Loading...
1/*
2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/sched.h>
20#include <linux/slab.h>
21#include "ctree.h"
22#include "disk-io.h"
23#include "transaction.h"
24#include "print-tree.h"
25#include "locking.h"
26
27static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
28 *root, struct btrfs_path *path, int level);
29static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
30 *root, struct btrfs_key *ins_key,
31 struct btrfs_path *path, int data_size, int extend);
32static int push_node_left(struct btrfs_trans_handle *trans,
33 struct btrfs_root *root, struct extent_buffer *dst,
34 struct extent_buffer *src, int empty);
35static int balance_node_right(struct btrfs_trans_handle *trans,
36 struct btrfs_root *root,
37 struct extent_buffer *dst_buf,
38 struct extent_buffer *src_buf);
39static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
40 struct btrfs_path *path, int level, int slot);
41
42struct btrfs_path *btrfs_alloc_path(void)
43{
44 struct btrfs_path *path;
45 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
46 return path;
47}
48
49/*
50 * set all locked nodes in the path to blocking locks. This should
51 * be done before scheduling
52 */
53noinline void btrfs_set_path_blocking(struct btrfs_path *p)
54{
55 int i;
56 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
57 if (!p->nodes[i] || !p->locks[i])
58 continue;
59 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
60 if (p->locks[i] == BTRFS_READ_LOCK)
61 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
62 else if (p->locks[i] == BTRFS_WRITE_LOCK)
63 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
64 }
65}
66
67/*
68 * reset all the locked nodes in the patch to spinning locks.
69 *
70 * held is used to keep lockdep happy, when lockdep is enabled
71 * we set held to a blocking lock before we go around and
72 * retake all the spinlocks in the path. You can safely use NULL
73 * for held
74 */
75noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
76 struct extent_buffer *held, int held_rw)
77{
78 int i;
79
80#ifdef CONFIG_DEBUG_LOCK_ALLOC
81 /* lockdep really cares that we take all of these spinlocks
82 * in the right order. If any of the locks in the path are not
83 * currently blocking, it is going to complain. So, make really
84 * really sure by forcing the path to blocking before we clear
85 * the path blocking.
86 */
87 if (held) {
88 btrfs_set_lock_blocking_rw(held, held_rw);
89 if (held_rw == BTRFS_WRITE_LOCK)
90 held_rw = BTRFS_WRITE_LOCK_BLOCKING;
91 else if (held_rw == BTRFS_READ_LOCK)
92 held_rw = BTRFS_READ_LOCK_BLOCKING;
93 }
94 btrfs_set_path_blocking(p);
95#endif
96
97 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
98 if (p->nodes[i] && p->locks[i]) {
99 btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
100 if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
101 p->locks[i] = BTRFS_WRITE_LOCK;
102 else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
103 p->locks[i] = BTRFS_READ_LOCK;
104 }
105 }
106
107#ifdef CONFIG_DEBUG_LOCK_ALLOC
108 if (held)
109 btrfs_clear_lock_blocking_rw(held, held_rw);
110#endif
111}
112
113/* this also releases the path */
114void btrfs_free_path(struct btrfs_path *p)
115{
116 if (!p)
117 return;
118 btrfs_release_path(p);
119 kmem_cache_free(btrfs_path_cachep, p);
120}
121
122/*
123 * path release drops references on the extent buffers in the path
124 * and it drops any locks held by this path
125 *
126 * It is safe to call this on paths that no locks or extent buffers held.
127 */
128noinline void btrfs_release_path(struct btrfs_path *p)
129{
130 int i;
131
132 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
133 p->slots[i] = 0;
134 if (!p->nodes[i])
135 continue;
136 if (p->locks[i]) {
137 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
138 p->locks[i] = 0;
139 }
140 free_extent_buffer(p->nodes[i]);
141 p->nodes[i] = NULL;
142 }
143}
144
145/*
146 * safely gets a reference on the root node of a tree. A lock
147 * is not taken, so a concurrent writer may put a different node
148 * at the root of the tree. See btrfs_lock_root_node for the
149 * looping required.
150 *
151 * The extent buffer returned by this has a reference taken, so
152 * it won't disappear. It may stop being the root of the tree
153 * at any time because there are no locks held.
154 */
155struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
156{
157 struct extent_buffer *eb;
158
159 rcu_read_lock();
160 eb = rcu_dereference(root->node);
161 extent_buffer_get(eb);
162 rcu_read_unlock();
163 return eb;
164}
165
166/* loop around taking references on and locking the root node of the
167 * tree until you end up with a lock on the root. A locked buffer
168 * is returned, with a reference held.
169 */
170struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
171{
172 struct extent_buffer *eb;
173
174 while (1) {
175 eb = btrfs_root_node(root);
176 btrfs_tree_lock(eb);
177 if (eb == root->node)
178 break;
179 btrfs_tree_unlock(eb);
180 free_extent_buffer(eb);
181 }
182 return eb;
183}
184
185/* loop around taking references on and locking the root node of the
186 * tree until you end up with a lock on the root. A locked buffer
187 * is returned, with a reference held.
188 */
189struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
190{
191 struct extent_buffer *eb;
192
193 while (1) {
194 eb = btrfs_root_node(root);
195 btrfs_tree_read_lock(eb);
196 if (eb == root->node)
197 break;
198 btrfs_tree_read_unlock(eb);
199 free_extent_buffer(eb);
200 }
201 return eb;
202}
203
204/* cowonly root (everything not a reference counted cow subvolume), just get
205 * put onto a simple dirty list. transaction.c walks this to make sure they
206 * get properly updated on disk.
207 */
208static void add_root_to_dirty_list(struct btrfs_root *root)
209{
210 if (root->track_dirty && list_empty(&root->dirty_list)) {
211 list_add(&root->dirty_list,
212 &root->fs_info->dirty_cowonly_roots);
213 }
214}
215
216/*
217 * used by snapshot creation to make a copy of a root for a tree with
218 * a given objectid. The buffer with the new root node is returned in
219 * cow_ret, and this func returns zero on success or a negative error code.
220 */
221int btrfs_copy_root(struct btrfs_trans_handle *trans,
222 struct btrfs_root *root,
223 struct extent_buffer *buf,
224 struct extent_buffer **cow_ret, u64 new_root_objectid)
225{
226 struct extent_buffer *cow;
227 int ret = 0;
228 int level;
229 struct btrfs_disk_key disk_key;
230
231 WARN_ON(root->ref_cows && trans->transid !=
232 root->fs_info->running_transaction->transid);
233 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
234
235 level = btrfs_header_level(buf);
236 if (level == 0)
237 btrfs_item_key(buf, &disk_key, 0);
238 else
239 btrfs_node_key(buf, &disk_key, 0);
240
241 cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
242 new_root_objectid, &disk_key, level,
243 buf->start, 0);
244 if (IS_ERR(cow))
245 return PTR_ERR(cow);
246
247 copy_extent_buffer(cow, buf, 0, 0, cow->len);
248 btrfs_set_header_bytenr(cow, cow->start);
249 btrfs_set_header_generation(cow, trans->transid);
250 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
251 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
252 BTRFS_HEADER_FLAG_RELOC);
253 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
254 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
255 else
256 btrfs_set_header_owner(cow, new_root_objectid);
257
258 write_extent_buffer(cow, root->fs_info->fsid,
259 (unsigned long)btrfs_header_fsid(cow),
260 BTRFS_FSID_SIZE);
261
262 WARN_ON(btrfs_header_generation(buf) > trans->transid);
263 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
264 ret = btrfs_inc_ref(trans, root, cow, 1);
265 else
266 ret = btrfs_inc_ref(trans, root, cow, 0);
267
268 if (ret)
269 return ret;
270
271 btrfs_mark_buffer_dirty(cow);
272 *cow_ret = cow;
273 return 0;
274}
275
276/*
277 * check if the tree block can be shared by multiple trees
278 */
279int btrfs_block_can_be_shared(struct btrfs_root *root,
280 struct extent_buffer *buf)
281{
282 /*
283 * Tree blocks not in refernece counted trees and tree roots
284 * are never shared. If a block was allocated after the last
285 * snapshot and the block was not allocated by tree relocation,
286 * we know the block is not shared.
287 */
288 if (root->ref_cows &&
289 buf != root->node && buf != root->commit_root &&
290 (btrfs_header_generation(buf) <=
291 btrfs_root_last_snapshot(&root->root_item) ||
292 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
293 return 1;
294#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
295 if (root->ref_cows &&
296 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
297 return 1;
298#endif
299 return 0;
300}
301
302static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
303 struct btrfs_root *root,
304 struct extent_buffer *buf,
305 struct extent_buffer *cow,
306 int *last_ref)
307{
308 u64 refs;
309 u64 owner;
310 u64 flags;
311 u64 new_flags = 0;
312 int ret;
313
314 /*
315 * Backrefs update rules:
316 *
317 * Always use full backrefs for extent pointers in tree block
318 * allocated by tree relocation.
319 *
320 * If a shared tree block is no longer referenced by its owner
321 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
322 * use full backrefs for extent pointers in tree block.
323 *
324 * If a tree block is been relocating
325 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
326 * use full backrefs for extent pointers in tree block.
327 * The reason for this is some operations (such as drop tree)
328 * are only allowed for blocks use full backrefs.
329 */
330
331 if (btrfs_block_can_be_shared(root, buf)) {
332 ret = btrfs_lookup_extent_info(trans, root, buf->start,
333 buf->len, &refs, &flags);
334 BUG_ON(ret);
335 BUG_ON(refs == 0);
336 } else {
337 refs = 1;
338 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
339 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
340 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
341 else
342 flags = 0;
343 }
344
345 owner = btrfs_header_owner(buf);
346 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
347 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
348
349 if (refs > 1) {
350 if ((owner == root->root_key.objectid ||
351 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
352 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
353 ret = btrfs_inc_ref(trans, root, buf, 1);
354 BUG_ON(ret);
355
356 if (root->root_key.objectid ==
357 BTRFS_TREE_RELOC_OBJECTID) {
358 ret = btrfs_dec_ref(trans, root, buf, 0);
359 BUG_ON(ret);
360 ret = btrfs_inc_ref(trans, root, cow, 1);
361 BUG_ON(ret);
362 }
363 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
364 } else {
365
366 if (root->root_key.objectid ==
367 BTRFS_TREE_RELOC_OBJECTID)
368 ret = btrfs_inc_ref(trans, root, cow, 1);
369 else
370 ret = btrfs_inc_ref(trans, root, cow, 0);
371 BUG_ON(ret);
372 }
373 if (new_flags != 0) {
374 ret = btrfs_set_disk_extent_flags(trans, root,
375 buf->start,
376 buf->len,
377 new_flags, 0);
378 BUG_ON(ret);
379 }
380 } else {
381 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
382 if (root->root_key.objectid ==
383 BTRFS_TREE_RELOC_OBJECTID)
384 ret = btrfs_inc_ref(trans, root, cow, 1);
385 else
386 ret = btrfs_inc_ref(trans, root, cow, 0);
387 BUG_ON(ret);
388 ret = btrfs_dec_ref(trans, root, buf, 1);
389 BUG_ON(ret);
390 }
391 clean_tree_block(trans, root, buf);
392 *last_ref = 1;
393 }
394 return 0;
395}
396
397/*
398 * does the dirty work in cow of a single block. The parent block (if
399 * supplied) is updated to point to the new cow copy. The new buffer is marked
400 * dirty and returned locked. If you modify the block it needs to be marked
401 * dirty again.
402 *
403 * search_start -- an allocation hint for the new block
404 *
405 * empty_size -- a hint that you plan on doing more cow. This is the size in
406 * bytes the allocator should try to find free next to the block it returns.
407 * This is just a hint and may be ignored by the allocator.
408 */
409static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
410 struct btrfs_root *root,
411 struct extent_buffer *buf,
412 struct extent_buffer *parent, int parent_slot,
413 struct extent_buffer **cow_ret,
414 u64 search_start, u64 empty_size)
415{
416 struct btrfs_disk_key disk_key;
417 struct extent_buffer *cow;
418 int level;
419 int last_ref = 0;
420 int unlock_orig = 0;
421 u64 parent_start;
422
423 if (*cow_ret == buf)
424 unlock_orig = 1;
425
426 btrfs_assert_tree_locked(buf);
427
428 WARN_ON(root->ref_cows && trans->transid !=
429 root->fs_info->running_transaction->transid);
430 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
431
432 level = btrfs_header_level(buf);
433
434 if (level == 0)
435 btrfs_item_key(buf, &disk_key, 0);
436 else
437 btrfs_node_key(buf, &disk_key, 0);
438
439 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
440 if (parent)
441 parent_start = parent->start;
442 else
443 parent_start = 0;
444 } else
445 parent_start = 0;
446
447 cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
448 root->root_key.objectid, &disk_key,
449 level, search_start, empty_size);
450 if (IS_ERR(cow))
451 return PTR_ERR(cow);
452
453 /* cow is set to blocking by btrfs_init_new_buffer */
454
455 copy_extent_buffer(cow, buf, 0, 0, cow->len);
456 btrfs_set_header_bytenr(cow, cow->start);
457 btrfs_set_header_generation(cow, trans->transid);
458 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
459 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
460 BTRFS_HEADER_FLAG_RELOC);
461 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
462 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
463 else
464 btrfs_set_header_owner(cow, root->root_key.objectid);
465
466 write_extent_buffer(cow, root->fs_info->fsid,
467 (unsigned long)btrfs_header_fsid(cow),
468 BTRFS_FSID_SIZE);
469
470 update_ref_for_cow(trans, root, buf, cow, &last_ref);
471
472 if (root->ref_cows)
473 btrfs_reloc_cow_block(trans, root, buf, cow);
474
475 if (buf == root->node) {
476 WARN_ON(parent && parent != buf);
477 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
478 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
479 parent_start = buf->start;
480 else
481 parent_start = 0;
482
483 extent_buffer_get(cow);
484 rcu_assign_pointer(root->node, cow);
485
486 btrfs_free_tree_block(trans, root, buf, parent_start,
487 last_ref);
488 free_extent_buffer(buf);
489 add_root_to_dirty_list(root);
490 } else {
491 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
492 parent_start = parent->start;
493 else
494 parent_start = 0;
495
496 WARN_ON(trans->transid != btrfs_header_generation(parent));
497 btrfs_set_node_blockptr(parent, parent_slot,
498 cow->start);
499 btrfs_set_node_ptr_generation(parent, parent_slot,
500 trans->transid);
501 btrfs_mark_buffer_dirty(parent);
502 btrfs_free_tree_block(trans, root, buf, parent_start,
503 last_ref);
504 }
505 if (unlock_orig)
506 btrfs_tree_unlock(buf);
507 free_extent_buffer(buf);
508 btrfs_mark_buffer_dirty(cow);
509 *cow_ret = cow;
510 return 0;
511}
512
513static inline int should_cow_block(struct btrfs_trans_handle *trans,
514 struct btrfs_root *root,
515 struct extent_buffer *buf)
516{
517 if (btrfs_header_generation(buf) == trans->transid &&
518 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
519 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
520 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
521 return 0;
522 return 1;
523}
524
525/*
526 * cows a single block, see __btrfs_cow_block for the real work.
527 * This version of it has extra checks so that a block isn't cow'd more than
528 * once per transaction, as long as it hasn't been written yet
529 */
530noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
531 struct btrfs_root *root, struct extent_buffer *buf,
532 struct extent_buffer *parent, int parent_slot,
533 struct extent_buffer **cow_ret)
534{
535 u64 search_start;
536 int ret;
537
538 if (trans->transaction != root->fs_info->running_transaction) {
539 printk(KERN_CRIT "trans %llu running %llu\n",
540 (unsigned long long)trans->transid,
541 (unsigned long long)
542 root->fs_info->running_transaction->transid);
543 WARN_ON(1);
544 }
545 if (trans->transid != root->fs_info->generation) {
546 printk(KERN_CRIT "trans %llu running %llu\n",
547 (unsigned long long)trans->transid,
548 (unsigned long long)root->fs_info->generation);
549 WARN_ON(1);
550 }
551
552 if (!should_cow_block(trans, root, buf)) {
553 *cow_ret = buf;
554 return 0;
555 }
556
557 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
558
559 if (parent)
560 btrfs_set_lock_blocking(parent);
561 btrfs_set_lock_blocking(buf);
562
563 ret = __btrfs_cow_block(trans, root, buf, parent,
564 parent_slot, cow_ret, search_start, 0);
565
566 trace_btrfs_cow_block(root, buf, *cow_ret);
567
568 return ret;
569}
570
571/*
572 * helper function for defrag to decide if two blocks pointed to by a
573 * node are actually close by
574 */
575static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
576{
577 if (blocknr < other && other - (blocknr + blocksize) < 32768)
578 return 1;
579 if (blocknr > other && blocknr - (other + blocksize) < 32768)
580 return 1;
581 return 0;
582}
583
584/*
585 * compare two keys in a memcmp fashion
586 */
587static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
588{
589 struct btrfs_key k1;
590
591 btrfs_disk_key_to_cpu(&k1, disk);
592
593 return btrfs_comp_cpu_keys(&k1, k2);
594}
595
596/*
597 * same as comp_keys only with two btrfs_key's
598 */
599int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
600{
601 if (k1->objectid > k2->objectid)
602 return 1;
603 if (k1->objectid < k2->objectid)
604 return -1;
605 if (k1->type > k2->type)
606 return 1;
607 if (k1->type < k2->type)
608 return -1;
609 if (k1->offset > k2->offset)
610 return 1;
611 if (k1->offset < k2->offset)
612 return -1;
613 return 0;
614}
615
616/*
617 * this is used by the defrag code to go through all the
618 * leaves pointed to by a node and reallocate them so that
619 * disk order is close to key order
620 */
621int btrfs_realloc_node(struct btrfs_trans_handle *trans,
622 struct btrfs_root *root, struct extent_buffer *parent,
623 int start_slot, int cache_only, u64 *last_ret,
624 struct btrfs_key *progress)
625{
626 struct extent_buffer *cur;
627 u64 blocknr;
628 u64 gen;
629 u64 search_start = *last_ret;
630 u64 last_block = 0;
631 u64 other;
632 u32 parent_nritems;
633 int end_slot;
634 int i;
635 int err = 0;
636 int parent_level;
637 int uptodate;
638 u32 blocksize;
639 int progress_passed = 0;
640 struct btrfs_disk_key disk_key;
641
642 parent_level = btrfs_header_level(parent);
643 if (cache_only && parent_level != 1)
644 return 0;
645
646 if (trans->transaction != root->fs_info->running_transaction)
647 WARN_ON(1);
648 if (trans->transid != root->fs_info->generation)
649 WARN_ON(1);
650
651 parent_nritems = btrfs_header_nritems(parent);
652 blocksize = btrfs_level_size(root, parent_level - 1);
653 end_slot = parent_nritems;
654
655 if (parent_nritems == 1)
656 return 0;
657
658 btrfs_set_lock_blocking(parent);
659
660 for (i = start_slot; i < end_slot; i++) {
661 int close = 1;
662
663 btrfs_node_key(parent, &disk_key, i);
664 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
665 continue;
666
667 progress_passed = 1;
668 blocknr = btrfs_node_blockptr(parent, i);
669 gen = btrfs_node_ptr_generation(parent, i);
670 if (last_block == 0)
671 last_block = blocknr;
672
673 if (i > 0) {
674 other = btrfs_node_blockptr(parent, i - 1);
675 close = close_blocks(blocknr, other, blocksize);
676 }
677 if (!close && i < end_slot - 2) {
678 other = btrfs_node_blockptr(parent, i + 1);
679 close = close_blocks(blocknr, other, blocksize);
680 }
681 if (close) {
682 last_block = blocknr;
683 continue;
684 }
685
686 cur = btrfs_find_tree_block(root, blocknr, blocksize);
687 if (cur)
688 uptodate = btrfs_buffer_uptodate(cur, gen);
689 else
690 uptodate = 0;
691 if (!cur || !uptodate) {
692 if (cache_only) {
693 free_extent_buffer(cur);
694 continue;
695 }
696 if (!cur) {
697 cur = read_tree_block(root, blocknr,
698 blocksize, gen);
699 if (!cur)
700 return -EIO;
701 } else if (!uptodate) {
702 btrfs_read_buffer(cur, gen);
703 }
704 }
705 if (search_start == 0)
706 search_start = last_block;
707
708 btrfs_tree_lock(cur);
709 btrfs_set_lock_blocking(cur);
710 err = __btrfs_cow_block(trans, root, cur, parent, i,
711 &cur, search_start,
712 min(16 * blocksize,
713 (end_slot - i) * blocksize));
714 if (err) {
715 btrfs_tree_unlock(cur);
716 free_extent_buffer(cur);
717 break;
718 }
719 search_start = cur->start;
720 last_block = cur->start;
721 *last_ret = search_start;
722 btrfs_tree_unlock(cur);
723 free_extent_buffer(cur);
724 }
725 return err;
726}
727
728/*
729 * The leaf data grows from end-to-front in the node.
730 * this returns the address of the start of the last item,
731 * which is the stop of the leaf data stack
732 */
733static inline unsigned int leaf_data_end(struct btrfs_root *root,
734 struct extent_buffer *leaf)
735{
736 u32 nr = btrfs_header_nritems(leaf);
737 if (nr == 0)
738 return BTRFS_LEAF_DATA_SIZE(root);
739 return btrfs_item_offset_nr(leaf, nr - 1);
740}
741
742
743/*
744 * search for key in the extent_buffer. The items start at offset p,
745 * and they are item_size apart. There are 'max' items in p.
746 *
747 * the slot in the array is returned via slot, and it points to
748 * the place where you would insert key if it is not found in
749 * the array.
750 *
751 * slot may point to max if the key is bigger than all of the keys
752 */
753static noinline int generic_bin_search(struct extent_buffer *eb,
754 unsigned long p,
755 int item_size, struct btrfs_key *key,
756 int max, int *slot)
757{
758 int low = 0;
759 int high = max;
760 int mid;
761 int ret;
762 struct btrfs_disk_key *tmp = NULL;
763 struct btrfs_disk_key unaligned;
764 unsigned long offset;
765 char *kaddr = NULL;
766 unsigned long map_start = 0;
767 unsigned long map_len = 0;
768 int err;
769
770 while (low < high) {
771 mid = (low + high) / 2;
772 offset = p + mid * item_size;
773
774 if (!kaddr || offset < map_start ||
775 (offset + sizeof(struct btrfs_disk_key)) >
776 map_start + map_len) {
777
778 err = map_private_extent_buffer(eb, offset,
779 sizeof(struct btrfs_disk_key),
780 &kaddr, &map_start, &map_len);
781
782 if (!err) {
783 tmp = (struct btrfs_disk_key *)(kaddr + offset -
784 map_start);
785 } else {
786 read_extent_buffer(eb, &unaligned,
787 offset, sizeof(unaligned));
788 tmp = &unaligned;
789 }
790
791 } else {
792 tmp = (struct btrfs_disk_key *)(kaddr + offset -
793 map_start);
794 }
795 ret = comp_keys(tmp, key);
796
797 if (ret < 0)
798 low = mid + 1;
799 else if (ret > 0)
800 high = mid;
801 else {
802 *slot = mid;
803 return 0;
804 }
805 }
806 *slot = low;
807 return 1;
808}
809
810/*
811 * simple bin_search frontend that does the right thing for
812 * leaves vs nodes
813 */
814static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
815 int level, int *slot)
816{
817 if (level == 0) {
818 return generic_bin_search(eb,
819 offsetof(struct btrfs_leaf, items),
820 sizeof(struct btrfs_item),
821 key, btrfs_header_nritems(eb),
822 slot);
823 } else {
824 return generic_bin_search(eb,
825 offsetof(struct btrfs_node, ptrs),
826 sizeof(struct btrfs_key_ptr),
827 key, btrfs_header_nritems(eb),
828 slot);
829 }
830 return -1;
831}
832
833int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
834 int level, int *slot)
835{
836 return bin_search(eb, key, level, slot);
837}
838
839static void root_add_used(struct btrfs_root *root, u32 size)
840{
841 spin_lock(&root->accounting_lock);
842 btrfs_set_root_used(&root->root_item,
843 btrfs_root_used(&root->root_item) + size);
844 spin_unlock(&root->accounting_lock);
845}
846
847static void root_sub_used(struct btrfs_root *root, u32 size)
848{
849 spin_lock(&root->accounting_lock);
850 btrfs_set_root_used(&root->root_item,
851 btrfs_root_used(&root->root_item) - size);
852 spin_unlock(&root->accounting_lock);
853}
854
855/* given a node and slot number, this reads the blocks it points to. The
856 * extent buffer is returned with a reference taken (but unlocked).
857 * NULL is returned on error.
858 */
859static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
860 struct extent_buffer *parent, int slot)
861{
862 int level = btrfs_header_level(parent);
863 if (slot < 0)
864 return NULL;
865 if (slot >= btrfs_header_nritems(parent))
866 return NULL;
867
868 BUG_ON(level == 0);
869
870 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
871 btrfs_level_size(root, level - 1),
872 btrfs_node_ptr_generation(parent, slot));
873}
874
875/*
876 * node level balancing, used to make sure nodes are in proper order for
877 * item deletion. We balance from the top down, so we have to make sure
878 * that a deletion won't leave an node completely empty later on.
879 */
880static noinline int balance_level(struct btrfs_trans_handle *trans,
881 struct btrfs_root *root,
882 struct btrfs_path *path, int level)
883{
884 struct extent_buffer *right = NULL;
885 struct extent_buffer *mid;
886 struct extent_buffer *left = NULL;
887 struct extent_buffer *parent = NULL;
888 int ret = 0;
889 int wret;
890 int pslot;
891 int orig_slot = path->slots[level];
892 u64 orig_ptr;
893
894 if (level == 0)
895 return 0;
896
897 mid = path->nodes[level];
898
899 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
900 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
901 WARN_ON(btrfs_header_generation(mid) != trans->transid);
902
903 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
904
905 if (level < BTRFS_MAX_LEVEL - 1)
906 parent = path->nodes[level + 1];
907 pslot = path->slots[level + 1];
908
909 /*
910 * deal with the case where there is only one pointer in the root
911 * by promoting the node below to a root
912 */
913 if (!parent) {
914 struct extent_buffer *child;
915
916 if (btrfs_header_nritems(mid) != 1)
917 return 0;
918
919 /* promote the child to a root */
920 child = read_node_slot(root, mid, 0);
921 BUG_ON(!child);
922 btrfs_tree_lock(child);
923 btrfs_set_lock_blocking(child);
924 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
925 if (ret) {
926 btrfs_tree_unlock(child);
927 free_extent_buffer(child);
928 goto enospc;
929 }
930
931 rcu_assign_pointer(root->node, child);
932
933 add_root_to_dirty_list(root);
934 btrfs_tree_unlock(child);
935
936 path->locks[level] = 0;
937 path->nodes[level] = NULL;
938 clean_tree_block(trans, root, mid);
939 btrfs_tree_unlock(mid);
940 /* once for the path */
941 free_extent_buffer(mid);
942
943 root_sub_used(root, mid->len);
944 btrfs_free_tree_block(trans, root, mid, 0, 1);
945 /* once for the root ptr */
946 free_extent_buffer(mid);
947 return 0;
948 }
949 if (btrfs_header_nritems(mid) >
950 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
951 return 0;
952
953 btrfs_header_nritems(mid);
954
955 left = read_node_slot(root, parent, pslot - 1);
956 if (left) {
957 btrfs_tree_lock(left);
958 btrfs_set_lock_blocking(left);
959 wret = btrfs_cow_block(trans, root, left,
960 parent, pslot - 1, &left);
961 if (wret) {
962 ret = wret;
963 goto enospc;
964 }
965 }
966 right = read_node_slot(root, parent, pslot + 1);
967 if (right) {
968 btrfs_tree_lock(right);
969 btrfs_set_lock_blocking(right);
970 wret = btrfs_cow_block(trans, root, right,
971 parent, pslot + 1, &right);
972 if (wret) {
973 ret = wret;
974 goto enospc;
975 }
976 }
977
978 /* first, try to make some room in the middle buffer */
979 if (left) {
980 orig_slot += btrfs_header_nritems(left);
981 wret = push_node_left(trans, root, left, mid, 1);
982 if (wret < 0)
983 ret = wret;
984 btrfs_header_nritems(mid);
985 }
986
987 /*
988 * then try to empty the right most buffer into the middle
989 */
990 if (right) {
991 wret = push_node_left(trans, root, mid, right, 1);
992 if (wret < 0 && wret != -ENOSPC)
993 ret = wret;
994 if (btrfs_header_nritems(right) == 0) {
995 clean_tree_block(trans, root, right);
996 btrfs_tree_unlock(right);
997 wret = del_ptr(trans, root, path, level + 1, pslot +
998 1);
999 if (wret)
1000 ret = wret;
1001 root_sub_used(root, right->len);
1002 btrfs_free_tree_block(trans, root, right, 0, 1);
1003 free_extent_buffer(right);
1004 right = NULL;
1005 } else {
1006 struct btrfs_disk_key right_key;
1007 btrfs_node_key(right, &right_key, 0);
1008 btrfs_set_node_key(parent, &right_key, pslot + 1);
1009 btrfs_mark_buffer_dirty(parent);
1010 }
1011 }
1012 if (btrfs_header_nritems(mid) == 1) {
1013 /*
1014 * we're not allowed to leave a node with one item in the
1015 * tree during a delete. A deletion from lower in the tree
1016 * could try to delete the only pointer in this node.
1017 * So, pull some keys from the left.
1018 * There has to be a left pointer at this point because
1019 * otherwise we would have pulled some pointers from the
1020 * right
1021 */
1022 BUG_ON(!left);
1023 wret = balance_node_right(trans, root, mid, left);
1024 if (wret < 0) {
1025 ret = wret;
1026 goto enospc;
1027 }
1028 if (wret == 1) {
1029 wret = push_node_left(trans, root, left, mid, 1);
1030 if (wret < 0)
1031 ret = wret;
1032 }
1033 BUG_ON(wret == 1);
1034 }
1035 if (btrfs_header_nritems(mid) == 0) {
1036 clean_tree_block(trans, root, mid);
1037 btrfs_tree_unlock(mid);
1038 wret = del_ptr(trans, root, path, level + 1, pslot);
1039 if (wret)
1040 ret = wret;
1041 root_sub_used(root, mid->len);
1042 btrfs_free_tree_block(trans, root, mid, 0, 1);
1043 free_extent_buffer(mid);
1044 mid = NULL;
1045 } else {
1046 /* update the parent key to reflect our changes */
1047 struct btrfs_disk_key mid_key;
1048 btrfs_node_key(mid, &mid_key, 0);
1049 btrfs_set_node_key(parent, &mid_key, pslot);
1050 btrfs_mark_buffer_dirty(parent);
1051 }
1052
1053 /* update the path */
1054 if (left) {
1055 if (btrfs_header_nritems(left) > orig_slot) {
1056 extent_buffer_get(left);
1057 /* left was locked after cow */
1058 path->nodes[level] = left;
1059 path->slots[level + 1] -= 1;
1060 path->slots[level] = orig_slot;
1061 if (mid) {
1062 btrfs_tree_unlock(mid);
1063 free_extent_buffer(mid);
1064 }
1065 } else {
1066 orig_slot -= btrfs_header_nritems(left);
1067 path->slots[level] = orig_slot;
1068 }
1069 }
1070 /* double check we haven't messed things up */
1071 if (orig_ptr !=
1072 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1073 BUG();
1074enospc:
1075 if (right) {
1076 btrfs_tree_unlock(right);
1077 free_extent_buffer(right);
1078 }
1079 if (left) {
1080 if (path->nodes[level] != left)
1081 btrfs_tree_unlock(left);
1082 free_extent_buffer(left);
1083 }
1084 return ret;
1085}
1086
1087/* Node balancing for insertion. Here we only split or push nodes around
1088 * when they are completely full. This is also done top down, so we
1089 * have to be pessimistic.
1090 */
1091static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1092 struct btrfs_root *root,
1093 struct btrfs_path *path, int level)
1094{
1095 struct extent_buffer *right = NULL;
1096 struct extent_buffer *mid;
1097 struct extent_buffer *left = NULL;
1098 struct extent_buffer *parent = NULL;
1099 int ret = 0;
1100 int wret;
1101 int pslot;
1102 int orig_slot = path->slots[level];
1103
1104 if (level == 0)
1105 return 1;
1106
1107 mid = path->nodes[level];
1108 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1109
1110 if (level < BTRFS_MAX_LEVEL - 1)
1111 parent = path->nodes[level + 1];
1112 pslot = path->slots[level + 1];
1113
1114 if (!parent)
1115 return 1;
1116
1117 left = read_node_slot(root, parent, pslot - 1);
1118
1119 /* first, try to make some room in the middle buffer */
1120 if (left) {
1121 u32 left_nr;
1122
1123 btrfs_tree_lock(left);
1124 btrfs_set_lock_blocking(left);
1125
1126 left_nr = btrfs_header_nritems(left);
1127 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1128 wret = 1;
1129 } else {
1130 ret = btrfs_cow_block(trans, root, left, parent,
1131 pslot - 1, &left);
1132 if (ret)
1133 wret = 1;
1134 else {
1135 wret = push_node_left(trans, root,
1136 left, mid, 0);
1137 }
1138 }
1139 if (wret < 0)
1140 ret = wret;
1141 if (wret == 0) {
1142 struct btrfs_disk_key disk_key;
1143 orig_slot += left_nr;
1144 btrfs_node_key(mid, &disk_key, 0);
1145 btrfs_set_node_key(parent, &disk_key, pslot);
1146 btrfs_mark_buffer_dirty(parent);
1147 if (btrfs_header_nritems(left) > orig_slot) {
1148 path->nodes[level] = left;
1149 path->slots[level + 1] -= 1;
1150 path->slots[level] = orig_slot;
1151 btrfs_tree_unlock(mid);
1152 free_extent_buffer(mid);
1153 } else {
1154 orig_slot -=
1155 btrfs_header_nritems(left);
1156 path->slots[level] = orig_slot;
1157 btrfs_tree_unlock(left);
1158 free_extent_buffer(left);
1159 }
1160 return 0;
1161 }
1162 btrfs_tree_unlock(left);
1163 free_extent_buffer(left);
1164 }
1165 right = read_node_slot(root, parent, pslot + 1);
1166
1167 /*
1168 * then try to empty the right most buffer into the middle
1169 */
1170 if (right) {
1171 u32 right_nr;
1172
1173 btrfs_tree_lock(right);
1174 btrfs_set_lock_blocking(right);
1175
1176 right_nr = btrfs_header_nritems(right);
1177 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1178 wret = 1;
1179 } else {
1180 ret = btrfs_cow_block(trans, root, right,
1181 parent, pslot + 1,
1182 &right);
1183 if (ret)
1184 wret = 1;
1185 else {
1186 wret = balance_node_right(trans, root,
1187 right, mid);
1188 }
1189 }
1190 if (wret < 0)
1191 ret = wret;
1192 if (wret == 0) {
1193 struct btrfs_disk_key disk_key;
1194
1195 btrfs_node_key(right, &disk_key, 0);
1196 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1197 btrfs_mark_buffer_dirty(parent);
1198
1199 if (btrfs_header_nritems(mid) <= orig_slot) {
1200 path->nodes[level] = right;
1201 path->slots[level + 1] += 1;
1202 path->slots[level] = orig_slot -
1203 btrfs_header_nritems(mid);
1204 btrfs_tree_unlock(mid);
1205 free_extent_buffer(mid);
1206 } else {
1207 btrfs_tree_unlock(right);
1208 free_extent_buffer(right);
1209 }
1210 return 0;
1211 }
1212 btrfs_tree_unlock(right);
1213 free_extent_buffer(right);
1214 }
1215 return 1;
1216}
1217
1218/*
1219 * readahead one full node of leaves, finding things that are close
1220 * to the block in 'slot', and triggering ra on them.
