xref: /openbmc/linux/fs/btrfs/ctree.c (revision 41a0f85e)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007,2008 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/rbtree.h>
9 #include <linux/mm.h>
10 #include <linux/error-injection.h>
11 #include "messages.h"
12 #include "ctree.h"
13 #include "disk-io.h"
14 #include "transaction.h"
15 #include "print-tree.h"
16 #include "locking.h"
17 #include "volumes.h"
18 #include "qgroup.h"
19 #include "tree-mod-log.h"
20 #include "tree-checker.h"
21 #include "fs.h"
22 #include "accessors.h"
23 #include "extent-tree.h"
24 #include "relocation.h"
25 #include "file-item.h"
26 
27 static struct kmem_cache *btrfs_path_cachep;
28 
29 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
30 		      *root, struct btrfs_path *path, int level);
31 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
32 		      const struct btrfs_key *ins_key, struct btrfs_path *path,
33 		      int data_size, int extend);
34 static int push_node_left(struct btrfs_trans_handle *trans,
35 			  struct extent_buffer *dst,
36 			  struct extent_buffer *src, int empty);
37 static int balance_node_right(struct btrfs_trans_handle *trans,
38 			      struct extent_buffer *dst_buf,
39 			      struct extent_buffer *src_buf);
40 
41 static const struct btrfs_csums {
42 	u16		size;
43 	const char	name[10];
44 	const char	driver[12];
45 } btrfs_csums[] = {
46 	[BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" },
47 	[BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" },
48 	[BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" },
49 	[BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b",
50 				     .driver = "blake2b-256" },
51 };
52 
53 /*
54  * The leaf data grows from end-to-front in the node.  this returns the address
55  * of the start of the last item, which is the stop of the leaf data stack.
56  */
leaf_data_end(const struct extent_buffer * leaf)57 static unsigned int leaf_data_end(const struct extent_buffer *leaf)
58 {
59 	u32 nr = btrfs_header_nritems(leaf);
60 
61 	if (nr == 0)
62 		return BTRFS_LEAF_DATA_SIZE(leaf->fs_info);
63 	return btrfs_item_offset(leaf, nr - 1);
64 }
65 
66 /*
67  * Move data in a @leaf (using memmove, safe for overlapping ranges).
68  *
69  * @leaf:	leaf that we're doing a memmove on
70  * @dst_offset:	item data offset we're moving to
71  * @src_offset:	item data offset were' moving from
72  * @len:	length of the data we're moving
73  *
74  * Wrapper around memmove_extent_buffer() that takes into account the header on
75  * the leaf.  The btrfs_item offset's start directly after the header, so we
76  * have to adjust any offsets to account for the header in the leaf.  This
77  * handles that math to simplify the callers.
78  */
memmove_leaf_data(const struct extent_buffer * leaf,unsigned long dst_offset,unsigned long src_offset,unsigned long len)79 static inline void memmove_leaf_data(const struct extent_buffer *leaf,
80 				     unsigned long dst_offset,
81 				     unsigned long src_offset,
82 				     unsigned long len)
83 {
84 	memmove_extent_buffer(leaf, btrfs_item_nr_offset(leaf, 0) + dst_offset,
85 			      btrfs_item_nr_offset(leaf, 0) + src_offset, len);
86 }
87 
88 /*
89  * Copy item data from @src into @dst at the given @offset.
90  *
91  * @dst:	destination leaf that we're copying into
92  * @src:	source leaf that we're copying from
93  * @dst_offset:	item data offset we're copying to
94  * @src_offset:	item data offset were' copying from
95  * @len:	length of the data we're copying
96  *
97  * Wrapper around copy_extent_buffer() that takes into account the header on
98  * the leaf.  The btrfs_item offset's start directly after the header, so we
99  * have to adjust any offsets to account for the header in the leaf.  This
100  * handles that math to simplify the callers.
101  */
copy_leaf_data(const struct extent_buffer * dst,const struct extent_buffer * src,unsigned long dst_offset,unsigned long src_offset,unsigned long len)102 static inline void copy_leaf_data(const struct extent_buffer *dst,
103 				  const struct extent_buffer *src,
104 				  unsigned long dst_offset,
105 				  unsigned long src_offset, unsigned long len)
106 {
107 	copy_extent_buffer(dst, src, btrfs_item_nr_offset(dst, 0) + dst_offset,
108 			   btrfs_item_nr_offset(src, 0) + src_offset, len);
109 }
110 
111 /*
112  * Move items in a @leaf (using memmove).
113  *
114  * @dst:	destination leaf for the items
115  * @dst_item:	the item nr we're copying into
116  * @src_item:	the item nr we're copying from
117  * @nr_items:	the number of items to copy
118  *
119  * Wrapper around memmove_extent_buffer() that does the math to get the
120  * appropriate offsets into the leaf from the item numbers.
121  */
memmove_leaf_items(const struct extent_buffer * leaf,int dst_item,int src_item,int nr_items)122 static inline void memmove_leaf_items(const struct extent_buffer *leaf,
123 				      int dst_item, int src_item, int nr_items)
124 {
125 	memmove_extent_buffer(leaf, btrfs_item_nr_offset(leaf, dst_item),
126 			      btrfs_item_nr_offset(leaf, src_item),
127 			      nr_items * sizeof(struct btrfs_item));
128 }
129 
130 /*
131  * Copy items from @src into @dst at the given @offset.
132  *
133  * @dst:	destination leaf for the items
134  * @src:	source leaf for the items
135  * @dst_item:	the item nr we're copying into
136  * @src_item:	the item nr we're copying from
137  * @nr_items:	the number of items to copy
138  *
139  * Wrapper around copy_extent_buffer() that does the math to get the
140  * appropriate offsets into the leaf from the item numbers.
141  */
copy_leaf_items(const struct extent_buffer * dst,const struct extent_buffer * src,int dst_item,int src_item,int nr_items)142 static inline void copy_leaf_items(const struct extent_buffer *dst,
143 				   const struct extent_buffer *src,
144 				   int dst_item, int src_item, int nr_items)
145 {
146 	copy_extent_buffer(dst, src, btrfs_item_nr_offset(dst, dst_item),
147 			      btrfs_item_nr_offset(src, src_item),
148 			      nr_items * sizeof(struct btrfs_item));
149 }
150 
151 /* This exists for btrfs-progs usages. */
btrfs_csum_type_size(u16 type)152 u16 btrfs_csum_type_size(u16 type)
153 {
154 	return btrfs_csums[type].size;
155 }
156 
btrfs_super_csum_size(const struct btrfs_super_block * s)157 int btrfs_super_csum_size(const struct btrfs_super_block *s)
158 {
159 	u16 t = btrfs_super_csum_type(s);
160 	/*
161 	 * csum type is validated at mount time
162 	 */
163 	return btrfs_csum_type_size(t);
164 }
165 
btrfs_super_csum_name(u16 csum_type)166 const char *btrfs_super_csum_name(u16 csum_type)
167 {
168 	/* csum type is validated at mount time */
169 	return btrfs_csums[csum_type].name;
170 }
171 
172 /*
173  * Return driver name if defined, otherwise the name that's also a valid driver
174  * name
175  */
btrfs_super_csum_driver(u16 csum_type)176 const char *btrfs_super_csum_driver(u16 csum_type)
177 {
178 	/* csum type is validated at mount time */
179 	return btrfs_csums[csum_type].driver[0] ?
180 		btrfs_csums[csum_type].driver :
181 		btrfs_csums[csum_type].name;
182 }
183 
btrfs_get_num_csums(void)184 size_t __attribute_const__ btrfs_get_num_csums(void)
185 {
186 	return ARRAY_SIZE(btrfs_csums);
187 }
188 
btrfs_alloc_path(void)189 struct btrfs_path *btrfs_alloc_path(void)
190 {
191 	might_sleep();
192 
193 	return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
194 }
195 
196 /* this also releases the path */
btrfs_free_path(struct btrfs_path * p)197 void btrfs_free_path(struct btrfs_path *p)
198 {
199 	if (!p)
200 		return;
201 	btrfs_release_path(p);
202 	kmem_cache_free(btrfs_path_cachep, p);
203 }
204 
205 /*
206  * path release drops references on the extent buffers in the path
207  * and it drops any locks held by this path
208  *
209  * It is safe to call this on paths that no locks or extent buffers held.
210  */
btrfs_release_path(struct btrfs_path * p)211 noinline void btrfs_release_path(struct btrfs_path *p)
212 {
213 	int i;
214 
215 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
216 		p->slots[i] = 0;
217 		if (!p->nodes[i])
218 			continue;
219 		if (p->locks[i]) {
220 			btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
221 			p->locks[i] = 0;
222 		}
223 		free_extent_buffer(p->nodes[i]);
224 		p->nodes[i] = NULL;
225 	}
226 }
227 
228 /*
229  * We want the transaction abort to print stack trace only for errors where the
230  * cause could be a bug, eg. due to ENOSPC, and not for common errors that are
231  * caused by external factors.
232  */
abort_should_print_stack(int errno)233 bool __cold abort_should_print_stack(int errno)
234 {
235 	switch (errno) {
236 	case -EIO:
237 	case -EROFS:
238 	case -ENOMEM:
239 		return false;
240 	}
241 	return true;
242 }
243 
244 /*
245  * safely gets a reference on the root node of a tree.  A lock
246  * is not taken, so a concurrent writer may put a different node
247  * at the root of the tree.  See btrfs_lock_root_node for the
248  * looping required.
249  *
250  * The extent buffer returned by this has a reference taken, so
251  * it won't disappear.  It may stop being the root of the tree
252  * at any time because there are no locks held.
253  */
btrfs_root_node(struct btrfs_root * root)254 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
255 {
256 	struct extent_buffer *eb;
257 
258 	while (1) {
259 		rcu_read_lock();
260 		eb = rcu_dereference(root->node);
261 
262 		/*
263 		 * RCU really hurts here, we could free up the root node because
264 		 * it was COWed but we may not get the new root node yet so do
265 		 * the inc_not_zero dance and if it doesn't work then
266 		 * synchronize_rcu and try again.
267 		 */
268 		if (atomic_inc_not_zero(&eb->refs)) {
269 			rcu_read_unlock();
270 			break;
271 		}
272 		rcu_read_unlock();
273 		synchronize_rcu();
274 	}
275 	return eb;
276 }
277 
278 /*
279  * Cowonly root (not-shareable trees, everything not subvolume or reloc roots),
280  * just get put onto a simple dirty list.  Transaction walks this list to make
281  * sure they get properly updated on disk.
282  */
add_root_to_dirty_list(struct btrfs_root * root)283 static void add_root_to_dirty_list(struct btrfs_root *root)
284 {
285 	struct btrfs_fs_info *fs_info = root->fs_info;
286 
287 	if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
288 	    !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
289 		return;
290 
291 	spin_lock(&fs_info->trans_lock);
292 	if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
293 		/* Want the extent tree to be the last on the list */
294 		if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
295 			list_move_tail(&root->dirty_list,
296 				       &fs_info->dirty_cowonly_roots);
297 		else
298 			list_move(&root->dirty_list,
299 				  &fs_info->dirty_cowonly_roots);
300 	}
301 	spin_unlock(&fs_info->trans_lock);
302 }
303 
304 /*
305  * used by snapshot creation to make a copy of a root for a tree with
306  * a given objectid.  The buffer with the new root node is returned in
307  * cow_ret, and this func returns zero on success or a negative error code.
308  */
btrfs_copy_root(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,struct extent_buffer ** cow_ret,u64 new_root_objectid)309 int btrfs_copy_root(struct btrfs_trans_handle *trans,
310 		      struct btrfs_root *root,
311 		      struct extent_buffer *buf,
312 		      struct extent_buffer **cow_ret, u64 new_root_objectid)
313 {
314 	struct btrfs_fs_info *fs_info = root->fs_info;
315 	struct extent_buffer *cow;
316 	int ret = 0;
317 	int level;
318 	struct btrfs_disk_key disk_key;
319 
320 	WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
321 		trans->transid != fs_info->running_transaction->transid);
322 	WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
323 		trans->transid != root->last_trans);
324 
325 	level = btrfs_header_level(buf);
326 	if (level == 0)
327 		btrfs_item_key(buf, &disk_key, 0);
328 	else
329 		btrfs_node_key(buf, &disk_key, 0);
330 
331 	cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
332 				     &disk_key, level, buf->start, 0,
333 				     BTRFS_NESTING_NEW_ROOT);
334 	if (IS_ERR(cow))
335 		return PTR_ERR(cow);
336 
337 	copy_extent_buffer_full(cow, buf);
338 	btrfs_set_header_bytenr(cow, cow->start);
339 	btrfs_set_header_generation(cow, trans->transid);
340 	btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
341 	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
342 				     BTRFS_HEADER_FLAG_RELOC);
343 	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
344 		btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
345 	else
346 		btrfs_set_header_owner(cow, new_root_objectid);
347 
348 	write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
349 
350 	WARN_ON(btrfs_header_generation(buf) > trans->transid);
351 	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
352 		ret = btrfs_inc_ref(trans, root, cow, 1);
353 	else
354 		ret = btrfs_inc_ref(trans, root, cow, 0);
355 	if (ret) {
356 		btrfs_tree_unlock(cow);
357 		free_extent_buffer(cow);
358 		btrfs_abort_transaction(trans, ret);
359 		return ret;
360 	}
361 
362 	btrfs_mark_buffer_dirty(trans, cow);
363 	*cow_ret = cow;
364 	return 0;
365 }
366 
367 /*
368  * check if the tree block can be shared by multiple trees
369  */
btrfs_block_can_be_shared(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf)370 int btrfs_block_can_be_shared(struct btrfs_trans_handle *trans,
371 			      struct btrfs_root *root,
372 			      struct extent_buffer *buf)
373 {
374 	/*
375 	 * Tree blocks not in shareable trees and tree roots are never shared.
376 	 * If a block was allocated after the last snapshot and the block was
377 	 * not allocated by tree relocation, we know the block is not shared.
378 	 */
379 	if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
380 	    buf != root->node &&
381 	    (btrfs_header_generation(buf) <=
382 	     btrfs_root_last_snapshot(&root->root_item) ||
383 	     btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) {
384 		if (buf != root->commit_root)
385 			return 1;
386 		/*
387 		 * An extent buffer that used to be the commit root may still be
388 		 * shared because the tree height may have increased and it
389 		 * became a child of a higher level root. This can happen when
390 		 * snapshotting a subvolume created in the current transaction.
391 		 */
392 		if (btrfs_header_generation(buf) == trans->transid)
393 			return 1;
394 	}
395 
396 	return 0;
397 }
398 
update_ref_for_cow(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,struct extent_buffer * cow,int * last_ref)399 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
400 				       struct btrfs_root *root,
401 				       struct extent_buffer *buf,
402 				       struct extent_buffer *cow,
403 				       int *last_ref)
404 {
405 	struct btrfs_fs_info *fs_info = root->fs_info;
406 	u64 refs;
407 	u64 owner;
408 	u64 flags;
409 	u64 new_flags = 0;
410 	int ret;
411 
412 	/*
413 	 * Backrefs update rules:
414 	 *
415 	 * Always use full backrefs for extent pointers in tree block
416 	 * allocated by tree relocation.
417 	 *
418 	 * If a shared tree block is no longer referenced by its owner
419 	 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
420 	 * use full backrefs for extent pointers in tree block.
421 	 *
422 	 * If a tree block is been relocating
423 	 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
424 	 * use full backrefs for extent pointers in tree block.
425 	 * The reason for this is some operations (such as drop tree)
426 	 * are only allowed for blocks use full backrefs.
427 	 */
428 
429 	if (btrfs_block_can_be_shared(trans, root, buf)) {
430 		ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
431 					       btrfs_header_level(buf), 1,
432 					       &refs, &flags);
433 		if (ret)
434 			return ret;
435 		if (unlikely(refs == 0)) {
436 			btrfs_crit(fs_info,
437 		"found 0 references for tree block at bytenr %llu level %d root %llu",
438 				   buf->start, btrfs_header_level(buf),
439 				   btrfs_root_id(root));
440 			ret = -EUCLEAN;
441 			btrfs_abort_transaction(trans, ret);
442 			return ret;
443 		}
444 	} else {
445 		refs = 1;
446 		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
447 		    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
448 			flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
449 		else
450 			flags = 0;
451 	}
452 
453 	owner = btrfs_header_owner(buf);
454 	if (unlikely(owner == BTRFS_TREE_RELOC_OBJECTID &&
455 		     !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))) {
456 		btrfs_crit(fs_info,
457 "found tree block at bytenr %llu level %d root %llu refs %llu flags %llx without full backref flag set",
458 			   buf->start, btrfs_header_level(buf),
459 			   btrfs_root_id(root), refs, flags);
460 		ret = -EUCLEAN;
461 		btrfs_abort_transaction(trans, ret);
462 		return ret;
463 	}
464 
465 	if (refs > 1) {
466 		if ((owner == root->root_key.objectid ||
467 		     root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
468 		    !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
469 			ret = btrfs_inc_ref(trans, root, buf, 1);
470 			if (ret)
471 				return ret;
472 
473 			if (root->root_key.objectid ==
474 			    BTRFS_TREE_RELOC_OBJECTID) {
475 				ret = btrfs_dec_ref(trans, root, buf, 0);
476 				if (ret)
477 					return ret;
478 				ret = btrfs_inc_ref(trans, root, cow, 1);
479 				if (ret)
480 					return ret;
481 			}
482 			new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
483 		} else {
484 
485 			if (root->root_key.objectid ==
486 			    BTRFS_TREE_RELOC_OBJECTID)
487 				ret = btrfs_inc_ref(trans, root, cow, 1);
488 			else
489 				ret = btrfs_inc_ref(trans, root, cow, 0);
490 			if (ret)
491 				return ret;
492 		}
493 		if (new_flags != 0) {
494 			ret = btrfs_set_disk_extent_flags(trans, buf, new_flags);
495 			if (ret)
496 				return ret;
497 		}
498 	} else {
499 		if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
500 			if (root->root_key.objectid ==
501 			    BTRFS_TREE_RELOC_OBJECTID)
502 				ret = btrfs_inc_ref(trans, root, cow, 1);
503 			else
504 				ret = btrfs_inc_ref(trans, root, cow, 0);
505 			if (ret)
506 				return ret;
507 			ret = btrfs_dec_ref(trans, root, buf, 1);
508 			if (ret)
509 				return ret;
510 		}
511 		btrfs_clear_buffer_dirty(trans, buf);
512 		*last_ref = 1;
513 	}
514 	return 0;
515 }
516 
517 /*
518  * does the dirty work in cow of a single block.  The parent block (if
519  * supplied) is updated to point to the new cow copy.  The new buffer is marked
520  * dirty and returned locked.  If you modify the block it needs to be marked
521  * dirty again.
522  *
523  * search_start -- an allocation hint for the new block
524  *
525  * empty_size -- a hint that you plan on doing more cow.  This is the size in
526  * bytes the allocator should try to find free next to the block it returns.
527  * This is just a hint and may be ignored by the allocator.
528  */
__btrfs_cow_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,struct extent_buffer * parent,int parent_slot,struct extent_buffer ** cow_ret,u64 search_start,u64 empty_size,enum btrfs_lock_nesting nest)529 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
530 			     struct btrfs_root *root,
531 			     struct extent_buffer *buf,
532 			     struct extent_buffer *parent, int parent_slot,
533 			     struct extent_buffer **cow_ret,
534 			     u64 search_start, u64 empty_size,
535 			     enum btrfs_lock_nesting nest)
536 {
537 	struct btrfs_fs_info *fs_info = root->fs_info;
538 	struct btrfs_disk_key disk_key;
539 	struct extent_buffer *cow;
540 	int level, ret;
541 	int last_ref = 0;
542 	int unlock_orig = 0;
543 	u64 parent_start = 0;
544 
545 	if (*cow_ret == buf)
546 		unlock_orig = 1;
547 
548 	btrfs_assert_tree_write_locked(buf);
549 
550 	WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
551 		trans->transid != fs_info->running_transaction->transid);
552 	WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
553 		trans->transid != root->last_trans);
554 
555 	level = btrfs_header_level(buf);
556 
557 	if (level == 0)
558 		btrfs_item_key(buf, &disk_key, 0);
559 	else
560 		btrfs_node_key(buf, &disk_key, 0);
561 
562 	if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
563 		parent_start = parent->start;
564 
565 	cow = btrfs_alloc_tree_block(trans, root, parent_start,
566 				     root->root_key.objectid, &disk_key, level,
567 				     search_start, empty_size, nest);
568 	if (IS_ERR(cow))
569 		return PTR_ERR(cow);
570 
571 	/* cow is set to blocking by btrfs_init_new_buffer */
572 
573 	copy_extent_buffer_full(cow, buf);
574 	btrfs_set_header_bytenr(cow, cow->start);
575 	btrfs_set_header_generation(cow, trans->transid);
576 	btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
577 	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
578 				     BTRFS_HEADER_FLAG_RELOC);
579 	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
580 		btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
581 	else
582 		btrfs_set_header_owner(cow, root->root_key.objectid);
583 
584 	write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
585 
586 	ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
587 	if (ret) {
588 		btrfs_tree_unlock(cow);
589 		free_extent_buffer(cow);
590 		btrfs_abort_transaction(trans, ret);
591 		return ret;
592 	}
593 
594 	if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
595 		ret = btrfs_reloc_cow_block(trans, root, buf, cow);
596 		if (ret) {
597 			btrfs_tree_unlock(cow);
598 			free_extent_buffer(cow);
599 			btrfs_abort_transaction(trans, ret);
600 			return ret;
601 		}
602 	}
603 
604 	if (buf == root->node) {
605 		WARN_ON(parent && parent != buf);
606 		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
607 		    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
608 			parent_start = buf->start;
609 
610 		ret = btrfs_tree_mod_log_insert_root(root->node, cow, true);
611 		if (ret < 0) {
612 			btrfs_tree_unlock(cow);
613 			free_extent_buffer(cow);
614 			btrfs_abort_transaction(trans, ret);
615 			return ret;
616 		}
617 		atomic_inc(&cow->refs);
618 		rcu_assign_pointer(root->node, cow);
619 
620 		btrfs_free_tree_block(trans, btrfs_root_id(root), buf,
621 				      parent_start, last_ref);
622 		free_extent_buffer(buf);
623 		add_root_to_dirty_list(root);
624 	} else {
625 		WARN_ON(trans->transid != btrfs_header_generation(parent));
626 		ret = btrfs_tree_mod_log_insert_key(parent, parent_slot,
627 						    BTRFS_MOD_LOG_KEY_REPLACE);
628 		if (ret) {
629 			btrfs_tree_unlock(cow);
630 			free_extent_buffer(cow);
631 			btrfs_abort_transaction(trans, ret);
632 			return ret;
633 		}
634 		btrfs_set_node_blockptr(parent, parent_slot,
635 					cow->start);
636 		btrfs_set_node_ptr_generation(parent, parent_slot,
637 					      trans->transid);
638 		btrfs_mark_buffer_dirty(trans, parent);
639 		if (last_ref) {
640 			ret = btrfs_tree_mod_log_free_eb(buf);
641 			if (ret) {
642 				btrfs_tree_unlock(cow);
643 				free_extent_buffer(cow);
644 				btrfs_abort_transaction(trans, ret);
645 				return ret;
646 			}
647 		}
648 		btrfs_free_tree_block(trans, btrfs_root_id(root), buf,
649 				      parent_start, last_ref);
650 	}
651 	if (unlock_orig)
652 		btrfs_tree_unlock(buf);
653 	free_extent_buffer_stale(buf);
654 	btrfs_mark_buffer_dirty(trans, cow);
655 	*cow_ret = cow;
656 	return 0;
657 }
658 
should_cow_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf)659 static inline int should_cow_block(struct btrfs_trans_handle *trans,
660 				   struct btrfs_root *root,
661 				   struct extent_buffer *buf)
662 {
663 	if (btrfs_is_testing(root->fs_info))
664 		return 0;
665 
666 	/* Ensure we can see the FORCE_COW bit */
667 	smp_mb__before_atomic();
668 
669 	/*
670 	 * We do not need to cow a block if
671 	 * 1) this block is not created or changed in this transaction;
672 	 * 2) this block does not belong to TREE_RELOC tree;
673 	 * 3) the root is not forced COW.
