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 struct extent_buffer *old;
1422 struct tree_mod_root *old_root = NULL;
1423 u64 old_generation = 0;
1426 eb_root = btrfs_read_lock_root_node(root);
1427 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1431 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1432 old_root = &tm->old_root;
1433 old_generation = tm->generation;
1434 logical = old_root->logical;
1436 logical = eb_root->start;
1439 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1440 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1441 btrfs_tree_read_unlock(eb_root);
1442 free_extent_buffer(eb_root);
1443 old = read_tree_block(root, logical, 0);
1444 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1446 free_extent_buffer(old);
1447 btrfs_warn(root->fs_info,
1448 "failed to read tree block %llu from get_old_root", logical);
1450 eb = btrfs_clone_extent_buffer(old);
1451 free_extent_buffer(old);
1453 } else if (old_root) {
1454 btrfs_tree_read_unlock(eb_root);
1455 free_extent_buffer(eb_root);
1456 eb = alloc_dummy_extent_buffer(root->fs_info, logical);
1458 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1459 eb = btrfs_clone_extent_buffer(eb_root);
1460 btrfs_tree_read_unlock_blocking(eb_root);
1461 free_extent_buffer(eb_root);
1466 extent_buffer_get(eb);
1467 btrfs_tree_read_lock(eb);
1469 btrfs_set_header_bytenr(eb, eb->start);
1470 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1471 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1472 btrfs_set_header_level(eb, old_root->level);
1473 btrfs_set_header_generation(eb, old_generation);
1476 __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1478 WARN_ON(btrfs_header_level(eb) != 0);
1479 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1484 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1486 struct tree_mod_elem *tm;
1488 struct extent_buffer *eb_root = btrfs_root_node(root);
1490 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1491 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1492 level = tm->old_root.level;
1494 level = btrfs_header_level(eb_root);
1496 free_extent_buffer(eb_root);
1501 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1502 struct btrfs_root *root,
1503 struct extent_buffer *buf)
1505 if (btrfs_test_is_dummy_root(root))
1508 /* ensure we can see the force_cow */
1512 * We do not need to cow a block if
1513 * 1) this block is not created or changed in this transaction;
1514 * 2) this block does not belong to TREE_RELOC tree;
1515 * 3) the root is not forced COW.
1517 * What is forced COW:
1518 * when we create snapshot during commiting the transaction,
1519 * after we've finished coping src root, we must COW the shared
1520 * block to ensure the metadata consistency.
1522 if (btrfs_header_generation(buf) == trans->transid &&
1523 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1524 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1525 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1526 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1532 * cows a single block, see __btrfs_cow_block for the real work.
1533 * This version of it has extra checks so that a block isn't cow'd more than
1534 * once per transaction, as long as it hasn't been written yet
1536 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1537 struct btrfs_root *root, struct extent_buffer *buf,
1538 struct extent_buffer *parent, int parent_slot,
1539 struct extent_buffer **cow_ret)
1544 if (trans->transaction != root->fs_info->running_transaction)
1545 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1547 root->fs_info->running_transaction->transid);
1549 if (trans->transid != root->fs_info->generation)
1550 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1551 trans->transid, root->fs_info->generation);
1553 if (!should_cow_block(trans, root, buf)) {
1554 trans->dirty = true;
1559 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
1562 btrfs_set_lock_blocking(parent);
1563 btrfs_set_lock_blocking(buf);
1565 ret = __btrfs_cow_block(trans, root, buf, parent,
1566 parent_slot, cow_ret, search_start, 0);
1568 trace_btrfs_cow_block(root, buf, *cow_ret);
1574 * helper function for defrag to decide if two blocks pointed to by a
1575 * node are actually close by
1577 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1579 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1581 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1587 * compare two keys in a memcmp fashion
1589 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1591 struct btrfs_key k1;
1593 btrfs_disk_key_to_cpu(&k1, disk);
1595 return btrfs_comp_cpu_keys(&k1, k2);
1599 * same as comp_keys only with two btrfs_key's
1601 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1603 if (k1->objectid > k2->objectid)
1605 if (k1->objectid < k2->objectid)
1607 if (k1->type > k2->type)
1609 if (k1->type < k2->type)
1611 if (k1->offset > k2->offset)
1613 if (k1->offset < k2->offset)
1619 * this is used by the defrag code to go through all the
1620 * leaves pointed to by a node and reallocate them so that
1621 * disk order is close to key order
1623 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1624 struct btrfs_root *root, struct extent_buffer *parent,
1625 int start_slot, u64 *last_ret,
1626 struct btrfs_key *progress)
1628 struct extent_buffer *cur;
1631 u64 search_start = *last_ret;
1641 int progress_passed = 0;
1642 struct btrfs_disk_key disk_key;
1644 parent_level = btrfs_header_level(parent);
1646 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1647 WARN_ON(trans->transid != root->fs_info->generation);
1649 parent_nritems = btrfs_header_nritems(parent);
1650 blocksize = root->nodesize;
1651 end_slot = parent_nritems - 1;
1653 if (parent_nritems <= 1)
1656 btrfs_set_lock_blocking(parent);
1658 for (i = start_slot; i <= end_slot; i++) {
1661 btrfs_node_key(parent, &disk_key, i);
1662 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1665 progress_passed = 1;
1666 blocknr = btrfs_node_blockptr(parent, i);
1667 gen = btrfs_node_ptr_generation(parent, i);
1668 if (last_block == 0)
1669 last_block = blocknr;
1672 other = btrfs_node_blockptr(parent, i - 1);
1673 close = close_blocks(blocknr, other, blocksize);
1675 if (!close && i < end_slot) {
1676 other = btrfs_node_blockptr(parent, i + 1);
1677 close = close_blocks(blocknr, other, blocksize);
1680 last_block = blocknr;
1684 cur = btrfs_find_tree_block(root->fs_info, blocknr);
1686 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1689 if (!cur || !uptodate) {
1691 cur = read_tree_block(root, blocknr, gen);
1693 return PTR_ERR(cur);
1694 } else if (!extent_buffer_uptodate(cur)) {
1695 free_extent_buffer(cur);
1698 } else if (!uptodate) {
1699 err = btrfs_read_buffer(cur, gen);
1701 free_extent_buffer(cur);
1706 if (search_start == 0)
1707 search_start = last_block;
1709 btrfs_tree_lock(cur);
1710 btrfs_set_lock_blocking(cur);
1711 err = __btrfs_cow_block(trans, root, cur, parent, i,
1714 (end_slot - i) * blocksize));
1716 btrfs_tree_unlock(cur);
1717 free_extent_buffer(cur);
1720 search_start = cur->start;
1721 last_block = cur->start;
1722 *last_ret = search_start;
1723 btrfs_tree_unlock(cur);
1724 free_extent_buffer(cur);
1730 * The leaf data grows from end-to-front in the node.
1731 * this returns the address of the start of the last item,
1732 * which is the stop of the leaf data stack
1734 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1735 struct extent_buffer *leaf)
1737 u32 nr = btrfs_header_nritems(leaf);
1739 return BTRFS_LEAF_DATA_SIZE(root);
1740 return btrfs_item_offset_nr(leaf, nr - 1);
1745 * search for key in the extent_buffer. The items start at offset p,
1746 * and they are item_size apart. There are 'max' items in p.
1748 * the slot in the array is returned via slot, and it points to
1749 * the place where you would insert key if it is not found in
1752 * slot may point to max if the key is bigger than all of the keys
1754 static noinline int generic_bin_search(struct extent_buffer *eb,
1756 int item_size, struct btrfs_key *key,
1763 struct btrfs_disk_key *tmp = NULL;
1764 struct btrfs_disk_key unaligned;
1765 unsigned long offset;
1767 unsigned long map_start = 0;
1768 unsigned long map_len = 0;
1771 while (low < high) {
1772 mid = (low + high) / 2;
1773 offset = p + mid * item_size;
1775 if (!kaddr || offset < map_start ||
1776 (offset + sizeof(struct btrfs_disk_key)) >
1777 map_start + map_len) {
1779 err = map_private_extent_buffer(eb, offset,
1780 sizeof(struct btrfs_disk_key),
1781 &kaddr, &map_start, &map_len);
1784 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1787 read_extent_buffer(eb, &unaligned,
1788 offset, sizeof(unaligned));
1793 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1796 ret = comp_keys(tmp, key);
1812 * simple bin_search frontend that does the right thing for
1815 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1816 int level, int *slot)
1819 return generic_bin_search(eb,
1820 offsetof(struct btrfs_leaf, items),
1821 sizeof(struct btrfs_item),
1822 key, btrfs_header_nritems(eb),
1825 return generic_bin_search(eb,
1826 offsetof(struct btrfs_node, ptrs),
1827 sizeof(struct btrfs_key_ptr),
1828 key, btrfs_header_nritems(eb),
1832 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1833 int level, int *slot)
1835 return bin_search(eb, key, level, slot);
1838 static void root_add_used(struct btrfs_root *root, u32 size)
1840 spin_lock(&root->accounting_lock);
1841 btrfs_set_root_used(&root->root_item,
1842 btrfs_root_used(&root->root_item) + size);
1843 spin_unlock(&root->accounting_lock);
1846 static void root_sub_used(struct btrfs_root *root, u32 size)
1848 spin_lock(&root->accounting_lock);
1849 btrfs_set_root_used(&root->root_item,
1850 btrfs_root_used(&root->root_item) - size);
1851 spin_unlock(&root->accounting_lock);
1854 /* given a node and slot number, this reads the blocks it points to. The
1855 * extent buffer is returned with a reference taken (but unlocked).
1856 * NULL is returned on error.
1858 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1859 struct extent_buffer *parent, int slot)
1861 int level = btrfs_header_level(parent);
1862 struct extent_buffer *eb;
1866 if (slot >= btrfs_header_nritems(parent))
1871 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1872 btrfs_node_ptr_generation(parent, slot));
1873 if (IS_ERR(eb) || !extent_buffer_uptodate(eb)) {
1875 free_extent_buffer(eb);
1883 * node level balancing, used to make sure nodes are in proper order for
1884 * item deletion. We balance from the top down, so we have to make sure
1885 * that a deletion won't leave an node completely empty later on.
1887 static noinline int balance_level(struct btrfs_trans_handle *trans,
1888 struct btrfs_root *root,
1889 struct btrfs_path *path, int level)
1891 struct extent_buffer *right = NULL;
1892 struct extent_buffer *mid;
1893 struct extent_buffer *left = NULL;
1894 struct extent_buffer *parent = NULL;
1898 int orig_slot = path->slots[level];
1904 mid = path->nodes[level];
1906 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1907 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1908 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1910 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1912 if (level < BTRFS_MAX_LEVEL - 1) {
1913 parent = path->nodes[level + 1];
1914 pslot = path->slots[level + 1];
1918 * deal with the case where there is only one pointer in the root
1919 * by promoting the node below to a root
1922 struct extent_buffer *child;
1924 if (btrfs_header_nritems(mid) != 1)
1927 /* promote the child to a root */
1928 child = read_node_slot(root, mid, 0);
1931 btrfs_std_error(root->fs_info, ret, NULL);
1935 btrfs_tree_lock(child);
1936 btrfs_set_lock_blocking(child);
1937 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1939 btrfs_tree_unlock(child);
1940 free_extent_buffer(child);
1944 tree_mod_log_set_root_pointer(root, child, 1);
1945 rcu_assign_pointer(root->node, child);
1947 add_root_to_dirty_list(root);
1948 btrfs_tree_unlock(child);
1950 path->locks[level] = 0;
1951 path->nodes[level] = NULL;
1952 clean_tree_block(trans, root->fs_info, mid);
1953 btrfs_tree_unlock(mid);
1954 /* once for the path */
1955 free_extent_buffer(mid);
1957 root_sub_used(root, mid->len);
1958 btrfs_free_tree_block(trans, root, mid, 0, 1);
1959 /* once for the root ptr */
1960 free_extent_buffer_stale(mid);
1963 if (btrfs_header_nritems(mid) >
1964 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1967 left = read_node_slot(root, parent, pslot - 1);
1969 btrfs_tree_lock(left);
1970 btrfs_set_lock_blocking(left);
1971 wret = btrfs_cow_block(trans, root, left,
1972 parent, pslot - 1, &left);
1978 right = read_node_slot(root, parent, pslot + 1);
1980 btrfs_tree_lock(right);
1981 btrfs_set_lock_blocking(right);
1982 wret = btrfs_cow_block(trans, root, right,
1983 parent, pslot + 1, &right);
1990 /* first, try to make some room in the middle buffer */
1992 orig_slot += btrfs_header_nritems(left);
1993 wret = push_node_left(trans, root, left, mid, 1);
1999 * then try to empty the right most buffer into the middle
2002 wret = push_node_left(trans, root, mid, right, 1);
2003 if (wret < 0 && wret != -ENOSPC)
2005 if (btrfs_header_nritems(right) == 0) {
2006 clean_tree_block(trans, root->fs_info, right);
2007 btrfs_tree_unlock(right);
2008 del_ptr(root, path, level + 1, pslot + 1);
2009 root_sub_used(root, right->len);
2010 btrfs_free_tree_block(trans, root, right, 0, 1);
2011 free_extent_buffer_stale(right);
2014 struct btrfs_disk_key right_key;
2015 btrfs_node_key(right, &right_key, 0);
2016 tree_mod_log_set_node_key(root->fs_info, parent,
2018 btrfs_set_node_key(parent, &right_key, pslot + 1);
2019 btrfs_mark_buffer_dirty(parent);
2022 if (btrfs_header_nritems(mid) == 1) {
2024 * we're not allowed to leave a node with one item in the
2025 * tree during a delete. A deletion from lower in the tree
2026 * could try to delete the only pointer in this node.