1221 */
1222static void reada_for_search(struct btrfs_root *root,
1223 struct btrfs_path *path,
1224 int level, int slot, u64 objectid)
1225{
1226 struct extent_buffer *node;
1227 struct btrfs_disk_key disk_key;
1228 u32 nritems;
1229 u64 search;
1230 u64 target;
1231 u64 nread = 0;
1232 u64 gen;
1233 int direction = path->reada;
1234 struct extent_buffer *eb;
1235 u32 nr;
1236 u32 blocksize;
1237 u32 nscan = 0;
1238
1239 if (level != 1)
1240 return;
1241
1242 if (!path->nodes[level])
1243 return;
1244
1245 node = path->nodes[level];
1246
1247 search = btrfs_node_blockptr(node, slot);
1248 blocksize = btrfs_level_size(root, level - 1);
1249 eb = btrfs_find_tree_block(root, search, blocksize);
1250 if (eb) {
1251 free_extent_buffer(eb);
1252 return;
1253 }
1254
1255 target = search;
1256
1257 nritems = btrfs_header_nritems(node);
1258 nr = slot;
1259
1260 while (1) {
1261 if (direction < 0) {
1262 if (nr == 0)
1263 break;
1264 nr--;
1265 } else if (direction > 0) {
1266 nr++;
1267 if (nr >= nritems)
1268 break;
1269 }
1270 if (path->reada < 0 && objectid) {
1271 btrfs_node_key(node, &disk_key, nr);
1272 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1273 break;
1274 }
1275 search = btrfs_node_blockptr(node, nr);
1276 if ((search <= target && target - search <= 65536) ||
1277 (search > target && search - target <= 65536)) {
1278 gen = btrfs_node_ptr_generation(node, nr);
1279 readahead_tree_block(root, search, blocksize, gen);
1280 nread += blocksize;
1281 }
1282 nscan++;
1283 if ((nread > 65536 || nscan > 32))
1284 break;
1285 }
1286}
1287
1288/*
1289 * returns -EAGAIN if it had to drop the path, or zero if everything was in
1290 * cache
1291 */
1292static noinline int reada_for_balance(struct btrfs_root *root,
1293 struct btrfs_path *path, int level)
1294{
1295 int slot;
1296 int nritems;
1297 struct extent_buffer *parent;
1298 struct extent_buffer *eb;
1299 u64 gen;
1300 u64 block1 = 0;
1301 u64 block2 = 0;
1302 int ret = 0;
1303 int blocksize;
1304
1305 parent = path->nodes[level + 1];
1306 if (!parent)
1307 return 0;
1308
1309 nritems = btrfs_header_nritems(parent);
1310 slot = path->slots[level + 1];
1311 blocksize = btrfs_level_size(root, level);
1312
1313 if (slot > 0) {
1314 block1 = btrfs_node_blockptr(parent, slot - 1);
1315 gen = btrfs_node_ptr_generation(parent, slot - 1);
1316 eb = btrfs_find_tree_block(root, block1, blocksize);
1317 if (eb && btrfs_buffer_uptodate(eb, gen))
1318 block1 = 0;
1319 free_extent_buffer(eb);
1320 }
1321 if (slot + 1 < nritems) {
1322 block2 = btrfs_node_blockptr(parent, slot + 1);
1323 gen = btrfs_node_ptr_generation(parent, slot + 1);
1324 eb = btrfs_find_tree_block(root, block2, blocksize);
1325 if (eb && btrfs_buffer_uptodate(eb, gen))
1326 block2 = 0;
1327 free_extent_buffer(eb);
1328 }
1329 if (block1 || block2) {
1330 ret = -EAGAIN;
1331
1332 /* release the whole path */
1333 btrfs_release_path(path);
1334
1335 /* read the blocks */
1336 if (block1)
1337 readahead_tree_block(root, block1, blocksize, 0);
1338 if (block2)
1339 readahead_tree_block(root, block2, blocksize, 0);
1340
1341 if (block1) {
1342 eb = read_tree_block(root, block1, blocksize, 0);
1343 free_extent_buffer(eb);
1344 }
1345 if (block2) {
1346 eb = read_tree_block(root, block2, blocksize, 0);
1347 free_extent_buffer(eb);
1348 }
1349 }
1350 return ret;
1351}
1352
1353
1354/*
1355 * when we walk down the tree, it is usually safe to unlock the higher layers
1356 * in the tree. The exceptions are when our path goes through slot 0, because
1357 * operations on the tree might require changing key pointers higher up in the
1358 * tree.
1359 *
1360 * callers might also have set path->keep_locks, which tells this code to keep
1361 * the lock if the path points to the last slot in the block. This is part of
1362 * walking through the tree, and selecting the next slot in the higher block.
1363 *
1364 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1365 * if lowest_unlock is 1, level 0 won't be unlocked
1366 */
1367static noinline void unlock_up(struct btrfs_path *path, int level,
1368 int lowest_unlock)
1369{
1370 int i;
1371 int skip_level = level;
1372 int no_skips = 0;
1373 struct extent_buffer *t;
1374
1375 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1376 if (!path->nodes[i])
1377 break;
1378 if (!path->locks[i])
1379 break;
1380 if (!no_skips && path->slots[i] == 0) {
1381 skip_level = i + 1;
1382 continue;
1383 }
1384 if (!no_skips && path->keep_locks) {
1385 u32 nritems;
1386 t = path->nodes[i];
1387 nritems = btrfs_header_nritems(t);
1388 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1389 skip_level = i + 1;
1390 continue;
1391 }
1392 }
1393 if (skip_level < i && i >= lowest_unlock)
1394 no_skips = 1;
1395
1396 t = path->nodes[i];
1397 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1398 btrfs_tree_unlock_rw(t, path->locks[i]);
1399 path->locks[i] = 0;
1400 }
1401 }
1402}
1403
1404/*
1405 * This releases any locks held in the path starting at level and
1406 * going all the way up to the root.
1407 *
1408 * btrfs_search_slot will keep the lock held on higher nodes in a few
1409 * corner cases, such as COW of the block at slot zero in the node. This
1410 * ignores those rules, and it should only be called when there are no
1411 * more updates to be done higher up in the tree.
1412 */
1413noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1414{
1415 int i;
1416
1417 if (path->keep_locks)
1418 return;
1419
1420 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1421 if (!path->nodes[i])
1422 continue;
1423 if (!path->locks[i])
1424 continue;
1425 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
1426 path->locks[i] = 0;
1427 }
1428}
1429
1430/*
1431 * helper function for btrfs_search_slot. The goal is to find a block
1432 * in cache without setting the path to blocking. If we find the block
1433 * we return zero and the path is unchanged.
1434 *
1435 * If we can't find the block, we set the path blocking and do some
1436 * reada. -EAGAIN is returned and the search must be repeated.
1437 */
1438static int
1439read_block_for_search(struct btrfs_trans_handle *trans,
1440 struct btrfs_root *root, struct btrfs_path *p,
1441 struct extent_buffer **eb_ret, int level, int slot,
1442 struct btrfs_key *key)
1443{
1444 u64 blocknr;
1445 u64 gen;
1446 u32 blocksize;
1447 struct extent_buffer *b = *eb_ret;
1448 struct extent_buffer *tmp;
1449 int ret;
1450
1451 blocknr = btrfs_node_blockptr(b, slot);
1452 gen = btrfs_node_ptr_generation(b, slot);
1453 blocksize = btrfs_level_size(root, level - 1);
1454
1455 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1456 if (tmp) {
1457 if (btrfs_buffer_uptodate(tmp, 0)) {
1458 if (btrfs_buffer_uptodate(tmp, gen)) {
1459 /*
1460 * we found an up to date block without
1461 * sleeping, return
1462 * right away
1463 */
1464 *eb_ret = tmp;
1465 return 0;
1466 }
1467 /* the pages were up to date, but we failed
1468 * the generation number check. Do a full
1469 * read for the generation number that is correct.
1470 * We must do this without dropping locks so
1471 * we can trust our generation number
1472 */
1473 free_extent_buffer(tmp);
1474 btrfs_set_path_blocking(p);
1475
1476 tmp = read_tree_block(root, blocknr, blocksize, gen);
1477 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1478 *eb_ret = tmp;
1479 return 0;
1480 }
1481 free_extent_buffer(tmp);
1482 btrfs_release_path(p);
1483 return -EIO;
1484 }
1485 }
1486
1487 /*
1488 * reduce lock contention at high levels
1489 * of the btree by dropping locks before
1490 * we read. Don't release the lock on the current
1491 * level because we need to walk this node to figure
1492 * out which blocks to read.
1493 */
1494 btrfs_unlock_up_safe(p, level + 1);
1495 btrfs_set_path_blocking(p);
1496
1497 free_extent_buffer(tmp);
1498 if (p->reada)
1499 reada_for_search(root, p, level, slot, key->objectid);
1500
1501 btrfs_release_path(p);
1502
1503 ret = -EAGAIN;
1504 tmp = read_tree_block(root, blocknr, blocksize, 0);
1505 if (tmp) {
1506 /*
1507 * If the read above didn't mark this buffer up to date,
1508 * it will never end up being up to date. Set ret to EIO now
1509 * and give up so that our caller doesn't loop forever
1510 * on our EAGAINs.
1511 */
1512 if (!btrfs_buffer_uptodate(tmp, 0))
1513 ret = -EIO;
1514 free_extent_buffer(tmp);
1515 }
1516 return ret;
1517}
1518
1519/*
1520 * helper function for btrfs_search_slot. This does all of the checks
1521 * for node-level blocks and does any balancing required based on
1522 * the ins_len.
1523 *
1524 * If no extra work was required, zero is returned. If we had to
1525 * drop the path, -EAGAIN is returned and btrfs_search_slot must
1526 * start over
1527 */
1528static int
1529setup_nodes_for_search(struct btrfs_trans_handle *trans,
1530 struct btrfs_root *root, struct btrfs_path *p,
1531 struct extent_buffer *b, int level, int ins_len,
1532 int *write_lock_level)
1533{
1534 int ret;
1535 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1536 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1537 int sret;
1538
1539 if (*write_lock_level < level + 1) {
1540 *write_lock_level = level + 1;
1541 btrfs_release_path(p);
1542 goto again;
1543 }
1544
1545 sret = reada_for_balance(root, p, level);
1546 if (sret)
1547 goto again;
1548
1549 btrfs_set_path_blocking(p);
1550 sret = split_node(trans, root, p, level);
1551 btrfs_clear_path_blocking(p, NULL, 0);
1552
1553 BUG_ON(sret > 0);
1554 if (sret) {
1555 ret = sret;
1556 goto done;
1557 }
1558 b = p->nodes[level];
1559 } else if (ins_len < 0 && btrfs_header_nritems(b) <
1560 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
1561 int sret;
1562
1563 if (*write_lock_level < level + 1) {
1564 *write_lock_level = level + 1;
1565 btrfs_release_path(p);
1566 goto again;
1567 }
1568
1569 sret = reada_for_balance(root, p, level);
1570 if (sret)
1571 goto again;
1572
1573 btrfs_set_path_blocking(p);
1574 sret = balance_level(trans, root, p, level);
1575 btrfs_clear_path_blocking(p, NULL, 0);
1576
1577 if (sret) {
1578 ret = sret;
1579 goto done;
1580 }
1581 b = p->nodes[level];
1582 if (!b) {
1583 btrfs_release_path(p);
1584 goto again;
1585 }
1586 BUG_ON(btrfs_header_nritems(b) == 1);
1587 }
1588 return 0;
1589
1590again:
1591 ret = -EAGAIN;
1592done:
1593 return ret;
1594}
1595
1596/*
1597 * look for key in the tree. path is filled in with nodes along the way
1598 * if key is found, we return zero and you can find the item in the leaf
1599 * level of the path (level 0)
1600 *
1601 * If the key isn't found, the path points to the slot where it should
1602 * be inserted, and 1 is returned. If there are other errors during the
1603 * search a negative error number is returned.
1604 *
1605 * if ins_len > 0, nodes and leaves will be split as we walk down the
1606 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1607 * possible)
1608 */
1609int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1610 *root, struct btrfs_key *key, struct btrfs_path *p, int
1611 ins_len, int cow)
1612{
1613 struct extent_buffer *b;
1614 int slot;
1615 int ret;
1616 int err;
1617 int level;
1618 int lowest_unlock = 1;
1619 int root_lock;
1620 /* everything at write_lock_level or lower must be write locked */
1621 int write_lock_level = 0;
1622 u8 lowest_level = 0;
1623
1624 lowest_level = p->lowest_level;
1625 WARN_ON(lowest_level && ins_len > 0);
1626 WARN_ON(p->nodes[0] != NULL);
1627
1628 if (ins_len < 0) {
1629 lowest_unlock = 2;
1630
1631 /* when we are removing items, we might have to go up to level
1632 * two as we update tree pointers Make sure we keep write
1633 * for those levels as well
1634 */
1635 write_lock_level = 2;
1636 } else if (ins_len > 0) {
1637 /*
1638 * for inserting items, make sure we have a write lock on
1639 * level 1 so we can update keys
1640 */
1641 write_lock_level = 1;
1642 }
1643
1644 if (!cow)
1645 write_lock_level = -1;
1646
1647 if (cow && (p->keep_locks || p->lowest_level))
1648 write_lock_level = BTRFS_MAX_LEVEL;
1649
1650again:
1651 /*
1652 * we try very hard to do read locks on the root
1653 */
1654 root_lock = BTRFS_READ_LOCK;
1655 level = 0;
1656 if (p->search_commit_root) {
1657 /*
1658 * the commit roots are read only
1659 * so we always do read locks
1660 */
1661 b = root->commit_root;
1662 extent_buffer_get(b);
1663 level = btrfs_header_level(b);
1664 if (!p->skip_locking)
1665 btrfs_tree_read_lock(b);
1666 } else {
1667 if (p->skip_locking) {
1668 b = btrfs_root_node(root);
1669 level = btrfs_header_level(b);
1670 } else {
1671 /* we don't know the level of the root node
1672 * until we actually have it read locked
1673 */
1674 b = btrfs_read_lock_root_node(root);
1675 level = btrfs_header_level(b);
1676 if (level <= write_lock_level) {
1677 /* whoops, must trade for write lock */
1678 btrfs_tree_read_unlock(b);
1679 free_extent_buffer(b);
1680 b = btrfs_lock_root_node(root);
1681 root_lock = BTRFS_WRITE_LOCK;
1682
1683 /* the level might have changed, check again */
1684 level = btrfs_header_level(b);
1685 }
1686 }
1687 }
1688 p->nodes[level] = b;
1689 if (!p->skip_locking)
1690 p->locks[level] = root_lock;
1691
1692 while (b) {
1693 level = btrfs_header_level(b);
1694
1695 /*
1696 * setup the path here so we can release it under lock
1697 * contention with the cow code
1698 */
1699 if (cow) {
1700 /*
1701 * if we don't really need to cow this block
1702 * then we don't want to set the path blocking,
1703 * so we test it here
1704 */
1705 if (!should_cow_block(trans, root, b))
1706 goto cow_done;
1707
1708 btrfs_set_path_blocking(p);
1709
1710 /*
1711 * must have write locks on this node and the
1712 * parent
1713 */
1714 if (level + 1 > write_lock_level) {
1715 write_lock_level = level + 1;
1716 btrfs_release_path(p);
1717 goto again;
1718 }
1719
1720 err = btrfs_cow_block(trans, root, b,
1721 p->nodes[level + 1],
1722 p->slots[level + 1], &b);
1723 if (err) {
1724 ret = err;
1725 goto done;
1726 }
1727 }
1728cow_done:
1729 BUG_ON(!cow && ins_len);
1730
1731 p->nodes[level] = b;
1732 btrfs_clear_path_blocking(p, NULL, 0);
1733
1734 /*
1735 * we have a lock on b and as long as we aren't changing
1736 * the tree, there is no way to for the items in b to change.
1737 * It is safe to drop the lock on our parent before we
1738 * go through the expensive btree search on b.
1739 *
1740 * If cow is true, then we might be changing slot zero,
1741 * which may require changing the parent. So, we can't
1742 * drop the lock until after we know which slot we're
1743 * operating on.
1744 */
1745 if (!cow)
1746 btrfs_unlock_up_safe(p, level + 1);
1747
1748 ret = bin_search(b, key, level, &slot);
1749
1750 if (level != 0) {
1751 int dec = 0;
1752 if (ret && slot > 0) {
1753 dec = 1;
1754 slot -= 1;
1755 }
1756 p->slots[level] = slot;
1757 err = setup_nodes_for_search(trans, root, p, b, level,
1758 ins_len, &write_lock_level);
1759 if (err == -EAGAIN)
1760 goto again;
1761 if (err) {
1762 ret = err;
1763 goto done;
1764 }
1765 b = p->nodes[level];
1766 slot = p->slots[level];
1767
1768 /*
1769 * slot 0 is special, if we change the key
1770 * we have to update the parent pointer
1771 * which means we must have a write lock
1772 * on the parent
1773 */
1774 if (slot == 0 && cow &&
1775 write_lock_level < level + 1) {
1776 write_lock_level = level + 1;
1777 btrfs_release_path(p);
1778 goto again;
1779 }
1780
1781 unlock_up(p, level, lowest_unlock);
1782
1783 if (level == lowest_level) {
1784 if (dec)
1785 p->slots[level]++;
1786 goto done;
1787 }
1788
1789 err = read_block_for_search(trans, root, p,
1790 &b, level, slot, key);
1791 if (err == -EAGAIN)
1792 goto again;
1793 if (err) {
1794 ret = err;
1795 goto done;
1796 }
1797
1798 if (!p->skip_locking) {
1799 level = btrfs_header_level(b);
1800 if (level <= write_lock_level) {
1801 err = btrfs_try_tree_write_lock(b);
1802 if (!err) {
1803 btrfs_set_path_blocking(p);
1804 btrfs_tree_lock(b);
1805 btrfs_clear_path_blocking(p, b,
1806 BTRFS_WRITE_LOCK);
1807 }
1808 p->locks[level] = BTRFS_WRITE_LOCK;
1809 } else {
1810 err = btrfs_try_tree_read_lock(b);
1811 if (!err) {
1812 btrfs_set_path_blocking(p);
1813 btrfs_tree_read_lock(b);
1814 btrfs_clear_path_blocking(p, b,
1815 BTRFS_READ_LOCK);
1816 }
1817 p->locks[level] = BTRFS_READ_LOCK;
1818 }
1819 p->nodes[level] = b;
1820 }
1821 } else {
1822 p->slots[level] = slot;
1823 if (ins_len > 0 &&
1824 btrfs_leaf_free_space(root, b) < ins_len) {
1825 if (write_lock_level < 1) {
1826 write_lock_level = 1;
1827 btrfs_release_path(p);
1828 goto again;
1829 }
1830
1831 btrfs_set_path_blocking(p);
1832 err = split_leaf(trans, root, key,
1833 p, ins_len, ret == 0);
1834 btrfs_clear_path_blocking(p, NULL, 0);
1835
1836 BUG_ON(err > 0);
1837 if (err) {
1838 ret = err;
1839 goto done;
1840 }
1841 }
1842 if (!p->search_for_split)
1843 unlock_up(p, level, lowest_unlock);
1844 goto done;
1845 }
1846 }
1847 ret = 1;
1848done:
1849 /*
1850 * we don't really know what they plan on doing with the path
1851 * from here on, so for now just mark it as blocking
1852 */
1853 if (!p->leave_spinning)
1854 btrfs_set_path_blocking(p);
1855 if (ret < 0)
1856 btrfs_release_path(p);
1857 return ret;
1858}
1859
1860/*
1861 * adjust the pointers going up the tree, starting at level
1862 * making sure the right key of each node is points to 'key'.
1863 * This is used after shifting pointers to the left, so it stops
1864 * fixing up pointers when a given leaf/node is not in slot 0 of the
1865 * higher levels
1866 *
1867 * If this fails to write a tree block, it returns -1, but continues
1868 * fixing up the blocks in ram so the tree is consistent.
1869 */
1870static int fixup_low_keys(struct btrfs_trans_handle *trans,
1871 struct btrfs_root *root, struct btrfs_path *path,
1872 struct btrfs_disk_key *key, int level)
1873{
1874 int i;
1875 int ret = 0;
1876 struct extent_buffer *t;
1877
1878 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1879 int tslot = path->slots[i];
1880 if (!path->nodes[i])
1881 break;
1882 t = path->nodes[i];
1883 btrfs_set_node_key(t, key, tslot);
1884 btrfs_mark_buffer_dirty(path->nodes[i]);
1885 if (tslot != 0)
1886 break;
1887 }
1888 return ret;
1889}
1890
1891/*
1892 * update item key.
1893 *
1894 * This function isn't completely safe. It's the caller's responsibility
1895 * that the new key won't break the order
1896 */
1897int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1898 struct btrfs_root *root, struct btrfs_path *path,
1899 struct btrfs_key *new_key)
1900{
1901 struct btrfs_disk_key disk_key;
1902 struct extent_buffer *eb;
1903 int slot;
1904
1905 eb = path->nodes[0];
1906 slot = path->slots[0];
1907 if (slot > 0) {
1908 btrfs_item_key(eb, &disk_key, slot - 1);
1909 if (comp_keys(&disk_key, new_key) >= 0)
1910 return -1;
1911 }
1912 if (slot < btrfs_header_nritems(eb) - 1) {
1913 btrfs_item_key(eb, &disk_key, slot + 1);
1914 if (comp_keys(&disk_key, new_key) <= 0)
1915 return -1;
1916 }
1917
1918 btrfs_cpu_key_to_disk(&disk_key, new_key);
1919 btrfs_set_item_key(eb, &disk_key, slot);
1920 btrfs_mark_buffer_dirty(eb);
1921 if (slot == 0)
1922 fixup_low_keys(trans, root, path, &disk_key, 1);
1923 return 0;
1924}
1925
1926/*
1927 * try to push data from one node into the next node left in the
1928 * tree.
1929 *
1930 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1931 * error, and > 0 if there was no room in the left hand block.
1932 */
1933static int push_node_left(struct btrfs_trans_handle *trans,
1934 struct btrfs_root *root, struct extent_buffer *dst,
1935 struct extent_buffer *src, int empty)
1936{
1937 int push_items = 0;
1938 int src_nritems;
1939 int dst_nritems;
1940 int ret = 0;
1941
1942 src_nritems = btrfs_header_nritems(src);
1943 dst_nritems = btrfs_header_nritems(dst);
1944 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1945 WARN_ON(btrfs_header_generation(src) != trans->transid);
1946 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1947
1948 if (!empty && src_nritems <= 8)
1949 return 1;
1950
1951 if (push_items <= 0)
1952 return 1;
1953
1954 if (empty) {
1955 push_items = min(src_nritems, push_items);
1956 if (push_items < src_nritems) {
1957 /* leave at least 8 pointers in the node if
1958 * we aren't going to empty it
1959 */
1960 if (src_nritems - push_items < 8) {
1961 if (push_items <= 8)
1962 return 1;
1963 push_items -= 8;
1964 }
1965 }
1966 } else
1967 push_items = min(src_nritems - 8, push_items);
1968
1969 copy_extent_buffer(dst, src,
1970 btrfs_node_key_ptr_offset(dst_nritems),
1971 btrfs_node_key_ptr_offset(0),
1972 push_items * sizeof(struct btrfs_key_ptr));
1973
1974 if (push_items < src_nritems) {
1975 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1976 btrfs_node_key_ptr_offset(push_items),
1977 (src_nritems - push_items) *
1978 sizeof(struct btrfs_key_ptr));
1979 }
1980 btrfs_set_header_nritems(src, src_nritems - push_items);
1981 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1982 btrfs_mark_buffer_dirty(src);
1983 btrfs_mark_buffer_dirty(dst);
1984
1985 return ret;
1986}
1987
1988/*
1989 * try to push data from one node into the next node right in the
1990 * tree.
1991 *
1992 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1993 * error, and > 0 if there was no room in the right hand block.
1994 *
1995 * this will only push up to 1/2 the contents of the left node over
1996 */
1997static int balance_node_right(struct btrfs_trans_handle *trans,
1998 struct btrfs_root *root,
1999 struct extent_buffer *dst,
2000 struct extent_buffer *src)
2001{
2002 int push_items = 0;
2003 int max_push;
2004 int src_nritems;
2005 int dst_nritems;
2006 int ret = 0;
2007
2008 WARN_ON(btrfs_header_generation(src) != trans->transid);
2009 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2010
2011 src_nritems = btrfs_header_nritems(src);
2012 dst_nritems = btrfs_header_nritems(dst);
2013 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2014 if (push_items <= 0)
2015 return 1;
2016
2017 if (src_nritems < 4)
2018 return 1;
2019
2020 max_push = src_nritems / 2 + 1;
2021 /* don't try to empty the node */
2022 if (max_push >= src_nritems)
2023 return 1;
2024
2025 if (max_push < push_items)
2026 push_items = max_push;
2027
2028 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2029 btrfs_node_key_ptr_offset(0),
2030 (dst_nritems) *
2031 sizeof(struct btrfs_key_ptr));
2032
2033 copy_extent_buffer(dst, src,
2034 btrfs_node_key_ptr_offset(0),
2035 btrfs_node_key_ptr_offset(src_nritems - push_items),
2036 push_items * sizeof(struct btrfs_key_ptr));
2037
2038 btrfs_set_header_nritems(src, src_nritems - push_items);
2039 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2040
2041 btrfs_mark_buffer_dirty(src);
2042 btrfs_mark_buffer_dirty(dst);
2043
2044 return ret;
2045}
2046
2047/*
2048 * helper function to insert a new root level in the tree.
2049 * A new node is allocated, and a single item is inserted to
2050 * point to the existing root
2051 *
2052 * returns zero on success or < 0 on failure.
2053 */
2054static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2055 struct btrfs_root *root,
2056 struct btrfs_path *path, int level)
2057{
2058 u64 lower_gen;
2059 struct extent_buffer *lower;
2060 struct extent_buffer *c;
2061 struct extent_buffer *old;
2062 struct btrfs_disk_key lower_key;
2063
2064 BUG_ON(path->nodes[level]);
2065 BUG_ON(path->nodes[level-1] != root->node);
2066
2067 lower = path->nodes[level-1];
2068 if (level == 1)
2069 btrfs_item_key(lower, &lower_key, 0);
2070 else
2071 btrfs_node_key(lower, &lower_key, 0);
2072
2073 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2074 root->root_key.objectid, &lower_key,
2075 level, root->node->start, 0);
2076 if (IS_ERR(c))
2077 return PTR_ERR(c);
2078
2079 root_add_used(root, root->nodesize);
2080
2081 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
2082 btrfs_set_header_nritems(c, 1);
2083 btrfs_set_header_level(c, level);
2084 btrfs_set_header_bytenr(c, c->start);
2085 btrfs_set_header_generation(c, trans->transid);
2086 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
2087 btrfs_set_header_owner(c, root->root_key.objectid);
2088
2089 write_extent_buffer(c, root->fs_info->fsid,
2090 (unsigned long)btrfs_header_fsid(c),
2091 BTRFS_FSID_SIZE);
2092
2093 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2094 (unsigned long)btrfs_header_chunk_tree_uuid(c),
2095 BTRFS_UUID_SIZE);
2096
2097 btrfs_set_node_key(c, &lower_key, 0);
2098 btrfs_set_node_blockptr(c, 0, lower->start);
2099 lower_gen = btrfs_header_generation(lower);
2100 WARN_ON(lower_gen != trans->transid);
2101
2102 btrfs_set_node_ptr_generation(c, 0, lower_gen);
2103
2104 btrfs_mark_buffer_dirty(c);
2105
2106 old = root->node;
2107 rcu_assign_pointer(root->node, c);
2108
2109 /* the super has an extra ref to root->node */
2110 free_extent_buffer(old);
2111
2112 add_root_to_dirty_list(root);
2113 extent_buffer_get(c);
2114 path->nodes[level] = c;
2115 path->locks[level] = BTRFS_WRITE_LOCK;
2116 path->slots[level] = 0;
2117 return 0;
2118}
2119
2120/*
2121 * worker function to insert a single pointer in a node.
2122 * the node should have enough room for the pointer already
2123 *
2124 * slot and level indicate where you want the key to go, and
2125 * blocknr is the block the key points to.
2126 *
2127 * returns zero on success and < 0 on any error
2128 */
2129static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2130 *root, struct btrfs_path *path, struct btrfs_disk_key
2131 *key, u64 bytenr, int slot, int level)
2132{
2133 struct extent_buffer *lower;
2134 int nritems;
2135
2136 BUG_ON(!path->nodes[level]);
2137 btrfs_assert_tree_locked(path->nodes[level]);
2138 lower = path->nodes[level];
2139 nritems = btrfs_header_nritems(lower);
2140 BUG_ON(slot > nritems);
2141 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2142 BUG();
2143 if (slot != nritems) {
2144 memmove_extent_buffer(lower,
2145 btrfs_node_key_ptr_offset(slot + 1),
2146 btrfs_node_key_ptr_offset(slot),
2147 (nritems - slot) * sizeof(struct btrfs_key_ptr));
2148 }
2149 btrfs_set_node_key(lower, key, slot);
2150 btrfs_set_node_blockptr(lower, slot, bytenr);
2151 WARN_ON(trans->transid == 0);
2152 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2153 btrfs_set_header_nritems(lower, nritems + 1);
2154 btrfs_mark_buffer_dirty(lower);
2155 return 0;
2156}
2157
2158/*
2159 * split the node at the specified level in path in two.
2160 * The path is corrected to point to the appropriate node after the split
2161 *
2162 * Before splitting this tries to make some room in the node by pushing
2163 * left and right, if either one works, it returns right away.
2164 *
2165 * returns 0 on success and < 0 on failure
2166 */
2167static noinline int split_node(struct btrfs_trans_handle *trans,
2168 struct btrfs_root *root,
2169 struct btrfs_path *path, int level)
2170{
2171 struct extent_buffer *c;
2172 struct extent_buffer *split;
2173 struct btrfs_disk_key disk_key;
2174 int mid;
2175 int ret;
2176 int wret;
2177 u32 c_nritems;
2178
2179 c = path->nodes[level];
2180 WARN_ON(btrfs_header_generation(c) != trans->transid);
2181 if (c == root->node) {
2182 /* trying to split the root, lets make a new one */
2183 ret = insert_new_root(trans, root, path, level + 1);
2184 if (ret)
2185 return ret;
2186 } else {
2187 ret = push_nodes_for_insert(trans, root, path, level);
2188 c = path->nodes[level];
2189 if (!ret && btrfs_header_nritems(c) <
2190 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2191 return 0;
2192 if (ret < 0)
2193 return ret;
2194 }
2195
2196 c_nritems = btrfs_header_nritems(c);
2197 mid = (c_nritems + 1) / 2;
2198 btrfs_node_key(c, &disk_key, mid);
2199
2200 split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2201 root->root_key.objectid,
2202 &disk_key, level, c->start, 0);
2203 if (IS_ERR(split))
2204 return PTR_ERR(split);
2205
2206 root_add_used(root, root->nodesize);
2207
2208 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
2209 btrfs_set_header_level(split, btrfs_header_level(c));
2210 btrfs_set_header_bytenr(split, split->start);
2211 btrfs_set_header_generation(split, trans->transid);
2212 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
2213 btrfs_set_header_owner(split, root->root_key.objectid);
2214 write_extent_buffer(split, root->fs_info->fsid,
2215 (unsigned long)btrfs_header_fsid(split),
2216 BTRFS_FSID_SIZE);
2217 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2218 (unsigned long)btrfs_header_chunk_tree_uuid(split),
2219 BTRFS_UUID_SIZE);
2220
2221
2222 copy_extent_buffer(split, c,
2223 btrfs_node_key_ptr_offset(0),
2224 btrfs_node_key_ptr_offset(mid),
2225 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2226 btrfs_set_header_nritems(split, c_nritems - mid);
2227 btrfs_set_header_nritems(c, mid);
2228 ret = 0;
2229
2230 btrfs_mark_buffer_dirty(c);
2231 btrfs_mark_buffer_dirty(split);
2232
2233 wret = insert_ptr(trans, root, path, &disk_key, split->start,
2234 path->slots[level + 1] + 1,
2235 level + 1);
2236 if (wret)
2237 ret = wret;
2238
2239 if (path->slots[level] >= mid) {
2240 path->slots[level] -= mid;
2241 btrfs_tree_unlock(c);
2242 free_extent_buffer(c);
2243 path->nodes[level] = split;
2244 path->slots[level + 1] += 1;
2245 } else {
2246 btrfs_tree_unlock(split);
2247 free_extent_buffer(split);
2248 }
2249 return ret;
2250}
2251
2252/*
2253 * how many bytes are required to store the items in a leaf. start
2254 * and nr indicate which items in the leaf to check. This totals up the
2255 * space used both by the item structs and the item data
2256 */
2257static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2258{
2259 int data_len;
2260 int nritems = btrfs_header_nritems(l);
2261 int end = min(nritems, start + nr) - 1;
2262
2263 if (!nr)
2264 return 0;
2265 data_len = btrfs_item_end_nr(l, start);
2266 data_len = data_len - btrfs_item_offset_nr(l, end);
2267 data_len += sizeof(struct btrfs_item) * nr;
2268 WARN_ON(data_len < 0);
2269 return data_len;
2270}
2271
2272/*
2273 * The space between the end of the leaf items and
2274 * the start of the leaf data. IOW, how much room
2275 * the leaf has left for both items and data
2276 */
2277noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2278 struct extent_buffer *leaf)
2279{
2280 int nritems = btrfs_header_nritems(leaf);
2281 int ret;
2282 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2283 if (ret < 0) {
2284 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2285 "used %d nritems %d\n",
2286 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2287 leaf_space_used(leaf, 0, nritems), nritems);
2288 }
2289 return ret;
2290}
2291
2292/*
2293 * min slot controls the lowest index we're willing to push to the
2294 * right. We'll push up to and including min_slot, but no lower
2295 */
2296static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2297 struct btrfs_root *root,
2298 struct btrfs_path *path,
2299 int data_size, int empty,
2300 struct extent_buffer *right,
2301 int free_space, u32 left_nritems,
2302 u32 min_slot)
2303{
2304 struct extent_buffer *left = path->nodes[0];
2305 struct extent_buffer *upper = path->nodes[1];
2306 struct btrfs_disk_key disk_key;
2307 int slot;
2308 u32 i;
2309 int push_space = 0;
2310 int push_items = 0;
2311 struct btrfs_item *item;
2312 u32 nr;
2313 u32 right_nritems;
2314 u32 data_end;
2315 u32 this_item_size;
2316
2317 if (empty)
2318 nr = 0;
2319 else
2320 nr = max_t(u32, 1, min_slot);
2321
2322 if (path->slots[0] >= left_nritems)
2323 push_space += data_size;
2324
2325 slot = path->slots[1];
2326 i = left_nritems - 1;
2327 while (i >= nr) {
2328 item = btrfs_item_nr(left, i);
2329
2330 if (!empty && push_items > 0) {
2331 if (path->slots[0] > i)
2332 break;
2333 if (path->slots[0] == i) {
2334 int space = btrfs_leaf_free_space(root, left);
2335 if (space + push_space * 2 > free_space)
2336 break;
2337 }
2338 }
2339
2340 if (path->slots[0] == i)
2341 push_space += data_size;
2342
2343 this_item_size = btrfs_item_size(left, item);
2344 if (this_item_size + sizeof(*item) + push_space > free_space)
2345 break;
2346
2347 push_items++;
2348 push_space += this_item_size + sizeof(*item);
2349 if (i == 0)
2350 break;
2351 i--;
2352 }
2353
2354 if (push_items == 0)
2355 goto out_unlock;
2356
2357 if (!empty && push_items == left_nritems)
2358 WARN_ON(1);
2359
2360 /* push left to right */
2361 right_nritems = btrfs_header_nritems(right);
2362
2363 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2364 push_space -= leaf_data_end(root, left);
2365
2366 /* make room in the right data area */
2367 data_end = leaf_data_end(root, right);
2368 memmove_extent_buffer(right,
2369 btrfs_leaf_data(right) + data_end - push_space,
2370 btrfs_leaf_data(right) + data_end,
2371 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2372
2373 /* copy from the left data area */
2374 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2375 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2376 btrfs_leaf_data(left) + leaf_data_end(root, left),
2377 push_space);
2378
2379 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2380 btrfs_item_nr_offset(0),
2381 right_nritems * sizeof(struct btrfs_item));
2382
2383 /* copy the items from left to right */
2384 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2385 btrfs_item_nr_offset(left_nritems - push_items),
2386 push_items * sizeof(struct btrfs_item));
2387
2388 /* update the item pointers */
2389 right_nritems += push_items;
2390 btrfs_set_header_nritems(right, right_nritems);
2391 push_space = BTRFS_LEAF_DATA_SIZE(root);
2392 for (i = 0; i < right_nritems; i++) {
2393 item = btrfs_item_nr(right, i);
2394 push_space -= btrfs_item_size(right, item);
2395 btrfs_set_item_offset(right, item, push_space);
2396 }
2397
2398 left_nritems -= push_items;
2399 btrfs_set_header_nritems(left, left_nritems);
2400
2401 if (left_nritems)
2402 btrfs_mark_buffer_dirty(left);
2403 else
2404 clean_tree_block(trans, root, left);
2405
2406 btrfs_mark_buffer_dirty(right);
2407
2408 btrfs_item_key(right, &disk_key, 0);
2409 btrfs_set_node_key(upper, &disk_key, slot + 1);
2410 btrfs_mark_buffer_dirty(upper);
2411
2412 /* then fixup the leaf pointer in the path */
2413 if (path->slots[0] >= left_nritems) {
2414 path->slots[0] -= left_nritems;
2415 if (btrfs_header_nritems(path->nodes[0]) == 0)
2416 clean_tree_block(trans, root, path->nodes[0]);
2417 btrfs_tree_unlock(path->nodes[0]);
2418 free_extent_buffer(path->nodes[0]);
2419 path->nodes[0] = right;
2420 path->slots[1] += 1;
2421 } else {
2422 btrfs_tree_unlock(right);
2423 free_extent_buffer(right);
2424 }
2425 return 0;
2426
2427out_unlock:
2428 btrfs_tree_unlock(right);
2429 free_extent_buffer(right);
2430 return 1;
2431}
2432
2433/*
2434 * push some data in the path leaf to the right, trying to free up at
2435 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2436 *
2437 * returns 1 if the push failed because the other node didn't have enough
2438 * room, 0 if everything worked out and < 0 if there were major errors.
