Linux 5.6.13
[linux/fpc-iii.git] / fs / btrfs / ctree.c
blobf2ec1a9bae287169e4ee6bb0494d14f4dc674cf8
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 */
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/rbtree.h>
9 #include <linux/mm.h>
10 #include "ctree.h"
11 #include "disk-io.h"
12 #include "transaction.h"
13 #include "print-tree.h"
14 #include "locking.h"
15 #include "volumes.h"
16 #include "qgroup.h"
18 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
19 *root, struct btrfs_path *path, int level);
20 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
21 const struct btrfs_key *ins_key, struct btrfs_path *path,
22 int data_size, int extend);
23 static int push_node_left(struct btrfs_trans_handle *trans,
24 struct extent_buffer *dst,
25 struct extent_buffer *src, int empty);
26 static int balance_node_right(struct btrfs_trans_handle *trans,
27 struct extent_buffer *dst_buf,
28 struct extent_buffer *src_buf);
29 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
30 int level, int slot);
32 static const struct btrfs_csums {
33 u16 size;
34 const char *name;
35 const char *driver;
36 } btrfs_csums[] = {
37 [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" },
38 [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" },
39 [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" },
40 [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b",
41 .driver = "blake2b-256" },
44 int btrfs_super_csum_size(const struct btrfs_super_block *s)
46 u16 t = btrfs_super_csum_type(s);
48 * csum type is validated at mount time
50 return btrfs_csums[t].size;
53 const char *btrfs_super_csum_name(u16 csum_type)
55 /* csum type is validated at mount time */
56 return btrfs_csums[csum_type].name;
60 * Return driver name if defined, otherwise the name that's also a valid driver
61 * name
63 const char *btrfs_super_csum_driver(u16 csum_type)
65 /* csum type is validated at mount time */
66 return btrfs_csums[csum_type].driver ?:
67 btrfs_csums[csum_type].name;
70 size_t __const btrfs_get_num_csums(void)
72 return ARRAY_SIZE(btrfs_csums);
75 struct btrfs_path *btrfs_alloc_path(void)
77 return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
80 /* this also releases the path */
81 void btrfs_free_path(struct btrfs_path *p)
83 if (!p)
84 return;
85 btrfs_release_path(p);
86 kmem_cache_free(btrfs_path_cachep, p);
90 * path release drops references on the extent buffers in the path
91 * and it drops any locks held by this path
93 * It is safe to call this on paths that no locks or extent buffers held.
95 noinline void btrfs_release_path(struct btrfs_path *p)
97 int i;
99 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
100 p->slots[i] = 0;
101 if (!p->nodes[i])
102 continue;
103 if (p->locks[i]) {
104 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
105 p->locks[i] = 0;
107 free_extent_buffer(p->nodes[i]);
108 p->nodes[i] = NULL;
113 * safely gets a reference on the root node of a tree. A lock
114 * is not taken, so a concurrent writer may put a different node
115 * at the root of the tree. See btrfs_lock_root_node for the
116 * looping required.
118 * The extent buffer returned by this has a reference taken, so
119 * it won't disappear. It may stop being the root of the tree
120 * at any time because there are no locks held.
122 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
124 struct extent_buffer *eb;
126 while (1) {
127 rcu_read_lock();
128 eb = rcu_dereference(root->node);
131 * RCU really hurts here, we could free up the root node because
132 * it was COWed but we may not get the new root node yet so do
133 * the inc_not_zero dance and if it doesn't work then
134 * synchronize_rcu and try again.
136 if (atomic_inc_not_zero(&eb->refs)) {
137 rcu_read_unlock();
138 break;
140 rcu_read_unlock();
141 synchronize_rcu();
143 return eb;
146 /* loop around taking references on and locking the root node of the
147 * tree until you end up with a lock on the root. A locked buffer
148 * is returned, with a reference held.
150 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
152 struct extent_buffer *eb;
154 while (1) {
155 eb = btrfs_root_node(root);
156 btrfs_tree_lock(eb);
157 if (eb == root->node)
158 break;
159 btrfs_tree_unlock(eb);
160 free_extent_buffer(eb);
162 return eb;
165 /* loop around taking references on and locking the root node of the
166 * tree until you end up with a lock on the root. A locked buffer
167 * is returned, with a reference held.
169 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
171 struct extent_buffer *eb;
173 while (1) {
174 eb = btrfs_root_node(root);
175 btrfs_tree_read_lock(eb);
176 if (eb == root->node)
177 break;
178 btrfs_tree_read_unlock(eb);
179 free_extent_buffer(eb);
181 return eb;
184 /* cowonly root (everything not a reference counted cow subvolume), just get
185 * put onto a simple dirty list. transaction.c walks this to make sure they
186 * get properly updated on disk.
188 static void add_root_to_dirty_list(struct btrfs_root *root)
190 struct btrfs_fs_info *fs_info = root->fs_info;
192 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
193 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
194 return;
196 spin_lock(&fs_info->trans_lock);
197 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
198 /* Want the extent tree to be the last on the list */
199 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
200 list_move_tail(&root->dirty_list,
201 &fs_info->dirty_cowonly_roots);
202 else
203 list_move(&root->dirty_list,
204 &fs_info->dirty_cowonly_roots);
206 spin_unlock(&fs_info->trans_lock);
210 * used by snapshot creation to make a copy of a root for a tree with
211 * a given objectid. The buffer with the new root node is returned in
212 * cow_ret, and this func returns zero on success or a negative error code.
214 int btrfs_copy_root(struct btrfs_trans_handle *trans,
215 struct btrfs_root *root,
216 struct extent_buffer *buf,
217 struct extent_buffer **cow_ret, u64 new_root_objectid)
219 struct btrfs_fs_info *fs_info = root->fs_info;
220 struct extent_buffer *cow;
221 int ret = 0;
222 int level;
223 struct btrfs_disk_key disk_key;
225 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
226 trans->transid != fs_info->running_transaction->transid);
227 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
228 trans->transid != root->last_trans);
230 level = btrfs_header_level(buf);
231 if (level == 0)
232 btrfs_item_key(buf, &disk_key, 0);
233 else
234 btrfs_node_key(buf, &disk_key, 0);
236 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
237 &disk_key, level, buf->start, 0);
238 if (IS_ERR(cow))
239 return PTR_ERR(cow);
241 copy_extent_buffer_full(cow, buf);
242 btrfs_set_header_bytenr(cow, cow->start);
243 btrfs_set_header_generation(cow, trans->transid);
244 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
245 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
246 BTRFS_HEADER_FLAG_RELOC);
247 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
248 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
249 else
250 btrfs_set_header_owner(cow, new_root_objectid);
252 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
254 WARN_ON(btrfs_header_generation(buf) > trans->transid);
255 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
256 ret = btrfs_inc_ref(trans, root, cow, 1);
257 else
258 ret = btrfs_inc_ref(trans, root, cow, 0);
260 if (ret)
261 return ret;
263 btrfs_mark_buffer_dirty(cow);
264 *cow_ret = cow;
265 return 0;
268 enum mod_log_op {
269 MOD_LOG_KEY_REPLACE,
270 MOD_LOG_KEY_ADD,
271 MOD_LOG_KEY_REMOVE,
272 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
273 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
274 MOD_LOG_MOVE_KEYS,
275 MOD_LOG_ROOT_REPLACE,
278 struct tree_mod_root {
279 u64 logical;
280 u8 level;
283 struct tree_mod_elem {
284 struct rb_node node;
285 u64 logical;
286 u64 seq;
287 enum mod_log_op op;
289 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
290 int slot;
292 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
293 u64 generation;
295 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
296 struct btrfs_disk_key key;
297 u64 blockptr;
299 /* this is used for op == MOD_LOG_MOVE_KEYS */
300 struct {
301 int dst_slot;
302 int nr_items;
303 } move;
305 /* this is used for op == MOD_LOG_ROOT_REPLACE */
306 struct tree_mod_root old_root;
310 * Pull a new tree mod seq number for our operation.
312 static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info)
314 return atomic64_inc_return(&fs_info->tree_mod_seq);
318 * This adds a new blocker to the tree mod log's blocker list if the @elem
319 * passed does not already have a sequence number set. So when a caller expects
320 * to record tree modifications, it should ensure to set elem->seq to zero
321 * before calling btrfs_get_tree_mod_seq.
322 * Returns a fresh, unused tree log modification sequence number, even if no new
323 * blocker was added.
325 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
326 struct seq_list *elem)
328 write_lock(&fs_info->tree_mod_log_lock);
329 if (!elem->seq) {
330 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
331 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
333 write_unlock(&fs_info->tree_mod_log_lock);
335 return elem->seq;
338 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
339 struct seq_list *elem)
341 struct rb_root *tm_root;
342 struct rb_node *node;
343 struct rb_node *next;
344 struct seq_list *cur_elem;
345 struct tree_mod_elem *tm;
346 u64 min_seq = (u64)-1;
347 u64 seq_putting = elem->seq;
349 if (!seq_putting)
350 return;
352 write_lock(&fs_info->tree_mod_log_lock);
353 list_del(&elem->list);
354 elem->seq = 0;
356 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
357 if (cur_elem->seq < min_seq) {
358 if (seq_putting > cur_elem->seq) {
360 * blocker with lower sequence number exists, we
361 * cannot remove anything from the log
363 write_unlock(&fs_info->tree_mod_log_lock);
364 return;
366 min_seq = cur_elem->seq;
371 * anything that's lower than the lowest existing (read: blocked)
372 * sequence number can be removed from the tree.
374 tm_root = &fs_info->tree_mod_log;
375 for (node = rb_first(tm_root); node; node = next) {
376 next = rb_next(node);
377 tm = rb_entry(node, struct tree_mod_elem, node);
378 if (tm->seq >= min_seq)
379 continue;
380 rb_erase(node, tm_root);
381 kfree(tm);
383 write_unlock(&fs_info->tree_mod_log_lock);
387 * key order of the log:
388 * node/leaf start address -> sequence
390 * The 'start address' is the logical address of the *new* root node
391 * for root replace operations, or the logical address of the affected
392 * block for all other operations.
394 static noinline int
395 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
397 struct rb_root *tm_root;
398 struct rb_node **new;
399 struct rb_node *parent = NULL;
400 struct tree_mod_elem *cur;
402 lockdep_assert_held_write(&fs_info->tree_mod_log_lock);
404 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
406 tm_root = &fs_info->tree_mod_log;
407 new = &tm_root->rb_node;
408 while (*new) {
409 cur = rb_entry(*new, struct tree_mod_elem, node);
410 parent = *new;
411 if (cur->logical < tm->logical)
412 new = &((*new)->rb_left);
413 else if (cur->logical > tm->logical)
414 new = &((*new)->rb_right);
415 else if (cur->seq < tm->seq)
416 new = &((*new)->rb_left);
417 else if (cur->seq > tm->seq)
418 new = &((*new)->rb_right);
419 else
420 return -EEXIST;
423 rb_link_node(&tm->node, parent, new);
424 rb_insert_color(&tm->node, tm_root);
425 return 0;
429 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
430 * returns zero with the tree_mod_log_lock acquired. The caller must hold
431 * this until all tree mod log insertions are recorded in the rb tree and then
432 * write unlock fs_info::tree_mod_log_lock.
434 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
435 struct extent_buffer *eb) {
436 smp_mb();
437 if (list_empty(&(fs_info)->tree_mod_seq_list))
438 return 1;
439 if (eb && btrfs_header_level(eb) == 0)
440 return 1;
442 write_lock(&fs_info->tree_mod_log_lock);
443 if (list_empty(&(fs_info)->tree_mod_seq_list)) {
444 write_unlock(&fs_info->tree_mod_log_lock);
445 return 1;
448 return 0;
451 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
452 static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info,
453 struct extent_buffer *eb)
455 smp_mb();
456 if (list_empty(&(fs_info)->tree_mod_seq_list))
457 return 0;
458 if (eb && btrfs_header_level(eb) == 0)
459 return 0;
461 return 1;
464 static struct tree_mod_elem *
465 alloc_tree_mod_elem(struct extent_buffer *eb, int slot,
466 enum mod_log_op op, gfp_t flags)
468 struct tree_mod_elem *tm;
470 tm = kzalloc(sizeof(*tm), flags);
471 if (!tm)
472 return NULL;
474 tm->logical = eb->start;
475 if (op != MOD_LOG_KEY_ADD) {
476 btrfs_node_key(eb, &tm->key, slot);
477 tm->blockptr = btrfs_node_blockptr(eb, slot);
479 tm->op = op;
480 tm->slot = slot;
481 tm->generation = btrfs_node_ptr_generation(eb, slot);
482 RB_CLEAR_NODE(&tm->node);
484 return tm;
487 static noinline int tree_mod_log_insert_key(struct extent_buffer *eb, int slot,
488 enum mod_log_op op, gfp_t flags)
490 struct tree_mod_elem *tm;
491 int ret;
493 if (!tree_mod_need_log(eb->fs_info, eb))
494 return 0;
496 tm = alloc_tree_mod_elem(eb, slot, op, flags);
497 if (!tm)
498 return -ENOMEM;
500 if (tree_mod_dont_log(eb->fs_info, eb)) {
501 kfree(tm);
502 return 0;
505 ret = __tree_mod_log_insert(eb->fs_info, tm);
506 write_unlock(&eb->fs_info->tree_mod_log_lock);
507 if (ret)
508 kfree(tm);
510 return ret;
513 static noinline int tree_mod_log_insert_move(struct extent_buffer *eb,
514 int dst_slot, int src_slot, int nr_items)
516 struct tree_mod_elem *tm = NULL;
517 struct tree_mod_elem **tm_list = NULL;
518 int ret = 0;
519 int i;
520 int locked = 0;
522 if (!tree_mod_need_log(eb->fs_info, eb))
523 return 0;
525 tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS);
526 if (!tm_list)
527 return -ENOMEM;
529 tm = kzalloc(sizeof(*tm), GFP_NOFS);
530 if (!tm) {
531 ret = -ENOMEM;
532 goto free_tms;
535 tm->logical = eb->start;
536 tm->slot = src_slot;
537 tm->move.dst_slot = dst_slot;
538 tm->move.nr_items = nr_items;
539 tm->op = MOD_LOG_MOVE_KEYS;
541 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
542 tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot,
543 MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS);
544 if (!tm_list[i]) {
545 ret = -ENOMEM;
546 goto free_tms;
550 if (tree_mod_dont_log(eb->fs_info, eb))
551 goto free_tms;
552 locked = 1;
555 * When we override something during the move, we log these removals.
556 * This can only happen when we move towards the beginning of the
557 * buffer, i.e. dst_slot < src_slot.
559 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
560 ret = __tree_mod_log_insert(eb->fs_info, tm_list[i]);
561 if (ret)
562 goto free_tms;
565 ret = __tree_mod_log_insert(eb->fs_info, tm);
566 if (ret)
567 goto free_tms;
568 write_unlock(&eb->fs_info->tree_mod_log_lock);
569 kfree(tm_list);
571 return 0;
572 free_tms:
573 for (i = 0; i < nr_items; i++) {
574 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
575 rb_erase(&tm_list[i]->node, &eb->fs_info->tree_mod_log);
576 kfree(tm_list[i]);
578 if (locked)
579 write_unlock(&eb->fs_info->tree_mod_log_lock);
580 kfree(tm_list);
581 kfree(tm);
583 return ret;
586 static inline int
587 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
588 struct tree_mod_elem **tm_list,
589 int nritems)
591 int i, j;
592 int ret;
594 for (i = nritems - 1; i >= 0; i--) {
595 ret = __tree_mod_log_insert(fs_info, tm_list[i]);
596 if (ret) {
597 for (j = nritems - 1; j > i; j--)
598 rb_erase(&tm_list[j]->node,
599 &fs_info->tree_mod_log);
600 return ret;
604 return 0;
607 static noinline int tree_mod_log_insert_root(struct extent_buffer *old_root,
608 struct extent_buffer *new_root, int log_removal)
610 struct btrfs_fs_info *fs_info = old_root->fs_info;
611 struct tree_mod_elem *tm = NULL;
612 struct tree_mod_elem **tm_list = NULL;
613 int nritems = 0;
614 int ret = 0;
615 int i;
617 if (!tree_mod_need_log(fs_info, NULL))
618 return 0;
620 if (log_removal && btrfs_header_level(old_root) > 0) {
621 nritems = btrfs_header_nritems(old_root);
622 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *),
623 GFP_NOFS);
624 if (!tm_list) {
625 ret = -ENOMEM;
626 goto free_tms;
628 for (i = 0; i < nritems; i++) {
629 tm_list[i] = alloc_tree_mod_elem(old_root, i,
630 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
631 if (!tm_list[i]) {
632 ret = -ENOMEM;
633 goto free_tms;
638 tm = kzalloc(sizeof(*tm), GFP_NOFS);
639 if (!tm) {
640 ret = -ENOMEM;
641 goto free_tms;
644 tm->logical = new_root->start;
645 tm->old_root.logical = old_root->start;
646 tm->old_root.level = btrfs_header_level(old_root);
647 tm->generation = btrfs_header_generation(old_root);
648 tm->op = MOD_LOG_ROOT_REPLACE;
650 if (tree_mod_dont_log(fs_info, NULL))
651 goto free_tms;
653 if (tm_list)
654 ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems);
655 if (!ret)
656 ret = __tree_mod_log_insert(fs_info, tm);
658 write_unlock(&fs_info->tree_mod_log_lock);
659 if (ret)
660 goto free_tms;
661 kfree(tm_list);
663 return ret;
665 free_tms:
666 if (tm_list) {
667 for (i = 0; i < nritems; i++)
668 kfree(tm_list[i]);
669 kfree(tm_list);
671 kfree(tm);
673 return ret;
676 static struct tree_mod_elem *
677 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
678 int smallest)
680 struct rb_root *tm_root;
681 struct rb_node *node;
682 struct tree_mod_elem *cur = NULL;
683 struct tree_mod_elem *found = NULL;
685 read_lock(&fs_info->tree_mod_log_lock);
686 tm_root = &fs_info->tree_mod_log;
687 node = tm_root->rb_node;
688 while (node) {
689 cur = rb_entry(node, struct tree_mod_elem, node);
690 if (cur->logical < start) {
691 node = node->rb_left;
692 } else if (cur->logical > start) {
693 node = node->rb_right;
694 } else if (cur->seq < min_seq) {
695 node = node->rb_left;
696 } else if (!smallest) {
697 /* we want the node with the highest seq */
698 if (found)
699 BUG_ON(found->seq > cur->seq);
700 found = cur;
701 node = node->rb_left;
702 } else if (cur->seq > min_seq) {
703 /* we want the node with the smallest seq */
704 if (found)
705 BUG_ON(found->seq < cur->seq);
706 found = cur;
707 node = node->rb_right;
708 } else {
709 found = cur;
710 break;
713 read_unlock(&fs_info->tree_mod_log_lock);
715 return found;
719 * this returns the element from the log with the smallest time sequence
720 * value that's in the log (the oldest log item). any element with a time
721 * sequence lower than min_seq will be ignored.
723 static struct tree_mod_elem *
724 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
725 u64 min_seq)
727 return __tree_mod_log_search(fs_info, start, min_seq, 1);
731 * this returns the element from the log with the largest time sequence
732 * value that's in the log (the most recent log item). any element with
733 * a time sequence lower than min_seq will be ignored.
