2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/rbtree.h>
24 #include "transaction.h"
25 #include "print-tree.h"
28 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
29 *root, struct btrfs_path *path, int level);
30 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
31 *root, struct btrfs_key *ins_key,
32 struct btrfs_path *path, int data_size, int extend);
33 static int push_node_left(struct btrfs_trans_handle *trans,
34 struct btrfs_root *root, struct extent_buffer *dst,
35 struct extent_buffer *src, int empty);
36 static int balance_node_right(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 struct extent_buffer *dst_buf,
39 struct extent_buffer *src_buf);
40 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
42 static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
43 struct extent_buffer *eb);
45 struct btrfs_path *btrfs_alloc_path(void)
47 struct btrfs_path *path;
48 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
53 * set all locked nodes in the path to blocking locks. This should
54 * be done before scheduling
56 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
59 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
60 if (!p->nodes[i] || !p->locks[i])
62 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
63 if (p->locks[i] == BTRFS_READ_LOCK)
64 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
65 else if (p->locks[i] == BTRFS_WRITE_LOCK)
66 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
71 * reset all the locked nodes in the patch to spinning locks.
73 * held is used to keep lockdep happy, when lockdep is enabled
74 * we set held to a blocking lock before we go around and
75 * retake all the spinlocks in the path. You can safely use NULL
78 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
79 struct extent_buffer *held, int held_rw)
83 #ifdef CONFIG_DEBUG_LOCK_ALLOC
84 /* lockdep really cares that we take all of these spinlocks
85 * in the right order. If any of the locks in the path are not
86 * currently blocking, it is going to complain. So, make really
87 * really sure by forcing the path to blocking before we clear
91 btrfs_set_lock_blocking_rw(held, held_rw);
92 if (held_rw == BTRFS_WRITE_LOCK)
93 held_rw = BTRFS_WRITE_LOCK_BLOCKING;
94 else if (held_rw == BTRFS_READ_LOCK)
95 held_rw = BTRFS_READ_LOCK_BLOCKING;
97 btrfs_set_path_blocking(p);
100 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
101 if (p->nodes[i] && p->locks[i]) {
102 btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
103 if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
104 p->locks[i] = BTRFS_WRITE_LOCK;
105 else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
106 p->locks[i] = BTRFS_READ_LOCK;
110 #ifdef CONFIG_DEBUG_LOCK_ALLOC
112 btrfs_clear_lock_blocking_rw(held, held_rw);
116 /* this also releases the path */
117 void btrfs_free_path(struct btrfs_path *p)
121 btrfs_release_path(p);
122 kmem_cache_free(btrfs_path_cachep, p);
126 * path release drops references on the extent buffers in the path
127 * and it drops any locks held by this path
129 * It is safe to call this on paths that no locks or extent buffers held.
131 noinline void btrfs_release_path(struct btrfs_path *p)
135 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
140 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
143 free_extent_buffer(p->nodes[i]);
149 * safely gets a reference on the root node of a tree. A lock
150 * is not taken, so a concurrent writer may put a different node
151 * at the root of the tree. See btrfs_lock_root_node for the
154 * The extent buffer returned by this has a reference taken, so
155 * it won't disappear. It may stop being the root of the tree
156 * at any time because there are no locks held.
158 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
160 struct extent_buffer *eb;
164 eb = rcu_dereference(root->node);
167 * RCU really hurts here, we could free up the root node because
168 * it was cow'ed but we may not get the new root node yet so do
169 * the inc_not_zero dance and if it doesn't work then
170 * synchronize_rcu and try again.
172 if (atomic_inc_not_zero(&eb->refs)) {
182 /* loop around taking references on and locking the root node of the
183 * tree until you end up with a lock on the root. A locked buffer
184 * is returned, with a reference held.
186 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
188 struct extent_buffer *eb;
191 eb = btrfs_root_node(root);
193 if (eb == root->node)
195 btrfs_tree_unlock(eb);
196 free_extent_buffer(eb);
201 /* loop around taking references on and locking the root node of the
202 * tree until you end up with a lock on the root. A locked buffer
203 * is returned, with a reference held.
205 static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
207 struct extent_buffer *eb;
210 eb = btrfs_root_node(root);
211 btrfs_tree_read_lock(eb);
212 if (eb == root->node)
214 btrfs_tree_read_unlock(eb);
215 free_extent_buffer(eb);
220 /* cowonly root (everything not a reference counted cow subvolume), just get
221 * put onto a simple dirty list. transaction.c walks this to make sure they
222 * get properly updated on disk.
224 static void add_root_to_dirty_list(struct btrfs_root *root)
226 spin_lock(&root->fs_info->trans_lock);
227 if (test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state) &&
228 list_empty(&root->dirty_list)) {
229 list_add(&root->dirty_list,
230 &root->fs_info->dirty_cowonly_roots);
232 spin_unlock(&root->fs_info->trans_lock);
236 * used by snapshot creation to make a copy of a root for a tree with
237 * a given objectid. The buffer with the new root node is returned in
238 * cow_ret, and this func returns zero on success or a negative error code.
240 int btrfs_copy_root(struct btrfs_trans_handle *trans,
241 struct btrfs_root *root,
242 struct extent_buffer *buf,
243 struct extent_buffer **cow_ret, u64 new_root_objectid)
245 struct extent_buffer *cow;
248 struct btrfs_disk_key disk_key;
250 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
251 trans->transid != root->fs_info->running_transaction->transid);
252 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
253 trans->transid != root->last_trans);
255 level = btrfs_header_level(buf);
257 btrfs_item_key(buf, &disk_key, 0);
259 btrfs_node_key(buf, &disk_key, 0);
261 cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
262 new_root_objectid, &disk_key, level,
267 copy_extent_buffer(cow, buf, 0, 0, cow->len);
268 btrfs_set_header_bytenr(cow, cow->start);
269 btrfs_set_header_generation(cow, trans->transid);
270 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
271 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
272 BTRFS_HEADER_FLAG_RELOC);
273 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
274 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
276 btrfs_set_header_owner(cow, new_root_objectid);
278 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
281 WARN_ON(btrfs_header_generation(buf) > trans->transid);
282 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
283 ret = btrfs_inc_ref(trans, root, cow, 1);
285 ret = btrfs_inc_ref(trans, root, cow, 0);
290 btrfs_mark_buffer_dirty(cow);
299 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
300 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
302 MOD_LOG_ROOT_REPLACE,
305 struct tree_mod_move {
310 struct tree_mod_root {
315 struct tree_mod_elem {
317 u64 index; /* shifted logical */
321 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
324 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
327 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
328 struct btrfs_disk_key key;
331 /* this is used for op == MOD_LOG_MOVE_KEYS */
332 struct tree_mod_move move;
334 /* this is used for op == MOD_LOG_ROOT_REPLACE */
335 struct tree_mod_root old_root;
338 static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
340 read_lock(&fs_info->tree_mod_log_lock);
343 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
345 read_unlock(&fs_info->tree_mod_log_lock);
348 static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
350 write_lock(&fs_info->tree_mod_log_lock);
353 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
355 write_unlock(&fs_info->tree_mod_log_lock);
359 * Pull a new tree mod seq number for our operation.
361 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
363 return atomic64_inc_return(&fs_info->tree_mod_seq);
367 * This adds a new blocker to the tree mod log's blocker list if the @elem
368 * passed does not already have a sequence number set. So when a caller expects
369 * to record tree modifications, it should ensure to set elem->seq to zero
370 * before calling btrfs_get_tree_mod_seq.
371 * Returns a fresh, unused tree log modification sequence number, even if no new
374 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
375 struct seq_list *elem)
377 tree_mod_log_write_lock(fs_info);
378 spin_lock(&fs_info->tree_mod_seq_lock);
380 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
381 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
383 spin_unlock(&fs_info->tree_mod_seq_lock);
384 tree_mod_log_write_unlock(fs_info);
389 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
390 struct seq_list *elem)
392 struct rb_root *tm_root;
393 struct rb_node *node;
394 struct rb_node *next;
395 struct seq_list *cur_elem;
396 struct tree_mod_elem *tm;
397 u64 min_seq = (u64)-1;
398 u64 seq_putting = elem->seq;
403 spin_lock(&fs_info->tree_mod_seq_lock);
404 list_del(&elem->list);
407 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
408 if (cur_elem->seq < min_seq) {
409 if (seq_putting > cur_elem->seq) {
411 * blocker with lower sequence number exists, we
412 * cannot remove anything from the log
414 spin_unlock(&fs_info->tree_mod_seq_lock);
417 min_seq = cur_elem->seq;
420 spin_unlock(&fs_info->tree_mod_seq_lock);
423 * anything that's lower than the lowest existing (read: blocked)
424 * sequence number can be removed from the tree.
426 tree_mod_log_write_lock(fs_info);
427 tm_root = &fs_info->tree_mod_log;
428 for (node = rb_first(tm_root); node; node = next) {
429 next = rb_next(node);
430 tm = container_of(node, struct tree_mod_elem, node);
431 if (tm->seq > min_seq)
433 rb_erase(node, tm_root);
436 tree_mod_log_write_unlock(fs_info);
440 * key order of the log:
443 * the index is the shifted logical of the *new* root node for root replace
444 * operations, or the shifted logical of the affected block for all other
447 * Note: must be called with write lock (tree_mod_log_write_lock).
450 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
452 struct rb_root *tm_root;
453 struct rb_node **new;
454 struct rb_node *parent = NULL;
455 struct tree_mod_elem *cur;
459 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
461 tm_root = &fs_info->tree_mod_log;
462 new = &tm_root->rb_node;
464 cur = container_of(*new, struct tree_mod_elem, node);
466 if (cur->index < tm->index)
467 new = &((*new)->rb_left);
468 else if (cur->index > tm->index)
469 new = &((*new)->rb_right);
470 else if (cur->seq < tm->seq)
471 new = &((*new)->rb_left);
472 else if (cur->seq > tm->seq)
473 new = &((*new)->rb_right);
478 rb_link_node(&tm->node, parent, new);
479 rb_insert_color(&tm->node, tm_root);
484 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
485 * returns zero with the tree_mod_log_lock acquired. The caller must hold
486 * this until all tree mod log insertions are recorded in the rb tree and then
487 * call tree_mod_log_write_unlock() to release.
489 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
490 struct extent_buffer *eb) {
492 if (list_empty(&(fs_info)->tree_mod_seq_list))
494 if (eb && btrfs_header_level(eb) == 0)
497 tree_mod_log_write_lock(fs_info);
498 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
499 tree_mod_log_write_unlock(fs_info);
506 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
507 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
508 struct extent_buffer *eb)
511 if (list_empty(&(fs_info)->tree_mod_seq_list))
513 if (eb && btrfs_header_level(eb) == 0)
519 static struct tree_mod_elem *
520 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
521 enum mod_log_op op, gfp_t flags)
523 struct tree_mod_elem *tm;
525 tm = kzalloc(sizeof(*tm), flags);
529 tm->index = eb->start >> PAGE_CACHE_SHIFT;
530 if (op != MOD_LOG_KEY_ADD) {
531 btrfs_node_key(eb, &tm->key, slot);
532 tm->blockptr = btrfs_node_blockptr(eb, slot);
536 tm->generation = btrfs_node_ptr_generation(eb, slot);
537 RB_CLEAR_NODE(&tm->node);
543 tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
544 struct extent_buffer *eb, int slot,
545 enum mod_log_op op, gfp_t flags)
547 struct tree_mod_elem *tm;
550 if (!tree_mod_need_log(fs_info, eb))
553 tm = alloc_tree_mod_elem(eb, slot, op, flags);
557 if (tree_mod_dont_log(fs_info, eb)) {
562 ret = __tree_mod_log_insert(fs_info, tm);
563 tree_mod_log_write_unlock(fs_info);
571 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
572 struct extent_buffer *eb, int dst_slot, int src_slot,
573 int nr_items, gfp_t flags)
575 struct tree_mod_elem *tm = NULL;
576 struct tree_mod_elem **tm_list = NULL;
581 if (!tree_mod_need_log(fs_info, eb))
584 tm_list = kzalloc(nr_items * sizeof(struct tree_mod_elem *), flags);
588 tm = kzalloc(sizeof(*tm), flags);
594 tm->index = eb->start >> PAGE_CACHE_SHIFT;
596 tm->move.dst_slot = dst_slot;
597 tm->move.nr_items = nr_items;
598 tm->op = MOD_LOG_MOVE_KEYS;
600 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
601 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
602 MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags);
609 if (tree_mod_dont_log(fs_info, eb))
614 * When we override something during the move, we log these removals.
615 * This can only happen when we move towards the beginning of the
616 * buffer, i.e. dst_slot < src_slot.
618 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
619 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
624 ret = __tree_mod_log_insert(fs_info, tm);
627 tree_mod_log_write_unlock(fs_info);
632 for (i = 0; i < nr_items; i++) {
633 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
634 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
638 tree_mod_log_write_unlock(fs_info);
646 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
647 struct tree_mod_elem **tm_list,
653 for (i = nritems - 1; i >= 0; i--) {
654 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
656 for (j = nritems - 1; j > i; j--)
657 rb_erase(&tm_list[j]->node,
658 &fs_info->tree_mod_log);
667 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
668 struct extent_buffer *old_root,
669 struct extent_buffer *new_root, gfp_t flags,
672 struct tree_mod_elem *tm = NULL;
673 struct tree_mod_elem **tm_list = NULL;
678 if (!tree_mod_need_log(fs_info, NULL))
681 if (log_removal && btrfs_header_level(old_root) > 0) {
682 nritems = btrfs_header_nritems(old_root);
683 tm_list = kzalloc(nritems * sizeof(struct tree_mod_elem *),
689 for (i = 0; i < nritems; i++) {
690 tm_list[i] = alloc_tree_mod_elem(old_root, i,
691 MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags);
699 tm = kzalloc(sizeof(*tm), flags);
705 tm->index = new_root->start >> PAGE_CACHE_SHIFT;
706 tm->old_root.logical = old_root->start;
707 tm->old_root.level = btrfs_header_level(old_root);
708 tm->generation = btrfs_header_generation(old_root);
709 tm->op = MOD_LOG_ROOT_REPLACE;
711 if (tree_mod_dont_log(fs_info, NULL))
715 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
717 ret = __tree_mod_log_insert(fs_info, tm);
719 tree_mod_log_write_unlock(fs_info);
728 for (i = 0; i < nritems; i++)
737 static struct tree_mod_elem *
738 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
741 struct rb_root *tm_root;
742 struct rb_node *node;
743 struct tree_mod_elem *cur = NULL;
744 struct tree_mod_elem *found = NULL;
745 u64 index = start >> PAGE_CACHE_SHIFT;
747 tree_mod_log_read_lock(fs_info);
748 tm_root = &fs_info->tree_mod_log;
749 node = tm_root->rb_node;
751 cur = container_of(node, struct tree_mod_elem, node);
752 if (cur->index < index) {
753 node = node->rb_left;
754 } else if (cur->index > index) {
755 node = node->rb_right;
756 } else if (cur->seq < min_seq) {
757 node = node->rb_left;
758 } else if (!smallest) {
759 /* we want the node with the highest seq */
761 BUG_ON(found->seq > cur->seq);
763 node = node->rb_left;
764 } else if (cur->seq > min_seq) {
765 /* we want the node with the smallest seq */
767 BUG_ON(found->seq < cur->seq);
769 node = node->rb_right;
775 tree_mod_log_read_unlock(fs_info);
781 * this returns the element from the log with the smallest time sequence
782 * value that's in the log (the oldest log item). any element with a time
783 * sequence lower than min_seq will be ignored.
785 static struct tree_mod_elem *
786 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
789 return __tree_mod_log_search(fs_info, start, min_seq, 1);
793 * this returns the element from the log with the largest time sequence
794 * value that's in the log (the most recent log item). any element with
795 * a time sequence lower than min_seq will be ignored.
