_key, mid);
3515 split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level,
3518 return PTR_ERR(split);
3520 root_add_used(root, fs_info->nodesize);
3521 ASSERT(btrfs_header_level(c) == level);
3523 ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid);
3525 btrfs_abort_transaction(trans, ret);
3528 copy_extent_buffer(split, c,
3529 btrfs_node_key_ptr_offset(0),
3530 btrfs_node_key_ptr_offset(mid),
3531 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3532 btrfs_set_header_nritems(split, c_nritems - mid);
3533 btrfs_set_header_nritems(c, mid);
3536 btrfs_mark_buffer_dirty(c);
3537 btrfs_mark_buffer_dirty(split);
3539 insert_ptr(trans, fs_info, path, &disk_key, split->start,
3540 path->slots[level + 1] + 1, level + 1);
3542 if (path->slots[level] >= mid) {
3543 path->slots[level] -= mid;
3544 btrfs_tree_unlock(c);
3545 free_extent_buffer(c);
3546 path->nodes[level] = split;
3547 path->slots[level + 1] += 1;
3549 btrfs_tree_unlock(split);
3550 free_extent_buffer(split);
3556 * how many bytes are required to store the items in a leaf. start
3557 * and nr indicate which items in the leaf to check. This totals up the
3558 * space used both by the item structs and the item data
3560 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3562 struct btrfs_item *start_item;
3563 struct btrfs_item *end_item;
3564 struct btrfs_map_token token;
3566 int nritems = btrfs_header_nritems(l);
3567 int end = min(nritems, start + nr) - 1;
3571 btrfs_init_map_token(&token);
3572 start_item = btrfs_item_nr(start);
3573 end_item = btrfs_item_nr(end);
3574 data_len = btrfs_token_item_offset(l, start_item, &token) +
3575 btrfs_token_item_size(l, start_item, &token);
3576 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3577 data_len += sizeof(struct btrfs_item) * nr;
3578 WARN_ON(data_len < 0);
3583 * The space between the end of the leaf items and
3584 * the start of the leaf data. IOW, how much room
3585 * the leaf has left for both items and data
3587 noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info,
3588 struct extent_buffer *leaf)
3590 int nritems = btrfs_header_nritems(leaf);
3593 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3596 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3598 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3599 leaf_space_used(leaf, 0, nritems), nritems);
3605 * min slot controls the lowest index we're willing to push to the
3606 * right. We'll push up to and including min_slot, but no lower
3608 static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info,
3609 struct btrfs_path *path,
3610 int data_size, int empty,
3611 struct extent_buffer *right,
3612 int free_space, u32 left_nritems,
3615 struct extent_buffer *left = path->nodes[0];
3616 struct extent_buffer *upper = path->nodes[1];
3617 struct btrfs_map_token token;
3618 struct btrfs_disk_key disk_key;
3623 struct btrfs_item *item;
3629 btrfs_init_map_token(&token);
3634 nr = max_t(u32, 1, min_slot);
3636 if (path->slots[0] >= left_nritems)
3637 push_space += data_size;
3639 slot = path->slots[1];
3640 i = left_nritems - 1;
3642 item = btrfs_item_nr(i);
3644 if (!empty && push_items > 0) {
3645 if (path->slots[0] > i)
3647 if (path->slots[0] == i) {
3648 int space = btrfs_leaf_free_space(fs_info, left);
3649 if (space + push_space * 2 > free_space)
3654 if (path->slots[0] == i)
3655 push_space += data_size;
3657 this_item_size = btrfs_item_size(left, item);
3658 if (this_item_size + sizeof(*item) + push_space > free_space)
3662 push_space += this_item_size + sizeof(*item);
3668 if (push_items == 0)
3671 WARN_ON(!empty && push_items == left_nritems);
3673 /* push left to right */
3674 right_nritems = btrfs_header_nritems(right);
3676 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3677 push_space -= leaf_data_end(fs_info, left);
3679 /* make room in the right data area */
3680 data_end = leaf_data_end(fs_info, right);
3681 memmove_extent_buffer(right,
3682 BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3683 BTRFS_LEAF_DATA_OFFSET + data_end,
3684 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3686 /* copy from the left data area */
3687 copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3688 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3689 BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left),
3692 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3693 btrfs_item_nr_offset(0),
3694 right_nritems * sizeof(struct btrfs_item));
3696 /* copy the items from left to right */
3697 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3698 btrfs_item_nr_offset(left_nritems - push_items),
3699 push_items * sizeof(struct btrfs_item));
3701 /* update the item pointers */
3702 right_nritems += push_items;
3703 btrfs_set_header_nritems(right, right_nritems);
3704 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3705 for (i = 0; i < right_nritems; i++) {
3706 item = btrfs_item_nr(i);
3707 push_space -= btrfs_token_item_size(right, item, &token);
3708 btrfs_set_token_item_offset(right, item, push_space, &token);
3711 left_nritems -= push_items;
3712 btrfs_set_header_nritems(left, left_nritems);
3715 btrfs_mark_buffer_dirty(left);
3717 clean_tree_block(fs_info, left);
3719 btrfs_mark_buffer_dirty(right);
3721 btrfs_item_key(right, &disk_key, 0);
3722 btrfs_set_node_key(upper, &disk_key, slot + 1);
3723 btrfs_mark_buffer_dirty(upper);
3725 /* then fixup the leaf pointer in the path */
3726 if (path->slots[0] >= left_nritems) {
3727 path->slots[0] -= left_nritems;
3728 if (btrfs_header_nritems(path->nodes[0]) == 0)
3729 clean_tree_block(fs_info, path->nodes[0]);
3730 btrfs_tree_unlock(path->nodes[0]);
3731 free_extent_buffer(path->nodes[0]);
3732 path->nodes[0] = right;
3733 path->slots[1] += 1;
3735 btrfs_tree_unlock(right);
3736 free_extent_buffer(right);
3741 btrfs_tree_unlock(right);
3742 free_extent_buffer(right);
3747 * push some data in the path leaf to the right, trying to free up at
3748 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3750 * returns 1 if the push failed because the other node didn't have enough
3751 * room, 0 if everything worked out and < 0 if there were major errors.
3753 * this will push starting from min_slot to the end of the leaf. It won't
3754 * push any slot lower than min_slot
3756 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3757 *root, struct btrfs_path *path,
3758 int min_data_size, int data_size,
3759 int empty, u32 min_slot)
3761 struct btrfs_fs_info *fs_info = root->fs_info;
3762 struct extent_buffer *left = path->nodes[0];
3763 struct extent_buffer *right;
3764 struct extent_buffer *upper;
3770 if (!path->nodes[1])
3773 slot = path->slots[1];
3774 upper = path->nodes[1];
3775 if (slot >= btrfs_header_nritems(upper) - 1)
3778 btrfs_assert_tree_locked(path->nodes[1]);
3780 right = read_node_slot(fs_info, upper, slot + 1);
3782 * slot + 1 is not valid or we fail to read the right node,
3783 * no big deal, just return.
