2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/rbtree.h>
22 #include <linux/vmalloc.h>
25 #include "transaction.h"
26 #include "print-tree.h"
29 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
30 *root, struct btrfs_path *path, int level);
31 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
32 *root, struct btrfs_key *ins_key,
33 struct btrfs_path *path, int data_size, int extend);
34 static int push_node_left(struct btrfs_trans_handle *trans,
35 struct btrfs_fs_info *fs_info,
36 struct extent_buffer *dst,
37 struct extent_buffer *src, int empty);
38 static int balance_node_right(struct btrfs_trans_handle *trans,
39 struct btrfs_fs_info *fs_info,
40 struct extent_buffer *dst_buf,
41 struct extent_buffer *src_buf);
42 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
44 static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
45 struct extent_buffer *eb);
47 struct btrfs_path *btrfs_alloc_path(void)
49 return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
53 * set all locked nodes in the path to blocking locks. This should
54 * be done before scheduling
56 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
59 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
60 if (!p->nodes[i] || !p->locks[i])
62 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
63 if (p->locks[i] == BTRFS_READ_LOCK)
64 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
65 else if (p->locks[i] == BTRFS_WRITE_LOCK)
66 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
71 * reset all the locked nodes in the patch to spinning locks.
73 * held is used to keep lockdep happy, when lockdep is enabled
74 * we set held to a blocking lock before we go around and
75 * retake all the spinlocks in the path. You can safely use NULL
78 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
79 struct extent_buffer *held, int held_rw)
84 btrfs_set_lock_blocking_rw(held, held_rw);
85 if (held_rw == BTRFS_WRITE_LOCK)
86 held_rw = BTRFS_WRITE_LOCK_BLOCKING;
87 else if (held_rw == BTRFS_READ_LOCK)
88 held_rw = BTRFS_READ_LOCK_BLOCKING;
90 btrfs_set_path_blocking(p);
92 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
93 if (p->nodes[i] && p->locks[i]) {
94 btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
95 if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
96 p->locks[i] = BTRFS_WRITE_LOCK;
97 else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
98 p->locks[i] = BTRFS_READ_LOCK;
103 btrfs_clear_lock_blocking_rw(held, held_rw);
106 /* this also releases the path */
107 void btrfs_free_path(struct btrfs_path *p)
111 btrfs_release_path(p);
112 kmem_cache_free(btrfs_path_cachep, p);
116 * path release drops references on the extent buffers in the path
117 * and it drops any locks held by this path
119 * It is safe to call this on paths that no locks or extent buffers held.
121 noinline void btrfs_release_path(struct btrfs_path *p)
125 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
130 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
133 free_extent_buffer(p->nodes[i]);
139 * safely gets a reference on the root node of a tree. A lock
140 * is not taken, so a concurrent writer may put a different node
141 * at the root of the tree. See btrfs_lock_root_node for the
144 * The extent buffer returned by this has a reference taken, so
145 * it won't disappear. It may stop being the root of the tree
146 * at any time because there are no locks held.
148 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
150 struct extent_buffer *eb;
154 eb = rcu_dereference(root->node);
157 * RCU really hurts here, we could free up the root node because
158 * it was COWed but we may not get the new root node yet so do
159 * the inc_not_zero dance and if it doesn't work then
160 * synchronize_rcu and try again.
162 if (atomic_inc_not_zero(&eb->refs)) {
172 /* loop around taking references on and locking the root node of the
173 * tree until you end up with a lock on the root. A locked buffer
174 * is returned, with a reference held.
176 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
178 struct extent_buffer *eb;
181 eb = btrfs_root_node(root);
183 if (eb == root->node)
185 btrfs_tree_unlock(eb);
186 free_extent_buffer(eb);
191 /* loop around taking references on and locking the root node of the
192 * tree until you end up with a lock on the root. A locked buffer
193 * is returned, with a reference held.
195 static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
197 struct extent_buffer *eb;
200 eb = btrfs_root_node(root);
201 btrfs_tree_read_lock(eb);
202 if (eb == root->node)
204 btrfs_tree_read_unlock(eb);
205 free_extent_buffer(eb);
210 /* cowonly root (everything not a reference counted cow subvolume), just get
211 * put onto a simple dirty list. transaction.c walks this to make sure they
212 * get properly updated on disk.
214 static void add_root_to_dirty_list(struct btrfs_root *root)
216 struct btrfs_fs_info *fs_info = root->fs_info;
218 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
219 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
222 spin_lock(&fs_info->trans_lock);
223 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
224 /* Want the extent tree to be the last on the list */
225 if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID)
226 list_move_tail(&root->dirty_list,
227 &fs_info->dirty_cowonly_roots);
229 list_move(&root->dirty_list,
230 &fs_info->dirty_cowonly_roots);
232 spin_unlock(&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 btrfs_fs_info *fs_info = root->fs_info;
246 struct extent_buffer *cow;
249 struct btrfs_disk_key disk_key;
251 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
252 trans->transid != fs_info->running_transaction->transid);
253 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
254 trans->transid != root->last_trans);
256 level = btrfs_header_level(buf);
258 btrfs_item_key(buf, &disk_key, 0);
260 btrfs_node_key(buf, &disk_key, 0);
262 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
263 &disk_key, level, buf->start, 0);
267 copy_extent_buffer_full(cow, buf);
268 btrfs_set_header_bytenr(cow, cow->start);
269 btrfs_set_header_generation(cow, trans->transid);
270 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
271 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
272 BTRFS_HEADER_FLAG_RELOC);
273 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
274 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
276 btrfs_set_header_owner(cow, new_root_objectid);
278 write_extent_buffer_fsid(cow, fs_info->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 {
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:
440 * node/leaf start address -> sequence
442 * The 'start address' is the logical address of the *new* root node
443 * for root replace operations, or the logical address of the affected
444 * block for all other operations.
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->logical < tm->logical)
466 new = &((*new)->rb_left);
467 else if (cur->logical > tm->logical)
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->logical = eb->start;
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 = kcalloc(nr_items, sizeof(struct tree_mod_elem *), flags);
587 tm = kzalloc(sizeof(*tm), flags);
593 tm->logical = eb->start;
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 = kcalloc(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->logical = new_root->start;
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;
745 tree_mod_log_read_lock(fs_info);
746 tm_root = &fs_info->tree_mod_log;
747 node = tm_root->rb_node;
749 cur = container_of(node, struct tree_mod_elem, node);
750 if (cur->logical < start) {
751 node = node->rb_left;
752 } else if (cur->logical > start) {
753 node = node->rb_right;
754 } else if (cur->seq < min_seq) {
755 node = node->rb_left;
756 } else if (!smallest) {
757 /* we want the node with the highest seq */
759 BUG_ON(found->seq > cur->seq);
761 node = node->rb_left;
762 } else if (cur->seq > min_seq) {
763 /* we want the node with the smallest seq */
765 BUG_ON(found->seq < cur->seq);
767 node = node->rb_right;
773 tree_mod_log_read_unlock(fs_info);
779 * this returns the element from the log with the smallest time sequence
780 * value that's in the log (the oldest log item). any element with a time
781 * sequence lower than min_seq will be ignored.
783 static struct tree_mod_elem *
784 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
787 return __tree_mod_log_search(fs_info, start, min_seq, 1);
791 * this returns the element from the log with the largest time sequence
792 * value that's in the log (the most recent log item). any element with
793 * a time sequence lower than min_seq will be ignored.
795 static struct tree_mod_elem *
796 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
798 return __tree_mod_log_search(fs_info, start, min_seq, 0);
802 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
803 struct extent_buffer *src, unsigned long dst_offset,
804 unsigned long src_offset, int nr_items)
807 struct tree_mod_elem **tm_list = NULL;
808 struct tree_mod_elem **tm_list_add, **tm_list_rem;
812 if (!tree_mod_need_log(fs_info, NULL))
815 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
818 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
823 tm_list_add = tm_list;
824 tm_list_rem = tm_list + nr_items;
825 for (i = 0; i < nr_items; i++) {
826 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
827 MOD_LOG_KEY_REMOVE, GFP_NOFS);
828 if (!tm_list_rem[i]) {
833 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
834 MOD_LOG_KEY_ADD, GFP_NOFS);
835 if (!tm_list_add[i]) {
841 if (tree_mod_dont_log(fs_info, NULL))
845 for (i = 0; i < nr_items; i++) {
846 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
849 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
854 tree_mod_log_write_unlock(fs_info);
860 for (i = 0; i < nr_items * 2; i++) {
861 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
862 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
866 tree_mod_log_write_unlock(fs_info);
873 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
874 int dst_offset, int src_offset, int nr_items)
877 ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
883 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
884 struct extent_buffer *eb, int slot, int atomic)
888 ret = tree_mod_log_insert_key(fs_info, eb, slot,
890 atomic ? GFP_ATOMIC : GFP_NOFS);
895 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
897 struct tree_mod_elem **tm_list = NULL;
902 if (btrfs_header_level(eb) == 0)
905 if (!tree_mod_need_log(fs_info, NULL))
908 nritems = btrfs_header_nritems(eb);
909 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
913 for (i = 0; i < nritems; i++) {
914 tm_list[i] = alloc_tree_mod_elem(eb, i,
915 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
922 if (tree_mod_dont_log(fs_info, eb))
925 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
926 tree_mod_log_write_unlock(fs_info);
934 for (i = 0; i < nritems; i++)
942 tree_mod_log_set_root_pointer(struct btrfs_root *root,
943 struct extent_buffer *new_root_node,
947 ret = tree_mod_log_insert_root(root->fs_info, root->node,
948 new_root_node, GFP_NOFS, log_removal);
953 * check if the tree block can be shared by multiple trees
955 int btrfs_block_can_be_shared(struct btrfs_root *root,
956 struct extent_buffer *buf)
959 * Tree blocks not in reference counted trees and tree roots
960 * are never shared. If a block was allocated after the last
961 * snapshot and the block was not allocated by tree relocation,
962 * we know the block is not shared.
964 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
965 buf != root->node && buf != root->commit_root &&
966 (btrfs_header_generation(buf) <=
967 btrfs_root_last_snapshot(&root->root_item) ||
968 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
970 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
971 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
972 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
978 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
979 struct btrfs_root *root,
980 struct extent_buffer *buf,
981 struct extent_buffer *cow,
984 struct btrfs_fs_info *fs_info = root->fs_info;
992 * Backrefs update rules:
994 * Always use full backrefs for extent pointers in tree block
995 * allocated by tree relocation.
997 * If a shared tree block is no longer referenced by its owner
998 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
999 * use full backrefs for extent pointers in tree block.
1001 * If a tree block is been relocating
1002 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1003 * use full backrefs for extent pointers in tree block.
1004 * The reason for this is some operations (such as drop tree)
1005 * are only allowed for blocks use full backrefs.
1008 if (btrfs_block_can_be_shared(root, buf)) {
1009 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
1010 btrfs_header_level(buf), 1,
1016 btrfs_handle_fs_error(fs_info, ret, NULL);
1021 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1022 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1023 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1028 owner = btrfs_header_owner(buf);
1029 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
1030 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
1033 if ((owner == root->root_key.objectid ||
1034 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
1035 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1036 ret = btrfs_inc_ref(trans, root, buf, 1);
1037 BUG_ON(ret); /* -ENOMEM */
1039 if (root->root_key.objectid ==
1040 BTRFS_TREE_RELOC_OBJECTID) {
1041 ret = btrfs_dec_ref(trans, root, buf, 0);
1042 BUG_ON(ret); /* -ENOMEM */
1043 ret = btrfs_inc_ref(trans, root, cow, 1);
1044 BUG_ON(ret); /* -ENOMEM */
1046 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
1049 if (root->root_key.objectid ==
1050 BTRFS_TREE_RELOC_OBJECTID)
1051 ret = btrfs_inc_ref(trans, root, cow, 1);
1053 ret = btrfs_inc_ref(trans, root, cow, 0);
1054 BUG_ON(ret); /* -ENOMEM */
1056 if (new_flags != 0) {
1057 int level = btrfs_header_level(buf);
1059 ret = btrfs_set_disk_extent_flags(trans, fs_info,
1062 new_flags, level, 0);
1067 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
1068 if (root->root_key.objectid ==
1069 BTRFS_TREE_RELOC_OBJECTID)
1070 ret = btrfs_inc_ref(trans, root, cow, 1);
1072 ret = btrfs_inc_ref(trans, root, cow, 0);
1073 BUG_ON(ret); /* -ENOMEM */
1074 ret = btrfs_dec_ref(trans, root, buf, 1);
1075 BUG_ON(ret); /* -ENOMEM */
1077 clean_tree_block(trans, fs_info, buf);
1084 * does the dirty work in cow of a single block. The parent block (if
1085 * supplied) is updated to point to the new cow copy. The new buffer is marked
1086 * dirty and returned locked. If you modify the block it needs to be marked
1089 * search_start -- an allocation hint for the new block
1091 * empty_size -- a hint that you plan on doing more cow. This is the size in
1092 * bytes the allocator should try to find free next to the block it returns.
1093 * This is just a hint and may be ignored by the allocator.
1095 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1096 struct btrfs_root *root,
1097 struct extent_buffer *buf,
1098 struct extent_buffer *parent, int parent_slot,
1099 struct extent_buffer **cow_ret,
1100 u64 search_start, u64 empty_size)
1102 struct btrfs_fs_info *fs_info = root->fs_info;
1103 struct btrfs_disk_key disk_key;
1104 struct extent_buffer *cow;
1107 int unlock_orig = 0;
1108 u64 parent_start = 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 != 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) && parent)
1128 parent_start = parent->start;
1130 cow = btrfs_alloc_tree_block(trans, root, parent_start,
1131 root->root_key.objectid, &disk_key, level,
1132 search_start, empty_size);
1134 return PTR_ERR(cow);
1136 /* cow is set to blocking by btrfs_init_new_buffer */
1138 copy_extent_buffer_full(cow, buf);
1139 btrfs_set_header_bytenr(cow, cow->start);
1140 btrfs_set_header_generation(cow, trans->transid);
1141 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1142 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1143 BTRFS_HEADER_FLAG_RELOC);
1144 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1145 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1147 btrfs_set_header_owner(cow, root->root_key.objectid);
1149 write_extent_buffer_fsid(cow, fs_info->fsid);
1151 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1153 btrfs_abort_transaction(trans, ret);
1157 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1158 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1160 btrfs_abort_transaction(trans, ret);
1165 if (buf == root->node) {
1166 WARN_ON(parent && parent != buf);
1167 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1168 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1169 parent_start = buf->start;
1171 extent_buffer_get(cow);
1172 tree_mod_log_set_root_pointer(root, cow, 1);
1173 rcu_assign_pointer(root->node, cow);
1175 btrfs_free_tree_block(trans, root, buf, parent_start,
1177 free_extent_buffer(buf);
1178 add_root_to_dirty_list(root);
1180 WARN_ON(trans->transid != btrfs_header_generation(parent));
1181 tree_mod_log_insert_key(fs_info, parent, parent_slot,
1182 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1183 btrfs_set_node_blockptr(parent, parent_slot,
1185 btrfs_set_node_ptr_generation(parent, parent_slot,
1187 btrfs_mark_buffer_dirty(parent);
1189 ret = tree_mod_log_free_eb(fs_info, buf);
1191 btrfs_abort_transaction(trans, ret);
1195 btrfs_free_tree_block(trans, root, buf, parent_start,
1199 btrfs_tree_unlock(buf);
1200 free_extent_buffer_stale(buf);
1201 btrfs_mark_buffer_dirty(cow);
1207 * returns the logical address of the oldest predecessor of the given root.
