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"
17 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
18 *root, struct btrfs_path *path, int level);
19 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
20 const struct btrfs_key *ins_key, struct btrfs_path *path,
21 int data_size, int extend);
22 static int push_node_left(struct btrfs_trans_handle *trans,
23 struct btrfs_fs_info *fs_info,
24 struct extent_buffer *dst,
25 struct extent_buffer *src, int empty);
26 static int balance_node_right(struct btrfs_trans_handle *trans,
27 struct btrfs_fs_info *fs_info,
28 struct extent_buffer *dst_buf,
29 struct extent_buffer *src_buf);
30 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
33 struct btrfs_path *btrfs_alloc_path(void)
35 return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
39 * set all locked nodes in the path to blocking locks. This should
40 * be done before scheduling
42 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
45 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
46 if (!p->nodes[i] || !p->locks[i])
48 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
49 if (p->locks[i] == BTRFS_READ_LOCK)
50 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
51 else if (p->locks[i] == BTRFS_WRITE_LOCK)
52 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
56 /* this also releases the path */
57 void btrfs_free_path(struct btrfs_path *p)
61 btrfs_release_path(p);
62 kmem_cache_free(btrfs_path_cachep, p);
66 * path release drops references on the extent buffers in the path
67 * and it drops any locks held by this path
69 * It is safe to call this on paths that no locks or extent buffers held.
71 noinline void btrfs_release_path(struct btrfs_path *p)
75 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
80 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
83 free_extent_buffer(p->nodes[i]);
89 * safely gets a reference on the root node of a tree. A lock
90 * is not taken, so a concurrent writer may put a different node
91 * at the root of the tree. See btrfs_lock_root_node for the
94 * The extent buffer returned by this has a reference taken, so
95 * it won't disappear. It may stop being the root of the tree
96 * at any time because there are no locks held.
98 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
100 struct extent_buffer *eb;
104 eb = rcu_dereference(root->node);
107 * RCU really hurts here, we could free up the root node because
108 * it was COWed but we may not get the new root node yet so do
109 * the inc_not_zero dance and if it doesn't work then
110 * synchronize_rcu and try again.
112 if (atomic_inc_not_zero(&eb->refs)) {
122 /* loop around taking references on and locking the root node of the
123 * tree until you end up with a lock on the root. A locked buffer
124 * is returned, with a reference held.
126 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
128 struct extent_buffer *eb;
131 eb = btrfs_root_node(root);
133 if (eb == root->node)
135 btrfs_tree_unlock(eb);
136 free_extent_buffer(eb);
141 /* loop around taking references on and locking the root node of the
142 * tree until you end up with a lock on the root. A locked buffer
143 * is returned, with a reference held.
145 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
147 struct extent_buffer *eb;
150 eb = btrfs_root_node(root);
151 btrfs_tree_read_lock(eb);
152 if (eb == root->node)
154 btrfs_tree_read_unlock(eb);
155 free_extent_buffer(eb);
160 /* cowonly root (everything not a reference counted cow subvolume), just get
161 * put onto a simple dirty list. transaction.c walks this to make sure they
162 * get properly updated on disk.
164 static void add_root_to_dirty_list(struct btrfs_root *root)
166 struct btrfs_fs_info *fs_info = root->fs_info;
168 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
169 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
172 spin_lock(&fs_info->trans_lock);
173 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
174 /* Want the extent tree to be the last on the list */
175 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
176 list_move_tail(&root->dirty_list,
177 &fs_info->dirty_cowonly_roots);
179 list_move(&root->dirty_list,
180 &fs_info->dirty_cowonly_roots);
182 spin_unlock(&fs_info->trans_lock);
186 * used by snapshot creation to make a copy of a root for a tree with
187 * a given objectid. The buffer with the new root node is returned in
188 * cow_ret, and this func returns zero on success or a negative error code.
190 int btrfs_copy_root(struct btrfs_trans_handle *trans,
191 struct btrfs_root *root,
192 struct extent_buffer *buf,
193 struct extent_buffer **cow_ret, u64 new_root_objectid)
195 struct btrfs_fs_info *fs_info = root->fs_info;
196 struct extent_buffer *cow;
199 struct btrfs_disk_key disk_key;
201 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
202 trans->transid != fs_info->running_transaction->transid);
203 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
204 trans->transid != root->last_trans);
206 level = btrfs_header_level(buf);
208 btrfs_item_key(buf, &disk_key, 0);
210 btrfs_node_key(buf, &disk_key, 0);
212 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
213 &disk_key, level, buf->start, 0);
217 copy_extent_buffer_full(cow, buf);
218 btrfs_set_header_bytenr(cow, cow->start);
219 btrfs_set_header_generation(cow, trans->transid);
220 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
221 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
222 BTRFS_HEADER_FLAG_RELOC);
223 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
224 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
226 btrfs_set_header_owner(cow, new_root_objectid);
228 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
230 WARN_ON(btrfs_header_generation(buf) > trans->transid);
231 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
232 ret = btrfs_inc_ref(trans, root, cow, 1);
234 ret = btrfs_inc_ref(trans, root, cow, 0);
239 btrfs_mark_buffer_dirty(cow);
248 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
249 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
251 MOD_LOG_ROOT_REPLACE,
254 struct tree_mod_root {
259 struct tree_mod_elem {
265 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
268 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
271 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
272 struct btrfs_disk_key key;
275 /* this is used for op == MOD_LOG_MOVE_KEYS */
281 /* this is used for op == MOD_LOG_ROOT_REPLACE */
282 struct tree_mod_root old_root;
286 * Pull a new tree mod seq number for our operation.
288 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
290 return atomic64_inc_return(&fs_info->tree_mod_seq);
294 * This adds a new blocker to the tree mod log's blocker list if the @elem
295 * passed does not already have a sequence number set. So when a caller expects
296 * to record tree modifications, it should ensure to set elem->seq to zero
297 * before calling btrfs_get_tree_mod_seq.
298 * Returns a fresh, unused tree log modification sequence number, even if no new
301 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
302 struct seq_list *elem)
304 write_lock(&fs_info->tree_mod_log_lock);
305 spin_lock(&fs_info->tree_mod_seq_lock);
307 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
308 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
310 spin_unlock(&fs_info->tree_mod_seq_lock);
311 write_unlock(&fs_info->tree_mod_log_lock);
316 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
317 struct seq_list *elem)
319 struct rb_root *tm_root;
320 struct rb_node *node;
321 struct rb_node *next;
322 struct seq_list *cur_elem;
323 struct tree_mod_elem *tm;
324 u64 min_seq = (u64)-1;
325 u64 seq_putting = elem->seq;
330 spin_lock(&fs_info->tree_mod_seq_lock);
331 list_del(&elem->list);
334 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
335 if (cur_elem->seq < min_seq) {
336 if (seq_putting > cur_elem->seq) {
338 * blocker with lower sequence number exists, we
339 * cannot remove anything from the log
341 spin_unlock(&fs_info->tree_mod_seq_lock);
344 min_seq = cur_elem->seq;
347 spin_unlock(&fs_info->tree_mod_seq_lock);
350 * anything that's lower than the lowest existing (read: blocked)
351 * sequence number can be removed from the tree.
353 write_lock(&fs_info->tree_mod_log_lock);
354 tm_root = &fs_info->tree_mod_log;
355 for (node = rb_first(tm_root); node; node = next) {
356 next = rb_next(node);
357 tm = rb_entry(node, struct tree_mod_elem, node);
358 if (tm->seq > min_seq)
360 rb_erase(node, tm_root);
363 write_unlock(&fs_info->tree_mod_log_lock);
367 * key order of the log:
368 * node/leaf start address -> sequence
370 * The 'start address' is the logical address of the *new* root node
371 * for root replace operations, or the logical address of the affected
372 * block for all other operations.
374 * Note: must be called with write lock for fs_info::tree_mod_log_lock.
377 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
379 struct rb_root *tm_root;
380 struct rb_node **new;
381 struct rb_node *parent = NULL;
382 struct tree_mod_elem *cur;
384 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
386 tm_root = &fs_info->tree_mod_log;
387 new = &tm_root->rb_node;
389 cur = rb_entry(*new, struct tree_mod_elem, node);
391 if (cur->logical < tm->logical)
392 new = &((*new)->rb_left);
393 else if (cur->logical > tm->logical)
394 new = &((*new)->rb_right);
395 else if (cur->seq < tm->seq)
396 new = &((*new)->rb_left);
397 else if (cur->seq > tm->seq)
398 new = &((*new)->rb_right);
403 rb_link_node(&tm->node, parent, new);
404 rb_insert_color(&tm->node, tm_root);
409 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
410 * returns zero with the tree_mod_log_lock acquired. The caller must hold
411 * this until all tree mod log insertions are recorded in the rb tree and then
412 * write unlock fs_info::tree_mod_log_lock.
414 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
415 struct extent_buffer *eb) {
417 if (list_empty(&(fs_info)->tree_mod_seq_list))
419 if (eb && btrfs_header_level(eb) == 0)
422 write_lock(&fs_info->tree_mod_log_lock);
423 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
424 write_unlock(&fs_info->tree_mod_log_lock);
431 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
432 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
433 struct extent_buffer *eb)
436 if (list_empty(&(fs_info)->tree_mod_seq_list))
438 if (eb && btrfs_header_level(eb) == 0)
444 static struct tree_mod_elem *
445 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
446 enum mod_log_op op, gfp_t flags)
448 struct tree_mod_elem *tm;
450 tm = kzalloc(sizeof(*tm), flags);
454 tm->logical = eb->start;
455 if (op != MOD_LOG_KEY_ADD) {
456 btrfs_node_key(eb, &tm->key, slot);
457 tm->blockptr = btrfs_node_blockptr(eb, slot);
461 tm->generation = btrfs_node_ptr_generation(eb, slot);
462 RB_CLEAR_NODE(&tm->node);
467 static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot,
468 enum mod_log_op op, gfp_t flags)
470 struct tree_mod_elem *tm;
473 if (!tree_mod_need_log(eb->fs_info, eb))
476 tm = alloc_tree_mod_elem(eb, slot, op, flags);
480 if (tree_mod_dont_log(eb->fs_info, eb)) {
485 ret = __tree_mod_log_insert(eb->fs_info, tm);
486 write_unlock(&eb->fs_info->tree_mod_log_lock);
493 static noinline int tree_mod_log_insert_move(struct extent_buffer *eb,
494 int dst_slot, int src_slot, int nr_items)
496 struct tree_mod_elem *tm = NULL;
497 struct tree_mod_elem **tm_list = NULL;
502 if (!tree_mod_need_log(eb->fs_info, eb))
505 tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
509 tm = kzalloc(sizeof(*tm), GFP_NOFS);
515 tm->logical = eb->start;
517 tm->move.dst_slot = dst_slot;
518 tm->move.nr_items = nr_items;
519 tm->op = MOD_LOG_MOVE_KEYS;
521 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
522 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
523 MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
530 if (tree_mod_dont_log(eb->fs_info, eb))
535 * When we override something during the move, we log these removals.
536 * This can only happen when we move towards the beginning of the
537 * buffer, i.e. dst_slot < src_slot.
539 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
540 ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]);
545 ret = __tree_mod_log_insert(eb->fs_info, tm);
548 write_unlock(&eb->fs_info->tree_mod_log_lock);
553 for (i = 0; i < nr_items; i++) {
554 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
555 rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log);
559 write_unlock(&eb->fs_info->tree_mod_log_lock);
567 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
568 struct tree_mod_elem **tm_list,
574 for (i = nritems - 1; i >= 0; i--) {
575 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
577 for (j = nritems - 1; j > i; j--)
578 rb_erase(&tm_list[j]->node,
579 &fs_info->tree_mod_log);
587 static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root,
588 struct extent_buffer *new_root, int log_removal)
590 struct btrfs_fs_info *fs_info = old_root->fs_info;
591 struct tree_mod_elem *tm = NULL;
592 struct tree_mod_elem **tm_list = NULL;
597 if (!tree_mod_need_log(fs_info, NULL))
600 if (log_removal && btrfs_header_level(old_root) > 0) {
601 nritems = btrfs_header_nritems(old_root);
602 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
608 for (i = 0; i < nritems; i++) {
609 tm_list[i] = alloc_tree_mod_elem(old_root, i,
610 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
618 tm = kzalloc(sizeof(*tm), GFP_NOFS);
624 tm->logical = new_root->start;
625 tm->old_root.logical = old_root->start;
626 tm->old_root.level = btrfs_header_level(old_root);
627 tm->generation = btrfs_header_generation(old_root);
628 tm->op = MOD_LOG_ROOT_REPLACE;
630 if (tree_mod_dont_log(fs_info, NULL))
634 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
636 ret = __tree_mod_log_insert(fs_info, tm);
638 write_unlock(&fs_info->tree_mod_log_lock);
647 for (i = 0; i < nritems; i++)
656 static struct tree_mod_elem *
657 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
660 struct rb_root *tm_root;
661 struct rb_node *node;
662 struct tree_mod_elem *cur = NULL;
663 struct tree_mod_elem *found = NULL;
665 read_lock(&fs_info->tree_mod_log_lock);
666 tm_root = &fs_info->tree_mod_log;
667 node = tm_root->rb_node;
669 cur = rb_entry(node, struct tree_mod_elem, node);
670 if (cur->logical < start) {
671 node = node->rb_left;
672 } else if (cur->logical > start) {
673 node = node->rb_right;
674 } else if (cur->seq < min_seq) {
675 node = node->rb_left;
676 } else if (!smallest) {
677 /* we want the node with the highest seq */
679 BUG_ON(found->seq > cur->seq);
681 node = node->rb_left;
682 } else if (cur->seq > min_seq) {
683 /* we want the node with the smallest seq */
685 BUG_ON(found->seq < cur->seq);
687 node = node->rb_right;
693 read_unlock(&fs_info->tree_mod_log_lock);
699 * this returns the element from the log with the smallest time sequence
700 * value that's in the log (the oldest log item). any element with a time
701 * sequence lower than min_seq will be ignored.
703 static struct tree_mod_elem *
704 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
707 return __tree_mod_log_search(fs_info, start, min_seq, 1);
711 * this returns the element from the log with the largest time sequence
712 * value that's in the log (the most recent log item). any element with
713 * a time sequence lower than min_seq will be ignored.
715 static struct tree_mod_elem *
716 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
718 return __tree_mod_log_search(fs_info, start, min_seq, 0);
722 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
723 struct extent_buffer *src, unsigned long dst_offset,
724 unsigned long src_offset, int nr_items)
727 struct tree_mod_elem **tm_list = NULL;
728 struct tree_mod_elem **tm_list_add, **tm_list_rem;
732 if (!tree_mod_need_log(fs_info, NULL))
735 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
738 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
743 tm_list_add = tm_list;
744 tm_list_rem = tm_list + nr_items;
745 for (i = 0; i < nr_items; i++) {
746 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
747 MOD_LOG_KEY_REMOVE, GFP_NOFS);
748 if (!tm_list_rem[i]) {
753 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
754 MOD_LOG_KEY_ADD, GFP_NOFS);
755 if (!tm_list_add[i]) {
761 if (tree_mod_dont_log(fs_info, NULL))
765 for (i = 0; i < nr_items; i++) {
766 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
769 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
774 write_unlock(&fs_info->tree_mod_log_lock);
780 for (i = 0; i < nr_items * 2; i++) {
781 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
782 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
786 write_unlock(&fs_info->tree_mod_log_lock);
792 static noinline int tree_mod_log_free_eb(struct extent_buffer *eb)
794 struct tree_mod_elem **tm_list = NULL;
799 if (btrfs_header_level(eb) == 0)
802 if (!tree_mod_need_log(eb->fs_info, NULL))
805 nritems = btrfs_header_nritems(eb);
806 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
810 for (i = 0; i < nritems; i++) {
811 tm_list[i] = alloc_tree_mod_elem(eb, i,
812 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
819 if (tree_mod_dont_log(eb->fs_info, eb))
822 ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems);
823 write_unlock(&eb->fs_info->tree_mod_log_lock);
831 for (i = 0; i < nritems; i++)
839 * check if the tree block can be shared by multiple trees
841 int btrfs_block_can_be_shared(struct btrfs_root *root,
842 struct extent_buffer *buf)
845 * Tree blocks not in reference counted trees and tree roots
846 * are never shared. If a block was allocated after the last
847 * snapshot and the block was not allocated by tree relocation,
848 * we know the block is not shared.