3788 btrfs_tree_lock(right);
3789 btrfs_set_lock_blocking_write(right);
3791 free_space = btrfs_leaf_free_space(fs_info, right);
3792 if (free_space < data_size)
3795 /* cow and double check */
3796 ret = btrfs_cow_block(trans, root, right, upper,
3801 free_space = btrfs_leaf_free_space(fs_info, right);
3802 if (free_space < data_size)
3805 left_nritems = btrfs_header_nritems(left);
3806 if (left_nritems == 0)
3809 if (path->slots[0] == left_nritems && !empty) {
3810 /* Key greater than all keys in the leaf, right neighbor has
3811 * enough room for it and we're not emptying our leaf to delete
3812 * it, therefore use right neighbor to insert the new item and
3813 * no need to touch/dirty our left leaf. */
3814 btrfs_tree_unlock(left);
3815 free_extent_buffer(left);
3816 path->nodes[0] = right;
3822 return __push_leaf_right(fs_info, path, min_data_size, empty,
3823 right, free_space, left_nritems, min_slot);
3825 btrfs_tree_unlock(right);
3826 free_extent_buffer(right);
3831 * push some data in the path leaf to the left, trying to free up at
3832 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3834 * max_slot can put a limit on how far into the leaf we'll push items. The
3835 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3838 static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info,
3839 struct btrfs_path *path, int data_size,
3840 int empty, struct extent_buffer *left,
3841 int free_space, u32 right_nritems,
3844 struct btrfs_disk_key disk_key;
3845 struct extent_buffer *right = path->nodes[0];
3849 struct btrfs_item *item;
3850 u32 old_left_nritems;
3854 u32 old_left_item_size;
3855 struct btrfs_map_token token;
3857 btrfs_init_map_token(&token);
3860 nr = min(right_nritems, max_slot);
3862 nr = min(right_nritems - 1, max_slot);
3864 for (i = 0; i < nr; i++) {
3865 item = btrfs_item_nr(i);
3867 if (!empty && push_items > 0) {
3868 if (path->slots[0] < i)
3870 if (path->slots[0] == i) {
3871 int space = btrfs_leaf_free_space(fs_info, right);
3872 if (space + push_space * 2 > free_space)
3877 if (path->slots[0] == i)
3878 push_space += data_size;
3880 this_item_size = btrfs_item_size(right, item);
3881 if (this_item_size + sizeof(*item) + push_space > free_space)
3885 push_space += this_item_size + sizeof(*item);
3888 if (push_items == 0) {
3892 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3894 /* push data from right to left */
3895 copy_extent_buffer(left, right,
3896 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3897 btrfs_item_nr_offset(0),
3898 push_items * sizeof(struct btrfs_item));
3900 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3901 btrfs_item_offset_nr(right, push_items - 1);
3903 copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3904 leaf_data_end(fs_info, left) - push_space,
3905 BTRFS_LEAF_DATA_OFFSET +
3906 btrfs_item_offset_nr(right, push_items - 1),
3908 old_left_nritems = btrfs_header_nritems(left);
3909 BUG_ON(old_left_nritems <= 0);
3911 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3912 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3915 item = btrfs_item_nr(i);
3917 ioff = btrfs_token_item_offset(left, item, &token);
3918 btrfs_set_token_item_offset(left, item,
3919 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3922 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3924 /* fixup right node */
3925 if (push_items > right_nritems)
3926 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3929 if (push_items < right_nritems) {
3930 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3931 leaf_data_end(fs_info, right);
3932 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3933 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3934 BTRFS_LEAF_DATA_OFFSET +
3935 leaf_data_end(fs_info, right), push_space);
3937 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3938 btrfs_item_nr_offset(push_items),
3939 (btrfs_header_nritems(right) - push_items) *
3940 sizeof(struct btrfs_item));
3942 right_nritems -= push_items;
3943 btrfs_set_header_nritems(right, right_nritems);
3944 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3945 for (i = 0; i < right_nritems; i++) {
3946 item = btrfs_item_nr(i);
3948 push_space = push_space - btrfs_token_item_size(right,
3950 btrfs_set_token_item_offset(right, item, push_space, &token);
3953 btrfs_mark_buffer_dirty(left);
3955 btrfs_mark_buffer_dirty(right);
3957 clean_tree_block(fs_info, right);
3959 btrfs_item_key(right, &disk_key, 0);
3960 fixup_low_keys(path, &disk_key, 1);
3962 /* then fixup the leaf pointer in the path */
3963 if (path->slots[0] < push_items) {
3964 path->slots[0] += old_left_nritems;
3965 btrfs_tree_unlock(path->nodes[0]);
3966 free_extent_buffer(path->nodes[0]);
3967 path->nodes[0] = left;
3968 path->slots[1] -= 1;
3970 btrfs_tree_unlock(left);
3971 free_extent_buffer(left);
3972 path->slots[0] -= push_items;
3974 BUG_ON(path->slots[0] < 0);
3977 btrfs_tree_unlock(left);
3978 free_extent_buffer(left);
3983 * push some data in the path leaf to the left, trying to free up at
3984 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3986 * max_slot can put a limit on how far into the leaf we'll push items. The
3987 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3990 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3991 *root, struct btrfs_path *path, int min_data_size,
3992 int data_size, int empty, u32 max_slot)
3994 struct btrfs_fs_info *fs_info = root->fs_info;
3995 struct extent_buffer *right = path->nodes[0];
3996 struct extent_buffer *left;
4002 slot = path->slots[1];
4005 if (!path->nodes[1])
4008 right_nritems = btrfs_header_nritems(right);
4009 if (right_nritems == 0)
4012 btrfs_assert_tree_locked(path->nodes[1]);
4014 left = read_node_slot(fs_info, path->nodes[1], slot - 1);
4016 * slot - 1 is not valid or we fail to read the left node,
4017 * no big deal, just return.
4022 btrfs_tree_lock(left);
4023 btrfs_set_lock_blocking_write(left);
4025 free_space = btrfs_leaf_free_space(fs_info, left);
4026 if (free_space < data_size) {
4031 /* cow and double check */
4032 ret = btrfs_cow_block(trans, root, left,
4033 path->nodes[1], slot - 1, &left);
4035 /* we hit -ENOSPC, but it isn't fatal here */
4041 free_space = btrfs_leaf_free_space(fs_info, left);
4042 if (free_space < data_size) {
4047 return __push_leaf_left(fs_info, path, min_data_size,
4048 empty, left, free_space, right_nritems,
4051 btrfs_tree_unlock(left);
4052 free_extent_buffer(left);
4057 * split the path's leaf in two, making sure there is at least data_size
4058 * available for the resulting leaf level of the path.