735 static struct tree_mod_elem *
736 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
738 return __tree_mod_log_search(fs_info, start, min_seq, 0);
741 static noinline int tree_mod_log_eb_copy(struct extent_buffer *dst,
742 struct extent_buffer *src, unsigned long dst_offset,
743 unsigned long src_offset, int nr_items)
745 struct btrfs_fs_info *fs_info = dst->fs_info;
746 int ret = 0;
747 struct tree_mod_elem **tm_list = NULL;
748 struct tree_mod_elem **tm_list_add, **tm_list_rem;
749 int i;
750 int locked = 0;
752 if (!tree_mod_need_log(fs_info, NULL))
753 return 0;
755 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0)
756 return 0;
758 tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *),
759 GFP_NOFS);
760 if (!tm_list)
761 return -ENOMEM;
763 tm_list_add = tm_list;
764 tm_list_rem = tm_list + nr_items;
765 for (i = 0; i < nr_items; i++) {
766 tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset,
767 MOD_LOG_KEY_REMOVE, GFP_NOFS);
768 if (!tm_list_rem[i]) {
769 ret = -ENOMEM;
770 goto free_tms;
773 tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset,
774 MOD_LOG_KEY_ADD, GFP_NOFS);
775 if (!tm_list_add[i]) {
776 ret = -ENOMEM;
777 goto free_tms;
781 if (tree_mod_dont_log(fs_info, NULL))
782 goto free_tms;
783 locked = 1;
785 for (i = 0; i < nr_items; i++) {
786 ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]);
787 if (ret)
788 goto free_tms;
789 ret = __tree_mod_log_insert(fs_info, tm_list_add[i]);
790 if (ret)
791 goto free_tms;
794 write_unlock(&fs_info->tree_mod_log_lock);
795 kfree(tm_list);
797 return 0;
799 free_tms:
800 for (i = 0; i < nr_items * 2; i++) {
801 if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node))
802 rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log);
803 kfree(tm_list[i]);
805 if (locked)
806 write_unlock(&fs_info->tree_mod_log_lock);
807 kfree(tm_list);
809 return ret;
812 static noinline int tree_mod_log_free_eb(struct extent_buffer *eb)
814 struct tree_mod_elem **tm_list = NULL;
815 int nritems = 0;
816 int i;
817 int ret = 0;
819 if (btrfs_header_level(eb) == 0)
820 return 0;
822 if (!tree_mod_need_log(eb->fs_info, NULL))
823 return 0;
825 nritems = btrfs_header_nritems(eb);
826 tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS);
827 if (!tm_list)
828 return -ENOMEM;
830 for (i = 0; i < nritems; i++) {
831 tm_list[i] = alloc_tree_mod_elem(eb, i,
832 MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS);
833 if (!tm_list[i]) {
834 ret = -ENOMEM;
835 goto free_tms;
839 if (tree_mod_dont_log(eb->fs_info, eb))
840 goto free_tms;
842 ret = __tree_mod_log_free_eb(eb->fs_info, tm_list, nritems);
843 write_unlock(&eb->fs_info->tree_mod_log_lock);
844 if (ret)
845 goto free_tms;
846 kfree(tm_list);
848 return 0;
850 free_tms:
851 for (i = 0; i < nritems; i++)
852 kfree(tm_list[i]);
853 kfree(tm_list);
855 return ret;
859 * check if the tree block can be shared by multiple trees
861 int btrfs_block_can_be_shared(struct btrfs_root *root,
862 struct extent_buffer *buf)
865 * Tree blocks not in reference counted trees and tree roots
866 * are never shared. If a block was allocated after the last
867 * snapshot and the block was not allocated by tree relocation,
868 * we know the block is not shared.
870 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
871 buf != root->node && buf != root->commit_root &&
872 (btrfs_header_generation(buf) <=
873 btrfs_root_last_snapshot(&root->root_item) ||
874 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
875 return 1;
877 return 0;
880 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
881 struct btrfs_root *root,
882 struct extent_buffer *buf,
883 struct extent_buffer *cow,
884 int *last_ref)
886 struct btrfs_fs_info *fs_info = root->fs_info;
887 u64 refs;
888 u64 owner;
889 u64 flags;
890 u64 new_flags = 0;
891 int ret;
894 * Backrefs update rules:
896 * Always use full backrefs for extent pointers in tree block
897 * allocated by tree relocation.
899 * If a shared tree block is no longer referenced by its owner
900 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
901 * use full backrefs for extent pointers in tree block.
903 * If a tree block is been relocating
904 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
905 * use full backrefs for extent pointers in tree block.
906 * The reason for this is some operations (such as drop tree)
907 * are only allowed for blocks use full backrefs.
910 if (btrfs_block_can_be_shared(root, buf)) {
911 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
912 btrfs_header_level(buf), 1,
913 &refs, &flags);
914 if (ret)
915 return ret;
916 if (refs == 0) {
917 ret = -EROFS;
918 btrfs_handle_fs_error(fs_info, ret, NULL);
919 return ret;
921 } else {
922 refs = 1;
923 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
924 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
925 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
926 else
927 flags = 0;
930 owner = btrfs_header_owner(buf);
931 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
932 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
934 if (refs > 1) {
935 if ((owner == root->root_key.objectid ||
936 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
937 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
938 ret = btrfs_inc_ref(trans, root, buf, 1);
939 if (ret)
940 return ret;
942 if (root->root_key.objectid ==
943 BTRFS_TREE_RELOC_OBJECTID) {
944 ret = btrfs_dec_ref(trans, root, buf, 0);
945 if (ret)
946 return ret;
947 ret = btrfs_inc_ref(trans, root, cow, 1);
948 if (ret)
949 return ret;
951 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
952 } else {
954 if (root->root_key.objectid ==
955 BTRFS_TREE_RELOC_OBJECTID)
956 ret = btrfs_inc_ref(trans, root, cow, 1);
957 else
958 ret = btrfs_inc_ref(trans, root, cow, 0);
959 if (ret)
960 return ret;
962 if (new_flags != 0) {
963 int level = btrfs_header_level(buf);
965 ret = btrfs_set_disk_extent_flags(trans,
966 buf->start,
967 buf->len,
968 new_flags, level, 0);
969 if (ret)
970 return ret;
972 } else {
973 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
974 if (root->root_key.objectid ==
975 BTRFS_TREE_RELOC_OBJECTID)
976 ret = btrfs_inc_ref(trans, root, cow, 1);
977 else
978 ret = btrfs_inc_ref(trans, root, cow, 0);
979 if (ret)
980 return ret;
981 ret = btrfs_dec_ref(trans, root, buf, 1);
982 if (ret)
983 return ret;
985 btrfs_clean_tree_block(buf);
986 *last_ref = 1;
988 return 0;
991 static struct extent_buffer *alloc_tree_block_no_bg_flush(
992 struct btrfs_trans_handle *trans,
993 struct btrfs_root *root,
994 u64 parent_start,
995 const struct btrfs_disk_key *disk_key,
996 int level,
997 u64 hint,
998 u64 empty_size)
1000 struct btrfs_fs_info *fs_info = root->fs_info;
1001 struct extent_buffer *ret;
1004 * If we are COWing a node/leaf from the extent, chunk, device or free
1005 * space trees, make sure that we do not finish block group creation of
1006 * pending block groups. We do this to avoid a deadlock.
1007 * COWing can result in allocation of a new chunk, and flushing pending
1008 * block groups (btrfs_create_pending_block_groups()) can be triggered
1009 * when finishing allocation of a new chunk. Creation of a pending block
1010 * group modifies the extent, chunk, device and free space trees,
1011 * therefore we could deadlock with ourselves since we are holding a
1012 * lock on an extent buffer that btrfs_create_pending_block_groups() may
1013 * try to COW later.
1014 * For similar reasons, we also need to delay flushing pending block
1015 * groups when splitting a leaf or node, from one of those trees, since
1016 * we are holding a write lock on it and its parent or when inserting a
1017 * new root node for one of those trees.
1019 if (root == fs_info->extent_root ||
1020 root == fs_info->chunk_root ||
1021 root == fs_info->dev_root ||
1022 root == fs_info->free_space_root)
1023 trans->can_flush_pending_bgs = false;
1025 ret = btrfs_alloc_tree_block(trans, root, parent_start,
1026 root->root_key.objectid, disk_key, level,
1027 hint, empty_size);
1028 trans->can_flush_pending_bgs = true;
1030 return ret;
1034 * does the dirty work in cow of a single block. The parent block (if
1035 * supplied) is updated to point to the new cow copy. The new buffer is marked
1036 * dirty and returned locked. If you modify the block it needs to be marked
1037 * dirty again.
1039 * search_start -- an allocation hint for the new block
1041 * empty_size -- a hint that you plan on doing more cow. This is the size in
1042 * bytes the allocator should try to find free next to the block it returns.
1043 * This is just a hint and may be ignored by the allocator.
1045 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
1046 struct btrfs_root *root,
1047 struct extent_buffer *buf,
1048 struct extent_buffer *parent, int parent_slot,
1049 struct extent_buffer **cow_ret,
1050 u64 search_start, u64 empty_size)
1052 struct btrfs_fs_info *fs_info = root->fs_info;
1053 struct btrfs_disk_key disk_key;
1054 struct extent_buffer *cow;
1055 int level, ret;
1056 int last_ref = 0;
1057 int unlock_orig = 0;
1058 u64 parent_start = 0;
1060 if (*cow_ret == buf)
1061 unlock_orig = 1;
1063 btrfs_assert_tree_locked(buf);
1065 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1066 trans->transid != fs_info->running_transaction->transid);
1067 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
1068 trans->transid != root->last_trans);
1070 level = btrfs_header_level(buf);
1072 if (level == 0)
1073 btrfs_item_key(buf, &disk_key, 0);
1074 else
1075 btrfs_node_key(buf, &disk_key, 0);
1077 if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
1078 parent_start = parent->start;
1080 cow = alloc_tree_block_no_bg_flush(trans, root, parent_start, &disk_key,
1081 level, search_start, empty_size);
1082 if (IS_ERR(cow))
1083 return PTR_ERR(cow);
1085 /* cow is set to blocking by btrfs_init_new_buffer */
1087 copy_extent_buffer_full(cow, buf);
1088 btrfs_set_header_bytenr(cow, cow->start);
1089 btrfs_set_header_generation(cow, trans->transid);
1090 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1091 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1092 BTRFS_HEADER_FLAG_RELOC);
1093 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1094 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1095 else
1096 btrfs_set_header_owner(cow, root->root_key.objectid);
1098 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
1100 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1101 if (ret) {
1102 btrfs_abort_transaction(trans, ret);
1103 return ret;
1106 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
1107 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
1108 if (ret) {
1109 btrfs_abort_transaction(trans, ret);
1110 return ret;
1114 if (buf == root->node) {
1115 WARN_ON(parent && parent != buf);
1116 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1117 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1118 parent_start = buf->start;
1120 atomic_inc(&cow->refs);
1121 ret = tree_mod_log_insert_root(root->node, cow, 1);
1122 BUG_ON(ret < 0);
1123 rcu_assign_pointer(root->node, cow);
1125 btrfs_free_tree_block(trans, root, buf, parent_start,
1126 last_ref);
1127 free_extent_buffer(buf);
1128 add_root_to_dirty_list(root);
1129 } else {
1130 WARN_ON(trans->transid != btrfs_header_generation(parent));
1131 tree_mod_log_insert_key(parent, parent_slot,
1132 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1133 btrfs_set_node_blockptr(parent, parent_slot,
1134 cow->start);
1135 btrfs_set_node_ptr_generation(parent, parent_slot,
1136 trans->transid);
1137 btrfs_mark_buffer_dirty(parent);
1138 if (last_ref) {
1139 ret = tree_mod_log_free_eb(buf);
1140 if (ret) {
1141 btrfs_abort_transaction(trans, ret);
1142 return ret;
1145 btrfs_free_tree_block(trans, root, buf, parent_start,
1146 last_ref);
1148 if (unlock_orig)
1149 btrfs_tree_unlock(buf);
1150 free_extent_buffer_stale(buf);
1151 btrfs_mark_buffer_dirty(cow);
1152 *cow_ret = cow;
1153 return 0;
1157 * returns the logical address of the oldest predecessor of the given root.
1158 * entries older than time_seq are ignored.
1160 static struct tree_mod_elem *__tree_mod_log_oldest_root(
1161 struct extent_buffer *eb_root, u64 time_seq)
1163 struct tree_mod_elem *tm;
1164 struct tree_mod_elem *found = NULL;
1165 u64 root_logical = eb_root->start;
1166 int looped = 0;
1168 if (!time_seq)
1169 return NULL;
1172 * the very last operation that's logged for a root is the
1173 * replacement operation (if it is replaced at all). this has
1174 * the logical address of the *new* root, making it the very
1175 * first operation that's logged for this root.
1177 while (1) {
1178 tm = tree_mod_log_search_oldest(eb_root->fs_info, root_logical,
1179 time_seq);
1180 if (!looped && !tm)
1181 return NULL;
1183 * if there are no tree operation for the oldest root, we simply
1184 * return it. this should only happen if that (old) root is at
1185 * level 0.
1187 if (!tm)
1188 break;
1191 * if there's an operation that's not a root replacement, we
1192 * found the oldest version of our root. normally, we'll find a
1193 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1195 if (tm->op != MOD_LOG_ROOT_REPLACE)
1196 break;
1198 found = tm;
1199 root_logical = tm->old_root.logical;
1200 looped = 1;
1203 /* if there's no old root to return, return what we found instead */
1204 if (!found)
1205 found = tm;
1207 return found;
1211 * tm is a pointer to the first operation to rewind within eb. then, all
1212 * previous operations will be rewound (until we reach something older than
1213 * time_seq).
1215 static void
1216 __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1217 u64 time_seq, struct tree_mod_elem *first_tm)
1219 u32 n;
1220 struct rb_node *next;
1221 struct tree_mod_elem *tm = first_tm;
1222 unsigned long o_dst;
1223 unsigned long o_src;
1224 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1226 n = btrfs_header_nritems(eb);
1227 read_lock(&fs_info->tree_mod_log_lock);
1228 while (tm && tm->seq >= time_seq) {
1230 * all the operations are recorded with the operator used for
1231 * the modification. as we're going backwards, we do the
1232 * opposite of each operation here.
1234 switch (tm->op) {
1235 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1236 BUG_ON(tm->slot < n);
1237 /* Fallthrough */
1238 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1239 case MOD_LOG_KEY_REMOVE:
1240 btrfs_set_node_key(eb, &tm->key, tm->slot);
1241 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1242 btrfs_set_node_ptr_generation(eb, tm->slot,
1243 tm->generation);
1244 n++;
1245 break;
1246 case MOD_LOG_KEY_REPLACE:
1247 BUG_ON(tm->slot >= n);
1248 btrfs_set_node_key(eb, &tm->key, tm->slot);
1249 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1250 btrfs_set_node_ptr_generation(eb, tm->slot,
1251 tm->generation);
1252 break;
1253 case MOD_LOG_KEY_ADD:
1254 /* if a move operation is needed it's in the log */
1255 n--;
1256 break;
1257 case MOD_LOG_MOVE_KEYS:
1258 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1259 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1260 memmove_extent_buffer(eb, o_dst, o_src,
1261 tm->move.nr_items * p_size);
1262 break;
1263 case MOD_LOG_ROOT_REPLACE:
1265 * this operation is special. for roots, this must be
1266 * handled explicitly before rewinding.
1267 * for non-roots, this operation may exist if the node
1268 * was a root: root A -> child B; then A gets empty and
1269 * B is promoted to the new root. in the mod log, we'll
1270 * have a root-replace operation for B, a tree block
1271 * that is no root. we simply ignore that operation.
1273 break;
1275 next = rb_next(&tm->node);
1276 if (!next)
1277 break;
1278 tm = rb_entry(next, struct tree_mod_elem, node);
1279 if (tm->logical != first_tm->logical)
1280 break;
1282 read_unlock(&fs_info->tree_mod_log_lock);
1283 btrfs_set_header_nritems(eb, n);
1287 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1288 * is returned. If rewind operations happen, a fresh buffer is returned. The
1289 * returned buffer is always read-locked. If the returned buffer is not the
1290 * input buffer, the lock on the input buffer is released and the input buffer
1291 * is freed (its refcount is decremented).
1293 static struct extent_buffer *
1294 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
1295 struct extent_buffer *eb, u64 time_seq)
1297 struct extent_buffer *eb_rewin;
1298 struct tree_mod_elem *tm;
1300 if (!time_seq)
1301 return eb;
1303 if (btrfs_header_level(eb) == 0)
1304 return eb;
1306 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1307 if (!tm)
1308 return eb;
1310 btrfs_set_path_blocking(path);
1311 btrfs_set_lock_blocking_read(eb);
1313 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1314 BUG_ON(tm->slot != 0);
1315 eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start);
1316 if (!eb_rewin) {
1317 btrfs_tree_read_unlock_blocking(eb);
1318 free_extent_buffer(eb);
1319 return NULL;
1321 btrfs_set_header_bytenr(eb_rewin, eb->start);
1322 btrfs_set_header_backref_rev(eb_rewin,
1323 btrfs_header_backref_rev(eb));
1324 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1325 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1326 } else {
1327 eb_rewin = btrfs_clone_extent_buffer(eb);
1328 if (!eb_rewin) {
1329 btrfs_tree_read_unlock_blocking(eb);
1330 free_extent_buffer(eb);
1331 return NULL;
1335 btrfs_tree_read_unlock_blocking(eb);
1336 free_extent_buffer(eb);
1338 btrfs_tree_read_lock(eb_rewin);
1339 __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm);
1340 WARN_ON(btrfs_header_nritems(eb_rewin) >
1341 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1343 return eb_rewin;
1347 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1348 * value. If there are no changes, the current root->root_node is returned. If
1349 * anything changed in between, there's a fresh buffer allocated on which the
1350 * rewind operations are done. In any case, the returned buffer is read locked.
1351 * Returns NULL on error (with no locks held).
1353 static inline struct extent_buffer *
1354 get_old_root(struct btrfs_root *root, u64 time_seq)
1356 struct btrfs_fs_info *fs_info = root->fs_info;
1357 struct tree_mod_elem *tm;
1358 struct extent_buffer *eb = NULL;
1359 struct extent_buffer *eb_root;
1360 u64 eb_root_owner = 0;
1361 struct extent_buffer *old;
1362 struct tree_mod_root *old_root = NULL;
1363 u64 old_generation = 0;
1364 u64 logical;
1365 int level;
1367 eb_root = btrfs_read_lock_root_node(root);
1368 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1369 if (!tm)
1370 return eb_root;
1372 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1373 old_root = &tm->old_root;
1374 old_generation = tm->generation;
1375 logical = old_root->logical;
1376 level = old_root->level;
1377 } else {
1378 logical = eb_root->start;
1379 level = btrfs_header_level(eb_root);
1382 tm = tree_mod_log_search(fs_info, logical, time_seq);
1383 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1384 btrfs_tree_read_unlock(eb_root);
1385 free_extent_buffer(eb_root);
1386 old = read_tree_block(fs_info, logical, 0, level, NULL);
1387 if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) {
1388 if (!IS_ERR(old))
1389 free_extent_buffer(old);
1390 btrfs_warn(fs_info,
1391 "failed to read tree block %llu from get_old_root",
1392 logical);
1393 } else {
1394 eb = btrfs_clone_extent_buffer(old);
1395 free_extent_buffer(old);
1397 } else if (old_root) {
1398 eb_root_owner = btrfs_header_owner(eb_root);
1399 btrfs_tree_read_unlock(eb_root);
1400 free_extent_buffer(eb_root);
1401 eb = alloc_dummy_extent_buffer(fs_info, logical);
1402 } else {
1403 btrfs_set_lock_blocking_read(eb_root);
1404 eb = btrfs_clone_extent_buffer(eb_root);
1405 btrfs_tree_read_unlock_blocking(eb_root);
1406 free_extent_buffer(eb_root);
1409 if (!eb)
1410 return NULL;
1411 btrfs_tree_read_lock(eb);
1412 if (old_root) {
1413 btrfs_set_header_bytenr(eb, eb->start);
1414 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1415 btrfs_set_header_owner(eb, eb_root_owner);
1416 btrfs_set_header_level(eb, old_root->level);
1417 btrfs_set_header_generation(eb, old_generation);
1419 if (tm)
1420 __tree_mod_log_rewind(fs_info, eb, time_seq, tm);
1421 else
1422 WARN_ON(btrfs_header_level(eb) != 0);
1423 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1425 return eb;
1428 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1430 struct tree_mod_elem *tm;
1431 int level;
1432 struct extent_buffer *eb_root = btrfs_root_node(root);
1434 tm = __tree_mod_log_oldest_root(eb_root, time_seq);
1435 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1436 level = tm->old_root.level;
1437 } else {
1438 level = btrfs_header_level(eb_root);
1440 free_extent_buffer(eb_root);
1442 return level;
1445 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1446 struct btrfs_root *root,
1447 struct extent_buffer *buf)
1449 if (btrfs_is_testing(root->fs_info))
1450 return 0;
1452 /* Ensure we can see the FORCE_COW bit */
1453 smp_mb__before_atomic();
1456 * We do not need to cow a block if
1457 * 1) this block is not created or changed in this transaction;
1458 * 2) this block does not belong to TREE_RELOC tree;
1459 * 3) the root is not forced COW.
