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, device or free
1020 * space trees, make sure that we do not finish block group creation of
1021 * pending block 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, device and free space trees,
1026 * therefore we could deadlock with ourselves since we are holding a
1027 * lock on an extent buffer that btrfs_create_pending_block_groups() may
1030 if (root == fs_info->extent_root ||
1031 root == fs_info->chunk_root ||
1032 root == fs_info->dev_root ||
1033 root == fs_info->free_space_root)
1034 trans->can_flush_pending_bgs = false;
1036 cow = btrfs_alloc_tree_block(trans, root, parent_start,
1037 root->root_key.objectid, &disk_key, level,
1038 search_start, empty_size);
1039 trans->can_flush_pending_bgs = true;
1041 return PTR_ERR(cow);
1043 /* cow is set to blocking by btrfs_init_new_buffer */
1045 copy_extent_buffer_full(cow, buf);
1046 btrfs_set_header_bytenr(cow, cow->start);
1047 btrfs_set_header_generation(cow, trans->transid);
1048 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1049 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1050 BTRFS_HEADER_FLAG_RELOC);
1051 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1052 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1054 btrfs_set_header_owner(cow, root->root_key.objectid);
1056 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
1058 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1060 btrfs_abort_transaction(trans, ret);
1064 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1065 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1067 btrfs_abort_transaction(trans, ret);
1072 if (buf == root->node) {
1073 WARN_ON(parent && parent != buf);
1074 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1075 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1076 parent_start = buf->start;
1078 extent_buffer_get(cow);
1079 ret = tree_mod_log_insert_root(root->node, cow, 1);
1081 rcu_assign_pointer(root->node, cow);
1083 btrfs_free_tree_block(trans, root, buf, parent_start,
1085 free_extent_buffer(buf);
1086 add_root_to_dirty_list(root);
1088 WARN_ON(trans->transid != btrfs_header_generation(parent));
1089 tree_mod_log_insert_key(parent, parent_slot,
1090 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1091 btrfs_set_node_blockptr(parent, parent_slot,
1093 btrfs_set_node_ptr_generation(parent, parent_slot,
1095 btrfs_mark_buffer_dirty(parent);
1097 ret = tree_mod_log_free_eb(buf);
1099 btrfs_abort_transaction(trans, ret);
1103 btrfs_free_tree_block(trans, root, buf, parent_start,
1107 btrfs_tree_unlock(buf);
1108 free_extent_buffer_stale(buf);
1109 btrfs_mark_buffer_dirty(cow);
1115 * returns the logical address of the oldest predecessor of the given root.
1116 * entries older than time_seq are ignored.
1118 static struct tree_mod_elem *__tree_mod_log_oldest_root(
1119 struct extent_buffer *eb_root, u64 time_seq)
1121 struct tree_mod_elem *tm;
1122 struct tree_mod_elem *found = NULL;
1123 u64 root_logical = eb_root->start;
1130 * the very last operation that's logged for a root is the
1131 * replacement operation (if it is replaced at all). this has
1132 * the logical address of the *new* root, making it the very
1133 * first operation that's logged for this root.
1136 tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
1141 * if there are no tree operation for the oldest root, we simply
1142 * return it. this should only happen if that (old) root is at
1149 * if there's an operation that's not a root replacement, we
1150 * found the oldest version of our root. normally, we'll find a
1151 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1153 if (tm->op != MOD_LOG_ROOT_REPLACE)
1157 root_logical = tm->old_root.logical;
1161 /* if there's no old root to return, return what we found instead */
1169 * tm is a pointer to the first operation to rewind within eb. then, all
1170 * previous operations will be rewound (until we reach something older than
1174 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1175 u64 time_seq, struct tree_mod_elem *first_tm)
1178 struct rb_node *next;
1179 struct tree_mod_elem *tm = first_tm;
1180 unsigned long o_dst;
1181 unsigned long o_src;
1182 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1184 n = btrfs_header_nritems(eb);
1185 read_lock(&fs_info->tree_mod_log_lock);
1186 while (tm && tm->seq >= time_seq) {
1188 * all the operations are recorded with the operator used for
1189 * the modification. as we're going backwards, we do the
1190 * opposite of each operation here.
1193 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1194 BUG_ON(tm->slot < n);
1196 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1197 case MOD_LOG_KEY_REMOVE:
1198 btrfs_set_node_key(eb, &tm->key, tm->slot);
1199 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1200 btrfs_set_node_ptr_generation(eb, tm->slot,
1204 case MOD_LOG_KEY_REPLACE:
1205 BUG_ON(tm->slot >= n);
1206 btrfs_set_node_key(eb, &tm->key, tm->slot);
1207 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1208 btrfs_set_node_ptr_generation(eb, tm->slot,
1211 case MOD_LOG_KEY_ADD:
1212 /* if a move operation is needed it's in the log */
1215 case MOD_LOG_MOVE_KEYS:
1216 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1217 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1218 memmove_extent_buffer(eb, o_dst, o_src,
1219 tm->move.nr_items * p_size);
1221 case MOD_LOG_ROOT_REPLACE:
1223 * this operation is special. for roots, this must be
1224 * handled explicitly before rewinding.
1225 * for non-roots, this operation may exist if the node
1226 * was a root: root A -> child B; then A gets empty and
1227 * B is promoted to the new root. in the mod log, we'll
1228 * have a root-replace operation for B, a tree block
1229 * that is no root. we simply ignore that operation.
1233 next = rb_next(&tm->node);
1236 tm = rb_entry(next, struct tree_mod_elem, node);
1237 if (tm->logical != first_tm->logical)
1240 read_unlock(&fs_info->tree_mod_log_lock);
1241 btrfs_set_header_nritems(eb, n);
1245 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1246 * is returned. If rewind operations happen, a fresh buffer is returned. The
1247 * returned buffer is always read-locked. If the returned buffer is not the
1248 * input buffer, the lock on the input buffer is released and the input buffer
1249 * is freed (its refcount is decremented).
1251 static struct extent_buffer *
1252 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1253 struct extent_buffer *eb, u64 time_seq)
1255 struct extent_buffer *eb_rewin;
1256 struct tree_mod_elem *tm;
1261 if (btrfs_header_level(eb) == 0)
1264 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1268 btrfs_set_path_blocking(path);
1269 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1271 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1272 BUG_ON(tm->slot != 0);
1273 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1275 btrfs_tree_read_unlock_blocking(eb);
1276 free_extent_buffer(eb);
1279 btrfs_set_header_bytenr(eb_rewin, eb->start);
1280 btrfs_set_header_backref_rev(eb_rewin,
1281 btrfs_header_backref_rev(eb));
1282 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1283 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1285 eb_rewin = btrfs_clone_extent_buffer(eb);
1287 btrfs_tree_read_unlock_blocking(eb);
1288 free_extent_buffer(eb);
1293 btrfs_tree_read_unlock_blocking(eb);
1294 free_extent_buffer(eb);
1296 btrfs_tree_read_lock(eb_rewin);
1297 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1298 WARN_ON(btrfs_header_nritems(eb_rewin) >
1299 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1305 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1306 * value. If there are no changes, the current root->root_node is returned. If
1307 * anything changed in between, there's a fresh buffer allocated on which the
1308 * rewind operations are done. In any case, the returned buffer is read locked.
1309 * Returns NULL on error (with no locks held).
1311 static inline struct extent_buffer *
1312 get_old_root(struct btrfs_root *root, u64 time_seq)
1314 struct btrfs_fs_info *fs_info = root->fs_info;
1315 struct tree_mod_elem *tm;
1316 struct extent_buffer *eb = NULL;
1317 struct extent_buffer *eb_root;
1318 struct extent_buffer *old;
1319 struct tree_mod_root *old_root = NULL;
1320 u64 old_generation = 0;
1324 eb_root = btrfs_read_lock_root_node(root);
1325 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1329 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1330 old_root = &tm->old_root;
1331 old_generation = tm->generation;
1332 logical = old_root->logical;
1333 level = old_root->level;
1335 logical = eb_root->start;
1336 level = btrfs_header_level(eb_root);
1339 tm = tree_mod_log_search(fs_info, logical, time_seq);
1340 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1341 btrfs_tree_read_unlock(eb_root);
1342 free_extent_buffer(eb_root);
1343 old = read_tree_block(fs_info, logical, 0, level, NULL);
1344 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1346 free_extent_buffer(old);
1348 "failed to read tree block %llu from get_old_root",
1351 eb = btrfs_clone_extent_buffer(old);
1352 free_extent_buffer(old);
1354 } else if (old_root) {
1355 btrfs_tree_read_unlock(eb_root);
1356 free_extent_buffer(eb_root);
1357 eb = alloc_dummy_extent_buffer(fs_info, logical);
1359 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1360 eb = btrfs_clone_extent_buffer(eb_root);
1361 btrfs_tree_read_unlock_blocking(eb_root);
1362 free_extent_buffer(eb_root);
1367 btrfs_tree_read_lock(eb);
1369 btrfs_set_header_bytenr(eb, eb->start);
1370 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1371 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1372 btrfs_set_header_level(eb, old_root->level);
1373 btrfs_set_header_generation(eb, old_generation);
1376 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1378 WARN_ON(btrfs_header_level(eb) != 0);
1379 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1384 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1386 struct tree_mod_elem *tm;
1388 struct extent_buffer *eb_root = btrfs_root_node(root);
1390 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1391 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1392 level = tm->old_root.level;
1394 level = btrfs_header_level(eb_root);
1396 free_extent_buffer(eb_root);
1401 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root,
1403 struct extent_buffer *buf)
1405 if (btrfs_is_testing(root->fs_info))
1408 /* Ensure we can see the FORCE_COW bit */
1409 smp_mb__before_atomic();
1412 * We do not need to cow a block if
1413 * 1) this block is not created or changed in this transaction;
1414 * 2) this block does not belong to TREE_RELOC tree;
1415 * 3) the root is not forced COW.
1417 * What is forced COW:
1418 * when we create snapshot during committing the transaction,
1419 * after we've finished copying src root, we must COW the shared
1420 * block to ensure the metadata consistency.
1422 if (btrfs_header_generation(buf) == trans->transid &&
1423 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1424 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1425 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1426 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1432 * cows a single block, see __btrfs_cow_block for the real work.
1433 * This version of it has extra checks so that a block isn't COWed more than
1434 * once per transaction, as long as it hasn't been written yet
1436 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1437 struct btrfs_root *root, struct extent_buffer *buf,
1438 struct extent_buffer *parent, int parent_slot,
1439 struct extent_buffer **cow_ret)
1441 struct btrfs_fs_info *fs_info = root->fs_info;
1445 if (test_bit(BTRFS_ROOT_DELETING, &root->state))
1447 "COW'ing blocks on a fs root that's being dropped");
1449 if (trans->transaction != fs_info->running_transaction)
1450 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1452 fs_info->running_transaction->transid);
1454 if (trans->transid != fs_info->generation)
1455 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1456 trans->transid, fs_info->generation);
1458 if (!should_cow_block(trans, root, buf)) {
1459 trans->dirty = true;
1464 search_start = buf->start & ~((u64)SZ_1G - 1);
1467 btrfs_set_lock_blocking(parent);
1468 btrfs_set_lock_blocking(buf);
1470 ret = __btrfs_cow_block(trans, root, buf, parent,
1471 parent_slot, cow_ret, search_start, 0);
1473 trace_btrfs_cow_block(root, buf, *cow_ret);
1479 * helper function for defrag to decide if two blocks pointed to by a
1480 * node are actually close by
1482 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1484 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1486 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1492 * compare two keys in a memcmp fashion
1494 static int comp_keys(const struct btrfs_disk_key *disk,
1495 const struct btrfs_key *k2)
1497 struct btrfs_key k1;
1499 btrfs_disk_key_to_cpu(&k1, disk);
1501 return btrfs_comp_cpu_keys(&k1, k2);
1505 * same as comp_keys only with two btrfs_key's
1507 int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1509 if (k1->objectid > k2->objectid)
1511 if (k1->objectid < k2->objectid)
1513 if (k1->type > k2->type)
1515 if (k1->type < k2->type)
1517 if (k1->offset > k2->offset)
1519 if (k1->offset < k2->offset)
1525 * this is used by the defrag code to go through all the
1526 * leaves pointed to by a node and reallocate them so that
1527 * disk order is close to key order
1529 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1530 struct btrfs_root *root, struct extent_buffer *parent,
1531 int start_slot, u64 *last_ret,
1532 struct btrfs_key *progress)
1534 struct btrfs_fs_info *fs_info = root->fs_info;
1535 struct extent_buffer *cur;
1538 u64 search_start = *last_ret;
1548 int progress_passed = 0;
1549 struct btrfs_disk_key disk_key;
1551 parent_level = btrfs_header_level(parent);
1553 WARN_ON(trans->transaction != fs_info->running_transaction);
1554 WARN_ON(trans->transid != fs_info->generation);
1556 parent_nritems = btrfs_header_nritems(parent);
1557 blocksize = fs_info->nodesize;
1558 end_slot = parent_nritems - 1;
1560 if (parent_nritems <= 1)
1563 btrfs_set_lock_blocking(parent);
1565 for (i = start_slot; i <= end_slot; i++) {
1566 struct btrfs_key first_key;
1569 btrfs_node_key(parent, &disk_key, i);
1570 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1573 progress_passed = 1;
1574 blocknr = btrfs_node_blockptr(parent, i);
1575 gen = btrfs_node_ptr_generation(parent, i);
1576 btrfs_node_key_to_cpu(parent, &first_key, i);
1577 if (last_block == 0)
1578 last_block = blocknr;
1581 other = btrfs_node_blockptr(parent, i - 1);
1582 close = close_blocks(blocknr, other, blocksize);
1584 if (!close && i < end_slot) {
1585 other = btrfs_node_blockptr(parent, i + 1);
1586 close = close_blocks(blocknr, other, blocksize);
1589 last_block = blocknr;
1593 cur = find_extent_buffer(fs_info, blocknr);
1595 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1598 if (!cur || !uptodate) {
1600 cur = read_tree_block(fs_info, blocknr, gen,
1604 return PTR_ERR(cur);
1605 } else if (!extent_buffer_uptodate(cur)) {
1606 free_extent_buffer(cur);
1609 } else if (!uptodate) {
1610 err = btrfs_read_buffer(cur, gen,
1611 parent_level - 1,&first_key);
1613 free_extent_buffer(cur);
1618 if (search_start == 0)
1619 search_start = last_block;
1621 btrfs_tree_lock(cur);
1622 btrfs_set_lock_blocking(cur);
1623 err = __btrfs_cow_block(trans, root, cur, parent, i,
1626 (end_slot - i) * blocksize));
1628 btrfs_tree_unlock(cur);
1629 free_extent_buffer(cur);
1632 search_start = cur->start;
1633 last_block = cur->start;
1634 *last_ret = search_start;
1635 btrfs_tree_unlock(cur);
1636 free_extent_buffer(cur);
1642 * search for key in the extent_buffer. The items start at offset p,
1643 * and they are item_size apart. There are 'max' items in p.