797 static struct tree_mod_elem *
798 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
800 return __tree_mod_log_search(fs_info, start, min_seq, 0);
804 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
805 struct extent_buffer *src, unsigned long dst_offset,
806 unsigned long src_offset, int nr_items)
809 struct tree_mod_elem **tm_list = NULL;
810 struct tree_mod_elem **tm_list_add, **tm_list_rem;
814 if (!tree_mod_need_log(fs_info, NULL))
817 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
820 tm_list = kzalloc(nr_items * 2 * sizeof(struct tree_mod_elem *),
825 tm_list_add = tm_list;
826 tm_list_rem = tm_list + nr_items;
827 for (i = 0; i < nr_items; i++) {
828 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
829 MOD_LOG_KEY_REMOVE, GFP_NOFS);
830 if (!tm_list_rem[i]) {
835 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
836 MOD_LOG_KEY_ADD, GFP_NOFS);
837 if (!tm_list_add[i]) {
843 if (tree_mod_dont_log(fs_info, NULL))
847 for (i = 0; i < nr_items; i++) {
848 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
851 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
856 tree_mod_log_write_unlock(fs_info);
862 for (i = 0; i < nr_items * 2; i++) {
863 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
864 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
868 tree_mod_log_write_unlock(fs_info);
875 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
876 int dst_offset, int src_offset, int nr_items)
879 ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
885 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
886 struct extent_buffer *eb, int slot, int atomic)
890 ret = tree_mod_log_insert_key(fs_info, eb, slot,
892 atomic ? GFP_ATOMIC : GFP_NOFS);
897 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
899 struct tree_mod_elem **tm_list = NULL;
904 if (btrfs_header_level(eb) == 0)
907 if (!tree_mod_need_log(fs_info, NULL))
910 nritems = btrfs_header_nritems(eb);
911 tm_list = kzalloc(nritems * sizeof(struct tree_mod_elem *),
916 for (i = 0; i < nritems; i++) {
917 tm_list[i] = alloc_tree_mod_elem(eb, i,
918 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
925 if (tree_mod_dont_log(fs_info, eb))
928 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
929 tree_mod_log_write_unlock(fs_info);
937 for (i = 0; i < nritems; i++)
945 tree_mod_log_set_root_pointer(struct btrfs_root *root,
946 struct extent_buffer *new_root_node,
950 ret = tree_mod_log_insert_root(root->fs_info, root->node,
951 new_root_node, GFP_NOFS, log_removal);
956 * check if the tree block can be shared by multiple trees
958 int btrfs_block_can_be_shared(struct btrfs_root *root,
959 struct extent_buffer *buf)
962 * Tree blocks not in refernece counted trees and tree roots
963 * are never shared. If a block was allocated after the last
964 * snapshot and the block was not allocated by tree relocation,
965 * we know the block is not shared.
967 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
968 buf != root->node && buf != root->commit_root &&
969 (btrfs_header_generation(buf) <=
970 btrfs_root_last_snapshot(&root->root_item) ||
971 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
973 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
974 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
975 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
981 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
982 struct btrfs_root *root,
983 struct extent_buffer *buf,
984 struct extent_buffer *cow,
994 * Backrefs update rules:
996 * Always use full backrefs for extent pointers in tree block
997 * allocated by tree relocation.
999 * If a shared tree block is no longer referenced by its owner
1000 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
1001 * use full backrefs for extent pointers in tree block.
1003 * If a tree block is been relocating
1004 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1005 * use full backrefs for extent pointers in tree block.
1006 * The reason for this is some operations (such as drop tree)
1007 * are only allowed for blocks use full backrefs.
1010 if (btrfs_block_can_be_shared(root, buf)) {
1011 ret = btrfs_lookup_extent_info(trans, root, buf->start,
1012 btrfs_header_level(buf), 1,
1018 btrfs_std_error(root->fs_info, ret);
1023 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1024 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1025 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1030 owner = btrfs_header_owner(buf);
1031 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
1032 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
1035 if ((owner == root->root_key.objectid ||
1036 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
1037 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1038 ret = btrfs_inc_ref(trans, root, buf, 1);
1039 BUG_ON(ret); /* -ENOMEM */
1041 if (root->root_key.objectid ==
1042 BTRFS_TREE_RELOC_OBJECTID) {
1043 ret = btrfs_dec_ref(trans, root, buf, 0);
1044 BUG_ON(ret); /* -ENOMEM */
1045 ret = btrfs_inc_ref(trans, root, cow, 1);
1046 BUG_ON(ret); /* -ENOMEM */
1048 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
1051 if (root->root_key.objectid ==
1052 BTRFS_TREE_RELOC_OBJECTID)
1053 ret = btrfs_inc_ref(trans, root, cow, 1);
1055 ret = btrfs_inc_ref(trans, root, cow, 0);
1056 BUG_ON(ret); /* -ENOMEM */
1058 if (new_flags != 0) {
1059 int level = btrfs_header_level(buf);
1061 ret = btrfs_set_disk_extent_flags(trans, root,
1064 new_flags, level, 0);
1069 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
1070 if (root->root_key.objectid ==
1071 BTRFS_TREE_RELOC_OBJECTID)
1072 ret = btrfs_inc_ref(trans, root, cow, 1);
1074 ret = btrfs_inc_ref(trans, root, cow, 0);
1075 BUG_ON(ret); /* -ENOMEM */
1076 ret = btrfs_dec_ref(trans, root, buf, 1);
1077 BUG_ON(ret); /* -ENOMEM */
1079 clean_tree_block(trans, root, buf);
1086 * does the dirty work in cow of a single block. The parent block (if
1087 * supplied) is updated to point to the new cow copy. The new buffer is marked
1088 * dirty and returned locked. If you modify the block it needs to be marked
1091 * search_start -- an allocation hint for the new block
1093 * empty_size -- a hint that you plan on doing more cow. This is the size in
1094 * bytes the allocator should try to find free next to the block it returns.
1095 * This is just a hint and may be ignored by the allocator.
1097 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1098 struct btrfs_root *root,
1099 struct extent_buffer *buf,
1100 struct extent_buffer *parent, int parent_slot,
1101 struct extent_buffer **cow_ret,
1102 u64 search_start, u64 empty_size)
1104 struct btrfs_disk_key disk_key;
1105 struct extent_buffer *cow;
1108 int unlock_orig = 0;
1111 if (*cow_ret == buf)
1114 btrfs_assert_tree_locked(buf);
1116 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1117 trans->transid != root->fs_info->running_transaction->transid);
1118 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1119 trans->transid != root->last_trans);
1121 level = btrfs_header_level(buf);
1124 btrfs_item_key(buf, &disk_key, 0);
1126 btrfs_node_key(buf, &disk_key, 0);
1128 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1130 parent_start = parent->start;
1136 cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
1137 root->root_key.objectid, &disk_key,
1138 level, search_start, empty_size);
1140 return PTR_ERR(cow);
1142 /* cow is set to blocking by btrfs_init_new_buffer */
1144 copy_extent_buffer(cow, buf, 0, 0, cow->len);
1145 btrfs_set_header_bytenr(cow, cow->start);
1146 btrfs_set_header_generation(cow, trans->transid);
1147 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1148 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1149 BTRFS_HEADER_FLAG_RELOC);
1150 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1151 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1153 btrfs_set_header_owner(cow, root->root_key.objectid);
1155 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
1158 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1160 btrfs_abort_transaction(trans, root, ret);
1164 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1165 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1170 if (buf == root->node) {
1171 WARN_ON(parent && parent != buf);
1172 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1173 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1174 parent_start = buf->start;
1178 extent_buffer_get(cow);
1179 tree_mod_log_set_root_pointer(root, cow, 1);
1180 rcu_assign_pointer(root->node, cow);
1182 btrfs_free_tree_block(trans, root, buf, parent_start,
1184 free_extent_buffer(buf);
1185 add_root_to_dirty_list(root);
1187 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1188 parent_start = parent->start;
1192 WARN_ON(trans->transid != btrfs_header_generation(parent));
1193 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1194 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1195 btrfs_set_node_blockptr(parent, parent_slot,
1197 btrfs_set_node_ptr_generation(parent, parent_slot,
1199 btrfs_mark_buffer_dirty(parent);
1201 ret = tree_mod_log_free_eb(root->fs_info, buf);
1203 btrfs_abort_transaction(trans, root, ret);
1207 btrfs_free_tree_block(trans, root, buf, parent_start,
1211 btrfs_tree_unlock(buf);
1212 free_extent_buffer_stale(buf);
1213 btrfs_mark_buffer_dirty(cow);
1219 * returns the logical address of the oldest predecessor of the given root.
1220 * entries older than time_seq are ignored.
1222 static struct tree_mod_elem *
1223 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1224 struct extent_buffer *eb_root, u64 time_seq)
1226 struct tree_mod_elem *tm;
1227 struct tree_mod_elem *found = NULL;
1228 u64 root_logical = eb_root->start;
1235 * the very last operation that's logged for a root is the replacement
1236 * operation (if it is replaced at all). this has the index of the *new*
1237 * root, making it the very first operation that's logged for this root.
1240 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1245 * if there are no tree operation for the oldest root, we simply
1246 * return it. this should only happen if that (old) root is at
1253 * if there's an operation that's not a root replacement, we
1254 * found the oldest version of our root. normally, we'll find a
1255 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1257 if (tm->op != MOD_LOG_ROOT_REPLACE)
1261 root_logical = tm->old_root.logical;
1265 /* if there's no old root to return, return what we found instead */
1273 * tm is a pointer to the first operation to rewind within eb. then, all
1274 * previous operations will be rewinded (until we reach something older than
1278 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1279 u64 time_seq, struct tree_mod_elem *first_tm)
1282 struct rb_node *next;
1283 struct tree_mod_elem *tm = first_tm;
1284 unsigned long o_dst;
1285 unsigned long o_src;
1286 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1288 n = btrfs_header_nritems(eb);
1289 tree_mod_log_read_lock(fs_info);
1290 while (tm && tm->seq >= time_seq) {
1292 * all the operations are recorded with the operator used for
1293 * the modification. as we're going backwards, we do the
1294 * opposite of each operation here.
1297 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1298 BUG_ON(tm->slot < n);
1300 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1301 case MOD_LOG_KEY_REMOVE:
1302 btrfs_set_node_key(eb, &tm->key, tm->slot);
1303 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1304 btrfs_set_node_ptr_generation(eb, tm->slot,
1308 case MOD_LOG_KEY_REPLACE:
1309 BUG_ON(tm->slot >= n);
1310 btrfs_set_node_key(eb, &tm->key, tm->slot);
1311 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1312 btrfs_set_node_ptr_generation(eb, tm->slot,
1315 case MOD_LOG_KEY_ADD:
1316 /* if a move operation is needed it's in the log */
1319 case MOD_LOG_MOVE_KEYS:
1320 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1321 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1322 memmove_extent_buffer(eb, o_dst, o_src,
1323 tm->move.nr_items * p_size);
1325 case MOD_LOG_ROOT_REPLACE:
1327 * this operation is special. for roots, this must be
1328 * handled explicitly before rewinding.
1329 * for non-roots, this operation may exist if the node
1330 * was a root: root A -> child B; then A gets empty and
1331 * B is promoted to the new root. in the mod log, we'll
1332 * have a root-replace operation for B, a tree block
1333 * that is no root. we simply ignore that operation.
1337 next = rb_next(&tm->node);
1340 tm = container_of(next, struct tree_mod_elem, node);
1341 if (tm->index != first_tm->index)
1344 tree_mod_log_read_unlock(fs_info);
1345 btrfs_set_header_nritems(eb, n);
1349 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1350 * is returned. If rewind operations happen, a fresh buffer is returned. The
1351 * returned buffer is always read-locked. If the returned buffer is not the
1352 * input buffer, the lock on the input buffer is released and the input buffer
1353 * is freed (its refcount is decremented).
1355 static struct extent_buffer *
1356 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1357 struct extent_buffer *eb, u64 time_seq)
1359 struct extent_buffer *eb_rewin;
1360 struct tree_mod_elem *tm;
1365 if (btrfs_header_level(eb) == 0)
1368 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1372 btrfs_set_path_blocking(path);
1373 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1375 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1376 BUG_ON(tm->slot != 0);
1377 eb_rewin = alloc_dummy_extent_buffer(eb->start,
1378 fs_info->tree_root->nodesize);
1380 btrfs_tree_read_unlock_blocking(eb);
1381 free_extent_buffer(eb);
1384 btrfs_set_header_bytenr(eb_rewin, eb->start);
1385 btrfs_set_header_backref_rev(eb_rewin,
1386 btrfs_header_backref_rev(eb));
1387 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1388 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1390 eb_rewin = btrfs_clone_extent_buffer(eb);
1392 btrfs_tree_read_unlock_blocking(eb);
1393 free_extent_buffer(eb);
1398 btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
1399 btrfs_tree_read_unlock_blocking(eb);
1400 free_extent_buffer(eb);
1402 extent_buffer_get(eb_rewin);
1403 btrfs_tree_read_lock(eb_rewin);
1404 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1405 WARN_ON(btrfs_header_nritems(eb_rewin) >
1406 BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1412 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1413 * value. If there are no changes, the current root->root_node is returned. If
1414 * anything changed in between, there's a fresh buffer allocated on which the
1415 * rewind operations are done. In any case, the returned buffer is read locked.
1416 * Returns NULL on error (with no locks held).
1418 static inline struct extent_buffer *
1419 get_old_root(struct btrfs_root *root, u64 time_seq)
1421 struct tree_mod_elem *tm;
1422 struct extent_buffer *eb = NULL;
1423 struct extent_buffer *eb_root;
1424 struct extent_buffer *old;
1425 struct tree_mod_root *old_root = NULL;
1426 u64 old_generation = 0;
1429 eb_root = btrfs_read_lock_root_node(root);
1430 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1434 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1435 old_root = &tm->old_root;
1436 old_generation = tm->generation;
1437 logical = old_root->logical;
1439 logical = eb_root->start;
1442 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1443 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1444 btrfs_tree_read_unlock(eb_root);
1445 free_extent_buffer(eb_root);
1446 old = read_tree_block(root, logical, 0);
1447 if (WARN_ON(!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(logical, root->nodesize);
1460 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1461 eb = btrfs_clone_extent_buffer(eb_root);
1462 btrfs_tree_read_unlock_blocking(eb_root);
1463 free_extent_buffer(eb_root);
1468 extent_buffer_get(eb);
1469 btrfs_tree_read_lock(eb);
1471 btrfs_set_header_bytenr(eb, eb->start);
1472 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1473 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1474 btrfs_set_header_level(eb, old_root->level);
1475 btrfs_set_header_generation(eb, old_generation);
1478 __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1480 WARN_ON(btrfs_header_level(eb) != 0);
1481 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1486 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1488 struct tree_mod_elem *tm;
1490 struct extent_buffer *eb_root = btrfs_root_node(root);
1492 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1493 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1494 level = tm->old_root.level;
1496 level = btrfs_header_level(eb_root);
1498 free_extent_buffer(eb_root);
1503 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1504 struct btrfs_root *root,
1505 struct extent_buffer *buf)
1507 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1508 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
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 commiting 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 cow'd 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)(1024 * 1024 * 1024) - 1);
1564 btrfs_set_lock_blocking(parent);
1565 btrfs_set_lock_blocking(buf);
1567 ret = __btrfs_cow_block(trans, root, buf, parent,
1568 parent_slot, cow_ret, search_start, 0);
1570 trace_btrfs_cow_block(root, buf, *cow_ret);
1576 * helper function for defrag to decide if two blocks pointed to by a
1577 * node are actually close by
1579 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1581 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1583 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1589 * compare two keys in a memcmp fashion
1591 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1593 struct btrfs_key k1;
1595 btrfs_disk_key_to_cpu(&k1, disk);
1597 return btrfs_comp_cpu_keys(&k1, k2);
1601 * same as comp_keys only with two btrfs_key's
1603 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1605 if (k1->objectid > k2->objectid)
1607 if (k1->objectid < k2->objectid)
1609 if (k1->type > k2->type)
1611 if (k1->type < k2->type)
1613 if (k1->offset > k2->offset)
1615 if (k1->offset < k2->offset)
1621 * this is used by the defrag code to go through all the
1622 * leaves pointed to by a node and reallocate them so that
1623 * disk order is close to key order
1625 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1626 struct btrfs_root *root, struct extent_buffer *parent,
1627 int start_slot, u64 *last_ret,
1628 struct btrfs_key *progress)
1630 struct extent_buffer *cur;
1633 u64 search_start = *last_ret;
1643 int progress_passed = 0;
1644 struct btrfs_disk_key disk_key;
1646 parent_level = btrfs_header_level(parent);
1648 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1649 WARN_ON(trans->transid != root->fs_info->generation);
1651 parent_nritems = btrfs_header_nritems(parent);
1652 blocksize = root->nodesize;
1653 end_slot = parent_nritems;
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 - 2) {
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, blocknr, blocksize);
1688 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1691 if (!cur || !uptodate) {
1693 cur = read_tree_block(root, blocknr, gen);
1694 if (!cur || !extent_buffer_uptodate(cur)) {
1695 free_extent_buffer(cur);
1698 } else if (!uptodate) {
1699 err = btrfs_read_buffer(cur, gen);
1701 free_extent_buffer(cur);
1706 if (search_start == 0)
1707 search_start = last_block;
1709 btrfs_tree_lock(cur);
1710 btrfs_set_lock_blocking(cur);
1711 err = __btrfs_cow_block(trans, root, cur, parent, i,
1714 (end_slot - i) * blocksize));
1716 btrfs_tree_unlock(cur);
1717 free_extent_buffer(cur);
1720 search_start = cur->start;
1721 last_block = cur->start;
1722 *last_ret = search_start;
1723 btrfs_tree_unlock(cur);
1724 free_extent_buffer(cur);
1730 * The leaf data grows from end-to-front in the node.