1461 * What is forced COW:
1462 * when we create snapshot during committing the transaction,
1463 * after we've finished copying src root, we must COW the shared
1464 * block to ensure the metadata consistency.
1466 if (btrfs_header_generation(buf) == trans->transid &&
1467 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1468 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1469 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1470 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
1471 return 0;
1472 return 1;
1476 * cows a single block, see __btrfs_cow_block for the real work.
1477 * This version of it has extra checks so that a block isn't COWed more than
1478 * once per transaction, as long as it hasn't been written yet
1480 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1481 struct btrfs_root *root, struct extent_buffer *buf,
1482 struct extent_buffer *parent, int parent_slot,
1483 struct extent_buffer **cow_ret)
1485 struct btrfs_fs_info *fs_info = root->fs_info;
1486 u64 search_start;
1487 int ret;
1489 if (test_bit(BTRFS_ROOT_DELETING, &root->state))
1490 btrfs_err(fs_info,
1491 "COW'ing blocks on a fs root that's being dropped");
1493 if (trans->transaction != fs_info->running_transaction)
1494 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1495 trans->transid,
1496 fs_info->running_transaction->transid);
1498 if (trans->transid != fs_info->generation)
1499 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1500 trans->transid, fs_info->generation);
1502 if (!should_cow_block(trans, root, buf)) {
1503 trans->dirty = true;
1504 *cow_ret = buf;
1505 return 0;
1508 search_start = buf->start & ~((u64)SZ_1G - 1);
1510 if (parent)
1511 btrfs_set_lock_blocking_write(parent);
1512 btrfs_set_lock_blocking_write(buf);
1515 * Before CoWing this block for later modification, check if it's
1516 * the subtree root and do the delayed subtree trace if needed.
1518 * Also We don't care about the error, as it's handled internally.
1520 btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
1521 ret = __btrfs_cow_block(trans, root, buf, parent,
1522 parent_slot, cow_ret, search_start, 0);
1524 trace_btrfs_cow_block(root, buf, *cow_ret);
1526 return ret;
1530 * helper function for defrag to decide if two blocks pointed to by a
1531 * node are actually close by
1533 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1535 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1536 return 1;
1537 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1538 return 1;
1539 return 0;
1543 * compare two keys in a memcmp fashion
1545 static int comp_keys(const struct btrfs_disk_key *disk,
1546 const struct btrfs_key *k2)
1548 struct btrfs_key k1;
1550 btrfs_disk_key_to_cpu(&k1, disk);
1552 return btrfs_comp_cpu_keys(&k1, k2);
1556 * same as comp_keys only with two btrfs_key's
1558 int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
1560 if (k1->objectid > k2->objectid)
1561 return 1;
1562 if (k1->objectid < k2->objectid)
1563 return -1;
1564 if (k1->type > k2->type)
1565 return 1;
1566 if (k1->type < k2->type)
1567 return -1;
1568 if (k1->offset > k2->offset)
1569 return 1;
1570 if (k1->offset < k2->offset)
1571 return -1;
1572 return 0;
1576 * this is used by the defrag code to go through all the
1577 * leaves pointed to by a node and reallocate them so that
1578 * disk order is close to key order
1580 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1581 struct btrfs_root *root, struct extent_buffer *parent,
1582 int start_slot, u64 *last_ret,
1583 struct btrfs_key *progress)
1585 struct btrfs_fs_info *fs_info = root->fs_info;
1586 struct extent_buffer *cur;
1587 u64 blocknr;
1588 u64 gen;
1589 u64 search_start = *last_ret;
1590 u64 last_block = 0;
1591 u64 other;
1592 u32 parent_nritems;
1593 int end_slot;
1594 int i;
1595 int err = 0;
1596 int parent_level;
1597 int uptodate;
1598 u32 blocksize;
1599 int progress_passed = 0;
1600 struct btrfs_disk_key disk_key;
1602 parent_level = btrfs_header_level(parent);
1604 WARN_ON(trans->transaction != fs_info->running_transaction);
1605 WARN_ON(trans->transid != fs_info->generation);
1607 parent_nritems = btrfs_header_nritems(parent);
1608 blocksize = fs_info->nodesize;
1609 end_slot = parent_nritems - 1;
1611 if (parent_nritems <= 1)
1612 return 0;
1614 btrfs_set_lock_blocking_write(parent);
1616 for (i = start_slot; i <= end_slot; i++) {
1617 struct btrfs_key first_key;
1618 int close = 1;
1620 btrfs_node_key(parent, &disk_key, i);
1621 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1622 continue;
1624 progress_passed = 1;
1625 blocknr = btrfs_node_blockptr(parent, i);
1626 gen = btrfs_node_ptr_generation(parent, i);
1627 btrfs_node_key_to_cpu(parent, &first_key, i);
1628 if (last_block == 0)
1629 last_block = blocknr;
1631 if (i > 0) {
1632 other = btrfs_node_blockptr(parent, i - 1);
1633 close = close_blocks(blocknr, other, blocksize);
1635 if (!close && i < end_slot) {
1636 other = btrfs_node_blockptr(parent, i + 1);
1637 close = close_blocks(blocknr, other, blocksize);
1639 if (close) {
1640 last_block = blocknr;
1641 continue;
1644 cur = find_extent_buffer(fs_info, blocknr);
1645 if (cur)
1646 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1647 else
1648 uptodate = 0;
1649 if (!cur || !uptodate) {
1650 if (!cur) {
1651 cur = read_tree_block(fs_info, blocknr, gen,
1652 parent_level - 1,
1653 &first_key);
1654 if (IS_ERR(cur)) {
1655 return PTR_ERR(cur);
1656 } else if (!extent_buffer_uptodate(cur)) {
1657 free_extent_buffer(cur);
1658 return -EIO;
1660 } else if (!uptodate) {
1661 err = btrfs_read_buffer(cur, gen,
1662 parent_level - 1,&first_key);
1663 if (err) {
1664 free_extent_buffer(cur);
1665 return err;
1669 if (search_start == 0)
1670 search_start = last_block;
1672 btrfs_tree_lock(cur);
1673 btrfs_set_lock_blocking_write(cur);
1674 err = __btrfs_cow_block(trans, root, cur, parent, i,
1675 &cur, search_start,
1676 min(16 * blocksize,
1677 (end_slot - i) * blocksize));
1678 if (err) {
1679 btrfs_tree_unlock(cur);
1680 free_extent_buffer(cur);
1681 break;
1683 search_start = cur->start;
1684 last_block = cur->start;
1685 *last_ret = search_start;
1686 btrfs_tree_unlock(cur);
1687 free_extent_buffer(cur);
1689 return err;
1693 * search for key in the extent_buffer. The items start at offset p,
1694 * and they are item_size apart. There are 'max' items in p.
1696 * the slot in the array is returned via slot, and it points to
1697 * the place where you would insert key if it is not found in
1698 * the array.
1700 * slot may point to max if the key is bigger than all of the keys
1702 static noinline int generic_bin_search(struct extent_buffer *eb,
1703 unsigned long p, int item_size,
1704 const struct btrfs_key *key,
1705 int max, int *slot)
1707 int low = 0;
1708 int high = max;
1709 int mid;
1710 int ret;
1711 struct btrfs_disk_key *tmp = NULL;
1712 struct btrfs_disk_key unaligned;
1713 unsigned long offset;
1714 char *kaddr = NULL;
1715 unsigned long map_start = 0;
1716 unsigned long map_len = 0;
1717 int err;
1719 if (low > high) {
1720 btrfs_err(eb->fs_info,
1721 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1722 __func__, low, high, eb->start,
1723 btrfs_header_owner(eb), btrfs_header_level(eb));
1724 return -EINVAL;
1727 while (low < high) {
1728 mid = (low + high) / 2;
1729 offset = p + mid * item_size;
1731 if (!kaddr || offset < map_start ||
1732 (offset + sizeof(struct btrfs_disk_key)) >
1733 map_start + map_len) {
1735 err = map_private_extent_buffer(eb, offset,
1736 sizeof(struct btrfs_disk_key),
1737 &kaddr, &map_start, &map_len);
1739 if (!err) {
1740 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1741 map_start);
1742 } else if (err == 1) {
1743 read_extent_buffer(eb, &unaligned,
1744 offset, sizeof(unaligned));
1745 tmp = &unaligned;
1746 } else {
1747 return err;
1750 } else {
1751 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1752 map_start);
1754 ret = comp_keys(tmp, key);
1756 if (ret < 0)
1757 low = mid + 1;
1758 else if (ret > 0)
1759 high = mid;
1760 else {
1761 *slot = mid;
1762 return 0;
1765 *slot = low;
1766 return 1;
1770 * simple bin_search frontend that does the right thing for
1771 * leaves vs nodes
1773 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
1774 int level, int *slot)
1776 if (level == 0)
1777 return generic_bin_search(eb,
1778 offsetof(struct btrfs_leaf, items),
1779 sizeof(struct btrfs_item),
1780 key, btrfs_header_nritems(eb),
1781 slot);
1782 else
1783 return generic_bin_search(eb,
1784 offsetof(struct btrfs_node, ptrs),
1785 sizeof(struct btrfs_key_ptr),
1786 key, btrfs_header_nritems(eb),
1787 slot);
1790 static void root_add_used(struct btrfs_root *root, u32 size)
1792 spin_lock(&root->accounting_lock);
1793 btrfs_set_root_used(&root->root_item,
1794 btrfs_root_used(&root->root_item) + size);
1795 spin_unlock(&root->accounting_lock);
1798 static void root_sub_used(struct btrfs_root *root, u32 size)
1800 spin_lock(&root->accounting_lock);
1801 btrfs_set_root_used(&root->root_item,
1802 btrfs_root_used(&root->root_item) - size);
1803 spin_unlock(&root->accounting_lock);
1806 /* given a node and slot number, this reads the blocks it points to. The
1807 * extent buffer is returned with a reference taken (but unlocked).
1809 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
1810 int slot)
1812 int level = btrfs_header_level(parent);
1813 struct extent_buffer *eb;
1814 struct btrfs_key first_key;
1816 if (slot < 0 || slot >= btrfs_header_nritems(parent))
1817 return ERR_PTR(-ENOENT);
1819 BUG_ON(level == 0);
1821 btrfs_node_key_to_cpu(parent, &first_key, slot);
1822 eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
1823 btrfs_node_ptr_generation(parent, slot),
1824 level - 1, &first_key);
1825 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
1826 free_extent_buffer(eb);
1827 eb = ERR_PTR(-EIO);
1830 return eb;
1834 * node level balancing, used to make sure nodes are in proper order for
1835 * item deletion. We balance from the top down, so we have to make sure
1836 * that a deletion won't leave an node completely empty later on.
1838 static noinline int balance_level(struct btrfs_trans_handle *trans,
1839 struct btrfs_root *root,
1840 struct btrfs_path *path, int level)
1842 struct btrfs_fs_info *fs_info = root->fs_info;
1843 struct extent_buffer *right = NULL;
1844 struct extent_buffer *mid;
1845 struct extent_buffer *left = NULL;
1846 struct extent_buffer *parent = NULL;
1847 int ret = 0;
1848 int wret;
1849 int pslot;
1850 int orig_slot = path->slots[level];
1851 u64 orig_ptr;
1853 ASSERT(level > 0);
1855 mid = path->nodes[level];
1857 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1858 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1859 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1861 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1863 if (level < BTRFS_MAX_LEVEL - 1) {
1864 parent = path->nodes[level + 1];
1865 pslot = path->slots[level + 1];
1869 * deal with the case where there is only one pointer in the root
1870 * by promoting the node below to a root
1872 if (!parent) {
1873 struct extent_buffer *child;
1875 if (btrfs_header_nritems(mid) != 1)
1876 return 0;
1878 /* promote the child to a root */
1879 child = btrfs_read_node_slot(mid, 0);
1880 if (IS_ERR(child)) {
1881 ret = PTR_ERR(child);
1882 btrfs_handle_fs_error(fs_info, ret, NULL);
1883 goto enospc;
1886 btrfs_tree_lock(child);
1887 btrfs_set_lock_blocking_write(child);
1888 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1889 if (ret) {
1890 btrfs_tree_unlock(child);
1891 free_extent_buffer(child);
1892 goto enospc;
1895 ret = tree_mod_log_insert_root(root->node, child, 1);
1896 BUG_ON(ret < 0);
1897 rcu_assign_pointer(root->node, child);
1899 add_root_to_dirty_list(root);
1900 btrfs_tree_unlock(child);
1902 path->locks[level] = 0;
1903 path->nodes[level] = NULL;
1904 btrfs_clean_tree_block(mid);
1905 btrfs_tree_unlock(mid);
1906 /* once for the path */
1907 free_extent_buffer(mid);
1909 root_sub_used(root, mid->len);
1910 btrfs_free_tree_block(trans, root, mid, 0, 1);
1911 /* once for the root ptr */
1912 free_extent_buffer_stale(mid);
1913 return 0;
1915 if (btrfs_header_nritems(mid) >
1916 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
1917 return 0;
1919 left = btrfs_read_node_slot(parent, pslot - 1);
1920 if (IS_ERR(left))
1921 left = NULL;
1923 if (left) {
1924 btrfs_tree_lock(left);
1925 btrfs_set_lock_blocking_write(left);
1926 wret = btrfs_cow_block(trans, root, left,
1927 parent, pslot - 1, &left);
1928 if (wret) {
1929 ret = wret;
1930 goto enospc;
1934 right = btrfs_read_node_slot(parent, pslot + 1);
1935 if (IS_ERR(right))
1936 right = NULL;
1938 if (right) {
1939 btrfs_tree_lock(right);
1940 btrfs_set_lock_blocking_write(right);
1941 wret = btrfs_cow_block(trans, root, right,
1942 parent, pslot + 1, &right);
1943 if (wret) {
1944 ret = wret;
1945 goto enospc;
1949 /* first, try to make some room in the middle buffer */
1950 if (left) {
1951 orig_slot += btrfs_header_nritems(left);
1952 wret = push_node_left(trans, left, mid, 1);
1953 if (wret < 0)
1954 ret = wret;
1958 * then try to empty the right most buffer into the middle
1960 if (right) {
1961 wret = push_node_left(trans, mid, right, 1);
1962 if (wret < 0 && wret != -ENOSPC)
1963 ret = wret;
1964 if (btrfs_header_nritems(right) == 0) {
1965 btrfs_clean_tree_block(right);
1966 btrfs_tree_unlock(right);
1967 del_ptr(root, path, level + 1, pslot + 1);
1968 root_sub_used(root, right->len);
1969 btrfs_free_tree_block(trans, root, right, 0, 1);
1970 free_extent_buffer_stale(right);
1971 right = NULL;
1972 } else {
1973 struct btrfs_disk_key right_key;
1974 btrfs_node_key(right, &right_key, 0);
1975 ret = tree_mod_log_insert_key(parent, pslot + 1,
1976 MOD_LOG_KEY_REPLACE, GFP_NOFS);
1977 BUG_ON(ret < 0);
1978 btrfs_set_node_key(parent, &right_key, pslot + 1);
1979 btrfs_mark_buffer_dirty(parent);
1982 if (btrfs_header_nritems(mid) == 1) {
1984 * we're not allowed to leave a node with one item in the
1985 * tree during a delete. A deletion from lower in the tree
1986 * could try to delete the only pointer in this node.
1987 * So, pull some keys from the left.
1988 * There has to be a left pointer at this point because
1989 * otherwise we would have pulled some pointers from the
1990 * right
1992 if (!left) {
1993 ret = -EROFS;
1994 btrfs_handle_fs_error(fs_info, ret, NULL);
1995 goto enospc;
1997 wret = balance_node_right(trans, mid, left);
1998 if (wret < 0) {
1999 ret = wret;
2000 goto enospc;
2002 if (wret == 1) {
2003 wret = push_node_left(trans, left, mid, 1);
2004 if (wret < 0)
2005 ret = wret;
2007 BUG_ON(wret == 1);
2009 if (btrfs_header_nritems(mid) == 0) {
2010 btrfs_clean_tree_block(mid);
2011 btrfs_tree_unlock(mid);
2012 del_ptr(root, path, level + 1, pslot);
2013 root_sub_used(root, mid->len);
2014 btrfs_free_tree_block(trans, root, mid, 0, 1);
2015 free_extent_buffer_stale(mid);
2016 mid = NULL;
2017 } else {
2018 /* update the parent key to reflect our changes */
2019 struct btrfs_disk_key mid_key;
2020 btrfs_node_key(mid, &mid_key, 0);
2021 ret = tree_mod_log_insert_key(parent, pslot,
2022 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2023 BUG_ON(ret < 0);
2024 btrfs_set_node_key(parent, &mid_key, pslot);
2025 btrfs_mark_buffer_dirty(parent);
2028 /* update the path */
2029 if (left) {
2030 if (btrfs_header_nritems(left) > orig_slot) {
2031 atomic_inc(&left->refs);
2032 /* left was locked after cow */
2033 path->nodes[level] = left;
2034 path->slots[level + 1] -= 1;
2035 path->slots[level] = orig_slot;
2036 if (mid) {
2037 btrfs_tree_unlock(mid);
2038 free_extent_buffer(mid);
2040 } else {
2041 orig_slot -= btrfs_header_nritems(left);
2042 path->slots[level] = orig_slot;
2045 /* double check we haven't messed things up */
2046 if (orig_ptr !=
2047 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
2048 BUG();
2049 enospc:
2050 if (right) {
2051 btrfs_tree_unlock(right);
2052 free_extent_buffer(right);
2054 if (left) {
2055 if (path->nodes[level] != left)
2056 btrfs_tree_unlock(left);
2057 free_extent_buffer(left);
2059 return ret;
2062 /* Node balancing for insertion. Here we only split or push nodes around
2063 * when they are completely full. This is also done top down, so we
2064 * have to be pessimistic.
