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>
22 #include <linux/vmalloc.h>
25 #include "transaction.h"
26 #include "print-tree.h"
29 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
30 *root, struct btrfs_path *path, int level);
31 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
32 *root, struct btrfs_key *ins_key,
33 struct btrfs_path *path, int data_size, int extend);
34 static int push_node_left(struct btrfs_trans_handle *trans,
35 struct btrfs_root *root, struct extent_buffer *dst,
36 struct extent_buffer *src, int empty);
37 static int balance_node_right(struct btrfs_trans_handle *trans,
38 struct btrfs_root *root,
39 struct extent_buffer *dst_buf,
40 struct extent_buffer *src_buf);
41 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
43 static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
44 struct extent_buffer *eb);
46 struct btrfs_path *btrfs_alloc_path(void)
48 struct btrfs_path *path;
49 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
54 * set all locked nodes in the path to blocking locks. This should
55 * be done before scheduling
57 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
60 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
61 if (!p->nodes[i] || !p->locks[i])
63 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
64 if (p->locks[i] == BTRFS_READ_LOCK)
65 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
66 else if (p->locks[i] == BTRFS_WRITE_LOCK)
67 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
72 * reset all the locked nodes in the patch to spinning locks.
74 * held is used to keep lockdep happy, when lockdep is enabled
75 * we set held to a blocking lock before we go around and
76 * retake all the spinlocks in the path. You can safely use NULL
79 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
80 struct extent_buffer *held, int held_rw)
85 btrfs_set_lock_blocking_rw(held, held_rw);
86 if (held_rw == BTRFS_WRITE_LOCK)
87 held_rw = BTRFS_WRITE_LOCK_BLOCKING;
88 else if (held_rw == BTRFS_READ_LOCK)
89 held_rw = BTRFS_READ_LOCK_BLOCKING;
91 btrfs_set_path_blocking(p);
93 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
94 if (p->nodes[i] && p->locks[i]) {
95 btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
96 if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
97 p->locks[i] = BTRFS_WRITE_LOCK;
98 else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
99 p->locks[i] = BTRFS_READ_LOCK;
104 btrfs_clear_lock_blocking_rw(held, held_rw);
107 /* this also releases the path */
108 void btrfs_free_path(struct btrfs_path *p)
112 btrfs_release_path(p);
113 kmem_cache_free(btrfs_path_cachep, p);
117 * path release drops references on the extent buffers in the path
118 * and it drops any locks held by this path
120 * It is safe to call this on paths that no locks or extent buffers held.
122 noinline void btrfs_release_path(struct btrfs_path *p)
126 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
131 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
134 free_extent_buffer(p->nodes[i]);
140 * safely gets a reference on the root node of a tree. A lock
141 * is not taken, so a concurrent writer may put a different node
142 * at the root of the tree. See btrfs_lock_root_node for the
145 * The extent buffer returned by this has a reference taken, so
146 * it won't disappear. It may stop being the root of the tree
147 * at any time because there are no locks held.
149 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
151 struct extent_buffer *eb;
155 eb = rcu_dereference(root->node);
158 * RCU really hurts here, we could free up the root node because
159 * it was COWed but we may not get the new root node yet so do
160 * the inc_not_zero dance and if it doesn't work then
161 * synchronize_rcu and try again.
163 if (atomic_inc_not_zero(&eb->refs)) {
173 /* loop around taking references on and locking the root node of the
174 * tree until you end up with a lock on the root. A locked buffer
175 * is returned, with a reference held.
177 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
179 struct extent_buffer *eb;
182 eb = btrfs_root_node(root);
184 if (eb == root->node)
186 btrfs_tree_unlock(eb);
187 free_extent_buffer(eb);
192 /* loop around taking references on and locking the root node of the
193 * tree until you end up with a lock on the root. A locked buffer
194 * is returned, with a reference held.
196 static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
198 struct extent_buffer *eb;
201 eb = btrfs_root_node(root);
202 btrfs_tree_read_lock(eb);
203 if (eb == root->node)
205 btrfs_tree_read_unlock(eb);
206 free_extent_buffer(eb);
211 /* cowonly root (everything not a reference counted cow subvolume), just get
212 * put onto a simple dirty list. transaction.c walks this to make sure they
213 * get properly updated on disk.
215 static void add_root_to_dirty_list(struct btrfs_root *root)
217 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
218 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
221 spin_lock(&root->fs_info->trans_lock);
222 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
223 /* Want the extent tree to be the last on the list */
224 if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID)
225 list_move_tail(&root->dirty_list,
226 &root->fs_info->dirty_cowonly_roots);
228 list_move(&root->dirty_list,
229 &root->fs_info->dirty_cowonly_roots);
231 spin_unlock(&root->fs_info->trans_lock);
235 * used by snapshot creation to make a copy of a root for a tree with
236 * a given objectid. The buffer with the new root node is returned in
237 * cow_ret, and this func returns zero on success or a negative error code.
239 int btrfs_copy_root(struct btrfs_trans_handle *trans,
240 struct btrfs_root *root,
241 struct extent_buffer *buf,
242 struct extent_buffer **cow_ret, u64 new_root_objectid)
244 struct extent_buffer *cow;
247 struct btrfs_disk_key disk_key;
249 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
250 trans->transid != root->fs_info->running_transaction->transid);
251 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
252 trans->transid != root->last_trans);
254 level = btrfs_header_level(buf);
256 btrfs_item_key(buf, &disk_key, 0);
258 btrfs_node_key(buf, &disk_key, 0);
260 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
261 &disk_key, level, buf->start, 0);
265 copy_extent_buffer(cow, buf, 0, 0, cow->len);
266 btrfs_set_header_bytenr(cow, cow->start);
267 btrfs_set_header_generation(cow, trans->transid);
268 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
269 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
270 BTRFS_HEADER_FLAG_RELOC);
271 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
272 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
274 btrfs_set_header_owner(cow, new_root_objectid);
276 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
279 WARN_ON(btrfs_header_generation(buf) > trans->transid);
280 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
281 ret = btrfs_inc_ref(trans, root, cow, 1);
283 ret = btrfs_inc_ref(trans, root, cow, 0);
288 btrfs_mark_buffer_dirty(cow);
297 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
298 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
300 MOD_LOG_ROOT_REPLACE,
303 struct tree_mod_move {
308 struct tree_mod_root {
313 struct tree_mod_elem {
319 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
322 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
325 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
326 struct btrfs_disk_key key;
329 /* this is used for op == MOD_LOG_MOVE_KEYS */
330 struct tree_mod_move move;
332 /* this is used for op == MOD_LOG_ROOT_REPLACE */
333 struct tree_mod_root old_root;
336 static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
338 read_lock(&fs_info->tree_mod_log_lock);
341 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
343 read_unlock(&fs_info->tree_mod_log_lock);
346 static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
348 write_lock(&fs_info->tree_mod_log_lock);
351 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
353 write_unlock(&fs_info->tree_mod_log_lock);
357 * Pull a new tree mod seq number for our operation.
359 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
361 return atomic64_inc_return(&fs_info->tree_mod_seq);
365 * This adds a new blocker to the tree mod log's blocker list if the @elem
366 * passed does not already have a sequence number set. So when a caller expects
367 * to record tree modifications, it should ensure to set elem->seq to zero
368 * before calling btrfs_get_tree_mod_seq.
369 * Returns a fresh, unused tree log modification sequence number, even if no new
372 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
373 struct seq_list *elem)
375 tree_mod_log_write_lock(fs_info);
376 spin_lock(&fs_info->tree_mod_seq_lock);
378 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
379 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
381 spin_unlock(&fs_info->tree_mod_seq_lock);
382 tree_mod_log_write_unlock(fs_info);
387 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
388 struct seq_list *elem)
390 struct rb_root *tm_root;
391 struct rb_node *node;
392 struct rb_node *next;
393 struct seq_list *cur_elem;
394 struct tree_mod_elem *tm;
395 u64 min_seq = (u64)-1;
396 u64 seq_putting = elem->seq;
401 spin_lock(&fs_info->tree_mod_seq_lock);
402 list_del(&elem->list);
405 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
406 if (cur_elem->seq < min_seq) {
407 if (seq_putting > cur_elem->seq) {
409 * blocker with lower sequence number exists, we
410 * cannot remove anything from the log
412 spin_unlock(&fs_info->tree_mod_seq_lock);
415 min_seq = cur_elem->seq;
418 spin_unlock(&fs_info->tree_mod_seq_lock);
421 * anything that's lower than the lowest existing (read: blocked)
422 * sequence number can be removed from the tree.
424 tree_mod_log_write_lock(fs_info);
425 tm_root = &fs_info->tree_mod_log;
426 for (node = rb_first(tm_root); node; node = next) {
427 next = rb_next(node);
428 tm = container_of(node, struct tree_mod_elem, node);
429 if (tm->seq > min_seq)
431 rb_erase(node, tm_root);
434 tree_mod_log_write_unlock(fs_info);
438 * key order of the log:
439 * node/leaf start address -> sequence
441 * The 'start address' is the logical address of the *new* root node
442 * for root replace operations, or the logical address of the affected
443 * block for all other operations.
445 * Note: must be called with write lock (tree_mod_log_write_lock).
448 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
450 struct rb_root *tm_root;
451 struct rb_node **new;
452 struct rb_node *parent = NULL;
453 struct tree_mod_elem *cur;
457 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
459 tm_root = &fs_info->tree_mod_log;
460 new = &tm_root->rb_node;
462 cur = container_of(*new, struct tree_mod_elem, node);
464 if (cur->logical < tm->logical)
465 new = &((*new)->rb_left);
466 else if (cur->logical > tm->logical)
467 new = &((*new)->rb_right);
468 else if (cur->seq < tm->seq)
469 new = &((*new)->rb_left);
470 else if (cur->seq > tm->seq)
471 new = &((*new)->rb_right);
476 rb_link_node(&tm->node, parent, new);
477 rb_insert_color(&tm->node, tm_root);
482 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
483 * returns zero with the tree_mod_log_lock acquired. The caller must hold
484 * this until all tree mod log insertions are recorded in the rb tree and then
485 * call tree_mod_log_write_unlock() to release.
487 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
488 struct extent_buffer *eb) {
490 if (list_empty(&(fs_info)->tree_mod_seq_list))
492 if (eb && btrfs_header_level(eb) == 0)
495 tree_mod_log_write_lock(fs_info);
496 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
497 tree_mod_log_write_unlock(fs_info);
504 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
505 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
506 struct extent_buffer *eb)
509 if (list_empty(&(fs_info)->tree_mod_seq_list))
511 if (eb && btrfs_header_level(eb) == 0)
517 static struct tree_mod_elem *
518 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
519 enum mod_log_op op, gfp_t flags)
521 struct tree_mod_elem *tm;
523 tm = kzalloc(sizeof(*tm), flags);
527 tm->logical = eb->start;
528 if (op != MOD_LOG_KEY_ADD) {
529 btrfs_node_key(eb, &tm->key, slot);
530 tm->blockptr = btrfs_node_blockptr(eb, slot);
534 tm->generation = btrfs_node_ptr_generation(eb, slot);
535 RB_CLEAR_NODE(&tm->node);
541 tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
542 struct extent_buffer *eb, int slot,
543 enum mod_log_op op, gfp_t flags)
545 struct tree_mod_elem *tm;
548 if (!tree_mod_need_log(fs_info, eb))
551 tm = alloc_tree_mod_elem(eb, slot, op, flags);
555 if (tree_mod_dont_log(fs_info, eb)) {
560 ret = __tree_mod_log_insert(fs_info, tm);
561 tree_mod_log_write_unlock(fs_info);
569 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
570 struct extent_buffer *eb, int dst_slot, int src_slot,
571 int nr_items, gfp_t flags)
573 struct tree_mod_elem *tm = NULL;
574 struct tree_mod_elem **tm_list = NULL;
579 if (!tree_mod_need_log(fs_info, eb))
582 tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), flags);
586 tm = kzalloc(sizeof(*tm), flags);
592 tm->logical = eb->start;
594 tm->move.dst_slot = dst_slot;
595 tm->move.nr_items = nr_items;
596 tm->op = MOD_LOG_MOVE_KEYS;
598 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
599 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
600 MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags);
607 if (tree_mod_dont_log(fs_info, eb))
612 * When we override something during the move, we log these removals.
613 * This can only happen when we move towards the beginning of the
614 * buffer, i.e. dst_slot < src_slot.
616 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
617 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
622 ret = __tree_mod_log_insert(fs_info, tm);
625 tree_mod_log_write_unlock(fs_info);
630 for (i = 0; i < nr_items; i++) {
631 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
632 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
636 tree_mod_log_write_unlock(fs_info);
644 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
645 struct tree_mod_elem **tm_list,
651 for (i = nritems - 1; i >= 0; i--) {
652 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
654 for (j = nritems - 1; j > i; j--)
655 rb_erase(&tm_list[j]->node,
656 &fs_info->tree_mod_log);
665 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
666 struct extent_buffer *old_root,
667 struct extent_buffer *new_root, gfp_t flags,
670 struct tree_mod_elem *tm = NULL;
671 struct tree_mod_elem **tm_list = NULL;
676 if (!tree_mod_need_log(fs_info, NULL))
679 if (log_removal && btrfs_header_level(old_root) > 0) {
680 nritems = btrfs_header_nritems(old_root);
681 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
687 for (i = 0; i < nritems; i++) {
688 tm_list[i] = alloc_tree_mod_elem(old_root, i,
689 MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags);
697 tm = kzalloc(sizeof(*tm), flags);
703 tm->logical = new_root->start;
704 tm->old_root.logical = old_root->start;
705 tm->old_root.level = btrfs_header_level(old_root);
706 tm->generation = btrfs_header_generation(old_root);
707 tm->op = MOD_LOG_ROOT_REPLACE;
709 if (tree_mod_dont_log(fs_info, NULL))
713 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
715 ret = __tree_mod_log_insert(fs_info, tm);
717 tree_mod_log_write_unlock(fs_info);
726 for (i = 0; i < nritems; i++)
735 static struct tree_mod_elem *
736 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
739 struct rb_root *tm_root;
740 struct rb_node *node;
741 struct tree_mod_elem *cur = NULL;
742 struct tree_mod_elem *found = NULL;
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->logical < start) {
750 node = node->rb_left;
751 } else if (cur->logical > start) {
752 node = node->rb_right;
753 } else if (cur->seq < min_seq) {
754 node = node->rb_left;
755 } else if (!smallest) {
756 /* we want the node with the highest seq */
758 BUG_ON(found->seq > cur->seq);
760 node = node->rb_left;
761 } else if (cur->seq > min_seq) {
762 /* we want the node with the smallest seq */
764 BUG_ON(found->seq < cur->seq);
766 node = node->rb_right;
772 tree_mod_log_read_unlock(fs_info);
778 * this returns the element from the log with the smallest time sequence
779 * value that's in the log (the oldest log item). any element with a time
780 * sequence lower than min_seq will be ignored.
782 static struct tree_mod_elem *
783 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
786 return __tree_mod_log_search(fs_info, start, min_seq, 1);
790 * this returns the element from the log with the largest time sequence
791 * value that's in the log (the most recent log item). any element with
792 * a time sequence lower than min_seq will be ignored.
794 static struct tree_mod_elem *
795 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
797 return __tree_mod_log_search(fs_info, start, min_seq, 0);
801 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
802 struct extent_buffer *src, unsigned long dst_offset,
803 unsigned long src_offset, int nr_items)
806 struct tree_mod_elem **tm_list = NULL;
807 struct tree_mod_elem **tm_list_add, **tm_list_rem;
811 if (!tree_mod_need_log(fs_info, NULL))
814 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
817 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
822 tm_list_add = tm_list;
823 tm_list_rem = tm_list + nr_items;
824 for (i = 0; i < nr_items; i++) {
825 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
826 MOD_LOG_KEY_REMOVE, GFP_NOFS);
827 if (!tm_list_rem[i]) {
832 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
833 MOD_LOG_KEY_ADD, GFP_NOFS);
834 if (!tm_list_add[i]) {
840 if (tree_mod_dont_log(fs_info, NULL))
844 for (i = 0; i < nr_items; i++) {
845 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
848 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
853 tree_mod_log_write_unlock(fs_info);
859 for (i = 0; i < nr_items * 2; i++) {
860 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
861 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
865 tree_mod_log_write_unlock(fs_info);
872 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
873 int dst_offset, int src_offset, int nr_items)
876 ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
882 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
883 struct extent_buffer *eb, int slot, int atomic)
887 ret = tree_mod_log_insert_key(fs_info, eb, slot,
889 atomic ? GFP_ATOMIC : GFP_NOFS);
894 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
896 struct tree_mod_elem **tm_list = NULL;
901 if (btrfs_header_level(eb) == 0)
904 if (!tree_mod_need_log(fs_info, NULL))
907 nritems = btrfs_header_nritems(eb);
908 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
912 for (i = 0; i < nritems; i++) {
913 tm_list[i] = alloc_tree_mod_elem(eb, i,
914 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
921 if (tree_mod_dont_log(fs_info, eb))
924 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
925 tree_mod_log_write_unlock(fs_info);
933 for (i = 0; i < nritems; i++)
941 tree_mod_log_set_root_pointer(struct btrfs_root *root,
942 struct extent_buffer *new_root_node,
946 ret = tree_mod_log_insert_root(root->fs_info, root->node,
947 new_root_node, GFP_NOFS, log_removal);
952 * check if the tree block can be shared by multiple trees
954 int btrfs_block_can_be_shared(struct btrfs_root *root,
955 struct extent_buffer *buf)
958 * Tree blocks not in reference counted trees and tree roots
959 * are never shared. If a block was allocated after the last
960 * snapshot and the block was not allocated by tree relocation,
961 * we know the block is not shared.