4060 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4061 struct btrfs_fs_info *fs_info,
4062 struct btrfs_path *path,
4063 struct extent_buffer *l,
4064 struct extent_buffer *right,
4065 int slot, int mid, int nritems)
4070 struct btrfs_disk_key disk_key;
4071 struct btrfs_map_token token;
4073 btrfs_init_map_token(&token);
4075 nritems = nritems - mid;
4076 btrfs_set_header_nritems(right, nritems);
4077 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l);
4079 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4080 btrfs_item_nr_offset(mid),
4081 nritems * sizeof(struct btrfs_item));
4083 copy_extent_buffer(right, l,
4084 BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4085 data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4086 leaf_data_end(fs_info, l), data_copy_size);
4088 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4090 for (i = 0; i < nritems; i++) {
4091 struct btrfs_item *item = btrfs_item_nr(i);
4094 ioff = btrfs_token_item_offset(right, item, &token);
4095 btrfs_set_token_item_offset(right, item,
4096 ioff + rt_data_off, &token);
4099 btrfs_set_header_nritems(l, mid);
4100 btrfs_item_key(right, &disk_key, 0);
4101 insert_ptr(trans, fs_info, path, &disk_key, right->start,
4102 path->slots[1] + 1, 1);
4104 btrfs_mark_buffer_dirty(right);
4105 btrfs_mark_buffer_dirty(l);
4106 BUG_ON(path->slots[0] != slot);
4109 btrfs_tree_unlock(path->nodes[0]);
4110 free_extent_buffer(path->nodes[0]);
4111 path->nodes[0] = right;
4112 path->slots[0] -= mid;
4113 path->slots[1] += 1;
4115 btrfs_tree_unlock(right);
4116 free_extent_buffer(right);
4119 BUG_ON(path->slots[0] < 0);
4123 * double splits happen when we need to insert a big item in the middle
4124 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4125 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4128 * We avoid this by trying to push the items on either side of our target
4129 * into the adjacent leaves. If all goes well we can avoid the double split
4132 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4133 struct btrfs_root *root,
4134 struct btrfs_path *path,
4137 struct btrfs_fs_info *fs_info = root->fs_info;
4142 int space_needed = data_size;
4144 slot = path->slots[0];
4145 if (slot < btrfs_header_nritems(path->nodes[0]))
4146 space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4149 * try to push all the items after our slot into the
4152 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4159 nritems = btrfs_header_nritems(path->nodes[0]);
4161 * our goal is to get our slot at the start or end of a leaf. If
4162 * we've done so we're done
4164 if (path->slots[0] == 0 || path->slots[0] == nritems)
4167 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4170 /* try to push all the items before our slot into the next leaf */
4171 slot = path->slots[0];
4172 space_needed = data_size;
4174 space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4175 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4188 * split the path's leaf in two, making sure there is at least data_size
4189 * available for the resulting leaf level of the path.
4191 * returns 0 if all went well and < 0 on failure.
4193 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4194 struct btrfs_root *root,
4195 const struct btrfs_key *ins_key,
4196 struct btrfs_path *path, int data_size,
4199 struct btrfs_disk_key disk_key;
4200 struct extent_buffer *l;
4204 struct extent_buffer *right;
4205 struct btrfs_fs_info *fs_info = root->fs_info;
4209 int num_doubles = 0;
4210 int tried_avoid_double = 0;
4213 slot = path->slots[0];
4214 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4215 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4218 /* first try to make some room by pushing left and right */
4219 if (data_size && path->nodes[1]) {
4220 int space_needed = data_size;
4222 if (slot < btrfs_header_nritems(l))
4223 space_needed -= btrfs_leaf_free_space(fs_info, l);
4225 wret = push_leaf_right(trans, root, path, space_needed,
4226 space_needed, 0, 0);
4230 space_needed = data_size;
4232 space_needed -= btrfs_leaf_free_space(fs_info,
4234 wret = push_leaf_left(trans, root, path, space_needed,
4235 space_needed, 0, (u32)-1);
4241 /* did the pushes work? */
4242 if (btrfs_leaf_free_space(fs_info, l) >= data_size)
4246 if (!path->nodes[1]) {
4247 ret = insert_new_root(trans, root, path, 1);
4254 slot = path->slots[0];
4255 nritems = btrfs_header_nritems(l);
4256 mid = (nritems + 1) / 2;
4260 leaf_space_used(l, mid, nritems - mid) + data_size >
4261 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4262 if (slot >= nritems) {
4266 if (mid != nritems &&
4267 leaf_space_used(l, mid, nritems - mid) +
4268 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4269 if (data_size && !tried_avoid_double)
4270 goto push_for_double;
4276 if (leaf_space_used(l, 0, mid) + data_size >
4277 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4278 if (!extend && data_size && slot == 0) {
4280 } else if ((extend || !data_size) && slot == 0) {
4284 if (mid != nritems &&
4285 leaf_space_used(l, mid, nritems - mid) +
4286 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4287 if (data_size && !tried_avoid_double)
4288 goto push_for_double;
4296 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4298 btrfs_item_key(l, &disk_key, mid);
4300 right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0,
4303 return PTR_ERR(right);
4305 root_add_used(root, fs_info->nodesize);
4309 btrfs_set_header_nritems(right, 0);
4310 insert_ptr(trans, fs_info, path, &disk_key,
4311 right->start, path->slots[1] + 1, 1);
4312 btrfs_tree_unlock(path->nodes[0]);
4313 free_extent_buffer(path->nodes[0]);
4314 path->nodes[0] = right;
4316 path->slots[1] += 1;
4318 btrfs_set_header_nritems(right, 0);
4319 insert_ptr(trans, fs_info, path, &disk_key,
4320 right->start, path->slots[1], 1);
4321 btrfs_tree_unlock(path->nodes[0]);
4322 free_extent_buffer(path->nodes[0]);
4323 path->nodes[0] = right;
4325 if (path->slots[1] == 0)
4326 fixup_low_keys(path, &disk_key, 1);
4329 * We create a new leaf 'right' for the required ins_len and
4330 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4331 * the content of ins_len to 'right'.
4336 copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems);
4339 BUG_ON(num_doubles != 0);
4347 push_for_double_split(trans, root, path, data_size);
4348 tried_avoid_double = 1;
4349 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4354 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4355 struct btrfs_root *root,
4356 struct btrfs_path *path, int ins_len)
4358 struct btrfs_fs_info *fs_info = root->fs_info;
4359 struct btrfs_key key;
4360 struct extent_buffer *leaf;
4361 struct btrfs_file_extent_item *fi;
4366 leaf = path->nodes[0];
4367 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4369 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4370 key.type != BTRFS_EXTENT_CSUM_KEY);
4372 if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len)
4375 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4376 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4377 fi = btrfs_item_ptr(leaf, path->slots[0],
4378 struct btrfs_file_extent_item);
4379 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4381 btrfs_release_path(path);
4383 path->keep_locks = 1;
4384 path->search_for_split = 1;
4385 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4386 path->search_for_split = 0;
4393 leaf = path->nodes[0];
4394 /* if our item isn't there, return now */
4395 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4398 /* the leaf has changed, it now has room. return now */
4399 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len)
4402 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4403 fi = btrfs_item_ptr(leaf, path->slots[0],
4404 struct btrfs_file_extent_item);
4405 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4409 btrfs_set_path_blocking(path);
4410 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4414 path->keep_locks = 0;
4415 btrfs_unlock_up_safe(path, 1);
4418 path->keep_locks = 0;
4422 static noinline int split_item(struct btrfs_fs_info *fs_info,
4423 struct btrfs_path *path,
4424 const struct btrfs_key *new_key,
4425 unsigned long split_offset)
4427 struct extent_buffer *leaf;
4428 struct btrfs_item *item;
4429 struct btrfs_item *new_item;
4435 struct btrfs_disk_key disk_key;
4437 leaf = path->nodes[0];
4438 BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item));
4440 btrfs_set_path_blocking(path);
4442 item = btrfs_item_nr(path->slots[0]);
4443 orig_offset = btrfs_item_offset(leaf, item);
4444 item_size = btrfs_item_size(leaf, item);
4446 buf = kmalloc(item_size, GFP_NOFS);
4450 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4451 path->slots[0]), item_size);
4453 slot = path->slots[0] + 1;
4454 nritems = btrfs_header_nritems(leaf);
4455 if (slot != nritems) {
4456 /* shift the items */
4457 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4458 btrfs_item_nr_offset(slot),
4459 (nritems - slot) * sizeof(struct btrfs_item));
4462 btrfs_cpu_key_to_disk(&disk_key, new_key);
4463 btrfs_set_item_key(leaf, &disk_key, slot);
4465 new_item = btrfs_item_nr(slot);
4467 btrfs_set_item_offset(leaf, new_item, orig_offset);
4468 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4470 btrfs_set_item_offset(leaf, item,
4471 orig_offset + item_size - split_offset);
4472 btrfs_set_item_size(leaf, item, split_offset);
4474 btrfs_set_header_nritems(leaf, nritems + 1);
4476 /* write the data for the start of the original item */
4477 write_extent_buffer(leaf, buf,
4478 btrfs_item_ptr_offset(leaf, path->slots[0]),
4481 /* write the data for the new item */
4482 write_extent_buffer(leaf, buf + split_offset,
4483 btrfs_item_ptr_offset(leaf, slot),
4484 item_size - split_offset);
4485 btrfs_mark_buffer_dirty(leaf);
4487 BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0);
4493 * This function splits a single item into two items,
4494 * giving 'new_key' to the new item and splitting the
4495 * old one at split_offset (from the start of the item).