1731 * this returns the address of the start of the last item,
1732 * which is the stop of the leaf data stack
1734 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1735 struct extent_buffer *leaf)
1737 u32 nr = btrfs_header_nritems(leaf);
1739 return BTRFS_LEAF_DATA_SIZE(root);
1740 return btrfs_item_offset_nr(leaf, nr - 1);
1745 * search for key in the extent_buffer. The items start at offset p,
1746 * and they are item_size apart. There are 'max' items in p.
1748 * the slot in the array is returned via slot, and it points to
1749 * the place where you would insert key if it is not found in
1752 * slot may point to max if the key is bigger than all of the keys
1754 static noinline int generic_bin_search(struct extent_buffer *eb,
1756 int item_size, struct btrfs_key *key,
1763 struct btrfs_disk_key *tmp = NULL;
1764 struct btrfs_disk_key unaligned;
1765 unsigned long offset;
1767 unsigned long map_start = 0;
1768 unsigned long map_len = 0;
1771 while (low < high) {
1772 mid = (low + high) / 2;
1773 offset = p + mid * item_size;
1775 if (!kaddr || offset < map_start ||
1776 (offset + sizeof(struct btrfs_disk_key)) >
1777 map_start + map_len) {
1779 err = map_private_extent_buffer(eb, offset,
1780 sizeof(struct btrfs_disk_key),
1781 &kaddr, &map_start, &map_len);
1784 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1787 read_extent_buffer(eb, &unaligned,
1788 offset, sizeof(unaligned));
1793 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1796 ret = comp_keys(tmp, key);
1812 * simple bin_search frontend that does the right thing for
1815 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1816 int level, int *slot)
1819 return generic_bin_search(eb,
1820 offsetof(struct btrfs_leaf, items),
1821 sizeof(struct btrfs_item),
1822 key, btrfs_header_nritems(eb),
1825 return generic_bin_search(eb,
1826 offsetof(struct btrfs_node, ptrs),
1827 sizeof(struct btrfs_key_ptr),
1828 key, btrfs_header_nritems(eb),
1832 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1833 int level, int *slot)
1835 return bin_search(eb, key, level, slot);
1838 static void root_add_used(struct btrfs_root *root, u32 size)
1840 spin_lock(&root->accounting_lock);
1841 btrfs_set_root_used(&root->root_item,
1842 btrfs_root_used(&root->root_item) + size);
1843 spin_unlock(&root->accounting_lock);
1846 static void root_sub_used(struct btrfs_root *root, u32 size)
1848 spin_lock(&root->accounting_lock);
1849 btrfs_set_root_used(&root->root_item,
1850 btrfs_root_used(&root->root_item) - size);
1851 spin_unlock(&root->accounting_lock);
1854 /* given a node and slot number, this reads the blocks it points to. The
1855 * extent buffer is returned with a reference taken (but unlocked).
1856 * NULL is returned on error.
1858 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1859 struct extent_buffer *parent, int slot)
1861 int level = btrfs_header_level(parent);
1862 struct extent_buffer *eb;
1866 if (slot >= btrfs_header_nritems(parent))
1871 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1872 btrfs_node_ptr_generation(parent, slot));
1873 if (eb && !extent_buffer_uptodate(eb)) {
1874 free_extent_buffer(eb);
1882 * node level balancing, used to make sure nodes are in proper order for
1883 * item deletion. We balance from the top down, so we have to make sure
1884 * that a deletion won't leave an node completely empty later on.
1886 static noinline int balance_level(struct btrfs_trans_handle *trans,
1887 struct btrfs_root *root,
1888 struct btrfs_path *path, int level)
1890 struct extent_buffer *right = NULL;
1891 struct extent_buffer *mid;
1892 struct extent_buffer *left = NULL;
1893 struct extent_buffer *parent = NULL;
1897 int orig_slot = path->slots[level];
1903 mid = path->nodes[level];
1905 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1906 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1907 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1909 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1911 if (level < BTRFS_MAX_LEVEL - 1) {
1912 parent = path->nodes[level + 1];
1913 pslot = path->slots[level + 1];
1917 * deal with the case where there is only one pointer in the root
1918 * by promoting the node below to a root
1921 struct extent_buffer *child;
1923 if (btrfs_header_nritems(mid) != 1)
1926 /* promote the child to a root */
1927 child = read_node_slot(root, mid, 0);
1930 btrfs_std_error(root->fs_info, ret);
1934 btrfs_tree_lock(child);
1935 btrfs_set_lock_blocking(child);
1936 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1938 btrfs_tree_unlock(child);
1939 free_extent_buffer(child);
1943 tree_mod_log_set_root_pointer(root, child, 1);
1944 rcu_assign_pointer(root->node, child);
1946 add_root_to_dirty_list(root);
1947 btrfs_tree_unlock(child);
1949 path->locks[level] = 0;
1950 path->nodes[level] = NULL;
1951 clean_tree_block(trans, root, mid);
1952 btrfs_tree_unlock(mid);
1953 /* once for the path */
1954 free_extent_buffer(mid);
1956 root_sub_used(root, mid->len);
1957 btrfs_free_tree_block(trans, root, mid, 0, 1);
1958 /* once for the root ptr */
1959 free_extent_buffer_stale(mid);
1962 if (btrfs_header_nritems(mid) >
1963 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1966 left = read_node_slot(root, parent, pslot - 1);
1968 btrfs_tree_lock(left);
1969 btrfs_set_lock_blocking(left);
1970 wret = btrfs_cow_block(trans, root, left,
1971 parent, pslot - 1, &left);
1977 right = read_node_slot(root, parent, pslot + 1);
1979 btrfs_tree_lock(right);
1980 btrfs_set_lock_blocking(right);
1981 wret = btrfs_cow_block(trans, root, right,
1982 parent, pslot + 1, &right);
1989 /* first, try to make some room in the middle buffer */
1991 orig_slot += btrfs_header_nritems(left);
1992 wret = push_node_left(trans, root, left, mid, 1);
1998 * then try to empty the right most buffer into the middle
2001 wret = push_node_left(trans, root, mid, right, 1);
2002 if (wret < 0 && wret != -ENOSPC)
2004 if (btrfs_header_nritems(right) == 0) {
2005 clean_tree_block(trans, root, right);
2006 btrfs_tree_unlock(right);
2007 del_ptr(root, path, level + 1, pslot + 1);
2008 root_sub_used(root, right->len);
2009 btrfs_free_tree_block(trans, root, right, 0, 1);
2010 free_extent_buffer_stale(right);
2013 struct btrfs_disk_key right_key;
2014 btrfs_node_key(right, &right_key, 0);
2015 tree_mod_log_set_node_key(root->fs_info, parent,
2017 btrfs_set_node_key(parent, &right_key, pslot + 1);
2018 btrfs_mark_buffer_dirty(parent);
2021 if (btrfs_header_nritems(mid) == 1) {
2023 * we're not allowed to leave a node with one item in the
2024 * tree during a delete. A deletion from lower in the tree
2025 * could try to delete the only pointer in this node.
2026 * So, pull some keys from the left.
2027 * There has to be a left pointer at this point because
2028 * otherwise we would have pulled some pointers from the
2033 btrfs_std_error(root->fs_info, ret);
2036 wret = balance_node_right(trans, root, mid, left);
2042 wret = push_node_left(trans, root, left, mid, 1);
2048 if (btrfs_header_nritems(mid) == 0) {
2049 clean_tree_block(trans, root, mid);
2050 btrfs_tree_unlock(mid);
2051 del_ptr(root, path, level + 1, pslot);
2052 root_sub_used(root, mid->len);
2053 btrfs_free_tree_block(trans, root, mid, 0, 1);
2054 free_extent_buffer_stale(mid);
2057 /* update the parent key to reflect our changes */
2058 struct btrfs_disk_key mid_key;
2059 btrfs_node_key(mid, &mid_key, 0);
2060 tree_mod_log_set_node_key(root->fs_info, parent,
2062 btrfs_set_node_key(parent, &mid_key, pslot);
2063 btrfs_mark_buffer_dirty(parent);
2066 /* update the path */
2068 if (btrfs_header_nritems(left) > orig_slot) {
2069 extent_buffer_get(left);
2070 /* left was locked after cow */
2071 path->nodes[level] = left;
2072 path->slots[level + 1] -= 1;
2073 path->slots[level] = orig_slot;
2075 btrfs_tree_unlock(mid);
2076 free_extent_buffer(mid);
2079 orig_slot -= btrfs_header_nritems(left);
2080 path->slots[level] = orig_slot;
2083 /* double check we haven't messed things up */
2085 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2089 btrfs_tree_unlock(right);
2090 free_extent_buffer(right);
2093 if (path->nodes[level] != left)
2094 btrfs_tree_unlock(left);
2095 free_extent_buffer(left);
2100 /* Node balancing for insertion. Here we only split or push nodes around
2101 * when they are completely full. This is also done top down, so we
2102 * have to be pessimistic.
2104 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2105 struct btrfs_root *root,
2106 struct btrfs_path *path, int level)
2108 struct extent_buffer *right = NULL;
2109 struct extent_buffer *mid;
2110 struct extent_buffer *left = NULL;
2111 struct extent_buffer *parent = NULL;
2115 int orig_slot = path->slots[level];
2120 mid = path->nodes[level];
2121 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2123 if (level < BTRFS_MAX_LEVEL - 1) {
2124 parent = path->nodes[level + 1];
2125 pslot = path->slots[level + 1];
2131 left = read_node_slot(root, parent, pslot - 1);
2133 /* first, try to make some room in the middle buffer */
2137 btrfs_tree_lock(left);
2138 btrfs_set_lock_blocking(left);
2140 left_nr = btrfs_header_nritems(left);
2141 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2144 ret = btrfs_cow_block(trans, root, left, parent,
2149 wret = push_node_left(trans, root,
2156 struct btrfs_disk_key disk_key;
2157 orig_slot += left_nr;
2158 btrfs_node_key(mid, &disk_key, 0);
2159 tree_mod_log_set_node_key(root->fs_info, parent,
2161 btrfs_set_node_key(parent, &disk_key, pslot);
2162 btrfs_mark_buffer_dirty(parent);
2163 if (btrfs_header_nritems(left) > orig_slot) {
2164 path->nodes[level] = left;
2165 path->slots[level + 1] -= 1;
2166 path->slots[level] = orig_slot;
2167 btrfs_tree_unlock(mid);
2168 free_extent_buffer(mid);
2171 btrfs_header_nritems(left);
2172 path->slots[level] = orig_slot;
2173 btrfs_tree_unlock(left);
2174 free_extent_buffer(left);
2178 btrfs_tree_unlock(left);
2179 free_extent_buffer(left);
2181 right = read_node_slot(root, parent, pslot + 1);
2184 * then try to empty the right most buffer into the middle
2189 btrfs_tree_lock(right);
2190 btrfs_set_lock_blocking(right);
2192 right_nr = btrfs_header_nritems(right);
2193 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2196 ret = btrfs_cow_block(trans, root, right,
2202 wret = balance_node_right(trans, root,
2209 struct btrfs_disk_key disk_key;
2211 btrfs_node_key(right, &disk_key, 0);
2212 tree_mod_log_set_node_key(root->fs_info, parent,
2214 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2215 btrfs_mark_buffer_dirty(parent);
2217 if (btrfs_header_nritems(mid) <= orig_slot) {
2218 path->nodes[level] = right;
2219 path->slots[level + 1] += 1;
2220 path->slots[level] = orig_slot -
2221 btrfs_header_nritems(mid);
2222 btrfs_tree_unlock(mid);
2223 free_extent_buffer(mid);
2225 btrfs_tree_unlock(right);
2226 free_extent_buffer(right);
2230 btrfs_tree_unlock(right);
2231 free_extent_buffer(right);
2237 * readahead one full node of leaves, finding things that are close
2238 * to the block in 'slot', and triggering ra on them.
2240 static void reada_for_search(struct btrfs_root *root,
2241 struct btrfs_path *path,
2242 int level, int slot, u64 objectid)
2244 struct extent_buffer *node;
2245 struct btrfs_disk_key disk_key;
2251 int direction = path->reada;
2252 struct extent_buffer *eb;
2260 if (!path->nodes[level])
2263 node = path->nodes[level];
2265 search = btrfs_node_blockptr(node, slot);
2266 blocksize = root->nodesize;
2267 eb = btrfs_find_tree_block(root, search, blocksize);
2269 free_extent_buffer(eb);
2275 nritems = btrfs_header_nritems(node);
2279 if (direction < 0) {
2283 } else if (direction > 0) {
2288 if (path->reada < 0 && objectid) {
2289 btrfs_node_key(node, &disk_key, nr);
2290 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2293 search = btrfs_node_blockptr(node, nr);
2294 if ((search <= target && target - search <= 65536) ||
2295 (search > target && search - target <= 65536)) {
2296 gen = btrfs_node_ptr_generation(node, nr);
2297 readahead_tree_block(root, search, blocksize);
2301 if ((nread > 65536 || nscan > 32))
2306 static noinline void reada_for_balance(struct btrfs_root *root,
2307 struct btrfs_path *path, int level)
2311 struct extent_buffer *parent;
2312 struct extent_buffer *eb;
2318 parent = path->nodes[level + 1];
2322 nritems = btrfs_header_nritems(parent);
2323 slot = path->slots[level + 1];
2324 blocksize = root->nodesize;
2327 block1 = btrfs_node_blockptr(parent, slot - 1);
2328 gen = btrfs_node_ptr_generation(parent, slot - 1);
2329 eb = btrfs_find_tree_block(root, block1, blocksize);
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, block2, blocksize);
2343 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2345 free_extent_buffer(eb);
2349 readahead_tree_block(root, block1, blocksize);
2351 readahead_tree_block(root, block2, blocksize);
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)
2454 struct extent_buffer *b = *eb_ret;
2455 struct extent_buffer *tmp;
2458 blocknr = btrfs_node_blockptr(b, slot);
2459 gen = btrfs_node_ptr_generation(b, slot);
2460 blocksize = root->nodesize;
2462 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
2464 /* first we do an atomic uptodate check */
2465 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2470 /* the pages were up to date, but we failed
2471 * the generation number check. Do a full
2472 * read for the generation number that is correct.
2473 * We must do this without dropping locks so
2474 * we can trust our generation number
2476 btrfs_set_path_blocking(p);
2478 /* now we're allowed to do a blocking uptodate check */
2479 ret = btrfs_read_buffer(tmp, gen);
2484 free_extent_buffer(tmp);
2485 btrfs_release_path(p);
2490 * reduce lock contention at high levels
2491 * of the btree by dropping locks before
2492 * we read. Don't release the lock on the current
2493 * level because we need to walk this node to figure
2494 * out which blocks to read.