850 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
851 buf != root->node && buf != root->commit_root &&
852 (btrfs_header_generation(buf) <=
853 btrfs_root_last_snapshot(&root->root_item) ||
854 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
860 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
861 struct btrfs_root *root,
862 struct extent_buffer *buf,
863 struct extent_buffer *cow,
866 struct btrfs_fs_info *fs_info = root->fs_info;
874 * Backrefs update rules:
876 * Always use full backrefs for extent pointers in tree block
877 * allocated by tree relocation.
879 * If a shared tree block is no longer referenced by its owner
880 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
881 * use full backrefs for extent pointers in tree block.
883 * If a tree block is been relocating
884 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
885 * use full backrefs for extent pointers in tree block.
886 * The reason for this is some operations (such as drop tree)
887 * are only allowed for blocks use full backrefs.
890 if (btrfs_block_can_be_shared(root, buf)) {
891 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
892 btrfs_header_level(buf), 1,
898 btrfs_handle_fs_error(fs_info, ret, NULL);
903 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
904 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
905 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
910 owner = btrfs_header_owner(buf);
911 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
912 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
915 if ((owner == root->root_key.objectid ||
916 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
917 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
918 ret = btrfs_inc_ref(trans, root, buf, 1);
922 if (root->root_key.objectid ==
923 BTRFS_TREE_RELOC_OBJECTID) {
924 ret = btrfs_dec_ref(trans, root, buf, 0);
927 ret = btrfs_inc_ref(trans, root, cow, 1);
931 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
934 if (root->root_key.objectid ==
935 BTRFS_TREE_RELOC_OBJECTID)
936 ret = btrfs_inc_ref(trans, root, cow, 1);
938 ret = btrfs_inc_ref(trans, root, cow, 0);
942 if (new_flags != 0) {
943 int level = btrfs_header_level(buf);
945 ret = btrfs_set_disk_extent_flags(trans, fs_info,
948 new_flags, level, 0);
953 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
954 if (root->root_key.objectid ==
955 BTRFS_TREE_RELOC_OBJECTID)
956 ret = btrfs_inc_ref(trans, root, cow, 1);
958 ret = btrfs_inc_ref(trans, root, cow, 0);
961 ret = btrfs_dec_ref(trans, root, buf, 1);
965 clean_tree_block(fs_info, buf);
972 * does the dirty work in cow of a single block. The parent block (if
973 * supplied) is updated to point to the new cow copy. The new buffer is marked
974 * dirty and returned locked. If you modify the block it needs to be marked
977 * search_start -- an allocation hint for the new block
979 * empty_size -- a hint that you plan on doing more cow. This is the size in
980 * bytes the allocator should try to find free next to the block it returns.
981 * This is just a hint and may be ignored by the allocator.
983 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
984 struct btrfs_root *root,
985 struct extent_buffer *buf,
986 struct extent_buffer *parent, int parent_slot,
987 struct extent_buffer **cow_ret,
988 u64 search_start, u64 empty_size)
990 struct btrfs_fs_info *fs_info = root->fs_info;
991 struct btrfs_disk_key disk_key;
992 struct extent_buffer *cow;
996 u64 parent_start = 0;
1001 btrfs_assert_tree_locked(buf);
1003 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1004 trans->transid != fs_info->running_transaction->transid);
1005 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1006 trans->transid != root->last_trans);
1008 level = btrfs_header_level(buf);
1011 btrfs_item_key(buf, &disk_key, 0);
1013 btrfs_node_key(buf, &disk_key, 0);
1015 if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
1016 parent_start = parent->start;
1019 * If we are COWing a node/leaf from the extent, chunk or device trees,
1020 * make sure that we do not finish block group creation of pending block
1021 * groups. We do this to avoid a deadlock.
1022 * COWing can result in allocation of a new chunk, and flushing pending
1023 * block groups (btrfs_create_pending_block_groups()) can be triggered
1024 * when finishing allocation of a new chunk. Creation of a pending block
1025 * group modifies the extent, chunk and device trees, therefore we could
1026 * deadlock with ourselves since we are holding a lock on an extent
1027 * buffer that btrfs_create_pending_block_groups() may try to COW later.
1029 if (root == fs_info->extent_root ||
1030 root == fs_info->chunk_root ||
1031 root == fs_info->dev_root)
1032 trans->can_flush_pending_bgs = false;
1034 cow = btrfs_alloc_tree_block(trans, root, parent_start,
1035 root->root_key.objectid, &disk_key, level,
1036 search_start, empty_size);
1037 trans->can_flush_pending_bgs = true;
1039 return PTR_ERR(cow);
1041 /* cow is set to blocking by btrfs_init_new_buffer */
1043 copy_extent_buffer_full(cow, buf);
1044 btrfs_set_header_bytenr(cow, cow->start);
1045 btrfs_set_header_generation(cow, trans->transid);
1046 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1047 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1048 BTRFS_HEADER_FLAG_RELOC);
1049 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1050 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1052 btrfs_set_header_owner(cow, root->root_key.objectid);
1054 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
1056 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1058 btrfs_abort_transaction(trans, ret);
1062 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1063 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1065 btrfs_abort_transaction(trans, ret);
1070 if (buf == root->node) {
1071 WARN_ON(parent && parent != buf);
1072 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1073 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1074 parent_start = buf->start;
1076 extent_buffer_get(cow);
1077 ret = tree_mod_log_insert_root(root->node, cow, 1);
1079 rcu_assign_pointer(root->node, cow);
1081 btrfs_free_tree_block(trans, root, buf, parent_start,
1083 free_extent_buffer(buf);
1084 add_root_to_dirty_list(root);
1086 WARN_ON(trans->transid != btrfs_header_generation(parent));
1087 tree_mod_log_insert_key(parent, parent_slot,
1088 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1089 btrfs_set_node_blockptr(parent, parent_slot,
1091 btrfs_set_node_ptr_generation(parent, parent_slot,
1093 btrfs_mark_buffer_dirty(parent);
1095 ret = tree_mod_log_free_eb(buf);
1097 btrfs_abort_transaction(trans, ret);
1101 btrfs_free_tree_block(trans, root, buf, parent_start,
1105 btrfs_tree_unlock(buf);
1106 free_extent_buffer_stale(buf);
1107 btrfs_mark_buffer_dirty(cow);
1113 * returns the logical address of the oldest predecessor of the given root.
1114 * entries older than time_seq are ignored.
1116 static struct tree_mod_elem *__tree_mod_log_oldest_root(
1117 struct extent_buffer *eb_root, u64 time_seq)
1119 struct tree_mod_elem *tm;
1120 struct tree_mod_elem *found = NULL;
1121 u64 root_logical = eb_root->start;
1128 * the very last operation that's logged for a root is the
1129 * replacement operation (if it is replaced at all). this has
1130 * the logical address of the *new* root, making it the very
1131 * first operation that's logged for this root.
1134 tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
1139 * if there are no tree operation for the oldest root, we simply
1140 * return it. this should only happen if that (old) root is at
1147 * if there's an operation that's not a root replacement, we
1148 * found the oldest version of our root. normally, we'll find a
1149 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1151 if (tm->op != MOD_LOG_ROOT_REPLACE)
1155 root_logical = tm->old_root.logical;
1159 /* if there's no old root to return, return what we found instead */
1167 * tm is a pointer to the first operation to rewind within eb. then, all
1168 * previous operations will be rewound (until we reach something older than
1172 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1173 u64 time_seq, struct tree_mod_elem *first_tm)
1176 struct rb_node *next;
1177 struct tree_mod_elem *tm = first_tm;
1178 unsigned long o_dst;
1179 unsigned long o_src;
1180 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1182 n = btrfs_header_nritems(eb);
1183 read_lock(&fs_info->tree_mod_log_lock);
1184 while (tm && tm->seq >= time_seq) {
1186 * all the operations are recorded with the operator used for
1187 * the modification. as we're going backwards, we do the
1188 * opposite of each operation here.
1191 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1192 BUG_ON(tm->slot < n);
1194 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1195 case MOD_LOG_KEY_REMOVE:
1196 btrfs_set_node_key(eb, &tm->key, tm->slot);
1197 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1198 btrfs_set_node_ptr_generation(eb, tm->slot,
1202 case MOD_LOG_KEY_REPLACE:
1203 BUG_ON(tm->slot >= n);
1204 btrfs_set_node_key(eb, &tm->key, tm->slot);
1205 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1206 btrfs_set_node_ptr_generation(eb, tm->slot,
1209 case MOD_LOG_KEY_ADD:
1210 /* if a move operation is needed it's in the log */
1213 case MOD_LOG_MOVE_KEYS:
1214 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1215 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1216 memmove_extent_buffer(eb, o_dst, o_src,
1217 tm->move.nr_items * p_size);
1219 case MOD_LOG_ROOT_REPLACE:
1221 * this operation is special. for roots, this must be
1222 * handled explicitly before rewinding.
1223 * for non-roots, this operation may exist if the node
1224 * was a root: root A -> child B; then A gets empty and
1225 * B is promoted to the new root. in the mod log, we'll
1226 * have a root-replace operation for B, a tree block
1227 * that is no root. we simply ignore that operation.
1231 next = rb_next(&tm->node);
1234 tm = rb_entry(next, struct tree_mod_elem, node);
1235 if (tm->logical != first_tm->logical)
1238 read_unlock(&fs_info->tree_mod_log_lock);
1239 btrfs_set_header_nritems(eb, n);
1243 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1244 * is returned. If rewind operations happen, a fresh buffer is returned. The
1245 * returned buffer is always read-locked. If the returned buffer is not the
1246 * input buffer, the lock on the input buffer is released and the input buffer
1247 * is freed (its refcount is decremented).
1249 static struct extent_buffer *
1250 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1251 struct extent_buffer *eb, u64 time_seq)
1253 struct extent_buffer *eb_rewin;
1254 struct tree_mod_elem *tm;
1259 if (btrfs_header_level(eb) == 0)
1262 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1266 btrfs_set_path_blocking(path);
1267 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1269 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1270 BUG_ON(tm->slot != 0);
1271 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1273 btrfs_tree_read_unlock_blocking(eb);
1274 free_extent_buffer(eb);
1277 btrfs_set_header_bytenr(eb_rewin, eb->start);
1278 btrfs_set_header_backref_rev(eb_rewin,
1279 btrfs_header_backref_rev(eb));
1280 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1281 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1283 eb_rewin = btrfs_clone_extent_buffer(eb);
1285 btrfs_tree_read_unlock_blocking(eb);
1286 free_extent_buffer(eb);
1291 btrfs_tree_read_unlock_blocking(eb);
1292 free_extent_buffer(eb);
1294 btrfs_tree_read_lock(eb_rewin);
1295 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1296 WARN_ON(btrfs_header_nritems(eb_rewin) >
1297 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1303 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1304 * value. If there are no changes, the current root->root_node is returned. If
1305 * anything changed in between, there's a fresh buffer allocated on which the
1306 * rewind operations are done. In any case, the returned buffer is read locked.
1307 * Returns NULL on error (with no locks held).
1309 static inline struct extent_buffer *
1310 get_old_root(struct btrfs_root *root, u64 time_seq)
1312 struct btrfs_fs_info *fs_info = root->fs_info;
1313 struct tree_mod_elem *tm;
1314 struct extent_buffer *eb = NULL;
1315 struct extent_buffer *eb_root;
1316 struct extent_buffer *old;
1317 struct tree_mod_root *old_root = NULL;
1318 u64 old_generation = 0;
1322 eb_root = btrfs_read_lock_root_node(root);
1323 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1327 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1328 old_root = &tm->old_root;
1329 old_generation = tm->generation;
1330 logical = old_root->logical;
1331 level = old_root->level;
1333 logical = eb_root->start;
1334 level = btrfs_header_level(eb_root);
1337 tm = tree_mod_log_search(fs_info, logical, time_seq);
1338 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1339 btrfs_tree_read_unlock(eb_root);
1340 free_extent_buffer(eb_root);
1341 old = read_tree_block(fs_info, logical, 0, level, NULL);
1342 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1344 free_extent_buffer(old);
1346 "failed to read tree block %llu from get_old_root",
1349 eb = btrfs_clone_extent_buffer(old);
1350 free_extent_buffer(old);
1352 } else if (old_root) {
1353 btrfs_tree_read_unlock(eb_root);
1354 free_extent_buffer(eb_root);
1355 eb = alloc_dummy_extent_buffer(fs_info, logical);
1357 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1358 eb = btrfs_clone_extent_buffer(eb_root);
1359 btrfs_tree_read_unlock_blocking(eb_root);
1360 free_extent_buffer(eb_root);
1365 btrfs_tree_read_lock(eb);
1367 btrfs_set_header_bytenr(eb, eb->start);
1368 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1369 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1370 btrfs_set_header_level(eb, old_root->level);
1371 btrfs_set_header_generation(eb, old_generation);
1374 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1376 WARN_ON(btrfs_header_level(eb) != 0);
1377 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1382 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1384 struct tree_mod_elem *tm;
1386 struct extent_buffer *eb_root = btrfs_root_node(root);
1388 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1389 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1390 level = tm->old_root.level;
1392 level = btrfs_header_level(eb_root);
1394 free_extent_buffer(eb_root);
1399 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1400 struct btrfs_root *root,
1401 struct extent_buffer *buf)
1403 if (btrfs_is_testing(root->fs_info))
1406 /* Ensure we can see the FORCE_COW bit */
1407 smp_mb__before_atomic();
1410 * We do not need to cow a block if
1411 * 1) this block is not created or changed in this transaction;
1412 * 2) this block does not belong to TREE_RELOC tree;
1413 * 3) the root is not forced COW.
1415 * What is forced COW:
1416 * when we create snapshot during committing the transaction,
1417 * after we've finished copying src root, we must COW the shared
1418 * block to ensure the metadata consistency.
1420 if (btrfs_header_generation(buf) == trans->transid &&
1421 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1422 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1423 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1424 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1430 * cows a single block, see __btrfs_cow_block for the real work.
1431 * This version of it has extra checks so that a block isn't COWed more than
1432 * once per transaction, as long as it hasn't been written yet
1434 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1435 struct btrfs_root *root, struct extent_buffer *buf,
1436 struct extent_buffer *parent, int parent_slot,
1437 struct extent_buffer **cow_ret)
1439 struct btrfs_fs_info *fs_info = root->fs_info;
1443 if (test_bit(BTRFS_ROOT_DELETING, &root->state))
1445 "COW'ing blocks on a fs root that's being dropped");
1447 if (trans->transaction != fs_info->running_transaction)
1448 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1450 fs_info->running_transaction->transid);
1452 if (trans->transid != fs_info->generation)
1453 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1454 trans->transid, fs_info->generation);
1456 if (!should_cow_block(trans, root, buf)) {
1457 trans->dirty = true;
1462 search_start = buf->start & ~((u64)SZ_1G - 1);
1465 btrfs_set_lock_blocking(parent);
1466 btrfs_set_lock_blocking(buf);
1468 ret = __btrfs_cow_block(trans, root, buf, parent,
1469 parent_slot, cow_ret, search_start, 0);
1471 trace_btrfs_cow_block(root, buf, *cow_ret);
1477 * helper function for defrag to decide if two blocks pointed to by a
1478 * node are actually close by
1480 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1482 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1484 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1490 * compare two keys in a memcmp fashion
1492 static int comp_keys(const struct btrfs_disk_key *disk,
1493 const struct btrfs_key *k2)
1495 struct btrfs_key k1;
1497 btrfs_disk_key_to_cpu(&k1, disk);
1499 return btrfs_comp_cpu_keys(&k1, k2);
1503 * same as comp_keys only with two btrfs_key's
1505 int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1507 if (k1->objectid > k2->objectid)
1509 if (k1->objectid < k2->objectid)
1511 if (k1->type > k2->type)
1513 if (k1->type < k2->type)
1515 if (k1->offset > k2->offset)
1517 if (k1->offset < k2->offset)
1523 * this is used by the defrag code to go through all the
1524 * leaves pointed to by a node and reallocate them so that
1525 * disk order is close to key order
1527 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1528 struct btrfs_root *root, struct extent_buffer *parent,
1529 int start_slot, u64 *last_ret,
1530 struct btrfs_key *progress)
1532 struct btrfs_fs_info *fs_info = root->fs_info;
1533 struct extent_buffer *cur;
1536 u64 search_start = *last_ret;
1546 int progress_passed = 0;
1547 struct btrfs_disk_key disk_key;
1549 parent_level = btrfs_header_level(parent);
1551 WARN_ON(trans->transaction != fs_info->running_transaction);
1552 WARN_ON(trans->transid != fs_info->generation);
1554 parent_nritems = btrfs_header_nritems(parent);
1555 blocksize = fs_info->nodesize;
1556 end_slot = parent_nritems - 1;
1558 if (parent_nritems <= 1)
1561 btrfs_set_lock_blocking(parent);
1563 for (i = start_slot; i <= end_slot; i++) {
1564 struct btrfs_key first_key;
1567 btrfs_node_key(parent, &disk_key, i);
1568 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1571 progress_passed = 1;
1572 blocknr = btrfs_node_blockptr(parent, i);
1573 gen = btrfs_node_ptr_generation(parent, i);
1574 btrfs_node_key_to_cpu(parent, &first_key, i);
1575 if (last_block == 0)
1576 last_block = blocknr;
1579 other = btrfs_node_blockptr(parent, i - 1);
1580 close = close_blocks(blocknr, other, blocksize);
1582 if (!close && i < end_slot) {
1583 other = btrfs_node_blockptr(parent, i + 1);
1584 close = close_blocks(blocknr, other, blocksize);
1587 last_block = blocknr;
1591 cur = find_extent_buffer(fs_info, blocknr);
1593 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1596 if (!cur || !uptodate) {
1598 cur = read_tree_block(fs_info, blocknr, gen,
1602 return PTR_ERR(cur);
1603 } else if (!extent_buffer_uptodate(cur)) {
1604 free_extent_buffer(cur);
1607 } else if (!uptodate) {
1608 err = btrfs_read_buffer(cur, gen,
1609 parent_level - 1,&first_key);
1611 free_extent_buffer(cur);
1616 if (search_start == 0)
1617 search_start = last_block;
1619 btrfs_tree_lock(cur);
1620 btrfs_set_lock_blocking(cur);
1621 err = __btrfs_cow_block(trans, root, cur, parent, i,
1624 (end_slot - i) * blocksize));
1626 btrfs_tree_unlock(cur);
1627 free_extent_buffer(cur);
1630 search_start = cur->start;
1631 last_block = cur->start;
1632 *last_ret = search_start;
1633 btrfs_tree_unlock(cur);
1634 free_extent_buffer(cur);
1640 * search for key in the extent_buffer. The items start at offset p,
1641 * and they are item_size apart. There are 'max' items in p.