1645 * the slot in the array is returned via slot, and it points to
1646 * the place where you would insert key if it is not found in
1649 * slot may point to max if the key is bigger than all of the keys
1651 static noinline int generic_bin_search(struct extent_buffer *eb,
1652 unsigned long p, int item_size,
1653 const struct btrfs_key *key,
1660 struct btrfs_disk_key *tmp = NULL;
1661 struct btrfs_disk_key unaligned;
1662 unsigned long offset;
1664 unsigned long map_start = 0;
1665 unsigned long map_len = 0;
1669 btrfs_err(eb->fs_info,
1670 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1671 __func__, low, high, eb->start,
1672 btrfs_header_owner(eb), btrfs_header_level(eb));
1676 while (low < high) {
1677 mid = (low + high) / 2;
1678 offset = p + mid * item_size;
1680 if (!kaddr || offset < map_start ||
1681 (offset + sizeof(struct btrfs_disk_key)) >
1682 map_start + map_len) {
1684 err = map_private_extent_buffer(eb, offset,
1685 sizeof(struct btrfs_disk_key),
1686 &kaddr, &map_start, &map_len);
1689 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1691 } else if (err == 1) {
1692 read_extent_buffer(eb, &unaligned,
1693 offset, sizeof(unaligned));
1700 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1703 ret = comp_keys(tmp, key);
1719 * simple bin_search frontend that does the right thing for
1722 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1723 int level, int *slot)
1726 return generic_bin_search(eb,
1727 offsetof(struct btrfs_leaf, items),
1728 sizeof(struct btrfs_item),
1729 key, btrfs_header_nritems(eb),
1732 return generic_bin_search(eb,
1733 offsetof(struct btrfs_node, ptrs),
1734 sizeof(struct btrfs_key_ptr),
1735 key, btrfs_header_nritems(eb),
1739 static void root_add_used(struct btrfs_root *root, u32 size)
1741 spin_lock(&root->accounting_lock);
1742 btrfs_set_root_used(&root->root_item,
1743 btrfs_root_used(&root->root_item) + size);
1744 spin_unlock(&root->accounting_lock);
1747 static void root_sub_used(struct btrfs_root *root, u32 size)
1749 spin_lock(&root->accounting_lock);
1750 btrfs_set_root_used(&root->root_item,
1751 btrfs_root_used(&root->root_item) - size);
1752 spin_unlock(&root->accounting_lock);
1755 /* given a node and slot number, this reads the blocks it points to. The
1756 * extent buffer is returned with a reference taken (but unlocked).
1758 static noinline struct extent_buffer *
1759 read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent,
1762 int level = btrfs_header_level(parent);
1763 struct extent_buffer *eb;
1764 struct btrfs_key first_key;
1766 if (slot < 0 || slot >= btrfs_header_nritems(parent))
1767 return ERR_PTR(-ENOENT);
1771 btrfs_node_key_to_cpu(parent, &first_key, slot);
1772 eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot),
1773 btrfs_node_ptr_generation(parent, slot),
1774 level - 1, &first_key);
1775 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1776 free_extent_buffer(eb);
1784 * node level balancing, used to make sure nodes are in proper order for
1785 * item deletion. We balance from the top down, so we have to make sure
1786 * that a deletion won't leave an node completely empty later on.
1788 static noinline int balance_level(struct btrfs_trans_handle *trans,
1789 struct btrfs_root *root,
1790 struct btrfs_path *path, int level)
1792 struct btrfs_fs_info *fs_info = root->fs_info;
1793 struct extent_buffer *right = NULL;
1794 struct extent_buffer *mid;
1795 struct extent_buffer *left = NULL;
1796 struct extent_buffer *parent = NULL;
1800 int orig_slot = path->slots[level];
1805 mid = path->nodes[level];
1807 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1808 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1809 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1811 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1813 if (level < BTRFS_MAX_LEVEL - 1) {
1814 parent = path->nodes[level + 1];
1815 pslot = path->slots[level + 1];
1819 * deal with the case where there is only one pointer in the root
1820 * by promoting the node below to a root
1823 struct extent_buffer *child;
1825 if (btrfs_header_nritems(mid) != 1)
1828 /* promote the child to a root */
1829 child = read_node_slot(fs_info, mid, 0);
1830 if (IS_ERR(child)) {
1831 ret = PTR_ERR(child);
1832 btrfs_handle_fs_error(fs_info, ret, NULL);
1836 btrfs_tree_lock(child);
1837 btrfs_set_lock_blocking(child);
1838 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1840 btrfs_tree_unlock(child);
1841 free_extent_buffer(child);
1845 ret = tree_mod_log_insert_root(root->node, child, 1);
1847 rcu_assign_pointer(root->node, child);
1849 add_root_to_dirty_list(root);
1850 btrfs_tree_unlock(child);
1852 path->locks[level] = 0;
1853 path->nodes[level] = NULL;
1854 clean_tree_block(fs_info, mid);
1855 btrfs_tree_unlock(mid);
1856 /* once for the path */
1857 free_extent_buffer(mid);
1859 root_sub_used(root, mid->len);
1860 btrfs_free_tree_block(trans, root, mid, 0, 1);
1861 /* once for the root ptr */
1862 free_extent_buffer_stale(mid);
1865 if (btrfs_header_nritems(mid) >
1866 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1869 left = read_node_slot(fs_info, parent, pslot - 1);
1874 btrfs_tree_lock(left);
1875 btrfs_set_lock_blocking(left);
1876 wret = btrfs_cow_block(trans, root, left,
1877 parent, pslot - 1, &left);
1884 right = read_node_slot(fs_info, parent, pslot + 1);
1889 btrfs_tree_lock(right);
1890 btrfs_set_lock_blocking(right);
1891 wret = btrfs_cow_block(trans, root, right,
1892 parent, pslot + 1, &right);
1899 /* first, try to make some room in the middle buffer */
1901 orig_slot += btrfs_header_nritems(left);
1902 wret = push_node_left(trans, fs_info, left, mid, 1);
1908 * then try to empty the right most buffer into the middle
1911 wret = push_node_left(trans, fs_info, mid, right, 1);
1912 if (wret < 0 && wret != -ENOSPC)
1914 if (btrfs_header_nritems(right) == 0) {
1915 clean_tree_block(fs_info, right);
1916 btrfs_tree_unlock(right);
1917 del_ptr(root, path, level + 1, pslot + 1);
1918 root_sub_used(root, right->len);
1919 btrfs_free_tree_block(trans, root, right, 0, 1);
1920 free_extent_buffer_stale(right);
1923 struct btrfs_disk_key right_key;
1924 btrfs_node_key(right, &right_key, 0);
1925 ret = tree_mod_log_insert_key(parent, pslot + 1,
1926 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1928 btrfs_set_node_key(parent, &right_key, pslot + 1);
1929 btrfs_mark_buffer_dirty(parent);
1932 if (btrfs_header_nritems(mid) == 1) {
1934 * we're not allowed to leave a node with one item in the
1935 * tree during a delete. A deletion from lower in the tree
1936 * could try to delete the only pointer in this node.
1937 * So, pull some keys from the left.
1938 * There has to be a left pointer at this point because
1939 * otherwise we would have pulled some pointers from the
1944 btrfs_handle_fs_error(fs_info, ret, NULL);
1947 wret = balance_node_right(trans, fs_info, mid, left);
1953 wret = push_node_left(trans, fs_info, left, mid, 1);
1959 if (btrfs_header_nritems(mid) == 0) {
1960 clean_tree_block(fs_info, mid);
1961 btrfs_tree_unlock(mid);
1962 del_ptr(root, path, level + 1, pslot);
1963 root_sub_used(root, mid->len);
1964 btrfs_free_tree_block(trans, root, mid, 0, 1);
1965 free_extent_buffer_stale(mid);
1968 /* update the parent key to reflect our changes */
1969 struct btrfs_disk_key mid_key;
1970 btrfs_node_key(mid, &mid_key, 0);
1971 ret = tree_mod_log_insert_key(parent, pslot,
1972 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1974 btrfs_set_node_key(parent, &mid_key, pslot);
1975 btrfs_mark_buffer_dirty(parent);
1978 /* update the path */
1980 if (btrfs_header_nritems(left) > orig_slot) {
1981 extent_buffer_get(left);
1982 /* left was locked after cow */
1983 path->nodes[level] = left;
1984 path->slots[level + 1] -= 1;
1985 path->slots[level] = orig_slot;
1987 btrfs_tree_unlock(mid);
1988 free_extent_buffer(mid);
1991 orig_slot -= btrfs_header_nritems(left);
1992 path->slots[level] = orig_slot;
1995 /* double check we haven't messed things up */
1997 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2001 btrfs_tree_unlock(right);
2002 free_extent_buffer(right);
2005 if (path->nodes[level] != left)
2006 btrfs_tree_unlock(left);
2007 free_extent_buffer(left);
2012 /* Node balancing for insertion. Here we only split or push nodes around
2013 * when they are completely full. This is also done top down, so we
2014 * have to be pessimistic.
2016 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2017 struct btrfs_root *root,
2018 struct btrfs_path *path, int level)
2020 struct btrfs_fs_info *fs_info = root->fs_info;
2021 struct extent_buffer *right = NULL;
2022 struct extent_buffer *mid;
2023 struct extent_buffer *left = NULL;
2024 struct extent_buffer *parent = NULL;
2028 int orig_slot = path->slots[level];
2033 mid = path->nodes[level];
2034 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2036 if (level < BTRFS_MAX_LEVEL - 1) {
2037 parent = path->nodes[level + 1];
2038 pslot = path->slots[level + 1];
2044 left = read_node_slot(fs_info, parent, pslot - 1);
2048 /* first, try to make some room in the middle buffer */
2052 btrfs_tree_lock(left);
2053 btrfs_set_lock_blocking(left);
2055 left_nr = btrfs_header_nritems(left);
2056 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2059 ret = btrfs_cow_block(trans, root, left, parent,
2064 wret = push_node_left(trans, fs_info,
2071 struct btrfs_disk_key disk_key;
2072 orig_slot += left_nr;
2073 btrfs_node_key(mid, &disk_key, 0);
2074 ret = tree_mod_log_insert_key(parent, pslot,
2075 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2077 btrfs_set_node_key(parent, &disk_key, pslot);
2078 btrfs_mark_buffer_dirty(parent);
2079 if (btrfs_header_nritems(left) > orig_slot) {
2080 path->nodes[level] = left;
2081 path->slots[level + 1] -= 1;
2082 path->slots[level] = orig_slot;
2083 btrfs_tree_unlock(mid);
2084 free_extent_buffer(mid);
2087 btrfs_header_nritems(left);
2088 path->slots[level] = orig_slot;
2089 btrfs_tree_unlock(left);
2090 free_extent_buffer(left);
2094 btrfs_tree_unlock(left);
2095 free_extent_buffer(left);
2097 right = read_node_slot(fs_info, parent, pslot + 1);
2102 * then try to empty the right most buffer into the middle
2107 btrfs_tree_lock(right);
2108 btrfs_set_lock_blocking(right);
2110 right_nr = btrfs_header_nritems(right);
2111 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2114 ret = btrfs_cow_block(trans, root, right,
2120 wret = balance_node_right(trans, fs_info,
2127 struct btrfs_disk_key disk_key;
2129 btrfs_node_key(right, &disk_key, 0);
2130 ret = tree_mod_log_insert_key(parent, pslot + 1,
2131 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2133 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2134 btrfs_mark_buffer_dirty(parent);
2136 if (btrfs_header_nritems(mid) <= orig_slot) {
2137 path->nodes[level] = right;
2138 path->slots[level + 1] += 1;
2139 path->slots[level] = orig_slot -
2140 btrfs_header_nritems(mid);
2141 btrfs_tree_unlock(mid);
2142 free_extent_buffer(mid);
2144 btrfs_tree_unlock(right);
2145 free_extent_buffer(right);
2149 btrfs_tree_unlock(right);
2150 free_extent_buffer(right);
2156 * readahead one full node of leaves, finding things that are close
2157 * to the block in 'slot', and triggering ra on them.
