]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/btrfs/ctree.c
Merge tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
[mirror_ubuntu-bionic-kernel.git] / fs / btrfs / ctree.c
1 /*
2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/rbtree.h>
22 #include <linux/vmalloc.h>
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "transaction.h"
26 #include "print-tree.h"
27 #include "locking.h"
28
29 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
30 *root, struct btrfs_path *path, int level);
31 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
32 *root, struct btrfs_key *ins_key,
33 struct btrfs_path *path, int data_size, int extend);
34 static int push_node_left(struct btrfs_trans_handle *trans,
35 struct btrfs_root *root, struct extent_buffer *dst,
36 struct extent_buffer *src, int empty);
37 static int balance_node_right(struct btrfs_trans_handle *trans,
38 struct btrfs_root *root,
39 struct extent_buffer *dst_buf,
40 struct extent_buffer *src_buf);
41 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
42 int level, int slot);
43 static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
44 struct extent_buffer *eb);
45
46 struct btrfs_path *btrfs_alloc_path(void)
47 {
48 struct btrfs_path *path;
49 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
50 return path;
51 }
52
53 /*
54 * set all locked nodes in the path to blocking locks. This should
55 * be done before scheduling
56 */
57 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
58 {
59 int i;
60 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
61 if (!p->nodes[i] || !p->locks[i])
62 continue;
63 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
64 if (p->locks[i] == BTRFS_READ_LOCK)
65 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
66 else if (p->locks[i] == BTRFS_WRITE_LOCK)
67 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
68 }
69 }
70
71 /*
72 * reset all the locked nodes in the patch to spinning locks.
73 *
74 * held is used to keep lockdep happy, when lockdep is enabled
75 * we set held to a blocking lock before we go around and
76 * retake all the spinlocks in the path. You can safely use NULL
77 * for held
78 */
79 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
80 struct extent_buffer *held, int held_rw)
81 {
82 int i;
83
84 if (held) {
85 btrfs_set_lock_blocking_rw(held, held_rw);
86 if (held_rw == BTRFS_WRITE_LOCK)
87 held_rw = BTRFS_WRITE_LOCK_BLOCKING;
88 else if (held_rw == BTRFS_READ_LOCK)
89 held_rw = BTRFS_READ_LOCK_BLOCKING;
90 }
91 btrfs_set_path_blocking(p);
92
93 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
94 if (p->nodes[i] && p->locks[i]) {
95 btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
96 if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
97 p->locks[i] = BTRFS_WRITE_LOCK;
98 else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
99 p->locks[i] = BTRFS_READ_LOCK;
100 }
101 }
102
103 if (held)
104 btrfs_clear_lock_blocking_rw(held, held_rw);
105 }
106
107 /* this also releases the path */
108 void btrfs_free_path(struct btrfs_path *p)
109 {
110 if (!p)
111 return;
112 btrfs_release_path(p);
113 kmem_cache_free(btrfs_path_cachep, p);
114 }
115
116 /*
117 * path release drops references on the extent buffers in the path
118 * and it drops any locks held by this path
119 *
120 * It is safe to call this on paths that no locks or extent buffers held.
121 */
122 noinline void btrfs_release_path(struct btrfs_path *p)
123 {
124 int i;
125
126 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
127 p->slots[i] = 0;
128 if (!p->nodes[i])
129 continue;
130 if (p->locks[i]) {
131 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
132 p->locks[i] = 0;
133 }
134 free_extent_buffer(p->nodes[i]);
135 p->nodes[i] = NULL;
136 }
137 }
138
139 /*
140 * safely gets a reference on the root node of a tree. A lock
141 * is not taken, so a concurrent writer may put a different node
142 * at the root of the tree. See btrfs_lock_root_node for the
143 * looping required.
144 *
145 * The extent buffer returned by this has a reference taken, so
146 * it won't disappear. It may stop being the root of the tree
147 * at any time because there are no locks held.
148 */
149 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
150 {
151 struct extent_buffer *eb;
152
153 while (1) {
154 rcu_read_lock();
155 eb = rcu_dereference(root->node);
156
157 /*
158 * RCU really hurts here, we could free up the root node because
159 * it was COWed but we may not get the new root node yet so do
160 * the inc_not_zero dance and if it doesn't work then
161 * synchronize_rcu and try again.
162 */
163 if (atomic_inc_not_zero(&eb->refs)) {
164 rcu_read_unlock();
165 break;
166 }
167 rcu_read_unlock();
168 synchronize_rcu();
169 }
170 return eb;
171 }
172
173 /* loop around taking references on and locking the root node of the
174 * tree until you end up with a lock on the root. A locked buffer
175 * is returned, with a reference held.
176 */
177 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
178 {
179 struct extent_buffer *eb;
180
181 while (1) {
182 eb = btrfs_root_node(root);
183 btrfs_tree_lock(eb);
184 if (eb == root->node)
185 break;
186 btrfs_tree_unlock(eb);
187 free_extent_buffer(eb);
188 }
189 return eb;
190 }
191
192 /* loop around taking references on and locking the root node of the
193 * tree until you end up with a lock on the root. A locked buffer
194 * is returned, with a reference held.
195 */
196 static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
197 {
198 struct extent_buffer *eb;
199
200 while (1) {
201 eb = btrfs_root_node(root);
202 btrfs_tree_read_lock(eb);
203 if (eb == root->node)
204 break;
205 btrfs_tree_read_unlock(eb);
206 free_extent_buffer(eb);
207 }
208 return eb;
209 }
210
211 /* cowonly root (everything not a reference counted cow subvolume), just get
212 * put onto a simple dirty list. transaction.c walks this to make sure they
213 * get properly updated on disk.
214 */
215 static void add_root_to_dirty_list(struct btrfs_root *root)
216 {
217 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
218 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
219 return;
220
221 spin_lock(&root->fs_info->trans_lock);
222 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
223 /* Want the extent tree to be the last on the list */
224 if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID)
225 list_move_tail(&root->dirty_list,
226 &root->fs_info->dirty_cowonly_roots);
227 else
228 list_move(&root->dirty_list,
229 &root->fs_info->dirty_cowonly_roots);
230 }
231 spin_unlock(&root->fs_info->trans_lock);
232 }
233
234 /*
235 * used by snapshot creation to make a copy of a root for a tree with
236 * a given objectid. The buffer with the new root node is returned in
237 * cow_ret, and this func returns zero on success or a negative error code.
238 */
239 int btrfs_copy_root(struct btrfs_trans_handle *trans,
240 struct btrfs_root *root,
241 struct extent_buffer *buf,
242 struct extent_buffer **cow_ret, u64 new_root_objectid)
243 {
244 struct extent_buffer *cow;
245 int ret = 0;
246 int level;
247 struct btrfs_disk_key disk_key;
248
249 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
250 trans->transid != root->fs_info->running_transaction->transid);
251 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
252 trans->transid != root->last_trans);
253
254 level = btrfs_header_level(buf);
255 if (level == 0)
256 btrfs_item_key(buf, &disk_key, 0);
257 else
258 btrfs_node_key(buf, &disk_key, 0);
259
260 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
261 &disk_key, level, buf->start, 0);
262 if (IS_ERR(cow))
263 return PTR_ERR(cow);
264
265 copy_extent_buffer(cow, buf, 0, 0, cow->len);
266 btrfs_set_header_bytenr(cow, cow->start);
267 btrfs_set_header_generation(cow, trans->transid);
268 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
269 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
270 BTRFS_HEADER_FLAG_RELOC);
271 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
272 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
273 else
274 btrfs_set_header_owner(cow, new_root_objectid);
275
276 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
277 BTRFS_FSID_SIZE);
278
279 WARN_ON(btrfs_header_generation(buf) > trans->transid);
280 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
281 ret = btrfs_inc_ref(trans, root, cow, 1);
282 else
283 ret = btrfs_inc_ref(trans, root, cow, 0);
284
285 if (ret)
286 return ret;
287
288 btrfs_mark_buffer_dirty(cow);
289 *cow_ret = cow;
290 return 0;
291 }
292
293 enum mod_log_op {
294 MOD_LOG_KEY_REPLACE,
295 MOD_LOG_KEY_ADD,
296 MOD_LOG_KEY_REMOVE,
297 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
298 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
299 MOD_LOG_MOVE_KEYS,
300 MOD_LOG_ROOT_REPLACE,
301 };
302
303 struct tree_mod_move {
304 int dst_slot;
305 int nr_items;
306 };
307
308 struct tree_mod_root {
309 u64 logical;
310 u8 level;
311 };
312
313 struct tree_mod_elem {
314 struct rb_node node;
315 u64 logical;
316 u64 seq;
317 enum mod_log_op op;
318
319 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
320 int slot;
321
322 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
323 u64 generation;
324
325 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
326 struct btrfs_disk_key key;
327 u64 blockptr;
328
329 /* this is used for op == MOD_LOG_MOVE_KEYS */
330 struct tree_mod_move move;
331
332 /* this is used for op == MOD_LOG_ROOT_REPLACE */
333 struct tree_mod_root old_root;
334 };
335
336 static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
337 {
338 read_lock(&fs_info->tree_mod_log_lock);
339 }
340
341 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
342 {
343 read_unlock(&fs_info->tree_mod_log_lock);
344 }
345
346 static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
347 {
348 write_lock(&fs_info->tree_mod_log_lock);
349 }
350
351 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
352 {
353 write_unlock(&fs_info->tree_mod_log_lock);
354 }
355
356 /*
357 * Pull a new tree mod seq number for our operation.
358 */
359 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
360 {
361 return atomic64_inc_return(&fs_info->tree_mod_seq);
362 }
363
364 /*
365 * This adds a new blocker to the tree mod log's blocker list if the @elem
366 * passed does not already have a sequence number set. So when a caller expects
367 * to record tree modifications, it should ensure to set elem->seq to zero
368 * before calling btrfs_get_tree_mod_seq.
369 * Returns a fresh, unused tree log modification sequence number, even if no new
370 * blocker was added.
371 */
372 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
373 struct seq_list *elem)
374 {
375 tree_mod_log_write_lock(fs_info);
376 spin_lock(&fs_info->tree_mod_seq_lock);
377 if (!elem->seq) {
378 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
379 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
380 }
381 spin_unlock(&fs_info->tree_mod_seq_lock);
382 tree_mod_log_write_unlock(fs_info);
383
384 return elem->seq;
385 }
386
387 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
388 struct seq_list *elem)
389 {
390 struct rb_root *tm_root;
391 struct rb_node *node;
392 struct rb_node *next;
393 struct seq_list *cur_elem;
394 struct tree_mod_elem *tm;
395 u64 min_seq = (u64)-1;
396 u64 seq_putting = elem->seq;
397
398 if (!seq_putting)
399 return;
400
401 spin_lock(&fs_info->tree_mod_seq_lock);
402 list_del(&elem->list);
403 elem->seq = 0;
404
405 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
406 if (cur_elem->seq < min_seq) {
407 if (seq_putting > cur_elem->seq) {
408 /*
409 * blocker with lower sequence number exists, we
410 * cannot remove anything from the log
411 */
412 spin_unlock(&fs_info->tree_mod_seq_lock);
413 return;
414 }
415 min_seq = cur_elem->seq;
416 }
417 }
418 spin_unlock(&fs_info->tree_mod_seq_lock);
419
420 /*
421 * anything that's lower than the lowest existing (read: blocked)
422 * sequence number can be removed from the tree.
423 */
424 tree_mod_log_write_lock(fs_info);
425 tm_root = &fs_info->tree_mod_log;
426 for (node = rb_first(tm_root); node; node = next) {
427 next = rb_next(node);
428 tm = container_of(node, struct tree_mod_elem, node);
429 if (tm->seq > min_seq)
430 continue;
431 rb_erase(node, tm_root);
432 kfree(tm);
433 }
434 tree_mod_log_write_unlock(fs_info);
435 }
436
437 /*
438 * key order of the log:
439 * node/leaf start address -> sequence
440 *
441 * The 'start address' is the logical address of the *new* root node
442 * for root replace operations, or the logical address of the affected
443 * block for all other operations.
444 *
445 * Note: must be called with write lock (tree_mod_log_write_lock).
446 */
447 static noinline int
448 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
449 {
450 struct rb_root *tm_root;
451 struct rb_node **new;
452 struct rb_node *parent = NULL;
453 struct tree_mod_elem *cur;
454
455 BUG_ON(!tm);
456
457 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
458
459 tm_root = &fs_info->tree_mod_log;
460 new = &tm_root->rb_node;
461 while (*new) {
462 cur = container_of(*new, struct tree_mod_elem, node);
463 parent = *new;
464 if (cur->logical < tm->logical)
465 new = &((*new)->rb_left);
466 else if (cur->logical > tm->logical)
467 new = &((*new)->rb_right);
468 else if (cur->seq < tm->seq)
469 new = &((*new)->rb_left);
470 else if (cur->seq > tm->seq)
471 new = &((*new)->rb_right);
472 else
473 return -EEXIST;
474 }
475
476 rb_link_node(&tm->node, parent, new);
477 rb_insert_color(&tm->node, tm_root);
478 return 0;
479 }
480
481 /*
482 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
483 * returns zero with the tree_mod_log_lock acquired. The caller must hold
484 * this until all tree mod log insertions are recorded in the rb tree and then
485 * call tree_mod_log_write_unlock() to release.
486 */
487 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
488 struct extent_buffer *eb) {
489 smp_mb();
490 if (list_empty(&(fs_info)->tree_mod_seq_list))
491 return 1;
492 if (eb && btrfs_header_level(eb) == 0)
493 return 1;
494
495 tree_mod_log_write_lock(fs_info);
496 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
497 tree_mod_log_write_unlock(fs_info);
498 return 1;
499 }
500
501 return 0;
502 }
503
504 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
505 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
506 struct extent_buffer *eb)
507 {
508 smp_mb();
509 if (list_empty(&(fs_info)->tree_mod_seq_list))
510 return 0;
511 if (eb && btrfs_header_level(eb) == 0)
512 return 0;
513
514 return 1;
515 }
516
517 static struct tree_mod_elem *
518 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
519 enum mod_log_op op, gfp_t flags)
520 {
521 struct tree_mod_elem *tm;
522
523 tm = kzalloc(sizeof(*tm), flags);
524 if (!tm)
525 return NULL;
526
527 tm->logical = eb->start;
528 if (op != MOD_LOG_KEY_ADD) {
529 btrfs_node_key(eb, &tm->key, slot);
530 tm->blockptr = btrfs_node_blockptr(eb, slot);
531 }
532 tm->op = op;
533 tm->slot = slot;
534 tm->generation = btrfs_node_ptr_generation(eb, slot);
535 RB_CLEAR_NODE(&tm->node);
536
537 return tm;
538 }
539
540 static noinline int
541 tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
542 struct extent_buffer *eb, int slot,
543 enum mod_log_op op, gfp_t flags)
544 {
545 struct tree_mod_elem *tm;
546 int ret;
547
548 if (!tree_mod_need_log(fs_info, eb))
549 return 0;
550
551 tm = alloc_tree_mod_elem(eb, slot, op, flags);
552 if (!tm)
553 return -ENOMEM;
554
555 if (tree_mod_dont_log(fs_info, eb)) {
556 kfree(tm);
557 return 0;
558 }
559
560 ret = __tree_mod_log_insert(fs_info, tm);
561 tree_mod_log_write_unlock(fs_info);
562 if (ret)
563 kfree(tm);
564
565 return ret;
566 }
567
568 static noinline int
569 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
570 struct extent_buffer *eb, int dst_slot, int src_slot,
571 int nr_items, gfp_t flags)
572 {
573 struct tree_mod_elem *tm = NULL;
574 struct tree_mod_elem **tm_list = NULL;
575 int ret = 0;
576 int i;
577 int locked = 0;
578
579 if (!tree_mod_need_log(fs_info, eb))
580 return 0;
581
582 tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), flags);
583 if (!tm_list)
584 return -ENOMEM;
585
586 tm = kzalloc(sizeof(*tm), flags);
587 if (!tm) {
588 ret = -ENOMEM;
589 goto free_tms;
590 }
591
592 tm->logical = eb->start;
593 tm->slot = src_slot;
594 tm->move.dst_slot = dst_slot;
595 tm->move.nr_items = nr_items;
596 tm->op = MOD_LOG_MOVE_KEYS;
597
598 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
599 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
600 MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags);
601 if (!tm_list[i]) {
602 ret = -ENOMEM;
603 goto free_tms;
604 }
605 }
606
607 if (tree_mod_dont_log(fs_info, eb))
608 goto free_tms;
609 locked = 1;
610
611 /*
612 * When we override something during the move, we log these removals.
613 * This can only happen when we move towards the beginning of the
614 * buffer, i.e. dst_slot < src_slot.
615 */
616 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
617 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
618 if (ret)
619 goto free_tms;
620 }
621
622 ret = __tree_mod_log_insert(fs_info, tm);
623 if (ret)
624 goto free_tms;
625 tree_mod_log_write_unlock(fs_info);
626 kfree(tm_list);
627
628 return 0;
629 free_tms:
630 for (i = 0; i < nr_items; i++) {
631 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
632 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
633 kfree(tm_list[i]);
634 }
635 if (locked)
636 tree_mod_log_write_unlock(fs_info);
637 kfree(tm_list);
638 kfree(tm);
639
640 return ret;
641 }
642
643 static inline int
644 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
645 struct tree_mod_elem **tm_list,
646 int nritems)
647 {
648 int i, j;
649 int ret;
650
651 for (i = nritems - 1; i >= 0; i--) {
652 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
653 if (ret) {
654 for (j = nritems - 1; j > i; j--)
655 rb_erase(&tm_list[j]->node,
656 &fs_info->tree_mod_log);
657 return ret;
658 }
659 }
660
661 return 0;
662 }
663
664 static noinline int
665 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
666 struct extent_buffer *old_root,
667 struct extent_buffer *new_root, gfp_t flags,
668 int log_removal)
669 {
670 struct tree_mod_elem *tm = NULL;
671 struct tree_mod_elem **tm_list = NULL;
672 int nritems = 0;
673 int ret = 0;
674 int i;
675
676 if (!tree_mod_need_log(fs_info, NULL))
677 return 0;
678
679 if (log_removal && btrfs_header_level(old_root) > 0) {
680 nritems = btrfs_header_nritems(old_root);
681 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
682 flags);
683 if (!tm_list) {
684 ret = -ENOMEM;
685 goto free_tms;
686 }
687 for (i = 0; i < nritems; i++) {
688 tm_list[i] = alloc_tree_mod_elem(old_root, i,
689 MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags);
690 if (!tm_list[i]) {
691 ret = -ENOMEM;
692 goto free_tms;
693 }
694 }
695 }
696
697 tm = kzalloc(sizeof(*tm), flags);
698 if (!tm) {
699 ret = -ENOMEM;
700 goto free_tms;
701 }
702
703 tm->logical = new_root->start;
704 tm->old_root.logical = old_root->start;
705 tm->old_root.level = btrfs_header_level(old_root);
706 tm->generation = btrfs_header_generation(old_root);
707 tm->op = MOD_LOG_ROOT_REPLACE;
708
709 if (tree_mod_dont_log(fs_info, NULL))
710 goto free_tms;
711
712 if (tm_list)
713 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
714 if (!ret)
715 ret = __tree_mod_log_insert(fs_info, tm);
716
717 tree_mod_log_write_unlock(fs_info);
718 if (ret)
719 goto free_tms;
720 kfree(tm_list);
721
722 return ret;
723
724 free_tms:
725 if (tm_list) {
726 for (i = 0; i < nritems; i++)
727 kfree(tm_list[i]);
728 kfree(tm_list);
729 }
730 kfree(tm);
731
732 return ret;
733 }
734
735 static struct tree_mod_elem *
736 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
737 int smallest)
738 {
739 struct rb_root *tm_root;
740 struct rb_node *node;
741 struct tree_mod_elem *cur = NULL;
742 struct tree_mod_elem *found = NULL;
743
744 tree_mod_log_read_lock(fs_info);
745 tm_root = &fs_info->tree_mod_log;
746 node = tm_root->rb_node;
747 while (node) {
748 cur = container_of(node, struct tree_mod_elem, node);
749 if (cur->logical < start) {
750 node = node->rb_left;
751 } else if (cur->logical > start) {
752 node = node->rb_right;
753 } else if (cur->seq < min_seq) {
754 node = node->rb_left;
755 } else if (!smallest) {
756 /* we want the node with the highest seq */
757 if (found)
758 BUG_ON(found->seq > cur->seq);
759 found = cur;
760 node = node->rb_left;
761 } else if (cur->seq > min_seq) {
762 /* we want the node with the smallest seq */
763 if (found)
764 BUG_ON(found->seq < cur->seq);
765 found = cur;
766 node = node->rb_right;
767 } else {
768 found = cur;
769 break;
770 }
771 }
772 tree_mod_log_read_unlock(fs_info);
773
774 return found;
775 }
776
777 /*
778 * this returns the element from the log with the smallest time sequence
779 * value that's in the log (the oldest log item). any element with a time
780 * sequence lower than min_seq will be ignored.
781 */
782 static struct tree_mod_elem *
783 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
784 u64 min_seq)
785 {
786 return __tree_mod_log_search(fs_info, start, min_seq, 1);
787 }
788
789 /*
790 * this returns the element from the log with the largest time sequence
791 * value that's in the log (the most recent log item). any element with
792 * a time sequence lower than min_seq will be ignored.