1643 * the slot in the array is returned via slot, and it points to
1644 * the place where you would insert key if it is not found in
1647 * slot may point to max if the key is bigger than all of the keys
1649 static noinline int generic_bin_search(struct extent_buffer *eb,
1650 unsigned long p, int item_size,
1651 const struct btrfs_key *key,
1658 struct btrfs_disk_key *tmp = NULL;
1659 struct btrfs_disk_key unaligned;
1660 unsigned long offset;
1662 unsigned long map_start = 0;
1663 unsigned long map_len = 0;
1667 btrfs_err(eb->fs_info,
1668 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1669 __func__, low, high, eb->start,
1670 btrfs_header_owner(eb), btrfs_header_level(eb));
1674 while (low < high) {
1675 mid = (low + high) / 2;
1676 offset = p + mid * item_size;
1678 if (!kaddr || offset < map_start ||
1679 (offset + sizeof(struct btrfs_disk_key)) >
1680 map_start + map_len) {
1682 err = map_private_extent_buffer(eb, offset,
1683 sizeof(struct btrfs_disk_key),
1684 &kaddr, &map_start, &map_len);
1687 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1689 } else if (err == 1) {
1690 read_extent_buffer(eb, &unaligned,
1691 offset, sizeof(unaligned));
1698 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1701 ret = comp_keys(tmp, key);
1717 * simple bin_search frontend that does the right thing for
1720 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1721 int level, int *slot)
1724 return generic_bin_search(eb,
1725 offsetof(struct btrfs_leaf, items),
1726 sizeof(struct btrfs_item),
1727 key, btrfs_header_nritems(eb),
1730 return generic_bin_search(eb,
1731 offsetof(struct btrfs_node, ptrs),
1732 sizeof(struct btrfs_key_ptr),
1733 key, btrfs_header_nritems(eb),
1737 static void root_add_used(struct btrfs_root *root, u32 size)
1739 spin_lock(&root->accounting_lock);
1740 btrfs_set_root_used(&root->root_item,
1741 btrfs_root_used(&root->root_item) + size);
1742 spin_unlock(&root->accounting_lock);
1745 static void root_sub_used(struct btrfs_root *root, u32 size)
1747 spin_lock(&root->accounting_lock);
1748 btrfs_set_root_used(&root->root_item,
1749 btrfs_root_used(&root->root_item) - size);
1750 spin_unlock(&root->accounting_lock);
1753 /* given a node and slot number, this reads the blocks it points to. The
1754 * extent buffer is returned with a reference taken (but unlocked).
1756 static noinline struct extent_buffer *
1757 read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent,
1760 int level = btrfs_header_level(parent);
1761 struct extent_buffer *eb;
1762 struct btrfs_key first_key;
1764 if (slot < 0 || slot >= btrfs_header_nritems(parent))
1765 return ERR_PTR(-ENOENT);
1769 btrfs_node_key_to_cpu(parent, &first_key, slot);
1770 eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot),
1771 btrfs_node_ptr_generation(parent, slot),
1772 level - 1, &first_key);
1773 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1774 free_extent_buffer(eb);
1782 * node level balancing, used to make sure nodes are in proper order for
1783 * item deletion. We balance from the top down, so we have to make sure
1784 * that a deletion won't leave an node completely empty later on.
1786 static noinline int balance_level(struct btrfs_trans_handle *trans,
1787 struct btrfs_root *root,
1788 struct btrfs_path *path, int level)
1790 struct btrfs_fs_info *fs_info = root->fs_info;
1791 struct extent_buffer *right = NULL;
1792 struct extent_buffer *mid;
1793 struct extent_buffer *left = NULL;
1794 struct extent_buffer *parent = NULL;
1798 int orig_slot = path->slots[level];
1803 mid = path->nodes[level];
1805 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1806 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1807 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1809 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1811 if (level < BTRFS_MAX_LEVEL - 1) {
1812 parent = path->nodes[level + 1];
1813 pslot = path->slots[level + 1];
1817 * deal with the case where there is only one pointer in the root
1818 * by promoting the node below to a root
1821 struct extent_buffer *child;
1823 if (btrfs_header_nritems(mid) != 1)
1826 /* promote the child to a root */
1827 child = read_node_slot(fs_info, mid, 0);
1828 if (IS_ERR(child)) {
1829 ret = PTR_ERR(child);
1830 btrfs_handle_fs_error(fs_info, ret, NULL);
1834 btrfs_tree_lock(child);
1835 btrfs_set_lock_blocking(child);
1836 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1838 btrfs_tree_unlock(child);
1839 free_extent_buffer(child);
1843 ret = tree_mod_log_insert_root(root->node, child, 1);
1845 rcu_assign_pointer(root->node, child);
1847 add_root_to_dirty_list(root);
1848 btrfs_tree_unlock(child);
1850 path->locks[level] = 0;
1851 path->nodes[level] = NULL;
1852 clean_tree_block(fs_info, mid);
1853 btrfs_tree_unlock(mid);
1854 /* once for the path */
1855 free_extent_buffer(mid);
1857 root_sub_used(root, mid->len);
1858 btrfs_free_tree_block(trans, root, mid, 0, 1);
1859 /* once for the root ptr */
1860 free_extent_buffer_stale(mid);
1863 if (btrfs_header_nritems(mid) >
1864 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1867 left = read_node_slot(fs_info, parent, pslot - 1);
1872 btrfs_tree_lock(left);
1873 btrfs_set_lock_blocking(left);
1874 wret = btrfs_cow_block(trans, root, left,
1875 parent, pslot - 1, &left);
1882 right = read_node_slot(fs_info, parent, pslot + 1);
1887 btrfs_tree_lock(right);
1888 btrfs_set_lock_blocking(right);
1889 wret = btrfs_cow_block(trans, root, right,
1890 parent, pslot + 1, &right);
1897 /* first, try to make some room in the middle buffer */
1899 orig_slot += btrfs_header_nritems(left);
1900 wret = push_node_left(trans, fs_info, left, mid, 1);
1906 * then try to empty the right most buffer into the middle
1909 wret = push_node_left(trans, fs_info, mid, right, 1);
1910 if (wret < 0 && wret != -ENOSPC)
1912 if (btrfs_header_nritems(right) == 0) {
1913 clean_tree_block(fs_info, right);
1914 btrfs_tree_unlock(right);
1915 del_ptr(root, path, level + 1, pslot + 1);
1916 root_sub_used(root, right->len);
1917 btrfs_free_tree_block(trans, root, right, 0, 1);
1918 free_extent_buffer_stale(right);
1921 struct btrfs_disk_key right_key;
1922 btrfs_node_key(right, &right_key, 0);
1923 ret = tree_mod_log_insert_key(parent, pslot + 1,
1924 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1926 btrfs_set_node_key(parent, &right_key, pslot + 1);
1927 btrfs_mark_buffer_dirty(parent);
1930 if (btrfs_header_nritems(mid) == 1) {
1932 * we're not allowed to leave a node with one item in the
1933 * tree during a delete. A deletion from lower in the tree
1934 * could try to delete the only pointer in this node.
1935 * So, pull some keys from the left.
1936 * There has to be a left pointer at this point because
1937 * otherwise we would have pulled some pointers from the
1942 btrfs_handle_fs_error(fs_info, ret, NULL);
1945 wret = balance_node_right(trans, fs_info, mid, left);
1951 wret = push_node_left(trans, fs_info, left, mid, 1);
1957 if (btrfs_header_nritems(mid) == 0) {
1958 clean_tree_block(fs_info, mid);
1959 btrfs_tree_unlock(mid);
1960 del_ptr(root, path, level + 1, pslot);
1961 root_sub_used(root, mid->len);
1962 btrfs_free_tree_block(trans, root, mid, 0, 1);
1963 free_extent_buffer_stale(mid);
1966 /* update the parent key to reflect our changes */
1967 struct btrfs_disk_key mid_key;
1968 btrfs_node_key(mid, &mid_key, 0);
1969 ret = tree_mod_log_insert_key(parent, pslot,
1970 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1972 btrfs_set_node_key(parent, &mid_key, pslot);
1973 btrfs_mark_buffer_dirty(parent);
1976 /* update the path */
1978 if (btrfs_header_nritems(left) > orig_slot) {
1979 extent_buffer_get(left);
1980 /* left was locked after cow */
1981 path->nodes[level] = left;
1982 path->slots[level + 1] -= 1;
1983 path->slots[level] = orig_slot;
1985 btrfs_tree_unlock(mid);
1986 free_extent_buffer(mid);
1989 orig_slot -= btrfs_header_nritems(left);
1990 path->slots[level] = orig_slot;
1993 /* double check we haven't messed things up */
1995 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1999 btrfs_tree_unlock(right);
2000 free_extent_buffer(right);
2003 if (path->nodes[level] != left)
2004 btrfs_tree_unlock(left);
2005 free_extent_buffer(left);
2010 /* Node balancing for insertion. Here we only split or push nodes around
2011 * when they are completely full. This is also done top down, so we
2012 * have to be pessimistic.
2014 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2015 struct btrfs_root *root,
2016 struct btrfs_path *path, int level)
2018 struct btrfs_fs_info *fs_info = root->fs_info;
2019 struct extent_buffer *right = NULL;
2020 struct extent_buffer *mid;
2021 struct extent_buffer *left = NULL;
2022 struct extent_buffer *parent = NULL;
2026 int orig_slot = path->slots[level];
2031 mid = path->nodes[level];
2032 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2034 if (level < BTRFS_MAX_LEVEL - 1) {
2035 parent = path->nodes[level + 1];
2036 pslot = path->slots[level + 1];
2042 left = read_node_slot(fs_info, parent, pslot - 1);
2046 /* first, try to make some room in the middle buffer */
2050 btrfs_tree_lock(left);
2051 btrfs_set_lock_blocking(left);
2053 left_nr = btrfs_header_nritems(left);
2054 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2057 ret = btrfs_cow_block(trans, root, left, parent,
2062 wret = push_node_left(trans, fs_info,
2069 struct btrfs_disk_key disk_key;
2070 orig_slot += left_nr;
2071 btrfs_node_key(mid, &disk_key, 0);
2072 ret = tree_mod_log_insert_key(parent, pslot,
2073 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2075 btrfs_set_node_key(parent, &disk_key, pslot);
2076 btrfs_mark_buffer_dirty(parent);
2077 if (btrfs_header_nritems(left) > orig_slot) {
2078 path->nodes[level] = left;
2079 path->slots[level + 1] -= 1;
2080 path->slots[level] = orig_slot;
2081 btrfs_tree_unlock(mid);
2082 free_extent_buffer(mid);
2085 btrfs_header_nritems(left);
2086 path->slots[level] = orig_slot;
2087 btrfs_tree_unlock(left);
2088 free_extent_buffer(left);
2092 btrfs_tree_unlock(left);
2093 free_extent_buffer(left);
2095 right = read_node_slot(fs_info, parent, pslot + 1);
2100 * then try to empty the right most buffer into the middle
2105 btrfs_tree_lock(right);
2106 btrfs_set_lock_blocking(right);
2108 right_nr = btrfs_header_nritems(right);
2109 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2112 ret = btrfs_cow_block(trans, root, right,
2118 wret = balance_node_right(trans, fs_info,
2125 struct btrfs_disk_key disk_key;
2127 btrfs_node_key(right, &disk_key, 0);
2128 ret = tree_mod_log_insert_key(parent, pslot + 1,
2129 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2131 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2132 btrfs_mark_buffer_dirty(parent);
2134 if (btrfs_header_nritems(mid) <= orig_slot) {
2135 path->nodes[level] = right;
2136 path->slots[level + 1] += 1;
2137 path->slots[level] = orig_slot -
2138 btrfs_header_nritems(mid);
2139 btrfs_tree_unlock(mid);
2140 free_extent_buffer(mid);
2142 btrfs_tree_unlock(right);
2143 free_extent_buffer(right);
2147 btrfs_tree_unlock(right);
2148 free_extent_buffer(right);
2154 * readahead one full node of leaves, finding things that are close
2155 * to the block in 'slot', and triggering ra on them.
2157 static void reada_for_search(struct btrfs_fs_info *fs_info,
2158 struct btrfs_path *path,
2159 int level, int slot, u64 objectid)
2161 struct extent_buffer *node;
2162 struct btrfs_disk_key disk_key;
2167 struct extent_buffer *eb;
2175 if (!path->nodes[level])
2178 node = path->nodes[level];
2180 search = btrfs_node_blockptr(node, slot);
2181 blocksize = fs_info->nodesize;
2182 eb = find_extent_buffer(fs_info, search);
2184 free_extent_buffer(eb);
2190 nritems = btrfs_header_nritems(node);
2194 if (path->reada == READA_BACK) {
2198 } else if (path->reada == READA_FORWARD) {
2203 if (path->reada == READA_BACK && objectid) {
2204 btrfs_node_key(node, &disk_key, nr);
2205 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2208 search = btrfs_node_blockptr(node, nr);
2209 if ((search <= target && target - search <= 65536) ||
2210 (search > target && search - target <= 65536)) {
2211 readahead_tree_block(fs_info, search);
2215 if ((nread > 65536 || nscan > 32))
2220 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2221 struct btrfs_path *path, int level)
2225 struct extent_buffer *parent;
2226 struct extent_buffer *eb;
2231 parent = path->nodes[level + 1];
2235 nritems = btrfs_header_nritems(parent);
2236 slot = path->slots[level + 1];
2239 block1 = btrfs_node_blockptr(parent, slot - 1);
2240 gen = btrfs_node_ptr_generation(parent, slot - 1);
2241 eb = find_extent_buffer(fs_info, block1);
2243 * if we get -eagain from btrfs_buffer_uptodate, we
2244 * don't want to return eagain here. That will loop
2247 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2249 free_extent_buffer(eb);
2251 if (slot + 1 < nritems) {
2252 block2 = btrfs_node_blockptr(parent, slot + 1);
2253 gen = btrfs_node_ptr_generation(parent, slot + 1);
2254 eb = find_extent_buffer(fs_info, block2);
2255 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2257 free_extent_buffer(eb);
2261 readahead_tree_block(fs_info, block1);
2263 readahead_tree_block(fs_info, block2);
2268 * when we walk down the tree, it is usually safe to unlock the higher layers
2269 * in the tree. The exceptions are when our path goes through slot 0, because
2270 * operations on the tree might require changing key pointers higher up in the
2273 * callers might also have set path->keep_locks, which tells this code to keep
2274 * the lock if the path points to the last slot in the block. This is part of
2275 * walking through the tree, and selecting the next slot in the higher block.