1208 * entries older than time_seq are ignored.
1210 static struct tree_mod_elem *
1211 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1212 struct extent_buffer *eb_root, u64 time_seq)
1214 struct tree_mod_elem *tm;
1215 struct tree_mod_elem *found = NULL;
1216 u64 root_logical = eb_root->start;
1223 * the very last operation that's logged for a root is the
1224 * replacement operation (if it is replaced at all). this has
1225 * the logical address of the *new* root, making it the very
1226 * first operation that's logged for this root.
1229 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1234 * if there are no tree operation for the oldest root, we simply
1235 * return it. this should only happen if that (old) root is at
1242 * if there's an operation that's not a root replacement, we
1243 * found the oldest version of our root. normally, we'll find a
1244 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1246 if (tm->op != MOD_LOG_ROOT_REPLACE)
1250 root_logical = tm->old_root.logical;
1254 /* if there's no old root to return, return what we found instead */
1262 * tm is a pointer to the first operation to rewind within eb. then, all
1263 * previous operations will be rewound (until we reach something older than
1267 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1268 u64 time_seq, struct tree_mod_elem *first_tm)
1271 struct rb_node *next;
1272 struct tree_mod_elem *tm = first_tm;
1273 unsigned long o_dst;
1274 unsigned long o_src;
1275 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1277 n = btrfs_header_nritems(eb);
1278 tree_mod_log_read_lock(fs_info);
1279 while (tm && tm->seq >= time_seq) {
1281 * all the operations are recorded with the operator used for
1282 * the modification. as we're going backwards, we do the
1283 * opposite of each operation here.
1286 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1287 BUG_ON(tm->slot < n);
1289 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1290 case MOD_LOG_KEY_REMOVE:
1291 btrfs_set_node_key(eb, &tm->key, tm->slot);
1292 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1293 btrfs_set_node_ptr_generation(eb, tm->slot,
1297 case MOD_LOG_KEY_REPLACE:
1298 BUG_ON(tm->slot >= n);
1299 btrfs_set_node_key(eb, &tm->key, tm->slot);
1300 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1301 btrfs_set_node_ptr_generation(eb, tm->slot,
1304 case MOD_LOG_KEY_ADD:
1305 /* if a move operation is needed it's in the log */
1308 case MOD_LOG_MOVE_KEYS:
1309 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1310 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1311 memmove_extent_buffer(eb, o_dst, o_src,
1312 tm->move.nr_items * p_size);
1314 case MOD_LOG_ROOT_REPLACE:
1316 * this operation is special. for roots, this must be
1317 * handled explicitly before rewinding.
1318 * for non-roots, this operation may exist if the node
1319 * was a root: root A -> child B; then A gets empty and
1320 * B is promoted to the new root. in the mod log, we'll
1321 * have a root-replace operation for B, a tree block
1322 * that is no root. we simply ignore that operation.
1326 next = rb_next(&tm->node);
1329 tm = container_of(next, struct tree_mod_elem, node);
1330 if (tm->logical != first_tm->logical)
1333 tree_mod_log_read_unlock(fs_info);
1334 btrfs_set_header_nritems(eb, n);
1338 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1339 * is returned. If rewind operations happen, a fresh buffer is returned. The
1340 * returned buffer is always read-locked. If the returned buffer is not the
1341 * input buffer, the lock on the input buffer is released and the input buffer
1342 * is freed (its refcount is decremented).
1344 static struct extent_buffer *
1345 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1346 struct extent_buffer *eb, u64 time_seq)
1348 struct extent_buffer *eb_rewin;
1349 struct tree_mod_elem *tm;
1354 if (btrfs_header_level(eb) == 0)
1357 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1361 btrfs_set_path_blocking(path);
1362 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1364 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1365 BUG_ON(tm->slot != 0);
1366 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1368 btrfs_tree_read_unlock_blocking(eb);
1369 free_extent_buffer(eb);
1372 btrfs_set_header_bytenr(eb_rewin, eb->start);
1373 btrfs_set_header_backref_rev(eb_rewin,
1374 btrfs_header_backref_rev(eb));
1375 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1376 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1378 eb_rewin = btrfs_clone_extent_buffer(eb);
1380 btrfs_tree_read_unlock_blocking(eb);
1381 free_extent_buffer(eb);
1386 btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
1387 btrfs_tree_read_unlock_blocking(eb);
1388 free_extent_buffer(eb);
1390 extent_buffer_get(eb_rewin);
1391 btrfs_tree_read_lock(eb_rewin);
1392 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1393 WARN_ON(btrfs_header_nritems(eb_rewin) >
1394 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1400 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1401 * value. If there are no changes, the current root->root_node is returned. If
1402 * anything changed in between, there's a fresh buffer allocated on which the
1403 * rewind operations are done. In any case, the returned buffer is read locked.
1404 * Returns NULL on error (with no locks held).
1406 static inline struct extent_buffer *
1407 get_old_root(struct btrfs_root *root, u64 time_seq)
1409 struct btrfs_fs_info *fs_info = root->fs_info;
1410 struct tree_mod_elem *tm;
1411 struct extent_buffer *eb = NULL;
1412 struct extent_buffer *eb_root;
1413 struct extent_buffer *old;
1414 struct tree_mod_root *old_root = NULL;
1415 u64 old_generation = 0;
1418 eb_root = btrfs_read_lock_root_node(root);
1419 tm = __tree_mod_log_oldest_root(fs_info, eb_root, time_seq);
1423 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1424 old_root = &tm->old_root;
1425 old_generation = tm->generation;
1426 logical = old_root->logical;
1428 logical = eb_root->start;
1431 tm = tree_mod_log_search(fs_info, logical, time_seq);
1432 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1433 btrfs_tree_read_unlock(eb_root);
1434 free_extent_buffer(eb_root);
1435 old = read_tree_block(fs_info, logical, 0);
1436 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1438 free_extent_buffer(old);
1440 "failed to read tree block %llu from get_old_root",
1443 eb = btrfs_clone_extent_buffer(old);
1444 free_extent_buffer(old);
1446 } else if (old_root) {
1447 btrfs_tree_read_unlock(eb_root);
1448 free_extent_buffer(eb_root);
1449 eb = alloc_dummy_extent_buffer(fs_info, logical);
1451 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1452 eb = btrfs_clone_extent_buffer(eb_root);
1453 btrfs_tree_read_unlock_blocking(eb_root);
1454 free_extent_buffer(eb_root);
1459 extent_buffer_get(eb);
1460 btrfs_tree_read_lock(eb);
1462 btrfs_set_header_bytenr(eb, eb->start);
1463 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1464 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1465 btrfs_set_header_level(eb, old_root->level);
1466 btrfs_set_header_generation(eb, old_generation);
1469 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1471 WARN_ON(btrfs_header_level(eb) != 0);
1472 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1477 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1479 struct tree_mod_elem *tm;
1481 struct extent_buffer *eb_root = btrfs_root_node(root);
1483 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1484 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1485 level = tm->old_root.level;
1487 level = btrfs_header_level(eb_root);
1489 free_extent_buffer(eb_root);
1494 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1495 struct btrfs_root *root,
1496 struct extent_buffer *buf)
1498 if (btrfs_is_testing(root->fs_info))
1501 /* ensure we can see the force_cow */
1505 * We do not need to cow a block if
1506 * 1) this block is not created or changed in this transaction;
1507 * 2) this block does not belong to TREE_RELOC tree;
1508 * 3) the root is not forced COW.
1510 * What is forced COW:
1511 * when we create snapshot during committing the transaction,
1512 * after we've finished coping src root, we must COW the shared
1513 * block to ensure the metadata consistency.
1515 if (btrfs_header_generation(buf) == trans->transid &&
1516 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1517 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1518 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1519 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1525 * cows a single block, see __btrfs_cow_block for the real work.
1526 * This version of it has extra checks so that a block isn't COWed more than
1527 * once per transaction, as long as it hasn't been written yet
1529 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1530 struct btrfs_root *root, struct extent_buffer *buf,
1531 struct extent_buffer *parent, int parent_slot,
1532 struct extent_buffer **cow_ret)
1534 struct btrfs_fs_info *fs_info = root->fs_info;
1538 if (trans->transaction != fs_info->running_transaction)
1539 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1541 fs_info->running_transaction->transid);
1543 if (trans->transid != fs_info->generation)
1544 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1545 trans->transid, fs_info->generation);
1547 if (!should_cow_block(trans, root, buf)) {
1548 trans->dirty = true;
1553 search_start = buf->start & ~((u64)SZ_1G - 1);
1556 btrfs_set_lock_blocking(parent);
1557 btrfs_set_lock_blocking(buf);
1559 ret = __btrfs_cow_block(trans, root, buf, parent,
1560 parent_slot, cow_ret, search_start, 0);
1562 trace_btrfs_cow_block(root, buf, *cow_ret);
1568 * helper function for defrag to decide if two blocks pointed to by a
1569 * node are actually close by
1571 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1573 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1575 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1581 * compare two keys in a memcmp fashion
1583 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1585 struct btrfs_key k1;
1587 btrfs_disk_key_to_cpu(&k1, disk);
1589 return btrfs_comp_cpu_keys(&k1, k2);
1593 * same as comp_keys only with two btrfs_key's
1595 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1597 if (k1->objectid > k2->objectid)
1599 if (k1->objectid < k2->objectid)
1601 if (k1->type > k2->type)
1603 if (k1->type < k2->type)
1605 if (k1->offset > k2->offset)
1607 if (k1->offset < k2->offset)
1613 * this is used by the defrag code to go through all the
1614 * leaves pointed to by a node and reallocate them so that
1615 * disk order is close to key order
1617 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1618 struct btrfs_root *root, struct extent_buffer *parent,
1619 int start_slot, u64 *last_ret,
1620 struct btrfs_key *progress)
1622 struct btrfs_fs_info *fs_info = root->fs_info;
1623 struct extent_buffer *cur;
1626 u64 search_start = *last_ret;
1636 int progress_passed = 0;
1637 struct btrfs_disk_key disk_key;
1639 parent_level = btrfs_header_level(parent);
1641 WARN_ON(trans->transaction != fs_info->running_transaction);
1642 WARN_ON(trans->transid != fs_info->generation);
1644 parent_nritems = btrfs_header_nritems(parent);
1645 blocksize = fs_info->nodesize;
1646 end_slot = parent_nritems - 1;
1648 if (parent_nritems <= 1)
1651 btrfs_set_lock_blocking(parent);
1653 for (i = start_slot; i <= end_slot; i++) {
1656 btrfs_node_key(parent, &disk_key, i);
1657 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1660 progress_passed = 1;
1661 blocknr = btrfs_node_blockptr(parent, i);
1662 gen = btrfs_node_ptr_generation(parent, i);
1663 if (last_block == 0)
1664 last_block = blocknr;
1667 other = btrfs_node_blockptr(parent, i - 1);
1668 close = close_blocks(blocknr, other, blocksize);
1670 if (!close && i < end_slot) {
1671 other = btrfs_node_blockptr(parent, i + 1);
1672 close = close_blocks(blocknr, other, blocksize);
1675 last_block = blocknr;
1679 cur = find_extent_buffer(fs_info, blocknr);
1681 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1684 if (!cur || !uptodate) {
1686 cur = read_tree_block(fs_info, blocknr, gen);
1688 return PTR_ERR(cur);
1689 } else if (!extent_buffer_uptodate(cur)) {
1690 free_extent_buffer(cur);
1693 } else if (!uptodate) {
1694 err = btrfs_read_buffer(cur, gen);
1696 free_extent_buffer(cur);
1701 if (search_start == 0)
1702 search_start = last_block;
1704 btrfs_tree_lock(cur);
1705 btrfs_set_lock_blocking(cur);
1706 err = __btrfs_cow_block(trans, root, cur, parent, i,
1709 (end_slot - i) * blocksize));
1711 btrfs_tree_unlock(cur);
1712 free_extent_buffer(cur);
1715 search_start = cur->start;
1716 last_block = cur->start;
1717 *last_ret = search_start;
1718 btrfs_tree_unlock(cur);
1719 free_extent_buffer(cur);
1725 * search for key in the extent_buffer. The items start at offset p,
1726 * and they are item_size apart. There are 'max' items in p.
1728 * the slot in the array is returned via slot, and it points to
1729 * the place where you would insert key if it is not found in
1732 * slot may point to max if the key is bigger than all of the keys
1734 static noinline int generic_bin_search(struct extent_buffer *eb,
1736 int item_size, struct btrfs_key *key,
1743 struct btrfs_disk_key *tmp = NULL;
1744 struct btrfs_disk_key unaligned;
1745 unsigned long offset;
1747 unsigned long map_start = 0;
1748 unsigned long map_len = 0;
1752 btrfs_err(eb->fs_info,
1753 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1754 __func__, low, high, eb->start,
1755 btrfs_header_owner(eb), btrfs_header_level(eb));
1759 while (low < high) {
1760 mid = (low + high) / 2;
1761 offset = p + mid * item_size;
1763 if (!kaddr || offset < map_start ||
1764 (offset + sizeof(struct btrfs_disk_key)) >
1765 map_start + map_len) {
1767 err = map_private_extent_buffer(eb, offset,
1768 sizeof(struct btrfs_disk_key),
1769 &kaddr, &map_start, &map_len);
1772 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1774 } else if (err == 1) {
1775 read_extent_buffer(eb, &unaligned,
1776 offset, sizeof(unaligned));
1783 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1786 ret = comp_keys(tmp, key);
1802 * simple bin_search frontend that does the right thing for
1805 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1806 int level, int *slot)
1809 return generic_bin_search(eb,
1810 offsetof(struct btrfs_leaf, items),
1811 sizeof(struct btrfs_item),
1812 key, btrfs_header_nritems(eb),
1815 return generic_bin_search(eb,
1816 offsetof(struct btrfs_node, ptrs),
1817 sizeof(struct btrfs_key_ptr),
1818 key, btrfs_header_nritems(eb),
1822 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1823 int level, int *slot)
1825 return bin_search(eb, key, level, slot);
1828 static void root_add_used(struct btrfs_root *root, u32 size)
1830 spin_lock(&root->accounting_lock);
1831 btrfs_set_root_used(&root->root_item,
1832 btrfs_root_used(&root->root_item) + size);
1833 spin_unlock(&root->accounting_lock);
1836 static void root_sub_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 /* given a node and slot number, this reads the blocks it points to. The
1845 * extent buffer is returned with a reference taken (but unlocked).
1847 static noinline struct extent_buffer *
1848 read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent,
1851 int level = btrfs_header_level(parent);
1852 struct extent_buffer *eb;
1854 if (slot < 0 || slot >= btrfs_header_nritems(parent))
1855 return ERR_PTR(-ENOENT);
1859 eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot),
1860 btrfs_node_ptr_generation(parent, slot));
1861 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1862 free_extent_buffer(eb);
1870 * node level balancing, used to make sure nodes are in proper order for
1871 * item deletion. We balance from the top down, so we have to make sure
1872 * that a deletion won't leave an node completely empty later on.
