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 = kzalloc(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 = kzalloc(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 = kzalloc(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 = kzalloc(nritems * sizeof(struct tree_mod_elem *),
913 for (i = 0; i < nritems; i++) {
914 tm_list[i] = alloc_tree_mod_elem(eb, i,
915 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
922 if (tree_mod_dont_log(fs_info, eb))
925 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
926 tree_mod_log_write_unlock(fs_info);
934 for (i = 0; i < nritems; i++)
942 tree_mod_log_set_root_pointer(struct btrfs_root *root,
943 struct extent_buffer *new_root_node,
947 ret = tree_mod_log_insert_root(root->fs_info, root->node,
948 new_root_node, GFP_NOFS, log_removal);
953 * check if the tree block can be shared by multiple trees
955 int btrfs_block_can_be_shared(struct btrfs_root *root,
956 struct extent_buffer *buf)
959 * Tree blocks not in refernece counted trees and tree roots
960 * are never shared. If a block was allocated after the last
961 * snapshot and the block was not allocated by tree relocation,
962 * we know the block is not shared.
964 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
965 buf != root->node && buf != root->commit_root &&
966 (btrfs_header_generation(buf) <=
967 btrfs_root_last_snapshot(&root->root_item) ||
968 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
970 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
971 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
972 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
978 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
979 struct btrfs_root *root,
980 struct extent_buffer *buf,
981 struct extent_buffer *cow,
991 * Backrefs update rules:
993 * Always use full backrefs for extent pointers in tree block
994 * allocated by tree relocation.
996 * If a shared tree block is no longer referenced by its owner
997 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
998 * use full backrefs for extent pointers in tree block.
1000 * If a tree block is been relocating
1001 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1002 * use full backrefs for extent pointers in tree block.
1003 * The reason for this is some operations (such as drop tree)
1004 * are only allowed for blocks use full backrefs.
1007 if (btrfs_block_can_be_shared(root, buf)) {
1008 ret = btrfs_lookup_extent_info(trans, root, buf->start,
1009 btrfs_header_level(buf), 1,
1015 btrfs_std_error(root->fs_info, ret);
1020 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1021 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1022 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1027 owner = btrfs_header_owner(buf);
1028 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
1029 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
1032 if ((owner == root->root_key.objectid ||
1033 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
1034 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1035 ret = btrfs_inc_ref(trans, root, buf, 1);
1036 BUG_ON(ret); /* -ENOMEM */
1038 if (root->root_key.objectid ==
1039 BTRFS_TREE_RELOC_OBJECTID) {
1040 ret = btrfs_dec_ref(trans, root, buf, 0);
1041 BUG_ON(ret); /* -ENOMEM */
1042 ret = btrfs_inc_ref(trans, root, cow, 1);
1043 BUG_ON(ret); /* -ENOMEM */
1045 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
1048 if (root->root_key.objectid ==
1049 BTRFS_TREE_RELOC_OBJECTID)
1050 ret = btrfs_inc_ref(trans, root, cow, 1);
1052 ret = btrfs_inc_ref(trans, root, cow, 0);
1053 BUG_ON(ret); /* -ENOMEM */
1055 if (new_flags != 0) {
1056 int level = btrfs_header_level(buf);
1058 ret = btrfs_set_disk_extent_flags(trans, root,
1061 new_flags, level, 0);
1066 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
1067 if (root->root_key.objectid ==
1068 BTRFS_TREE_RELOC_OBJECTID)
1069 ret = btrfs_inc_ref(trans, root, cow, 1);
1071 ret = btrfs_inc_ref(trans, root, cow, 0);
1072 BUG_ON(ret); /* -ENOMEM */
1073 ret = btrfs_dec_ref(trans, root, buf, 1);
1074 BUG_ON(ret); /* -ENOMEM */
1076 clean_tree_block(trans, root, buf);
1083 * does the dirty work in cow of a single block. The parent block (if
1084 * supplied) is updated to point to the new cow copy. The new buffer is marked
1085 * dirty and returned locked. If you modify the block it needs to be marked
1088 * search_start -- an allocation hint for the new block
1090 * empty_size -- a hint that you plan on doing more cow. This is the size in
1091 * bytes the allocator should try to find free next to the block it returns.
1092 * This is just a hint and may be ignored by the allocator.
1094 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1095 struct btrfs_root *root,
1096 struct extent_buffer *buf,
1097 struct extent_buffer *parent, int parent_slot,
1098 struct extent_buffer **cow_ret,
1099 u64 search_start, u64 empty_size)
1101 struct btrfs_disk_key disk_key;
1102 struct extent_buffer *cow;
1105 int unlock_orig = 0;
1108 if (*cow_ret == buf)
1111 btrfs_assert_tree_locked(buf);
1113 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1114 trans->transid != root->fs_info->running_transaction->transid);
1115 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1116 trans->transid != root->last_trans);
1118 level = btrfs_header_level(buf);
1121 btrfs_item_key(buf, &disk_key, 0);
1123 btrfs_node_key(buf, &disk_key, 0);
1125 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1127 parent_start = parent->start;
1133 cow = btrfs_alloc_tree_block(trans, root, parent_start,
1134 root->root_key.objectid, &disk_key, level,
1135 search_start, empty_size);
1137 return PTR_ERR(cow);
1139 /* cow is set to blocking by btrfs_init_new_buffer */
1141 copy_extent_buffer(cow, buf, 0, 0, cow->len);
1142 btrfs_set_header_bytenr(cow, cow->start);
1143 btrfs_set_header_generation(cow, trans->transid);
1144 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1145 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1146 BTRFS_HEADER_FLAG_RELOC);
1147 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1148 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1150 btrfs_set_header_owner(cow, root->root_key.objectid);
1152 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
1155 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1157 btrfs_abort_transaction(trans, root, ret);
1161 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1162 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1167 if (buf == root->node) {
1168 WARN_ON(parent && parent != buf);
1169 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1170 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1171 parent_start = buf->start;
1175 extent_buffer_get(cow);
1176 tree_mod_log_set_root_pointer(root, cow, 1);
1177 rcu_assign_pointer(root->node, cow);
1179 btrfs_free_tree_block(trans, root, buf, parent_start,
1181 free_extent_buffer(buf);
1182 add_root_to_dirty_list(root);
1184 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1185 parent_start = parent->start;
1189 WARN_ON(trans->transid != btrfs_header_generation(parent));
1190 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1191 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1192 btrfs_set_node_blockptr(parent, parent_slot,
1194 btrfs_set_node_ptr_generation(parent, parent_slot,
1196 btrfs_mark_buffer_dirty(parent);
1198 ret = tree_mod_log_free_eb(root->fs_info, buf);
1200 btrfs_abort_transaction(trans, root, ret);
1204 btrfs_free_tree_block(trans, root, buf, parent_start,
1208 btrfs_tree_unlock(buf);
1209 free_extent_buffer_stale(buf);
1210 btrfs_mark_buffer_dirty(cow);
1216 * returns the logical address of the oldest predecessor of the given root.
1217 * entries older than time_seq are ignored.
1219 static struct tree_mod_elem *
1220 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1221 struct extent_buffer *eb_root, u64 time_seq)
1223 struct tree_mod_elem *tm;
1224 struct tree_mod_elem *found = NULL;
1225 u64 root_logical = eb_root->start;
1232 * the very last operation that's logged for a root is the replacement
1233 * operation (if it is replaced at all). this has the index of the *new*
1234 * root, making it the very first operation that's logged for this root.
1237 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1242 * if there are no tree operation for the oldest root, we simply
1243 * return it. this should only happen if that (old) root is at
1250 * if there's an operation that's not a root replacement, we
1251 * found the oldest version of our root. normally, we'll find a
1252 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1254 if (tm->op != MOD_LOG_ROOT_REPLACE)
1258 root_logical = tm->old_root.logical;
1262 /* if there's no old root to return, return what we found instead */
1270 * tm is a pointer to the first operation to rewind within eb. then, all
1271 * previous operations will be rewinded (until we reach something older than
1275 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1276 u64 time_seq, struct tree_mod_elem *first_tm)
1279 struct rb_node *next;
1280 struct tree_mod_elem *tm = first_tm;
1281 unsigned long o_dst;
1282 unsigned long o_src;
1283 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1285 n = btrfs_header_nritems(eb);
1286 tree_mod_log_read_lock(fs_info);
1287 while (tm && tm->seq >= time_seq) {
1289 * all the operations are recorded with the operator used for
1290 * the modification. as we're going backwards, we do the
1291 * opposite of each operation here.
1294 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1295 BUG_ON(tm->slot < n);
1297 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1298 case MOD_LOG_KEY_REMOVE:
1299 btrfs_set_node_key(eb, &tm->key, tm->slot);
1300 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1301 btrfs_set_node_ptr_generation(eb, tm->slot,
1305 case MOD_LOG_KEY_REPLACE:
1306 BUG_ON(tm->slot >= n);
1307 btrfs_set_node_key(eb, &tm->key, tm->slot);
1308 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1309 btrfs_set_node_ptr_generation(eb, tm->slot,
1312 case MOD_LOG_KEY_ADD:
1313 /* if a move operation is needed it's in the log */
1316 case MOD_LOG_MOVE_KEYS:
1317 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1318 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1319 memmove_extent_buffer(eb, o_dst, o_src,
1320 tm->move.nr_items * p_size);
1322 case MOD_LOG_ROOT_REPLACE:
1324 * this operation is special. for roots, this must be
1325 * handled explicitly before rewinding.
1326 * for non-roots, this operation may exist if the node
1327 * was a root: root A -> child B; then A gets empty and
1328 * B is promoted to the new root. in the mod log, we'll
1329 * have a root-replace operation for B, a tree block
1330 * that is no root. we simply ignore that operation.
1334 next = rb_next(&tm->node);
1337 tm = container_of(next, struct tree_mod_elem, node);
1338 if (tm->index != first_tm->index)
1341 tree_mod_log_read_unlock(fs_info);
1342 btrfs_set_header_nritems(eb, n);
1346 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1347 * is returned. If rewind operations happen, a fresh buffer is returned. The
1348 * returned buffer is always read-locked. If the returned buffer is not the
1349 * input buffer, the lock on the input buffer is released and the input buffer
1350 * is freed (its refcount is decremented).
1352 static struct extent_buffer *
1353 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1354 struct extent_buffer *eb, u64 time_seq)
1356 struct extent_buffer *eb_rewin;
1357 struct tree_mod_elem *tm;
1362 if (btrfs_header_level(eb) == 0)
1365 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1369 btrfs_set_path_blocking(path);
1370 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1372 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1373 BUG_ON(tm->slot != 0);
1374 eb_rewin = alloc_dummy_extent_buffer(eb->start,
1375 fs_info->tree_root->nodesize);
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(!old || !extent_buffer_uptodate(old))) {
1445 free_extent_buffer(old);
1446 btrfs_warn(root->fs_info,
1447 "failed to read tree block %llu from get_old_root", logical);
1449 eb = btrfs_clone_extent_buffer(old);
1450 free_extent_buffer(old);
1452 } else if (old_root) {
1453 btrfs_tree_read_unlock(eb_root);
1454 free_extent_buffer(eb_root);
1455 eb = alloc_dummy_extent_buffer(logical, root->nodesize);
1457 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1458 eb = btrfs_clone_extent_buffer(eb_root);
1459 btrfs_tree_read_unlock_blocking(eb_root);
1460 free_extent_buffer(eb_root);
1465 extent_buffer_get(eb);
1466 btrfs_tree_read_lock(eb);
1468 btrfs_set_header_bytenr(eb, eb->start);
1469 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1470 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1471 btrfs_set_header_level(eb, old_root->level);
1472 btrfs_set_header_generation(eb, old_generation);
1475 __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1477 WARN_ON(btrfs_header_level(eb) != 0);
1478 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1483 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1485 struct tree_mod_elem *tm;
1487 struct extent_buffer *eb_root = btrfs_root_node(root);
1489 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1490 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1491 level = tm->old_root.level;
1493 level = btrfs_header_level(eb_root);
1495 free_extent_buffer(eb_root);
1500 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1501 struct btrfs_root *root,
1502 struct extent_buffer *buf)
1504 if (btrfs_test_is_dummy_root(root))
1507 /* ensure we can see the force_cow */
1511 * We do not need to cow a block if
1512 * 1) this block is not created or changed in this transaction;
1513 * 2) this block does not belong to TREE_RELOC tree;
1514 * 3) the root is not forced COW.
1516 * What is forced COW:
1517 * when we create snapshot during commiting the transaction,
1518 * after we've finished coping src root, we must COW the shared
1519 * block to ensure the metadata consistency.
1521 if (btrfs_header_generation(buf) == trans->transid &&
1522 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1523 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1524 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1525 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1531 * cows a single block, see __btrfs_cow_block for the real work.