2066 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
2067 struct btrfs_root *root,
2068 struct btrfs_path *path, int level)
2070 struct btrfs_fs_info *fs_info = root->fs_info;
2071 struct extent_buffer *right = NULL;
2072 struct extent_buffer *mid;
2073 struct extent_buffer *left = NULL;
2074 struct extent_buffer *parent = NULL;
2075 int ret = 0;
2076 int wret;
2077 int pslot;
2078 int orig_slot = path->slots[level];
2080 if (level == 0)
2081 return 1;
2083 mid = path->nodes[level];
2084 WARN_ON(btrfs_header_generation(mid) != trans->transid);
2086 if (level < BTRFS_MAX_LEVEL - 1) {
2087 parent = path->nodes[level + 1];
2088 pslot = path->slots[level + 1];
2091 if (!parent)
2092 return 1;
2094 left = btrfs_read_node_slot(parent, pslot - 1);
2095 if (IS_ERR(left))
2096 left = NULL;
2098 /* first, try to make some room in the middle buffer */
2099 if (left) {
2100 u32 left_nr;
2102 btrfs_tree_lock(left);
2103 btrfs_set_lock_blocking_write(left);
2105 left_nr = btrfs_header_nritems(left);
2106 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2107 wret = 1;
2108 } else {
2109 ret = btrfs_cow_block(trans, root, left, parent,
2110 pslot - 1, &left);
2111 if (ret)
2112 wret = 1;
2113 else {
2114 wret = push_node_left(trans, left, mid, 0);
2117 if (wret < 0)
2118 ret = wret;
2119 if (wret == 0) {
2120 struct btrfs_disk_key disk_key;
2121 orig_slot += left_nr;
2122 btrfs_node_key(mid, &disk_key, 0);
2123 ret = tree_mod_log_insert_key(parent, pslot,
2124 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2125 BUG_ON(ret < 0);
2126 btrfs_set_node_key(parent, &disk_key, pslot);
2127 btrfs_mark_buffer_dirty(parent);
2128 if (btrfs_header_nritems(left) > orig_slot) {
2129 path->nodes[level] = left;
2130 path->slots[level + 1] -= 1;
2131 path->slots[level] = orig_slot;
2132 btrfs_tree_unlock(mid);
2133 free_extent_buffer(mid);
2134 } else {
2135 orig_slot -=
2136 btrfs_header_nritems(left);
2137 path->slots[level] = orig_slot;
2138 btrfs_tree_unlock(left);
2139 free_extent_buffer(left);
2141 return 0;
2143 btrfs_tree_unlock(left);
2144 free_extent_buffer(left);
2146 right = btrfs_read_node_slot(parent, pslot + 1);
2147 if (IS_ERR(right))
2148 right = NULL;
2151 * then try to empty the right most buffer into the middle
2153 if (right) {
2154 u32 right_nr;
2156 btrfs_tree_lock(right);
2157 btrfs_set_lock_blocking_write(right);
2159 right_nr = btrfs_header_nritems(right);
2160 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
2161 wret = 1;
2162 } else {
2163 ret = btrfs_cow_block(trans, root, right,
2164 parent, pslot + 1,
2165 &right);
2166 if (ret)
2167 wret = 1;
2168 else {
2169 wret = balance_node_right(trans, right, mid);
2172 if (wret < 0)
2173 ret = wret;
2174 if (wret == 0) {
2175 struct btrfs_disk_key disk_key;
2177 btrfs_node_key(right, &disk_key, 0);
2178 ret = tree_mod_log_insert_key(parent, pslot + 1,
2179 MOD_LOG_KEY_REPLACE, GFP_NOFS);
2180 BUG_ON(ret < 0);
2181 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2182 btrfs_mark_buffer_dirty(parent);
2184 if (btrfs_header_nritems(mid) <= orig_slot) {
2185 path->nodes[level] = right;
2186 path->slots[level + 1] += 1;
2187 path->slots[level] = orig_slot -
2188 btrfs_header_nritems(mid);
2189 btrfs_tree_unlock(mid);
2190 free_extent_buffer(mid);
2191 } else {
2192 btrfs_tree_unlock(right);
2193 free_extent_buffer(right);
2195 return 0;
2197 btrfs_tree_unlock(right);
2198 free_extent_buffer(right);
2200 return 1;
2204 * readahead one full node of leaves, finding things that are close
2205 * to the block in 'slot', and triggering ra on them.
2207 static void reada_for_search(struct btrfs_fs_info *fs_info,
2208 struct btrfs_path *path,
2209 int level, int slot, u64 objectid)
2211 struct extent_buffer *node;
2212 struct btrfs_disk_key disk_key;
2213 u32 nritems;
2214 u64 search;
2215 u64 target;
2216 u64 nread = 0;
2217 struct extent_buffer *eb;
2218 u32 nr;
2219 u32 blocksize;
2220 u32 nscan = 0;
2222 if (level != 1)
2223 return;
2225 if (!path->nodes[level])
2226 return;
2228 node = path->nodes[level];
2230 search = btrfs_node_blockptr(node, slot);
2231 blocksize = fs_info->nodesize;
2232 eb = find_extent_buffer(fs_info, search);
2233 if (eb) {
2234 free_extent_buffer(eb);
2235 return;
2238 target = search;
2240 nritems = btrfs_header_nritems(node);
2241 nr = slot;
2243 while (1) {
2244 if (path->reada == READA_BACK) {
2245 if (nr == 0)
2246 break;
2247 nr--;
2248 } else if (path->reada == READA_FORWARD) {
2249 nr++;
2250 if (nr >= nritems)
2251 break;
2253 if (path->reada == READA_BACK && objectid) {
2254 btrfs_node_key(node, &disk_key, nr);
2255 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2256 break;
2258 search = btrfs_node_blockptr(node, nr);
2259 if ((search <= target && target - search <= 65536) ||
2260 (search > target && search - target <= 65536)) {
2261 readahead_tree_block(fs_info, search);
2262 nread += blocksize;
2264 nscan++;
2265 if ((nread > 65536 || nscan > 32))
2266 break;
2270 static noinline void reada_for_balance(struct btrfs_fs_info *fs_info,
2271 struct btrfs_path *path, int level)
2273 int slot;
2274 int nritems;
2275 struct extent_buffer *parent;
2276 struct extent_buffer *eb;
2277 u64 gen;
2278 u64 block1 = 0;
2279 u64 block2 = 0;
2281 parent = path->nodes[level + 1];
2282 if (!parent)
2283 return;
2285 nritems = btrfs_header_nritems(parent);
2286 slot = path->slots[level + 1];
2288 if (slot > 0) {
2289 block1 = btrfs_node_blockptr(parent, slot - 1);
2290 gen = btrfs_node_ptr_generation(parent, slot - 1);
2291 eb = find_extent_buffer(fs_info, block1);
2293 * if we get -eagain from btrfs_buffer_uptodate, we
2294 * don't want to return eagain here. That will loop
2295 * forever
2297 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2298 block1 = 0;
2299 free_extent_buffer(eb);
2301 if (slot + 1 < nritems) {
2302 block2 = btrfs_node_blockptr(parent, slot + 1);
2303 gen = btrfs_node_ptr_generation(parent, slot + 1);
2304 eb = find_extent_buffer(fs_info, block2);
2305 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2306 block2 = 0;
2307 free_extent_buffer(eb);
2310 if (block1)
2311 readahead_tree_block(fs_info, block1);
2312 if (block2)
2313 readahead_tree_block(fs_info, block2);
2318 * when we walk down the tree, it is usually safe to unlock the higher layers
2319 * in the tree. The exceptions are when our path goes through slot 0, because
2320 * operations on the tree might require changing key pointers higher up in the
2321 * tree.
2323 * callers might also have set path->keep_locks, which tells this code to keep
2324 * the lock if the path points to the last slot in the block. This is part of
2325 * walking through the tree, and selecting the next slot in the higher block.
2327 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2328 * if lowest_unlock is 1, level 0 won't be unlocked
2330 static noinline void unlock_up(struct btrfs_path *path, int level,
2331 int lowest_unlock, int min_write_lock_level,
2332 int *write_lock_level)
2334 int i;
2335 int skip_level = level;
2336 int no_skips = 0;
2337 struct extent_buffer *t;
2339 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2340 if (!path->nodes[i])
2341 break;
2342 if (!path->locks[i])
2343 break;
2344 if (!no_skips && path->slots[i] == 0) {
2345 skip_level = i + 1;
2346 continue;
2348 if (!no_skips && path->keep_locks) {
2349 u32 nritems;
2350 t = path->nodes[i];
2351 nritems = btrfs_header_nritems(t);
2352 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2353 skip_level = i + 1;
2354 continue;
2357 if (skip_level < i && i >= lowest_unlock)
2358 no_skips = 1;
2360 t = path->nodes[i];
2361 if (i >= lowest_unlock && i > skip_level) {
2362 btrfs_tree_unlock_rw(t, path->locks[i]);
2363 path->locks[i] = 0;
2364 if (write_lock_level &&
2365 i > min_write_lock_level &&
2366 i <= *write_lock_level) {
2367 *write_lock_level = i - 1;
2374 * helper function for btrfs_search_slot. The goal is to find a block
2375 * in cache without setting the path to blocking. If we find the block
2376 * we return zero and the path is unchanged.
2378 * If we can't find the block, we set the path blocking and do some
2379 * reada. -EAGAIN is returned and the search must be repeated.
2381 static int
2382 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
2383 struct extent_buffer **eb_ret, int level, int slot,
2384 const struct btrfs_key *key)
2386 struct btrfs_fs_info *fs_info = root->fs_info;
2387 u64 blocknr;
2388 u64 gen;
2389 struct extent_buffer *b = *eb_ret;
2390 struct extent_buffer *tmp;
2391 struct btrfs_key first_key;
2392 int ret;
2393 int parent_level;
2395 blocknr = btrfs_node_blockptr(b, slot);
2396 gen = btrfs_node_ptr_generation(b, slot);
2397 parent_level = btrfs_header_level(b);
2398 btrfs_node_key_to_cpu(b, &first_key, slot);
2400 tmp = find_extent_buffer(fs_info, blocknr);
2401 if (tmp) {
2402 /* first we do an atomic uptodate check */
2403 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2405 * Do extra check for first_key, eb can be stale due to
2406 * being cached, read from scrub, or have multiple
2407 * parents (shared tree blocks).
2409 if (btrfs_verify_level_key(tmp,
2410 parent_level - 1, &first_key, gen)) {
2411 free_extent_buffer(tmp);
2412 return -EUCLEAN;
2414 *eb_ret = tmp;
2415 return 0;
2418 /* the pages were up to date, but we failed
2419 * the generation number check. Do a full
2420 * read for the generation number that is correct.
2421 * We must do this without dropping locks so
2422 * we can trust our generation number
2424 btrfs_set_path_blocking(p);
2426 /* now we're allowed to do a blocking uptodate check */
2427 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
2428 if (!ret) {
2429 *eb_ret = tmp;
2430 return 0;
2432 free_extent_buffer(tmp);
2433 btrfs_release_path(p);
2434 return -EIO;
2438 * reduce lock contention at high levels
2439 * of the btree by dropping locks before
2440 * we read. Don't release the lock on the current
2441 * level because we need to walk this node to figure
2442 * out which blocks to read.
2444 btrfs_unlock_up_safe(p, level + 1);
2445 btrfs_set_path_blocking(p);
2447 if (p->reada != READA_NONE)
2448 reada_for_search(fs_info, p, level, slot, key->objectid);
2450 ret = -EAGAIN;
2451 tmp = read_tree_block(fs_info, blocknr, gen, parent_level - 1,
2452 &first_key);
2453 if (!IS_ERR(tmp)) {
2455 * If the read above didn't mark this buffer up to date,
2456 * it will never end up being up to date. Set ret to EIO now
2457 * and give up so that our caller doesn't loop forever
2458 * on our EAGAINs.
2460 if (!extent_buffer_uptodate(tmp))
2461 ret = -EIO;
2462 free_extent_buffer(tmp);
2463 } else {
2464 ret = PTR_ERR(tmp);
2467 btrfs_release_path(p);
2468 return ret;
2472 * helper function for btrfs_search_slot. This does all of the checks
2473 * for node-level blocks and does any balancing required based on
2474 * the ins_len.
2476 * If no extra work was required, zero is returned. If we had to
2477 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2478 * start over
2480 static int
2481 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2482 struct btrfs_root *root, struct btrfs_path *p,
2483 struct extent_buffer *b, int level, int ins_len,
2484 int *write_lock_level)
2486 struct btrfs_fs_info *fs_info = root->fs_info;
2487 int ret;
2489 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2490 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
2491 int sret;
2493 if (*write_lock_level < level + 1) {
2494 *write_lock_level = level + 1;
2495 btrfs_release_path(p);
2496 goto again;
2499 btrfs_set_path_blocking(p);
2500 reada_for_balance(fs_info, p, level);
2501 sret = split_node(trans, root, p, level);
2503 BUG_ON(sret > 0);
2504 if (sret) {
2505 ret = sret;
2506 goto done;
2508 b = p->nodes[level];
2509 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2510 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
2511 int sret;
2513 if (*write_lock_level < level + 1) {
2514 *write_lock_level = level + 1;
2515 btrfs_release_path(p);
2516 goto again;
2519 btrfs_set_path_blocking(p);
2520 reada_for_balance(fs_info, p, level);
2521 sret = balance_level(trans, root, p, level);
2523 if (sret) {
2524 ret = sret;
2525 goto done;
2527 b = p->nodes[level];
2528 if (!b) {
2529 btrfs_release_path(p);
2530 goto again;
2532 BUG_ON(btrfs_header_nritems(b) == 1);
2534 return 0;
2536 again:
2537 ret = -EAGAIN;
2538 done:
2539 return ret;
2542 static int key_search(struct extent_buffer *b, const struct btrfs_key *key,
2543 int level, int *prev_cmp, int *slot)
2545 if (*prev_cmp != 0) {
2546 *prev_cmp = btrfs_bin_search(b, key, level, slot);
2547 return *prev_cmp;
2550 *slot = 0;
2552 return 0;
2555 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
2556 u64 iobjectid, u64 ioff, u8 key_type,
2557 struct btrfs_key *found_key)
2559 int ret;
2560 struct btrfs_key key;
2561 struct extent_buffer *eb;
2563 ASSERT(path);
2564 ASSERT(found_key);
2566 key.type = key_type;
2567 key.objectid = iobjectid;
2568 key.offset = ioff;
2570 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
2571 if (ret < 0)
2572 return ret;
2574 eb = path->nodes[0];
2575 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
2576 ret = btrfs_next_leaf(fs_root, path);
2577 if (ret)
2578 return ret;
2579 eb = path->nodes[0];
2582 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
2583 if (found_key->type != key.type ||
2584 found_key->objectid != key.objectid)
2585 return 1;
2587 return 0;
2590 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
2591 struct btrfs_path *p,
2592 int write_lock_level)
2594 struct btrfs_fs_info *fs_info = root->fs_info;
2595 struct extent_buffer *b;
2596 int root_lock;
2597 int level = 0;
2599 /* We try very hard to do read locks on the root */
2600 root_lock = BTRFS_READ_LOCK;
2602 if (p->search_commit_root) {
2604 * The commit roots are read only so we always do read locks,
2605 * and we always must hold the commit_root_sem when doing
2606 * searches on them, the only exception is send where we don't
2607 * want to block transaction commits for a long time, so
2608 * we need to clone the commit root in order to avoid races
2609 * with transaction commits that create a snapshot of one of
2610 * the roots used by a send operation.
2612 if (p->need_commit_sem) {
2613 down_read(&fs_info->commit_root_sem);
2614 b = btrfs_clone_extent_buffer(root->commit_root);
2615 up_read(&fs_info->commit_root_sem);
2616 if (!b)
2617 return ERR_PTR(-ENOMEM);
2619 } else {
2620 b = root->commit_root;
2621 atomic_inc(&b->refs);
2623 level = btrfs_header_level(b);
2625 * Ensure that all callers have set skip_locking when
2626 * p->search_commit_root = 1.
2628 ASSERT(p->skip_locking == 1);
2630 goto out;
2633 if (p->skip_locking) {
2634 b = btrfs_root_node(root);
2635 level = btrfs_header_level(b);
2636 goto out;
2640 * If the level is set to maximum, we can skip trying to get the read
2641 * lock.
2643 if (write_lock_level < BTRFS_MAX_LEVEL) {
2645 * We don't know the level of the root node until we actually
2646 * have it read locked
2648 b = btrfs_read_lock_root_node(root);
2649 level = btrfs_header_level(b);
2650 if (level > write_lock_level)
2651 goto out;
2653 /* Whoops, must trade for write lock */
2654 btrfs_tree_read_unlock(b);
2655 free_extent_buffer(b);
2658 b = btrfs_lock_root_node(root);
2659 root_lock = BTRFS_WRITE_LOCK;
2661 /* The level might have changed, check again */
2662 level = btrfs_header_level(b);
2664 out:
2665 p->nodes[level] = b;
2666 if (!p->skip_locking)
2667 p->locks[level] = root_lock;
2669 * Callers are responsible for dropping b's references.
2671 return b;
2676 * btrfs_search_slot - look for a key in a tree and perform necessary
2677 * modifications to preserve tree invariants.
2679 * @trans: Handle of transaction, used when modifying the tree
2680 * @p: Holds all btree nodes along the search path
2681 * @root: The root node of the tree
2682 * @key: The key we are looking for
2683 * @ins_len: Indicates purpose of search, for inserts it is 1, for
2684 * deletions it's -1. 0 for plain searches
2685 * @cow: boolean should CoW operations be performed. Must always be 1
2686 * when modifying the tree.
2688 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2689 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2691 * If @key is found, 0 is returned and you can find the item in the leaf level
2692 * of the path (level 0)
2694 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2695 * points to the slot where it should be inserted
2697 * If an error is encountered while searching the tree a negative error number
2698 * is returned
2700 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2701 const struct btrfs_key *key, struct btrfs_path *p,
2702 int ins_len, int cow)
2704 struct extent_buffer *b;
2705 int slot;
2706 int ret;
2707 int err;
2708 int level;
2709 int lowest_unlock = 1;
2710 /* everything at write_lock_level or lower must be write locked */
2711 int write_lock_level = 0;
2712 u8 lowest_level = 0;
2713 int min_write_lock_level;
2714 int prev_cmp;
2716 lowest_level = p->lowest_level;
2717 WARN_ON(lowest_level && ins_len > 0);
2718 WARN_ON(p->nodes[0] != NULL);
2719 BUG_ON(!cow && ins_len);
2721 if (ins_len < 0) {
2722 lowest_unlock = 2;
2724 /* when we are removing items, we might have to go up to level
2725 * two as we update tree pointers Make sure we keep write
2726 * for those levels as well
2728 write_lock_level = 2;
2729 } else if (ins_len > 0) {
2731 * for inserting items, make sure we have a write lock on
2732 * level 1 so we can update keys
2734 write_lock_level = 1;
2737 if (!cow)
2738 write_lock_level = -1;
2740 if (cow && (p->keep_locks || p->lowest_level))
2741 write_lock_level = BTRFS_MAX_LEVEL;
2743 min_write_lock_level = write_lock_level;
2745 again:
2746 prev_cmp = -1;
2747 b = btrfs_search_slot_get_root(root, p, write_lock_level);
2748 if (IS_ERR(b)) {
2749 ret = PTR_ERR(b);
2750 goto done;
2753 while (b) {
2754 int dec = 0;
2756 level = btrfs_header_level(b);
2758 if (cow) {
2759 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
2762 * if we don't really need to cow this block
2763 * then we don't want to set the path blocking,
2764 * so we test it here
2766 if (!should_cow_block(trans, root, b)) {
2767 trans->dirty = true;
2768 goto cow_done;
2772 * must have write locks on this node and the
2773 * parent
2775 if (level > write_lock_level ||
2776 (level + 1 > write_lock_level &&
2777 level + 1 < BTRFS_MAX_LEVEL &&
2778 p->nodes[level + 1])) {
2779 write_lock_level = level + 1;
2780 btrfs_release_path(p);
2781 goto again;
2784 btrfs_set_path_blocking(p);
2785 if (last_level)
2786 err = btrfs_cow_block(trans, root, b, NULL, 0,
2787 &b);
2788 else
2789 err = btrfs_cow_block(trans, root, b,
2790 p->nodes[level + 1],
2791 p->slots[level + 1], &b);
2792 if (err) {
2793 ret = err;
2794 goto done;
2797 cow_done:
2798 p->nodes[level] = b;
2800 * Leave path with blocking locks to avoid massive
2801 * lock context switch, this is made on purpose.
2805 * we have a lock on b and as long as we aren't changing
2806 * the tree, there is no way to for the items in b to change.
2807 * It is safe to drop the lock on our parent before we
2808 * go through the expensive btree search on b.
2810 * If we're inserting or deleting (ins_len != 0), then we might
2811 * be changing slot zero, which may require changing the parent.