1874 static noinline int balance_level(struct btrfs_trans_handle *trans,
1875 struct btrfs_root *root,
1876 struct btrfs_path *path, int level)
1878 struct btrfs_fs_info *fs_info = root->fs_info;
1879 struct extent_buffer *right = NULL;
1880 struct extent_buffer *mid;
1881 struct extent_buffer *left = NULL;
1882 struct extent_buffer *parent = NULL;
1886 int orig_slot = path->slots[level];
1892 mid = path->nodes[level];
1894 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1895 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1896 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1898 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1900 if (level < BTRFS_MAX_LEVEL - 1) {
1901 parent = path->nodes[level + 1];
1902 pslot = path->slots[level + 1];
1906 * deal with the case where there is only one pointer in the root
1907 * by promoting the node below to a root
1910 struct extent_buffer *child;
1912 if (btrfs_header_nritems(mid) != 1)
1915 /* promote the child to a root */
1916 child = read_node_slot(fs_info, mid, 0);
1917 if (IS_ERR(child)) {
1918 ret = PTR_ERR(child);
1919 btrfs_handle_fs_error(fs_info, ret, NULL);
1923 btrfs_tree_lock(child);
1924 btrfs_set_lock_blocking(child);
1925 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1927 btrfs_tree_unlock(child);
1928 free_extent_buffer(child);
1932 tree_mod_log_set_root_pointer(root, child, 1);
1933 rcu_assign_pointer(root->node, child);
1935 add_root_to_dirty_list(root);
1936 btrfs_tree_unlock(child);
1938 path->locks[level] = 0;
1939 path->nodes[level] = NULL;
1940 clean_tree_block(trans, fs_info, mid);
1941 btrfs_tree_unlock(mid);
1942 /* once for the path */
1943 free_extent_buffer(mid);
1945 root_sub_used(root, mid->len);
1946 btrfs_free_tree_block(trans, root, mid, 0, 1);
1947 /* once for the root ptr */
1948 free_extent_buffer_stale(mid);
1951 if (btrfs_header_nritems(mid) >
1952 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1955 left = read_node_slot(fs_info, parent, pslot - 1);
1960 btrfs_tree_lock(left);
1961 btrfs_set_lock_blocking(left);
1962 wret = btrfs_cow_block(trans, root, left,
1963 parent, pslot - 1, &left);
1970 right = read_node_slot(fs_info, parent, pslot + 1);
1975 btrfs_tree_lock(right);
1976 btrfs_set_lock_blocking(right);
1977 wret = btrfs_cow_block(trans, root, right,
1978 parent, pslot + 1, &right);
1985 /* first, try to make some room in the middle buffer */
1987 orig_slot += btrfs_header_nritems(left);
1988 wret = push_node_left(trans, fs_info, left, mid, 1);
1994 * then try to empty the right most buffer into the middle
1997 wret = push_node_left(trans, fs_info, mid, right, 1);
1998 if (wret < 0 && wret != -ENOSPC)
2000 if (btrfs_header_nritems(right) == 0) {
2001 clean_tree_block(trans, fs_info, right);
2002 btrfs_tree_unlock(right);
2003 del_ptr(root, path, level + 1, pslot + 1);
2004 root_sub_used(root, right->len);
2005 btrfs_free_tree_block(trans, root, right, 0, 1);
2006 free_extent_buffer_stale(right);
2009 struct btrfs_disk_key right_key;
2010 btrfs_node_key(right, &right_key, 0);
2011 tree_mod_log_set_node_key(fs_info, parent,
2013 btrfs_set_node_key(parent, &right_key, pslot + 1);
2014 btrfs_mark_buffer_dirty(parent);
2017 if (btrfs_header_nritems(mid) == 1) {
2019 * we're not allowed to leave a node with one item in the
2020 * tree during a delete. A deletion from lower in the tree
2021 * could try to delete the only pointer in this node.
2022 * So, pull some keys from the left.
2023 * There has to be a left pointer at this point because
2024 * otherwise we would have pulled some pointers from the
2029 btrfs_handle_fs_error(fs_info, ret, NULL);
2032 wret = balance_node_right(trans, fs_info, mid, left);
2038 wret = push_node_left(trans, fs_info, left, mid, 1);
2044 if (btrfs_header_nritems(mid) == 0) {
2045 clean_tree_block(trans, fs_info, mid);
2046 btrfs_tree_unlock(mid);
2047 del_ptr(root, path, level + 1, pslot);
2048 root_sub_used(root, mid->len);
2049 btrfs_free_tree_block(trans, root, mid, 0, 1);
2050 free_extent_buffer_stale(mid);
2053 /* update the parent key to reflect our changes */
2054 struct btrfs_disk_key mid_key;
2055 btrfs_node_key(mid, &mid_key, 0);
2056 tree_mod_log_set_node_key(fs_info, parent, pslot, 0);
2057 btrfs_set_node_key(parent, &mid_key, pslot);
2058 btrfs_mark_buffer_dirty(parent);
2061 /* update the path */
2063 if (btrfs_header_nritems(left) > orig_slot) {
2064 extent_buffer_get(left);
2065 /* left was locked after cow */
2066 path->nodes[level] = left;
2067 path->slots[level + 1] -= 1;
2068 path->slots[level] = orig_slot;
2070 btrfs_tree_unlock(mid);
2071 free_extent_buffer(mid);
2074 orig_slot -= btrfs_header_nritems(left);
2075 path->slots[level] = orig_slot;
2078 /* double check we haven't messed things up */
2080 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2084 btrfs_tree_unlock(right);
2085 free_extent_buffer(right);
2088 if (path->nodes[level] != left)
2089 btrfs_tree_unlock(left);
2090 free_extent_buffer(left);
2095 /* Node balancing for insertion. Here we only split or push nodes around
2096 * when they are completely full. This is also done top down, so we
2097 * have to be pessimistic.
2099 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct btrfs_path *path, int level)
2103 struct btrfs_fs_info *fs_info = root->fs_info;
2104 struct extent_buffer *right = NULL;
2105 struct extent_buffer *mid;
2106 struct extent_buffer *left = NULL;
2107 struct extent_buffer *parent = NULL;
2111 int orig_slot = path->slots[level];
2116 mid = path->nodes[level];
2117 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2119 if (level < BTRFS_MAX_LEVEL - 1) {
2120 parent = path->nodes[level + 1];
2121 pslot = path->slots[level + 1];
2127 left = read_node_slot(fs_info, 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(fs_info) - 1) {
2142 ret = btrfs_cow_block(trans, root, left, parent,
2147 wret = push_node_left(trans, fs_info,
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(fs_info, parent, pslot, 0);
2158 btrfs_set_node_key(parent, &disk_key, pslot);
2159 btrfs_mark_buffer_dirty(parent);
2160 if (btrfs_header_nritems(left) > orig_slot) {
2161 path->nodes[level] = left;
2162 path->slots[level + 1] -= 1;
2163 path->slots[level] = orig_slot;
2164 btrfs_tree_unlock(mid);
2165 free_extent_buffer(mid);
2168 btrfs_header_nritems(left);
2169 path->slots[level] = orig_slot;
2170 btrfs_tree_unlock(left);
2171 free_extent_buffer(left);
2175 btrfs_tree_unlock(left);
2176 free_extent_buffer(left);
2178 right = read_node_slot(fs_info, parent, pslot + 1);
2183 * then try to empty the right most buffer into the middle
2188 btrfs_tree_lock(right);
2189 btrfs_set_lock_blocking(right);
2191 right_nr = btrfs_header_nritems(right);
2192 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2195 ret = btrfs_cow_block(trans, root, right,
2201 wret = balance_node_right(trans, fs_info,
2208 struct btrfs_disk_key disk_key;
2210 btrfs_node_key(right, &disk_key, 0);
2211 tree_mod_log_set_node_key(fs_info, parent,
2213 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2214 btrfs_mark_buffer_dirty(parent);
2216 if (btrfs_header_nritems(mid) <= orig_slot) {
2217 path->nodes[level] = right;
2218 path->slots[level + 1] += 1;
2219 path->slots[level] = orig_slot -
2220 btrfs_header_nritems(mid);
2221 btrfs_tree_unlock(mid);
2222 free_extent_buffer(mid);
2224 btrfs_tree_unlock(right);
2225 free_extent_buffer(right);
2229 btrfs_tree_unlock(right);
2230 free_extent_buffer(right);
2236 * readahead one full node of leaves, finding things that are close
2237 * to the block in 'slot', and triggering ra on them.
2239 static void reada_for_search(struct btrfs_fs_info *fs_info,
2240 struct btrfs_path *path,
2241 int level, int slot, u64 objectid)
2243 struct extent_buffer *node;
2244 struct btrfs_disk_key disk_key;
2249 struct extent_buffer *eb;
2257 if (!path->nodes[level])
2260 node = path->nodes[level];
2262 search = btrfs_node_blockptr(node, slot);
2263 blocksize = fs_info->nodesize;
2264 eb = find_extent_buffer(fs_info, search);
2266 free_extent_buffer(eb);
2272 nritems = btrfs_header_nritems(node);
2276 if (path->reada == READA_BACK) {
2280 } else if (path->reada == READA_FORWARD) {
2285 if (path->reada == READA_BACK && objectid) {
2286 btrfs_node_key(node, &disk_key, nr);
2287 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2290 search = btrfs_node_blockptr(node, nr);
2291 if ((search <= target && target - search <= 65536) ||
2292 (search > target && search - target <= 65536)) {
2293 readahead_tree_block(fs_info, search);
2297 if ((nread > 65536 || nscan > 32))
2302 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2303 struct btrfs_path *path, int level)
2307 struct extent_buffer *parent;
2308 struct extent_buffer *eb;
2313 parent = path->nodes[level + 1];
2317 nritems = btrfs_header_nritems(parent);
2318 slot = path->slots[level + 1];
2321 block1 = btrfs_node_blockptr(parent, slot - 1);
2322 gen = btrfs_node_ptr_generation(parent, slot - 1);
2323 eb = find_extent_buffer(fs_info, block1);
2325 * if we get -eagain from btrfs_buffer_uptodate, we
2326 * don't want to return eagain here. That will loop
2329 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2331 free_extent_buffer(eb);
2333 if (slot + 1 < nritems) {
2334 block2 = btrfs_node_blockptr(parent, slot + 1);
2335 gen = btrfs_node_ptr_generation(parent, slot + 1);
2336 eb = find_extent_buffer(fs_info, block2);
2337 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2339 free_extent_buffer(eb);
2343 readahead_tree_block(fs_info, block1);
2345 readahead_tree_block(fs_info, block2);
2350 * when we walk down the tree, it is usually safe to unlock the higher layers
2351 * in the tree. The exceptions are when our path goes through slot 0, because
2352 * operations on the tree might require changing key pointers higher up in the
2355 * callers might also have set path->keep_locks, which tells this code to keep
2356 * the lock if the path points to the last slot in the block. This is part of
2357 * walking through the tree, and selecting the next slot in the higher block.
2359 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2360 * if lowest_unlock is 1, level 0 won't be unlocked
2362 static noinline void unlock_up(struct btrfs_path *path, int level,
2363 int lowest_unlock, int min_write_lock_level,
2364 int *write_lock_level)
2367 int skip_level = level;
2369 struct extent_buffer *t;
2371 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2372 if (!path->nodes[i])
2374 if (!path->locks[i])
2376 if (!no_skips && path->slots[i] == 0) {
2380 if (!no_skips && path->keep_locks) {
2383 nritems = btrfs_header_nritems(t);
2384 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2389 if (skip_level < i && i >= lowest_unlock)
2393 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2394 btrfs_tree_unlock_rw(t, path->locks[i]);
2396 if (write_lock_level &&
2397 i > min_write_lock_level &&
2398 i <= *write_lock_level) {
2399 *write_lock_level = i - 1;
2406 * This releases any locks held in the path starting at level and
2407 * going all the way up to the root.
2409 * btrfs_search_slot will keep the lock held on higher nodes in a few
2410 * corner cases, such as COW of the block at slot zero in the node. This
2411 * ignores those rules, and it should only be called when there are no
2412 * more updates to be done higher up in the tree.
2414 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2418 if (path->keep_locks)
2421 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2422 if (!path->nodes[i])
2424 if (!path->locks[i])
2426 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2432 * helper function for btrfs_search_slot. The goal is to find a block
2433 * in cache without setting the path to blocking. If we find the block
2434 * we return zero and the path is unchanged.
2436 * If we can't find the block, we set the path blocking and do some
2437 * reada. -EAGAIN is returned and the search must be repeated.
2440 read_block_for_search(struct btrfs_trans_handle *trans,
2441 struct btrfs_root *root, struct btrfs_path *p,
2442 struct extent_buffer **eb_ret, int level, int slot,
2443 struct btrfs_key *key, u64 time_seq)
2445 struct btrfs_fs_info *fs_info = root->fs_info;
2448 struct extent_buffer *b = *eb_ret;
2449 struct extent_buffer *tmp;
2452 blocknr = btrfs_node_blockptr(b, slot);
2453 gen = btrfs_node_ptr_generation(b, slot);
2455 tmp = find_extent_buffer(fs_info, blocknr);
2457 /* first we do an atomic uptodate check */
2458 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2463 /* the pages were up to date, but we failed
2464 * the generation number check. Do a full
2465 * read for the generation number that is correct.
2466 * We must do this without dropping locks so
2467 * we can trust our generation number
2469 btrfs_set_path_blocking(p);
2471 /* now we're allowed to do a blocking uptodate check */
2472 ret = btrfs_read_buffer(tmp, gen);
2477 free_extent_buffer(tmp);
2478 btrfs_release_path(p);
2483 * reduce lock contention at high levels
2484 * of the btree by dropping locks before
2485 * we read. Don't release the lock on the current
2486 * level because we need to walk this node to figure
2487 * out which blocks to read.
2489 btrfs_unlock_up_safe(p, level + 1);
2490 btrfs_set_path_blocking(p);
2492 free_extent_buffer(tmp);
2493 if (p->reada != READA_NONE)
2494 reada_for_search(fs_info, p, level, slot, key->objectid);
2496 btrfs_release_path(p);
2499 tmp = read_tree_block(fs_info, blocknr, 0);
2502 * If the read above didn't mark this buffer up to date,
2503 * it will never end up being up to date. Set ret to EIO now
2504 * and give up so that our caller doesn't loop forever
2507 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2509 free_extent_buffer(tmp);
2517 * helper function for btrfs_search_slot. This does all of the checks
2518 * for node-level blocks and does any balancing required based on
2521 * If no extra work was required, zero is returned. If we had to
2522 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2526 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2527 struct btrfs_root *root, struct btrfs_path *p,
2528 struct extent_buffer *b, int level, int ins_len,
2529 int *write_lock_level)
2531 struct btrfs_fs_info *fs_info = root->fs_info;
2534 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2535 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2538 if (*write_lock_level < level + 1) {
2539 *write_lock_level = level + 1;
2540 btrfs_release_path(p);
2544 btrfs_set_path_blocking(p);
2545 reada_for_balance(fs_info, p, level);
2546 sret = split_node(trans, root, p, level);
2547 btrfs_clear_path_blocking(p, NULL, 0);
2554 b = p->nodes[level];
2555 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2556 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2559 if (*write_lock_level < level + 1) {
2560 *write_lock_level = level + 1;
2561 btrfs_release_path(p);
2565 btrfs_set_path_blocking(p);
2566 reada_for_balance(fs_info, p, level);
2567 sret = balance_level(trans, root, p, level);
2568 btrfs_clear_path_blocking(p, NULL, 0);
2574 b = p->nodes[level];
2576 btrfs_release_path(p);
2579 BUG_ON(btrfs_header_nritems(b) == 1);
2589 static void key_search_validate(struct extent_buffer *b,
2590 struct btrfs_key *key,
2593 #ifdef CONFIG_BTRFS_ASSERT
2594 struct btrfs_disk_key disk_key;
2596 btrfs_cpu_key_to_disk(&disk_key, key);
2599 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2600 offsetof(struct btrfs_leaf, items[0].key),
2603 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2604 offsetof(struct btrfs_node, ptrs[0].key),
2609 static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2610 int level, int *prev_cmp, int *slot)
2612 if (*prev_cmp != 0) {
2613 *prev_cmp = bin_search(b, key, level, slot);
2617 key_search_validate(b, key, level);
2623 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2624 u64 iobjectid, u64 ioff, u8 key_type,
2625 struct btrfs_key *found_key)
2628 struct btrfs_key key;
2629 struct extent_buffer *eb;
2634 key.type = key_type;
2635 key.objectid = iobjectid;
2638 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2642 eb = path->nodes[0];
2643 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2644 ret = btrfs_next_leaf(fs_root, path);
2647 eb = path->nodes[0];
2650 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2651 if (found_key->type != key.type ||
2652 found_key->objectid != key.objectid)
2659 * look for key in the tree. path is filled in with nodes along the way
2660 * if key is found, we return zero and you can find the item in the leaf
2661 * level of the path (level 0)
2663 * If the key isn't found, the path points to the slot where it should
2664 * be inserted, and 1 is returned. If there are other errors during the
2665 * search a negative error number is returned.
