xref: /openbmc/linux/fs/btrfs/ctree.c (revision e6dec923)
1 /*
2  * Copyright (C) 2007,2008 Oracle.  All rights reserved.
3  *
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.
7  *
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.
12  *
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.
17  */
18 
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/rbtree.h>
22 #include <linux/mm.h>
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "transaction.h"
26 #include "print-tree.h"
27 #include "locking.h"
28 
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 *root,
32 		      const struct btrfs_key *ins_key, struct btrfs_path *path,
33 		      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,
43 		    int level, int slot);
44 static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
45 				 struct extent_buffer *eb);
46 
47 struct btrfs_path *btrfs_alloc_path(void)
48 {
49 	return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
50 }
51 
52 /*
53  * set all locked nodes in the path to blocking locks.  This should
54  * be done before scheduling
55  */
56 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
57 {
58 	int i;
59 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
60 		if (!p->nodes[i] || !p->locks[i])
61 			continue;
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;
67 	}
68 }
69 
70 /*
71  * reset all the locked nodes in the patch to spinning locks.
72  *
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
76  * for held
77  */
78 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
79 					struct extent_buffer *held, int held_rw)
80 {
81 	int i;
82 
83 	if (held) {
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;
89 	}
90 	btrfs_set_path_blocking(p);
91 
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;
99 		}
100 	}
101 
102 	if (held)
103 		btrfs_clear_lock_blocking_rw(held, held_rw);
104 }
105 
106 /* this also releases the path */
107 void btrfs_free_path(struct btrfs_path *p)
108 {
109 	if (!p)
110 		return;
111 	btrfs_release_path(p);
112 	kmem_cache_free(btrfs_path_cachep, p);
113 }
114 
115 /*
116  * path release drops references on the extent buffers in the path
117  * and it drops any locks held by this path
118  *
119  * It is safe to call this on paths that no locks or extent buffers held.
120  */
121 noinline void btrfs_release_path(struct btrfs_path *p)
122 {
123 	int i;
124 
125 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
126 		p->slots[i] = 0;
127 		if (!p->nodes[i])
128 			continue;
129 		if (p->locks[i]) {
130 			btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
131 			p->locks[i] = 0;
132 		}
133 		free_extent_buffer(p->nodes[i]);
134 		p->nodes[i] = NULL;
135 	}
136 }
137 
138 /*
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
142  * looping required.
143  *
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.
147  */
148 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
149 {
150 	struct extent_buffer *eb;
151 
152 	while (1) {
153 		rcu_read_lock();
154 		eb = rcu_dereference(root->node);
155 
156 		/*
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.
161 		 */
162 		if (atomic_inc_not_zero(&eb->refs)) {
163 			rcu_read_unlock();
164 			break;
165 		}
166 		rcu_read_unlock();
167 		synchronize_rcu();
168 	}
169 	return eb;
170 }
171 
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.
175  */
176 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
177 {
178 	struct extent_buffer *eb;
179 
180 	while (1) {
181 		eb = btrfs_root_node(root);
182 		btrfs_tree_lock(eb);
183 		if (eb == root->node)
184 			break;
185 		btrfs_tree_unlock(eb);
186 		free_extent_buffer(eb);
187 	}
188 	return eb;
189 }
190 
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.
194  */
195 static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
196 {
197 	struct extent_buffer *eb;
198 
199 	while (1) {
200 		eb = btrfs_root_node(root);
201 		btrfs_tree_read_lock(eb);
202 		if (eb == root->node)
203 			break;
204 		btrfs_tree_read_unlock(eb);
205 		free_extent_buffer(eb);
206 	}
207 	return eb;
208 }
209 
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.
213  */
214 static void add_root_to_dirty_list(struct btrfs_root *root)
215 {
216 	struct btrfs_fs_info *fs_info = root->fs_info;
217 
218 	if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
219 	    !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
220 		return;
221 
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);
228 		else
229 			list_move(&root->dirty_list,
230 				  &fs_info->dirty_cowonly_roots);
231 	}
232 	spin_unlock(&fs_info->trans_lock);
233 }
234 
235 /*
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.
239  */
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)
244 {
245 	struct btrfs_fs_info *fs_info = root->fs_info;
246 	struct extent_buffer *cow;
247 	int ret = 0;
248 	int level;
249 	struct btrfs_disk_key disk_key;
250 
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);
255 
256 	level = btrfs_header_level(buf);
257 	if (level == 0)
258 		btrfs_item_key(buf, &disk_key, 0);
259 	else
260 		btrfs_node_key(buf, &disk_key, 0);
261 
262 	cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
263 			&disk_key, level, buf->start, 0);
264 	if (IS_ERR(cow))
265 		return PTR_ERR(cow);
266 
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);
275 	else
276 		btrfs_set_header_owner(cow, new_root_objectid);
277 
278 	write_extent_buffer_fsid(cow, fs_info->fsid);
279 
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);
283 	else
284 		ret = btrfs_inc_ref(trans, root, cow, 0);
285 
286 	if (ret)
287 		return ret;
288 
289 	btrfs_mark_buffer_dirty(cow);
290 	*cow_ret = cow;
291 	return 0;
292 }
293 
294 enum mod_log_op {
295 	MOD_LOG_KEY_REPLACE,
296 	MOD_LOG_KEY_ADD,
297 	MOD_LOG_KEY_REMOVE,
298 	MOD_LOG_KEY_REMOVE_WHILE_FREEING,
299 	MOD_LOG_KEY_REMOVE_WHILE_MOVING,
300 	MOD_LOG_MOVE_KEYS,
301 	MOD_LOG_ROOT_REPLACE,
302 };
303 
304 struct tree_mod_move {
305 	int dst_slot;
306 	int nr_items;
307 };
308 
309 struct tree_mod_root {
310 	u64 logical;
311 	u8 level;
312 };
313 
314 struct tree_mod_elem {
315 	struct rb_node node;
316 	u64 logical;
317 	u64 seq;
318 	enum mod_log_op op;
319 
320 	/* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
321 	int slot;
322 
323 	/* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
324 	u64 generation;
325 
326 	/* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
327 	struct btrfs_disk_key key;
328 	u64 blockptr;
329 
330 	/* this is used for op == MOD_LOG_MOVE_KEYS */
331 	struct tree_mod_move move;
332 
333 	/* this is used for op == MOD_LOG_ROOT_REPLACE */
334 	struct tree_mod_root old_root;
335 };
336 
337 static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
338 {
339 	read_lock(&fs_info->tree_mod_log_lock);
340 }
341 
342 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
343 {
344 	read_unlock(&fs_info->tree_mod_log_lock);
345 }
346 
347 static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
348 {
349 	write_lock(&fs_info->tree_mod_log_lock);
350 }
351 
352 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
353 {
354 	write_unlock(&fs_info->tree_mod_log_lock);
355 }
356 
357 /*
358  * Pull a new tree mod seq number for our operation.
359  */
360 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
361 {
362 	return atomic64_inc_return(&fs_info->tree_mod_seq);
363 }
364 
365 /*
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
371  * blocker was added.
372  */
373 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
374 			   struct seq_list *elem)
375 {
376 	tree_mod_log_write_lock(fs_info);
377 	spin_lock(&fs_info->tree_mod_seq_lock);
378 	if (!elem->seq) {
379 		elem->seq = btrfs_inc_tree_mod_seq(fs_info);
380 		list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
381 	}
382 	spin_unlock(&fs_info->tree_mod_seq_lock);
383 	tree_mod_log_write_unlock(fs_info);
384 
385 	return elem->seq;
386 }
387 
388 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
389 			    struct seq_list *elem)
390 {
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;
398 
399 	if (!seq_putting)
400 		return;
401 
402 	spin_lock(&fs_info->tree_mod_seq_lock);
403 	list_del(&elem->list);
404 	elem->seq = 0;
405 
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) {
409 				/*
410 				 * blocker with lower sequence number exists, we
411 				 * cannot remove anything from the log
412 				 */
413 				spin_unlock(&fs_info->tree_mod_seq_lock);
414 				return;
415 			}
416 			min_seq = cur_elem->seq;
417 		}
418 	}
419 	spin_unlock(&fs_info->tree_mod_seq_lock);
420 
421 	/*
422 	 * anything that's lower than the lowest existing (read: blocked)
423 	 * sequence number can be removed from the tree.
424 	 */
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 = rb_entry(node, struct tree_mod_elem, node);
430 		if (tm->seq > min_seq)
431 			continue;
432 		rb_erase(node, tm_root);
433 		kfree(tm);
434 	}
435 	tree_mod_log_write_unlock(fs_info);
436 }
437 
438 /*
439  * key order of the log:
440  *       node/leaf start address -> sequence
441  *
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.
445  *
446  * Note: must be called with write lock (tree_mod_log_write_lock).
447  */
448 static noinline int
449 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
450 {
451 	struct rb_root *tm_root;
452 	struct rb_node **new;
453 	struct rb_node *parent = NULL;
454 	struct tree_mod_elem *cur;
455 
456 	tm->seq = btrfs_inc_tree_mod_seq(fs_info);
457 
458 	tm_root = &fs_info->tree_mod_log;
459 	new = &tm_root->rb_node;
460 	while (*new) {
461 		cur = rb_entry(*new, struct tree_mod_elem, node);
462 		parent = *new;
463 		if (cur->logical < tm->logical)
464 			new = &((*new)->rb_left);
465 		else if (cur->logical > tm->logical)
466 			new = &((*new)->rb_right);
467 		else if (cur->seq < tm->seq)
468 			new = &((*new)->rb_left);
469 		else if (cur->seq > tm->seq)
470 			new = &((*new)->rb_right);
471 		else
472 			return -EEXIST;
473 	}
474 
475 	rb_link_node(&tm->node, parent, new);
476 	rb_insert_color(&tm->node, tm_root);
477 	return 0;
478 }
479 
480 /*
481  * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
482  * returns zero with the tree_mod_log_lock acquired. The caller must hold
483  * this until all tree mod log insertions are recorded in the rb tree and then
484  * call tree_mod_log_write_unlock() to release.
485  */
486 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
487 				    struct extent_buffer *eb) {
488 	smp_mb();
489 	if (list_empty(&(fs_info)->tree_mod_seq_list))
490 		return 1;
491 	if (eb && btrfs_header_level(eb) == 0)
492 		return 1;
493 
494 	tree_mod_log_write_lock(fs_info);
495 	if (list_empty(&(fs_info)->tree_mod_seq_list)) {
496 		tree_mod_log_write_unlock(fs_info);
497 		return 1;
498 	}
499 
500 	return 0;
501 }
502 
503 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
504 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
505 				    struct extent_buffer *eb)
506 {
507 	smp_mb();
508 	if (list_empty(&(fs_info)->tree_mod_seq_list))
509 		return 0;
510 	if (eb && btrfs_header_level(eb) == 0)
511 		return 0;
512 
513 	return 1;
514 }
515 
516 static struct tree_mod_elem *
517 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
518 		    enum mod_log_op op, gfp_t flags)
519 {
520 	struct tree_mod_elem *tm;
521 
522 	tm = kzalloc(sizeof(*tm), flags);
523 	if (!tm)
524 		return NULL;
525 
526 	tm->logical = eb->start;
527 	if (op != MOD_LOG_KEY_ADD) {
528 		btrfs_node_key(eb, &tm->key, slot);
529 		tm->blockptr = btrfs_node_blockptr(eb, slot);
530 	}
531 	tm->op = op;
532 	tm->slot = slot;
533 	tm->generation = btrfs_node_ptr_generation(eb, slot);
534 	RB_CLEAR_NODE(&tm->node);
535 
536 	return tm;
537 }
538 
539 static noinline int
540 tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
541 			struct extent_buffer *eb, int slot,
542 			enum mod_log_op op, gfp_t flags)
543 {
544 	struct tree_mod_elem *tm;
545 	int ret;
546 
547 	if (!tree_mod_need_log(fs_info, eb))
548 		return 0;
549 
550 	tm = alloc_tree_mod_elem(eb, slot, op, flags);
551 	if (!tm)
552 		return -ENOMEM;
553 
554 	if (tree_mod_dont_log(fs_info, eb)) {
555 		kfree(tm);
556 		return 0;
557 	}
558 
559 	ret = __tree_mod_log_insert(fs_info, tm);
560 	tree_mod_log_write_unlock(fs_info);
561 	if (ret)
562 		kfree(tm);
563 
564 	return ret;
565 }
566 
567 static noinline int
568 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
569 			 struct extent_buffer *eb, int dst_slot, int src_slot,
570 			 int nr_items)
571 {
572 	struct tree_mod_elem *tm = NULL;
573 	struct tree_mod_elem **tm_list = NULL;
574 	int ret = 0;
575 	int i;
576 	int locked = 0;
577 
578 	if (!tree_mod_need_log(fs_info, eb))
579 		return 0;
580 
581 	tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
582 	if (!tm_list)
583 		return -ENOMEM;
584 
585 	tm = kzalloc(sizeof(*tm), GFP_NOFS);
586 	if (!tm) {
587 		ret = -ENOMEM;
588 		goto free_tms;
589 	}
590 
591 	tm->logical = eb->start;
592 	tm->slot = src_slot;
593 	tm->move.dst_slot = dst_slot;
594 	tm->move.nr_items = nr_items;
595 	tm->op = MOD_LOG_MOVE_KEYS;
596 
597 	for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
598 		tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
599 		    MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
600 		if (!tm_list[i]) {
601 			ret = -ENOMEM;
602 			goto free_tms;
603 		}
604 	}
605 
606 	if (tree_mod_dont_log(fs_info, eb))
607 		goto free_tms;
608 	locked = 1;
609 
610 	/*
611 	 * When we override something during the move, we log these removals.
612 	 * This can only happen when we move towards the beginning of the
613 	 * buffer, i.e. dst_slot < src_slot.
614 	 */
615 	for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
616 		ret = __tree_mod_log_insert(fs_info, tm_list[i]);
617 		if (ret)
618 			goto free_tms;
619 	}
620 
621 	ret = __tree_mod_log_insert(fs_info, tm);
622 	if (ret)
623 		goto free_tms;
624 	tree_mod_log_write_unlock(fs_info);
625 	kfree(tm_list);
626 
627 	return 0;
628 free_tms:
629 	for (i = 0; i < nr_items; i++) {
630 		if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
631 			rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
632 		kfree(tm_list[i]);
633 	}
634 	if (locked)
635 		tree_mod_log_write_unlock(fs_info);
636 	kfree(tm_list);
637 	kfree(tm);
638 
639 	return ret;
640 }
641 
642 static inline int
643 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
644 		       struct tree_mod_elem **tm_list,
645 		       int nritems)
646 {
647 	int i, j;
648 	int ret;
649 
650 	for (i = nritems - 1; i >= 0; i--) {
651 		ret = __tree_mod_log_insert(fs_info, tm_list[i]);
652 		if (ret) {
653 			for (j = nritems - 1; j > i; j--)
654 				rb_erase(&tm_list[j]->node,
655 					 &fs_info->tree_mod_log);
656 			return ret;
657 		}
658 	}
659 
660 	return 0;
661 }
662 
663 static noinline int
664 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
665 			 struct extent_buffer *old_root,
666 			 struct extent_buffer *new_root,
667 			 int log_removal)
668 {
669 	struct tree_mod_elem *tm = NULL;
670 	struct tree_mod_elem **tm_list = NULL;
671 	int nritems = 0;
672 	int ret = 0;
673 	int i;
674 
675 	if (!tree_mod_need_log(fs_info, NULL))
676 		return 0;
677 
678 	if (log_removal && btrfs_header_level(old_root) > 0) {
679 		nritems = btrfs_header_nritems(old_root);
680 		tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
681 				  GFP_NOFS);
682 		if (!tm_list) {
683 			ret = -ENOMEM;
684 			goto free_tms;
685 		}
686 		for (i = 0; i < nritems; i++) {
687 			tm_list[i] = alloc_tree_mod_elem(old_root, i,
688 			    MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
689 			if (!tm_list[i]) {
690 				ret = -ENOMEM;
691 				goto free_tms;
692 			}
693 		}
694 	}
695 
696 	tm = kzalloc(sizeof(*tm), GFP_NOFS);
697 	if (!tm) {
698 		ret = -ENOMEM;
699 		goto free_tms;
700 	}
701 
702 	tm->logical = new_root->start;
703 	tm->old_root.logical = old_root->start;
704 	tm->old_root.level = btrfs_header_level(old_root);
705 	tm->generation = btrfs_header_generation(old_root);
706 	tm->op = MOD_LOG_ROOT_REPLACE;
707 
708 	if (tree_mod_dont_log(fs_info, NULL))
709 		goto free_tms;
710 
711 	if (tm_list)
712 		ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
713 	if (!ret)
714 		ret = __tree_mod_log_insert(fs_info, tm);
715 
716 	tree_mod_log_write_unlock(fs_info);
717 	if (ret)
718 		goto free_tms;
719 	kfree(tm_list);
720 
721 	return ret;
722 
723 free_tms:
724 	if (tm_list) {
725 		for (i = 0; i < nritems; i++)
726 			kfree(tm_list[i]);
727 		kfree(tm_list);
728 	}
729 	kfree(tm);
730 
731 	return ret;
732 }
733 
734 static struct tree_mod_elem *
735 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
736 		      int smallest)
737 {
738 	struct rb_root *tm_root;
739 	struct rb_node *node;
740 	struct tree_mod_elem *cur = NULL;
741 	struct tree_mod_elem *found = NULL;
742 
743 	tree_mod_log_read_lock(fs_info);
744 	tm_root = &fs_info->tree_mod_log;
745 	node = tm_root->rb_node;
746 	while (node) {
747 		cur = rb_entry(node, struct tree_mod_elem, node);
748 		if (cur->logical < start) {
749 			node = node->rb_left;
750 		} else if (cur->logical > start) {
751 			node = node->rb_right;
752 		} else if (cur->seq < min_seq) {
753 			node = node->rb_left;
754 		} else if (!smallest) {
755 			/* we want the node with the highest seq */
756 			if (found)
757 				BUG_ON(found->seq > cur->seq);
758 			found = cur;
759 			node = node->rb_left;
760 		} else if (cur->seq > min_seq) {
761 			/* we want the node with the smallest seq */
762 			if (found)
763 				BUG_ON(found->seq < cur->seq);
764 			found = cur;
765 			node = node->rb_right;
766 		} else {
767 			found = cur;
768 			break;
769 		}
770 	}
771 	tree_mod_log_read_unlock(fs_info);
772 
773 	return found;
774 }
775 
776 /*
777  * this returns the element from the log with the smallest time sequence
778  * value that's in the log (the oldest log item). any element with a time
779  * sequence lower than min_seq will be ignored.
780  */
781 static struct tree_mod_elem *
782 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
783 			   u64 min_seq)
784 {
785 	return __tree_mod_log_search(fs_info, start, min_seq, 1);
786 }
787 
788 /*
789  * this returns the element from the log with the largest time sequence
790  * value that's in the log (the most recent log item). any element with
791  * a time sequence lower than min_seq will be ignored.
792  */
793 static struct tree_mod_elem *
794 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
795 {
796 	return __tree_mod_log_search(fs_info, start, min_seq, 0);
797 }
798 
799 static noinline int
800 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
801 		     struct extent_buffer *src, unsigned long dst_offset,
802 		     unsigned long src_offset, int nr_items)
803 {
804 	int ret = 0;
805 	struct tree_mod_elem **tm_list = NULL;
806 	struct tree_mod_elem **tm_list_add, **tm_list_rem;
807 	int i;
808 	int locked = 0;
809 
810 	if (!tree_mod_need_log(fs_info, NULL))
811 		return 0;
812 
813 	if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
814 		return 0;
815 
816 	tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
817 			  GFP_NOFS);
818 	if (!tm_list)
819 		return -ENOMEM;
820 
821 	tm_list_add = tm_list;
822 	tm_list_rem = tm_list + nr_items;
823 	for (i = 0; i < nr_items; i++) {
824 		tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
825 		    MOD_LOG_KEY_REMOVE, GFP_NOFS);
826 		if (!tm_list_rem[i]) {
827 			ret = -ENOMEM;
828 			goto free_tms;
829 		}
830 
831 		tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
832 		    MOD_LOG_KEY_ADD, GFP_NOFS);
833 		if (!tm_list_add[i]) {
834 			ret = -ENOMEM;
835 			goto free_tms;
836 		}
837 	}
838 
839 	if (tree_mod_dont_log(fs_info, NULL))
840 		goto free_tms;
841 	locked = 1;
842 
843 	for (i = 0; i < nr_items; i++) {
844 		ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
845 		if (ret)
846 			goto free_tms;
847 		ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
848 		if (ret)
849 			goto free_tms;
850 	}
851 
852 	tree_mod_log_write_unlock(fs_info);
853 	kfree(tm_list);
854 
855 	return 0;
856 
857 free_tms:
858 	for (i = 0; i < nr_items * 2; i++) {
859 		if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
860 			rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
861 		kfree(tm_list[i]);
862 	}
863 	if (locked)
864 		tree_mod_log_write_unlock(fs_info);
865 	kfree(tm_list);
866 
867 	return ret;
868 }
869 
870 static inline void
871 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
872 		     int dst_offset, int src_offset, int nr_items)
873 {
874 	int ret;
875 	ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
876 				       nr_items);
877 	BUG_ON(ret < 0);
878 }
879 
880 static noinline void
881 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
882 			  struct extent_buffer *eb, int slot, int atomic)
883 {
884 	int ret;
885 
886 	ret = tree_mod_log_insert_key(fs_info, eb, slot,
887 					MOD_LOG_KEY_REPLACE,
888 					atomic ? GFP_ATOMIC : GFP_NOFS);
889 	BUG_ON(ret < 0);
890 }
891 
892 static noinline int
893 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
894 {
895 	struct tree_mod_elem **tm_list = NULL;
896 	int nritems = 0;
897 	int i;
898 	int ret = 0;
899 
900 	if (btrfs_header_level(eb) == 0)
901 		return 0;
902 
903 	if (!tree_mod_need_log(fs_info, NULL))
904 		return 0;
905 
906 	nritems = btrfs_header_nritems(eb);
907 	tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
908 	if (!tm_list)
909 		return -ENOMEM;
910 
911 	for (i = 0; i < nritems; i++) {
912 		tm_list[i] = alloc_tree_mod_elem(eb, i,
913 		    MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
914 		if (!tm_list[i]) {
915 			ret = -ENOMEM;
916 			goto free_tms;
917 		}
918 	}
919 
920 	if (tree_mod_dont_log(fs_info, eb))
921 		goto free_tms;
922 
923 	ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
924 	tree_mod_log_write_unlock(fs_info);
925 	if (ret)
926 		goto free_tms;
927 	kfree(tm_list);
928 
929 	return 0;
930 
931 free_tms:
932 	for (i = 0; i < nritems; i++)
933 		kfree(tm_list[i]);
934 	kfree(tm_list);
935 
936 	return ret;
937 }
938 
939 static noinline void
940 tree_mod_log_set_root_pointer(struct btrfs_root *root,
941 			      struct extent_buffer *new_root_node,
942 			      int log_removal)
943 {
944 	int ret;
945 	ret = tree_mod_log_insert_root(root->fs_info, root->node,
946 				       new_root_node, log_removal);
947 	BUG_ON(ret < 0);
948 }
949 
950 /*
951  * check if the tree block can be shared by multiple trees
952  */
953 int btrfs_block_can_be_shared(struct btrfs_root *root,
954 			      struct extent_buffer *buf)
955 {
956 	/*
957 	 * Tree blocks not in reference counted trees and tree roots
958 	 * are never shared. If a block was allocated after the last
959 	 * snapshot and the block was not allocated by tree relocation,
960 	 * we know the block is not shared.
961 	 */
962 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
963 	    buf != root->node && buf != root->commit_root &&
964 	    (btrfs_header_generation(buf) <=
965 	     btrfs_root_last_snapshot(&root->root_item) ||
966 	     btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
967 		return 1;
968 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
969 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
970 	    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
971 		return 1;
972 #endif
973 	return 0;
974 }
975 
976 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
977 				       struct btrfs_root *root,
978 				       struct extent_buffer *buf,
979 				       struct extent_buffer *cow,
980 				       int *last_ref)
981 {
982 	struct btrfs_fs_info *fs_info = root->fs_info;
983 	u64 refs;
984 	u64 owner;
985 	u64 flags;
986 	u64 new_flags = 0;
987 	int ret;
988 
989 	/*
990 	 * Backrefs update rules:
991 	 *
992 	 * Always use full backrefs for extent pointers in tree block
993 	 * allocated by tree relocation.