4497 * The path may be released by this operation. After
4498 * the split, the path is pointing to the old item. The
4499 * new item is going to be in the same node as the old one.
4501 * Note, the item being split must be smaller enough to live alone on
4502 * a tree block with room for one extra struct btrfs_item
4504 * This allows us to split the item in place, keeping a lock on the
4505 * leaf the entire time.
4507 int btrfs_split_item(struct btrfs_trans_handle *trans,
4508 struct btrfs_root *root,
4509 struct btrfs_path *path,
4510 const struct btrfs_key *new_key,
4511 unsigned long split_offset)
4514 ret = setup_leaf_for_split(trans, root, path,
4515 sizeof(struct btrfs_item));
4519 ret = split_item(root->fs_info, path, new_key, split_offset);
4524 * This function duplicate a item, giving 'new_key' to the new item.
4525 * It guarantees both items live in the same tree leaf and the new item
4526 * is contiguous with the original item.
4528 * This allows us to split file extent in place, keeping a lock on the
4529 * leaf the entire time.
4531 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4532 struct btrfs_root *root,
4533 struct btrfs_path *path,
4534 const struct btrfs_key *new_key)
4536 struct extent_buffer *leaf;
4540 leaf = path->nodes[0];
4541 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4542 ret = setup_leaf_for_split(trans, root, path,
4543 item_size + sizeof(struct btrfs_item));
4548 setup_items_for_insert(root, path, new_key, &item_size,
4549 item_size, item_size +
4550 sizeof(struct btrfs_item), 1);
4551 leaf = path->nodes[0];
4552 memcpy_extent_buffer(leaf,
4553 btrfs_item_ptr_offset(leaf, path->slots[0]),
4554 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4560 * make the item pointed to by the path smaller. new_size indicates
4561 * how small to make it, and from_end tells us if we just chop bytes
4562 * off the end of the item or if we shift the item to chop bytes off
4565 void btrfs_truncate_item(struct btrfs_fs_info *fs_info,
4566 struct btrfs_path *path, u32 new_size, int from_end)
4569 struct extent_buffer *leaf;
4570 struct btrfs_item *item;
4572 unsigned int data_end;
4573 unsigned int old_data_start;
4574 unsigned int old_size;
4575 unsigned int size_diff;
4577 struct btrfs_map_token token;
4579 btrfs_init_map_token(&token);
4581 leaf = path->nodes[0];
4582 slot = path->slots[0];
4584 old_size = btrfs_item_size_nr(leaf, slot);
4585 if (old_size == new_size)
4588 nritems = btrfs_header_nritems(leaf);
4589 data_end = leaf_data_end(fs_info, leaf);
4591 old_data_start = btrfs_item_offset_nr(leaf, slot);
4593 size_diff = old_size - new_size;
4596 BUG_ON(slot >= nritems);
4599 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4601 /* first correct the data pointers */
4602 for (i = slot; i < nritems; i++) {
4604 item = btrfs_item_nr(i);
4606 ioff = btrfs_token_item_offset(leaf, item, &token);
4607 btrfs_set_token_item_offset(leaf, item,
4608 ioff + size_diff, &token);
4611 /* shift the data */
4613 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4614 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4615 data_end, old_data_start + new_size - data_end);
4617 struct btrfs_disk_key disk_key;
4620 btrfs_item_key(leaf, &disk_key, slot);
4622 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4624 struct btrfs_file_extent_item *fi;
4626 fi = btrfs_item_ptr(leaf, slot,
4627 struct btrfs_file_extent_item);
4628 fi = (struct btrfs_file_extent_item *)(
4629 (unsigned long)fi - size_diff);
4631 if (btrfs_file_extent_type(leaf, fi) ==
4632 BTRFS_FILE_EXTENT_INLINE) {
4633 ptr = btrfs_item_ptr_offset(leaf, slot);
4634 memmove_extent_buffer(leaf, ptr,
4636 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4640 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4641 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4642 data_end, old_data_start - data_end);
4644 offset = btrfs_disk_key_offset(&disk_key);
4645 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4646 btrfs_set_item_key(leaf, &disk_key, slot);
4648 fixup_low_keys(path, &disk_key, 1);
4651 item = btrfs_item_nr(slot);
4652 btrfs_set_item_size(leaf, item, new_size);
4653 btrfs_mark_buffer_dirty(leaf);
4655 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4656 btrfs_print_leaf(leaf);
4662 * make the item pointed to by the path bigger, data_size is the added size.