2159 static void reada_for_search(struct btrfs_fs_info *fs_info,
2160 struct btrfs_path *path,
2161 int level, int slot, u64 objectid)
2163 struct extent_buffer *node;
2164 struct btrfs_disk_key disk_key;
2169 struct extent_buffer *eb;
2177 if (!path->nodes[level])
2180 node = path->nodes[level];
2182 search = btrfs_node_blockptr(node, slot);
2183 blocksize = fs_info->nodesize;
2184 eb = find_extent_buffer(fs_info, search);
2186 free_extent_buffer(eb);
2192 nritems = btrfs_header_nritems(node);
2196 if (path->reada == READA_BACK) {
2200 } else if (path->reada == READA_FORWARD) {
2205 if (path->reada == READA_BACK && objectid) {
2206 btrfs_node_key(node, &disk_key, nr);
2207 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2210 search = btrfs_node_blockptr(node, nr);
2211 if ((search <= target && target - search <= 65536) ||
2212 (search > target && search - target <= 65536)) {
2213 readahead_tree_block(fs_info, search);
2217 if ((nread > 65536 || nscan > 32))
2222 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2223 struct btrfs_path *path, int level)
2227 struct extent_buffer *parent;
2228 struct extent_buffer *eb;
2233 parent = path->nodes[level + 1];
2237 nritems = btrfs_header_nritems(parent);
2238 slot = path->slots[level + 1];
2241 block1 = btrfs_node_blockptr(parent, slot - 1);
2242 gen = btrfs_node_ptr_generation(parent, slot - 1);
2243 eb = find_extent_buffer(fs_info, block1);
2245 * if we get -eagain from btrfs_buffer_uptodate, we
2246 * don't want to return eagain here. That will loop
2249 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2251 free_extent_buffer(eb);
2253 if (slot + 1 < nritems) {
2254 block2 = btrfs_node_blockptr(parent, slot + 1);
2255 gen = btrfs_node_ptr_generation(parent, slot + 1);
2256 eb = find_extent_buffer(fs_info, block2);
2257 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2259 free_extent_buffer(eb);
2263 readahead_tree_block(fs_info, block1);
2265 readahead_tree_block(fs_info, block2);
2270 * when we walk down the tree, it is usually safe to unlock the higher layers
2271 * in the tree. The exceptions are when our path goes through slot 0, because
2272 * operations on the tree might require changing key pointers higher up in the
2275 * callers might also have set path->keep_locks, which tells this code to keep
2276 * the lock if the path points to the last slot in the block. This is part of
2277 * walking through the tree, and selecting the next slot in the higher block.
2279 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2280 * if lowest_unlock is 1, level 0 won't be unlocked
2282 static noinline void unlock_up(struct btrfs_path *path, int level,
2283 int lowest_unlock, int min_write_lock_level,
2284 int *write_lock_level)
2287 int skip_level = level;
2289 struct extent_buffer *t;
2291 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2292 if (!path->nodes[i])
2294 if (!path->locks[i])
2296 if (!no_skips && path->slots[i] == 0) {
2300 if (!no_skips && path->keep_locks) {
2303 nritems = btrfs_header_nritems(t);
2304 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2309 if (skip_level < i && i >= lowest_unlock)
2313 if (i >= lowest_unlock && i > skip_level) {
2314 btrfs_tree_unlock_rw(t, path->locks[i]);
2316 if (write_lock_level &&
2317 i > min_write_lock_level &&
2318 i <= *write_lock_level) {
2319 *write_lock_level = i - 1;
2326 * This releases any locks held in the path starting at level and
2327 * going all the way up to the root.
2329 * btrfs_search_slot will keep the lock held on higher nodes in a few
2330 * corner cases, such as COW of the block at slot zero in the node. This
2331 * ignores those rules, and it should only be called when there are no
2332 * more updates to be done higher up in the tree.
2334 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2338 if (path->keep_locks)
2341 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2342 if (!path->nodes[i])
2344 if (!path->locks[i])
2346 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2352 * helper function for btrfs_search_slot. The goal is to find a block
2353 * in cache without setting the path to blocking. If we find the block
2354 * we return zero and the path is unchanged.
2356 * If we can't find the block, we set the path blocking and do some
2357 * reada. -EAGAIN is returned and the search must be repeated.
2360 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2361 struct extent_buffer **eb_ret, int level, int slot,
2362 const struct btrfs_key *key)
2364 struct btrfs_fs_info *fs_info = root->fs_info;
2367 struct extent_buffer *b = *eb_ret;
2368 struct extent_buffer *tmp;
2369 struct btrfs_key first_key;
2373 blocknr = btrfs_node_blockptr(b, slot);
2374 gen = btrfs_node_ptr_generation(b, slot);
2375 parent_level = btrfs_header_level(b);
2376 btrfs_node_key_to_cpu(b, &first_key, slot);
2378 tmp = find_extent_buffer(fs_info, blocknr);
2380 /* first we do an atomic uptodate check */
2381 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2386 /* the pages were up to date, but we failed
2387 * the generation number check. Do a full
2388 * read for the generation number that is correct.
2389 * We must do this without dropping locks so
2390 * we can trust our generation number
2392 btrfs_set_path_blocking(p);
2394 /* now we're allowed to do a blocking uptodate check */
2395 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2400 free_extent_buffer(tmp);
2401 btrfs_release_path(p);
2406 * reduce lock contention at high levels
2407 * of the btree by dropping locks before
2408 * we read. Don't release the lock on the current
2409 * level because we need to walk this node to figure
2410 * out which blocks to read.
2412 btrfs_unlock_up_safe(p, level + 1);
2413 btrfs_set_path_blocking(p);
2415 if (p->reada != READA_NONE)
2416 reada_for_search(fs_info, p, level, slot, key->objectid);
2419 tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2423 * If the read above didn't mark this buffer up to date,
2424 * it will never end up being up to date. Set ret to EIO now
2425 * and give up so that our caller doesn't loop forever
2428 if (!extent_buffer_uptodate(tmp))
2430 free_extent_buffer(tmp);
2435 btrfs_release_path(p);
2440 * helper function for btrfs_search_slot. This does all of the checks
2441 * for node-level blocks and does any balancing required based on
2444 * If no extra work was required, zero is returned. If we had to
2445 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2449 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2450 struct btrfs_root *root, struct btrfs_path *p,
2451 struct extent_buffer *b, int level, int ins_len,
2452 int *write_lock_level)
2454 struct btrfs_fs_info *fs_info = root->fs_info;
2457 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2458 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2461 if (*write_lock_level < level + 1) {
2462 *write_lock_level = level + 1;
2463 btrfs_release_path(p);
2467 btrfs_set_path_blocking(p);
2468 reada_for_balance(fs_info, p, level);
2469 sret = split_node(trans, root, p, level);
2476 b = p->nodes[level];
2477 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2478 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2481 if (*write_lock_level < level + 1) {
2482 *write_lock_level = level + 1;
2483 btrfs_release_path(p);
2487 btrfs_set_path_blocking(p);
2488 reada_for_balance(fs_info, p, level);
2489 sret = balance_level(trans, root, p, level);
2495 b = p->nodes[level];
2497 btrfs_release_path(p);
2500 BUG_ON(btrfs_header_nritems(b) == 1);
2510 static void key_search_validate(struct extent_buffer *b,
2511 const struct btrfs_key *key,
2514 #ifdef CONFIG_BTRFS_ASSERT
2515 struct btrfs_disk_key disk_key;
2517 btrfs_cpu_key_to_disk(&disk_key, key);
2520 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2521 offsetof(struct btrfs_leaf, items[0].key),
2524 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2525 offsetof(struct btrfs_node, ptrs[0].key),
2530 static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
2531 int level, int *prev_cmp, int *slot)
2533 if (*prev_cmp != 0) {
2534 *prev_cmp = btrfs_bin_search(b, key, level, slot);
2538 key_search_validate(b, key, level);
2544 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2545 u64 iobjectid, u64 ioff, u8 key_type,
2546 struct btrfs_key *found_key)
2549 struct btrfs_key key;
2550 struct extent_buffer *eb;
2555 key.type = key_type;
2556 key.objectid = iobjectid;
2559 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2563 eb = path->nodes[0];
2564 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2565 ret = btrfs_next_leaf(fs_root, path);
2568 eb = path->nodes[0];
2571 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2572 if (found_key->type != key.type ||
2573 found_key->objectid != key.objectid)
2579 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2580 struct btrfs_path *p,
2581 int write_lock_level)
2583 struct btrfs_fs_info *fs_info = root->fs_info;
2584 struct extent_buffer *b;
2588 /* We try very hard to do read locks on the root */
2589 root_lock = BTRFS_READ_LOCK;
2591 if (p->search_commit_root) {
2593 * The commit roots are read only so we always do read locks,
2594 * and we always must hold the commit_root_sem when doing
2595 * searches on them, the only exception is send where we don't
2596 * want to block transaction commits for a long time, so
2597 * we need to clone the commit root in order to avoid races
2598 * with transaction commits that create a snapshot of one of
2599 * the roots used by a send operation.
2601 if (p->need_commit_sem) {
2602 down_read(&fs_info->commit_root_sem);
2603 b = btrfs_clone_extent_buffer(root->commit_root);
2604 up_read(&fs_info->commit_root_sem);
2606 return ERR_PTR(-ENOMEM);
2609 b = root->commit_root;
2610 extent_buffer_get(b);
2612 level = btrfs_header_level(b);
2614 * Ensure that all callers have set skip_locking when
2615 * p->search_commit_root = 1.
2617 ASSERT(p->skip_locking == 1);
2622 if (p->skip_locking) {
2623 b = btrfs_root_node(root);
2624 level = btrfs_header_level(b);
2629 * If the level is set to maximum, we can skip trying to get the read
2632 if (write_lock_level < BTRFS_MAX_LEVEL) {
2634 * We don't know the level of the root node until we actually
2635 * have it read locked
2637 b = btrfs_read_lock_root_node(root);
2638 level = btrfs_header_level(b);
2639 if (level > write_lock_level)
2642 /* Whoops, must trade for write lock */
2643 btrfs_tree_read_unlock(b);
2644 free_extent_buffer(b);
2647 b = btrfs_lock_root_node(root);
2648 root_lock = BTRFS_WRITE_LOCK;
2650 /* The level might have changed, check again */
2651 level = btrfs_header_level(b);
2654 p->nodes[level] = b;
2655 if (!p->skip_locking)
2656 p->locks[level] = root_lock;
2658 * Callers are responsible for dropping b's references.
2665 * btrfs_search_slot - look for a key in a tree and perform necessary
2666 * modifications to preserve tree invariants.
2668 * @trans: Handle of transaction, used when modifying the tree
2669 * @p: Holds all btree nodes along the search path
2670 * @root: The root node of the tree
2671 * @key: The key we are looking for
2672 * @ins_len: Indicates purpose of search, for inserts it is 1, for
2673 * deletions it's -1. 0 for plain searches
2674 * @cow: boolean should CoW operations be performed. Must always be 1
2675 * when modifying the tree.
2677 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2678 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2680 * If @key is found, 0 is returned and you can find the item in the leaf level
2681 * of the path (level 0)
2683 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2684 * points to the slot where it should be inserted
2686 * If an error is encountered while searching the tree a negative error number
2689 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2690 const struct btrfs_key *key, struct btrfs_path *p,
2691 int ins_len, int cow)
2693 struct btrfs_fs_info *fs_info = root->fs_info;
2694 struct extent_buffer *b;
2699 int lowest_unlock = 1;
2700 /* everything at write_lock_level or lower must be write locked */
2701 int write_lock_level = 0;
2702 u8 lowest_level = 0;
2703 int min_write_lock_level;
2706 lowest_level = p->lowest_level;
2707 WARN_ON(lowest_level && ins_len > 0);
2708 WARN_ON(p->nodes[0] != NULL);
2709 BUG_ON(!cow && ins_len);
2714 /* when we are removing items, we might have to go up to level
2715 * two as we update tree pointers Make sure we keep write
2716 * for those levels as well
2718 write_lock_level = 2;
2719 } else if (ins_len > 0) {
2721 * for inserting items, make sure we have a write lock on
2722 * level 1 so we can update keys
2724 write_lock_level = 1;
2728 write_lock_level = -1;
2730 if (cow && (p->keep_locks || p->lowest_level))
2731 write_lock_level = BTRFS_MAX_LEVEL;
2733 min_write_lock_level = write_lock_level;
2737 b = btrfs_search_slot_get_root(root, p, write_lock_level);
2744 level = btrfs_header_level(b);
2747 * setup the path here so we can release it under lock
2748 * contention with the cow code
2751 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2754 * if we don't really need to cow this block
2755 * then we don't want to set the path blocking,
2756 * so we test it here
2758 if (!should_cow_block(trans, root, b)) {
2759 trans->dirty = true;
2764 * must have write locks on this node and the
2767 if (level > write_lock_level ||
2768 (level + 1 > write_lock_level &&
2769 level + 1 < BTRFS_MAX_LEVEL &&
2770 p->nodes[level + 1])) {
2771 write_lock_level = level + 1;
2772 btrfs_release_path(p);
2776 btrfs_set_path_blocking(p);
2778 err = btrfs_cow_block(trans, root, b, NULL, 0,
2781 err = btrfs_cow_block(trans, root, b,
2782 p->nodes[level + 1],
2783 p->slots[level + 1], &b);
2790 p->nodes[level] = b;
2792 * Leave path with blocking locks to avoid massive
2793 * lock context switch, this is made on purpose.
2797 * we have a lock on b and as long as we aren't changing
2798 * the tree, there is no way to for the items in b to change.
2799 * It is safe to drop the lock on our parent before we
2800 * go through the expensive btree search on b.
2802 * If we're inserting or deleting (ins_len != 0), then we might
2803 * be changing slot zero, which may require changing the parent.
2804 * So, we can't drop the lock until after we know which slot
2805 * we're operating on.
2807 if (!ins_len && !p->keep_locks) {
2810 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2811 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2816 ret = key_search(b, key, level, &prev_cmp, &slot);
2822 if (ret && slot > 0) {
2826 p->slots[level] = slot;
2827 err = setup_nodes_for_search(trans, root, p, b, level,
2828 ins_len, &write_lock_level);
2835 b = p->nodes[level];
2836 slot = p->slots[level];
2839 * slot 0 is special, if we change the key
2840 * we have to update the parent pointer
2841 * which means we must have a write lock
2844 if (slot == 0 && ins_len &&
2845 write_lock_level < level + 1) {
2846 write_lock_level = level + 1;
2847 btrfs_release_path(p);
2851 unlock_up(p, level, lowest_unlock,
2852 min_write_lock_level, &write_lock_level);
2854 if (level == lowest_level) {
2860 err = read_block_for_search(root, p, &b, level,
2869 if (!p->skip_locking) {
2870 level = btrfs_header_level(b);
2871 if (level <= write_lock_level) {
2872 err = btrfs_try_tree_write_lock(b);
2874 btrfs_set_path_blocking(p);
2877 p->locks[level] = BTRFS_WRITE_LOCK;
2879 err = btrfs_tree_read_lock_atomic(b);
2881 btrfs_set_path_blocking(p);
2882 btrfs_tree_read_lock(b);
2884 p->locks[level] = BTRFS_READ_LOCK;
2886 p->nodes[level] = b;
2889 p->slots[level] = slot;
2891 btrfs_leaf_free_space(fs_info, b) < ins_len) {
2892 if (write_lock_level < 1) {
2893 write_lock_level = 1;
2894 btrfs_release_path(p);
2898 btrfs_set_path_blocking(p);
2899 err = split_leaf(trans, root, key,
2900 p, ins_len, ret == 0);
2908 if (!p->search_for_split)
2909 unlock_up(p, level, lowest_unlock,
2910 min_write_lock_level, NULL);
2917 * we don't really know what they plan on doing with the path
2918 * from here on, so for now just mark it as blocking
2920 if (!p->leave_spinning)
2921 btrfs_set_path_blocking(p);
2922 if (ret < 0 && !p->skip_release_on_error)
2923 btrfs_release_path(p);
2928 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2929 * current state of the tree together with the operations recorded in the tree
2930 * modification log to search for the key in a previous version of this tree, as
2931 * denoted by the time_seq parameter.