2439 *
2440 * this will push starting from min_slot to the end of the leaf. It won't
2441 * push any slot lower than min_slot
2442 */
2443static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2444 *root, struct btrfs_path *path,
2445 int min_data_size, int data_size,
2446 int empty, u32 min_slot)
2447{
2448 struct extent_buffer *left = path->nodes[0];
2449 struct extent_buffer *right;
2450 struct extent_buffer *upper;
2451 int slot;
2452 int free_space;
2453 u32 left_nritems;
2454 int ret;
2455
2456 if (!path->nodes[1])
2457 return 1;
2458
2459 slot = path->slots[1];
2460 upper = path->nodes[1];
2461 if (slot >= btrfs_header_nritems(upper) - 1)
2462 return 1;
2463
2464 btrfs_assert_tree_locked(path->nodes[1]);
2465
2466 right = read_node_slot(root, upper, slot + 1);
2467 if (right == NULL)
2468 return 1;
2469
2470 btrfs_tree_lock(right);
2471 btrfs_set_lock_blocking(right);
2472
2473 free_space = btrfs_leaf_free_space(root, right);
2474 if (free_space < data_size)
2475 goto out_unlock;
2476
2477 /* cow and double check */
2478 ret = btrfs_cow_block(trans, root, right, upper,
2479 slot + 1, &right);
2480 if (ret)
2481 goto out_unlock;
2482
2483 free_space = btrfs_leaf_free_space(root, right);
2484 if (free_space < data_size)
2485 goto out_unlock;
2486
2487 left_nritems = btrfs_header_nritems(left);
2488 if (left_nritems == 0)
2489 goto out_unlock;
2490
2491 return __push_leaf_right(trans, root, path, min_data_size, empty,
2492 right, free_space, left_nritems, min_slot);
2493out_unlock:
2494 btrfs_tree_unlock(right);
2495 free_extent_buffer(right);
2496 return 1;
2497}
2498
2499/*
2500 * push some data in the path leaf to the left, trying to free up at
2501 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2502 *
2503 * max_slot can put a limit on how far into the leaf we'll push items. The
2504 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
2505 * items
2506 */
2507static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2508 struct btrfs_root *root,
2509 struct btrfs_path *path, int data_size,
2510 int empty, struct extent_buffer *left,
2511 int free_space, u32 right_nritems,
2512 u32 max_slot)
2513{
2514 struct btrfs_disk_key disk_key;
2515 struct extent_buffer *right = path->nodes[0];
2516 int i;
2517 int push_space = 0;
2518 int push_items = 0;
2519 struct btrfs_item *item;
2520 u32 old_left_nritems;
2521 u32 nr;
2522 int ret = 0;
2523 int wret;
2524 u32 this_item_size;
2525 u32 old_left_item_size;
2526
2527 if (empty)
2528 nr = min(right_nritems, max_slot);
2529 else
2530 nr = min(right_nritems - 1, max_slot);
2531
2532 for (i = 0; i < nr; i++) {
2533 item = btrfs_item_nr(right, i);
2534
2535 if (!empty && push_items > 0) {
2536 if (path->slots[0] < i)
2537 break;
2538 if (path->slots[0] == i) {
2539 int space = btrfs_leaf_free_space(root, right);
2540 if (space + push_space * 2 > free_space)
2541 break;
2542 }
2543 }
2544
2545 if (path->slots[0] == i)
2546 push_space += data_size;
2547
2548 this_item_size = btrfs_item_size(right, item);
2549 if (this_item_size + sizeof(*item) + push_space > free_space)
2550 break;
2551
2552 push_items++;
2553 push_space += this_item_size + sizeof(*item);
2554 }
2555
2556 if (push_items == 0) {
2557 ret = 1;
2558 goto out;
2559 }
2560 if (!empty && push_items == btrfs_header_nritems(right))
2561 WARN_ON(1);
2562
2563 /* push data from right to left */
2564 copy_extent_buffer(left, right,
2565 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2566 btrfs_item_nr_offset(0),
2567 push_items * sizeof(struct btrfs_item));
2568
2569 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2570 btrfs_item_offset_nr(right, push_items - 1);
2571
2572 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2573 leaf_data_end(root, left) - push_space,
2574 btrfs_leaf_data(right) +
2575 btrfs_item_offset_nr(right, push_items - 1),
2576 push_space);
2577 old_left_nritems = btrfs_header_nritems(left);
2578 BUG_ON(old_left_nritems <= 0);
2579
2580 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2581 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2582 u32 ioff;
2583
2584 item = btrfs_item_nr(left, i);
2585
2586 ioff = btrfs_item_offset(left, item);
2587 btrfs_set_item_offset(left, item,
2588 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2589 }
2590 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2591
2592 /* fixup right node */
2593 if (push_items > right_nritems) {
2594 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2595 right_nritems);
2596 WARN_ON(1);
2597 }
2598
2599 if (push_items < right_nritems) {
2600 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2601 leaf_data_end(root, right);
2602 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2603 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2604 btrfs_leaf_data(right) +
2605 leaf_data_end(root, right), push_space);
2606
2607 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2608 btrfs_item_nr_offset(push_items),
2609 (btrfs_header_nritems(right) - push_items) *
2610 sizeof(struct btrfs_item));
2611 }
2612 right_nritems -= push_items;
2613 btrfs_set_header_nritems(right, right_nritems);
2614 push_space = BTRFS_LEAF_DATA_SIZE(root);
2615 for (i = 0; i < right_nritems; i++) {
2616 item = btrfs_item_nr(right, i);
2617
2618 push_space = push_space - btrfs_item_size(right, item);
2619 btrfs_set_item_offset(right, item, push_space);
2620 }
2621
2622 btrfs_mark_buffer_dirty(left);
2623 if (right_nritems)
2624 btrfs_mark_buffer_dirty(right);
2625 else
2626 clean_tree_block(trans, root, right);
2627
2628 btrfs_item_key(right, &disk_key, 0);
2629 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2630 if (wret)
2631 ret = wret;
2632
2633 /* then fixup the leaf pointer in the path */
2634 if (path->slots[0] < push_items) {
2635 path->slots[0] += old_left_nritems;
2636 btrfs_tree_unlock(path->nodes[0]);
2637 free_extent_buffer(path->nodes[0]);
2638 path->nodes[0] = left;
2639 path->slots[1] -= 1;
2640 } else {
2641 btrfs_tree_unlock(left);
2642 free_extent_buffer(left);
2643 path->slots[0] -= push_items;
2644 }
2645 BUG_ON(path->slots[0] < 0);
2646 return ret;
2647out:
2648 btrfs_tree_unlock(left);
2649 free_extent_buffer(left);
2650 return ret;
2651}
2652
2653/*
2654 * push some data in the path leaf to the left, trying to free up at
2655 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2656 *
2657 * max_slot can put a limit on how far into the leaf we'll push items. The
2658 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
2659 * items
2660 */
2661static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2662 *root, struct btrfs_path *path, int min_data_size,
2663 int data_size, int empty, u32 max_slot)
2664{
2665 struct extent_buffer *right = path->nodes[0];
2666 struct extent_buffer *left;
2667 int slot;
2668 int free_space;
2669 u32 right_nritems;
2670 int ret = 0;
2671
2672 slot = path->slots[1];
2673 if (slot == 0)
2674 return 1;
2675 if (!path->nodes[1])
2676 return 1;
2677
2678 right_nritems = btrfs_header_nritems(right);
2679 if (right_nritems == 0)
2680 return 1;
2681
2682 btrfs_assert_tree_locked(path->nodes[1]);
2683
2684 left = read_node_slot(root, path->nodes[1], slot - 1);
2685 if (left == NULL)
2686 return 1;
2687
2688 btrfs_tree_lock(left);
2689 btrfs_set_lock_blocking(left);
2690
2691 free_space = btrfs_leaf_free_space(root, left);
2692 if (free_space < data_size) {
2693 ret = 1;
2694 goto out;
2695 }
2696
2697 /* cow and double check */
2698 ret = btrfs_cow_block(trans, root, left,
2699 path->nodes[1], slot - 1, &left);
2700 if (ret) {
2701 /* we hit -ENOSPC, but it isn't fatal here */
2702 ret = 1;
2703 goto out;
2704 }
2705
2706 free_space = btrfs_leaf_free_space(root, left);
2707 if (free_space < data_size) {
2708 ret = 1;
2709 goto out;
2710 }
2711
2712 return __push_leaf_left(trans, root, path, min_data_size,
2713 empty, left, free_space, right_nritems,
2714 max_slot);
2715out:
2716 btrfs_tree_unlock(left);
2717 free_extent_buffer(left);
2718 return ret;
2719}
2720
2721/*
2722 * split the path's leaf in two, making sure there is at least data_size
2723 * available for the resulting leaf level of the path.
2724 *
2725 * returns 0 if all went well and < 0 on failure.
2726 */
2727static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2728 struct btrfs_root *root,
2729 struct btrfs_path *path,
2730 struct extent_buffer *l,
2731 struct extent_buffer *right,
2732 int slot, int mid, int nritems)
2733{
2734 int data_copy_size;
2735 int rt_data_off;
2736 int i;
2737 int ret = 0;
2738 int wret;
2739 struct btrfs_disk_key disk_key;
2740
2741 nritems = nritems - mid;
2742 btrfs_set_header_nritems(right, nritems);
2743 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2744
2745 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2746 btrfs_item_nr_offset(mid),
2747 nritems * sizeof(struct btrfs_item));
2748
2749 copy_extent_buffer(right, l,
2750 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2751 data_copy_size, btrfs_leaf_data(l) +
2752 leaf_data_end(root, l), data_copy_size);
2753
2754 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2755 btrfs_item_end_nr(l, mid);
2756
2757 for (i = 0; i < nritems; i++) {
2758 struct btrfs_item *item = btrfs_item_nr(right, i);
2759 u32 ioff;
2760
2761 ioff = btrfs_item_offset(right, item);
2762 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2763 }
2764
2765 btrfs_set_header_nritems(l, mid);
2766 ret = 0;
2767 btrfs_item_key(right, &disk_key, 0);
2768 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2769 path->slots[1] + 1, 1);
2770 if (wret)
2771 ret = wret;
2772
2773 btrfs_mark_buffer_dirty(right);
2774 btrfs_mark_buffer_dirty(l);
2775 BUG_ON(path->slots[0] != slot);
2776
2777 if (mid <= slot) {
2778 btrfs_tree_unlock(path->nodes[0]);
2779 free_extent_buffer(path->nodes[0]);
2780 path->nodes[0] = right;
2781 path->slots[0] -= mid;
2782 path->slots[1] += 1;
2783 } else {
2784 btrfs_tree_unlock(right);
2785 free_extent_buffer(right);
2786 }
2787
2788 BUG_ON(path->slots[0] < 0);
2789
2790 return ret;
2791}
2792
2793/*
2794 * double splits happen when we need to insert a big item in the middle
2795 * of a leaf. A double split can leave us with 3 mostly empty leaves:
2796 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
2797 * A B C
2798 *
2799 * We avoid this by trying to push the items on either side of our target
2800 * into the adjacent leaves. If all goes well we can avoid the double split
2801 * completely.
2802 */
2803static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
2804 struct btrfs_root *root,
2805 struct btrfs_path *path,
2806 int data_size)
2807{
2808 int ret;
2809 int progress = 0;
2810 int slot;
2811 u32 nritems;
2812
2813 slot = path->slots[0];
2814
2815 /*
2816 * try to push all the items after our slot into the
2817 * right leaf
2818 */
2819 ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot);
2820 if (ret < 0)
2821 return ret;
2822
2823 if (ret == 0)
2824 progress++;
2825
2826 nritems = btrfs_header_nritems(path->nodes[0]);
2827 /*
2828 * our goal is to get our slot at the start or end of a leaf. If
2829 * we've done so we're done
2830 */
2831 if (path->slots[0] == 0 || path->slots[0] == nritems)
2832 return 0;
2833
2834 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
2835 return 0;
2836
2837 /* try to push all the items before our slot into the next leaf */
2838 slot = path->slots[0];
2839 ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot);
2840 if (ret < 0)
2841 return ret;
2842
2843 if (ret == 0)
2844 progress++;
2845
2846 if (progress)
2847 return 0;
2848 return 1;
2849}
2850
2851/*
2852 * split the path's leaf in two, making sure there is at least data_size
2853 * available for the resulting leaf level of the path.
2854 *
2855 * returns 0 if all went well and < 0 on failure.
2856 */
2857static noinline int split_leaf(struct btrfs_trans_handle *trans,
2858 struct btrfs_root *root,
2859 struct btrfs_key *ins_key,
2860 struct btrfs_path *path, int data_size,
2861 int extend)
2862{
2863 struct btrfs_disk_key disk_key;
2864 struct extent_buffer *l;
2865 u32 nritems;
2866 int mid;
2867 int slot;
2868 struct extent_buffer *right;
2869 int ret = 0;
2870 int wret;
2871 int split;
2872 int num_doubles = 0;
2873 int tried_avoid_double = 0;
2874
2875 l = path->nodes[0];
2876 slot = path->slots[0];
2877 if (extend && data_size + btrfs_item_size_nr(l, slot) +
2878 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2879 return -EOVERFLOW;
2880
2881 /* first try to make some room by pushing left and right */
2882 if (data_size) {
2883 wret = push_leaf_right(trans, root, path, data_size,
2884 data_size, 0, 0);
2885 if (wret < 0)
2886 return wret;
2887 if (wret) {
2888 wret = push_leaf_left(trans, root, path, data_size,
2889 data_size, 0, (u32)-1);
2890 if (wret < 0)
2891 return wret;
2892 }
2893 l = path->nodes[0];
2894
2895 /* did the pushes work? */
2896 if (btrfs_leaf_free_space(root, l) >= data_size)
2897 return 0;
2898 }
2899
2900 if (!path->nodes[1]) {
2901 ret = insert_new_root(trans, root, path, 1);
2902 if (ret)
2903 return ret;
2904 }
2905again:
2906 split = 1;
2907 l = path->nodes[0];
2908 slot = path->slots[0];
2909 nritems = btrfs_header_nritems(l);
2910 mid = (nritems + 1) / 2;
2911
2912 if (mid <= slot) {
2913 if (nritems == 1 ||
2914 leaf_space_used(l, mid, nritems - mid) + data_size >
2915 BTRFS_LEAF_DATA_SIZE(root)) {
2916 if (slot >= nritems) {
2917 split = 0;
2918 } else {
2919 mid = slot;
2920 if (mid != nritems &&
2921 leaf_space_used(l, mid, nritems - mid) +
2922 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2923 if (data_size && !tried_avoid_double)
2924 goto push_for_double;
2925 split = 2;
2926 }
2927 }
2928 }
2929 } else {
2930 if (leaf_space_used(l, 0, mid) + data_size >
2931 BTRFS_LEAF_DATA_SIZE(root)) {
2932 if (!extend && data_size && slot == 0) {
2933 split = 0;
2934 } else if ((extend || !data_size) && slot == 0) {
2935 mid = 1;
2936 } else {
2937 mid = slot;
2938 if (mid != nritems &&
2939 leaf_space_used(l, mid, nritems - mid) +
2940 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2941 if (data_size && !tried_avoid_double)
2942 goto push_for_double;
2943 split = 2 ;
2944 }
2945 }
2946 }
2947 }
2948
2949 if (split == 0)
2950 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2951 else
2952 btrfs_item_key(l, &disk_key, mid);
2953
2954 right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
2955 root->root_key.objectid,
2956 &disk_key, 0, l->start, 0);
2957 if (IS_ERR(right))
2958 return PTR_ERR(right);
2959
2960 root_add_used(root, root->leafsize);
2961
2962 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2963 btrfs_set_header_bytenr(right, right->start);
2964 btrfs_set_header_generation(right, trans->transid);
2965 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2966 btrfs_set_header_owner(right, root->root_key.objectid);
2967 btrfs_set_header_level(right, 0);
2968 write_extent_buffer(right, root->fs_info->fsid,
2969 (unsigned long)btrfs_header_fsid(right),
2970 BTRFS_FSID_SIZE);
2971
2972 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2973 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2974 BTRFS_UUID_SIZE);
2975
2976 if (split == 0) {
2977 if (mid <= slot) {
2978 btrfs_set_header_nritems(right, 0);
2979 wret = insert_ptr(trans, root, path,
2980 &disk_key, right->start,
2981 path->slots[1] + 1, 1);
2982 if (wret)
2983 ret = wret;
2984
2985 btrfs_tree_unlock(path->nodes[0]);
2986 free_extent_buffer(path->nodes[0]);
2987 path->nodes[0] = right;
2988 path->slots[0] = 0;
2989 path->slots[1] += 1;
2990 } else {
2991 btrfs_set_header_nritems(right, 0);
2992 wret = insert_ptr(trans, root, path,
2993 &disk_key,
2994 right->start,
2995 path->slots[1], 1);
2996 if (wret)
2997 ret = wret;
2998 btrfs_tree_unlock(path->nodes[0]);
2999 free_extent_buffer(path->nodes[0]);
3000 path->nodes[0] = right;
3001 path->slots[0] = 0;
3002 if (path->slots[1] == 0) {
3003 wret = fixup_low_keys(trans, root,
3004 path, &disk_key, 1);
3005 if (wret)
3006 ret = wret;
3007 }
3008 }
3009 btrfs_mark_buffer_dirty(right);
3010 return ret;
3011 }
3012
3013 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
3014 BUG_ON(ret);
3015
3016 if (split == 2) {
3017 BUG_ON(num_doubles != 0);
3018 num_doubles++;
3019 goto again;
3020 }
3021
3022 return ret;
3023
3024push_for_double:
3025 push_for_double_split(trans, root, path, data_size);
3026 tried_avoid_double = 1;
3027 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
3028 return 0;
3029 goto again;
3030}
3031
3032static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3033 struct btrfs_root *root,
3034 struct btrfs_path *path, int ins_len)
3035{
3036 struct btrfs_key key;
3037 struct extent_buffer *leaf;
3038 struct btrfs_file_extent_item *fi;
3039 u64 extent_len = 0;
3040 u32 item_size;
3041 int ret;
3042
3043 leaf = path->nodes[0];
3044 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3045
3046 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3047 key.type != BTRFS_EXTENT_CSUM_KEY);
3048
3049 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
3050 return 0;
3051
3052 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3053 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3054 fi = btrfs_item_ptr(leaf, path->slots[0],
3055 struct btrfs_file_extent_item);
3056 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3057 }
3058 btrfs_release_path(path);
3059
3060 path->keep_locks = 1;
3061 path->search_for_split = 1;
3062 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3063 path->search_for_split = 0;
3064 if (ret < 0)
3065 goto err;
3066
3067 ret = -EAGAIN;
3068 leaf = path->nodes[0];
3069 /* if our item isn't there or got smaller, return now */
3070 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3071 goto err;
3072
3073 /* the leaf has changed, it now has room. return now */
3074 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
3075 goto err;
3076
3077 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3078 fi = btrfs_item_ptr(leaf, path->slots[0],
3079 struct btrfs_file_extent_item);
3080 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3081 goto err;
3082 }
3083
3084 btrfs_set_path_blocking(path);
3085 ret = split_leaf(trans, root, &key, path, ins_len, 1);
3086 if (ret)
3087 goto err;
3088
3089 path->keep_locks = 0;
3090 btrfs_unlock_up_safe(path, 1);
3091 return 0;
3092err:
3093 path->keep_locks = 0;
3094 return ret;
3095}
3096
3097static noinline int split_item(struct btrfs_trans_handle *trans,
3098 struct btrfs_root *root,
3099 struct btrfs_path *path,
3100 struct btrfs_key *new_key,
3101 unsigned long split_offset)
3102{
3103 struct extent_buffer *leaf;
3104 struct btrfs_item *item;
3105 struct btrfs_item *new_item;
3106 int slot;
3107 char *buf;
3108 u32 nritems;
3109 u32 item_size;
3110 u32 orig_offset;
3111 struct btrfs_disk_key disk_key;
3112
3113 leaf = path->nodes[0];
3114 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3115
3116 btrfs_set_path_blocking(path);
3117
3118 item = btrfs_item_nr(leaf, path->slots[0]);
3119 orig_offset = btrfs_item_offset(leaf, item);
3120 item_size = btrfs_item_size(leaf, item);
3121
3122 buf = kmalloc(item_size, GFP_NOFS);
3123 if (!buf)
3124 return -ENOMEM;
3125
3126 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3127 path->slots[0]), item_size);
3128
3129 slot = path->slots[0] + 1;
3130 nritems = btrfs_header_nritems(leaf);
3131 if (slot != nritems) {
3132 /* shift the items */
3133 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3134 btrfs_item_nr_offset(slot),
3135 (nritems - slot) * sizeof(struct btrfs_item));
3136 }
3137
3138 btrfs_cpu_key_to_disk(&disk_key, new_key);
3139 btrfs_set_item_key(leaf, &disk_key, slot);
3140
3141 new_item = btrfs_item_nr(leaf, slot);
3142
3143 btrfs_set_item_offset(leaf, new_item, orig_offset);
3144 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3145
3146 btrfs_set_item_offset(leaf, item,
3147 orig_offset + item_size - split_offset);
3148 btrfs_set_item_size(leaf, item, split_offset);
3149
3150 btrfs_set_header_nritems(leaf, nritems + 1);
3151
3152 /* write the data for the start of the original item */
3153 write_extent_buffer(leaf, buf,
3154 btrfs_item_ptr_offset(leaf, path->slots[0]),
3155 split_offset);
3156
3157 /* write the data for the new item */
3158 write_extent_buffer(leaf, buf + split_offset,
3159 btrfs_item_ptr_offset(leaf, slot),
3160 item_size - split_offset);
3161 btrfs_mark_buffer_dirty(leaf);
3162
3163 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
3164 kfree(buf);
3165 return 0;
3166}
3167
3168/*
3169 * This function splits a single item into two items,
3170 * giving 'new_key' to the new item and splitting the
3171 * old one at split_offset (from the start of the item).
3172 *
3173 * The path may be released by this operation. After
3174 * the split, the path is pointing to the old item. The
3175 * new item is going to be in the same node as the old one.
3176 *
3177 * Note, the item being split must be smaller enough to live alone on
3178 * a tree block with room for one extra struct btrfs_item
3179 *
3180 * This allows us to split the item in place, keeping a lock on the
3181 * leaf the entire time.
3182 */
3183int btrfs_split_item(struct btrfs_trans_handle *trans,
3184 struct btrfs_root *root,
3185 struct btrfs_path *path,
3186 struct btrfs_key *new_key,
3187 unsigned long split_offset)
3188{
3189 int ret;
3190 ret = setup_leaf_for_split(trans, root, path,
3191 sizeof(struct btrfs_item));
3192 if (ret)
3193 return ret;
3194
3195 ret = split_item(trans, root, path, new_key, split_offset);
3196 return ret;
3197}
3198
3199/*
3200 * This function duplicate a item, giving 'new_key' to the new item.
3201 * It guarantees both items live in the same tree leaf and the new item
3202 * is contiguous with the original item.
3203 *
3204 * This allows us to split file extent in place, keeping a lock on the
3205 * leaf the entire time.
3206 */
3207int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
3208 struct btrfs_root *root,
3209 struct btrfs_path *path,
3210 struct btrfs_key *new_key)
3211{
3212 struct extent_buffer *leaf;
3213 int ret;
3214 u32 item_size;
3215
3216 leaf = path->nodes[0];
3217 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3218 ret = setup_leaf_for_split(trans, root, path,
3219 item_size + sizeof(struct btrfs_item));
3220 if (ret)
3221 return ret;
3222
3223 path->slots[0]++;
3224 ret = setup_items_for_insert(trans, root, path, new_key, &item_size,
3225 item_size, item_size +
3226 sizeof(struct btrfs_item), 1);
3227 BUG_ON(ret);
3228
3229 leaf = path->nodes[0];
3230 memcpy_extent_buffer(leaf,
3231 btrfs_item_ptr_offset(leaf, path->slots[0]),
3232 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
3233 item_size);
3234 return 0;
3235}
3236
3237/*
3238 * make the item pointed to by the path smaller. new_size indicates
3239 * how small to make it, and from_end tells us if we just chop bytes
3240 * off the end of the item or if we shift the item to chop bytes off
3241 * the front.
3242 */
3243int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3244 struct btrfs_root *root,
3245 struct btrfs_path *path,
3246 u32 new_size, int from_end)
3247{
3248 int slot;
3249 struct extent_buffer *leaf;
3250 struct btrfs_item *item;
3251 u32 nritems;
3252 unsigned int data_end;
3253 unsigned int old_data_start;
3254 unsigned int old_size;
3255 unsigned int size_diff;
3256 int i;
3257
3258 leaf = path->nodes[0];
3259 slot = path->slots[0];
3260
3261 old_size = btrfs_item_size_nr(leaf, slot);
3262 if (old_size == new_size)
3263 return 0;
3264
3265 nritems = btrfs_header_nritems(leaf);
3266 data_end = leaf_data_end(root, leaf);
3267
3268 old_data_start = btrfs_item_offset_nr(leaf, slot);
3269
3270 size_diff = old_size - new_size;
3271
3272 BUG_ON(slot < 0);
3273 BUG_ON(slot >= nritems);
3274
3275 /*
3276 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3277 */
3278 /* first correct the data pointers */
3279 for (i = slot; i < nritems; i++) {
3280 u32 ioff;
3281 item = btrfs_item_nr(leaf, i);
3282
3283 ioff = btrfs_item_offset(leaf, item);
3284 btrfs_set_item_offset(leaf, item, ioff + size_diff);
3285 }
3286
3287 /* shift the data */
3288 if (from_end) {
3289 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3290 data_end + size_diff, btrfs_leaf_data(leaf) +
3291 data_end, old_data_start + new_size - data_end);
3292 } else {
3293 struct btrfs_disk_key disk_key;
3294 u64 offset;
3295
3296 btrfs_item_key(leaf, &disk_key, slot);
3297
3298 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3299 unsigned long ptr;
3300 struct btrfs_file_extent_item *fi;
3301
3302 fi = btrfs_item_ptr(leaf, slot,
3303 struct btrfs_file_extent_item);
3304 fi = (struct btrfs_file_extent_item *)(
3305 (unsigned long)fi - size_diff);
3306
3307 if (btrfs_file_extent_type(leaf, fi) ==
3308 BTRFS_FILE_EXTENT_INLINE) {
3309 ptr = btrfs_item_ptr_offset(leaf, slot);
3310 memmove_extent_buffer(leaf, ptr,
3311 (unsigned long)fi,
3312 offsetof(struct btrfs_file_extent_item,
3313 disk_bytenr));
3314 }
3315 }
3316
3317 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3318 data_end + size_diff, btrfs_leaf_data(leaf) +
3319 data_end, old_data_start - data_end);
3320
3321 offset = btrfs_disk_key_offset(&disk_key);
3322 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3323 btrfs_set_item_key(leaf, &disk_key, slot);
3324 if (slot == 0)
3325 fixup_low_keys(trans, root, path, &disk_key, 1);
3326 }
3327
3328 item = btrfs_item_nr(leaf, slot);
3329 btrfs_set_item_size(leaf, item, new_size);
3330 btrfs_mark_buffer_dirty(leaf);
3331
3332 if (btrfs_leaf_free_space(root, leaf) < 0) {
3333 btrfs_print_leaf(root, leaf);
3334 BUG();
3335 }
3336 return 0;
3337}
3338
3339/*
3340 * make the item pointed to by the path bigger, data_size is the new size.
3341 */
3342int btrfs_extend_item(struct btrfs_trans_handle *trans,
3343 struct btrfs_root *root, struct btrfs_path *path,
3344 u32 data_size)
3345{
3346 int slot;
3347 struct extent_buffer *leaf;
3348 struct btrfs_item *item;
3349 u32 nritems;
3350 unsigned int data_end;
3351 unsigned int old_data;
3352 unsigned int old_size;
3353 int i;
3354
3355 leaf = path->nodes[0];
3356
3357 nritems = btrfs_header_nritems(leaf);
3358 data_end = leaf_data_end(root, leaf);
3359
3360 if (btrfs_leaf_free_space(root, leaf) < data_size) {
3361 btrfs_print_leaf(root, leaf);
3362 BUG();
3363 }
3364 slot = path->slots[0];
3365 old_data = btrfs_item_end_nr(leaf, slot);
3366
3367 BUG_ON(slot < 0);
3368 if (slot >= nritems) {
3369 btrfs_print_leaf(root, leaf);
3370 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3371 slot, nritems);
3372 BUG_ON(1);
3373 }
3374
3375 /*
3376 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3377 */
3378 /* first correct the data pointers */
3379 for (i = slot; i < nritems; i++) {
3380 u32 ioff;
3381 item = btrfs_item_nr(leaf, i);
3382
3383 ioff = btrfs_item_offset(leaf, item);
3384 btrfs_set_item_offset(leaf, item, ioff - data_size);
3385 }
3386
3387 /* shift the data */
3388 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3389 data_end - data_size, btrfs_leaf_data(leaf) +
3390 data_end, old_data - data_end);
3391
3392 data_end = old_data;
3393 old_size = btrfs_item_size_nr(leaf, slot);
3394 item = btrfs_item_nr(leaf, slot);
3395 btrfs_set_item_size(leaf, item, old_size + data_size);
3396 btrfs_mark_buffer_dirty(leaf);
3397
3398 if (btrfs_leaf_free_space(root, leaf) < 0) {
3399 btrfs_print_leaf(root, leaf);
3400 BUG();
3401 }
3402 return 0;
3403}
3404
3405/*
3406 * Given a key and some data, insert items into the tree.
3407 * This does all the path init required, making room in the tree if needed.
3408 * Returns the number of keys that were inserted.
3409 */
3410int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3411 struct btrfs_root *root,
3412 struct btrfs_path *path,
3413 struct btrfs_key *cpu_key, u32 *data_size,
3414 int nr)
3415{
3416 struct extent_buffer *leaf;
3417 struct btrfs_item *item;
3418 int ret = 0;
3419 int slot;
3420 int i;
3421 u32 nritems;
3422 u32 total_data = 0;
3423 u32 total_size = 0;
3424 unsigned int data_end;
3425 struct btrfs_disk_key disk_key;
3426 struct btrfs_key found_key;
3427
3428 for (i = 0; i < nr; i++) {
3429 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3430 BTRFS_LEAF_DATA_SIZE(root)) {
3431 break;
3432 nr = i;
3433 }
3434 total_data += data_size[i];
3435 total_size += data_size[i] + sizeof(struct btrfs_item);
3436 }
3437 BUG_ON(nr == 0);
3438
3439 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3440 if (ret == 0)
3441 return -EEXIST;
3442 if (ret < 0)
3443 goto out;
3444
3445 leaf = path->nodes[0];
3446
3447 nritems = btrfs_header_nritems(leaf);
3448 data_end = leaf_data_end(root, leaf);
3449
3450 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3451 for (i = nr; i >= 0; i--) {
3452 total_data -= data_size[i];
3453 total_size -= data_size[i] + sizeof(struct btrfs_item);
3454 if (total_size < btrfs_leaf_free_space(root, leaf))
3455 break;
3456 }
3457 nr = i;
3458 }
3459
3460 slot = path->slots[0];
3461 BUG_ON(slot < 0);
3462
3463 if (slot != nritems) {
3464 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3465
3466 item = btrfs_item_nr(leaf, slot);
3467 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3468
3469 /* figure out how many keys we can insert in here */
3470 total_data = data_size[0];
3471 for (i = 1; i < nr; i++) {
3472 if (btrfs_comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3473 break;
3474 total_data += data_size[i];
3475 }
3476 nr = i;
3477
3478 if (old_data < data_end) {
3479 btrfs_print_leaf(root, leaf);
3480 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3481 slot, old_data, data_end);
3482 BUG_ON(1);
3483 }
3484 /*
3485 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3486 */
3487 /* first correct the data pointers */
3488 for (i = slot; i < nritems; i++) {
3489 u32 ioff;
3490
3491 item = btrfs_item_nr(leaf, i);
3492 ioff = btrfs_item_offset(leaf, item);
3493 btrfs_set_item_offset(leaf, item, ioff - total_data);
3494 }
3495 /* shift the items */
3496 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3497 btrfs_item_nr_offset(slot),
3498 (nritems - slot) * sizeof(struct btrfs_item));
3499
3500 /* shift the data */
3501 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3502 data_end - total_data, btrfs_leaf_data(leaf) +
3503 data_end, old_data - data_end);
3504 data_end = old_data;
3505 } else {
3506 /*
3507 * this sucks but it has to be done, if we are inserting at
3508 * the end of the leaf only insert 1 of the items, since we
3509 * have no way of knowing whats on the next leaf and we'd have
3510 * to drop our current locks to figure it out
3511 */
3512 nr = 1;
3513 }
3514
3515 /* setup the item for the new data */
3516 for (i = 0; i < nr; i++) {
3517 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3518 btrfs_set_item_key(leaf, &disk_key, slot + i);
3519 item = btrfs_item_nr(leaf, slot + i);
3520 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3521 data_end -= data_size[i];
3522 btrfs_set_item_size(leaf, item, data_size[i]);
3523 }
3524 btrfs_set_header_nritems(leaf, nritems + nr);
3525 btrfs_mark_buffer_dirty(leaf);
3526
3527 ret = 0;
3528 if (slot == 0) {
3529 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3530 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3531 }
3532
3533 if (btrfs_leaf_free_space(root, leaf) < 0) {
3534 btrfs_print_leaf(root, leaf);
3535 BUG();
3536 }
3537out:
3538 if (!ret)
3539 ret = nr;
3540 return ret;
3541}
3542
3543/*
3544 * this is a helper for btrfs_insert_empty_items, the main goal here is
3545 * to save stack depth by doing the bulk of the work in a function
3546 * that doesn't call btrfs_search_slot
3547 */
3548int setup_items_for_insert(struct btrfs_trans_handle *trans,
3549 struct btrfs_root *root, struct btrfs_path *path,
3550 struct btrfs_key *cpu_key, u32 *data_size,
3551 u32 total_data, u32 total_size, int nr)
3552{
3553 struct btrfs_item *item;
3554 int i;
3555 u32 nritems;
3556 unsigned int data_end;
3557 struct btrfs_disk_key disk_key;
3558 int ret;
3559 struct extent_buffer *leaf;
3560 int slot;
3561
3562 leaf = path->nodes[0];
3563 slot = path->slots[0];
3564
3565 nritems = btrfs_header_nritems(leaf);
3566 data_end = leaf_data_end(root, leaf);
3567
3568 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3569 btrfs_print_leaf(root, leaf);
3570 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3571 total_size, btrfs_leaf_free_space(root, leaf));
3572 BUG();
3573 }
3574
3575 if (slot != nritems) {
3576 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3577
3578 if (old_data < data_end) {
3579 btrfs_print_leaf(root, leaf);
3580 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3581 slot, old_data, data_end);
3582 BUG_ON(1);
3583 }
3584 /*
3585 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3586 */
3587 /* first correct the data pointers */
3588 for (i = slot; i < nritems; i++) {
3589 u32 ioff;
3590
3591 item = btrfs_item_nr(leaf, i);
3592 ioff = btrfs_item_offset(leaf, item);
3593 btrfs_set_item_offset(leaf, item, ioff - total_data);
3594 }
3595 /* shift the items */
3596 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3597 btrfs_item_nr_offset(slot),
3598 (nritems - slot) * sizeof(struct btrfs_item));
3599
3600 /* shift the data */
3601 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3602 data_end - total_data, btrfs_leaf_data(leaf) +
3603 data_end, old_data - data_end);
3604 data_end = old_data;
3605 }
3606
3607 /* setup the item for the new data */
3608 for (i = 0; i < nr; i++) {
3609 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3610 btrfs_set_item_key(leaf, &disk_key, slot + i);
3611 item = btrfs_item_nr(leaf, slot + i);
3612 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3613 data_end -= data_size[i];
3614 btrfs_set_item_size(leaf, item, data_size[i]);
3615 }
3616
3617 btrfs_set_header_nritems(leaf, nritems + nr);
3618
3619 ret = 0;
3620 if (slot == 0) {
3621 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3622 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3623 }
3624 btrfs_unlock_up_safe(path, 1);
3625 btrfs_mark_buffer_dirty(leaf);
3626
3627 if (btrfs_leaf_free_space(root, leaf) < 0) {
3628 btrfs_print_leaf(root, leaf);
3629 BUG();
3630 }
3631 return ret;
3632}
3633
3634/*
3635 * Given a key and some data, insert items into the tree.