793 */
794 static struct tree_mod_elem *
795 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
796 {
797 return __tree_mod_log_search(fs_info, start, min_seq, 0);
798 }
799
800 static noinline int
801 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
802 struct extent_buffer *src, unsigned long dst_offset,
803 unsigned long src_offset, int nr_items)
804 {
805 int ret = 0;
806 struct tree_mod_elem **tm_list = NULL;
807 struct tree_mod_elem **tm_list_add, **tm_list_rem;
808 int i;
809 int locked = 0;
810
811 if (!tree_mod_need_log(fs_info, NULL))
812 return 0;
813
814 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
815 return 0;
816
817 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
818 GFP_NOFS);
819 if (!tm_list)
820 return -ENOMEM;
821
822 tm_list_add = tm_list;
823 tm_list_rem = tm_list + nr_items;
824 for (i = 0; i < nr_items; i++) {
825 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
826 MOD_LOG_KEY_REMOVE, GFP_NOFS);
827 if (!tm_list_rem[i]) {
828 ret = -ENOMEM;
829 goto free_tms;
830 }
831
832 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
833 MOD_LOG_KEY_ADD, GFP_NOFS);
834 if (!tm_list_add[i]) {
835 ret = -ENOMEM;
836 goto free_tms;
837 }
838 }
839
840 if (tree_mod_dont_log(fs_info, NULL))
841 goto free_tms;
842 locked = 1;
843
844 for (i = 0; i < nr_items; i++) {
845 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
846 if (ret)
847 goto free_tms;
848 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
849 if (ret)
850 goto free_tms;
851 }
852
853 tree_mod_log_write_unlock(fs_info);
854 kfree(tm_list);
855
856 return 0;
857
858 free_tms:
859 for (i = 0; i < nr_items * 2; i++) {
860 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
861 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
862 kfree(tm_list[i]);
863 }
864 if (locked)
865 tree_mod_log_write_unlock(fs_info);
866 kfree(tm_list);
867
868 return ret;
869 }
870
871 static inline void
872 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
873 int dst_offset, int src_offset, int nr_items)
874 {
875 int ret;
876 ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
877 nr_items, GFP_NOFS);
878 BUG_ON(ret < 0);
879 }
880
881 static noinline void
882 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
883 struct extent_buffer *eb, int slot, int atomic)
884 {
885 int ret;
886
887 ret = tree_mod_log_insert_key(fs_info, eb, slot,
888 MOD_LOG_KEY_REPLACE,
889 atomic ? GFP_ATOMIC : GFP_NOFS);
890 BUG_ON(ret < 0);
891 }
892
893 static noinline int
894 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
895 {
896 struct tree_mod_elem **tm_list = NULL;
897 int nritems = 0;
898 int i;
899 int ret = 0;
900
901 if (btrfs_header_level(eb) == 0)
902 return 0;
903
904 if (!tree_mod_need_log(fs_info, NULL))
905 return 0;
906
907 nritems = btrfs_header_nritems(eb);
908 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
909 if (!tm_list)
910 return -ENOMEM;
911
912 for (i = 0; i < nritems; i++) {
913 tm_list[i] = alloc_tree_mod_elem(eb, i,
914 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
915 if (!tm_list[i]) {
916 ret = -ENOMEM;
917 goto free_tms;
918 }
919 }
920
921 if (tree_mod_dont_log(fs_info, eb))
922 goto free_tms;
923
924 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
925 tree_mod_log_write_unlock(fs_info);
926 if (ret)
927 goto free_tms;
928 kfree(tm_list);
929
930 return 0;
931
932 free_tms:
933 for (i = 0; i < nritems; i++)
934 kfree(tm_list[i]);
935 kfree(tm_list);
936
937 return ret;
938 }
939
940 static noinline void
941 tree_mod_log_set_root_pointer(struct btrfs_root *root,
942 struct extent_buffer *new_root_node,
943 int log_removal)
944 {
945 int ret;
946 ret = tree_mod_log_insert_root(root->fs_info, root->node,
947 new_root_node, GFP_NOFS, log_removal);
948 BUG_ON(ret < 0);
949 }
950
951 /*
952 * check if the tree block can be shared by multiple trees
953 */
954 int btrfs_block_can_be_shared(struct btrfs_root *root,
955 struct extent_buffer *buf)
956 {
957 /*
958 * Tree blocks not in reference counted trees and tree roots
959 * are never shared. If a block was allocated after the last
960 * snapshot and the block was not allocated by tree relocation,
961 * we know the block is not shared.
962 */
963 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
964 buf != root->node && buf != root->commit_root &&
965 (btrfs_header_generation(buf) <=
966 btrfs_root_last_snapshot(&root->root_item) ||
967 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
968 return 1;
969 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
970 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
971 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
972 return 1;
973 #endif
974 return 0;
975 }
976
977 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
978 struct btrfs_root *root,
979 struct extent_buffer *buf,
980 struct extent_buffer *cow,
981 int *last_ref)
982 {
983 u64 refs;
984 u64 owner;
985 u64 flags;
986 u64 new_flags = 0;
987 int ret;
988
989 /*
990 * Backrefs update rules:
991 *
992 * Always use full backrefs for extent pointers in tree block
993 * allocated by tree relocation.
994 *
995 * If a shared tree block is no longer referenced by its owner
996 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
997 * use full backrefs for extent pointers in tree block.
998 *
999 * If a tree block is been relocating
1000 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1001 * use full backrefs for extent pointers in tree block.
1002 * The reason for this is some operations (such as drop tree)
1003 * are only allowed for blocks use full backrefs.
1004 */
1005
1006 if (btrfs_block_can_be_shared(root, buf)) {
1007 ret = btrfs_lookup_extent_info(trans, root, buf->start,
1008 btrfs_header_level(buf), 1,
1009 &refs, &flags);
1010 if (ret)
1011 return ret;
1012 if (refs == 0) {
1013 ret = -EROFS;
1014 btrfs_handle_fs_error(root->fs_info, ret, NULL);
1015 return ret;
1016 }
1017 } else {
1018 refs = 1;
1019 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1020 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1021 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1022 else
1023 flags = 0;
1024 }
1025
1026 owner = btrfs_header_owner(buf);
1027 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
1028 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
1029
1030 if (refs > 1) {
1031 if ((owner == root->root_key.objectid ||
1032 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
1033 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1034 ret = btrfs_inc_ref(trans, root, buf, 1);
1035 BUG_ON(ret); /* -ENOMEM */
1036
1037 if (root->root_key.objectid ==
1038 BTRFS_TREE_RELOC_OBJECTID) {
1039 ret = btrfs_dec_ref(trans, root, buf, 0);
1040 BUG_ON(ret); /* -ENOMEM */
1041 ret = btrfs_inc_ref(trans, root, cow, 1);
1042 BUG_ON(ret); /* -ENOMEM */
1043 }
1044 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
1045 } else {
1046
1047 if (root->root_key.objectid ==
1048 BTRFS_TREE_RELOC_OBJECTID)
1049 ret = btrfs_inc_ref(trans, root, cow, 1);
1050 else
1051 ret = btrfs_inc_ref(trans, root, cow, 0);
1052 BUG_ON(ret); /* -ENOMEM */
1053 }
1054 if (new_flags != 0) {
1055 int level = btrfs_header_level(buf);
1056
1057 ret = btrfs_set_disk_extent_flags(trans, root,
1058 buf->start,
1059 buf->len,
1060 new_flags, level, 0);
1061 if (ret)
1062 return ret;
1063 }
1064 } else {
1065 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
1066 if (root->root_key.objectid ==
1067 BTRFS_TREE_RELOC_OBJECTID)
1068 ret = btrfs_inc_ref(trans, root, cow, 1);
1069 else
1070 ret = btrfs_inc_ref(trans, root, cow, 0);
1071 BUG_ON(ret); /* -ENOMEM */
1072 ret = btrfs_dec_ref(trans, root, buf, 1);
1073 BUG_ON(ret); /* -ENOMEM */
1074 }
1075 clean_tree_block(trans, root->fs_info, buf);
1076 *last_ref = 1;
1077 }
1078 return 0;
1079 }
1080
1081 /*
1082 * does the dirty work in cow of a single block. The parent block (if
1083 * supplied) is updated to point to the new cow copy. The new buffer is marked
1084 * dirty and returned locked. If you modify the block it needs to be marked
1085 * dirty again.
1086 *
1087 * search_start -- an allocation hint for the new block
1088 *
1089 * empty_size -- a hint that you plan on doing more cow. This is the size in
1090 * bytes the allocator should try to find free next to the block it returns.
1091 * This is just a hint and may be ignored by the allocator.
1092 */
1093 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1094 struct btrfs_root *root,
1095 struct extent_buffer *buf,
1096 struct extent_buffer *parent, int parent_slot,
1097 struct extent_buffer **cow_ret,
1098 u64 search_start, u64 empty_size)
1099 {
1100 struct btrfs_disk_key disk_key;
1101 struct extent_buffer *cow;
1102 int level, ret;
1103 int last_ref = 0;
1104 int unlock_orig = 0;
1105 u64 parent_start;
1106
1107 if (*cow_ret == buf)
1108 unlock_orig = 1;
1109
1110 btrfs_assert_tree_locked(buf);
1111
1112 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1113 trans->transid != root->fs_info->running_transaction->transid);
1114 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1115 trans->transid != root->last_trans);
1116
1117 level = btrfs_header_level(buf);
1118
1119 if (level == 0)
1120 btrfs_item_key(buf, &disk_key, 0);
1121 else
1122 btrfs_node_key(buf, &disk_key, 0);
1123
1124 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1125 if (parent)
1126 parent_start = parent->start;
1127 else
1128 parent_start = 0;
1129 } else
1130 parent_start = 0;
1131
1132 cow = btrfs_alloc_tree_block(trans, root, parent_start,
1133 root->root_key.objectid, &disk_key, level,
1134 search_start, empty_size);
1135 if (IS_ERR(cow))
1136 return PTR_ERR(cow);
1137
1138 /* cow is set to blocking by btrfs_init_new_buffer */
1139
1140 copy_extent_buffer(cow, buf, 0, 0, cow->len);
1141 btrfs_set_header_bytenr(cow, cow->start);
1142 btrfs_set_header_generation(cow, trans->transid);
1143 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1144 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1145 BTRFS_HEADER_FLAG_RELOC);
1146 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1147 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1148 else
1149 btrfs_set_header_owner(cow, root->root_key.objectid);
1150
1151 write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
1152 BTRFS_FSID_SIZE);
1153
1154 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1155 if (ret) {
1156 btrfs_abort_transaction(trans, root, ret);
1157 return ret;
1158 }
1159
1160 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1161 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1162 if (ret) {
1163 btrfs_abort_transaction(trans, root, ret);
1164 return ret;
1165 }
1166 }
1167
1168 if (buf == root->node) {
1169 WARN_ON(parent && parent != buf);
1170 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1171 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1172 parent_start = buf->start;
1173 else
1174 parent_start = 0;
1175
1176 extent_buffer_get(cow);
1177 tree_mod_log_set_root_pointer(root, cow, 1);
1178 rcu_assign_pointer(root->node, cow);
1179
1180 btrfs_free_tree_block(trans, root, buf, parent_start,
1181 last_ref);
1182 free_extent_buffer(buf);
1183 add_root_to_dirty_list(root);
1184 } else {
1185 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1186 parent_start = parent->start;
1187 else
1188 parent_start = 0;
1189
1190 WARN_ON(trans->transid != btrfs_header_generation(parent));
1191 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1192 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1193 btrfs_set_node_blockptr(parent, parent_slot,
1194 cow->start);
1195 btrfs_set_node_ptr_generation(parent, parent_slot,
1196 trans->transid);
1197 btrfs_mark_buffer_dirty(parent);
1198 if (last_ref) {
1199 ret = tree_mod_log_free_eb(root->fs_info, buf);
1200 if (ret) {
1201 btrfs_abort_transaction(trans, root, ret);
1202 return ret;
1203 }
1204 }
1205 btrfs_free_tree_block(trans, root, buf, parent_start,
1206 last_ref);
1207 }
1208 if (unlock_orig)
1209 btrfs_tree_unlock(buf);
1210 free_extent_buffer_stale(buf);
1211 btrfs_mark_buffer_dirty(cow);
1212 *cow_ret = cow;
1213 return 0;
1214 }
1215
1216 /*
1217 * returns the logical address of the oldest predecessor of the given root.
1218 * entries older than time_seq are ignored.
1219 */
1220 static struct tree_mod_elem *
1221 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1222 struct extent_buffer *eb_root, u64 time_seq)
1223 {
1224 struct tree_mod_elem *tm;
1225 struct tree_mod_elem *found = NULL;
1226 u64 root_logical = eb_root->start;
1227 int looped = 0;
1228
1229 if (!time_seq)
1230 return NULL;
1231
1232 /*
1233 * the very last operation that's logged for a root is the
1234 * replacement operation (if it is replaced at all). this has
1235 * the logical address of the *new* root, making it the very
1236 * first operation that's logged for this root.
1237 */
1238 while (1) {
1239 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1240 time_seq);
1241 if (!looped && !tm)
1242 return NULL;
1243 /*
1244 * if there are no tree operation for the oldest root, we simply
1245 * return it. this should only happen if that (old) root is at
1246 * level 0.
1247 */
1248 if (!tm)
1249 break;
1250
1251 /*
1252 * if there's an operation that's not a root replacement, we
1253 * found the oldest version of our root. normally, we'll find a
1254 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1255 */
1256 if (tm->op != MOD_LOG_ROOT_REPLACE)
1257 break;
1258
1259 found = tm;
1260 root_logical = tm->old_root.logical;
1261 looped = 1;
1262 }
1263
1264 /* if there's no old root to return, return what we found instead */
1265 if (!found)
1266 found = tm;
1267
1268 return found;
1269 }
1270
1271 /*
1272 * tm is a pointer to the first operation to rewind within eb. then, all
1273 * previous operations will be rewound (until we reach something older than
1274 * time_seq).
1275 */
1276 static void
1277 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1278 u64 time_seq, struct tree_mod_elem *first_tm)
1279 {
1280 u32 n;
1281 struct rb_node *next;
1282 struct tree_mod_elem *tm = first_tm;
1283 unsigned long o_dst;
1284 unsigned long o_src;
1285 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1286
1287 n = btrfs_header_nritems(eb);
1288 tree_mod_log_read_lock(fs_info);
1289 while (tm && tm->seq >= time_seq) {
1290 /*
1291 * all the operations are recorded with the operator used for
1292 * the modification. as we're going backwards, we do the
1293 * opposite of each operation here.
1294 */
1295 switch (tm->op) {
1296 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1297 BUG_ON(tm->slot < n);
1298 /* Fallthrough */
1299 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1300 case MOD_LOG_KEY_REMOVE:
1301 btrfs_set_node_key(eb, &tm->key, tm->slot);
1302 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1303 btrfs_set_node_ptr_generation(eb, tm->slot,
1304 tm->generation);
1305 n++;
1306 break;
1307 case MOD_LOG_KEY_REPLACE:
1308 BUG_ON(tm->slot >= n);
1309 btrfs_set_node_key(eb, &tm->key, tm->slot);
1310 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1311 btrfs_set_node_ptr_generation(eb, tm->slot,
1312 tm->generation);
1313 break;
1314 case MOD_LOG_KEY_ADD:
1315 /* if a move operation is needed it's in the log */
1316 n--;
1317 break;
1318 case MOD_LOG_MOVE_KEYS:
1319 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1320 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1321 memmove_extent_buffer(eb, o_dst, o_src,
1322 tm->move.nr_items * p_size);
1323 break;
1324 case MOD_LOG_ROOT_REPLACE:
1325 /*
1326 * this operation is special. for roots, this must be
1327 * handled explicitly before rewinding.
1328 * for non-roots, this operation may exist if the node
1329 * was a root: root A -> child B; then A gets empty and
1330 * B is promoted to the new root. in the mod log, we'll
1331 * have a root-replace operation for B, a tree block
1332 * that is no root. we simply ignore that operation.
1333 */
1334 break;
1335 }
1336 next = rb_next(&tm->node);
1337 if (!next)
1338 break;
1339 tm = container_of(next, struct tree_mod_elem, node);
1340 if (tm->logical != first_tm->logical)
1341 break;
1342 }
1343 tree_mod_log_read_unlock(fs_info);
1344 btrfs_set_header_nritems(eb, n);
1345 }
1346
1347 /*
1348 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1349 * is returned. If rewind operations happen, a fresh buffer is returned. The
1350 * returned buffer is always read-locked. If the returned buffer is not the
1351 * input buffer, the lock on the input buffer is released and the input buffer
1352 * is freed (its refcount is decremented).
1353 */
1354 static struct extent_buffer *
1355 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1356 struct extent_buffer *eb, u64 time_seq)
1357 {
1358 struct extent_buffer *eb_rewin;
1359 struct tree_mod_elem *tm;
1360
1361 if (!time_seq)
1362 return eb;
1363
1364 if (btrfs_header_level(eb) == 0)
1365 return eb;
1366
1367 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1368 if (!tm)
1369 return eb;
1370
1371 btrfs_set_path_blocking(path);
1372 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1373
1374 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1375 BUG_ON(tm->slot != 0);
1376 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start,
1377 eb->len);
1378 if (!eb_rewin) {
1379 btrfs_tree_read_unlock_blocking(eb);
1380 free_extent_buffer(eb);
1381 return NULL;
1382 }
1383 btrfs_set_header_bytenr(eb_rewin, eb->start);
1384 btrfs_set_header_backref_rev(eb_rewin,
1385 btrfs_header_backref_rev(eb));
1386 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1387 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1388 } else {
1389 eb_rewin = btrfs_clone_extent_buffer(eb);
1390 if (!eb_rewin) {
1391 btrfs_tree_read_unlock_blocking(eb);
1392 free_extent_buffer(eb);
1393 return NULL;
1394 }
1395 }
1396
1397 btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
1398 btrfs_tree_read_unlock_blocking(eb);
1399 free_extent_buffer(eb);
1400
1401 extent_buffer_get(eb_rewin);
1402 btrfs_tree_read_lock(eb_rewin);
1403 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1404 WARN_ON(btrfs_header_nritems(eb_rewin) >
1405 BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1406
1407 return eb_rewin;
1408 }
1409
1410 /*
1411 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1412 * value. If there are no changes, the current root->root_node is returned. If
1413 * anything changed in between, there's a fresh buffer allocated on which the
1414 * rewind operations are done. In any case, the returned buffer is read locked.
1415 * Returns NULL on error (with no locks held).
1416 */
1417 static inline struct extent_buffer *
1418 get_old_root(struct btrfs_root *root, u64 time_seq)
1419 {
1420 struct tree_mod_elem *tm;
1421 struct extent_buffer *eb = NULL;
1422 struct extent_buffer *eb_root;
1423 struct extent_buffer *old;
1424 struct tree_mod_root *old_root = NULL;
1425 u64 old_generation = 0;
1426 u64 logical;
1427
1428 eb_root = btrfs_read_lock_root_node(root);
1429 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1430 if (!tm)
1431 return eb_root;
1432
1433 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1434 old_root = &tm->old_root;
1435 old_generation = tm->generation;
1436 logical = old_root->logical;
1437 } else {
1438 logical = eb_root->start;
1439 }
1440
1441 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1442 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1443 btrfs_tree_read_unlock(eb_root);
1444 free_extent_buffer(eb_root);
1445 old = read_tree_block(root, logical, 0);
1446 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1447 if (!IS_ERR(old))
1448 free_extent_buffer(old);
1449 btrfs_warn(root->fs_info,
1450 "failed to read tree block %llu from get_old_root", logical);
1451 } else {
1452 eb = btrfs_clone_extent_buffer(old);
1453 free_extent_buffer(old);
1454 }
1455 } else if (old_root) {
1456 btrfs_tree_read_unlock(eb_root);
1457 free_extent_buffer(eb_root);
1458 eb = alloc_dummy_extent_buffer(root->fs_info, logical,
1459 root->nodesize);
1460 } else {
1461 btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1462 eb = btrfs_clone_extent_buffer(eb_root);
1463 btrfs_tree_read_unlock_blocking(eb_root);
1464 free_extent_buffer(eb_root);
1465 }
1466
1467 if (!eb)
1468 return NULL;
1469 extent_buffer_get(eb);
1470 btrfs_tree_read_lock(eb);
1471 if (old_root) {
1472 btrfs_set_header_bytenr(eb, eb->start);
1473 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1474 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1475 btrfs_set_header_level(eb, old_root->level);
1476 btrfs_set_header_generation(eb, old_generation);
1477 }
1478 if (tm)
1479 __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1480 else
1481 WARN_ON(btrfs_header_level(eb) != 0);
1482 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1483
1484 return eb;
1485 }
1486
1487 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1488 {
1489 struct tree_mod_elem *tm;
1490 int level;
1491 struct extent_buffer *eb_root = btrfs_root_node(root);
1492
1493 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1494 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1495 level = tm->old_root.level;
1496 } else {
1497 level = btrfs_header_level(eb_root);
1498 }
1499 free_extent_buffer(eb_root);
1500
1501 return level;
1502 }
1503
1504 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1505 struct btrfs_root *root,
1506 struct extent_buffer *buf)
1507 {
1508 if (btrfs_test_is_dummy_root(root))
1509 return 0;
1510
1511 /* ensure we can see the force_cow */
1512 smp_rmb();
1513
1514 /*
1515 * We do not need to cow a block if
1516 * 1) this block is not created or changed in this transaction;
1517 * 2) this block does not belong to TREE_RELOC tree;
1518 * 3) the root is not forced COW.
1519 *
1520 * What is forced COW:
1521 * when we create snapshot during committing the transaction,
1522 * after we've finished coping src root, we must COW the shared
1523 * block to ensure the metadata consistency.
1524 */
1525 if (btrfs_header_generation(buf) == trans->transid &&
1526 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1527 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1528 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1529 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1530 return 0;
1531 return 1;
1532 }
1533
1534 /*
1535 * cows a single block, see __btrfs_cow_block for the real work.