1532 * This version of it has extra checks so that a block isn't cow'd more than
1533 * once per transaction, as long as it hasn't been written yet
1535 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1536 struct btrfs_root *root, struct extent_buffer *buf,
1537 struct extent_buffer *parent, int parent_slot,
1538 struct extent_buffer **cow_ret)
1543 if (trans->transaction != root->fs_info->running_transaction)
1544 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1546 root->fs_info->running_transaction->transid);
1548 if (trans->transid != root->fs_info->generation)
1549 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1550 trans->transid, root->fs_info->generation);
1552 if (!should_cow_block(trans, root, buf)) {
1557 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
1560 btrfs_set_lock_blocking(parent);
1561 btrfs_set_lock_blocking(buf);
1563 ret = __btrfs_cow_block(trans, root, buf, parent,
1564 parent_slot, cow_ret, search_start, 0);
1566 trace_btrfs_cow_block(root, buf, *cow_ret);
1572 * helper function for defrag to decide if two blocks pointed to by a
1573 * node are actually close by
1575 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1577 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1579 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1585 * compare two keys in a memcmp fashion
1587 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1589 struct btrfs_key k1;
1591 btrfs_disk_key_to_cpu(&k1, disk);
1593 return btrfs_comp_cpu_keys(&k1, k2);
1597 * same as comp_keys only with two btrfs_key's
1599 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1601 if (k1->objectid > k2->objectid)
1603 if (k1->objectid < k2->objectid)
1605 if (k1->type > k2->type)
1607 if (k1->type < k2->type)
1609 if (k1->offset > k2->offset)
1611 if (k1->offset < k2->offset)
1617 * this is used by the defrag code to go through all the
1618 * leaves pointed to by a node and reallocate them so that
1619 * disk order is close to key order
1621 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1622 struct btrfs_root *root, struct extent_buffer *parent,
1623 int start_slot, u64 *last_ret,
1624 struct btrfs_key *progress)
1626 struct extent_buffer *cur;
1629 u64 search_start = *last_ret;
1639 int progress_passed = 0;
1640 struct btrfs_disk_key disk_key;
1642 parent_level = btrfs_header_level(parent);
1644 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1645 WARN_ON(trans->transid != root->fs_info->generation);
1647 parent_nritems = btrfs_header_nritems(parent);
1648 blocksize = root->nodesize;
1649 end_slot = parent_nritems;
1651 if (parent_nritems == 1)
1654 btrfs_set_lock_blocking(parent);
1656 for (i = start_slot; i < end_slot; i++) {
1659 btrfs_node_key(parent, &disk_key, i);
1660 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1663 progress_passed = 1;
1664 blocknr = btrfs_node_blockptr(parent, i);
1665 gen = btrfs_node_ptr_generation(parent, i);
1666 if (last_block == 0)
1667 last_block = blocknr;
1670 other = btrfs_node_blockptr(parent, i - 1);
1671 close = close_blocks(blocknr, other, blocksize);
1673 if (!close && i < end_slot - 2) {
1674 other = btrfs_node_blockptr(parent, i + 1);
1675 close = close_blocks(blocknr, other, blocksize);
1678 last_block = blocknr;
1682 cur = btrfs_find_tree_block(root, blocknr);
1684 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1687 if (!cur || !uptodate) {
1689 cur = read_tree_block(root, blocknr, gen);
1690 if (!cur || !extent_buffer_uptodate(cur)) {
1691 free_extent_buffer(cur);
1694 } else if (!uptodate) {
1695 err = btrfs_read_buffer(cur, gen);
1697 free_extent_buffer(cur);
1702 if (search_start == 0)
1703 search_start = last_block;
1705 btrfs_tree_lock(cur);
1706 btrfs_set_lock_blocking(cur);
1707 err = __btrfs_cow_block(trans, root, cur, parent, i,
1710 (end_slot - i) * blocksize));
1712 btrfs_tree_unlock(cur);
1713 free_extent_buffer(cur);
1716 search_start = cur->start;
1717 last_block = cur->start;
1718 *last_ret = search_start;
1719 btrfs_tree_unlock(cur);
1720 free_extent_buffer(cur);
1726 * The leaf data grows from end-to-front in the node.
1727 * this returns the address of the start of the last item,
1728 * which is the stop of the leaf data stack
1730 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1731 struct extent_buffer *leaf)
1733 u32 nr = btrfs_header_nritems(leaf);
1735 return BTRFS_LEAF_DATA_SIZE(root);
1736 return btrfs_item_offset_nr(leaf, nr - 1);
1741 * search for key in the extent_buffer. The items start at offset p,
1742 * and they are item_size apart. There are 'max' items in p.
1744 * the slot in the array is returned via slot, and it points to
1745 * the place where you would insert key if it is not found in
1748 * slot may point to max if the key is bigger than all of the keys
1750 static noinline int generic_bin_search(struct extent_buffer *eb,
1752 int item_size, struct btrfs_key *key,
1759 struct btrfs_disk_key *tmp = NULL;
1760 struct btrfs_disk_key unaligned;
1761 unsigned long offset;
1763 unsigned long map_start = 0;
1764 unsigned long map_len = 0;
1767 while (low < high) {
1768 mid = (low + high) / 2;
1769 offset = p + mid * item_size;
1771 if (!kaddr || offset < map_start ||
1772 (offset + sizeof(struct btrfs_disk_key)) >
1773 map_start + map_len) {
1775 err = map_private_extent_buffer(eb, offset,
1776 sizeof(struct btrfs_disk_key),
1777 &kaddr, &map_start, &map_len);
1780 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1783 read_extent_buffer(eb, &unaligned,
1784 offset, sizeof(unaligned));
1789 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1792 ret = comp_keys(tmp, key);
1808 * simple bin_search frontend that does the right thing for
1811 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1812 int level, int *slot)
1815 return generic_bin_search(eb,
1816 offsetof(struct btrfs_leaf, items),
1817 sizeof(struct btrfs_item),
1818 key, btrfs_header_nritems(eb),
1821 return generic_bin_search(eb,
1822 offsetof(struct btrfs_node, ptrs),
1823 sizeof(struct btrfs_key_ptr),
1824 key, btrfs_header_nritems(eb),
1828 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1829 int level, int *slot)
1831 return bin_search(eb, key, level, slot);
1834 static void root_add_used(struct btrfs_root *root, u32 size)
1836 spin_lock(&root->accounting_lock);
1837 btrfs_set_root_used(&root->root_item,
1838 btrfs_root_used(&root->root_item) + size);
1839 spin_unlock(&root->accounting_lock);
1842 static void root_sub_used(struct btrfs_root *root, u32 size)
1844 spin_lock(&root->accounting_lock);
1845 btrfs_set_root_used(&root->root_item,
1846 btrfs_root_used(&root->root_item) - size);
1847 spin_unlock(&root->accounting_lock);
1850 /* given a node and slot number, this reads the blocks it points to. The
1851 * extent buffer is returned with a reference taken (but unlocked).
1852 * NULL is returned on error.
1854 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1855 struct extent_buffer *parent, int slot)
1857 int level = btrfs_header_level(parent);
1858 struct extent_buffer *eb;
1862 if (slot >= btrfs_header_nritems(parent))
1867 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1868 btrfs_node_ptr_generation(parent, slot));
1869 if (eb && !extent_buffer_uptodate(eb)) {
1870 free_extent_buffer(eb);
1878 * node level balancing, used to make sure nodes are in proper order for
1879 * item deletion. We balance from the top down, so we have to make sure
1880 * that a deletion won't leave an node completely empty later on.
1882 static noinline int balance_level(struct btrfs_trans_handle *trans,
1883 struct btrfs_root *root,
1884 struct btrfs_path *path, int level)
1886 struct extent_buffer *right = NULL;
1887 struct extent_buffer *mid;
1888 struct extent_buffer *left = NULL;
1889 struct extent_buffer *parent = NULL;
1893 int orig_slot = path->slots[level];
1899 mid = path->nodes[level];
1901 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1902 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1903 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1905 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1907 if (level < BTRFS_MAX_LEVEL - 1) {
1908 parent = path->nodes[level + 1];
1909 pslot = path->slots[level + 1];
1913 * deal with the case where there is only one pointer in the root
1914 * by promoting the node below to a root
1917 struct extent_buffer *child;
1919 if (btrfs_header_nritems(mid) != 1)
1922 /* promote the child to a root */
1923 child = read_node_slot(root, mid, 0);
1926 btrfs_std_error(root->fs_info, ret);
1930 btrfs_tree_lock(child);
1931 btrfs_set_lock_blocking(child);
1932 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1934 btrfs_tree_unlock(child);
1935 free_extent_buffer(child);
1939 tree_mod_log_set_root_pointer(root, child, 1);
1940 rcu_assign_pointer(root->node, child);
1942 add_root_to_dirty_list(root);
1943 btrfs_tree_unlock(child);
1945 path->locks[level] = 0;
1946 path->nodes[level] = NULL;
1947 clean_tree_block(trans, root, mid);
1948 btrfs_tree_unlock(mid);
1949 /* once for the path */
1950 free_extent_buffer(mid);
1952 root_sub_used(root, mid->len);
1953 btrfs_free_tree_block(trans, root, mid, 0, 1);
1954 /* once for the root ptr */
1955 free_extent_buffer_stale(mid);
1958 if (btrfs_header_nritems(mid) >
1959 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1962 left = read_node_slot(root, parent, pslot - 1);
1964 btrfs_tree_lock(left);
1965 btrfs_set_lock_blocking(left);
1966 wret = btrfs_cow_block(trans, root, left,
1967 parent, pslot - 1, &left);
1973 right = read_node_slot(root, parent, pslot + 1);
1975 btrfs_tree_lock(right);
1976 btrfs_set_lock_blocking(right);
1977 wret = btrfs_cow_block(trans, root, right,
1978 parent, pslot + 1, &right);
1985 /* first, try to make some room in the middle buffer */
1987 orig_slot += btrfs_header_nritems(left);
1988 wret = push_node_left(trans, root, left, mid, 1);
1994 * then try to empty the right most buffer into the middle
1997 wret = push_node_left(trans, root, mid, right, 1);
1998 if (wret < 0 && wret != -ENOSPC)
2000 if (btrfs_header_nritems(right) == 0) {
2001 clean_tree_block(trans, root, right);
2002 btrfs_tree_unlock(right);
2003 del_ptr(root, path, level + 1, pslot + 1);
2004 root_sub_used(root, right->len);
2005 btrfs_free_tree_block(trans, root, right, 0, 1);
2006 free_extent_buffer_stale(right);
2009 struct btrfs_disk_key right_key;
2010 btrfs_node_key(right, &right_key, 0);
2011 tree_mod_log_set_node_key(root->fs_info, parent,
2013 btrfs_set_node_key(parent, &right_key, pslot + 1);
2014 btrfs_mark_buffer_dirty(parent);
2017 if (btrfs_header_nritems(mid) == 1) {
2019 * we're not allowed to leave a node with one item in the
2020 * tree during a delete. A deletion from lower in the tree
2021 * could try to delete the only pointer in this node.
2022 * So, pull some keys from the left.
2023 * There has to be a left pointer at this point because
2024 * otherwise we would have pulled some pointers from the
2029 btrfs_std_error(root->fs_info, ret);
2032 wret = balance_node_right(trans, root, mid, left);
2038 wret = push_node_left(trans, root, left, mid, 1);
2044 if (btrfs_header_nritems(mid) == 0) {
2045 clean_tree_block(trans, root, mid);
2046 btrfs_tree_unlock(mid);
2047 del_ptr(root, path, level + 1, pslot);
2048 root_sub_used(root, mid->len);
2049 btrfs_free_tree_block(trans, root, mid, 0, 1);
2050 free_extent_buffer_stale(mid);
2053 /* update the parent key to reflect our changes */
2054 struct btrfs_disk_key mid_key;
2055 btrfs_node_key(mid, &mid_key, 0);
2056 tree_mod_log_set_node_key(root->fs_info, parent,
2058 btrfs_set_node_key(parent, &mid_key, pslot);
2059 btrfs_mark_buffer_dirty(parent);
2062 /* update the path */
2064 if (btrfs_header_nritems(left) > orig_slot) {
2065 extent_buffer_get(left);
2066 /* left was locked after cow */
2067 path->nodes[level] = left;
2068 path->slots[level + 1] -= 1;
2069 path->slots[level] = orig_slot;
2071 btrfs_tree_unlock(mid);
2072 free_extent_buffer(mid);
2075 orig_slot -= btrfs_header_nritems(left);
2076 path->slots[level] = orig_slot;
2079 /* double check we haven't messed things up */
2081 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2085 btrfs_tree_unlock(right);
2086 free_extent_buffer(right);
2089 if (path->nodes[level] != left)
2090 btrfs_tree_unlock(left);
2091 free_extent_buffer(left);
2096 /* Node balancing for insertion. Here we only split or push nodes around
2097 * when they are completely full. This is also done top down, so we
2098 * have to be pessimistic.
2100 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2101 struct btrfs_root *root,
2102 struct btrfs_path *path, int level)
2104 struct extent_buffer *right = NULL;
2105 struct extent_buffer *mid;
2106 struct extent_buffer *left = NULL;
2107 struct extent_buffer *parent = NULL;
2111 int orig_slot = path->slots[level];
2116 mid = path->nodes[level];
2117 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2119 if (level < BTRFS_MAX_LEVEL - 1) {
2120 parent = path->nodes[level + 1];
2121 pslot = path->slots[level + 1];
2127 left = read_node_slot(root, parent, pslot - 1);
2129 /* first, try to make some room in the middle buffer */
2133 btrfs_tree_lock(left);
2134 btrfs_set_lock_blocking(left);
2136 left_nr = btrfs_header_nritems(left);
2137 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2140 ret = btrfs_cow_block(trans, root, left, parent,
2145 wret = push_node_left(trans, root,
2152 struct btrfs_disk_key disk_key;
2153 orig_slot += left_nr;
2154 btrfs_node_key(mid, &disk_key, 0);
2155 tree_mod_log_set_node_key(root->fs_info, parent,
2157 btrfs_set_node_key(parent, &disk_key, pslot);
2158 btrfs_mark_buffer_dirty(parent);
2159 if (btrfs_header_nritems(left) > orig_slot) {
2160 path->nodes[level] = left;
2161 path->slots[level + 1] -= 1;
2162 path->slots[level] = orig_slot;
2163 btrfs_tree_unlock(mid);
2164 free_extent_buffer(mid);
2167 btrfs_header_nritems(left);
2168 path->slots[level] = orig_slot;
2169 btrfs_tree_unlock(left);
2170 free_extent_buffer(left);
2174 btrfs_tree_unlock(left);
2175 free_extent_buffer(left);
2177 right = read_node_slot(root, parent, pslot + 1);
2180 * then try to empty the right most buffer into the middle
2185 btrfs_tree_lock(right);
2186 btrfs_set_lock_blocking(right);
2188 right_nr = btrfs_header_nritems(right);
2189 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2192 ret = btrfs_cow_block(trans, root, right,
2198 wret = balance_node_right(trans, root,
2205 struct btrfs_disk_key disk_key;
2207 btrfs_node_key(right, &disk_key, 0);
2208 tree_mod_log_set_node_key(root->fs_info, parent,
2210 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2211 btrfs_mark_buffer_dirty(parent);
2213 if (btrfs_header_nritems(mid) <= orig_slot) {
2214 path->nodes[level] = right;
2215 path->slots[level + 1] += 1;
2216 path->slots[level] = orig_slot -
2217 btrfs_header_nritems(mid);
2218 btrfs_tree_unlock(mid);
2219 free_extent_buffer(mid);
2221 btrfs_tree_unlock(right);
2222 free_extent_buffer(right);
2226 btrfs_tree_unlock(right);
2227 free_extent_buffer(right);
2233 * readahead one full node of leaves, finding things that are close
2234 * to the block in 'slot', and triggering ra on them.