2812 * So, we can't drop the lock until after we know which slot
2813 * we're operating on.
2815 if (!ins_len && !p->keep_locks) {
2816 int u = level + 1;
2818 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
2819 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
2820 p->locks[u] = 0;
2824 ret = key_search(b, key, level, &prev_cmp, &slot);
2825 if (ret < 0)
2826 goto done;
2828 if (level == 0) {
2829 p->slots[level] = slot;
2830 if (ins_len > 0 &&
2831 btrfs_leaf_free_space(b) < ins_len) {
2832 if (write_lock_level < 1) {
2833 write_lock_level = 1;
2834 btrfs_release_path(p);
2835 goto again;
2838 btrfs_set_path_blocking(p);
2839 err = split_leaf(trans, root, key,
2840 p, ins_len, ret == 0);
2842 BUG_ON(err > 0);
2843 if (err) {
2844 ret = err;
2845 goto done;
2848 if (!p->search_for_split)
2849 unlock_up(p, level, lowest_unlock,
2850 min_write_lock_level, NULL);
2851 goto done;
2853 if (ret && slot > 0) {
2854 dec = 1;
2855 slot--;
2857 p->slots[level] = slot;
2858 err = setup_nodes_for_search(trans, root, p, b, level, ins_len,
2859 &write_lock_level);
2860 if (err == -EAGAIN)
2861 goto again;
2862 if (err) {
2863 ret = err;
2864 goto done;
2866 b = p->nodes[level];
2867 slot = p->slots[level];
2870 * Slot 0 is special, if we change the key we have to update
2871 * the parent pointer which means we must have a write lock on
2872 * the parent
2874 if (slot == 0 && ins_len && write_lock_level < level + 1) {
2875 write_lock_level = level + 1;
2876 btrfs_release_path(p);
2877 goto again;
2880 unlock_up(p, level, lowest_unlock, min_write_lock_level,
2881 &write_lock_level);
2883 if (level == lowest_level) {
2884 if (dec)
2885 p->slots[level]++;
2886 goto done;
2889 err = read_block_for_search(root, p, &b, level, slot, key);
2890 if (err == -EAGAIN)
2891 goto again;
2892 if (err) {
2893 ret = err;
2894 goto done;
2897 if (!p->skip_locking) {
2898 level = btrfs_header_level(b);
2899 if (level <= write_lock_level) {
2900 if (!btrfs_try_tree_write_lock(b)) {
2901 btrfs_set_path_blocking(p);
2902 btrfs_tree_lock(b);
2904 p->locks[level] = BTRFS_WRITE_LOCK;
2905 } else {
2906 if (!btrfs_tree_read_lock_atomic(b)) {
2907 btrfs_set_path_blocking(p);
2908 btrfs_tree_read_lock(b);
2910 p->locks[level] = BTRFS_READ_LOCK;
2912 p->nodes[level] = b;
2915 ret = 1;
2916 done:
2918 * we don't really know what they plan on doing with the path
2919 * from here on, so for now just mark it as blocking
2921 if (!p->leave_spinning)
2922 btrfs_set_path_blocking(p);
2923 if (ret < 0 && !p->skip_release_on_error)
2924 btrfs_release_path(p);
2925 return ret;
2929 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2930 * current state of the tree together with the operations recorded in the tree
2931 * modification log to search for the key in a previous version of this tree, as
2932 * denoted by the time_seq parameter.
2934 * Naturally, there is no support for insert, delete or cow operations.
2936 * The resulting path and return value will be set up as if we called
2937 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2939 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
2940 struct btrfs_path *p, u64 time_seq)
2942 struct btrfs_fs_info *fs_info = root->fs_info;
2943 struct extent_buffer *b;
2944 int slot;
2945 int ret;
2946 int err;
2947 int level;
2948 int lowest_unlock = 1;
2949 u8 lowest_level = 0;
2950 int prev_cmp = -1;
2952 lowest_level = p->lowest_level;
2953 WARN_ON(p->nodes[0] != NULL);
2955 if (p->search_commit_root) {
2956 BUG_ON(time_seq);
2957 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2960 again:
2961 b = get_old_root(root, time_seq);
2962 if (!b) {
2963 ret = -EIO;
2964 goto done;
2966 level = btrfs_header_level(b);
2967 p->locks[level] = BTRFS_READ_LOCK;
2969 while (b) {
2970 int dec = 0;
2972 level = btrfs_header_level(b);
2973 p->nodes[level] = b;
2976 * we have a lock on b and as long as we aren't changing
2977 * the tree, there is no way to for the items in b to change.
2978 * It is safe to drop the lock on our parent before we
2979 * go through the expensive btree search on b.
2981 btrfs_unlock_up_safe(p, level + 1);
2984 * Since we can unwind ebs we want to do a real search every
2985 * time.
2987 prev_cmp = -1;
2988 ret = key_search(b, key, level, &prev_cmp, &slot);
2989 if (ret < 0)
2990 goto done;
2992 if (level == 0) {
2993 p->slots[level] = slot;
2994 unlock_up(p, level, lowest_unlock, 0, NULL);
2995 goto done;
2998 if (ret && slot > 0) {
2999 dec = 1;
3000 slot--;
3002 p->slots[level] = slot;
3003 unlock_up(p, level, lowest_unlock, 0, NULL);
3005 if (level == lowest_level) {
3006 if (dec)
3007 p->slots[level]++;
3008 goto done;
3011 err = read_block_for_search(root, p, &b, level, slot, key);
3012 if (err == -EAGAIN)
3013 goto again;
3014 if (err) {
3015 ret = err;
3016 goto done;
3019 level = btrfs_header_level(b);
3020 if (!btrfs_tree_read_lock_atomic(b)) {
3021 btrfs_set_path_blocking(p);
3022 btrfs_tree_read_lock(b);
3024 b = tree_mod_log_rewind(fs_info, p, b, time_seq);
3025 if (!b) {
3026 ret = -ENOMEM;
3027 goto done;
3029 p->locks[level] = BTRFS_READ_LOCK;
3030 p->nodes[level] = b;
3032 ret = 1;
3033 done:
3034 if (!p->leave_spinning)
3035 btrfs_set_path_blocking(p);
3036 if (ret < 0)
3037 btrfs_release_path(p);
3039 return ret;
3043 * helper to use instead of search slot if no exact match is needed but
3044 * instead the next or previous item should be returned.
3045 * When find_higher is true, the next higher item is returned, the next lower
3046 * otherwise.
3047 * When return_any and find_higher are both true, and no higher item is found,
3048 * return the next lower instead.
3049 * When return_any is true and find_higher is false, and no lower item is found,
3050 * return the next higher instead.
3051 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3052 * < 0 on error
3054 int btrfs_search_slot_for_read(struct btrfs_root *root,
3055 const struct btrfs_key *key,
3056 struct btrfs_path *p, int find_higher,
3057 int return_any)
3059 int ret;
3060 struct extent_buffer *leaf;
3062 again:
3063 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
3064 if (ret <= 0)
3065 return ret;
3067 * a return value of 1 means the path is at the position where the
3068 * item should be inserted. Normally this is the next bigger item,
3069 * but in case the previous item is the last in a leaf, path points
3070 * to the first free slot in the previous leaf, i.e. at an invalid
3071 * item.
3073 leaf = p->nodes[0];
3075 if (find_higher) {
3076 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
3077 ret = btrfs_next_leaf(root, p);
3078 if (ret <= 0)
3079 return ret;
3080 if (!return_any)
3081 return 1;
3083 * no higher item found, return the next
3084 * lower instead
3086 return_any = 0;
3087 find_higher = 0;
3088 btrfs_release_path(p);
3089 goto again;
3091 } else {
3092 if (p->slots[0] == 0) {
3093 ret = btrfs_prev_leaf(root, p);
3094 if (ret < 0)
3095 return ret;
3096 if (!ret) {
3097 leaf = p->nodes[0];
3098 if (p->slots[0] == btrfs_header_nritems(leaf))
3099 p->slots[0]--;
3100 return 0;
3102 if (!return_any)
3103 return 1;
3105 * no lower item found, return the next
3106 * higher instead
3108 return_any = 0;
3109 find_higher = 1;
3110 btrfs_release_path(p);
3111 goto again;
3112 } else {
3113 --p->slots[0];
3116 return 0;
3120 * adjust the pointers going up the tree, starting at level
3121 * making sure the right key of each node is points to 'key'.
3122 * This is used after shifting pointers to the left, so it stops
3123 * fixing up pointers when a given leaf/node is not in slot 0 of the
3124 * higher levels
3127 static void fixup_low_keys(struct btrfs_path *path,
3128 struct btrfs_disk_key *key, int level)
3130 int i;
3131 struct extent_buffer *t;
3132 int ret;
3134 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
3135 int tslot = path->slots[i];
3137 if (!path->nodes[i])
3138 break;
3139 t = path->nodes[i];
3140 ret = tree_mod_log_insert_key(t, tslot, MOD_LOG_KEY_REPLACE,
3141 GFP_ATOMIC);
3142 BUG_ON(ret < 0);
3143 btrfs_set_node_key(t, key, tslot);
3144 btrfs_mark_buffer_dirty(path->nodes[i]);
3145 if (tslot != 0)
3146 break;
3151 * update item key.
3153 * This function isn't completely safe. It's the caller's responsibility
3154 * that the new key won't break the order
3156 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
3157 struct btrfs_path *path,
3158 const struct btrfs_key *new_key)
3160 struct btrfs_disk_key disk_key;
3161 struct extent_buffer *eb;
3162 int slot;
3164 eb = path->nodes[0];
3165 slot = path->slots[0];
3166 if (slot > 0) {
3167 btrfs_item_key(eb, &disk_key, slot - 1);
3168 if (unlikely(comp_keys(&disk_key, new_key) >= 0)) {
3169 btrfs_crit(fs_info,
3170 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3171 slot, btrfs_disk_key_objectid(&disk_key),
3172 btrfs_disk_key_type(&disk_key),
3173 btrfs_disk_key_offset(&disk_key),
3174 new_key->objectid, new_key->type,
3175 new_key->offset);
3176 btrfs_print_leaf(eb);
3177 BUG();
3180 if (slot < btrfs_header_nritems(eb) - 1) {
3181 btrfs_item_key(eb, &disk_key, slot + 1);
3182 if (unlikely(comp_keys(&disk_key, new_key) <= 0)) {
3183 btrfs_crit(fs_info,
3184 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3185 slot, btrfs_disk_key_objectid(&disk_key),
3186 btrfs_disk_key_type(&disk_key),
3187 btrfs_disk_key_offset(&disk_key),
3188 new_key->objectid, new_key->type,
3189 new_key->offset);
3190 btrfs_print_leaf(eb);
3191 BUG();
3195 btrfs_cpu_key_to_disk(&disk_key, new_key);
3196 btrfs_set_item_key(eb, &disk_key, slot);
3197 btrfs_mark_buffer_dirty(eb);
3198 if (slot == 0)
3199 fixup_low_keys(path, &disk_key, 1);
3203 * try to push data from one node into the next node left in the
3204 * tree.
3206 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3207 * error, and > 0 if there was no room in the left hand block.
3209 static int push_node_left(struct btrfs_trans_handle *trans,
3210 struct extent_buffer *dst,
3211 struct extent_buffer *src, int empty)
3213 struct btrfs_fs_info *fs_info = trans->fs_info;
3214 int push_items = 0;
3215 int src_nritems;
3216 int dst_nritems;
3217 int ret = 0;
3219 src_nritems = btrfs_header_nritems(src);
3220 dst_nritems = btrfs_header_nritems(dst);
3221 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3222 WARN_ON(btrfs_header_generation(src) != trans->transid);
3223 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3225 if (!empty && src_nritems <= 8)
3226 return 1;
3228 if (push_items <= 0)
3229 return 1;
3231 if (empty) {
3232 push_items = min(src_nritems, push_items);
3233 if (push_items < src_nritems) {
3234 /* leave at least 8 pointers in the node if
3235 * we aren't going to empty it
3237 if (src_nritems - push_items < 8) {
3238 if (push_items <= 8)
3239 return 1;
3240 push_items -= 8;
3243 } else
3244 push_items = min(src_nritems - 8, push_items);
3246 ret = tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
3247 if (ret) {
3248 btrfs_abort_transaction(trans, ret);
3249 return ret;
3251 copy_extent_buffer(dst, src,
3252 btrfs_node_key_ptr_offset(dst_nritems),
3253 btrfs_node_key_ptr_offset(0),
3254 push_items * sizeof(struct btrfs_key_ptr));
3256 if (push_items < src_nritems) {
3258 * Don't call tree_mod_log_insert_move here, key removal was
3259 * already fully logged by tree_mod_log_eb_copy above.
3261 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3262 btrfs_node_key_ptr_offset(push_items),
3263 (src_nritems - push_items) *
3264 sizeof(struct btrfs_key_ptr));
3266 btrfs_set_header_nritems(src, src_nritems - push_items);
3267 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3268 btrfs_mark_buffer_dirty(src);
3269 btrfs_mark_buffer_dirty(dst);
3271 return ret;
3275 * try to push data from one node into the next node right in the
3276 * tree.
3278 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3279 * error, and > 0 if there was no room in the right hand block.
3281 * this will only push up to 1/2 the contents of the left node over
3283 static int balance_node_right(struct btrfs_trans_handle *trans,
3284 struct extent_buffer *dst,
3285 struct extent_buffer *src)
3287 struct btrfs_fs_info *fs_info = trans->fs_info;
3288 int push_items = 0;
3289 int max_push;
3290 int src_nritems;
3291 int dst_nritems;
3292 int ret = 0;
3294 WARN_ON(btrfs_header_generation(src) != trans->transid);
3295 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3297 src_nritems = btrfs_header_nritems(src);
3298 dst_nritems = btrfs_header_nritems(dst);
3299 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
3300 if (push_items <= 0)
3301 return 1;
3303 if (src_nritems < 4)
3304 return 1;
3306 max_push = src_nritems / 2 + 1;
3307 /* don't try to empty the node */
3308 if (max_push >= src_nritems)
3309 return 1;
3311 if (max_push < push_items)
3312 push_items = max_push;
3314 ret = tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
3315 BUG_ON(ret < 0);
3316 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3317 btrfs_node_key_ptr_offset(0),
3318 (dst_nritems) *
3319 sizeof(struct btrfs_key_ptr));
3321 ret = tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
3322 push_items);
3323 if (ret) {
3324 btrfs_abort_transaction(trans, ret);
3325 return ret;
3327 copy_extent_buffer(dst, src,
3328 btrfs_node_key_ptr_offset(0),
3329 btrfs_node_key_ptr_offset(src_nritems - push_items),
3330 push_items * sizeof(struct btrfs_key_ptr));
3332 btrfs_set_header_nritems(src, src_nritems - push_items);
3333 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3335 btrfs_mark_buffer_dirty(src);
3336 btrfs_mark_buffer_dirty(dst);
3338 return ret;
3342 * helper function to insert a new root level in the tree.
3343 * A new node is allocated, and a single item is inserted to
3344 * point to the existing root
3346 * returns zero on success or < 0 on failure.
3348 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3349 struct btrfs_root *root,
3350 struct btrfs_path *path, int level)
3352 struct btrfs_fs_info *fs_info = root->fs_info;
3353 u64 lower_gen;
3354 struct extent_buffer *lower;
3355 struct extent_buffer *c;
3356 struct extent_buffer *old;
3357 struct btrfs_disk_key lower_key;
3358 int ret;
3360 BUG_ON(path->nodes[level]);
3361 BUG_ON(path->nodes[level-1] != root->node);
3363 lower = path->nodes[level-1];
3364 if (level == 1)
3365 btrfs_item_key(lower, &lower_key, 0);
3366 else
3367 btrfs_node_key(lower, &lower_key, 0);
3369 c = alloc_tree_block_no_bg_flush(trans, root, 0, &lower_key, level,
3370 root->node->start, 0);
3371 if (IS_ERR(c))
3372 return PTR_ERR(c);
3374 root_add_used(root, fs_info->nodesize);
3376 btrfs_set_header_nritems(c, 1);
3377 btrfs_set_node_key(c, &lower_key, 0);
3378 btrfs_set_node_blockptr(c, 0, lower->start);
3379 lower_gen = btrfs_header_generation(lower);
3380 WARN_ON(lower_gen != trans->transid);
3382 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3384 btrfs_mark_buffer_dirty(c);
3386 old = root->node;
3387 ret = tree_mod_log_insert_root(root->node, c, 0);
3388 BUG_ON(ret < 0);
3389 rcu_assign_pointer(root->node, c);
3391 /* the super has an extra ref to root->node */
3392 free_extent_buffer(old);
3394 add_root_to_dirty_list(root);
3395 atomic_inc(&c->refs);
3396 path->nodes[level] = c;
3397 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
3398 path->slots[level] = 0;
3399 return 0;
3403 * worker function to insert a single pointer in a node.
3404 * the node should have enough room for the pointer already
3406 * slot and level indicate where you want the key to go, and
3407 * blocknr is the block the key points to.
3409 static void insert_ptr(struct btrfs_trans_handle *trans,
3410 struct btrfs_path *path,
3411 struct btrfs_disk_key *key, u64 bytenr,
3412 int slot, int level)
3414 struct extent_buffer *lower;
3415 int nritems;
3416 int ret;
3418 BUG_ON(!path->nodes[level]);
3419 btrfs_assert_tree_locked(path->nodes[level]);
3420 lower = path->nodes[level];
3421 nritems = btrfs_header_nritems(lower);
3422 BUG_ON(slot > nritems);
3423 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
3424 if (slot != nritems) {
3425 if (level) {
3426 ret = tree_mod_log_insert_move(lower, slot + 1, slot,
3427 nritems - slot);
3428 BUG_ON(ret < 0);
3430 memmove_extent_buffer(lower,
3431 btrfs_node_key_ptr_offset(slot + 1),
3432 btrfs_node_key_ptr_offset(slot),
3433 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3435 if (level) {
3436 ret = tree_mod_log_insert_key(lower, slot, MOD_LOG_KEY_ADD,
3437 GFP_NOFS);
3438 BUG_ON(ret < 0);
3440 btrfs_set_node_key(lower, key, slot);
3441 btrfs_set_node_blockptr(lower, slot, bytenr);
3442 WARN_ON(trans->transid == 0);
3443 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3444 btrfs_set_header_nritems(lower, nritems + 1);
3445 btrfs_mark_buffer_dirty(lower);
3449 * split the node at the specified level in path in two.
3450 * The path is corrected to point to the appropriate node after the split
3452 * Before splitting this tries to make some room in the node by pushing
3453 * left and right, if either one works, it returns right away.
3455 * returns 0 on success and < 0 on failure
3457 static noinline int split_node(struct btrfs_trans_handle *trans,
3458 struct btrfs_root *root,
3459 struct btrfs_path *path, int level)
3461 struct btrfs_fs_info *fs_info = root->fs_info;
3462 struct extent_buffer *c;
3463 struct extent_buffer *split;
3464 struct btrfs_disk_key disk_key;
3465 int mid;
3466 int ret;
3467 u32 c_nritems;
3469 c = path->nodes[level];
3470 WARN_ON(btrfs_header_generation(c) != trans->transid);
3471 if (c == root->node) {
3473 * trying to split the root, lets make a new one
3475 * tree mod log: We don't log_removal old root in
3476 * insert_new_root, because that root buffer will be kept as a
3477 * normal node. We are going to log removal of half of the
3478 * elements below with tree_mod_log_eb_copy. We're holding a
3479 * tree lock on the buffer, which is why we cannot race with
3480 * other tree_mod_log users.