2667 * if ins_len > 0, nodes and leaves will be split as we walk down the
2668 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2671 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2672 *root, struct btrfs_key *key, struct btrfs_path *p, int
2675 struct btrfs_fs_info *fs_info = root->fs_info;
2676 struct extent_buffer *b;
2681 int lowest_unlock = 1;
2683 /* everything at write_lock_level or lower must be write locked */
2684 int write_lock_level = 0;
2685 u8 lowest_level = 0;
2686 int min_write_lock_level;
2689 lowest_level = p->lowest_level;
2690 WARN_ON(lowest_level && ins_len > 0);
2691 WARN_ON(p->nodes[0] != NULL);
2692 BUG_ON(!cow && ins_len);
2697 /* when we are removing items, we might have to go up to level
2698 * two as we update tree pointers Make sure we keep write
2699 * for those levels as well
2701 write_lock_level = 2;
2702 } else if (ins_len > 0) {
2704 * for inserting items, make sure we have a write lock on
2705 * level 1 so we can update keys
2707 write_lock_level = 1;
2711 write_lock_level = -1;
2713 if (cow && (p->keep_locks || p->lowest_level))
2714 write_lock_level = BTRFS_MAX_LEVEL;
2716 min_write_lock_level = write_lock_level;
2721 * we try very hard to do read locks on the root
2723 root_lock = BTRFS_READ_LOCK;
2725 if (p->search_commit_root) {
2727 * the commit roots are read only
2728 * so we always do read locks
2730 if (p->need_commit_sem)
2731 down_read(&fs_info->commit_root_sem);
2732 b = root->commit_root;
2733 extent_buffer_get(b);
2734 level = btrfs_header_level(b);
2735 if (p->need_commit_sem)
2736 up_read(&fs_info->commit_root_sem);
2737 if (!p->skip_locking)
2738 btrfs_tree_read_lock(b);
2740 if (p->skip_locking) {
2741 b = btrfs_root_node(root);
2742 level = btrfs_header_level(b);
2744 /* we don't know the level of the root node
2745 * until we actually have it read locked
2747 b = btrfs_read_lock_root_node(root);
2748 level = btrfs_header_level(b);
2749 if (level <= write_lock_level) {
2750 /* whoops, must trade for write lock */
2751 btrfs_tree_read_unlock(b);
2752 free_extent_buffer(b);
2753 b = btrfs_lock_root_node(root);
2754 root_lock = BTRFS_WRITE_LOCK;
2756 /* the level might have changed, check again */
2757 level = btrfs_header_level(b);
2761 p->nodes[level] = b;
2762 if (!p->skip_locking)
2763 p->locks[level] = root_lock;
2766 level = btrfs_header_level(b);
2769 * setup the path here so we can release it under lock
2770 * contention with the cow code
2774 * if we don't really need to cow this block
2775 * then we don't want to set the path blocking,
2776 * so we test it here
2778 if (!should_cow_block(trans, root, b)) {
2779 trans->dirty = true;
2784 * must have write locks on this node and the
2787 if (level > write_lock_level ||
2788 (level + 1 > write_lock_level &&
2789 level + 1 < BTRFS_MAX_LEVEL &&
2790 p->nodes[level + 1])) {
2791 write_lock_level = level + 1;
2792 btrfs_release_path(p);
2796 btrfs_set_path_blocking(p);
2797 err = btrfs_cow_block(trans, root, b,
2798 p->nodes[level + 1],
2799 p->slots[level + 1], &b);
2806 p->nodes[level] = b;
2807 btrfs_clear_path_blocking(p, NULL, 0);
2810 * we have a lock on b and as long as we aren't changing
2811 * the tree, there is no way to for the items in b to change.
2812 * It is safe to drop the lock on our parent before we
2813 * go through the expensive btree search on b.
2815 * If we're inserting or deleting (ins_len != 0), then we might
2816 * be changing slot zero, which may require changing the parent.
2817 * So, we can't drop the lock until after we know which slot
2818 * we're operating on.
2820 if (!ins_len && !p->keep_locks) {
2823 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2824 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2829 ret = key_search(b, key, level, &prev_cmp, &slot);
2835 if (ret && slot > 0) {
2839 p->slots[level] = slot;
2840 err = setup_nodes_for_search(trans, root, p, b, level,
2841 ins_len, &write_lock_level);
2848 b = p->nodes[level];
2849 slot = p->slots[level];
2852 * slot 0 is special, if we change the key
2853 * we have to update the parent pointer
2854 * which means we must have a write lock
2857 if (slot == 0 && ins_len &&
2858 write_lock_level < level + 1) {
2859 write_lock_level = level + 1;
2860 btrfs_release_path(p);
2864 unlock_up(p, level, lowest_unlock,
2865 min_write_lock_level, &write_lock_level);
2867 if (level == lowest_level) {
2873 err = read_block_for_search(trans, root, p,
2874 &b, level, slot, key, 0);
2882 if (!p->skip_locking) {
2883 level = btrfs_header_level(b);
2884 if (level <= write_lock_level) {
2885 err = btrfs_try_tree_write_lock(b);
2887 btrfs_set_path_blocking(p);
2889 btrfs_clear_path_blocking(p, b,
2892 p->locks[level] = BTRFS_WRITE_LOCK;
2894 err = btrfs_tree_read_lock_atomic(b);
2896 btrfs_set_path_blocking(p);
2897 btrfs_tree_read_lock(b);
2898 btrfs_clear_path_blocking(p, b,
2901 p->locks[level] = BTRFS_READ_LOCK;
2903 p->nodes[level] = b;
2906 p->slots[level] = slot;
2908 btrfs_leaf_free_space(fs_info, b) < ins_len) {
2909 if (write_lock_level < 1) {
2910 write_lock_level = 1;
2911 btrfs_release_path(p);
2915 btrfs_set_path_blocking(p);
2916 err = split_leaf(trans, root, key,
2917 p, ins_len, ret == 0);
2918 btrfs_clear_path_blocking(p, NULL, 0);
2926 if (!p->search_for_split)
2927 unlock_up(p, level, lowest_unlock,
2928 min_write_lock_level, &write_lock_level);
2935 * we don't really know what they plan on doing with the path
2936 * from here on, so for now just mark it as blocking
2938 if (!p->leave_spinning)
2939 btrfs_set_path_blocking(p);
2940 if (ret < 0 && !p->skip_release_on_error)
2941 btrfs_release_path(p);
2946 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2947 * current state of the tree together with the operations recorded in the tree
2948 * modification log to search for the key in a previous version of this tree, as
2949 * denoted by the time_seq parameter.
2951 * Naturally, there is no support for insert, delete or cow operations.
2953 * The resulting path and return value will be set up as if we called
2954 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2956 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2957 struct btrfs_path *p, u64 time_seq)
2959 struct btrfs_fs_info *fs_info = root->fs_info;
2960 struct extent_buffer *b;
2965 int lowest_unlock = 1;
2966 u8 lowest_level = 0;
2969 lowest_level = p->lowest_level;
2970 WARN_ON(p->nodes[0] != NULL);
2972 if (p->search_commit_root) {
2974 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2978 b = get_old_root(root, time_seq);
2979 level = btrfs_header_level(b);
2980 p->locks[level] = BTRFS_READ_LOCK;
2983 level = btrfs_header_level(b);
2984 p->nodes[level] = b;
2985 btrfs_clear_path_blocking(p, NULL, 0);
2988 * we have a lock on b and as long as we aren't changing
2989 * the tree, there is no way to for the items in b to change.
2990 * It is safe to drop the lock on our parent before we
2991 * go through the expensive btree search on b.
2993 btrfs_unlock_up_safe(p, level + 1);
2996 * Since we can unwind ebs we want to do a real search every
3000 ret = key_search(b, key, level, &prev_cmp, &slot);
3004 if (ret && slot > 0) {
3008 p->slots[level] = slot;
3009 unlock_up(p, level, lowest_unlock, 0, NULL);
3011 if (level == lowest_level) {
3017 err = read_block_for_search(NULL, root, p, &b, level,
3018 slot, key, time_seq);
3026 level = btrfs_header_level(b);
3027 err = btrfs_tree_read_lock_atomic(b);
3029 btrfs_set_path_blocking(p);
3030 btrfs_tree_read_lock(b);
3031 btrfs_clear_path_blocking(p, b,
3034 b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3039 p->locks[level] = BTRFS_READ_LOCK;
3040 p->nodes[level] = b;
3042 p->slots[level] = slot;
3043 unlock_up(p, level, lowest_unlock, 0, NULL);
3049 if (!p->leave_spinning)
3050 btrfs_set_path_blocking(p);
3052 btrfs_release_path(p);
3058 * helper to use instead of search slot if no exact match is needed but
3059 * instead the next or previous item should be returned.
3060 * When find_higher is true, the next higher item is returned, the next lower
3062 * When return_any and find_higher are both true, and no higher item is found,
3063 * return the next lower instead.
3064 * When return_any is true and find_higher is false, and no lower item is found,
3065 * return the next higher instead.
3066 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3069 int btrfs_search_slot_for_read(struct btrfs_root *root,
3070 struct btrfs_key *key, struct btrfs_path *p,
3071 int find_higher, int return_any)
3074 struct extent_buffer *leaf;
3077 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3081 * a return value of 1 means the path is at the position where the
3082 * item should be inserted. Normally this is the next bigger item,
3083 * but in case the previous item is the last in a leaf, path points
3084 * to the first free slot in the previous leaf, i.e. at an invalid
3090 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3091 ret = btrfs_next_leaf(root, p);
3097 * no higher item found, return the next
3102 btrfs_release_path(p);
3106 if (p->slots[0] == 0) {
3107 ret = btrfs_prev_leaf(root, p);
3112 if (p->slots[0] == btrfs_header_nritems(leaf))
3119 * no lower item found, return the next
3124 btrfs_release_path(p);
3134 * adjust the pointers going up the tree, starting at level
3135 * making sure the right key of each node is points to 'key'.
3136 * This is used after shifting pointers to the left, so it stops
3137 * fixing up pointers when a given leaf/node is not in slot 0 of the
3141 static void fixup_low_keys(struct btrfs_fs_info *fs_info,
3142 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(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_fs_info *fs_info,
3168 struct btrfs_path *path,
3169 struct btrfs_key *new_key)
3171 struct btrfs_disk_key disk_key;
3172 struct extent_buffer *eb;
3175 eb = path->nodes[0];
3176 slot = path->slots[0];
3178 btrfs_item_key(eb, &disk_key, slot - 1);
3179 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3181 if (slot < btrfs_header_nritems(eb) - 1) {
3182 btrfs_item_key(eb, &disk_key, slot + 1);
3183 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3186 btrfs_cpu_key_to_disk(&disk_key, new_key);
3187 btrfs_set_item_key(eb, &disk_key, slot);
3188 btrfs_mark_buffer_dirty(eb);
3190 fixup_low_keys(fs_info, path, &disk_key, 1);
3194 * try to push data from one node into the next node left in the
3197 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3198 * error, and > 0 if there was no room in the left hand block.
3200 static int push_node_left(struct btrfs_trans_handle *trans,
3201 struct btrfs_fs_info *fs_info,
3202 struct extent_buffer *dst,
3203 struct extent_buffer *src, int empty)
3210 src_nritems = btrfs_header_nritems(src);
3211 dst_nritems = btrfs_header_nritems(dst);
3212 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3213 WARN_ON(btrfs_header_generation(src) != trans->transid);
3214 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3216 if (!empty && src_nritems <= 8)
3219 if (push_items <= 0)
3223 push_items = min(src_nritems, push_items);
3224 if (push_items < src_nritems) {
3225 /* leave at least 8 pointers in the node if
3226 * we aren't going to empty it
3228 if (src_nritems - push_items < 8) {
3229 if (push_items <= 8)
3235 push_items = min(src_nritems - 8, push_items);
3237 ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0,
3240 btrfs_abort_transaction(trans, ret);
3243 copy_extent_buffer(dst, src,
3244 btrfs_node_key_ptr_offset(dst_nritems),
3245 btrfs_node_key_ptr_offset(0),
3246 push_items * sizeof(struct btrfs_key_ptr));
3248 if (push_items < src_nritems) {
3250 * don't call tree_mod_log_eb_move here, key removal was already
3251 * fully logged by tree_mod_log_eb_copy above.
3253 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3254 btrfs_node_key_ptr_offset(push_items),
3255 (src_nritems - push_items) *
3256 sizeof(struct btrfs_key_ptr));
3258 btrfs_set_header_nritems(src, src_nritems - push_items);
3259 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3260 btrfs_mark_buffer_dirty(src);
3261 btrfs_mark_buffer_dirty(dst);
3267 * try to push data from one node into the next node right in the
3270 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3271 * error, and > 0 if there was no room in the right hand block.
3273 * this will only push up to 1/2 the contents of the left node over
3275 static int balance_node_right(struct btrfs_trans_handle *trans,
3276 struct btrfs_fs_info *fs_info,
3277 struct extent_buffer *dst,
3278 struct extent_buffer *src)
3286 WARN_ON(btrfs_header_generation(src) != trans->transid);
3287 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3289 src_nritems = btrfs_header_nritems(src);
3290 dst_nritems = btrfs_header_nritems(dst);
3291 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3292 if (push_items <= 0)
3295 if (src_nritems < 4)
3298 max_push = src_nritems / 2 + 1;
3299 /* don't try to empty the node */
3300 if (max_push >= src_nritems)
3303 if (max_push < push_items)
3304 push_items = max_push;
3306 tree_mod_log_eb_move(fs_info, dst, push_items, 0, dst_nritems);
3307 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3308 btrfs_node_key_ptr_offset(0),
3310 sizeof(struct btrfs_key_ptr));
3312 ret = tree_mod_log_eb_copy(fs_info, dst, src, 0,
3313 src_nritems - push_items, push_items);
3315 btrfs_abort_transaction(trans, ret);
3318 copy_extent_buffer(dst, src,
3319 btrfs_node_key_ptr_offset(0),
3320 btrfs_node_key_ptr_offset(src_nritems - push_items),
3321 push_items * sizeof(struct btrfs_key_ptr));
3323 btrfs_set_header_nritems(src, src_nritems - push_items);
3324 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3326 btrfs_mark_buffer_dirty(src);
3327 btrfs_mark_buffer_dirty(dst);
3333 * helper function to insert a new root level in the tree.