2236 static void reada_for_search(struct btrfs_root *root,
2237 struct btrfs_path *path,
2238 int level, int slot, u64 objectid)
2240 struct extent_buffer *node;
2241 struct btrfs_disk_key disk_key;
2247 int direction = path->reada;
2248 struct extent_buffer *eb;
2256 if (!path->nodes[level])
2259 node = path->nodes[level];
2261 search = btrfs_node_blockptr(node, slot);
2262 blocksize = root->nodesize;
2263 eb = btrfs_find_tree_block(root, search);
2265 free_extent_buffer(eb);
2271 nritems = btrfs_header_nritems(node);
2275 if (direction < 0) {
2279 } else if (direction > 0) {
2284 if (path->reada < 0 && objectid) {
2285 btrfs_node_key(node, &disk_key, nr);
2286 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2289 search = btrfs_node_blockptr(node, nr);
2290 if ((search <= target && target - search <= 65536) ||
2291 (search > target && search - target <= 65536)) {
2292 gen = btrfs_node_ptr_generation(node, nr);
2293 readahead_tree_block(root, search, blocksize);
2297 if ((nread > 65536 || nscan > 32))
2302 static noinline void reada_for_balance(struct btrfs_root *root,
2303 struct btrfs_path *path, int level)
2307 struct extent_buffer *parent;
2308 struct extent_buffer *eb;
2314 parent = path->nodes[level + 1];
2318 nritems = btrfs_header_nritems(parent);
2319 slot = path->slots[level + 1];
2320 blocksize = root->nodesize;
2323 block1 = btrfs_node_blockptr(parent, slot - 1);
2324 gen = btrfs_node_ptr_generation(parent, slot - 1);
2325 eb = btrfs_find_tree_block(root, block1);
2327 * if we get -eagain from btrfs_buffer_uptodate, we
2328 * don't want to return eagain here. That will loop
2331 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2333 free_extent_buffer(eb);
2335 if (slot + 1 < nritems) {
2336 block2 = btrfs_node_blockptr(parent, slot + 1);
2337 gen = btrfs_node_ptr_generation(parent, slot + 1);
2338 eb = btrfs_find_tree_block(root, block2);
2339 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2341 free_extent_buffer(eb);
2345 readahead_tree_block(root, block1, blocksize);
2347 readahead_tree_block(root, block2, blocksize);
2352 * when we walk down the tree, it is usually safe to unlock the higher layers
2353 * in the tree. The exceptions are when our path goes through slot 0, because
2354 * operations on the tree might require changing key pointers higher up in the
2357 * callers might also have set path->keep_locks, which tells this code to keep
2358 * the lock if the path points to the last slot in the block. This is part of
2359 * walking through the tree, and selecting the next slot in the higher block.
2361 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2362 * if lowest_unlock is 1, level 0 won't be unlocked
2364 static noinline void unlock_up(struct btrfs_path *path, int level,
2365 int lowest_unlock, int min_write_lock_level,
2366 int *write_lock_level)
2369 int skip_level = level;
2371 struct extent_buffer *t;
2373 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2374 if (!path->nodes[i])
2376 if (!path->locks[i])
2378 if (!no_skips && path->slots[i] == 0) {
2382 if (!no_skips && path->keep_locks) {
2385 nritems = btrfs_header_nritems(t);
2386 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2391 if (skip_level < i && i >= lowest_unlock)
2395 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2396 btrfs_tree_unlock_rw(t, path->locks[i]);
2398 if (write_lock_level &&
2399 i > min_write_lock_level &&
2400 i <= *write_lock_level) {
2401 *write_lock_level = i - 1;
2408 * This releases any locks held in the path starting at level and
2409 * going all the way up to the root.
2411 * btrfs_search_slot will keep the lock held on higher nodes in a few
2412 * corner cases, such as COW of the block at slot zero in the node. This
2413 * ignores those rules, and it should only be called when there are no
2414 * more updates to be done higher up in the tree.
2416 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2420 if (path->keep_locks)
2423 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2424 if (!path->nodes[i])
2426 if (!path->locks[i])
2428 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2434 * helper function for btrfs_search_slot. The goal is to find a block
2435 * in cache without setting the path to blocking. If we find the block
2436 * we return zero and the path is unchanged.
2438 * If we can't find the block, we set the path blocking and do some
2439 * reada. -EAGAIN is returned and the search must be repeated.
2442 read_block_for_search(struct btrfs_trans_handle *trans,
2443 struct btrfs_root *root, struct btrfs_path *p,
2444 struct extent_buffer **eb_ret, int level, int slot,
2445 struct btrfs_key *key, u64 time_seq)
2449 struct extent_buffer *b = *eb_ret;
2450 struct extent_buffer *tmp;
2453 blocknr = btrfs_node_blockptr(b, slot);
2454 gen = btrfs_node_ptr_generation(b, slot);
2456 tmp = btrfs_find_tree_block(root, blocknr);
2458 /* first we do an atomic uptodate check */
2459 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2464 /* the pages were up to date, but we failed
2465 * the generation number check. Do a full
2466 * read for the generation number that is correct.
2467 * We must do this without dropping locks so
2468 * we can trust our generation number
2470 btrfs_set_path_blocking(p);
2472 /* now we're allowed to do a blocking uptodate check */
2473 ret = btrfs_read_buffer(tmp, gen);
2478 free_extent_buffer(tmp);
2479 btrfs_release_path(p);
2484 * reduce lock contention at high levels
2485 * of the btree by dropping locks before
2486 * we read. Don't release the lock on the current
2487 * level because we need to walk this node to figure
2488 * out which blocks to read.
2490 btrfs_unlock_up_safe(p, level + 1);
2491 btrfs_set_path_blocking(p);
2493 free_extent_buffer(tmp);
2495 reada_for_search(root, p, level, slot, key->objectid);
2497 btrfs_release_path(p);
2500 tmp = read_tree_block(root, blocknr, 0);
2503 * If the read above didn't mark this buffer up to date,
2504 * it will never end up being up to date. Set ret to EIO now
2505 * and give up so that our caller doesn't loop forever
2508 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2510 free_extent_buffer(tmp);
2516 * helper function for btrfs_search_slot. This does all of the checks
2517 * for node-level blocks and does any balancing required based on
2520 * If no extra work was required, zero is returned. If we had to
2521 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2525 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2526 struct btrfs_root *root, struct btrfs_path *p,
2527 struct extent_buffer *b, int level, int ins_len,
2528 int *write_lock_level)
2531 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2532 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2535 if (*write_lock_level < level + 1) {
2536 *write_lock_level = level + 1;
2537 btrfs_release_path(p);
2541 btrfs_set_path_blocking(p);
2542 reada_for_balance(root, p, level);
2543 sret = split_node(trans, root, p, level);
2544 btrfs_clear_path_blocking(p, NULL, 0);
2551 b = p->nodes[level];
2552 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2553 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2556 if (*write_lock_level < level + 1) {
2557 *write_lock_level = level + 1;
2558 btrfs_release_path(p);
2562 btrfs_set_path_blocking(p);
2563 reada_for_balance(root, p, level);
2564 sret = balance_level(trans, root, p, level);
2565 btrfs_clear_path_blocking(p, NULL, 0);
2571 b = p->nodes[level];
2573 btrfs_release_path(p);
2576 BUG_ON(btrfs_header_nritems(b) == 1);
2586 static void key_search_validate(struct extent_buffer *b,
2587 struct btrfs_key *key,
2590 #ifdef CONFIG_BTRFS_ASSERT
2591 struct btrfs_disk_key disk_key;
2593 btrfs_cpu_key_to_disk(&disk_key, key);
2596 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2597 offsetof(struct btrfs_leaf, items[0].key),
2600 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2601 offsetof(struct btrfs_node, ptrs[0].key),
2606 static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2607 int level, int *prev_cmp, int *slot)
2609 if (*prev_cmp != 0) {
2610 *prev_cmp = bin_search(b, key, level, slot);
2614 key_search_validate(b, key, level);
2620 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2621 u64 iobjectid, u64 ioff, u8 key_type,
2622 struct btrfs_key *found_key)
2625 struct btrfs_key key;
2626 struct extent_buffer *eb;
2631 key.type = key_type;
2632 key.objectid = iobjectid;
2635 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2639 eb = path->nodes[0];
2640 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2641 ret = btrfs_next_leaf(fs_root, path);
2644 eb = path->nodes[0];
2647 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2648 if (found_key->type != key.type ||
2649 found_key->objectid != key.objectid)
2656 * look for key in the tree. path is filled in with nodes along the way
2657 * if key is found, we return zero and you can find the item in the leaf
2658 * level of the path (level 0)
2660 * If the key isn't found, the path points to the slot where it should
2661 * be inserted, and 1 is returned. If there are other errors during the
2662 * search a negative error number is returned.
2664 * if ins_len > 0, nodes and leaves will be split as we walk down the
2665 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2668 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2669 *root, struct btrfs_key *key, struct btrfs_path *p, int
2672 struct extent_buffer *b;
2677 int lowest_unlock = 1;
2679 /* everything at write_lock_level or lower must be write locked */
2680 int write_lock_level = 0;
2681 u8 lowest_level = 0;
2682 int min_write_lock_level;
2685 lowest_level = p->lowest_level;
2686 WARN_ON(lowest_level && ins_len > 0);
2687 WARN_ON(p->nodes[0] != NULL);
2688 BUG_ON(!cow && ins_len);
2693 /* when we are removing items, we might have to go up to level
2694 * two as we update tree pointers Make sure we keep write
2695 * for those levels as well
2697 write_lock_level = 2;
2698 } else if (ins_len > 0) {
2700 * for inserting items, make sure we have a write lock on
2701 * level 1 so we can update keys
2703 write_lock_level = 1;
2707 write_lock_level = -1;
2709 if (cow && (p->keep_locks || p->lowest_level))
2710 write_lock_level = BTRFS_MAX_LEVEL;
2712 min_write_lock_level = write_lock_level;
2717 * we try very hard to do read locks on the root
2719 root_lock = BTRFS_READ_LOCK;
2721 if (p->search_commit_root) {
2723 * the commit roots are read only
2724 * so we always do read locks
2726 if (p->need_commit_sem)
2727 down_read(&root->fs_info->commit_root_sem);
2728 b = root->commit_root;
2729 extent_buffer_get(b);
2730 level = btrfs_header_level(b);
2731 if (p->need_commit_sem)
2732 up_read(&root->fs_info->commit_root_sem);
2733 if (!p->skip_locking)
2734 btrfs_tree_read_lock(b);
2736 if (p->skip_locking) {
2737 b = btrfs_root_node(root);
2738 level = btrfs_header_level(b);
2740 /* we don't know the level of the root node
2741 * until we actually have it read locked
2743 b = btrfs_read_lock_root_node(root);
2744 level = btrfs_header_level(b);
2745 if (level <= write_lock_level) {
2746 /* whoops, must trade for write lock */
2747 btrfs_tree_read_unlock(b);
2748 free_extent_buffer(b);
2749 b = btrfs_lock_root_node(root);
2750 root_lock = BTRFS_WRITE_LOCK;
2752 /* the level might have changed, check again */
2753 level = btrfs_header_level(b);
2757 p->nodes[level] = b;
2758 if (!p->skip_locking)
2759 p->locks[level] = root_lock;
2762 level = btrfs_header_level(b);
2765 * setup the path here so we can release it under lock
2766 * contention with the cow code
2770 * if we don't really need to cow this block
2771 * then we don't want to set the path blocking,
2772 * so we test it here
2774 if (!should_cow_block(trans, root, b))
2778 * must have write locks on this node and the
2781 if (level > write_lock_level ||
2782 (level + 1 > write_lock_level &&
2783 level + 1 < BTRFS_MAX_LEVEL &&
2784 p->nodes[level + 1])) {
2785 write_lock_level = level + 1;
2786 btrfs_release_path(p);
2790 btrfs_set_path_blocking(p);
2791 err = btrfs_cow_block(trans, root, b,
2792 p->nodes[level + 1],
2793 p->slots[level + 1], &b);
2800 p->nodes[level] = b;
2801 btrfs_clear_path_blocking(p, NULL, 0);
2804 * we have a lock on b and as long as we aren't changing
2805 * the tree, there is no way to for the items in b to change.
2806 * It is safe to drop the lock on our parent before we
2807 * go through the expensive btree search on b.
2809 * If we're inserting or deleting (ins_len != 0), then we might
2810 * be changing slot zero, which may require changing the parent.
2811 * So, we can't drop the lock until after we know which slot
2812 * we're operating on.
2814 if (!ins_len && !p->keep_locks) {
2817 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2818 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2823 ret = key_search(b, key, level, &prev_cmp, &slot);
2827 if (ret && slot > 0) {
2831 p->slots[level] = slot;
2832 err = setup_nodes_for_search(trans, root, p, b, level,
2833 ins_len, &write_lock_level);
2840 b = p->nodes[level];
2841 slot = p->slots[level];
2844 * slot 0 is special, if we change the key
2845 * we have to update the parent pointer
2846 * which means we must have a write lock
2849 if (slot == 0 && ins_len &&
2850 write_lock_level < level + 1) {
2851 write_lock_level = level + 1;
2852 btrfs_release_path(p);
2856 unlock_up(p, level, lowest_unlock,
2857 min_write_lock_level, &write_lock_level);
2859 if (level == lowest_level) {
2865 err = read_block_for_search(trans, root, p,
2866 &b, level, slot, key, 0);
2874 if (!p->skip_locking) {
2875 level = btrfs_header_level(b);
2876 if (level <= write_lock_level) {
2877 err = btrfs_try_tree_write_lock(b);
2879 btrfs_set_path_blocking(p);
2881 btrfs_clear_path_blocking(p, b,
2884 p->locks[level] = BTRFS_WRITE_LOCK;
2886 err = btrfs_tree_read_lock_atomic(b);
2888 btrfs_set_path_blocking(p);
2889 btrfs_tree_read_lock(b);
2890 btrfs_clear_path_blocking(p, b,
2893 p->locks[level] = BTRFS_READ_LOCK;
2895 p->nodes[level] = b;
2898 p->slots[level] = slot;
2900 btrfs_leaf_free_space(root, b) < ins_len) {
2901 if (write_lock_level < 1) {
2902 write_lock_level = 1;
2903 btrfs_release_path(p);
2907 btrfs_set_path_blocking(p);
2908 err = split_leaf(trans, root, key,
2909 p, ins_len, ret == 0);
2910 btrfs_clear_path_blocking(p, NULL, 0);
2918 if (!p->search_for_split)
2919 unlock_up(p, level, lowest_unlock,
2920 min_write_lock_level, &write_lock_level);
2927 * we don't really know what they plan on doing with the path
2928 * from here on, so for now just mark it as blocking
2930 if (!p->leave_spinning)
2931 btrfs_set_path_blocking(p);
2932 if (ret < 0 && !p->skip_release_on_error)
2933 btrfs_release_path(p);
2938 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2939 * current state of the tree together with the operations recorded in the tree
2940 * modification log to search for the key in a previous version of this tree, as
2941 * denoted by the time_seq parameter.
