• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 "ctree.h"
11 #include "disk-io.h"
12 #include "transaction.h"
13 #include "print-tree.h"
14 #include "locking.h"
15 #include "volumes.h"
16 #include "qgroup.h"
17 
18 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
19 		      *root, struct btrfs_path *path, int level);
20 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
21 		      const struct btrfs_key *ins_key, struct btrfs_path *path,
22 		      int data_size, int extend);
23 static int push_node_left(struct btrfs_trans_handle *trans,
24 			  struct extent_buffer *dst,
25 			  struct extent_buffer *src, int empty);
26 static int balance_node_right(struct btrfs_trans_handle *trans,
27 			      struct extent_buffer *dst_buf,
28 			      struct extent_buffer *src_buf);
29 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
30 		    int level, int slot);
31 
32 static const struct btrfs_csums {
33 	u16		size;
34 	const char	name[10];
35 	const char	driver[12];
36 } btrfs_csums[] = {
37 	[BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" },
38 	[BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" },
39 	[BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" },
40 	[BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b",
41 				     .driver = "blake2b-256" },
42 };
43 
btrfs_super_csum_size(const struct btrfs_super_block * s)44 int btrfs_super_csum_size(const struct btrfs_super_block *s)
45 {
46 	u16 t = btrfs_super_csum_type(s);
47 	/*
48 	 * csum type is validated at mount time
49 	 */
50 	return btrfs_csums[t].size;
51 }
52 
btrfs_super_csum_name(u16 csum_type)53 const char *btrfs_super_csum_name(u16 csum_type)
54 {
55 	/* csum type is validated at mount time */
56 	return btrfs_csums[csum_type].name;
57 }
58 
59 /*
60  * Return driver name if defined, otherwise the name that's also a valid driver
61  * name
62  */
btrfs_super_csum_driver(u16 csum_type)63 const char *btrfs_super_csum_driver(u16 csum_type)
64 {
65 	/* csum type is validated at mount time */
66 	return btrfs_csums[csum_type].driver[0] ?
67 		btrfs_csums[csum_type].driver :
68 		btrfs_csums[csum_type].name;
69 }
70 
btrfs_get_num_csums(void)71 size_t __attribute_const__ btrfs_get_num_csums(void)
72 {
73 	return ARRAY_SIZE(btrfs_csums);
74 }
75 
btrfs_alloc_path(void)76 struct btrfs_path *btrfs_alloc_path(void)
77 {
78 	return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
79 }
80 
81 /* this also releases the path */
btrfs_free_path(struct btrfs_path * p)82 void btrfs_free_path(struct btrfs_path *p)
83 {
84 	if (!p)
85 		return;
86 	btrfs_release_path(p);
87 	kmem_cache_free(btrfs_path_cachep, p);
88 }
89 
90 /*
91  * path release drops references on the extent buffers in the path
92  * and it drops any locks held by this path
93  *
94  * It is safe to call this on paths that no locks or extent buffers held.
95  */
btrfs_release_path(struct btrfs_path * p)96 noinline void btrfs_release_path(struct btrfs_path *p)
97 {
98 	int i;
99 
100 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
101 		p->slots[i] = 0;
102 		if (!p->nodes[i])
103 			continue;
104 		if (p->locks[i]) {
105 			btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
106 			p->locks[i] = 0;
107 		}
108 		free_extent_buffer(p->nodes[i]);
109 		p->nodes[i] = NULL;
110 	}
111 }
112 
113 /*
114  * safely gets a reference on the root node of a tree.  A lock
115  * is not taken, so a concurrent writer may put a different node
116  * at the root of the tree.  See btrfs_lock_root_node for the
117  * looping required.
118  *
119  * The extent buffer returned by this has a reference taken, so
120  * it won't disappear.  It may stop being the root of the tree
121  * at any time because there are no locks held.
122  */
btrfs_root_node(struct btrfs_root * root)123 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
124 {
125 	struct extent_buffer *eb;
126 
127 	while (1) {
128 		rcu_read_lock();
129 		eb = rcu_dereference(root->node);
130 
131 		/*
132 		 * RCU really hurts here, we could free up the root node because
133 		 * it was COWed but we may not get the new root node yet so do
134 		 * the inc_not_zero dance and if it doesn't work then
135 		 * synchronize_rcu and try again.
136 		 */
137 		if (atomic_inc_not_zero(&eb->refs)) {
138 			rcu_read_unlock();
139 			break;
140 		}
141 		rcu_read_unlock();
142 		synchronize_rcu();
143 	}
144 	return eb;
145 }
146 
147 /*
148  * Cowonly root (not-shareable trees, everything not subvolume or reloc roots),
149  * just get put onto a simple dirty list.  Transaction walks this list to make
150  * sure they get properly updated on disk.
151  */
add_root_to_dirty_list(struct btrfs_root * root)152 static void add_root_to_dirty_list(struct btrfs_root *root)
153 {
154 	struct btrfs_fs_info *fs_info = root->fs_info;
155 
156 	if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
157 	    !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
158 		return;
159 
160 	spin_lock(&fs_info->trans_lock);
161 	if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
162 		/* Want the extent tree to be the last on the list */
163 		if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
164 			list_move_tail(&root->dirty_list,
165 				       &fs_info->dirty_cowonly_roots);
166 		else
167 			list_move(&root->dirty_list,
168 				  &fs_info->dirty_cowonly_roots);
169 	}
170 	spin_unlock(&fs_info->trans_lock);
171 }
172 
173 /*
174  * used by snapshot creation to make a copy of a root for a tree with
175  * a given objectid.  The buffer with the new root node is returned in
176  * cow_ret, and this func returns zero on success or a negative error code.
177  */
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)178 int btrfs_copy_root(struct btrfs_trans_handle *trans,
179 		      struct btrfs_root *root,
180 		      struct extent_buffer *buf,
181 		      struct extent_buffer **cow_ret, u64 new_root_objectid)
182 {
183 	struct btrfs_fs_info *fs_info = root->fs_info;
184 	struct extent_buffer *cow;
185 	int ret = 0;
186 	int level;
187 	struct btrfs_disk_key disk_key;
188 
189 	WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
190 		trans->transid != fs_info->running_transaction->transid);
191 	WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
192 		trans->transid != root->last_trans);
193 
194 	level = btrfs_header_level(buf);
195 	if (level == 0)
196 		btrfs_item_key(buf, &disk_key, 0);
197 	else
198 		btrfs_node_key(buf, &disk_key, 0);
199 
200 	cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
201 				     &disk_key, level, buf->start, 0,
202 				     BTRFS_NESTING_NEW_ROOT);
203 	if (IS_ERR(cow))
204 		return PTR_ERR(cow);
205 
206 	copy_extent_buffer_full(cow, buf);
207 	btrfs_set_header_bytenr(cow, cow->start);
208 	btrfs_set_header_generation(cow, trans->transid);
209 	btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
210 	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
211 				     BTRFS_HEADER_FLAG_RELOC);
212 	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
213 		btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
214 	else
215 		btrfs_set_header_owner(cow, new_root_objectid);
216 
217 	write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
218 
219 	WARN_ON(btrfs_header_generation(buf) > trans->transid);
220 	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
221 		ret = btrfs_inc_ref(trans, root, cow, 1);
222 	else
223 		ret = btrfs_inc_ref(trans, root, cow, 0);
224 	if (ret) {
225 		btrfs_tree_unlock(cow);
226 		free_extent_buffer(cow);
227 		btrfs_abort_transaction(trans, ret);
228 		return ret;
229 	}
230 
231 	btrfs_mark_buffer_dirty(cow);
232 	*cow_ret = cow;
233 	return 0;
234 }
235 
236 enum mod_log_op {
237 	MOD_LOG_KEY_REPLACE,
238 	MOD_LOG_KEY_ADD,
239 	MOD_LOG_KEY_REMOVE,
240 	MOD_LOG_KEY_REMOVE_WHILE_FREEING,
241 	MOD_LOG_KEY_REMOVE_WHILE_MOVING,
242 	MOD_LOG_MOVE_KEYS,
243 	MOD_LOG_ROOT_REPLACE,
244 };
245 
246 struct tree_mod_root {
247 	u64 logical;
248 	u8 level;
249 };
250 
251 struct tree_mod_elem {
252 	struct rb_node node;
253 	u64 logical;
254 	u64 seq;
255 	enum mod_log_op op;
256 
257 	/* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
258 	int slot;
259 
260 	/* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
261 	u64 generation;
262 
263 	/* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
264 	struct btrfs_disk_key key;
265 	u64 blockptr;
266 
267 	/* this is used for op == MOD_LOG_MOVE_KEYS */
268 	struct {
269 		int dst_slot;
270 		int nr_items;
271 	} move;
272 
273 	/* this is used for op == MOD_LOG_ROOT_REPLACE */
274 	struct tree_mod_root old_root;
275 };
276 
277 /*
278  * Pull a new tree mod seq number for our operation.
279  */
btrfs_inc_tree_mod_seq(struct btrfs_fs_info * fs_info)280 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
281 {
282 	return atomic64_inc_return(&fs_info->tree_mod_seq);
283 }
284 
285 /*
286  * This adds a new blocker to the tree mod log's blocker list if the @elem
287  * passed does not already have a sequence number set. So when a caller expects
288  * to record tree modifications, it should ensure to set elem->seq to zero
289  * before calling btrfs_get_tree_mod_seq.
290  * Returns a fresh, unused tree log modification sequence number, even if no new
291  * blocker was added.
292  */
btrfs_get_tree_mod_seq(struct btrfs_fs_info * fs_info,struct seq_list * elem)293 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
294 			   struct seq_list *elem)
295 {
296 	write_lock(&fs_info->tree_mod_log_lock);
297 	if (!elem->seq) {
298 		elem->seq = btrfs_inc_tree_mod_seq(fs_info);
299 		list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
300 	}
301 	write_unlock(&fs_info->tree_mod_log_lock);
302 
303 	return elem->seq;
304 }
305 
btrfs_put_tree_mod_seq(struct btrfs_fs_info * fs_info,struct seq_list * elem)306 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
307 			    struct seq_list *elem)
308 {
309 	struct rb_root *tm_root;
310 	struct rb_node *node;
311 	struct rb_node *next;
312 	struct tree_mod_elem *tm;
313 	u64 min_seq = (u64)-1;
314 	u64 seq_putting = elem->seq;
315 
316 	if (!seq_putting)
317 		return;
318 
319 	write_lock(&fs_info->tree_mod_log_lock);
320 	list_del(&elem->list);
321 	elem->seq = 0;
322 
323 	if (!list_empty(&fs_info->tree_mod_seq_list)) {
324 		struct seq_list *first;
325 
326 		first = list_first_entry(&fs_info->tree_mod_seq_list,
327 					 struct seq_list, list);
328 		if (seq_putting > first->seq) {
329 			/*
330 			 * Blocker with lower sequence number exists, we
331 			 * cannot remove anything from the log.
332 			 */
333 			write_unlock(&fs_info->tree_mod_log_lock);
334 			return;
335 		}
336 		min_seq = first->seq;
337 	}
338 
339 	/*
340 	 * anything that's lower than the lowest existing (read: blocked)
341 	 * sequence number can be removed from the tree.
342 	 */
343 	tm_root = &fs_info->tree_mod_log;
344 	for (node = rb_first(tm_root); node; node = next) {
345 		next = rb_next(node);
346 		tm = rb_entry(node, struct tree_mod_elem, node);
347 		if (tm->seq >= min_seq)
348 			continue;
349 		rb_erase(node, tm_root);
350 		kfree(tm);
351 	}
352 	write_unlock(&fs_info->tree_mod_log_lock);
353 }
354 
355 /*
356  * key order of the log:
357  *       node/leaf start address -> sequence
358  *
359  * The 'start address' is the logical address of the *new* root node
360  * for root replace operations, or the logical address of the affected
361  * block for all other operations.
362  */
363 static noinline int
__tree_mod_log_insert(struct btrfs_fs_info * fs_info,struct tree_mod_elem * tm)364 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
365 {
366 	struct rb_root *tm_root;
367 	struct rb_node **new;
368 	struct rb_node *parent = NULL;
369 	struct tree_mod_elem *cur;
370 
371 	lockdep_assert_held_write(&fs_info->tree_mod_log_lock);
372 
373 	tm->seq = btrfs_inc_tree_mod_seq(fs_info);
374 
375 	tm_root = &fs_info->tree_mod_log;
376 	new = &tm_root->rb_node;
377 	while (*new) {
378 		cur = rb_entry(*new, struct tree_mod_elem, node);
379 		parent = *new;
380 		if (cur->logical < tm->logical)
381 			new = &((*new)->rb_left);
382 		else if (cur->logical > tm->logical)
383 			new = &((*new)->rb_right);
384 		else if (cur->seq < tm->seq)
385 			new = &((*new)->rb_left);
386 		else if (cur->seq > tm->seq)
387 			new = &((*new)->rb_right);
388 		else
389 			return -EEXIST;
390 	}
391 
392 	rb_link_node(&tm->node, parent, new);
393 	rb_insert_color(&tm->node, tm_root);
394 	return 0;
395 }
396 
397 /*
398  * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
399  * returns zero with the tree_mod_log_lock acquired. The caller must hold
400  * this until all tree mod log insertions are recorded in the rb tree and then
401  * write unlock fs_info::tree_mod_log_lock.
402  */
tree_mod_dont_log(struct btrfs_fs_info * fs_info,struct extent_buffer * eb)403 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
404 				    struct extent_buffer *eb) {
405 	smp_mb();
406 	if (list_empty(&(fs_info)->tree_mod_seq_list))
407 		return 1;
408 	if (eb && btrfs_header_level(eb) == 0)
409 		return 1;
410 
411 	write_lock(&fs_info->tree_mod_log_lock);
412 	if (list_empty(&(fs_info)->tree_mod_seq_list)) {
413 		write_unlock(&fs_info->tree_mod_log_lock);
414 		return 1;
415 	}
416 
417 	return 0;
418 }
419 
420 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
tree_mod_need_log(const struct btrfs_fs_info * fs_info,struct extent_buffer * eb)421 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
422 				    struct extent_buffer *eb)
423 {
424 	smp_mb();
425 	if (list_empty(&(fs_info)->tree_mod_seq_list))
426 		return 0;
427 	if (eb && btrfs_header_level(eb) == 0)
428 		return 0;
429 
430 	return 1;
431 }
432 
433 static struct tree_mod_elem *
alloc_tree_mod_elem(struct extent_buffer * eb,int slot,enum mod_log_op op,gfp_t flags)434 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
435 		    enum mod_log_op op, gfp_t flags)
436 {
437 	struct tree_mod_elem *tm;
438 
439 	tm = kzalloc(sizeof(*tm), flags);
440 	if (!tm)
441 		return NULL;
442 
443 	tm->logical = eb->start;
444 	if (op != MOD_LOG_KEY_ADD) {
445 		btrfs_node_key(eb, &tm->key, slot);
446 		tm->blockptr = btrfs_node_blockptr(eb, slot);
447 	}
448 	tm->op = op;
449 	tm->slot = slot;
450 	tm->generation = btrfs_node_ptr_generation(eb, slot);
451 	RB_CLEAR_NODE(&tm->node);
452 
453 	return tm;
454 }
455 
tree_mod_log_insert_key(struct extent_buffer * eb,int slot,enum mod_log_op op,gfp_t flags)456 static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot,
457 		enum mod_log_op op, gfp_t flags)
458 {
459 	struct tree_mod_elem *tm;
460 	int ret;
461 
462 	if (!tree_mod_need_log(eb->fs_info, eb))
463 		return 0;
464 
465 	tm = alloc_tree_mod_elem(eb, slot, op, flags);
466 	if (!tm)
467 		return -ENOMEM;
468 
469 	if (tree_mod_dont_log(eb->fs_info, eb)) {
470 		kfree(tm);
471 		return 0;
472 	}
473 
474 	ret = __tree_mod_log_insert(eb->fs_info, tm);
475 	write_unlock(&eb->fs_info->tree_mod_log_lock);
476 	if (ret)
477 		kfree(tm);
478 
479 	return ret;
480 }
481 
tree_mod_log_insert_move(struct extent_buffer * eb,int dst_slot,int src_slot,int nr_items)482 static noinline int tree_mod_log_insert_move(struct extent_buffer *eb,
483 		int dst_slot, int src_slot, int nr_items)
484 {
485 	struct tree_mod_elem *tm = NULL;
486 	struct tree_mod_elem **tm_list = NULL;
487 	int ret = 0;
488 	int i;
489 	int locked = 0;
490 
491 	if (!tree_mod_need_log(eb->fs_info, eb))
492 		return 0;
493 
494 	tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
495 	if (!tm_list)
496 		return -ENOMEM;
497 
498 	tm = kzalloc(sizeof(*tm), GFP_NOFS);
499 	if (!tm) {
500 		ret = -ENOMEM;
501 		goto free_tms;
502 	}
503 
504 	tm->logical = eb->start;
505 	tm->slot = src_slot;
506 	tm->move.dst_slot = dst_slot;
507 	tm->move.nr_items = nr_items;
508 	tm->op = MOD_LOG_MOVE_KEYS;
509 
510 	for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
511 		tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
512 		    MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
513 		if (!tm_list[i]) {
514 			ret = -ENOMEM;
515 			goto free_tms;
516 		}
517 	}
518 
519 	if (tree_mod_dont_log(eb->fs_info, eb))
520 		goto free_tms;
521 	locked = 1;
522 
523 	/*
524 	 * When we override something during the move, we log these removals.
525 	 * This can only happen when we move towards the beginning of the
526 	 * buffer, i.e. dst_slot < src_slot.
527 	 */
528 	for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
529 		ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]);
530 		if (ret)
531 			goto free_tms;
532 	}
533 
534 	ret = __tree_mod_log_insert(eb->fs_info, tm);
535 	if (ret)
536 		goto free_tms;
537 	write_unlock(&eb->fs_info->tree_mod_log_lock);
538 	kfree(tm_list);
539 
540 	return 0;
541 free_tms:
542 	for (i = 0; i < nr_items; i++) {
543 		if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
544 			rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log);
545 		kfree(tm_list[i]);
546 	}
547 	if (locked)
548 		write_unlock(&eb->fs_info->tree_mod_log_lock);
549 	kfree(tm_list);
550 	kfree(tm);
551 
552 	return ret;
553 }
554 
555 static inline int
__tree_mod_log_free_eb(struct btrfs_fs_info * fs_info,struct tree_mod_elem ** tm_list,int nritems)556 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
557 		       struct tree_mod_elem **tm_list,
558 		       int nritems)
559 {
560 	int i, j;
561 	int ret;
562 
563 	for (i = nritems - 1; i >= 0; i--) {
564 		ret = __tree_mod_log_insert(fs_info, tm_list[i]);
565 		if (ret) {
566 			for (j = nritems - 1; j > i; j--)
567 				rb_erase(&tm_list[j]->node,
568 					 &fs_info->tree_mod_log);
569 			return ret;
570 		}
571 	}
572 
573 	return 0;
574 }
575 
tree_mod_log_insert_root(struct extent_buffer * old_root,struct extent_buffer * new_root,int log_removal)576 static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root,
577 			 struct extent_buffer *new_root, int log_removal)
578 {
579 	struct btrfs_fs_info *fs_info = old_root->fs_info;
580 	struct tree_mod_elem *tm = NULL;
581 	struct tree_mod_elem **tm_list = NULL;
582 	int nritems = 0;
583 	int ret = 0;
584 	int i;
585 
586 	if (!tree_mod_need_log(fs_info, NULL))
587 		return 0;
588 
589 	if (log_removal && btrfs_header_level(old_root) > 0) {
590 		nritems = btrfs_header_nritems(old_root);
591 		tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
592 				  GFP_NOFS);
593 		if (!tm_list) {
594 			ret = -ENOMEM;
595 			goto free_tms;
596 		}
597 		for (i = 0; i < nritems; i++) {
598 			tm_list[i] = alloc_tree_mod_elem(old_root, i,
599 			    MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
600 			if (!tm_list[i]) {
601 				ret = -ENOMEM;
602 				goto free_tms;
603 			}
604 		}
605 	}
606 
607 	tm = kzalloc(sizeof(*tm), GFP_NOFS);
608 	if (!tm) {
609 		ret = -ENOMEM;
610 		goto free_tms;
611 	}
612 
613 	tm->logical = new_root->start;
614 	tm->old_root.logical = old_root->start;
615 	tm->old_root.level = btrfs_header_level(old_root);
616 	tm->generation = btrfs_header_generation(old_root);
617 	tm->op = MOD_LOG_ROOT_REPLACE;
618 
619 	if (tree_mod_dont_log(fs_info, NULL))
620 		goto free_tms;
621 
622 	if (tm_list)
623 		ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
624 	if (!ret)
625 		ret = __tree_mod_log_insert(fs_info, tm);
626 
627 	write_unlock(&fs_info->tree_mod_log_lock);
628 	if (ret)
629 		goto free_tms;
630 	kfree(tm_list);
631 
632 	return ret;
633 
634 free_tms:
635 	if (tm_list) {
636 		for (i = 0; i < nritems; i++)
637 			kfree(tm_list[i]);
638 		kfree(tm_list);
639 	}
640 	kfree(tm);
641 
642 	return ret;
643 }
644 
645 static struct tree_mod_elem *
__tree_mod_log_search(struct btrfs_fs_info * fs_info,u64 start,u64 min_seq,int smallest)646 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
647 		      int smallest)
648 {
649 	struct rb_root *tm_root;
650 	struct rb_node *node;
651 	struct tree_mod_elem *cur = NULL;
652 	struct tree_mod_elem *found = NULL;
653 
654 	read_lock(&fs_info->tree_mod_log_lock);
655 	tm_root = &fs_info->tree_mod_log;
656 	node = tm_root->rb_node;
657 	while (node) {
658 		cur = rb_entry(node, struct tree_mod_elem, node);
659 		if (cur->logical < start) {
660 			node = node->rb_left;
661 		} else if (cur->logical > start) {
662 			node = node->rb_right;
663 		} else if (cur->seq < min_seq) {
664 			node = node->rb_left;
665 		} else if (!smallest) {
666 			/* we want the node with the highest seq */
667 			if (found)
668 				BUG_ON(found->seq > cur->seq);
669 			found = cur;
670 			node = node->rb_left;
671 		} else if (cur->seq > min_seq) {
672 			/* we want the node with the smallest seq */
673 			if (found)
674 				BUG_ON(found->seq < cur->seq);
675 			found = cur;
676 			node = node->rb_right;
677 		} else {
678 			found = cur;
679 			break;
680 		}
681 	}
682 	read_unlock(&fs_info->tree_mod_log_lock);
683 
684 	return found;
685 }
686 
687 /*
688  * this returns the element from the log with the smallest time sequence
689  * value that's in the log (the oldest log item). any element with a time
690  * sequence lower than min_seq will be ignored.
691  */
692 static struct tree_mod_elem *
tree_mod_log_search_oldest(struct btrfs_fs_info * fs_info,u64 start,u64 min_seq)693 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
694 			   u64 min_seq)
695 {
696 	return __tree_mod_log_search(fs_info, start, min_seq, 1);
697 }
698 
699 /*
700  * this returns the element from the log with the largest time sequence
701  * value that's in the log (the most recent log item). any element with
702  * a time sequence lower than min_seq will be ignored.