2933 * Naturally, there is no support for insert, delete or cow operations.
2935 * The resulting path and return value will be set up as if we called
2936 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2938 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2939 struct btrfs_path *p, u64 time_seq)
2941 struct btrfs_fs_info *fs_info = root->fs_info;
2942 struct extent_buffer *b;
2947 int lowest_unlock = 1;
2948 u8 lowest_level = 0;
2951 lowest_level = p->lowest_level;
2952 WARN_ON(p->nodes[0] != NULL);
2954 if (p->search_commit_root) {
2956 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2960 b = get_old_root(root, time_seq);
2965 level = btrfs_header_level(b);
2966 p->locks[level] = BTRFS_READ_LOCK;
2969 level = btrfs_header_level(b);
2970 p->nodes[level] = b;
2973 * we have a lock on b and as long as we aren't changing
2974 * the tree, there is no way to for the items in b to change.
2975 * It is safe to drop the lock on our parent before we
2976 * go through the expensive btree search on b.
2978 btrfs_unlock_up_safe(p, level + 1);
2981 * Since we can unwind ebs we want to do a real search every
2985 ret = key_search(b, key, level, &prev_cmp, &slot);
2989 if (ret && slot > 0) {
2993 p->slots[level] = slot;
2994 unlock_up(p, level, lowest_unlock, 0, NULL);
2996 if (level == lowest_level) {
3002 err = read_block_for_search(root, p, &b, level,
3011 level = btrfs_header_level(b);
3012 err = btrfs_tree_read_lock_atomic(b);
3014 btrfs_set_path_blocking(p);
3015 btrfs_tree_read_lock(b);
3017 b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3022 p->locks[level] = BTRFS_READ_LOCK;
3023 p->nodes[level] = b;
3025 p->slots[level] = slot;
3026 unlock_up(p, level, lowest_unlock, 0, NULL);
3032 if (!p->leave_spinning)
3033 btrfs_set_path_blocking(p);
3035 btrfs_release_path(p);
3041 * helper to use instead of search slot if no exact match is needed but
3042 * instead the next or previous item should be returned.
3043 * When find_higher is true, the next higher item is returned, the next lower
3045 * When return_any and find_higher are both true, and no higher item is found,
3046 * return the next lower instead.
3047 * When return_any is true and find_higher is false, and no lower item is found,
3048 * return the next higher instead.
3049 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3052 int btrfs_search_slot_for_read(struct btrfs_root *root,
3053 const struct btrfs_key *key,
3054 struct btrfs_path *p, int find_higher,
3058 struct extent_buffer *leaf;
3061 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3065 * a return value of 1 means the path is at the position where the
3066 * item should be inserted. Normally this is the next bigger item,
3067 * but in case the previous item is the last in a leaf, path points
3068 * to the first free slot in the previous leaf, i.e. at an invalid
3074 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3075 ret = btrfs_next_leaf(root, p);
3081 * no higher item found, return the next
3086 btrfs_release_path(p);
3090 if (p->slots[0] == 0) {
3091 ret = btrfs_prev_leaf(root, p);
3096 if (p->slots[0] == btrfs_header_nritems(leaf))
3103 * no lower item found, return the next
3108 btrfs_release_path(p);
3118 * adjust the pointers going up the tree, starting at level
3119 * making sure the right key of each node is points to 'key'.
3120 * This is used after shifting pointers to the left, so it stops
3121 * fixing up pointers when a given leaf/node is not in slot 0 of the
3125 static void fixup_low_keys(struct btrfs_path *path,
3126 struct btrfs_disk_key *key, int level)
3129 struct extent_buffer *t;
3132 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3133 int tslot = path->slots[i];
3135 if (!path->nodes[i])
3138 ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3141 btrfs_set_node_key(t, key, tslot);
3142 btrfs_mark_buffer_dirty(path->nodes[i]);
3151 * This function isn't completely safe. It's the caller's responsibility
3152 * that the new key won't break the order
3154 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3155 struct btrfs_path *path,
3156 const struct btrfs_key *new_key)
3158 struct btrfs_disk_key disk_key;
3159 struct extent_buffer *eb;
3162 eb = path->nodes[0];
3163 slot = path->slots[0];
3165 btrfs_item_key(eb, &disk_key, slot - 1);
3166 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3168 if (slot < btrfs_header_nritems(eb) - 1) {
3169 btrfs_item_key(eb, &disk_key, slot + 1);
3170 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3173 btrfs_cpu_key_to_disk(&disk_key, new_key);
3174 btrfs_set_item_key(eb, &disk_key, slot);
3175 btrfs_mark_buffer_dirty(eb);
3177 fixup_low_keys(path, &disk_key, 1);
3181 * try to push data from one node into the next node left in the
3184 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3185 * error, and > 0 if there was no room in the left hand block.
3187 static int push_node_left(struct btrfs_trans_handle *trans,
3188 struct btrfs_fs_info *fs_info,
3189 struct extent_buffer *dst,
3190 struct extent_buffer *src, int empty)
3197 src_nritems = btrfs_header_nritems(src);
3198 dst_nritems = btrfs_header_nritems(dst);
3199 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3200 WARN_ON(btrfs_header_generation(src) != trans->transid);
3201 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3203 if (!empty && src_nritems <= 8)
3206 if (push_items <= 0)
3210 push_items = min(src_nritems, push_items);
3211 if (push_items < src_nritems) {
3212 /* leave at least 8 pointers in the node if
3213 * we aren't going to empty it
3215 if (src_nritems - push_items < 8) {
3216 if (push_items <= 8)
3222 push_items = min(src_nritems - 8, push_items);
3224 ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0,
3227 btrfs_abort_transaction(trans, ret);
3230 copy_extent_buffer(dst, src,
3231 btrfs_node_key_ptr_offset(dst_nritems),
3232 btrfs_node_key_ptr_offset(0),
3233 push_items * sizeof(struct btrfs_key_ptr));
3235 if (push_items < src_nritems) {
3237 * Don't call tree_mod_log_insert_move here, key removal was
3238 * already fully logged by tree_mod_log_eb_copy above.
3240 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3241 btrfs_node_key_ptr_offset(push_items),
3242 (src_nritems - push_items) *
3243 sizeof(struct btrfs_key_ptr));
3245 btrfs_set_header_nritems(src, src_nritems - push_items);
3246 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3247 btrfs_mark_buffer_dirty(src);
3248 btrfs_mark_buffer_dirty(dst);
3254 * try to push data from one node into the next node right in the
3257 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3258 * error, and > 0 if there was no room in the right hand block.
3260 * this will only push up to 1/2 the contents of the left node over
3262 static int balance_node_right(struct btrfs_trans_handle *trans,
3263 struct btrfs_fs_info *fs_info,
3264 struct extent_buffer *dst,
3265 struct extent_buffer *src)
3273 WARN_ON(btrfs_header_generation(src) != trans->transid);
3274 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3276 src_nritems = btrfs_header_nritems(src);
3277 dst_nritems = btrfs_header_nritems(dst);
3278 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3279 if (push_items <= 0)
3282 if (src_nritems < 4)
3285 max_push = src_nritems / 2 + 1;
3286 /* don't try to empty the node */
3287 if (max_push >= src_nritems)
3290 if (max_push < push_items)
3291 push_items = max_push;
3293 ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3295 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3296 btrfs_node_key_ptr_offset(0),
3298 sizeof(struct btrfs_key_ptr));
3300 ret = tree_mod_log_eb_copy(fs_info, dst, src, 0,
3301 src_nritems - push_items, push_items);
3303 btrfs_abort_transaction(trans, ret);
3306 copy_extent_buffer(dst, src,
3307 btrfs_node_key_ptr_offset(0),
3308 btrfs_node_key_ptr_offset(src_nritems - push_items),
3309 push_items * sizeof(struct btrfs_key_ptr));
3311 btrfs_set_header_nritems(src, src_nritems - push_items);
3312 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3314 btrfs_mark_buffer_dirty(src);
3315 btrfs_mark_buffer_dirty(dst);
3321 * helper function to insert a new root level in the tree.
3322 * A new node is allocated, and a single item is inserted to
3323 * point to the existing root
3325 * returns zero on success or < 0 on failure.
3327 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3328 struct btrfs_root *root,
3329 struct btrfs_path *path, int level)
3331 struct btrfs_fs_info *fs_info = root->fs_info;
3333 struct extent_buffer *lower;
3334 struct extent_buffer *c;
3335 struct extent_buffer *old;
3336 struct btrfs_disk_key lower_key;
3339 BUG_ON(path->nodes[level]);
3340 BUG_ON(path->nodes[level-1] != root->node);
3342 lower = path->nodes[level-1];
3344 btrfs_item_key(lower, &lower_key, 0);
3346 btrfs_node_key(lower, &lower_key, 0);
3348 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3349 &lower_key, level, root->node->start, 0);
3353 root_add_used(root, fs_info->nodesize);
3355 btrfs_set_header_nritems(c, 1);
3356 btrfs_set_node_key(c, &lower_key, 0);
3357 btrfs_set_node_blockptr(c, 0, lower->start);
3358 lower_gen = btrfs_header_generation(lower);
3359 WARN_ON(lower_gen != trans->transid);
3361 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3363 btrfs_mark_buffer_dirty(c);
3366 ret = tree_mod_log_insert_root(root->node, c, 0);
3368 rcu_assign_pointer(root->node, c);
3370 /* the super has an extra ref to root->node */
3371 free_extent_buffer(old);
3373 add_root_to_dirty_list(root);
3374 extent_buffer_get(c);
3375 path->nodes[level] = c;
3376 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3377 path->slots[level] = 0;
3382 * worker function to insert a single pointer in a node.
3383 * the node should have enough room for the pointer already
3385 * slot and level indicate where you want the key to go, and
3386 * blocknr is the block the key points to.
3388 static void insert_ptr(struct btrfs_trans_handle *trans,
3389 struct btrfs_fs_info *fs_info, struct btrfs_path *path,
3390 struct btrfs_disk_key *key, u64 bytenr,
3391 int slot, int level)
3393 struct extent_buffer *lower;
3397 BUG_ON(!path->nodes[level]);
3398 btrfs_assert_tree_locked(path->nodes[level]);
3399 lower = path->nodes[level];
3400 nritems = btrfs_header_nritems(lower);
3401 BUG_ON(slot > nritems);
3402 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info));
3403 if (slot != nritems) {
3405 ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3409 memmove_extent_buffer(lower,
3410 btrfs_node_key_ptr_offset(slot + 1),
3411 btrfs_node_key_ptr_offset(slot),
3412 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3415 ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3419 btrfs_set_node_key(lower, key, slot);
3420 btrfs_set_node_blockptr(lower, slot, bytenr);
3421 WARN_ON(trans->transid == 0);
3422 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3423 btrfs_set_header_nritems(lower, nritems + 1);
3424 btrfs_mark_buffer_dirty(lower);
3428 * split the node at the specified level in path in two.
3429 * The path is corrected to point to the appropriate node after the split
3431 * Before splitting this tries to make some room in the node by pushing
3432 * left and right, if either one works, it returns right away.
3434 * returns 0 on success and < 0 on failure
3436 static noinline int split_node(struct btrfs_trans_handle *trans,
3437 struct btrfs_root *root,
3438 struct btrfs_path *path, int level)
3440 struct btrfs_fs_info *fs_info = root->fs_info;
3441 struct extent_buffer *c;
3442 struct extent_buffer *split;
3443 struct btrfs_disk_key disk_key;
3448 c = path->nodes[level];
3449 WARN_ON(btrfs_header_generation(c) != trans->transid);
3450 if (c == root->node) {
3452 * trying to split the root, lets make a new one
3454 * tree mod log: We don't log_removal old root in
3455 * insert_new_root, because that root buffer will be kept as a
3456 * normal node. We are going to log removal of half of the
3457 * elements below with tree_mod_log_eb_copy. We're holding a
3458 * tree lock on the buffer, which is why we cannot race with
3459 * other tree_mod_log users.