2496 btrfs_unlock_up_safe(p, level + 1);
2497 btrfs_set_path_blocking(p);
2499 free_extent_buffer(tmp);
2501 reada_for_search(root, p, level, slot, key->objectid);
2503 btrfs_release_path(p);
2506 tmp = read_tree_block(root, blocknr, 0);
2509 * If the read above didn't mark this buffer up to date,
2510 * it will never end up being up to date. Set ret to EIO now
2511 * and give up so that our caller doesn't loop forever
2514 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2516 free_extent_buffer(tmp);
2522 * helper function for btrfs_search_slot. This does all of the checks
2523 * for node-level blocks and does any balancing required based on
2526 * If no extra work was required, zero is returned. If we had to
2527 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2531 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2532 struct btrfs_root *root, struct btrfs_path *p,
2533 struct extent_buffer *b, int level, int ins_len,
2534 int *write_lock_level)
2537 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2538 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2541 if (*write_lock_level < level + 1) {
2542 *write_lock_level = level + 1;
2543 btrfs_release_path(p);
2547 btrfs_set_path_blocking(p);
2548 reada_for_balance(root, p, level);
2549 sret = split_node(trans, root, p, level);
2550 btrfs_clear_path_blocking(p, NULL, 0);
2557 b = p->nodes[level];
2558 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2559 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2562 if (*write_lock_level < level + 1) {
2563 *write_lock_level = level + 1;
2564 btrfs_release_path(p);
2568 btrfs_set_path_blocking(p);
2569 reada_for_balance(root, p, level);
2570 sret = balance_level(trans, root, p, level);
2571 btrfs_clear_path_blocking(p, NULL, 0);
2577 b = p->nodes[level];
2579 btrfs_release_path(p);
2582 BUG_ON(btrfs_header_nritems(b) == 1);
2592 static void key_search_validate(struct extent_buffer *b,
2593 struct btrfs_key *key,
2596 #ifdef CONFIG_BTRFS_ASSERT
2597 struct btrfs_disk_key disk_key;
2599 btrfs_cpu_key_to_disk(&disk_key, key);
2602 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2603 offsetof(struct btrfs_leaf, items[0].key),
2606 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2607 offsetof(struct btrfs_node, ptrs[0].key),
2612 static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2613 int level, int *prev_cmp, int *slot)
2615 if (*prev_cmp != 0) {
2616 *prev_cmp = bin_search(b, key, level, slot);
2620 key_search_validate(b, key, level);
2626 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path,
2627 u64 iobjectid, u64 ioff, u8 key_type,
2628 struct btrfs_key *found_key)
2631 struct btrfs_key key;
2632 struct extent_buffer *eb;
2633 struct btrfs_path *path;
2635 key.type = key_type;
2636 key.objectid = iobjectid;
2639 if (found_path == NULL) {
2640 path = btrfs_alloc_path();
2646 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2647 if ((ret < 0) || (found_key == NULL)) {
2648 if (path != found_path)
2649 btrfs_free_path(path);
2653 eb = path->nodes[0];
2654 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2655 ret = btrfs_next_leaf(fs_root, path);
2658 eb = path->nodes[0];
2661 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2662 if (found_key->type != key.type ||
2663 found_key->objectid != key.objectid)
2670 * look for key in the tree. path is filled in with nodes along the way
2671 * if key is found, we return zero and you can find the item in the leaf
2672 * level of the path (level 0)
2674 * If the key isn't found, the path points to the slot where it should
2675 * be inserted, and 1 is returned. If there are other errors during the
2676 * search a negative error number is returned.
2678 * if ins_len > 0, nodes and leaves will be split as we walk down the
2679 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2682 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2683 *root, struct btrfs_key *key, struct btrfs_path *p, int
2686 struct extent_buffer *b;
2691 int lowest_unlock = 1;
2693 /* everything at write_lock_level or lower must be write locked */
2694 int write_lock_level = 0;
2695 u8 lowest_level = 0;
2696 int min_write_lock_level;
2699 lowest_level = p->lowest_level;
2700 WARN_ON(lowest_level && ins_len > 0);
2701 WARN_ON(p->nodes[0] != NULL);
2702 BUG_ON(!cow && ins_len);
2707 /* when we are removing items, we might have to go up to level
2708 * two as we update tree pointers Make sure we keep write
2709 * for those levels as well
2711 write_lock_level = 2;
2712 } else if (ins_len > 0) {
2714 * for inserting items, make sure we have a write lock on
2715 * level 1 so we can update keys
2717 write_lock_level = 1;
2721 write_lock_level = -1;
2723 if (cow && (p->keep_locks || p->lowest_level))
2724 write_lock_level = BTRFS_MAX_LEVEL;
2726 min_write_lock_level = write_lock_level;
2731 * we try very hard to do read locks on the root
2733 root_lock = BTRFS_READ_LOCK;
2735 if (p->search_commit_root) {
2737 * the commit roots are read only
2738 * so we always do read locks
2740 if (p->need_commit_sem)
2741 down_read(&root->fs_info->commit_root_sem);
2742 b = root->commit_root;
2743 extent_buffer_get(b);
2744 level = btrfs_header_level(b);
2745 if (p->need_commit_sem)
2746 up_read(&root->fs_info->commit_root_sem);
2747 if (!p->skip_locking)
2748 btrfs_tree_read_lock(b);
2750 if (p->skip_locking) {
2751 b = btrfs_root_node(root);
2752 level = btrfs_header_level(b);
2754 /* we don't know the level of the root node
2755 * until we actually have it read locked
2757 b = btrfs_read_lock_root_node(root);
2758 level = btrfs_header_level(b);
2759 if (level <= write_lock_level) {
2760 /* whoops, must trade for write lock */
2761 btrfs_tree_read_unlock(b);
2762 free_extent_buffer(b);
2763 b = btrfs_lock_root_node(root);
2764 root_lock = BTRFS_WRITE_LOCK;
2766 /* the level might have changed, check again */
2767 level = btrfs_header_level(b);
2771 p->nodes[level] = b;
2772 if (!p->skip_locking)
2773 p->locks[level] = root_lock;
2776 level = btrfs_header_level(b);
2779 * setup the path here so we can release it under lock
2780 * contention with the cow code
2784 * if we don't really need to cow this block
2785 * then we don't want to set the path blocking,
2786 * so we test it here
2788 if (!should_cow_block(trans, root, b))
2792 * must have write locks on this node and the
2795 if (level > write_lock_level ||
2796 (level + 1 > write_lock_level &&
2797 level + 1 < BTRFS_MAX_LEVEL &&
2798 p->nodes[level + 1])) {
2799 write_lock_level = level + 1;
2800 btrfs_release_path(p);
2804 btrfs_set_path_blocking(p);
2805 err = btrfs_cow_block(trans, root, b,
2806 p->nodes[level + 1],
2807 p->slots[level + 1], &b);
2814 p->nodes[level] = b;
2815 btrfs_clear_path_blocking(p, NULL, 0);
2818 * we have a lock on b and as long as we aren't changing
2819 * the tree, there is no way to for the items in b to change.
2820 * It is safe to drop the lock on our parent before we
2821 * go through the expensive btree search on b.
2823 * If we're inserting or deleting (ins_len != 0), then we might
2824 * be changing slot zero, which may require changing the parent.
2825 * So, we can't drop the lock until after we know which slot
2826 * we're operating on.
2828 if (!ins_len && !p->keep_locks) {
2831 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2832 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2837 ret = key_search(b, key, level, &prev_cmp, &slot);
2841 if (ret && slot > 0) {
2845 p->slots[level] = slot;
2846 err = setup_nodes_for_search(trans, root, p, b, level,
2847 ins_len, &write_lock_level);
2854 b = p->nodes[level];
2855 slot = p->slots[level];
2858 * slot 0 is special, if we change the key
2859 * we have to update the parent pointer
2860 * which means we must have a write lock
2863 if (slot == 0 && ins_len &&
2864 write_lock_level < level + 1) {
2865 write_lock_level = level + 1;
2866 btrfs_release_path(p);
2870 unlock_up(p, level, lowest_unlock,
2871 min_write_lock_level, &write_lock_level);
2873 if (level == lowest_level) {
2879 err = read_block_for_search(trans, root, p,
2880 &b, level, slot, key, 0);
2888 if (!p->skip_locking) {
2889 level = btrfs_header_level(b);
2890 if (level <= write_lock_level) {
2891 err = btrfs_try_tree_write_lock(b);
2893 btrfs_set_path_blocking(p);
2895 btrfs_clear_path_blocking(p, b,
2898 p->locks[level] = BTRFS_WRITE_LOCK;
2900 err = btrfs_try_tree_read_lock(b);
2902 btrfs_set_path_blocking(p);
2903 btrfs_tree_read_lock(b);
2904 btrfs_clear_path_blocking(p, b,
2907 p->locks[level] = BTRFS_READ_LOCK;
2909 p->nodes[level] = b;
2912 p->slots[level] = slot;
2914 btrfs_leaf_free_space(root, b) < ins_len) {
2915 if (write_lock_level < 1) {
2916 write_lock_level = 1;
2917 btrfs_release_path(p);
2921 btrfs_set_path_blocking(p);
2922 err = split_leaf(trans, root, key,
2923 p, ins_len, ret == 0);
2924 btrfs_clear_path_blocking(p, NULL, 0);
2932 if (!p->search_for_split)
2933 unlock_up(p, level, lowest_unlock,
2934 min_write_lock_level, &write_lock_level);
2941 * we don't really know what they plan on doing with the path
2942 * from here on, so for now just mark it as blocking
2944 if (!p->leave_spinning)
2945 btrfs_set_path_blocking(p);
2947 btrfs_release_path(p);
2952 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2953 * current state of the tree together with the operations recorded in the tree
2954 * modification log to search for the key in a previous version of this tree, as
2955 * denoted by the time_seq parameter.
2957 * Naturally, there is no support for insert, delete or cow operations.
2959 * The resulting path and return value will be set up as if we called
2960 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2962 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2963 struct btrfs_path *p, u64 time_seq)
2965 struct extent_buffer *b;
2970 int lowest_unlock = 1;
2971 u8 lowest_level = 0;
2974 lowest_level = p->lowest_level;
2975 WARN_ON(p->nodes[0] != NULL);
2977 if (p->search_commit_root) {
2979 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2983 b = get_old_root(root, time_seq);
2984 level = btrfs_header_level(b);
2985 p->locks[level] = BTRFS_READ_LOCK;
2988 level = btrfs_header_level(b);
2989 p->nodes[level] = b;
2990 btrfs_clear_path_blocking(p, NULL, 0);
2993 * we have a lock on b and as long as we aren't changing
2994 * the tree, there is no way to for the items in b to change.
2995 * It is safe to drop the lock on our parent before we
2996 * go through the expensive btree search on b.
2998 btrfs_unlock_up_safe(p, level + 1);
3001 * Since we can unwind eb's we want to do a real search every
3005 ret = key_search(b, key, level, &prev_cmp, &slot);
3009 if (ret && slot > 0) {
3013 p->slots[level] = slot;
3014 unlock_up(p, level, lowest_unlock, 0, NULL);
3016 if (level == lowest_level) {
3022 err = read_block_for_search(NULL, root, p, &b, level,
3023 slot, key, time_seq);
3031 level = btrfs_header_level(b);
3032 err = btrfs_try_tree_read_lock(b);
3034 btrfs_set_path_blocking(p);
3035 btrfs_tree_read_lock(b);
3036 btrfs_clear_path_blocking(p, b,
3039 b = tree_mod_log_rewind(root->fs_info, p, b, time_seq);
3044 p->locks[level] = BTRFS_READ_LOCK;
3045 p->nodes[level] = b;
3047 p->slots[level] = slot;
3048 unlock_up(p, level, lowest_unlock, 0, NULL);
3054 if (!p->leave_spinning)
3055 btrfs_set_path_blocking(p);
3057 btrfs_release_path(p);
3063 * helper to use instead of search slot if no exact match is needed but
3064 * instead the next or previous item should be returned.
3065 * When find_higher is true, the next higher item is returned, the next lower
3067 * When return_any and find_higher are both true, and no higher item is found,
3068 * return the next lower instead.
3069 * When return_any is true and find_higher is false, and no lower item is found,
3070 * return the next higher instead.
3071 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3074 int btrfs_search_slot_for_read(struct btrfs_root *root,
3075 struct btrfs_key *key, struct btrfs_path *p,
3076 int find_higher, int return_any)
3079 struct extent_buffer *leaf;
3082 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3086 * a return value of 1 means the path is at the position where the
3087 * item should be inserted. Normally this is the next bigger item,
3088 * but in case the previous item is the last in a leaf, path points
3089 * to the first free slot in the previous leaf, i.e. at an invalid
3095 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3096 ret = btrfs_next_leaf(root, p);
3102 * no higher item found, return the next
3107 btrfs_release_path(p);
3111 if (p->slots[0] == 0) {
3112 ret = btrfs_prev_leaf(root, p);
3117 if (p->slots[0] == btrfs_header_nritems(leaf))
3124 * no lower item found, return the next
3129 btrfs_release_path(p);
3139 * adjust the pointers going up the tree, starting at level
3140 * making sure the right key of each node is points to 'key'.
3141 * This is used after shifting pointers to the left, so it stops
3142 * fixing up pointers when a given leaf/node is not in slot 0 of the
3146 static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
3147 struct btrfs_disk_key *key, int level)
3150 struct extent_buffer *t;
3152 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3153 int tslot = path->slots[i];
3154 if (!path->nodes[i])
3157 tree_mod_log_set_node_key(root->fs_info, t, tslot, 1);
3158 btrfs_set_node_key(t, key, tslot);
3159 btrfs_mark_buffer_dirty(path->nodes[i]);
3168 * This function isn't completely safe. It's the caller's responsibility
3169 * that the new key won't break the order
3171 void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
3172 struct btrfs_key *new_key)
3174 struct btrfs_disk_key disk_key;
3175 struct extent_buffer *eb;
3178 eb = path->nodes[0];
3179 slot = path->slots[0];
3181 btrfs_item_key(eb, &disk_key, slot - 1);
3182 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3184 if (slot < btrfs_header_nritems(eb) - 1) {
3185 btrfs_item_key(eb, &disk_key, slot + 1);
3186 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3189 btrfs_cpu_key_to_disk(&disk_key, new_key);
3190 btrfs_set_item_key(eb, &disk_key, slot);
3191 btrfs_mark_buffer_dirty(eb);
3193 fixup_low_keys(root, path, &disk_key, 1);
3197 * try to push data from one node into the next node left in the
3200 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3201 * error, and > 0 if there was no room in the left hand block.
3203 static int push_node_left(struct btrfs_trans_handle *trans,
3204 struct btrfs_root *root, struct extent_buffer *dst,
3205 struct extent_buffer *src, int empty)
3212 src_nritems = btrfs_header_nritems(src);
3213 dst_nritems = btrfs_header_nritems(dst);
3214 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3215 WARN_ON(btrfs_header_generation(src) != trans->transid);
3216 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3218 if (!empty && src_nritems <= 8)
3221 if (push_items <= 0)
3225 push_items = min(src_nritems, push_items);
3226 if (push_items < src_nritems) {
3227 /* leave at least 8 pointers in the node if
3228 * we aren't going to empty it
3230 if (src_nritems - push_items < 8) {
3231 if (push_items <= 8)
3237 push_items = min(src_nritems - 8, push_items);
3239 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3242 btrfs_abort_transaction(trans, root, ret);
3245 copy_extent_buffer(dst, src,
3246 btrfs_node_key_ptr_offset(dst_nritems),
3247 btrfs_node_key_ptr_offset(0),
3248 push_items * sizeof(struct btrfs_key_ptr));
3250 if (push_items < src_nritems) {
3252 * don't call tree_mod_log_eb_move here, key removal was already
3253 * fully logged by tree_mod_log_eb_copy above.