703  */
704 static struct tree_mod_elem *
tree_mod_log_search(struct btrfs_fs_info * fs_info,u64 start,u64 min_seq)705 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
706 {
707 	return __tree_mod_log_search(fs_info, start, min_seq, 0);
708 }
709 
tree_mod_log_eb_copy(struct extent_buffer * dst,struct extent_buffer * src,unsigned long dst_offset,unsigned long src_offset,int nr_items)710 static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst,
711 		     struct extent_buffer *src, unsigned long dst_offset,
712 		     unsigned long src_offset, int nr_items)
713 {
714 	struct btrfs_fs_info *fs_info = dst->fs_info;
715 	int ret = 0;
716 	struct tree_mod_elem **tm_list = NULL;
717 	struct tree_mod_elem **tm_list_add, **tm_list_rem;
718 	int i;
719 	int locked = 0;
720 
721 	if (!tree_mod_need_log(fs_info, NULL))
722 		return 0;
723 
724 	if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
725 		return 0;
726 
727 	tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
728 			  GFP_NOFS);
729 	if (!tm_list)
730 		return -ENOMEM;
731 
732 	tm_list_add = tm_list;
733 	tm_list_rem = tm_list + nr_items;
734 	for (i = 0; i < nr_items; i++) {
735 		tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
736 		    MOD_LOG_KEY_REMOVE, GFP_NOFS);
737 		if (!tm_list_rem[i]) {
738 			ret = -ENOMEM;
739 			goto free_tms;
740 		}
741 
742 		tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
743 		    MOD_LOG_KEY_ADD, GFP_NOFS);
744 		if (!tm_list_add[i]) {
745 			ret = -ENOMEM;
746 			goto free_tms;
747 		}
748 	}
749 
750 	if (tree_mod_dont_log(fs_info, NULL))
751 		goto free_tms;
752 	locked = 1;
753 
754 	for (i = 0; i < nr_items; i++) {
755 		ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
756 		if (ret)
757 			goto free_tms;
758 		ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
759 		if (ret)
760 			goto free_tms;
761 	}
762 
763 	write_unlock(&fs_info->tree_mod_log_lock);
764 	kfree(tm_list);
765 
766 	return 0;
767 
768 free_tms:
769 	for (i = 0; i < nr_items * 2; i++) {
770 		if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
771 			rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
772 		kfree(tm_list[i]);
773 	}
774 	if (locked)
775 		write_unlock(&fs_info->tree_mod_log_lock);
776 	kfree(tm_list);
777 
778 	return ret;
779 }
780 
tree_mod_log_free_eb(struct extent_buffer * eb)781 static noinline int tree_mod_log_free_eb(struct extent_buffer *eb)
782 {
783 	struct tree_mod_elem **tm_list = NULL;
784 	int nritems = 0;
785 	int i;
786 	int ret = 0;
787 
788 	if (btrfs_header_level(eb) == 0)
789 		return 0;
790 
791 	if (!tree_mod_need_log(eb->fs_info, NULL))
792 		return 0;
793 
794 	nritems = btrfs_header_nritems(eb);
795 	tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
796 	if (!tm_list)
797 		return -ENOMEM;
798 
799 	for (i = 0; i < nritems; i++) {
800 		tm_list[i] = alloc_tree_mod_elem(eb, i,
801 		    MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
802 		if (!tm_list[i]) {
803 			ret = -ENOMEM;
804 			goto free_tms;
805 		}
806 	}
807 
808 	if (tree_mod_dont_log(eb->fs_info, eb))
809 		goto free_tms;
810 
811 	ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems);
812 	write_unlock(&eb->fs_info->tree_mod_log_lock);
813 	if (ret)
814 		goto free_tms;
815 	kfree(tm_list);
816 
817 	return 0;
818 
819 free_tms:
820 	for (i = 0; i < nritems; i++)
821 		kfree(tm_list[i]);
822 	kfree(tm_list);
823 
824 	return ret;
825 }
826 
827 /*
828  * check if the tree block can be shared by multiple trees
829  */
btrfs_block_can_be_shared(struct btrfs_root * root,struct extent_buffer * buf)830 int btrfs_block_can_be_shared(struct btrfs_root *root,
831 			      struct extent_buffer *buf)
832 {
833 	/*
834 	 * Tree blocks not in shareable trees and tree roots are never shared.
835 	 * If a block was allocated after the last snapshot and the block was
836 	 * not allocated by tree relocation, we know the block is not shared.
837 	 */
838 	if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
839 	    buf != root->node && buf != root->commit_root &&
840 	    (btrfs_header_generation(buf) <=
841 	     btrfs_root_last_snapshot(&root->root_item) ||
842 	     btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
843 		return 1;
844 
845 	return 0;
846 }
847 
update_ref_for_cow(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf,struct extent_buffer * cow,int * last_ref)848 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
849 				       struct btrfs_root *root,
850 				       struct extent_buffer *buf,
851 				       struct extent_buffer *cow,
852 				       int *last_ref)
853 {
854 	struct btrfs_fs_info *fs_info = root->fs_info;
855 	u64 refs;
856 	u64 owner;
857 	u64 flags;
858 	u64 new_flags = 0;
859 	int ret;
860 
861 	/*
862 	 * Backrefs update rules:
863 	 *
864 	 * Always use full backrefs for extent pointers in tree block
865 	 * allocated by tree relocation.
866 	 *
867 	 * If a shared tree block is no longer referenced by its owner
868 	 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
869 	 * use full backrefs for extent pointers in tree block.
870 	 *
871 	 * If a tree block is been relocating
872 	 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
873 	 * use full backrefs for extent pointers in tree block.
874 	 * The reason for this is some operations (such as drop tree)
875 	 * are only allowed for blocks use full backrefs.
876 	 */
877 
878 	if (btrfs_block_can_be_shared(root, buf)) {
879 		ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
880 					       btrfs_header_level(buf), 1,
881 					       &refs, &flags);
882 		if (ret)
883 			return ret;
884 		if (refs == 0) {
885 			ret = -EROFS;
886 			btrfs_handle_fs_error(fs_info, ret, NULL);
887 			return ret;
888 		}
889 	} else {
890 		refs = 1;
891 		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
892 		    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
893 			flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
894 		else
895 			flags = 0;
896 	}
897 
898 	owner = btrfs_header_owner(buf);
899 	BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
900 	       !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
901 
902 	if (refs > 1) {
903 		if ((owner == root->root_key.objectid ||
904 		     root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
905 		    !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
906 			ret = btrfs_inc_ref(trans, root, buf, 1);
907 			if (ret)
908 				return ret;
909 
910 			if (root->root_key.objectid ==
911 			    BTRFS_TREE_RELOC_OBJECTID) {
912 				ret = btrfs_dec_ref(trans, root, buf, 0);
913 				if (ret)
914 					return ret;
915 				ret = btrfs_inc_ref(trans, root, cow, 1);
916 				if (ret)
917 					return ret;
918 			}
919 			new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
920 		} else {
921 
922 			if (root->root_key.objectid ==
923 			    BTRFS_TREE_RELOC_OBJECTID)
924 				ret = btrfs_inc_ref(trans, root, cow, 1);
925 			else
926 				ret = btrfs_inc_ref(trans, root, cow, 0);
927 			if (ret)
928 				return ret;
929 		}
930 		if (new_flags != 0) {
931 			int level = btrfs_header_level(buf);
932 
933 			ret = btrfs_set_disk_extent_flags(trans, buf,
934 							  new_flags, level, 0);
935 			if (ret)
936 				return ret;
937 		}
938 	} else {
939 		if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
940 			if (root->root_key.objectid ==
941 			    BTRFS_TREE_RELOC_OBJECTID)
942 				ret = btrfs_inc_ref(trans, root, cow, 1);
943 			else
944 				ret = btrfs_inc_ref(trans, root, cow, 0);
945 			if (ret)
946 				return ret;
947 			ret = btrfs_dec_ref(trans, root, buf, 1);
948 			if (ret)
949 				return ret;
950 		}
951 		btrfs_clean_tree_block(buf);
952 		*last_ref = 1;
953 	}
954 	return 0;
955 }
956 
alloc_tree_block_no_bg_flush(struct btrfs_trans_handle * trans,struct btrfs_root * root,u64 parent_start,const struct btrfs_disk_key * disk_key,int level,u64 hint,u64 empty_size,enum btrfs_lock_nesting nest)957 static struct extent_buffer *alloc_tree_block_no_bg_flush(
958 					  struct btrfs_trans_handle *trans,
959 					  struct btrfs_root *root,
960 					  u64 parent_start,
961 					  const struct btrfs_disk_key *disk_key,
962 					  int level,
963 					  u64 hint,
964 					  u64 empty_size,
965 					  enum btrfs_lock_nesting nest)
966 {
967 	struct btrfs_fs_info *fs_info = root->fs_info;
968 	struct extent_buffer *ret;
969 
970 	/*
971 	 * If we are COWing a node/leaf from the extent, chunk, device or free
972 	 * space trees, make sure that we do not finish block group creation of
973 	 * pending block groups. We do this to avoid a deadlock.
974 	 * COWing can result in allocation of a new chunk, and flushing pending
975 	 * block groups (btrfs_create_pending_block_groups()) can be triggered
976 	 * when finishing allocation of a new chunk. Creation of a pending block
977 	 * group modifies the extent, chunk, device and free space trees,
978 	 * therefore we could deadlock with ourselves since we are holding a
979 	 * lock on an extent buffer that btrfs_create_pending_block_groups() may
980 	 * try to COW later.
981 	 * For similar reasons, we also need to delay flushing pending block
982 	 * groups when splitting a leaf or node, from one of those trees, since
983 	 * we are holding a write lock on it and its parent or when inserting a
984 	 * new root node for one of those trees.
985 	 */
986 	if (root == fs_info->extent_root ||
987 	    root == fs_info->chunk_root ||
988 	    root == fs_info->dev_root ||
989 	    root == fs_info->free_space_root)
990 		trans->can_flush_pending_bgs = false;
991 
992 	ret = btrfs_alloc_tree_block(trans, root, parent_start,
993 				     root->root_key.objectid, disk_key, level,
994 				     hint, empty_size, nest);
995 	trans->can_flush_pending_bgs = true;
996 
997 	return ret;
998 }
999 
1000 /*
1001  * does the dirty work in cow of a single block.  The parent block (if
1002  * supplied) is updated to point to the new cow copy.  The new buffer is marked
1003  * dirty and returned locked.  If you modify the block it needs to be marked
1004  * dirty again.
1005  *
1006  * search_start -- an allocation hint for the new block
1007  *
1008  * empty_size -- a hint that you plan on doing more cow.  This is the size in
1009  * bytes the allocator should try to find free next to the block it returns.
1010  * This is just a hint and may be ignored by the allocator.
1011  */
__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)1012 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1013 			     struct btrfs_root *root,
1014 			     struct extent_buffer *buf,
1015 			     struct extent_buffer *parent, int parent_slot,
1016 			     struct extent_buffer **cow_ret,
1017 			     u64 search_start, u64 empty_size,
1018 			     enum btrfs_lock_nesting nest)
1019 {
1020 	struct btrfs_fs_info *fs_info = root->fs_info;
1021 	struct btrfs_disk_key disk_key;
1022 	struct extent_buffer *cow;
1023 	int level, ret;
1024 	int last_ref = 0;
1025 	int unlock_orig = 0;
1026 	u64 parent_start = 0;
1027 
1028 	if (*cow_ret == buf)
1029 		unlock_orig = 1;
1030 
1031 	btrfs_assert_tree_locked(buf);
1032 
1033 	WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
1034 		trans->transid != fs_info->running_transaction->transid);
1035 	WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
1036 		trans->transid != root->last_trans);
1037 
1038 	level = btrfs_header_level(buf);
1039 
1040 	if (level == 0)
1041 		btrfs_item_key(buf, &disk_key, 0);
1042 	else
1043 		btrfs_node_key(buf, &disk_key, 0);
1044 
1045 	if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
1046 		parent_start = parent->start;
1047 
1048 	cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key,
1049 					   level, search_start, empty_size, nest);
1050 	if (IS_ERR(cow))
1051 		return PTR_ERR(cow);
1052 
1053 	/* cow is set to blocking by btrfs_init_new_buffer */
1054 
1055 	copy_extent_buffer_full(cow, buf);
1056 	btrfs_set_header_bytenr(cow, cow->start);
1057 	btrfs_set_header_generation(cow, trans->transid);
1058 	btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1059 	btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1060 				     BTRFS_HEADER_FLAG_RELOC);
1061 	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1062 		btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1063 	else
1064 		btrfs_set_header_owner(cow, root->root_key.objectid);
1065 
1066 	write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
1067 
1068 	ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1069 	if (ret) {
1070 		btrfs_tree_unlock(cow);
1071 		free_extent_buffer(cow);
1072 		btrfs_abort_transaction(trans, ret);
1073 		return ret;
1074 	}
1075 
1076 	if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
1077 		ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1078 		if (ret) {
1079 			btrfs_tree_unlock(cow);
1080 			free_extent_buffer(cow);
1081 			btrfs_abort_transaction(trans, ret);
1082 			return ret;
1083 		}
1084 	}
1085 
1086 	if (buf == root->node) {
1087 		WARN_ON(parent && parent != buf);
1088 		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1089 		    btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1090 			parent_start = buf->start;
1091 
1092 		atomic_inc(&cow->refs);
1093 		ret = tree_mod_log_insert_root(root->node, cow, 1);
1094 		BUG_ON(ret < 0);
1095 		rcu_assign_pointer(root->node, cow);
1096 
1097 		btrfs_free_tree_block(trans, root, buf, parent_start,
1098 				      last_ref);
1099 		free_extent_buffer(buf);
1100 		add_root_to_dirty_list(root);
1101 	} else {
1102 		WARN_ON(trans->transid != btrfs_header_generation(parent));
1103 		tree_mod_log_insert_key(parent, parent_slot,
1104 					MOD_LOG_KEY_REPLACE, GFP_NOFS);
1105 		btrfs_set_node_blockptr(parent, parent_slot,
1106 					cow->start);
1107 		btrfs_set_node_ptr_generation(parent, parent_slot,
1108 					      trans->transid);
1109 		btrfs_mark_buffer_dirty(parent);
1110 		if (last_ref) {
1111 			ret = tree_mod_log_free_eb(buf);
1112 			if (ret) {
1113 				btrfs_tree_unlock(cow);
1114 				free_extent_buffer(cow);
1115 				btrfs_abort_transaction(trans, ret);
1116 				return ret;
1117 			}
1118 		}
1119 		btrfs_free_tree_block(trans, root, buf, parent_start,
1120 				      last_ref);
1121 	}
1122 	if (unlock_orig)
1123 		btrfs_tree_unlock(buf);
1124 	free_extent_buffer_stale(buf);
1125 	btrfs_mark_buffer_dirty(cow);
1126 	*cow_ret = cow;
1127 	return 0;
1128 }
1129 
1130 /*
1131  * returns the logical address of the oldest predecessor of the given root.
1132  * entries older than time_seq are ignored.
1133  */
__tree_mod_log_oldest_root(struct extent_buffer * eb_root,u64 time_seq)1134 static struct tree_mod_elem *__tree_mod_log_oldest_root(
1135 		struct extent_buffer *eb_root, u64 time_seq)
1136 {
1137 	struct tree_mod_elem *tm;
1138 	struct tree_mod_elem *found = NULL;
1139 	u64 root_logical = eb_root->start;
1140 	int looped = 0;
1141 
1142 	if (!time_seq)
1143 		return NULL;
1144 
1145 	/*
1146 	 * the very last operation that's logged for a root is the
1147 	 * replacement operation (if it is replaced at all). this has
1148 	 * the logical address of the *new* root, making it the very
1149 	 * first operation that's logged for this root.
1150 	 */
1151 	while (1) {
1152 		tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
1153 						time_seq);
1154 		if (!looped && !tm)
1155 			return NULL;
1156 		/*
1157 		 * if there are no tree operation for the oldest root, we simply
1158 		 * return it. this should only happen if that (old) root is at
1159 		 * level 0.
1160 		 */
1161 		if (!tm)
1162 			break;
1163 
1164 		/*
1165 		 * if there's an operation that's not a root replacement, we
1166 		 * found the oldest version of our root. normally, we'll find a
1167 		 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1168 		 */
1169 		if (tm->op != MOD_LOG_ROOT_REPLACE)
1170 			break;
1171 
1172 		found = tm;
1173 		root_logical = tm->old_root.logical;
1174 		looped = 1;
1175 	}
1176 
1177 	/* if there's no old root to return, return what we found instead */
1178 	if (!found)
1179 		found = tm;
1180 
1181 	return found;
1182 }
1183 
1184 /*
1185  * tm is a pointer to the first operation to rewind within eb. then, all
1186  * previous operations will be rewound (until we reach something older than
1187  * time_seq).
1188  */
1189 static void
__tree_mod_log_rewind(struct btrfs_fs_info * fs_info,struct extent_buffer * eb,u64 time_seq,struct tree_mod_elem * first_tm)1190 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1191 		      u64 time_seq, struct tree_mod_elem *first_tm)
1192 {
1193 	u32 n;
1194 	struct rb_node *next;
1195 	struct tree_mod_elem *tm = first_tm;
1196 	unsigned long o_dst;
1197 	unsigned long o_src;
1198 	unsigned long p_size = sizeof(struct btrfs_key_ptr);
1199 
1200 	n = btrfs_header_nritems(eb);
1201 	read_lock(&fs_info->tree_mod_log_lock);
1202 	while (tm && tm->seq >= time_seq) {
1203 		/*
1204 		 * all the operations are recorded with the operator used for
1205 		 * the modification. as we're going backwards, we do the
1206 		 * opposite of each operation here.
1207 		 */
1208 		switch (tm->op) {
1209 		case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1210 			BUG_ON(tm->slot < n);
1211 			fallthrough;
1212 		case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1213 		case MOD_LOG_KEY_REMOVE:
1214 			btrfs_set_node_key(eb, &tm->key, tm->slot);
1215 			btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1216 			btrfs_set_node_ptr_generation(eb, tm->slot,
1217 						      tm->generation);
1218 			n++;
1219 			break;
1220 		case MOD_LOG_KEY_REPLACE:
1221 			BUG_ON(tm->slot >= n);
1222 			btrfs_set_node_key(eb, &tm->key, tm->slot);
1223 			btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1224 			btrfs_set_node_ptr_generation(eb, tm->slot,
1225 						      tm->generation);
1226 			break;
1227 		case MOD_LOG_KEY_ADD:
1228 			/* if a move operation is needed it's in the log */
1229 			n--;
1230 			break;
1231 		case MOD_LOG_MOVE_KEYS:
1232 			o_dst = btrfs_node_key_ptr_offset(tm->slot);
1233 			o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1234 			memmove_extent_buffer(eb, o_dst, o_src,
1235 					      tm->move.nr_items * p_size);
1236 			break;
1237 		case MOD_LOG_ROOT_REPLACE:
1238 			/*
1239 			 * this operation is special. for roots, this must be
1240 			 * handled explicitly before rewinding.
1241 			 * for non-roots, this operation may exist if the node
1242 			 * was a root: root A -> child B; then A gets empty and
1243 			 * B is promoted to the new root. in the mod log, we'll
1244 			 * have a root-replace operation for B, a tree block
1245 			 * that is no root. we simply ignore that operation.
1246 			 */
1247 			break;
1248 		}
1249 		next = rb_next(&tm->node);
1250 		if (!next)
1251 			break;
1252 		tm = rb_entry(next, struct tree_mod_elem, node);
1253 		if (tm->logical != first_tm->logical)
1254 			break;
1255 	}
1256 	read_unlock(&fs_info->tree_mod_log_lock);
1257 	btrfs_set_header_nritems(eb, n);
1258 }
1259 
1260 /*
1261  * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1262  * is returned. If rewind operations happen, a fresh buffer is returned. The
1263  * returned buffer is always read-locked. If the returned buffer is not the
1264  * input buffer, the lock on the input buffer is released and the input buffer
1265  * is freed (its refcount is decremented).
1266  */
1267 static struct extent_buffer *
tree_mod_log_rewind(struct btrfs_fs_info * fs_info,struct btrfs_path * path,struct extent_buffer * eb,u64 time_seq)1268 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1269 		    struct extent_buffer *eb, u64 time_seq)
1270 {
1271 	struct extent_buffer *eb_rewin;
1272 	struct tree_mod_elem *tm;
1273 
1274 	if (!time_seq)
1275 		return eb;
1276 
1277 	if (btrfs_header_level(eb) == 0)
1278 		return eb;
1279 
1280 	tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1281 	if (!tm)
1282 		return eb;
1283 
1284 	btrfs_set_path_blocking(path);
1285 	btrfs_set_lock_blocking_read(eb);
1286 
1287 	if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1288 		BUG_ON(tm->slot != 0);
1289 		eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1290 		if (!eb_rewin) {
1291 			btrfs_tree_read_unlock_blocking(eb);
1292 			free_extent_buffer(eb);
1293 			return NULL;
1294 		}
1295 		btrfs_set_header_bytenr(eb_rewin, eb->start);
1296 		btrfs_set_header_backref_rev(eb_rewin,
1297 					     btrfs_header_backref_rev(eb));
1298 		btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1299 		btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1300 	} else {
1301 		eb_rewin = btrfs_clone_extent_buffer(eb);
1302 		if (!eb_rewin) {
1303 			btrfs_tree_read_unlock_blocking(eb);
1304 			free_extent_buffer(eb);
1305 			return NULL;
1306 		}
1307 	}
1308 
1309 	btrfs_tree_read_unlock_blocking(eb);
1310 	free_extent_buffer(eb);
1311 
1312 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb_rewin),
1313 				       eb_rewin, btrfs_header_level(eb_rewin));
1314 	btrfs_tree_read_lock(eb_rewin);
1315 	__tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1316 	WARN_ON(btrfs_header_nritems(eb_rewin) >
1317 		BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1318 
1319 	return eb_rewin;
1320 }
1321 
1322 /*
1323  * get_old_root() rewinds the state of @root's root node to the given @time_seq
1324  * value. If there are no changes, the current root->root_node is returned. If
1325  * anything changed in between, there's a fresh buffer allocated on which the
1326  * rewind operations are done. In any case, the returned buffer is read locked.
1327  * Returns NULL on error (with no locks held).
1328  */
1329 static inline struct extent_buffer *
get_old_root(struct btrfs_root * root,u64 time_seq)1330 get_old_root(struct btrfs_root *root, u64 time_seq)
1331 {
1332 	struct btrfs_fs_info *fs_info = root->fs_info;
1333 	struct tree_mod_elem *tm;
1334 	struct extent_buffer *eb = NULL;
1335 	struct extent_buffer *eb_root;
1336 	u64 eb_root_owner = 0;
1337 	struct extent_buffer *old;
1338 	struct tree_mod_root *old_root = NULL;
1339 	u64 old_generation = 0;
1340 	u64 logical;
1341 	int level;
1342 
1343 	eb_root = btrfs_read_lock_root_node(root);
1344 	tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1345 	if (!tm)
1346 		return eb_root;
1347 
1348 	if (tm->op == MOD_LOG_ROOT_REPLACE) {
1349 		old_root = &tm->old_root;
1350 		old_generation = tm->generation;
1351 		logical = old_root->logical;
1352 		level = old_root->level;
1353 	} else {
1354 		logical = eb_root->start;
1355 		level = btrfs_header_level(eb_root);
1356 	}
1357 
1358 	tm = tree_mod_log_search(fs_info, logical, time_seq);
1359 	if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1360 		btrfs_tree_read_unlock(eb_root);
1361 		free_extent_buffer(eb_root);
1362 		old = read_tree_block(fs_info, logical, 0, level, NULL);
1363 		if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1364 			if (!IS_ERR(old))
1365 				free_extent_buffer(old);
1366 			btrfs_warn(fs_info,
1367 				   "failed to read tree block %llu from get_old_root",
1368 				   logical);
1369 		} else {
1370 			struct tree_mod_elem *tm2;
1371 
1372 			btrfs_tree_read_lock(old);
1373 			eb = btrfs_clone_extent_buffer(old);
1374 			/*
1375 			 * After the lookup for the most recent tree mod operation
1376 			 * above and before we locked and cloned the extent buffer
1377 			 * 'old', a new tree mod log operation may have been added.
1378 			 * So lookup for a more recent one to make sure the number
1379 			 * of mod log operations we replay is consistent with the
1380 			 * number of items we have in the cloned extent buffer,
1381 			 * otherwise we can hit a BUG_ON when rewinding the extent
1382 			 * buffer.
1383 			 */
1384 			tm2 = tree_mod_log_search(fs_info, logical, time_seq);
1385 			btrfs_tree_read_unlock(old);
1386 			free_extent_buffer(old);
1387 			ASSERT(tm2);
1388 			ASSERT(tm2 == tm || tm2->seq > tm->seq);
1389 			if (!tm2 || tm2->seq < tm->seq) {
1390 				free_extent_buffer(eb);
1391 				return NULL;
1392 			}
1393 			tm = tm2;
1394 		}
1395 	} else if (old_root) {
1396 		eb_root_owner = btrfs_header_owner(eb_root);
1397 		btrfs_tree_read_unlock(eb_root);
1398 		free_extent_buffer(eb_root);
1399 		eb = alloc_dummy_extent_buffer(fs_info, logical);
1400 	} else {
1401 		btrfs_set_lock_blocking_read(eb_root);
1402 		eb = btrfs_clone_extent_buffer(eb_root);
1403 		btrfs_tree_read_unlock_blocking(eb_root);
1404 		free_extent_buffer(eb_root);
1405 	}
1406 
1407 	if (!eb)
1408 		return NULL;
1409 	if (old_root) {
1410 		btrfs_set_header_bytenr(eb, eb->start);
1411 		btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1412 		btrfs_set_header_owner(eb, eb_root_owner);
1413 		btrfs_set_header_level(eb, old_root->level);
1414 		btrfs_set_header_generation(eb, old_generation);
1415 	}
1416 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), eb,
1417 				       btrfs_header_level(eb));
1418 	btrfs_tree_read_lock(eb);
1419 	if (tm)
1420 		__tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1421 	else
1422 		WARN_ON(btrfs_header_level(eb) != 0);
1423 	WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1424 
1425 	return eb;
1426 }
1427 
btrfs_old_root_level(struct btrfs_root * root,u64 time_seq)1428 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1429 {
1430 	struct tree_mod_elem *tm;
1431 	int level;
1432 	struct extent_buffer *eb_root = btrfs_root_node(root);
1433 
1434 	tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1435 	if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1436 		level = tm->old_root.level;
1437 	} else {
1438 		level = btrfs_header_level(eb_root);
1439 	}
1440 	free_extent_buffer(eb_root);
1441 
1442 	return level;
1443 }
1444 
should_cow_block(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct extent_buffer * buf)1445 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1446 				   struct btrfs_root *root,
1447 				   struct extent_buffer *buf)
1448 {
1449 	if (btrfs_is_testing(root->fs_info))
1450 		return 0;
1451 
1452 	/* Ensure we can see the FORCE_COW bit */
1453 	smp_mb__before_atomic();
1454 
1455 	/*
1456 	 * We do not need to cow a block if
1457 	 * 1) this block is not created or changed in this transaction;
1458 	 * 2) this block does not belong to TREE_RELOC tree;
1459 	 * 3) the root is not forced COW.
