2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/rbtree.h>
24 #include "transaction.h"
25 #include "print-tree.h"
28 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
29 *root, struct btrfs_path *path, int level);
30 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
31 *root, struct btrfs_key *ins_key,
32 struct btrfs_path *path, int data_size, int extend);
33 static int push_node_left(struct btrfs_trans_handle *trans,
34 struct btrfs_root *root, struct extent_buffer *dst,
35 struct extent_buffer *src, int empty);
36 static int balance_node_right(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 struct extent_buffer *dst_buf,
39 struct extent_buffer *src_buf);
40 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
42 static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
43 struct extent_buffer *eb);
45 struct btrfs_path *btrfs_alloc_path(void)
47 struct btrfs_path *path;
48 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
53 * set all locked nodes in the path to blocking locks. This should
54 * be done before scheduling
56 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
59 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
60 if (!p->nodes[i] || !p->locks[i])
62 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
63 if (p->locks[i] == BTRFS_READ_LOCK)
64 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
65 else if (p->locks[i] == BTRFS_WRITE_LOCK)
66 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
71 * reset all the locked nodes in the patch to spinning locks.
73 * held is used to keep lockdep happy, when lockdep is enabled
74 * we set held to a blocking lock before we go around and
75 * retake all the spinlocks in the path. You can safely use NULL
78 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
79 struct extent_buffer *held, int held_rw)
84 btrfs_set_lock_blocking_rw(held, held_rw);
85 if (held_rw == BTRFS_WRITE_LOCK)
86 held_rw = BTRFS_WRITE_LOCK_BLOCKING;
87 else if (held_rw == BTRFS_READ_LOCK)
88 held_rw = BTRFS_READ_LOCK_BLOCKING;
90 btrfs_set_path_blocking(p);
92 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
93 if (p->nodes[i] && p->locks[i]) {
94 btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
95 if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
96 p->locks[i] = BTRFS_WRITE_LOCK;
97 else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
98 p->locks[i] = BTRFS_READ_LOCK;
103 btrfs_clear_lock_blocking_rw(held, held_rw);
106 /* this also releases the path */
107 void btrfs_free_path(struct btrfs_path *p)
111 btrfs_release_path(p);
112 kmem_cache_free(btrfs_path_cachep, p);
116 * path release drops references on the extent buffers in the path
117 * and it drops any locks held by this path
119 * It is safe to call this on paths that no locks or extent buffers held.
121 noinline void btrfs_release_path(struct btrfs_path *p)
125 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
130 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
133 free_extent_buffer(p->nodes[i]);
139 * safely gets a reference on the root node of a tree. A lock
140 * is not taken, so a concurrent writer may put a different node
141 * at the root of the tree. See btrfs_lock_root_node for the
144 * The extent buffer returned by this has a reference taken, so
145 * it won't disappear. It may stop being the root of the tree
146 * at any time because there are no locks held.
148 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
150 struct extent_buffer *eb;
154 eb = rcu_dereference(root->node);
157 * RCU really hurts here, we could free up the root node because
158 * it was cow'ed but we may not get the new root node yet so do
159 * the inc_not_zero dance and if it doesn't work then
160 * synchronize_rcu and try again.
162 if (atomic_inc_not_zero(&eb->refs)) {
172 /* loop around taking references on and locking the root node of the
173 * tree until you end up with a lock on the root. A locked buffer
174 * is returned, with a reference held.
176 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
178 struct extent_buffer *eb;
181 eb = btrfs_root_node(root);
183 if (eb == root->node)
185 btrfs_tree_unlock(eb);
186 free_extent_buffer(eb);
191 /* loop around taking references on and locking the root node of the
192 * tree until you end up with a lock on the root. A locked buffer
193 * is returned, with a reference held.
195 static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
197 struct extent_buffer *eb;
200 eb = btrfs_root_node(root);
201 btrfs_tree_read_lock(eb);
202 if (eb == root->node)
204 btrfs_tree_read_unlock(eb);
205 free_extent_buffer(eb);
210 /* cowonly root (everything not a reference counted cow subvolume), just get
211 * put onto a simple dirty list. transaction.c walks this to make sure they
212 * get properly updated on disk.
214 static void add_root_to_dirty_list(struct btrfs_root *root)
216 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
217 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
220 spin_lock(&root->fs_info->trans_lock);
221 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
222 /* Want the extent tree to be the last on the list */
223 if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID)
224 list_move_tail(&root->dirty_list,
225 &root->fs_info->dirty_cowonly_roots);
227 list_move(&root->dirty_list,
228 &root->fs_info->dirty_cowonly_roots);
230 spin_unlock(&root->fs_info->trans_lock);
234 * used by snapshot creation to make a copy of a root for a tree with
235 * a given objectid. The buffer with the new root node is returned in
236 * cow_ret, and this func returns zero on success or a negative error code.
238 int btrfs_copy_root(struct btrfs_trans_handle *trans,
239 struct btrfs_root *root,
240 struct extent_buffer *buf,
241 struct extent_buffer **cow_ret, u64 new_root_objectid)
243 struct extent_buffer *cow;
246 struct btrfs_disk_key disk_key;
248 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
249 trans->transid != root->fs_info->running_transaction->transid);
250 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
251 trans->transid != root->last_trans);
253 level = btrfs_header_level(buf);
255 btrfs_item_key(buf, &disk_key, 0);
257 btrfs_node_key(buf, &disk_key, 0);
259 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
260 &disk_key, level, buf->start, 0);
264 copy_extent_buffer(cow, buf, 0, 0, cow->len);
265 btrfs_set_header_bytenr(cow, cow->start);
266 btrfs_set_header_generation(cow, trans->transid);
267 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
268 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
269 BTRFS_HEADER_FLAG_RELOC);
270 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
271 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
273 btrfs_set_header_owner(cow, new_root_objectid);
275 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
278 WARN_ON(btrfs_header_generation(buf) > trans->transid);
279 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
280 ret = btrfs_inc_ref(trans, root, cow, 1);
282 ret = btrfs_inc_ref(trans, root, cow, 0);
287 btrfs_mark_buffer_dirty(cow);
296 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
297 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
299 MOD_LOG_ROOT_REPLACE,
302 struct tree_mod_move {
307 struct tree_mod_root {
312 struct tree_mod_elem {
314 u64 index; /* shifted logical */
318 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
321 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
324 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
325 struct btrfs_disk_key key;
328 /* this is used for op == MOD_LOG_MOVE_KEYS */
329 struct tree_mod_move move;
331 /* this is used for op == MOD_LOG_ROOT_REPLACE */
332 struct tree_mod_root old_root;
335 static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
337 read_lock(&fs_info->tree_mod_log_lock);
340 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
342 read_unlock(&fs_info->tree_mod_log_lock);
345 static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
347 write_lock(&fs_info->tree_mod_log_lock);
350 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
352 write_unlock(&fs_info->tree_mod_log_lock);
356 * Pull a new tree mod seq number for our operation.
358 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
360 return atomic64_inc_return(&fs_info->tree_mod_seq);
364 * This adds a new blocker to the tree mod log's blocker list if the @elem
365 * passed does not already have a sequence number set. So when a caller expects
366 * to record tree modifications, it should ensure to set elem->seq to zero
367 * before calling btrfs_get_tree_mod_seq.
368 * Returns a fresh, unused tree log modification sequence number, even if no new
371 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
372 struct seq_list *elem)
374 tree_mod_log_write_lock(fs_info);
375 spin_lock(&fs_info->tree_mod_seq_lock);
377 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
378 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
380 spin_unlock(&fs_info->tree_mod_seq_lock);
381 tree_mod_log_write_unlock(fs_info);
386 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
387 struct seq_list *elem)
389 struct rb_root *tm_root;
390 struct rb_node *node;
391 struct rb_node *next;
392 struct seq_list *cur_elem;
393 struct tree_mod_elem *tm;
394 u64 min_seq = (u64)-1;
395 u64 seq_putting = elem->seq;
400 spin_lock(&fs_info->tree_mod_seq_lock);
401 list_del(&elem->list);
404 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
405 if (cur_elem->seq < min_seq) {
406 if (seq_putting > cur_elem->seq) {
408 * blocker with lower sequence number exists, we
409 * cannot remove anything from the log
411 spin_unlock(&fs_info->tree_mod_seq_lock);
414 min_seq = cur_elem->seq;
417 spin_unlock(&fs_info->tree_mod_seq_lock);
420 * anything that's lower than the lowest existing (read: blocked)
421 * sequence number can be removed from the tree.
423 tree_mod_log_write_lock(fs_info);
424 tm_root = &fs_info->tree_mod_log;
425 for (node = rb_first(tm_root); node; node = next) {
426 next = rb_next(node);
427 tm = container_of(node, struct tree_mod_elem, node);
428 if (tm->seq > min_seq)
430 rb_erase(node, tm_root);
433 tree_mod_log_write_unlock(fs_info);
437 * key order of the log:
440 * the index is the shifted logical of the *new* root node for root replace
441 * operations, or the shifted logical of the affected block for all other
444 * Note: must be called with write lock (tree_mod_log_write_lock).
447 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
449 struct rb_root *tm_root;
450 struct rb_node **new;
451 struct rb_node *parent = NULL;
452 struct tree_mod_elem *cur;
456 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
458 tm_root = &fs_info->tree_mod_log;
459 new = &tm_root->rb_node;
461 cur = container_of(*new, struct tree_mod_elem, node);
463 if (cur->index < tm->index)
464 new = &((*new)->rb_left);
465 else if (cur->index > tm->index)
466 new = &((*new)->rb_right);
467 else if (cur->seq < tm->seq)
468 new = &((*new)->rb_left);
469 else if (cur->seq > tm->seq)
470 new = &((*new)->rb_right);
475 rb_link_node(&tm->node, parent, new);
476 rb_insert_color(&tm->node, tm_root);
481 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
482 * returns zero with the tree_mod_log_lock acquired. The caller must hold
483 * this until all tree mod log insertions are recorded in the rb tree and then
484 * call tree_mod_log_write_unlock() to release.
486 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
487 struct extent_buffer *eb) {
489 if (list_empty(&(fs_info)->tree_mod_seq_list))
491 if (eb && btrfs_header_level(eb) == 0)
494 tree_mod_log_write_lock(fs_info);
495 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
496 tree_mod_log_write_unlock(fs_info);
503 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
504 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
505 struct extent_buffer *eb)
508 if (list_empty(&(fs_info)->tree_mod_seq_list))
510 if (eb && btrfs_header_level(eb) == 0)
516 static struct tree_mod_elem *
517 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
518 enum mod_log_op op, gfp_t flags)
520 struct tree_mod_elem *tm;
522 tm = kzalloc(sizeof(*tm), flags);
526 tm->index = eb->start >> PAGE_CACHE_SHIFT;
527 if (op != MOD_LOG_KEY_ADD) {
528 btrfs_node_key(eb, &tm->key, slot);
529 tm->blockptr = btrfs_node_blockptr(eb, slot);
533 tm->generation = btrfs_node_ptr_generation(eb, slot);
534 RB_CLEAR_NODE(&tm->node);
540 tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
541 struct extent_buffer *eb, int slot,
542 enum mod_log_op op, gfp_t flags)
544 struct tree_mod_elem *tm;
547 if (!tree_mod_need_log(fs_info, eb))
550 tm = alloc_tree_mod_elem(eb, slot, op, flags);
554 if (tree_mod_dont_log(fs_info, eb)) {
559 ret = __tree_mod_log_insert(fs_info, tm);
560 tree_mod_log_write_unlock(fs_info);
568 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
569 struct extent_buffer *eb, int dst_slot, int src_slot,
570 int nr_items, gfp_t flags)
572 struct tree_mod_elem *tm = NULL;
573 struct tree_mod_elem **tm_list = NULL;
578 if (!tree_mod_need_log(fs_info, eb))
581 tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), flags);
585 tm = kzalloc(sizeof(*tm), flags);
591 tm->index = eb->start >> PAGE_CACHE_SHIFT;
593 tm->move.dst_slot = dst_slot;
594 tm->move.nr_items = nr_items;
595 tm->op = MOD_LOG_MOVE_KEYS;
597 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
598 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
599 MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags);
606 if (tree_mod_dont_log(fs_info, eb))
611 * When we override something during the move, we log these removals.
612 * This can only happen when we move towards the beginning of the
613 * buffer, i.e. dst_slot < src_slot.
615 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
616 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
621 ret = __tree_mod_log_insert(fs_info, tm);
624 tree_mod_log_write_unlock(fs_info);
629 for (i = 0; i < nr_items; i++) {
630 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
631 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
635 tree_mod_log_write_unlock(fs_info);
643 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
644 struct tree_mod_elem **tm_list,
650 for (i = nritems - 1; i >= 0; i--) {
651 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
653 for (j = nritems - 1; j > i; j--)
654 rb_erase(&tm_list[j]->node,
655 &fs_info->tree_mod_log);
664 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
665 struct extent_buffer *old_root,
666 struct extent_buffer *new_root, gfp_t flags,
669 struct tree_mod_elem *tm = NULL;
670 struct tree_mod_elem **tm_list = NULL;
675 if (!tree_mod_need_log(fs_info, NULL))
678 if (log_removal && btrfs_header_level(old_root) > 0) {
679 nritems = btrfs_header_nritems(old_root);
680 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
686 for (i = 0; i < nritems; i++) {
687 tm_list[i] = alloc_tree_mod_elem(old_root, i,
688 MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags);
696 tm = kzalloc(sizeof(*tm), flags);
702 tm->index = new_root->start >> PAGE_CACHE_SHIFT;
703 tm->old_root.logical = old_root->start;
704 tm->old_root.level = btrfs_header_level(old_root);
705 tm->generation = btrfs_header_generation(old_root);
706 tm->op = MOD_LOG_ROOT_REPLACE;
708 if (tree_mod_dont_log(fs_info, NULL))
712 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
714 ret = __tree_mod_log_insert(fs_info, tm);
716 tree_mod_log_write_unlock(fs_info);
725 for (i = 0; i < nritems; i++)
734 static struct tree_mod_elem *
735 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
738 struct rb_root *tm_root;
739 struct rb_node *node;
740 struct tree_mod_elem *cur = NULL;
741 struct tree_mod_elem *found = NULL;
742 u64 index = start >> PAGE_CACHE_SHIFT;
744 tree_mod_log_read_lock(fs_info);
745 tm_root = &fs_info->tree_mod_log;
746 node = tm_root->rb_node;
748 cur = container_of(node, struct tree_mod_elem, node);
749 if (cur->index < index) {
750 node = node->rb_left;
751 } else if (cur->index > index) {
752 node = node->rb_right;
753 } else if (cur->seq < min_seq) {
754 node = node->rb_left;
755 } else if (!smallest) {
756 /* we want the node with the highest seq */
758 BUG_ON(found->seq > cur->seq);
760 node = node->rb_left;
761 } else if (cur->seq > min_seq) {
762 /* we want the node with the smallest seq */
764 BUG_ON(found->seq < cur->seq);
766 node = node->rb_right;
772 tree_mod_log_read_unlock(fs_info);
778 * this returns the element from the log with the smallest time sequence
779 * value that's in the log (the oldest log item). any element with a time
780 * sequence lower than min_seq will be ignored.
782 static struct tree_mod_elem *
783 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
786 return __tree_mod_log_search(fs_info, start, min_seq, 1);
790 * this returns the element from the log with the largest time sequence
791 * value that's in the log (the most recent log item). any element with
792 * a time sequence lower than min_seq will be ignored.