1536 * This version of it has extra checks so that a block isn't COWed more than
1537 * once per transaction, as long as it hasn't been written yet
1538 */
1539 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1540 struct btrfs_root *root, struct extent_buffer *buf,
1541 struct extent_buffer *parent, int parent_slot,
1542 struct extent_buffer **cow_ret)
1543 {
1544 u64 search_start;
1545 int ret;
1546
1547 if (trans->transaction != root->fs_info->running_transaction)
1548 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1549 trans->transid,
1550 root->fs_info->running_transaction->transid);
1551
1552 if (trans->transid != root->fs_info->generation)
1553 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1554 trans->transid, root->fs_info->generation);
1555
1556 if (!should_cow_block(trans, root, buf)) {
1557 trans->dirty = true;
1558 *cow_ret = buf;
1559 return 0;
1560 }
1561
1562 search_start = buf->start & ~((u64)SZ_1G - 1);
1563
1564 if (parent)
1565 btrfs_set_lock_blocking(parent);
1566 btrfs_set_lock_blocking(buf);
1567
1568 ret = __btrfs_cow_block(trans, root, buf, parent,
1569 parent_slot, cow_ret, search_start, 0);
1570
1571 trace_btrfs_cow_block(root, buf, *cow_ret);
1572
1573 return ret;
1574 }
1575
1576 /*
1577 * helper function for defrag to decide if two blocks pointed to by a
1578 * node are actually close by
1579 */
1580 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1581 {
1582 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1583 return 1;
1584 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1585 return 1;
1586 return 0;
1587 }
1588
1589 /*
1590 * compare two keys in a memcmp fashion
1591 */
1592 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1593 {
1594 struct btrfs_key k1;
1595
1596 btrfs_disk_key_to_cpu(&k1, disk);
1597
1598 return btrfs_comp_cpu_keys(&k1, k2);
1599 }
1600
1601 /*
1602 * same as comp_keys only with two btrfs_key's
1603 */
1604 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1605 {
1606 if (k1->objectid > k2->objectid)
1607 return 1;
1608 if (k1->objectid < k2->objectid)
1609 return -1;
1610 if (k1->type > k2->type)
1611 return 1;
1612 if (k1->type < k2->type)
1613 return -1;
1614 if (k1->offset > k2->offset)
1615 return 1;
1616 if (k1->offset < k2->offset)
1617 return -1;
1618 return 0;
1619 }
1620
1621 /*
1622 * this is used by the defrag code to go through all the
1623 * leaves pointed to by a node and reallocate them so that
1624 * disk order is close to key order
1625 */
1626 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1627 struct btrfs_root *root, struct extent_buffer *parent,
1628 int start_slot, u64 *last_ret,
1629 struct btrfs_key *progress)
1630 {
1631 struct extent_buffer *cur;
1632 u64 blocknr;
1633 u64 gen;
1634 u64 search_start = *last_ret;
1635 u64 last_block = 0;
1636 u64 other;
1637 u32 parent_nritems;
1638 int end_slot;
1639 int i;
1640 int err = 0;
1641 int parent_level;
1642 int uptodate;
1643 u32 blocksize;
1644 int progress_passed = 0;
1645 struct btrfs_disk_key disk_key;
1646
1647 parent_level = btrfs_header_level(parent);
1648
1649 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1650 WARN_ON(trans->transid != root->fs_info->generation);
1651
1652 parent_nritems = btrfs_header_nritems(parent);
1653 blocksize = root->nodesize;
1654 end_slot = parent_nritems - 1;
1655
1656 if (parent_nritems <= 1)
1657 return 0;
1658
1659 btrfs_set_lock_blocking(parent);
1660
1661 for (i = start_slot; i <= end_slot; i++) {
1662 int close = 1;
1663
1664 btrfs_node_key(parent, &disk_key, i);
1665 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1666 continue;
1667
1668 progress_passed = 1;
1669 blocknr = btrfs_node_blockptr(parent, i);
1670 gen = btrfs_node_ptr_generation(parent, i);
1671 if (last_block == 0)
1672 last_block = blocknr;
1673
1674 if (i > 0) {
1675 other = btrfs_node_blockptr(parent, i - 1);
1676 close = close_blocks(blocknr, other, blocksize);
1677 }
1678 if (!close && i < end_slot) {
1679 other = btrfs_node_blockptr(parent, i + 1);
1680 close = close_blocks(blocknr, other, blocksize);
1681 }
1682 if (close) {
1683 last_block = blocknr;
1684 continue;
1685 }
1686
1687 cur = btrfs_find_tree_block(root->fs_info, blocknr);
1688 if (cur)
1689 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1690 else
1691 uptodate = 0;
1692 if (!cur || !uptodate) {
1693 if (!cur) {
1694 cur = read_tree_block(root, blocknr, gen);
1695 if (IS_ERR(cur)) {
1696 return PTR_ERR(cur);
1697 } else if (!extent_buffer_uptodate(cur)) {
1698 free_extent_buffer(cur);
1699 return -EIO;
1700 }
1701 } else if (!uptodate) {
1702 err = btrfs_read_buffer(cur, gen);
1703 if (err) {
1704 free_extent_buffer(cur);
1705 return err;
1706 }
1707 }
1708 }
1709 if (search_start == 0)
1710 search_start = last_block;
1711
1712 btrfs_tree_lock(cur);
1713 btrfs_set_lock_blocking(cur);
1714 err = __btrfs_cow_block(trans, root, cur, parent, i,
1715 &cur, search_start,
1716 min(16 * blocksize,
1717 (end_slot - i) * blocksize));
1718 if (err) {
1719 btrfs_tree_unlock(cur);
1720 free_extent_buffer(cur);
1721 break;
1722 }
1723 search_start = cur->start;
1724 last_block = cur->start;
1725 *last_ret = search_start;
1726 btrfs_tree_unlock(cur);
1727 free_extent_buffer(cur);
1728 }
1729 return err;
1730 }
1731
1732 /*
1733 * The leaf data grows from end-to-front in the node.
1734 * this returns the address of the start of the last item,
1735 * which is the stop of the leaf data stack
1736 */
1737 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1738 struct extent_buffer *leaf)
1739 {
1740 u32 nr = btrfs_header_nritems(leaf);
1741 if (nr == 0)
1742 return BTRFS_LEAF_DATA_SIZE(root);
1743 return btrfs_item_offset_nr(leaf, nr - 1);
1744 }
1745
1746
1747 /*
1748 * search for key in the extent_buffer. The items start at offset p,
1749 * and they are item_size apart. There are 'max' items in p.
1750 *
1751 * the slot in the array is returned via slot, and it points to
1752 * the place where you would insert key if it is not found in
1753 * the array.
1754 *
1755 * slot may point to max if the key is bigger than all of the keys
1756 */
1757 static noinline int generic_bin_search(struct extent_buffer *eb,
1758 unsigned long p,
1759 int item_size, struct btrfs_key *key,
1760 int max, int *slot)
1761 {
1762 int low = 0;
1763 int high = max;
1764 int mid;
1765 int ret;
1766 struct btrfs_disk_key *tmp = NULL;
1767 struct btrfs_disk_key unaligned;
1768 unsigned long offset;
1769 char *kaddr = NULL;
1770 unsigned long map_start = 0;
1771 unsigned long map_len = 0;
1772 int err;
1773
1774 while (low < high) {
1775 mid = (low + high) / 2;
1776 offset = p + mid * item_size;
1777
1778 if (!kaddr || offset < map_start ||
1779 (offset + sizeof(struct btrfs_disk_key)) >
1780 map_start + map_len) {
1781
1782 err = map_private_extent_buffer(eb, offset,
1783 sizeof(struct btrfs_disk_key),
1784 &kaddr, &map_start, &map_len);
1785
1786 if (!err) {
1787 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1788 map_start);
1789 } else {
1790 read_extent_buffer(eb, &unaligned,
1791 offset, sizeof(unaligned));
1792 tmp = &unaligned;
1793 }
1794
1795 } else {
1796 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1797 map_start);
1798 }
1799 ret = comp_keys(tmp, key);
1800
1801 if (ret < 0)
1802 low = mid + 1;
1803 else if (ret > 0)
1804 high = mid;
1805 else {
1806 *slot = mid;
1807 return 0;
1808 }
1809 }
1810 *slot = low;
1811 return 1;
1812 }
1813
1814 /*
1815 * simple bin_search frontend that does the right thing for
1816 * leaves vs nodes
1817 */
1818 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1819 int level, int *slot)
1820 {
1821 if (level == 0)
1822 return generic_bin_search(eb,
1823 offsetof(struct btrfs_leaf, items),
1824 sizeof(struct btrfs_item),
1825 key, btrfs_header_nritems(eb),
1826 slot);
1827 else
1828 return generic_bin_search(eb,
1829 offsetof(struct btrfs_node, ptrs),
1830 sizeof(struct btrfs_key_ptr),
1831 key, btrfs_header_nritems(eb),
1832 slot);
1833 }
1834
1835 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1836 int level, int *slot)
1837 {
1838 return bin_search(eb, key, level, slot);
1839 }
1840
1841 static void root_add_used(struct btrfs_root *root, u32 size)
1842 {
1843 spin_lock(&root->accounting_lock);
1844 btrfs_set_root_used(&root->root_item,
1845 btrfs_root_used(&root->root_item) + size);
1846 spin_unlock(&root->accounting_lock);
1847 }
1848
1849 static void root_sub_used(struct btrfs_root *root, u32 size)
1850 {
1851 spin_lock(&root->accounting_lock);
1852 btrfs_set_root_used(&root->root_item,
1853 btrfs_root_used(&root->root_item) - size);
1854 spin_unlock(&root->accounting_lock);
1855 }
1856
1857 /* given a node and slot number, this reads the blocks it points to. The
1858 * extent buffer is returned with a reference taken (but unlocked).
1859 * NULL is returned on error.
1860 */
1861 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1862 struct extent_buffer *parent, int slot)
1863 {
1864 int level = btrfs_header_level(parent);
1865 struct extent_buffer *eb;
1866
1867 if (slot < 0)
1868 return NULL;
1869 if (slot >= btrfs_header_nritems(parent))
1870 return NULL;
1871
1872 BUG_ON(level == 0);
1873
1874 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1875 btrfs_node_ptr_generation(parent, slot));
1876 if (IS_ERR(eb) || !extent_buffer_uptodate(eb)) {
1877 if (!IS_ERR(eb))
1878 free_extent_buffer(eb);
1879 eb = NULL;
1880 }
1881
1882 return eb;
1883 }
1884
1885 /*
1886 * node level balancing, used to make sure nodes are in proper order for
1887 * item deletion. We balance from the top down, so we have to make sure
1888 * that a deletion won't leave an node completely empty later on.
1889 */
1890 static noinline int balance_level(struct btrfs_trans_handle *trans,
1891 struct btrfs_root *root,
1892 struct btrfs_path *path, int level)
1893 {
1894 struct extent_buffer *right = NULL;
1895 struct extent_buffer *mid;
1896 struct extent_buffer *left = NULL;
1897 struct extent_buffer *parent = NULL;
1898 int ret = 0;
1899 int wret;
1900 int pslot;
1901 int orig_slot = path->slots[level];
1902 u64 orig_ptr;
1903
1904 if (level == 0)
1905 return 0;
1906
1907 mid = path->nodes[level];
1908
1909 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1910 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1911 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1912
1913 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1914
1915 if (level < BTRFS_MAX_LEVEL - 1) {
1916 parent = path->nodes[level + 1];
1917 pslot = path->slots[level + 1];
1918 }
1919
1920 /*
1921 * deal with the case where there is only one pointer in the root
1922 * by promoting the node below to a root
1923 */
1924 if (!parent) {
1925 struct extent_buffer *child;
1926
1927 if (btrfs_header_nritems(mid) != 1)
1928 return 0;
1929
1930 /* promote the child to a root */
1931 child = read_node_slot(root, mid, 0);
1932 if (!child) {
1933 ret = -EROFS;
1934 btrfs_handle_fs_error(root->fs_info, ret, NULL);
1935 goto enospc;
1936 }
1937
1938 btrfs_tree_lock(child);
1939 btrfs_set_lock_blocking(child);
1940 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1941 if (ret) {
1942 btrfs_tree_unlock(child);
1943 free_extent_buffer(child);
1944 goto enospc;
1945 }
1946
1947 tree_mod_log_set_root_pointer(root, child, 1);
1948 rcu_assign_pointer(root->node, child);
1949
1950 add_root_to_dirty_list(root);
1951 btrfs_tree_unlock(child);
1952
1953 path->locks[level] = 0;
1954 path->nodes[level] = NULL;
1955 clean_tree_block(trans, root->fs_info, mid);
1956 btrfs_tree_unlock(mid);
1957 /* once for the path */
1958 free_extent_buffer(mid);
1959
1960 root_sub_used(root, mid->len);
1961 btrfs_free_tree_block(trans, root, mid, 0, 1);
1962 /* once for the root ptr */
1963 free_extent_buffer_stale(mid);
1964 return 0;
1965 }
1966 if (btrfs_header_nritems(mid) >
1967 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1968 return 0;
1969
1970 left = read_node_slot(root, parent, pslot - 1);
1971 if (left) {
1972 btrfs_tree_lock(left);
1973 btrfs_set_lock_blocking(left);
1974 wret = btrfs_cow_block(trans, root, left,
1975 parent, pslot - 1, &left);
1976 if (wret) {
1977 ret = wret;
1978 goto enospc;
1979 }
1980 }
1981 right = read_node_slot(root, parent, pslot + 1);
1982 if (right) {
1983 btrfs_tree_lock(right);
1984 btrfs_set_lock_blocking(right);
1985 wret = btrfs_cow_block(trans, root, right,
1986 parent, pslot + 1, &right);
1987 if (wret) {
1988 ret = wret;
1989 goto enospc;
1990 }
1991 }
1992
1993 /* first, try to make some room in the middle buffer */
1994 if (left) {
1995 orig_slot += btrfs_header_nritems(left);
1996 wret = push_node_left(trans, root, left, mid, 1);
1997 if (wret < 0)
1998 ret = wret;
1999 }
2000
2001 /*
2002 * then try to empty the right most buffer into the middle
2003 */
2004 if (right) {
2005 wret = push_node_left(trans, root, mid, right, 1);
2006 if (wret < 0 && wret != -ENOSPC)
2007 ret = wret;
2008 if (btrfs_header_nritems(right) == 0) {
2009 clean_tree_block(trans, root->fs_info, right);
2010 btrfs_tree_unlock(right);
2011 del_ptr(root, path, level + 1, pslot + 1);
2012 root_sub_used(root, right->len);
2013 btrfs_free_tree_block(trans, root, right, 0, 1);
2014 free_extent_buffer_stale(right);
2015 right = NULL;
2016 } else {
2017 struct btrfs_disk_key right_key;
2018 btrfs_node_key(right, &right_key, 0);
2019 tree_mod_log_set_node_key(root->fs_info, parent,
2020 pslot + 1, 0);
2021 btrfs_set_node_key(parent, &right_key, pslot + 1);
2022 btrfs_mark_buffer_dirty(parent);
2023 }
2024 }
2025 if (btrfs_header_nritems(mid) == 1) {
2026 /*
2027 * we're not allowed to leave a node with one item in the
2028 * tree during a delete. A deletion from lower in the tree
2029 * could try to delete the only pointer in this node.
2030 * So, pull some keys from the left.
2031 * There has to be a left pointer at this point because
2032 * otherwise we would have pulled some pointers from the
2033 * right
2034 */
2035 if (!left) {
2036 ret = -EROFS;
2037 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2038 goto enospc;
2039 }
2040 wret = balance_node_right(trans, root, mid, left);
2041 if (wret < 0) {
2042 ret = wret;
2043 goto enospc;
2044 }
2045 if (wret == 1) {
2046 wret = push_node_left(trans, root, left, mid, 1);
2047 if (wret < 0)
2048 ret = wret;
2049 }
2050 BUG_ON(wret == 1);
2051 }
2052 if (btrfs_header_nritems(mid) == 0) {
2053 clean_tree_block(trans, root->fs_info, mid);
2054 btrfs_tree_unlock(mid);
2055 del_ptr(root, path, level + 1, pslot);
2056 root_sub_used(root, mid->len);
2057 btrfs_free_tree_block(trans, root, mid, 0, 1);
2058 free_extent_buffer_stale(mid);
2059 mid = NULL;
2060 } else {
2061 /* update the parent key to reflect our changes */
2062 struct btrfs_disk_key mid_key;
2063 btrfs_node_key(mid, &mid_key, 0);
2064 tree_mod_log_set_node_key(root->fs_info, parent,
2065 pslot, 0);
2066 btrfs_set_node_key(parent, &mid_key, pslot);
2067 btrfs_mark_buffer_dirty(parent);
2068 }
2069
2070 /* update the path */
2071 if (left) {
2072 if (btrfs_header_nritems(left) > orig_slot) {
2073 extent_buffer_get(left);
2074 /* left was locked after cow */
2075 path->nodes[level] = left;
2076 path->slots[level + 1] -= 1;
2077 path->slots[level] = orig_slot;
2078 if (mid) {
2079 btrfs_tree_unlock(mid);
2080 free_extent_buffer(mid);
2081 }
2082 } else {
2083 orig_slot -= btrfs_header_nritems(left);
2084 path->slots[level] = orig_slot;
2085 }
2086 }
2087 /* double check we haven't messed things up */
2088 if (orig_ptr !=
2089 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2090 BUG();
2091 enospc:
2092 if (right) {
2093 btrfs_tree_unlock(right);
2094 free_extent_buffer(right);
2095 }
2096 if (left) {
2097 if (path->nodes[level] != left)
2098 btrfs_tree_unlock(left);
2099 free_extent_buffer(left);
2100 }
2101 return ret;
2102 }
2103
2104 /* Node balancing for insertion. Here we only split or push nodes around
2105 * when they are completely full. This is also done top down, so we
2106 * have to be pessimistic.
2107 */
2108 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2109 struct btrfs_root *root,
2110 struct btrfs_path *path, int level)
2111 {
2112 struct extent_buffer *right = NULL;
2113 struct extent_buffer *mid;
2114 struct extent_buffer *left = NULL;
2115 struct extent_buffer *parent = NULL;
2116 int ret = 0;
2117 int wret;
2118 int pslot;
2119 int orig_slot = path->slots[level];
2120
2121 if (level == 0)
2122 return 1;
2123
2124 mid = path->nodes[level];
2125 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2126
2127 if (level < BTRFS_MAX_LEVEL - 1) {
2128 parent = path->nodes[level + 1];
2129 pslot = path->slots[level + 1];
2130 }
2131
2132 if (!parent)
2133 return 1;
2134
2135 left = read_node_slot(root, parent, pslot - 1);
2136
2137 /* first, try to make some room in the middle buffer */
2138 if (left) {
2139 u32 left_nr;
2140
2141 btrfs_tree_lock(left);
2142 btrfs_set_lock_blocking(left);
2143
2144 left_nr = btrfs_header_nritems(left);
2145 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2146 wret = 1;
2147 } else {
2148 ret = btrfs_cow_block(trans, root, left, parent,
2149 pslot - 1, &left);
2150 if (ret)
2151 wret = 1;
2152 else {
2153 wret = push_node_left(trans, root,
2154 left, mid, 0);
2155 }
2156 }
2157 if (wret < 0)
2158 ret = wret;
2159 if (wret == 0) {
2160 struct btrfs_disk_key disk_key;
2161 orig_slot += left_nr;
2162 btrfs_node_key(mid, &disk_key, 0);
2163 tree_mod_log_set_node_key(root->fs_info, parent,
2164 pslot, 0);
2165 btrfs_set_node_key(parent, &disk_key, pslot);
2166 btrfs_mark_buffer_dirty(parent);
2167 if (btrfs_header_nritems(left) > orig_slot) {
2168 path->nodes[level] = left;
2169 path->slots[level + 1] -= 1;
2170 path->slots[level] = orig_slot;
2171 btrfs_tree_unlock(mid);
2172 free_extent_buffer(mid);
2173 } else {
2174 orig_slot -=
2175 btrfs_header_nritems(left);
2176 path->slots[level] = orig_slot;
2177 btrfs_tree_unlock(left);
2178 free_extent_buffer(left);
2179 }
2180 return 0;
2181 }
2182 btrfs_tree_unlock(left);
2183 free_extent_buffer(left);
2184 }
2185 right = read_node_slot(root, parent, pslot + 1);
2186
2187 /*
2188 * then try to empty the right most buffer into the middle
2189 */
2190 if (right) {
2191 u32 right_nr;
2192
2193 btrfs_tree_lock(right);
2194 btrfs_set_lock_blocking(right);
2195
2196 right_nr = btrfs_header_nritems(right);
2197 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2198 wret = 1;
2199 } else {
2200 ret = btrfs_cow_block(trans, root, right,
2201 parent, pslot + 1,
2202 &right);
2203 if (ret)
2204 wret = 1;
2205 else {
2206 wret = balance_node_right(trans, root,
2207 right, mid);
2208 }
2209 }
2210 if (wret < 0)
2211 ret = wret;
2212 if (wret == 0) {
2213 struct btrfs_disk_key disk_key;
2214
2215 btrfs_node_key(right, &disk_key, 0);
2216 tree_mod_log_set_node_key(root->fs_info, parent,
2217 pslot + 1, 0);
2218 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2219 btrfs_mark_buffer_dirty(parent);
2220
2221 if (btrfs_header_nritems(mid) <= orig_slot) {
2222 path->nodes[level] = right;
2223 path->slots[level + 1] += 1;
2224 path->slots[level] = orig_slot -
2225 btrfs_header_nritems(mid);
2226 btrfs_tree_unlock(mid);
2227 free_extent_buffer(mid);
2228 } else {
2229 btrfs_tree_unlock(right);
2230 free_extent_buffer(right);
2231 }
2232 return 0;
2233 }
2234 btrfs_tree_unlock(right);
2235 free_extent_buffer(right);
2236 }
2237 return 1;
2238 }
2239
2240 /*
2241 * readahead one full node of leaves, finding things that are close
2242 * to the block in 'slot', and triggering ra on them.
2243 */
2244 static void reada_for_search(struct btrfs_root *root,
2245 struct btrfs_path *path,
2246 int level, int slot, u64 objectid)
2247 {
2248 struct extent_buffer *node;
2249 struct btrfs_disk_key disk_key;
2250 u32 nritems;
2251 u64 search;
2252 u64 target;
2253 u64 nread = 0;
2254 u64 gen;
2255 struct extent_buffer *eb;
2256 u32 nr;
2257 u32 blocksize;
2258 u32 nscan = 0;
2259
2260 if (level != 1)
2261 return;
2262
2263 if (!path->nodes[level])
2264 return;
2265
2266 node = path->nodes[level];
2267
2268 search = btrfs_node_blockptr(node, slot);
2269 blocksize = root->nodesize;
2270 eb = btrfs_find_tree_block(root->fs_info, search);
2271 if (eb) {
2272 free_extent_buffer(eb);
2273 return;
2274 }
2275
2276 target = search;
2277
2278 nritems = btrfs_header_nritems(node);
2279 nr = slot;
2280
2281 while (1) {
2282 if (path->reada == READA_BACK) {
2283 if (nr == 0)
2284 break;
2285 nr--;
2286 } else if (path->reada == READA_FORWARD) {
2287 nr++;
2288 if (nr >= nritems)
2289 break;
2290 }
2291 if (path->reada == READA_BACK && objectid) {
2292 btrfs_node_key(node, &disk_key, nr);
2293 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2294 break;
2295 }
2296 search = btrfs_node_blockptr(node, nr);
2297 if ((search <= target && target - search <= 65536) ||
2298 (search > target && search - target <= 65536)) {
2299 gen = btrfs_node_ptr_generation(node, nr);
2300 readahead_tree_block(root, search);
2301 nread += blocksize;
2302 }
2303 nscan++;
2304 if ((nread > 65536 || nscan > 32))
2305 break;
2306 }
2307 }
2308
2309 static noinline void reada_for_balance(struct btrfs_root *root,
2310 struct btrfs_path *path, int level)
2311 {
2312 int slot;
2313 int nritems;
2314 struct extent_buffer *parent;
2315 struct extent_buffer *eb;
2316 u64 gen;
2317 u64 block1 = 0;
2318 u64 block2 = 0;
2319
2320 parent = path->nodes[level + 1];
2321 if (!parent)
2322 return;
2323
2324 nritems = btrfs_header_nritems(parent);
2325 slot = path->slots[level + 1];
2326
2327 if (slot > 0) {
2328 block1 = btrfs_node_blockptr(parent, slot - 1);
2329 gen = btrfs_node_ptr_generation(parent, slot - 1);
2330 eb = btrfs_find_tree_block(root->fs_info, block1);
2331 /*
2332 * if we get -eagain from btrfs_buffer_uptodate, we
2333 * don't want to return eagain here. That will loop
2334 * forever
2335 */
2336 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2337 block1 = 0;
2338 free_extent_buffer(eb);
2339 }
2340 if (slot + 1 < nritems) {
2341 block2 = btrfs_node_blockptr(parent, slot + 1);
2342 gen = btrfs_node_ptr_generation(parent, slot + 1);
2343 eb = btrfs_find_tree_block(root->fs_info, block2);
2344 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2345 block2 = 0;
2346 free_extent_buffer(eb);
2347 }
2348
2349 if (block1)
2350 readahead_tree_block(root, block1);
2351 if (block2)
2352 readahead_tree_block(root, block2);
2353 }
2354
2355
2356 /*
2357 * when we walk down the tree, it is usually safe to unlock the higher layers
2358 * in the tree. The exceptions are when our path goes through slot 0, because
2359 * operations on the tree might require changing key pointers higher up in the
2360 * tree.