674 	 *
675 	 * What is forced COW:
676 	 *    when we create snapshot during committing the transaction,
677 	 *    after we've finished copying src root, we must COW the shared
678 	 *    block to ensure the metadata consistency.
679 	 */
680 	if (btrfs_header_generation(buf) == trans->transid &&
681 	    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
682 	    !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
683 	      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
684 	    !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
685 		return 0;
686 	return 1;
687 }
688 
689 /*
690  * cows a single block, see __btrfs_cow_block for the real work.
691  * This version of it has extra checks so that a block isn't COWed more than
692  * once per transaction, as long as it hasn't been written yet
693  */
btrfs_cow_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,struct extent_buffer * parent,int parent_slot,struct extent_buffer ** cow_ret,enum btrfs_lock_nesting nest)694 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
695 		    struct btrfs_root *root, struct extent_buffer *buf,
696 		    struct extent_buffer *parent, int parent_slot,
697 		    struct extent_buffer **cow_ret,
698 		    enum btrfs_lock_nesting nest)
699 {
700 	struct btrfs_fs_info *fs_info = root->fs_info;
701 	u64 search_start;
702 	int ret;
703 
704 	if (unlikely(test_bit(BTRFS_ROOT_DELETING, &root->state))) {
705 		btrfs_abort_transaction(trans, -EUCLEAN);
706 		btrfs_crit(fs_info,
707 		   "attempt to COW block %llu on root %llu that is being deleted",
708 			   buf->start, btrfs_root_id(root));
709 		return -EUCLEAN;
710 	}
711 
712 	/*
713 	 * COWing must happen through a running transaction, which always
714 	 * matches the current fs generation (it's a transaction with a state
715 	 * less than TRANS_STATE_UNBLOCKED). If it doesn't, then turn the fs
716 	 * into error state to prevent the commit of any transaction.
717 	 */
718 	if (unlikely(trans->transaction != fs_info->running_transaction ||
719 		     trans->transid != fs_info->generation)) {
720 		btrfs_abort_transaction(trans, -EUCLEAN);
721 		btrfs_crit(fs_info,
722 "unexpected transaction when attempting to COW block %llu on root %llu, transaction %llu running transaction %llu fs generation %llu",
723 			   buf->start, btrfs_root_id(root), trans->transid,
724 			   fs_info->running_transaction->transid,
725 			   fs_info->generation);
726 		return -EUCLEAN;
727 	}
728 
729 	if (!should_cow_block(trans, root, buf)) {
730 		*cow_ret = buf;
731 		return 0;
732 	}
733 
734 	search_start = buf->start & ~((u64)SZ_1G - 1);
735 
736 	/*
737 	 * Before CoWing this block for later modification, check if it's
738 	 * the subtree root and do the delayed subtree trace if needed.
739 	 *
740 	 * Also We don't care about the error, as it's handled internally.
741 	 */
742 	btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
743 	ret = __btrfs_cow_block(trans, root, buf, parent,
744 				 parent_slot, cow_ret, search_start, 0, nest);
745 
746 	trace_btrfs_cow_block(root, buf, *cow_ret);
747 
748 	return ret;
749 }
750 ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO);
751 
752 /*
753  * helper function for defrag to decide if two blocks pointed to by a
754  * node are actually close by
755  */
close_blocks(u64 blocknr,u64 other,u32 blocksize)756 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
757 {
758 	if (blocknr < other && other - (blocknr + blocksize) < 32768)
759 		return 1;
760 	if (blocknr > other && blocknr - (other + blocksize) < 32768)
761 		return 1;
762 	return 0;
763 }
764 
765 #ifdef __LITTLE_ENDIAN
766 
767 /*
768  * Compare two keys, on little-endian the disk order is same as CPU order and
769  * we can avoid the conversion.
770  */
comp_keys(const struct btrfs_disk_key * disk_key,const struct btrfs_key * k2)771 static int comp_keys(const struct btrfs_disk_key *disk_key,
772 		     const struct btrfs_key *k2)
773 {
774 	const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key;
775 
776 	return btrfs_comp_cpu_keys(k1, k2);
777 }
778 
779 #else
780 
781 /*
782  * compare two keys in a memcmp fashion
783  */
comp_keys(const struct btrfs_disk_key * disk,const struct btrfs_key * k2)784 static int comp_keys(const struct btrfs_disk_key *disk,
785 		     const struct btrfs_key *k2)
786 {
787 	struct btrfs_key k1;
788 
789 	btrfs_disk_key_to_cpu(&k1, disk);
790 
791 	return btrfs_comp_cpu_keys(&k1, k2);
792 }
793 #endif
794 
795 /*
796  * same as comp_keys only with two btrfs_key's
797  */
btrfs_comp_cpu_keys(const struct btrfs_key * k1,const struct btrfs_key * k2)798 int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
799 {
800 	if (k1->objectid > k2->objectid)
801 		return 1;
802 	if (k1->objectid < k2->objectid)
803 		return -1;
804 	if (k1->type > k2->type)
805 		return 1;
806 	if (k1->type < k2->type)
807 		return -1;
808 	if (k1->offset > k2->offset)
809 		return 1;
810 	if (k1->offset < k2->offset)
811 		return -1;
812 	return 0;
813 }
814 
815 /*
816  * this is used by the defrag code to go through all the
817  * leaves pointed to by a node and reallocate them so that
818  * disk order is close to key order
819  */
btrfs_realloc_node(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * parent,int start_slot,u64 * last_ret,struct btrfs_key * progress)820 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
821 		       struct btrfs_root *root, struct extent_buffer *parent,
822 		       int start_slot, u64 *last_ret,
823 		       struct btrfs_key *progress)
824 {
825 	struct btrfs_fs_info *fs_info = root->fs_info;
826 	struct extent_buffer *cur;
827 	u64 blocknr;
828 	u64 search_start = *last_ret;
829 	u64 last_block = 0;
830 	u64 other;
831 	u32 parent_nritems;
832 	int end_slot;
833 	int i;
834 	int err = 0;
835 	u32 blocksize;
836 	int progress_passed = 0;
837 	struct btrfs_disk_key disk_key;
838 
839 	/*
840 	 * COWing must happen through a running transaction, which always
841 	 * matches the current fs generation (it's a transaction with a state
842 	 * less than TRANS_STATE_UNBLOCKED). If it doesn't, then turn the fs
843 	 * into error state to prevent the commit of any transaction.
844 	 */
845 	if (unlikely(trans->transaction != fs_info->running_transaction ||
846 		     trans->transid != fs_info->generation)) {
847 		btrfs_abort_transaction(trans, -EUCLEAN);
848 		btrfs_crit(fs_info,
849 "unexpected transaction when attempting to reallocate parent %llu for root %llu, transaction %llu running transaction %llu fs generation %llu",
850 			   parent->start, btrfs_root_id(root), trans->transid,
851 			   fs_info->running_transaction->transid,
852 			   fs_info->generation);
853 		return -EUCLEAN;
854 	}
855 
856 	parent_nritems = btrfs_header_nritems(parent);
857 	blocksize = fs_info->nodesize;
858 	end_slot = parent_nritems - 1;
859 
860 	if (parent_nritems <= 1)
861 		return 0;
862 
863 	for (i = start_slot; i <= end_slot; i++) {
864 		int close = 1;
865 
866 		btrfs_node_key(parent, &disk_key, i);
867 		if (!progress_passed && comp_keys(&disk_key, progress) < 0)
868 			continue;
869 
870 		progress_passed = 1;
871 		blocknr = btrfs_node_blockptr(parent, i);
872 		if (last_block == 0)
873 			last_block = blocknr;
874 
875 		if (i > 0) {
876 			other = btrfs_node_blockptr(parent, i - 1);
877 			close = close_blocks(blocknr, other, blocksize);
878 		}
879 		if (!close && i < end_slot) {
880 			other = btrfs_node_blockptr(parent, i + 1);
881 			close = close_blocks(blocknr, other, blocksize);
882 		}
883 		if (close) {
884 			last_block = blocknr;
885 			continue;
886 		}
887 
888 		cur = btrfs_read_node_slot(parent, i);
889 		if (IS_ERR(cur))
890 			return PTR_ERR(cur);
891 		if (search_start == 0)
892 			search_start = last_block;
893 
894 		btrfs_tree_lock(cur);
895 		err = __btrfs_cow_block(trans, root, cur, parent, i,
896 					&cur, search_start,
897 					min(16 * blocksize,
898 					    (end_slot - i) * blocksize),
899 					BTRFS_NESTING_COW);
900 		if (err) {
901 			btrfs_tree_unlock(cur);
902 			free_extent_buffer(cur);
903 			break;
904 		}
905 		search_start = cur->start;
906 		last_block = cur->start;
907 		*last_ret = search_start;
908 		btrfs_tree_unlock(cur);
909 		free_extent_buffer(cur);
910 	}
911 	return err;
912 }
913 
914 /*
915  * Search for a key in the given extent_buffer.
916  *
917  * The lower boundary for the search is specified by the slot number @first_slot.
918  * Use a value of 0 to search over the whole extent buffer. Works for both
919  * leaves and nodes.
920  *
921  * The slot in the extent buffer is returned via @slot. If the key exists in the
922  * extent buffer, then @slot will point to the slot where the key is, otherwise
923  * it points to the slot where you would insert the key.
924  *
925  * Slot may point to the total number of items (i.e. one position beyond the last
926  * key) if the key is bigger than the last key in the extent buffer.
927  */
btrfs_bin_search(struct extent_buffer * eb,int first_slot,const struct btrfs_key * key,int * slot)928 int btrfs_bin_search(struct extent_buffer *eb, int first_slot,
929 		     const struct btrfs_key *key, int *slot)
930 {
931 	unsigned long p;
932 	int item_size;
933 	/*
934 	 * Use unsigned types for the low and high slots, so that we get a more
935 	 * efficient division in the search loop below.
936 	 */
937 	u32 low = first_slot;
938 	u32 high = btrfs_header_nritems(eb);
939 	int ret;
940 	const int key_size = sizeof(struct btrfs_disk_key);
941 
942 	if (unlikely(low > high)) {
943 		btrfs_err(eb->fs_info,
944 		 "%s: low (%u) > high (%u) eb %llu owner %llu level %d",
945 			  __func__, low, high, eb->start,
946 			  btrfs_header_owner(eb), btrfs_header_level(eb));
947 		return -EINVAL;
948 	}
949 
950 	if (btrfs_header_level(eb) == 0) {
951 		p = offsetof(struct btrfs_leaf, items);
952 		item_size = sizeof(struct btrfs_item);
953 	} else {
954 		p = offsetof(struct btrfs_node, ptrs);
955 		item_size = sizeof(struct btrfs_key_ptr);
956 	}
957 
958 	while (low < high) {
959 		unsigned long oip;
960 		unsigned long offset;
961 		struct btrfs_disk_key *tmp;
962 		struct btrfs_disk_key unaligned;
963 		int mid;
964 
965 		mid = (low + high) / 2;
966 		offset = p + mid * item_size;
967 		oip = offset_in_page(offset);
968 
969 		if (oip + key_size <= PAGE_SIZE) {
970 			const unsigned long idx = get_eb_page_index(offset);
971 			char *kaddr = page_address(eb->pages[idx]);
972 
973 			oip = get_eb_offset_in_page(eb, offset);
974 			tmp = (struct btrfs_disk_key *)(kaddr + oip);
975 		} else {
976 			read_extent_buffer(eb, &unaligned, offset, key_size);
977 			tmp = &unaligned;
978 		}
979 
980 		ret = comp_keys(tmp, key);
981 
982 		if (ret < 0)
983 			low = mid + 1;
984 		else if (ret > 0)
985 			high = mid;
986 		else {
987 			*slot = mid;
988 			return 0;
989 		}
990 	}
991 	*slot = low;
992 	return 1;
993 }
994 
root_add_used(struct btrfs_root * root,u32 size)995 static void root_add_used(struct btrfs_root *root, u32 size)
996 {
997 	spin_lock(&root->accounting_lock);
998 	btrfs_set_root_used(&root->root_item,
999 			    btrfs_root_used(&root->root_item) + size);
1000 	spin_unlock(&root->accounting_lock);
1001 }
1002 
root_sub_used(struct btrfs_root * root,u32 size)1003 static void root_sub_used(struct btrfs_root *root, u32 size)
1004 {
1005 	spin_lock(&root->accounting_lock);
1006 	btrfs_set_root_used(&root->root_item,
1007 			    btrfs_root_used(&root->root_item) - size);
1008 	spin_unlock(&root->accounting_lock);
1009 }
1010 
1011 /* given a node and slot number, this reads the blocks it points to.  The
1012  * extent buffer is returned with a reference taken (but unlocked).
1013  */
btrfs_read_node_slot(struct extent_buffer * parent,int slot)1014 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
1015 					   int slot)
1016 {
1017 	int level = btrfs_header_level(parent);
1018 	struct btrfs_tree_parent_check check = { 0 };
1019 	struct extent_buffer *eb;
1020 
1021 	if (slot < 0 || slot >= btrfs_header_nritems(parent))
1022 		return ERR_PTR(-ENOENT);
1023 
1024 	ASSERT(level);
1025 
1026 	check.level = level - 1;
1027 	check.transid = btrfs_node_ptr_generation(parent, slot);
1028 	check.owner_root = btrfs_header_owner(parent);
1029 	check.has_first_key = true;
1030 	btrfs_node_key_to_cpu(parent, &check.first_key, slot);
1031 
1032 	eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
1033 			     &check);
1034 	if (IS_ERR(eb))
1035 		return eb;
1036 	if (!extent_buffer_uptodate(eb)) {
1037 		free_extent_buffer(eb);
1038 		return ERR_PTR(-EIO);
1039 	}
1040 
1041 	return eb;
1042 }
1043 
1044 /*
1045  * node level balancing, used to make sure nodes are in proper order for
1046  * item deletion.  We balance from the top down, so we have to make sure
1047  * that a deletion won't leave an node completely empty later on.
1048  */
balance_level(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int level)1049 static noinline int balance_level(struct btrfs_trans_handle *trans,
1050 			 struct btrfs_root *root,
1051 			 struct btrfs_path *path, int level)
1052 {
1053 	struct btrfs_fs_info *fs_info = root->fs_info;
1054 	struct extent_buffer *right = NULL;
1055 	struct extent_buffer *mid;
1056 	struct extent_buffer *left = NULL;
1057 	struct extent_buffer *parent = NULL;
1058 	int ret = 0;
1059 	int wret;
1060 	int pslot;
1061 	int orig_slot = path->slots[level];
1062 	u64 orig_ptr;
1063 
1064 	ASSERT(level > 0);
1065 
1066 	mid = path->nodes[level];
1067 
1068 	WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK);
1069 	WARN_ON(btrfs_header_generation(mid) != trans->transid);
1070 
1071 	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1072 
1073 	if (level < BTRFS_MAX_LEVEL - 1) {
1074 		parent = path->nodes[level + 1];
1075 		pslot = path->slots[level + 1];
1076 	}
1077 
1078 	/*
1079 	 * deal with the case where there is only one pointer in the root
1080 	 * by promoting the node below to a root
1081 	 */
1082 	if (!parent) {
1083 		struct extent_buffer *child;
1084 
1085 		if (btrfs_header_nritems(mid) != 1)
1086 			return 0;
1087 
1088 		/* promote the child to a root */
1089 		child = btrfs_read_node_slot(mid, 0);
1090 		if (IS_ERR(child)) {
1091 			ret = PTR_ERR(child);
1092 			goto out;
1093 		}
1094 
1095 		btrfs_tree_lock(child);
1096 		ret = btrfs_cow_block(trans, root, child, mid, 0, &child,
1097 				      BTRFS_NESTING_COW);
1098 		if (ret) {
1099 			btrfs_tree_unlock(child);
1100 			free_extent_buffer(child);
1101 			goto out;
1102 		}
1103 
1104 		ret = btrfs_tree_mod_log_insert_root(root->node, child, true);
1105 		if (ret < 0) {
1106 			btrfs_tree_unlock(child);
1107 			free_extent_buffer(child);
1108 			btrfs_abort_transaction(trans, ret);
1109 			goto out;
1110 		}
1111 		rcu_assign_pointer(root->node, child);
1112 
1113 		add_root_to_dirty_list(root);
1114 		btrfs_tree_unlock(child);
1115 
1116 		path->locks[level] = 0;
1117 		path->nodes[level] = NULL;
1118 		btrfs_clear_buffer_dirty(trans, mid);
1119 		btrfs_tree_unlock(mid);
1120 		/* once for the path */
1121 		free_extent_buffer(mid);
1122 
1123 		root_sub_used(root, mid->len);
1124 		btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1);
1125 		/* once for the root ptr */
1126 		free_extent_buffer_stale(mid);
1127 		return 0;
1128 	}
1129 	if (btrfs_header_nritems(mid) >
1130 	    BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1131 		return 0;
1132 
1133 	if (pslot) {
1134 		left = btrfs_read_node_slot(parent, pslot - 1);
1135 		if (IS_ERR(left)) {
1136 			ret = PTR_ERR(left);
1137 			left = NULL;
1138 			goto out;
1139 		}
1140 
1141 		__btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
1142 		wret = btrfs_cow_block(trans, root, left,
1143 				       parent, pslot - 1, &left,
1144 				       BTRFS_NESTING_LEFT_COW);
1145 		if (wret) {
1146 			ret = wret;
1147 			goto out;
1148 		}
1149 	}
1150 
1151 	if (pslot + 1 < btrfs_header_nritems(parent)) {
1152 		right = btrfs_read_node_slot(parent, pslot + 1);
1153 		if (IS_ERR(right)) {
1154 			ret = PTR_ERR(right);
1155 			right = NULL;
1156 			goto out;
1157 		}
1158 
1159 		__btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
1160 		wret = btrfs_cow_block(trans, root, right,
1161 				       parent, pslot + 1, &right,
1162 				       BTRFS_NESTING_RIGHT_COW);
1163 		if (wret) {
1164 			ret = wret;
1165 			goto out;
1166 		}
1167 	}
1168 
1169 	/* first, try to make some room in the middle buffer */
1170 	if (left) {
1171 		orig_slot += btrfs_header_nritems(left);
1172 		wret = push_node_left(trans, left, mid, 1);
1173 		if (wret < 0)
1174 			ret = wret;
1175 	}
1176 
1177 	/*
1178 	 * then try to empty the right most buffer into the middle
1179 	 */
1180 	if (right) {
1181 		wret = push_node_left(trans, mid, right, 1);
1182 		if (wret < 0 && wret != -ENOSPC)
1183 			ret = wret;
1184 		if (btrfs_header_nritems(right) == 0) {
1185 			btrfs_clear_buffer_dirty(trans, right);
1186 			btrfs_tree_unlock(right);
1187 			ret = btrfs_del_ptr(trans, root, path, level + 1, pslot + 1);
1188 			if (ret < 0) {
1189 				free_extent_buffer_stale(right);
1190 				right = NULL;
1191 				goto out;
1192 			}
1193 			root_sub_used(root, right->len);
1194 			btrfs_free_tree_block(trans, btrfs_root_id(root), right,
1195 					      0, 1);
1196 			free_extent_buffer_stale(right);
1197 			right = NULL;
1198 		} else {
1199 			struct btrfs_disk_key right_key;
1200 			btrfs_node_key(right, &right_key, 0);
1201 			ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1,
1202 					BTRFS_MOD_LOG_KEY_REPLACE);
1203 			if (ret < 0) {
1204 				btrfs_abort_transaction(trans, ret);
1205 				goto out;
1206 			}
1207 			btrfs_set_node_key(parent, &right_key, pslot + 1);
1208 			btrfs_mark_buffer_dirty(trans, parent);
1209 		}
1210 	}
1211 	if (btrfs_header_nritems(mid) == 1) {
1212 		/*
1213 		 * we're not allowed to leave a node with one item in the
1214 		 * tree during a delete.  A deletion from lower in the tree
1215 		 * could try to delete the only pointer in this node.
1216 		 * So, pull some keys from the left.
1217 		 * There has to be a left pointer at this point because
1218 		 * otherwise we would have pulled some pointers from the
1219 		 * right
1220 		 */
1221 		if (unlikely(!left)) {
1222 			btrfs_crit(fs_info,
1223 "missing left child when middle child only has 1 item, parent bytenr %llu level %d mid bytenr %llu root %llu",
1224 				   parent->start, btrfs_header_level(parent),
1225 				   mid->start, btrfs_root_id(root));
1226 			ret = -EUCLEAN;
1227 			btrfs_abort_transaction(trans, ret);
1228 			goto out;
1229 		}
1230 		wret = balance_node_right(trans, mid, left);
1231 		if (wret < 0) {
1232 			ret = wret;
1233 			goto out;
1234 		}
1235 		if (wret == 1) {
1236 			wret = push_node_left(trans, left, mid, 1);
1237 			if (wret < 0)
1238 				ret = wret;
1239 		}
1240 		BUG_ON(wret == 1);
1241 	}
1242 	if (btrfs_header_nritems(mid) == 0) {
1243 		btrfs_clear_buffer_dirty(trans, mid);
1244 		btrfs_tree_unlock(mid);
1245 		ret = btrfs_del_ptr(trans, root, path, level + 1, pslot);
1246 		if (ret < 0) {
1247 			free_extent_buffer_stale(mid);
1248 			mid = NULL;
1249 			goto out;
1250 		}
1251 		root_sub_used(root, mid->len);
1252 		btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1);
1253 		free_extent_buffer_stale(mid);
1254 		mid = NULL;
1255 	} else {
1256 		/* update the parent key to reflect our changes */
1257 		struct btrfs_disk_key mid_key;
1258 		btrfs_node_key(mid, &mid_key, 0);
1259 		ret = btrfs_tree_mod_log_insert_key(parent, pslot,
1260 						    BTRFS_MOD_LOG_KEY_REPLACE);
1261 		if (ret < 0) {
1262 			btrfs_abort_transaction(trans, ret);
1263 			goto out;
1264 		}
1265 		btrfs_set_node_key(parent, &mid_key, pslot);
1266 		btrfs_mark_buffer_dirty(trans, parent);
1267 	}
1268 
1269 	/* update the path */
1270 	if (left) {
1271 		if (btrfs_header_nritems(left) > orig_slot) {
1272 			atomic_inc(&left->refs);
1273 			/* left was locked after cow */
1274 			path->nodes[level] = left;
1275 			path->slots[level + 1] -= 1;
1276 			path->slots[level] = orig_slot;
1277 			if (mid) {
1278 				btrfs_tree_unlock(mid);
1279 				free_extent_buffer(mid);
1280 			}
1281 		} else {
1282 			orig_slot -= btrfs_header_nritems(left);
1283 			path->slots[level] = orig_slot;
1284 		}
1285 	}
1286 	/* double check we haven't messed things up */
1287 	if (orig_ptr !=
1288 	    btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1289 		BUG();
1290 out:
1291 	if (right) {
1292 		btrfs_tree_unlock(right);
1293 		free_extent_buffer(right);
1294 	}
1295 	if (left) {
1296 		if (path->nodes[level] != left)
1297 			btrfs_tree_unlock(left);
1298 		free_extent_buffer(left);
1299 	}
1300 	return ret;
1301 }
1302 
1303 /* Node balancing for insertion.  Here we only split or push nodes around
1304  * when they are completely full.  This is also done top down, so we
1305  * have to be pessimistic.