794 static struct tree_mod_elem *
795 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
797 return __tree_mod_log_search(fs_info, start, min_seq, 0);
801 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
802 struct extent_buffer *src, unsigned long dst_offset,
803 unsigned long src_offset, int nr_items)
806 struct tree_mod_elem **tm_list = NULL;
807 struct tree_mod_elem **tm_list_add, **tm_list_rem;
811 if (!tree_mod_need_log(fs_info, NULL))
814 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
817 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
822 tm_list_add = tm_list;
823 tm_list_rem = tm_list + nr_items;
824 for (i = 0; i < nr_items; i++) {
825 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
826 MOD_LOG_KEY_REMOVE, GFP_NOFS);
827 if (!tm_list_rem[i]) {
832 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
833 MOD_LOG_KEY_ADD, GFP_NOFS);
834 if (!tm_list_add[i]) {
840 if (tree_mod_dont_log(fs_info, NULL))
844 for (i = 0; i < nr_items; i++) {
845 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
848 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
853 tree_mod_log_write_unlock(fs_info);
859 for (i = 0; i < nr_items * 2; i++) {
860 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
861 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
865 tree_mod_log_write_unlock(fs_info);
872 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
873 int dst_offset, int src_offset, int nr_items)
876 ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
882 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
883 struct extent_buffer *eb, int slot, int atomic)
887 ret = tree_mod_log_insert_key(fs_info, eb, slot,
889 atomic ? GFP_ATOMIC : GFP_NOFS);
894 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
896 struct tree_mod_elem **tm_list = NULL;
901 if (btrfs_header_level(eb) == 0)
904 if (!tree_mod_need_log(fs_info, NULL))
907 nritems = btrfs_header_nritems(eb);
908 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
912 for (i = 0; i < nritems; i++) {
913 tm_list[i] = alloc_tree_mod_elem(eb, i,
914 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
921 if (tree_mod_dont_log(fs_info, eb))
924 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
925 tree_mod_log_write_unlock(fs_info);
933 for (i = 0; i < nritems; i++)
941 tree_mod_log_set_root_pointer(struct btrfs_root *root,
942 struct extent_buffer *new_root_node,
946 ret = tree_mod_log_insert_root(root->fs_info, root->node,
947 new_root_node, GFP_NOFS, log_removal);
952 * check if the tree block can be shared by multiple trees
954 int btrfs_block_can_be_shared(struct btrfs_root *root,
955 struct extent_buffer *buf)
958 * Tree blocks not in refernece counted trees and tree roots
959 * are never shared. If a block was allocated after the last
960 * snapshot and the block was not allocated by tree relocation,
961 * we know the block is not shared.
963 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
964 buf != root->node && buf != root->commit_root &&
965 (btrfs_header_generation(buf) <=
966 btrfs_root_last_snapshot(&root->root_item) ||
967 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
969 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
970 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
971 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
977 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
978 struct btrfs_root *root,
979 struct extent_buffer *buf,
980 struct extent_buffer *cow,
990 * Backrefs update rules:
992 * Always use full backrefs for extent pointers in tree block
993 * allocated by tree relocation.
995 * If a shared tree block is no longer referenced by its owner
996 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
997 * use full backrefs for extent pointers in tree block.
999 * If a tree block is been relocating
1000 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1001 * use full backrefs for extent pointers in tree block.
1002 * The reason for this is some operations (such as drop tree)
1003 * are only allowed for blocks use full backrefs.
1006 if (btrfs_block_can_be_shared(root, buf)) {
1007 ret = btrfs_lookup_extent_info(trans, root, buf->start,
1008 btrfs_header_level(buf), 1,
1014 btrfs_std_error(root->fs_info, ret, NULL);
1019 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1020 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1021 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1026 owner = btrfs_header_owner(buf);
1027 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
1028 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
1031 if ((owner == root->root_key.objectid ||
1032 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
1033 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1034 ret = btrfs_inc_ref(trans, root, buf, 1);
1035 BUG_ON(ret); /* -ENOMEM */
1037 if (root->root_key.objectid ==
1038 BTRFS_TREE_RELOC_OBJECTID) {
1039 ret = btrfs_dec_ref(trans, root, buf, 0);
1040 BUG_ON(ret); /* -ENOMEM */
1041 ret = btrfs_inc_ref(trans, root, cow, 1);
1042 BUG_ON(ret); /* -ENOMEM */
1044 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
1047 if (root->root_key.objectid ==
1048 BTRFS_TREE_RELOC_OBJECTID)
1049 ret = btrfs_inc_ref(trans, root, cow, 1);
1051 ret = btrfs_inc_ref(trans, root, cow, 0);
1052 BUG_ON(ret); /* -ENOMEM */
1054 if (new_flags != 0) {
1055 int level = btrfs_header_level(buf);
1057 ret = btrfs_set_disk_extent_flags(trans, root,
1060 new_flags, level, 0);
1065 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
1066 if (root->root_key.objectid ==
1067 BTRFS_TREE_RELOC_OBJECTID)
1068 ret = btrfs_inc_ref(trans, root, cow, 1);
1070 ret = btrfs_inc_ref(trans, root, cow, 0);
1071 BUG_ON(ret); /* -ENOMEM */
1072 ret = btrfs_dec_ref(trans, root, buf, 1);
1073 BUG_ON(ret); /* -ENOMEM */
1075 clean_tree_block(trans, root->fs_info, buf);
1082 * does the dirty work in cow of a single block. The parent block (if
1083 * supplied) is updated to point to the new cow copy. The new buffer is marked
1084 * dirty and returned locked. If you modify the block it needs to be marked
1087 * search_start -- an allocation hint for the new block
1089 * empty_size -- a hint that you plan on doing more cow. This is the size in
1090 * bytes the allocator should try to find free next to the block it returns.
1091 * This is just a hint and may be ignored by the allocator.
1093 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1094 struct btrfs_root *root,
1095 struct extent_buffer *buf,
1096 struct extent_buffer *parent, int parent_slot,
1097 struct extent_buffer **cow_ret,
1098 u64 search_start, u64 empty_size)
1100 struct btrfs_disk_key disk_key;
1101 struct extent_buffer *cow;
1104 int unlock_orig = 0;
1107 if (*cow_ret == buf)
1110 btrfs_assert_tree_locked(buf);
1112 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1113 trans->transid != root->fs_info->running_transaction->transid);
1114 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1115 trans->transid != root->last_trans);
1117 level = btrfs_header_level(buf);
1120 btrfs_item_key(buf, &disk_key, 0);
1122 btrfs_node_key(buf, &disk_key, 0);
1124 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1126 parent_start = parent->start;
1132 cow = btrfs_alloc_tree_block(trans, root, parent_start,
1133 root->root_key.objectid, &disk_key, level,
1134 search_start, empty_size);
1136 return PTR_ERR(cow);
1138 /* cow is set to blocking by btrfs_init_new_buffer */
1140 copy_extent_buffer(cow, buf, 0, 0, cow->len);
1141 btrfs_set_header_bytenr(cow, cow->start);
1142 btrfs_set_header_generation(cow, trans->transid);
1143 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1144 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1145 BTRFS_HEADER_FLAG_RELOC);
1146 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1147 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1149 btrfs_set_header_owner(cow, root->root_key.objectid);
1151 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
1154 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1156 btrfs_abort_transaction(trans, root, ret);
1160 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1161 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1163 btrfs_abort_transaction(trans, root, ret);
1168 if (buf == root->node) {
1169 WARN_ON(parent && parent != buf);
1170 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1171 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1172 parent_start = buf->start;
1176 extent_buffer_get(cow);
1177 tree_mod_log_set_root_pointer(root, cow, 1);
1178 rcu_assign_pointer(root->node, cow);
1180 btrfs_free_tree_block(trans, root, buf, parent_start,
1182 free_extent_buffer(buf);
1183 add_root_to_dirty_list(root);
1185 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1186 parent_start = parent->start;
1190 WARN_ON(trans->transid != btrfs_header_generation(parent));
1191 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1192 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1193 btrfs_set_node_blockptr(parent, parent_slot,
1195 btrfs_set_node_ptr_generation(parent, parent_slot,
1197 btrfs_mark_buffer_dirty(parent);
1199 ret = tree_mod_log_free_eb(root->fs_info, buf);
1201 btrfs_abort_transaction(trans, root, ret);
1205 btrfs_free_tree_block(trans, root, buf, parent_start,
1209 btrfs_tree_unlock(buf);
1210 free_extent_buffer_stale(buf);
1211 btrfs_mark_buffer_dirty(cow);
1217 * returns the logical address of the oldest predecessor of the given root.
1218 * entries older than time_seq are ignored.
1220 static struct tree_mod_elem *
1221 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1222 struct extent_buffer *eb_root, u64 time_seq)
1224 struct tree_mod_elem *tm;
1225 struct tree_mod_elem *found = NULL;
1226 u64 root_logical = eb_root->start;
1233 * the very last operation that's logged for a root is the replacement
1234 * operation (if it is replaced at all). this has the index of the *new*
1235 * root, making it the very first operation that's logged for this root.
1238 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1243 * if there are no tree operation for the oldest root, we simply
1244 * return it. this should only happen if that (old) root is at
1251 * if there's an operation that's not a root replacement, we
1252 * found the oldest version of our root. normally, we'll find a
1253 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1255 if (tm->op != MOD_LOG_ROOT_REPLACE)
1259 root_logical = tm->old_root.logical;
1263 /* if there's no old root to return, return what we found instead */
1271 * tm is a pointer to the first operation to rewind within eb. then, all
1272 * previous operations will be rewinded (until we reach something older than
1276 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1277 u64 time_seq, struct tree_mod_elem *first_tm)
1280 struct rb_node *next;
1281 struct tree_mod_elem *tm = first_tm;
1282 unsigned long o_dst;
1283 unsigned long o_src;
1284 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1286 n = btrfs_header_nritems(eb);
1287 tree_mod_log_read_lock(fs_info);
1288 while (tm && tm->seq >= time_seq) {
1290 * all the operations are recorded with the operator used for
1291 * the modification. as we're going backwards, we do the
1292 * opposite of each operation here.
1295 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1296 BUG_ON(tm->slot < n);
1298 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1299 case MOD_LOG_KEY_REMOVE:
1300 btrfs_set_node_key(eb, &tm->key, tm->slot);
1301 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1302 btrfs_set_node_ptr_generation(eb, tm->slot,
1306 case MOD_LOG_KEY_REPLACE:
1307 BUG_ON(tm->slot >= n);
1308 btrfs_set_node_key(eb, &tm->key, tm->slot);
1309 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1310 btrfs_set_node_ptr_generation(eb, tm->slot,
1313 case MOD_LOG_KEY_ADD:
1314 /* if a move operation is needed it's in the log */
1317 case MOD_LOG_MOVE_KEYS:
1318 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1319 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1320 memmove_extent_buffer(eb, o_dst, o_src,
1321 tm->move.nr_items * p_size);
1323 case MOD_LOG_ROOT_REPLACE:
1325 * this operation is special. for roots, this must be
1326 * handled explicitly before rewinding.
1327 * for non-roots, this operation may exist if the node
1328 * was a root: root A -> child B; then A gets empty and
1329 * B is promoted to the new root. in the mod log, we'll
1330 * have a root-replace operation for B, a tree block
1331 * that is no root. we simply ignore that operation.
1335 next = rb_next(&tm->node);
1338 tm = container_of(next, struct tree_mod_elem, node);
1339 if (tm->index != first_tm->index)
1342 tree_mod_log_read_unlock(fs_info);
1343 btrfs_set_header_nritems(eb, n);
1347 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1348 * is returned. If rewind operations happen, a fresh buffer is returned. The
1349 * returned buffer is always read-locked. If the returned buffer is not the
1350 * input buffer, the lock on the input buffer is released and the input buffer
1351 * is freed (its refcount is decremented).
1353 static struct extent_buffer *
1354 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1355 struct extent_buffer *eb, u64 time_seq)
1357 struct extent_buffer *eb_rewin;
1358 struct tree_mod_elem *tm;
1363 if (btrfs_header_level(eb) == 0)
1366 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1370 btrfs_set_path_blocking(path);
1371 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1373 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1374 BUG_ON(tm->slot != 0);
1375 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1377 btrfs_tree_read_unlock_blocking(eb);
1378 free_extent_buffer(eb);
1381 btrfs_set_header_bytenr(eb_rewin, eb->start);
1382 btrfs_set_header_backref_rev(eb_rewin,
1383 btrfs_header_backref_rev(eb));
1384 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1385 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1387 eb_rewin = btrfs_clone_extent_buffer(eb);
1389 btrfs_tree_read_unlock_blocking(eb);
1390 free_extent_buffer(eb);
1395 btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
1396 btrfs_tree_read_unlock_blocking(eb);
1397 free_extent_buffer(eb);
1399 extent_buffer_get(eb_rewin);
1400 btrfs_tree_read_lock(eb_rewin);
1401 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1402 WARN_ON(btrfs_header_nritems(eb_rewin) >
1403 BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1409 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1410 * value. If there are no changes, the current root->root_node is returned. If
1411 * anything changed in between, there's a fresh buffer allocated on which the
1412 * rewind operations are done. In any case, the returned buffer is read locked.
1413 * Returns NULL on error (with no locks held).
1415 static inline struct extent_buffer *
1416 get_old_root(struct btrfs_root *root, u64 time_seq)
1418 struct tree_mod_elem *tm;
1419 struct extent_buffer *eb = NULL;
1420 struct extent_buffer *eb_root;
1421 u64 eb_root_owner = 0;
1422 struct extent_buffer *old;
1423 struct tree_mod_root *old_root = NULL;
1424 u64 old_generation = 0;
1427 eb_root = btrfs_read_lock_root_node(root);
1428 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1432 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1433 old_root = &tm->old_root;
1434 old_generation = tm->generation;
1435 logical = old_root->logical;
1437 logical = eb_root->start;
1440 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1441 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1442 btrfs_tree_read_unlock(eb_root);
1443 free_extent_buffer(eb_root);
1444 old = read_tree_block(root, logical, 0);
1445 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1447 free_extent_buffer(old);
1448 btrfs_warn(root->fs_info,
1449 "failed to read tree block %llu from get_old_root", logical);
1451 eb = btrfs_clone_extent_buffer(old);
1452 free_extent_buffer(old);
1454 } else if (old_root) {
1455 eb_root_owner = btrfs_header_owner(eb_root);
1456 btrfs_tree_read_unlock(eb_root);
1457 free_extent_buffer(eb_root);
1458 eb = alloc_dummy_extent_buffer(root->fs_info, logical);
1460 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1461 eb = btrfs_clone_extent_buffer(eb_root);
1462 btrfs_tree_read_unlock_blocking(eb_root);
1463 free_extent_buffer(eb_root);
1468 extent_buffer_get(eb);
1469 btrfs_tree_read_lock(eb);
1471 btrfs_set_header_bytenr(eb, eb->start);
1472 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1473 btrfs_set_header_owner(eb, eb_root_owner);
1474 btrfs_set_header_level(eb, old_root->level);
1475 btrfs_set_header_generation(eb, old_generation);
1478 __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1480 WARN_ON(btrfs_header_level(eb) != 0);
1481 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1486 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1488 struct tree_mod_elem *tm;
1490 struct extent_buffer *eb_root = btrfs_root_node(root);
1492 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1493 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1494 level = tm->old_root.level;
1496 level = btrfs_header_level(eb_root);
1498 free_extent_buffer(eb_root);
1503 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1504 struct btrfs_root *root,
1505 struct extent_buffer *buf)
1507 if (btrfs_test_is_dummy_root(root))
1510 /* ensure we can see the force_cow */
1514 * We do not need to cow a block if
1515 * 1) this block is not created or changed in this transaction;
1516 * 2) this block does not belong to TREE_RELOC tree;
1517 * 3) the root is not forced COW.
1519 * What is forced COW:
1520 * when we create snapshot during commiting the transaction,
1521 * after we've finished coping src root, we must COW the shared
1522 * block to ensure the metadata consistency.
1524 if (btrfs_header_generation(buf) == trans->transid &&
1525 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1526 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1527 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1528 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1534 * cows a single block, see __btrfs_cow_block for the real work.