963 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
964 buf != root->node && buf != root->commit_root &&
965 (btrfs_header_generation(buf) <=
966 btrfs_root_last_snapshot(&root->root_item) ||
967 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
969 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
970 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
971 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
977 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
978 struct btrfs_root *root,
979 struct extent_buffer *buf,
980 struct extent_buffer *cow,
990 * Backrefs update rules:
992 * Always use full backrefs for extent pointers in tree block
993 * allocated by tree relocation.
995 * If a shared tree block is no longer referenced by its owner
996 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
997 * use full backrefs for extent pointers in tree block.
999 * If a tree block is been relocating
1000 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1001 * use full backrefs for extent pointers in tree block.
1002 * The reason for this is some operations (such as drop tree)
1003 * are only allowed for blocks use full backrefs.
1006 if (btrfs_block_can_be_shared(root, buf)) {
1007 ret = btrfs_lookup_extent_info(trans, root, buf->start,
1008 btrfs_header_level(buf), 1,
1014 btrfs_handle_fs_error(root->fs_info, ret, NULL);
1019 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1020 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1021 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1026 owner = btrfs_header_owner(buf);
1027 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
1028 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
1031 if ((owner == root->root_key.objectid ||
1032 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
1033 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1034 ret = btrfs_inc_ref(trans, root, buf, 1);
1035 BUG_ON(ret); /* -ENOMEM */
1037 if (root->root_key.objectid ==
1038 BTRFS_TREE_RELOC_OBJECTID) {
1039 ret = btrfs_dec_ref(trans, root, buf, 0);
1040 BUG_ON(ret); /* -ENOMEM */
1041 ret = btrfs_inc_ref(trans, root, cow, 1);
1042 BUG_ON(ret); /* -ENOMEM */
1044 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
1047 if (root->root_key.objectid ==
1048 BTRFS_TREE_RELOC_OBJECTID)
1049 ret = btrfs_inc_ref(trans, root, cow, 1);
1051 ret = btrfs_inc_ref(trans, root, cow, 0);
1052 BUG_ON(ret); /* -ENOMEM */
1054 if (new_flags != 0) {
1055 int level = btrfs_header_level(buf);
1057 ret = btrfs_set_disk_extent_flags(trans, root,
1060 new_flags, level, 0);
1065 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
1066 if (root->root_key.objectid ==
1067 BTRFS_TREE_RELOC_OBJECTID)
1068 ret = btrfs_inc_ref(trans, root, cow, 1);
1070 ret = btrfs_inc_ref(trans, root, cow, 0);
1071 BUG_ON(ret); /* -ENOMEM */
1072 ret = btrfs_dec_ref(trans, root, buf, 1);
1073 BUG_ON(ret); /* -ENOMEM */
1075 clean_tree_block(trans, root->fs_info, buf);
1082 * does the dirty work in cow of a single block. The parent block (if
1083 * supplied) is updated to point to the new cow copy. The new buffer is marked
1084 * dirty and returned locked. If you modify the block it needs to be marked
1087 * search_start -- an allocation hint for the new block
1089 * empty_size -- a hint that you plan on doing more cow. This is the size in
1090 * bytes the allocator should try to find free next to the block it returns.
1091 * This is just a hint and may be ignored by the allocator.
1093 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1094 struct btrfs_root *root,
1095 struct extent_buffer *buf,
1096 struct extent_buffer *parent, int parent_slot,
1097 struct extent_buffer **cow_ret,
1098 u64 search_start, u64 empty_size)
1100 struct btrfs_disk_key disk_key;
1101 struct extent_buffer *cow;
1104 int unlock_orig = 0;
1107 if (*cow_ret == buf)
1110 btrfs_assert_tree_locked(buf);
1112 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1113 trans->transid != root->fs_info->running_transaction->transid);
1114 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1115 trans->transid != root->last_trans);
1117 level = btrfs_header_level(buf);
1120 btrfs_item_key(buf, &disk_key, 0);
1122 btrfs_node_key(buf, &disk_key, 0);
1124 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1126 parent_start = parent->start;
1132 cow = btrfs_alloc_tree_block(trans, root, parent_start,
1133 root->root_key.objectid, &disk_key, level,
1134 search_start, empty_size);
1136 return PTR_ERR(cow);
1138 /* cow is set to blocking by btrfs_init_new_buffer */
1140 copy_extent_buffer(cow, buf, 0, 0, cow->len);
1141 btrfs_set_header_bytenr(cow, cow->start);
1142 btrfs_set_header_generation(cow, trans->transid);
1143 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1144 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1145 BTRFS_HEADER_FLAG_RELOC);
1146 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1147 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1149 btrfs_set_header_owner(cow, root->root_key.objectid);
1151 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
1154 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1156 btrfs_abort_transaction(trans, root, ret);
1160 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1161 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1163 btrfs_abort_transaction(trans, root, ret);
1168 if (buf == root->node) {
1169 WARN_ON(parent && parent != buf);
1170 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1171 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1172 parent_start = buf->start;
1176 extent_buffer_get(cow);
1177 tree_mod_log_set_root_pointer(root, cow, 1);
1178 rcu_assign_pointer(root->node, cow);
1180 btrfs_free_tree_block(trans, root, buf, parent_start,
1182 free_extent_buffer(buf);
1183 add_root_to_dirty_list(root);
1185 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1186 parent_start = parent->start;
1190 WARN_ON(trans->transid != btrfs_header_generation(parent));
1191 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1192 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1193 btrfs_set_node_blockptr(parent, parent_slot,
1195 btrfs_set_node_ptr_generation(parent, parent_slot,
1197 btrfs_mark_buffer_dirty(parent);
1199 ret = tree_mod_log_free_eb(root->fs_info, buf);
1201 btrfs_abort_transaction(trans, root, ret);
1205 btrfs_free_tree_block(trans, root, buf, parent_start,
1209 btrfs_tree_unlock(buf);
1210 free_extent_buffer_stale(buf);
1211 btrfs_mark_buffer_dirty(cow);
1217 * returns the logical address of the oldest predecessor of the given root.
1218 * entries older than time_seq are ignored.
1220 static struct tree_mod_elem *
1221 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1222 struct extent_buffer *eb_root, u64 time_seq)
1224 struct tree_mod_elem *tm;
1225 struct tree_mod_elem *found = NULL;
1226 u64 root_logical = eb_root->start;
1233 * the very last operation that's logged for a root is the
1234 * replacement operation (if it is replaced at all). this has
1235 * the logical address of the *new* root, making it the very
1236 * first operation that's logged for this root.
1239 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1244 * if there are no tree operation for the oldest root, we simply
1245 * return it. this should only happen if that (old) root is at
1252 * if there's an operation that's not a root replacement, we
1253 * found the oldest version of our root. normally, we'll find a
1254 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1256 if (tm->op != MOD_LOG_ROOT_REPLACE)
1260 root_logical = tm->old_root.logical;
1264 /* if there's no old root to return, return what we found instead */
1272 * tm is a pointer to the first operation to rewind within eb. then, all
1273 * previous operations will be rewound (until we reach something older than
1277 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1278 u64 time_seq, struct tree_mod_elem *first_tm)
1281 struct rb_node *next;
1282 struct tree_mod_elem *tm = first_tm;
1283 unsigned long o_dst;
1284 unsigned long o_src;
1285 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1287 n = btrfs_header_nritems(eb);
1288 tree_mod_log_read_lock(fs_info);
1289 while (tm && tm->seq >= time_seq) {
1291 * all the operations are recorded with the operator used for
1292 * the modification. as we're going backwards, we do the
1293 * opposite of each operation here.
1296 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1297 BUG_ON(tm->slot < n);
1299 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1300 case MOD_LOG_KEY_REMOVE:
1301 btrfs_set_node_key(eb, &tm->key, tm->slot);
1302 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1303 btrfs_set_node_ptr_generation(eb, tm->slot,
1307 case MOD_LOG_KEY_REPLACE:
1308 BUG_ON(tm->slot >= n);
1309 btrfs_set_node_key(eb, &tm->key, tm->slot);
1310 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1311 btrfs_set_node_ptr_generation(eb, tm->slot,
1314 case MOD_LOG_KEY_ADD:
1315 /* if a move operation is needed it's in the log */
1318 case MOD_LOG_MOVE_KEYS:
1319 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1320 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1321 memmove_extent_buffer(eb, o_dst, o_src,
1322 tm->move.nr_items * p_size);
1324 case MOD_LOG_ROOT_REPLACE:
1326 * this operation is special. for roots, this must be
1327 * handled explicitly before rewinding.
1328 * for non-roots, this operation may exist if the node
1329 * was a root: root A -> child B; then A gets empty and
1330 * B is promoted to the new root. in the mod log, we'll
1331 * have a root-replace operation for B, a tree block
1332 * that is no root. we simply ignore that operation.
1336 next = rb_next(&tm->node);
1339 tm = container_of(next, struct tree_mod_elem, node);
1340 if (tm->logical != first_tm->logical)
1343 tree_mod_log_read_unlock(fs_info);
1344 btrfs_set_header_nritems(eb, n);
1348 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1349 * is returned. If rewind operations happen, a fresh buffer is returned. The
1350 * returned buffer is always read-locked. If the returned buffer is not the
1351 * input buffer, the lock on the input buffer is released and the input buffer
1352 * is freed (its refcount is decremented).
1354 static struct extent_buffer *
1355 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1356 struct extent_buffer *eb, u64 time_seq)
1358 struct extent_buffer *eb_rewin;
1359 struct tree_mod_elem *tm;
1364 if (btrfs_header_level(eb) == 0)
1367 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1371 btrfs_set_path_blocking(path);
1372 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1374 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1375 BUG_ON(tm->slot != 0);
1376 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start,
1379 btrfs_tree_read_unlock_blocking(eb);
1380 free_extent_buffer(eb);
1383 btrfs_set_header_bytenr(eb_rewin, eb->start);
1384 btrfs_set_header_backref_rev(eb_rewin,
1385 btrfs_header_backref_rev(eb));
1386 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1387 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1389 eb_rewin = btrfs_clone_extent_buffer(eb);
1391 btrfs_tree_read_unlock_blocking(eb);
1392 free_extent_buffer(eb);
1397 btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
1398 btrfs_tree_read_unlock_blocking(eb);
1399 free_extent_buffer(eb);
1401 extent_buffer_get(eb_rewin);
1402 btrfs_tree_read_lock(eb_rewin);
1403 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1404 WARN_ON(btrfs_header_nritems(eb_rewin) >
1405 BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1411 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1412 * value. If there are no changes, the current root->root_node is returned. If
1413 * anything changed in between, there's a fresh buffer allocated on which the
1414 * rewind operations are done. In any case, the returned buffer is read locked.
1415 * Returns NULL on error (with no locks held).
1417 static inline struct extent_buffer *
1418 get_old_root(struct btrfs_root *root, u64 time_seq)
1420 struct tree_mod_elem *tm;
1421 struct extent_buffer *eb = NULL;
1422 struct extent_buffer *eb_root;
1423 struct extent_buffer *old;
1424 struct tree_mod_root *old_root = NULL;
1425 u64 old_generation = 0;
1428 eb_root = btrfs_read_lock_root_node(root);
1429 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1433 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1434 old_root = &tm->old_root;
1435 old_generation = tm->generation;
1436 logical = old_root->logical;
1438 logical = eb_root->start;
1441 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1442 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1443 btrfs_tree_read_unlock(eb_root);
1444 free_extent_buffer(eb_root);
1445 old = read_tree_block(root, logical, 0);
1446 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1448 free_extent_buffer(old);
1449 btrfs_warn(root->fs_info,
1450 "failed to read tree block %llu from get_old_root", logical);
1452 eb = btrfs_clone_extent_buffer(old);
1453 free_extent_buffer(old);
1455 } else if (old_root) {
1456 btrfs_tree_read_unlock(eb_root);
1457 free_extent_buffer(eb_root);
1458 eb = alloc_dummy_extent_buffer(root->fs_info, logical,
1461 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1462 eb = btrfs_clone_extent_buffer(eb_root);
1463 btrfs_tree_read_unlock_blocking(eb_root);
1464 free_extent_buffer(eb_root);
1469 extent_buffer_get(eb);
1470 btrfs_tree_read_lock(eb);
1472 btrfs_set_header_bytenr(eb, eb->start);
1473 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1474 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1475 btrfs_set_header_level(eb, old_root->level);
1476 btrfs_set_header_generation(eb, old_generation);
1479 __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1481 WARN_ON(btrfs_header_level(eb) != 0);
1482 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1487 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1489 struct tree_mod_elem *tm;
1491 struct extent_buffer *eb_root = btrfs_root_node(root);
1493 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1494 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1495 level = tm->old_root.level;
1497 level = btrfs_header_level(eb_root);
1499 free_extent_buffer(eb_root);
1504 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1505 struct btrfs_root *root,
1506 struct extent_buffer *buf)
1508 if (btrfs_test_is_dummy_root(root))
1511 /* ensure we can see the force_cow */
1515 * We do not need to cow a block if
1516 * 1) this block is not created or changed in this transaction;
1517 * 2) this block does not belong to TREE_RELOC tree;
1518 * 3) the root is not forced COW.
1520 * What is forced COW:
1521 * when we create snapshot during committing the transaction,
1522 * after we've finished coping src root, we must COW the shared
1523 * block to ensure the metadata consistency.
1525 if (btrfs_header_generation(buf) == trans->transid &&
1526 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1527 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1528 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1529 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1535 * cows a single block, see __btrfs_cow_block for the real work.