3255 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3256 btrfs_node_key_ptr_offset(push_items),
3257 (src_nritems - push_items) *
3258 sizeof(struct btrfs_key_ptr));
3260 btrfs_set_header_nritems(src, src_nritems - push_items);
3261 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3262 btrfs_mark_buffer_dirty(src);
3263 btrfs_mark_buffer_dirty(dst);
3269 * try to push data from one node into the next node right in the
3272 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3273 * error, and > 0 if there was no room in the right hand block.
3275 * this will only push up to 1/2 the contents of the left node over
3277 static int balance_node_right(struct btrfs_trans_handle *trans,
3278 struct btrfs_root *root,
3279 struct extent_buffer *dst,
3280 struct extent_buffer *src)
3288 WARN_ON(btrfs_header_generation(src) != trans->transid);
3289 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3291 src_nritems = btrfs_header_nritems(src);
3292 dst_nritems = btrfs_header_nritems(dst);
3293 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3294 if (push_items <= 0)
3297 if (src_nritems < 4)
3300 max_push = src_nritems / 2 + 1;
3301 /* don't try to empty the node */
3302 if (max_push >= src_nritems)
3305 if (max_push < push_items)
3306 push_items = max_push;
3308 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3309 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3310 btrfs_node_key_ptr_offset(0),
3312 sizeof(struct btrfs_key_ptr));
3314 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3315 src_nritems - push_items, push_items);
3317 btrfs_abort_transaction(trans, root, ret);
3320 copy_extent_buffer(dst, src,
3321 btrfs_node_key_ptr_offset(0),
3322 btrfs_node_key_ptr_offset(src_nritems - push_items),
3323 push_items * sizeof(struct btrfs_key_ptr));
3325 btrfs_set_header_nritems(src, src_nritems - push_items);
3326 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3328 btrfs_mark_buffer_dirty(src);
3329 btrfs_mark_buffer_dirty(dst);
3335 * helper function to insert a new root level in the tree.
3336 * A new node is allocated, and a single item is inserted to
3337 * point to the existing root
3339 * returns zero on success or < 0 on failure.
3341 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3342 struct btrfs_root *root,
3343 struct btrfs_path *path, int level)
3346 struct extent_buffer *lower;
3347 struct extent_buffer *c;
3348 struct extent_buffer *old;
3349 struct btrfs_disk_key lower_key;
3351 BUG_ON(path->nodes[level]);
3352 BUG_ON(path->nodes[level-1] != root->node);
3354 lower = path->nodes[level-1];
3356 btrfs_item_key(lower, &lower_key, 0);
3358 btrfs_node_key(lower, &lower_key, 0);
3360 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3361 root->root_key.objectid, &lower_key,
3362 level, root->node->start, 0);
3366 root_add_used(root, root->nodesize);
3368 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3369 btrfs_set_header_nritems(c, 1);
3370 btrfs_set_header_level(c, level);
3371 btrfs_set_header_bytenr(c, c->start);
3372 btrfs_set_header_generation(c, trans->transid);
3373 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3374 btrfs_set_header_owner(c, root->root_key.objectid);
3376 write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(),
3379 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3380 btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE);
3382 btrfs_set_node_key(c, &lower_key, 0);
3383 btrfs_set_node_blockptr(c, 0, lower->start);
3384 lower_gen = btrfs_header_generation(lower);
3385 WARN_ON(lower_gen != trans->transid);
3387 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3389 btrfs_mark_buffer_dirty(c);
3392 tree_mod_log_set_root_pointer(root, c, 0);
3393 rcu_assign_pointer(root->node, c);
3395 /* the super has an extra ref to root->node */
3396 free_extent_buffer(old);
3398 add_root_to_dirty_list(root);
3399 extent_buffer_get(c);
3400 path->nodes[level] = c;
3401 path->locks[level] = BTRFS_WRITE_LOCK;
3402 path->slots[level] = 0;
3407 * worker function to insert a single pointer in a node.
3408 * the node should have enough room for the pointer already
3410 * slot and level indicate where you want the key to go, and
3411 * blocknr is the block the key points to.
3413 static void insert_ptr(struct btrfs_trans_handle *trans,
3414 struct btrfs_root *root, struct btrfs_path *path,
3415 struct btrfs_disk_key *key, u64 bytenr,
3416 int slot, int level)
3418 struct extent_buffer *lower;
3422 BUG_ON(!path->nodes[level]);
3423 btrfs_assert_tree_locked(path->nodes[level]);
3424 lower = path->nodes[level];
3425 nritems = btrfs_header_nritems(lower);
3426 BUG_ON(slot > nritems);
3427 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3428 if (slot != nritems) {
3430 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3431 slot, nritems - slot);
3432 memmove_extent_buffer(lower,
3433 btrfs_node_key_ptr_offset(slot + 1),
3434 btrfs_node_key_ptr_offset(slot),
3435 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3438 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3439 MOD_LOG_KEY_ADD, GFP_NOFS);
3442 btrfs_set_node_key(lower, key, slot);
3443 btrfs_set_node_blockptr(lower, slot, bytenr);
3444 WARN_ON(trans->transid == 0);
3445 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3446 btrfs_set_header_nritems(lower, nritems + 1);
3447 btrfs_mark_buffer_dirty(lower);
3451 * split the node at the specified level in path in two.
3452 * The path is corrected to point to the appropriate node after the split
3454 * Before splitting this tries to make some room in the node by pushing
3455 * left and right, if either one works, it returns right away.
3457 * returns 0 on success and < 0 on failure
3459 static noinline int split_node(struct btrfs_trans_handle *trans,
3460 struct btrfs_root *root,
3461 struct btrfs_path *path, int level)
3463 struct extent_buffer *c;
3464 struct extent_buffer *split;
3465 struct btrfs_disk_key disk_key;
3470 c = path->nodes[level];
3471 WARN_ON(btrfs_header_generation(c) != trans->transid);
3472 if (c == root->node) {
3474 * trying to split the root, lets make a new one
3476 * tree mod log: We don't log_removal old root in
3477 * insert_new_root, because that root buffer will be kept as a
3478 * normal node. We are going to log removal of half of the
3479 * elements below with tree_mod_log_eb_copy. We're holding a
3480 * tree lock on the buffer, which is why we cannot race with
3481 * other tree_mod_log users.
3483 ret = insert_new_root(trans, root, path, level + 1);
3487 ret = push_nodes_for_insert(trans, root, path, level);
3488 c = path->nodes[level];
3489 if (!ret && btrfs_header_nritems(c) <
3490 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3496 c_nritems = btrfs_header_nritems(c);
3497 mid = (c_nritems + 1) / 2;
3498 btrfs_node_key(c, &disk_key, mid);
3500 split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3501 root->root_key.objectid,
3502 &disk_key, level, c->start, 0);
3504 return PTR_ERR(split);
3506 root_add_used(root, root->nodesize);
3508 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3509 btrfs_set_header_level(split, btrfs_header_level(c));
3510 btrfs_set_header_bytenr(split, split->start);
3511 btrfs_set_header_generation(split, trans->transid);
3512 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3513 btrfs_set_header_owner(split, root->root_key.objectid);
3514 write_extent_buffer(split, root->fs_info->fsid,
3515 btrfs_header_fsid(), BTRFS_FSID_SIZE);
3516 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3517 btrfs_header_chunk_tree_uuid(split),
3520 ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0,
3521 mid, c_nritems - mid);
3523 btrfs_abort_transaction(trans, root, ret);
3526 copy_extent_buffer(split, c,
3527 btrfs_node_key_ptr_offset(0),
3528 btrfs_node_key_ptr_offset(mid),
3529 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3530 btrfs_set_header_nritems(split, c_nritems - mid);
3531 btrfs_set_header_nritems(c, mid);
3534 btrfs_mark_buffer_dirty(c);
3535 btrfs_mark_buffer_dirty(split);
3537 insert_ptr(trans, root, path, &disk_key, split->start,
3538 path->slots[level + 1] + 1, level + 1);
3540 if (path->slots[level] >= mid) {
3541 path->slots[level] -= mid;
3542 btrfs_tree_unlock(c);
3543 free_extent_buffer(c);
3544 path->nodes[level] = split;
3545 path->slots[level + 1] += 1;
3547 btrfs_tree_unlock(split);
3548 free_extent_buffer(split);
3554 * how many bytes are required to store the items in a leaf. start
3555 * and nr indicate which items in the leaf to check. This totals up the
3556 * space used both by the item structs and the item data
3558 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3560 struct btrfs_item *start_item;
3561 struct btrfs_item *end_item;
3562 struct btrfs_map_token token;
3564 int nritems = btrfs_header_nritems(l);
3565 int end = min(nritems, start + nr) - 1;
3569 btrfs_init_map_token(&token);
3570 start_item = btrfs_item_nr(start);
3571 end_item = btrfs_item_nr(end);
3572 data_len = btrfs_token_item_offset(l, start_item, &token) +
3573 btrfs_token_item_size(l, start_item, &token);
3574 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3575 data_len += sizeof(struct btrfs_item) * nr;
3576 WARN_ON(data_len < 0);
3581 * The space between the end of the leaf items and
3582 * the start of the leaf data. IOW, how much room
3583 * the leaf has left for both items and data
3585 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3586 struct extent_buffer *leaf)
3588 int nritems = btrfs_header_nritems(leaf);
3590 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3592 btrfs_crit(root->fs_info,
3593 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3594 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3595 leaf_space_used(leaf, 0, nritems), nritems);
3601 * min slot controls the lowest index we're willing to push to the
3602 * right. We'll push up to and including min_slot, but no lower
3604 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3605 struct btrfs_root *root,
3606 struct btrfs_path *path,
3607 int data_size, int empty,
3608 struct extent_buffer *right,
3609 int free_space, u32 left_nritems,
3612 struct extent_buffer *left = path->nodes[0];
3613 struct extent_buffer *upper = path->nodes[1];
3614 struct btrfs_map_token token;
3615 struct btrfs_disk_key disk_key;
3620 struct btrfs_item *item;
3626 btrfs_init_map_token(&token);
3631 nr = max_t(u32, 1, min_slot);
3633 if (path->slots[0] >= left_nritems)
3634 push_space += data_size;
3636 slot = path->slots[1];
3637 i = left_nritems - 1;
3639 item = btrfs_item_nr(i);
3641 if (!empty && push_items > 0) {
3642 if (path->slots[0] > i)
3644 if (path->slots[0] == i) {
3645 int space = btrfs_leaf_free_space(root, left);
3646 if (space + push_space * 2 > free_space)
3651 if (path->slots[0] == i)
3652 push_space += data_size;
3654 this_item_size = btrfs_item_size(left, item);
3655 if (this_item_size + sizeof(*item) + push_space > free_space)
3659 push_space += this_item_size + sizeof(*item);
3665 if (push_items == 0)
3668 WARN_ON(!empty && push_items == left_nritems);
3670 /* push left to right */
3671 right_nritems = btrfs_header_nritems(right);
3673 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3674 push_space -= leaf_data_end(root, left);
3676 /* make room in the right data area */
3677 data_end = leaf_data_end(root, right);
3678 memmove_extent_buffer(right,
3679 btrfs_leaf_data(right) + data_end - push_space,
3680 btrfs_leaf_data(right) + data_end,
3681 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3683 /* copy from the left data area */
3684 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3685 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3686 btrfs_leaf_data(left) + leaf_data_end(root, left),
3689 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3690 btrfs_item_nr_offset(0),
3691 right_nritems * sizeof(struct btrfs_item));
3693 /* copy the items from left to right */
3694 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3695 btrfs_item_nr_offset(left_nritems - push_items),
3696 push_items * sizeof(struct btrfs_item));
3698 /* update the item pointers */
3699 right_nritems += push_items;
3700 btrfs_set_header_nritems(right, right_nritems);
3701 push_space = BTRFS_LEAF_DATA_SIZE(root);
3702 for (i = 0; i < right_nritems; i++) {
3703 item = btrfs_item_nr(i);
3704 push_space -= btrfs_token_item_size(right, item, &token);
3705 btrfs_set_token_item_offset(right, item, push_space, &token);
3708 left_nritems -= push_items;
3709 btrfs_set_header_nritems(left, left_nritems);
3712 btrfs_mark_buffer_dirty(left);
3714 clean_tree_block(trans, root, left);
3716 btrfs_mark_buffer_dirty(right);
3718 btrfs_item_key(right, &disk_key, 0);
3719 btrfs_set_node_key(upper, &disk_key, slot + 1);
3720 btrfs_mark_buffer_dirty(upper);
3722 /* then fixup the leaf pointer in the path */
3723 if (path->slots[0] >= left_nritems) {
3724 path->slots[0] -= left_nritems;
3725 if (btrfs_header_nritems(path->nodes[0]) == 0)
3726 clean_tree_block(trans, root, path->nodes[0]);
3727 btrfs_tree_unlock(path->nodes[0]);
3728 free_extent_buffer(path->nodes[0]);
3729 path->nodes[0] = right;
3730 path->slots[1] += 1;
3732 btrfs_tree_unlock(right);
3733 free_extent_buffer(right);
3738 btrfs_tree_unlock(right);
3739 free_extent_buffer(right);
3744 * push some data in the path leaf to the right, trying to free up at
3745 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3747 * returns 1 if the push failed because the other node didn't have enough
3748 * room, 0 if everything worked out and < 0 if there were major errors.