2277 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2278 * if lowest_unlock is 1, level 0 won't be unlocked
2280 static noinline void unlock_up(struct btrfs_path *path, int level,
2281 int lowest_unlock, int min_write_lock_level,
2282 int *write_lock_level)
2285 int skip_level = level;
2287 struct extent_buffer *t;
2289 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2290 if (!path->nodes[i])
2292 if (!path->locks[i])
2294 if (!no_skips && path->slots[i] == 0) {
2298 if (!no_skips && path->keep_locks) {
2301 nritems = btrfs_header_nritems(t);
2302 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2307 if (skip_level < i && i >= lowest_unlock)
2311 if (i >= lowest_unlock && i > skip_level) {
2312 btrfs_tree_unlock_rw(t, path->locks[i]);
2314 if (write_lock_level &&
2315 i > min_write_lock_level &&
2316 i <= *write_lock_level) {
2317 *write_lock_level = i - 1;
2324 * This releases any locks held in the path starting at level and
2325 * going all the way up to the root.
2327 * btrfs_search_slot will keep the lock held on higher nodes in a few
2328 * corner cases, such as COW of the block at slot zero in the node. This
2329 * ignores those rules, and it should only be called when there are no
2330 * more updates to be done higher up in the tree.
2332 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2336 if (path->keep_locks)
2339 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2340 if (!path->nodes[i])
2342 if (!path->locks[i])
2344 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2350 * helper function for btrfs_search_slot. The goal is to find a block
2351 * in cache without setting the path to blocking. If we find the block
2352 * we return zero and the path is unchanged.
2354 * If we can't find the block, we set the path blocking and do some
2355 * reada. -EAGAIN is returned and the search must be repeated.
2358 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2359 struct extent_buffer **eb_ret, int level, int slot,
2360 const struct btrfs_key *key)
2362 struct btrfs_fs_info *fs_info = root->fs_info;
2365 struct extent_buffer *b = *eb_ret;
2366 struct extent_buffer *tmp;
2367 struct btrfs_key first_key;
2371 blocknr = btrfs_node_blockptr(b, slot);
2372 gen = btrfs_node_ptr_generation(b, slot);
2373 parent_level = btrfs_header_level(b);
2374 btrfs_node_key_to_cpu(b, &first_key, slot);
2376 tmp = find_extent_buffer(fs_info, blocknr);
2378 /* first we do an atomic uptodate check */
2379 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2384 /* the pages were up to date, but we failed
2385 * the generation number check. Do a full
2386 * read for the generation number that is correct.
2387 * We must do this without dropping locks so
2388 * we can trust our generation number
2390 btrfs_set_path_blocking(p);
2392 /* now we're allowed to do a blocking uptodate check */
2393 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2398 free_extent_buffer(tmp);
2399 btrfs_release_path(p);
2404 * reduce lock contention at high levels
2405 * of the btree by dropping locks before
2406 * we read. Don't release the lock on the current
2407 * level because we need to walk this node to figure
2408 * out which blocks to read.
2410 btrfs_unlock_up_safe(p, level + 1);
2411 btrfs_set_path_blocking(p);
2413 if (p->reada != READA_NONE)
2414 reada_for_search(fs_info, p, level, slot, key->objectid);
2417 tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2421 * If the read above didn't mark this buffer up to date,
2422 * it will never end up being up to date. Set ret to EIO now
2423 * and give up so that our caller doesn't loop forever
2426 if (!extent_buffer_uptodate(tmp))
2428 free_extent_buffer(tmp);
2433 btrfs_release_path(p);
2438 * helper function for btrfs_search_slot. This does all of the checks
2439 * for node-level blocks and does any balancing required based on
2442 * If no extra work was required, zero is returned. If we had to
2443 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2447 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2448 struct btrfs_root *root, struct btrfs_path *p,
2449 struct extent_buffer *b, int level, int ins_len,
2450 int *write_lock_level)
2452 struct btrfs_fs_info *fs_info = root->fs_info;
2455 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2456 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2459 if (*write_lock_level < level + 1) {
2460 *write_lock_level = level + 1;
2461 btrfs_release_path(p);
2465 btrfs_set_path_blocking(p);
2466 reada_for_balance(fs_info, p, level);
2467 sret = split_node(trans, root, p, level);
2474 b = p->nodes[level];
2475 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2476 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2479 if (*write_lock_level < level + 1) {
2480 *write_lock_level = level + 1;
2481 btrfs_release_path(p);
2485 btrfs_set_path_blocking(p);
2486 reada_for_balance(fs_info, p, level);
2487 sret = balance_level(trans, root, p, level);
2493 b = p->nodes[level];
2495 btrfs_release_path(p);
2498 BUG_ON(btrfs_header_nritems(b) == 1);
2508 static void key_search_validate(struct extent_buffer *b,
2509 const struct btrfs_key *key,
2512 #ifdef CONFIG_BTRFS_ASSERT
2513 struct btrfs_disk_key disk_key;
2515 btrfs_cpu_key_to_disk(&disk_key, key);
2518 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2519 offsetof(struct btrfs_leaf, items[0].key),
2522 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2523 offsetof(struct btrfs_node, ptrs[0].key),
2528 static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
2529 int level, int *prev_cmp, int *slot)
2531 if (*prev_cmp != 0) {
2532 *prev_cmp = btrfs_bin_search(b, key, level, slot);
2536 key_search_validate(b, key, level);
2542 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2543 u64 iobjectid, u64 ioff, u8 key_type,
2544 struct btrfs_key *found_key)
2547 struct btrfs_key key;
2548 struct extent_buffer *eb;
2553 key.type = key_type;
2554 key.objectid = iobjectid;
2557 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2561 eb = path->nodes[0];
2562 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2563 ret = btrfs_next_leaf(fs_root, path);
2566 eb = path->nodes[0];
2569 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2570 if (found_key->type != key.type ||
2571 found_key->objectid != key.objectid)
2577 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2578 struct btrfs_path *p,
2579 int write_lock_level)
2581 struct btrfs_fs_info *fs_info = root->fs_info;
2582 struct extent_buffer *b;
2586 /* We try very hard to do read locks on the root */
2587 root_lock = BTRFS_READ_LOCK;
2589 if (p->search_commit_root) {
2591 * The commit roots are read only so we always do read locks,
2592 * and we always must hold the commit_root_sem when doing
2593 * searches on them, the only exception is send where we don't
2594 * want to block transaction commits for a long time, so
2595 * we need to clone the commit root in order to avoid races
2596 * with transaction commits that create a snapshot of one of
2597 * the roots used by a send operation.
2599 if (p->need_commit_sem) {
2600 down_read(&fs_info->commit_root_sem);
2601 b = btrfs_clone_extent_buffer(root->commit_root);
2602 up_read(&fs_info->commit_root_sem);
2604 return ERR_PTR(-ENOMEM);
2607 b = root->commit_root;
2608 extent_buffer_get(b);
2610 level = btrfs_header_level(b);
2612 * Ensure that all callers have set skip_locking when
2613 * p->search_commit_root = 1.
2615 ASSERT(p->skip_locking == 1);
2620 if (p->skip_locking) {
2621 b = btrfs_root_node(root);
2622 level = btrfs_header_level(b);
2627 * If the level is set to maximum, we can skip trying to get the read
2630 if (write_lock_level < BTRFS_MAX_LEVEL) {
2632 * We don't know the level of the root node until we actually
2633 * have it read locked
2635 b = btrfs_read_lock_root_node(root);
2636 level = btrfs_header_level(b);
2637 if (level > write_lock_level)
2640 /* Whoops, must trade for write lock */
2641 btrfs_tree_read_unlock(b);
2642 free_extent_buffer(b);
2645 b = btrfs_lock_root_node(root);
2646 root_lock = BTRFS_WRITE_LOCK;
2648 /* The level might have changed, check again */
2649 level = btrfs_header_level(b);
2652 p->nodes[level] = b;
2653 if (!p->skip_locking)
2654 p->locks[level] = root_lock;
2656 * Callers are responsible for dropping b's references.
2663 * btrfs_search_slot - look for a key in a tree and perform necessary
2664 * modifications to preserve tree invariants.
2666 * @trans: Handle of transaction, used when modifying the tree
2667 * @p: Holds all btree nodes along the search path
2668 * @root: The root node of the tree
2669 * @key: The key we are looking for
2670 * @ins_len: Indicates purpose of search, for inserts it is 1, for
2671 * deletions it's -1. 0 for plain searches
2672 * @cow: boolean should CoW operations be performed. Must always be 1
2673 * when modifying the tree.
2675 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2676 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2678 * If @key is found, 0 is returned and you can find the item in the leaf level
2679 * of the path (level 0)
2681 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2682 * points to the slot where it should be inserted
2684 * If an error is encountered while searching the tree a negative error number
2687 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2688 const struct btrfs_key *key, struct btrfs_path *p,
2689 int ins_len, int cow)
2691 struct btrfs_fs_info *fs_info = root->fs_info;
2692 struct extent_buffer *b;
2697 int lowest_unlock = 1;
2698 /* everything at write_lock_level or lower must be write locked */
2699 int write_lock_level = 0;
2700 u8 lowest_level = 0;
2701 int min_write_lock_level;
2704 lowest_level = p->lowest_level;
2705 WARN_ON(lowest_level && ins_len > 0);
2706 WARN_ON(p->nodes[0] != NULL);
2707 BUG_ON(!cow && ins_len);
2712 /* when we are removing items, we might have to go up to level
2713 * two as we update tree pointers Make sure we keep write
2714 * for those levels as well
2716 write_lock_level = 2;
2717 } else if (ins_len > 0) {
2719 * for inserting items, make sure we have a write lock on
2720 * level 1 so we can update keys
2722 write_lock_level = 1;
2726 write_lock_level = -1;
2728 if (cow && (p->keep_locks || p->lowest_level))
2729 write_lock_level = BTRFS_MAX_LEVEL;
2731 min_write_lock_level = write_lock_level;
2735 b = btrfs_search_slot_get_root(root, p, write_lock_level);
2742 level = btrfs_header_level(b);
2745 * setup the path here so we can release it under lock
2746 * contention with the cow code
2749 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2752 * if we don't really need to cow this block
2753 * then we don't want to set the path blocking,
2754 * so we test it here
2756 if (!should_cow_block(trans, root, b)) {
2757 trans->dirty = true;
2762 * must have write locks on this node and the
2765 if (level > write_lock_level ||
2766 (level + 1 > write_lock_level &&
2767 level + 1 < BTRFS_MAX_LEVEL &&
2768 p->nodes[level + 1])) {
2769 write_lock_level = level + 1;
2770 btrfs_release_path(p);
2774 btrfs_set_path_blocking(p);
2776 err = btrfs_cow_block(trans, root, b, NULL, 0,
2779 err = btrfs_cow_block(trans, root, b,
2780 p->nodes[level + 1],
2781 p->slots[level + 1], &b);
2788 p->nodes[level] = b;
2790 * Leave path with blocking locks to avoid massive
2791 * lock context switch, this is made on purpose.
2795 * we have a lock on b and as long as we aren't changing
2796 * the tree, there is no way to for the items in b to change.
2797 * It is safe to drop the lock on our parent before we
2798 * go through the expensive btree search on b.
2800 * If we're inserting or deleting (ins_len != 0), then we might
2801 * be changing slot zero, which may require changing the parent.
2802 * So, we can't drop the lock until after we know which slot
2803 * we're operating on.
2805 if (!ins_len && !p->keep_locks) {
2808 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2809 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2814 ret = key_search(b, key, level, &prev_cmp, &slot);
2820 if (ret && slot > 0) {
2824 p->slots[level] = slot;
2825 err = setup_nodes_for_search(trans, root, p, b, level,
2826 ins_len, &write_lock_level);
2833 b = p->nodes[level];
2834 slot = p->slots[level];
2837 * slot 0 is special, if we change the key
2838 * we have to update the parent pointer
2839 * which means we must have a write lock
2842 if (slot == 0 && ins_len &&
2843 write_lock_level < level + 1) {
2844 write_lock_level = level + 1;
2845 btrfs_release_path(p);
2849 unlock_up(p, level, lowest_unlock,
2850 min_write_lock_level, &write_lock_level);
2852 if (level == lowest_level) {
2858 err = read_block_for_search(root, p, &b, level,
2867 if (!p->skip_locking) {
2868 level = btrfs_header_level(b);
2869 if (level <= write_lock_level) {
2870 err = btrfs_try_tree_write_lock(b);
2872 btrfs_set_path_blocking(p);
2875 p->locks[level] = BTRFS_WRITE_LOCK;
2877 err = btrfs_tree_read_lock_atomic(b);
2879 btrfs_set_path_blocking(p);
2880 btrfs_tree_read_lock(b);
2882 p->locks[level] = BTRFS_READ_LOCK;
2884 p->nodes[level] = b;
2887 p->slots[level] = slot;
2889 btrfs_leaf_free_space(fs_info, b) < ins_len) {
2890 if (write_lock_level < 1) {
2891 write_lock_level = 1;
2892 btrfs_release_path(p);
2896 btrfs_set_path_blocking(p);
2897 err = split_leaf(trans, root, key,
2898 p, ins_len, ret == 0);
2906 if (!p->search_for_split)
2907 unlock_up(p, level, lowest_unlock,
2908 min_write_lock_level, NULL);
2915 * we don't really know what they plan on doing with the path
2916 * from here on, so for now just mark it as blocking
2918 if (!p->leave_spinning)
2919 btrfs_set_path_blocking(p);
2920 if (ret < 0 && !p->skip_release_on_error)
2921 btrfs_release_path(p);
2926 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2927 * current state of the tree together with the operations recorded in the tree
2928 * modification log to search for the key in a previous version of this tree, as
2929 * denoted by the time_seq parameter.
2931 * Naturally, there is no support for insert, delete or cow operations.
2933 * The resulting path and return value will be set up as if we called
2934 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2936 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2937 struct btrfs_path *p, u64 time_seq)
2939 struct btrfs_fs_info *fs_info = root->fs_info;
2940 struct extent_buffer *b;
2945 int lowest_unlock = 1;
2946 u8 lowest_level = 0;
2949 lowest_level = p->lowest_level;
2950 WARN_ON(p->nodes[0] != NULL);
2952 if (p->search_commit_root) {
2954 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2958 b = get_old_root(root, time_seq);
2963 level = btrfs_header_level(b);
2964 p->locks[level] = BTRFS_READ_LOCK;
2967 level = btrfs_header_level(b);
2968 p->nodes[level] = b;
2971 * we have a lock on b and as long as we aren't changing
2972 * the tree, there is no way to for the items in b to change.
2973 * It is safe to drop the lock on our parent before we
2974 * go through the expensive btree search on b.
2976 btrfs_unlock_up_safe(p, level + 1);
2979 * Since we can unwind ebs we want to do a real search every
2983 ret = key_search(b, key, level, &prev_cmp, &slot);
2987 if (ret && slot > 0) {
2991 p->slots[level] = slot;
2992 unlock_up(p, level, lowest_unlock, 0, NULL);
2994 if (level == lowest_level) {
3000 err = read_block_for_search(root, p, &b, level,
3009 level = btrfs_header_level(b);
3010 err = btrfs_tree_read_lock_atomic(b);
3012 btrfs_set_path_blocking(p);
3013 btrfs_tree_read_lock(b);
3015 b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3020 p->locks[level] = BTRFS_READ_LOCK;
3021 p->nodes[level] = b;
3023 p->slots[level] = slot;
3024 unlock_up(p, level, lowest_unlock, 0, NULL);
3030 if (!p->leave_spinning)
3031 btrfs_set_path_blocking(p);
3033 btrfs_release_path(p);
3039 * helper to use instead of search slot if no exact match is needed but
3040 * instead the next or previous item should be returned.
3041 * When find_higher is true, the next higher item is returned, the next lower
3043 * When return_any and find_higher are both true, and no higher item is found,
3044 * return the next lower instead.
3045 * When return_any is true and find_higher is false, and no lower item is found,
3046 * return the next higher instead.
3047 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3050 int btrfs_search_slot_for_read(struct btrfs_root *root,
3051 const struct btrfs_key *key,
3052 struct btrfs_path *p, int find_higher,
3056 struct extent_buffer *leaf;
3059 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3063 * a return value of 1 means the path is at the position where the
3064 * item should be inserted. Normally this is the next bigger item,
3065 * but in case the previous item is the last in a leaf, path points
3066 * to the first free slot in the previous leaf, i.e. at an invalid
3072 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3073 ret = btrfs_next_leaf(root, p);
3079 * no higher item found, return the next
3084 btrfs_release_path(p);
3088 if (p->slots[0] == 0) {
3089 ret = btrfs_prev_leaf(root, p);
3094 if (p->slots[0] == btrfs_header_nritems(leaf))
3101 * no lower item found, return the next
3106 btrfs_release_path(p);
3116 * adjust the pointers going up the tree, starting at level
3117 * making sure the right key of each node is points to 'key'.