2027 * So, pull some keys from the left.
2028 * There has to be a left pointer at this point because
2029 * otherwise we would have pulled some pointers from the
2034 btrfs_std_error(root->fs_info, ret, NULL);
2037 wret = balance_node_right(trans, root, mid, left);
2043 wret = push_node_left(trans, root, left, mid, 1);
2049 if (btrfs_header_nritems(mid) == 0) {
2050 clean_tree_block(trans, root->fs_info, mid);
2051 btrfs_tree_unlock(mid);
2052 del_ptr(root, path, level + 1, pslot);
2053 root_sub_used(root, mid->len);
2054 btrfs_free_tree_block(trans, root, mid, 0, 1);
2055 free_extent_buffer_stale(mid);
2058 /* update the parent key to reflect our changes */
2059 struct btrfs_disk_key mid_key;
2060 btrfs_node_key(mid, &mid_key, 0);
2061 tree_mod_log_set_node_key(root->fs_info, parent,
2063 btrfs_set_node_key(parent, &mid_key, pslot);
2064 btrfs_mark_buffer_dirty(parent);
2067 /* update the path */
2069 if (btrfs_header_nritems(left) > orig_slot) {
2070 extent_buffer_get(left);
2071 /* left was locked after cow */
2072 path->nodes[level] = left;
2073 path->slots[level + 1] -= 1;
2074 path->slots[level] = orig_slot;
2076 btrfs_tree_unlock(mid);
2077 free_extent_buffer(mid);
2080 orig_slot -= btrfs_header_nritems(left);
2081 path->slots[level] = orig_slot;
2084 /* double check we haven't messed things up */
2086 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2090 btrfs_tree_unlock(right);
2091 free_extent_buffer(right);
2094 if (path->nodes[level] != left)
2095 btrfs_tree_unlock(left);
2096 free_extent_buffer(left);
2101 /* Node balancing for insertion. Here we only split or push nodes around
2102 * when they are completely full. This is also done top down, so we
2103 * have to be pessimistic.
2105 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2106 struct btrfs_root *root,
2107 struct btrfs_path *path, int level)
2109 struct extent_buffer *right = NULL;
2110 struct extent_buffer *mid;
2111 struct extent_buffer *left = NULL;
2112 struct extent_buffer *parent = NULL;
2116 int orig_slot = path->slots[level];
2121 mid = path->nodes[level];
2122 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2124 if (level < BTRFS_MAX_LEVEL - 1) {
2125 parent = path->nodes[level + 1];
2126 pslot = path->slots[level + 1];
2132 left = read_node_slot(root, parent, pslot - 1);
2134 /* first, try to make some room in the middle buffer */
2138 btrfs_tree_lock(left);
2139 btrfs_set_lock_blocking(left);
2141 left_nr = btrfs_header_nritems(left);
2142 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2145 ret = btrfs_cow_block(trans, root, left, parent,
2150 wret = push_node_left(trans, root,
2157 struct btrfs_disk_key disk_key;
2158 orig_slot += left_nr;
2159 btrfs_node_key(mid, &disk_key, 0);
2160 tree_mod_log_set_node_key(root->fs_info, parent,
2162 btrfs_set_node_key(parent, &disk_key, pslot);
2163 btrfs_mark_buffer_dirty(parent);
2164 if (btrfs_header_nritems(left) > orig_slot) {
2165 path->nodes[level] = left;
2166 path->slots[level + 1] -= 1;
2167 path->slots[level] = orig_slot;
2168 btrfs_tree_unlock(mid);
2169 free_extent_buffer(mid);
2172 btrfs_header_nritems(left);
2173 path->slots[level] = orig_slot;
2174 btrfs_tree_unlock(left);
2175 free_extent_buffer(left);
2179 btrfs_tree_unlock(left);
2180 free_extent_buffer(left);
2182 right = read_node_slot(root, parent, pslot + 1);
2185 * then try to empty the right most buffer into the middle
2190 btrfs_tree_lock(right);
2191 btrfs_set_lock_blocking(right);
2193 right_nr = btrfs_header_nritems(right);
2194 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2197 ret = btrfs_cow_block(trans, root, right,
2203 wret = balance_node_right(trans, root,
2210 struct btrfs_disk_key disk_key;
2212 btrfs_node_key(right, &disk_key, 0);
2213 tree_mod_log_set_node_key(root->fs_info, parent,
2215 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2216 btrfs_mark_buffer_dirty(parent);
2218 if (btrfs_header_nritems(mid) <= orig_slot) {
2219 path->nodes[level] = right;
2220 path->slots[level + 1] += 1;
2221 path->slots[level] = orig_slot -
2222 btrfs_header_nritems(mid);
2223 btrfs_tree_unlock(mid);
2224 free_extent_buffer(mid);
2226 btrfs_tree_unlock(right);
2227 free_extent_buffer(right);
2231 btrfs_tree_unlock(right);
2232 free_extent_buffer(right);
2238 * readahead one full node of leaves, finding things that are close
2239 * to the block in 'slot', and triggering ra on them.
2241 static void reada_for_search(struct btrfs_root *root,
2242 struct btrfs_path *path,
2243 int level, int slot, u64 objectid)
2245 struct extent_buffer *node;
2246 struct btrfs_disk_key disk_key;
2252 int direction = path->reada;
2253 struct extent_buffer *eb;
2261 if (!path->nodes[level])
2264 node = path->nodes[level];
2266 search = btrfs_node_blockptr(node, slot);
2267 blocksize = root->nodesize;
2268 eb = btrfs_find_tree_block(root->fs_info, search);
2270 free_extent_buffer(eb);
2276 nritems = btrfs_header_nritems(node);
2280 if (direction < 0) {
2284 } else if (direction > 0) {
2289 if (path->reada < 0 && objectid) {
2290 btrfs_node_key(node, &disk_key, nr);
2291 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2294 search = btrfs_node_blockptr(node, nr);
2295 if ((search <= target && target - search <= 65536) ||
2296 (search > target && search - target <= 65536)) {
2297 gen = btrfs_node_ptr_generation(node, nr);
2298 readahead_tree_block(root, search);
2302 if ((nread > 65536 || nscan > 32))
2307 static noinline void reada_for_balance(struct btrfs_root *root,
2308 struct btrfs_path *path, int level)
2312 struct extent_buffer *parent;
2313 struct extent_buffer *eb;
2318 parent = path->nodes[level + 1];
2322 nritems = btrfs_header_nritems(parent);
2323 slot = path->slots[level + 1];
2326 block1 = btrfs_node_blockptr(parent, slot - 1);
2327 gen = btrfs_node_ptr_generation(parent, slot - 1);
2328 eb = btrfs_find_tree_block(root->fs_info, block1);
2330 * if we get -eagain from btrfs_buffer_uptodate, we
2331 * don't want to return eagain here. That will loop
2334 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2336 free_extent_buffer(eb);
2338 if (slot + 1 < nritems) {
2339 block2 = btrfs_node_blockptr(parent, slot + 1);
2340 gen = btrfs_node_ptr_generation(parent, slot + 1);
2341 eb = btrfs_find_tree_block(root->fs_info, block2);
2342 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2344 free_extent_buffer(eb);
2348 readahead_tree_block(root, block1);
2350 readahead_tree_block(root, block2);
2355 * when we walk down the tree, it is usually safe to unlock the higher layers
2356 * in the tree. The exceptions are when our path goes through slot 0, because
2357 * operations on the tree might require changing key pointers higher up in the
2360 * callers might also have set path->keep_locks, which tells this code to keep
2361 * the lock if the path points to the last slot in the block. This is part of
2362 * walking through the tree, and selecting the next slot in the higher block.
2364 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2365 * if lowest_unlock is 1, level 0 won't be unlocked
2367 static noinline void unlock_up(struct btrfs_path *path, int level,
2368 int lowest_unlock, int min_write_lock_level,
2369 int *write_lock_level)
2372 int skip_level = level;
2374 struct extent_buffer *t;
2376 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2377 if (!path->nodes[i])
2379 if (!path->locks[i])
2381 if (!no_skips && path->slots[i] == 0) {
2385 if (!no_skips && path->keep_locks) {
2388 nritems = btrfs_header_nritems(t);
2389 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2394 if (skip_level < i && i >= lowest_unlock)
2398 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2399 btrfs_tree_unlock_rw(t, path->locks[i]);
2401 if (write_lock_level &&
2402 i > min_write_lock_level &&
2403 i <= *write_lock_level) {
2404 *write_lock_level = i - 1;
2411 * This releases any locks held in the path starting at level and
2412 * going all the way up to the root.
2414 * btrfs_search_slot will keep the lock held on higher nodes in a few
2415 * corner cases, such as COW of the block at slot zero in the node. This
2416 * ignores those rules, and it should only be called when there are no
2417 * more updates to be done higher up in the tree.
2419 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2423 if (path->keep_locks)
2426 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2427 if (!path->nodes[i])
2429 if (!path->locks[i])
2431 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2437 * helper function for btrfs_search_slot. The goal is to find a block
2438 * in cache without setting the path to blocking. If we find the block
2439 * we return zero and the path is unchanged.
2441 * If we can't find the block, we set the path blocking and do some
2442 * reada. -EAGAIN is returned and the search must be repeated.
2445 read_block_for_search(struct btrfs_trans_handle *trans,
2446 struct btrfs_root *root, struct btrfs_path *p,
2447 struct extent_buffer **eb_ret, int level, int slot,
2448 struct btrfs_key *key, u64 time_seq)
2452 struct extent_buffer *b = *eb_ret;
2453 struct extent_buffer *tmp;
2456 blocknr = btrfs_node_blockptr(b, slot);
2457 gen = btrfs_node_ptr_generation(b, slot);
2459 tmp = btrfs_find_tree_block(root->fs_info, blocknr);
2461 /* first we do an atomic uptodate check */
2462 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2467 /* the pages were up to date, but we failed
2468 * the generation number check. Do a full
2469 * read for the generation number that is correct.
2470 * We must do this without dropping locks so
2471 * we can trust our generation number
2473 btrfs_set_path_blocking(p);
2475 /* now we're allowed to do a blocking uptodate check */
2476 ret = btrfs_read_buffer(tmp, gen);
2481 free_extent_buffer(tmp);
2482 btrfs_release_path(p);
2487 * reduce lock contention at high levels
2488 * of the btree by dropping locks before
2489 * we read. Don't release the lock on the current
2490 * level because we need to walk this node to figure
2491 * out which blocks to read.
2493 btrfs_unlock_up_safe(p, level + 1);
2494 btrfs_set_path_blocking(p);
2496 free_extent_buffer(tmp);
2498 reada_for_search(root, p, level, slot, key->objectid);
2501 tmp = read_tree_block(root, blocknr, gen);
2504 * If the read above didn't mark this buffer up to date,
2505 * it will never end up being up to date. Set ret to EIO now
2506 * and give up so that our caller doesn't loop forever
2509 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2511 free_extent_buffer(tmp);
2514 btrfs_release_path(p);
2519 * helper function for btrfs_search_slot. This does all of the checks
2520 * for node-level blocks and does any balancing required based on
2523 * If no extra work was required, zero is returned. If we had to
2524 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2528 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2529 struct btrfs_root *root, struct btrfs_path *p,
2530 struct extent_buffer *b, int level, int ins_len,
2531 int *write_lock_level)
2534 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2535 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2538 if (*write_lock_level < level + 1) {
2539 *write_lock_level = level + 1;
2540 btrfs_release_path(p);
2544 btrfs_set_path_blocking(p);
2545 reada_for_balance(root, p, level);
2546 sret = split_node(trans, root, p, level);
2547 btrfs_clear_path_blocking(p, NULL, 0);
2554 b = p->nodes[level];
2555 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2556 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2559 if (*write_lock_level < level + 1) {
2560 *write_lock_level = level + 1;
2561 btrfs_release_path(p);
2565 btrfs_set_path_blocking(p);
2566 reada_for_balance(root, p, level);
2567 sret = balance_level(trans, root, p, level);
2568 btrfs_clear_path_blocking(p, NULL, 0);
2574 b = p->nodes[level];
2576 btrfs_release_path(p);
2579 BUG_ON(btrfs_header_nritems(b) == 1);
2589 static void key_search_validate(struct extent_buffer *b,
2590 struct btrfs_key *key,
2593 #ifdef CONFIG_BTRFS_ASSERT
2594 struct btrfs_disk_key disk_key;
2596 btrfs_cpu_key_to_disk(&disk_key, key);
2599 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2600 offsetof(struct btrfs_leaf, items[0].key),
2603 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2604 offsetof(struct btrfs_node, ptrs[0].key),
2609 static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2610 int level, int *prev_cmp, int *slot)
2612 if (*prev_cmp != 0) {
2613 *prev_cmp = bin_search(b, key, level, slot);
2617 key_search_validate(b, key, level);
2623 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2624 u64 iobjectid, u64 ioff, u8 key_type,
2625 struct btrfs_key *found_key)
2628 struct btrfs_key key;
2629 struct extent_buffer *eb;
2634 key.type = key_type;
2635 key.objectid = iobjectid;
2638 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2642 eb = path->nodes[0];
2643 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2644 ret = btrfs_next_leaf(fs_root, path);
2647 eb = path->nodes[0];
2650 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2651 if (found_key->type != key.type ||
2652 found_key->objectid != key.objectid)
2659 * look for key in the tree. path is filled in with nodes along the way
2660 * if key is found, we return zero and you can find the item in the leaf
2661 * level of the path (level 0)
2663 * If the key isn't found, the path points to the slot where it should
2664 * be inserted, and 1 is returned. If there are other errors during the
2665 * search a negative error number is returned.