3750 * this will push starting from min_slot to the end of the leaf. It won't
3751 * push any slot lower than min_slot
3753 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3754 *root, struct btrfs_path *path,
3755 int min_data_size, int data_size,
3756 int empty, u32 min_slot)
3758 struct extent_buffer *left = path->nodes[0];
3759 struct extent_buffer *right;
3760 struct extent_buffer *upper;
3766 if (!path->nodes[1])
3769 slot = path->slots[1];
3770 upper = path->nodes[1];
3771 if (slot >= btrfs_header_nritems(upper) - 1)
3774 btrfs_assert_tree_locked(path->nodes[1]);
3776 right = read_node_slot(root, upper, slot + 1);
3780 btrfs_tree_lock(right);
3781 btrfs_set_lock_blocking(right);
3783 free_space = btrfs_leaf_free_space(root, right);
3784 if (free_space < data_size)
3787 /* cow and double check */
3788 ret = btrfs_cow_block(trans, root, right, upper,
3793 free_space = btrfs_leaf_free_space(root, right);
3794 if (free_space < data_size)
3797 left_nritems = btrfs_header_nritems(left);
3798 if (left_nritems == 0)
3801 if (path->slots[0] == left_nritems && !empty) {
3802 /* Key greater than all keys in the leaf, right neighbor has
3803 * enough room for it and we're not emptying our leaf to delete
3804 * it, therefore use right neighbor to insert the new item and
3805 * no need to touch/dirty our left leaft. */
3806 btrfs_tree_unlock(left);
3807 free_extent_buffer(left);
3808 path->nodes[0] = right;
3814 return __push_leaf_right(trans, root, path, min_data_size, empty,
3815 right, free_space, left_nritems, min_slot);
3817 btrfs_tree_unlock(right);
3818 free_extent_buffer(right);
3823 * push some data in the path leaf to the left, trying to free up at
3824 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3826 * max_slot can put a limit on how far into the leaf we'll push items. The
3827 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3830 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3831 struct btrfs_root *root,
3832 struct btrfs_path *path, int data_size,
3833 int empty, struct extent_buffer *left,
3834 int free_space, u32 right_nritems,
3837 struct btrfs_disk_key disk_key;
3838 struct extent_buffer *right = path->nodes[0];
3842 struct btrfs_item *item;
3843 u32 old_left_nritems;
3847 u32 old_left_item_size;
3848 struct btrfs_map_token token;
3850 btrfs_init_map_token(&token);
3853 nr = min(right_nritems, max_slot);
3855 nr = min(right_nritems - 1, max_slot);
3857 for (i = 0; i < nr; i++) {
3858 item = btrfs_item_nr(i);
3860 if (!empty && push_items > 0) {
3861 if (path->slots[0] < i)
3863 if (path->slots[0] == i) {
3864 int space = btrfs_leaf_free_space(root, right);
3865 if (space + push_space * 2 > free_space)
3870 if (path->slots[0] == i)
3871 push_space += data_size;
3873 this_item_size = btrfs_item_size(right, item);
3874 if (this_item_size + sizeof(*item) + push_space > free_space)
3878 push_space += this_item_size + sizeof(*item);
3881 if (push_items == 0) {
3885 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3887 /* push data from right to left */
3888 copy_extent_buffer(left, right,
3889 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3890 btrfs_item_nr_offset(0),
3891 push_items * sizeof(struct btrfs_item));
3893 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3894 btrfs_item_offset_nr(right, push_items - 1);
3896 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3897 leaf_data_end(root, left) - push_space,
3898 btrfs_leaf_data(right) +
3899 btrfs_item_offset_nr(right, push_items - 1),
3901 old_left_nritems = btrfs_header_nritems(left);
3902 BUG_ON(old_left_nritems <= 0);
3904 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3905 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3908 item = btrfs_item_nr(i);
3910 ioff = btrfs_token_item_offset(left, item, &token);
3911 btrfs_set_token_item_offset(left, item,
3912 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3915 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3917 /* fixup right node */
3918 if (push_items > right_nritems)
3919 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3922 if (push_items < right_nritems) {
3923 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3924 leaf_data_end(root, right);
3925 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3926 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3927 btrfs_leaf_data(right) +
3928 leaf_data_end(root, right), push_space);
3930 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3931 btrfs_item_nr_offset(push_items),
3932 (btrfs_header_nritems(right) - push_items) *
3933 sizeof(struct btrfs_item));
3935 right_nritems -= push_items;
3936 btrfs_set_header_nritems(right, right_nritems);
3937 push_space = BTRFS_LEAF_DATA_SIZE(root);
3938 for (i = 0; i < right_nritems; i++) {
3939 item = btrfs_item_nr(i);
3941 push_space = push_space - btrfs_token_item_size(right,
3943 btrfs_set_token_item_offset(right, item, push_space, &token);
3946 btrfs_mark_buffer_dirty(left);
3948 btrfs_mark_buffer_dirty(right);
3950 clean_tree_block(trans, root, right);
3952 btrfs_item_key(right, &disk_key, 0);
3953 fixup_low_keys(root, path, &disk_key, 1);
3955 /* then fixup the leaf pointer in the path */
3956 if (path->slots[0] < push_items) {
3957 path->slots[0] += old_left_nritems;
3958 btrfs_tree_unlock(path->nodes[0]);
3959 free_extent_buffer(path->nodes[0]);
3960 path->nodes[0] = left;
3961 path->slots[1] -= 1;
3963 btrfs_tree_unlock(left);
3964 free_extent_buffer(left);
3965 path->slots[0] -= push_items;
3967 BUG_ON(path->slots[0] < 0);
3970 btrfs_tree_unlock(left);
3971 free_extent_buffer(left);
3976 * push some data in the path leaf to the left, trying to free up at
3977 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3979 * max_slot can put a limit on how far into the leaf we'll push items. The
3980 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3983 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3984 *root, struct btrfs_path *path, int min_data_size,
3985 int data_size, int empty, u32 max_slot)
3987 struct extent_buffer *right = path->nodes[0];
3988 struct extent_buffer *left;
3994 slot = path->slots[1];
3997 if (!path->nodes[1])
4000 right_nritems = btrfs_header_nritems(right);
4001 if (right_nritems == 0)
4004 btrfs_assert_tree_locked(path->nodes[1]);
4006 left = read_node_slot(root, path->nodes[1], slot - 1);
4010 btrfs_tree_lock(left);
4011 btrfs_set_lock_blocking(left);
4013 free_space = btrfs_leaf_free_space(root, left);
4014 if (free_space < data_size) {
4019 /* cow and double check */
4020 ret = btrfs_cow_block(trans, root, left,
4021 path->nodes[1], slot - 1, &left);
4023 /* we hit -ENOSPC, but it isn't fatal here */
4029 free_space = btrfs_leaf_free_space(root, left);
4030 if (free_space < data_size) {
4035 return __push_leaf_left(trans, root, path, min_data_size,
4036 empty, left, free_space, right_nritems,
4039 btrfs_tree_unlock(left);
4040 free_extent_buffer(left);
4045 * split the path's leaf in two, making sure there is at least data_size
4046 * available for the resulting leaf level of the path.
4048 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4049 struct btrfs_root *root,
4050 struct btrfs_path *path,
4051 struct extent_buffer *l,
4052 struct extent_buffer *right,
4053 int slot, int mid, int nritems)
4058 struct btrfs_disk_key disk_key;
4059 struct btrfs_map_token token;
4061 btrfs_init_map_token(&token);
4063 nritems = nritems - mid;
4064 btrfs_set_header_nritems(right, nritems);
4065 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
4067 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4068 btrfs_item_nr_offset(mid),
4069 nritems * sizeof(struct btrfs_item));
4071 copy_extent_buffer(right, l,
4072 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
4073 data_copy_size, btrfs_leaf_data(l) +
4074 leaf_data_end(root, l), data_copy_size);
4076 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
4077 btrfs_item_end_nr(l, mid);
4079 for (i = 0; i < nritems; i++) {
4080 struct btrfs_item *item = btrfs_item_nr(i);
4083 ioff = btrfs_token_item_offset(right, item, &token);
4084 btrfs_set_token_item_offset(right, item,
4085 ioff + rt_data_off, &token);
4088 btrfs_set_header_nritems(l, mid);
4089 btrfs_item_key(right, &disk_key, 0);
4090 insert_ptr(trans, root, path, &disk_key, right->start,
4091 path->slots[1] + 1, 1);
4093 btrfs_mark_buffer_dirty(right);
4094 btrfs_mark_buffer_dirty(l);
4095 BUG_ON(path->slots[0] != slot);
4098 btrfs_tree_unlock(path->nodes[0]);
4099 free_extent_buffer(path->nodes[0]);
4100 path->nodes[0] = right;
4101 path->slots[0] -= mid;
4102 path->slots[1] += 1;
4104 btrfs_tree_unlock(right);
4105 free_extent_buffer(right);
4108 BUG_ON(path->slots[0] < 0);
4112 * double splits happen when we need to insert a big item in the middle
4113 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4114 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4117 * We avoid this by trying to push the items on either side of our target
4118 * into the adjacent leaves. If all goes well we can avoid the double split
4121 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4122 struct btrfs_root *root,
4123 struct btrfs_path *path,
4130 int space_needed = data_size;
4132 slot = path->slots[0];
4133 if (slot < btrfs_header_nritems(path->nodes[0]))
4134 space_needed -= btrfs_leaf_free_space(root, path->nodes[0]);
4137 * try to push all the items after our slot into the
4140 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4147 nritems = btrfs_header_nritems(path->nodes[0]);
4149 * our goal is to get our slot at the start or end of a leaf. If
4150 * we've done so we're done
4152 if (path->slots[0] == 0 || path->slots[0] == nritems)
4155 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4158 /* try to push all the items before our slot into the next leaf */
4159 slot = path->slots[0];
4160 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4173 * split the path's leaf in two, making sure there is at least data_size
4174 * available for the resulting leaf level of the path.
4176 * returns 0 if all went well and < 0 on failure.
4178 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4179 struct btrfs_root *root,
4180 struct btrfs_key *ins_key,
4181 struct btrfs_path *path, int data_size,
4184 struct btrfs_disk_key disk_key;
4185 struct extent_buffer *l;
4189 struct extent_buffer *right;
4193 int num_doubles = 0;
4194 int tried_avoid_double = 0;
4197 slot = path->slots[0];
4198 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4199 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
4202 /* first try to make some room by pushing left and right */
4203 if (data_size && path->nodes[1]) {
4204 int space_needed = data_size;
4206 if (slot < btrfs_header_nritems(l))
4207 space_needed -= btrfs_leaf_free_space(root, l);
4209 wret = push_leaf_right(trans, root, path, space_needed,
4210 space_needed, 0, 0);
4214 wret = push_leaf_left(trans, root, path, space_needed,
4215 space_needed, 0, (u32)-1);
4221 /* did the pushes work? */
4222 if (btrfs_leaf_free_space(root, l) >= data_size)
4226 if (!path->nodes[1]) {
4227 ret = insert_new_root(trans, root, path, 1);
4234 slot = path->slots[0];
4235 nritems = btrfs_header_nritems(l);
4236 mid = (nritems + 1) / 2;
4240 leaf_space_used(l, mid, nritems - mid) + data_size >
4241 BTRFS_LEAF_DATA_SIZE(root)) {
4242 if (slot >= nritems) {
4246 if (mid != nritems &&
4247 leaf_space_used(l, mid, nritems - mid) +
4248 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4249 if (data_size && !tried_avoid_double)
4250 goto push_for_double;
4256 if (leaf_space_used(l, 0, mid) + data_size >
4257 BTRFS_LEAF_DATA_SIZE(root)) {
4258 if (!extend && data_size && slot == 0) {
4260 } else if ((extend || !data_size) && slot == 0) {
4264 if (mid != nritems &&
4265 leaf_space_used(l, mid, nritems - mid) +
4266 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4267 if (data_size && !tried_avoid_double)
4268 goto push_for_double;
4276 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4278 btrfs_item_key(l, &disk_key, mid);
4280 right = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
4281 root->root_key.objectid,
4282 &disk_key, 0, l->start, 0);
4284 return PTR_ERR(right);
4286 root_add_used(root, root->nodesize);
4288 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4289 btrfs_set_header_bytenr(right, right->start);
4290 btrfs_set_header_generation(right, trans->transid);
4291 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4292 btrfs_set_header_owner(right, root->root_key.objectid);
4293 btrfs_set_header_level(right, 0);
4294 write_extent_buffer(right, root->fs_info->fsid,
4295 btrfs_header_fsid(), BTRFS_FSID_SIZE);
4297 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
4298 btrfs_header_chunk_tree_uuid(right),
4303 btrfs_set_header_nritems(right, 0);
4304 insert_ptr(trans, root, path, &disk_key, right->start,
4305 path->slots[1] + 1, 1);
4306 btrfs_tree_unlock(path->nodes[0]);
4307 free_extent_buffer(path->nodes[0]);
4308 path->nodes[0] = right;
4310 path->slots[1] += 1;
4312 btrfs_set_header_nritems(right, 0);
4313 insert_ptr(trans, root, path, &disk_key, right->start,
4315 btrfs_tree_unlock(path->nodes[0]);
4316 free_extent_buffer(path->nodes[0]);
4317 path->nodes[0] = right;
4319 if (path->slots[1] == 0)
4320 fixup_low_keys(root, path, &disk_key, 1);
4322 btrfs_mark_buffer_dirty(right);
4326 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4329 BUG_ON(num_doubles != 0);
4337 push_for_double_split(trans, root, path, data_size);
4338 tried_avoid_double = 1;
4339 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4344 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4345 struct btrfs_root *root,
4346 struct btrfs_path *path, int ins_len)
4348 struct btrfs_key key;
4349 struct extent_buffer *leaf;
4350 struct btrfs_file_extent_item *fi;
4355 leaf = path->nodes[0];
4356 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4358 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4359 key.type != BTRFS_EXTENT_CSUM_KEY);
4361 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4364 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4365 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4366 fi = btrfs_item_ptr(leaf, path->slots[0],
4367 struct btrfs_file_extent_item);
4368 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4370 btrfs_release_path(path);
4372 path->keep_locks = 1;
4373 path->search_for_split = 1;
4374 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4375 path->search_for_split = 0;
4380 leaf = path->nodes[0];
4381 /* if our item isn't there or got smaller, return now */
4382 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4385 /* the leaf has changed, it now has room. return now */
4386 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4389 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4390 fi = btrfs_item_ptr(leaf, path->slots[0],
4391 struct btrfs_file_extent_item);
4392 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4396 btrfs_set_path_blocking(path);
4397 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4401 path->keep_locks = 0;
4402 btrfs_unlock_up_safe(path, 1);
4405 path->keep_locks = 0;
4409 static noinline int split_item(struct btrfs_trans_handle *trans,
4410 struct btrfs_root *root,
4411 struct btrfs_path *path,
4412 struct btrfs_key *new_key,
4413 unsigned long split_offset)
4415 struct extent_buffer *leaf;
4416 struct btrfs_item *item;
4417 struct btrfs_item *new_item;
4423 struct btrfs_disk_key disk_key;
4425 leaf = path->nodes[0];
4426 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4428 btrfs_set_path_blocking(path);
4430 item = btrfs_item_nr(path->slots[0]);
4431 orig_offset = btrfs_item_offset(leaf, item);
4432 item_size = btrfs_item_size(leaf, item);
4434 buf = kmalloc(item_size, GFP_NOFS);
4438 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4439 path->slots[0]), item_size);
4441 slot = path->slots[0] + 1;
4442 nritems = btrfs_header_nritems(leaf);
4443 if (slot != nritems) {
4444 /* shift the items */
4445 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4446 btrfs_item_nr_offset(slot),
4447 (nritems - slot) * sizeof(struct btrfs_item));
4450 btrfs_cpu_key_to_disk(&disk_key, new_key);
4451 btrfs_set_item_key(leaf, &disk_key, slot);
4453 new_item = btrfs_item_nr(slot);
4455 btrfs_set_item_offset(leaf, new_item, orig_offset);
4456 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4458 btrfs_set_item_offset(leaf, item,
4459 orig_offset + item_size - split_offset);
4460 btrfs_set_item_size(leaf, item, split_offset);
4462 btrfs_set_header_nritems(leaf, nritems + 1);
4464 /* write the data for the start of the original item */
4465 write_extent_buffer(leaf, buf,
4466 btrfs_item_ptr_offset(leaf, path->slots[0]),
4469 /* write the data for the new item */
4470 write_extent_buffer(leaf, buf + split_offset,
4471 btrfs_item_ptr_offset(leaf, slot),
4472 item_size - split_offset);
4473 btrfs_mark_buffer_dirty(leaf);
4475 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4481 * This function splits a single item into two items,
4482 * giving 'new_key' to the new item and splitting the
4483 * old one at split_offset (from the start of the item).
4485 * The path may be released by this operation. After
4486 * the split, the path is pointing to the old item. The
4487 * new item is going to be in the same node as the old one.
4489 * Note, the item being split must be smaller enough to live alone on
4490 * a tree block with room for one extra struct btrfs_item
4492 * This allows us to split the item in place, keeping a lock on the
4493 * leaf the entire time.
4495 int btrfs_split_item(struct btrfs_trans_handle *trans,
4496 struct btrfs_root *root,
4497 struct btrfs_path *path,
4498 struct btrfs_key *new_key,
4499 unsigned long split_offset)
4502 ret = setup_leaf_for_split(trans, root, path,
4503 sizeof(struct btrfs_item));
4507 ret = split_item(trans, root, path, new_key, split_offset);
4512 * This function duplicate a item, giving 'new_key' to the new item.
4513 * It guarantees both items live in the same tree leaf and the new item
4514 * is contiguous with the original item.
4516 * This allows us to split file extent in place, keeping a lock on the
4517 * leaf the entire time.