2361 *
2362 * callers might also have set path->keep_locks, which tells this code to keep
2363 * the lock if the path points to the last slot in the block. This is part of
2364 * walking through the tree, and selecting the next slot in the higher block.
2365 *
2366 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2367 * if lowest_unlock is 1, level 0 won't be unlocked
2368 */
2369 static noinline void unlock_up(struct btrfs_path *path, int level,
2370 int lowest_unlock, int min_write_lock_level,
2371 int *write_lock_level)
2372 {
2373 int i;
2374 int skip_level = level;
2375 int no_skips = 0;
2376 struct extent_buffer *t;
2377
2378 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2379 if (!path->nodes[i])
2380 break;
2381 if (!path->locks[i])
2382 break;
2383 if (!no_skips && path->slots[i] == 0) {
2384 skip_level = i + 1;
2385 continue;
2386 }
2387 if (!no_skips && path->keep_locks) {
2388 u32 nritems;
2389 t = path->nodes[i];
2390 nritems = btrfs_header_nritems(t);
2391 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2392 skip_level = i + 1;
2393 continue;
2394 }
2395 }
2396 if (skip_level < i && i >= lowest_unlock)
2397 no_skips = 1;
2398
2399 t = path->nodes[i];
2400 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2401 btrfs_tree_unlock_rw(t, path->locks[i]);
2402 path->locks[i] = 0;
2403 if (write_lock_level &&
2404 i > min_write_lock_level &&
2405 i <= *write_lock_level) {
2406 *write_lock_level = i - 1;
2407 }
2408 }
2409 }
2410 }
2411
2412 /*
2413 * This releases any locks held in the path starting at level and
2414 * going all the way up to the root.
2415 *
2416 * btrfs_search_slot will keep the lock held on higher nodes in a few
2417 * corner cases, such as COW of the block at slot zero in the node. This
2418 * ignores those rules, and it should only be called when there are no
2419 * more updates to be done higher up in the tree.
2420 */
2421 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2422 {
2423 int i;
2424
2425 if (path->keep_locks)
2426 return;
2427
2428 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2429 if (!path->nodes[i])
2430 continue;
2431 if (!path->locks[i])
2432 continue;
2433 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2434 path->locks[i] = 0;
2435 }
2436 }
2437
2438 /*
2439 * helper function for btrfs_search_slot. The goal is to find a block
2440 * in cache without setting the path to blocking. If we find the block
2441 * we return zero and the path is unchanged.
2442 *
2443 * If we can't find the block, we set the path blocking and do some
2444 * reada. -EAGAIN is returned and the search must be repeated.
2445 */
2446 static int
2447 read_block_for_search(struct btrfs_trans_handle *trans,
2448 struct btrfs_root *root, struct btrfs_path *p,
2449 struct extent_buffer **eb_ret, int level, int slot,
2450 struct btrfs_key *key, u64 time_seq)
2451 {
2452 u64 blocknr;
2453 u64 gen;
2454 struct extent_buffer *b = *eb_ret;
2455 struct extent_buffer *tmp;
2456 int ret;
2457
2458 blocknr = btrfs_node_blockptr(b, slot);
2459 gen = btrfs_node_ptr_generation(b, slot);
2460
2461 tmp = btrfs_find_tree_block(root->fs_info, blocknr);
2462 if (tmp) {
2463 /* first we do an atomic uptodate check */
2464 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2465 *eb_ret = tmp;
2466 return 0;
2467 }
2468
2469 /* the pages were up to date, but we failed
2470 * the generation number check. Do a full
2471 * read for the generation number that is correct.
2472 * We must do this without dropping locks so
2473 * we can trust our generation number
2474 */
2475 btrfs_set_path_blocking(p);
2476
2477 /* now we're allowed to do a blocking uptodate check */
2478 ret = btrfs_read_buffer(tmp, gen);
2479 if (!ret) {
2480 *eb_ret = tmp;
2481 return 0;
2482 }
2483 free_extent_buffer(tmp);
2484 btrfs_release_path(p);
2485 return -EIO;
2486 }
2487
2488 /*
2489 * reduce lock contention at high levels
2490 * of the btree by dropping locks before
2491 * we read. Don't release the lock on the current
2492 * level because we need to walk this node to figure
2493 * out which blocks to read.
2494 */
2495 btrfs_unlock_up_safe(p, level + 1);
2496 btrfs_set_path_blocking(p);
2497
2498 free_extent_buffer(tmp);
2499 if (p->reada != READA_NONE)
2500 reada_for_search(root, p, level, slot, key->objectid);
2501
2502 btrfs_release_path(p);
2503
2504 ret = -EAGAIN;
2505 tmp = read_tree_block(root, blocknr, 0);
2506 if (!IS_ERR(tmp)) {
2507 /*
2508 * If the read above didn't mark this buffer up to date,
2509 * it will never end up being up to date. Set ret to EIO now
2510 * and give up so that our caller doesn't loop forever
2511 * on our EAGAINs.
2512 */
2513 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2514 ret = -EIO;
2515 free_extent_buffer(tmp);
2516 } else {
2517 ret = PTR_ERR(tmp);
2518 }
2519 return ret;
2520 }
2521
2522 /*
2523 * helper function for btrfs_search_slot. This does all of the checks
2524 * for node-level blocks and does any balancing required based on
2525 * the ins_len.
2526 *
2527 * If no extra work was required, zero is returned. If we had to
2528 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2529 * start over
2530 */
2531 static int
2532 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2533 struct btrfs_root *root, struct btrfs_path *p,
2534 struct extent_buffer *b, int level, int ins_len,
2535 int *write_lock_level)
2536 {
2537 int ret;
2538 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2539 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2540 int sret;
2541
2542 if (*write_lock_level < level + 1) {
2543 *write_lock_level = level + 1;
2544 btrfs_release_path(p);
2545 goto again;
2546 }
2547
2548 btrfs_set_path_blocking(p);
2549 reada_for_balance(root, p, level);
2550 sret = split_node(trans, root, p, level);
2551 btrfs_clear_path_blocking(p, NULL, 0);
2552
2553 BUG_ON(sret > 0);
2554 if (sret) {
2555 ret = sret;
2556 goto done;
2557 }
2558 b = p->nodes[level];
2559 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2560 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2561 int sret;
2562
2563 if (*write_lock_level < level + 1) {
2564 *write_lock_level = level + 1;
2565 btrfs_release_path(p);
2566 goto again;
2567 }
2568
2569 btrfs_set_path_blocking(p);
2570 reada_for_balance(root, p, level);
2571 sret = balance_level(trans, root, p, level);
2572 btrfs_clear_path_blocking(p, NULL, 0);
2573
2574 if (sret) {
2575 ret = sret;
2576 goto done;
2577 }
2578 b = p->nodes[level];
2579 if (!b) {
2580 btrfs_release_path(p);
2581 goto again;
2582 }
2583 BUG_ON(btrfs_header_nritems(b) == 1);
2584 }
2585 return 0;
2586
2587 again:
2588 ret = -EAGAIN;
2589 done:
2590 return ret;
2591 }
2592
2593 static void key_search_validate(struct extent_buffer *b,
2594 struct btrfs_key *key,
2595 int level)
2596 {
2597 #ifdef CONFIG_BTRFS_ASSERT
2598 struct btrfs_disk_key disk_key;
2599
2600 btrfs_cpu_key_to_disk(&disk_key, key);
2601
2602 if (level == 0)
2603 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2604 offsetof(struct btrfs_leaf, items[0].key),
2605 sizeof(disk_key)));
2606 else
2607 ASSERT(!memcmp_extent_buffer(b, &disk_key,
2608 offsetof(struct btrfs_node, ptrs[0].key),
2609 sizeof(disk_key)));
2610 #endif
2611 }
2612
2613 static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2614 int level, int *prev_cmp, int *slot)
2615 {
2616 if (*prev_cmp != 0) {
2617 *prev_cmp = bin_search(b, key, level, slot);
2618 return *prev_cmp;
2619 }
2620
2621 key_search_validate(b, key, level);
2622 *slot = 0;
2623
2624 return 0;
2625 }
2626
2627 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2628 u64 iobjectid, u64 ioff, u8 key_type,
2629 struct btrfs_key *found_key)
2630 {
2631 int ret;
2632 struct btrfs_key key;
2633 struct extent_buffer *eb;
2634
2635 ASSERT(path);
2636 ASSERT(found_key);
2637
2638 key.type = key_type;
2639 key.objectid = iobjectid;
2640 key.offset = ioff;
2641
2642 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2643 if (ret < 0)
2644 return ret;
2645
2646 eb = path->nodes[0];
2647 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2648 ret = btrfs_next_leaf(fs_root, path);
2649 if (ret)
2650 return ret;
2651 eb = path->nodes[0];
2652 }
2653
2654 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2655 if (found_key->type != key.type ||
2656 found_key->objectid != key.objectid)
2657 return 1;
2658
2659 return 0;
2660 }
2661
2662 /*
2663 * look for key in the tree. path is filled in with nodes along the way
2664 * if key is found, we return zero and you can find the item in the leaf
2665 * level of the path (level 0)
2666 *
2667 * If the key isn't found, the path points to the slot where it should
2668 * be inserted, and 1 is returned. If there are other errors during the
2669 * search a negative error number is returned.
2670 *
2671 * if ins_len > 0, nodes and leaves will be split as we walk down the
2672 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2673 * possible)
2674 */
2675 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2676 *root, struct btrfs_key *key, struct btrfs_path *p, int
2677 ins_len, int cow)
2678 {
2679 struct extent_buffer *b;
2680 int slot;
2681 int ret;
2682 int err;
2683 int level;
2684 int lowest_unlock = 1;
2685 int root_lock;
2686 /* everything at write_lock_level or lower must be write locked */
2687 int write_lock_level = 0;
2688 u8 lowest_level = 0;
2689 int min_write_lock_level;
2690 int prev_cmp;
2691
2692 lowest_level = p->lowest_level;
2693 WARN_ON(lowest_level && ins_len > 0);
2694 WARN_ON(p->nodes[0] != NULL);
2695 BUG_ON(!cow && ins_len);
2696
2697 if (ins_len < 0) {
2698 lowest_unlock = 2;
2699
2700 /* when we are removing items, we might have to go up to level
2701 * two as we update tree pointers Make sure we keep write
2702 * for those levels as well
2703 */
2704 write_lock_level = 2;
2705 } else if (ins_len > 0) {
2706 /*
2707 * for inserting items, make sure we have a write lock on
2708 * level 1 so we can update keys
2709 */
2710 write_lock_level = 1;
2711 }
2712
2713 if (!cow)
2714 write_lock_level = -1;
2715
2716 if (cow && (p->keep_locks || p->lowest_level))
2717 write_lock_level = BTRFS_MAX_LEVEL;
2718
2719 min_write_lock_level = write_lock_level;
2720
2721 again:
2722 prev_cmp = -1;
2723 /*
2724 * we try very hard to do read locks on the root
2725 */
2726 root_lock = BTRFS_READ_LOCK;
2727 level = 0;
2728 if (p->search_commit_root) {
2729 /*
2730 * the commit roots are read only
2731 * so we always do read locks
2732 */
2733 if (p->need_commit_sem)
2734 down_read(&root->fs_info->commit_root_sem);
2735 b = root->commit_root;
2736 extent_buffer_get(b);
2737 level = btrfs_header_level(b);
2738 if (p->need_commit_sem)
2739 up_read(&root->fs_info->commit_root_sem);
2740 if (!p->skip_locking)
2741 btrfs_tree_read_lock(b);
2742 } else {
2743 if (p->skip_locking) {
2744 b = btrfs_root_node(root);
2745 level = btrfs_header_level(b);
2746 } else {
2747 /* we don't know the level of the root node
2748 * until we actually have it read locked
2749 */
2750 b = btrfs_read_lock_root_node(root);
2751 level = btrfs_header_level(b);
2752 if (level <= write_lock_level) {
2753 /* whoops, must trade for write lock */
2754 btrfs_tree_read_unlock(b);
2755 free_extent_buffer(b);
2756 b = btrfs_lock_root_node(root);
2757 root_lock = BTRFS_WRITE_LOCK;
2758
2759 /* the level might have changed, check again */
2760 level = btrfs_header_level(b);
2761 }
2762 }
2763 }
2764 p->nodes[level] = b;
2765 if (!p->skip_locking)
2766 p->locks[level] = root_lock;
2767
2768 while (b) {
2769 level = btrfs_header_level(b);
2770
2771 /*
2772 * setup the path here so we can release it under lock
2773 * contention with the cow code
2774 */
2775 if (cow) {
2776 /*
2777 * if we don't really need to cow this block
2778 * then we don't want to set the path blocking,
2779 * so we test it here
2780 */
2781 if (!should_cow_block(trans, root, b)) {
2782 trans->dirty = true;
2783 goto cow_done;
2784 }
2785
2786 /*
2787 * must have write locks on this node and the
2788 * parent
2789 */
2790 if (level > write_lock_level ||
2791 (level + 1 > write_lock_level &&
2792 level + 1 < BTRFS_MAX_LEVEL &&
2793 p->nodes[level + 1])) {
2794 write_lock_level = level + 1;
2795 btrfs_release_path(p);
2796 goto again;
2797 }
2798
2799 btrfs_set_path_blocking(p);
2800 err = btrfs_cow_block(trans, root, b,
2801 p->nodes[level + 1],
2802 p->slots[level + 1], &b);
2803 if (err) {
2804 ret = err;
2805 goto done;
2806 }
2807 }
2808 cow_done:
2809 p->nodes[level] = b;
2810 btrfs_clear_path_blocking(p, NULL, 0);
2811
2812 /*
2813 * we have a lock on b and as long as we aren't changing
2814 * the tree, there is no way to for the items in b to change.
2815 * It is safe to drop the lock on our parent before we
2816 * go through the expensive btree search on b.
2817 *
2818 * If we're inserting or deleting (ins_len != 0), then we might
2819 * be changing slot zero, which may require changing the parent.
2820 * So, we can't drop the lock until after we know which slot
2821 * we're operating on.
2822 */
2823 if (!ins_len && !p->keep_locks) {
2824 int u = level + 1;
2825
2826 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2827 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2828 p->locks[u] = 0;
2829 }
2830 }
2831
2832 ret = key_search(b, key, level, &prev_cmp, &slot);
2833
2834 if (level != 0) {
2835 int dec = 0;
2836 if (ret && slot > 0) {
2837 dec = 1;
2838 slot -= 1;
2839 }
2840 p->slots[level] = slot;
2841 err = setup_nodes_for_search(trans, root, p, b, level,
2842 ins_len, &write_lock_level);
2843 if (err == -EAGAIN)
2844 goto again;
2845 if (err) {
2846 ret = err;
2847 goto done;
2848 }
2849 b = p->nodes[level];
2850 slot = p->slots[level];
2851
2852 /*
2853 * slot 0 is special, if we change the key
2854 * we have to update the parent pointer
2855 * which means we must have a write lock
2856 * on the parent
2857 */
2858 if (slot == 0 && ins_len &&
2859 write_lock_level < level + 1) {
2860 write_lock_level = level + 1;
2861 btrfs_release_path(p);
2862 goto again;
2863 }
2864
2865 unlock_up(p, level, lowest_unlock,
2866 min_write_lock_level, &write_lock_level);
2867
2868 if (level == lowest_level) {
2869 if (dec)
2870 p->slots[level]++;
2871 goto done;
2872 }
2873
2874 err = read_block_for_search(trans, root, p,
2875 &b, level, slot, key, 0);
2876 if (err == -EAGAIN)
2877 goto again;
2878 if (err) {
2879 ret = err;
2880 goto done;
2881 }
2882
2883 if (!p->skip_locking) {
2884 level = btrfs_header_level(b);
2885 if (level <= write_lock_level) {
2886 err = btrfs_try_tree_write_lock(b);
2887 if (!err) {
2888 btrfs_set_path_blocking(p);
2889 btrfs_tree_lock(b);
2890 btrfs_clear_path_blocking(p, b,
2891 BTRFS_WRITE_LOCK);
2892 }
2893 p->locks[level] = BTRFS_WRITE_LOCK;
2894 } else {
2895 err = btrfs_tree_read_lock_atomic(b);
2896 if (!err) {
2897 btrfs_set_path_blocking(p);
2898 btrfs_tree_read_lock(b);
2899 btrfs_clear_path_blocking(p, b,
2900 BTRFS_READ_LOCK);
2901 }
2902 p->locks[level] = BTRFS_READ_LOCK;
2903 }
2904 p->nodes[level] = b;
2905 }
2906 } else {
2907 p->slots[level] = slot;
2908 if (ins_len > 0 &&
2909 btrfs_leaf_free_space(root, b) < ins_len) {
2910 if (write_lock_level < 1) {
2911 write_lock_level = 1;
2912 btrfs_release_path(p);
2913 goto again;
2914 }
2915
2916 btrfs_set_path_blocking(p);
2917 err = split_leaf(trans, root, key,
2918 p, ins_len, ret == 0);
2919 btrfs_clear_path_blocking(p, NULL, 0);
2920
2921 BUG_ON(err > 0);
2922 if (err) {
2923 ret = err;
2924 goto done;
2925 }
2926 }
2927 if (!p->search_for_split)
2928 unlock_up(p, level, lowest_unlock,
2929 min_write_lock_level, &write_lock_level);
2930 goto done;
2931 }
2932 }
2933 ret = 1;
2934 done:
2935 /*
2936 * we don't really know what they plan on doing with the path
2937 * from here on, so for now just mark it as blocking
2938 */
2939 if (!p->leave_spinning)
2940 btrfs_set_path_blocking(p);
2941 if (ret < 0 && !p->skip_release_on_error)
2942 btrfs_release_path(p);
2943 return ret;
2944 }
2945
2946 /*
2947 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2948 * current state of the tree together with the operations recorded in the tree
2949 * modification log to search for the key in a previous version of this tree, as
2950 * denoted by the time_seq parameter.
2951 *
2952 * Naturally, there is no support for insert, delete or cow operations.
2953 *
2954 * The resulting path and return value will be set up as if we called
2955 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2956 */
2957 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2958 struct btrfs_path *p, u64 time_seq)
2959 {
2960 struct extent_buffer *b;
2961 int slot;
2962 int ret;
2963 int err;
2964 int level;
2965 int lowest_unlock = 1;
2966 u8 lowest_level = 0;
2967 int prev_cmp = -1;
2968
2969 lowest_level = p->lowest_level;
2970 WARN_ON(p->nodes[0] != NULL);
2971
2972 if (p->search_commit_root) {
2973 BUG_ON(time_seq);
2974 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2975 }
2976
2977 again:
2978 b = get_old_root(root, time_seq);
2979 level = btrfs_header_level(b);
2980 p->locks[level] = BTRFS_READ_LOCK;
2981
2982 while (b) {
2983 level = btrfs_header_level(b);
2984 p->nodes[level] = b;
2985 btrfs_clear_path_blocking(p, NULL, 0);
2986
2987 /*
2988 * we have a lock on b and as long as we aren't changing
2989 * the tree, there is no way to for the items in b to change.
2990 * It is safe to drop the lock on our parent before we
2991 * go through the expensive btree search on b.
2992 */
2993 btrfs_unlock_up_safe(p, level + 1);
2994
2995 /*
2996 * Since we can unwind ebs we want to do a real search every
2997 * time.
2998 */
2999 prev_cmp = -1;
3000 ret = key_search(b, key, level, &prev_cmp, &slot);
3001
3002 if (level != 0) {
3003 int dec = 0;
3004 if (ret && slot > 0) {
3005 dec = 1;
3006 slot -= 1;
3007 }
3008 p->slots[level] = slot;
3009 unlock_up(p, level, lowest_unlock, 0, NULL);
3010
3011 if (level == lowest_level) {
3012 if (dec)
3013 p->slots[level]++;
3014 goto done;
3015 }
3016
3017 err = read_block_for_search(NULL, root, p, &b, level,
3018 slot, key, time_seq);
3019 if (err == -EAGAIN)
3020 goto again;
3021 if (err) {
3022 ret = err;
3023 goto done;
3024 }
3025
3026 level = btrfs_header_level(b);
3027 err = btrfs_tree_read_lock_atomic(b);
3028 if (!err) {
3029 btrfs_set_path_blocking(p);
3030 btrfs_tree_read_lock(b);
3031 btrfs_clear_path_blocking(p, b,
3032 BTRFS_READ_LOCK);
3033 }
3034 b = tree_mod_log_rewind(root->fs_info, p, b, time_seq);
3035 if (!b) {
3036 ret = -ENOMEM;
3037 goto done;
3038 }
3039 p->locks[level] = BTRFS_READ_LOCK;
3040 p->nodes[level] = b;
3041 } else {
3042 p->slots[level] = slot;
3043 unlock_up(p, level, lowest_unlock, 0, NULL);
3044 goto done;
3045 }
3046 }
3047 ret = 1;
3048 done:
3049 if (!p->leave_spinning)
3050 btrfs_set_path_blocking(p);
3051 if (ret < 0)
3052 btrfs_release_path(p);
3053
3054 return ret;
3055 }
3056
3057 /*
3058 * helper to use instead of search slot if no exact match is needed but
3059 * instead the next or previous item should be returned.
3060 * When find_higher is true, the next higher item is returned, the next lower
3061 * otherwise.
3062 * When return_any and find_higher are both true, and no higher item is found,
3063 * return the next lower instead.