2943 * Naturally, there is no support for insert, delete or cow operations.
2945 * The resulting path and return value will be set up as if we called
2946 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2948 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2949 struct btrfs_path *p, u64 time_seq)
2951 struct extent_buffer *b;
2956 int lowest_unlock = 1;
2957 u8 lowest_level = 0;
2960 lowest_level = p->lowest_level;
2961 WARN_ON(p->nodes[0] != NULL);
2963 if (p->search_commit_root) {
2965 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2969 b = get_old_root(root, time_seq);
2970 level = btrfs_header_level(b);
2971 p->locks[level] = BTRFS_READ_LOCK;
2974 level = btrfs_header_level(b);
2975 p->nodes[level] = b;
2976 btrfs_clear_path_blocking(p, NULL, 0);
2979 * we have a lock on b and as long as we aren't changing
2980 * the tree, there is no way to for the items in b to change.
2981 * It is safe to drop the lock on our parent before we
2982 * go through the expensive btree search on b.
2984 btrfs_unlock_up_safe(p, level + 1);
2987 * Since we can unwind eb's we want to do a real search every
2991 ret = key_search(b, key, level, &prev_cmp, &slot);
2995 if (ret && slot > 0) {
2999 p->slots[level] = slot;
3000 unlock_up(p, level, lowest_unlock, 0, NULL);
3002 if (level == lowest_level) {
3008 err = read_block_for_search(NULL, root, p, &b, level,
3009 slot, key, time_seq);
3017 level = btrfs_header_level(b);
3018 err = btrfs_tree_read_lock_atomic(b);
3020 btrfs_set_path_blocking(p);
3021 btrfs_tree_read_lock(b);
3022 btrfs_clear_path_blocking(p, b,
3025 b = tree_mod_log_rewind(root->fs_info, p, b, time_seq);
3030 p->locks[level] = BTRFS_READ_LOCK;
3031 p->nodes[level] = b;
3033 p->slots[level] = slot;
3034 unlock_up(p, level, lowest_unlock, 0, NULL);
3040 if (!p->leave_spinning)
3041 btrfs_set_path_blocking(p);
3043 btrfs_release_path(p);
3049 * helper to use instead of search slot if no exact match is needed but
3050 * instead the next or previous item should be returned.
3051 * When find_higher is true, the next higher item is returned, the next lower
3053 * When return_any and find_higher are both true, and no higher item is found,
3054 * return the next lower instead.
3055 * When return_any is true and find_higher is false, and no lower item is found,
3056 * return the next higher instead.
3057 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3060 int btrfs_search_slot_for_read(struct btrfs_root *root,
3061 struct btrfs_key *key, struct btrfs_path *p,
3062 int find_higher, int return_any)
3065 struct extent_buffer *leaf;
3068 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3072 * a return value of 1 means the path is at the position where the
3073 * item should be inserted. Normally this is the next bigger item,
3074 * but in case the previous item is the last in a leaf, path points
3075 * to the first free slot in the previous leaf, i.e. at an invalid
3081 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3082 ret = btrfs_next_leaf(root, p);
3088 * no higher item found, return the next
3093 btrfs_release_path(p);
3097 if (p->slots[0] == 0) {
3098 ret = btrfs_prev_leaf(root, p);
3103 if (p->slots[0] == btrfs_header_nritems(leaf))
3110 * no lower item found, return the next
3115 btrfs_release_path(p);
3125 * adjust the pointers going up the tree, starting at level
3126 * making sure the right key of each node is points to 'key'.
3127 * This is used after shifting pointers to the left, so it stops
3128 * fixing up pointers when a given leaf/node is not in slot 0 of the
3132 static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
3133 struct btrfs_disk_key *key, int level)
3136 struct extent_buffer *t;
3138 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3139 int tslot = path->slots[i];
3140 if (!path->nodes[i])
3143 tree_mod_log_set_node_key(root->fs_info, t, tslot, 1);
3144 btrfs_set_node_key(t, key, tslot);
3145 btrfs_mark_buffer_dirty(path->nodes[i]);
3154 * This function isn't completely safe. It's the caller's responsibility
3155 * that the new key won't break the order
3157 void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
3158 struct btrfs_key *new_key)
3160 struct btrfs_disk_key disk_key;
3161 struct extent_buffer *eb;
3164 eb = path->nodes[0];
3165 slot = path->slots[0];
3167 btrfs_item_key(eb, &disk_key, slot - 1);
3168 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3170 if (slot < btrfs_header_nritems(eb) - 1) {
3171 btrfs_item_key(eb, &disk_key, slot + 1);
3172 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3175 btrfs_cpu_key_to_disk(&disk_key, new_key);
3176 btrfs_set_item_key(eb, &disk_key, slot);
3177 btrfs_mark_buffer_dirty(eb);
3179 fixup_low_keys(root, path, &disk_key, 1);
3183 * try to push data from one node into the next node left in the
3186 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3187 * error, and > 0 if there was no room in the left hand block.
3189 static int push_node_left(struct btrfs_trans_handle *trans,
3190 struct btrfs_root *root, struct extent_buffer *dst,
3191 struct extent_buffer *src, int empty)
3198 src_nritems = btrfs_header_nritems(src);
3199 dst_nritems = btrfs_header_nritems(dst);
3200 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3201 WARN_ON(btrfs_header_generation(src) != trans->transid);
3202 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3204 if (!empty && src_nritems <= 8)
3207 if (push_items <= 0)
3211 push_items = min(src_nritems, push_items);
3212 if (push_items < src_nritems) {
3213 /* leave at least 8 pointers in the node if
3214 * we aren't going to empty it
3216 if (src_nritems - push_items < 8) {
3217 if (push_items <= 8)
3223 push_items = min(src_nritems - 8, push_items);
3225 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3228 btrfs_abort_transaction(trans, root, ret);
3231 copy_extent_buffer(dst, src,
3232 btrfs_node_key_ptr_offset(dst_nritems),
3233 btrfs_node_key_ptr_offset(0),
3234 push_items * sizeof(struct btrfs_key_ptr));
3236 if (push_items < src_nritems) {
3238 * don't call tree_mod_log_eb_move here, key removal was already
3239 * fully logged by tree_mod_log_eb_copy above.
3241 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3242 btrfs_node_key_ptr_offset(push_items),
3243 (src_nritems - push_items) *
3244 sizeof(struct btrfs_key_ptr));
3246 btrfs_set_header_nritems(src, src_nritems - push_items);
3247 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3248 btrfs_mark_buffer_dirty(src);
3249 btrfs_mark_buffer_dirty(dst);
3255 * try to push data from one node into the next node right in the
3258 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3259 * error, and > 0 if there was no room in the right hand block.
3261 * this will only push up to 1/2 the contents of the left node over
3263 static int balance_node_right(struct btrfs_trans_handle *trans,
3264 struct btrfs_root *root,
3265 struct extent_buffer *dst,
3266 struct extent_buffer *src)
3274 WARN_ON(btrfs_header_generation(src) != trans->transid);
3275 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3277 src_nritems = btrfs_header_nritems(src);
3278 dst_nritems = btrfs_header_nritems(dst);
3279 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3280 if (push_items <= 0)
3283 if (src_nritems < 4)
3286 max_push = src_nritems / 2 + 1;
3287 /* don't try to empty the node */
3288 if (max_push >= src_nritems)
3291 if (max_push < push_items)
3292 push_items = max_push;
3294 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3295 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3296 btrfs_node_key_ptr_offset(0),
3298 sizeof(struct btrfs_key_ptr));
3300 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3301 src_nritems - push_items, push_items);
3303 btrfs_abort_transaction(trans, root, ret);
3306 copy_extent_buffer(dst, src,
3307 btrfs_node_key_ptr_offset(0),
3308 btrfs_node_key_ptr_offset(src_nritems - push_items),
3309 push_items * sizeof(struct btrfs_key_ptr));
3311 btrfs_set_header_nritems(src, src_nritems - push_items);
3312 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3314 btrfs_mark_buffer_dirty(src);
3315 btrfs_mark_buffer_dirty(dst);
3321 * helper function to insert a new root level in the tree.
3322 * A new node is allocated, and a single item is inserted to
3323 * point to the existing root
3325 * returns zero on success or < 0 on failure.
3327 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3328 struct btrfs_root *root,
3329 struct btrfs_path *path, int level)
3332 struct extent_buffer *lower;
3333 struct extent_buffer *c;
3334 struct extent_buffer *old;
3335 struct btrfs_disk_key lower_key;
3337 BUG_ON(path->nodes[level]);
3338 BUG_ON(path->nodes[level-1] != root->node);
3340 lower = path->nodes[level-1];
3342 btrfs_item_key(lower, &lower_key, 0);
3344 btrfs_node_key(lower, &lower_key, 0);
3346 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3347 &lower_key, level, root->node->start, 0);
3351 root_add_used(root, root->nodesize);
3353 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3354 btrfs_set_header_nritems(c, 1);
3355 btrfs_set_header_level(c, level);
3356 btrfs_set_header_bytenr(c, c->start);
3357 btrfs_set_header_generation(c, trans->transid);
3358 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3359 btrfs_set_header_owner(c, root->root_key.objectid);
3361 write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(),
3364 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3365 btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE);
3367 btrfs_set_node_key(c, &lower_key, 0);
3368 btrfs_set_node_blockptr(c, 0, lower->start);
3369 lower_gen = btrfs_header_generation(lower);
3370 WARN_ON(lower_gen != trans->transid);
3372 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3374 btrfs_mark_buffer_dirty(c);
3377 tree_mod_log_set_root_pointer(root, c, 0);
3378 rcu_assign_pointer(root->node, c);
3380 /* the super has an extra ref to root->node */
3381 free_extent_buffer(old);
3383 add_root_to_dirty_list(root);
3384 extent_buffer_get(c);
3385 path->nodes[level] = c;
3386 path->locks[level] = BTRFS_WRITE_LOCK;
3387 path->slots[level] = 0;
3392 * worker function to insert a single pointer in a node.
3393 * the node should have enough room for the pointer already
3395 * slot and level indicate where you want the key to go, and
3396 * blocknr is the block the key points to.
3398 static void insert_ptr(struct btrfs_trans_handle *trans,
3399 struct btrfs_root *root, struct btrfs_path *path,
3400 struct btrfs_disk_key *key, u64 bytenr,
3401 int slot, int level)
3403 struct extent_buffer *lower;
3407 BUG_ON(!path->nodes[level]);
3408 btrfs_assert_tree_locked(path->nodes[level]);
3409 lower = path->nodes[level];
3410 nritems = btrfs_header_nritems(lower);
3411 BUG_ON(slot > nritems);
3412 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3413 if (slot != nritems) {
3415 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3416 slot, nritems - slot);
3417 memmove_extent_buffer(lower,
3418 btrfs_node_key_ptr_offset(slot + 1),
3419 btrfs_node_key_ptr_offset(slot),
3420 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3423 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3424 MOD_LOG_KEY_ADD, GFP_NOFS);
3427 btrfs_set_node_key(lower, key, slot);
3428 btrfs_set_node_blockptr(lower, slot, bytenr);
3429 WARN_ON(trans->transid == 0);
3430 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3431 btrfs_set_header_nritems(lower, nritems + 1);
3432 btrfs_mark_buffer_dirty(lower);
3436 * split the node at the specified level in path in two.
3437 * The path is corrected to point to the appropriate node after the split
3439 * Before splitting this tries to make some room in the node by pushing
3440 * left and right, if either one works, it returns right away.
3442 * returns 0 on success and < 0 on failure
3444 static noinline int split_node(struct btrfs_trans_handle *trans,
3445 struct btrfs_root *root,
3446 struct btrfs_path *path, int level)
3448 struct extent_buffer *c;
3449 struct extent_buffer *split;
3450 struct btrfs_disk_key disk_key;
3455 c = path->nodes[level];
3456 WARN_ON(btrfs_header_generation(c) != trans->transid);
3457 if (c == root->node) {
3459 * trying to split the root, lets make a new one
3461 * tree mod log: We don't log_removal old root in
3462 * insert_new_root, because that root buffer will be kept as a
3463 * normal node. We are going to log removal of half of the
3464 * elements below with tree_mod_log_eb_copy. We're holding a
3465 * tree lock on the buffer, which is why we cannot race with
3466 * other tree_mod_log users.
