xref: /openbmc/linux/fs/btrfs/ctree.c (revision b6dcefde)
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 "ctree.h"
21 #include "disk-io.h"
22 #include "transaction.h"
23 #include "print-tree.h"
24 #include "locking.h"
25 
26 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
27 		      *root, struct btrfs_path *path, int level);
28 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
29 		      *root, struct btrfs_key *ins_key,
30 		      struct btrfs_path *path, int data_size, int extend);
31 static int push_node_left(struct btrfs_trans_handle *trans,
32 			  struct btrfs_root *root, struct extent_buffer *dst,
33 			  struct extent_buffer *src, int empty);
34 static int balance_node_right(struct btrfs_trans_handle *trans,
35 			      struct btrfs_root *root,
36 			      struct extent_buffer *dst_buf,
37 			      struct extent_buffer *src_buf);
38 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
39 		   struct btrfs_path *path, int level, int slot);
40 static int setup_items_for_insert(struct btrfs_trans_handle *trans,
41 			struct btrfs_root *root, struct btrfs_path *path,
42 			struct btrfs_key *cpu_key, u32 *data_size,
43 			u32 total_data, u32 total_size, int nr);
44 
45 
46 struct btrfs_path *btrfs_alloc_path(void)
47 {
48 	struct btrfs_path *path;
49 	path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
50 	if (path)
51 		path->reada = 1;
52 	return path;
53 }
54 
55 /*
56  * set all locked nodes in the path to blocking locks.  This should
57  * be done before scheduling
58  */
59 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
60 {
61 	int i;
62 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
63 		if (p->nodes[i] && p->locks[i])
64 			btrfs_set_lock_blocking(p->nodes[i]);
65 	}
66 }
67 
68 /*
69  * reset all the locked nodes in the patch to spinning locks.
70  *
71  * held is used to keep lockdep happy, when lockdep is enabled
72  * we set held to a blocking lock before we go around and
73  * retake all the spinlocks in the path.  You can safely use NULL
74  * for held
75  */
76 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
77 					struct extent_buffer *held)
78 {
79 	int i;
80 
81 #ifdef CONFIG_DEBUG_LOCK_ALLOC
82 	/* lockdep really cares that we take all of these spinlocks
83 	 * in the right order.  If any of the locks in the path are not
84 	 * currently blocking, it is going to complain.  So, make really
85 	 * really sure by forcing the path to blocking before we clear
86 	 * the path blocking.
87 	 */
88 	if (held)
89 		btrfs_set_lock_blocking(held);
90 	btrfs_set_path_blocking(p);
91 #endif
92 
93 	for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
94 		if (p->nodes[i] && p->locks[i])
95 			btrfs_clear_lock_blocking(p->nodes[i]);
96 	}
97 
98 #ifdef CONFIG_DEBUG_LOCK_ALLOC
99 	if (held)
100 		btrfs_clear_lock_blocking(held);
101 #endif
102 }
103 
104 /* this also releases the path */
105 void btrfs_free_path(struct btrfs_path *p)
106 {
107 	btrfs_release_path(NULL, p);
108 	kmem_cache_free(btrfs_path_cachep, p);
109 }
110 
111 /*
112  * path release drops references on the extent buffers in the path
113  * and it drops any locks held by this path
114  *
115  * It is safe to call this on paths that no locks or extent buffers held.
116  */
117 noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
118 {
119 	int i;
120 
121 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
122 		p->slots[i] = 0;
123 		if (!p->nodes[i])
124 			continue;
125 		if (p->locks[i]) {
126 			btrfs_tree_unlock(p->nodes[i]);
127 			p->locks[i] = 0;
128 		}
129 		free_extent_buffer(p->nodes[i]);
130 		p->nodes[i] = NULL;
131 	}
132 }
133 
134 /*
135  * safely gets a reference on the root node of a tree.  A lock
136  * is not taken, so a concurrent writer may put a different node
137  * at the root of the tree.  See btrfs_lock_root_node for the
138  * looping required.
139  *
140  * The extent buffer returned by this has a reference taken, so
141  * it won't disappear.  It may stop being the root of the tree
142  * at any time because there are no locks held.
143  */
144 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
145 {
146 	struct extent_buffer *eb;
147 	spin_lock(&root->node_lock);
148 	eb = root->node;
149 	extent_buffer_get(eb);
150 	spin_unlock(&root->node_lock);
151 	return eb;
152 }
153 
154 /* loop around taking references on and locking the root node of the
155  * tree until you end up with a lock on the root.  A locked buffer
156  * is returned, with a reference held.
157  */
158 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
159 {
160 	struct extent_buffer *eb;
161 
162 	while (1) {
163 		eb = btrfs_root_node(root);
164 		btrfs_tree_lock(eb);
165 
166 		spin_lock(&root->node_lock);
167 		if (eb == root->node) {
168 			spin_unlock(&root->node_lock);
169 			break;
170 		}
171 		spin_unlock(&root->node_lock);
172 
173 		btrfs_tree_unlock(eb);
174 		free_extent_buffer(eb);
175 	}
176 	return eb;
177 }
178 
179 /* cowonly root (everything not a reference counted cow subvolume), just get
180  * put onto a simple dirty list.  transaction.c walks this to make sure they
181  * get properly updated on disk.
182  */
183 static void add_root_to_dirty_list(struct btrfs_root *root)
184 {
185 	if (root->track_dirty && list_empty(&root->dirty_list)) {
186 		list_add(&root->dirty_list,
187 			 &root->fs_info->dirty_cowonly_roots);
188 	}
189 }
190 
191 /*
192  * used by snapshot creation to make a copy of a root for a tree with
193  * a given objectid.  The buffer with the new root node is returned in
194  * cow_ret, and this func returns zero on success or a negative error code.
195  */
196 int btrfs_copy_root(struct btrfs_trans_handle *trans,
197 		      struct btrfs_root *root,
198 		      struct extent_buffer *buf,
199 		      struct extent_buffer **cow_ret, u64 new_root_objectid)
200 {
201 	struct extent_buffer *cow;
202 	u32 nritems;
203 	int ret = 0;
204 	int level;
205 	struct btrfs_disk_key disk_key;
206 
207 	WARN_ON(root->ref_cows && trans->transid !=
208 		root->fs_info->running_transaction->transid);
209 	WARN_ON(root->ref_cows && trans->transid != root->last_trans);
210 
211 	level = btrfs_header_level(buf);
212 	nritems = btrfs_header_nritems(buf);
213 	if (level == 0)
214 		btrfs_item_key(buf, &disk_key, 0);
215 	else
216 		btrfs_node_key(buf, &disk_key, 0);
217 
218 	cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
219 				     new_root_objectid, &disk_key, level,
220 				     buf->start, 0);
221 	if (IS_ERR(cow))
222 		return PTR_ERR(cow);
223 
224 	copy_extent_buffer(cow, buf, 0, 0, cow->len);
225 	btrfs_set_header_bytenr(cow, cow->start);
226 	btrfs_set_header_generation(cow, trans->transid);
227 	btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
228 	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
229 				     BTRFS_HEADER_FLAG_RELOC);
230 	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
231 		btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
232 	else
233 		btrfs_set_header_owner(cow, new_root_objectid);
234 
235 	write_extent_buffer(cow, root->fs_info->fsid,
236 			    (unsigned long)btrfs_header_fsid(cow),
237 			    BTRFS_FSID_SIZE);
238 
239 	WARN_ON(btrfs_header_generation(buf) > trans->transid);
240 	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
241 		ret = btrfs_inc_ref(trans, root, cow, 1);
242 	else
243 		ret = btrfs_inc_ref(trans, root, cow, 0);
244 
245 	if (ret)
246 		return ret;
247 
248 	btrfs_mark_buffer_dirty(cow);
249 	*cow_ret = cow;
250 	return 0;
251 }
252 
253 /*
254  * check if the tree block can be shared by multiple trees
255  */
256 int btrfs_block_can_be_shared(struct btrfs_root *root,
257 			      struct extent_buffer *buf)
258 {
259 	/*
260 	 * Tree blocks not in refernece counted trees and tree roots
261 	 * are never shared. If a block was allocated after the last
262 	 * snapshot and the block was not allocated by tree relocation,
263 	 * we know the block is not shared.
264 	 */
265 	if (root->ref_cows &&
266 	    buf != root->node && buf != root->commit_root &&
267 	    (btrfs_header_generation(buf) <=
268 	     btrfs_root_last_snapshot(&root->root_item) ||
269 	     btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
270 		return 1;
271 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
272 	if (root->ref_cows &&
273 	    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
274 		return 1;
275 #endif
276 	return 0;
277 }
278 
279 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
280 				       struct btrfs_root *root,
281 				       struct extent_buffer *buf,
282 				       struct extent_buffer *cow)
283 {
284 	u64 refs;
285 	u64 owner;
286 	u64 flags;
287 	u64 new_flags = 0;
288 	int ret;
289 
290 	/*
291 	 * Backrefs update rules:
292 	 *
293 	 * Always use full backrefs for extent pointers in tree block
294 	 * allocated by tree relocation.
295 	 *
296 	 * If a shared tree block is no longer referenced by its owner
297 	 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
298 	 * use full backrefs for extent pointers in tree block.
299 	 *
300 	 * If a tree block is been relocating
301 	 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
302 	 * use full backrefs for extent pointers in tree block.
303 	 * The reason for this is some operations (such as drop tree)
304 	 * are only allowed for blocks use full backrefs.
305 	 */
306 
307 	if (btrfs_block_can_be_shared(root, buf)) {
308 		ret = btrfs_lookup_extent_info(trans, root, buf->start,
309 					       buf->len, &refs, &flags);
310 		BUG_ON(ret);
311 		BUG_ON(refs == 0);
312 	} else {
313 		refs = 1;
314 		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
315 		    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
316 			flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
317 		else
318 			flags = 0;
319 	}
320 
321 	owner = btrfs_header_owner(buf);
322 	BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
323 	       !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
324 
325 	if (refs > 1) {
326 		if ((owner == root->root_key.objectid ||
327 		     root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
328 		    !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
329 			ret = btrfs_inc_ref(trans, root, buf, 1);
330 			BUG_ON(ret);
331 
332 			if (root->root_key.objectid ==
333 			    BTRFS_TREE_RELOC_OBJECTID) {
334 				ret = btrfs_dec_ref(trans, root, buf, 0);
335 				BUG_ON(ret);
336 				ret = btrfs_inc_ref(trans, root, cow, 1);
337 				BUG_ON(ret);
338 			}
339 			new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
340 		} else {
341 
342 			if (root->root_key.objectid ==
343 			    BTRFS_TREE_RELOC_OBJECTID)
344 				ret = btrfs_inc_ref(trans, root, cow, 1);
345 			else
346 				ret = btrfs_inc_ref(trans, root, cow, 0);
347 			BUG_ON(ret);
348 		}
349 		if (new_flags != 0) {
350 			ret = btrfs_set_disk_extent_flags(trans, root,
351 							  buf->start,
352 							  buf->len,
353 							  new_flags, 0);
354 			BUG_ON(ret);
355 		}
356 	} else {
357 		if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
358 			if (root->root_key.objectid ==
359 			    BTRFS_TREE_RELOC_OBJECTID)
360 				ret = btrfs_inc_ref(trans, root, cow, 1);
361 			else
362 				ret = btrfs_inc_ref(trans, root, cow, 0);
363 			BUG_ON(ret);
364 			ret = btrfs_dec_ref(trans, root, buf, 1);
365 			BUG_ON(ret);
366 		}
367 		clean_tree_block(trans, root, buf);
368 	}
369 	return 0;
370 }
371 
372 /*
373  * does the dirty work in cow of a single block.  The parent block (if
374  * supplied) is updated to point to the new cow copy.  The new buffer is marked
375  * dirty and returned locked.  If you modify the block it needs to be marked
376  * dirty again.
377  *
378  * search_start -- an allocation hint for the new block
379  *
380  * empty_size -- a hint that you plan on doing more cow.  This is the size in
381  * bytes the allocator should try to find free next to the block it returns.
382  * This is just a hint and may be ignored by the allocator.
383  */
384 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
385 			     struct btrfs_root *root,
386 			     struct extent_buffer *buf,
387 			     struct extent_buffer *parent, int parent_slot,
388 			     struct extent_buffer **cow_ret,
389 			     u64 search_start, u64 empty_size)
390 {
391 	struct btrfs_disk_key disk_key;
392 	struct extent_buffer *cow;
393 	int level;
394 	int unlock_orig = 0;
395 	u64 parent_start;
396 
397 	if (*cow_ret == buf)
398 		unlock_orig = 1;
399 
400 	btrfs_assert_tree_locked(buf);
401 
402 	WARN_ON(root->ref_cows && trans->transid !=
403 		root->fs_info->running_transaction->transid);
404 	WARN_ON(root->ref_cows && trans->transid != root->last_trans);
405 
406 	level = btrfs_header_level(buf);
407 
408 	if (level == 0)
409 		btrfs_item_key(buf, &disk_key, 0);
410 	else
411 		btrfs_node_key(buf, &disk_key, 0);
412 
413 	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
414 		if (parent)
415 			parent_start = parent->start;
416 		else
417 			parent_start = 0;
418 	} else
419 		parent_start = 0;
420 
421 	cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
422 				     root->root_key.objectid, &disk_key,
423 				     level, search_start, empty_size);
424 	if (IS_ERR(cow))
425 		return PTR_ERR(cow);
426 
427 	/* cow is set to blocking by btrfs_init_new_buffer */
428 
429 	copy_extent_buffer(cow, buf, 0, 0, cow->len);
430 	btrfs_set_header_bytenr(cow, cow->start);
431 	btrfs_set_header_generation(cow, trans->transid);
432 	btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
433 	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
434 				     BTRFS_HEADER_FLAG_RELOC);
435 	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
436 		btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
437 	else
438 		btrfs_set_header_owner(cow, root->root_key.objectid);
439 
440 	write_extent_buffer(cow, root->fs_info->fsid,
441 			    (unsigned long)btrfs_header_fsid(cow),
442 			    BTRFS_FSID_SIZE);
443 
444 	update_ref_for_cow(trans, root, buf, cow);
445 
446 	if (buf == root->node) {
447 		WARN_ON(parent && parent != buf);
448 		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
449 		    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
450 			parent_start = buf->start;
451 		else
452 			parent_start = 0;
453 
454 		spin_lock(&root->node_lock);
455 		root->node = cow;
456 		extent_buffer_get(cow);
457 		spin_unlock(&root->node_lock);
458 
459 		btrfs_free_tree_block(trans, root, buf->start, buf->len,
460 				parent_start, root->root_key.objectid, level);
461 		free_extent_buffer(buf);
462 		add_root_to_dirty_list(root);
463 	} else {
464 		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
465 			parent_start = parent->start;
466 		else
467 			parent_start = 0;
468 
469 		WARN_ON(trans->transid != btrfs_header_generation(parent));
470 		btrfs_set_node_blockptr(parent, parent_slot,
471 					cow->start);
472 		btrfs_set_node_ptr_generation(parent, parent_slot,
473 					      trans->transid);
474 		btrfs_mark_buffer_dirty(parent);
475 		btrfs_free_tree_block(trans, root, buf->start, buf->len,
476 				parent_start, root->root_key.objectid, level);
477 	}
478 	if (unlock_orig)
479 		btrfs_tree_unlock(buf);
480 	free_extent_buffer(buf);
481 	btrfs_mark_buffer_dirty(cow);
482 	*cow_ret = cow;
483 	return 0;
484 }
485 
486 static inline int should_cow_block(struct btrfs_trans_handle *trans,
487 				   struct btrfs_root *root,
488 				   struct extent_buffer *buf)
489 {
490 	if (btrfs_header_generation(buf) == trans->transid &&
491 	    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
492 	    !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
493 	      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
494 		return 0;
495 	return 1;
496 }
497 
498 /*
499  * cows a single block, see __btrfs_cow_block for the real work.