3334 * A new node is allocated, and a single item is inserted to
3335 * point to the existing root
3337 * returns zero on success or < 0 on failure.
3339 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3340 struct btrfs_root *root,
3341 struct btrfs_path *path, int level)
3343 struct btrfs_fs_info *fs_info = root->fs_info;
3345 struct extent_buffer *lower;
3346 struct extent_buffer *c;
3347 struct extent_buffer *old;
3348 struct btrfs_disk_key lower_key;
3350 BUG_ON(path->nodes[level]);
3351 BUG_ON(path->nodes[level-1] != root->node);
3353 lower = path->nodes[level-1];
3355 btrfs_item_key(lower, &lower_key, 0);
3357 btrfs_node_key(lower, &lower_key, 0);
3359 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3360 &lower_key, level, root->node->start, 0);
3364 root_add_used(root, fs_info->nodesize);
3366 memzero_extent_buffer(c, 0, sizeof(struct btrfs_header));
3367 btrfs_set_header_nritems(c, 1);
3368 btrfs_set_header_level(c, level);
3369 btrfs_set_header_bytenr(c, c->start);
3370 btrfs_set_header_generation(c, trans->transid);
3371 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3372 btrfs_set_header_owner(c, root->root_key.objectid);
3374 write_extent_buffer_fsid(c, fs_info->fsid);
3375 write_extent_buffer_chunk_tree_uuid(c, fs_info->chunk_tree_uuid);
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_BLOCKING;
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_fs_info *fs_info, 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(fs_info));
3423 if (slot != nritems) {
3425 tree_mod_log_eb_move(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(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 btrfs_fs_info *fs_info = root->fs_info;
3459 struct extent_buffer *c;
3460 struct extent_buffer *split;
3461 struct btrfs_disk_key disk_key;
3466 c = path->nodes[level];
3467 WARN_ON(btrfs_header_generation(c) != trans->transid);
3468 if (c == root->node) {
3470 * trying to split the root, lets make a new one
3472 * tree mod log: We don't log_removal old root in
3473 * insert_new_root, because that root buffer will be kept as a
3474 * normal node. We are going to log removal of half of the
3475 * elements below with tree_mod_log_eb_copy. We're holding a
3476 * tree lock on the buffer, which is why we cannot race with
3477 * other tree_mod_log users.
3479 ret = insert_new_root(trans, root, path, level + 1);
3483 ret = push_nodes_for_insert(trans, root, path, level);
3484 c = path->nodes[level];
3485 if (!ret && btrfs_header_nritems(c) <
3486 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3492 c_nritems = btrfs_header_nritems(c);
3493 mid = (c_nritems + 1) / 2;
3494 btrfs_node_key(c, &disk_key, mid);
3496 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3497 &disk_key, level, c->start, 0);
3499 return PTR_ERR(split);
3501 root_add_used(root, fs_info->nodesize);
3503 memzero_extent_buffer(split, 0, sizeof(struct btrfs_header));
3504 btrfs_set_header_level(split, btrfs_header_level(c));
3505 btrfs_set_header_bytenr(split, split->start);
3506 btrfs_set_header_generation(split, trans->transid);
3507 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3508 btrfs_set_header_owner(split, root->root_key.objectid);
3509 write_extent_buffer_fsid(split, fs_info->fsid);
3510 write_extent_buffer_chunk_tree_uuid(split, fs_info->chunk_tree_uuid);
3512 ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid);
3514 btrfs_abort_transaction(trans, ret);
3517 copy_extent_buffer(split, c,
3518 btrfs_node_key_ptr_offset(0),
3519 btrfs_node_key_ptr_offset(mid),
3520 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3521 btrfs_set_header_nritems(split, c_nritems - mid);
3522 btrfs_set_header_nritems(c, mid);
3525 btrfs_mark_buffer_dirty(c);
3526 btrfs_mark_buffer_dirty(split);
3528 insert_ptr(trans, fs_info, path, &disk_key, split->start,
3529 path->slots[level + 1] + 1, level + 1);
3531 if (path->slots[level] >= mid) {
3532 path->slots[level] -= mid;
3533 btrfs_tree_unlock(c);
3534 free_extent_buffer(c);
3535 path->nodes[level] = split;
3536 path->slots[level + 1] += 1;
3538 btrfs_tree_unlock(split);
3539 free_extent_buffer(split);
3545 * how many bytes are required to store the items in a leaf. start
3546 * and nr indicate which items in the leaf to check. This totals up the
3547 * space used both by the item structs and the item data
3549 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3551 struct btrfs_item *start_item;
3552 struct btrfs_item *end_item;
3553 struct btrfs_map_token token;
3555 int nritems = btrfs_header_nritems(l);
3556 int end = min(nritems, start + nr) - 1;
3560 btrfs_init_map_token(&token);
3561 start_item = btrfs_item_nr(start);
3562 end_item = btrfs_item_nr(end);
3563 data_len = btrfs_token_item_offset(l, start_item, &token) +
3564 btrfs_token_item_size(l, start_item, &token);
3565 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3566 data_len += sizeof(struct btrfs_item) * nr;
3567 WARN_ON(data_len < 0);
3572 * The space between the end of the leaf items and
3573 * the start of the leaf data. IOW, how much room
3574 * the leaf has left for both items and data
3576 noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info,
3577 struct extent_buffer *leaf)
3579 int nritems = btrfs_header_nritems(leaf);
3582 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3585 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3587 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3588 leaf_space_used(leaf, 0, nritems), nritems);
3594 * min slot controls the lowest index we're willing to push to the
3595 * right. We'll push up to and including min_slot, but no lower
3597 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3598 struct btrfs_fs_info *fs_info,
3599 struct btrfs_path *path,
3600 int data_size, int empty,
3601 struct extent_buffer *right,
3602 int free_space, u32 left_nritems,
3605 struct extent_buffer *left = path->nodes[0];
3606 struct extent_buffer *upper = path->nodes[1];
3607 struct btrfs_map_token token;
3608 struct btrfs_disk_key disk_key;
3613 struct btrfs_item *item;
3619 btrfs_init_map_token(&token);
3624 nr = max_t(u32, 1, min_slot);
3626 if (path->slots[0] >= left_nritems)
3627 push_space += data_size;
3629 slot = path->slots[1];
3630 i = left_nritems - 1;
3632 item = btrfs_item_nr(i);
3634 if (!empty && push_items > 0) {
3635 if (path->slots[0] > i)
3637 if (path->slots[0] == i) {
3638 int space = btrfs_leaf_free_space(fs_info, left);
3639 if (space + push_space * 2 > free_space)
3644 if (path->slots[0] == i)
3645 push_space += data_size;
3647 this_item_size = btrfs_item_size(left, item);
3648 if (this_item_size + sizeof(*item) + push_space > free_space)
3652 push_space += this_item_size + sizeof(*item);
3658 if (push_items == 0)
3661 WARN_ON(!empty && push_items == left_nritems);
3663 /* push left to right */
3664 right_nritems = btrfs_header_nritems(right);
3666 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3667 push_space -= leaf_data_end(fs_info, left);
3669 /* make room in the right data area */
3670 data_end = leaf_data_end(fs_info, right);
3671 memmove_extent_buffer(right,
3672 btrfs_leaf_data(right) + data_end - push_space,
3673 btrfs_leaf_data(right) + data_end,
3674 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3676 /* copy from the left data area */
3677 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3678 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3679 btrfs_leaf_data(left) + leaf_data_end(fs_info, left),
3682 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3683 btrfs_item_nr_offset(0),
3684 right_nritems * sizeof(struct btrfs_item));
3686 /* copy the items from left to right */
3687 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3688 btrfs_item_nr_offset(left_nritems - push_items),
3689 push_items * sizeof(struct btrfs_item));
3691 /* update the item pointers */
3692 right_nritems += push_items;
3693 btrfs_set_header_nritems(right, right_nritems);
3694 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3695 for (i = 0; i < right_nritems; i++) {
3696 item = btrfs_item_nr(i);
3697 push_space -= btrfs_token_item_size(right, item, &token);
3698 btrfs_set_token_item_offset(right, item, push_space, &token);
3701 left_nritems -= push_items;
3702 btrfs_set_header_nritems(left, left_nritems);
3705 btrfs_mark_buffer_dirty(left);
3707 clean_tree_block(trans, fs_info, left);
3709 btrfs_mark_buffer_dirty(right);
3711 btrfs_item_key(right, &disk_key, 0);
3712 btrfs_set_node_key(upper, &disk_key, slot + 1);
3713 btrfs_mark_buffer_dirty(upper);
3715 /* then fixup the leaf pointer in the path */
3716 if (path->slots[0] >= left_nritems) {
3717 path->slots[0] -= left_nritems;
3718 if (btrfs_header_nritems(path->nodes[0]) == 0)
3719 clean_tree_block(trans, fs_info, path->nodes[0]);
3720 btrfs_tree_unlock(path->nodes[0]);
3721 free_extent_buffer(path->nodes[0]);
3722 path->nodes[0] = right;
3723 path->slots[1] += 1;
3725 btrfs_tree_unlock(right);
3726 free_extent_buffer(right);
3731 btrfs_tree_unlock(right);
3732 free_extent_buffer(right);
3737 * push some data in the path leaf to the right, trying to free up at
3738 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3740 * returns 1 if the push failed because the other node didn't have enough
3741 * room, 0 if everything worked out and < 0 if there were major errors.
3743 * this will push starting from min_slot to the end of the leaf. It won't
3744 * push any slot lower than min_slot
3746 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3747 *root, struct btrfs_path *path,
3748 int min_data_size, int data_size,
3749 int empty, u32 min_slot)
3751 struct btrfs_fs_info *fs_info = root->fs_info;
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(fs_info, upper, slot + 1);
3772 * slot + 1 is not valid or we fail to read the right node,
3773 * no big deal, just return.
3778 btrfs_tree_lock(right);
3779 btrfs_set_lock_blocking(right);
3781 free_space = btrfs_leaf_free_space(fs_info, right);
3782 if (free_space < data_size)
3785 /* cow and double check */
3786 ret = btrfs_cow_block(trans, root, right, upper,
3791 free_space = btrfs_leaf_free_space(fs_info, right);
3792 if (free_space < data_size)
3795 left_nritems = btrfs_header_nritems(left);
3796 if (left_nritems == 0)
3799 if (path->slots[0] == left_nritems && !empty) {
3800 /* Key greater than all keys in the leaf, right neighbor has
3801 * enough room for it and we're not emptying our leaf to delete
3802 * it, therefore use right neighbor to insert the new item and
3803 * no need to touch/dirty our left leaft. */
3804 btrfs_tree_unlock(left);
3805 free_extent_buffer(left);
3806 path->nodes[0] = right;
3812 return __push_leaf_right(trans, fs_info, path, min_data_size, empty,
3813 right, free_space, left_nritems, min_slot);
3815 btrfs_tree_unlock(right);
3816 free_extent_buffer(right);
3821 * push some data in the path leaf to the left, trying to free up at
3822 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3824 * max_slot can put a limit on how far into the leaf we'll push items. The
3825 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3828 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3829 struct btrfs_fs_info *fs_info,
3830 struct btrfs_path *path, int data_size,
3831 int empty, struct extent_buffer *left,
3832 int free_space, u32 right_nritems,
3835 struct btrfs_disk_key disk_key;
3836 struct extent_buffer *right = path->nodes[0];
3840 struct btrfs_item *item;
3841 u32 old_left_nritems;
3845 u32 old_left_item_size;
3846 struct btrfs_map_token token;
3848 btrfs_init_map_token(&token);
3851 nr = min(right_nritems, max_slot);
3853 nr = min(right_nritems - 1, max_slot);
3855 for (i = 0; i < nr; i++) {
3856 item = btrfs_item_nr(i);
3858 if (!empty && push_items > 0) {
3859 if (path->slots[0] < i)
3861 if (path->slots[0] == i) {
3862 int space = btrfs_leaf_free_space(fs_info, right);
3863 if (space + push_space * 2 > free_space)
3868 if (path->slots[0] == i)
3869 push_space += data_size;
3871 this_item_size = btrfs_item_size(right, item);
3872 if (this_item_size + sizeof(*item) + push_space > free_space)
3876 push_space += this_item_size + sizeof(*item);
3879 if (push_items == 0) {
3883 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3885 /* push data from right to left */
3886 copy_extent_buffer(left, right,
3887 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3888 btrfs_item_nr_offset(0),
3889 push_items * sizeof(struct btrfs_item));
3891 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3892 btrfs_item_offset_nr(right, push_items - 1);
3894 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3895 leaf_data_end(fs_info, left) - push_space,
3896 btrfs_leaf_data(right) +
3897 btrfs_item_offset_nr(right, push_items - 1),
3899 old_left_nritems = btrfs_header_nritems(left);
3900 BUG_ON(old_left_nritems <= 0);
3902 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3903 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3906 item = btrfs_item_nr(i);
3908 ioff = btrfs_token_item_offset(left, item, &token);
3909 btrfs_set_token_item_offset(left, item,
3910 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3913 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3915 /* fixup right node */
3916 if (push_items > right_nritems)
3917 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3920 if (push_items < right_nritems) {
3921 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3922 leaf_data_end(fs_info, right);
3923 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3924 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3925 btrfs_leaf_data(right) +
3926 leaf_data_end(fs_info, right), push_space);
3928 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3929 btrfs_item_nr_offset(push_items),
3930 (btrfs_header_nritems(right) - push_items) *
3931 sizeof(struct btrfs_item));
3933 right_nritems -= push_items;
3934 btrfs_set_header_nritems(right, right_nritems);
3935 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3936 for (i = 0; i < right_nritems; i++) {
3937 item = btrfs_item_nr(i);
3939 push_space = push_space - btrfs_token_item_size(right,
3941 btrfs_set_token_item_offset(right, item, push_space, &token);
3944 btrfs_mark_buffer_dirty(left);
3946 btrfs_mark_buffer_dirty(right);
3948 clean_tree_block(trans, fs_info, right);
3950 btrfs_item_key(right, &disk_key, 0);
3951 fixup_low_keys(fs_info, path, &disk_key, 1);
3953 /* then fixup the leaf pointer in the path */
3954 if (path->slots[0] < push_items) {
3955 path->slots[0] += old_left_nritems;
3956 btrfs_tree_unlock(path->nodes[0]);
3957 free_extent_buffer(path->nodes[0]);
3958 path->nodes[0] = left;
3959 path->slots[1] -= 1;
3961 btrfs_tree_unlock(left);
3962 free_extent_buffer(left);
3963 path->slots[0] -= push_items;
3965 BUG_ON(path->slots[0] < 0);
3968 btrfs_tree_unlock(left);
3969 free_extent_buffer(left);
3974 * push some data in the path leaf to the left, trying to free up at
3975 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3977 * max_slot can put a limit on how far into the leaf we'll push items. The
3978 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3981 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3982 *root, struct btrfs_path *path, int min_data_size,
3983 int data_size, int empty, u32 max_slot)
3985 struct btrfs_fs_info *fs_info = root->fs_info;
3986 struct extent_buffer *right = path->nodes[0];
3987 struct extent_buffer *left;
3993 slot = path->slots[1];
3996 if (!path->nodes[1])
3999 right_nritems = btrfs_header_nritems(right);
4000 if (right_nritems == 0)
4003 btrfs_assert_tree_locked(path->nodes[1]);
4005 left = read_node_slot(fs_info, path->nodes[1], slot - 1);
4007 * slot - 1 is not valid or we fail to read the left node,
4008 * no big deal, just return.
