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_tree_block(trans, root, 0, new_root_objectid,
262 &disk_key, level, buf->start, 0);
266 copy_extent_buffer(cow, buf, 0, 0, cow->len);
267 btrfs_set_header_bytenr(cow, cow->start);
268 btrfs_set_header_generation(cow, trans->transid);
269 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
270 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
271 BTRFS_HEADER_FLAG_RELOC);
272 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
273 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
275 btrfs_set_header_owner(cow, new_root_objectid);
277 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
280 WARN_ON(btrfs_header_generation(buf) > trans->transid);
281 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
282 ret = btrfs_inc_ref(trans, root, cow, 1);
284 ret = btrfs_inc_ref(trans, root, cow, 0);
289 btrfs_mark_buffer_dirty(cow);
298 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
299 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
301 MOD_LOG_ROOT_REPLACE,
304 struct tree_mod_move {
309 struct tree_mod_root {
314 struct tree_mod_elem {
316 u64 index; /* shifted logical */
320 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
323 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
326 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
327 struct btrfs_disk_key key;
330 /* this is used for op == MOD_LOG_MOVE_KEYS */
331 struct tree_mod_move move;
333 /* this is used for op == MOD_LOG_ROOT_REPLACE */
334 struct tree_mod_root old_root;
337 static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
339 read_lock(&fs_info->tree_mod_log_lock);
342 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
344 read_unlock(&fs_info->tree_mod_log_lock);
347 static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
349 write_lock(&fs_info->tree_mod_log_lock);
352 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
354 write_unlock(&fs_info->tree_mod_log_lock);
358 * Pull a new tree mod seq number for our operation.
360 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
362 return atomic64_inc_return(&fs_info->tree_mod_seq);
366 * This adds a new blocker to the tree mod log's blocker list if the @elem
367 * passed does not already have a sequence number set. So when a caller expects
368 * to record tree modifications, it should ensure to set elem->seq to zero
369 * before calling btrfs_get_tree_mod_seq.
370 * Returns a fresh, unused tree log modification sequence number, even if no new
373 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
374 struct seq_list *elem)
376 tree_mod_log_write_lock(fs_info);
377 spin_lock(&fs_info->tree_mod_seq_lock);
379 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
380 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
382 spin_unlock(&fs_info->tree_mod_seq_lock);
383 tree_mod_log_write_unlock(fs_info);
388 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
389 struct seq_list *elem)
391 struct rb_root *tm_root;
392 struct rb_node *node;
393 struct rb_node *next;
394 struct seq_list *cur_elem;
395 struct tree_mod_elem *tm;
396 u64 min_seq = (u64)-1;
397 u64 seq_putting = elem->seq;
402 spin_lock(&fs_info->tree_mod_seq_lock);
403 list_del(&elem->list);
406 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
407 if (cur_elem->seq < min_seq) {
408 if (seq_putting > cur_elem->seq) {
410 * blocker with lower sequence number exists, we
411 * cannot remove anything from the log
413 spin_unlock(&fs_info->tree_mod_seq_lock);
416 min_seq = cur_elem->seq;
419 spin_unlock(&fs_info->tree_mod_seq_lock);
422 * anything that's lower than the lowest existing (read: blocked)
423 * sequence number can be removed from the tree.
425 tree_mod_log_write_lock(fs_info);
426 tm_root = &fs_info->tree_mod_log;
427 for (node = rb_first(tm_root); node; node = next) {
428 next = rb_next(node);
429 tm = container_of(node, struct tree_mod_elem, node);
430 if (tm->seq > min_seq)
432 rb_erase(node, tm_root);
435 tree_mod_log_write_unlock(fs_info);
439 * key order of the log:
442 * the index is the shifted logical of the *new* root node for root replace
443 * operations, or the shifted logical of the affected block for all other
446 * Note: must be called with write lock (tree_mod_log_write_lock).
449 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
451 struct rb_root *tm_root;
452 struct rb_node **new;
453 struct rb_node *parent = NULL;
454 struct tree_mod_elem *cur;
458 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
460 tm_root = &fs_info->tree_mod_log;
461 new = &tm_root->rb_node;
463 cur = container_of(*new, struct tree_mod_elem, node);
465 if (cur->index < tm->index)
466 new = &((*new)->rb_left);
467 else if (cur->index > tm->index)
468 new = &((*new)->rb_right);
469 else if (cur->seq < tm->seq)
470 new = &((*new)->rb_left);
471 else if (cur->seq > tm->seq)
472 new = &((*new)->rb_right);
477 rb_link_node(&tm->node, parent, new);
478 rb_insert_color(&tm->node, tm_root);
483 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
484 * returns zero with the tree_mod_log_lock acquired. The caller must hold
485 * this until all tree mod log insertions are recorded in the rb tree and then
486 * call tree_mod_log_write_unlock() to release.
488 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
489 struct extent_buffer *eb) {
491 if (list_empty(&(fs_info)->tree_mod_seq_list))
493 if (eb && btrfs_header_level(eb) == 0)
496 tree_mod_log_write_lock(fs_info);
497 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
498 tree_mod_log_write_unlock(fs_info);
505 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
506 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
507 struct extent_buffer *eb)
510 if (list_empty(&(fs_info)->tree_mod_seq_list))
512 if (eb && btrfs_header_level(eb) == 0)
518 static struct tree_mod_elem *
519 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
520 enum mod_log_op op, gfp_t flags)
522 struct tree_mod_elem *tm;
524 tm = kzalloc(sizeof(*tm), flags);
528 tm->index = eb->start >> PAGE_CACHE_SHIFT;
529 if (op != MOD_LOG_KEY_ADD) {
530 btrfs_node_key(eb, &tm->key, slot);
531 tm->blockptr = btrfs_node_blockptr(eb, slot);
535 tm->generation = btrfs_node_ptr_generation(eb, slot);
536 RB_CLEAR_NODE(&tm->node);
542 tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
543 struct extent_buffer *eb, int slot,
544 enum mod_log_op op, gfp_t flags)
546 struct tree_mod_elem *tm;
549 if (!tree_mod_need_log(fs_info, eb))
552 tm = alloc_tree_mod_elem(eb, slot, op, flags);
556 if (tree_mod_dont_log(fs_info, eb)) {
561 ret = __tree_mod_log_insert(fs_info, tm);
562 tree_mod_log_write_unlock(fs_info);
570 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
571 struct extent_buffer *eb, int dst_slot, int src_slot,
572 int nr_items, gfp_t flags)
574 struct tree_mod_elem *tm = NULL;
575 struct tree_mod_elem **tm_list = NULL;
580 if (!tree_mod_need_log(fs_info, eb))
583 tm_list = kzalloc(nr_items * sizeof(struct tree_mod_elem *), flags);
587 tm = kzalloc(sizeof(*tm), flags);
593 tm->index = eb->start >> PAGE_CACHE_SHIFT;
595 tm->move.dst_slot = dst_slot;
596 tm->move.nr_items = nr_items;
597 tm->op = MOD_LOG_MOVE_KEYS;
599 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
600 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
601 MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags);
608 if (tree_mod_dont_log(fs_info, eb))
613 * When we override something during the move, we log these removals.
614 * This can only happen when we move towards the beginning of the
615 * buffer, i.e. dst_slot < src_slot.
617 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
618 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
623 ret = __tree_mod_log_insert(fs_info, tm);
626 tree_mod_log_write_unlock(fs_info);
631 for (i = 0; i < nr_items; i++) {
632 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
633 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
637 tree_mod_log_write_unlock(fs_info);
645 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
646 struct tree_mod_elem **tm_list,
652 for (i = nritems - 1; i >= 0; i--) {
653 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
655 for (j = nritems - 1; j > i; j--)
656 rb_erase(&tm_list[j]->node,
657 &fs_info->tree_mod_log);
666 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
667 struct extent_buffer *old_root,
668 struct extent_buffer *new_root, gfp_t flags,
671 struct tree_mod_elem *tm = NULL;
672 struct tree_mod_elem **tm_list = NULL;
677 if (!tree_mod_need_log(fs_info, NULL))
680 if (log_removal && btrfs_header_level(old_root) > 0) {
681 nritems = btrfs_header_nritems(old_root);
682 tm_list = kzalloc(nritems * sizeof(struct tree_mod_elem *),
688 for (i = 0; i < nritems; i++) {
689 tm_list[i] = alloc_tree_mod_elem(old_root, i,
690 MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags);
698 tm = kzalloc(sizeof(*tm), flags);
704 tm->index = new_root->start >> PAGE_CACHE_SHIFT;
705 tm->old_root.logical = old_root->start;
706 tm->old_root.level = btrfs_header_level(old_root);
707 tm->generation = btrfs_header_generation(old_root);
708 tm->op = MOD_LOG_ROOT_REPLACE;
710 if (tree_mod_dont_log(fs_info, NULL))
714 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
716 ret = __tree_mod_log_insert(fs_info, tm);
718 tree_mod_log_write_unlock(fs_info);
727 for (i = 0; i < nritems; i++)
736 static struct tree_mod_elem *
737 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
740 struct rb_root *tm_root;
741 struct rb_node *node;
742 struct tree_mod_elem *cur = NULL;
743 struct tree_mod_elem *found = NULL;
744 u64 index = start >> PAGE_CACHE_SHIFT;
746 tree_mod_log_read_lock(fs_info);
747 tm_root = &fs_info->tree_mod_log;
748 node = tm_root->rb_node;
750 cur = container_of(node, struct tree_mod_elem, node);
751 if (cur->index < index) {
752 node = node->rb_left;
753 } else if (cur->index > index) {
754 node = node->rb_right;
755 } else if (cur->seq < min_seq) {
756 node = node->rb_left;
757 } else if (!smallest) {
758 /* we want the node with the highest seq */
760 BUG_ON(found->seq > cur->seq);
762 node = node->rb_left;
763 } else if (cur->seq > min_seq) {
764 /* we want the node with the smallest seq */
766 BUG_ON(found->seq < cur->seq);
768 node = node->rb_right;
774 tree_mod_log_read_unlock(fs_info);
780 * this returns the element from the log with the smallest time sequence
781 * value that's in the log (the oldest log item). any element with a time
782 * sequence lower than min_seq will be ignored.
784 static struct tree_mod_elem *
785 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
788 return __tree_mod_log_search(fs_info, start, min_seq, 1);
792 * this returns the element from the log with the largest time sequence
793 * value that's in the log (the most recent log item). any element with
794 * a time sequence lower than min_seq will be ignored.
796 static struct tree_mod_elem *
797 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
799 return __tree_mod_log_search(fs_info, start, min_seq, 0);
803 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
804 struct extent_buffer *src, unsigned long dst_offset,
805 unsigned long src_offset, int nr_items)
808 struct tree_mod_elem **tm_list = NULL;
809 struct tree_mod_elem **tm_list_add, **tm_list_rem;
813 if (!tree_mod_need_log(fs_info, NULL))
816 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
819 tm_list = kzalloc(nr_items * 2 * sizeof(struct tree_mod_elem *),
824 tm_list_add = tm_list;
825 tm_list_rem = tm_list + nr_items;
826 for (i = 0; i < nr_items; i++) {
827 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
828 MOD_LOG_KEY_REMOVE, GFP_NOFS);
829 if (!tm_list_rem[i]) {
834 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
835 MOD_LOG_KEY_ADD, GFP_NOFS);
836 if (!tm_list_add[i]) {
842 if (tree_mod_dont_log(fs_info, NULL))
846 for (i = 0; i < nr_items; i++) {
847 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
850 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
855 tree_mod_log_write_unlock(fs_info);
861 for (i = 0; i < nr_items * 2; i++) {
862 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
863 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
867 tree_mod_log_write_unlock(fs_info);
874 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
875 int dst_offset, int src_offset, int nr_items)
878 ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
884 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
885 struct extent_buffer *eb, int slot, int atomic)
889 ret = tree_mod_log_insert_key(fs_info, eb, slot,
891 atomic ? GFP_ATOMIC : GFP_NOFS);
896 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
898 struct tree_mod_elem **tm_list = NULL;
903 if (btrfs_header_level(eb) == 0)
906 if (!tree_mod_need_log(fs_info, NULL))
909 nritems = btrfs_header_nritems(eb);
910 tm_list = kzalloc(nritems * sizeof(struct tree_mod_elem *),
915 for (i = 0; i < nritems; i++) {
916 tm_list[i] = alloc_tree_mod_elem(eb, i,
917 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
924 if (tree_mod_dont_log(fs_info, eb))
927 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
928 tree_mod_log_write_unlock(fs_info);
936 for (i = 0; i < nritems; i++)
944 tree_mod_log_set_root_pointer(struct btrfs_root *root,
945 struct extent_buffer *new_root_node,
949 ret = tree_mod_log_insert_root(root->fs_info, root->node,
950 new_root_node, GFP_NOFS, log_removal);
955 * check if the tree block can be shared by multiple trees
957 int btrfs_block_can_be_shared(struct btrfs_root *root,
958 struct extent_buffer *buf)
961 * Tree blocks not in refernece counted trees and tree roots
962 * are never shared. If a block was allocated after the last
963 * snapshot and the block was not allocated by tree relocation,
964 * we know the block is not shared.
966 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
967 buf != root->node && buf != root->commit_root &&
968 (btrfs_header_generation(buf) <=
969 btrfs_root_last_snapshot(&root->root_item) ||
970 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
972 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
973 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
974 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
980 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
981 struct btrfs_root *root,
982 struct extent_buffer *buf,
983 struct extent_buffer *cow,
993 * Backrefs update rules:
995 * Always use full backrefs for extent pointers in tree block
996 * allocated by tree relocation.
998 * If a shared tree block is no longer referenced by its owner
999 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
1000 * use full backrefs for extent pointers in tree block.
1002 * If a tree block is been relocating
1003 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1004 * use full backrefs for extent pointers in tree block.
1005 * The reason for this is some operations (such as drop tree)
1006 * are only allowed for blocks use full backrefs.
1009 if (btrfs_block_can_be_shared(root, buf)) {
1010 ret = btrfs_lookup_extent_info(trans, root, buf->start,
1011 btrfs_header_level(buf), 1,
1017 btrfs_std_error(root->fs_info, ret);
1022 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1023 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1024 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1029 owner = btrfs_header_owner(buf);
1030 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
1031 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
1034 if ((owner == root->root_key.objectid ||
1035 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
1036 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1037 ret = btrfs_inc_ref(trans, root, buf, 1);
1038 BUG_ON(ret); /* -ENOMEM */
1040 if (root->root_key.objectid ==
1041 BTRFS_TREE_RELOC_OBJECTID) {
1042 ret = btrfs_dec_ref(trans, root, buf, 0);
1043 BUG_ON(ret); /* -ENOMEM */
1044 ret = btrfs_inc_ref(trans, root, cow, 1);
1045 BUG_ON(ret); /* -ENOMEM */
1047 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
1050 if (root->root_key.objectid ==
1051 BTRFS_TREE_RELOC_OBJECTID)
1052 ret = btrfs_inc_ref(trans, root, cow, 1);
1054 ret = btrfs_inc_ref(trans, root, cow, 0);
1055 BUG_ON(ret); /* -ENOMEM */
1057 if (new_flags != 0) {
1058 int level = btrfs_header_level(buf);
1060 ret = btrfs_set_disk_extent_flags(trans, root,
1063 new_flags, level, 0);
1068 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
1069 if (root->root_key.objectid ==
1070 BTRFS_TREE_RELOC_OBJECTID)
1071 ret = btrfs_inc_ref(trans, root, cow, 1);
1073 ret = btrfs_inc_ref(trans, root, cow, 0);
1074 BUG_ON(ret); /* -ENOMEM */
1075 ret = btrfs_dec_ref(trans, root, buf, 1);
1076 BUG_ON(ret); /* -ENOMEM */
1078 clean_tree_block(trans, root, buf);
1085 * does the dirty work in cow of a single block. The parent block (if
1086 * supplied) is updated to point to the new cow copy. The new buffer is marked
1087 * dirty and returned locked. If you modify the block it needs to be marked
1090 * search_start -- an allocation hint for the new block
1092 * empty_size -- a hint that you plan on doing more cow. This is the size in
1093 * bytes the allocator should try to find free next to the block it returns.