3064 * When return_any is true and find_higher is false, and no lower item is found,
3065 * return the next higher instead.
3066 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3067 * < 0 on error
3068 */
3069 int btrfs_search_slot_for_read(struct btrfs_root *root,
3070 struct btrfs_key *key, struct btrfs_path *p,
3071 int find_higher, int return_any)
3072 {
3073 int ret;
3074 struct extent_buffer *leaf;
3075
3076 again:
3077 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3078 if (ret <= 0)
3079 return ret;
3080 /*
3081 * a return value of 1 means the path is at the position where the
3082 * item should be inserted. Normally this is the next bigger item,
3083 * but in case the previous item is the last in a leaf, path points
3084 * to the first free slot in the previous leaf, i.e. at an invalid
3085 * item.
3086 */
3087 leaf = p->nodes[0];
3088
3089 if (find_higher) {
3090 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3091 ret = btrfs_next_leaf(root, p);
3092 if (ret <= 0)
3093 return ret;
3094 if (!return_any)
3095 return 1;
3096 /*
3097 * no higher item found, return the next
3098 * lower instead
3099 */
3100 return_any = 0;
3101 find_higher = 0;
3102 btrfs_release_path(p);
3103 goto again;
3104 }
3105 } else {
3106 if (p->slots[0] == 0) {
3107 ret = btrfs_prev_leaf(root, p);
3108 if (ret < 0)
3109 return ret;
3110 if (!ret) {
3111 leaf = p->nodes[0];
3112 if (p->slots[0] == btrfs_header_nritems(leaf))
3113 p->slots[0]--;
3114 return 0;
3115 }
3116 if (!return_any)
3117 return 1;
3118 /*
3119 * no lower item found, return the next
3120 * higher instead
3121 */
3122 return_any = 0;
3123 find_higher = 1;
3124 btrfs_release_path(p);
3125 goto again;
3126 } else {
3127 --p->slots[0];
3128 }
3129 }
3130 return 0;
3131 }
3132
3133 /*
3134 * adjust the pointers going up the tree, starting at level
3135 * making sure the right key of each node is points to 'key'.
3136 * This is used after shifting pointers to the left, so it stops
3137 * fixing up pointers when a given leaf/node is not in slot 0 of the
3138 * higher levels
3139 *
3140 */
3141 static void fixup_low_keys(struct btrfs_fs_info *fs_info,
3142 struct btrfs_path *path,
3143 struct btrfs_disk_key *key, int level)
3144 {
3145 int i;
3146 struct extent_buffer *t;
3147
3148 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3149 int tslot = path->slots[i];
3150 if (!path->nodes[i])
3151 break;
3152 t = path->nodes[i];
3153 tree_mod_log_set_node_key(fs_info, t, tslot, 1);
3154 btrfs_set_node_key(t, key, tslot);
3155 btrfs_mark_buffer_dirty(path->nodes[i]);
3156 if (tslot != 0)
3157 break;
3158 }
3159 }
3160
3161 /*
3162 * update item key.
3163 *
3164 * This function isn't completely safe. It's the caller's responsibility
3165 * that the new key won't break the order
3166 */
3167 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3168 struct btrfs_path *path,
3169 struct btrfs_key *new_key)
3170 {
3171 struct btrfs_disk_key disk_key;
3172 struct extent_buffer *eb;
3173 int slot;
3174
3175 eb = path->nodes[0];
3176 slot = path->slots[0];
3177 if (slot > 0) {
3178 btrfs_item_key(eb, &disk_key, slot - 1);
3179 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3180 }
3181 if (slot < btrfs_header_nritems(eb) - 1) {
3182 btrfs_item_key(eb, &disk_key, slot + 1);
3183 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3184 }
3185
3186 btrfs_cpu_key_to_disk(&disk_key, new_key);
3187 btrfs_set_item_key(eb, &disk_key, slot);
3188 btrfs_mark_buffer_dirty(eb);
3189 if (slot == 0)
3190 fixup_low_keys(fs_info, path, &disk_key, 1);
3191 }
3192
3193 /*
3194 * try to push data from one node into the next node left in the
3195 * tree.
3196 *
3197 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3198 * error, and > 0 if there was no room in the left hand block.
3199 */
3200 static int push_node_left(struct btrfs_trans_handle *trans,
3201 struct btrfs_root *root, struct extent_buffer *dst,
3202 struct extent_buffer *src, int empty)
3203 {
3204 int push_items = 0;
3205 int src_nritems;
3206 int dst_nritems;
3207 int ret = 0;
3208
3209 src_nritems = btrfs_header_nritems(src);
3210 dst_nritems = btrfs_header_nritems(dst);
3211 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3212 WARN_ON(btrfs_header_generation(src) != trans->transid);
3213 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3214
3215 if (!empty && src_nritems <= 8)
3216 return 1;
3217
3218 if (push_items <= 0)
3219 return 1;
3220
3221 if (empty) {
3222 push_items = min(src_nritems, push_items);
3223 if (push_items < src_nritems) {
3224 /* leave at least 8 pointers in the node if
3225 * we aren't going to empty it
3226 */
3227 if (src_nritems - push_items < 8) {
3228 if (push_items <= 8)
3229 return 1;
3230 push_items -= 8;
3231 }
3232 }
3233 } else
3234 push_items = min(src_nritems - 8, push_items);
3235
3236 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3237 push_items);
3238 if (ret) {
3239 btrfs_abort_transaction(trans, root, ret);
3240 return ret;
3241 }
3242 copy_extent_buffer(dst, src,
3243 btrfs_node_key_ptr_offset(dst_nritems),
3244 btrfs_node_key_ptr_offset(0),
3245 push_items * sizeof(struct btrfs_key_ptr));
3246
3247 if (push_items < src_nritems) {
3248 /*
3249 * don't call tree_mod_log_eb_move here, key removal was already
3250 * fully logged by tree_mod_log_eb_copy above.
3251 */
3252 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3253 btrfs_node_key_ptr_offset(push_items),
3254 (src_nritems - push_items) *
3255 sizeof(struct btrfs_key_ptr));
3256 }
3257 btrfs_set_header_nritems(src, src_nritems - push_items);
3258 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3259 btrfs_mark_buffer_dirty(src);
3260 btrfs_mark_buffer_dirty(dst);
3261
3262 return ret;
3263 }
3264
3265 /*
3266 * try to push data from one node into the next node right in the
3267 * tree.
3268 *
3269 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3270 * error, and > 0 if there was no room in the right hand block.
3271 *
3272 * this will only push up to 1/2 the contents of the left node over
3273 */
3274 static int balance_node_right(struct btrfs_trans_handle *trans,
3275 struct btrfs_root *root,
3276 struct extent_buffer *dst,
3277 struct extent_buffer *src)
3278 {
3279 int push_items = 0;
3280 int max_push;
3281 int src_nritems;
3282 int dst_nritems;
3283 int ret = 0;
3284
3285 WARN_ON(btrfs_header_generation(src) != trans->transid);
3286 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3287
3288 src_nritems = btrfs_header_nritems(src);
3289 dst_nritems = btrfs_header_nritems(dst);
3290 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3291 if (push_items <= 0)
3292 return 1;
3293
3294 if (src_nritems < 4)
3295 return 1;
3296
3297 max_push = src_nritems / 2 + 1;
3298 /* don't try to empty the node */
3299 if (max_push >= src_nritems)
3300 return 1;
3301
3302 if (max_push < push_items)
3303 push_items = max_push;
3304
3305 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3306 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3307 btrfs_node_key_ptr_offset(0),
3308 (dst_nritems) *
3309 sizeof(struct btrfs_key_ptr));
3310
3311 ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3312 src_nritems - push_items, push_items);
3313 if (ret) {
3314 btrfs_abort_transaction(trans, root, ret);
3315 return ret;
3316 }
3317 copy_extent_buffer(dst, src,
3318 btrfs_node_key_ptr_offset(0),
3319 btrfs_node_key_ptr_offset(src_nritems - push_items),
3320 push_items * sizeof(struct btrfs_key_ptr));
3321
3322 btrfs_set_header_nritems(src, src_nritems - push_items);
3323 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3324
3325 btrfs_mark_buffer_dirty(src);
3326 btrfs_mark_buffer_dirty(dst);
3327
3328 return ret;
3329 }
3330
3331 /*
3332 * helper function to insert a new root level in the tree.
3333 * A new node is allocated, and a single item is inserted to
3334 * point to the existing root
3335 *
3336 * returns zero on success or < 0 on failure.
3337 */
3338 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3339 struct btrfs_root *root,
3340 struct btrfs_path *path, int level)
3341 {
3342 u64 lower_gen;
3343 struct extent_buffer *lower;
3344 struct extent_buffer *c;
3345 struct extent_buffer *old;
3346 struct btrfs_disk_key lower_key;
3347
3348 BUG_ON(path->nodes[level]);
3349 BUG_ON(path->nodes[level-1] != root->node);
3350
3351 lower = path->nodes[level-1];
3352 if (level == 1)
3353 btrfs_item_key(lower, &lower_key, 0);
3354 else
3355 btrfs_node_key(lower, &lower_key, 0);
3356
3357 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3358 &lower_key, level, root->node->start, 0);
3359 if (IS_ERR(c))
3360 return PTR_ERR(c);
3361
3362 root_add_used(root, root->nodesize);
3363
3364 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3365 btrfs_set_header_nritems(c, 1);
3366 btrfs_set_header_level(c, level);
3367 btrfs_set_header_bytenr(c, c->start);
3368 btrfs_set_header_generation(c, trans->transid);
3369 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3370 btrfs_set_header_owner(c, root->root_key.objectid);
3371
3372 write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(),
3373 BTRFS_FSID_SIZE);
3374
3375 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3376 btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE);
3377
3378 btrfs_set_node_key(c, &lower_key, 0);
3379 btrfs_set_node_blockptr(c, 0, lower->start);
3380 lower_gen = btrfs_header_generation(lower);
3381 WARN_ON(lower_gen != trans->transid);
3382
3383 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3384
3385 btrfs_mark_buffer_dirty(c);
3386
3387 old = root->node;
3388 tree_mod_log_set_root_pointer(root, c, 0);
3389 rcu_assign_pointer(root->node, c);
3390
3391 /* the super has an extra ref to root->node */
3392 free_extent_buffer(old);
3393
3394 add_root_to_dirty_list(root);
3395 extent_buffer_get(c);
3396 path->nodes[level] = c;
3397 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3398 path->slots[level] = 0;
3399 return 0;
3400 }
3401
3402 /*
3403 * worker function to insert a single pointer in a node.
3404 * the node should have enough room for the pointer already
3405 *
3406 * slot and level indicate where you want the key to go, and
3407 * blocknr is the block the key points to.
3408 */
3409 static void insert_ptr(struct btrfs_trans_handle *trans,
3410 struct btrfs_root *root, struct btrfs_path *path,
3411 struct btrfs_disk_key *key, u64 bytenr,
3412 int slot, int level)
3413 {
3414 struct extent_buffer *lower;
3415 int nritems;
3416 int ret;
3417
3418 BUG_ON(!path->nodes[level]);
3419 btrfs_assert_tree_locked(path->nodes[level]);
3420 lower = path->nodes[level];
3421 nritems = btrfs_header_nritems(lower);
3422 BUG_ON(slot > nritems);
3423 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3424 if (slot != nritems) {
3425 if (level)
3426 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3427 slot, nritems - slot);
3428 memmove_extent_buffer(lower,
3429 btrfs_node_key_ptr_offset(slot + 1),
3430 btrfs_node_key_ptr_offset(slot),
3431 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3432 }
3433 if (level) {
3434 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3435 MOD_LOG_KEY_ADD, GFP_NOFS);
3436 BUG_ON(ret < 0);
3437 }
3438 btrfs_set_node_key(lower, key, slot);
3439 btrfs_set_node_blockptr(lower, slot, bytenr);
3440 WARN_ON(trans->transid == 0);
3441 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3442 btrfs_set_header_nritems(lower, nritems + 1);
3443 btrfs_mark_buffer_dirty(lower);
3444 }
3445
3446 /*
3447 * split the node at the specified level in path in two.
3448 * The path is corrected to point to the appropriate node after the split
3449 *
3450 * Before splitting this tries to make some room in the node by pushing
3451 * left and right, if either one works, it returns right away.
3452 *
3453 * returns 0 on success and < 0 on failure
3454 */
3455 static noinline int split_node(struct btrfs_trans_handle *trans,
3456 struct btrfs_root *root,
3457 struct btrfs_path *path, int level)
3458 {
3459 struct extent_buffer *c;
3460 struct extent_buffer *split;
3461 struct btrfs_disk_key disk_key;
3462 int mid;
3463 int ret;
3464 u32 c_nritems;
3465
3466 c = path->nodes[level];
3467 WARN_ON(btrfs_header_generation(c) != trans->transid);
3468 if (c == root->node) {
3469 /*
3470 * trying to split the root, lets make a new one
3471 *
3472 * tree mod log: We don't log_removal old root in
3473 * insert_new_root, because that root buffer will be kept as a
3474 * normal node. We are going to log removal of half of the
3475 * elements below with tree_mod_log_eb_copy. We're holding a
3476 * tree lock on the buffer, which is why we cannot race with
3477 * other tree_mod_log users.
3478 */
3479 ret = insert_new_root(trans, root, path, level + 1);
3480 if (ret)
3481 return ret;
3482 } else {
3483 ret = push_nodes_for_insert(trans, root, path, level);
3484 c = path->nodes[level];
3485 if (!ret && btrfs_header_nritems(c) <
3486 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3487 return 0;
3488 if (ret < 0)
3489 return ret;
3490 }
3491
3492 c_nritems = btrfs_header_nritems(c);
3493 mid = (c_nritems + 1) / 2;
3494 btrfs_node_key(c, &disk_key, mid);
3495
3496 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3497 &disk_key, level, c->start, 0);
3498 if (IS_ERR(split))
3499 return PTR_ERR(split);
3500
3501 root_add_used(root, root->nodesize);
3502
3503 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3504 btrfs_set_header_level(split, btrfs_header_level(c));
3505 btrfs_set_header_bytenr(split, split->start);
3506 btrfs_set_header_generation(split, trans->transid);
3507 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3508 btrfs_set_header_owner(split, root->root_key.objectid);
3509 write_extent_buffer(split, root->fs_info->fsid,
3510 btrfs_header_fsid(), BTRFS_FSID_SIZE);
3511 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3512 btrfs_header_chunk_tree_uuid(split),
3513 BTRFS_UUID_SIZE);
3514
3515 ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0,
3516 mid, c_nritems - mid);
3517 if (ret) {
3518 btrfs_abort_transaction(trans, root, ret);
3519 return ret;
3520 }
3521 copy_extent_buffer(split, c,
3522 btrfs_node_key_ptr_offset(0),
3523 btrfs_node_key_ptr_offset(mid),
3524 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3525 btrfs_set_header_nritems(split, c_nritems - mid);
3526 btrfs_set_header_nritems(c, mid);
3527 ret = 0;
3528
3529 btrfs_mark_buffer_dirty(c);
3530 btrfs_mark_buffer_dirty(split);
3531
3532 insert_ptr(trans, root, path, &disk_key, split->start,
3533 path->slots[level + 1] + 1, level + 1);
3534
3535 if (path->slots[level] >= mid) {
3536 path->slots[level] -= mid;
3537 btrfs_tree_unlock(c);
3538 free_extent_buffer(c);
3539 path->nodes[level] = split;
3540 path->slots[level + 1] += 1;
3541 } else {
3542 btrfs_tree_unlock(split);
3543 free_extent_buffer(split);
3544 }
3545 return ret;
3546 }
3547
3548 /*
3549 * how many bytes are required to store the items in a leaf. start
3550 * and nr indicate which items in the leaf to check. This totals up the
3551 * space used both by the item structs and the item data
3552 */
3553 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3554 {
3555 struct btrfs_item *start_item;
3556 struct btrfs_item *end_item;
3557 struct btrfs_map_token token;
3558 int data_len;
3559 int nritems = btrfs_header_nritems(l);
3560 int end = min(nritems, start + nr) - 1;
3561
3562 if (!nr)
3563 return 0;
3564 btrfs_init_map_token(&token);
3565 start_item = btrfs_item_nr(start);
3566 end_item = btrfs_item_nr(end);
3567 data_len = btrfs_token_item_offset(l, start_item, &token) +
3568 btrfs_token_item_size(l, start_item, &token);
3569 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3570 data_len += sizeof(struct btrfs_item) * nr;
3571 WARN_ON(data_len < 0);
3572 return data_len;
3573 }
3574
3575 /*
3576 * The space between the end of the leaf items and
3577 * the start of the leaf data. IOW, how much room
3578 * the leaf has left for both items and data
3579 */
3580 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3581 struct extent_buffer *leaf)
3582 {
3583 int nritems = btrfs_header_nritems(leaf);
3584 int ret;
3585 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3586 if (ret < 0) {
3587 btrfs_crit(root->fs_info,
3588 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3589 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3590 leaf_space_used(leaf, 0, nritems), nritems);
3591 }
3592 return ret;
3593 }
3594
3595 /*
3596 * min slot controls the lowest index we're willing to push to the
3597 * right. We'll push up to and including min_slot, but no lower
3598 */
3599 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3600 struct btrfs_root *root,
3601 struct btrfs_path *path,
3602 int data_size, int empty,
3603 struct extent_buffer *right,
3604 int free_space, u32 left_nritems,
3605 u32 min_slot)
3606 {
3607 struct extent_buffer *left = path->nodes[0];
3608 struct extent_buffer *upper = path->nodes[1];
3609 struct btrfs_map_token token;
3610 struct btrfs_disk_key disk_key;
3611 int slot;
3612 u32 i;
3613 int push_space = 0;
3614 int push_items = 0;
3615 struct btrfs_item *item;
3616 u32 nr;
3617 u32 right_nritems;
3618 u32 data_end;
3619 u32 this_item_size;
3620
3621 btrfs_init_map_token(&token);
3622
3623 if (empty)
3624 nr = 0;
3625 else
3626 nr = max_t(u32, 1, min_slot);
3627
3628 if (path->slots[0] >= left_nritems)
3629 push_space += data_size;
3630
3631 slot = path->slots[1];
3632 i = left_nritems - 1;
3633 while (i >= nr) {
3634 item = btrfs_item_nr(i);
3635
3636 if (!empty && push_items > 0) {
3637 if (path->slots[0] > i)
3638 break;
3639 if (path->slots[0] == i) {
3640 int space = btrfs_leaf_free_space(root, left);
3641 if (space + push_space * 2 > free_space)
3642 break;
3643 }
3644 }
3645
3646 if (path->slots[0] == i)
3647 push_space += data_size;
3648
3649 this_item_size = btrfs_item_size(left, item);
3650 if (this_item_size + sizeof(*item) + push_space > free_space)
3651 break;
3652
3653 push_items++;
3654 push_space += this_item_size + sizeof(*item);
3655 if (i == 0)
3656 break;
3657 i--;
3658 }
3659
3660 if (push_items == 0)
3661 goto out_unlock;
3662
3663 WARN_ON(!empty && push_items == left_nritems);
3664
3665 /* push left to right */
3666 right_nritems = btrfs_header_nritems(right);
3667
3668 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3669 push_space -= leaf_data_end(root, left);
3670
3671 /* make room in the right data area */
3672 data_end = leaf_data_end(root, right);
3673 memmove_extent_buffer(right,
3674 btrfs_leaf_data(right) + data_end - push_space,
3675 btrfs_leaf_data(right) + data_end,
3676 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3677
3678 /* copy from the left data area */
3679 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3680 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3681 btrfs_leaf_data(left) + leaf_data_end(root, left),
3682 push_space);
3683
3684 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3685 btrfs_item_nr_offset(0),
3686 right_nritems * sizeof(struct btrfs_item));
3687
3688 /* copy the items from left to right */
3689 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3690 btrfs_item_nr_offset(left_nritems - push_items),
3691 push_items * sizeof(struct btrfs_item));
3692
3693 /* update the item pointers */
3694 right_nritems += push_items;
3695 btrfs_set_header_nritems(right, right_nritems);
3696 push_space = BTRFS_LEAF_DATA_SIZE(root);
3697 for (i = 0; i < right_nritems; i++) {
3698 item = btrfs_item_nr(i);
3699 push_space -= btrfs_token_item_size(right, item, &token);
3700 btrfs_set_token_item_offset(right, item, push_space, &token);
3701 }
3702
3703 left_nritems -= push_items;
3704 btrfs_set_header_nritems(left, left_nritems);
3705
3706 if (left_nritems)
3707 btrfs_mark_buffer_dirty(left);
3708 else
3709 clean_tree_block(trans, root->fs_info, left);
3710
3711 btrfs_mark_buffer_dirty(right);
3712
3713 btrfs_item_key(right, &disk_key, 0);
3714 btrfs_set_node_key(upper, &disk_key, slot + 1);
3715 btrfs_mark_buffer_dirty(upper);
3716
3717 /* then fixup the leaf pointer in the path */
3718 if (path->slots[0] >= left_nritems) {
3719 path->slots[0] -= left_nritems;
3720 if (btrfs_header_nritems(path->nodes[0]) == 0)
3721 clean_tree_block(trans, root->fs_info, path->nodes[0]);
3722 btrfs_tree_unlock(path->nodes[0]);
3723 free_extent_buffer(path->nodes[0]);
3724 path->nodes[0] = right;
3725 path->slots[1] += 1;
3726 } else {
3727 btrfs_tree_unlock(right);
3728 free_extent_buffer(right);
3729 }
3730 return 0;
3731
3732 out_unlock:
3733 btrfs_tree_unlock(right);
3734 free_extent_buffer(right);
3735 return 1;
3736 }
3737
3738 /*
3739 * push some data in the path leaf to the right, trying to free up at
3740 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3741 *
3742 * returns 1 if the push failed because the other node didn't have enough
3743 * room, 0 if everything worked out and < 0 if there were major errors.