4013 btrfs_tree_lock(left);
4014 btrfs_set_lock_blocking(left);
4016 free_space = btrfs_leaf_free_space(fs_info, left);
4017 if (free_space < data_size) {
4022 /* cow and double check */
4023 ret = btrfs_cow_block(trans, root, left,
4024 path->nodes[1], slot - 1, &left);
4026 /* we hit -ENOSPC, but it isn't fatal here */
4032 free_space = btrfs_leaf_free_space(fs_info, left);
4033 if (free_space < data_size) {
4038 return __push_leaf_left(trans, fs_info, path, min_data_size,
4039 empty, left, free_space, right_nritems,
4042 btrfs_tree_unlock(left);
4043 free_extent_buffer(left);
4048 * split the path's leaf in two, making sure there is at least data_size
4049 * available for the resulting leaf level of the path.
4051 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4052 struct btrfs_fs_info *fs_info,
4053 struct btrfs_path *path,
4054 struct extent_buffer *l,
4055 struct extent_buffer *right,
4056 int slot, int mid, int nritems)
4061 struct btrfs_disk_key disk_key;
4062 struct btrfs_map_token token;
4064 btrfs_init_map_token(&token);
4066 nritems = nritems - mid;
4067 btrfs_set_header_nritems(right, nritems);
4068 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l);
4070 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4071 btrfs_item_nr_offset(mid),
4072 nritems * sizeof(struct btrfs_item));
4074 copy_extent_buffer(right, l,
4075 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(fs_info) -
4076 data_copy_size, btrfs_leaf_data(l) +
4077 leaf_data_end(fs_info, l), data_copy_size);
4079 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4081 for (i = 0; i < nritems; i++) {
4082 struct btrfs_item *item = btrfs_item_nr(i);
4085 ioff = btrfs_token_item_offset(right, item, &token);
4086 btrfs_set_token_item_offset(right, item,
4087 ioff + rt_data_off, &token);
4090 btrfs_set_header_nritems(l, mid);
4091 btrfs_item_key(right, &disk_key, 0);
4092 insert_ptr(trans, fs_info, path, &disk_key, right->start,
4093 path->slots[1] + 1, 1);
4095 btrfs_mark_buffer_dirty(right);
4096 btrfs_mark_buffer_dirty(l);
4097 BUG_ON(path->slots[0] != slot);
4100 btrfs_tree_unlock(path->nodes[0]);
4101 free_extent_buffer(path->nodes[0]);
4102 path->nodes[0] = right;
4103 path->slots[0] -= mid;
4104 path->slots[1] += 1;
4106 btrfs_tree_unlock(right);
4107 free_extent_buffer(right);
4110 BUG_ON(path->slots[0] < 0);
4114 * double splits happen when we need to insert a big item in the middle
4115 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4116 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4119 * We avoid this by trying to push the items on either side of our target
4120 * into the adjacent leaves. If all goes well we can avoid the double split
4123 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4124 struct btrfs_root *root,
4125 struct btrfs_path *path,
4128 struct btrfs_fs_info *fs_info = root->fs_info;
4133 int space_needed = data_size;
4135 slot = path->slots[0];
4136 if (slot < btrfs_header_nritems(path->nodes[0]))
4137 space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4140 * try to push all the items after our slot into the
4143 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4150 nritems = btrfs_header_nritems(path->nodes[0]);
4152 * our goal is to get our slot at the start or end of a leaf. If
4153 * we've done so we're done
4155 if (path->slots[0] == 0 || path->slots[0] == nritems)
4158 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4161 /* try to push all the items before our slot into the next leaf */
4162 slot = path->slots[0];
4163 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4176 * split the path's leaf in two, making sure there is at least data_size
4177 * available for the resulting leaf level of the path.
4179 * returns 0 if all went well and < 0 on failure.
4181 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4182 struct btrfs_root *root,
4183 struct btrfs_key *ins_key,
4184 struct btrfs_path *path, int data_size,
4187 struct btrfs_disk_key disk_key;
4188 struct extent_buffer *l;
4192 struct extent_buffer *right;
4193 struct btrfs_fs_info *fs_info = root->fs_info;
4197 int num_doubles = 0;
4198 int tried_avoid_double = 0;
4201 slot = path->slots[0];
4202 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4203 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4206 /* first try to make some room by pushing left and right */
4207 if (data_size && path->nodes[1]) {
4208 int space_needed = data_size;
4210 if (slot < btrfs_header_nritems(l))
4211 space_needed -= btrfs_leaf_free_space(fs_info, l);
4213 wret = push_leaf_right(trans, root, path, space_needed,
4214 space_needed, 0, 0);
4218 wret = push_leaf_left(trans, root, path, space_needed,
4219 space_needed, 0, (u32)-1);
4225 /* did the pushes work? */
4226 if (btrfs_leaf_free_space(fs_info, l) >= data_size)
4230 if (!path->nodes[1]) {
4231 ret = insert_new_root(trans, root, path, 1);
4238 slot = path->slots[0];
4239 nritems = btrfs_header_nritems(l);
4240 mid = (nritems + 1) / 2;
4244 leaf_space_used(l, mid, nritems - mid) + data_size >
4245 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4246 if (slot >= nritems) {
4250 if (mid != nritems &&
4251 leaf_space_used(l, mid, nritems - mid) +
4252 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4253 if (data_size && !tried_avoid_double)
4254 goto push_for_double;
4260 if (leaf_space_used(l, 0, mid) + data_size >
4261 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4262 if (!extend && data_size && slot == 0) {
4264 } else if ((extend || !data_size) && slot == 0) {
4268 if (mid != nritems &&
4269 leaf_space_used(l, mid, nritems - mid) +
4270 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4271 if (data_size && !tried_avoid_double)
4272 goto push_for_double;
4280 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4282 btrfs_item_key(l, &disk_key, mid);
4284 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4285 &disk_key, 0, l->start, 0);
4287 return PTR_ERR(right);
4289 root_add_used(root, fs_info->nodesize);
4291 memzero_extent_buffer(right, 0, sizeof(struct btrfs_header));
4292 btrfs_set_header_bytenr(right, right->start);
4293 btrfs_set_header_generation(right, trans->transid);
4294 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4295 btrfs_set_header_owner(right, root->root_key.objectid);
4296 btrfs_set_header_level(right, 0);
4297 write_extent_buffer_fsid(right, fs_info->fsid);
4298 write_extent_buffer_chunk_tree_uuid(right, fs_info->chunk_tree_uuid);
4302 btrfs_set_header_nritems(right, 0);
4303 insert_ptr(trans, fs_info, path, &disk_key,
4304 right->start, path->slots[1] + 1, 1);
4305 btrfs_tree_unlock(path->nodes[0]);
4306 free_extent_buffer(path->nodes[0]);
4307 path->nodes[0] = right;
4309 path->slots[1] += 1;
4311 btrfs_set_header_nritems(right, 0);
4312 insert_ptr(trans, fs_info, path, &disk_key,
4313 right->start, path->slots[1], 1);
4314 btrfs_tree_unlock(path->nodes[0]);
4315 free_extent_buffer(path->nodes[0]);
4316 path->nodes[0] = right;
4318 if (path->slots[1] == 0)
4319 fixup_low_keys(fs_info, path, &disk_key, 1);
4322 * We create a new leaf 'right' for the required ins_len and
4323 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4324 * the content of ins_len to 'right'.
4329 copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems);
4332 BUG_ON(num_doubles != 0);
4340 push_for_double_split(trans, root, path, data_size);
4341 tried_avoid_double = 1;
4342 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4347 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4348 struct btrfs_root *root,
4349 struct btrfs_path *path, int ins_len)
4351 struct btrfs_fs_info *fs_info = root->fs_info;
4352 struct btrfs_key key;
4353 struct extent_buffer *leaf;
4354 struct btrfs_file_extent_item *fi;
4359 leaf = path->nodes[0];
4360 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4362 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4363 key.type != BTRFS_EXTENT_CSUM_KEY);
4365 if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len)
4368 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4369 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4370 fi = btrfs_item_ptr(leaf, path->slots[0],
4371 struct btrfs_file_extent_item);
4372 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4374 btrfs_release_path(path);
4376 path->keep_locks = 1;
4377 path->search_for_split = 1;
4378 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4379 path->search_for_split = 0;
4386 leaf = path->nodes[0];
4387 /* if our item isn't there, return now */
4388 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4391 /* the leaf has changed, it now has room. return now */
4392 if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len)
4395 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4396 fi = btrfs_item_ptr(leaf, path->slots[0],
4397 struct btrfs_file_extent_item);
4398 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4402 btrfs_set_path_blocking(path);
4403 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4407 path->keep_locks = 0;
4408 btrfs_unlock_up_safe(path, 1);
4411 path->keep_locks = 0;
4415 static noinline int split_item(struct btrfs_trans_handle *trans,
4416 struct btrfs_fs_info *fs_info,
4417 struct btrfs_path *path,
4418 struct btrfs_key *new_key,
4419 unsigned long split_offset)
4421 struct extent_buffer *leaf;
4422 struct btrfs_item *item;
4423 struct btrfs_item *new_item;
4429 struct btrfs_disk_key disk_key;
4431 leaf = path->nodes[0];
4432 BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item));
4434 btrfs_set_path_blocking(path);
4436 item = btrfs_item_nr(path->slots[0]);
4437 orig_offset = btrfs_item_offset(leaf, item);
4438 item_size = btrfs_item_size(leaf, item);
4440 buf = kmalloc(item_size, GFP_NOFS);
4444 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4445 path->slots[0]), item_size);
4447 slot = path->slots[0] + 1;
4448 nritems = btrfs_header_nritems(leaf);
4449 if (slot != nritems) {
4450 /* shift the items */
4451 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4452 btrfs_item_nr_offset(slot),
4453 (nritems - slot) * sizeof(struct btrfs_item));
4456 btrfs_cpu_key_to_disk(&disk_key, new_key);
4457 btrfs_set_item_key(leaf, &disk_key, slot);
4459 new_item = btrfs_item_nr(slot);
4461 btrfs_set_item_offset(leaf, new_item, orig_offset);
4462 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4464 btrfs_set_item_offset(leaf, item,
4465 orig_offset + item_size - split_offset);
4466 btrfs_set_item_size(leaf, item, split_offset);
4468 btrfs_set_header_nritems(leaf, nritems + 1);
4470 /* write the data for the start of the original item */
4471 write_extent_buffer(leaf, buf,
4472 btrfs_item_ptr_offset(leaf, path->slots[0]),
4475 /* write the data for the new item */
4476 write_extent_buffer(leaf, buf + split_offset,
4477 btrfs_item_ptr_offset(leaf, slot),
4478 item_size - split_offset);
4479 btrfs_mark_buffer_dirty(leaf);
4481 BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0);
4487 * This function splits a single item into two items,
4488 * giving 'new_key' to the new item and splitting the
4489 * old one at split_offset (from the start of the item).
4491 * The path may be released by this operation. After
4492 * the split, the path is pointing to the old item. The
4493 * new item is going to be in the same node as the old one.
4495 * Note, the item being split must be smaller enough to live alone on
4496 * a tree block with room for one extra struct btrfs_item
4498 * This allows us to split the item in place, keeping a lock on the
4499 * leaf the entire time.
4501 int btrfs_split_item(struct btrfs_trans_handle *trans,
4502 struct btrfs_root *root,
4503 struct btrfs_path *path,
4504 struct btrfs_key *new_key,
4505 unsigned long split_offset)
4508 ret = setup_leaf_for_split(trans, root, path,
4509 sizeof(struct btrfs_item));
4513 ret = split_item(trans, root->fs_info, path, new_key, split_offset);
4518 * This function duplicate a item, giving 'new_key' to the new item.
4519 * It guarantees both items live in the same tree leaf and the new item
4520 * is contiguous with the original item.
4522 * This allows us to split file extent in place, keeping a lock on the
4523 * leaf the entire time.
4525 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4526 struct btrfs_root *root,
4527 struct btrfs_path *path,
4528 struct btrfs_key *new_key)
4530 struct extent_buffer *leaf;
4534 leaf = path->nodes[0];
4535 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4536 ret = setup_leaf_for_split(trans, root, path,
4537 item_size + sizeof(struct btrfs_item));
4542 setup_items_for_insert(root, path, new_key, &item_size,
4543 item_size, item_size +
4544 sizeof(struct btrfs_item), 1);
4545 leaf = path->nodes[0];
4546 memcpy_extent_buffer(leaf,
4547 btrfs_item_ptr_offset(leaf, path->slots[0]),
4548 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4554 * make the item pointed to by the path smaller. new_size indicates
4555 * how small to make it, and from_end tells us if we just chop bytes
4556 * off the end of the item or if we shift the item to chop bytes off
4559 void btrfs_truncate_item(struct btrfs_fs_info *fs_info,
4560 struct btrfs_path *path, u32 new_size, int from_end)
4563 struct extent_buffer *leaf;
4564 struct btrfs_item *item;
4566 unsigned int data_end;
4567 unsigned int old_data_start;
4568 unsigned int old_size;
4569 unsigned int size_diff;
4571 struct btrfs_map_token token;
4573 btrfs_init_map_token(&token);
4575 leaf = path->nodes[0];
4576 slot = path->slots[0];
4578 old_size = btrfs_item_size_nr(leaf, slot);
4579 if (old_size == new_size)
4582 nritems = btrfs_header_nritems(leaf);
4583 data_end = leaf_data_end(fs_info, leaf);
4585 old_data_start = btrfs_item_offset_nr(leaf, slot);
4587 size_diff = old_size - new_size;
4590 BUG_ON(slot >= nritems);
4593 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4595 /* first correct the data pointers */
4596 for (i = slot; i < nritems; i++) {
4598 item = btrfs_item_nr(i);
4600 ioff = btrfs_token_item_offset(leaf, item, &token);
4601 btrfs_set_token_item_offset(leaf, item,
4602 ioff + size_diff, &token);
4605 /* shift the data */
4607 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4608 data_end + size_diff, btrfs_leaf_data(leaf) +
4609 data_end, old_data_start + new_size - data_end);
4611 struct btrfs_disk_key disk_key;
4614 btrfs_item_key(leaf, &disk_key, slot);
4616 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4618 struct btrfs_file_extent_item *fi;
4620 fi = btrfs_item_ptr(leaf, slot,
4621 struct btrfs_file_extent_item);
4622 fi = (struct btrfs_file_extent_item *)(
4623 (unsigned long)fi - size_diff);
4625 if (btrfs_file_extent_type(leaf, fi) ==
4626 BTRFS_FILE_EXTENT_INLINE) {
4627 ptr = btrfs_item_ptr_offset(leaf, slot);
4628 memmove_extent_buffer(leaf, ptr,
4630 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4634 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4635 data_end + size_diff, btrfs_leaf_data(leaf) +
4636 data_end, old_data_start - data_end);
4638 offset = btrfs_disk_key_offset(&disk_key);
4639 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4640 btrfs_set_item_key(leaf, &disk_key, slot);
4642 fixup_low_keys(fs_info, path, &disk_key, 1);
4645 item = btrfs_item_nr(slot);
4646 btrfs_set_item_size(leaf, item, new_size);
4647 btrfs_mark_buffer_dirty(leaf);
4649 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4650 btrfs_print_leaf(fs_info, leaf);
4656 * make the item pointed to by the path bigger, data_size is the added size.