2667 * if ins_len > 0, nodes and leaves will be split as we walk down the
2668 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2671 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2672 *root, struct btrfs_key *key, struct btrfs_path *p, int
2675 struct extent_buffer *b;
2680 int lowest_unlock = 1;
2682 /* everything at write_lock_level or lower must be write locked */
2683 int write_lock_level = 0;
2684 u8 lowest_level = 0;
2685 int min_write_lock_level;
2688 lowest_level = p->lowest_level;
2689 WARN_ON(lowest_level && ins_len > 0);
2690 WARN_ON(p->nodes[0] != NULL);
2691 BUG_ON(!cow && ins_len);
2696 /* when we are removing items, we might have to go up to level
2697 * two as we update tree pointers Make sure we keep write
2698 * for those levels as well
2700 write_lock_level = 2;
2701 } else if (ins_len > 0) {
2703 * for inserting items, make sure we have a write lock on
2704 * level 1 so we can update keys
2706 write_lock_level = 1;
2710 write_lock_level = -1;
2712 if (cow && (p->keep_locks || p->lowest_level))
2713 write_lock_level = BTRFS_MAX_LEVEL;
2715 min_write_lock_level = write_lock_level;
2720 * we try very hard to do read locks on the root
2722 root_lock = BTRFS_READ_LOCK;
2724 if (p->search_commit_root) {
2726 * the commit roots are read only
2727 * so we always do read locks
2729 if (p->need_commit_sem)
2730 down_read(&root->fs_info->commit_root_sem);
2731 b = root->commit_root;
2732 extent_buffer_get(b);
2733 level = btrfs_header_level(b);
2734 if (p->need_commit_sem)
2735 up_read(&root->fs_info->commit_root_sem);
2736 if (!p->skip_locking)
2737 btrfs_tree_read_lock(b);
2739 if (p->skip_locking) {
2740 b = btrfs_root_node(root);
2741 level = btrfs_header_level(b);
2743 /* we don't know the level of the root node
2744 * until we actually have it read locked
2746 b = btrfs_read_lock_root_node(root);
2747 level = btrfs_header_level(b);
2748 if (level <= write_lock_level) {
2749 /* whoops, must trade for write lock */
2750 btrfs_tree_read_unlock(b);
2751 free_extent_buffer(b);
2752 b = btrfs_lock_root_node(root);
2753 root_lock = BTRFS_WRITE_LOCK;
2755 /* the level might have changed, check again */
2756 level = btrfs_header_level(b);
2760 p->nodes[level] = b;
2761 if (!p->skip_locking)
2762 p->locks[level] = root_lock;
2765 level = btrfs_header_level(b);
2768 * setup the path here so we can release it under lock
2769 * contention with the cow code
2772 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2775 * if we don't really need to cow this block
2776 * then we don't want to set the path blocking,
2777 * so we test it here
2779 if (!should_cow_block(trans, root, b)) {
2780 trans->dirty = true;
2785 * must have write locks on this node and the
2788 if (level > write_lock_level ||
2789 (level + 1 > write_lock_level &&
2790 level + 1 < BTRFS_MAX_LEVEL &&
2791 p->nodes[level + 1])) {
2792 write_lock_level = level + 1;
2793 btrfs_release_path(p);
2797 btrfs_set_path_blocking(p);
2799 err = btrfs_cow_block(trans, root, b, NULL, 0,
2802 err = btrfs_cow_block(trans, root, b,
2803 p->nodes[level + 1],
2804 p->slots[level + 1], &b);
2811 p->nodes[level] = b;
2812 btrfs_clear_path_blocking(p, NULL, 0);
2815 * we have a lock on b and as long as we aren't changing
2816 * the tree, there is no way to for the items in b to change.
2817 * It is safe to drop the lock on our parent before we
2818 * go through the expensive btree search on b.
2820 * If we're inserting or deleting (ins_len != 0), then we might
2821 * be changing slot zero, which may require changing the parent.
2822 * So, we can't drop the lock until after we know which slot
2823 * we're operating on.
2825 if (!ins_len && !p->keep_locks) {
2828 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2829 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2834 ret = key_search(b, key, level, &prev_cmp, &slot);
2838 if (ret && slot > 0) {
2842 p->slots[level] = slot;
2843 err = setup_nodes_for_search(trans, root, p, b, level,
2844 ins_len, &write_lock_level);
2851 b = p->nodes[level];
2852 slot = p->slots[level];
2855 * slot 0 is special, if we change the key
2856 * we have to update the parent pointer
2857 * which means we must have a write lock
2860 if (slot == 0 && ins_len &&
2861 write_lock_level < level + 1) {
2862 write_lock_level = level + 1;
2863 btrfs_release_path(p);
2867 unlock_up(p, level, lowest_unlock,
2868 min_write_lock_level, &write_lock_level);
2870 if (level == lowest_level) {
2876 err = read_block_for_search(trans, root, p,
2877 &b, level, slot, key, 0);
2885 if (!p->skip_locking) {
2886 level = btrfs_header_level(b);
2887 if (level <= write_lock_level) {
2888 err = btrfs_try_tree_write_lock(b);
2890 btrfs_set_path_blocking(p);
2892 btrfs_clear_path_blocking(p, b,
2895 p->locks[level] = BTRFS_WRITE_LOCK;
2897 err = btrfs_tree_read_lock_atomic(b);
2899 btrfs_set_path_blocking(p);
2900 btrfs_tree_read_lock(b);
2901 btrfs_clear_path_blocking(p, b,
2904 p->locks[level] = BTRFS_READ_LOCK;
2906 p->nodes[level] = b;
2909 p->slots[level] = slot;
2911 btrfs_leaf_free_space(root, b) < ins_len) {
2912 if (write_lock_level < 1) {
2913 write_lock_level = 1;
2914 btrfs_release_path(p);
2918 btrfs_set_path_blocking(p);
2919 err = split_leaf(trans, root, key,
2920 p, ins_len, ret == 0);
2921 btrfs_clear_path_blocking(p, NULL, 0);
2929 if (!p->search_for_split)
2930 unlock_up(p, level, lowest_unlock,
2931 min_write_lock_level, &write_lock_level);
2938 * we don't really know what they plan on doing with the path
2939 * from here on, so for now just mark it as blocking
2941 if (!p->leave_spinning)
2942 btrfs_set_path_blocking(p);
2943 if (ret < 0 && !p->skip_release_on_error)
2944 btrfs_release_path(p);
2949 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2950 * current state of the tree together with the operations recorded in the tree
2951 * modification log to search for the key in a previous version of this tree, as
2952 * denoted by the time_seq parameter.
2954 * Naturally, there is no support for insert, delete or cow operations.
2956 * The resulting path and return value will be set up as if we called
2957 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2959 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2960 struct btrfs_path *p, u64 time_seq)
2962 struct extent_buffer *b;
2967 int lowest_unlock = 1;
2968 u8 lowest_level = 0;
2971 lowest_level = p->lowest_level;
2972 WARN_ON(p->nodes[0] != NULL);
2974 if (p->search_commit_root) {
2976 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2980 b = get_old_root(root, time_seq);
2981 level = btrfs_header_level(b);
2982 p->locks[level] = BTRFS_READ_LOCK;
2985 level = btrfs_header_level(b);
2986 p->nodes[level] = b;
2987 btrfs_clear_path_blocking(p, NULL, 0);
2990 * we have a lock on b and as long as we aren't changing
2991 * the tree, there is no way to for the items in b to change.
2992 * It is safe to drop the lock on our parent before we
2993 * go through the expensive btree search on b.
2995 btrfs_unlock_up_safe(p, level + 1);
2998 * Since we can unwind eb's we want to do a real search every
3002 ret = key_search(b, key, level, &prev_cmp, &slot);
3006 if (ret && slot > 0) {
3010 p->slots[level] = slot;
3011 unlock_up(p, level, lowest_unlock, 0, NULL);
3013 if (level == lowest_level) {
3019 err = read_block_for_search(NULL, root, p, &b, level,
3020 slot, key, time_seq);
3028 level = btrfs_header_level(b);
3029 err = btrfs_tree_read_lock_atomic(b);
3031 btrfs_set_path_blocking(p);
3032 btrfs_tree_read_lock(b);
3033 btrfs_clear_path_blocking(p, b,
3036 b = tree_mod_log_rewind(root->fs_info, p, b, time_seq);
3041 p->locks[level] = BTRFS_READ_LOCK;
3042 p->nodes[level] = b;
3044 p->slots[level] = slot;
3045 unlock_up(p, level, lowest_unlock, 0, NULL);
3051 if (!p->leave_spinning)
3052 btrfs_set_path_blocking(p);
3054 btrfs_release_path(p);
3060 * helper to use instead of search slot if no exact match is needed but
3061 * instead the next or previous item should be returned.
3062 * When find_higher is true, the next higher item is returned, the next lower
3064 * When return_any and find_higher are both true, and no higher item is found,
3065 * return the next lower instead.
3066 * When return_any is true and find_higher is false, and no lower item is found,
3067 * return the next higher instead.
3068 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3071 int btrfs_search_slot_for_read(struct btrfs_root *root,
3072 struct btrfs_key *key, struct btrfs_path *p,
3073 int find_higher, int return_any)
3076 struct extent_buffer *leaf;
3079 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3083 * a return value of 1 means the path is at the position where the
3084 * item should be inserted. Normally this is the next bigger item,
3085 * but in case the previous item is the last in a leaf, path points
3086 * to the first free slot in the previous leaf, i.e. at an invalid
3092 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3093 ret = btrfs_next_leaf(root, p);
3099 * no higher item found, return the next
3104 btrfs_release_path(p);
3108 if (p->slots[0] == 0) {
3109 ret = btrfs_prev_leaf(root, p);
3114 if (p->slots[0] == btrfs_header_nritems(leaf))
3121 * no lower item found, return the next
3126 btrfs_release_path(p);
3136 * adjust the pointers going up the tree, starting at level
3137 * making sure the right key of each node is points to 'key'.
3138 * This is used after shifting pointers to the left, so it stops
3139 * fixing up pointers when a given leaf/node is not in slot 0 of the
3143 static void fixup_low_keys(struct btrfs_fs_info *fs_info,
3144 struct btrfs_path *path,
3145 struct btrfs_disk_key *key, int level)
3148 struct extent_buffer *t;
3150 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3151 int tslot = path->slots[i];
3152 if (!path->nodes[i])
3155 tree_mod_log_set_node_key(fs_info, t, tslot, 1);
3156 btrfs_set_node_key(t, key, tslot);
3157 btrfs_mark_buffer_dirty(path->nodes[i]);
3166 * This function isn't completely safe. It's the caller's responsibility
3167 * that the new key won't break the order
3169 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3170 struct btrfs_path *path,
3171 struct btrfs_key *new_key)
3173 struct btrfs_disk_key disk_key;
3174 struct extent_buffer *eb;
3177 eb = path->nodes[0];
3178 slot = path->slots[0];
3180 btrfs_item_key(eb, &disk_key, slot - 1);
3181 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3183 if (slot < btrfs_header_nritems(eb) - 1) {
3184 btrfs_item_key(eb, &disk_key, slot + 1);
3185 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3188 btrfs_cpu_key_to_disk(&disk_key, new_key);
3189 btrfs_set_item_key(eb, &disk_key, slot);
3190 btrfs_mark_buffer_dirty(eb);
3192 fixup_low_keys(fs_info, path, &disk_key, 1);
3196 * try to push data from one node into the next node left in the
3199 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3200 * error, and > 0 if there was no room in the left hand block.