500  * This version of it has extra checks so that a block isn't cow'd more than
501  * once per transaction, as long as it hasn't been written yet
502  */
503 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
504 		    struct btrfs_root *root, struct extent_buffer *buf,
505 		    struct extent_buffer *parent, int parent_slot,
506 		    struct extent_buffer **cow_ret)
507 {
508 	u64 search_start;
509 	int ret;
510 
511 	if (trans->transaction != root->fs_info->running_transaction) {
512 		printk(KERN_CRIT "trans %llu running %llu\n",
513 		       (unsigned long long)trans->transid,
514 		       (unsigned long long)
515 		       root->fs_info->running_transaction->transid);
516 		WARN_ON(1);
517 	}
518 	if (trans->transid != root->fs_info->generation) {
519 		printk(KERN_CRIT "trans %llu running %llu\n",
520 		       (unsigned long long)trans->transid,
521 		       (unsigned long long)root->fs_info->generation);
522 		WARN_ON(1);
523 	}
524 
525 	if (!should_cow_block(trans, root, buf)) {
526 		*cow_ret = buf;
527 		return 0;
528 	}
529 
530 	search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
531 
532 	if (parent)
533 		btrfs_set_lock_blocking(parent);
534 	btrfs_set_lock_blocking(buf);
535 
536 	ret = __btrfs_cow_block(trans, root, buf, parent,
537 				 parent_slot, cow_ret, search_start, 0);
538 	return ret;
539 }
540 
541 /*
542  * helper function for defrag to decide if two blocks pointed to by a
543  * node are actually close by
544  */
545 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
546 {
547 	if (blocknr < other && other - (blocknr + blocksize) < 32768)
548 		return 1;
549 	if (blocknr > other && blocknr - (other + blocksize) < 32768)
550 		return 1;
551 	return 0;
552 }
553 
554 /*
555  * compare two keys in a memcmp fashion
556  */
557 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
558 {
559 	struct btrfs_key k1;
560 
561 	btrfs_disk_key_to_cpu(&k1, disk);
562 
563 	return btrfs_comp_cpu_keys(&k1, k2);
564 }
565 
566 /*
567  * same as comp_keys only with two btrfs_key's
568  */
569 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
570 {
571 	if (k1->objectid > k2->objectid)
572 		return 1;
573 	if (k1->objectid < k2->objectid)
574 		return -1;
575 	if (k1->type > k2->type)
576 		return 1;
577 	if (k1->type < k2->type)
578 		return -1;
579 	if (k1->offset > k2->offset)
580 		return 1;
581 	if (k1->offset < k2->offset)
582 		return -1;
583 	return 0;
584 }
585 
586 /*
587  * this is used by the defrag code to go through all the
588  * leaves pointed to by a node and reallocate them so that
589  * disk order is close to key order
590  */
591 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
592 		       struct btrfs_root *root, struct extent_buffer *parent,
593 		       int start_slot, int cache_only, u64 *last_ret,
594 		       struct btrfs_key *progress)
595 {
596 	struct extent_buffer *cur;
597 	u64 blocknr;
598 	u64 gen;
599 	u64 search_start = *last_ret;
600 	u64 last_block = 0;
601 	u64 other;
602 	u32 parent_nritems;
603 	int end_slot;
604 	int i;
605 	int err = 0;
606 	int parent_level;
607 	int uptodate;
608 	u32 blocksize;
609 	int progress_passed = 0;
610 	struct btrfs_disk_key disk_key;
611 
612 	parent_level = btrfs_header_level(parent);
613 	if (cache_only && parent_level != 1)
614 		return 0;
615 
616 	if (trans->transaction != root->fs_info->running_transaction)
617 		WARN_ON(1);
618 	if (trans->transid != root->fs_info->generation)
619 		WARN_ON(1);
620 
621 	parent_nritems = btrfs_header_nritems(parent);
622 	blocksize = btrfs_level_size(root, parent_level - 1);
623 	end_slot = parent_nritems;
624 
625 	if (parent_nritems == 1)
626 		return 0;
627 
628 	btrfs_set_lock_blocking(parent);
629 
630 	for (i = start_slot; i < end_slot; i++) {
631 		int close = 1;
632 
633 		if (!parent->map_token) {
634 			map_extent_buffer(parent,
635 					btrfs_node_key_ptr_offset(i),
636 					sizeof(struct btrfs_key_ptr),
637 					&parent->map_token, &parent->kaddr,
638 					&parent->map_start, &parent->map_len,
639 					KM_USER1);
640 		}
641 		btrfs_node_key(parent, &disk_key, i);
642 		if (!progress_passed && comp_keys(&disk_key, progress) < 0)
643 			continue;
644 
645 		progress_passed = 1;
646 		blocknr = btrfs_node_blockptr(parent, i);
647 		gen = btrfs_node_ptr_generation(parent, i);
648 		if (last_block == 0)
649 			last_block = blocknr;
650 
651 		if (i > 0) {
652 			other = btrfs_node_blockptr(parent, i - 1);
653 			close = close_blocks(blocknr, other, blocksize);
654 		}
655 		if (!close && i < end_slot - 2) {
656 			other = btrfs_node_blockptr(parent, i + 1);
657 			close = close_blocks(blocknr, other, blocksize);
658 		}
659 		if (close) {
660 			last_block = blocknr;
661 			continue;
662 		}
663 		if (parent->map_token) {
664 			unmap_extent_buffer(parent, parent->map_token,
665 					    KM_USER1);
666 			parent->map_token = NULL;
667 		}
668 
669 		cur = btrfs_find_tree_block(root, blocknr, blocksize);
670 		if (cur)
671 			uptodate = btrfs_buffer_uptodate(cur, gen);
672 		else
673 			uptodate = 0;
674 		if (!cur || !uptodate) {
675 			if (cache_only) {
676 				free_extent_buffer(cur);
677 				continue;
678 			}
679 			if (!cur) {
680 				cur = read_tree_block(root, blocknr,
681 							 blocksize, gen);
682 			} else if (!uptodate) {
683 				btrfs_read_buffer(cur, gen);
684 			}
685 		}
686 		if (search_start == 0)
687 			search_start = last_block;
688 
689 		btrfs_tree_lock(cur);
690 		btrfs_set_lock_blocking(cur);
691 		err = __btrfs_cow_block(trans, root, cur, parent, i,
692 					&cur, search_start,
693 					min(16 * blocksize,
694 					    (end_slot - i) * blocksize));
695 		if (err) {
696 			btrfs_tree_unlock(cur);
697 			free_extent_buffer(cur);
698 			break;
699 		}
700 		search_start = cur->start;
701 		last_block = cur->start;
702 		*last_ret = search_start;
703 		btrfs_tree_unlock(cur);
704 		free_extent_buffer(cur);
705 	}
706 	if (parent->map_token) {
707 		unmap_extent_buffer(parent, parent->map_token,
708 				    KM_USER1);
709 		parent->map_token = NULL;
710 	}
711 	return err;
712 }
713 
714 /*
715  * The leaf data grows from end-to-front in the node.
716  * this returns the address of the start of the last item,
717  * which is the stop of the leaf data stack
718  */
719 static inline unsigned int leaf_data_end(struct btrfs_root *root,
720 					 struct extent_buffer *leaf)
721 {
722 	u32 nr = btrfs_header_nritems(leaf);
723 	if (nr == 0)
724 		return BTRFS_LEAF_DATA_SIZE(root);
725 	return btrfs_item_offset_nr(leaf, nr - 1);
726 }
727 
728 /*
729  * extra debugging checks to make sure all the items in a key are
730  * well formed and in the proper order
731  */
732 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
733 		      int level)
734 {
735 	struct extent_buffer *parent = NULL;
736 	struct extent_buffer *node = path->nodes[level];
737 	struct btrfs_disk_key parent_key;
738 	struct btrfs_disk_key node_key;
739 	int parent_slot;
740 	int slot;
741 	struct btrfs_key cpukey;
742 	u32 nritems = btrfs_header_nritems(node);
743 
744 	if (path->nodes[level + 1])
745 		parent = path->nodes[level + 1];
746 
747 	slot = path->slots[level];
748 	BUG_ON(nritems == 0);
749 	if (parent) {
750 		parent_slot = path->slots[level + 1];
751 		btrfs_node_key(parent, &parent_key, parent_slot);
752 		btrfs_node_key(node, &node_key, 0);
753 		BUG_ON(memcmp(&parent_key, &node_key,
754 			      sizeof(struct btrfs_disk_key)));
755 		BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
756 		       btrfs_header_bytenr(node));
757 	}
758 	BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
759 	if (slot != 0) {
760 		btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
761 		btrfs_node_key(node, &node_key, slot);
762 		BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
763 	}
764 	if (slot < nritems - 1) {
765 		btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
766 		btrfs_node_key(node, &node_key, slot);
767 		BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
768 	}
769 	return 0;
770 }
771 
772 /*
773  * extra checking to make sure all the items in a leaf are
774  * well formed and in the proper order
775  */
776 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
777 		      int level)
778 {
779 	struct extent_buffer *leaf = path->nodes[level];
780 	struct extent_buffer *parent = NULL;
781 	int parent_slot;
782 	struct btrfs_key cpukey;
783 	struct btrfs_disk_key parent_key;
784 	struct btrfs_disk_key leaf_key;
785 	int slot = path->slots[0];
786 
787 	u32 nritems = btrfs_header_nritems(leaf);
788 
789 	if (path->nodes[level + 1])
790 		parent = path->nodes[level + 1];
791 
792 	if (nritems == 0)
793 		return 0;
794 
795 	if (parent) {
796 		parent_slot = path->slots[level + 1];
797 		btrfs_node_key(parent, &parent_key, parent_slot);
798 		btrfs_item_key(leaf, &leaf_key, 0);
799 
800 		BUG_ON(memcmp(&parent_key, &leaf_key,
801 		       sizeof(struct btrfs_disk_key)));
802 		BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
803 		       btrfs_header_bytenr(leaf));
804 	}
805 	if (slot != 0 && slot < nritems - 1) {
806 		btrfs_item_key(leaf, &leaf_key, slot);
807 		btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
808 		if (comp_keys(&leaf_key, &cpukey) <= 0) {
809 			btrfs_print_leaf(root, leaf);
810 			printk(KERN_CRIT "slot %d offset bad key\n", slot);
811 			BUG_ON(1);
812 		}
813 		if (btrfs_item_offset_nr(leaf, slot - 1) !=
814 		       btrfs_item_end_nr(leaf, slot)) {
815 			btrfs_print_leaf(root, leaf);
816 			printk(KERN_CRIT "slot %d offset bad\n", slot);
817 			BUG_ON(1);
818 		}
819 	}
820 	if (slot < nritems - 1) {
821 		btrfs_item_key(leaf, &leaf_key, slot);
822 		btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
823 		BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
824 		if (btrfs_item_offset_nr(leaf, slot) !=
825 			btrfs_item_end_nr(leaf, slot + 1)) {
826 			btrfs_print_leaf(root, leaf);
827 			printk(KERN_CRIT "slot %d offset bad\n", slot);
828 			BUG_ON(1);
829 		}
830 	}
831 	BUG_ON(btrfs_item_offset_nr(leaf, 0) +
832 	       btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
833 	return 0;
834 }
835 
836 static noinline int check_block(struct btrfs_root *root,
837 				struct btrfs_path *path, int level)
838 {
839 	return 0;
840 	if (level == 0)
841 		return check_leaf(root, path, level);
842 	return check_node(root, path, level);
843 }
844 
845 /*
846  * search for key in the extent_buffer.  The items start at offset p,
847  * and they are item_size apart.  There are 'max' items in p.
848  *
849  * the slot in the array is returned via slot, and it points to
850  * the place where you would insert key if it is not found in
851  * the array.
852  *
853  * slot may point to max if the key is bigger than all of the keys
854  */
855 static noinline int generic_bin_search(struct extent_buffer *eb,
856 				       unsigned long p,
857 				       int item_size, struct btrfs_key *key,
858 				       int max, int *slot)
859 {
860 	int low = 0;
861 	int high = max;
862 	int mid;
863 	int ret;
864 	struct btrfs_disk_key *tmp = NULL;
865 	struct btrfs_disk_key unaligned;
866 	unsigned long offset;
867 	char *map_token = NULL;
868 	char *kaddr = NULL;
869 	unsigned long map_start = 0;
870 	unsigned long map_len = 0;
871 	int err;
872 
873 	while (low < high) {
874 		mid = (low + high) / 2;
875 		offset = p + mid * item_size;
876 
877 		if (!map_token || offset < map_start ||
878 		    (offset + sizeof(struct btrfs_disk_key)) >
879 		    map_start + map_len) {
880 			if (map_token) {
881 				unmap_extent_buffer(eb, map_token, KM_USER0);
882 				map_token = NULL;
883 			}
884 
885 			err = map_private_extent_buffer(eb, offset,
886 						sizeof(struct btrfs_disk_key),
887 						&map_token, &kaddr,
888 						&map_start, &map_len, KM_USER0);
889 
890 			if (!err) {
891 				tmp = (struct btrfs_disk_key *)(kaddr + offset -
892 							map_start);
893 			} else {
894 				read_extent_buffer(eb, &unaligned,
895 						   offset, sizeof(unaligned));
896 				tmp = &unaligned;
897 			}
898 
899 		} else {
900 			tmp = (struct btrfs_disk_key *)(kaddr + offset -
901 							map_start);
902 		}
903 		ret = comp_keys(tmp, key);
904 
905 		if (ret < 0)
906 			low = mid + 1;
907 		else if (ret > 0)
908 			high = mid;
909 		else {
910 			*slot = mid;
911 			if (map_token)
912 				unmap_extent_buffer(eb, map_token, KM_USER0);
913 			return 0;
914 		}
915 	}
916 	*slot = low;
917 	if (map_token)
918 		unmap_extent_buffer(eb, map_token, KM_USER0);
919 	return 1;
920 }
921 
922 /*
923  * simple bin_search frontend that does the right thing for
924  * leaves vs nodes
925  */
926 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
927 		      int level, int *slot)
928 {
929 	if (level == 0) {
930 		return generic_bin_search(eb,
931 					  offsetof(struct btrfs_leaf, items),
932 					  sizeof(struct btrfs_item),
933 					  key, btrfs_header_nritems(eb),
934 					  slot);
935 	} else {
936 		return generic_bin_search(eb,
937 					  offsetof(struct btrfs_node, ptrs),
938 					  sizeof(struct btrfs_key_ptr),
939 					  key, btrfs_header_nritems(eb),
940 					  slot);
941 	}
942 	return -1;
943 }
944 
945 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
946 		     int level, int *slot)
947 {
948 	return bin_search(eb, key, level, slot);
949 }
950 
951 /* given a node and slot number, this reads the blocks it points to.  The
952  * extent buffer is returned with a reference taken (but unlocked).
953  * NULL is returned on error.
954  */
955 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
956 				   struct extent_buffer *parent, int slot)
957 {
958 	int level = btrfs_header_level(parent);
959 	if (slot < 0)
960 		return NULL;
961 	if (slot >= btrfs_header_nritems(parent))
962 		return NULL;
963 
964 	BUG_ON(level == 0);
965 
966 	return read_tree_block(root, btrfs_node_blockptr(parent, slot),
967 		       btrfs_level_size(root, level - 1),
968 		       btrfs_node_ptr_generation(parent, slot));
969 }
970 
971 /*
972  * node level balancing, used to make sure nodes are in proper order for
973  * item deletion.  We balance from the top down, so we have to make sure
974  * that a deletion won't leave an node completely empty later on.
975  */
976 static noinline int balance_level(struct btrfs_trans_handle *trans,
977 			 struct btrfs_root *root,
978 			 struct btrfs_path *path, int level)
979 {
980 	struct extent_buffer *right = NULL;
981 	struct extent_buffer *mid;
982 	struct extent_buffer *left = NULL;
983 	struct extent_buffer *parent = NULL;
984 	int ret = 0;
985 	int wret;
986 	int pslot;
987 	int orig_slot = path->slots[level];
988 	int err_on_enospc = 0;
989 	u64 orig_ptr;
990 
991 	if (level == 0)
992 		return 0;
993 
994 	mid = path->nodes[level];
995 
996 	WARN_ON(!path->locks[level]);
997 	WARN_ON(btrfs_header_generation(mid) != trans->transid);
998 
999 	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1000 
1001 	if (level < BTRFS_MAX_LEVEL - 1)
1002 		parent = path->nodes[level + 1];
1003 	pslot = path->slots[level + 1];
1004 
1005 	/*
1006 	 * deal with the case where there is only one pointer in the root
1007 	 * by promoting the node below to a root
1008 	 */
1009 	if (!parent) {
1010 		struct extent_buffer *child;
1011 
1012 		if (btrfs_header_nritems(mid) != 1)
1013 			return 0;
1014 
1015 		/* promote the child to a root */
1016 		child = read_node_slot(root, mid, 0);
1017 		BUG_ON(!child);
1018 		btrfs_tree_lock(child);
1019 		btrfs_set_lock_blocking(child);
1020 		ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1021 		BUG_ON(ret);
1022 
1023 		spin_lock(&root->node_lock);
1024 		root->node = child;
1025 		spin_unlock(&root->node_lock);
1026 
1027 		add_root_to_dirty_list(root);
1028 		btrfs_tree_unlock(child);
1029 
1030 		path->locks[level] = 0;
1031 		path->nodes[level] = NULL;
1032 		clean_tree_block(trans, root, mid);
1033 		btrfs_tree_unlock(mid);
1034 		/* once for the path */
1035 		free_extent_buffer(mid);
1036 		ret = btrfs_free_tree_block(trans, root, mid->start, mid->len,
1037 					    0, root->root_key.objectid, level);
1038 		/* once for the root ptr */
1039 		free_extent_buffer(mid);
1040 		return ret;
1041 	}
1042 	if (btrfs_header_nritems(mid) >
1043 	    BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1044 		return 0;
1045 
1046 	if (btrfs_header_nritems(mid) < 2)
1047 		err_on_enospc = 1;
1048 
1049 	left = read_node_slot(root, parent, pslot - 1);
1050 	if (left) {
1051 		btrfs_tree_lock(left);
1052 		btrfs_set_lock_blocking(left);
1053 		wret = btrfs_cow_block(trans, root, left,
1054 				       parent, pslot - 1, &left);
1055 		if (wret) {
1056 			ret = wret;
1057 			goto enospc;
1058 		}
1059 	}
1060 	right = read_node_slot(root, parent, pslot + 1);
1061 	if (right) {
1062 		btrfs_tree_lock(right);
1063 		btrfs_set_lock_blocking(right);
1064 		wret = btrfs_cow_block(trans, root, right,
1065 				       parent, pslot + 1, &right);
1066 		if (wret) {
1067 			ret = wret;
1068 			goto enospc;
1069 		}
1070 	}
1071 
1072 	/* first, try to make some room in the middle buffer */
1073 	if (left) {
1074 		orig_slot += btrfs_header_nritems(left);
1075 		wret = push_node_left(trans, root, left, mid, 1);
1076 		if (wret < 0)
1077 			ret = wret;
1078 		if (btrfs_header_nritems(mid) < 2)
1079 			err_on_enospc = 1;
1080 	}
1081 
1082 	/*
1083 	 * then try to empty the right most buffer into the middle
1084 	 */
1085 	if (right) {
1086 		wret = push_node_left(trans, root, mid, right, 1);
1087 		if (wret < 0 && wret != -ENOSPC)
1088 			ret = wret;
1089 		if (btrfs_header_nritems(right) == 0) {
1090 			u64 bytenr = right->start;
1091 			u32 blocksize = right->len;
1092 
1093 			clean_tree_block(trans, root, right);
1094 			btrfs_tree_unlock(right);
1095 			free_extent_buffer(right);
1096 			right = NULL;
1097 			wret = del_ptr(trans, root, path, level + 1, pslot +
1098 				       1);
1099 			if (wret)
1100 				ret = wret;
1101 			wret = btrfs_free_tree_block(trans, root,
1102 						     bytenr, blocksize, 0,
1103 						     root->root_key.objectid,
1104 						     level);
1105 			if (wret)
1106 				ret = wret;
1107 		} else {
1108 			struct btrfs_disk_key right_key;
1109 			btrfs_node_key(right, &right_key, 0);
1110 			btrfs_set_node_key(parent, &right_key, pslot + 1);
1111 			btrfs_mark_buffer_dirty(parent);
1112 		}
1113 	}
1114 	if (btrfs_header_nritems(mid) == 1) {
1115 		/*
1116 		 * we're not allowed to leave a node with one item in the
1117 		 * tree during a delete.  A deletion from lower in the tree
1118 		 * could try to delete the only pointer in this node.
1119 		 * So, pull some keys from the left.