3636 * This does all the path init required, making room in the tree if needed.
3637 */
3638int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3639 struct btrfs_root *root,
3640 struct btrfs_path *path,
3641 struct btrfs_key *cpu_key, u32 *data_size,
3642 int nr)
3643{
3644 int ret = 0;
3645 int slot;
3646 int i;
3647 u32 total_size = 0;
3648 u32 total_data = 0;
3649
3650 for (i = 0; i < nr; i++)
3651 total_data += data_size[i];
3652
3653 total_size = total_data + (nr * sizeof(struct btrfs_item));
3654 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3655 if (ret == 0)
3656 return -EEXIST;
3657 if (ret < 0)
3658 goto out;
3659
3660 slot = path->slots[0];
3661 BUG_ON(slot < 0);
3662
3663 ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3664 total_data, total_size, nr);
3665
3666out:
3667 return ret;
3668}
3669
3670/*
3671 * Given a key and some data, insert an item into the tree.
3672 * This does all the path init required, making room in the tree if needed.
3673 */
3674int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3675 *root, struct btrfs_key *cpu_key, void *data, u32
3676 data_size)
3677{
3678 int ret = 0;
3679 struct btrfs_path *path;
3680 struct extent_buffer *leaf;
3681 unsigned long ptr;
3682
3683 path = btrfs_alloc_path();
3684 if (!path)
3685 return -ENOMEM;
3686 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3687 if (!ret) {
3688 leaf = path->nodes[0];
3689 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3690 write_extent_buffer(leaf, data, ptr, data_size);
3691 btrfs_mark_buffer_dirty(leaf);
3692 }
3693 btrfs_free_path(path);
3694 return ret;
3695}
3696
3697/*
3698 * delete the pointer from a given node.
3699 *
3700 * the tree should have been previously balanced so the deletion does not
3701 * empty a node.
3702 */
3703static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3704 struct btrfs_path *path, int level, int slot)
3705{
3706 struct extent_buffer *parent = path->nodes[level];
3707 u32 nritems;
3708 int ret = 0;
3709 int wret;
3710
3711 nritems = btrfs_header_nritems(parent);
3712 if (slot != nritems - 1) {
3713 memmove_extent_buffer(parent,
3714 btrfs_node_key_ptr_offset(slot),
3715 btrfs_node_key_ptr_offset(slot + 1),
3716 sizeof(struct btrfs_key_ptr) *
3717 (nritems - slot - 1));
3718 }
3719 nritems--;
3720 btrfs_set_header_nritems(parent, nritems);
3721 if (nritems == 0 && parent == root->node) {
3722 BUG_ON(btrfs_header_level(root->node) != 1);
3723 /* just turn the root into a leaf and break */
3724 btrfs_set_header_level(root->node, 0);
3725 } else if (slot == 0) {
3726 struct btrfs_disk_key disk_key;
3727
3728 btrfs_node_key(parent, &disk_key, 0);
3729 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3730 if (wret)
3731 ret = wret;
3732 }
3733 btrfs_mark_buffer_dirty(parent);
3734 return ret;
3735}
3736
3737/*
3738 * a helper function to delete the leaf pointed to by path->slots[1] and
3739 * path->nodes[1].
3740 *
3741 * This deletes the pointer in path->nodes[1] and frees the leaf
3742 * block extent. zero is returned if it all worked out, < 0 otherwise.
3743 *
3744 * The path must have already been setup for deleting the leaf, including
3745 * all the proper balancing. path->nodes[1] must be locked.
3746 */
3747static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3748 struct btrfs_root *root,
3749 struct btrfs_path *path,
3750 struct extent_buffer *leaf)
3751{
3752 int ret;
3753
3754 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
3755 ret = del_ptr(trans, root, path, 1, path->slots[1]);
3756 if (ret)
3757 return ret;
3758
3759 /*
3760 * btrfs_free_extent is expensive, we want to make sure we
3761 * aren't holding any locks when we call it
3762 */
3763 btrfs_unlock_up_safe(path, 0);
3764
3765 root_sub_used(root, leaf->len);
3766
3767 btrfs_free_tree_block(trans, root, leaf, 0, 1);
3768 return 0;
3769}
3770/*
3771 * delete the item at the leaf level in path. If that empties
3772 * the leaf, remove it from the tree
3773 */
3774int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3775 struct btrfs_path *path, int slot, int nr)
3776{
3777 struct extent_buffer *leaf;
3778 struct btrfs_item *item;
3779 int last_off;
3780 int dsize = 0;
3781 int ret = 0;
3782 int wret;
3783 int i;
3784 u32 nritems;
3785
3786 leaf = path->nodes[0];
3787 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3788
3789 for (i = 0; i < nr; i++)
3790 dsize += btrfs_item_size_nr(leaf, slot + i);
3791
3792 nritems = btrfs_header_nritems(leaf);
3793
3794 if (slot + nr != nritems) {
3795 int data_end = leaf_data_end(root, leaf);
3796
3797 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3798 data_end + dsize,
3799 btrfs_leaf_data(leaf) + data_end,
3800 last_off - data_end);
3801
3802 for (i = slot + nr; i < nritems; i++) {
3803 u32 ioff;
3804
3805 item = btrfs_item_nr(leaf, i);
3806 ioff = btrfs_item_offset(leaf, item);
3807 btrfs_set_item_offset(leaf, item, ioff + dsize);
3808 }
3809
3810 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3811 btrfs_item_nr_offset(slot + nr),
3812 sizeof(struct btrfs_item) *
3813 (nritems - slot - nr));
3814 }
3815 btrfs_set_header_nritems(leaf, nritems - nr);
3816 nritems -= nr;
3817
3818 /* delete the leaf if we've emptied it */
3819 if (nritems == 0) {
3820 if (leaf == root->node) {
3821 btrfs_set_header_level(leaf, 0);
3822 } else {
3823 btrfs_set_path_blocking(path);
3824 clean_tree_block(trans, root, leaf);
3825 ret = btrfs_del_leaf(trans, root, path, leaf);
3826 BUG_ON(ret);
3827 }
3828 } else {
3829 int used = leaf_space_used(leaf, 0, nritems);
3830 if (slot == 0) {
3831 struct btrfs_disk_key disk_key;
3832
3833 btrfs_item_key(leaf, &disk_key, 0);
3834 wret = fixup_low_keys(trans, root, path,
3835 &disk_key, 1);
3836 if (wret)
3837 ret = wret;
3838 }
3839
3840 /* delete the leaf if it is mostly empty */
3841 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
3842 /* push_leaf_left fixes the path.
3843 * make sure the path still points to our leaf
3844 * for possible call to del_ptr below
3845 */
3846 slot = path->slots[1];
3847 extent_buffer_get(leaf);
3848
3849 btrfs_set_path_blocking(path);
3850 wret = push_leaf_left(trans, root, path, 1, 1,
3851 1, (u32)-1);
3852 if (wret < 0 && wret != -ENOSPC)
3853 ret = wret;
3854
3855 if (path->nodes[0] == leaf &&
3856 btrfs_header_nritems(leaf)) {
3857 wret = push_leaf_right(trans, root, path, 1,
3858 1, 1, 0);
3859 if (wret < 0 && wret != -ENOSPC)
3860 ret = wret;
3861 }
3862
3863 if (btrfs_header_nritems(leaf) == 0) {
3864 path->slots[1] = slot;
3865 ret = btrfs_del_leaf(trans, root, path, leaf);
3866 BUG_ON(ret);
3867 free_extent_buffer(leaf);
3868 } else {
3869 /* if we're still in the path, make sure
3870 * we're dirty. Otherwise, one of the
3871 * push_leaf functions must have already
3872 * dirtied this buffer
3873 */
3874 if (path->nodes[0] == leaf)
3875 btrfs_mark_buffer_dirty(leaf);
3876 free_extent_buffer(leaf);
3877 }
3878 } else {
3879 btrfs_mark_buffer_dirty(leaf);
3880 }
3881 }
3882 return ret;
3883}
3884
3885/*
3886 * search the tree again to find a leaf with lesser keys
3887 * returns 0 if it found something or 1 if there are no lesser leaves.
3888 * returns < 0 on io errors.
3889 *
3890 * This may release the path, and so you may lose any locks held at the
3891 * time you call it.
3892 */
3893int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3894{
3895 struct btrfs_key key;
3896 struct btrfs_disk_key found_key;
3897 int ret;
3898
3899 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3900
3901 if (key.offset > 0)
3902 key.offset--;
3903 else if (key.type > 0)
3904 key.type--;
3905 else if (key.objectid > 0)
3906 key.objectid--;
3907 else
3908 return 1;
3909
3910 btrfs_release_path(path);
3911 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3912 if (ret < 0)
3913 return ret;
3914 btrfs_item_key(path->nodes[0], &found_key, 0);
3915 ret = comp_keys(&found_key, &key);
3916 if (ret < 0)
3917 return 0;
3918 return 1;
3919}
3920
3921/*
3922 * A helper function to walk down the tree starting at min_key, and looking
3923 * for nodes or leaves that are either in cache or have a minimum
3924 * transaction id. This is used by the btree defrag code, and tree logging
3925 *
3926 * This does not cow, but it does stuff the starting key it finds back
3927 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3928 * key and get a writable path.
3929 *
3930 * This does lock as it descends, and path->keep_locks should be set
3931 * to 1 by the caller.
3932 *
3933 * This honors path->lowest_level to prevent descent past a given level
3934 * of the tree.
3935 *
3936 * min_trans indicates the oldest transaction that you are interested
3937 * in walking through. Any nodes or leaves older than min_trans are
3938 * skipped over (without reading them).
3939 *
3940 * returns zero if something useful was found, < 0 on error and 1 if there
3941 * was nothing in the tree that matched the search criteria.
3942 */
3943int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3944 struct btrfs_key *max_key,
3945 struct btrfs_path *path, int cache_only,
3946 u64 min_trans)
3947{
3948 struct extent_buffer *cur;
3949 struct btrfs_key found_key;
3950 int slot;
3951 int sret;
3952 u32 nritems;
3953 int level;
3954 int ret = 1;
3955
3956 WARN_ON(!path->keep_locks);
3957again:
3958 cur = btrfs_read_lock_root_node(root);
3959 level = btrfs_header_level(cur);
3960 WARN_ON(path->nodes[level]);
3961 path->nodes[level] = cur;
3962 path->locks[level] = BTRFS_READ_LOCK;
3963
3964 if (btrfs_header_generation(cur) < min_trans) {
3965 ret = 1;
3966 goto out;
3967 }
3968 while (1) {
3969 nritems = btrfs_header_nritems(cur);
3970 level = btrfs_header_level(cur);
3971 sret = bin_search(cur, min_key, level, &slot);
3972
3973 /* at the lowest level, we're done, setup the path and exit */
3974 if (level == path->lowest_level) {
3975 if (slot >= nritems)
3976 goto find_next_key;
3977 ret = 0;
3978 path->slots[level] = slot;
3979 btrfs_item_key_to_cpu(cur, &found_key, slot);
3980 goto out;
3981 }
3982 if (sret && slot > 0)
3983 slot--;
3984 /*
3985 * check this node pointer against the cache_only and
3986 * min_trans parameters. If it isn't in cache or is too
3987 * old, skip to the next one.
3988 */
3989 while (slot < nritems) {
3990 u64 blockptr;
3991 u64 gen;
3992 struct extent_buffer *tmp;
3993 struct btrfs_disk_key disk_key;
3994
3995 blockptr = btrfs_node_blockptr(cur, slot);
3996 gen = btrfs_node_ptr_generation(cur, slot);
3997 if (gen < min_trans) {
3998 slot++;
3999 continue;
4000 }
4001 if (!cache_only)
4002 break;
4003
4004 if (max_key) {
4005 btrfs_node_key(cur, &disk_key, slot);
4006 if (comp_keys(&disk_key, max_key) >= 0) {
4007 ret = 1;
4008 goto out;
4009 }
4010 }
4011
4012 tmp = btrfs_find_tree_block(root, blockptr,
4013 btrfs_level_size(root, level - 1));
4014
4015 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
4016 free_extent_buffer(tmp);
4017 break;
4018 }
4019 if (tmp)
4020 free_extent_buffer(tmp);
4021 slot++;
4022 }
4023find_next_key:
4024 /*
4025 * we didn't find a candidate key in this node, walk forward
4026 * and find another one
4027 */
4028 if (slot >= nritems) {
4029 path->slots[level] = slot;
4030 btrfs_set_path_blocking(path);
4031 sret = btrfs_find_next_key(root, path, min_key, level,
4032 cache_only, min_trans);
4033 if (sret == 0) {
4034 btrfs_release_path(path);
4035 goto again;
4036 } else {
4037 goto out;
4038 }
4039 }
4040 /* save our key for returning back */
4041 btrfs_node_key_to_cpu(cur, &found_key, slot);
4042 path->slots[level] = slot;
4043 if (level == path->lowest_level) {
4044 ret = 0;
4045 unlock_up(path, level, 1);
4046 goto out;
4047 }
4048 btrfs_set_path_blocking(path);
4049 cur = read_node_slot(root, cur, slot);
4050 BUG_ON(!cur);
4051
4052 btrfs_tree_read_lock(cur);
4053
4054 path->locks[level - 1] = BTRFS_READ_LOCK;
4055 path->nodes[level - 1] = cur;
4056 unlock_up(path, level, 1);
4057 btrfs_clear_path_blocking(path, NULL, 0);
4058 }
4059out:
4060 if (ret == 0)
4061 memcpy(min_key, &found_key, sizeof(found_key));
4062 btrfs_set_path_blocking(path);
4063 return ret;
4064}
4065
4066/*
4067 * this is similar to btrfs_next_leaf, but does not try to preserve
4068 * and fixup the path. It looks for and returns the next key in the
4069 * tree based on the current path and the cache_only and min_trans
4070 * parameters.
4071 *
4072 * 0 is returned if another key is found, < 0 if there are any errors
4073 * and 1 is returned if there are no higher keys in the tree
4074 *
4075 * path->keep_locks should be set to 1 on the search made before
4076 * calling this function.
4077 */
4078int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4079 struct btrfs_key *key, int level,
4080 int cache_only, u64 min_trans)
4081{
4082 int slot;
4083 struct extent_buffer *c;
4084
4085 WARN_ON(!path->keep_locks);
4086 while (level < BTRFS_MAX_LEVEL) {
4087 if (!path->nodes[level])
4088 return 1;
4089
4090 slot = path->slots[level] + 1;
4091 c = path->nodes[level];
4092next:
4093 if (slot >= btrfs_header_nritems(c)) {
4094 int ret;
4095 int orig_lowest;
4096 struct btrfs_key cur_key;
4097 if (level + 1 >= BTRFS_MAX_LEVEL ||
4098 !path->nodes[level + 1])
4099 return 1;
4100
4101 if (path->locks[level + 1]) {
4102 level++;
4103 continue;
4104 }
4105
4106 slot = btrfs_header_nritems(c) - 1;
4107 if (level == 0)
4108 btrfs_item_key_to_cpu(c, &cur_key, slot);
4109 else
4110 btrfs_node_key_to_cpu(c, &cur_key, slot);
4111
4112 orig_lowest = path->lowest_level;
4113 btrfs_release_path(path);
4114 path->lowest_level = level;
4115 ret = btrfs_search_slot(NULL, root, &cur_key, path,
4116 0, 0);
4117 path->lowest_level = orig_lowest;
4118 if (ret < 0)
4119 return ret;
4120
4121 c = path->nodes[level];
4122 slot = path->slots[level];
4123 if (ret == 0)
4124 slot++;
4125 goto next;
4126 }
4127
4128 if (level == 0)
4129 btrfs_item_key_to_cpu(c, key, slot);
4130 else {
4131 u64 blockptr = btrfs_node_blockptr(c, slot);
4132 u64 gen = btrfs_node_ptr_generation(c, slot);
4133
4134 if (cache_only) {
4135 struct extent_buffer *cur;
4136 cur = btrfs_find_tree_block(root, blockptr,
4137 btrfs_level_size(root, level - 1));
4138 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4139 slot++;
4140 if (cur)
4141 free_extent_buffer(cur);
4142 goto next;
4143 }
4144 free_extent_buffer(cur);
4145 }
4146 if (gen < min_trans) {
4147 slot++;
4148 goto next;
4149 }
4150 btrfs_node_key_to_cpu(c, key, slot);
4151 }
4152 return 0;
4153 }
4154 return 1;
4155}
4156
4157/*
4158 * search the tree again to find a leaf with greater keys
4159 * returns 0 if it found something or 1 if there are no greater leaves.
4160 * returns < 0 on io errors.
4161 */
4162int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4163{
4164 int slot;
4165 int level;
4166 struct extent_buffer *c;
4167 struct extent_buffer *next;
4168 struct btrfs_key key;
4169 u32 nritems;
4170 int ret;
4171 int old_spinning = path->leave_spinning;
4172 int next_rw_lock = 0;
4173
4174 nritems = btrfs_header_nritems(path->nodes[0]);
4175 if (nritems == 0)
4176 return 1;
4177
4178 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4179again:
4180 level = 1;
4181 next = NULL;
4182 next_rw_lock = 0;
4183 btrfs_release_path(path);
4184
4185 path->keep_locks = 1;
4186 path->leave_spinning = 1;
4187
4188 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4189 path->keep_locks = 0;
4190
4191 if (ret < 0)
4192 return ret;
4193
4194 nritems = btrfs_header_nritems(path->nodes[0]);
4195 /*
4196 * by releasing the path above we dropped all our locks. A balance
4197 * could have added more items next to the key that used to be
4198 * at the very end of the block. So, check again here and
4199 * advance the path if there are now more items available.
4200 */
4201 if (nritems > 0 && path->slots[0] < nritems - 1) {
4202 if (ret == 0)
4203 path->slots[0]++;
4204 ret = 0;
4205 goto done;
4206 }
4207
4208 while (level < BTRFS_MAX_LEVEL) {
4209 if (!path->nodes[level]) {
4210 ret = 1;
4211 goto done;
4212 }
4213
4214 slot = path->slots[level] + 1;
4215 c = path->nodes[level];
4216 if (slot >= btrfs_header_nritems(c)) {
4217 level++;
4218 if (level == BTRFS_MAX_LEVEL) {
4219 ret = 1;
4220 goto done;
4221 }
4222 continue;
4223 }
4224
4225 if (next) {
4226 btrfs_tree_unlock_rw(next, next_rw_lock);
4227 free_extent_buffer(next);
4228 }
4229
4230 next = c;
4231 next_rw_lock = path->locks[level];
4232 ret = read_block_for_search(NULL, root, path, &next, level,
4233 slot, &key);
4234 if (ret == -EAGAIN)
4235 goto again;
4236
4237 if (ret < 0) {
4238 btrfs_release_path(path);
4239 goto done;
4240 }
4241
4242 if (!path->skip_locking) {
4243 ret = btrfs_try_tree_read_lock(next);
4244 if (!ret) {
4245 btrfs_set_path_blocking(path);
4246 btrfs_tree_read_lock(next);
4247 btrfs_clear_path_blocking(path, next,
4248 BTRFS_READ_LOCK);
4249 }
4250 next_rw_lock = BTRFS_READ_LOCK;
4251 }
4252 break;
4253 }
4254 path->slots[level] = slot;
4255 while (1) {
4256 level--;
4257 c = path->nodes[level];
4258 if (path->locks[level])
4259 btrfs_tree_unlock_rw(c, path->locks[level]);
4260
4261 free_extent_buffer(c);
4262 path->nodes[level] = next;
4263 path->slots[level] = 0;
4264 if (!path->skip_locking)
4265 path->locks[level] = next_rw_lock;
4266 if (!level)
4267 break;
4268
4269 ret = read_block_for_search(NULL, root, path, &next, level,
4270 0, &key);
4271 if (ret == -EAGAIN)
4272 goto again;
4273
4274 if (ret < 0) {
4275 btrfs_release_path(path);
4276 goto done;
4277 }
4278
4279 if (!path->skip_locking) {
4280 ret = btrfs_try_tree_read_lock(next);
4281 if (!ret) {
4282 btrfs_set_path_blocking(path);
4283 btrfs_tree_read_lock(next);
4284 btrfs_clear_path_blocking(path, next,
4285 BTRFS_READ_LOCK);
4286 }
4287 next_rw_lock = BTRFS_READ_LOCK;
4288 }
4289 }
4290 ret = 0;
4291done:
4292 unlock_up(path, 0, 1);
4293 path->leave_spinning = old_spinning;
4294 if (!old_spinning)
4295 btrfs_set_path_blocking(path);
4296
4297 return ret;
4298}
4299
4300/*
4301 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4302 * searching until it gets past min_objectid or finds an item of 'type'
4303 *
4304 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4305 */
4306int btrfs_previous_item(struct btrfs_root *root,
4307 struct btrfs_path *path, u64 min_objectid,
4308 int type)
4309{
4310 struct btrfs_key found_key;
4311 struct extent_buffer *leaf;
4312 u32 nritems;
4313 int ret;
4314
4315 while (1) {
4316 if (path->slots[0] == 0) {
4317 btrfs_set_path_blocking(path);
4318 ret = btrfs_prev_leaf(root, path);
4319 if (ret != 0)
4320 return ret;
4321 } else {
4322 path->slots[0]--;
4323 }
4324 leaf = path->nodes[0];
4325 nritems = btrfs_header_nritems(leaf);
4326 if (nritems == 0)
4327 return 1;
4328 if (path->slots[0] == nritems)
4329 path->slots[0]--;
4330
4331 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4332 if (found_key.objectid < min_objectid)
4333 break;
4334 if (found_key.type == type)
4335 return 0;
4336 if (found_key.objectid == min_objectid &&
4337 found_key.type < type)
4338 break;
4339 }
4340 return 1;
4341}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 */
5
6#include <linux/sched.h>
7#include <linux/slab.h>
8#include <linux/rbtree.h>
9#include <linux/mm.h>
10#include <linux/error-injection.h>
11#include "messages.h"
12#include "ctree.h"
13#include "disk-io.h"
14#include "transaction.h"
15#include "print-tree.h"
16#include "locking.h"
17#include "volumes.h"
18#include "qgroup.h"
19#include "tree-mod-log.h"
20#include "tree-checker.h"
21#include "fs.h"
22#include "accessors.h"
23#include "extent-tree.h"
24#include "relocation.h"
25#include "file-item.h"
26
27static struct kmem_cache *btrfs_path_cachep;
28
29static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
30 *root, struct btrfs_path *path, int level);
31static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
32 const struct btrfs_key *ins_key, struct btrfs_path *path,
33 int data_size, int extend);
34static int push_node_left(struct btrfs_trans_handle *trans,
35 struct extent_buffer *dst,
36 struct extent_buffer *src, int empty);
37static int balance_node_right(struct btrfs_trans_handle *trans,
38 struct extent_buffer *dst_buf,
39 struct extent_buffer *src_buf);
40static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
41 int level, int slot);
42
43static const struct btrfs_csums {
44 u16 size;
45 const char name[10];
46 const char driver[12];
47} btrfs_csums[] = {
48 [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" },
49 [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" },
50 [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" },
51 [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b",
52 .driver = "blake2b-256" },
53};
54
55/*
56 * The leaf data grows from end-to-front in the node. this returns the address
57 * of the start of the last item, which is the stop of the leaf data stack.
58 */
59static unsigned int leaf_data_end(const struct extent_buffer *leaf)
60{
61 u32 nr = btrfs_header_nritems(leaf);
62
63 if (nr == 0)
64 return BTRFS_LEAF_DATA_SIZE(leaf->fs_info);
65 return btrfs_item_offset(leaf, nr - 1);
66}
67
68/*
69 * Move data in a @leaf (using memmove, safe for overlapping ranges).
70 *
71 * @leaf: leaf that we're doing a memmove on
72 * @dst_offset: item data offset we're moving to
73 * @src_offset: item data offset were' moving from
74 * @len: length of the data we're moving
75 *
76 * Wrapper around memmove_extent_buffer() that takes into account the header on
77 * the leaf. The btrfs_item offset's start directly after the header, so we
78 * have to adjust any offsets to account for the header in the leaf. This
79 * handles that math to simplify the callers.
80 */
81static inline void memmove_leaf_data(const struct extent_buffer *leaf,
82 unsigned long dst_offset,
83 unsigned long src_offset,
84 unsigned long len)
85{
86 memmove_extent_buffer(leaf, btrfs_item_nr_offset(leaf, 0) + dst_offset,
87 btrfs_item_nr_offset(leaf, 0) + src_offset, len);
88}
89
90/*
91 * Copy item data from @src into @dst at the given @offset.
92 *
93 * @dst: destination leaf that we're copying into
94 * @src: source leaf that we're copying from
95 * @dst_offset: item data offset we're copying to
96 * @src_offset: item data offset were' copying from
97 * @len: length of the data we're copying
98 *
99 * Wrapper around copy_extent_buffer() that takes into account the header on
100 * the leaf. The btrfs_item offset's start directly after the header, so we
101 * have to adjust any offsets to account for the header in the leaf. This
102 * handles that math to simplify the callers.
103 */
104static inline void copy_leaf_data(const struct extent_buffer *dst,
105 const struct extent_buffer *src,
106 unsigned long dst_offset,
107 unsigned long src_offset, unsigned long len)
108{
109 copy_extent_buffer(dst, src, btrfs_item_nr_offset(dst, 0) + dst_offset,
110 btrfs_item_nr_offset(src, 0) + src_offset, len);
111}
112
113/*
114 * Move items in a @leaf (using memmove).
115 *
116 * @dst: destination leaf for the items
117 * @dst_item: the item nr we're copying into
118 * @src_item: the item nr we're copying from
119 * @nr_items: the number of items to copy
120 *
121 * Wrapper around memmove_extent_buffer() that does the math to get the
122 * appropriate offsets into the leaf from the item numbers.
123 */
124static inline void memmove_leaf_items(const struct extent_buffer *leaf,
125 int dst_item, int src_item, int nr_items)
126{
127 memmove_extent_buffer(leaf, btrfs_item_nr_offset(leaf, dst_item),
128 btrfs_item_nr_offset(leaf, src_item),
129 nr_items * sizeof(struct btrfs_item));
130}
131
132/*
133 * Copy items from @src into @dst at the given @offset.
134 *
135 * @dst: destination leaf for the items
136 * @src: source leaf for the items
137 * @dst_item: the item nr we're copying into
138 * @src_item: the item nr we're copying from
139 * @nr_items: the number of items to copy
140 *
141 * Wrapper around copy_extent_buffer() that does the math to get the
142 * appropriate offsets into the leaf from the item numbers.
143 */
144static inline void copy_leaf_items(const struct extent_buffer *dst,
145 const struct extent_buffer *src,
146 int dst_item, int src_item, int nr_items)
147{
148 copy_extent_buffer(dst, src, btrfs_item_nr_offset(dst, dst_item),
149 btrfs_item_nr_offset(src, src_item),
150 nr_items * sizeof(struct btrfs_item));
151}
152
153int btrfs_super_csum_size(const struct btrfs_super_block *s)
154{
155 u16 t = btrfs_super_csum_type(s);
156 /*
157 * csum type is validated at mount time
158 */
159 return btrfs_csums[t].size;
160}
161
162const char *btrfs_super_csum_name(u16 csum_type)
163{
164 /* csum type is validated at mount time */
165 return btrfs_csums[csum_type].name;
166}
167
168/*
169 * Return driver name if defined, otherwise the name that's also a valid driver
170 * name
171 */
172const char *btrfs_super_csum_driver(u16 csum_type)
173{
174 /* csum type is validated at mount time */
175 return btrfs_csums[csum_type].driver[0] ?
176 btrfs_csums[csum_type].driver :
177 btrfs_csums[csum_type].name;
178}
179
180size_t __attribute_const__ btrfs_get_num_csums(void)
181{
182 return ARRAY_SIZE(btrfs_csums);
183}
184
185struct btrfs_path *btrfs_alloc_path(void)
186{
187 might_sleep();
188
189 return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
190}
191
192/* this also releases the path */
193void btrfs_free_path(struct btrfs_path *p)
194{
195 if (!p)
196 return;
197 btrfs_release_path(p);
198 kmem_cache_free(btrfs_path_cachep, p);
199}
200
201/*
202 * path release drops references on the extent buffers in the path
203 * and it drops any locks held by this path
204 *
205 * It is safe to call this on paths that no locks or extent buffers held.
206 */
207noinline void btrfs_release_path(struct btrfs_path *p)
208{
209 int i;
210
211 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
212 p->slots[i] = 0;
213 if (!p->nodes[i])
214 continue;
215 if (p->locks[i]) {
216 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
217 p->locks[i] = 0;
218 }
219 free_extent_buffer(p->nodes[i]);
220 p->nodes[i] = NULL;
221 }
222}
223
224/*
225 * We want the transaction abort to print stack trace only for errors where the
226 * cause could be a bug, eg. due to ENOSPC, and not for common errors that are
227 * caused by external factors.
228 */
229bool __cold abort_should_print_stack(int errno)
230{
231 switch (errno) {
232 case -EIO:
233 case -EROFS:
234 case -ENOMEM:
235 return false;
236 }
237 return true;
238}
239
240/*
241 * safely gets a reference on the root node of a tree. A lock
242 * is not taken, so a concurrent writer may put a different node
243 * at the root of the tree. See btrfs_lock_root_node for the
244 * looping required.
245 *
246 * The extent buffer returned by this has a reference taken, so
247 * it won't disappear. It may stop being the root of the tree
248 * at any time because there are no locks held.
249 */
250struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
251{
252 struct extent_buffer *eb;
253
254 while (1) {
255 rcu_read_lock();
256 eb = rcu_dereference(root->node);
257
258 /*
259 * RCU really hurts here, we could free up the root node because
260 * it was COWed but we may not get the new root node yet so do
261 * the inc_not_zero dance and if it doesn't work then
262 * synchronize_rcu and try again.
263 */
264 if (atomic_inc_not_zero(&eb->refs)) {
265 rcu_read_unlock();
266 break;
267 }
268 rcu_read_unlock();
269 synchronize_rcu();
270 }
271 return eb;
272}
273
274/*
275 * Cowonly root (not-shareable trees, everything not subvolume or reloc roots),
276 * just get put onto a simple dirty list. Transaction walks this list to make
277 * sure they get properly updated on disk.
278 */
279static void add_root_to_dirty_list(struct btrfs_root *root)
280{
281 struct btrfs_fs_info *fs_info = root->fs_info;
282
283 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
284 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
285 return;
286
287 spin_lock(&fs_info->trans_lock);
288 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
289 /* Want the extent tree to be the last on the list */
290 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
291 list_move_tail(&root->dirty_list,
292 &fs_info->dirty_cowonly_roots);
293 else
294 list_move(&root->dirty_list,
295 &fs_info->dirty_cowonly_roots);
296 }
297 spin_unlock(&fs_info->trans_lock);
298}
299
300/*
301 * used by snapshot creation to make a copy of a root for a tree with
302 * a given objectid. The buffer with the new root node is returned in
303 * cow_ret, and this func returns zero on success or a negative error code.
304 */
305int btrfs_copy_root(struct btrfs_trans_handle *trans,
306 struct btrfs_root *root,
307 struct extent_buffer *buf,
308 struct extent_buffer **cow_ret, u64 new_root_objectid)
309{
310 struct btrfs_fs_info *fs_info = root->fs_info;
311 struct extent_buffer *cow;
312 int ret = 0;
313 int level;
314 struct btrfs_disk_key disk_key;
315
316 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
317 trans->transid != fs_info->running_transaction->transid);
318 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
319 trans->transid != root->last_trans);
320
321 level = btrfs_header_level(buf);
322 if (level == 0)
323 btrfs_item_key(buf, &disk_key, 0);
324 else
325 btrfs_node_key(buf, &disk_key, 0);
326
327 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
328 &disk_key, level, buf->start, 0,
329 BTRFS_NESTING_NEW_ROOT);
330 if (IS_ERR(cow))
331 return PTR_ERR(cow);
332
333 copy_extent_buffer_full(cow, buf);
334 btrfs_set_header_bytenr(cow, cow->start);
335 btrfs_set_header_generation(cow, trans->transid);
336 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
337 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
338 BTRFS_HEADER_FLAG_RELOC);
339 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
340 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
341 else
342 btrfs_set_header_owner(cow, new_root_objectid);
343
344 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
345
346 WARN_ON(btrfs_header_generation(buf) > trans->transid);
347 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
348 ret = btrfs_inc_ref(trans, root, cow, 1);
349 else
350 ret = btrfs_inc_ref(trans, root, cow, 0);
351 if (ret) {
352 btrfs_tree_unlock(cow);
353 free_extent_buffer(cow);
354 btrfs_abort_transaction(trans, ret);
355 return ret;
356 }
357
358 btrfs_mark_buffer_dirty(cow);
359 *cow_ret = cow;
360 return 0;
361}
362
363/*
364 * check if the tree block can be shared by multiple trees
365 */
366int btrfs_block_can_be_shared(struct btrfs_root *root,
367 struct extent_buffer *buf)
368{
369 /*
370 * Tree blocks not in shareable trees and tree roots are never shared.
371 * If a block was allocated after the last snapshot and the block was
372 * not allocated by tree relocation, we know the block is not shared.
373 */
374 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
375 buf != root->node && buf != root->commit_root &&
376 (btrfs_header_generation(buf) <=
377 btrfs_root_last_snapshot(&root->root_item) ||
378 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
379 return 1;
380
381 return 0;
382}
383
384static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
385 struct btrfs_root *root,
386 struct extent_buffer *buf,
387 struct extent_buffer *cow,
388 int *last_ref)
389{
390 struct btrfs_fs_info *fs_info = root->fs_info;
391 u64 refs;
392 u64 owner;
393 u64 flags;
394 u64 new_flags = 0;
395 int ret;
396
397 /*
398 * Backrefs update rules:
399 *
400 * Always use full backrefs for extent pointers in tree block
401 * allocated by tree relocation.
402 *
403 * If a shared tree block is no longer referenced by its owner
404 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
405 * use full backrefs for extent pointers in tree block.
406 *
407 * If a tree block is been relocating
408 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
409 * use full backrefs for extent pointers in tree block.
410 * The reason for this is some operations (such as drop tree)
411 * are only allowed for blocks use full backrefs.