3744 *
3745 * this will push starting from min_slot to the end of the leaf. It won't
3746 * push any slot lower than min_slot
3747 */
3748 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3749 *root, struct btrfs_path *path,
3750 int min_data_size, int data_size,
3751 int empty, u32 min_slot)
3752 {
3753 struct extent_buffer *left = path->nodes[0];
3754 struct extent_buffer *right;
3755 struct extent_buffer *upper;
3756 int slot;
3757 int free_space;
3758 u32 left_nritems;
3759 int ret;
3760
3761 if (!path->nodes[1])
3762 return 1;
3763
3764 slot = path->slots[1];
3765 upper = path->nodes[1];
3766 if (slot >= btrfs_header_nritems(upper) - 1)
3767 return 1;
3768
3769 btrfs_assert_tree_locked(path->nodes[1]);
3770
3771 right = read_node_slot(root, upper, slot + 1);
3772 if (right == NULL)
3773 return 1;
3774
3775 btrfs_tree_lock(right);
3776 btrfs_set_lock_blocking(right);
3777
3778 free_space = btrfs_leaf_free_space(root, right);
3779 if (free_space < data_size)
3780 goto out_unlock;
3781
3782 /* cow and double check */
3783 ret = btrfs_cow_block(trans, root, right, upper,
3784 slot + 1, &right);
3785 if (ret)
3786 goto out_unlock;
3787
3788 free_space = btrfs_leaf_free_space(root, right);
3789 if (free_space < data_size)
3790 goto out_unlock;
3791
3792 left_nritems = btrfs_header_nritems(left);
3793 if (left_nritems == 0)
3794 goto out_unlock;
3795
3796 if (path->slots[0] == left_nritems && !empty) {
3797 /* Key greater than all keys in the leaf, right neighbor has
3798 * enough room for it and we're not emptying our leaf to delete
3799 * it, therefore use right neighbor to insert the new item and
3800 * no need to touch/dirty our left leaft. */
3801 btrfs_tree_unlock(left);
3802 free_extent_buffer(left);
3803 path->nodes[0] = right;
3804 path->slots[0] = 0;
3805 path->slots[1]++;
3806 return 0;
3807 }
3808
3809 return __push_leaf_right(trans, root, path, min_data_size, empty,
3810 right, free_space, left_nritems, min_slot);
3811 out_unlock:
3812 btrfs_tree_unlock(right);
3813 free_extent_buffer(right);
3814 return 1;
3815 }
3816
3817 /*
3818 * push some data in the path leaf to the left, trying to free up at
3819 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3820 *
3821 * max_slot can put a limit on how far into the leaf we'll push items. The
3822 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3823 * items
3824 */
3825 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3826 struct btrfs_root *root,
3827 struct btrfs_path *path, int data_size,
3828 int empty, struct extent_buffer *left,
3829 int free_space, u32 right_nritems,
3830 u32 max_slot)
3831 {
3832 struct btrfs_disk_key disk_key;
3833 struct extent_buffer *right = path->nodes[0];
3834 int i;
3835 int push_space = 0;
3836 int push_items = 0;
3837 struct btrfs_item *item;
3838 u32 old_left_nritems;
3839 u32 nr;
3840 int ret = 0;
3841 u32 this_item_size;
3842 u32 old_left_item_size;
3843 struct btrfs_map_token token;
3844
3845 btrfs_init_map_token(&token);
3846
3847 if (empty)
3848 nr = min(right_nritems, max_slot);
3849 else
3850 nr = min(right_nritems - 1, max_slot);
3851
3852 for (i = 0; i < nr; i++) {
3853 item = btrfs_item_nr(i);
3854
3855 if (!empty && push_items > 0) {
3856 if (path->slots[0] < i)
3857 break;
3858 if (path->slots[0] == i) {
3859 int space = btrfs_leaf_free_space(root, right);
3860 if (space + push_space * 2 > free_space)
3861 break;
3862 }
3863 }
3864
3865 if (path->slots[0] == i)
3866 push_space += data_size;
3867
3868 this_item_size = btrfs_item_size(right, item);
3869 if (this_item_size + sizeof(*item) + push_space > free_space)
3870 break;
3871
3872 push_items++;
3873 push_space += this_item_size + sizeof(*item);
3874 }
3875
3876 if (push_items == 0) {
3877 ret = 1;
3878 goto out;
3879 }
3880 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3881
3882 /* push data from right to left */
3883 copy_extent_buffer(left, right,
3884 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3885 btrfs_item_nr_offset(0),
3886 push_items * sizeof(struct btrfs_item));
3887
3888 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3889 btrfs_item_offset_nr(right, push_items - 1);
3890
3891 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3892 leaf_data_end(root, left) - push_space,
3893 btrfs_leaf_data(right) +
3894 btrfs_item_offset_nr(right, push_items - 1),
3895 push_space);
3896 old_left_nritems = btrfs_header_nritems(left);
3897 BUG_ON(old_left_nritems <= 0);
3898
3899 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3900 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3901 u32 ioff;
3902
3903 item = btrfs_item_nr(i);
3904
3905 ioff = btrfs_token_item_offset(left, item, &token);
3906 btrfs_set_token_item_offset(left, item,
3907 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3908 &token);
3909 }
3910 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3911
3912 /* fixup right node */
3913 if (push_items > right_nritems)
3914 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3915 right_nritems);
3916
3917 if (push_items < right_nritems) {
3918 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3919 leaf_data_end(root, right);
3920 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3921 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3922 btrfs_leaf_data(right) +
3923 leaf_data_end(root, right), push_space);
3924
3925 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3926 btrfs_item_nr_offset(push_items),
3927 (btrfs_header_nritems(right) - push_items) *
3928 sizeof(struct btrfs_item));
3929 }
3930 right_nritems -= push_items;
3931 btrfs_set_header_nritems(right, right_nritems);
3932 push_space = BTRFS_LEAF_DATA_SIZE(root);
3933 for (i = 0; i < right_nritems; i++) {
3934 item = btrfs_item_nr(i);
3935
3936 push_space = push_space - btrfs_token_item_size(right,
3937 item, &token);
3938 btrfs_set_token_item_offset(right, item, push_space, &token);
3939 }
3940
3941 btrfs_mark_buffer_dirty(left);
3942 if (right_nritems)
3943 btrfs_mark_buffer_dirty(right);
3944 else
3945 clean_tree_block(trans, root->fs_info, right);
3946
3947 btrfs_item_key(right, &disk_key, 0);
3948 fixup_low_keys(root->fs_info, path, &disk_key, 1);
3949
3950 /* then fixup the leaf pointer in the path */
3951 if (path->slots[0] < push_items) {
3952 path->slots[0] += old_left_nritems;
3953 btrfs_tree_unlock(path->nodes[0]);
3954 free_extent_buffer(path->nodes[0]);
3955 path->nodes[0] = left;
3956 path->slots[1] -= 1;
3957 } else {
3958 btrfs_tree_unlock(left);
3959 free_extent_buffer(left);
3960 path->slots[0] -= push_items;
3961 }
3962 BUG_ON(path->slots[0] < 0);
3963 return ret;
3964 out:
3965 btrfs_tree_unlock(left);
3966 free_extent_buffer(left);
3967 return ret;
3968 }
3969
3970 /*
3971 * push some data in the path leaf to the left, trying to free up at
3972 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3973 *
3974 * max_slot can put a limit on how far into the leaf we'll push items. The
3975 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3976 * items
3977 */
3978 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3979 *root, struct btrfs_path *path, int min_data_size,
3980 int data_size, int empty, u32 max_slot)
3981 {
3982 struct extent_buffer *right = path->nodes[0];
3983 struct extent_buffer *left;
3984 int slot;
3985 int free_space;
3986 u32 right_nritems;
3987 int ret = 0;
3988
3989 slot = path->slots[1];
3990 if (slot == 0)
3991 return 1;
3992 if (!path->nodes[1])
3993 return 1;
3994
3995 right_nritems = btrfs_header_nritems(right);
3996 if (right_nritems == 0)
3997 return 1;
3998
3999 btrfs_assert_tree_locked(path->nodes[1]);
4000
4001 left = read_node_slot(root, path->nodes[1], slot - 1);
4002 if (left == NULL)
4003 return 1;
4004
4005 btrfs_tree_lock(left);
4006 btrfs_set_lock_blocking(left);
4007
4008 free_space = btrfs_leaf_free_space(root, left);
4009 if (free_space < data_size) {
4010 ret = 1;
4011 goto out;
4012 }
4013
4014 /* cow and double check */
4015 ret = btrfs_cow_block(trans, root, left,
4016 path->nodes[1], slot - 1, &left);
4017 if (ret) {
4018 /* we hit -ENOSPC, but it isn't fatal here */
4019 if (ret == -ENOSPC)
4020 ret = 1;
4021 goto out;
4022 }
4023
4024 free_space = btrfs_leaf_free_space(root, left);
4025 if (free_space < data_size) {
4026 ret = 1;
4027 goto out;
4028 }
4029
4030 return __push_leaf_left(trans, root, path, min_data_size,
4031 empty, left, free_space, right_nritems,
4032 max_slot);
4033 out:
4034 btrfs_tree_unlock(left);
4035 free_extent_buffer(left);
4036 return ret;
4037 }
4038
4039 /*
4040 * split the path's leaf in two, making sure there is at least data_size
4041 * available for the resulting leaf level of the path.
4042 */
4043 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4044 struct btrfs_root *root,
4045 struct btrfs_path *path,
4046 struct extent_buffer *l,
4047 struct extent_buffer *right,
4048 int slot, int mid, int nritems)
4049 {
4050 int data_copy_size;
4051 int rt_data_off;
4052 int i;
4053 struct btrfs_disk_key disk_key;
4054 struct btrfs_map_token token;
4055
4056 btrfs_init_map_token(&token);
4057
4058 nritems = nritems - mid;
4059 btrfs_set_header_nritems(right, nritems);
4060 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
4061
4062 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4063 btrfs_item_nr_offset(mid),
4064 nritems * sizeof(struct btrfs_item));
4065
4066 copy_extent_buffer(right, l,
4067 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
4068 data_copy_size, btrfs_leaf_data(l) +
4069 leaf_data_end(root, l), data_copy_size);
4070
4071 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
4072 btrfs_item_end_nr(l, mid);
4073
4074 for (i = 0; i < nritems; i++) {
4075 struct btrfs_item *item = btrfs_item_nr(i);
4076 u32 ioff;
4077
4078 ioff = btrfs_token_item_offset(right, item, &token);
4079 btrfs_set_token_item_offset(right, item,
4080 ioff + rt_data_off, &token);
4081 }
4082
4083 btrfs_set_header_nritems(l, mid);
4084 btrfs_item_key(right, &disk_key, 0);
4085 insert_ptr(trans, root, path, &disk_key, right->start,
4086 path->slots[1] + 1, 1);
4087
4088 btrfs_mark_buffer_dirty(right);
4089 btrfs_mark_buffer_dirty(l);
4090 BUG_ON(path->slots[0] != slot);
4091
4092 if (mid <= slot) {
4093 btrfs_tree_unlock(path->nodes[0]);
4094 free_extent_buffer(path->nodes[0]);
4095 path->nodes[0] = right;
4096 path->slots[0] -= mid;
4097 path->slots[1] += 1;
4098 } else {
4099 btrfs_tree_unlock(right);
4100 free_extent_buffer(right);
4101 }
4102
4103 BUG_ON(path->slots[0] < 0);
4104 }
4105
4106 /*
4107 * double splits happen when we need to insert a big item in the middle
4108 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4109 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4110 * A B C
4111 *
4112 * We avoid this by trying to push the items on either side of our target
4113 * into the adjacent leaves. If all goes well we can avoid the double split
4114 * completely.
4115 */
4116 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4117 struct btrfs_root *root,
4118 struct btrfs_path *path,
4119 int data_size)
4120 {
4121 int ret;
4122 int progress = 0;
4123 int slot;
4124 u32 nritems;
4125 int space_needed = data_size;
4126
4127 slot = path->slots[0];
4128 if (slot < btrfs_header_nritems(path->nodes[0]))
4129 space_needed -= btrfs_leaf_free_space(root, path->nodes[0]);
4130
4131 /*
4132 * try to push all the items after our slot into the
4133 * right leaf
4134 */
4135 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4136 if (ret < 0)
4137 return ret;
4138
4139 if (ret == 0)
4140 progress++;
4141
4142 nritems = btrfs_header_nritems(path->nodes[0]);
4143 /*
4144 * our goal is to get our slot at the start or end of a leaf. If
4145 * we've done so we're done
4146 */
4147 if (path->slots[0] == 0 || path->slots[0] == nritems)
4148 return 0;
4149
4150 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4151 return 0;
4152
4153 /* try to push all the items before our slot into the next leaf */
4154 slot = path->slots[0];
4155 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4156 if (ret < 0)
4157 return ret;
4158
4159 if (ret == 0)
4160 progress++;
4161
4162 if (progress)
4163 return 0;
4164 return 1;
4165 }
4166
4167 /*
4168 * split the path's leaf in two, making sure there is at least data_size
4169 * available for the resulting leaf level of the path.
4170 *
4171 * returns 0 if all went well and < 0 on failure.
4172 */
4173 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4174 struct btrfs_root *root,
4175 struct btrfs_key *ins_key,
4176 struct btrfs_path *path, int data_size,
4177 int extend)
4178 {
4179 struct btrfs_disk_key disk_key;
4180 struct extent_buffer *l;
4181 u32 nritems;
4182 int mid;
4183 int slot;
4184 struct extent_buffer *right;
4185 struct btrfs_fs_info *fs_info = root->fs_info;
4186 int ret = 0;
4187 int wret;
4188 int split;
4189 int num_doubles = 0;
4190 int tried_avoid_double = 0;
4191
4192 l = path->nodes[0];
4193 slot = path->slots[0];
4194 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4195 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
4196 return -EOVERFLOW;
4197
4198 /* first try to make some room by pushing left and right */
4199 if (data_size && path->nodes[1]) {
4200 int space_needed = data_size;
4201
4202 if (slot < btrfs_header_nritems(l))
4203 space_needed -= btrfs_leaf_free_space(root, l);
4204
4205 wret = push_leaf_right(trans, root, path, space_needed,
4206 space_needed, 0, 0);
4207 if (wret < 0)
4208 return wret;
4209 if (wret) {
4210 wret = push_leaf_left(trans, root, path, space_needed,
4211 space_needed, 0, (u32)-1);
4212 if (wret < 0)
4213 return wret;
4214 }
4215 l = path->nodes[0];
4216
4217 /* did the pushes work? */
4218 if (btrfs_leaf_free_space(root, l) >= data_size)
4219 return 0;
4220 }
4221
4222 if (!path->nodes[1]) {
4223 ret = insert_new_root(trans, root, path, 1);
4224 if (ret)
4225 return ret;
4226 }
4227 again:
4228 split = 1;
4229 l = path->nodes[0];
4230 slot = path->slots[0];
4231 nritems = btrfs_header_nritems(l);
4232 mid = (nritems + 1) / 2;
4233
4234 if (mid <= slot) {
4235 if (nritems == 1 ||
4236 leaf_space_used(l, mid, nritems - mid) + data_size >
4237 BTRFS_LEAF_DATA_SIZE(root)) {
4238 if (slot >= nritems) {
4239 split = 0;
4240 } else {
4241 mid = slot;
4242 if (mid != nritems &&
4243 leaf_space_used(l, mid, nritems - mid) +
4244 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4245 if (data_size && !tried_avoid_double)
4246 goto push_for_double;
4247 split = 2;
4248 }
4249 }
4250 }
4251 } else {
4252 if (leaf_space_used(l, 0, mid) + data_size >
4253 BTRFS_LEAF_DATA_SIZE(root)) {
4254 if (!extend && data_size && slot == 0) {
4255 split = 0;
4256 } else if ((extend || !data_size) && slot == 0) {
4257 mid = 1;
4258 } else {
4259 mid = slot;
4260 if (mid != nritems &&
4261 leaf_space_used(l, mid, nritems - mid) +
4262 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4263 if (data_size && !tried_avoid_double)
4264 goto push_for_double;
4265 split = 2;
4266 }
4267 }
4268 }
4269 }
4270
4271 if (split == 0)
4272 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4273 else
4274 btrfs_item_key(l, &disk_key, mid);
4275
4276 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
4277 &disk_key, 0, l->start, 0);
4278 if (IS_ERR(right))
4279 return PTR_ERR(right);
4280
4281 root_add_used(root, root->nodesize);
4282
4283 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4284 btrfs_set_header_bytenr(right, right->start);
4285 btrfs_set_header_generation(right, trans->transid);
4286 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4287 btrfs_set_header_owner(right, root->root_key.objectid);
4288 btrfs_set_header_level(right, 0);
4289 write_extent_buffer(right, fs_info->fsid,
4290 btrfs_header_fsid(), BTRFS_FSID_SIZE);
4291
4292 write_extent_buffer(right, fs_info->chunk_tree_uuid,
4293 btrfs_header_chunk_tree_uuid(right),
4294 BTRFS_UUID_SIZE);
4295
4296 if (split == 0) {
4297 if (mid <= slot) {
4298 btrfs_set_header_nritems(right, 0);
4299 insert_ptr(trans, root, path, &disk_key, right->start,
4300 path->slots[1] + 1, 1);
4301 btrfs_tree_unlock(path->nodes[0]);
4302 free_extent_buffer(path->nodes[0]);
4303 path->nodes[0] = right;
4304 path->slots[0] = 0;
4305 path->slots[1] += 1;
4306 } else {
4307 btrfs_set_header_nritems(right, 0);
4308 insert_ptr(trans, root, path, &disk_key, right->start,
4309 path->slots[1], 1);
4310 btrfs_tree_unlock(path->nodes[0]);
4311 free_extent_buffer(path->nodes[0]);
4312 path->nodes[0] = right;
4313 path->slots[0] = 0;
4314 if (path->slots[1] == 0)
4315 fixup_low_keys(fs_info, path, &disk_key, 1);
4316 }
4317 btrfs_mark_buffer_dirty(right);
4318 return ret;
4319 }
4320
4321 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4322
4323 if (split == 2) {
4324 BUG_ON(num_doubles != 0);
4325 num_doubles++;
4326 goto again;
4327 }
4328
4329 return 0;
4330
4331 push_for_double:
4332 push_for_double_split(trans, root, path, data_size);
4333 tried_avoid_double = 1;
4334 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4335 return 0;
4336 goto again;
4337 }
4338
4339 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4340 struct btrfs_root *root,
4341 struct btrfs_path *path, int ins_len)
4342 {
4343 struct btrfs_key key;
4344 struct extent_buffer *leaf;
4345 struct btrfs_file_extent_item *fi;
4346 u64 extent_len = 0;
4347 u32 item_size;
4348 int ret;
4349
4350 leaf = path->nodes[0];
4351 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4352
4353 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4354 key.type != BTRFS_EXTENT_CSUM_KEY);
4355
4356 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4357 return 0;
4358
4359 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4360 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4361 fi = btrfs_item_ptr(leaf, path->slots[0],
4362 struct btrfs_file_extent_item);
4363 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4364 }
4365 btrfs_release_path(path);
4366
4367 path->keep_locks = 1;
4368 path->search_for_split = 1;
4369 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4370 path->search_for_split = 0;
4371 if (ret > 0)
4372 ret = -EAGAIN;
4373 if (ret < 0)
4374 goto err;
4375
4376 ret = -EAGAIN;
4377 leaf = path->nodes[0];
4378 /* if our item isn't there, return now */
4379 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4380 goto err;
4381
4382 /* the leaf has changed, it now has room. return now */
4383 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4384 goto err;
4385
4386 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4387 fi = btrfs_item_ptr(leaf, path->slots[0],
4388 struct btrfs_file_extent_item);
4389 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4390 goto err;
4391 }
4392
4393 btrfs_set_path_blocking(path);
4394 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4395 if (ret)
4396 goto err;
4397
4398 path->keep_locks = 0;
4399 btrfs_unlock_up_safe(path, 1);
4400 return 0;
4401 err:
4402 path->keep_locks = 0;
4403 return ret;
4404 }
4405
4406 static noinline int split_item(struct btrfs_trans_handle *trans,
4407 struct btrfs_root *root,
4408 struct btrfs_path *path,
4409 struct btrfs_key *new_key,
4410 unsigned long split_offset)
4411 {
4412 struct extent_buffer *leaf;
4413 struct btrfs_item *item;
4414 struct btrfs_item *new_item;
4415 int slot;
4416 char *buf;
4417 u32 nritems;
4418 u32 item_size;
4419 u32 orig_offset;
4420 struct btrfs_disk_key disk_key;
4421
4422 leaf = path->nodes[0];
4423 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4424
4425 btrfs_set_path_blocking(path);
4426
4427 item = btrfs_item_nr(path->slots[0]);
4428 orig_offset = btrfs_item_offset(leaf, item);
4429 item_size = btrfs_item_size(leaf, item);
4430
4431 buf = kmalloc(item_size, GFP_NOFS);
4432 if (!buf)
4433 return -ENOMEM;
4434
4435 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4436 path->slots[0]), item_size);
4437
4438 slot = path->slots[0] + 1;
4439 nritems = btrfs_header_nritems(leaf);
4440 if (slot != nritems) {
4441 /* shift the items */
4442 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4443 btrfs_item_nr_offset(slot),
4444 (nritems - slot) * sizeof(struct btrfs_item));
4445 }
4446
4447 btrfs_cpu_key_to_disk(&disk_key, new_key);
4448 btrfs_set_item_key(leaf, &disk_key, slot);
4449
4450 new_item = btrfs_item_nr(slot);
4451
4452 btrfs_set_item_offset(leaf, new_item, orig_offset);
4453 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4454
4455 btrfs_set_item_offset(leaf, item,
4456 orig_offset + item_size - split_offset);
4457 btrfs_set_item_size(leaf, item, split_offset);
4458
4459 btrfs_set_header_nritems(leaf, nritems + 1);
4460
4461 /* write the data for the start of the original item */
4462 write_extent_buffer(leaf, buf,
4463 btrfs_item_ptr_offset(leaf, path->slots[0]),
4464 split_offset);
4465
4466 /* write the data for the new item */
4467 write_extent_buffer(leaf, buf + split_offset,
4468 btrfs_item_ptr_offset(leaf, slot),
4469 item_size - split_offset);
4470 btrfs_mark_buffer_dirty(leaf);
4471
4472 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4473 kfree(buf);
4474 return 0;
4475 }
4476
4477 /*
4478 * This function splits a single item into two items,
4479 * giving 'new_key' to the new item and splitting the
4480 * old one at split_offset (from the start of the item).
4481 *
4482 * The path may be released by this operation. After
4483 * the split, the path is pointing to the old item. The
4484 * new item is going to be in the same node as the old one.
4485 *
4486 * Note, the item being split must be smaller enough to live alone on
4487 * a tree block with room for one extra struct btrfs_item
4488 *
4489 * This allows us to split the item in place, keeping a lock on the
4490 * leaf the entire time.
4491 */
4492 int btrfs_split_item(struct btrfs_trans_handle *trans,
4493 struct btrfs_root *root,
4494 struct btrfs_path *path,
4495 struct btrfs_key *new_key,
4496 unsigned long split_offset)
4497 {
4498 int ret;
4499 ret = setup_leaf_for_split(trans, root, path,
4500 sizeof(struct btrfs_item));
4501 if (ret)
4502 return ret;
4503
4504 ret = split_item(trans, root, path, new_key, split_offset);
4505 return ret;
4506 }
4507
4508 /*
4509 * This function duplicate a item, giving 'new_key' to the new item.