1536 * This version of it has extra checks so that a block isn't COWed more than
1537 * once per transaction, as long as it hasn't been written yet
1539 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1540 struct btrfs_root *root, struct extent_buffer *buf,
1541 struct extent_buffer *parent, int parent_slot,
1542 struct extent_buffer **cow_ret)
1547 if (trans->transaction != root->fs_info->running_transaction)
1548 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1550 root->fs_info->running_transaction->transid);
1552 if (trans->transid != root->fs_info->generation)
1553 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1554 trans->transid, root->fs_info->generation);
1556 if (!should_cow_block(trans, root, buf)) {
1561 search_start = buf->start & ~((u64)SZ_1G - 1);
1564 btrfs_set_lock_blocking(parent);
1565 btrfs_set_lock_blocking(buf);
1567 ret = __btrfs_cow_block(trans, root, buf, parent,
1568 parent_slot, cow_ret, search_start, 0);
1570 trace_btrfs_cow_block(root, buf, *cow_ret);
1576 * helper function for defrag to decide if two blocks pointed to by a
1577 * node are actually close by
1579 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1581 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1583 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1589 * compare two keys in a memcmp fashion
1591 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1593 struct btrfs_key k1;
1595 btrfs_disk_key_to_cpu(&k1, disk);
1597 return btrfs_comp_cpu_keys(&k1, k2);
1601 * same as comp_keys only with two btrfs_key's
1603 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1605 if (k1->objectid > k2->objectid)
1607 if (k1->objectid < k2->objectid)
1609 if (k1->type > k2->type)
1611 if (k1->type < k2->type)
1613 if (k1->offset > k2->offset)
1615 if (k1->offset < k2->offset)
1621 * this is used by the defrag code to go through all the
1622 * leaves pointed to by a node and reallocate them so that
1623 * disk order is close to key order
1625 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1626 struct btrfs_root *root, struct extent_buffer *parent,
1627 int start_slot, u64 *last_ret,
1628 struct btrfs_key *progress)
1630 struct extent_buffer *cur;
1633 u64 search_start = *last_ret;
1643 int progress_passed = 0;
1644 struct btrfs_disk_key disk_key;
1646 parent_level = btrfs_header_level(parent);
1648 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1649 WARN_ON(trans->transid != root->fs_info->generation);
1651 parent_nritems = btrfs_header_nritems(parent);
1652 blocksize = root->nodesize;
1653 end_slot = parent_nritems - 1;
1655 if (parent_nritems <= 1)
1658 btrfs_set_lock_blocking(parent);
1660 for (i = start_slot; i <= end_slot; i++) {
1663 btrfs_node_key(parent, &disk_key, i);
1664 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1667 progress_passed = 1;
1668 blocknr = btrfs_node_blockptr(parent, i);
1669 gen = btrfs_node_ptr_generation(parent, i);
1670 if (last_block == 0)
1671 last_block = blocknr;
1674 other = btrfs_node_blockptr(parent, i - 1);
1675 close = close_blocks(blocknr, other, blocksize);
1677 if (!close && i < end_slot) {
1678 other = btrfs_node_blockptr(parent, i + 1);
1679 close = close_blocks(blocknr, other, blocksize);
1682 last_block = blocknr;
1686 cur = btrfs_find_tree_block(root->fs_info, blocknr);
1688 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1691 if (!cur || !uptodate) {
1693 cur = read_tree_block(root, blocknr, gen);
1695 return PTR_ERR(cur);
1696 } else if (!extent_buffer_uptodate(cur)) {
1697 free_extent_buffer(cur);
1700 } else if (!uptodate) {
1701 err = btrfs_read_buffer(cur, gen);
1703 free_extent_buffer(cur);
1708 if (search_start == 0)
1709 search_start = last_block;
1711 btrfs_tree_lock(cur);
1712 btrfs_set_lock_blocking(cur);
1713 err = __btrfs_cow_block(trans, root, cur, parent, i,
1716 (end_slot - i) * blocksize));
1718 btrfs_tree_unlock(cur);
1719 free_extent_buffer(cur);
1722 search_start = cur->start;
1723 last_block = cur->start;
1724 *last_ret = search_start;
1725 btrfs_tree_unlock(cur);
1726 free_extent_buffer(cur);
1732 * The leaf data grows from end-to-front in the node.
1733 * this returns the address of the start of the last item,
1734 * which is the stop of the leaf data stack
1736 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1737 struct extent_buffer *leaf)
1739 u32 nr = btrfs_header_nritems(leaf);
1741 return BTRFS_LEAF_DATA_SIZE(root);
1742 return btrfs_item_offset_nr(leaf, nr - 1);
1747 * search for key in the extent_buffer. The items start at offset p,
1748 * and they are item_size apart. There are 'max' items in p.
1750 * the slot in the array is returned via slot, and it points to
1751 * the place where you would insert key if it is not found in
1754 * slot may point to max if the key is bigger than all of the keys
1756 static noinline int generic_bin_search(struct extent_buffer *eb,
1758 int item_size, struct btrfs_key *key,
1765 struct btrfs_disk_key *tmp = NULL;
1766 struct btrfs_disk_key unaligned;
1767 unsigned long offset;
1769 unsigned long map_start = 0;
1770 unsigned long map_len = 0;
1773 while (low < high) {
1774 mid = (low + high) / 2;
1775 offset = p + mid * item_size;
1777 if (!kaddr || offset < map_start ||
1778 (offset + sizeof(struct btrfs_disk_key)) >
1779 map_start + map_len) {
1781 err = map_private_extent_buffer(eb, offset,
1782 sizeof(struct btrfs_disk_key),
1783 &kaddr, &map_start, &map_len);
1786 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1789 read_extent_buffer(eb, &unaligned,
1790 offset, sizeof(unaligned));
1795 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1798 ret = comp_keys(tmp, key);
1814 * simple bin_search frontend that does the right thing for
1817 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1818 int level, int *slot)
1821 return generic_bin_search(eb,
1822 offsetof(struct btrfs_leaf, items),
1823 sizeof(struct btrfs_item),
1824 key, btrfs_header_nritems(eb),
1827 return generic_bin_search(eb,
1828 offsetof(struct btrfs_node, ptrs),
1829 sizeof(struct btrfs_key_ptr),
1830 key, btrfs_header_nritems(eb),
1834 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1835 int level, int *slot)
1837 return bin_search(eb, key, level, slot);
1840 static void root_add_used(struct btrfs_root *root, u32 size)
1842 spin_lock(&root->accounting_lock);
1843 btrfs_set_root_used(&root->root_item,
1844 btrfs_root_used(&root->root_item) + size);
1845 spin_unlock(&root->accounting_lock);
1848 static void root_sub_used(struct btrfs_root *root, u32 size)
1850 spin_lock(&root->accounting_lock);
1851 btrfs_set_root_used(&root->root_item,
1852 btrfs_root_used(&root->root_item) - size);
1853 spin_unlock(&root->accounting_lock);
1856 /* given a node and slot number, this reads the blocks it points to. The
1857 * extent buffer is returned with a reference taken (but unlocked).
1858 * NULL is returned on error.
1860 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1861 struct extent_buffer *parent, int slot)
1863 int level = btrfs_header_level(parent);
1864 struct extent_buffer *eb;
1868 if (slot >= btrfs_header_nritems(parent))
1873 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1874 btrfs_node_ptr_generation(parent, slot));
1875 if (IS_ERR(eb) || !extent_buffer_uptodate(eb)) {
1877 free_extent_buffer(eb);
1885 * node level balancing, used to make sure nodes are in proper order for
1886 * item deletion. We balance from the top down, so we have to make sure
1887 * that a deletion won't leave an node completely empty later on.
1889 static noinline int balance_level(struct btrfs_trans_handle *trans,
1890 struct btrfs_root *root,
1891 struct btrfs_path *path, int level)
1893 struct extent_buffer *right = NULL;
1894 struct extent_buffer *mid;
1895 struct extent_buffer *left = NULL;
1896 struct extent_buffer *parent = NULL;
1900 int orig_slot = path->slots[level];
1906 mid = path->nodes[level];
1908 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1909 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1910 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1912 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1914 if (level < BTRFS_MAX_LEVEL - 1) {
1915 parent = path->nodes[level + 1];
1916 pslot = path->slots[level + 1];
1920 * deal with the case where there is only one pointer in the root
1921 * by promoting the node below to a root
1924 struct extent_buffer *child;
1926 if (btrfs_header_nritems(mid) != 1)
1929 /* promote the child to a root */
1930 child = read_node_slot(root, mid, 0);
1933 btrfs_handle_fs_error(root->fs_info, ret, NULL);
1937 btrfs_tree_lock(child);
1938 btrfs_set_lock_blocking(child);
1939 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1941 btrfs_tree_unlock(child);
1942 free_extent_buffer(child);
1946 tree_mod_log_set_root_pointer(root, child, 1);
1947 rcu_assign_pointer(root->node, child);
1949 add_root_to_dirty_list(root);
1950 btrfs_tree_unlock(child);
1952 path->locks[level] = 0;
1953 path->nodes[level] = NULL;
1954 clean_tree_block(trans, root->fs_info, mid);
1955 btrfs_tree_unlock(mid);
1956 /* once for the path */
1957 free_extent_buffer(mid);
1959 root_sub_used(root, mid->len);
1960 btrfs_free_tree_block(trans, root, mid, 0, 1);
1961 /* once for the root ptr */
1962 free_extent_buffer_stale(mid);
1965 if (btrfs_header_nritems(mid) >
1966 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1969 left = read_node_slot(root, parent, pslot - 1);
1971 btrfs_tree_lock(left);
1972 btrfs_set_lock_blocking(left);
1973 wret = btrfs_cow_block(trans, root, left,
1974 parent, pslot - 1, &left);
1980 right = read_node_slot(root, parent, pslot + 1);
1982 btrfs_tree_lock(right);
1983 btrfs_set_lock_blocking(right);
1984 wret = btrfs_cow_block(trans, root, right,
1985 parent, pslot + 1, &right);
1992 /* first, try to make some room in the middle buffer */
1994 orig_slot += btrfs_header_nritems(left);
1995 wret = push_node_left(trans, root, left, mid, 1);
2001 * then try to empty the right most buffer into the middle
2004 wret = push_node_left(trans, root, mid, right, 1);
2005 if (wret < 0 && wret != -ENOSPC)
2007 if (btrfs_header_nritems(right) == 0) {
2008 clean_tree_block(trans, root->fs_info, right);
2009 btrfs_tree_unlock(right);
2010 del_ptr(root, path, level + 1, pslot + 1);
2011 root_sub_used(root, right->len);
2012 btrfs_free_tree_block(trans, root, right, 0, 1);
2013 free_extent_buffer_stale(right);
2016 struct btrfs_disk_key right_key;
2017 btrfs_node_key(right, &right_key, 0);
2018 tree_mod_log_set_node_key(root->fs_info, parent,
2020 btrfs_set_node_key(parent, &right_key, pslot + 1);
2021 btrfs_mark_buffer_dirty(parent);
2024 if (btrfs_header_nritems(mid) == 1) {
2026 * we're not allowed to leave a node with one item in the
2027 * tree during a delete. A deletion from lower in the tree
2028 * could try to delete the only pointer in this node.
2029 * So, pull some keys from the left.
2030 * There has to be a left pointer at this point because
2031 * otherwise we would have pulled some pointers from the
2036 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2039 wret = balance_node_right(trans, root, mid, left);
2045 wret = push_node_left(trans, root, left, mid, 1);
2051 if (btrfs_header_nritems(mid) == 0) {
2052 clean_tree_block(trans, root->fs_info, mid);
2053 btrfs_tree_unlock(mid);
2054 del_ptr(root, path, level + 1, pslot);
2055 root_sub_used(root, mid->len);
2056 btrfs_free_tree_block(trans, root, mid, 0, 1);
2057 free_extent_buffer_stale(mid);
2060 /* update the parent key to reflect our changes */
2061 struct btrfs_disk_key mid_key;
2062 btrfs_node_key(mid, &mid_key, 0);
2063 tree_mod_log_set_node_key(root->fs_info, parent,
2065 btrfs_set_node_key(parent, &mid_key, pslot);
2066 btrfs_mark_buffer_dirty(parent);
2069 /* update the path */
2071 if (btrfs_header_nritems(left) > orig_slot) {
2072 extent_buffer_get(left);
2073 /* left was locked after cow */
2074 path->nodes[level] = left;
2075 path->slots[level + 1] -= 1;
2076 path->slots[level] = orig_slot;
2078 btrfs_tree_unlock(mid);
2079 free_extent_buffer(mid);
2082 orig_slot -= btrfs_header_nritems(left);
2083 path->slots[level] = orig_slot;
2086 /* double check we haven't messed things up */
2088 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2092 btrfs_tree_unlock(right);
2093 free_extent_buffer(right);
2096 if (path->nodes[level] != left)
2097 btrfs_tree_unlock(left);
2098 free_extent_buffer(left);
2103 /* Node balancing for insertion. Here we only split or push nodes around
2104 * when they are completely full. This is also done top down, so we
2105 * have to be pessimistic.
2107 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2108 struct btrfs_root *root,
2109 struct btrfs_path *path, int level)
2111 struct extent_buffer *right = NULL;
2112 struct extent_buffer *mid;
2113 struct extent_buffer *left = NULL;
2114 struct extent_buffer *parent = NULL;
2118 int orig_slot = path->slots[level];
2123 mid = path->nodes[level];
2124 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2126 if (level < BTRFS_MAX_LEVEL - 1) {
2127 parent = path->nodes[level + 1];
2128 pslot = path->slots[level + 1];
2134 left = read_node_slot(root, parent, pslot - 1);
2136 /* first, try to make some room in the middle buffer */
2140 btrfs_tree_lock(left);
2141 btrfs_set_lock_blocking(left);
2143 left_nr = btrfs_header_nritems(left);
2144 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2147 ret = btrfs_cow_block(trans, root, left, parent,
2152 wret = push_node_left(trans, root,
2159 struct btrfs_disk_key disk_key;
2160 orig_slot += left_nr;
2161 btrfs_node_key(mid, &disk_key, 0);
2162 tree_mod_log_set_node_key(root->fs_info, parent,
2164 btrfs_set_node_key(parent, &disk_key, pslot);
2165 btrfs_mark_buffer_dirty(parent);
2166 if (btrfs_header_nritems(left) > orig_slot) {
2167 path->nodes[level] = left;
2168 path->slots[level + 1] -= 1;
2169 path->slots[level] = orig_slot;
2170 btrfs_tree_unlock(mid);
2171 free_extent_buffer(mid);
2174 btrfs_header_nritems(left);
2175 path->slots[level] = orig_slot;
2176 btrfs_tree_unlock(left);
2177 free_extent_buffer(left);
2181 btrfs_tree_unlock(left);
2182 free_extent_buffer(left);
2184 right = read_node_slot(root, parent, pslot + 1);
2187 * then try to empty the right most buffer into the middle
2192 btrfs_tree_lock(right);
2193 btrfs_set_lock_blocking(right);
2195 right_nr = btrfs_header_nritems(right);
2196 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2199 ret = btrfs_cow_block(trans, root, right,
2205 wret = balance_node_right(trans, root,
2212 struct btrfs_disk_key disk_key;
2214 btrfs_node_key(right, &disk_key, 0);
2215 tree_mod_log_set_node_key(root->fs_info, parent,
2217 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2218 btrfs_mark_buffer_dirty(parent);
2220 if (btrfs_header_nritems(mid) <= orig_slot) {
2221 path->nodes[level] = right;
2222 path->slots[level + 1] += 1;
2223 path->slots[level] = orig_slot -
2224 btrfs_header_nritems(mid);
2225 btrfs_tree_unlock(mid);
2226 free_extent_buffer(mid);
2228 btrfs_tree_unlock(right);
2229 free_extent_buffer(right);
2233 btrfs_tree_unlock(right);
2234 free_extent_buffer(right);
2240 * readahead one full node of leaves, finding things that are close
2241 * to the block in 'slot', and triggering ra on them.
2243 static void reada_for_search(struct btrfs_root *root,
2244 struct btrfs_path *path,
2245 int level, int slot, u64 objectid)
2247 struct extent_buffer *node;
2248 struct btrfs_disk_key disk_key;
2254 struct extent_buffer *eb;
2262 if (!path->nodes[level])
2265 node = path->nodes[level];
2267 search = btrfs_node_blockptr(node, slot);
2268 blocksize = root->nodesize;
2269 eb = btrfs_find_tree_block(root->fs_info, search);
2271 free_extent_buffer(eb);
2277 nritems = btrfs_header_nritems(node);
2281 if (path->reada == READA_BACK) {
2285 } else if (path->reada == READA_FORWARD) {
2290 if (path->reada == READA_BACK && objectid) {
2291 btrfs_node_key(node, &disk_key, nr);
2292 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2295 search = btrfs_node_blockptr(node, nr);
2296 if ((search <= target && target - search <= 65536) ||
2297 (search > target && search - target <= 65536)) {
2298 gen = btrfs_node_ptr_generation(node, nr);
2299 readahead_tree_block(root, search);
2303 if ((nread > 65536 || nscan > 32))
2308 static noinline void reada_for_balance(struct btrfs_root *root,
2309 struct btrfs_path *path, int level)
2313 struct extent_buffer *parent;
2314 struct extent_buffer *eb;
2319 parent = path->nodes[level + 1];
2323 nritems = btrfs_header_nritems(parent);
2324 slot = path->slots[level + 1];
2327 block1 = btrfs_node_blockptr(parent, slot - 1);
2328 gen = btrfs_node_ptr_generation(parent, slot - 1);
2329 eb = btrfs_find_tree_block(root->fs_info, block1);
2331 * if we get -eagain from btrfs_buffer_uptodate, we
2332 * don't want to return eagain here. That will loop
2335 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2337 free_extent_buffer(eb);
2339 if (slot + 1 < nritems) {
2340 block2 = btrfs_node_blockptr(parent, slot + 1);
2341 gen = btrfs_node_ptr_generation(parent, slot + 1);
2342 eb = btrfs_find_tree_block(root->fs_info, block2);
2343 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2345 free_extent_buffer(eb);
2349 readahead_tree_block(root, block1);
2351 readahead_tree_block(root, block2);
2356 * when we walk down the tree, it is usually safe to unlock the higher layers
2357 * in the tree. The exceptions are when our path goes through slot 0, because
2358 * operations on the tree might require changing key pointers higher up in the
2361 * callers might also have set path->keep_locks, which tells this code to keep
2362 * the lock if the path points to the last slot in the block. This is part of
2363 * walking through the tree, and selecting the next slot in the higher block.