3468 ret = insert_new_root(trans, root, path, level + 1);
3472 ret = push_nodes_for_insert(trans, root, path, level);
3473 c = path->nodes[level];
3474 if (!ret && btrfs_header_nritems(c) <
3475 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3481 c_nritems = btrfs_header_nritems(c);
3482 mid = (c_nritems + 1) / 2;
3483 btrfs_node_key(c, &disk_key, mid);
3485 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3486 &disk_key, level, c->start, 0);
3488 return PTR_ERR(split);
3490 root_add_used(root, root->nodesize);
3492 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3493 btrfs_set_header_level(split, btrfs_header_level(c));
3494 btrfs_set_header_bytenr(split, split->start);
3495 btrfs_set_header_generation(split, trans->transid);
3496 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3497 btrfs_set_header_owner(split, root->root_key.objectid);
3498 write_extent_buffer(split, root->fs_info->fsid,
3499 btrfs_header_fsid(), BTRFS_FSID_SIZE);
3500 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3501 btrfs_header_chunk_tree_uuid(split),
3504 ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0,
3505 mid, c_nritems - mid);
3507 btrfs_abort_transaction(trans, root, ret);
3510 copy_extent_buffer(split, c,
3511 btrfs_node_key_ptr_offset(0),
3512 btrfs_node_key_ptr_offset(mid),
3513 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3514 btrfs_set_header_nritems(split, c_nritems - mid);
3515 btrfs_set_header_nritems(c, mid);
3518 btrfs_mark_buffer_dirty(c);
3519 btrfs_mark_buffer_dirty(split);
3521 insert_ptr(trans, root, path, &disk_key, split->start,
3522 path->slots[level + 1] + 1, level + 1);
3524 if (path->slots[level] >= mid) {
3525 path->slots[level] -= mid;
3526 btrfs_tree_unlock(c);
3527 free_extent_buffer(c);
3528 path->nodes[level] = split;
3529 path->slots[level + 1] += 1;
3531 btrfs_tree_unlock(split);
3532 free_extent_buffer(split);
3538 * how many bytes are required to store the items in a leaf. start
3539 * and nr indicate which items in the leaf to check. This totals up the
3540 * space used both by the item structs and the item data
3542 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3544 struct btrfs_item *start_item;
3545 struct btrfs_item *end_item;
3546 struct btrfs_map_token token;
3548 int nritems = btrfs_header_nritems(l);
3549 int end = min(nritems, start + nr) - 1;
3553 btrfs_init_map_token(&token);
3554 start_item = btrfs_item_nr(start);
3555 end_item = btrfs_item_nr(end);
3556 data_len = btrfs_token_item_offset(l, start_item, &token) +
3557 btrfs_token_item_size(l, start_item, &token);
3558 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3559 data_len += sizeof(struct btrfs_item) * nr;
3560 WARN_ON(data_len < 0);
3565 * The space between the end of the leaf items and
3566 * the start of the leaf data. IOW, how much room
3567 * the leaf has left for both items and data
3569 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3570 struct extent_buffer *leaf)
3572 int nritems = btrfs_header_nritems(leaf);
3574 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3576 btrfs_crit(root->fs_info,
3577 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3578 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3579 leaf_space_used(leaf, 0, nritems), nritems);
3585 * min slot controls the lowest index we're willing to push to the
3586 * right. We'll push up to and including min_slot, but no lower
3588 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3589 struct btrfs_root *root,
3590 struct btrfs_path *path,
3591 int data_size, int empty,
3592 struct extent_buffer *right,
3593 int free_space, u32 left_nritems,
3596 struct extent_buffer *left = path->nodes[0];
3597 struct extent_buffer *upper = path->nodes[1];
3598 struct btrfs_map_token token;
3599 struct btrfs_disk_key disk_key;
3604 struct btrfs_item *item;
3610 btrfs_init_map_token(&token);
3615 nr = max_t(u32, 1, min_slot);
3617 if (path->slots[0] >= left_nritems)
3618 push_space += data_size;
3620 slot = path->slots[1];
3621 i = left_nritems - 1;
3623 item = btrfs_item_nr(i);
3625 if (!empty && push_items > 0) {
3626 if (path->slots[0] > i)
3628 if (path->slots[0] == i) {
3629 int space = btrfs_leaf_free_space(root, left);
3630 if (space + push_space * 2 > free_space)
3635 if (path->slots[0] == i)
3636 push_space += data_size;
3638 this_item_size = btrfs_item_size(left, item);
3639 if (this_item_size + sizeof(*item) + push_space > free_space)
3643 push_space += this_item_size + sizeof(*item);
3649 if (push_items == 0)
3652 WARN_ON(!empty && push_items == left_nritems);
3654 /* push left to right */
3655 right_nritems = btrfs_header_nritems(right);
3657 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3658 push_space -= leaf_data_end(root, left);
3660 /* make room in the right data area */
3661 data_end = leaf_data_end(root, right);
3662 memmove_extent_buffer(right,
3663 btrfs_leaf_data(right) + data_end - push_space,
3664 btrfs_leaf_data(right) + data_end,
3665 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3667 /* copy from the left data area */
3668 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3669 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3670 btrfs_leaf_data(left) + leaf_data_end(root, left),
3673 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3674 btrfs_item_nr_offset(0),
3675 right_nritems * sizeof(struct btrfs_item));
3677 /* copy the items from left to right */
3678 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3679 btrfs_item_nr_offset(left_nritems - push_items),
3680 push_items * sizeof(struct btrfs_item));
3682 /* update the item pointers */
3683 right_nritems += push_items;
3684 btrfs_set_header_nritems(right, right_nritems);
3685 push_space = BTRFS_LEAF_DATA_SIZE(root);
3686 for (i = 0; i < right_nritems; i++) {
3687 item = btrfs_item_nr(i);
3688 push_space -= btrfs_token_item_size(right, item, &token);
3689 btrfs_set_token_item_offset(right, item, push_space, &token);
3692 left_nritems -= push_items;
3693 btrfs_set_header_nritems(left, left_nritems);
3696 btrfs_mark_buffer_dirty(left);
3698 clean_tree_block(trans, root, left);
3700 btrfs_mark_buffer_dirty(right);
3702 btrfs_item_key(right, &disk_key, 0);
3703 btrfs_set_node_key(upper, &disk_key, slot + 1);
3704 btrfs_mark_buffer_dirty(upper);
3706 /* then fixup the leaf pointer in the path */
3707 if (path->slots[0] >= left_nritems) {
3708 path->slots[0] -= left_nritems;
3709 if (btrfs_header_nritems(path->nodes[0]) == 0)
3710 clean_tree_block(trans, root, path->nodes[0]);
3711 btrfs_tree_unlock(path->nodes[0]);
3712 free_extent_buffer(path->nodes[0]);
3713 path->nodes[0] = right;
3714 path->slots[1] += 1;
3716 btrfs_tree_unlock(right);
3717 free_extent_buffer(right);
3722 btrfs_tree_unlock(right);
3723 free_extent_buffer(right);
3728 * push some data in the path leaf to the right, trying to free up at
3729 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3731 * returns 1 if the push failed because the other node didn't have enough
3732 * room, 0 if everything worked out and < 0 if there were major errors.
3734 * this will push starting from min_slot to the end of the leaf. It won't
3735 * push any slot lower than min_slot
3737 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3738 *root, struct btrfs_path *path,
3739 int min_data_size, int data_size,
3740 int empty, u32 min_slot)
3742 struct extent_buffer *left = path->nodes[0];
3743 struct extent_buffer *right;
3744 struct extent_buffer *upper;
3750 if (!path->nodes[1])
3753 slot = path->slots[1];
3754 upper = path->nodes[1];
3755 if (slot >= btrfs_header_nritems(upper) - 1)
3758 btrfs_assert_tree_locked(path->nodes[1]);
3760 right = read_node_slot(root, upper, slot + 1);
3764 btrfs_tree_lock(right);
3765 btrfs_set_lock_blocking(right);
3767 free_space = btrfs_leaf_free_space(root, right);
3768 if (free_space < data_size)
3771 /* cow and double check */
3772 ret = btrfs_cow_block(trans, root, right, upper,
3777 free_space = btrfs_leaf_free_space(root, right);
3778 if (free_space < data_size)
3781 left_nritems = btrfs_header_nritems(left);
3782 if (left_nritems == 0)
3785 if (path->slots[0] == left_nritems && !empty) {
3786 /* Key greater than all keys in the leaf, right neighbor has
3787 * enough room for it and we're not emptying our leaf to delete
3788 * it, therefore use right neighbor to insert the new item and
3789 * no need to touch/dirty our left leaft. */
3790 btrfs_tree_unlock(left);
3791 free_extent_buffer(left);
3792 path->nodes[0] = right;
3798 return __push_leaf_right(trans, root, path, min_data_size, empty,
3799 right, free_space, left_nritems, min_slot);
3801 btrfs_tree_unlock(right);
3802 free_extent_buffer(right);
3807 * push some data in the path leaf to the left, trying to free up at
3808 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3810 * max_slot can put a limit on how far into the leaf we'll push items. The
3811 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3814 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3815 struct btrfs_root *root,
3816 struct btrfs_path *path, int data_size,
3817 int empty, struct extent_buffer *left,
3818 int free_space, u32 right_nritems,
3821 struct btrfs_disk_key disk_key;
3822 struct extent_buffer *right = path->nodes[0];
3826 struct btrfs_item *item;
3827 u32 old_left_nritems;
3831 u32 old_left_item_size;
3832 struct btrfs_map_token token;
3834 btrfs_init_map_token(&token);
3837 nr = min(right_nritems, max_slot);
3839 nr = min(right_nritems - 1, max_slot);
3841 for (i = 0; i < nr; i++) {
3842 item = btrfs_item_nr(i);
3844 if (!empty && push_items > 0) {
3845 if (path->slots[0] < i)
3847 if (path->slots[0] == i) {
3848 int space = btrfs_leaf_free_space(root, right);
3849 if (space + push_space * 2 > free_space)
3854 if (path->slots[0] == i)
3855 push_space += data_size;
3857 this_item_size = btrfs_item_size(right, item);
3858 if (this_item_size + sizeof(*item) + push_space > free_space)
3862 push_space += this_item_size + sizeof(*item);
3865 if (push_items == 0) {
3869 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3871 /* push data from right to left */
3872 copy_extent_buffer(left, right,
3873 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3874 btrfs_item_nr_offset(0),
3875 push_items * sizeof(struct btrfs_item));
3877 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3878 btrfs_item_offset_nr(right, push_items - 1);
3880 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3881 leaf_data_end(root, left) - push_space,
3882 btrfs_leaf_data(right) +
3883 btrfs_item_offset_nr(right, push_items - 1),
3885 old_left_nritems = btrfs_header_nritems(left);
3886 BUG_ON(old_left_nritems <= 0);
3888 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3889 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3892 item = btrfs_item_nr(i);
3894 ioff = btrfs_token_item_offset(left, item, &token);
3895 btrfs_set_token_item_offset(left, item,
3896 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3899 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3901 /* fixup right node */
3902 if (push_items > right_nritems)
3903 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3906 if (push_items < right_nritems) {
3907 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3908 leaf_data_end(root, right);
3909 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3910 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3911 btrfs_leaf_data(right) +
3912 leaf_data_end(root, right), push_space);
3914 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3915 btrfs_item_nr_offset(push_items),
3916 (btrfs_header_nritems(right) - push_items) *
3917 sizeof(struct btrfs_item));
3919 right_nritems -= push_items;
3920 btrfs_set_header_nritems(right, right_nritems);
3921 push_space = BTRFS_LEAF_DATA_SIZE(root);
3922 for (i = 0; i < right_nritems; i++) {
3923 item = btrfs_item_nr(i);
3925 push_space = push_space - btrfs_token_item_size(right,
3927 btrfs_set_token_item_offset(right, item, push_space, &token);
3930 btrfs_mark_buffer_dirty(left);
3932 btrfs_mark_buffer_dirty(right);
3934 clean_tree_block(trans, root, right);
3936 btrfs_item_key(right, &disk_key, 0);
3937 fixup_low_keys(root, path, &disk_key, 1);
3939 /* then fixup the leaf pointer in the path */
3940 if (path->slots[0] < push_items) {
3941 path->slots[0] += old_left_nritems;
3942 btrfs_tree_unlock(path->nodes[0]);
3943 free_extent_buffer(path->nodes[0]);
3944 path->nodes[0] = left;
3945 path->slots[1] -= 1;
3947 btrfs_tree_unlock(left);
3948 free_extent_buffer(left);
3949 path->slots[0] -= push_items;
3951 BUG_ON(path->slots[0] < 0);
3954 btrfs_tree_unlock(left);
3955 free_extent_buffer(left);
3960 * push some data in the path leaf to the left, trying to free up at
3961 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3963 * max_slot can put a limit on how far into the leaf we'll push items. The
3964 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3967 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3968 *root, struct btrfs_path *path, int min_data_size,
3969 int data_size, int empty, u32 max_slot)
3971 struct extent_buffer *right = path->nodes[0];
3972 struct extent_buffer *left;
3978 slot = path->slots[1];
3981 if (!path->nodes[1])
3984 right_nritems = btrfs_header_nritems(right);
3985 if (right_nritems == 0)
3988 btrfs_assert_tree_locked(path->nodes[1]);
3990 left = read_node_slot(root, path->nodes[1], slot - 1);
3994 btrfs_tree_lock(left);
3995 btrfs_set_lock_blocking(left);
3997 free_space = btrfs_leaf_free_space(root, left);
3998 if (free_space < data_size) {
4003 /* cow and double check */
4004 ret = btrfs_cow_block(trans, root, left,
4005 path->nodes[1], slot - 1, &left);
4007 /* we hit -ENOSPC, but it isn't fatal here */
4013 free_space = btrfs_leaf_free_space(root, left);
4014 if (free_space < data_size) {
4019 return __push_leaf_left(trans, root, path, min_data_size,
4020 empty, left, free_space, right_nritems,
4023 btrfs_tree_unlock(left);
4024 free_extent_buffer(left);
4029 * split the path's leaf in two, making sure there is at least data_size
4030 * available for the resulting leaf level of the path.