3461 ret = insert_new_root(trans, root, path, level + 1);
3465 ret = push_nodes_for_insert(trans, root, path, level);
3466 c = path->nodes[level];
3467 if (!ret && btrfs_header_nritems(c) <
3468 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3474 c_nritems = btrfs_header_nritems(c);
3475 mid = (c_nritems + 1) / 2;
3476 btrfs_node_key(c, &disk_key, mid);
3478 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3479 &disk_key, level, c->start, 0);
3481 return PTR_ERR(split);
3483 root_add_used(root, fs_info->nodesize);
3484 ASSERT(btrfs_header_level(c) == level);
3486 ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid);
3488 btrfs_abort_transaction(trans, ret);
3491 copy_extent_buffer(split, c,
3492 btrfs_node_key_ptr_offset(0),
3493 btrfs_node_key_ptr_offset(mid),
3494 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3495 btrfs_set_header_nritems(split, c_nritems - mid);
3496 btrfs_set_header_nritems(c, mid);
3499 btrfs_mark_buffer_dirty(c);
3500 btrfs_mark_buffer_dirty(split);
3502 insert_ptr(trans, fs_info, path, &disk_key, split->start,
3503 path->slots[level + 1] + 1, level + 1);
3505 if (path->slots[level] >= mid) {
3506 path->slots[level] -= mid;
3507 btrfs_tree_unlock(c);
3508 free_extent_buffer(c);
3509 path->nodes[level] = split;
3510 path->slots[level + 1] += 1;
3512 btrfs_tree_unlock(split);
3513 free_extent_buffer(split);
3519 * how many bytes are required to store the items in a leaf. start
3520 * and nr indicate which items in the leaf to check. This totals up the
3521 * space used both by the item structs and the item data
3523 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3525 struct btrfs_item *start_item;
3526 struct btrfs_item *end_item;
3527 struct btrfs_map_token token;
3529 int nritems = btrfs_header_nritems(l);
3530 int end = min(nritems, start + nr) - 1;
3534 btrfs_init_map_token(&token);
3535 start_item = btrfs_item_nr(start);
3536 end_item = btrfs_item_nr(end);
3537 data_len = btrfs_token_item_offset(l, start_item, &token) +
3538 btrfs_token_item_size(l, start_item, &token);
3539 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3540 data_len += sizeof(struct btrfs_item) * nr;
3541 WARN_ON(data_len < 0);
3546 * The space between the end of the leaf items and
3547 * the start of the leaf data. IOW, how much room
3548 * the leaf has left for both items and data
3550 noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info,
3551 struct extent_buffer *leaf)
3553 int nritems = btrfs_header_nritems(leaf);
3556 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3559 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3561 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3562 leaf_space_used(leaf, 0, nritems), nritems);
3568 * min slot controls the lowest index we're willing to push to the
3569 * right. We'll push up to and including min_slot, but no lower
3571 static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info,
3572 struct btrfs_path *path,
3573 int data_size, int empty,
3574 struct extent_buffer *right,
3575 int free_space, u32 left_nritems,
3578 struct extent_buffer *left = path->nodes[0];
3579 struct extent_buffer *upper = path->nodes[1];
3580 struct btrfs_map_token token;
3581 struct btrfs_disk_key disk_key;
3586 struct btrfs_item *item;
3592 btrfs_init_map_token(&token);
3597 nr = max_t(u32, 1, min_slot);
3599 if (path->slots[0] >= left_nritems)
3600 push_space += data_size;
3602 slot = path->slots[1];
3603 i = left_nritems - 1;
3605 item = btrfs_item_nr(i);
3607 if (!empty && push_items > 0) {
3608 if (path->slots[0] > i)
3610 if (path->slots[0] == i) {
3611 int space = btrfs_leaf_free_space(fs_info, left);
3612 if (space + push_space * 2 > free_space)
3617 if (path->slots[0] == i)
3618 push_space += data_size;
3620 this_item_size = btrfs_item_size(left, item);
3621 if (this_item_size + sizeof(*item) + push_space > free_space)
3625 push_space += this_item_size + sizeof(*item);
3631 if (push_items == 0)
3634 WARN_ON(!empty && push_items == left_nritems);
3636 /* push left to right */
3637 right_nritems = btrfs_header_nritems(right);
3639 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3640 push_space -= leaf_data_end(fs_info, left);
3642 /* make room in the right data area */
3643 data_end = leaf_data_end(fs_info, right);
3644 memmove_extent_buffer(right,
3645 BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3646 BTRFS_LEAF_DATA_OFFSET + data_end,
3647 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3649 /* copy from the left data area */
3650 copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3651 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3652 BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left),
3655 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3656 btrfs_item_nr_offset(0),
3657 right_nritems * sizeof(struct btrfs_item));
3659 /* copy the items from left to right */
3660 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3661 btrfs_item_nr_offset(left_nritems - push_items),
3662 push_items * sizeof(struct btrfs_item));
3664 /* update the item pointers */
3665 right_nritems += push_items;
3666 btrfs_set_header_nritems(right, right_nritems);
3667 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3668 for (i = 0; i < right_nritems; i++) {
3669 item = btrfs_item_nr(i);
3670 push_space -= btrfs_token_item_size(right, item, &token);
3671 btrfs_set_token_item_offset(right, item, push_space, &token);
3674 left_nritems -= push_items;
3675 btrfs_set_header_nritems(left, left_nritems);
3678 btrfs_mark_buffer_dirty(left);
3680 clean_tree_block(fs_info, left);
3682 btrfs_mark_buffer_dirty(right);
3684 btrfs_item_key(right, &disk_key, 0);
3685 btrfs_set_node_key(upper, &disk_key, slot + 1);
3686 btrfs_mark_buffer_dirty(upper);
3688 /* then fixup the leaf pointer in the path */
3689 if (path->slots[0] >= left_nritems) {
3690 path->slots[0] -= left_nritems;
3691 if (btrfs_header_nritems(path->nodes[0]) == 0)
3692 clean_tree_block(fs_info, path->nodes[0]);
3693 btrfs_tree_unlock(path->nodes[0]);
3694 free_extent_buffer(path->nodes[0]);
3695 path->nodes[0] = right;
3696 path->slots[1] += 1;
3698 btrfs_tree_unlock(right);
3699 free_extent_buffer(right);
3704 btrfs_tree_unlock(right);
3705 free_extent_buffer(right);
3710 * push some data in the path leaf to the right, trying to free up at
3711 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3713 * returns 1 if the push failed because the other node didn't have enough
3714 * room, 0 if everything worked out and < 0 if there were major errors.
3716 * this will push starting from min_slot to the end of the leaf. It won't
3717 * push any slot lower than min_slot
3719 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3720 *root, struct btrfs_path *path,
3721 int min_data_size, int data_size,
3722 int empty, u32 min_slot)
3724 struct btrfs_fs_info *fs_info = root->fs_info;
3725 struct extent_buffer *left = path->nodes[0];
3726 struct extent_buffer *right;
3727 struct extent_buffer *upper;
3733 if (!path->nodes[1])
3736 slot = path->slots[1];
3737 upper = path->nodes[1];
3738 if (slot >= btrfs_header_nritems(upper) - 1)
3741 btrfs_assert_tree_locked(path->nodes[1]);
3743 right = read_node_slot(fs_info, upper, slot + 1);
3745 * slot + 1 is not valid or we fail to read the right node,
3746 * no big deal, just return.
3751 btrfs_tree_lock(right);
3752 btrfs_set_lock_blocking(right);
3754 free_space = btrfs_leaf_free_space(fs_info, right);
3755 if (free_space < data_size)
3758 /* cow and double check */
3759 ret = btrfs_cow_block(trans, root, right, upper,
3764 free_space = btrfs_leaf_free_space(fs_info, right);
3765 if (free_space < data_size)
3768 left_nritems = btrfs_header_nritems(left);
3769 if (left_nritems == 0)
3772 if (path->slots[0] == left_nritems && !empty) {
3773 /* Key greater than all keys in the leaf, right neighbor has
3774 * enough room for it and we're not emptying our leaf to delete
3775 * it, therefore use right neighbor to insert the new item and
3776 * no need to touch/dirty our left leaf. */
3777 btrfs_tree_unlock(left);
3778 free_extent_buffer(left);
3779 path->nodes[0] = right;
3785 return __push_leaf_right(fs_info, path, min_data_size, empty,
3786 right, free_space, left_nritems, min_slot);
3788 btrfs_tree_unlock(right);
3789 free_extent_buffer(right);
3794 * push some data in the path leaf to the left, trying to free up at
3795 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3797 * max_slot can put a limit on how far into the leaf we'll push items. The
3798 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3801 static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info,
3802 struct btrfs_path *path, int data_size,
3803 int empty, struct extent_buffer *left,
3804 int free_space, u32 right_nritems,
3807 struct btrfs_disk_key disk_key;
3808 struct extent_buffer *right = path->nodes[0];
3812 struct btrfs_item *item;
3813 u32 old_left_nritems;
3817 u32 old_left_item_size;
3818 struct btrfs_map_token token;
3820 btrfs_init_map_token(&token);
3823 nr = min(right_nritems, max_slot);
3825 nr = min(right_nritems - 1, max_slot);
3827 for (i = 0; i < nr; i++) {
3828 item = btrfs_item_nr(i);
3830 if (!empty && push_items > 0) {
3831 if (path->slots[0] < i)
3833 if (path->slots[0] == i) {
3834 int space = btrfs_leaf_free_space(fs_info, right);
3835 if (space + push_space * 2 > free_space)
3840 if (path->slots[0] == i)
3841 push_space += data_size;
3843 this_item_size = btrfs_item_size(right, item);
3844 if (this_item_size + sizeof(*item) + push_space > free_space)
3848 push_space += this_item_size + sizeof(*item);
3851 if (push_items == 0) {
3855 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3857 /* push data from right to left */
3858 copy_extent_buffer(left, right,
3859 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3860 btrfs_item_nr_offset(0),
3861 push_items * sizeof(struct btrfs_item));
3863 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3864 btrfs_item_offset_nr(right, push_items - 1);
3866 copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3867 leaf_data_end(fs_info, left) - push_space,
3868 BTRFS_LEAF_DATA_OFFSET +
3869 btrfs_item_offset_nr(right, push_items - 1),
3871 old_left_nritems = btrfs_header_nritems(left);
3872 BUG_ON(old_left_nritems <= 0);
3874 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3875 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3878 item = btrfs_item_nr(i);
3880 ioff = btrfs_token_item_offset(left, item, &token);
3881 btrfs_set_token_item_offset(left, item,
3882 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3885 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3887 /* fixup right node */
3888 if (push_items > right_nritems)
3889 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3892 if (push_items < right_nritems) {
3893 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3894 leaf_data_end(fs_info, right);
3895 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3896 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3897 BTRFS_LEAF_DATA_OFFSET +
3898 leaf_data_end(fs_info, right), push_space);
3900 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3901 btrfs_item_nr_offset(push_items),
3902 (btrfs_header_nritems(right) - push_items) *
3903 sizeof(struct btrfs_item));
3905 right_nritems -= push_items;
3906 btrfs_set_header_nritems(right, right_nritems);
3907 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3908 for (i = 0; i < right_nritems; i++) {
3909 item = btrfs_item_nr(i);
3911 push_space = push_space - btrfs_token_item_size(right,
3913 btrfs_set_token_item_offset(right, item, push_space, &token);
3916 btrfs_mark_buffer_dirty(left);
3918 btrfs_mark_buffer_dirty(right);
3920 clean_tree_block(fs_info, right);
3922 btrfs_item_key(right, &disk_key, 0);
3923 fixup_low_keys(path, &disk_key, 1);
3925 /* then fixup the leaf pointer in the path */
3926 if (path->slots[0] < push_items) {
3927 path->slots[0] += old_left_nritems;
3928 btrfs_tree_unlock(path->nodes[0]);
3929 free_extent_buffer(path->nodes[0]);
3930 path->nodes[0] = left;
3931 path->slots[1] -= 1;
3933 btrfs_tree_unlock(left);
3934 free_extent_buffer(left);
3935 path->slots[0] -= push_items;
3937 BUG_ON(path->slots[0] < 0);
3940 btrfs_tree_unlock(left);
3941 free_extent_buffer(left);
3946 * push some data in the path leaf to the left, trying to free up at
3947 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3949 * max_slot can put a limit on how far into the leaf we'll push items. The
3950 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3953 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3954 *root, struct btrfs_path *path, int min_data_size,
3955 int data_size, int empty, u32 max_slot)
3957 struct btrfs_fs_info *fs_info = root->fs_info;
3958 struct extent_buffer *right = path->nodes[0];
3959 struct extent_buffer *left;
3965 slot = path->slots[1];
3968 if (!path->nodes[1])
3971 right_nritems = btrfs_header_nritems(right);
3972 if (right_nritems == 0)
3975 btrfs_assert_tree_locked(path->nodes[1]);
3977 left = read_node_slot(fs_info, path->nodes[1], slot - 1);
3979 * slot - 1 is not valid or we fail to read the left node,
3980 * no big deal, just return.
3985 btrfs_tree_lock(left);
3986 btrfs_set_lock_blocking(left);
3988 free_space = btrfs_leaf_free_space(fs_info, left);
3989 if (free_space < data_size) {
3994 /* cow and double check */
3995 ret = btrfs_cow_block(trans, root, left,
3996 path->nodes[1], slot - 1, &left);
3998 /* we hit -ENOSPC, but it isn't fatal here */
4004 free_space = btrfs_leaf_free_space(fs_info, left);
4005 if (free_space < data_size) {
4010 return __push_leaf_left(fs_info, path, min_data_size,
4011 empty, left, free_space, right_nritems,
4014 btrfs_tree_unlock(left);
4015 free_extent_buffer(left);
4020 * split the path's leaf in two, making sure there is at least data_size
4021 * available for the resulting leaf level of the path.