1094 * This is just a hint and may be ignored by the allocator.
1096 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1097 struct btrfs_root *root,
1098 struct extent_buffer *buf,
1099 struct extent_buffer *parent, int parent_slot,
1100 struct extent_buffer **cow_ret,
1101 u64 search_start, u64 empty_size)
1103 struct btrfs_disk_key disk_key;
1104 struct extent_buffer *cow;
1107 int unlock_orig = 0;
1110 if (*cow_ret == buf)
1113 btrfs_assert_tree_locked(buf);
1115 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1116 trans->transid != root->fs_info->running_transaction->transid);
1117 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1118 trans->transid != root->last_trans);
1120 level = btrfs_header_level(buf);
1123 btrfs_item_key(buf, &disk_key, 0);
1125 btrfs_node_key(buf, &disk_key, 0);
1127 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1129 parent_start = parent->start;
1135 cow = btrfs_alloc_tree_block(trans, root, parent_start,
1136 root->root_key.objectid, &disk_key, level,
1137 search_start, empty_size);
1139 return PTR_ERR(cow);
1141 /* cow is set to blocking by btrfs_init_new_buffer */
1143 copy_extent_buffer(cow, buf, 0, 0, cow->len);
1144 btrfs_set_header_bytenr(cow, cow->start);
1145 btrfs_set_header_generation(cow, trans->transid);
1146 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1147 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1148 BTRFS_HEADER_FLAG_RELOC);
1149 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1150 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1152 btrfs_set_header_owner(cow, root->root_key.objectid);
1154 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
1157 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1159 btrfs_abort_transaction(trans, root, ret);
1163 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1164 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1169 if (buf == root->node) {
1170 WARN_ON(parent && parent != buf);
1171 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1172 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1173 parent_start = buf->start;
1177 extent_buffer_get(cow);
1178 tree_mod_log_set_root_pointer(root, cow, 1);
1179 rcu_assign_pointer(root->node, cow);
1181 btrfs_free_tree_block(trans, root, buf, parent_start,
1183 free_extent_buffer(buf);
1184 add_root_to_dirty_list(root);
1186 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1187 parent_start = parent->start;
1191 WARN_ON(trans->transid != btrfs_header_generation(parent));
1192 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1193 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1194 btrfs_set_node_blockptr(parent, parent_slot,
1196 btrfs_set_node_ptr_generation(parent, parent_slot,
1198 btrfs_mark_buffer_dirty(parent);
1200 ret = tree_mod_log_free_eb(root->fs_info, buf);
1202 btrfs_abort_transaction(trans, root, ret);
1206 btrfs_free_tree_block(trans, root, buf, parent_start,
1210 btrfs_tree_unlock(buf);
1211 free_extent_buffer_stale(buf);
1212 btrfs_mark_buffer_dirty(cow);
1218 * returns the logical address of the oldest predecessor of the given root.
1219 * entries older than time_seq are ignored.
1221 static struct tree_mod_elem *
1222 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1223 struct extent_buffer *eb_root, u64 time_seq)
1225 struct tree_mod_elem *tm;
1226 struct tree_mod_elem *found = NULL;
1227 u64 root_logical = eb_root->start;
1234 * the very last operation that's logged for a root is the replacement
1235 * operation (if it is replaced at all). this has the index of the *new*
1236 * root, making it the very first operation that's logged for this root.
1239 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1244 * if there are no tree operation for the oldest root, we simply
1245 * return it. this should only happen if that (old) root is at
1252 * if there's an operation that's not a root replacement, we
1253 * found the oldest version of our root. normally, we'll find a
1254 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1256 if (tm->op != MOD_LOG_ROOT_REPLACE)
1260 root_logical = tm->old_root.logical;
1264 /* if there's no old root to return, return what we found instead */
1272 * tm is a pointer to the first operation to rewind within eb. then, all
1273 * previous operations will be rewinded (until we reach something older than
1277 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1278 u64 time_seq, struct tree_mod_elem *first_tm)
1281 struct rb_node *next;
1282 struct tree_mod_elem *tm = first_tm;
1283 unsigned long o_dst;
1284 unsigned long o_src;
1285 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1287 n = btrfs_header_nritems(eb);
1288 tree_mod_log_read_lock(fs_info);
1289 while (tm && tm->seq >= time_seq) {
1291 * all the operations are recorded with the operator used for
1292 * the modification. as we're going backwards, we do the
1293 * opposite of each operation here.
1296 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1297 BUG_ON(tm->slot < n);
1299 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1300 case MOD_LOG_KEY_REMOVE:
1301 btrfs_set_node_key(eb, &tm->key, tm->slot);
1302 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1303 btrfs_set_node_ptr_generation(eb, tm->slot,
1307 case MOD_LOG_KEY_REPLACE:
1308 BUG_ON(tm->slot >= n);
1309 btrfs_set_node_key(eb, &tm->key, tm->slot);
1310 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1311 btrfs_set_node_ptr_generation(eb, tm->slot,
1314 case MOD_LOG_KEY_ADD:
1315 /* if a move operation is needed it's in the log */
1318 case MOD_LOG_MOVE_KEYS:
1319 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1320 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1321 memmove_extent_buffer(eb, o_dst, o_src,
1322 tm->move.nr_items * p_size);
1324 case MOD_LOG_ROOT_REPLACE:
1326 * this operation is special. for roots, this must be
1327 * handled explicitly before rewinding.
1328 * for non-roots, this operation may exist if the node
1329 * was a root: root A -> child B; then A gets empty and
1330 * B is promoted to the new root. in the mod log, we'll
1331 * have a root-replace operation for B, a tree block
1332 * that is no root. we simply ignore that operation.
1336 next = rb_next(&tm->node);
1339 tm = container_of(next, struct tree_mod_elem, node);
1340 if (tm->index != first_tm->index)
1343 tree_mod_log_read_unlock(fs_info);
1344 btrfs_set_header_nritems(eb, n);
1348 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1349 * is returned. If rewind operations happen, a fresh buffer is returned. The
1350 * returned buffer is always read-locked. If the returned buffer is not the
1351 * input buffer, the lock on the input buffer is released and the input buffer
1352 * is freed (its refcount is decremented).
1354 static struct extent_buffer *
1355 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1356 struct extent_buffer *eb, u64 time_seq)
1358 struct extent_buffer *eb_rewin;
1359 struct tree_mod_elem *tm;
1364 if (btrfs_header_level(eb) == 0)
1367 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1371 btrfs_set_path_blocking(path);
1372 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1374 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1375 BUG_ON(tm->slot != 0);
1376 eb_rewin = alloc_dummy_extent_buffer(eb->start,
1377 fs_info->tree_root->nodesize);
1379 btrfs_tree_read_unlock_blocking(eb);
1380 free_extent_buffer(eb);
1383 btrfs_set_header_bytenr(eb_rewin, eb->start);
1384 btrfs_set_header_backref_rev(eb_rewin,
1385 btrfs_header_backref_rev(eb));
1386 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1387 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1389 eb_rewin = btrfs_clone_extent_buffer(eb);
1391 btrfs_tree_read_unlock_blocking(eb);
1392 free_extent_buffer(eb);
1397 btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
1398 btrfs_tree_read_unlock_blocking(eb);
1399 free_extent_buffer(eb);
1401 extent_buffer_get(eb_rewin);
1402 btrfs_tree_read_lock(eb_rewin);
1403 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1404 WARN_ON(btrfs_header_nritems(eb_rewin) >
1405 BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1411 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1412 * value. If there are no changes, the current root->root_node is returned. If
1413 * anything changed in between, there's a fresh buffer allocated on which the
1414 * rewind operations are done. In any case, the returned buffer is read locked.
1415 * Returns NULL on error (with no locks held).
1417 static inline struct extent_buffer *
1418 get_old_root(struct btrfs_root *root, u64 time_seq)
1420 struct tree_mod_elem *tm;
1421 struct extent_buffer *eb = NULL;
1422 struct extent_buffer *eb_root;
1423 struct extent_buffer *old;
1424 struct tree_mod_root *old_root = NULL;
1425 u64 old_generation = 0;
1428 eb_root = btrfs_read_lock_root_node(root);
1429 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1433 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1434 old_root = &tm->old_root;
1435 old_generation = tm->generation;
1436 logical = old_root->logical;
1438 logical = eb_root->start;
1441 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1442 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1443 btrfs_tree_read_unlock(eb_root);
1444 free_extent_buffer(eb_root);
1445 old = read_tree_block(root, logical, 0);
1446 if (WARN_ON(!old || !extent_buffer_uptodate(old))) {
1447 free_extent_buffer(old);
1448 btrfs_warn(root->fs_info,
1449 "failed to read tree block %llu from get_old_root", logical);
1451 eb = btrfs_clone_extent_buffer(old);
1452 free_extent_buffer(old);
1454 } else if (old_root) {
1455 btrfs_tree_read_unlock(eb_root);
1456 free_extent_buffer(eb_root);
1457 eb = alloc_dummy_extent_buffer(logical, root->nodesize);
1459 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1460 eb = btrfs_clone_extent_buffer(eb_root);
1461 btrfs_tree_read_unlock_blocking(eb_root);
1462 free_extent_buffer(eb_root);
1467 extent_buffer_get(eb);
1468 btrfs_tree_read_lock(eb);
1470 btrfs_set_header_bytenr(eb, eb->start);
1471 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1472 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1473 btrfs_set_header_level(eb, old_root->level);
1474 btrfs_set_header_generation(eb, old_generation);
1477 __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1479 WARN_ON(btrfs_header_level(eb) != 0);
1480 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1485 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1487 struct tree_mod_elem *tm;
1489 struct extent_buffer *eb_root = btrfs_root_node(root);
1491 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1492 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1493 level = tm->old_root.level;
1495 level = btrfs_header_level(eb_root);
1497 free_extent_buffer(eb_root);
1502 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1503 struct btrfs_root *root,
1504 struct extent_buffer *buf)
1506 if (btrfs_test_is_dummy_root(root))
1509 /* ensure we can see the force_cow */
1513 * We do not need to cow a block if
1514 * 1) this block is not created or changed in this transaction;
1515 * 2) this block does not belong to TREE_RELOC tree;
1516 * 3) the root is not forced COW.
1518 * What is forced COW:
1519 * when we create snapshot during commiting the transaction,
1520 * after we've finished coping src root, we must COW the shared
1521 * block to ensure the metadata consistency.
1523 if (btrfs_header_generation(buf) == trans->transid &&
1524 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1525 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1526 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1527 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1533 * cows a single block, see __btrfs_cow_block for the real work.
1534 * This version of it has extra checks so that a block isn't cow'd more than
1535 * once per transaction, as long as it hasn't been written yet
1537 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1538 struct btrfs_root *root, struct extent_buffer *buf,
1539 struct extent_buffer *parent, int parent_slot,
1540 struct extent_buffer **cow_ret)
1545 if (trans->transaction != root->fs_info->running_transaction)
1546 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1548 root->fs_info->running_transaction->transid);
1550 if (trans->transid != root->fs_info->generation)
1551 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1552 trans->transid, root->fs_info->generation);
1554 if (!should_cow_block(trans, root, buf)) {
1559 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
1562 btrfs_set_lock_blocking(parent);
1563 btrfs_set_lock_blocking(buf);
1565 ret = __btrfs_cow_block(trans, root, buf, parent,
1566 parent_slot, cow_ret, search_start, 0);
1568 trace_btrfs_cow_block(root, buf, *cow_ret);
1574 * helper function for defrag to decide if two blocks pointed to by a
1575 * node are actually close by
1577 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1579 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1581 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1587 * compare two keys in a memcmp fashion
1589 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1591 struct btrfs_key k1;
1593 btrfs_disk_key_to_cpu(&k1, disk);
1595 return btrfs_comp_cpu_keys(&k1, k2);
1599 * same as comp_keys only with two btrfs_key's
1601 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1603 if (k1->objectid > k2->objectid)
1605 if (k1->objectid < k2->objectid)
1607 if (k1->type > k2->type)
1609 if (k1->type < k2->type)
1611 if (k1->offset > k2->offset)
1613 if (k1->offset < k2->offset)
1619 * this is used by the defrag code to go through all the
1620 * leaves pointed to by a node and reallocate them so that
1621 * disk order is close to key order
1623 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1624 struct btrfs_root *root, struct extent_buffer *parent,
1625 int start_slot, u64 *last_ret,
1626 struct btrfs_key *progress)
1628 struct extent_buffer *cur;
1631 u64 search_start = *last_ret;
1641 int progress_passed = 0;
1642 struct btrfs_disk_key disk_key;
1644 parent_level = btrfs_header_level(parent);
1646 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1647 WARN_ON(trans->transid != root->fs_info->generation);
1649 parent_nritems = btrfs_header_nritems(parent);
1650 blocksize = root->nodesize;
1651 end_slot = parent_nritems;
1653 if (parent_nritems == 1)
1656 btrfs_set_lock_blocking(parent);
1658 for (i = start_slot; i < end_slot; i++) {
1661 btrfs_node_key(parent, &disk_key, i);
1662 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1665 progress_passed = 1;
1666 blocknr = btrfs_node_blockptr(parent, i);
1667 gen = btrfs_node_ptr_generation(parent, i);
1668 if (last_block == 0)
1669 last_block = blocknr;
1672 other = btrfs_node_blockptr(parent, i - 1);
1673 close = close_blocks(blocknr, other, blocksize);
1675 if (!close && i < end_slot - 2) {
1676 other = btrfs_node_blockptr(parent, i + 1);
1677 close = close_blocks(blocknr, other, blocksize);
1680 last_block = blocknr;
1684 cur = btrfs_find_tree_block(root, blocknr);
1686 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1689 if (!cur || !uptodate) {
1691 cur = read_tree_block(root, blocknr, gen);
1692 if (!cur || !extent_buffer_uptodate(cur)) {
1693 free_extent_buffer(cur);
1696 } else if (!uptodate) {
1697 err = btrfs_read_buffer(cur, gen);
1699 free_extent_buffer(cur);
1704 if (search_start == 0)
1705 search_start = last_block;
1707 btrfs_tree_lock(cur);
1708 btrfs_set_lock_blocking(cur);
1709 err = __btrfs_cow_block(trans, root, cur, parent, i,
1712 (end_slot - i) * blocksize));
1714 btrfs_tree_unlock(cur);
1715 free_extent_buffer(cur);
1718 search_start = cur->start;
1719 last_block = cur->start;
1720 *last_ret = search_start;
1721 btrfs_tree_unlock(cur);
1722 free_extent_buffer(cur);
1728 * The leaf data grows from end-to-front in the node.
1729 * this returns the address of the start of the last item,
1730 * which is the stop of the leaf data stack
1732 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1733 struct extent_buffer *leaf)
1735 u32 nr = btrfs_header_nritems(leaf);
1737 return BTRFS_LEAF_DATA_SIZE(root);
1738 return btrfs_item_offset_nr(leaf, nr - 1);
1743 * search for key in the extent_buffer. The items start at offset p,
1744 * and they are item_size apart. There are 'max' items in p.