412 */
413
414 if (btrfs_block_can_be_shared(root, buf)) {
415 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
416 btrfs_header_level(buf), 1,
417 &refs, &flags);
418 if (ret)
419 return ret;
420 if (refs == 0) {
421 ret = -EROFS;
422 btrfs_handle_fs_error(fs_info, ret, NULL);
423 return ret;
424 }
425 } else {
426 refs = 1;
427 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
428 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
429 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
430 else
431 flags = 0;
432 }
433
434 owner = btrfs_header_owner(buf);
435 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
436 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
437
438 if (refs > 1) {
439 if ((owner == root->root_key.objectid ||
440 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
441 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
442 ret = btrfs_inc_ref(trans, root, buf, 1);
443 if (ret)
444 return ret;
445
446 if (root->root_key.objectid ==
447 BTRFS_TREE_RELOC_OBJECTID) {
448 ret = btrfs_dec_ref(trans, root, buf, 0);
449 if (ret)
450 return ret;
451 ret = btrfs_inc_ref(trans, root, cow, 1);
452 if (ret)
453 return ret;
454 }
455 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
456 } else {
457
458 if (root->root_key.objectid ==
459 BTRFS_TREE_RELOC_OBJECTID)
460 ret = btrfs_inc_ref(trans, root, cow, 1);
461 else
462 ret = btrfs_inc_ref(trans, root, cow, 0);
463 if (ret)
464 return ret;
465 }
466 if (new_flags != 0) {
467 int level = btrfs_header_level(buf);
468
469 ret = btrfs_set_disk_extent_flags(trans, buf,
470 new_flags, level);
471 if (ret)
472 return ret;
473 }
474 } else {
475 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
476 if (root->root_key.objectid ==
477 BTRFS_TREE_RELOC_OBJECTID)
478 ret = btrfs_inc_ref(trans, root, cow, 1);
479 else
480 ret = btrfs_inc_ref(trans, root, cow, 0);
481 if (ret)
482 return ret;
483 ret = btrfs_dec_ref(trans, root, buf, 1);
484 if (ret)
485 return ret;
486 }
487 btrfs_clean_tree_block(buf);
488 *last_ref = 1;
489 }
490 return 0;
491}
492
493/*
494 * does the dirty work in cow of a single block. The parent block (if
495 * supplied) is updated to point to the new cow copy. The new buffer is marked
496 * dirty and returned locked. If you modify the block it needs to be marked
497 * dirty again.
498 *
499 * search_start -- an allocation hint for the new block
500 *
501 * empty_size -- a hint that you plan on doing more cow. This is the size in
502 * bytes the allocator should try to find free next to the block it returns.
503 * This is just a hint and may be ignored by the allocator.
504 */
505static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
506 struct btrfs_root *root,
507 struct extent_buffer *buf,
508 struct extent_buffer *parent, int parent_slot,
509 struct extent_buffer **cow_ret,
510 u64 search_start, u64 empty_size,
511 enum btrfs_lock_nesting nest)
512{
513 struct btrfs_fs_info *fs_info = root->fs_info;
514 struct btrfs_disk_key disk_key;
515 struct extent_buffer *cow;
516 int level, ret;
517 int last_ref = 0;
518 int unlock_orig = 0;
519 u64 parent_start = 0;
520
521 if (*cow_ret == buf)
522 unlock_orig = 1;
523
524 btrfs_assert_tree_write_locked(buf);
525
526 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
527 trans->transid != fs_info->running_transaction->transid);
528 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
529 trans->transid != root->last_trans);
530
531 level = btrfs_header_level(buf);
532
533 if (level == 0)
534 btrfs_item_key(buf, &disk_key, 0);
535 else
536 btrfs_node_key(buf, &disk_key, 0);
537
538 if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
539 parent_start = parent->start;
540
541 cow = btrfs_alloc_tree_block(trans, root, parent_start,
542 root->root_key.objectid, &disk_key, level,
543 search_start, empty_size, nest);
544 if (IS_ERR(cow))
545 return PTR_ERR(cow);
546
547 /* cow is set to blocking by btrfs_init_new_buffer */
548
549 copy_extent_buffer_full(cow, buf);
550 btrfs_set_header_bytenr(cow, cow->start);
551 btrfs_set_header_generation(cow, trans->transid);
552 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
553 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
554 BTRFS_HEADER_FLAG_RELOC);
555 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
556 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
557 else
558 btrfs_set_header_owner(cow, root->root_key.objectid);
559
560 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
561
562 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
563 if (ret) {
564 btrfs_tree_unlock(cow);
565 free_extent_buffer(cow);
566 btrfs_abort_transaction(trans, ret);
567 return ret;
568 }
569
570 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
571 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
572 if (ret) {
573 btrfs_tree_unlock(cow);
574 free_extent_buffer(cow);
575 btrfs_abort_transaction(trans, ret);
576 return ret;
577 }
578 }
579
580 if (buf == root->node) {
581 WARN_ON(parent && parent != buf);
582 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
583 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
584 parent_start = buf->start;
585
586 atomic_inc(&cow->refs);
587 ret = btrfs_tree_mod_log_insert_root(root->node, cow, true);
588 BUG_ON(ret < 0);
589 rcu_assign_pointer(root->node, cow);
590
591 btrfs_free_tree_block(trans, btrfs_root_id(root), buf,
592 parent_start, last_ref);
593 free_extent_buffer(buf);
594 add_root_to_dirty_list(root);
595 } else {
596 WARN_ON(trans->transid != btrfs_header_generation(parent));
597 btrfs_tree_mod_log_insert_key(parent, parent_slot,
598 BTRFS_MOD_LOG_KEY_REPLACE);
599 btrfs_set_node_blockptr(parent, parent_slot,
600 cow->start);
601 btrfs_set_node_ptr_generation(parent, parent_slot,
602 trans->transid);
603 btrfs_mark_buffer_dirty(parent);
604 if (last_ref) {
605 ret = btrfs_tree_mod_log_free_eb(buf);
606 if (ret) {
607 btrfs_tree_unlock(cow);
608 free_extent_buffer(cow);
609 btrfs_abort_transaction(trans, ret);
610 return ret;
611 }
612 }
613 btrfs_free_tree_block(trans, btrfs_root_id(root), buf,
614 parent_start, last_ref);
615 }
616 if (unlock_orig)
617 btrfs_tree_unlock(buf);
618 free_extent_buffer_stale(buf);
619 btrfs_mark_buffer_dirty(cow);
620 *cow_ret = cow;
621 return 0;
622}
623
624static inline int should_cow_block(struct btrfs_trans_handle *trans,
625 struct btrfs_root *root,
626 struct extent_buffer *buf)
627{
628 if (btrfs_is_testing(root->fs_info))
629 return 0;
630
631 /* Ensure we can see the FORCE_COW bit */
632 smp_mb__before_atomic();
633
634 /*
635 * We do not need to cow a block if
636 * 1) this block is not created or changed in this transaction;
637 * 2) this block does not belong to TREE_RELOC tree;
638 * 3) the root is not forced COW.
639 *
640 * What is forced COW:
641 * when we create snapshot during committing the transaction,
642 * after we've finished copying src root, we must COW the shared
643 * block to ensure the metadata consistency.
644 */
645 if (btrfs_header_generation(buf) == trans->transid &&
646 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
647 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
648 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
649 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
650 return 0;
651 return 1;
652}
653
654/*
655 * cows a single block, see __btrfs_cow_block for the real work.
656 * This version of it has extra checks so that a block isn't COWed more than
657 * once per transaction, as long as it hasn't been written yet
658 */
659noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
660 struct btrfs_root *root, struct extent_buffer *buf,
661 struct extent_buffer *parent, int parent_slot,
662 struct extent_buffer **cow_ret,
663 enum btrfs_lock_nesting nest)
664{
665 struct btrfs_fs_info *fs_info = root->fs_info;
666 u64 search_start;
667 int ret;
668
669 if (test_bit(BTRFS_ROOT_DELETING, &root->state))
670 btrfs_err(fs_info,
671 "COW'ing blocks on a fs root that's being dropped");
672
673 if (trans->transaction != fs_info->running_transaction)
674 WARN(1, KERN_CRIT "trans %llu running %llu\n",
675 trans->transid,
676 fs_info->running_transaction->transid);
677
678 if (trans->transid != fs_info->generation)
679 WARN(1, KERN_CRIT "trans %llu running %llu\n",
680 trans->transid, fs_info->generation);
681
682 if (!should_cow_block(trans, root, buf)) {
683 *cow_ret = buf;
684 return 0;
685 }
686
687 search_start = buf->start & ~((u64)SZ_1G - 1);
688
689 /*
690 * Before CoWing this block for later modification, check if it's
691 * the subtree root and do the delayed subtree trace if needed.
692 *
693 * Also We don't care about the error, as it's handled internally.
694 */
695 btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
696 ret = __btrfs_cow_block(trans, root, buf, parent,
697 parent_slot, cow_ret, search_start, 0, nest);
698
699 trace_btrfs_cow_block(root, buf, *cow_ret);
700
701 return ret;
702}
703ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO);
704
705/*
706 * helper function for defrag to decide if two blocks pointed to by a
707 * node are actually close by
708 */
709static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
710{
711 if (blocknr < other && other - (blocknr + blocksize) < 32768)
712 return 1;
713 if (blocknr > other && blocknr - (other + blocksize) < 32768)
714 return 1;
715 return 0;
716}
717
718#ifdef __LITTLE_ENDIAN
719
720/*
721 * Compare two keys, on little-endian the disk order is same as CPU order and
722 * we can avoid the conversion.
723 */
724static int comp_keys(const struct btrfs_disk_key *disk_key,
725 const struct btrfs_key *k2)
726{
727 const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key;
728
729 return btrfs_comp_cpu_keys(k1, k2);
730}
731
732#else
733
734/*
735 * compare two keys in a memcmp fashion
736 */
737static int comp_keys(const struct btrfs_disk_key *disk,
738 const struct btrfs_key *k2)
739{
740 struct btrfs_key k1;
741
742 btrfs_disk_key_to_cpu(&k1, disk);
743
744 return btrfs_comp_cpu_keys(&k1, k2);
745}
746#endif
747
748/*
749 * same as comp_keys only with two btrfs_key's
750 */
751int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
752{
753 if (k1->objectid > k2->objectid)
754 return 1;
755 if (k1->objectid < k2->objectid)
756 return -1;
757 if (k1->type > k2->type)
758 return 1;
759 if (k1->type < k2->type)
760 return -1;
761 if (k1->offset > k2->offset)
762 return 1;
763 if (k1->offset < k2->offset)
764 return -1;
765 return 0;
766}
767
768/*
769 * this is used by the defrag code to go through all the
770 * leaves pointed to by a node and reallocate them so that
771 * disk order is close to key order
772 */
773int btrfs_realloc_node(struct btrfs_trans_handle *trans,
774 struct btrfs_root *root, struct extent_buffer *parent,
775 int start_slot, u64 *last_ret,
776 struct btrfs_key *progress)
777{
778 struct btrfs_fs_info *fs_info = root->fs_info;
779 struct extent_buffer *cur;
780 u64 blocknr;
781 u64 search_start = *last_ret;
782 u64 last_block = 0;
783 u64 other;
784 u32 parent_nritems;
785 int end_slot;
786 int i;
787 int err = 0;
788 u32 blocksize;
789 int progress_passed = 0;
790 struct btrfs_disk_key disk_key;
791
792 WARN_ON(trans->transaction != fs_info->running_transaction);
793 WARN_ON(trans->transid != fs_info->generation);
794
795 parent_nritems = btrfs_header_nritems(parent);
796 blocksize = fs_info->nodesize;
797 end_slot = parent_nritems - 1;
798
799 if (parent_nritems <= 1)
800 return 0;
801
802 for (i = start_slot; i <= end_slot; i++) {
803 int close = 1;
804
805 btrfs_node_key(parent, &disk_key, i);
806 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
807 continue;
808
809 progress_passed = 1;
810 blocknr = btrfs_node_blockptr(parent, i);
811 if (last_block == 0)
812 last_block = blocknr;
813
814 if (i > 0) {
815 other = btrfs_node_blockptr(parent, i - 1);
816 close = close_blocks(blocknr, other, blocksize);
817 }
818 if (!close && i < end_slot) {
819 other = btrfs_node_blockptr(parent, i + 1);
820 close = close_blocks(blocknr, other, blocksize);
821 }
822 if (close) {
823 last_block = blocknr;
824 continue;
825 }
826
827 cur = btrfs_read_node_slot(parent, i);
828 if (IS_ERR(cur))
829 return PTR_ERR(cur);
830 if (search_start == 0)
831 search_start = last_block;
832
833 btrfs_tree_lock(cur);
834 err = __btrfs_cow_block(trans, root, cur, parent, i,
835 &cur, search_start,
836 min(16 * blocksize,
837 (end_slot - i) * blocksize),
838 BTRFS_NESTING_COW);
839 if (err) {
840 btrfs_tree_unlock(cur);
841 free_extent_buffer(cur);
842 break;
843 }
844 search_start = cur->start;
845 last_block = cur->start;
846 *last_ret = search_start;
847 btrfs_tree_unlock(cur);
848 free_extent_buffer(cur);
849 }
850 return err;
851}
852
853/*
854 * Search for a key in the given extent_buffer.
855 *
856 * The lower boundary for the search is specified by the slot number @low. Use a
857 * value of 0 to search over the whole extent buffer.
858 *
859 * The slot in the extent buffer is returned via @slot. If the key exists in the
860 * extent buffer, then @slot will point to the slot where the key is, otherwise
861 * it points to the slot where you would insert the key.
862 *
863 * Slot may point to the total number of items (i.e. one position beyond the last
864 * key) if the key is bigger than the last key in the extent buffer.
865 */
866static noinline int generic_bin_search(struct extent_buffer *eb, int low,
867 const struct btrfs_key *key, int *slot)
868{
869 unsigned long p;
870 int item_size;
871 int high = btrfs_header_nritems(eb);
872 int ret;
873 const int key_size = sizeof(struct btrfs_disk_key);
874
875 if (low > high) {
876 btrfs_err(eb->fs_info,
877 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
878 __func__, low, high, eb->start,
879 btrfs_header_owner(eb), btrfs_header_level(eb));
880 return -EINVAL;
881 }
882
883 if (btrfs_header_level(eb) == 0) {
884 p = offsetof(struct btrfs_leaf, items);
885 item_size = sizeof(struct btrfs_item);
886 } else {
887 p = offsetof(struct btrfs_node, ptrs);
888 item_size = sizeof(struct btrfs_key_ptr);
889 }
890
891 while (low < high) {
892 unsigned long oip;
893 unsigned long offset;
894 struct btrfs_disk_key *tmp;
895 struct btrfs_disk_key unaligned;
896 int mid;
897
898 mid = (low + high) / 2;
899 offset = p + mid * item_size;
900 oip = offset_in_page(offset);
901
902 if (oip + key_size <= PAGE_SIZE) {
903 const unsigned long idx = get_eb_page_index(offset);
904 char *kaddr = page_address(eb->pages[idx]);
905
906 oip = get_eb_offset_in_page(eb, offset);
907 tmp = (struct btrfs_disk_key *)(kaddr + oip);
908 } else {
909 read_extent_buffer(eb, &unaligned, offset, key_size);
910 tmp = &unaligned;
911 }
912
913 ret = comp_keys(tmp, key);
914
915 if (ret < 0)
916 low = mid + 1;
917 else if (ret > 0)
918 high = mid;
919 else {
920 *slot = mid;
921 return 0;
922 }
923 }
924 *slot = low;
925 return 1;
926}
927
928/*
929 * Simple binary search on an extent buffer. Works for both leaves and nodes, and
930 * always searches over the whole range of keys (slot 0 to slot 'nritems - 1').
931 */
932int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
933 int *slot)
934{
935 return generic_bin_search(eb, 0, key, slot);
936}
937
938static void root_add_used(struct btrfs_root *root, u32 size)
939{
940 spin_lock(&root->accounting_lock);
941 btrfs_set_root_used(&root->root_item,
942 btrfs_root_used(&root->root_item) + size);
943 spin_unlock(&root->accounting_lock);
944}
945
946static void root_sub_used(struct btrfs_root *root, u32 size)
947{
948 spin_lock(&root->accounting_lock);
949 btrfs_set_root_used(&root->root_item,
950 btrfs_root_used(&root->root_item) - size);
951 spin_unlock(&root->accounting_lock);
952}
953
954/* given a node and slot number, this reads the blocks it points to. The
955 * extent buffer is returned with a reference taken (but unlocked).
956 */
957struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
958 int slot)
959{
960 int level = btrfs_header_level(parent);
961 struct btrfs_tree_parent_check check = { 0 };
962 struct extent_buffer *eb;
963
964 if (slot < 0 || slot >= btrfs_header_nritems(parent))
965 return ERR_PTR(-ENOENT);
966
967 BUG_ON(level == 0);
968
969 check.level = level - 1;
970 check.transid = btrfs_node_ptr_generation(parent, slot);
971 check.owner_root = btrfs_header_owner(parent);
972 check.has_first_key = true;
973 btrfs_node_key_to_cpu(parent, &check.first_key, slot);
974
975 eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
976 &check);
977 if (IS_ERR(eb))
978 return eb;
979 if (!extent_buffer_uptodate(eb)) {
980 free_extent_buffer(eb);
981 return ERR_PTR(-EIO);
982 }
983
984 return eb;
985}
986
987/*
988 * node level balancing, used to make sure nodes are in proper order for
989 * item deletion. We balance from the top down, so we have to make sure
990 * that a deletion won't leave an node completely empty later on.
991 */
992static noinline int balance_level(struct btrfs_trans_handle *trans,
993 struct btrfs_root *root,
994 struct btrfs_path *path, int level)
995{
996 struct btrfs_fs_info *fs_info = root->fs_info;
997 struct extent_buffer *right = NULL;
998 struct extent_buffer *mid;
999 struct extent_buffer *left = NULL;
1000 struct extent_buffer *parent = NULL;
1001 int ret = 0;
1002 int wret;
1003 int pslot;
1004 int orig_slot = path->slots[level];
1005 u64 orig_ptr;
1006
1007 ASSERT(level > 0);
1008
1009 mid = path->nodes[level];
1010
1011 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK);
1012 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1013
1014 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1015
1016 if (level < BTRFS_MAX_LEVEL - 1) {
1017 parent = path->nodes[level + 1];
1018 pslot = path->slots[level + 1];
1019 }
1020
1021 /*
1022 * deal with the case where there is only one pointer in the root
1023 * by promoting the node below to a root
1024 */
1025 if (!parent) {
1026 struct extent_buffer *child;
1027
1028 if (btrfs_header_nritems(mid) != 1)
1029 return 0;
1030
1031 /* promote the child to a root */
1032 child = btrfs_read_node_slot(mid, 0);
1033 if (IS_ERR(child)) {
1034 ret = PTR_ERR(child);
1035 btrfs_handle_fs_error(fs_info, ret, NULL);
1036 goto enospc;
1037 }
1038
1039 btrfs_tree_lock(child);
1040 ret = btrfs_cow_block(trans, root, child, mid, 0, &child,
1041 BTRFS_NESTING_COW);
1042 if (ret) {
1043 btrfs_tree_unlock(child);
1044 free_extent_buffer(child);
1045 goto enospc;
1046 }
1047
1048 ret = btrfs_tree_mod_log_insert_root(root->node, child, true);
1049 BUG_ON(ret < 0);
1050 rcu_assign_pointer(root->node, child);
1051
1052 add_root_to_dirty_list(root);
1053 btrfs_tree_unlock(child);
1054
1055 path->locks[level] = 0;
1056 path->nodes[level] = NULL;
1057 btrfs_clean_tree_block(mid);
1058 btrfs_tree_unlock(mid);
1059 /* once for the path */
1060 free_extent_buffer(mid);
1061
1062 root_sub_used(root, mid->len);
1063 btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1);
1064 /* once for the root ptr */
1065 free_extent_buffer_stale(mid);
1066 return 0;
1067 }
1068 if (btrfs_header_nritems(mid) >
1069 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1070 return 0;
1071
1072 left = btrfs_read_node_slot(parent, pslot - 1);
1073 if (IS_ERR(left))
1074 left = NULL;
1075
1076 if (left) {
1077 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
1078 wret = btrfs_cow_block(trans, root, left,
1079 parent, pslot - 1, &left,
1080 BTRFS_NESTING_LEFT_COW);
1081 if (wret) {
1082 ret = wret;
1083 goto enospc;
1084 }
1085 }
1086
1087 right = btrfs_read_node_slot(parent, pslot + 1);
1088 if (IS_ERR(right))
1089 right = NULL;
1090
1091 if (right) {
1092 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
1093 wret = btrfs_cow_block(trans, root, right,
1094 parent, pslot + 1, &right,
1095 BTRFS_NESTING_RIGHT_COW);
1096 if (wret) {
1097 ret = wret;
1098 goto enospc;
1099 }
1100 }
1101
1102 /* first, try to make some room in the middle buffer */
1103 if (left) {
1104 orig_slot += btrfs_header_nritems(left);
1105 wret = push_node_left(trans, left, mid, 1);
1106 if (wret < 0)
1107 ret = wret;
1108 }
1109
1110 /*
1111 * then try to empty the right most buffer into the middle
1112 */
1113 if (right) {
1114 wret = push_node_left(trans, mid, right, 1);
1115 if (wret < 0 && wret != -ENOSPC)
1116 ret = wret;
1117 if (btrfs_header_nritems(right) == 0) {
1118 btrfs_clean_tree_block(right);
1119 btrfs_tree_unlock(right);
1120 del_ptr(root, path, level + 1, pslot + 1);
1121 root_sub_used(root, right->len);
1122 btrfs_free_tree_block(trans, btrfs_root_id(root), right,
1123 0, 1);
1124 free_extent_buffer_stale(right);
1125 right = NULL;
1126 } else {
1127 struct btrfs_disk_key right_key;
1128 btrfs_node_key(right, &right_key, 0);
1129 ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1,
1130 BTRFS_MOD_LOG_KEY_REPLACE);
1131 BUG_ON(ret < 0);
1132 btrfs_set_node_key(parent, &right_key, pslot + 1);
1133 btrfs_mark_buffer_dirty(parent);
1134 }
1135 }
1136 if (btrfs_header_nritems(mid) == 1) {
1137 /*
1138 * we're not allowed to leave a node with one item in the
1139 * tree during a delete. A deletion from lower in the tree
1140 * could try to delete the only pointer in this node.
1141 * So, pull some keys from the left.
1142 * There has to be a left pointer at this point because
1143 * otherwise we would have pulled some pointers from the
1144 * right
1145 */
1146 if (!left) {
1147 ret = -EROFS;
1148 btrfs_handle_fs_error(fs_info, ret, NULL);
1149 goto enospc;
1150 }
1151 wret = balance_node_right(trans, mid, left);
1152 if (wret < 0) {
1153 ret = wret;
1154 goto enospc;
1155 }
1156 if (wret == 1) {
1157 wret = push_node_left(trans, left, mid, 1);
1158 if (wret < 0)
1159 ret = wret;
1160 }
1161 BUG_ON(wret == 1);
1162 }
1163 if (btrfs_header_nritems(mid) == 0) {
1164 btrfs_clean_tree_block(mid);
1165 btrfs_tree_unlock(mid);
1166 del_ptr(root, path, level + 1, pslot);
1167 root_sub_used(root, mid->len);
1168 btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1);
1169 free_extent_buffer_stale(mid);
1170 mid = NULL;
1171 } else {
1172 /* update the parent key to reflect our changes */
1173 struct btrfs_disk_key mid_key;
1174 btrfs_node_key(mid, &mid_key, 0);
1175 ret = btrfs_tree_mod_log_insert_key(parent, pslot,
1176 BTRFS_MOD_LOG_KEY_REPLACE);
1177 BUG_ON(ret < 0);
1178 btrfs_set_node_key(parent, &mid_key, pslot);
1179 btrfs_mark_buffer_dirty(parent);
1180 }
1181
1182 /* update the path */
1183 if (left) {
1184 if (btrfs_header_nritems(left) > orig_slot) {
1185 atomic_inc(&left->refs);
1186 /* left was locked after cow */
1187 path->nodes[level] = left;
1188 path->slots[level + 1] -= 1;
1189 path->slots[level] = orig_slot;
1190 if (mid) {
1191 btrfs_tree_unlock(mid);
1192 free_extent_buffer(mid);
1193 }
1194 } else {
1195 orig_slot -= btrfs_header_nritems(left);
1196 path->slots[level] = orig_slot;
1197 }
1198 }
1199 /* double check we haven't messed things up */
1200 if (orig_ptr !=
1201 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1202 BUG();
1203enospc:
1204 if (right) {
1205 btrfs_tree_unlock(right);
1206 free_extent_buffer(right);
1207 }
1208 if (left) {
1209 if (path->nodes[level] != left)
1210 btrfs_tree_unlock(left);
1211 free_extent_buffer(left);
1212 }
1213 return ret;
1214}
1215
1216/* Node balancing for insertion. Here we only split or push nodes around
1217 * when they are completely full. This is also done top down, so we
1218 * have to be pessimistic.
1219 */
1220static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1221 struct btrfs_root *root,
1222 struct btrfs_path *path, int level)
1223{
1224 struct btrfs_fs_info *fs_info = root->fs_info;
1225 struct extent_buffer *right = NULL;
1226 struct extent_buffer *mid;
1227 struct extent_buffer *left = NULL;
1228 struct extent_buffer *parent = NULL;
1229 int ret = 0;
1230 int wret;
1231 int pslot;
1232 int orig_slot = path->slots[level];
1233
1234 if (level == 0)
1235 return 1;
1236
1237 mid = path->nodes[level];
1238 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1239
1240 if (level < BTRFS_MAX_LEVEL - 1) {
1241 parent = path->nodes[level + 1];
1242 pslot = path->slots[level + 1];
1243 }
1244
1245 if (!parent)
1246 return 1;
1247
1248 left = btrfs_read_node_slot(parent, pslot - 1);
1249 if (IS_ERR(left))
1250 left = NULL;
1251
1252 /* first, try to make some room in the middle buffer */
1253 if (left) {
1254 u32 left_nr;
1255
1256 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
1257
1258 left_nr = btrfs_header_nritems(left);
1259 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
1260 wret = 1;
1261 } else {
1262 ret = btrfs_cow_block(trans, root, left, parent,
1263 pslot - 1, &left,
1264 BTRFS_NESTING_LEFT_COW);
1265 if (ret)
1266 wret = 1;
1267 else {
1268 wret = push_node_left(trans, left, mid, 0);
1269 }
1270 }
1271 if (wret < 0)
1272 ret = wret;
1273 if (wret == 0) {
1274 struct btrfs_disk_key disk_key;
1275 orig_slot += left_nr;
1276 btrfs_node_key(mid, &disk_key, 0);
1277 ret = btrfs_tree_mod_log_insert_key(parent, pslot,
1278 BTRFS_MOD_LOG_KEY_REPLACE);
1279 BUG_ON(ret < 0);
1280 btrfs_set_node_key(parent, &disk_key, pslot);
1281 btrfs_mark_buffer_dirty(parent);
1282 if (btrfs_header_nritems(left) > orig_slot) {
1283 path->nodes[level] = left;
1284 path->slots[level + 1] -= 1;
1285 path->slots[level] = orig_slot;
1286 btrfs_tree_unlock(mid);
1287 free_extent_buffer(mid);
1288 } else {
1289 orig_slot -=
1290 btrfs_header_nritems(left);
1291 path->slots[level] = orig_slot;
1292 btrfs_tree_unlock(left);
1293 free_extent_buffer(left);
1294 }
1295 return 0;
1296 }
1297 btrfs_tree_unlock(left);
1298 free_extent_buffer(left);
1299 }
1300 right = btrfs_read_node_slot(parent, pslot + 1);
1301 if (IS_ERR(right))
1302 right = NULL;
1303
1304 /*
1305 * then try to empty the right most buffer into the middle
1306 */
1307 if (right) {
1308 u32 right_nr;
1309
1310 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
1311
1312 right_nr = btrfs_header_nritems(right);
1313 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
1314 wret = 1;
1315 } else {
1316 ret = btrfs_cow_block(trans, root, right,
1317 parent, pslot + 1,
1318 &right, BTRFS_NESTING_RIGHT_COW);
1319 if (ret)
1320 wret = 1;
1321 else {
1322 wret = balance_node_right(trans, right, mid);
1323 }
1324 }
1325 if (wret < 0)
1326 ret = wret;
1327 if (wret == 0) {
1328 struct btrfs_disk_key disk_key;
1329
1330 btrfs_node_key(right, &disk_key, 0);
1331 ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1,
1332 BTRFS_MOD_LOG_KEY_REPLACE);
1333 BUG_ON(ret < 0);
1334 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1335 btrfs_mark_buffer_dirty(parent);
1336
1337 if (btrfs_header_nritems(mid) <= orig_slot) {
1338 path->nodes[level] = right;
1339 path->slots[level + 1] += 1;
1340 path->slots[level] = orig_slot -
1341 btrfs_header_nritems(mid);
1342 btrfs_tree_unlock(mid);
1343 free_extent_buffer(mid);
1344 } else {
1345 btrfs_tree_unlock(right);
1346 free_extent_buffer(right);
1347 }
1348 return 0;
1349 }
1350 btrfs_tree_unlock(right);
1351 free_extent_buffer(right);
1352 }
1353 return 1;
1354}
1355
1356/*
1357 * readahead one full node of leaves, finding things that are close
1358 * to the block in 'slot', and triggering ra on them.
1359 */
1360static void reada_for_search(struct btrfs_fs_info *fs_info,
1361 struct btrfs_path *path,
1362 int level, int slot, u64 objectid)
1363{
1364 struct extent_buffer *node;
1365 struct btrfs_disk_key disk_key;
1366 u32 nritems;
1367 u64 search;
1368 u64 target;
1369 u64 nread = 0;
1370 u64 nread_max;
1371 u32 nr;
1372 u32 blocksize;
1373 u32 nscan = 0;
1374
1375 if (level != 1 && path->reada != READA_FORWARD_ALWAYS)
1376 return;
1377
1378 if (!path->nodes[level])
1379 return;
1380
1381 node = path->nodes[level];
1382
1383 /*
1384 * Since the time between visiting leaves is much shorter than the time
1385 * between visiting nodes, limit read ahead of nodes to 1, to avoid too
1386 * much IO at once (possibly random).
1387 */
1388 if (path->reada == READA_FORWARD_ALWAYS) {
1389 if (level > 1)
1390 nread_max = node->fs_info->nodesize;
1391 else
1392 nread_max = SZ_128K;
1393 } else {
1394 nread_max = SZ_64K;
1395 }
1396
1397 search = btrfs_node_blockptr(node, slot);
1398 blocksize = fs_info->nodesize;
1399 if (path->reada != READA_FORWARD_ALWAYS) {
1400 struct extent_buffer *eb;
1401
1402 eb = find_extent_buffer(fs_info, search);
1403 if (eb) {
1404 free_extent_buffer(eb);
1405 return;
1406 }
1407 }
1408
1409 target = search;
1410
1411 nritems = btrfs_header_nritems(node);
1412 nr = slot;
1413
1414 while (1) {
1415 if (path->reada == READA_BACK) {
1416 if (nr == 0)
1417 break;
1418 nr--;
1419 } else if (path->reada == READA_FORWARD ||
1420 path->reada == READA_FORWARD_ALWAYS) {
1421 nr++;
1422 if (nr >= nritems)
1423 break;
1424 }
1425 if (path->reada == READA_BACK && objectid) {
1426 btrfs_node_key(node, &disk_key, nr);
1427 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1428 break;
1429 }
1430 search = btrfs_node_blockptr(node, nr);
1431 if (path->reada == READA_FORWARD_ALWAYS ||
1432 (search <= target && target - search <= 65536) ||
1433 (search > target && search - target <= 65536)) {
1434 btrfs_readahead_node_child(node, nr);
1435 nread += blocksize;
1436 }
1437 nscan++;
1438 if (nread > nread_max || nscan > 32)
1439 break;
1440 }
1441}
1442
1443static noinline void reada_for_balance(struct btrfs_path *path, int level)
1444{
1445 struct extent_buffer *parent;
1446 int slot;
1447 int nritems;
1448
1449 parent = path->nodes[level + 1];
1450 if (!parent)
1451 return;
1452
1453 nritems = btrfs_header_nritems(parent);
1454 slot = path->slots[level + 1];
1455
1456 if (slot > 0)
1457 btrfs_readahead_node_child(parent, slot - 1);
1458 if (slot + 1 < nritems)
1459 btrfs_readahead_node_child(parent, slot + 1);
1460}
1461
1462
1463/*
1464 * when we walk down the tree, it is usually safe to unlock the higher layers
1465 * in the tree. The exceptions are when our path goes through slot 0, because
1466 * operations on the tree might require changing key pointers higher up in the
1467 * tree.
1468 *
1469 * callers might also have set path->keep_locks, which tells this code to keep
1470 * the lock if the path points to the last slot in the block. This is part of
1471 * walking through the tree, and selecting the next slot in the higher block.
1472 *
1473 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1474 * if lowest_unlock is 1, level 0 won't be unlocked
1475 */
1476static noinline void unlock_up(struct btrfs_path *path, int level,
1477 int lowest_unlock, int min_write_lock_level,
1478 int *write_lock_level)
1479{
1480 int i;
1481 int skip_level = level;
1482 bool check_skip = true;
1483
1484 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1485 if (!path->nodes[i])
1486 break;
1487 if (!path->locks[i])
1488 break;
1489
1490 if (check_skip) {
1491 if (path->slots[i] == 0) {
1492 skip_level = i + 1;
1493 continue;
1494 }
1495
1496 if (path->keep_locks) {
1497 u32 nritems;
1498
1499 nritems = btrfs_header_nritems(path->nodes[i]);
1500 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1501 skip_level = i + 1;
1502 continue;
1503 }
1504 }
1505 }
1506
1507 if (i >= lowest_unlock && i > skip_level) {
1508 check_skip = false;
1509 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
1510 path->locks[i] = 0;
1511 if (write_lock_level &&
1512 i > min_write_lock_level &&
1513 i <= *write_lock_level) {
1514 *write_lock_level = i - 1;
1515 }
1516 }
1517 }
1518}
1519
1520/*
1521 * Helper function for btrfs_search_slot() and other functions that do a search
1522 * on a btree. The goal is to find a tree block in the cache (the radix tree at
1523 * fs_info->buffer_radix), but if we can't find it, or it's not up to date, read
1524 * its pages from disk.
1525 *
1526 * Returns -EAGAIN, with the path unlocked, if the caller needs to repeat the
1527 * whole btree search, starting again from the current root node.
1528 */
1529static int
1530read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
1531 struct extent_buffer **eb_ret, int level, int slot,
1532 const struct btrfs_key *key)
1533{
1534 struct btrfs_fs_info *fs_info = root->fs_info;
1535 struct btrfs_tree_parent_check check = { 0 };
1536 u64 blocknr;
1537 u64 gen;
1538 struct extent_buffer *tmp;
1539 int ret;
1540 int parent_level;
1541 bool unlock_up;
1542
1543 unlock_up = ((level + 1 < BTRFS_MAX_LEVEL) && p->locks[level + 1]);
1544 blocknr = btrfs_node_blockptr(*eb_ret, slot);
1545 gen = btrfs_node_ptr_generation(*eb_ret, slot);
1546 parent_level = btrfs_header_level(*eb_ret);
1547 btrfs_node_key_to_cpu(*eb_ret, &check.first_key, slot);
1548 check.has_first_key = true;
1549 check.level = parent_level - 1;
1550 check.transid = gen;
1551 check.owner_root = root->root_key.objectid;
1552
1553 /*
1554 * If we need to read an extent buffer from disk and we are holding locks
1555 * on upper level nodes, we unlock all the upper nodes before reading the
1556 * extent buffer, and then return -EAGAIN to the caller as it needs to
1557 * restart the search. We don't release the lock on the current level
1558 * because we need to walk this node to figure out which blocks to read.
1559 */
1560 tmp = find_extent_buffer(fs_info, blocknr);
1561 if (tmp) {
1562 if (p->reada == READA_FORWARD_ALWAYS)
1563 reada_for_search(fs_info, p, level, slot, key->objectid);
1564
1565 /* first we do an atomic uptodate check */
1566 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
1567 /*
1568 * Do extra check for first_key, eb can be stale due to
1569 * being cached, read from scrub, or have multiple
1570 * parents (shared tree blocks).
1571 */
1572 if (btrfs_verify_level_key(tmp,
1573 parent_level - 1, &check.first_key, gen)) {
1574 free_extent_buffer(tmp);
1575 return -EUCLEAN;
1576 }
1577 *eb_ret = tmp;
1578 return 0;
1579 }
1580
1581 if (p->nowait) {
1582 free_extent_buffer(tmp);
1583 return -EAGAIN;
1584 }
1585
1586 if (unlock_up)
1587 btrfs_unlock_up_safe(p, level + 1);
1588
1589 /* now we're allowed to do a blocking uptodate check */
1590 ret = btrfs_read_extent_buffer(tmp, &check);
1591 if (ret) {
1592 free_extent_buffer(tmp);
1593 btrfs_release_path(p);
1594 return -EIO;
1595 }
1596 if (btrfs_check_eb_owner(tmp, root->root_key.objectid)) {
1597 free_extent_buffer(tmp);
1598 btrfs_release_path(p);
1599 return -EUCLEAN;
1600 }
1601
1602 if (unlock_up)
1603 ret = -EAGAIN;
1604
1605 goto out;
1606 } else if (p->nowait) {
1607 return -EAGAIN;
1608 }
1609
1610 if (unlock_up) {
1611 btrfs_unlock_up_safe(p, level + 1);
1612 ret = -EAGAIN;
1613 } else {
1614 ret = 0;
1615 }
1616
1617 if (p->reada != READA_NONE)
1618 reada_for_search(fs_info, p, level, slot, key->objectid);
1619
1620 tmp = read_tree_block(fs_info, blocknr, &check);
1621 if (IS_ERR(tmp)) {
1622 btrfs_release_path(p);
1623 return PTR_ERR(tmp);
1624 }
1625 /*
1626 * If the read above didn't mark this buffer up to date,
1627 * it will never end up being up to date. Set ret to EIO now
1628 * and give up so that our caller doesn't loop forever
1629 * on our EAGAINs.