1306  */
push_nodes_for_insert(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int level)1307 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1308 					  struct btrfs_root *root,
1309 					  struct btrfs_path *path, int level)
1310 {
1311 	struct btrfs_fs_info *fs_info = root->fs_info;
1312 	struct extent_buffer *right = NULL;
1313 	struct extent_buffer *mid;
1314 	struct extent_buffer *left = NULL;
1315 	struct extent_buffer *parent = NULL;
1316 	int ret = 0;
1317 	int wret;
1318 	int pslot;
1319 	int orig_slot = path->slots[level];
1320 
1321 	if (level == 0)
1322 		return 1;
1323 
1324 	mid = path->nodes[level];
1325 	WARN_ON(btrfs_header_generation(mid) != trans->transid);
1326 
1327 	if (level < BTRFS_MAX_LEVEL - 1) {
1328 		parent = path->nodes[level + 1];
1329 		pslot = path->slots[level + 1];
1330 	}
1331 
1332 	if (!parent)
1333 		return 1;
1334 
1335 	/* first, try to make some room in the middle buffer */
1336 	if (pslot) {
1337 		u32 left_nr;
1338 
1339 		left = btrfs_read_node_slot(parent, pslot - 1);
1340 		if (IS_ERR(left))
1341 			return PTR_ERR(left);
1342 
1343 		__btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
1344 
1345 		left_nr = btrfs_header_nritems(left);
1346 		if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
1347 			wret = 1;
1348 		} else {
1349 			ret = btrfs_cow_block(trans, root, left, parent,
1350 					      pslot - 1, &left,
1351 					      BTRFS_NESTING_LEFT_COW);
1352 			if (ret)
1353 				wret = 1;
1354 			else {
1355 				wret = push_node_left(trans, left, mid, 0);
1356 			}
1357 		}
1358 		if (wret < 0)
1359 			ret = wret;
1360 		if (wret == 0) {
1361 			struct btrfs_disk_key disk_key;
1362 			orig_slot += left_nr;
1363 			btrfs_node_key(mid, &disk_key, 0);
1364 			ret = btrfs_tree_mod_log_insert_key(parent, pslot,
1365 					BTRFS_MOD_LOG_KEY_REPLACE);
1366 			if (ret < 0) {
1367 				btrfs_tree_unlock(left);
1368 				free_extent_buffer(left);
1369 				btrfs_abort_transaction(trans, ret);
1370 				return ret;
1371 			}
1372 			btrfs_set_node_key(parent, &disk_key, pslot);
1373 			btrfs_mark_buffer_dirty(trans, parent);
1374 			if (btrfs_header_nritems(left) > orig_slot) {
1375 				path->nodes[level] = left;
1376 				path->slots[level + 1] -= 1;
1377 				path->slots[level] = orig_slot;
1378 				btrfs_tree_unlock(mid);
1379 				free_extent_buffer(mid);
1380 			} else {
1381 				orig_slot -=
1382 					btrfs_header_nritems(left);
1383 				path->slots[level] = orig_slot;
1384 				btrfs_tree_unlock(left);
1385 				free_extent_buffer(left);
1386 			}
1387 			return 0;
1388 		}
1389 		btrfs_tree_unlock(left);
1390 		free_extent_buffer(left);
1391 	}
1392 
1393 	/*
1394 	 * then try to empty the right most buffer into the middle
1395 	 */
1396 	if (pslot + 1 < btrfs_header_nritems(parent)) {
1397 		u32 right_nr;
1398 
1399 		right = btrfs_read_node_slot(parent, pslot + 1);
1400 		if (IS_ERR(right))
1401 			return PTR_ERR(right);
1402 
1403 		__btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
1404 
1405 		right_nr = btrfs_header_nritems(right);
1406 		if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
1407 			wret = 1;
1408 		} else {
1409 			ret = btrfs_cow_block(trans, root, right,
1410 					      parent, pslot + 1,
1411 					      &right, BTRFS_NESTING_RIGHT_COW);
1412 			if (ret)
1413 				wret = 1;
1414 			else {
1415 				wret = balance_node_right(trans, right, mid);
1416 			}
1417 		}
1418 		if (wret < 0)
1419 			ret = wret;
1420 		if (wret == 0) {
1421 			struct btrfs_disk_key disk_key;
1422 
1423 			btrfs_node_key(right, &disk_key, 0);
1424 			ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1,
1425 					BTRFS_MOD_LOG_KEY_REPLACE);
1426 			if (ret < 0) {
1427 				btrfs_tree_unlock(right);
1428 				free_extent_buffer(right);
1429 				btrfs_abort_transaction(trans, ret);
1430 				return ret;
1431 			}
1432 			btrfs_set_node_key(parent, &disk_key, pslot + 1);
1433 			btrfs_mark_buffer_dirty(trans, parent);
1434 
1435 			if (btrfs_header_nritems(mid) <= orig_slot) {
1436 				path->nodes[level] = right;
1437 				path->slots[level + 1] += 1;
1438 				path->slots[level] = orig_slot -
1439 					btrfs_header_nritems(mid);
1440 				btrfs_tree_unlock(mid);
1441 				free_extent_buffer(mid);
1442 			} else {
1443 				btrfs_tree_unlock(right);
1444 				free_extent_buffer(right);
1445 			}
1446 			return 0;
1447 		}
1448 		btrfs_tree_unlock(right);
1449 		free_extent_buffer(right);
1450 	}
1451 	return 1;
1452 }
1453 
1454 /*
1455  * readahead one full node of leaves, finding things that are close
1456  * to the block in 'slot', and triggering ra on them.
1457  */
reada_for_search(struct btrfs_fs_info * fs_info,struct btrfs_path * path,int level,int slot,u64 objectid)1458 static void reada_for_search(struct btrfs_fs_info *fs_info,
1459 			     struct btrfs_path *path,
1460 			     int level, int slot, u64 objectid)
1461 {
1462 	struct extent_buffer *node;
1463 	struct btrfs_disk_key disk_key;
1464 	u32 nritems;
1465 	u64 search;
1466 	u64 target;
1467 	u64 nread = 0;
1468 	u64 nread_max;
1469 	u32 nr;
1470 	u32 blocksize;
1471 	u32 nscan = 0;
1472 
1473 	if (level != 1 && path->reada != READA_FORWARD_ALWAYS)
1474 		return;
1475 
1476 	if (!path->nodes[level])
1477 		return;
1478 
1479 	node = path->nodes[level];
1480 
1481 	/*
1482 	 * Since the time between visiting leaves is much shorter than the time
1483 	 * between visiting nodes, limit read ahead of nodes to 1, to avoid too
1484 	 * much IO at once (possibly random).
1485 	 */
1486 	if (path->reada == READA_FORWARD_ALWAYS) {
1487 		if (level > 1)
1488 			nread_max = node->fs_info->nodesize;
1489 		else
1490 			nread_max = SZ_128K;
1491 	} else {
1492 		nread_max = SZ_64K;
1493 	}
1494 
1495 	search = btrfs_node_blockptr(node, slot);
1496 	blocksize = fs_info->nodesize;
1497 	if (path->reada != READA_FORWARD_ALWAYS) {
1498 		struct extent_buffer *eb;
1499 
1500 		eb = find_extent_buffer(fs_info, search);
1501 		if (eb) {
1502 			free_extent_buffer(eb);
1503 			return;
1504 		}
1505 	}
1506 
1507 	target = search;
1508 
1509 	nritems = btrfs_header_nritems(node);
1510 	nr = slot;
1511 
1512 	while (1) {
1513 		if (path->reada == READA_BACK) {
1514 			if (nr == 0)
1515 				break;
1516 			nr--;
1517 		} else if (path->reada == READA_FORWARD ||
1518 			   path->reada == READA_FORWARD_ALWAYS) {
1519 			nr++;
1520 			if (nr >= nritems)
1521 				break;
1522 		}
1523 		if (path->reada == READA_BACK && objectid) {
1524 			btrfs_node_key(node, &disk_key, nr);
1525 			if (btrfs_disk_key_objectid(&disk_key) != objectid)
1526 				break;
1527 		}
1528 		search = btrfs_node_blockptr(node, nr);
1529 		if (path->reada == READA_FORWARD_ALWAYS ||
1530 		    (search <= target && target - search <= 65536) ||
1531 		    (search > target && search - target <= 65536)) {
1532 			btrfs_readahead_node_child(node, nr);
1533 			nread += blocksize;
1534 		}
1535 		nscan++;
1536 		if (nread > nread_max || nscan > 32)
1537 			break;
1538 	}
1539 }
1540 
reada_for_balance(struct btrfs_path * path,int level)1541 static noinline void reada_for_balance(struct btrfs_path *path, int level)
1542 {
1543 	struct extent_buffer *parent;
1544 	int slot;
1545 	int nritems;
1546 
1547 	parent = path->nodes[level + 1];
1548 	if (!parent)
1549 		return;
1550 
1551 	nritems = btrfs_header_nritems(parent);
1552 	slot = path->slots[level + 1];
1553 
1554 	if (slot > 0)
1555 		btrfs_readahead_node_child(parent, slot - 1);
1556 	if (slot + 1 < nritems)
1557 		btrfs_readahead_node_child(parent, slot + 1);
1558 }
1559 
1560 
1561 /*
1562  * when we walk down the tree, it is usually safe to unlock the higher layers
1563  * in the tree.  The exceptions are when our path goes through slot 0, because
1564  * operations on the tree might require changing key pointers higher up in the
1565  * tree.
1566  *
1567  * callers might also have set path->keep_locks, which tells this code to keep
1568  * the lock if the path points to the last slot in the block.  This is part of
1569  * walking through the tree, and selecting the next slot in the higher block.
1570  *
1571  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
1572  * if lowest_unlock is 1, level 0 won't be unlocked
1573  */
unlock_up(struct btrfs_path * path,int level,int lowest_unlock,int min_write_lock_level,int * write_lock_level)1574 static noinline void unlock_up(struct btrfs_path *path, int level,
1575 			       int lowest_unlock, int min_write_lock_level,
1576 			       int *write_lock_level)
1577 {
1578 	int i;
1579 	int skip_level = level;
1580 	bool check_skip = true;
1581 
1582 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1583 		if (!path->nodes[i])
1584 			break;
1585 		if (!path->locks[i])
1586 			break;
1587 
1588 		if (check_skip) {
1589 			if (path->slots[i] == 0) {
1590 				skip_level = i + 1;
1591 				continue;
1592 			}
1593 
1594 			if (path->keep_locks) {
1595 				u32 nritems;
1596 
1597 				nritems = btrfs_header_nritems(path->nodes[i]);
1598 				if (nritems < 1 || path->slots[i] >= nritems - 1) {
1599 					skip_level = i + 1;
1600 					continue;
1601 				}
1602 			}
1603 		}
1604 
1605 		if (i >= lowest_unlock && i > skip_level) {
1606 			check_skip = false;
1607 			btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
1608 			path->locks[i] = 0;
1609 			if (write_lock_level &&
1610 			    i > min_write_lock_level &&
1611 			    i <= *write_lock_level) {
1612 				*write_lock_level = i - 1;
1613 			}
1614 		}
1615 	}
1616 }
1617 
1618 /*
1619  * Helper function for btrfs_search_slot() and other functions that do a search
1620  * on a btree. The goal is to find a tree block in the cache (the radix tree at
1621  * fs_info->buffer_radix), but if we can't find it, or it's not up to date, read
1622  * its pages from disk.
1623  *
1624  * Returns -EAGAIN, with the path unlocked, if the caller needs to repeat the
1625  * whole btree search, starting again from the current root node.
1626  */
1627 static int
read_block_for_search(struct btrfs_root * root,struct btrfs_path * p,struct extent_buffer ** eb_ret,int level,int slot,const struct btrfs_key * key)1628 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
1629 		      struct extent_buffer **eb_ret, int level, int slot,
1630 		      const struct btrfs_key *key)
1631 {
1632 	struct btrfs_fs_info *fs_info = root->fs_info;
1633 	struct btrfs_tree_parent_check check = { 0 };
1634 	u64 blocknr;
1635 	u64 gen;
1636 	struct extent_buffer *tmp;
1637 	int ret;
1638 	int parent_level;
1639 	bool unlock_up;
1640 
1641 	unlock_up = ((level + 1 < BTRFS_MAX_LEVEL) && p->locks[level + 1]);
1642 	blocknr = btrfs_node_blockptr(*eb_ret, slot);
1643 	gen = btrfs_node_ptr_generation(*eb_ret, slot);
1644 	parent_level = btrfs_header_level(*eb_ret);
1645 	btrfs_node_key_to_cpu(*eb_ret, &check.first_key, slot);
1646 	check.has_first_key = true;
1647 	check.level = parent_level - 1;
1648 	check.transid = gen;
1649 	check.owner_root = root->root_key.objectid;
1650 
1651 	/*
1652 	 * If we need to read an extent buffer from disk and we are holding locks
1653 	 * on upper level nodes, we unlock all the upper nodes before reading the
1654 	 * extent buffer, and then return -EAGAIN to the caller as it needs to
1655 	 * restart the search. We don't release the lock on the current level
1656 	 * because we need to walk this node to figure out which blocks to read.
1657 	 */
1658 	tmp = find_extent_buffer(fs_info, blocknr);
1659 	if (tmp) {
1660 		if (p->reada == READA_FORWARD_ALWAYS)
1661 			reada_for_search(fs_info, p, level, slot, key->objectid);
1662 
1663 		/* first we do an atomic uptodate check */
1664 		if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
1665 			/*
1666 			 * Do extra check for first_key, eb can be stale due to
1667 			 * being cached, read from scrub, or have multiple
1668 			 * parents (shared tree blocks).
1669 			 */
1670 			if (btrfs_verify_level_key(tmp,
1671 					parent_level - 1, &check.first_key, gen)) {
1672 				free_extent_buffer(tmp);
1673 				return -EUCLEAN;
1674 			}
1675 			*eb_ret = tmp;
1676 			return 0;
1677 		}
1678 
1679 		if (p->nowait) {
1680 			free_extent_buffer(tmp);
1681 			return -EAGAIN;
1682 		}
1683 
1684 		if (unlock_up)
1685 			btrfs_unlock_up_safe(p, level + 1);
1686 
1687 		/* now we're allowed to do a blocking uptodate check */
1688 		ret = btrfs_read_extent_buffer(tmp, &check);
1689 		if (ret) {
1690 			free_extent_buffer(tmp);
1691 			btrfs_release_path(p);
1692 			return -EIO;
1693 		}
1694 		if (btrfs_check_eb_owner(tmp, root->root_key.objectid)) {
1695 			free_extent_buffer(tmp);
1696 			btrfs_release_path(p);
1697 			return -EUCLEAN;
1698 		}
1699 
1700 		if (unlock_up)
1701 			ret = -EAGAIN;
1702 
1703 		goto out;
1704 	} else if (p->nowait) {
1705 		return -EAGAIN;
1706 	}
1707 
1708 	if (unlock_up) {
1709 		btrfs_unlock_up_safe(p, level + 1);
1710 		ret = -EAGAIN;
1711 	} else {
1712 		ret = 0;
1713 	}
1714 
1715 	if (p->reada != READA_NONE)
1716 		reada_for_search(fs_info, p, level, slot, key->objectid);
1717 
1718 	tmp = read_tree_block(fs_info, blocknr, &check);
1719 	if (IS_ERR(tmp)) {
1720 		btrfs_release_path(p);
1721 		return PTR_ERR(tmp);
1722 	}
1723 	/*
1724 	 * If the read above didn't mark this buffer up to date,
1725 	 * it will never end up being up to date.  Set ret to EIO now
1726 	 * and give up so that our caller doesn't loop forever
1727 	 * on our EAGAINs.
1728 	 */
1729 	if (!extent_buffer_uptodate(tmp))
1730 		ret = -EIO;
1731 
1732 out:
1733 	if (ret == 0) {
1734 		*eb_ret = tmp;
1735 	} else {
1736 		free_extent_buffer(tmp);
1737 		btrfs_release_path(p);
1738 	}
1739 
1740 	return ret;
1741 }
1742 
1743 /*
1744  * helper function for btrfs_search_slot.  This does all of the checks
1745  * for node-level blocks and does any balancing required based on
1746  * the ins_len.
1747  *
1748  * If no extra work was required, zero is returned.  If we had to
1749  * drop the path, -EAGAIN is returned and btrfs_search_slot must
1750  * start over
1751  */
1752 static int
setup_nodes_for_search(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * p,struct extent_buffer * b,int level,int ins_len,int * write_lock_level)1753 setup_nodes_for_search(struct btrfs_trans_handle *trans,
1754 		       struct btrfs_root *root, struct btrfs_path *p,
1755 		       struct extent_buffer *b, int level, int ins_len,
1756 		       int *write_lock_level)
1757 {
1758 	struct btrfs_fs_info *fs_info = root->fs_info;
1759 	int ret = 0;
1760 
1761 	if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1762 	    BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
1763 
1764 		if (*write_lock_level < level + 1) {
1765 			*write_lock_level = level + 1;
1766 			btrfs_release_path(p);
1767 			return -EAGAIN;
1768 		}
1769 
1770 		reada_for_balance(p, level);
1771 		ret = split_node(trans, root, p, level);
1772 
1773 		b = p->nodes[level];
1774 	} else if (ins_len < 0 && btrfs_header_nritems(b) <
1775 		   BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
1776 
1777 		if (*write_lock_level < level + 1) {
1778 			*write_lock_level = level + 1;
1779 			btrfs_release_path(p);
1780 			return -EAGAIN;
1781 		}
1782 
1783 		reada_for_balance(p, level);
1784 		ret = balance_level(trans, root, p, level);
1785 		if (ret)
1786 			return ret;
1787 
1788 		b = p->nodes[level];
1789 		if (!b) {
1790 			btrfs_release_path(p);
1791 			return -EAGAIN;
1792 		}
1793 		BUG_ON(btrfs_header_nritems(b) == 1);
1794 	}
1795 	return ret;
1796 }
1797 
btrfs_find_item(struct btrfs_root * fs_root,struct btrfs_path * path,u64 iobjectid,u64 ioff,u8 key_type,struct btrfs_key * found_key)1798 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
1799 		u64 iobjectid, u64 ioff, u8 key_type,
1800 		struct btrfs_key *found_key)
1801 {
1802 	int ret;
1803 	struct btrfs_key key;
1804 	struct extent_buffer *eb;
1805 
1806 	ASSERT(path);
1807 	ASSERT(found_key);
1808 
1809 	key.type = key_type;
1810 	key.objectid = iobjectid;
1811 	key.offset = ioff;
1812 
1813 	ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1814 	if (ret < 0)
1815 		return ret;
1816 
1817 	eb = path->nodes[0];
1818 	if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1819 		ret = btrfs_next_leaf(fs_root, path);
1820 		if (ret)
1821 			return ret;
1822 		eb = path->nodes[0];
1823 	}
1824 
1825 	btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1826 	if (found_key->type != key.type ||
1827 			found_key->objectid != key.objectid)
1828 		return 1;
1829 
1830 	return 0;
1831 }
1832 
btrfs_search_slot_get_root(struct btrfs_root * root,struct btrfs_path * p,int write_lock_level)1833 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
1834 							struct btrfs_path *p,
1835 							int write_lock_level)
1836 {
1837 	struct extent_buffer *b;
1838 	int root_lock = 0;
1839 	int level = 0;
1840 
1841 	if (p->search_commit_root) {
1842 		b = root->commit_root;
1843 		atomic_inc(&b->refs);
1844 		level = btrfs_header_level(b);
1845 		/*
1846 		 * Ensure that all callers have set skip_locking when
1847 		 * p->search_commit_root = 1.
1848 		 */
1849 		ASSERT(p->skip_locking == 1);
1850 
1851 		goto out;
1852 	}
1853 
1854 	if (p->skip_locking) {
1855 		b = btrfs_root_node(root);
1856 		level = btrfs_header_level(b);
1857 		goto out;
1858 	}
1859 
1860 	/* We try very hard to do read locks on the root */
1861 	root_lock = BTRFS_READ_LOCK;
1862 
1863 	/*
1864 	 * If the level is set to maximum, we can skip trying to get the read
1865 	 * lock.
1866 	 */
1867 	if (write_lock_level < BTRFS_MAX_LEVEL) {
1868 		/*
1869 		 * We don't know the level of the root node until we actually
1870 		 * have it read locked
1871 		 */
1872 		if (p->nowait) {
1873 			b = btrfs_try_read_lock_root_node(root);
1874 			if (IS_ERR(b))
1875 				return b;
1876 		} else {
1877 			b = btrfs_read_lock_root_node(root);
1878 		}
1879 		level = btrfs_header_level(b);
1880 		if (level > write_lock_level)
1881 			goto out;
1882 
1883 		/* Whoops, must trade for write lock */
1884 		btrfs_tree_read_unlock(b);
1885 		free_extent_buffer(b);
1886 	}
1887 
1888 	b = btrfs_lock_root_node(root);
1889 	root_lock = BTRFS_WRITE_LOCK;
1890 
1891 	/* The level might have changed, check again */
1892 	level = btrfs_header_level(b);
1893 
1894 out:
1895 	/*
1896 	 * The root may have failed to write out at some point, and thus is no
1897 	 * longer valid, return an error in this case.
1898 	 */
1899 	if (!extent_buffer_uptodate(b)) {
1900 		if (root_lock)
1901 			btrfs_tree_unlock_rw(b, root_lock);
1902 		free_extent_buffer(b);
1903 		return ERR_PTR(-EIO);
1904 	}
1905 
1906 	p->nodes[level] = b;
1907 	if (!p->skip_locking)
1908 		p->locks[level] = root_lock;
1909 	/*
1910 	 * Callers are responsible for dropping b's references.
1911 	 */
1912 	return b;
1913 }
1914 
1915 /*
1916  * Replace the extent buffer at the lowest level of the path with a cloned
1917  * version. The purpose is to be able to use it safely, after releasing the
1918  * commit root semaphore, even if relocation is happening in parallel, the
1919  * transaction used for relocation is committed and the extent buffer is
1920  * reallocated in the next transaction.
1921  *
1922  * This is used in a context where the caller does not prevent transaction
1923  * commits from happening, either by holding a transaction handle or holding
1924  * some lock, while it's doing searches through a commit root.
1925  * At the moment it's only used for send operations.
1926  */
finish_need_commit_sem_search(struct btrfs_path * path)1927 static int finish_need_commit_sem_search(struct btrfs_path *path)
1928 {
1929 	const int i = path->lowest_level;
1930 	const int slot = path->slots[i];
1931 	struct extent_buffer *lowest = path->nodes[i];
1932 	struct extent_buffer *clone;
1933 
1934 	ASSERT(path->need_commit_sem);
1935 
1936 	if (!lowest)
1937 		return 0;
1938 
1939 	lockdep_assert_held_read(&lowest->fs_info->commit_root_sem);
1940 
1941 	clone = btrfs_clone_extent_buffer(lowest);
1942 	if (!clone)
1943 		return -ENOMEM;
1944 
1945 	btrfs_release_path(path);
1946 	path->nodes[i] = clone;
1947 	path->slots[i] = slot;
1948 
1949 	return 0;
1950 }
1951 
search_for_key_slot(struct extent_buffer * eb,int search_low_slot,const struct btrfs_key * key,int prev_cmp,int * slot)1952 static inline int search_for_key_slot(struct extent_buffer *eb,
1953 				      int search_low_slot,
1954 				      const struct btrfs_key *key,
1955 				      int prev_cmp,
1956 				      int *slot)
1957 {
1958 	/*
1959 	 * If a previous call to btrfs_bin_search() on a parent node returned an
1960 	 * exact match (prev_cmp == 0), we can safely assume the target key will
1961 	 * always be at slot 0 on lower levels, since each key pointer
1962 	 * (struct btrfs_key_ptr) refers to the lowest key accessible from the
1963 	 * subtree it points to. Thus we can skip searching lower levels.