1746 * the slot in the array is returned via slot, and it points to
1747 * the place where you would insert key if it is not found in
1750 * slot may point to max if the key is bigger than all of the keys
1752 static noinline int generic_bin_search(struct extent_buffer *eb,
1754 int item_size, struct btrfs_key *key,
1761 struct btrfs_disk_key *tmp = NULL;
1762 struct btrfs_disk_key unaligned;
1763 unsigned long offset;
1765 unsigned long map_start = 0;
1766 unsigned long map_len = 0;
1769 while (low < high) {
1770 mid = (low + high) / 2;
1771 offset = p + mid * item_size;
1773 if (!kaddr || offset < map_start ||
1774 (offset + sizeof(struct btrfs_disk_key)) >
1775 map_start + map_len) {
1777 err = map_private_extent_buffer(eb, offset,
1778 sizeof(struct btrfs_disk_key),
1779 &kaddr, &map_start, &map_len);
1782 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1785 read_extent_buffer(eb, &unaligned,
1786 offset, sizeof(unaligned));
1791 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1794 ret = comp_keys(tmp, key);
1810 * simple bin_search frontend that does the right thing for
1813 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1814 int level, int *slot)
1817 return generic_bin_search(eb,
1818 offsetof(struct btrfs_leaf, items),
1819 sizeof(struct btrfs_item),
1820 key, btrfs_header_nritems(eb),
1823 return generic_bin_search(eb,
1824 offsetof(struct btrfs_node, ptrs),
1825 sizeof(struct btrfs_key_ptr),
1826 key, btrfs_header_nritems(eb),
1830 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1831 int level, int *slot)
1833 return bin_search(eb, key, level, slot);
1836 static void root_add_used(struct btrfs_root *root, u32 size)
1838 spin_lock(&root->accounting_lock);
1839 btrfs_set_root_used(&root->root_item,
1840 btrfs_root_used(&root->root_item) + size);
1841 spin_unlock(&root->accounting_lock);
1844 static void root_sub_used(struct btrfs_root *root, u32 size)
1846 spin_lock(&root->accounting_lock);
1847 btrfs_set_root_used(&root->root_item,
1848 btrfs_root_used(&root->root_item) - size);
1849 spin_unlock(&root->accounting_lock);
1852 /* given a node and slot number, this reads the blocks it points to. The
1853 * extent buffer is returned with a reference taken (but unlocked).
1854 * NULL is returned on error.
1856 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1857 struct extent_buffer *parent, int slot)
1859 int level = btrfs_header_level(parent);
1860 struct extent_buffer *eb;
1864 if (slot >= btrfs_header_nritems(parent))
1869 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1870 btrfs_node_ptr_generation(parent, slot));
1871 if (eb && !extent_buffer_uptodate(eb)) {
1872 free_extent_buffer(eb);
1880 * node level balancing, used to make sure nodes are in proper order for
1881 * item deletion. We balance from the top down, so we have to make sure
1882 * that a deletion won't leave an node completely empty later on.
1884 static noinline int balance_level(struct btrfs_trans_handle *trans,
1885 struct btrfs_root *root,
1886 struct btrfs_path *path, int level)
1888 struct extent_buffer *right = NULL;
1889 struct extent_buffer *mid;
1890 struct extent_buffer *left = NULL;
1891 struct extent_buffer *parent = NULL;
1895 int orig_slot = path->slots[level];
1901 mid = path->nodes[level];
1903 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1904 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1905 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1907 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1909 if (level < BTRFS_MAX_LEVEL - 1) {
1910 parent = path->nodes[level + 1];
1911 pslot = path->slots[level + 1];
1915 * deal with the case where there is only one pointer in the root
1916 * by promoting the node below to a root
1919 struct extent_buffer *child;
1921 if (btrfs_header_nritems(mid) != 1)
1924 /* promote the child to a root */
1925 child = read_node_slot(root, mid, 0);
1928 btrfs_std_error(root->fs_info, ret);
1932 btrfs_tree_lock(child);
1933 btrfs_set_lock_blocking(child);
1934 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1936 btrfs_tree_unlock(child);
1937 free_extent_buffer(child);
1941 tree_mod_log_set_root_pointer(root, child, 1);
1942 rcu_assign_pointer(root->node, child);
1944 add_root_to_dirty_list(root);
1945 btrfs_tree_unlock(child);
1947 path->locks[level] = 0;
1948 path->nodes[level] = NULL;
1949 clean_tree_block(trans, root, mid);
1950 btrfs_tree_unlock(mid);
1951 /* once for the path */
1952 free_extent_buffer(mid);
1954 root_sub_used(root, mid->len);
1955 btrfs_free_tree_block(trans, root, mid, 0, 1);
1956 /* once for the root ptr */
1957 free_extent_buffer_stale(mid);
1960 if (btrfs_header_nritems(mid) >
1961 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1964 left = read_node_slot(root, parent, pslot - 1);
1966 btrfs_tree_lock(left);
1967 btrfs_set_lock_blocking(left);
1968 wret = btrfs_cow_block(trans, root, left,
1969 parent, pslot - 1, &left);
1975 right = read_node_slot(root, parent, pslot + 1);
1977 btrfs_tree_lock(right);
1978 btrfs_set_lock_blocking(right);
1979 wret = btrfs_cow_block(trans, root, right,
1980 parent, pslot + 1, &right);
1987 /* first, try to make some room in the middle buffer */
1989 orig_slot += btrfs_header_nritems(left);
1990 wret = push_node_left(trans, root, left, mid, 1);
1996 * then try to empty the right most buffer into the middle
1999 wret = push_node_left(trans, root, mid, right, 1);
2000 if (wret < 0 && wret != -ENOSPC)
2002 if (btrfs_header_nritems(right) == 0) {
2003 clean_tree_block(trans, root, right);
2004 btrfs_tree_unlock(right);
2005 del_ptr(root, path, level + 1, pslot + 1);
2006 root_sub_used(root, right->len);
2007 btrfs_free_tree_block(trans, root, right, 0, 1);
2008 free_extent_buffer_stale(right);
2011 struct btrfs_disk_key right_key;
2012 btrfs_node_key(right, &right_key, 0);
2013 tree_mod_log_set_node_key(root->fs_info, parent,
2015 btrfs_set_node_key(parent, &right_key, pslot + 1);
2016 btrfs_mark_buffer_dirty(parent);
2019 if (btrfs_header_nritems(mid) == 1) {
2021 * we're not allowed to leave a node with one item in the
2022 * tree during a delete. A deletion from lower in the tree
2023 * could try to delete the only pointer in this node.
2024 * So, pull some keys from the left.
2025 * There has to be a left pointer at this point because
2026 * otherwise we would have pulled some pointers from the
2031 btrfs_std_error(root->fs_info, ret);
2034 wret = balance_node_right(trans, root, mid, left);
2040 wret = push_node_left(trans, root, left, mid, 1);
2046 if (btrfs_header_nritems(mid) == 0) {
2047 clean_tree_block(trans, root, mid);
2048 btrfs_tree_unlock(mid);
2049 del_ptr(root, path, level + 1, pslot);
2050 root_sub_used(root, mid->len);
2051 btrfs_free_tree_block(trans, root, mid, 0, 1);
2052 free_extent_buffer_stale(mid);
2055 /* update the parent key to reflect our changes */
2056 struct btrfs_disk_key mid_key;
2057 btrfs_node_key(mid, &mid_key, 0);
2058 tree_mod_log_set_node_key(root->fs_info, parent,
2060 btrfs_set_node_key(parent, &mid_key, pslot);
2061 btrfs_mark_buffer_dirty(parent);
2064 /* update the path */
2066 if (btrfs_header_nritems(left) > orig_slot) {
2067 extent_buffer_get(left);
2068 /* left was locked after cow */
2069 path->nodes[level] = left;
2070 path->slots[level + 1] -= 1;
2071 path->slots[level] = orig_slot;
2073 btrfs_tree_unlock(mid);
2074 free_extent_buffer(mid);
2077 orig_slot -= btrfs_header_nritems(left);
2078 path->slots[level] = orig_slot;
2081 /* double check we haven't messed things up */
2083 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2087 btrfs_tree_unlock(right);
2088 free_extent_buffer(right);
2091 if (path->nodes[level] != left)
2092 btrfs_tree_unlock(left);
2093 free_extent_buffer(left);
2098 /* Node balancing for insertion. Here we only split or push nodes around
2099 * when they are completely full. This is also done top down, so we
2100 * have to be pessimistic.
2102 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2103 struct btrfs_root *root,
2104 struct btrfs_path *path, int level)
2106 struct extent_buffer *right = NULL;
2107 struct extent_buffer *mid;
2108 struct extent_buffer *left = NULL;
2109 struct extent_buffer *parent = NULL;
2113 int orig_slot = path->slots[level];
2118 mid = path->nodes[level];
2119 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2121 if (level < BTRFS_MAX_LEVEL - 1) {
2122 parent = path->nodes[level + 1];
2123 pslot = path->slots[level + 1];
2129 left = read_node_slot(root, parent, pslot - 1);
2131 /* first, try to make some room in the middle buffer */
2135 btrfs_tree_lock(left);
2136 btrfs_set_lock_blocking(left);
2138 left_nr = btrfs_header_nritems(left);
2139 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2142 ret = btrfs_cow_block(trans, root, left, parent,
2147 wret = push_node_left(trans, root,
2154 struct btrfs_disk_key disk_key;
2155 orig_slot += left_nr;
2156 btrfs_node_key(mid, &disk_key, 0);
2157 tree_mod_log_set_node_key(root->fs_info, parent,
2159 btrfs_set_node_key(parent, &disk_key, pslot);
2160 btrfs_mark_buffer_dirty(parent);
2161 if (btrfs_header_nritems(left) > orig_slot) {
2162 path->nodes[level] = left;
2163 path->slots[level + 1] -= 1;
2164 path->slots[level] = orig_slot;
2165 btrfs_tree_unlock(mid);
2166 free_extent_buffer(mid);
2169 btrfs_header_nritems(left);
2170 path->slots[level] = orig_slot;
2171 btrfs_tree_unlock(left);
2172 free_extent_buffer(left);
2176 btrfs_tree_unlock(left);
2177 free_extent_buffer(left);
2179 right = read_node_slot(root, parent, pslot + 1);
2182 * then try to empty the right most buffer into the middle
2187 btrfs_tree_lock(right);
2188 btrfs_set_lock_blocking(right);
2190 right_nr = btrfs_header_nritems(right);
2191 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2194 ret = btrfs_cow_block(trans, root, right,
2200 wret = balance_node_right(trans, root,
2207 struct btrfs_disk_key disk_key;
2209 btrfs_node_key(right, &disk_key, 0);
2210 tree_mod_log_set_node_key(root->fs_info, parent,
2212 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2213 btrfs_mark_buffer_dirty(parent);
2215 if (btrfs_header_nritems(mid) <= orig_slot) {
2216 path->nodes[level] = right;
2217 path->slots[level + 1] += 1;
2218 path->slots[level] = orig_slot -
2219 btrfs_header_nritems(mid);
2220 btrfs_tree_unlock(mid);
2221 free_extent_buffer(mid);
2223 btrfs_tree_unlock(right);
2224 free_extent_buffer(right);
2228 btrfs_tree_unlock(right);
2229 free_extent_buffer(right);
2235 * readahead one full node of leaves, finding things that are close
2236 * to the block in 'slot', and triggering ra on them.
2238 static void reada_for_search(struct btrfs_root *root,
2239 struct btrfs_path *path,
2240 int level, int slot, u64 objectid)
2242 struct extent_buffer *node;
2243 struct btrfs_disk_key disk_key;
2249 int direction = path->reada;
2250 struct extent_buffer *eb;
2258 if (!path->nodes[level])
2261 node = path->nodes[level];
2263 search = btrfs_node_blockptr(node, slot);
2264 blocksize = root->nodesize;
2265 eb = btrfs_find_tree_block(root, search);
2267 free_extent_buffer(eb);
2273 nritems = btrfs_header_nritems(node);
2277 if (direction < 0) {
2281 } else if (direction > 0) {
2286 if (path->reada < 0 && objectid) {
2287 btrfs_node_key(node, &disk_key, nr);
2288 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2291 search = btrfs_node_blockptr(node, nr);
2292 if ((search <= target && target - search <= 65536) ||
2293 (search > target && search - target <= 65536)) {
2294 gen = btrfs_node_ptr_generation(node, nr);
2295 readahead_tree_block(root, search, blocksize);
2299 if ((nread > 65536 || nscan > 32))
2304 static noinline void reada_for_balance(struct btrfs_root *root,
2305 struct btrfs_path *path, int level)
2309 struct extent_buffer *parent;
2310 struct extent_buffer *eb;
2316 parent = path->nodes[level + 1];
2320 nritems = btrfs_header_nritems(parent);
2321 slot = path->slots[level + 1];
2322 blocksize = root->nodesize;
2325 block1 = btrfs_node_blockptr(parent, slot - 1);
2326 gen = btrfs_node_ptr_generation(parent, slot - 1);
2327 eb = btrfs_find_tree_block(root, block1);
2329 * if we get -eagain from btrfs_buffer_uptodate, we
2330 * don't want to return eagain here. That will loop
2333 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2335 free_extent_buffer(eb);
2337 if (slot + 1 < nritems) {
2338 block2 = btrfs_node_blockptr(parent, slot + 1);
2339 gen = btrfs_node_ptr_generation(parent, slot + 1);
2340 eb = btrfs_find_tree_block(root, block2);
2341 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2343 free_extent_buffer(eb);
2347 readahead_tree_block(root, block1, blocksize);
2349 readahead_tree_block(root, block2, blocksize);
2354 * when we walk down the tree, it is usually safe to unlock the higher layers
2355 * in the tree. The exceptions are when our path goes through slot 0, because
2356 * operations on the tree might require changing key pointers higher up in the
2359 * callers might also have set path->keep_locks, which tells this code to keep
2360 * the lock if the path points to the last slot in the block. This is part of
2361 * walking through the tree, and selecting the next slot in the higher block.
2363 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2364 * if lowest_unlock is 1, level 0 won't be unlocked
2366 static noinline void unlock_up(struct btrfs_path *path, int level,
2367 int lowest_unlock, int min_write_lock_level,
2368 int *write_lock_level)
2371 int skip_level = level;
2373 struct extent_buffer *t;
2375 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2376 if (!path->nodes[i])
2378 if (!path->locks[i])
2380 if (!no_skips && path->slots[i] == 0) {
2384 if (!no_skips && path->keep_locks) {
2387 nritems = btrfs_header_nritems(t);
2388 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2393 if (skip_level < i && i >= lowest_unlock)
2397 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2398 btrfs_tree_unlock_rw(t, path->locks[i]);
2400 if (write_lock_level &&
2401 i > min_write_lock_level &&
2402 i <= *write_lock_level) {
2403 *write_lock_level = i - 1;
2410 * This releases any locks held in the path starting at level and
2411 * going all the way up to the root.
2413 * btrfs_search_slot will keep the lock held on higher nodes in a few
2414 * corner cases, such as COW of the block at slot zero in the node. This
2415 * ignores those rules, and it should only be called when there are no
2416 * more updates to be done higher up in the tree.
2418 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2422 if (path->keep_locks)
2425 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2426 if (!path->nodes[i])
2428 if (!path->locks[i])
2430 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2436 * helper function for btrfs_search_slot. The goal is to find a block
2437 * in cache without setting the path to blocking. If we find the block
2438 * we return zero and the path is unchanged.
2440 * If we can't find the block, we set the path blocking and do some
2441 * reada. -EAGAIN is returned and the search must be repeated.
2444 read_block_for_search(struct btrfs_trans_handle *trans,
2445 struct btrfs_root *root, struct btrfs_path *p,
2446 struct extent_buffer **eb_ret, int level, int slot,
2447 struct btrfs_key *key, u64 time_seq)
2451 struct extent_buffer *b = *eb_ret;
2452 struct extent_buffer *tmp;
2455 blocknr = btrfs_node_blockptr(b, slot);
2456 gen = btrfs_node_ptr_generation(b, slot);
2458 tmp = btrfs_find_tree_block(root, blocknr);
2460 /* first we do an atomic uptodate check */
2461 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2466 /* the pages were up to date, but we failed
2467 * the generation number check. Do a full
2468 * read for the generation number that is correct.
2469 * We must do this without dropping locks so
2470 * we can trust our generation number
2472 btrfs_set_path_blocking(p);
2474 /* now we're allowed to do a blocking uptodate check */
2475 ret = btrfs_read_buffer(tmp, gen);
2480 free_extent_buffer(tmp);
2481 btrfs_release_path(p);
2486 * reduce lock contention at high levels
2487 * of the btree by dropping locks before
2488 * we read. Don't release the lock on the current
2489 * level because we need to walk this node to figure
2490 * out which blocks to read.