1120 		 * There has to be a left pointer at this point because
1121 		 * otherwise we would have pulled some pointers from the
1122 		 * right
1123 		 */
1124 		BUG_ON(!left);
1125 		wret = balance_node_right(trans, root, mid, left);
1126 		if (wret < 0) {
1127 			ret = wret;
1128 			goto enospc;
1129 		}
1130 		if (wret == 1) {
1131 			wret = push_node_left(trans, root, left, mid, 1);
1132 			if (wret < 0)
1133 				ret = wret;
1134 		}
1135 		BUG_ON(wret == 1);
1136 	}
1137 	if (btrfs_header_nritems(mid) == 0) {
1138 		/* we've managed to empty the middle node, drop it */
1139 		u64 bytenr = mid->start;
1140 		u32 blocksize = mid->len;
1141 
1142 		clean_tree_block(trans, root, mid);
1143 		btrfs_tree_unlock(mid);
1144 		free_extent_buffer(mid);
1145 		mid = NULL;
1146 		wret = del_ptr(trans, root, path, level + 1, pslot);
1147 		if (wret)
1148 			ret = wret;
1149 		wret = btrfs_free_tree_block(trans, root, bytenr, blocksize,
1150 					 0, root->root_key.objectid, level);
1151 		if (wret)
1152 			ret = wret;
1153 	} else {
1154 		/* update the parent key to reflect our changes */
1155 		struct btrfs_disk_key mid_key;
1156 		btrfs_node_key(mid, &mid_key, 0);
1157 		btrfs_set_node_key(parent, &mid_key, pslot);
1158 		btrfs_mark_buffer_dirty(parent);
1159 	}
1160 
1161 	/* update the path */
1162 	if (left) {
1163 		if (btrfs_header_nritems(left) > orig_slot) {
1164 			extent_buffer_get(left);
1165 			/* left was locked after cow */
1166 			path->nodes[level] = left;
1167 			path->slots[level + 1] -= 1;
1168 			path->slots[level] = orig_slot;
1169 			if (mid) {
1170 				btrfs_tree_unlock(mid);
1171 				free_extent_buffer(mid);
1172 			}
1173 		} else {
1174 			orig_slot -= btrfs_header_nritems(left);
1175 			path->slots[level] = orig_slot;
1176 		}
1177 	}
1178 	/* double check we haven't messed things up */
1179 	check_block(root, path, level);
1180 	if (orig_ptr !=
1181 	    btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1182 		BUG();
1183 enospc:
1184 	if (right) {
1185 		btrfs_tree_unlock(right);
1186 		free_extent_buffer(right);
1187 	}
1188 	if (left) {
1189 		if (path->nodes[level] != left)
1190 			btrfs_tree_unlock(left);
1191 		free_extent_buffer(left);
1192 	}
1193 	return ret;
1194 }
1195 
1196 /* Node balancing for insertion.  Here we only split or push nodes around
1197  * when they are completely full.  This is also done top down, so we
1198  * have to be pessimistic.
1199  */
1200 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1201 					  struct btrfs_root *root,
1202 					  struct btrfs_path *path, int level)
1203 {
1204 	struct extent_buffer *right = NULL;
1205 	struct extent_buffer *mid;
1206 	struct extent_buffer *left = NULL;
1207 	struct extent_buffer *parent = NULL;
1208 	int ret = 0;
1209 	int wret;
1210 	int pslot;
1211 	int orig_slot = path->slots[level];
1212 	u64 orig_ptr;
1213 
1214 	if (level == 0)
1215 		return 1;
1216 
1217 	mid = path->nodes[level];
1218 	WARN_ON(btrfs_header_generation(mid) != trans->transid);
1219 	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1220 
1221 	if (level < BTRFS_MAX_LEVEL - 1)
1222 		parent = path->nodes[level + 1];
1223 	pslot = path->slots[level + 1];
1224 
1225 	if (!parent)
1226 		return 1;
1227 
1228 	left = read_node_slot(root, parent, pslot - 1);
1229 
1230 	/* first, try to make some room in the middle buffer */
1231 	if (left) {
1232 		u32 left_nr;
1233 
1234 		btrfs_tree_lock(left);
1235 		btrfs_set_lock_blocking(left);
1236 
1237 		left_nr = btrfs_header_nritems(left);
1238 		if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1239 			wret = 1;
1240 		} else {
1241 			ret = btrfs_cow_block(trans, root, left, parent,
1242 					      pslot - 1, &left);
1243 			if (ret)
1244 				wret = 1;
1245 			else {
1246 				wret = push_node_left(trans, root,
1247 						      left, mid, 0);
1248 			}
1249 		}
1250 		if (wret < 0)
1251 			ret = wret;
1252 		if (wret == 0) {
1253 			struct btrfs_disk_key disk_key;
1254 			orig_slot += left_nr;
1255 			btrfs_node_key(mid, &disk_key, 0);
1256 			btrfs_set_node_key(parent, &disk_key, pslot);
1257 			btrfs_mark_buffer_dirty(parent);
1258 			if (btrfs_header_nritems(left) > orig_slot) {
1259 				path->nodes[level] = left;
1260 				path->slots[level + 1] -= 1;
1261 				path->slots[level] = orig_slot;
1262 				btrfs_tree_unlock(mid);
1263 				free_extent_buffer(mid);
1264 			} else {
1265 				orig_slot -=
1266 					btrfs_header_nritems(left);
1267 				path->slots[level] = orig_slot;
1268 				btrfs_tree_unlock(left);
1269 				free_extent_buffer(left);
1270 			}
1271 			return 0;
1272 		}
1273 		btrfs_tree_unlock(left);
1274 		free_extent_buffer(left);
1275 	}
1276 	right = read_node_slot(root, parent, pslot + 1);
1277 
1278 	/*
1279 	 * then try to empty the right most buffer into the middle
1280 	 */
1281 	if (right) {
1282 		u32 right_nr;
1283 
1284 		btrfs_tree_lock(right);
1285 		btrfs_set_lock_blocking(right);
1286 
1287 		right_nr = btrfs_header_nritems(right);
1288 		if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1289 			wret = 1;
1290 		} else {
1291 			ret = btrfs_cow_block(trans, root, right,
1292 					      parent, pslot + 1,
1293 					      &right);
1294 			if (ret)
1295 				wret = 1;
1296 			else {
1297 				wret = balance_node_right(trans, root,
1298 							  right, mid);
1299 			}
1300 		}
1301 		if (wret < 0)
1302 			ret = wret;
1303 		if (wret == 0) {
1304 			struct btrfs_disk_key disk_key;
1305 
1306 			btrfs_node_key(right, &disk_key, 0);
1307 			btrfs_set_node_key(parent, &disk_key, pslot + 1);
1308 			btrfs_mark_buffer_dirty(parent);
1309 
1310 			if (btrfs_header_nritems(mid) <= orig_slot) {
1311 				path->nodes[level] = right;
1312 				path->slots[level + 1] += 1;
1313 				path->slots[level] = orig_slot -
1314 					btrfs_header_nritems(mid);
1315 				btrfs_tree_unlock(mid);
1316 				free_extent_buffer(mid);
1317 			} else {
1318 				btrfs_tree_unlock(right);
1319 				free_extent_buffer(right);
1320 			}
1321 			return 0;
1322 		}
1323 		btrfs_tree_unlock(right);
1324 		free_extent_buffer(right);
1325 	}
1326 	return 1;
1327 }
1328 
1329 /*
1330  * readahead one full node of leaves, finding things that are close
1331  * to the block in 'slot', and triggering ra on them.
1332  */
1333 static void reada_for_search(struct btrfs_root *root,
1334 			     struct btrfs_path *path,
1335 			     int level, int slot, u64 objectid)
1336 {
1337 	struct extent_buffer *node;
1338 	struct btrfs_disk_key disk_key;
1339 	u32 nritems;
1340 	u64 search;
1341 	u64 target;
1342 	u64 nread = 0;
1343 	int direction = path->reada;
1344 	struct extent_buffer *eb;
1345 	u32 nr;
1346 	u32 blocksize;
1347 	u32 nscan = 0;
1348 
1349 	if (level != 1)
1350 		return;
1351 
1352 	if (!path->nodes[level])
1353 		return;
1354 
1355 	node = path->nodes[level];
1356 
1357 	search = btrfs_node_blockptr(node, slot);
1358 	blocksize = btrfs_level_size(root, level - 1);
1359 	eb = btrfs_find_tree_block(root, search, blocksize);
1360 	if (eb) {
1361 		free_extent_buffer(eb);
1362 		return;
1363 	}
1364 
1365 	target = search;
1366 
1367 	nritems = btrfs_header_nritems(node);
1368 	nr = slot;
1369 	while (1) {
1370 		if (direction < 0) {
1371 			if (nr == 0)
1372 				break;
1373 			nr--;
1374 		} else if (direction > 0) {
1375 			nr++;
1376 			if (nr >= nritems)
1377 				break;
1378 		}
1379 		if (path->reada < 0 && objectid) {
1380 			btrfs_node_key(node, &disk_key, nr);
1381 			if (btrfs_disk_key_objectid(&disk_key) != objectid)
1382 				break;
1383 		}
1384 		search = btrfs_node_blockptr(node, nr);
1385 		if ((search <= target && target - search <= 65536) ||
1386 		    (search > target && search - target <= 65536)) {
1387 			readahead_tree_block(root, search, blocksize,
1388 				     btrfs_node_ptr_generation(node, nr));
1389 			nread += blocksize;
1390 		}
1391 		nscan++;
1392 		if ((nread > 65536 || nscan > 32))
1393 			break;
1394 	}
1395 }
1396 
1397 /*
1398  * returns -EAGAIN if it had to drop the path, or zero if everything was in
1399  * cache
1400  */
1401 static noinline int reada_for_balance(struct btrfs_root *root,
1402 				      struct btrfs_path *path, int level)
1403 {
1404 	int slot;
1405 	int nritems;
1406 	struct extent_buffer *parent;
1407 	struct extent_buffer *eb;
1408 	u64 gen;
1409 	u64 block1 = 0;
1410 	u64 block2 = 0;
1411 	int ret = 0;
1412 	int blocksize;
1413 
1414 	parent = path->nodes[level + 1];
1415 	if (!parent)
1416 		return 0;
1417 
1418 	nritems = btrfs_header_nritems(parent);
1419 	slot = path->slots[level + 1];
1420 	blocksize = btrfs_level_size(root, level);
1421 
1422 	if (slot > 0) {
1423 		block1 = btrfs_node_blockptr(parent, slot - 1);
1424 		gen = btrfs_node_ptr_generation(parent, slot - 1);
1425 		eb = btrfs_find_tree_block(root, block1, blocksize);
1426 		if (eb && btrfs_buffer_uptodate(eb, gen))
1427 			block1 = 0;
1428 		free_extent_buffer(eb);
1429 	}
1430 	if (slot + 1 < nritems) {
1431 		block2 = btrfs_node_blockptr(parent, slot + 1);
1432 		gen = btrfs_node_ptr_generation(parent, slot + 1);
1433 		eb = btrfs_find_tree_block(root, block2, blocksize);
1434 		if (eb && btrfs_buffer_uptodate(eb, gen))
1435 			block2 = 0;
1436 		free_extent_buffer(eb);
1437 	}
1438 	if (block1 || block2) {
1439 		ret = -EAGAIN;
1440 
1441 		/* release the whole path */
1442 		btrfs_release_path(root, path);
1443 
1444 		/* read the blocks */
1445 		if (block1)
1446 			readahead_tree_block(root, block1, blocksize, 0);
1447 		if (block2)
1448 			readahead_tree_block(root, block2, blocksize, 0);
1449 
1450 		if (block1) {
1451 			eb = read_tree_block(root, block1, blocksize, 0);
1452 			free_extent_buffer(eb);
1453 		}
1454 		if (block2) {
1455 			eb = read_tree_block(root, block2, blocksize, 0);
1456 			free_extent_buffer(eb);
1457 		}
1458 	}
1459 	return ret;
1460 }
1461 
1462 
1463 /*
1464  * when we walk down the tree, it is usually safe to unlock the higher layers
1465  * in the tree.  The exceptions are when our path goes through slot 0, because
1466  * operations on the tree might require changing key pointers higher up in the
1467  * tree.
1468  *
1469  * callers might also have set path->keep_locks, which tells this code to keep
1470  * the lock if the path points to the last slot in the block.  This is part of
1471  * walking through the tree, and selecting the next slot in the higher block.
1472  *
1473  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
1474  * if lowest_unlock is 1, level 0 won't be unlocked
1475  */
1476 static noinline void unlock_up(struct btrfs_path *path, int level,
1477 			       int lowest_unlock)
1478 {
1479 	int i;
1480 	int skip_level = level;
1481 	int no_skips = 0;
1482 	struct extent_buffer *t;
1483 
1484 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1485 		if (!path->nodes[i])
1486 			break;
1487 		if (!path->locks[i])
1488 			break;
1489 		if (!no_skips && path->slots[i] == 0) {
1490 			skip_level = i + 1;
1491 			continue;
1492 		}
1493 		if (!no_skips && path->keep_locks) {
1494 			u32 nritems;
1495 			t = path->nodes[i];
1496 			nritems = btrfs_header_nritems(t);
1497 			if (nritems < 1 || path->slots[i] >= nritems - 1) {
1498 				skip_level = i + 1;
1499 				continue;
1500 			}
1501 		}
1502 		if (skip_level < i && i >= lowest_unlock)
1503 			no_skips = 1;
1504 
1505 		t = path->nodes[i];
1506 		if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1507 			btrfs_tree_unlock(t);
1508 			path->locks[i] = 0;
1509 		}
1510 	}
1511 }
1512 
1513 /*
1514  * This releases any locks held in the path starting at level and
1515  * going all the way up to the root.
1516  *
1517  * btrfs_search_slot will keep the lock held on higher nodes in a few
1518  * corner cases, such as COW of the block at slot zero in the node.  This
1519  * ignores those rules, and it should only be called when there are no
1520  * more updates to be done higher up in the tree.
1521  */
1522 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1523 {
1524 	int i;
1525 
1526 	if (path->keep_locks)
1527 		return;
1528 
1529 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1530 		if (!path->nodes[i])
1531 			continue;
1532 		if (!path->locks[i])
1533 			continue;
1534 		btrfs_tree_unlock(path->nodes[i]);
1535 		path->locks[i] = 0;
1536 	}
1537 }
1538 
1539 /*
1540  * helper function for btrfs_search_slot.  The goal is to find a block
1541  * in cache without setting the path to blocking.  If we find the block
1542  * we return zero and the path is unchanged.
1543  *
1544  * If we can't find the block, we set the path blocking and do some
1545  * reada.  -EAGAIN is returned and the search must be repeated.
1546  */
1547 static int
1548 read_block_for_search(struct btrfs_trans_handle *trans,
1549 		       struct btrfs_root *root, struct btrfs_path *p,
1550 		       struct extent_buffer **eb_ret, int level, int slot,
1551 		       struct btrfs_key *key)
1552 {
1553 	u64 blocknr;
1554 	u64 gen;
1555 	u32 blocksize;
1556 	struct extent_buffer *b = *eb_ret;
1557 	struct extent_buffer *tmp;
1558 	int ret;
1559 
1560 	blocknr = btrfs_node_blockptr(b, slot);
1561 	gen = btrfs_node_ptr_generation(b, slot);
1562 	blocksize = btrfs_level_size(root, level - 1);
1563 
1564 	tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1565 	if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1566 		/*
1567 		 * we found an up to date block without sleeping, return
1568 		 * right away
1569 		 */
1570 		*eb_ret = tmp;
1571 		return 0;
1572 	}
1573 
1574 	/*
1575 	 * reduce lock contention at high levels
1576 	 * of the btree by dropping locks before
1577 	 * we read.  Don't release the lock on the current
1578 	 * level because we need to walk this node to figure
1579 	 * out which blocks to read.
1580 	 */
1581 	btrfs_unlock_up_safe(p, level + 1);
1582 	btrfs_set_path_blocking(p);
1583 
1584 	if (tmp)
1585 		free_extent_buffer(tmp);
1586 	if (p->reada)
1587 		reada_for_search(root, p, level, slot, key->objectid);
1588 
1589 	btrfs_release_path(NULL, p);
1590 
1591 	ret = -EAGAIN;
1592 	tmp = read_tree_block(root, blocknr, blocksize, gen);
1593 	if (tmp) {
1594 		/*
1595 		 * If the read above didn't mark this buffer up to date,
1596 		 * it will never end up being up to date.  Set ret to EIO now
1597 		 * and give up so that our caller doesn't loop forever
1598 		 * on our EAGAINs.
1599 		 */
1600 		if (!btrfs_buffer_uptodate(tmp, 0))
1601 			ret = -EIO;
1602 		free_extent_buffer(tmp);
1603 	}
1604 	return ret;
1605 }
1606 
1607 /*
1608  * helper function for btrfs_search_slot.  This does all of the checks
1609  * for node-level blocks and does any balancing required based on
1610  * the ins_len.
1611  *
1612  * If no extra work was required, zero is returned.  If we had to
1613  * drop the path, -EAGAIN is returned and btrfs_search_slot must
1614  * start over
1615  */
1616 static int
1617 setup_nodes_for_search(struct btrfs_trans_handle *trans,
1618 		       struct btrfs_root *root, struct btrfs_path *p,
1619 		       struct extent_buffer *b, int level, int ins_len)
1620 {
1621 	int ret;
1622 	if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1623 	    BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1624 		int sret;
1625 
1626 		sret = reada_for_balance(root, p, level);
1627 		if (sret)
1628 			goto again;
1629 
1630 		btrfs_set_path_blocking(p);
1631 		sret = split_node(trans, root, p, level);
1632 		btrfs_clear_path_blocking(p, NULL);
1633 
1634 		BUG_ON(sret > 0);
1635 		if (sret) {
1636 			ret = sret;
1637 			goto done;
1638 		}
1639 		b = p->nodes[level];
1640 	} else if (ins_len < 0 && btrfs_header_nritems(b) <
1641 		   BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
1642 		int sret;
1643 
1644 		sret = reada_for_balance(root, p, level);
1645 		if (sret)
1646 			goto again;
1647 
1648 		btrfs_set_path_blocking(p);
1649 		sret = balance_level(trans, root, p, level);
1650 		btrfs_clear_path_blocking(p, NULL);
1651 
1652 		if (sret) {
1653 			ret = sret;
1654 			goto done;
1655 		}
1656 		b = p->nodes[level];
1657 		if (!b) {
1658 			btrfs_release_path(NULL, p);
1659 			goto again;
1660 		}
1661 		BUG_ON(btrfs_header_nritems(b) == 1);
1662 	}
1663 	return 0;
1664 
1665 again:
1666 	ret = -EAGAIN;
1667 done:
1668 	return ret;
1669 }
1670 
1671 /*
1672  * look for key in the tree.  path is filled in with nodes along the way
1673  * if key is found, we return zero and you can find the item in the leaf
1674  * level of the path (level 0)
1675  *
1676  * If the key isn't found, the path points to the slot where it should
1677  * be inserted, and 1 is returned.  If there are other errors during the
1678  * search a negative error number is returned.