2365 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2366 * if lowest_unlock is 1, level 0 won't be unlocked
2368 static noinline void unlock_up(struct btrfs_path *path, int level,
2369 int lowest_unlock, int min_write_lock_level,
2370 int *write_lock_level)
2373 int skip_level = level;
2375 struct extent_buffer *t;
2377 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2378 if (!path->nodes[i])
2380 if (!path->locks[i])
2382 if (!no_skips && path->slots[i] == 0) {
2386 if (!no_skips && path->keep_locks) {
2389 nritems = btrfs_header_nritems(t);
2390 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2395 if (skip_level < i && i >= lowest_unlock)
2399 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2400 btrfs_tree_unlock_rw(t, path->locks[i]);
2402 if (write_lock_level &&
2403 i > min_write_lock_level &&
2404 i <= *write_lock_level) {
2405 *write_lock_level = i - 1;
2412 * This releases any locks held in the path starting at level and
2413 * going all the way up to the root.
2415 * btrfs_search_slot will keep the lock held on higher nodes in a few
2416 * corner cases, such as COW of the block at slot zero in the node. This
2417 * ignores those rules, and it should only be called when there are no
2418 * more updates to be done higher up in the tree.
2420 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2424 if (path->keep_locks)
2427 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2428 if (!path->nodes[i])
2430 if (!path->locks[i])
2432 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2438 * helper function for btrfs_search_slot. The goal is to find a block
2439 * in cache without setting the path to blocking. If we find the block
2440 * we return zero and the path is unchanged.
2442 * If we can't find the block, we set the path blocking and do some
2443 * reada. -EAGAIN is returned and the search must be repeated.
2446 read_block_for_search(struct btrfs_trans_handle *trans,
2447 struct btrfs_root *root, struct btrfs_path *p,
2448 struct extent_buffer **eb_ret, int level, int slot,
2449 struct btrfs_key *key, u64 time_seq)
2453 struct extent_buffer *b = *eb_ret;
2454 struct extent_buffer *tmp;
2457 blocknr = btrfs_node_blockptr(b, slot);
2458 gen = btrfs_node_ptr_generation(b, slot);
2460 tmp = btrfs_find_tree_block(root->fs_info, blocknr);
2462 /* first we do an atomic uptodate check */
2463 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2468 /* the pages were up to date, but we failed
2469 * the generation number check. Do a full
2470 * read for the generation number that is correct.
2471 * We must do this without dropping locks so
2472 * we can trust our generation number
2474 btrfs_set_path_blocking(p);
2476 /* now we're allowed to do a blocking uptodate check */
2477 ret = btrfs_read_buffer(tmp, gen);
2482 free_extent_buffer(tmp);
2483 btrfs_release_path(p);
2488 * reduce lock contention at high levels
2489 * of the btree by dropping locks before
2490 * we read. Don't release the lock on the current
2491 * level because we need to walk this node to figure
2492 * out which blocks to read.
2494 btrfs_unlock_up_safe(p, level + 1);
2495 btrfs_set_path_blocking(p);
2497 free_extent_buffer(tmp);
2498 if (p->reada != READA_NONE)
2499 reada_for_search(root, p, level, slot, key->objectid);
2501 btrfs_release_path(p);
2504 tmp = read_tree_block(root, blocknr, 0);
2507 * If the read above didn't mark this buffer up to date,
2508 * it will never end up being up to date. Set ret to EIO now
2509 * and give up so that our caller doesn't loop forever
2512 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2514 free_extent_buffer(tmp);
2520 * helper function for btrfs_search_slot. This does all of the checks
2521 * for node-level blocks and does any balancing required based on
2524 * If no extra work was required, zero is returned. If we had to
2525 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2529 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2530 struct btrfs_root *root, struct btrfs_path *p,
2531 struct extent_buffer *b, int level, int ins_len,
2532 int *write_lock_level)
2535 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2536 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2539 if (*write_lock_level < level + 1) {
2540 *write_lock_level = level + 1;
2541 btrfs_release_path(p);
2545 btrfs_set_path_blocking(p);
2546 reada_for_balance(root, p, level);
2547 sret = split_node(trans, root, p, level);
2548 btrfs_clear_path_blocking(p, NULL, 0);
2555 b = p->nodes[level];
2556 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2557 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2560 if (*write_lock_level < level + 1) {
2561 *write_lock_level = level + 1;
2562 btrfs_release_path(p);
2566 btrfs_set_path_blocking(p);
2567 reada_for_balance(root, p, level);
2568 sret = balance_level(trans, root, p, level);
2569 btrfs_clear_path_blocking(p, NULL, 0);
2575 b = p->nodes[level];
2577 btrfs_release_path(p);
2580 BUG_ON(btrfs_header_nritems(b) == 1);
2590 static void key_search_validate(struct extent_buffer *b,
2591 struct btrfs_key *key,
2594 #ifdef CONFIG_BTRFS_ASSERT
2595 struct btrfs_disk_key disk_key;
2597 btrfs_cpu_key_to_disk(&disk_key, key);
2600 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2601 offsetof(struct btrfs_leaf, items[0].key),
2604 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2605 offsetof(struct btrfs_node, ptrs[0].key),
2610 static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2611 int level, int *prev_cmp, int *slot)
2613 if (*prev_cmp != 0) {
2614 *prev_cmp = bin_search(b, key, level, slot);
2618 key_search_validate(b, key, level);
2624 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2625 u64 iobjectid, u64 ioff, u8 key_type,
2626 struct btrfs_key *found_key)
2629 struct btrfs_key key;
2630 struct extent_buffer *eb;
2635 key.type = key_type;
2636 key.objectid = iobjectid;
2639 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2643 eb = path->nodes[0];
2644 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2645 ret = btrfs_next_leaf(fs_root, path);
2648 eb = path->nodes[0];
2651 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2652 if (found_key->type != key.type ||
2653 found_key->objectid != key.objectid)
2660 * look for key in the tree. path is filled in with nodes along the way
2661 * if key is found, we return zero and you can find the item in the leaf
2662 * level of the path (level 0)
2664 * If the key isn't found, the path points to the slot where it should
2665 * be inserted, and 1 is returned. If there are other errors during the
2666 * search a negative error number is returned.
2668 * if ins_len > 0, nodes and leaves will be split as we walk down the
2669 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2672 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2673 *root, struct btrfs_key *key, struct btrfs_path *p, int
2676 struct extent_buffer *b;
2681 int lowest_unlock = 1;
2683 /* everything at write_lock_level or lower must be write locked */
2684 int write_lock_level = 0;
2685 u8 lowest_level = 0;
2686 int min_write_lock_level;
2689 lowest_level = p->lowest_level;
2690 WARN_ON(lowest_level && ins_len > 0);
2691 WARN_ON(p->nodes[0] != NULL);
2692 BUG_ON(!cow && ins_len);
2697 /* when we are removing items, we might have to go up to level
2698 * two as we update tree pointers Make sure we keep write
2699 * for those levels as well
2701 write_lock_level = 2;
2702 } else if (ins_len > 0) {
2704 * for inserting items, make sure we have a write lock on
2705 * level 1 so we can update keys
2707 write_lock_level = 1;
2711 write_lock_level = -1;
2713 if (cow && (p->keep_locks || p->lowest_level))
2714 write_lock_level = BTRFS_MAX_LEVEL;
2716 min_write_lock_level = write_lock_level;
2721 * we try very hard to do read locks on the root
2723 root_lock = BTRFS_READ_LOCK;
2725 if (p->search_commit_root) {
2727 * the commit roots are read only
2728 * so we always do read locks
2730 if (p->need_commit_sem)
2731 down_read(&root->fs_info->commit_root_sem);
2732 b = root->commit_root;
2733 extent_buffer_get(b);
2734 level = btrfs_header_level(b);
2735 if (p->need_commit_sem)
2736 up_read(&root->fs_info->commit_root_sem);
2737 if (!p->skip_locking)
2738 btrfs_tree_read_lock(b);
2740 if (p->skip_locking) {
2741 b = btrfs_root_node(root);
2742 level = btrfs_header_level(b);
2744 /* we don't know the level of the root node
2745 * until we actually have it read locked
2747 b = btrfs_read_lock_root_node(root);
2748 level = btrfs_header_level(b);
2749 if (level <= write_lock_level) {
2750 /* whoops, must trade for write lock */
2751 btrfs_tree_read_unlock(b);
2752 free_extent_buffer(b);
2753 b = btrfs_lock_root_node(root);
2754 root_lock = BTRFS_WRITE_LOCK;
2756 /* the level might have changed, check again */
2757 level = btrfs_header_level(b);
2761 p->nodes[level] = b;
2762 if (!p->skip_locking)
2763 p->locks[level] = root_lock;
2766 level = btrfs_header_level(b);
2769 * setup the path here so we can release it under lock
2770 * contention with the cow code
2774 * if we don't really need to cow this block
2775 * then we don't want to set the path blocking,
2776 * so we test it here
2778 if (!should_cow_block(trans, root, b))
2782 * must have write locks on this node and the
2785 if (level > write_lock_level ||
2786 (level + 1 > write_lock_level &&
2787 level + 1 < BTRFS_MAX_LEVEL &&
2788 p->nodes[level + 1])) {
2789 write_lock_level = level + 1;
2790 btrfs_release_path(p);
2794 btrfs_set_path_blocking(p);
2795 err = btrfs_cow_block(trans, root, b,
2796 p->nodes[level + 1],
2797 p->slots[level + 1], &b);
2804 p->nodes[level] = b;
2805 btrfs_clear_path_blocking(p, NULL, 0);
2808 * we have a lock on b and as long as we aren't changing
2809 * the tree, there is no way to for the items in b to change.
2810 * It is safe to drop the lock on our parent before we
2811 * go through the expensive btree search on b.
2813 * If we're inserting or deleting (ins_len != 0), then we might
2814 * be changing slot zero, which may require changing the parent.
2815 * So, we can't drop the lock until after we know which slot
2816 * we're operating on.
2818 if (!ins_len && !p->keep_locks) {
2821 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2822 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2827 ret = key_search(b, key, level, &prev_cmp, &slot);
2831 if (ret && slot > 0) {
2835 p->slots[level] = slot;
2836 err = setup_nodes_for_search(trans, root, p, b, level,
2837 ins_len, &write_lock_level);
2844 b = p->nodes[level];
2845 slot = p->slots[level];
2848 * slot 0 is special, if we change the key
2849 * we have to update the parent pointer
2850 * which means we must have a write lock
2853 if (slot == 0 && ins_len &&
2854 write_lock_level < level + 1) {
2855 write_lock_level = level + 1;
2856 btrfs_release_path(p);
2860 unlock_up(p, level, lowest_unlock,
2861 min_write_lock_level, &write_lock_level);
2863 if (level == lowest_level) {
2869 err = read_block_for_search(trans, root, p,
2870 &b, level, slot, key, 0);
2878 if (!p->skip_locking) {
2879 level = btrfs_header_level(b);
2880 if (level <= write_lock_level) {
2881 err = btrfs_try_tree_write_lock(b);
2883 btrfs_set_path_blocking(p);
2885 btrfs_clear_path_blocking(p, b,
2888 p->locks[level] = BTRFS_WRITE_LOCK;
2890 err = btrfs_tree_read_lock_atomic(b);
2892 btrfs_set_path_blocking(p);
2893 btrfs_tree_read_lock(b);
2894 btrfs_clear_path_blocking(p, b,
2897 p->locks[level] = BTRFS_READ_LOCK;
2899 p->nodes[level] = b;
2902 p->slots[level] = slot;
2904 btrfs_leaf_free_space(root, b) < ins_len) {
2905 if (write_lock_level < 1) {
2906 write_lock_level = 1;
2907 btrfs_release_path(p);
2911 btrfs_set_path_blocking(p);
2912 err = split_leaf(trans, root, key,
2913 p, ins_len, ret == 0);
2914 btrfs_clear_path_blocking(p, NULL, 0);
2922 if (!p->search_for_split)
2923 unlock_up(p, level, lowest_unlock,
2924 min_write_lock_level, &write_lock_level);
2931 * we don't really know what they plan on doing with the path
2932 * from here on, so for now just mark it as blocking
2934 if (!p->leave_spinning)
2935 btrfs_set_path_blocking(p);
2936 if (ret < 0 && !p->skip_release_on_error)
2937 btrfs_release_path(p);
2942 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2943 * current state of the tree together with the operations recorded in the tree
2944 * modification log to search for the key in a previous version of this tree, as
2945 * denoted by the time_seq parameter.
2947 * Naturally, there is no support for insert, delete or cow operations.
2949 * The resulting path and return value will be set up as if we called
2950 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2952 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2953 struct btrfs_path *p, u64 time_seq)
2955 struct extent_buffer *b;
2960 int lowest_unlock = 1;
2961 u8 lowest_level = 0;
2964 lowest_level = p->lowest_level;
2965 WARN_ON(p->nodes[0] != NULL);
2967 if (p->search_commit_root) {
2969 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2973 b = get_old_root(root, time_seq);
2974 level = btrfs_header_level(b);
2975 p->locks[level] = BTRFS_READ_LOCK;
2978 level = btrfs_header_level(b);
2979 p->nodes[level] = b;
2980 btrfs_clear_path_blocking(p, NULL, 0);
2983 * we have a lock on b and as long as we aren't changing
2984 * the tree, there is no way to for the items in b to change.
2985 * It is safe to drop the lock on our parent before we
2986 * go through the expensive btree search on b.
2988 btrfs_unlock_up_safe(p, level + 1);
2991 * Since we can unwind ebs we want to do a real search every
2995 ret = key_search(b, key, level, &prev_cmp, &slot);
2999 if (ret && slot > 0) {
3003 p->slots[level] = slot;
3004 unlock_up(p, level, lowest_unlock, 0, NULL);
3006 if (level == lowest_level) {
3012 err = read_block_for_search(NULL, root, p, &b, level,
3013 slot, key, time_seq);
3021 level = btrfs_header_level(b);
3022 err = btrfs_tree_read_lock_atomic(b);
3024 btrfs_set_path_blocking(p);
3025 btrfs_tree_read_lock(b);
3026 btrfs_clear_path_blocking(p, b,
3029 b = tree_mod_log_rewind(root->fs_info, p, b, time_seq);
3034 p->locks[level] = BTRFS_READ_LOCK;
3035 p->nodes[level] = b;
3037 p->slots[level] = slot;
3038 unlock_up(p, level, lowest_unlock, 0, NULL);
3044 if (!p->leave_spinning)
3045 btrfs_set_path_blocking(p);
3047 btrfs_release_path(p);
3053 * helper to use instead of search slot if no exact match is needed but
3054 * instead the next or previous item should be returned.
3055 * When find_higher is true, the next higher item is returned, the next lower
3057 * When return_any and find_higher are both true, and no higher item is found,
3058 * return the next lower instead.
3059 * When return_any is true and find_higher is false, and no lower item is found,
3060 * return the next higher instead.
3061 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3064 int btrfs_search_slot_for_read(struct btrfs_root *root,
3065 struct btrfs_key *key, struct btrfs_path *p,
3066 int find_higher, int return_any)
3069 struct extent_buffer *leaf;
3072 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3076 * a return value of 1 means the path is at the position where the
3077 * item should be inserted. Normally this is the next bigger item,
3078 * but in case the previous item is the last in a leaf, path points
3079 * to the first free slot in the previous leaf, i.e. at an invalid
3085 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3086 ret = btrfs_next_leaf(root, p);
3092 * no higher item found, return the next
3097 btrfs_release_path(p);
3101 if (p->slots[0] == 0) {
3102 ret = btrfs_prev_leaf(root, p);
3107 if (p->slots[0] == btrfs_header_nritems(leaf))
3114 * no lower item found, return the next
3119 btrfs_release_path(p);
3129 * adjust the pointers going up the tree, starting at level
3130 * making sure the right key of each node is points to 'key'.
3131 * This is used after shifting pointers to the left, so it stops
3132 * fixing up pointers when a given leaf/node is not in slot 0 of the
3136 static void fixup_low_keys(struct btrfs_fs_info *fs_info,
3137 struct btrfs_path *path,
3138 struct btrfs_disk_key *key, int level)
3141 struct extent_buffer *t;
3143 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3144 int tslot = path->slots[i];
3145 if (!path->nodes[i])
3148 tree_mod_log_set_node_key(fs_info, t, tslot, 1);
3149 btrfs_set_node_key(t, key, tslot);
3150 btrfs_mark_buffer_dirty(path->nodes[i]);
3159 * This function isn't completely safe. It's the caller's responsibility
3160 * that the new key won't break the order
3162 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3163 struct btrfs_path *path,
3164 struct btrfs_key *new_key)
3166 struct btrfs_disk_key disk_key;
3167 struct extent_buffer *eb;
3170 eb = path->nodes[0];
3171 slot = path->slots[0];
3173 btrfs_item_key(eb, &disk_key, slot - 1);
3174 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3176 if (slot < btrfs_header_nritems(eb) - 1) {
3177 btrfs_item_key(eb, &disk_key, slot + 1);
3178 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3181 btrfs_cpu_key_to_disk(&disk_key, new_key);
3182 btrfs_set_item_key(eb, &disk_key, slot);
3183 btrfs_mark_buffer_dirty(eb);
3185 fixup_low_keys(fs_info, path, &disk_key, 1);
3189 * try to push data from one node into the next node left in the
3192 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3193 * error, and > 0 if there was no room in the left hand block.