3202 static int push_node_left(struct btrfs_trans_handle *trans,
3203 struct btrfs_root *root, struct extent_buffer *dst,
3204 struct extent_buffer *src, int empty)
3211 src_nritems = btrfs_header_nritems(src);
3212 dst_nritems = btrfs_header_nritems(dst);
3213 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3214 WARN_ON(btrfs_header_generation(src) != trans->transid);
3215 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3217 if (!empty && src_nritems <= 8)
3220 if (push_items <= 0)
3224 push_items = min(src_nritems, push_items);
3225 if (push_items < src_nritems) {
3226 /* leave at least 8 pointers in the node if
3227 * we aren't going to empty it
3229 if (src_nritems - push_items < 8) {
3230 if (push_items <= 8)
3236 push_items = min(src_nritems - 8, push_items);
3238 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3241 btrfs_abort_transaction(trans, root, ret);
3244 copy_extent_buffer(dst, src,
3245 btrfs_node_key_ptr_offset(dst_nritems),
3246 btrfs_node_key_ptr_offset(0),
3247 push_items * sizeof(struct btrfs_key_ptr));
3249 if (push_items < src_nritems) {
3251 * don't call tree_mod_log_eb_move here, key removal was already
3252 * fully logged by tree_mod_log_eb_copy above.
3254 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3255 btrfs_node_key_ptr_offset(push_items),
3256 (src_nritems - push_items) *
3257 sizeof(struct btrfs_key_ptr));
3259 btrfs_set_header_nritems(src, src_nritems - push_items);
3260 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3261 btrfs_mark_buffer_dirty(src);
3262 btrfs_mark_buffer_dirty(dst);
3268 * try to push data from one node into the next node right in the
3271 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3272 * error, and > 0 if there was no room in the right hand block.
3274 * this will only push up to 1/2 the contents of the left node over
3276 static int balance_node_right(struct btrfs_trans_handle *trans,
3277 struct btrfs_root *root,
3278 struct extent_buffer *dst,
3279 struct extent_buffer *src)
3287 WARN_ON(btrfs_header_generation(src) != trans->transid);
3288 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3290 src_nritems = btrfs_header_nritems(src);
3291 dst_nritems = btrfs_header_nritems(dst);
3292 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3293 if (push_items <= 0)
3296 if (src_nritems < 4)
3299 max_push = src_nritems / 2 + 1;
3300 /* don't try to empty the node */
3301 if (max_push >= src_nritems)
3304 if (max_push < push_items)
3305 push_items = max_push;
3307 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3308 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3309 btrfs_node_key_ptr_offset(0),
3311 sizeof(struct btrfs_key_ptr));
3313 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3314 src_nritems - push_items, push_items);
3316 btrfs_abort_transaction(trans, root, ret);
3319 copy_extent_buffer(dst, src,
3320 btrfs_node_key_ptr_offset(0),
3321 btrfs_node_key_ptr_offset(src_nritems - push_items),
3322 push_items * sizeof(struct btrfs_key_ptr));
3324 btrfs_set_header_nritems(src, src_nritems - push_items);
3325 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3327 btrfs_mark_buffer_dirty(src);
3328 btrfs_mark_buffer_dirty(dst);
3334 * helper function to insert a new root level in the tree.
3335 * A new node is allocated, and a single item is inserted to
3336 * point to the existing root
3338 * returns zero on success or < 0 on failure.
3340 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3341 struct btrfs_root *root,
3342 struct btrfs_path *path, int level)
3345 struct extent_buffer *lower;
3346 struct extent_buffer *c;
3347 struct extent_buffer *old;
3348 struct btrfs_disk_key lower_key;
3350 BUG_ON(path->nodes[level]);
3351 BUG_ON(path->nodes[level-1] != root->node);
3353 lower = path->nodes[level-1];
3355 btrfs_item_key(lower, &lower_key, 0);
3357 btrfs_node_key(lower, &lower_key, 0);
3359 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3360 &lower_key, level, root->node->start, 0);
3364 root_add_used(root, root->nodesize);
3366 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3367 btrfs_set_header_nritems(c, 1);
3368 btrfs_set_header_level(c, level);
3369 btrfs_set_header_bytenr(c, c->start);
3370 btrfs_set_header_generation(c, trans->transid);
3371 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3372 btrfs_set_header_owner(c, root->root_key.objectid);
3374 write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(),
3377 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3378 btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE);
3380 btrfs_set_node_key(c, &lower_key, 0);
3381 btrfs_set_node_blockptr(c, 0, lower->start);
3382 lower_gen = btrfs_header_generation(lower);
3383 WARN_ON(lower_gen != trans->transid);
3385 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3387 btrfs_mark_buffer_dirty(c);
3390 tree_mod_log_set_root_pointer(root, c, 0);
3391 rcu_assign_pointer(root->node, c);
3393 /* the super has an extra ref to root->node */
3394 free_extent_buffer(old);
3396 add_root_to_dirty_list(root);
3397 extent_buffer_get(c);
3398 path->nodes[level] = c;
3399 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3400 path->slots[level] = 0;
3405 * worker function to insert a single pointer in a node.
3406 * the node should have enough room for the pointer already
3408 * slot and level indicate where you want the key to go, and
3409 * blocknr is the block the key points to.
3411 static void insert_ptr(struct btrfs_trans_handle *trans,
3412 struct btrfs_root *root, struct btrfs_path *path,
3413 struct btrfs_disk_key *key, u64 bytenr,
3414 int slot, int level)
3416 struct extent_buffer *lower;
3420 BUG_ON(!path->nodes[level]);
3421 btrfs_assert_tree_locked(path->nodes[level]);
3422 lower = path->nodes[level];
3423 nritems = btrfs_header_nritems(lower);
3424 BUG_ON(slot > nritems);
3425 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3426 if (slot != nritems) {
3428 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3429 slot, nritems - slot);
3430 memmove_extent_buffer(lower,
3431 btrfs_node_key_ptr_offset(slot + 1),
3432 btrfs_node_key_ptr_offset(slot),
3433 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3436 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3437 MOD_LOG_KEY_ADD, GFP_NOFS);
3440 btrfs_set_node_key(lower, key, slot);
3441 btrfs_set_node_blockptr(lower, slot, bytenr);
3442 WARN_ON(trans->transid == 0);
3443 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3444 btrfs_set_header_nritems(lower, nritems + 1);
3445 btrfs_mark_buffer_dirty(lower);
3449 * split the node at the specified level in path in two.
3450 * The path is corrected to point to the appropriate node after the split
3452 * Before splitting this tries to make some room in the node by pushing
3453 * left and right, if either one works, it returns right away.
3455 * returns 0 on success and < 0 on failure
3457 static noinline int split_node(struct btrfs_trans_handle *trans,
3458 struct btrfs_root *root,
3459 struct btrfs_path *path, int level)
3461 struct extent_buffer *c;
3462 struct extent_buffer *split;
3463 struct btrfs_disk_key disk_key;
3468 c = path->nodes[level];
3469 WARN_ON(btrfs_header_generation(c) != trans->transid);
3470 if (c == root->node) {
3472 * trying to split the root, lets make a new one
3474 * tree mod log: We don't log_removal old root in
3475 * insert_new_root, because that root buffer will be kept as a
3476 * normal node. We are going to log removal of half of the
3477 * elements below with tree_mod_log_eb_copy. We're holding a
3478 * tree lock on the buffer, which is why we cannot race with
3479 * other tree_mod_log users.
3481 ret = insert_new_root(trans, root, path, level + 1);
3485 ret = push_nodes_for_insert(trans, root, path, level);
3486 c = path->nodes[level];
3487 if (!ret && btrfs_header_nritems(c) <
3488 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3494 c_nritems = btrfs_header_nritems(c);
3495 mid = (c_nritems + 1) / 2;
3496 btrfs_node_key(c, &disk_key, mid);
3498 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3499 &disk_key, level, c->start, 0);
3501 return PTR_ERR(split);
3503 root_add_used(root, root->nodesize);
3505 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3506 btrfs_set_header_level(split, btrfs_header_level(c));
3507 btrfs_set_header_bytenr(split, split->start);
3508 btrfs_set_header_generation(split, trans->transid);
3509 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3510 btrfs_set_header_owner(split, root->root_key.objectid);
3511 write_extent_buffer(split, root->fs_info->fsid,
3512 btrfs_header_fsid(), BTRFS_FSID_SIZE);
3513 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3514 btrfs_header_chunk_tree_uuid(split),
3517 ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0,
3518 mid, c_nritems - mid);
3520 btrfs_abort_transaction(trans, root, ret);
3523 copy_extent_buffer(split, c,
3524 btrfs_node_key_ptr_offset(0),
3525 btrfs_node_key_ptr_offset(mid),
3526 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3527 btrfs_set_header_nritems(split, c_nritems - mid);
3528 btrfs_set_header_nritems(c, mid);
3531 btrfs_mark_buffer_dirty(c);
3532 btrfs_mark_buffer_dirty(split);
3534 insert_ptr(trans, root, path, &disk_key, split->start,
3535 path->slots[level + 1] + 1, level + 1);
3537 if (path->slots[level] >= mid) {
3538 path->slots[level] -= mid;
3539 btrfs_tree_unlock(c);
3540 free_extent_buffer(c);
3541 path->nodes[level] = split;
3542 path->slots[level + 1] += 1;
3544 btrfs_tree_unlock(split);
3545 free_extent_buffer(split);
3551 * how many bytes are required to store the items in a leaf. start
3552 * and nr indicate which items in the leaf to check. This totals up the
3553 * space used both by the item structs and the item data
3555 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3557 struct btrfs_item *start_item;
3558 struct btrfs_item *end_item;
3559 struct btrfs_map_token token;
3561 int nritems = btrfs_header_nritems(l);
3562 int end = min(nritems, start + nr) - 1;
3566 btrfs_init_map_token(&token);
3567 start_item = btrfs_item_nr(start);
3568 end_item = btrfs_item_nr(end);
3569 data_len = btrfs_token_item_offset(l, start_item, &token) +
3570 btrfs_token_item_size(l, start_item, &token);
3571 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3572 data_len += sizeof(struct btrfs_item) * nr;
3573 WARN_ON(data_len < 0);
3578 * The space between the end of the leaf items and
3579 * the start of the leaf data. IOW, how much room
3580 * the leaf has left for both items and data
3582 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3583 struct extent_buffer *leaf)
3585 int nritems = btrfs_header_nritems(leaf);
3587 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3589 btrfs_crit(root->fs_info,
3590 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3591 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3592 leaf_space_used(leaf, 0, nritems), nritems);
3598 * min slot controls the lowest index we're willing to push to the
3599 * right. We'll push up to and including min_slot, but no lower
3601 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3602 struct btrfs_root *root,
3603 struct btrfs_path *path,
3604 int data_size, int empty,
3605 struct extent_buffer *right,
3606 int free_space, u32 left_nritems,
3609 struct extent_buffer *left = path->nodes[0];
3610 struct extent_buffer *upper = path->nodes[1];
3611 struct btrfs_map_token token;
3612 struct btrfs_disk_key disk_key;
3617 struct btrfs_item *item;
3623 btrfs_init_map_token(&token);
3628 nr = max_t(u32, 1, min_slot);
3630 if (path->slots[0] >= left_nritems)
3631 push_space += data_size;
3633 slot = path->slots[1];
3634 i = left_nritems - 1;
3636 item = btrfs_item_nr(i);
3638 if (!empty && push_items > 0) {
3639 if (path->slots[0] > i)
3641 if (path->slots[0] == i) {
3642 int space = btrfs_leaf_free_space(root, left);
3643 if (space + push_space * 2 > free_space)
3648 if (path->slots[0] == i)
3649 push_space += data_size;
3651 this_item_size = btrfs_item_size(left, item);
3652 if (this_item_size + sizeof(*item) + push_space > free_space)
3656 push_space += this_item_size + sizeof(*item);
3662 if (push_items == 0)
3665 WARN_ON(!empty && push_items == left_nritems);
3667 /* push left to right */
3668 right_nritems = btrfs_header_nritems(right);
3670 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3671 push_space -= leaf_data_end(root, left);
3673 /* make room in the right data area */
3674 data_end = leaf_data_end(root, right);
3675 memmove_extent_buffer(right,
3676 btrfs_leaf_data(right) + data_end - push_space,
3677 btrfs_leaf_data(right) + data_end,
3678 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3680 /* copy from the left data area */
3681 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3682 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3683 btrfs_leaf_data(left) + leaf_data_end(root, left),
3686 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3687 btrfs_item_nr_offset(0),
3688 right_nritems * sizeof(struct btrfs_item));
3690 /* copy the items from left to right */
3691 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3692 btrfs_item_nr_offset(left_nritems - push_items),
3693 push_items * sizeof(struct btrfs_item));
3695 /* update the item pointers */
3696 right_nritems += push_items;
3697 btrfs_set_header_nritems(right, right_nritems);
3698 push_space = BTRFS_LEAF_DATA_SIZE(root);
3699 for (i = 0; i < right_nritems; i++) {
3700 item = btrfs_item_nr(i);
3701 push_space -= btrfs_token_item_size(right, item, &token);
3702 btrfs_set_token_item_offset(right, item, push_space, &token);
3705 left_nritems -= push_items;
3706 btrfs_set_header_nritems(left, left_nritems);
3709 btrfs_mark_buffer_dirty(left);
3711 clean_tree_block(trans, root->fs_info, left);
3713 btrfs_mark_buffer_dirty(right);
3715 btrfs_item_key(right, &disk_key, 0);
3716 btrfs_set_node_key(upper, &disk_key, slot + 1);
3717 btrfs_mark_buffer_dirty(upper);
3719 /* then fixup the leaf pointer in the path */
3720 if (path->slots[0] >= left_nritems) {
3721 path->slots[0] -= left_nritems;
3722 if (btrfs_header_nritems(path->nodes[0]) == 0)
3723 clean_tree_block(trans, root->fs_info, path->nodes[0]);
3724 btrfs_tree_unlock(path->nodes[0]);
3725 free_extent_buffer(path->nodes[0]);
3726 path->nodes[0] = right;
3727 path->slots[1] += 1;
3729 btrfs_tree_unlock(right);
3730 free_extent_buffer(right);
3735 btrfs_tree_unlock(right);
3736 free_extent_buffer(right);
3741 * push some data in the path leaf to the right, trying to free up at
3742 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3744 * returns 1 if the push failed because the other node didn't have enough
3745 * room, 0 if everything worked out and < 0 if there were major errors.