1535 * This version of it has extra checks so that a block isn't cow'd more than
1536 * once per transaction, as long as it hasn't been written yet
1538 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1539 struct btrfs_root *root, struct extent_buffer *buf,
1540 struct extent_buffer *parent, int parent_slot,
1541 struct extent_buffer **cow_ret)
1546 if (trans->transaction != root->fs_info->running_transaction)
1547 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1549 root->fs_info->running_transaction->transid);
1551 if (trans->transid != root->fs_info->generation)
1552 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1553 trans->transid, root->fs_info->generation);
1555 if (!should_cow_block(trans, root, buf)) {
1556 trans->dirty = true;
1561 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
1564 btrfs_set_lock_blocking(parent);
1565 btrfs_set_lock_blocking(buf);
1567 ret = __btrfs_cow_block(trans, root, buf, parent,
1568 parent_slot, cow_ret, search_start, 0);
1570 trace_btrfs_cow_block(root, buf, *cow_ret);
1576 * helper function for defrag to decide if two blocks pointed to by a
1577 * node are actually close by
1579 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1581 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1583 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1589 * compare two keys in a memcmp fashion
1591 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1593 struct btrfs_key k1;
1595 btrfs_disk_key_to_cpu(&k1, disk);
1597 return btrfs_comp_cpu_keys(&k1, k2);
1601 * same as comp_keys only with two btrfs_key's
1603 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1605 if (k1->objectid > k2->objectid)
1607 if (k1->objectid < k2->objectid)
1609 if (k1->type > k2->type)
1611 if (k1->type < k2->type)
1613 if (k1->offset > k2->offset)
1615 if (k1->offset < k2->offset)
1621 * this is used by the defrag code to go through all the
1622 * leaves pointed to by a node and reallocate them so that
1623 * disk order is close to key order
1625 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1626 struct btrfs_root *root, struct extent_buffer *parent,
1627 int start_slot, u64 *last_ret,
1628 struct btrfs_key *progress)
1630 struct extent_buffer *cur;
1633 u64 search_start = *last_ret;
1643 int progress_passed = 0;
1644 struct btrfs_disk_key disk_key;
1646 parent_level = btrfs_header_level(parent);
1648 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1649 WARN_ON(trans->transid != root->fs_info->generation);
1651 parent_nritems = btrfs_header_nritems(parent);
1652 blocksize = root->nodesize;
1653 end_slot = parent_nritems - 1;
1655 if (parent_nritems <= 1)
1658 btrfs_set_lock_blocking(parent);
1660 for (i = start_slot; i <= end_slot; i++) {
1663 btrfs_node_key(parent, &disk_key, i);
1664 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1667 progress_passed = 1;
1668 blocknr = btrfs_node_blockptr(parent, i);
1669 gen = btrfs_node_ptr_generation(parent, i);
1670 if (last_block == 0)
1671 last_block = blocknr;
1674 other = btrfs_node_blockptr(parent, i - 1);
1675 close = close_blocks(blocknr, other, blocksize);
1677 if (!close && i < end_slot) {
1678 other = btrfs_node_blockptr(parent, i + 1);
1679 close = close_blocks(blocknr, other, blocksize);
1682 last_block = blocknr;
1686 cur = btrfs_find_tree_block(root->fs_info, blocknr);
1688 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1691 if (!cur || !uptodate) {
1693 cur = read_tree_block(root, blocknr, gen);
1695 return PTR_ERR(cur);
1696 } else if (!extent_buffer_uptodate(cur)) {
1697 free_extent_buffer(cur);
1700 } else if (!uptodate) {
1701 err = btrfs_read_buffer(cur, gen);
1703 free_extent_buffer(cur);
1708 if (search_start == 0)
1709 search_start = last_block;
1711 btrfs_tree_lock(cur);
1712 btrfs_set_lock_blocking(cur);
1713 err = __btrfs_cow_block(trans, root, cur, parent, i,
1716 (end_slot - i) * blocksize));
1718 btrfs_tree_unlock(cur);
1719 free_extent_buffer(cur);
1722 search_start = cur->start;
1723 last_block = cur->start;
1724 *last_ret = search_start;
1725 btrfs_tree_unlock(cur);
1726 free_extent_buffer(cur);
1733 * search for key in the extent_buffer. The items start at offset p,
1734 * and they are item_size apart. There are 'max' items in p.
1736 * the slot in the array is returned via slot, and it points to
1737 * the place where you would insert key if it is not found in
1740 * slot may point to max if the key is bigger than all of the keys
1742 static noinline int generic_bin_search(struct extent_buffer *eb,
1744 int item_size, struct btrfs_key *key,
1751 struct btrfs_disk_key *tmp = NULL;
1752 struct btrfs_disk_key unaligned;
1753 unsigned long offset;
1755 unsigned long map_start = 0;
1756 unsigned long map_len = 0;
1759 while (low < high) {
1760 mid = (low + high) / 2;
1761 offset = p + mid * item_size;
1763 if (!kaddr || offset < map_start ||
1764 (offset + sizeof(struct btrfs_disk_key)) >
1765 map_start + map_len) {
1767 err = map_private_extent_buffer(eb, offset,
1768 sizeof(struct btrfs_disk_key),
1769 &kaddr, &map_start, &map_len);
1772 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1775 read_extent_buffer(eb, &unaligned,
1776 offset, sizeof(unaligned));
1781 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1784 ret = comp_keys(tmp, key);
1800 * simple bin_search frontend that does the right thing for
1803 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1804 int level, int *slot)
1807 return generic_bin_search(eb,
1808 offsetof(struct btrfs_leaf, items),
1809 sizeof(struct btrfs_item),
1810 key, btrfs_header_nritems(eb),
1813 return generic_bin_search(eb,
1814 offsetof(struct btrfs_node, ptrs),
1815 sizeof(struct btrfs_key_ptr),
1816 key, btrfs_header_nritems(eb),
1820 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1821 int level, int *slot)
1823 return bin_search(eb, key, level, slot);
1826 static void root_add_used(struct btrfs_root *root, u32 size)
1828 spin_lock(&root->accounting_lock);
1829 btrfs_set_root_used(&root->root_item,
1830 btrfs_root_used(&root->root_item) + size);
1831 spin_unlock(&root->accounting_lock);
1834 static void root_sub_used(struct btrfs_root *root, u32 size)
1836 spin_lock(&root->accounting_lock);
1837 btrfs_set_root_used(&root->root_item,
1838 btrfs_root_used(&root->root_item) - size);
1839 spin_unlock(&root->accounting_lock);
1842 /* given a node and slot number, this reads the blocks it points to. The
1843 * extent buffer is returned with a reference taken (but unlocked).
1844 * NULL is returned on error.
1846 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1847 struct extent_buffer *parent, int slot)
1849 int level = btrfs_header_level(parent);
1850 struct extent_buffer *eb;
1854 if (slot >= btrfs_header_nritems(parent))
1859 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1860 btrfs_node_ptr_generation(parent, slot));
1861 if (IS_ERR(eb) || !extent_buffer_uptodate(eb)) {
1863 free_extent_buffer(eb);
1871 * node level balancing, used to make sure nodes are in proper order for
1872 * item deletion. We balance from the top down, so we have to make sure
1873 * that a deletion won't leave an node completely empty later on.
1875 static noinline int balance_level(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *root,
1877 struct btrfs_path *path, int level)
1879 struct extent_buffer *right = NULL;
1880 struct extent_buffer *mid;
1881 struct extent_buffer *left = NULL;
1882 struct extent_buffer *parent = NULL;
1886 int orig_slot = path->slots[level];
1892 mid = path->nodes[level];
1894 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1895 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1896 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1898 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1900 if (level < BTRFS_MAX_LEVEL - 1) {
1901 parent = path->nodes[level + 1];
1902 pslot = path->slots[level + 1];
1906 * deal with the case where there is only one pointer in the root
1907 * by promoting the node below to a root
1910 struct extent_buffer *child;
1912 if (btrfs_header_nritems(mid) != 1)
1915 /* promote the child to a root */
1916 child = read_node_slot(root, mid, 0);
1919 btrfs_std_error(root->fs_info, ret, NULL);
1923 btrfs_tree_lock(child);
1924 btrfs_set_lock_blocking(child);
1925 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1927 btrfs_tree_unlock(child);
1928 free_extent_buffer(child);
1932 tree_mod_log_set_root_pointer(root, child, 1);
1933 rcu_assign_pointer(root->node, child);
1935 add_root_to_dirty_list(root);
1936 btrfs_tree_unlock(child);
1938 path->locks[level] = 0;
1939 path->nodes[level] = NULL;
1940 clean_tree_block(trans, root->fs_info, mid);
1941 btrfs_tree_unlock(mid);
1942 /* once for the path */
1943 free_extent_buffer(mid);
1945 root_sub_used(root, mid->len);
1946 btrfs_free_tree_block(trans, root, mid, 0, 1);
1947 /* once for the root ptr */
1948 free_extent_buffer_stale(mid);
1951 if (btrfs_header_nritems(mid) >
1952 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1955 left = read_node_slot(root, parent, pslot - 1);
1957 btrfs_tree_lock(left);
1958 btrfs_set_lock_blocking(left);
1959 wret = btrfs_cow_block(trans, root, left,
1960 parent, pslot - 1, &left);
1966 right = read_node_slot(root, parent, pslot + 1);
1968 btrfs_tree_lock(right);
1969 btrfs_set_lock_blocking(right);
1970 wret = btrfs_cow_block(trans, root, right,
1971 parent, pslot + 1, &right);
1978 /* first, try to make some room in the middle buffer */
1980 orig_slot += btrfs_header_nritems(left);
1981 wret = push_node_left(trans, root, left, mid, 1);
1987 * then try to empty the right most buffer into the middle
1990 wret = push_node_left(trans, root, mid, right, 1);
1991 if (wret < 0 && wret != -ENOSPC)
1993 if (btrfs_header_nritems(right) == 0) {
1994 clean_tree_block(trans, root->fs_info, right);
1995 btrfs_tree_unlock(right);
1996 del_ptr(root, path, level + 1, pslot + 1);
1997 root_sub_used(root, right->len);
1998 btrfs_free_tree_block(trans, root, right, 0, 1);
1999 free_extent_buffer_stale(right);
2002 struct btrfs_disk_key right_key;
2003 btrfs_node_key(right, &right_key, 0);
2004 tree_mod_log_set_node_key(root->fs_info, parent,
2006 btrfs_set_node_key(parent, &right_key, pslot + 1);
2007 btrfs_mark_buffer_dirty(parent);
2010 if (btrfs_header_nritems(mid) == 1) {
2012 * we're not allowed to leave a node with one item in the
2013 * tree during a delete. A deletion from lower in the tree
2014 * could try to delete the only pointer in this node.
2015 * So, pull some keys from the left.
2016 * There has to be a left pointer at this point because
2017 * otherwise we would have pulled some pointers from the
2022 btrfs_std_error(root->fs_info, ret, NULL);
2025 wret = balance_node_right(trans, root, mid, left);
2031 wret = push_node_left(trans, root, left, mid, 1);
2037 if (btrfs_header_nritems(mid) == 0) {
2038 clean_tree_block(trans, root->fs_info, mid);
2039 btrfs_tree_unlock(mid);
2040 del_ptr(root, path, level + 1, pslot);
2041 root_sub_used(root, mid->len);
2042 btrfs_free_tree_block(trans, root, mid, 0, 1);
2043 free_extent_buffer_stale(mid);
2046 /* update the parent key to reflect our changes */
2047 struct btrfs_disk_key mid_key;
2048 btrfs_node_key(mid, &mid_key, 0);
2049 tree_mod_log_set_node_key(root->fs_info, parent,
2051 btrfs_set_node_key(parent, &mid_key, pslot);
2052 btrfs_mark_buffer_dirty(parent);
2055 /* update the path */
2057 if (btrfs_header_nritems(left) > orig_slot) {
2058 extent_buffer_get(left);
2059 /* left was locked after cow */
2060 path->nodes[level] = left;
2061 path->slots[level + 1] -= 1;
2062 path->slots[level] = orig_slot;
2064 btrfs_tree_unlock(mid);
2065 free_extent_buffer(mid);
2068 orig_slot -= btrfs_header_nritems(left);
2069 path->slots[level] = orig_slot;
2072 /* double check we haven't messed things up */
2074 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2078 btrfs_tree_unlock(right);
2079 free_extent_buffer(right);
2082 if (path->nodes[level] != left)
2083 btrfs_tree_unlock(left);
2084 free_extent_buffer(left);
2089 /* Node balancing for insertion. Here we only split or push nodes around
2090 * when they are completely full. This is also done top down, so we
2091 * have to be pessimistic.
2093 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2094 struct btrfs_root *root,
2095 struct btrfs_path *path, int level)
2097 struct extent_buffer *right = NULL;
2098 struct extent_buffer *mid;
2099 struct extent_buffer *left = NULL;
2100 struct extent_buffer *parent = NULL;
2104 int orig_slot = path->slots[level];
2109 mid = path->nodes[level];
2110 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2112 if (level < BTRFS_MAX_LEVEL - 1) {
2113 parent = path->nodes[level + 1];
2114 pslot = path->slots[level + 1];
2120 left = read_node_slot(root, parent, pslot - 1);
2122 /* first, try to make some room in the middle buffer */
2126 btrfs_tree_lock(left);
2127 btrfs_set_lock_blocking(left);
2129 left_nr = btrfs_header_nritems(left);
2130 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2133 ret = btrfs_cow_block(trans, root, left, parent,
2138 wret = push_node_left(trans, root,
2145 struct btrfs_disk_key disk_key;
2146 orig_slot += left_nr;
2147 btrfs_node_key(mid, &disk_key, 0);
2148 tree_mod_log_set_node_key(root->fs_info, parent,
2150 btrfs_set_node_key(parent, &disk_key, pslot);
2151 btrfs_mark_buffer_dirty(parent);
2152 if (btrfs_header_nritems(left) > orig_slot) {
2153 path->nodes[level] = left;
2154 path->slots[level + 1] -= 1;
2155 path->slots[level] = orig_slot;
2156 btrfs_tree_unlock(mid);
2157 free_extent_buffer(mid);
2160 btrfs_header_nritems(left);
2161 path->slots[level] = orig_slot;
2162 btrfs_tree_unlock(left);
2163 free_extent_buffer(left);
2167 btrfs_tree_unlock(left);
2168 free_extent_buffer(left);
2170 right = read_node_slot(root, parent, pslot + 1);
2173 * then try to empty the right most buffer into the middle
2178 btrfs_tree_lock(right);
2179 btrfs_set_lock_blocking(right);
2181 right_nr = btrfs_header_nritems(right);
2182 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2185 ret = btrfs_cow_block(trans, root, right,
2191 wret = balance_node_right(trans, root,
2198 struct btrfs_disk_key disk_key;
2200 btrfs_node_key(right, &disk_key, 0);
2201 tree_mod_log_set_node_key(root->fs_info, parent,
2203 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2204 btrfs_mark_buffer_dirty(parent);
2206 if (btrfs_header_nritems(mid) <= orig_slot) {
2207 path->nodes[level] = right;
2208 path->slots[level + 1] += 1;
2209 path->slots[level] = orig_slot -
2210 btrfs_header_nritems(mid);
2211 btrfs_tree_unlock(mid);
2212 free_extent_buffer(mid);
2214 btrfs_tree_unlock(right);
2215 free_extent_buffer(right);
2219 btrfs_tree_unlock(right);
2220 free_extent_buffer(right);
2226 * readahead one full node of leaves, finding things that are close
2227 * to the block in 'slot', and triggering ra on them.
2229 static void reada_for_search(struct btrfs_root *root,
2230 struct btrfs_path *path,
2231 int level, int slot, u64 objectid)
2233 struct extent_buffer *node;
2234 struct btrfs_disk_key disk_key;
2240 int direction = path->reada;
2241 struct extent_buffer *eb;
2249 if (!path->nodes[level])
2252 node = path->nodes[level];
2254 search = btrfs_node_blockptr(node, slot);
2255 blocksize = root->nodesize;
2256 eb = btrfs_find_tree_block(root->fs_info, search);
2258 free_extent_buffer(eb);
2264 nritems = btrfs_header_nritems(node);
2268 if (direction < 0) {
2272 } else if (direction > 0) {
2277 if (path->reada < 0 && objectid) {
2278 btrfs_node_key(node, &disk_key, nr);
2279 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2282 search = btrfs_node_blockptr(node, nr);
2283 if ((search <= target && target - search <= 65536) ||
2284 (search > target && search - target <= 65536)) {
2285 gen = btrfs_node_ptr_generation(node, nr);
2286 readahead_tree_block(root, search);
2290 if ((nread > 65536 || nscan > 32))
2295 static noinline void reada_for_balance(struct btrfs_root *root,
2296 struct btrfs_path *path, int level)
2300 struct extent_buffer *parent;
2301 struct extent_buffer *eb;
2306 parent = path->nodes[level + 1];
2310 nritems = btrfs_header_nritems(parent);
2311 slot = path->slots[level + 1];
2314 block1 = btrfs_node_blockptr(parent, slot - 1);
2315 gen = btrfs_node_ptr_generation(parent, slot - 1);
2316 eb = btrfs_find_tree_block(root->fs_info, block1);
2318 * if we get -eagain from btrfs_buffer_uptodate, we
2319 * don't want to return eagain here. That will loop
2322 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2324 free_extent_buffer(eb);
2326 if (slot + 1 < nritems) {
2327 block2 = btrfs_node_blockptr(parent, slot + 1);
2328 gen = btrfs_node_ptr_generation(parent, slot + 1);
2329 eb = btrfs_find_tree_block(root->fs_info, block2);
2330 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2332 free_extent_buffer(eb);
2336 readahead_tree_block(root, block1);
2338 readahead_tree_block(root, block2);
2343 * when we walk down the tree, it is usually safe to unlock the higher layers
2344 * in the tree. The exceptions are when our path goes through slot 0, because
2345 * operations on the tree might require changing key pointers higher up in the
2348 * callers might also have set path->keep_locks, which tells this code to keep
2349 * the lock if the path points to the last slot in the block. This is part of
2350 * walking through the tree, and selecting the next slot in the higher block.