1630 */
1631 if (!extent_buffer_uptodate(tmp))
1632 ret = -EIO;
1633
1634out:
1635 if (ret == 0) {
1636 *eb_ret = tmp;
1637 } else {
1638 free_extent_buffer(tmp);
1639 btrfs_release_path(p);
1640 }
1641
1642 return ret;
1643}
1644
1645/*
1646 * helper function for btrfs_search_slot. This does all of the checks
1647 * for node-level blocks and does any balancing required based on
1648 * the ins_len.
1649 *
1650 * If no extra work was required, zero is returned. If we had to
1651 * drop the path, -EAGAIN is returned and btrfs_search_slot must
1652 * start over
1653 */
1654static int
1655setup_nodes_for_search(struct btrfs_trans_handle *trans,
1656 struct btrfs_root *root, struct btrfs_path *p,
1657 struct extent_buffer *b, int level, int ins_len,
1658 int *write_lock_level)
1659{
1660 struct btrfs_fs_info *fs_info = root->fs_info;
1661 int ret = 0;
1662
1663 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1664 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
1665
1666 if (*write_lock_level < level + 1) {
1667 *write_lock_level = level + 1;
1668 btrfs_release_path(p);
1669 return -EAGAIN;
1670 }
1671
1672 reada_for_balance(p, level);
1673 ret = split_node(trans, root, p, level);
1674
1675 b = p->nodes[level];
1676 } else if (ins_len < 0 && btrfs_header_nritems(b) <
1677 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
1678
1679 if (*write_lock_level < level + 1) {
1680 *write_lock_level = level + 1;
1681 btrfs_release_path(p);
1682 return -EAGAIN;
1683 }
1684
1685 reada_for_balance(p, level);
1686 ret = balance_level(trans, root, p, level);
1687 if (ret)
1688 return ret;
1689
1690 b = p->nodes[level];
1691 if (!b) {
1692 btrfs_release_path(p);
1693 return -EAGAIN;
1694 }
1695 BUG_ON(btrfs_header_nritems(b) == 1);
1696 }
1697 return ret;
1698}
1699
1700int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
1701 u64 iobjectid, u64 ioff, u8 key_type,
1702 struct btrfs_key *found_key)
1703{
1704 int ret;
1705 struct btrfs_key key;
1706 struct extent_buffer *eb;
1707
1708 ASSERT(path);
1709 ASSERT(found_key);
1710
1711 key.type = key_type;
1712 key.objectid = iobjectid;
1713 key.offset = ioff;
1714
1715 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1716 if (ret < 0)
1717 return ret;
1718
1719 eb = path->nodes[0];
1720 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1721 ret = btrfs_next_leaf(fs_root, path);
1722 if (ret)
1723 return ret;
1724 eb = path->nodes[0];
1725 }
1726
1727 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1728 if (found_key->type != key.type ||
1729 found_key->objectid != key.objectid)
1730 return 1;
1731
1732 return 0;
1733}
1734
1735static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
1736 struct btrfs_path *p,
1737 int write_lock_level)
1738{
1739 struct extent_buffer *b;
1740 int root_lock = 0;
1741 int level = 0;
1742
1743 if (p->search_commit_root) {
1744 b = root->commit_root;
1745 atomic_inc(&b->refs);
1746 level = btrfs_header_level(b);
1747 /*
1748 * Ensure that all callers have set skip_locking when
1749 * p->search_commit_root = 1.
1750 */
1751 ASSERT(p->skip_locking == 1);
1752
1753 goto out;
1754 }
1755
1756 if (p->skip_locking) {
1757 b = btrfs_root_node(root);
1758 level = btrfs_header_level(b);
1759 goto out;
1760 }
1761
1762 /* We try very hard to do read locks on the root */
1763 root_lock = BTRFS_READ_LOCK;
1764
1765 /*
1766 * If the level is set to maximum, we can skip trying to get the read
1767 * lock.
1768 */
1769 if (write_lock_level < BTRFS_MAX_LEVEL) {
1770 /*
1771 * We don't know the level of the root node until we actually
1772 * have it read locked
1773 */
1774 if (p->nowait) {
1775 b = btrfs_try_read_lock_root_node(root);
1776 if (IS_ERR(b))
1777 return b;
1778 } else {
1779 b = btrfs_read_lock_root_node(root);
1780 }
1781 level = btrfs_header_level(b);
1782 if (level > write_lock_level)
1783 goto out;
1784
1785 /* Whoops, must trade for write lock */
1786 btrfs_tree_read_unlock(b);
1787 free_extent_buffer(b);
1788 }
1789
1790 b = btrfs_lock_root_node(root);
1791 root_lock = BTRFS_WRITE_LOCK;
1792
1793 /* The level might have changed, check again */
1794 level = btrfs_header_level(b);
1795
1796out:
1797 /*
1798 * The root may have failed to write out at some point, and thus is no
1799 * longer valid, return an error in this case.
1800 */
1801 if (!extent_buffer_uptodate(b)) {
1802 if (root_lock)
1803 btrfs_tree_unlock_rw(b, root_lock);
1804 free_extent_buffer(b);
1805 return ERR_PTR(-EIO);
1806 }
1807
1808 p->nodes[level] = b;
1809 if (!p->skip_locking)
1810 p->locks[level] = root_lock;
1811 /*
1812 * Callers are responsible for dropping b's references.
1813 */
1814 return b;
1815}
1816
1817/*
1818 * Replace the extent buffer at the lowest level of the path with a cloned
1819 * version. The purpose is to be able to use it safely, after releasing the
1820 * commit root semaphore, even if relocation is happening in parallel, the
1821 * transaction used for relocation is committed and the extent buffer is
1822 * reallocated in the next transaction.
1823 *
1824 * This is used in a context where the caller does not prevent transaction
1825 * commits from happening, either by holding a transaction handle or holding
1826 * some lock, while it's doing searches through a commit root.
1827 * At the moment it's only used for send operations.
1828 */
1829static int finish_need_commit_sem_search(struct btrfs_path *path)
1830{
1831 const int i = path->lowest_level;
1832 const int slot = path->slots[i];
1833 struct extent_buffer *lowest = path->nodes[i];
1834 struct extent_buffer *clone;
1835
1836 ASSERT(path->need_commit_sem);
1837
1838 if (!lowest)
1839 return 0;
1840
1841 lockdep_assert_held_read(&lowest->fs_info->commit_root_sem);
1842
1843 clone = btrfs_clone_extent_buffer(lowest);
1844 if (!clone)
1845 return -ENOMEM;
1846
1847 btrfs_release_path(path);
1848 path->nodes[i] = clone;
1849 path->slots[i] = slot;
1850
1851 return 0;
1852}
1853
1854static inline int search_for_key_slot(struct extent_buffer *eb,
1855 int search_low_slot,
1856 const struct btrfs_key *key,
1857 int prev_cmp,
1858 int *slot)
1859{
1860 /*
1861 * If a previous call to btrfs_bin_search() on a parent node returned an
1862 * exact match (prev_cmp == 0), we can safely assume the target key will
1863 * always be at slot 0 on lower levels, since each key pointer
1864 * (struct btrfs_key_ptr) refers to the lowest key accessible from the
1865 * subtree it points to. Thus we can skip searching lower levels.
1866 */
1867 if (prev_cmp == 0) {
1868 *slot = 0;
1869 return 0;
1870 }
1871
1872 return generic_bin_search(eb, search_low_slot, key, slot);
1873}
1874
1875static int search_leaf(struct btrfs_trans_handle *trans,
1876 struct btrfs_root *root,
1877 const struct btrfs_key *key,
1878 struct btrfs_path *path,
1879 int ins_len,
1880 int prev_cmp)
1881{
1882 struct extent_buffer *leaf = path->nodes[0];
1883 int leaf_free_space = -1;
1884 int search_low_slot = 0;
1885 int ret;
1886 bool do_bin_search = true;
1887
1888 /*
1889 * If we are doing an insertion, the leaf has enough free space and the
1890 * destination slot for the key is not slot 0, then we can unlock our
1891 * write lock on the parent, and any other upper nodes, before doing the
1892 * binary search on the leaf (with search_for_key_slot()), allowing other
1893 * tasks to lock the parent and any other upper nodes.
1894 */
1895 if (ins_len > 0) {
1896 /*
1897 * Cache the leaf free space, since we will need it later and it
1898 * will not change until then.
1899 */
1900 leaf_free_space = btrfs_leaf_free_space(leaf);
1901
1902 /*
1903 * !path->locks[1] means we have a single node tree, the leaf is
1904 * the root of the tree.
1905 */
1906 if (path->locks[1] && leaf_free_space >= ins_len) {
1907 struct btrfs_disk_key first_key;
1908
1909 ASSERT(btrfs_header_nritems(leaf) > 0);
1910 btrfs_item_key(leaf, &first_key, 0);
1911
1912 /*
1913 * Doing the extra comparison with the first key is cheap,
1914 * taking into account that the first key is very likely
1915 * already in a cache line because it immediately follows
1916 * the extent buffer's header and we have recently accessed
1917 * the header's level field.
1918 */
1919 ret = comp_keys(&first_key, key);
1920 if (ret < 0) {
1921 /*
1922 * The first key is smaller than the key we want
1923 * to insert, so we are safe to unlock all upper
1924 * nodes and we have to do the binary search.
1925 *
1926 * We do use btrfs_unlock_up_safe() and not
1927 * unlock_up() because the later does not unlock
1928 * nodes with a slot of 0 - we can safely unlock
1929 * any node even if its slot is 0 since in this
1930 * case the key does not end up at slot 0 of the
1931 * leaf and there's no need to split the leaf.
1932 */
1933 btrfs_unlock_up_safe(path, 1);
1934 search_low_slot = 1;
1935 } else {
1936 /*
1937 * The first key is >= then the key we want to
1938 * insert, so we can skip the binary search as
1939 * the target key will be at slot 0.
1940 *
1941 * We can not unlock upper nodes when the key is
1942 * less than the first key, because we will need
1943 * to update the key at slot 0 of the parent node
1944 * and possibly of other upper nodes too.
1945 * If the key matches the first key, then we can
1946 * unlock all the upper nodes, using
1947 * btrfs_unlock_up_safe() instead of unlock_up()
1948 * as stated above.
1949 */
1950 if (ret == 0)
1951 btrfs_unlock_up_safe(path, 1);
1952 /*
1953 * ret is already 0 or 1, matching the result of
1954 * a btrfs_bin_search() call, so there is no need
1955 * to adjust it.
1956 */
1957 do_bin_search = false;
1958 path->slots[0] = 0;
1959 }
1960 }
1961 }
1962
1963 if (do_bin_search) {
1964 ret = search_for_key_slot(leaf, search_low_slot, key,
1965 prev_cmp, &path->slots[0]);
1966 if (ret < 0)
1967 return ret;
1968 }
1969
1970 if (ins_len > 0) {
1971 /*
1972 * Item key already exists. In this case, if we are allowed to
1973 * insert the item (for example, in dir_item case, item key
1974 * collision is allowed), it will be merged with the original
1975 * item. Only the item size grows, no new btrfs item will be
1976 * added. If search_for_extension is not set, ins_len already
1977 * accounts the size btrfs_item, deduct it here so leaf space
1978 * check will be correct.
1979 */
1980 if (ret == 0 && !path->search_for_extension) {
1981 ASSERT(ins_len >= sizeof(struct btrfs_item));
1982 ins_len -= sizeof(struct btrfs_item);
1983 }
1984
1985 ASSERT(leaf_free_space >= 0);
1986
1987 if (leaf_free_space < ins_len) {
1988 int err;
1989
1990 err = split_leaf(trans, root, key, path, ins_len,
1991 (ret == 0));
1992 ASSERT(err <= 0);
1993 if (WARN_ON(err > 0))
1994 err = -EUCLEAN;
1995 if (err)
1996 ret = err;
1997 }
1998 }
1999
2000 return ret;
2001}
2002
2003/*
2004 * btrfs_search_slot - look for a key in a tree and perform necessary
2005 * modifications to preserve tree invariants.
2006 *
2007 * @trans: Handle of transaction, used when modifying the tree
2008 * @p: Holds all btree nodes along the search path
2009 * @root: The root node of the tree
2010 * @key: The key we are looking for
2011 * @ins_len: Indicates purpose of search:
2012 * >0 for inserts it's size of item inserted (*)
2013 * <0 for deletions
2014 * 0 for plain searches, not modifying the tree
2015 *
2016 * (*) If size of item inserted doesn't include
2017 * sizeof(struct btrfs_item), then p->search_for_extension must
2018 * be set.
2019 * @cow: boolean should CoW operations be performed. Must always be 1
2020 * when modifying the tree.
2021 *
2022 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2023 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2024 *
2025 * If @key is found, 0 is returned and you can find the item in the leaf level
2026 * of the path (level 0)
2027 *
2028 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2029 * points to the slot where it should be inserted
2030 *
2031 * If an error is encountered while searching the tree a negative error number
2032 * is returned
2033 */
2034int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2035 const struct btrfs_key *key, struct btrfs_path *p,
2036 int ins_len, int cow)
2037{
2038 struct btrfs_fs_info *fs_info = root->fs_info;
2039 struct extent_buffer *b;
2040 int slot;
2041 int ret;
2042 int err;
2043 int level;
2044 int lowest_unlock = 1;
2045 /* everything at write_lock_level or lower must be write locked */
2046 int write_lock_level = 0;
2047 u8 lowest_level = 0;
2048 int min_write_lock_level;
2049 int prev_cmp;
2050
2051 might_sleep();
2052
2053 lowest_level = p->lowest_level;
2054 WARN_ON(lowest_level && ins_len > 0);
2055 WARN_ON(p->nodes[0] != NULL);
2056 BUG_ON(!cow && ins_len);
2057
2058 /*
2059 * For now only allow nowait for read only operations. There's no
2060 * strict reason why we can't, we just only need it for reads so it's
2061 * only implemented for reads.
2062 */
2063 ASSERT(!p->nowait || !cow);
2064
2065 if (ins_len < 0) {
2066 lowest_unlock = 2;
2067
2068 /* when we are removing items, we might have to go up to level
2069 * two as we update tree pointers Make sure we keep write
2070 * for those levels as well
2071 */
2072 write_lock_level = 2;
2073 } else if (ins_len > 0) {
2074 /*
2075 * for inserting items, make sure we have a write lock on
2076 * level 1 so we can update keys
2077 */
2078 write_lock_level = 1;
2079 }
2080
2081 if (!cow)
2082 write_lock_level = -1;
2083
2084 if (cow && (p->keep_locks || p->lowest_level))
2085 write_lock_level = BTRFS_MAX_LEVEL;
2086
2087 min_write_lock_level = write_lock_level;
2088
2089 if (p->need_commit_sem) {
2090 ASSERT(p->search_commit_root);
2091 if (p->nowait) {
2092 if (!down_read_trylock(&fs_info->commit_root_sem))
2093 return -EAGAIN;
2094 } else {
2095 down_read(&fs_info->commit_root_sem);
2096 }
2097 }
2098
2099again:
2100 prev_cmp = -1;
2101 b = btrfs_search_slot_get_root(root, p, write_lock_level);
2102 if (IS_ERR(b)) {
2103 ret = PTR_ERR(b);
2104 goto done;
2105 }
2106
2107 while (b) {
2108 int dec = 0;
2109
2110 level = btrfs_header_level(b);
2111
2112 if (cow) {
2113 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2114
2115 /*
2116 * if we don't really need to cow this block
2117 * then we don't want to set the path blocking,
2118 * so we test it here
2119 */
2120 if (!should_cow_block(trans, root, b))
2121 goto cow_done;
2122
2123 /*
2124 * must have write locks on this node and the
2125 * parent
2126 */
2127 if (level > write_lock_level ||
2128 (level + 1 > write_lock_level &&
2129 level + 1 < BTRFS_MAX_LEVEL &&
2130 p->nodes[level + 1])) {
2131 write_lock_level = level + 1;
2132 btrfs_release_path(p);
2133 goto again;
2134 }
2135
2136 if (last_level)
2137 err = btrfs_cow_block(trans, root, b, NULL, 0,
2138 &b,
2139 BTRFS_NESTING_COW);
2140 else
2141 err = btrfs_cow_block(trans, root, b,
2142 p->nodes[level + 1],
2143 p->slots[level + 1], &b,
2144 BTRFS_NESTING_COW);
2145 if (err) {
2146 ret = err;
2147 goto done;
2148 }
2149 }
2150cow_done:
2151 p->nodes[level] = b;
2152
2153 /*
2154 * we have a lock on b and as long as we aren't changing
2155 * the tree, there is no way to for the items in b to change.
2156 * It is safe to drop the lock on our parent before we
2157 * go through the expensive btree search on b.
2158 *
2159 * If we're inserting or deleting (ins_len != 0), then we might
2160 * be changing slot zero, which may require changing the parent.
2161 * So, we can't drop the lock until after we know which slot
2162 * we're operating on.
2163 */
2164 if (!ins_len && !p->keep_locks) {
2165 int u = level + 1;
2166
2167 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2168 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2169 p->locks[u] = 0;
2170 }
2171 }
2172
2173 if (level == 0) {
2174 if (ins_len > 0)
2175 ASSERT(write_lock_level >= 1);
2176
2177 ret = search_leaf(trans, root, key, p, ins_len, prev_cmp);
2178 if (!p->search_for_split)
2179 unlock_up(p, level, lowest_unlock,
2180 min_write_lock_level, NULL);
2181 goto done;
2182 }
2183
2184 ret = search_for_key_slot(b, 0, key, prev_cmp, &slot);
2185 if (ret < 0)
2186 goto done;
2187 prev_cmp = ret;
2188
2189 if (ret && slot > 0) {
2190 dec = 1;
2191 slot--;
2192 }
2193 p->slots[level] = slot;
2194 err = setup_nodes_for_search(trans, root, p, b, level, ins_len,
2195 &write_lock_level);
2196 if (err == -EAGAIN)
2197 goto again;
2198 if (err) {
2199 ret = err;
2200 goto done;
2201 }
2202 b = p->nodes[level];
2203 slot = p->slots[level];
2204
2205 /*
2206 * Slot 0 is special, if we change the key we have to update
2207 * the parent pointer which means we must have a write lock on
2208 * the parent
2209 */
2210 if (slot == 0 && ins_len && write_lock_level < level + 1) {
2211 write_lock_level = level + 1;
2212 btrfs_release_path(p);
2213 goto again;
2214 }
2215
2216 unlock_up(p, level, lowest_unlock, min_write_lock_level,
2217 &write_lock_level);
2218
2219 if (level == lowest_level) {
2220 if (dec)
2221 p->slots[level]++;
2222 goto done;
2223 }
2224
2225 err = read_block_for_search(root, p, &b, level, slot, key);
2226 if (err == -EAGAIN)
2227 goto again;
2228 if (err) {
2229 ret = err;
2230 goto done;
2231 }
2232
2233 if (!p->skip_locking) {
2234 level = btrfs_header_level(b);
2235
2236 btrfs_maybe_reset_lockdep_class(root, b);
2237
2238 if (level <= write_lock_level) {
2239 btrfs_tree_lock(b);
2240 p->locks[level] = BTRFS_WRITE_LOCK;
2241 } else {
2242 if (p->nowait) {
2243 if (!btrfs_try_tree_read_lock(b)) {
2244 free_extent_buffer(b);
2245 ret = -EAGAIN;
2246 goto done;
2247 }
2248 } else {
2249 btrfs_tree_read_lock(b);
2250 }
2251 p->locks[level] = BTRFS_READ_LOCK;
2252 }
2253 p->nodes[level] = b;
2254 }
2255 }
2256 ret = 1;
2257done:
2258 if (ret < 0 && !p->skip_release_on_error)
2259 btrfs_release_path(p);
2260
2261 if (p->need_commit_sem) {
2262 int ret2;
2263
2264 ret2 = finish_need_commit_sem_search(p);
2265 up_read(&fs_info->commit_root_sem);
2266 if (ret2)
2267 ret = ret2;
2268 }
2269
2270 return ret;
2271}
2272ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO);
2273
2274/*
2275 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2276 * current state of the tree together with the operations recorded in the tree
2277 * modification log to search for the key in a previous version of this tree, as
2278 * denoted by the time_seq parameter.
2279 *
2280 * Naturally, there is no support for insert, delete or cow operations.
2281 *
2282 * The resulting path and return value will be set up as if we called
2283 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2284 */
2285int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2286 struct btrfs_path *p, u64 time_seq)
2287{
2288 struct btrfs_fs_info *fs_info = root->fs_info;
2289 struct extent_buffer *b;
2290 int slot;
2291 int ret;
2292 int err;
2293 int level;
2294 int lowest_unlock = 1;
2295 u8 lowest_level = 0;
2296
2297 lowest_level = p->lowest_level;
2298 WARN_ON(p->nodes[0] != NULL);
2299 ASSERT(!p->nowait);
2300
2301 if (p->search_commit_root) {
2302 BUG_ON(time_seq);
2303 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2304 }
2305
2306again:
2307 b = btrfs_get_old_root(root, time_seq);
2308 if (!b) {
2309 ret = -EIO;
2310 goto done;
2311 }
2312 level = btrfs_header_level(b);
2313 p->locks[level] = BTRFS_READ_LOCK;
2314
2315 while (b) {
2316 int dec = 0;
2317
2318 level = btrfs_header_level(b);
2319 p->nodes[level] = b;
2320
2321 /*
2322 * we have a lock on b and as long as we aren't changing
2323 * the tree, there is no way to for the items in b to change.
2324 * It is safe to drop the lock on our parent before we
2325 * go through the expensive btree search on b.
2326 */
2327 btrfs_unlock_up_safe(p, level + 1);
2328
2329 ret = btrfs_bin_search(b, key, &slot);
2330 if (ret < 0)
2331 goto done;
2332
2333 if (level == 0) {
2334 p->slots[level] = slot;
2335 unlock_up(p, level, lowest_unlock, 0, NULL);
2336 goto done;
2337 }
2338
2339 if (ret && slot > 0) {
2340 dec = 1;
2341 slot--;
2342 }
2343 p->slots[level] = slot;
2344 unlock_up(p, level, lowest_unlock, 0, NULL);
2345
2346 if (level == lowest_level) {
2347 if (dec)
2348 p->slots[level]++;
2349 goto done;
2350 }
2351
2352 err = read_block_for_search(root, p, &b, level, slot, key);
2353 if (err == -EAGAIN)
2354 goto again;
2355 if (err) {
2356 ret = err;
2357 goto done;
2358 }
2359
2360 level = btrfs_header_level(b);
2361 btrfs_tree_read_lock(b);
2362 b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq);
2363 if (!b) {
2364 ret = -ENOMEM;
2365 goto done;
2366 }
2367 p->locks[level] = BTRFS_READ_LOCK;
2368 p->nodes[level] = b;
2369 }
2370 ret = 1;
2371done:
2372 if (ret < 0)
2373 btrfs_release_path(p);
2374
2375 return ret;
2376}
2377
2378/*
2379 * helper to use instead of search slot if no exact match is needed but
2380 * instead the next or previous item should be returned.
2381 * When find_higher is true, the next higher item is returned, the next lower
2382 * otherwise.
2383 * When return_any and find_higher are both true, and no higher item is found,
2384 * return the next lower instead.
2385 * When return_any is true and find_higher is false, and no lower item is found,
2386 * return the next higher instead.
2387 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2388 * < 0 on error
2389 */
2390int btrfs_search_slot_for_read(struct btrfs_root *root,
2391 const struct btrfs_key *key,
2392 struct btrfs_path *p, int find_higher,
2393 int return_any)
2394{
2395 int ret;
2396 struct extent_buffer *leaf;
2397
2398again:
2399 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
2400 if (ret <= 0)
2401 return ret;
2402 /*
2403 * a return value of 1 means the path is at the position where the
2404 * item should be inserted. Normally this is the next bigger item,
2405 * but in case the previous item is the last in a leaf, path points
2406 * to the first free slot in the previous leaf, i.e. at an invalid
2407 * item.
2408 */
2409 leaf = p->nodes[0];
2410
2411 if (find_higher) {
2412 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
2413 ret = btrfs_next_leaf(root, p);
2414 if (ret <= 0)
2415 return ret;
2416 if (!return_any)
2417 return 1;
2418 /*
2419 * no higher item found, return the next
2420 * lower instead
2421 */
2422 return_any = 0;
2423 find_higher = 0;
2424 btrfs_release_path(p);
2425 goto again;
2426 }
2427 } else {
2428 if (p->slots[0] == 0) {
2429 ret = btrfs_prev_leaf(root, p);
2430 if (ret < 0)
2431 return ret;
2432 if (!ret) {
2433 leaf = p->nodes[0];
2434 if (p->slots[0] == btrfs_header_nritems(leaf))
2435 p->slots[0]--;
2436 return 0;
2437 }
2438 if (!return_any)
2439 return 1;
2440 /*
2441 * no lower item found, return the next
2442 * higher instead
2443 */
2444 return_any = 0;
2445 find_higher = 1;
2446 btrfs_release_path(p);
2447 goto again;
2448 } else {
2449 --p->slots[0];
2450 }
2451 }
2452 return 0;
2453}
2454
2455/*
2456 * Execute search and call btrfs_previous_item to traverse backwards if the item
2457 * was not found.
2458 *
2459 * Return 0 if found, 1 if not found and < 0 if error.
2460 */
2461int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key,
2462 struct btrfs_path *path)
2463{
2464 int ret;
2465
2466 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2467 if (ret > 0)
2468 ret = btrfs_previous_item(root, path, key->objectid, key->type);
2469
2470 if (ret == 0)
2471 btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]);
2472
2473 return ret;
2474}
2475
2476/*
2477 * Search for a valid slot for the given path.
2478 *
2479 * @root: The root node of the tree.
2480 * @key: Will contain a valid item if found.
2481 * @path: The starting point to validate the slot.
2482 *
2483 * Return: 0 if the item is valid
2484 * 1 if not found
2485 * <0 if error.
2486 */
2487int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key,
2488 struct btrfs_path *path)
2489{
2490 while (1) {
2491 int ret;
2492 const int slot = path->slots[0];
2493 const struct extent_buffer *leaf = path->nodes[0];
2494
2495 /* This is where we start walking the path. */
2496 if (slot >= btrfs_header_nritems(leaf)) {
2497 /*
2498 * If we've reached the last slot in this leaf we need
2499 * to go to the next leaf and reset the path.
2500 */
2501 ret = btrfs_next_leaf(root, path);
2502 if (ret)
2503 return ret;
2504 continue;
2505 }
2506 /* Store the found, valid item in @key. */
2507 btrfs_item_key_to_cpu(leaf, key, slot);
2508 break;
2509 }
2510 return 0;
2511}
2512
2513/*
2514 * adjust the pointers going up the tree, starting at level
2515 * making sure the right key of each node is points to 'key'.
2516 * This is used after shifting pointers to the left, so it stops
2517 * fixing up pointers when a given leaf/node is not in slot 0 of the
2518 * higher levels
2519 *
2520 */
2521static void fixup_low_keys(struct btrfs_path *path,
2522 struct btrfs_disk_key *key, int level)
2523{
2524 int i;
2525 struct extent_buffer *t;
2526 int ret;
2527
2528 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2529 int tslot = path->slots[i];
2530
2531 if (!path->nodes[i])
2532 break;
2533 t = path->nodes[i];
2534 ret = btrfs_tree_mod_log_insert_key(t, tslot,
2535 BTRFS_MOD_LOG_KEY_REPLACE);
2536 BUG_ON(ret < 0);
2537 btrfs_set_node_key(t, key, tslot);
2538 btrfs_mark_buffer_dirty(path->nodes[i]);
2539 if (tslot != 0)
2540 break;
2541 }
2542}
2543
2544/*
2545 * update item key.
2546 *
2547 * This function isn't completely safe. It's the caller's responsibility
2548 * that the new key won't break the order
2549 */
2550void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
2551 struct btrfs_path *path,
2552 const struct btrfs_key *new_key)
2553{
2554 struct btrfs_disk_key disk_key;
2555 struct extent_buffer *eb;
2556 int slot;
2557
2558 eb = path->nodes[0];
2559 slot = path->slots[0];
2560 if (slot > 0) {
2561 btrfs_item_key(eb, &disk_key, slot - 1);
2562 if (unlikely(comp_keys(&disk_key, new_key) >= 0)) {
2563 btrfs_crit(fs_info,
2564 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
2565 slot, btrfs_disk_key_objectid(&disk_key),
2566 btrfs_disk_key_type(&disk_key),
2567 btrfs_disk_key_offset(&disk_key),
2568 new_key->objectid, new_key->type,
2569 new_key->offset);
2570 btrfs_print_leaf(eb);
2571 BUG();
2572 }
2573 }
2574 if (slot < btrfs_header_nritems(eb) - 1) {
2575 btrfs_item_key(eb, &disk_key, slot + 1);
2576 if (unlikely(comp_keys(&disk_key, new_key) <= 0)) {
2577 btrfs_crit(fs_info,
2578 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
2579 slot, btrfs_disk_key_objectid(&disk_key),
2580 btrfs_disk_key_type(&disk_key),
2581 btrfs_disk_key_offset(&disk_key),
2582 new_key->objectid, new_key->type,
2583 new_key->offset);
2584 btrfs_print_leaf(eb);
2585 BUG();
2586 }
2587 }
2588
2589 btrfs_cpu_key_to_disk(&disk_key, new_key);
2590 btrfs_set_item_key(eb, &disk_key, slot);
2591 btrfs_mark_buffer_dirty(eb);
2592 if (slot == 0)
2593 fixup_low_keys(path, &disk_key, 1);
2594}
2595
2596/*
2597 * Check key order of two sibling extent buffers.
2598 *
2599 * Return true if something is wrong.
2600 * Return false if everything is fine.
2601 *
2602 * Tree-checker only works inside one tree block, thus the following
2603 * corruption can not be detected by tree-checker:
2604 *
2605 * Leaf @left | Leaf @right
2606 * --------------------------------------------------------------
2607 * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 |
2608 *
2609 * Key f6 in leaf @left itself is valid, but not valid when the next
2610 * key in leaf @right is 7.
2611 * This can only be checked at tree block merge time.
2612 * And since tree checker has ensured all key order in each tree block
2613 * is correct, we only need to bother the last key of @left and the first
2614 * key of @right.
2615 */
2616static bool check_sibling_keys(struct extent_buffer *left,
2617 struct extent_buffer *right)
2618{
2619 struct btrfs_key left_last;
2620 struct btrfs_key right_first;
2621 int level = btrfs_header_level(left);
2622 int nr_left = btrfs_header_nritems(left);
2623 int nr_right = btrfs_header_nritems(right);
2624
2625 /* No key to check in one of the tree blocks */
2626 if (!nr_left || !nr_right)
2627 return false;
2628
2629 if (level) {
2630 btrfs_node_key_to_cpu(left, &left_last, nr_left - 1);
2631 btrfs_node_key_to_cpu(right, &right_first, 0);
2632 } else {
2633 btrfs_item_key_to_cpu(left, &left_last, nr_left - 1);
2634 btrfs_item_key_to_cpu(right, &right_first, 0);
2635 }
2636
2637 if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) {
2638 btrfs_crit(left->fs_info,
2639"bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)",
2640 left_last.objectid, left_last.type,
2641 left_last.offset, right_first.objectid,
2642 right_first.type, right_first.offset);
2643 return true;
2644 }
2645 return false;
2646}
2647
2648/*
2649 * try to push data from one node into the next node left in the
2650 * tree.
2651 *
2652 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2653 * error, and > 0 if there was no room in the left hand block.
2654 */
2655static int push_node_left(struct btrfs_trans_handle *trans,
2656 struct extent_buffer *dst,
2657 struct extent_buffer *src, int empty)
2658{
2659 struct btrfs_fs_info *fs_info = trans->fs_info;
2660 int push_items = 0;
2661 int src_nritems;
2662 int dst_nritems;
2663 int ret = 0;
2664
2665 src_nritems = btrfs_header_nritems(src);
2666 dst_nritems = btrfs_header_nritems(dst);
2667 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
2668 WARN_ON(btrfs_header_generation(src) != trans->transid);
2669 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2670
2671 if (!empty && src_nritems <= 8)
2672 return 1;
2673
2674 if (push_items <= 0)
2675 return 1;
2676
2677 if (empty) {
2678 push_items = min(src_nritems, push_items);
2679 if (push_items < src_nritems) {
2680 /* leave at least 8 pointers in the node if
2681 * we aren't going to empty it
2682 */
2683 if (src_nritems - push_items < 8) {
2684 if (push_items <= 8)
2685 return 1;
2686 push_items -= 8;
2687 }
2688 }
2689 } else
2690 push_items = min(src_nritems - 8, push_items);
2691
2692 /* dst is the left eb, src is the middle eb */
2693 if (check_sibling_keys(dst, src)) {
2694 ret = -EUCLEAN;
2695 btrfs_abort_transaction(trans, ret);
2696 return ret;
2697 }
2698 ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
2699 if (ret) {
2700 btrfs_abort_transaction(trans, ret);
2701 return ret;
2702 }
2703 copy_extent_buffer(dst, src,
2704 btrfs_node_key_ptr_offset(dst, dst_nritems),
2705 btrfs_node_key_ptr_offset(src, 0),
2706 push_items * sizeof(struct btrfs_key_ptr));
2707
2708 if (push_items < src_nritems) {
2709 /*
2710 * Don't call btrfs_tree_mod_log_insert_move() here, key removal
2711 * was already fully logged by btrfs_tree_mod_log_eb_copy() above.
2712 */
2713 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(src, 0),
2714 btrfs_node_key_ptr_offset(src, push_items),
2715 (src_nritems - push_items) *
2716 sizeof(struct btrfs_key_ptr));
2717 }
2718 btrfs_set_header_nritems(src, src_nritems - push_items);
2719 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2720 btrfs_mark_buffer_dirty(src);
2721 btrfs_mark_buffer_dirty(dst);
2722
2723 return ret;
2724}
2725
2726/*
2727 * try to push data from one node into the next node right in the
2728 * tree.
2729 *
2730 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
2731 * error, and > 0 if there was no room in the right hand block.
2732 *
2733 * this will only push up to 1/2 the contents of the left node over
2734 */
2735static int balance_node_right(struct btrfs_trans_handle *trans,
2736 struct extent_buffer *dst,
2737 struct extent_buffer *src)
2738{
2739 struct btrfs_fs_info *fs_info = trans->fs_info;
2740 int push_items = 0;
2741 int max_push;
2742 int src_nritems;
2743 int dst_nritems;
2744 int ret = 0;
2745
2746 WARN_ON(btrfs_header_generation(src) != trans->transid);
2747 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2748
2749 src_nritems = btrfs_header_nritems(src);
2750 dst_nritems = btrfs_header_nritems(dst);
2751 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
2752 if (push_items <= 0)
2753 return 1;
2754
2755 if (src_nritems < 4)
2756 return 1;
2757
2758 max_push = src_nritems / 2 + 1;
2759 /* don't try to empty the node */
2760 if (max_push >= src_nritems)
2761 return 1;
2762
2763 if (max_push < push_items)
2764 push_items = max_push;
2765
2766 /* dst is the right eb, src is the middle eb */
2767 if (check_sibling_keys(src, dst)) {
2768 ret = -EUCLEAN;
2769 btrfs_abort_transaction(trans, ret);
2770 return ret;
2771 }
2772 ret = btrfs_tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
2773 BUG_ON(ret < 0);
2774 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(dst, push_items),
2775 btrfs_node_key_ptr_offset(dst, 0),
2776 (dst_nritems) *
2777 sizeof(struct btrfs_key_ptr));
2778
2779 ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
2780 push_items);
2781 if (ret) {
2782 btrfs_abort_transaction(trans, ret);
2783 return ret;
2784 }
2785 copy_extent_buffer(dst, src,
2786 btrfs_node_key_ptr_offset(dst, 0),
2787 btrfs_node_key_ptr_offset(src, src_nritems - push_items),
2788 push_items * sizeof(struct btrfs_key_ptr));
2789
2790 btrfs_set_header_nritems(src, src_nritems - push_items);
2791 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2792
2793 btrfs_mark_buffer_dirty(src);
2794 btrfs_mark_buffer_dirty(dst);
2795
2796 return ret;
2797}
2798
2799/*
2800 * helper function to insert a new root level in the tree.