1460 	 *
1461 	 * What is forced COW:
1462 	 *    when we create snapshot during committing the transaction,
1463 	 *    after we've finished copying src root, we must COW the shared
1464 	 *    block to ensure the metadata consistency.
1465 	 */
1466 	if (btrfs_header_generation(buf) == trans->transid &&
1467 	    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1468 	    !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1469 	      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1470 	    !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1471 		return 0;
1472 	return 1;
1473 }
1474 
1475 /*
1476  * cows a single block, see __btrfs_cow_block for the real work.
1477  * This version of it has extra checks so that a block isn't COWed more than
1478  * once per transaction, as long as it hasn't been written yet
1479  */
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)1480 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1481 		    struct btrfs_root *root, struct extent_buffer *buf,
1482 		    struct extent_buffer *parent, int parent_slot,
1483 		    struct extent_buffer **cow_ret,
1484 		    enum btrfs_lock_nesting nest)
1485 {
1486 	struct btrfs_fs_info *fs_info = root->fs_info;
1487 	u64 search_start;
1488 	int ret;
1489 
1490 	if (test_bit(BTRFS_ROOT_DELETING, &root->state))
1491 		btrfs_err(fs_info,
1492 			"COW'ing blocks on a fs root that's being dropped");
1493 
1494 	if (trans->transaction != fs_info->running_transaction)
1495 		WARN(1, KERN_CRIT "trans %llu running %llu\n",
1496 		       trans->transid,
1497 		       fs_info->running_transaction->transid);
1498 
1499 	if (trans->transid != fs_info->generation)
1500 		WARN(1, KERN_CRIT "trans %llu running %llu\n",
1501 		       trans->transid, fs_info->generation);
1502 
1503 	if (!should_cow_block(trans, root, buf)) {
1504 		trans->dirty = true;
1505 		*cow_ret = buf;
1506 		return 0;
1507 	}
1508 
1509 	search_start = buf->start & ~((u64)SZ_1G - 1);
1510 
1511 	if (parent)
1512 		btrfs_set_lock_blocking_write(parent);
1513 	btrfs_set_lock_blocking_write(buf);
1514 
1515 	/*
1516 	 * Before CoWing this block for later modification, check if it's
1517 	 * the subtree root and do the delayed subtree trace if needed.
1518 	 *
1519 	 * Also We don't care about the error, as it's handled internally.
1520 	 */
1521 	btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
1522 	ret = __btrfs_cow_block(trans, root, buf, parent,
1523 				 parent_slot, cow_ret, search_start, 0, nest);
1524 
1525 	trace_btrfs_cow_block(root, buf, *cow_ret);
1526 
1527 	return ret;
1528 }
1529 
1530 /*
1531  * helper function for defrag to decide if two blocks pointed to by a
1532  * node are actually close by
1533  */
close_blocks(u64 blocknr,u64 other,u32 blocksize)1534 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1535 {
1536 	if (blocknr < other && other - (blocknr + blocksize) < 32768)
1537 		return 1;
1538 	if (blocknr > other && blocknr - (other + blocksize) < 32768)
1539 		return 1;
1540 	return 0;
1541 }
1542 
1543 #ifdef __LITTLE_ENDIAN
1544 
1545 /*
1546  * Compare two keys, on little-endian the disk order is same as CPU order and
1547  * we can avoid the conversion.
1548  */
comp_keys(const struct btrfs_disk_key * disk_key,const struct btrfs_key * k2)1549 static int comp_keys(const struct btrfs_disk_key *disk_key,
1550 		     const struct btrfs_key *k2)
1551 {
1552 	const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key;
1553 
1554 	return btrfs_comp_cpu_keys(k1, k2);
1555 }
1556 
1557 #else
1558 
1559 /*
1560  * compare two keys in a memcmp fashion
1561  */
comp_keys(const struct btrfs_disk_key * disk,const struct btrfs_key * k2)1562 static int comp_keys(const struct btrfs_disk_key *disk,
1563 		     const struct btrfs_key *k2)
1564 {
1565 	struct btrfs_key k1;
1566 
1567 	btrfs_disk_key_to_cpu(&k1, disk);
1568 
1569 	return btrfs_comp_cpu_keys(&k1, k2);
1570 }
1571 #endif
1572 
1573 /*
1574  * same as comp_keys only with two btrfs_key's
1575  */
btrfs_comp_cpu_keys(const struct btrfs_key * k1,const struct btrfs_key * k2)1576 int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1577 {
1578 	if (k1->objectid > k2->objectid)
1579 		return 1;
1580 	if (k1->objectid < k2->objectid)
1581 		return -1;
1582 	if (k1->type > k2->type)
1583 		return 1;
1584 	if (k1->type < k2->type)
1585 		return -1;
1586 	if (k1->offset > k2->offset)
1587 		return 1;
1588 	if (k1->offset < k2->offset)
1589 		return -1;
1590 	return 0;
1591 }
1592 
1593 /*
1594  * this is used by the defrag code to go through all the
1595  * leaves pointed to by a node and reallocate them so that
1596  * disk order is close to key order
1597  */
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)1598 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1599 		       struct btrfs_root *root, struct extent_buffer *parent,
1600 		       int start_slot, u64 *last_ret,
1601 		       struct btrfs_key *progress)
1602 {
1603 	struct btrfs_fs_info *fs_info = root->fs_info;
1604 	struct extent_buffer *cur;
1605 	u64 blocknr;
1606 	u64 gen;
1607 	u64 search_start = *last_ret;
1608 	u64 last_block = 0;
1609 	u64 other;
1610 	u32 parent_nritems;
1611 	int end_slot;
1612 	int i;
1613 	int err = 0;
1614 	int parent_level;
1615 	int uptodate;
1616 	u32 blocksize;
1617 	int progress_passed = 0;
1618 	struct btrfs_disk_key disk_key;
1619 
1620 	parent_level = btrfs_header_level(parent);
1621 
1622 	WARN_ON(trans->transaction != fs_info->running_transaction);
1623 	WARN_ON(trans->transid != fs_info->generation);
1624 
1625 	parent_nritems = btrfs_header_nritems(parent);
1626 	blocksize = fs_info->nodesize;
1627 	end_slot = parent_nritems - 1;
1628 
1629 	if (parent_nritems <= 1)
1630 		return 0;
1631 
1632 	btrfs_set_lock_blocking_write(parent);
1633 
1634 	for (i = start_slot; i <= end_slot; i++) {
1635 		struct btrfs_key first_key;
1636 		int close = 1;
1637 
1638 		btrfs_node_key(parent, &disk_key, i);
1639 		if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1640 			continue;
1641 
1642 		progress_passed = 1;
1643 		blocknr = btrfs_node_blockptr(parent, i);
1644 		gen = btrfs_node_ptr_generation(parent, i);
1645 		btrfs_node_key_to_cpu(parent, &first_key, i);
1646 		if (last_block == 0)
1647 			last_block = blocknr;
1648 
1649 		if (i > 0) {
1650 			other = btrfs_node_blockptr(parent, i - 1);
1651 			close = close_blocks(blocknr, other, blocksize);
1652 		}
1653 		if (!close && i < end_slot) {
1654 			other = btrfs_node_blockptr(parent, i + 1);
1655 			close = close_blocks(blocknr, other, blocksize);
1656 		}
1657 		if (close) {
1658 			last_block = blocknr;
1659 			continue;
1660 		}
1661 
1662 		cur = find_extent_buffer(fs_info, blocknr);
1663 		if (cur)
1664 			uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1665 		else
1666 			uptodate = 0;
1667 		if (!cur || !uptodate) {
1668 			if (!cur) {
1669 				cur = read_tree_block(fs_info, blocknr, gen,
1670 						      parent_level - 1,
1671 						      &first_key);
1672 				if (IS_ERR(cur)) {
1673 					return PTR_ERR(cur);
1674 				} else if (!extent_buffer_uptodate(cur)) {
1675 					free_extent_buffer(cur);
1676 					return -EIO;
1677 				}
1678 			} else if (!uptodate) {
1679 				err = btrfs_read_buffer(cur, gen,
1680 						parent_level - 1,&first_key);
1681 				if (err) {
1682 					free_extent_buffer(cur);
1683 					return err;
1684 				}
1685 			}
1686 		}
1687 		if (search_start == 0)
1688 			search_start = last_block;
1689 
1690 		btrfs_tree_lock(cur);
1691 		btrfs_set_lock_blocking_write(cur);
1692 		err = __btrfs_cow_block(trans, root, cur, parent, i,
1693 					&cur, search_start,
1694 					min(16 * blocksize,
1695 					    (end_slot - i) * blocksize),
1696 					BTRFS_NESTING_COW);
1697 		if (err) {
1698 			btrfs_tree_unlock(cur);
1699 			free_extent_buffer(cur);
1700 			break;
1701 		}
1702 		search_start = cur->start;
1703 		last_block = cur->start;
1704 		*last_ret = search_start;
1705 		btrfs_tree_unlock(cur);
1706 		free_extent_buffer(cur);
1707 	}
1708 	return err;
1709 }
1710 
1711 /*
1712  * search for key in the extent_buffer.  The items start at offset p,
1713  * and they are item_size apart.  There are 'max' items in p.
1714  *
1715  * the slot in the array is returned via slot, and it points to
1716  * the place where you would insert key if it is not found in
1717  * the array.
1718  *
1719  * slot may point to max if the key is bigger than all of the keys
1720  */
generic_bin_search(struct extent_buffer * eb,unsigned long p,int item_size,const struct btrfs_key * key,int max,int * slot)1721 static noinline int generic_bin_search(struct extent_buffer *eb,
1722 				       unsigned long p, int item_size,
1723 				       const struct btrfs_key *key,
1724 				       int max, int *slot)
1725 {
1726 	int low = 0;
1727 	int high = max;
1728 	int ret;
1729 	const int key_size = sizeof(struct btrfs_disk_key);
1730 
1731 	if (low > high) {
1732 		btrfs_err(eb->fs_info,
1733 		 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1734 			  __func__, low, high, eb->start,
1735 			  btrfs_header_owner(eb), btrfs_header_level(eb));
1736 		return -EINVAL;
1737 	}
1738 
1739 	while (low < high) {
1740 		unsigned long oip;
1741 		unsigned long offset;
1742 		struct btrfs_disk_key *tmp;
1743 		struct btrfs_disk_key unaligned;
1744 		int mid;
1745 
1746 		mid = (low + high) / 2;
1747 		offset = p + mid * item_size;
1748 		oip = offset_in_page(offset);
1749 
1750 		if (oip + key_size <= PAGE_SIZE) {
1751 			const unsigned long idx = offset >> PAGE_SHIFT;
1752 			char *kaddr = page_address(eb->pages[idx]);
1753 
1754 			tmp = (struct btrfs_disk_key *)(kaddr + oip);
1755 		} else {
1756 			read_extent_buffer(eb, &unaligned, offset, key_size);
1757 			tmp = &unaligned;
1758 		}
1759 
1760 		ret = comp_keys(tmp, key);
1761 
1762 		if (ret < 0)
1763 			low = mid + 1;
1764 		else if (ret > 0)
1765 			high = mid;
1766 		else {
1767 			*slot = mid;
1768 			return 0;
1769 		}
1770 	}
1771 	*slot = low;
1772 	return 1;
1773 }
1774 
1775 /*
1776  * simple bin_search frontend that does the right thing for
1777  * leaves vs nodes
1778  */
btrfs_bin_search(struct extent_buffer * eb,const struct btrfs_key * key,int * slot)1779 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1780 		     int *slot)
1781 {
1782 	if (btrfs_header_level(eb) == 0)
1783 		return generic_bin_search(eb,
1784 					  offsetof(struct btrfs_leaf, items),
1785 					  sizeof(struct btrfs_item),
1786 					  key, btrfs_header_nritems(eb),
1787 					  slot);
1788 	else
1789 		return generic_bin_search(eb,
1790 					  offsetof(struct btrfs_node, ptrs),
1791 					  sizeof(struct btrfs_key_ptr),
1792 					  key, btrfs_header_nritems(eb),
1793 					  slot);
1794 }
1795 
root_add_used(struct btrfs_root * root,u32 size)1796 static void root_add_used(struct btrfs_root *root, u32 size)
1797 {
1798 	spin_lock(&root->accounting_lock);
1799 	btrfs_set_root_used(&root->root_item,
1800 			    btrfs_root_used(&root->root_item) + size);
1801 	spin_unlock(&root->accounting_lock);
1802 }
1803 
root_sub_used(struct btrfs_root * root,u32 size)1804 static void root_sub_used(struct btrfs_root *root, u32 size)
1805 {
1806 	spin_lock(&root->accounting_lock);
1807 	btrfs_set_root_used(&root->root_item,
1808 			    btrfs_root_used(&root->root_item) - size);
1809 	spin_unlock(&root->accounting_lock);
1810 }
1811 
1812 /* given a node and slot number, this reads the blocks it points to.  The
1813  * extent buffer is returned with a reference taken (but unlocked).
1814  */
btrfs_read_node_slot(struct extent_buffer * parent,int slot)1815 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
1816 					   int slot)
1817 {
1818 	int level = btrfs_header_level(parent);
1819 	struct extent_buffer *eb;
1820 	struct btrfs_key first_key;
1821 
1822 	if (slot < 0 || slot >= btrfs_header_nritems(parent))
1823 		return ERR_PTR(-ENOENT);
1824 
1825 	BUG_ON(level == 0);
1826 
1827 	btrfs_node_key_to_cpu(parent, &first_key, slot);
1828 	eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
1829 			     btrfs_node_ptr_generation(parent, slot),
1830 			     level - 1, &first_key);
1831 	if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1832 		free_extent_buffer(eb);
1833 		eb = ERR_PTR(-EIO);
1834 	}
1835 
1836 	return eb;
1837 }
1838 
1839 /*
1840  * node level balancing, used to make sure nodes are in proper order for
1841  * item deletion.  We balance from the top down, so we have to make sure
1842  * that a deletion won't leave an node completely empty later on.
1843  */
balance_level(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int level)1844 static noinline int balance_level(struct btrfs_trans_handle *trans,
1845 			 struct btrfs_root *root,
1846 			 struct btrfs_path *path, int level)
1847 {
1848 	struct btrfs_fs_info *fs_info = root->fs_info;
1849 	struct extent_buffer *right = NULL;
1850 	struct extent_buffer *mid;
1851 	struct extent_buffer *left = NULL;
1852 	struct extent_buffer *parent = NULL;
1853 	int ret = 0;
1854 	int wret;
1855 	int pslot;
1856 	int orig_slot = path->slots[level];
1857 	u64 orig_ptr;
1858 
1859 	ASSERT(level > 0);
1860 
1861 	mid = path->nodes[level];
1862 
1863 	WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1864 		path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1865 	WARN_ON(btrfs_header_generation(mid) != trans->transid);
1866 
1867 	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1868 
1869 	if (level < BTRFS_MAX_LEVEL - 1) {
1870 		parent = path->nodes[level + 1];
1871 		pslot = path->slots[level + 1];
1872 	}
1873 
1874 	/*
1875 	 * deal with the case where there is only one pointer in the root
1876 	 * by promoting the node below to a root
1877 	 */
1878 	if (!parent) {
1879 		struct extent_buffer *child;
1880 
1881 		if (btrfs_header_nritems(mid) != 1)
1882 			return 0;
1883 
1884 		/* promote the child to a root */
1885 		child = btrfs_read_node_slot(mid, 0);
1886 		if (IS_ERR(child)) {
1887 			ret = PTR_ERR(child);
1888 			btrfs_handle_fs_error(fs_info, ret, NULL);
1889 			goto enospc;
1890 		}
1891 
1892 		btrfs_tree_lock(child);
1893 		btrfs_set_lock_blocking_write(child);
1894 		ret = btrfs_cow_block(trans, root, child, mid, 0, &child,
1895 				      BTRFS_NESTING_COW);
1896 		if (ret) {
1897 			btrfs_tree_unlock(child);
1898 			free_extent_buffer(child);
1899 			goto enospc;
1900 		}
1901 
1902 		ret = tree_mod_log_insert_root(root->node, child, 1);
1903 		BUG_ON(ret < 0);
1904 		rcu_assign_pointer(root->node, child);
1905 
1906 		add_root_to_dirty_list(root);
1907 		btrfs_tree_unlock(child);
1908 
1909 		path->locks[level] = 0;
1910 		path->nodes[level] = NULL;
1911 		btrfs_clean_tree_block(mid);
1912 		btrfs_tree_unlock(mid);
1913 		/* once for the path */
1914 		free_extent_buffer(mid);
1915 
1916 		root_sub_used(root, mid->len);
1917 		btrfs_free_tree_block(trans, root, mid, 0, 1);
1918 		/* once for the root ptr */
1919 		free_extent_buffer_stale(mid);
1920 		return 0;
1921 	}
1922 	if (btrfs_header_nritems(mid) >
1923 	    BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1924 		return 0;
1925 
1926 	left = btrfs_read_node_slot(parent, pslot - 1);
1927 	if (IS_ERR(left))
1928 		left = NULL;
1929 
1930 	if (left) {
1931 		__btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
1932 		btrfs_set_lock_blocking_write(left);
1933 		wret = btrfs_cow_block(trans, root, left,
1934 				       parent, pslot - 1, &left,
1935 				       BTRFS_NESTING_LEFT_COW);
1936 		if (wret) {
1937 			ret = wret;
1938 			goto enospc;
1939 		}
1940 	}
1941 
1942 	right = btrfs_read_node_slot(parent, pslot + 1);
1943 	if (IS_ERR(right))
1944 		right = NULL;
1945 
1946 	if (right) {
1947 		__btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
1948 		btrfs_set_lock_blocking_write(right);
1949 		wret = btrfs_cow_block(trans, root, right,
1950 				       parent, pslot + 1, &right,
1951 				       BTRFS_NESTING_RIGHT_COW);
1952 		if (wret) {
1953 			ret = wret;
1954 			goto enospc;
1955 		}
1956 	}
1957 
1958 	/* first, try to make some room in the middle buffer */
1959 	if (left) {
1960 		orig_slot += btrfs_header_nritems(left);
1961 		wret = push_node_left(trans, left, mid, 1);
1962 		if (wret < 0)
1963 			ret = wret;
1964 	}
1965 
1966 	/*
1967 	 * then try to empty the right most buffer into the middle
1968 	 */
1969 	if (right) {
1970 		wret = push_node_left(trans, mid, right, 1);
1971 		if (wret < 0 && wret != -ENOSPC)
1972 			ret = wret;
1973 		if (btrfs_header_nritems(right) == 0) {
1974 			btrfs_clean_tree_block(right);
1975 			btrfs_tree_unlock(right);
1976 			del_ptr(root, path, level + 1, pslot + 1);
1977 			root_sub_used(root, right->len);
1978 			btrfs_free_tree_block(trans, root, right, 0, 1);
1979 			free_extent_buffer_stale(right);
1980 			right = NULL;
1981 		} else {
1982 			struct btrfs_disk_key right_key;
1983 			btrfs_node_key(right, &right_key, 0);
1984 			ret = tree_mod_log_insert_key(parent, pslot + 1,
1985 					MOD_LOG_KEY_REPLACE, GFP_NOFS);
1986 			BUG_ON(ret < 0);
1987 			btrfs_set_node_key(parent, &right_key, pslot + 1);
1988 			btrfs_mark_buffer_dirty(parent);
1989 		}
1990 	}
1991 	if (btrfs_header_nritems(mid) == 1) {
1992 		/*
1993 		 * we're not allowed to leave a node with one item in the
1994 		 * tree during a delete.  A deletion from lower in the tree
1995 		 * could try to delete the only pointer in this node.
1996 		 * So, pull some keys from the left.
1997 		 * There has to be a left pointer at this point because
1998 		 * otherwise we would have pulled some pointers from the
1999 		 * right
2000 		 */
2001 		if (!left) {
2002 			ret = -EROFS;
2003 			btrfs_handle_fs_error(fs_info, ret, NULL);
2004 			goto enospc;
2005 		}
2006 		wret = balance_node_right(trans, mid, left);
2007 		if (wret < 0) {
2008 			ret = wret;
2009 			goto enospc;
2010 		}
2011 		if (wret == 1) {
2012 			wret = push_node_left(trans, left, mid, 1);
2013 			if (wret < 0)
2014 				ret = wret;
2015 		}
2016 		BUG_ON(wret == 1);
2017 	}
2018 	if (btrfs_header_nritems(mid) == 0) {
2019 		btrfs_clean_tree_block(mid);
2020 		btrfs_tree_unlock(mid);
2021 		del_ptr(root, path, level + 1, pslot);
2022 		root_sub_used(root, mid->len);
2023 		btrfs_free_tree_block(trans, root, mid, 0, 1);
2024 		free_extent_buffer_stale(mid);
2025 		mid = NULL;
2026 	} else {
2027 		/* update the parent key to reflect our changes */
2028 		struct btrfs_disk_key mid_key;
2029 		btrfs_node_key(mid, &mid_key, 0);
2030 		ret = tree_mod_log_insert_key(parent, pslot,
2031 				MOD_LOG_KEY_REPLACE, GFP_NOFS);
2032 		BUG_ON(ret < 0);
2033 		btrfs_set_node_key(parent, &mid_key, pslot);
2034 		btrfs_mark_buffer_dirty(parent);
2035 	}
2036 
2037 	/* update the path */
2038 	if (left) {
2039 		if (btrfs_header_nritems(left) > orig_slot) {
2040 			atomic_inc(&left->refs);
2041 			/* left was locked after cow */
2042 			path->nodes[level] = left;
2043 			path->slots[level + 1] -= 1;
2044 			path->slots[level] = orig_slot;
2045 			if (mid) {
2046 				btrfs_tree_unlock(mid);
2047 				free_extent_buffer(mid);
2048 			}
2049 		} else {
2050 			orig_slot -= btrfs_header_nritems(left);
2051 			path->slots[level] = orig_slot;
2052 		}
2053 	}
2054 	/* double check we haven't messed things up */
2055 	if (orig_ptr !=
2056 	    btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2057 		BUG();
2058 enospc:
2059 	if (right) {
2060 		btrfs_tree_unlock(right);
2061 		free_extent_buffer(right);
2062 	}
2063 	if (left) {
2064 		if (path->nodes[level] != left)
2065 			btrfs_tree_unlock(left);
2066 		free_extent_buffer(left);
2067 	}
2068 	return ret;
2069 }
2070 
2071 /* Node balancing for insertion.  Here we only split or push nodes around
2072  * when they are completely full.  This is also done top down, so we
2073  * have to be pessimistic.
2074  */
push_nodes_for_insert(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int level)2075 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2076 					  struct btrfs_root *root,
2077 					  struct btrfs_path *path, int level)
2078 {
2079 	struct btrfs_fs_info *fs_info = root->fs_info;
2080 	struct extent_buffer *right = NULL;
2081 	struct extent_buffer *mid;
2082 	struct extent_buffer *left = NULL;
2083 	struct extent_buffer *parent = NULL;
2084 	int ret = 0;
2085 	int wret;
2086 	int pslot;
2087 	int orig_slot = path->slots[level];
2088 
2089 	if (level == 0)
2090 		return 1;
2091 
2092 	mid = path->nodes[level];
2093 	WARN_ON(btrfs_header_generation(mid) != trans->transid);
2094 
2095 	if (level < BTRFS_MAX_LEVEL - 1) {
2096 		parent = path->nodes[level + 1];
2097 		pslot = path->slots[level + 1];
2098 	}
2099 
2100 	if (!parent)
2101 		return 1;
2102 
2103 	left = btrfs_read_node_slot(parent, pslot - 1);
2104 	if (IS_ERR(left))
2105 		left = NULL;
2106 
2107 	/* first, try to make some room in the middle buffer */
2108 	if (left) {
2109 		u32 left_nr;
2110 
2111 		__btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
2112 		btrfs_set_lock_blocking_write(left);
2113 
2114 		left_nr = btrfs_header_nritems(left);
2115 		if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2116 			wret = 1;
2117 		} else {
2118 			ret = btrfs_cow_block(trans, root, left, parent,
2119 					      pslot - 1, &left,
2120 					      BTRFS_NESTING_LEFT_COW);
2121 			if (ret)
2122 				wret = 1;
2123 			else {
2124 				wret = push_node_left(trans, left, mid, 0);
2125 			}
2126 		}
2127 		if (wret < 0)
2128 			ret = wret;
2129 		if (wret == 0) {
2130 			struct btrfs_disk_key disk_key;
2131 			orig_slot += left_nr;
2132 			btrfs_node_key(mid, &disk_key, 0);
2133 			ret = tree_mod_log_insert_key(parent, pslot,
2134 					MOD_LOG_KEY_REPLACE, GFP_NOFS);
2135 			BUG_ON(ret < 0);
2136 			btrfs_set_node_key(parent, &disk_key, pslot);
2137 			btrfs_mark_buffer_dirty(parent);
2138 			if (btrfs_header_nritems(left) > orig_slot) {
2139 				path->nodes[level] = left;
2140 				path->slots[level + 1] -= 1;
2141 				path->slots[level] = orig_slot;
2142 				btrfs_tree_unlock(mid);
2143 				free_extent_buffer(mid);
2144 			} else {
2145 				orig_slot -=
2146 					btrfs_header_nritems(left);
2147 				path->slots[level] = orig_slot;
2148 				btrfs_tree_unlock(left);
2149 				free_extent_buffer(left);
2150 			}
2151 			return 0;
2152 		}
2153 		btrfs_tree_unlock(left);
2154 		free_extent_buffer(left);
2155 	}
2156 	right = btrfs_read_node_slot(parent, pslot + 1);
2157 	if (IS_ERR(right))
2158 		right = NULL;
2159 
2160 	/*
2161 	 * then try to empty the right most buffer into the middle
2162 	 */
2163 	if (right) {
2164 		u32 right_nr;
2165 
2166 		__btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
2167 		btrfs_set_lock_blocking_write(right);
2168 
2169 		right_nr = btrfs_header_nritems(right);
2170 		if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2171 			wret = 1;
2172 		} else {
2173 			ret = btrfs_cow_block(trans, root, right,
2174 					      parent, pslot + 1,
2175 					      &right, BTRFS_NESTING_RIGHT_COW);
2176 			if (ret)
2177 				wret = 1;
2178 			else {
2179 				wret = balance_node_right(trans, right, mid);
2180 			}
2181 		}
2182 		if (wret < 0)
2183 			ret = wret;
2184 		if (wret == 0) {
2185 			struct btrfs_disk_key disk_key;
2186 
2187 			btrfs_node_key(right, &disk_key, 0);
2188 			ret = tree_mod_log_insert_key(parent, pslot + 1,
2189 					MOD_LOG_KEY_REPLACE, GFP_NOFS);
2190 			BUG_ON(ret < 0);
2191 			btrfs_set_node_key(parent, &disk_key, pslot + 1);
2192 			btrfs_mark_buffer_dirty(parent);
2193 
2194 			if (btrfs_header_nritems(mid) <= orig_slot) {
2195 				path->nodes[level] = right;
2196 				path->slots[level + 1] += 1;
2197 				path->slots[level] = orig_slot -
2198 					btrfs_header_nritems(mid);
2199 				btrfs_tree_unlock(mid);
2200 				free_extent_buffer(mid);
2201 			} else {
2202 				btrfs_tree_unlock(right);
2203 				free_extent_buffer(right);
2204 			}
2205 			return 0;
2206 		}
2207 		btrfs_tree_unlock(right);
2208 		free_extent_buffer(right);
2209 	}
2210 	return 1;
2211 }
2212 
2213 /*
2214  * readahead one full node of leaves, finding things that are close
2215  * to the block in 'slot', and triggering ra on them.