3195 static int push_node_left(struct btrfs_trans_handle *trans,
3196 struct btrfs_root *root, struct extent_buffer *dst,
3197 struct extent_buffer *src, int empty)
3204 src_nritems = btrfs_header_nritems(src);
3205 dst_nritems = btrfs_header_nritems(dst);
3206 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3207 WARN_ON(btrfs_header_generation(src) != trans->transid);
3208 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3210 if (!empty && src_nritems <= 8)
3213 if (push_items <= 0)
3217 push_items = min(src_nritems, push_items);
3218 if (push_items < src_nritems) {
3219 /* leave at least 8 pointers in the node if
3220 * we aren't going to empty it
3222 if (src_nritems - push_items < 8) {
3223 if (push_items <= 8)
3229 push_items = min(src_nritems - 8, push_items);
3231 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3234 btrfs_abort_transaction(trans, root, ret);
3237 copy_extent_buffer(dst, src,
3238 btrfs_node_key_ptr_offset(dst_nritems),
3239 btrfs_node_key_ptr_offset(0),
3240 push_items * sizeof(struct btrfs_key_ptr));
3242 if (push_items < src_nritems) {
3244 * don't call tree_mod_log_eb_move here, key removal was already
3245 * fully logged by tree_mod_log_eb_copy above.
3247 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3248 btrfs_node_key_ptr_offset(push_items),
3249 (src_nritems - push_items) *
3250 sizeof(struct btrfs_key_ptr));
3252 btrfs_set_header_nritems(src, src_nritems - push_items);
3253 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3254 btrfs_mark_buffer_dirty(src);
3255 btrfs_mark_buffer_dirty(dst);
3261 * try to push data from one node into the next node right in the
3264 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3265 * error, and > 0 if there was no room in the right hand block.
3267 * this will only push up to 1/2 the contents of the left node over
3269 static int balance_node_right(struct btrfs_trans_handle *trans,
3270 struct btrfs_root *root,
3271 struct extent_buffer *dst,
3272 struct extent_buffer *src)
3280 WARN_ON(btrfs_header_generation(src) != trans->transid);
3281 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3283 src_nritems = btrfs_header_nritems(src);
3284 dst_nritems = btrfs_header_nritems(dst);
3285 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3286 if (push_items <= 0)
3289 if (src_nritems < 4)
3292 max_push = src_nritems / 2 + 1;
3293 /* don't try to empty the node */
3294 if (max_push >= src_nritems)
3297 if (max_push < push_items)
3298 push_items = max_push;
3300 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3301 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3302 btrfs_node_key_ptr_offset(0),
3304 sizeof(struct btrfs_key_ptr));
3306 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3307 src_nritems - push_items, push_items);
3309 btrfs_abort_transaction(trans, root, ret);
3312 copy_extent_buffer(dst, src,
3313 btrfs_node_key_ptr_offset(0),
3314 btrfs_node_key_ptr_offset(src_nritems - push_items),
3315 push_items * sizeof(struct btrfs_key_ptr));
3317 btrfs_set_header_nritems(src, src_nritems - push_items);
3318 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3320 btrfs_mark_buffer_dirty(src);
3321 btrfs_mark_buffer_dirty(dst);
3327 * helper function to insert a new root level in the tree.
3328 * A new node is allocated, and a single item is inserted to
3329 * point to the existing root
3331 * returns zero on success or < 0 on failure.
3333 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3334 struct btrfs_root *root,
3335 struct btrfs_path *path, int level)
3338 struct extent_buffer *lower;
3339 struct extent_buffer *c;
3340 struct extent_buffer *old;
3341 struct btrfs_disk_key lower_key;
3343 BUG_ON(path->nodes[level]);
3344 BUG_ON(path->nodes[level-1] != root->node);
3346 lower = path->nodes[level-1];
3348 btrfs_item_key(lower, &lower_key, 0);
3350 btrfs_node_key(lower, &lower_key, 0);
3352 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3353 &lower_key, level, root->node->start, 0);
3357 root_add_used(root, root->nodesize);
3359 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3360 btrfs_set_header_nritems(c, 1);
3361 btrfs_set_header_level(c, level);
3362 btrfs_set_header_bytenr(c, c->start);
3363 btrfs_set_header_generation(c, trans->transid);
3364 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3365 btrfs_set_header_owner(c, root->root_key.objectid);
3367 write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(),
3370 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3371 btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE);
3373 btrfs_set_node_key(c, &lower_key, 0);
3374 btrfs_set_node_blockptr(c, 0, lower->start);
3375 lower_gen = btrfs_header_generation(lower);
3376 WARN_ON(lower_gen != trans->transid);
3378 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3380 btrfs_mark_buffer_dirty(c);
3383 tree_mod_log_set_root_pointer(root, c, 0);
3384 rcu_assign_pointer(root->node, c);
3386 /* the super has an extra ref to root->node */
3387 free_extent_buffer(old);
3389 add_root_to_dirty_list(root);
3390 extent_buffer_get(c);
3391 path->nodes[level] = c;
3392 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3393 path->slots[level] = 0;
3398 * worker function to insert a single pointer in a node.
3399 * the node should have enough room for the pointer already
3401 * slot and level indicate where you want the key to go, and
3402 * blocknr is the block the key points to.
3404 static void insert_ptr(struct btrfs_trans_handle *trans,
3405 struct btrfs_root *root, struct btrfs_path *path,
3406 struct btrfs_disk_key *key, u64 bytenr,
3407 int slot, int level)
3409 struct extent_buffer *lower;
3413 BUG_ON(!path->nodes[level]);
3414 btrfs_assert_tree_locked(path->nodes[level]);
3415 lower = path->nodes[level];
3416 nritems = btrfs_header_nritems(lower);
3417 BUG_ON(slot > nritems);
3418 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3419 if (slot != nritems) {
3421 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3422 slot, nritems - slot);
3423 memmove_extent_buffer(lower,
3424 btrfs_node_key_ptr_offset(slot + 1),
3425 btrfs_node_key_ptr_offset(slot),
3426 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3429 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3430 MOD_LOG_KEY_ADD, GFP_NOFS);
3433 btrfs_set_node_key(lower, key, slot);
3434 btrfs_set_node_blockptr(lower, slot, bytenr);
3435 WARN_ON(trans->transid == 0);
3436 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3437 btrfs_set_header_nritems(lower, nritems + 1);
3438 btrfs_mark_buffer_dirty(lower);
3442 * split the node at the specified level in path in two.
3443 * The path is corrected to point to the appropriate node after the split
3445 * Before splitting this tries to make some room in the node by pushing
3446 * left and right, if either one works, it returns right away.
3448 * returns 0 on success and < 0 on failure
3450 static noinline int split_node(struct btrfs_trans_handle *trans,
3451 struct btrfs_root *root,
3452 struct btrfs_path *path, int level)
3454 struct extent_buffer *c;
3455 struct extent_buffer *split;
3456 struct btrfs_disk_key disk_key;
3461 c = path->nodes[level];
3462 WARN_ON(btrfs_header_generation(c) != trans->transid);
3463 if (c == root->node) {
3465 * trying to split the root, lets make a new one
3467 * tree mod log: We don't log_removal old root in
3468 * insert_new_root, because that root buffer will be kept as a
3469 * normal node. We are going to log removal of half of the
3470 * elements below with tree_mod_log_eb_copy. We're holding a
3471 * tree lock on the buffer, which is why we cannot race with
3472 * other tree_mod_log users.
3474 ret = insert_new_root(trans, root, path, level + 1);
3478 ret = push_nodes_for_insert(trans, root, path, level);
3479 c = path->nodes[level];
3480 if (!ret && btrfs_header_nritems(c) <
3481 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3487 c_nritems = btrfs_header_nritems(c);
3488 mid = (c_nritems + 1) / 2;
3489 btrfs_node_key(c, &disk_key, mid);
3491 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3492 &disk_key, level, c->start, 0);
3494 return PTR_ERR(split);
3496 root_add_used(root, root->nodesize);
3498 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3499 btrfs_set_header_level(split, btrfs_header_level(c));
3500 btrfs_set_header_bytenr(split, split->start);
3501 btrfs_set_header_generation(split, trans->transid);
3502 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3503 btrfs_set_header_owner(split, root->root_key.objectid);
3504 write_extent_buffer(split, root->fs_info->fsid,
3505 btrfs_header_fsid(), BTRFS_FSID_SIZE);
3506 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3507 btrfs_header_chunk_tree_uuid(split),
3510 ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0,
3511 mid, c_nritems - mid);
3513 btrfs_abort_transaction(trans, root, ret);
3516 copy_extent_buffer(split, c,
3517 btrfs_node_key_ptr_offset(0),
3518 btrfs_node_key_ptr_offset(mid),
3519 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3520 btrfs_set_header_nritems(split, c_nritems - mid);
3521 btrfs_set_header_nritems(c, mid);
3524 btrfs_mark_buffer_dirty(c);
3525 btrfs_mark_buffer_dirty(split);
3527 insert_ptr(trans, root, path, &disk_key, split->start,
3528 path->slots[level + 1] + 1, level + 1);
3530 if (path->slots[level] >= mid) {
3531 path->slots[level] -= mid;
3532 btrfs_tree_unlock(c);
3533 free_extent_buffer(c);
3534 path->nodes[level] = split;
3535 path->slots[level + 1] += 1;
3537 btrfs_tree_unlock(split);
3538 free_extent_buffer(split);
3544 * how many bytes are required to store the items in a leaf. start
3545 * and nr indicate which items in the leaf to check. This totals up the
3546 * space used both by the item structs and the item data
3548 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3550 struct btrfs_item *start_item;
3551 struct btrfs_item *end_item;
3552 struct btrfs_map_token token;
3554 int nritems = btrfs_header_nritems(l);
3555 int end = min(nritems, start + nr) - 1;
3559 btrfs_init_map_token(&token);
3560 start_item = btrfs_item_nr(start);
3561 end_item = btrfs_item_nr(end);
3562 data_len = btrfs_token_item_offset(l, start_item, &token) +
3563 btrfs_token_item_size(l, start_item, &token);
3564 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3565 data_len += sizeof(struct btrfs_item) * nr;
3566 WARN_ON(data_len < 0);
3571 * The space between the end of the leaf items and
3572 * the start of the leaf data. IOW, how much room
3573 * the leaf has left for both items and data
3575 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3576 struct extent_buffer *leaf)
3578 int nritems = btrfs_header_nritems(leaf);
3580 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3582 btrfs_crit(root->fs_info,
3583 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3584 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3585 leaf_space_used(leaf, 0, nritems), nritems);
3591 * min slot controls the lowest index we're willing to push to the
3592 * right. We'll push up to and including min_slot, but no lower
3594 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3595 struct btrfs_root *root,
3596 struct btrfs_path *path,
3597 int data_size, int empty,
3598 struct extent_buffer *right,
3599 int free_space, u32 left_nritems,
3602 struct extent_buffer *left = path->nodes[0];
3603 struct extent_buffer *upper = path->nodes[1];
3604 struct btrfs_map_token token;
3605 struct btrfs_disk_key disk_key;
3610 struct btrfs_item *item;
3616 btrfs_init_map_token(&token);
3621 nr = max_t(u32, 1, min_slot);
3623 if (path->slots[0] >= left_nritems)
3624 push_space += data_size;
3626 slot = path->slots[1];
3627 i = left_nritems - 1;
3629 item = btrfs_item_nr(i);
3631 if (!empty && push_items > 0) {
3632 if (path->slots[0] > i)
3634 if (path->slots[0] == i) {
3635 int space = btrfs_leaf_free_space(root, left);
3636 if (space + push_space * 2 > free_space)
3641 if (path->slots[0] == i)
3642 push_space += data_size;
3644 this_item_size = btrfs_item_size(left, item);
3645 if (this_item_size + sizeof(*item) + push_space > free_space)
3649 push_space += this_item_size + sizeof(*item);
3655 if (push_items == 0)
3658 WARN_ON(!empty && push_items == left_nritems);
3660 /* push left to right */
3661 right_nritems = btrfs_header_nritems(right);
3663 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3664 push_space -= leaf_data_end(root, left);
3666 /* make room in the right data area */
3667 data_end = leaf_data_end(root, right);
3668 memmove_extent_buffer(right,
3669 btrfs_leaf_data(right) + data_end - push_space,
3670 btrfs_leaf_data(right) + data_end,
3671 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3673 /* copy from the left data area */
3674 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3675 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3676 btrfs_leaf_data(left) + leaf_data_end(root, left),
3679 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3680 btrfs_item_nr_offset(0),
3681 right_nritems * sizeof(struct btrfs_item));
3683 /* copy the items from left to right */
3684 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3685 btrfs_item_nr_offset(left_nritems - push_items),
3686 push_items * sizeof(struct btrfs_item));
3688 /* update the item pointers */
3689 right_nritems += push_items;
3690 btrfs_set_header_nritems(right, right_nritems);
3691 push_space = BTRFS_LEAF_DATA_SIZE(root);
3692 for (i = 0; i < right_nritems; i++) {
3693 item = btrfs_item_nr(i);
3694 push_space -= btrfs_token_item_size(right, item, &token);
3695 btrfs_set_token_item_offset(right, item, push_space, &token);
3698 left_nritems -= push_items;
3699 btrfs_set_header_nritems(left, left_nritems);
3702 btrfs_mark_buffer_dirty(left);
3704 clean_tree_block(trans, root->fs_info, left);
3706 btrfs_mark_buffer_dirty(right);
3708 btrfs_item_key(right, &disk_key, 0);
3709 btrfs_set_node_key(upper, &disk_key, slot + 1);
3710 btrfs_mark_buffer_dirty(upper);
3712 /* then fixup the leaf pointer in the path */
3713 if (path->slots[0] >= left_nritems) {
3714 path->slots[0] -= left_nritems;
3715 if (btrfs_header_nritems(path->nodes[0]) == 0)
3716 clean_tree_block(trans, root->fs_info, path->nodes[0]);
3717 btrfs_tree_unlock(path->nodes[0]);
3718 free_extent_buffer(path->nodes[0]);
3719 path->nodes[0] = right;
3720 path->slots[1] += 1;
3722 btrfs_tree_unlock(right);
3723 free_extent_buffer(right);
3728 btrfs_tree_unlock(right);
3729 free_extent_buffer(right);
3734 * push some data in the path leaf to the right, trying to free up at
3735 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3737 * returns 1 if the push failed because the other node didn't have enough
3738 * room, 0 if everything worked out and < 0 if there were major errors.