2492 btrfs_unlock_up_safe(p, level + 1);
2493 btrfs_set_path_blocking(p);
2495 free_extent_buffer(tmp);
2497 reada_for_search(root, p, level, slot, key->objectid);
2499 btrfs_release_path(p);
2502 tmp = read_tree_block(root, blocknr, 0);
2505 * If the read above didn't mark this buffer up to date,
2506 * it will never end up being up to date. Set ret to EIO now
2507 * and give up so that our caller doesn't loop forever
2510 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2512 free_extent_buffer(tmp);
2518 * helper function for btrfs_search_slot. This does all of the checks
2519 * for node-level blocks and does any balancing required based on
2522 * If no extra work was required, zero is returned. If we had to
2523 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2527 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2528 struct btrfs_root *root, struct btrfs_path *p,
2529 struct extent_buffer *b, int level, int ins_len,
2530 int *write_lock_level)
2533 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2534 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2537 if (*write_lock_level < level + 1) {
2538 *write_lock_level = level + 1;
2539 btrfs_release_path(p);
2543 btrfs_set_path_blocking(p);
2544 reada_for_balance(root, p, level);
2545 sret = split_node(trans, root, p, level);
2546 btrfs_clear_path_blocking(p, NULL, 0);
2553 b = p->nodes[level];
2554 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2555 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2558 if (*write_lock_level < level + 1) {
2559 *write_lock_level = level + 1;
2560 btrfs_release_path(p);
2564 btrfs_set_path_blocking(p);
2565 reada_for_balance(root, p, level);
2566 sret = balance_level(trans, root, p, level);
2567 btrfs_clear_path_blocking(p, NULL, 0);
2573 b = p->nodes[level];
2575 btrfs_release_path(p);
2578 BUG_ON(btrfs_header_nritems(b) == 1);
2588 static void key_search_validate(struct extent_buffer *b,
2589 struct btrfs_key *key,
2592 #ifdef CONFIG_BTRFS_ASSERT
2593 struct btrfs_disk_key disk_key;
2595 btrfs_cpu_key_to_disk(&disk_key, key);
2598 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2599 offsetof(struct btrfs_leaf, items[0].key),
2602 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2603 offsetof(struct btrfs_node, ptrs[0].key),
2608 static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2609 int level, int *prev_cmp, int *slot)
2611 if (*prev_cmp != 0) {
2612 *prev_cmp = bin_search(b, key, level, slot);
2616 key_search_validate(b, key, level);
2622 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path,
2623 u64 iobjectid, u64 ioff, u8 key_type,
2624 struct btrfs_key *found_key)
2627 struct btrfs_key key;
2628 struct extent_buffer *eb;
2629 struct btrfs_path *path;
2631 key.type = key_type;
2632 key.objectid = iobjectid;
2635 if (found_path == NULL) {
2636 path = btrfs_alloc_path();
2642 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2643 if ((ret < 0) || (found_key == NULL)) {
2644 if (path != found_path)
2645 btrfs_free_path(path);
2649 eb = path->nodes[0];
2650 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2651 ret = btrfs_next_leaf(fs_root, path);
2654 eb = path->nodes[0];
2657 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2658 if (found_key->type != key.type ||
2659 found_key->objectid != key.objectid)
2666 * look for key in the tree. path is filled in with nodes along the way
2667 * if key is found, we return zero and you can find the item in the leaf
2668 * level of the path (level 0)
2670 * If the key isn't found, the path points to the slot where it should
2671 * be inserted, and 1 is returned. If there are other errors during the
2672 * search a negative error number is returned.
2674 * if ins_len > 0, nodes and leaves will be split as we walk down the
2675 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2678 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2679 *root, struct btrfs_key *key, struct btrfs_path *p, int
2682 struct extent_buffer *b;
2687 int lowest_unlock = 1;
2689 /* everything at write_lock_level or lower must be write locked */
2690 int write_lock_level = 0;
2691 u8 lowest_level = 0;
2692 int min_write_lock_level;
2695 lowest_level = p->lowest_level;
2696 WARN_ON(lowest_level && ins_len > 0);
2697 WARN_ON(p->nodes[0] != NULL);
2698 BUG_ON(!cow && ins_len);
2703 /* when we are removing items, we might have to go up to level
2704 * two as we update tree pointers Make sure we keep write
2705 * for those levels as well
2707 write_lock_level = 2;
2708 } else if (ins_len > 0) {
2710 * for inserting items, make sure we have a write lock on
2711 * level 1 so we can update keys
2713 write_lock_level = 1;
2717 write_lock_level = -1;
2719 if (cow && (p->keep_locks || p->lowest_level))
2720 write_lock_level = BTRFS_MAX_LEVEL;
2722 min_write_lock_level = write_lock_level;
2727 * we try very hard to do read locks on the root
2729 root_lock = BTRFS_READ_LOCK;
2731 if (p->search_commit_root) {
2733 * the commit roots are read only
2734 * so we always do read locks
2736 if (p->need_commit_sem)
2737 down_read(&root->fs_info->commit_root_sem);
2738 b = root->commit_root;
2739 extent_buffer_get(b);
2740 level = btrfs_header_level(b);
2741 if (p->need_commit_sem)
2742 up_read(&root->fs_info->commit_root_sem);
2743 if (!p->skip_locking)
2744 btrfs_tree_read_lock(b);
2746 if (p->skip_locking) {
2747 b = btrfs_root_node(root);
2748 level = btrfs_header_level(b);
2750 /* we don't know the level of the root node
2751 * until we actually have it read locked
2753 b = btrfs_read_lock_root_node(root);
2754 level = btrfs_header_level(b);
2755 if (level <= write_lock_level) {
2756 /* whoops, must trade for write lock */
2757 btrfs_tree_read_unlock(b);
2758 free_extent_buffer(b);
2759 b = btrfs_lock_root_node(root);
2760 root_lock = BTRFS_WRITE_LOCK;
2762 /* the level might have changed, check again */
2763 level = btrfs_header_level(b);
2767 p->nodes[level] = b;
2768 if (!p->skip_locking)
2769 p->locks[level] = root_lock;
2772 level = btrfs_header_level(b);
2775 * setup the path here so we can release it under lock
2776 * contention with the cow code
2780 * if we don't really need to cow this block
2781 * then we don't want to set the path blocking,
2782 * so we test it here
2784 if (!should_cow_block(trans, root, b))
2788 * must have write locks on this node and the
2791 if (level > write_lock_level ||
2792 (level + 1 > write_lock_level &&
2793 level + 1 < BTRFS_MAX_LEVEL &&
2794 p->nodes[level + 1])) {
2795 write_lock_level = level + 1;
2796 btrfs_release_path(p);
2800 btrfs_set_path_blocking(p);
2801 err = btrfs_cow_block(trans, root, b,
2802 p->nodes[level + 1],
2803 p->slots[level + 1], &b);
2810 p->nodes[level] = b;
2811 btrfs_clear_path_blocking(p, NULL, 0);
2814 * we have a lock on b and as long as we aren't changing
2815 * the tree, there is no way to for the items in b to change.
2816 * It is safe to drop the lock on our parent before we
2817 * go through the expensive btree search on b.
2819 * If we're inserting or deleting (ins_len != 0), then we might
2820 * be changing slot zero, which may require changing the parent.
2821 * So, we can't drop the lock until after we know which slot
2822 * we're operating on.
2824 if (!ins_len && !p->keep_locks) {
2827 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2828 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2833 ret = key_search(b, key, level, &prev_cmp, &slot);
2837 if (ret && slot > 0) {
2841 p->slots[level] = slot;
2842 err = setup_nodes_for_search(trans, root, p, b, level,
2843 ins_len, &write_lock_level);
2850 b = p->nodes[level];
2851 slot = p->slots[level];
2854 * slot 0 is special, if we change the key
2855 * we have to update the parent pointer
2856 * which means we must have a write lock
2859 if (slot == 0 && ins_len &&
2860 write_lock_level < level + 1) {
2861 write_lock_level = level + 1;
2862 btrfs_release_path(p);
2866 unlock_up(p, level, lowest_unlock,
2867 min_write_lock_level, &write_lock_level);
2869 if (level == lowest_level) {
2875 err = read_block_for_search(trans, root, p,
2876 &b, level, slot, key, 0);
2884 if (!p->skip_locking) {
2885 level = btrfs_header_level(b);
2886 if (level <= write_lock_level) {
2887 err = btrfs_try_tree_write_lock(b);
2889 btrfs_set_path_blocking(p);
2891 btrfs_clear_path_blocking(p, b,
2894 p->locks[level] = BTRFS_WRITE_LOCK;
2896 err = btrfs_try_tree_read_lock(b);
2898 btrfs_set_path_blocking(p);
2899 btrfs_tree_read_lock(b);
2900 btrfs_clear_path_blocking(p, b,
2903 p->locks[level] = BTRFS_READ_LOCK;
2905 p->nodes[level] = b;
2908 p->slots[level] = slot;
2910 btrfs_leaf_free_space(root, b) < ins_len) {
2911 if (write_lock_level < 1) {
2912 write_lock_level = 1;
2913 btrfs_release_path(p);
2917 btrfs_set_path_blocking(p);
2918 err = split_leaf(trans, root, key,
2919 p, ins_len, ret == 0);
2920 btrfs_clear_path_blocking(p, NULL, 0);
2928 if (!p->search_for_split)
2929 unlock_up(p, level, lowest_unlock,
2930 min_write_lock_level, &write_lock_level);
2937 * we don't really know what they plan on doing with the path
2938 * from here on, so for now just mark it as blocking
2940 if (!p->leave_spinning)
2941 btrfs_set_path_blocking(p);
2943 btrfs_release_path(p);
2948 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2949 * current state of the tree together with the operations recorded in the tree
2950 * modification log to search for the key in a previous version of this tree, as
2951 * denoted by the time_seq parameter.
2953 * Naturally, there is no support for insert, delete or cow operations.
2955 * The resulting path and return value will be set up as if we called
2956 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2958 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2959 struct btrfs_path *p, u64 time_seq)
2961 struct extent_buffer *b;
2966 int lowest_unlock = 1;
2967 u8 lowest_level = 0;
2970 lowest_level = p->lowest_level;
2971 WARN_ON(p->nodes[0] != NULL);
2973 if (p->search_commit_root) {
2975 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2979 b = get_old_root(root, time_seq);
2980 level = btrfs_header_level(b);
2981 p->locks[level] = BTRFS_READ_LOCK;
2984 level = btrfs_header_level(b);
2985 p->nodes[level] = b;
2986 btrfs_clear_path_blocking(p, NULL, 0);
2989 * we have a lock on b and as long as we aren't changing
2990 * the tree, there is no way to for the items in b to change.
2991 * It is safe to drop the lock on our parent before we
2992 * go through the expensive btree search on b.
2994 btrfs_unlock_up_safe(p, level + 1);
2997 * Since we can unwind eb's we want to do a real search every
3001 ret = key_search(b, key, level, &prev_cmp, &slot);
3005 if (ret && slot > 0) {
3009 p->slots[level] = slot;
3010 unlock_up(p, level, lowest_unlock, 0, NULL);
3012 if (level == lowest_level) {
3018 err = read_block_for_search(NULL, root, p, &b, level,
3019 slot, key, time_seq);
3027 level = btrfs_header_level(b);
3028 err = btrfs_try_tree_read_lock(b);
3030 btrfs_set_path_blocking(p);
3031 btrfs_tree_read_lock(b);
3032 btrfs_clear_path_blocking(p, b,
3035 b = tree_mod_log_rewind(root->fs_info, p, b, time_seq);
3040 p->locks[level] = BTRFS_READ_LOCK;
3041 p->nodes[level] = b;
3043 p->slots[level] = slot;
3044 unlock_up(p, level, lowest_unlock, 0, NULL);
3050 if (!p->leave_spinning)
3051 btrfs_set_path_blocking(p);
3053 btrfs_release_path(p);
3059 * helper to use instead of search slot if no exact match is needed but
3060 * instead the next or previous item should be returned.
3061 * When find_higher is true, the next higher item is returned, the next lower
3063 * When return_any and find_higher are both true, and no higher item is found,
3064 * return the next lower instead.
3065 * When return_any is true and find_higher is false, and no lower item is found,
3066 * return the next higher instead.
3067 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3070 int btrfs_search_slot_for_read(struct btrfs_root *root,
3071 struct btrfs_key *key, struct btrfs_path *p,
3072 int find_higher, int return_any)
3075 struct extent_buffer *leaf;
3078 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3082 * a return value of 1 means the path is at the position where the
3083 * item should be inserted. Normally this is the next bigger item,
3084 * but in case the previous item is the last in a leaf, path points
3085 * to the first free slot in the previous leaf, i.e. at an invalid
3091 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3092 ret = btrfs_next_leaf(root, p);
3098 * no higher item found, return the next
3103 btrfs_release_path(p);
3107 if (p->slots[0] == 0) {
3108 ret = btrfs_prev_leaf(root, p);
3113 if (p->slots[0] == btrfs_header_nritems(leaf))
3120 * no lower item found, return the next
3125 btrfs_release_path(p);
3135 * adjust the pointers going up the tree, starting at level
3136 * making sure the right key of each node is points to 'key'.
3137 * This is used after shifting pointers to the left, so it stops
3138 * fixing up pointers when a given leaf/node is not in slot 0 of the
3142 static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
3143 struct btrfs_disk_key *key, int level)
3146 struct extent_buffer *t;
3148 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3149 int tslot = path->slots[i];
3150 if (!path->nodes[i])
3153 tree_mod_log_set_node_key(root->fs_info, t, tslot, 1);
3154 btrfs_set_node_key(t, key, tslot);
3155 btrfs_mark_buffer_dirty(path->nodes[i]);
3164 * This function isn't completely safe. It's the caller's responsibility
3165 * that the new key won't break the order
3167 void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
3168 struct btrfs_key *new_key)
3170 struct btrfs_disk_key disk_key;
3171 struct extent_buffer *eb;
3174 eb = path->nodes[0];
3175 slot = path->slots[0];
3177 btrfs_item_key(eb, &disk_key, slot - 1);
3178 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3180 if (slot < btrfs_header_nritems(eb) - 1) {
3181 btrfs_item_key(eb, &disk_key, slot + 1);
3182 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3185 btrfs_cpu_key_to_disk(&disk_key, new_key);
3186 btrfs_set_item_key(eb, &disk_key, slot);
3187 btrfs_mark_buffer_dirty(eb);
3189 fixup_low_keys(root, path, &disk_key, 1);
3193 * try to push data from one node into the next node left in the
3196 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3197 * error, and > 0 if there was no room in the left hand block.
3199 static int push_node_left(struct btrfs_trans_handle *trans,
3200 struct btrfs_root *root, struct extent_buffer *dst,
3201 struct extent_buffer *src, int empty)
3208 src_nritems = btrfs_header_nritems(src);
3209 dst_nritems = btrfs_header_nritems(dst);
3210 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3211 WARN_ON(btrfs_header_generation(src) != trans->transid);
3212 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3214 if (!empty && src_nritems <= 8)
3217 if (push_items <= 0)
3221 push_items = min(src_nritems, push_items);
3222 if (push_items < src_nritems) {
3223 /* leave at least 8 pointers in the node if
3224 * we aren't going to empty it
3226 if (src_nritems - push_items < 8) {
3227 if (push_items <= 8)
3233 push_items = min(src_nritems - 8, push_items);
3235 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3238 btrfs_abort_transaction(trans, root, ret);
3241 copy_extent_buffer(dst, src,
3242 btrfs_node_key_ptr_offset(dst_nritems),
3243 btrfs_node_key_ptr_offset(0),
3244 push_items * sizeof(struct btrfs_key_ptr));
3246 if (push_items < src_nritems) {
3248 * don't call tree_mod_log_eb_move here, key removal was already
3249 * fully logged by tree_mod_log_eb_copy above.
3251 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3252 btrfs_node_key_ptr_offset(push_items),
3253 (src_nritems - push_items) *
3254 sizeof(struct btrfs_key_ptr));
3256 btrfs_set_header_nritems(src, src_nritems - push_items);
3257 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3258 btrfs_mark_buffer_dirty(src);
3259 btrfs_mark_buffer_dirty(dst);
3265 * try to push data from one node into the next node right in the
3268 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3269 * error, and > 0 if there was no room in the right hand block.