3118 * This is used after shifting pointers to the left, so it stops
3119 * fixing up pointers when a given leaf/node is not in slot 0 of the
3123 static void fixup_low_keys(struct btrfs_path *path,
3124 struct btrfs_disk_key *key, int level)
3127 struct extent_buffer *t;
3130 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3131 int tslot = path->slots[i];
3133 if (!path->nodes[i])
3136 ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3139 btrfs_set_node_key(t, key, tslot);
3140 btrfs_mark_buffer_dirty(path->nodes[i]);
3149 * This function isn't completely safe. It's the caller's responsibility
3150 * that the new key won't break the order
3152 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3153 struct btrfs_path *path,
3154 const struct btrfs_key *new_key)
3156 struct btrfs_disk_key disk_key;
3157 struct extent_buffer *eb;
3160 eb = path->nodes[0];
3161 slot = path->slots[0];
3163 btrfs_item_key(eb, &disk_key, slot - 1);
3164 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3166 if (slot < btrfs_header_nritems(eb) - 1) {
3167 btrfs_item_key(eb, &disk_key, slot + 1);
3168 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3171 btrfs_cpu_key_to_disk(&disk_key, new_key);
3172 btrfs_set_item_key(eb, &disk_key, slot);
3173 btrfs_mark_buffer_dirty(eb);
3175 fixup_low_keys(path, &disk_key, 1);
3179 * try to push data from one node into the next node left in the
3182 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3183 * error, and > 0 if there was no room in the left hand block.
3185 static int push_node_left(struct btrfs_trans_handle *trans,
3186 struct btrfs_fs_info *fs_info,
3187 struct extent_buffer *dst,
3188 struct extent_buffer *src, int empty)
3195 src_nritems = btrfs_header_nritems(src);
3196 dst_nritems = btrfs_header_nritems(dst);
3197 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3198 WARN_ON(btrfs_header_generation(src) != trans->transid);
3199 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3201 if (!empty && src_nritems <= 8)
3204 if (push_items <= 0)
3208 push_items = min(src_nritems, push_items);
3209 if (push_items < src_nritems) {
3210 /* leave at least 8 pointers in the node if
3211 * we aren't going to empty it
3213 if (src_nritems - push_items < 8) {
3214 if (push_items <= 8)
3220 push_items = min(src_nritems - 8, push_items);
3222 ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0,
3225 btrfs_abort_transaction(trans, ret);
3228 copy_extent_buffer(dst, src,
3229 btrfs_node_key_ptr_offset(dst_nritems),
3230 btrfs_node_key_ptr_offset(0),
3231 push_items * sizeof(struct btrfs_key_ptr));
3233 if (push_items < src_nritems) {
3235 * Don't call tree_mod_log_insert_move here, key removal was
3236 * already fully logged by tree_mod_log_eb_copy above.
3238 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3239 btrfs_node_key_ptr_offset(push_items),
3240 (src_nritems - push_items) *
3241 sizeof(struct btrfs_key_ptr));
3243 btrfs_set_header_nritems(src, src_nritems - push_items);
3244 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3245 btrfs_mark_buffer_dirty(src);
3246 btrfs_mark_buffer_dirty(dst);
3252 * try to push data from one node into the next node right in the
3255 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3256 * error, and > 0 if there was no room in the right hand block.
3258 * this will only push up to 1/2 the contents of the left node over
3260 static int balance_node_right(struct btrfs_trans_handle *trans,
3261 struct btrfs_fs_info *fs_info,
3262 struct extent_buffer *dst,
3263 struct extent_buffer *src)
3271 WARN_ON(btrfs_header_generation(src) != trans->transid);
3272 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3274 src_nritems = btrfs_header_nritems(src);
3275 dst_nritems = btrfs_header_nritems(dst);
3276 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3277 if (push_items <= 0)
3280 if (src_nritems < 4)
3283 max_push = src_nritems / 2 + 1;
3284 /* don't try to empty the node */
3285 if (max_push >= src_nritems)
3288 if (max_push < push_items)
3289 push_items = max_push;
3291 ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3293 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3294 btrfs_node_key_ptr_offset(0),
3296 sizeof(struct btrfs_key_ptr));
3298 ret = tree_mod_log_eb_copy(fs_info, dst, src, 0,
3299 src_nritems - push_items, push_items);
3301 btrfs_abort_transaction(trans, ret);
3304 copy_extent_buffer(dst, src,
3305 btrfs_node_key_ptr_offset(0),
3306 btrfs_node_key_ptr_offset(src_nritems - push_items),
3307 push_items * sizeof(struct btrfs_key_ptr));
3309 btrfs_set_header_nritems(src, src_nritems - push_items);
3310 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3312 btrfs_mark_buffer_dirty(src);
3313 btrfs_mark_buffer_dirty(dst);
3319 * helper function to insert a new root level in the tree.
3320 * A new node is allocated, and a single item is inserted to
3321 * point to the existing root
3323 * returns zero on success or < 0 on failure.
3325 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3326 struct btrfs_root *root,
3327 struct btrfs_path *path, int level)
3329 struct btrfs_fs_info *fs_info = root->fs_info;
3331 struct extent_buffer *lower;
3332 struct extent_buffer *c;
3333 struct extent_buffer *old;
3334 struct btrfs_disk_key lower_key;
3337 BUG_ON(path->nodes[level]);
3338 BUG_ON(path->nodes[level-1] != root->node);
3340 lower = path->nodes[level-1];
3342 btrfs_item_key(lower, &lower_key, 0);
3344 btrfs_node_key(lower, &lower_key, 0);
3346 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3347 &lower_key, level, root->node->start, 0);
3351 root_add_used(root, fs_info->nodesize);
3353 btrfs_set_header_nritems(c, 1);
3354 btrfs_set_node_key(c, &lower_key, 0);
3355 btrfs_set_node_blockptr(c, 0, lower->start);
3356 lower_gen = btrfs_header_generation(lower);
3357 WARN_ON(lower_gen != trans->transid);
3359 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3361 btrfs_mark_buffer_dirty(c);
3364 ret = tree_mod_log_insert_root(root->node, c, 0);
3366 rcu_assign_pointer(root->node, c);
3368 /* the super has an extra ref to root->node */
3369 free_extent_buffer(old);
3371 add_root_to_dirty_list(root);
3372 extent_buffer_get(c);
3373 path->nodes[level] = c;
3374 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3375 path->slots[level] = 0;
3380 * worker function to insert a single pointer in a node.
3381 * the node should have enough room for the pointer already
3383 * slot and level indicate where you want the key to go, and
3384 * blocknr is the block the key points to.
3386 static void insert_ptr(struct btrfs_trans_handle *trans,
3387 struct btrfs_fs_info *fs_info, struct btrfs_path *path,
3388 struct btrfs_disk_key *key, u64 bytenr,
3389 int slot, int level)
3391 struct extent_buffer *lower;
3395 BUG_ON(!path->nodes[level]);
3396 btrfs_assert_tree_locked(path->nodes[level]);
3397 lower = path->nodes[level];
3398 nritems = btrfs_header_nritems(lower);
3399 BUG_ON(slot > nritems);
3400 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info));
3401 if (slot != nritems) {
3403 ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3407 memmove_extent_buffer(lower,
3408 btrfs_node_key_ptr_offset(slot + 1),
3409 btrfs_node_key_ptr_offset(slot),
3410 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3413 ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3417 btrfs_set_node_key(lower, key, slot);
3418 btrfs_set_node_blockptr(lower, slot, bytenr);
3419 WARN_ON(trans->transid == 0);
3420 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3421 btrfs_set_header_nritems(lower, nritems + 1);
3422 btrfs_mark_buffer_dirty(lower);
3426 * split the node at the specified level in path in two.
3427 * The path is corrected to point to the appropriate node after the split
3429 * Before splitting this tries to make some room in the node by pushing
3430 * left and right, if either one works, it returns right away.
3432 * returns 0 on success and < 0 on failure
3434 static noinline int split_node(struct btrfs_trans_handle *trans,
3435 struct btrfs_root *root,
3436 struct btrfs_path *path, int level)
3438 struct btrfs_fs_info *fs_info = root->fs_info;
3439 struct extent_buffer *c;
3440 struct extent_buffer *split;
3441 struct btrfs_disk_key disk_key;
3446 c = path->nodes[level];
3447 WARN_ON(btrfs_header_generation(c) != trans->transid);
3448 if (c == root->node) {
3450 * trying to split the root, lets make a new one
3452 * tree mod log: We don't log_removal old root in
3453 * insert_new_root, because that root buffer will be kept as a
3454 * normal node. We are going to log removal of half of the
3455 * elements below with tree_mod_log_eb_copy. We're holding a
3456 * tree lock on the buffer, which is why we cannot race with
3457 * other tree_mod_log users.
3459 ret = insert_new_root(trans, root, path, level + 1);
3463 ret = push_nodes_for_insert(trans, root, path, level);
3464 c = path->nodes[level];
3465 if (!ret && btrfs_header_nritems(c) <
3466 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3472 c_nritems = btrfs_header_nritems(c);
3473 mid = (c_nritems + 1) / 2;
3474 btrfs_node_key(c, &disk_key, mid);
3476 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3477 &disk_key, level, c->start, 0);
3479 return PTR_ERR(split);
3481 root_add_used(root, fs_info->nodesize);
3482 ASSERT(btrfs_header_level(c) == level);
3484 ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid);
3486 btrfs_abort_transaction(trans, ret);
3489 copy_extent_buffer(split, c,
3490 btrfs_node_key_ptr_offset(0),
3491 btrfs_node_key_ptr_offset(mid),
3492 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3493 btrfs_set_header_nritems(split, c_nritems - mid);
3494 btrfs_set_header_nritems(c, mid);
3497 btrfs_mark_buffer_dirty(c);
3498 btrfs_mark_buffer_dirty(split);
3500 insert_ptr(trans, fs_info, path, &disk_key, split->start,
3501 path->slots[level + 1] + 1, level + 1);
3503 if (path->slots[level] >= mid) {
3504 path->slots[level] -= mid;
3505 btrfs_tree_unlock(c);
3506 free_extent_buffer(c);
3507 path->nodes[level] = split;
3508 path->slots[level + 1] += 1;
3510 btrfs_tree_unlock(split);
3511 free_extent_buffer(split);
3517 * how many bytes are required to store the items in a leaf. start
3518 * and nr indicate which items in the leaf to check. This totals up the
3519 * space used both by the item structs and the item data
3521 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3523 struct btrfs_item *start_item;
3524 struct btrfs_item *end_item;
3525 struct btrfs_map_token token;
3527 int nritems = btrfs_header_nritems(l);
3528 int end = min(nritems, start + nr) - 1;
3532 btrfs_init_map_token(&token);
3533 start_item = btrfs_item_nr(start);
3534 end_item = btrfs_item_nr(end);
3535 data_len = btrfs_token_item_offset(l, start_item, &token) +
3536 btrfs_token_item_size(l, start_item, &token);
3537 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3538 data_len += sizeof(struct btrfs_item) * nr;
3539 WARN_ON(data_len < 0);
3544 * The space between the end of the leaf items and
3545 * the start of the leaf data. IOW, how much room
3546 * the leaf has left for both items and data
3548 noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info,
3549 struct extent_buffer *leaf)
3551 int nritems = btrfs_header_nritems(leaf);
3554 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3557 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3559 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3560 leaf_space_used(leaf, 0, nritems), nritems);
3566 * min slot controls the lowest index we're willing to push to the
3567 * right. We'll push up to and including min_slot, but no lower
3569 static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info,
3570 struct btrfs_path *path,
3571 int data_size, int empty,
3572 struct extent_buffer *right,
3573 int free_space, u32 left_nritems,
3576 struct extent_buffer *left = path->nodes[0];
3577 struct extent_buffer *upper = path->nodes[1];
3578 struct btrfs_map_token token;
3579 struct btrfs_disk_key disk_key;
3584 struct btrfs_item *item;
3590 btrfs_init_map_token(&token);
3595 nr = max_t(u32, 1, min_slot);
3597 if (path->slots[0] >= left_nritems)
3598 push_space += data_size;
3600 slot = path->slots[1];
3601 i = left_nritems - 1;
3603 item = btrfs_item_nr(i);
3605 if (!empty && push_items > 0) {
3606 if (path->slots[0] > i)
3608 if (path->slots[0] == i) {
3609 int space = btrfs_leaf_free_space(fs_info, left);
3610 if (space + push_space * 2 > free_space)
3615 if (path->slots[0] == i)
3616 push_space += data_size;
3618 this_item_size = btrfs_item_size(left, item);
3619 if (this_item_size + sizeof(*item) + push_space > free_space)
3623 push_space += this_item_size + sizeof(*item);
3629 if (push_items == 0)
3632 WARN_ON(!empty && push_items == left_nritems);
3634 /* push left to right */
3635 right_nritems = btrfs_header_nritems(right);
3637 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3638 push_space -= leaf_data_end(fs_info, left);
3640 /* make room in the right data area */
3641 data_end = leaf_data_end(fs_info, right);
3642 memmove_extent_buffer(right,
3643 BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3644 BTRFS_LEAF_DATA_OFFSET + data_end,
3645 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3647 /* copy from the left data area */
3648 copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3649 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3650 BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left),
3653 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3654 btrfs_item_nr_offset(0),
3655 right_nritems * sizeof(struct btrfs_item));
3657 /* copy the items from left to right */
3658 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3659 btrfs_item_nr_offset(left_nritems - push_items),
3660 push_items * sizeof(struct btrfs_item));
3662 /* update the item pointers */
3663 right_nritems += push_items;
3664 btrfs_set_header_nritems(right, right_nritems);
3665 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3666 for (i = 0; i < right_nritems; i++) {
3667 item = btrfs_item_nr(i);
3668 push_space -= btrfs_token_item_size(right, item, &token);
3669 btrfs_set_token_item_offset(right, item, push_space, &token);
3672 left_nritems -= push_items;
3673 btrfs_set_header_nritems(left, left_nritems);
3676 btrfs_mark_buffer_dirty(left);
3678 clean_tree_block(fs_info, left);
3680 btrfs_mark_buffer_dirty(right);
3682 btrfs_item_key(right, &disk_key, 0);
3683 btrfs_set_node_key(upper, &disk_key, slot + 1);
3684 btrfs_mark_buffer_dirty(upper);
3686 /* then fixup the leaf pointer in the path */
3687 if (path->slots[0] >= left_nritems) {
3688 path->slots[0] -= left_nritems;
3689 if (btrfs_header_nritems(path->nodes[0]) == 0)
3690 clean_tree_block(fs_info, path->nodes[0]);
3691 btrfs_tree_unlock(path->nodes[0]);
3692 free_extent_buffer(path->nodes[0]);
3693 path->nodes[0] = right;
3694 path->slots[1] += 1;
3696 btrfs_tree_unlock(right);
3697 free_extent_buffer(right);
3702 btrfs_tree_unlock(right);
3703 free_extent_buffer(right);
3708 * push some data in the path leaf to the right, trying to free up at
3709 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3711 * returns 1 if the push failed because the other node didn't have enough
3712 * room, 0 if everything worked out and < 0 if there were major errors.
3714 * this will push starting from min_slot to the end of the leaf. It won't
3715 * push any slot lower than min_slot
3717 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3718 *root, struct btrfs_path *path,
3719 int min_data_size, int data_size,
3720 int empty, u32 min_slot)
3722 struct btrfs_fs_info *fs_info = root->fs_info;
3723 struct extent_buffer *left = path->nodes[0];
3724 struct extent_buffer *right;
3725 struct extent_buffer *upper;
3731 if (!path->nodes[1])
3734 slot = path->slots[1];
3735 upper = path->nodes[1];
3736 if (slot >= btrfs_header_nritems(upper) - 1)
3739 btrfs_assert_tree_locked(path->nodes[1]);
3741 right = read_node_slot(fs_info, upper, slot + 1);
3743 * slot + 1 is not valid or we fail to read the right node,
3744 * no big deal, just return.
