Line data Source code
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 "ctree.h"
23 : #include "disk-io.h"
24 : #include "transaction.h"
25 : #include "print-tree.h"
26 : #include "locking.h"
27 :
28 : static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
29 : *root, struct btrfs_path *path, int level);
30 : static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
31 : *root, struct btrfs_key *ins_key,
32 : struct btrfs_path *path, int data_size, int extend);
33 : static int push_node_left(struct btrfs_trans_handle *trans,
34 : struct btrfs_root *root, struct extent_buffer *dst,
35 : struct extent_buffer *src, int empty);
36 : static int balance_node_right(struct btrfs_trans_handle *trans,
37 : struct btrfs_root *root,
38 : struct extent_buffer *dst_buf,
39 : struct extent_buffer *src_buf);
40 : static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
41 : int level, int slot);
42 : static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
43 : struct extent_buffer *eb);
44 :
45 1854047 : struct btrfs_path *btrfs_alloc_path(void)
46 : {
47 : struct btrfs_path *path;
48 1855831 : path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
49 1854158 : return path;
50 : }
51 :
52 : /*
53 : * set all locked nodes in the path to blocking locks. This should
54 : * be done before scheduling
55 : */
56 2594767 : noinline void btrfs_set_path_blocking(struct btrfs_path *p)
57 : {
58 : int i;
59 23375417 : for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
60 20777480 : if (!p->nodes[i] || !p->locks[i])
61 19637098 : continue;
62 1140382 : btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
63 1143552 : if (p->locks[i] == BTRFS_READ_LOCK)
64 409973 : p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
65 733579 : else if (p->locks[i] == BTRFS_WRITE_LOCK)
66 713014 : p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
67 : }
68 2597937 : }
69 :
70 : /*
71 : * reset all the locked nodes in the patch to spinning locks.
72 : *
73 : * held is used to keep lockdep happy, when lockdep is enabled
74 : * we set held to a blocking lock before we go around and
75 : * retake all the spinlocks in the path. You can safely use NULL
76 : * for held
77 : */
78 5212087 : noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
79 : struct extent_buffer *held, int held_rw)
80 : {
81 : int i;
82 :
83 : #ifdef CONFIG_DEBUG_LOCK_ALLOC
84 : /* lockdep really cares that we take all of these spinlocks
85 : * in the right order. If any of the locks in the path are not
86 : * currently blocking, it is going to complain. So, make really
87 : * really sure by forcing the path to blocking before we clear
88 : * the path blocking.
89 : */
90 : if (held) {
91 : btrfs_set_lock_blocking_rw(held, held_rw);
92 : if (held_rw == BTRFS_WRITE_LOCK)
93 : held_rw = BTRFS_WRITE_LOCK_BLOCKING;
94 : else if (held_rw == BTRFS_READ_LOCK)
95 : held_rw = BTRFS_READ_LOCK_BLOCKING;
96 : }
97 : btrfs_set_path_blocking(p);
98 : #endif
99 :
100 46861138 : for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
101 41652998 : if (p->nodes[i] && p->locks[i]) {
102 3786076 : btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
103 3782129 : if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
104 326601 : p->locks[i] = BTRFS_WRITE_LOCK;
105 3455528 : else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
106 5404 : p->locks[i] = BTRFS_READ_LOCK;
107 : }
108 : }
109 :
110 : #ifdef CONFIG_DEBUG_LOCK_ALLOC
111 : if (held)
112 : btrfs_clear_lock_blocking_rw(held, held_rw);
113 : #endif
114 5208140 : }
115 :
116 : /* this also releases the path */
117 1855866 : void btrfs_free_path(struct btrfs_path *p)
118 : {
119 1855866 : if (!p)
120 1855760 : return;
121 1855874 : btrfs_release_path(p);
122 1855739 : kmem_cache_free(btrfs_path_cachep, p);
123 : }
124 :
125 : /*
126 : * path release drops references on the extent buffers in the path
127 : * and it drops any locks held by this path
128 : *
129 : * It is safe to call this on paths that no locks or extent buffers held.
130 : */
131 4493213 : noinline void btrfs_release_path(struct btrfs_path *p)
132 : {
133 : int i;
134 :
135 40453389 : for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
136 35957959 : p->slots[i] = 0;
137 35957959 : if (!p->nodes[i])
138 30953674 : continue;
139 5004285 : if (p->locks[i]) {
140 1384137 : btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
141 1384409 : p->locks[i] = 0;
142 : }
143 5004557 : free_extent_buffer(p->nodes[i]);
144 5006502 : p->nodes[i] = NULL;
145 : }
146 4495430 : }
147 :
148 : /*
149 : * safely gets a reference on the root node of a tree. A lock
150 : * is not taken, so a concurrent writer may put a different node
151 : * at the root of the tree. See btrfs_lock_root_node for the
152 : * looping required.
153 : *
154 : * The extent buffer returned by this has a reference taken, so
155 : * it won't disappear. It may stop being the root of the tree
156 : * at any time because there are no locks held.
157 : */
158 1872073 : struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
159 : {
160 : struct extent_buffer *eb;
161 :
162 : while (1) {
163 : rcu_read_lock();
164 1872156 : eb = rcu_dereference(root->node);
165 :
166 : /*
167 : * RCU really hurts here, we could free up the root node because
168 : * it was cow'ed but we may not get the new root node yet so do
169 : * the inc_not_zero dance and if it doesn't work then
170 : * synchronize_rcu and try again.
171 : */
172 1873751 : if (atomic_inc_not_zero(&eb->refs)) {
173 : rcu_read_unlock();
174 : break;
175 : }
176 : rcu_read_unlock();
177 : synchronize_rcu();
178 : }
179 1873755 : return eb;
180 : }
181 :
182 : /* loop around taking references on and locking the root node of the
183 : * tree until you end up with a lock on the root. A locked buffer
184 : * is returned, with a reference held.
185 : */
186 635185 : struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
187 : {
188 : struct extent_buffer *eb;
189 :
190 : while (1) {
191 635210 : eb = btrfs_root_node(root);
192 635225 : btrfs_tree_lock(eb);
193 635245 : if (eb == root->node)
194 : break;
195 25 : btrfs_tree_unlock(eb);
196 25 : free_extent_buffer(eb);
197 25 : }
198 635220 : return eb;
199 : }
200 :
201 : /* loop around taking references on and locking the root node of the
202 : * tree until you end up with a lock on the root. A locked buffer
203 : * is returned, with a reference held.
204 : */
205 1211496 : static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
206 : {
207 : struct extent_buffer *eb;
208 :
209 : while (1) {
210 1211636 : eb = btrfs_root_node(root);
211 1212206 : btrfs_tree_read_lock(eb);
212 1212141 : if (eb == root->node)
213 : break;
214 140 : btrfs_tree_read_unlock(eb);
215 140 : free_extent_buffer(eb);
216 140 : }
217 1212001 : return eb;
218 : }
219 :
220 : /* cowonly root (everything not a reference counted cow subvolume), just get
221 : * put onto a simple dirty list. transaction.c walks this to make sure they
222 : * get properly updated on disk.
223 : */
224 13659 : static void add_root_to_dirty_list(struct btrfs_root *root)
225 : {
226 13659 : spin_lock(&root->fs_info->trans_lock);
227 19174 : if (test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state) &&
228 5515 : list_empty(&root->dirty_list)) {
229 4833 : list_add(&root->dirty_list,
230 4833 : &root->fs_info->dirty_cowonly_roots);
231 : }
232 13659 : spin_unlock(&root->fs_info->trans_lock);
233 13659 : }
234 :
235 : /*
236 : * used by snapshot creation to make a copy of a root for a tree with
237 : * a given objectid. The buffer with the new root node is returned in
238 : * cow_ret, and this func returns zero on success or a negative error code.
239 : */
240 585 : int btrfs_copy_root(struct btrfs_trans_handle *trans,
241 : struct btrfs_root *root,
242 1170 : struct extent_buffer *buf,
243 : struct extent_buffer **cow_ret, u64 new_root_objectid)
244 : {
245 1316 : struct extent_buffer *cow;
246 : int ret = 0;
247 : int level;
248 : struct btrfs_disk_key disk_key;
249 :
250 585 : WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
251 : trans->transid != root->fs_info->running_transaction->transid);
252 585 : WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
253 : trans->transid != root->last_trans);
254 :
255 585 : level = btrfs_header_level(buf);
256 585 : if (level == 0)
257 : btrfs_item_key(buf, &disk_key, 0);
258 : else
259 79 : btrfs_node_key(buf, &disk_key, 0);
260 :
261 585 : cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
262 : new_root_objectid, &disk_key, level,
263 : buf->start, 0);
264 585 : if (IS_ERR(cow))
265 0 : return PTR_ERR(cow);
266 :
267 585 : copy_extent_buffer(cow, buf, 0, 0, cow->len);
268 585 : btrfs_set_header_bytenr(cow, cow->start);
269 585 : btrfs_set_header_generation(cow, trans->transid);
270 : btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
271 : btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
272 : BTRFS_HEADER_FLAG_RELOC);
273 585 : if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
274 : btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
275 : else
276 : btrfs_set_header_owner(cow, new_root_objectid);
277 :
278 585 : write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
279 : BTRFS_FSID_SIZE);
280 :
281 585 : WARN_ON(btrfs_header_generation(buf) > trans->transid);
282 585 : if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
283 439 : ret = btrfs_inc_ref(trans, root, cow, 1);
284 : else
285 146 : ret = btrfs_inc_ref(trans, root, cow, 0);
286 :
287 585 : if (ret)
288 : return ret;
289 :
290 585 : btrfs_mark_buffer_dirty(cow);
291 585 : *cow_ret = cow;
292 585 : return 0;
293 : }
294 :
295 : enum mod_log_op {
296 : MOD_LOG_KEY_REPLACE,
297 : MOD_LOG_KEY_ADD,
298 : MOD_LOG_KEY_REMOVE,
299 : MOD_LOG_KEY_REMOVE_WHILE_FREEING,
300 : MOD_LOG_KEY_REMOVE_WHILE_MOVING,
301 : MOD_LOG_MOVE_KEYS,
302 : MOD_LOG_ROOT_REPLACE,
303 : };
304 :
305 : struct tree_mod_move {
306 : int dst_slot;
307 : int nr_items;
308 : };
309 :
310 : struct tree_mod_root {
311 : u64 logical;
312 : u8 level;
313 : };
314 :
315 : struct tree_mod_elem {
316 : struct rb_node node;
317 : u64 index; /* shifted logical */
318 : u64 seq;
319 : enum mod_log_op op;
320 :
321 : /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
322 : int slot;
323 :
324 : /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
325 : u64 generation;
326 :
327 : /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
328 : struct btrfs_disk_key key;
329 : u64 blockptr;
330 :
331 : /* this is used for op == MOD_LOG_MOVE_KEYS */
332 : struct tree_mod_move move;
333 :
334 : /* this is used for op == MOD_LOG_ROOT_REPLACE */
335 : struct tree_mod_root old_root;
336 : };
337 :
338 : static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
339 : {
340 21988 : read_lock(&fs_info->tree_mod_log_lock);
341 : }
342 :
343 : static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
344 : {
345 : read_unlock(&fs_info->tree_mod_log_lock);
346 : }
347 :
348 : static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
349 : {
350 10596 : write_lock(&fs_info->tree_mod_log_lock);
351 : }
352 :
353 : static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
354 : {
355 : write_unlock(&fs_info->tree_mod_log_lock);
356 : }
357 :
358 : /*
359 : * Pull a new tree mod seq number for our operation.
360 : */
361 : static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
362 : {
363 15698 : return atomic64_inc_return(&fs_info->tree_mod_seq);
364 : }
365 :
366 : /*
367 : * This adds a new blocker to the tree mod log's blocker list if the @elem
368 : * passed does not already have a sequence number set. So when a caller expects
369 : * to record tree modifications, it should ensure to set elem->seq to zero
370 : * before calling btrfs_get_tree_mod_seq.
371 : * Returns a fresh, unused tree log modification sequence number, even if no new
372 : * blocker was added.
373 : */
374 4917 : u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
375 : struct seq_list *elem)
376 : {
377 : tree_mod_log_write_lock(fs_info);
378 : spin_lock(&fs_info->tree_mod_seq_lock);
379 4917 : if (!elem->seq) {
380 4917 : elem->seq = btrfs_inc_tree_mod_seq(fs_info);
381 4917 : list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
382 : }
383 : spin_unlock(&fs_info->tree_mod_seq_lock);
384 : tree_mod_log_write_unlock(fs_info);
385 :
386 4917 : return elem->seq;
387 : }
388 :
389 17271 : void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
390 : struct seq_list *elem)
391 : {
392 : struct rb_root *tm_root;
393 : struct rb_node *node;
394 : struct rb_node *next;
395 : struct seq_list *cur_elem;
396 : struct tree_mod_elem *tm;
397 : u64 min_seq = (u64)-1;
398 17271 : u64 seq_putting = elem->seq;
399 :
400 17271 : if (!seq_putting)
401 : return;
402 :
403 : spin_lock(&fs_info->tree_mod_seq_lock);
404 4917 : list_del(&elem->list);
405 4917 : elem->seq = 0;
406 :
407 30418 : list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
408 26763 : if (cur_elem->seq < min_seq) {
409 2262 : if (seq_putting > cur_elem->seq) {
410 : /*
411 : * blocker with lower sequence number exists, we
412 : * cannot remove anything from the log
413 : */
414 : spin_unlock(&fs_info->tree_mod_seq_lock);
415 : return;
416 : }
417 : min_seq = cur_elem->seq;
418 : }
419 : }
420 : spin_unlock(&fs_info->tree_mod_seq_lock);
421 :
422 : /*
423 : * anything that's lower than the lowest existing (read: blocked)
424 : * sequence number can be removed from the tree.
425 : */
426 : tree_mod_log_write_lock(fs_info);
427 3655 : tm_root = &fs_info->tree_mod_log;
428 216105 : for (node = rb_first(tm_root); node; node = next) {
429 208795 : next = rb_next(node);
430 : tm = container_of(node, struct tree_mod_elem, node);
431 208795 : if (tm->seq > min_seq)
432 205863 : continue;
433 2932 : rb_erase(node, tm_root);
434 2932 : kfree(tm);
435 : }
436 : tree_mod_log_write_unlock(fs_info);
437 : }
438 :
439 : /*
440 : * key order of the log:
441 : * index -> sequence
442 : *
443 : * the index is the shifted logical of the *new* root node for root replace
444 : * operations, or the shifted logical of the affected block for all other
445 : * operations.
446 : *
447 : * Note: must be called with write lock (tree_mod_log_write_lock).
448 : */
449 : static noinline int
450 2932 : __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
451 : {
452 : struct rb_root *tm_root;
453 : struct rb_node **new;
454 : struct rb_node *parent = NULL;
455 : struct tree_mod_elem *cur;
456 :
457 2932 : BUG_ON(!tm);
458 :
459 2932 : tm->seq = btrfs_inc_tree_mod_seq(fs_info);
460 :
461 2932 : tm_root = &fs_info->tree_mod_log;
462 2932 : new = &tm_root->rb_node;
463 32774 : while (*new) {
464 : cur = container_of(*new, struct tree_mod_elem, node);
465 : parent = *new;
466 26910 : if (cur->index < tm->index)
467 352 : new = &((*new)->rb_left);
468 26558 : else if (cur->index > tm->index)
469 4232 : new = &((*new)->rb_right);
470 22326 : else if (cur->seq < tm->seq)
471 22326 : new = &((*new)->rb_left);
472 0 : else if (cur->seq > tm->seq)
473 0 : new = &((*new)->rb_right);
474 : else
475 : return -EEXIST;
476 : }
477 :
478 2932 : rb_link_node(&tm->node, parent, new);
479 2932 : rb_insert_color(&tm->node, tm_root);
480 2932 : return 0;
481 : }
482 :
483 : /*
484 : * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
485 : * returns zero with the tree_mod_log_lock acquired. The caller must hold
486 : * this until all tree mod log insertions are recorded in the rb tree and then
487 : * call tree_mod_log_write_unlock() to release.
488 : */
489 2024 : static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
490 2011 : struct extent_buffer *eb) {
491 2024 : smp_mb();
492 4048 : if (list_empty(&(fs_info)->tree_mod_seq_list))
493 : return 1;
494 4035 : if (eb && btrfs_header_level(eb) == 0)
495 : return 1;
496 :
497 : tree_mod_log_write_lock(fs_info);
498 2024 : if (list_empty(&(fs_info)->tree_mod_seq_list)) {
499 : tree_mod_log_write_unlock(fs_info);
500 0 : return 1;
501 : }
502 :
503 : return 0;
504 : }
505 :
506 : /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
507 : static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
508 2011 : struct extent_buffer *eb)
509 : {
510 79644 : smp_mb();
511 159290 : if (list_empty(&(fs_info)->tree_mod_seq_list))
512 : return 0;
513 4022 : if (eb && btrfs_header_level(eb) == 0)
514 : return 0;
515 :
516 : return 1;
517 : }
518 :
519 : static struct tree_mod_elem *
520 2697 : alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
521 : enum mod_log_op op, gfp_t flags)
522 : {
523 : struct tree_mod_elem *tm;
524 :
525 2697 : tm = kzalloc(sizeof(*tm), flags);
526 2697 : if (!tm)
527 : return NULL;
528 :
529 2697 : tm->index = eb->start >> PAGE_CACHE_SHIFT;
530 2697 : if (op != MOD_LOG_KEY_ADD) {
531 2448 : btrfs_node_key(eb, &tm->key, slot);
532 2448 : tm->blockptr = btrfs_node_blockptr(eb, slot);
533 : }
534 2697 : tm->op = op;
535 2697 : tm->slot = slot;
536 2697 : tm->generation = btrfs_node_ptr_generation(eb, slot);
537 2697 : RB_CLEAR_NODE(&tm->node);
538 :
539 2697 : return tm;
540 : }
541 :
542 : static noinline int
543 58000 : tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
544 : struct extent_buffer *eb, int slot,
545 : enum mod_log_op op, gfp_t flags)
546 : {
547 : struct tree_mod_elem *tm;
548 : int ret;
549 :
550 58001 : if (!tree_mod_need_log(fs_info, eb))
551 : return 0;
552 :
553 1783 : tm = alloc_tree_mod_elem(eb, slot, op, flags);
554 1783 : if (!tm)
555 : return -ENOMEM;
556 :
557 1783 : if (tree_mod_dont_log(fs_info, eb)) {
558 0 : kfree(tm);
559 0 : return 0;
560 : }
561 :
562 1783 : ret = __tree_mod_log_insert(fs_info, tm);
563 : tree_mod_log_write_unlock(fs_info);
564 1783 : if (ret)
565 0 : kfree(tm);
566 :
567 1783 : return ret;
568 : }
569 :
570 : static noinline int
571 4522 : tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
572 : struct extent_buffer *eb, int dst_slot, int src_slot,
573 : int nr_items, gfp_t flags)
574 : {
575 : struct tree_mod_elem *tm = NULL;
576 : struct tree_mod_elem **tm_list = NULL;
577 : int ret = 0;
578 : int i;
579 : int locked = 0;
580 :
581 4522 : if (!tree_mod_need_log(fs_info, eb))
582 : return 0;
583 :
584 228 : tm_list = kzalloc(nr_items * sizeof(struct tree_mod_elem *), flags);
585 228 : if (!tm_list)
586 : return -ENOMEM;
587 :
588 228 : tm = kzalloc(sizeof(*tm), flags);
589 228 : if (!tm) {
590 : ret = -ENOMEM;
591 : goto free_tms;
592 : }
593 :
594 228 : tm->index = eb->start >> PAGE_CACHE_SHIFT;
595 228 : tm->slot = src_slot;
596 228 : tm->move.dst_slot = dst_slot;
597 228 : tm->move.nr_items = nr_items;
598 228 : tm->op = MOD_LOG_MOVE_KEYS;
599 :
600 455 : for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
601 227 : tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
602 : MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags);
603 227 : if (!tm_list[i]) {
604 : ret = -ENOMEM;
605 : goto free_tms;
606 : }
607 : }
608 :
609 228 : if (tree_mod_dont_log(fs_info, eb))
610 : goto free_tms;
611 : locked = 1;
612 :
613 : /*
614 : * When we override something during the move, we log these removals.
615 : * This can only happen when we move towards the beginning of the
616 : * buffer, i.e. dst_slot < src_slot.
617 : */
618 227 : for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
619 227 : ret = __tree_mod_log_insert(fs_info, tm_list[i]);
620 227 : if (ret)
621 : goto free_tms;
622 : }
623 :
624 228 : ret = __tree_mod_log_insert(fs_info, tm);
625 228 : if (ret)
626 : goto free_tms;
627 : tree_mod_log_write_unlock(fs_info);
628 228 : kfree(tm_list);
629 :
630 228 : return 0;
631 : free_tms:
632 0 : for (i = 0; i < nr_items; i++) {
633 0 : if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
634 0 : rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
635 0 : kfree(tm_list[i]);
636 : }
637 0 : if (locked)
638 : tree_mod_log_write_unlock(fs_info);
639 0 : kfree(tm_list);
640 0 : kfree(tm);
641 :
642 0 : return ret;
643 : }
644 :
645 : static inline int
646 6 : __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
647 : struct tree_mod_elem **tm_list,
648 : int nritems)
649 : {
650 : int i, j;
651 : int ret;
652 :
653 227 : for (i = nritems - 1; i >= 0; i--) {
654 221 : ret = __tree_mod_log_insert(fs_info, tm_list[i]);
655 221 : if (ret) {
656 0 : for (j = nritems - 1; j > i; j--)
657 0 : rb_erase(&tm_list[j]->node,
658 : &fs_info->tree_mod_log);
659 : return ret;
660 : }
661 : }
662 :
663 : return 0;
664 : }
665 :
666 : static noinline int
667 13659 : tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
668 21 : struct extent_buffer *old_root,
669 : struct extent_buffer *new_root, gfp_t flags,
670 : int log_removal)
671 : {
672 : struct tree_mod_elem *tm = NULL;
673 : struct tree_mod_elem **tm_list = NULL;
674 : int nritems = 0;
675 : int ret = 0;
676 : int i;
677 :
678 13659 : if (!tree_mod_need_log(fs_info, NULL))
679 : return 0;
680 :
681 14 : if (log_removal && btrfs_header_level(old_root) > 0) {
682 6 : nritems = btrfs_header_nritems(old_root);
683 6 : tm_list = kzalloc(nritems * sizeof(struct tree_mod_elem *),
684 : flags);
685 6 : if (!tm_list) {
686 : ret = -ENOMEM;
687 : goto free_tms;
688 : }
689 221 : for (i = 0; i < nritems; i++) {
690 221 : tm_list[i] = alloc_tree_mod_elem(old_root, i,
691 : MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags);
692 221 : if (!tm_list[i]) {
693 : ret = -ENOMEM;
694 : goto free_tms;
695 : }
696 : }
697 : }
698 :
699 7 : tm = kzalloc(sizeof(*tm), flags);
700 7 : if (!tm) {
701 : ret = -ENOMEM;
702 : goto free_tms;
703 : }
704 :
705 7 : tm->index = new_root->start >> PAGE_CACHE_SHIFT;
706 7 : tm->old_root.logical = old_root->start;
707 7 : tm->old_root.level = btrfs_header_level(old_root);
708 7 : tm->generation = btrfs_header_generation(old_root);
709 7 : tm->op = MOD_LOG_ROOT_REPLACE;
710 :
711 7 : if (tree_mod_dont_log(fs_info, NULL))
712 : goto free_tms;
713 :
714 7 : if (tm_list)
715 6 : ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
716 7 : if (!ret)
717 7 : ret = __tree_mod_log_insert(fs_info, tm);
718 :
719 : tree_mod_log_write_unlock(fs_info);
720 7 : if (ret)
721 : goto free_tms;
722 7 : kfree(tm_list);
723 :
724 : return ret;
725 :
726 : free_tms:
727 0 : if (tm_list) {
728 0 : for (i = 0; i < nritems; i++)
729 0 : kfree(tm_list[i]);
730 0 : kfree(tm_list);
731 : }
732 0 : kfree(tm);
733 :
734 : return ret;
735 : }
736 :
737 : static struct tree_mod_elem *
738 21967 : __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
739 : int smallest)
740 : {
741 : struct rb_root *tm_root;
742 : struct rb_node *node;
743 : struct tree_mod_elem *cur = NULL;
744 : struct tree_mod_elem *found = NULL;
745 21967 : u64 index = start >> PAGE_CACHE_SHIFT;
746 :
747 : tree_mod_log_read_lock(fs_info);
748 : tm_root = &fs_info->tree_mod_log;
749 21967 : node = tm_root->rb_node;
750 57305 : while (node) {
751 : cur = container_of(node, struct tree_mod_elem, node);
752 13371 : if (cur->index < index) {
753 1048 : node = node->rb_left;
754 12323 : } else if (cur->index > index) {
755 12219 : node = node->rb_right;
756 104 : } else if (cur->seq < min_seq) {
757 0 : node = node->rb_left;
758 104 : } else if (!smallest) {
759 : /* we want the node with the highest seq */
760 39 : if (found)
761 18 : BUG_ON(found->seq > cur->seq);
762 : found = cur;
763 39 : node = node->rb_left;
764 65 : } else if (cur->seq > min_seq) {
765 : /* we want the node with the smallest seq */
766 65 : if (found)
767 16 : BUG_ON(found->seq < cur->seq);
768 : found = cur;
769 65 : node = node->rb_right;
770 : } else {
771 : found = cur;
772 : break;
773 : }
774 : }
775 : tree_mod_log_read_unlock(fs_info);
776 :
777 21967 : return found;
778 : }
779 :
780 : /*
781 : * this returns the element from the log with the smallest time sequence
782 : * value that's in the log (the oldest log item). any element with a time
783 : * sequence lower than min_seq will be ignored.
784 : */
785 : static struct tree_mod_elem *
786 : tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
787 : u64 min_seq)
788 : {
789 20677 : return __tree_mod_log_search(fs_info, start, min_seq, 1);
790 : }
791 :
792 : /*
793 : * this returns the element from the log with the largest time sequence
794 : * value that's in the log (the most recent log item). any element with
795 : * a time sequence lower than min_seq will be ignored.