2216  */
reada_for_search(struct btrfs_fs_info * fs_info,struct btrfs_path * path,int level,int slot,u64 objectid)2217 static void reada_for_search(struct btrfs_fs_info *fs_info,
2218 			     struct btrfs_path *path,
2219 			     int level, int slot, u64 objectid)
2220 {
2221 	struct extent_buffer *node;
2222 	struct btrfs_disk_key disk_key;
2223 	u32 nritems;
2224 	u64 search;
2225 	u64 target;
2226 	u64 nread = 0;
2227 	struct extent_buffer *eb;
2228 	u32 nr;
2229 	u32 blocksize;
2230 	u32 nscan = 0;
2231 
2232 	if (level != 1)
2233 		return;
2234 
2235 	if (!path->nodes[level])
2236 		return;
2237 
2238 	node = path->nodes[level];
2239 
2240 	search = btrfs_node_blockptr(node, slot);
2241 	blocksize = fs_info->nodesize;
2242 	eb = find_extent_buffer(fs_info, search);
2243 	if (eb) {
2244 		free_extent_buffer(eb);
2245 		return;
2246 	}
2247 
2248 	target = search;
2249 
2250 	nritems = btrfs_header_nritems(node);
2251 	nr = slot;
2252 
2253 	while (1) {
2254 		if (path->reada == READA_BACK) {
2255 			if (nr == 0)
2256 				break;
2257 			nr--;
2258 		} else if (path->reada == READA_FORWARD) {
2259 			nr++;
2260 			if (nr >= nritems)
2261 				break;
2262 		}
2263 		if (path->reada == READA_BACK && objectid) {
2264 			btrfs_node_key(node, &disk_key, nr);
2265 			if (btrfs_disk_key_objectid(&disk_key) != objectid)
2266 				break;
2267 		}
2268 		search = btrfs_node_blockptr(node, nr);
2269 		if ((search <= target && target - search <= 65536) ||
2270 		    (search > target && search - target <= 65536)) {
2271 			readahead_tree_block(fs_info, search);
2272 			nread += blocksize;
2273 		}
2274 		nscan++;
2275 		if ((nread > 65536 || nscan > 32))
2276 			break;
2277 	}
2278 }
2279 
reada_for_balance(struct btrfs_fs_info * fs_info,struct btrfs_path * path,int level)2280 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2281 				       struct btrfs_path *path, int level)
2282 {
2283 	int slot;
2284 	int nritems;
2285 	struct extent_buffer *parent;
2286 	struct extent_buffer *eb;
2287 	u64 gen;
2288 	u64 block1 = 0;
2289 	u64 block2 = 0;
2290 
2291 	parent = path->nodes[level + 1];
2292 	if (!parent)
2293 		return;
2294 
2295 	nritems = btrfs_header_nritems(parent);
2296 	slot = path->slots[level + 1];
2297 
2298 	if (slot > 0) {
2299 		block1 = btrfs_node_blockptr(parent, slot - 1);
2300 		gen = btrfs_node_ptr_generation(parent, slot - 1);
2301 		eb = find_extent_buffer(fs_info, block1);
2302 		/*
2303 		 * if we get -eagain from btrfs_buffer_uptodate, we
2304 		 * don't want to return eagain here.  That will loop
2305 		 * forever
2306 		 */
2307 		if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2308 			block1 = 0;
2309 		free_extent_buffer(eb);
2310 	}
2311 	if (slot + 1 < nritems) {
2312 		block2 = btrfs_node_blockptr(parent, slot + 1);
2313 		gen = btrfs_node_ptr_generation(parent, slot + 1);
2314 		eb = find_extent_buffer(fs_info, block2);
2315 		if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2316 			block2 = 0;
2317 		free_extent_buffer(eb);
2318 	}
2319 
2320 	if (block1)
2321 		readahead_tree_block(fs_info, block1);
2322 	if (block2)
2323 		readahead_tree_block(fs_info, block2);
2324 }
2325 
2326 
2327 /*
2328  * when we walk down the tree, it is usually safe to unlock the higher layers
2329  * in the tree.  The exceptions are when our path goes through slot 0, because
2330  * operations on the tree might require changing key pointers higher up in the
2331  * tree.
2332  *
2333  * callers might also have set path->keep_locks, which tells this code to keep
2334  * the lock if the path points to the last slot in the block.  This is part of
2335  * walking through the tree, and selecting the next slot in the higher block.
2336  *
2337  * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
2338  * if lowest_unlock is 1, level 0 won't be unlocked
2339  */
unlock_up(struct btrfs_path * path,int level,int lowest_unlock,int min_write_lock_level,int * write_lock_level)2340 static noinline void unlock_up(struct btrfs_path *path, int level,
2341 			       int lowest_unlock, int min_write_lock_level,
2342 			       int *write_lock_level)
2343 {
2344 	int i;
2345 	int skip_level = level;
2346 	int no_skips = 0;
2347 	struct extent_buffer *t;
2348 
2349 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2350 		if (!path->nodes[i])
2351 			break;
2352 		if (!path->locks[i])
2353 			break;
2354 		if (!no_skips && path->slots[i] == 0) {
2355 			skip_level = i + 1;
2356 			continue;
2357 		}
2358 		if (!no_skips && path->keep_locks) {
2359 			u32 nritems;
2360 			t = path->nodes[i];
2361 			nritems = btrfs_header_nritems(t);
2362 			if (nritems < 1 || path->slots[i] >= nritems - 1) {
2363 				skip_level = i + 1;
2364 				continue;
2365 			}
2366 		}
2367 		if (skip_level < i && i >= lowest_unlock)
2368 			no_skips = 1;
2369 
2370 		t = path->nodes[i];
2371 		if (i >= lowest_unlock && i > skip_level) {
2372 			btrfs_tree_unlock_rw(t, path->locks[i]);
2373 			path->locks[i] = 0;
2374 			if (write_lock_level &&
2375 			    i > min_write_lock_level &&
2376 			    i <= *write_lock_level) {
2377 				*write_lock_level = i - 1;
2378 			}
2379 		}
2380 	}
2381 }
2382 
2383 /*
2384  * helper function for btrfs_search_slot.  The goal is to find a block
2385  * in cache without setting the path to blocking.  If we find the block
2386  * we return zero and the path is unchanged.
2387  *
2388  * If we can't find the block, we set the path blocking and do some
2389  * reada.  -EAGAIN is returned and the search must be repeated.
2390  */
2391 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)2392 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2393 		      struct extent_buffer **eb_ret, int level, int slot,
2394 		      const struct btrfs_key *key)
2395 {
2396 	struct btrfs_fs_info *fs_info = root->fs_info;
2397 	u64 blocknr;
2398 	u64 gen;
2399 	struct extent_buffer *tmp;
2400 	struct btrfs_key first_key;
2401 	int ret;
2402 	int parent_level;
2403 
2404 	blocknr = btrfs_node_blockptr(*eb_ret, slot);
2405 	gen = btrfs_node_ptr_generation(*eb_ret, slot);
2406 	parent_level = btrfs_header_level(*eb_ret);
2407 	btrfs_node_key_to_cpu(*eb_ret, &first_key, slot);
2408 
2409 	tmp = find_extent_buffer(fs_info, blocknr);
2410 	if (tmp) {
2411 		/* first we do an atomic uptodate check */
2412 		if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2413 			/*
2414 			 * Do extra check for first_key, eb can be stale due to
2415 			 * being cached, read from scrub, or have multiple
2416 			 * parents (shared tree blocks).
2417 			 */
2418 			if (btrfs_verify_level_key(tmp,
2419 					parent_level - 1, &first_key, gen)) {
2420 				free_extent_buffer(tmp);
2421 				return -EUCLEAN;
2422 			}
2423 			*eb_ret = tmp;
2424 			return 0;
2425 		}
2426 
2427 		/* the pages were up to date, but we failed
2428 		 * the generation number check.  Do a full
2429 		 * read for the generation number that is correct.
2430 		 * We must do this without dropping locks so
2431 		 * we can trust our generation number
2432 		 */
2433 		btrfs_set_path_blocking(p);
2434 
2435 		/* now we're allowed to do a blocking uptodate check */
2436 		ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2437 		if (!ret) {
2438 			*eb_ret = tmp;
2439 			return 0;
2440 		}
2441 		free_extent_buffer(tmp);
2442 		btrfs_release_path(p);
2443 		return -EIO;
2444 	}
2445 
2446 	/*
2447 	 * reduce lock contention at high levels
2448 	 * of the btree by dropping locks before
2449 	 * we read.  Don't release the lock on the current
2450 	 * level because we need to walk this node to figure
2451 	 * out which blocks to read.
2452 	 */
2453 	btrfs_unlock_up_safe(p, level + 1);
2454 	btrfs_set_path_blocking(p);
2455 
2456 	if (p->reada != READA_NONE)
2457 		reada_for_search(fs_info, p, level, slot, key->objectid);
2458 
2459 	ret = -EAGAIN;
2460 	tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2461 			      &first_key);
2462 	if (!IS_ERR(tmp)) {
2463 		/*
2464 		 * If the read above didn't mark this buffer up to date,
2465 		 * it will never end up being up to date.  Set ret to EIO now
2466 		 * and give up so that our caller doesn't loop forever
2467 		 * on our EAGAINs.
2468 		 */
2469 		if (!extent_buffer_uptodate(tmp))
2470 			ret = -EIO;
2471 		free_extent_buffer(tmp);
2472 	} else {
2473 		ret = PTR_ERR(tmp);
2474 	}
2475 
2476 	btrfs_release_path(p);
2477 	return ret;
2478 }
2479 
2480 /*
2481  * helper function for btrfs_search_slot.  This does all of the checks
2482  * for node-level blocks and does any balancing required based on
2483  * the ins_len.
2484  *
2485  * If no extra work was required, zero is returned.  If we had to
2486  * drop the path, -EAGAIN is returned and btrfs_search_slot must
2487  * start over
2488  */
2489 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)2490 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2491 		       struct btrfs_root *root, struct btrfs_path *p,
2492 		       struct extent_buffer *b, int level, int ins_len,
2493 		       int *write_lock_level)
2494 {
2495 	struct btrfs_fs_info *fs_info = root->fs_info;
2496 	int ret;
2497 
2498 	if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2499 	    BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2500 		int sret;
2501 
2502 		if (*write_lock_level < level + 1) {
2503 			*write_lock_level = level + 1;
2504 			btrfs_release_path(p);
2505 			goto again;
2506 		}
2507 
2508 		btrfs_set_path_blocking(p);
2509 		reada_for_balance(fs_info, p, level);
2510 		sret = split_node(trans, root, p, level);
2511 
2512 		BUG_ON(sret > 0);
2513 		if (sret) {
2514 			ret = sret;
2515 			goto done;
2516 		}
2517 		b = p->nodes[level];
2518 	} else if (ins_len < 0 && btrfs_header_nritems(b) <
2519 		   BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2520 		int sret;
2521 
2522 		if (*write_lock_level < level + 1) {
2523 			*write_lock_level = level + 1;
2524 			btrfs_release_path(p);
2525 			goto again;
2526 		}
2527 
2528 		btrfs_set_path_blocking(p);
2529 		reada_for_balance(fs_info, p, level);
2530 		sret = balance_level(trans, root, p, level);
2531 
2532 		if (sret) {
2533 			ret = sret;
2534 			goto done;
2535 		}
2536 		b = p->nodes[level];
2537 		if (!b) {
2538 			btrfs_release_path(p);
2539 			goto again;
2540 		}
2541 		BUG_ON(btrfs_header_nritems(b) == 1);
2542 	}
2543 	return 0;
2544 
2545 again:
2546 	ret = -EAGAIN;
2547 done:
2548 	return ret;
2549 }
2550 
btrfs_find_item(struct btrfs_root * fs_root,struct btrfs_path * path,u64 iobjectid,u64 ioff,u8 key_type,struct btrfs_key * found_key)2551 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2552 		u64 iobjectid, u64 ioff, u8 key_type,
2553 		struct btrfs_key *found_key)
2554 {
2555 	int ret;
2556 	struct btrfs_key key;
2557 	struct extent_buffer *eb;
2558 
2559 	ASSERT(path);
2560 	ASSERT(found_key);
2561 
2562 	key.type = key_type;
2563 	key.objectid = iobjectid;
2564 	key.offset = ioff;
2565 
2566 	ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2567 	if (ret < 0)
2568 		return ret;
2569 
2570 	eb = path->nodes[0];
2571 	if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2572 		ret = btrfs_next_leaf(fs_root, path);
2573 		if (ret)
2574 			return ret;
2575 		eb = path->nodes[0];
2576 	}
2577 
2578 	btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2579 	if (found_key->type != key.type ||
2580 			found_key->objectid != key.objectid)
2581 		return 1;
2582 
2583 	return 0;
2584 }
2585 
btrfs_search_slot_get_root(struct btrfs_root * root,struct btrfs_path * p,int write_lock_level)2586 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2587 							struct btrfs_path *p,
2588 							int write_lock_level)
2589 {
2590 	struct btrfs_fs_info *fs_info = root->fs_info;
2591 	struct extent_buffer *b;
2592 	int root_lock = 0;
2593 	int level = 0;
2594 
2595 	if (p->search_commit_root) {
2596 		/*
2597 		 * The commit roots are read only so we always do read locks,
2598 		 * and we always must hold the commit_root_sem when doing
2599 		 * searches on them, the only exception is send where we don't
2600 		 * want to block transaction commits for a long time, so
2601 		 * we need to clone the commit root in order to avoid races
2602 		 * with transaction commits that create a snapshot of one of
2603 		 * the roots used by a send operation.
2604 		 */
2605 		if (p->need_commit_sem) {
2606 			down_read(&fs_info->commit_root_sem);
2607 			b = btrfs_clone_extent_buffer(root->commit_root);
2608 			up_read(&fs_info->commit_root_sem);
2609 			if (!b)
2610 				return ERR_PTR(-ENOMEM);
2611 
2612 		} else {
2613 			b = root->commit_root;
2614 			atomic_inc(&b->refs);
2615 		}
2616 		level = btrfs_header_level(b);
2617 		/*
2618 		 * Ensure that all callers have set skip_locking when
2619 		 * p->search_commit_root = 1.
2620 		 */
2621 		ASSERT(p->skip_locking == 1);
2622 
2623 		goto out;
2624 	}
2625 
2626 	if (p->skip_locking) {
2627 		b = btrfs_root_node(root);
2628 		level = btrfs_header_level(b);
2629 		goto out;
2630 	}
2631 
2632 	/* We try very hard to do read locks on the root */
2633 	root_lock = BTRFS_READ_LOCK;
2634 
2635 	/*
2636 	 * If the level is set to maximum, we can skip trying to get the read
2637 	 * lock.
2638 	 */
2639 	if (write_lock_level < BTRFS_MAX_LEVEL) {
2640 		/*
2641 		 * We don't know the level of the root node until we actually
2642 		 * have it read locked
2643 		 */
2644 		b = __btrfs_read_lock_root_node(root, p->recurse);
2645 		level = btrfs_header_level(b);
2646 		if (level > write_lock_level)
2647 			goto out;
2648 
2649 		/* Whoops, must trade for write lock */
2650 		btrfs_tree_read_unlock(b);
2651 		free_extent_buffer(b);
2652 	}
2653 
2654 	b = btrfs_lock_root_node(root);
2655 	root_lock = BTRFS_WRITE_LOCK;
2656 
2657 	/* The level might have changed, check again */
2658 	level = btrfs_header_level(b);
2659 
2660 out:
2661 	/*
2662 	 * The root may have failed to write out at some point, and thus is no
2663 	 * longer valid, return an error in this case.
2664 	 */
2665 	if (!extent_buffer_uptodate(b)) {
2666 		if (root_lock)
2667 			btrfs_tree_unlock_rw(b, root_lock);
2668 		free_extent_buffer(b);
2669 		return ERR_PTR(-EIO);
2670 	}
2671 
2672 	p->nodes[level] = b;
2673 	if (!p->skip_locking)
2674 		p->locks[level] = root_lock;
2675 	/*
2676 	 * Callers are responsible for dropping b's references.
2677 	 */
2678 	return b;
2679 }
2680 
2681 
2682 /*
2683  * btrfs_search_slot - look for a key in a tree and perform necessary
2684  * modifications to preserve tree invariants.
2685  *
2686  * @trans:	Handle of transaction, used when modifying the tree
2687  * @p:		Holds all btree nodes along the search path
2688  * @root:	The root node of the tree
2689  * @key:	The key we are looking for
2690  * @ins_len:	Indicates purpose of search, for inserts it is 1, for
2691  *		deletions it's -1. 0 for plain searches
2692  * @cow:	boolean should CoW operations be performed. Must always be 1
2693  *		when modifying the tree.
2694  *
2695  * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2696  * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2697  *
2698  * If @key is found, 0 is returned and you can find the item in the leaf level
2699  * of the path (level 0)
2700  *
2701  * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2702  * points to the slot where it should be inserted
2703  *
2704  * If an error is encountered while searching the tree a negative error number
2705  * is returned
2706  */
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)2707 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2708 		      const struct btrfs_key *key, struct btrfs_path *p,
2709 		      int ins_len, int cow)
2710 {
2711 	struct extent_buffer *b;
2712 	int slot;
2713 	int ret;
2714 	int err;
2715 	int level;
2716 	int lowest_unlock = 1;
2717 	/* everything at write_lock_level or lower must be write locked */
2718 	int write_lock_level = 0;
2719 	u8 lowest_level = 0;
2720 	int min_write_lock_level;
2721 	int prev_cmp;
2722 
2723 	lowest_level = p->lowest_level;
2724 	WARN_ON(lowest_level && ins_len > 0);
2725 	WARN_ON(p->nodes[0] != NULL);
2726 	BUG_ON(!cow && ins_len);
2727 
2728 	if (ins_len < 0) {
2729 		lowest_unlock = 2;
2730 
2731 		/* when we are removing items, we might have to go up to level
2732 		 * two as we update tree pointers  Make sure we keep write
2733 		 * for those levels as well
2734 		 */
2735 		write_lock_level = 2;
2736 	} else if (ins_len > 0) {
2737 		/*
2738 		 * for inserting items, make sure we have a write lock on
2739 		 * level 1 so we can update keys
2740 		 */
2741 		write_lock_level = 1;
2742 	}
2743 
2744 	if (!cow)
2745 		write_lock_level = -1;
2746 
2747 	if (cow && (p->keep_locks || p->lowest_level))
2748 		write_lock_level = BTRFS_MAX_LEVEL;
2749 
2750 	min_write_lock_level = write_lock_level;
2751 
2752 again:
2753 	prev_cmp = -1;
2754 	b = btrfs_search_slot_get_root(root, p, write_lock_level);
2755 	if (IS_ERR(b)) {
2756 		ret = PTR_ERR(b);
2757 		goto done;
2758 	}
2759 
2760 	while (b) {
2761 		int dec = 0;
2762 
2763 		level = btrfs_header_level(b);
2764 
2765 		if (cow) {
2766 			bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2767 
2768 			/*
2769 			 * if we don't really need to cow this block
2770 			 * then we don't want to set the path blocking,
2771 			 * so we test it here
2772 			 */
2773 			if (!should_cow_block(trans, root, b)) {
2774 				trans->dirty = true;
2775 				goto cow_done;
2776 			}
2777 
2778 			/*
2779 			 * must have write locks on this node and the
2780 			 * parent
2781 			 */
2782 			if (level > write_lock_level ||
2783 			    (level + 1 > write_lock_level &&
2784 			    level + 1 < BTRFS_MAX_LEVEL &&
2785 			    p->nodes[level + 1])) {
2786 				write_lock_level = level + 1;
2787 				btrfs_release_path(p);
2788 				goto again;
2789 			}
2790 
2791 			btrfs_set_path_blocking(p);
2792 			if (last_level)
2793 				err = btrfs_cow_block(trans, root, b, NULL, 0,
2794 						      &b,
2795 						      BTRFS_NESTING_COW);
2796 			else
2797 				err = btrfs_cow_block(trans, root, b,
2798 						      p->nodes[level + 1],
2799 						      p->slots[level + 1], &b,
2800 						      BTRFS_NESTING_COW);
2801 			if (err) {
2802 				ret = err;
2803 				goto done;
2804 			}
2805 		}
2806 cow_done:
2807 		p->nodes[level] = b;
2808 		/*
2809 		 * Leave path with blocking locks to avoid massive
2810 		 * lock context switch, this is made on purpose.
2811 		 */
2812 
2813 		/*
2814 		 * we have a lock on b and as long as we aren't changing
2815 		 * the tree, there is no way to for the items in b to change.
2816 		 * It is safe to drop the lock on our parent before we
2817 		 * go through the expensive btree search on b.
2818 		 *
2819 		 * If we're inserting or deleting (ins_len != 0), then we might
2820 		 * be changing slot zero, which may require changing the parent.
2821 		 * So, we can't drop the lock until after we know which slot
2822 		 * we're operating on.