4664 void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
4668 struct extent_buffer *leaf;
4669 struct btrfs_item *item;
4671 unsigned int data_end;
4672 unsigned int old_data;
4673 unsigned int old_size;
4675 struct btrfs_map_token token;
4677 btrfs_init_map_token(&token);
4679 leaf = path->nodes[0];
4681 nritems = btrfs_header_nritems(leaf);
4682 data_end = leaf_data_end(fs_info, leaf);
4684 if (btrfs_leaf_free_space(fs_info, leaf) < data_size) {
4685 btrfs_print_leaf(leaf);
4688 slot = path->slots[0];
4689 old_data = btrfs_item_end_nr(leaf, slot);
4692 if (slot >= nritems) {
4693 btrfs_print_leaf(leaf);
4694 btrfs_crit(fs_info, "slot %d too large, nritems %d",
4700 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4702 /* first correct the data pointers */
4703 for (i = slot; i < nritems; i++) {
4705 item = btrfs_item_nr(i);
4707 ioff = btrfs_token_item_offset(leaf, item, &token);
4708 btrfs_set_token_item_offset(leaf, item,
4709 ioff - data_size, &token);
4712 /* shift the data */
4713 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4714 data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4715 data_end, old_data - data_end);
4717 data_end = old_data;
4718 old_size = btrfs_item_size_nr(leaf, slot);
4719 item = btrfs_item_nr(slot);
4720 btrfs_set_item_size(leaf, item, old_size + data_size);
4721 btrfs_mark_buffer_dirty(leaf);
4723 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4724 btrfs_print_leaf(leaf);
4730 * this is a helper for btrfs_insert_empty_items, the main goal here is
4731 * to save stack depth by doing the bulk of the work in a function
4732 * that doesn't call btrfs_search_slot
4734 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4735 const struct btrfs_key *cpu_key, u32 *data_size,
4736 u32 total_data, u32 total_size, int nr)
4738 struct btrfs_fs_info *fs_info = root->fs_info;
4739 struct btrfs_item *item;
4742 unsigned int data_end;
4743 struct btrfs_disk_key disk_key;
4744 struct extent_buffer *leaf;
4746 struct btrfs_map_token token;
4748 if (path->slots[0] == 0) {
4749 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4750 fixup_low_keys(path, &disk_key, 1);
4752 btrfs_unlock_up_safe(path, 1);
4754 btrfs_init_map_token(&token);
4756 leaf = path->nodes[0];
4757 slot = path->slots[0];
4759 nritems = btrfs_header_nritems(leaf);
4760 data_end = leaf_data_end(fs_info, leaf);
4762 if (btrfs_leaf_free_space(fs_info, leaf) < total_size) {
4763 btrfs_print_leaf(leaf);
4764 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4765 total_size, btrfs_leaf_free_space(fs_info, leaf));
4769 if (slot != nritems) {
4770 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4772 if (old_data < data_end) {
4773 btrfs_print_leaf(leaf);
4774 btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4775 slot, old_data, data_end);
4779 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4781 /* first correct the data pointers */
4782 for (i = slot; i < nritems; i++) {
4785 item = btrfs_item_nr(i);
4786 ioff = btrfs_token_item_offset(leaf, item, &token);
4787 btrfs_set_token_item_offset(leaf, item,
4788 ioff - total_data, &token);
4790 /* shift the items */
4791 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4792 btrfs_item_nr_offset(slot),
4793 (nritems - slot) * sizeof(struct btrfs_item));
4795 /* shift the data */
4796 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4797 data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4798 data_end, old_data - data_end);
4799 data_end = old_data;
4802 /* setup the item for the new data */
4803 for (i = 0; i < nr; i++) {
4804 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4805 btrfs_set_item_key(leaf, &disk_key, slot + i);
4806 item = btrfs_item_nr(slot + i);
4807 btrfs_set_token_item_offset(leaf, item,
4808 data_end - data_size[i], &token);
4809 data_end -= data_size[i];
4810 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4813 btrfs_set_header_nritems(leaf, nritems + nr);
4814 btrfs_mark_buffer_dirty(leaf);
4816 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4817 btrfs_print_leaf(leaf);
4823 * Given a key and some data, insert items into the tree.
4824 * This does all the path init required, making room in the tree if needed.
4826 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4827 struct btrfs_root *root,
4828 struct btrfs_path *path,
4829 const struct btrfs_key *cpu_key, u32 *data_size,
4838 for (i = 0; i < nr; i++)
4839 total_data += data_size[i];
4841 total_size = total_data + (nr * sizeof(struct btrfs_item));
4842 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4848 slot = path->slots[0];
4851 setup_items_for_insert(root, path, cpu_key, data_size,
4852 total_data, total_size, nr);
4857 * Given a key and some data, insert an item into the tree.
4858 * This does all the path init required, making room in the tree if needed.
4860 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4861 const struct btrfs_key *cpu_key, void *data,
4865 struct btrfs_path *path;
4866 struct extent_buffer *leaf;
4869 path = btrfs_alloc_path();
4872 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4874 leaf = path->nodes[0];
4875 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4876 write_extent_buffer(leaf, data, ptr, data_size);
4877 btrfs_mark_buffer_dirty(leaf);
4879 btrfs_free_path(path);
4884 * delete the pointer from a given node.
4886 * the tree should have been previously balanced so the deletion does not
4889 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4890 int level, int slot)
4892 struct extent_buffer *parent = path->nodes[level];
4896 nritems = btrfs_header_nritems(parent);
4897 if (slot != nritems - 1) {
4899 ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4900 nritems - slot - 1);
4903 memmove_extent_buffer(parent,
4904 btrfs_node_key_ptr_offset(slot),
4905 btrfs_node_key_ptr_offset(slot + 1),
4906 sizeof(struct btrfs_key_ptr) *
4907 (nritems - slot - 1));
4909 ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4915 btrfs_set_header_nritems(parent, nritems);
4916 if (nritems == 0 && parent == root->node) {
4917 BUG_ON(btrfs_header_level(root->node) != 1);
4918 /* just turn the root into a leaf and break */
4919 btrfs_set_header_level(root->node, 0);
4920 } else if (slot == 0) {
4921 struct btrfs_disk_key disk_key;
4923 btrfs_node_key(parent, &disk_key, 0);
4924 fixup_low_keys(path, &disk_key, level + 1);
4926 btrfs_mark_buffer_dirty(parent);
4930 * a helper function to delete the leaf pointed to by path->slots[1] and
4933 * This deletes the pointer in path->nodes[1] and frees the leaf
4934 * block extent. zero is returned if it all worked out, < 0 otherwise.
4936 * The path must have already been setup for deleting the leaf, including
4937 * all the proper balancing. path->nodes[1] must be locked.
4939 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4940 struct btrfs_root *root,
4941 struct btrfs_path *path,
4942 struct extent_buffer *leaf)
4944 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4945 del_ptr(root, path, 1, path->slots[1]);
4948 * btrfs_free_extent is expensive, we want to make sure we
4949 * aren't holding any locks when we call it
4951 btrfs_unlock_up_safe(path, 0);
4953 root_sub_used(root, leaf->len);
4955 extent_buffer_get(leaf);
4956 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4957 free_extent_buffer_stale(leaf);
4960 * delete the item at the leaf level in path. If that empties
4961 * the leaf, remove it from the tree
4963 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4964 struct btrfs_path *path, int slot, int nr)
4966 struct btrfs_fs_info *fs_info = root->fs_info;
4967 struct extent_buffer *leaf;
4968 struct btrfs_item *item;
4975 struct btrfs_map_token token;
4977 btrfs_init_map_token(&token);
4979 leaf = path->nodes[0];
4980 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4982 for (i = 0; i < nr; i++)
4983 dsize += btrfs_item_size_nr(leaf, slot + i);
4985 nritems = btrfs_header_nritems(leaf);
4987 if (slot + nr != nritems) {
4988 int data_end = leaf_data_end(fs_info, leaf);
4990 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4992 BTRFS_LEAF_DATA_OFFSET + data_end,
4993 last_off - data_end);
4995 for (i = slot + nr; i < nritems; i++) {
4998 item = btrfs_item_nr(i);
4999 ioff = btrfs_token_item_offset(leaf, item, &token);
5000 btrfs_set_token_item_offset(leaf, item,
5001 ioff + dsize, &token);
5004 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
5005 btrfs_item_nr_offset(slot + nr),
5006 sizeof(struct btrfs_item) *
5007 (nritems - slot - nr));
5009 btrfs_set_header_nritems(leaf, nritems - nr);
5012 /* delete the leaf if we've emptied it */
5014 if (leaf == root->node) {
5015 btrfs_set_header_level(leaf, 0);
5017 btrfs_set_path_blocking(path);
5018 clean_tree_block(fs_info, leaf);
5019 btrfs_del_leaf(trans, root, path, leaf);
5022 int used = leaf_space_used(leaf, 0, nritems);
5024 struct btrfs_disk_key disk_key;
5026 btrfs_item_key(leaf, &disk_key, 0);
5027 fixup_low_keys(path, &disk_key, 1);
5030 /* delete the leaf if it is mostly empty */
5031 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
5032 /* push_leaf_left fixes the path.