2352 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2353 * if lowest_unlock is 1, level 0 won't be unlocked
2355 static noinline void unlock_up(struct btrfs_path *path, int level,
2356 int lowest_unlock, int min_write_lock_level,
2357 int *write_lock_level)
2360 int skip_level = level;
2362 struct extent_buffer *t;
2364 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2365 if (!path->nodes[i])
2367 if (!path->locks[i])
2369 if (!no_skips && path->slots[i] == 0) {
2373 if (!no_skips && path->keep_locks) {
2376 nritems = btrfs_header_nritems(t);
2377 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2382 if (skip_level < i && i >= lowest_unlock)
2386 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2387 btrfs_tree_unlock_rw(t, path->locks[i]);
2389 if (write_lock_level &&
2390 i > min_write_lock_level &&
2391 i <= *write_lock_level) {
2392 *write_lock_level = i - 1;
2399 * This releases any locks held in the path starting at level and
2400 * going all the way up to the root.
2402 * btrfs_search_slot will keep the lock held on higher nodes in a few
2403 * corner cases, such as COW of the block at slot zero in the node. This
2404 * ignores those rules, and it should only be called when there are no
2405 * more updates to be done higher up in the tree.
2407 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2411 if (path->keep_locks)
2414 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2415 if (!path->nodes[i])
2417 if (!path->locks[i])
2419 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2425 * helper function for btrfs_search_slot. The goal is to find a block
2426 * in cache without setting the path to blocking. If we find the block
2427 * we return zero and the path is unchanged.
2429 * If we can't find the block, we set the path blocking and do some
2430 * reada. -EAGAIN is returned and the search must be repeated.
2433 read_block_for_search(struct btrfs_trans_handle *trans,
2434 struct btrfs_root *root, struct btrfs_path *p,
2435 struct extent_buffer **eb_ret, int level, int slot,
2436 struct btrfs_key *key, u64 time_seq)
2440 struct extent_buffer *b = *eb_ret;
2441 struct extent_buffer *tmp;
2444 blocknr = btrfs_node_blockptr(b, slot);
2445 gen = btrfs_node_ptr_generation(b, slot);
2447 tmp = btrfs_find_tree_block(root->fs_info, blocknr);
2449 /* first we do an atomic uptodate check */
2450 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2455 /* the pages were up to date, but we failed
2456 * the generation number check. Do a full
2457 * read for the generation number that is correct.
2458 * We must do this without dropping locks so
2459 * we can trust our generation number
2461 btrfs_set_path_blocking(p);
2463 /* now we're allowed to do a blocking uptodate check */
2464 ret = btrfs_read_buffer(tmp, gen);
2469 free_extent_buffer(tmp);
2470 btrfs_release_path(p);
2475 * reduce lock contention at high levels
2476 * of the btree by dropping locks before
2477 * we read. Don't release the lock on the current
2478 * level because we need to walk this node to figure
2479 * out which blocks to read.
2481 btrfs_unlock_up_safe(p, level + 1);
2482 btrfs_set_path_blocking(p);
2484 free_extent_buffer(tmp);
2486 reada_for_search(root, p, level, slot, key->objectid);
2489 tmp = read_tree_block(root, blocknr, gen);
2492 * If the read above didn't mark this buffer up to date,
2493 * it will never end up being up to date. Set ret to EIO now
2494 * and give up so that our caller doesn't loop forever
2497 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2499 free_extent_buffer(tmp);
2502 btrfs_release_path(p);
2507 * helper function for btrfs_search_slot. This does all of the checks
2508 * for node-level blocks and does any balancing required based on
2511 * If no extra work was required, zero is returned. If we had to
2512 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2516 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2517 struct btrfs_root *root, struct btrfs_path *p,
2518 struct extent_buffer *b, int level, int ins_len,
2519 int *write_lock_level)
2522 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2523 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2526 if (*write_lock_level < level + 1) {
2527 *write_lock_level = level + 1;
2528 btrfs_release_path(p);
2532 btrfs_set_path_blocking(p);
2533 reada_for_balance(root, p, level);
2534 sret = split_node(trans, root, p, level);
2535 btrfs_clear_path_blocking(p, NULL, 0);
2542 b = p->nodes[level];
2543 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2544 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2547 if (*write_lock_level < level + 1) {
2548 *write_lock_level = level + 1;
2549 btrfs_release_path(p);
2553 btrfs_set_path_blocking(p);
2554 reada_for_balance(root, p, level);
2555 sret = balance_level(trans, root, p, level);
2556 btrfs_clear_path_blocking(p, NULL, 0);
2562 b = p->nodes[level];
2564 btrfs_release_path(p);
2567 BUG_ON(btrfs_header_nritems(b) == 1);
2577 static void key_search_validate(struct extent_buffer *b,
2578 struct btrfs_key *key,
2581 #ifdef CONFIG_BTRFS_ASSERT
2582 struct btrfs_disk_key disk_key;
2584 btrfs_cpu_key_to_disk(&disk_key, key);
2587 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2588 offsetof(struct btrfs_leaf, items[0].key),
2591 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2592 offsetof(struct btrfs_node, ptrs[0].key),
2597 static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2598 int level, int *prev_cmp, int *slot)
2600 if (*prev_cmp != 0) {
2601 *prev_cmp = bin_search(b, key, level, slot);
2605 key_search_validate(b, key, level);
2611 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2612 u64 iobjectid, u64 ioff, u8 key_type,
2613 struct btrfs_key *found_key)
2616 struct btrfs_key key;
2617 struct extent_buffer *eb;
2622 key.type = key_type;
2623 key.objectid = iobjectid;
2626 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2630 eb = path->nodes[0];
2631 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2632 ret = btrfs_next_leaf(fs_root, path);
2635 eb = path->nodes[0];
2638 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2639 if (found_key->type != key.type ||
2640 found_key->objectid != key.objectid)
2647 * look for key in the tree. path is filled in with nodes along the way
2648 * if key is found, we return zero and you can find the item in the leaf
2649 * level of the path (level 0)
2651 * If the key isn't found, the path points to the slot where it should
2652 * be inserted, and 1 is returned. If there are other errors during the
2653 * search a negative error number is returned.
2655 * if ins_len > 0, nodes and leaves will be split as we walk down the
2656 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2659 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2660 *root, struct btrfs_key *key, struct btrfs_path *p, int
2663 struct extent_buffer *b;
2668 int lowest_unlock = 1;
2670 /* everything at write_lock_level or lower must be write locked */
2671 int write_lock_level = 0;
2672 u8 lowest_level = 0;
2673 int min_write_lock_level;
2676 lowest_level = p->lowest_level;
2677 WARN_ON(lowest_level && ins_len > 0);
2678 WARN_ON(p->nodes[0] != NULL);
2679 BUG_ON(!cow && ins_len);
2684 /* when we are removing items, we might have to go up to level
2685 * two as we update tree pointers Make sure we keep write
2686 * for those levels as well
2688 write_lock_level = 2;
2689 } else if (ins_len > 0) {
2691 * for inserting items, make sure we have a write lock on
2692 * level 1 so we can update keys
2694 write_lock_level = 1;
2698 write_lock_level = -1;
2700 if (cow && (p->keep_locks || p->lowest_level))
2701 write_lock_level = BTRFS_MAX_LEVEL;
2703 min_write_lock_level = write_lock_level;
2708 * we try very hard to do read locks on the root
2710 root_lock = BTRFS_READ_LOCK;
2712 if (p->search_commit_root) {
2714 * the commit roots are read only
2715 * so we always do read locks
2717 if (p->need_commit_sem)
2718 down_read(&root->fs_info->commit_root_sem);
2719 b = root->commit_root;
2720 extent_buffer_get(b);
2721 level = btrfs_header_level(b);
2722 if (p->need_commit_sem)
2723 up_read(&root->fs_info->commit_root_sem);
2724 if (!p->skip_locking)
2725 btrfs_tree_read_lock(b);
2727 if (p->skip_locking) {
2728 b = btrfs_root_node(root);
2729 level = btrfs_header_level(b);
2731 /* we don't know the level of the root node
2732 * until we actually have it read locked
2734 b = btrfs_read_lock_root_node(root);
2735 level = btrfs_header_level(b);
2736 if (level <= write_lock_level) {
2737 /* whoops, must trade for write lock */
2738 btrfs_tree_read_unlock(b);
2739 free_extent_buffer(b);
2740 b = btrfs_lock_root_node(root);
2741 root_lock = BTRFS_WRITE_LOCK;
2743 /* the level might have changed, check again */
2744 level = btrfs_header_level(b);
2748 p->nodes[level] = b;
2749 if (!p->skip_locking)
2750 p->locks[level] = root_lock;
2753 level = btrfs_header_level(b);
2756 * setup the path here so we can release it under lock
2757 * contention with the cow code
2760 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2763 * if we don't really need to cow this block
2764 * then we don't want to set the path blocking,
2765 * so we test it here
2767 if (!should_cow_block(trans, root, b)) {
2768 trans->dirty = true;
2773 * must have write locks on this node and the
2776 if (level > write_lock_level ||
2777 (level + 1 > write_lock_level &&
2778 level + 1 < BTRFS_MAX_LEVEL &&
2779 p->nodes[level + 1])) {
2780 write_lock_level = level + 1;
2781 btrfs_release_path(p);
2785 btrfs_set_path_blocking(p);
2787 err = btrfs_cow_block(trans, root, b, NULL, 0,
2790 err = btrfs_cow_block(trans, root, b,
2791 p->nodes[level + 1],
2792 p->slots[level + 1], &b);
2799 p->nodes[level] = b;
2800 btrfs_clear_path_blocking(p, NULL, 0);
2803 * we have a lock on b and as long as we aren't changing
2804 * the tree, there is no way to for the items in b to change.
2805 * It is safe to drop the lock on our parent before we
2806 * go through the expensive btree search on b.
2808 * If we're inserting or deleting (ins_len != 0), then we might
2809 * be changing slot zero, which may require changing the parent.
2810 * So, we can't drop the lock until after we know which slot
2811 * we're operating on.
2813 if (!ins_len && !p->keep_locks) {
2816 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2817 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2822 ret = key_search(b, key, level, &prev_cmp, &slot);
2826 if (ret && slot > 0) {
2830 p->slots[level] = slot;
2831 err = setup_nodes_for_search(trans, root, p, b, level,
2832 ins_len, &write_lock_level);
2839 b = p->nodes[level];
2840 slot = p->slots[level];
2843 * slot 0 is special, if we change the key
2844 * we have to update the parent pointer
2845 * which means we must have a write lock
2848 if (slot == 0 && ins_len &&
2849 write_lock_level < level + 1) {
2850 write_lock_level = level + 1;
2851 btrfs_release_path(p);
2855 unlock_up(p, level, lowest_unlock,
2856 min_write_lock_level, &write_lock_level);
2858 if (level == lowest_level) {
2864 err = read_block_for_search(trans, root, p,
2865 &b, level, slot, key, 0);
2873 if (!p->skip_locking) {
2874 level = btrfs_header_level(b);
2875 if (level <= write_lock_level) {
2876 err = btrfs_try_tree_write_lock(b);
2878 btrfs_set_path_blocking(p);
2880 btrfs_clear_path_blocking(p, b,
2883 p->locks[level] = BTRFS_WRITE_LOCK;
2885 err = btrfs_tree_read_lock_atomic(b);
2887 btrfs_set_path_blocking(p);
2888 btrfs_tree_read_lock(b);
2889 btrfs_clear_path_blocking(p, b,
2892 p->locks[level] = BTRFS_READ_LOCK;
2894 p->nodes[level] = b;
2897 p->slots[level] = slot;
2899 btrfs_leaf_free_space(root, b) < ins_len) {
2900 if (write_lock_level < 1) {
2901 write_lock_level = 1;
2902 btrfs_release_path(p);
2906 btrfs_set_path_blocking(p);
2907 err = split_leaf(trans, root, key,
2908 p, ins_len, ret == 0);
2909 btrfs_clear_path_blocking(p, NULL, 0);
2917 if (!p->search_for_split)
2918 unlock_up(p, level, lowest_unlock,
2919 min_write_lock_level, &write_lock_level);
2926 * we don't really know what they plan on doing with the path
2927 * from here on, so for now just mark it as blocking
2929 if (!p->leave_spinning)
2930 btrfs_set_path_blocking(p);
2931 if (ret < 0 && !p->skip_release_on_error)
2932 btrfs_release_path(p);
2937 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2938 * current state of the tree together with the operations recorded in the tree
2939 * modification log to search for the key in a previous version of this tree, as
2940 * denoted by the time_seq parameter.
2942 * Naturally, there is no support for insert, delete or cow operations.
2944 * The resulting path and return value will be set up as if we called
2945 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2947 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2948 struct btrfs_path *p, u64 time_seq)
2950 struct extent_buffer *b;
2955 int lowest_unlock = 1;
2956 u8 lowest_level = 0;
2959 lowest_level = p->lowest_level;
2960 WARN_ON(p->nodes[0] != NULL);
2962 if (p->search_commit_root) {
2964 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2968 b = get_old_root(root, time_seq);
2969 level = btrfs_header_level(b);
2970 p->locks[level] = BTRFS_READ_LOCK;
2973 level = btrfs_header_level(b);
2974 p->nodes[level] = b;
2975 btrfs_clear_path_blocking(p, NULL, 0);
2978 * we have a lock on b and as long as we aren't changing
2979 * the tree, there is no way to for the items in b to change.
2980 * It is safe to drop the lock on our parent before we
2981 * go through the expensive btree search on b.
2983 btrfs_unlock_up_safe(p, level + 1);
2986 * Since we can unwind eb's we want to do a real search every
2990 ret = key_search(b, key, level, &prev_cmp, &slot);
2994 if (ret && slot > 0) {
2998 p->slots[level] = slot;
2999 unlock_up(p, level, lowest_unlock, 0, NULL);
3001 if (level == lowest_level) {
3007 err = read_block_for_search(NULL, root, p, &b, level,
3008 slot, key, time_seq);
3016 level = btrfs_header_level(b);
3017 err = btrfs_tree_read_lock_atomic(b);
3019 btrfs_set_path_blocking(p);
3020 btrfs_tree_read_lock(b);
3021 btrfs_clear_path_blocking(p, b,
3024 b = tree_mod_log_rewind(root->fs_info, p, b, time_seq);
3029 p->locks[level] = BTRFS_READ_LOCK;
3030 p->nodes[level] = b;
3032 p->slots[level] = slot;
3033 unlock_up(p, level, lowest_unlock, 0, NULL);
3039 if (!p->leave_spinning)
3040 btrfs_set_path_blocking(p);
3042 btrfs_release_path(p);
3048 * helper to use instead of search slot if no exact match is needed but
3049 * instead the next or previous item should be returned.
3050 * When find_higher is true, the next higher item is returned, the next lower
3052 * When return_any and find_higher are both true, and no higher item is found,
3053 * return the next lower instead.
3054 * When return_any is true and find_higher is false, and no lower item is found,
3055 * return the next higher instead.
3056 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3059 int btrfs_search_slot_for_read(struct btrfs_root *root,
3060 struct btrfs_key *key, struct btrfs_path *p,
3061 int find_higher, int return_any)
3064 struct extent_buffer *leaf;
3067 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3071 * a return value of 1 means the path is at the position where the
3072 * item should be inserted. Normally this is the next bigger item,
3073 * but in case the previous item is the last in a leaf, path points
3074 * to the first free slot in the previous leaf, i.e. at an invalid
3080 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3081 ret = btrfs_next_leaf(root, p);
3087 * no higher item found, return the next
3092 btrfs_release_path(p);
3096 if (p->slots[0] == 0) {
3097 ret = btrfs_prev_leaf(root, p);
3102 if (p->slots[0] == btrfs_header_nritems(leaf))
3109 * no lower item found, return the next
3114 btrfs_release_path(p);
3124 * adjust the pointers going up the tree, starting at level
3125 * making sure the right key of each node is points to 'key'.
3126 * This is used after shifting pointers to the left, so it stops
3127 * fixing up pointers when a given leaf/node is not in slot 0 of the
3131 static void fixup_low_keys(struct btrfs_fs_info *fs_info,
3132 struct btrfs_path *path,
3133 struct btrfs_disk_key *key, int level)
3136 struct extent_buffer *t;
3138 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3139 int tslot = path->slots[i];
3140 if (!path->nodes[i])
3143 tree_mod_log_set_node_key(fs_info, t, tslot, 1);
3144 btrfs_set_node_key(t, key, tslot);
3145 btrfs_mark_buffer_dirty(path->nodes[i]);
3154 * This function isn't completely safe. It's the caller's responsibility
3155 * that the new key won't break the order
3157 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3158 struct btrfs_path *path,
3159 struct btrfs_key *new_key)
3161 struct btrfs_disk_key disk_key;
3162 struct extent_buffer *eb;
3165 eb = path->nodes[0];
3166 slot = path->slots[0];
3168 btrfs_item_key(eb, &disk_key, slot - 1);
3169 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3171 if (slot < btrfs_header_nritems(eb) - 1) {
3172 btrfs_item_key(eb, &disk_key, slot + 1);
3173 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3176 btrfs_cpu_key_to_disk(&disk_key, new_key);
3177 btrfs_set_item_key(eb, &disk_key, slot);
3178 btrfs_mark_buffer_dirty(eb);
3180 fixup_low_keys(fs_info, path, &disk_key, 1);
3184 * try to push data from one node into the next node left in the
3187 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3188 * error, and > 0 if there was no room in the left hand block.