3747 * this will push starting from min_slot to the end of the leaf. It won't
3748 * push any slot lower than min_slot
3750 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3751 *root, struct btrfs_path *path,
3752 int min_data_size, int data_size,
3753 int empty, u32 min_slot)
3755 struct extent_buffer *left = path->nodes[0];
3756 struct extent_buffer *right;
3757 struct extent_buffer *upper;
3763 if (!path->nodes[1])
3766 slot = path->slots[1];
3767 upper = path->nodes[1];
3768 if (slot >= btrfs_header_nritems(upper) - 1)
3771 btrfs_assert_tree_locked(path->nodes[1]);
3773 right = read_node_slot(root, upper, slot + 1);
3777 btrfs_tree_lock(right);
3778 btrfs_set_lock_blocking(right);
3780 free_space = btrfs_leaf_free_space(root, right);
3781 if (free_space < data_size)
3784 /* cow and double check */
3785 ret = btrfs_cow_block(trans, root, right, upper,
3790 free_space = btrfs_leaf_free_space(root, right);
3791 if (free_space < data_size)
3794 left_nritems = btrfs_header_nritems(left);
3795 if (left_nritems == 0)
3798 if (path->slots[0] == left_nritems && !empty) {
3799 /* Key greater than all keys in the leaf, right neighbor has
3800 * enough room for it and we're not emptying our leaf to delete
3801 * it, therefore use right neighbor to insert the new item and
3802 * no need to touch/dirty our left leaft. */
3803 btrfs_tree_unlock(left);
3804 free_extent_buffer(left);
3805 path->nodes[0] = right;
3811 return __push_leaf_right(trans, root, path, min_data_size, empty,
3812 right, free_space, left_nritems, min_slot);
3814 btrfs_tree_unlock(right);
3815 free_extent_buffer(right);
3820 * push some data in the path leaf to the left, trying to free up at
3821 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3823 * max_slot can put a limit on how far into the leaf we'll push items. The
3824 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3827 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3828 struct btrfs_root *root,
3829 struct btrfs_path *path, int data_size,
3830 int empty, struct extent_buffer *left,
3831 int free_space, u32 right_nritems,
3834 struct btrfs_disk_key disk_key;
3835 struct extent_buffer *right = path->nodes[0];
3839 struct btrfs_item *item;
3840 u32 old_left_nritems;
3844 u32 old_left_item_size;
3845 struct btrfs_map_token token;
3847 btrfs_init_map_token(&token);
3850 nr = min(right_nritems, max_slot);
3852 nr = min(right_nritems - 1, max_slot);
3854 for (i = 0; i < nr; i++) {
3855 item = btrfs_item_nr(i);
3857 if (!empty && push_items > 0) {
3858 if (path->slots[0] < i)
3860 if (path->slots[0] == i) {
3861 int space = btrfs_leaf_free_space(root, right);
3862 if (space + push_space * 2 > free_space)
3867 if (path->slots[0] == i)
3868 push_space += data_size;
3870 this_item_size = btrfs_item_size(right, item);
3871 if (this_item_size + sizeof(*item) + push_space > free_space)
3875 push_space += this_item_size + sizeof(*item);
3878 if (push_items == 0) {
3882 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3884 /* push data from right to left */
3885 copy_extent_buffer(left, right,
3886 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3887 btrfs_item_nr_offset(0),
3888 push_items * sizeof(struct btrfs_item));
3890 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3891 btrfs_item_offset_nr(right, push_items - 1);
3893 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3894 leaf_data_end(root, left) - push_space,
3895 btrfs_leaf_data(right) +
3896 btrfs_item_offset_nr(right, push_items - 1),
3898 old_left_nritems = btrfs_header_nritems(left);
3899 BUG_ON(old_left_nritems <= 0);
3901 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3902 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3905 item = btrfs_item_nr(i);
3907 ioff = btrfs_token_item_offset(left, item, &token);
3908 btrfs_set_token_item_offset(left, item,
3909 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3912 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3914 /* fixup right node */
3915 if (push_items > right_nritems)
3916 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3919 if (push_items < right_nritems) {
3920 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3921 leaf_data_end(root, right);
3922 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3923 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3924 btrfs_leaf_data(right) +
3925 leaf_data_end(root, right), push_space);
3927 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3928 btrfs_item_nr_offset(push_items),
3929 (btrfs_header_nritems(right) - push_items) *
3930 sizeof(struct btrfs_item));
3932 right_nritems -= push_items;
3933 btrfs_set_header_nritems(right, right_nritems);
3934 push_space = BTRFS_LEAF_DATA_SIZE(root);
3935 for (i = 0; i < right_nritems; i++) {
3936 item = btrfs_item_nr(i);
3938 push_space = push_space - btrfs_token_item_size(right,
3940 btrfs_set_token_item_offset(right, item, push_space, &token);
3943 btrfs_mark_buffer_dirty(left);
3945 btrfs_mark_buffer_dirty(right);
3947 clean_tree_block(trans, root->fs_info, right);
3949 btrfs_item_key(right, &disk_key, 0);
3950 fixup_low_keys(root->fs_info, path, &disk_key, 1);
3952 /* then fixup the leaf pointer in the path */
3953 if (path->slots[0] < push_items) {
3954 path->slots[0] += old_left_nritems;
3955 btrfs_tree_unlock(path->nodes[0]);
3956 free_extent_buffer(path->nodes[0]);
3957 path->nodes[0] = left;
3958 path->slots[1] -= 1;
3960 btrfs_tree_unlock(left);
3961 free_extent_buffer(left);
3962 path->slots[0] -= push_items;
3964 BUG_ON(path->slots[0] < 0);
3967 btrfs_tree_unlock(left);
3968 free_extent_buffer(left);
3973 * push some data in the path leaf to the left, trying to free up at
3974 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3976 * max_slot can put a limit on how far into the leaf we'll push items. The
3977 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3980 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3981 *root, struct btrfs_path *path, int min_data_size,
3982 int data_size, int empty, u32 max_slot)
3984 struct extent_buffer *right = path->nodes[0];
3985 struct extent_buffer *left;
3991 slot = path->slots[1];
3994 if (!path->nodes[1])
3997 right_nritems = btrfs_header_nritems(right);
3998 if (right_nritems == 0)
4001 btrfs_assert_tree_locked(path->nodes[1]);
4003 left = read_node_slot(root, path->nodes[1], slot - 1);
4007 btrfs_tree_lock(left);
4008 btrfs_set_lock_blocking(left);
4010 free_space = btrfs_leaf_free_space(root, left);
4011 if (free_space < data_size) {
4016 /* cow and double check */
4017 ret = btrfs_cow_block(trans, root, left,
4018 path->nodes[1], slot - 1, &left);
4020 /* we hit -ENOSPC, but it isn't fatal here */
4026 free_space = btrfs_leaf_free_space(root, left);
4027 if (free_space < data_size) {
4032 return __push_leaf_left(trans, root, path, min_data_size,
4033 empty, left, free_space, right_nritems,
4036 btrfs_tree_unlock(left);
4037 free_extent_buffer(left);
4042 * split the path's leaf in two, making sure there is at least data_size
4043 * available for the resulting leaf level of the path.
4045 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4046 struct btrfs_root *root,
4047 struct btrfs_path *path,
4048 struct extent_buffer *l,
4049 struct extent_buffer *right,
4050 int slot, int mid, int nritems)
4055 struct btrfs_disk_key disk_key;
4056 struct btrfs_map_token token;
4058 btrfs_init_map_token(&token);
4060 nritems = nritems - mid;
4061 btrfs_set_header_nritems(right, nritems);
4062 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
4064 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4065 btrfs_item_nr_offset(mid),
4066 nritems * sizeof(struct btrfs_item));
4068 copy_extent_buffer(right, l,
4069 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
4070 data_copy_size, btrfs_leaf_data(l) +
4071 leaf_data_end(root, l), data_copy_size);
4073 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
4074 btrfs_item_end_nr(l, mid);
4076 for (i = 0; i < nritems; i++) {
4077 struct btrfs_item *item = btrfs_item_nr(i);
4080 ioff = btrfs_token_item_offset(right, item, &token);
4081 btrfs_set_token_item_offset(right, item,
4082 ioff + rt_data_off, &token);
4085 btrfs_set_header_nritems(l, mid);
4086 btrfs_item_key(right, &disk_key, 0);
4087 insert_ptr(trans, root, path, &disk_key, right->start,
4088 path->slots[1] + 1, 1);
4090 btrfs_mark_buffer_dirty(right);
4091 btrfs_mark_buffer_dirty(l);
4092 BUG_ON(path->slots[0] != slot);
4095 btrfs_tree_unlock(path->nodes[0]);
4096 free_extent_buffer(path->nodes[0]);
4097 path->nodes[0] = right;
4098 path->slots[0] -= mid;
4099 path->slots[1] += 1;
4101 btrfs_tree_unlock(right);
4102 free_extent_buffer(right);
4105 BUG_ON(path->slots[0] < 0);
4109 * double splits happen when we need to insert a big item in the middle
4110 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4111 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4114 * We avoid this by trying to push the items on either side of our target
4115 * into the adjacent leaves. If all goes well we can avoid the double split
4118 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4119 struct btrfs_root *root,
4120 struct btrfs_path *path,
4127 int space_needed = data_size;
4129 slot = path->slots[0];
4130 if (slot < btrfs_header_nritems(path->nodes[0]))
4131 space_needed -= btrfs_leaf_free_space(root, path->nodes[0]);
4134 * try to push all the items after our slot into the
4137 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4144 nritems = btrfs_header_nritems(path->nodes[0]);
4146 * our goal is to get our slot at the start or end of a leaf. If
4147 * we've done so we're done
4149 if (path->slots[0] == 0 || path->slots[0] == nritems)
4152 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4155 /* try to push all the items before our slot into the next leaf */
4156 slot = path->slots[0];
4157 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4170 * split the path's leaf in two, making sure there is at least data_size
4171 * available for the resulting leaf level of the path.
4173 * returns 0 if all went well and < 0 on failure.