796 : */
797 : static struct tree_mod_elem *
798 : tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
799 : {
800 1290 : return __tree_mod_log_search(fs_info, start, min_seq, 0);
801 : }
802 :
803 : static noinline int
804 217 : tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
805 0 : struct extent_buffer *src, unsigned long dst_offset,
806 : unsigned long src_offset, int nr_items)
807 : {
808 : int ret = 0;
809 : struct tree_mod_elem **tm_list = NULL;
810 : struct tree_mod_elem **tm_list_add, **tm_list_rem;
811 : int i;
812 : int locked = 0;
813 :
814 211 : if (!tree_mod_need_log(fs_info, NULL))
815 : return 0;
816 :
817 6 : if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
818 : return 0;
819 :
820 6 : tm_list = kzalloc(nr_items * 2 * sizeof(struct tree_mod_elem *),
821 : GFP_NOFS);
822 6 : if (!tm_list)
823 : return -ENOMEM;
824 :
825 : tm_list_add = tm_list;
826 6 : tm_list_rem = tm_list + nr_items;
827 239 : for (i = 0; i < nr_items; i++) {
828 233 : tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
829 : MOD_LOG_KEY_REMOVE, GFP_NOFS);
830 233 : if (!tm_list_rem[i]) {
831 : ret = -ENOMEM;
832 : goto free_tms;
833 : }
834 :
835 233 : tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
836 : MOD_LOG_KEY_ADD, GFP_NOFS);
837 233 : if (!tm_list_add[i]) {
838 : ret = -ENOMEM;
839 : goto free_tms;
840 : }
841 : }
842 :
843 6 : if (tree_mod_dont_log(fs_info, NULL))
844 : goto free_tms;
845 : locked = 1;
846 :
847 233 : for (i = 0; i < nr_items; i++) {
848 233 : ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
849 233 : if (ret)
850 : goto free_tms;
851 233 : ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
852 233 : if (ret)
853 : goto free_tms;
854 : }
855 :
856 : tree_mod_log_write_unlock(fs_info);
857 6 : kfree(tm_list);
858 :
859 6 : return 0;
860 :
861 : free_tms:
862 0 : for (i = 0; i < nr_items * 2; i++) {
863 0 : if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
864 0 : rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
865 0 : kfree(tm_list[i]);
866 : }
867 0 : if (locked)
868 : tree_mod_log_write_unlock(fs_info);
869 0 : kfree(tm_list);
870 :
871 0 : return ret;
872 : }
873 :
874 : static inline void
875 4522 : tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
876 : int dst_offset, int src_offset, int nr_items)
877 : {
878 : int ret;
879 4522 : ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
880 : nr_items, GFP_NOFS);
881 4522 : BUG_ON(ret < 0);
882 4522 : }
883 :
884 : static noinline void
885 13775 : tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
886 : struct extent_buffer *eb, int slot, int atomic)
887 : {
888 : int ret;
889 :
890 13775 : ret = tree_mod_log_insert_key(fs_info, eb, slot,
891 : MOD_LOG_KEY_REPLACE,
892 : atomic ? GFP_ATOMIC : GFP_NOFS);
893 13775 : BUG_ON(ret < 0);
894 13775 : }
895 :
896 : static noinline int
897 38288 : tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
898 : {
899 : struct tree_mod_elem **tm_list = NULL;
900 : int nritems = 0;
901 : int i;
902 : int ret = 0;
903 :
904 38288 : if (btrfs_header_level(eb) == 0)
905 : return 0;
906 :
907 3252 : if (!tree_mod_need_log(fs_info, NULL))
908 : return 0;
909 :
910 0 : nritems = btrfs_header_nritems(eb);
911 0 : tm_list = kzalloc(nritems * sizeof(struct tree_mod_elem *),
912 : GFP_NOFS);
913 0 : if (!tm_list)
914 : return -ENOMEM;
915 :
916 0 : for (i = 0; i < nritems; i++) {
917 0 : tm_list[i] = alloc_tree_mod_elem(eb, i,
918 : MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
919 0 : if (!tm_list[i]) {
920 : ret = -ENOMEM;
921 : goto free_tms;
922 : }
923 : }
924 :
925 0 : if (tree_mod_dont_log(fs_info, eb))
926 : goto free_tms;
927 :
928 0 : ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
929 : tree_mod_log_write_unlock(fs_info);
930 0 : if (ret)
931 : goto free_tms;
932 0 : kfree(tm_list);
933 :
934 0 : return 0;
935 :
936 : free_tms:
937 0 : for (i = 0; i < nritems; i++)
938 0 : kfree(tm_list[i]);
939 0 : kfree(tm_list);
940 :
941 0 : return ret;
942 : }
943 :
944 : static noinline void
945 13659 : tree_mod_log_set_root_pointer(struct btrfs_root *root,
946 : struct extent_buffer *new_root_node,
947 : int log_removal)
948 : {
949 : int ret;
950 13659 : ret = tree_mod_log_insert_root(root->fs_info, root->node,
951 : new_root_node, GFP_NOFS, log_removal);
952 13658 : BUG_ON(ret < 0);
953 13658 : }
954 :
955 : /*
956 : * check if the tree block can be shared by multiple trees
957 : */
958 52290 : int btrfs_block_can_be_shared(struct btrfs_root *root,
959 19589 : struct extent_buffer *buf)
960 : {
961 : /*
962 : * Tree blocks not in refernece counted trees and tree roots
963 : * are never shared. If a block was allocated after the last
964 : * snapshot and the block was not allocated by tree relocation,
965 : * we know the block is not shared.
966 : */
967 75123 : if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
968 62011 : buf != root->node && buf != root->commit_root &&
969 : (btrfs_header_generation(buf) <=
970 18647 : btrfs_root_last_snapshot(&root->root_item) ||
971 : btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
972 : return 1;
973 : #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
974 73225 : if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
975 : btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
976 : return 1;
977 : #endif
978 51341 : return 0;
979 : }
980 :
981 52284 : static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
982 : struct btrfs_root *root,
983 53645 : struct extent_buffer *buf,
984 : struct extent_buffer *cow,
985 : int *last_ref)
986 : {
987 : u64 refs;
988 : u64 owner;
989 : u64 flags;
990 : u64 new_flags = 0;
991 : int ret;
992 :
993 : /*
994 : * Backrefs update rules:
995 : *
996 : * Always use full backrefs for extent pointers in tree block
997 : * allocated by tree relocation.
998 : *
999 : * If a shared tree block is no longer referenced by its owner
1000 : * tree (btrfs_header_owner(buf) == root->root_key.objectid),
1001 : * use full backrefs for extent pointers in tree block.
1002 : *
1003 : * If a tree block is been relocating
1004 : * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1005 : * use full backrefs for extent pointers in tree block.
1006 : * The reason for this is some operations (such as drop tree)
1007 : * are only allowed for blocks use full backrefs.
1008 : */
1009 :
1010 52284 : if (btrfs_block_can_be_shared(root, buf)) {
1011 949 : ret = btrfs_lookup_extent_info(trans, root, buf->start,
1012 : btrfs_header_level(buf), 1,
1013 : &refs, &flags);
1014 949 : if (ret)
1015 : return ret;
1016 949 : if (refs == 0) {
1017 : ret = -EROFS;
1018 0 : btrfs_std_error(root->fs_info, ret);
1019 0 : return ret;
1020 : }
1021 : } else {
1022 51335 : refs = 1;
1023 102668 : if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1024 : btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1025 2 : flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1026 : else
1027 51333 : flags = 0;
1028 : }
1029 :
1030 : owner = btrfs_header_owner(buf);
1031 52284 : BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
1032 : !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
1033 :
1034 52284 : if (refs > 1) {
1035 470 : if ((owner == root->root_key.objectid ||
1036 419 : root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
1037 419 : !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1038 412 : ret = btrfs_inc_ref(trans, root, buf, 1);
1039 412 : BUG_ON(ret); /* -ENOMEM */
1040 :
1041 412 : if (root->root_key.objectid ==
1042 : BTRFS_TREE_RELOC_OBJECTID) {
1043 9 : ret = btrfs_dec_ref(trans, root, buf, 0);
1044 9 : BUG_ON(ret); /* -ENOMEM */
1045 9 : ret = btrfs_inc_ref(trans, root, cow, 1);
1046 9 : BUG_ON(ret); /* -ENOMEM */
1047 : }
1048 : new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
1049 : } else {
1050 :
1051 58 : if (root->root_key.objectid ==
1052 : BTRFS_TREE_RELOC_OBJECTID)
1053 7 : ret = btrfs_inc_ref(trans, root, cow, 1);
1054 : else
1055 51 : ret = btrfs_inc_ref(trans, root, cow, 0);
1056 58 : BUG_ON(ret); /* -ENOMEM */
1057 : }
1058 470 : if (new_flags != 0) {
1059 412 : int level = btrfs_header_level(buf);
1060 :
1061 412 : ret = btrfs_set_disk_extent_flags(trans, root,
1062 : buf->start,
1063 412 : buf->len,
1064 : new_flags, level, 0);
1065 412 : if (ret)
1066 0 : return ret;
1067 : }
1068 : } else {
1069 51814 : if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
1070 332 : if (root->root_key.objectid ==
1071 : BTRFS_TREE_RELOC_OBJECTID)
1072 2 : ret = btrfs_inc_ref(trans, root, cow, 1);
1073 : else
1074 330 : ret = btrfs_inc_ref(trans, root, cow, 0);
1075 332 : BUG_ON(ret); /* -ENOMEM */
1076 332 : ret = btrfs_dec_ref(trans, root, buf, 1);
1077 332 : BUG_ON(ret); /* -ENOMEM */
1078 : }
1079 51814 : clean_tree_block(trans, root, buf);
1080 51814 : *last_ref = 1;
1081 : }
1082 : return 0;
1083 : }
1084 :
1085 : /*
1086 : * does the dirty work in cow of a single block. The parent block (if
1087 : * supplied) is updated to point to the new cow copy. The new buffer is marked
1088 : * dirty and returned locked. If you modify the block it needs to be marked
1089 : * dirty again.
1090 : *
1091 : * search_start -- an allocation hint for the new block
1092 : *
1093 : * empty_size -- a hint that you plan on doing more cow. This is the size in
1094 : * bytes the allocator should try to find free next to the block it returns.
1095 : * This is just a hint and may be ignored by the allocator.
1096 : */
1097 52285 : static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1098 13527 : struct btrfs_root *root,
1099 52284 : struct extent_buffer *buf,
1100 38757 : struct extent_buffer *parent, int parent_slot,
1101 : struct extent_buffer **cow_ret,
1102 : u64 search_start, u64 empty_size)
1103 : {
1104 : struct btrfs_disk_key disk_key;
1105 156840 : struct extent_buffer *cow;
1106 : int level, ret;
1107 52285 : int last_ref = 0;
1108 : int unlock_orig = 0;
1109 : u64 parent_start;
1110 :
1111 52285 : if (*cow_ret == buf)
1112 : unlock_orig = 1;
1113 :
1114 52285 : btrfs_assert_tree_locked(buf);
1115 :
1116 52283 : WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1117 : trans->transid != root->fs_info->running_transaction->transid);
1118 52284 : WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1119 : trans->transid != root->last_trans);
1120 :
1121 52284 : level = btrfs_header_level(buf);
1122 :
1123 52284 : if (level == 0)
1124 : btrfs_item_key(buf, &disk_key, 0);
1125 : else
1126 7972 : btrfs_node_key(buf, &disk_key, 0);
1127 :
1128 52284 : if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1129 18 : if (parent)
1130 16 : parent_start = parent->start;
1131 : else
1132 : parent_start = 0;
1133 : } else
1134 : parent_start = 0;
1135 :
1136 52284 : cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
1137 : root->root_key.objectid, &disk_key,
1138 : level, search_start, empty_size);
1139 52286 : if (IS_ERR(cow))
1140 0 : return PTR_ERR(cow);
1141 :
1142 : /* cow is set to blocking by btrfs_init_new_buffer */
1143 :
1144 52286 : copy_extent_buffer(cow, buf, 0, 0, cow->len);
1145 52286 : btrfs_set_header_bytenr(cow, cow->start);
1146 52286 : btrfs_set_header_generation(cow, trans->transid);
1147 : btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1148 : btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1149 : BTRFS_HEADER_FLAG_RELOC);
1150 52286 : if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1151 : btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1152 : else
1153 : btrfs_set_header_owner(cow, root->root_key.objectid);
1154 :
1155 52286 : write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(),
1156 : BTRFS_FSID_SIZE);
1157 :
1158 52284 : ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1159 52284 : if (ret) {
1160 0 : btrfs_abort_transaction(trans, root, ret);
1161 0 : return ret;
1162 : }
1163 :
1164 52284 : if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1165 22831 : ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1166 22831 : if (ret)
1167 : return ret;
1168 : }
1169 :
1170 52285 : if (buf == root->node) {
1171 13528 : WARN_ON(parent && parent != buf);
1172 27054 : if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1173 : btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1174 2 : parent_start = buf->start;
1175 : else
1176 : parent_start = 0;
1177 :
1178 : extent_buffer_get(cow);
1179 13527 : tree_mod_log_set_root_pointer(root, cow, 1);
1180 13527 : rcu_assign_pointer(root->node, cow);
1181 :
1182 13527 : btrfs_free_tree_block(trans, root, buf, parent_start,
1183 : last_ref);
1184 13528 : free_extent_buffer(buf);
1185 13528 : add_root_to_dirty_list(root);
1186 : } else {
1187 38757 : if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1188 16 : parent_start = parent->start;
1189 : else
1190 : parent_start = 0;
1191 :
1192 77514 : WARN_ON(trans->transid != btrfs_header_generation(parent));
1193 38757 : tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1194 : MOD_LOG_KEY_REPLACE, GFP_NOFS);
1195 38757 : btrfs_set_node_blockptr(parent, parent_slot,
1196 : cow->start);
1197 38757 : btrfs_set_node_ptr_generation(parent, parent_slot,
1198 : trans->transid);
1199 38756 : btrfs_mark_buffer_dirty(parent);
1200 38758 : if (last_ref) {
1201 38288 : ret = tree_mod_log_free_eb(root->fs_info, buf);
1202 38288 : if (ret) {
1203 0 : btrfs_abort_transaction(trans, root, ret);
1204 0 : return ret;
1205 : }
1206 : }
1207 38758 : btrfs_free_tree_block(trans, root, buf, parent_start,
1208 : last_ref);
1209 : }
1210 52286 : if (unlock_orig)
1211 52286 : btrfs_tree_unlock(buf);
1212 52286 : free_extent_buffer_stale(buf);
1213 52284 : btrfs_mark_buffer_dirty(cow);
1214 52285 : *cow_ret = cow;
1215 52285 : return 0;
1216 : }
1217 :
1218 : /*
1219 : * returns the logical address of the oldest predecessor of the given root.
1220 : * entries older than time_seq are ignored.
1221 : */
1222 : static struct tree_mod_elem *
1223 20675 : __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1224 : struct extent_buffer *eb_root, u64 time_seq)
1225 : {
1226 : struct tree_mod_elem *tm;
1227 : struct tree_mod_elem *found = NULL;
1228 20675 : u64 root_logical = eb_root->start;
1229 : int looped = 0;
1230 :
1231 20675 : if (!time_seq)
1232 : return NULL;
1233 :
1234 : /*
1235 : * the very last operation that's logged for a root is the replacement
1236 : * operation (if it is replaced at all). this has the index of the *new*
1237 : * root, making it the very first operation that's logged for this root.
1238 : */
1239 : while (1) {
1240 : tm = tree_mod_log_search_oldest(fs_info, root_logical,
1241 : time_seq);
1242 20677 : if (!looped && !tm)
1243 : return NULL;
1244 : /*
1245 : * if there are no tree operation for the oldest root, we simply
1246 : * return it. this should only happen if that (old) root is at
1247 : * level 0.
1248 : */
1249 49 : if (!tm)
1250 : break;
1251 :
1252 : /*
1253 : * if there's an operation that's not a root replacement, we
1254 : * found the oldest version of our root. normally, we'll find a
1255 : * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1256 : */
1257 49 : if (tm->op != MOD_LOG_ROOT_REPLACE)
1258 : break;
1259 :
1260 : found = tm;
1261 2 : root_logical = tm->old_root.logical;
1262 : looped = 1;
1263 : }
1264 :
1265 : /* if there's no old root to return, return what we found instead */
1266 47 : if (!found)
1267 : found = tm;
1268 :
1269 : return found;
1270 : }
1271 :
1272 : /*
1273 : * tm is a pointer to the first operation to rewind within eb. then, all
1274 : * previous operations will be rewinded (until we reach something older than
1275 : * time_seq).
1276 : */
1277 : static void
1278 42 : __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1279 : u64 time_seq, struct tree_mod_elem *first_tm)
1280 : {
1281 : u32 n;
1282 : struct rb_node *next;
1283 : struct tree_mod_elem *tm = first_tm;
1284 : unsigned long o_dst;
1285 : unsigned long o_src;
1286 : unsigned long p_size = sizeof(struct btrfs_key_ptr);
1287 :
1288 : n = btrfs_header_nritems(eb);
1289 : tree_mod_log_read_lock(fs_info);
1290 238 : while (tm && tm->seq >= time_seq) {
1291 : /*
1292 : * all the operations are recorded with the operator used for
1293 : * the modification. as we're going backwards, we do the
1294 : * opposite of each operation here.
1295 : */
1296 238 : switch (tm->op) {
1297 : case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1298 208 : BUG_ON(tm->slot < n);
1299 : /* Fallthrough */
1300 : case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1301 : case MOD_LOG_KEY_REMOVE:
1302 208 : btrfs_set_node_key(eb, &tm->key, tm->slot);
1303 208 : btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1304 208 : btrfs_set_node_ptr_generation(eb, tm->slot,
1305 : tm->generation);
1306 208 : n++;
1307 208 : break;
1308 : case MOD_LOG_KEY_REPLACE:
1309 27 : BUG_ON(tm->slot >= n);
1310 27 : btrfs_set_node_key(eb, &tm->key, tm->slot);
1311 27 : btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1312 27 : btrfs_set_node_ptr_generation(eb, tm->slot,
1313 : tm->generation);
1314 : break;
1315 : case MOD_LOG_KEY_ADD:
1316 : /* if a move operation is needed it's in the log */
1317 3 : n--;
1318 3 : break;
1319 : case MOD_LOG_MOVE_KEYS:
1320 0 : o_dst = btrfs_node_key_ptr_offset(tm->slot);
1321 0 : o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1322 0 : memmove_extent_buffer(eb, o_dst, o_src,
1323 0 : tm->move.nr_items * p_size);
1324 0 : break;
1325 : case MOD_LOG_ROOT_REPLACE:
1326 : /*
1327 : * this operation is special. for roots, this must be
1328 : * handled explicitly before rewinding.
1329 : * for non-roots, this operation may exist if the node
1330 : * was a root: root A -> child B; then A gets empty and
1331 : * B is promoted to the new root. in the mod log, we'll
1332 : * have a root-replace operation for B, a tree block
1333 : * that is no root. we simply ignore that operation.
1334 : */
1335 : break;
1336 : }
1337 238 : next = rb_next(&tm->node);
1338 238 : if (!next)
1339 : break;
1340 : tm = container_of(next, struct tree_mod_elem, node);
1341 217 : if (tm->index != first_tm->index)
1342 : break;
1343 : }
1344 : tree_mod_log_read_unlock(fs_info);
1345 : btrfs_set_header_nritems(eb, n);
1346 21 : }
1347 :
1348 : /*
1349 : * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1350 : * is returned. If rewind operations happen, a fresh buffer is returned. The
1351 : * returned buffer is always read-locked. If the returned buffer is not the
1352 : * input buffer, the lock on the input buffer is released and the input buffer
1353 : * is freed (its refcount is decremented).
1354 : */
1355 : static struct extent_buffer *
1356 3188 : tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1357 3188 : struct extent_buffer *eb, u64 time_seq)
1358 : {
1359 0 : struct extent_buffer *eb_rewin;
1360 : struct tree_mod_elem *tm;
1361 :
1362 3188 : if (!time_seq)
1363 : return eb;
1364 :
1365 3188 : if (btrfs_header_level(eb) == 0)
1366 : return eb;
1367 :
1368 1269 : tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1369 1269 : if (!tm)
1370 : return eb;
1371 :
1372 0 : btrfs_set_path_blocking(path);
1373 0 : btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1374 :
1375 0 : if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1376 0 : BUG_ON(tm->slot != 0);
1377 0 : eb_rewin = alloc_dummy_extent_buffer(eb->start,
1378 0 : fs_info->tree_root->nodesize);
1379 0 : if (!eb_rewin) {
1380 0 : btrfs_tree_read_unlock_blocking(eb);
1381 0 : free_extent_buffer(eb);
1382 0 : return NULL;
1383 : }
1384 0 : btrfs_set_header_bytenr(eb_rewin, eb->start);
1385 : btrfs_set_header_backref_rev(eb_rewin,
1386 : btrfs_header_backref_rev(eb));
1387 : btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1388 : btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1389 : } else {
1390 0 : eb_rewin = btrfs_clone_extent_buffer(eb);
1391 0 : if (!eb_rewin) {
1392 0 : btrfs_tree_read_unlock_blocking(eb);
1393 0 : free_extent_buffer(eb);
1394 0 : return NULL;
1395 : }
1396 : }
1397 :
1398 0 : btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK);
1399 0 : btrfs_tree_read_unlock_blocking(eb);
1400 0 : free_extent_buffer(eb);
1401 :
1402 : extent_buffer_get(eb_rewin);
1403 0 : btrfs_tree_read_lock(eb_rewin);
1404 0 : __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1405 0 : WARN_ON(btrfs_header_nritems(eb_rewin) >
1406 : BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1407 :
1408 0 : return eb_rewin;
1409 : }
1410 :
1411 : /*
1412 : * get_old_root() rewinds the state of @root's root node to the given @time_seq
1413 : * value. If there are no changes, the current root->root_node is returned. If
1414 : * anything changed in between, there's a fresh buffer allocated on which the
1415 : * rewind operations are done. In any case, the returned buffer is read locked.
1416 : * Returns NULL on error (with no locks held).
1417 : */
1418 : static inline struct extent_buffer *
1419 7947 : get_old_root(struct btrfs_root *root, u64 time_seq)
1420 : {
1421 : struct tree_mod_elem *tm;
1422 25 : struct extent_buffer *eb = NULL;
1423 7948 : struct extent_buffer *eb_root;
1424 : struct extent_buffer *old;
1425 : struct tree_mod_root *old_root = NULL;
1426 : u64 old_generation = 0;
1427 : u64 logical;
1428 : u32 blocksize;
1429 :
1430 7947 : eb_root = btrfs_read_lock_root_node(root);
1431 15894 : tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1432 7947 : if (!tm)
1433 : return eb_root;
1434 :
1435 21 : if (tm->op == MOD_LOG_ROOT_REPLACE) {
1436 1 : old_root = &tm->old_root;
1437 1 : old_generation = tm->generation;
1438 1 : logical = old_root->logical;
1439 : } else {
1440 20 : logical = eb_root->start;
1441 : }
1442 :
1443 21 : tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1444 21 : if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1445 0 : btrfs_tree_read_unlock(eb_root);
1446 0 : free_extent_buffer(eb_root);
1447 0 : blocksize = btrfs_level_size(root, old_root->level);
1448 0 : old = read_tree_block(root, logical, blocksize, 0);
1449 0 : if (WARN_ON(!old || !extent_buffer_uptodate(old))) {
1450 0 : free_extent_buffer(old);
1451 0 : btrfs_warn(root->fs_info,
1452 : "failed to read tree block %llu from get_old_root", logical);
1453 : } else {
1454 0 : eb = btrfs_clone_extent_buffer(old);
1455 0 : free_extent_buffer(old);
1456 : }
1457 21 : } else if (old_root) {
1458 1 : btrfs_tree_read_unlock(eb_root);
1459 1 : free_extent_buffer(eb_root);
1460 1 : eb = alloc_dummy_extent_buffer(logical, root->nodesize);
1461 : } else {
1462 20 : btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK);
1463 20 : eb = btrfs_clone_extent_buffer(eb_root);
1464 20 : btrfs_tree_read_unlock_blocking(eb_root);
1465 20 : free_extent_buffer(eb_root);
1466 : }
1467 :
1468 21 : if (!eb)
1469 : return NULL;
1470 : extent_buffer_get(eb);
1471 21 : btrfs_tree_read_lock(eb);
1472 21 : if (old_root) {
1473 1 : btrfs_set_header_bytenr(eb, eb->start);
1474 : btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1475 : btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1476 1 : btrfs_set_header_level(eb, old_root->level);
1477 : btrfs_set_header_generation(eb, old_generation);
1478 : }
1479 21 : if (tm)
1480 21 : __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm);
1481 : else
1482 0 : WARN_ON(btrfs_header_level(eb) != 0);
1483 21 : WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1484 :
1485 21 : return eb;
1486 : }
1487 :
1488 12728 : int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1489 : {
1490 : struct tree_mod_elem *tm;
1491 : int level;
1492 38183 : struct extent_buffer *eb_root = btrfs_root_node(root);
1493 :
1494 25456 : tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1495 12728 : if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1496 1 : level = tm->old_root.level;
1497 : } else {
1498 12727 : level = btrfs_header_level(eb_root);
1499 : }
1500 12728 : free_extent_buffer(eb_root);
1501 :
1502 12728 : return level;
1503 : }
1504 :
1505 1600370 : static inline int should_cow_block(struct btrfs_trans_handle *trans,
1506 : struct btrfs_root *root,
1507 1600379 : struct extent_buffer *buf)
1508 : {
1509 : #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1510 : if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
1511 : return 0;
1512 : #endif
1513 : /* ensure we can see the force_cow */
1514 1600370 : smp_rmb();
1515 :
1516 : /*
1517 : * We do not need to cow a block if
1518 : * 1) this block is not created or changed in this transaction;
1519 : * 2) this block does not belong to TREE_RELOC tree;
1520 : * 3) the root is not forced COW.
1521 : *
1522 : * What is forced COW:
1523 : * when we create snapshot during commiting the transaction,
1524 : * after we've finished coping src root, we must COW the shared
1525 : * block to ensure the metadata consistency.
1526 : */
1527 3119011 : if (btrfs_header_generation(buf) == trans->transid &&
1528 1490830 : !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1529 2981638 : !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1530 1490838 : btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1531 : !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1532 : return 0;
1533 : return 1;
1534 : }
1535 :
1536 : /*
1537 : * cows a single block, see __btrfs_cow_block for the real work.