2823 		 */
2824 		if (!ins_len && !p->keep_locks) {
2825 			int u = level + 1;
2826 
2827 			if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2828 				btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2829 				p->locks[u] = 0;
2830 			}
2831 		}
2832 
2833 		/*
2834 		 * If btrfs_bin_search returns an exact match (prev_cmp == 0)
2835 		 * we can safely assume the target key will always be in slot 0
2836 		 * on lower levels due to the invariants BTRFS' btree provides,
2837 		 * namely that a btrfs_key_ptr entry always points to the
2838 		 * lowest key in the child node, thus we can skip searching
2839 		 * lower levels
2840 		 */
2841 		if (prev_cmp == 0) {
2842 			slot = 0;
2843 			ret = 0;
2844 		} else {
2845 			ret = btrfs_bin_search(b, key, &slot);
2846 			prev_cmp = ret;
2847 			if (ret < 0)
2848 				goto done;
2849 		}
2850 
2851 		if (level == 0) {
2852 			p->slots[level] = slot;
2853 			if (ins_len > 0 &&
2854 			    btrfs_leaf_free_space(b) < ins_len) {
2855 				if (write_lock_level < 1) {
2856 					write_lock_level = 1;
2857 					btrfs_release_path(p);
2858 					goto again;
2859 				}
2860 
2861 				btrfs_set_path_blocking(p);
2862 				err = split_leaf(trans, root, key,
2863 						 p, ins_len, ret == 0);
2864 
2865 				BUG_ON(err > 0);
2866 				if (err) {
2867 					ret = err;
2868 					goto done;
2869 				}
2870 			}
2871 			if (!p->search_for_split)
2872 				unlock_up(p, level, lowest_unlock,
2873 					  min_write_lock_level, NULL);
2874 			goto done;
2875 		}
2876 		if (ret && slot > 0) {
2877 			dec = 1;
2878 			slot--;
2879 		}
2880 		p->slots[level] = slot;
2881 		err = setup_nodes_for_search(trans, root, p, b, level, ins_len,
2882 					     &write_lock_level);
2883 		if (err == -EAGAIN)
2884 			goto again;
2885 		if (err) {
2886 			ret = err;
2887 			goto done;
2888 		}
2889 		b = p->nodes[level];
2890 		slot = p->slots[level];
2891 
2892 		/*
2893 		 * Slot 0 is special, if we change the key we have to update
2894 		 * the parent pointer which means we must have a write lock on
2895 		 * the parent
2896 		 */
2897 		if (slot == 0 && ins_len && write_lock_level < level + 1) {
2898 			write_lock_level = level + 1;
2899 			btrfs_release_path(p);
2900 			goto again;
2901 		}
2902 
2903 		unlock_up(p, level, lowest_unlock, min_write_lock_level,
2904 			  &write_lock_level);
2905 
2906 		if (level == lowest_level) {
2907 			if (dec)
2908 				p->slots[level]++;
2909 			goto done;
2910 		}
2911 
2912 		err = read_block_for_search(root, p, &b, level, slot, key);
2913 		if (err == -EAGAIN)
2914 			goto again;
2915 		if (err) {
2916 			ret = err;
2917 			goto done;
2918 		}
2919 
2920 		if (!p->skip_locking) {
2921 			level = btrfs_header_level(b);
2922 			if (level <= write_lock_level) {
2923 				if (!btrfs_try_tree_write_lock(b)) {
2924 					btrfs_set_path_blocking(p);
2925 					btrfs_tree_lock(b);
2926 				}
2927 				p->locks[level] = BTRFS_WRITE_LOCK;
2928 			} else {
2929 				if (!btrfs_tree_read_lock_atomic(b)) {
2930 					btrfs_set_path_blocking(p);
2931 					__btrfs_tree_read_lock(b, BTRFS_NESTING_NORMAL,
2932 							       p->recurse);
2933 				}
2934 				p->locks[level] = BTRFS_READ_LOCK;
2935 			}
2936 			p->nodes[level] = b;
2937 		}
2938 	}
2939 	ret = 1;
2940 done:
2941 	/*
2942 	 * we don't really know what they plan on doing with the path
2943 	 * from here on, so for now just mark it as blocking
2944 	 */
2945 	if (!p->leave_spinning)
2946 		btrfs_set_path_blocking(p);
2947 	if (ret < 0 && !p->skip_release_on_error)
2948 		btrfs_release_path(p);
2949 	return ret;
2950 }
2951 
2952 /*
2953  * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2954  * current state of the tree together with the operations recorded in the tree
2955  * modification log to search for the key in a previous version of this tree, as
2956  * denoted by the time_seq parameter.
2957  *
2958  * Naturally, there is no support for insert, delete or cow operations.
2959  *
2960  * The resulting path and return value will be set up as if we called
2961  * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2962  */
btrfs_search_old_slot(struct btrfs_root * root,const struct btrfs_key * key,struct btrfs_path * p,u64 time_seq)2963 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2964 			  struct btrfs_path *p, u64 time_seq)
2965 {
2966 	struct btrfs_fs_info *fs_info = root->fs_info;
2967 	struct extent_buffer *b;
2968 	int slot;
2969 	int ret;
2970 	int err;
2971 	int level;
2972 	int lowest_unlock = 1;
2973 	u8 lowest_level = 0;
2974 
2975 	lowest_level = p->lowest_level;
2976 	WARN_ON(p->nodes[0] != NULL);
2977 
2978 	if (p->search_commit_root) {
2979 		BUG_ON(time_seq);
2980 		return btrfs_search_slot(NULL, root, key, p, 0, 0);
2981 	}
2982 
2983 again:
2984 	b = get_old_root(root, time_seq);
2985 	if (!b) {
2986 		ret = -EIO;
2987 		goto done;
2988 	}
2989 	level = btrfs_header_level(b);
2990 	p->locks[level] = BTRFS_READ_LOCK;
2991 
2992 	while (b) {
2993 		int dec = 0;
2994 
2995 		level = btrfs_header_level(b);
2996 		p->nodes[level] = b;
2997 
2998 		/*
2999 		 * we have a lock on b and as long as we aren't changing
3000 		 * the tree, there is no way to for the items in b to change.
3001 		 * It is safe to drop the lock on our parent before we
3002 		 * go through the expensive btree search on b.
3003 		 */
3004 		btrfs_unlock_up_safe(p, level + 1);
3005 
3006 		ret = btrfs_bin_search(b, key, &slot);
3007 		if (ret < 0)
3008 			goto done;
3009 
3010 		if (level == 0) {
3011 			p->slots[level] = slot;
3012 			unlock_up(p, level, lowest_unlock, 0, NULL);
3013 			goto done;
3014 		}
3015 
3016 		if (ret && slot > 0) {
3017 			dec = 1;
3018 			slot--;
3019 		}
3020 		p->slots[level] = slot;
3021 		unlock_up(p, level, lowest_unlock, 0, NULL);
3022 
3023 		if (level == lowest_level) {
3024 			if (dec)
3025 				p->slots[level]++;
3026 			goto done;
3027 		}
3028 
3029 		err = read_block_for_search(root, p, &b, level, slot, key);
3030 		if (err == -EAGAIN)
3031 			goto again;
3032 		if (err) {
3033 			ret = err;
3034 			goto done;
3035 		}
3036 
3037 		level = btrfs_header_level(b);
3038 		if (!btrfs_tree_read_lock_atomic(b)) {
3039 			btrfs_set_path_blocking(p);
3040 			btrfs_tree_read_lock(b);
3041 		}
3042 		b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3043 		if (!b) {
3044 			ret = -ENOMEM;
3045 			goto done;
3046 		}
3047 		p->locks[level] = BTRFS_READ_LOCK;
3048 		p->nodes[level] = b;
3049 	}
3050 	ret = 1;
3051 done:
3052 	if (!p->leave_spinning)
3053 		btrfs_set_path_blocking(p);
3054 	if (ret < 0)
3055 		btrfs_release_path(p);
3056 
3057 	return ret;
3058 }
3059 
3060 /*
3061  * helper to use instead of search slot if no exact match is needed but
3062  * instead the next or previous item should be returned.
3063  * When find_higher is true, the next higher item is returned, the next lower
3064  * otherwise.
3065  * When return_any and find_higher are both true, and no higher item is found,
3066  * return the next lower instead.
3067  * When return_any is true and find_higher is false, and no lower item is found,
3068  * return the next higher instead.
3069  * It returns 0 if any item is found, 1 if none is found (tree empty), and
3070  * < 0 on error
3071  */
btrfs_search_slot_for_read(struct btrfs_root * root,const struct btrfs_key * key,struct btrfs_path * p,int find_higher,int return_any)3072 int btrfs_search_slot_for_read(struct btrfs_root *root,
3073 			       const struct btrfs_key *key,
3074 			       struct btrfs_path *p, int find_higher,
3075 			       int return_any)
3076 {
3077 	int ret;
3078 	struct extent_buffer *leaf;
3079 
3080 again:
3081 	ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3082 	if (ret <= 0)
3083 		return ret;
3084 	/*
3085 	 * a return value of 1 means the path is at the position where the
3086 	 * item should be inserted. Normally this is the next bigger item,
3087 	 * but in case the previous item is the last in a leaf, path points
3088 	 * to the first free slot in the previous leaf, i.e. at an invalid
3089 	 * item.
3090 	 */
3091 	leaf = p->nodes[0];
3092 
3093 	if (find_higher) {
3094 		if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3095 			ret = btrfs_next_leaf(root, p);
3096 			if (ret <= 0)
3097 				return ret;
3098 			if (!return_any)
3099 				return 1;
3100 			/*
3101 			 * no higher item found, return the next
3102 			 * lower instead
3103 			 */
3104 			return_any = 0;
3105 			find_higher = 0;
3106 			btrfs_release_path(p);
3107 			goto again;
3108 		}
3109 	} else {
3110 		if (p->slots[0] == 0) {
3111 			ret = btrfs_prev_leaf(root, p);
3112 			if (ret < 0)
3113 				return ret;
3114 			if (!ret) {
3115 				leaf = p->nodes[0];
3116 				if (p->slots[0] == btrfs_header_nritems(leaf))
3117 					p->slots[0]--;
3118 				return 0;
3119 			}
3120 			if (!return_any)
3121 				return 1;
3122 			/*
3123 			 * no lower item found, return the next
3124 			 * higher instead
3125 			 */
3126 			return_any = 0;
3127 			find_higher = 1;
3128 			btrfs_release_path(p);
3129 			goto again;
3130 		} else {
3131 			--p->slots[0];
3132 		}
3133 	}
3134 	return 0;
3135 }
3136 
3137 /*
3138  * adjust the pointers going up the tree, starting at level
3139  * making sure the right key of each node is points to 'key'.
3140  * This is used after shifting pointers to the left, so it stops
3141  * fixing up pointers when a given leaf/node is not in slot 0 of the
3142  * higher levels
3143  *
3144  */
fixup_low_keys(struct btrfs_path * path,struct btrfs_disk_key * key,int level)3145 static void fixup_low_keys(struct btrfs_path *path,
3146 			   struct btrfs_disk_key *key, int level)
3147 {
3148 	int i;
3149 	struct extent_buffer *t;
3150 	int ret;
3151 
3152 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3153 		int tslot = path->slots[i];
3154 
3155 		if (!path->nodes[i])
3156 			break;
3157 		t = path->nodes[i];
3158 		ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3159 				GFP_ATOMIC);
3160 		BUG_ON(ret < 0);
3161 		btrfs_set_node_key(t, key, tslot);
3162 		btrfs_mark_buffer_dirty(path->nodes[i]);
3163 		if (tslot != 0)
3164 			break;
3165 	}
3166 }
3167 
3168 /*
3169  * update item key.
3170  *
3171  * This function isn't completely safe. It's the caller's responsibility
3172  * that the new key won't break the order
3173  */
btrfs_set_item_key_safe(struct btrfs_fs_info * fs_info,struct btrfs_path * path,const struct btrfs_key * new_key)3174 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3175 			     struct btrfs_path *path,
3176 			     const struct btrfs_key *new_key)
3177 {
3178 	struct btrfs_disk_key disk_key;
3179 	struct extent_buffer *eb;
3180 	int slot;
3181 
3182 	eb = path->nodes[0];
3183 	slot = path->slots[0];
3184 	if (slot > 0) {
3185 		btrfs_item_key(eb, &disk_key, slot - 1);
3186 		if (unlikely(comp_keys(&disk_key, new_key) >= 0)) {
3187 			btrfs_crit(fs_info,
3188 		"slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3189 				   slot, btrfs_disk_key_objectid(&disk_key),
3190 				   btrfs_disk_key_type(&disk_key),
3191 				   btrfs_disk_key_offset(&disk_key),
3192 				   new_key->objectid, new_key->type,
3193 				   new_key->offset);
3194 			btrfs_print_leaf(eb);
3195 			BUG();
3196 		}
3197 	}
3198 	if (slot < btrfs_header_nritems(eb) - 1) {
3199 		btrfs_item_key(eb, &disk_key, slot + 1);
3200 		if (unlikely(comp_keys(&disk_key, new_key) <= 0)) {
3201 			btrfs_crit(fs_info,
3202 		"slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3203 				   slot, btrfs_disk_key_objectid(&disk_key),
3204 				   btrfs_disk_key_type(&disk_key),
3205 				   btrfs_disk_key_offset(&disk_key),
3206 				   new_key->objectid, new_key->type,
3207 				   new_key->offset);
3208 			btrfs_print_leaf(eb);
3209 			BUG();
3210 		}
3211 	}
3212 
3213 	btrfs_cpu_key_to_disk(&disk_key, new_key);
3214 	btrfs_set_item_key(eb, &disk_key, slot);
3215 	btrfs_mark_buffer_dirty(eb);
3216 	if (slot == 0)
3217 		fixup_low_keys(path, &disk_key, 1);
3218 }
3219 
3220 /*
3221  * Check key order of two sibling extent buffers.
3222  *
3223  * Return true if something is wrong.
3224  * Return false if everything is fine.
3225  *
3226  * Tree-checker only works inside one tree block, thus the following
3227  * corruption can not be detected by tree-checker:
3228  *
3229  * Leaf @left			| Leaf @right
3230  * --------------------------------------------------------------
3231  * | 1 | 2 | 3 | 4 | 5 | f6 |   | 7 | 8 |
3232  *
3233  * Key f6 in leaf @left itself is valid, but not valid when the next
3234  * key in leaf @right is 7.
3235  * This can only be checked at tree block merge time.
3236  * And since tree checker has ensured all key order in each tree block
3237  * is correct, we only need to bother the last key of @left and the first
3238  * key of @right.
3239  */
check_sibling_keys(struct extent_buffer * left,struct extent_buffer * right)3240 static bool check_sibling_keys(struct extent_buffer *left,
3241 			       struct extent_buffer *right)
3242 {
3243 	struct btrfs_key left_last;
3244 	struct btrfs_key right_first;
3245 	int level = btrfs_header_level(left);
3246 	int nr_left = btrfs_header_nritems(left);
3247 	int nr_right = btrfs_header_nritems(right);
3248 
3249 	/* No key to check in one of the tree blocks */
3250 	if (!nr_left || !nr_right)
3251 		return false;
3252 
3253 	if (level) {
3254 		btrfs_node_key_to_cpu(left, &left_last, nr_left - 1);
3255 		btrfs_node_key_to_cpu(right, &right_first, 0);
3256 	} else {
3257 		btrfs_item_key_to_cpu(left, &left_last, nr_left - 1);
3258 		btrfs_item_key_to_cpu(right, &right_first, 0);
3259 	}
3260 
3261 	if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) {
3262 		btrfs_crit(left->fs_info,
3263 "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)",
3264 			   left_last.objectid, left_last.type,
3265 			   left_last.offset, right_first.objectid,
3266 			   right_first.type, right_first.offset);
3267 		return true;
3268 	}
3269 	return false;
3270 }
3271 
3272 /*
3273  * try to push data from one node into the next node left in the
3274  * tree.
3275  *
3276  * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3277  * error, and > 0 if there was no room in the left hand block.
3278  */
push_node_left(struct btrfs_trans_handle * trans,struct extent_buffer * dst,struct extent_buffer * src,int empty)3279 static int push_node_left(struct btrfs_trans_handle *trans,
3280 			  struct extent_buffer *dst,
3281 			  struct extent_buffer *src, int empty)
3282 {
3283 	struct btrfs_fs_info *fs_info = trans->fs_info;
3284 	int push_items = 0;
3285 	int src_nritems;
3286 	int dst_nritems;
3287 	int ret = 0;
3288 
3289 	src_nritems = btrfs_header_nritems(src);
3290 	dst_nritems = btrfs_header_nritems(dst);
3291 	push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3292 	WARN_ON(btrfs_header_generation(src) != trans->transid);
3293 	WARN_ON(btrfs_header_generation(dst) != trans->transid);
3294 
3295 	if (!empty && src_nritems <= 8)
3296 		return 1;
3297 
3298 	if (push_items <= 0)
3299 		return 1;
3300 
3301 	if (empty) {
3302 		push_items = min(src_nritems, push_items);
3303 		if (push_items < src_nritems) {
3304 			/* leave at least 8 pointers in the node if
3305 			 * we aren't going to empty it
3306 			 */
3307 			if (src_nritems - push_items < 8) {
3308 				if (push_items <= 8)
3309 					return 1;
3310 				push_items -= 8;
3311 			}
3312 		}
3313 	} else
3314 		push_items = min(src_nritems - 8, push_items);
3315 
3316 	/* dst is the left eb, src is the middle eb */
3317 	if (check_sibling_keys(dst, src)) {
3318 		ret = -EUCLEAN;
3319 		btrfs_abort_transaction(trans, ret);
3320 		return ret;
3321 	}
3322 	ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
3323 	if (ret) {
3324 		btrfs_abort_transaction(trans, ret);
3325 		return ret;
3326 	}
3327 	copy_extent_buffer(dst, src,
3328 			   btrfs_node_key_ptr_offset(dst_nritems),
3329 			   btrfs_node_key_ptr_offset(0),
3330 			   push_items * sizeof(struct btrfs_key_ptr));
3331 
3332 	if (push_items < src_nritems) {
3333 		/*
3334 		 * Don't call tree_mod_log_insert_move here, key removal was
3335 		 * already fully logged by tree_mod_log_eb_copy above.
3336 		 */
3337 		memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3338 				      btrfs_node_key_ptr_offset(push_items),
3339 				      (src_nritems - push_items) *
3340 				      sizeof(struct btrfs_key_ptr));
3341 	}
3342 	btrfs_set_header_nritems(src, src_nritems - push_items);
3343 	btrfs_set_header_nritems(dst, dst_nritems + push_items);
3344 	btrfs_mark_buffer_dirty(src);
3345 	btrfs_mark_buffer_dirty(dst);
3346 
3347 	return ret;
3348 }
3349 
3350 /*
3351  * try to push data from one node into the next node right in the
3352  * tree.
3353  *
3354  * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3355  * error, and > 0 if there was no room in the right hand block.
3356  *
3357  * this will  only push up to 1/2 the contents of the left node over
3358  */
balance_node_right(struct btrfs_trans_handle * trans,struct extent_buffer * dst,struct extent_buffer * src)3359 static int balance_node_right(struct btrfs_trans_handle *trans,
3360 			      struct extent_buffer *dst,
3361 			      struct extent_buffer *src)
3362 {
3363 	struct btrfs_fs_info *fs_info = trans->fs_info;
3364 	int push_items = 0;
3365 	int max_push;
3366 	int src_nritems;
3367 	int dst_nritems;
3368 	int ret = 0;
3369 
3370 	WARN_ON(btrfs_header_generation(src) != trans->transid);
3371 	WARN_ON(btrfs_header_generation(dst) != trans->transid);
3372 
3373 	src_nritems = btrfs_header_nritems(src);
3374 	dst_nritems = btrfs_header_nritems(dst);
3375 	push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3376 	if (push_items <= 0)
3377 		return 1;
3378 
3379 	if (src_nritems < 4)
3380 		return 1;
3381 
3382 	max_push = src_nritems / 2 + 1;
3383 	/* don't try to empty the node */
3384 	if (max_push >= src_nritems)
3385 		return 1;
3386 
3387 	if (max_push < push_items)
3388 		push_items = max_push;
3389 
3390 	/* dst is the right eb, src is the middle eb */
3391 	if (check_sibling_keys(src, dst)) {
3392 		ret = -EUCLEAN;
3393 		btrfs_abort_transaction(trans, ret);
3394 		return ret;
3395 	}
3396 	ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3397 	BUG_ON(ret < 0);
3398 	memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3399 				      btrfs_node_key_ptr_offset(0),
3400 				      (dst_nritems) *
3401 				      sizeof(struct btrfs_key_ptr));
3402 
3403 	ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
3404 				   push_items);
3405 	if (ret) {
3406 		btrfs_abort_transaction(trans, ret);
3407 		return ret;
3408 	}
3409 	copy_extent_buffer(dst, src,
3410 			   btrfs_node_key_ptr_offset(0),
3411 			   btrfs_node_key_ptr_offset(src_nritems - push_items),
3412 			   push_items * sizeof(struct btrfs_key_ptr));
3413 
3414 	btrfs_set_header_nritems(src, src_nritems - push_items);
3415 	btrfs_set_header_nritems(dst, dst_nritems + push_items);
3416 
3417 	btrfs_mark_buffer_dirty(src);
3418 	btrfs_mark_buffer_dirty(dst);
3419 
3420 	return ret;
3421 }
3422 
3423 /*
3424  * helper function to insert a new root level in the tree.
3425  * A new node is allocated, and a single item is inserted to
3426  * point to the existing root
3427  *
3428  * returns zero on success or < 0 on failure.
3429  */
insert_new_root(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int level)3430 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3431 			   struct btrfs_root *root,
3432 			   struct btrfs_path *path, int level)
3433 {
3434 	struct btrfs_fs_info *fs_info = root->fs_info;
3435 	u64 lower_gen;
3436 	struct extent_buffer *lower;
3437 	struct extent_buffer *c;
3438 	struct extent_buffer *old;
3439 	struct btrfs_disk_key lower_key;
3440 	int ret;
3441 
3442 	BUG_ON(path->nodes[level]);
3443 	BUG_ON(path->nodes[level-1] != root->node);
3444 
3445 	lower = path->nodes[level-1];
3446 	if (level == 1)
3447 		btrfs_item_key(lower, &lower_key, 0);
3448 	else
3449 		btrfs_node_key(lower, &lower_key, 0);
3450 
3451 	c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level,
3452 					 root->node->start, 0,
3453 					 BTRFS_NESTING_NEW_ROOT);
3454 	if (IS_ERR(c))
3455 		return PTR_ERR(c);
3456 
3457 	root_add_used(root, fs_info->nodesize);
3458 
3459 	btrfs_set_header_nritems(c, 1);
3460 	btrfs_set_node_key(c, &lower_key, 0);
3461 	btrfs_set_node_blockptr(c, 0, lower->start);
3462 	lower_gen = btrfs_header_generation(lower);
3463 	WARN_ON(lower_gen != trans->transid);
3464 
3465 	btrfs_set_node_ptr_generation(c, 0, lower_gen);
3466 
3467 	btrfs_mark_buffer_dirty(c);
3468 
3469 	old = root->node;
3470 	ret = tree_mod_log_insert_root(root->node, c, 0);
3471 	BUG_ON(ret < 0);
3472 	rcu_assign_pointer(root->node, c);
3473 
3474 	/* the super has an extra ref to root->node */
3475 	free_extent_buffer(old);
3476 
3477 	add_root_to_dirty_list(root);
3478 	atomic_inc(&c->refs);
3479 	path->nodes[level] = c;
3480 	path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3481 	path->slots[level] = 0;
3482 	return 0;
3483 }
3484 
3485 /*
3486  * worker function to insert a single pointer in a node.
3487  * the node should have enough room for the pointer already
3488  *
3489  * slot and level indicate where you want the key to go, and
3490  * blocknr is the block the key points to.
3491  */
insert_ptr(struct btrfs_trans_handle * trans,struct btrfs_path * path,struct btrfs_disk_key * key,u64 bytenr,int slot,int level)3492 static void insert_ptr(struct btrfs_trans_handle *trans,
3493 		       struct btrfs_path *path,
3494 		       struct btrfs_disk_key *key, u64 bytenr,
3495 		       int slot, int level)
3496 {
3497 	struct extent_buffer *lower;
3498 	int nritems;
3499 	int ret;
3500 
3501 	BUG_ON(!path->nodes[level]);
3502 	btrfs_assert_tree_locked(path->nodes[level]);
3503 	lower = path->nodes[level];
3504 	nritems = btrfs_header_nritems(lower);
3505 	BUG_ON(slot > nritems);
3506 	BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
3507 	if (slot != nritems) {
3508 		if (level) {
3509 			ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3510 					nritems - slot);
3511 			BUG_ON(ret < 0);
3512 		}
3513 		memmove_extent_buffer(lower,
3514 			      btrfs_node_key_ptr_offset(slot + 1),
3515 			      btrfs_node_key_ptr_offset(slot),
3516 			      (nritems - slot) * sizeof(struct btrfs_key_ptr));
3517 	}
3518 	if (level) {
3519 		ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3520 				GFP_NOFS);
3521 		BUG_ON(ret < 0);
3522 	}
3523 	btrfs_set_node_key(lower, key, slot);
3524 	btrfs_set_node_blockptr(lower, slot, bytenr);
3525 	WARN_ON(trans->transid == 0);
3526 	btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3527 	btrfs_set_header_nritems(lower, nritems + 1);
3528 	btrfs_mark_buffer_dirty(lower);
3529 }
3530 
3531 /*
3532  * split the node at the specified level in path in two.
3533  * The path is corrected to point to the appropriate node after the split
3534  *
3535  * Before splitting this tries to make some room in the node by pushing
3536  * left and right, if either one works, it returns right away.
3537  *
3538  * returns 0 on success and < 0 on failure
3539  */
split_node(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int level)3540 static noinline int split_node(struct btrfs_trans_handle *trans,
3541 			       struct btrfs_root *root,
3542 			       struct btrfs_path *path, int level)
3543 {
3544 	struct btrfs_fs_info *fs_info = root->fs_info;
3545 	struct extent_buffer *c;
3546 	struct extent_buffer *split;
3547 	struct btrfs_disk_key disk_key;
3548 	int mid;
3549 	int ret;
3550 	u32 c_nritems;
3551 
3552 	c = path->nodes[level];
3553 	WARN_ON(btrfs_header_generation(c) != trans->transid);
3554 	if (c == root->node) {
3555 		/*
3556 		 * trying to split the root, lets make a new one
3557 		 *
3558 		 * tree mod log: We don't log_removal old root in
3559 		 * insert_new_root, because that root buffer will be kept as a
3560 		 * normal node. We are going to log removal of half of the
3561 		 * elements below with tree_mod_log_eb_copy. We're holding a
3562 		 * tree lock on the buffer, which is why we cannot race with
3563 		 * other tree_mod_log users.