2801 * A new node is allocated, and a single item is inserted to
2802 * point to the existing root
2803 *
2804 * returns zero on success or < 0 on failure.
2805 */
2806static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2807 struct btrfs_root *root,
2808 struct btrfs_path *path, int level)
2809{
2810 struct btrfs_fs_info *fs_info = root->fs_info;
2811 u64 lower_gen;
2812 struct extent_buffer *lower;
2813 struct extent_buffer *c;
2814 struct extent_buffer *old;
2815 struct btrfs_disk_key lower_key;
2816 int ret;
2817
2818 BUG_ON(path->nodes[level]);
2819 BUG_ON(path->nodes[level-1] != root->node);
2820
2821 lower = path->nodes[level-1];
2822 if (level == 1)
2823 btrfs_item_key(lower, &lower_key, 0);
2824 else
2825 btrfs_node_key(lower, &lower_key, 0);
2826
2827 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
2828 &lower_key, level, root->node->start, 0,
2829 BTRFS_NESTING_NEW_ROOT);
2830 if (IS_ERR(c))
2831 return PTR_ERR(c);
2832
2833 root_add_used(root, fs_info->nodesize);
2834
2835 btrfs_set_header_nritems(c, 1);
2836 btrfs_set_node_key(c, &lower_key, 0);
2837 btrfs_set_node_blockptr(c, 0, lower->start);
2838 lower_gen = btrfs_header_generation(lower);
2839 WARN_ON(lower_gen != trans->transid);
2840
2841 btrfs_set_node_ptr_generation(c, 0, lower_gen);
2842
2843 btrfs_mark_buffer_dirty(c);
2844
2845 old = root->node;
2846 ret = btrfs_tree_mod_log_insert_root(root->node, c, false);
2847 BUG_ON(ret < 0);
2848 rcu_assign_pointer(root->node, c);
2849
2850 /* the super has an extra ref to root->node */
2851 free_extent_buffer(old);
2852
2853 add_root_to_dirty_list(root);
2854 atomic_inc(&c->refs);
2855 path->nodes[level] = c;
2856 path->locks[level] = BTRFS_WRITE_LOCK;
2857 path->slots[level] = 0;
2858 return 0;
2859}
2860
2861/*
2862 * worker function to insert a single pointer in a node.
2863 * the node should have enough room for the pointer already
2864 *
2865 * slot and level indicate where you want the key to go, and
2866 * blocknr is the block the key points to.
2867 */
2868static void insert_ptr(struct btrfs_trans_handle *trans,
2869 struct btrfs_path *path,
2870 struct btrfs_disk_key *key, u64 bytenr,
2871 int slot, int level)
2872{
2873 struct extent_buffer *lower;
2874 int nritems;
2875 int ret;
2876
2877 BUG_ON(!path->nodes[level]);
2878 btrfs_assert_tree_write_locked(path->nodes[level]);
2879 lower = path->nodes[level];
2880 nritems = btrfs_header_nritems(lower);
2881 BUG_ON(slot > nritems);
2882 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
2883 if (slot != nritems) {
2884 if (level) {
2885 ret = btrfs_tree_mod_log_insert_move(lower, slot + 1,
2886 slot, nritems - slot);
2887 BUG_ON(ret < 0);
2888 }
2889 memmove_extent_buffer(lower,
2890 btrfs_node_key_ptr_offset(lower, slot + 1),
2891 btrfs_node_key_ptr_offset(lower, slot),
2892 (nritems - slot) * sizeof(struct btrfs_key_ptr));
2893 }
2894 if (level) {
2895 ret = btrfs_tree_mod_log_insert_key(lower, slot,
2896 BTRFS_MOD_LOG_KEY_ADD);
2897 BUG_ON(ret < 0);
2898 }
2899 btrfs_set_node_key(lower, key, slot);
2900 btrfs_set_node_blockptr(lower, slot, bytenr);
2901 WARN_ON(trans->transid == 0);
2902 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2903 btrfs_set_header_nritems(lower, nritems + 1);
2904 btrfs_mark_buffer_dirty(lower);
2905}
2906
2907/*
2908 * split the node at the specified level in path in two.
2909 * The path is corrected to point to the appropriate node after the split
2910 *
2911 * Before splitting this tries to make some room in the node by pushing
2912 * left and right, if either one works, it returns right away.
2913 *
2914 * returns 0 on success and < 0 on failure
2915 */
2916static noinline int split_node(struct btrfs_trans_handle *trans,
2917 struct btrfs_root *root,
2918 struct btrfs_path *path, int level)
2919{
2920 struct btrfs_fs_info *fs_info = root->fs_info;
2921 struct extent_buffer *c;
2922 struct extent_buffer *split;
2923 struct btrfs_disk_key disk_key;
2924 int mid;
2925 int ret;
2926 u32 c_nritems;
2927
2928 c = path->nodes[level];
2929 WARN_ON(btrfs_header_generation(c) != trans->transid);
2930 if (c == root->node) {
2931 /*
2932 * trying to split the root, lets make a new one
2933 *
2934 * tree mod log: We don't log_removal old root in
2935 * insert_new_root, because that root buffer will be kept as a
2936 * normal node. We are going to log removal of half of the
2937 * elements below with btrfs_tree_mod_log_eb_copy(). We're
2938 * holding a tree lock on the buffer, which is why we cannot
2939 * race with other tree_mod_log users.
2940 */
2941 ret = insert_new_root(trans, root, path, level + 1);
2942 if (ret)
2943 return ret;
2944 } else {
2945 ret = push_nodes_for_insert(trans, root, path, level);
2946 c = path->nodes[level];
2947 if (!ret && btrfs_header_nritems(c) <
2948 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
2949 return 0;
2950 if (ret < 0)
2951 return ret;
2952 }
2953
2954 c_nritems = btrfs_header_nritems(c);
2955 mid = (c_nritems + 1) / 2;
2956 btrfs_node_key(c, &disk_key, mid);
2957
2958 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
2959 &disk_key, level, c->start, 0,
2960 BTRFS_NESTING_SPLIT);
2961 if (IS_ERR(split))
2962 return PTR_ERR(split);
2963
2964 root_add_used(root, fs_info->nodesize);
2965 ASSERT(btrfs_header_level(c) == level);
2966
2967 ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
2968 if (ret) {
2969 btrfs_abort_transaction(trans, ret);
2970 return ret;
2971 }
2972 copy_extent_buffer(split, c,
2973 btrfs_node_key_ptr_offset(split, 0),
2974 btrfs_node_key_ptr_offset(c, mid),
2975 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2976 btrfs_set_header_nritems(split, c_nritems - mid);
2977 btrfs_set_header_nritems(c, mid);
2978
2979 btrfs_mark_buffer_dirty(c);
2980 btrfs_mark_buffer_dirty(split);
2981
2982 insert_ptr(trans, path, &disk_key, split->start,
2983 path->slots[level + 1] + 1, level + 1);
2984
2985 if (path->slots[level] >= mid) {
2986 path->slots[level] -= mid;
2987 btrfs_tree_unlock(c);
2988 free_extent_buffer(c);
2989 path->nodes[level] = split;
2990 path->slots[level + 1] += 1;
2991 } else {
2992 btrfs_tree_unlock(split);
2993 free_extent_buffer(split);
2994 }
2995 return 0;
2996}
2997
2998/*
2999 * how many bytes are required to store the items in a leaf. start
3000 * and nr indicate which items in the leaf to check. This totals up the
3001 * space used both by the item structs and the item data
3002 */
3003static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3004{
3005 int data_len;
3006 int nritems = btrfs_header_nritems(l);
3007 int end = min(nritems, start + nr) - 1;
3008
3009 if (!nr)
3010 return 0;
3011 data_len = btrfs_item_offset(l, start) + btrfs_item_size(l, start);
3012 data_len = data_len - btrfs_item_offset(l, end);
3013 data_len += sizeof(struct btrfs_item) * nr;
3014 WARN_ON(data_len < 0);
3015 return data_len;
3016}
3017
3018/*
3019 * The space between the end of the leaf items and
3020 * the start of the leaf data. IOW, how much room
3021 * the leaf has left for both items and data
3022 */
3023noinline int btrfs_leaf_free_space(struct extent_buffer *leaf)
3024{
3025 struct btrfs_fs_info *fs_info = leaf->fs_info;
3026 int nritems = btrfs_header_nritems(leaf);
3027 int ret;
3028
3029 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3030 if (ret < 0) {
3031 btrfs_crit(fs_info,
3032 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3033 ret,
3034 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3035 leaf_space_used(leaf, 0, nritems), nritems);
3036 }
3037 return ret;
3038}
3039
3040/*
3041 * min slot controls the lowest index we're willing to push to the
3042 * right. We'll push up to and including min_slot, but no lower
3043 */
3044static noinline int __push_leaf_right(struct btrfs_path *path,
3045 int data_size, int empty,
3046 struct extent_buffer *right,
3047 int free_space, u32 left_nritems,
3048 u32 min_slot)
3049{
3050 struct btrfs_fs_info *fs_info = right->fs_info;
3051 struct extent_buffer *left = path->nodes[0];
3052 struct extent_buffer *upper = path->nodes[1];
3053 struct btrfs_map_token token;
3054 struct btrfs_disk_key disk_key;
3055 int slot;
3056 u32 i;
3057 int push_space = 0;
3058 int push_items = 0;
3059 u32 nr;
3060 u32 right_nritems;
3061 u32 data_end;
3062 u32 this_item_size;
3063
3064 if (empty)
3065 nr = 0;
3066 else
3067 nr = max_t(u32, 1, min_slot);
3068
3069 if (path->slots[0] >= left_nritems)
3070 push_space += data_size;
3071
3072 slot = path->slots[1];
3073 i = left_nritems - 1;
3074 while (i >= nr) {
3075 if (!empty && push_items > 0) {
3076 if (path->slots[0] > i)
3077 break;
3078 if (path->slots[0] == i) {
3079 int space = btrfs_leaf_free_space(left);
3080
3081 if (space + push_space * 2 > free_space)
3082 break;
3083 }
3084 }
3085
3086 if (path->slots[0] == i)
3087 push_space += data_size;
3088
3089 this_item_size = btrfs_item_size(left, i);
3090 if (this_item_size + sizeof(struct btrfs_item) +
3091 push_space > free_space)
3092 break;
3093
3094 push_items++;
3095 push_space += this_item_size + sizeof(struct btrfs_item);
3096 if (i == 0)
3097 break;
3098 i--;
3099 }
3100
3101 if (push_items == 0)
3102 goto out_unlock;
3103
3104 WARN_ON(!empty && push_items == left_nritems);
3105
3106 /* push left to right */
3107 right_nritems = btrfs_header_nritems(right);
3108
3109 push_space = btrfs_item_data_end(left, left_nritems - push_items);
3110 push_space -= leaf_data_end(left);
3111
3112 /* make room in the right data area */
3113 data_end = leaf_data_end(right);
3114 memmove_leaf_data(right, data_end - push_space, data_end,
3115 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3116
3117 /* copy from the left data area */
3118 copy_leaf_data(right, left, BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3119 leaf_data_end(left), push_space);
3120
3121 memmove_leaf_items(right, push_items, 0, right_nritems);
3122
3123 /* copy the items from left to right */
3124 copy_leaf_items(right, left, 0, left_nritems - push_items, push_items);
3125
3126 /* update the item pointers */
3127 btrfs_init_map_token(&token, right);
3128 right_nritems += push_items;
3129 btrfs_set_header_nritems(right, right_nritems);
3130 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3131 for (i = 0; i < right_nritems; i++) {
3132 push_space -= btrfs_token_item_size(&token, i);
3133 btrfs_set_token_item_offset(&token, i, push_space);
3134 }
3135
3136 left_nritems -= push_items;
3137 btrfs_set_header_nritems(left, left_nritems);
3138
3139 if (left_nritems)
3140 btrfs_mark_buffer_dirty(left);
3141 else
3142 btrfs_clean_tree_block(left);
3143
3144 btrfs_mark_buffer_dirty(right);
3145
3146 btrfs_item_key(right, &disk_key, 0);
3147 btrfs_set_node_key(upper, &disk_key, slot + 1);
3148 btrfs_mark_buffer_dirty(upper);
3149
3150 /* then fixup the leaf pointer in the path */
3151 if (path->slots[0] >= left_nritems) {
3152 path->slots[0] -= left_nritems;
3153 if (btrfs_header_nritems(path->nodes[0]) == 0)
3154 btrfs_clean_tree_block(path->nodes[0]);
3155 btrfs_tree_unlock(path->nodes[0]);
3156 free_extent_buffer(path->nodes[0]);
3157 path->nodes[0] = right;
3158 path->slots[1] += 1;
3159 } else {
3160 btrfs_tree_unlock(right);
3161 free_extent_buffer(right);
3162 }
3163 return 0;
3164
3165out_unlock:
3166 btrfs_tree_unlock(right);
3167 free_extent_buffer(right);
3168 return 1;
3169}
3170
3171/*
3172 * push some data in the path leaf to the right, trying to free up at
3173 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3174 *
3175 * returns 1 if the push failed because the other node didn't have enough
3176 * room, 0 if everything worked out and < 0 if there were major errors.
3177 *
3178 * this will push starting from min_slot to the end of the leaf. It won't
3179 * push any slot lower than min_slot
3180 */
3181static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3182 *root, struct btrfs_path *path,
3183 int min_data_size, int data_size,
3184 int empty, u32 min_slot)
3185{
3186 struct extent_buffer *left = path->nodes[0];
3187 struct extent_buffer *right;
3188 struct extent_buffer *upper;
3189 int slot;
3190 int free_space;
3191 u32 left_nritems;
3192 int ret;
3193
3194 if (!path->nodes[1])
3195 return 1;
3196
3197 slot = path->slots[1];
3198 upper = path->nodes[1];
3199 if (slot >= btrfs_header_nritems(upper) - 1)
3200 return 1;
3201
3202 btrfs_assert_tree_write_locked(path->nodes[1]);
3203
3204 right = btrfs_read_node_slot(upper, slot + 1);
3205 /*
3206 * slot + 1 is not valid or we fail to read the right node,
3207 * no big deal, just return.
3208 */
3209 if (IS_ERR(right))
3210 return 1;
3211
3212 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
3213
3214 free_space = btrfs_leaf_free_space(right);
3215 if (free_space < data_size)
3216 goto out_unlock;
3217
3218 ret = btrfs_cow_block(trans, root, right, upper,
3219 slot + 1, &right, BTRFS_NESTING_RIGHT_COW);
3220 if (ret)
3221 goto out_unlock;
3222
3223 left_nritems = btrfs_header_nritems(left);
3224 if (left_nritems == 0)
3225 goto out_unlock;
3226
3227 if (check_sibling_keys(left, right)) {
3228 ret = -EUCLEAN;
3229 btrfs_tree_unlock(right);
3230 free_extent_buffer(right);
3231 return ret;
3232 }
3233 if (path->slots[0] == left_nritems && !empty) {
3234 /* Key greater than all keys in the leaf, right neighbor has
3235 * enough room for it and we're not emptying our leaf to delete
3236 * it, therefore use right neighbor to insert the new item and
3237 * no need to touch/dirty our left leaf. */
3238 btrfs_tree_unlock(left);
3239 free_extent_buffer(left);
3240 path->nodes[0] = right;
3241 path->slots[0] = 0;
3242 path->slots[1]++;
3243 return 0;
3244 }
3245
3246 return __push_leaf_right(path, min_data_size, empty,
3247 right, free_space, left_nritems, min_slot);
3248out_unlock:
3249 btrfs_tree_unlock(right);
3250 free_extent_buffer(right);
3251 return 1;
3252}
3253
3254/*
3255 * push some data in the path leaf to the left, trying to free up at
3256 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3257 *
3258 * max_slot can put a limit on how far into the leaf we'll push items. The
3259 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3260 * items
3261 */
3262static noinline int __push_leaf_left(struct btrfs_path *path, int data_size,
3263 int empty, struct extent_buffer *left,
3264 int free_space, u32 right_nritems,
3265 u32 max_slot)
3266{
3267 struct btrfs_fs_info *fs_info = left->fs_info;
3268 struct btrfs_disk_key disk_key;
3269 struct extent_buffer *right = path->nodes[0];
3270 int i;
3271 int push_space = 0;
3272 int push_items = 0;
3273 u32 old_left_nritems;
3274 u32 nr;
3275 int ret = 0;
3276 u32 this_item_size;
3277 u32 old_left_item_size;
3278 struct btrfs_map_token token;
3279
3280 if (empty)
3281 nr = min(right_nritems, max_slot);
3282 else
3283 nr = min(right_nritems - 1, max_slot);
3284
3285 for (i = 0; i < nr; i++) {
3286 if (!empty && push_items > 0) {
3287 if (path->slots[0] < i)
3288 break;
3289 if (path->slots[0] == i) {
3290 int space = btrfs_leaf_free_space(right);
3291
3292 if (space + push_space * 2 > free_space)
3293 break;
3294 }
3295 }
3296
3297 if (path->slots[0] == i)
3298 push_space += data_size;
3299
3300 this_item_size = btrfs_item_size(right, i);
3301 if (this_item_size + sizeof(struct btrfs_item) + push_space >
3302 free_space)
3303 break;
3304
3305 push_items++;
3306 push_space += this_item_size + sizeof(struct btrfs_item);
3307 }
3308
3309 if (push_items == 0) {
3310 ret = 1;
3311 goto out;
3312 }
3313 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3314
3315 /* push data from right to left */
3316 copy_leaf_items(left, right, btrfs_header_nritems(left), 0, push_items);
3317
3318 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3319 btrfs_item_offset(right, push_items - 1);
3320
3321 copy_leaf_data(left, right, leaf_data_end(left) - push_space,
3322 btrfs_item_offset(right, push_items - 1), push_space);
3323 old_left_nritems = btrfs_header_nritems(left);
3324 BUG_ON(old_left_nritems <= 0);
3325
3326 btrfs_init_map_token(&token, left);
3327 old_left_item_size = btrfs_item_offset(left, old_left_nritems - 1);
3328 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3329 u32 ioff;
3330
3331 ioff = btrfs_token_item_offset(&token, i);
3332 btrfs_set_token_item_offset(&token, i,
3333 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size));
3334 }
3335 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3336
3337 /* fixup right node */
3338 if (push_items > right_nritems)
3339 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3340 right_nritems);
3341
3342 if (push_items < right_nritems) {
3343 push_space = btrfs_item_offset(right, push_items - 1) -
3344 leaf_data_end(right);
3345 memmove_leaf_data(right,
3346 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3347 leaf_data_end(right), push_space);
3348
3349 memmove_leaf_items(right, 0, push_items,
3350 btrfs_header_nritems(right) - push_items);
3351 }
3352
3353 btrfs_init_map_token(&token, right);
3354 right_nritems -= push_items;
3355 btrfs_set_header_nritems(right, right_nritems);
3356 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3357 for (i = 0; i < right_nritems; i++) {
3358 push_space = push_space - btrfs_token_item_size(&token, i);
3359 btrfs_set_token_item_offset(&token, i, push_space);
3360 }
3361
3362 btrfs_mark_buffer_dirty(left);
3363 if (right_nritems)
3364 btrfs_mark_buffer_dirty(right);
3365 else
3366 btrfs_clean_tree_block(right);
3367
3368 btrfs_item_key(right, &disk_key, 0);
3369 fixup_low_keys(path, &disk_key, 1);
3370
3371 /* then fixup the leaf pointer in the path */
3372 if (path->slots[0] < push_items) {
3373 path->slots[0] += old_left_nritems;
3374 btrfs_tree_unlock(path->nodes[0]);
3375 free_extent_buffer(path->nodes[0]);
3376 path->nodes[0] = left;
3377 path->slots[1] -= 1;
3378 } else {
3379 btrfs_tree_unlock(left);
3380 free_extent_buffer(left);
3381 path->slots[0] -= push_items;
3382 }
3383 BUG_ON(path->slots[0] < 0);
3384 return ret;
3385out:
3386 btrfs_tree_unlock(left);
3387 free_extent_buffer(left);
3388 return ret;
3389}
3390
3391/*
3392 * push some data in the path leaf to the left, trying to free up at
3393 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3394 *
3395 * max_slot can put a limit on how far into the leaf we'll push items. The
3396 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3397 * items
3398 */
3399static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3400 *root, struct btrfs_path *path, int min_data_size,
3401 int data_size, int empty, u32 max_slot)
3402{
3403 struct extent_buffer *right = path->nodes[0];
3404 struct extent_buffer *left;
3405 int slot;
3406 int free_space;
3407 u32 right_nritems;
3408 int ret = 0;
3409
3410 slot = path->slots[1];
3411 if (slot == 0)
3412 return 1;
3413 if (!path->nodes[1])
3414 return 1;
3415
3416 right_nritems = btrfs_header_nritems(right);
3417 if (right_nritems == 0)
3418 return 1;
3419
3420 btrfs_assert_tree_write_locked(path->nodes[1]);
3421
3422 left = btrfs_read_node_slot(path->nodes[1], slot - 1);
3423 /*
3424 * slot - 1 is not valid or we fail to read the left node,
3425 * no big deal, just return.
3426 */
3427 if (IS_ERR(left))
3428 return 1;
3429
3430 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
3431
3432 free_space = btrfs_leaf_free_space(left);
3433 if (free_space < data_size) {
3434 ret = 1;
3435 goto out;
3436 }
3437
3438 ret = btrfs_cow_block(trans, root, left,
3439 path->nodes[1], slot - 1, &left,
3440 BTRFS_NESTING_LEFT_COW);
3441 if (ret) {
3442 /* we hit -ENOSPC, but it isn't fatal here */
3443 if (ret == -ENOSPC)
3444 ret = 1;
3445 goto out;
3446 }
3447
3448 if (check_sibling_keys(left, right)) {
3449 ret = -EUCLEAN;
3450 goto out;
3451 }
3452 return __push_leaf_left(path, min_data_size,
3453 empty, left, free_space, right_nritems,
3454 max_slot);
3455out:
3456 btrfs_tree_unlock(left);
3457 free_extent_buffer(left);
3458 return ret;
3459}
3460
3461/*
3462 * split the path's leaf in two, making sure there is at least data_size
3463 * available for the resulting leaf level of the path.
3464 */
3465static noinline void copy_for_split(struct btrfs_trans_handle *trans,
3466 struct btrfs_path *path,
3467 struct extent_buffer *l,
3468 struct extent_buffer *right,
3469 int slot, int mid, int nritems)
3470{
3471 struct btrfs_fs_info *fs_info = trans->fs_info;
3472 int data_copy_size;
3473 int rt_data_off;
3474 int i;
3475 struct btrfs_disk_key disk_key;
3476 struct btrfs_map_token token;
3477
3478 nritems = nritems - mid;
3479 btrfs_set_header_nritems(right, nritems);
3480 data_copy_size = btrfs_item_data_end(l, mid) - leaf_data_end(l);
3481
3482 copy_leaf_items(right, l, 0, mid, nritems);
3483
3484 copy_leaf_data(right, l, BTRFS_LEAF_DATA_SIZE(fs_info) - data_copy_size,
3485 leaf_data_end(l), data_copy_size);
3486
3487 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_data_end(l, mid);
3488
3489 btrfs_init_map_token(&token, right);
3490 for (i = 0; i < nritems; i++) {
3491 u32 ioff;
3492
3493 ioff = btrfs_token_item_offset(&token, i);
3494 btrfs_set_token_item_offset(&token, i, ioff + rt_data_off);
3495 }
3496
3497 btrfs_set_header_nritems(l, mid);
3498 btrfs_item_key(right, &disk_key, 0);
3499 insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
3500
3501 btrfs_mark_buffer_dirty(right);
3502 btrfs_mark_buffer_dirty(l);
3503 BUG_ON(path->slots[0] != slot);
3504
3505 if (mid <= slot) {
3506 btrfs_tree_unlock(path->nodes[0]);
3507 free_extent_buffer(path->nodes[0]);
3508 path->nodes[0] = right;
3509 path->slots[0] -= mid;
3510 path->slots[1] += 1;
3511 } else {
3512 btrfs_tree_unlock(right);
3513 free_extent_buffer(right);
3514 }
3515
3516 BUG_ON(path->slots[0] < 0);
3517}
3518
3519/*
3520 * double splits happen when we need to insert a big item in the middle
3521 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3522 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3523 * A B C
3524 *
3525 * We avoid this by trying to push the items on either side of our target
3526 * into the adjacent leaves. If all goes well we can avoid the double split
3527 * completely.
3528 */
3529static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
3530 struct btrfs_root *root,
3531 struct btrfs_path *path,
3532 int data_size)
3533{
3534 int ret;
3535 int progress = 0;
3536 int slot;
3537 u32 nritems;
3538 int space_needed = data_size;
3539
3540 slot = path->slots[0];
3541 if (slot < btrfs_header_nritems(path->nodes[0]))
3542 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
3543
3544 /*
3545 * try to push all the items after our slot into the
3546 * right leaf
3547 */
3548 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
3549 if (ret < 0)
3550 return ret;
3551
3552 if (ret == 0)
3553 progress++;
3554
3555 nritems = btrfs_header_nritems(path->nodes[0]);
3556 /*
3557 * our goal is to get our slot at the start or end of a leaf. If
3558 * we've done so we're done
3559 */
3560 if (path->slots[0] == 0 || path->slots[0] == nritems)
3561 return 0;
3562
3563 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
3564 return 0;
3565
3566 /* try to push all the items before our slot into the next leaf */
3567 slot = path->slots[0];
3568 space_needed = data_size;
3569 if (slot > 0)
3570 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
3571 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
3572 if (ret < 0)
3573 return ret;
3574
3575 if (ret == 0)
3576 progress++;
3577
3578 if (progress)
3579 return 0;
3580 return 1;
3581}
3582
3583/*
3584 * split the path's leaf in two, making sure there is at least data_size
3585 * available for the resulting leaf level of the path.
3586 *
3587 * returns 0 if all went well and < 0 on failure.
3588 */
3589static noinline int split_leaf(struct btrfs_trans_handle *trans,
3590 struct btrfs_root *root,
3591 const struct btrfs_key *ins_key,
3592 struct btrfs_path *path, int data_size,
3593 int extend)
3594{
3595 struct btrfs_disk_key disk_key;
3596 struct extent_buffer *l;
3597 u32 nritems;
3598 int mid;
3599 int slot;
3600 struct extent_buffer *right;
3601 struct btrfs_fs_info *fs_info = root->fs_info;
3602 int ret = 0;
3603 int wret;
3604 int split;
3605 int num_doubles = 0;
3606 int tried_avoid_double = 0;
3607
3608 l = path->nodes[0];
3609 slot = path->slots[0];
3610 if (extend && data_size + btrfs_item_size(l, slot) +
3611 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
3612 return -EOVERFLOW;
3613
3614 /* first try to make some room by pushing left and right */
3615 if (data_size && path->nodes[1]) {
3616 int space_needed = data_size;
3617
3618 if (slot < btrfs_header_nritems(l))
3619 space_needed -= btrfs_leaf_free_space(l);
3620
3621 wret = push_leaf_right(trans, root, path, space_needed,
3622 space_needed, 0, 0);
3623 if (wret < 0)
3624 return wret;
3625 if (wret) {
3626 space_needed = data_size;
3627 if (slot > 0)
3628 space_needed -= btrfs_leaf_free_space(l);
3629 wret = push_leaf_left(trans, root, path, space_needed,
3630 space_needed, 0, (u32)-1);
3631 if (wret < 0)
3632 return wret;
3633 }
3634 l = path->nodes[0];
3635
3636 /* did the pushes work? */
3637 if (btrfs_leaf_free_space(l) >= data_size)
3638 return 0;
3639 }
3640
3641 if (!path->nodes[1]) {
3642 ret = insert_new_root(trans, root, path, 1);
3643 if (ret)
3644 return ret;
3645 }
3646again:
3647 split = 1;
3648 l = path->nodes[0];
3649 slot = path->slots[0];
3650 nritems = btrfs_header_nritems(l);
3651 mid = (nritems + 1) / 2;
3652
3653 if (mid <= slot) {
3654 if (nritems == 1 ||
3655 leaf_space_used(l, mid, nritems - mid) + data_size >
3656 BTRFS_LEAF_DATA_SIZE(fs_info)) {
3657 if (slot >= nritems) {
3658 split = 0;
3659 } else {
3660 mid = slot;
3661 if (mid != nritems &&
3662 leaf_space_used(l, mid, nritems - mid) +
3663 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
3664 if (data_size && !tried_avoid_double)
3665 goto push_for_double;
3666 split = 2;
3667 }
3668 }
3669 }
3670 } else {
3671 if (leaf_space_used(l, 0, mid) + data_size >
3672 BTRFS_LEAF_DATA_SIZE(fs_info)) {
3673 if (!extend && data_size && slot == 0) {
3674 split = 0;
3675 } else if ((extend || !data_size) && slot == 0) {
3676 mid = 1;
3677 } else {
3678 mid = slot;
3679 if (mid != nritems &&
3680 leaf_space_used(l, mid, nritems - mid) +
3681 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
3682 if (data_size && !tried_avoid_double)
3683 goto push_for_double;
3684 split = 2;
3685 }
3686 }
3687 }
3688 }
3689
3690 if (split == 0)
3691 btrfs_cpu_key_to_disk(&disk_key, ins_key);
3692 else
3693 btrfs_item_key(l, &disk_key, mid);
3694
3695 /*
3696 * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double
3697 * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES
3698 * subclasses, which is 8 at the time of this patch, and we've maxed it
3699 * out. In the future we could add a
3700 * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just
3701 * use BTRFS_NESTING_NEW_ROOT.
3702 */
3703 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3704 &disk_key, 0, l->start, 0,
3705 num_doubles ? BTRFS_NESTING_NEW_ROOT :
3706 BTRFS_NESTING_SPLIT);
3707 if (IS_ERR(right))
3708 return PTR_ERR(right);
3709
3710 root_add_used(root, fs_info->nodesize);
3711
3712 if (split == 0) {
3713 if (mid <= slot) {
3714 btrfs_set_header_nritems(right, 0);
3715 insert_ptr(trans, path, &disk_key,
3716 right->start, path->slots[1] + 1, 1);
3717 btrfs_tree_unlock(path->nodes[0]);
3718 free_extent_buffer(path->nodes[0]);
3719 path->nodes[0] = right;
3720 path->slots[0] = 0;
3721 path->slots[1] += 1;
3722 } else {
3723 btrfs_set_header_nritems(right, 0);
3724 insert_ptr(trans, path, &disk_key,
3725 right->start, path->slots[1], 1);
3726 btrfs_tree_unlock(path->nodes[0]);
3727 free_extent_buffer(path->nodes[0]);
3728 path->nodes[0] = right;
3729 path->slots[0] = 0;
3730 if (path->slots[1] == 0)
3731 fixup_low_keys(path, &disk_key, 1);
3732 }
3733 /*
3734 * We create a new leaf 'right' for the required ins_len and
3735 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
3736 * the content of ins_len to 'right'.
3737 */
3738 return ret;
3739 }
3740
3741 copy_for_split(trans, path, l, right, slot, mid, nritems);
3742
3743 if (split == 2) {
3744 BUG_ON(num_doubles != 0);
3745 num_doubles++;
3746 goto again;
3747 }
3748
3749 return 0;
3750
3751push_for_double:
3752 push_for_double_split(trans, root, path, data_size);
3753 tried_avoid_double = 1;
3754 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
3755 return 0;
3756 goto again;
3757}
3758
3759static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3760 struct btrfs_root *root,
3761 struct btrfs_path *path, int ins_len)
3762{
3763 struct btrfs_key key;
3764 struct extent_buffer *leaf;
3765 struct btrfs_file_extent_item *fi;
3766 u64 extent_len = 0;
3767 u32 item_size;
3768 int ret;
3769
3770 leaf = path->nodes[0];
3771 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3772
3773 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3774 key.type != BTRFS_EXTENT_CSUM_KEY);
3775
3776 if (btrfs_leaf_free_space(leaf) >= ins_len)
3777 return 0;
3778
3779 item_size = btrfs_item_size(leaf, path->slots[0]);
3780 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3781 fi = btrfs_item_ptr(leaf, path->slots[0],
3782 struct btrfs_file_extent_item);
3783 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3784 }
3785 btrfs_release_path(path);
3786
3787 path->keep_locks = 1;
3788 path->search_for_split = 1;
3789 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3790 path->search_for_split = 0;
3791 if (ret > 0)
3792 ret = -EAGAIN;
3793 if (ret < 0)
3794 goto err;
3795
3796 ret = -EAGAIN;
3797 leaf = path->nodes[0];
3798 /* if our item isn't there, return now */
3799 if (item_size != btrfs_item_size(leaf, path->slots[0]))
3800 goto err;
3801
3802 /* the leaf has changed, it now has room. return now */
3803 if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
3804 goto err;
3805
3806 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3807 fi = btrfs_item_ptr(leaf, path->slots[0],
3808 struct btrfs_file_extent_item);
3809 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3810 goto err;
3811 }
3812
3813 ret = split_leaf(trans, root, &key, path, ins_len, 1);
3814 if (ret)
3815 goto err;
3816
3817 path->keep_locks = 0;
3818 btrfs_unlock_up_safe(path, 1);
3819 return 0;
3820err:
3821 path->keep_locks = 0;
3822 return ret;
3823}
3824
3825static noinline int split_item(struct btrfs_path *path,
3826 const struct btrfs_key *new_key,
3827 unsigned long split_offset)
3828{
3829 struct extent_buffer *leaf;
3830 int orig_slot, slot;
3831 char *buf;
3832 u32 nritems;
3833 u32 item_size;
3834 u32 orig_offset;
3835 struct btrfs_disk_key disk_key;
3836
3837 leaf = path->nodes[0];
3838 BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item));
3839
3840 orig_slot = path->slots[0];
3841 orig_offset = btrfs_item_offset(leaf, path->slots[0]);
3842 item_size = btrfs_item_size(leaf, path->slots[0]);
3843
3844 buf = kmalloc(item_size, GFP_NOFS);
3845 if (!buf)
3846 return -ENOMEM;
3847
3848 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3849 path->slots[0]), item_size);
3850
3851 slot = path->slots[0] + 1;
3852 nritems = btrfs_header_nritems(leaf);
3853 if (slot != nritems) {
3854 /* shift the items */
3855 memmove_leaf_items(leaf, slot + 1, slot, nritems - slot);
3856 }
3857
3858 btrfs_cpu_key_to_disk(&disk_key, new_key);
3859 btrfs_set_item_key(leaf, &disk_key, slot);
3860
3861 btrfs_set_item_offset(leaf, slot, orig_offset);
3862 btrfs_set_item_size(leaf, slot, item_size - split_offset);
3863
3864 btrfs_set_item_offset(leaf, orig_slot,
3865 orig_offset + item_size - split_offset);
3866 btrfs_set_item_size(leaf, orig_slot, split_offset);
3867
3868 btrfs_set_header_nritems(leaf, nritems + 1);
3869
3870 /* write the data for the start of the original item */
3871 write_extent_buffer(leaf, buf,
3872 btrfs_item_ptr_offset(leaf, path->slots[0]),
3873 split_offset);
3874
3875 /* write the data for the new item */
3876 write_extent_buffer(leaf, buf + split_offset,
3877 btrfs_item_ptr_offset(leaf, slot),
3878 item_size - split_offset);
3879 btrfs_mark_buffer_dirty(leaf);
3880
3881 BUG_ON(btrfs_leaf_free_space(leaf) < 0);
3882 kfree(buf);
3883 return 0;
3884}
3885
3886/*
3887 * This function splits a single item into two items,
3888 * giving 'new_key' to the new item and splitting the
3889 * old one at split_offset (from the start of the item).
3890 *
3891 * The path may be released by this operation. After
3892 * the split, the path is pointing to the old item. The
3893 * new item is going to be in the same node as the old one.
3894 *
3895 * Note, the item being split must be smaller enough to live alone on
3896 * a tree block with room for one extra struct btrfs_item
3897 *
3898 * This allows us to split the item in place, keeping a lock on the
3899 * leaf the entire time.