3482 ret = insert_new_root(trans, root, path, level + 1);
3483 if (ret)
3484 return ret;
3485 } else {
3486 ret = push_nodes_for_insert(trans, root, path, level);
3487 c = path->nodes[level];
3488 if (!ret && btrfs_header_nritems(c) <
3489 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
3490 return 0;
3491 if (ret < 0)
3492 return ret;
3495 c_nritems = btrfs_header_nritems(c);
3496 mid = (c_nritems + 1) / 2;
3497 btrfs_node_key(c, &disk_key, mid);
3499 split = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, level,
3500 c->start, 0);
3501 if (IS_ERR(split))
3502 return PTR_ERR(split);
3504 root_add_used(root, fs_info->nodesize);
3505 ASSERT(btrfs_header_level(c) == level);
3507 ret = tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
3508 if (ret) {
3509 btrfs_abort_transaction(trans, ret);
3510 return ret;
3512 copy_extent_buffer(split, c,
3513 btrfs_node_key_ptr_offset(0),
3514 btrfs_node_key_ptr_offset(mid),
3515 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3516 btrfs_set_header_nritems(split, c_nritems - mid);
3517 btrfs_set_header_nritems(c, mid);
3518 ret = 0;
3520 btrfs_mark_buffer_dirty(c);
3521 btrfs_mark_buffer_dirty(split);
3523 insert_ptr(trans, path, &disk_key, split->start,
3524 path->slots[level + 1] + 1, level + 1);
3526 if (path->slots[level] >= mid) {
3527 path->slots[level] -= mid;
3528 btrfs_tree_unlock(c);
3529 free_extent_buffer(c);
3530 path->nodes[level] = split;
3531 path->slots[level + 1] += 1;
3532 } else {
3533 btrfs_tree_unlock(split);
3534 free_extent_buffer(split);
3536 return ret;
3540 * how many bytes are required to store the items in a leaf. start
3541 * and nr indicate which items in the leaf to check. This totals up the
3542 * space used both by the item structs and the item data
3544 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3546 struct btrfs_item *start_item;
3547 struct btrfs_item *end_item;
3548 struct btrfs_map_token token;
3549 int data_len;
3550 int nritems = btrfs_header_nritems(l);
3551 int end = min(nritems, start + nr) - 1;
3553 if (!nr)
3554 return 0;
3555 btrfs_init_map_token(&token, l);
3556 start_item = btrfs_item_nr(start);
3557 end_item = btrfs_item_nr(end);
3558 data_len = btrfs_token_item_offset(l, start_item, &token) +
3559 btrfs_token_item_size(l, start_item, &token);
3560 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3561 data_len += sizeof(struct btrfs_item) * nr;
3562 WARN_ON(data_len < 0);
3563 return data_len;
3567 * The space between the end of the leaf items and
3568 * the start of the leaf data. IOW, how much room
3569 * the leaf has left for both items and data
3571 noinline int btrfs_leaf_free_space(struct extent_buffer *leaf)
3573 struct btrfs_fs_info *fs_info = leaf->fs_info;
3574 int nritems = btrfs_header_nritems(leaf);
3575 int ret;
3577 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
3578 if (ret < 0) {
3579 btrfs_crit(fs_info,
3580 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3581 ret,
3582 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
3583 leaf_space_used(leaf, 0, nritems), nritems);
3585 return ret;
3589 * min slot controls the lowest index we're willing to push to the
3590 * right. We'll push up to and including min_slot, but no lower
3592 static noinline int __push_leaf_right(struct btrfs_path *path,
3593 int data_size, int empty,
3594 struct extent_buffer *right,
3595 int free_space, u32 left_nritems,
3596 u32 min_slot)
3598 struct btrfs_fs_info *fs_info = right->fs_info;
3599 struct extent_buffer *left = path->nodes[0];
3600 struct extent_buffer *upper = path->nodes[1];
3601 struct btrfs_map_token token;
3602 struct btrfs_disk_key disk_key;
3603 int slot;
3604 u32 i;
3605 int push_space = 0;
3606 int push_items = 0;
3607 struct btrfs_item *item;
3608 u32 nr;
3609 u32 right_nritems;
3610 u32 data_end;
3611 u32 this_item_size;
3613 if (empty)
3614 nr = 0;
3615 else
3616 nr = max_t(u32, 1, min_slot);
3618 if (path->slots[0] >= left_nritems)
3619 push_space += data_size;
3621 slot = path->slots[1];
3622 i = left_nritems - 1;
3623 while (i >= nr) {
3624 item = btrfs_item_nr(i);
3626 if (!empty && push_items > 0) {
3627 if (path->slots[0] > i)
3628 break;
3629 if (path->slots[0] == i) {
3630 int space = btrfs_leaf_free_space(left);
3632 if (space + push_space * 2 > free_space)
3633 break;
3637 if (path->slots[0] == i)
3638 push_space += data_size;
3640 this_item_size = btrfs_item_size(left, item);
3641 if (this_item_size + sizeof(*item) + push_space > free_space)
3642 break;
3644 push_items++;
3645 push_space += this_item_size + sizeof(*item);
3646 if (i == 0)
3647 break;
3648 i--;
3651 if (push_items == 0)
3652 goto out_unlock;
3654 WARN_ON(!empty && push_items == left_nritems);
3656 /* push left to right */
3657 right_nritems = btrfs_header_nritems(right);
3659 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3660 push_space -= leaf_data_end(left);
3662 /* make room in the right data area */
3663 data_end = leaf_data_end(right);
3664 memmove_extent_buffer(right,
3665 BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
3666 BTRFS_LEAF_DATA_OFFSET + data_end,
3667 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
3669 /* copy from the left data area */
3670 copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
3671 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3672 BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left),
3673 push_space);
3675 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3676 btrfs_item_nr_offset(0),
3677 right_nritems * sizeof(struct btrfs_item));
3679 /* copy the items from left to right */
3680 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3681 btrfs_item_nr_offset(left_nritems - push_items),
3682 push_items * sizeof(struct btrfs_item));
3684 /* update the item pointers */
3685 btrfs_init_map_token(&token, right);
3686 right_nritems += push_items;
3687 btrfs_set_header_nritems(right, right_nritems);
3688 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3689 for (i = 0; i < right_nritems; i++) {
3690 item = btrfs_item_nr(i);
3691 push_space -= btrfs_token_item_size(right, item, &token);
3692 btrfs_set_token_item_offset(right, item, push_space, &token);
3695 left_nritems -= push_items;
3696 btrfs_set_header_nritems(left, left_nritems);
3698 if (left_nritems)
3699 btrfs_mark_buffer_dirty(left);
3700 else
3701 btrfs_clean_tree_block(left);
3703 btrfs_mark_buffer_dirty(right);
3705 btrfs_item_key(right, &disk_key, 0);
3706 btrfs_set_node_key(upper, &disk_key, slot + 1);
3707 btrfs_mark_buffer_dirty(upper);
3709 /* then fixup the leaf pointer in the path */
3710 if (path->slots[0] >= left_nritems) {
3711 path->slots[0] -= left_nritems;
3712 if (btrfs_header_nritems(path->nodes[0]) == 0)
3713 btrfs_clean_tree_block(path->nodes[0]);
3714 btrfs_tree_unlock(path->nodes[0]);
3715 free_extent_buffer(path->nodes[0]);
3716 path->nodes[0] = right;
3717 path->slots[1] += 1;
3718 } else {
3719 btrfs_tree_unlock(right);
3720 free_extent_buffer(right);
3722 return 0;
3724 out_unlock:
3725 btrfs_tree_unlock(right);
3726 free_extent_buffer(right);
3727 return 1;
3731 * push some data in the path leaf to the right, trying to free up at
3732 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3734 * returns 1 if the push failed because the other node didn't have enough
3735 * room, 0 if everything worked out and < 0 if there were major errors.
3737 * this will push starting from min_slot to the end of the leaf. It won't
3738 * push any slot lower than min_slot
3740 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3741 *root, struct btrfs_path *path,
3742 int min_data_size, int data_size,
3743 int empty, u32 min_slot)
3745 struct extent_buffer *left = path->nodes[0];
3746 struct extent_buffer *right;
3747 struct extent_buffer *upper;
3748 int slot;
3749 int free_space;
3750 u32 left_nritems;
3751 int ret;
3753 if (!path->nodes[1])
3754 return 1;
3756 slot = path->slots[1];
3757 upper = path->nodes[1];
3758 if (slot >= btrfs_header_nritems(upper) - 1)
3759 return 1;
3761 btrfs_assert_tree_locked(path->nodes[1]);
3763 right = btrfs_read_node_slot(upper, slot + 1);
3765 * slot + 1 is not valid or we fail to read the right node,
3766 * no big deal, just return.
3768 if (IS_ERR(right))
3769 return 1;
3771 btrfs_tree_lock(right);
3772 btrfs_set_lock_blocking_write(right);
3774 free_space = btrfs_leaf_free_space(right);
3775 if (free_space < data_size)
3776 goto out_unlock;
3778 /* cow and double check */
3779 ret = btrfs_cow_block(trans, root, right, upper,
3780 slot + 1, &right);
3781 if (ret)
3782 goto out_unlock;
3784 free_space = btrfs_leaf_free_space(right);
3785 if (free_space < data_size)
3786 goto out_unlock;
3788 left_nritems = btrfs_header_nritems(left);
3789 if (left_nritems == 0)
3790 goto out_unlock;
3792 if (path->slots[0] == left_nritems && !empty) {
3793 /* Key greater than all keys in the leaf, right neighbor has
3794 * enough room for it and we're not emptying our leaf to delete
3795 * it, therefore use right neighbor to insert the new item and
3796 * no need to touch/dirty our left leaf. */
3797 btrfs_tree_unlock(left);
3798 free_extent_buffer(left);
3799 path->nodes[0] = right;
3800 path->slots[0] = 0;
3801 path->slots[1]++;
3802 return 0;
3805 return __push_leaf_right(path, min_data_size, empty,
3806 right, free_space, left_nritems, min_slot);
3807 out_unlock:
3808 btrfs_tree_unlock(right);
3809 free_extent_buffer(right);
3810 return 1;
3814 * push some data in the path leaf to the left, trying to free up at
3815 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3817 * max_slot can put a limit on how far into the leaf we'll push items. The
3818 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3819 * items
3821 static noinline int __push_leaf_left(struct btrfs_path *path, int data_size,
3822 int empty, struct extent_buffer *left,
3823 int free_space, u32 right_nritems,
3824 u32 max_slot)
3826 struct btrfs_fs_info *fs_info = left->fs_info;
3827 struct btrfs_disk_key disk_key;
3828 struct extent_buffer *right = path->nodes[0];
3829 int i;
3830 int push_space = 0;
3831 int push_items = 0;
3832 struct btrfs_item *item;
3833 u32 old_left_nritems;
3834 u32 nr;
3835 int ret = 0;
3836 u32 this_item_size;
3837 u32 old_left_item_size;
3838 struct btrfs_map_token token;
3840 if (empty)
3841 nr = min(right_nritems, max_slot);
3842 else
3843 nr = min(right_nritems - 1, max_slot);
3845 for (i = 0; i < nr; i++) {
3846 item = btrfs_item_nr(i);
3848 if (!empty && push_items > 0) {
3849 if (path->slots[0] < i)
3850 break;
3851 if (path->slots[0] == i) {
3852 int space = btrfs_leaf_free_space(right);
3854 if (space + push_space * 2 > free_space)
3855 break;
3859 if (path->slots[0] == i)
3860 push_space += data_size;
3862 this_item_size = btrfs_item_size(right, item);
3863 if (this_item_size + sizeof(*item) + push_space > free_space)
3864 break;
3866 push_items++;
3867 push_space += this_item_size + sizeof(*item);
3870 if (push_items == 0) {
3871 ret = 1;
3872 goto out;
3874 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
3876 /* push data from right to left */
3877 copy_extent_buffer(left, right,
3878 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3879 btrfs_item_nr_offset(0),
3880 push_items * sizeof(struct btrfs_item));
3882 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
3883 btrfs_item_offset_nr(right, push_items - 1);
3885 copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
3886 leaf_data_end(left) - push_space,
3887 BTRFS_LEAF_DATA_OFFSET +
3888 btrfs_item_offset_nr(right, push_items - 1),
3889 push_space);
3890 old_left_nritems = btrfs_header_nritems(left);
3891 BUG_ON(old_left_nritems <= 0);
3893 btrfs_init_map_token(&token, left);
3894 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3895 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3896 u32 ioff;
3898 item = btrfs_item_nr(i);
3900 ioff = btrfs_token_item_offset(left, item, &token);
3901 btrfs_set_token_item_offset(left, item,
3902 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size),
3903 &token);
3905 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3907 /* fixup right node */
3908 if (push_items > right_nritems)
3909 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3910 right_nritems);
3912 if (push_items < right_nritems) {
3913 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3914 leaf_data_end(right);
3915 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
3916 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
3917 BTRFS_LEAF_DATA_OFFSET +
3918 leaf_data_end(right), push_space);
3920 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3921 btrfs_item_nr_offset(push_items),
3922 (btrfs_header_nritems(right) - push_items) *
3923 sizeof(struct btrfs_item));
3926 btrfs_init_map_token(&token, right);
3927 right_nritems -= push_items;
3928 btrfs_set_header_nritems(right, right_nritems);
3929 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
3930 for (i = 0; i < right_nritems; i++) {
3931 item = btrfs_item_nr(i);
3933 push_space = push_space - btrfs_token_item_size(right,
3934 item, &token);
3935 btrfs_set_token_item_offset(right, item, push_space, &token);
3938 btrfs_mark_buffer_dirty(left);
3939 if (right_nritems)
3940 btrfs_mark_buffer_dirty(right);
3941 else
3942 btrfs_clean_tree_block(right);
3944 btrfs_item_key(right, &disk_key, 0);
3945 fixup_low_keys(path, &disk_key, 1);
3947 /* then fixup the leaf pointer in the path */
3948 if (path->slots[0] < push_items) {
3949 path->slots[0] += old_left_nritems;
3950 btrfs_tree_unlock(path->nodes[0]);
3951 free_extent_buffer(path->nodes[0]);
3952 path->nodes[0] = left;
3953 path->slots[1] -= 1;
3954 } else {
3955 btrfs_tree_unlock(left);
3956 free_extent_buffer(left);
3957 path->slots[0] -= push_items;
3959 BUG_ON(path->slots[0] < 0);
3960 return ret;
3961 out:
3962 btrfs_tree_unlock(left);
3963 free_extent_buffer(left);
3964 return ret;
3968 * push some data in the path leaf to the left, trying to free up at
3969 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3971 * max_slot can put a limit on how far into the leaf we'll push items. The
3972 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3973 * items
3975 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3976 *root, struct btrfs_path *path, int min_data_size,
3977 int data_size, int empty, u32 max_slot)
3979 struct extent_buffer *right = path->nodes[0];
3980 struct extent_buffer *left;
3981 int slot;
3982 int free_space;
3983 u32 right_nritems;
3984 int ret = 0;
3986 slot = path->slots[1];
3987 if (slot == 0)
3988 return 1;
3989 if (!path->nodes[1])
3990 return 1;
3992 right_nritems = btrfs_header_nritems(right);
3993 if (right_nritems == 0)
3994 return 1;
3996 btrfs_assert_tree_locked(path->nodes[1]);
3998 left = btrfs_read_node_slot(path->nodes[1], slot - 1);
4000 * slot - 1 is not valid or we fail to read the left node,
4001 * no big deal, just return.
4003 if (IS_ERR(left))
4004 return 1;
4006 btrfs_tree_lock(left);
4007 btrfs_set_lock_blocking_write(left);
4009 free_space = btrfs_leaf_free_space(left);
4010 if (free_space < data_size) {
4011 ret = 1;
4012 goto out;
4015 /* cow and double check */
4016 ret = btrfs_cow_block(trans, root, left,
4017 path->nodes[1], slot - 1, &left);
4018 if (ret) {
4019 /* we hit -ENOSPC, but it isn't fatal here */
4020 if (ret == -ENOSPC)
4021 ret = 1;
4022 goto out;
4025 free_space = btrfs_leaf_free_space(left);
4026 if (free_space < data_size) {
4027 ret = 1;
4028 goto out;
4031 return __push_leaf_left(path, min_data_size,
4032 empty, left, free_space, right_nritems,
4033 max_slot);
4034 out:
4035 btrfs_tree_unlock(left);
4036 free_extent_buffer(left);
4037 return ret;
4041 * split the path's leaf in two, making sure there is at least data_size
4042 * available for the resulting leaf level of the path.
4044 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
4045 struct btrfs_path *path,
4046 struct extent_buffer *l,
4047 struct extent_buffer *right,
4048 int slot, int mid, int nritems)
4050 struct btrfs_fs_info *fs_info = trans->fs_info;
4051 int data_copy_size;
4052 int rt_data_off;
4053 int i;
4054 struct btrfs_disk_key disk_key;
4055 struct btrfs_map_token token;
4057 nritems = nritems - mid;
4058 btrfs_set_header_nritems(right, nritems);
4059 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l);
4061 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
4062 btrfs_item_nr_offset(mid),
4063 nritems * sizeof(struct btrfs_item));
4065 copy_extent_buffer(right, l,
4066 BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
4067 data_copy_size, BTRFS_LEAF_DATA_OFFSET +
4068 leaf_data_end(l), data_copy_size);
4070 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
4072 btrfs_init_map_token(&token, right);
4073 for (i = 0; i < nritems; i++) {
4074 struct btrfs_item *item = btrfs_item_nr(i);
4075 u32 ioff;
4077 ioff = btrfs_token_item_offset(right, item, &token);
4078 btrfs_set_token_item_offset(right, item,
4079 ioff + rt_data_off, &token);
4082 btrfs_set_header_nritems(l, mid);
4083 btrfs_item_key(right, &disk_key, 0);
4084 insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
4086 btrfs_mark_buffer_dirty(right);
4087 btrfs_mark_buffer_dirty(l);
4088 BUG_ON(path->slots[0] != slot);
4090 if (mid <= slot) {
4091 btrfs_tree_unlock(path->nodes[0]);
4092 free_extent_buffer(path->nodes[0]);
4093 path->nodes[0] = right;
4094 path->slots[0] -= mid;
4095 path->slots[1] += 1;
4096 } else {
4097 btrfs_tree_unlock(right);
4098 free_extent_buffer(right);
4101 BUG_ON(path->slots[0] < 0);
4105 * double splits happen when we need to insert a big item in the middle
4106 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4107 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4108 * A B C
4110 * We avoid this by trying to push the items on either side of our target
4111 * into the adjacent leaves. If all goes well we can avoid the double split
4112 * completely.
4114 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
4115 struct btrfs_root *root,
4116 struct btrfs_path *path,
4117 int data_size)
4119 int ret;
4120 int progress = 0;
4121 int slot;
4122 u32 nritems;
4123 int space_needed = data_size;
4125 slot = path->slots[0];
4126 if (slot < btrfs_header_nritems(path->nodes[0]))
4127 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4130 * try to push all the items after our slot into the
4131 * right leaf
4133 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
4134 if (ret < 0)
4135 return ret;
4137 if (ret == 0)
4138 progress++;
4140 nritems = btrfs_header_nritems(path->nodes[0]);
4142 * our goal is to get our slot at the start or end of a leaf. If
4143 * we've done so we're done
4145 if (path->slots[0] == 0 || path->slots[0] == nritems)
4146 return 0;
4148 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4149 return 0;
4151 /* try to push all the items before our slot into the next leaf */
4152 slot = path->slots[0];
4153 space_needed = data_size;
4154 if (slot > 0)
4155 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
4156 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
4157 if (ret < 0)
4158 return ret;
4160 if (ret == 0)
4161 progress++;
4163 if (progress)
4164 return 0;
4165 return 1;
4169 * split the path's leaf in two, making sure there is at least data_size
4170 * available for the resulting leaf level of the path.
4172 * returns 0 if all went well and < 0 on failure.