1964 	 */
1965 	if (prev_cmp == 0) {
1966 		*slot = 0;
1967 		return 0;
1968 	}
1969 
1970 	return btrfs_bin_search(eb, search_low_slot, key, slot);
1971 }
1972 
search_leaf(struct btrfs_trans_handle * trans,struct btrfs_root * root,const struct btrfs_key * key,struct btrfs_path * path,int ins_len,int prev_cmp)1973 static int search_leaf(struct btrfs_trans_handle *trans,
1974 		       struct btrfs_root *root,
1975 		       const struct btrfs_key *key,
1976 		       struct btrfs_path *path,
1977 		       int ins_len,
1978 		       int prev_cmp)
1979 {
1980 	struct extent_buffer *leaf = path->nodes[0];
1981 	int leaf_free_space = -1;
1982 	int search_low_slot = 0;
1983 	int ret;
1984 	bool do_bin_search = true;
1985 
1986 	/*
1987 	 * If we are doing an insertion, the leaf has enough free space and the
1988 	 * destination slot for the key is not slot 0, then we can unlock our
1989 	 * write lock on the parent, and any other upper nodes, before doing the
1990 	 * binary search on the leaf (with search_for_key_slot()), allowing other
1991 	 * tasks to lock the parent and any other upper nodes.
1992 	 */
1993 	if (ins_len > 0) {
1994 		/*
1995 		 * Cache the leaf free space, since we will need it later and it
1996 		 * will not change until then.
1997 		 */
1998 		leaf_free_space = btrfs_leaf_free_space(leaf);
1999 
2000 		/*
2001 		 * !path->locks[1] means we have a single node tree, the leaf is
2002 		 * the root of the tree.
2003 		 */
2004 		if (path->locks[1] && leaf_free_space >= ins_len) {
2005 			struct btrfs_disk_key first_key;
2006 
2007 			ASSERT(btrfs_header_nritems(leaf) > 0);
2008 			btrfs_item_key(leaf, &first_key, 0);
2009 
2010 			/*
2011 			 * Doing the extra comparison with the first key is cheap,
2012 			 * taking into account that the first key is very likely
2013 			 * already in a cache line because it immediately follows
2014 			 * the extent buffer's header and we have recently accessed
2015 			 * the header's level field.
2016 			 */
2017 			ret = comp_keys(&first_key, key);
2018 			if (ret < 0) {
2019 				/*
2020 				 * The first key is smaller than the key we want
2021 				 * to insert, so we are safe to unlock all upper
2022 				 * nodes and we have to do the binary search.
2023 				 *
2024 				 * We do use btrfs_unlock_up_safe() and not
2025 				 * unlock_up() because the later does not unlock
2026 				 * nodes with a slot of 0 - we can safely unlock
2027 				 * any node even if its slot is 0 since in this
2028 				 * case the key does not end up at slot 0 of the
2029 				 * leaf and there's no need to split the leaf.
2030 				 */
2031 				btrfs_unlock_up_safe(path, 1);
2032 				search_low_slot = 1;
2033 			} else {
2034 				/*
2035 				 * The first key is >= then the key we want to
2036 				 * insert, so we can skip the binary search as
2037 				 * the target key will be at slot 0.
2038 				 *
2039 				 * We can not unlock upper nodes when the key is
2040 				 * less than the first key, because we will need
2041 				 * to update the key at slot 0 of the parent node
2042 				 * and possibly of other upper nodes too.
2043 				 * If the key matches the first key, then we can
2044 				 * unlock all the upper nodes, using
2045 				 * btrfs_unlock_up_safe() instead of unlock_up()
2046 				 * as stated above.
2047 				 */
2048 				if (ret == 0)
2049 					btrfs_unlock_up_safe(path, 1);
2050 				/*
2051 				 * ret is already 0 or 1, matching the result of
2052 				 * a btrfs_bin_search() call, so there is no need
2053 				 * to adjust it.
2054 				 */
2055 				do_bin_search = false;
2056 				path->slots[0] = 0;
2057 			}
2058 		}
2059 	}
2060 
2061 	if (do_bin_search) {
2062 		ret = search_for_key_slot(leaf, search_low_slot, key,
2063 					  prev_cmp, &path->slots[0]);
2064 		if (ret < 0)
2065 			return ret;
2066 	}
2067 
2068 	if (ins_len > 0) {
2069 		/*
2070 		 * Item key already exists. In this case, if we are allowed to
2071 		 * insert the item (for example, in dir_item case, item key
2072 		 * collision is allowed), it will be merged with the original
2073 		 * item. Only the item size grows, no new btrfs item will be
2074 		 * added. If search_for_extension is not set, ins_len already
2075 		 * accounts the size btrfs_item, deduct it here so leaf space
2076 		 * check will be correct.
2077 		 */
2078 		if (ret == 0 && !path->search_for_extension) {
2079 			ASSERT(ins_len >= sizeof(struct btrfs_item));
2080 			ins_len -= sizeof(struct btrfs_item);
2081 		}
2082 
2083 		ASSERT(leaf_free_space >= 0);
2084 
2085 		if (leaf_free_space < ins_len) {
2086 			int err;
2087 
2088 			err = split_leaf(trans, root, key, path, ins_len,
2089 					 (ret == 0));
2090 			ASSERT(err <= 0);
2091 			if (WARN_ON(err > 0))
2092 				err = -EUCLEAN;
2093 			if (err)
2094 				ret = err;
2095 		}
2096 	}
2097 
2098 	return ret;
2099 }
2100 
2101 /*
2102  * btrfs_search_slot - look for a key in a tree and perform necessary
2103  * modifications to preserve tree invariants.
2104  *
2105  * @trans:	Handle of transaction, used when modifying the tree
2106  * @p:		Holds all btree nodes along the search path
2107  * @root:	The root node of the tree
2108  * @key:	The key we are looking for
2109  * @ins_len:	Indicates purpose of search:
2110  *              >0  for inserts it's size of item inserted (*)
2111  *              <0  for deletions
2112  *               0  for plain searches, not modifying the tree
2113  *
2114  *              (*) If size of item inserted doesn't include
2115  *              sizeof(struct btrfs_item), then p->search_for_extension must
2116  *              be set.
2117  * @cow:	boolean should CoW operations be performed. Must always be 1
2118  *		when modifying the tree.
2119  *
2120  * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2121  * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2122  *
2123  * If @key is found, 0 is returned and you can find the item in the leaf level
2124  * of the path (level 0)
2125  *
2126  * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2127  * points to the slot where it should be inserted
2128  *
2129  * If an error is encountered while searching the tree a negative error number
2130  * is returned
2131  */
btrfs_search_slot(struct btrfs_trans_handle * trans,struct btrfs_root * root,const struct btrfs_key * key,struct btrfs_path * p,int ins_len,int cow)2132 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2133 		      const struct btrfs_key *key, struct btrfs_path *p,
2134 		      int ins_len, int cow)
2135 {
2136 	struct btrfs_fs_info *fs_info = root->fs_info;
2137 	struct extent_buffer *b;
2138 	int slot;
2139 	int ret;
2140 	int err;
2141 	int level;
2142 	int lowest_unlock = 1;
2143 	/* everything at write_lock_level or lower must be write locked */
2144 	int write_lock_level = 0;
2145 	u8 lowest_level = 0;
2146 	int min_write_lock_level;
2147 	int prev_cmp;
2148 
2149 	might_sleep();
2150 
2151 	lowest_level = p->lowest_level;
2152 	WARN_ON(lowest_level && ins_len > 0);
2153 	WARN_ON(p->nodes[0] != NULL);
2154 	BUG_ON(!cow && ins_len);
2155 
2156 	/*
2157 	 * For now only allow nowait for read only operations.  There's no
2158 	 * strict reason why we can't, we just only need it for reads so it's
2159 	 * only implemented for reads.
2160 	 */
2161 	ASSERT(!p->nowait || !cow);
2162 
2163 	if (ins_len < 0) {
2164 		lowest_unlock = 2;
2165 
2166 		/* when we are removing items, we might have to go up to level
2167 		 * two as we update tree pointers  Make sure we keep write
2168 		 * for those levels as well
2169 		 */
2170 		write_lock_level = 2;
2171 	} else if (ins_len > 0) {
2172 		/*
2173 		 * for inserting items, make sure we have a write lock on
2174 		 * level 1 so we can update keys
2175 		 */
2176 		write_lock_level = 1;
2177 	}
2178 
2179 	if (!cow)
2180 		write_lock_level = -1;
2181 
2182 	if (cow && (p->keep_locks || p->lowest_level))
2183 		write_lock_level = BTRFS_MAX_LEVEL;
2184 
2185 	min_write_lock_level = write_lock_level;
2186 
2187 	if (p->need_commit_sem) {
2188 		ASSERT(p->search_commit_root);
2189 		if (p->nowait) {
2190 			if (!down_read_trylock(&fs_info->commit_root_sem))
2191 				return -EAGAIN;
2192 		} else {
2193 			down_read(&fs_info->commit_root_sem);
2194 		}
2195 	}
2196 
2197 again:
2198 	prev_cmp = -1;
2199 	b = btrfs_search_slot_get_root(root, p, write_lock_level);
2200 	if (IS_ERR(b)) {
2201 		ret = PTR_ERR(b);
2202 		goto done;
2203 	}
2204 
2205 	while (b) {
2206 		int dec = 0;
2207 
2208 		level = btrfs_header_level(b);
2209 
2210 		if (cow) {
2211 			bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2212 
2213 			/*
2214 			 * if we don't really need to cow this block
2215 			 * then we don't want to set the path blocking,
2216 			 * so we test it here
2217 			 */
2218 			if (!should_cow_block(trans, root, b))
2219 				goto cow_done;
2220 
2221 			/*
2222 			 * must have write locks on this node and the
2223 			 * parent
2224 			 */
2225 			if (level > write_lock_level ||
2226 			    (level + 1 > write_lock_level &&
2227 			    level + 1 < BTRFS_MAX_LEVEL &&
2228 			    p->nodes[level + 1])) {
2229 				write_lock_level = level + 1;
2230 				btrfs_release_path(p);
2231 				goto again;
2232 			}
2233 
2234 			if (last_level)
2235 				err = btrfs_cow_block(trans, root, b, NULL, 0,
2236 						      &b,
2237 						      BTRFS_NESTING_COW);
2238 			else
2239 				err = btrfs_cow_block(trans, root, b,
2240 						      p->nodes[level + 1],
2241 						      p->slots[level + 1], &b,
2242 						      BTRFS_NESTING_COW);
2243 			if (err) {
2244 				ret = err;
2245 				goto done;
2246 			}
2247 		}
2248 cow_done:
2249 		p->nodes[level] = b;
2250 
2251 		/*
2252 		 * we have a lock on b and as long as we aren't changing
2253 		 * the tree, there is no way to for the items in b to change.
2254 		 * It is safe to drop the lock on our parent before we
2255 		 * go through the expensive btree search on b.
2256 		 *
2257 		 * If we're inserting or deleting (ins_len != 0), then we might
2258 		 * be changing slot zero, which may require changing the parent.
2259 		 * So, we can't drop the lock until after we know which slot
2260 		 * we're operating on.
2261 		 */
2262 		if (!ins_len && !p->keep_locks) {
2263 			int u = level + 1;
2264 
2265 			if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2266 				btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2267 				p->locks[u] = 0;
2268 			}
2269 		}
2270 
2271 		if (level == 0) {
2272 			if (ins_len > 0)
2273 				ASSERT(write_lock_level >= 1);
2274 
2275 			ret = search_leaf(trans, root, key, p, ins_len, prev_cmp);
2276 			if (!p->search_for_split)
2277 				unlock_up(p, level, lowest_unlock,
2278 					  min_write_lock_level, NULL);
2279 			goto done;
2280 		}
2281 
2282 		ret = search_for_key_slot(b, 0, key, prev_cmp, &slot);
2283 		if (ret < 0)
2284 			goto done;
2285 		prev_cmp = ret;
2286 
2287 		if (ret && slot > 0) {
2288 			dec = 1;
2289 			slot--;
2290 		}
2291 		p->slots[level] = slot;
2292 		err = setup_nodes_for_search(trans, root, p, b, level, ins_len,
2293 					     &write_lock_level);
2294 		if (err == -EAGAIN)
2295 			goto again;
2296 		if (err) {
2297 			ret = err;
2298 			goto done;
2299 		}
2300 		b = p->nodes[level];
2301 		slot = p->slots[level];
2302 
2303 		/*
2304 		 * Slot 0 is special, if we change the key we have to update
2305 		 * the parent pointer which means we must have a write lock on
2306 		 * the parent
2307 		 */
2308 		if (slot == 0 && ins_len && write_lock_level < level + 1) {
2309 			write_lock_level = level + 1;
2310 			btrfs_release_path(p);
2311 			goto again;
2312 		}
2313 
2314 		unlock_up(p, level, lowest_unlock, min_write_lock_level,
2315 			  &write_lock_level);
2316 
2317 		if (level == lowest_level) {
2318 			if (dec)
2319 				p->slots[level]++;
2320 			goto done;
2321 		}
2322 
2323 		err = read_block_for_search(root, p, &b, level, slot, key);
2324 		if (err == -EAGAIN)
2325 			goto again;
2326 		if (err) {
2327 			ret = err;
2328 			goto done;
2329 		}
2330 
2331 		if (!p->skip_locking) {
2332 			level = btrfs_header_level(b);
2333 
2334 			btrfs_maybe_reset_lockdep_class(root, b);
2335 
2336 			if (level <= write_lock_level) {
2337 				btrfs_tree_lock(b);
2338 				p->locks[level] = BTRFS_WRITE_LOCK;
2339 			} else {
2340 				if (p->nowait) {
2341 					if (!btrfs_try_tree_read_lock(b)) {
2342 						free_extent_buffer(b);
2343 						ret = -EAGAIN;
2344 						goto done;
2345 					}
2346 				} else {
2347 					btrfs_tree_read_lock(b);
2348 				}
2349 				p->locks[level] = BTRFS_READ_LOCK;
2350 			}
2351 			p->nodes[level] = b;
2352 		}
2353 	}
2354 	ret = 1;
2355 done:
2356 	if (ret < 0 && !p->skip_release_on_error)
2357 		btrfs_release_path(p);
2358 
2359 	if (p->need_commit_sem) {
2360 		int ret2;
2361 
2362 		ret2 = finish_need_commit_sem_search(p);
2363 		up_read(&fs_info->commit_root_sem);
2364 		if (ret2)
2365 			ret = ret2;
2366 	}
2367 
2368 	return ret;
2369 }
2370 ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO);
2371 
2372 /*
2373  * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2374  * current state of the tree together with the operations recorded in the tree
2375  * modification log to search for the key in a previous version of this tree, as
2376  * denoted by the time_seq parameter.
2377  *
2378  * Naturally, there is no support for insert, delete or cow operations.
2379  *
2380  * The resulting path and return value will be set up as if we called
2381  * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2382  */
btrfs_search_old_slot(struct btrfs_root * root,const struct btrfs_key * key,struct btrfs_path * p,u64 time_seq)2383 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2384 			  struct btrfs_path *p, u64 time_seq)
2385 {
2386 	struct btrfs_fs_info *fs_info = root->fs_info;
2387 	struct extent_buffer *b;
2388 	int slot;
2389 	int ret;
2390 	int err;
2391 	int level;
2392 	int lowest_unlock = 1;
2393 	u8 lowest_level = 0;
2394 
2395 	lowest_level = p->lowest_level;
2396 	WARN_ON(p->nodes[0] != NULL);
2397 	ASSERT(!p->nowait);
2398 
2399 	if (p->search_commit_root) {
2400 		BUG_ON(time_seq);
2401 		return btrfs_search_slot(NULL, root, key, p, 0, 0);
2402 	}
2403 
2404 again:
2405 	b = btrfs_get_old_root(root, time_seq);
2406 	if (!b) {
2407 		ret = -EIO;
2408 		goto done;
2409 	}
2410 	level = btrfs_header_level(b);
2411 	p->locks[level] = BTRFS_READ_LOCK;
2412 
2413 	while (b) {
2414 		int dec = 0;
2415 
2416 		level = btrfs_header_level(b);
2417 		p->nodes[level] = b;
2418 
2419 		/*
2420 		 * we have a lock on b and as long as we aren't changing
2421 		 * the tree, there is no way to for the items in b to change.
2422 		 * It is safe to drop the lock on our parent before we
2423 		 * go through the expensive btree search on b.
2424 		 */
2425 		btrfs_unlock_up_safe(p, level + 1);
2426 
2427 		ret = btrfs_bin_search(b, 0, key, &slot);
2428 		if (ret < 0)
2429 			goto done;
2430 
2431 		if (level == 0) {
2432 			p->slots[level] = slot;
2433 			unlock_up(p, level, lowest_unlock, 0, NULL);
2434 			goto done;
2435 		}
2436 
2437 		if (ret && slot > 0) {
2438 			dec = 1;
2439 			slot--;
2440 		}
2441 		p->slots[level] = slot;
2442 		unlock_up(p, level, lowest_unlock, 0, NULL);
2443 
2444 		if (level == lowest_level) {
2445 			if (dec)
2446 				p->slots[level]++;
2447 			goto done;
2448 		}
2449 
2450 		err = read_block_for_search(root, p, &b, level, slot, key);
2451 		if (err == -EAGAIN)
2452 			goto again;
2453 		if (err) {
2454 			ret = err;
2455 			goto done;
2456 		}
2457 
2458 		level = btrfs_header_level(b);
2459 		btrfs_tree_read_lock(b);
2460 		b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq);
2461 		if (!b) {
2462 			ret = -ENOMEM;
2463 			goto done;
2464 		}
2465 		p->locks[level] = BTRFS_READ_LOCK;
2466 		p->nodes[level] = b;
2467 	}
2468 	ret = 1;
2469 done:
2470 	if (ret < 0)
2471 		btrfs_release_path(p);
2472 
2473 	return ret;
2474 }
2475 
2476 /*
2477  * Search the tree again to find a leaf with smaller keys.
2478  * Returns 0 if it found something.
2479  * Returns 1 if there are no smaller keys.
2480  * Returns < 0 on error.
2481  *
2482  * This may release the path, and so you may lose any locks held at the
2483  * time you call it.
2484  */
btrfs_prev_leaf(struct btrfs_root * root,struct btrfs_path * path)2485 static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2486 {
2487 	struct btrfs_key key;
2488 	struct btrfs_key orig_key;
2489 	struct btrfs_disk_key found_key;
2490 	int ret;
2491 
2492 	btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
2493 	orig_key = key;
2494 
2495 	if (key.offset > 0) {
2496 		key.offset--;
2497 	} else if (key.type > 0) {
2498 		key.type--;
2499 		key.offset = (u64)-1;
2500 	} else if (key.objectid > 0) {
2501 		key.objectid--;
2502 		key.type = (u8)-1;
2503 		key.offset = (u64)-1;
2504 	} else {
2505 		return 1;
2506 	}
2507 
2508 	btrfs_release_path(path);
2509 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2510 	if (ret <= 0)
2511 		return ret;
2512 
2513 	/*
2514 	 * Previous key not found. Even if we were at slot 0 of the leaf we had
2515 	 * before releasing the path and calling btrfs_search_slot(), we now may
2516 	 * be in a slot pointing to the same original key - this can happen if
2517 	 * after we released the path, one of more items were moved from a
2518 	 * sibling leaf into the front of the leaf we had due to an insertion
2519 	 * (see push_leaf_right()).
2520 	 * If we hit this case and our slot is > 0 and just decrement the slot
2521 	 * so that the caller does not process the same key again, which may or
2522 	 * may not break the caller, depending on its logic.
2523 	 */
2524 	if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
2525 		btrfs_item_key(path->nodes[0], &found_key, path->slots[0]);
2526 		ret = comp_keys(&found_key, &orig_key);
2527 		if (ret == 0) {
2528 			if (path->slots[0] > 0) {
2529 				path->slots[0]--;
2530 				return 0;
2531 			}
2532 			/*
2533 			 * At slot 0, same key as before, it means orig_key is
2534 			 * the lowest, leftmost, key in the tree. We're done.
2535 			 */
2536 			return 1;
2537 		}
2538 	}
2539 
2540 	btrfs_item_key(path->nodes[0], &found_key, 0);
2541 	ret = comp_keys(&found_key, &key);
2542 	/*
2543 	 * We might have had an item with the previous key in the tree right
2544 	 * before we released our path. And after we released our path, that
2545 	 * item might have been pushed to the first slot (0) of the leaf we
2546 	 * were holding due to a tree balance. Alternatively, an item with the
2547 	 * previous key can exist as the only element of a leaf (big fat item).
2548 	 * Therefore account for these 2 cases, so that our callers (like
2549 	 * btrfs_previous_item) don't miss an existing item with a key matching
2550 	 * the previous key we computed above.
2551 	 */
2552 	if (ret <= 0)
2553 		return 0;
2554 	return 1;
2555 }
2556 
2557 /*
2558  * helper to use instead of search slot if no exact match is needed but
2559  * instead the next or previous item should be returned.
2560  * When find_higher is true, the next higher item is returned, the next lower
2561  * otherwise.
2562  * When return_any and find_higher are both true, and no higher item is found,
2563  * return the next lower instead.
2564  * When return_any is true and find_higher is false, and no lower item is found,
2565  * return the next higher instead.
2566  * It returns 0 if any item is found, 1 if none is found (tree empty), and
2567  * < 0 on error
2568  */
btrfs_search_slot_for_read(struct btrfs_root * root,const struct btrfs_key * key,struct btrfs_path * p,int find_higher,int return_any)2569 int btrfs_search_slot_for_read(struct btrfs_root *root,
2570 			       const struct btrfs_key *key,
2571 			       struct btrfs_path *p, int find_higher,
2572 			       int return_any)
2573 {
2574 	int ret;
2575 	struct extent_buffer *leaf;
2576 
2577 again:
2578 	ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
2579 	if (ret <= 0)
2580 		return ret;
2581 	/*
2582 	 * a return value of 1 means the path is at the position where the
2583 	 * item should be inserted. Normally this is the next bigger item,
2584 	 * but in case the previous item is the last in a leaf, path points
2585 	 * to the first free slot in the previous leaf, i.e. at an invalid
2586 	 * item.
2587 	 */
2588 	leaf = p->nodes[0];
2589 
2590 	if (find_higher) {
2591 		if (p->slots[0] >= btrfs_header_nritems(leaf)) {
2592 			ret = btrfs_next_leaf(root, p);
2593 			if (ret <= 0)
2594 				return ret;
2595 			if (!return_any)
2596 				return 1;
2597 			/*
2598 			 * no higher item found, return the next
2599 			 * lower instead
2600 			 */
2601 			return_any = 0;
2602 			find_higher = 0;
2603 			btrfs_release_path(p);
2604 			goto again;
2605 		}
2606 	} else {
2607 		if (p->slots[0] == 0) {
2608 			ret = btrfs_prev_leaf(root, p);
2609 			if (ret < 0)
2610 				return ret;
2611 			if (!ret) {
2612 				leaf = p->nodes[0];
2613 				if (p->slots[0] == btrfs_header_nritems(leaf))
2614 					p->slots[0]--;
2615 				return 0;
2616 			}
2617 			if (!return_any)
2618 				return 1;
2619 			/*
2620 			 * no lower item found, return the next
2621 			 * higher instead
2622 			 */
2623 			return_any = 0;
2624 			find_higher = 1;
2625 			btrfs_release_path(p);
2626 			goto again;
2627 		} else {
2628 			--p->slots[0];
2629 		}
2630 	}
2631 	return 0;
2632 }
2633 
2634 /*
2635  * Execute search and call btrfs_previous_item to traverse backwards if the item
2636  * was not found.
2637  *
2638  * Return 0 if found, 1 if not found and < 0 if error.
2639  */
btrfs_search_backwards(struct btrfs_root * root,struct btrfs_key * key,struct btrfs_path * path)2640 int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key,
2641 			   struct btrfs_path *path)
2642 {
2643 	int ret;
2644 
2645 	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2646 	if (ret > 0)
2647 		ret = btrfs_previous_item(root, path, key->objectid, key->type);
2648 
2649 	if (ret == 0)
2650 		btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]);
2651 
2652 	return ret;
2653 }
2654 
2655 /*
2656  * Search for a valid slot for the given path.
2657  *
2658  * @root:	The root node of the tree.
2659  * @key:	Will contain a valid item if found.