4023 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4024 struct btrfs_fs_info *fs_info,
4025 struct btrfs_path *path,
4026 struct extent_buffer *l,
4027 struct extent_buffer *right,
4028 int slot, int mid, int nritems)
4033 struct btrfs_disk_key disk_key;
4034 struct btrfs_map_token token;
4036 btrfs_init_map_token(&token);
4038 nritems = nritems - mid;
4039 btrfs_set_header_nritems(right, nritems);
4040 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l);
4042 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4043 btrfs_item_nr_offset(mid),
4044 nritems * sizeof(struct btrfs_item));
4046 copy_extent_buffer(right, l,
4047 BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4048 data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4049 leaf_data_end(fs_info, l), data_copy_size);
4051 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4053 for (i = 0; i < nritems; i++) {
4054 struct btrfs_item *item = btrfs_item_nr(i);
4057 ioff = btrfs_token_item_offset(right, item, &token);
4058 btrfs_set_token_item_offset(right, item,
4059 ioff + rt_data_off, &token);
4062 btrfs_set_header_nritems(l, mid);
4063 btrfs_item_key(right, &disk_key, 0);
4064 insert_ptr(trans, fs_info, path, &disk_key, right->start,
4065 path->slots[1] + 1, 1);
4067 btrfs_mark_buffer_dirty(right);
4068 btrfs_mark_buffer_dirty(l);
4069 BUG_ON(path->slots[0] != slot);
4072 btrfs_tree_unlock(path->nodes[0]);
4073 free_extent_buffer(path->nodes[0]);
4074 path->nodes[0] = right;
4075 path->slots[0] -= mid;
4076 path->slots[1] += 1;
4078 btrfs_tree_unlock(right);
4079 free_extent_buffer(right);
4082 BUG_ON(path->slots[0] < 0);
4086 * double splits happen when we need to insert a big item in the middle
4087 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4088 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4091 * We avoid this by trying to push the items on either side of our target
4092 * into the adjacent leaves. If all goes well we can avoid the double split
4095 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4096 struct btrfs_root *root,
4097 struct btrfs_path *path,
4100 struct btrfs_fs_info *fs_info = root->fs_info;
4105 int space_needed = data_size;
4107 slot = path->slots[0];
4108 if (slot < btrfs_header_nritems(path->nodes[0]))
4109 space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4112 * try to push all the items after our slot into the
4115 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4122 nritems = btrfs_header_nritems(path->nodes[0]);
4124 * our goal is to get our slot at the start or end of a leaf. If
4125 * we've done so we're done
4127 if (path->slots[0] == 0 || path->slots[0] == nritems)
4130 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4133 /* try to push all the items before our slot into the next leaf */
4134 slot = path->slots[0];
4135 space_needed = data_size;
4137 space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4138 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4151 * split the path's leaf in two, making sure there is at least data_size
4152 * available for the resulting leaf level of the path.
4154 * returns 0 if all went well and < 0 on failure.
4156 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4157 struct btrfs_root *root,
4158 const struct btrfs_key *ins_key,
4159 struct btrfs_path *path, int data_size,
4162 struct btrfs_disk_key disk_key;
4163 struct extent_buffer *l;
4167 struct extent_buffer *right;
4168 struct btrfs_fs_info *fs_info = root->fs_info;
4172 int num_doubles = 0;
4173 int tried_avoid_double = 0;
4176 slot = path->slots[0];
4177 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4178 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4181 /* first try to make some room by pushing left and right */
4182 if (data_size && path->nodes[1]) {
4183 int space_needed = data_size;
4185 if (slot < btrfs_header_nritems(l))
4186 space_needed -= btrfs_leaf_free_space(fs_info, l);
4188 wret = push_leaf_right(trans, root, path, space_needed,
4189 space_needed, 0, 0);
4193 space_needed = data_size;
4195 space_needed -= btrfs_leaf_free_space(fs_info,
4197 wret = push_leaf_left(trans, root, path, space_needed,
4198 space_needed, 0, (u32)-1);
4204 /* did the pushes work? */
4205 if (btrfs_leaf_free_space(fs_info, l) >= data_size)
4209 if (!path->nodes[1]) {
4210 ret = insert_new_root(trans, root, path, 1);
4217 slot = path->slots[0];
4218 nritems = btrfs_header_nritems(l);
4219 mid = (nritems + 1) / 2;
4223 leaf_space_used(l, mid, nritems - mid) + data_size >
4224 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4225 if (slot >= nritems) {
4229 if (mid != nritems &&
4230 leaf_space_used(l, mid, nritems - mid) +
4231 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4232 if (data_size && !tried_avoid_double)
4233 goto push_for_double;
4239 if (leaf_space_used(l, 0, mid) + data_size >
4240 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4241 if (!extend && data_size && slot == 0) {
4243 } else if ((extend || !data_size) && slot == 0) {
4247 if (mid != nritems &&
4248 leaf_space_used(l, mid, nritems - mid) +
4249 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4250 if (data_size && !tried_avoid_double)
4251 goto push_for_double;
4259 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4261 btrfs_item_key(l, &disk_key, mid);
4263 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4264 &disk_key, 0, l->start, 0);
4266 return PTR_ERR(right);
4268 root_add_used(root, fs_info->nodesize);
4272 btrfs_set_header_nritems(right, 0);
4273 insert_ptr(trans, fs_info, path, &disk_key,
4274 right->start, path->slots[1] + 1, 1);
4275 btrfs_tree_unlock(path->nodes[0]);
4276 free_extent_buffer(path->nodes[0]);
4277 path->nodes[0] = right;
4279 path->slots[1] += 1;
4281 btrfs_set_header_nritems(right, 0);
4282 insert_ptr(trans, fs_info, path, &disk_key,
4283 right->start, path->slots[1], 1);
4284 btrfs_tree_unlock(path->nodes[0]);
4285 free_extent_buffer(path->nodes[0]);
4286 path->nodes[0] = right;
4288 if (path->slots[1] == 0)
4289 fixup_low_keys(path, &disk_key, 1);
4292 * We create a new leaf 'right' for the required ins_len and
4293 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4294 * the content of ins_len to 'right'.
4299 copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems);
4302 BUG_ON(num_doubles != 0);
4310 push_for_double_split(trans, root, path, data_size);
4311 tried_avoid_double = 1;
4312 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4317 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4318 struct btrfs_root *root,
4319 struct btrfs_path *path, int ins_len)
4321 struct btrfs_fs_info *fs_info = root->fs_info;
4322 struct btrfs_key key;
4323 struct extent_buffer *leaf;
4324 struct btrfs_file_extent_item *fi;
4329 leaf = path->nodes[0];
4330 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4332 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4333 key.type != BTRFS_EXTENT_CSUM_KEY);
4335 if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len)
4338 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4339 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4340 fi = btrfs_item_ptr(leaf, path->slots[0],
4341 struct btrfs_file_extent_item);
4342 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4344 btrfs_release_path(path);
4346 path->keep_locks = 1;
4347 path->search_for_split = 1;
4348 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4349 path->search_for_split = 0;
4356 leaf = path->nodes[0];
4357 /* if our item isn't there, return now */
4358 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4361 /* the leaf has changed, it now has room. return now */
4362 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len)
4365 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4366 fi = btrfs_item_ptr(leaf, path->slots[0],
4367 struct btrfs_file_extent_item);
4368 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4372 btrfs_set_path_blocking(path);
4373 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4377 path->keep_locks = 0;
4378 btrfs_unlock_up_safe(path, 1);
4381 path->keep_locks = 0;
4385 static noinline int split_item(struct btrfs_fs_info *fs_info,
4386 struct btrfs_path *path,
4387 const struct btrfs_key *new_key,
4388 unsigned long split_offset)
4390 struct extent_buffer *leaf;
4391 struct btrfs_item *item;
4392 struct btrfs_item *new_item;
4398 struct btrfs_disk_key disk_key;
4400 leaf = path->nodes[0];
4401 BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item));
4403 btrfs_set_path_blocking(path);
4405 item = btrfs_item_nr(path->slots[0]);
4406 orig_offset = btrfs_item_offset(leaf, item);
4407 item_size = btrfs_item_size(leaf, item);
4409 buf = kmalloc(item_size, GFP_NOFS);
4413 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4414 path->slots[0]), item_size);
4416 slot = path->slots[0] + 1;
4417 nritems = btrfs_header_nritems(leaf);
4418 if (slot != nritems) {
4419 /* shift the items */
4420 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4421 btrfs_item_nr_offset(slot),
4422 (nritems - slot) * sizeof(struct btrfs_item));
4425 btrfs_cpu_key_to_disk(&disk_key, new_key);
4426 btrfs_set_item_key(leaf, &disk_key, slot);
4428 new_item = btrfs_item_nr(slot);
4430 btrfs_set_item_offset(leaf, new_item, orig_offset);
4431 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4433 btrfs_set_item_offset(leaf, item,
4434 orig_offset + item_size - split_offset);
4435 btrfs_set_item_size(leaf, item, split_offset);
4437 btrfs_set_header_nritems(leaf, nritems + 1);
4439 /* write the data for the start of the original item */
4440 write_extent_buffer(leaf, buf,
4441 btrfs_item_ptr_offset(leaf, path->slots[0]),
4444 /* write the data for the new item */
4445 write_extent_buffer(leaf, buf + split_offset,
4446 btrfs_item_ptr_offset(leaf, slot),
4447 item_size - split_offset);
4448 btrfs_mark_buffer_dirty(leaf);
4450 BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0);
4456 * This function splits a single item into two items,
4457 * giving 'new_key' to the new item and splitting the
4458 * old one at split_offset (from the start of the item).
4460 * The path may be released by this operation. After
4461 * the split, the path is pointing to the old item. The
4462 * new item is going to be in the same node as the old one.
4464 * Note, the item being split must be smaller enough to live alone on
4465 * a tree block with room for one extra struct btrfs_item
4467 * This allows us to split the item in place, keeping a lock on the
4468 * leaf the entire time.
4470 int btrfs_split_item(struct btrfs_trans_handle *trans,
4471 struct btrfs_root *root,
4472 struct btrfs_path *path,
4473 const struct btrfs_key *new_key,
4474 unsigned long split_offset)
4477 ret = setup_leaf_for_split(trans, root, path,
4478 sizeof(struct btrfs_item));
4482 ret = split_item(root->fs_info, path, new_key, split_offset);
4487 * This function duplicate a item, giving 'new_key' to the new item.
4488 * It guarantees both items live in the same tree leaf and the new item
4489 * is contiguous with the original item.
4491 * This allows us to split file extent in place, keeping a lock on the
4492 * leaf the entire time.
4494 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4495 struct btrfs_root *root,
4496 struct btrfs_path *path,
4497 const struct btrfs_key *new_key)
4499 struct extent_buffer *leaf;
4503 leaf = path->nodes[0];
4504 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4505 ret = setup_leaf_for_split(trans, root, path,
4506 item_size + sizeof(struct btrfs_item));
4511 setup_items_for_insert(root, path, new_key, &item_size,
4512 item_size, item_size +
4513 sizeof(struct btrfs_item), 1);
4514 leaf = path->nodes[0];
4515 memcpy_extent_buffer(leaf,
4516 btrfs_item_ptr_offset(leaf, path->slots[0]),
4517 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4523 * make the item pointed to by the path smaller. new_size indicates
4524 * how small to make it, and from_end tells us if we just chop bytes
4525 * off the end of the item or if we shift the item to chop bytes off
4528 void btrfs_truncate_item(struct btrfs_fs_info *fs_info,
4529 struct btrfs_path *path, u32 new_size, int from_end)
4532 struct extent_buffer *leaf;
4533 struct btrfs_item *item;
4535 unsigned int data_end;
4536 unsigned int old_data_start;
4537 unsigned int old_size;
4538 unsigned int size_diff;
4540 struct btrfs_map_token token;
4542 btrfs_init_map_token(&token);
4544 leaf = path->nodes[0];
4545 slot = path->slots[0];
4547 old_size = btrfs_item_size_nr(leaf, slot);
4548 if (old_size == new_size)
4551 nritems = btrfs_header_nritems(leaf);
4552 data_end = leaf_data_end(fs_info, leaf);
4554 old_data_start = btrfs_item_offset_nr(leaf, slot);
4556 size_diff = old_size - new_size;
4559 BUG_ON(slot >= nritems);
4562 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4564 /* first correct the data pointers */
4565 for (i = slot; i < nritems; i++) {
4567 item = btrfs_item_nr(i);
4569 ioff = btrfs_token_item_offset(leaf, item, &token);
4570 btrfs_set_token_item_offset(leaf, item,
4571 ioff + size_diff, &token);
4574 /* shift the data */
4576 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4577 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4578 data_end, old_data_start + new_size - data_end);
4580 struct btrfs_disk_key disk_key;
4583 btrfs_item_key(leaf, &disk_key, slot);
4585 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4587 struct btrfs_file_extent_item *fi;
4589 fi = btrfs_item_ptr(leaf, slot,
4590 struct btrfs_file_extent_item);
4591 fi = (struct btrfs_file_extent_item *)(
4592 (unsigned long)fi - size_diff);
4594 if (btrfs_file_extent_type(leaf, fi) ==
4595 BTRFS_FILE_EXTENT_INLINE) {
4596 ptr = btrfs_item_ptr_offset(leaf, slot);
4597 memmove_extent_buffer(leaf, ptr,
4599 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4603 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4604 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4605 data_end, old_data_start - data_end);
4607 offset = btrfs_disk_key_offset(&disk_key);
4608 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4609 btrfs_set_item_key(leaf, &disk_key, slot);
4611 fixup_low_keys(path, &disk_key, 1);
4614 item = btrfs_item_nr(slot);
4615 btrfs_set_item_size(leaf, item, new_size);
4616 btrfs_mark_buffer_dirty(leaf);
4618 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4619 btrfs_print_leaf(leaf);
4625 * make the item pointed to by the path bigger, data_size is the added size.