3271 * this will only push up to 1/2 the contents of the left node over
3273 static int balance_node_right(struct btrfs_trans_handle *trans,
3274 struct btrfs_root *root,
3275 struct extent_buffer *dst,
3276 struct extent_buffer *src)
3284 WARN_ON(btrfs_header_generation(src) != trans->transid);
3285 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3287 src_nritems = btrfs_header_nritems(src);
3288 dst_nritems = btrfs_header_nritems(dst);
3289 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3290 if (push_items <= 0)
3293 if (src_nritems < 4)
3296 max_push = src_nritems / 2 + 1;
3297 /* don't try to empty the node */
3298 if (max_push >= src_nritems)
3301 if (max_push < push_items)
3302 push_items = max_push;
3304 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3305 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3306 btrfs_node_key_ptr_offset(0),
3308 sizeof(struct btrfs_key_ptr));
3310 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3311 src_nritems - push_items, push_items);
3313 btrfs_abort_transaction(trans, root, ret);
3316 copy_extent_buffer(dst, src,
3317 btrfs_node_key_ptr_offset(0),
3318 btrfs_node_key_ptr_offset(src_nritems - push_items),
3319 push_items * sizeof(struct btrfs_key_ptr));
3321 btrfs_set_header_nritems(src, src_nritems - push_items);
3322 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3324 btrfs_mark_buffer_dirty(src);
3325 btrfs_mark_buffer_dirty(dst);
3331 * helper function to insert a new root level in the tree.
3332 * A new node is allocated, and a single item is inserted to
3333 * point to the existing root
3335 * returns zero on success or < 0 on failure.
3337 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3338 struct btrfs_root *root,
3339 struct btrfs_path *path, int level)
3342 struct extent_buffer *lower;
3343 struct extent_buffer *c;
3344 struct extent_buffer *old;
3345 struct btrfs_disk_key lower_key;
3347 BUG_ON(path->nodes[level]);
3348 BUG_ON(path->nodes[level-1] != root->node);
3350 lower = path->nodes[level-1];
3352 btrfs_item_key(lower, &lower_key, 0);
3354 btrfs_node_key(lower, &lower_key, 0);
3356 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3357 &lower_key, level, root->node->start, 0);
3361 root_add_used(root, root->nodesize);
3363 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3364 btrfs_set_header_nritems(c, 1);
3365 btrfs_set_header_level(c, level);
3366 btrfs_set_header_bytenr(c, c->start);
3367 btrfs_set_header_generation(c, trans->transid);
3368 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3369 btrfs_set_header_owner(c, root->root_key.objectid);
3371 write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(),
3374 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3375 btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE);
3377 btrfs_set_node_key(c, &lower_key, 0);
3378 btrfs_set_node_blockptr(c, 0, lower->start);
3379 lower_gen = btrfs_header_generation(lower);
3380 WARN_ON(lower_gen != trans->transid);
3382 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3384 btrfs_mark_buffer_dirty(c);
3387 tree_mod_log_set_root_pointer(root, c, 0);
3388 rcu_assign_pointer(root->node, c);
3390 /* the super has an extra ref to root->node */
3391 free_extent_buffer(old);
3393 add_root_to_dirty_list(root);
3394 extent_buffer_get(c);
3395 path->nodes[level] = c;
3396 path->locks[level] = BTRFS_WRITE_LOCK;
3397 path->slots[level] = 0;
3402 * worker function to insert a single pointer in a node.
3403 * the node should have enough room for the pointer already
3405 * slot and level indicate where you want the key to go, and
3406 * blocknr is the block the key points to.
3408 static void insert_ptr(struct btrfs_trans_handle *trans,
3409 struct btrfs_root *root, struct btrfs_path *path,
3410 struct btrfs_disk_key *key, u64 bytenr,
3411 int slot, int level)
3413 struct extent_buffer *lower;
3417 BUG_ON(!path->nodes[level]);
3418 btrfs_assert_tree_locked(path->nodes[level]);
3419 lower = path->nodes[level];
3420 nritems = btrfs_header_nritems(lower);
3421 BUG_ON(slot > nritems);
3422 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3423 if (slot != nritems) {
3425 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3426 slot, nritems - slot);
3427 memmove_extent_buffer(lower,
3428 btrfs_node_key_ptr_offset(slot + 1),
3429 btrfs_node_key_ptr_offset(slot),
3430 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3433 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3434 MOD_LOG_KEY_ADD, GFP_NOFS);
3437 btrfs_set_node_key(lower, key, slot);
3438 btrfs_set_node_blockptr(lower, slot, bytenr);
3439 WARN_ON(trans->transid == 0);
3440 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3441 btrfs_set_header_nritems(lower, nritems + 1);
3442 btrfs_mark_buffer_dirty(lower);
3446 * split the node at the specified level in path in two.
3447 * The path is corrected to point to the appropriate node after the split
3449 * Before splitting this tries to make some room in the node by pushing
3450 * left and right, if either one works, it returns right away.
3452 * returns 0 on success and < 0 on failure
3454 static noinline int split_node(struct btrfs_trans_handle *trans,
3455 struct btrfs_root *root,
3456 struct btrfs_path *path, int level)
3458 struct extent_buffer *c;
3459 struct extent_buffer *split;
3460 struct btrfs_disk_key disk_key;
3465 c = path->nodes[level];
3466 WARN_ON(btrfs_header_generation(c) != trans->transid);
3467 if (c == root->node) {
3469 * trying to split the root, lets make a new one
3471 * tree mod log: We don't log_removal old root in
3472 * insert_new_root, because that root buffer will be kept as a
3473 * normal node. We are going to log removal of half of the
3474 * elements below with tree_mod_log_eb_copy. We're holding a
3475 * tree lock on the buffer, which is why we cannot race with
3476 * other tree_mod_log users.
3478 ret = insert_new_root(trans, root, path, level + 1);
3482 ret = push_nodes_for_insert(trans, root, path, level);
3483 c = path->nodes[level];
3484 if (!ret && btrfs_header_nritems(c) <
3485 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3491 c_nritems = btrfs_header_nritems(c);
3492 mid = (c_nritems + 1) / 2;
3493 btrfs_node_key(c, &disk_key, mid);
3495 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3496 &disk_key, level, c->start, 0);
3498 return PTR_ERR(split);
3500 root_add_used(root, root->nodesize);
3502 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3503 btrfs_set_header_level(split, btrfs_header_level(c));
3504 btrfs_set_header_bytenr(split, split->start);
3505 btrfs_set_header_generation(split, trans->transid);
3506 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3507 btrfs_set_header_owner(split, root->root_key.objectid);
3508 write_extent_buffer(split, root->fs_info->fsid,
3509 btrfs_header_fsid(), BTRFS_FSID_SIZE);
3510 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3511 btrfs_header_chunk_tree_uuid(split),
3514 ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0,
3515 mid, c_nritems - mid);
3517 btrfs_abort_transaction(trans, root, ret);
3520 copy_extent_buffer(split, c,
3521 btrfs_node_key_ptr_offset(0),
3522 btrfs_node_key_ptr_offset(mid),
3523 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3524 btrfs_set_header_nritems(split, c_nritems - mid);
3525 btrfs_set_header_nritems(c, mid);
3528 btrfs_mark_buffer_dirty(c);
3529 btrfs_mark_buffer_dirty(split);
3531 insert_ptr(trans, root, path, &disk_key, split->start,
3532 path->slots[level + 1] + 1, level + 1);
3534 if (path->slots[level] >= mid) {
3535 path->slots[level] -= mid;
3536 btrfs_tree_unlock(c);
3537 free_extent_buffer(c);
3538 path->nodes[level] = split;
3539 path->slots[level + 1] += 1;
3541 btrfs_tree_unlock(split);
3542 free_extent_buffer(split);
3548 * how many bytes are required to store the items in a leaf. start
3549 * and nr indicate which items in the leaf to check. This totals up the
3550 * space used both by the item structs and the item data
3552 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3554 struct btrfs_item *start_item;
3555 struct btrfs_item *end_item;
3556 struct btrfs_map_token token;
3558 int nritems = btrfs_header_nritems(l);
3559 int end = min(nritems, start + nr) - 1;
3563 btrfs_init_map_token(&token);
3564 start_item = btrfs_item_nr(start);
3565 end_item = btrfs_item_nr(end);
3566 data_len = btrfs_token_item_offset(l, start_item, &token) +
3567 btrfs_token_item_size(l, start_item, &token);
3568 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3569 data_len += sizeof(struct btrfs_item) * nr;
3570 WARN_ON(data_len < 0);
3575 * The space between the end of the leaf items and
3576 * the start of the leaf data. IOW, how much room
3577 * the leaf has left for both items and data
3579 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3580 struct extent_buffer *leaf)
3582 int nritems = btrfs_header_nritems(leaf);
3584 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3586 btrfs_crit(root->fs_info,
3587 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3588 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3589 leaf_space_used(leaf, 0, nritems), nritems);
3595 * min slot controls the lowest index we're willing to push to the
3596 * right. We'll push up to and including min_slot, but no lower
3598 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3599 struct btrfs_root *root,
3600 struct btrfs_path *path,
3601 int data_size, int empty,
3602 struct extent_buffer *right,
3603 int free_space, u32 left_nritems,
3606 struct extent_buffer *left = path->nodes[0];
3607 struct extent_buffer *upper = path->nodes[1];
3608 struct btrfs_map_token token;
3609 struct btrfs_disk_key disk_key;
3614 struct btrfs_item *item;
3620 btrfs_init_map_token(&token);
3625 nr = max_t(u32, 1, min_slot);
3627 if (path->slots[0] >= left_nritems)
3628 push_space += data_size;
3630 slot = path->slots[1];
3631 i = left_nritems - 1;
3633 item = btrfs_item_nr(i);
3635 if (!empty && push_items > 0) {
3636 if (path->slots[0] > i)
3638 if (path->slots[0] == i) {
3639 int space = btrfs_leaf_free_space(root, left);
3640 if (space + push_space * 2 > free_space)
3645 if (path->slots[0] == i)
3646 push_space += data_size;
3648 this_item_size = btrfs_item_size(left, item);
3649 if (this_item_size + sizeof(*item) + push_space > free_space)
3653 push_space += this_item_size + sizeof(*item);
3659 if (push_items == 0)
3662 WARN_ON(!empty && push_items == left_nritems);
3664 /* push left to right */
3665 right_nritems = btrfs_header_nritems(right);
3667 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3668 push_space -= leaf_data_end(root, left);
3670 /* make room in the right data area */
3671 data_end = leaf_data_end(root, right);
3672 memmove_extent_buffer(right,
3673 btrfs_leaf_data(right) + data_end - push_space,
3674 btrfs_leaf_data(right) + data_end,
3675 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3677 /* copy from the left data area */
3678 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3679 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3680 btrfs_leaf_data(left) + leaf_data_end(root, left),
3683 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3684 btrfs_item_nr_offset(0),
3685 right_nritems * sizeof(struct btrfs_item));
3687 /* copy the items from left to right */
3688 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3689 btrfs_item_nr_offset(left_nritems - push_items),
3690 push_items * sizeof(struct btrfs_item));
3692 /* update the item pointers */
3693 right_nritems += push_items;
3694 btrfs_set_header_nritems(right, right_nritems);
3695 push_space = BTRFS_LEAF_DATA_SIZE(root);
3696 for (i = 0; i < right_nritems; i++) {
3697 item = btrfs_item_nr(i);
3698 push_space -= btrfs_token_item_size(right, item, &token);
3699 btrfs_set_token_item_offset(right, item, push_space, &token);
3702 left_nritems -= push_items;
3703 btrfs_set_header_nritems(left, left_nritems);
3706 btrfs_mark_buffer_dirty(left);
3708 clean_tree_block(trans, root, left);
3710 btrfs_mark_buffer_dirty(right);
3712 btrfs_item_key(right, &disk_key, 0);
3713 btrfs_set_node_key(upper, &disk_key, slot + 1);
3714 btrfs_mark_buffer_dirty(upper);
3716 /* then fixup the leaf pointer in the path */
3717 if (path->slots[0] >= left_nritems) {
3718 path->slots[0] -= left_nritems;
3719 if (btrfs_header_nritems(path->nodes[0]) == 0)
3720 clean_tree_block(trans, root, path->nodes[0]);
3721 btrfs_tree_unlock(path->nodes[0]);
3722 free_extent_buffer(path->nodes[0]);
3723 path->nodes[0] = right;
3724 path->slots[1] += 1;
3726 btrfs_tree_unlock(right);
3727 free_extent_buffer(right);
3732 btrfs_tree_unlock(right);
3733 free_extent_buffer(right);
3738 * push some data in the path leaf to the right, trying to free up at
3739 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3741 * returns 1 if the push failed because the other node didn't have enough
3742 * room, 0 if everything worked out and < 0 if there were major errors.