1679  *
1680  * if ins_len > 0, nodes and leaves will be split as we walk down the
1681  * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
1682  * possible)
1683  */
1684 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1685 		      *root, struct btrfs_key *key, struct btrfs_path *p, int
1686 		      ins_len, int cow)
1687 {
1688 	struct extent_buffer *b;
1689 	int slot;
1690 	int ret;
1691 	int err;
1692 	int level;
1693 	int lowest_unlock = 1;
1694 	u8 lowest_level = 0;
1695 
1696 	lowest_level = p->lowest_level;
1697 	WARN_ON(lowest_level && ins_len > 0);
1698 	WARN_ON(p->nodes[0] != NULL);
1699 
1700 	if (ins_len < 0)
1701 		lowest_unlock = 2;
1702 
1703 again:
1704 	if (p->search_commit_root) {
1705 		b = root->commit_root;
1706 		extent_buffer_get(b);
1707 		if (!p->skip_locking)
1708 			btrfs_tree_lock(b);
1709 	} else {
1710 		if (p->skip_locking)
1711 			b = btrfs_root_node(root);
1712 		else
1713 			b = btrfs_lock_root_node(root);
1714 	}
1715 
1716 	while (b) {
1717 		level = btrfs_header_level(b);
1718 
1719 		/*
1720 		 * setup the path here so we can release it under lock
1721 		 * contention with the cow code
1722 		 */
1723 		p->nodes[level] = b;
1724 		if (!p->skip_locking)
1725 			p->locks[level] = 1;
1726 
1727 		if (cow) {
1728 			/*
1729 			 * if we don't really need to cow this block
1730 			 * then we don't want to set the path blocking,
1731 			 * so we test it here
1732 			 */
1733 			if (!should_cow_block(trans, root, b))
1734 				goto cow_done;
1735 
1736 			btrfs_set_path_blocking(p);
1737 
1738 			err = btrfs_cow_block(trans, root, b,
1739 					      p->nodes[level + 1],
1740 					      p->slots[level + 1], &b);
1741 			if (err) {
1742 				free_extent_buffer(b);
1743 				ret = err;
1744 				goto done;
1745 			}
1746 		}
1747 cow_done:
1748 		BUG_ON(!cow && ins_len);
1749 		if (level != btrfs_header_level(b))
1750 			WARN_ON(1);
1751 		level = btrfs_header_level(b);
1752 
1753 		p->nodes[level] = b;
1754 		if (!p->skip_locking)
1755 			p->locks[level] = 1;
1756 
1757 		btrfs_clear_path_blocking(p, NULL);
1758 
1759 		/*
1760 		 * we have a lock on b and as long as we aren't changing
1761 		 * the tree, there is no way to for the items in b to change.
1762 		 * It is safe to drop the lock on our parent before we
1763 		 * go through the expensive btree search on b.
1764 		 *
1765 		 * If cow is true, then we might be changing slot zero,
1766 		 * which may require changing the parent.  So, we can't
1767 		 * drop the lock until after we know which slot we're
1768 		 * operating on.
1769 		 */
1770 		if (!cow)
1771 			btrfs_unlock_up_safe(p, level + 1);
1772 
1773 		ret = check_block(root, p, level);
1774 		if (ret) {
1775 			ret = -1;
1776 			goto done;
1777 		}
1778 
1779 		ret = bin_search(b, key, level, &slot);
1780 
1781 		if (level != 0) {
1782 			int dec = 0;
1783 			if (ret && slot > 0) {
1784 				dec = 1;
1785 				slot -= 1;
1786 			}
1787 			p->slots[level] = slot;
1788 			err = setup_nodes_for_search(trans, root, p, b, level,
1789 						     ins_len);
1790 			if (err == -EAGAIN)
1791 				goto again;
1792 			if (err) {
1793 				ret = err;
1794 				goto done;
1795 			}
1796 			b = p->nodes[level];
1797 			slot = p->slots[level];
1798 
1799 			unlock_up(p, level, lowest_unlock);
1800 
1801 			if (level == lowest_level) {
1802 				if (dec)
1803 					p->slots[level]++;
1804 				goto done;
1805 			}
1806 
1807 			err = read_block_for_search(trans, root, p,
1808 						    &b, level, slot, key);
1809 			if (err == -EAGAIN)
1810 				goto again;
1811 			if (err) {
1812 				ret = err;
1813 				goto done;
1814 			}
1815 
1816 			if (!p->skip_locking) {
1817 				btrfs_clear_path_blocking(p, NULL);
1818 				err = btrfs_try_spin_lock(b);
1819 
1820 				if (!err) {
1821 					btrfs_set_path_blocking(p);
1822 					btrfs_tree_lock(b);
1823 					btrfs_clear_path_blocking(p, b);
1824 				}
1825 			}
1826 		} else {
1827 			p->slots[level] = slot;
1828 			if (ins_len > 0 &&
1829 			    btrfs_leaf_free_space(root, b) < ins_len) {
1830 				btrfs_set_path_blocking(p);
1831 				err = split_leaf(trans, root, key,
1832 						 p, ins_len, ret == 0);
1833 				btrfs_clear_path_blocking(p, NULL);
1834 
1835 				BUG_ON(err > 0);
1836 				if (err) {
1837 					ret = err;
1838 					goto done;
1839 				}
1840 			}
1841 			if (!p->search_for_split)
1842 				unlock_up(p, level, lowest_unlock);
1843 			goto done;
1844 		}
1845 	}
1846 	ret = 1;
1847 done:
1848 	/*
1849 	 * we don't really know what they plan on doing with the path
1850 	 * from here on, so for now just mark it as blocking
1851 	 */
1852 	if (!p->leave_spinning)
1853 		btrfs_set_path_blocking(p);
1854 	if (ret < 0)
1855 		btrfs_release_path(root, p);
1856 	return ret;
1857 }
1858 
1859 /*
1860  * adjust the pointers going up the tree, starting at level
1861  * making sure the right key of each node is points to 'key'.
1862  * This is used after shifting pointers to the left, so it stops
1863  * fixing up pointers when a given leaf/node is not in slot 0 of the
1864  * higher levels
1865  *
1866  * If this fails to write a tree block, it returns -1, but continues
1867  * fixing up the blocks in ram so the tree is consistent.
1868  */
1869 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1870 			  struct btrfs_root *root, struct btrfs_path *path,
1871 			  struct btrfs_disk_key *key, int level)
1872 {
1873 	int i;
1874 	int ret = 0;
1875 	struct extent_buffer *t;
1876 
1877 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1878 		int tslot = path->slots[i];
1879 		if (!path->nodes[i])
1880 			break;
1881 		t = path->nodes[i];
1882 		btrfs_set_node_key(t, key, tslot);
1883 		btrfs_mark_buffer_dirty(path->nodes[i]);
1884 		if (tslot != 0)
1885 			break;
1886 	}
1887 	return ret;
1888 }
1889 
1890 /*
1891  * update item key.
1892  *
1893  * This function isn't completely safe. It's the caller's responsibility
1894  * that the new key won't break the order
1895  */
1896 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1897 			    struct btrfs_root *root, struct btrfs_path *path,
1898 			    struct btrfs_key *new_key)
1899 {
1900 	struct btrfs_disk_key disk_key;
1901 	struct extent_buffer *eb;
1902 	int slot;
1903 
1904 	eb = path->nodes[0];
1905 	slot = path->slots[0];
1906 	if (slot > 0) {
1907 		btrfs_item_key(eb, &disk_key, slot - 1);
1908 		if (comp_keys(&disk_key, new_key) >= 0)
1909 			return -1;
1910 	}
1911 	if (slot < btrfs_header_nritems(eb) - 1) {
1912 		btrfs_item_key(eb, &disk_key, slot + 1);
1913 		if (comp_keys(&disk_key, new_key) <= 0)
1914 			return -1;
1915 	}
1916 
1917 	btrfs_cpu_key_to_disk(&disk_key, new_key);
1918 	btrfs_set_item_key(eb, &disk_key, slot);
1919 	btrfs_mark_buffer_dirty(eb);
1920 	if (slot == 0)
1921 		fixup_low_keys(trans, root, path, &disk_key, 1);
1922 	return 0;
1923 }
1924 
1925 /*
1926  * try to push data from one node into the next node left in the
1927  * tree.
1928  *
1929  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1930  * error, and > 0 if there was no room in the left hand block.
1931  */
1932 static int push_node_left(struct btrfs_trans_handle *trans,
1933 			  struct btrfs_root *root, struct extent_buffer *dst,
1934 			  struct extent_buffer *src, int empty)
1935 {
1936 	int push_items = 0;
1937 	int src_nritems;
1938 	int dst_nritems;
1939 	int ret = 0;
1940 
1941 	src_nritems = btrfs_header_nritems(src);
1942 	dst_nritems = btrfs_header_nritems(dst);
1943 	push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1944 	WARN_ON(btrfs_header_generation(src) != trans->transid);
1945 	WARN_ON(btrfs_header_generation(dst) != trans->transid);
1946 
1947 	if (!empty && src_nritems <= 8)
1948 		return 1;
1949 
1950 	if (push_items <= 0)
1951 		return 1;
1952 
1953 	if (empty) {
1954 		push_items = min(src_nritems, push_items);
1955 		if (push_items < src_nritems) {
1956 			/* leave at least 8 pointers in the node if
1957 			 * we aren't going to empty it
1958 			 */
1959 			if (src_nritems - push_items < 8) {
1960 				if (push_items <= 8)
1961 					return 1;
1962 				push_items -= 8;
1963 			}
1964 		}
1965 	} else
1966 		push_items = min(src_nritems - 8, push_items);
1967 
1968 	copy_extent_buffer(dst, src,
1969 			   btrfs_node_key_ptr_offset(dst_nritems),
1970 			   btrfs_node_key_ptr_offset(0),
1971 			   push_items * sizeof(struct btrfs_key_ptr));
1972 
1973 	if (push_items < src_nritems) {
1974 		memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1975 				      btrfs_node_key_ptr_offset(push_items),
1976 				      (src_nritems - push_items) *
1977 				      sizeof(struct btrfs_key_ptr));
1978 	}
1979 	btrfs_set_header_nritems(src, src_nritems - push_items);
1980 	btrfs_set_header_nritems(dst, dst_nritems + push_items);
1981 	btrfs_mark_buffer_dirty(src);
1982 	btrfs_mark_buffer_dirty(dst);
1983 
1984 	return ret;
1985 }
1986 
1987 /*
1988  * try to push data from one node into the next node right in the
1989  * tree.
1990  *
1991  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1992  * error, and > 0 if there was no room in the right hand block.
1993  *
1994  * this will  only push up to 1/2 the contents of the left node over
1995  */
1996 static int balance_node_right(struct btrfs_trans_handle *trans,
1997 			      struct btrfs_root *root,
1998 			      struct extent_buffer *dst,
1999 			      struct extent_buffer *src)
2000 {
2001 	int push_items = 0;
2002 	int max_push;
2003 	int src_nritems;
2004 	int dst_nritems;
2005 	int ret = 0;
2006 
2007 	WARN_ON(btrfs_header_generation(src) != trans->transid);
2008 	WARN_ON(btrfs_header_generation(dst) != trans->transid);
2009 
2010 	src_nritems = btrfs_header_nritems(src);
2011 	dst_nritems = btrfs_header_nritems(dst);
2012 	push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2013 	if (push_items <= 0)
2014 		return 1;
2015 
2016 	if (src_nritems < 4)
2017 		return 1;
2018 
2019 	max_push = src_nritems / 2 + 1;
2020 	/* don't try to empty the node */
2021 	if (max_push >= src_nritems)
2022 		return 1;
2023 
2024 	if (max_push < push_items)
2025 		push_items = max_push;
2026 
2027 	memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2028 				      btrfs_node_key_ptr_offset(0),
2029 				      (dst_nritems) *
2030 				      sizeof(struct btrfs_key_ptr));
2031 
2032 	copy_extent_buffer(dst, src,
2033 			   btrfs_node_key_ptr_offset(0),
2034 			   btrfs_node_key_ptr_offset(src_nritems - push_items),
2035 			   push_items * sizeof(struct btrfs_key_ptr));
2036 
2037 	btrfs_set_header_nritems(src, src_nritems - push_items);
2038 	btrfs_set_header_nritems(dst, dst_nritems + push_items);
2039 
2040 	btrfs_mark_buffer_dirty(src);
2041 	btrfs_mark_buffer_dirty(dst);
2042 
2043 	return ret;
2044 }
2045 
2046 /*
2047  * helper function to insert a new root level in the tree.
2048  * A new node is allocated, and a single item is inserted to
2049  * point to the existing root
2050  *
2051  * returns zero on success or < 0 on failure.
2052  */
2053 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2054 			   struct btrfs_root *root,
2055 			   struct btrfs_path *path, int level)
2056 {
2057 	u64 lower_gen;
2058 	struct extent_buffer *lower;
2059 	struct extent_buffer *c;
2060 	struct extent_buffer *old;
2061 	struct btrfs_disk_key lower_key;
2062 
2063 	BUG_ON(path->nodes[level]);
2064 	BUG_ON(path->nodes[level-1] != root->node);
2065 
2066 	lower = path->nodes[level-1];
2067 	if (level == 1)
2068 		btrfs_item_key(lower, &lower_key, 0);
2069 	else
2070 		btrfs_node_key(lower, &lower_key, 0);
2071 
2072 	c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2073 				   root->root_key.objectid, &lower_key,
2074 				   level, root->node->start, 0);
2075 	if (IS_ERR(c))
2076 		return PTR_ERR(c);
2077 
2078 	memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
2079 	btrfs_set_header_nritems(c, 1);
2080 	btrfs_set_header_level(c, level);
2081 	btrfs_set_header_bytenr(c, c->start);
2082 	btrfs_set_header_generation(c, trans->transid);
2083 	btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
2084 	btrfs_set_header_owner(c, root->root_key.objectid);
2085 
2086 	write_extent_buffer(c, root->fs_info->fsid,
2087 			    (unsigned long)btrfs_header_fsid(c),
2088 			    BTRFS_FSID_SIZE);
2089 
2090 	write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2091 			    (unsigned long)btrfs_header_chunk_tree_uuid(c),
2092 			    BTRFS_UUID_SIZE);
2093 
2094 	btrfs_set_node_key(c, &lower_key, 0);
2095 	btrfs_set_node_blockptr(c, 0, lower->start);
2096 	lower_gen = btrfs_header_generation(lower);
2097 	WARN_ON(lower_gen != trans->transid);
2098 
2099 	btrfs_set_node_ptr_generation(c, 0, lower_gen);
2100 
2101 	btrfs_mark_buffer_dirty(c);
2102 
2103 	spin_lock(&root->node_lock);
2104 	old = root->node;
2105 	root->node = c;
2106 	spin_unlock(&root->node_lock);
2107 
2108 	/* the super has an extra ref to root->node */
2109 	free_extent_buffer(old);
2110 
2111 	add_root_to_dirty_list(root);
2112 	extent_buffer_get(c);
2113 	path->nodes[level] = c;
2114 	path->locks[level] = 1;
2115 	path->slots[level] = 0;
2116 	return 0;
2117 }
2118 
2119 /*
2120  * worker function to insert a single pointer in a node.
2121  * the node should have enough room for the pointer already
2122  *
2123  * slot and level indicate where you want the key to go, and
2124  * blocknr is the block the key points to.
2125  *
2126  * returns zero on success and < 0 on any error
2127  */
2128 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2129 		      *root, struct btrfs_path *path, struct btrfs_disk_key
2130 		      *key, u64 bytenr, int slot, int level)
2131 {
2132 	struct extent_buffer *lower;
2133 	int nritems;
2134 
2135 	BUG_ON(!path->nodes[level]);
2136 	lower = path->nodes[level];
2137 	nritems = btrfs_header_nritems(lower);
2138 	BUG_ON(slot > nritems);
2139 	if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2140 		BUG();
2141 	if (slot != nritems) {
2142 		memmove_extent_buffer(lower,
2143 			      btrfs_node_key_ptr_offset(slot + 1),
2144 			      btrfs_node_key_ptr_offset(slot),
2145 			      (nritems - slot) * sizeof(struct btrfs_key_ptr));
2146 	}
2147 	btrfs_set_node_key(lower, key, slot);
2148 	btrfs_set_node_blockptr(lower, slot, bytenr);
2149 	WARN_ON(trans->transid == 0);
2150 	btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2151 	btrfs_set_header_nritems(lower, nritems + 1);
2152 	btrfs_mark_buffer_dirty(lower);
2153 	return 0;
2154 }
2155 
2156 /*
2157  * split the node at the specified level in path in two.
2158  * The path is corrected to point to the appropriate node after the split
2159  *
2160  * Before splitting this tries to make some room in the node by pushing
2161  * left and right, if either one works, it returns right away.