4510 * It guarantees both items live in the same tree leaf and the new item
4511 * is contiguous with the original item.
4512 *
4513 * This allows us to split file extent in place, keeping a lock on the
4514 * leaf the entire time.
4515 */
4516 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4517 struct btrfs_root *root,
4518 struct btrfs_path *path,
4519 struct btrfs_key *new_key)
4520 {
4521 struct extent_buffer *leaf;
4522 int ret;
4523 u32 item_size;
4524
4525 leaf = path->nodes[0];
4526 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4527 ret = setup_leaf_for_split(trans, root, path,
4528 item_size + sizeof(struct btrfs_item));
4529 if (ret)
4530 return ret;
4531
4532 path->slots[0]++;
4533 setup_items_for_insert(root, path, new_key, &item_size,
4534 item_size, item_size +
4535 sizeof(struct btrfs_item), 1);
4536 leaf = path->nodes[0];
4537 memcpy_extent_buffer(leaf,
4538 btrfs_item_ptr_offset(leaf, path->slots[0]),
4539 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4540 item_size);
4541 return 0;
4542 }
4543
4544 /*
4545 * make the item pointed to by the path smaller. new_size indicates
4546 * how small to make it, and from_end tells us if we just chop bytes
4547 * off the end of the item or if we shift the item to chop bytes off
4548 * the front.
4549 */
4550 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4551 u32 new_size, int from_end)
4552 {
4553 int slot;
4554 struct extent_buffer *leaf;
4555 struct btrfs_item *item;
4556 u32 nritems;
4557 unsigned int data_end;
4558 unsigned int old_data_start;
4559 unsigned int old_size;
4560 unsigned int size_diff;
4561 int i;
4562 struct btrfs_map_token token;
4563
4564 btrfs_init_map_token(&token);
4565
4566 leaf = path->nodes[0];
4567 slot = path->slots[0];
4568
4569 old_size = btrfs_item_size_nr(leaf, slot);
4570 if (old_size == new_size)
4571 return;
4572
4573 nritems = btrfs_header_nritems(leaf);
4574 data_end = leaf_data_end(root, leaf);
4575
4576 old_data_start = btrfs_item_offset_nr(leaf, slot);
4577
4578 size_diff = old_size - new_size;
4579
4580 BUG_ON(slot < 0);
4581 BUG_ON(slot >= nritems);
4582
4583 /*
4584 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4585 */
4586 /* first correct the data pointers */
4587 for (i = slot; i < nritems; i++) {
4588 u32 ioff;
4589 item = btrfs_item_nr(i);
4590
4591 ioff = btrfs_token_item_offset(leaf, item, &token);
4592 btrfs_set_token_item_offset(leaf, item,
4593 ioff + size_diff, &token);
4594 }
4595
4596 /* shift the data */
4597 if (from_end) {
4598 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4599 data_end + size_diff, btrfs_leaf_data(leaf) +
4600 data_end, old_data_start + new_size - data_end);
4601 } else {
4602 struct btrfs_disk_key disk_key;
4603 u64 offset;
4604
4605 btrfs_item_key(leaf, &disk_key, slot);
4606
4607 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4608 unsigned long ptr;
4609 struct btrfs_file_extent_item *fi;
4610
4611 fi = btrfs_item_ptr(leaf, slot,
4612 struct btrfs_file_extent_item);
4613 fi = (struct btrfs_file_extent_item *)(
4614 (unsigned long)fi - size_diff);
4615
4616 if (btrfs_file_extent_type(leaf, fi) ==
4617 BTRFS_FILE_EXTENT_INLINE) {
4618 ptr = btrfs_item_ptr_offset(leaf, slot);
4619 memmove_extent_buffer(leaf, ptr,
4620 (unsigned long)fi,
4621 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4622 }
4623 }
4624
4625 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4626 data_end + size_diff, btrfs_leaf_data(leaf) +
4627 data_end, old_data_start - data_end);
4628
4629 offset = btrfs_disk_key_offset(&disk_key);
4630 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4631 btrfs_set_item_key(leaf, &disk_key, slot);
4632 if (slot == 0)
4633 fixup_low_keys(root->fs_info, path, &disk_key, 1);
4634 }
4635
4636 item = btrfs_item_nr(slot);
4637 btrfs_set_item_size(leaf, item, new_size);
4638 btrfs_mark_buffer_dirty(leaf);
4639
4640 if (btrfs_leaf_free_space(root, leaf) < 0) {
4641 btrfs_print_leaf(root, leaf);
4642 BUG();
4643 }
4644 }
4645
4646 /*
4647 * make the item pointed to by the path bigger, data_size is the added size.
4648 */
4649 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4650 u32 data_size)
4651 {
4652 int slot;
4653 struct extent_buffer *leaf;
4654 struct btrfs_item *item;
4655 u32 nritems;
4656 unsigned int data_end;
4657 unsigned int old_data;
4658 unsigned int old_size;
4659 int i;
4660 struct btrfs_map_token token;
4661
4662 btrfs_init_map_token(&token);
4663
4664 leaf = path->nodes[0];
4665
4666 nritems = btrfs_header_nritems(leaf);
4667 data_end = leaf_data_end(root, leaf);
4668
4669 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4670 btrfs_print_leaf(root, leaf);
4671 BUG();
4672 }
4673 slot = path->slots[0];
4674 old_data = btrfs_item_end_nr(leaf, slot);
4675
4676 BUG_ON(slot < 0);
4677 if (slot >= nritems) {
4678 btrfs_print_leaf(root, leaf);
4679 btrfs_crit(root->fs_info, "slot %d too large, nritems %d",
4680 slot, nritems);
4681 BUG_ON(1);
4682 }
4683
4684 /*
4685 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4686 */
4687 /* first correct the data pointers */
4688 for (i = slot; i < nritems; i++) {
4689 u32 ioff;
4690 item = btrfs_item_nr(i);
4691
4692 ioff = btrfs_token_item_offset(leaf, item, &token);
4693 btrfs_set_token_item_offset(leaf, item,
4694 ioff - data_size, &token);
4695 }
4696
4697 /* shift the data */
4698 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4699 data_end - data_size, btrfs_leaf_data(leaf) +
4700 data_end, old_data - data_end);
4701
4702 data_end = old_data;
4703 old_size = btrfs_item_size_nr(leaf, slot);
4704 item = btrfs_item_nr(slot);
4705 btrfs_set_item_size(leaf, item, old_size + data_size);
4706 btrfs_mark_buffer_dirty(leaf);
4707
4708 if (btrfs_leaf_free_space(root, leaf) < 0) {
4709 btrfs_print_leaf(root, leaf);
4710 BUG();
4711 }
4712 }
4713
4714 /*
4715 * this is a helper for btrfs_insert_empty_items, the main goal here is
4716 * to save stack depth by doing the bulk of the work in a function
4717 * that doesn't call btrfs_search_slot
4718 */
4719 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4720 struct btrfs_key *cpu_key, u32 *data_size,
4721 u32 total_data, u32 total_size, int nr)
4722 {
4723 struct btrfs_item *item;
4724 int i;
4725 u32 nritems;
4726 unsigned int data_end;
4727 struct btrfs_disk_key disk_key;
4728 struct extent_buffer *leaf;
4729 int slot;
4730 struct btrfs_map_token token;
4731
4732 if (path->slots[0] == 0) {
4733 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4734 fixup_low_keys(root->fs_info, path, &disk_key, 1);
4735 }
4736 btrfs_unlock_up_safe(path, 1);
4737
4738 btrfs_init_map_token(&token);
4739
4740 leaf = path->nodes[0];
4741 slot = path->slots[0];
4742
4743 nritems = btrfs_header_nritems(leaf);
4744 data_end = leaf_data_end(root, leaf);
4745
4746 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4747 btrfs_print_leaf(root, leaf);
4748 btrfs_crit(root->fs_info, "not enough freespace need %u have %d",
4749 total_size, btrfs_leaf_free_space(root, leaf));
4750 BUG();
4751 }
4752
4753 if (slot != nritems) {
4754 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4755
4756 if (old_data < data_end) {
4757 btrfs_print_leaf(root, leaf);
4758 btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d",
4759 slot, old_data, data_end);
4760 BUG_ON(1);
4761 }
4762 /*
4763 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4764 */
4765 /* first correct the data pointers */
4766 for (i = slot; i < nritems; i++) {
4767 u32 ioff;
4768
4769 item = btrfs_item_nr( i);
4770 ioff = btrfs_token_item_offset(leaf, item, &token);
4771 btrfs_set_token_item_offset(leaf, item,
4772 ioff - total_data, &token);
4773 }
4774 /* shift the items */
4775 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4776 btrfs_item_nr_offset(slot),
4777 (nritems - slot) * sizeof(struct btrfs_item));
4778
4779 /* shift the data */
4780 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4781 data_end - total_data, btrfs_leaf_data(leaf) +
4782 data_end, old_data - data_end);
4783 data_end = old_data;
4784 }
4785
4786 /* setup the item for the new data */
4787 for (i = 0; i < nr; i++) {
4788 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4789 btrfs_set_item_key(leaf, &disk_key, slot + i);
4790 item = btrfs_item_nr(slot + i);
4791 btrfs_set_token_item_offset(leaf, item,
4792 data_end - data_size[i], &token);
4793 data_end -= data_size[i];
4794 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4795 }
4796
4797 btrfs_set_header_nritems(leaf, nritems + nr);
4798 btrfs_mark_buffer_dirty(leaf);
4799
4800 if (btrfs_leaf_free_space(root, leaf) < 0) {
4801 btrfs_print_leaf(root, leaf);
4802 BUG();
4803 }
4804 }
4805
4806 /*
4807 * Given a key and some data, insert items into the tree.
4808 * This does all the path init required, making room in the tree if needed.
4809 */
4810 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4811 struct btrfs_root *root,
4812 struct btrfs_path *path,
4813 struct btrfs_key *cpu_key, u32 *data_size,
4814 int nr)
4815 {
4816 int ret = 0;
4817 int slot;
4818 int i;
4819 u32 total_size = 0;
4820 u32 total_data = 0;
4821
4822 for (i = 0; i < nr; i++)
4823 total_data += data_size[i];
4824
4825 total_size = total_data + (nr * sizeof(struct btrfs_item));
4826 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4827 if (ret == 0)
4828 return -EEXIST;
4829 if (ret < 0)
4830 return ret;
4831
4832 slot = path->slots[0];
4833 BUG_ON(slot < 0);
4834
4835 setup_items_for_insert(root, path, cpu_key, data_size,
4836 total_data, total_size, nr);
4837 return 0;
4838 }
4839
4840 /*
4841 * Given a key and some data, insert an item into the tree.
4842 * This does all the path init required, making room in the tree if needed.
4843 */
4844 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4845 *root, struct btrfs_key *cpu_key, void *data, u32
4846 data_size)
4847 {
4848 int ret = 0;
4849 struct btrfs_path *path;
4850 struct extent_buffer *leaf;
4851 unsigned long ptr;
4852
4853 path = btrfs_alloc_path();
4854 if (!path)
4855 return -ENOMEM;
4856 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4857 if (!ret) {
4858 leaf = path->nodes[0];
4859 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4860 write_extent_buffer(leaf, data, ptr, data_size);
4861 btrfs_mark_buffer_dirty(leaf);
4862 }
4863 btrfs_free_path(path);
4864 return ret;
4865 }
4866
4867 /*
4868 * delete the pointer from a given node.
4869 *
4870 * the tree should have been previously balanced so the deletion does not
4871 * empty a node.
4872 */
4873 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4874 int level, int slot)
4875 {
4876 struct extent_buffer *parent = path->nodes[level];
4877 u32 nritems;
4878 int ret;
4879
4880 nritems = btrfs_header_nritems(parent);
4881 if (slot != nritems - 1) {
4882 if (level)
4883 tree_mod_log_eb_move(root->fs_info, parent, slot,
4884 slot + 1, nritems - slot - 1);
4885 memmove_extent_buffer(parent,
4886 btrfs_node_key_ptr_offset(slot),
4887 btrfs_node_key_ptr_offset(slot + 1),
4888 sizeof(struct btrfs_key_ptr) *
4889 (nritems - slot - 1));
4890 } else if (level) {
4891 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4892 MOD_LOG_KEY_REMOVE, GFP_NOFS);
4893 BUG_ON(ret < 0);
4894 }
4895
4896 nritems--;
4897 btrfs_set_header_nritems(parent, nritems);
4898 if (nritems == 0 && parent == root->node) {
4899 BUG_ON(btrfs_header_level(root->node) != 1);
4900 /* just turn the root into a leaf and break */
4901 btrfs_set_header_level(root->node, 0);
4902 } else if (slot == 0) {
4903 struct btrfs_disk_key disk_key;
4904
4905 btrfs_node_key(parent, &disk_key, 0);
4906 fixup_low_keys(root->fs_info, path, &disk_key, level + 1);
4907 }
4908 btrfs_mark_buffer_dirty(parent);
4909 }
4910
4911 /*
4912 * a helper function to delete the leaf pointed to by path->slots[1] and
4913 * path->nodes[1].
4914 *
4915 * This deletes the pointer in path->nodes[1] and frees the leaf
4916 * block extent. zero is returned if it all worked out, < 0 otherwise.
4917 *
4918 * The path must have already been setup for deleting the leaf, including
4919 * all the proper balancing. path->nodes[1] must be locked.
4920 */
4921 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4922 struct btrfs_root *root,
4923 struct btrfs_path *path,
4924 struct extent_buffer *leaf)
4925 {
4926 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4927 del_ptr(root, path, 1, path->slots[1]);
4928
4929 /*
4930 * btrfs_free_extent is expensive, we want to make sure we
4931 * aren't holding any locks when we call it
4932 */
4933 btrfs_unlock_up_safe(path, 0);
4934
4935 root_sub_used(root, leaf->len);
4936
4937 extent_buffer_get(leaf);
4938 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4939 free_extent_buffer_stale(leaf);
4940 }
4941 /*
4942 * delete the item at the leaf level in path. If that empties
4943 * the leaf, remove it from the tree
4944 */
4945 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4946 struct btrfs_path *path, int slot, int nr)
4947 {
4948 struct extent_buffer *leaf;
4949 struct btrfs_item *item;
4950 u32 last_off;
4951 u32 dsize = 0;
4952 int ret = 0;
4953 int wret;
4954 int i;
4955 u32 nritems;
4956 struct btrfs_map_token token;
4957
4958 btrfs_init_map_token(&token);
4959
4960 leaf = path->nodes[0];
4961 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4962
4963 for (i = 0; i < nr; i++)
4964 dsize += btrfs_item_size_nr(leaf, slot + i);
4965
4966 nritems = btrfs_header_nritems(leaf);
4967
4968 if (slot + nr != nritems) {
4969 int data_end = leaf_data_end(root, leaf);
4970
4971 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4972 data_end + dsize,
4973 btrfs_leaf_data(leaf) + data_end,
4974 last_off - data_end);
4975
4976 for (i = slot + nr; i < nritems; i++) {
4977 u32 ioff;
4978
4979 item = btrfs_item_nr(i);
4980 ioff = btrfs_token_item_offset(leaf, item, &token);
4981 btrfs_set_token_item_offset(leaf, item,
4982 ioff + dsize, &token);
4983 }
4984
4985 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4986 btrfs_item_nr_offset(slot + nr),
4987 sizeof(struct btrfs_item) *
4988 (nritems - slot - nr));
4989 }
4990 btrfs_set_header_nritems(leaf, nritems - nr);
4991 nritems -= nr;
4992
4993 /* delete the leaf if we've emptied it */
4994 if (nritems == 0) {
4995 if (leaf == root->node) {
4996 btrfs_set_header_level(leaf, 0);
4997 } else {
4998 btrfs_set_path_blocking(path);
4999 clean_tree_block(trans, root->fs_info, leaf);
5000 btrfs_del_leaf(trans, root, path, leaf);
5001 }
5002 } else {
5003 int used = leaf_space_used(leaf, 0, nritems);
5004 if (slot == 0) {
5005 struct btrfs_disk_key disk_key;
5006
5007 btrfs_item_key(leaf, &disk_key, 0);
5008 fixup_low_keys(root->fs_info, path, &disk_key, 1);
5009 }
5010
5011 /* delete the leaf if it is mostly empty */
5012 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
5013 /* push_leaf_left fixes the path.
5014 * make sure the path still points to our leaf
5015 * for possible call to del_ptr below
5016 */
5017 slot = path->slots[1];
5018 extent_buffer_get(leaf);
5019
5020 btrfs_set_path_blocking(path);
5021 wret = push_leaf_left(trans, root, path, 1, 1,
5022 1, (u32)-1);
5023 if (wret < 0 && wret != -ENOSPC)
5024 ret = wret;
5025
5026 if (path->nodes[0] == leaf &&
5027 btrfs_header_nritems(leaf)) {
5028 wret = push_leaf_right(trans, root, path, 1,
5029 1, 1, 0);
5030 if (wret < 0 && wret != -ENOSPC)
5031 ret = wret;
5032 }
5033
5034 if (btrfs_header_nritems(leaf) == 0) {
5035 path->slots[1] = slot;
5036 btrfs_del_leaf(trans, root, path, leaf);
5037 free_extent_buffer(leaf);
5038 ret = 0;
5039 } else {
5040 /* if we're still in the path, make sure
5041 * we're dirty. Otherwise, one of the
5042 * push_leaf functions must have already
5043 * dirtied this buffer
5044 */
5045 if (path->nodes[0] == leaf)
5046 btrfs_mark_buffer_dirty(leaf);
5047 free_extent_buffer(leaf);
5048 }
5049 } else {
5050 btrfs_mark_buffer_dirty(leaf);
5051 }
5052 }
5053 return ret;
5054 }
5055
5056 /*
5057 * search the tree again to find a leaf with lesser keys
5058 * returns 0 if it found something or 1 if there are no lesser leaves.
5059 * returns < 0 on io errors.
5060 *
5061 * This may release the path, and so you may lose any locks held at the
5062 * time you call it.
5063 */
5064 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5065 {
5066 struct btrfs_key key;
5067 struct btrfs_disk_key found_key;
5068 int ret;
5069
5070 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5071
5072 if (key.offset > 0) {
5073 key.offset--;
5074 } else if (key.type > 0) {
5075 key.type--;
5076 key.offset = (u64)-1;
5077 } else if (key.objectid > 0) {
5078 key.objectid--;
5079 key.type = (u8)-1;
5080 key.offset = (u64)-1;
5081 } else {
5082 return 1;
5083 }
5084
5085 btrfs_release_path(path);
5086 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5087 if (ret < 0)
5088 return ret;
5089 btrfs_item_key(path->nodes[0], &found_key, 0);
5090 ret = comp_keys(&found_key, &key);
5091 /*
5092 * We might have had an item with the previous key in the tree right
5093 * before we released our path. And after we released our path, that
5094 * item might have been pushed to the first slot (0) of the leaf we
5095 * were holding due to a tree balance. Alternatively, an item with the
5096 * previous key can exist as the only element of a leaf (big fat item).
5097 * Therefore account for these 2 cases, so that our callers (like
5098 * btrfs_previous_item) don't miss an existing item with a key matching
5099 * the previous key we computed above.
5100 */
5101 if (ret <= 0)
5102 return 0;
5103 return 1;
5104 }
5105
5106 /*
5107 * A helper function to walk down the tree starting at min_key, and looking
5108 * for nodes or leaves that are have a minimum transaction id.
5109 * This is used by the btree defrag code, and tree logging
5110 *
5111 * This does not cow, but it does stuff the starting key it finds back
5112 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5113 * key and get a writable path.
5114 *
5115 * This does lock as it descends, and path->keep_locks should be set
5116 * to 1 by the caller.
5117 *
5118 * This honors path->lowest_level to prevent descent past a given level
5119 * of the tree.
5120 *
5121 * min_trans indicates the oldest transaction that you are interested
5122 * in walking through. Any nodes or leaves older than min_trans are
5123 * skipped over (without reading them).
5124 *
5125 * returns zero if something useful was found, < 0 on error and 1 if there
5126 * was nothing in the tree that matched the search criteria.
5127 */
5128 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5129 struct btrfs_path *path,
5130 u64 min_trans)
5131 {
5132 struct extent_buffer *cur;
5133 struct btrfs_key found_key;
5134 int slot;
5135 int sret;
5136 u32 nritems;
5137 int level;
5138 int ret = 1;
5139 int keep_locks = path->keep_locks;
5140
5141 path->keep_locks = 1;
5142 again:
5143 cur = btrfs_read_lock_root_node(root);
5144 level = btrfs_header_level(cur);
5145 WARN_ON(path->nodes[level]);
5146 path->nodes[level] = cur;
5147 path->locks[level] = BTRFS_READ_LOCK;
5148
5149 if (btrfs_header_generation(cur) < min_trans) {
5150 ret = 1;
5151 goto out;
5152 }
5153 while (1) {
5154 nritems = btrfs_header_nritems(cur);
5155 level = btrfs_header_level(cur);
5156 sret = bin_search(cur, min_key, level, &slot);
5157
5158 /* at the lowest level, we're done, setup the path and exit */
5159 if (level == path->lowest_level) {
5160 if (slot >= nritems)
5161 goto find_next_key;
5162 ret = 0;
5163 path->slots[level] = slot;
5164 btrfs_item_key_to_cpu(cur, &found_key, slot);
5165 goto out;
5166 }
5167 if (sret && slot > 0)
5168 slot--;
5169 /*
5170 * check this node pointer against the min_trans parameters.
5171 * If it is too old, old, skip to the next one.