3744 * this will push starting from min_slot to the end of the leaf. It won't
3745 * push any slot lower than min_slot
3747 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3748 *root, struct btrfs_path *path,
3749 int min_data_size, int data_size,
3750 int empty, u32 min_slot)
3752 struct extent_buffer *left = path->nodes[0];
3753 struct extent_buffer *right;
3754 struct extent_buffer *upper;
3760 if (!path->nodes[1])
3763 slot = path->slots[1];
3764 upper = path->nodes[1];
3765 if (slot >= btrfs_header_nritems(upper) - 1)
3768 btrfs_assert_tree_locked(path->nodes[1]);
3770 right = read_node_slot(root, upper, slot + 1);
3774 btrfs_tree_lock(right);
3775 btrfs_set_lock_blocking(right);
3777 free_space = btrfs_leaf_free_space(root, right);
3778 if (free_space < data_size)
3781 /* cow and double check */
3782 ret = btrfs_cow_block(trans, root, right, upper,
3787 free_space = btrfs_leaf_free_space(root, right);
3788 if (free_space < data_size)
3791 left_nritems = btrfs_header_nritems(left);
3792 if (left_nritems == 0)
3795 if (path->slots[0] == left_nritems && !empty) {
3796 /* Key greater than all keys in the leaf, right neighbor has
3797 * enough room for it and we're not emptying our leaf to delete
3798 * it, therefore use right neighbor to insert the new item and
3799 * no need to touch/dirty our left leaft. */
3800 btrfs_tree_unlock(left);
3801 free_extent_buffer(left);
3802 path->nodes[0] = right;
3808 return __push_leaf_right(trans, root, path, min_data_size, empty,
3809 right, free_space, left_nritems, min_slot);
3811 btrfs_tree_unlock(right);
3812 free_extent_buffer(right);
3817 * push some data in the path leaf to the left, trying to free up at
3818 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3820 * max_slot can put a limit on how far into the leaf we'll push items. The
3821 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3824 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3825 struct btrfs_root *root,
3826 struct btrfs_path *path, int data_size,
3827 int empty, struct extent_buffer *left,
3828 int free_space, u32 right_nritems,
3831 struct btrfs_disk_key disk_key;
3832 struct extent_buffer *right = path->nodes[0];
3836 struct btrfs_item *item;
3837 u32 old_left_nritems;
3841 u32 old_left_item_size;
3842 struct btrfs_map_token token;
3844 btrfs_init_map_token(&token);
3847 nr = min(right_nritems, max_slot);
3849 nr = min(right_nritems - 1, max_slot);
3851 for (i = 0; i < nr; i++) {
3852 item = btrfs_item_nr(i);
3854 if (!empty && push_items > 0) {
3855 if (path->slots[0] < i)
3857 if (path->slots[0] == i) {
3858 int space = btrfs_leaf_free_space(root, right);
3859 if (space + push_space * 2 > free_space)
3864 if (path->slots[0] == i)
3865 push_space += data_size;
3867 this_item_size = btrfs_item_size(right, item);
3868 if (this_item_size + sizeof(*item) + push_space > free_space)
3872 push_space += this_item_size + sizeof(*item);
3875 if (push_items == 0) {
3879 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3881 /* push data from right to left */
3882 copy_extent_buffer(left, right,
3883 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3884 btrfs_item_nr_offset(0),
3885 push_items * sizeof(struct btrfs_item));
3887 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3888 btrfs_item_offset_nr(right, push_items - 1);
3890 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3891 leaf_data_end(root, left) - push_space,
3892 btrfs_leaf_data(right) +
3893 btrfs_item_offset_nr(right, push_items - 1),
3895 old_left_nritems = btrfs_header_nritems(left);
3896 BUG_ON(old_left_nritems <= 0);
3898 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3899 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3902 item = btrfs_item_nr(i);
3904 ioff = btrfs_token_item_offset(left, item, &token);
3905 btrfs_set_token_item_offset(left, item,
3906 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3909 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3911 /* fixup right node */
3912 if (push_items > right_nritems)
3913 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3916 if (push_items < right_nritems) {
3917 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3918 leaf_data_end(root, right);
3919 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3920 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3921 btrfs_leaf_data(right) +
3922 leaf_data_end(root, right), push_space);
3924 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3925 btrfs_item_nr_offset(push_items),
3926 (btrfs_header_nritems(right) - push_items) *
3927 sizeof(struct btrfs_item));
3929 right_nritems -= push_items;
3930 btrfs_set_header_nritems(right, right_nritems);
3931 push_space = BTRFS_LEAF_DATA_SIZE(root);
3932 for (i = 0; i < right_nritems; i++) {
3933 item = btrfs_item_nr(i);
3935 push_space = push_space - btrfs_token_item_size(right,
3937 btrfs_set_token_item_offset(right, item, push_space, &token);
3940 btrfs_mark_buffer_dirty(left);
3942 btrfs_mark_buffer_dirty(right);
3944 clean_tree_block(trans, root, right);
3946 btrfs_item_key(right, &disk_key, 0);
3947 fixup_low_keys(root, path, &disk_key, 1);
3949 /* then fixup the leaf pointer in the path */
3950 if (path->slots[0] < push_items) {
3951 path->slots[0] += old_left_nritems;
3952 btrfs_tree_unlock(path->nodes[0]);
3953 free_extent_buffer(path->nodes[0]);
3954 path->nodes[0] = left;
3955 path->slots[1] -= 1;
3957 btrfs_tree_unlock(left);
3958 free_extent_buffer(left);
3959 path->slots[0] -= push_items;
3961 BUG_ON(path->slots[0] < 0);
3964 btrfs_tree_unlock(left);
3965 free_extent_buffer(left);
3970 * push some data in the path leaf to the left, trying to free up at
3971 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3973 * max_slot can put a limit on how far into the leaf we'll push items. The
3974 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3977 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3978 *root, struct btrfs_path *path, int min_data_size,
3979 int data_size, int empty, u32 max_slot)
3981 struct extent_buffer *right = path->nodes[0];
3982 struct extent_buffer *left;
3988 slot = path->slots[1];
3991 if (!path->nodes[1])
3994 right_nritems = btrfs_header_nritems(right);
3995 if (right_nritems == 0)
3998 btrfs_assert_tree_locked(path->nodes[1]);
4000 left = read_node_slot(root, path->nodes[1], slot - 1);
4004 btrfs_tree_lock(left);
4005 btrfs_set_lock_blocking(left);
4007 free_space = btrfs_leaf_free_space(root, left);
4008 if (free_space < data_size) {
4013 /* cow and double check */
4014 ret = btrfs_cow_block(trans, root, left,
4015 path->nodes[1], slot - 1, &left);
4017 /* we hit -ENOSPC, but it isn't fatal here */
4023 free_space = btrfs_leaf_free_space(root, left);
4024 if (free_space < data_size) {
4029 return __push_leaf_left(trans, root, path, min_data_size,
4030 empty, left, free_space, right_nritems,
4033 btrfs_tree_unlock(left);
4034 free_extent_buffer(left);
4039 * split the path's leaf in two, making sure there is at least data_size
4040 * available for the resulting leaf level of the path.
4042 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4043 struct btrfs_root *root,
4044 struct btrfs_path *path,
4045 struct extent_buffer *l,
4046 struct extent_buffer *right,
4047 int slot, int mid, int nritems)
4052 struct btrfs_disk_key disk_key;
4053 struct btrfs_map_token token;
4055 btrfs_init_map_token(&token);
4057 nritems = nritems - mid;
4058 btrfs_set_header_nritems(right, nritems);
4059 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
4061 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4062 btrfs_item_nr_offset(mid),
4063 nritems * sizeof(struct btrfs_item));
4065 copy_extent_buffer(right, l,
4066 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
4067 data_copy_size, btrfs_leaf_data(l) +
4068 leaf_data_end(root, l), data_copy_size);
4070 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
4071 btrfs_item_end_nr(l, mid);
4073 for (i = 0; i < nritems; i++) {
4074 struct btrfs_item *item = btrfs_item_nr(i);
4077 ioff = btrfs_token_item_offset(right, item, &token);
4078 btrfs_set_token_item_offset(right, item,
4079 ioff + rt_data_off, &token);
4082 btrfs_set_header_nritems(l, mid);
4083 btrfs_item_key(right, &disk_key, 0);
4084 insert_ptr(trans, root, path, &disk_key, right->start,
4085 path->slots[1] + 1, 1);
4087 btrfs_mark_buffer_dirty(right);
4088 btrfs_mark_buffer_dirty(l);
4089 BUG_ON(path->slots[0] != slot);
4092 btrfs_tree_unlock(path->nodes[0]);
4093 free_extent_buffer(path->nodes[0]);
4094 path->nodes[0] = right;
4095 path->slots[0] -= mid;
4096 path->slots[1] += 1;
4098 btrfs_tree_unlock(right);
4099 free_extent_buffer(right);
4102 BUG_ON(path->slots[0] < 0);
4106 * double splits happen when we need to insert a big item in the middle
4107 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4108 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4111 * We avoid this by trying to push the items on either side of our target
4112 * into the adjacent leaves. If all goes well we can avoid the double split
4115 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4116 struct btrfs_root *root,
4117 struct btrfs_path *path,
4124 int space_needed = data_size;
4126 slot = path->slots[0];
4127 if (slot < btrfs_header_nritems(path->nodes[0]))
4128 space_needed -= btrfs_leaf_free_space(root, path->nodes[0]);
4131 * try to push all the items after our slot into the
4134 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4141 nritems = btrfs_header_nritems(path->nodes[0]);
4143 * our goal is to get our slot at the start or end of a leaf. If
4144 * we've done so we're done
4146 if (path->slots[0] == 0 || path->slots[0] == nritems)
4149 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4152 /* try to push all the items before our slot into the next leaf */
4153 slot = path->slots[0];
4154 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4167 * split the path's leaf in two, making sure there is at least data_size
4168 * available for the resulting leaf level of the path.
4170 * returns 0 if all went well and < 0 on failure.
4172 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4173 struct btrfs_root *root,
4174 struct btrfs_key *ins_key,
4175 struct btrfs_path *path, int data_size,
4178 struct btrfs_disk_key disk_key;
4179 struct extent_buffer *l;
4183 struct extent_buffer *right;
4187 int num_doubles = 0;
4188 int tried_avoid_double = 0;
4191 slot = path->slots[0];
4192 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4193 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
4196 /* first try to make some room by pushing left and right */
4197 if (data_size && path->nodes[1]) {
4198 int space_needed = data_size;
4200 if (slot < btrfs_header_nritems(l))
4201 space_needed -= btrfs_leaf_free_space(root, l);
4203 wret = push_leaf_right(trans, root, path, space_needed,
4204 space_needed, 0, 0);
4208 wret = push_leaf_left(trans, root, path, space_needed,
4209 space_needed, 0, (u32)-1);
4215 /* did the pushes work? */
4216 if (btrfs_leaf_free_space(root, l) >= data_size)
4220 if (!path->nodes[1]) {
4221 ret = insert_new_root(trans, root, path, 1);
4228 slot = path->slots[0];
4229 nritems = btrfs_header_nritems(l);
4230 mid = (nritems + 1) / 2;
4234 leaf_space_used(l, mid, nritems - mid) + data_size >
4235 BTRFS_LEAF_DATA_SIZE(root)) {
4236 if (slot >= nritems) {
4240 if (mid != nritems &&
4241 leaf_space_used(l, mid, nritems - mid) +
4242 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4243 if (data_size && !tried_avoid_double)
4244 goto push_for_double;
4250 if (leaf_space_used(l, 0, mid) + data_size >
4251 BTRFS_LEAF_DATA_SIZE(root)) {
4252 if (!extend && data_size && slot == 0) {
4254 } else if ((extend || !data_size) && slot == 0) {
4258 if (mid != nritems &&
4259 leaf_space_used(l, mid, nritems - mid) +
4260 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4261 if (data_size && !tried_avoid_double)
4262 goto push_for_double;
4270 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4272 btrfs_item_key(l, &disk_key, mid);
4274 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4275 &disk_key, 0, l->start, 0);
4277 return PTR_ERR(right);
4279 root_add_used(root, root->nodesize);
4281 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4282 btrfs_set_header_bytenr(right, right->start);
4283 btrfs_set_header_generation(right, trans->transid);
4284 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4285 btrfs_set_header_owner(right, root->root_key.objectid);
4286 btrfs_set_header_level(right, 0);
4287 write_extent_buffer(right, root->fs_info->fsid,
4288 btrfs_header_fsid(), BTRFS_FSID_SIZE);
4290 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
4291 btrfs_header_chunk_tree_uuid(right),
4296 btrfs_set_header_nritems(right, 0);
4297 insert_ptr(trans, root, path, &disk_key, right->start,
4298 path->slots[1] + 1, 1);
4299 btrfs_tree_unlock(path->nodes[0]);
4300 free_extent_buffer(path->nodes[0]);
4301 path->nodes[0] = right;
4303 path->slots[1] += 1;
4305 btrfs_set_header_nritems(right, 0);
4306 insert_ptr(trans, root, path, &disk_key, right->start,
4308 btrfs_tree_unlock(path->nodes[0]);
4309 free_extent_buffer(path->nodes[0]);
4310 path->nodes[0] = right;
4312 if (path->slots[1] == 0)
4313 fixup_low_keys(root, path, &disk_key, 1);
4315 btrfs_mark_buffer_dirty(right);
4319 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4322 BUG_ON(num_doubles != 0);
4330 push_for_double_split(trans, root, path, data_size);
4331 tried_avoid_double = 1;
4332 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4337 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4338 struct btrfs_root *root,
4339 struct btrfs_path *path, int ins_len)
4341 struct btrfs_key key;
4342 struct extent_buffer *leaf;
4343 struct btrfs_file_extent_item *fi;
4348 leaf = path->nodes[0];
4349 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4351 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4352 key.type != BTRFS_EXTENT_CSUM_KEY);
4354 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4357 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4358 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4359 fi = btrfs_item_ptr(leaf, path->slots[0],
4360 struct btrfs_file_extent_item);
4361 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4363 btrfs_release_path(path);
4365 path->keep_locks = 1;
4366 path->search_for_split = 1;
4367 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4368 path->search_for_split = 0;
4373 leaf = path->nodes[0];
4374 /* if our item isn't there or got smaller, return now */
4375 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4378 /* the leaf has changed, it now has room. return now */
4379 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4382 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4383 fi = btrfs_item_ptr(leaf, path->slots[0],
4384 struct btrfs_file_extent_item);
4385 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4389 btrfs_set_path_blocking(path);
4390 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4394 path->keep_locks = 0;
4395 btrfs_unlock_up_safe(path, 1);
4398 path->keep_locks = 0;
4402 static noinline int split_item(struct btrfs_trans_handle *trans,
4403 struct btrfs_root *root,
4404 struct btrfs_path *path,
4405 struct btrfs_key *new_key,
4406 unsigned long split_offset)
4408 struct extent_buffer *leaf;
4409 struct btrfs_item *item;
4410 struct btrfs_item *new_item;
4416 struct btrfs_disk_key disk_key;
4418 leaf = path->nodes[0];
4419 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4421 btrfs_set_path_blocking(path);
4423 item = btrfs_item_nr(path->slots[0]);
4424 orig_offset = btrfs_item_offset(leaf, item);
4425 item_size = btrfs_item_size(leaf, item);
4427 buf = kmalloc(item_size, GFP_NOFS);
4431 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4432 path->slots[0]), item_size);
4434 slot = path->slots[0] + 1;
4435 nritems = btrfs_header_nritems(leaf);
4436 if (slot != nritems) {
4437 /* shift the items */
4438 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4439 btrfs_item_nr_offset(slot),
4440 (nritems - slot) * sizeof(struct btrfs_item));
4443 btrfs_cpu_key_to_disk(&disk_key, new_key);
4444 btrfs_set_item_key(leaf, &disk_key, slot);
4446 new_item = btrfs_item_nr(slot);
4448 btrfs_set_item_offset(leaf, new_item, orig_offset);
4449 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4451 btrfs_set_item_offset(leaf, item,
4452 orig_offset + item_size - split_offset);
4453 btrfs_set_item_size(leaf, item, split_offset);
4455 btrfs_set_header_nritems(leaf, nritems + 1);
4457 /* write the data for the start of the original item */
4458 write_extent_buffer(leaf, buf,
4459 btrfs_item_ptr_offset(leaf, path->slots[0]),
4462 /* write the data for the new item */
4463 write_extent_buffer(leaf, buf + split_offset,
4464 btrfs_item_ptr_offset(leaf, slot),
4465 item_size - split_offset);
4466 btrfs_mark_buffer_dirty(leaf);
4468 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4474 * This function splits a single item into two items,
4475 * giving 'new_key' to the new item and splitting the
4476 * old one at split_offset (from the start of the item).
4478 * The path may be released by this operation. After
4479 * the split, the path is pointing to the old item. The
4480 * new item is going to be in the same node as the old one.
4482 * Note, the item being split must be smaller enough to live alone on
4483 * a tree block with room for one extra struct btrfs_item
4485 * This allows us to split the item in place, keeping a lock on the
4486 * leaf the entire time.
4488 int btrfs_split_item(struct btrfs_trans_handle *trans,
4489 struct btrfs_root *root,
4490 struct btrfs_path *path,
4491 struct btrfs_key *new_key,
4492 unsigned long split_offset)
4495 ret = setup_leaf_for_split(trans, root, path,
4496 sizeof(struct btrfs_item));
4500 ret = split_item(trans, root, path, new_key, split_offset);
4505 * This function duplicate a item, giving 'new_key' to the new item.
4506 * It guarantees both items live in the same tree leaf and the new item
4507 * is contiguous with the original item.
4509 * This allows us to split file extent in place, keeping a lock on the
4510 * leaf the entire time.