5033 * make sure the path still points to our leaf
5034 * for possible call to del_ptr below
5036 slot = path->slots[1];
5037 extent_buffer_get(leaf);
5039 btrfs_set_path_blocking(path);
5040 wret = push_leaf_left(trans, root, path, 1, 1,
5042 if (wret < 0 && wret != -ENOSPC)
5045 if (path->nodes[0] == leaf &&
5046 btrfs_header_nritems(leaf)) {
5047 wret = push_leaf_right(trans, root, path, 1,
5049 if (wret < 0 && wret != -ENOSPC)
5053 if (btrfs_header_nritems(leaf) == 0) {
5054 path->slots[1] = slot;
5055 btrfs_del_leaf(trans, root, path, leaf);
5056 free_extent_buffer(leaf);
5059 /* if we're still in the path, make sure
5060 * we're dirty. Otherwise, one of the
5061 * push_leaf functions must have already
5062 * dirtied this buffer
5064 if (path->nodes[0] == leaf)
5065 btrfs_mark_buffer_dirty(leaf);
5066 free_extent_buffer(leaf);
5069 btrfs_mark_buffer_dirty(leaf);
5076 * search the tree again to find a leaf with lesser keys
5077 * returns 0 if it found something or 1 if there are no lesser leaves.
5078 * returns < 0 on io errors.
5080 * This may release the path, and so you may lose any locks held at the
5083 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5085 struct btrfs_key key;
5086 struct btrfs_disk_key found_key;
5089 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5091 if (key.offset > 0) {
5093 } else if (key.type > 0) {
5095 key.offset = (u64)-1;
5096 } else if (key.objectid > 0) {
5099 key.offset = (u64)-1;
5104 btrfs_release_path(path);
5105 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5108 btrfs_item_key(path->nodes[0], &found_key, 0);
5109 ret = comp_keys(&found_key, &key);
5111 * We might have had an item with the previous key in the tree right
5112 * before we released our path. And after we released our path, that
5113 * item might have been pushed to the first slot (0) of the leaf we
5114 * were holding due to a tree balance. Alternatively, an item with the
5115 * previous key can exist as the only element of a leaf (big fat item).
5116 * Therefore account for these 2 cases, so that our callers (like
5117 * btrfs_previous_item) don't miss an existing item with a key matching
5118 * the previous key we computed above.
5126 * A helper function to walk down the tree starting at min_key, and looking
5127 * for nodes or leaves that are have a minimum transaction id.
5128 * This is used by the btree defrag code, and tree logging
5130 * This does not cow, but it does stuff the starting key it finds back
5131 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5132 * key and get a writable path.
5134 * This honors path->lowest_level to prevent descent past a given level
5137 * min_trans indicates the oldest transaction that you are interested
5138 * in walking through. Any nodes or leaves older than min_trans are
5139 * skipped over (without reading them).
5141 * returns zero if something useful was found, < 0 on error and 1 if there
5142 * was nothing in the tree that matched the search criteria.
5144 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5145 struct btrfs_path *path,
5148 struct btrfs_fs_info *fs_info = root->fs_info;
5149 struct extent_buffer *cur;
5150 struct btrfs_key found_key;
5156 int keep_locks = path->keep_locks;
5158 path->keep_locks = 1;
5160 cur = btrfs_read_lock_root_node(root);
5161 level = btrfs_header_level(cur);
5162 WARN_ON(path->nodes[level]);
5163 path->nodes[level] = cur;
5164 path->locks[level] = BTRFS_READ_LOCK;
5166 if (btrfs_header_generation(cur) < min_trans) {
5171 nritems = btrfs_header_nritems(cur);
5172 level = btrfs_header_level(cur);
5173 sret = btrfs_bin_search(cur, min_key, level, &slot);
5175 /* at the lowest level, we're done, setup the path and exit */
5176 if (level == path->lowest_level) {
5177 if (slot >= nritems)
5180 path->slots[level] = slot;
5181 btrfs_item_key_to_cpu(cur, &found_key, slot);
5184 if (sret && slot > 0)
5187 * check this node pointer against the min_trans parameters.
5188 * If it is too old, old, skip to the next one.
5190 while (slot < nritems) {
5193 gen = btrfs_node_ptr_generation(cur, slot);
5194 if (gen < min_trans) {
5202 * we didn't find a candidate key in this node, walk forward
5203 * and find another one
5205 if (slot >= nritems) {
5206 path->slots[level] = slot;
5207 btrfs_set_path_blocking(path);
5208 sret = btrfs_find_next_key(root, path, min_key, level,
5211 btrfs_release_path(path);
5217 /* save our key for returning back */
5218 btrfs_node_key_to_cpu(cur, &found_key, slot);
5219 path->slots[level] = slot;
5220 if (level == path->lowest_level) {
5224 btrfs_set_path_blocking(path);
5225 cur = read_node_slot(fs_info, cur, slot);
5231 btrfs_tree_read_lock(cur);
5233 path->locks[level - 1] = BTRFS_READ_LOCK;
5234 path->nodes[level - 1] = cur;
5235 unlock_up(path, level, 1, 0, NULL);
5238 path->keep_locks = keep_locks;
5240 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5241 btrfs_set_path_blocking(path);
5242 memcpy(min_key, &found_key, sizeof(found_key));
5247 static int tree_move_down(struct btrfs_fs_info *fs_info,
5248 struct btrfs_path *path,
5251 struct extent_buffer *eb;
5253 BUG_ON(*level == 0);
5254 eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]);
5258 path->nodes[*level - 1] = eb;
5259 path->slots[*level - 1] = 0;
5264 static int tree_move_next_or_upnext(struct btrfs_path *path,
5265 int *level, int root_level)
5269 nritems = btrfs_header_nritems(path->nodes[*level]);
5271 path->slots[*level]++;
5273 while (path->slots[*level] >= nritems) {
5274 if (*level == root_level)
5278 path->slots[*level] = 0;
5279 free_extent_buffer(path->nodes[*level]);
5280 path->nodes[*level] = NULL;
5282 path->slots[*level]++;
5284 nritems = btrfs_header_nritems(path->nodes[*level]);
5291 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5294 static int tree_advance(struct btrfs_fs_info *fs_info,
5295 struct btrfs_path *path,
5296 int *level, int root_level,
5298 struct btrfs_key *key)
5302 if (*level == 0 || !allow_down) {
5303 ret = tree_move_next_or_upnext(path, level, root_level);
5305 ret = tree_move_down(fs_info, path, level);
5309 btrfs_item_key_to_cpu(path->nodes[*level], key,
5310 path->slots[*level]);
5312 btrfs_node_key_to_cpu(path->nodes[*level], key,
5313 path->slots[*level]);
5318 static int tree_compare_item(struct btrfs_path *left_path,
5319 struct btrfs_path *right_path,
5324 unsigned long off1, off2;
5326 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5327 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5331 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5332 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5333 right_path->slots[0]);
5335 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5337 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5344 #define ADVANCE_ONLY_NEXT -1
5347 * This function compares two trees and calls the provided callback for
5348 * every changed/new/deleted item it finds.