3190 static int push_node_left(struct btrfs_trans_handle *trans,
3191 struct btrfs_root *root, struct extent_buffer *dst,
3192 struct extent_buffer *src, int empty)
3199 src_nritems = btrfs_header_nritems(src);
3200 dst_nritems = btrfs_header_nritems(dst);
3201 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3202 WARN_ON(btrfs_header_generation(src) != trans->transid);
3203 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3205 if (!empty && src_nritems <= 8)
3208 if (push_items <= 0)
3212 push_items = min(src_nritems, push_items);
3213 if (push_items < src_nritems) {
3214 /* leave at least 8 pointers in the node if
3215 * we aren't going to empty it
3217 if (src_nritems - push_items < 8) {
3218 if (push_items <= 8)
3224 push_items = min(src_nritems - 8, push_items);
3226 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3229 btrfs_abort_transaction(trans, root, ret);
3232 copy_extent_buffer(dst, src,
3233 btrfs_node_key_ptr_offset(dst_nritems),
3234 btrfs_node_key_ptr_offset(0),
3235 push_items * sizeof(struct btrfs_key_ptr));
3237 if (push_items < src_nritems) {
3239 * don't call tree_mod_log_eb_move here, key removal was already
3240 * fully logged by tree_mod_log_eb_copy above.
3242 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3243 btrfs_node_key_ptr_offset(push_items),
3244 (src_nritems - push_items) *
3245 sizeof(struct btrfs_key_ptr));
3247 btrfs_set_header_nritems(src, src_nritems - push_items);
3248 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3249 btrfs_mark_buffer_dirty(src);
3250 btrfs_mark_buffer_dirty(dst);
3256 * try to push data from one node into the next node right in the
3259 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3260 * error, and > 0 if there was no room in the right hand block.
3262 * this will only push up to 1/2 the contents of the left node over
3264 static int balance_node_right(struct btrfs_trans_handle *trans,
3265 struct btrfs_root *root,
3266 struct extent_buffer *dst,
3267 struct extent_buffer *src)
3275 WARN_ON(btrfs_header_generation(src) != trans->transid);
3276 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3278 src_nritems = btrfs_header_nritems(src);
3279 dst_nritems = btrfs_header_nritems(dst);
3280 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3281 if (push_items <= 0)
3284 if (src_nritems < 4)
3287 max_push = src_nritems / 2 + 1;
3288 /* don't try to empty the node */
3289 if (max_push >= src_nritems)
3292 if (max_push < push_items)
3293 push_items = max_push;
3295 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3296 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3297 btrfs_node_key_ptr_offset(0),
3299 sizeof(struct btrfs_key_ptr));
3301 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3302 src_nritems - push_items, push_items);
3304 btrfs_abort_transaction(trans, root, ret);
3307 copy_extent_buffer(dst, src,
3308 btrfs_node_key_ptr_offset(0),
3309 btrfs_node_key_ptr_offset(src_nritems - push_items),
3310 push_items * sizeof(struct btrfs_key_ptr));
3312 btrfs_set_header_nritems(src, src_nritems - push_items);
3313 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3315 btrfs_mark_buffer_dirty(src);
3316 btrfs_mark_buffer_dirty(dst);
3322 * helper function to insert a new root level in the tree.
3323 * A new node is allocated, and a single item is inserted to
3324 * point to the existing root
3326 * returns zero on success or < 0 on failure.
3328 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3329 struct btrfs_root *root,
3330 struct btrfs_path *path, int level)
3333 struct extent_buffer *lower;
3334 struct extent_buffer *c;
3335 struct extent_buffer *old;
3336 struct btrfs_disk_key lower_key;
3338 BUG_ON(path->nodes[level]);
3339 BUG_ON(path->nodes[level-1] != root->node);
3341 lower = path->nodes[level-1];
3343 btrfs_item_key(lower, &lower_key, 0);
3345 btrfs_node_key(lower, &lower_key, 0);
3347 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3348 &lower_key, level, root->node->start, 0);
3352 root_add_used(root, root->nodesize);
3354 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3355 btrfs_set_header_nritems(c, 1);
3356 btrfs_set_header_level(c, level);
3357 btrfs_set_header_bytenr(c, c->start);
3358 btrfs_set_header_generation(c, trans->transid);
3359 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3360 btrfs_set_header_owner(c, root->root_key.objectid);
3362 write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(),
3365 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3366 btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE);
3368 btrfs_set_node_key(c, &lower_key, 0);
3369 btrfs_set_node_blockptr(c, 0, lower->start);
3370 lower_gen = btrfs_header_generation(lower);
3371 WARN_ON(lower_gen != trans->transid);
3373 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3375 btrfs_mark_buffer_dirty(c);
3378 tree_mod_log_set_root_pointer(root, c, 0);
3379 rcu_assign_pointer(root->node, c);
3381 /* the super has an extra ref to root->node */
3382 free_extent_buffer(old);
3384 add_root_to_dirty_list(root);
3385 extent_buffer_get(c);
3386 path->nodes[level] = c;
3387 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3388 path->slots[level] = 0;
3393 * worker function to insert a single pointer in a node.
3394 * the node should have enough room for the pointer already
3396 * slot and level indicate where you want the key to go, and
3397 * blocknr is the block the key points to.
3399 static void insert_ptr(struct btrfs_trans_handle *trans,
3400 struct btrfs_root *root, struct btrfs_path *path,
3401 struct btrfs_disk_key *key, u64 bytenr,
3402 int slot, int level)
3404 struct extent_buffer *lower;
3408 BUG_ON(!path->nodes[level]);
3409 btrfs_assert_tree_locked(path->nodes[level]);
3410 lower = path->nodes[level];
3411 nritems = btrfs_header_nritems(lower);
3412 BUG_ON(slot > nritems);
3413 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3414 if (slot != nritems) {
3416 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3417 slot, nritems - slot);
3418 memmove_extent_buffer(lower,
3419 btrfs_node_key_ptr_offset(slot + 1),
3420 btrfs_node_key_ptr_offset(slot),
3421 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3424 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3425 MOD_LOG_KEY_ADD, GFP_NOFS);
3428 btrfs_set_node_key(lower, key, slot);
3429 btrfs_set_node_blockptr(lower, slot, bytenr);
3430 WARN_ON(trans->transid == 0);
3431 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3432 btrfs_set_header_nritems(lower, nritems + 1);
3433 btrfs_mark_buffer_dirty(lower);
3437 * split the node at the specified level in path in two.
3438 * The path is corrected to point to the appropriate node after the split
3440 * Before splitting this tries to make some room in the node by pushing
3441 * left and right, if either one works, it returns right away.
3443 * returns 0 on success and < 0 on failure
3445 static noinline int split_node(struct btrfs_trans_handle *trans,
3446 struct btrfs_root *root,
3447 struct btrfs_path *path, int level)
3449 struct extent_buffer *c;
3450 struct extent_buffer *split;
3451 struct btrfs_disk_key disk_key;
3456 c = path->nodes[level];
3457 WARN_ON(btrfs_header_generation(c) != trans->transid);
3458 if (c == root->node) {
3460 * trying to split the root, lets make a new one
3462 * tree mod log: We don't log_removal old root in
3463 * insert_new_root, because that root buffer will be kept as a
3464 * normal node. We are going to log removal of half of the
3465 * elements below with tree_mod_log_eb_copy. We're holding a
3466 * tree lock on the buffer, which is why we cannot race with
3467 * other tree_mod_log users.
3469 ret = insert_new_root(trans, root, path, level + 1);
3473 ret = push_nodes_for_insert(trans, root, path, level);
3474 c = path->nodes[level];
3475 if (!ret && btrfs_header_nritems(c) <
3476 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3482 c_nritems = btrfs_header_nritems(c);
3483 mid = (c_nritems + 1) / 2;
3484 btrfs_node_key(c, &disk_key, mid);
3486 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3487 &disk_key, level, c->start, 0);
3489 return PTR_ERR(split);
3491 root_add_used(root, root->nodesize);
3493 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3494 btrfs_set_header_level(split, btrfs_header_level(c));
3495 btrfs_set_header_bytenr(split, split->start);
3496 btrfs_set_header_generation(split, trans->transid);
3497 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3498 btrfs_set_header_owner(split, root->root_key.objectid);
3499 write_extent_buffer(split, root->fs_info->fsid,
3500 btrfs_header_fsid(), BTRFS_FSID_SIZE);
3501 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3502 btrfs_header_chunk_tree_uuid(split),
3505 ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0,
3506 mid, c_nritems - mid);
3508 btrfs_abort_transaction(trans, root, ret);
3511 copy_extent_buffer(split, c,
3512 btrfs_node_key_ptr_offset(0),
3513 btrfs_node_key_ptr_offset(mid),
3514 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3515 btrfs_set_header_nritems(split, c_nritems - mid);
3516 btrfs_set_header_nritems(c, mid);
3519 btrfs_mark_buffer_dirty(c);
3520 btrfs_mark_buffer_dirty(split);
3522 insert_ptr(trans, root, path, &disk_key, split->start,
3523 path->slots[level + 1] + 1, level + 1);
3525 if (path->slots[level] >= mid) {
3526 path->slots[level] -= mid;
3527 btrfs_tree_unlock(c);
3528 free_extent_buffer(c);
3529 path->nodes[level] = split;
3530 path->slots[level + 1] += 1;
3532 btrfs_tree_unlock(split);
3533 free_extent_buffer(split);
3539 * how many bytes are required to store the items in a leaf. start
3540 * and nr indicate which items in the leaf to check. This totals up the
3541 * space used both by the item structs and the item data
3543 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3545 struct btrfs_item *start_item;
3546 struct btrfs_item *end_item;
3547 struct btrfs_map_token token;
3549 int nritems = btrfs_header_nritems(l);
3550 int end = min(nritems, start + nr) - 1;
3554 btrfs_init_map_token(&token);
3555 start_item = btrfs_item_nr(start);
3556 end_item = btrfs_item_nr(end);
3557 data_len = btrfs_token_item_offset(l, start_item, &token) +
3558 btrfs_token_item_size(l, start_item, &token);
3559 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3560 data_len += sizeof(struct btrfs_item) * nr;
3561 WARN_ON(data_len < 0);
3566 * The space between the end of the leaf items and
3567 * the start of the leaf data. IOW, how much room
3568 * the leaf has left for both items and data
3570 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3571 struct extent_buffer *leaf)
3573 int nritems = btrfs_header_nritems(leaf);
3575 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3577 btrfs_crit(root->fs_info,
3578 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3579 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3580 leaf_space_used(leaf, 0, nritems), nritems);
3586 * min slot controls the lowest index we're willing to push to the
3587 * right. We'll push up to and including min_slot, but no lower
3589 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3590 struct btrfs_root *root,
3591 struct btrfs_path *path,
3592 int data_size, int empty,
3593 struct extent_buffer *right,
3594 int free_space, u32 left_nritems,
3597 struct extent_buffer *left = path->nodes[0];
3598 struct extent_buffer *upper = path->nodes[1];
3599 struct btrfs_map_token token;
3600 struct btrfs_disk_key disk_key;
3605 struct btrfs_item *item;
3611 btrfs_init_map_token(&token);
3616 nr = max_t(u32, 1, min_slot);
3618 if (path->slots[0] >= left_nritems)
3619 push_space += data_size;
3621 slot = path->slots[1];
3622 i = left_nritems - 1;
3624 item = btrfs_item_nr(i);
3626 if (!empty && push_items > 0) {
3627 if (path->slots[0] > i)
3629 if (path->slots[0] == i) {
3630 int space = btrfs_leaf_free_space(root, left);
3631 if (space + push_space * 2 > free_space)
3636 if (path->slots[0] == i)
3637 push_space += data_size;
3639 this_item_size = btrfs_item_size(left, item);
3640 if (this_item_size + sizeof(*item) + push_space > free_space)
3644 push_space += this_item_size + sizeof(*item);
3650 if (push_items == 0)
3653 WARN_ON(!empty && push_items == left_nritems);
3655 /* push left to right */
3656 right_nritems = btrfs_header_nritems(right);
3658 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3659 push_space -= leaf_data_end(root, left);
3661 /* make room in the right data area */
3662 data_end = leaf_data_end(root, right);
3663 memmove_extent_buffer(right,
3664 btrfs_leaf_data(right) + data_end - push_space,
3665 btrfs_leaf_data(right) + data_end,
3666 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3668 /* copy from the left data area */
3669 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3670 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3671 btrfs_leaf_data(left) + leaf_data_end(root, left),
3674 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3675 btrfs_item_nr_offset(0),
3676 right_nritems * sizeof(struct btrfs_item));
3678 /* copy the items from left to right */
3679 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3680 btrfs_item_nr_offset(left_nritems - push_items),
3681 push_items * sizeof(struct btrfs_item));
3683 /* update the item pointers */
3684 right_nritems += push_items;
3685 btrfs_set_header_nritems(right, right_nritems);
3686 push_space = BTRFS_LEAF_DATA_SIZE(root);
3687 for (i = 0; i < right_nritems; i++) {
3688 item = btrfs_item_nr(i);
3689 push_space -= btrfs_token_item_size(right, item, &token);
3690 btrfs_set_token_item_offset(right, item, push_space, &token);
3693 left_nritems -= push_items;
3694 btrfs_set_header_nritems(left, left_nritems);
3697 btrfs_mark_buffer_dirty(left);
3699 clean_tree_block(trans, root->fs_info, left);
3701 btrfs_mark_buffer_dirty(right);
3703 btrfs_item_key(right, &disk_key, 0);
3704 btrfs_set_node_key(upper, &disk_key, slot + 1);
3705 btrfs_mark_buffer_dirty(upper);
3707 /* then fixup the leaf pointer in the path */
3708 if (path->slots[0] >= left_nritems) {
3709 path->slots[0] -= left_nritems;
3710 if (btrfs_header_nritems(path->nodes[0]) == 0)
3711 clean_tree_block(trans, root->fs_info, path->nodes[0]);
3712 btrfs_tree_unlock(path->nodes[0]);
3713 free_extent_buffer(path->nodes[0]);
3714 path->nodes[0] = right;
3715 path->slots[1] += 1;
3717 btrfs_tree_unlock(right);
3718 free_extent_buffer(right);
3723 btrfs_tree_unlock(right);
3724 free_extent_buffer(right);
3729 * push some data in the path leaf to the right, trying to free up at
3730 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3732 * returns 1 if the push failed because the other node didn't have enough
3733 * room, 0 if everything worked out and < 0 if there were major errors.