994 	 *
995 	 * If a shared tree block is no longer referenced by its owner
996 	 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
997 	 * use full backrefs for extent pointers in tree block.
998 	 *
999 	 * If a tree block is been relocating
1000 	 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1001 	 * use full backrefs for extent pointers in tree block.
1002 	 * The reason for this is some operations (such as drop tree)
1003 	 * are only allowed for blocks use full backrefs.
1004 	 */
1005 
1006 	if (btrfs_block_can_be_shared(root, buf)) {
1007 		ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
1008 					       btrfs_header_level(buf), 1,
1009 					       &refs, &flags);
1010 		if (ret)
1011 			return ret;
1012 		if (refs == 0) {
1013 			ret = -EROFS;
1014 			btrfs_handle_fs_error(fs_info, ret, NULL);
1015 			return ret;
1016 		}
1017 	} else {
1018 		refs = 1;
1019 		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1020 		    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1021 			flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1022 		else
1023 			flags = 0;
1024 	}
1025 
1026 	owner = btrfs_header_owner(buf);
1027 	BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
1028 	       !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
1029 
1030 	if (refs > 1) {
1031 		if ((owner == root->root_key.objectid ||
1032 		     root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
1033 		    !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1034 			ret = btrfs_inc_ref(trans, root, buf, 1);
1035 			BUG_ON(ret); /* -ENOMEM */
1036 
1037 			if (root->root_key.objectid ==
1038 			    BTRFS_TREE_RELOC_OBJECTID) {
1039 				ret = btrfs_dec_ref(trans, root, buf, 0);
1040 				BUG_ON(ret); /* -ENOMEM */
1041 				ret = btrfs_inc_ref(trans, root, cow, 1);
1042 				BUG_ON(ret); /* -ENOMEM */
1043 			}
1044 			new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
1045 		} else {
1046 
1047 			if (root->root_key.objectid ==
1048 			    BTRFS_TREE_RELOC_OBJECTID)
1049 				ret = btrfs_inc_ref(trans, root, cow, 1);
1050 			else
1051 				ret = btrfs_inc_ref(trans, root, cow, 0);
1052 			BUG_ON(ret); /* -ENOMEM */
1053 		}
1054 		if (new_flags != 0) {
1055 			int level = btrfs_header_level(buf);
1056 
1057 			ret = btrfs_set_disk_extent_flags(trans, fs_info,
1058 							  buf->start,
1059 							  buf->len,
1060 							  new_flags, level, 0);
1061 			if (ret)
1062 				return ret;
1063 		}
1064 	} else {
1065 		if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
1066 			if (root->root_key.objectid ==
1067 			    BTRFS_TREE_RELOC_OBJECTID)
1068 				ret = btrfs_inc_ref(trans, root, cow, 1);
1069 			else
1070 				ret = btrfs_inc_ref(trans, root, cow, 0);
1071 			BUG_ON(ret); /* -ENOMEM */
1072 			ret = btrfs_dec_ref(trans, root, buf, 1);
1073 			BUG_ON(ret); /* -ENOMEM */
1074 		}
1075 		clean_tree_block(fs_info, buf);
1076 		*last_ref = 1;
1077 	}
1078 	return 0;
1079 }
1080 
1081 /*
1082  * does the dirty work in cow of a single block.  The parent block (if
1083  * supplied) is updated to point to the new cow copy.  The new buffer is marked
1084  * dirty and returned locked.  If you modify the block it needs to be marked
1085  * dirty again.
1086  *
1087  * search_start -- an allocation hint for the new block
1088  *
1089  * empty_size -- a hint that you plan on doing more cow.  This is the size in
1090  * bytes the allocator should try to find free next to the block it returns.
1091  * This is just a hint and may be ignored by the allocator.
1092  */
1093 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1094 			     struct btrfs_root *root,
1095 			     struct extent_buffer *buf,
1096 			     struct extent_buffer *parent, int parent_slot,
1097 			     struct extent_buffer **cow_ret,
1098 			     u64 search_start, u64 empty_size)
1099 {
1100 	struct btrfs_fs_info *fs_info = root->fs_info;
1101 	struct btrfs_disk_key disk_key;
1102 	struct extent_buffer *cow;
1103 	int level, ret;
1104 	int last_ref = 0;
1105 	int unlock_orig = 0;
1106 	u64 parent_start = 0;
1107 
1108 	if (*cow_ret == buf)
1109 		unlock_orig = 1;
1110 
1111 	btrfs_assert_tree_locked(buf);
1112 
1113 	WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1114 		trans->transid != fs_info->running_transaction->transid);
1115 	WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1116 		trans->transid != root->last_trans);
1117 
1118 	level = btrfs_header_level(buf);
1119 
1120 	if (level == 0)
1121 		btrfs_item_key(buf, &disk_key, 0);
1122 	else
1123 		btrfs_node_key(buf, &disk_key, 0);
1124 
1125 	if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
1126 		parent_start = parent->start;
1127 
1128 	cow = btrfs_alloc_tree_block(trans, root, parent_start,
1129 			root->root_key.objectid, &disk_key, level,
1130 			search_start, empty_size);
1131 	if (IS_ERR(cow))
1132 		return PTR_ERR(cow);
1133 
1134 	/* cow is set to blocking by btrfs_init_new_buffer */
1135 
1136 	copy_extent_buffer_full(cow, buf);
1137 	btrfs_set_header_bytenr(cow, cow->start);
1138 	btrfs_set_header_generation(cow, trans->transid);
1139 	btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1140 	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1141 				     BTRFS_HEADER_FLAG_RELOC);
1142 	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1143 		btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1144 	else
1145 		btrfs_set_header_owner(cow, root->root_key.objectid);
1146 
1147 	write_extent_buffer_fsid(cow, fs_info->fsid);
1148 
1149 	ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1150 	if (ret) {
1151 		btrfs_abort_transaction(trans, ret);
1152 		return ret;
1153 	}
1154 
1155 	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1156 		ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1157 		if (ret) {
1158 			btrfs_abort_transaction(trans, ret);
1159 			return ret;
1160 		}
1161 	}
1162 
1163 	if (buf == root->node) {
1164 		WARN_ON(parent && parent != buf);
1165 		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1166 		    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1167 			parent_start = buf->start;
1168 
1169 		extent_buffer_get(cow);
1170 		tree_mod_log_set_root_pointer(root, cow, 1);
1171 		rcu_assign_pointer(root->node, cow);
1172 
1173 		btrfs_free_tree_block(trans, root, buf, parent_start,
1174 				      last_ref);
1175 		free_extent_buffer(buf);
1176 		add_root_to_dirty_list(root);
1177 	} else {
1178 		WARN_ON(trans->transid != btrfs_header_generation(parent));
1179 		tree_mod_log_insert_key(fs_info, parent, parent_slot,
1180 					MOD_LOG_KEY_REPLACE, GFP_NOFS);
1181 		btrfs_set_node_blockptr(parent, parent_slot,
1182 					cow->start);
1183 		btrfs_set_node_ptr_generation(parent, parent_slot,
1184 					      trans->transid);
1185 		btrfs_mark_buffer_dirty(parent);
1186 		if (last_ref) {
1187 			ret = tree_mod_log_free_eb(fs_info, buf);
1188 			if (ret) {
1189 				btrfs_abort_transaction(trans, ret);
1190 				return ret;
1191 			}
1192 		}
1193 		btrfs_free_tree_block(trans, root, buf, parent_start,
1194 				      last_ref);
1195 	}
1196 	if (unlock_orig)
1197 		btrfs_tree_unlock(buf);
1198 	free_extent_buffer_stale(buf);
1199 	btrfs_mark_buffer_dirty(cow);
1200 	*cow_ret = cow;
1201 	return 0;
1202 }
1203 
1204 /*
1205  * returns the logical address of the oldest predecessor of the given root.
1206  * entries older than time_seq are ignored.
1207  */
1208 static struct tree_mod_elem *
1209 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1210 			   struct extent_buffer *eb_root, u64 time_seq)
1211 {
1212 	struct tree_mod_elem *tm;
1213 	struct tree_mod_elem *found = NULL;
1214 	u64 root_logical = eb_root->start;
1215 	int looped = 0;
1216 
1217 	if (!time_seq)
1218 		return NULL;
1219 
1220 	/*
1221 	 * the very last operation that's logged for a root is the
1222 	 * replacement operation (if it is replaced at all). this has
1223 	 * the logical address of the *new* root, making it the very
1224 	 * first operation that's logged for this root.
1225 	 */
1226 	while (1) {
1227 		tm = tree_mod_log_search_oldest(fs_info, root_logical,
1228 						time_seq);
1229 		if (!looped && !tm)
1230 			return NULL;
1231 		/*
1232 		 * if there are no tree operation for the oldest root, we simply
1233 		 * return it. this should only happen if that (old) root is at
1234 		 * level 0.
1235 		 */
1236 		if (!tm)
1237 			break;
1238 
1239 		/*
1240 		 * if there's an operation that's not a root replacement, we
1241 		 * found the oldest version of our root. normally, we'll find a
1242 		 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1243 		 */
1244 		if (tm->op != MOD_LOG_ROOT_REPLACE)
1245 			break;
1246 
1247 		found = tm;
1248 		root_logical = tm->old_root.logical;
1249 		looped = 1;
1250 	}
1251 
1252 	/* if there's no old root to return, return what we found instead */
1253 	if (!found)
1254 		found = tm;
1255 
1256 	return found;
1257 }
1258 
1259 /*
1260  * tm is a pointer to the first operation to rewind within eb. then, all
1261  * previous operations will be rewound (until we reach something older than
1262  * time_seq).
1263  */
1264 static void
1265 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1266 		      u64 time_seq, struct tree_mod_elem *first_tm)
1267 {
1268 	u32 n;
1269 	struct rb_node *next;
1270 	struct tree_mod_elem *tm = first_tm;
1271 	unsigned long o_dst;
1272 	unsigned long o_src;
1273 	unsigned long p_size = sizeof(struct btrfs_key_ptr);
1274 
1275 	n = btrfs_header_nritems(eb);
1276 	tree_mod_log_read_lock(fs_info);
1277 	while (tm && tm->seq >= time_seq) {
1278 		/*
1279 		 * all the operations are recorded with the operator used for
1280 		 * the modification. as we're going backwards, we do the
1281 		 * opposite of each operation here.
1282 		 */
1283 		switch (tm->op) {
1284 		case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1285 			BUG_ON(tm->slot < n);
1286 			/* Fallthrough */
1287 		case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1288 		case MOD_LOG_KEY_REMOVE:
1289 			btrfs_set_node_key(eb, &tm->key, tm->slot);
1290 			btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1291 			btrfs_set_node_ptr_generation(eb, tm->slot,
1292 						      tm->generation);
1293 			n++;
1294 			break;
1295 		case MOD_LOG_KEY_REPLACE:
1296 			BUG_ON(tm->slot >= n);
1297 			btrfs_set_node_key(eb, &tm->key, tm->slot);
1298 			btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1299 			btrfs_set_node_ptr_generation(eb, tm->slot,
1300 						      tm->generation);
1301 			break;
1302 		case MOD_LOG_KEY_ADD:
1303 			/* if a move operation is needed it's in the log */
1304 			n--;
1305 			break;
1306 		case MOD_LOG_MOVE_KEYS:
1307 			o_dst = btrfs_node_key_ptr_offset(tm->slot);
1308 			o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1309 			memmove_extent_buffer(eb, o_dst, o_src,
1310 					      tm->move.nr_items * p_size);
1311 			break;
1312 		case MOD_LOG_ROOT_REPLACE:
1313 			/*
1314 			 * this operation is special. for roots, this must be
1315 			 * handled explicitly before rewinding.
1316 			 * for non-roots, this operation may exist if the node
1317 			 * was a root: root A -> child B; then A gets empty and
1318 			 * B is promoted to the new root. in the mod log, we'll
1319 			 * have a root-replace operation for B, a tree block
1320 			 * that is no root. we simply ignore that operation.
1321 			 */
1322 			break;
1323 		}
1324 		next = rb_next(&tm->node);
1325 		if (!next)
1326 			break;
1327 		tm = rb_entry(next, struct tree_mod_elem, node);
1328 		if (tm->logical != first_tm->logical)
1329 			break;
1330 	}
1331 	tree_mod_log_read_unlock(fs_info);
1332 	btrfs_set_header_nritems(eb, n);
1333 }
1334 
1335 /*
1336  * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1337  * is returned. If rewind operations happen, a fresh buffer is returned. The
1338  * returned buffer is always read-locked. If the returned buffer is not the
1339  * input buffer, the lock on the input buffer is released and the input buffer
1340  * is freed (its refcount is decremented).
1341  */
1342 static struct extent_buffer *
1343 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1344 		    struct extent_buffer *eb, u64 time_seq)
1345 {
1346 	struct extent_buffer *eb_rewin;
1347 	struct tree_mod_elem *tm;
1348 
1349 	if (!time_seq)
1350 		return eb;
1351 
1352 	if (btrfs_header_level(eb) == 0)
1353 		return eb;
1354 
1355 	tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1356 	if (!tm)
1357 		return eb;
1358 
1359 	btrfs_set_path_blocking(path);
1360 	btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1361 
1362 	if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1363 		BUG_ON(tm->slot != 0);
1364 		eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1365 		if (!eb_rewin) {
1366 			btrfs_tree_read_unlock_blocking(eb);
1367 			free_extent_buffer(eb);
1368 			return NULL;
1369 		}
1370 		btrfs_set_header_bytenr(eb_rewin, eb->start);
1371 		btrfs_set_header_backref_rev(eb_rewin,
1372 					     btrfs_header_backref_rev(eb));
1373 		btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1374 		btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1375 	} else {
1376 		eb_rewin = btrfs_clone_extent_buffer(eb);
1377 		if (!eb_rewin) {
1378 			btrfs_tree_read_unlock_blocking(eb);
1379 			free_extent_buffer(eb);
1380 			return NULL;
1381 		}
1382 	}
1383 
1384 	btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
1385 	btrfs_tree_read_unlock_blocking(eb);
1386 	free_extent_buffer(eb);
1387 
1388 	extent_buffer_get(eb_rewin);
1389 	btrfs_tree_read_lock(eb_rewin);
1390 	__tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1391 	WARN_ON(btrfs_header_nritems(eb_rewin) >
1392 		BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1393 
1394 	return eb_rewin;
1395 }
1396 
1397 /*
1398  * get_old_root() rewinds the state of @root's root node to the given @time_seq
1399  * value. If there are no changes, the current root->root_node is returned. If
1400  * anything changed in between, there's a fresh buffer allocated on which the
1401  * rewind operations are done. In any case, the returned buffer is read locked.
1402  * Returns NULL on error (with no locks held).
1403  */
1404 static inline struct extent_buffer *
1405 get_old_root(struct btrfs_root *root, u64 time_seq)
1406 {
1407 	struct btrfs_fs_info *fs_info = root->fs_info;
1408 	struct tree_mod_elem *tm;
1409 	struct extent_buffer *eb = NULL;
1410 	struct extent_buffer *eb_root;
1411 	struct extent_buffer *old;
1412 	struct tree_mod_root *old_root = NULL;
1413 	u64 old_generation = 0;
1414 	u64 logical;
1415 
1416 	eb_root = btrfs_read_lock_root_node(root);
1417 	tm = __tree_mod_log_oldest_root(fs_info, eb_root, time_seq);
1418 	if (!tm)
1419 		return eb_root;
1420 
1421 	if (tm->op == MOD_LOG_ROOT_REPLACE) {
1422 		old_root = &tm->old_root;
1423 		old_generation = tm->generation;
1424 		logical = old_root->logical;
1425 	} else {
1426 		logical = eb_root->start;
1427 	}
1428 
1429 	tm = tree_mod_log_search(fs_info, logical, time_seq);
1430 	if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1431 		btrfs_tree_read_unlock(eb_root);
1432 		free_extent_buffer(eb_root);
1433 		old = read_tree_block(fs_info, logical, 0);
1434 		if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1435 			if (!IS_ERR(old))
1436 				free_extent_buffer(old);
1437 			btrfs_warn(fs_info,
1438 				   "failed to read tree block %llu from get_old_root",
1439 				   logical);
1440 		} else {
1441 			eb = btrfs_clone_extent_buffer(old);
1442 			free_extent_buffer(old);
1443 		}
1444 	} else if (old_root) {
1445 		btrfs_tree_read_unlock(eb_root);
1446 		free_extent_buffer(eb_root);
1447 		eb = alloc_dummy_extent_buffer(fs_info, logical);
1448 	} else {
1449 		btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1450 		eb = btrfs_clone_extent_buffer(eb_root);
1451 		btrfs_tree_read_unlock_blocking(eb_root);
1452 		free_extent_buffer(eb_root);
1453 	}
1454 
1455 	if (!eb)
1456 		return NULL;
1457 	extent_buffer_get(eb);
1458 	btrfs_tree_read_lock(eb);
1459 	if (old_root) {
1460 		btrfs_set_header_bytenr(eb, eb->start);
1461 		btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1462 		btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1463 		btrfs_set_header_level(eb, old_root->level);
1464 		btrfs_set_header_generation(eb, old_generation);
1465 	}
1466 	if (tm)
1467 		__tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1468 	else
1469 		WARN_ON(btrfs_header_level(eb) != 0);
1470 	WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1471 
1472 	return eb;
1473 }
1474 
1475 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1476 {
1477 	struct tree_mod_elem *tm;
1478 	int level;
1479 	struct extent_buffer *eb_root = btrfs_root_node(root);
1480 
1481 	tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1482 	if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1483 		level = tm->old_root.level;
1484 	} else {
1485 		level = btrfs_header_level(eb_root);
1486 	}
1487 	free_extent_buffer(eb_root);
1488 
1489 	return level;
1490 }
1491 
1492 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1493 				   struct btrfs_root *root,
1494 				   struct extent_buffer *buf)
1495 {
1496 	if (btrfs_is_testing(root->fs_info))
1497 		return 0;
1498 
1499 	/* ensure we can see the force_cow */
1500 	smp_rmb();
1501 
1502 	/*
1503 	 * We do not need to cow a block if
1504 	 * 1) this block is not created or changed in this transaction;
1505 	 * 2) this block does not belong to TREE_RELOC tree;
1506 	 * 3) the root is not forced COW.
1507 	 *
1508 	 * What is forced COW:
1509 	 *    when we create snapshot during committing the transaction,
1510 	 *    after we've finished coping src root, we must COW the shared
1511 	 *    block to ensure the metadata consistency.
1512 	 */
1513 	if (btrfs_header_generation(buf) == trans->transid &&
1514 	    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1515 	    !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1516 	      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1517 	    !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1518 		return 0;
1519 	return 1;
1520 }
1521 
1522 /*
1523  * cows a single block, see __btrfs_cow_block for the real work.
1524  * This version of it has extra checks so that a block isn't COWed more than
1525  * once per transaction, as long as it hasn't been written yet
1526  */
1527 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1528 		    struct btrfs_root *root, struct extent_buffer *buf,
1529 		    struct extent_buffer *parent, int parent_slot,
1530 		    struct extent_buffer **cow_ret)
1531 {
1532 	struct btrfs_fs_info *fs_info = root->fs_info;
1533 	u64 search_start;
1534 	int ret;
1535 
1536 	if (trans->transaction != fs_info->running_transaction)
1537 		WARN(1, KERN_CRIT "trans %llu running %llu\n",
1538 		       trans->transid,
1539 		       fs_info->running_transaction->transid);
1540 
1541 	if (trans->transid != fs_info->generation)
1542 		WARN(1, KERN_CRIT "trans %llu running %llu\n",
1543 		       trans->transid, fs_info->generation);
1544 
1545 	if (!should_cow_block(trans, root, buf)) {
1546 		trans->dirty = true;
1547 		*cow_ret = buf;
1548 		return 0;
1549 	}
1550 
1551 	search_start = buf->start & ~((u64)SZ_1G - 1);
1552 
1553 	if (parent)
1554 		btrfs_set_lock_blocking(parent);
1555 	btrfs_set_lock_blocking(buf);
1556 
1557 	ret = __btrfs_cow_block(trans, root, buf, parent,
1558 				 parent_slot, cow_ret, search_start, 0);
1559 
1560 	trace_btrfs_cow_block(root, buf, *cow_ret);
1561 
1562 	return ret;
1563 }
1564 
1565 /*
1566  * helper function for defrag to decide if two blocks pointed to by a
1567  * node are actually close by
1568  */
1569 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1570 {
1571 	if (blocknr < other && other - (blocknr + blocksize) < 32768)
1572 		return 1;
1573 	if (blocknr > other && blocknr - (other + blocksize) < 32768)
1574 		return 1;
1575 	return 0;
1576 }
1577 
1578 /*
1579  * compare two keys in a memcmp fashion
1580  */
1581 static int comp_keys(const struct btrfs_disk_key *disk,
1582 		     const struct btrfs_key *k2)
1583 {
1584 	struct btrfs_key k1;
1585 
1586 	btrfs_disk_key_to_cpu(&k1, disk);
1587 
1588 	return btrfs_comp_cpu_keys(&k1, k2);
1589 }
1590 
1591 /*
1592  * same as comp_keys only with two btrfs_key's
1593  */
1594 int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1595 {
1596 	if (k1->objectid > k2->objectid)
1597 		return 1;
1598 	if (k1->objectid < k2->objectid)
1599 		return -1;
1600 	if (k1->type > k2->type)
1601 		return 1;
1602 	if (k1->type < k2->type)
1603 		return -1;
1604 	if (k1->offset > k2->offset)
1605 		return 1;
1606 	if (k1->offset < k2->offset)
1607 		return -1;
1608 	return 0;
1609 }
1610 
1611 /*
1612  * this is used by the defrag code to go through all the
1613  * leaves pointed to by a node and reallocate them so that
1614  * disk order is close to key order
1615  */
1616 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1617 		       struct btrfs_root *root, struct extent_buffer *parent,
1618 		       int start_slot, u64 *last_ret,
1619 		       struct btrfs_key *progress)
1620 {
1621 	struct btrfs_fs_info *fs_info = root->fs_info;
1622 	struct extent_buffer *cur;
1623 	u64 blocknr;
1624 	u64 gen;
1625 	u64 search_start = *last_ret;
1626 	u64 last_block = 0;
1627 	u64 other;
1628 	u32 parent_nritems;
1629 	int end_slot;
1630 	int i;
1631 	int err = 0;
1632 	int parent_level;
1633 	int uptodate;
1634 	u32 blocksize;
1635 	int progress_passed = 0;
1636 	struct btrfs_disk_key disk_key;
1637 
1638 	parent_level = btrfs_header_level(parent);
1639 
1640 	WARN_ON(trans->transaction != fs_info->running_transaction);
1641 	WARN_ON(trans->transid != fs_info->generation);
1642 
1643 	parent_nritems = btrfs_header_nritems(parent);
1644 	blocksize = fs_info->nodesize;
1645 	end_slot = parent_nritems - 1;
1646 
1647 	if (parent_nritems <= 1)
1648 		return 0;
1649 
1650 	btrfs_set_lock_blocking(parent);
1651 
1652 	for (i = start_slot; i <= end_slot; i++) {
1653 		int close = 1;
1654 
1655 		btrfs_node_key(parent, &disk_key, i);
1656 		if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1657 			continue;
1658 
1659 		progress_passed = 1;
1660 		blocknr = btrfs_node_blockptr(parent, i);
1661 		gen = btrfs_node_ptr_generation(parent, i);
1662 		if (last_block == 0)
1663 			last_block = blocknr;
1664 
1665 		if (i > 0) {
1666 			other = btrfs_node_blockptr(parent, i - 1);
1667 			close = close_blocks(blocknr, other, blocksize);
1668 		}
1669 		if (!close && i < end_slot) {
1670 			other = btrfs_node_blockptr(parent, i + 1);
1671 			close = close_blocks(blocknr, other, blocksize);
1672 		}
1673 		if (close) {
1674 			last_block = blocknr;
1675 			continue;
1676 		}
1677 
1678 		cur = find_extent_buffer(fs_info, blocknr);
1679 		if (cur)
1680 			uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1681 		else
1682 			uptodate = 0;
1683 		if (!cur || !uptodate) {
1684 			if (!cur) {
1685 				cur = read_tree_block(fs_info, blocknr, gen);
1686 				if (IS_ERR(cur)) {
1687 					return PTR_ERR(cur);
1688 				} else if (!extent_buffer_uptodate(cur)) {
1689 					free_extent_buffer(cur);
1690 					return -EIO;
1691 				}
1692 			} else if (!uptodate) {
1693 				err = btrfs_read_buffer(cur, gen);
1694 				if (err) {
1695 					free_extent_buffer(cur);
1696 					return err;
1697 				}
1698 			}
1699 		}
1700 		if (search_start == 0)
1701 			search_start = last_block;
1702 
1703 		btrfs_tree_lock(cur);
1704 		btrfs_set_lock_blocking(cur);
1705 		err = __btrfs_cow_block(trans, root, cur, parent, i,
1706 					&cur, search_start,
1707 					min(16 * blocksize,
1708 					    (end_slot - i) * blocksize));
1709 		if (err) {
1710 			btrfs_tree_unlock(cur);
1711 			free_extent_buffer(cur);
1712 			break;
1713 		}
1714 		search_start = cur->start;
1715 		last_block = cur->start;
1716 		*last_ret = search_start;
1717 		btrfs_tree_unlock(cur);
1718 		free_extent_buffer(cur);
1719 	}
1720 	return err;
1721 }
1722 
1723 /*
1724  * search for key in the extent_buffer.  The items start at offset p,
1725  * and they are item_size apart.  There are 'max' items in p.