1538 : * This version of it has extra checks so that a block isn't cow'd more than
1539 : * once per transaction, as long as it hasn't been written yet
1540 : */
1541 66091 : noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1542 : struct btrfs_root *root, struct extent_buffer *buf,
1543 : struct extent_buffer *parent, int parent_slot,
1544 : struct extent_buffer **cow_ret)
1545 : {
1546 : u64 search_start;
1547 : int ret;
1548 :
1549 66091 : if (trans->transaction != root->fs_info->running_transaction)
1550 0 : WARN(1, KERN_CRIT "trans %llu running %llu\n",
1551 : trans->transid,
1552 : root->fs_info->running_transaction->transid);
1553 :
1554 66091 : if (trans->transid != root->fs_info->generation)
1555 0 : WARN(1, KERN_CRIT "trans %llu running %llu\n",
1556 : trans->transid, root->fs_info->generation);
1557 :
1558 66091 : if (!should_cow_block(trans, root, buf)) {
1559 13805 : *cow_ret = buf;
1560 13805 : return 0;
1561 : }
1562 :
1563 52286 : search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
1564 :
1565 52286 : if (parent)
1566 : btrfs_set_lock_blocking(parent);
1567 : btrfs_set_lock_blocking(buf);
1568 :
1569 52285 : ret = __btrfs_cow_block(trans, root, buf, parent,
1570 : parent_slot, cow_ret, search_start, 0);
1571 :
1572 52284 : trace_btrfs_cow_block(root, buf, *cow_ret);
1573 :
1574 52284 : return ret;
1575 : }
1576 :
1577 : /*
1578 : * helper function for defrag to decide if two blocks pointed to by a
1579 : * node are actually close by
1580 : */
1581 : static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1582 : {
1583 0 : if (blocknr < other && other - (blocknr + blocksize) < 32768)
1584 : return 1;
1585 0 : if (blocknr > other && blocknr - (other + blocksize) < 32768)
1586 : return 1;
1587 : return 0;
1588 : }
1589 :
1590 : /*
1591 : * compare two keys in a memcmp fashion
1592 : */
1593 : static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1594 : {
1595 : struct btrfs_key k1;
1596 :
1597 : btrfs_disk_key_to_cpu(&k1, disk);
1598 :
1599 : return btrfs_comp_cpu_keys(&k1, k2);
1600 : }
1601 :
1602 : /*
1603 : * same as comp_keys only with two btrfs_key's
1604 : */
1605 275675 : int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1606 : {
1607 28437019 : if (k1->objectid > k2->objectid)
1608 : return 1;
1609 20136305 : if (k1->objectid < k2->objectid)
1610 : return -1;
1611 12375604 : if (k1->type > k2->type)
1612 : return 1;
1613 11863499 : if (k1->type < k2->type)
1614 : return -1;
1615 11495661 : if (k1->offset > k2->offset)
1616 : return 1;
1617 7012757 : if (k1->offset < k2->offset)
1618 : return -1;
1619 6644 : return 0;
1620 : }
1621 :
1622 : /*
1623 : * this is used by the defrag code to go through all the
1624 : * leaves pointed to by a node and reallocate them so that
1625 : * disk order is close to key order
1626 : */
1627 0 : int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1628 0 : struct btrfs_root *root, struct extent_buffer *parent,
1629 : int start_slot, u64 *last_ret,
1630 : struct btrfs_key *progress)
1631 : {
1632 : struct extent_buffer *cur;
1633 : u64 blocknr;
1634 : u64 gen;
1635 0 : u64 search_start = *last_ret;
1636 : u64 last_block = 0;
1637 : u64 other;
1638 : u32 parent_nritems;
1639 : int end_slot;
1640 : int i;
1641 : int err = 0;
1642 : int parent_level;
1643 : int uptodate;
1644 : u32 blocksize;
1645 : int progress_passed = 0;
1646 : struct btrfs_disk_key disk_key;
1647 :
1648 : parent_level = btrfs_header_level(parent);
1649 :
1650 0 : WARN_ON(trans->transaction != root->fs_info->running_transaction);
1651 0 : WARN_ON(trans->transid != root->fs_info->generation);
1652 :
1653 : parent_nritems = btrfs_header_nritems(parent);
1654 : blocksize = btrfs_level_size(root, parent_level - 1);
1655 0 : end_slot = parent_nritems;
1656 :
1657 0 : if (parent_nritems == 1)
1658 : return 0;
1659 :
1660 : btrfs_set_lock_blocking(parent);
1661 :
1662 0 : for (i = start_slot; i < end_slot; i++) {
1663 : int close = 1;
1664 :
1665 0 : btrfs_node_key(parent, &disk_key, i);
1666 0 : if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1667 0 : continue;
1668 :
1669 : progress_passed = 1;
1670 : blocknr = btrfs_node_blockptr(parent, i);
1671 : gen = btrfs_node_ptr_generation(parent, i);
1672 0 : if (last_block == 0)
1673 : last_block = blocknr;
1674 :
1675 0 : if (i > 0) {
1676 0 : other = btrfs_node_blockptr(parent, i - 1);
1677 : close = close_blocks(blocknr, other, blocksize);
1678 : }
1679 0 : if (!close && i < end_slot - 2) {
1680 0 : other = btrfs_node_blockptr(parent, i + 1);
1681 : close = close_blocks(blocknr, other, blocksize);
1682 : }
1683 0 : if (close) {
1684 : last_block = blocknr;
1685 0 : continue;
1686 : }
1687 :
1688 0 : cur = btrfs_find_tree_block(root, blocknr, blocksize);
1689 0 : if (cur)
1690 0 : uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1691 : else
1692 : uptodate = 0;
1693 0 : if (!cur || !uptodate) {
1694 0 : if (!cur) {
1695 0 : cur = read_tree_block(root, blocknr,
1696 : blocksize, gen);
1697 0 : if (!cur || !extent_buffer_uptodate(cur)) {
1698 0 : free_extent_buffer(cur);
1699 0 : return -EIO;
1700 : }
1701 0 : } else if (!uptodate) {
1702 0 : err = btrfs_read_buffer(cur, gen);
1703 0 : if (err) {
1704 0 : free_extent_buffer(cur);
1705 0 : return err;
1706 : }
1707 : }
1708 : }
1709 0 : if (search_start == 0)
1710 : search_start = last_block;
1711 :
1712 0 : btrfs_tree_lock(cur);
1713 0 : btrfs_set_lock_blocking(cur);
1714 0 : err = __btrfs_cow_block(trans, root, cur, parent, i,
1715 : &cur, search_start,
1716 0 : min(16 * blocksize,
1717 : (end_slot - i) * blocksize));
1718 0 : if (err) {
1719 0 : btrfs_tree_unlock(cur);
1720 0 : free_extent_buffer(cur);
1721 0 : break;
1722 : }
1723 0 : search_start = cur->start;
1724 : last_block = cur->start;
1725 0 : *last_ret = search_start;
1726 0 : btrfs_tree_unlock(cur);
1727 0 : free_extent_buffer(cur);
1728 : }
1729 0 : 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 524875 : static inline unsigned int leaf_data_end(struct btrfs_root *root,
1738 524875 : struct extent_buffer *leaf)
1739 : {
1740 : u32 nr = btrfs_header_nritems(leaf);
1741 524875 : if (nr == 0)
1742 1779 : return BTRFS_LEAF_DATA_SIZE(root);
1743 523096 : 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 4978020 : 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 4978020 : char *kaddr = NULL;
1770 4978020 : unsigned long map_start = 0;
1771 4978020 : unsigned long map_len = 0;
1772 : int err;
1773 :
1774 37022947 : while (low < high) {
1775 28139641 : mid = (low + high) / 2;
1776 28139641 : offset = p + mid * item_size;
1777 :
1778 51123818 : if (!kaddr || offset < map_start ||
1779 22984177 : (offset + sizeof(struct btrfs_disk_key)) >
1780 22984177 : map_start + map_len) {
1781 :
1782 5503721 : err = map_private_extent_buffer(eb, offset,
1783 : sizeof(struct btrfs_disk_key),
1784 : &kaddr, &map_start, &map_len);
1785 :
1786 5515117 : if (!err) {
1787 5476843 : tmp = (struct btrfs_disk_key *)(kaddr + offset -
1788 : map_start);
1789 : } else {
1790 38274 : read_extent_buffer(eb, &unaligned,
1791 : offset, sizeof(unaligned));
1792 : tmp = &unaligned;
1793 : }
1794 :
1795 : } else {
1796 22635920 : tmp = (struct btrfs_disk_key *)(kaddr + offset -
1797 : map_start);
1798 : }
1799 : ret = comp_keys(tmp, key);
1800 :
1801 28145279 : if (ret < 0)
1802 13819832 : low = mid + 1;
1803 14325447 : else if (ret > 0)
1804 : high = mid;
1805 : else {
1806 1078372 : *slot = mid;
1807 1078372 : return 0;
1808 : }
1809 : }
1810 3905286 : *slot = low;
1811 3905286 : return 1;
1812 : }
1813 :
1814 : /*
1815 : * simple bin_search frontend that does the right thing for
1816 : * leaves vs nodes
1817 : */
1818 9970782 : static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1819 : int level, int *slot)
1820 : {
1821 4985391 : if (level == 0)
1822 2131536 : 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 2853855 : 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 247 : int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1836 : int level, int *slot)
1837 : {
1838 247 : 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 5411 : 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 1403 : 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 43874 : static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1862 21977 : struct extent_buffer *parent, int slot)
1863 : {
1864 21977 : int level = btrfs_header_level(parent);
1865 : struct extent_buffer *eb;
1866 :
1867 21977 : if (slot < 0)
1868 : return NULL;
1869 43826 : if (slot >= btrfs_header_nritems(parent))
1870 : return NULL;
1871 :
1872 21897 : BUG_ON(level == 0);
1873 :
1874 21897 : eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1875 : btrfs_level_size(root, level - 1),
1876 : btrfs_node_ptr_generation(parent, slot));
1877 21897 : if (eb && !extent_buffer_uptodate(eb)) {
1878 0 : free_extent_buffer(eb);
1879 : eb = NULL;
1880 : }
1881 :
1882 21897 : 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 190233 : static noinline int balance_level(struct btrfs_trans_handle *trans,
1891 32 : struct btrfs_root *root,
1892 : struct btrfs_path *path, int level)
1893 : {
1894 190233 : struct extent_buffer *right = NULL;
1895 380487 : struct extent_buffer *mid;
1896 190233 : struct extent_buffer *left = NULL;
1897 : struct extent_buffer *parent = NULL;
1898 : int ret = 0;
1899 : int wret;
1900 : int pslot;
1901 190233 : int orig_slot = path->slots[level];
1902 : u64 orig_ptr;
1903 :
1904 190233 : if (level == 0)
1905 : return 0;
1906 :
1907 190233 : mid = path->nodes[level];
1908 :
1909 190233 : WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1910 : path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1911 190233 : WARN_ON(btrfs_header_generation(mid) != trans->transid);
1912 :
1913 : orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1914 :
1915 190232 : if (level < BTRFS_MAX_LEVEL - 1) {
1916 190233 : parent = path->nodes[level + 1];
1917 190233 : 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 190232 : if (!parent) {
1925 : struct extent_buffer *child;
1926 :
1927 178923 : if (btrfs_header_nritems(mid) != 1)
1928 178923 : return 0;
1929 :
1930 : /* promote the child to a root */
1931 32 : child = read_node_slot(root, mid, 0);
1932 32 : if (!child) {
1933 : ret = -EROFS;
1934 0 : btrfs_std_error(root->fs_info, ret);
1935 0 : goto enospc;
1936 : }
1937 :
1938 32 : btrfs_tree_lock(child);
1939 32 : btrfs_set_lock_blocking(child);
1940 32 : ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1941 32 : if (ret) {
1942 0 : btrfs_tree_unlock(child);
1943 0 : free_extent_buffer(child);
1944 0 : goto enospc;
1945 : }
1946 :
1947 64 : tree_mod_log_set_root_pointer(root, child, 1);
1948 32 : rcu_assign_pointer(root->node, child);
1949 :
1950 32 : add_root_to_dirty_list(root);
1951 32 : btrfs_tree_unlock(child);
1952 :
1953 32 : path->locks[level] = 0;
1954 32 : path->nodes[level] = NULL;
1955 32 : clean_tree_block(trans, root, mid);
1956 32 : btrfs_tree_unlock(mid);
1957 : /* once for the path */
1958 32 : free_extent_buffer(mid);
1959 :
1960 32 : root_sub_used(root, mid->len);
1961 32 : btrfs_free_tree_block(trans, root, mid, 0, 1);
1962 : /* once for the root ptr */
1963 32 : free_extent_buffer_stale(mid);
1964 32 : return 0;
1965 : }
1966 22618 : if (btrfs_header_nritems(mid) >
1967 11309 : BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1968 : return 0;
1969 :
1970 11 : left = read_node_slot(root, parent, pslot - 1);
1971 27 : if (left) {
1972 8 : btrfs_tree_lock(left);
1973 8 : btrfs_set_lock_blocking(left);
1974 8 : wret = btrfs_cow_block(trans, root, left,
1975 : parent, pslot - 1, &left);
1976 8 : if (wret) {
1977 : ret = wret;
1978 : goto enospc;
1979 : }
1980 : }
1981 11 : right = read_node_slot(root, parent, pslot + 1);
1982 20 : if (right) {
1983 9 : btrfs_tree_lock(right);
1984 9 : btrfs_set_lock_blocking(right);
1985 9 : wret = btrfs_cow_block(trans, root, right,
1986 : parent, pslot + 1, &right);
1987 9 : if (wret) {
1988 : ret = wret;
1989 : goto enospc;
1990 : }
1991 : }
1992 :
1993 : /* first, try to make some room in the middle buffer */
1994 11 : if (left) {
1995 8 : orig_slot += btrfs_header_nritems(left);
1996 8 : wret = push_node_left(trans, root, left, mid, 1);
1997 8 : if (wret < 0)
1998 : ret = wret;
1999 : }
2000 :
2001 : /*
2002 : * then try to empty the right most buffer into the middle
2003 : */
2004 11 : if (right) {
2005 9 : wret = push_node_left(trans, root, mid, right, 1);
2006 9 : if (wret < 0 && wret != -ENOSPC)
2007 : ret = wret;
2008 18 : if (btrfs_header_nritems(right) == 0) {
2009 9 : clean_tree_block(trans, root, right);
2010 9 : btrfs_tree_unlock(right);
2011 9 : del_ptr(root, path, level + 1, pslot + 1);
2012 9 : root_sub_used(root, right->len);
2013 9 : btrfs_free_tree_block(trans, root, right, 0, 1);
2014 9 : free_extent_buffer_stale(right);
2015 9 : right = NULL;
2016 : } else {
2017 : struct btrfs_disk_key right_key;
2018 0 : btrfs_node_key(right, &right_key, 0);
2019 0 : 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 0 : btrfs_mark_buffer_dirty(parent);
2023 : }
2024 : }
2025 11 : 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 0 : if (!left) {
2036 : ret = -EROFS;
2037 0 : btrfs_std_error(root->fs_info, ret);
2038 : goto enospc;
2039 : }
2040 0 : wret = balance_node_right(trans, root, mid, left);
2041 0 : if (wret < 0) {
2042 : ret = wret;
2043 : goto enospc;
2044 : }
2045 0 : if (wret == 1) {
2046 0 : wret = push_node_left(trans, root, left, mid, 1);
2047 0 : if (wret < 0)
2048 : ret = wret;
2049 : }
2050 0 : BUG_ON(wret == 1);
2051 : }
2052 11 : if (btrfs_header_nritems(mid) == 0) {
2053 2 : clean_tree_block(trans, root, mid);
2054 2 : btrfs_tree_unlock(mid);
2055 2 : del_ptr(root, path, level + 1, pslot);
2056 2 : root_sub_used(root, mid->len);
2057 2 : btrfs_free_tree_block(trans, root, mid, 0, 1);
2058 2 : 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 9 : btrfs_node_key(mid, &mid_key, 0);
2064 9 : tree_mod_log_set_node_key(root->fs_info, parent,
2065 : pslot, 0);
2066 : btrfs_set_node_key(parent, &mid_key, pslot);
2067 9 : btrfs_mark_buffer_dirty(parent);
2068 : }
2069 :
2070 : /* update the path */
2071 11 : if (left) {
2072 8 : if (btrfs_header_nritems(left) > orig_slot) {
2073 : extent_buffer_get(left);
2074 : /* left was locked after cow */
2075 8 : path->nodes[level] = left;
2076 8 : path->slots[level + 1] -= 1;
2077 8 : path->slots[level] = orig_slot;
2078 8 : if (mid) {
2079 6 : btrfs_tree_unlock(mid);
2080 6 : free_extent_buffer(mid);
2081 : }
2082 : } else {
2083 0 : orig_slot -= btrfs_header_nritems(left);
2084 0 : path->slots[level] = orig_slot;
2085 : }
2086 : }
2087 : /* double check we haven't messed things up */
2088 11 : if (orig_ptr !=
2089 11 : btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2090 0 : BUG();
2091 : enospc:
2092 11 : if (right) {
2093 0 : btrfs_tree_unlock(right);
2094 0 : free_extent_buffer(right);
2095 : }
2096 11 : if (left) {
2097 8 : if (path->nodes[level] != left)
2098 0 : btrfs_tree_unlock(left);
2099 8 : free_extent_buffer(left);
2100 : }
2101 11 : 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 177 : 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 177 : struct extent_buffer *right = NULL;
2113 247 : struct extent_buffer *mid;
2114 177 : struct extent_buffer *left = NULL;
2115 : struct extent_buffer *parent = NULL;
2116 : int ret = 0;
2117 : int wret;
2118 : int pslot;
2119 177 : int orig_slot = path->slots[level];
2120 :
2121 177 : if (level == 0)
2122 : return 1;
2123 :
2124 177 : mid = path->nodes[level];
2125 177 : WARN_ON(btrfs_header_generation(mid) != trans->transid);
2126 :
2127 177 : if (level < BTRFS_MAX_LEVEL - 1) {
2128 177 : parent = path->nodes[level + 1];
2129 177 : pslot = path->slots[level + 1];
2130 : }
2131 :
2132 177 : if (!parent)
2133 : return 1;
2134 :
2135 177 : left = read_node_slot(root, parent, pslot - 1);
2136 :
2137 : /* first, try to make some room in the middle buffer */
2138 390 : if (left) {
2139 : u32 left_nr;
2140 :
2141 116 : btrfs_tree_lock(left);
2142 116 : btrfs_set_lock_blocking(left);
2143 :
2144 116 : left_nr = btrfs_header_nritems(left);
2145 116 : if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2146 : wret = 1;
2147 : } else {
2148 97 : ret = btrfs_cow_block(trans, root, left, parent,
2149 : pslot - 1, &left);
2150 97 : if (ret)
2151 : wret = 1;
2152 : else {
2153 97 : wret = push_node_left(trans, root,
2154 : left, mid, 0);
2155 : }
2156 : }
2157 : if (wret < 0)
2158 : ret = wret;
2159 116 : if (wret == 0) {
2160 : struct btrfs_disk_key disk_key;
2161 97 : orig_slot += left_nr;
2162 97 : btrfs_node_key(mid, &disk_key, 0);
2163 97 : tree_mod_log_set_node_key(root->fs_info, parent,
2164 : pslot, 0);
2165 : btrfs_set_node_key(parent, &disk_key, pslot);
2166 97 : btrfs_mark_buffer_dirty(parent);
2167 194 : if (btrfs_header_nritems(left) > orig_slot) {
2168 9 : path->nodes[level] = left;
2169 9 : path->slots[level + 1] -= 1;
2170 9 : path->slots[level] = orig_slot;
2171 9 : btrfs_tree_unlock(mid);
2172 9 : free_extent_buffer(mid);
2173 : } else {
2174 88 : orig_slot -=
2175 : btrfs_header_nritems(left);
2176 88 : path->slots[level] = orig_slot;
2177 88 : btrfs_tree_unlock(left);
2178 88 : free_extent_buffer(left);
2179 : }
2180 : return 0;
2181 : }
2182 19 : btrfs_tree_unlock(left);
2183 19 : free_extent_buffer(left);
2184 : }
2185 80 : right = read_node_slot(root, parent, pslot + 1);
2186 :
2187 : /*
2188 : * then try to empty the right most buffer into the middle
2189 : */
2190 146 : if (right) {
2191 : u32 right_nr;
2192 :
2193 66 : btrfs_tree_lock(right);
2194 66 : btrfs_set_lock_blocking(right);
2195 :
2196 66 : right_nr = btrfs_header_nritems(right);
2197 66 : if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2198 : wret = 1;
2199 : } else {
2200 65 : ret = btrfs_cow_block(trans, root, right,
2201 : parent, pslot + 1,
2202 : &right);
2203 65 : if (ret)
2204 : wret = 1;
2205 : else {
2206 65 : wret = balance_node_right(trans, root,
2207 : right, mid);
2208 : }
2209 : }
2210 : if (wret < 0)
2211 : ret = wret;
2212 66 : if (wret == 0) {
2213 : struct btrfs_disk_key disk_key;
2214 :
2215 65 : btrfs_node_key(right, &disk_key, 0);
2216 65 : 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 65 : btrfs_mark_buffer_dirty(parent);
2220 :
2221 65 : if (btrfs_header_nritems(mid) <= orig_slot) {
2222 5 : path->nodes[level] = right;
2223 5 : path->slots[level + 1] += 1;
2224 5 : path->slots[level] = orig_slot -
2225 : btrfs_header_nritems(mid);
2226 5 : btrfs_tree_unlock(mid);
2227 5 : free_extent_buffer(mid);
2228 : } else {
2229 60 : btrfs_tree_unlock(right);
2230 60 : free_extent_buffer(right);
2231 : }
2232 : return 0;
2233 : }
2234 1 : btrfs_tree_unlock(right);
2235 1 : 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 230 : static void reada_for_search(struct btrfs_root *root,
2245 : struct btrfs_path *path,
2246 : int level, int slot, u64 objectid)
2247 : {
2248 114 : 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 116 : int direction = path->reada;
2256 : struct extent_buffer *eb;
2257 : u32 nr;
2258 : u32 blocksize;
2259 : u32 nscan = 0;
2260 :
2261 116 : if (level != 1)
2262 2 : return;
2263 :
2264 114 : if (!path->nodes[level])
2265 : return;
2266 :
2267 : node = path->nodes[level];
2268 :
2269 : search = btrfs_node_blockptr(node, slot);
2270 : blocksize = btrfs_level_size(root, level - 1);
2271 114 : eb = btrfs_find_tree_block(root, search, blocksize);
2272 114 : if (eb) {
2273 0 : free_extent_buffer(eb);
2274 0 : return;
2275 : }
2276 :
2277 : target = search;
2278 :
2279 : nritems = btrfs_header_nritems(node);
2280 114 : nr = slot;
2281 :
2282 : while (1) {
2283 1308 : if (direction < 0) {
2284 37 : if (nr == 0)
2285 : break;
2286 37 : nr--;
2287 1271 : } else if (direction > 0) {
2288 1271 : nr++;
2289 1271 : if (nr >= nritems)
2290 : break;
2291 : }
2292 1254 : if (path->reada < 0 && objectid) {
2293 37 : btrfs_node_key(node, &disk_key, nr);
2294 37 : if (btrfs_disk_key_objectid(&disk_key) != objectid)
2295 : break;
2296 : }
2297 1217 : search = btrfs_node_blockptr(node, nr);
2298 1217 : if ((search <= target && target - search <= 65536) ||
2299 744 : (search > target && search - target <= 65536)) {
2300 : gen = btrfs_node_ptr_generation(node, nr);
2301 142 : readahead_tree_block(root, search, blocksize, gen);
2302 142 : nread += blocksize;
2303 : }
2304 1217 : nscan++;
2305 1217 : if ((nread > 65536 || nscan > 32))
2306 : break;
2307 : }
2308 : }
2309 :
2310 201896 : static noinline void reada_for_balance(struct btrfs_root *root,
2311 : struct btrfs_path *path, int level)
2312 : {
2313 : int slot;
2314 : int nritems;
2315 11486 : struct extent_buffer *parent;
2316 : struct extent_buffer *eb;
2317 : u64 gen;
2318 : u64 block1 = 0;
2319 : u64 block2 = 0;
2320 : int blocksize;
2321 :
2322 190410 : parent = path->nodes[level + 1];
2323 190410 : if (!parent)
2324 190410 : return;
2325 :
2326 11486 : nritems = btrfs_header_nritems(parent);
2327 11486 : slot = path->slots[level + 1];
2328 : blocksize = btrfs_level_size(root, level);
2329 :
2330 11486 : if (slot > 0) {
2331 9446 : block1 = btrfs_node_blockptr(parent, slot - 1);
2332 : gen = btrfs_node_ptr_generation(parent, slot - 1);
2333 9446 : eb = btrfs_find_tree_block(root, block1, blocksize);
2334 : /*
2335 : * if we get -eagain from btrfs_buffer_uptodate, we
2336 : * don't want to return eagain here. That will loop
2337 : * forever
2338 : */
2339 9446 : if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2340 : block1 = 0;
2341 9446 : free_extent_buffer(eb);
2342 : }
2343 11486 : if (slot + 1 < nritems) {
2344 : block2 = btrfs_node_blockptr(parent, slot + 1);
2345 : gen = btrfs_node_ptr_generation(parent, slot + 1);
2346 7098 : eb = btrfs_find_tree_block(root, block2, blocksize);
2347 7098 : if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2348 : block2 = 0;
2349 7098 : free_extent_buffer(eb);
2350 : }
2351 :
2352 11486 : if (block1)
2353 0 : readahead_tree_block(root, block1, blocksize, 0);
2354 11486 : if (block2)
2355 0 : readahead_tree_block(root, block2, blocksize, 0);
2356 : }
2357 :
2358 :
2359 : /*
2360 : * when we walk down the tree, it is usually safe to unlock the higher layers
2361 : * in the tree. The exceptions are when our path goes through slot 0, because
2362 : * operations on the tree might require changing key pointers higher up in the
2363 : * tree.
2364 : *
2365 : * callers might also have set path->keep_locks, which tells this code to keep
2366 : * the lock if the path points to the last slot in the block. This is part of
2367 : * walking through the tree, and selecting the next slot in the higher block.