3735 * this will push starting from min_slot to the end of the leaf. It won't
3736 * push any slot lower than min_slot
3738 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3739 *root, struct btrfs_path *path,
3740 int min_data_size, int data_size,
3741 int empty, u32 min_slot)
3743 struct extent_buffer *left = path->nodes[0];
3744 struct extent_buffer *right;
3745 struct extent_buffer *upper;
3751 if (!path->nodes[1])
3754 slot = path->slots[1];
3755 upper = path->nodes[1];
3756 if (slot >= btrfs_header_nritems(upper) - 1)
3759 btrfs_assert_tree_locked(path->nodes[1]);
3761 right = read_node_slot(root, upper, slot + 1);
3765 btrfs_tree_lock(right);
3766 btrfs_set_lock_blocking(right);
3768 free_space = btrfs_leaf_free_space(root, right);
3769 if (free_space < data_size)
3772 /* cow and double check */
3773 ret = btrfs_cow_block(trans, root, right, upper,
3778 free_space = btrfs_leaf_free_space(root, right);
3779 if (free_space < data_size)
3782 left_nritems = btrfs_header_nritems(left);
3783 if (left_nritems == 0)
3786 if (path->slots[0] == left_nritems && !empty) {
3787 /* Key greater than all keys in the leaf, right neighbor has
3788 * enough room for it and we're not emptying our leaf to delete
3789 * it, therefore use right neighbor to insert the new item and
3790 * no need to touch/dirty our left leaft. */
3791 btrfs_tree_unlock(left);
3792 free_extent_buffer(left);
3793 path->nodes[0] = right;
3799 return __push_leaf_right(trans, root, path, min_data_size, empty,
3800 right, free_space, left_nritems, min_slot);
3802 btrfs_tree_unlock(right);
3803 free_extent_buffer(right);
3808 * push some data in the path leaf to the left, trying to free up at
3809 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3811 * max_slot can put a limit on how far into the leaf we'll push items. The
3812 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3815 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3816 struct btrfs_root *root,
3817 struct btrfs_path *path, int data_size,
3818 int empty, struct extent_buffer *left,
3819 int free_space, u32 right_nritems,
3822 struct btrfs_disk_key disk_key;
3823 struct extent_buffer *right = path->nodes[0];
3827 struct btrfs_item *item;
3828 u32 old_left_nritems;
3832 u32 old_left_item_size;
3833 struct btrfs_map_token token;
3835 btrfs_init_map_token(&token);
3838 nr = min(right_nritems, max_slot);
3840 nr = min(right_nritems - 1, max_slot);
3842 for (i = 0; i < nr; i++) {
3843 item = btrfs_item_nr(i);
3845 if (!empty && push_items > 0) {
3846 if (path->slots[0] < i)
3848 if (path->slots[0] == i) {
3849 int space = btrfs_leaf_free_space(root, right);
3850 if (space + push_space * 2 > free_space)
3855 if (path->slots[0] == i)
3856 push_space += data_size;
3858 this_item_size = btrfs_item_size(right, item);
3859 if (this_item_size + sizeof(*item) + push_space > free_space)
3863 push_space += this_item_size + sizeof(*item);
3866 if (push_items == 0) {
3870 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3872 /* push data from right to left */
3873 copy_extent_buffer(left, right,
3874 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3875 btrfs_item_nr_offset(0),
3876 push_items * sizeof(struct btrfs_item));
3878 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3879 btrfs_item_offset_nr(right, push_items - 1);
3881 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3882 leaf_data_end(root, left) - push_space,
3883 btrfs_leaf_data(right) +
3884 btrfs_item_offset_nr(right, push_items - 1),
3886 old_left_nritems = btrfs_header_nritems(left);
3887 BUG_ON(old_left_nritems <= 0);
3889 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3890 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3893 item = btrfs_item_nr(i);
3895 ioff = btrfs_token_item_offset(left, item, &token);
3896 btrfs_set_token_item_offset(left, item,
3897 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3900 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3902 /* fixup right node */
3903 if (push_items > right_nritems)
3904 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3907 if (push_items < right_nritems) {
3908 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3909 leaf_data_end(root, right);
3910 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3911 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3912 btrfs_leaf_data(right) +
3913 leaf_data_end(root, right), push_space);
3915 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3916 btrfs_item_nr_offset(push_items),
3917 (btrfs_header_nritems(right) - push_items) *
3918 sizeof(struct btrfs_item));
3920 right_nritems -= push_items;
3921 btrfs_set_header_nritems(right, right_nritems);
3922 push_space = BTRFS_LEAF_DATA_SIZE(root);
3923 for (i = 0; i < right_nritems; i++) {
3924 item = btrfs_item_nr(i);
3926 push_space = push_space - btrfs_token_item_size(right,
3928 btrfs_set_token_item_offset(right, item, push_space, &token);
3931 btrfs_mark_buffer_dirty(left);
3933 btrfs_mark_buffer_dirty(right);
3935 clean_tree_block(trans, root->fs_info, right);
3937 btrfs_item_key(right, &disk_key, 0);
3938 fixup_low_keys(root->fs_info, path, &disk_key, 1);
3940 /* then fixup the leaf pointer in the path */
3941 if (path->slots[0] < push_items) {
3942 path->slots[0] += old_left_nritems;
3943 btrfs_tree_unlock(path->nodes[0]);
3944 free_extent_buffer(path->nodes[0]);
3945 path->nodes[0] = left;
3946 path->slots[1] -= 1;
3948 btrfs_tree_unlock(left);
3949 free_extent_buffer(left);
3950 path->slots[0] -= push_items;
3952 BUG_ON(path->slots[0] < 0);
3955 btrfs_tree_unlock(left);
3956 free_extent_buffer(left);
3961 * push some data in the path leaf to the left, trying to free up at
3962 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3964 * max_slot can put a limit on how far into the leaf we'll push items. The
3965 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3968 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3969 *root, struct btrfs_path *path, int min_data_size,
3970 int data_size, int empty, u32 max_slot)
3972 struct extent_buffer *right = path->nodes[0];
3973 struct extent_buffer *left;
3979 slot = path->slots[1];
3982 if (!path->nodes[1])
3985 right_nritems = btrfs_header_nritems(right);
3986 if (right_nritems == 0)
3989 btrfs_assert_tree_locked(path->nodes[1]);
3991 left = read_node_slot(root, path->nodes[1], slot - 1);
3995 btrfs_tree_lock(left);
3996 btrfs_set_lock_blocking(left);
3998 free_space = btrfs_leaf_free_space(root, left);
3999 if (free_space < data_size) {
4004 /* cow and double check */
4005 ret = btrfs_cow_block(trans, root, left,
4006 path->nodes[1], slot - 1, &left);
4008 /* we hit -ENOSPC, but it isn't fatal here */
4014 free_space = btrfs_leaf_free_space(root, left);
4015 if (free_space < data_size) {
4020 return __push_leaf_left(trans, root, path, min_data_size,
4021 empty, left, free_space, right_nritems,
4024 btrfs_tree_unlock(left);
4025 free_extent_buffer(left);
4030 * split the path's leaf in two, making sure there is at least data_size
4031 * available for the resulting leaf level of the path.
4033 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4034 struct btrfs_root *root,
4035 struct btrfs_path *path,
4036 struct extent_buffer *l,
4037 struct extent_buffer *right,
4038 int slot, int mid, int nritems)
4043 struct btrfs_disk_key disk_key;
4044 struct btrfs_map_token token;
4046 btrfs_init_map_token(&token);
4048 nritems = nritems - mid;
4049 btrfs_set_header_nritems(right, nritems);
4050 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
4052 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4053 btrfs_item_nr_offset(mid),
4054 nritems * sizeof(struct btrfs_item));
4056 copy_extent_buffer(right, l,
4057 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
4058 data_copy_size, btrfs_leaf_data(l) +
4059 leaf_data_end(root, l), data_copy_size);
4061 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
4062 btrfs_item_end_nr(l, mid);
4064 for (i = 0; i < nritems; i++) {
4065 struct btrfs_item *item = btrfs_item_nr(i);
4068 ioff = btrfs_token_item_offset(right, item, &token);
4069 btrfs_set_token_item_offset(right, item,
4070 ioff + rt_data_off, &token);
4073 btrfs_set_header_nritems(l, mid);
4074 btrfs_item_key(right, &disk_key, 0);
4075 insert_ptr(trans, root, path, &disk_key, right->start,
4076 path->slots[1] + 1, 1);
4078 btrfs_mark_buffer_dirty(right);
4079 btrfs_mark_buffer_dirty(l);
4080 BUG_ON(path->slots[0] != slot);
4083 btrfs_tree_unlock(path->nodes[0]);
4084 free_extent_buffer(path->nodes[0]);
4085 path->nodes[0] = right;
4086 path->slots[0] -= mid;
4087 path->slots[1] += 1;
4089 btrfs_tree_unlock(right);
4090 free_extent_buffer(right);
4093 BUG_ON(path->slots[0] < 0);
4097 * double splits happen when we need to insert a big item in the middle
4098 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4099 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4102 * We avoid this by trying to push the items on either side of our target
4103 * into the adjacent leaves. If all goes well we can avoid the double split
4106 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4107 struct btrfs_root *root,
4108 struct btrfs_path *path,
4115 int space_needed = data_size;
4117 slot = path->slots[0];
4118 if (slot < btrfs_header_nritems(path->nodes[0]))
4119 space_needed -= btrfs_leaf_free_space(root, path->nodes[0]);
4122 * try to push all the items after our slot into the
4125 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4132 nritems = btrfs_header_nritems(path->nodes[0]);
4134 * our goal is to get our slot at the start or end of a leaf. If
4135 * we've done so we're done
4137 if (path->slots[0] == 0 || path->slots[0] == nritems)
4140 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4143 /* try to push all the items before our slot into the next leaf */
4144 slot = path->slots[0];
4145 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4158 * split the path's leaf in two, making sure there is at least data_size
4159 * available for the resulting leaf level of the path.
4161 * returns 0 if all went well and < 0 on failure.
4163 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4164 struct btrfs_root *root,
4165 struct btrfs_key *ins_key,
4166 struct btrfs_path *path, int data_size,
4169 struct btrfs_disk_key disk_key;
4170 struct extent_buffer *l;
4174 struct extent_buffer *right;
4175 struct btrfs_fs_info *fs_info = root->fs_info;
4179 int num_doubles = 0;
4180 int tried_avoid_double = 0;
4183 slot = path->slots[0];
4184 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4185 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
4188 /* first try to make some room by pushing left and right */
4189 if (data_size && path->nodes[1]) {
4190 int space_needed = data_size;
4192 if (slot < btrfs_header_nritems(l))
4193 space_needed -= btrfs_leaf_free_space(root, l);
4195 wret = push_leaf_right(trans, root, path, space_needed,
4196 space_needed, 0, 0);
4200 wret = push_leaf_left(trans, root, path, space_needed,
4201 space_needed, 0, (u32)-1);
4207 /* did the pushes work? */
4208 if (btrfs_leaf_free_space(root, l) >= data_size)
4212 if (!path->nodes[1]) {
4213 ret = insert_new_root(trans, root, path, 1);
4220 slot = path->slots[0];
4221 nritems = btrfs_header_nritems(l);
4222 mid = (nritems + 1) / 2;
4226 leaf_space_used(l, mid, nritems - mid) + data_size >
4227 BTRFS_LEAF_DATA_SIZE(root)) {
4228 if (slot >= nritems) {
4232 if (mid != nritems &&
4233 leaf_space_used(l, mid, nritems - mid) +
4234 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4235 if (data_size && !tried_avoid_double)
4236 goto push_for_double;
4242 if (leaf_space_used(l, 0, mid) + data_size >
4243 BTRFS_LEAF_DATA_SIZE(root)) {
4244 if (!extend && data_size && slot == 0) {
4246 } else if ((extend || !data_size) && slot == 0) {
4250 if (mid != nritems &&
4251 leaf_space_used(l, mid, nritems - mid) +
4252 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4253 if (data_size && !tried_avoid_double)
4254 goto push_for_double;
4262 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4264 btrfs_item_key(l, &disk_key, mid);
4266 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4267 &disk_key, 0, l->start, 0);
4269 return PTR_ERR(right);
4271 root_add_used(root, root->nodesize);
4273 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4274 btrfs_set_header_bytenr(right, right->start);
4275 btrfs_set_header_generation(right, trans->transid);
4276 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4277 btrfs_set_header_owner(right, root->root_key.objectid);
4278 btrfs_set_header_level(right, 0);
4279 write_extent_buffer(right, fs_info->fsid,
4280 btrfs_header_fsid(), BTRFS_FSID_SIZE);
4282 write_extent_buffer(right, fs_info->chunk_tree_uuid,
4283 btrfs_header_chunk_tree_uuid(right),
4288 btrfs_set_header_nritems(right, 0);
4289 insert_ptr(trans, root, path, &disk_key, right->start,
4290 path->slots[1] + 1, 1);
4291 btrfs_tree_unlock(path->nodes[0]);
4292 free_extent_buffer(path->nodes[0]);
4293 path->nodes[0] = right;
4295 path->slots[1] += 1;
4297 btrfs_set_header_nritems(right, 0);
4298 insert_ptr(trans, root, path, &disk_key, right->start,
4300 btrfs_tree_unlock(path->nodes[0]);
4301 free_extent_buffer(path->nodes[0]);
4302 path->nodes[0] = right;
4304 if (path->slots[1] == 0)
4305 fixup_low_keys(fs_info, path, &disk_key, 1);
4307 btrfs_mark_buffer_dirty(right);
4311 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4314 BUG_ON(num_doubles != 0);
4322 push_for_double_split(trans, root, path, data_size);
4323 tried_avoid_double = 1;
4324 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4329 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4330 struct btrfs_root *root,
4331 struct btrfs_path *path, int ins_len)
4333 struct btrfs_key key;
4334 struct extent_buffer *leaf;
4335 struct btrfs_file_extent_item *fi;
4340 leaf = path->nodes[0];
4341 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4343 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4344 key.type != BTRFS_EXTENT_CSUM_KEY);
4346 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4349 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4350 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4351 fi = btrfs_item_ptr(leaf, path->slots[0],
4352 struct btrfs_file_extent_item);
4353 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4355 btrfs_release_path(path);
4357 path->keep_locks = 1;
4358 path->search_for_split = 1;
4359 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4360 path->search_for_split = 0;
4367 leaf = path->nodes[0];
4368 /* if our item isn't there, return now */
4369 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4372 /* the leaf has changed, it now has room. return now */
4373 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4376 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4377 fi = btrfs_item_ptr(leaf, path->slots[0],
4378 struct btrfs_file_extent_item);
4379 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4383 btrfs_set_path_blocking(path);
4384 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4388 path->keep_locks = 0;
4389 btrfs_unlock_up_safe(path, 1);
4392 path->keep_locks = 0;
4396 static noinline int split_item(struct btrfs_trans_handle *trans,
4397 struct btrfs_root *root,
4398 struct btrfs_path *path,
4399 struct btrfs_key *new_key,
4400 unsigned long split_offset)
4402 struct extent_buffer *leaf;
4403 struct btrfs_item *item;
4404 struct btrfs_item *new_item;
4410 struct btrfs_disk_key disk_key;
4412 leaf = path->nodes[0];
4413 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4415 btrfs_set_path_blocking(path);
4417 item = btrfs_item_nr(path->slots[0]);
4418 orig_offset = btrfs_item_offset(leaf, item);
4419 item_size = btrfs_item_size(leaf, item);
4421 buf = kmalloc(item_size, GFP_NOFS);
4425 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4426 path->slots[0]), item_size);
4428 slot = path->slots[0] + 1;
4429 nritems = btrfs_header_nritems(leaf);
4430 if (slot != nritems) {
4431 /* shift the items */
4432 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4433 btrfs_item_nr_offset(slot),
4434 (nritems - slot) * sizeof(struct btrfs_item));
4437 btrfs_cpu_key_to_disk(&disk_key, new_key);
4438 btrfs_set_item_key(leaf, &disk_key, slot);
4440 new_item = btrfs_item_nr(slot);
4442 btrfs_set_item_offset(leaf, new_item, orig_offset);
4443 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4445 btrfs_set_item_offset(leaf, item,
4446 orig_offset + item_size - split_offset);
4447 btrfs_set_item_size(leaf, item, split_offset);
4449 btrfs_set_header_nritems(leaf, nritems + 1);
4451 /* write the data for the start of the original item */
4452 write_extent_buffer(leaf, buf,
4453 btrfs_item_ptr_offset(leaf, path->slots[0]),
4456 /* write the data for the new item */
4457 write_extent_buffer(leaf, buf + split_offset,
4458 btrfs_item_ptr_offset(leaf, slot),
4459 item_size - split_offset);
4460 btrfs_mark_buffer_dirty(leaf);
4462 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4468 * This function splits a single item into two items,
4469 * giving 'new_key' to the new item and splitting the
4470 * old one at split_offset (from the start of the item).
4472 * The path may be released by this operation. After
4473 * the split, the path is pointing to the old item. The
4474 * new item is going to be in the same node as the old one.
4476 * Note, the item being split must be smaller enough to live alone on
4477 * a tree block with room for one extra struct btrfs_item
4479 * This allows us to split the item in place, keeping a lock on the
4480 * leaf the entire time.
4482 int btrfs_split_item(struct btrfs_trans_handle *trans,
4483 struct btrfs_root *root,
4484 struct btrfs_path *path,
4485 struct btrfs_key *new_key,
4486 unsigned long split_offset)
4489 ret = setup_leaf_for_split(trans, root, path,
4490 sizeof(struct btrfs_item));
4494 ret = split_item(trans, root, path, new_key, split_offset);
4499 * This function duplicate a item, giving 'new_key' to the new item.
4500 * It guarantees both items live in the same tree leaf and the new item
4501 * is contiguous with the original item.
4503 * This allows us to split file extent in place, keeping a lock on the
4504 * leaf the entire time.