1726  *
1727  * the slot in the array is returned via slot, and it points to
1728  * the place where you would insert key if it is not found in
1729  * the array.
1730  *
1731  * slot may point to max if the key is bigger than all of the keys
1732  */
1733 static noinline int generic_bin_search(struct extent_buffer *eb,
1734 				       unsigned long p, int item_size,
1735 				       const struct btrfs_key *key,
1736 				       int max, int *slot)
1737 {
1738 	int low = 0;
1739 	int high = max;
1740 	int mid;
1741 	int ret;
1742 	struct btrfs_disk_key *tmp = NULL;
1743 	struct btrfs_disk_key unaligned;
1744 	unsigned long offset;
1745 	char *kaddr = NULL;
1746 	unsigned long map_start = 0;
1747 	unsigned long map_len = 0;
1748 	int err;
1749 
1750 	if (low > high) {
1751 		btrfs_err(eb->fs_info,
1752 		 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1753 			  __func__, low, high, eb->start,
1754 			  btrfs_header_owner(eb), btrfs_header_level(eb));
1755 		return -EINVAL;
1756 	}
1757 
1758 	while (low < high) {
1759 		mid = (low + high) / 2;
1760 		offset = p + mid * item_size;
1761 
1762 		if (!kaddr || offset < map_start ||
1763 		    (offset + sizeof(struct btrfs_disk_key)) >
1764 		    map_start + map_len) {
1765 
1766 			err = map_private_extent_buffer(eb, offset,
1767 						sizeof(struct btrfs_disk_key),
1768 						&kaddr, &map_start, &map_len);
1769 
1770 			if (!err) {
1771 				tmp = (struct btrfs_disk_key *)(kaddr + offset -
1772 							map_start);
1773 			} else if (err == 1) {
1774 				read_extent_buffer(eb, &unaligned,
1775 						   offset, sizeof(unaligned));
1776 				tmp = &unaligned;
1777 			} else {
1778 				return err;
1779 			}
1780 
1781 		} else {
1782 			tmp = (struct btrfs_disk_key *)(kaddr + offset -
1783 							map_start);
1784 		}
1785 		ret = comp_keys(tmp, key);
1786 
1787 		if (ret < 0)
1788 			low = mid + 1;
1789 		else if (ret > 0)
1790 			high = mid;
1791 		else {
1792 			*slot = mid;
1793 			return 0;
1794 		}
1795 	}
1796 	*slot = low;
1797 	return 1;
1798 }
1799 
1800 /*
1801  * simple bin_search frontend that does the right thing for
1802  * leaves vs nodes
1803  */
1804 static int bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1805 		      int level, int *slot)
1806 {
1807 	if (level == 0)
1808 		return generic_bin_search(eb,
1809 					  offsetof(struct btrfs_leaf, items),
1810 					  sizeof(struct btrfs_item),
1811 					  key, btrfs_header_nritems(eb),
1812 					  slot);
1813 	else
1814 		return generic_bin_search(eb,
1815 					  offsetof(struct btrfs_node, ptrs),
1816 					  sizeof(struct btrfs_key_ptr),
1817 					  key, btrfs_header_nritems(eb),
1818 					  slot);
1819 }
1820 
1821 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1822 		     int level, int *slot)
1823 {
1824 	return bin_search(eb, key, level, slot);
1825 }
1826 
1827 static void root_add_used(struct btrfs_root *root, u32 size)
1828 {
1829 	spin_lock(&root->accounting_lock);
1830 	btrfs_set_root_used(&root->root_item,
1831 			    btrfs_root_used(&root->root_item) + size);
1832 	spin_unlock(&root->accounting_lock);
1833 }
1834 
1835 static void root_sub_used(struct btrfs_root *root, u32 size)
1836 {
1837 	spin_lock(&root->accounting_lock);
1838 	btrfs_set_root_used(&root->root_item,
1839 			    btrfs_root_used(&root->root_item) - size);
1840 	spin_unlock(&root->accounting_lock);
1841 }
1842 
1843 /* given a node and slot number, this reads the blocks it points to.  The
1844  * extent buffer is returned with a reference taken (but unlocked).
1845  */
1846 static noinline struct extent_buffer *
1847 read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent,
1848 	       int slot)
1849 {
1850 	int level = btrfs_header_level(parent);
1851 	struct extent_buffer *eb;
1852 
1853 	if (slot < 0 || slot >= btrfs_header_nritems(parent))
1854 		return ERR_PTR(-ENOENT);
1855 
1856 	BUG_ON(level == 0);
1857 
1858 	eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot),
1859 			     btrfs_node_ptr_generation(parent, slot));
1860 	if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1861 		free_extent_buffer(eb);
1862 		eb = ERR_PTR(-EIO);
1863 	}
1864 
1865 	return eb;
1866 }
1867 
1868 /*
1869  * node level balancing, used to make sure nodes are in proper order for
1870  * item deletion.  We balance from the top down, so we have to make sure
1871  * that a deletion won't leave an node completely empty later on.
1872  */
1873 static noinline int balance_level(struct btrfs_trans_handle *trans,
1874 			 struct btrfs_root *root,
1875 			 struct btrfs_path *path, int level)
1876 {
1877 	struct btrfs_fs_info *fs_info = root->fs_info;
1878 	struct extent_buffer *right = NULL;
1879 	struct extent_buffer *mid;
1880 	struct extent_buffer *left = NULL;
1881 	struct extent_buffer *parent = NULL;
1882 	int ret = 0;
1883 	int wret;
1884 	int pslot;
1885 	int orig_slot = path->slots[level];
1886 	u64 orig_ptr;
1887 
1888 	if (level == 0)
1889 		return 0;
1890 
1891 	mid = path->nodes[level];
1892 
1893 	WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1894 		path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1895 	WARN_ON(btrfs_header_generation(mid) != trans->transid);
1896 
1897 	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1898 
1899 	if (level < BTRFS_MAX_LEVEL - 1) {
1900 		parent = path->nodes[level + 1];
1901 		pslot = path->slots[level + 1];
1902 	}
1903 
1904 	/*
1905 	 * deal with the case where there is only one pointer in the root
1906 	 * by promoting the node below to a root
1907 	 */
1908 	if (!parent) {
1909 		struct extent_buffer *child;
1910 
1911 		if (btrfs_header_nritems(mid) != 1)
1912 			return 0;
1913 
1914 		/* promote the child to a root */
1915 		child = read_node_slot(fs_info, mid, 0);
1916 		if (IS_ERR(child)) {
1917 			ret = PTR_ERR(child);
1918 			btrfs_handle_fs_error(fs_info, ret, NULL);
1919 			goto enospc;
1920 		}
1921 
1922 		btrfs_tree_lock(child);
1923 		btrfs_set_lock_blocking(child);
1924 		ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1925 		if (ret) {
1926 			btrfs_tree_unlock(child);
1927 			free_extent_buffer(child);
1928 			goto enospc;
1929 		}
1930 
1931 		tree_mod_log_set_root_pointer(root, child, 1);
1932 		rcu_assign_pointer(root->node, child);
1933 
1934 		add_root_to_dirty_list(root);
1935 		btrfs_tree_unlock(child);
1936 
1937 		path->locks[level] = 0;
1938 		path->nodes[level] = NULL;
1939 		clean_tree_block(fs_info, mid);
1940 		btrfs_tree_unlock(mid);
1941 		/* once for the path */
1942 		free_extent_buffer(mid);
1943 
1944 		root_sub_used(root, mid->len);
1945 		btrfs_free_tree_block(trans, root, mid, 0, 1);
1946 		/* once for the root ptr */
1947 		free_extent_buffer_stale(mid);
1948 		return 0;
1949 	}
1950 	if (btrfs_header_nritems(mid) >
1951 	    BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1952 		return 0;
1953 
1954 	left = read_node_slot(fs_info, parent, pslot - 1);
1955 	if (IS_ERR(left))
1956 		left = NULL;
1957 
1958 	if (left) {
1959 		btrfs_tree_lock(left);
1960 		btrfs_set_lock_blocking(left);
1961 		wret = btrfs_cow_block(trans, root, left,
1962 				       parent, pslot - 1, &left);
1963 		if (wret) {
1964 			ret = wret;
1965 			goto enospc;
1966 		}
1967 	}
1968 
1969 	right = read_node_slot(fs_info, parent, pslot + 1);
1970 	if (IS_ERR(right))
1971 		right = NULL;
1972 
1973 	if (right) {
1974 		btrfs_tree_lock(right);
1975 		btrfs_set_lock_blocking(right);
1976 		wret = btrfs_cow_block(trans, root, right,
1977 				       parent, pslot + 1, &right);
1978 		if (wret) {
1979 			ret = wret;
1980 			goto enospc;
1981 		}
1982 	}
1983 
1984 	/* first, try to make some room in the middle buffer */
1985 	if (left) {
1986 		orig_slot += btrfs_header_nritems(left);
1987 		wret = push_node_left(trans, fs_info, left, mid, 1);
1988 		if (wret < 0)
1989 			ret = wret;
1990 	}
1991 
1992 	/*
1993 	 * then try to empty the right most buffer into the middle
1994 	 */
1995 	if (right) {
1996 		wret = push_node_left(trans, fs_info, mid, right, 1);
1997 		if (wret < 0 && wret != -ENOSPC)
1998 			ret = wret;
1999 		if (btrfs_header_nritems(right) == 0) {
2000 			clean_tree_block(fs_info, right);
2001 			btrfs_tree_unlock(right);
2002 			del_ptr(root, path, level + 1, pslot + 1);
2003 			root_sub_used(root, right->len);
2004 			btrfs_free_tree_block(trans, root, right, 0, 1);
2005 			free_extent_buffer_stale(right);
2006 			right = NULL;
2007 		} else {
2008 			struct btrfs_disk_key right_key;
2009 			btrfs_node_key(right, &right_key, 0);
2010 			tree_mod_log_set_node_key(fs_info, parent,
2011 						  pslot + 1, 0);
2012 			btrfs_set_node_key(parent, &right_key, pslot + 1);
2013 			btrfs_mark_buffer_dirty(parent);
2014 		}
2015 	}
2016 	if (btrfs_header_nritems(mid) == 1) {
2017 		/*
2018 		 * we're not allowed to leave a node with one item in the
2019 		 * tree during a delete.  A deletion from lower in the tree
2020 		 * could try to delete the only pointer in this node.
2021 		 * So, pull some keys from the left.
2022 		 * There has to be a left pointer at this point because
2023 		 * otherwise we would have pulled some pointers from the
2024 		 * right
2025 		 */
2026 		if (!left) {
2027 			ret = -EROFS;
2028 			btrfs_handle_fs_error(fs_info, ret, NULL);
2029 			goto enospc;
2030 		}
2031 		wret = balance_node_right(trans, fs_info, mid, left);
2032 		if (wret < 0) {
2033 			ret = wret;
2034 			goto enospc;
2035 		}
2036 		if (wret == 1) {
2037 			wret = push_node_left(trans, fs_info, left, mid, 1);
2038 			if (wret < 0)
2039 				ret = wret;
2040 		}
2041 		BUG_ON(wret == 1);
2042 	}
2043 	if (btrfs_header_nritems(mid) == 0) {
2044 		clean_tree_block(fs_info, mid);
2045 		btrfs_tree_unlock(mid);
2046 		del_ptr(root, path, level + 1, pslot);
2047 		root_sub_used(root, mid->len);
2048 		btrfs_free_tree_block(trans, root, mid, 0, 1);
2049 		free_extent_buffer_stale(mid);
2050 		mid = NULL;
2051 	} else {
2052 		/* update the parent key to reflect our changes */
2053 		struct btrfs_disk_key mid_key;
2054 		btrfs_node_key(mid, &mid_key, 0);
2055 		tree_mod_log_set_node_key(fs_info, parent, pslot, 0);
2056 		btrfs_set_node_key(parent, &mid_key, pslot);
2057 		btrfs_mark_buffer_dirty(parent);
2058 	}
2059 
2060 	/* update the path */
2061 	if (left) {
2062 		if (btrfs_header_nritems(left) > orig_slot) {
2063 			extent_buffer_get(left);
2064 			/* left was locked after cow */
2065 			path->nodes[level] = left;
2066 			path->slots[level + 1] -= 1;
2067 			path->slots[level] = orig_slot;
2068 			if (mid) {
2069 				btrfs_tree_unlock(mid);
2070 				free_extent_buffer(mid);
2071 			}
2072 		} else {
2073 			orig_slot -= btrfs_header_nritems(left);
2074 			path->slots[level] = orig_slot;
2075 		}
2076 	}
2077 	/* double check we haven't messed things up */
2078 	if (orig_ptr !=
2079 	    btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2080 		BUG();
2081 enospc:
2082 	if (right) {
2083 		btrfs_tree_unlock(right);
2084 		free_extent_buffer(right);
2085 	}
2086 	if (left) {
2087 		if (path->nodes[level] != left)
2088 			btrfs_tree_unlock(left);
2089 		free_extent_buffer(left);
2090 	}
2091 	return ret;
2092 }
2093 
2094 /* Node balancing for insertion.  Here we only split or push nodes around
2095  * when they are completely full.  This is also done top down, so we
2096  * have to be pessimistic.
2097  */
2098 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2099 					  struct btrfs_root *root,
2100 					  struct btrfs_path *path, int level)
2101 {
2102 	struct btrfs_fs_info *fs_info = root->fs_info;
2103 	struct extent_buffer *right = NULL;
2104 	struct extent_buffer *mid;
2105 	struct extent_buffer *left = NULL;
2106 	struct extent_buffer *parent = NULL;
2107 	int ret = 0;
2108 	int wret;
2109 	int pslot;
2110 	int orig_slot = path->slots[level];
2111 
2112 	if (level == 0)
2113 		return 1;
2114 
2115 	mid = path->nodes[level];
2116 	WARN_ON(btrfs_header_generation(mid) != trans->transid);
2117 
2118 	if (level < BTRFS_MAX_LEVEL - 1) {
2119 		parent = path->nodes[level + 1];
2120 		pslot = path->slots[level + 1];
2121 	}
2122 
2123 	if (!parent)
2124 		return 1;
2125 
2126 	left = read_node_slot(fs_info, parent, pslot - 1);
2127 	if (IS_ERR(left))
2128 		left = NULL;
2129 
2130 	/* first, try to make some room in the middle buffer */
2131 	if (left) {
2132 		u32 left_nr;
2133 
2134 		btrfs_tree_lock(left);
2135 		btrfs_set_lock_blocking(left);
2136 
2137 		left_nr = btrfs_header_nritems(left);
2138 		if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2139 			wret = 1;
2140 		} else {
2141 			ret = btrfs_cow_block(trans, root, left, parent,
2142 					      pslot - 1, &left);
2143 			if (ret)
2144 				wret = 1;
2145 			else {
2146 				wret = push_node_left(trans, fs_info,
2147 						      left, mid, 0);
2148 			}
2149 		}
2150 		if (wret < 0)
2151 			ret = wret;
2152 		if (wret == 0) {
2153 			struct btrfs_disk_key disk_key;
2154 			orig_slot += left_nr;
2155 			btrfs_node_key(mid, &disk_key, 0);
2156 			tree_mod_log_set_node_key(fs_info, parent, pslot, 0);
2157 			btrfs_set_node_key(parent, &disk_key, pslot);
2158 			btrfs_mark_buffer_dirty(parent);
2159 			if (btrfs_header_nritems(left) > orig_slot) {
2160 				path->nodes[level] = left;
2161 				path->slots[level + 1] -= 1;
2162 				path->slots[level] = orig_slot;
2163 				btrfs_tree_unlock(mid);
2164 				free_extent_buffer(mid);
2165 			} else {
2166 				orig_slot -=
2167 					btrfs_header_nritems(left);
2168 				path->slots[level] = orig_slot;
2169 				btrfs_tree_unlock(left);
2170 				free_extent_buffer(left);
2171 			}
2172 			return 0;
2173 		}
2174 		btrfs_tree_unlock(left);
2175 		free_extent_buffer(left);
2176 	}
2177 	right = read_node_slot(fs_info, parent, pslot + 1);
2178 	if (IS_ERR(right))
2179 		right = NULL;
2180 
2181 	/*
2182 	 * then try to empty the right most buffer into the middle
2183 	 */
2184 	if (right) {
2185 		u32 right_nr;
2186 
2187 		btrfs_tree_lock(right);
2188 		btrfs_set_lock_blocking(right);
2189 
2190 		right_nr = btrfs_header_nritems(right);
2191 		if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2192 			wret = 1;
2193 		} else {
2194 			ret = btrfs_cow_block(trans, root, right,
2195 					      parent, pslot + 1,
2196 					      &right);
2197 			if (ret)
2198 				wret = 1;
2199 			else {
2200 				wret = balance_node_right(trans, fs_info,
2201 							  right, mid);
2202 			}
2203 		}
2204 		if (wret < 0)
2205 			ret = wret;
2206 		if (wret == 0) {
2207 			struct btrfs_disk_key disk_key;
2208 
2209 			btrfs_node_key(right, &disk_key, 0);
2210 			tree_mod_log_set_node_key(fs_info, parent,
2211 						  pslot + 1, 0);
2212 			btrfs_set_node_key(parent, &disk_key, pslot + 1);
2213 			btrfs_mark_buffer_dirty(parent);
2214 
2215 			if (btrfs_header_nritems(mid) <= orig_slot) {
2216 				path->nodes[level] = right;
2217 				path->slots[level + 1] += 1;
2218 				path->slots[level] = orig_slot -
2219 					btrfs_header_nritems(mid);
2220 				btrfs_tree_unlock(mid);
2221 				free_extent_buffer(mid);
2222 			} else {
2223 				btrfs_tree_unlock(right);
2224 				free_extent_buffer(right);
2225 			}
2226 			return 0;
2227 		}
2228 		btrfs_tree_unlock(right);
2229 		free_extent_buffer(right);
2230 	}
2231 	return 1;
2232 }
2233 
2234 /*
2235  * readahead one full node of leaves, finding things that are close
2236  * to the block in 'slot', and triggering ra on them.
2237  */
2238 static void reada_for_search(struct btrfs_fs_info *fs_info,
2239 			     struct btrfs_path *path,
2240 			     int level, int slot, u64 objectid)
2241 {
2242 	struct extent_buffer *node;
2243 	struct btrfs_disk_key disk_key;
2244 	u32 nritems;
2245 	u64 search;
2246 	u64 target;
2247 	u64 nread = 0;
2248 	struct extent_buffer *eb;
2249 	u32 nr;
2250 	u32 blocksize;
2251 	u32 nscan = 0;
2252 
2253 	if (level != 1)
2254 		return;
2255 
2256 	if (!path->nodes[level])
2257 		return;
2258 
2259 	node = path->nodes[level];
2260 
2261 	search = btrfs_node_blockptr(node, slot);
2262 	blocksize = fs_info->nodesize;
2263 	eb = find_extent_buffer(fs_info, search);
2264 	if (eb) {
2265 		free_extent_buffer(eb);
2266 		return;
2267 	}
2268 
2269 	target = search;
2270 
2271 	nritems = btrfs_header_nritems(node);
2272 	nr = slot;
2273 
2274 	while (1) {
2275 		if (path->reada == READA_BACK) {
2276 			if (nr == 0)
2277 				break;
2278 			nr--;
2279 		} else if (path->reada == READA_FORWARD) {
2280 			nr++;
2281 			if (nr >= nritems)
2282 				break;
2283 		}
2284 		if (path->reada == READA_BACK && objectid) {
2285 			btrfs_node_key(node, &disk_key, nr);
2286 			if (btrfs_disk_key_objectid(&disk_key) != objectid)
2287 				break;
2288 		}
2289 		search = btrfs_node_blockptr(node, nr);
2290 		if ((search <= target && target - search <= 65536) ||
2291 		    (search > target && search - target <= 65536)) {
2292 			readahead_tree_block(fs_info, search);
2293 			nread += blocksize;
2294 		}
2295 		nscan++;
2296 		if ((nread > 65536 || nscan > 32))
2297 			break;
2298 	}
2299 }
2300 
2301 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2302 				       struct btrfs_path *path, int level)
2303 {
2304 	int slot;
2305 	int nritems;
2306 	struct extent_buffer *parent;
2307 	struct extent_buffer *eb;
2308 	u64 gen;
2309 	u64 block1 = 0;
2310 	u64 block2 = 0;
2311 
2312 	parent = path->nodes[level + 1];
2313 	if (!parent)
2314 		return;
2315 
2316 	nritems = btrfs_header_nritems(parent);
2317 	slot = path->slots[level + 1];
2318 
2319 	if (slot > 0) {
2320 		block1 = btrfs_node_blockptr(parent, slot - 1);
2321 		gen = btrfs_node_ptr_generation(parent, slot - 1);
2322 		eb = find_extent_buffer(fs_info, block1);
2323 		/*
2324 		 * if we get -eagain from btrfs_buffer_uptodate, we
2325 		 * don't want to return eagain here.  That will loop
2326 		 * forever
2327 		 */
2328 		if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2329 			block1 = 0;
2330 		free_extent_buffer(eb);
2331 	}
2332 	if (slot + 1 < nritems) {
2333 		block2 = btrfs_node_blockptr(parent, slot + 1);
2334 		gen = btrfs_node_ptr_generation(parent, slot + 1);
2335 		eb = find_extent_buffer(fs_info, block2);
2336 		if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2337 			block2 = 0;
2338 		free_extent_buffer(eb);
2339 	}
2340 
2341 	if (block1)
2342 		readahead_tree_block(fs_info, block1);
2343 	if (block2)
2344 		readahead_tree_block(fs_info, block2);
2345 }
2346 
2347 
2348 /*
2349  * when we walk down the tree, it is usually safe to unlock the higher layers
2350  * in the tree.  The exceptions are when our path goes through slot 0, because
2351  * operations on the tree might require changing key pointers higher up in the
2352  * tree.
2353  *
2354  * callers might also have set path->keep_locks, which tells this code to keep
2355  * the lock if the path points to the last slot in the block.  This is part of
2356  * walking through the tree, and selecting the next slot in the higher block.
2357  *
2358  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
2359  * if lowest_unlock is 1, level 0 won't be unlocked
2360  */
2361 static noinline void unlock_up(struct btrfs_path *path, int level,
2362 			       int lowest_unlock, int min_write_lock_level,
2363 			       int *write_lock_level)
2364 {
2365 	int i;
2366 	int skip_level = level;
2367 	int no_skips = 0;
2368 	struct extent_buffer *t;
2369 
2370 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2371 		if (!path->nodes[i])
2372 			break;
2373 		if (!path->locks[i])
2374 			break;
2375 		if (!no_skips && path->slots[i] == 0) {
2376 			skip_level = i + 1;
2377 			continue;
2378 		}
2379 		if (!no_skips && path->keep_locks) {
2380 			u32 nritems;
2381 			t = path->nodes[i];
2382 			nritems = btrfs_header_nritems(t);
2383 			if (nritems < 1 || path->slots[i] >= nritems - 1) {
2384 				skip_level = i + 1;
2385 				continue;
2386 			}
2387 		}
2388 		if (skip_level < i && i >= lowest_unlock)
2389 			no_skips = 1;
2390 
2391 		t = path->nodes[i];
2392 		if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2393 			btrfs_tree_unlock_rw(t, path->locks[i]);
2394 			path->locks[i] = 0;
2395 			if (write_lock_level &&
2396 			    i > min_write_lock_level &&
2397 			    i <= *write_lock_level) {
2398 				*write_lock_level = i - 1;
2399 			}
2400 		}
2401 	}
2402 }
2403 
2404 /*
2405  * This releases any locks held in the path starting at level and
2406  * going all the way up to the root.