3749 btrfs_tree_lock(right);
3750 btrfs_set_lock_blocking(right);
3752 free_space = btrfs_leaf_free_space(fs_info, right);
3753 if (free_space < data_size)
3756 /* cow and double check */
3757 ret = btrfs_cow_block(trans, root, right, upper,
3762 free_space = btrfs_leaf_free_space(fs_info, right);
3763 if (free_space < data_size)
3766 left_nritems = btrfs_header_nritems(left);
3767 if (left_nritems == 0)
3770 if (path->slots[0] == left_nritems && !empty) {
3771 /* Key greater than all keys in the leaf, right neighbor has
3772 * enough room for it and we're not emptying our leaf to delete
3773 * it, therefore use right neighbor to insert the new item and
3774 * no need to touch/dirty our left leaf. */
3775 btrfs_tree_unlock(left);
3776 free_extent_buffer(left);
3777 path->nodes[0] = right;
3783 return __push_leaf_right(fs_info, path, min_data_size, empty,
3784 right, free_space, left_nritems, min_slot);
3786 btrfs_tree_unlock(right);
3787 free_extent_buffer(right);
3792 * push some data in the path leaf to the left, trying to free up at
3793 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3795 * max_slot can put a limit on how far into the leaf we'll push items. The
3796 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3799 static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info,
3800 struct btrfs_path *path, int data_size,
3801 int empty, struct extent_buffer *left,
3802 int free_space, u32 right_nritems,
3805 struct btrfs_disk_key disk_key;
3806 struct extent_buffer *right = path->nodes[0];
3810 struct btrfs_item *item;
3811 u32 old_left_nritems;
3815 u32 old_left_item_size;
3816 struct btrfs_map_token token;
3818 btrfs_init_map_token(&token);
3821 nr = min(right_nritems, max_slot);
3823 nr = min(right_nritems - 1, max_slot);
3825 for (i = 0; i < nr; i++) {
3826 item = btrfs_item_nr(i);
3828 if (!empty && push_items > 0) {
3829 if (path->slots[0] < i)
3831 if (path->slots[0] == i) {
3832 int space = btrfs_leaf_free_space(fs_info, right);
3833 if (space + push_space * 2 > free_space)
3838 if (path->slots[0] == i)
3839 push_space += data_size;
3841 this_item_size = btrfs_item_size(right, item);
3842 if (this_item_size + sizeof(*item) + push_space > free_space)
3846 push_space += this_item_size + sizeof(*item);
3849 if (push_items == 0) {
3853 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3855 /* push data from right to left */
3856 copy_extent_buffer(left, right,
3857 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3858 btrfs_item_nr_offset(0),
3859 push_items * sizeof(struct btrfs_item));
3861 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3862 btrfs_item_offset_nr(right, push_items - 1);
3864 copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3865 leaf_data_end(fs_info, left) - push_space,
3866 BTRFS_LEAF_DATA_OFFSET +
3867 btrfs_item_offset_nr(right, push_items - 1),
3869 old_left_nritems = btrfs_header_nritems(left);
3870 BUG_ON(old_left_nritems <= 0);
3872 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3873 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3876 item = btrfs_item_nr(i);
3878 ioff = btrfs_token_item_offset(left, item, &token);
3879 btrfs_set_token_item_offset(left, item,
3880 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3883 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3885 /* fixup right node */
3886 if (push_items > right_nritems)
3887 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3890 if (push_items < right_nritems) {
3891 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3892 leaf_data_end(fs_info, right);
3893 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3894 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3895 BTRFS_LEAF_DATA_OFFSET +
3896 leaf_data_end(fs_info, right), push_space);
3898 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3899 btrfs_item_nr_offset(push_items),
3900 (btrfs_header_nritems(right) - push_items) *
3901 sizeof(struct btrfs_item));
3903 right_nritems -= push_items;
3904 btrfs_set_header_nritems(right, right_nritems);
3905 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3906 for (i = 0; i < right_nritems; i++) {
3907 item = btrfs_item_nr(i);
3909 push_space = push_space - btrfs_token_item_size(right,
3911 btrfs_set_token_item_offset(right, item, push_space, &token);
3914 btrfs_mark_buffer_dirty(left);
3916 btrfs_mark_buffer_dirty(right);
3918 clean_tree_block(fs_info, right);
3920 btrfs_item_key(right, &disk_key, 0);
3921 fixup_low_keys(path, &disk_key, 1);
3923 /* then fixup the leaf pointer in the path */
3924 if (path->slots[0] < push_items) {
3925 path->slots[0] += old_left_nritems;
3926 btrfs_tree_unlock(path->nodes[0]);
3927 free_extent_buffer(path->nodes[0]);
3928 path->nodes[0] = left;
3929 path->slots[1] -= 1;
3931 btrfs_tree_unlock(left);
3932 free_extent_buffer(left);
3933 path->slots[0] -= push_items;
3935 BUG_ON(path->slots[0] < 0);
3938 btrfs_tree_unlock(left);
3939 free_extent_buffer(left);
3944 * push some data in the path leaf to the left, trying to free up at
3945 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3947 * max_slot can put a limit on how far into the leaf we'll push items. The
3948 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3951 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3952 *root, struct btrfs_path *path, int min_data_size,
3953 int data_size, int empty, u32 max_slot)
3955 struct btrfs_fs_info *fs_info = root->fs_info;
3956 struct extent_buffer *right = path->nodes[0];
3957 struct extent_buffer *left;
3963 slot = path->slots[1];
3966 if (!path->nodes[1])
3969 right_nritems = btrfs_header_nritems(right);
3970 if (right_nritems == 0)
3973 btrfs_assert_tree_locked(path->nodes[1]);
3975 left = read_node_slot(fs_info, path->nodes[1], slot - 1);
3977 * slot - 1 is not valid or we fail to read the left node,
3978 * no big deal, just return.
3983 btrfs_tree_lock(left);
3984 btrfs_set_lock_blocking(left);
3986 free_space = btrfs_leaf_free_space(fs_info, left);
3987 if (free_space < data_size) {
3992 /* cow and double check */
3993 ret = btrfs_cow_block(trans, root, left,
3994 path->nodes[1], slot - 1, &left);
3996 /* we hit -ENOSPC, but it isn't fatal here */
4002 free_space = btrfs_leaf_free_space(fs_info, left);
4003 if (free_space < data_size) {
4008 return __push_leaf_left(fs_info, path, min_data_size,
4009 empty, left, free_space, right_nritems,
4012 btrfs_tree_unlock(left);
4013 free_extent_buffer(left);
4018 * split the path's leaf in two, making sure there is at least data_size
4019 * available for the resulting leaf level of the path.
4021 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4022 struct btrfs_fs_info *fs_info,
4023 struct btrfs_path *path,
4024 struct extent_buffer *l,
4025 struct extent_buffer *right,
4026 int slot, int mid, int nritems)
4031 struct btrfs_disk_key disk_key;
4032 struct btrfs_map_token token;
4034 btrfs_init_map_token(&token);
4036 nritems = nritems - mid;
4037 btrfs_set_header_nritems(right, nritems);
4038 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l);
4040 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4041 btrfs_item_nr_offset(mid),
4042 nritems * sizeof(struct btrfs_item));
4044 copy_extent_buffer(right, l,
4045 BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4046 data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4047 leaf_data_end(fs_info, l), data_copy_size);
4049 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4051 for (i = 0; i < nritems; i++) {
4052 struct btrfs_item *item = btrfs_item_nr(i);
4055 ioff = btrfs_token_item_offset(right, item, &token);
4056 btrfs_set_token_item_offset(right, item,
4057 ioff + rt_data_off, &token);
4060 btrfs_set_header_nritems(l, mid);
4061 btrfs_item_key(right, &disk_key, 0);
4062 insert_ptr(trans, fs_info, path, &disk_key, right->start,
4063 path->slots[1] + 1, 1);
4065 btrfs_mark_buffer_dirty(right);
4066 btrfs_mark_buffer_dirty(l);
4067 BUG_ON(path->slots[0] != slot);
4070 btrfs_tree_unlock(path->nodes[0]);
4071 free_extent_buffer(path->nodes[0]);
4072 path->nodes[0] = right;
4073 path->slots[0] -= mid;
4074 path->slots[1] += 1;
4076 btrfs_tree_unlock(right);
4077 free_extent_buffer(right);
4080 BUG_ON(path->slots[0] < 0);
4084 * double splits happen when we need to insert a big item in the middle
4085 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4086 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4089 * We avoid this by trying to push the items on either side of our target
4090 * into the adjacent leaves. If all goes well we can avoid the double split
4093 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4094 struct btrfs_root *root,
4095 struct btrfs_path *path,
4098 struct btrfs_fs_info *fs_info = root->fs_info;
4103 int space_needed = data_size;
4105 slot = path->slots[0];
4106 if (slot < btrfs_header_nritems(path->nodes[0]))
4107 space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4110 * try to push all the items after our slot into the
4113 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4120 nritems = btrfs_header_nritems(path->nodes[0]);
4122 * our goal is to get our slot at the start or end of a leaf. If
4123 * we've done so we're done
4125 if (path->slots[0] == 0 || path->slots[0] == nritems)
4128 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4131 /* try to push all the items before our slot into the next leaf */
4132 slot = path->slots[0];
4133 space_needed = data_size;
4135 space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4136 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4149 * split the path's leaf in two, making sure there is at least data_size
4150 * available for the resulting leaf level of the path.
4152 * returns 0 if all went well and < 0 on failure.
4154 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4155 struct btrfs_root *root,
4156 const struct btrfs_key *ins_key,
4157 struct btrfs_path *path, int data_size,
4160 struct btrfs_disk_key disk_key;
4161 struct extent_buffer *l;
4165 struct extent_buffer *right;
4166 struct btrfs_fs_info *fs_info = root->fs_info;
4170 int num_doubles = 0;
4171 int tried_avoid_double = 0;
4174 slot = path->slots[0];
4175 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4176 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4179 /* first try to make some room by pushing left and right */
4180 if (data_size && path->nodes[1]) {
4181 int space_needed = data_size;
4183 if (slot < btrfs_header_nritems(l))
4184 space_needed -= btrfs_leaf_free_space(fs_info, l);
4186 wret = push_leaf_right(trans, root, path, space_needed,
4187 space_needed, 0, 0);
4191 space_needed = data_size;
4193 space_needed -= btrfs_leaf_free_space(fs_info,
4195 wret = push_leaf_left(trans, root, path, space_needed,
4196 space_needed, 0, (u32)-1);
4202 /* did the pushes work? */
4203 if (btrfs_leaf_free_space(fs_info, l) >= data_size)
4207 if (!path->nodes[1]) {
4208 ret = insert_new_root(trans, root, path, 1);
4215 slot = path->slots[0];
4216 nritems = btrfs_header_nritems(l);
4217 mid = (nritems + 1) / 2;
4221 leaf_space_used(l, mid, nritems - mid) + data_size >
4222 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4223 if (slot >= nritems) {
4227 if (mid != nritems &&
4228 leaf_space_used(l, mid, nritems - mid) +
4229 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4230 if (data_size && !tried_avoid_double)
4231 goto push_for_double;
4237 if (leaf_space_used(l, 0, mid) + data_size >
4238 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4239 if (!extend && data_size && slot == 0) {
4241 } else if ((extend || !data_size) && slot == 0) {
4245 if (mid != nritems &&
4246 leaf_space_used(l, mid, nritems - mid) +
4247 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4248 if (data_size && !tried_avoid_double)
4249 goto push_for_double;
4257 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4259 btrfs_item_key(l, &disk_key, mid);
4261 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4262 &disk_key, 0, l->start, 0);
4264 return PTR_ERR(right);
4266 root_add_used(root, fs_info->nodesize);
4270 btrfs_set_header_nritems(right, 0);
4271 insert_ptr(trans, fs_info, path, &disk_key,
4272 right->start, path->slots[1] + 1, 1);
4273 btrfs_tree_unlock(path->nodes[0]);
4274 free_extent_buffer(path->nodes[0]);
4275 path->nodes[0] = right;
4277 path->slots[1] += 1;
4279 btrfs_set_header_nritems(right, 0);
4280 insert_ptr(trans, fs_info, path, &disk_key,
4281 right->start, path->slots[1], 1);
4282 btrfs_tree_unlock(path->nodes[0]);
4283 free_extent_buffer(path->nodes[0]);
4284 path->nodes[0] = right;
4286 if (path->slots[1] == 0)
4287 fixup_low_keys(path, &disk_key, 1);
4290 * We create a new leaf 'right' for the required ins_len and
4291 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4292 * the content of ins_len to 'right'.
4297 copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems);
4300 BUG_ON(num_doubles != 0);
4308 push_for_double_split(trans, root, path, data_size);
4309 tried_avoid_double = 1;
4310 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4315 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4316 struct btrfs_root *root,
4317 struct btrfs_path *path, int ins_len)
4319 struct btrfs_fs_info *fs_info = root->fs_info;
4320 struct btrfs_key key;
4321 struct extent_buffer *leaf;
4322 struct btrfs_file_extent_item *fi;
4327 leaf = path->nodes[0];
4328 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4330 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4331 key.type != BTRFS_EXTENT_CSUM_KEY);
4333 if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len)
4336 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4337 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4338 fi = btrfs_item_ptr(leaf, path->slots[0],
4339 struct btrfs_file_extent_item);
4340 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4342 btrfs_release_path(path);
4344 path->keep_locks = 1;
4345 path->search_for_split = 1;
4346 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4347 path->search_for_split = 0;
4354 leaf = path->nodes[0];
4355 /* if our item isn't there, return now */
4356 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4359 /* the leaf has changed, it now has room. return now */
4360 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len)
4363 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4364 fi = btrfs_item_ptr(leaf, path->slots[0],
4365 struct btrfs_file_extent_item);
4366 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4370 btrfs_set_path_blocking(path);
4371 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4375 path->keep_locks = 0;
4376 btrfs_unlock_up_safe(path, 1);
4379 path->keep_locks = 0;
4383 static noinline int split_item(struct btrfs_fs_info *fs_info,
4384 struct btrfs_path *path,
4385 const struct btrfs_key *new_key,
4386 unsigned long split_offset)
4388 struct extent_buffer *leaf;
4389 struct btrfs_item *item;
4390 struct btrfs_item *new_item;
4396 struct btrfs_disk_key disk_key;
4398 leaf = path->nodes[0];
4399 BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item));
4401 btrfs_set_path_blocking(path);
4403 item = btrfs_item_nr(path->slots[0]);
4404 orig_offset = btrfs_item_offset(leaf, item);
4405 item_size = btrfs_item_size(leaf, item);
4407 buf = kmalloc(item_size, GFP_NOFS);
4411 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4412 path->slots[0]), item_size);
4414 slot = path->slots[0] + 1;
4415 nritems = btrfs_header_nritems(leaf);
4416 if (slot != nritems) {
4417 /* shift the items */
4418 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4419 btrfs_item_nr_offset(slot),
4420 (nritems - slot) * sizeof(struct btrfs_item));
4423 btrfs_cpu_key_to_disk(&disk_key, new_key);
4424 btrfs_set_item_key(leaf, &disk_key, slot);
4426 new_item = btrfs_item_nr(slot);
4428 btrfs_set_item_offset(leaf, new_item, orig_offset);
4429 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4431 btrfs_set_item_offset(leaf, item,
4432 orig_offset + item_size - split_offset);
4433 btrfs_set_item_size(leaf, item, split_offset);
4435 btrfs_set_header_nritems(leaf, nritems + 1);
4437 /* write the data for the start of the original item */
4438 write_extent_buffer(leaf, buf,
4439 btrfs_item_ptr_offset(leaf, path->slots[0]),
4442 /* write the data for the new item */
4443 write_extent_buffer(leaf, buf + split_offset,
4444 btrfs_item_ptr_offset(leaf, slot),
4445 item_size - split_offset);
4446 btrfs_mark_buffer_dirty(leaf);
4448 BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0);
4454 * This function splits a single item into two items,
4455 * giving 'new_key' to the new item and splitting the
4456 * old one at split_offset (from the start of the item).
4458 * The path may be released by this operation. After
4459 * the split, the path is pointing to the old item. The
4460 * new item is going to be in the same node as the old one.
4462 * Note, the item being split must be smaller enough to live alone on
4463 * a tree block with room for one extra struct btrfs_item
4465 * This allows us to split the item in place, keeping a lock on the
4466 * leaf the entire time.
4468 int btrfs_split_item(struct btrfs_trans_handle *trans,
4469 struct btrfs_root *root,
4470 struct btrfs_path *path,
4471 const struct btrfs_key *new_key,
4472 unsigned long split_offset)
4475 ret = setup_leaf_for_split(trans, root, path,
4476 sizeof(struct btrfs_item));
4480 ret = split_item(root->fs_info, path, new_key, split_offset);
4485 * This function duplicate a item, giving 'new_key' to the new item.