4627 void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
4631 struct extent_buffer *leaf;
4632 struct btrfs_item *item;
4634 unsigned int data_end;
4635 unsigned int old_data;
4636 unsigned int old_size;
4638 struct btrfs_map_token token;
4640 btrfs_init_map_token(&token);
4642 leaf = path->nodes[0];
4644 nritems = btrfs_header_nritems(leaf);
4645 data_end = leaf_data_end(fs_info, leaf);
4647 if (btrfs_leaf_free_space(fs_info, leaf) < data_size) {
4648 btrfs_print_leaf(leaf);
4651 slot = path->slots[0];
4652 old_data = btrfs_item_end_nr(leaf, slot);
4655 if (slot >= nritems) {
4656 btrfs_print_leaf(leaf);
4657 btrfs_crit(fs_info, "slot %d too large, nritems %d",
4663 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4665 /* first correct the data pointers */
4666 for (i = slot; i < nritems; i++) {
4668 item = btrfs_item_nr(i);
4670 ioff = btrfs_token_item_offset(leaf, item, &token);
4671 btrfs_set_token_item_offset(leaf, item,
4672 ioff - data_size, &token);
4675 /* shift the data */
4676 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4677 data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4678 data_end, old_data - data_end);
4680 data_end = old_data;
4681 old_size = btrfs_item_size_nr(leaf, slot);
4682 item = btrfs_item_nr(slot);
4683 btrfs_set_item_size(leaf, item, old_size + data_size);
4684 btrfs_mark_buffer_dirty(leaf);
4686 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4687 btrfs_print_leaf(leaf);
4693 * this is a helper for btrfs_insert_empty_items, the main goal here is
4694 * to save stack depth by doing the bulk of the work in a function
4695 * that doesn't call btrfs_search_slot
4697 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4698 const struct btrfs_key *cpu_key, u32 *data_size,
4699 u32 total_data, u32 total_size, int nr)
4701 struct btrfs_fs_info *fs_info = root->fs_info;
4702 struct btrfs_item *item;
4705 unsigned int data_end;
4706 struct btrfs_disk_key disk_key;
4707 struct extent_buffer *leaf;
4709 struct btrfs_map_token token;
4711 if (path->slots[0] == 0) {
4712 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4713 fixup_low_keys(path, &disk_key, 1);
4715 btrfs_unlock_up_safe(path, 1);
4717 btrfs_init_map_token(&token);
4719 leaf = path->nodes[0];
4720 slot = path->slots[0];
4722 nritems = btrfs_header_nritems(leaf);
4723 data_end = leaf_data_end(fs_info, leaf);
4725 if (btrfs_leaf_free_space(fs_info, leaf) < total_size) {
4726 btrfs_print_leaf(leaf);
4727 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4728 total_size, btrfs_leaf_free_space(fs_info, leaf));
4732 if (slot != nritems) {
4733 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4735 if (old_data < data_end) {
4736 btrfs_print_leaf(leaf);
4737 btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4738 slot, old_data, data_end);
4742 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4744 /* first correct the data pointers */
4745 for (i = slot; i < nritems; i++) {
4748 item = btrfs_item_nr(i);
4749 ioff = btrfs_token_item_offset(leaf, item, &token);
4750 btrfs_set_token_item_offset(leaf, item,
4751 ioff - total_data, &token);
4753 /* shift the items */
4754 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4755 btrfs_item_nr_offset(slot),
4756 (nritems - slot) * sizeof(struct btrfs_item));
4758 /* shift the data */
4759 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4760 data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4761 data_end, old_data - data_end);
4762 data_end = old_data;
4765 /* setup the item for the new data */
4766 for (i = 0; i < nr; i++) {
4767 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4768 btrfs_set_item_key(leaf, &disk_key, slot + i);
4769 item = btrfs_item_nr(slot + i);
4770 btrfs_set_token_item_offset(leaf, item,
4771 data_end - data_size[i], &token);
4772 data_end -= data_size[i];
4773 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4776 btrfs_set_header_nritems(leaf, nritems + nr);
4777 btrfs_mark_buffer_dirty(leaf);
4779 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4780 btrfs_print_leaf(leaf);
4786 * Given a key and some data, insert items into the tree.
4787 * This does all the path init required, making room in the tree if needed.
4789 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4790 struct btrfs_root *root,
4791 struct btrfs_path *path,
4792 const struct btrfs_key *cpu_key, u32 *data_size,
4801 for (i = 0; i < nr; i++)
4802 total_data += data_size[i];
4804 total_size = total_data + (nr * sizeof(struct btrfs_item));
4805 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4811 slot = path->slots[0];
4814 setup_items_for_insert(root, path, cpu_key, data_size,
4815 total_data, total_size, nr);
4820 * Given a key and some data, insert an item into the tree.
4821 * This does all the path init required, making room in the tree if needed.
4823 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4824 const struct btrfs_key *cpu_key, void *data,
4828 struct btrfs_path *path;
4829 struct extent_buffer *leaf;
4832 path = btrfs_alloc_path();
4835 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4837 leaf = path->nodes[0];
4838 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4839 write_extent_buffer(leaf, data, ptr, data_size);
4840 btrfs_mark_buffer_dirty(leaf);
4842 btrfs_free_path(path);
4847 * delete the pointer from a given node.
4849 * the tree should have been previously balanced so the deletion does not
4852 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4853 int level, int slot)
4855 struct extent_buffer *parent = path->nodes[level];
4859 nritems = btrfs_header_nritems(parent);
4860 if (slot != nritems - 1) {
4862 ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4863 nritems - slot - 1);
4866 memmove_extent_buffer(parent,
4867 btrfs_node_key_ptr_offset(slot),
4868 btrfs_node_key_ptr_offset(slot + 1),
4869 sizeof(struct btrfs_key_ptr) *
4870 (nritems - slot - 1));
4872 ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4878 btrfs_set_header_nritems(parent, nritems);
4879 if (nritems == 0 && parent == root->node) {
4880 BUG_ON(btrfs_header_level(root->node) != 1);
4881 /* just turn the root into a leaf and break */
4882 btrfs_set_header_level(root->node, 0);
4883 } else if (slot == 0) {
4884 struct btrfs_disk_key disk_key;
4886 btrfs_node_key(parent, &disk_key, 0);
4887 fixup_low_keys(path, &disk_key, level + 1);
4889 btrfs_mark_buffer_dirty(parent);
4893 * a helper function to delete the leaf pointed to by path->slots[1] and
4896 * This deletes the pointer in path->nodes[1] and frees the leaf
4897 * block extent. zero is returned if it all worked out, < 0 otherwise.
4899 * The path must have already been setup for deleting the leaf, including
4900 * all the proper balancing. path->nodes[1] must be locked.
4902 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4903 struct btrfs_root *root,
4904 struct btrfs_path *path,
4905 struct extent_buffer *leaf)
4907 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4908 del_ptr(root, path, 1, path->slots[1]);
4911 * btrfs_free_extent is expensive, we want to make sure we
4912 * aren't holding any locks when we call it
4914 btrfs_unlock_up_safe(path, 0);
4916 root_sub_used(root, leaf->len);
4918 extent_buffer_get(leaf);
4919 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4920 free_extent_buffer_stale(leaf);
4923 * delete the item at the leaf level in path. If that empties
4924 * the leaf, remove it from the tree
4926 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4927 struct btrfs_path *path, int slot, int nr)
4929 struct btrfs_fs_info *fs_info = root->fs_info;
4930 struct extent_buffer *leaf;
4931 struct btrfs_item *item;
4938 struct btrfs_map_token token;
4940 btrfs_init_map_token(&token);
4942 leaf = path->nodes[0];
4943 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4945 for (i = 0; i < nr; i++)
4946 dsize += btrfs_item_size_nr(leaf, slot + i);
4948 nritems = btrfs_header_nritems(leaf);
4950 if (slot + nr != nritems) {
4951 int data_end = leaf_data_end(fs_info, leaf);
4953 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4955 BTRFS_LEAF_DATA_OFFSET + data_end,
4956 last_off - data_end);
4958 for (i = slot + nr; i < nritems; i++) {
4961 item = btrfs_item_nr(i);
4962 ioff = btrfs_token_item_offset(leaf, item, &token);
4963 btrfs_set_token_item_offset(leaf, item,
4964 ioff + dsize, &token);
4967 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4968 btrfs_item_nr_offset(slot + nr),
4969 sizeof(struct btrfs_item) *
4970 (nritems - slot - nr));
4972 btrfs_set_header_nritems(leaf, nritems - nr);
4975 /* delete the leaf if we've emptied it */
4977 if (leaf == root->node) {
4978 btrfs_set_header_level(leaf, 0);
4980 btrfs_set_path_blocking(path);
4981 clean_tree_block(fs_info, leaf);
4982 btrfs_del_leaf(trans, root, path, leaf);
4985 int used = leaf_space_used(leaf, 0, nritems);
4987 struct btrfs_disk_key disk_key;
4989 btrfs_item_key(leaf, &disk_key, 0);
4990 fixup_low_keys(path, &disk_key, 1);
4993 /* delete the leaf if it is mostly empty */
4994 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
4995 /* push_leaf_left fixes the path.
4996 * make sure the path still points to our leaf
4997 * for possible call to del_ptr below
4999 slot = path->slots[1];
5000 extent_buffer_get(leaf);
5002 btrfs_set_path_blocking(path);
5003 wret = push_leaf_left(trans, root, path, 1, 1,
5005 if (wret < 0 && wret != -ENOSPC)
5008 if (path->nodes[0] == leaf &&
5009 btrfs_header_nritems(leaf)) {
5010 wret = push_leaf_right(trans, root, path, 1,
5012 if (wret < 0 && wret != -ENOSPC)
5016 if (btrfs_header_nritems(leaf) == 0) {
5017 path->slots[1] = slot;
5018 btrfs_del_leaf(trans, root, path, leaf);
5019 free_extent_buffer(leaf);
5022 /* if we're still in the path, make sure
5023 * we're dirty. Otherwise, one of the
5024 * push_leaf functions must have already
5025 * dirtied this buffer
5027 if (path->nodes[0] == leaf)
5028 btrfs_mark_buffer_dirty(leaf);
5029 free_extent_buffer(leaf);
5032 btrfs_mark_buffer_dirty(leaf);
5039 * search the tree again to find a leaf with lesser keys
5040 * returns 0 if it found something or 1 if there are no lesser leaves.
5041 * returns < 0 on io errors.
5043 * This may release the path, and so you may lose any locks held at the
5046 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5048 struct btrfs_key key;
5049 struct btrfs_disk_key found_key;
5052 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5054 if (key.offset > 0) {
5056 } else if (key.type > 0) {
5058 key.offset = (u64)-1;
5059 } else if (key.objectid > 0) {
5062 key.offset = (u64)-1;
5067 btrfs_release_path(path);
5068 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5071 btrfs_item_key(path->nodes[0], &found_key, 0);
5072 ret = comp_keys(&found_key, &key);
5074 * We might have had an item with the previous key in the tree right
5075 * before we released our path. And after we released our path, that
5076 * item might have been pushed to the first slot (0) of the leaf we
5077 * were holding due to a tree balance. Alternatively, an item with the
5078 * previous key can exist as the only element of a leaf (big fat item).
5079 * Therefore account for these 2 cases, so that our callers (like
5080 * btrfs_previous_item) don't miss an existing item with a key matching
5081 * the previous key we computed above.
5089 * A helper function to walk down the tree starting at min_key, and looking
5090 * for nodes or leaves that are have a minimum transaction id.
5091 * This is used by the btree defrag code, and tree logging
5093 * This does not cow, but it does stuff the starting key it finds back
5094 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5095 * key and get a writable path.
5097 * This honors path->lowest_level to prevent descent past a given level
5100 * min_trans indicates the oldest transaction that you are interested
5101 * in walking through. Any nodes or leaves older than min_trans are
5102 * skipped over (without reading them).
5104 * returns zero if something useful was found, < 0 on error and 1 if there
5105 * was nothing in the tree that matched the search criteria.
5107 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5108 struct btrfs_path *path,
5111 struct btrfs_fs_info *fs_info = root->fs_info;
5112 struct extent_buffer *cur;
5113 struct btrfs_key found_key;
5119 int keep_locks = path->keep_locks;
5121 path->keep_locks = 1;
5123 cur = btrfs_read_lock_root_node(root);
5124 level = btrfs_header_level(cur);
5125 WARN_ON(path->nodes[level]);
5126 path->nodes[level] = cur;
5127 path->locks[level] = BTRFS_READ_LOCK;
5129 if (btrfs_header_generation(cur) < min_trans) {
5134 nritems = btrfs_header_nritems(cur);
5135 level = btrfs_header_level(cur);
5136 sret = btrfs_bin_search(cur, min_key, level, &slot);
5138 /* at the lowest level, we're done, setup the path and exit */
5139 if (level == path->lowest_level) {
5140 if (slot >= nritems)
5143 path->slots[level] = slot;
5144 btrfs_item_key_to_cpu(cur, &found_key, slot);
5147 if (sret && slot > 0)
5150 * check this node pointer against the min_trans parameters.
5151 * If it is too old, old, skip to the next one.
5153 while (slot < nritems) {
5156 gen = btrfs_node_ptr_generation(cur, slot);
5157 if (gen < min_trans) {
5165 * we didn't find a candidate key in this node, walk forward
5166 * and find another one
5168 if (slot >= nritems) {
5169 path->slots[level] = slot;
5170 btrfs_set_path_blocking(path);
5171 sret = btrfs_find_next_key(root, path, min_key, level,
5174 btrfs_release_path(path);
5180 /* save our key for returning back */
5181 btrfs_node_key_to_cpu(cur, &found_key, slot);
5182 path->slots[level] = slot;
5183 if (level == path->lowest_level) {
5187 btrfs_set_path_blocking(path);
5188 cur = read_node_slot(fs_info, cur, slot);
5194 btrfs_tree_read_lock(cur);
5196 path->locks[level - 1] = BTRFS_READ_LOCK;
5197 path->nodes[level - 1] = cur;
5198 unlock_up(path, level, 1, 0, NULL);
5201 path->keep_locks = keep_locks;
5203 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5204 btrfs_set_path_blocking(path);
5205 memcpy(min_key, &found_key, sizeof(found_key));
5210 static int tree_move_down(struct btrfs_fs_info *fs_info,
5211 struct btrfs_path *path,
5214 struct extent_buffer *eb;
5216 BUG_ON(*level == 0);
5217 eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]);
5221 path->nodes[*level - 1] = eb;
5222 path->slots[*level - 1] = 0;
5227 static int tree_move_next_or_upnext(struct btrfs_path *path,
5228 int *level, int root_level)
5232 nritems = btrfs_header_nritems(path->nodes[*level]);
5234 path->slots[*level]++;
5236 while (path->slots[*level] >= nritems) {
5237 if (*level == root_level)
5241 path->slots[*level] = 0;
5242 free_extent_buffer(path->nodes[*level]);
5243 path->nodes[*level] = NULL;
5245 path->slots[*level]++;
5247 nritems = btrfs_header_nritems(path->nodes[*level]);
5254 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5257 static int tree_advance(struct btrfs_fs_info *fs_info,
5258 struct btrfs_path *path,
5259 int *level, int root_level,
5261 struct btrfs_key *key)
5265 if (*level == 0 || !allow_down) {
5266 ret = tree_move_next_or_upnext(path, level, root_level);
5268 ret = tree_move_down(fs_info, path, level);
5272 btrfs_item_key_to_cpu(path->nodes[*level], key,
5273 path->slots[*level]);
5275 btrfs_node_key_to_cpu(path->nodes[*level], key,
5276 path->slots[*level]);
5281 static int tree_compare_item(struct btrfs_path *left_path,
5282 struct btrfs_path *right_path,
5287 unsigned long off1, off2;
5289 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5290 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5294 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5295 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5296 right_path->slots[0]);
5298 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5300 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5307 #define ADVANCE_ONLY_NEXT -1
5310 * This function compares two trees and calls the provided callback for
5311 * every changed/new/deleted item it finds.