2407  *
2408  * btrfs_search_slot will keep the lock held on higher nodes in a few
2409  * corner cases, such as COW of the block at slot zero in the node.  This
2410  * ignores those rules, and it should only be called when there are no
2411  * more updates to be done higher up in the tree.
2412  */
2413 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2414 {
2415 	int i;
2416 
2417 	if (path->keep_locks)
2418 		return;
2419 
2420 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2421 		if (!path->nodes[i])
2422 			continue;
2423 		if (!path->locks[i])
2424 			continue;
2425 		btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2426 		path->locks[i] = 0;
2427 	}
2428 }
2429 
2430 /*
2431  * helper function for btrfs_search_slot.  The goal is to find a block
2432  * in cache without setting the path to blocking.  If we find the block
2433  * we return zero and the path is unchanged.
2434  *
2435  * If we can't find the block, we set the path blocking and do some
2436  * reada.  -EAGAIN is returned and the search must be repeated.
2437  */
2438 static int
2439 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2440 		      struct extent_buffer **eb_ret, int level, int slot,
2441 		      const struct btrfs_key *key)
2442 {
2443 	struct btrfs_fs_info *fs_info = root->fs_info;
2444 	u64 blocknr;
2445 	u64 gen;
2446 	struct extent_buffer *b = *eb_ret;
2447 	struct extent_buffer *tmp;
2448 	int ret;
2449 
2450 	blocknr = btrfs_node_blockptr(b, slot);
2451 	gen = btrfs_node_ptr_generation(b, slot);
2452 
2453 	tmp = find_extent_buffer(fs_info, blocknr);
2454 	if (tmp) {
2455 		/* first we do an atomic uptodate check */
2456 		if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2457 			*eb_ret = tmp;
2458 			return 0;
2459 		}
2460 
2461 		/* the pages were up to date, but we failed
2462 		 * the generation number check.  Do a full
2463 		 * read for the generation number that is correct.
2464 		 * We must do this without dropping locks so
2465 		 * we can trust our generation number
2466 		 */
2467 		btrfs_set_path_blocking(p);
2468 
2469 		/* now we're allowed to do a blocking uptodate check */
2470 		ret = btrfs_read_buffer(tmp, gen);
2471 		if (!ret) {
2472 			*eb_ret = tmp;
2473 			return 0;
2474 		}
2475 		free_extent_buffer(tmp);
2476 		btrfs_release_path(p);
2477 		return -EIO;
2478 	}
2479 
2480 	/*
2481 	 * reduce lock contention at high levels
2482 	 * of the btree by dropping locks before
2483 	 * we read.  Don't release the lock on the current
2484 	 * level because we need to walk this node to figure
2485 	 * out which blocks to read.
2486 	 */
2487 	btrfs_unlock_up_safe(p, level + 1);
2488 	btrfs_set_path_blocking(p);
2489 
2490 	free_extent_buffer(tmp);
2491 	if (p->reada != READA_NONE)
2492 		reada_for_search(fs_info, p, level, slot, key->objectid);
2493 
2494 	btrfs_release_path(p);
2495 
2496 	ret = -EAGAIN;
2497 	tmp = read_tree_block(fs_info, blocknr, 0);
2498 	if (!IS_ERR(tmp)) {
2499 		/*
2500 		 * If the read above didn't mark this buffer up to date,
2501 		 * it will never end up being up to date.  Set ret to EIO now
2502 		 * and give up so that our caller doesn't loop forever
2503 		 * on our EAGAINs.
2504 		 */
2505 		if (!btrfs_buffer_uptodate(tmp, 0, 0))
2506 			ret = -EIO;
2507 		free_extent_buffer(tmp);
2508 	} else {
2509 		ret = PTR_ERR(tmp);
2510 	}
2511 	return ret;
2512 }
2513 
2514 /*
2515  * helper function for btrfs_search_slot.  This does all of the checks
2516  * for node-level blocks and does any balancing required based on
2517  * the ins_len.
2518  *
2519  * If no extra work was required, zero is returned.  If we had to
2520  * drop the path, -EAGAIN is returned and btrfs_search_slot must
2521  * start over
2522  */
2523 static int
2524 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2525 		       struct btrfs_root *root, struct btrfs_path *p,
2526 		       struct extent_buffer *b, int level, int ins_len,
2527 		       int *write_lock_level)
2528 {
2529 	struct btrfs_fs_info *fs_info = root->fs_info;
2530 	int ret;
2531 
2532 	if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2533 	    BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2534 		int sret;
2535 
2536 		if (*write_lock_level < level + 1) {
2537 			*write_lock_level = level + 1;
2538 			btrfs_release_path(p);
2539 			goto again;
2540 		}
2541 
2542 		btrfs_set_path_blocking(p);
2543 		reada_for_balance(fs_info, p, level);
2544 		sret = split_node(trans, root, p, level);
2545 		btrfs_clear_path_blocking(p, NULL, 0);
2546 
2547 		BUG_ON(sret > 0);
2548 		if (sret) {
2549 			ret = sret;
2550 			goto done;
2551 		}
2552 		b = p->nodes[level];
2553 	} else if (ins_len < 0 && btrfs_header_nritems(b) <
2554 		   BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2555 		int sret;
2556 
2557 		if (*write_lock_level < level + 1) {
2558 			*write_lock_level = level + 1;
2559 			btrfs_release_path(p);
2560 			goto again;
2561 		}
2562 
2563 		btrfs_set_path_blocking(p);
2564 		reada_for_balance(fs_info, p, level);
2565 		sret = balance_level(trans, root, p, level);
2566 		btrfs_clear_path_blocking(p, NULL, 0);
2567 
2568 		if (sret) {
2569 			ret = sret;
2570 			goto done;
2571 		}
2572 		b = p->nodes[level];
2573 		if (!b) {
2574 			btrfs_release_path(p);
2575 			goto again;
2576 		}
2577 		BUG_ON(btrfs_header_nritems(b) == 1);
2578 	}
2579 	return 0;
2580 
2581 again:
2582 	ret = -EAGAIN;
2583 done:
2584 	return ret;
2585 }
2586 
2587 static void key_search_validate(struct extent_buffer *b,
2588 				const struct btrfs_key *key,
2589 				int level)
2590 {
2591 #ifdef CONFIG_BTRFS_ASSERT
2592 	struct btrfs_disk_key disk_key;
2593 
2594 	btrfs_cpu_key_to_disk(&disk_key, key);
2595 
2596 	if (level == 0)
2597 		ASSERT(!memcmp_extent_buffer(b, &disk_key,
2598 		    offsetof(struct btrfs_leaf, items[0].key),
2599 		    sizeof(disk_key)));
2600 	else
2601 		ASSERT(!memcmp_extent_buffer(b, &disk_key,
2602 		    offsetof(struct btrfs_node, ptrs[0].key),
2603 		    sizeof(disk_key)));
2604 #endif
2605 }
2606 
2607 static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
2608 		      int level, int *prev_cmp, int *slot)
2609 {
2610 	if (*prev_cmp != 0) {
2611 		*prev_cmp = bin_search(b, key, level, slot);
2612 		return *prev_cmp;
2613 	}
2614 
2615 	key_search_validate(b, key, level);
2616 	*slot = 0;
2617 
2618 	return 0;
2619 }
2620 
2621 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2622 		u64 iobjectid, u64 ioff, u8 key_type,
2623 		struct btrfs_key *found_key)
2624 {
2625 	int ret;
2626 	struct btrfs_key key;
2627 	struct extent_buffer *eb;
2628 
2629 	ASSERT(path);
2630 	ASSERT(found_key);
2631 
2632 	key.type = key_type;
2633 	key.objectid = iobjectid;
2634 	key.offset = ioff;
2635 
2636 	ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2637 	if (ret < 0)
2638 		return ret;
2639 
2640 	eb = path->nodes[0];
2641 	if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2642 		ret = btrfs_next_leaf(fs_root, path);
2643 		if (ret)
2644 			return ret;
2645 		eb = path->nodes[0];
2646 	}
2647 
2648 	btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2649 	if (found_key->type != key.type ||
2650 			found_key->objectid != key.objectid)
2651 		return 1;
2652 
2653 	return 0;
2654 }
2655 
2656 /*
2657  * look for key in the tree.  path is filled in with nodes along the way
2658  * if key is found, we return zero and you can find the item in the leaf
2659  * level of the path (level 0)
2660  *
2661  * If the key isn't found, the path points to the slot where it should
2662  * be inserted, and 1 is returned.  If there are other errors during the
2663  * search a negative error number is returned.
2664  *
2665  * if ins_len > 0, nodes and leaves will be split as we walk down the
2666  * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
2667  * possible)
2668  */
2669 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2670 		      const struct btrfs_key *key, struct btrfs_path *p,
2671 		      int ins_len, int cow)
2672 {
2673 	struct btrfs_fs_info *fs_info = root->fs_info;
2674 	struct extent_buffer *b;
2675 	int slot;
2676 	int ret;
2677 	int err;
2678 	int level;
2679 	int lowest_unlock = 1;
2680 	int root_lock;
2681 	/* everything at write_lock_level or lower must be write locked */
2682 	int write_lock_level = 0;
2683 	u8 lowest_level = 0;
2684 	int min_write_lock_level;
2685 	int prev_cmp;
2686 
2687 	lowest_level = p->lowest_level;
2688 	WARN_ON(lowest_level && ins_len > 0);
2689 	WARN_ON(p->nodes[0] != NULL);
2690 	BUG_ON(!cow && ins_len);
2691 
2692 	if (ins_len < 0) {
2693 		lowest_unlock = 2;
2694 
2695 		/* when we are removing items, we might have to go up to level
2696 		 * two as we update tree pointers  Make sure we keep write
2697 		 * for those levels as well
2698 		 */
2699 		write_lock_level = 2;
2700 	} else if (ins_len > 0) {
2701 		/*
2702 		 * for inserting items, make sure we have a write lock on
2703 		 * level 1 so we can update keys
2704 		 */
2705 		write_lock_level = 1;
2706 	}
2707 
2708 	if (!cow)
2709 		write_lock_level = -1;
2710 
2711 	if (cow && (p->keep_locks || p->lowest_level))
2712 		write_lock_level = BTRFS_MAX_LEVEL;
2713 
2714 	min_write_lock_level = write_lock_level;
2715 
2716 again:
2717 	prev_cmp = -1;
2718 	/*
2719 	 * we try very hard to do read locks on the root
2720 	 */
2721 	root_lock = BTRFS_READ_LOCK;
2722 	level = 0;
2723 	if (p->search_commit_root) {
2724 		/*
2725 		 * the commit roots are read only
2726 		 * so we always do read locks
2727 		 */
2728 		if (p->need_commit_sem)
2729 			down_read(&fs_info->commit_root_sem);
2730 		b = root->commit_root;
2731 		extent_buffer_get(b);
2732 		level = btrfs_header_level(b);
2733 		if (p->need_commit_sem)
2734 			up_read(&fs_info->commit_root_sem);
2735 		if (!p->skip_locking)
2736 			btrfs_tree_read_lock(b);
2737 	} else {
2738 		if (p->skip_locking) {
2739 			b = btrfs_root_node(root);
2740 			level = btrfs_header_level(b);
2741 		} else {
2742 			/* we don't know the level of the root node
2743 			 * until we actually have it read locked
2744 			 */
2745 			b = btrfs_read_lock_root_node(root);
2746 			level = btrfs_header_level(b);
2747 			if (level <= write_lock_level) {
2748 				/* whoops, must trade for write lock */
2749 				btrfs_tree_read_unlock(b);
2750 				free_extent_buffer(b);
2751 				b = btrfs_lock_root_node(root);
2752 				root_lock = BTRFS_WRITE_LOCK;
2753 
2754 				/* the level might have changed, check again */
2755 				level = btrfs_header_level(b);
2756 			}
2757 		}
2758 	}
2759 	p->nodes[level] = b;
2760 	if (!p->skip_locking)
2761 		p->locks[level] = root_lock;
2762 
2763 	while (b) {
2764 		level = btrfs_header_level(b);
2765 
2766 		/*
2767 		 * setup the path here so we can release it under lock
2768 		 * contention with the cow code
2769 		 */
2770 		if (cow) {
2771 			/*
2772 			 * if we don't really need to cow this block
2773 			 * then we don't want to set the path blocking,
2774 			 * so we test it here
2775 			 */
2776 			if (!should_cow_block(trans, root, b)) {
2777 				trans->dirty = true;
2778 				goto cow_done;
2779 			}
2780 
2781 			/*
2782 			 * must have write locks on this node and the
2783 			 * parent
2784 			 */
2785 			if (level > write_lock_level ||
2786 			    (level + 1 > write_lock_level &&
2787 			    level + 1 < BTRFS_MAX_LEVEL &&
2788 			    p->nodes[level + 1])) {
2789 				write_lock_level = level + 1;
2790 				btrfs_release_path(p);
2791 				goto again;
2792 			}
2793 
2794 			btrfs_set_path_blocking(p);
2795 			err = btrfs_cow_block(trans, root, b,
2796 					      p->nodes[level + 1],
2797 					      p->slots[level + 1], &b);
2798 			if (err) {
2799 				ret = err;
2800 				goto done;
2801 			}
2802 		}
2803 cow_done:
2804 		p->nodes[level] = b;
2805 		btrfs_clear_path_blocking(p, NULL, 0);
2806 
2807 		/*
2808 		 * we have a lock on b and as long as we aren't changing
2809 		 * the tree, there is no way to for the items in b to change.
2810 		 * It is safe to drop the lock on our parent before we
2811 		 * go through the expensive btree search on b.
2812 		 *
2813 		 * If we're inserting or deleting (ins_len != 0), then we might
2814 		 * be changing slot zero, which may require changing the parent.
2815 		 * So, we can't drop the lock until after we know which slot
2816 		 * we're operating on.
2817 		 */
2818 		if (!ins_len && !p->keep_locks) {
2819 			int u = level + 1;
2820 
2821 			if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2822 				btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2823 				p->locks[u] = 0;
2824 			}
2825 		}
2826 
2827 		ret = key_search(b, key, level, &prev_cmp, &slot);
2828 		if (ret < 0)
2829 			goto done;
2830 
2831 		if (level != 0) {
2832 			int dec = 0;
2833 			if (ret && slot > 0) {
2834 				dec = 1;
2835 				slot -= 1;
2836 			}
2837 			p->slots[level] = slot;
2838 			err = setup_nodes_for_search(trans, root, p, b, level,
2839 					     ins_len, &write_lock_level);
2840 			if (err == -EAGAIN)
2841 				goto again;
2842 			if (err) {
2843 				ret = err;
2844 				goto done;
2845 			}
2846 			b = p->nodes[level];
2847 			slot = p->slots[level];
2848 
2849 			/*
2850 			 * slot 0 is special, if we change the key
2851 			 * we have to update the parent pointer
2852 			 * which means we must have a write lock
2853 			 * on the parent
2854 			 */
2855 			if (slot == 0 && ins_len &&
2856 			    write_lock_level < level + 1) {
2857 				write_lock_level = level + 1;
2858 				btrfs_release_path(p);
2859 				goto again;
2860 			}
2861 
2862 			unlock_up(p, level, lowest_unlock,
2863 				  min_write_lock_level, &write_lock_level);
2864 
2865 			if (level == lowest_level) {
2866 				if (dec)
2867 					p->slots[level]++;
2868 				goto done;
2869 			}
2870 
2871 			err = read_block_for_search(root, p, &b, level,
2872 						    slot, key);
2873 			if (err == -EAGAIN)
2874 				goto again;
2875 			if (err) {
2876 				ret = err;
2877 				goto done;
2878 			}
2879 
2880 			if (!p->skip_locking) {
2881 				level = btrfs_header_level(b);
2882 				if (level <= write_lock_level) {
2883 					err = btrfs_try_tree_write_lock(b);
2884 					if (!err) {
2885 						btrfs_set_path_blocking(p);
2886 						btrfs_tree_lock(b);
2887 						btrfs_clear_path_blocking(p, b,
2888 								  BTRFS_WRITE_LOCK);
2889 					}
2890 					p->locks[level] = BTRFS_WRITE_LOCK;
2891 				} else {
2892 					err = btrfs_tree_read_lock_atomic(b);
2893 					if (!err) {
2894 						btrfs_set_path_blocking(p);
2895 						btrfs_tree_read_lock(b);
2896 						btrfs_clear_path_blocking(p, b,
2897 								  BTRFS_READ_LOCK);
2898 					}
2899 					p->locks[level] = BTRFS_READ_LOCK;
2900 				}
2901 				p->nodes[level] = b;
2902 			}
2903 		} else {
2904 			p->slots[level] = slot;
2905 			if (ins_len > 0 &&
2906 			    btrfs_leaf_free_space(fs_info, b) < ins_len) {
2907 				if (write_lock_level < 1) {
2908 					write_lock_level = 1;
2909 					btrfs_release_path(p);
2910 					goto again;
2911 				}
2912 
2913 				btrfs_set_path_blocking(p);
2914 				err = split_leaf(trans, root, key,
2915 						 p, ins_len, ret == 0);
2916 				btrfs_clear_path_blocking(p, NULL, 0);
2917 
2918 				BUG_ON(err > 0);
2919 				if (err) {
2920 					ret = err;
2921 					goto done;
2922 				}
2923 			}
2924 			if (!p->search_for_split)
2925 				unlock_up(p, level, lowest_unlock,
2926 					  min_write_lock_level, &write_lock_level);
2927 			goto done;
2928 		}
2929 	}
2930 	ret = 1;
2931 done:
2932 	/*
2933 	 * we don't really know what they plan on doing with the path
2934 	 * from here on, so for now just mark it as blocking
2935 	 */
2936 	if (!p->leave_spinning)
2937 		btrfs_set_path_blocking(p);
2938 	if (ret < 0 && !p->skip_release_on_error)
2939 		btrfs_release_path(p);
2940 	return ret;
2941 }
2942 
2943 /*
2944  * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2945  * current state of the tree together with the operations recorded in the tree
2946  * modification log to search for the key in a previous version of this tree, as
2947  * denoted by the time_seq parameter.
2948  *
2949  * Naturally, there is no support for insert, delete or cow operations.
2950  *
2951  * The resulting path and return value will be set up as if we called
2952  * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2953  */
2954 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2955 			  struct btrfs_path *p, u64 time_seq)
2956 {
2957 	struct btrfs_fs_info *fs_info = root->fs_info;
2958 	struct extent_buffer *b;
2959 	int slot;
2960 	int ret;
2961 	int err;
2962 	int level;
2963 	int lowest_unlock = 1;
2964 	u8 lowest_level = 0;
2965 	int prev_cmp = -1;
2966 
2967 	lowest_level = p->lowest_level;
2968 	WARN_ON(p->nodes[0] != NULL);
2969 
2970 	if (p->search_commit_root) {
2971 		BUG_ON(time_seq);
2972 		return btrfs_search_slot(NULL, root, key, p, 0, 0);
2973 	}
2974 
2975 again:
2976 	b = get_old_root(root, time_seq);
2977 	level = btrfs_header_level(b);
2978 	p->locks[level] = BTRFS_READ_LOCK;
2979 
2980 	while (b) {
2981 		level = btrfs_header_level(b);
2982 		p->nodes[level] = b;
2983 		btrfs_clear_path_blocking(p, NULL, 0);
2984 
2985 		/*
2986 		 * we have a lock on b and as long as we aren't changing
2987 		 * the tree, there is no way to for the items in b to change.
2988 		 * It is safe to drop the lock on our parent before we
2989 		 * go through the expensive btree search on b.
2990 		 */
2991 		btrfs_unlock_up_safe(p, level + 1);
2992 
2993 		/*
2994 		 * Since we can unwind ebs we want to do a real search every
2995 		 * time.
2996 		 */
2997 		prev_cmp = -1;
2998 		ret = key_search(b, key, level, &prev_cmp, &slot);
2999 
3000 		if (level != 0) {
3001 			int dec = 0;
3002 			if (ret && slot > 0) {
3003 				dec = 1;
3004 				slot -= 1;
3005 			}
3006 			p->slots[level] = slot;
3007 			unlock_up(p, level, lowest_unlock, 0, NULL);
3008 
3009 			if (level == lowest_level) {
3010 				if (dec)
3011 					p->slots[level]++;
3012 				goto done;
3013 			}
3014 
3015 			err = read_block_for_search(root, p, &b, level,
3016 						    slot, key);
3017 			if (err == -EAGAIN)
3018 				goto again;
3019 			if (err) {
3020 				ret = err;
3021 				goto done;
3022 			}
3023 
3024 			level = btrfs_header_level(b);
3025 			err = btrfs_tree_read_lock_atomic(b);
3026 			if (!err) {
3027 				btrfs_set_path_blocking(p);
3028 				btrfs_tree_read_lock(b);
3029 				btrfs_clear_path_blocking(p, b,
3030 							  BTRFS_READ_LOCK);
3031 			}
3032 			b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3033 			if (!b) {
3034 				ret = -ENOMEM;
3035 				goto done;
3036 			}
3037 			p->locks[level] = BTRFS_READ_LOCK;
3038 			p->nodes[level] = b;
3039 		} else {
3040 			p->slots[level] = slot;
3041 			unlock_up(p, level, lowest_unlock, 0, NULL);
3042 			goto done;
3043 		}
3044 	}
3045 	ret = 1;
3046 done:
3047 	if (!p->leave_spinning)
3048 		btrfs_set_path_blocking(p);
3049 	if (ret < 0)
3050 		btrfs_release_path(p);
3051 
3052 	return ret;
3053 }
3054 
3055 /*
3056  * helper to use instead of search slot if no exact match is needed but
3057  * instead the next or previous item should be returned.
3058  * When find_higher is true, the next higher item is returned, the next lower
3059  * otherwise.
3060  * When return_any and find_higher are both true, and no higher item is found,
3061  * return the next lower instead.
3062  * When return_any is true and find_higher is false, and no lower item is found,
3063  * return the next higher instead.
3064  * It returns 0 if any item is found, 1 if none is found (tree empty), and
3065  * < 0 on error
3066  */
3067 int btrfs_search_slot_for_read(struct btrfs_root *root,
3068 			       const struct btrfs_key *key,
3069 			       struct btrfs_path *p, int find_higher,
3070 			       int return_any)
3071 {
3072 	int ret;
3073 	struct extent_buffer *leaf;
3074 
3075 again:
3076 	ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3077 	if (ret <= 0)
3078 		return ret;
3079 	/*
3080 	 * a return value of 1 means the path is at the position where the
3081 	 * item should be inserted. Normally this is the next bigger item,
3082 	 * but in case the previous item is the last in a leaf, path points
3083 	 * to the first free slot in the previous leaf, i.e. at an invalid
3084 	 * item.
3085 	 */
3086 	leaf = p->nodes[0];
3087 
3088 	if (find_higher) {
3089 		if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3090 			ret = btrfs_next_leaf(root, p);
3091 			if (ret <= 0)
3092 				return ret;
3093 			if (!return_any)
3094 				return 1;
3095 			/*
3096 			 * no higher item found, return the next
3097 			 * lower instead
3098 			 */
3099 			return_any = 0;
3100 			find_higher = 0;
3101 			btrfs_release_path(p);
3102 			goto again;
3103 		}
3104 	} else {
3105 		if (p->slots[0] == 0) {
3106 			ret = btrfs_prev_leaf(root, p);
3107 			if (ret < 0)
3108 				return ret;
3109 			if (!ret) {
3110 				leaf = p->nodes[0];
3111 				if (p->slots[0] == btrfs_header_nritems(leaf))
3112 					p->slots[0]--;
3113 				return 0;
3114 			}
3115 			if (!return_any)
3116 				return 1;
3117 			/*
3118 			 * no lower item found, return the next
3119 			 * higher instead
3120 			 */
3121 			return_any = 0;
3122 			find_higher = 1;
3123 			btrfs_release_path(p);
3124 			goto again;
3125 		} else {
3126 			--p->slots[0];
3127 		}
3128 	}
3129 	return 0;
3130 }
3131 
3132 /*
3133  * adjust the pointers going up the tree, starting at level
3134  * making sure the right key of each node is points to 'key'.