4175 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4176 struct btrfs_root *root,
4177 struct btrfs_key *ins_key,
4178 struct btrfs_path *path, int data_size,
4181 struct btrfs_disk_key disk_key;
4182 struct extent_buffer *l;
4186 struct extent_buffer *right;
4187 struct btrfs_fs_info *fs_info = root->fs_info;
4191 int num_doubles = 0;
4192 int tried_avoid_double = 0;
4195 slot = path->slots[0];
4196 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4197 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
4200 /* first try to make some room by pushing left and right */
4201 if (data_size && path->nodes[1]) {
4202 int space_needed = data_size;
4204 if (slot < btrfs_header_nritems(l))
4205 space_needed -= btrfs_leaf_free_space(root, l);
4207 wret = push_leaf_right(trans, root, path, space_needed,
4208 space_needed, 0, 0);
4212 wret = push_leaf_left(trans, root, path, space_needed,
4213 space_needed, 0, (u32)-1);
4219 /* did the pushes work? */
4220 if (btrfs_leaf_free_space(root, l) >= data_size)
4224 if (!path->nodes[1]) {
4225 ret = insert_new_root(trans, root, path, 1);
4232 slot = path->slots[0];
4233 nritems = btrfs_header_nritems(l);
4234 mid = (nritems + 1) / 2;
4238 leaf_space_used(l, mid, nritems - mid) + data_size >
4239 BTRFS_LEAF_DATA_SIZE(root)) {
4240 if (slot >= nritems) {
4244 if (mid != nritems &&
4245 leaf_space_used(l, mid, nritems - mid) +
4246 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4247 if (data_size && !tried_avoid_double)
4248 goto push_for_double;
4254 if (leaf_space_used(l, 0, mid) + data_size >
4255 BTRFS_LEAF_DATA_SIZE(root)) {
4256 if (!extend && data_size && slot == 0) {
4258 } else if ((extend || !data_size) && slot == 0) {
4262 if (mid != nritems &&
4263 leaf_space_used(l, mid, nritems - mid) +
4264 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4265 if (data_size && !tried_avoid_double)
4266 goto push_for_double;
4274 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4276 btrfs_item_key(l, &disk_key, mid);
4278 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4279 &disk_key, 0, l->start, 0);
4281 return PTR_ERR(right);
4283 root_add_used(root, root->nodesize);
4285 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4286 btrfs_set_header_bytenr(right, right->start);
4287 btrfs_set_header_generation(right, trans->transid);
4288 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4289 btrfs_set_header_owner(right, root->root_key.objectid);
4290 btrfs_set_header_level(right, 0);
4291 write_extent_buffer(right, fs_info->fsid,
4292 btrfs_header_fsid(), BTRFS_FSID_SIZE);
4294 write_extent_buffer(right, fs_info->chunk_tree_uuid,
4295 btrfs_header_chunk_tree_uuid(right),
4300 btrfs_set_header_nritems(right, 0);
4301 insert_ptr(trans, root, path, &disk_key, right->start,
4302 path->slots[1] + 1, 1);
4303 btrfs_tree_unlock(path->nodes[0]);
4304 free_extent_buffer(path->nodes[0]);
4305 path->nodes[0] = right;
4307 path->slots[1] += 1;
4309 btrfs_set_header_nritems(right, 0);
4310 insert_ptr(trans, root, path, &disk_key, right->start,
4312 btrfs_tree_unlock(path->nodes[0]);
4313 free_extent_buffer(path->nodes[0]);
4314 path->nodes[0] = right;
4316 if (path->slots[1] == 0)
4317 fixup_low_keys(fs_info, path, &disk_key, 1);
4319 btrfs_mark_buffer_dirty(right);
4323 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4326 BUG_ON(num_doubles != 0);
4334 push_for_double_split(trans, root, path, data_size);
4335 tried_avoid_double = 1;
4336 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4341 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4342 struct btrfs_root *root,
4343 struct btrfs_path *path, int ins_len)
4345 struct btrfs_key key;
4346 struct extent_buffer *leaf;
4347 struct btrfs_file_extent_item *fi;
4352 leaf = path->nodes[0];
4353 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4355 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4356 key.type != BTRFS_EXTENT_CSUM_KEY);
4358 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4361 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4362 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4363 fi = btrfs_item_ptr(leaf, path->slots[0],
4364 struct btrfs_file_extent_item);
4365 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4367 btrfs_release_path(path);
4369 path->keep_locks = 1;
4370 path->search_for_split = 1;
4371 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4372 path->search_for_split = 0;
4379 leaf = path->nodes[0];
4380 /* if our item isn't there, return now */
4381 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4384 /* the leaf has changed, it now has room. return now */
4385 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4388 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4389 fi = btrfs_item_ptr(leaf, path->slots[0],
4390 struct btrfs_file_extent_item);
4391 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4395 btrfs_set_path_blocking(path);
4396 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4400 path->keep_locks = 0;
4401 btrfs_unlock_up_safe(path, 1);
4404 path->keep_locks = 0;
4408 static noinline int split_item(struct btrfs_trans_handle *trans,
4409 struct btrfs_root *root,
4410 struct btrfs_path *path,
4411 struct btrfs_key *new_key,
4412 unsigned long split_offset)
4414 struct extent_buffer *leaf;
4415 struct btrfs_item *item;
4416 struct btrfs_item *new_item;
4422 struct btrfs_disk_key disk_key;
4424 leaf = path->nodes[0];
4425 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4427 btrfs_set_path_blocking(path);
4429 item = btrfs_item_nr(path->slots[0]);
4430 orig_offset = btrfs_item_offset(leaf, item);
4431 item_size = btrfs_item_size(leaf, item);
4433 buf = kmalloc(item_size, GFP_NOFS);
4437 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4438 path->slots[0]), item_size);
4440 slot = path->slots[0] + 1;
4441 nritems = btrfs_header_nritems(leaf);
4442 if (slot != nritems) {
4443 /* shift the items */
4444 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4445 btrfs_item_nr_offset(slot),
4446 (nritems - slot) * sizeof(struct btrfs_item));
4449 btrfs_cpu_key_to_disk(&disk_key, new_key);
4450 btrfs_set_item_key(leaf, &disk_key, slot);
4452 new_item = btrfs_item_nr(slot);
4454 btrfs_set_item_offset(leaf, new_item, orig_offset);
4455 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4457 btrfs_set_item_offset(leaf, item,
4458 orig_offset + item_size - split_offset);
4459 btrfs_set_item_size(leaf, item, split_offset);
4461 btrfs_set_header_nritems(leaf, nritems + 1);
4463 /* write the data for the start of the original item */
4464 write_extent_buffer(leaf, buf,
4465 btrfs_item_ptr_offset(leaf, path->slots[0]),
4468 /* write the data for the new item */
4469 write_extent_buffer(leaf, buf + split_offset,
4470 btrfs_item_ptr_offset(leaf, slot),
4471 item_size - split_offset);
4472 btrfs_mark_buffer_dirty(leaf);
4474 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4480 * This function splits a single item into two items,
4481 * giving 'new_key' to the new item and splitting the
4482 * old one at split_offset (from the start of the item).
4484 * The path may be released by this operation. After
4485 * the split, the path is pointing to the old item. The
4486 * new item is going to be in the same node as the old one.
4488 * Note, the item being split must be smaller enough to live alone on
4489 * a tree block with room for one extra struct btrfs_item
4491 * This allows us to split the item in place, keeping a lock on the
4492 * leaf the entire time.
4494 int btrfs_split_item(struct btrfs_trans_handle *trans,
4495 struct btrfs_root *root,
4496 struct btrfs_path *path,
4497 struct btrfs_key *new_key,
4498 unsigned long split_offset)
4501 ret = setup_leaf_for_split(trans, root, path,
4502 sizeof(struct btrfs_item));
4506 ret = split_item(trans, root, path, new_key, split_offset);
4511 * This function duplicate a item, giving 'new_key' to the new item.
4512 * It guarantees both items live in the same tree leaf and the new item
4513 * is contiguous with the original item.
4515 * This allows us to split file extent in place, keeping a lock on the
4516 * leaf the entire time.
4518 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4519 struct btrfs_root *root,
4520 struct btrfs_path *path,
4521 struct btrfs_key *new_key)
4523 struct extent_buffer *leaf;
4527 leaf = path->nodes[0];
4528 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4529 ret = setup_leaf_for_split(trans, root, path,
4530 item_size + sizeof(struct btrfs_item));
4535 setup_items_for_insert(root, path, new_key, &item_size,
4536 item_size, item_size +
4537 sizeof(struct btrfs_item), 1);
4538 leaf = path->nodes[0];
4539 memcpy_extent_buffer(leaf,
4540 btrfs_item_ptr_offset(leaf, path->slots[0]),
4541 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4547 * make the item pointed to by the path smaller. new_size indicates
4548 * how small to make it, and from_end tells us if we just chop bytes
4549 * off the end of the item or if we shift the item to chop bytes off
4552 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4553 u32 new_size, int from_end)
4556 struct extent_buffer *leaf;
4557 struct btrfs_item *item;
4559 unsigned int data_end;
4560 unsigned int old_data_start;
4561 unsigned int old_size;
4562 unsigned int size_diff;
4564 struct btrfs_map_token token;
4566 btrfs_init_map_token(&token);
4568 leaf = path->nodes[0];
4569 slot = path->slots[0];
4571 old_size = btrfs_item_size_nr(leaf, slot);
4572 if (old_size == new_size)
4575 nritems = btrfs_header_nritems(leaf);
4576 data_end = leaf_data_end(root, leaf);
4578 old_data_start = btrfs_item_offset_nr(leaf, slot);
4580 size_diff = old_size - new_size;
4583 BUG_ON(slot >= nritems);
4586 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4588 /* first correct the data pointers */
4589 for (i = slot; i < nritems; i++) {
4591 item = btrfs_item_nr(i);
4593 ioff = btrfs_token_item_offset(leaf, item, &token);
4594 btrfs_set_token_item_offset(leaf, item,
4595 ioff + size_diff, &token);
4598 /* shift the data */
4600 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4601 data_end + size_diff, btrfs_leaf_data(leaf) +
4602 data_end, old_data_start + new_size - data_end);
4604 struct btrfs_disk_key disk_key;
4607 btrfs_item_key(leaf, &disk_key, slot);
4609 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4611 struct btrfs_file_extent_item *fi;
4613 fi = btrfs_item_ptr(leaf, slot,
4614 struct btrfs_file_extent_item);
4615 fi = (struct btrfs_file_extent_item *)(
4616 (unsigned long)fi - size_diff);
4618 if (btrfs_file_extent_type(leaf, fi) ==
4619 BTRFS_FILE_EXTENT_INLINE) {
4620 ptr = btrfs_item_ptr_offset(leaf, slot);
4621 memmove_extent_buffer(leaf, ptr,
4623 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4627 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4628 data_end + size_diff, btrfs_leaf_data(leaf) +
4629 data_end, old_data_start - data_end);
4631 offset = btrfs_disk_key_offset(&disk_key);
4632 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4633 btrfs_set_item_key(leaf, &disk_key, slot);
4635 fixup_low_keys(root->fs_info, path, &disk_key, 1);
4638 item = btrfs_item_nr(slot);
4639 btrfs_set_item_size(leaf, item, new_size);
4640 btrfs_mark_buffer_dirty(leaf);
4642 if (btrfs_leaf_free_space(root, leaf) < 0) {
4643 btrfs_print_leaf(root, leaf);
4649 * make the item pointed to by the path bigger, data_size is the added size.
4651 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4655 struct extent_buffer *leaf;
4656 struct btrfs_item *item;
4658 unsigned int data_end;
4659 unsigned int old_data;
4660 unsigned int old_size;
4662 struct btrfs_map_token token;
4664 btrfs_init_map_token(&token);
4666 leaf = path->nodes[0];
4668 nritems = btrfs_header_nritems(leaf);
4669 data_end = leaf_data_end(root, leaf);
4671 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4672 btrfs_print_leaf(root, leaf);
4675 slot = path->slots[0];
4676 old_data = btrfs_item_end_nr(leaf, slot);
4679 if (slot >= nritems) {
4680 btrfs_print_leaf(root, leaf);
4681 btrfs_crit(root->fs_info, "slot %d too large, nritems %d",
4687 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4689 /* first correct the data pointers */
4690 for (i = slot; i < nritems; i++) {
4692 item = btrfs_item_nr(i);
4694 ioff = btrfs_token_item_offset(leaf, item, &token);
4695 btrfs_set_token_item_offset(leaf, item,
4696 ioff - data_size, &token);
4699 /* shift the data */
4700 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4701 data_end - data_size, btrfs_leaf_data(leaf) +
4702 data_end, old_data - data_end);
4704 data_end = old_data;
4705 old_size = btrfs_item_size_nr(leaf, slot);
4706 item = btrfs_item_nr(slot);
4707 btrfs_set_item_size(leaf, item, old_size + data_size);
4708 btrfs_mark_buffer_dirty(leaf);
4710 if (btrfs_leaf_free_space(root, leaf) < 0) {
4711 btrfs_print_leaf(root, leaf);
4717 * this is a helper for btrfs_insert_empty_items, the main goal here is
4718 * to save stack depth by doing the bulk of the work in a function
4719 * that doesn't call btrfs_search_slot
4721 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4722 struct btrfs_key *cpu_key, u32 *data_size,
4723 u32 total_data, u32 total_size, int nr)
4725 struct btrfs_item *item;
4728 unsigned int data_end;
4729 struct btrfs_disk_key disk_key;
4730 struct extent_buffer *leaf;
4732 struct btrfs_map_token token;
4734 if (path->slots[0] == 0) {
4735 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4736 fixup_low_keys(root->fs_info, path, &disk_key, 1);
4738 btrfs_unlock_up_safe(path, 1);
4740 btrfs_init_map_token(&token);
4742 leaf = path->nodes[0];
4743 slot = path->slots[0];
4745 nritems = btrfs_header_nritems(leaf);
4746 data_end = leaf_data_end(root, leaf);
4748 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4749 btrfs_print_leaf(root, leaf);
4750 btrfs_crit(root->fs_info, "not enough freespace need %u have %d",
4751 total_size, btrfs_leaf_free_space(root, leaf));
4755 if (slot != nritems) {
4756 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4758 if (old_data < data_end) {
4759 btrfs_print_leaf(root, leaf);
4760 btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d",
4761 slot, old_data, data_end);
4765 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4767 /* first correct the data pointers */
4768 for (i = slot; i < nritems; i++) {
4771 item = btrfs_item_nr( i);
4772 ioff = btrfs_token_item_offset(leaf, item, &token);
4773 btrfs_set_token_item_offset(leaf, item,
4774 ioff - total_data, &token);
4776 /* shift the items */
4777 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4778 btrfs_item_nr_offset(slot),
4779 (nritems - slot) * sizeof(struct btrfs_item));
4781 /* shift the data */
4782 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4783 data_end - total_data, btrfs_leaf_data(leaf) +
4784 data_end, old_data - data_end);
4785 data_end = old_data;
4788 /* setup the item for the new data */
4789 for (i = 0; i < nr; i++) {
4790 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4791 btrfs_set_item_key(leaf, &disk_key, slot + i);
4792 item = btrfs_item_nr(slot + i);
4793 btrfs_set_token_item_offset(leaf, item,
4794 data_end - data_size[i], &token);
4795 data_end -= data_size[i];
4796 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4799 btrfs_set_header_nritems(leaf, nritems + nr);
4800 btrfs_mark_buffer_dirty(leaf);
4802 if (btrfs_leaf_free_space(root, leaf) < 0) {
4803 btrfs_print_leaf(root, leaf);
4809 * Given a key and some data, insert items into the tree.