5312 * If shared tree blocks are encountered, whole subtrees are skipped, making
5313 * the compare pretty fast on snapshotted subvolumes.
5315 * This currently works on commit roots only. As commit roots are read only,
5316 * we don't do any locking. The commit roots are protected with transactions.
5317 * Transactions are ended and rejoined when a commit is tried in between.
5319 * This function checks for modifications done to the trees while comparing.
5320 * If it detects a change, it aborts immediately.
5322 int btrfs_compare_trees(struct btrfs_root *left_root,
5323 struct btrfs_root *right_root,
5324 btrfs_changed_cb_t changed_cb, void *ctx)
5326 struct btrfs_fs_info *fs_info = left_root->fs_info;
5329 struct btrfs_path *left_path = NULL;
5330 struct btrfs_path *right_path = NULL;
5331 struct btrfs_key left_key;
5332 struct btrfs_key right_key;
5333 char *tmp_buf = NULL;
5334 int left_root_level;
5335 int right_root_level;
5338 int left_end_reached;
5339 int right_end_reached;
5347 left_path = btrfs_alloc_path();
5352 right_path = btrfs_alloc_path();
5358 tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
5364 left_path->search_commit_root = 1;
5365 left_path->skip_locking = 1;
5366 right_path->search_commit_root = 1;
5367 right_path->skip_locking = 1;
5370 * Strategy: Go to the first items of both trees. Then do
5372 * If both trees are at level 0
5373 * Compare keys of current items
5374 * If left < right treat left item as new, advance left tree
5376 * If left > right treat right item as deleted, advance right tree
5378 * If left == right do deep compare of items, treat as changed if
5379 * needed, advance both trees and repeat
5380 * If both trees are at the same level but not at level 0
5381 * Compare keys of current nodes/leafs
5382 * If left < right advance left tree and repeat
5383 * If left > right advance right tree and repeat
5384 * If left == right compare blockptrs of the next nodes/leafs
5385 * If they match advance both trees but stay at the same level
5387 * If they don't match advance both trees while allowing to go
5389 * If tree levels are different
5390 * Advance the tree that needs it and repeat
5392 * Advancing a tree means:
5393 * If we are at level 0, try to go to the next slot. If that's not
5394 * possible, go one level up and repeat. Stop when we found a level
5395 * where we could go to the next slot. We may at this point be on a
5398 * If we are not at level 0 and not on shared tree blocks, go one
5401 * If we are not at level 0 and on shared tree blocks, go one slot to
5402 * the right if possible or go up and right.
5405 down_read(&fs_info->commit_root_sem);
5406 left_level = btrfs_header_level(left_root->commit_root);
5407 left_root_level = left_level;
5408 left_path->nodes[left_level] =
5409 btrfs_clone_extent_buffer(left_root->commit_root);
5410 if (!left_path->nodes[left_level]) {
5411 up_read(&fs_info->commit_root_sem);
5416 right_level = btrfs_header_level(right_root->commit_root);
5417 right_root_level = right_level;
5418 right_path->nodes[right_level] =
5419 btrfs_clone_extent_buffer(right_root->commit_root);
5420 if (!right_path->nodes[right_level]) {
5421 up_read(&fs_info->commit_root_sem);
5425 up_read(&fs_info->commit_root_sem);
5427 if (left_level == 0)
5428 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5429 &left_key, left_path->slots[left_level]);
5431 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5432 &left_key, left_path->slots[left_level]);
5433 if (right_level == 0)
5434 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5435 &right_key, right_path->slots[right_level]);
5437 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5438 &right_key, right_path->slots[right_level]);
5440 left_end_reached = right_end_reached = 0;
5441 advance_left = advance_right = 0;
5444 if (advance_left && !left_end_reached) {
5445 ret = tree_advance(fs_info, left_path, &left_level,
5447 advance_left != ADVANCE_ONLY_NEXT,
5450 left_end_reached = ADVANCE;
5455 if (advance_right && !right_end_reached) {
5456 ret = tree_advance(fs_info, right_path, &right_level,
5458 advance_right != ADVANCE_ONLY_NEXT,
5461 right_end_reached = ADVANCE;
5467 if (left_end_reached && right_end_reached) {
5470 } else if (left_end_reached) {
5471 if (right_level == 0) {
5472 ret = changed_cb(left_path, right_path,
5474 BTRFS_COMPARE_TREE_DELETED,
5479 advance_right = ADVANCE;
5481 } else if (right_end_reached) {
5482 if (left_level == 0) {
5483 ret = changed_cb(left_path, right_path,
5485 BTRFS_COMPARE_TREE_NEW,
5490 advance_left = ADVANCE;
5494 if (left_level == 0 && right_level == 0) {
5495 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5497 ret = changed_cb(left_path, right_path,
5499 BTRFS_COMPARE_TREE_NEW,
5503 advance_left = ADVANCE;
5504 } else if (cmp > 0) {
5505 ret = changed_cb(left_path, right_path,
5507 BTRFS_COMPARE_TREE_DELETED,
5511 advance_right = ADVANCE;
5513 enum btrfs_compare_tree_result result;
5515 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5516 ret = tree_compare_item(left_path, right_path,
5519 result = BTRFS_COMPARE_TREE_CHANGED;
5521 result = BTRFS_COMPARE_TREE_SAME;
5522 ret = changed_cb(left_path, right_path,
5523 &left_key, result, ctx);
5526 advance_left = ADVANCE;
5527 advance_right = ADVANCE;
5529 } else if (left_level == right_level) {
5530 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5532 advance_left = ADVANCE;
5533 } else if (cmp > 0) {
5534 advance_right = ADVANCE;
5536 left_blockptr = btrfs_node_blockptr(
5537 left_path->nodes[left_level],
5538 left_path->slots[left_level]);
5539 right_blockptr = btrfs_node_blockptr(
5540 right_path->nodes[right_level],
5541 right_path->slots[right_level]);
5542 left_gen = btrfs_node_ptr_generation(
5543 left_path->nodes[left_level],
5544 left_path->slots[left_level]);
5545 right_gen = btrfs_node_ptr_generation(
5546 right_path->nodes[right_level],
5547 right_path->slots[right_level]);
5548 if (left_blockptr == right_blockptr &&
5549 left_gen == right_gen) {
5551 * As we're on a shared block, don't
5552 * allow to go deeper.
5554 advance_left = ADVANCE_ONLY_NEXT;
5555 advance_right = ADVANCE_ONLY_NEXT;
5557 advance_left = ADVANCE;
5558 advance_right = ADVANCE;
5561 } else if (left_level < right_level) {
5562 advance_right = ADVANCE;
5564 advance_left = ADVANCE;
5569 btrfs_free_path(left_path);
5570 btrfs_free_path(right_path);
5576 * this is similar to btrfs_next_leaf, but does not try to preserve
5577 * and fixup the path. It looks for and returns the next key in the
5578 * tree based on the current path and the min_trans parameters.
5580 * 0 is returned if another key is found, < 0 if there are any errors
5581 * and 1 is returned if there are no higher keys in the tree
5583 * path->keep_locks should be set to 1 on the search made before
5584 * calling this function.
5586 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5587 struct btrfs_key *key, int level, u64 min_trans)
5590 struct extent_buffer *c;
5592 WARN_ON(!path->keep_locks);
5593 while (level < BTRFS_MAX_LEVEL) {
5594 if (!path->nodes[level])
5597 slot = path->slots[level] + 1;
5598 c = path->nodes[level];
5600 if (slot >= btrfs_header_nritems(c)) {
5603 struct btrfs_key cur_key;
5604 if (level + 1 >= BTRFS_MAX_LEVEL ||
5605 !path->nodes[level + 1])
5608 if (path->locks[level + 1]) {
5613 slot = btrfs_header_nritems(c) - 1;
5615 btrfs_item_key_to_cpu(c, &cur_key, slot);
5617 btrfs_node_key_to_cpu(c, &cur_key, slot);
5619 orig_lowest = path->lowest_level;
5620 btrfs_release_path(path);
5621 path->lowest_level = level;
5622 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5624 path->lowest_level = orig_lowest;
5628 c = path->nodes[level];
5629 slot = path->slots[level];
5636 btrfs_item_key_to_cpu(c, key, slot);
5638 u64 gen = btrfs_node_ptr_generation(c, slot);
5640 if (gen < min_trans) {
5644 btrfs_node_key_to_cpu(c, key, slot);
5652 * search the tree again to find a leaf with greater keys
5653 * returns 0 if it found something or 1 if there are no greater leaves.
5654 * returns < 0 on io errors.
5656 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5658 return btrfs_next_old_leaf(root, path, 0);
5661 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5666 struct extent_buffer *c;
5667 struct extent_buffer *next;
5668 struct btrfs_key key;
5671 int old_spinning = path->leave_spinning;
5672 int next_rw_lock = 0;
5674 nritems = btrfs_header_nritems(path->nodes[0]);
5678 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5683 btrfs_release_path(path);
5685 path->keep_locks = 1;
5686 path->leave_spinning = 1;
5689 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5691 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5692 path->keep_locks = 0;
5697 nritems = btrfs_header_nritems(path->nodes[0]);
5699 * by releasing the path above we dropped all our locks. A balance
5700 * could have added more items next to the key that used to be
5701 * at the very end of the block. So, check again here and
5702 * advance the path if there are now more items available.
5704 if (nritems > 0 && path->slots[0] < nritems - 1) {
5711 * So the above check misses one case:
5712 * - after releasing the path above, someone has removed the item that
5713 * used to be at the very end of the block, and balance between leafs
5714 * gets another one with bigger key.offset to replace it.
5716 * This one should be returned as well, or we can get leaf corruption
5717 * later(esp. in __btrfs_drop_extents()).
5719 * And a bit more explanation about this check,
5720 * with ret > 0, the key isn't found, the path points to the slot
5721 * where it should be inserted, so the path->slots[0] item must be the
5724 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5729 while (level < BTRFS_MAX_LEVEL) {
5730 if (!path->nodes[level]) {
5735 slot = path->slots[level] + 1;
5736 c = path->nodes[level];
5737 if (slot >= btrfs_header_nritems(c)) {
5739 if (level == BTRFS_MAX_LEVEL) {
5747 btrfs_tree_unlock_rw(next, next_rw_lock);
5748 free_extent_buffer(next);
5752 next_rw_lock = path->locks[level];
5753 ret = read_block_for_search(root, path, &next, level,
5759 btrfs_release_path(path);
5763 if (!path->skip_locking) {
5764 ret = btrfs_try_tree_read_lock(next);
5765 if (!ret && time_seq) {
5767 * If we don't get the lock, we may be racing
5768 * with push_leaf_left, holding that lock while
5769 * itself waiting for the leaf we've currently
5770 * locked. To solve this situation, we give up
5771 * on our lock and cycle.
5773 free_extent_buffer(next);
5774 btrfs_release_path(path);
5779 btrfs_set_path_blocking(path);
5780 btrfs_tree_read_lock(next);
5782 next_rw_lock = BTRFS_READ_LOCK;
5786 path->slots[level] = slot;
5789 c = path->nodes[level];
5790 if (path->locks[level])
5791 btrfs_tree_unlock_rw(c, path->locks[level]);
5793 free_extent_buffer(c);
5794 path->nodes[level] = next;
5795 path->slots[level] = 0;
5796 if (!path->skip_locking)
5797 path->locks[level] = next_rw_lock;
5801 ret = read_block_for_search(root, path, &next, level,
5807 btrfs_release_path(path);
5811 if (!path->skip_locking) {
5812 ret = btrfs_try_tree_read_lock(next);
5814 btrfs_set_path_blocking(path);
5815 btrfs_tree_read_lock(next);
5817 next_rw_lock = BTRFS_READ_LOCK;
5822 unlock_up(path, 0, 1, 0, NULL);
5823 path->leave_spinning = old_spinning;
5825 btrfs_set_path_blocking(path);
5831 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5832 * searching until it gets past min_objectid or finds an item of 'type'
5834 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5836 int btrfs_previous_item(struct btrfs_root *root,
5837 struct btrfs_path *path, u64 min_objectid,
5840 struct btrfs_key found_key;
5841 struct extent_buffer *leaf;
5846 if (path->slots[0] == 0) {
5847 btrfs_set_path_blocking(path);
5848 ret = btrfs_prev_leaf(root, path);
5854 leaf = path->nodes[0];
5855 nritems = btrfs_header_nritems(leaf);
5858 if (path->slots[0] == nritems)
5861 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5862 if (found_key.objectid < min_objectid)
5864 if (found_key.type == type)
5866 if (found_key.objectid == min_objectid &&
5867 found_key.type < type)
5874 * search in extent tree to find a previous Metadata/Data extent item with
5877 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5879 int btrfs_previous_extent_item(struct btrfs_root *root,
5880 struct btrfs_path *path, u64 min_objectid)
5882 struct btrfs_key found_key;
5883 struct extent_buffer *leaf;
5888 if (path->slots[0] == 0) {
5889 btrfs_set_path_blocking(path);
5890 ret = btrfs_prev_leaf(root, path);
5896 leaf = path->nodes[0];
5897 nritems = btrfs_header_nritems(leaf);
5900 if (path->slots[0] == nritems)
5903 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5904 if (found_key.objectid < min_objectid)
5906 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5907 found_key.type == BTRFS_METADATA_ITEM_KEY)
5909 if (found_key.objectid == min_objectid &&
5910 found_key.type < BTRFS_EXTENT_ITEM_KEY)