4032 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4033 struct btrfs_root *root,
4034 struct btrfs_path *path,
4035 struct extent_buffer *l,
4036 struct extent_buffer *right,
4037 int slot, int mid, int nritems)
4042 struct btrfs_disk_key disk_key;
4043 struct btrfs_map_token token;
4045 btrfs_init_map_token(&token);
4047 nritems = nritems - mid;
4048 btrfs_set_header_nritems(right, nritems);
4049 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
4051 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4052 btrfs_item_nr_offset(mid),
4053 nritems * sizeof(struct btrfs_item));
4055 copy_extent_buffer(right, l,
4056 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
4057 data_copy_size, btrfs_leaf_data(l) +
4058 leaf_data_end(root, l), data_copy_size);
4060 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
4061 btrfs_item_end_nr(l, mid);
4063 for (i = 0; i < nritems; i++) {
4064 struct btrfs_item *item = btrfs_item_nr(i);
4067 ioff = btrfs_token_item_offset(right, item, &token);
4068 btrfs_set_token_item_offset(right, item,
4069 ioff + rt_data_off, &token);
4072 btrfs_set_header_nritems(l, mid);
4073 btrfs_item_key(right, &disk_key, 0);
4074 insert_ptr(trans, root, path, &disk_key, right->start,
4075 path->slots[1] + 1, 1);
4077 btrfs_mark_buffer_dirty(right);
4078 btrfs_mark_buffer_dirty(l);
4079 BUG_ON(path->slots[0] != slot);
4082 btrfs_tree_unlock(path->nodes[0]);
4083 free_extent_buffer(path->nodes[0]);
4084 path->nodes[0] = right;
4085 path->slots[0] -= mid;
4086 path->slots[1] += 1;
4088 btrfs_tree_unlock(right);
4089 free_extent_buffer(right);
4092 BUG_ON(path->slots[0] < 0);
4096 * double splits happen when we need to insert a big item in the middle
4097 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4098 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4101 * We avoid this by trying to push the items on either side of our target
4102 * into the adjacent leaves. If all goes well we can avoid the double split
4105 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4106 struct btrfs_root *root,
4107 struct btrfs_path *path,
4114 int space_needed = data_size;
4116 slot = path->slots[0];
4117 if (slot < btrfs_header_nritems(path->nodes[0]))
4118 space_needed -= btrfs_leaf_free_space(root, path->nodes[0]);
4121 * try to push all the items after our slot into the
4124 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4131 nritems = btrfs_header_nritems(path->nodes[0]);
4133 * our goal is to get our slot at the start or end of a leaf. If
4134 * we've done so we're done
4136 if (path->slots[0] == 0 || path->slots[0] == nritems)
4139 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4142 /* try to push all the items before our slot into the next leaf */
4143 slot = path->slots[0];
4144 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4157 * split the path's leaf in two, making sure there is at least data_size
4158 * available for the resulting leaf level of the path.
4160 * returns 0 if all went well and < 0 on failure.
4162 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4163 struct btrfs_root *root,
4164 struct btrfs_key *ins_key,
4165 struct btrfs_path *path, int data_size,
4168 struct btrfs_disk_key disk_key;
4169 struct extent_buffer *l;
4173 struct extent_buffer *right;
4177 int num_doubles = 0;
4178 int tried_avoid_double = 0;
4181 slot = path->slots[0];
4182 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4183 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
4186 /* first try to make some room by pushing left and right */
4187 if (data_size && path->nodes[1]) {
4188 int space_needed = data_size;
4190 if (slot < btrfs_header_nritems(l))
4191 space_needed -= btrfs_leaf_free_space(root, l);
4193 wret = push_leaf_right(trans, root, path, space_needed,
4194 space_needed, 0, 0);
4198 wret = push_leaf_left(trans, root, path, space_needed,
4199 space_needed, 0, (u32)-1);
4205 /* did the pushes work? */
4206 if (btrfs_leaf_free_space(root, l) >= data_size)
4210 if (!path->nodes[1]) {
4211 ret = insert_new_root(trans, root, path, 1);
4218 slot = path->slots[0];
4219 nritems = btrfs_header_nritems(l);
4220 mid = (nritems + 1) / 2;
4224 leaf_space_used(l, mid, nritems - mid) + data_size >
4225 BTRFS_LEAF_DATA_SIZE(root)) {
4226 if (slot >= nritems) {
4230 if (mid != nritems &&
4231 leaf_space_used(l, mid, nritems - mid) +
4232 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4233 if (data_size && !tried_avoid_double)
4234 goto push_for_double;
4240 if (leaf_space_used(l, 0, mid) + data_size >
4241 BTRFS_LEAF_DATA_SIZE(root)) {
4242 if (!extend && data_size && slot == 0) {
4244 } else if ((extend || !data_size) && slot == 0) {
4248 if (mid != nritems &&
4249 leaf_space_used(l, mid, nritems - mid) +
4250 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4251 if (data_size && !tried_avoid_double)
4252 goto push_for_double;
4260 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4262 btrfs_item_key(l, &disk_key, mid);
4264 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4265 &disk_key, 0, l->start, 0);
4267 return PTR_ERR(right);
4269 root_add_used(root, root->nodesize);
4271 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4272 btrfs_set_header_bytenr(right, right->start);
4273 btrfs_set_header_generation(right, trans->transid);
4274 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4275 btrfs_set_header_owner(right, root->root_key.objectid);
4276 btrfs_set_header_level(right, 0);
4277 write_extent_buffer(right, root->fs_info->fsid,
4278 btrfs_header_fsid(), BTRFS_FSID_SIZE);
4280 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
4281 btrfs_header_chunk_tree_uuid(right),
4286 btrfs_set_header_nritems(right, 0);
4287 insert_ptr(trans, root, path, &disk_key, right->start,
4288 path->slots[1] + 1, 1);
4289 btrfs_tree_unlock(path->nodes[0]);
4290 free_extent_buffer(path->nodes[0]);
4291 path->nodes[0] = right;
4293 path->slots[1] += 1;
4295 btrfs_set_header_nritems(right, 0);
4296 insert_ptr(trans, root, path, &disk_key, right->start,
4298 btrfs_tree_unlock(path->nodes[0]);
4299 free_extent_buffer(path->nodes[0]);
4300 path->nodes[0] = right;
4302 if (path->slots[1] == 0)
4303 fixup_low_keys(root, path, &disk_key, 1);
4305 btrfs_mark_buffer_dirty(right);
4309 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4312 BUG_ON(num_doubles != 0);
4320 push_for_double_split(trans, root, path, data_size);
4321 tried_avoid_double = 1;
4322 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4327 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4328 struct btrfs_root *root,
4329 struct btrfs_path *path, int ins_len)
4331 struct btrfs_key key;
4332 struct extent_buffer *leaf;
4333 struct btrfs_file_extent_item *fi;
4338 leaf = path->nodes[0];
4339 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4341 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4342 key.type != BTRFS_EXTENT_CSUM_KEY);
4344 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4347 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4348 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4349 fi = btrfs_item_ptr(leaf, path->slots[0],
4350 struct btrfs_file_extent_item);
4351 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4353 btrfs_release_path(path);
4355 path->keep_locks = 1;
4356 path->search_for_split = 1;
4357 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4358 path->search_for_split = 0;
4363 leaf = path->nodes[0];
4364 /* if our item isn't there or got smaller, return now */
4365 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4368 /* the leaf has changed, it now has room. return now */
4369 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4372 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4373 fi = btrfs_item_ptr(leaf, path->slots[0],
4374 struct btrfs_file_extent_item);
4375 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4379 btrfs_set_path_blocking(path);
4380 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4384 path->keep_locks = 0;
4385 btrfs_unlock_up_safe(path, 1);
4388 path->keep_locks = 0;
4392 static noinline int split_item(struct btrfs_trans_handle *trans,
4393 struct btrfs_root *root,
4394 struct btrfs_path *path,
4395 struct btrfs_key *new_key,
4396 unsigned long split_offset)
4398 struct extent_buffer *leaf;
4399 struct btrfs_item *item;
4400 struct btrfs_item *new_item;
4406 struct btrfs_disk_key disk_key;
4408 leaf = path->nodes[0];
4409 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4411 btrfs_set_path_blocking(path);
4413 item = btrfs_item_nr(path->slots[0]);
4414 orig_offset = btrfs_item_offset(leaf, item);
4415 item_size = btrfs_item_size(leaf, item);
4417 buf = kmalloc(item_size, GFP_NOFS);
4421 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4422 path->slots[0]), item_size);
4424 slot = path->slots[0] + 1;
4425 nritems = btrfs_header_nritems(leaf);
4426 if (slot != nritems) {
4427 /* shift the items */
4428 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4429 btrfs_item_nr_offset(slot),
4430 (nritems - slot) * sizeof(struct btrfs_item));
4433 btrfs_cpu_key_to_disk(&disk_key, new_key);
4434 btrfs_set_item_key(leaf, &disk_key, slot);
4436 new_item = btrfs_item_nr(slot);
4438 btrfs_set_item_offset(leaf, new_item, orig_offset);
4439 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4441 btrfs_set_item_offset(leaf, item,
4442 orig_offset + item_size - split_offset);
4443 btrfs_set_item_size(leaf, item, split_offset);
4445 btrfs_set_header_nritems(leaf, nritems + 1);
4447 /* write the data for the start of the original item */
4448 write_extent_buffer(leaf, buf,
4449 btrfs_item_ptr_offset(leaf, path->slots[0]),
4452 /* write the data for the new item */
4453 write_extent_buffer(leaf, buf + split_offset,
4454 btrfs_item_ptr_offset(leaf, slot),
4455 item_size - split_offset);
4456 btrfs_mark_buffer_dirty(leaf);
4458 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4464 * This function splits a single item into two items,
4465 * giving 'new_key' to the new item and splitting the
4466 * old one at split_offset (from the start of the item).
4468 * The path may be released by this operation. After
4469 * the split, the path is pointing to the old item. The
4470 * new item is going to be in the same node as the old one.
4472 * Note, the item being split must be smaller enough to live alone on
4473 * a tree block with room for one extra struct btrfs_item
4475 * This allows us to split the item in place, keeping a lock on the
4476 * leaf the entire time.
4478 int btrfs_split_item(struct btrfs_trans_handle *trans,
4479 struct btrfs_root *root,
4480 struct btrfs_path *path,
4481 struct btrfs_key *new_key,
4482 unsigned long split_offset)
4485 ret = setup_leaf_for_split(trans, root, path,
4486 sizeof(struct btrfs_item));
4490 ret = split_item(trans, root, path, new_key, split_offset);
4495 * This function duplicate a item, giving 'new_key' to the new item.
4496 * It guarantees both items live in the same tree leaf and the new item
4497 * is contiguous with the original item.
4499 * This allows us to split file extent in place, keeping a lock on the
4500 * leaf the entire time.
4502 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4503 struct btrfs_root *root,
4504 struct btrfs_path *path,
4505 struct btrfs_key *new_key)
4507 struct extent_buffer *leaf;
4511 leaf = path->nodes[0];
4512 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4513 ret = setup_leaf_for_split(trans, root, path,
4514 item_size + sizeof(struct btrfs_item));
4519 setup_items_for_insert(root, path, new_key, &item_size,
4520 item_size, item_size +
4521 sizeof(struct btrfs_item), 1);
4522 leaf = path->nodes[0];
4523 memcpy_extent_buffer(leaf,
4524 btrfs_item_ptr_offset(leaf, path->slots[0]),
4525 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4531 * make the item pointed to by the path smaller. new_size indicates
4532 * how small to make it, and from_end tells us if we just chop bytes
4533 * off the end of the item or if we shift the item to chop bytes off
4536 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4537 u32 new_size, int from_end)
4540 struct extent_buffer *leaf;
4541 struct btrfs_item *item;
4543 unsigned int data_end;
4544 unsigned int old_data_start;
4545 unsigned int old_size;
4546 unsigned int size_diff;
4548 struct btrfs_map_token token;
4550 btrfs_init_map_token(&token);
4552 leaf = path->nodes[0];
4553 slot = path->slots[0];
4555 old_size = btrfs_item_size_nr(leaf, slot);
4556 if (old_size == new_size)
4559 nritems = btrfs_header_nritems(leaf);
4560 data_end = leaf_data_end(root, leaf);
4562 old_data_start = btrfs_item_offset_nr(leaf, slot);
4564 size_diff = old_size - new_size;
4567 BUG_ON(slot >= nritems);
4570 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4572 /* first correct the data pointers */
4573 for (i = slot; i < nritems; i++) {
4575 item = btrfs_item_nr(i);
4577 ioff = btrfs_token_item_offset(leaf, item, &token);
4578 btrfs_set_token_item_offset(leaf, item,
4579 ioff + size_diff, &token);
4582 /* shift the data */
4584 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4585 data_end + size_diff, btrfs_leaf_data(leaf) +
4586 data_end, old_data_start + new_size - data_end);
4588 struct btrfs_disk_key disk_key;
4591 btrfs_item_key(leaf, &disk_key, slot);
4593 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4595 struct btrfs_file_extent_item *fi;
4597 fi = btrfs_item_ptr(leaf, slot,
4598 struct btrfs_file_extent_item);
4599 fi = (struct btrfs_file_extent_item *)(
4600 (unsigned long)fi - size_diff);
4602 if (btrfs_file_extent_type(leaf, fi) ==
4603 BTRFS_FILE_EXTENT_INLINE) {
4604 ptr = btrfs_item_ptr_offset(leaf, slot);
4605 memmove_extent_buffer(leaf, ptr,
4607 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4611 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4612 data_end + size_diff, btrfs_leaf_data(leaf) +
4613 data_end, old_data_start - data_end);
4615 offset = btrfs_disk_key_offset(&disk_key);
4616 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4617 btrfs_set_item_key(leaf, &disk_key, slot);
4619 fixup_low_keys(root, path, &disk_key, 1);
4622 item = btrfs_item_nr(slot);
4623 btrfs_set_item_size(leaf, item, new_size);
4624 btrfs_mark_buffer_dirty(leaf);
4626 if (btrfs_leaf_free_space(root, leaf) < 0) {
4627 btrfs_print_leaf(root, leaf);
4633 * make the item pointed to by the path bigger, data_size is the added size.