4810 * This does all the path init required, making room in the tree if needed.
4812 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4813 struct btrfs_root *root,
4814 struct btrfs_path *path,
4815 struct btrfs_key *cpu_key, u32 *data_size,
4824 for (i = 0; i < nr; i++)
4825 total_data += data_size[i];
4827 total_size = total_data + (nr * sizeof(struct btrfs_item));
4828 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4834 slot = path->slots[0];
4837 setup_items_for_insert(root, path, cpu_key, data_size,
4838 total_data, total_size, nr);
4843 * Given a key and some data, insert an item into the tree.
4844 * This does all the path init required, making room in the tree if needed.
4846 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4847 *root, struct btrfs_key *cpu_key, void *data, u32
4851 struct btrfs_path *path;
4852 struct extent_buffer *leaf;
4855 path = btrfs_alloc_path();
4858 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4860 leaf = path->nodes[0];
4861 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4862 write_extent_buffer(leaf, data, ptr, data_size);
4863 btrfs_mark_buffer_dirty(leaf);
4865 btrfs_free_path(path);
4870 * delete the pointer from a given node.
4872 * the tree should have been previously balanced so the deletion does not
4875 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4876 int level, int slot)
4878 struct extent_buffer *parent = path->nodes[level];
4882 nritems = btrfs_header_nritems(parent);
4883 if (slot != nritems - 1) {
4885 tree_mod_log_eb_move(root->fs_info, parent, slot,
4886 slot + 1, nritems - slot - 1);
4887 memmove_extent_buffer(parent,
4888 btrfs_node_key_ptr_offset(slot),
4889 btrfs_node_key_ptr_offset(slot + 1),
4890 sizeof(struct btrfs_key_ptr) *
4891 (nritems - slot - 1));
4893 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4894 MOD_LOG_KEY_REMOVE, GFP_NOFS);
4899 btrfs_set_header_nritems(parent, nritems);
4900 if (nritems == 0 && parent == root->node) {
4901 BUG_ON(btrfs_header_level(root->node) != 1);
4902 /* just turn the root into a leaf and break */
4903 btrfs_set_header_level(root->node, 0);
4904 } else if (slot == 0) {
4905 struct btrfs_disk_key disk_key;
4907 btrfs_node_key(parent, &disk_key, 0);
4908 fixup_low_keys(root->fs_info, path, &disk_key, level + 1);
4910 btrfs_mark_buffer_dirty(parent);
4914 * a helper function to delete the leaf pointed to by path->slots[1] and
4917 * This deletes the pointer in path->nodes[1] and frees the leaf
4918 * block extent. zero is returned if it all worked out, < 0 otherwise.
4920 * The path must have already been setup for deleting the leaf, including
4921 * all the proper balancing. path->nodes[1] must be locked.
4923 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4924 struct btrfs_root *root,
4925 struct btrfs_path *path,
4926 struct extent_buffer *leaf)
4928 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4929 del_ptr(root, path, 1, path->slots[1]);
4932 * btrfs_free_extent is expensive, we want to make sure we
4933 * aren't holding any locks when we call it
4935 btrfs_unlock_up_safe(path, 0);
4937 root_sub_used(root, leaf->len);
4939 extent_buffer_get(leaf);
4940 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4941 free_extent_buffer_stale(leaf);
4944 * delete the item at the leaf level in path. If that empties
4945 * the leaf, remove it from the tree
4947 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4948 struct btrfs_path *path, int slot, int nr)
4950 struct extent_buffer *leaf;
4951 struct btrfs_item *item;
4958 struct btrfs_map_token token;
4960 btrfs_init_map_token(&token);
4962 leaf = path->nodes[0];
4963 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4965 for (i = 0; i < nr; i++)
4966 dsize += btrfs_item_size_nr(leaf, slot + i);
4968 nritems = btrfs_header_nritems(leaf);
4970 if (slot + nr != nritems) {
4971 int data_end = leaf_data_end(root, leaf);
4973 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4975 btrfs_leaf_data(leaf) + data_end,
4976 last_off - data_end);
4978 for (i = slot + nr; i < nritems; i++) {
4981 item = btrfs_item_nr(i);
4982 ioff = btrfs_token_item_offset(leaf, item, &token);
4983 btrfs_set_token_item_offset(leaf, item,
4984 ioff + dsize, &token);
4987 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4988 btrfs_item_nr_offset(slot + nr),
4989 sizeof(struct btrfs_item) *
4990 (nritems - slot - nr));
4992 btrfs_set_header_nritems(leaf, nritems - nr);
4995 /* delete the leaf if we've emptied it */
4997 if (leaf == root->node) {
4998 btrfs_set_header_level(leaf, 0);
5000 btrfs_set_path_blocking(path);
5001 clean_tree_block(trans, root->fs_info, leaf);
5002 btrfs_del_leaf(trans, root, path, leaf);
5005 int used = leaf_space_used(leaf, 0, nritems);
5007 struct btrfs_disk_key disk_key;
5009 btrfs_item_key(leaf, &disk_key, 0);
5010 fixup_low_keys(root->fs_info, path, &disk_key, 1);
5013 /* delete the leaf if it is mostly empty */
5014 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
5015 /* push_leaf_left fixes the path.
5016 * make sure the path still points to our leaf
5017 * for possible call to del_ptr below
5019 slot = path->slots[1];
5020 extent_buffer_get(leaf);
5022 btrfs_set_path_blocking(path);
5023 wret = push_leaf_left(trans, root, path, 1, 1,
5025 if (wret < 0 && wret != -ENOSPC)
5028 if (path->nodes[0] == leaf &&
5029 btrfs_header_nritems(leaf)) {
5030 wret = push_leaf_right(trans, root, path, 1,
5032 if (wret < 0 && wret != -ENOSPC)
5036 if (btrfs_header_nritems(leaf) == 0) {
5037 path->slots[1] = slot;
5038 btrfs_del_leaf(trans, root, path, leaf);
5039 free_extent_buffer(leaf);
5042 /* if we're still in the path, make sure
5043 * we're dirty. Otherwise, one of the
5044 * push_leaf functions must have already
5045 * dirtied this buffer
5047 if (path->nodes[0] == leaf)
5048 btrfs_mark_buffer_dirty(leaf);
5049 free_extent_buffer(leaf);
5052 btrfs_mark_buffer_dirty(leaf);
5059 * search the tree again to find a leaf with lesser keys
5060 * returns 0 if it found something or 1 if there are no lesser leaves.
5061 * returns < 0 on io errors.
5063 * This may release the path, and so you may lose any locks held at the
5066 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5068 struct btrfs_key key;
5069 struct btrfs_disk_key found_key;
5072 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5074 if (key.offset > 0) {
5076 } else if (key.type > 0) {
5078 key.offset = (u64)-1;
5079 } else if (key.objectid > 0) {
5082 key.offset = (u64)-1;
5087 btrfs_release_path(path);
5088 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5091 btrfs_item_key(path->nodes[0], &found_key, 0);
5092 ret = comp_keys(&found_key, &key);
5094 * We might have had an item with the previous key in the tree right
5095 * before we released our path. And after we released our path, that
5096 * item might have been pushed to the first slot (0) of the leaf we
5097 * were holding due to a tree balance. Alternatively, an item with the
5098 * previous key can exist as the only element of a leaf (big fat item).
5099 * Therefore account for these 2 cases, so that our callers (like
5100 * btrfs_previous_item) don't miss an existing item with a key matching
5101 * the previous key we computed above.
5109 * A helper function to walk down the tree starting at min_key, and looking
5110 * for nodes or leaves that are have a minimum transaction id.
5111 * This is used by the btree defrag code, and tree logging
5113 * This does not cow, but it does stuff the starting key it finds back
5114 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5115 * key and get a writable path.
5117 * This does lock as it descends, and path->keep_locks should be set
5118 * to 1 by the caller.
5120 * This honors path->lowest_level to prevent descent past a given level
5123 * min_trans indicates the oldest transaction that you are interested
5124 * in walking through. Any nodes or leaves older than min_trans are
5125 * skipped over (without reading them).
5127 * returns zero if something useful was found, < 0 on error and 1 if there
5128 * was nothing in the tree that matched the search criteria.
5130 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5131 struct btrfs_path *path,
5134 struct extent_buffer *cur;
5135 struct btrfs_key found_key;
5141 int keep_locks = path->keep_locks;
5143 path->keep_locks = 1;
5145 cur = btrfs_read_lock_root_node(root);
5146 level = btrfs_header_level(cur);
5147 WARN_ON(path->nodes[level]);
5148 path->nodes[level] = cur;
5149 path->locks[level] = BTRFS_READ_LOCK;
5151 if (btrfs_header_generation(cur) < min_trans) {
5156 nritems = btrfs_header_nritems(cur);
5157 level = btrfs_header_level(cur);
5158 sret = bin_search(cur, min_key, level, &slot);
5160 /* at the lowest level, we're done, setup the path and exit */
5161 if (level == path->lowest_level) {
5162 if (slot >= nritems)
5165 path->slots[level] = slot;
5166 btrfs_item_key_to_cpu(cur, &found_key, slot);
5169 if (sret && slot > 0)
5172 * check this node pointer against the min_trans parameters.
5173 * If it is too old, old, skip to the next one.
5175 while (slot < nritems) {
5178 gen = btrfs_node_ptr_generation(cur, slot);
5179 if (gen < min_trans) {
5187 * we didn't find a candidate key in this node, walk forward
5188 * and find another one
5190 if (slot >= nritems) {
5191 path->slots[level] = slot;
5192 btrfs_set_path_blocking(path);
5193 sret = btrfs_find_next_key(root, path, min_key, level,
5196 btrfs_release_path(path);
5202 /* save our key for returning back */
5203 btrfs_node_key_to_cpu(cur, &found_key, slot);
5204 path->slots[level] = slot;
5205 if (level == path->lowest_level) {
5209 btrfs_set_path_blocking(path);
5210 cur = read_node_slot(root, cur, slot);
5211 BUG_ON(!cur); /* -ENOMEM */
5213 btrfs_tree_read_lock(cur);
5215 path->locks[level - 1] = BTRFS_READ_LOCK;
5216 path->nodes[level - 1] = cur;
5217 unlock_up(path, level, 1, 0, NULL);
5218 btrfs_clear_path_blocking(path, NULL, 0);
5221 path->keep_locks = keep_locks;
5223 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5224 btrfs_set_path_blocking(path);
5225 memcpy(min_key, &found_key, sizeof(found_key));
5230 static void tree_move_down(struct btrfs_root *root,
5231 struct btrfs_path *path,
5232 int *level, int root_level)
5234 BUG_ON(*level == 0);
5235 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
5236 path->slots[*level]);
5237 path->slots[*level - 1] = 0;
5241 static int tree_move_next_or_upnext(struct btrfs_root *root,
5242 struct btrfs_path *path,
5243 int *level, int root_level)
5247 nritems = btrfs_header_nritems(path->nodes[*level]);
5249 path->slots[*level]++;
5251 while (path->slots[*level] >= nritems) {
5252 if (*level == root_level)
5256 path->slots[*level] = 0;
5257 free_extent_buffer(path->nodes[*level]);
5258 path->nodes[*level] = NULL;
5260 path->slots[*level]++;
5262 nritems = btrfs_header_nritems(path->nodes[*level]);
5269 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5272 static int tree_advance(struct btrfs_root *root,
5273 struct btrfs_path *path,
5274 int *level, int root_level,
5276 struct btrfs_key *key)
5280 if (*level == 0 || !allow_down) {
5281 ret = tree_move_next_or_upnext(root, path, level, root_level);
5283 tree_move_down(root, path, level, root_level);
5288 btrfs_item_key_to_cpu(path->nodes[*level], key,
5289 path->slots[*level]);
5291 btrfs_node_key_to_cpu(path->nodes[*level], key,
5292 path->slots[*level]);
5297 static int tree_compare_item(struct btrfs_root *left_root,
5298 struct btrfs_path *left_path,
5299 struct btrfs_path *right_path,
5304 unsigned long off1, off2;
5306 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5307 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5311 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5312 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5313 right_path->slots[0]);
5315 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5317 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5324 #define ADVANCE_ONLY_NEXT -1
5327 * This function compares two trees and calls the provided callback for
5328 * every changed/new/deleted item it finds.
5329 * If shared tree blocks are encountered, whole subtrees are skipped, making
5330 * the compare pretty fast on snapshotted subvolumes.
5332 * This currently works on commit roots only. As commit roots are read only,
5333 * we don't do any locking. The commit roots are protected with transactions.
5334 * Transactions are ended and rejoined when a commit is tried in between.
5336 * This function checks for modifications done to the trees while comparing.
5337 * If it detects a change, it aborts immediately.