2162  *
2163  * returns 0 on success and < 0 on failure
2164  */
2165 static noinline int split_node(struct btrfs_trans_handle *trans,
2166 			       struct btrfs_root *root,
2167 			       struct btrfs_path *path, int level)
2168 {
2169 	struct extent_buffer *c;
2170 	struct extent_buffer *split;
2171 	struct btrfs_disk_key disk_key;
2172 	int mid;
2173 	int ret;
2174 	int wret;
2175 	u32 c_nritems;
2176 
2177 	c = path->nodes[level];
2178 	WARN_ON(btrfs_header_generation(c) != trans->transid);
2179 	if (c == root->node) {
2180 		/* trying to split the root, lets make a new one */
2181 		ret = insert_new_root(trans, root, path, level + 1);
2182 		if (ret)
2183 			return ret;
2184 	} else {
2185 		ret = push_nodes_for_insert(trans, root, path, level);
2186 		c = path->nodes[level];
2187 		if (!ret && btrfs_header_nritems(c) <
2188 		    BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2189 			return 0;
2190 		if (ret < 0)
2191 			return ret;
2192 	}
2193 
2194 	c_nritems = btrfs_header_nritems(c);
2195 	mid = (c_nritems + 1) / 2;
2196 	btrfs_node_key(c, &disk_key, mid);
2197 
2198 	split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2199 					root->root_key.objectid,
2200 					&disk_key, level, c->start, 0);
2201 	if (IS_ERR(split))
2202 		return PTR_ERR(split);
2203 
2204 	memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
2205 	btrfs_set_header_level(split, btrfs_header_level(c));
2206 	btrfs_set_header_bytenr(split, split->start);
2207 	btrfs_set_header_generation(split, trans->transid);
2208 	btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
2209 	btrfs_set_header_owner(split, root->root_key.objectid);
2210 	write_extent_buffer(split, root->fs_info->fsid,
2211 			    (unsigned long)btrfs_header_fsid(split),
2212 			    BTRFS_FSID_SIZE);
2213 	write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2214 			    (unsigned long)btrfs_header_chunk_tree_uuid(split),
2215 			    BTRFS_UUID_SIZE);
2216 
2217 
2218 	copy_extent_buffer(split, c,
2219 			   btrfs_node_key_ptr_offset(0),
2220 			   btrfs_node_key_ptr_offset(mid),
2221 			   (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2222 	btrfs_set_header_nritems(split, c_nritems - mid);
2223 	btrfs_set_header_nritems(c, mid);
2224 	ret = 0;
2225 
2226 	btrfs_mark_buffer_dirty(c);
2227 	btrfs_mark_buffer_dirty(split);
2228 
2229 	wret = insert_ptr(trans, root, path, &disk_key, split->start,
2230 			  path->slots[level + 1] + 1,
2231 			  level + 1);
2232 	if (wret)
2233 		ret = wret;
2234 
2235 	if (path->slots[level] >= mid) {
2236 		path->slots[level] -= mid;
2237 		btrfs_tree_unlock(c);
2238 		free_extent_buffer(c);
2239 		path->nodes[level] = split;
2240 		path->slots[level + 1] += 1;
2241 	} else {
2242 		btrfs_tree_unlock(split);
2243 		free_extent_buffer(split);
2244 	}
2245 	return ret;
2246 }
2247 
2248 /*
2249  * how many bytes are required to store the items in a leaf.  start
2250  * and nr indicate which items in the leaf to check.  This totals up the
2251  * space used both by the item structs and the item data
2252  */
2253 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2254 {
2255 	int data_len;
2256 	int nritems = btrfs_header_nritems(l);
2257 	int end = min(nritems, start + nr) - 1;
2258 
2259 	if (!nr)
2260 		return 0;
2261 	data_len = btrfs_item_end_nr(l, start);
2262 	data_len = data_len - btrfs_item_offset_nr(l, end);
2263 	data_len += sizeof(struct btrfs_item) * nr;
2264 	WARN_ON(data_len < 0);
2265 	return data_len;
2266 }
2267 
2268 /*
2269  * The space between the end of the leaf items and
2270  * the start of the leaf data.  IOW, how much room
2271  * the leaf has left for both items and data
2272  */
2273 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2274 				   struct extent_buffer *leaf)
2275 {
2276 	int nritems = btrfs_header_nritems(leaf);
2277 	int ret;
2278 	ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2279 	if (ret < 0) {
2280 		printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2281 		       "used %d nritems %d\n",
2282 		       ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2283 		       leaf_space_used(leaf, 0, nritems), nritems);
2284 	}
2285 	return ret;
2286 }
2287 
2288 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2289 				      struct btrfs_root *root,
2290 				      struct btrfs_path *path,
2291 				      int data_size, int empty,
2292 				      struct extent_buffer *right,
2293 				      int free_space, u32 left_nritems)
2294 {
2295 	struct extent_buffer *left = path->nodes[0];
2296 	struct extent_buffer *upper = path->nodes[1];
2297 	struct btrfs_disk_key disk_key;
2298 	int slot;
2299 	u32 i;
2300 	int push_space = 0;
2301 	int push_items = 0;
2302 	struct btrfs_item *item;
2303 	u32 nr;
2304 	u32 right_nritems;
2305 	u32 data_end;
2306 	u32 this_item_size;
2307 
2308 	if (empty)
2309 		nr = 0;
2310 	else
2311 		nr = 1;
2312 
2313 	if (path->slots[0] >= left_nritems)
2314 		push_space += data_size;
2315 
2316 	slot = path->slots[1];
2317 	i = left_nritems - 1;
2318 	while (i >= nr) {
2319 		item = btrfs_item_nr(left, i);
2320 
2321 		if (!empty && push_items > 0) {
2322 			if (path->slots[0] > i)
2323 				break;
2324 			if (path->slots[0] == i) {
2325 				int space = btrfs_leaf_free_space(root, left);
2326 				if (space + push_space * 2 > free_space)
2327 					break;
2328 			}
2329 		}
2330 
2331 		if (path->slots[0] == i)
2332 			push_space += data_size;
2333 
2334 		if (!left->map_token) {
2335 			map_extent_buffer(left, (unsigned long)item,
2336 					sizeof(struct btrfs_item),
2337 					&left->map_token, &left->kaddr,
2338 					&left->map_start, &left->map_len,
2339 					KM_USER1);
2340 		}
2341 
2342 		this_item_size = btrfs_item_size(left, item);
2343 		if (this_item_size + sizeof(*item) + push_space > free_space)
2344 			break;
2345 
2346 		push_items++;
2347 		push_space += this_item_size + sizeof(*item);
2348 		if (i == 0)
2349 			break;
2350 		i--;
2351 	}
2352 	if (left->map_token) {
2353 		unmap_extent_buffer(left, left->map_token, KM_USER1);
2354 		left->map_token = NULL;
2355 	}
2356 
2357 	if (push_items == 0)
2358 		goto out_unlock;
2359 
2360 	if (!empty && push_items == left_nritems)
2361 		WARN_ON(1);
2362 
2363 	/* push left to right */
2364 	right_nritems = btrfs_header_nritems(right);
2365 
2366 	push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2367 	push_space -= leaf_data_end(root, left);
2368 
2369 	/* make room in the right data area */
2370 	data_end = leaf_data_end(root, right);
2371 	memmove_extent_buffer(right,
2372 			      btrfs_leaf_data(right) + data_end - push_space,
2373 			      btrfs_leaf_data(right) + data_end,
2374 			      BTRFS_LEAF_DATA_SIZE(root) - data_end);
2375 
2376 	/* copy from the left data area */
2377 	copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2378 		     BTRFS_LEAF_DATA_SIZE(root) - push_space,
2379 		     btrfs_leaf_data(left) + leaf_data_end(root, left),
2380 		     push_space);
2381 
2382 	memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2383 			      btrfs_item_nr_offset(0),
2384 			      right_nritems * sizeof(struct btrfs_item));
2385 
2386 	/* copy the items from left to right */
2387 	copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2388 		   btrfs_item_nr_offset(left_nritems - push_items),
2389 		   push_items * sizeof(struct btrfs_item));
2390 
2391 	/* update the item pointers */
2392 	right_nritems += push_items;
2393 	btrfs_set_header_nritems(right, right_nritems);
2394 	push_space = BTRFS_LEAF_DATA_SIZE(root);
2395 	for (i = 0; i < right_nritems; i++) {
2396 		item = btrfs_item_nr(right, i);
2397 		if (!right->map_token) {
2398 			map_extent_buffer(right, (unsigned long)item,
2399 					sizeof(struct btrfs_item),
2400 					&right->map_token, &right->kaddr,
2401 					&right->map_start, &right->map_len,
2402 					KM_USER1);
2403 		}
2404 		push_space -= btrfs_item_size(right, item);
2405 		btrfs_set_item_offset(right, item, push_space);
2406 	}
2407 
2408 	if (right->map_token) {
2409 		unmap_extent_buffer(right, right->map_token, KM_USER1);
2410 		right->map_token = NULL;
2411 	}
2412 	left_nritems -= push_items;
2413 	btrfs_set_header_nritems(left, left_nritems);
2414 
2415 	if (left_nritems)
2416 		btrfs_mark_buffer_dirty(left);
2417 	btrfs_mark_buffer_dirty(right);
2418 
2419 	btrfs_item_key(right, &disk_key, 0);
2420 	btrfs_set_node_key(upper, &disk_key, slot + 1);
2421 	btrfs_mark_buffer_dirty(upper);
2422 
2423 	/* then fixup the leaf pointer in the path */
2424 	if (path->slots[0] >= left_nritems) {
2425 		path->slots[0] -= left_nritems;
2426 		if (btrfs_header_nritems(path->nodes[0]) == 0)
2427 			clean_tree_block(trans, root, path->nodes[0]);
2428 		btrfs_tree_unlock(path->nodes[0]);
2429 		free_extent_buffer(path->nodes[0]);
2430 		path->nodes[0] = right;
2431 		path->slots[1] += 1;
2432 	} else {
2433 		btrfs_tree_unlock(right);
2434 		free_extent_buffer(right);
2435 	}
2436 	return 0;
2437 
2438 out_unlock:
2439 	btrfs_tree_unlock(right);
2440 	free_extent_buffer(right);
2441 	return 1;
2442 }
2443 
2444 /*
2445  * push some data in the path leaf to the right, trying to free up at
2446  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2447  *
2448  * returns 1 if the push failed because the other node didn't have enough
2449  * room, 0 if everything worked out and < 0 if there were major errors.
2450  */
2451 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2452 			   *root, struct btrfs_path *path, int data_size,
2453 			   int empty)
2454 {
2455 	struct extent_buffer *left = path->nodes[0];
2456 	struct extent_buffer *right;
2457 	struct extent_buffer *upper;
2458 	int slot;
2459 	int free_space;
2460 	u32 left_nritems;
2461 	int ret;
2462 
2463 	if (!path->nodes[1])
2464 		return 1;
2465 
2466 	slot = path->slots[1];
2467 	upper = path->nodes[1];
2468 	if (slot >= btrfs_header_nritems(upper) - 1)
2469 		return 1;
2470 
2471 	btrfs_assert_tree_locked(path->nodes[1]);
2472 
2473 	right = read_node_slot(root, upper, slot + 1);
2474 	btrfs_tree_lock(right);
2475 	btrfs_set_lock_blocking(right);
2476 
2477 	free_space = btrfs_leaf_free_space(root, right);
2478 	if (free_space < data_size)
2479 		goto out_unlock;
2480 
2481 	/* cow and double check */
2482 	ret = btrfs_cow_block(trans, root, right, upper,
2483 			      slot + 1, &right);
2484 	if (ret)
2485 		goto out_unlock;
2486 
2487 	free_space = btrfs_leaf_free_space(root, right);
2488 	if (free_space < data_size)
2489 		goto out_unlock;
2490 
2491 	left_nritems = btrfs_header_nritems(left);
2492 	if (left_nritems == 0)
2493 		goto out_unlock;
2494 
2495 	return __push_leaf_right(trans, root, path, data_size, empty,
2496 				right, free_space, left_nritems);
2497 out_unlock:
2498 	btrfs_tree_unlock(right);
2499 	free_extent_buffer(right);
2500 	return 1;
2501 }
2502 
2503 /*
2504  * push some data in the path leaf to the left, trying to free up at
2505  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2506  */
2507 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2508 				     struct btrfs_root *root,
2509 				     struct btrfs_path *path, int data_size,
2510 				     int empty, struct extent_buffer *left,
2511 				     int free_space, int right_nritems)
2512 {
2513 	struct btrfs_disk_key disk_key;
2514 	struct extent_buffer *right = path->nodes[0];
2515 	int slot;
2516 	int i;
2517 	int push_space = 0;
2518 	int push_items = 0;
2519 	struct btrfs_item *item;
2520 	u32 old_left_nritems;
2521 	u32 nr;
2522 	int ret = 0;
2523 	int wret;
2524 	u32 this_item_size;
2525 	u32 old_left_item_size;
2526 
2527 	slot = path->slots[1];
2528 
2529 	if (empty)
2530 		nr = right_nritems;
2531 	else
2532 		nr = right_nritems - 1;
2533 
2534 	for (i = 0; i < nr; i++) {
2535 		item = btrfs_item_nr(right, i);
2536 		if (!right->map_token) {
2537 			map_extent_buffer(right, (unsigned long)item,
2538 					sizeof(struct btrfs_item),
2539 					&right->map_token, &right->kaddr,
2540 					&right->map_start, &right->map_len,
2541 					KM_USER1);
2542 		}
2543 
2544 		if (!empty && push_items > 0) {
2545 			if (path->slots[0] < i)
2546 				break;
2547 			if (path->slots[0] == i) {
2548 				int space = btrfs_leaf_free_space(root, right);
2549 				if (space + push_space * 2 > free_space)
2550 					break;
2551 			}
2552 		}
2553 
2554 		if (path->slots[0] == i)
2555 			push_space += data_size;
2556 
2557 		this_item_size = btrfs_item_size(right, item);
2558 		if (this_item_size + sizeof(*item) + push_space > free_space)
2559 			break;
2560 
2561 		push_items++;
2562 		push_space += this_item_size + sizeof(*item);
2563 	}
2564 
2565 	if (right->map_token) {
2566 		unmap_extent_buffer(right, right->map_token, KM_USER1);
2567 		right->map_token = NULL;
2568 	}
2569 
2570 	if (push_items == 0) {
2571 		ret = 1;
2572 		goto out;
2573 	}
2574 	if (!empty && push_items == btrfs_header_nritems(right))
2575 		WARN_ON(1);
2576 
2577 	/* push data from right to left */
2578 	copy_extent_buffer(left, right,
2579 			   btrfs_item_nr_offset(btrfs_header_nritems(left)),
2580 			   btrfs_item_nr_offset(0),
2581 			   push_items * sizeof(struct btrfs_item));
2582 
2583 	push_space = BTRFS_LEAF_DATA_SIZE(root) -
2584 		     btrfs_item_offset_nr(right, push_items - 1);
2585 
2586 	copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2587 		     leaf_data_end(root, left) - push_space,
2588 		     btrfs_leaf_data(right) +
2589 		     btrfs_item_offset_nr(right, push_items - 1),
2590 		     push_space);
2591 	old_left_nritems = btrfs_header_nritems(left);
2592 	BUG_ON(old_left_nritems <= 0);
2593 
2594 	old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2595 	for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2596 		u32 ioff;
2597 
2598 		item = btrfs_item_nr(left, i);
2599 		if (!left->map_token) {
2600 			map_extent_buffer(left, (unsigned long)item,
2601 					sizeof(struct btrfs_item),
2602 					&left->map_token, &left->kaddr,
2603 					&left->map_start, &left->map_len,
2604 					KM_USER1);
2605 		}
2606 
2607 		ioff = btrfs_item_offset(left, item);
2608 		btrfs_set_item_offset(left, item,
2609 		      ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2610 	}
2611 	btrfs_set_header_nritems(left, old_left_nritems + push_items);
2612 	if (left->map_token) {
2613 		unmap_extent_buffer(left, left->map_token, KM_USER1);
2614 		left->map_token = NULL;
2615 	}
2616 
2617 	/* fixup right node */
2618 	if (push_items > right_nritems) {
2619 		printk(KERN_CRIT "push items %d nr %u\n", push_items,
2620 		       right_nritems);
2621 		WARN_ON(1);
2622 	}
2623 
2624 	if (push_items < right_nritems) {
2625 		push_space = btrfs_item_offset_nr(right, push_items - 1) -
2626 						  leaf_data_end(root, right);
2627 		memmove_extent_buffer(right, btrfs_leaf_data(right) +
2628 				      BTRFS_LEAF_DATA_SIZE(root) - push_space,
2629 				      btrfs_leaf_data(right) +
2630 				      leaf_data_end(root, right), push_space);
2631 
2632 		memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2633 			      btrfs_item_nr_offset(push_items),
2634 			     (btrfs_header_nritems(right) - push_items) *
2635 			     sizeof(struct btrfs_item));
2636 	}
2637 	right_nritems -= push_items;
2638 	btrfs_set_header_nritems(right, right_nritems);
2639 	push_space = BTRFS_LEAF_DATA_SIZE(root);
2640 	for (i = 0; i < right_nritems; i++) {
2641 		item = btrfs_item_nr(right, i);
2642 
2643 		if (!right->map_token) {
2644 			map_extent_buffer(right, (unsigned long)item,
2645 					sizeof(struct btrfs_item),
2646 					&right->map_token, &right->kaddr,
2647 					&right->map_start, &right->map_len,
2648 					KM_USER1);
2649 		}
2650 
2651 		push_space = push_space - btrfs_item_size(right, item);
2652 		btrfs_set_item_offset(right, item, push_space);
2653 	}
2654 	if (right->map_token) {
2655 		unmap_extent_buffer(right, right->map_token, KM_USER1);
2656 		right->map_token = NULL;
2657 	}
2658 
2659 	btrfs_mark_buffer_dirty(left);
2660 	if (right_nritems)
2661 		btrfs_mark_buffer_dirty(right);
2662 
2663 	btrfs_item_key(right, &disk_key, 0);
2664 	wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2665 	if (wret)
2666 		ret = wret;
2667 
2668 	/* then fixup the leaf pointer in the path */
2669 	if (path->slots[0] < push_items) {
2670 		path->slots[0] += old_left_nritems;
2671 		if (btrfs_header_nritems(path->nodes[0]) == 0)
2672 			clean_tree_block(trans, root, path->nodes[0]);
2673 		btrfs_tree_unlock(path->nodes[0]);
2674 		free_extent_buffer(path->nodes[0]);
2675 		path->nodes[0] = left;
2676 		path->slots[1] -= 1;
2677 	} else {
2678 		btrfs_tree_unlock(left);
2679 		free_extent_buffer(left);
2680 		path->slots[0] -= push_items;
2681 	}
2682 	BUG_ON(path->slots[0] < 0);
2683 	return ret;
2684 out:
2685 	btrfs_tree_unlock(left);
2686 	free_extent_buffer(left);
2687 	return ret;
2688 }
2689 
2690 /*
2691  * push some data in the path leaf to the left, trying to free up at
2692  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
2693  */
2694 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2695 			  *root, struct btrfs_path *path, int data_size,
2696 			  int empty)
2697 {
2698 	struct extent_buffer *right = path->nodes[0];
2699 	struct extent_buffer *left;
2700 	int slot;
2701 	int free_space;
2702 	u32 right_nritems;
2703 	int ret = 0;
2704 
2705 	slot = path->slots[1];
2706 	if (slot == 0)
2707 		return 1;
2708 	if (!path->nodes[1])
2709 		return 1;
2710 
2711 	right_nritems = btrfs_header_nritems(right);
2712 	if (right_nritems == 0)
2713 		return 1;
2714 
2715 	btrfs_assert_tree_locked(path->nodes[1]);
2716 
2717 	left = read_node_slot(root, path->nodes[1], slot - 1);
2718 	btrfs_tree_lock(left);
2719 	btrfs_set_lock_blocking(left);
2720 
2721 	free_space = btrfs_leaf_free_space(root, left);
2722 	if (free_space < data_size) {
2723 		ret = 1;
2724 		goto out;
2725 	}
2726 
2727 	/* cow and double check */
2728 	ret = btrfs_cow_block(trans, root, left,
2729 			      path->nodes[1], slot - 1, &left);
2730 	if (ret) {
2731 		/* we hit -ENOSPC, but it isn't fatal here */
2732 		ret = 1;
2733 		goto out;
2734 	}
2735 
2736 	free_space = btrfs_leaf_free_space(root, left);
2737 	if (free_space < data_size) {
2738 		ret = 1;
2739 		goto out;
2740 	}
2741 
2742 	return __push_leaf_left(trans, root, path, data_size,
2743 			       empty, left, free_space, right_nritems);
2744 out:
2745 	btrfs_tree_unlock(left);
2746 	free_extent_buffer(left);
2747 	return ret;
2748 }
2749 
2750 /*
2751  * split the path's leaf in two, making sure there is at least data_size
2752  * available for the resulting leaf level of the path.