2368 : *
2369 : * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2370 : * if lowest_unlock is 1, level 0 won't be unlocked
2371 : */
2372 4948754 : static noinline void unlock_up(struct btrfs_path *path, int level,
2373 : int lowest_unlock, int min_write_lock_level,
2374 : int *write_lock_level)
2375 : {
2376 : int i;
2377 : int skip_level = level;
2378 : int no_skips = 0;
2379 105520 : struct extent_buffer *t;
2380 :
2381 7861945 : for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2382 7860331 : if (!path->nodes[i])
2383 : break;
2384 6213724 : if (!path->locks[i])
2385 : break;
2386 2913116 : if (!no_skips && path->slots[i] == 0) {
2387 262583 : skip_level = i + 1;
2388 262583 : continue;
2389 : }
2390 2650533 : if (!no_skips && path->keep_locks) {
2391 : u32 nritems;
2392 : t = path->nodes[i];
2393 : nritems = btrfs_header_nritems(t);
2394 105520 : if (nritems < 1 || path->slots[i] >= nritems - 1) {
2395 55929 : skip_level = i + 1;
2396 55929 : continue;
2397 : }
2398 : }
2399 2594604 : if (skip_level < i && i >= lowest_unlock)
2400 : no_skips = 1;
2401 :
2402 : t = path->nodes[i];
2403 2594604 : if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2404 345066 : btrfs_tree_unlock_rw(t, path->locks[i]);
2405 345141 : path->locks[i] = 0;
2406 690282 : if (write_lock_level &&
2407 392278 : i > min_write_lock_level &&
2408 47137 : i <= *write_lock_level) {
2409 21259 : *write_lock_level = i - 1;
2410 : }
2411 : }
2412 : }
2413 4948829 : }
2414 :
2415 : /*
2416 : * This releases any locks held in the path starting at level and
2417 : * going all the way up to the root.
2418 : *
2419 : * btrfs_search_slot will keep the lock held on higher nodes in a few
2420 : * corner cases, such as COW of the block at slot zero in the node. This
2421 : * ignores those rules, and it should only be called when there are no
2422 : * more updates to be done higher up in the tree.
2423 : */
2424 295102 : noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2425 : {
2426 : int i;
2427 :
2428 295102 : if (path->keep_locks)
2429 295296 : return;
2430 :
2431 2054782 : for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2432 2054588 : if (!path->nodes[i])
2433 1749008 : continue;
2434 305580 : if (!path->locks[i])
2435 239717 : continue;
2436 65863 : btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2437 66057 : path->locks[i] = 0;
2438 : }
2439 : }
2440 :
2441 : /*
2442 : * helper function for btrfs_search_slot. The goal is to find a block
2443 : * in cache without setting the path to blocking. If we find the block
2444 : * we return zero and the path is unchanged.
2445 : *
2446 : * If we can't find the block, we set the path blocking and do some
2447 : * reada. -EAGAIN is returned and the search must be repeated.
2448 : */
2449 : static int
2450 2208274 : read_block_for_search(struct btrfs_trans_handle *trans,
2451 2207946 : struct btrfs_root *root, struct btrfs_path *p,
2452 : struct extent_buffer **eb_ret, int level, int slot,
2453 : struct btrfs_key *key, u64 time_seq)
2454 : {
2455 : u64 blocknr;
2456 : u64 gen;
2457 : u32 blocksize;
2458 2208274 : struct extent_buffer *b = *eb_ret;
2459 : struct extent_buffer *tmp;
2460 : int ret;
2461 :
2462 : blocknr = btrfs_node_blockptr(b, slot);
2463 : gen = btrfs_node_ptr_generation(b, slot);
2464 : blocksize = btrfs_level_size(root, level - 1);
2465 :
2466 2207946 : tmp = btrfs_find_tree_block(root, blocknr, blocksize);
2467 2208699 : if (tmp) {
2468 : /* first we do an atomic uptodate check */
2469 2208304 : if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2470 2208130 : *eb_ret = tmp;
2471 : return 0;
2472 : }
2473 :
2474 : /* the pages were up to date, but we failed
2475 : * the generation number check. Do a full
2476 : * read for the generation number that is correct.
2477 : * We must do this without dropping locks so
2478 : * we can trust our generation number
2479 : */
2480 3 : btrfs_set_path_blocking(p);
2481 :
2482 : /* now we're allowed to do a blocking uptodate check */
2483 3 : ret = btrfs_read_buffer(tmp, gen);
2484 3 : if (!ret) {
2485 3 : *eb_ret = tmp;
2486 : return 0;
2487 : }
2488 0 : free_extent_buffer(tmp);
2489 0 : btrfs_release_path(p);
2490 : return -EIO;
2491 : }
2492 :
2493 : /*
2494 : * reduce lock contention at high levels
2495 : * of the btree by dropping locks before
2496 : * we read. Don't release the lock on the current
2497 : * level because we need to walk this node to figure
2498 : * out which blocks to read.
2499 : */
2500 395 : btrfs_unlock_up_safe(p, level + 1);
2501 395 : btrfs_set_path_blocking(p);
2502 :
2503 395 : free_extent_buffer(tmp);
2504 395 : if (p->reada)
2505 116 : reada_for_search(root, p, level, slot, key->objectid);
2506 :
2507 395 : btrfs_release_path(p);
2508 :
2509 : ret = -EAGAIN;
2510 395 : tmp = read_tree_block(root, blocknr, blocksize, 0);
2511 395 : if (tmp) {
2512 : /*
2513 : * If the read above didn't mark this buffer up to date,
2514 : * it will never end up being up to date. Set ret to EIO now
2515 : * and give up so that our caller doesn't loop forever
2516 : * on our EAGAINs.
2517 : */
2518 395 : if (!btrfs_buffer_uptodate(tmp, 0, 0))
2519 : ret = -EIO;
2520 395 : free_extent_buffer(tmp);
2521 : }
2522 : return ret;
2523 : }
2524 :
2525 : /*
2526 : * helper function for btrfs_search_slot. This does all of the checks
2527 : * for node-level blocks and does any balancing required based on
2528 : * the ins_len.
2529 : *
2530 : * If no extra work was required, zero is returned. If we had to
2531 : * drop the path, -EAGAIN is returned and btrfs_search_slot must
2532 : * start over
2533 : */
2534 : static int
2535 2844551 : setup_nodes_for_search(struct btrfs_trans_handle *trans,
2536 : struct btrfs_root *root, struct btrfs_path *p,
2537 833904 : struct extent_buffer *b, int level, int ins_len,
2538 : int *write_lock_level)
2539 : {
2540 : int ret;
2541 3528429 : if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2542 341939 : BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2543 : int sret;
2544 :
2545 299 : if (*write_lock_level < level + 1) {
2546 119 : *write_lock_level = level + 1;
2547 119 : btrfs_release_path(p);
2548 119 : goto again;
2549 : }
2550 :
2551 180 : btrfs_set_path_blocking(p);
2552 180 : reada_for_balance(root, p, level);
2553 180 : sret = split_node(trans, root, p, level);
2554 180 : btrfs_clear_path_blocking(p, NULL, 0);
2555 :
2556 180 : BUG_ON(sret > 0);
2557 180 : if (sret) {
2558 : ret = sret;
2559 : goto done;
2560 : }
2561 : b = p->nodes[level];
2562 3447782 : } else if (ins_len < 0 && btrfs_header_nritems(b) <
2563 301765 : BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2564 : int sret;
2565 :
2566 250402 : if (*write_lock_level < level + 1) {
2567 60182 : *write_lock_level = level + 1;
2568 60182 : btrfs_release_path(p);
2569 60182 : goto again;
2570 : }
2571 :
2572 190220 : btrfs_set_path_blocking(p);
2573 190230 : reada_for_balance(root, p, level);
2574 190233 : sret = balance_level(trans, root, p, level);
2575 190232 : btrfs_clear_path_blocking(p, NULL, 0);
2576 :
2577 190231 : if (sret) {
2578 : ret = sret;
2579 : goto done;
2580 : }
2581 190232 : b = p->nodes[level];
2582 190232 : if (!b) {
2583 32 : btrfs_release_path(p);
2584 32 : goto again;
2585 : }
2586 190200 : BUG_ON(btrfs_header_nritems(b) == 1);
2587 : }
2588 : return 0;
2589 :
2590 : again:
2591 : ret = -EAGAIN;
2592 : done:
2593 60333 : return ret;
2594 : }
2595 :
2596 : static void key_search_validate(struct extent_buffer *b,
2597 : struct btrfs_key *key,
2598 : int level)
2599 : {
2600 : #ifdef CONFIG_BTRFS_ASSERT
2601 : struct btrfs_disk_key disk_key;
2602 :
2603 : btrfs_cpu_key_to_disk(&disk_key, key);
2604 :
2605 : if (level == 0)
2606 : ASSERT(!memcmp_extent_buffer(b, &disk_key,
2607 : offsetof(struct btrfs_leaf, items[0].key),
2608 : sizeof(disk_key)));
2609 : else
2610 : ASSERT(!memcmp_extent_buffer(b, &disk_key,
2611 : offsetof(struct btrfs_node, ptrs[0].key),
2612 : sizeof(disk_key)));
2613 : #endif
2614 : }
2615 :
2616 : static int key_search(struct extent_buffer *b, struct btrfs_key *key,
2617 : int level, int *prev_cmp, int *slot)
2618 : {
2619 4980558 : if (*prev_cmp != 0) {
2620 4980672 : *prev_cmp = bin_search(b, key, level, slot);
2621 : return *prev_cmp;
2622 : }
2623 :
2624 : key_search_validate(b, key, level);
2625 11021 : *slot = 0;
2626 :
2627 : return 0;
2628 : }
2629 :
2630 6052 : int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path,
2631 : u64 iobjectid, u64 ioff, u8 key_type,
2632 : struct btrfs_key *found_key)
2633 : {
2634 : int ret;
2635 : struct btrfs_key key;
2636 5227 : struct extent_buffer *eb;
2637 : struct btrfs_path *path;
2638 :
2639 6052 : key.type = key_type;
2640 6052 : key.objectid = iobjectid;
2641 6052 : key.offset = ioff;
2642 :
2643 6052 : if (found_path == NULL) {
2644 : path = btrfs_alloc_path();
2645 616 : if (!path)
2646 : return -ENOMEM;
2647 : } else
2648 : path = found_path;
2649 :
2650 6052 : ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2651 6052 : if ((ret < 0) || (found_key == NULL)) {
2652 821 : if (path != found_path)
2653 616 : btrfs_free_path(path);
2654 821 : return ret;
2655 : }
2656 :
2657 5231 : eb = path->nodes[0];
2658 10458 : if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2659 : ret = btrfs_next_leaf(fs_root, path);
2660 26 : if (ret)
2661 : return ret;
2662 26 : eb = path->nodes[0];
2663 : }
2664 :
2665 5231 : btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2666 9999 : if (found_key->type != key.type ||
2667 4768 : found_key->objectid != key.objectid)
2668 : return 1;
2669 :
2670 4768 : return 0;
2671 : }
2672 :
2673 : /*
2674 : * look for key in the tree. path is filled in with nodes along the way
2675 : * if key is found, we return zero and you can find the item in the leaf
2676 : * level of the path (level 0)
2677 : *
2678 : * If the key isn't found, the path points to the slot where it should
2679 : * be inserted, and 1 is returned. If there are other errors during the
2680 : * search a negative error number is returned.
2681 : *
2682 : * if ins_len > 0, nodes and leaves will be split as we walk down the
2683 : * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2684 : * possible)
2685 : */
2686 2713725 : int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2687 : *root, struct btrfs_key *key, struct btrfs_path *p, int
2688 : ins_len, int cow)
2689 : {
2690 : struct extent_buffer *b;
2691 : int slot;
2692 : int ret;
2693 : int err;
2694 : int level;
2695 : int lowest_unlock = 1;
2696 : int root_lock;
2697 : /* everything at write_lock_level or lower must be write locked */
2698 2713725 : int write_lock_level = 0;
2699 : u8 lowest_level = 0;
2700 : int min_write_lock_level;
2701 : int prev_cmp;
2702 :
2703 2713725 : lowest_level = p->lowest_level;
2704 2713725 : WARN_ON(lowest_level && ins_len > 0);
2705 2713711 : WARN_ON(p->nodes[0] != NULL);
2706 2713711 : BUG_ON(!cow && ins_len);
2707 :
2708 2713711 : if (ins_len < 0) {
2709 : lowest_unlock = 2;
2710 :
2711 : /* when we are removing items, we might have to go up to level
2712 : * two as we update tree pointers Make sure we keep write
2713 : * for those levels as well
2714 : */
2715 215608 : write_lock_level = 2;
2716 2498103 : } else if (ins_len > 0) {
2717 : /*
2718 : * for inserting items, make sure we have a write lock on
2719 : * level 1 so we can update keys
2720 : */
2721 311751 : write_lock_level = 1;
2722 : }
2723 :
2724 2713711 : if (!cow)
2725 1995650 : write_lock_level = -1;
2726 :
2727 2713711 : if (cow && (p->keep_locks || p->lowest_level))
2728 17677 : write_lock_level = BTRFS_MAX_LEVEL;
2729 :
2730 2713711 : min_write_lock_level = write_lock_level;
2731 :
2732 : again:
2733 : prev_cmp = -1;
2734 : /*
2735 : * we try very hard to do read locks on the root
2736 : */
2737 : root_lock = BTRFS_READ_LOCK;
2738 : level = 0;
2739 2811656 : if (p->search_commit_root) {
2740 : /*
2741 : * the commit roots are read only
2742 : * so we always do read locks
2743 : */
2744 1610652 : if (p->need_commit_sem)
2745 7783 : down_read(&root->fs_info->commit_root_sem);
2746 1610652 : b = root->commit_root;
2747 9582965 : extent_buffer_get(b);
2748 3221362 : level = btrfs_header_level(b);
2749 1610681 : if (p->need_commit_sem)
2750 7783 : up_read(&root->fs_info->commit_root_sem);
2751 1610576 : if (!p->skip_locking)
2752 0 : btrfs_tree_read_lock(b);
2753 : } else {
2754 1201004 : if (p->skip_locking) {
2755 0 : b = btrfs_root_node(root);
2756 0 : level = btrfs_header_level(b);
2757 : } else {
2758 : /* we don't know the level of the root node
2759 : * until we actually have it read locked
2760 : */
2761 1201004 : b = btrfs_read_lock_root_node(root);
2762 1201432 : level = btrfs_header_level(b);
2763 1201432 : if (level <= write_lock_level) {
2764 : /* whoops, must trade for write lock */
2765 632025 : btrfs_tree_read_unlock(b);
2766 632011 : free_extent_buffer(b);
2767 632021 : b = btrfs_lock_root_node(root);
2768 : root_lock = BTRFS_WRITE_LOCK;
2769 :
2770 : /* the level might have changed, check again */
2771 632051 : level = btrfs_header_level(b);
2772 : }
2773 : }
2774 : }
2775 2812034 : p->nodes[level] = b;
2776 2812034 : if (!p->skip_locking)
2777 1201364 : p->locks[level] = root_lock;
2778 :
2779 4991657 : while (b) {
2780 4990304 : level = btrfs_header_level(b);
2781 :
2782 : /*
2783 : * setup the path here so we can release it under lock
2784 : * contention with the cow code
2785 : */
2786 4990304 : if (cow) {
2787 : /*
2788 : * if we don't really need to cow this block
2789 : * then we don't want to set the path blocking,
2790 : * so we test it here
2791 : */
2792 1534341 : if (!should_cow_block(trans, root, b))
2793 : goto cow_done;
2794 :
2795 58705 : btrfs_set_path_blocking(p);
2796 :
2797 : /*
2798 : * must have write locks on this node and the
2799 : * parent
2800 : */
2801 115296 : if (level > write_lock_level ||
2802 68755 : (level + 1 > write_lock_level &&
2803 12164 : level + 1 < BTRFS_MAX_LEVEL &&
2804 12164 : p->nodes[level + 1])) {
2805 9533 : write_lock_level = level + 1;
2806 9533 : btrfs_release_path(p);
2807 9533 : goto again;
2808 : }
2809 :
2810 49172 : err = btrfs_cow_block(trans, root, b,
2811 : p->nodes[level + 1],
2812 : p->slots[level + 1], &b);
2813 49170 : if (err) {
2814 : ret = err;
2815 : goto done;
2816 : }
2817 : }
2818 : cow_done:
2819 4981692 : p->nodes[level] = b;
2820 4981692 : btrfs_clear_path_blocking(p, NULL, 0);
2821 :
2822 : /*
2823 : * we have a lock on b and as long as we aren't changing
2824 : * the tree, there is no way to for the items in b to change.
2825 : * It is safe to drop the lock on our parent before we
2826 : * go through the expensive btree search on b.
2827 : *
2828 : * If we're inserting or deleting (ins_len != 0), then we might
2829 : * be changing slot zero, which may require changing the parent.
2830 : * So, we can't drop the lock until after we know which slot
2831 : * we're operating on.
2832 : */
2833 4980504 : if (!ins_len && !p->keep_locks) {
2834 3722115 : int u = level + 1;
2835 :
2836 3722115 : if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2837 569200 : btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2838 569254 : p->locks[u] = 0;
2839 : }
2840 : }
2841 :
2842 4980558 : ret = key_search(b, key, level, &prev_cmp, &slot);
2843 :
2844 4980582 : if (level != 0) {
2845 : int dec = 0;
2846 2844572 : if (ret && slot > 0) {
2847 : dec = 1;
2848 2251151 : slot -= 1;
2849 : }
2850 2844572 : p->slots[level] = slot;
2851 2844572 : err = setup_nodes_for_search(trans, root, p, b, level,
2852 : ins_len, &write_lock_level);
2853 2844566 : if (err == -EAGAIN)
2854 : goto again;
2855 2784346 : if (err) {
2856 : ret = err;
2857 : goto done;
2858 : }
2859 2784348 : b = p->nodes[level];
2860 2784348 : slot = p->slots[level];
2861 :
2862 : /*
2863 : * slot 0 is special, if we change the key
2864 : * we have to update the parent pointer
2865 : * which means we must have a write lock
2866 : * on the parent
2867 : */
2868 2896461 : if (slot == 0 && ins_len &&
2869 112113 : write_lock_level < level + 1) {
2870 27869 : write_lock_level = level + 1;
2871 27869 : btrfs_release_path(p);
2872 27868 : goto again;
2873 : }
2874 :
2875 2756479 : unlock_up(p, level, lowest_unlock,
2876 : min_write_lock_level, &write_lock_level);
2877 :
2878 2756520 : if (level == lowest_level) {
2879 577272 : if (dec)
2880 0 : p->slots[level]++;
2881 : goto done;
2882 : }
2883 :
2884 2179248 : err = read_block_for_search(trans, root, p,
2885 : &b, level, slot, key, 0);
2886 2179444 : if (err == -EAGAIN)
2887 : goto again;
2888 2179120 : if (err) {
2889 : ret = err;
2890 : goto done;
2891 : }
2892 :
2893 2179371 : if (!p->skip_locking) {
2894 2296994 : level = btrfs_header_level(b);
2895 1148497 : if (level <= write_lock_level) {
2896 701151 : err = btrfs_try_tree_write_lock(b);
2897 701355 : if (!err) {
2898 8794 : btrfs_set_path_blocking(p);
2899 8797 : btrfs_tree_lock(b);
2900 8799 : btrfs_clear_path_blocking(p, b,
2901 : BTRFS_WRITE_LOCK);
2902 : }
2903 701300 : p->locks[level] = BTRFS_WRITE_LOCK;
2904 : } else {
2905 447346 : err = btrfs_try_tree_read_lock(b);
2906 447453 : if (!err) {
2907 1171 : btrfs_set_path_blocking(p);
2908 1171 : btrfs_tree_read_lock(b);
2909 1171 : btrfs_clear_path_blocking(p, b,
2910 : BTRFS_READ_LOCK);
2911 : }
2912 447449 : p->locks[level] = BTRFS_READ_LOCK;
2913 : }
2914 1148749 : p->nodes[level] = b;
2915 : }
2916 : } else {
2917 2136010 : p->slots[level] = slot;
2918 2447718 : if (ins_len > 0 &&
2919 311731 : btrfs_leaf_free_space(root, b) < ins_len) {
2920 13324 : if (write_lock_level < 1) {
2921 0 : write_lock_level = 1;
2922 0 : btrfs_release_path(p);
2923 0 : goto again;
2924 : }
2925 :
2926 13324 : btrfs_set_path_blocking(p);
2927 13324 : err = split_leaf(trans, root, key,
2928 : p, ins_len, ret == 0);
2929 13324 : btrfs_clear_path_blocking(p, NULL, 0);
2930 :
2931 13324 : BUG_ON(err > 0);
2932 13324 : if (err) {
2933 : ret = err;
2934 : goto done;
2935 : }
2936 : }
2937 2134241 : if (!p->search_for_split)
2938 2134116 : unlock_up(p, level, lowest_unlock,
2939 : min_write_lock_level, &write_lock_level);
2940 : goto done;
2941 : }
2942 : }
2943 : ret = 1;
2944 : done:
2945 : /*
2946 : * we don't really know what they plan on doing with the path
2947 : * from here on, so for now just mark it as blocking
2948 : */
2949 2713474 : if (!p->leave_spinning)
2950 2234443 : btrfs_set_path_blocking(p);
2951 2713702 : if (ret < 0)
2952 1928 : btrfs_release_path(p);
2953 2713702 : return ret;
2954 : }
2955 :
2956 : /*
2957 : * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2958 : * current state of the tree together with the operations recorded in the tree
2959 : * modification log to search for the key in a previous version of this tree, as
2960 : * denoted by the time_seq parameter.
2961 : *
2962 : * Naturally, there is no support for insert, delete or cow operations.
2963 : *
2964 : * The resulting path and return value will be set up as if we called
2965 : * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2966 : */
2967 608193 : int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2968 : struct btrfs_path *p, u64 time_seq)
2969 : {
2970 : struct extent_buffer *b;
2971 : int slot;
2972 : int ret;
2973 : int err;
2974 : int level;
2975 : int lowest_unlock = 1;
2976 : u8 lowest_level = 0;
2977 : int prev_cmp = -1;
2978 :
2979 608193 : lowest_level = p->lowest_level;
2980 608193 : WARN_ON(p->nodes[0] != NULL);
2981 :
2982 608193 : if (p->search_commit_root) {
2983 600267 : BUG_ON(time_seq);
2984 600267 : return btrfs_search_slot(NULL, root, key, p, 0, 0);
2985 : }
2986 :
2987 : again:
2988 7947 : b = get_old_root(root, time_seq);
2989 30217 : level = btrfs_header_level(b);
2990 7947 : p->locks[level] = BTRFS_READ_LOCK;
2991 :
2992 19082 : while (b) {
2993 11135 : level = btrfs_header_level(b);
2994 11135 : p->nodes[level] = b;
2995 11135 : btrfs_clear_path_blocking(p, NULL, 0);
2996 :
2997 : /*
2998 : * we have a lock on b and as long as we aren't changing
2999 : * the tree, there is no way to for the items in b to change.
3000 : * It is safe to drop the lock on our parent before we
3001 : * go through the expensive btree search on b.
3002 : */
3003 11135 : btrfs_unlock_up_safe(p, level + 1);
3004 :
3005 : /*
3006 : * Since we can unwind eb's we want to do a real search every
3007 : * time.
3008 : */
3009 : prev_cmp = -1;
3010 11135 : ret = key_search(b, key, level, &prev_cmp, &slot);
3011 :
3012 11135 : if (level != 0) {
3013 : int dec = 0;
3014 8093 : if (ret && slot > 0) {
3015 : dec = 1;
3016 3267 : slot -= 1;
3017 : }
3018 8093 : p->slots[level] = slot;
3019 8093 : unlock_up(p, level, lowest_unlock, 0, NULL);
3020 :
3021 8093 : if (level == lowest_level) {
3022 4884 : if (dec)
3023 90 : p->slots[level]++;
3024 : goto done;
3025 : }
3026 :
3027 3209 : err = read_block_for_search(NULL, root, p, &b, level,
3028 : slot, key, time_seq);
3029 3209 : if (err == -EAGAIN)
3030 : goto again;
3031 3188 : if (err) {
3032 : ret = err;
3033 : goto done;
3034 : }
3035 :
3036 6376 : level = btrfs_header_level(b);
3037 3188 : err = btrfs_try_tree_read_lock(b);
3038 3188 : if (!err) {
3039 0 : btrfs_set_path_blocking(p);
3040 0 : btrfs_tree_read_lock(b);
3041 0 : btrfs_clear_path_blocking(p, b,
3042 : BTRFS_READ_LOCK);
3043 : }
3044 3188 : b = tree_mod_log_rewind(root->fs_info, p, b, time_seq);
3045 3188 : if (!b) {
3046 : ret = -ENOMEM;
3047 : goto done;
3048 : }
3049 3188 : p->locks[level] = BTRFS_READ_LOCK;
3050 3188 : p->nodes[level] = b;
3051 : } else {
3052 3042 : p->slots[level] = slot;
3053 3042 : unlock_up(p, level, lowest_unlock, 0, NULL);
3054 3042 : goto done;
3055 : }
3056 : }
3057 : ret = 1;
3058 : done:
3059 7926 : if (!p->leave_spinning)
3060 7855 : btrfs_set_path_blocking(p);
3061 7926 : if (ret < 0)
3062 0 : btrfs_release_path(p);
3063 :
3064 7926 : return ret;
3065 : }
3066 :
3067 : /*
3068 : * helper to use instead of search slot if no exact match is needed but
3069 : * instead the next or previous item should be returned.
3070 : * When find_higher is true, the next higher item is returned, the next lower
3071 : * otherwise.
3072 : * When return_any and find_higher are both true, and no higher item is found,
3073 : * return the next lower instead.
3074 : * When return_any is true and find_higher is false, and no lower item is found,
3075 : * return the next higher instead.
3076 : * It returns 0 if any item is found, 1 if none is found (tree empty), and
3077 : * < 0 on error
3078 : */
3079 1085 : int btrfs_search_slot_for_read(struct btrfs_root *root,
3080 : struct btrfs_key *key, struct btrfs_path *p,
3081 : int find_higher, int return_any)
3082 : {
3083 : int ret;
3084 1007 : struct extent_buffer *leaf;
3085 :
3086 : again:
3087 1085 : ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3088 1085 : if (ret <= 0)
3089 : return ret;
3090 : /*
3091 : * a return value of 1 means the path is at the position where the
3092 : * item should be inserted. Normally this is the next bigger item,
3093 : * but in case the previous item is the last in a leaf, path points
3094 : * to the first free slot in the previous leaf, i.e. at an invalid
3095 : * item.