4658 void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
4662 struct extent_buffer *leaf;
4663 struct btrfs_item *item;
4665 unsigned int data_end;
4666 unsigned int old_data;
4667 unsigned int old_size;
4669 struct btrfs_map_token token;
4671 btrfs_init_map_token(&token);
4673 leaf = path->nodes[0];
4675 nritems = btrfs_header_nritems(leaf);
4676 data_end = leaf_data_end(fs_info, leaf);
4678 if (btrfs_leaf_free_space(fs_info, leaf) < data_size) {
4679 btrfs_print_leaf(fs_info, leaf);
4682 slot = path->slots[0];
4683 old_data = btrfs_item_end_nr(leaf, slot);
4686 if (slot >= nritems) {
4687 btrfs_print_leaf(fs_info, leaf);
4688 btrfs_crit(fs_info, "slot %d too large, nritems %d",
4694 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4696 /* first correct the data pointers */
4697 for (i = slot; i < nritems; i++) {
4699 item = btrfs_item_nr(i);
4701 ioff = btrfs_token_item_offset(leaf, item, &token);
4702 btrfs_set_token_item_offset(leaf, item,
4703 ioff - data_size, &token);
4706 /* shift the data */
4707 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4708 data_end - data_size, btrfs_leaf_data(leaf) +
4709 data_end, old_data - data_end);
4711 data_end = old_data;
4712 old_size = btrfs_item_size_nr(leaf, slot);
4713 item = btrfs_item_nr(slot);
4714 btrfs_set_item_size(leaf, item, old_size + data_size);
4715 btrfs_mark_buffer_dirty(leaf);
4717 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4718 btrfs_print_leaf(fs_info, leaf);
4724 * this is a helper for btrfs_insert_empty_items, the main goal here is
4725 * to save stack depth by doing the bulk of the work in a function
4726 * that doesn't call btrfs_search_slot
4728 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4729 struct btrfs_key *cpu_key, u32 *data_size,
4730 u32 total_data, u32 total_size, int nr)
4732 struct btrfs_fs_info *fs_info = root->fs_info;
4733 struct btrfs_item *item;
4736 unsigned int data_end;
4737 struct btrfs_disk_key disk_key;
4738 struct extent_buffer *leaf;
4740 struct btrfs_map_token token;
4742 if (path->slots[0] == 0) {
4743 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4744 fixup_low_keys(fs_info, path, &disk_key, 1);
4746 btrfs_unlock_up_safe(path, 1);
4748 btrfs_init_map_token(&token);
4750 leaf = path->nodes[0];
4751 slot = path->slots[0];
4753 nritems = btrfs_header_nritems(leaf);
4754 data_end = leaf_data_end(fs_info, leaf);
4756 if (btrfs_leaf_free_space(fs_info, leaf) < total_size) {
4757 btrfs_print_leaf(fs_info, leaf);
4758 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4759 total_size, btrfs_leaf_free_space(fs_info, leaf));
4763 if (slot != nritems) {
4764 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4766 if (old_data < data_end) {
4767 btrfs_print_leaf(fs_info, leaf);
4768 btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4769 slot, old_data, data_end);
4773 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4775 /* first correct the data pointers */
4776 for (i = slot; i < nritems; i++) {
4779 item = btrfs_item_nr(i);
4780 ioff = btrfs_token_item_offset(leaf, item, &token);
4781 btrfs_set_token_item_offset(leaf, item,
4782 ioff - total_data, &token);
4784 /* shift the items */
4785 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4786 btrfs_item_nr_offset(slot),
4787 (nritems - slot) * sizeof(struct btrfs_item));
4789 /* shift the data */
4790 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4791 data_end - total_data, btrfs_leaf_data(leaf) +
4792 data_end, old_data - data_end);
4793 data_end = old_data;
4796 /* setup the item for the new data */
4797 for (i = 0; i < nr; i++) {
4798 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4799 btrfs_set_item_key(leaf, &disk_key, slot + i);
4800 item = btrfs_item_nr(slot + i);
4801 btrfs_set_token_item_offset(leaf, item,
4802 data_end - data_size[i], &token);
4803 data_end -= data_size[i];
4804 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4807 btrfs_set_header_nritems(leaf, nritems + nr);
4808 btrfs_mark_buffer_dirty(leaf);
4810 if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4811 btrfs_print_leaf(fs_info, leaf);
4817 * Given a key and some data, insert items into the tree.
4818 * This does all the path init required, making room in the tree if needed.
4820 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4821 struct btrfs_root *root,
4822 struct btrfs_path *path,
4823 struct btrfs_key *cpu_key, u32 *data_size,
4832 for (i = 0; i < nr; i++)
4833 total_data += data_size[i];
4835 total_size = total_data + (nr * sizeof(struct btrfs_item));
4836 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4842 slot = path->slots[0];
4845 setup_items_for_insert(root, path, cpu_key, data_size,
4846 total_data, total_size, nr);
4851 * Given a key and some data, insert an item into the tree.
4852 * This does all the path init required, making room in the tree if needed.
4854 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4855 *root, struct btrfs_key *cpu_key, void *data, u32
4859 struct btrfs_path *path;
4860 struct extent_buffer *leaf;
4863 path = btrfs_alloc_path();
4866 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4868 leaf = path->nodes[0];
4869 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4870 write_extent_buffer(leaf, data, ptr, data_size);
4871 btrfs_mark_buffer_dirty(leaf);
4873 btrfs_free_path(path);
4878 * delete the pointer from a given node.
4880 * the tree should have been previously balanced so the deletion does not
4883 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4884 int level, int slot)
4886 struct btrfs_fs_info *fs_info = root->fs_info;
4887 struct extent_buffer *parent = path->nodes[level];
4891 nritems = btrfs_header_nritems(parent);
4892 if (slot != nritems - 1) {
4894 tree_mod_log_eb_move(fs_info, parent, slot,
4895 slot + 1, nritems - slot - 1);
4896 memmove_extent_buffer(parent,
4897 btrfs_node_key_ptr_offset(slot),
4898 btrfs_node_key_ptr_offset(slot + 1),
4899 sizeof(struct btrfs_key_ptr) *
4900 (nritems - slot - 1));
4902 ret = tree_mod_log_insert_key(fs_info, parent, slot,
4903 MOD_LOG_KEY_REMOVE, GFP_NOFS);
4908 btrfs_set_header_nritems(parent, nritems);
4909 if (nritems == 0 && parent == root->node) {
4910 BUG_ON(btrfs_header_level(root->node) != 1);
4911 /* just turn the root into a leaf and break */
4912 btrfs_set_header_level(root->node, 0);
4913 } else if (slot == 0) {
4914 struct btrfs_disk_key disk_key;
4916 btrfs_node_key(parent, &disk_key, 0);
4917 fixup_low_keys(fs_info, path, &disk_key, level + 1);
4919 btrfs_mark_buffer_dirty(parent);
4923 * a helper function to delete the leaf pointed to by path->slots[1] and
4926 * This deletes the pointer in path->nodes[1] and frees the leaf
4927 * block extent. zero is returned if it all worked out, < 0 otherwise.
4929 * The path must have already been setup for deleting the leaf, including
4930 * all the proper balancing. path->nodes[1] must be locked.
4932 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4933 struct btrfs_root *root,
4934 struct btrfs_path *path,
4935 struct extent_buffer *leaf)
4937 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4938 del_ptr(root, path, 1, path->slots[1]);
4941 * btrfs_free_extent is expensive, we want to make sure we
4942 * aren't holding any locks when we call it
4944 btrfs_unlock_up_safe(path, 0);
4946 root_sub_used(root, leaf->len);
4948 extent_buffer_get(leaf);
4949 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4950 free_extent_buffer_stale(leaf);
4953 * delete the item at the leaf level in path. If that empties
4954 * the leaf, remove it from the tree
4956 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4957 struct btrfs_path *path, int slot, int nr)
4959 struct btrfs_fs_info *fs_info = root->fs_info;
4960 struct extent_buffer *leaf;
4961 struct btrfs_item *item;
4968 struct btrfs_map_token token;
4970 btrfs_init_map_token(&token);
4972 leaf = path->nodes[0];
4973 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4975 for (i = 0; i < nr; i++)
4976 dsize += btrfs_item_size_nr(leaf, slot + i);
4978 nritems = btrfs_header_nritems(leaf);
4980 if (slot + nr != nritems) {
4981 int data_end = leaf_data_end(fs_info, leaf);
4983 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4985 btrfs_leaf_data(leaf) + data_end,
4986 last_off - data_end);
4988 for (i = slot + nr; i < nritems; i++) {
4991 item = btrfs_item_nr(i);
4992 ioff = btrfs_token_item_offset(leaf, item, &token);
4993 btrfs_set_token_item_offset(leaf, item,
4994 ioff + dsize, &token);
4997 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4998 btrfs_item_nr_offset(slot + nr),
4999 sizeof(struct btrfs_item) *
5000 (nritems - slot - nr));
5002 btrfs_set_header_nritems(leaf, nritems - nr);
5005 /* delete the leaf if we've emptied it */
5007 if (leaf == root->node) {
5008 btrfs_set_header_level(leaf, 0);
5010 btrfs_set_path_blocking(path);
5011 clean_tree_block(trans, fs_info, leaf);
5012 btrfs_del_leaf(trans, root, path, leaf);
5015 int used = leaf_space_used(leaf, 0, nritems);
5017 struct btrfs_disk_key disk_key;
5019 btrfs_item_key(leaf, &disk_key, 0);
5020 fixup_low_keys(fs_info, path, &disk_key, 1);
5023 /* delete the leaf if it is mostly empty */
5024 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
5025 /* push_leaf_left fixes the path.
5026 * make sure the path still points to our leaf
5027 * for possible call to del_ptr below
5029 slot = path->slots[1];
5030 extent_buffer_get(leaf);
5032 btrfs_set_path_blocking(path);
5033 wret = push_leaf_left(trans, root, path, 1, 1,
5035 if (wret < 0 && wret != -ENOSPC)
5038 if (path->nodes[0] == leaf &&
5039 btrfs_header_nritems(leaf)) {
5040 wret = push_leaf_right(trans, root, path, 1,
5042 if (wret < 0 && wret != -ENOSPC)
5046 if (btrfs_header_nritems(leaf) == 0) {
5047 path->slots[1] = slot;
5048 btrfs_del_leaf(trans, root, path, leaf);
5049 free_extent_buffer(leaf);
5052 /* if we're still in the path, make sure
5053 * we're dirty. Otherwise, one of the
5054 * push_leaf functions must have already
5055 * dirtied this buffer
5057 if (path->nodes[0] == leaf)
5058 btrfs_mark_buffer_dirty(leaf);
5059 free_extent_buffer(leaf);
5062 btrfs_mark_buffer_dirty(leaf);
5069 * search the tree again to find a leaf with lesser keys
5070 * returns 0 if it found something or 1 if there are no lesser leaves.
5071 * returns < 0 on io errors.
5073 * This may release the path, and so you may lose any locks held at the
5076 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5078 struct btrfs_key key;
5079 struct btrfs_disk_key found_key;
5082 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5084 if (key.offset > 0) {
5086 } else if (key.type > 0) {
5088 key.offset = (u64)-1;
5089 } else if (key.objectid > 0) {
5092 key.offset = (u64)-1;
5097 btrfs_release_path(path);
5098 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5101 btrfs_item_key(path->nodes[0], &found_key, 0);
5102 ret = comp_keys(&found_key, &key);
5104 * We might have had an item with the previous key in the tree right
5105 * before we released our path. And after we released our path, that
5106 * item might have been pushed to the first slot (0) of the leaf we
5107 * were holding due to a tree balance. Alternatively, an item with the
5108 * previous key can exist as the only element of a leaf (big fat item).
5109 * Therefore account for these 2 cases, so that our callers (like
5110 * btrfs_previous_item) don't miss an existing item with a key matching
5111 * the previous key we computed above.
5119 * A helper function to walk down the tree starting at min_key, and looking
5120 * for nodes or leaves that are have a minimum transaction id.
5121 * This is used by the btree defrag code, and tree logging
5123 * This does not cow, but it does stuff the starting key it finds back
5124 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5125 * key and get a writable path.
5127 * This does lock as it descends, and path->keep_locks should be set
5128 * to 1 by the caller.
5130 * This honors path->lowest_level to prevent descent past a given level
5133 * min_trans indicates the oldest transaction that you are interested
5134 * in walking through. Any nodes or leaves older than min_trans are
5135 * skipped over (without reading them).
5137 * returns zero if something useful was found, < 0 on error and 1 if there
5138 * was nothing in the tree that matched the search criteria.
5140 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5141 struct btrfs_path *path,
5144 struct btrfs_fs_info *fs_info = root->fs_info;
5145 struct extent_buffer *cur;
5146 struct btrfs_key found_key;
5152 int keep_locks = path->keep_locks;
5154 path->keep_locks = 1;
5156 cur = btrfs_read_lock_root_node(root);
5157 level = btrfs_header_level(cur);
5158 WARN_ON(path->nodes[level]);
5159 path->nodes[level] = cur;
5160 path->locks[level] = BTRFS_READ_LOCK;
5162 if (btrfs_header_generation(cur) < min_trans) {
5167 nritems = btrfs_header_nritems(cur);
5168 level = btrfs_header_level(cur);
5169 sret = bin_search(cur, min_key, level, &slot);
5171 /* at the lowest level, we're done, setup the path and exit */
5172 if (level == path->lowest_level) {
5173 if (slot >= nritems)
5176 path->slots[level] = slot;
5177 btrfs_item_key_to_cpu(cur, &found_key, slot);
5180 if (sret && slot > 0)
5183 * check this node pointer against the min_trans parameters.
5184 * If it is too old, old, skip to the next one.