3135  * This is used after shifting pointers to the left, so it stops
3136  * fixing up pointers when a given leaf/node is not in slot 0 of the
3137  * higher levels
3138  *
3139  */
3140 static void fixup_low_keys(struct btrfs_fs_info *fs_info,
3141 			   struct btrfs_path *path,
3142 			   struct btrfs_disk_key *key, int level)
3143 {
3144 	int i;
3145 	struct extent_buffer *t;
3146 
3147 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3148 		int tslot = path->slots[i];
3149 		if (!path->nodes[i])
3150 			break;
3151 		t = path->nodes[i];
3152 		tree_mod_log_set_node_key(fs_info, t, tslot, 1);
3153 		btrfs_set_node_key(t, key, tslot);
3154 		btrfs_mark_buffer_dirty(path->nodes[i]);
3155 		if (tslot != 0)
3156 			break;
3157 	}
3158 }
3159 
3160 /*
3161  * update item key.
3162  *
3163  * This function isn't completely safe. It's the caller's responsibility
3164  * that the new key won't break the order
3165  */
3166 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3167 			     struct btrfs_path *path,
3168 			     const struct btrfs_key *new_key)
3169 {
3170 	struct btrfs_disk_key disk_key;
3171 	struct extent_buffer *eb;
3172 	int slot;
3173 
3174 	eb = path->nodes[0];
3175 	slot = path->slots[0];
3176 	if (slot > 0) {
3177 		btrfs_item_key(eb, &disk_key, slot - 1);
3178 		BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3179 	}
3180 	if (slot < btrfs_header_nritems(eb) - 1) {
3181 		btrfs_item_key(eb, &disk_key, slot + 1);
3182 		BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3183 	}
3184 
3185 	btrfs_cpu_key_to_disk(&disk_key, new_key);
3186 	btrfs_set_item_key(eb, &disk_key, slot);
3187 	btrfs_mark_buffer_dirty(eb);
3188 	if (slot == 0)
3189 		fixup_low_keys(fs_info, path, &disk_key, 1);
3190 }
3191 
3192 /*
3193  * try to push data from one node into the next node left in the
3194  * tree.
3195  *
3196  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3197  * error, and > 0 if there was no room in the left hand block.
3198  */
3199 static int push_node_left(struct btrfs_trans_handle *trans,
3200 			  struct btrfs_fs_info *fs_info,
3201 			  struct extent_buffer *dst,
3202 			  struct extent_buffer *src, int empty)
3203 {
3204 	int push_items = 0;
3205 	int src_nritems;
3206 	int dst_nritems;
3207 	int ret = 0;
3208 
3209 	src_nritems = btrfs_header_nritems(src);
3210 	dst_nritems = btrfs_header_nritems(dst);
3211 	push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3212 	WARN_ON(btrfs_header_generation(src) != trans->transid);
3213 	WARN_ON(btrfs_header_generation(dst) != trans->transid);
3214 
3215 	if (!empty && src_nritems <= 8)
3216 		return 1;
3217 
3218 	if (push_items <= 0)
3219 		return 1;
3220 
3221 	if (empty) {
3222 		push_items = min(src_nritems, push_items);
3223 		if (push_items < src_nritems) {
3224 			/* leave at least 8 pointers in the node if
3225 			 * we aren't going to empty it
3226 			 */
3227 			if (src_nritems - push_items < 8) {
3228 				if (push_items <= 8)
3229 					return 1;
3230 				push_items -= 8;
3231 			}
3232 		}
3233 	} else
3234 		push_items = min(src_nritems - 8, push_items);
3235 
3236 	ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0,
3237 				   push_items);
3238 	if (ret) {
3239 		btrfs_abort_transaction(trans, ret);
3240 		return ret;
3241 	}
3242 	copy_extent_buffer(dst, src,
3243 			   btrfs_node_key_ptr_offset(dst_nritems),
3244 			   btrfs_node_key_ptr_offset(0),
3245 			   push_items * sizeof(struct btrfs_key_ptr));
3246 
3247 	if (push_items < src_nritems) {
3248 		/*
3249 		 * don't call tree_mod_log_eb_move here, key removal was already
3250 		 * fully logged by tree_mod_log_eb_copy above.
3251 		 */
3252 		memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3253 				      btrfs_node_key_ptr_offset(push_items),
3254 				      (src_nritems - push_items) *
3255 				      sizeof(struct btrfs_key_ptr));
3256 	}
3257 	btrfs_set_header_nritems(src, src_nritems - push_items);
3258 	btrfs_set_header_nritems(dst, dst_nritems + push_items);
3259 	btrfs_mark_buffer_dirty(src);
3260 	btrfs_mark_buffer_dirty(dst);
3261 
3262 	return ret;
3263 }
3264 
3265 /*
3266  * try to push data from one node into the next node right in the
3267  * tree.
3268  *
3269  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3270  * error, and > 0 if there was no room in the right hand block.
3271  *
3272  * this will  only push up to 1/2 the contents of the left node over
3273  */
3274 static int balance_node_right(struct btrfs_trans_handle *trans,
3275 			      struct btrfs_fs_info *fs_info,
3276 			      struct extent_buffer *dst,
3277 			      struct extent_buffer *src)
3278 {
3279 	int push_items = 0;
3280 	int max_push;
3281 	int src_nritems;
3282 	int dst_nritems;
3283 	int ret = 0;
3284 
3285 	WARN_ON(btrfs_header_generation(src) != trans->transid);
3286 	WARN_ON(btrfs_header_generation(dst) != trans->transid);
3287 
3288 	src_nritems = btrfs_header_nritems(src);
3289 	dst_nritems = btrfs_header_nritems(dst);
3290 	push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3291 	if (push_items <= 0)
3292 		return 1;
3293 
3294 	if (src_nritems < 4)
3295 		return 1;
3296 
3297 	max_push = src_nritems / 2 + 1;
3298 	/* don't try to empty the node */
3299 	if (max_push >= src_nritems)
3300 		return 1;
3301 
3302 	if (max_push < push_items)
3303 		push_items = max_push;
3304 
3305 	tree_mod_log_eb_move(fs_info, dst, push_items, 0, dst_nritems);
3306 	memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3307 				      btrfs_node_key_ptr_offset(0),
3308 				      (dst_nritems) *
3309 				      sizeof(struct btrfs_key_ptr));
3310 
3311 	ret = tree_mod_log_eb_copy(fs_info, dst, src, 0,
3312 				   src_nritems - push_items, push_items);
3313 	if (ret) {
3314 		btrfs_abort_transaction(trans, ret);
3315 		return ret;
3316 	}
3317 	copy_extent_buffer(dst, src,
3318 			   btrfs_node_key_ptr_offset(0),
3319 			   btrfs_node_key_ptr_offset(src_nritems - push_items),
3320 			   push_items * sizeof(struct btrfs_key_ptr));
3321 
3322 	btrfs_set_header_nritems(src, src_nritems - push_items);
3323 	btrfs_set_header_nritems(dst, dst_nritems + push_items);
3324 
3325 	btrfs_mark_buffer_dirty(src);
3326 	btrfs_mark_buffer_dirty(dst);
3327 
3328 	return ret;
3329 }
3330 
3331 /*
3332  * helper function to insert a new root level in the tree.
3333  * A new node is allocated, and a single item is inserted to
3334  * point to the existing root
3335  *
3336  * returns zero on success or < 0 on failure.
3337  */
3338 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3339 			   struct btrfs_root *root,
3340 			   struct btrfs_path *path, int level)
3341 {
3342 	struct btrfs_fs_info *fs_info = root->fs_info;
3343 	u64 lower_gen;
3344 	struct extent_buffer *lower;
3345 	struct extent_buffer *c;
3346 	struct extent_buffer *old;
3347 	struct btrfs_disk_key lower_key;
3348 
3349 	BUG_ON(path->nodes[level]);
3350 	BUG_ON(path->nodes[level-1] != root->node);
3351 
3352 	lower = path->nodes[level-1];
3353 	if (level == 1)
3354 		btrfs_item_key(lower, &lower_key, 0);
3355 	else
3356 		btrfs_node_key(lower, &lower_key, 0);
3357 
3358 	c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3359 				   &lower_key, level, root->node->start, 0);
3360 	if (IS_ERR(c))
3361 		return PTR_ERR(c);
3362 
3363 	root_add_used(root, fs_info->nodesize);
3364 
3365 	memzero_extent_buffer(c, 0, sizeof(struct btrfs_header));
3366 	btrfs_set_header_nritems(c, 1);
3367 	btrfs_set_header_level(c, level);
3368 	btrfs_set_header_bytenr(c, c->start);
3369 	btrfs_set_header_generation(c, trans->transid);
3370 	btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3371 	btrfs_set_header_owner(c, root->root_key.objectid);
3372 
3373 	write_extent_buffer_fsid(c, fs_info->fsid);
3374 	write_extent_buffer_chunk_tree_uuid(c, fs_info->chunk_tree_uuid);
3375 
3376 	btrfs_set_node_key(c, &lower_key, 0);
3377 	btrfs_set_node_blockptr(c, 0, lower->start);
3378 	lower_gen = btrfs_header_generation(lower);
3379 	WARN_ON(lower_gen != trans->transid);
3380 
3381 	btrfs_set_node_ptr_generation(c, 0, lower_gen);
3382 
3383 	btrfs_mark_buffer_dirty(c);
3384 
3385 	old = root->node;
3386 	tree_mod_log_set_root_pointer(root, c, 0);
3387 	rcu_assign_pointer(root->node, c);
3388 
3389 	/* the super has an extra ref to root->node */
3390 	free_extent_buffer(old);
3391 
3392 	add_root_to_dirty_list(root);
3393 	extent_buffer_get(c);
3394 	path->nodes[level] = c;
3395 	path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3396 	path->slots[level] = 0;
3397 	return 0;
3398 }
3399 
3400 /*
3401  * worker function to insert a single pointer in a node.
3402  * the node should have enough room for the pointer already
3403  *
3404  * slot and level indicate where you want the key to go, and
3405  * blocknr is the block the key points to.
3406  */
3407 static void insert_ptr(struct btrfs_trans_handle *trans,
3408 		       struct btrfs_fs_info *fs_info, struct btrfs_path *path,
3409 		       struct btrfs_disk_key *key, u64 bytenr,
3410 		       int slot, int level)
3411 {
3412 	struct extent_buffer *lower;
3413 	int nritems;
3414 	int ret;
3415 
3416 	BUG_ON(!path->nodes[level]);
3417 	btrfs_assert_tree_locked(path->nodes[level]);
3418 	lower = path->nodes[level];
3419 	nritems = btrfs_header_nritems(lower);
3420 	BUG_ON(slot > nritems);
3421 	BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info));
3422 	if (slot != nritems) {
3423 		if (level)
3424 			tree_mod_log_eb_move(fs_info, lower, slot + 1,
3425 					     slot, nritems - slot);
3426 		memmove_extent_buffer(lower,
3427 			      btrfs_node_key_ptr_offset(slot + 1),
3428 			      btrfs_node_key_ptr_offset(slot),
3429 			      (nritems - slot) * sizeof(struct btrfs_key_ptr));
3430 	}
3431 	if (level) {
3432 		ret = tree_mod_log_insert_key(fs_info, lower, slot,
3433 					      MOD_LOG_KEY_ADD, GFP_NOFS);
3434 		BUG_ON(ret < 0);
3435 	}
3436 	btrfs_set_node_key(lower, key, slot);
3437 	btrfs_set_node_blockptr(lower, slot, bytenr);
3438 	WARN_ON(trans->transid == 0);
3439 	btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3440 	btrfs_set_header_nritems(lower, nritems + 1);
3441 	btrfs_mark_buffer_dirty(lower);
3442 }
3443 
3444 /*
3445  * split the node at the specified level in path in two.
3446  * The path is corrected to point to the appropriate node after the split
3447  *
3448  * Before splitting this tries to make some room in the node by pushing
3449  * left and right, if either one works, it returns right away.
3450  *
3451  * returns 0 on success and < 0 on failure
3452  */
3453 static noinline int split_node(struct btrfs_trans_handle *trans,
3454 			       struct btrfs_root *root,
3455 			       struct btrfs_path *path, int level)
3456 {
3457 	struct btrfs_fs_info *fs_info = root->fs_info;
3458 	struct extent_buffer *c;
3459 	struct extent_buffer *split;
3460 	struct btrfs_disk_key disk_key;
3461 	int mid;
3462 	int ret;
3463 	u32 c_nritems;
3464 
3465 	c = path->nodes[level];
3466 	WARN_ON(btrfs_header_generation(c) != trans->transid);
3467 	if (c == root->node) {
3468 		/*
3469 		 * trying to split the root, lets make a new one
3470 		 *
3471 		 * tree mod log: We don't log_removal old root in
3472 		 * insert_new_root, because that root buffer will be kept as a
3473 		 * normal node. We are going to log removal of half of the
3474 		 * elements below with tree_mod_log_eb_copy. We're holding a
3475 		 * tree lock on the buffer, which is why we cannot race with
3476 		 * other tree_mod_log users.
3477 		 */
3478 		ret = insert_new_root(trans, root, path, level + 1);
3479 		if (ret)
3480 			return ret;
3481 	} else {
3482 		ret = push_nodes_for_insert(trans, root, path, level);
3483 		c = path->nodes[level];
3484 		if (!ret && btrfs_header_nritems(c) <
3485 		    BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3486 			return 0;
3487 		if (ret < 0)
3488 			return ret;
3489 	}
3490 
3491 	c_nritems = btrfs_header_nritems(c);
3492 	mid = (c_nritems + 1) / 2;
3493 	btrfs_node_key(c, &disk_key, mid);
3494 
3495 	split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3496 			&disk_key, level, c->start, 0);
3497 	if (IS_ERR(split))
3498 		return PTR_ERR(split);
3499 
3500 	root_add_used(root, fs_info->nodesize);
3501 
3502 	memzero_extent_buffer(split, 0, sizeof(struct btrfs_header));
3503 	btrfs_set_header_level(split, btrfs_header_level(c));
3504 	btrfs_set_header_bytenr(split, split->start);
3505 	btrfs_set_header_generation(split, trans->transid);
3506 	btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3507 	btrfs_set_header_owner(split, root->root_key.objectid);
3508 	write_extent_buffer_fsid(split, fs_info->fsid);
3509 	write_extent_buffer_chunk_tree_uuid(split, fs_info->chunk_tree_uuid);
3510 
3511 	ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid);
3512 	if (ret) {
3513 		btrfs_abort_transaction(trans, ret);
3514 		return ret;
3515 	}
3516 	copy_extent_buffer(split, c,
3517 			   btrfs_node_key_ptr_offset(0),
3518 			   btrfs_node_key_ptr_offset(mid),
3519 			   (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3520 	btrfs_set_header_nritems(split, c_nritems - mid);
3521 	btrfs_set_header_nritems(c, mid);
3522 	ret = 0;
3523 
3524 	btrfs_mark_buffer_dirty(c);
3525 	btrfs_mark_buffer_dirty(split);
3526 
3527 	insert_ptr(trans, fs_info, path, &disk_key, split->start,
3528 		   path->slots[level + 1] + 1, level + 1);
3529 
3530 	if (path->slots[level] >= mid) {
3531 		path->slots[level] -= mid;
3532 		btrfs_tree_unlock(c);
3533 		free_extent_buffer(c);
3534 		path->nodes[level] = split;
3535 		path->slots[level + 1] += 1;
3536 	} else {
3537 		btrfs_tree_unlock(split);
3538 		free_extent_buffer(split);
3539 	}
3540 	return ret;
3541 }
3542 
3543 /*
3544  * how many bytes are required to store the items in a leaf.  start
3545  * and nr indicate which items in the leaf to check.  This totals up the
3546  * space used both by the item structs and the item data
3547  */
3548 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3549 {
3550 	struct btrfs_item *start_item;
3551 	struct btrfs_item *end_item;
3552 	struct btrfs_map_token token;
3553 	int data_len;
3554 	int nritems = btrfs_header_nritems(l);
3555 	int end = min(nritems, start + nr) - 1;
3556 
3557 	if (!nr)
3558 		return 0;
3559 	btrfs_init_map_token(&token);
3560 	start_item = btrfs_item_nr(start);
3561 	end_item = btrfs_item_nr(end);
3562 	data_len = btrfs_token_item_offset(l, start_item, &token) +
3563 		btrfs_token_item_size(l, start_item, &token);
3564 	data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3565 	data_len += sizeof(struct btrfs_item) * nr;
3566 	WARN_ON(data_len < 0);
3567 	return data_len;
3568 }
3569 
3570 /*
3571  * The space between the end of the leaf items and
3572  * the start of the leaf data.  IOW, how much room
3573  * the leaf has left for both items and data
3574  */
3575 noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info,
3576 				   struct extent_buffer *leaf)
3577 {
3578 	int nritems = btrfs_header_nritems(leaf);
3579 	int ret;
3580 
3581 	ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3582 	if (ret < 0) {
3583 		btrfs_crit(fs_info,
3584 			   "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3585 			   ret,
3586 			   (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3587 			   leaf_space_used(leaf, 0, nritems), nritems);
3588 	}
3589 	return ret;
3590 }
3591 
3592 /*
3593  * min slot controls the lowest index we're willing to push to the
3594  * right.  We'll push up to and including min_slot, but no lower
3595  */
3596 static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info,
3597 				      struct btrfs_path *path,
3598 				      int data_size, int empty,
3599 				      struct extent_buffer *right,
3600 				      int free_space, u32 left_nritems,
3601 				      u32 min_slot)
3602 {
3603 	struct extent_buffer *left = path->nodes[0];
3604 	struct extent_buffer *upper = path->nodes[1];
3605 	struct btrfs_map_token token;
3606 	struct btrfs_disk_key disk_key;
3607 	int slot;
3608 	u32 i;
3609 	int push_space = 0;
3610 	int push_items = 0;
3611 	struct btrfs_item *item;
3612 	u32 nr;
3613 	u32 right_nritems;
3614 	u32 data_end;
3615 	u32 this_item_size;
3616 
3617 	btrfs_init_map_token(&token);
3618 
3619 	if (empty)
3620 		nr = 0;
3621 	else
3622 		nr = max_t(u32, 1, min_slot);
3623 
3624 	if (path->slots[0] >= left_nritems)
3625 		push_space += data_size;
3626 
3627 	slot = path->slots[1];
3628 	i = left_nritems - 1;
3629 	while (i >= nr) {
3630 		item = btrfs_item_nr(i);
3631 
3632 		if (!empty && push_items > 0) {
3633 			if (path->slots[0] > i)
3634 				break;
3635 			if (path->slots[0] == i) {
3636 				int space = btrfs_leaf_free_space(fs_info, left);
3637 				if (space + push_space * 2 > free_space)
3638 					break;
3639 			}
3640 		}
3641 
3642 		if (path->slots[0] == i)
3643 			push_space += data_size;
3644 
3645 		this_item_size = btrfs_item_size(left, item);
3646 		if (this_item_size + sizeof(*item) + push_space > free_space)
3647 			break;
3648 
3649 		push_items++;
3650 		push_space += this_item_size + sizeof(*item);
3651 		if (i == 0)
3652 			break;
3653 		i--;
3654 	}
3655 
3656 	if (push_items == 0)
3657 		goto out_unlock;
3658 
3659 	WARN_ON(!empty && push_items == left_nritems);
3660 
3661 	/* push left to right */
3662 	right_nritems = btrfs_header_nritems(right);
3663 
3664 	push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3665 	push_space -= leaf_data_end(fs_info, left);
3666 
3667 	/* make room in the right data area */
3668 	data_end = leaf_data_end(fs_info, right);
3669 	memmove_extent_buffer(right,
3670 			      BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3671 			      BTRFS_LEAF_DATA_OFFSET + data_end,
3672 			      BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3673 
3674 	/* copy from the left data area */
3675 	copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3676 		     BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3677 		     BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left),
3678 		     push_space);
3679 
3680 	memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3681 			      btrfs_item_nr_offset(0),
3682 			      right_nritems * sizeof(struct btrfs_item));
3683 
3684 	/* copy the items from left to right */
3685 	copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3686 		   btrfs_item_nr_offset(left_nritems - push_items),
3687 		   push_items * sizeof(struct btrfs_item));
3688 
3689 	/* update the item pointers */
3690 	right_nritems += push_items;
3691 	btrfs_set_header_nritems(right, right_nritems);
3692 	push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3693 	for (i = 0; i < right_nritems; i++) {
3694 		item = btrfs_item_nr(i);
3695 		push_space -= btrfs_token_item_size(right, item, &token);
3696 		btrfs_set_token_item_offset(right, item, push_space, &token);
3697 	}
3698 
3699 	left_nritems -= push_items;
3700 	btrfs_set_header_nritems(left, left_nritems);
3701 
3702 	if (left_nritems)
3703 		btrfs_mark_buffer_dirty(left);
3704 	else
3705 		clean_tree_block(fs_info, left);
3706 
3707 	btrfs_mark_buffer_dirty(right);
3708 
3709 	btrfs_item_key(right, &disk_key, 0);
3710 	btrfs_set_node_key(upper, &disk_key, slot + 1);
3711 	btrfs_mark_buffer_dirty(upper);
3712 
3713 	/* then fixup the leaf pointer in the path */
3714 	if (path->slots[0] >= left_nritems) {
3715 		path->slots[0] -= left_nritems;
3716 		if (btrfs_header_nritems(path->nodes[0]) == 0)
3717 			clean_tree_block(fs_info, path->nodes[0]);
3718 		btrfs_tree_unlock(path->nodes[0]);
3719 		free_extent_buffer(path->nodes[0]);
3720 		path->nodes[0] = right;
3721 		path->slots[1] += 1;
3722 	} else {
3723 		btrfs_tree_unlock(right);
3724 		free_extent_buffer(right);
3725 	}
3726 	return 0;
3727 
3728 out_unlock:
3729 	btrfs_tree_unlock(right);
3730 	free_extent_buffer(right);
3731 	return 1;
3732 }
3733 
3734 /*
3735  * push some data in the path leaf to the right, trying to free up at
3736  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3737  *
3738  * returns 1 if the push failed because the other node didn't have enough
3739  * room, 0 if everything worked out and < 0 if there were major errors.
3740  *
3741  * this will push starting from min_slot to the end of the leaf.  It won't
3742  * push any slot lower than min_slot
3743  */
3744 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3745 			   *root, struct btrfs_path *path,
3746 			   int min_data_size, int data_size,
3747 			   int empty, u32 min_slot)
3748 {
3749 	struct btrfs_fs_info *fs_info = root->fs_info;
3750 	struct extent_buffer *left = path->nodes[0];
3751 	struct extent_buffer *right;
3752 	struct extent_buffer *upper;
3753 	int slot;
3754 	int free_space;
3755 	u32 left_nritems;
3756 	int ret;
3757 
3758 	if (!path->nodes[1])
3759 		return 1;
3760 
3761 	slot = path->slots[1];
3762 	upper = path->nodes[1];
3763 	if (slot >= btrfs_header_nritems(upper) - 1)
3764 		return 1;
3765 
3766 	btrfs_assert_tree_locked(path->nodes[1]);
3767 
3768 	right = read_node_slot(fs_info, upper, slot + 1);
3769 	/*
3770 	 * slot + 1 is not valid or we fail to read the right node,
3771 	 * no big deal, just return.