4519 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4520 struct btrfs_root *root,
4521 struct btrfs_path *path,
4522 struct btrfs_key *new_key)
4524 struct extent_buffer *leaf;
4528 leaf = path->nodes[0];
4529 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4530 ret = setup_leaf_for_split(trans, root, path,
4531 item_size + sizeof(struct btrfs_item));
4536 setup_items_for_insert(root, path, new_key, &item_size,
4537 item_size, item_size +
4538 sizeof(struct btrfs_item), 1);
4539 leaf = path->nodes[0];
4540 memcpy_extent_buffer(leaf,
4541 btrfs_item_ptr_offset(leaf, path->slots[0]),
4542 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4548 * make the item pointed to by the path smaller. new_size indicates
4549 * how small to make it, and from_end tells us if we just chop bytes
4550 * off the end of the item or if we shift the item to chop bytes off
4553 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4554 u32 new_size, int from_end)
4557 struct extent_buffer *leaf;
4558 struct btrfs_item *item;
4560 unsigned int data_end;
4561 unsigned int old_data_start;
4562 unsigned int old_size;
4563 unsigned int size_diff;
4565 struct btrfs_map_token token;
4567 btrfs_init_map_token(&token);
4569 leaf = path->nodes[0];
4570 slot = path->slots[0];
4572 old_size = btrfs_item_size_nr(leaf, slot);
4573 if (old_size == new_size)
4576 nritems = btrfs_header_nritems(leaf);
4577 data_end = leaf_data_end(root, leaf);
4579 old_data_start = btrfs_item_offset_nr(leaf, slot);
4581 size_diff = old_size - new_size;
4584 BUG_ON(slot >= nritems);
4587 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4589 /* first correct the data pointers */
4590 for (i = slot; i < nritems; i++) {
4592 item = btrfs_item_nr(i);
4594 ioff = btrfs_token_item_offset(leaf, item, &token);
4595 btrfs_set_token_item_offset(leaf, item,
4596 ioff + size_diff, &token);
4599 /* shift the data */
4601 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4602 data_end + size_diff, btrfs_leaf_data(leaf) +
4603 data_end, old_data_start + new_size - data_end);
4605 struct btrfs_disk_key disk_key;
4608 btrfs_item_key(leaf, &disk_key, slot);
4610 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4612 struct btrfs_file_extent_item *fi;
4614 fi = btrfs_item_ptr(leaf, slot,
4615 struct btrfs_file_extent_item);
4616 fi = (struct btrfs_file_extent_item *)(
4617 (unsigned long)fi - size_diff);
4619 if (btrfs_file_extent_type(leaf, fi) ==
4620 BTRFS_FILE_EXTENT_INLINE) {
4621 ptr = btrfs_item_ptr_offset(leaf, slot);
4622 memmove_extent_buffer(leaf, ptr,
4624 offsetof(struct btrfs_file_extent_item,
4629 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4630 data_end + size_diff, btrfs_leaf_data(leaf) +
4631 data_end, old_data_start - data_end);
4633 offset = btrfs_disk_key_offset(&disk_key);
4634 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4635 btrfs_set_item_key(leaf, &disk_key, slot);
4637 fixup_low_keys(root, path, &disk_key, 1);
4640 item = btrfs_item_nr(slot);
4641 btrfs_set_item_size(leaf, item, new_size);
4642 btrfs_mark_buffer_dirty(leaf);
4644 if (btrfs_leaf_free_space(root, leaf) < 0) {
4645 btrfs_print_leaf(root, leaf);
4651 * make the item pointed to by the path bigger, data_size is the added size.
4653 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4657 struct extent_buffer *leaf;
4658 struct btrfs_item *item;
4660 unsigned int data_end;
4661 unsigned int old_data;
4662 unsigned int old_size;
4664 struct btrfs_map_token token;
4666 btrfs_init_map_token(&token);
4668 leaf = path->nodes[0];
4670 nritems = btrfs_header_nritems(leaf);
4671 data_end = leaf_data_end(root, leaf);
4673 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4674 btrfs_print_leaf(root, leaf);
4677 slot = path->slots[0];
4678 old_data = btrfs_item_end_nr(leaf, slot);
4681 if (slot >= nritems) {
4682 btrfs_print_leaf(root, leaf);
4683 btrfs_crit(root->fs_info, "slot %d too large, nritems %d",
4689 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4691 /* first correct the data pointers */
4692 for (i = slot; i < nritems; i++) {
4694 item = btrfs_item_nr(i);
4696 ioff = btrfs_token_item_offset(leaf, item, &token);
4697 btrfs_set_token_item_offset(leaf, item,
4698 ioff - data_size, &token);
4701 /* shift the data */
4702 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4703 data_end - data_size, btrfs_leaf_data(leaf) +
4704 data_end, old_data - data_end);
4706 data_end = old_data;
4707 old_size = btrfs_item_size_nr(leaf, slot);
4708 item = btrfs_item_nr(slot);
4709 btrfs_set_item_size(leaf, item, old_size + data_size);
4710 btrfs_mark_buffer_dirty(leaf);
4712 if (btrfs_leaf_free_space(root, leaf) < 0) {
4713 btrfs_print_leaf(root, leaf);
4719 * this is a helper for btrfs_insert_empty_items, the main goal here is
4720 * to save stack depth by doing the bulk of the work in a function
4721 * that doesn't call btrfs_search_slot
4723 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4724 struct btrfs_key *cpu_key, u32 *data_size,
4725 u32 total_data, u32 total_size, int nr)
4727 struct btrfs_item *item;
4730 unsigned int data_end;
4731 struct btrfs_disk_key disk_key;
4732 struct extent_buffer *leaf;
4734 struct btrfs_map_token token;
4736 if (path->slots[0] == 0) {
4737 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4738 fixup_low_keys(root, path, &disk_key, 1);
4740 btrfs_unlock_up_safe(path, 1);
4742 btrfs_init_map_token(&token);
4744 leaf = path->nodes[0];
4745 slot = path->slots[0];
4747 nritems = btrfs_header_nritems(leaf);
4748 data_end = leaf_data_end(root, leaf);
4750 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4751 btrfs_print_leaf(root, leaf);
4752 btrfs_crit(root->fs_info, "not enough freespace need %u have %d",
4753 total_size, btrfs_leaf_free_space(root, leaf));
4757 if (slot != nritems) {
4758 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4760 if (old_data < data_end) {
4761 btrfs_print_leaf(root, leaf);
4762 btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d",
4763 slot, old_data, data_end);
4767 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4769 /* first correct the data pointers */
4770 for (i = slot; i < nritems; i++) {
4773 item = btrfs_item_nr( i);
4774 ioff = btrfs_token_item_offset(leaf, item, &token);
4775 btrfs_set_token_item_offset(leaf, item,
4776 ioff - total_data, &token);
4778 /* shift the items */
4779 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4780 btrfs_item_nr_offset(slot),
4781 (nritems - slot) * sizeof(struct btrfs_item));
4783 /* shift the data */
4784 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4785 data_end - total_data, btrfs_leaf_data(leaf) +
4786 data_end, old_data - data_end);
4787 data_end = old_data;
4790 /* setup the item for the new data */
4791 for (i = 0; i < nr; i++) {
4792 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4793 btrfs_set_item_key(leaf, &disk_key, slot + i);
4794 item = btrfs_item_nr(slot + i);
4795 btrfs_set_token_item_offset(leaf, item,
4796 data_end - data_size[i], &token);
4797 data_end -= data_size[i];
4798 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4801 btrfs_set_header_nritems(leaf, nritems + nr);
4802 btrfs_mark_buffer_dirty(leaf);
4804 if (btrfs_leaf_free_space(root, leaf) < 0) {
4805 btrfs_print_leaf(root, leaf);
4811 * Given a key and some data, insert items into the tree.
4812 * This does all the path init required, making room in the tree if needed.
4814 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4815 struct btrfs_root *root,
4816 struct btrfs_path *path,
4817 struct btrfs_key *cpu_key, u32 *data_size,
4826 for (i = 0; i < nr; i++)
4827 total_data += data_size[i];
4829 total_size = total_data + (nr * sizeof(struct btrfs_item));
4830 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4836 slot = path->slots[0];
4839 setup_items_for_insert(root, path, cpu_key, data_size,
4840 total_data, total_size, nr);
4845 * Given a key and some data, insert an item into the tree.
4846 * This does all the path init required, making room in the tree if needed.
4848 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4849 *root, struct btrfs_key *cpu_key, void *data, u32
4853 struct btrfs_path *path;
4854 struct extent_buffer *leaf;
4857 path = btrfs_alloc_path();
4860 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4862 leaf = path->nodes[0];
4863 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4864 write_extent_buffer(leaf, data, ptr, data_size);
4865 btrfs_mark_buffer_dirty(leaf);
4867 btrfs_free_path(path);
4872 * delete the pointer from a given node.
4874 * the tree should have been previously balanced so the deletion does not
4877 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4878 int level, int slot)
4880 struct extent_buffer *parent = path->nodes[level];
4884 nritems = btrfs_header_nritems(parent);
4885 if (slot != nritems - 1) {
4887 tree_mod_log_eb_move(root->fs_info, parent, slot,
4888 slot + 1, nritems - slot - 1);
4889 memmove_extent_buffer(parent,
4890 btrfs_node_key_ptr_offset(slot),
4891 btrfs_node_key_ptr_offset(slot + 1),
4892 sizeof(struct btrfs_key_ptr) *
4893 (nritems - slot - 1));
4895 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4896 MOD_LOG_KEY_REMOVE, GFP_NOFS);
4901 btrfs_set_header_nritems(parent, nritems);
4902 if (nritems == 0 && parent == root->node) {
4903 BUG_ON(btrfs_header_level(root->node) != 1);
4904 /* just turn the root into a leaf and break */
4905 btrfs_set_header_level(root->node, 0);
4906 } else if (slot == 0) {
4907 struct btrfs_disk_key disk_key;
4909 btrfs_node_key(parent, &disk_key, 0);
4910 fixup_low_keys(root, path, &disk_key, level + 1);
4912 btrfs_mark_buffer_dirty(parent);
4916 * a helper function to delete the leaf pointed to by path->slots[1] and
4919 * This deletes the pointer in path->nodes[1] and frees the leaf
4920 * block extent. zero is returned if it all worked out, < 0 otherwise.
4922 * The path must have already been setup for deleting the leaf, including
4923 * all the proper balancing. path->nodes[1] must be locked.
4925 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4926 struct btrfs_root *root,
4927 struct btrfs_path *path,
4928 struct extent_buffer *leaf)
4930 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4931 del_ptr(root, path, 1, path->slots[1]);
4934 * btrfs_free_extent is expensive, we want to make sure we
4935 * aren't holding any locks when we call it
4937 btrfs_unlock_up_safe(path, 0);
4939 root_sub_used(root, leaf->len);
4941 extent_buffer_get(leaf);
4942 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4943 free_extent_buffer_stale(leaf);
4946 * delete the item at the leaf level in path. If that empties
4947 * the leaf, remove it from the tree
4949 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4950 struct btrfs_path *path, int slot, int nr)
4952 struct extent_buffer *leaf;
4953 struct btrfs_item *item;
4960 struct btrfs_map_token token;
4962 btrfs_init_map_token(&token);
4964 leaf = path->nodes[0];
4965 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4967 for (i = 0; i < nr; i++)
4968 dsize += btrfs_item_size_nr(leaf, slot + i);
4970 nritems = btrfs_header_nritems(leaf);
4972 if (slot + nr != nritems) {
4973 int data_end = leaf_data_end(root, leaf);
4975 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4977 btrfs_leaf_data(leaf) + data_end,
4978 last_off - data_end);
4980 for (i = slot + nr; i < nritems; i++) {
4983 item = btrfs_item_nr(i);
4984 ioff = btrfs_token_item_offset(leaf, item, &token);
4985 btrfs_set_token_item_offset(leaf, item,
4986 ioff + dsize, &token);
4989 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4990 btrfs_item_nr_offset(slot + nr),
4991 sizeof(struct btrfs_item) *
4992 (nritems - slot - nr));
4994 btrfs_set_header_nritems(leaf, nritems - nr);
4997 /* delete the leaf if we've emptied it */
4999 if (leaf == root->node) {
5000 btrfs_set_header_level(leaf, 0);
5002 btrfs_set_path_blocking(path);
5003 clean_tree_block(trans, root, leaf);
5004 btrfs_del_leaf(trans, root, path, leaf);
5007 int used = leaf_space_used(leaf, 0, nritems);
5009 struct btrfs_disk_key disk_key;
5011 btrfs_item_key(leaf, &disk_key, 0);
5012 fixup_low_keys(root, path, &disk_key, 1);
5015 /* delete the leaf if it is mostly empty */
5016 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
5017 /* push_leaf_left fixes the path.
5018 * make sure the path still points to our leaf
5019 * for possible call to del_ptr below
5021 slot = path->slots[1];
5022 extent_buffer_get(leaf);
5024 btrfs_set_path_blocking(path);
5025 wret = push_leaf_left(trans, root, path, 1, 1,
5027 if (wret < 0 && wret != -ENOSPC)
5030 if (path->nodes[0] == leaf &&
5031 btrfs_header_nritems(leaf)) {
5032 wret = push_leaf_right(trans, root, path, 1,
5034 if (wret < 0 && wret != -ENOSPC)
5038 if (btrfs_header_nritems(leaf) == 0) {
5039 path->slots[1] = slot;
5040 btrfs_del_leaf(trans, root, path, leaf);
5041 free_extent_buffer(leaf);
5044 /* if we're still in the path, make sure
5045 * we're dirty. Otherwise, one of the
5046 * push_leaf functions must have already
5047 * dirtied this buffer
5049 if (path->nodes[0] == leaf)
5050 btrfs_mark_buffer_dirty(leaf);
5051 free_extent_buffer(leaf);
5054 btrfs_mark_buffer_dirty(leaf);
5061 * search the tree again to find a leaf with lesser keys
5062 * returns 0 if it found something or 1 if there are no lesser leaves.
5063 * returns < 0 on io errors.
5065 * This may release the path, and so you may lose any locks held at the
5068 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5070 struct btrfs_key key;
5071 struct btrfs_disk_key found_key;
5074 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5076 if (key.offset > 0) {
5078 } else if (key.type > 0) {
5080 key.offset = (u64)-1;
5081 } else if (key.objectid > 0) {
5084 key.offset = (u64)-1;
5089 btrfs_release_path(path);
5090 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5093 btrfs_item_key(path->nodes[0], &found_key, 0);
5094 ret = comp_keys(&found_key, &key);
5096 * We might have had an item with the previous key in the tree right
5097 * before we released our path. And after we released our path, that
5098 * item might have been pushed to the first slot (0) of the leaf we
5099 * were holding due to a tree balance. Alternatively, an item with the
5100 * previous key can exist as the only element of a leaf (big fat item).
5101 * Therefore account for these 2 cases, so that our callers (like
5102 * btrfs_previous_item) don't miss an existing item with a key matching
5103 * the previous key we computed above.
5111 * A helper function to walk down the tree starting at min_key, and looking
5112 * for nodes or leaves that are have a minimum transaction id.
5113 * This is used by the btree defrag code, and tree logging
5115 * This does not cow, but it does stuff the starting key it finds back
5116 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5117 * key and get a writable path.
5119 * This does lock as it descends, and path->keep_locks should be set
5120 * to 1 by the caller.
5122 * This honors path->lowest_level to prevent descent past a given level
5125 * min_trans indicates the oldest transaction that you are interested
5126 * in walking through. Any nodes or leaves older than min_trans are
5127 * skipped over (without reading them).
5129 * returns zero if something useful was found, < 0 on error and 1 if there
5130 * was nothing in the tree that matched the search criteria.
5132 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5133 struct btrfs_path *path,
5136 struct extent_buffer *cur;
5137 struct btrfs_key found_key;
5143 int keep_locks = path->keep_locks;
5145 path->keep_locks = 1;
5147 cur = btrfs_read_lock_root_node(root);
5148 level = btrfs_header_level(cur);
5149 WARN_ON(path->nodes[level]);
5150 path->nodes[level] = cur;
5151 path->locks[level] = BTRFS_READ_LOCK;
5153 if (btrfs_header_generation(cur) < min_trans) {
5158 nritems = btrfs_header_nritems(cur);
5159 level = btrfs_header_level(cur);
5160 sret = bin_search(cur, min_key, level, &slot);
5162 /* at the lowest level, we're done, setup the path and exit */
5163 if (level == path->lowest_level) {
5164 if (slot >= nritems)
5167 path->slots[level] = slot;
5168 btrfs_item_key_to_cpu(cur, &found_key, slot);
5171 if (sret && slot > 0)
5174 * check this node pointer against the min_trans parameters.
5175 * If it is too old, old, skip to the next one.