3096 : */
3097 1056 : leaf = p->nodes[0];
3098 :
3099 1056 : if (find_higher) {
3100 2014 : if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3101 : ret = btrfs_next_leaf(root, p);
3102 73 : if (ret <= 0)
3103 : return ret;
3104 16 : if (!return_any)
3105 : return 1;
3106 : /*
3107 : * no higher item found, return the next
3108 : * lower instead
3109 : */
3110 : return_any = 0;
3111 : find_higher = 0;
3112 0 : btrfs_release_path(p);
3113 0 : goto again;
3114 : }
3115 : } else {
3116 49 : if (p->slots[0] == 0) {
3117 0 : ret = btrfs_prev_leaf(root, p);
3118 0 : if (ret < 0)
3119 : return ret;
3120 0 : if (!ret) {
3121 0 : leaf = p->nodes[0];
3122 0 : if (p->slots[0] == btrfs_header_nritems(leaf))
3123 0 : p->slots[0]--;
3124 : return 0;
3125 : }
3126 0 : if (!return_any)
3127 : return 1;
3128 : /*
3129 : * no lower item found, return the next
3130 : * higher instead
3131 : */
3132 : return_any = 0;
3133 : find_higher = 1;
3134 0 : btrfs_release_path(p);
3135 0 : goto again;
3136 : } else {
3137 49 : --p->slots[0];
3138 : }
3139 : }
3140 : return 0;
3141 : }
3142 :
3143 : /*
3144 : * adjust the pointers going up the tree, starting at level
3145 : * making sure the right key of each node is points to 'key'.
3146 : * This is used after shifting pointers to the left, so it stops
3147 : * fixing up pointers when a given leaf/node is not in slot 0 of the
3148 : * higher levels
3149 : *
3150 : */
3151 15232 : static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
3152 : struct btrfs_disk_key *key, int level)
3153 : {
3154 : int i;
3155 : struct extent_buffer *t;
3156 :
3157 2593 : for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3158 17825 : int tslot = path->slots[i];
3159 17825 : if (!path->nodes[i])
3160 : break;
3161 : t = path->nodes[i];
3162 13604 : tree_mod_log_set_node_key(root->fs_info, t, tslot, 1);
3163 : btrfs_set_node_key(t, key, tslot);
3164 13604 : btrfs_mark_buffer_dirty(path->nodes[i]);
3165 13604 : if (tslot != 0)
3166 : break;
3167 : }
3168 15232 : }
3169 :
3170 : /*
3171 : * update item key.
3172 : *
3173 : * This function isn't completely safe. It's the caller's responsibility
3174 : * that the new key won't break the order
3175 : */
3176 8488 : void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
3177 : struct btrfs_key *new_key)
3178 : {
3179 : struct btrfs_disk_key disk_key;
3180 8488 : struct extent_buffer *eb;
3181 : int slot;
3182 :
3183 8488 : eb = path->nodes[0];
3184 8488 : slot = path->slots[0];
3185 8488 : if (slot > 0) {
3186 7328 : btrfs_item_key(eb, &disk_key, slot - 1);
3187 7328 : BUG_ON(comp_keys(&disk_key, new_key) >= 0);
3188 : }
3189 16976 : if (slot < btrfs_header_nritems(eb) - 1) {
3190 7257 : btrfs_item_key(eb, &disk_key, slot + 1);
3191 7257 : BUG_ON(comp_keys(&disk_key, new_key) <= 0);
3192 : }
3193 :
3194 : btrfs_cpu_key_to_disk(&disk_key, new_key);
3195 : btrfs_set_item_key(eb, &disk_key, slot);
3196 8488 : btrfs_mark_buffer_dirty(eb);
3197 8488 : if (slot == 0)
3198 1160 : fixup_low_keys(root, path, &disk_key, 1);
3199 8488 : }
3200 :
3201 : /*
3202 : * try to push data from one node into the next node left in the
3203 : * tree.
3204 : *
3205 : * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3206 : * error, and > 0 if there was no room in the left hand block.
3207 : */
3208 114 : static int push_node_left(struct btrfs_trans_handle *trans,
3209 342 : struct btrfs_root *root, struct extent_buffer *dst,
3210 228 : struct extent_buffer *src, int empty)
3211 : {
3212 : int push_items = 0;
3213 : int src_nritems;
3214 : int dst_nritems;
3215 : int ret = 0;
3216 :
3217 114 : src_nritems = btrfs_header_nritems(src);
3218 114 : dst_nritems = btrfs_header_nritems(dst);
3219 114 : push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3220 114 : WARN_ON(btrfs_header_generation(src) != trans->transid);
3221 114 : WARN_ON(btrfs_header_generation(dst) != trans->transid);
3222 :
3223 114 : if (!empty && src_nritems <= 8)
3224 : return 1;
3225 :
3226 114 : if (push_items <= 0)
3227 : return 1;
3228 :
3229 114 : if (empty) {
3230 17 : push_items = min(src_nritems, push_items);
3231 17 : if (push_items < src_nritems) {
3232 : /* leave at least 8 pointers in the node if
3233 : * we aren't going to empty it
3234 : */
3235 0 : if (src_nritems - push_items < 8) {
3236 0 : if (push_items <= 8)
3237 : return 1;
3238 0 : push_items -= 8;
3239 : }
3240 : }
3241 : } else
3242 97 : push_items = min(src_nritems - 8, push_items);
3243 :
3244 114 : ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3245 : push_items);
3246 114 : if (ret) {
3247 0 : btrfs_abort_transaction(trans, root, ret);
3248 0 : return ret;
3249 : }
3250 228 : copy_extent_buffer(dst, src,
3251 : btrfs_node_key_ptr_offset(dst_nritems),
3252 : btrfs_node_key_ptr_offset(0),
3253 : push_items * sizeof(struct btrfs_key_ptr));
3254 :
3255 114 : if (push_items < src_nritems) {
3256 : /*
3257 : * don't call tree_mod_log_eb_move here, key removal was already
3258 : * fully logged by tree_mod_log_eb_copy above.
3259 : */
3260 194 : memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3261 : btrfs_node_key_ptr_offset(push_items),
3262 97 : (src_nritems - push_items) *
3263 : sizeof(struct btrfs_key_ptr));
3264 : }
3265 114 : btrfs_set_header_nritems(src, src_nritems - push_items);
3266 114 : btrfs_set_header_nritems(dst, dst_nritems + push_items);
3267 114 : btrfs_mark_buffer_dirty(src);
3268 114 : btrfs_mark_buffer_dirty(dst);
3269 :
3270 114 : return ret;
3271 : }
3272 :
3273 : /*
3274 : * try to push data from one node into the next node right in the
3275 : * tree.
3276 : *
3277 : * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3278 : * error, and > 0 if there was no room in the right hand block.
3279 : *
3280 : * this will only push up to 1/2 the contents of the left node over
3281 : */
3282 65 : static int balance_node_right(struct btrfs_trans_handle *trans,
3283 : struct btrfs_root *root,
3284 195 : struct extent_buffer *dst,
3285 195 : struct extent_buffer *src)
3286 : {
3287 : int push_items = 0;
3288 : int max_push;
3289 : int src_nritems;
3290 : int dst_nritems;
3291 : int ret = 0;
3292 :
3293 65 : WARN_ON(btrfs_header_generation(src) != trans->transid);
3294 65 : WARN_ON(btrfs_header_generation(dst) != trans->transid);
3295 :
3296 65 : src_nritems = btrfs_header_nritems(src);
3297 65 : dst_nritems = btrfs_header_nritems(dst);
3298 65 : push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3299 65 : if (push_items <= 0)
3300 : return 1;
3301 :
3302 65 : if (src_nritems < 4)
3303 : return 1;
3304 :
3305 65 : max_push = src_nritems / 2 + 1;
3306 : /* don't try to empty the node */
3307 65 : if (max_push >= src_nritems)
3308 : return 1;
3309 :
3310 65 : if (max_push < push_items)
3311 : push_items = max_push;
3312 :
3313 65 : tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3314 130 : memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3315 : btrfs_node_key_ptr_offset(0),
3316 : (dst_nritems) *
3317 : sizeof(struct btrfs_key_ptr));
3318 :
3319 65 : ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3320 65 : src_nritems - push_items, push_items);
3321 65 : if (ret) {
3322 0 : btrfs_abort_transaction(trans, root, ret);
3323 0 : return ret;
3324 : }
3325 65 : copy_extent_buffer(dst, src,
3326 : btrfs_node_key_ptr_offset(0),
3327 : btrfs_node_key_ptr_offset(src_nritems - push_items),
3328 : push_items * sizeof(struct btrfs_key_ptr));
3329 :
3330 65 : btrfs_set_header_nritems(src, src_nritems - push_items);
3331 65 : btrfs_set_header_nritems(dst, dst_nritems + push_items);
3332 :
3333 65 : btrfs_mark_buffer_dirty(src);
3334 65 : btrfs_mark_buffer_dirty(dst);
3335 :
3336 65 : return ret;
3337 : }
3338 :
3339 : /*
3340 : * helper function to insert a new root level in the tree.
3341 : * A new node is allocated, and a single item is inserted to
3342 : * point to the existing root
3343 : *
3344 : * returns zero on success or < 0 on failure.
3345 : */
3346 99 : static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3347 99 : struct btrfs_root *root,
3348 : struct btrfs_path *path, int level)
3349 : {
3350 : u64 lower_gen;
3351 99 : struct extent_buffer *lower;
3352 495 : struct extent_buffer *c;
3353 : struct extent_buffer *old;
3354 : struct btrfs_disk_key lower_key;
3355 :
3356 99 : BUG_ON(path->nodes[level]);
3357 99 : BUG_ON(path->nodes[level-1] != root->node);
3358 :
3359 : lower = path->nodes[level-1];
3360 99 : if (level == 1)
3361 : btrfs_item_key(lower, &lower_key, 0);
3362 : else
3363 3 : btrfs_node_key(lower, &lower_key, 0);
3364 :
3365 99 : c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3366 : root->root_key.objectid, &lower_key,
3367 99 : level, root->node->start, 0);
3368 99 : if (IS_ERR(c))
3369 0 : return PTR_ERR(c);
3370 :
3371 99 : root_add_used(root, root->nodesize);
3372 :
3373 99 : memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3374 : btrfs_set_header_nritems(c, 1);
3375 99 : btrfs_set_header_level(c, level);
3376 99 : btrfs_set_header_bytenr(c, c->start);
3377 99 : btrfs_set_header_generation(c, trans->transid);
3378 : btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3379 99 : btrfs_set_header_owner(c, root->root_key.objectid);
3380 :
3381 99 : write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(),
3382 : BTRFS_FSID_SIZE);
3383 :
3384 99 : write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3385 : btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE);
3386 :
3387 : btrfs_set_node_key(c, &lower_key, 0);
3388 99 : btrfs_set_node_blockptr(c, 0, lower->start);
3389 : lower_gen = btrfs_header_generation(lower);
3390 99 : WARN_ON(lower_gen != trans->transid);
3391 :
3392 : btrfs_set_node_ptr_generation(c, 0, lower_gen);
3393 :
3394 99 : btrfs_mark_buffer_dirty(c);
3395 :
3396 99 : old = root->node;
3397 99 : tree_mod_log_set_root_pointer(root, c, 0);
3398 99 : rcu_assign_pointer(root->node, c);
3399 :
3400 : /* the super has an extra ref to root->node */
3401 99 : free_extent_buffer(old);
3402 :
3403 99 : add_root_to_dirty_list(root);
3404 : extent_buffer_get(c);
3405 99 : path->nodes[level] = c;
3406 99 : path->locks[level] = BTRFS_WRITE_LOCK;
3407 99 : path->slots[level] = 0;
3408 99 : return 0;
3409 : }
3410 :
3411 : /*
3412 : * worker function to insert a single pointer in a node.
3413 : * the node should have enough room for the pointer already
3414 : *
3415 : * slot and level indicate where you want the key to go, and
3416 : * blocknr is the block the key points to.
3417 : */
3418 5312 : static void insert_ptr(struct btrfs_trans_handle *trans,
3419 : struct btrfs_root *root, struct btrfs_path *path,
3420 : struct btrfs_disk_key *key, u64 bytenr,
3421 : int slot, int level)
3422 : {
3423 10624 : struct extent_buffer *lower;
3424 : int nritems;
3425 : int ret;
3426 :
3427 5312 : BUG_ON(!path->nodes[level]);
3428 5312 : btrfs_assert_tree_locked(path->nodes[level]);
3429 5312 : lower = path->nodes[level];
3430 5312 : nritems = btrfs_header_nritems(lower);
3431 5312 : BUG_ON(slot > nritems);
3432 5312 : BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3433 5312 : if (slot != nritems) {
3434 3242 : if (level)
3435 3242 : tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3436 : slot, nritems - slot);
3437 9726 : memmove_extent_buffer(lower,
3438 : btrfs_node_key_ptr_offset(slot + 1),
3439 : btrfs_node_key_ptr_offset(slot),
3440 3242 : (nritems - slot) * sizeof(struct btrfs_key_ptr));
3441 : }
3442 5312 : if (level) {
3443 5312 : ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3444 : MOD_LOG_KEY_ADD, GFP_NOFS);
3445 5312 : BUG_ON(ret < 0);
3446 : }
3447 : btrfs_set_node_key(lower, key, slot);
3448 : btrfs_set_node_blockptr(lower, slot, bytenr);
3449 5312 : WARN_ON(trans->transid == 0);
3450 5312 : btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3451 5312 : btrfs_set_header_nritems(lower, nritems + 1);
3452 5312 : btrfs_mark_buffer_dirty(lower);
3453 5312 : }
3454 :
3455 : /*
3456 : * split the node at the specified level in path in two.
3457 : * The path is corrected to point to the appropriate node after the split
3458 : *
3459 : * Before splitting this tries to make some room in the node by pushing
3460 : * left and right, if either one works, it returns right away.
3461 : *
3462 : * returns 0 on success and < 0 on failure
3463 : */
3464 180 : static noinline int split_node(struct btrfs_trans_handle *trans,
3465 : struct btrfs_root *root,
3466 : struct btrfs_path *path, int level)
3467 : {
3468 438 : struct extent_buffer *c;
3469 160 : struct extent_buffer *split;
3470 : struct btrfs_disk_key disk_key;
3471 : int mid;
3472 : int ret;
3473 : u32 c_nritems;
3474 :
3475 180 : c = path->nodes[level];
3476 180 : WARN_ON(btrfs_header_generation(c) != trans->transid);
3477 180 : if (c == root->node) {
3478 : /*
3479 : * trying to split the root, lets make a new one
3480 : *
3481 : * tree mod log: We don't log_removal old root in
3482 : * insert_new_root, because that root buffer will be kept as a
3483 : * normal node. We are going to log removal of half of the
3484 : * elements below with tree_mod_log_eb_copy. We're holding a
3485 : * tree lock on the buffer, which is why we cannot race with
3486 : * other tree_mod_log users.
3487 : */
3488 3 : ret = insert_new_root(trans, root, path, level + 1);
3489 3 : if (ret)
3490 : return ret;
3491 : } else {
3492 177 : ret = push_nodes_for_insert(trans, root, path, level);
3493 177 : c = path->nodes[level];
3494 501 : if (!ret && btrfs_header_nritems(c) <
3495 162 : BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3496 : return 0;
3497 29 : if (ret < 0)
3498 : return ret;
3499 : }
3500 :
3501 : c_nritems = btrfs_header_nritems(c);
3502 32 : mid = (c_nritems + 1) / 2;
3503 32 : btrfs_node_key(c, &disk_key, mid);
3504 :
3505 32 : split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3506 : root->root_key.objectid,
3507 : &disk_key, level, c->start, 0);
3508 32 : if (IS_ERR(split))
3509 0 : return PTR_ERR(split);
3510 :
3511 32 : root_add_used(root, root->nodesize);
3512 :
3513 32 : memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3514 : btrfs_set_header_level(split, btrfs_header_level(c));
3515 32 : btrfs_set_header_bytenr(split, split->start);
3516 32 : btrfs_set_header_generation(split, trans->transid);
3517 : btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3518 32 : btrfs_set_header_owner(split, root->root_key.objectid);
3519 32 : write_extent_buffer(split, root->fs_info->fsid,
3520 : btrfs_header_fsid(), BTRFS_FSID_SIZE);
3521 32 : write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3522 : btrfs_header_chunk_tree_uuid(split),
3523 : BTRFS_UUID_SIZE);
3524 :
3525 32 : ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0,
3526 32 : mid, c_nritems - mid);
3527 32 : if (ret) {
3528 0 : btrfs_abort_transaction(trans, root, ret);
3529 0 : return ret;
3530 : }
3531 64 : copy_extent_buffer(split, c,
3532 : btrfs_node_key_ptr_offset(0),
3533 : btrfs_node_key_ptr_offset(mid),
3534 : (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3535 : btrfs_set_header_nritems(split, c_nritems - mid);
3536 : btrfs_set_header_nritems(c, mid);
3537 : ret = 0;
3538 :
3539 32 : btrfs_mark_buffer_dirty(c);
3540 32 : btrfs_mark_buffer_dirty(split);
3541 :
3542 32 : insert_ptr(trans, root, path, &disk_key, split->start,
3543 32 : path->slots[level + 1] + 1, level + 1);
3544 :
3545 32 : if (path->slots[level] >= mid) {
3546 26 : path->slots[level] -= mid;
3547 26 : btrfs_tree_unlock(c);
3548 26 : free_extent_buffer(c);
3549 26 : path->nodes[level] = split;
3550 26 : path->slots[level + 1] += 1;
3551 : } else {
3552 6 : btrfs_tree_unlock(split);
3553 6 : free_extent_buffer(split);
3554 : }
3555 : return ret;
3556 : }
3557 :
3558 : /*
3559 : * how many bytes are required to store the items in a leaf. start
3560 : * and nr indicate which items in the leaf to check. This totals up the
3561 : * space used both by the item structs and the item data
3562 : */
3563 1472520 : static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3564 : {
3565 : struct btrfs_item *start_item;
3566 : struct btrfs_item *end_item;
3567 : struct btrfs_map_token token;
3568 : int data_len;
3569 1472520 : int nritems = btrfs_header_nritems(l);
3570 1472520 : int end = min(nritems, start + nr) - 1;
3571 :
3572 1472520 : if (!nr)
3573 : return 0;
3574 : btrfs_init_map_token(&token);
3575 : start_item = btrfs_item_nr(start);
3576 : end_item = btrfs_item_nr(end);
3577 1469729 : data_len = btrfs_token_item_offset(l, start_item, &token) +
3578 : btrfs_token_item_size(l, start_item, &token);
3579 1469192 : data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3580 1469192 : data_len += sizeof(struct btrfs_item) * nr;
3581 1469192 : WARN_ON(data_len < 0);
3582 1469233 : return data_len;
3583 : }
3584 :
3585 : /*
3586 : * The space between the end of the leaf items and
3587 : * the start of the leaf data. IOW, how much room
3588 : * the leaf has left for both items and data
3589 : */
3590 1340412 : noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3591 1340412 : struct extent_buffer *leaf)
3592 : {
3593 1340412 : int nritems = btrfs_header_nritems(leaf);
3594 : int ret;
3595 1340412 : ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3596 1339636 : if (ret < 0) {
3597 0 : btrfs_crit(root->fs_info,
3598 : "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3599 : ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3600 : leaf_space_used(leaf, 0, nritems), nritems);
3601 : }
3602 1339636 : return ret;
3603 : }
3604 :
3605 : /*
3606 : * min slot controls the lowest index we're willing to push to the
3607 : * right. We'll push up to and including min_slot, but no lower
3608 : */
3609 5925 : static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3610 : struct btrfs_root *root,
3611 : struct btrfs_path *path,
3612 : int data_size, int empty,
3613 8795 : struct extent_buffer *right,
3614 : int free_space, u32 left_nritems,
3615 : u32 min_slot)
3616 : {
3617 10322 : struct extent_buffer *left = path->nodes[0];
3618 5925 : struct extent_buffer *upper = path->nodes[1];
3619 : struct btrfs_map_token token;
3620 : struct btrfs_disk_key disk_key;
3621 : int slot;
3622 : u32 i;
3623 : int push_space = 0;
3624 : int push_items = 0;
3625 : struct btrfs_item *item;
3626 : u32 nr;
3627 : u32 right_nritems;
3628 : u32 data_end;
3629 : u32 this_item_size;
3630 :
3631 : btrfs_init_map_token(&token);
3632 :
3633 5925 : if (empty)
3634 : nr = 0;
3635 : else
3636 4988 : nr = max_t(u32, 1, min_slot);
3637 :
3638 5925 : if (path->slots[0] >= left_nritems)
3639 : push_space += data_size;
3640 :
3641 5925 : slot = path->slots[1];
3642 5925 : i = left_nritems - 1;
3643 120382 : while (i >= nr) {
3644 114453 : item = btrfs_item_nr(i);
3645 :
3646 114453 : if (!empty && push_items > 0) {
3647 101420 : if (path->slots[0] > i)
3648 : break;
3649 100659 : if (path->slots[0] == i) {
3650 1249 : int space = btrfs_leaf_free_space(root, left);
3651 1249 : if (space + push_space * 2 > free_space)
3652 : break;
3653 : }
3654 : }
3655 :
3656 113137 : if (path->slots[0] == i)
3657 1158 : push_space += data_size;
3658 :
3659 : this_item_size = btrfs_item_size(left, item);
3660 113137 : if (this_item_size + sizeof(*item) + push_space > free_space)
3661 : break;
3662 :
3663 108846 : push_items++;
3664 108846 : push_space += this_item_size + sizeof(*item);
3665 108846 : if (i == 0)
3666 : break;
3667 108532 : i--;
3668 : }
3669 :
3670 5925 : if (push_items == 0)
3671 : goto out_unlock;
3672 :
3673 4398 : WARN_ON(!empty && push_items == left_nritems);
3674 :
3675 : /* push left to right */
3676 : right_nritems = btrfs_header_nritems(right);
3677 :
3678 4398 : push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3679 4398 : push_space -= leaf_data_end(root, left);
3680 :
3681 : /* make room in the right data area */
3682 4398 : data_end = leaf_data_end(root, right);
3683 4398 : memmove_extent_buffer(right,
3684 : btrfs_leaf_data(right) + data_end - push_space,
3685 : btrfs_leaf_data(right) + data_end,
3686 4398 : BTRFS_LEAF_DATA_SIZE(root) - data_end);
3687 :
3688 : /* copy from the left data area */
3689 13194 : copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3690 8796 : BTRFS_LEAF_DATA_SIZE(root) - push_space,
3691 4398 : btrfs_leaf_data(left) + leaf_data_end(root, left),
3692 : push_space);
3693 :
3694 8796 : memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3695 : btrfs_item_nr_offset(0),
3696 : right_nritems * sizeof(struct btrfs_item));
3697 :
3698 : /* copy the items from left to right */
3699 4398 : copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3700 : btrfs_item_nr_offset(left_nritems - push_items),
3701 : push_items * sizeof(struct btrfs_item));
3702 :
3703 : /* update the item pointers */
3704 4397 : right_nritems += push_items;
3705 : btrfs_set_header_nritems(right, right_nritems);
3706 4397 : push_space = BTRFS_LEAF_DATA_SIZE(root);
3707 442960 : for (i = 0; i < right_nritems; i++) {
3708 438563 : item = btrfs_item_nr(i);
3709 877126 : push_space -= btrfs_token_item_size(right, item, &token);
3710 : btrfs_set_token_item_offset(right, item, push_space, &token);
3711 : }
3712 :
3713 : left_nritems -= push_items;
3714 : btrfs_set_header_nritems(left, left_nritems);
3715 :
3716 4397 : if (left_nritems)
3717 4083 : btrfs_mark_buffer_dirty(left);
3718 : else
3719 314 : clean_tree_block(trans, root, left);
3720 :
3721 4398 : btrfs_mark_buffer_dirty(right);
3722 :
3723 : btrfs_item_key(right, &disk_key, 0);
3724 4398 : btrfs_set_node_key(upper, &disk_key, slot + 1);
3725 4397 : btrfs_mark_buffer_dirty(upper);
3726 :
3727 : /* then fixup the leaf pointer in the path */
3728 4398 : if (path->slots[0] >= left_nritems) {
3729 1164 : path->slots[0] -= left_nritems;
3730 2328 : if (btrfs_header_nritems(path->nodes[0]) == 0)
3731 314 : clean_tree_block(trans, root, path->nodes[0]);
3732 1164 : btrfs_tree_unlock(path->nodes[0]);
3733 1164 : free_extent_buffer(path->nodes[0]);
3734 1164 : path->nodes[0] = right;
3735 1164 : path->slots[1] += 1;
3736 : } else {
3737 3234 : btrfs_tree_unlock(right);
3738 3234 : free_extent_buffer(right);
3739 : }
3740 : return 0;
3741 :
3742 : out_unlock:
3743 1527 : btrfs_tree_unlock(right);
3744 1527 : free_extent_buffer(right);
3745 1527 : return 1;
3746 : }
3747 :
3748 : /*
3749 : * push some data in the path leaf to the right, trying to free up at
3750 : * least data_size bytes. returns zero if the push worked, nonzero otherwise
3751 : *
3752 : * returns 1 if the push failed because the other node didn't have enough
3753 : * room, 0 if everything worked out and < 0 if there were major errors.