2753  *
2754  * returns 0 if all went well and < 0 on failure.
2755  */
2756 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2757 			       struct btrfs_root *root,
2758 			       struct btrfs_path *path,
2759 			       struct extent_buffer *l,
2760 			       struct extent_buffer *right,
2761 			       int slot, int mid, int nritems)
2762 {
2763 	int data_copy_size;
2764 	int rt_data_off;
2765 	int i;
2766 	int ret = 0;
2767 	int wret;
2768 	struct btrfs_disk_key disk_key;
2769 
2770 	nritems = nritems - mid;
2771 	btrfs_set_header_nritems(right, nritems);
2772 	data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2773 
2774 	copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2775 			   btrfs_item_nr_offset(mid),
2776 			   nritems * sizeof(struct btrfs_item));
2777 
2778 	copy_extent_buffer(right, l,
2779 		     btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2780 		     data_copy_size, btrfs_leaf_data(l) +
2781 		     leaf_data_end(root, l), data_copy_size);
2782 
2783 	rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2784 		      btrfs_item_end_nr(l, mid);
2785 
2786 	for (i = 0; i < nritems; i++) {
2787 		struct btrfs_item *item = btrfs_item_nr(right, i);
2788 		u32 ioff;
2789 
2790 		if (!right->map_token) {
2791 			map_extent_buffer(right, (unsigned long)item,
2792 					sizeof(struct btrfs_item),
2793 					&right->map_token, &right->kaddr,
2794 					&right->map_start, &right->map_len,
2795 					KM_USER1);
2796 		}
2797 
2798 		ioff = btrfs_item_offset(right, item);
2799 		btrfs_set_item_offset(right, item, ioff + rt_data_off);
2800 	}
2801 
2802 	if (right->map_token) {
2803 		unmap_extent_buffer(right, right->map_token, KM_USER1);
2804 		right->map_token = NULL;
2805 	}
2806 
2807 	btrfs_set_header_nritems(l, mid);
2808 	ret = 0;
2809 	btrfs_item_key(right, &disk_key, 0);
2810 	wret = insert_ptr(trans, root, path, &disk_key, right->start,
2811 			  path->slots[1] + 1, 1);
2812 	if (wret)
2813 		ret = wret;
2814 
2815 	btrfs_mark_buffer_dirty(right);
2816 	btrfs_mark_buffer_dirty(l);
2817 	BUG_ON(path->slots[0] != slot);
2818 
2819 	if (mid <= slot) {
2820 		btrfs_tree_unlock(path->nodes[0]);
2821 		free_extent_buffer(path->nodes[0]);
2822 		path->nodes[0] = right;
2823 		path->slots[0] -= mid;
2824 		path->slots[1] += 1;
2825 	} else {
2826 		btrfs_tree_unlock(right);
2827 		free_extent_buffer(right);
2828 	}
2829 
2830 	BUG_ON(path->slots[0] < 0);
2831 
2832 	return ret;
2833 }
2834 
2835 /*
2836  * split the path's leaf in two, making sure there is at least data_size
2837  * available for the resulting leaf level of the path.
2838  *
2839  * returns 0 if all went well and < 0 on failure.
2840  */
2841 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2842 			       struct btrfs_root *root,
2843 			       struct btrfs_key *ins_key,
2844 			       struct btrfs_path *path, int data_size,
2845 			       int extend)
2846 {
2847 	struct btrfs_disk_key disk_key;
2848 	struct extent_buffer *l;
2849 	u32 nritems;
2850 	int mid;
2851 	int slot;
2852 	struct extent_buffer *right;
2853 	int ret = 0;
2854 	int wret;
2855 	int split;
2856 	int num_doubles = 0;
2857 
2858 	l = path->nodes[0];
2859 	slot = path->slots[0];
2860 	if (extend && data_size + btrfs_item_size_nr(l, slot) +
2861 	    sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2862 		return -EOVERFLOW;
2863 
2864 	/* first try to make some room by pushing left and right */
2865 	if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2866 		wret = push_leaf_right(trans, root, path, data_size, 0);
2867 		if (wret < 0)
2868 			return wret;
2869 		if (wret) {
2870 			wret = push_leaf_left(trans, root, path, data_size, 0);
2871 			if (wret < 0)
2872 				return wret;
2873 		}
2874 		l = path->nodes[0];
2875 
2876 		/* did the pushes work? */
2877 		if (btrfs_leaf_free_space(root, l) >= data_size)
2878 			return 0;
2879 	}
2880 
2881 	if (!path->nodes[1]) {
2882 		ret = insert_new_root(trans, root, path, 1);
2883 		if (ret)
2884 			return ret;
2885 	}
2886 again:
2887 	split = 1;
2888 	l = path->nodes[0];
2889 	slot = path->slots[0];
2890 	nritems = btrfs_header_nritems(l);
2891 	mid = (nritems + 1) / 2;
2892 
2893 	if (mid <= slot) {
2894 		if (nritems == 1 ||
2895 		    leaf_space_used(l, mid, nritems - mid) + data_size >
2896 			BTRFS_LEAF_DATA_SIZE(root)) {
2897 			if (slot >= nritems) {
2898 				split = 0;
2899 			} else {
2900 				mid = slot;
2901 				if (mid != nritems &&
2902 				    leaf_space_used(l, mid, nritems - mid) +
2903 				    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2904 					split = 2;
2905 				}
2906 			}
2907 		}
2908 	} else {
2909 		if (leaf_space_used(l, 0, mid) + data_size >
2910 			BTRFS_LEAF_DATA_SIZE(root)) {
2911 			if (!extend && data_size && slot == 0) {
2912 				split = 0;
2913 			} else if ((extend || !data_size) && slot == 0) {
2914 				mid = 1;
2915 			} else {
2916 				mid = slot;
2917 				if (mid != nritems &&
2918 				    leaf_space_used(l, mid, nritems - mid) +
2919 				    data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2920 					split = 2 ;
2921 				}
2922 			}
2923 		}
2924 	}
2925 
2926 	if (split == 0)
2927 		btrfs_cpu_key_to_disk(&disk_key, ins_key);
2928 	else
2929 		btrfs_item_key(l, &disk_key, mid);
2930 
2931 	right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
2932 					root->root_key.objectid,
2933 					&disk_key, 0, l->start, 0);
2934 	if (IS_ERR(right)) {
2935 		BUG_ON(1);
2936 		return PTR_ERR(right);
2937 	}
2938 
2939 	memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2940 	btrfs_set_header_bytenr(right, right->start);
2941 	btrfs_set_header_generation(right, trans->transid);
2942 	btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2943 	btrfs_set_header_owner(right, root->root_key.objectid);
2944 	btrfs_set_header_level(right, 0);
2945 	write_extent_buffer(right, root->fs_info->fsid,
2946 			    (unsigned long)btrfs_header_fsid(right),
2947 			    BTRFS_FSID_SIZE);
2948 
2949 	write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2950 			    (unsigned long)btrfs_header_chunk_tree_uuid(right),
2951 			    BTRFS_UUID_SIZE);
2952 
2953 	if (split == 0) {
2954 		if (mid <= slot) {
2955 			btrfs_set_header_nritems(right, 0);
2956 			wret = insert_ptr(trans, root, path,
2957 					  &disk_key, right->start,
2958 					  path->slots[1] + 1, 1);
2959 			if (wret)
2960 				ret = wret;
2961 
2962 			btrfs_tree_unlock(path->nodes[0]);
2963 			free_extent_buffer(path->nodes[0]);
2964 			path->nodes[0] = right;
2965 			path->slots[0] = 0;
2966 			path->slots[1] += 1;
2967 		} else {
2968 			btrfs_set_header_nritems(right, 0);
2969 			wret = insert_ptr(trans, root, path,
2970 					  &disk_key,
2971 					  right->start,
2972 					  path->slots[1], 1);
2973 			if (wret)
2974 				ret = wret;
2975 			btrfs_tree_unlock(path->nodes[0]);
2976 			free_extent_buffer(path->nodes[0]);
2977 			path->nodes[0] = right;
2978 			path->slots[0] = 0;
2979 			if (path->slots[1] == 0) {
2980 				wret = fixup_low_keys(trans, root,
2981 						path, &disk_key, 1);
2982 				if (wret)
2983 					ret = wret;
2984 			}
2985 		}
2986 		btrfs_mark_buffer_dirty(right);
2987 		return ret;
2988 	}
2989 
2990 	ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2991 	BUG_ON(ret);
2992 
2993 	if (split == 2) {
2994 		BUG_ON(num_doubles != 0);
2995 		num_doubles++;
2996 		goto again;
2997 	}
2998 
2999 	return ret;
3000 }
3001 
3002 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3003 					 struct btrfs_root *root,
3004 					 struct btrfs_path *path, int ins_len)
3005 {
3006 	struct btrfs_key key;
3007 	struct extent_buffer *leaf;
3008 	struct btrfs_file_extent_item *fi;
3009 	u64 extent_len = 0;
3010 	u32 item_size;
3011 	int ret;
3012 
3013 	leaf = path->nodes[0];
3014 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3015 
3016 	BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3017 	       key.type != BTRFS_EXTENT_CSUM_KEY);
3018 
3019 	if (btrfs_leaf_free_space(root, leaf) >= ins_len)
3020 		return 0;
3021 
3022 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3023 	if (key.type == BTRFS_EXTENT_DATA_KEY) {
3024 		fi = btrfs_item_ptr(leaf, path->slots[0],
3025 				    struct btrfs_file_extent_item);
3026 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3027 	}
3028 	btrfs_release_path(root, path);
3029 
3030 	path->keep_locks = 1;
3031 	path->search_for_split = 1;
3032 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3033 	path->search_for_split = 0;
3034 	if (ret < 0)
3035 		goto err;
3036 
3037 	ret = -EAGAIN;
3038 	leaf = path->nodes[0];
3039 	/* if our item isn't there or got smaller, return now */
3040 	if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3041 		goto err;
3042 
3043 	if (key.type == BTRFS_EXTENT_DATA_KEY) {
3044 		fi = btrfs_item_ptr(leaf, path->slots[0],
3045 				    struct btrfs_file_extent_item);
3046 		if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3047 			goto err;
3048 	}
3049 
3050 	btrfs_set_path_blocking(path);
3051 	ret = split_leaf(trans, root, &key, path, ins_len, 1);
3052 	BUG_ON(ret);
3053 
3054 	path->keep_locks = 0;
3055 	btrfs_unlock_up_safe(path, 1);
3056 	return 0;
3057 err:
3058 	path->keep_locks = 0;
3059 	return ret;
3060 }
3061 
3062 static noinline int split_item(struct btrfs_trans_handle *trans,
3063 			       struct btrfs_root *root,
3064 			       struct btrfs_path *path,
3065 			       struct btrfs_key *new_key,
3066 			       unsigned long split_offset)
3067 {
3068 	struct extent_buffer *leaf;
3069 	struct btrfs_item *item;
3070 	struct btrfs_item *new_item;
3071 	int slot;
3072 	char *buf;
3073 	u32 nritems;
3074 	u32 item_size;
3075 	u32 orig_offset;
3076 	struct btrfs_disk_key disk_key;
3077 
3078 	leaf = path->nodes[0];
3079 	BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3080 
3081 	btrfs_set_path_blocking(path);
3082 
3083 	item = btrfs_item_nr(leaf, path->slots[0]);
3084 	orig_offset = btrfs_item_offset(leaf, item);
3085 	item_size = btrfs_item_size(leaf, item);
3086 
3087 	buf = kmalloc(item_size, GFP_NOFS);
3088 	if (!buf)
3089 		return -ENOMEM;
3090 
3091 	read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3092 			    path->slots[0]), item_size);
3093 
3094 	slot = path->slots[0] + 1;
3095 	nritems = btrfs_header_nritems(leaf);
3096 	if (slot != nritems) {
3097 		/* shift the items */
3098 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3099 				btrfs_item_nr_offset(slot),
3100 				(nritems - slot) * sizeof(struct btrfs_item));
3101 	}
3102 
3103 	btrfs_cpu_key_to_disk(&disk_key, new_key);
3104 	btrfs_set_item_key(leaf, &disk_key, slot);
3105 
3106 	new_item = btrfs_item_nr(leaf, slot);
3107 
3108 	btrfs_set_item_offset(leaf, new_item, orig_offset);
3109 	btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3110 
3111 	btrfs_set_item_offset(leaf, item,
3112 			      orig_offset + item_size - split_offset);
3113 	btrfs_set_item_size(leaf, item, split_offset);
3114 
3115 	btrfs_set_header_nritems(leaf, nritems + 1);
3116 
3117 	/* write the data for the start of the original item */
3118 	write_extent_buffer(leaf, buf,
3119 			    btrfs_item_ptr_offset(leaf, path->slots[0]),
3120 			    split_offset);
3121 
3122 	/* write the data for the new item */
3123 	write_extent_buffer(leaf, buf + split_offset,
3124 			    btrfs_item_ptr_offset(leaf, slot),
3125 			    item_size - split_offset);
3126 	btrfs_mark_buffer_dirty(leaf);
3127 
3128 	BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
3129 	kfree(buf);
3130 	return 0;
3131 }
3132 
3133 /*
3134  * This function splits a single item into two items,
3135  * giving 'new_key' to the new item and splitting the
3136  * old one at split_offset (from the start of the item).
3137  *
3138  * The path may be released by this operation.  After
3139  * the split, the path is pointing to the old item.  The
3140  * new item is going to be in the same node as the old one.
3141  *
3142  * Note, the item being split must be smaller enough to live alone on
3143  * a tree block with room for one extra struct btrfs_item
3144  *
3145  * This allows us to split the item in place, keeping a lock on the
3146  * leaf the entire time.
3147  */
3148 int btrfs_split_item(struct btrfs_trans_handle *trans,
3149 		     struct btrfs_root *root,
3150 		     struct btrfs_path *path,
3151 		     struct btrfs_key *new_key,
3152 		     unsigned long split_offset)
3153 {
3154 	int ret;
3155 	ret = setup_leaf_for_split(trans, root, path,
3156 				   sizeof(struct btrfs_item));
3157 	if (ret)
3158 		return ret;
3159 
3160 	ret = split_item(trans, root, path, new_key, split_offset);
3161 	return ret;
3162 }
3163 
3164 /*
3165  * This function duplicate a item, giving 'new_key' to the new item.
3166  * It guarantees both items live in the same tree leaf and the new item
3167  * is contiguous with the original item.
3168  *
3169  * This allows us to split file extent in place, keeping a lock on the
3170  * leaf the entire time.
3171  */
3172 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
3173 			 struct btrfs_root *root,
3174 			 struct btrfs_path *path,
3175 			 struct btrfs_key *new_key)
3176 {
3177 	struct extent_buffer *leaf;
3178 	int ret;
3179 	u32 item_size;
3180 
3181 	leaf = path->nodes[0];
3182 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3183 	ret = setup_leaf_for_split(trans, root, path,
3184 				   item_size + sizeof(struct btrfs_item));
3185 	if (ret)
3186 		return ret;
3187 
3188 	path->slots[0]++;
3189 	ret = setup_items_for_insert(trans, root, path, new_key, &item_size,
3190 				     item_size, item_size +
3191 				     sizeof(struct btrfs_item), 1);
3192 	BUG_ON(ret);
3193 
3194 	leaf = path->nodes[0];
3195 	memcpy_extent_buffer(leaf,
3196 			     btrfs_item_ptr_offset(leaf, path->slots[0]),
3197 			     btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
3198 			     item_size);
3199 	return 0;
3200 }
3201 
3202 /*
3203  * make the item pointed to by the path smaller.  new_size indicates
3204  * how small to make it, and from_end tells us if we just chop bytes
3205  * off the end of the item or if we shift the item to chop bytes off
3206  * the front.
3207  */
3208 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3209 			struct btrfs_root *root,
3210 			struct btrfs_path *path,
3211 			u32 new_size, int from_end)
3212 {
3213 	int ret = 0;
3214 	int slot;
3215 	int slot_orig;
3216 	struct extent_buffer *leaf;
3217 	struct btrfs_item *item;
3218 	u32 nritems;
3219 	unsigned int data_end;
3220 	unsigned int old_data_start;
3221 	unsigned int old_size;
3222 	unsigned int size_diff;
3223 	int i;
3224 
3225 	slot_orig = path->slots[0];
3226 	leaf = path->nodes[0];
3227 	slot = path->slots[0];
3228 
3229 	old_size = btrfs_item_size_nr(leaf, slot);
3230 	if (old_size == new_size)
3231 		return 0;
3232 
3233 	nritems = btrfs_header_nritems(leaf);
3234 	data_end = leaf_data_end(root, leaf);
3235 
3236 	old_data_start = btrfs_item_offset_nr(leaf, slot);
3237 
3238 	size_diff = old_size - new_size;
3239 
3240 	BUG_ON(slot < 0);
3241 	BUG_ON(slot >= nritems);
3242 
3243 	/*
3244 	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3245 	 */
3246 	/* first correct the data pointers */
3247 	for (i = slot; i < nritems; i++) {
3248 		u32 ioff;
3249 		item = btrfs_item_nr(leaf, i);
3250 
3251 		if (!leaf->map_token) {
3252 			map_extent_buffer(leaf, (unsigned long)item,
3253 					sizeof(struct btrfs_item),
3254 					&leaf->map_token, &leaf->kaddr,
3255 					&leaf->map_start, &leaf->map_len,
3256 					KM_USER1);
3257 		}
3258 
3259 		ioff = btrfs_item_offset(leaf, item);
3260 		btrfs_set_item_offset(leaf, item, ioff + size_diff);
3261 	}
3262 
3263 	if (leaf->map_token) {
3264 		unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3265 		leaf->map_token = NULL;
3266 	}
3267 
3268 	/* shift the data */
3269 	if (from_end) {
3270 		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3271 			      data_end + size_diff, btrfs_leaf_data(leaf) +
3272 			      data_end, old_data_start + new_size - data_end);
3273 	} else {
3274 		struct btrfs_disk_key disk_key;
3275 		u64 offset;
3276 
3277 		btrfs_item_key(leaf, &disk_key, slot);
3278 
3279 		if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3280 			unsigned long ptr;
3281 			struct btrfs_file_extent_item *fi;
3282 
3283 			fi = btrfs_item_ptr(leaf, slot,
3284 					    struct btrfs_file_extent_item);
3285 			fi = (struct btrfs_file_extent_item *)(
3286 			     (unsigned long)fi - size_diff);
3287 
3288 			if (btrfs_file_extent_type(leaf, fi) ==
3289 			    BTRFS_FILE_EXTENT_INLINE) {
3290 				ptr = btrfs_item_ptr_offset(leaf, slot);
3291 				memmove_extent_buffer(leaf, ptr,
3292 				      (unsigned long)fi,
3293 				      offsetof(struct btrfs_file_extent_item,
3294 						 disk_bytenr));
3295 			}
3296 		}
3297 
3298 		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3299 			      data_end + size_diff, btrfs_leaf_data(leaf) +
3300 			      data_end, old_data_start - data_end);
3301 
3302 		offset = btrfs_disk_key_offset(&disk_key);
3303 		btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3304 		btrfs_set_item_key(leaf, &disk_key, slot);
3305 		if (slot == 0)
3306 			fixup_low_keys(trans, root, path, &disk_key, 1);
3307 	}
3308 
3309 	item = btrfs_item_nr(leaf, slot);
3310 	btrfs_set_item_size(leaf, item, new_size);
3311 	btrfs_mark_buffer_dirty(leaf);
3312 
3313 	ret = 0;
3314 	if (btrfs_leaf_free_space(root, leaf) < 0) {
3315 		btrfs_print_leaf(root, leaf);
3316 		BUG();
3317 	}
3318 	return ret;
3319 }
3320 
3321 /*
3322  * make the item pointed to by the path bigger, data_size is the new size.