4512 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4513 struct btrfs_root *root,
4514 struct btrfs_path *path,
4515 struct btrfs_key *new_key)
4517 struct extent_buffer *leaf;
4521 leaf = path->nodes[0];
4522 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4523 ret = setup_leaf_for_split(trans, root, path,
4524 item_size + sizeof(struct btrfs_item));
4529 setup_items_for_insert(root, path, new_key, &item_size,
4530 item_size, item_size +
4531 sizeof(struct btrfs_item), 1);
4532 leaf = path->nodes[0];
4533 memcpy_extent_buffer(leaf,
4534 btrfs_item_ptr_offset(leaf, path->slots[0]),
4535 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4541 * make the item pointed to by the path smaller. new_size indicates
4542 * how small to make it, and from_end tells us if we just chop bytes
4543 * off the end of the item or if we shift the item to chop bytes off
4546 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4547 u32 new_size, int from_end)
4550 struct extent_buffer *leaf;
4551 struct btrfs_item *item;
4553 unsigned int data_end;
4554 unsigned int old_data_start;
4555 unsigned int old_size;
4556 unsigned int size_diff;
4558 struct btrfs_map_token token;
4560 btrfs_init_map_token(&token);
4562 leaf = path->nodes[0];
4563 slot = path->slots[0];
4565 old_size = btrfs_item_size_nr(leaf, slot);
4566 if (old_size == new_size)
4569 nritems = btrfs_header_nritems(leaf);
4570 data_end = leaf_data_end(root, leaf);
4572 old_data_start = btrfs_item_offset_nr(leaf, slot);
4574 size_diff = old_size - new_size;
4577 BUG_ON(slot >= nritems);
4580 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4582 /* first correct the data pointers */
4583 for (i = slot; i < nritems; i++) {
4585 item = btrfs_item_nr(i);
4587 ioff = btrfs_token_item_offset(leaf, item, &token);
4588 btrfs_set_token_item_offset(leaf, item,
4589 ioff + size_diff, &token);
4592 /* shift the data */
4594 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4595 data_end + size_diff, btrfs_leaf_data(leaf) +
4596 data_end, old_data_start + new_size - data_end);
4598 struct btrfs_disk_key disk_key;
4601 btrfs_item_key(leaf, &disk_key, slot);
4603 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4605 struct btrfs_file_extent_item *fi;
4607 fi = btrfs_item_ptr(leaf, slot,
4608 struct btrfs_file_extent_item);
4609 fi = (struct btrfs_file_extent_item *)(
4610 (unsigned long)fi - size_diff);
4612 if (btrfs_file_extent_type(leaf, fi) ==
4613 BTRFS_FILE_EXTENT_INLINE) {
4614 ptr = btrfs_item_ptr_offset(leaf, slot);
4615 memmove_extent_buffer(leaf, ptr,
4617 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4621 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4622 data_end + size_diff, btrfs_leaf_data(leaf) +
4623 data_end, old_data_start - data_end);
4625 offset = btrfs_disk_key_offset(&disk_key);
4626 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4627 btrfs_set_item_key(leaf, &disk_key, slot);
4629 fixup_low_keys(root, path, &disk_key, 1);
4632 item = btrfs_item_nr(slot);
4633 btrfs_set_item_size(leaf, item, new_size);
4634 btrfs_mark_buffer_dirty(leaf);
4636 if (btrfs_leaf_free_space(root, leaf) < 0) {
4637 btrfs_print_leaf(root, leaf);
4643 * make the item pointed to by the path bigger, data_size is the added size.
4645 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4649 struct extent_buffer *leaf;
4650 struct btrfs_item *item;
4652 unsigned int data_end;
4653 unsigned int old_data;
4654 unsigned int old_size;
4656 struct btrfs_map_token token;
4658 btrfs_init_map_token(&token);
4660 leaf = path->nodes[0];
4662 nritems = btrfs_header_nritems(leaf);
4663 data_end = leaf_data_end(root, leaf);
4665 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4666 btrfs_print_leaf(root, leaf);
4669 slot = path->slots[0];
4670 old_data = btrfs_item_end_nr(leaf, slot);
4673 if (slot >= nritems) {
4674 btrfs_print_leaf(root, leaf);
4675 btrfs_crit(root->fs_info, "slot %d too large, nritems %d",
4681 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4683 /* first correct the data pointers */
4684 for (i = slot; i < nritems; i++) {
4686 item = btrfs_item_nr(i);
4688 ioff = btrfs_token_item_offset(leaf, item, &token);
4689 btrfs_set_token_item_offset(leaf, item,
4690 ioff - data_size, &token);
4693 /* shift the data */
4694 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4695 data_end - data_size, btrfs_leaf_data(leaf) +
4696 data_end, old_data - data_end);
4698 data_end = old_data;
4699 old_size = btrfs_item_size_nr(leaf, slot);
4700 item = btrfs_item_nr(slot);
4701 btrfs_set_item_size(leaf, item, old_size + data_size);
4702 btrfs_mark_buffer_dirty(leaf);
4704 if (btrfs_leaf_free_space(root, leaf) < 0) {
4705 btrfs_print_leaf(root, leaf);
4711 * this is a helper for btrfs_insert_empty_items, the main goal here is
4712 * to save stack depth by doing the bulk of the work in a function
4713 * that doesn't call btrfs_search_slot
4715 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4716 struct btrfs_key *cpu_key, u32 *data_size,
4717 u32 total_data, u32 total_size, int nr)
4719 struct btrfs_item *item;
4722 unsigned int data_end;
4723 struct btrfs_disk_key disk_key;
4724 struct extent_buffer *leaf;
4726 struct btrfs_map_token token;
4728 if (path->slots[0] == 0) {
4729 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4730 fixup_low_keys(root, path, &disk_key, 1);
4732 btrfs_unlock_up_safe(path, 1);
4734 btrfs_init_map_token(&token);
4736 leaf = path->nodes[0];
4737 slot = path->slots[0];
4739 nritems = btrfs_header_nritems(leaf);
4740 data_end = leaf_data_end(root, leaf);
4742 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4743 btrfs_print_leaf(root, leaf);
4744 btrfs_crit(root->fs_info, "not enough freespace need %u have %d",
4745 total_size, btrfs_leaf_free_space(root, leaf));
4749 if (slot != nritems) {
4750 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4752 if (old_data < data_end) {
4753 btrfs_print_leaf(root, leaf);
4754 btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d",
4755 slot, old_data, data_end);
4759 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4761 /* first correct the data pointers */
4762 for (i = slot; i < nritems; i++) {
4765 item = btrfs_item_nr( i);
4766 ioff = btrfs_token_item_offset(leaf, item, &token);
4767 btrfs_set_token_item_offset(leaf, item,
4768 ioff - total_data, &token);
4770 /* shift the items */
4771 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4772 btrfs_item_nr_offset(slot),
4773 (nritems - slot) * sizeof(struct btrfs_item));
4775 /* shift the data */
4776 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4777 data_end - total_data, btrfs_leaf_data(leaf) +
4778 data_end, old_data - data_end);
4779 data_end = old_data;
4782 /* setup the item for the new data */
4783 for (i = 0; i < nr; i++) {
4784 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4785 btrfs_set_item_key(leaf, &disk_key, slot + i);
4786 item = btrfs_item_nr(slot + i);
4787 btrfs_set_token_item_offset(leaf, item,
4788 data_end - data_size[i], &token);
4789 data_end -= data_size[i];
4790 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4793 btrfs_set_header_nritems(leaf, nritems + nr);
4794 btrfs_mark_buffer_dirty(leaf);
4796 if (btrfs_leaf_free_space(root, leaf) < 0) {
4797 btrfs_print_leaf(root, leaf);
4803 * Given a key and some data, insert items into the tree.
4804 * This does all the path init required, making room in the tree if needed.
4806 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4807 struct btrfs_root *root,
4808 struct btrfs_path *path,
4809 struct btrfs_key *cpu_key, u32 *data_size,
4818 for (i = 0; i < nr; i++)
4819 total_data += data_size[i];
4821 total_size = total_data + (nr * sizeof(struct btrfs_item));
4822 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4828 slot = path->slots[0];
4831 setup_items_for_insert(root, path, cpu_key, data_size,
4832 total_data, total_size, nr);
4837 * Given a key and some data, insert an item into the tree.
4838 * This does all the path init required, making room in the tree if needed.
4840 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4841 *root, struct btrfs_key *cpu_key, void *data, u32
4845 struct btrfs_path *path;
4846 struct extent_buffer *leaf;
4849 path = btrfs_alloc_path();
4852 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4854 leaf = path->nodes[0];
4855 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4856 write_extent_buffer(leaf, data, ptr, data_size);
4857 btrfs_mark_buffer_dirty(leaf);
4859 btrfs_free_path(path);
4864 * delete the pointer from a given node.
4866 * the tree should have been previously balanced so the deletion does not
4869 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4870 int level, int slot)
4872 struct extent_buffer *parent = path->nodes[level];
4876 nritems = btrfs_header_nritems(parent);
4877 if (slot != nritems - 1) {
4879 tree_mod_log_eb_move(root->fs_info, parent, slot,
4880 slot + 1, nritems - slot - 1);
4881 memmove_extent_buffer(parent,
4882 btrfs_node_key_ptr_offset(slot),
4883 btrfs_node_key_ptr_offset(slot + 1),
4884 sizeof(struct btrfs_key_ptr) *
4885 (nritems - slot - 1));
4887 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4888 MOD_LOG_KEY_REMOVE, GFP_NOFS);
4893 btrfs_set_header_nritems(parent, nritems);
4894 if (nritems == 0 && parent == root->node) {
4895 BUG_ON(btrfs_header_level(root->node) != 1);
4896 /* just turn the root into a leaf and break */
4897 btrfs_set_header_level(root->node, 0);
4898 } else if (slot == 0) {
4899 struct btrfs_disk_key disk_key;
4901 btrfs_node_key(parent, &disk_key, 0);
4902 fixup_low_keys(root, path, &disk_key, level + 1);
4904 btrfs_mark_buffer_dirty(parent);
4908 * a helper function to delete the leaf pointed to by path->slots[1] and
4911 * This deletes the pointer in path->nodes[1] and frees the leaf
4912 * block extent. zero is returned if it all worked out, < 0 otherwise.
4914 * The path must have already been setup for deleting the leaf, including
4915 * all the proper balancing. path->nodes[1] must be locked.
4917 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4918 struct btrfs_root *root,
4919 struct btrfs_path *path,
4920 struct extent_buffer *leaf)
4922 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4923 del_ptr(root, path, 1, path->slots[1]);
4926 * btrfs_free_extent is expensive, we want to make sure we
4927 * aren't holding any locks when we call it
4929 btrfs_unlock_up_safe(path, 0);
4931 root_sub_used(root, leaf->len);
4933 extent_buffer_get(leaf);
4934 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4935 free_extent_buffer_stale(leaf);
4938 * delete the item at the leaf level in path. If that empties
4939 * the leaf, remove it from the tree
4941 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4942 struct btrfs_path *path, int slot, int nr)
4944 struct extent_buffer *leaf;
4945 struct btrfs_item *item;
4952 struct btrfs_map_token token;
4954 btrfs_init_map_token(&token);
4956 leaf = path->nodes[0];
4957 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4959 for (i = 0; i < nr; i++)
4960 dsize += btrfs_item_size_nr(leaf, slot + i);
4962 nritems = btrfs_header_nritems(leaf);
4964 if (slot + nr != nritems) {
4965 int data_end = leaf_data_end(root, leaf);
4967 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4969 btrfs_leaf_data(leaf) + data_end,
4970 last_off - data_end);
4972 for (i = slot + nr; i < nritems; i++) {
4975 item = btrfs_item_nr(i);
4976 ioff = btrfs_token_item_offset(leaf, item, &token);
4977 btrfs_set_token_item_offset(leaf, item,
4978 ioff + dsize, &token);
4981 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4982 btrfs_item_nr_offset(slot + nr),
4983 sizeof(struct btrfs_item) *
4984 (nritems - slot - nr));
4986 btrfs_set_header_nritems(leaf, nritems - nr);
4989 /* delete the leaf if we've emptied it */
4991 if (leaf == root->node) {
4992 btrfs_set_header_level(leaf, 0);
4994 btrfs_set_path_blocking(path);
4995 clean_tree_block(trans, root, leaf);
4996 btrfs_del_leaf(trans, root, path, leaf);
4999 int used = leaf_space_used(leaf, 0, nritems);
5001 struct btrfs_disk_key disk_key;
5003 btrfs_item_key(leaf, &disk_key, 0);
5004 fixup_low_keys(root, path, &disk_key, 1);
5007 /* delete the leaf if it is mostly empty */
5008 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
5009 /* push_leaf_left fixes the path.
5010 * make sure the path still points to our leaf
5011 * for possible call to del_ptr below
5013 slot = path->slots[1];
5014 extent_buffer_get(leaf);
5016 btrfs_set_path_blocking(path);
5017 wret = push_leaf_left(trans, root, path, 1, 1,
5019 if (wret < 0 && wret != -ENOSPC)
5022 if (path->nodes[0] == leaf &&
5023 btrfs_header_nritems(leaf)) {
5024 wret = push_leaf_right(trans, root, path, 1,
5026 if (wret < 0 && wret != -ENOSPC)
5030 if (btrfs_header_nritems(leaf) == 0) {
5031 path->slots[1] = slot;
5032 btrfs_del_leaf(trans, root, path, leaf);
5033 free_extent_buffer(leaf);
5036 /* if we're still in the path, make sure
5037 * we're dirty. Otherwise, one of the
5038 * push_leaf functions must have already
5039 * dirtied this buffer
5041 if (path->nodes[0] == leaf)
5042 btrfs_mark_buffer_dirty(leaf);
5043 free_extent_buffer(leaf);
5046 btrfs_mark_buffer_dirty(leaf);
5053 * search the tree again to find a leaf with lesser keys
5054 * returns 0 if it found something or 1 if there are no lesser leaves.
5055 * returns < 0 on io errors.
5057 * This may release the path, and so you may lose any locks held at the
5060 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5062 struct btrfs_key key;
5063 struct btrfs_disk_key found_key;
5066 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5068 if (key.offset > 0) {
5070 } else if (key.type > 0) {
5072 key.offset = (u64)-1;
5073 } else if (key.objectid > 0) {
5076 key.offset = (u64)-1;
5081 btrfs_release_path(path);
5082 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5085 btrfs_item_key(path->nodes[0], &found_key, 0);
5086 ret = comp_keys(&found_key, &key);
5088 * We might have had an item with the previous key in the tree right
5089 * before we released our path. And after we released our path, that
5090 * item might have been pushed to the first slot (0) of the leaf we
5091 * were holding due to a tree balance. Alternatively, an item with the
5092 * previous key can exist as the only element of a leaf (big fat item).
5093 * Therefore account for these 2 cases, so that our callers (like
5094 * btrfs_previous_item) don't miss an existing item with a key matching
5095 * the previous key we computed above.
5103 * A helper function to walk down the tree starting at min_key, and looking
5104 * for nodes or leaves that are have a minimum transaction id.
5105 * This is used by the btree defrag code, and tree logging
5107 * This does not cow, but it does stuff the starting key it finds back
5108 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5109 * key and get a writable path.
5111 * This does lock as it descends, and path->keep_locks should be set
5112 * to 1 by the caller.
5114 * This honors path->lowest_level to prevent descent past a given level
5117 * min_trans indicates the oldest transaction that you are interested
5118 * in walking through. Any nodes or leaves older than min_trans are
5119 * skipped over (without reading them).
5121 * returns zero if something useful was found, < 0 on error and 1 if there
5122 * was nothing in the tree that matched the search criteria.
5124 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5125 struct btrfs_path *path,
5128 struct extent_buffer *cur;
5129 struct btrfs_key found_key;
5135 int keep_locks = path->keep_locks;
5137 path->keep_locks = 1;
5139 cur = btrfs_read_lock_root_node(root);
5140 level = btrfs_header_level(cur);
5141 WARN_ON(path->nodes[level]);
5142 path->nodes[level] = cur;
5143 path->locks[level] = BTRFS_READ_LOCK;
5145 if (btrfs_header_generation(cur) < min_trans) {
5150 nritems = btrfs_header_nritems(cur);
5151 level = btrfs_header_level(cur);
5152 sret = bin_search(cur, min_key, level, &slot);
5154 /* at the lowest level, we're done, setup the path and exit */
5155 if (level == path->lowest_level) {
5156 if (slot >= nritems)
5159 path->slots[level] = slot;
5160 btrfs_item_key_to_cpu(cur, &found_key, slot);
5163 if (sret && slot > 0)
5166 * check this node pointer against the min_trans parameters.
5167 * If it is too old, old, skip to the next one.