3740 * this will push starting from min_slot to the end of the leaf. It won't
3741 * push any slot lower than min_slot
3743 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3744 *root, struct btrfs_path *path,
3745 int min_data_size, int data_size,
3746 int empty, u32 min_slot)
3748 struct extent_buffer *left = path->nodes[0];
3749 struct extent_buffer *right;
3750 struct extent_buffer *upper;
3756 if (!path->nodes[1])
3759 slot = path->slots[1];
3760 upper = path->nodes[1];
3761 if (slot >= btrfs_header_nritems(upper) - 1)
3764 btrfs_assert_tree_locked(path->nodes[1]);
3766 right = read_node_slot(root, upper, slot + 1);
3770 btrfs_tree_lock(right);
3771 btrfs_set_lock_blocking(right);
3773 free_space = btrfs_leaf_free_space(root, right);
3774 if (free_space < data_size)
3777 /* cow and double check */
3778 ret = btrfs_cow_block(trans, root, right, upper,
3783 free_space = btrfs_leaf_free_space(root, right);
3784 if (free_space < data_size)
3787 left_nritems = btrfs_header_nritems(left);
3788 if (left_nritems == 0)
3791 if (path->slots[0] == left_nritems && !empty) {
3792 /* Key greater than all keys in the leaf, right neighbor has
3793 * enough room for it and we're not emptying our leaf to delete
3794 * it, therefore use right neighbor to insert the new item and
3795 * no need to touch/dirty our left leaft. */
3796 btrfs_tree_unlock(left);
3797 free_extent_buffer(left);
3798 path->nodes[0] = right;
3804 return __push_leaf_right(trans, root, path, min_data_size, empty,
3805 right, free_space, left_nritems, min_slot);
3807 btrfs_tree_unlock(right);
3808 free_extent_buffer(right);
3813 * push some data in the path leaf to the left, trying to free up at
3814 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3816 * max_slot can put a limit on how far into the leaf we'll push items. The
3817 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3820 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3821 struct btrfs_root *root,
3822 struct btrfs_path *path, int data_size,
3823 int empty, struct extent_buffer *left,
3824 int free_space, u32 right_nritems,
3827 struct btrfs_disk_key disk_key;
3828 struct extent_buffer *right = path->nodes[0];
3832 struct btrfs_item *item;
3833 u32 old_left_nritems;
3837 u32 old_left_item_size;
3838 struct btrfs_map_token token;
3840 btrfs_init_map_token(&token);
3843 nr = min(right_nritems, max_slot);
3845 nr = min(right_nritems - 1, max_slot);
3847 for (i = 0; i < nr; i++) {
3848 item = btrfs_item_nr(i);
3850 if (!empty && push_items > 0) {
3851 if (path->slots[0] < i)
3853 if (path->slots[0] == i) {
3854 int space = btrfs_leaf_free_space(root, right);
3855 if (space + push_space * 2 > free_space)
3860 if (path->slots[0] == i)
3861 push_space += data_size;
3863 this_item_size = btrfs_item_size(right, item);
3864 if (this_item_size + sizeof(*item) + push_space > free_space)
3868 push_space += this_item_size + sizeof(*item);
3871 if (push_items == 0) {
3875 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3877 /* push data from right to left */
3878 copy_extent_buffer(left, right,
3879 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3880 btrfs_item_nr_offset(0),
3881 push_items * sizeof(struct btrfs_item));
3883 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3884 btrfs_item_offset_nr(right, push_items - 1);
3886 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3887 leaf_data_end(root, left) - push_space,
3888 btrfs_leaf_data(right) +
3889 btrfs_item_offset_nr(right, push_items - 1),
3891 old_left_nritems = btrfs_header_nritems(left);
3892 BUG_ON(old_left_nritems <= 0);
3894 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3895 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3898 item = btrfs_item_nr(i);
3900 ioff = btrfs_token_item_offset(left, item, &token);
3901 btrfs_set_token_item_offset(left, item,
3902 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3905 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3907 /* fixup right node */
3908 if (push_items > right_nritems)
3909 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3912 if (push_items < right_nritems) {
3913 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3914 leaf_data_end(root, right);
3915 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3916 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3917 btrfs_leaf_data(right) +
3918 leaf_data_end(root, right), push_space);
3920 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3921 btrfs_item_nr_offset(push_items),
3922 (btrfs_header_nritems(right) - push_items) *
3923 sizeof(struct btrfs_item));
3925 right_nritems -= push_items;
3926 btrfs_set_header_nritems(right, right_nritems);
3927 push_space = BTRFS_LEAF_DATA_SIZE(root);
3928 for (i = 0; i < right_nritems; i++) {
3929 item = btrfs_item_nr(i);
3931 push_space = push_space - btrfs_token_item_size(right,
3933 btrfs_set_token_item_offset(right, item, push_space, &token);
3936 btrfs_mark_buffer_dirty(left);
3938 btrfs_mark_buffer_dirty(right);
3940 clean_tree_block(trans, root->fs_info, right);
3942 btrfs_item_key(right, &disk_key, 0);
3943 fixup_low_keys(root->fs_info, path, &disk_key, 1);
3945 /* then fixup the leaf pointer in the path */
3946 if (path->slots[0] < push_items) {
3947 path->slots[0] += old_left_nritems;
3948 btrfs_tree_unlock(path->nodes[0]);
3949 free_extent_buffer(path->nodes[0]);
3950 path->nodes[0] = left;
3951 path->slots[1] -= 1;
3953 btrfs_tree_unlock(left);
3954 free_extent_buffer(left);
3955 path->slots[0] -= push_items;
3957 BUG_ON(path->slots[0] < 0);
3960 btrfs_tree_unlock(left);
3961 free_extent_buffer(left);
3966 * push some data in the path leaf to the left, trying to free up at
3967 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3969 * max_slot can put a limit on how far into the leaf we'll push items. The
3970 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3973 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3974 *root, struct btrfs_path *path, int min_data_size,
3975 int data_size, int empty, u32 max_slot)
3977 struct extent_buffer *right = path->nodes[0];
3978 struct extent_buffer *left;
3984 slot = path->slots[1];
3987 if (!path->nodes[1])
3990 right_nritems = btrfs_header_nritems(right);
3991 if (right_nritems == 0)
3994 btrfs_assert_tree_locked(path->nodes[1]);
3996 left = read_node_slot(root, path->nodes[1], slot - 1);
4000 btrfs_tree_lock(left);
4001 btrfs_set_lock_blocking(left);
4003 free_space = btrfs_leaf_free_space(root, left);
4004 if (free_space < data_size) {
4009 /* cow and double check */
4010 ret = btrfs_cow_block(trans, root, left,
4011 path->nodes[1], slot - 1, &left);
4013 /* we hit -ENOSPC, but it isn't fatal here */
4019 free_space = btrfs_leaf_free_space(root, left);
4020 if (free_space < data_size) {
4025 return __push_leaf_left(trans, root, path, min_data_size,
4026 empty, left, free_space, right_nritems,
4029 btrfs_tree_unlock(left);
4030 free_extent_buffer(left);
4035 * split the path's leaf in two, making sure there is at least data_size
4036 * available for the resulting leaf level of the path.
4038 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4039 struct btrfs_root *root,
4040 struct btrfs_path *path,
4041 struct extent_buffer *l,
4042 struct extent_buffer *right,
4043 int slot, int mid, int nritems)
4048 struct btrfs_disk_key disk_key;
4049 struct btrfs_map_token token;
4051 btrfs_init_map_token(&token);
4053 nritems = nritems - mid;
4054 btrfs_set_header_nritems(right, nritems);
4055 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
4057 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4058 btrfs_item_nr_offset(mid),
4059 nritems * sizeof(struct btrfs_item));
4061 copy_extent_buffer(right, l,
4062 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
4063 data_copy_size, btrfs_leaf_data(l) +
4064 leaf_data_end(root, l), data_copy_size);
4066 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
4067 btrfs_item_end_nr(l, mid);
4069 for (i = 0; i < nritems; i++) {
4070 struct btrfs_item *item = btrfs_item_nr(i);
4073 ioff = btrfs_token_item_offset(right, item, &token);
4074 btrfs_set_token_item_offset(right, item,
4075 ioff + rt_data_off, &token);
4078 btrfs_set_header_nritems(l, mid);
4079 btrfs_item_key(right, &disk_key, 0);
4080 insert_ptr(trans, root, path, &disk_key, right->start,
4081 path->slots[1] + 1, 1);
4083 btrfs_mark_buffer_dirty(right);
4084 btrfs_mark_buffer_dirty(l);
4085 BUG_ON(path->slots[0] != slot);
4088 btrfs_tree_unlock(path->nodes[0]);
4089 free_extent_buffer(path->nodes[0]);
4090 path->nodes[0] = right;
4091 path->slots[0] -= mid;
4092 path->slots[1] += 1;
4094 btrfs_tree_unlock(right);
4095 free_extent_buffer(right);
4098 BUG_ON(path->slots[0] < 0);
4102 * double splits happen when we need to insert a big item in the middle
4103 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4104 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4107 * We avoid this by trying to push the items on either side of our target
4108 * into the adjacent leaves. If all goes well we can avoid the double split
4111 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4112 struct btrfs_root *root,
4113 struct btrfs_path *path,
4120 int space_needed = data_size;
4122 slot = path->slots[0];
4123 if (slot < btrfs_header_nritems(path->nodes[0]))
4124 space_needed -= btrfs_leaf_free_space(root, path->nodes[0]);
4127 * try to push all the items after our slot into the
4130 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4137 nritems = btrfs_header_nritems(path->nodes[0]);
4139 * our goal is to get our slot at the start or end of a leaf. If
4140 * we've done so we're done
4142 if (path->slots[0] == 0 || path->slots[0] == nritems)
4145 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4148 /* try to push all the items before our slot into the next leaf */
4149 slot = path->slots[0];
4150 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4163 * split the path's leaf in two, making sure there is at least data_size
4164 * available for the resulting leaf level of the path.
4166 * returns 0 if all went well and < 0 on failure.
4168 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4169 struct btrfs_root *root,
4170 struct btrfs_key *ins_key,
4171 struct btrfs_path *path, int data_size,
4174 struct btrfs_disk_key disk_key;
4175 struct extent_buffer *l;
4179 struct extent_buffer *right;
4180 struct btrfs_fs_info *fs_info = root->fs_info;
4184 int num_doubles = 0;
4185 int tried_avoid_double = 0;
4188 slot = path->slots[0];
4189 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4190 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
4193 /* first try to make some room by pushing left and right */
4194 if (data_size && path->nodes[1]) {
4195 int space_needed = data_size;
4197 if (slot < btrfs_header_nritems(l))
4198 space_needed -= btrfs_leaf_free_space(root, l);
4200 wret = push_leaf_right(trans, root, path, space_needed,
4201 space_needed, 0, 0);
4205 wret = push_leaf_left(trans, root, path, space_needed,
4206 space_needed, 0, (u32)-1);
4212 /* did the pushes work? */
4213 if (btrfs_leaf_free_space(root, l) >= data_size)
4217 if (!path->nodes[1]) {
4218 ret = insert_new_root(trans, root, path, 1);
4225 slot = path->slots[0];
4226 nritems = btrfs_header_nritems(l);
4227 mid = (nritems + 1) / 2;
4231 leaf_space_used(l, mid, nritems - mid) + data_size >
4232 BTRFS_LEAF_DATA_SIZE(root)) {
4233 if (slot >= nritems) {
4237 if (mid != nritems &&
4238 leaf_space_used(l, mid, nritems - mid) +
4239 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4240 if (data_size && !tried_avoid_double)
4241 goto push_for_double;
4247 if (leaf_space_used(l, 0, mid) + data_size >
4248 BTRFS_LEAF_DATA_SIZE(root)) {
4249 if (!extend && data_size && slot == 0) {
4251 } else if ((extend || !data_size) && slot == 0) {
4255 if (mid != nritems &&
4256 leaf_space_used(l, mid, nritems - mid) +
4257 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4258 if (data_size && !tried_avoid_double)
4259 goto push_for_double;
4267 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4269 btrfs_item_key(l, &disk_key, mid);
4271 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4272 &disk_key, 0, l->start, 0);
4274 return PTR_ERR(right);
4276 root_add_used(root, root->nodesize);
4278 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4279 btrfs_set_header_bytenr(right, right->start);
4280 btrfs_set_header_generation(right, trans->transid);
4281 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4282 btrfs_set_header_owner(right, root->root_key.objectid);
4283 btrfs_set_header_level(right, 0);
4284 write_extent_buffer(right, fs_info->fsid,
4285 btrfs_header_fsid(), BTRFS_FSID_SIZE);
4287 write_extent_buffer(right, fs_info->chunk_tree_uuid,
4288 btrfs_header_chunk_tree_uuid(right),
4293 btrfs_set_header_nritems(right, 0);
4294 insert_ptr(trans, root, path, &disk_key, right->start,
4295 path->slots[1] + 1, 1);
4296 btrfs_tree_unlock(path->nodes[0]);
4297 free_extent_buffer(path->nodes[0]);
4298 path->nodes[0] = right;
4300 path->slots[1] += 1;
4302 btrfs_set_header_nritems(right, 0);
4303 insert_ptr(trans, root, path, &disk_key, right->start,
4305 btrfs_tree_unlock(path->nodes[0]);
4306 free_extent_buffer(path->nodes[0]);
4307 path->nodes[0] = right;
4309 if (path->slots[1] == 0)
4310 fixup_low_keys(fs_info, path, &disk_key, 1);
4312 btrfs_mark_buffer_dirty(right);
4316 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4319 BUG_ON(num_doubles != 0);
4327 push_for_double_split(trans, root, path, data_size);
4328 tried_avoid_double = 1;
4329 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4334 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4335 struct btrfs_root *root,
4336 struct btrfs_path *path, int ins_len)
4338 struct btrfs_key key;
4339 struct extent_buffer *leaf;
4340 struct btrfs_file_extent_item *fi;
4345 leaf = path->nodes[0];
4346 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4348 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4349 key.type != BTRFS_EXTENT_CSUM_KEY);
4351 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4354 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4355 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4356 fi = btrfs_item_ptr(leaf, path->slots[0],
4357 struct btrfs_file_extent_item);
4358 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4360 btrfs_release_path(path);
4362 path->keep_locks = 1;
4363 path->search_for_split = 1;
4364 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4365 path->search_for_split = 0;
4372 leaf = path->nodes[0];
4373 /* if our item isn't there, return now */
4374 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4377 /* the leaf has changed, it now has room. return now */
4378 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4381 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4382 fi = btrfs_item_ptr(leaf, path->slots[0],
4383 struct btrfs_file_extent_item);
4384 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4388 btrfs_set_path_blocking(path);
4389 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4393 path->keep_locks = 0;
4394 btrfs_unlock_up_safe(path, 1);
4397 path->keep_locks = 0;
4401 static noinline int split_item(struct btrfs_trans_handle *trans,
4402 struct btrfs_root *root,
4403 struct btrfs_path *path,
4404 struct btrfs_key *new_key,
4405 unsigned long split_offset)
4407 struct extent_buffer *leaf;
4408 struct btrfs_item *item;
4409 struct btrfs_item *new_item;
4415 struct btrfs_disk_key disk_key;
4417 leaf = path->nodes[0];
4418 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4420 btrfs_set_path_blocking(path);
4422 item = btrfs_item_nr(path->slots[0]);
4423 orig_offset = btrfs_item_offset(leaf, item);
4424 item_size = btrfs_item_size(leaf, item);
4426 buf = kmalloc(item_size, GFP_NOFS);
4430 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4431 path->slots[0]), item_size);
4433 slot = path->slots[0] + 1;
4434 nritems = btrfs_header_nritems(leaf);
4435 if (slot != nritems) {
4436 /* shift the items */
4437 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4438 btrfs_item_nr_offset(slot),
4439 (nritems - slot) * sizeof(struct btrfs_item));
4442 btrfs_cpu_key_to_disk(&disk_key, new_key);
4443 btrfs_set_item_key(leaf, &disk_key, slot);
4445 new_item = btrfs_item_nr(slot);
4447 btrfs_set_item_offset(leaf, new_item, orig_offset);
4448 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4450 btrfs_set_item_offset(leaf, item,
4451 orig_offset + item_size - split_offset);
4452 btrfs_set_item_size(leaf, item, split_offset);
4454 btrfs_set_header_nritems(leaf, nritems + 1);
4456 /* write the data for the start of the original item */
4457 write_extent_buffer(leaf, buf,
4458 btrfs_item_ptr_offset(leaf, path->slots[0]),
4461 /* write the data for the new item */
4462 write_extent_buffer(leaf, buf + split_offset,
4463 btrfs_item_ptr_offset(leaf, slot),
4464 item_size - split_offset);
4465 btrfs_mark_buffer_dirty(leaf);
4467 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4473 * This function splits a single item into two items,
4474 * giving 'new_key' to the new item and splitting the
4475 * old one at split_offset (from the start of the item).
4477 * The path may be released by this operation. After
4478 * the split, the path is pointing to the old item. The
4479 * new item is going to be in the same node as the old one.
4481 * Note, the item being split must be smaller enough to live alone on
4482 * a tree block with room for one extra struct btrfs_item
4484 * This allows us to split the item in place, keeping a lock on the
4485 * leaf the entire time.
4487 int btrfs_split_item(struct btrfs_trans_handle *trans,
4488 struct btrfs_root *root,
4489 struct btrfs_path *path,
4490 struct btrfs_key *new_key,
4491 unsigned long split_offset)
4494 ret = setup_leaf_for_split(trans, root, path,
4495 sizeof(struct btrfs_item));
4499 ret = split_item(trans, root, path, new_key, split_offset);
4504 * This function duplicate a item, giving 'new_key' to the new item.
4505 * It guarantees both items live in the same tree leaf and the new item
4506 * is contiguous with the original item.
4508 * This allows us to split file extent in place, keeping a lock on the
4509 * leaf the entire time.