4635 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4639 struct extent_buffer *leaf;
4640 struct btrfs_item *item;
4642 unsigned int data_end;
4643 unsigned int old_data;
4644 unsigned int old_size;
4646 struct btrfs_map_token token;
4648 btrfs_init_map_token(&token);
4650 leaf = path->nodes[0];
4652 nritems = btrfs_header_nritems(leaf);
4653 data_end = leaf_data_end(root, leaf);
4655 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4656 btrfs_print_leaf(root, leaf);
4659 slot = path->slots[0];
4660 old_data = btrfs_item_end_nr(leaf, slot);
4663 if (slot >= nritems) {
4664 btrfs_print_leaf(root, leaf);
4665 btrfs_crit(root->fs_info, "slot %d too large, nritems %d",
4671 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4673 /* first correct the data pointers */
4674 for (i = slot; i < nritems; i++) {
4676 item = btrfs_item_nr(i);
4678 ioff = btrfs_token_item_offset(leaf, item, &token);
4679 btrfs_set_token_item_offset(leaf, item,
4680 ioff - data_size, &token);
4683 /* shift the data */
4684 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4685 data_end - data_size, btrfs_leaf_data(leaf) +
4686 data_end, old_data - data_end);
4688 data_end = old_data;
4689 old_size = btrfs_item_size_nr(leaf, slot);
4690 item = btrfs_item_nr(slot);
4691 btrfs_set_item_size(leaf, item, old_size + data_size);
4692 btrfs_mark_buffer_dirty(leaf);
4694 if (btrfs_leaf_free_space(root, leaf) < 0) {
4695 btrfs_print_leaf(root, leaf);
4701 * this is a helper for btrfs_insert_empty_items, the main goal here is
4702 * to save stack depth by doing the bulk of the work in a function
4703 * that doesn't call btrfs_search_slot
4705 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4706 struct btrfs_key *cpu_key, u32 *data_size,
4707 u32 total_data, u32 total_size, int nr)
4709 struct btrfs_item *item;
4712 unsigned int data_end;
4713 struct btrfs_disk_key disk_key;
4714 struct extent_buffer *leaf;
4716 struct btrfs_map_token token;
4718 if (path->slots[0] == 0) {
4719 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4720 fixup_low_keys(root, path, &disk_key, 1);
4722 btrfs_unlock_up_safe(path, 1);
4724 btrfs_init_map_token(&token);
4726 leaf = path->nodes[0];
4727 slot = path->slots[0];
4729 nritems = btrfs_header_nritems(leaf);
4730 data_end = leaf_data_end(root, leaf);
4732 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4733 btrfs_print_leaf(root, leaf);
4734 btrfs_crit(root->fs_info, "not enough freespace need %u have %d",
4735 total_size, btrfs_leaf_free_space(root, leaf));
4739 if (slot != nritems) {
4740 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4742 if (old_data < data_end) {
4743 btrfs_print_leaf(root, leaf);
4744 btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d",
4745 slot, old_data, data_end);
4749 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4751 /* first correct the data pointers */
4752 for (i = slot; i < nritems; i++) {
4755 item = btrfs_item_nr( i);
4756 ioff = btrfs_token_item_offset(leaf, item, &token);
4757 btrfs_set_token_item_offset(leaf, item,
4758 ioff - total_data, &token);
4760 /* shift the items */
4761 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4762 btrfs_item_nr_offset(slot),
4763 (nritems - slot) * sizeof(struct btrfs_item));
4765 /* shift the data */
4766 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4767 data_end - total_data, btrfs_leaf_data(leaf) +
4768 data_end, old_data - data_end);
4769 data_end = old_data;
4772 /* setup the item for the new data */
4773 for (i = 0; i < nr; i++) {
4774 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4775 btrfs_set_item_key(leaf, &disk_key, slot + i);
4776 item = btrfs_item_nr(slot + i);
4777 btrfs_set_token_item_offset(leaf, item,
4778 data_end - data_size[i], &token);
4779 data_end -= data_size[i];
4780 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4783 btrfs_set_header_nritems(leaf, nritems + nr);
4784 btrfs_mark_buffer_dirty(leaf);
4786 if (btrfs_leaf_free_space(root, leaf) < 0) {
4787 btrfs_print_leaf(root, leaf);
4793 * Given a key and some data, insert items into the tree.
4794 * This does all the path init required, making room in the tree if needed.
4796 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4797 struct btrfs_root *root,
4798 struct btrfs_path *path,
4799 struct btrfs_key *cpu_key, u32 *data_size,
4808 for (i = 0; i < nr; i++)
4809 total_data += data_size[i];
4811 total_size = total_data + (nr * sizeof(struct btrfs_item));
4812 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4818 slot = path->slots[0];
4821 setup_items_for_insert(root, path, cpu_key, data_size,
4822 total_data, total_size, nr);
4827 * Given a key and some data, insert an item into the tree.
4828 * This does all the path init required, making room in the tree if needed.
4830 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4831 *root, struct btrfs_key *cpu_key, void *data, u32
4835 struct btrfs_path *path;
4836 struct extent_buffer *leaf;
4839 path = btrfs_alloc_path();
4842 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4844 leaf = path->nodes[0];
4845 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4846 write_extent_buffer(leaf, data, ptr, data_size);
4847 btrfs_mark_buffer_dirty(leaf);
4849 btrfs_free_path(path);
4854 * delete the pointer from a given node.
4856 * the tree should have been previously balanced so the deletion does not
4859 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4860 int level, int slot)
4862 struct extent_buffer *parent = path->nodes[level];
4866 nritems = btrfs_header_nritems(parent);
4867 if (slot != nritems - 1) {
4869 tree_mod_log_eb_move(root->fs_info, parent, slot,
4870 slot + 1, nritems - slot - 1);
4871 memmove_extent_buffer(parent,
4872 btrfs_node_key_ptr_offset(slot),
4873 btrfs_node_key_ptr_offset(slot + 1),
4874 sizeof(struct btrfs_key_ptr) *
4875 (nritems - slot - 1));
4877 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4878 MOD_LOG_KEY_REMOVE, GFP_NOFS);
4883 btrfs_set_header_nritems(parent, nritems);
4884 if (nritems == 0 && parent == root->node) {
4885 BUG_ON(btrfs_header_level(root->node) != 1);
4886 /* just turn the root into a leaf and break */
4887 btrfs_set_header_level(root->node, 0);
4888 } else if (slot == 0) {
4889 struct btrfs_disk_key disk_key;
4891 btrfs_node_key(parent, &disk_key, 0);
4892 fixup_low_keys(root, path, &disk_key, level + 1);
4894 btrfs_mark_buffer_dirty(parent);
4898 * a helper function to delete the leaf pointed to by path->slots[1] and
4901 * This deletes the pointer in path->nodes[1] and frees the leaf
4902 * block extent. zero is returned if it all worked out, < 0 otherwise.
4904 * The path must have already been setup for deleting the leaf, including
4905 * all the proper balancing. path->nodes[1] must be locked.
4907 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4908 struct btrfs_root *root,
4909 struct btrfs_path *path,
4910 struct extent_buffer *leaf)
4912 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4913 del_ptr(root, path, 1, path->slots[1]);
4916 * btrfs_free_extent is expensive, we want to make sure we
4917 * aren't holding any locks when we call it
4919 btrfs_unlock_up_safe(path, 0);
4921 root_sub_used(root, leaf->len);
4923 extent_buffer_get(leaf);
4924 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4925 free_extent_buffer_stale(leaf);
4928 * delete the item at the leaf level in path. If that empties
4929 * the leaf, remove it from the tree
4931 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4932 struct btrfs_path *path, int slot, int nr)
4934 struct extent_buffer *leaf;
4935 struct btrfs_item *item;
4942 struct btrfs_map_token token;
4944 btrfs_init_map_token(&token);
4946 leaf = path->nodes[0];
4947 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4949 for (i = 0; i < nr; i++)
4950 dsize += btrfs_item_size_nr(leaf, slot + i);
4952 nritems = btrfs_header_nritems(leaf);
4954 if (slot + nr != nritems) {
4955 int data_end = leaf_data_end(root, leaf);
4957 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4959 btrfs_leaf_data(leaf) + data_end,
4960 last_off - data_end);
4962 for (i = slot + nr; i < nritems; i++) {
4965 item = btrfs_item_nr(i);
4966 ioff = btrfs_token_item_offset(leaf, item, &token);
4967 btrfs_set_token_item_offset(leaf, item,
4968 ioff + dsize, &token);
4971 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4972 btrfs_item_nr_offset(slot + nr),
4973 sizeof(struct btrfs_item) *
4974 (nritems - slot - nr));
4976 btrfs_set_header_nritems(leaf, nritems - nr);
4979 /* delete the leaf if we've emptied it */
4981 if (leaf == root->node) {
4982 btrfs_set_header_level(leaf, 0);
4984 btrfs_set_path_blocking(path);
4985 clean_tree_block(trans, root, leaf);
4986 btrfs_del_leaf(trans, root, path, leaf);
4989 int used = leaf_space_used(leaf, 0, nritems);
4991 struct btrfs_disk_key disk_key;
4993 btrfs_item_key(leaf, &disk_key, 0);
4994 fixup_low_keys(root, path, &disk_key, 1);
4997 /* delete the leaf if it is mostly empty */
4998 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
4999 /* push_leaf_left fixes the path.
5000 * make sure the path still points to our leaf
5001 * for possible call to del_ptr below
5003 slot = path->slots[1];
5004 extent_buffer_get(leaf);
5006 btrfs_set_path_blocking(path);
5007 wret = push_leaf_left(trans, root, path, 1, 1,
5009 if (wret < 0 && wret != -ENOSPC)
5012 if (path->nodes[0] == leaf &&
5013 btrfs_header_nritems(leaf)) {
5014 wret = push_leaf_right(trans, root, path, 1,
5016 if (wret < 0 && wret != -ENOSPC)
5020 if (btrfs_header_nritems(leaf) == 0) {
5021 path->slots[1] = slot;
5022 btrfs_del_leaf(trans, root, path, leaf);
5023 free_extent_buffer(leaf);
5026 /* if we're still in the path, make sure
5027 * we're dirty. Otherwise, one of the
5028 * push_leaf functions must have already
5029 * dirtied this buffer
5031 if (path->nodes[0] == leaf)
5032 btrfs_mark_buffer_dirty(leaf);
5033 free_extent_buffer(leaf);
5036 btrfs_mark_buffer_dirty(leaf);
5043 * search the tree again to find a leaf with lesser keys
5044 * returns 0 if it found something or 1 if there are no lesser leaves.
5045 * returns < 0 on io errors.
5047 * This may release the path, and so you may lose any locks held at the
5050 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5052 struct btrfs_key key;
5053 struct btrfs_disk_key found_key;
5056 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5058 if (key.offset > 0) {
5060 } else if (key.type > 0) {
5062 key.offset = (u64)-1;
5063 } else if (key.objectid > 0) {
5066 key.offset = (u64)-1;
5071 btrfs_release_path(path);
5072 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5075 btrfs_item_key(path->nodes[0], &found_key, 0);
5076 ret = comp_keys(&found_key, &key);
5078 * We might have had an item with the previous key in the tree right
5079 * before we released our path. And after we released our path, that
5080 * item might have been pushed to the first slot (0) of the leaf we
5081 * were holding due to a tree balance. Alternatively, an item with the
5082 * previous key can exist as the only element of a leaf (big fat item).
5083 * Therefore account for these 2 cases, so that our callers (like
5084 * btrfs_previous_item) don't miss an existing item with a key matching
5085 * the previous key we computed above.
5093 * A helper function to walk down the tree starting at min_key, and looking
5094 * for nodes or leaves that are have a minimum transaction id.
5095 * This is used by the btree defrag code, and tree logging
5097 * This does not cow, but it does stuff the starting key it finds back
5098 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5099 * key and get a writable path.
5101 * This does lock as it descends, and path->keep_locks should be set
5102 * to 1 by the caller.
5104 * This honors path->lowest_level to prevent descent past a given level
5107 * min_trans indicates the oldest transaction that you are interested
5108 * in walking through. Any nodes or leaves older than min_trans are
5109 * skipped over (without reading them).
5111 * returns zero if something useful was found, < 0 on error and 1 if there
5112 * was nothing in the tree that matched the search criteria.
5114 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5115 struct btrfs_path *path,
5118 struct extent_buffer *cur;
5119 struct btrfs_key found_key;
5125 int keep_locks = path->keep_locks;
5127 path->keep_locks = 1;
5129 cur = btrfs_read_lock_root_node(root);
5130 level = btrfs_header_level(cur);
5131 WARN_ON(path->nodes[level]);
5132 path->nodes[level] = cur;
5133 path->locks[level] = BTRFS_READ_LOCK;
5135 if (btrfs_header_generation(cur) < min_trans) {
5140 nritems = btrfs_header_nritems(cur);
5141 level = btrfs_header_level(cur);
5142 sret = bin_search(cur, min_key, level, &slot);
5144 /* at the lowest level, we're done, setup the path and exit */
5145 if (level == path->lowest_level) {
5146 if (slot >= nritems)
5149 path->slots[level] = slot;
5150 btrfs_item_key_to_cpu(cur, &found_key, slot);
5153 if (sret && slot > 0)
5156 * check this node pointer against the min_trans parameters.
5157 * If it is too old, old, skip to the next one.
5159 while (slot < nritems) {
5162 gen = btrfs_node_ptr_generation(cur, slot);
5163 if (gen < min_trans) {
5171 * we didn't find a candidate key in this node, walk forward
5172 * and find another one
5174 if (slot >= nritems) {
5175 path->slots[level] = slot;
5176 btrfs_set_path_blocking(path);
5177 sret = btrfs_find_next_key(root, path, min_key, level,
5180 btrfs_release_path(path);
5186 /* save our key for returning back */
5187 btrfs_node_key_to_cpu(cur, &found_key, slot);
5188 path->slots[level] = slot;
5189 if (level == path->lowest_level) {
5193 btrfs_set_path_blocking(path);
5194 cur = read_node_slot(root, cur, slot);
5195 BUG_ON(!cur); /* -ENOMEM */
5197 btrfs_tree_read_lock(cur);
5199 path->locks[level - 1] = BTRFS_READ_LOCK;
5200 path->nodes[level - 1] = cur;
5201 unlock_up(path, level, 1, 0, NULL);
5202 btrfs_clear_path_blocking(path, NULL, 0);
5205 path->keep_locks = keep_locks;
5207 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5208 btrfs_set_path_blocking(path);
5209 memcpy(min_key, &found_key, sizeof(found_key));
5214 static void tree_move_down(struct btrfs_root *root,
5215 struct btrfs_path *path,
5216 int *level, int root_level)
5218 BUG_ON(*level == 0);
5219 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
5220 path->slots[*level]);
5221 path->slots[*level - 1] = 0;
5225 static int tree_move_next_or_upnext(struct btrfs_root *root,
5226 struct btrfs_path *path,
5227 int *level, int root_level)
5231 nritems = btrfs_header_nritems(path->nodes[*level]);
5233 path->slots[*level]++;
5235 while (path->slots[*level] >= nritems) {
5236 if (*level == root_level)
5240 path->slots[*level] = 0;
5241 free_extent_buffer(path->nodes[*level]);
5242 path->nodes[*level] = NULL;
5244 path->slots[*level]++;
5246 nritems = btrfs_header_nritems(path->nodes[*level]);
5253 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5256 static int tree_advance(struct btrfs_root *root,
5257 struct btrfs_path *path,
5258 int *level, int root_level,
5260 struct btrfs_key *key)
5264 if (*level == 0 || !allow_down) {
5265 ret = tree_move_next_or_upnext(root, path, level, root_level);
5267 tree_move_down(root, path, level, root_level);
5272 btrfs_item_key_to_cpu(path->nodes[*level], key,
5273 path->slots[*level]);
5275 btrfs_node_key_to_cpu(path->nodes[*level], key,
5276 path->slots[*level]);
5281 static int tree_compare_item(struct btrfs_root *left_root,
5282 struct btrfs_path *left_path,
5283 struct btrfs_path *right_path,
5288 unsigned long off1, off2;
5290 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5291 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5295 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5296 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5297 right_path->slots[0]);
5299 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5301 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5308 #define ADVANCE_ONLY_NEXT -1
5311 * This function compares two trees and calls the provided callback for
5312 * every changed/new/deleted item it finds.