3900 */
3901int btrfs_split_item(struct btrfs_trans_handle *trans,
3902 struct btrfs_root *root,
3903 struct btrfs_path *path,
3904 const struct btrfs_key *new_key,
3905 unsigned long split_offset)
3906{
3907 int ret;
3908 ret = setup_leaf_for_split(trans, root, path,
3909 sizeof(struct btrfs_item));
3910 if (ret)
3911 return ret;
3912
3913 ret = split_item(path, new_key, split_offset);
3914 return ret;
3915}
3916
3917/*
3918 * make the item pointed to by the path smaller. new_size indicates
3919 * how small to make it, and from_end tells us if we just chop bytes
3920 * off the end of the item or if we shift the item to chop bytes off
3921 * the front.
3922 */
3923void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end)
3924{
3925 int slot;
3926 struct extent_buffer *leaf;
3927 u32 nritems;
3928 unsigned int data_end;
3929 unsigned int old_data_start;
3930 unsigned int old_size;
3931 unsigned int size_diff;
3932 int i;
3933 struct btrfs_map_token token;
3934
3935 leaf = path->nodes[0];
3936 slot = path->slots[0];
3937
3938 old_size = btrfs_item_size(leaf, slot);
3939 if (old_size == new_size)
3940 return;
3941
3942 nritems = btrfs_header_nritems(leaf);
3943 data_end = leaf_data_end(leaf);
3944
3945 old_data_start = btrfs_item_offset(leaf, slot);
3946
3947 size_diff = old_size - new_size;
3948
3949 BUG_ON(slot < 0);
3950 BUG_ON(slot >= nritems);
3951
3952 /*
3953 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3954 */
3955 /* first correct the data pointers */
3956 btrfs_init_map_token(&token, leaf);
3957 for (i = slot; i < nritems; i++) {
3958 u32 ioff;
3959
3960 ioff = btrfs_token_item_offset(&token, i);
3961 btrfs_set_token_item_offset(&token, i, ioff + size_diff);
3962 }
3963
3964 /* shift the data */
3965 if (from_end) {
3966 memmove_leaf_data(leaf, data_end + size_diff, data_end,
3967 old_data_start + new_size - data_end);
3968 } else {
3969 struct btrfs_disk_key disk_key;
3970 u64 offset;
3971
3972 btrfs_item_key(leaf, &disk_key, slot);
3973
3974 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3975 unsigned long ptr;
3976 struct btrfs_file_extent_item *fi;
3977
3978 fi = btrfs_item_ptr(leaf, slot,
3979 struct btrfs_file_extent_item);
3980 fi = (struct btrfs_file_extent_item *)(
3981 (unsigned long)fi - size_diff);
3982
3983 if (btrfs_file_extent_type(leaf, fi) ==
3984 BTRFS_FILE_EXTENT_INLINE) {
3985 ptr = btrfs_item_ptr_offset(leaf, slot);
3986 memmove_extent_buffer(leaf, ptr,
3987 (unsigned long)fi,
3988 BTRFS_FILE_EXTENT_INLINE_DATA_START);
3989 }
3990 }
3991
3992 memmove_leaf_data(leaf, data_end + size_diff, data_end,
3993 old_data_start - data_end);
3994
3995 offset = btrfs_disk_key_offset(&disk_key);
3996 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3997 btrfs_set_item_key(leaf, &disk_key, slot);
3998 if (slot == 0)
3999 fixup_low_keys(path, &disk_key, 1);
4000 }
4001
4002 btrfs_set_item_size(leaf, slot, new_size);
4003 btrfs_mark_buffer_dirty(leaf);
4004
4005 if (btrfs_leaf_free_space(leaf) < 0) {
4006 btrfs_print_leaf(leaf);
4007 BUG();
4008 }
4009}
4010
4011/*
4012 * make the item pointed to by the path bigger, data_size is the added size.
4013 */
4014void btrfs_extend_item(struct btrfs_path *path, u32 data_size)
4015{
4016 int slot;
4017 struct extent_buffer *leaf;
4018 u32 nritems;
4019 unsigned int data_end;
4020 unsigned int old_data;
4021 unsigned int old_size;
4022 int i;
4023 struct btrfs_map_token token;
4024
4025 leaf = path->nodes[0];
4026
4027 nritems = btrfs_header_nritems(leaf);
4028 data_end = leaf_data_end(leaf);
4029
4030 if (btrfs_leaf_free_space(leaf) < data_size) {
4031 btrfs_print_leaf(leaf);
4032 BUG();
4033 }
4034 slot = path->slots[0];
4035 old_data = btrfs_item_data_end(leaf, slot);
4036
4037 BUG_ON(slot < 0);
4038 if (slot >= nritems) {
4039 btrfs_print_leaf(leaf);
4040 btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d",
4041 slot, nritems);
4042 BUG();
4043 }
4044
4045 /*
4046 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4047 */
4048 /* first correct the data pointers */
4049 btrfs_init_map_token(&token, leaf);
4050 for (i = slot; i < nritems; i++) {
4051 u32 ioff;
4052
4053 ioff = btrfs_token_item_offset(&token, i);
4054 btrfs_set_token_item_offset(&token, i, ioff - data_size);
4055 }
4056
4057 /* shift the data */
4058 memmove_leaf_data(leaf, data_end - data_size, data_end,
4059 old_data - data_end);
4060
4061 data_end = old_data;
4062 old_size = btrfs_item_size(leaf, slot);
4063 btrfs_set_item_size(leaf, slot, old_size + data_size);
4064 btrfs_mark_buffer_dirty(leaf);
4065
4066 if (btrfs_leaf_free_space(leaf) < 0) {
4067 btrfs_print_leaf(leaf);
4068 BUG();
4069 }
4070}
4071
4072/*
4073 * Make space in the node before inserting one or more items.
4074 *
4075 * @root: root we are inserting items to
4076 * @path: points to the leaf/slot where we are going to insert new items
4077 * @batch: information about the batch of items to insert
4078 *
4079 * Main purpose is to save stack depth by doing the bulk of the work in a
4080 * function that doesn't call btrfs_search_slot
4081 */
4082static void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4083 const struct btrfs_item_batch *batch)
4084{
4085 struct btrfs_fs_info *fs_info = root->fs_info;
4086 int i;
4087 u32 nritems;
4088 unsigned int data_end;
4089 struct btrfs_disk_key disk_key;
4090 struct extent_buffer *leaf;
4091 int slot;
4092 struct btrfs_map_token token;
4093 u32 total_size;
4094
4095 /*
4096 * Before anything else, update keys in the parent and other ancestors
4097 * if needed, then release the write locks on them, so that other tasks
4098 * can use them while we modify the leaf.
4099 */
4100 if (path->slots[0] == 0) {
4101 btrfs_cpu_key_to_disk(&disk_key, &batch->keys[0]);
4102 fixup_low_keys(path, &disk_key, 1);
4103 }
4104 btrfs_unlock_up_safe(path, 1);
4105
4106 leaf = path->nodes[0];
4107 slot = path->slots[0];
4108
4109 nritems = btrfs_header_nritems(leaf);
4110 data_end = leaf_data_end(leaf);
4111 total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item));
4112
4113 if (btrfs_leaf_free_space(leaf) < total_size) {
4114 btrfs_print_leaf(leaf);
4115 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4116 total_size, btrfs_leaf_free_space(leaf));
4117 BUG();
4118 }
4119
4120 btrfs_init_map_token(&token, leaf);
4121 if (slot != nritems) {
4122 unsigned int old_data = btrfs_item_data_end(leaf, slot);
4123
4124 if (old_data < data_end) {
4125 btrfs_print_leaf(leaf);
4126 btrfs_crit(fs_info,
4127 "item at slot %d with data offset %u beyond data end of leaf %u",
4128 slot, old_data, data_end);
4129 BUG();
4130 }
4131 /*
4132 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4133 */
4134 /* first correct the data pointers */
4135 for (i = slot; i < nritems; i++) {
4136 u32 ioff;
4137
4138 ioff = btrfs_token_item_offset(&token, i);
4139 btrfs_set_token_item_offset(&token, i,
4140 ioff - batch->total_data_size);
4141 }
4142 /* shift the items */
4143 memmove_leaf_items(leaf, slot + batch->nr, slot, nritems - slot);
4144
4145 /* shift the data */
4146 memmove_leaf_data(leaf, data_end - batch->total_data_size,
4147 data_end, old_data - data_end);
4148 data_end = old_data;
4149 }
4150
4151 /* setup the item for the new data */
4152 for (i = 0; i < batch->nr; i++) {
4153 btrfs_cpu_key_to_disk(&disk_key, &batch->keys[i]);
4154 btrfs_set_item_key(leaf, &disk_key, slot + i);
4155 data_end -= batch->data_sizes[i];
4156 btrfs_set_token_item_offset(&token, slot + i, data_end);
4157 btrfs_set_token_item_size(&token, slot + i, batch->data_sizes[i]);
4158 }
4159
4160 btrfs_set_header_nritems(leaf, nritems + batch->nr);
4161 btrfs_mark_buffer_dirty(leaf);
4162
4163 if (btrfs_leaf_free_space(leaf) < 0) {
4164 btrfs_print_leaf(leaf);
4165 BUG();
4166 }
4167}
4168
4169/*
4170 * Insert a new item into a leaf.
4171 *
4172 * @root: The root of the btree.
4173 * @path: A path pointing to the target leaf and slot.
4174 * @key: The key of the new item.
4175 * @data_size: The size of the data associated with the new key.
4176 */
4177void btrfs_setup_item_for_insert(struct btrfs_root *root,
4178 struct btrfs_path *path,
4179 const struct btrfs_key *key,
4180 u32 data_size)
4181{
4182 struct btrfs_item_batch batch;
4183
4184 batch.keys = key;
4185 batch.data_sizes = &data_size;
4186 batch.total_data_size = data_size;
4187 batch.nr = 1;
4188
4189 setup_items_for_insert(root, path, &batch);
4190}
4191
4192/*
4193 * Given a key and some data, insert items into the tree.
4194 * This does all the path init required, making room in the tree if needed.
4195 */
4196int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4197 struct btrfs_root *root,
4198 struct btrfs_path *path,
4199 const struct btrfs_item_batch *batch)
4200{
4201 int ret = 0;
4202 int slot;
4203 u32 total_size;
4204
4205 total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item));
4206 ret = btrfs_search_slot(trans, root, &batch->keys[0], path, total_size, 1);
4207 if (ret == 0)
4208 return -EEXIST;
4209 if (ret < 0)
4210 return ret;
4211
4212 slot = path->slots[0];
4213 BUG_ON(slot < 0);
4214
4215 setup_items_for_insert(root, path, batch);
4216 return 0;
4217}
4218
4219/*
4220 * Given a key and some data, insert an item into the tree.
4221 * This does all the path init required, making room in the tree if needed.
4222 */
4223int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4224 const struct btrfs_key *cpu_key, void *data,
4225 u32 data_size)
4226{
4227 int ret = 0;
4228 struct btrfs_path *path;
4229 struct extent_buffer *leaf;
4230 unsigned long ptr;
4231
4232 path = btrfs_alloc_path();
4233 if (!path)
4234 return -ENOMEM;
4235 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4236 if (!ret) {
4237 leaf = path->nodes[0];
4238 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4239 write_extent_buffer(leaf, data, ptr, data_size);
4240 btrfs_mark_buffer_dirty(leaf);
4241 }
4242 btrfs_free_path(path);
4243 return ret;
4244}
4245
4246/*
4247 * This function duplicates an item, giving 'new_key' to the new item.
4248 * It guarantees both items live in the same tree leaf and the new item is
4249 * contiguous with the original item.
4250 *
4251 * This allows us to split a file extent in place, keeping a lock on the leaf
4252 * the entire time.
4253 */
4254int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4255 struct btrfs_root *root,
4256 struct btrfs_path *path,
4257 const struct btrfs_key *new_key)
4258{
4259 struct extent_buffer *leaf;
4260 int ret;
4261 u32 item_size;
4262
4263 leaf = path->nodes[0];
4264 item_size = btrfs_item_size(leaf, path->slots[0]);
4265 ret = setup_leaf_for_split(trans, root, path,
4266 item_size + sizeof(struct btrfs_item));
4267 if (ret)
4268 return ret;
4269
4270 path->slots[0]++;
4271 btrfs_setup_item_for_insert(root, path, new_key, item_size);
4272 leaf = path->nodes[0];
4273 memcpy_extent_buffer(leaf,
4274 btrfs_item_ptr_offset(leaf, path->slots[0]),
4275 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4276 item_size);
4277 return 0;
4278}
4279
4280/*
4281 * delete the pointer from a given node.
4282 *
4283 * the tree should have been previously balanced so the deletion does not
4284 * empty a node.
4285 */
4286static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4287 int level, int slot)
4288{
4289 struct extent_buffer *parent = path->nodes[level];
4290 u32 nritems;
4291 int ret;
4292
4293 nritems = btrfs_header_nritems(parent);
4294 if (slot != nritems - 1) {
4295 if (level) {
4296 ret = btrfs_tree_mod_log_insert_move(parent, slot,
4297 slot + 1, nritems - slot - 1);
4298 BUG_ON(ret < 0);
4299 }
4300 memmove_extent_buffer(parent,
4301 btrfs_node_key_ptr_offset(parent, slot),
4302 btrfs_node_key_ptr_offset(parent, slot + 1),
4303 sizeof(struct btrfs_key_ptr) *
4304 (nritems - slot - 1));
4305 } else if (level) {
4306 ret = btrfs_tree_mod_log_insert_key(parent, slot,
4307 BTRFS_MOD_LOG_KEY_REMOVE);
4308 BUG_ON(ret < 0);
4309 }
4310
4311 nritems--;
4312 btrfs_set_header_nritems(parent, nritems);
4313 if (nritems == 0 && parent == root->node) {
4314 BUG_ON(btrfs_header_level(root->node) != 1);
4315 /* just turn the root into a leaf and break */
4316 btrfs_set_header_level(root->node, 0);
4317 } else if (slot == 0) {
4318 struct btrfs_disk_key disk_key;
4319
4320 btrfs_node_key(parent, &disk_key, 0);
4321 fixup_low_keys(path, &disk_key, level + 1);
4322 }
4323 btrfs_mark_buffer_dirty(parent);
4324}
4325
4326/*
4327 * a helper function to delete the leaf pointed to by path->slots[1] and
4328 * path->nodes[1].
4329 *
4330 * This deletes the pointer in path->nodes[1] and frees the leaf
4331 * block extent. zero is returned if it all worked out, < 0 otherwise.
4332 *
4333 * The path must have already been setup for deleting the leaf, including
4334 * all the proper balancing. path->nodes[1] must be locked.
4335 */
4336static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4337 struct btrfs_root *root,
4338 struct btrfs_path *path,
4339 struct extent_buffer *leaf)
4340{
4341 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4342 del_ptr(root, path, 1, path->slots[1]);
4343
4344 /*
4345 * btrfs_free_extent is expensive, we want to make sure we
4346 * aren't holding any locks when we call it
4347 */
4348 btrfs_unlock_up_safe(path, 0);
4349
4350 root_sub_used(root, leaf->len);
4351
4352 atomic_inc(&leaf->refs);
4353 btrfs_free_tree_block(trans, btrfs_root_id(root), leaf, 0, 1);
4354 free_extent_buffer_stale(leaf);
4355}
4356/*
4357 * delete the item at the leaf level in path. If that empties
4358 * the leaf, remove it from the tree
4359 */
4360int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4361 struct btrfs_path *path, int slot, int nr)
4362{
4363 struct btrfs_fs_info *fs_info = root->fs_info;
4364 struct extent_buffer *leaf;
4365 int ret = 0;
4366 int wret;
4367 u32 nritems;
4368
4369 leaf = path->nodes[0];
4370 nritems = btrfs_header_nritems(leaf);
4371
4372 if (slot + nr != nritems) {
4373 const u32 last_off = btrfs_item_offset(leaf, slot + nr - 1);
4374 const int data_end = leaf_data_end(leaf);
4375 struct btrfs_map_token token;
4376 u32 dsize = 0;
4377 int i;
4378
4379 for (i = 0; i < nr; i++)
4380 dsize += btrfs_item_size(leaf, slot + i);
4381
4382 memmove_leaf_data(leaf, data_end + dsize, data_end,
4383 last_off - data_end);
4384
4385 btrfs_init_map_token(&token, leaf);
4386 for (i = slot + nr; i < nritems; i++) {
4387 u32 ioff;
4388
4389 ioff = btrfs_token_item_offset(&token, i);
4390 btrfs_set_token_item_offset(&token, i, ioff + dsize);
4391 }
4392
4393 memmove_leaf_items(leaf, slot, slot + nr, nritems - slot - nr);
4394 }
4395 btrfs_set_header_nritems(leaf, nritems - nr);
4396 nritems -= nr;
4397
4398 /* delete the leaf if we've emptied it */
4399 if (nritems == 0) {
4400 if (leaf == root->node) {
4401 btrfs_set_header_level(leaf, 0);
4402 } else {
4403 btrfs_clean_tree_block(leaf);
4404 btrfs_del_leaf(trans, root, path, leaf);
4405 }
4406 } else {
4407 int used = leaf_space_used(leaf, 0, nritems);
4408 if (slot == 0) {
4409 struct btrfs_disk_key disk_key;
4410
4411 btrfs_item_key(leaf, &disk_key, 0);
4412 fixup_low_keys(path, &disk_key, 1);
4413 }
4414
4415 /*
4416 * Try to delete the leaf if it is mostly empty. We do this by
4417 * trying to move all its items into its left and right neighbours.
4418 * If we can't move all the items, then we don't delete it - it's
4419 * not ideal, but future insertions might fill the leaf with more
4420 * items, or items from other leaves might be moved later into our
4421 * leaf due to deletions on those leaves.
4422 */
4423 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
4424 u32 min_push_space;
4425
4426 /* push_leaf_left fixes the path.
4427 * make sure the path still points to our leaf
4428 * for possible call to del_ptr below
4429 */
4430 slot = path->slots[1];
4431 atomic_inc(&leaf->refs);
4432 /*
4433 * We want to be able to at least push one item to the
4434 * left neighbour leaf, and that's the first item.
4435 */
4436 min_push_space = sizeof(struct btrfs_item) +
4437 btrfs_item_size(leaf, 0);
4438 wret = push_leaf_left(trans, root, path, 0,
4439 min_push_space, 1, (u32)-1);
4440 if (wret < 0 && wret != -ENOSPC)
4441 ret = wret;
4442
4443 if (path->nodes[0] == leaf &&
4444 btrfs_header_nritems(leaf)) {
4445 /*
4446 * If we were not able to push all items from our
4447 * leaf to its left neighbour, then attempt to
4448 * either push all the remaining items to the
4449 * right neighbour or none. There's no advantage
4450 * in pushing only some items, instead of all, as
4451 * it's pointless to end up with a leaf having
4452 * too few items while the neighbours can be full
4453 * or nearly full.
4454 */
4455 nritems = btrfs_header_nritems(leaf);
4456 min_push_space = leaf_space_used(leaf, 0, nritems);
4457 wret = push_leaf_right(trans, root, path, 0,
4458 min_push_space, 1, 0);
4459 if (wret < 0 && wret != -ENOSPC)
4460 ret = wret;
4461 }
4462
4463 if (btrfs_header_nritems(leaf) == 0) {
4464 path->slots[1] = slot;
4465 btrfs_del_leaf(trans, root, path, leaf);
4466 free_extent_buffer(leaf);
4467 ret = 0;
4468 } else {
4469 /* if we're still in the path, make sure
4470 * we're dirty. Otherwise, one of the
4471 * push_leaf functions must have already
4472 * dirtied this buffer
4473 */
4474 if (path->nodes[0] == leaf)
4475 btrfs_mark_buffer_dirty(leaf);
4476 free_extent_buffer(leaf);
4477 }
4478 } else {
4479 btrfs_mark_buffer_dirty(leaf);
4480 }
4481 }
4482 return ret;
4483}
4484
4485/*
4486 * search the tree again to find a leaf with lesser keys
4487 * returns 0 if it found something or 1 if there are no lesser leaves.
4488 * returns < 0 on io errors.
4489 *
4490 * This may release the path, and so you may lose any locks held at the
4491 * time you call it.
4492 */
4493int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
4494{
4495 struct btrfs_key key;
4496 struct btrfs_disk_key found_key;
4497 int ret;
4498
4499 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
4500
4501 if (key.offset > 0) {
4502 key.offset--;
4503 } else if (key.type > 0) {
4504 key.type--;
4505 key.offset = (u64)-1;
4506 } else if (key.objectid > 0) {
4507 key.objectid--;
4508 key.type = (u8)-1;
4509 key.offset = (u64)-1;
4510 } else {
4511 return 1;
4512 }
4513
4514 btrfs_release_path(path);
4515 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4516 if (ret < 0)
4517 return ret;
4518 btrfs_item_key(path->nodes[0], &found_key, 0);
4519 ret = comp_keys(&found_key, &key);
4520 /*
4521 * We might have had an item with the previous key in the tree right
4522 * before we released our path. And after we released our path, that
4523 * item might have been pushed to the first slot (0) of the leaf we
4524 * were holding due to a tree balance. Alternatively, an item with the
4525 * previous key can exist as the only element of a leaf (big fat item).
4526 * Therefore account for these 2 cases, so that our callers (like
4527 * btrfs_previous_item) don't miss an existing item with a key matching
4528 * the previous key we computed above.
4529 */
4530 if (ret <= 0)
4531 return 0;
4532 return 1;
4533}
4534
4535/*
4536 * A helper function to walk down the tree starting at min_key, and looking
4537 * for nodes or leaves that are have a minimum transaction id.
4538 * This is used by the btree defrag code, and tree logging
4539 *
4540 * This does not cow, but it does stuff the starting key it finds back
4541 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4542 * key and get a writable path.
4543 *
4544 * This honors path->lowest_level to prevent descent past a given level
4545 * of the tree.
4546 *
4547 * min_trans indicates the oldest transaction that you are interested
4548 * in walking through. Any nodes or leaves older than min_trans are
4549 * skipped over (without reading them).
4550 *
4551 * returns zero if something useful was found, < 0 on error and 1 if there
4552 * was nothing in the tree that matched the search criteria.
4553 */
4554int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
4555 struct btrfs_path *path,
4556 u64 min_trans)
4557{
4558 struct extent_buffer *cur;
4559 struct btrfs_key found_key;
4560 int slot;
4561 int sret;
4562 u32 nritems;
4563 int level;
4564 int ret = 1;
4565 int keep_locks = path->keep_locks;
4566
4567 ASSERT(!path->nowait);
4568 path->keep_locks = 1;
4569again:
4570 cur = btrfs_read_lock_root_node(root);
4571 level = btrfs_header_level(cur);
4572 WARN_ON(path->nodes[level]);
4573 path->nodes[level] = cur;
4574 path->locks[level] = BTRFS_READ_LOCK;
4575
4576 if (btrfs_header_generation(cur) < min_trans) {
4577 ret = 1;
4578 goto out;
4579 }
4580 while (1) {
4581 nritems = btrfs_header_nritems(cur);
4582 level = btrfs_header_level(cur);
4583 sret = btrfs_bin_search(cur, min_key, &slot);
4584 if (sret < 0) {
4585 ret = sret;
4586 goto out;
4587 }
4588
4589 /* at the lowest level, we're done, setup the path and exit */
4590 if (level == path->lowest_level) {
4591 if (slot >= nritems)
4592 goto find_next_key;
4593 ret = 0;
4594 path->slots[level] = slot;
4595 btrfs_item_key_to_cpu(cur, &found_key, slot);
4596 goto out;
4597 }
4598 if (sret && slot > 0)
4599 slot--;
4600 /*
4601 * check this node pointer against the min_trans parameters.
4602 * If it is too old, skip to the next one.
4603 */
4604 while (slot < nritems) {
4605 u64 gen;
4606
4607 gen = btrfs_node_ptr_generation(cur, slot);
4608 if (gen < min_trans) {
4609 slot++;
4610 continue;
4611 }
4612 break;
4613 }
4614find_next_key:
4615 /*
4616 * we didn't find a candidate key in this node, walk forward
4617 * and find another one
4618 */
4619 if (slot >= nritems) {
4620 path->slots[level] = slot;
4621 sret = btrfs_find_next_key(root, path, min_key, level,
4622 min_trans);
4623 if (sret == 0) {
4624 btrfs_release_path(path);
4625 goto again;
4626 } else {
4627 goto out;
4628 }
4629 }
4630 /* save our key for returning back */
4631 btrfs_node_key_to_cpu(cur, &found_key, slot);
4632 path->slots[level] = slot;
4633 if (level == path->lowest_level) {
4634 ret = 0;
4635 goto out;
4636 }
4637 cur = btrfs_read_node_slot(cur, slot);
4638 if (IS_ERR(cur)) {
4639 ret = PTR_ERR(cur);
4640 goto out;
4641 }
4642
4643 btrfs_tree_read_lock(cur);
4644
4645 path->locks[level - 1] = BTRFS_READ_LOCK;
4646 path->nodes[level - 1] = cur;
4647 unlock_up(path, level, 1, 0, NULL);
4648 }
4649out:
4650 path->keep_locks = keep_locks;
4651 if (ret == 0) {
4652 btrfs_unlock_up_safe(path, path->lowest_level + 1);
4653 memcpy(min_key, &found_key, sizeof(found_key));
4654 }
4655 return ret;
4656}
4657
4658/*
4659 * this is similar to btrfs_next_leaf, but does not try to preserve
4660 * and fixup the path. It looks for and returns the next key in the
4661 * tree based on the current path and the min_trans parameters.
4662 *
4663 * 0 is returned if another key is found, < 0 if there are any errors
4664 * and 1 is returned if there are no higher keys in the tree
4665 *
4666 * path->keep_locks should be set to 1 on the search made before
4667 * calling this function.
4668 */
4669int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4670 struct btrfs_key *key, int level, u64 min_trans)
4671{
4672 int slot;
4673 struct extent_buffer *c;
4674
4675 WARN_ON(!path->keep_locks && !path->skip_locking);
4676 while (level < BTRFS_MAX_LEVEL) {
4677 if (!path->nodes[level])
4678 return 1;
4679
4680 slot = path->slots[level] + 1;
4681 c = path->nodes[level];
4682next:
4683 if (slot >= btrfs_header_nritems(c)) {
4684 int ret;
4685 int orig_lowest;
4686 struct btrfs_key cur_key;
4687 if (level + 1 >= BTRFS_MAX_LEVEL ||
4688 !path->nodes[level + 1])
4689 return 1;
4690
4691 if (path->locks[level + 1] || path->skip_locking) {
4692 level++;
4693 continue;
4694 }
4695
4696 slot = btrfs_header_nritems(c) - 1;
4697 if (level == 0)
4698 btrfs_item_key_to_cpu(c, &cur_key, slot);
4699 else
4700 btrfs_node_key_to_cpu(c, &cur_key, slot);
4701
4702 orig_lowest = path->lowest_level;
4703 btrfs_release_path(path);
4704 path->lowest_level = level;
4705 ret = btrfs_search_slot(NULL, root, &cur_key, path,
4706 0, 0);
4707 path->lowest_level = orig_lowest;
4708 if (ret < 0)
4709 return ret;
4710
4711 c = path->nodes[level];
4712 slot = path->slots[level];
4713 if (ret == 0)
4714 slot++;
4715 goto next;
4716 }
4717
4718 if (level == 0)
4719 btrfs_item_key_to_cpu(c, key, slot);
4720 else {
4721 u64 gen = btrfs_node_ptr_generation(c, slot);
4722
4723 if (gen < min_trans) {
4724 slot++;
4725 goto next;
4726 }
4727 btrfs_node_key_to_cpu(c, key, slot);
4728 }
4729 return 0;
4730 }
4731 return 1;
4732}
4733
4734int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
4735 u64 time_seq)
4736{
4737 int slot;
4738 int level;
4739 struct extent_buffer *c;
4740 struct extent_buffer *next;
4741 struct btrfs_fs_info *fs_info = root->fs_info;
4742 struct btrfs_key key;
4743 bool need_commit_sem = false;
4744 u32 nritems;
4745 int ret;
4746 int i;
4747
4748 /*
4749 * The nowait semantics are used only for write paths, where we don't
4750 * use the tree mod log and sequence numbers.
4751 */
4752 if (time_seq)
4753 ASSERT(!path->nowait);
4754
4755 nritems = btrfs_header_nritems(path->nodes[0]);
4756 if (nritems == 0)
4757 return 1;
4758
4759 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4760again:
4761 level = 1;
4762 next = NULL;
4763 btrfs_release_path(path);
4764
4765 path->keep_locks = 1;
4766
4767 if (time_seq) {
4768 ret = btrfs_search_old_slot(root, &key, path, time_seq);
4769 } else {
4770 if (path->need_commit_sem) {
4771 path->need_commit_sem = 0;
4772 need_commit_sem = true;
4773 if (path->nowait) {
4774 if (!down_read_trylock(&fs_info->commit_root_sem)) {
4775 ret = -EAGAIN;
4776 goto done;
4777 }
4778 } else {
4779 down_read(&fs_info->commit_root_sem);
4780 }
4781 }
4782 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4783 }
4784 path->keep_locks = 0;
4785
4786 if (ret < 0)
4787 goto done;
4788
4789 nritems = btrfs_header_nritems(path->nodes[0]);
4790 /*
4791 * by releasing the path above we dropped all our locks. A balance
4792 * could have added more items next to the key that used to be
4793 * at the very end of the block. So, check again here and
4794 * advance the path if there are now more items available.
4795 */
4796 if (nritems > 0 && path->slots[0] < nritems - 1) {
4797 if (ret == 0)
4798 path->slots[0]++;
4799 ret = 0;
4800 goto done;
4801 }
4802 /*
4803 * So the above check misses one case:
4804 * - after releasing the path above, someone has removed the item that
4805 * used to be at the very end of the block, and balance between leafs
4806 * gets another one with bigger key.offset to replace it.
4807 *
4808 * This one should be returned as well, or we can get leaf corruption
4809 * later(esp. in __btrfs_drop_extents()).
4810 *
4811 * And a bit more explanation about this check,
4812 * with ret > 0, the key isn't found, the path points to the slot
4813 * where it should be inserted, so the path->slots[0] item must be the
4814 * bigger one.
4815 */
4816 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
4817 ret = 0;
4818 goto done;
4819 }
4820
4821 while (level < BTRFS_MAX_LEVEL) {
4822 if (!path->nodes[level]) {
4823 ret = 1;
4824 goto done;
4825 }
4826
4827 slot = path->slots[level] + 1;
4828 c = path->nodes[level];
4829 if (slot >= btrfs_header_nritems(c)) {
4830 level++;
4831 if (level == BTRFS_MAX_LEVEL) {
4832 ret = 1;
4833 goto done;
4834 }
4835 continue;
4836 }
4837
4838
4839 /*
4840 * Our current level is where we're going to start from, and to
4841 * make sure lockdep doesn't complain we need to drop our locks
4842 * and nodes from 0 to our current level.
4843 */
4844 for (i = 0; i < level; i++) {
4845 if (path->locks[level]) {
4846 btrfs_tree_read_unlock(path->nodes[i]);
4847 path->locks[i] = 0;
4848 }
4849 free_extent_buffer(path->nodes[i]);
4850 path->nodes[i] = NULL;
4851 }
4852
4853 next = c;
4854 ret = read_block_for_search(root, path, &next, level,
4855 slot, &key);
4856 if (ret == -EAGAIN && !path->nowait)
4857 goto again;
4858
4859 if (ret < 0) {
4860 btrfs_release_path(path);
4861 goto done;
4862 }
4863
4864 if (!path->skip_locking) {
4865 ret = btrfs_try_tree_read_lock(next);
4866 if (!ret && path->nowait) {
4867 ret = -EAGAIN;
4868 goto done;
4869 }
4870 if (!ret && time_seq) {
4871 /*
4872 * If we don't get the lock, we may be racing
4873 * with push_leaf_left, holding that lock while
4874 * itself waiting for the leaf we've currently
4875 * locked. To solve this situation, we give up
4876 * on our lock and cycle.
4877 */
4878 free_extent_buffer(next);
4879 btrfs_release_path(path);
4880 cond_resched();
4881 goto again;
4882 }
4883 if (!ret)
4884 btrfs_tree_read_lock(next);
4885 }
4886 break;
4887 }
4888 path->slots[level] = slot;
4889 while (1) {
4890 level--;
4891 path->nodes[level] = next;
4892 path->slots[level] = 0;
4893 if (!path->skip_locking)
4894 path->locks[level] = BTRFS_READ_LOCK;
4895 if (!level)
4896 break;
4897
4898 ret = read_block_for_search(root, path, &next, level,
4899 0, &key);
4900 if (ret == -EAGAIN && !path->nowait)
4901 goto again;
4902
4903 if (ret < 0) {
4904 btrfs_release_path(path);
4905 goto done;
4906 }
4907
4908 if (!path->skip_locking) {
4909 if (path->nowait) {
4910 if (!btrfs_try_tree_read_lock(next)) {
4911 ret = -EAGAIN;
4912 goto done;
4913 }
4914 } else {
4915 btrfs_tree_read_lock(next);
4916 }
4917 }
4918 }
4919 ret = 0;
4920done:
4921 unlock_up(path, 0, 1, 0, NULL);
4922 if (need_commit_sem) {
4923 int ret2;
4924
4925 path->need_commit_sem = 1;
4926 ret2 = finish_need_commit_sem_search(path);
4927 up_read(&fs_info->commit_root_sem);
4928 if (ret2)
4929 ret = ret2;
4930 }
4931
4932 return ret;
4933}
4934
4935int btrfs_next_old_item(struct btrfs_root *root, struct btrfs_path *path, u64 time_seq)
4936{
4937 path->slots[0]++;
4938 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
4939 return btrfs_next_old_leaf(root, path, time_seq);
4940 return 0;
4941}
4942
4943/*
4944 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4945 * searching until it gets past min_objectid or finds an item of 'type'
4946 *
4947 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4948 */
4949int btrfs_previous_item(struct btrfs_root *root,
4950 struct btrfs_path *path, u64 min_objectid,
4951 int type)
4952{
4953 struct btrfs_key found_key;
4954 struct extent_buffer *leaf;
4955 u32 nritems;
4956 int ret;
4957
4958 while (1) {
4959 if (path->slots[0] == 0) {
4960 ret = btrfs_prev_leaf(root, path);
4961 if (ret != 0)
4962 return ret;
4963 } else {
4964 path->slots[0]--;
4965 }
4966 leaf = path->nodes[0];
4967 nritems = btrfs_header_nritems(leaf);
4968 if (nritems == 0)
4969 return 1;
4970 if (path->slots[0] == nritems)
4971 path->slots[0]--;
4972
4973 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4974 if (found_key.objectid < min_objectid)
4975 break;
4976 if (found_key.type == type)
4977 return 0;
4978 if (found_key.objectid == min_objectid &&
4979 found_key.type < type)
4980 break;
4981 }
4982 return 1;
4983}
4984
4985/*
4986 * search in extent tree to find a previous Metadata/Data extent item with
4987 * min objecitd.
4988 *
4989 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4990 */
4991int btrfs_previous_extent_item(struct btrfs_root *root,
4992 struct btrfs_path *path, u64 min_objectid)
4993{
4994 struct btrfs_key found_key;
4995 struct extent_buffer *leaf;
4996 u32 nritems;
4997 int ret;
4998
4999 while (1) {
5000 if (path->slots[0] == 0) {
5001 ret = btrfs_prev_leaf(root, path);
5002 if (ret != 0)
5003 return ret;
5004 } else {
5005 path->slots[0]--;
5006 }
5007 leaf = path->nodes[0];
5008 nritems = btrfs_header_nritems(leaf);
5009 if (nritems == 0)
5010 return 1;
5011 if (path->slots[0] == nritems)
5012 path->slots[0]--;
5013
5014 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5015 if (found_key.objectid < min_objectid)
5016 break;
5017 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5018 found_key.type == BTRFS_METADATA_ITEM_KEY)
5019 return 0;
5020 if (found_key.objectid == min_objectid &&
5021 found_key.type < BTRFS_EXTENT_ITEM_KEY)
5022 break;
5023 }
5024 return 1;
5025}
5026
5027int __init btrfs_ctree_init(void)
5028{
5029 btrfs_path_cachep = kmem_cache_create("btrfs_path",
5030 sizeof(struct btrfs_path), 0,
5031 SLAB_MEM_SPREAD, NULL);
5032 if (!btrfs_path_cachep)
5033 return -ENOMEM;
5034 return 0;
5035}
5036
5037void __cold btrfs_ctree_exit(void)
5038{
5039 kmem_cache_destroy(btrfs_path_cachep);
5040}