4511 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4512 struct btrfs_root *root,
4513 struct btrfs_path *path,
4514 struct btrfs_key *new_key)
4516 struct extent_buffer *leaf;
4520 leaf = path->nodes[0];
4521 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4522 ret = setup_leaf_for_split(trans, root, path,
4523 item_size + sizeof(struct btrfs_item));
4528 setup_items_for_insert(root, path, new_key, &item_size,
4529 item_size, item_size +
4530 sizeof(struct btrfs_item), 1);
4531 leaf = path->nodes[0];
4532 memcpy_extent_buffer(leaf,
4533 btrfs_item_ptr_offset(leaf, path->slots[0]),
4534 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4540 * make the item pointed to by the path smaller. new_size indicates
4541 * how small to make it, and from_end tells us if we just chop bytes
4542 * off the end of the item or if we shift the item to chop bytes off
4545 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4546 u32 new_size, int from_end)
4549 struct extent_buffer *leaf;
4550 struct btrfs_item *item;
4552 unsigned int data_end;
4553 unsigned int old_data_start;
4554 unsigned int old_size;
4555 unsigned int size_diff;
4557 struct btrfs_map_token token;
4559 btrfs_init_map_token(&token);
4561 leaf = path->nodes[0];
4562 slot = path->slots[0];
4564 old_size = btrfs_item_size_nr(leaf, slot);
4565 if (old_size == new_size)
4568 nritems = btrfs_header_nritems(leaf);
4569 data_end = leaf_data_end(root, leaf);
4571 old_data_start = btrfs_item_offset_nr(leaf, slot);
4573 size_diff = old_size - new_size;
4576 BUG_ON(slot >= nritems);
4579 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4581 /* first correct the data pointers */
4582 for (i = slot; i < nritems; i++) {
4584 item = btrfs_item_nr(i);
4586 ioff = btrfs_token_item_offset(leaf, item, &token);
4587 btrfs_set_token_item_offset(leaf, item,
4588 ioff + size_diff, &token);
4591 /* shift the data */
4593 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4594 data_end + size_diff, btrfs_leaf_data(leaf) +
4595 data_end, old_data_start + new_size - data_end);
4597 struct btrfs_disk_key disk_key;
4600 btrfs_item_key(leaf, &disk_key, slot);
4602 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4604 struct btrfs_file_extent_item *fi;
4606 fi = btrfs_item_ptr(leaf, slot,
4607 struct btrfs_file_extent_item);
4608 fi = (struct btrfs_file_extent_item *)(
4609 (unsigned long)fi - size_diff);
4611 if (btrfs_file_extent_type(leaf, fi) ==
4612 BTRFS_FILE_EXTENT_INLINE) {
4613 ptr = btrfs_item_ptr_offset(leaf, slot);
4614 memmove_extent_buffer(leaf, ptr,
4616 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4620 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4621 data_end + size_diff, btrfs_leaf_data(leaf) +
4622 data_end, old_data_start - data_end);
4624 offset = btrfs_disk_key_offset(&disk_key);
4625 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4626 btrfs_set_item_key(leaf, &disk_key, slot);
4628 fixup_low_keys(root->fs_info, path, &disk_key, 1);
4631 item = btrfs_item_nr(slot);
4632 btrfs_set_item_size(leaf, item, new_size);
4633 btrfs_mark_buffer_dirty(leaf);
4635 if (btrfs_leaf_free_space(root, leaf) < 0) {
4636 btrfs_print_leaf(root, leaf);
4642 * make the item pointed to by the path bigger, data_size is the added size.
4644 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4648 struct extent_buffer *leaf;
4649 struct btrfs_item *item;
4651 unsigned int data_end;
4652 unsigned int old_data;
4653 unsigned int old_size;
4655 struct btrfs_map_token token;
4657 btrfs_init_map_token(&token);
4659 leaf = path->nodes[0];
4661 nritems = btrfs_header_nritems(leaf);
4662 data_end = leaf_data_end(root, leaf);
4664 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4665 btrfs_print_leaf(root, leaf);
4668 slot = path->slots[0];
4669 old_data = btrfs_item_end_nr(leaf, slot);
4672 if (slot >= nritems) {
4673 btrfs_print_leaf(root, leaf);
4674 btrfs_crit(root->fs_info, "slot %d too large, nritems %d",
4680 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4682 /* first correct the data pointers */
4683 for (i = slot; i < nritems; i++) {
4685 item = btrfs_item_nr(i);
4687 ioff = btrfs_token_item_offset(leaf, item, &token);
4688 btrfs_set_token_item_offset(leaf, item,
4689 ioff - data_size, &token);
4692 /* shift the data */
4693 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4694 data_end - data_size, btrfs_leaf_data(leaf) +
4695 data_end, old_data - data_end);
4697 data_end = old_data;
4698 old_size = btrfs_item_size_nr(leaf, slot);
4699 item = btrfs_item_nr(slot);
4700 btrfs_set_item_size(leaf, item, old_size + data_size);
4701 btrfs_mark_buffer_dirty(leaf);
4703 if (btrfs_leaf_free_space(root, leaf) < 0) {
4704 btrfs_print_leaf(root, leaf);
4710 * this is a helper for btrfs_insert_empty_items, the main goal here is
4711 * to save stack depth by doing the bulk of the work in a function
4712 * that doesn't call btrfs_search_slot
4714 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4715 struct btrfs_key *cpu_key, u32 *data_size,
4716 u32 total_data, u32 total_size, int nr)
4718 struct btrfs_item *item;
4721 unsigned int data_end;
4722 struct btrfs_disk_key disk_key;
4723 struct extent_buffer *leaf;
4725 struct btrfs_map_token token;
4727 if (path->slots[0] == 0) {
4728 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4729 fixup_low_keys(root->fs_info, path, &disk_key, 1);
4731 btrfs_unlock_up_safe(path, 1);
4733 btrfs_init_map_token(&token);
4735 leaf = path->nodes[0];
4736 slot = path->slots[0];
4738 nritems = btrfs_header_nritems(leaf);
4739 data_end = leaf_data_end(root, leaf);
4741 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4742 btrfs_print_leaf(root, leaf);
4743 btrfs_crit(root->fs_info, "not enough freespace need %u have %d",
4744 total_size, btrfs_leaf_free_space(root, leaf));
4748 if (slot != nritems) {
4749 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4751 if (old_data < data_end) {
4752 btrfs_print_leaf(root, leaf);
4753 btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d",
4754 slot, old_data, data_end);
4758 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4760 /* first correct the data pointers */
4761 for (i = slot; i < nritems; i++) {
4764 item = btrfs_item_nr( i);
4765 ioff = btrfs_token_item_offset(leaf, item, &token);
4766 btrfs_set_token_item_offset(leaf, item,
4767 ioff - total_data, &token);
4769 /* shift the items */
4770 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4771 btrfs_item_nr_offset(slot),
4772 (nritems - slot) * sizeof(struct btrfs_item));
4774 /* shift the data */
4775 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4776 data_end - total_data, btrfs_leaf_data(leaf) +
4777 data_end, old_data - data_end);
4778 data_end = old_data;
4781 /* setup the item for the new data */
4782 for (i = 0; i < nr; i++) {
4783 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4784 btrfs_set_item_key(leaf, &disk_key, slot + i);
4785 item = btrfs_item_nr(slot + i);
4786 btrfs_set_token_item_offset(leaf, item,
4787 data_end - data_size[i], &token);
4788 data_end -= data_size[i];
4789 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4792 btrfs_set_header_nritems(leaf, nritems + nr);
4793 btrfs_mark_buffer_dirty(leaf);
4795 if (btrfs_leaf_free_space(root, leaf) < 0) {
4796 btrfs_print_leaf(root, leaf);
4802 * Given a key and some data, insert items into the tree.
4803 * This does all the path init required, making room in the tree if needed.
4805 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4806 struct btrfs_root *root,
4807 struct btrfs_path *path,
4808 struct btrfs_key *cpu_key, u32 *data_size,
4817 for (i = 0; i < nr; i++)
4818 total_data += data_size[i];
4820 total_size = total_data + (nr * sizeof(struct btrfs_item));
4821 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4827 slot = path->slots[0];
4830 setup_items_for_insert(root, path, cpu_key, data_size,
4831 total_data, total_size, nr);
4836 * Given a key and some data, insert an item into the tree.
4837 * This does all the path init required, making room in the tree if needed.
4839 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4840 *root, struct btrfs_key *cpu_key, void *data, u32
4844 struct btrfs_path *path;
4845 struct extent_buffer *leaf;
4848 path = btrfs_alloc_path();
4851 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4853 leaf = path->nodes[0];
4854 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4855 write_extent_buffer(leaf, data, ptr, data_size);
4856 btrfs_mark_buffer_dirty(leaf);
4858 btrfs_free_path(path);
4863 * delete the pointer from a given node.
4865 * the tree should have been previously balanced so the deletion does not
4868 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4869 int level, int slot)
4871 struct extent_buffer *parent = path->nodes[level];
4875 nritems = btrfs_header_nritems(parent);
4876 if (slot != nritems - 1) {
4878 tree_mod_log_eb_move(root->fs_info, parent, slot,
4879 slot + 1, nritems - slot - 1);
4880 memmove_extent_buffer(parent,
4881 btrfs_node_key_ptr_offset(slot),
4882 btrfs_node_key_ptr_offset(slot + 1),
4883 sizeof(struct btrfs_key_ptr) *
4884 (nritems - slot - 1));
4886 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4887 MOD_LOG_KEY_REMOVE, GFP_NOFS);
4892 btrfs_set_header_nritems(parent, nritems);
4893 if (nritems == 0 && parent == root->node) {
4894 BUG_ON(btrfs_header_level(root->node) != 1);
4895 /* just turn the root into a leaf and break */
4896 btrfs_set_header_level(root->node, 0);
4897 } else if (slot == 0) {
4898 struct btrfs_disk_key disk_key;
4900 btrfs_node_key(parent, &disk_key, 0);
4901 fixup_low_keys(root->fs_info, path, &disk_key, level + 1);
4903 btrfs_mark_buffer_dirty(parent);
4907 * a helper function to delete the leaf pointed to by path->slots[1] and
4910 * This deletes the pointer in path->nodes[1] and frees the leaf
4911 * block extent. zero is returned if it all worked out, < 0 otherwise.
4913 * The path must have already been setup for deleting the leaf, including
4914 * all the proper balancing. path->nodes[1] must be locked.
4916 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4917 struct btrfs_root *root,
4918 struct btrfs_path *path,
4919 struct extent_buffer *leaf)
4921 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4922 del_ptr(root, path, 1, path->slots[1]);
4925 * btrfs_free_extent is expensive, we want to make sure we
4926 * aren't holding any locks when we call it
4928 btrfs_unlock_up_safe(path, 0);
4930 root_sub_used(root, leaf->len);
4932 extent_buffer_get(leaf);
4933 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4934 free_extent_buffer_stale(leaf);
4937 * delete the item at the leaf level in path. If that empties
4938 * the leaf, remove it from the tree
4940 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4941 struct btrfs_path *path, int slot, int nr)
4943 struct extent_buffer *leaf;
4944 struct btrfs_item *item;
4951 struct btrfs_map_token token;
4953 btrfs_init_map_token(&token);
4955 leaf = path->nodes[0];
4956 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4958 for (i = 0; i < nr; i++)
4959 dsize += btrfs_item_size_nr(leaf, slot + i);
4961 nritems = btrfs_header_nritems(leaf);
4963 if (slot + nr != nritems) {
4964 int data_end = leaf_data_end(root, leaf);
4966 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4968 btrfs_leaf_data(leaf) + data_end,
4969 last_off - data_end);
4971 for (i = slot + nr; i < nritems; i++) {
4974 item = btrfs_item_nr(i);
4975 ioff = btrfs_token_item_offset(leaf, item, &token);
4976 btrfs_set_token_item_offset(leaf, item,
4977 ioff + dsize, &token);
4980 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4981 btrfs_item_nr_offset(slot + nr),
4982 sizeof(struct btrfs_item) *
4983 (nritems - slot - nr));
4985 btrfs_set_header_nritems(leaf, nritems - nr);
4988 /* delete the leaf if we've emptied it */
4990 if (leaf == root->node) {
4991 btrfs_set_header_level(leaf, 0);
4993 btrfs_set_path_blocking(path);
4994 clean_tree_block(trans, root->fs_info, leaf);
4995 btrfs_del_leaf(trans, root, path, leaf);
4998 int used = leaf_space_used(leaf, 0, nritems);
5000 struct btrfs_disk_key disk_key;
5002 btrfs_item_key(leaf, &disk_key, 0);
5003 fixup_low_keys(root->fs_info, path, &disk_key, 1);
5006 /* delete the leaf if it is mostly empty */
5007 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
5008 /* push_leaf_left fixes the path.
5009 * make sure the path still points to our leaf
5010 * for possible call to del_ptr below
5012 slot = path->slots[1];
5013 extent_buffer_get(leaf);
5015 btrfs_set_path_blocking(path);
5016 wret = push_leaf_left(trans, root, path, 1, 1,
5018 if (wret < 0 && wret != -ENOSPC)
5021 if (path->nodes[0] == leaf &&
5022 btrfs_header_nritems(leaf)) {
5023 wret = push_leaf_right(trans, root, path, 1,
5025 if (wret < 0 && wret != -ENOSPC)
5029 if (btrfs_header_nritems(leaf) == 0) {
5030 path->slots[1] = slot;
5031 btrfs_del_leaf(trans, root, path, leaf);
5032 free_extent_buffer(leaf);
5035 /* if we're still in the path, make sure
5036 * we're dirty. Otherwise, one of the
5037 * push_leaf functions must have already
5038 * dirtied this buffer
5040 if (path->nodes[0] == leaf)
5041 btrfs_mark_buffer_dirty(leaf);
5042 free_extent_buffer(leaf);
5045 btrfs_mark_buffer_dirty(leaf);
5052 * search the tree again to find a leaf with lesser keys
5053 * returns 0 if it found something or 1 if there are no lesser leaves.
5054 * returns < 0 on io errors.
5056 * This may release the path, and so you may lose any locks held at the
5059 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5061 struct btrfs_key key;
5062 struct btrfs_disk_key found_key;
5065 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5067 if (key.offset > 0) {
5069 } else if (key.type > 0) {
5071 key.offset = (u64)-1;
5072 } else if (key.objectid > 0) {
5075 key.offset = (u64)-1;
5080 btrfs_release_path(path);
5081 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5084 btrfs_item_key(path->nodes[0], &found_key, 0);
5085 ret = comp_keys(&found_key, &key);
5087 * We might have had an item with the previous key in the tree right
5088 * before we released our path. And after we released our path, that
5089 * item might have been pushed to the first slot (0) of the leaf we
5090 * were holding due to a tree balance. Alternatively, an item with the
5091 * previous key can exist as the only element of a leaf (big fat item).
5092 * Therefore account for these 2 cases, so that our callers (like
5093 * btrfs_previous_item) don't miss an existing item with a key matching
5094 * the previous key we computed above.
5102 * A helper function to walk down the tree starting at min_key, and looking
5103 * for nodes or leaves that are have a minimum transaction id.
5104 * This is used by the btree defrag code, and tree logging
5106 * This does not cow, but it does stuff the starting key it finds back
5107 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5108 * key and get a writable path.
5110 * This does lock as it descends, and path->keep_locks should be set
5111 * to 1 by the caller.
5113 * This honors path->lowest_level to prevent descent past a given level
5116 * min_trans indicates the oldest transaction that you are interested
5117 * in walking through. Any nodes or leaves older than min_trans are
5118 * skipped over (without reading them).
5120 * returns zero if something useful was found, < 0 on error and 1 if there
5121 * was nothing in the tree that matched the search criteria.
5123 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5124 struct btrfs_path *path,
5127 struct extent_buffer *cur;
5128 struct btrfs_key found_key;
5134 int keep_locks = path->keep_locks;
5136 path->keep_locks = 1;
5138 cur = btrfs_read_lock_root_node(root);
5139 level = btrfs_header_level(cur);
5140 WARN_ON(path->nodes[level]);
5141 path->nodes[level] = cur;
5142 path->locks[level] = BTRFS_READ_LOCK;
5144 if (btrfs_header_generation(cur) < min_trans) {
5149 nritems = btrfs_header_nritems(cur);
5150 level = btrfs_header_level(cur);
5151 sret = bin_search(cur, min_key, level, &slot);
5153 /* at the lowest level, we're done, setup the path and exit */
5154 if (level == path->lowest_level) {
5155 if (slot >= nritems)
5158 path->slots[level] = slot;
5159 btrfs_item_key_to_cpu(cur, &found_key, slot);
5162 if (sret && slot > 0)
5165 * check this node pointer against the min_trans parameters.
5166 * If it is too old, old, skip to the next one.