3323  */
3324 int btrfs_extend_item(struct btrfs_trans_handle *trans,
3325 		      struct btrfs_root *root, struct btrfs_path *path,
3326 		      u32 data_size)
3327 {
3328 	int ret = 0;
3329 	int slot;
3330 	int slot_orig;
3331 	struct extent_buffer *leaf;
3332 	struct btrfs_item *item;
3333 	u32 nritems;
3334 	unsigned int data_end;
3335 	unsigned int old_data;
3336 	unsigned int old_size;
3337 	int i;
3338 
3339 	slot_orig = path->slots[0];
3340 	leaf = path->nodes[0];
3341 
3342 	nritems = btrfs_header_nritems(leaf);
3343 	data_end = leaf_data_end(root, leaf);
3344 
3345 	if (btrfs_leaf_free_space(root, leaf) < data_size) {
3346 		btrfs_print_leaf(root, leaf);
3347 		BUG();
3348 	}
3349 	slot = path->slots[0];
3350 	old_data = btrfs_item_end_nr(leaf, slot);
3351 
3352 	BUG_ON(slot < 0);
3353 	if (slot >= nritems) {
3354 		btrfs_print_leaf(root, leaf);
3355 		printk(KERN_CRIT "slot %d too large, nritems %d\n",
3356 		       slot, nritems);
3357 		BUG_ON(1);
3358 	}
3359 
3360 	/*
3361 	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3362 	 */
3363 	/* first correct the data pointers */
3364 	for (i = slot; i < nritems; i++) {
3365 		u32 ioff;
3366 		item = btrfs_item_nr(leaf, i);
3367 
3368 		if (!leaf->map_token) {
3369 			map_extent_buffer(leaf, (unsigned long)item,
3370 					sizeof(struct btrfs_item),
3371 					&leaf->map_token, &leaf->kaddr,
3372 					&leaf->map_start, &leaf->map_len,
3373 					KM_USER1);
3374 		}
3375 		ioff = btrfs_item_offset(leaf, item);
3376 		btrfs_set_item_offset(leaf, item, ioff - data_size);
3377 	}
3378 
3379 	if (leaf->map_token) {
3380 		unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3381 		leaf->map_token = NULL;
3382 	}
3383 
3384 	/* shift the data */
3385 	memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3386 		      data_end - data_size, btrfs_leaf_data(leaf) +
3387 		      data_end, old_data - data_end);
3388 
3389 	data_end = old_data;
3390 	old_size = btrfs_item_size_nr(leaf, slot);
3391 	item = btrfs_item_nr(leaf, slot);
3392 	btrfs_set_item_size(leaf, item, old_size + data_size);
3393 	btrfs_mark_buffer_dirty(leaf);
3394 
3395 	ret = 0;
3396 	if (btrfs_leaf_free_space(root, leaf) < 0) {
3397 		btrfs_print_leaf(root, leaf);
3398 		BUG();
3399 	}
3400 	return ret;
3401 }
3402 
3403 /*
3404  * Given a key and some data, insert items into the tree.
3405  * This does all the path init required, making room in the tree if needed.
3406  * Returns the number of keys that were inserted.
3407  */
3408 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3409 			    struct btrfs_root *root,
3410 			    struct btrfs_path *path,
3411 			    struct btrfs_key *cpu_key, u32 *data_size,
3412 			    int nr)
3413 {
3414 	struct extent_buffer *leaf;
3415 	struct btrfs_item *item;
3416 	int ret = 0;
3417 	int slot;
3418 	int i;
3419 	u32 nritems;
3420 	u32 total_data = 0;
3421 	u32 total_size = 0;
3422 	unsigned int data_end;
3423 	struct btrfs_disk_key disk_key;
3424 	struct btrfs_key found_key;
3425 
3426 	for (i = 0; i < nr; i++) {
3427 		if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3428 		    BTRFS_LEAF_DATA_SIZE(root)) {
3429 			break;
3430 			nr = i;
3431 		}
3432 		total_data += data_size[i];
3433 		total_size += data_size[i] + sizeof(struct btrfs_item);
3434 	}
3435 	BUG_ON(nr == 0);
3436 
3437 	ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3438 	if (ret == 0)
3439 		return -EEXIST;
3440 	if (ret < 0)
3441 		goto out;
3442 
3443 	leaf = path->nodes[0];
3444 
3445 	nritems = btrfs_header_nritems(leaf);
3446 	data_end = leaf_data_end(root, leaf);
3447 
3448 	if (btrfs_leaf_free_space(root, leaf) < total_size) {
3449 		for (i = nr; i >= 0; i--) {
3450 			total_data -= data_size[i];
3451 			total_size -= data_size[i] + sizeof(struct btrfs_item);
3452 			if (total_size < btrfs_leaf_free_space(root, leaf))
3453 				break;
3454 		}
3455 		nr = i;
3456 	}
3457 
3458 	slot = path->slots[0];
3459 	BUG_ON(slot < 0);
3460 
3461 	if (slot != nritems) {
3462 		unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3463 
3464 		item = btrfs_item_nr(leaf, slot);
3465 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
3466 
3467 		/* figure out how many keys we can insert in here */
3468 		total_data = data_size[0];
3469 		for (i = 1; i < nr; i++) {
3470 			if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3471 				break;
3472 			total_data += data_size[i];
3473 		}
3474 		nr = i;
3475 
3476 		if (old_data < data_end) {
3477 			btrfs_print_leaf(root, leaf);
3478 			printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3479 			       slot, old_data, data_end);
3480 			BUG_ON(1);
3481 		}
3482 		/*
3483 		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3484 		 */
3485 		/* first correct the data pointers */
3486 		WARN_ON(leaf->map_token);
3487 		for (i = slot; i < nritems; i++) {
3488 			u32 ioff;
3489 
3490 			item = btrfs_item_nr(leaf, i);
3491 			if (!leaf->map_token) {
3492 				map_extent_buffer(leaf, (unsigned long)item,
3493 					sizeof(struct btrfs_item),
3494 					&leaf->map_token, &leaf->kaddr,
3495 					&leaf->map_start, &leaf->map_len,
3496 					KM_USER1);
3497 			}
3498 
3499 			ioff = btrfs_item_offset(leaf, item);
3500 			btrfs_set_item_offset(leaf, item, ioff - total_data);
3501 		}
3502 		if (leaf->map_token) {
3503 			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3504 			leaf->map_token = NULL;
3505 		}
3506 
3507 		/* shift the items */
3508 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3509 			      btrfs_item_nr_offset(slot),
3510 			      (nritems - slot) * sizeof(struct btrfs_item));
3511 
3512 		/* shift the data */
3513 		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3514 			      data_end - total_data, btrfs_leaf_data(leaf) +
3515 			      data_end, old_data - data_end);
3516 		data_end = old_data;
3517 	} else {
3518 		/*
3519 		 * this sucks but it has to be done, if we are inserting at
3520 		 * the end of the leaf only insert 1 of the items, since we
3521 		 * have no way of knowing whats on the next leaf and we'd have
3522 		 * to drop our current locks to figure it out
3523 		 */
3524 		nr = 1;
3525 	}
3526 
3527 	/* setup the item for the new data */
3528 	for (i = 0; i < nr; i++) {
3529 		btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3530 		btrfs_set_item_key(leaf, &disk_key, slot + i);
3531 		item = btrfs_item_nr(leaf, slot + i);
3532 		btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3533 		data_end -= data_size[i];
3534 		btrfs_set_item_size(leaf, item, data_size[i]);
3535 	}
3536 	btrfs_set_header_nritems(leaf, nritems + nr);
3537 	btrfs_mark_buffer_dirty(leaf);
3538 
3539 	ret = 0;
3540 	if (slot == 0) {
3541 		btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3542 		ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3543 	}
3544 
3545 	if (btrfs_leaf_free_space(root, leaf) < 0) {
3546 		btrfs_print_leaf(root, leaf);
3547 		BUG();
3548 	}
3549 out:
3550 	if (!ret)
3551 		ret = nr;
3552 	return ret;
3553 }
3554 
3555 /*
3556  * this is a helper for btrfs_insert_empty_items, the main goal here is
3557  * to save stack depth by doing the bulk of the work in a function
3558  * that doesn't call btrfs_search_slot
3559  */
3560 static noinline_for_stack int
3561 setup_items_for_insert(struct btrfs_trans_handle *trans,
3562 		      struct btrfs_root *root, struct btrfs_path *path,
3563 		      struct btrfs_key *cpu_key, u32 *data_size,
3564 		      u32 total_data, u32 total_size, int nr)
3565 {
3566 	struct btrfs_item *item;
3567 	int i;
3568 	u32 nritems;
3569 	unsigned int data_end;
3570 	struct btrfs_disk_key disk_key;
3571 	int ret;
3572 	struct extent_buffer *leaf;
3573 	int slot;
3574 
3575 	leaf = path->nodes[0];
3576 	slot = path->slots[0];
3577 
3578 	nritems = btrfs_header_nritems(leaf);
3579 	data_end = leaf_data_end(root, leaf);
3580 
3581 	if (btrfs_leaf_free_space(root, leaf) < total_size) {
3582 		btrfs_print_leaf(root, leaf);
3583 		printk(KERN_CRIT "not enough freespace need %u have %d\n",
3584 		       total_size, btrfs_leaf_free_space(root, leaf));
3585 		BUG();
3586 	}
3587 
3588 	if (slot != nritems) {
3589 		unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3590 
3591 		if (old_data < data_end) {
3592 			btrfs_print_leaf(root, leaf);
3593 			printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3594 			       slot, old_data, data_end);
3595 			BUG_ON(1);
3596 		}
3597 		/*
3598 		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3599 		 */
3600 		/* first correct the data pointers */
3601 		WARN_ON(leaf->map_token);
3602 		for (i = slot; i < nritems; i++) {
3603 			u32 ioff;
3604 
3605 			item = btrfs_item_nr(leaf, i);
3606 			if (!leaf->map_token) {
3607 				map_extent_buffer(leaf, (unsigned long)item,
3608 					sizeof(struct btrfs_item),
3609 					&leaf->map_token, &leaf->kaddr,
3610 					&leaf->map_start, &leaf->map_len,
3611 					KM_USER1);
3612 			}
3613 
3614 			ioff = btrfs_item_offset(leaf, item);
3615 			btrfs_set_item_offset(leaf, item, ioff - total_data);
3616 		}
3617 		if (leaf->map_token) {
3618 			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3619 			leaf->map_token = NULL;
3620 		}
3621 
3622 		/* shift the items */
3623 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3624 			      btrfs_item_nr_offset(slot),
3625 			      (nritems - slot) * sizeof(struct btrfs_item));
3626 
3627 		/* shift the data */
3628 		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3629 			      data_end - total_data, btrfs_leaf_data(leaf) +
3630 			      data_end, old_data - data_end);
3631 		data_end = old_data;
3632 	}
3633 
3634 	/* setup the item for the new data */
3635 	for (i = 0; i < nr; i++) {
3636 		btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3637 		btrfs_set_item_key(leaf, &disk_key, slot + i);
3638 		item = btrfs_item_nr(leaf, slot + i);
3639 		btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3640 		data_end -= data_size[i];
3641 		btrfs_set_item_size(leaf, item, data_size[i]);
3642 	}
3643 
3644 	btrfs_set_header_nritems(leaf, nritems + nr);
3645 
3646 	ret = 0;
3647 	if (slot == 0) {
3648 		struct btrfs_disk_key disk_key;
3649 		btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3650 		ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3651 	}
3652 	btrfs_unlock_up_safe(path, 1);
3653 	btrfs_mark_buffer_dirty(leaf);
3654 
3655 	if (btrfs_leaf_free_space(root, leaf) < 0) {
3656 		btrfs_print_leaf(root, leaf);
3657 		BUG();
3658 	}
3659 	return ret;
3660 }
3661 
3662 /*
3663  * Given a key and some data, insert items into the tree.
3664  * This does all the path init required, making room in the tree if needed.
3665  */
3666 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3667 			    struct btrfs_root *root,
3668 			    struct btrfs_path *path,
3669 			    struct btrfs_key *cpu_key, u32 *data_size,
3670 			    int nr)
3671 {
3672 	struct extent_buffer *leaf;
3673 	int ret = 0;
3674 	int slot;
3675 	int i;
3676 	u32 total_size = 0;
3677 	u32 total_data = 0;
3678 
3679 	for (i = 0; i < nr; i++)
3680 		total_data += data_size[i];
3681 
3682 	total_size = total_data + (nr * sizeof(struct btrfs_item));
3683 	ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3684 	if (ret == 0)
3685 		return -EEXIST;
3686 	if (ret < 0)
3687 		goto out;
3688 
3689 	leaf = path->nodes[0];
3690 	slot = path->slots[0];
3691 	BUG_ON(slot < 0);
3692 
3693 	ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3694 			       total_data, total_size, nr);
3695 
3696 out:
3697 	return ret;
3698 }
3699 
3700 /*
3701  * Given a key and some data, insert an item into the tree.
3702  * This does all the path init required, making room in the tree if needed.
3703  */
3704 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3705 		      *root, struct btrfs_key *cpu_key, void *data, u32
3706 		      data_size)
3707 {
3708 	int ret = 0;
3709 	struct btrfs_path *path;
3710 	struct extent_buffer *leaf;
3711 	unsigned long ptr;
3712 
3713 	path = btrfs_alloc_path();
3714 	BUG_ON(!path);
3715 	ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3716 	if (!ret) {
3717 		leaf = path->nodes[0];
3718 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3719 		write_extent_buffer(leaf, data, ptr, data_size);
3720 		btrfs_mark_buffer_dirty(leaf);
3721 	}
3722 	btrfs_free_path(path);
3723 	return ret;
3724 }
3725 
3726 /*
3727  * delete the pointer from a given node.
3728  *
3729  * the tree should have been previously balanced so the deletion does not
3730  * empty a node.
3731  */
3732 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3733 		   struct btrfs_path *path, int level, int slot)
3734 {
3735 	struct extent_buffer *parent = path->nodes[level];
3736 	u32 nritems;
3737 	int ret = 0;
3738 	int wret;
3739 
3740 	nritems = btrfs_header_nritems(parent);
3741 	if (slot != nritems - 1) {
3742 		memmove_extent_buffer(parent,
3743 			      btrfs_node_key_ptr_offset(slot),
3744 			      btrfs_node_key_ptr_offset(slot + 1),
3745 			      sizeof(struct btrfs_key_ptr) *
3746 			      (nritems - slot - 1));
3747 	}
3748 	nritems--;
3749 	btrfs_set_header_nritems(parent, nritems);
3750 	if (nritems == 0 && parent == root->node) {
3751 		BUG_ON(btrfs_header_level(root->node) != 1);
3752 		/* just turn the root into a leaf and break */
3753 		btrfs_set_header_level(root->node, 0);
3754 	} else if (slot == 0) {
3755 		struct btrfs_disk_key disk_key;
3756 
3757 		btrfs_node_key(parent, &disk_key, 0);
3758 		wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3759 		if (wret)
3760 			ret = wret;
3761 	}
3762 	btrfs_mark_buffer_dirty(parent);
3763 	return ret;
3764 }
3765 
3766 /*
3767  * a helper function to delete the leaf pointed to by path->slots[1] and
3768  * path->nodes[1].
3769  *
3770  * This deletes the pointer in path->nodes[1] and frees the leaf
3771  * block extent.  zero is returned if it all worked out, < 0 otherwise.
3772  *
3773  * The path must have already been setup for deleting the leaf, including
3774  * all the proper balancing.  path->nodes[1] must be locked.