2660  * @path:	The starting point to validate the slot.
2661  *
2662  * Return: 0  if the item is valid
2663  *         1  if not found
2664  *         <0 if error.
2665  */
btrfs_get_next_valid_item(struct btrfs_root * root,struct btrfs_key * key,struct btrfs_path * path)2666 int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key,
2667 			      struct btrfs_path *path)
2668 {
2669 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2670 		int ret;
2671 
2672 		ret = btrfs_next_leaf(root, path);
2673 		if (ret)
2674 			return ret;
2675 	}
2676 
2677 	btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]);
2678 	return 0;
2679 }
2680 
2681 /*
2682  * adjust the pointers going up the tree, starting at level
2683  * making sure the right key of each node is points to 'key'.
2684  * This is used after shifting pointers to the left, so it stops
2685  * fixing up pointers when a given leaf/node is not in slot 0 of the
2686  * higher levels
2687  *
2688  */
fixup_low_keys(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_disk_key * key,int level)2689 static void fixup_low_keys(struct btrfs_trans_handle *trans,
2690 			   struct btrfs_path *path,
2691 			   struct btrfs_disk_key *key, int level)
2692 {
2693 	int i;
2694 	struct extent_buffer *t;
2695 	int ret;
2696 
2697 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2698 		int tslot = path->slots[i];
2699 
2700 		if (!path->nodes[i])
2701 			break;
2702 		t = path->nodes[i];
2703 		ret = btrfs_tree_mod_log_insert_key(t, tslot,
2704 						    BTRFS_MOD_LOG_KEY_REPLACE);
2705 		BUG_ON(ret < 0);
2706 		btrfs_set_node_key(t, key, tslot);
2707 		btrfs_mark_buffer_dirty(trans, path->nodes[i]);
2708 		if (tslot != 0)
2709 			break;
2710 	}
2711 }
2712 
2713 /*
2714  * update item key.
2715  *
2716  * This function isn't completely safe. It's the caller's responsibility
2717  * that the new key won't break the order
2718  */
btrfs_set_item_key_safe(struct btrfs_trans_handle * trans,struct btrfs_path * path,const struct btrfs_key * new_key)2719 void btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
2720 			     struct btrfs_path *path,
2721 			     const struct btrfs_key *new_key)
2722 {
2723 	struct btrfs_fs_info *fs_info = trans->fs_info;
2724 	struct btrfs_disk_key disk_key;
2725 	struct extent_buffer *eb;
2726 	int slot;
2727 
2728 	eb = path->nodes[0];
2729 	slot = path->slots[0];
2730 	if (slot > 0) {
2731 		btrfs_item_key(eb, &disk_key, slot - 1);
2732 		if (unlikely(comp_keys(&disk_key, new_key) >= 0)) {
2733 			btrfs_print_leaf(eb);
2734 			btrfs_crit(fs_info,
2735 		"slot %u key (%llu %u %llu) new key (%llu %u %llu)",
2736 				   slot, btrfs_disk_key_objectid(&disk_key),
2737 				   btrfs_disk_key_type(&disk_key),
2738 				   btrfs_disk_key_offset(&disk_key),
2739 				   new_key->objectid, new_key->type,
2740 				   new_key->offset);
2741 			BUG();
2742 		}
2743 	}
2744 	if (slot < btrfs_header_nritems(eb) - 1) {
2745 		btrfs_item_key(eb, &disk_key, slot + 1);
2746 		if (unlikely(comp_keys(&disk_key, new_key) <= 0)) {
2747 			btrfs_print_leaf(eb);
2748 			btrfs_crit(fs_info,
2749 		"slot %u key (%llu %u %llu) new key (%llu %u %llu)",
2750 				   slot, btrfs_disk_key_objectid(&disk_key),
2751 				   btrfs_disk_key_type(&disk_key),
2752 				   btrfs_disk_key_offset(&disk_key),
2753 				   new_key->objectid, new_key->type,
2754 				   new_key->offset);
2755 			BUG();
2756 		}
2757 	}
2758 
2759 	btrfs_cpu_key_to_disk(&disk_key, new_key);
2760 	btrfs_set_item_key(eb, &disk_key, slot);
2761 	btrfs_mark_buffer_dirty(trans, eb);
2762 	if (slot == 0)
2763 		fixup_low_keys(trans, path, &disk_key, 1);
2764 }
2765 
2766 /*
2767  * Check key order of two sibling extent buffers.
2768  *
2769  * Return true if something is wrong.
2770  * Return false if everything is fine.
2771  *
2772  * Tree-checker only works inside one tree block, thus the following
2773  * corruption can not be detected by tree-checker:
2774  *
2775  * Leaf @left			| Leaf @right
2776  * --------------------------------------------------------------
2777  * | 1 | 2 | 3 | 4 | 5 | f6 |   | 7 | 8 |
2778  *
2779  * Key f6 in leaf @left itself is valid, but not valid when the next
2780  * key in leaf @right is 7.
2781  * This can only be checked at tree block merge time.
2782  * And since tree checker has ensured all key order in each tree block
2783  * is correct, we only need to bother the last key of @left and the first
2784  * key of @right.
2785  */
check_sibling_keys(struct extent_buffer * left,struct extent_buffer * right)2786 static bool check_sibling_keys(struct extent_buffer *left,
2787 			       struct extent_buffer *right)
2788 {
2789 	struct btrfs_key left_last;
2790 	struct btrfs_key right_first;
2791 	int level = btrfs_header_level(left);
2792 	int nr_left = btrfs_header_nritems(left);
2793 	int nr_right = btrfs_header_nritems(right);
2794 
2795 	/* No key to check in one of the tree blocks */
2796 	if (!nr_left || !nr_right)
2797 		return false;
2798 
2799 	if (level) {
2800 		btrfs_node_key_to_cpu(left, &left_last, nr_left - 1);
2801 		btrfs_node_key_to_cpu(right, &right_first, 0);
2802 	} else {
2803 		btrfs_item_key_to_cpu(left, &left_last, nr_left - 1);
2804 		btrfs_item_key_to_cpu(right, &right_first, 0);
2805 	}
2806 
2807 	if (unlikely(btrfs_comp_cpu_keys(&left_last, &right_first) >= 0)) {
2808 		btrfs_crit(left->fs_info, "left extent buffer:");
2809 		btrfs_print_tree(left, false);
2810 		btrfs_crit(left->fs_info, "right extent buffer:");
2811 		btrfs_print_tree(right, false);
2812 		btrfs_crit(left->fs_info,
2813 "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)",
2814 			   left_last.objectid, left_last.type,
2815 			   left_last.offset, right_first.objectid,
2816 			   right_first.type, right_first.offset);
2817 		return true;
2818 	}
2819 	return false;
2820 }
2821 
2822 /*
2823  * try to push data from one node into the next node left in the
2824  * tree.
2825  *
2826  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2827  * error, and > 0 if there was no room in the left hand block.
2828  */
push_node_left(struct btrfs_trans_handle * trans,struct extent_buffer * dst,struct extent_buffer * src,int empty)2829 static int push_node_left(struct btrfs_trans_handle *trans,
2830 			  struct extent_buffer *dst,
2831 			  struct extent_buffer *src, int empty)
2832 {
2833 	struct btrfs_fs_info *fs_info = trans->fs_info;
2834 	int push_items = 0;
2835 	int src_nritems;
2836 	int dst_nritems;
2837 	int ret = 0;
2838 
2839 	src_nritems = btrfs_header_nritems(src);
2840 	dst_nritems = btrfs_header_nritems(dst);
2841 	push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
2842 	WARN_ON(btrfs_header_generation(src) != trans->transid);
2843 	WARN_ON(btrfs_header_generation(dst) != trans->transid);
2844 
2845 	if (!empty && src_nritems <= 8)
2846 		return 1;
2847 
2848 	if (push_items <= 0)
2849 		return 1;
2850 
2851 	if (empty) {
2852 		push_items = min(src_nritems, push_items);
2853 		if (push_items < src_nritems) {
2854 			/* leave at least 8 pointers in the node if
2855 			 * we aren't going to empty it
2856 			 */
2857 			if (src_nritems - push_items < 8) {
2858 				if (push_items <= 8)
2859 					return 1;
2860 				push_items -= 8;
2861 			}
2862 		}
2863 	} else
2864 		push_items = min(src_nritems - 8, push_items);
2865 
2866 	/* dst is the left eb, src is the middle eb */
2867 	if (check_sibling_keys(dst, src)) {
2868 		ret = -EUCLEAN;
2869 		btrfs_abort_transaction(trans, ret);
2870 		return ret;
2871 	}
2872 	ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
2873 	if (ret) {
2874 		btrfs_abort_transaction(trans, ret);
2875 		return ret;
2876 	}
2877 	copy_extent_buffer(dst, src,
2878 			   btrfs_node_key_ptr_offset(dst, dst_nritems),
2879 			   btrfs_node_key_ptr_offset(src, 0),
2880 			   push_items * sizeof(struct btrfs_key_ptr));
2881 
2882 	if (push_items < src_nritems) {
2883 		/*
2884 		 * btrfs_tree_mod_log_eb_copy handles logging the move, so we
2885 		 * don't need to do an explicit tree mod log operation for it.
2886 		 */
2887 		memmove_extent_buffer(src, btrfs_node_key_ptr_offset(src, 0),
2888 				      btrfs_node_key_ptr_offset(src, push_items),
2889 				      (src_nritems - push_items) *
2890 				      sizeof(struct btrfs_key_ptr));
2891 	}
2892 	btrfs_set_header_nritems(src, src_nritems - push_items);
2893 	btrfs_set_header_nritems(dst, dst_nritems + push_items);
2894 	btrfs_mark_buffer_dirty(trans, src);
2895 	btrfs_mark_buffer_dirty(trans, dst);
2896 
2897 	return ret;
2898 }
2899 
2900 /*
2901  * try to push data from one node into the next node right in the
2902  * tree.
2903  *
2904  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
2905  * error, and > 0 if there was no room in the right hand block.
2906  *
2907  * this will  only push up to 1/2 the contents of the left node over
2908  */
balance_node_right(struct btrfs_trans_handle * trans,struct extent_buffer * dst,struct extent_buffer * src)2909 static int balance_node_right(struct btrfs_trans_handle *trans,
2910 			      struct extent_buffer *dst,
2911 			      struct extent_buffer *src)
2912 {
2913 	struct btrfs_fs_info *fs_info = trans->fs_info;
2914 	int push_items = 0;
2915 	int max_push;
2916 	int src_nritems;
2917 	int dst_nritems;
2918 	int ret = 0;
2919 
2920 	WARN_ON(btrfs_header_generation(src) != trans->transid);
2921 	WARN_ON(btrfs_header_generation(dst) != trans->transid);
2922 
2923 	src_nritems = btrfs_header_nritems(src);
2924 	dst_nritems = btrfs_header_nritems(dst);
2925 	push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
2926 	if (push_items <= 0)
2927 		return 1;
2928 
2929 	if (src_nritems < 4)
2930 		return 1;
2931 
2932 	max_push = src_nritems / 2 + 1;
2933 	/* don't try to empty the node */
2934 	if (max_push >= src_nritems)
2935 		return 1;
2936 
2937 	if (max_push < push_items)
2938 		push_items = max_push;
2939 
2940 	/* dst is the right eb, src is the middle eb */
2941 	if (check_sibling_keys(src, dst)) {
2942 		ret = -EUCLEAN;
2943 		btrfs_abort_transaction(trans, ret);
2944 		return ret;
2945 	}
2946 
2947 	/*
2948 	 * btrfs_tree_mod_log_eb_copy handles logging the move, so we don't
2949 	 * need to do an explicit tree mod log operation for it.
2950 	 */
2951 	memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(dst, push_items),
2952 				      btrfs_node_key_ptr_offset(dst, 0),
2953 				      (dst_nritems) *
2954 				      sizeof(struct btrfs_key_ptr));
2955 
2956 	ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
2957 					 push_items);
2958 	if (ret) {
2959 		btrfs_abort_transaction(trans, ret);
2960 		return ret;
2961 	}
2962 	copy_extent_buffer(dst, src,
2963 			   btrfs_node_key_ptr_offset(dst, 0),
2964 			   btrfs_node_key_ptr_offset(src, src_nritems - push_items),
2965 			   push_items * sizeof(struct btrfs_key_ptr));
2966 
2967 	btrfs_set_header_nritems(src, src_nritems - push_items);
2968 	btrfs_set_header_nritems(dst, dst_nritems + push_items);
2969 
2970 	btrfs_mark_buffer_dirty(trans, src);
2971 	btrfs_mark_buffer_dirty(trans, dst);
2972 
2973 	return ret;
2974 }
2975 
2976 /*
2977  * helper function to insert a new root level in the tree.
2978  * A new node is allocated, and a single item is inserted to
2979  * point to the existing root
2980  *
2981  * returns zero on success or < 0 on failure.
2982  */
insert_new_root(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int level)2983 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2984 			   struct btrfs_root *root,
2985 			   struct btrfs_path *path, int level)
2986 {
2987 	struct btrfs_fs_info *fs_info = root->fs_info;
2988 	u64 lower_gen;
2989 	struct extent_buffer *lower;
2990 	struct extent_buffer *c;
2991 	struct extent_buffer *old;
2992 	struct btrfs_disk_key lower_key;
2993 	int ret;
2994 
2995 	BUG_ON(path->nodes[level]);
2996 	BUG_ON(path->nodes[level-1] != root->node);
2997 
2998 	lower = path->nodes[level-1];
2999 	if (level == 1)
3000 		btrfs_item_key(lower, &lower_key, 0);
3001 	else
3002 		btrfs_node_key(lower, &lower_key, 0);
3003 
3004 	c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3005 				   &lower_key, level, root->node->start, 0,
3006 				   BTRFS_NESTING_NEW_ROOT);
3007 	if (IS_ERR(c))
3008 		return PTR_ERR(c);
3009 
3010 	root_add_used(root, fs_info->nodesize);
3011 
3012 	btrfs_set_header_nritems(c, 1);
3013 	btrfs_set_node_key(c, &lower_key, 0);
3014 	btrfs_set_node_blockptr(c, 0, lower->start);
3015 	lower_gen = btrfs_header_generation(lower);
3016 	WARN_ON(lower_gen != trans->transid);
3017 
3018 	btrfs_set_node_ptr_generation(c, 0, lower_gen);
3019 
3020 	btrfs_mark_buffer_dirty(trans, c);
3021 
3022 	old = root->node;
3023 	ret = btrfs_tree_mod_log_insert_root(root->node, c, false);
3024 	if (ret < 0) {
3025 		btrfs_free_tree_block(trans, btrfs_root_id(root), c, 0, 1);
3026 		btrfs_tree_unlock(c);
3027 		free_extent_buffer(c);
3028 		return ret;
3029 	}
3030 	rcu_assign_pointer(root->node, c);
3031 
3032 	/* the super has an extra ref to root->node */
3033 	free_extent_buffer(old);
3034 
3035 	add_root_to_dirty_list(root);
3036 	atomic_inc(&c->refs);
3037 	path->nodes[level] = c;
3038 	path->locks[level] = BTRFS_WRITE_LOCK;
3039 	path->slots[level] = 0;
3040 	return 0;
3041 }
3042 
3043 /*
3044  * worker function to insert a single pointer in a node.
3045  * the node should have enough room for the pointer already
3046  *
3047  * slot and level indicate where you want the key to go, and
3048  * blocknr is the block the key points to.
3049  */
insert_ptr(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_disk_key * key,u64 bytenr,int slot,int level)3050 static int insert_ptr(struct btrfs_trans_handle *trans,
3051 		      struct btrfs_path *path,
3052 		      struct btrfs_disk_key *key, u64 bytenr,
3053 		      int slot, int level)
3054 {
3055 	struct extent_buffer *lower;
3056 	int nritems;
3057 	int ret;
3058 
3059 	BUG_ON(!path->nodes[level]);
3060 	btrfs_assert_tree_write_locked(path->nodes[level]);
3061 	lower = path->nodes[level];
3062 	nritems = btrfs_header_nritems(lower);
3063 	BUG_ON(slot > nritems);
3064 	BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
3065 	if (slot != nritems) {
3066 		if (level) {
3067 			ret = btrfs_tree_mod_log_insert_move(lower, slot + 1,
3068 					slot, nritems - slot);
3069 			if (ret < 0) {
3070 				btrfs_abort_transaction(trans, ret);
3071 				return ret;
3072 			}
3073 		}
3074 		memmove_extent_buffer(lower,
3075 			      btrfs_node_key_ptr_offset(lower, slot + 1),
3076 			      btrfs_node_key_ptr_offset(lower, slot),
3077 			      (nritems - slot) * sizeof(struct btrfs_key_ptr));
3078 	}
3079 	if (level) {
3080 		ret = btrfs_tree_mod_log_insert_key(lower, slot,
3081 						    BTRFS_MOD_LOG_KEY_ADD);
3082 		if (ret < 0) {
3083 			btrfs_abort_transaction(trans, ret);
3084 			return ret;
3085 		}
3086 	}
3087 	btrfs_set_node_key(lower, key, slot);
3088 	btrfs_set_node_blockptr(lower, slot, bytenr);
3089 	WARN_ON(trans->transid == 0);
3090 	btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3091 	btrfs_set_header_nritems(lower, nritems + 1);
3092 	btrfs_mark_buffer_dirty(trans, lower);
3093 
3094 	return 0;
3095 }
3096 
3097 /*
3098  * split the node at the specified level in path in two.
3099  * The path is corrected to point to the appropriate node after the split
3100  *
3101  * Before splitting this tries to make some room in the node by pushing
3102  * left and right, if either one works, it returns right away.
3103  *
3104  * returns 0 on success and < 0 on failure
3105  */
split_node(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int level)3106 static noinline int split_node(struct btrfs_trans_handle *trans,
3107 			       struct btrfs_root *root,
3108 			       struct btrfs_path *path, int level)
3109 {
3110 	struct btrfs_fs_info *fs_info = root->fs_info;
3111 	struct extent_buffer *c;
3112 	struct extent_buffer *split;
3113 	struct btrfs_disk_key disk_key;
3114 	int mid;
3115 	int ret;
3116 	u32 c_nritems;
3117 
3118 	c = path->nodes[level];
3119 	WARN_ON(btrfs_header_generation(c) != trans->transid);
3120 	if (c == root->node) {
3121 		/*
3122 		 * trying to split the root, lets make a new one
3123 		 *
3124 		 * tree mod log: We don't log_removal old root in
3125 		 * insert_new_root, because that root buffer will be kept as a
3126 		 * normal node. We are going to log removal of half of the
3127 		 * elements below with btrfs_tree_mod_log_eb_copy(). We're
3128 		 * holding a tree lock on the buffer, which is why we cannot
3129 		 * race with other tree_mod_log users.
3130 		 */
3131 		ret = insert_new_root(trans, root, path, level + 1);
3132 		if (ret)
3133 			return ret;
3134 	} else {
3135 		ret = push_nodes_for_insert(trans, root, path, level);
3136 		c = path->nodes[level];
3137 		if (!ret && btrfs_header_nritems(c) <
3138 		    BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3139 			return 0;
3140 		if (ret < 0)
3141 			return ret;
3142 	}
3143 
3144 	c_nritems = btrfs_header_nritems(c);
3145 	mid = (c_nritems + 1) / 2;
3146 	btrfs_node_key(c, &disk_key, mid);
3147 
3148 	split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3149 				       &disk_key, level, c->start, 0,
3150 				       BTRFS_NESTING_SPLIT);
3151 	if (IS_ERR(split))
3152 		return PTR_ERR(split);
3153 
3154 	root_add_used(root, fs_info->nodesize);
3155 	ASSERT(btrfs_header_level(c) == level);
3156 
3157 	ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
3158 	if (ret) {
3159 		btrfs_tree_unlock(split);
3160 		free_extent_buffer(split);
3161 		btrfs_abort_transaction(trans, ret);
3162 		return ret;
3163 	}
3164 	copy_extent_buffer(split, c,
3165 			   btrfs_node_key_ptr_offset(split, 0),
3166 			   btrfs_node_key_ptr_offset(c, mid),
3167 			   (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3168 	btrfs_set_header_nritems(split, c_nritems - mid);
3169 	btrfs_set_header_nritems(c, mid);
3170 
3171 	btrfs_mark_buffer_dirty(trans, c);
3172 	btrfs_mark_buffer_dirty(trans, split);
3173 
3174 	ret = insert_ptr(trans, path, &disk_key, split->start,
3175 			 path->slots[level + 1] + 1, level + 1);
3176 	if (ret < 0) {
3177 		btrfs_tree_unlock(split);
3178 		free_extent_buffer(split);
3179 		return ret;
3180 	}
3181 
3182 	if (path->slots[level] >= mid) {
3183 		path->slots[level] -= mid;
3184 		btrfs_tree_unlock(c);
3185 		free_extent_buffer(c);
3186 		path->nodes[level] = split;
3187 		path->slots[level + 1] += 1;
3188 	} else {
3189 		btrfs_tree_unlock(split);
3190 		free_extent_buffer(split);
3191 	}
3192 	return 0;
3193 }
3194 
3195 /*
3196  * how many bytes are required to store the items in a leaf.  start
3197  * and nr indicate which items in the leaf to check.  This totals up the
3198  * space used both by the item structs and the item data
3199  */
leaf_space_used(const struct extent_buffer * l,int start,int nr)3200 static int leaf_space_used(const struct extent_buffer *l, int start, int nr)
3201 {
3202 	int data_len;
3203 	int nritems = btrfs_header_nritems(l);
3204 	int end = min(nritems, start + nr) - 1;
3205 
3206 	if (!nr)
3207 		return 0;
3208 	data_len = btrfs_item_offset(l, start) + btrfs_item_size(l, start);
3209 	data_len = data_len - btrfs_item_offset(l, end);
3210 	data_len += sizeof(struct btrfs_item) * nr;
3211 	WARN_ON(data_len < 0);
3212 	return data_len;
3213 }
3214 
3215 /*
3216  * The space between the end of the leaf items and
3217  * the start of the leaf data.  IOW, how much room
3218  * the leaf has left for both items and data
3219  */
btrfs_leaf_free_space(const struct extent_buffer * leaf)3220 int btrfs_leaf_free_space(const struct extent_buffer *leaf)
3221 {
3222 	struct btrfs_fs_info *fs_info = leaf->fs_info;
3223 	int nritems = btrfs_header_nritems(leaf);
3224 	int ret;
3225 
3226 	ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3227 	if (ret < 0) {
3228 		btrfs_crit(fs_info,
3229 			   "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3230 			   ret,
3231 			   (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3232 			   leaf_space_used(leaf, 0, nritems), nritems);
3233 	}
3234 	return ret;
3235 }
3236 
3237 /*
3238  * min slot controls the lowest index we're willing to push to the
3239  * right.  We'll push up to and including min_slot, but no lower
3240  */
__push_leaf_right(struct btrfs_trans_handle * trans,struct btrfs_path * path,int data_size,int empty,struct extent_buffer * right,int free_space,u32 left_nritems,u32 min_slot)3241 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3242 				      struct btrfs_path *path,
3243 				      int data_size, int empty,
3244 				      struct extent_buffer *right,
3245 				      int free_space, u32 left_nritems,
3246 				      u32 min_slot)
3247 {
3248 	struct btrfs_fs_info *fs_info = right->fs_info;
3249 	struct extent_buffer *left = path->nodes[0];
3250 	struct extent_buffer *upper = path->nodes[1];
3251 	struct btrfs_map_token token;
3252 	struct btrfs_disk_key disk_key;
3253 	int slot;
3254 	u32 i;
3255 	int push_space = 0;
3256 	int push_items = 0;
3257 	u32 nr;
3258 	u32 right_nritems;
3259 	u32 data_end;
3260 	u32 this_item_size;
3261 
3262 	if (empty)
3263 		nr = 0;
3264 	else
3265 		nr = max_t(u32, 1, min_slot);
3266 
3267 	if (path->slots[0] >= left_nritems)
3268 		push_space += data_size;
3269 
3270 	slot = path->slots[1];
3271 	i = left_nritems - 1;
3272 	while (i >= nr) {
3273 		if (!empty && push_items > 0) {
3274 			if (path->slots[0] > i)
3275 				break;
3276 			if (path->slots[0] == i) {
3277 				int space = btrfs_leaf_free_space(left);
3278 
3279 				if (space + push_space * 2 > free_space)
3280 					break;
3281 			}
3282 		}
3283 
3284 		if (path->slots[0] == i)
3285 			push_space += data_size;
3286 
3287 		this_item_size = btrfs_item_size(left, i);
3288 		if (this_item_size + sizeof(struct btrfs_item) +
3289 		    push_space > free_space)
3290 			break;
3291 
3292 		push_items++;
3293 		push_space += this_item_size + sizeof(struct btrfs_item);
3294 		if (i == 0)
3295 			break;
3296 		i--;
3297 	}
3298 
3299 	if (push_items == 0)
3300 		goto out_unlock;
3301 
3302 	WARN_ON(!empty && push_items == left_nritems);
3303 
3304 	/* push left to right */
3305 	right_nritems = btrfs_header_nritems(right);
3306 
3307 	push_space = btrfs_item_data_end(left, left_nritems - push_items);
3308 	push_space -= leaf_data_end(left);
3309 
3310 	/* make room in the right data area */
3311 	data_end = leaf_data_end(right);
3312 	memmove_leaf_data(right, data_end - push_space, data_end,
3313 			  BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3314 
3315 	/* copy from the left data area */
3316 	copy_leaf_data(right, left, BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3317 		       leaf_data_end(left), push_space);
3318 
3319 	memmove_leaf_items(right, push_items, 0, right_nritems);
3320 
3321 	/* copy the items from left to right */
3322 	copy_leaf_items(right, left, 0, left_nritems - push_items, push_items);
3323 
3324 	/* update the item pointers */
3325 	btrfs_init_map_token(&token, right);
3326 	right_nritems += push_items;
3327 	btrfs_set_header_nritems(right, right_nritems);
3328 	push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3329 	for (i = 0; i < right_nritems; i++) {
3330 		push_space -= btrfs_token_item_size(&token, i);
3331 		btrfs_set_token_item_offset(&token, i, push_space);
3332 	}
3333 
3334 	left_nritems -= push_items;
3335 	btrfs_set_header_nritems(left, left_nritems);
3336 
3337 	if (left_nritems)
3338 		btrfs_mark_buffer_dirty(trans, left);
3339 	else
3340 		btrfs_clear_buffer_dirty(trans, left);
3341 
3342 	btrfs_mark_buffer_dirty(trans, right);
3343 
3344 	btrfs_item_key(right, &disk_key, 0);
3345 	btrfs_set_node_key(upper, &disk_key, slot + 1);
3346 	btrfs_mark_buffer_dirty(trans, upper);
3347 
3348 	/* then fixup the leaf pointer in the path */
3349 	if (path->slots[0] >= left_nritems) {
3350 		path->slots[0] -= left_nritems;
3351 		if (btrfs_header_nritems(path->nodes[0]) == 0)
3352 			btrfs_clear_buffer_dirty(trans, path->nodes[0]);
3353 		btrfs_tree_unlock(path->nodes[0]);
3354 		free_extent_buffer(path->nodes[0]);
3355 		path->nodes[0] = right;
3356 		path->slots[1] += 1;
3357 	} else {
3358 		btrfs_tree_unlock(right);
3359 		free_extent_buffer(right);
3360 	}
3361 	return 0;
3362 
3363 out_unlock:
3364 	btrfs_tree_unlock(right);
3365 	free_extent_buffer(right);
3366 	return 1;
3367 }
3368 
3369 /*
3370  * push some data in the path leaf to the right, trying to free up at
3371  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3372  *
3373  * returns 1 if the push failed because the other node didn't have enough
3374  * room, 0 if everything worked out and < 0 if there were major errors.