4506 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4507 struct btrfs_root *root,
4508 struct btrfs_path *path,
4509 struct btrfs_key *new_key)
4511 struct extent_buffer *leaf;
4515 leaf = path->nodes[0];
4516 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4517 ret = setup_leaf_for_split(trans, root, path,
4518 item_size + sizeof(struct btrfs_item));
4523 setup_items_for_insert(root, path, new_key, &item_size,
4524 item_size, item_size +
4525 sizeof(struct btrfs_item), 1);
4526 leaf = path->nodes[0];
4527 memcpy_extent_buffer(leaf,
4528 btrfs_item_ptr_offset(leaf, path->slots[0]),
4529 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4535 * make the item pointed to by the path smaller. new_size indicates
4536 * how small to make it, and from_end tells us if we just chop bytes
4537 * off the end of the item or if we shift the item to chop bytes off
4540 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4541 u32 new_size, int from_end)
4544 struct extent_buffer *leaf;
4545 struct btrfs_item *item;
4547 unsigned int data_end;
4548 unsigned int old_data_start;
4549 unsigned int old_size;
4550 unsigned int size_diff;
4552 struct btrfs_map_token token;
4554 btrfs_init_map_token(&token);
4556 leaf = path->nodes[0];
4557 slot = path->slots[0];
4559 old_size = btrfs_item_size_nr(leaf, slot);
4560 if (old_size == new_size)
4563 nritems = btrfs_header_nritems(leaf);
4564 data_end = leaf_data_end(root, leaf);
4566 old_data_start = btrfs_item_offset_nr(leaf, slot);
4568 size_diff = old_size - new_size;
4571 BUG_ON(slot >= nritems);
4574 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4576 /* first correct the data pointers */
4577 for (i = slot; i < nritems; i++) {
4579 item = btrfs_item_nr(i);
4581 ioff = btrfs_token_item_offset(leaf, item, &token);
4582 btrfs_set_token_item_offset(leaf, item,
4583 ioff + size_diff, &token);
4586 /* shift the data */
4588 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4589 data_end + size_diff, btrfs_leaf_data(leaf) +
4590 data_end, old_data_start + new_size - data_end);
4592 struct btrfs_disk_key disk_key;
4595 btrfs_item_key(leaf, &disk_key, slot);
4597 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4599 struct btrfs_file_extent_item *fi;
4601 fi = btrfs_item_ptr(leaf, slot,
4602 struct btrfs_file_extent_item);
4603 fi = (struct btrfs_file_extent_item *)(
4604 (unsigned long)fi - size_diff);
4606 if (btrfs_file_extent_type(leaf, fi) ==
4607 BTRFS_FILE_EXTENT_INLINE) {
4608 ptr = btrfs_item_ptr_offset(leaf, slot);
4609 memmove_extent_buffer(leaf, ptr,
4611 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4615 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4616 data_end + size_diff, btrfs_leaf_data(leaf) +
4617 data_end, old_data_start - data_end);
4619 offset = btrfs_disk_key_offset(&disk_key);
4620 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4621 btrfs_set_item_key(leaf, &disk_key, slot);
4623 fixup_low_keys(root->fs_info, path, &disk_key, 1);
4626 item = btrfs_item_nr(slot);
4627 btrfs_set_item_size(leaf, item, new_size);
4628 btrfs_mark_buffer_dirty(leaf);
4630 if (btrfs_leaf_free_space(root, leaf) < 0) {
4631 btrfs_print_leaf(root, leaf);
4637 * make the item pointed to by the path bigger, data_size is the added size.
4639 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4643 struct extent_buffer *leaf;
4644 struct btrfs_item *item;
4646 unsigned int data_end;
4647 unsigned int old_data;
4648 unsigned int old_size;
4650 struct btrfs_map_token token;
4652 btrfs_init_map_token(&token);
4654 leaf = path->nodes[0];
4656 nritems = btrfs_header_nritems(leaf);
4657 data_end = leaf_data_end(root, leaf);
4659 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4660 btrfs_print_leaf(root, leaf);
4663 slot = path->slots[0];
4664 old_data = btrfs_item_end_nr(leaf, slot);
4667 if (slot >= nritems) {
4668 btrfs_print_leaf(root, leaf);
4669 btrfs_crit(root->fs_info, "slot %d too large, nritems %d",
4675 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4677 /* first correct the data pointers */
4678 for (i = slot; i < nritems; i++) {
4680 item = btrfs_item_nr(i);
4682 ioff = btrfs_token_item_offset(leaf, item, &token);
4683 btrfs_set_token_item_offset(leaf, item,
4684 ioff - data_size, &token);
4687 /* shift the data */
4688 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4689 data_end - data_size, btrfs_leaf_data(leaf) +
4690 data_end, old_data - data_end);
4692 data_end = old_data;
4693 old_size = btrfs_item_size_nr(leaf, slot);
4694 item = btrfs_item_nr(slot);
4695 btrfs_set_item_size(leaf, item, old_size + data_size);
4696 btrfs_mark_buffer_dirty(leaf);
4698 if (btrfs_leaf_free_space(root, leaf) < 0) {
4699 btrfs_print_leaf(root, leaf);
4705 * this is a helper for btrfs_insert_empty_items, the main goal here is
4706 * to save stack depth by doing the bulk of the work in a function
4707 * that doesn't call btrfs_search_slot
4709 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4710 struct btrfs_key *cpu_key, u32 *data_size,
4711 u32 total_data, u32 total_size, int nr)
4713 struct btrfs_item *item;
4716 unsigned int data_end;
4717 struct btrfs_disk_key disk_key;
4718 struct extent_buffer *leaf;
4720 struct btrfs_map_token token;
4722 if (path->slots[0] == 0) {
4723 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4724 fixup_low_keys(root->fs_info, path, &disk_key, 1);
4726 btrfs_unlock_up_safe(path, 1);
4728 btrfs_init_map_token(&token);
4730 leaf = path->nodes[0];
4731 slot = path->slots[0];
4733 nritems = btrfs_header_nritems(leaf);
4734 data_end = leaf_data_end(root, leaf);
4736 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4737 btrfs_print_leaf(root, leaf);
4738 btrfs_crit(root->fs_info, "not enough freespace need %u have %d",
4739 total_size, btrfs_leaf_free_space(root, leaf));
4743 if (slot != nritems) {
4744 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4746 if (old_data < data_end) {
4747 btrfs_print_leaf(root, leaf);
4748 btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d",
4749 slot, old_data, data_end);
4753 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4755 /* first correct the data pointers */
4756 for (i = slot; i < nritems; i++) {
4759 item = btrfs_item_nr( i);
4760 ioff = btrfs_token_item_offset(leaf, item, &token);
4761 btrfs_set_token_item_offset(leaf, item,
4762 ioff - total_data, &token);
4764 /* shift the items */
4765 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4766 btrfs_item_nr_offset(slot),
4767 (nritems - slot) * sizeof(struct btrfs_item));
4769 /* shift the data */
4770 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4771 data_end - total_data, btrfs_leaf_data(leaf) +
4772 data_end, old_data - data_end);
4773 data_end = old_data;
4776 /* setup the item for the new data */
4777 for (i = 0; i < nr; i++) {
4778 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4779 btrfs_set_item_key(leaf, &disk_key, slot + i);
4780 item = btrfs_item_nr(slot + i);
4781 btrfs_set_token_item_offset(leaf, item,
4782 data_end - data_size[i], &token);
4783 data_end -= data_size[i];
4784 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4787 btrfs_set_header_nritems(leaf, nritems + nr);
4788 btrfs_mark_buffer_dirty(leaf);
4790 if (btrfs_leaf_free_space(root, leaf) < 0) {
4791 btrfs_print_leaf(root, leaf);
4797 * Given a key and some data, insert items into the tree.
4798 * This does all the path init required, making room in the tree if needed.
4800 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4801 struct btrfs_root *root,
4802 struct btrfs_path *path,
4803 struct btrfs_key *cpu_key, u32 *data_size,
4812 for (i = 0; i < nr; i++)
4813 total_data += data_size[i];
4815 total_size = total_data + (nr * sizeof(struct btrfs_item));
4816 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4822 slot = path->slots[0];
4825 setup_items_for_insert(root, path, cpu_key, data_size,
4826 total_data, total_size, nr);
4831 * Given a key and some data, insert an item into the tree.
4832 * This does all the path init required, making room in the tree if needed.
4834 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4835 *root, struct btrfs_key *cpu_key, void *data, u32
4839 struct btrfs_path *path;
4840 struct extent_buffer *leaf;
4843 path = btrfs_alloc_path();
4846 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4848 leaf = path->nodes[0];
4849 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4850 write_extent_buffer(leaf, data, ptr, data_size);
4851 btrfs_mark_buffer_dirty(leaf);
4853 btrfs_free_path(path);
4858 * delete the pointer from a given node.
4860 * the tree should have been previously balanced so the deletion does not
4863 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4864 int level, int slot)
4866 struct extent_buffer *parent = path->nodes[level];
4870 nritems = btrfs_header_nritems(parent);
4871 if (slot != nritems - 1) {
4873 tree_mod_log_eb_move(root->fs_info, parent, slot,
4874 slot + 1, nritems - slot - 1);
4875 memmove_extent_buffer(parent,
4876 btrfs_node_key_ptr_offset(slot),
4877 btrfs_node_key_ptr_offset(slot + 1),
4878 sizeof(struct btrfs_key_ptr) *
4879 (nritems - slot - 1));
4881 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4882 MOD_LOG_KEY_REMOVE, GFP_NOFS);
4887 btrfs_set_header_nritems(parent, nritems);
4888 if (nritems == 0 && parent == root->node) {
4889 BUG_ON(btrfs_header_level(root->node) != 1);
4890 /* just turn the root into a leaf and break */
4891 btrfs_set_header_level(root->node, 0);
4892 } else if (slot == 0) {
4893 struct btrfs_disk_key disk_key;
4895 btrfs_node_key(parent, &disk_key, 0);
4896 fixup_low_keys(root->fs_info, path, &disk_key, level + 1);
4898 btrfs_mark_buffer_dirty(parent);
4902 * a helper function to delete the leaf pointed to by path->slots[1] and
4905 * This deletes the pointer in path->nodes[1] and frees the leaf
4906 * block extent. zero is returned if it all worked out, < 0 otherwise.
4908 * The path must have already been setup for deleting the leaf, including
4909 * all the proper balancing. path->nodes[1] must be locked.
4911 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4912 struct btrfs_root *root,
4913 struct btrfs_path *path,
4914 struct extent_buffer *leaf)
4916 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4917 del_ptr(root, path, 1, path->slots[1]);
4920 * btrfs_free_extent is expensive, we want to make sure we
4921 * aren't holding any locks when we call it
4923 btrfs_unlock_up_safe(path, 0);
4925 root_sub_used(root, leaf->len);
4927 extent_buffer_get(leaf);
4928 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4929 free_extent_buffer_stale(leaf);
4932 * delete the item at the leaf level in path. If that empties
4933 * the leaf, remove it from the tree
4935 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4936 struct btrfs_path *path, int slot, int nr)
4938 struct extent_buffer *leaf;
4939 struct btrfs_item *item;
4946 struct btrfs_map_token token;
4948 btrfs_init_map_token(&token);
4950 leaf = path->nodes[0];
4951 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4953 for (i = 0; i < nr; i++)
4954 dsize += btrfs_item_size_nr(leaf, slot + i);
4956 nritems = btrfs_header_nritems(leaf);
4958 if (slot + nr != nritems) {
4959 int data_end = leaf_data_end(root, leaf);
4961 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4963 btrfs_leaf_data(leaf) + data_end,
4964 last_off - data_end);
4966 for (i = slot + nr; i < nritems; i++) {
4969 item = btrfs_item_nr(i);
4970 ioff = btrfs_token_item_offset(leaf, item, &token);
4971 btrfs_set_token_item_offset(leaf, item,
4972 ioff + dsize, &token);
4975 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4976 btrfs_item_nr_offset(slot + nr),
4977 sizeof(struct btrfs_item) *
4978 (nritems - slot - nr));
4980 btrfs_set_header_nritems(leaf, nritems - nr);
4983 /* delete the leaf if we've emptied it */
4985 if (leaf == root->node) {
4986 btrfs_set_header_level(leaf, 0);
4988 btrfs_set_path_blocking(path);
4989 clean_tree_block(trans, root->fs_info, leaf);
4990 btrfs_del_leaf(trans, root, path, leaf);
4993 int used = leaf_space_used(leaf, 0, nritems);
4995 struct btrfs_disk_key disk_key;
4997 btrfs_item_key(leaf, &disk_key, 0);
4998 fixup_low_keys(root->fs_info, path, &disk_key, 1);
5001 /* delete the leaf if it is mostly empty */
5002 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
5003 /* push_leaf_left fixes the path.
5004 * make sure the path still points to our leaf
5005 * for possible call to del_ptr below
5007 slot = path->slots[1];
5008 extent_buffer_get(leaf);
5010 btrfs_set_path_blocking(path);
5011 wret = push_leaf_left(trans, root, path, 1, 1,
5013 if (wret < 0 && wret != -ENOSPC)
5016 if (path->nodes[0] == leaf &&
5017 btrfs_header_nritems(leaf)) {
5018 wret = push_leaf_right(trans, root, path, 1,
5020 if (wret < 0 && wret != -ENOSPC)
5024 if (btrfs_header_nritems(leaf) == 0) {
5025 path->slots[1] = slot;
5026 btrfs_del_leaf(trans, root, path, leaf);
5027 free_extent_buffer(leaf);
5030 /* if we're still in the path, make sure
5031 * we're dirty. Otherwise, one of the
5032 * push_leaf functions must have already
5033 * dirtied this buffer
5035 if (path->nodes[0] == leaf)
5036 btrfs_mark_buffer_dirty(leaf);
5037 free_extent_buffer(leaf);
5040 btrfs_mark_buffer_dirty(leaf);
5047 * search the tree again to find a leaf with lesser keys
5048 * returns 0 if it found something or 1 if there are no lesser leaves.
5049 * returns < 0 on io errors.
5051 * This may release the path, and so you may lose any locks held at the
5054 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5056 struct btrfs_key key;
5057 struct btrfs_disk_key found_key;
5060 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5062 if (key.offset > 0) {
5064 } else if (key.type > 0) {
5066 key.offset = (u64)-1;
5067 } else if (key.objectid > 0) {
5070 key.offset = (u64)-1;
5075 btrfs_release_path(path);
5076 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5079 btrfs_item_key(path->nodes[0], &found_key, 0);
5080 ret = comp_keys(&found_key, &key);
5082 * We might have had an item with the previous key in the tree right
5083 * before we released our path. And after we released our path, that
5084 * item might have been pushed to the first slot (0) of the leaf we
5085 * were holding due to a tree balance. Alternatively, an item with the
5086 * previous key can exist as the only element of a leaf (big fat item).
5087 * Therefore account for these 2 cases, so that our callers (like
5088 * btrfs_previous_item) don't miss an existing item with a key matching
5089 * the previous key we computed above.
5097 * A helper function to walk down the tree starting at min_key, and looking
5098 * for nodes or leaves that are have a minimum transaction id.
5099 * This is used by the btree defrag code, and tree logging
5101 * This does not cow, but it does stuff the starting key it finds back
5102 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5103 * key and get a writable path.
5105 * This does lock as it descends, and path->keep_locks should be set
5106 * to 1 by the caller.
5108 * This honors path->lowest_level to prevent descent past a given level
5111 * min_trans indicates the oldest transaction that you are interested
5112 * in walking through. Any nodes or leaves older than min_trans are
5113 * skipped over (without reading them).
5115 * returns zero if something useful was found, < 0 on error and 1 if there
5116 * was nothing in the tree that matched the search criteria.
5118 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5119 struct btrfs_path *path,
5122 struct extent_buffer *cur;
5123 struct btrfs_key found_key;
5129 int keep_locks = path->keep_locks;
5131 path->keep_locks = 1;
5133 cur = btrfs_read_lock_root_node(root);
5134 level = btrfs_header_level(cur);
5135 WARN_ON(path->nodes[level]);
5136 path->nodes[level] = cur;
5137 path->locks[level] = BTRFS_READ_LOCK;
5139 if (btrfs_header_generation(cur) < min_trans) {
5144 nritems = btrfs_header_nritems(cur);
5145 level = btrfs_header_level(cur);
5146 sret = bin_search(cur, min_key, level, &slot);
5148 /* at the lowest level, we're done, setup the path and exit */
5149 if (level == path->lowest_level) {
5150 if (slot >= nritems)
5153 path->slots[level] = slot;
5154 btrfs_item_key_to_cpu(cur, &found_key, slot);
5157 if (sret && slot > 0)
5160 * check this node pointer against the min_trans parameters.