3564 		 */
3565 		ret = insert_new_root(trans, root, path, level + 1);
3566 		if (ret)
3567 			return ret;
3568 	} else {
3569 		ret = push_nodes_for_insert(trans, root, path, level);
3570 		c = path->nodes[level];
3571 		if (!ret && btrfs_header_nritems(c) <
3572 		    BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3573 			return 0;
3574 		if (ret < 0)
3575 			return ret;
3576 	}
3577 
3578 	c_nritems = btrfs_header_nritems(c);
3579 	mid = (c_nritems + 1) / 2;
3580 	btrfs_node_key(c, &disk_key, mid);
3581 
3582 	split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level,
3583 					     c->start, 0, BTRFS_NESTING_SPLIT);
3584 	if (IS_ERR(split))
3585 		return PTR_ERR(split);
3586 
3587 	root_add_used(root, fs_info->nodesize);
3588 	ASSERT(btrfs_header_level(c) == level);
3589 
3590 	ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
3591 	if (ret) {
3592 		btrfs_tree_unlock(split);
3593 		free_extent_buffer(split);
3594 		btrfs_abort_transaction(trans, ret);
3595 		return ret;
3596 	}
3597 	copy_extent_buffer(split, c,
3598 			   btrfs_node_key_ptr_offset(0),
3599 			   btrfs_node_key_ptr_offset(mid),
3600 			   (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3601 	btrfs_set_header_nritems(split, c_nritems - mid);
3602 	btrfs_set_header_nritems(c, mid);
3603 	ret = 0;
3604 
3605 	btrfs_mark_buffer_dirty(c);
3606 	btrfs_mark_buffer_dirty(split);
3607 
3608 	insert_ptr(trans, path, &disk_key, split->start,
3609 		   path->slots[level + 1] + 1, level + 1);
3610 
3611 	if (path->slots[level] >= mid) {
3612 		path->slots[level] -= mid;
3613 		btrfs_tree_unlock(c);
3614 		free_extent_buffer(c);
3615 		path->nodes[level] = split;
3616 		path->slots[level + 1] += 1;
3617 	} else {
3618 		btrfs_tree_unlock(split);
3619 		free_extent_buffer(split);
3620 	}
3621 	return ret;
3622 }
3623 
3624 /*
3625  * how many bytes are required to store the items in a leaf.  start
3626  * and nr indicate which items in the leaf to check.  This totals up the
3627  * space used both by the item structs and the item data
3628  */
leaf_space_used(struct extent_buffer * l,int start,int nr)3629 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3630 {
3631 	struct btrfs_item *start_item;
3632 	struct btrfs_item *end_item;
3633 	int data_len;
3634 	int nritems = btrfs_header_nritems(l);
3635 	int end = min(nritems, start + nr) - 1;
3636 
3637 	if (!nr)
3638 		return 0;
3639 	start_item = btrfs_item_nr(start);
3640 	end_item = btrfs_item_nr(end);
3641 	data_len = btrfs_item_offset(l, start_item) +
3642 		   btrfs_item_size(l, start_item);
3643 	data_len = data_len - btrfs_item_offset(l, end_item);
3644 	data_len += sizeof(struct btrfs_item) * nr;
3645 	WARN_ON(data_len < 0);
3646 	return data_len;
3647 }
3648 
3649 /*
3650  * The space between the end of the leaf items and
3651  * the start of the leaf data.  IOW, how much room
3652  * the leaf has left for both items and data
3653  */
btrfs_leaf_free_space(struct extent_buffer * leaf)3654 noinline int btrfs_leaf_free_space(struct extent_buffer *leaf)
3655 {
3656 	struct btrfs_fs_info *fs_info = leaf->fs_info;
3657 	int nritems = btrfs_header_nritems(leaf);
3658 	int ret;
3659 
3660 	ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3661 	if (ret < 0) {
3662 		btrfs_crit(fs_info,
3663 			   "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3664 			   ret,
3665 			   (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3666 			   leaf_space_used(leaf, 0, nritems), nritems);
3667 	}
3668 	return ret;
3669 }
3670 
3671 /*
3672  * min slot controls the lowest index we're willing to push to the
3673  * right.  We'll push up to and including min_slot, but no lower
3674  */
__push_leaf_right(struct btrfs_path * path,int data_size,int empty,struct extent_buffer * right,int free_space,u32 left_nritems,u32 min_slot)3675 static noinline int __push_leaf_right(struct btrfs_path *path,
3676 				      int data_size, int empty,
3677 				      struct extent_buffer *right,
3678 				      int free_space, u32 left_nritems,
3679 				      u32 min_slot)
3680 {
3681 	struct btrfs_fs_info *fs_info = right->fs_info;
3682 	struct extent_buffer *left = path->nodes[0];
3683 	struct extent_buffer *upper = path->nodes[1];
3684 	struct btrfs_map_token token;
3685 	struct btrfs_disk_key disk_key;
3686 	int slot;
3687 	u32 i;
3688 	int push_space = 0;
3689 	int push_items = 0;
3690 	struct btrfs_item *item;
3691 	u32 nr;
3692 	u32 right_nritems;
3693 	u32 data_end;
3694 	u32 this_item_size;
3695 
3696 	if (empty)
3697 		nr = 0;
3698 	else
3699 		nr = max_t(u32, 1, min_slot);
3700 
3701 	if (path->slots[0] >= left_nritems)
3702 		push_space += data_size;
3703 
3704 	slot = path->slots[1];
3705 	i = left_nritems - 1;
3706 	while (i >= nr) {
3707 		item = btrfs_item_nr(i);
3708 
3709 		if (!empty && push_items > 0) {
3710 			if (path->slots[0] > i)
3711 				break;
3712 			if (path->slots[0] == i) {
3713 				int space = btrfs_leaf_free_space(left);
3714 
3715 				if (space + push_space * 2 > free_space)
3716 					break;
3717 			}
3718 		}
3719 
3720 		if (path->slots[0] == i)
3721 			push_space += data_size;
3722 
3723 		this_item_size = btrfs_item_size(left, item);
3724 		if (this_item_size + sizeof(*item) + push_space > free_space)
3725 			break;
3726 
3727 		push_items++;
3728 		push_space += this_item_size + sizeof(*item);
3729 		if (i == 0)
3730 			break;
3731 		i--;
3732 	}
3733 
3734 	if (push_items == 0)
3735 		goto out_unlock;
3736 
3737 	WARN_ON(!empty && push_items == left_nritems);
3738 
3739 	/* push left to right */
3740 	right_nritems = btrfs_header_nritems(right);
3741 
3742 	push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3743 	push_space -= leaf_data_end(left);
3744 
3745 	/* make room in the right data area */
3746 	data_end = leaf_data_end(right);
3747 	memmove_extent_buffer(right,
3748 			      BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3749 			      BTRFS_LEAF_DATA_OFFSET + data_end,
3750 			      BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3751 
3752 	/* copy from the left data area */
3753 	copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3754 		     BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3755 		     BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left),
3756 		     push_space);
3757 
3758 	memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3759 			      btrfs_item_nr_offset(0),
3760 			      right_nritems * sizeof(struct btrfs_item));
3761 
3762 	/* copy the items from left to right */
3763 	copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3764 		   btrfs_item_nr_offset(left_nritems - push_items),
3765 		   push_items * sizeof(struct btrfs_item));
3766 
3767 	/* update the item pointers */
3768 	btrfs_init_map_token(&token, right);
3769 	right_nritems += push_items;
3770 	btrfs_set_header_nritems(right, right_nritems);
3771 	push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3772 	for (i = 0; i < right_nritems; i++) {
3773 		item = btrfs_item_nr(i);
3774 		push_space -= btrfs_token_item_size(&token, item);
3775 		btrfs_set_token_item_offset(&token, item, push_space);
3776 	}
3777 
3778 	left_nritems -= push_items;
3779 	btrfs_set_header_nritems(left, left_nritems);
3780 
3781 	if (left_nritems)
3782 		btrfs_mark_buffer_dirty(left);
3783 	else
3784 		btrfs_clean_tree_block(left);
3785 
3786 	btrfs_mark_buffer_dirty(right);
3787 
3788 	btrfs_item_key(right, &disk_key, 0);
3789 	btrfs_set_node_key(upper, &disk_key, slot + 1);
3790 	btrfs_mark_buffer_dirty(upper);
3791 
3792 	/* then fixup the leaf pointer in the path */
3793 	if (path->slots[0] >= left_nritems) {
3794 		path->slots[0] -= left_nritems;
3795 		if (btrfs_header_nritems(path->nodes[0]) == 0)
3796 			btrfs_clean_tree_block(path->nodes[0]);
3797 		btrfs_tree_unlock(path->nodes[0]);
3798 		free_extent_buffer(path->nodes[0]);
3799 		path->nodes[0] = right;
3800 		path->slots[1] += 1;
3801 	} else {
3802 		btrfs_tree_unlock(right);
3803 		free_extent_buffer(right);
3804 	}
3805 	return 0;
3806 
3807 out_unlock:
3808 	btrfs_tree_unlock(right);
3809 	free_extent_buffer(right);
3810 	return 1;
3811 }
3812 
3813 /*
3814  * push some data in the path leaf to the right, trying to free up at
3815  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3816  *
3817  * returns 1 if the push failed because the other node didn't have enough
3818  * room, 0 if everything worked out and < 0 if there were major errors.
3819  *
3820  * this will push starting from min_slot to the end of the leaf.  It won't
3821  * push any slot lower than min_slot
3822  */
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)3823 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3824 			   *root, struct btrfs_path *path,
3825 			   int min_data_size, int data_size,
3826 			   int empty, u32 min_slot)
3827 {
3828 	struct extent_buffer *left = path->nodes[0];
3829 	struct extent_buffer *right;
3830 	struct extent_buffer *upper;
3831 	int slot;
3832 	int free_space;
3833 	u32 left_nritems;
3834 	int ret;
3835 
3836 	if (!path->nodes[1])
3837 		return 1;
3838 
3839 	slot = path->slots[1];
3840 	upper = path->nodes[1];
3841 	if (slot >= btrfs_header_nritems(upper) - 1)
3842 		return 1;
3843 
3844 	btrfs_assert_tree_locked(path->nodes[1]);
3845 
3846 	right = btrfs_read_node_slot(upper, slot + 1);
3847 	/*
3848 	 * slot + 1 is not valid or we fail to read the right node,
3849 	 * no big deal, just return.
3850 	 */
3851 	if (IS_ERR(right))
3852 		return 1;
3853 
3854 	__btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
3855 	btrfs_set_lock_blocking_write(right);
3856 
3857 	free_space = btrfs_leaf_free_space(right);
3858 	if (free_space < data_size)
3859 		goto out_unlock;
3860 
3861 	/* cow and double check */
3862 	ret = btrfs_cow_block(trans, root, right, upper,
3863 			      slot + 1, &right, BTRFS_NESTING_RIGHT_COW);
3864 	if (ret)
3865 		goto out_unlock;
3866 
3867 	free_space = btrfs_leaf_free_space(right);
3868 	if (free_space < data_size)
3869 		goto out_unlock;
3870 
3871 	left_nritems = btrfs_header_nritems(left);
3872 	if (left_nritems == 0)
3873 		goto out_unlock;
3874 
3875 	if (check_sibling_keys(left, right)) {
3876 		ret = -EUCLEAN;
3877 		btrfs_abort_transaction(trans, ret);
3878 		btrfs_tree_unlock(right);
3879 		free_extent_buffer(right);
3880 		return ret;
3881 	}
3882 	if (path->slots[0] == left_nritems && !empty) {
3883 		/* Key greater than all keys in the leaf, right neighbor has
3884 		 * enough room for it and we're not emptying our leaf to delete
3885 		 * it, therefore use right neighbor to insert the new item and
3886 		 * no need to touch/dirty our left leaf. */
3887 		btrfs_tree_unlock(left);
3888 		free_extent_buffer(left);
3889 		path->nodes[0] = right;
3890 		path->slots[0] = 0;
3891 		path->slots[1]++;
3892 		return 0;
3893 	}
3894 
3895 	return __push_leaf_right(path, min_data_size, empty,
3896 				right, free_space, left_nritems, min_slot);
3897 out_unlock:
3898 	btrfs_tree_unlock(right);
3899 	free_extent_buffer(right);
3900 	return 1;
3901 }
3902 
3903 /*
3904  * push some data in the path leaf to the left, trying to free up at
3905  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
3906  *
3907  * max_slot can put a limit on how far into the leaf we'll push items.  The
3908  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us do all the
3909  * items
3910  */
__push_leaf_left(struct btrfs_path * path,int data_size,int empty,struct extent_buffer * left,int free_space,u32 right_nritems,u32 max_slot)3911 static noinline int __push_leaf_left(struct btrfs_path *path, int data_size,
3912 				     int empty, struct extent_buffer *left,
3913 				     int free_space, u32 right_nritems,
3914 				     u32 max_slot)
3915 {
3916 	struct btrfs_fs_info *fs_info = left->fs_info;
3917 	struct btrfs_disk_key disk_key;
3918 	struct extent_buffer *right = path->nodes[0];
3919 	int i;
3920 	int push_space = 0;
3921 	int push_items = 0;
3922 	struct btrfs_item *item;
3923 	u32 old_left_nritems;
3924 	u32 nr;
3925 	int ret = 0;
3926 	u32 this_item_size;
3927 	u32 old_left_item_size;
3928 	struct btrfs_map_token token;
3929 
3930 	if (empty)
3931 		nr = min(right_nritems, max_slot);
3932 	else
3933 		nr = min(right_nritems - 1, max_slot);
3934 
3935 	for (i = 0; i < nr; i++) {
3936 		item = btrfs_item_nr(i);
3937 
3938 		if (!empty && push_items > 0) {
3939 			if (path->slots[0] < i)
3940 				break;
3941 			if (path->slots[0] == i) {
3942 				int space = btrfs_leaf_free_space(right);
3943 
3944 				if (space + push_space * 2 > free_space)
3945 					break;
3946 			}
3947 		}
3948 
3949 		if (path->slots[0] == i)
3950 			push_space += data_size;
3951 
3952 		this_item_size = btrfs_item_size(right, item);
3953 		if (this_item_size + sizeof(*item) + push_space > free_space)
3954 			break;
3955 
3956 		push_items++;
3957 		push_space += this_item_size + sizeof(*item);
3958 	}
3959 
3960 	if (push_items == 0) {
3961 		ret = 1;
3962 		goto out;
3963 	}
3964 	WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3965 
3966 	/* push data from right to left */
3967 	copy_extent_buffer(left, right,
3968 			   btrfs_item_nr_offset(btrfs_header_nritems(left)),
3969 			   btrfs_item_nr_offset(0),
3970 			   push_items * sizeof(struct btrfs_item));
3971 
3972 	push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3973 		     btrfs_item_offset_nr(right, push_items - 1);
3974 
3975 	copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3976 		     leaf_data_end(left) - push_space,
3977 		     BTRFS_LEAF_DATA_OFFSET +
3978 		     btrfs_item_offset_nr(right, push_items - 1),
3979 		     push_space);
3980 	old_left_nritems = btrfs_header_nritems(left);
3981 	BUG_ON(old_left_nritems <= 0);
3982 
3983 	btrfs_init_map_token(&token, left);
3984 	old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3985 	for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3986 		u32 ioff;
3987 
3988 		item = btrfs_item_nr(i);
3989 
3990 		ioff = btrfs_token_item_offset(&token, item);
3991 		btrfs_set_token_item_offset(&token, item,
3992 		      ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size));
3993 	}
3994 	btrfs_set_header_nritems(left, old_left_nritems + push_items);
3995 
3996 	/* fixup right node */
3997 	if (push_items > right_nritems)
3998 		WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3999 		       right_nritems);
4000 
4001 	if (push_items < right_nritems) {
4002 		push_space = btrfs_item_offset_nr(right, push_items - 1) -
4003 						  leaf_data_end(right);
4004 		memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
4005 				      BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
4006 				      BTRFS_LEAF_DATA_OFFSET +
4007 				      leaf_data_end(right), push_space);
4008 
4009 		memmove_extent_buffer(right, btrfs_item_nr_offset(0),
4010 			      btrfs_item_nr_offset(push_items),
4011 			     (btrfs_header_nritems(right) - push_items) *
4012 			     sizeof(struct btrfs_item));
4013 	}
4014 
4015 	btrfs_init_map_token(&token, right);
4016 	right_nritems -= push_items;
4017 	btrfs_set_header_nritems(right, right_nritems);
4018 	push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
4019 	for (i = 0; i < right_nritems; i++) {
4020 		item = btrfs_item_nr(i);
4021 
4022 		push_space = push_space - btrfs_token_item_size(&token, item);
4023 		btrfs_set_token_item_offset(&token, item, push_space);
4024 	}
4025 
4026 	btrfs_mark_buffer_dirty(left);
4027 	if (right_nritems)
4028 		btrfs_mark_buffer_dirty(right);
4029 	else
4030 		btrfs_clean_tree_block(right);
4031 
4032 	btrfs_item_key(right, &disk_key, 0);
4033 	fixup_low_keys(path, &disk_key, 1);
4034 
4035 	/* then fixup the leaf pointer in the path */
4036 	if (path->slots[0] < push_items) {
4037 		path->slots[0] += old_left_nritems;
4038 		btrfs_tree_unlock(path->nodes[0]);
4039 		free_extent_buffer(path->nodes[0]);
4040 		path->nodes[0] = left;
4041 		path->slots[1] -= 1;
4042 	} else {
4043 		btrfs_tree_unlock(left);
4044 		free_extent_buffer(left);
4045 		path->slots[0] -= push_items;
4046 	}
4047 	BUG_ON(path->slots[0] < 0);
4048 	return ret;
4049 out:
4050 	btrfs_tree_unlock(left);
4051 	free_extent_buffer(left);
4052 	return ret;
4053 }
4054 
4055 /*
4056  * push some data in the path leaf to the left, trying to free up at
4057  * least data_size bytes.  returns zero if the push worked, nonzero otherwise
4058  *
4059  * max_slot can put a limit on how far into the leaf we'll push items.  The
4060  * item at 'max_slot' won't be touched.  Use (u32)-1 to make us push all the
4061  * items
4062  */
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)4063 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
4064 			  *root, struct btrfs_path *path, int min_data_size,
4065 			  int data_size, int empty, u32 max_slot)
4066 {
4067 	struct extent_buffer *right = path->nodes[0];
4068 	struct extent_buffer *left;
4069 	int slot;
4070 	int free_space;
4071 	u32 right_nritems;
4072 	int ret = 0;
4073 
4074 	slot = path->slots[1];
4075 	if (slot == 0)
4076 		return 1;
4077 	if (!path->nodes[1])
4078 		return 1;
4079 
4080 	right_nritems = btrfs_header_nritems(right);
4081 	if (right_nritems == 0)
4082 		return 1;
4083 
4084 	btrfs_assert_tree_locked(path->nodes[1]);
4085 
4086 	left = btrfs_read_node_slot(path->nodes[1], slot - 1);
4087 	/*
4088 	 * slot - 1 is not valid or we fail to read the left node,
4089 	 * no big deal, just return.
4090 	 */
4091 	if (IS_ERR(left))
4092 		return 1;
4093 
4094 	__btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
4095 	btrfs_set_lock_blocking_write(left);
4096 
4097 	free_space = btrfs_leaf_free_space(left);
4098 	if (free_space < data_size) {
4099 		ret = 1;
4100 		goto out;
4101 	}
4102 
4103 	/* cow and double check */
4104 	ret = btrfs_cow_block(trans, root, left,
4105 			      path->nodes[1], slot - 1, &left,
4106 			      BTRFS_NESTING_LEFT_COW);
4107 	if (ret) {
4108 		/* we hit -ENOSPC, but it isn't fatal here */
4109 		if (ret == -ENOSPC)
4110 			ret = 1;
4111 		goto out;
4112 	}
4113 
4114 	free_space = btrfs_leaf_free_space(left);
4115 	if (free_space < data_size) {
4116 		ret = 1;
4117 		goto out;
4118 	}
4119 
4120 	if (check_sibling_keys(left, right)) {
4121 		ret = -EUCLEAN;
4122 		btrfs_abort_transaction(trans, ret);
4123 		goto out;
4124 	}
4125 	return __push_leaf_left(path, min_data_size,
4126 			       empty, left, free_space, right_nritems,
4127 			       max_slot);
4128 out:
4129 	btrfs_tree_unlock(left);
4130 	free_extent_buffer(left);
4131 	return ret;
4132 }
4133 
4134 /*
4135  * split the path's leaf in two, making sure there is at least data_size
4136  * available for the resulting leaf level of the path.
4137  */
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)4138 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4139 				    struct btrfs_path *path,
4140 				    struct extent_buffer *l,
4141 				    struct extent_buffer *right,
4142 				    int slot, int mid, int nritems)
4143 {
4144 	struct btrfs_fs_info *fs_info = trans->fs_info;
4145 	int data_copy_size;
4146 	int rt_data_off;
4147 	int i;
4148 	struct btrfs_disk_key disk_key;
4149 	struct btrfs_map_token token;
4150 
4151 	nritems = nritems - mid;
4152 	btrfs_set_header_nritems(right, nritems);
4153 	data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l);
4154 
4155 	copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4156 			   btrfs_item_nr_offset(mid),
4157 			   nritems * sizeof(struct btrfs_item));
4158 
4159 	copy_extent_buffer(right, l,
4160 		     BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4161 		     data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4162 		     leaf_data_end(l), data_copy_size);
4163 
4164 	rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4165 
4166 	btrfs_init_map_token(&token, right);
4167 	for (i = 0; i < nritems; i++) {
4168 		struct btrfs_item *item = btrfs_item_nr(i);
4169 		u32 ioff;
4170 
4171 		ioff = btrfs_token_item_offset(&token, item);
4172 		btrfs_set_token_item_offset(&token, item, ioff + rt_data_off);
4173 	}
4174 
4175 	btrfs_set_header_nritems(l, mid);
4176 	btrfs_item_key(right, &disk_key, 0);
4177 	insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
4178 
4179 	btrfs_mark_buffer_dirty(right);
4180 	btrfs_mark_buffer_dirty(l);
4181 	BUG_ON(path->slots[0] != slot);
4182 
4183 	if (mid <= slot) {
4184 		btrfs_tree_unlock(path->nodes[0]);
4185 		free_extent_buffer(path->nodes[0]);
4186 		path->nodes[0] = right;
4187 		path->slots[0] -= mid;
4188 		path->slots[1] += 1;
4189 	} else {
4190 		btrfs_tree_unlock(right);
4191 		free_extent_buffer(right);
4192 	}
4193 
4194 	BUG_ON(path->slots[0] < 0);
4195 }
4196 
4197 /*
4198  * double splits happen when we need to insert a big item in the middle
4199  * of a leaf.  A double split can leave us with 3 mostly empty leaves:
4200  * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4201  *          A                 B                 C
4202  *
4203  * We avoid this by trying to push the items on either side of our target
4204  * into the adjacent leaves.  If all goes well we can avoid the double split
4205  * completely.
4206  */
push_for_double_split(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int data_size)4207 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4208 					  struct btrfs_root *root,
4209 					  struct btrfs_path *path,
4210 					  int data_size)
4211 {
4212 	int ret;
4213 	int progress = 0;
4214 	int slot;
4215 	u32 nritems;
4216 	int space_needed = data_size;
4217 
4218 	slot = path->slots[0];
4219 	if (slot < btrfs_header_nritems(path->nodes[0]))
4220 		space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4221 
4222 	/*
4223 	 * try to push all the items after our slot into the
4224 	 * right leaf
4225 	 */
4226 	ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4227 	if (ret < 0)
4228 		return ret;
4229 
4230 	if (ret == 0)
4231 		progress++;
4232 
4233 	nritems = btrfs_header_nritems(path->nodes[0]);
4234 	/*
4235 	 * our goal is to get our slot at the start or end of a leaf.  If
4236 	 * we've done so we're done
4237 	 */
4238 	if (path->slots[0] == 0 || path->slots[0] == nritems)
4239 		return 0;
4240 
4241 	if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4242 		return 0;
4243 
4244 	/* try to push all the items before our slot into the next leaf */
4245 	slot = path->slots[0];
4246 	space_needed = data_size;
4247 	if (slot > 0)
4248 		space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4249 	ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4250 	if (ret < 0)
4251 		return ret;
4252 
4253 	if (ret == 0)
4254 		progress++;
4255 
4256 	if (progress)
4257 		return 0;
4258 	return 1;
4259 }
4260 
4261 /*
4262  * split the path's leaf in two, making sure there is at least data_size
4263  * available for the resulting leaf level of the path.
4264  *
4265  * returns 0 if all went well and < 0 on failure.