3772 	 */
3773 	if (IS_ERR(right))
3774 		return 1;
3775 
3776 	btrfs_tree_lock(right);
3777 	btrfs_set_lock_blocking(right);
3778 
3779 	free_space = btrfs_leaf_free_space(fs_info, right);
3780 	if (free_space < data_size)
3781 		goto out_unlock;
3782 
3783 	/* cow and double check */
3784 	ret = btrfs_cow_block(trans, root, right, upper,
3785 			      slot + 1, &right);
3786 	if (ret)
3787 		goto out_unlock;
3788 
3789 	free_space = btrfs_leaf_free_space(fs_info, right);
3790 	if (free_space < data_size)
3791 		goto out_unlock;
3792 
3793 	left_nritems = btrfs_header_nritems(left);
3794 	if (left_nritems == 0)
3795 		goto out_unlock;
3796 
3797 	if (path->slots[0] == left_nritems && !empty) {
3798 		/* Key greater than all keys in the leaf, right neighbor has
3799 		 * enough room for it and we're not emptying our leaf to delete
3800 		 * it, therefore use right neighbor to insert the new item and
3801 		 * no need to touch/dirty our left leaft. */
3802 		btrfs_tree_unlock(left);
3803 		free_extent_buffer(left);
3804 		path->nodes[0] = right;
3805 		path->slots[0] = 0;
3806 		path->slots[1]++;
3807 		return 0;
3808 	}
3809 
3810 	return __push_leaf_right(fs_info, path, min_data_size, empty,
3811 				right, free_space, left_nritems, min_slot);
3812 out_unlock:
3813 	btrfs_tree_unlock(right);
3814 	free_extent_buffer(right);
3815 	return 1;
3816 }
3817 
3818 /*
3819  * push some data in the path leaf to the left, trying to free up at
3820  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3821  *
3822  * max_slot can put a limit on how far into the leaf we'll push items.  The
3823  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us do all the
3824  * items
3825  */
3826 static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info,
3827 				     struct btrfs_path *path, int data_size,
3828 				     int empty, struct extent_buffer *left,
3829 				     int free_space, u32 right_nritems,
3830 				     u32 max_slot)
3831 {
3832 	struct btrfs_disk_key disk_key;
3833 	struct extent_buffer *right = path->nodes[0];
3834 	int i;
3835 	int push_space = 0;
3836 	int push_items = 0;
3837 	struct btrfs_item *item;
3838 	u32 old_left_nritems;
3839 	u32 nr;
3840 	int ret = 0;
3841 	u32 this_item_size;
3842 	u32 old_left_item_size;
3843 	struct btrfs_map_token token;
3844 
3845 	btrfs_init_map_token(&token);
3846 
3847 	if (empty)
3848 		nr = min(right_nritems, max_slot);
3849 	else
3850 		nr = min(right_nritems - 1, max_slot);
3851 
3852 	for (i = 0; i < nr; i++) {
3853 		item = btrfs_item_nr(i);
3854 
3855 		if (!empty && push_items > 0) {
3856 			if (path->slots[0] < i)
3857 				break;
3858 			if (path->slots[0] == i) {
3859 				int space = btrfs_leaf_free_space(fs_info, right);
3860 				if (space + push_space * 2 > free_space)
3861 					break;
3862 			}
3863 		}
3864 
3865 		if (path->slots[0] == i)
3866 			push_space += data_size;
3867 
3868 		this_item_size = btrfs_item_size(right, item);
3869 		if (this_item_size + sizeof(*item) + push_space > free_space)
3870 			break;
3871 
3872 		push_items++;
3873 		push_space += this_item_size + sizeof(*item);
3874 	}
3875 
3876 	if (push_items == 0) {
3877 		ret = 1;
3878 		goto out;
3879 	}
3880 	WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3881 
3882 	/* push data from right to left */
3883 	copy_extent_buffer(left, right,
3884 			   btrfs_item_nr_offset(btrfs_header_nritems(left)),
3885 			   btrfs_item_nr_offset(0),
3886 			   push_items * sizeof(struct btrfs_item));
3887 
3888 	push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3889 		     btrfs_item_offset_nr(right, push_items - 1);
3890 
3891 	copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3892 		     leaf_data_end(fs_info, left) - push_space,
3893 		     BTRFS_LEAF_DATA_OFFSET +
3894 		     btrfs_item_offset_nr(right, push_items - 1),
3895 		     push_space);
3896 	old_left_nritems = btrfs_header_nritems(left);
3897 	BUG_ON(old_left_nritems <= 0);
3898 
3899 	old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3900 	for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3901 		u32 ioff;
3902 
3903 		item = btrfs_item_nr(i);
3904 
3905 		ioff = btrfs_token_item_offset(left, item, &token);
3906 		btrfs_set_token_item_offset(left, item,
3907 		      ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3908 		      &token);
3909 	}
3910 	btrfs_set_header_nritems(left, old_left_nritems + push_items);
3911 
3912 	/* fixup right node */
3913 	if (push_items > right_nritems)
3914 		WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3915 		       right_nritems);
3916 
3917 	if (push_items < right_nritems) {
3918 		push_space = btrfs_item_offset_nr(right, push_items - 1) -
3919 						  leaf_data_end(fs_info, right);
3920 		memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3921 				      BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3922 				      BTRFS_LEAF_DATA_OFFSET +
3923 				      leaf_data_end(fs_info, right), push_space);
3924 
3925 		memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3926 			      btrfs_item_nr_offset(push_items),
3927 			     (btrfs_header_nritems(right) - push_items) *
3928 			     sizeof(struct btrfs_item));
3929 	}
3930 	right_nritems -= push_items;
3931 	btrfs_set_header_nritems(right, right_nritems);
3932 	push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3933 	for (i = 0; i < right_nritems; i++) {
3934 		item = btrfs_item_nr(i);
3935 
3936 		push_space = push_space - btrfs_token_item_size(right,
3937 								item, &token);
3938 		btrfs_set_token_item_offset(right, item, push_space, &token);
3939 	}
3940 
3941 	btrfs_mark_buffer_dirty(left);
3942 	if (right_nritems)
3943 		btrfs_mark_buffer_dirty(right);
3944 	else
3945 		clean_tree_block(fs_info, right);
3946 
3947 	btrfs_item_key(right, &disk_key, 0);
3948 	fixup_low_keys(fs_info, path, &disk_key, 1);
3949 
3950 	/* then fixup the leaf pointer in the path */
3951 	if (path->slots[0] < push_items) {
3952 		path->slots[0] += old_left_nritems;
3953 		btrfs_tree_unlock(path->nodes[0]);
3954 		free_extent_buffer(path->nodes[0]);
3955 		path->nodes[0] = left;
3956 		path->slots[1] -= 1;
3957 	} else {
3958 		btrfs_tree_unlock(left);
3959 		free_extent_buffer(left);
3960 		path->slots[0] -= push_items;
3961 	}
3962 	BUG_ON(path->slots[0] < 0);
3963 	return ret;
3964 out:
3965 	btrfs_tree_unlock(left);
3966 	free_extent_buffer(left);
3967 	return ret;
3968 }
3969 
3970 /*
3971  * push some data in the path leaf to the left, trying to free up at
3972  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3973  *
3974  * max_slot can put a limit on how far into the leaf we'll push items.  The
3975  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us push all the
3976  * items
3977  */
3978 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3979 			  *root, struct btrfs_path *path, int min_data_size,
3980 			  int data_size, int empty, u32 max_slot)
3981 {
3982 	struct btrfs_fs_info *fs_info = root->fs_info;
3983 	struct extent_buffer *right = path->nodes[0];
3984 	struct extent_buffer *left;
3985 	int slot;
3986 	int free_space;
3987 	u32 right_nritems;
3988 	int ret = 0;
3989 
3990 	slot = path->slots[1];
3991 	if (slot == 0)
3992 		return 1;
3993 	if (!path->nodes[1])
3994 		return 1;
3995 
3996 	right_nritems = btrfs_header_nritems(right);
3997 	if (right_nritems == 0)
3998 		return 1;
3999 
4000 	btrfs_assert_tree_locked(path->nodes[1]);
4001 
4002 	left = read_node_slot(fs_info, path->nodes[1], slot - 1);
4003 	/*
4004 	 * slot - 1 is not valid or we fail to read the left node,
4005 	 * no big deal, just return.
4006 	 */
4007 	if (IS_ERR(left))
4008 		return 1;
4009 
4010 	btrfs_tree_lock(left);
4011 	btrfs_set_lock_blocking(left);
4012 
4013 	free_space = btrfs_leaf_free_space(fs_info, left);
4014 	if (free_space < data_size) {
4015 		ret = 1;
4016 		goto out;
4017 	}
4018 
4019 	/* cow and double check */
4020 	ret = btrfs_cow_block(trans, root, left,
4021 			      path->nodes[1], slot - 1, &left);
4022 	if (ret) {
4023 		/* we hit -ENOSPC, but it isn't fatal here */
4024 		if (ret == -ENOSPC)
4025 			ret = 1;
4026 		goto out;
4027 	}
4028 
4029 	free_space = btrfs_leaf_free_space(fs_info, left);
4030 	if (free_space < data_size) {
4031 		ret = 1;
4032 		goto out;
4033 	}
4034 
4035 	return __push_leaf_left(fs_info, path, min_data_size,
4036 			       empty, left, free_space, right_nritems,
4037 			       max_slot);
4038 out:
4039 	btrfs_tree_unlock(left);
4040 	free_extent_buffer(left);
4041 	return ret;
4042 }
4043 
4044 /*
4045  * split the path's leaf in two, making sure there is at least data_size
4046  * available for the resulting leaf level of the path.
4047  */
4048 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4049 				    struct btrfs_fs_info *fs_info,
4050 				    struct btrfs_path *path,
4051 				    struct extent_buffer *l,
4052 				    struct extent_buffer *right,
4053 				    int slot, int mid, int nritems)
4054 {
4055 	int data_copy_size;
4056 	int rt_data_off;
4057 	int i;
4058 	struct btrfs_disk_key disk_key;
4059 	struct btrfs_map_token token;
4060 
4061 	btrfs_init_map_token(&token);
4062 
4063 	nritems = nritems - mid;
4064 	btrfs_set_header_nritems(right, nritems);
4065 	data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l);
4066 
4067 	copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4068 			   btrfs_item_nr_offset(mid),
4069 			   nritems * sizeof(struct btrfs_item));
4070 
4071 	copy_extent_buffer(right, l,
4072 		     BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4073 		     data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4074 		     leaf_data_end(fs_info, l), data_copy_size);
4075 
4076 	rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4077 
4078 	for (i = 0; i < nritems; i++) {
4079 		struct btrfs_item *item = btrfs_item_nr(i);
4080 		u32 ioff;
4081 
4082 		ioff = btrfs_token_item_offset(right, item, &token);
4083 		btrfs_set_token_item_offset(right, item,
4084 					    ioff + rt_data_off, &token);
4085 	}
4086 
4087 	btrfs_set_header_nritems(l, mid);
4088 	btrfs_item_key(right, &disk_key, 0);
4089 	insert_ptr(trans, fs_info, path, &disk_key, right->start,
4090 		   path->slots[1] + 1, 1);
4091 
4092 	btrfs_mark_buffer_dirty(right);
4093 	btrfs_mark_buffer_dirty(l);
4094 	BUG_ON(path->slots[0] != slot);
4095 
4096 	if (mid <= slot) {
4097 		btrfs_tree_unlock(path->nodes[0]);
4098 		free_extent_buffer(path->nodes[0]);
4099 		path->nodes[0] = right;
4100 		path->slots[0] -= mid;
4101 		path->slots[1] += 1;
4102 	} else {
4103 		btrfs_tree_unlock(right);
4104 		free_extent_buffer(right);
4105 	}
4106 
4107 	BUG_ON(path->slots[0] < 0);
4108 }
4109 
4110 /*
4111  * double splits happen when we need to insert a big item in the middle
4112  * of a leaf.  A double split can leave us with 3 mostly empty leaves:
4113  * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4114  *          A                 B                 C
4115  *
4116  * We avoid this by trying to push the items on either side of our target
4117  * into the adjacent leaves.  If all goes well we can avoid the double split
4118  * completely.
4119  */
4120 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4121 					  struct btrfs_root *root,
4122 					  struct btrfs_path *path,
4123 					  int data_size)
4124 {
4125 	struct btrfs_fs_info *fs_info = root->fs_info;
4126 	int ret;
4127 	int progress = 0;
4128 	int slot;
4129 	u32 nritems;
4130 	int space_needed = data_size;
4131 
4132 	slot = path->slots[0];
4133 	if (slot < btrfs_header_nritems(path->nodes[0]))
4134 		space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4135 
4136 	/*
4137 	 * try to push all the items after our slot into the
4138 	 * right leaf
4139 	 */
4140 	ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4141 	if (ret < 0)
4142 		return ret;
4143 
4144 	if (ret == 0)
4145 		progress++;
4146 
4147 	nritems = btrfs_header_nritems(path->nodes[0]);
4148 	/*
4149 	 * our goal is to get our slot at the start or end of a leaf.  If
4150 	 * we've done so we're done
4151 	 */
4152 	if (path->slots[0] == 0 || path->slots[0] == nritems)
4153 		return 0;
4154 
4155 	if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4156 		return 0;
4157 
4158 	/* try to push all the items before our slot into the next leaf */
4159 	slot = path->slots[0];
4160 	space_needed = data_size;
4161 	if (slot > 0)
4162 		space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]);
4163 	ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4164 	if (ret < 0)
4165 		return ret;
4166 
4167 	if (ret == 0)
4168 		progress++;
4169 
4170 	if (progress)
4171 		return 0;
4172 	return 1;
4173 }
4174 
4175 /*
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.
4178  *
4179  * returns 0 if all went well and < 0 on failure.
4180  */
4181 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4182 			       struct btrfs_root *root,
4183 			       const struct btrfs_key *ins_key,
4184 			       struct btrfs_path *path, int data_size,
4185 			       int extend)
4186 {
4187 	struct btrfs_disk_key disk_key;
4188 	struct extent_buffer *l;
4189 	u32 nritems;
4190 	int mid;
4191 	int slot;
4192 	struct extent_buffer *right;
4193 	struct btrfs_fs_info *fs_info = root->fs_info;
4194 	int ret = 0;
4195 	int wret;
4196 	int split;
4197 	int num_doubles = 0;
4198 	int tried_avoid_double = 0;
4199 
4200 	l = path->nodes[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))
4204 		return -EOVERFLOW;
4205 
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;
4209 
4210 		if (slot < btrfs_header_nritems(l))
4211 			space_needed -= btrfs_leaf_free_space(fs_info, l);
4212 
4213 		wret = push_leaf_right(trans, root, path, space_needed,
4214 				       space_needed, 0, 0);
4215 		if (wret < 0)
4216 			return wret;
4217 		if (wret) {
4218 			space_needed = data_size;
4219 			if (slot > 0)
4220 				space_needed -= btrfs_leaf_free_space(fs_info,
4221 								      l);
4222 			wret = push_leaf_left(trans, root, path, space_needed,
4223 					      space_needed, 0, (u32)-1);
4224 			if (wret < 0)
4225 				return wret;
4226 		}
4227 		l = path->nodes[0];
4228 
4229 		/* did the pushes work? */
4230 		if (btrfs_leaf_free_space(fs_info, l) >= data_size)
4231 			return 0;
4232 	}
4233 
4234 	if (!path->nodes[1]) {
4235 		ret = insert_new_root(trans, root, path, 1);
4236 		if (ret)
4237 			return ret;
4238 	}
4239 again:
4240 	split = 1;
4241 	l = path->nodes[0];
4242 	slot = path->slots[0];
4243 	nritems = btrfs_header_nritems(l);
4244 	mid = (nritems + 1) / 2;
4245 
4246 	if (mid <= slot) {
4247 		if (nritems == 1 ||
4248 		    leaf_space_used(l, mid, nritems - mid) + data_size >
4249 			BTRFS_LEAF_DATA_SIZE(fs_info)) {
4250 			if (slot >= nritems) {
4251 				split = 0;
4252 			} else {
4253 				mid = slot;
4254 				if (mid != nritems &&
4255 				    leaf_space_used(l, mid, nritems - mid) +
4256 				    data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4257 					if (data_size && !tried_avoid_double)
4258 						goto push_for_double;
4259 					split = 2;
4260 				}
4261 			}
4262 		}
4263 	} else {
4264 		if (leaf_space_used(l, 0, mid) + data_size >
4265 			BTRFS_LEAF_DATA_SIZE(fs_info)) {
4266 			if (!extend && data_size && slot == 0) {
4267 				split = 0;
4268 			} else if ((extend || !data_size) && slot == 0) {
4269 				mid = 1;
4270 			} else {
4271 				mid = slot;
4272 				if (mid != nritems &&
4273 				    leaf_space_used(l, mid, nritems - mid) +
4274 				    data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4275 					if (data_size && !tried_avoid_double)
4276 						goto push_for_double;
4277 					split = 2;
4278 				}
4279 			}
4280 		}
4281 	}
4282 
4283 	if (split == 0)
4284 		btrfs_cpu_key_to_disk(&disk_key, ins_key);
4285 	else
4286 		btrfs_item_key(l, &disk_key, mid);
4287 
4288 	right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4289 			&disk_key, 0, l->start, 0);
4290 	if (IS_ERR(right))
4291 		return PTR_ERR(right);
4292 
4293 	root_add_used(root, fs_info->nodesize);
4294 
4295 	memzero_extent_buffer(right, 0, sizeof(struct btrfs_header));
4296 	btrfs_set_header_bytenr(right, right->start);
4297 	btrfs_set_header_generation(right, trans->transid);
4298 	btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4299 	btrfs_set_header_owner(right, root->root_key.objectid);
4300 	btrfs_set_header_level(right, 0);
4301 	write_extent_buffer_fsid(right, fs_info->fsid);
4302 	write_extent_buffer_chunk_tree_uuid(right, fs_info->chunk_tree_uuid);
4303 
4304 	if (split == 0) {
4305 		if (mid <= slot) {
4306 			btrfs_set_header_nritems(right, 0);
4307 			insert_ptr(trans, fs_info, path, &disk_key,
4308 				   right->start, path->slots[1] + 1, 1);
4309 			btrfs_tree_unlock(path->nodes[0]);
4310 			free_extent_buffer(path->nodes[0]);
4311 			path->nodes[0] = right;
4312 			path->slots[0] = 0;
4313 			path->slots[1] += 1;
4314 		} else {
4315 			btrfs_set_header_nritems(right, 0);
4316 			insert_ptr(trans, fs_info, path, &disk_key,
4317 				   right->start, path->slots[1], 1);
4318 			btrfs_tree_unlock(path->nodes[0]);
4319 			free_extent_buffer(path->nodes[0]);
4320 			path->nodes[0] = right;
4321 			path->slots[0] = 0;
4322 			if (path->slots[1] == 0)
4323 				fixup_low_keys(fs_info, path, &disk_key, 1);
4324 		}
4325 		/*
4326 		 * We create a new leaf 'right' for the required ins_len and
4327 		 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4328 		 * the content of ins_len to 'right'.
4329 		 */
4330 		return ret;
4331 	}
4332 
4333 	copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems);
4334 
4335 	if (split == 2) {
4336 		BUG_ON(num_doubles != 0);
4337 		num_doubles++;
4338 		goto again;
4339 	}
4340 
4341 	return 0;
4342 
4343 push_for_double:
4344 	push_for_double_split(trans, root, path, data_size);
4345 	tried_avoid_double = 1;
4346 	if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size)
4347 		return 0;
4348 	goto again;
4349 }
4350 
4351 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4352 					 struct btrfs_root *root,
4353 					 struct btrfs_path *path, int ins_len)
4354 {
4355 	struct btrfs_fs_info *fs_info = root->fs_info;
4356 	struct btrfs_key key;
4357 	struct extent_buffer *leaf;
4358 	struct btrfs_file_extent_item *fi;
4359 	u64 extent_len = 0;
4360 	u32 item_size;
4361 	int ret;
4362 
4363 	leaf = path->nodes[0];
4364 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4365 
4366 	BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4367 	       key.type != BTRFS_EXTENT_CSUM_KEY);
4368 
4369 	if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len)
4370 		return 0;
4371 
4372 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4373 	if (key.type == BTRFS_EXTENT_DATA_KEY) {
4374 		fi = btrfs_item_ptr(leaf, path->slots[0],
4375 				    struct btrfs_file_extent_item);
4376 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4377 	}
4378 	btrfs_release_path(path);
4379 
4380 	path->keep_locks = 1;
4381 	path->search_for_split = 1;
4382 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4383 	path->search_for_split = 0;
4384 	if (ret > 0)
4385 		ret = -EAGAIN;
4386 	if (ret < 0)
4387 		goto err;
4388 
4389 	ret = -EAGAIN;
4390 	leaf = path->nodes[0];
4391 	/* if our item isn't there, return now */
4392 	if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4393 		goto err;
4394 
4395 	/* the leaf has  changed, it now has room.  return now */
4396 	if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len)
4397 		goto err;
4398 
4399 	if (key.type == BTRFS_EXTENT_DATA_KEY) {
4400 		fi = btrfs_item_ptr(leaf, path->slots[0],
4401 				    struct btrfs_file_extent_item);
4402 		if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4403 			goto err;
4404 	}
4405 
4406 	btrfs_set_path_blocking(path);
4407 	ret = split_leaf(trans, root, &key, path, ins_len, 1);
4408 	if (ret)
4409 		goto err;
4410 
4411 	path->keep_locks = 0;
4412 	btrfs_unlock_up_safe(path, 1);
4413 	return 0;
4414 err:
4415 	path->keep_locks = 0;
4416 	return ret;
4417 }
4418 
4419 static noinline int split_item(struct btrfs_fs_info *fs_info,
4420 			       struct btrfs_path *path,
4421 			       const struct btrfs_key *new_key,
4422 			       unsigned long split_offset)
4423 {
4424 	struct extent_buffer *leaf;
4425 	struct btrfs_item *item;
4426 	struct btrfs_item *new_item;
4427 	int slot;
4428 	char *buf;
4429 	u32 nritems;
4430 	u32 item_size;
4431 	u32 orig_offset;
4432 	struct btrfs_disk_key disk_key;
4433 
4434 	leaf = path->nodes[0];
4435 	BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item));
4436 
4437 	btrfs_set_path_blocking(path);
4438 
4439 	item = btrfs_item_nr(path->slots[0]);
4440 	orig_offset = btrfs_item_offset(leaf, item);
4441 	item_size = btrfs_item_size(leaf, item);
4442 
4443 	buf = kmalloc(item_size, GFP_NOFS);
4444 	if (!buf)
4445 		return -ENOMEM;
4446 
4447 	read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4448 			    path->slots[0]), item_size);
4449 
4450 	slot = path->slots[0] + 1;
4451 	nritems = btrfs_header_nritems(leaf);
4452 	if (slot != nritems) {
4453 		/* shift the items */
4454 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4455 				btrfs_item_nr_offset(slot),
4456 				(nritems - slot) * sizeof(struct btrfs_item));
4457 	}
4458 
4459 	btrfs_cpu_key_to_disk(&disk_key, new_key);
4460 	btrfs_set_item_key(leaf, &disk_key, slot);
4461 
4462 	new_item = btrfs_item_nr(slot);
4463 
4464 	btrfs_set_item_offset(leaf, new_item, orig_offset);
4465 	btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4466 
4467 	btrfs_set_item_offset(leaf, item,
4468 			      orig_offset + item_size - split_offset);
4469 	btrfs_set_item_size(leaf, item, split_offset);
4470 
4471 	btrfs_set_header_nritems(leaf, nritems + 1);
4472 
4473 	/* write the data for the start of the original item */
4474 	write_extent_buffer(leaf, buf,
4475 			    btrfs_item_ptr_offset(leaf, path->slots[0]),
4476 			    split_offset);
4477 
4478 	/* write the data for the new item */
4479 	write_extent_buffer(leaf, buf + split_offset,
4480 			    btrfs_item_ptr_offset(leaf, slot),
4481 			    item_size - split_offset);
4482 	btrfs_mark_buffer_dirty(leaf);
4483 
4484 	BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0);
4485 	kfree(buf);
4486 	return 0;
4487 }
4488 
4489 /*
4490  * This function splits a single item into two items,
4491  * giving 'new_key' to the new item and splitting the
4492  * old one at split_offset (from the start of the item).
4493  *
4494  * The path may be released by this operation.  After
4495  * the split, the path is pointing to the old item.  The
4496  * new item is going to be in the same node as the old one.
4497  *
4498  * Note, the item being split must be smaller enough to live alone on
4499  * a tree block with room for one extra struct btrfs_item
4500  *
4501  * This allows us to split the item in place, keeping a lock on the
4502  * leaf the entire time.
4503  */
4504 int btrfs_split_item(struct btrfs_trans_handle *trans,
4505 		     struct btrfs_root *root,
4506 		     struct btrfs_path *path,
4507 		     const struct btrfs_key *new_key,
4508 		     unsigned long split_offset)
4509 {
4510 	int ret;
4511 	ret = setup_leaf_for_split(trans, root, path,
4512 				   sizeof(struct btrfs_item));
4513 	if (ret)
4514 		return ret;
4515 
4516 	ret = split_item(root->fs_info, path, new_key, split_offset);
4517 	return ret;
4518 }
4519 
4520 /*
4521  * This function duplicate a item, giving 'new_key' to the new item.
4522  * It guarantees both items live in the same tree leaf and the new item
4523  * is contiguous with the original item.
4524  *
4525  * This allows us to split file extent in place, keeping a lock on the
4526  * leaf the entire time.