5349 * If shared tree blocks are encountered, whole subtrees are skipped, making
5350 * the compare pretty fast on snapshotted subvolumes.
5352 * This currently works on commit roots only. As commit roots are read only,
5353 * we don't do any locking. The commit roots are protected with transactions.
5354 * Transactions are ended and rejoined when a commit is tried in between.
5356 * This function checks for modifications done to the trees while comparing.
5357 * If it detects a change, it aborts immediately.
5359 int btrfs_compare_trees(struct btrfs_root *left_root,
5360 struct btrfs_root *right_root,
5361 btrfs_changed_cb_t changed_cb, void *ctx)
5363 struct btrfs_fs_info *fs_info = left_root->fs_info;
5366 struct btrfs_path *left_path = NULL;
5367 struct btrfs_path *right_path = NULL;
5368 struct btrfs_key left_key;
5369 struct btrfs_key right_key;
5370 char *tmp_buf = NULL;
5371 int left_root_level;
5372 int right_root_level;
5375 int left_end_reached;
5376 int right_end_reached;
5384 left_path = btrfs_alloc_path();
5389 right_path = btrfs_alloc_path();
5395 tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
5401 left_path->search_commit_root = 1;
5402 left_path->skip_locking = 1;
5403 right_path->search_commit_root = 1;
5404 right_path->skip_locking = 1;
5407 * Strategy: Go to the first items of both trees. Then do
5409 * If both trees are at level 0
5410 * Compare keys of current items
5411 * If left < right treat left item as new, advance left tree
5413 * If left > right treat right item as deleted, advance right tree
5415 * If left == right do deep compare of items, treat as changed if
5416 * needed, advance both trees and repeat
5417 * If both trees are at the same level but not at level 0
5418 * Compare keys of current nodes/leafs
5419 * If left < right advance left tree and repeat
5420 * If left > right advance right tree and repeat
5421 * If left == right compare blockptrs of the next nodes/leafs
5422 * If they match advance both trees but stay at the same level
5424 * If they don't match advance both trees while allowing to go
5426 * If tree levels are different
5427 * Advance the tree that needs it and repeat
5429 * Advancing a tree means:
5430 * If we are at level 0, try to go to the next slot. If that's not
5431 * possible, go one level up and repeat. Stop when we found a level
5432 * where we could go to the next slot. We may at this point be on a
5435 * If we are not at level 0 and not on shared tree blocks, go one
5438 * If we are not at level 0 and on shared tree blocks, go one slot to
5439 * the right if possible or go up and right.
5442 down_read(&fs_info->commit_root_sem);
5443 left_level = btrfs_header_level(left_root->commit_root);
5444 left_root_level = left_level;
5445 left_path->nodes[left_level] =
5446 btrfs_clone_extent_buffer(left_root->commit_root);
5447 if (!left_path->nodes[left_level]) {
5448 up_read(&fs_info->commit_root_sem);
5453 right_level = btrfs_header_level(right_root->commit_root);
5454 right_root_level = right_level;
5455 right_path->nodes[right_level] =
5456 btrfs_clone_extent_buffer(right_root->commit_root);
5457 if (!right_path->nodes[right_level]) {
5458 up_read(&fs_info->commit_root_sem);
5462 up_read(&fs_info->commit_root_sem);
5464 if (left_level == 0)
5465 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5466 &left_key, left_path->slots[left_level]);
5468 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5469 &left_key, left_path->slots[left_level]);
5470 if (right_level == 0)
5471 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5472 &right_key, right_path->slots[right_level]);
5474 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5475 &right_key, right_path->slots[right_level]);
5477 left_end_reached = right_end_reached = 0;
5478 advance_left = advance_right = 0;
5481 if (advance_left && !left_end_reached) {
5482 ret = tree_advance(fs_info, left_path, &left_level,
5484 advance_left != ADVANCE_ONLY_NEXT,
5487 left_end_reached = ADVANCE;
5492 if (advance_right && !right_end_reached) {
5493 ret = tree_advance(fs_info, right_path, &right_level,
5495 advance_right != ADVANCE_ONLY_NEXT,
5498 right_end_reached = ADVANCE;
5504 if (left_end_reached && right_end_reached) {
5507 } else if (left_end_reached) {
5508 if (right_level == 0) {
5509 ret = changed_cb(left_path, right_path,
5511 BTRFS_COMPARE_TREE_DELETED,
5516 advance_right = ADVANCE;
5518 } else if (right_end_reached) {
5519 if (left_level == 0) {
5520 ret = changed_cb(left_path, right_path,
5522 BTRFS_COMPARE_TREE_NEW,
5527 advance_left = ADVANCE;
5531 if (left_level == 0 && right_level == 0) {
5532 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5534 ret = changed_cb(left_path, right_path,
5536 BTRFS_COMPARE_TREE_NEW,
5540 advance_left = ADVANCE;
5541 } else if (cmp > 0) {
5542 ret = changed_cb(left_path, right_path,
5544 BTRFS_COMPARE_TREE_DELETED,
5548 advance_right = ADVANCE;
5550 enum btrfs_compare_tree_result result;
5552 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5553 ret = tree_compare_item(left_path, right_path,
5556 result = BTRFS_COMPARE_TREE_CHANGED;
5558 result = BTRFS_COMPARE_TREE_SAME;
5559 ret = changed_cb(left_path, right_path,
5560 &left_key, result, ctx);
5563 advance_left = ADVANCE;
5564 advance_right = ADVANCE;
5566 } else if (left_level == right_level) {
5567 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5569 advance_left = ADVANCE;
5570 } else if (cmp > 0) {
5571 advance_right = ADVANCE;
5573 left_blockptr = btrfs_node_blockptr(
5574 left_path->nodes[left_level],
5575 left_path->slots[left_level]);
5576 right_blockptr = btrfs_node_blockptr(
5577 right_path->nodes[right_level],
5578 right_path->slots[right_level]);
5579 left_gen = btrfs_node_ptr_generation(
5580 left_path->nodes[left_level],
5581 left_path->slots[left_level]);
5582 right_gen = btrfs_node_ptr_generation(
5583 right_path->nodes[right_level],
5584 right_path->slots[right_level]);
5585 if (left_blockptr == right_blockptr &&
5586 left_gen == right_gen) {
5588 * As we're on a shared block, don't
5589 * allow to go deeper.