4266  */
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)4267 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4268 			       struct btrfs_root *root,
4269 			       const struct btrfs_key *ins_key,
4270 			       struct btrfs_path *path, int data_size,
4271 			       int extend)
4272 {
4273 	struct btrfs_disk_key disk_key;
4274 	struct extent_buffer *l;
4275 	u32 nritems;
4276 	int mid;
4277 	int slot;
4278 	struct extent_buffer *right;
4279 	struct btrfs_fs_info *fs_info = root->fs_info;
4280 	int ret = 0;
4281 	int wret;
4282 	int split;
4283 	int num_doubles = 0;
4284 	int tried_avoid_double = 0;
4285 
4286 	l = path->nodes[0];
4287 	slot = path->slots[0];
4288 	if (extend && data_size + btrfs_item_size_nr(l, slot) +
4289 	    sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4290 		return -EOVERFLOW;
4291 
4292 	/* first try to make some room by pushing left and right */
4293 	if (data_size && path->nodes[1]) {
4294 		int space_needed = data_size;
4295 
4296 		if (slot < btrfs_header_nritems(l))
4297 			space_needed -= btrfs_leaf_free_space(l);
4298 
4299 		wret = push_leaf_right(trans, root, path, space_needed,
4300 				       space_needed, 0, 0);
4301 		if (wret < 0)
4302 			return wret;
4303 		if (wret) {
4304 			space_needed = data_size;
4305 			if (slot > 0)
4306 				space_needed -= btrfs_leaf_free_space(l);
4307 			wret = push_leaf_left(trans, root, path, space_needed,
4308 					      space_needed, 0, (u32)-1);
4309 			if (wret < 0)
4310 				return wret;
4311 		}
4312 		l = path->nodes[0];
4313 
4314 		/* did the pushes work? */
4315 		if (btrfs_leaf_free_space(l) >= data_size)
4316 			return 0;
4317 	}
4318 
4319 	if (!path->nodes[1]) {
4320 		ret = insert_new_root(trans, root, path, 1);
4321 		if (ret)
4322 			return ret;
4323 	}
4324 again:
4325 	split = 1;
4326 	l = path->nodes[0];
4327 	slot = path->slots[0];
4328 	nritems = btrfs_header_nritems(l);
4329 	mid = (nritems + 1) / 2;
4330 
4331 	if (mid <= slot) {
4332 		if (nritems == 1 ||
4333 		    leaf_space_used(l, mid, nritems - mid) + data_size >
4334 			BTRFS_LEAF_DATA_SIZE(fs_info)) {
4335 			if (slot >= nritems) {
4336 				split = 0;
4337 			} else {
4338 				mid = slot;
4339 				if (mid != nritems &&
4340 				    leaf_space_used(l, mid, nritems - mid) +
4341 				    data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4342 					if (data_size && !tried_avoid_double)
4343 						goto push_for_double;
4344 					split = 2;
4345 				}
4346 			}
4347 		}
4348 	} else {
4349 		if (leaf_space_used(l, 0, mid) + data_size >
4350 			BTRFS_LEAF_DATA_SIZE(fs_info)) {
4351 			if (!extend && data_size && slot == 0) {
4352 				split = 0;
4353 			} else if ((extend || !data_size) && slot == 0) {
4354 				mid = 1;
4355 			} else {
4356 				mid = slot;
4357 				if (mid != nritems &&
4358 				    leaf_space_used(l, mid, nritems - mid) +
4359 				    data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4360 					if (data_size && !tried_avoid_double)
4361 						goto push_for_double;
4362 					split = 2;
4363 				}
4364 			}
4365 		}
4366 	}
4367 
4368 	if (split == 0)
4369 		btrfs_cpu_key_to_disk(&disk_key, ins_key);
4370 	else
4371 		btrfs_item_key(l, &disk_key, mid);
4372 
4373 	/*
4374 	 * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double
4375 	 * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES
4376 	 * subclasses, which is 8 at the time of this patch, and we've maxed it
4377 	 * out.  In the future we could add a
4378 	 * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just
4379 	 * use BTRFS_NESTING_NEW_ROOT.
4380 	 */
4381 	right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0,
4382 					     l->start, 0, num_doubles ?
4383 					     BTRFS_NESTING_NEW_ROOT :
4384 					     BTRFS_NESTING_SPLIT);
4385 	if (IS_ERR(right))
4386 		return PTR_ERR(right);
4387 
4388 	root_add_used(root, fs_info->nodesize);
4389 
4390 	if (split == 0) {
4391 		if (mid <= slot) {
4392 			btrfs_set_header_nritems(right, 0);
4393 			insert_ptr(trans, path, &disk_key,
4394 				   right->start, path->slots[1] + 1, 1);
4395 			btrfs_tree_unlock(path->nodes[0]);
4396 			free_extent_buffer(path->nodes[0]);
4397 			path->nodes[0] = right;
4398 			path->slots[0] = 0;
4399 			path->slots[1] += 1;
4400 		} else {
4401 			btrfs_set_header_nritems(right, 0);
4402 			insert_ptr(trans, path, &disk_key,
4403 				   right->start, path->slots[1], 1);
4404 			btrfs_tree_unlock(path->nodes[0]);
4405 			free_extent_buffer(path->nodes[0]);
4406 			path->nodes[0] = right;
4407 			path->slots[0] = 0;
4408 			if (path->slots[1] == 0)
4409 				fixup_low_keys(path, &disk_key, 1);
4410 		}
4411 		/*
4412 		 * We create a new leaf 'right' for the required ins_len and
4413 		 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4414 		 * the content of ins_len to 'right'.
4415 		 */
4416 		return ret;
4417 	}
4418 
4419 	copy_for_split(trans, path, l, right, slot, mid, nritems);
4420 
4421 	if (split == 2) {
4422 		BUG_ON(num_doubles != 0);
4423 		num_doubles++;
4424 		goto again;
4425 	}
4426 
4427 	return 0;
4428 
4429 push_for_double:
4430 	push_for_double_split(trans, root, path, data_size);
4431 	tried_avoid_double = 1;
4432 	if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4433 		return 0;
4434 	goto again;
4435 }
4436 
setup_leaf_for_split(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int ins_len)4437 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4438 					 struct btrfs_root *root,
4439 					 struct btrfs_path *path, int ins_len)
4440 {
4441 	struct btrfs_key key;
4442 	struct extent_buffer *leaf;
4443 	struct btrfs_file_extent_item *fi;
4444 	u64 extent_len = 0;
4445 	u32 item_size;
4446 	int ret;
4447 
4448 	leaf = path->nodes[0];
4449 	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4450 
4451 	BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4452 	       key.type != BTRFS_EXTENT_CSUM_KEY);
4453 
4454 	if (btrfs_leaf_free_space(leaf) >= ins_len)
4455 		return 0;
4456 
4457 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4458 	if (key.type == BTRFS_EXTENT_DATA_KEY) {
4459 		fi = btrfs_item_ptr(leaf, path->slots[0],
4460 				    struct btrfs_file_extent_item);
4461 		extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4462 	}
4463 	btrfs_release_path(path);
4464 
4465 	path->keep_locks = 1;
4466 	path->search_for_split = 1;
4467 	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4468 	path->search_for_split = 0;
4469 	if (ret > 0)
4470 		ret = -EAGAIN;
4471 	if (ret < 0)
4472 		goto err;
4473 
4474 	ret = -EAGAIN;
4475 	leaf = path->nodes[0];
4476 	/* if our item isn't there, return now */
4477 	if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4478 		goto err;
4479 
4480 	/* the leaf has  changed, it now has room.  return now */
4481 	if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
4482 		goto err;
4483 
4484 	if (key.type == BTRFS_EXTENT_DATA_KEY) {
4485 		fi = btrfs_item_ptr(leaf, path->slots[0],
4486 				    struct btrfs_file_extent_item);
4487 		if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4488 			goto err;
4489 	}
4490 
4491 	btrfs_set_path_blocking(path);
4492 	ret = split_leaf(trans, root, &key, path, ins_len, 1);
4493 	if (ret)
4494 		goto err;
4495 
4496 	path->keep_locks = 0;
4497 	btrfs_unlock_up_safe(path, 1);
4498 	return 0;
4499 err:
4500 	path->keep_locks = 0;
4501 	return ret;
4502 }
4503 
split_item(struct btrfs_path * path,const struct btrfs_key * new_key,unsigned long split_offset)4504 static noinline int split_item(struct btrfs_path *path,
4505 			       const struct btrfs_key *new_key,
4506 			       unsigned long split_offset)
4507 {
4508 	struct extent_buffer *leaf;
4509 	struct btrfs_item *item;
4510 	struct btrfs_item *new_item;
4511 	int slot;
4512 	char *buf;
4513 	u32 nritems;
4514 	u32 item_size;
4515 	u32 orig_offset;
4516 	struct btrfs_disk_key disk_key;
4517 
4518 	leaf = path->nodes[0];
4519 	BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item));
4520 
4521 	btrfs_set_path_blocking(path);
4522 
4523 	item = btrfs_item_nr(path->slots[0]);
4524 	orig_offset = btrfs_item_offset(leaf, item);
4525 	item_size = btrfs_item_size(leaf, item);
4526 
4527 	buf = kmalloc(item_size, GFP_NOFS);
4528 	if (!buf)
4529 		return -ENOMEM;
4530 
4531 	read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4532 			    path->slots[0]), item_size);
4533 
4534 	slot = path->slots[0] + 1;
4535 	nritems = btrfs_header_nritems(leaf);
4536 	if (slot != nritems) {
4537 		/* shift the items */
4538 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4539 				btrfs_item_nr_offset(slot),
4540 				(nritems - slot) * sizeof(struct btrfs_item));
4541 	}
4542 
4543 	btrfs_cpu_key_to_disk(&disk_key, new_key);
4544 	btrfs_set_item_key(leaf, &disk_key, slot);
4545 
4546 	new_item = btrfs_item_nr(slot);
4547 
4548 	btrfs_set_item_offset(leaf, new_item, orig_offset);
4549 	btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4550 
4551 	btrfs_set_item_offset(leaf, item,
4552 			      orig_offset + item_size - split_offset);
4553 	btrfs_set_item_size(leaf, item, split_offset);
4554 
4555 	btrfs_set_header_nritems(leaf, nritems + 1);
4556 
4557 	/* write the data for the start of the original item */
4558 	write_extent_buffer(leaf, buf,
4559 			    btrfs_item_ptr_offset(leaf, path->slots[0]),
4560 			    split_offset);
4561 
4562 	/* write the data for the new item */
4563 	write_extent_buffer(leaf, buf + split_offset,
4564 			    btrfs_item_ptr_offset(leaf, slot),
4565 			    item_size - split_offset);
4566 	btrfs_mark_buffer_dirty(leaf);
4567 
4568 	BUG_ON(btrfs_leaf_free_space(leaf) < 0);
4569 	kfree(buf);
4570 	return 0;
4571 }
4572 
4573 /*
4574  * This function splits a single item into two items,
4575  * giving 'new_key' to the new item and splitting the
4576  * old one at split_offset (from the start of the item).
4577  *
4578  * The path may be released by this operation.  After
4579  * the split, the path is pointing to the old item.  The
4580  * new item is going to be in the same node as the old one.
4581  *
4582  * Note, the item being split must be smaller enough to live alone on
4583  * a tree block with room for one extra struct btrfs_item
4584  *
4585  * This allows us to split the item in place, keeping a lock on the
4586  * leaf the entire time.
4587  */
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)4588 int btrfs_split_item(struct btrfs_trans_handle *trans,
4589 		     struct btrfs_root *root,
4590 		     struct btrfs_path *path,
4591 		     const struct btrfs_key *new_key,
4592 		     unsigned long split_offset)
4593 {
4594 	int ret;
4595 	ret = setup_leaf_for_split(trans, root, path,
4596 				   sizeof(struct btrfs_item));
4597 	if (ret)
4598 		return ret;
4599 
4600 	ret = split_item(path, new_key, split_offset);
4601 	return ret;
4602 }
4603 
4604 /*
4605  * This function duplicate a item, giving 'new_key' to the new item.
4606  * It guarantees both items live in the same tree leaf and the new item
4607  * is contiguous with the original item.
4608  *
4609  * This allows us to split file extent in place, keeping a lock on the
4610  * leaf the entire time.
4611  */
btrfs_duplicate_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,const struct btrfs_key * new_key)4612 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4613 			 struct btrfs_root *root,
4614 			 struct btrfs_path *path,
4615 			 const struct btrfs_key *new_key)
4616 {
4617 	struct extent_buffer *leaf;
4618 	int ret;
4619 	u32 item_size;
4620 
4621 	leaf = path->nodes[0];
4622 	item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4623 	ret = setup_leaf_for_split(trans, root, path,
4624 				   item_size + sizeof(struct btrfs_item));
4625 	if (ret)
4626 		return ret;
4627 
4628 	path->slots[0]++;
4629 	setup_items_for_insert(root, path, new_key, &item_size, 1);
4630 	leaf = path->nodes[0];
4631 	memcpy_extent_buffer(leaf,
4632 			     btrfs_item_ptr_offset(leaf, path->slots[0]),
4633 			     btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4634 			     item_size);
4635 	return 0;
4636 }
4637 
4638 /*
4639  * make the item pointed to by the path smaller.  new_size indicates
4640  * how small to make it, and from_end tells us if we just chop bytes
4641  * off the end of the item or if we shift the item to chop bytes off
4642  * the front.
4643  */
btrfs_truncate_item(struct btrfs_path * path,u32 new_size,int from_end)4644 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end)
4645 {
4646 	int slot;
4647 	struct extent_buffer *leaf;
4648 	struct btrfs_item *item;
4649 	u32 nritems;
4650 	unsigned int data_end;
4651 	unsigned int old_data_start;
4652 	unsigned int old_size;
4653 	unsigned int size_diff;
4654 	int i;
4655 	struct btrfs_map_token token;
4656 
4657 	leaf = path->nodes[0];
4658 	slot = path->slots[0];
4659 
4660 	old_size = btrfs_item_size_nr(leaf, slot);
4661 	if (old_size == new_size)
4662 		return;
4663 
4664 	nritems = btrfs_header_nritems(leaf);
4665 	data_end = leaf_data_end(leaf);
4666 
4667 	old_data_start = btrfs_item_offset_nr(leaf, slot);
4668 
4669 	size_diff = old_size - new_size;
4670 
4671 	BUG_ON(slot < 0);
4672 	BUG_ON(slot >= nritems);
4673 
4674 	/*
4675 	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4676 	 */
4677 	/* first correct the data pointers */
4678 	btrfs_init_map_token(&token, leaf);
4679 	for (i = slot; i < nritems; i++) {
4680 		u32 ioff;
4681 		item = btrfs_item_nr(i);
4682 
4683 		ioff = btrfs_token_item_offset(&token, item);
4684 		btrfs_set_token_item_offset(&token, item, ioff + size_diff);
4685 	}
4686 
4687 	/* shift the data */
4688 	if (from_end) {
4689 		memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4690 			      data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4691 			      data_end, old_data_start + new_size - data_end);
4692 	} else {
4693 		struct btrfs_disk_key disk_key;
4694 		u64 offset;
4695 
4696 		btrfs_item_key(leaf, &disk_key, slot);
4697 
4698 		if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4699 			unsigned long ptr;
4700 			struct btrfs_file_extent_item *fi;
4701 
4702 			fi = btrfs_item_ptr(leaf, slot,
4703 					    struct btrfs_file_extent_item);
4704 			fi = (struct btrfs_file_extent_item *)(
4705 			     (unsigned long)fi - size_diff);
4706 
4707 			if (btrfs_file_extent_type(leaf, fi) ==
4708 			    BTRFS_FILE_EXTENT_INLINE) {
4709 				ptr = btrfs_item_ptr_offset(leaf, slot);
4710 				memmove_extent_buffer(leaf, ptr,
4711 				      (unsigned long)fi,
4712 				      BTRFS_FILE_EXTENT_INLINE_DATA_START);
4713 			}
4714 		}
4715 
4716 		memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4717 			      data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4718 			      data_end, old_data_start - data_end);
4719 
4720 		offset = btrfs_disk_key_offset(&disk_key);
4721 		btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4722 		btrfs_set_item_key(leaf, &disk_key, slot);
4723 		if (slot == 0)
4724 			fixup_low_keys(path, &disk_key, 1);
4725 	}
4726 
4727 	item = btrfs_item_nr(slot);
4728 	btrfs_set_item_size(leaf, item, new_size);
4729 	btrfs_mark_buffer_dirty(leaf);
4730 
4731 	if (btrfs_leaf_free_space(leaf) < 0) {
4732 		btrfs_print_leaf(leaf);
4733 		BUG();
4734 	}
4735 }
4736 
4737 /*
4738  * make the item pointed to by the path bigger, data_size is the added size.
4739  */
btrfs_extend_item(struct btrfs_path * path,u32 data_size)4740 void btrfs_extend_item(struct btrfs_path *path, u32 data_size)
4741 {
4742 	int slot;
4743 	struct extent_buffer *leaf;
4744 	struct btrfs_item *item;
4745 	u32 nritems;
4746 	unsigned int data_end;
4747 	unsigned int old_data;
4748 	unsigned int old_size;
4749 	int i;
4750 	struct btrfs_map_token token;
4751 
4752 	leaf = path->nodes[0];
4753 
4754 	nritems = btrfs_header_nritems(leaf);
4755 	data_end = leaf_data_end(leaf);
4756 
4757 	if (btrfs_leaf_free_space(leaf) < data_size) {
4758 		btrfs_print_leaf(leaf);
4759 		BUG();
4760 	}
4761 	slot = path->slots[0];
4762 	old_data = btrfs_item_end_nr(leaf, slot);
4763 
4764 	BUG_ON(slot < 0);
4765 	if (slot >= nritems) {
4766 		btrfs_print_leaf(leaf);
4767 		btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d",
4768 			   slot, nritems);
4769 		BUG();
4770 	}
4771 
4772 	/*
4773 	 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4774 	 */
4775 	/* first correct the data pointers */
4776 	btrfs_init_map_token(&token, leaf);
4777 	for (i = slot; i < nritems; i++) {
4778 		u32 ioff;
4779 		item = btrfs_item_nr(i);
4780 
4781 		ioff = btrfs_token_item_offset(&token, item);
4782 		btrfs_set_token_item_offset(&token, item, ioff - data_size);
4783 	}
4784 
4785 	/* shift the data */
4786 	memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4787 		      data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4788 		      data_end, old_data - data_end);
4789 
4790 	data_end = old_data;
4791 	old_size = btrfs_item_size_nr(leaf, slot);
4792 	item = btrfs_item_nr(slot);
4793 	btrfs_set_item_size(leaf, item, old_size + data_size);
4794 	btrfs_mark_buffer_dirty(leaf);
4795 
4796 	if (btrfs_leaf_free_space(leaf) < 0) {
4797 		btrfs_print_leaf(leaf);
4798 		BUG();
4799 	}
4800 }
4801 
4802 /**
4803  * setup_items_for_insert - Helper called before inserting one or more items
4804  * to a leaf. Main purpose is to save stack depth by doing the bulk of the work
4805  * in a function that doesn't call btrfs_search_slot
4806  *
4807  * @root:	root we are inserting items to
4808  * @path:	points to the leaf/slot where we are going to insert new items
4809  * @cpu_key:	array of keys for items to be inserted
4810  * @data_size:	size of the body of each item we are going to insert
4811  * @nr:		size of @cpu_key/@data_size arrays
4812  */
setup_items_for_insert(struct btrfs_root * root,struct btrfs_path * path,const struct btrfs_key * cpu_key,u32 * data_size,int nr)4813 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4814 			    const struct btrfs_key *cpu_key, u32 *data_size,
4815 			    int nr)
4816 {
4817 	struct btrfs_fs_info *fs_info = root->fs_info;
4818 	struct btrfs_item *item;
4819 	int i;
4820 	u32 nritems;
4821 	unsigned int data_end;
4822 	struct btrfs_disk_key disk_key;
4823 	struct extent_buffer *leaf;
4824 	int slot;
4825 	struct btrfs_map_token token;
4826 	u32 total_size;
4827 	u32 total_data = 0;
4828 
4829 	for (i = 0; i < nr; i++)
4830 		total_data += data_size[i];
4831 	total_size = total_data + (nr * sizeof(struct btrfs_item));
4832 
4833 	if (path->slots[0] == 0) {
4834 		btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4835 		fixup_low_keys(path, &disk_key, 1);
4836 	}
4837 	btrfs_unlock_up_safe(path, 1);
4838 
4839 	leaf = path->nodes[0];
4840 	slot = path->slots[0];
4841 
4842 	nritems = btrfs_header_nritems(leaf);
4843 	data_end = leaf_data_end(leaf);
4844 
4845 	if (btrfs_leaf_free_space(leaf) < total_size) {
4846 		btrfs_print_leaf(leaf);
4847 		btrfs_crit(fs_info, "not enough freespace need %u have %d",
4848 			   total_size, btrfs_leaf_free_space(leaf));
4849 		BUG();
4850 	}
4851 
4852 	btrfs_init_map_token(&token, leaf);
4853 	if (slot != nritems) {
4854 		unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4855 
4856 		if (old_data < data_end) {
4857 			btrfs_print_leaf(leaf);
4858 			btrfs_crit(fs_info,
4859 		"item at slot %d with data offset %u beyond data end of leaf %u",
4860 				   slot, old_data, data_end);
4861 			BUG();
4862 		}
4863 		/*
4864 		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4865 		 */
4866 		/* first correct the data pointers */
4867 		for (i = slot; i < nritems; i++) {
4868 			u32 ioff;
4869 
4870 			item = btrfs_item_nr(i);
4871 			ioff = btrfs_token_item_offset(&token, item);
4872 			btrfs_set_token_item_offset(&token, item,
4873 						    ioff - total_data);
4874 		}
4875 		/* shift the items */
4876 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4877 			      btrfs_item_nr_offset(slot),
4878 			      (nritems - slot) * sizeof(struct btrfs_item));
4879 
4880 		/* shift the data */
4881 		memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4882 			      data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4883 			      data_end, old_data - data_end);
4884 		data_end = old_data;
4885 	}
4886 
4887 	/* setup the item for the new data */
4888 	for (i = 0; i < nr; i++) {
4889 		btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4890 		btrfs_set_item_key(leaf, &disk_key, slot + i);
4891 		item = btrfs_item_nr(slot + i);
4892 		data_end -= data_size[i];
4893 		btrfs_set_token_item_offset(&token, item, data_end);
4894 		btrfs_set_token_item_size(&token, item, data_size[i]);
4895 	}
4896 
4897 	btrfs_set_header_nritems(leaf, nritems + nr);
4898 	btrfs_mark_buffer_dirty(leaf);
4899 
4900 	if (btrfs_leaf_free_space(leaf) < 0) {
4901 		btrfs_print_leaf(leaf);
4902 		BUG();
4903 	}
4904 }
4905 
4906 /*
4907  * Given a key and some data, insert items into the tree.
4908  * This does all the path init required, making room in the tree if needed.
4909  */
btrfs_insert_empty_items(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,const struct btrfs_key * cpu_key,u32 * data_size,int nr)4910 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4911 			    struct btrfs_root *root,
4912 			    struct btrfs_path *path,
4913 			    const struct btrfs_key *cpu_key, u32 *data_size,
4914 			    int nr)
4915 {
4916 	int ret = 0;
4917 	int slot;
4918 	int i;
4919 	u32 total_size = 0;
4920 	u32 total_data = 0;
4921 
4922 	for (i = 0; i < nr; i++)
4923 		total_data += data_size[i];
4924 
4925 	total_size = total_data + (nr * sizeof(struct btrfs_item));
4926 	ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4927 	if (ret == 0)
4928 		return -EEXIST;
4929 	if (ret < 0)
4930 		return ret;
4931 
4932 	slot = path->slots[0];
4933 	BUG_ON(slot < 0);
4934 
4935 	setup_items_for_insert(root, path, cpu_key, data_size, nr);
4936 	return 0;
4937 }
4938 
4939 /*
4940  * Given a key and some data, insert an item into the tree.
4941  * This does all the path init required, making room in the tree if needed.
4942  */
btrfs_insert_item(struct btrfs_trans_handle * trans,struct btrfs_root * root,const struct btrfs_key * cpu_key,void * data,u32 data_size)4943 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4944 		      const struct btrfs_key *cpu_key, void *data,
4945 		      u32 data_size)
4946 {
4947 	int ret = 0;
4948 	struct btrfs_path *path;
4949 	struct extent_buffer *leaf;
4950 	unsigned long ptr;
4951 
4952 	path = btrfs_alloc_path();
4953 	if (!path)
4954 		return -ENOMEM;
4955 	ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4956 	if (!ret) {
4957 		leaf = path->nodes[0];
4958 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4959 		write_extent_buffer(leaf, data, ptr, data_size);
4960 		btrfs_mark_buffer_dirty(leaf);
4961 	}
4962 	btrfs_free_path(path);
4963 	return ret;
4964 }
4965 
4966 /*
4967  * delete the pointer from a given node.
4968  *
4969  * the tree should have been previously balanced so the deletion does not
4970  * empty a node.
4971  */
del_ptr(struct btrfs_root * root,struct btrfs_path * path,int level,int slot)4972 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4973 		    int level, int slot)
4974 {
4975 	struct extent_buffer *parent = path->nodes[level];
4976 	u32 nritems;
4977 	int ret;
4978 
4979 	nritems = btrfs_header_nritems(parent);
4980 	if (slot != nritems - 1) {
4981 		if (level) {
4982 			ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4983 					nritems - slot - 1);
4984 			BUG_ON(ret < 0);
4985 		}
4986 		memmove_extent_buffer(parent,
4987 			      btrfs_node_key_ptr_offset(slot),
4988 			      btrfs_node_key_ptr_offset(slot + 1),
4989 			      sizeof(struct btrfs_key_ptr) *
4990 			      (nritems - slot - 1));
4991 	} else if (level) {
4992 		ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4993 				GFP_NOFS);
4994 		BUG_ON(ret < 0);
4995 	}
4996 
4997 	nritems--;
4998 	btrfs_set_header_nritems(parent, nritems);
4999 	if (nritems == 0 && parent == root->node) {
5000 		BUG_ON(btrfs_header_level(root->node) != 1);
5001 		/* just turn the root into a leaf and break */
5002 		btrfs_set_header_level(root->node, 0);
5003 	} else if (slot == 0) {
5004 		struct btrfs_disk_key disk_key;
5005 
5006 		btrfs_node_key(parent, &disk_key, 0);
5007 		fixup_low_keys(path, &disk_key, level + 1);
5008 	}
5009 	btrfs_mark_buffer_dirty(parent);
5010 }
5011 
5012 /*
5013  * a helper function to delete the leaf pointed to by path->slots[1] and
5014  * path->nodes[1].