5313 * If shared tree blocks are encountered, whole subtrees are skipped, making
5314 * the compare pretty fast on snapshotted subvolumes.
5316 * This currently works on commit roots only. As commit roots are read only,
5317 * we don't do any locking. The commit roots are protected with transactions.
5318 * Transactions are ended and rejoined when a commit is tried in between.
5320 * This function checks for modifications done to the trees while comparing.
5321 * If it detects a change, it aborts immediately.
5323 int btrfs_compare_trees(struct btrfs_root *left_root,
5324 struct btrfs_root *right_root,
5325 btrfs_changed_cb_t changed_cb, void *ctx)
5329 struct btrfs_path *left_path = NULL;
5330 struct btrfs_path *right_path = NULL;
5331 struct btrfs_key left_key;
5332 struct btrfs_key right_key;
5333 char *tmp_buf = NULL;
5334 int left_root_level;
5335 int right_root_level;
5338 int left_end_reached;
5339 int right_end_reached;
5347 left_path = btrfs_alloc_path();
5352 right_path = btrfs_alloc_path();
5358 tmp_buf = kmalloc(left_root->nodesize, GFP_NOFS);
5364 left_path->search_commit_root = 1;
5365 left_path->skip_locking = 1;
5366 right_path->search_commit_root = 1;
5367 right_path->skip_locking = 1;
5370 * Strategy: Go to the first items of both trees. Then do
5372 * If both trees are at level 0
5373 * Compare keys of current items
5374 * If left < right treat left item as new, advance left tree
5376 * If left > right treat right item as deleted, advance right tree
5378 * If left == right do deep compare of items, treat as changed if
5379 * needed, advance both trees and repeat
5380 * If both trees are at the same level but not at level 0
5381 * Compare keys of current nodes/leafs
5382 * If left < right advance left tree and repeat
5383 * If left > right advance right tree and repeat
5384 * If left == right compare blockptrs of the next nodes/leafs
5385 * If they match advance both trees but stay at the same level
5387 * If they don't match advance both trees while allowing to go
5389 * If tree levels are different
5390 * Advance the tree that needs it and repeat
5392 * Advancing a tree means:
5393 * If we are at level 0, try to go to the next slot. If that's not
5394 * possible, go one level up and repeat. Stop when we found a level
5395 * where we could go to the next slot. We may at this point be on a
5398 * If we are not at level 0 and not on shared tree blocks, go one
5401 * If we are not at level 0 and on shared tree blocks, go one slot to
5402 * the right if possible or go up and right.
5405 down_read(&left_root->fs_info->commit_root_sem);
5406 left_level = btrfs_header_level(left_root->commit_root);
5407 left_root_level = left_level;
5408 left_path->nodes[left_level] = left_root->commit_root;
5409 extent_buffer_get(left_path->nodes[left_level]);
5411 right_level = btrfs_header_level(right_root->commit_root);
5412 right_root_level = right_level;
5413 right_path->nodes[right_level] = right_root->commit_root;
5414 extent_buffer_get(right_path->nodes[right_level]);
5415 up_read(&left_root->fs_info->commit_root_sem);
5417 if (left_level == 0)
5418 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5419 &left_key, left_path->slots[left_level]);
5421 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5422 &left_key, left_path->slots[left_level]);
5423 if (right_level == 0)
5424 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5425 &right_key, right_path->slots[right_level]);
5427 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5428 &right_key, right_path->slots[right_level]);
5430 left_end_reached = right_end_reached = 0;
5431 advance_left = advance_right = 0;
5434 if (advance_left && !left_end_reached) {
5435 ret = tree_advance(left_root, left_path, &left_level,
5437 advance_left != ADVANCE_ONLY_NEXT,
5440 left_end_reached = ADVANCE;
5443 if (advance_right && !right_end_reached) {
5444 ret = tree_advance(right_root, right_path, &right_level,
5446 advance_right != ADVANCE_ONLY_NEXT,
5449 right_end_reached = ADVANCE;
5453 if (left_end_reached && right_end_reached) {
5456 } else if (left_end_reached) {
5457 if (right_level == 0) {
5458 ret = changed_cb(left_root, right_root,
5459 left_path, right_path,
5461 BTRFS_COMPARE_TREE_DELETED,
5466 advance_right = ADVANCE;
5468 } else if (right_end_reached) {
5469 if (left_level == 0) {
5470 ret = changed_cb(left_root, right_root,
5471 left_path, right_path,
5473 BTRFS_COMPARE_TREE_NEW,
5478 advance_left = ADVANCE;
5482 if (left_level == 0 && right_level == 0) {
5483 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5485 ret = changed_cb(left_root, right_root,
5486 left_path, right_path,
5488 BTRFS_COMPARE_TREE_NEW,
5492 advance_left = ADVANCE;
5493 } else if (cmp > 0) {
5494 ret = changed_cb(left_root, right_root,
5495 left_path, right_path,
5497 BTRFS_COMPARE_TREE_DELETED,
5501 advance_right = ADVANCE;
5503 enum btrfs_compare_tree_result result;
5505 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5506 ret = tree_compare_item(left_root, left_path,
5507 right_path, tmp_buf);
5509 result = BTRFS_COMPARE_TREE_CHANGED;
5511 result = BTRFS_COMPARE_TREE_SAME;
5512 ret = changed_cb(left_root, right_root,
5513 left_path, right_path,
5514 &left_key, result, ctx);
5517 advance_left = ADVANCE;
5518 advance_right = ADVANCE;
5520 } else if (left_level == right_level) {
5521 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5523 advance_left = ADVANCE;
5524 } else if (cmp > 0) {
5525 advance_right = ADVANCE;
5527 left_blockptr = btrfs_node_blockptr(
5528 left_path->nodes[left_level],
5529 left_path->slots[left_level]);
5530 right_blockptr = btrfs_node_blockptr(
5531 right_path->nodes[right_level],
5532 right_path->slots[right_level]);
5533 left_gen = btrfs_node_ptr_generation(
5534 left_path->nodes[left_level],
5535 left_path->slots[left_level]);
5536 right_gen = btrfs_node_ptr_generation(
5537 right_path->nodes[right_level],
5538 right_path->slots[right_level]);
5539 if (left_blockptr == right_blockptr &&
5540 left_gen == right_gen) {
5542 * As we're on a shared block, don't
5543 * allow to go deeper.
5545 advance_left = ADVANCE_ONLY_NEXT;
5546 advance_right = ADVANCE_ONLY_NEXT;
5548 advance_left = ADVANCE;
5549 advance_right = ADVANCE;
5552 } else if (left_level < right_level) {
5553 advance_right = ADVANCE;
5555 advance_left = ADVANCE;
5560 btrfs_free_path(left_path);
5561 btrfs_free_path(right_path);
5567 * this is similar to btrfs_next_leaf, but does not try to preserve
5568 * and fixup the path. It looks for and returns the next key in the
5569 * tree based on the current path and the min_trans parameters.
5571 * 0 is returned if another key is found, < 0 if there are any errors
5572 * and 1 is returned if there are no higher keys in the tree
5574 * path->keep_locks should be set to 1 on the search made before
5575 * calling this function.
5577 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5578 struct btrfs_key *key, int level, u64 min_trans)
5581 struct extent_buffer *c;
5583 WARN_ON(!path->keep_locks);
5584 while (level < BTRFS_MAX_LEVEL) {
5585 if (!path->nodes[level])
5588 slot = path->slots[level] + 1;
5589 c = path->nodes[level];
5591 if (slot >= btrfs_header_nritems(c)) {
5594 struct btrfs_key cur_key;
5595 if (level + 1 >= BTRFS_MAX_LEVEL ||
5596 !path->nodes[level + 1])
5599 if (path->locks[level + 1]) {
5604 slot = btrfs_header_nritems(c) - 1;
5606 btrfs_item_key_to_cpu(c, &cur_key, slot);
5608 btrfs_node_key_to_cpu(c, &cur_key, slot);
5610 orig_lowest = path->lowest_level;
5611 btrfs_release_path(path);
5612 path->lowest_level = level;
5613 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5615 path->lowest_level = orig_lowest;
5619 c = path->nodes[level];
5620 slot = path->slots[level];
5627 btrfs_item_key_to_cpu(c, key, slot);
5629 u64 gen = btrfs_node_ptr_generation(c, slot);
5631 if (gen < min_trans) {
5635 btrfs_node_key_to_cpu(c, key, slot);
5643 * search the tree again to find a leaf with greater keys
5644 * returns 0 if it found something or 1 if there are no greater leaves.
5645 * returns < 0 on io errors.
5647 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5649 return btrfs_next_old_leaf(root, path, 0);
5652 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5657 struct extent_buffer *c;
5658 struct extent_buffer *next;
5659 struct btrfs_key key;
5662 int old_spinning = path->leave_spinning;
5663 int next_rw_lock = 0;
5665 nritems = btrfs_header_nritems(path->nodes[0]);
5669 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5674 btrfs_release_path(path);
5676 path->keep_locks = 1;
5677 path->leave_spinning = 1;
5680 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5682 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5683 path->keep_locks = 0;
5688 nritems = btrfs_header_nritems(path->nodes[0]);
5690 * by releasing the path above we dropped all our locks. A balance
5691 * could have added more items next to the key that used to be
5692 * at the very end of the block. So, check again here and
5693 * advance the path if there are now more items available.
5695 if (nritems > 0 && path->slots[0] < nritems - 1) {
5702 * So the above check misses one case:
5703 * - after releasing the path above, someone has removed the item that
5704 * used to be at the very end of the block, and balance between leafs
5705 * gets another one with bigger key.offset to replace it.
5707 * This one should be returned as well, or we can get leaf corruption
5708 * later(esp. in __btrfs_drop_extents()).
5710 * And a bit more explanation about this check,
5711 * with ret > 0, the key isn't found, the path points to the slot
5712 * where it should be inserted, so the path->slots[0] item must be the
5715 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5720 while (level < BTRFS_MAX_LEVEL) {
5721 if (!path->nodes[level]) {
5726 slot = path->slots[level] + 1;
5727 c = path->nodes[level];
5728 if (slot >= btrfs_header_nritems(c)) {
5730 if (level == BTRFS_MAX_LEVEL) {
5738 btrfs_tree_unlock_rw(next, next_rw_lock);
5739 free_extent_buffer(next);
5743 next_rw_lock = path->locks[level];
5744 ret = read_block_for_search(NULL, root, path, &next, level,
5750 btrfs_release_path(path);
5754 if (!path->skip_locking) {
5755 ret = btrfs_try_tree_read_lock(next);
5756 if (!ret && time_seq) {
5758 * If we don't get the lock, we may be racing
5759 * with push_leaf_left, holding that lock while
5760 * itself waiting for the leaf we've currently
5761 * locked. To solve this situation, we give up
5762 * on our lock and cycle.
5764 free_extent_buffer(next);
5765 btrfs_release_path(path);
5770 btrfs_set_path_blocking(path);
5771 btrfs_tree_read_lock(next);
5772 btrfs_clear_path_blocking(path, next,
5775 next_rw_lock = BTRFS_READ_LOCK;
5779 path->slots[level] = slot;
5782 c = path->nodes[level];
5783 if (path->locks[level])
5784 btrfs_tree_unlock_rw(c, path->locks[level]);
5786 free_extent_buffer(c);
5787 path->nodes[level] = next;
5788 path->slots[level] = 0;
5789 if (!path->skip_locking)
5790 path->locks[level] = next_rw_lock;
5794 ret = read_block_for_search(NULL, root, path, &next, level,
5800 btrfs_release_path(path);
5804 if (!path->skip_locking) {
5805 ret = btrfs_try_tree_read_lock(next);
5807 btrfs_set_path_blocking(path);
5808 btrfs_tree_read_lock(next);
5809 btrfs_clear_path_blocking(path, next,
5812 next_rw_lock = BTRFS_READ_LOCK;
5817 unlock_up(path, 0, 1, 0, NULL);
5818 path->leave_spinning = old_spinning;
5820 btrfs_set_path_blocking(path);
5826 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5827 * searching until it gets past min_objectid or finds an item of 'type'
5829 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5831 int btrfs_previous_item(struct btrfs_root *root,
5832 struct btrfs_path *path, u64 min_objectid,
5835 struct btrfs_key found_key;
5836 struct extent_buffer *leaf;
5841 if (path->slots[0] == 0) {
5842 btrfs_set_path_blocking(path);
5843 ret = btrfs_prev_leaf(root, path);
5849 leaf = path->nodes[0];
5850 nritems = btrfs_header_nritems(leaf);
5853 if (path->slots[0] == nritems)
5856 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5857 if (found_key.objectid < min_objectid)
5859 if (found_key.type == type)
5861 if (found_key.objectid == min_objectid &&
5862 found_key.type < type)
5869 * search in extent tree to find a previous Metadata/Data extent item with
5872 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5874 int btrfs_previous_extent_item(struct btrfs_root *root,
5875 struct btrfs_path *path, u64 min_objectid)
5877 struct btrfs_key found_key;
5878 struct extent_buffer *leaf;
5883 if (path->slots[0] == 0) {
5884 btrfs_set_path_blocking(path);
5885 ret = btrfs_prev_leaf(root, path);
5891 leaf = path->nodes[0];
5892 nritems = btrfs_header_nritems(leaf);
5895 if (path->slots[0] == nritems)
5898 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5899 if (found_key.objectid < min_objectid)
5901 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5902 found_key.type == BTRFS_METADATA_ITEM_KEY)
5904 if (found_key.objectid == min_objectid &&
5905 found_key.type < BTRFS_EXTENT_ITEM_KEY)