5339 int btrfs_compare_trees(struct btrfs_root *left_root,
5340 struct btrfs_root *right_root,
5341 btrfs_changed_cb_t changed_cb, void *ctx)
5345 struct btrfs_path *left_path = NULL;
5346 struct btrfs_path *right_path = NULL;
5347 struct btrfs_key left_key;
5348 struct btrfs_key right_key;
5349 char *tmp_buf = NULL;
5350 int left_root_level;
5351 int right_root_level;
5354 int left_end_reached;
5355 int right_end_reached;
5363 left_path = btrfs_alloc_path();
5368 right_path = btrfs_alloc_path();
5374 tmp_buf = kmalloc(left_root->nodesize, GFP_NOFS);
5380 left_path->search_commit_root = 1;
5381 left_path->skip_locking = 1;
5382 right_path->search_commit_root = 1;
5383 right_path->skip_locking = 1;
5386 * Strategy: Go to the first items of both trees. Then do
5388 * If both trees are at level 0
5389 * Compare keys of current items
5390 * If left < right treat left item as new, advance left tree
5392 * If left > right treat right item as deleted, advance right tree
5394 * If left == right do deep compare of items, treat as changed if
5395 * needed, advance both trees and repeat
5396 * If both trees are at the same level but not at level 0
5397 * Compare keys of current nodes/leafs
5398 * If left < right advance left tree and repeat
5399 * If left > right advance right tree and repeat
5400 * If left == right compare blockptrs of the next nodes/leafs
5401 * If they match advance both trees but stay at the same level
5403 * If they don't match advance both trees while allowing to go
5405 * If tree levels are different
5406 * Advance the tree that needs it and repeat
5408 * Advancing a tree means:
5409 * If we are at level 0, try to go to the next slot. If that's not
5410 * possible, go one level up and repeat. Stop when we found a level
5411 * where we could go to the next slot. We may at this point be on a
5414 * If we are not at level 0 and not on shared tree blocks, go one
5417 * If we are not at level 0 and on shared tree blocks, go one slot to
5418 * the right if possible or go up and right.
5421 down_read(&left_root->fs_info->commit_root_sem);
5422 left_level = btrfs_header_level(left_root->commit_root);
5423 left_root_level = left_level;
5424 left_path->nodes[left_level] = left_root->commit_root;
5425 extent_buffer_get(left_path->nodes[left_level]);
5427 right_level = btrfs_header_level(right_root->commit_root);
5428 right_root_level = right_level;
5429 right_path->nodes[right_level] = right_root->commit_root;
5430 extent_buffer_get(right_path->nodes[right_level]);
5431 up_read(&left_root->fs_info->commit_root_sem);
5433 if (left_level == 0)
5434 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5435 &left_key, left_path->slots[left_level]);
5437 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5438 &left_key, left_path->slots[left_level]);
5439 if (right_level == 0)
5440 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5441 &right_key, right_path->slots[right_level]);
5443 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5444 &right_key, right_path->slots[right_level]);
5446 left_end_reached = right_end_reached = 0;
5447 advance_left = advance_right = 0;
5450 if (advance_left && !left_end_reached) {
5451 ret = tree_advance(left_root, left_path, &left_level,
5453 advance_left != ADVANCE_ONLY_NEXT,
5456 left_end_reached = ADVANCE;
5459 if (advance_right && !right_end_reached) {
5460 ret = tree_advance(right_root, right_path, &right_level,
5462 advance_right != ADVANCE_ONLY_NEXT,
5465 right_end_reached = ADVANCE;
5469 if (left_end_reached && right_end_reached) {
5472 } else if (left_end_reached) {
5473 if (right_level == 0) {
5474 ret = changed_cb(left_root, right_root,
5475 left_path, right_path,
5477 BTRFS_COMPARE_TREE_DELETED,
5482 advance_right = ADVANCE;
5484 } else if (right_end_reached) {
5485 if (left_level == 0) {
5486 ret = changed_cb(left_root, right_root,
5487 left_path, right_path,
5489 BTRFS_COMPARE_TREE_NEW,
5494 advance_left = ADVANCE;
5498 if (left_level == 0 && right_level == 0) {
5499 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5501 ret = changed_cb(left_root, right_root,
5502 left_path, right_path,
5504 BTRFS_COMPARE_TREE_NEW,
5508 advance_left = ADVANCE;
5509 } else if (cmp > 0) {
5510 ret = changed_cb(left_root, right_root,
5511 left_path, right_path,
5513 BTRFS_COMPARE_TREE_DELETED,
5517 advance_right = ADVANCE;
5519 enum btrfs_compare_tree_result result;
5521 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5522 ret = tree_compare_item(left_root, left_path,
5523 right_path, tmp_buf);
5525 result = BTRFS_COMPARE_TREE_CHANGED;
5527 result = BTRFS_COMPARE_TREE_SAME;
5528 ret = changed_cb(left_root, right_root,
5529 left_path, right_path,
5530 &left_key, result, ctx);
5533 advance_left = ADVANCE;
5534 advance_right = ADVANCE;
5536 } else if (left_level == right_level) {
5537 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5539 advance_left = ADVANCE;
5540 } else if (cmp > 0) {
5541 advance_right = ADVANCE;
5543 left_blockptr = btrfs_node_blockptr(
5544 left_path->nodes[left_level],
5545 left_path->slots[left_level]);
5546 right_blockptr = btrfs_node_blockptr(
5547 right_path->nodes[right_level],
5548 right_path->slots[right_level]);
5549 left_gen = btrfs_node_ptr_generation(
5550 left_path->nodes[left_level],
5551 left_path->slots[left_level]);
5552 right_gen = btrfs_node_ptr_generation(
5553 right_path->nodes[right_level],
5554 right_path->slots[right_level]);
5555 if (left_blockptr == right_blockptr &&
5556 left_gen == right_gen) {
5558 * As we're on a shared block, don't
5559 * allow to go deeper.
5561 advance_left = ADVANCE_ONLY_NEXT;
5562 advance_right = ADVANCE_ONLY_NEXT;
5564 advance_left = ADVANCE;
5565 advance_right = ADVANCE;
5568 } else if (left_level < right_level) {
5569 advance_right = ADVANCE;
5571 advance_left = ADVANCE;
5576 btrfs_free_path(left_path);
5577 btrfs_free_path(right_path);
5583 * this is similar to btrfs_next_leaf, but does not try to preserve
5584 * and fixup the path. It looks for and returns the next key in the
5585 * tree based on the current path and the min_trans parameters.
5587 * 0 is returned if another key is found, < 0 if there are any errors
5588 * and 1 is returned if there are no higher keys in the tree
5590 * path->keep_locks should be set to 1 on the search made before
5591 * calling this function.
5593 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5594 struct btrfs_key *key, int level, u64 min_trans)
5597 struct extent_buffer *c;
5599 WARN_ON(!path->keep_locks);
5600 while (level < BTRFS_MAX_LEVEL) {
5601 if (!path->nodes[level])
5604 slot = path->slots[level] + 1;
5605 c = path->nodes[level];
5607 if (slot >= btrfs_header_nritems(c)) {
5610 struct btrfs_key cur_key;
5611 if (level + 1 >= BTRFS_MAX_LEVEL ||
5612 !path->nodes[level + 1])
5615 if (path->locks[level + 1]) {
5620 slot = btrfs_header_nritems(c) - 1;
5622 btrfs_item_key_to_cpu(c, &cur_key, slot);
5624 btrfs_node_key_to_cpu(c, &cur_key, slot);
5626 orig_lowest = path->lowest_level;
5627 btrfs_release_path(path);
5628 path->lowest_level = level;
5629 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5631 path->lowest_level = orig_lowest;
5635 c = path->nodes[level];
5636 slot = path->slots[level];
5643 btrfs_item_key_to_cpu(c, key, slot);
5645 u64 gen = btrfs_node_ptr_generation(c, slot);
5647 if (gen < min_trans) {
5651 btrfs_node_key_to_cpu(c, key, slot);
5659 * search the tree again to find a leaf with greater keys
5660 * returns 0 if it found something or 1 if there are no greater leaves.
5661 * returns < 0 on io errors.
5663 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5665 return btrfs_next_old_leaf(root, path, 0);
5668 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5673 struct extent_buffer *c;
5674 struct extent_buffer *next;
5675 struct btrfs_key key;
5678 int old_spinning = path->leave_spinning;
5679 int next_rw_lock = 0;
5681 nritems = btrfs_header_nritems(path->nodes[0]);
5685 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5690 btrfs_release_path(path);
5692 path->keep_locks = 1;
5693 path->leave_spinning = 1;
5696 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5698 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5699 path->keep_locks = 0;
5704 nritems = btrfs_header_nritems(path->nodes[0]);
5706 * by releasing the path above we dropped all our locks. A balance
5707 * could have added more items next to the key that used to be
5708 * at the very end of the block. So, check again here and
5709 * advance the path if there are now more items available.
5711 if (nritems > 0 && path->slots[0] < nritems - 1) {
5718 * So the above check misses one case:
5719 * - after releasing the path above, someone has removed the item that
5720 * used to be at the very end of the block, and balance between leafs
5721 * gets another one with bigger key.offset to replace it.
5723 * This one should be returned as well, or we can get leaf corruption
5724 * later(esp. in __btrfs_drop_extents()).
5726 * And a bit more explanation about this check,
5727 * with ret > 0, the key isn't found, the path points to the slot
5728 * where it should be inserted, so the path->slots[0] item must be the
5731 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5736 while (level < BTRFS_MAX_LEVEL) {
5737 if (!path->nodes[level]) {
5742 slot = path->slots[level] + 1;
5743 c = path->nodes[level];
5744 if (slot >= btrfs_header_nritems(c)) {
5746 if (level == BTRFS_MAX_LEVEL) {
5754 btrfs_tree_unlock_rw(next, next_rw_lock);
5755 free_extent_buffer(next);
5759 next_rw_lock = path->locks[level];
5760 ret = read_block_for_search(NULL, root, path, &next, level,
5766 btrfs_release_path(path);
5770 if (!path->skip_locking) {
5771 ret = btrfs_try_tree_read_lock(next);
5772 if (!ret && time_seq) {
5774 * If we don't get the lock, we may be racing
5775 * with push_leaf_left, holding that lock while
5776 * itself waiting for the leaf we've currently
5777 * locked. To solve this situation, we give up
5778 * on our lock and cycle.
5780 free_extent_buffer(next);
5781 btrfs_release_path(path);
5786 btrfs_set_path_blocking(path);
5787 btrfs_tree_read_lock(next);
5788 btrfs_clear_path_blocking(path, next,
5791 next_rw_lock = BTRFS_READ_LOCK;
5795 path->slots[level] = slot;
5798 c = path->nodes[level];
5799 if (path->locks[level])
5800 btrfs_tree_unlock_rw(c, path->locks[level]);
5802 free_extent_buffer(c);
5803 path->nodes[level] = next;
5804 path->slots[level] = 0;
5805 if (!path->skip_locking)
5806 path->locks[level] = next_rw_lock;
5810 ret = read_block_for_search(NULL, root, path, &next, level,
5816 btrfs_release_path(path);
5820 if (!path->skip_locking) {
5821 ret = btrfs_try_tree_read_lock(next);
5823 btrfs_set_path_blocking(path);
5824 btrfs_tree_read_lock(next);
5825 btrfs_clear_path_blocking(path, next,
5828 next_rw_lock = BTRFS_READ_LOCK;
5833 unlock_up(path, 0, 1, 0, NULL);
5834 path->leave_spinning = old_spinning;
5836 btrfs_set_path_blocking(path);
5842 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5843 * searching until it gets past min_objectid or finds an item of 'type'
5845 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5847 int btrfs_previous_item(struct btrfs_root *root,
5848 struct btrfs_path *path, u64 min_objectid,
5851 struct btrfs_key found_key;
5852 struct extent_buffer *leaf;
5857 if (path->slots[0] == 0) {
5858 btrfs_set_path_blocking(path);
5859 ret = btrfs_prev_leaf(root, path);
5865 leaf = path->nodes[0];
5866 nritems = btrfs_header_nritems(leaf);
5869 if (path->slots[0] == nritems)
5872 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5873 if (found_key.objectid < min_objectid)
5875 if (found_key.type == type)
5877 if (found_key.objectid == min_objectid &&
5878 found_key.type < type)
5885 * search in extent tree to find a previous Metadata/Data extent item with
5888 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5890 int btrfs_previous_extent_item(struct btrfs_root *root,
5891 struct btrfs_path *path, u64 min_objectid)
5893 struct btrfs_key found_key;
5894 struct extent_buffer *leaf;
5899 if (path->slots[0] == 0) {
5900 btrfs_set_path_blocking(path);
5901 ret = btrfs_prev_leaf(root, path);
5907 leaf = path->nodes[0];
5908 nritems = btrfs_header_nritems(leaf);
5911 if (path->slots[0] == nritems)
5914 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5915 if (found_key.objectid < min_objectid)
5917 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5918 found_key.type == BTRFS_METADATA_ITEM_KEY)
5920 if (found_key.objectid == min_objectid &&
5921 found_key.type < BTRFS_EXTENT_ITEM_KEY)