4486 * It guarantees both items live in the same tree leaf and the new item
4487 * is contiguous with the original item.
4489 * This allows us to split file extent in place, keeping a lock on the
4490 * leaf the entire time.
4492 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4493 struct btrfs_root *root,
4494 struct btrfs_path *path,
4495 const struct btrfs_key *new_key)
4497 struct extent_buffer *leaf;
4501 leaf = path->nodes[0];
4502 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4503 ret = setup_leaf_for_split(trans, root, path,
4504 item_size + sizeof(struct btrfs_item));
4509 setup_items_for_insert(root, path, new_key, &item_size,
4510 item_size, item_size +
4511 sizeof(struct btrfs_item), 1);
4512 leaf = path->nodes[0];
4513 memcpy_extent_buffer(leaf,
4514 btrfs_item_ptr_offset(leaf, path->slots[0]),
4515 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4521 * make the item pointed to by the path smaller. new_size indicates
4522 * how small to make it, and from_end tells us if we just chop bytes
4523 * off the end of the item or if we shift the item to chop bytes off
4526 void btrfs_truncate_item(struct btrfs_fs_info *fs_info,
4527 struct btrfs_path *path, u32 new_size, int from_end)
4530 struct extent_buffer *leaf;
4531 struct btrfs_item *item;
4533 unsigned int data_end;
4534 unsigned int old_data_start;
4535 unsigned int old_size;
4536 unsigned int size_diff;
4538 struct btrfs_map_token token;
4540 btrfs_init_map_token(&token);
4542 leaf = path->nodes[0];
4543 slot = path->slots[0];
4545 old_size = btrfs_item_size_nr(leaf, slot);
4546 if (old_size == new_size)
4549 nritems = btrfs_header_nritems(leaf);
4550 data_end = leaf_data_end(fs_info, leaf);
4552 old_data_start = btrfs_item_offset_nr(leaf, slot);
4554 size_diff = old_size - new_size;
4557 BUG_ON(slot >= nritems);
4560 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4562 /* first correct the data pointers */
4563 for (i = slot; i < nritems; i++) {
4565 item = btrfs_item_nr(i);
4567 ioff = btrfs_token_item_offset(leaf, item, &token);
4568 btrfs_set_token_item_offset(leaf, item,
4569 ioff + size_diff, &token);
4572 /* shift the data */
4574 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4575 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4576 data_end, old_data_start + new_size - data_end);
4578 struct btrfs_disk_key disk_key;
4581 btrfs_item_key(leaf, &disk_key, slot);
4583 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4585 struct btrfs_file_extent_item *fi;
4587 fi = btrfs_item_ptr(leaf, slot,
4588 struct btrfs_file_extent_item);
4589 fi = (struct btrfs_file_extent_item *)(
4590 (unsigned long)fi - size_diff);
4592 if (btrfs_file_extent_type(leaf, fi) ==
4593 BTRFS_FILE_EXTENT_INLINE) {
4594 ptr = btrfs_item_ptr_offset(leaf, slot);
4595 memmove_extent_buffer(leaf, ptr,
4597 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4601 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4602 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4603 data_end, old_data_start - data_end);
4605 offset = btrfs_disk_key_offset(&disk_key);
4606 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4607 btrfs_set_item_key(leaf, &disk_key, slot);
4609 fixup_low_keys(path, &disk_key, 1);
4612 item = btrfs_item_nr(slot);
4613 btrfs_set_item_size(leaf, item, new_size);
4614 btrfs_mark_buffer_dirty(leaf);
4616 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4617 btrfs_print_leaf(leaf);
4623 * make the item pointed to by the path bigger, data_size is the added size.
4625 void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
4629 struct extent_buffer *leaf;
4630 struct btrfs_item *item;
4632 unsigned int data_end;
4633 unsigned int old_data;
4634 unsigned int old_size;
4636 struct btrfs_map_token token;
4638 btrfs_init_map_token(&token);
4640 leaf = path->nodes[0];
4642 nritems = btrfs_header_nritems(leaf);
4643 data_end = leaf_data_end(fs_info, leaf);
4645 if (btrfs_leaf_free_space(fs_info, leaf) < data_size) {
4646 btrfs_print_leaf(leaf);
4649 slot = path->slots[0];
4650 old_data = btrfs_item_end_nr(leaf, slot);
4653 if (slot >= nritems) {
4654 btrfs_print_leaf(leaf);
4655 btrfs_crit(fs_info, "slot %d too large, nritems %d",
4661 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4663 /* first correct the data pointers */
4664 for (i = slot; i < nritems; i++) {
4666 item = btrfs_item_nr(i);
4668 ioff = btrfs_token_item_offset(leaf, item, &token);
4669 btrfs_set_token_item_offset(leaf, item,
4670 ioff - data_size, &token);
4673 /* shift the data */
4674 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4675 data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4676 data_end, old_data - data_end);
4678 data_end = old_data;
4679 old_size = btrfs_item_size_nr(leaf, slot);
4680 item = btrfs_item_nr(slot);
4681 btrfs_set_item_size(leaf, item, old_size + data_size);
4682 btrfs_mark_buffer_dirty(leaf);
4684 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4685 btrfs_print_leaf(leaf);
4691 * this is a helper for btrfs_insert_empty_items, the main goal here is
4692 * to save stack depth by doing the bulk of the work in a function
4693 * that doesn't call btrfs_search_slot
4695 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4696 const struct btrfs_key *cpu_key, u32 *data_size,
4697 u32 total_data, u32 total_size, int nr)
4699 struct btrfs_fs_info *fs_info = root->fs_info;
4700 struct btrfs_item *item;
4703 unsigned int data_end;
4704 struct btrfs_disk_key disk_key;
4705 struct extent_buffer *leaf;
4707 struct btrfs_map_token token;
4709 if (path->slots[0] == 0) {
4710 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4711 fixup_low_keys(path, &disk_key, 1);
4713 btrfs_unlock_up_safe(path, 1);
4715 btrfs_init_map_token(&token);
4717 leaf = path->nodes[0];
4718 slot = path->slots[0];
4720 nritems = btrfs_header_nritems(leaf);
4721 data_end = leaf_data_end(fs_info, leaf);
4723 if (btrfs_leaf_free_space(fs_info, leaf) < total_size) {
4724 btrfs_print_leaf(leaf);
4725 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4726 total_size, btrfs_leaf_free_space(fs_info, leaf));
4730 if (slot != nritems) {
4731 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4733 if (old_data < data_end) {
4734 btrfs_print_leaf(leaf);
4735 btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4736 slot, old_data, data_end);
4740 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4742 /* first correct the data pointers */
4743 for (i = slot; i < nritems; i++) {
4746 item = btrfs_item_nr(i);
4747 ioff = btrfs_token_item_offset(leaf, item, &token);
4748 btrfs_set_token_item_offset(leaf, item,
4749 ioff - total_data, &token);
4751 /* shift the items */
4752 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4753 btrfs_item_nr_offset(slot),
4754 (nritems - slot) * sizeof(struct btrfs_item));
4756 /* shift the data */
4757 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4758 data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4759 data_end, old_data - data_end);
4760 data_end = old_data;
4763 /* setup the item for the new data */
4764 for (i = 0; i < nr; i++) {
4765 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4766 btrfs_set_item_key(leaf, &disk_key, slot + i);
4767 item = btrfs_item_nr(slot + i);
4768 btrfs_set_token_item_offset(leaf, item,
4769 data_end - data_size[i], &token);
4770 data_end -= data_size[i];
4771 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4774 btrfs_set_header_nritems(leaf, nritems + nr);
4775 btrfs_mark_buffer_dirty(leaf);
4777 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4778 btrfs_print_leaf(leaf);
4784 * Given a key and some data, insert items into the tree.
4785 * This does all the path init required, making room in the tree if needed.
4787 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4788 struct btrfs_root *root,
4789 struct btrfs_path *path,
4790 const struct btrfs_key *cpu_key, u32 *data_size,
4799 for (i = 0; i < nr; i++)
4800 total_data += data_size[i];
4802 total_size = total_data + (nr * sizeof(struct btrfs_item));
4803 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4809 slot = path->slots[0];
4812 setup_items_for_insert(root, path, cpu_key, data_size,
4813 total_data, total_size, nr);
4818 * Given a key and some data, insert an item into the tree.
4819 * This does all the path init required, making room in the tree if needed.
4821 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4822 const struct btrfs_key *cpu_key, void *data,
4826 struct btrfs_path *path;
4827 struct extent_buffer *leaf;
4830 path = btrfs_alloc_path();
4833 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4835 leaf = path->nodes[0];
4836 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4837 write_extent_buffer(leaf, data, ptr, data_size);
4838 btrfs_mark_buffer_dirty(leaf);
4840 btrfs_free_path(path);
4845 * delete the pointer from a given node.
4847 * the tree should have been previously balanced so the deletion does not
4850 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4851 int level, int slot)
4853 struct extent_buffer *parent = path->nodes[level];
4857 nritems = btrfs_header_nritems(parent);
4858 if (slot != nritems - 1) {
4860 ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4861 nritems - slot - 1);
4864 memmove_extent_buffer(parent,
4865 btrfs_node_key_ptr_offset(slot),
4866 btrfs_node_key_ptr_offset(slot + 1),
4867 sizeof(struct btrfs_key_ptr) *
4868 (nritems - slot - 1));
4870 ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4876 btrfs_set_header_nritems(parent, nritems);
4877 if (nritems == 0 && parent == root->node) {
4878 BUG_ON(btrfs_header_level(root->node) != 1);
4879 /* just turn the root into a leaf and break */
4880 btrfs_set_header_level(root->node, 0);
4881 } else if (slot == 0) {
4882 struct btrfs_disk_key disk_key;
4884 btrfs_node_key(parent, &disk_key, 0);
4885 fixup_low_keys(path, &disk_key, level + 1);
4887 btrfs_mark_buffer_dirty(parent);
4891 * a helper function to delete the leaf pointed to by path->slots[1] and
4894 * This deletes the pointer in path->nodes[1] and frees the leaf
4895 * block extent. zero is returned if it all worked out, < 0 otherwise.
4897 * The path must have already been setup for deleting the leaf, including
4898 * all the proper balancing. path->nodes[1] must be locked.
4900 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4901 struct btrfs_root *root,
4902 struct btrfs_path *path,
4903 struct extent_buffer *leaf)
4905 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4906 del_ptr(root, path, 1, path->slots[1]);
4909 * btrfs_free_extent is expensive, we want to make sure we
4910 * aren't holding any locks when we call it
4912 btrfs_unlock_up_safe(path, 0);
4914 root_sub_used(root, leaf->len);
4916 extent_buffer_get(leaf);
4917 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4918 free_extent_buffer_stale(leaf);
4921 * delete the item at the leaf level in path. If that empties
4922 * the leaf, remove it from the tree
4924 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4925 struct btrfs_path *path, int slot, int nr)
4927 struct btrfs_fs_info *fs_info = root->fs_info;
4928 struct extent_buffer *leaf;
4929 struct btrfs_item *item;
4936 struct btrfs_map_token token;
4938 btrfs_init_map_token(&token);
4940 leaf = path->nodes[0];
4941 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4943 for (i = 0; i < nr; i++)
4944 dsize += btrfs_item_size_nr(leaf, slot + i);
4946 nritems = btrfs_header_nritems(leaf);
4948 if (slot + nr != nritems) {
4949 int data_end = leaf_data_end(fs_info, leaf);
4951 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4953 BTRFS_LEAF_DATA_OFFSET + data_end,
4954 last_off - data_end);
4956 for (i = slot + nr; i < nritems; i++) {
4959 item = btrfs_item_nr(i);
4960 ioff = btrfs_token_item_offset(leaf, item, &token);
4961 btrfs_set_token_item_offset(leaf, item,
4962 ioff + dsize, &token);
4965 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4966 btrfs_item_nr_offset(slot + nr),
4967 sizeof(struct btrfs_item) *
4968 (nritems - slot - nr));
4970 btrfs_set_header_nritems(leaf, nritems - nr);
4973 /* delete the leaf if we've emptied it */
4975 if (leaf == root->node) {
4976 btrfs_set_header_level(leaf, 0);
4978 btrfs_set_path_blocking(path);
4979 clean_tree_block(fs_info, leaf);
4980 btrfs_del_leaf(trans, root, path, leaf);
4983 int used = leaf_space_used(leaf, 0, nritems);
4985 struct btrfs_disk_key disk_key;
4987 btrfs_item_key(leaf, &disk_key, 0);
4988 fixup_low_keys(path, &disk_key, 1);
4991 /* delete the leaf if it is mostly empty */
4992 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
4993 /* push_leaf_left fixes the path.
4994 * make sure the path still points to our leaf
4995 * for possible call to del_ptr below
4997 slot = path->slots[1];
4998 extent_buffer_get(leaf);
5000 btrfs_set_path_blocking(path);
5001 wret = push_leaf_left(trans, root, path, 1, 1,
5003 if (wret < 0 && wret != -ENOSPC)
5006 if (path->nodes[0] == leaf &&
5007 btrfs_header_nritems(leaf)) {
5008 wret = push_leaf_right(trans, root, path, 1,
5010 if (wret < 0 && wret != -ENOSPC)
5014 if (btrfs_header_nritems(leaf) == 0) {
5015 path->slots[1] = slot;
5016 btrfs_del_leaf(trans, root, path, leaf);
5017 free_extent_buffer(leaf);
5020 /* if we're still in the path, make sure
5021 * we're dirty. Otherwise, one of the
5022 * push_leaf functions must have already
5023 * dirtied this buffer
5025 if (path->nodes[0] == leaf)
5026 btrfs_mark_buffer_dirty(leaf);
5027 free_extent_buffer(leaf);
5030 btrfs_mark_buffer_dirty(leaf);
5037 * search the tree again to find a leaf with lesser keys
5038 * returns 0 if it found something or 1 if there are no lesser leaves.
5039 * returns < 0 on io errors.
5041 * This may release the path, and so you may lose any locks held at the
5044 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5046 struct btrfs_key key;
5047 struct btrfs_disk_key found_key;
5050 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5052 if (key.offset > 0) {
5054 } else if (key.type > 0) {
5056 key.offset = (u64)-1;
5057 } else if (key.objectid > 0) {
5060 key.offset = (u64)-1;
5065 btrfs_release_path(path);
5066 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5069 btrfs_item_key(path->nodes[0], &found_key, 0);
5070 ret = comp_keys(&found_key, &key);
5072 * We might have had an item with the previous key in the tree right
5073 * before we released our path. And after we released our path, that
5074 * item might have been pushed to the first slot (0) of the leaf we
5075 * were holding due to a tree balance. Alternatively, an item with the
5076 * previous key can exist as the only element of a leaf (big fat item).
5077 * Therefore account for these 2 cases, so that our callers (like
5078 * btrfs_previous_item) don't miss an existing item with a key matching
5079 * the previous key we computed above.
5087 * A helper function to walk down the tree starting at min_key, and looking
5088 * for nodes or leaves that are have a minimum transaction id.
5089 * This is used by the btree defrag code, and tree logging
5091 * This does not cow, but it does stuff the starting key it finds back
5092 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5093 * key and get a writable path.
5095 * This honors path->lowest_level to prevent descent past a given level
5098 * min_trans indicates the oldest transaction that you are interested
5099 * in walking through. Any nodes or leaves older than min_trans are
5100 * skipped over (without reading them).
5102 * returns zero if something useful was found, < 0 on error and 1 if there
5103 * was nothing in the tree that matched the search criteria.
5105 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5106 struct btrfs_path *path,
5109 struct btrfs_fs_info *fs_info = root->fs_info;
5110 struct extent_buffer *cur;
5111 struct btrfs_key found_key;
5117 int keep_locks = path->keep_locks;
5119 path->keep_locks = 1;
5121 cur = btrfs_read_lock_root_node(root);
5122 level = btrfs_header_level(cur);
5123 WARN_ON(path->nodes[level]);
5124 path->nodes[level] = cur;
5125 path->locks[level] = BTRFS_READ_LOCK;
5127 if (btrfs_header_generation(cur) < min_trans) {
5132 nritems = btrfs_header_nritems(cur);
5133 level = btrfs_header_level(cur);
5134 sret = btrfs_bin_search(cur, min_key, level, &slot);
5136 /* at the lowest level, we're done, setup the path and exit */
5137 if (level == path->lowest_level) {
5138 if (slot >= nritems)
5141 path->slots[level] = slot;
5142 btrfs_item_key_to_cpu(cur, &found_key, slot);
5145 if (sret && slot > 0)
5148 * check this node pointer against the min_trans parameters.