4527  */
4528 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4529 			 struct btrfs_root *root,
4530 			 struct btrfs_path *path,
4531 			 const struct btrfs_key *new_key)
4532 {
4533 	struct extent_buffer *leaf;
4534 	int ret;
4535 	u32 item_size;
4536 
4537 	leaf = path->nodes[0];
4538 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4539 	ret = setup_leaf_for_split(trans, root, path,
4540 				   item_size + sizeof(struct btrfs_item));
4541 	if (ret)
4542 		return ret;
4543 
4544 	path->slots[0]++;
4545 	setup_items_for_insert(root, path, new_key, &item_size,
4546 			       item_size, item_size +
4547 			       sizeof(struct btrfs_item), 1);
4548 	leaf = path->nodes[0];
4549 	memcpy_extent_buffer(leaf,
4550 			     btrfs_item_ptr_offset(leaf, path->slots[0]),
4551 			     btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4552 			     item_size);
4553 	return 0;
4554 }
4555 
4556 /*
4557  * make the item pointed to by the path smaller.  new_size indicates
4558  * how small to make it, and from_end tells us if we just chop bytes
4559  * off the end of the item or if we shift the item to chop bytes off
4560  * the front.
4561  */
4562 void btrfs_truncate_item(struct btrfs_fs_info *fs_info,
4563 			 struct btrfs_path *path, u32 new_size, int from_end)
4564 {
4565 	int slot;
4566 	struct extent_buffer *leaf;
4567 	struct btrfs_item *item;
4568 	u32 nritems;
4569 	unsigned int data_end;
4570 	unsigned int old_data_start;
4571 	unsigned int old_size;
4572 	unsigned int size_diff;
4573 	int i;
4574 	struct btrfs_map_token token;
4575 
4576 	btrfs_init_map_token(&token);
4577 
4578 	leaf = path->nodes[0];
4579 	slot = path->slots[0];
4580 
4581 	old_size = btrfs_item_size_nr(leaf, slot);
4582 	if (old_size == new_size)
4583 		return;
4584 
4585 	nritems = btrfs_header_nritems(leaf);
4586 	data_end = leaf_data_end(fs_info, leaf);
4587 
4588 	old_data_start = btrfs_item_offset_nr(leaf, slot);
4589 
4590 	size_diff = old_size - new_size;
4591 
4592 	BUG_ON(slot < 0);
4593 	BUG_ON(slot >= nritems);
4594 
4595 	/*
4596 	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4597 	 */
4598 	/* first correct the data pointers */
4599 	for (i = slot; i < nritems; i++) {
4600 		u32 ioff;
4601 		item = btrfs_item_nr(i);
4602 
4603 		ioff = btrfs_token_item_offset(leaf, item, &token);
4604 		btrfs_set_token_item_offset(leaf, item,
4605 					    ioff + size_diff, &token);
4606 	}
4607 
4608 	/* shift the data */
4609 	if (from_end) {
4610 		memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4611 			      data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4612 			      data_end, old_data_start + new_size - data_end);
4613 	} else {
4614 		struct btrfs_disk_key disk_key;
4615 		u64 offset;
4616 
4617 		btrfs_item_key(leaf, &disk_key, slot);
4618 
4619 		if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4620 			unsigned long ptr;
4621 			struct btrfs_file_extent_item *fi;
4622 
4623 			fi = btrfs_item_ptr(leaf, slot,
4624 					    struct btrfs_file_extent_item);
4625 			fi = (struct btrfs_file_extent_item *)(
4626 			     (unsigned long)fi - size_diff);
4627 
4628 			if (btrfs_file_extent_type(leaf, fi) ==
4629 			    BTRFS_FILE_EXTENT_INLINE) {
4630 				ptr = btrfs_item_ptr_offset(leaf, slot);
4631 				memmove_extent_buffer(leaf, ptr,
4632 				      (unsigned long)fi,
4633 				      BTRFS_FILE_EXTENT_INLINE_DATA_START);
4634 			}
4635 		}
4636 
4637 		memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4638 			      data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4639 			      data_end, old_data_start - data_end);
4640 
4641 		offset = btrfs_disk_key_offset(&disk_key);
4642 		btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4643 		btrfs_set_item_key(leaf, &disk_key, slot);
4644 		if (slot == 0)
4645 			fixup_low_keys(fs_info, path, &disk_key, 1);
4646 	}
4647 
4648 	item = btrfs_item_nr(slot);
4649 	btrfs_set_item_size(leaf, item, new_size);
4650 	btrfs_mark_buffer_dirty(leaf);
4651 
4652 	if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4653 		btrfs_print_leaf(fs_info, leaf);
4654 		BUG();
4655 	}
4656 }
4657 
4658 /*
4659  * make the item pointed to by the path bigger, data_size is the added size.
4660  */
4661 void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
4662 		       u32 data_size)
4663 {
4664 	int slot;
4665 	struct extent_buffer *leaf;
4666 	struct btrfs_item *item;
4667 	u32 nritems;
4668 	unsigned int data_end;
4669 	unsigned int old_data;
4670 	unsigned int old_size;
4671 	int i;
4672 	struct btrfs_map_token token;
4673 
4674 	btrfs_init_map_token(&token);
4675 
4676 	leaf = path->nodes[0];
4677 
4678 	nritems = btrfs_header_nritems(leaf);
4679 	data_end = leaf_data_end(fs_info, leaf);
4680 
4681 	if (btrfs_leaf_free_space(fs_info, leaf) < data_size) {
4682 		btrfs_print_leaf(fs_info, leaf);
4683 		BUG();
4684 	}
4685 	slot = path->slots[0];
4686 	old_data = btrfs_item_end_nr(leaf, slot);
4687 
4688 	BUG_ON(slot < 0);
4689 	if (slot >= nritems) {
4690 		btrfs_print_leaf(fs_info, leaf);
4691 		btrfs_crit(fs_info, "slot %d too large, nritems %d",
4692 			   slot, nritems);
4693 		BUG_ON(1);
4694 	}
4695 
4696 	/*
4697 	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4698 	 */
4699 	/* first correct the data pointers */
4700 	for (i = slot; i < nritems; i++) {
4701 		u32 ioff;
4702 		item = btrfs_item_nr(i);
4703 
4704 		ioff = btrfs_token_item_offset(leaf, item, &token);
4705 		btrfs_set_token_item_offset(leaf, item,
4706 					    ioff - data_size, &token);
4707 	}
4708 
4709 	/* shift the data */
4710 	memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4711 		      data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4712 		      data_end, old_data - data_end);
4713 
4714 	data_end = old_data;
4715 	old_size = btrfs_item_size_nr(leaf, slot);
4716 	item = btrfs_item_nr(slot);
4717 	btrfs_set_item_size(leaf, item, old_size + data_size);
4718 	btrfs_mark_buffer_dirty(leaf);
4719 
4720 	if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4721 		btrfs_print_leaf(fs_info, leaf);
4722 		BUG();
4723 	}
4724 }
4725 
4726 /*
4727  * this is a helper for btrfs_insert_empty_items, the main goal here is
4728  * to save stack depth by doing the bulk of the work in a function
4729  * that doesn't call btrfs_search_slot
4730  */
4731 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4732 			    const struct btrfs_key *cpu_key, u32 *data_size,
4733 			    u32 total_data, u32 total_size, int nr)
4734 {
4735 	struct btrfs_fs_info *fs_info = root->fs_info;
4736 	struct btrfs_item *item;
4737 	int i;
4738 	u32 nritems;
4739 	unsigned int data_end;
4740 	struct btrfs_disk_key disk_key;
4741 	struct extent_buffer *leaf;
4742 	int slot;
4743 	struct btrfs_map_token token;
4744 
4745 	if (path->slots[0] == 0) {
4746 		btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4747 		fixup_low_keys(fs_info, path, &disk_key, 1);
4748 	}
4749 	btrfs_unlock_up_safe(path, 1);
4750 
4751 	btrfs_init_map_token(&token);
4752 
4753 	leaf = path->nodes[0];
4754 	slot = path->slots[0];
4755 
4756 	nritems = btrfs_header_nritems(leaf);
4757 	data_end = leaf_data_end(fs_info, leaf);
4758 
4759 	if (btrfs_leaf_free_space(fs_info, leaf) < total_size) {
4760 		btrfs_print_leaf(fs_info, leaf);
4761 		btrfs_crit(fs_info, "not enough freespace need %u have %d",
4762 			   total_size, btrfs_leaf_free_space(fs_info, leaf));
4763 		BUG();
4764 	}
4765 
4766 	if (slot != nritems) {
4767 		unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4768 
4769 		if (old_data < data_end) {
4770 			btrfs_print_leaf(fs_info, leaf);
4771 			btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4772 				   slot, old_data, data_end);
4773 			BUG_ON(1);
4774 		}
4775 		/*
4776 		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4777 		 */
4778 		/* first correct the data pointers */
4779 		for (i = slot; i < nritems; i++) {
4780 			u32 ioff;
4781 
4782 			item = btrfs_item_nr(i);
4783 			ioff = btrfs_token_item_offset(leaf, item, &token);
4784 			btrfs_set_token_item_offset(leaf, item,
4785 						    ioff - total_data, &token);
4786 		}
4787 		/* shift the items */
4788 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4789 			      btrfs_item_nr_offset(slot),
4790 			      (nritems - slot) * sizeof(struct btrfs_item));
4791 
4792 		/* shift the data */
4793 		memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4794 			      data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4795 			      data_end, old_data - data_end);
4796 		data_end = old_data;
4797 	}
4798 
4799 	/* setup the item for the new data */
4800 	for (i = 0; i < nr; i++) {
4801 		btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4802 		btrfs_set_item_key(leaf, &disk_key, slot + i);
4803 		item = btrfs_item_nr(slot + i);
4804 		btrfs_set_token_item_offset(leaf, item,
4805 					    data_end - data_size[i], &token);
4806 		data_end -= data_size[i];
4807 		btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4808 	}
4809 
4810 	btrfs_set_header_nritems(leaf, nritems + nr);
4811 	btrfs_mark_buffer_dirty(leaf);
4812 
4813 	if (btrfs_leaf_free_space(fs_info, leaf) < 0) {
4814 		btrfs_print_leaf(fs_info, leaf);
4815 		BUG();
4816 	}
4817 }
4818 
4819 /*
4820  * Given a key and some data, insert items into the tree.
4821  * This does all the path init required, making room in the tree if needed.
4822  */
4823 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4824 			    struct btrfs_root *root,
4825 			    struct btrfs_path *path,
4826 			    const struct btrfs_key *cpu_key, u32 *data_size,
4827 			    int nr)
4828 {
4829 	int ret = 0;
4830 	int slot;
4831 	int i;
4832 	u32 total_size = 0;
4833 	u32 total_data = 0;
4834 
4835 	for (i = 0; i < nr; i++)
4836 		total_data += data_size[i];
4837 
4838 	total_size = total_data + (nr * sizeof(struct btrfs_item));
4839 	ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4840 	if (ret == 0)
4841 		return -EEXIST;
4842 	if (ret < 0)
4843 		return ret;
4844 
4845 	slot = path->slots[0];
4846 	BUG_ON(slot < 0);
4847 
4848 	setup_items_for_insert(root, path, cpu_key, data_size,
4849 			       total_data, total_size, nr);
4850 	return 0;
4851 }
4852 
4853 /*
4854  * Given a key and some data, insert an item into the tree.
4855  * This does all the path init required, making room in the tree if needed.
4856  */
4857 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4858 		      const struct btrfs_key *cpu_key, void *data,
4859 		      u32 data_size)
4860 {
4861 	int ret = 0;
4862 	struct btrfs_path *path;
4863 	struct extent_buffer *leaf;
4864 	unsigned long ptr;
4865 
4866 	path = btrfs_alloc_path();
4867 	if (!path)
4868 		return -ENOMEM;
4869 	ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4870 	if (!ret) {
4871 		leaf = path->nodes[0];
4872 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4873 		write_extent_buffer(leaf, data, ptr, data_size);
4874 		btrfs_mark_buffer_dirty(leaf);
4875 	}
4876 	btrfs_free_path(path);
4877 	return ret;
4878 }
4879 
4880 /*
4881  * delete the pointer from a given node.
4882  *
4883  * the tree should have been previously balanced so the deletion does not
4884  * empty a node.
4885  */
4886 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4887 		    int level, int slot)
4888 {
4889 	struct btrfs_fs_info *fs_info = root->fs_info;
4890 	struct extent_buffer *parent = path->nodes[level];
4891 	u32 nritems;
4892 	int ret;
4893 
4894 	nritems = btrfs_header_nritems(parent);
4895 	if (slot != nritems - 1) {
4896 		if (level)
4897 			tree_mod_log_eb_move(fs_info, parent, slot,
4898 					     slot + 1, nritems - slot - 1);
4899 		memmove_extent_buffer(parent,
4900 			      btrfs_node_key_ptr_offset(slot),
4901 			      btrfs_node_key_ptr_offset(slot + 1),
4902 			      sizeof(struct btrfs_key_ptr) *
4903 			      (nritems - slot - 1));
4904 	} else if (level) {
4905 		ret = tree_mod_log_insert_key(fs_info, parent, slot,
4906 					      MOD_LOG_KEY_REMOVE, GFP_NOFS);
4907 		BUG_ON(ret < 0);
4908 	}
4909 
4910 	nritems--;
4911 	btrfs_set_header_nritems(parent, nritems);
4912 	if (nritems == 0 && parent == root->node) {
4913 		BUG_ON(btrfs_header_level(root->node) != 1);
4914 		/* just turn the root into a leaf and break */
4915 		btrfs_set_header_level(root->node, 0);
4916 	} else if (slot == 0) {
4917 		struct btrfs_disk_key disk_key;
4918 
4919 		btrfs_node_key(parent, &disk_key, 0);
4920 		fixup_low_keys(fs_info, path, &disk_key, level + 1);
4921 	}
4922 	btrfs_mark_buffer_dirty(parent);
4923 }
4924 
4925 /*
4926  * a helper function to delete the leaf pointed to by path->slots[1] and
4927  * path->nodes[1].
4928  *
4929  * This deletes the pointer in path->nodes[1] and frees the leaf
4930  * block extent.  zero is returned if it all worked out, < 0 otherwise.
4931  *
4932  * The path must have already been setup for deleting the leaf, including
4933  * all the proper balancing.  path->nodes[1] must be locked.
4934  */
4935 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4936 				    struct btrfs_root *root,
4937 				    struct btrfs_path *path,
4938 				    struct extent_buffer *leaf)
4939 {
4940 	WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4941 	del_ptr(root, path, 1, path->slots[1]);
4942 
4943 	/*
4944 	 * btrfs_free_extent is expensive, we want to make sure we
4945 	 * aren't holding any locks when we call it
4946 	 */
4947 	btrfs_unlock_up_safe(path, 0);
4948 
4949 	root_sub_used(root, leaf->len);
4950 
4951 	extent_buffer_get(leaf);
4952 	btrfs_free_tree_block(trans, root, leaf, 0, 1);
4953 	free_extent_buffer_stale(leaf);
4954 }
4955 /*
4956  * delete the item at the leaf level in path.  If that empties
4957  * the leaf, remove it from the tree
4958  */
4959 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4960 		    struct btrfs_path *path, int slot, int nr)
4961 {
4962 	struct btrfs_fs_info *fs_info = root->fs_info;
4963 	struct extent_buffer *leaf;
4964 	struct btrfs_item *item;
4965 	u32 last_off;
4966 	u32 dsize = 0;
4967 	int ret = 0;
4968 	int wret;
4969 	int i;
4970 	u32 nritems;
4971 	struct btrfs_map_token token;
4972 
4973 	btrfs_init_map_token(&token);
4974 
4975 	leaf = path->nodes[0];
4976 	last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4977 
4978 	for (i = 0; i < nr; i++)
4979 		dsize += btrfs_item_size_nr(leaf, slot + i);
4980 
4981 	nritems = btrfs_header_nritems(leaf);
4982 
4983 	if (slot + nr != nritems) {
4984 		int data_end = leaf_data_end(fs_info, leaf);
4985 
4986 		memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4987 			      data_end + dsize,
4988 			      BTRFS_LEAF_DATA_OFFSET + data_end,
4989 			      last_off - data_end);
4990 
4991 		for (i = slot + nr; i < nritems; i++) {
4992 			u32 ioff;
4993 
4994 			item = btrfs_item_nr(i);
4995 			ioff = btrfs_token_item_offset(leaf, item, &token);
4996 			btrfs_set_token_item_offset(leaf, item,
4997 						    ioff + dsize, &token);
4998 		}
4999 
5000 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
5001 			      btrfs_item_nr_offset(slot + nr),
5002 			      sizeof(struct btrfs_item) *
5003 			      (nritems - slot - nr));
5004 	}
5005 	btrfs_set_header_nritems(leaf, nritems - nr);
5006 	nritems -= nr;
5007 
5008 	/* delete the leaf if we've emptied it */
5009 	if (nritems == 0) {
5010 		if (leaf == root->node) {
5011 			btrfs_set_header_level(leaf, 0);
5012 		} else {
5013 			btrfs_set_path_blocking(path);
5014 			clean_tree_block(fs_info, leaf);
5015 			btrfs_del_leaf(trans, root, path, leaf);
5016 		}
5017 	} else {
5018 		int used = leaf_space_used(leaf, 0, nritems);
5019 		if (slot == 0) {
5020 			struct btrfs_disk_key disk_key;
5021 
5022 			btrfs_item_key(leaf, &disk_key, 0);
5023 			fixup_low_keys(fs_info, path, &disk_key, 1);
5024 		}
5025 
5026 		/* delete the leaf if it is mostly empty */
5027 		if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
5028 			/* push_leaf_left fixes the path.
5029 			 * make sure the path still points to our leaf
5030 			 * for possible call to del_ptr below
5031 			 */
5032 			slot = path->slots[1];
5033 			extent_buffer_get(leaf);
5034 
5035 			btrfs_set_path_blocking(path);
5036 			wret = push_leaf_left(trans, root, path, 1, 1,
5037 					      1, (u32)-1);
5038 			if (wret < 0 && wret != -ENOSPC)
5039 				ret = wret;
5040 
5041 			if (path->nodes[0] == leaf &&
5042 			    btrfs_header_nritems(leaf)) {
5043 				wret = push_leaf_right(trans, root, path, 1,
5044 						       1, 1, 0);
5045 				if (wret < 0 && wret != -ENOSPC)
5046 					ret = wret;
5047 			}
5048 
5049 			if (btrfs_header_nritems(leaf) == 0) {
5050 				path->slots[1] = slot;
5051 				btrfs_del_leaf(trans, root, path, leaf);
5052 				free_extent_buffer(leaf);
5053 				ret = 0;
5054 			} else {
5055 				/* if we're still in the path, make sure
5056 				 * we're dirty.  Otherwise, one of the
5057 				 * push_leaf functions must have already
5058 				 * dirtied this buffer
5059 				 */
5060 				if (path->nodes[0] == leaf)
5061 					btrfs_mark_buffer_dirty(leaf);
5062 				free_extent_buffer(leaf);
5063 			}
5064 		} else {
5065 			btrfs_mark_buffer_dirty(leaf);
5066 		}
5067 	}
5068 	return ret;
5069 }
5070 
5071 /*
5072  * search the tree again to find a leaf with lesser keys
5073  * returns 0 if it found something or 1 if there are no lesser leaves.
5074  * returns < 0 on io errors.
5075  *
5076  * This may release the path, and so you may lose any locks held at the
5077  * time you call it.
5078  */
5079 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5080 {
5081 	struct btrfs_key key;
5082 	struct btrfs_disk_key found_key;
5083 	int ret;
5084 
5085 	btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5086 
5087 	if (key.offset > 0) {
5088 		key.offset--;
5089 	} else if (key.type > 0) {
5090 		key.type--;
5091 		key.offset = (u64)-1;
5092 	} else if (key.objectid > 0) {
5093 		key.objectid--;
5094 		key.type = (u8)-1;
5095 		key.offset = (u64)-1;
5096 	} else {
5097 		return 1;
5098 	}
5099 
5100 	btrfs_release_path(path);
5101 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5102 	if (ret < 0)
5103 		return ret;
5104 	btrfs_item_key(path->nodes[0], &found_key, 0);
5105 	ret = comp_keys(&found_key, &key);
5106 	/*
5107 	 * We might have had an item with the previous key in the tree right
5108 	 * before we released our path. And after we released our path, that
5109 	 * item might have been pushed to the first slot (0) of the leaf we
5110 	 * were holding due to a tree balance. Alternatively, an item with the
5111 	 * previous key can exist as the only element of a leaf (big fat item).
5112 	 * Therefore account for these 2 cases, so that our callers (like
5113 	 * btrfs_previous_item) don't miss an existing item with a key matching
5114 	 * the previous key we computed above.
5115 	 */
5116 	if (ret <= 0)
5117 		return 0;
5118 	return 1;
5119 }
5120 
5121 /*
5122  * A helper function to walk down the tree starting at min_key, and looking
5123  * for nodes or leaves that are have a minimum transaction id.
5124  * This is used by the btree defrag code, and tree logging
5125  *
5126  * This does not cow, but it does stuff the starting key it finds back
5127  * into min_key, so you can call btrfs_search_slot with cow=1 on the
5128  * key and get a writable path.
5129  *
5130  * This does lock as it descends, and path->keep_locks should be set
5131  * to 1 by the caller.
5132  *
5133  * This honors path->lowest_level to prevent descent past a given level
5134  * of the tree.
5135  *
5136  * min_trans indicates the oldest transaction that you are interested
5137  * in walking through.  Any nodes or leaves older than min_trans are
5138  * skipped over (without reading them).
5139  *
5140  * returns zero if something useful was found, < 0 on error and 1 if there
5141  * was nothing in the tree that matched the search criteria.
5142  */
5143 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5144 			 struct btrfs_path *path,
5145 			 u64 min_trans)
5146 {
5147 	struct btrfs_fs_info *fs_info = root->fs_info;
5148 	struct extent_buffer *cur;
5149 	struct btrfs_key found_key;
5150 	int slot;
5151 	int sret;
5152 	u32 nritems;
5153 	int level;
5154 	int ret = 1;
5155 	int keep_locks = path->keep_locks;
5156 
5157 	path->keep_locks = 1;
5158 again:
5159 	cur = btrfs_read_lock_root_node(root);
5160 	level = btrfs_header_level(cur);
5161 	WARN_ON(path->nodes[level]);
5162 	path->nodes[level] = cur;
5163 	path->locks[level] = BTRFS_READ_LOCK;
5164 
5165 	if (btrfs_header_generation(cur) < min_trans) {
5166 		ret = 1;
5167 		goto out;
5168 	}
5169 	while (1) {
5170 		nritems = btrfs_header_nritems(cur);
5171 		level = btrfs_header_level(cur);
5172 		sret = bin_search(cur, min_key, level, &slot);
5173 
5174 		/* at the lowest level, we're done, setup the path and exit */
5175 		if (level == path->lowest_level) {
5176 			if (slot >= nritems)
5177 				goto find_next_key;
5178 			ret = 0;
5179 			path->slots[level] = slot;
5180 			btrfs_item_key_to_cpu(cur, &found_key, slot);
5181 			goto out;
5182 		}
5183 		if (sret && slot > 0)
5184 			slot--;
5185 		/*
5186 		 * check this node pointer against the min_trans parameters.
5187 		 * If it is too old, old, skip to the next one.