4174 static noinline int split_leaf(struct btrfs_trans_handle *trans,
4175 struct btrfs_root *root,
4176 const struct btrfs_key *ins_key,
4177 struct btrfs_path *path, int data_size,
4178 int extend)
4180 struct btrfs_disk_key disk_key;
4181 struct extent_buffer *l;
4182 u32 nritems;
4183 int mid;
4184 int slot;
4185 struct extent_buffer *right;
4186 struct btrfs_fs_info *fs_info = root->fs_info;
4187 int ret = 0;
4188 int wret;
4189 int split;
4190 int num_doubles = 0;
4191 int tried_avoid_double = 0;
4193 l = path->nodes[0];
4194 slot = path->slots[0];
4195 if (extend && data_size + btrfs_item_size_nr(l, slot) +
4196 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
4197 return -EOVERFLOW;
4199 /* first try to make some room by pushing left and right */
4200 if (data_size && path->nodes[1]) {
4201 int space_needed = data_size;
4203 if (slot < btrfs_header_nritems(l))
4204 space_needed -= btrfs_leaf_free_space(l);
4206 wret = push_leaf_right(trans, root, path, space_needed,
4207 space_needed, 0, 0);
4208 if (wret < 0)
4209 return wret;
4210 if (wret) {
4211 space_needed = data_size;
4212 if (slot > 0)
4213 space_needed -= btrfs_leaf_free_space(l);
4214 wret = push_leaf_left(trans, root, path, space_needed,
4215 space_needed, 0, (u32)-1);
4216 if (wret < 0)
4217 return wret;
4219 l = path->nodes[0];
4221 /* did the pushes work? */
4222 if (btrfs_leaf_free_space(l) >= data_size)
4223 return 0;
4226 if (!path->nodes[1]) {
4227 ret = insert_new_root(trans, root, path, 1);
4228 if (ret)
4229 return ret;
4231 again:
4232 split = 1;
4233 l = path->nodes[0];
4234 slot = path->slots[0];
4235 nritems = btrfs_header_nritems(l);
4236 mid = (nritems + 1) / 2;
4238 if (mid <= slot) {
4239 if (nritems == 1 ||
4240 leaf_space_used(l, mid, nritems - mid) + data_size >
4241 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4242 if (slot >= nritems) {
4243 split = 0;
4244 } else {
4245 mid = slot;
4246 if (mid != nritems &&
4247 leaf_space_used(l, mid, nritems - mid) +
4248 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4249 if (data_size && !tried_avoid_double)
4250 goto push_for_double;
4251 split = 2;
4255 } else {
4256 if (leaf_space_used(l, 0, mid) + data_size >
4257 BTRFS_LEAF_DATA_SIZE(fs_info)) {
4258 if (!extend && data_size && slot == 0) {
4259 split = 0;
4260 } else if ((extend || !data_size) && slot == 0) {
4261 mid = 1;
4262 } else {
4263 mid = slot;
4264 if (mid != nritems &&
4265 leaf_space_used(l, mid, nritems - mid) +
4266 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
4267 if (data_size && !tried_avoid_double)
4268 goto push_for_double;
4269 split = 2;
4275 if (split == 0)
4276 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4277 else
4278 btrfs_item_key(l, &disk_key, mid);
4280 right = alloc_tree_block_no_bg_flush(trans, root, 0, &disk_key, 0,
4281 l->start, 0);
4282 if (IS_ERR(right))
4283 return PTR_ERR(right);
4285 root_add_used(root, fs_info->nodesize);
4287 if (split == 0) {
4288 if (mid <= slot) {
4289 btrfs_set_header_nritems(right, 0);
4290 insert_ptr(trans, path, &disk_key,
4291 right->start, path->slots[1] + 1, 1);
4292 btrfs_tree_unlock(path->nodes[0]);
4293 free_extent_buffer(path->nodes[0]);
4294 path->nodes[0] = right;
4295 path->slots[0] = 0;
4296 path->slots[1] += 1;
4297 } else {
4298 btrfs_set_header_nritems(right, 0);
4299 insert_ptr(trans, path, &disk_key,
4300 right->start, path->slots[1], 1);
4301 btrfs_tree_unlock(path->nodes[0]);
4302 free_extent_buffer(path->nodes[0]);
4303 path->nodes[0] = right;
4304 path->slots[0] = 0;
4305 if (path->slots[1] == 0)
4306 fixup_low_keys(path, &disk_key, 1);
4309 * We create a new leaf 'right' for the required ins_len and
4310 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4311 * the content of ins_len to 'right'.
4313 return ret;
4316 copy_for_split(trans, path, l, right, slot, mid, nritems);
4318 if (split == 2) {
4319 BUG_ON(num_doubles != 0);
4320 num_doubles++;
4321 goto again;
4324 return 0;
4326 push_for_double:
4327 push_for_double_split(trans, root, path, data_size);
4328 tried_avoid_double = 1;
4329 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
4330 return 0;
4331 goto again;
4334 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4335 struct btrfs_root *root,
4336 struct btrfs_path *path, int ins_len)
4338 struct btrfs_key key;
4339 struct extent_buffer *leaf;
4340 struct btrfs_file_extent_item *fi;
4341 u64 extent_len = 0;
4342 u32 item_size;
4343 int ret;
4345 leaf = path->nodes[0];
4346 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4348 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4349 key.type != BTRFS_EXTENT_CSUM_KEY);
4351 if (btrfs_leaf_free_space(leaf) >= ins_len)
4352 return 0;
4354 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4355 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4356 fi = btrfs_item_ptr(leaf, path->slots[0],
4357 struct btrfs_file_extent_item);
4358 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4360 btrfs_release_path(path);
4362 path->keep_locks = 1;
4363 path->search_for_split = 1;
4364 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4365 path->search_for_split = 0;
4366 if (ret > 0)
4367 ret = -EAGAIN;
4368 if (ret < 0)
4369 goto err;
4371 ret = -EAGAIN;
4372 leaf = path->nodes[0];
4373 /* if our item isn't there, return now */
4374 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4375 goto err;
4377 /* the leaf has changed, it now has room. return now */
4378 if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
4379 goto err;
4381 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4382 fi = btrfs_item_ptr(leaf, path->slots[0],
4383 struct btrfs_file_extent_item);
4384 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4385 goto err;
4388 btrfs_set_path_blocking(path);
4389 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4390 if (ret)
4391 goto err;
4393 path->keep_locks = 0;
4394 btrfs_unlock_up_safe(path, 1);
4395 return 0;
4396 err:
4397 path->keep_locks = 0;
4398 return ret;
4401 static noinline int split_item(struct btrfs_path *path,
4402 const struct btrfs_key *new_key,
4403 unsigned long split_offset)
4405 struct extent_buffer *leaf;
4406 struct btrfs_item *item;
4407 struct btrfs_item *new_item;
4408 int slot;
4409 char *buf;
4410 u32 nritems;
4411 u32 item_size;
4412 u32 orig_offset;
4413 struct btrfs_disk_key disk_key;
4415 leaf = path->nodes[0];
4416 BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item));
4418 btrfs_set_path_blocking(path);
4420 item = btrfs_item_nr(path->slots[0]);
4421 orig_offset = btrfs_item_offset(leaf, item);
4422 item_size = btrfs_item_size(leaf, item);
4424 buf = kmalloc(item_size, GFP_NOFS);
4425 if (!buf)
4426 return -ENOMEM;
4428 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4429 path->slots[0]), item_size);
4431 slot = path->slots[0] + 1;
4432 nritems = btrfs_header_nritems(leaf);
4433 if (slot != nritems) {
4434 /* shift the items */
4435 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4436 btrfs_item_nr_offset(slot),
4437 (nritems - slot) * sizeof(struct btrfs_item));
4440 btrfs_cpu_key_to_disk(&disk_key, new_key);
4441 btrfs_set_item_key(leaf, &disk_key, slot);
4443 new_item = btrfs_item_nr(slot);
4445 btrfs_set_item_offset(leaf, new_item, orig_offset);
4446 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4448 btrfs_set_item_offset(leaf, item,
4449 orig_offset + item_size - split_offset);
4450 btrfs_set_item_size(leaf, item, split_offset);
4452 btrfs_set_header_nritems(leaf, nritems + 1);
4454 /* write the data for the start of the original item */
4455 write_extent_buffer(leaf, buf,
4456 btrfs_item_ptr_offset(leaf, path->slots[0]),
4457 split_offset);
4459 /* write the data for the new item */
4460 write_extent_buffer(leaf, buf + split_offset,
4461 btrfs_item_ptr_offset(leaf, slot),
4462 item_size - split_offset);
4463 btrfs_mark_buffer_dirty(leaf);
4465 BUG_ON(btrfs_leaf_free_space(leaf) < 0);
4466 kfree(buf);
4467 return 0;
4471 * This function splits a single item into two items,
4472 * giving 'new_key' to the new item and splitting the
4473 * old one at split_offset (from the start of the item).
4475 * The path may be released by this operation. After
4476 * the split, the path is pointing to the old item. The
4477 * new item is going to be in the same node as the old one.
4479 * Note, the item being split must be smaller enough to live alone on
4480 * a tree block with room for one extra struct btrfs_item
4482 * This allows us to split the item in place, keeping a lock on the
4483 * leaf the entire time.
4485 int btrfs_split_item(struct btrfs_trans_handle *trans,
4486 struct btrfs_root *root,
4487 struct btrfs_path *path,
4488 const struct btrfs_key *new_key,
4489 unsigned long split_offset)
4491 int ret;
4492 ret = setup_leaf_for_split(trans, root, path,
4493 sizeof(struct btrfs_item));
4494 if (ret)
4495 return ret;
4497 ret = split_item(path, new_key, split_offset);
4498 return ret;
4502 * This function duplicate a item, giving 'new_key' to the new item.
4503 * It guarantees both items live in the same tree leaf and the new item
4504 * is contiguous with the original item.
4506 * This allows us to split file extent in place, keeping a lock on the
4507 * leaf the entire time.
4509 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4510 struct btrfs_root *root,
4511 struct btrfs_path *path,
4512 const struct btrfs_key *new_key)
4514 struct extent_buffer *leaf;
4515 int ret;
4516 u32 item_size;
4518 leaf = path->nodes[0];
4519 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4520 ret = setup_leaf_for_split(trans, root, path,
4521 item_size + sizeof(struct btrfs_item));
4522 if (ret)
4523 return ret;
4525 path->slots[0]++;
4526 setup_items_for_insert(root, path, new_key, &item_size,
4527 item_size, item_size +
4528 sizeof(struct btrfs_item), 1);
4529 leaf = path->nodes[0];
4530 memcpy_extent_buffer(leaf,
4531 btrfs_item_ptr_offset(leaf, path->slots[0]),
4532 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4533 item_size);
4534 return 0;
4538 * make the item pointed to by the path smaller. new_size indicates
4539 * how small to make it, and from_end tells us if we just chop bytes
4540 * off the end of the item or if we shift the item to chop bytes off
4541 * the front.
4543 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end)
4545 int slot;
4546 struct extent_buffer *leaf;
4547 struct btrfs_item *item;
4548 u32 nritems;
4549 unsigned int data_end;
4550 unsigned int old_data_start;
4551 unsigned int old_size;
4552 unsigned int size_diff;
4553 int i;
4554 struct btrfs_map_token token;
4556 leaf = path->nodes[0];
4557 slot = path->slots[0];
4559 old_size = btrfs_item_size_nr(leaf, slot);
4560 if (old_size == new_size)
4561 return;
4563 nritems = btrfs_header_nritems(leaf);
4564 data_end = leaf_data_end(leaf);
4566 old_data_start = btrfs_item_offset_nr(leaf, slot);
4568 size_diff = old_size - new_size;
4570 BUG_ON(slot < 0);
4571 BUG_ON(slot >= nritems);
4574 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4576 /* first correct the data pointers */
4577 btrfs_init_map_token(&token, leaf);
4578 for (i = slot; i < nritems; i++) {
4579 u32 ioff;
4580 item = btrfs_item_nr(i);
4582 ioff = btrfs_token_item_offset(leaf, item, &token);
4583 btrfs_set_token_item_offset(leaf, item,
4584 ioff + size_diff, &token);
4587 /* shift the data */
4588 if (from_end) {
4589 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4590 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4591 data_end, old_data_start + new_size - data_end);
4592 } else {
4593 struct btrfs_disk_key disk_key;
4594 u64 offset;
4596 btrfs_item_key(leaf, &disk_key, slot);
4598 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4599 unsigned long ptr;
4600 struct btrfs_file_extent_item *fi;
4602 fi = btrfs_item_ptr(leaf, slot,
4603 struct btrfs_file_extent_item);
4604 fi = (struct btrfs_file_extent_item *)(
4605 (unsigned long)fi - size_diff);
4607 if (btrfs_file_extent_type(leaf, fi) ==
4608 BTRFS_FILE_EXTENT_INLINE) {
4609 ptr = btrfs_item_ptr_offset(leaf, slot);
4610 memmove_extent_buffer(leaf, ptr,
4611 (unsigned long)fi,
4612 BTRFS_FILE_EXTENT_INLINE_DATA_START);
4616 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4617 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
4618 data_end, old_data_start - data_end);
4620 offset = btrfs_disk_key_offset(&disk_key);
4621 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4622 btrfs_set_item_key(leaf, &disk_key, slot);
4623 if (slot == 0)
4624 fixup_low_keys(path, &disk_key, 1);
4627 item = btrfs_item_nr(slot);
4628 btrfs_set_item_size(leaf, item, new_size);
4629 btrfs_mark_buffer_dirty(leaf);
4631 if (btrfs_leaf_free_space(leaf) < 0) {
4632 btrfs_print_leaf(leaf);
4633 BUG();
4638 * make the item pointed to by the path bigger, data_size is the added size.
4640 void btrfs_extend_item(struct btrfs_path *path, u32 data_size)
4642 int slot;
4643 struct extent_buffer *leaf;
4644 struct btrfs_item *item;
4645 u32 nritems;
4646 unsigned int data_end;
4647 unsigned int old_data;
4648 unsigned int old_size;
4649 int i;
4650 struct btrfs_map_token token;
4652 leaf = path->nodes[0];
4654 nritems = btrfs_header_nritems(leaf);
4655 data_end = leaf_data_end(leaf);
4657 if (btrfs_leaf_free_space(leaf) < data_size) {
4658 btrfs_print_leaf(leaf);
4659 BUG();
4661 slot = path->slots[0];
4662 old_data = btrfs_item_end_nr(leaf, slot);
4664 BUG_ON(slot < 0);
4665 if (slot >= nritems) {
4666 btrfs_print_leaf(leaf);
4667 btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d",
4668 slot, nritems);
4669 BUG();
4673 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4675 /* first correct the data pointers */
4676 btrfs_init_map_token(&token, leaf);
4677 for (i = slot; i < nritems; i++) {
4678 u32 ioff;
4679 item = btrfs_item_nr(i);
4681 ioff = btrfs_token_item_offset(leaf, item, &token);
4682 btrfs_set_token_item_offset(leaf, item,
4683 ioff - data_size, &token);
4686 /* shift the data */
4687 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4688 data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
4689 data_end, old_data - data_end);
4691 data_end = old_data;
4692 old_size = btrfs_item_size_nr(leaf, slot);
4693 item = btrfs_item_nr(slot);
4694 btrfs_set_item_size(leaf, item, old_size + data_size);
4695 btrfs_mark_buffer_dirty(leaf);
4697 if (btrfs_leaf_free_space(leaf) < 0) {
4698 btrfs_print_leaf(leaf);
4699 BUG();
4704 * this is a helper for btrfs_insert_empty_items, the main goal here is
4705 * to save stack depth by doing the bulk of the work in a function
4706 * that doesn't call btrfs_search_slot
4708 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4709 const struct btrfs_key *cpu_key, u32 *data_size,
4710 u32 total_data, u32 total_size, int nr)
4712 struct btrfs_fs_info *fs_info = root->fs_info;
4713 struct btrfs_item *item;
4714 int i;
4715 u32 nritems;
4716 unsigned int data_end;
4717 struct btrfs_disk_key disk_key;
4718 struct extent_buffer *leaf;
4719 int slot;
4720 struct btrfs_map_token token;
4722 if (path->slots[0] == 0) {
4723 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4724 fixup_low_keys(path, &disk_key, 1);
4726 btrfs_unlock_up_safe(path, 1);
4728 leaf = path->nodes[0];
4729 slot = path->slots[0];
4731 nritems = btrfs_header_nritems(leaf);
4732 data_end = leaf_data_end(leaf);
4734 if (btrfs_leaf_free_space(leaf) < total_size) {
4735 btrfs_print_leaf(leaf);
4736 btrfs_crit(fs_info, "not enough freespace need %u have %d",
4737 total_size, btrfs_leaf_free_space(leaf));
4738 BUG();
4741 btrfs_init_map_token(&token, leaf);
4742 if (slot != nritems) {
4743 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4745 if (old_data < data_end) {
4746 btrfs_print_leaf(leaf);
4747 btrfs_crit(fs_info, "slot %d old_data %d data_end %d",
4748 slot, old_data, data_end);
4749 BUG();
4752 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4754 /* first correct the data pointers */
4755 for (i = slot; i < nritems; i++) {
4756 u32 ioff;
4758 item = btrfs_item_nr(i);
4759 ioff = btrfs_token_item_offset(leaf, item, &token);
4760 btrfs_set_token_item_offset(leaf, item,
4761 ioff - total_data, &token);
4763 /* shift the items */
4764 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4765 btrfs_item_nr_offset(slot),
4766 (nritems - slot) * sizeof(struct btrfs_item));
4768 /* shift the data */
4769 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4770 data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
4771 data_end, old_data - data_end);
4772 data_end = old_data;
4775 /* setup the item for the new data */
4776 for (i = 0; i < nr; i++) {
4777 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4778 btrfs_set_item_key(leaf, &disk_key, slot + i);
4779 item = btrfs_item_nr(slot + i);
4780 btrfs_set_token_item_offset(leaf, item,
4781 data_end - data_size[i], &token);
4782 data_end -= data_size[i];
4783 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4786 btrfs_set_header_nritems(leaf, nritems + nr);
4787 btrfs_mark_buffer_dirty(leaf);
4789 if (btrfs_leaf_free_space(leaf) < 0) {
4790 btrfs_print_leaf(leaf);
4791 BUG();
4796 * Given a key and some data, insert items into the tree.
4797 * This does all the path init required, making room in the tree if needed.
4799 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4800 struct btrfs_root *root,
4801 struct btrfs_path *path,
4802 const struct btrfs_key *cpu_key, u32 *data_size,
4803 int nr)
4805 int ret = 0;
4806 int slot;
4807 int i;
4808 u32 total_size = 0;
4809 u32 total_data = 0;
4811 for (i = 0; i < nr; i++)
4812 total_data += data_size[i];
4814 total_size = total_data + (nr * sizeof(struct btrfs_item));
4815 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4816 if (ret == 0)
4817 return -EEXIST;
4818 if (ret < 0)
4819 return ret;
4821 slot = path->slots[0];
4822 BUG_ON(slot < 0);
4824 setup_items_for_insert(root, path, cpu_key, data_size,
4825 total_data, total_size, nr);
4826 return 0;
4830 * Given a key and some data, insert an item into the tree.
4831 * This does all the path init required, making room in the tree if needed.
4833 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4834 const struct btrfs_key *cpu_key, void *data,
4835 u32 data_size)
4837 int ret = 0;
4838 struct btrfs_path *path;
4839 struct extent_buffer *leaf;
4840 unsigned long ptr;
4842 path = btrfs_alloc_path();
4843 if (!path)
4844 return -ENOMEM;
4845 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4846 if (!ret) {
4847 leaf = path->nodes[0];
4848 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4849 write_extent_buffer(leaf, data, ptr, data_size);
4850 btrfs_mark_buffer_dirty(leaf);
4852 btrfs_free_path(path);
4853 return ret;
4857 * delete the pointer from a given node.
4859 * the tree should have been previously balanced so the deletion does not
4860 * empty a node.