3375  *
3376  * this will push starting from min_slot to the end of the leaf.  It won't
3377  * push any slot lower than min_slot
3378  */
push_leaf_right(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int min_data_size,int data_size,int empty,u32 min_slot)3379 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3380 			   *root, struct btrfs_path *path,
3381 			   int min_data_size, int data_size,
3382 			   int empty, u32 min_slot)
3383 {
3384 	struct extent_buffer *left = path->nodes[0];
3385 	struct extent_buffer *right;
3386 	struct extent_buffer *upper;
3387 	int slot;
3388 	int free_space;
3389 	u32 left_nritems;
3390 	int ret;
3391 
3392 	if (!path->nodes[1])
3393 		return 1;
3394 
3395 	slot = path->slots[1];
3396 	upper = path->nodes[1];
3397 	if (slot >= btrfs_header_nritems(upper) - 1)
3398 		return 1;
3399 
3400 	btrfs_assert_tree_write_locked(path->nodes[1]);
3401 
3402 	right = btrfs_read_node_slot(upper, slot + 1);
3403 	if (IS_ERR(right))
3404 		return PTR_ERR(right);
3405 
3406 	__btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
3407 
3408 	free_space = btrfs_leaf_free_space(right);
3409 	if (free_space < data_size)
3410 		goto out_unlock;
3411 
3412 	ret = btrfs_cow_block(trans, root, right, upper,
3413 			      slot + 1, &right, BTRFS_NESTING_RIGHT_COW);
3414 	if (ret)
3415 		goto out_unlock;
3416 
3417 	left_nritems = btrfs_header_nritems(left);
3418 	if (left_nritems == 0)
3419 		goto out_unlock;
3420 
3421 	if (check_sibling_keys(left, right)) {
3422 		ret = -EUCLEAN;
3423 		btrfs_abort_transaction(trans, ret);
3424 		btrfs_tree_unlock(right);
3425 		free_extent_buffer(right);
3426 		return ret;
3427 	}
3428 	if (path->slots[0] == left_nritems && !empty) {
3429 		/* Key greater than all keys in the leaf, right neighbor has
3430 		 * enough room for it and we're not emptying our leaf to delete
3431 		 * it, therefore use right neighbor to insert the new item and
3432 		 * no need to touch/dirty our left leaf. */
3433 		btrfs_tree_unlock(left);
3434 		free_extent_buffer(left);
3435 		path->nodes[0] = right;
3436 		path->slots[0] = 0;
3437 		path->slots[1]++;
3438 		return 0;
3439 	}
3440 
3441 	return __push_leaf_right(trans, path, min_data_size, empty, right,
3442 				 free_space, left_nritems, min_slot);
3443 out_unlock:
3444 	btrfs_tree_unlock(right);
3445 	free_extent_buffer(right);
3446 	return 1;
3447 }
3448 
3449 /*
3450  * push some data in the path leaf to the left, trying to free up at
3451  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3452  *
3453  * max_slot can put a limit on how far into the leaf we'll push items.  The
3454  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us do all the
3455  * items
3456  */
__push_leaf_left(struct btrfs_trans_handle * trans,struct btrfs_path * path,int data_size,int empty,struct extent_buffer * left,int free_space,u32 right_nritems,u32 max_slot)3457 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3458 				     struct btrfs_path *path, int data_size,
3459 				     int empty, struct extent_buffer *left,
3460 				     int free_space, u32 right_nritems,
3461 				     u32 max_slot)
3462 {
3463 	struct btrfs_fs_info *fs_info = left->fs_info;
3464 	struct btrfs_disk_key disk_key;
3465 	struct extent_buffer *right = path->nodes[0];
3466 	int i;
3467 	int push_space = 0;
3468 	int push_items = 0;
3469 	u32 old_left_nritems;
3470 	u32 nr;
3471 	int ret = 0;
3472 	u32 this_item_size;
3473 	u32 old_left_item_size;
3474 	struct btrfs_map_token token;
3475 
3476 	if (empty)
3477 		nr = min(right_nritems, max_slot);
3478 	else
3479 		nr = min(right_nritems - 1, max_slot);
3480 
3481 	for (i = 0; i < nr; i++) {
3482 		if (!empty && push_items > 0) {
3483 			if (path->slots[0] < i)
3484 				break;
3485 			if (path->slots[0] == i) {
3486 				int space = btrfs_leaf_free_space(right);
3487 
3488 				if (space + push_space * 2 > free_space)
3489 					break;
3490 			}
3491 		}
3492 
3493 		if (path->slots[0] == i)
3494 			push_space += data_size;
3495 
3496 		this_item_size = btrfs_item_size(right, i);
3497 		if (this_item_size + sizeof(struct btrfs_item) + push_space >
3498 		    free_space)
3499 			break;
3500 
3501 		push_items++;
3502 		push_space += this_item_size + sizeof(struct btrfs_item);
3503 	}
3504 
3505 	if (push_items == 0) {
3506 		ret = 1;
3507 		goto out;
3508 	}
3509 	WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3510 
3511 	/* push data from right to left */
3512 	copy_leaf_items(left, right, btrfs_header_nritems(left), 0, push_items);
3513 
3514 	push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3515 		     btrfs_item_offset(right, push_items - 1);
3516 
3517 	copy_leaf_data(left, right, leaf_data_end(left) - push_space,
3518 		       btrfs_item_offset(right, push_items - 1), push_space);
3519 	old_left_nritems = btrfs_header_nritems(left);
3520 	BUG_ON(old_left_nritems <= 0);
3521 
3522 	btrfs_init_map_token(&token, left);
3523 	old_left_item_size = btrfs_item_offset(left, old_left_nritems - 1);
3524 	for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3525 		u32 ioff;
3526 
3527 		ioff = btrfs_token_item_offset(&token, i);
3528 		btrfs_set_token_item_offset(&token, i,
3529 		      ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size));
3530 	}
3531 	btrfs_set_header_nritems(left, old_left_nritems + push_items);
3532 
3533 	/* fixup right node */
3534 	if (push_items > right_nritems)
3535 		WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3536 		       right_nritems);
3537 
3538 	if (push_items < right_nritems) {
3539 		push_space = btrfs_item_offset(right, push_items - 1) -
3540 						  leaf_data_end(right);
3541 		memmove_leaf_data(right,
3542 				  BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3543 				  leaf_data_end(right), push_space);
3544 
3545 		memmove_leaf_items(right, 0, push_items,
3546 				   btrfs_header_nritems(right) - push_items);
3547 	}
3548 
3549 	btrfs_init_map_token(&token, right);
3550 	right_nritems -= push_items;
3551 	btrfs_set_header_nritems(right, right_nritems);
3552 	push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3553 	for (i = 0; i < right_nritems; i++) {
3554 		push_space = push_space - btrfs_token_item_size(&token, i);
3555 		btrfs_set_token_item_offset(&token, i, push_space);
3556 	}
3557 
3558 	btrfs_mark_buffer_dirty(trans, left);
3559 	if (right_nritems)
3560 		btrfs_mark_buffer_dirty(trans, right);
3561 	else
3562 		btrfs_clear_buffer_dirty(trans, right);
3563 
3564 	btrfs_item_key(right, &disk_key, 0);
3565 	fixup_low_keys(trans, path, &disk_key, 1);
3566 
3567 	/* then fixup the leaf pointer in the path */
3568 	if (path->slots[0] < push_items) {
3569 		path->slots[0] += old_left_nritems;
3570 		btrfs_tree_unlock(path->nodes[0]);
3571 		free_extent_buffer(path->nodes[0]);
3572 		path->nodes[0] = left;
3573 		path->slots[1] -= 1;
3574 	} else {
3575 		btrfs_tree_unlock(left);
3576 		free_extent_buffer(left);
3577 		path->slots[0] -= push_items;
3578 	}
3579 	BUG_ON(path->slots[0] < 0);
3580 	return ret;
3581 out:
3582 	btrfs_tree_unlock(left);
3583 	free_extent_buffer(left);
3584 	return ret;
3585 }
3586 
3587 /*
3588  * push some data in the path leaf to the left, trying to free up at
3589  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3590  *
3591  * max_slot can put a limit on how far into the leaf we'll push items.  The
3592  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us push all the
3593  * items
3594  */
push_leaf_left(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int min_data_size,int data_size,int empty,u32 max_slot)3595 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3596 			  *root, struct btrfs_path *path, int min_data_size,
3597 			  int data_size, int empty, u32 max_slot)
3598 {
3599 	struct extent_buffer *right = path->nodes[0];
3600 	struct extent_buffer *left;
3601 	int slot;
3602 	int free_space;
3603 	u32 right_nritems;
3604 	int ret = 0;
3605 
3606 	slot = path->slots[1];
3607 	if (slot == 0)
3608 		return 1;
3609 	if (!path->nodes[1])
3610 		return 1;
3611 
3612 	right_nritems = btrfs_header_nritems(right);
3613 	if (right_nritems == 0)
3614 		return 1;
3615 
3616 	btrfs_assert_tree_write_locked(path->nodes[1]);
3617 
3618 	left = btrfs_read_node_slot(path->nodes[1], slot - 1);
3619 	if (IS_ERR(left))
3620 		return PTR_ERR(left);
3621 
3622 	__btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
3623 
3624 	free_space = btrfs_leaf_free_space(left);
3625 	if (free_space < data_size) {
3626 		ret = 1;
3627 		goto out;
3628 	}
3629 
3630 	ret = btrfs_cow_block(trans, root, left,
3631 			      path->nodes[1], slot - 1, &left,
3632 			      BTRFS_NESTING_LEFT_COW);
3633 	if (ret) {
3634 		/* we hit -ENOSPC, but it isn't fatal here */
3635 		if (ret == -ENOSPC)
3636 			ret = 1;
3637 		goto out;
3638 	}
3639 
3640 	if (check_sibling_keys(left, right)) {
3641 		ret = -EUCLEAN;
3642 		btrfs_abort_transaction(trans, ret);
3643 		goto out;
3644 	}
3645 	return __push_leaf_left(trans, path, min_data_size, empty, left,
3646 				free_space, right_nritems, max_slot);
3647 out:
3648 	btrfs_tree_unlock(left);
3649 	free_extent_buffer(left);
3650 	return ret;
3651 }
3652 
3653 /*
3654  * split the path's leaf in two, making sure there is at least data_size
3655  * available for the resulting leaf level of the path.
3656  */
copy_for_split(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct extent_buffer * l,struct extent_buffer * right,int slot,int mid,int nritems)3657 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
3658 				   struct btrfs_path *path,
3659 				   struct extent_buffer *l,
3660 				   struct extent_buffer *right,
3661 				   int slot, int mid, int nritems)
3662 {
3663 	struct btrfs_fs_info *fs_info = trans->fs_info;
3664 	int data_copy_size;
3665 	int rt_data_off;
3666 	int i;
3667 	int ret;
3668 	struct btrfs_disk_key disk_key;
3669 	struct btrfs_map_token token;
3670 
3671 	nritems = nritems - mid;
3672 	btrfs_set_header_nritems(right, nritems);
3673 	data_copy_size = btrfs_item_data_end(l, mid) - leaf_data_end(l);
3674 
3675 	copy_leaf_items(right, l, 0, mid, nritems);
3676 
3677 	copy_leaf_data(right, l, BTRFS_LEAF_DATA_SIZE(fs_info) - data_copy_size,
3678 		       leaf_data_end(l), data_copy_size);
3679 
3680 	rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_data_end(l, mid);
3681 
3682 	btrfs_init_map_token(&token, right);
3683 	for (i = 0; i < nritems; i++) {
3684 		u32 ioff;
3685 
3686 		ioff = btrfs_token_item_offset(&token, i);
3687 		btrfs_set_token_item_offset(&token, i, ioff + rt_data_off);
3688 	}
3689 
3690 	btrfs_set_header_nritems(l, mid);
3691 	btrfs_item_key(right, &disk_key, 0);
3692 	ret = insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
3693 	if (ret < 0)
3694 		return ret;
3695 
3696 	btrfs_mark_buffer_dirty(trans, right);
3697 	btrfs_mark_buffer_dirty(trans, l);
3698 	BUG_ON(path->slots[0] != slot);
3699 
3700 	if (mid <= slot) {
3701 		btrfs_tree_unlock(path->nodes[0]);
3702 		free_extent_buffer(path->nodes[0]);
3703 		path->nodes[0] = right;
3704 		path->slots[0] -= mid;
3705 		path->slots[1] += 1;
3706 	} else {
3707 		btrfs_tree_unlock(right);
3708 		free_extent_buffer(right);
3709 	}
3710 
3711 	BUG_ON(path->slots[0] < 0);
3712 
3713 	return 0;
3714 }
3715 
3716 /*
3717  * double splits happen when we need to insert a big item in the middle
3718  * of a leaf.  A double split can leave us with 3 mostly empty leaves:
3719  * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3720  *          A                 B                 C
3721  *
3722  * We avoid this by trying to push the items on either side of our target
3723  * into the adjacent leaves.  If all goes well we can avoid the double split
3724  * completely.
3725  */
push_for_double_split(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int data_size)3726 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
3727 					  struct btrfs_root *root,
3728 					  struct btrfs_path *path,
3729 					  int data_size)
3730 {
3731 	int ret;
3732 	int progress = 0;
3733 	int slot;
3734 	u32 nritems;
3735 	int space_needed = data_size;
3736 
3737 	slot = path->slots[0];
3738 	if (slot < btrfs_header_nritems(path->nodes[0]))
3739 		space_needed -= btrfs_leaf_free_space(path->nodes[0]);
3740 
3741 	/*
3742 	 * try to push all the items after our slot into the
3743 	 * right leaf
3744 	 */
3745 	ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
3746 	if (ret < 0)
3747 		return ret;
3748 
3749 	if (ret == 0)
3750 		progress++;
3751 
3752 	nritems = btrfs_header_nritems(path->nodes[0]);
3753 	/*
3754 	 * our goal is to get our slot at the start or end of a leaf.  If
3755 	 * we've done so we're done
3756 	 */
3757 	if (path->slots[0] == 0 || path->slots[0] == nritems)
3758 		return 0;
3759 
3760 	if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
3761 		return 0;
3762 
3763 	/* try to push all the items before our slot into the next leaf */
3764 	slot = path->slots[0];
3765 	space_needed = data_size;
3766 	if (slot > 0)
3767 		space_needed -= btrfs_leaf_free_space(path->nodes[0]);
3768 	ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
3769 	if (ret < 0)
3770 		return ret;
3771 
3772 	if (ret == 0)
3773 		progress++;
3774 
3775 	if (progress)
3776 		return 0;
3777 	return 1;
3778 }
3779 
3780 /*
3781  * split the path's leaf in two, making sure there is at least data_size
3782  * available for the resulting leaf level of the path.
3783  *
3784  * returns 0 if all went well and < 0 on failure.
3785  */
split_leaf(struct btrfs_trans_handle * trans,struct btrfs_root * root,const struct btrfs_key * ins_key,struct btrfs_path * path,int data_size,int extend)3786 static noinline int split_leaf(struct btrfs_trans_handle *trans,
3787 			       struct btrfs_root *root,
3788 			       const struct btrfs_key *ins_key,
3789 			       struct btrfs_path *path, int data_size,
3790 			       int extend)
3791 {
3792 	struct btrfs_disk_key disk_key;
3793 	struct extent_buffer *l;
3794 	u32 nritems;
3795 	int mid;
3796 	int slot;
3797 	struct extent_buffer *right;
3798 	struct btrfs_fs_info *fs_info = root->fs_info;
3799 	int ret = 0;
3800 	int wret;
3801 	int split;
3802 	int num_doubles = 0;
3803 	int tried_avoid_double = 0;
3804 
3805 	l = path->nodes[0];
3806 	slot = path->slots[0];
3807 	if (extend && data_size + btrfs_item_size(l, slot) +
3808 	    sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
3809 		return -EOVERFLOW;
3810 
3811 	/* first try to make some room by pushing left and right */
3812 	if (data_size && path->nodes[1]) {
3813 		int space_needed = data_size;
3814 
3815 		if (slot < btrfs_header_nritems(l))
3816 			space_needed -= btrfs_leaf_free_space(l);
3817 
3818 		wret = push_leaf_right(trans, root, path, space_needed,
3819 				       space_needed, 0, 0);
3820 		if (wret < 0)
3821 			return wret;
3822 		if (wret) {
3823 			space_needed = data_size;
3824 			if (slot > 0)
3825 				space_needed -= btrfs_leaf_free_space(l);
3826 			wret = push_leaf_left(trans, root, path, space_needed,
3827 					      space_needed, 0, (u32)-1);
3828 			if (wret < 0)
3829 				return wret;
3830 		}
3831 		l = path->nodes[0];
3832 
3833 		/* did the pushes work? */
3834 		if (btrfs_leaf_free_space(l) >= data_size)
3835 			return 0;
3836 	}
3837 
3838 	if (!path->nodes[1]) {
3839 		ret = insert_new_root(trans, root, path, 1);
3840 		if (ret)
3841 			return ret;
3842 	}
3843 again:
3844 	split = 1;
3845 	l = path->nodes[0];
3846 	slot = path->slots[0];
3847 	nritems = btrfs_header_nritems(l);
3848 	mid = (nritems + 1) / 2;
3849 
3850 	if (mid <= slot) {
3851 		if (nritems == 1 ||
3852 		    leaf_space_used(l, mid, nritems - mid) + data_size >
3853 			BTRFS_LEAF_DATA_SIZE(fs_info)) {
3854 			if (slot >= nritems) {
3855 				split = 0;
3856 			} else {
3857 				mid = slot;
3858 				if (mid != nritems &&
3859 				    leaf_space_used(l, mid, nritems - mid) +
3860 				    data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
3861 					if (data_size && !tried_avoid_double)
3862 						goto push_for_double;
3863 					split = 2;
3864 				}
3865 			}
3866 		}
3867 	} else {
3868 		if (leaf_space_used(l, 0, mid) + data_size >
3869 			BTRFS_LEAF_DATA_SIZE(fs_info)) {
3870 			if (!extend && data_size && slot == 0) {
3871 				split = 0;
3872 			} else if ((extend || !data_size) && slot == 0) {
3873 				mid = 1;
3874 			} else {
3875 				mid = slot;
3876 				if (mid != nritems &&
3877 				    leaf_space_used(l, mid, nritems - mid) +
3878 				    data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
3879 					if (data_size && !tried_avoid_double)
3880 						goto push_for_double;
3881 					split = 2;
3882 				}
3883 			}
3884 		}
3885 	}
3886 
3887 	if (split == 0)
3888 		btrfs_cpu_key_to_disk(&disk_key, ins_key);
3889 	else
3890 		btrfs_item_key(l, &disk_key, mid);
3891 
3892 	/*
3893 	 * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double
3894 	 * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES
3895 	 * subclasses, which is 8 at the time of this patch, and we've maxed it
3896 	 * out.  In the future we could add a
3897 	 * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just
3898 	 * use BTRFS_NESTING_NEW_ROOT.
3899 	 */
3900 	right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3901 				       &disk_key, 0, l->start, 0,
3902 				       num_doubles ? BTRFS_NESTING_NEW_ROOT :
3903 				       BTRFS_NESTING_SPLIT);
3904 	if (IS_ERR(right))
3905 		return PTR_ERR(right);
3906 
3907 	root_add_used(root, fs_info->nodesize);
3908 
3909 	if (split == 0) {
3910 		if (mid <= slot) {
3911 			btrfs_set_header_nritems(right, 0);
3912 			ret = insert_ptr(trans, path, &disk_key,
3913 					 right->start, path->slots[1] + 1, 1);
3914 			if (ret < 0) {
3915 				btrfs_tree_unlock(right);
3916 				free_extent_buffer(right);
3917 				return ret;
3918 			}
3919 			btrfs_tree_unlock(path->nodes[0]);
3920 			free_extent_buffer(path->nodes[0]);
3921 			path->nodes[0] = right;
3922 			path->slots[0] = 0;
3923 			path->slots[1] += 1;
3924 		} else {
3925 			btrfs_set_header_nritems(right, 0);
3926 			ret = insert_ptr(trans, path, &disk_key,
3927 					 right->start, path->slots[1], 1);
3928 			if (ret < 0) {
3929 				btrfs_tree_unlock(right);
3930 				free_extent_buffer(right);
3931 				return ret;
3932 			}
3933 			btrfs_tree_unlock(path->nodes[0]);
3934 			free_extent_buffer(path->nodes[0]);
3935 			path->nodes[0] = right;
3936 			path->slots[0] = 0;
3937 			if (path->slots[1] == 0)
3938 				fixup_low_keys(trans, path, &disk_key, 1);
3939 		}
3940 		/*
3941 		 * We create a new leaf 'right' for the required ins_len and
3942 		 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
3943 		 * the content of ins_len to 'right'.