5015  *
5016  * This deletes the pointer in path->nodes[1] and frees the leaf
5017  * block extent.  zero is returned if it all worked out, < 0 otherwise.
5018  *
5019  * The path must have already been setup for deleting the leaf, including
5020  * all the proper balancing.  path->nodes[1] must be locked.
5021  */
btrfs_del_leaf(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,struct extent_buffer * leaf)5022 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
5023 				    struct btrfs_root *root,
5024 				    struct btrfs_path *path,
5025 				    struct extent_buffer *leaf)
5026 {
5027 	WARN_ON(btrfs_header_generation(leaf) != trans->transid);
5028 	del_ptr(root, path, 1, path->slots[1]);
5029 
5030 	/*
5031 	 * btrfs_free_extent is expensive, we want to make sure we
5032 	 * aren't holding any locks when we call it
5033 	 */
5034 	btrfs_unlock_up_safe(path, 0);
5035 
5036 	root_sub_used(root, leaf->len);
5037 
5038 	atomic_inc(&leaf->refs);
5039 	btrfs_free_tree_block(trans, root, leaf, 0, 1);
5040 	free_extent_buffer_stale(leaf);
5041 }
5042 /*
5043  * delete the item at the leaf level in path.  If that empties
5044  * the leaf, remove it from the tree
5045  */
btrfs_del_items(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_path * path,int slot,int nr)5046 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5047 		    struct btrfs_path *path, int slot, int nr)
5048 {
5049 	struct btrfs_fs_info *fs_info = root->fs_info;
5050 	struct extent_buffer *leaf;
5051 	struct btrfs_item *item;
5052 	u32 last_off;
5053 	u32 dsize = 0;
5054 	int ret = 0;
5055 	int wret;
5056 	int i;
5057 	u32 nritems;
5058 
5059 	leaf = path->nodes[0];
5060 	last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
5061 
5062 	for (i = 0; i < nr; i++)
5063 		dsize += btrfs_item_size_nr(leaf, slot + i);
5064 
5065 	nritems = btrfs_header_nritems(leaf);
5066 
5067 	if (slot + nr != nritems) {
5068 		int data_end = leaf_data_end(leaf);
5069 		struct btrfs_map_token token;
5070 
5071 		memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
5072 			      data_end + dsize,
5073 			      BTRFS_LEAF_DATA_OFFSET + data_end,
5074 			      last_off - data_end);
5075 
5076 		btrfs_init_map_token(&token, leaf);
5077 		for (i = slot + nr; i < nritems; i++) {
5078 			u32 ioff;
5079 
5080 			item = btrfs_item_nr(i);
5081 			ioff = btrfs_token_item_offset(&token, item);
5082 			btrfs_set_token_item_offset(&token, item, ioff + dsize);
5083 		}
5084 
5085 		memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
5086 			      btrfs_item_nr_offset(slot + nr),
5087 			      sizeof(struct btrfs_item) *
5088 			      (nritems - slot - nr));
5089 	}
5090 	btrfs_set_header_nritems(leaf, nritems - nr);
5091 	nritems -= nr;
5092 
5093 	/* delete the leaf if we've emptied it */
5094 	if (nritems == 0) {
5095 		if (leaf == root->node) {
5096 			btrfs_set_header_level(leaf, 0);
5097 		} else {
5098 			btrfs_set_path_blocking(path);
5099 			btrfs_clean_tree_block(leaf);
5100 			btrfs_del_leaf(trans, root, path, leaf);
5101 		}
5102 	} else {
5103 		int used = leaf_space_used(leaf, 0, nritems);
5104 		if (slot == 0) {
5105 			struct btrfs_disk_key disk_key;
5106 
5107 			btrfs_item_key(leaf, &disk_key, 0);
5108 			fixup_low_keys(path, &disk_key, 1);
5109 		}
5110 
5111 		/* delete the leaf if it is mostly empty */
5112 		if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
5113 			/* push_leaf_left fixes the path.
5114 			 * make sure the path still points to our leaf
5115 			 * for possible call to del_ptr below
5116 			 */
5117 			slot = path->slots[1];
5118 			atomic_inc(&leaf->refs);
5119 
5120 			btrfs_set_path_blocking(path);
5121 			wret = push_leaf_left(trans, root, path, 1, 1,
5122 					      1, (u32)-1);
5123 			if (wret < 0 && wret != -ENOSPC)
5124 				ret = wret;
5125 
5126 			if (path->nodes[0] == leaf &&
5127 			    btrfs_header_nritems(leaf)) {
5128 				wret = push_leaf_right(trans, root, path, 1,
5129 						       1, 1, 0);
5130 				if (wret < 0 && wret != -ENOSPC)
5131 					ret = wret;
5132 			}
5133 
5134 			if (btrfs_header_nritems(leaf) == 0) {
5135 				path->slots[1] = slot;
5136 				btrfs_del_leaf(trans, root, path, leaf);
5137 				free_extent_buffer(leaf);
5138 				ret = 0;
5139 			} else {
5140 				/* if we're still in the path, make sure
5141 				 * we're dirty.  Otherwise, one of the
5142 				 * push_leaf functions must have already
5143 				 * dirtied this buffer
5144 				 */
5145 				if (path->nodes[0] == leaf)
5146 					btrfs_mark_buffer_dirty(leaf);
5147 				free_extent_buffer(leaf);
5148 			}
5149 		} else {
5150 			btrfs_mark_buffer_dirty(leaf);
5151 		}
5152 	}
5153 	return ret;
5154 }
5155 
5156 /*
5157  * search the tree again to find a leaf with lesser keys
5158  * returns 0 if it found something or 1 if there are no lesser leaves.
5159  * returns < 0 on io errors.
5160  *
5161  * This may release the path, and so you may lose any locks held at the
5162  * time you call it.
5163  */
btrfs_prev_leaf(struct btrfs_root * root,struct btrfs_path * path)5164 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5165 {
5166 	struct btrfs_key key;
5167 	struct btrfs_key orig_key;
5168 	struct btrfs_disk_key found_key;
5169 	int ret;
5170 
5171 	btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5172 	orig_key = key;
5173 
5174 	if (key.offset > 0) {
5175 		key.offset--;
5176 	} else if (key.type > 0) {
5177 		key.type--;
5178 		key.offset = (u64)-1;
5179 	} else if (key.objectid > 0) {
5180 		key.objectid--;
5181 		key.type = (u8)-1;
5182 		key.offset = (u64)-1;
5183 	} else {
5184 		return 1;
5185 	}
5186 
5187 	btrfs_release_path(path);
5188 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5189 	if (ret <= 0)
5190 		return ret;
5191 
5192 	/*
5193 	 * Previous key not found. Even if we were at slot 0 of the leaf we had
5194 	 * before releasing the path and calling btrfs_search_slot(), we now may
5195 	 * be in a slot pointing to the same original key - this can happen if
5196 	 * after we released the path, one of more items were moved from a
5197 	 * sibling leaf into the front of the leaf we had due to an insertion
5198 	 * (see push_leaf_right()).
5199 	 * If we hit this case and our slot is > 0 and just decrement the slot
5200 	 * so that the caller does not process the same key again, which may or
5201 	 * may not break the caller, depending on its logic.
5202 	 */
5203 	if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5204 		btrfs_item_key(path->nodes[0], &found_key, path->slots[0]);
5205 		ret = comp_keys(&found_key, &orig_key);
5206 		if (ret == 0) {
5207 			if (path->slots[0] > 0) {
5208 				path->slots[0]--;
5209 				return 0;
5210 			}
5211 			/*
5212 			 * At slot 0, same key as before, it means orig_key is
5213 			 * the lowest, leftmost, key in the tree. We're done.
5214 			 */
5215 			return 1;
5216 		}
5217 	}
5218 
5219 	btrfs_item_key(path->nodes[0], &found_key, 0);
5220 	ret = comp_keys(&found_key, &key);
5221 	/*
5222 	 * We might have had an item with the previous key in the tree right
5223 	 * before we released our path. And after we released our path, that
5224 	 * item might have been pushed to the first slot (0) of the leaf we
5225 	 * were holding due to a tree balance. Alternatively, an item with the
5226 	 * previous key can exist as the only element of a leaf (big fat item).
5227 	 * Therefore account for these 2 cases, so that our callers (like
5228 	 * btrfs_previous_item) don't miss an existing item with a key matching
5229 	 * the previous key we computed above.
5230 	 */
5231 	if (ret <= 0)
5232 		return 0;
5233 	return 1;
5234 }
5235 
5236 /*
5237  * A helper function to walk down the tree starting at min_key, and looking
5238  * for nodes or leaves that are have a minimum transaction id.
5239  * This is used by the btree defrag code, and tree logging
5240  *
5241  * This does not cow, but it does stuff the starting key it finds back
5242  * into min_key, so you can call btrfs_search_slot with cow=1 on the
5243  * key and get a writable path.
5244  *
5245  * This honors path->lowest_level to prevent descent past a given level
5246  * of the tree.
5247  *
5248  * min_trans indicates the oldest transaction that you are interested
5249  * in walking through.  Any nodes or leaves older than min_trans are
5250  * skipped over (without reading them).
5251  *
5252  * returns zero if something useful was found, < 0 on error and 1 if there
5253  * was nothing in the tree that matched the search criteria.
5254  */
btrfs_search_forward(struct btrfs_root * root,struct btrfs_key * min_key,struct btrfs_path * path,u64 min_trans)5255 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5256 			 struct btrfs_path *path,
5257 			 u64 min_trans)
5258 {
5259 	struct extent_buffer *cur;
5260 	struct btrfs_key found_key;
5261 	int slot;
5262 	int sret;
5263 	u32 nritems;
5264 	int level;
5265 	int ret = 1;
5266 	int keep_locks = path->keep_locks;
5267 
5268 	path->keep_locks = 1;
5269 again:
5270 	cur = btrfs_read_lock_root_node(root);
5271 	level = btrfs_header_level(cur);
5272 	WARN_ON(path->nodes[level]);
5273 	path->nodes[level] = cur;
5274 	path->locks[level] = BTRFS_READ_LOCK;
5275 
5276 	if (btrfs_header_generation(cur) < min_trans) {
5277 		ret = 1;
5278 		goto out;
5279 	}
5280 	while (1) {
5281 		nritems = btrfs_header_nritems(cur);
5282 		level = btrfs_header_level(cur);
5283 		sret = btrfs_bin_search(cur, min_key, &slot);
5284 		if (sret < 0) {
5285 			ret = sret;
5286 			goto out;
5287 		}
5288 
5289 		/* at the lowest level, we're done, setup the path and exit */
5290 		if (level == path->lowest_level) {
5291 			if (slot >= nritems)
5292 				goto find_next_key;
5293 			ret = 0;
5294 			path->slots[level] = slot;
5295 			btrfs_item_key_to_cpu(cur, &found_key, slot);
5296 			goto out;
5297 		}
5298 		if (sret && slot > 0)
5299 			slot--;
5300 		/*
5301 		 * check this node pointer against the min_trans parameters.
5302 		 * If it is too old, skip to the next one.
5303 		 */
5304 		while (slot < nritems) {
5305 			u64 gen;
5306 
5307 			gen = btrfs_node_ptr_generation(cur, slot);
5308 			if (gen < min_trans) {
5309 				slot++;
5310 				continue;
5311 			}
5312 			break;
5313 		}
5314 find_next_key:
5315 		/*
5316 		 * we didn't find a candidate key in this node, walk forward
5317 		 * and find another one
5318 		 */
5319 		if (slot >= nritems) {
5320 			path->slots[level] = slot;
5321 			btrfs_set_path_blocking(path);
5322 			sret = btrfs_find_next_key(root, path, min_key, level,
5323 						  min_trans);
5324 			if (sret == 0) {
5325 				btrfs_release_path(path);
5326 				goto again;
5327 			} else {
5328 				goto out;
5329 			}
5330 		}
5331 		/* save our key for returning back */
5332 		btrfs_node_key_to_cpu(cur, &found_key, slot);
5333 		path->slots[level] = slot;
5334 		if (level == path->lowest_level) {
5335 			ret = 0;
5336 			goto out;
5337 		}
5338 		btrfs_set_path_blocking(path);
5339 		cur = btrfs_read_node_slot(cur, slot);
5340 		if (IS_ERR(cur)) {
5341 			ret = PTR_ERR(cur);
5342 			goto out;
5343 		}
5344 
5345 		btrfs_tree_read_lock(cur);
5346 
5347 		path->locks[level - 1] = BTRFS_READ_LOCK;
5348 		path->nodes[level - 1] = cur;
5349 		unlock_up(path, level, 1, 0, NULL);
5350 	}
5351 out:
5352 	path->keep_locks = keep_locks;
5353 	if (ret == 0) {
5354 		btrfs_unlock_up_safe(path, path->lowest_level + 1);
5355 		btrfs_set_path_blocking(path);
5356 		memcpy(min_key, &found_key, sizeof(found_key));
5357 	}
5358 	return ret;
5359 }
5360 
5361 /*
5362  * this is similar to btrfs_next_leaf, but does not try to preserve
5363  * and fixup the path.  It looks for and returns the next key in the
5364  * tree based on the current path and the min_trans parameters.
5365  *
5366  * 0 is returned if another key is found, < 0 if there are any errors
5367  * and 1 is returned if there are no higher keys in the tree
5368  *
5369  * path->keep_locks should be set to 1 on the search made before
5370  * calling this function.
5371  */
btrfs_find_next_key(struct btrfs_root * root,struct btrfs_path * path,struct btrfs_key * key,int level,u64 min_trans)5372 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5373 			struct btrfs_key *key, int level, u64 min_trans)
5374 {
5375 	int slot;
5376 	struct extent_buffer *c;
5377 
5378 	WARN_ON(!path->keep_locks && !path->skip_locking);
5379 	while (level < BTRFS_MAX_LEVEL) {
5380 		if (!path->nodes[level])
5381 			return 1;
5382 
5383 		slot = path->slots[level] + 1;
5384 		c = path->nodes[level];
5385 next:
5386 		if (slot >= btrfs_header_nritems(c)) {
5387 			int ret;
5388 			int orig_lowest;
5389 			struct btrfs_key cur_key;
5390 			if (level + 1 >= BTRFS_MAX_LEVEL ||
5391 			    !path->nodes[level + 1])
5392 				return 1;
5393 
5394 			if (path->locks[level + 1] || path->skip_locking) {
5395 				level++;
5396 				continue;
5397 			}
5398 
5399 			slot = btrfs_header_nritems(c) - 1;
5400 			if (level == 0)
5401 				btrfs_item_key_to_cpu(c, &cur_key, slot);
5402 			else
5403 				btrfs_node_key_to_cpu(c, &cur_key, slot);
5404 
5405 			orig_lowest = path->lowest_level;
5406 			btrfs_release_path(path);
5407 			path->lowest_level = level;
5408 			ret = btrfs_search_slot(NULL, root, &cur_key, path,
5409 						0, 0);
5410 			path->lowest_level = orig_lowest;
5411 			if (ret < 0)
5412 				return ret;
5413 
5414 			c = path->nodes[level];
5415 			slot = path->slots[level];
5416 			if (ret == 0)
5417 				slot++;
5418 			goto next;
5419 		}
5420 
5421 		if (level == 0)
5422 			btrfs_item_key_to_cpu(c, key, slot);
5423 		else {
5424 			u64 gen = btrfs_node_ptr_generation(c, slot);
5425 
5426 			if (gen < min_trans) {
5427 				slot++;
5428 				goto next;
5429 			}
5430 			btrfs_node_key_to_cpu(c, key, slot);
5431 		}
5432 		return 0;
5433 	}
5434 	return 1;
5435 }
5436 
5437 /*
5438  * search the tree again to find a leaf with greater keys
5439  * returns 0 if it found something or 1 if there are no greater leaves.
5440  * returns < 0 on io errors.
5441  */
btrfs_next_leaf(struct btrfs_root * root,struct btrfs_path * path)5442 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5443 {
5444 	return btrfs_next_old_leaf(root, path, 0);
5445 }
5446 
btrfs_next_old_leaf(struct btrfs_root * root,struct btrfs_path * path,u64 time_seq)5447 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5448 			u64 time_seq)
5449 {
5450 	int slot;
5451 	int level;
5452 	struct extent_buffer *c;
5453 	struct extent_buffer *next;
5454 	struct btrfs_key key;
5455 	u32 nritems;
5456 	int ret;
5457 	int old_spinning = path->leave_spinning;
5458 	int next_rw_lock = 0;
5459 
5460 	nritems = btrfs_header_nritems(path->nodes[0]);
5461 	if (nritems == 0)
5462 		return 1;
5463 
5464 	btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5465 again:
5466 	level = 1;
5467 	next = NULL;
5468 	next_rw_lock = 0;
5469 	btrfs_release_path(path);
5470 
5471 	path->keep_locks = 1;
5472 	path->leave_spinning = 1;
5473 
5474 	if (time_seq)
5475 		ret = btrfs_search_old_slot(root, &key, path, time_seq);
5476 	else
5477 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5478 	path->keep_locks = 0;
5479 
5480 	if (ret < 0)
5481 		return ret;
5482 
5483 	nritems = btrfs_header_nritems(path->nodes[0]);
5484 	/*
5485 	 * by releasing the path above we dropped all our locks.  A balance
5486 	 * could have added more items next to the key that used to be
5487 	 * at the very end of the block.  So, check again here and
5488 	 * advance the path if there are now more items available.
5489 	 */
5490 	if (nritems > 0 && path->slots[0] < nritems - 1) {
5491 		if (ret == 0)
5492 			path->slots[0]++;
5493 		ret = 0;
5494 		goto done;
5495 	}
5496 	/*
5497 	 * So the above check misses one case:
5498 	 * - after releasing the path above, someone has removed the item that
5499 	 *   used to be at the very end of the block, and balance between leafs
5500 	 *   gets another one with bigger key.offset to replace it.
5501 	 *
5502 	 * This one should be returned as well, or we can get leaf corruption
5503 	 * later(esp. in __btrfs_drop_extents()).
5504 	 *
5505 	 * And a bit more explanation about this check,
5506 	 * with ret > 0, the key isn't found, the path points to the slot
5507 	 * where it should be inserted, so the path->slots[0] item must be the
5508 	 * bigger one.
5509 	 */
5510 	if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5511 		ret = 0;
5512 		goto done;
5513 	}
5514 
5515 	while (level < BTRFS_MAX_LEVEL) {
5516 		if (!path->nodes[level]) {
5517 			ret = 1;
5518 			goto done;
5519 		}
5520 
5521 		slot = path->slots[level] + 1;
5522 		c = path->nodes[level];
5523 		if (slot >= btrfs_header_nritems(c)) {
5524 			level++;
5525 			if (level == BTRFS_MAX_LEVEL) {
5526 				ret = 1;
5527 				goto done;
5528 			}
5529 			continue;
5530 		}
5531 
5532 		if (next) {
5533 			btrfs_tree_unlock_rw(next, next_rw_lock);
5534 			free_extent_buffer(next);
5535 		}
5536 
5537 		next = c;
5538 		next_rw_lock = path->locks[level];
5539 		ret = read_block_for_search(root, path, &next, level,
5540 					    slot, &key);
5541 		if (ret == -EAGAIN)
5542 			goto again;
5543 
5544 		if (ret < 0) {
5545 			btrfs_release_path(path);
5546 			goto done;
5547 		}
5548 
5549 		if (!path->skip_locking) {
5550 			ret = btrfs_try_tree_read_lock(next);
5551 			if (!ret && time_seq) {
5552 				/*
5553 				 * If we don't get the lock, we may be racing
5554 				 * with push_leaf_left, holding that lock while
5555 				 * itself waiting for the leaf we've currently
5556 				 * locked. To solve this situation, we give up
5557 				 * on our lock and cycle.
5558 				 */
5559 				free_extent_buffer(next);
5560 				btrfs_release_path(path);
5561 				cond_resched();
5562 				goto again;
5563 			}
5564 			if (!ret) {
5565 				btrfs_set_path_blocking(path);
5566 				__btrfs_tree_read_lock(next,
5567 						       BTRFS_NESTING_RIGHT,
5568 						       path->recurse);
5569 			}
5570 			next_rw_lock = BTRFS_READ_LOCK;
5571 		}
5572 		break;
5573 	}
5574 	path->slots[level] = slot;
5575 	while (1) {
5576 		level--;
5577 		c = path->nodes[level];
5578 		if (path->locks[level])
5579 			btrfs_tree_unlock_rw(c, path->locks[level]);
5580 
5581 		free_extent_buffer(c);
5582 		path->nodes[level] = next;
5583 		path->slots[level] = 0;
5584 		if (!path->skip_locking)
5585 			path->locks[level] = next_rw_lock;
5586 		if (!level)
5587 			break;
5588 
5589 		ret = read_block_for_search(root, path, &next, level,
5590 					    0, &key);
5591 		if (ret == -EAGAIN)
5592 			goto again;
5593 
5594 		if (ret < 0) {
5595 			btrfs_release_path(path);
5596 			goto done;
5597 		}
5598 
5599 		if (!path->skip_locking) {
5600 			ret = btrfs_try_tree_read_lock(next);
5601 			if (!ret) {
5602 				btrfs_set_path_blocking(path);
5603 				__btrfs_tree_read_lock(next,
5604 						       BTRFS_NESTING_RIGHT,
5605 						       path->recurse);
5606 			}
5607 			next_rw_lock = BTRFS_READ_LOCK;
5608 		}
5609 	}
5610 	ret = 0;
5611 done:
5612 	unlock_up(path, 0, 1, 0, NULL);
5613 	path->leave_spinning = old_spinning;
5614 	if (!old_spinning)
5615 		btrfs_set_path_blocking(path);
5616 
5617 	return ret;
5618 }
5619 
5620 /*
5621  * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5622  * searching until it gets past min_objectid or finds an item of 'type'
5623  *
5624  * returns 0 if something is found, 1 if nothing was found and < 0 on error
5625  */
btrfs_previous_item(struct btrfs_root * root,struct btrfs_path * path,u64 min_objectid,int type)5626 int btrfs_previous_item(struct btrfs_root *root,
5627 			struct btrfs_path *path, u64 min_objectid,
5628 			int type)
5629 {
5630 	struct btrfs_key found_key;
5631 	struct extent_buffer *leaf;
5632 	u32 nritems;
5633 	int ret;
5634 
5635 	while (1) {
5636 		if (path->slots[0] == 0) {
5637 			btrfs_set_path_blocking(path);
5638 			ret = btrfs_prev_leaf(root, path);
5639 			if (ret != 0)
5640 				return ret;
5641 		} else {
5642 			path->slots[0]--;
5643 		}
5644 		leaf = path->nodes[0];
5645 		nritems = btrfs_header_nritems(leaf);
5646 		if (nritems == 0)
5647 			return 1;
5648 		if (path->slots[0] == nritems)
5649 			path->slots[0]--;
5650 
5651 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5652 		if (found_key.objectid < min_objectid)
5653 			break;
5654 		if (found_key.type == type)
5655 			return 0;
5656 		if (found_key.objectid == min_objectid &&
5657 		    found_key.type < type)
5658 			break;
5659 	}
5660 	return 1;
5661 }
5662 
5663 /*
5664  * search in extent tree to find a previous Metadata/Data extent item with
5665  * min objecitd.
5666  *
5667  * returns 0 if something is found, 1 if nothing was found and < 0 on error
5668  */
btrfs_previous_extent_item(struct btrfs_root * root,struct btrfs_path * path,u64 min_objectid)5669 int btrfs_previous_extent_item(struct btrfs_root *root,
5670 			struct btrfs_path *path, u64 min_objectid)
5671 {
5672 	struct btrfs_key found_key;
5673 	struct extent_buffer *leaf;
5674 	u32 nritems;
5675 	int ret;
5676 
5677 	while (1) {
5678 		if (path->slots[0] == 0) {
5679 			btrfs_set_path_blocking(path);
5680 			ret = btrfs_prev_leaf(root, path);
5681 			if (ret != 0)
5682 				return ret;
5683 		} else {
5684 			path->slots[0]--;
5685 		}
5686 		leaf = path->nodes[0];
5687 		nritems = btrfs_header_nritems(leaf);
5688 		if (nritems == 0)
5689 			return 1;
5690 		if (path->slots[0] == nritems)
5691 			path->slots[0]--;
5692 
5693 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5694 		if (found_key.objectid < min_objectid)
5695 			break;
5696 		if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5697 		    found_key.type == BTRFS_METADATA_ITEM_KEY)
5698 			return 0;
5699 		if (found_key.objectid == min_objectid &&
5700 		    found_key.type < BTRFS_EXTENT_ITEM_KEY)
5701 			break;
5702 	}
5703 	return 1;
5704 }
5705