3775  */
3776 static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3777 				   struct btrfs_root *root,
3778 				   struct btrfs_path *path,
3779 				   struct extent_buffer *leaf)
3780 {
3781 	int ret;
3782 
3783 	WARN_ON(btrfs_header_generation(leaf) != trans->transid);
3784 	ret = del_ptr(trans, root, path, 1, path->slots[1]);
3785 	if (ret)
3786 		return ret;
3787 
3788 	/*
3789 	 * btrfs_free_extent is expensive, we want to make sure we
3790 	 * aren't holding any locks when we call it
3791 	 */
3792 	btrfs_unlock_up_safe(path, 0);
3793 
3794 	ret = btrfs_free_tree_block(trans, root, leaf->start, leaf->len,
3795 				    0, root->root_key.objectid, 0);
3796 	return ret;
3797 }
3798 /*
3799  * delete the item at the leaf level in path.  If that empties
3800  * the leaf, remove it from the tree
3801  */
3802 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3803 		    struct btrfs_path *path, int slot, int nr)
3804 {
3805 	struct extent_buffer *leaf;
3806 	struct btrfs_item *item;
3807 	int last_off;
3808 	int dsize = 0;
3809 	int ret = 0;
3810 	int wret;
3811 	int i;
3812 	u32 nritems;
3813 
3814 	leaf = path->nodes[0];
3815 	last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3816 
3817 	for (i = 0; i < nr; i++)
3818 		dsize += btrfs_item_size_nr(leaf, slot + i);
3819 
3820 	nritems = btrfs_header_nritems(leaf);
3821 
3822 	if (slot + nr != nritems) {
3823 		int data_end = leaf_data_end(root, leaf);
3824 
3825 		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3826 			      data_end + dsize,
3827 			      btrfs_leaf_data(leaf) + data_end,
3828 			      last_off - data_end);
3829 
3830 		for (i = slot + nr; i < nritems; i++) {
3831 			u32 ioff;
3832 
3833 			item = btrfs_item_nr(leaf, i);
3834 			if (!leaf->map_token) {
3835 				map_extent_buffer(leaf, (unsigned long)item,
3836 					sizeof(struct btrfs_item),
3837 					&leaf->map_token, &leaf->kaddr,
3838 					&leaf->map_start, &leaf->map_len,
3839 					KM_USER1);
3840 			}
3841 			ioff = btrfs_item_offset(leaf, item);
3842 			btrfs_set_item_offset(leaf, item, ioff + dsize);
3843 		}
3844 
3845 		if (leaf->map_token) {
3846 			unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3847 			leaf->map_token = NULL;
3848 		}
3849 
3850 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3851 			      btrfs_item_nr_offset(slot + nr),
3852 			      sizeof(struct btrfs_item) *
3853 			      (nritems - slot - nr));
3854 	}
3855 	btrfs_set_header_nritems(leaf, nritems - nr);
3856 	nritems -= nr;
3857 
3858 	/* delete the leaf if we've emptied it */
3859 	if (nritems == 0) {
3860 		if (leaf == root->node) {
3861 			btrfs_set_header_level(leaf, 0);
3862 		} else {
3863 			ret = btrfs_del_leaf(trans, root, path, leaf);
3864 			BUG_ON(ret);
3865 		}
3866 	} else {
3867 		int used = leaf_space_used(leaf, 0, nritems);
3868 		if (slot == 0) {
3869 			struct btrfs_disk_key disk_key;
3870 
3871 			btrfs_item_key(leaf, &disk_key, 0);
3872 			wret = fixup_low_keys(trans, root, path,
3873 					      &disk_key, 1);
3874 			if (wret)
3875 				ret = wret;
3876 		}
3877 
3878 		/* delete the leaf if it is mostly empty */
3879 		if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
3880 			/* push_leaf_left fixes the path.
3881 			 * make sure the path still points to our leaf
3882 			 * for possible call to del_ptr below
3883 			 */
3884 			slot = path->slots[1];
3885 			extent_buffer_get(leaf);
3886 
3887 			btrfs_set_path_blocking(path);
3888 			wret = push_leaf_left(trans, root, path, 1, 1);
3889 			if (wret < 0 && wret != -ENOSPC)
3890 				ret = wret;
3891 
3892 			if (path->nodes[0] == leaf &&
3893 			    btrfs_header_nritems(leaf)) {
3894 				wret = push_leaf_right(trans, root, path, 1, 1);
3895 				if (wret < 0 && wret != -ENOSPC)
3896 					ret = wret;
3897 			}
3898 
3899 			if (btrfs_header_nritems(leaf) == 0) {
3900 				path->slots[1] = slot;
3901 				ret = btrfs_del_leaf(trans, root, path, leaf);
3902 				BUG_ON(ret);
3903 				free_extent_buffer(leaf);
3904 			} else {
3905 				/* if we're still in the path, make sure
3906 				 * we're dirty.  Otherwise, one of the
3907 				 * push_leaf functions must have already
3908 				 * dirtied this buffer
3909 				 */
3910 				if (path->nodes[0] == leaf)
3911 					btrfs_mark_buffer_dirty(leaf);
3912 				free_extent_buffer(leaf);
3913 			}
3914 		} else {
3915 			btrfs_mark_buffer_dirty(leaf);
3916 		}
3917 	}
3918 	return ret;
3919 }
3920 
3921 /*
3922  * search the tree again to find a leaf with lesser keys
3923  * returns 0 if it found something or 1 if there are no lesser leaves.
3924  * returns < 0 on io errors.
3925  *
3926  * This may release the path, and so you may lose any locks held at the
3927  * time you call it.
3928  */
3929 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3930 {
3931 	struct btrfs_key key;
3932 	struct btrfs_disk_key found_key;
3933 	int ret;
3934 
3935 	btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3936 
3937 	if (key.offset > 0)
3938 		key.offset--;
3939 	else if (key.type > 0)
3940 		key.type--;
3941 	else if (key.objectid > 0)
3942 		key.objectid--;
3943 	else
3944 		return 1;
3945 
3946 	btrfs_release_path(root, path);
3947 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3948 	if (ret < 0)
3949 		return ret;
3950 	btrfs_item_key(path->nodes[0], &found_key, 0);
3951 	ret = comp_keys(&found_key, &key);
3952 	if (ret < 0)
3953 		return 0;
3954 	return 1;
3955 }
3956 
3957 /*
3958  * A helper function to walk down the tree starting at min_key, and looking
3959  * for nodes or leaves that are either in cache or have a minimum
3960  * transaction id.  This is used by the btree defrag code, and tree logging
3961  *
3962  * This does not cow, but it does stuff the starting key it finds back
3963  * into min_key, so you can call btrfs_search_slot with cow=1 on the
3964  * key and get a writable path.
3965  *
3966  * This does lock as it descends, and path->keep_locks should be set
3967  * to 1 by the caller.
3968  *
3969  * This honors path->lowest_level to prevent descent past a given level
3970  * of the tree.
3971  *
3972  * min_trans indicates the oldest transaction that you are interested
3973  * in walking through.  Any nodes or leaves older than min_trans are
3974  * skipped over (without reading them).
3975  *
3976  * returns zero if something useful was found, < 0 on error and 1 if there
3977  * was nothing in the tree that matched the search criteria.
3978  */
3979 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3980 			 struct btrfs_key *max_key,
3981 			 struct btrfs_path *path, int cache_only,
3982 			 u64 min_trans)
3983 {
3984 	struct extent_buffer *cur;
3985 	struct btrfs_key found_key;
3986 	int slot;
3987 	int sret;
3988 	u32 nritems;
3989 	int level;
3990 	int ret = 1;
3991 
3992 	WARN_ON(!path->keep_locks);
3993 again:
3994 	cur = btrfs_lock_root_node(root);
3995 	level = btrfs_header_level(cur);
3996 	WARN_ON(path->nodes[level]);
3997 	path->nodes[level] = cur;
3998 	path->locks[level] = 1;
3999 
4000 	if (btrfs_header_generation(cur) < min_trans) {
4001 		ret = 1;
4002 		goto out;
4003 	}
4004 	while (1) {
4005 		nritems = btrfs_header_nritems(cur);
4006 		level = btrfs_header_level(cur);
4007 		sret = bin_search(cur, min_key, level, &slot);
4008 
4009 		/* at the lowest level, we're done, setup the path and exit */
4010 		if (level == path->lowest_level) {
4011 			if (slot >= nritems)
4012 				goto find_next_key;
4013 			ret = 0;
4014 			path->slots[level] = slot;
4015 			btrfs_item_key_to_cpu(cur, &found_key, slot);
4016 			goto out;
4017 		}
4018 		if (sret && slot > 0)
4019 			slot--;
4020 		/*
4021 		 * check this node pointer against the cache_only and
4022 		 * min_trans parameters.  If it isn't in cache or is too
4023 		 * old, skip to the next one.
4024 		 */
4025 		while (slot < nritems) {
4026 			u64 blockptr;
4027 			u64 gen;
4028 			struct extent_buffer *tmp;
4029 			struct btrfs_disk_key disk_key;
4030 
4031 			blockptr = btrfs_node_blockptr(cur, slot);
4032 			gen = btrfs_node_ptr_generation(cur, slot);
4033 			if (gen < min_trans) {
4034 				slot++;
4035 				continue;
4036 			}
4037 			if (!cache_only)
4038 				break;
4039 
4040 			if (max_key) {
4041 				btrfs_node_key(cur, &disk_key, slot);
4042 				if (comp_keys(&disk_key, max_key) >= 0) {
4043 					ret = 1;
4044 					goto out;
4045 				}
4046 			}
4047 
4048 			tmp = btrfs_find_tree_block(root, blockptr,
4049 					    btrfs_level_size(root, level - 1));
4050 
4051 			if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
4052 				free_extent_buffer(tmp);
4053 				break;
4054 			}
4055 			if (tmp)
4056 				free_extent_buffer(tmp);
4057 			slot++;
4058 		}
4059 find_next_key:
4060 		/*
4061 		 * we didn't find a candidate key in this node, walk forward
4062 		 * and find another one
4063 		 */
4064 		if (slot >= nritems) {
4065 			path->slots[level] = slot;
4066 			btrfs_set_path_blocking(path);
4067 			sret = btrfs_find_next_key(root, path, min_key, level,
4068 						  cache_only, min_trans);
4069 			if (sret == 0) {
4070 				btrfs_release_path(root, path);
4071 				goto again;
4072 			} else {
4073 				goto out;
4074 			}
4075 		}
4076 		/* save our key for returning back */
4077 		btrfs_node_key_to_cpu(cur, &found_key, slot);
4078 		path->slots[level] = slot;
4079 		if (level == path->lowest_level) {
4080 			ret = 0;
4081 			unlock_up(path, level, 1);
4082 			goto out;
4083 		}
4084 		btrfs_set_path_blocking(path);
4085 		cur = read_node_slot(root, cur, slot);
4086 
4087 		btrfs_tree_lock(cur);
4088 
4089 		path->locks[level - 1] = 1;
4090 		path->nodes[level - 1] = cur;
4091 		unlock_up(path, level, 1);
4092 		btrfs_clear_path_blocking(path, NULL);
4093 	}
4094 out:
4095 	if (ret == 0)
4096 		memcpy(min_key, &found_key, sizeof(found_key));
4097 	btrfs_set_path_blocking(path);
4098 	return ret;
4099 }
4100 
4101 /*
4102  * this is similar to btrfs_next_leaf, but does not try to preserve
4103  * and fixup the path.  It looks for and returns the next key in the
4104  * tree based on the current path and the cache_only and min_trans
4105  * parameters.
4106  *
4107  * 0 is returned if another key is found, < 0 if there are any errors
4108  * and 1 is returned if there are no higher keys in the tree
4109  *
4110  * path->keep_locks should be set to 1 on the search made before
4111  * calling this function.
4112  */
4113 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4114 			struct btrfs_key *key, int level,
4115 			int cache_only, u64 min_trans)
4116 {
4117 	int slot;
4118 	struct extent_buffer *c;
4119 
4120 	WARN_ON(!path->keep_locks);
4121 	while (level < BTRFS_MAX_LEVEL) {
4122 		if (!path->nodes[level])
4123 			return 1;
4124 
4125 		slot = path->slots[level] + 1;
4126 		c = path->nodes[level];
4127 next:
4128 		if (slot >= btrfs_header_nritems(c)) {
4129 			int ret;
4130 			int orig_lowest;
4131 			struct btrfs_key cur_key;
4132 			if (level + 1 >= BTRFS_MAX_LEVEL ||
4133 			    !path->nodes[level + 1])
4134 				return 1;
4135 
4136 			if (path->locks[level + 1]) {
4137 				level++;
4138 				continue;
4139 			}
4140 
4141 			slot = btrfs_header_nritems(c) - 1;
4142 			if (level == 0)
4143 				btrfs_item_key_to_cpu(c, &cur_key, slot);
4144 			else
4145 				btrfs_node_key_to_cpu(c, &cur_key, slot);
4146 
4147 			orig_lowest = path->lowest_level;
4148 			btrfs_release_path(root, path);
4149 			path->lowest_level = level;
4150 			ret = btrfs_search_slot(NULL, root, &cur_key, path,
4151 						0, 0);
4152 			path->lowest_level = orig_lowest;
4153 			if (ret < 0)
4154 				return ret;
4155 
4156 			c = path->nodes[level];
4157 			slot = path->slots[level];
4158 			if (ret == 0)
4159 				slot++;
4160 			goto next;
4161 		}
4162 
4163 		if (level == 0)
4164 			btrfs_item_key_to_cpu(c, key, slot);
4165 		else {
4166 			u64 blockptr = btrfs_node_blockptr(c, slot);
4167 			u64 gen = btrfs_node_ptr_generation(c, slot);
4168 
4169 			if (cache_only) {
4170 				struct extent_buffer *cur;
4171 				cur = btrfs_find_tree_block(root, blockptr,
4172 					    btrfs_level_size(root, level - 1));
4173 				if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4174 					slot++;
4175 					if (cur)
4176 						free_extent_buffer(cur);
4177 					goto next;
4178 				}
4179 				free_extent_buffer(cur);
4180 			}
4181 			if (gen < min_trans) {
4182 				slot++;
4183 				goto next;
4184 			}
4185 			btrfs_node_key_to_cpu(c, key, slot);
4186 		}
4187 		return 0;
4188 	}
4189 	return 1;
4190 }
4191 
4192 /*
4193  * search the tree again to find a leaf with greater keys
4194  * returns 0 if it found something or 1 if there are no greater leaves.
4195  * returns < 0 on io errors.
4196  */
4197 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4198 {
4199 	int slot;
4200 	int level;
4201 	struct extent_buffer *c;
4202 	struct extent_buffer *next;
4203 	struct btrfs_key key;
4204 	u32 nritems;
4205 	int ret;
4206 	int old_spinning = path->leave_spinning;
4207 	int force_blocking = 0;
4208 
4209 	nritems = btrfs_header_nritems(path->nodes[0]);
4210 	if (nritems == 0)
4211 		return 1;
4212 
4213 	/*
4214 	 * we take the blocks in an order that upsets lockdep.  Using
4215 	 * blocking mode is the only way around it.
4216 	 */
4217 #ifdef CONFIG_DEBUG_LOCK_ALLOC
4218 	force_blocking = 1;
4219 #endif
4220 
4221 	btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4222 again:
4223 	level = 1;
4224 	next = NULL;
4225 	btrfs_release_path(root, path);
4226 
4227 	path->keep_locks = 1;
4228 
4229 	if (!force_blocking)
4230 		path->leave_spinning = 1;
4231 
4232 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4233 	path->keep_locks = 0;
4234 
4235 	if (ret < 0)
4236 		return ret;
4237 
4238 	nritems = btrfs_header_nritems(path->nodes[0]);
4239 	/*
4240 	 * by releasing the path above we dropped all our locks.  A balance
4241 	 * could have added more items next to the key that used to be
4242 	 * at the very end of the block.  So, check again here and
4243 	 * advance the path if there are now more items available.
4244 	 */
4245 	if (nritems > 0 && path->slots[0] < nritems - 1) {
4246 		if (ret == 0)
4247 			path->slots[0]++;
4248 		ret = 0;
4249 		goto done;
4250 	}
4251 
4252 	while (level < BTRFS_MAX_LEVEL) {
4253 		if (!path->nodes[level]) {
4254 			ret = 1;
4255 			goto done;
4256 		}
4257 
4258 		slot = path->slots[level] + 1;
4259 		c = path->nodes[level];
4260 		if (slot >= btrfs_header_nritems(c)) {
4261 			level++;
4262 			if (level == BTRFS_MAX_LEVEL) {
4263 				ret = 1;
4264 				goto done;
4265 			}
4266 			continue;
4267 		}
4268 
4269 		if (next) {
4270 			btrfs_tree_unlock(next);
4271 			free_extent_buffer(next);
4272 		}
4273 
4274 		next = c;
4275 		ret = read_block_for_search(NULL, root, path, &next, level,
4276 					    slot, &key);
4277 		if (ret == -EAGAIN)
4278 			goto again;
4279 
4280 		if (ret < 0) {
4281 			btrfs_release_path(root, path);
4282 			goto done;
4283 		}
4284 
4285 		if (!path->skip_locking) {
4286 			ret = btrfs_try_spin_lock(next);
4287 			if (!ret) {
4288 				btrfs_set_path_blocking(path);
4289 				btrfs_tree_lock(next);
4290 				if (!force_blocking)
4291 					btrfs_clear_path_blocking(path, next);
4292 			}
4293 			if (force_blocking)
4294 				btrfs_set_lock_blocking(next);
4295 		}
4296 		break;
4297 	}
4298 	path->slots[level] = slot;
4299 	while (1) {
4300 		level--;
4301 		c = path->nodes[level];
4302 		if (path->locks[level])
4303 			btrfs_tree_unlock(c);
4304 
4305 		free_extent_buffer(c);
4306 		path->nodes[level] = next;
4307 		path->slots[level] = 0;
4308 		if (!path->skip_locking)
4309 			path->locks[level] = 1;
4310 
4311 		if (!level)
4312 			break;
4313 
4314 		ret = read_block_for_search(NULL, root, path, &next, level,
4315 					    0, &key);
4316 		if (ret == -EAGAIN)
4317 			goto again;
4318 
4319 		if (ret < 0) {
4320 			btrfs_release_path(root, path);
4321 			goto done;
4322 		}
4323 
4324 		if (!path->skip_locking) {
4325 			btrfs_assert_tree_locked(path->nodes[level]);
4326 			ret = btrfs_try_spin_lock(next);
4327 			if (!ret) {
4328 				btrfs_set_path_blocking(path);
4329 				btrfs_tree_lock(next);
4330 				if (!force_blocking)
4331 					btrfs_clear_path_blocking(path, next);
4332 			}
4333 			if (force_blocking)
4334 				btrfs_set_lock_blocking(next);
4335 		}
4336 	}
4337 	ret = 0;
4338 done:
4339 	unlock_up(path, 0, 1);
4340 	path->leave_spinning = old_spinning;
4341 	if (!old_spinning)
4342 		btrfs_set_path_blocking(path);
4343 
4344 	return ret;
4345 }
4346 
4347 /*
4348  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4349  * searching until it gets past min_objectid or finds an item of 'type'
4350  *
4351  * returns 0 if something is found, 1 if nothing was found and < 0 on error
4352  */
4353 int btrfs_previous_item(struct btrfs_root *root,
4354 			struct btrfs_path *path, u64 min_objectid,
4355 			int type)
4356 {
4357 	struct btrfs_key found_key;
4358 	struct extent_buffer *leaf;
4359 	u32 nritems;
4360 	int ret;
4361 
4362 	while (1) {
4363 		if (path->slots[0] == 0) {
4364 			btrfs_set_path_blocking(path);
4365 			ret = btrfs_prev_leaf(root, path);
4366 			if (ret != 0)
4367 				return ret;
4368 		} else {
4369 			path->slots[0]--;
4370 		}
4371 		leaf = path->nodes[0];
4372 		nritems = btrfs_header_nritems(leaf);
4373 		if (nritems == 0)
4374 			return 1;
4375 		if (path->slots[0] == nritems)
4376 			path->slots[0]--;
4377 
4378 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4379 		if (found_key.objectid < min_objectid)
4380 			break;
4381 		if (found_key.type == type)
4382 			return 0;
4383 		if (found_key.objectid == min_objectid &&
4384 		    found_key.type < type)
4385 			break;
4386 	}
4387 	return 1;
4388 }
4389