3944 		 */
3945 		return ret;
3946 	}
3947 
3948 	ret = copy_for_split(trans, path, l, right, slot, mid, nritems);
3949 	if (ret < 0) {
3950 		btrfs_tree_unlock(right);
3951 		free_extent_buffer(right);
3952 		return ret;
3953 	}
3954 
3955 	if (split == 2) {
3956 		BUG_ON(num_doubles != 0);
3957 		num_doubles++;
3958 		goto again;
3959 	}
3960 
3961 	return 0;
3962 
3963 push_for_double:
3964 	push_for_double_split(trans, root, path, data_size);
3965 	tried_avoid_double = 1;
3966 	if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
3967 		return 0;
3968 	goto again;
3969 }
3970 
setup_leaf_for_split(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int ins_len)3971 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3972 					 struct btrfs_root *root,
3973 					 struct btrfs_path *path, int ins_len)
3974 {
3975 	struct btrfs_key key;
3976 	struct extent_buffer *leaf;
3977 	struct btrfs_file_extent_item *fi;
3978 	u64 extent_len = 0;
3979 	u32 item_size;
3980 	int ret;
3981 
3982 	leaf = path->nodes[0];
3983 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3984 
3985 	BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3986 	       key.type != BTRFS_EXTENT_CSUM_KEY);
3987 
3988 	if (btrfs_leaf_free_space(leaf) >= ins_len)
3989 		return 0;
3990 
3991 	item_size = btrfs_item_size(leaf, path->slots[0]);
3992 	if (key.type == BTRFS_EXTENT_DATA_KEY) {
3993 		fi = btrfs_item_ptr(leaf, path->slots[0],
3994 				    struct btrfs_file_extent_item);
3995 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3996 	}
3997 	btrfs_release_path(path);
3998 
3999 	path->keep_locks = 1;
4000 	path->search_for_split = 1;
4001 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4002 	path->search_for_split = 0;
4003 	if (ret > 0)
4004 		ret = -EAGAIN;
4005 	if (ret < 0)
4006 		goto err;
4007 
4008 	ret = -EAGAIN;
4009 	leaf = path->nodes[0];
4010 	/* if our item isn't there, return now */
4011 	if (item_size != btrfs_item_size(leaf, path->slots[0]))
4012 		goto err;
4013 
4014 	/* the leaf has  changed, it now has room.  return now */
4015 	if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
4016 		goto err;
4017 
4018 	if (key.type == BTRFS_EXTENT_DATA_KEY) {
4019 		fi = btrfs_item_ptr(leaf, path->slots[0],
4020 				    struct btrfs_file_extent_item);
4021 		if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4022 			goto err;
4023 	}
4024 
4025 	ret = split_leaf(trans, root, &key, path, ins_len, 1);
4026 	if (ret)
4027 		goto err;
4028 
4029 	path->keep_locks = 0;
4030 	btrfs_unlock_up_safe(path, 1);
4031 	return 0;
4032 err:
4033 	path->keep_locks = 0;
4034 	return ret;
4035 }
4036 
split_item(struct btrfs_trans_handle * trans,struct btrfs_path * path,const struct btrfs_key * new_key,unsigned long split_offset)4037 static noinline int split_item(struct btrfs_trans_handle *trans,
4038 			       struct btrfs_path *path,
4039 			       const struct btrfs_key *new_key,
4040 			       unsigned long split_offset)
4041 {
4042 	struct extent_buffer *leaf;
4043 	int orig_slot, slot;
4044 	char *buf;
4045 	u32 nritems;
4046 	u32 item_size;
4047 	u32 orig_offset;
4048 	struct btrfs_disk_key disk_key;
4049 
4050 	leaf = path->nodes[0];
4051 	/*
4052 	 * Shouldn't happen because the caller must have previously called
4053 	 * setup_leaf_for_split() to make room for the new item in the leaf.
4054 	 */
4055 	if (WARN_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)))
4056 		return -ENOSPC;
4057 
4058 	orig_slot = path->slots[0];
4059 	orig_offset = btrfs_item_offset(leaf, path->slots[0]);
4060 	item_size = btrfs_item_size(leaf, path->slots[0]);
4061 
4062 	buf = kmalloc(item_size, GFP_NOFS);
4063 	if (!buf)
4064 		return -ENOMEM;
4065 
4066 	read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4067 			    path->slots[0]), item_size);
4068 
4069 	slot = path->slots[0] + 1;
4070 	nritems = btrfs_header_nritems(leaf);
4071 	if (slot != nritems) {
4072 		/* shift the items */
4073 		memmove_leaf_items(leaf, slot + 1, slot, nritems - slot);
4074 	}
4075 
4076 	btrfs_cpu_key_to_disk(&disk_key, new_key);
4077 	btrfs_set_item_key(leaf, &disk_key, slot);
4078 
4079 	btrfs_set_item_offset(leaf, slot, orig_offset);
4080 	btrfs_set_item_size(leaf, slot, item_size - split_offset);
4081 
4082 	btrfs_set_item_offset(leaf, orig_slot,
4083 				 orig_offset + item_size - split_offset);
4084 	btrfs_set_item_size(leaf, orig_slot, split_offset);
4085 
4086 	btrfs_set_header_nritems(leaf, nritems + 1);
4087 
4088 	/* write the data for the start of the original item */
4089 	write_extent_buffer(leaf, buf,
4090 			    btrfs_item_ptr_offset(leaf, path->slots[0]),
4091 			    split_offset);
4092 
4093 	/* write the data for the new item */
4094 	write_extent_buffer(leaf, buf + split_offset,
4095 			    btrfs_item_ptr_offset(leaf, slot),
4096 			    item_size - split_offset);
4097 	btrfs_mark_buffer_dirty(trans, leaf);
4098 
4099 	BUG_ON(btrfs_leaf_free_space(leaf) < 0);
4100 	kfree(buf);
4101 	return 0;
4102 }
4103 
4104 /*
4105  * This function splits a single item into two items,
4106  * giving 'new_key' to the new item and splitting the
4107  * old one at split_offset (from the start of the item).
4108  *
4109  * The path may be released by this operation.  After
4110  * the split, the path is pointing to the old item.  The
4111  * new item is going to be in the same node as the old one.
4112  *
4113  * Note, the item being split must be smaller enough to live alone on
4114  * a tree block with room for one extra struct btrfs_item
4115  *
4116  * This allows us to split the item in place, keeping a lock on the
4117  * leaf the entire time.
4118  */
btrfs_split_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,const struct btrfs_key * new_key,unsigned long split_offset)4119 int btrfs_split_item(struct btrfs_trans_handle *trans,
4120 		     struct btrfs_root *root,
4121 		     struct btrfs_path *path,
4122 		     const struct btrfs_key *new_key,
4123 		     unsigned long split_offset)
4124 {
4125 	int ret;
4126 	ret = setup_leaf_for_split(trans, root, path,
4127 				   sizeof(struct btrfs_item));
4128 	if (ret)
4129 		return ret;
4130 
4131 	ret = split_item(trans, path, new_key, split_offset);
4132 	return ret;
4133 }
4134 
4135 /*
4136  * make the item pointed to by the path smaller.  new_size indicates
4137  * how small to make it, and from_end tells us if we just chop bytes
4138  * off the end of the item or if we shift the item to chop bytes off
4139  * the front.
4140  */
btrfs_truncate_item(struct btrfs_trans_handle * trans,struct btrfs_path * path,u32 new_size,int from_end)4141 void btrfs_truncate_item(struct btrfs_trans_handle *trans,
4142 			 struct btrfs_path *path, u32 new_size, int from_end)
4143 {
4144 	int slot;
4145 	struct extent_buffer *leaf;
4146 	u32 nritems;
4147 	unsigned int data_end;
4148 	unsigned int old_data_start;
4149 	unsigned int old_size;
4150 	unsigned int size_diff;
4151 	int i;
4152 	struct btrfs_map_token token;
4153 
4154 	leaf = path->nodes[0];
4155 	slot = path->slots[0];
4156 
4157 	old_size = btrfs_item_size(leaf, slot);
4158 	if (old_size == new_size)
4159 		return;
4160 
4161 	nritems = btrfs_header_nritems(leaf);
4162 	data_end = leaf_data_end(leaf);
4163 
4164 	old_data_start = btrfs_item_offset(leaf, slot);
4165 
4166 	size_diff = old_size - new_size;
4167 
4168 	BUG_ON(slot < 0);
4169 	BUG_ON(slot >= nritems);
4170 
4171 	/*
4172 	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4173 	 */
4174 	/* first correct the data pointers */
4175 	btrfs_init_map_token(&token, leaf);
4176 	for (i = slot; i < nritems; i++) {
4177 		u32 ioff;
4178 
4179 		ioff = btrfs_token_item_offset(&token, i);
4180 		btrfs_set_token_item_offset(&token, i, ioff + size_diff);
4181 	}
4182 
4183 	/* shift the data */
4184 	if (from_end) {
4185 		memmove_leaf_data(leaf, data_end + size_diff, data_end,
4186 				  old_data_start + new_size - data_end);
4187 	} else {
4188 		struct btrfs_disk_key disk_key;
4189 		u64 offset;
4190 
4191 		btrfs_item_key(leaf, &disk_key, slot);
4192 
4193 		if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4194 			unsigned long ptr;
4195 			struct btrfs_file_extent_item *fi;
4196 
4197 			fi = btrfs_item_ptr(leaf, slot,
4198 					    struct btrfs_file_extent_item);
4199 			fi = (struct btrfs_file_extent_item *)(
4200 			     (unsigned long)fi - size_diff);
4201 
4202 			if (btrfs_file_extent_type(leaf, fi) ==
4203 			    BTRFS_FILE_EXTENT_INLINE) {
4204 				ptr = btrfs_item_ptr_offset(leaf, slot);
4205 				memmove_extent_buffer(leaf, ptr,
4206 				      (unsigned long)fi,
4207 				      BTRFS_FILE_EXTENT_INLINE_DATA_START);
4208 			}
4209 		}
4210 
4211 		memmove_leaf_data(leaf, data_end + size_diff, data_end,
4212 				  old_data_start - data_end);
4213 
4214 		offset = btrfs_disk_key_offset(&disk_key);
4215 		btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4216 		btrfs_set_item_key(leaf, &disk_key, slot);
4217 		if (slot == 0)
4218 			fixup_low_keys(trans, path, &disk_key, 1);
4219 	}
4220 
4221 	btrfs_set_item_size(leaf, slot, new_size);
4222 	btrfs_mark_buffer_dirty(trans, leaf);
4223 
4224 	if (btrfs_leaf_free_space(leaf) < 0) {
4225 		btrfs_print_leaf(leaf);
4226 		BUG();
4227 	}
4228 }
4229 
4230 /*
4231  * make the item pointed to by the path bigger, data_size is the added size.
4232  */
btrfs_extend_item(struct btrfs_trans_handle * trans,struct btrfs_path * path,u32 data_size)4233 void btrfs_extend_item(struct btrfs_trans_handle *trans,
4234 		       struct btrfs_path *path, u32 data_size)
4235 {
4236 	int slot;
4237 	struct extent_buffer *leaf;
4238 	u32 nritems;
4239 	unsigned int data_end;
4240 	unsigned int old_data;
4241 	unsigned int old_size;
4242 	int i;
4243 	struct btrfs_map_token token;
4244 
4245 	leaf = path->nodes[0];
4246 
4247 	nritems = btrfs_header_nritems(leaf);
4248 	data_end = leaf_data_end(leaf);
4249 
4250 	if (btrfs_leaf_free_space(leaf) < data_size) {
4251 		btrfs_print_leaf(leaf);
4252 		BUG();
4253 	}
4254 	slot = path->slots[0];
4255 	old_data = btrfs_item_data_end(leaf, slot);
4256 
4257 	BUG_ON(slot < 0);
4258 	if (slot >= nritems) {
4259 		btrfs_print_leaf(leaf);
4260 		btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d",
4261 			   slot, nritems);
4262 		BUG();
4263 	}
4264 
4265 	/*
4266 	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4267 	 */
4268 	/* first correct the data pointers */
4269 	btrfs_init_map_token(&token, leaf);
4270 	for (i = slot; i < nritems; i++) {
4271 		u32 ioff;
4272 
4273 		ioff = btrfs_token_item_offset(&token, i);
4274 		btrfs_set_token_item_offset(&token, i, ioff - data_size);
4275 	}
4276 
4277 	/* shift the data */
4278 	memmove_leaf_data(leaf, data_end - data_size, data_end,
4279 			  old_data - data_end);
4280 
4281 	data_end = old_data;
4282 	old_size = btrfs_item_size(leaf, slot);
4283 	btrfs_set_item_size(leaf, slot, old_size + data_size);
4284 	btrfs_mark_buffer_dirty(trans, leaf);
4285 
4286 	if (btrfs_leaf_free_space(leaf) < 0) {
4287 		btrfs_print_leaf(leaf);
4288 		BUG();
4289 	}
4290 }
4291 
4292 /*
4293  * Make space in the node before inserting one or more items.
4294  *
4295  * @trans:	transaction handle
4296  * @root:	root we are inserting items to
4297  * @path:	points to the leaf/slot where we are going to insert new items
4298  * @batch:      information about the batch of items to insert
4299  *
4300  * Main purpose is to save stack depth by doing the bulk of the work in a
4301  * function that doesn't call btrfs_search_slot
4302  */
setup_items_for_insert(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,const struct btrfs_item_batch * batch)4303 static void setup_items_for_insert(struct btrfs_trans_handle *trans,
4304 				   struct btrfs_root *root, struct btrfs_path *path,
4305 				   const struct btrfs_item_batch *batch)
4306 {
4307 	struct btrfs_fs_info *fs_info = root->fs_info;
4308 	int i;
4309 	u32 nritems;
4310 	unsigned int data_end;
4311 	struct btrfs_disk_key disk_key;
4312 	struct extent_buffer *leaf;
4313 	int slot;
4314 	struct btrfs_map_token token;
4315 	u32 total_size;
4316 
4317 	/*
4318 	 * Before anything else, update keys in the parent and other ancestors
4319 	 * if needed, then release the write locks on them, so that other tasks
4320 	 * can use them while we modify the leaf.
4321 	 */
4322 	if (path->slots[0] == 0) {
4323 		btrfs_cpu_key_to_disk(&disk_key, &batch->keys[0]);
4324 		fixup_low_keys(trans, path, &disk_key, 1);
4325 	}
4326 	btrfs_unlock_up_safe(path, 1);
4327 
4328 	leaf = path->nodes[0];
4329 	slot = path->slots[0];
4330 
4331 	nritems = btrfs_header_nritems(leaf);
4332 	data_end = leaf_data_end(leaf);
4333 	total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item));
4334 
4335 	if (btrfs_leaf_free_space(leaf) < total_size) {
4336 		btrfs_print_leaf(leaf);
4337 		btrfs_crit(fs_info, "not enough freespace need %u have %d",
4338 			   total_size, btrfs_leaf_free_space(leaf));
4339 		BUG();
4340 	}
4341 
4342 	btrfs_init_map_token(&token, leaf);
4343 	if (slot != nritems) {
4344 		unsigned int old_data = btrfs_item_data_end(leaf, slot);
4345 
4346 		if (old_data < data_end) {
4347 			btrfs_print_leaf(leaf);
4348 			btrfs_crit(fs_info,
4349 		"item at slot %d with data offset %u beyond data end of leaf %u",
4350 				   slot, old_data, data_end);
4351 			BUG();
4352 		}
4353 		/*
4354 		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4355 		 */
4356 		/* first correct the data pointers */
4357 		for (i = slot; i < nritems; i++) {
4358 			u32 ioff;
4359 
4360 			ioff = btrfs_token_item_offset(&token, i);
4361 			btrfs_set_token_item_offset(&token, i,
4362 						       ioff - batch->total_data_size);
4363 		}
4364 		/* shift the items */
4365 		memmove_leaf_items(leaf, slot + batch->nr, slot, nritems - slot);
4366 
4367 		/* shift the data */
4368 		memmove_leaf_data(leaf, data_end - batch->total_data_size,
4369 				  data_end, old_data - data_end);
4370 		data_end = old_data;
4371 	}
4372 
4373 	/* setup the item for the new data */
4374 	for (i = 0; i < batch->nr; i++) {
4375 		btrfs_cpu_key_to_disk(&disk_key, &batch->keys[i]);
4376 		btrfs_set_item_key(leaf, &disk_key, slot + i);
4377 		data_end -= batch->data_sizes[i];
4378 		btrfs_set_token_item_offset(&token, slot + i, data_end);
4379 		btrfs_set_token_item_size(&token, slot + i, batch->data_sizes[i]);
4380 	}
4381 
4382 	btrfs_set_header_nritems(leaf, nritems + batch->nr);
4383 	btrfs_mark_buffer_dirty(trans, leaf);
4384 
4385 	if (btrfs_leaf_free_space(leaf) < 0) {
4386 		btrfs_print_leaf(leaf);
4387 		BUG();
4388 	}
4389 }
4390 
4391 /*
4392  * Insert a new item into a leaf.
4393  *
4394  * @trans:     Transaction handle.
4395  * @root:      The root of the btree.
4396  * @path:      A path pointing to the target leaf and slot.
4397  * @key:       The key of the new item.
4398  * @data_size: The size of the data associated with the new key.
4399  */
btrfs_setup_item_for_insert(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,const struct btrfs_key * key,u32 data_size)4400 void btrfs_setup_item_for_insert(struct btrfs_trans_handle *trans,
4401 				 struct btrfs_root *root,
4402 				 struct btrfs_path *path,
4403 				 const struct btrfs_key *key,
4404 				 u32 data_size)
4405 {
4406 	struct btrfs_item_batch batch;
4407 
4408 	batch.keys = key;
4409 	batch.data_sizes = &data_size;
4410 	batch.total_data_size = data_size;
4411 	batch.nr = 1;
4412 
4413 	setup_items_for_insert(trans, root, path, &batch);
4414 }
4415 
4416 /*
4417  * Given a key and some data, insert items into the tree.
4418  * This does all the path init required, making room in the tree if needed.
4419  */
btrfs_insert_empty_items(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,const struct btrfs_item_batch * batch)4420 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4421 			    struct btrfs_root *root,
4422 			    struct btrfs_path *path,
4423 			    const struct btrfs_item_batch *batch)
4424 {
4425 	int ret = 0;
4426 	int slot;
4427 	u32 total_size;
4428 
4429 	total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item));
4430 	ret = btrfs_search_slot(trans, root, &batch->keys[0], path, total_size, 1);
4431 	if (ret == 0)
4432 		return -EEXIST;
4433 	if (ret < 0)
4434 		return ret;
4435 
4436 	slot = path->slots[0];
4437 	BUG_ON(slot < 0);
4438 
4439 	setup_items_for_insert(trans, root, path, batch);
4440 	return 0;
4441 }
4442 
4443 /*
4444  * Given a key and some data, insert an item into the tree.
4445  * This does all the path init required, making room in the tree if needed.
4446  */
btrfs_insert_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,const struct btrfs_key * cpu_key,void * data,u32 data_size)4447 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4448 		      const struct btrfs_key *cpu_key, void *data,
4449 		      u32 data_size)
4450 {
4451 	int ret = 0;
4452 	struct btrfs_path *path;
4453 	struct extent_buffer *leaf;
4454 	unsigned long ptr;
4455 
4456 	path = btrfs_alloc_path();
4457 	if (!path)
4458 		return -ENOMEM;
4459 	ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4460 	if (!ret) {
4461 		leaf = path->nodes[0];
4462 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4463 		write_extent_buffer(leaf, data, ptr, data_size);
4464 		btrfs_mark_buffer_dirty(trans, leaf);
4465 	}
4466 	btrfs_free_path(path);
4467 	return ret;
4468 }
4469 
4470 /*
4471  * This function duplicates an item, giving 'new_key' to the new item.
4472  * It guarantees both items live in the same tree leaf and the new item is
4473  * contiguous with the original item.
4474  *
4475  * This allows us to split a file extent in place, keeping a lock on the leaf
4476  * the entire time.
4477  */
btrfs_duplicate_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,const struct btrfs_key * new_key)4478 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4479 			 struct btrfs_root *root,
4480 			 struct btrfs_path *path,
4481 			 const struct btrfs_key *new_key)
4482 {
4483 	struct extent_buffer *leaf;
4484 	int ret;
4485 	u32 item_size;
4486 
4487 	leaf = path->nodes[0];
4488 	item_size = btrfs_item_size(leaf, path->slots[0]);
4489 	ret = setup_leaf_for_split(trans, root, path,
4490 				   item_size + sizeof(struct btrfs_item));
4491 	if (ret)
4492 		return ret;
4493 
4494 	path->slots[0]++;
4495 	btrfs_setup_item_for_insert(trans, root, path, new_key, item_size);
4496 	leaf = path->nodes[0];
4497 	memcpy_extent_buffer(leaf,
4498 			     btrfs_item_ptr_offset(leaf, path->slots[0]),
4499 			     btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4500 			     item_size);
4501 	return 0;
4502 }
4503 
4504 /*
4505  * delete the pointer from a given node.
4506  *
4507  * the tree should have been previously balanced so the deletion does not
4508  * empty a node.
4509  *
4510  * This is exported for use inside btrfs-progs, don't un-export it.
4511  */
btrfs_del_ptr(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int level,int slot)4512 int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4513 		  struct btrfs_path *path, int level, int slot)
4514 {
4515 	struct extent_buffer *parent = path->nodes[level];
4516 	u32 nritems;
4517 	int ret;
4518 
4519 	nritems = btrfs_header_nritems(parent);
4520 	if (slot != nritems - 1) {
4521 		if (level) {
4522 			ret = btrfs_tree_mod_log_insert_move(parent, slot,
4523 					slot + 1, nritems - slot - 1);
4524 			if (ret < 0) {
4525 				btrfs_abort_transaction(trans, ret);
4526 				return ret;
4527 			}
4528 		}
4529 		memmove_extent_buffer(parent,
4530 			      btrfs_node_key_ptr_offset(parent, slot),
4531 			      btrfs_node_key_ptr_offset(parent, slot + 1),
4532 			      sizeof(struct btrfs_key_ptr) *
4533 			      (nritems - slot - 1));
4534 	} else if (level) {
4535 		ret = btrfs_tree_mod_log_insert_key(parent, slot,
4536 						    BTRFS_MOD_LOG_KEY_REMOVE);
4537 		if (ret < 0) {
4538 			btrfs_abort_transaction(trans, ret);
4539 			return ret;
4540 		}
4541 	}
4542 
4543 	nritems--;
4544 	btrfs_set_header_nritems(parent, nritems);
4545 	if (nritems == 0 && parent == root->node) {
4546 		BUG_ON(btrfs_header_level(root->node) != 1);
4547 		/* just turn the root into a leaf and break */
4548 		btrfs_set_header_level(root->node, 0);
4549 	} else if (slot == 0) {
4550 		struct btrfs_disk_key disk_key;
4551 
4552 		btrfs_node_key(parent, &disk_key, 0);
4553 		fixup_low_keys(trans, path, &disk_key, level + 1);
4554 	}
4555 	btrfs_mark_buffer_dirty(trans, parent);
4556 	return 0;
4557 }
4558 
4559 /*
4560  * a helper function to delete the leaf pointed to by path->slots[1] and
4561  * path->nodes[1].
4562  *
4563  * This deletes the pointer in path->nodes[1] and frees the leaf
4564  * block extent.  zero is returned if it all worked out, < 0 otherwise.