5169 while (slot < nritems) {
5172 gen = btrfs_node_ptr_generation(cur, slot);
5173 if (gen < min_trans) {
5181 * we didn't find a candidate key in this node, walk forward
5182 * and find another one
5184 if (slot >= nritems) {
5185 path->slots[level] = slot;
5186 btrfs_set_path_blocking(path);
5187 sret = btrfs_find_next_key(root, path, min_key, level,
5190 btrfs_release_path(path);
5196 /* save our key for returning back */
5197 btrfs_node_key_to_cpu(cur, &found_key, slot);
5198 path->slots[level] = slot;
5199 if (level == path->lowest_level) {
5203 btrfs_set_path_blocking(path);
5204 cur = read_node_slot(root, cur, slot);
5205 BUG_ON(!cur); /* -ENOMEM */
5207 btrfs_tree_read_lock(cur);
5209 path->locks[level - 1] = BTRFS_READ_LOCK;
5210 path->nodes[level - 1] = cur;
5211 unlock_up(path, level, 1, 0, NULL);
5212 btrfs_clear_path_blocking(path, NULL, 0);
5215 path->keep_locks = keep_locks;
5217 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5218 btrfs_set_path_blocking(path);
5219 memcpy(min_key, &found_key, sizeof(found_key));
5224 static void tree_move_down(struct btrfs_root *root,
5225 struct btrfs_path *path,
5226 int *level, int root_level)
5228 BUG_ON(*level == 0);
5229 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
5230 path->slots[*level]);
5231 path->slots[*level - 1] = 0;
5235 static int tree_move_next_or_upnext(struct btrfs_root *root,
5236 struct btrfs_path *path,
5237 int *level, int root_level)
5241 nritems = btrfs_header_nritems(path->nodes[*level]);
5243 path->slots[*level]++;
5245 while (path->slots[*level] >= nritems) {
5246 if (*level == root_level)
5250 path->slots[*level] = 0;
5251 free_extent_buffer(path->nodes[*level]);
5252 path->nodes[*level] = NULL;
5254 path->slots[*level]++;
5256 nritems = btrfs_header_nritems(path->nodes[*level]);
5263 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5266 static int tree_advance(struct btrfs_root *root,
5267 struct btrfs_path *path,
5268 int *level, int root_level,
5270 struct btrfs_key *key)
5274 if (*level == 0 || !allow_down) {
5275 ret = tree_move_next_or_upnext(root, path, level, root_level);
5277 tree_move_down(root, path, level, root_level);
5282 btrfs_item_key_to_cpu(path->nodes[*level], key,
5283 path->slots[*level]);
5285 btrfs_node_key_to_cpu(path->nodes[*level], key,
5286 path->slots[*level]);
5291 static int tree_compare_item(struct btrfs_root *left_root,
5292 struct btrfs_path *left_path,
5293 struct btrfs_path *right_path,
5298 unsigned long off1, off2;
5300 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5301 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5305 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5306 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5307 right_path->slots[0]);
5309 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5311 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5318 #define ADVANCE_ONLY_NEXT -1
5321 * This function compares two trees and calls the provided callback for
5322 * every changed/new/deleted item it finds.
5323 * If shared tree blocks are encountered, whole subtrees are skipped, making
5324 * the compare pretty fast on snapshotted subvolumes.
5326 * This currently works on commit roots only. As commit roots are read only,
5327 * we don't do any locking. The commit roots are protected with transactions.
5328 * Transactions are ended and rejoined when a commit is tried in between.
5330 * This function checks for modifications done to the trees while comparing.
5331 * If it detects a change, it aborts immediately.
5333 int btrfs_compare_trees(struct btrfs_root *left_root,
5334 struct btrfs_root *right_root,
5335 btrfs_changed_cb_t changed_cb, void *ctx)
5339 struct btrfs_path *left_path = NULL;
5340 struct btrfs_path *right_path = NULL;
5341 struct btrfs_key left_key;
5342 struct btrfs_key right_key;
5343 char *tmp_buf = NULL;
5344 int left_root_level;
5345 int right_root_level;
5348 int left_end_reached;
5349 int right_end_reached;
5357 left_path = btrfs_alloc_path();
5362 right_path = btrfs_alloc_path();
5368 tmp_buf = kmalloc(left_root->nodesize, GFP_NOFS);
5374 left_path->search_commit_root = 1;
5375 left_path->skip_locking = 1;
5376 right_path->search_commit_root = 1;
5377 right_path->skip_locking = 1;
5380 * Strategy: Go to the first items of both trees. Then do
5382 * If both trees are at level 0
5383 * Compare keys of current items
5384 * If left < right treat left item as new, advance left tree
5386 * If left > right treat right item as deleted, advance right tree
5388 * If left == right do deep compare of items, treat as changed if
5389 * needed, advance both trees and repeat
5390 * If both trees are at the same level but not at level 0
5391 * Compare keys of current nodes/leafs
5392 * If left < right advance left tree and repeat
5393 * If left > right advance right tree and repeat
5394 * If left == right compare blockptrs of the next nodes/leafs
5395 * If they match advance both trees but stay at the same level
5397 * If they don't match advance both trees while allowing to go
5399 * If tree levels are different
5400 * Advance the tree that needs it and repeat
5402 * Advancing a tree means:
5403 * If we are at level 0, try to go to the next slot. If that's not
5404 * possible, go one level up and repeat. Stop when we found a level
5405 * where we could go to the next slot. We may at this point be on a
5408 * If we are not at level 0 and not on shared tree blocks, go one
5411 * If we are not at level 0 and on shared tree blocks, go one slot to
5412 * the right if possible or go up and right.
5415 down_read(&left_root->fs_info->commit_root_sem);
5416 left_level = btrfs_header_level(left_root->commit_root);
5417 left_root_level = left_level;
5418 left_path->nodes[left_level] = left_root->commit_root;
5419 extent_buffer_get(left_path->nodes[left_level]);
5421 right_level = btrfs_header_level(right_root->commit_root);
5422 right_root_level = right_level;
5423 right_path->nodes[right_level] = right_root->commit_root;
5424 extent_buffer_get(right_path->nodes[right_level]);
5425 up_read(&left_root->fs_info->commit_root_sem);
5427 if (left_level == 0)
5428 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5429 &left_key, left_path->slots[left_level]);
5431 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5432 &left_key, left_path->slots[left_level]);
5433 if (right_level == 0)
5434 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5435 &right_key, right_path->slots[right_level]);
5437 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5438 &right_key, right_path->slots[right_level]);
5440 left_end_reached = right_end_reached = 0;
5441 advance_left = advance_right = 0;
5444 if (advance_left && !left_end_reached) {
5445 ret = tree_advance(left_root, left_path, &left_level,
5447 advance_left != ADVANCE_ONLY_NEXT,
5450 left_end_reached = ADVANCE;
5453 if (advance_right && !right_end_reached) {
5454 ret = tree_advance(right_root, right_path, &right_level,
5456 advance_right != ADVANCE_ONLY_NEXT,
5459 right_end_reached = ADVANCE;
5463 if (left_end_reached && right_end_reached) {
5466 } else if (left_end_reached) {
5467 if (right_level == 0) {
5468 ret = changed_cb(left_root, right_root,
5469 left_path, right_path,
5471 BTRFS_COMPARE_TREE_DELETED,
5476 advance_right = ADVANCE;
5478 } else if (right_end_reached) {
5479 if (left_level == 0) {
5480 ret = changed_cb(left_root, right_root,
5481 left_path, right_path,
5483 BTRFS_COMPARE_TREE_NEW,
5488 advance_left = ADVANCE;
5492 if (left_level == 0 && right_level == 0) {
5493 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5495 ret = changed_cb(left_root, right_root,
5496 left_path, right_path,
5498 BTRFS_COMPARE_TREE_NEW,
5502 advance_left = ADVANCE;
5503 } else if (cmp > 0) {
5504 ret = changed_cb(left_root, right_root,
5505 left_path, right_path,
5507 BTRFS_COMPARE_TREE_DELETED,
5511 advance_right = ADVANCE;
5513 enum btrfs_compare_tree_result result;
5515 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5516 ret = tree_compare_item(left_root, left_path,
5517 right_path, tmp_buf);
5519 result = BTRFS_COMPARE_TREE_CHANGED;
5521 result = BTRFS_COMPARE_TREE_SAME;
5522 ret = changed_cb(left_root, right_root,
5523 left_path, right_path,
5524 &left_key, result, ctx);
5527 advance_left = ADVANCE;
5528 advance_right = ADVANCE;
5530 } else if (left_level == right_level) {
5531 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5533 advance_left = ADVANCE;
5534 } else if (cmp > 0) {
5535 advance_right = ADVANCE;
5537 left_blockptr = btrfs_node_blockptr(
5538 left_path->nodes[left_level],
5539 left_path->slots[left_level]);
5540 right_blockptr = btrfs_node_blockptr(
5541 right_path->nodes[right_level],
5542 right_path->slots[right_level]);
5543 left_gen = btrfs_node_ptr_generation(
5544 left_path->nodes[left_level],
5545 left_path->slots[left_level]);
5546 right_gen = btrfs_node_ptr_generation(
5547 right_path->nodes[right_level],
5548 right_path->slots[right_level]);
5549 if (left_blockptr == right_blockptr &&
5550 left_gen == right_gen) {
5552 * As we're on a shared block, don't
5553 * allow to go deeper.
5555 advance_left = ADVANCE_ONLY_NEXT;
5556 advance_right = ADVANCE_ONLY_NEXT;
5558 advance_left = ADVANCE;
5559 advance_right = ADVANCE;
5562 } else if (left_level < right_level) {
5563 advance_right = ADVANCE;
5565 advance_left = ADVANCE;
5570 btrfs_free_path(left_path);
5571 btrfs_free_path(right_path);
5577 * this is similar to btrfs_next_leaf, but does not try to preserve
5578 * and fixup the path. It looks for and returns the next key in the
5579 * tree based on the current path and the min_trans parameters.
5581 * 0 is returned if another key is found, < 0 if there are any errors
5582 * and 1 is returned if there are no higher keys in the tree
5584 * path->keep_locks should be set to 1 on the search made before
5585 * calling this function.
5587 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5588 struct btrfs_key *key, int level, u64 min_trans)
5591 struct extent_buffer *c;
5593 WARN_ON(!path->keep_locks);
5594 while (level < BTRFS_MAX_LEVEL) {
5595 if (!path->nodes[level])
5598 slot = path->slots[level] + 1;
5599 c = path->nodes[level];
5601 if (slot >= btrfs_header_nritems(c)) {
5604 struct btrfs_key cur_key;
5605 if (level + 1 >= BTRFS_MAX_LEVEL ||
5606 !path->nodes[level + 1])
5609 if (path->locks[level + 1]) {
5614 slot = btrfs_header_nritems(c) - 1;
5616 btrfs_item_key_to_cpu(c, &cur_key, slot);
5618 btrfs_node_key_to_cpu(c, &cur_key, slot);
5620 orig_lowest = path->lowest_level;
5621 btrfs_release_path(path);
5622 path->lowest_level = level;
5623 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5625 path->lowest_level = orig_lowest;
5629 c = path->nodes[level];
5630 slot = path->slots[level];
5637 btrfs_item_key_to_cpu(c, key, slot);
5639 u64 gen = btrfs_node_ptr_generation(c, slot);
5641 if (gen < min_trans) {
5645 btrfs_node_key_to_cpu(c, key, slot);
5653 * search the tree again to find a leaf with greater keys
5654 * returns 0 if it found something or 1 if there are no greater leaves.
5655 * returns < 0 on io errors.
5657 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5659 return btrfs_next_old_leaf(root, path, 0);
5662 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5667 struct extent_buffer *c;
5668 struct extent_buffer *next;
5669 struct btrfs_key key;
5672 int old_spinning = path->leave_spinning;
5673 int next_rw_lock = 0;
5675 nritems = btrfs_header_nritems(path->nodes[0]);
5679 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5684 btrfs_release_path(path);
5686 path->keep_locks = 1;
5687 path->leave_spinning = 1;
5690 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5692 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5693 path->keep_locks = 0;
5698 nritems = btrfs_header_nritems(path->nodes[0]);
5700 * by releasing the path above we dropped all our locks. A balance
5701 * could have added more items next to the key that used to be
5702 * at the very end of the block. So, check again here and
5703 * advance the path if there are now more items available.
5705 if (nritems > 0 && path->slots[0] < nritems - 1) {
5712 * So the above check misses one case:
5713 * - after releasing the path above, someone has removed the item that
5714 * used to be at the very end of the block, and balance between leafs
5715 * gets another one with bigger key.offset to replace it.
5717 * This one should be returned as well, or we can get leaf corruption
5718 * later(esp. in __btrfs_drop_extents()).
5720 * And a bit more explanation about this check,
5721 * with ret > 0, the key isn't found, the path points to the slot
5722 * where it should be inserted, so the path->slots[0] item must be the
5725 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5730 while (level < BTRFS_MAX_LEVEL) {
5731 if (!path->nodes[level]) {
5736 slot = path->slots[level] + 1;
5737 c = path->nodes[level];
5738 if (slot >= btrfs_header_nritems(c)) {
5740 if (level == BTRFS_MAX_LEVEL) {
5748 btrfs_tree_unlock_rw(next, next_rw_lock);
5749 free_extent_buffer(next);
5753 next_rw_lock = path->locks[level];
5754 ret = read_block_for_search(NULL, root, path, &next, level,
5760 btrfs_release_path(path);
5764 if (!path->skip_locking) {
5765 ret = btrfs_try_tree_read_lock(next);
5766 if (!ret && time_seq) {
5768 * If we don't get the lock, we may be racing
5769 * with push_leaf_left, holding that lock while
5770 * itself waiting for the leaf we've currently
5771 * locked. To solve this situation, we give up
5772 * on our lock and cycle.
5774 free_extent_buffer(next);
5775 btrfs_release_path(path);
5780 btrfs_set_path_blocking(path);
5781 btrfs_tree_read_lock(next);
5782 btrfs_clear_path_blocking(path, next,
5785 next_rw_lock = BTRFS_READ_LOCK;
5789 path->slots[level] = slot;
5792 c = path->nodes[level];
5793 if (path->locks[level])
5794 btrfs_tree_unlock_rw(c, path->locks[level]);
5796 free_extent_buffer(c);
5797 path->nodes[level] = next;
5798 path->slots[level] = 0;
5799 if (!path->skip_locking)
5800 path->locks[level] = next_rw_lock;
5804 ret = read_block_for_search(NULL, root, path, &next, level,
5810 btrfs_release_path(path);
5814 if (!path->skip_locking) {
5815 ret = btrfs_try_tree_read_lock(next);
5817 btrfs_set_path_blocking(path);
5818 btrfs_tree_read_lock(next);
5819 btrfs_clear_path_blocking(path, next,
5822 next_rw_lock = BTRFS_READ_LOCK;
5827 unlock_up(path, 0, 1, 0, NULL);
5828 path->leave_spinning = old_spinning;
5830 btrfs_set_path_blocking(path);
5836 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5837 * searching until it gets past min_objectid or finds an item of 'type'
5839 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5841 int btrfs_previous_item(struct btrfs_root *root,
5842 struct btrfs_path *path, u64 min_objectid,
5845 struct btrfs_key found_key;
5846 struct extent_buffer *leaf;
5851 if (path->slots[0] == 0) {
5852 btrfs_set_path_blocking(path);
5853 ret = btrfs_prev_leaf(root, path);
5859 leaf = path->nodes[0];
5860 nritems = btrfs_header_nritems(leaf);
5863 if (path->slots[0] == nritems)
5866 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5867 if (found_key.objectid < min_objectid)
5869 if (found_key.type == type)
5871 if (found_key.objectid == min_objectid &&
5872 found_key.type < type)
5879 * search in extent tree to find a previous Metadata/Data extent item with
5882 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5884 int btrfs_previous_extent_item(struct btrfs_root *root,
5885 struct btrfs_path *path, u64 min_objectid)
5887 struct btrfs_key found_key;
5888 struct extent_buffer *leaf;
5893 if (path->slots[0] == 0) {
5894 btrfs_set_path_blocking(path);
5895 ret = btrfs_prev_leaf(root, path);
5901 leaf = path->nodes[0];
5902 nritems = btrfs_header_nritems(leaf);
5905 if (path->slots[0] == nritems)
5908 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5909 if (found_key.objectid < min_objectid)
5911 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5912 found_key.type == BTRFS_METADATA_ITEM_KEY)
5914 if (found_key.objectid == min_objectid &&
5915 found_key.type < BTRFS_EXTENT_ITEM_KEY)