5591 advance_left = ADVANCE_ONLY_NEXT;
5592 advance_right = ADVANCE_ONLY_NEXT;
5594 advance_left = ADVANCE;
5595 advance_right = ADVANCE;
5598 } else if (left_level < right_level) {
5599 advance_right = ADVANCE;
5601 advance_left = ADVANCE;
5606 btrfs_free_path(left_path);
5607 btrfs_free_path(right_path);
5613 * this is similar to btrfs_next_leaf, but does not try to preserve
5614 * and fixup the path. It looks for and returns the next key in the
5615 * tree based on the current path and the min_trans parameters.
5617 * 0 is returned if another key is found, < 0 if there are any errors
5618 * and 1 is returned if there are no higher keys in the tree
5620 * path->keep_locks should be set to 1 on the search made before
5621 * calling this function.
5623 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5624 struct btrfs_key *key, int level, u64 min_trans)
5627 struct extent_buffer *c;
5629 WARN_ON(!path->keep_locks);
5630 while (level < BTRFS_MAX_LEVEL) {
5631 if (!path->nodes[level])
5634 slot = path->slots[level] + 1;
5635 c = path->nodes[level];
5637 if (slot >= btrfs_header_nritems(c)) {
5640 struct btrfs_key cur_key;
5641 if (level + 1 >= BTRFS_MAX_LEVEL ||
5642 !path->nodes[level + 1])
5645 if (path->locks[level + 1]) {
5650 slot = btrfs_header_nritems(c) - 1;
5652 btrfs_item_key_to_cpu(c, &cur_key, slot);
5654 btrfs_node_key_to_cpu(c, &cur_key, slot);
5656 orig_lowest = path->lowest_level;
5657 btrfs_release_path(path);
5658 path->lowest_level = level;
5659 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5661 path->lowest_level = orig_lowest;
5665 c = path->nodes[level];
5666 slot = path->slots[level];
5673 btrfs_item_key_to_cpu(c, key, slot);
5675 u64 gen = btrfs_node_ptr_generation(c, slot);
5677 if (gen < min_trans) {
5681 btrfs_node_key_to_cpu(c, key, slot);
5689 * search the tree again to find a leaf with greater keys
5690 * returns 0 if it found something or 1 if there are no greater leaves.
5691 * returns < 0 on io errors.
5693 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5695 return btrfs_next_old_leaf(root, path, 0);
5698 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5703 struct extent_buffer *c;
5704 struct extent_buffer *next;
5705 struct btrfs_key key;
5708 int old_spinning = path->leave_spinning;
5709 int next_rw_lock = 0;
5711 nritems = btrfs_header_nritems(path->nodes[0]);
5715 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5720 btrfs_release_path(path);
5722 path->keep_locks = 1;
5723 path->leave_spinning = 1;
5726 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5728 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5729 path->keep_locks = 0;
5734 nritems = btrfs_header_nritems(path->nodes[0]);
5736 * by releasing the path above we dropped all our locks. A balance
5737 * could have added more items next to the key that used to be
5738 * at the very end of the block. So, check again here and
5739 * advance the path if there are now more items available.
5741 if (nritems > 0 && path->slots[0] < nritems - 1) {
5748 * So the above check misses one case:
5749 * - after releasing the path above, someone has removed the item that
5750 * used to be at the very end of the block, and balance between leafs
5751 * gets another one with bigger key.offset to replace it.
5753 * This one should be returned as well, or we can get leaf corruption
5754 * later(esp. in __btrfs_drop_extents()).
5756 * And a bit more explanation about this check,
5757 * with ret > 0, the key isn't found, the path points to the slot
5758 * where it should be inserted, so the path->slots[0] item must be the
5761 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5766 while (level < BTRFS_MAX_LEVEL) {
5767 if (!path->nodes[level]) {
5772 slot = path->slots[level] + 1;
5773 c = path->nodes[level];
5774 if (slot >= btrfs_header_nritems(c)) {
5776 if (level == BTRFS_MAX_LEVEL) {
5784 btrfs_tree_unlock_rw(next, next_rw_lock);
5785 free_extent_buffer(next);
5789 next_rw_lock = path->locks[level];
5790 ret = read_block_for_search(root, path, &next, level,
5796 btrfs_release_path(path);
5800 if (!path->skip_locking) {
5801 ret = btrfs_try_tree_read_lock(next);
5802 if (!ret && time_seq) {
5804 * If we don't get the lock, we may be racing
5805 * with push_leaf_left, holding that lock while
5806 * itself waiting for the leaf we've currently
5807 * locked. To solve this situation, we give up
5808 * on our lock and cycle.
5810 free_extent_buffer(next);
5811 btrfs_release_path(path);
5816 btrfs_set_path_blocking(path);
5817 btrfs_tree_read_lock(next);
5819 next_rw_lock = BTRFS_READ_LOCK;
5823 path->slots[level] = slot;
5826 c = path->nodes[level];
5827 if (path->locks[level])
5828 btrfs_tree_unlock_rw(c, path->locks[level]);
5830 free_extent_buffer(c);
5831 path->nodes[level] = next;
5832 path->slots[level] = 0;
5833 if (!path->skip_locking)
5834 path->locks[level] = next_rw_lock;
5838 ret = read_block_for_search(root, path, &next, level,
5844 btrfs_release_path(path);
5848 if (!path->skip_locking) {
5849 ret = btrfs_try_tree_read_lock(next);
5851 btrfs_set_path_blocking(path);
5852 btrfs_tree_read_lock(next);
5854 next_rw_lock = BTRFS_READ_LOCK;
5859 unlock_up(path, 0, 1, 0, NULL);
5860 path->leave_spinning = old_spinning;
5862 btrfs_set_path_blocking(path);
5868 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5869 * searching until it gets past min_objectid or finds an item of 'type'
5871 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5873 int btrfs_previous_item(struct btrfs_root *root,
5874 struct btrfs_path *path, u64 min_objectid,
5877 struct btrfs_key found_key;
5878 struct extent_buffer *leaf;
5883 if (path->slots[0] == 0) {
5884 btrfs_set_path_blocking(path);
5885 ret = btrfs_prev_leaf(root, path);
5891 leaf = path->nodes[0];
5892 nritems = btrfs_header_nritems(leaf);
5895 if (path->slots[0] == nritems)
5898 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5899 if (found_key.objectid < min_objectid)
5901 if (found_key.type == type)
5903 if (found_key.objectid == min_objectid &&
5904 found_key.type < type)
5911 * search in extent tree to find a previous Metadata/Data extent item with
5914 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5916 int btrfs_previous_extent_item(struct btrfs_root *root,
5917 struct btrfs_path *path, u64 min_objectid)
5919 struct btrfs_key found_key;
5920 struct extent_buffer *leaf;
5925 if (path->slots[0] == 0) {
5926 btrfs_set_path_blocking(path);
5927 ret = btrfs_prev_leaf(root, path);
5933 leaf = path->nodes[0];
5934 nritems = btrfs_header_nritems(leaf);
5937 if (path->slots[0] == nritems)
5940 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5941 if (found_key.objectid < min_objectid)
5943 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5944 found_key.type == BTRFS_METADATA_ITEM_KEY)
5946 if (found_key.objectid == min_objectid &&
5947 found_key.type < BTRFS_EXTENT_ITEM_KEY)