4565  *
4566  * The path must have already been setup for deleting the leaf, including
4567  * all the proper balancing.  path->nodes[1] must be locked.
4568  */
btrfs_del_leaf(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct extent_buffer * leaf)4569 static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
4570 				   struct btrfs_root *root,
4571 				   struct btrfs_path *path,
4572 				   struct extent_buffer *leaf)
4573 {
4574 	int ret;
4575 
4576 	WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4577 	ret = btrfs_del_ptr(trans, root, path, 1, path->slots[1]);
4578 	if (ret < 0)
4579 		return ret;
4580 
4581 	/*
4582 	 * btrfs_free_extent is expensive, we want to make sure we
4583 	 * aren't holding any locks when we call it
4584 	 */
4585 	btrfs_unlock_up_safe(path, 0);
4586 
4587 	root_sub_used(root, leaf->len);
4588 
4589 	atomic_inc(&leaf->refs);
4590 	btrfs_free_tree_block(trans, btrfs_root_id(root), leaf, 0, 1);
4591 	free_extent_buffer_stale(leaf);
4592 	return 0;
4593 }
4594 /*
4595  * delete the item at the leaf level in path.  If that empties
4596  * the leaf, remove it from the tree
4597  */
btrfs_del_items(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int slot,int nr)4598 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4599 		    struct btrfs_path *path, int slot, int nr)
4600 {
4601 	struct btrfs_fs_info *fs_info = root->fs_info;
4602 	struct extent_buffer *leaf;
4603 	int ret = 0;
4604 	int wret;
4605 	u32 nritems;
4606 
4607 	leaf = path->nodes[0];
4608 	nritems = btrfs_header_nritems(leaf);
4609 
4610 	if (slot + nr != nritems) {
4611 		const u32 last_off = btrfs_item_offset(leaf, slot + nr - 1);
4612 		const int data_end = leaf_data_end(leaf);
4613 		struct btrfs_map_token token;
4614 		u32 dsize = 0;
4615 		int i;
4616 
4617 		for (i = 0; i < nr; i++)
4618 			dsize += btrfs_item_size(leaf, slot + i);
4619 
4620 		memmove_leaf_data(leaf, data_end + dsize, data_end,
4621 				  last_off - data_end);
4622 
4623 		btrfs_init_map_token(&token, leaf);
4624 		for (i = slot + nr; i < nritems; i++) {
4625 			u32 ioff;
4626 
4627 			ioff = btrfs_token_item_offset(&token, i);
4628 			btrfs_set_token_item_offset(&token, i, ioff + dsize);
4629 		}
4630 
4631 		memmove_leaf_items(leaf, slot, slot + nr, nritems - slot - nr);
4632 	}
4633 	btrfs_set_header_nritems(leaf, nritems - nr);
4634 	nritems -= nr;
4635 
4636 	/* delete the leaf if we've emptied it */
4637 	if (nritems == 0) {
4638 		if (leaf == root->node) {
4639 			btrfs_set_header_level(leaf, 0);
4640 		} else {
4641 			btrfs_clear_buffer_dirty(trans, leaf);
4642 			ret = btrfs_del_leaf(trans, root, path, leaf);
4643 			if (ret < 0)
4644 				return ret;
4645 		}
4646 	} else {
4647 		int used = leaf_space_used(leaf, 0, nritems);
4648 		if (slot == 0) {
4649 			struct btrfs_disk_key disk_key;
4650 
4651 			btrfs_item_key(leaf, &disk_key, 0);
4652 			fixup_low_keys(trans, path, &disk_key, 1);
4653 		}
4654 
4655 		/*
4656 		 * Try to delete the leaf if it is mostly empty. We do this by
4657 		 * trying to move all its items into its left and right neighbours.
4658 		 * If we can't move all the items, then we don't delete it - it's
4659 		 * not ideal, but future insertions might fill the leaf with more
4660 		 * items, or items from other leaves might be moved later into our
4661 		 * leaf due to deletions on those leaves.
4662 		 */
4663 		if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
4664 			u32 min_push_space;
4665 
4666 			/* push_leaf_left fixes the path.
4667 			 * make sure the path still points to our leaf
4668 			 * for possible call to btrfs_del_ptr below
4669 			 */
4670 			slot = path->slots[1];
4671 			atomic_inc(&leaf->refs);
4672 			/*
4673 			 * We want to be able to at least push one item to the
4674 			 * left neighbour leaf, and that's the first item.
4675 			 */
4676 			min_push_space = sizeof(struct btrfs_item) +
4677 				btrfs_item_size(leaf, 0);
4678 			wret = push_leaf_left(trans, root, path, 0,
4679 					      min_push_space, 1, (u32)-1);
4680 			if (wret < 0 && wret != -ENOSPC)
4681 				ret = wret;
4682 
4683 			if (path->nodes[0] == leaf &&
4684 			    btrfs_header_nritems(leaf)) {
4685 				/*
4686 				 * If we were not able to push all items from our
4687 				 * leaf to its left neighbour, then attempt to
4688 				 * either push all the remaining items to the
4689 				 * right neighbour or none. There's no advantage
4690 				 * in pushing only some items, instead of all, as
4691 				 * it's pointless to end up with a leaf having
4692 				 * too few items while the neighbours can be full
4693 				 * or nearly full.
4694 				 */
4695 				nritems = btrfs_header_nritems(leaf);
4696 				min_push_space = leaf_space_used(leaf, 0, nritems);
4697 				wret = push_leaf_right(trans, root, path, 0,
4698 						       min_push_space, 1, 0);
4699 				if (wret < 0 && wret != -ENOSPC)
4700 					ret = wret;
4701 			}
4702 
4703 			if (btrfs_header_nritems(leaf) == 0) {
4704 				path->slots[1] = slot;
4705 				ret = btrfs_del_leaf(trans, root, path, leaf);
4706 				if (ret < 0)
4707 					return ret;
4708 				free_extent_buffer(leaf);
4709 				ret = 0;
4710 			} else {
4711 				/* if we're still in the path, make sure
4712 				 * we're dirty.  Otherwise, one of the
4713 				 * push_leaf functions must have already
4714 				 * dirtied this buffer
4715 				 */
4716 				if (path->nodes[0] == leaf)
4717 					btrfs_mark_buffer_dirty(trans, leaf);
4718 				free_extent_buffer(leaf);
4719 			}
4720 		} else {
4721 			btrfs_mark_buffer_dirty(trans, leaf);
4722 		}
4723 	}
4724 	return ret;
4725 }
4726 
4727 /*
4728  * A helper function to walk down the tree starting at min_key, and looking
4729  * for nodes or leaves that are have a minimum transaction id.
4730  * This is used by the btree defrag code, and tree logging
4731  *
4732  * This does not cow, but it does stuff the starting key it finds back
4733  * into min_key, so you can call btrfs_search_slot with cow=1 on the
4734  * key and get a writable path.
4735  *
4736  * This honors path->lowest_level to prevent descent past a given level
4737  * of the tree.
4738  *
4739  * min_trans indicates the oldest transaction that you are interested
4740  * in walking through.  Any nodes or leaves older than min_trans are
4741  * skipped over (without reading them).
4742  *
4743  * returns zero if something useful was found, < 0 on error and 1 if there
4744  * was nothing in the tree that matched the search criteria.
4745  */
btrfs_search_forward(struct btrfs_root * root,struct btrfs_key * min_key,struct btrfs_path * path,u64 min_trans)4746 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
4747 			 struct btrfs_path *path,
4748 			 u64 min_trans)
4749 {
4750 	struct extent_buffer *cur;
4751 	struct btrfs_key found_key;
4752 	int slot;
4753 	int sret;
4754 	u32 nritems;
4755 	int level;
4756 	int ret = 1;
4757 	int keep_locks = path->keep_locks;
4758 
4759 	ASSERT(!path->nowait);
4760 	path->keep_locks = 1;
4761 again:
4762 	cur = btrfs_read_lock_root_node(root);
4763 	level = btrfs_header_level(cur);
4764 	WARN_ON(path->nodes[level]);
4765 	path->nodes[level] = cur;
4766 	path->locks[level] = BTRFS_READ_LOCK;
4767 
4768 	if (btrfs_header_generation(cur) < min_trans) {
4769 		ret = 1;
4770 		goto out;
4771 	}
4772 	while (1) {
4773 		nritems = btrfs_header_nritems(cur);
4774 		level = btrfs_header_level(cur);
4775 		sret = btrfs_bin_search(cur, 0, min_key, &slot);
4776 		if (sret < 0) {
4777 			ret = sret;
4778 			goto out;
4779 		}
4780 
4781 		/* at the lowest level, we're done, setup the path and exit */
4782 		if (level == path->lowest_level) {
4783 			if (slot >= nritems)
4784 				goto find_next_key;
4785 			ret = 0;
4786 			path->slots[level] = slot;
4787 			btrfs_item_key_to_cpu(cur, &found_key, slot);
4788 			goto out;
4789 		}
4790 		if (sret && slot > 0)
4791 			slot--;
4792 		/*
4793 		 * check this node pointer against the min_trans parameters.
4794 		 * If it is too old, skip to the next one.
4795 		 */
4796 		while (slot < nritems) {
4797 			u64 gen;
4798 
4799 			gen = btrfs_node_ptr_generation(cur, slot);
4800 			if (gen < min_trans) {
4801 				slot++;
4802 				continue;
4803 			}
4804 			break;
4805 		}
4806 find_next_key:
4807 		/*
4808 		 * we didn't find a candidate key in this node, walk forward
4809 		 * and find another one
4810 		 */
4811 		if (slot >= nritems) {
4812 			path->slots[level] = slot;
4813 			sret = btrfs_find_next_key(root, path, min_key, level,
4814 						  min_trans);
4815 			if (sret == 0) {
4816 				btrfs_release_path(path);
4817 				goto again;
4818 			} else {
4819 				goto out;
4820 			}
4821 		}
4822 		/* save our key for returning back */
4823 		btrfs_node_key_to_cpu(cur, &found_key, slot);
4824 		path->slots[level] = slot;
4825 		if (level == path->lowest_level) {
4826 			ret = 0;
4827 			goto out;
4828 		}
4829 		cur = btrfs_read_node_slot(cur, slot);
4830 		if (IS_ERR(cur)) {
4831 			ret = PTR_ERR(cur);
4832 			goto out;
4833 		}
4834 
4835 		btrfs_tree_read_lock(cur);
4836 
4837 		path->locks[level - 1] = BTRFS_READ_LOCK;
4838 		path->nodes[level - 1] = cur;
4839 		unlock_up(path, level, 1, 0, NULL);
4840 	}
4841 out:
4842 	path->keep_locks = keep_locks;
4843 	if (ret == 0) {
4844 		btrfs_unlock_up_safe(path, path->lowest_level + 1);
4845 		memcpy(min_key, &found_key, sizeof(found_key));
4846 	}
4847 	return ret;
4848 }
4849 
4850 /*
4851  * this is similar to btrfs_next_leaf, but does not try to preserve
4852  * and fixup the path.  It looks for and returns the next key in the
4853  * tree based on the current path and the min_trans parameters.
4854  *
4855  * 0 is returned if another key is found, < 0 if there are any errors
4856  * and 1 is returned if there are no higher keys in the tree
4857  *
4858  * path->keep_locks should be set to 1 on the search made before
4859  * calling this function.
4860  */
btrfs_find_next_key(struct btrfs_root * root,struct btrfs_path * path,struct btrfs_key * key,int level,u64 min_trans)4861 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4862 			struct btrfs_key *key, int level, u64 min_trans)
4863 {
4864 	int slot;
4865 	struct extent_buffer *c;
4866 
4867 	WARN_ON(!path->keep_locks && !path->skip_locking);
4868 	while (level < BTRFS_MAX_LEVEL) {
4869 		if (!path->nodes[level])
4870 			return 1;
4871 
4872 		slot = path->slots[level] + 1;
4873 		c = path->nodes[level];
4874 next:
4875 		if (slot >= btrfs_header_nritems(c)) {
4876 			int ret;
4877 			int orig_lowest;
4878 			struct btrfs_key cur_key;
4879 			if (level + 1 >= BTRFS_MAX_LEVEL ||
4880 			    !path->nodes[level + 1])
4881 				return 1;
4882 
4883 			if (path->locks[level + 1] || path->skip_locking) {
4884 				level++;
4885 				continue;
4886 			}
4887 
4888 			slot = btrfs_header_nritems(c) - 1;
4889 			if (level == 0)
4890 				btrfs_item_key_to_cpu(c, &cur_key, slot);
4891 			else
4892 				btrfs_node_key_to_cpu(c, &cur_key, slot);
4893 
4894 			orig_lowest = path->lowest_level;
4895 			btrfs_release_path(path);
4896 			path->lowest_level = level;
4897 			ret = btrfs_search_slot(NULL, root, &cur_key, path,
4898 						0, 0);
4899 			path->lowest_level = orig_lowest;
4900 			if (ret < 0)
4901 				return ret;
4902 
4903 			c = path->nodes[level];
4904 			slot = path->slots[level];
4905 			if (ret == 0)
4906 				slot++;
4907 			goto next;
4908 		}
4909 
4910 		if (level == 0)
4911 			btrfs_item_key_to_cpu(c, key, slot);
4912 		else {
4913 			u64 gen = btrfs_node_ptr_generation(c, slot);
4914 
4915 			if (gen < min_trans) {
4916 				slot++;
4917 				goto next;
4918 			}
4919 			btrfs_node_key_to_cpu(c, key, slot);
4920 		}
4921 		return 0;
4922 	}
4923 	return 1;
4924 }
4925 
btrfs_next_old_leaf(struct btrfs_root * root,struct btrfs_path * path,u64 time_seq)4926 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
4927 			u64 time_seq)
4928 {
4929 	int slot;
4930 	int level;
4931 	struct extent_buffer *c;
4932 	struct extent_buffer *next;
4933 	struct btrfs_fs_info *fs_info = root->fs_info;
4934 	struct btrfs_key key;
4935 	bool need_commit_sem = false;
4936 	u32 nritems;
4937 	int ret;
4938 	int i;
4939 
4940 	/*
4941 	 * The nowait semantics are used only for write paths, where we don't
4942 	 * use the tree mod log and sequence numbers.
4943 	 */
4944 	if (time_seq)
4945 		ASSERT(!path->nowait);
4946 
4947 	nritems = btrfs_header_nritems(path->nodes[0]);
4948 	if (nritems == 0)
4949 		return 1;
4950 
4951 	btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4952 again:
4953 	level = 1;
4954 	next = NULL;
4955 	btrfs_release_path(path);
4956 
4957 	path->keep_locks = 1;
4958 
4959 	if (time_seq) {
4960 		ret = btrfs_search_old_slot(root, &key, path, time_seq);
4961 	} else {
4962 		if (path->need_commit_sem) {
4963 			path->need_commit_sem = 0;
4964 			need_commit_sem = true;
4965 			if (path->nowait) {
4966 				if (!down_read_trylock(&fs_info->commit_root_sem)) {
4967 					ret = -EAGAIN;
4968 					goto done;
4969 				}
4970 			} else {
4971 				down_read(&fs_info->commit_root_sem);
4972 			}
4973 		}
4974 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4975 	}
4976 	path->keep_locks = 0;
4977 
4978 	if (ret < 0)
4979 		goto done;
4980 
4981 	nritems = btrfs_header_nritems(path->nodes[0]);
4982 	/*
4983 	 * by releasing the path above we dropped all our locks.  A balance
4984 	 * could have added more items next to the key that used to be
4985 	 * at the very end of the block.  So, check again here and
4986 	 * advance the path if there are now more items available.
4987 	 */
4988 	if (nritems > 0 && path->slots[0] < nritems - 1) {
4989 		if (ret == 0)
4990 			path->slots[0]++;
4991 		ret = 0;
4992 		goto done;
4993 	}
4994 	/*
4995 	 * So the above check misses one case:
4996 	 * - after releasing the path above, someone has removed the item that
4997 	 *   used to be at the very end of the block, and balance between leafs
4998 	 *   gets another one with bigger key.offset to replace it.
4999 	 *
5000 	 * This one should be returned as well, or we can get leaf corruption
5001 	 * later(esp. in __btrfs_drop_extents()).
5002 	 *
5003 	 * And a bit more explanation about this check,
5004 	 * with ret > 0, the key isn't found, the path points to the slot
5005 	 * where it should be inserted, so the path->slots[0] item must be the
5006 	 * bigger one.
5007 	 */
5008 	if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5009 		ret = 0;
5010 		goto done;
5011 	}
5012 
5013 	while (level < BTRFS_MAX_LEVEL) {
5014 		if (!path->nodes[level]) {
5015 			ret = 1;
5016 			goto done;
5017 		}
5018 
5019 		slot = path->slots[level] + 1;
5020 		c = path->nodes[level];
5021 		if (slot >= btrfs_header_nritems(c)) {
5022 			level++;
5023 			if (level == BTRFS_MAX_LEVEL) {
5024 				ret = 1;
5025 				goto done;
5026 			}
5027 			continue;
5028 		}
5029 
5030 
5031 		/*
5032 		 * Our current level is where we're going to start from, and to
5033 		 * make sure lockdep doesn't complain we need to drop our locks
5034 		 * and nodes from 0 to our current level.
5035 		 */
5036 		for (i = 0; i < level; i++) {
5037 			if (path->locks[level]) {
5038 				btrfs_tree_read_unlock(path->nodes[i]);
5039 				path->locks[i] = 0;
5040 			}
5041 			free_extent_buffer(path->nodes[i]);
5042 			path->nodes[i] = NULL;
5043 		}
5044 
5045 		next = c;
5046 		ret = read_block_for_search(root, path, &next, level,
5047 					    slot, &key);
5048 		if (ret == -EAGAIN && !path->nowait)
5049 			goto again;
5050 
5051 		if (ret < 0) {
5052 			btrfs_release_path(path);
5053 			goto done;
5054 		}
5055 
5056 		if (!path->skip_locking) {
5057 			ret = btrfs_try_tree_read_lock(next);
5058 			if (!ret && path->nowait) {
5059 				ret = -EAGAIN;
5060 				goto done;
5061 			}
5062 			if (!ret && time_seq) {
5063 				/*
5064 				 * If we don't get the lock, we may be racing
5065 				 * with push_leaf_left, holding that lock while
5066 				 * itself waiting for the leaf we've currently
5067 				 * locked. To solve this situation, we give up
5068 				 * on our lock and cycle.
5069 				 */
5070 				free_extent_buffer(next);
5071 				btrfs_release_path(path);
5072 				cond_resched();
5073 				goto again;
5074 			}
5075 			if (!ret)
5076 				btrfs_tree_read_lock(next);
5077 		}
5078 		break;
5079 	}
5080 	path->slots[level] = slot;
5081 	while (1) {
5082 		level--;
5083 		path->nodes[level] = next;
5084 		path->slots[level] = 0;
5085 		if (!path->skip_locking)
5086 			path->locks[level] = BTRFS_READ_LOCK;
5087 		if (!level)
5088 			break;
5089 
5090 		ret = read_block_for_search(root, path, &next, level,
5091 					    0, &key);
5092 		if (ret == -EAGAIN && !path->nowait)
5093 			goto again;
5094 
5095 		if (ret < 0) {
5096 			btrfs_release_path(path);
5097 			goto done;
5098 		}
5099 
5100 		if (!path->skip_locking) {
5101 			if (path->nowait) {
5102 				if (!btrfs_try_tree_read_lock(next)) {
5103 					ret = -EAGAIN;
5104 					goto done;
5105 				}
5106 			} else {
5107 				btrfs_tree_read_lock(next);
5108 			}
5109 		}
5110 	}
5111 	ret = 0;
5112 done:
5113 	unlock_up(path, 0, 1, 0, NULL);
5114 	if (need_commit_sem) {
5115 		int ret2;
5116 
5117 		path->need_commit_sem = 1;
5118 		ret2 = finish_need_commit_sem_search(path);
5119 		up_read(&fs_info->commit_root_sem);
5120 		if (ret2)
5121 			ret = ret2;
5122 	}
5123 
5124 	return ret;
5125 }
5126 
btrfs_next_old_item(struct btrfs_root * root,struct btrfs_path * path,u64 time_seq)5127 int btrfs_next_old_item(struct btrfs_root *root, struct btrfs_path *path, u64 time_seq)
5128 {
5129 	path->slots[0]++;
5130 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
5131 		return btrfs_next_old_leaf(root, path, time_seq);
5132 	return 0;
5133 }
5134 
5135 /*
5136  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5137  * searching until it gets past min_objectid or finds an item of 'type'
5138  *
5139  * returns 0 if something is found, 1 if nothing was found and < 0 on error
5140  */
btrfs_previous_item(struct btrfs_root * root,struct btrfs_path * path,u64 min_objectid,int type)5141 int btrfs_previous_item(struct btrfs_root *root,
5142 			struct btrfs_path *path, u64 min_objectid,
5143 			int type)
5144 {
5145 	struct btrfs_key found_key;
5146 	struct extent_buffer *leaf;
5147 	u32 nritems;
5148 	int ret;
5149 
5150 	while (1) {
5151 		if (path->slots[0] == 0) {
5152 			ret = btrfs_prev_leaf(root, path);
5153 			if (ret != 0)
5154 				return ret;
5155 		} else {
5156 			path->slots[0]--;
5157 		}
5158 		leaf = path->nodes[0];
5159 		nritems = btrfs_header_nritems(leaf);
5160 		if (nritems == 0)
5161 			return 1;
5162 		if (path->slots[0] == nritems)
5163 			path->slots[0]--;
5164 
5165 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5166 		if (found_key.objectid < min_objectid)
5167 			break;
5168 		if (found_key.type == type)
5169 			return 0;
5170 		if (found_key.objectid == min_objectid &&
5171 		    found_key.type < type)
5172 			break;
5173 	}
5174 	return 1;
5175 }
5176 
5177 /*
5178  * search in extent tree to find a previous Metadata/Data extent item with
5179  * min objecitd.
5180  *
5181  * returns 0 if something is found, 1 if nothing was found and < 0 on error
5182  */
btrfs_previous_extent_item(struct btrfs_root * root,struct btrfs_path * path,u64 min_objectid)5183 int btrfs_previous_extent_item(struct btrfs_root *root,
5184 			struct btrfs_path *path, u64 min_objectid)
5185 {
5186 	struct btrfs_key found_key;
5187 	struct extent_buffer *leaf;
5188 	u32 nritems;
5189 	int ret;
5190 
5191 	while (1) {
5192 		if (path->slots[0] == 0) {
5193 			ret = btrfs_prev_leaf(root, path);
5194 			if (ret != 0)
5195 				return ret;
5196 		} else {
5197 			path->slots[0]--;
5198 		}
5199 		leaf = path->nodes[0];
5200 		nritems = btrfs_header_nritems(leaf);
5201 		if (nritems == 0)
5202 			return 1;
5203 		if (path->slots[0] == nritems)
5204 			path->slots[0]--;
5205 
5206 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5207 		if (found_key.objectid < min_objectid)
5208 			break;
5209 		if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5210 		    found_key.type == BTRFS_METADATA_ITEM_KEY)
5211 			return 0;
5212 		if (found_key.objectid == min_objectid &&
5213 		    found_key.type < BTRFS_EXTENT_ITEM_KEY)
5214 			break;
5215 	}
5216 	return 1;
5217 }
5218 
btrfs_ctree_init(void)5219 int __init btrfs_ctree_init(void)
5220 {
5221 	btrfs_path_cachep = kmem_cache_create("btrfs_path",
5222 			sizeof(struct btrfs_path), 0,
5223 			SLAB_MEM_SPREAD, NULL);
5224 	if (!btrfs_path_cachep)
5225 		return -ENOMEM;
5226 	return 0;
5227 }
5228 
btrfs_ctree_exit(void)5229 void __cold btrfs_ctree_exit(void)
5230 {
5231 	kmem_cache_destroy(btrfs_path_cachep);
5232 }
5233