5161 * If it is too old, old, skip to the next one.
5163 while (slot < nritems) {
5166 gen = btrfs_node_ptr_generation(cur, slot);
5167 if (gen < min_trans) {
5175 * we didn't find a candidate key in this node, walk forward
5176 * and find another one
5178 if (slot >= nritems) {
5179 path->slots[level] = slot;
5180 btrfs_set_path_blocking(path);
5181 sret = btrfs_find_next_key(root, path, min_key, level,
5184 btrfs_release_path(path);
5190 /* save our key for returning back */
5191 btrfs_node_key_to_cpu(cur, &found_key, slot);
5192 path->slots[level] = slot;
5193 if (level == path->lowest_level) {
5197 btrfs_set_path_blocking(path);
5198 cur = read_node_slot(root, cur, slot);
5199 BUG_ON(!cur); /* -ENOMEM */
5201 btrfs_tree_read_lock(cur);
5203 path->locks[level - 1] = BTRFS_READ_LOCK;
5204 path->nodes[level - 1] = cur;
5205 unlock_up(path, level, 1, 0, NULL);
5206 btrfs_clear_path_blocking(path, NULL, 0);
5209 path->keep_locks = keep_locks;
5211 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5212 btrfs_set_path_blocking(path);
5213 memcpy(min_key, &found_key, sizeof(found_key));
5218 static void tree_move_down(struct btrfs_root *root,
5219 struct btrfs_path *path,
5220 int *level, int root_level)
5222 BUG_ON(*level == 0);
5223 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
5224 path->slots[*level]);
5225 path->slots[*level - 1] = 0;
5229 static int tree_move_next_or_upnext(struct btrfs_root *root,
5230 struct btrfs_path *path,
5231 int *level, int root_level)
5235 nritems = btrfs_header_nritems(path->nodes[*level]);
5237 path->slots[*level]++;
5239 while (path->slots[*level] >= nritems) {
5240 if (*level == root_level)
5244 path->slots[*level] = 0;
5245 free_extent_buffer(path->nodes[*level]);
5246 path->nodes[*level] = NULL;
5248 path->slots[*level]++;
5250 nritems = btrfs_header_nritems(path->nodes[*level]);
5257 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5260 static int tree_advance(struct btrfs_root *root,
5261 struct btrfs_path *path,
5262 int *level, int root_level,
5264 struct btrfs_key *key)
5268 if (*level == 0 || !allow_down) {
5269 ret = tree_move_next_or_upnext(root, path, level, root_level);
5271 tree_move_down(root, path, level, root_level);
5276 btrfs_item_key_to_cpu(path->nodes[*level], key,
5277 path->slots[*level]);
5279 btrfs_node_key_to_cpu(path->nodes[*level], key,
5280 path->slots[*level]);
5285 static int tree_compare_item(struct btrfs_root *left_root,
5286 struct btrfs_path *left_path,
5287 struct btrfs_path *right_path,
5292 unsigned long off1, off2;
5294 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5295 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5299 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5300 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5301 right_path->slots[0]);
5303 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5305 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5312 #define ADVANCE_ONLY_NEXT -1
5315 * This function compares two trees and calls the provided callback for
5316 * every changed/new/deleted item it finds.
5317 * If shared tree blocks are encountered, whole subtrees are skipped, making
5318 * the compare pretty fast on snapshotted subvolumes.
5320 * This currently works on commit roots only. As commit roots are read only,
5321 * we don't do any locking. The commit roots are protected with transactions.
5322 * Transactions are ended and rejoined when a commit is tried in between.
5324 * This function checks for modifications done to the trees while comparing.
5325 * If it detects a change, it aborts immediately.
5327 int btrfs_compare_trees(struct btrfs_root *left_root,
5328 struct btrfs_root *right_root,
5329 btrfs_changed_cb_t changed_cb, void *ctx)
5333 struct btrfs_path *left_path = NULL;
5334 struct btrfs_path *right_path = NULL;
5335 struct btrfs_key left_key;
5336 struct btrfs_key right_key;
5337 char *tmp_buf = NULL;
5338 int left_root_level;
5339 int right_root_level;
5342 int left_end_reached;
5343 int right_end_reached;
5351 left_path = btrfs_alloc_path();
5356 right_path = btrfs_alloc_path();
5362 tmp_buf = kmalloc(left_root->nodesize, GFP_NOFS);
5368 left_path->search_commit_root = 1;
5369 left_path->skip_locking = 1;
5370 right_path->search_commit_root = 1;
5371 right_path->skip_locking = 1;
5374 * Strategy: Go to the first items of both trees. Then do
5376 * If both trees are at level 0
5377 * Compare keys of current items
5378 * If left < right treat left item as new, advance left tree
5380 * If left > right treat right item as deleted, advance right tree
5382 * If left == right do deep compare of items, treat as changed if
5383 * needed, advance both trees and repeat
5384 * If both trees are at the same level but not at level 0
5385 * Compare keys of current nodes/leafs
5386 * If left < right advance left tree and repeat
5387 * If left > right advance right tree and repeat
5388 * If left == right compare blockptrs of the next nodes/leafs
5389 * If they match advance both trees but stay at the same level
5391 * If they don't match advance both trees while allowing to go
5393 * If tree levels are different
5394 * Advance the tree that needs it and repeat
5396 * Advancing a tree means:
5397 * If we are at level 0, try to go to the next slot. If that's not
5398 * possible, go one level up and repeat. Stop when we found a level
5399 * where we could go to the next slot. We may at this point be on a
5402 * If we are not at level 0 and not on shared tree blocks, go one
5405 * If we are not at level 0 and on shared tree blocks, go one slot to
5406 * the right if possible or go up and right.
5409 down_read(&left_root->fs_info->commit_root_sem);
5410 left_level = btrfs_header_level(left_root->commit_root);
5411 left_root_level = left_level;
5412 left_path->nodes[left_level] = left_root->commit_root;
5413 extent_buffer_get(left_path->nodes[left_level]);
5415 right_level = btrfs_header_level(right_root->commit_root);
5416 right_root_level = right_level;
5417 right_path->nodes[right_level] = right_root->commit_root;
5418 extent_buffer_get(right_path->nodes[right_level]);
5419 up_read(&left_root->fs_info->commit_root_sem);
5421 if (left_level == 0)
5422 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5423 &left_key, left_path->slots[left_level]);
5425 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5426 &left_key, left_path->slots[left_level]);
5427 if (right_level == 0)
5428 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5429 &right_key, right_path->slots[right_level]);
5431 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5432 &right_key, right_path->slots[right_level]);
5434 left_end_reached = right_end_reached = 0;
5435 advance_left = advance_right = 0;
5438 if (advance_left && !left_end_reached) {
5439 ret = tree_advance(left_root, left_path, &left_level,
5441 advance_left != ADVANCE_ONLY_NEXT,
5444 left_end_reached = ADVANCE;
5447 if (advance_right && !right_end_reached) {
5448 ret = tree_advance(right_root, right_path, &right_level,
5450 advance_right != ADVANCE_ONLY_NEXT,
5453 right_end_reached = ADVANCE;
5457 if (left_end_reached && right_end_reached) {
5460 } else if (left_end_reached) {
5461 if (right_level == 0) {
5462 ret = changed_cb(left_root, right_root,
5463 left_path, right_path,
5465 BTRFS_COMPARE_TREE_DELETED,
5470 advance_right = ADVANCE;
5472 } else if (right_end_reached) {
5473 if (left_level == 0) {
5474 ret = changed_cb(left_root, right_root,
5475 left_path, right_path,
5477 BTRFS_COMPARE_TREE_NEW,
5482 advance_left = ADVANCE;
5486 if (left_level == 0 && right_level == 0) {
5487 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5489 ret = changed_cb(left_root, right_root,
5490 left_path, right_path,
5492 BTRFS_COMPARE_TREE_NEW,
5496 advance_left = ADVANCE;
5497 } else if (cmp > 0) {
5498 ret = changed_cb(left_root, right_root,
5499 left_path, right_path,
5501 BTRFS_COMPARE_TREE_DELETED,
5505 advance_right = ADVANCE;
5507 enum btrfs_compare_tree_result result;
5509 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5510 ret = tree_compare_item(left_root, left_path,
5511 right_path, tmp_buf);
5513 result = BTRFS_COMPARE_TREE_CHANGED;
5515 result = BTRFS_COMPARE_TREE_SAME;
5516 ret = changed_cb(left_root, right_root,
5517 left_path, right_path,
5518 &left_key, result, ctx);
5521 advance_left = ADVANCE;
5522 advance_right = ADVANCE;
5524 } else if (left_level == right_level) {
5525 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5527 advance_left = ADVANCE;
5528 } else if (cmp > 0) {
5529 advance_right = ADVANCE;
5531 left_blockptr = btrfs_node_blockptr(
5532 left_path->nodes[left_level],
5533 left_path->slots[left_level]);
5534 right_blockptr = btrfs_node_blockptr(
5535 right_path->nodes[right_level],
5536 right_path->slots[right_level]);
5537 left_gen = btrfs_node_ptr_generation(
5538 left_path->nodes[left_level],
5539 left_path->slots[left_level]);
5540 right_gen = btrfs_node_ptr_generation(
5541 right_path->nodes[right_level],
5542 right_path->slots[right_level]);
5543 if (left_blockptr == right_blockptr &&
5544 left_gen == right_gen) {
5546 * As we're on a shared block, don't
5547 * allow to go deeper.
5549 advance_left = ADVANCE_ONLY_NEXT;
5550 advance_right = ADVANCE_ONLY_NEXT;
5552 advance_left = ADVANCE;
5553 advance_right = ADVANCE;
5556 } else if (left_level < right_level) {
5557 advance_right = ADVANCE;
5559 advance_left = ADVANCE;
5564 btrfs_free_path(left_path);
5565 btrfs_free_path(right_path);
5571 * this is similar to btrfs_next_leaf, but does not try to preserve
5572 * and fixup the path. It looks for and returns the next key in the
5573 * tree based on the current path and the min_trans parameters.
5575 * 0 is returned if another key is found, < 0 if there are any errors
5576 * and 1 is returned if there are no higher keys in the tree
5578 * path->keep_locks should be set to 1 on the search made before
5579 * calling this function.
5581 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5582 struct btrfs_key *key, int level, u64 min_trans)
5585 struct extent_buffer *c;
5587 WARN_ON(!path->keep_locks);
5588 while (level < BTRFS_MAX_LEVEL) {
5589 if (!path->nodes[level])
5592 slot = path->slots[level] + 1;
5593 c = path->nodes[level];
5595 if (slot >= btrfs_header_nritems(c)) {
5598 struct btrfs_key cur_key;
5599 if (level + 1 >= BTRFS_MAX_LEVEL ||
5600 !path->nodes[level + 1])
5603 if (path->locks[level + 1]) {
5608 slot = btrfs_header_nritems(c) - 1;
5610 btrfs_item_key_to_cpu(c, &cur_key, slot);
5612 btrfs_node_key_to_cpu(c, &cur_key, slot);
5614 orig_lowest = path->lowest_level;
5615 btrfs_release_path(path);
5616 path->lowest_level = level;
5617 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5619 path->lowest_level = orig_lowest;
5623 c = path->nodes[level];
5624 slot = path->slots[level];
5631 btrfs_item_key_to_cpu(c, key, slot);
5633 u64 gen = btrfs_node_ptr_generation(c, slot);
5635 if (gen < min_trans) {
5639 btrfs_node_key_to_cpu(c, key, slot);
5647 * search the tree again to find a leaf with greater keys
5648 * returns 0 if it found something or 1 if there are no greater leaves.
5649 * returns < 0 on io errors.
5651 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5653 return btrfs_next_old_leaf(root, path, 0);
5656 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5661 struct extent_buffer *c;
5662 struct extent_buffer *next;
5663 struct btrfs_key key;
5666 int old_spinning = path->leave_spinning;
5667 int next_rw_lock = 0;
5669 nritems = btrfs_header_nritems(path->nodes[0]);
5673 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5678 btrfs_release_path(path);
5680 path->keep_locks = 1;
5681 path->leave_spinning = 1;
5684 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5686 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5687 path->keep_locks = 0;
5692 nritems = btrfs_header_nritems(path->nodes[0]);
5694 * by releasing the path above we dropped all our locks. A balance
5695 * could have added more items next to the key that used to be
5696 * at the very end of the block. So, check again here and
5697 * advance the path if there are now more items available.
5699 if (nritems > 0 && path->slots[0] < nritems - 1) {
5706 * So the above check misses one case:
5707 * - after releasing the path above, someone has removed the item that
5708 * used to be at the very end of the block, and balance between leafs
5709 * gets another one with bigger key.offset to replace it.
5711 * This one should be returned as well, or we can get leaf corruption
5712 * later(esp. in __btrfs_drop_extents()).
5714 * And a bit more explanation about this check,
5715 * with ret > 0, the key isn't found, the path points to the slot
5716 * where it should be inserted, so the path->slots[0] item must be the
5719 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5724 while (level < BTRFS_MAX_LEVEL) {
5725 if (!path->nodes[level]) {
5730 slot = path->slots[level] + 1;
5731 c = path->nodes[level];
5732 if (slot >= btrfs_header_nritems(c)) {
5734 if (level == BTRFS_MAX_LEVEL) {
5742 btrfs_tree_unlock_rw(next, next_rw_lock);
5743 free_extent_buffer(next);
5747 next_rw_lock = path->locks[level];
5748 ret = read_block_for_search(NULL, root, path, &next, level,
5754 btrfs_release_path(path);
5758 if (!path->skip_locking) {
5759 ret = btrfs_try_tree_read_lock(next);
5760 if (!ret && time_seq) {
5762 * If we don't get the lock, we may be racing
5763 * with push_leaf_left, holding that lock while
5764 * itself waiting for the leaf we've currently
5765 * locked. To solve this situation, we give up
5766 * on our lock and cycle.
5768 free_extent_buffer(next);
5769 btrfs_release_path(path);
5774 btrfs_set_path_blocking(path);
5775 btrfs_tree_read_lock(next);
5776 btrfs_clear_path_blocking(path, next,
5779 next_rw_lock = BTRFS_READ_LOCK;
5783 path->slots[level] = slot;
5786 c = path->nodes[level];
5787 if (path->locks[level])
5788 btrfs_tree_unlock_rw(c, path->locks[level]);
5790 free_extent_buffer(c);
5791 path->nodes[level] = next;
5792 path->slots[level] = 0;
5793 if (!path->skip_locking)
5794 path->locks[level] = next_rw_lock;
5798 ret = read_block_for_search(NULL, root, path, &next, level,
5804 btrfs_release_path(path);
5808 if (!path->skip_locking) {
5809 ret = btrfs_try_tree_read_lock(next);
5811 btrfs_set_path_blocking(path);
5812 btrfs_tree_read_lock(next);
5813 btrfs_clear_path_blocking(path, next,
5816 next_rw_lock = BTRFS_READ_LOCK;
5821 unlock_up(path, 0, 1, 0, NULL);
5822 path->leave_spinning = old_spinning;
5824 btrfs_set_path_blocking(path);
5830 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5831 * searching until it gets past min_objectid or finds an item of 'type'
5833 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5835 int btrfs_previous_item(struct btrfs_root *root,
5836 struct btrfs_path *path, u64 min_objectid,
5839 struct btrfs_key found_key;
5840 struct extent_buffer *leaf;
5845 if (path->slots[0] == 0) {
5846 btrfs_set_path_blocking(path);
5847 ret = btrfs_prev_leaf(root, path);
5853 leaf = path->nodes[0];
5854 nritems = btrfs_header_nritems(leaf);
5857 if (path->slots[0] == nritems)
5860 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5861 if (found_key.objectid < min_objectid)
5863 if (found_key.type == type)
5865 if (found_key.objectid == min_objectid &&
5866 found_key.type < type)
5873 * search in extent tree to find a previous Metadata/Data extent item with
5876 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5878 int btrfs_previous_extent_item(struct btrfs_root *root,
5879 struct btrfs_path *path, u64 min_objectid)
5881 struct btrfs_key found_key;
5882 struct extent_buffer *leaf;
5887 if (path->slots[0] == 0) {
5888 btrfs_set_path_blocking(path);
5889 ret = btrfs_prev_leaf(root, path);
5895 leaf = path->nodes[0];
5896 nritems = btrfs_header_nritems(leaf);
5899 if (path->slots[0] == nritems)
5902 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5903 if (found_key.objectid < min_objectid)
5905 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5906 found_key.type == BTRFS_METADATA_ITEM_KEY)
5908 if (found_key.objectid == min_objectid &&
5909 found_key.type < BTRFS_EXTENT_ITEM_KEY)