3754 : *
3755 : * this will push starting from min_slot to the end of the leaf. It won't
3756 : * push any slot lower than min_slot
3757 : */
3758 25319 : static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3759 : *root, struct btrfs_path *path,
3760 : int min_data_size, int data_size,
3761 : int empty, u32 min_slot)
3762 : {
3763 31370 : struct extent_buffer *left = path->nodes[0];
3764 : struct extent_buffer *right;
3765 14758 : struct extent_buffer *upper;
3766 : int slot;
3767 : int free_space;
3768 : u32 left_nritems;
3769 : int ret;
3770 :
3771 25319 : if (!path->nodes[1])
3772 : return 1;
3773 :
3774 14758 : slot = path->slots[1];
3775 : upper = path->nodes[1];
3776 29516 : if (slot >= btrfs_header_nritems(upper) - 1)
3777 : return 1;
3778 :
3779 9422 : btrfs_assert_tree_locked(path->nodes[1]);
3780 :
3781 9422 : right = read_node_slot(root, upper, slot + 1);
3782 9422 : if (right == NULL)
3783 : return 1;
3784 :
3785 9422 : btrfs_tree_lock(right);
3786 9422 : btrfs_set_lock_blocking(right);
3787 :
3788 9422 : free_space = btrfs_leaf_free_space(root, right);
3789 9422 : if (free_space < data_size)
3790 : goto out_unlock;
3791 :
3792 : /* cow and double check */
3793 6051 : ret = btrfs_cow_block(trans, root, right, upper,
3794 : slot + 1, &right);
3795 6051 : if (ret)
3796 : goto out_unlock;
3797 :
3798 6051 : free_space = btrfs_leaf_free_space(root, right);
3799 6051 : if (free_space < data_size)
3800 : goto out_unlock;
3801 :
3802 : left_nritems = btrfs_header_nritems(left);
3803 6051 : if (left_nritems == 0)
3804 : goto out_unlock;
3805 :
3806 6051 : if (path->slots[0] == left_nritems && !empty) {
3807 : /* Key greater than all keys in the leaf, right neighbor has
3808 : * enough room for it and we're not emptying our leaf to delete
3809 : * it, therefore use right neighbor to insert the new item and
3810 : * no need to touch/dirty our left leaft. */
3811 126 : btrfs_tree_unlock(left);
3812 126 : free_extent_buffer(left);
3813 126 : path->nodes[0] = right;
3814 126 : path->slots[0] = 0;
3815 126 : path->slots[1]++;
3816 126 : return 0;
3817 : }
3818 :
3819 5925 : return __push_leaf_right(trans, root, path, min_data_size, empty,
3820 : right, free_space, left_nritems, min_slot);
3821 : out_unlock:
3822 3371 : btrfs_tree_unlock(right);
3823 3371 : free_extent_buffer(right);
3824 3371 : return 1;
3825 : }
3826 :
3827 : /*
3828 : * push some data in the path leaf to the left, trying to free up at
3829 : * least data_size bytes. returns zero if the push worked, nonzero otherwise
3830 : *
3831 : * max_slot can put a limit on how far into the leaf we'll push items. The
3832 : * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3833 : * items
3834 : */
3835 7942 : static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3836 : struct btrfs_root *root,
3837 : struct btrfs_path *path, int data_size,
3838 15128 : int empty, struct extent_buffer *left,
3839 : int free_space, u32 right_nritems,
3840 : u32 max_slot)
3841 : {
3842 : struct btrfs_disk_key disk_key;
3843 20842 : struct extent_buffer *right = path->nodes[0];
3844 : int i;
3845 : int push_space = 0;
3846 : int push_items = 0;
3847 : struct btrfs_item *item;
3848 : u32 old_left_nritems;
3849 : u32 nr;
3850 : int ret = 0;
3851 : u32 this_item_size;
3852 : u32 old_left_item_size;
3853 : struct btrfs_map_token token;
3854 :
3855 : btrfs_init_map_token(&token);
3856 :
3857 7942 : if (empty)
3858 3080 : nr = min(right_nritems, max_slot);
3859 : else
3860 4862 : nr = min(right_nritems - 1, max_slot);
3861 :
3862 113871 : for (i = 0; i < nr; i++) {
3863 : item = btrfs_item_nr(i);
3864 :
3865 112954 : if (!empty && push_items > 0) {
3866 88912 : if (path->slots[0] < i)
3867 : break;
3868 88345 : if (path->slots[0] == i) {
3869 889 : int space = btrfs_leaf_free_space(root, right);
3870 889 : if (space + push_space * 2 > free_space)
3871 : break;
3872 : }
3873 : }
3874 :
3875 112132 : if (path->slots[0] == i)
3876 1635 : push_space += data_size;
3877 :
3878 : this_item_size = btrfs_item_size(right, item);
3879 112131 : if (this_item_size + sizeof(*item) + push_space > free_space)
3880 : break;
3881 :
3882 105929 : push_items++;
3883 105929 : push_space += this_item_size + sizeof(*item);
3884 : }
3885 :
3886 7941 : if (push_items == 0) {
3887 : ret = 1;
3888 : goto out;
3889 : }
3890 8716 : WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3891 :
3892 : /* push data from right to left */
3893 15126 : copy_extent_buffer(left, right,
3894 : btrfs_item_nr_offset(btrfs_header_nritems(left)),
3895 : btrfs_item_nr_offset(0),
3896 : push_items * sizeof(struct btrfs_item));
3897 :
3898 10086 : push_space = BTRFS_LEAF_DATA_SIZE(root) -
3899 5043 : btrfs_item_offset_nr(right, push_items - 1);
3900 :
3901 10086 : copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3902 5043 : leaf_data_end(root, left) - push_space,
3903 : btrfs_leaf_data(right) +
3904 : btrfs_item_offset_nr(right, push_items - 1),
3905 : push_space);
3906 : old_left_nritems = btrfs_header_nritems(left);
3907 5043 : BUG_ON(old_left_nritems <= 0);
3908 :
3909 5043 : old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3910 110984 : for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3911 : u32 ioff;
3912 :
3913 : item = btrfs_item_nr(i);
3914 :
3915 : ioff = btrfs_token_item_offset(left, item, &token);
3916 105941 : btrfs_set_token_item_offset(left, item,
3917 105941 : ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3918 : &token);
3919 : }
3920 : btrfs_set_header_nritems(left, old_left_nritems + push_items);
3921 :
3922 : /* fixup right node */
3923 5043 : if (push_items > right_nritems)
3924 0 : WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3925 : right_nritems);
3926 :
3927 5043 : if (push_items < right_nritems) {
3928 4183 : push_space = btrfs_item_offset_nr(right, push_items - 1) -
3929 4183 : leaf_data_end(root, right);
3930 12549 : memmove_extent_buffer(right, btrfs_leaf_data(right) +
3931 8366 : BTRFS_LEAF_DATA_SIZE(root) - push_space,
3932 : btrfs_leaf_data(right) +
3933 4183 : leaf_data_end(root, right), push_space);
3934 :
3935 8366 : memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3936 : btrfs_item_nr_offset(push_items),
3937 4183 : (btrfs_header_nritems(right) - push_items) *
3938 : sizeof(struct btrfs_item));
3939 : }
3940 5043 : right_nritems -= push_items;
3941 : btrfs_set_header_nritems(right, right_nritems);
3942 5043 : push_space = BTRFS_LEAF_DATA_SIZE(root);
3943 289376 : for (i = 0; i < right_nritems; i++) {
3944 : item = btrfs_item_nr(i);
3945 :
3946 568666 : push_space = push_space - btrfs_token_item_size(right,
3947 : item, &token);
3948 : btrfs_set_token_item_offset(right, item, push_space, &token);
3949 : }
3950 :
3951 5043 : btrfs_mark_buffer_dirty(left);
3952 5043 : if (right_nritems)
3953 4183 : btrfs_mark_buffer_dirty(right);
3954 : else
3955 860 : clean_tree_block(trans, root, right);
3956 :
3957 : btrfs_item_key(right, &disk_key, 0);
3958 5043 : fixup_low_keys(root, path, &disk_key, 1);
3959 :
3960 : /* then fixup the leaf pointer in the path */
3961 5043 : if (path->slots[0] < push_items) {
3962 1437 : path->slots[0] += old_left_nritems;
3963 1437 : btrfs_tree_unlock(path->nodes[0]);
3964 1437 : free_extent_buffer(path->nodes[0]);
3965 1437 : path->nodes[0] = left;
3966 1437 : path->slots[1] -= 1;
3967 : } else {
3968 3606 : btrfs_tree_unlock(left);
3969 3606 : free_extent_buffer(left);
3970 3606 : path->slots[0] -= push_items;
3971 : }
3972 5043 : BUG_ON(path->slots[0] < 0);
3973 : return ret;
3974 : out:
3975 2899 : btrfs_tree_unlock(left);
3976 2899 : free_extent_buffer(left);
3977 2899 : return ret;
3978 : }
3979 :
3980 : /*
3981 : * push some data in the path leaf to the left, trying to free up at
3982 : * least data_size bytes. returns zero if the push worked, nonzero otherwise
3983 : *
3984 : * max_slot can put a limit on how far into the leaf we'll push items. The
3985 : * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3986 : * items
3987 : */
3988 22351 : static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3989 : *root, struct btrfs_path *path, int min_data_size,
3990 : int data_size, int empty, u32 max_slot)
3991 : {
3992 33615 : struct extent_buffer *right = path->nodes[0];
3993 : struct extent_buffer *left;
3994 : int slot;
3995 : int free_space;
3996 : u32 right_nritems;
3997 : int ret = 0;
3998 :
3999 22351 : slot = path->slots[1];
4000 22351 : if (slot == 0)
4001 : return 1;
4002 11264 : if (!path->nodes[1])
4003 : return 1;
4004 :
4005 : right_nritems = btrfs_header_nritems(right);
4006 11264 : if (right_nritems == 0)
4007 : return 1;
4008 :
4009 11264 : btrfs_assert_tree_locked(path->nodes[1]);
4010 :
4011 11264 : left = read_node_slot(root, path->nodes[1], slot - 1);
4012 11264 : if (left == NULL)
4013 : return 1;
4014 :
4015 11264 : btrfs_tree_lock(left);
4016 11264 : btrfs_set_lock_blocking(left);
4017 :
4018 11264 : free_space = btrfs_leaf_free_space(root, left);
4019 11264 : if (free_space < data_size) {
4020 : ret = 1;
4021 : goto out;
4022 : }
4023 :
4024 : /* cow and double check */
4025 7942 : ret = btrfs_cow_block(trans, root, left,
4026 : path->nodes[1], slot - 1, &left);
4027 7942 : if (ret) {
4028 : /* we hit -ENOSPC, but it isn't fatal here */
4029 0 : if (ret == -ENOSPC)
4030 : ret = 1;
4031 : goto out;
4032 : }
4033 :
4034 7942 : free_space = btrfs_leaf_free_space(root, left);
4035 7942 : if (free_space < data_size) {
4036 : ret = 1;
4037 : goto out;
4038 : }
4039 :
4040 7942 : return __push_leaf_left(trans, root, path, min_data_size,
4041 : empty, left, free_space, right_nritems,
4042 : max_slot);
4043 : out:
4044 3322 : btrfs_tree_unlock(left);
4045 3322 : free_extent_buffer(left);
4046 3322 : return ret;
4047 : }
4048 :
4049 : /*
4050 : * split the path's leaf in two, making sure there is at least data_size
4051 : * available for the resulting leaf level of the path.
4052 : */
4053 4137 : static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4054 : struct btrfs_root *root,
4055 : struct btrfs_path *path,
4056 4137 : struct extent_buffer *l,
4057 4137 : struct extent_buffer *right,
4058 : int slot, int mid, int nritems)
4059 : {
4060 : int data_copy_size;
4061 : int rt_data_off;
4062 : int i;
4063 : struct btrfs_disk_key disk_key;
4064 : struct btrfs_map_token token;
4065 :
4066 : btrfs_init_map_token(&token);
4067 :
4068 4137 : nritems = nritems - mid;
4069 4137 : btrfs_set_header_nritems(right, nritems);
4070 4137 : data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
4071 :
4072 4137 : copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4073 : btrfs_item_nr_offset(mid),
4074 : nritems * sizeof(struct btrfs_item));
4075 :
4076 8274 : copy_extent_buffer(right, l,
4077 4137 : btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
4078 : data_copy_size, btrfs_leaf_data(l) +
4079 4137 : leaf_data_end(root, l), data_copy_size);
4080 :
4081 8275 : rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
4082 : btrfs_item_end_nr(l, mid);
4083 :
4084 140992 : for (i = 0; i < nritems; i++) {
4085 : struct btrfs_item *item = btrfs_item_nr(i);
4086 : u32 ioff;
4087 :
4088 : ioff = btrfs_token_item_offset(right, item, &token);
4089 136854 : btrfs_set_token_item_offset(right, item,
4090 : ioff + rt_data_off, &token);
4091 : }
4092 :
4093 4137 : btrfs_set_header_nritems(l, mid);
4094 : btrfs_item_key(right, &disk_key, 0);
4095 4137 : insert_ptr(trans, root, path, &disk_key, right->start,
4096 4137 : path->slots[1] + 1, 1);
4097 :
4098 4137 : btrfs_mark_buffer_dirty(right);
4099 4137 : btrfs_mark_buffer_dirty(l);
4100 4137 : BUG_ON(path->slots[0] != slot);
4101 :
4102 4137 : if (mid <= slot) {
4103 3295 : btrfs_tree_unlock(path->nodes[0]);
4104 3295 : free_extent_buffer(path->nodes[0]);
4105 3295 : path->nodes[0] = right;
4106 3295 : path->slots[0] -= mid;
4107 3295 : path->slots[1] += 1;
4108 : } else {
4109 842 : btrfs_tree_unlock(right);
4110 842 : free_extent_buffer(right);
4111 : }
4112 :
4113 4137 : BUG_ON(path->slots[0] < 0);
4114 4137 : }
4115 :
4116 : /*
4117 : * double splits happen when we need to insert a big item in the middle
4118 : * of a leaf. A double split can leave us with 3 mostly empty leaves:
4119 : * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4120 : * A B C
4121 : *
4122 : * We avoid this by trying to push the items on either side of our target
4123 : * into the adjacent leaves. If all goes well we can avoid the double split
4124 : * completely.
4125 : */
4126 946 : static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4127 : struct btrfs_root *root,
4128 : struct btrfs_path *path,
4129 : int data_size)
4130 : {
4131 : int ret;
4132 : int progress = 0;
4133 : int slot;
4134 : u32 nritems;
4135 : int space_needed = data_size;
4136 :
4137 946 : slot = path->slots[0];
4138 2838 : if (slot < btrfs_header_nritems(path->nodes[0]))
4139 946 : space_needed -= btrfs_leaf_free_space(root, path->nodes[0]);
4140 :
4141 : /*
4142 : * try to push all the items after our slot into the
4143 : * right leaf
4144 : */
4145 946 : ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4146 946 : if (ret < 0)
4147 : return ret;
4148 :
4149 946 : if (ret == 0)
4150 : progress++;
4151 :
4152 946 : nritems = btrfs_header_nritems(path->nodes[0]);
4153 : /*
4154 : * our goal is to get our slot at the start or end of a leaf. If
4155 : * we've done so we're done
4156 : */
4157 946 : if (path->slots[0] == 0 || path->slots[0] == nritems)
4158 : return 0;
4159 :
4160 943 : if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4161 : return 0;
4162 :
4163 : /* try to push all the items before our slot into the next leaf */
4164 943 : slot = path->slots[0];
4165 943 : ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4166 943 : if (ret < 0)
4167 : return ret;
4168 :
4169 943 : if (ret == 0)
4170 0 : progress++;
4171 :
4172 943 : if (progress)
4173 : return 0;
4174 943 : return 1;
4175 : }
4176 :
4177 : /*
4178 : * split the path's leaf in two, making sure there is at least data_size
4179 : * available for the resulting leaf level of the path.
4180 : *
4181 : * returns 0 if all went well and < 0 on failure.
4182 : */
4183 13447 : static noinline int split_leaf(struct btrfs_trans_handle *trans,
4184 : struct btrfs_root *root,
4185 : struct btrfs_key *ins_key,
4186 : struct btrfs_path *path, int data_size,
4187 : int extend)
4188 : {
4189 : struct btrfs_disk_key disk_key;
4190 17649 : struct extent_buffer *l;
4191 : u32 nritems;
4192 : int mid;
4193 : int slot;
4194 22263 : struct extent_buffer *right;
4195 : int ret = 0;
4196 : int wret;
4197 : int split;
4198 : int num_doubles = 0;
4199 : int tried_avoid_double = 0;
4200 :
4201 13447 : l = path->nodes[0];
4202 13447 : slot = path->slots[0];
4203 17875 : if (extend && data_size + btrfs_item_size_nr(l, slot) +
4204 2214 : sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
4205 : return -EOVERFLOW;
4206 :
4207 : /* first try to make some room by pushing left and right */
4208 11519 : if (data_size && path->nodes[1]) {
4209 : int space_needed = data_size;
4210 :
4211 22846 : if (slot < btrfs_header_nritems(l))
4212 8627 : space_needed -= btrfs_leaf_free_space(root, l);
4213 :
4214 11423 : wret = push_leaf_right(trans, root, path, space_needed,
4215 : space_needed, 0, 0);
4216 11423 : if (wret < 0)
4217 : return wret;
4218 11423 : if (wret) {
4219 7384 : wret = push_leaf_left(trans, root, path, space_needed,
4220 : space_needed, 0, (u32)-1);
4221 7384 : if (wret < 0)
4222 : return wret;
4223 : }
4224 11423 : l = path->nodes[0];
4225 :
4226 : /* did the pushes work? */
4227 11423 : if (btrfs_leaf_free_space(root, l) >= data_size)
4228 : return 0;
4229 : }
4230 :
4231 4337 : if (!path->nodes[1]) {
4232 96 : ret = insert_new_root(trans, root, path, 1);
4233 96 : if (ret)
4234 : return ret;
4235 : }
4236 : again:
4237 : split = 1;
4238 6226 : l = path->nodes[0];
4239 6226 : slot = path->slots[0];
4240 : nritems = btrfs_header_nritems(l);
4241 6226 : mid = (nritems + 1) / 2;
4242 :
4243 6226 : if (mid <= slot) {
4244 5390 : if (nritems == 1 ||
4245 2678 : leaf_space_used(l, mid, nritems - mid) + data_size >
4246 2678 : BTRFS_LEAF_DATA_SIZE(root)) {
4247 504 : if (slot >= nritems) {
4248 : split = 0;
4249 : } else {
4250 : mid = slot;
4251 652 : if (mid != nritems &&
4252 326 : leaf_space_used(l, mid, nritems - mid) +
4253 326 : data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4254 293 : if (data_size && !tried_avoid_double)
4255 : goto push_for_double;
4256 : split = 2;
4257 : }
4258 : }
4259 : }
4260 : } else {
4261 7028 : if (leaf_space_used(l, 0, mid) + data_size >
4262 3514 : BTRFS_LEAF_DATA_SIZE(root)) {
4263 2672 : if (!extend && data_size && slot == 0) {
4264 : split = 0;
4265 1707 : } else if ((extend || !data_size) && slot == 0) {
4266 : mid = 1;
4267 : } else {
4268 : mid = slot;
4269 3414 : if (mid != nritems &&
4270 1707 : leaf_space_used(l, mid, nritems - mid) +
4271 1707 : data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4272 1596 : if (data_size && !tried_avoid_double)
4273 : goto push_for_double;
4274 : split = 2;
4275 : }
4276 : }
4277 : }
4278 : }
4279 :
4280 5280 : if (split == 0)
4281 : btrfs_cpu_key_to_disk(&disk_key, ins_key);
4282 : else
4283 : btrfs_item_key(l, &disk_key, mid);
4284 :
4285 5280 : right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
4286 : root->root_key.objectid,
4287 : &disk_key, 0, l->start, 0);
4288 5280 : if (IS_ERR(right))
4289 0 : return PTR_ERR(right);
4290 :
4291 5280 : root_add_used(root, root->leafsize);
4292 :
4293 5280 : memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4294 5280 : btrfs_set_header_bytenr(right, right->start);
4295 5280 : btrfs_set_header_generation(right, trans->transid);
4296 : btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4297 5280 : btrfs_set_header_owner(right, root->root_key.objectid);
4298 : btrfs_set_header_level(right, 0);
4299 5280 : write_extent_buffer(right, root->fs_info->fsid,
4300 : btrfs_header_fsid(), BTRFS_FSID_SIZE);
4301 :
4302 5280 : write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
4303 : btrfs_header_chunk_tree_uuid(right),
4304 : BTRFS_UUID_SIZE);
4305 :
4306 5280 : if (split == 0) {
4307 1143 : if (mid <= slot) {
4308 : btrfs_set_header_nritems(right, 0);
4309 178 : insert_ptr(trans, root, path, &disk_key, right->start,
4310 178 : path->slots[1] + 1, 1);
4311 178 : btrfs_tree_unlock(path->nodes[0]);
4312 178 : free_extent_buffer(path->nodes[0]);
4313 178 : path->nodes[0] = right;
4314 178 : path->slots[0] = 0;
4315 178 : path->slots[1] += 1;
4316 : } else {
4317 : btrfs_set_header_nritems(right, 0);
4318 965 : insert_ptr(trans, root, path, &disk_key, right->start,
4319 : path->slots[1], 1);
4320 965 : btrfs_tree_unlock(path->nodes[0]);
4321 965 : free_extent_buffer(path->nodes[0]);
4322 965 : path->nodes[0] = right;
4323 965 : path->slots[0] = 0;
4324 965 : if (path->slots[1] == 0)
4325 0 : fixup_low_keys(root, path, &disk_key, 1);
4326 : }
4327 1143 : btrfs_mark_buffer_dirty(right);
4328 1143 : return ret;
4329 : }
4330 :
4331 4137 : copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4332 :
4333 4137 : if (split == 2) {
4334 943 : BUG_ON(num_doubles != 0);
4335 943 : num_doubles++;
4336 943 : goto again;
4337 : }
4338 :
4339 : return 0;
4340 :
4341 : push_for_double:
4342 946 : push_for_double_split(trans, root, path, data_size);
4343 : tried_avoid_double = 1;
4344 946 : if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4345 : return 0;
4346 : goto again;
4347 : }
4348 :
4349 3246 : static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4350 : struct btrfs_root *root,
4351 : struct btrfs_path *path, int ins_len)
4352 : {
4353 : struct btrfs_key key;
4354 : struct extent_buffer *leaf;
4355 : struct btrfs_file_extent_item *fi;
4356 : u64 extent_len = 0;
4357 : u32 item_size;
4358 : int ret;
4359 :
4360 3246 : leaf = path->nodes[0];
4361 3246 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4362 :
4363 3246 : BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4364 : key.type != BTRFS_EXTENT_CSUM_KEY);
4365 :
4366 3246 : if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4367 : return 0;
4368 :
4369 125 : item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4370 125 : if (key.type == BTRFS_EXTENT_DATA_KEY) {
4371 244 : fi = btrfs_item_ptr(leaf, path->slots[0],
4372 : struct btrfs_file_extent_item);
4373 : extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4374 : }
4375 125 : btrfs_release_path(path);
4376 :
4377 125 : path->keep_locks = 1;
4378 125 : path->search_for_split = 1;
4379 125 : ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4380 125 : path->search_for_split = 0;
4381 125 : if (ret < 0)
4382 : goto err;
4383 :
4384 : ret = -EAGAIN;
4385 125 : leaf = path->nodes[0];
4386 : /* if our item isn't there or got smaller, return now */
4387 250 : if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4388 : goto err;
4389 :
4390 : /* the leaf has changed, it now has room. return now */
4391 125 : if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4392 : goto err;
4393 :
4394 123 : if (key.type == BTRFS_EXTENT_DATA_KEY) {
4395 240 : fi = btrfs_item_ptr(leaf, path->slots[0],
4396 : struct btrfs_file_extent_item);
4397 120 : if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4398 : goto err;
4399 : }
4400 :
4401 123 : btrfs_set_path_blocking(path);
4402 123 : ret = split_leaf(trans, root, &key, path, ins_len, 1);
4403 123 : if (ret)
4404 : goto err;
4405 :
4406 123 : path->keep_locks = 0;
4407 123 : btrfs_unlock_up_safe(path, 1);
4408 123 : return 0;
4409 : err:
4410 2 : path->keep_locks = 0;
4411 2 : return ret;
4412 : }
4413 :
4414 1183 : static noinline int split_item(struct btrfs_trans_handle *trans,
4415 : struct btrfs_root *root,
4416 : struct btrfs_path *path,
4417 : struct btrfs_key *new_key,
4418 : unsigned long split_offset)
4419 : {
4420 2366 : struct extent_buffer *leaf;
4421 : struct btrfs_item *item;
4422 : struct btrfs_item *new_item;
4423 : int slot;
4424 : char *buf;
4425 : u32 nritems;
4426 : u32 item_size;
4427 : u32 orig_offset;
4428 : struct btrfs_disk_key disk_key;
4429 :
4430 1183 : leaf = path->nodes[0];
4431 1183 : BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4432 :
4433 1183 : btrfs_set_path_blocking(path);
4434 :
4435 1183 : item = btrfs_item_nr(path->slots[0]);
4436 : orig_offset = btrfs_item_offset(leaf, item);
4437 : item_size = btrfs_item_size(leaf, item);
4438 :
4439 1183 : buf = kmalloc(item_size, GFP_NOFS);
4440 1183 : if (!buf)
4441 : return -ENOMEM;
4442 :
4443 2366 : read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4444 : path->slots[0]), item_size);
4445 :
4446 1183 : slot = path->slots[0] + 1;
4447 : nritems = btrfs_header_nritems(leaf);
4448 1183 : if (slot != nritems) {
4449 : /* shift the items */
4450 3498 : memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4451 : btrfs_item_nr_offset(slot),
4452 1166 : (nritems - slot) * sizeof(struct btrfs_item));
4453 : }
4454 :
4455 : btrfs_cpu_key_to_disk(&disk_key, new_key);
4456 : btrfs_set_item_key(leaf, &disk_key, slot);
4457 :
4458 : new_item = btrfs_item_nr(slot);
4459 :
4460 : btrfs_set_item_offset(leaf, new_item, orig_offset);
4461 1183 : btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4462 :
4463 1183 : btrfs_set_item_offset(leaf, item,
4464 1183 : orig_offset + item_size - split_offset);
4465 : btrfs_set_item_size(leaf, item, split_offset);
4466 :
4467 1183 : btrfs_set_header_nritems(leaf, nritems + 1);
4468 :
4469 : /* write the data for the start of the original item */
4470 1183 : write_extent_buffer(leaf, buf,
4471 1183 : btrfs_item_ptr_offset(leaf, path->slots[0]),
4472 : split_offset);
4473 :
4474 : /* write the data for the new item */
4475 2366 : write_extent_buffer(leaf, buf + split_offset,
4476 : btrfs_item_ptr_offset(leaf, slot),
4477 : item_size - split_offset);
4478 1183 : btrfs_mark_buffer_dirty(leaf);
4479 :
4480 1183 : BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4481 1183 : kfree(buf);
4482 : return 0;
4483 : }
4484 :
4485 : /*
4486 : * This function splits a single item into two items,
4487 : * giving 'new_key' to the new item and splitting the
4488 : * old one at split_offset (from the start of the item).
4489 : *
4490 : * The path may be released by this operation. After
4491 : * the split, the path is pointing to the old item. The
4492 : * new item is going to be in the same node as the old one.
4493 : *
4494 : * Note, the item being split must be smaller enough to live alone on
4495 : * a tree block with room for one extra struct btrfs_item
4496 : *
4497 : * This allows us to split the item in place, keeping a lock on the
4498 : * leaf the entire time.