5186 while (slot < nritems) {
5189 gen = btrfs_node_ptr_generation(cur, slot);
5190 if (gen < min_trans) {
5198 * we didn't find a candidate key in this node, walk forward
5199 * and find another one
5201 if (slot >= nritems) {
5202 path->slots[level] = slot;
5203 btrfs_set_path_blocking(path);
5204 sret = btrfs_find_next_key(root, path, min_key, level,
5207 btrfs_release_path(path);
5213 /* save our key for returning back */
5214 btrfs_node_key_to_cpu(cur, &found_key, slot);
5215 path->slots[level] = slot;
5216 if (level == path->lowest_level) {
5220 btrfs_set_path_blocking(path);
5221 cur = read_node_slot(fs_info, cur, slot);
5227 btrfs_tree_read_lock(cur);
5229 path->locks[level - 1] = BTRFS_READ_LOCK;
5230 path->nodes[level - 1] = cur;
5231 unlock_up(path, level, 1, 0, NULL);
5232 btrfs_clear_path_blocking(path, NULL, 0);
5235 path->keep_locks = keep_locks;
5237 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5238 btrfs_set_path_blocking(path);
5239 memcpy(min_key, &found_key, sizeof(found_key));
5244 static int tree_move_down(struct btrfs_fs_info *fs_info,
5245 struct btrfs_path *path,
5246 int *level, int root_level)
5248 struct extent_buffer *eb;
5250 BUG_ON(*level == 0);
5251 eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]);
5255 path->nodes[*level - 1] = eb;
5256 path->slots[*level - 1] = 0;
5261 static int tree_move_next_or_upnext(struct btrfs_fs_info *fs_info,
5262 struct btrfs_path *path,
5263 int *level, int root_level)
5267 nritems = btrfs_header_nritems(path->nodes[*level]);
5269 path->slots[*level]++;
5271 while (path->slots[*level] >= nritems) {
5272 if (*level == root_level)
5276 path->slots[*level] = 0;
5277 free_extent_buffer(path->nodes[*level]);
5278 path->nodes[*level] = NULL;
5280 path->slots[*level]++;
5282 nritems = btrfs_header_nritems(path->nodes[*level]);
5289 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5292 static int tree_advance(struct btrfs_fs_info *fs_info,
5293 struct btrfs_path *path,
5294 int *level, int root_level,
5296 struct btrfs_key *key)
5300 if (*level == 0 || !allow_down) {
5301 ret = tree_move_next_or_upnext(fs_info, path, level,
5304 ret = tree_move_down(fs_info, path, level, root_level);
5308 btrfs_item_key_to_cpu(path->nodes[*level], key,
5309 path->slots[*level]);
5311 btrfs_node_key_to_cpu(path->nodes[*level], key,
5312 path->slots[*level]);
5317 static int tree_compare_item(struct btrfs_path *left_path,
5318 struct btrfs_path *right_path,
5323 unsigned long off1, off2;
5325 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5326 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5330 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5331 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5332 right_path->slots[0]);
5334 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5336 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5343 #define ADVANCE_ONLY_NEXT -1
5346 * This function compares two trees and calls the provided callback for
5347 * every changed/new/deleted item it finds.
5348 * If shared tree blocks are encountered, whole subtrees are skipped, making
5349 * the compare pretty fast on snapshotted subvolumes.
5351 * This currently works on commit roots only. As commit roots are read only,
5352 * we don't do any locking. The commit roots are protected with transactions.
5353 * Transactions are ended and rejoined when a commit is tried in between.
5355 * This function checks for modifications done to the trees while comparing.
5356 * If it detects a change, it aborts immediately.
5358 int btrfs_compare_trees(struct btrfs_root *left_root,
5359 struct btrfs_root *right_root,
5360 btrfs_changed_cb_t changed_cb, void *ctx)
5362 struct btrfs_fs_info *fs_info = left_root->fs_info;
5365 struct btrfs_path *left_path = NULL;
5366 struct btrfs_path *right_path = NULL;
5367 struct btrfs_key left_key;
5368 struct btrfs_key right_key;
5369 char *tmp_buf = NULL;
5370 int left_root_level;
5371 int right_root_level;
5374 int left_end_reached;
5375 int right_end_reached;
5383 left_path = btrfs_alloc_path();
5388 right_path = btrfs_alloc_path();
5394 tmp_buf = kmalloc(fs_info->nodesize, GFP_KERNEL | __GFP_NOWARN);
5396 tmp_buf = vmalloc(fs_info->nodesize);
5403 left_path->search_commit_root = 1;
5404 left_path->skip_locking = 1;
5405 right_path->search_commit_root = 1;
5406 right_path->skip_locking = 1;
5409 * Strategy: Go to the first items of both trees. Then do
5411 * If both trees are at level 0
5412 * Compare keys of current items
5413 * If left < right treat left item as new, advance left tree
5415 * If left > right treat right item as deleted, advance right tree
5417 * If left == right do deep compare of items, treat as changed if
5418 * needed, advance both trees and repeat
5419 * If both trees are at the same level but not at level 0
5420 * Compare keys of current nodes/leafs
5421 * If left < right advance left tree and repeat
5422 * If left > right advance right tree and repeat
5423 * If left == right compare blockptrs of the next nodes/leafs
5424 * If they match advance both trees but stay at the same level
5426 * If they don't match advance both trees while allowing to go
5428 * If tree levels are different
5429 * Advance the tree that needs it and repeat
5431 * Advancing a tree means:
5432 * If we are at level 0, try to go to the next slot. If that's not
5433 * possible, go one level up and repeat. Stop when we found a level
5434 * where we could go to the next slot. We may at this point be on a
5437 * If we are not at level 0 and not on shared tree blocks, go one
5440 * If we are not at level 0 and on shared tree blocks, go one slot to
5441 * the right if possible or go up and right.
5444 down_read(&fs_info->commit_root_sem);
5445 left_level = btrfs_header_level(left_root->commit_root);
5446 left_root_level = left_level;
5447 left_path->nodes[left_level] = left_root->commit_root;
5448 extent_buffer_get(left_path->nodes[left_level]);
5450 right_level = btrfs_header_level(right_root->commit_root);
5451 right_root_level = right_level;
5452 right_path->nodes[right_level] = right_root->commit_root;
5453 extent_buffer_get(right_path->nodes[right_level]);
5454 up_read(&fs_info->commit_root_sem);
5456 if (left_level == 0)
5457 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5458 &left_key, left_path->slots[left_level]);
5460 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5461 &left_key, left_path->slots[left_level]);
5462 if (right_level == 0)
5463 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5464 &right_key, right_path->slots[right_level]);
5466 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5467 &right_key, right_path->slots[right_level]);
5469 left_end_reached = right_end_reached = 0;
5470 advance_left = advance_right = 0;
5473 if (advance_left && !left_end_reached) {
5474 ret = tree_advance(fs_info, left_path, &left_level,
5476 advance_left != ADVANCE_ONLY_NEXT,
5479 left_end_reached = ADVANCE;
5484 if (advance_right && !right_end_reached) {
5485 ret = tree_advance(fs_info, right_path, &right_level,
5487 advance_right != ADVANCE_ONLY_NEXT,
5490 right_end_reached = ADVANCE;
5496 if (left_end_reached && right_end_reached) {
5499 } else if (left_end_reached) {
5500 if (right_level == 0) {
5501 ret = changed_cb(left_root, right_root,
5502 left_path, right_path,
5504 BTRFS_COMPARE_TREE_DELETED,
5509 advance_right = ADVANCE;
5511 } else if (right_end_reached) {
5512 if (left_level == 0) {
5513 ret = changed_cb(left_root, right_root,
5514 left_path, right_path,
5516 BTRFS_COMPARE_TREE_NEW,
5521 advance_left = ADVANCE;
5525 if (left_level == 0 && right_level == 0) {
5526 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5528 ret = changed_cb(left_root, right_root,
5529 left_path, right_path,
5531 BTRFS_COMPARE_TREE_NEW,
5535 advance_left = ADVANCE;
5536 } else if (cmp > 0) {
5537 ret = changed_cb(left_root, right_root,
5538 left_path, right_path,
5540 BTRFS_COMPARE_TREE_DELETED,
5544 advance_right = ADVANCE;
5546 enum btrfs_compare_tree_result result;
5548 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5549 ret = tree_compare_item(left_path, right_path,
5552 result = BTRFS_COMPARE_TREE_CHANGED;
5554 result = BTRFS_COMPARE_TREE_SAME;
5555 ret = changed_cb(left_root, right_root,
5556 left_path, right_path,
5557 &left_key, result, ctx);
5560 advance_left = ADVANCE;
5561 advance_right = ADVANCE;
5563 } else if (left_level == right_level) {
5564 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5566 advance_left = ADVANCE;
5567 } else if (cmp > 0) {
5568 advance_right = ADVANCE;
5570 left_blockptr = btrfs_node_blockptr(
5571 left_path->nodes[left_level],
5572 left_path->slots[left_level]);
5573 right_blockptr = btrfs_node_blockptr(
5574 right_path->nodes[right_level],
5575 right_path->slots[right_level]);
5576 left_gen = btrfs_node_ptr_generation(
5577 left_path->nodes[left_level],
5578 left_path->slots[left_level]);
5579 right_gen = btrfs_node_ptr_generation(
5580 right_path->nodes[right_level],
5581 right_path->slots[right_level]);
5582 if (left_blockptr == right_blockptr &&
5583 left_gen == right_gen) {
5585 * As we're on a shared block, don't
5586 * allow to go deeper.
5588 advance_left = ADVANCE_ONLY_NEXT;
5589 advance_right = ADVANCE_ONLY_NEXT;
5591 advance_left = ADVANCE;
5592 advance_right = ADVANCE;
5595 } else if (left_level < right_level) {
5596 advance_right = ADVANCE;
5598 advance_left = ADVANCE;
5603 btrfs_free_path(left_path);
5604 btrfs_free_path(right_path);
5610 * this is similar to btrfs_next_leaf, but does not try to preserve
5611 * and fixup the path. It looks for and returns the next key in the
5612 * tree based on the current path and the min_trans parameters.
5614 * 0 is returned if another key is found, < 0 if there are any errors
5615 * and 1 is returned if there are no higher keys in the tree
5617 * path->keep_locks should be set to 1 on the search made before
5618 * calling this function.
5620 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5621 struct btrfs_key *key, int level, u64 min_trans)
5624 struct extent_buffer *c;
5626 WARN_ON(!path->keep_locks);
5627 while (level < BTRFS_MAX_LEVEL) {
5628 if (!path->nodes[level])
5631 slot = path->slots[level] + 1;
5632 c = path->nodes[level];
5634 if (slot >= btrfs_header_nritems(c)) {
5637 struct btrfs_key cur_key;
5638 if (level + 1 >= BTRFS_MAX_LEVEL ||
5639 !path->nodes[level + 1])
5642 if (path->locks[level + 1]) {
5647 slot = btrfs_header_nritems(c) - 1;
5649 btrfs_item_key_to_cpu(c, &cur_key, slot);
5651 btrfs_node_key_to_cpu(c, &cur_key, slot);
5653 orig_lowest = path->lowest_level;
5654 btrfs_release_path(path);
5655 path->lowest_level = level;
5656 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5658 path->lowest_level = orig_lowest;
5662 c = path->nodes[level];
5663 slot = path->slots[level];
5670 btrfs_item_key_to_cpu(c, key, slot);
5672 u64 gen = btrfs_node_ptr_generation(c, slot);
5674 if (gen < min_trans) {
5678 btrfs_node_key_to_cpu(c, key, slot);
5686 * search the tree again to find a leaf with greater keys
5687 * returns 0 if it found something or 1 if there are no greater leaves.
5688 * returns < 0 on io errors.
5690 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5692 return btrfs_next_old_leaf(root, path, 0);
5695 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5700 struct extent_buffer *c;
5701 struct extent_buffer *next;
5702 struct btrfs_key key;
5705 int old_spinning = path->leave_spinning;
5706 int next_rw_lock = 0;
5708 nritems = btrfs_header_nritems(path->nodes[0]);
5712 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5717 btrfs_release_path(path);
5719 path->keep_locks = 1;
5720 path->leave_spinning = 1;
5723 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5725 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5726 path->keep_locks = 0;
5731 nritems = btrfs_header_nritems(path->nodes[0]);
5733 * by releasing the path above we dropped all our locks. A balance
5734 * could have added more items next to the key that used to be
5735 * at the very end of the block. So, check again here and
5736 * advance the path if there are now more items available.
5738 if (nritems > 0 && path->slots[0] < nritems - 1) {
5745 * So the above check misses one case:
5746 * - after releasing the path above, someone has removed the item that
5747 * used to be at the very end of the block, and balance between leafs
5748 * gets another one with bigger key.offset to replace it.
5750 * This one should be returned as well, or we can get leaf corruption
5751 * later(esp. in __btrfs_drop_extents()).
5753 * And a bit more explanation about this check,
5754 * with ret > 0, the key isn't found, the path points to the slot
5755 * where it should be inserted, so the path->slots[0] item must be the
5758 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5763 while (level < BTRFS_MAX_LEVEL) {
5764 if (!path->nodes[level]) {
5769 slot = path->slots[level] + 1;
5770 c = path->nodes[level];
5771 if (slot >= btrfs_header_nritems(c)) {
5773 if (level == BTRFS_MAX_LEVEL) {
5781 btrfs_tree_unlock_rw(next, next_rw_lock);
5782 free_extent_buffer(next);
5786 next_rw_lock = path->locks[level];
5787 ret = read_block_for_search(NULL, root, path, &next, level,
5793 btrfs_release_path(path);
5797 if (!path->skip_locking) {
5798 ret = btrfs_try_tree_read_lock(next);
5799 if (!ret && time_seq) {
5801 * If we don't get the lock, we may be racing
5802 * with push_leaf_left, holding that lock while
5803 * itself waiting for the leaf we've currently
5804 * locked. To solve this situation, we give up
5805 * on our lock and cycle.
5807 free_extent_buffer(next);
5808 btrfs_release_path(path);
5813 btrfs_set_path_blocking(path);
5814 btrfs_tree_read_lock(next);
5815 btrfs_clear_path_blocking(path, next,
5818 next_rw_lock = BTRFS_READ_LOCK;
5822 path->slots[level] = slot;
5825 c = path->nodes[level];
5826 if (path->locks[level])
5827 btrfs_tree_unlock_rw(c, path->locks[level]);
5829 free_extent_buffer(c);
5830 path->nodes[level] = next;
5831 path->slots[level] = 0;
5832 if (!path->skip_locking)
5833 path->locks[level] = next_rw_lock;
5837 ret = read_block_for_search(NULL, root, path, &next, level,
5843 btrfs_release_path(path);
5847 if (!path->skip_locking) {
5848 ret = btrfs_try_tree_read_lock(next);
5850 btrfs_set_path_blocking(path);
5851 btrfs_tree_read_lock(next);
5852 btrfs_clear_path_blocking(path, next,
5855 next_rw_lock = BTRFS_READ_LOCK;
5860 unlock_up(path, 0, 1, 0, NULL);
5861 path->leave_spinning = old_spinning;
5863 btrfs_set_path_blocking(path);
5869 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5870 * searching until it gets past min_objectid or finds an item of 'type'
5872 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5874 int btrfs_previous_item(struct btrfs_root *root,
5875 struct btrfs_path *path, u64 min_objectid,
5878 struct btrfs_key found_key;
5879 struct extent_buffer *leaf;
5884 if (path->slots[0] == 0) {
5885 btrfs_set_path_blocking(path);
5886 ret = btrfs_prev_leaf(root, path);
5892 leaf = path->nodes[0];
5893 nritems = btrfs_header_nritems(leaf);
5896 if (path->slots[0] == nritems)
5899 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5900 if (found_key.objectid < min_objectid)
5902 if (found_key.type == type)
5904 if (found_key.objectid == min_objectid &&
5905 found_key.type < type)
5912 * search in extent tree to find a previous Metadata/Data extent item with
5915 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5917 int btrfs_previous_extent_item(struct btrfs_root *root,
5918 struct btrfs_path *path, u64 min_objectid)
5920 struct btrfs_key found_key;
5921 struct extent_buffer *leaf;
5926 if (path->slots[0] == 0) {
5927 btrfs_set_path_blocking(path);
5928 ret = btrfs_prev_leaf(root, path);
5934 leaf = path->nodes[0];
5935 nritems = btrfs_header_nritems(leaf);
5938 if (path->slots[0] == nritems)
5941 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5942 if (found_key.objectid < min_objectid)
5944 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5945 found_key.type == BTRFS_METADATA_ITEM_KEY)
5947 if (found_key.objectid == min_objectid &&
5948 found_key.type < BTRFS_EXTENT_ITEM_KEY)