5177 while (slot < nritems) {
5180 gen = btrfs_node_ptr_generation(cur, slot);
5181 if (gen < min_trans) {
5189 * we didn't find a candidate key in this node, walk forward
5190 * and find another one
5192 if (slot >= nritems) {
5193 path->slots[level] = slot;
5194 btrfs_set_path_blocking(path);
5195 sret = btrfs_find_next_key(root, path, min_key, level,
5198 btrfs_release_path(path);
5204 /* save our key for returning back */
5205 btrfs_node_key_to_cpu(cur, &found_key, slot);
5206 path->slots[level] = slot;
5207 if (level == path->lowest_level) {
5211 btrfs_set_path_blocking(path);
5212 cur = read_node_slot(root, cur, slot);
5213 BUG_ON(!cur); /* -ENOMEM */
5215 btrfs_tree_read_lock(cur);
5217 path->locks[level - 1] = BTRFS_READ_LOCK;
5218 path->nodes[level - 1] = cur;
5219 unlock_up(path, level, 1, 0, NULL);
5220 btrfs_clear_path_blocking(path, NULL, 0);
5223 path->keep_locks = keep_locks;
5225 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5226 btrfs_set_path_blocking(path);
5227 memcpy(min_key, &found_key, sizeof(found_key));
5232 static void tree_move_down(struct btrfs_root *root,
5233 struct btrfs_path *path,
5234 int *level, int root_level)
5236 BUG_ON(*level == 0);
5237 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
5238 path->slots[*level]);
5239 path->slots[*level - 1] = 0;
5243 static int tree_move_next_or_upnext(struct btrfs_root *root,
5244 struct btrfs_path *path,
5245 int *level, int root_level)
5249 nritems = btrfs_header_nritems(path->nodes[*level]);
5251 path->slots[*level]++;
5253 while (path->slots[*level] >= nritems) {
5254 if (*level == root_level)
5258 path->slots[*level] = 0;
5259 free_extent_buffer(path->nodes[*level]);
5260 path->nodes[*level] = NULL;
5262 path->slots[*level]++;
5264 nritems = btrfs_header_nritems(path->nodes[*level]);
5271 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5274 static int tree_advance(struct btrfs_root *root,
5275 struct btrfs_path *path,
5276 int *level, int root_level,
5278 struct btrfs_key *key)
5282 if (*level == 0 || !allow_down) {
5283 ret = tree_move_next_or_upnext(root, path, level, root_level);
5285 tree_move_down(root, path, level, root_level);
5290 btrfs_item_key_to_cpu(path->nodes[*level], key,
5291 path->slots[*level]);
5293 btrfs_node_key_to_cpu(path->nodes[*level], key,
5294 path->slots[*level]);
5299 static int tree_compare_item(struct btrfs_root *left_root,
5300 struct btrfs_path *left_path,
5301 struct btrfs_path *right_path,
5306 unsigned long off1, off2;
5308 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5309 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5313 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5314 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5315 right_path->slots[0]);
5317 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5319 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5326 #define ADVANCE_ONLY_NEXT -1
5329 * This function compares two trees and calls the provided callback for
5330 * every changed/new/deleted item it finds.
5331 * If shared tree blocks are encountered, whole subtrees are skipped, making
5332 * the compare pretty fast on snapshotted subvolumes.
5334 * This currently works on commit roots only. As commit roots are read only,
5335 * we don't do any locking. The commit roots are protected with transactions.
5336 * Transactions are ended and rejoined when a commit is tried in between.
5338 * This function checks for modifications done to the trees while comparing.
5339 * If it detects a change, it aborts immediately.
5341 int btrfs_compare_trees(struct btrfs_root *left_root,
5342 struct btrfs_root *right_root,
5343 btrfs_changed_cb_t changed_cb, void *ctx)
5347 struct btrfs_path *left_path = NULL;
5348 struct btrfs_path *right_path = NULL;
5349 struct btrfs_key left_key;
5350 struct btrfs_key right_key;
5351 char *tmp_buf = NULL;
5352 int left_root_level;
5353 int right_root_level;
5356 int left_end_reached;
5357 int right_end_reached;
5365 left_path = btrfs_alloc_path();
5370 right_path = btrfs_alloc_path();
5376 tmp_buf = kmalloc(left_root->nodesize, GFP_NOFS);
5382 left_path->search_commit_root = 1;
5383 left_path->skip_locking = 1;
5384 right_path->search_commit_root = 1;
5385 right_path->skip_locking = 1;
5388 * Strategy: Go to the first items of both trees. Then do
5390 * If both trees are at level 0
5391 * Compare keys of current items
5392 * If left < right treat left item as new, advance left tree
5394 * If left > right treat right item as deleted, advance right tree
5396 * If left == right do deep compare of items, treat as changed if
5397 * needed, advance both trees and repeat
5398 * If both trees are at the same level but not at level 0
5399 * Compare keys of current nodes/leafs
5400 * If left < right advance left tree and repeat
5401 * If left > right advance right tree and repeat
5402 * If left == right compare blockptrs of the next nodes/leafs
5403 * If they match advance both trees but stay at the same level
5405 * If they don't match advance both trees while allowing to go
5407 * If tree levels are different
5408 * Advance the tree that needs it and repeat
5410 * Advancing a tree means:
5411 * If we are at level 0, try to go to the next slot. If that's not
5412 * possible, go one level up and repeat. Stop when we found a level
5413 * where we could go to the next slot. We may at this point be on a
5416 * If we are not at level 0 and not on shared tree blocks, go one
5419 * If we are not at level 0 and on shared tree blocks, go one slot to
5420 * the right if possible or go up and right.
5423 down_read(&left_root->fs_info->commit_root_sem);
5424 left_level = btrfs_header_level(left_root->commit_root);
5425 left_root_level = left_level;
5426 left_path->nodes[left_level] = left_root->commit_root;
5427 extent_buffer_get(left_path->nodes[left_level]);
5429 right_level = btrfs_header_level(right_root->commit_root);
5430 right_root_level = right_level;
5431 right_path->nodes[right_level] = right_root->commit_root;
5432 extent_buffer_get(right_path->nodes[right_level]);
5433 up_read(&left_root->fs_info->commit_root_sem);
5435 if (left_level == 0)
5436 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5437 &left_key, left_path->slots[left_level]);
5439 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5440 &left_key, left_path->slots[left_level]);
5441 if (right_level == 0)
5442 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5443 &right_key, right_path->slots[right_level]);
5445 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5446 &right_key, right_path->slots[right_level]);
5448 left_end_reached = right_end_reached = 0;
5449 advance_left = advance_right = 0;
5452 if (advance_left && !left_end_reached) {
5453 ret = tree_advance(left_root, left_path, &left_level,
5455 advance_left != ADVANCE_ONLY_NEXT,
5458 left_end_reached = ADVANCE;
5461 if (advance_right && !right_end_reached) {
5462 ret = tree_advance(right_root, right_path, &right_level,
5464 advance_right != ADVANCE_ONLY_NEXT,
5467 right_end_reached = ADVANCE;
5471 if (left_end_reached && right_end_reached) {
5474 } else if (left_end_reached) {
5475 if (right_level == 0) {
5476 ret = changed_cb(left_root, right_root,
5477 left_path, right_path,
5479 BTRFS_COMPARE_TREE_DELETED,
5484 advance_right = ADVANCE;
5486 } else if (right_end_reached) {
5487 if (left_level == 0) {
5488 ret = changed_cb(left_root, right_root,
5489 left_path, right_path,
5491 BTRFS_COMPARE_TREE_NEW,
5496 advance_left = ADVANCE;
5500 if (left_level == 0 && right_level == 0) {
5501 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5503 ret = changed_cb(left_root, right_root,
5504 left_path, right_path,
5506 BTRFS_COMPARE_TREE_NEW,
5510 advance_left = ADVANCE;
5511 } else if (cmp > 0) {
5512 ret = changed_cb(left_root, right_root,
5513 left_path, right_path,
5515 BTRFS_COMPARE_TREE_DELETED,
5519 advance_right = ADVANCE;
5521 enum btrfs_compare_tree_result cmp;
5523 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5524 ret = tree_compare_item(left_root, left_path,
5525 right_path, tmp_buf);
5527 cmp = BTRFS_COMPARE_TREE_CHANGED;
5529 cmp = BTRFS_COMPARE_TREE_SAME;
5530 ret = changed_cb(left_root, right_root,
5531 left_path, right_path,
5532 &left_key, cmp, ctx);
5535 advance_left = ADVANCE;
5536 advance_right = ADVANCE;
5538 } else if (left_level == right_level) {
5539 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5541 advance_left = ADVANCE;
5542 } else if (cmp > 0) {
5543 advance_right = ADVANCE;
5545 left_blockptr = btrfs_node_blockptr(
5546 left_path->nodes[left_level],
5547 left_path->slots[left_level]);
5548 right_blockptr = btrfs_node_blockptr(
5549 right_path->nodes[right_level],
5550 right_path->slots[right_level]);
5551 left_gen = btrfs_node_ptr_generation(
5552 left_path->nodes[left_level],
5553 left_path->slots[left_level]);
5554 right_gen = btrfs_node_ptr_generation(
5555 right_path->nodes[right_level],
5556 right_path->slots[right_level]);
5557 if (left_blockptr == right_blockptr &&
5558 left_gen == right_gen) {
5560 * As we're on a shared block, don't
5561 * allow to go deeper.
5563 advance_left = ADVANCE_ONLY_NEXT;
5564 advance_right = ADVANCE_ONLY_NEXT;
5566 advance_left = ADVANCE;
5567 advance_right = ADVANCE;
5570 } else if (left_level < right_level) {
5571 advance_right = ADVANCE;
5573 advance_left = ADVANCE;
5578 btrfs_free_path(left_path);
5579 btrfs_free_path(right_path);
5585 * this is similar to btrfs_next_leaf, but does not try to preserve
5586 * and fixup the path. It looks for and returns the next key in the
5587 * tree based on the current path and the min_trans parameters.
5589 * 0 is returned if another key is found, < 0 if there are any errors
5590 * and 1 is returned if there are no higher keys in the tree
5592 * path->keep_locks should be set to 1 on the search made before
5593 * calling this function.
5595 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5596 struct btrfs_key *key, int level, u64 min_trans)
5599 struct extent_buffer *c;
5601 WARN_ON(!path->keep_locks);
5602 while (level < BTRFS_MAX_LEVEL) {
5603 if (!path->nodes[level])
5606 slot = path->slots[level] + 1;
5607 c = path->nodes[level];
5609 if (slot >= btrfs_header_nritems(c)) {
5612 struct btrfs_key cur_key;
5613 if (level + 1 >= BTRFS_MAX_LEVEL ||
5614 !path->nodes[level + 1])
5617 if (path->locks[level + 1]) {
5622 slot = btrfs_header_nritems(c) - 1;
5624 btrfs_item_key_to_cpu(c, &cur_key, slot);
5626 btrfs_node_key_to_cpu(c, &cur_key, slot);
5628 orig_lowest = path->lowest_level;
5629 btrfs_release_path(path);
5630 path->lowest_level = level;
5631 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5633 path->lowest_level = orig_lowest;
5637 c = path->nodes[level];
5638 slot = path->slots[level];
5645 btrfs_item_key_to_cpu(c, key, slot);
5647 u64 gen = btrfs_node_ptr_generation(c, slot);
5649 if (gen < min_trans) {
5653 btrfs_node_key_to_cpu(c, key, slot);
5661 * search the tree again to find a leaf with greater keys
5662 * returns 0 if it found something or 1 if there are no greater leaves.
5663 * returns < 0 on io errors.
5665 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5667 return btrfs_next_old_leaf(root, path, 0);
5670 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5675 struct extent_buffer *c;
5676 struct extent_buffer *next;
5677 struct btrfs_key key;
5680 int old_spinning = path->leave_spinning;
5681 int next_rw_lock = 0;
5683 nritems = btrfs_header_nritems(path->nodes[0]);
5687 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5692 btrfs_release_path(path);
5694 path->keep_locks = 1;
5695 path->leave_spinning = 1;
5698 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5700 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5701 path->keep_locks = 0;
5706 nritems = btrfs_header_nritems(path->nodes[0]);
5708 * by releasing the path above we dropped all our locks. A balance
5709 * could have added more items next to the key that used to be
5710 * at the very end of the block. So, check again here and
5711 * advance the path if there are now more items available.
5713 if (nritems > 0 && path->slots[0] < nritems - 1) {
5720 * So the above check misses one case:
5721 * - after releasing the path above, someone has removed the item that
5722 * used to be at the very end of the block, and balance between leafs
5723 * gets another one with bigger key.offset to replace it.
5725 * This one should be returned as well, or we can get leaf corruption
5726 * later(esp. in __btrfs_drop_extents()).
5728 * And a bit more explanation about this check,
5729 * with ret > 0, the key isn't found, the path points to the slot
5730 * where it should be inserted, so the path->slots[0] item must be the
5733 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5738 while (level < BTRFS_MAX_LEVEL) {
5739 if (!path->nodes[level]) {
5744 slot = path->slots[level] + 1;
5745 c = path->nodes[level];
5746 if (slot >= btrfs_header_nritems(c)) {
5748 if (level == BTRFS_MAX_LEVEL) {
5756 btrfs_tree_unlock_rw(next, next_rw_lock);
5757 free_extent_buffer(next);
5761 next_rw_lock = path->locks[level];
5762 ret = read_block_for_search(NULL, root, path, &next, level,
5768 btrfs_release_path(path);
5772 if (!path->skip_locking) {
5773 ret = btrfs_try_tree_read_lock(next);
5774 if (!ret && time_seq) {
5776 * If we don't get the lock, we may be racing
5777 * with push_leaf_left, holding that lock while
5778 * itself waiting for the leaf we've currently
5779 * locked. To solve this situation, we give up
5780 * on our lock and cycle.
5782 free_extent_buffer(next);
5783 btrfs_release_path(path);
5788 btrfs_set_path_blocking(path);
5789 btrfs_tree_read_lock(next);
5790 btrfs_clear_path_blocking(path, next,
5793 next_rw_lock = BTRFS_READ_LOCK;
5797 path->slots[level] = slot;
5800 c = path->nodes[level];
5801 if (path->locks[level])
5802 btrfs_tree_unlock_rw(c, path->locks[level]);
5804 free_extent_buffer(c);
5805 path->nodes[level] = next;
5806 path->slots[level] = 0;
5807 if (!path->skip_locking)
5808 path->locks[level] = next_rw_lock;
5812 ret = read_block_for_search(NULL, root, path, &next, level,
5818 btrfs_release_path(path);
5822 if (!path->skip_locking) {
5823 ret = btrfs_try_tree_read_lock(next);
5825 btrfs_set_path_blocking(path);
5826 btrfs_tree_read_lock(next);
5827 btrfs_clear_path_blocking(path, next,
5830 next_rw_lock = BTRFS_READ_LOCK;
5835 unlock_up(path, 0, 1, 0, NULL);
5836 path->leave_spinning = old_spinning;
5838 btrfs_set_path_blocking(path);
5844 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5845 * searching until it gets past min_objectid or finds an item of 'type'
5847 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5849 int btrfs_previous_item(struct btrfs_root *root,
5850 struct btrfs_path *path, u64 min_objectid,
5853 struct btrfs_key found_key;
5854 struct extent_buffer *leaf;
5859 if (path->slots[0] == 0) {
5860 btrfs_set_path_blocking(path);
5861 ret = btrfs_prev_leaf(root, path);
5867 leaf = path->nodes[0];
5868 nritems = btrfs_header_nritems(leaf);
5871 if (path->slots[0] == nritems)
5874 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5875 if (found_key.objectid < min_objectid)
5877 if (found_key.type == type)
5879 if (found_key.objectid == min_objectid &&
5880 found_key.type < type)
5887 * search in extent tree to find a previous Metadata/Data extent item with
5890 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5892 int btrfs_previous_extent_item(struct btrfs_root *root,
5893 struct btrfs_path *path, u64 min_objectid)
5895 struct btrfs_key found_key;
5896 struct extent_buffer *leaf;
5901 if (path->slots[0] == 0) {
5902 btrfs_set_path_blocking(path);
5903 ret = btrfs_prev_leaf(root, path);
5909 leaf = path->nodes[0];
5910 nritems = btrfs_header_nritems(leaf);
5913 if (path->slots[0] == nritems)
5916 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5917 if (found_key.objectid < min_objectid)
5919 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5920 found_key.type == BTRFS_METADATA_ITEM_KEY)
5922 if (found_key.objectid == min_objectid &&
5923 found_key.type < BTRFS_EXTENT_ITEM_KEY)