4499 : */
4500 1183 : int btrfs_split_item(struct btrfs_trans_handle *trans,
4501 : struct btrfs_root *root,
4502 : struct btrfs_path *path,
4503 : struct btrfs_key *new_key,
4504 : unsigned long split_offset)
4505 : {
4506 : int ret;
4507 1183 : ret = setup_leaf_for_split(trans, root, path,
4508 : sizeof(struct btrfs_item));
4509 1183 : if (ret)
4510 : return ret;
4511 :
4512 1183 : ret = split_item(trans, root, path, new_key, split_offset);
4513 1183 : return ret;
4514 : }
4515 :
4516 : /*
4517 : * This function duplicate a item, giving 'new_key' to the new item.
4518 : * It guarantees both items live in the same tree leaf and the new item
4519 : * is contiguous with the original item.
4520 : *
4521 : * This allows us to split file extent in place, keeping a lock on the
4522 : * leaf the entire time.
4523 : */
4524 2063 : int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4525 : struct btrfs_root *root,
4526 : struct btrfs_path *path,
4527 : struct btrfs_key *new_key)
4528 : {
4529 : struct extent_buffer *leaf;
4530 : int ret;
4531 : u32 item_size;
4532 :
4533 2063 : leaf = path->nodes[0];
4534 4126 : item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4535 2063 : ret = setup_leaf_for_split(trans, root, path,
4536 : item_size + sizeof(struct btrfs_item));
4537 2063 : if (ret)
4538 : return ret;
4539 :
4540 2061 : path->slots[0]++;
4541 2061 : setup_items_for_insert(root, path, new_key, &item_size,
4542 : item_size, item_size +
4543 : sizeof(struct btrfs_item), 1);
4544 2061 : leaf = path->nodes[0];
4545 6183 : memcpy_extent_buffer(leaf,
4546 2061 : btrfs_item_ptr_offset(leaf, path->slots[0]),
4547 2061 : btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4548 : item_size);
4549 2061 : return 0;
4550 : }
4551 :
4552 : /*
4553 : * make the item pointed to by the path smaller. new_size indicates
4554 : * how small to make it, and from_end tells us if we just chop bytes
4555 : * off the end of the item or if we shift the item to chop bytes off
4556 : * the front.
4557 : */
4558 16006 : void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4559 : u32 new_size, int from_end)
4560 : {
4561 : int slot;
4562 16006 : struct extent_buffer *leaf;
4563 : struct btrfs_item *item;
4564 : u32 nritems;
4565 : unsigned int data_end;
4566 : unsigned int old_data_start;
4567 : unsigned int old_size;
4568 : unsigned int size_diff;
4569 : int i;
4570 : struct btrfs_map_token token;
4571 :
4572 : btrfs_init_map_token(&token);
4573 :
4574 16006 : leaf = path->nodes[0];
4575 16006 : slot = path->slots[0];
4576 :
4577 : old_size = btrfs_item_size_nr(leaf, slot);
4578 16006 : if (old_size == new_size)
4579 0 : return;
4580 :
4581 : nritems = btrfs_header_nritems(leaf);
4582 16006 : data_end = leaf_data_end(root, leaf);
4583 :
4584 : old_data_start = btrfs_item_offset_nr(leaf, slot);
4585 :
4586 16006 : size_diff = old_size - new_size;
4587 :
4588 16006 : BUG_ON(slot < 0);
4589 16006 : BUG_ON(slot >= nritems);
4590 :
4591 : /*
4592 : * item0..itemN ... dataN.offset..dataN.size .. data0.size
4593 : */
4594 : /* first correct the data pointers */
4595 1293333 : for (i = slot; i < nritems; i++) {
4596 : u32 ioff;
4597 : item = btrfs_item_nr(i);
4598 :
4599 : ioff = btrfs_token_item_offset(leaf, item, &token);
4600 1293333 : btrfs_set_token_item_offset(leaf, item,
4601 : ioff + size_diff, &token);
4602 : }
4603 :
4604 : /* shift the data */
4605 16006 : if (from_end) {
4606 10801 : memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4607 : data_end + size_diff, btrfs_leaf_data(leaf) +
4608 10801 : data_end, old_data_start + new_size - data_end);
4609 : } else {
4610 : struct btrfs_disk_key disk_key;
4611 : u64 offset;
4612 :
4613 : btrfs_item_key(leaf, &disk_key, slot);
4614 :
4615 5205 : if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4616 : unsigned long ptr;
4617 : struct btrfs_file_extent_item *fi;
4618 :
4619 0 : fi = btrfs_item_ptr(leaf, slot,
4620 : struct btrfs_file_extent_item);
4621 0 : fi = (struct btrfs_file_extent_item *)(
4622 0 : (unsigned long)fi - size_diff);
4623 :
4624 0 : if (btrfs_file_extent_type(leaf, fi) ==
4625 : BTRFS_FILE_EXTENT_INLINE) {
4626 0 : ptr = btrfs_item_ptr_offset(leaf, slot);
4627 0 : memmove_extent_buffer(leaf, ptr,
4628 : (unsigned long)fi,
4629 : offsetof(struct btrfs_file_extent_item,
4630 : disk_bytenr));
4631 : }
4632 : }
4633 :
4634 5205 : memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4635 : data_end + size_diff, btrfs_leaf_data(leaf) +
4636 5205 : data_end, old_data_start - data_end);
4637 :
4638 : offset = btrfs_disk_key_offset(&disk_key);
4639 5205 : btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4640 : btrfs_set_item_key(leaf, &disk_key, slot);
4641 5205 : if (slot == 0)
4642 1133 : fixup_low_keys(root, path, &disk_key, 1);
4643 : }
4644 :
4645 : item = btrfs_item_nr(slot);
4646 : btrfs_set_item_size(leaf, item, new_size);
4647 16006 : btrfs_mark_buffer_dirty(leaf);
4648 :
4649 16006 : if (btrfs_leaf_free_space(root, leaf) < 0) {
4650 0 : btrfs_print_leaf(root, leaf);
4651 0 : BUG();
4652 : }
4653 : }
4654 :
4655 : /*
4656 : * make the item pointed to by the path bigger, data_size is the added size.
4657 : */
4658 98804 : void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4659 : u32 data_size)
4660 : {
4661 : int slot;
4662 98804 : struct extent_buffer *leaf;
4663 : struct btrfs_item *item;
4664 : u32 nritems;
4665 : unsigned int data_end;
4666 : unsigned int old_data;
4667 : unsigned int old_size;
4668 : int i;
4669 : struct btrfs_map_token token;
4670 :
4671 : btrfs_init_map_token(&token);
4672 :
4673 98804 : leaf = path->nodes[0];
4674 :
4675 : nritems = btrfs_header_nritems(leaf);
4676 98804 : data_end = leaf_data_end(root, leaf);
4677 :
4678 98804 : if (btrfs_leaf_free_space(root, leaf) < data_size) {
4679 0 : btrfs_print_leaf(root, leaf);
4680 0 : BUG();
4681 : }
4682 98804 : slot = path->slots[0];
4683 : old_data = btrfs_item_end_nr(leaf, slot);
4684 :
4685 98823 : BUG_ON(slot < 0);
4686 98823 : if (slot >= nritems) {
4687 0 : btrfs_print_leaf(root, leaf);
4688 0 : btrfs_crit(root->fs_info, "slot %d too large, nritems %d",
4689 : slot, nritems);
4690 0 : BUG_ON(1);
4691 : }
4692 :
4693 : /*
4694 : * item0..itemN ... dataN.offset..dataN.size .. data0.size
4695 : */
4696 : /* first correct the data pointers */
4697 1700248 : for (i = slot; i < nritems; i++) {
4698 : u32 ioff;
4699 : item = btrfs_item_nr(i);
4700 :
4701 : ioff = btrfs_token_item_offset(leaf, item, &token);
4702 1700249 : btrfs_set_token_item_offset(leaf, item,
4703 : ioff - data_size, &token);
4704 : }
4705 :
4706 : /* shift the data */
4707 98804 : memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4708 : data_end - data_size, btrfs_leaf_data(leaf) +
4709 98804 : data_end, old_data - data_end);
4710 :
4711 : data_end = old_data;
4712 : old_size = btrfs_item_size_nr(leaf, slot);
4713 : item = btrfs_item_nr(slot);
4714 98803 : btrfs_set_item_size(leaf, item, old_size + data_size);
4715 98803 : btrfs_mark_buffer_dirty(leaf);
4716 :
4717 98803 : if (btrfs_leaf_free_space(root, leaf) < 0) {
4718 0 : btrfs_print_leaf(root, leaf);
4719 0 : BUG();
4720 : }
4721 98802 : }
4722 :
4723 : /*
4724 : * this is a helper for btrfs_insert_empty_items, the main goal here is
4725 : * to save stack depth by doing the bulk of the work in a function
4726 : * that doesn't call btrfs_search_slot
4727 : */
4728 264568 : void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4729 : struct btrfs_key *cpu_key, u32 *data_size,
4730 : u32 total_data, u32 total_size, int nr)
4731 : {
4732 : struct btrfs_item *item;
4733 : int i;
4734 : u32 nritems;
4735 : unsigned int data_end;
4736 : struct btrfs_disk_key disk_key;
4737 529126 : struct extent_buffer *leaf;
4738 : int slot;
4739 : struct btrfs_map_token token;
4740 :
4741 : btrfs_init_map_token(&token);
4742 :
4743 264568 : leaf = path->nodes[0];
4744 264568 : slot = path->slots[0];
4745 :
4746 : nritems = btrfs_header_nritems(leaf);
4747 264568 : data_end = leaf_data_end(root, leaf);
4748 :
4749 264561 : if (btrfs_leaf_free_space(root, leaf) < total_size) {
4750 0 : btrfs_print_leaf(root, leaf);
4751 0 : btrfs_crit(root->fs_info, "not enough freespace need %u have %d",
4752 : total_size, btrfs_leaf_free_space(root, leaf));
4753 0 : BUG();
4754 : }
4755 :
4756 264543 : if (slot != nritems) {
4757 : unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4758 :
4759 183630 : if (old_data < data_end) {
4760 0 : btrfs_print_leaf(root, leaf);
4761 0 : btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d",
4762 : slot, old_data, data_end);
4763 0 : BUG_ON(1);
4764 : }
4765 : /*
4766 : * item0..itemN ... dataN.offset..dataN.size .. data0.size
4767 : */
4768 : /* first correct the data pointers */
4769 10554215 : for (i = slot; i < nritems; i++) {
4770 : u32 ioff;
4771 :
4772 : item = btrfs_item_nr( i);
4773 : ioff = btrfs_token_item_offset(leaf, item, &token);
4774 10554227 : btrfs_set_token_item_offset(leaf, item,
4775 : ioff - total_data, &token);
4776 : }
4777 : /* shift the items */
4778 367242 : memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4779 : btrfs_item_nr_offset(slot),
4780 183621 : (nritems - slot) * sizeof(struct btrfs_item));
4781 :
4782 : /* shift the data */
4783 183622 : memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4784 : data_end - total_data, btrfs_leaf_data(leaf) +
4785 183622 : data_end, old_data - data_end);
4786 : data_end = old_data;
4787 : }
4788 :
4789 : /* setup the item for the new data */
4790 310631 : for (i = 0; i < nr; i++) {
4791 310640 : btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4792 310640 : btrfs_set_item_key(leaf, &disk_key, slot + i);
4793 : item = btrfs_item_nr(slot + i);
4794 310655 : btrfs_set_token_item_offset(leaf, item,
4795 310655 : data_end - data_size[i], &token);
4796 310644 : data_end -= data_size[i];
4797 : btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4798 : }
4799 :
4800 264558 : btrfs_set_header_nritems(leaf, nritems + nr);
4801 :
4802 264558 : if (slot == 0) {
4803 : btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4804 3663 : fixup_low_keys(root, path, &disk_key, 1);
4805 : }
4806 264558 : btrfs_unlock_up_safe(path, 1);
4807 264576 : btrfs_mark_buffer_dirty(leaf);
4808 :
4809 264596 : if (btrfs_leaf_free_space(root, leaf) < 0) {
4810 0 : btrfs_print_leaf(root, leaf);
4811 0 : BUG();
4812 : }
4813 264544 : }
4814 :
4815 : /*
4816 : * Given a key and some data, insert items into the tree.
4817 : * This does all the path init required, making room in the tree if needed.
4818 : */
4819 212207 : int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4820 : struct btrfs_root *root,
4821 : struct btrfs_path *path,
4822 : struct btrfs_key *cpu_key, u32 *data_size,
4823 : int nr)
4824 : {
4825 : int ret = 0;
4826 : int slot;
4827 : int i;
4828 : u32 total_size = 0;
4829 : u32 total_data = 0;
4830 :
4831 456894 : for (i = 0; i < nr; i++)
4832 244687 : total_data += data_size[i];
4833 :
4834 212207 : total_size = total_data + (nr * sizeof(struct btrfs_item));
4835 212207 : ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4836 212209 : if (ret == 0)
4837 : return -EEXIST;
4838 210124 : if (ret < 0)
4839 : return ret;
4840 :
4841 208196 : slot = path->slots[0];
4842 208196 : BUG_ON(slot < 0);
4843 :
4844 208196 : setup_items_for_insert(root, path, cpu_key, data_size,
4845 : total_data, total_size, nr);
4846 208179 : return 0;
4847 : }
4848 :
4849 : /*
4850 : * Given a key and some data, insert an item into the tree.
4851 : * This does all the path init required, making room in the tree if needed.
4852 : */
4853 1136 : int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4854 : *root, struct btrfs_key *cpu_key, void *data, u32
4855 : data_size)
4856 : {
4857 : int ret = 0;
4858 : struct btrfs_path *path;
4859 : struct extent_buffer *leaf;
4860 : unsigned long ptr;
4861 :
4862 : path = btrfs_alloc_path();
4863 1136 : if (!path)
4864 : return -ENOMEM;
4865 : ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4866 1136 : if (!ret) {
4867 1136 : leaf = path->nodes[0];
4868 2272 : ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4869 1136 : write_extent_buffer(leaf, data, ptr, data_size);
4870 1136 : btrfs_mark_buffer_dirty(leaf);
4871 : }
4872 1136 : btrfs_free_path(path);
4873 1136 : return ret;
4874 : }
4875 :
4876 : /*
4877 : * delete the pointer from a given node.
4878 : *
4879 : * the tree should have been previously balanced so the deletion does not
4880 : * empty a node.
4881 : */
4882 1371 : static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4883 : int level, int slot)
4884 : {
4885 2742 : struct extent_buffer *parent = path->nodes[level];
4886 : u32 nritems;
4887 : int ret;
4888 :
4889 : nritems = btrfs_header_nritems(parent);
4890 1371 : if (slot != nritems - 1) {
4891 1215 : if (level)
4892 1215 : tree_mod_log_eb_move(root->fs_info, parent, slot,
4893 1215 : slot + 1, nritems - slot - 1);
4894 2430 : memmove_extent_buffer(parent,
4895 : btrfs_node_key_ptr_offset(slot),
4896 : btrfs_node_key_ptr_offset(slot + 1),
4897 : sizeof(struct btrfs_key_ptr) *
4898 1215 : (nritems - slot - 1));
4899 156 : } else if (level) {
4900 156 : ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4901 : MOD_LOG_KEY_REMOVE, GFP_NOFS);
4902 156 : BUG_ON(ret < 0);
4903 : }
4904 :
4905 : nritems--;
4906 : btrfs_set_header_nritems(parent, nritems);
4907 1371 : if (nritems == 0 && parent == root->node) {
4908 0 : BUG_ON(btrfs_header_level(root->node) != 1);
4909 : /* just turn the root into a leaf and break */
4910 : btrfs_set_header_level(root->node, 0);
4911 1371 : } else if (slot == 0) {
4912 : struct btrfs_disk_key disk_key;
4913 :
4914 80 : btrfs_node_key(parent, &disk_key, 0);
4915 80 : fixup_low_keys(root, path, &disk_key, level + 1);
4916 : }
4917 1371 : btrfs_mark_buffer_dirty(parent);
4918 1371 : }
4919 :
4920 : /*
4921 : * a helper function to delete the leaf pointed to by path->slots[1] and
4922 : * path->nodes[1].
4923 : *
4924 : * This deletes the pointer in path->nodes[1] and frees the leaf
4925 : * block extent. zero is returned if it all worked out, < 0 otherwise.
4926 : *
4927 : * The path must have already been setup for deleting the leaf, including
4928 : * all the proper balancing. path->nodes[1] must be locked.
4929 : */
4930 1360 : static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4931 : struct btrfs_root *root,
4932 : struct btrfs_path *path,
4933 1360 : struct extent_buffer *leaf)
4934 : {
4935 1360 : WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4936 1360 : del_ptr(root, path, 1, path->slots[1]);
4937 :
4938 : /*
4939 : * btrfs_free_extent is expensive, we want to make sure we
4940 : * aren't holding any locks when we call it
4941 : */
4942 1360 : btrfs_unlock_up_safe(path, 0);
4943 :
4944 1360 : root_sub_used(root, leaf->len);
4945 :
4946 : extent_buffer_get(leaf);
4947 1360 : btrfs_free_tree_block(trans, root, leaf, 0, 1);
4948 1360 : free_extent_buffer_stale(leaf);
4949 1360 : }
4950 : /*
4951 : * delete the item at the leaf level in path. If that empties
4952 : * the leaf, remove it from the tree
4953 : */
4954 124167 : int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4955 : struct btrfs_path *path, int slot, int nr)
4956 : {
4957 275542 : struct extent_buffer *leaf;
4958 : struct btrfs_item *item;
4959 : int last_off;
4960 : int dsize = 0;
4961 : int ret = 0;
4962 : int wret;
4963 : int i;
4964 : u32 nritems;
4965 : struct btrfs_map_token token;
4966 :
4967 : btrfs_init_map_token(&token);
4968 :
4969 124167 : leaf = path->nodes[0];
4970 248335 : last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4971 :
4972 272602 : for (i = 0; i < nr; i++)
4973 296869 : dsize += btrfs_item_size_nr(leaf, slot + i);
4974 :
4975 : nritems = btrfs_header_nritems(leaf);
4976 :
4977 124167 : if (slot + nr != nritems) {
4978 110684 : int data_end = leaf_data_end(root, leaf);
4979 :
4980 110684 : memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4981 : data_end + dsize,
4982 : btrfs_leaf_data(leaf) + data_end,
4983 110684 : last_off - data_end);
4984 :
4985 6122663 : for (i = slot + nr; i < nritems; i++) {
4986 : u32 ioff;
4987 :
4988 : item = btrfs_item_nr(i);
4989 : ioff = btrfs_token_item_offset(leaf, item, &token);
4990 6011958 : btrfs_set_token_item_offset(leaf, item,
4991 : ioff + dsize, &token);
4992 : }
4993 :
4994 221370 : memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4995 : btrfs_item_nr_offset(slot + nr),
4996 : sizeof(struct btrfs_item) *
4997 110685 : (nritems - slot - nr));
4998 : }
4999 124168 : btrfs_set_header_nritems(leaf, nritems - nr);
5000 : nritems -= nr;
5001 :
5002 : /* delete the leaf if we've emptied it */
5003 124168 : if (nritems == 0) {
5004 214 : if (leaf == root->node) {
5005 : btrfs_set_header_level(leaf, 0);
5006 : } else {
5007 186 : btrfs_set_path_blocking(path);
5008 186 : clean_tree_block(trans, root, leaf);
5009 186 : btrfs_del_leaf(trans, root, path, leaf);
5010 : }
5011 : } else {
5012 123954 : int used = leaf_space_used(leaf, 0, nritems);
5013 123951 : if (slot == 0) {
5014 : struct btrfs_disk_key disk_key;
5015 :
5016 : btrfs_item_key(leaf, &disk_key, 0);
5017 4153 : fixup_low_keys(root, path, &disk_key, 1);
5018 : }
5019 :
5020 : /* delete the leaf if it is mostly empty */
5021 123951 : if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
5022 : /* push_leaf_left fixes the path.
5023 : * make sure the path still points to our leaf
5024 : * for possible call to del_ptr below
5025 : */
5026 14022 : slot = path->slots[1];
5027 : extent_buffer_get(leaf);
5028 :
5029 14024 : btrfs_set_path_blocking(path);
5030 14024 : wret = push_leaf_left(trans, root, path, 1, 1,
5031 : 1, (u32)-1);
5032 14024 : if (wret < 0 && wret != -ENOSPC)
5033 : ret = wret;
5034 :
5035 27179 : if (path->nodes[0] == leaf &&
5036 : btrfs_header_nritems(leaf)) {
5037 12950 : wret = push_leaf_right(trans, root, path, 1,
5038 : 1, 1, 0);
5039 12950 : if (wret < 0 && wret != -ENOSPC)
5040 : ret = wret;
5041 : }
5042 :
5043 14024 : if (btrfs_header_nritems(leaf) == 0) {
5044 1174 : path->slots[1] = slot;
5045 1174 : btrfs_del_leaf(trans, root, path, leaf);
5046 1174 : free_extent_buffer(leaf);
5047 : ret = 0;
5048 : } else {
5049 : /* if we're still in the path, make sure
5050 : * we're dirty. Otherwise, one of the
5051 : * push_leaf functions must have already
5052 : * dirtied this buffer
5053 : */
5054 12850 : if (path->nodes[0] == leaf)
5055 12551 : btrfs_mark_buffer_dirty(leaf);
5056 12850 : free_extent_buffer(leaf);
5057 : }
5058 : } else {
5059 109929 : btrfs_mark_buffer_dirty(leaf);
5060 : }
5061 : }
5062 124169 : return ret;
5063 : }
5064 :
5065 : /*
5066 : * search the tree again to find a leaf with lesser keys
5067 : * returns 0 if it found something or 1 if there are no lesser leaves.
5068 : * returns < 0 on io errors.
5069 : *
5070 : * This may release the path, and so you may lose any locks held at the
5071 : * time you call it.
5072 : */
5073 164 : int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5074 : {
5075 : struct btrfs_key key;
5076 : struct btrfs_disk_key found_key;
5077 : int ret;
5078 :
5079 164 : btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5080 :
5081 164 : if (key.offset > 0) {
5082 164 : key.offset--;
5083 0 : } else if (key.type > 0) {
5084 0 : key.type--;
5085 0 : key.offset = (u64)-1;
5086 0 : } else if (key.objectid > 0) {
5087 0 : key.objectid--;
5088 0 : key.type = (u8)-1;
5089 0 : key.offset = (u64)-1;
5090 : } else {
5091 : return 1;
5092 : }
5093 :
5094 164 : btrfs_release_path(path);
5095 164 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5096 164 : if (ret < 0)
5097 : return ret;
5098 164 : btrfs_item_key(path->nodes[0], &found_key, 0);
5099 : ret = comp_keys(&found_key, &key);
5100 : /*
5101 : * We might have had an item with the previous key in the tree right
5102 : * before we released our path. And after we released our path, that
5103 : * item might have been pushed to the first slot (0) of the leaf we
5104 : * were holding due to a tree balance. Alternatively, an item with the
5105 : * previous key can exist as the only element of a leaf (big fat item).
5106 : * Therefore account for these 2 cases, so that our callers (like
5107 : * btrfs_previous_item) don't miss an existing item with a key matching
5108 : * the previous key we computed above.
5109 : */
5110 164 : if (ret <= 0)
5111 : return 0;
5112 80 : return 1;
5113 : }
5114 :
5115 : /*
5116 : * A helper function to walk down the tree starting at min_key, and looking
5117 : * for nodes or leaves that are have a minimum transaction id.
5118 : * This is used by the btree defrag code, and tree logging
5119 : *
5120 : * This does not cow, but it does stuff the starting key it finds back
5121 : * into min_key, so you can call btrfs_search_slot with cow=1 on the
5122 : * key and get a writable path.
5123 : *
5124 : * This does lock as it descends, and path->keep_locks should be set
5125 : * to 1 by the caller.
5126 : *
5127 : * This honors path->lowest_level to prevent descent past a given level
5128 : * of the tree.
5129 : *
5130 : * min_trans indicates the oldest transaction that you are interested
5131 : * in walking through. Any nodes or leaves older than min_trans are
5132 : * skipped over (without reading them).
5133 : *
5134 : * returns zero if something useful was found, < 0 on error and 1 if there
5135 : * was nothing in the tree that matched the search criteria.
5136 : */
5137 2536 : int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5138 : struct btrfs_path *path,
5139 : u64 min_trans)
5140 : {
5141 8832 : struct extent_buffer *cur;
5142 : struct btrfs_key found_key;
5143 : int slot;
5144 : int sret;
5145 : u32 nritems;
5146 : int level;
5147 : int ret = 1;
5148 :
5149 2536 : WARN_ON(!path->keep_locks);
5150 : again:
5151 2627 : cur = btrfs_read_lock_root_node(root);
5152 2627 : level = btrfs_header_level(cur);
5153 2627 : WARN_ON(path->nodes[level]);
5154 2627 : path->nodes[level] = cur;
5155 2627 : path->locks[level] = BTRFS_READ_LOCK;
5156 :
5157 2627 : if (btrfs_header_generation(cur) < min_trans) {
5158 : ret = 1;
5159 : goto out;
5160 : }
5161 : while (1) {
5162 : nritems = btrfs_header_nritems(cur);
5163 3578 : level = btrfs_header_level(cur);
5164 3578 : sret = bin_search(cur, min_key, level, &slot);
5165 :
5166 : /* at the lowest level, we're done, setup the path and exit */
5167 3578 : if (level == path->lowest_level) {
5168 2623 : if (slot >= nritems)
5169 : goto find_next_key;
5170 : ret = 0;
5171 2277 : path->slots[level] = slot;
5172 2277 : btrfs_item_key_to_cpu(cur, &found_key, slot);
5173 2277 : goto out;
5174 : }
5175 955 : if (sret && slot > 0)
5176 851 : slot--;
5177 : /*
5178 : * check this node pointer against the min_trans parameters.
5179 : * If it is too old, old, skip to the next one.