5172 */
5173 while (slot < nritems) {
5174 u64 gen;
5175
5176 gen = btrfs_node_ptr_generation(cur, slot);
5177 if (gen < min_trans) {
5178 slot++;
5179 continue;
5180 }
5181 break;
5182 }
5183 find_next_key:
5184 /*
5185 * we didn't find a candidate key in this node, walk forward
5186 * and find another one
5187 */
5188 if (slot >= nritems) {
5189 path->slots[level] = slot;
5190 btrfs_set_path_blocking(path);
5191 sret = btrfs_find_next_key(root, path, min_key, level,
5192 min_trans);
5193 if (sret == 0) {
5194 btrfs_release_path(path);
5195 goto again;
5196 } else {
5197 goto out;
5198 }
5199 }
5200 /* save our key for returning back */
5201 btrfs_node_key_to_cpu(cur, &found_key, slot);
5202 path->slots[level] = slot;
5203 if (level == path->lowest_level) {
5204 ret = 0;
5205 goto out;
5206 }
5207 btrfs_set_path_blocking(path);
5208 cur = read_node_slot(root, cur, slot);
5209 BUG_ON(!cur); /* -ENOMEM */
5210
5211 btrfs_tree_read_lock(cur);
5212
5213 path->locks[level - 1] = BTRFS_READ_LOCK;
5214 path->nodes[level - 1] = cur;
5215 unlock_up(path, level, 1, 0, NULL);
5216 btrfs_clear_path_blocking(path, NULL, 0);
5217 }
5218 out:
5219 path->keep_locks = keep_locks;
5220 if (ret == 0) {
5221 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5222 btrfs_set_path_blocking(path);
5223 memcpy(min_key, &found_key, sizeof(found_key));
5224 }
5225 return ret;
5226 }
5227
5228 static void tree_move_down(struct btrfs_root *root,
5229 struct btrfs_path *path,
5230 int *level, int root_level)
5231 {
5232 BUG_ON(*level == 0);
5233 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
5234 path->slots[*level]);
5235 path->slots[*level - 1] = 0;
5236 (*level)--;
5237 }
5238
5239 static int tree_move_next_or_upnext(struct btrfs_root *root,
5240 struct btrfs_path *path,
5241 int *level, int root_level)
5242 {
5243 int ret = 0;
5244 int nritems;
5245 nritems = btrfs_header_nritems(path->nodes[*level]);
5246
5247 path->slots[*level]++;
5248
5249 while (path->slots[*level] >= nritems) {
5250 if (*level == root_level)
5251 return -1;
5252
5253 /* move upnext */
5254 path->slots[*level] = 0;
5255 free_extent_buffer(path->nodes[*level]);
5256 path->nodes[*level] = NULL;
5257 (*level)++;
5258 path->slots[*level]++;
5259
5260 nritems = btrfs_header_nritems(path->nodes[*level]);
5261 ret = 1;
5262 }
5263 return ret;
5264 }
5265
5266 /*
5267 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5268 * or down.
5269 */
5270 static int tree_advance(struct btrfs_root *root,
5271 struct btrfs_path *path,
5272 int *level, int root_level,
5273 int allow_down,
5274 struct btrfs_key *key)
5275 {
5276 int ret;
5277
5278 if (*level == 0 || !allow_down) {
5279 ret = tree_move_next_or_upnext(root, path, level, root_level);
5280 } else {
5281 tree_move_down(root, path, level, root_level);
5282 ret = 0;
5283 }
5284 if (ret >= 0) {
5285 if (*level == 0)
5286 btrfs_item_key_to_cpu(path->nodes[*level], key,
5287 path->slots[*level]);
5288 else
5289 btrfs_node_key_to_cpu(path->nodes[*level], key,
5290 path->slots[*level]);
5291 }
5292 return ret;
5293 }
5294
5295 static int tree_compare_item(struct btrfs_root *left_root,
5296 struct btrfs_path *left_path,
5297 struct btrfs_path *right_path,
5298 char *tmp_buf)
5299 {
5300 int cmp;
5301 int len1, len2;
5302 unsigned long off1, off2;
5303
5304 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5305 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5306 if (len1 != len2)
5307 return 1;
5308
5309 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5310 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5311 right_path->slots[0]);
5312
5313 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5314
5315 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5316 if (cmp)
5317 return 1;
5318 return 0;
5319 }
5320
5321 #define ADVANCE 1
5322 #define ADVANCE_ONLY_NEXT -1
5323
5324 /*
5325 * This function compares two trees and calls the provided callback for
5326 * every changed/new/deleted item it finds.
5327 * If shared tree blocks are encountered, whole subtrees are skipped, making
5328 * the compare pretty fast on snapshotted subvolumes.
5329 *
5330 * This currently works on commit roots only. As commit roots are read only,
5331 * we don't do any locking. The commit roots are protected with transactions.
5332 * Transactions are ended and rejoined when a commit is tried in between.
5333 *
5334 * This function checks for modifications done to the trees while comparing.
5335 * If it detects a change, it aborts immediately.
5336 */
5337 int btrfs_compare_trees(struct btrfs_root *left_root,
5338 struct btrfs_root *right_root,
5339 btrfs_changed_cb_t changed_cb, void *ctx)
5340 {
5341 int ret;
5342 int cmp;
5343 struct btrfs_path *left_path = NULL;
5344 struct btrfs_path *right_path = NULL;
5345 struct btrfs_key left_key;
5346 struct btrfs_key right_key;
5347 char *tmp_buf = NULL;
5348 int left_root_level;
5349 int right_root_level;
5350 int left_level;
5351 int right_level;
5352 int left_end_reached;
5353 int right_end_reached;
5354 int advance_left;
5355 int advance_right;
5356 u64 left_blockptr;
5357 u64 right_blockptr;
5358 u64 left_gen;
5359 u64 right_gen;
5360
5361 left_path = btrfs_alloc_path();
5362 if (!left_path) {
5363 ret = -ENOMEM;
5364 goto out;
5365 }
5366 right_path = btrfs_alloc_path();
5367 if (!right_path) {
5368 ret = -ENOMEM;
5369 goto out;
5370 }
5371
5372 tmp_buf = kmalloc(left_root->nodesize, GFP_KERNEL | __GFP_NOWARN);
5373 if (!tmp_buf) {
5374 tmp_buf = vmalloc(left_root->nodesize);
5375 if (!tmp_buf) {
5376 ret = -ENOMEM;
5377 goto out;
5378 }
5379 }
5380
5381 left_path->search_commit_root = 1;
5382 left_path->skip_locking = 1;
5383 right_path->search_commit_root = 1;
5384 right_path->skip_locking = 1;
5385
5386 /*
5387 * Strategy: Go to the first items of both trees. Then do
5388 *
5389 * If both trees are at level 0
5390 * Compare keys of current items
5391 * If left < right treat left item as new, advance left tree
5392 * and repeat
5393 * If left > right treat right item as deleted, advance right tree
5394 * and repeat
5395 * If left == right do deep compare of items, treat as changed if
5396 * needed, advance both trees and repeat
5397 * If both trees are at the same level but not at level 0
5398 * Compare keys of current nodes/leafs
5399 * If left < right advance left tree and repeat
5400 * If left > right advance right tree and repeat
5401 * If left == right compare blockptrs of the next nodes/leafs
5402 * If they match advance both trees but stay at the same level
5403 * and repeat
5404 * If they don't match advance both trees while allowing to go
5405 * deeper and repeat
5406 * If tree levels are different
5407 * Advance the tree that needs it and repeat
5408 *
5409 * Advancing a tree means:
5410 * If we are at level 0, try to go to the next slot. If that's not
5411 * possible, go one level up and repeat. Stop when we found a level
5412 * where we could go to the next slot. We may at this point be on a
5413 * node or a leaf.
5414 *
5415 * If we are not at level 0 and not on shared tree blocks, go one
5416 * level deeper.
5417 *
5418 * If we are not at level 0 and on shared tree blocks, go one slot to
5419 * the right if possible or go up and right.
5420 */
5421
5422 down_read(&left_root->fs_info->commit_root_sem);
5423 left_level = btrfs_header_level(left_root->commit_root);
5424 left_root_level = left_level;
5425 left_path->nodes[left_level] = left_root->commit_root;
5426 extent_buffer_get(left_path->nodes[left_level]);
5427
5428 right_level = btrfs_header_level(right_root->commit_root);
5429 right_root_level = right_level;
5430 right_path->nodes[right_level] = right_root->commit_root;
5431 extent_buffer_get(right_path->nodes[right_level]);
5432 up_read(&left_root->fs_info->commit_root_sem);
5433
5434 if (left_level == 0)
5435 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5436 &left_key, left_path->slots[left_level]);
5437 else
5438 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5439 &left_key, left_path->slots[left_level]);
5440 if (right_level == 0)
5441 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5442 &right_key, right_path->slots[right_level]);
5443 else
5444 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5445 &right_key, right_path->slots[right_level]);
5446
5447 left_end_reached = right_end_reached = 0;
5448 advance_left = advance_right = 0;
5449
5450 while (1) {
5451 if (advance_left && !left_end_reached) {
5452 ret = tree_advance(left_root, left_path, &left_level,
5453 left_root_level,
5454 advance_left != ADVANCE_ONLY_NEXT,
5455 &left_key);
5456 if (ret < 0)
5457 left_end_reached = ADVANCE;
5458 advance_left = 0;
5459 }
5460 if (advance_right && !right_end_reached) {
5461 ret = tree_advance(right_root, right_path, &right_level,
5462 right_root_level,
5463 advance_right != ADVANCE_ONLY_NEXT,
5464 &right_key);
5465 if (ret < 0)
5466 right_end_reached = ADVANCE;
5467 advance_right = 0;
5468 }
5469
5470 if (left_end_reached && right_end_reached) {
5471 ret = 0;
5472 goto out;
5473 } else if (left_end_reached) {
5474 if (right_level == 0) {
5475 ret = changed_cb(left_root, right_root,
5476 left_path, right_path,
5477 &right_key,
5478 BTRFS_COMPARE_TREE_DELETED,
5479 ctx);
5480 if (ret < 0)
5481 goto out;
5482 }
5483 advance_right = ADVANCE;
5484 continue;
5485 } else if (right_end_reached) {
5486 if (left_level == 0) {
5487 ret = changed_cb(left_root, right_root,
5488 left_path, right_path,
5489 &left_key,
5490 BTRFS_COMPARE_TREE_NEW,
5491 ctx);
5492 if (ret < 0)
5493 goto out;
5494 }
5495 advance_left = ADVANCE;
5496 continue;
5497 }
5498
5499 if (left_level == 0 && right_level == 0) {
5500 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5501 if (cmp < 0) {
5502 ret = changed_cb(left_root, right_root,
5503 left_path, right_path,
5504 &left_key,
5505 BTRFS_COMPARE_TREE_NEW,
5506 ctx);
5507 if (ret < 0)
5508 goto out;
5509 advance_left = ADVANCE;
5510 } else if (cmp > 0) {
5511 ret = changed_cb(left_root, right_root,
5512 left_path, right_path,
5513 &right_key,
5514 BTRFS_COMPARE_TREE_DELETED,
5515 ctx);
5516 if (ret < 0)
5517 goto out;
5518 advance_right = ADVANCE;
5519 } else {
5520 enum btrfs_compare_tree_result result;
5521
5522 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5523 ret = tree_compare_item(left_root, left_path,
5524 right_path, tmp_buf);
5525 if (ret)
5526 result = BTRFS_COMPARE_TREE_CHANGED;
5527 else
5528 result = BTRFS_COMPARE_TREE_SAME;
5529 ret = changed_cb(left_root, right_root,
5530 left_path, right_path,
5531 &left_key, result, ctx);
5532 if (ret < 0)
5533 goto out;
5534 advance_left = ADVANCE;
5535 advance_right = ADVANCE;
5536 }
5537 } else if (left_level == right_level) {
5538 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5539 if (cmp < 0) {
5540 advance_left = ADVANCE;
5541 } else if (cmp > 0) {
5542 advance_right = ADVANCE;
5543 } else {
5544 left_blockptr = btrfs_node_blockptr(
5545 left_path->nodes[left_level],
5546 left_path->slots[left_level]);
5547 right_blockptr = btrfs_node_blockptr(
5548 right_path->nodes[right_level],
5549 right_path->slots[right_level]);
5550 left_gen = btrfs_node_ptr_generation(
5551 left_path->nodes[left_level],
5552 left_path->slots[left_level]);
5553 right_gen = btrfs_node_ptr_generation(
5554 right_path->nodes[right_level],
5555 right_path->slots[right_level]);
5556 if (left_blockptr == right_blockptr &&
5557 left_gen == right_gen) {
5558 /*
5559 * As we're on a shared block, don't
5560 * allow to go deeper.
5561 */
5562 advance_left = ADVANCE_ONLY_NEXT;
5563 advance_right = ADVANCE_ONLY_NEXT;
5564 } else {
5565 advance_left = ADVANCE;
5566 advance_right = ADVANCE;
5567 }
5568 }
5569 } else if (left_level < right_level) {
5570 advance_right = ADVANCE;
5571 } else {
5572 advance_left = ADVANCE;
5573 }
5574 }
5575
5576 out:
5577 btrfs_free_path(left_path);
5578 btrfs_free_path(right_path);
5579 kvfree(tmp_buf);
5580 return ret;
5581 }
5582
5583 /*
5584 * this is similar to btrfs_next_leaf, but does not try to preserve
5585 * and fixup the path. It looks for and returns the next key in the
5586 * tree based on the current path and the min_trans parameters.
5587 *
5588 * 0 is returned if another key is found, < 0 if there are any errors
5589 * and 1 is returned if there are no higher keys in the tree
5590 *
5591 * path->keep_locks should be set to 1 on the search made before
5592 * calling this function.
5593 */
5594 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5595 struct btrfs_key *key, int level, u64 min_trans)
5596 {
5597 int slot;
5598 struct extent_buffer *c;
5599
5600 WARN_ON(!path->keep_locks);
5601 while (level < BTRFS_MAX_LEVEL) {
5602 if (!path->nodes[level])
5603 return 1;
5604
5605 slot = path->slots[level] + 1;
5606 c = path->nodes[level];
5607 next:
5608 if (slot >= btrfs_header_nritems(c)) {
5609 int ret;
5610 int orig_lowest;
5611 struct btrfs_key cur_key;
5612 if (level + 1 >= BTRFS_MAX_LEVEL ||
5613 !path->nodes[level + 1])
5614 return 1;
5615
5616 if (path->locks[level + 1]) {
5617 level++;
5618 continue;
5619 }
5620
5621 slot = btrfs_header_nritems(c) - 1;
5622 if (level == 0)
5623 btrfs_item_key_to_cpu(c, &cur_key, slot);
5624 else
5625 btrfs_node_key_to_cpu(c, &cur_key, slot);
5626
5627 orig_lowest = path->lowest_level;
5628 btrfs_release_path(path);
5629 path->lowest_level = level;
5630 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5631 0, 0);
5632 path->lowest_level = orig_lowest;
5633 if (ret < 0)
5634 return ret;
5635
5636 c = path->nodes[level];
5637 slot = path->slots[level];
5638 if (ret == 0)
5639 slot++;
5640 goto next;
5641 }
5642
5643 if (level == 0)
5644 btrfs_item_key_to_cpu(c, key, slot);
5645 else {
5646 u64 gen = btrfs_node_ptr_generation(c, slot);
5647
5648 if (gen < min_trans) {
5649 slot++;
5650 goto next;
5651 }
5652 btrfs_node_key_to_cpu(c, key, slot);
5653 }
5654 return 0;
5655 }
5656 return 1;
5657 }
5658
5659 /*
5660 * search the tree again to find a leaf with greater keys
5661 * returns 0 if it found something or 1 if there are no greater leaves.
5662 * returns < 0 on io errors.
5663 */
5664 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5665 {
5666 return btrfs_next_old_leaf(root, path, 0);
5667 }
5668
5669 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5670 u64 time_seq)
5671 {
5672 int slot;
5673 int level;
5674 struct extent_buffer *c;
5675 struct extent_buffer *next;
5676 struct btrfs_key key;
5677 u32 nritems;
5678 int ret;
5679 int old_spinning = path->leave_spinning;
5680 int next_rw_lock = 0;
5681
5682 nritems = btrfs_header_nritems(path->nodes[0]);
5683 if (nritems == 0)
5684 return 1;
5685
5686 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5687 again:
5688 level = 1;
5689 next = NULL;
5690 next_rw_lock = 0;
5691 btrfs_release_path(path);
5692
5693 path->keep_locks = 1;
5694 path->leave_spinning = 1;
5695
5696 if (time_seq)
5697 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5698 else
5699 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5700 path->keep_locks = 0;
5701
5702 if (ret < 0)
5703 return ret;
5704
5705 nritems = btrfs_header_nritems(path->nodes[0]);
5706 /*
5707 * by releasing the path above we dropped all our locks. A balance
5708 * could have added more items next to the key that used to be
5709 * at the very end of the block. So, check again here and
5710 * advance the path if there are now more items available.
5711 */
5712 if (nritems > 0 && path->slots[0] < nritems - 1) {
5713 if (ret == 0)
5714 path->slots[0]++;
5715 ret = 0;
5716 goto done;
5717 }
5718 /*
5719 * So the above check misses one case:
5720 * - after releasing the path above, someone has removed the item that
5721 * used to be at the very end of the block, and balance between leafs
5722 * gets another one with bigger key.offset to replace it.
5723 *
5724 * This one should be returned as well, or we can get leaf corruption
5725 * later(esp. in __btrfs_drop_extents()).
5726 *
5727 * And a bit more explanation about this check,
5728 * with ret > 0, the key isn't found, the path points to the slot
5729 * where it should be inserted, so the path->slots[0] item must be the
5730 * bigger one.
5731 */
5732 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5733 ret = 0;
5734 goto done;
5735 }
5736
5737 while (level < BTRFS_MAX_LEVEL) {
5738 if (!path->nodes[level]) {
5739 ret = 1;
5740 goto done;
5741 }
5742
5743 slot = path->slots[level] + 1;
5744 c = path->nodes[level];
5745 if (slot >= btrfs_header_nritems(c)) {
5746 level++;
5747 if (level == BTRFS_MAX_LEVEL) {
5748 ret = 1;
5749 goto done;
5750 }
5751 continue;
5752 }
5753
5754 if (next) {
5755 btrfs_tree_unlock_rw(next, next_rw_lock);
5756 free_extent_buffer(next);
5757 }
5758
5759 next = c;
5760 next_rw_lock = path->locks[level];
5761 ret = read_block_for_search(NULL, root, path, &next, level,
5762 slot, &key, 0);
5763 if (ret == -EAGAIN)
5764 goto again;
5765
5766 if (ret < 0) {
5767 btrfs_release_path(path);
5768 goto done;
5769 }
5770
5771 if (!path->skip_locking) {
5772 ret = btrfs_try_tree_read_lock(next);
5773 if (!ret && time_seq) {
5774 /*
5775 * If we don't get the lock, we may be racing
5776 * with push_leaf_left, holding that lock while
5777 * itself waiting for the leaf we've currently
5778 * locked. To solve this situation, we give up
5779 * on our lock and cycle.
5780 */
5781 free_extent_buffer(next);
5782 btrfs_release_path(path);
5783 cond_resched();
5784 goto again;
5785 }
5786 if (!ret) {
5787 btrfs_set_path_blocking(path);
5788 btrfs_tree_read_lock(next);
5789 btrfs_clear_path_blocking(path, next,
5790 BTRFS_READ_LOCK);
5791 }
5792 next_rw_lock = BTRFS_READ_LOCK;
5793 }
5794 break;
5795 }
5796 path->slots[level] = slot;
5797 while (1) {
5798 level--;
5799 c = path->nodes[level];
5800 if (path->locks[level])
5801 btrfs_tree_unlock_rw(c, path->locks[level]);
5802
5803 free_extent_buffer(c);
5804 path->nodes[level] = next;
5805 path->slots[level] = 0;
5806 if (!path->skip_locking)
5807 path->locks[level] = next_rw_lock;
5808 if (!level)
5809 break;
5810
5811 ret = read_block_for_search(NULL, root, path, &next, level,
5812 0, &key, 0);
5813 if (ret == -EAGAIN)
5814 goto again;
5815
5816 if (ret < 0) {
5817 btrfs_release_path(path);
5818 goto done;
5819 }
5820
5821 if (!path->skip_locking) {
5822 ret = btrfs_try_tree_read_lock(next);
5823 if (!ret) {
5824 btrfs_set_path_blocking(path);
5825 btrfs_tree_read_lock(next);
5826 btrfs_clear_path_blocking(path, next,
5827 BTRFS_READ_LOCK);
5828 }
5829 next_rw_lock = BTRFS_READ_LOCK;
5830 }
5831 }
5832 ret = 0;
5833 done:
5834 unlock_up(path, 0, 1, 0, NULL);
5835 path->leave_spinning = old_spinning;
5836 if (!old_spinning)
5837 btrfs_set_path_blocking(path);
5838
5839 return ret;
5840 }
5841
5842 /*
5843 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5844 * searching until it gets past min_objectid or finds an item of 'type'
5845 *
5846 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5847 */
5848 int btrfs_previous_item(struct btrfs_root *root,
5849 struct btrfs_path *path, u64 min_objectid,
5850 int type)
5851 {
5852 struct btrfs_key found_key;
5853 struct extent_buffer *leaf;
5854 u32 nritems;
5855 int ret;
5856
5857 while (1) {
5858 if (path->slots[0] == 0) {
5859 btrfs_set_path_blocking(path);
5860 ret = btrfs_prev_leaf(root, path);
5861 if (ret != 0)
5862 return ret;
5863 } else {
5864 path->slots[0]--;
5865 }
5866 leaf = path->nodes[0];
5867 nritems = btrfs_header_nritems(leaf);
5868 if (nritems == 0)
5869 return 1;
5870 if (path->slots[0] == nritems)
5871 path->slots[0]--;
5872
5873 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5874 if (found_key.objectid < min_objectid)
5875 break;
5876 if (found_key.type == type)
5877 return 0;
5878 if (found_key.objectid == min_objectid &&
5879 found_key.type < type)
5880 break;
5881 }
5882 return 1;
5883 }
5884
5885 /*
5886 * search in extent tree to find a previous Metadata/Data extent item with
5887 * min objecitd.
5888 *
5889 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5890 */
5891 int btrfs_previous_extent_item(struct btrfs_root *root,
5892 struct btrfs_path *path, u64 min_objectid)
5893 {
5894 struct btrfs_key found_key;
5895 struct extent_buffer *leaf;
5896 u32 nritems;
5897 int ret;
5898
5899 while (1) {
5900 if (path->slots[0] == 0) {
5901 btrfs_set_path_blocking(path);
5902 ret = btrfs_prev_leaf(root, path);
5903 if (ret != 0)
5904 return ret;
5905 } else {
5906 path->slots[0]--;
5907 }
5908 leaf = path->nodes[0];
5909 nritems = btrfs_header_nritems(leaf);
5910 if (nritems == 0)
5911 return 1;
5912 if (path->slots[0] == nritems)
5913 path->slots[0]--;
5914
5915 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5916 if (found_key.objectid < min_objectid)
5917 break;
5918 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5919 found_key.type == BTRFS_METADATA_ITEM_KEY)
5920 return 0;
5921 if (found_key.objectid == min_objectid &&
5922 found_key.type < BTRFS_EXTENT_ITEM_KEY)
5923 break;
5924 }
5925 return 1;
5926 }