5188 		 */
5189 		while (slot < nritems) {
5190 			u64 gen;
5191 
5192 			gen = btrfs_node_ptr_generation(cur, slot);
5193 			if (gen < min_trans) {
5194 				slot++;
5195 				continue;
5196 			}
5197 			break;
5198 		}
5199 find_next_key:
5200 		/*
5201 		 * we didn't find a candidate key in this node, walk forward
5202 		 * and find another one
5203 		 */
5204 		if (slot >= nritems) {
5205 			path->slots[level] = slot;
5206 			btrfs_set_path_blocking(path);
5207 			sret = btrfs_find_next_key(root, path, min_key, level,
5208 						  min_trans);
5209 			if (sret == 0) {
5210 				btrfs_release_path(path);
5211 				goto again;
5212 			} else {
5213 				goto out;
5214 			}
5215 		}
5216 		/* save our key for returning back */
5217 		btrfs_node_key_to_cpu(cur, &found_key, slot);
5218 		path->slots[level] = slot;
5219 		if (level == path->lowest_level) {
5220 			ret = 0;
5221 			goto out;
5222 		}
5223 		btrfs_set_path_blocking(path);
5224 		cur = read_node_slot(fs_info, cur, slot);
5225 		if (IS_ERR(cur)) {
5226 			ret = PTR_ERR(cur);
5227 			goto out;
5228 		}
5229 
5230 		btrfs_tree_read_lock(cur);
5231 
5232 		path->locks[level - 1] = BTRFS_READ_LOCK;
5233 		path->nodes[level - 1] = cur;
5234 		unlock_up(path, level, 1, 0, NULL);
5235 		btrfs_clear_path_blocking(path, NULL, 0);
5236 	}
5237 out:
5238 	path->keep_locks = keep_locks;
5239 	if (ret == 0) {
5240 		btrfs_unlock_up_safe(path, path->lowest_level + 1);
5241 		btrfs_set_path_blocking(path);
5242 		memcpy(min_key, &found_key, sizeof(found_key));
5243 	}
5244 	return ret;
5245 }
5246 
5247 static int tree_move_down(struct btrfs_fs_info *fs_info,
5248 			   struct btrfs_path *path,
5249 			   int *level)
5250 {
5251 	struct extent_buffer *eb;
5252 
5253 	BUG_ON(*level == 0);
5254 	eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]);
5255 	if (IS_ERR(eb))
5256 		return PTR_ERR(eb);
5257 
5258 	path->nodes[*level - 1] = eb;
5259 	path->slots[*level - 1] = 0;
5260 	(*level)--;
5261 	return 0;
5262 }
5263 
5264 static int tree_move_next_or_upnext(struct btrfs_path *path,
5265 				    int *level, int root_level)
5266 {
5267 	int ret = 0;
5268 	int nritems;
5269 	nritems = btrfs_header_nritems(path->nodes[*level]);
5270 
5271 	path->slots[*level]++;
5272 
5273 	while (path->slots[*level] >= nritems) {
5274 		if (*level == root_level)
5275 			return -1;
5276 
5277 		/* move upnext */
5278 		path->slots[*level] = 0;
5279 		free_extent_buffer(path->nodes[*level]);
5280 		path->nodes[*level] = NULL;
5281 		(*level)++;
5282 		path->slots[*level]++;
5283 
5284 		nritems = btrfs_header_nritems(path->nodes[*level]);
5285 		ret = 1;
5286 	}
5287 	return ret;
5288 }
5289 
5290 /*
5291  * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5292  * or down.
5293  */
5294 static int tree_advance(struct btrfs_fs_info *fs_info,
5295 			struct btrfs_path *path,
5296 			int *level, int root_level,
5297 			int allow_down,
5298 			struct btrfs_key *key)
5299 {
5300 	int ret;
5301 
5302 	if (*level == 0 || !allow_down) {
5303 		ret = tree_move_next_or_upnext(path, level, root_level);
5304 	} else {
5305 		ret = tree_move_down(fs_info, path, level);
5306 	}
5307 	if (ret >= 0) {
5308 		if (*level == 0)
5309 			btrfs_item_key_to_cpu(path->nodes[*level], key,
5310 					path->slots[*level]);
5311 		else
5312 			btrfs_node_key_to_cpu(path->nodes[*level], key,
5313 					path->slots[*level]);
5314 	}
5315 	return ret;
5316 }
5317 
5318 static int tree_compare_item(struct btrfs_path *left_path,
5319 			     struct btrfs_path *right_path,
5320 			     char *tmp_buf)
5321 {
5322 	int cmp;
5323 	int len1, len2;
5324 	unsigned long off1, off2;
5325 
5326 	len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5327 	len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5328 	if (len1 != len2)
5329 		return 1;
5330 
5331 	off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5332 	off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5333 				right_path->slots[0]);
5334 
5335 	read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5336 
5337 	cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5338 	if (cmp)
5339 		return 1;
5340 	return 0;
5341 }
5342 
5343 #define ADVANCE 1
5344 #define ADVANCE_ONLY_NEXT -1
5345 
5346 /*
5347  * This function compares two trees and calls the provided callback for
5348  * every changed/new/deleted item it finds.
5349  * If shared tree blocks are encountered, whole subtrees are skipped, making
5350  * the compare pretty fast on snapshotted subvolumes.
5351  *
5352  * This currently works on commit roots only. As commit roots are read only,
5353  * we don't do any locking. The commit roots are protected with transactions.
5354  * Transactions are ended and rejoined when a commit is tried in between.
5355  *
5356  * This function checks for modifications done to the trees while comparing.
5357  * If it detects a change, it aborts immediately.
5358  */
5359 int btrfs_compare_trees(struct btrfs_root *left_root,
5360 			struct btrfs_root *right_root,
5361 			btrfs_changed_cb_t changed_cb, void *ctx)
5362 {
5363 	struct btrfs_fs_info *fs_info = left_root->fs_info;
5364 	int ret;
5365 	int cmp;
5366 	struct btrfs_path *left_path = NULL;
5367 	struct btrfs_path *right_path = NULL;
5368 	struct btrfs_key left_key;
5369 	struct btrfs_key right_key;
5370 	char *tmp_buf = NULL;
5371 	int left_root_level;
5372 	int right_root_level;
5373 	int left_level;
5374 	int right_level;
5375 	int left_end_reached;
5376 	int right_end_reached;
5377 	int advance_left;
5378 	int advance_right;
5379 	u64 left_blockptr;
5380 	u64 right_blockptr;
5381 	u64 left_gen;
5382 	u64 right_gen;
5383 
5384 	left_path = btrfs_alloc_path();
5385 	if (!left_path) {
5386 		ret = -ENOMEM;
5387 		goto out;
5388 	}
5389 	right_path = btrfs_alloc_path();
5390 	if (!right_path) {
5391 		ret = -ENOMEM;
5392 		goto out;
5393 	}
5394 
5395 	tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
5396 	if (!tmp_buf) {
5397 		ret = -ENOMEM;
5398 		goto out;
5399 	}
5400 
5401 	left_path->search_commit_root = 1;
5402 	left_path->skip_locking = 1;
5403 	right_path->search_commit_root = 1;
5404 	right_path->skip_locking = 1;
5405 
5406 	/*
5407 	 * Strategy: Go to the first items of both trees. Then do
5408 	 *
5409 	 * If both trees are at level 0
5410 	 *   Compare keys of current items
5411 	 *     If left < right treat left item as new, advance left tree
5412 	 *       and repeat
5413 	 *     If left > right treat right item as deleted, advance right tree
5414 	 *       and repeat
5415 	 *     If left == right do deep compare of items, treat as changed if
5416 	 *       needed, advance both trees and repeat
5417 	 * If both trees are at the same level but not at level 0
5418 	 *   Compare keys of current nodes/leafs
5419 	 *     If left < right advance left tree and repeat
5420 	 *     If left > right advance right tree and repeat
5421 	 *     If left == right compare blockptrs of the next nodes/leafs
5422 	 *       If they match advance both trees but stay at the same level
5423 	 *         and repeat
5424 	 *       If they don't match advance both trees while allowing to go
5425 	 *         deeper and repeat
5426 	 * If tree levels are different
5427 	 *   Advance the tree that needs it and repeat
5428 	 *
5429 	 * Advancing a tree means:
5430 	 *   If we are at level 0, try to go to the next slot. If that's not
5431 	 *   possible, go one level up and repeat. Stop when we found a level
5432 	 *   where we could go to the next slot. We may at this point be on a
5433 	 *   node or a leaf.
5434 	 *
5435 	 *   If we are not at level 0 and not on shared tree blocks, go one
5436 	 *   level deeper.
5437 	 *
5438 	 *   If we are not at level 0 and on shared tree blocks, go one slot to
5439 	 *   the right if possible or go up and right.
5440 	 */
5441 
5442 	down_read(&fs_info->commit_root_sem);
5443 	left_level = btrfs_header_level(left_root->commit_root);
5444 	left_root_level = left_level;
5445 	left_path->nodes[left_level] = left_root->commit_root;
5446 	extent_buffer_get(left_path->nodes[left_level]);
5447 
5448 	right_level = btrfs_header_level(right_root->commit_root);
5449 	right_root_level = right_level;
5450 	right_path->nodes[right_level] = right_root->commit_root;
5451 	extent_buffer_get(right_path->nodes[right_level]);
5452 	up_read(&fs_info->commit_root_sem);
5453 
5454 	if (left_level == 0)
5455 		btrfs_item_key_to_cpu(left_path->nodes[left_level],
5456 				&left_key, left_path->slots[left_level]);
5457 	else
5458 		btrfs_node_key_to_cpu(left_path->nodes[left_level],
5459 				&left_key, left_path->slots[left_level]);
5460 	if (right_level == 0)
5461 		btrfs_item_key_to_cpu(right_path->nodes[right_level],
5462 				&right_key, right_path->slots[right_level]);
5463 	else
5464 		btrfs_node_key_to_cpu(right_path->nodes[right_level],
5465 				&right_key, right_path->slots[right_level]);
5466 
5467 	left_end_reached = right_end_reached = 0;
5468 	advance_left = advance_right = 0;
5469 
5470 	while (1) {
5471 		if (advance_left && !left_end_reached) {
5472 			ret = tree_advance(fs_info, left_path, &left_level,
5473 					left_root_level,
5474 					advance_left != ADVANCE_ONLY_NEXT,
5475 					&left_key);
5476 			if (ret == -1)
5477 				left_end_reached = ADVANCE;
5478 			else if (ret < 0)
5479 				goto out;
5480 			advance_left = 0;
5481 		}
5482 		if (advance_right && !right_end_reached) {
5483 			ret = tree_advance(fs_info, right_path, &right_level,
5484 					right_root_level,
5485 					advance_right != ADVANCE_ONLY_NEXT,
5486 					&right_key);
5487 			if (ret == -1)
5488 				right_end_reached = ADVANCE;
5489 			else if (ret < 0)
5490 				goto out;
5491 			advance_right = 0;
5492 		}
5493 
5494 		if (left_end_reached && right_end_reached) {
5495 			ret = 0;
5496 			goto out;
5497 		} else if (left_end_reached) {
5498 			if (right_level == 0) {
5499 				ret = changed_cb(left_root, right_root,
5500 						left_path, right_path,
5501 						&right_key,
5502 						BTRFS_COMPARE_TREE_DELETED,
5503 						ctx);
5504 				if (ret < 0)
5505 					goto out;
5506 			}
5507 			advance_right = ADVANCE;
5508 			continue;
5509 		} else if (right_end_reached) {
5510 			if (left_level == 0) {
5511 				ret = changed_cb(left_root, right_root,
5512 						left_path, right_path,
5513 						&left_key,
5514 						BTRFS_COMPARE_TREE_NEW,
5515 						ctx);
5516 				if (ret < 0)
5517 					goto out;
5518 			}
5519 			advance_left = ADVANCE;
5520 			continue;
5521 		}
5522 
5523 		if (left_level == 0 && right_level == 0) {
5524 			cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5525 			if (cmp < 0) {
5526 				ret = changed_cb(left_root, right_root,
5527 						left_path, right_path,
5528 						&left_key,
5529 						BTRFS_COMPARE_TREE_NEW,
5530 						ctx);
5531 				if (ret < 0)
5532 					goto out;
5533 				advance_left = ADVANCE;
5534 			} else if (cmp > 0) {
5535 				ret = changed_cb(left_root, right_root,
5536 						left_path, right_path,
5537 						&right_key,
5538 						BTRFS_COMPARE_TREE_DELETED,
5539 						ctx);
5540 				if (ret < 0)
5541 					goto out;
5542 				advance_right = ADVANCE;
5543 			} else {
5544 				enum btrfs_compare_tree_result result;
5545 
5546 				WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5547 				ret = tree_compare_item(left_path, right_path,
5548 							tmp_buf);
5549 				if (ret)
5550 					result = BTRFS_COMPARE_TREE_CHANGED;
5551 				else
5552 					result = BTRFS_COMPARE_TREE_SAME;
5553 				ret = changed_cb(left_root, right_root,
5554 						 left_path, right_path,
5555 						 &left_key, result, ctx);
5556 				if (ret < 0)
5557 					goto out;
5558 				advance_left = ADVANCE;
5559 				advance_right = ADVANCE;
5560 			}
5561 		} else if (left_level == right_level) {
5562 			cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5563 			if (cmp < 0) {
5564 				advance_left = ADVANCE;
5565 			} else if (cmp > 0) {
5566 				advance_right = ADVANCE;
5567 			} else {
5568 				left_blockptr = btrfs_node_blockptr(
5569 						left_path->nodes[left_level],
5570 						left_path->slots[left_level]);
5571 				right_blockptr = btrfs_node_blockptr(
5572 						right_path->nodes[right_level],
5573 						right_path->slots[right_level]);
5574 				left_gen = btrfs_node_ptr_generation(
5575 						left_path->nodes[left_level],
5576 						left_path->slots[left_level]);
5577 				right_gen = btrfs_node_ptr_generation(
5578 						right_path->nodes[right_level],
5579 						right_path->slots[right_level]);
5580 				if (left_blockptr == right_blockptr &&
5581 				    left_gen == right_gen) {
5582 					/*
5583 					 * As we're on a shared block, don't
5584 					 * allow to go deeper.
5585 					 */
5586 					advance_left = ADVANCE_ONLY_NEXT;
5587 					advance_right = ADVANCE_ONLY_NEXT;
5588 				} else {
5589 					advance_left = ADVANCE;
5590 					advance_right = ADVANCE;
5591 				}
5592 			}
5593 		} else if (left_level < right_level) {
5594 			advance_right = ADVANCE;
5595 		} else {
5596 			advance_left = ADVANCE;
5597 		}
5598 	}
5599 
5600 out:
5601 	btrfs_free_path(left_path);
5602 	btrfs_free_path(right_path);
5603 	kvfree(tmp_buf);
5604 	return ret;
5605 }
5606 
5607 /*
5608  * this is similar to btrfs_next_leaf, but does not try to preserve
5609  * and fixup the path.  It looks for and returns the next key in the
5610  * tree based on the current path and the min_trans parameters.
5611  *
5612  * 0 is returned if another key is found, < 0 if there are any errors
5613  * and 1 is returned if there are no higher keys in the tree
5614  *
5615  * path->keep_locks should be set to 1 on the search made before
5616  * calling this function.
5617  */
5618 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5619 			struct btrfs_key *key, int level, u64 min_trans)
5620 {
5621 	int slot;
5622 	struct extent_buffer *c;
5623 
5624 	WARN_ON(!path->keep_locks);
5625 	while (level < BTRFS_MAX_LEVEL) {
5626 		if (!path->nodes[level])
5627 			return 1;
5628 
5629 		slot = path->slots[level] + 1;
5630 		c = path->nodes[level];
5631 next:
5632 		if (slot >= btrfs_header_nritems(c)) {
5633 			int ret;
5634 			int orig_lowest;
5635 			struct btrfs_key cur_key;
5636 			if (level + 1 >= BTRFS_MAX_LEVEL ||
5637 			    !path->nodes[level + 1])
5638 				return 1;
5639 
5640 			if (path->locks[level + 1]) {
5641 				level++;
5642 				continue;
5643 			}
5644 
5645 			slot = btrfs_header_nritems(c) - 1;
5646 			if (level == 0)
5647 				btrfs_item_key_to_cpu(c, &cur_key, slot);
5648 			else
5649 				btrfs_node_key_to_cpu(c, &cur_key, slot);
5650 
5651 			orig_lowest = path->lowest_level;
5652 			btrfs_release_path(path);
5653 			path->lowest_level = level;
5654 			ret = btrfs_search_slot(NULL, root, &cur_key, path,
5655 						0, 0);
5656 			path->lowest_level = orig_lowest;
5657 			if (ret < 0)
5658 				return ret;
5659 
5660 			c = path->nodes[level];
5661 			slot = path->slots[level];
5662 			if (ret == 0)
5663 				slot++;
5664 			goto next;
5665 		}
5666 
5667 		if (level == 0)
5668 			btrfs_item_key_to_cpu(c, key, slot);
5669 		else {
5670 			u64 gen = btrfs_node_ptr_generation(c, slot);
5671 
5672 			if (gen < min_trans) {
5673 				slot++;
5674 				goto next;
5675 			}
5676 			btrfs_node_key_to_cpu(c, key, slot);
5677 		}
5678 		return 0;
5679 	}
5680 	return 1;
5681 }
5682 
5683 /*
5684  * search the tree again to find a leaf with greater keys
5685  * returns 0 if it found something or 1 if there are no greater leaves.
5686  * returns < 0 on io errors.
5687  */
5688 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5689 {
5690 	return btrfs_next_old_leaf(root, path, 0);
5691 }
5692 
5693 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5694 			u64 time_seq)
5695 {
5696 	int slot;
5697 	int level;
5698 	struct extent_buffer *c;
5699 	struct extent_buffer *next;
5700 	struct btrfs_key key;
5701 	u32 nritems;
5702 	int ret;
5703 	int old_spinning = path->leave_spinning;
5704 	int next_rw_lock = 0;
5705 
5706 	nritems = btrfs_header_nritems(path->nodes[0]);
5707 	if (nritems == 0)
5708 		return 1;
5709 
5710 	btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5711 again:
5712 	level = 1;
5713 	next = NULL;
5714 	next_rw_lock = 0;
5715 	btrfs_release_path(path);
5716 
5717 	path->keep_locks = 1;
5718 	path->leave_spinning = 1;
5719 
5720 	if (time_seq)
5721 		ret = btrfs_search_old_slot(root, &key, path, time_seq);
5722 	else
5723 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5724 	path->keep_locks = 0;
5725 
5726 	if (ret < 0)
5727 		return ret;
5728 
5729 	nritems = btrfs_header_nritems(path->nodes[0]);
5730 	/*
5731 	 * by releasing the path above we dropped all our locks.  A balance
5732 	 * could have added more items next to the key that used to be
5733 	 * at the very end of the block.  So, check again here and
5734 	 * advance the path if there are now more items available.
5735 	 */
5736 	if (nritems > 0 && path->slots[0] < nritems - 1) {
5737 		if (ret == 0)
5738 			path->slots[0]++;
5739 		ret = 0;
5740 		goto done;
5741 	}
5742 	/*
5743 	 * So the above check misses one case:
5744 	 * - after releasing the path above, someone has removed the item that
5745 	 *   used to be at the very end of the block, and balance between leafs
5746 	 *   gets another one with bigger key.offset to replace it.
5747 	 *
5748 	 * This one should be returned as well, or we can get leaf corruption
5749 	 * later(esp. in __btrfs_drop_extents()).
5750 	 *
5751 	 * And a bit more explanation about this check,
5752 	 * with ret > 0, the key isn't found, the path points to the slot
5753 	 * where it should be inserted, so the path->slots[0] item must be the
5754 	 * bigger one.
5755 	 */
5756 	if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5757 		ret = 0;
5758 		goto done;
5759 	}
5760 
5761 	while (level < BTRFS_MAX_LEVEL) {
5762 		if (!path->nodes[level]) {
5763 			ret = 1;
5764 			goto done;
5765 		}
5766 
5767 		slot = path->slots[level] + 1;
5768 		c = path->nodes[level];
5769 		if (slot >= btrfs_header_nritems(c)) {
5770 			level++;
5771 			if (level == BTRFS_MAX_LEVEL) {
5772 				ret = 1;
5773 				goto done;
5774 			}
5775 			continue;
5776 		}
5777 
5778 		if (next) {
5779 			btrfs_tree_unlock_rw(next, next_rw_lock);
5780 			free_extent_buffer(next);
5781 		}
5782 
5783 		next = c;
5784 		next_rw_lock = path->locks[level];
5785 		ret = read_block_for_search(root, path, &next, level,
5786 					    slot, &key);
5787 		if (ret == -EAGAIN)
5788 			goto again;
5789 
5790 		if (ret < 0) {
5791 			btrfs_release_path(path);
5792 			goto done;
5793 		}
5794 
5795 		if (!path->skip_locking) {
5796 			ret = btrfs_try_tree_read_lock(next);
5797 			if (!ret && time_seq) {
5798 				/*
5799 				 * If we don't get the lock, we may be racing
5800 				 * with push_leaf_left, holding that lock while
5801 				 * itself waiting for the leaf we've currently
5802 				 * locked. To solve this situation, we give up
5803 				 * on our lock and cycle.
5804 				 */
5805 				free_extent_buffer(next);
5806 				btrfs_release_path(path);
5807 				cond_resched();
5808 				goto again;
5809 			}
5810 			if (!ret) {
5811 				btrfs_set_path_blocking(path);
5812 				btrfs_tree_read_lock(next);
5813 				btrfs_clear_path_blocking(path, next,
5814 							  BTRFS_READ_LOCK);
5815 			}
5816 			next_rw_lock = BTRFS_READ_LOCK;
5817 		}
5818 		break;
5819 	}
5820 	path->slots[level] = slot;
5821 	while (1) {
5822 		level--;
5823 		c = path->nodes[level];
5824 		if (path->locks[level])
5825 			btrfs_tree_unlock_rw(c, path->locks[level]);
5826 
5827 		free_extent_buffer(c);
5828 		path->nodes[level] = next;
5829 		path->slots[level] = 0;
5830 		if (!path->skip_locking)
5831 			path->locks[level] = next_rw_lock;
5832 		if (!level)
5833 			break;
5834 
5835 		ret = read_block_for_search(root, path, &next, level,
5836 					    0, &key);
5837 		if (ret == -EAGAIN)
5838 			goto again;
5839 
5840 		if (ret < 0) {
5841 			btrfs_release_path(path);
5842 			goto done;
5843 		}
5844 
5845 		if (!path->skip_locking) {
5846 			ret = btrfs_try_tree_read_lock(next);
5847 			if (!ret) {
5848 				btrfs_set_path_blocking(path);
5849 				btrfs_tree_read_lock(next);
5850 				btrfs_clear_path_blocking(path, next,
5851 							  BTRFS_READ_LOCK);
5852 			}
5853 			next_rw_lock = BTRFS_READ_LOCK;
5854 		}
5855 	}
5856 	ret = 0;
5857 done:
5858 	unlock_up(path, 0, 1, 0, NULL);
5859 	path->leave_spinning = old_spinning;
5860 	if (!old_spinning)
5861 		btrfs_set_path_blocking(path);
5862 
5863 	return ret;
5864 }
5865 
5866 /*
5867  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5868  * searching until it gets past min_objectid or finds an item of 'type'
5869  *
5870  * returns 0 if something is found, 1 if nothing was found and < 0 on error
5871  */
5872 int btrfs_previous_item(struct btrfs_root *root,
5873 			struct btrfs_path *path, u64 min_objectid,
5874 			int type)
5875 {
5876 	struct btrfs_key found_key;
5877 	struct extent_buffer *leaf;
5878 	u32 nritems;
5879 	int ret;
5880 
5881 	while (1) {
5882 		if (path->slots[0] == 0) {
5883 			btrfs_set_path_blocking(path);
5884 			ret = btrfs_prev_leaf(root, path);
5885 			if (ret != 0)
5886 				return ret;
5887 		} else {
5888 			path->slots[0]--;
5889 		}
5890 		leaf = path->nodes[0];
5891 		nritems = btrfs_header_nritems(leaf);
5892 		if (nritems == 0)
5893 			return 1;
5894 		if (path->slots[0] == nritems)
5895 			path->slots[0]--;
5896 
5897 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5898 		if (found_key.objectid < min_objectid)
5899 			break;
5900 		if (found_key.type == type)
5901 			return 0;
5902 		if (found_key.objectid == min_objectid &&
5903 		    found_key.type < type)
5904 			break;
5905 	}
5906 	return 1;
5907 }
5908 
5909 /*
5910  * search in extent tree to find a previous Metadata/Data extent item with
5911  * min objecitd.
5912  *
5913  * returns 0 if something is found, 1 if nothing was found and < 0 on error
5914  */
5915 int btrfs_previous_extent_item(struct btrfs_root *root,
5916 			struct btrfs_path *path, u64 min_objectid)
5917 {
5918 	struct btrfs_key found_key;
5919 	struct extent_buffer *leaf;
5920 	u32 nritems;
5921 	int ret;
5922 
5923 	while (1) {
5924 		if (path->slots[0] == 0) {
5925 			btrfs_set_path_blocking(path);
5926 			ret = btrfs_prev_leaf(root, path);
5927 			if (ret != 0)
5928 				return ret;
5929 		} else {
5930 			path->slots[0]--;
5931 		}
5932 		leaf = path->nodes[0];
5933 		nritems = btrfs_header_nritems(leaf);
5934 		if (nritems == 0)
5935 			return 1;
5936 		if (path->slots[0] == nritems)
5937 			path->slots[0]--;
5938 
5939 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5940 		if (found_key.objectid < min_objectid)
5941 			break;
5942 		if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5943 		    found_key.type == BTRFS_METADATA_ITEM_KEY)
5944 			return 0;
5945 		if (found_key.objectid == min_objectid &&
5946 		    found_key.type < BTRFS_EXTENT_ITEM_KEY)
5947 			break;
5948 	}
5949 	return 1;
5950 }
5951