5168 while (slot < nritems) {
5171 gen = btrfs_node_ptr_generation(cur, slot);
5172 if (gen < min_trans) {
5180 * we didn't find a candidate key in this node, walk forward
5181 * and find another one
5183 if (slot >= nritems) {
5184 path->slots[level] = slot;
5185 btrfs_set_path_blocking(path);
5186 sret = btrfs_find_next_key(root, path, min_key, level,
5189 btrfs_release_path(path);
5195 /* save our key for returning back */
5196 btrfs_node_key_to_cpu(cur, &found_key, slot);
5197 path->slots[level] = slot;
5198 if (level == path->lowest_level) {
5202 btrfs_set_path_blocking(path);
5203 cur = read_node_slot(root, cur, slot);
5204 BUG_ON(!cur); /* -ENOMEM */
5206 btrfs_tree_read_lock(cur);
5208 path->locks[level - 1] = BTRFS_READ_LOCK;
5209 path->nodes[level - 1] = cur;
5210 unlock_up(path, level, 1, 0, NULL);
5211 btrfs_clear_path_blocking(path, NULL, 0);
5214 path->keep_locks = keep_locks;
5216 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5217 btrfs_set_path_blocking(path);
5218 memcpy(min_key, &found_key, sizeof(found_key));
5223 static void tree_move_down(struct btrfs_root *root,
5224 struct btrfs_path *path,
5225 int *level, int root_level)
5227 BUG_ON(*level == 0);
5228 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
5229 path->slots[*level]);
5230 path->slots[*level - 1] = 0;
5234 static int tree_move_next_or_upnext(struct btrfs_root *root,
5235 struct btrfs_path *path,
5236 int *level, int root_level)
5240 nritems = btrfs_header_nritems(path->nodes[*level]);
5242 path->slots[*level]++;
5244 while (path->slots[*level] >= nritems) {
5245 if (*level == root_level)
5249 path->slots[*level] = 0;
5250 free_extent_buffer(path->nodes[*level]);
5251 path->nodes[*level] = NULL;
5253 path->slots[*level]++;
5255 nritems = btrfs_header_nritems(path->nodes[*level]);
5262 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5265 static int tree_advance(struct btrfs_root *root,
5266 struct btrfs_path *path,
5267 int *level, int root_level,
5269 struct btrfs_key *key)
5273 if (*level == 0 || !allow_down) {
5274 ret = tree_move_next_or_upnext(root, path, level, root_level);
5276 tree_move_down(root, path, level, root_level);
5281 btrfs_item_key_to_cpu(path->nodes[*level], key,
5282 path->slots[*level]);
5284 btrfs_node_key_to_cpu(path->nodes[*level], key,
5285 path->slots[*level]);
5290 static int tree_compare_item(struct btrfs_root *left_root,
5291 struct btrfs_path *left_path,
5292 struct btrfs_path *right_path,
5297 unsigned long off1, off2;
5299 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5300 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5304 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5305 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5306 right_path->slots[0]);
5308 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5310 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5317 #define ADVANCE_ONLY_NEXT -1
5320 * This function compares two trees and calls the provided callback for
5321 * every changed/new/deleted item it finds.
5322 * If shared tree blocks are encountered, whole subtrees are skipped, making
5323 * the compare pretty fast on snapshotted subvolumes.
5325 * This currently works on commit roots only. As commit roots are read only,
5326 * we don't do any locking. The commit roots are protected with transactions.
5327 * Transactions are ended and rejoined when a commit is tried in between.
5329 * This function checks for modifications done to the trees while comparing.
5330 * If it detects a change, it aborts immediately.
5332 int btrfs_compare_trees(struct btrfs_root *left_root,
5333 struct btrfs_root *right_root,
5334 btrfs_changed_cb_t changed_cb, void *ctx)
5338 struct btrfs_path *left_path = NULL;
5339 struct btrfs_path *right_path = NULL;
5340 struct btrfs_key left_key;
5341 struct btrfs_key right_key;
5342 char *tmp_buf = NULL;
5343 int left_root_level;
5344 int right_root_level;
5347 int left_end_reached;
5348 int right_end_reached;
5356 left_path = btrfs_alloc_path();
5361 right_path = btrfs_alloc_path();
5367 tmp_buf = kmalloc(left_root->nodesize, GFP_KERNEL | __GFP_NOWARN);
5369 tmp_buf = vmalloc(left_root->nodesize);
5376 left_path->search_commit_root = 1;
5377 left_path->skip_locking = 1;
5378 right_path->search_commit_root = 1;
5379 right_path->skip_locking = 1;
5382 * Strategy: Go to the first items of both trees. Then do
5384 * If both trees are at level 0
5385 * Compare keys of current items
5386 * If left < right treat left item as new, advance left tree
5388 * If left > right treat right item as deleted, advance right tree
5390 * If left == right do deep compare of items, treat as changed if
5391 * needed, advance both trees and repeat
5392 * If both trees are at the same level but not at level 0
5393 * Compare keys of current nodes/leafs
5394 * If left < right advance left tree and repeat
5395 * If left > right advance right tree and repeat
5396 * If left == right compare blockptrs of the next nodes/leafs
5397 * If they match advance both trees but stay at the same level
5399 * If they don't match advance both trees while allowing to go
5401 * If tree levels are different
5402 * Advance the tree that needs it and repeat
5404 * Advancing a tree means:
5405 * If we are at level 0, try to go to the next slot. If that's not
5406 * possible, go one level up and repeat. Stop when we found a level
5407 * where we could go to the next slot. We may at this point be on a
5410 * If we are not at level 0 and not on shared tree blocks, go one
5413 * If we are not at level 0 and on shared tree blocks, go one slot to
5414 * the right if possible or go up and right.
5417 down_read(&left_root->fs_info->commit_root_sem);
5418 left_level = btrfs_header_level(left_root->commit_root);
5419 left_root_level = left_level;
5420 left_path->nodes[left_level] = left_root->commit_root;
5421 extent_buffer_get(left_path->nodes[left_level]);
5423 right_level = btrfs_header_level(right_root->commit_root);
5424 right_root_level = right_level;
5425 right_path->nodes[right_level] = right_root->commit_root;
5426 extent_buffer_get(right_path->nodes[right_level]);
5427 up_read(&left_root->fs_info->commit_root_sem);
5429 if (left_level == 0)
5430 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5431 &left_key, left_path->slots[left_level]);
5433 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5434 &left_key, left_path->slots[left_level]);
5435 if (right_level == 0)
5436 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5437 &right_key, right_path->slots[right_level]);
5439 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5440 &right_key, right_path->slots[right_level]);
5442 left_end_reached = right_end_reached = 0;
5443 advance_left = advance_right = 0;
5446 if (advance_left && !left_end_reached) {
5447 ret = tree_advance(left_root, left_path, &left_level,
5449 advance_left != ADVANCE_ONLY_NEXT,
5452 left_end_reached = ADVANCE;
5455 if (advance_right && !right_end_reached) {
5456 ret = tree_advance(right_root, right_path, &right_level,
5458 advance_right != ADVANCE_ONLY_NEXT,
5461 right_end_reached = ADVANCE;
5465 if (left_end_reached && right_end_reached) {
5468 } else if (left_end_reached) {
5469 if (right_level == 0) {
5470 ret = changed_cb(left_root, right_root,
5471 left_path, right_path,
5473 BTRFS_COMPARE_TREE_DELETED,
5478 advance_right = ADVANCE;
5480 } else if (right_end_reached) {
5481 if (left_level == 0) {
5482 ret = changed_cb(left_root, right_root,
5483 left_path, right_path,
5485 BTRFS_COMPARE_TREE_NEW,
5490 advance_left = ADVANCE;
5494 if (left_level == 0 && right_level == 0) {
5495 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5497 ret = changed_cb(left_root, right_root,
5498 left_path, right_path,
5500 BTRFS_COMPARE_TREE_NEW,
5504 advance_left = ADVANCE;
5505 } else if (cmp > 0) {
5506 ret = changed_cb(left_root, right_root,
5507 left_path, right_path,
5509 BTRFS_COMPARE_TREE_DELETED,
5513 advance_right = ADVANCE;
5515 enum btrfs_compare_tree_result result;
5517 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5518 ret = tree_compare_item(left_root, left_path,
5519 right_path, tmp_buf);
5521 result = BTRFS_COMPARE_TREE_CHANGED;
5523 result = BTRFS_COMPARE_TREE_SAME;
5524 ret = changed_cb(left_root, right_root,
5525 left_path, right_path,
5526 &left_key, result, ctx);
5529 advance_left = ADVANCE;
5530 advance_right = ADVANCE;
5532 } else if (left_level == right_level) {
5533 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5535 advance_left = ADVANCE;
5536 } else if (cmp > 0) {
5537 advance_right = ADVANCE;
5539 left_blockptr = btrfs_node_blockptr(
5540 left_path->nodes[left_level],
5541 left_path->slots[left_level]);
5542 right_blockptr = btrfs_node_blockptr(
5543 right_path->nodes[right_level],
5544 right_path->slots[right_level]);
5545 left_gen = btrfs_node_ptr_generation(
5546 left_path->nodes[left_level],
5547 left_path->slots[left_level]);
5548 right_gen = btrfs_node_ptr_generation(
5549 right_path->nodes[right_level],
5550 right_path->slots[right_level]);
5551 if (left_blockptr == right_blockptr &&
5552 left_gen == right_gen) {
5554 * As we're on a shared block, don't
5555 * allow to go deeper.
5557 advance_left = ADVANCE_ONLY_NEXT;
5558 advance_right = ADVANCE_ONLY_NEXT;
5560 advance_left = ADVANCE;
5561 advance_right = ADVANCE;
5564 } else if (left_level < right_level) {
5565 advance_right = ADVANCE;
5567 advance_left = ADVANCE;
5572 btrfs_free_path(left_path);
5573 btrfs_free_path(right_path);
5579 * this is similar to btrfs_next_leaf, but does not try to preserve
5580 * and fixup the path. It looks for and returns the next key in the
5581 * tree based on the current path and the min_trans parameters.
5583 * 0 is returned if another key is found, < 0 if there are any errors
5584 * and 1 is returned if there are no higher keys in the tree
5586 * path->keep_locks should be set to 1 on the search made before
5587 * calling this function.
5589 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5590 struct btrfs_key *key, int level, u64 min_trans)
5593 struct extent_buffer *c;
5595 WARN_ON(!path->keep_locks);
5596 while (level < BTRFS_MAX_LEVEL) {
5597 if (!path->nodes[level])
5600 slot = path->slots[level] + 1;
5601 c = path->nodes[level];
5603 if (slot >= btrfs_header_nritems(c)) {
5606 struct btrfs_key cur_key;
5607 if (level + 1 >= BTRFS_MAX_LEVEL ||
5608 !path->nodes[level + 1])
5611 if (path->locks[level + 1]) {
5616 slot = btrfs_header_nritems(c) - 1;
5618 btrfs_item_key_to_cpu(c, &cur_key, slot);
5620 btrfs_node_key_to_cpu(c, &cur_key, slot);
5622 orig_lowest = path->lowest_level;
5623 btrfs_release_path(path);
5624 path->lowest_level = level;
5625 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5627 path->lowest_level = orig_lowest;
5631 c = path->nodes[level];
5632 slot = path->slots[level];
5639 btrfs_item_key_to_cpu(c, key, slot);
5641 u64 gen = btrfs_node_ptr_generation(c, slot);
5643 if (gen < min_trans) {
5647 btrfs_node_key_to_cpu(c, key, slot);
5655 * search the tree again to find a leaf with greater keys
5656 * returns 0 if it found something or 1 if there are no greater leaves.
5657 * returns < 0 on io errors.
5659 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5661 return btrfs_next_old_leaf(root, path, 0);
5664 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5669 struct extent_buffer *c;
5670 struct extent_buffer *next;
5671 struct btrfs_key key;
5674 int old_spinning = path->leave_spinning;
5675 int next_rw_lock = 0;
5677 nritems = btrfs_header_nritems(path->nodes[0]);
5681 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5686 btrfs_release_path(path);
5688 path->keep_locks = 1;
5689 path->leave_spinning = 1;
5692 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5694 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5695 path->keep_locks = 0;
5700 nritems = btrfs_header_nritems(path->nodes[0]);
5702 * by releasing the path above we dropped all our locks. A balance
5703 * could have added more items next to the key that used to be
5704 * at the very end of the block. So, check again here and
5705 * advance the path if there are now more items available.
5707 if (nritems > 0 && path->slots[0] < nritems - 1) {
5714 * So the above check misses one case:
5715 * - after releasing the path above, someone has removed the item that
5716 * used to be at the very end of the block, and balance between leafs
5717 * gets another one with bigger key.offset to replace it.
5719 * This one should be returned as well, or we can get leaf corruption
5720 * later(esp. in __btrfs_drop_extents()).
5722 * And a bit more explanation about this check,
5723 * with ret > 0, the key isn't found, the path points to the slot
5724 * where it should be inserted, so the path->slots[0] item must be the
5727 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5732 while (level < BTRFS_MAX_LEVEL) {
5733 if (!path->nodes[level]) {
5738 slot = path->slots[level] + 1;
5739 c = path->nodes[level];
5740 if (slot >= btrfs_header_nritems(c)) {
5742 if (level == BTRFS_MAX_LEVEL) {
5750 btrfs_tree_unlock_rw(next, next_rw_lock);
5751 free_extent_buffer(next);
5755 next_rw_lock = path->locks[level];
5756 ret = read_block_for_search(NULL, root, path, &next, level,
5762 btrfs_release_path(path);
5766 if (!path->skip_locking) {
5767 ret = btrfs_try_tree_read_lock(next);
5768 if (!ret && time_seq) {
5770 * If we don't get the lock, we may be racing
5771 * with push_leaf_left, holding that lock while
5772 * itself waiting for the leaf we've currently
5773 * locked. To solve this situation, we give up
5774 * on our lock and cycle.
5776 free_extent_buffer(next);
5777 btrfs_release_path(path);
5782 btrfs_set_path_blocking(path);
5783 btrfs_tree_read_lock(next);
5784 btrfs_clear_path_blocking(path, next,
5787 next_rw_lock = BTRFS_READ_LOCK;
5791 path->slots[level] = slot;
5794 c = path->nodes[level];
5795 if (path->locks[level])
5796 btrfs_tree_unlock_rw(c, path->locks[level]);
5798 free_extent_buffer(c);
5799 path->nodes[level] = next;
5800 path->slots[level] = 0;
5801 if (!path->skip_locking)
5802 path->locks[level] = next_rw_lock;
5806 ret = read_block_for_search(NULL, root, path, &next, level,
5812 btrfs_release_path(path);
5816 if (!path->skip_locking) {
5817 ret = btrfs_try_tree_read_lock(next);
5819 btrfs_set_path_blocking(path);
5820 btrfs_tree_read_lock(next);
5821 btrfs_clear_path_blocking(path, next,
5824 next_rw_lock = BTRFS_READ_LOCK;
5829 unlock_up(path, 0, 1, 0, NULL);
5830 path->leave_spinning = old_spinning;
5832 btrfs_set_path_blocking(path);
5838 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5839 * searching until it gets past min_objectid or finds an item of 'type'
5841 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5843 int btrfs_previous_item(struct btrfs_root *root,
5844 struct btrfs_path *path, u64 min_objectid,
5847 struct btrfs_key found_key;
5848 struct extent_buffer *leaf;
5853 if (path->slots[0] == 0) {
5854 btrfs_set_path_blocking(path);
5855 ret = btrfs_prev_leaf(root, path);
5861 leaf = path->nodes[0];
5862 nritems = btrfs_header_nritems(leaf);
5865 if (path->slots[0] == nritems)
5868 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5869 if (found_key.objectid < min_objectid)
5871 if (found_key.type == type)
5873 if (found_key.objectid == min_objectid &&
5874 found_key.type < type)
5881 * search in extent tree to find a previous Metadata/Data extent item with
5884 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5886 int btrfs_previous_extent_item(struct btrfs_root *root,
5887 struct btrfs_path *path, u64 min_objectid)
5889 struct btrfs_key found_key;
5890 struct extent_buffer *leaf;
5895 if (path->slots[0] == 0) {
5896 btrfs_set_path_blocking(path);
5897 ret = btrfs_prev_leaf(root, path);
5903 leaf = path->nodes[0];
5904 nritems = btrfs_header_nritems(leaf);
5907 if (path->slots[0] == nritems)
5910 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5911 if (found_key.objectid < min_objectid)
5913 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5914 found_key.type == BTRFS_METADATA_ITEM_KEY)
5916 if (found_key.objectid == min_objectid &&
5917 found_key.type < BTRFS_EXTENT_ITEM_KEY)