5149 * If it is too old, old, skip to the next one.
5151 while (slot < nritems) {
5154 gen = btrfs_node_ptr_generation(cur, slot);
5155 if (gen < min_trans) {
5163 * we didn't find a candidate key in this node, walk forward
5164 * and find another one
5166 if (slot >= nritems) {
5167 path->slots[level] = slot;
5168 btrfs_set_path_blocking(path);
5169 sret = btrfs_find_next_key(root, path, min_key, level,
5172 btrfs_release_path(path);
5178 /* save our key for returning back */
5179 btrfs_node_key_to_cpu(cur, &found_key, slot);
5180 path->slots[level] = slot;
5181 if (level == path->lowest_level) {
5185 btrfs_set_path_blocking(path);
5186 cur = read_node_slot(fs_info, cur, slot);
5192 btrfs_tree_read_lock(cur);
5194 path->locks[level - 1] = BTRFS_READ_LOCK;
5195 path->nodes[level - 1] = cur;
5196 unlock_up(path, level, 1, 0, NULL);
5199 path->keep_locks = keep_locks;
5201 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5202 btrfs_set_path_blocking(path);
5203 memcpy(min_key, &found_key, sizeof(found_key));
5208 static int tree_move_down(struct btrfs_fs_info *fs_info,
5209 struct btrfs_path *path,
5212 struct extent_buffer *eb;
5214 BUG_ON(*level == 0);
5215 eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]);
5219 path->nodes[*level - 1] = eb;
5220 path->slots[*level - 1] = 0;
5225 static int tree_move_next_or_upnext(struct btrfs_path *path,
5226 int *level, int root_level)
5230 nritems = btrfs_header_nritems(path->nodes[*level]);
5232 path->slots[*level]++;
5234 while (path->slots[*level] >= nritems) {
5235 if (*level == root_level)
5239 path->slots[*level] = 0;
5240 free_extent_buffer(path->nodes[*level]);
5241 path->nodes[*level] = NULL;
5243 path->slots[*level]++;
5245 nritems = btrfs_header_nritems(path->nodes[*level]);
5252 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5255 static int tree_advance(struct btrfs_fs_info *fs_info,
5256 struct btrfs_path *path,
5257 int *level, int root_level,
5259 struct btrfs_key *key)
5263 if (*level == 0 || !allow_down) {
5264 ret = tree_move_next_or_upnext(path, level, root_level);
5266 ret = tree_move_down(fs_info, path, level);
5270 btrfs_item_key_to_cpu(path->nodes[*level], key,
5271 path->slots[*level]);
5273 btrfs_node_key_to_cpu(path->nodes[*level], key,
5274 path->slots[*level]);
5279 static int tree_compare_item(struct btrfs_path *left_path,
5280 struct btrfs_path *right_path,
5285 unsigned long off1, off2;
5287 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5288 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5292 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5293 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5294 right_path->slots[0]);
5296 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5298 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5305 #define ADVANCE_ONLY_NEXT -1
5308 * This function compares two trees and calls the provided callback for
5309 * every changed/new/deleted item it finds.
5310 * If shared tree blocks are encountered, whole subtrees are skipped, making
5311 * the compare pretty fast on snapshotted subvolumes.
5313 * This currently works on commit roots only. As commit roots are read only,
5314 * we don't do any locking. The commit roots are protected with transactions.
5315 * Transactions are ended and rejoined when a commit is tried in between.
5317 * This function checks for modifications done to the trees while comparing.
5318 * If it detects a change, it aborts immediately.
5320 int btrfs_compare_trees(struct btrfs_root *left_root,
5321 struct btrfs_root *right_root,
5322 btrfs_changed_cb_t changed_cb, void *ctx)
5324 struct btrfs_fs_info *fs_info = left_root->fs_info;
5327 struct btrfs_path *left_path = NULL;
5328 struct btrfs_path *right_path = NULL;
5329 struct btrfs_key left_key;
5330 struct btrfs_key right_key;
5331 char *tmp_buf = NULL;
5332 int left_root_level;
5333 int right_root_level;
5336 int left_end_reached;
5337 int right_end_reached;
5345 left_path = btrfs_alloc_path();
5350 right_path = btrfs_alloc_path();
5356 tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
5362 left_path->search_commit_root = 1;
5363 left_path->skip_locking = 1;
5364 right_path->search_commit_root = 1;
5365 right_path->skip_locking = 1;
5368 * Strategy: Go to the first items of both trees. Then do
5370 * If both trees are at level 0
5371 * Compare keys of current items
5372 * If left < right treat left item as new, advance left tree
5374 * If left > right treat right item as deleted, advance right tree
5376 * If left == right do deep compare of items, treat as changed if
5377 * needed, advance both trees and repeat
5378 * If both trees are at the same level but not at level 0
5379 * Compare keys of current nodes/leafs
5380 * If left < right advance left tree and repeat
5381 * If left > right advance right tree and repeat
5382 * If left == right compare blockptrs of the next nodes/leafs
5383 * If they match advance both trees but stay at the same level
5385 * If they don't match advance both trees while allowing to go
5387 * If tree levels are different
5388 * Advance the tree that needs it and repeat
5390 * Advancing a tree means:
5391 * If we are at level 0, try to go to the next slot. If that's not
5392 * possible, go one level up and repeat. Stop when we found a level
5393 * where we could go to the next slot. We may at this point be on a
5396 * If we are not at level 0 and not on shared tree blocks, go one
5399 * If we are not at level 0 and on shared tree blocks, go one slot to
5400 * the right if possible or go up and right.
5403 down_read(&fs_info->commit_root_sem);
5404 left_level = btrfs_header_level(left_root->commit_root);
5405 left_root_level = left_level;
5406 left_path->nodes[left_level] =
5407 btrfs_clone_extent_buffer(left_root->commit_root);
5408 if (!left_path->nodes[left_level]) {
5409 up_read(&fs_info->commit_root_sem);
5414 right_level = btrfs_header_level(right_root->commit_root);
5415 right_root_level = right_level;
5416 right_path->nodes[right_level] =
5417 btrfs_clone_extent_buffer(right_root->commit_root);
5418 if (!right_path->nodes[right_level]) {
5419 up_read(&fs_info->commit_root_sem);
5423 up_read(&fs_info->commit_root_sem);
5425 if (left_level == 0)
5426 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5427 &left_key, left_path->slots[left_level]);
5429 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5430 &left_key, left_path->slots[left_level]);
5431 if (right_level == 0)
5432 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5433 &right_key, right_path->slots[right_level]);
5435 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5436 &right_key, right_path->slots[right_level]);
5438 left_end_reached = right_end_reached = 0;
5439 advance_left = advance_right = 0;
5442 if (advance_left && !left_end_reached) {
5443 ret = tree_advance(fs_info, left_path, &left_level,
5445 advance_left != ADVANCE_ONLY_NEXT,
5448 left_end_reached = ADVANCE;
5453 if (advance_right && !right_end_reached) {
5454 ret = tree_advance(fs_info, right_path, &right_level,
5456 advance_right != ADVANCE_ONLY_NEXT,
5459 right_end_reached = ADVANCE;
5465 if (left_end_reached && right_end_reached) {
5468 } else if (left_end_reached) {
5469 if (right_level == 0) {
5470 ret = changed_cb(left_path, right_path,
5472 BTRFS_COMPARE_TREE_DELETED,
5477 advance_right = ADVANCE;
5479 } else if (right_end_reached) {
5480 if (left_level == 0) {
5481 ret = changed_cb(left_path, right_path,
5483 BTRFS_COMPARE_TREE_NEW,
5488 advance_left = ADVANCE;
5492 if (left_level == 0 && right_level == 0) {
5493 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5495 ret = changed_cb(left_path, right_path,
5497 BTRFS_COMPARE_TREE_NEW,
5501 advance_left = ADVANCE;
5502 } else if (cmp > 0) {
5503 ret = changed_cb(left_path, right_path,
5505 BTRFS_COMPARE_TREE_DELETED,
5509 advance_right = ADVANCE;
5511 enum btrfs_compare_tree_result result;
5513 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5514 ret = tree_compare_item(left_path, right_path,
5517 result = BTRFS_COMPARE_TREE_CHANGED;
5519 result = BTRFS_COMPARE_TREE_SAME;
5520 ret = changed_cb(left_path, right_path,
5521 &left_key, result, ctx);
5524 advance_left = ADVANCE;
5525 advance_right = ADVANCE;
5527 } else if (left_level == right_level) {
5528 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5530 advance_left = ADVANCE;
5531 } else if (cmp > 0) {
5532 advance_right = ADVANCE;
5534 left_blockptr = btrfs_node_blockptr(
5535 left_path->nodes[left_level],
5536 left_path->slots[left_level]);
5537 right_blockptr = btrfs_node_blockptr(
5538 right_path->nodes[right_level],
5539 right_path->slots[right_level]);
5540 left_gen = btrfs_node_ptr_generation(
5541 left_path->nodes[left_level],
5542 left_path->slots[left_level]);
5543 right_gen = btrfs_node_ptr_generation(
5544 right_path->nodes[right_level],
5545 right_path->slots[right_level]);
5546 if (left_blockptr == right_blockptr &&
5547 left_gen == right_gen) {
5549 * As we're on a shared block, don't
5550 * allow to go deeper.
5552 advance_left = ADVANCE_ONLY_NEXT;
5553 advance_right = ADVANCE_ONLY_NEXT;
5555 advance_left = ADVANCE;
5556 advance_right = ADVANCE;
5559 } else if (left_level < right_level) {
5560 advance_right = ADVANCE;
5562 advance_left = ADVANCE;
5567 btrfs_free_path(left_path);
5568 btrfs_free_path(right_path);
5574 * this is similar to btrfs_next_leaf, but does not try to preserve
5575 * and fixup the path. It looks for and returns the next key in the
5576 * tree based on the current path and the min_trans parameters.
5578 * 0 is returned if another key is found, < 0 if there are any errors
5579 * and 1 is returned if there are no higher keys in the tree
5581 * path->keep_locks should be set to 1 on the search made before
5582 * calling this function.
5584 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5585 struct btrfs_key *key, int level, u64 min_trans)
5588 struct extent_buffer *c;
5590 WARN_ON(!path->keep_locks);
5591 while (level < BTRFS_MAX_LEVEL) {
5592 if (!path->nodes[level])
5595 slot = path->slots[level] + 1;
5596 c = path->nodes[level];
5598 if (slot >= btrfs_header_nritems(c)) {
5601 struct btrfs_key cur_key;
5602 if (level + 1 >= BTRFS_MAX_LEVEL ||
5603 !path->nodes[level + 1])
5606 if (path->locks[level + 1]) {
5611 slot = btrfs_header_nritems(c) - 1;
5613 btrfs_item_key_to_cpu(c, &cur_key, slot);
5615 btrfs_node_key_to_cpu(c, &cur_key, slot);
5617 orig_lowest = path->lowest_level;
5618 btrfs_release_path(path);
5619 path->lowest_level = level;
5620 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5622 path->lowest_level = orig_lowest;
5626 c = path->nodes[level];
5627 slot = path->slots[level];
5634 btrfs_item_key_to_cpu(c, key, slot);
5636 u64 gen = btrfs_node_ptr_generation(c, slot);
5638 if (gen < min_trans) {
5642 btrfs_node_key_to_cpu(c, key, slot);
5650 * search the tree again to find a leaf with greater keys
5651 * returns 0 if it found something or 1 if there are no greater leaves.
5652 * returns < 0 on io errors.
5654 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5656 return btrfs_next_old_leaf(root, path, 0);
5659 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5664 struct extent_buffer *c;
5665 struct extent_buffer *next;
5666 struct btrfs_key key;
5669 int old_spinning = path->leave_spinning;
5670 int next_rw_lock = 0;
5672 nritems = btrfs_header_nritems(path->nodes[0]);
5676 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5681 btrfs_release_path(path);
5683 path->keep_locks = 1;
5684 path->leave_spinning = 1;
5687 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5689 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5690 path->keep_locks = 0;
5695 nritems = btrfs_header_nritems(path->nodes[0]);
5697 * by releasing the path above we dropped all our locks. A balance
5698 * could have added more items next to the key that used to be
5699 * at the very end of the block. So, check again here and
5700 * advance the path if there are now more items available.
5702 if (nritems > 0 && path->slots[0] < nritems - 1) {
5709 * So the above check misses one case:
5710 * - after releasing the path above, someone has removed the item that
5711 * used to be at the very end of the block, and balance between leafs
5712 * gets another one with bigger key.offset to replace it.
5714 * This one should be returned as well, or we can get leaf corruption
5715 * later(esp. in __btrfs_drop_extents()).
5717 * And a bit more explanation about this check,
5718 * with ret > 0, the key isn't found, the path points to the slot
5719 * where it should be inserted, so the path->slots[0] item must be the
5722 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5727 while (level < BTRFS_MAX_LEVEL) {
5728 if (!path->nodes[level]) {
5733 slot = path->slots[level] + 1;
5734 c = path->nodes[level];
5735 if (slot >= btrfs_header_nritems(c)) {
5737 if (level == BTRFS_MAX_LEVEL) {
5745 btrfs_tree_unlock_rw(next, next_rw_lock);
5746 free_extent_buffer(next);
5750 next_rw_lock = path->locks[level];
5751 ret = read_block_for_search(root, path, &next, level,
5757 btrfs_release_path(path);
5761 if (!path->skip_locking) {
5762 ret = btrfs_try_tree_read_lock(next);
5763 if (!ret && time_seq) {
5765 * If we don't get the lock, we may be racing
5766 * with push_leaf_left, holding that lock while
5767 * itself waiting for the leaf we've currently
5768 * locked. To solve this situation, we give up
5769 * on our lock and cycle.
5771 free_extent_buffer(next);
5772 btrfs_release_path(path);
5777 btrfs_set_path_blocking(path);
5778 btrfs_tree_read_lock(next);
5780 next_rw_lock = BTRFS_READ_LOCK;
5784 path->slots[level] = slot;
5787 c = path->nodes[level];
5788 if (path->locks[level])
5789 btrfs_tree_unlock_rw(c, path->locks[level]);
5791 free_extent_buffer(c);
5792 path->nodes[level] = next;
5793 path->slots[level] = 0;
5794 if (!path->skip_locking)
5795 path->locks[level] = next_rw_lock;
5799 ret = read_block_for_search(root, path, &next, level,
5805 btrfs_release_path(path);
5809 if (!path->skip_locking) {
5810 ret = btrfs_try_tree_read_lock(next);
5812 btrfs_set_path_blocking(path);
5813 btrfs_tree_read_lock(next);
5815 next_rw_lock = BTRFS_READ_LOCK;
5820 unlock_up(path, 0, 1, 0, NULL);
5821 path->leave_spinning = old_spinning;
5823 btrfs_set_path_blocking(path);
5829 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5830 * searching until it gets past min_objectid or finds an item of 'type'
5832 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5834 int btrfs_previous_item(struct btrfs_root *root,
5835 struct btrfs_path *path, u64 min_objectid,
5838 struct btrfs_key found_key;
5839 struct extent_buffer *leaf;
5844 if (path->slots[0] == 0) {
5845 btrfs_set_path_blocking(path);
5846 ret = btrfs_prev_leaf(root, path);
5852 leaf = path->nodes[0];
5853 nritems = btrfs_header_nritems(leaf);
5856 if (path->slots[0] == nritems)
5859 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5860 if (found_key.objectid < min_objectid)
5862 if (found_key.type == type)
5864 if (found_key.objectid == min_objectid &&
5865 found_key.type < type)
5872 * search in extent tree to find a previous Metadata/Data extent item with
5875 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5877 int btrfs_previous_extent_item(struct btrfs_root *root,
5878 struct btrfs_path *path, u64 min_objectid)
5880 struct btrfs_key found_key;
5881 struct extent_buffer *leaf;
5886 if (path->slots[0] == 0) {
5887 btrfs_set_path_blocking(path);
5888 ret = btrfs_prev_leaf(root, path);
5894 leaf = path->nodes[0];
5895 nritems = btrfs_header_nritems(leaf);
5898 if (path->slots[0] == nritems)
5901 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5902 if (found_key.objectid < min_objectid)
5904 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5905 found_key.type == BTRFS_METADATA_ITEM_KEY)
5907 if (found_key.objectid == min_objectid &&
5908 found_key.type < BTRFS_EXTENT_ITEM_KEY)