5180 : */
5181 976 : while (slot < nritems) {
5182 : u64 gen;
5183 :
5184 : gen = btrfs_node_ptr_generation(cur, slot);
5185 976 : if (gen < min_trans) {
5186 21 : slot++;
5187 21 : continue;
5188 : }
5189 : break;
5190 : }
5191 : find_next_key:
5192 : /*
5193 : * we didn't find a candidate key in this node, walk forward
5194 : * and find another one
5195 : */
5196 1301 : if (slot >= nritems) {
5197 346 : path->slots[level] = slot;
5198 346 : btrfs_set_path_blocking(path);
5199 346 : sret = btrfs_find_next_key(root, path, min_key, level,
5200 : min_trans);
5201 346 : if (sret == 0) {
5202 91 : btrfs_release_path(path);
5203 91 : goto again;
5204 : } else {
5205 : goto out;
5206 : }
5207 : }
5208 : /* save our key for returning back */
5209 : btrfs_node_key_to_cpu(cur, &found_key, slot);
5210 955 : path->slots[level] = slot;
5211 955 : if (level == path->lowest_level) {
5212 : ret = 0;
5213 0 : unlock_up(path, level, 1, 0, NULL);
5214 0 : goto out;
5215 : }
5216 955 : btrfs_set_path_blocking(path);
5217 955 : cur = read_node_slot(root, cur, slot);
5218 955 : BUG_ON(!cur); /* -ENOMEM */
5219 :
5220 955 : btrfs_tree_read_lock(cur);
5221 :
5222 955 : path->locks[level - 1] = BTRFS_READ_LOCK;
5223 955 : path->nodes[level - 1] = cur;
5224 955 : unlock_up(path, level, 1, 0, NULL);
5225 955 : btrfs_clear_path_blocking(path, NULL, 0);
5226 955 : }
5227 : out:
5228 2536 : if (ret == 0)
5229 2277 : memcpy(min_key, &found_key, sizeof(found_key));
5230 2536 : btrfs_set_path_blocking(path);
5231 2536 : return ret;
5232 : }
5233 :
5234 25 : static void tree_move_down(struct btrfs_root *root,
5235 : struct btrfs_path *path,
5236 : int *level, int root_level)
5237 : {
5238 25 : BUG_ON(*level == 0);
5239 25 : path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
5240 : path->slots[*level]);
5241 25 : path->slots[*level - 1] = 0;
5242 25 : (*level)--;
5243 25 : }
5244 :
5245 3684 : static int tree_move_next_or_upnext(struct btrfs_root *root,
5246 : struct btrfs_path *path,
5247 : int *level, int root_level)
5248 : {
5249 : int ret = 0;
5250 : int nritems;
5251 7393 : nritems = btrfs_header_nritems(path->nodes[*level]);
5252 :
5253 3684 : path->slots[*level]++;
5254 :
5255 3709 : while (path->slots[*level] >= nritems) {
5256 57 : if (*level == root_level)
5257 : return -1;
5258 :
5259 : /* move upnext */
5260 25 : path->slots[*level] = 0;
5261 25 : free_extent_buffer(path->nodes[*level]);
5262 25 : path->nodes[*level] = NULL;
5263 25 : (*level)++;
5264 25 : path->slots[*level]++;
5265 :
5266 50 : nritems = btrfs_header_nritems(path->nodes[*level]);
5267 : ret = 1;
5268 : }
5269 : return ret;
5270 : }
5271 :
5272 : /*
5273 : * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5274 : * or down.
5275 : */
5276 3709 : static int tree_advance(struct btrfs_root *root,
5277 : struct btrfs_path *path,
5278 : int *level, int root_level,
5279 : int allow_down,
5280 : struct btrfs_key *key)
5281 : {
5282 : int ret;
5283 :
5284 3709 : if (*level == 0 || !allow_down) {
5285 3684 : ret = tree_move_next_or_upnext(root, path, level, root_level);
5286 : } else {
5287 25 : tree_move_down(root, path, level, root_level);
5288 : ret = 0;
5289 : }
5290 3709 : if (ret >= 0) {
5291 3677 : if (*level == 0)
5292 3650 : btrfs_item_key_to_cpu(path->nodes[*level], key,
5293 : path->slots[*level]);
5294 : else
5295 27 : btrfs_node_key_to_cpu(path->nodes[*level], key,
5296 : path->slots[*level]);
5297 : }
5298 3709 : return ret;
5299 : }
5300 :
5301 807 : static int tree_compare_item(struct btrfs_root *left_root,
5302 : struct btrfs_path *left_path,
5303 : struct btrfs_path *right_path,
5304 : char *tmp_buf)
5305 : {
5306 : int cmp;
5307 : int len1, len2;
5308 : unsigned long off1, off2;
5309 :
5310 1614 : len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5311 1614 : len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5312 807 : if (len1 != len2)
5313 : return 1;
5314 :
5315 1586 : off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5316 1586 : off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5317 : right_path->slots[0]);
5318 :
5319 793 : read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5320 :
5321 793 : cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5322 793 : if (cmp)
5323 : return 1;
5324 : return 0;
5325 : }
5326 :
5327 : #define ADVANCE 1
5328 : #define ADVANCE_ONLY_NEXT -1
5329 :
5330 : /*
5331 : * This function compares two trees and calls the provided callback for
5332 : * every changed/new/deleted item it finds.
5333 : * If shared tree blocks are encountered, whole subtrees are skipped, making
5334 : * the compare pretty fast on snapshotted subvolumes.
5335 : *
5336 : * This currently works on commit roots only. As commit roots are read only,
5337 : * we don't do any locking. The commit roots are protected with transactions.
5338 : * Transactions are ended and rejoined when a commit is tried in between.
5339 : *
5340 : * This function checks for modifications done to the trees while comparing.
5341 : * If it detects a change, it aborts immediately.
5342 : */
5343 16 : int btrfs_compare_trees(struct btrfs_root *left_root,
5344 : struct btrfs_root *right_root,
5345 : btrfs_changed_cb_t changed_cb, void *ctx)
5346 : {
5347 : int ret;
5348 : int cmp;
5349 : struct btrfs_path *left_path = NULL;
5350 : struct btrfs_path *right_path = NULL;
5351 : struct btrfs_key left_key;
5352 : struct btrfs_key right_key;
5353 : char *tmp_buf = NULL;
5354 : int left_root_level;
5355 : int right_root_level;
5356 : int left_level;
5357 : int right_level;
5358 : int left_end_reached;
5359 : int right_end_reached;
5360 : int advance_left;
5361 : int advance_right;
5362 : u64 left_blockptr;
5363 : u64 right_blockptr;
5364 : u64 left_gen;
5365 : u64 right_gen;
5366 :
5367 : left_path = btrfs_alloc_path();
5368 16 : if (!left_path) {
5369 : ret = -ENOMEM;
5370 : goto out;
5371 : }
5372 : right_path = btrfs_alloc_path();
5373 16 : if (!right_path) {
5374 : ret = -ENOMEM;
5375 : goto out;
5376 : }
5377 :
5378 16 : tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS);
5379 16 : if (!tmp_buf) {
5380 : ret = -ENOMEM;
5381 : goto out;
5382 : }
5383 :
5384 16 : left_path->search_commit_root = 1;
5385 16 : left_path->skip_locking = 1;
5386 16 : right_path->search_commit_root = 1;
5387 16 : right_path->skip_locking = 1;
5388 :
5389 : /*
5390 : * Strategy: Go to the first items of both trees. Then do
5391 : *
5392 : * If both trees are at level 0
5393 : * Compare keys of current items
5394 : * If left < right treat left item as new, advance left tree
5395 : * and repeat
5396 : * If left > right treat right item as deleted, advance right tree
5397 : * and repeat
5398 : * If left == right do deep compare of items, treat as changed if
5399 : * needed, advance both trees and repeat
5400 : * If both trees are at the same level but not at level 0
5401 : * Compare keys of current nodes/leafs
5402 : * If left < right advance left tree and repeat
5403 : * If left > right advance right tree and repeat
5404 : * If left == right compare blockptrs of the next nodes/leafs
5405 : * If they match advance both trees but stay at the same level
5406 : * and repeat
5407 : * If they don't match advance both trees while allowing to go
5408 : * deeper and repeat
5409 : * If tree levels are different
5410 : * Advance the tree that needs it and repeat
5411 : *
5412 : * Advancing a tree means:
5413 : * If we are at level 0, try to go to the next slot. If that's not
5414 : * possible, go one level up and repeat. Stop when we found a level
5415 : * where we could go to the next slot. We may at this point be on a
5416 : * node or a leaf.
5417 : *
5418 : * If we are not at level 0 and not on shared tree blocks, go one
5419 : * level deeper.
5420 : *
5421 : * If we are not at level 0 and on shared tree blocks, go one slot to
5422 : * the right if possible or go up and right.
5423 : */
5424 :
5425 16 : down_read(&left_root->fs_info->commit_root_sem);
5426 32 : left_level = btrfs_header_level(left_root->commit_root);
5427 : left_root_level = left_level;
5428 16 : left_path->nodes[left_level] = left_root->commit_root;
5429 16 : extent_buffer_get(left_path->nodes[left_level]);
5430 :
5431 32 : right_level = btrfs_header_level(right_root->commit_root);
5432 : right_root_level = right_level;
5433 16 : right_path->nodes[right_level] = right_root->commit_root;
5434 16 : extent_buffer_get(right_path->nodes[right_level]);
5435 16 : up_read(&left_root->fs_info->commit_root_sem);
5436 :
5437 16 : if (left_level == 0)
5438 12 : btrfs_item_key_to_cpu(left_path->nodes[left_level],
5439 : &left_key, left_path->slots[left_level]);
5440 : else
5441 4 : btrfs_node_key_to_cpu(left_path->nodes[left_level],
5442 : &left_key, left_path->slots[left_level]);
5443 16 : if (right_level == 0)
5444 12 : btrfs_item_key_to_cpu(right_path->nodes[right_level],
5445 : &right_key, right_path->slots[right_level]);
5446 : else
5447 4 : btrfs_node_key_to_cpu(right_path->nodes[right_level],
5448 : &right_key, right_path->slots[right_level]);
5449 :
5450 : left_end_reached = right_end_reached = 0;
5451 : advance_left = advance_right = 0;
5452 :
5453 : while (1) {
5454 2907 : if (advance_left && !left_end_reached) {
5455 1234 : ret = tree_advance(left_root, left_path, &left_level,
5456 : left_root_level,
5457 : advance_left != ADVANCE_ONLY_NEXT,
5458 : &left_key);
5459 1234 : if (ret < 0)
5460 : left_end_reached = ADVANCE;
5461 : advance_left = 0;
5462 : }
5463 2907 : if (advance_right && !right_end_reached) {
5464 2475 : ret = tree_advance(right_root, right_path, &right_level,
5465 : right_root_level,
5466 : advance_right != ADVANCE_ONLY_NEXT,
5467 : &right_key);
5468 2475 : if (ret < 0)
5469 : right_end_reached = ADVANCE;
5470 : advance_right = 0;
5471 : }
5472 :
5473 2907 : if (left_end_reached && right_end_reached) {
5474 : ret = 0;
5475 : goto out;
5476 2891 : } else if (left_end_reached) {
5477 1421 : if (right_level == 0) {
5478 1413 : ret = changed_cb(left_root, right_root,
5479 : left_path, right_path,
5480 : &right_key,
5481 : BTRFS_COMPARE_TREE_DELETED,
5482 : ctx);
5483 1413 : if (ret < 0)
5484 : goto out;
5485 : }
5486 : advance_right = ADVANCE;
5487 1421 : continue;
5488 1470 : } else if (right_end_reached) {
5489 150 : if (left_level == 0) {
5490 149 : ret = changed_cb(left_root, right_root,
5491 : left_path, right_path,
5492 : &left_key,
5493 : BTRFS_COMPARE_TREE_NEW,
5494 : ctx);
5495 149 : if (ret < 0)
5496 : goto out;
5497 : }
5498 : advance_left = ADVANCE;
5499 150 : continue;
5500 : }
5501 :
5502 1320 : if (left_level == 0 && right_level == 0) {
5503 : cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5504 1305 : if (cmp < 0) {
5505 264 : ret = changed_cb(left_root, right_root,
5506 : left_path, right_path,
5507 : &left_key,
5508 : BTRFS_COMPARE_TREE_NEW,
5509 : ctx);
5510 264 : if (ret < 0)
5511 : goto out;
5512 : advance_left = ADVANCE;
5513 1041 : } else if (cmp > 0) {
5514 234 : ret = changed_cb(left_root, right_root,
5515 : left_path, right_path,
5516 : &right_key,
5517 : BTRFS_COMPARE_TREE_DELETED,
5518 : ctx);
5519 234 : if (ret < 0)
5520 : goto out;
5521 : advance_right = ADVANCE;
5522 : } else {
5523 : enum btrfs_compare_tree_result cmp;
5524 :
5525 807 : WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5526 807 : ret = tree_compare_item(left_root, left_path,
5527 : right_path, tmp_buf);
5528 807 : if (ret)
5529 : cmp = BTRFS_COMPARE_TREE_CHANGED;
5530 : else
5531 : cmp = BTRFS_COMPARE_TREE_SAME;
5532 807 : ret = changed_cb(left_root, right_root,
5533 : left_path, right_path,
5534 : &left_key, cmp, ctx);
5535 807 : if (ret < 0)
5536 : goto out;
5537 : advance_left = ADVANCE;
5538 : advance_right = ADVANCE;
5539 : }
5540 15 : } else if (left_level == right_level) {
5541 : cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5542 11 : if (cmp < 0) {
5543 : advance_left = ADVANCE;
5544 11 : } else if (cmp > 0) {
5545 : advance_right = ADVANCE;
5546 : } else {
5547 11 : left_blockptr = btrfs_node_blockptr(
5548 : left_path->nodes[left_level],
5549 : left_path->slots[left_level]);
5550 11 : right_blockptr = btrfs_node_blockptr(
5551 : right_path->nodes[right_level],
5552 : right_path->slots[right_level]);
5553 11 : left_gen = btrfs_node_ptr_generation(
5554 : left_path->nodes[left_level],
5555 : left_path->slots[left_level]);
5556 11 : right_gen = btrfs_node_ptr_generation(
5557 : right_path->nodes[right_level],
5558 : right_path->slots[right_level]);
5559 22 : if (left_blockptr == right_blockptr &&
5560 11 : left_gen == right_gen) {
5561 : /*
5562 : * As we're on a shared block, don't
5563 : * allow to go deeper.
5564 : */
5565 : advance_left = ADVANCE_ONLY_NEXT;
5566 : advance_right = ADVANCE_ONLY_NEXT;
5567 : } else {
5568 : advance_left = ADVANCE;
5569 : advance_right = ADVANCE;
5570 : }
5571 : }
5572 4 : } else if (left_level < right_level) {
5573 : advance_right = ADVANCE;
5574 : } else {
5575 : advance_left = ADVANCE;
5576 : }
5577 : }
5578 :
5579 : out:
5580 16 : btrfs_free_path(left_path);
5581 16 : btrfs_free_path(right_path);
5582 16 : kfree(tmp_buf);
5583 16 : return ret;
5584 : }
5585 :
5586 : /*
5587 : * this is similar to btrfs_next_leaf, but does not try to preserve
5588 : * and fixup the path. It looks for and returns the next key in the
5589 : * tree based on the current path and the min_trans parameters.
5590 : *
5591 : * 0 is returned if another key is found, < 0 if there are any errors
5592 : * and 1 is returned if there are no higher keys in the tree
5593 : *
5594 : * path->keep_locks should be set to 1 on the search made before
5595 : * calling this function.
5596 : */
5597 346 : int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5598 : struct btrfs_key *key, int level, u64 min_trans)
5599 : {
5600 : int slot;
5601 585 : struct extent_buffer *c;
5602 :
5603 346 : WARN_ON(!path->keep_locks);
5604 465 : while (level < BTRFS_MAX_LEVEL) {
5605 465 : if (!path->nodes[level])
5606 : return 1;
5607 :
5608 465 : slot = path->slots[level] + 1;
5609 : c = path->nodes[level];
5610 : next:
5611 1170 : if (slot >= btrfs_header_nritems(c)) {
5612 : int ret;
5613 : int orig_lowest;
5614 : struct btrfs_key cur_key;
5615 748 : if (level + 1 >= BTRFS_MAX_LEVEL ||
5616 374 : !path->nodes[level + 1])
5617 255 : return 1;
5618 :
5619 119 : if (path->locks[level + 1]) {
5620 : level++;
5621 119 : continue;
5622 : }
5623 :
5624 0 : slot = btrfs_header_nritems(c) - 1;
5625 0 : if (level == 0)
5626 0 : btrfs_item_key_to_cpu(c, &cur_key, slot);
5627 : else
5628 : btrfs_node_key_to_cpu(c, &cur_key, slot);
5629 :
5630 0 : orig_lowest = path->lowest_level;
5631 0 : btrfs_release_path(path);
5632 0 : path->lowest_level = level;
5633 0 : ret = btrfs_search_slot(NULL, root, &cur_key, path,
5634 : 0, 0);
5635 0 : path->lowest_level = orig_lowest;
5636 0 : if (ret < 0)
5637 : return ret;
5638 :
5639 0 : c = path->nodes[level];
5640 0 : slot = path->slots[level];
5641 0 : if (ret == 0)
5642 0 : slot++;
5643 0 : goto next;
5644 : }
5645 :
5646 211 : if (level == 0)
5647 0 : btrfs_item_key_to_cpu(c, key, slot);
5648 : else {
5649 : u64 gen = btrfs_node_ptr_generation(c, slot);
5650 :
5651 211 : if (gen < min_trans) {
5652 120 : slot++;
5653 120 : goto next;
5654 : }
5655 : btrfs_node_key_to_cpu(c, key, slot);
5656 : }
5657 : return 0;
5658 : }
5659 : return 1;
5660 : }
5661 :
5662 : /*
5663 : * search the tree again to find a leaf with greater keys
5664 : * returns 0 if it found something or 1 if there are no greater leaves.
5665 : * returns < 0 on io errors.
5666 : */
5667 25598 : int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5668 : {
5669 25697 : return btrfs_next_old_leaf(root, path, 0);
5670 : }
5671 :
5672 49082 : int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5673 : u64 time_seq)
5674 : {
5675 : int slot;
5676 : int level;
5677 38739 : struct extent_buffer *c;
5678 : struct extent_buffer *next;
5679 : struct btrfs_key key;
5680 : u32 nritems;
5681 : int ret;
5682 49082 : int old_spinning = path->leave_spinning;
5683 : int next_rw_lock = 0;
5684 :
5685 97032 : nritems = btrfs_header_nritems(path->nodes[0]);
5686 49082 : if (nritems == 0)
5687 : return 1;
5688 :
5689 47896 : btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5690 : again:
5691 : level = 1;
5692 47944 : next = NULL;
5693 : next_rw_lock = 0;
5694 47944 : btrfs_release_path(path);
5695 :
5696 47953 : path->keep_locks = 1;
5697 47953 : path->leave_spinning = 1;
5698 :
5699 47953 : if (time_seq)
5700 71 : ret = btrfs_search_old_slot(root, &key, path, time_seq);
5701 : else
5702 47882 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5703 47950 : path->keep_locks = 0;
5704 :
5705 47950 : if (ret < 0)
5706 : return ret;
5707 :
5708 47950 : nritems = btrfs_header_nritems(path->nodes[0]);
5709 : /*
5710 : * by releasing the path above we dropped all our locks. A balance
5711 : * could have added more items next to the key that used to be
5712 : * at the very end of the block. So, check again here and
5713 : * advance the path if there are now more items available.
5714 : */
5715 47950 : if (nritems > 0 && path->slots[0] < nritems - 1) {
5716 207 : if (ret == 0)
5717 207 : path->slots[0]++;
5718 : ret = 0;
5719 : goto done;
5720 : }
5721 : /*
5722 : * So the above check misses one case:
5723 : * - after releasing the path above, someone has removed the item that
5724 : * used to be at the very end of the block, and balance between leafs
5725 : * gets another one with bigger key.offset to replace it.
5726 : *
5727 : * This one should be returned as well, or we can get leaf corruption
5728 : * later(esp. in __btrfs_drop_extents()).
5729 : *
5730 : * And a bit more explanation about this check,
5731 : * with ret > 0, the key isn't found, the path points to the slot
5732 : * where it should be inserted, so the path->slots[0] item must be the
5733 : * bigger one.
5734 : */
5735 47743 : if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5736 : ret = 0;
5737 : goto done;
5738 : }
5739 :
5740 60633 : while (level < BTRFS_MAX_LEVEL) {
5741 60629 : if (!path->nodes[level]) {
5742 : ret = 1;
5743 : goto done;
5744 : }
5745 :
5746 38739 : slot = path->slots[level] + 1;
5747 : c = path->nodes[level];
5748 77478 : if (slot >= btrfs_header_nritems(c)) {
5749 12891 : level++;
5750 12891 : if (level == BTRFS_MAX_LEVEL) {
5751 : ret = 1;
5752 : goto done;
5753 : }
5754 12889 : continue;
5755 : }
5756 :
5757 25848 : if (next) {
5758 0 : btrfs_tree_unlock_rw(next, next_rw_lock);
5759 0 : free_extent_buffer(next);
5760 : }
5761 :
5762 25848 : next = c;
5763 25848 : next_rw_lock = path->locks[level];
5764 25848 : ret = read_block_for_search(NULL, root, path, &next, level,
5765 : slot, &key, 0);
5766 25848 : if (ret == -EAGAIN)
5767 : goto again;
5768 :
5769 25800 : if (ret < 0) {
5770 0 : btrfs_release_path(path);
5771 0 : goto done;
5772 : }
5773 :
5774 25800 : if (!path->skip_locking) {
5775 2814 : ret = btrfs_try_tree_read_lock(next);
5776 2814 : if (!ret && time_seq) {
5777 : /*
5778 : * If we don't get the lock, we may be racing
5779 : * with push_leaf_left, holding that lock while
5780 : * itself waiting for the leaf we've currently
5781 : * locked. To solve this situation, we give up
5782 : * on our lock and cycle.
5783 : */
5784 0 : free_extent_buffer(next);
5785 0 : btrfs_release_path(path);
5786 0 : cond_resched();
5787 0 : goto again;
5788 : }
5789 2814 : if (!ret) {
5790 0 : btrfs_set_path_blocking(path);
5791 0 : btrfs_tree_read_lock(next);
5792 0 : btrfs_clear_path_blocking(path, next,
5793 : BTRFS_READ_LOCK);
5794 : }
5795 : next_rw_lock = BTRFS_READ_LOCK;
5796 : }
5797 : break;
5798 : }
5799 25800 : path->slots[level] = slot;
5800 : while (1) {
5801 25824 : level--;
5802 25824 : c = path->nodes[level];
5803 25824 : if (path->locks[level])
5804 2836 : btrfs_tree_unlock_rw(c, path->locks[level]);
5805 :
5806 25824 : free_extent_buffer(c);
5807 25824 : path->nodes[level] = next;
5808 25824 : path->slots[level] = 0;
5809 25824 : if (!path->skip_locking)
5810 2836 : path->locks[level] = next_rw_lock;
5811 25824 : if (!level)
5812 : break;
5813 :
5814 24 : ret = read_block_for_search(NULL, root, path, &next, level,
5815 : 0, &key, 0);
5816 24 : if (ret == -EAGAIN)
5817 : goto again;
5818 :
5819 24 : if (ret < 0) {
5820 0 : btrfs_release_path(path);
5821 0 : goto done;
5822 : }
5823 :
5824 24 : if (!path->skip_locking) {
5825 22 : ret = btrfs_try_tree_read_lock(next);
5826 22 : if (!ret) {
5827 0 : btrfs_set_path_blocking(path);
5828 0 : btrfs_tree_read_lock(next);
5829 0 : btrfs_clear_path_blocking(path, next,
5830 : BTRFS_READ_LOCK);
5831 : }
5832 : next_rw_lock = BTRFS_READ_LOCK;
5833 : }
5834 : }
5835 : ret = 0;
5836 : done:
5837 47898 : unlock_up(path, 0, 1, 0, NULL);
5838 47901 : path->leave_spinning = old_spinning;
5839 47901 : if (!old_spinning)
5840 47832 : btrfs_set_path_blocking(path);
5841 :
5842 47920 : return ret;
5843 : }
5844 :
5845 : /*
5846 : * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5847 : * searching until it gets past min_objectid or finds an item of 'type'
5848 : *
5849 : * returns 0 if something is found, 1 if nothing was found and < 0 on error
5850 : */
5851 282 : int btrfs_previous_item(struct btrfs_root *root,
5852 : struct btrfs_path *path, u64 min_objectid,
5853 : int type)
5854 : {
5855 : struct btrfs_key found_key;
5856 282 : struct extent_buffer *leaf;
5857 : u32 nritems;
5858 : int ret;
5859 :
5860 : while (1) {
5861 320 : if (path->slots[0] == 0) {
5862 38 : btrfs_set_path_blocking(path);
5863 38 : ret = btrfs_prev_leaf(root, path);
5864 38 : if (ret != 0)
5865 : return ret;
5866 : } else {
5867 282 : path->slots[0]--;
5868 : }
5869 282 : leaf = path->nodes[0];
5870 : nritems = btrfs_header_nritems(leaf);
5871 282 : if (nritems == 0)
5872 : return 1;
5873 282 : if (path->slots[0] == nritems)
5874 0 : path->slots[0]--;
5875 :
5876 282 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5877 282 : if (found_key.objectid < min_objectid)
5878 : break;
5879 218 : if (found_key.type == type)
5880 : return 0;
5881 38 : if (found_key.objectid == min_objectid &&
5882 : found_key.type < type)
5883 : break;
5884 : }
5885 : return 1;
5886 : }
5887 :
5888 : /*
5889 : * search in extent tree to find a previous Metadata/Data extent item with
5890 : * min objecitd.
5891 : *
5892 : * returns 0 if something is found, 1 if nothing was found and < 0 on error
5893 : */
5894 33368 : int btrfs_previous_extent_item(struct btrfs_root *root,
5895 : struct btrfs_path *path, u64 min_objectid)
5896 : {
5897 : struct btrfs_key found_key;
5898 33697 : struct extent_buffer *leaf;
5899 : u32 nritems;
5900 : int ret;
5901 :
5902 : while (1) {
5903 33738 : if (path->slots[0] == 0) {
5904 105 : btrfs_set_path_blocking(path);
5905 105 : ret = btrfs_prev_leaf(root, path);
5906 105 : if (ret != 0)
5907 : return ret;
5908 : } else {
5909 33633 : path->slots[0]--;
5910 : }
5911 33697 : leaf = path->nodes[0];
5912 : nritems = btrfs_header_nritems(leaf);
5913 33697 : if (nritems == 0)
5914 : return 1;
5915 33697 : if (path->slots[0] == nritems)
5916 63 : path->slots[0]--;
5917 :
5918 33697 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5919 33697 : if (found_key.objectid < min_objectid)
5920 : break;
5921 33697 : if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5922 : found_key.type == BTRFS_METADATA_ITEM_KEY)
5923 : return 0;
5924 370 : if (found_key.objectid == min_objectid &&
5925 : found_key.type < BTRFS_EXTENT_ITEM_KEY)
5926 : break;
5927 : }
5928 : return 1;
5929 : }
|