4862 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4863 int level, int slot)
4865 struct extent_buffer *parent = path->nodes[level];
4866 u32 nritems;
4867 int ret;
4869 nritems = btrfs_header_nritems(parent);
4870 if (slot != nritems - 1) {
4871 if (level) {
4872 ret = tree_mod_log_insert_move(parent, slot, slot + 1,
4873 nritems - slot - 1);
4874 BUG_ON(ret < 0);
4876 memmove_extent_buffer(parent,
4877 btrfs_node_key_ptr_offset(slot),
4878 btrfs_node_key_ptr_offset(slot + 1),
4879 sizeof(struct btrfs_key_ptr) *
4880 (nritems - slot - 1));
4881 } else if (level) {
4882 ret = tree_mod_log_insert_key(parent, slot, MOD_LOG_KEY_REMOVE,
4883 GFP_NOFS);
4884 BUG_ON(ret < 0);
4887 nritems--;
4888 btrfs_set_header_nritems(parent, nritems);
4889 if (nritems == 0 && parent == root->node) {
4890 BUG_ON(btrfs_header_level(root->node) != 1);
4891 /* just turn the root into a leaf and break */
4892 btrfs_set_header_level(root->node, 0);
4893 } else if (slot == 0) {
4894 struct btrfs_disk_key disk_key;
4896 btrfs_node_key(parent, &disk_key, 0);
4897 fixup_low_keys(path, &disk_key, level + 1);
4899 btrfs_mark_buffer_dirty(parent);
4903 * a helper function to delete the leaf pointed to by path->slots[1] and
4904 * path->nodes[1].
4906 * This deletes the pointer in path->nodes[1] and frees the leaf
4907 * block extent. zero is returned if it all worked out, < 0 otherwise.
4909 * The path must have already been setup for deleting the leaf, including
4910 * all the proper balancing. path->nodes[1] must be locked.
4912 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4913 struct btrfs_root *root,
4914 struct btrfs_path *path,
4915 struct extent_buffer *leaf)
4917 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4918 del_ptr(root, path, 1, path->slots[1]);
4921 * btrfs_free_extent is expensive, we want to make sure we
4922 * aren't holding any locks when we call it
4924 btrfs_unlock_up_safe(path, 0);
4926 root_sub_used(root, leaf->len);
4928 atomic_inc(&leaf->refs);
4929 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4930 free_extent_buffer_stale(leaf);
4933 * delete the item at the leaf level in path. If that empties
4934 * the leaf, remove it from the tree
4936 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4937 struct btrfs_path *path, int slot, int nr)
4939 struct btrfs_fs_info *fs_info = root->fs_info;
4940 struct extent_buffer *leaf;
4941 struct btrfs_item *item;
4942 u32 last_off;
4943 u32 dsize = 0;
4944 int ret = 0;
4945 int wret;
4946 int i;
4947 u32 nritems;
4949 leaf = path->nodes[0];
4950 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4952 for (i = 0; i < nr; i++)
4953 dsize += btrfs_item_size_nr(leaf, slot + i);
4955 nritems = btrfs_header_nritems(leaf);
4957 if (slot + nr != nritems) {
4958 int data_end = leaf_data_end(leaf);
4959 struct btrfs_map_token token;
4961 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4962 data_end + dsize,
4963 BTRFS_LEAF_DATA_OFFSET + data_end,
4964 last_off - data_end);
4966 btrfs_init_map_token(&token, leaf);
4967 for (i = slot + nr; i < nritems; i++) {
4968 u32 ioff;
4970 item = btrfs_item_nr(i);
4971 ioff = btrfs_token_item_offset(leaf, item, &token);
4972 btrfs_set_token_item_offset(leaf, item,
4973 ioff + dsize, &token);
4976 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4977 btrfs_item_nr_offset(slot + nr),
4978 sizeof(struct btrfs_item) *
4979 (nritems - slot - nr));
4981 btrfs_set_header_nritems(leaf, nritems - nr);
4982 nritems -= nr;
4984 /* delete the leaf if we've emptied it */
4985 if (nritems == 0) {
4986 if (leaf == root->node) {
4987 btrfs_set_header_level(leaf, 0);
4988 } else {
4989 btrfs_set_path_blocking(path);
4990 btrfs_clean_tree_block(leaf);
4991 btrfs_del_leaf(trans, root, path, leaf);
4993 } else {
4994 int used = leaf_space_used(leaf, 0, nritems);
4995 if (slot == 0) {
4996 struct btrfs_disk_key disk_key;
4998 btrfs_item_key(leaf, &disk_key, 0);
4999 fixup_low_keys(path, &disk_key, 1);
5002 /* delete the leaf if it is mostly empty */
5003 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
5004 /* push_leaf_left fixes the path.
5005 * make sure the path still points to our leaf
5006 * for possible call to del_ptr below
5008 slot = path->slots[1];
5009 atomic_inc(&leaf->refs);
5011 btrfs_set_path_blocking(path);
5012 wret = push_leaf_left(trans, root, path, 1, 1,
5013 1, (u32)-1);
5014 if (wret < 0 && wret != -ENOSPC)
5015 ret = wret;
5017 if (path->nodes[0] == leaf &&
5018 btrfs_header_nritems(leaf)) {
5019 wret = push_leaf_right(trans, root, path, 1,
5020 1, 1, 0);
5021 if (wret < 0 && wret != -ENOSPC)
5022 ret = wret;
5025 if (btrfs_header_nritems(leaf) == 0) {
5026 path->slots[1] = slot;
5027 btrfs_del_leaf(trans, root, path, leaf);
5028 free_extent_buffer(leaf);
5029 ret = 0;
5030 } else {
5031 /* if we're still in the path, make sure
5032 * we're dirty. Otherwise, one of the
5033 * push_leaf functions must have already
5034 * dirtied this buffer
5036 if (path->nodes[0] == leaf)
5037 btrfs_mark_buffer_dirty(leaf);
5038 free_extent_buffer(leaf);
5040 } else {
5041 btrfs_mark_buffer_dirty(leaf);
5044 return ret;
5048 * search the tree again to find a leaf with lesser keys
5049 * returns 0 if it found something or 1 if there are no lesser leaves.
5050 * returns < 0 on io errors.
5052 * This may release the path, and so you may lose any locks held at the
5053 * time you call it.
5055 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
5057 struct btrfs_key key;
5058 struct btrfs_disk_key found_key;
5059 int ret;
5061 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
5063 if (key.offset > 0) {
5064 key.offset--;
5065 } else if (key.type > 0) {
5066 key.type--;
5067 key.offset = (u64)-1;
5068 } else if (key.objectid > 0) {
5069 key.objectid--;
5070 key.type = (u8)-1;
5071 key.offset = (u64)-1;
5072 } else {
5073 return 1;
5076 btrfs_release_path(path);
5077 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5078 if (ret < 0)
5079 return ret;
5080 btrfs_item_key(path->nodes[0], &found_key, 0);
5081 ret = comp_keys(&found_key, &key);
5083 * We might have had an item with the previous key in the tree right
5084 * before we released our path. And after we released our path, that
5085 * item might have been pushed to the first slot (0) of the leaf we
5086 * were holding due to a tree balance. Alternatively, an item with the
5087 * previous key can exist as the only element of a leaf (big fat item).
5088 * Therefore account for these 2 cases, so that our callers (like
5089 * btrfs_previous_item) don't miss an existing item with a key matching
5090 * the previous key we computed above.
5092 if (ret <= 0)
5093 return 0;
5094 return 1;
5098 * A helper function to walk down the tree starting at min_key, and looking
5099 * for nodes or leaves that are have a minimum transaction id.
5100 * This is used by the btree defrag code, and tree logging
5102 * This does not cow, but it does stuff the starting key it finds back
5103 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5104 * key and get a writable path.
5106 * This honors path->lowest_level to prevent descent past a given level
5107 * of the tree.
5109 * min_trans indicates the oldest transaction that you are interested
5110 * in walking through. Any nodes or leaves older than min_trans are
5111 * skipped over (without reading them).
5113 * returns zero if something useful was found, < 0 on error and 1 if there
5114 * was nothing in the tree that matched the search criteria.
5116 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
5117 struct btrfs_path *path,
5118 u64 min_trans)
5120 struct extent_buffer *cur;
5121 struct btrfs_key found_key;
5122 int slot;
5123 int sret;
5124 u32 nritems;
5125 int level;
5126 int ret = 1;
5127 int keep_locks = path->keep_locks;
5129 path->keep_locks = 1;
5130 again:
5131 cur = btrfs_read_lock_root_node(root);
5132 level = btrfs_header_level(cur);
5133 WARN_ON(path->nodes[level]);
5134 path->nodes[level] = cur;
5135 path->locks[level] = BTRFS_READ_LOCK;
5137 if (btrfs_header_generation(cur) < min_trans) {
5138 ret = 1;
5139 goto out;
5141 while (1) {
5142 nritems = btrfs_header_nritems(cur);
5143 level = btrfs_header_level(cur);
5144 sret = btrfs_bin_search(cur, min_key, level, &slot);
5145 if (sret < 0) {
5146 ret = sret;
5147 goto out;
5150 /* at the lowest level, we're done, setup the path and exit */
5151 if (level == path->lowest_level) {
5152 if (slot >= nritems)
5153 goto find_next_key;
5154 ret = 0;
5155 path->slots[level] = slot;
5156 btrfs_item_key_to_cpu(cur, &found_key, slot);
5157 goto out;
5159 if (sret && slot > 0)
5160 slot--;
5162 * check this node pointer against the min_trans parameters.
5163 * If it is too old, old, skip to the next one.
5165 while (slot < nritems) {
5166 u64 gen;
5168 gen = btrfs_node_ptr_generation(cur, slot);
5169 if (gen < min_trans) {
5170 slot++;
5171 continue;
5173 break;
5175 find_next_key:
5177 * we didn't find a candidate key in this node, walk forward
5178 * and find another one
5180 if (slot >= nritems) {
5181 path->slots[level] = slot;
5182 btrfs_set_path_blocking(path);
5183 sret = btrfs_find_next_key(root, path, min_key, level,
5184 min_trans);
5185 if (sret == 0) {
5186 btrfs_release_path(path);
5187 goto again;
5188 } else {
5189 goto out;
5192 /* save our key for returning back */
5193 btrfs_node_key_to_cpu(cur, &found_key, slot);
5194 path->slots[level] = slot;
5195 if (level == path->lowest_level) {
5196 ret = 0;
5197 goto out;
5199 btrfs_set_path_blocking(path);
5200 cur = btrfs_read_node_slot(cur, slot);
5201 if (IS_ERR(cur)) {
5202 ret = PTR_ERR(cur);
5203 goto out;
5206 btrfs_tree_read_lock(cur);
5208 path->locks[level - 1] = BTRFS_READ_LOCK;
5209 path->nodes[level - 1] = cur;
5210 unlock_up(path, level, 1, 0, NULL);
5212 out:
5213 path->keep_locks = keep_locks;
5214 if (ret == 0) {
5215 btrfs_unlock_up_safe(path, path->lowest_level + 1);
5216 btrfs_set_path_blocking(path);
5217 memcpy(min_key, &found_key, sizeof(found_key));
5219 return ret;
5223 * this is similar to btrfs_next_leaf, but does not try to preserve
5224 * and fixup the path. It looks for and returns the next key in the
5225 * tree based on the current path and the min_trans parameters.
5227 * 0 is returned if another key is found, < 0 if there are any errors
5228 * and 1 is returned if there are no higher keys in the tree
5230 * path->keep_locks should be set to 1 on the search made before
5231 * calling this function.
5233 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5234 struct btrfs_key *key, int level, u64 min_trans)
5236 int slot;
5237 struct extent_buffer *c;
5239 WARN_ON(!path->keep_locks && !path->skip_locking);
5240 while (level < BTRFS_MAX_LEVEL) {
5241 if (!path->nodes[level])
5242 return 1;
5244 slot = path->slots[level] + 1;
5245 c = path->nodes[level];
5246 next:
5247 if (slot >= btrfs_header_nritems(c)) {
5248 int ret;
5249 int orig_lowest;
5250 struct btrfs_key cur_key;
5251 if (level + 1 >= BTRFS_MAX_LEVEL ||
5252 !path->nodes[level + 1])
5253 return 1;
5255 if (path->locks[level + 1] || path->skip_locking) {
5256 level++;
5257 continue;
5260 slot = btrfs_header_nritems(c) - 1;
5261 if (level == 0)
5262 btrfs_item_key_to_cpu(c, &cur_key, slot);
5263 else
5264 btrfs_node_key_to_cpu(c, &cur_key, slot);
5266 orig_lowest = path->lowest_level;
5267 btrfs_release_path(path);
5268 path->lowest_level = level;
5269 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5270 0, 0);
5271 path->lowest_level = orig_lowest;
5272 if (ret < 0)
5273 return ret;
5275 c = path->nodes[level];
5276 slot = path->slots[level];
5277 if (ret == 0)
5278 slot++;
5279 goto next;
5282 if (level == 0)
5283 btrfs_item_key_to_cpu(c, key, slot);
5284 else {
5285 u64 gen = btrfs_node_ptr_generation(c, slot);
5287 if (gen < min_trans) {
5288 slot++;
5289 goto next;
5291 btrfs_node_key_to_cpu(c, key, slot);
5293 return 0;
5295 return 1;
5299 * search the tree again to find a leaf with greater keys
5300 * returns 0 if it found something or 1 if there are no greater leaves.
5301 * returns < 0 on io errors.
5303 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5305 return btrfs_next_old_leaf(root, path, 0);
5308 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5309 u64 time_seq)
5311 int slot;
5312 int level;
5313 struct extent_buffer *c;
5314 struct extent_buffer *next;
5315 struct btrfs_key key;
5316 u32 nritems;
5317 int ret;
5318 int old_spinning = path->leave_spinning;
5319 int next_rw_lock = 0;
5321 nritems = btrfs_header_nritems(path->nodes[0]);
5322 if (nritems == 0)
5323 return 1;
5325 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5326 again:
5327 level = 1;
5328 next = NULL;
5329 next_rw_lock = 0;
5330 btrfs_release_path(path);
5332 path->keep_locks = 1;
5333 path->leave_spinning = 1;
5335 if (time_seq)
5336 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5337 else
5338 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5339 path->keep_locks = 0;
5341 if (ret < 0)
5342 return ret;
5344 nritems = btrfs_header_nritems(path->nodes[0]);
5346 * by releasing the path above we dropped all our locks. A balance
5347 * could have added more items next to the key that used to be
5348 * at the very end of the block. So, check again here and
5349 * advance the path if there are now more items available.
5351 if (nritems > 0 && path->slots[0] < nritems - 1) {
5352 if (ret == 0)
5353 path->slots[0]++;
5354 ret = 0;
5355 goto done;
5358 * So the above check misses one case:
5359 * - after releasing the path above, someone has removed the item that
5360 * used to be at the very end of the block, and balance between leafs
5361 * gets another one with bigger key.offset to replace it.
5363 * This one should be returned as well, or we can get leaf corruption
5364 * later(esp. in __btrfs_drop_extents()).
5366 * And a bit more explanation about this check,
5367 * with ret > 0, the key isn't found, the path points to the slot
5368 * where it should be inserted, so the path->slots[0] item must be the
5369 * bigger one.
5371 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
5372 ret = 0;
5373 goto done;
5376 while (level < BTRFS_MAX_LEVEL) {
5377 if (!path->nodes[level]) {
5378 ret = 1;
5379 goto done;
5382 slot = path->slots[level] + 1;
5383 c = path->nodes[level];
5384 if (slot >= btrfs_header_nritems(c)) {
5385 level++;
5386 if (level == BTRFS_MAX_LEVEL) {
5387 ret = 1;
5388 goto done;
5390 continue;
5393 if (next) {
5394 btrfs_tree_unlock_rw(next, next_rw_lock);
5395 free_extent_buffer(next);
5398 next = c;
5399 next_rw_lock = path->locks[level];
5400 ret = read_block_for_search(root, path, &next, level,
5401 slot, &key);
5402 if (ret == -EAGAIN)
5403 goto again;
5405 if (ret < 0) {
5406 btrfs_release_path(path);
5407 goto done;
5410 if (!path->skip_locking) {
5411 ret = btrfs_try_tree_read_lock(next);
5412 if (!ret && time_seq) {
5414 * If we don't get the lock, we may be racing
5415 * with push_leaf_left, holding that lock while
5416 * itself waiting for the leaf we've currently
5417 * locked. To solve this situation, we give up
5418 * on our lock and cycle.
5420 free_extent_buffer(next);
5421 btrfs_release_path(path);
5422 cond_resched();
5423 goto again;
5425 if (!ret) {
5426 btrfs_set_path_blocking(path);
5427 btrfs_tree_read_lock(next);
5429 next_rw_lock = BTRFS_READ_LOCK;
5431 break;
5433 path->slots[level] = slot;
5434 while (1) {
5435 level--;
5436 c = path->nodes[level];
5437 if (path->locks[level])
5438 btrfs_tree_unlock_rw(c, path->locks[level]);
5440 free_extent_buffer(c);
5441 path->nodes[level] = next;
5442 path->slots[level] = 0;
5443 if (!path->skip_locking)
5444 path->locks[level] = next_rw_lock;
5445 if (!level)
5446 break;
5448 ret = read_block_for_search(root, path, &next, level,
5449 0, &key);
5450 if (ret == -EAGAIN)
5451 goto again;
5453 if (ret < 0) {
5454 btrfs_release_path(path);
5455 goto done;
5458 if (!path->skip_locking) {
5459 ret = btrfs_try_tree_read_lock(next);
5460 if (!ret) {
5461 btrfs_set_path_blocking(path);
5462 btrfs_tree_read_lock(next);
5464 next_rw_lock = BTRFS_READ_LOCK;
5467 ret = 0;
5468 done:
5469 unlock_up(path, 0, 1, 0, NULL);
5470 path->leave_spinning = old_spinning;
5471 if (!old_spinning)
5472 btrfs_set_path_blocking(path);
5474 return ret;
5478 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5479 * searching until it gets past min_objectid or finds an item of 'type'
5481 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5483 int btrfs_previous_item(struct btrfs_root *root,
5484 struct btrfs_path *path, u64 min_objectid,
5485 int type)
5487 struct btrfs_key found_key;
5488 struct extent_buffer *leaf;
5489 u32 nritems;
5490 int ret;
5492 while (1) {
5493 if (path->slots[0] == 0) {
5494 btrfs_set_path_blocking(path);
5495 ret = btrfs_prev_leaf(root, path);
5496 if (ret != 0)
5497 return ret;
5498 } else {
5499 path->slots[0]--;
5501 leaf = path->nodes[0];
5502 nritems = btrfs_header_nritems(leaf);
5503 if (nritems == 0)
5504 return 1;
5505 if (path->slots[0] == nritems)
5506 path->slots[0]--;
5508 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5509 if (found_key.objectid < min_objectid)
5510 break;
5511 if (found_key.type == type)
5512 return 0;
5513 if (found_key.objectid == min_objectid &&
5514 found_key.type < type)
5515 break;
5517 return 1;
5521 * search in extent tree to find a previous Metadata/Data extent item with
5522 * min objecitd.
5524 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5526 int btrfs_previous_extent_item(struct btrfs_root *root,
5527 struct btrfs_path *path, u64 min_objectid)
5529 struct btrfs_key found_key;
5530 struct extent_buffer *leaf;
5531 u32 nritems;
5532 int ret;
5534 while (1) {
5535 if (path->slots[0] == 0) {
5536 btrfs_set_path_blocking(path);
5537 ret = btrfs_prev_leaf(root, path);
5538 if (ret != 0)
5539 return ret;
5540 } else {
5541 path->slots[0]--;
5543 leaf = path->nodes[0];
5544 nritems = btrfs_header_nritems(leaf);
5545 if (nritems == 0)
5546 return 1;
5547 if (path->slots[0] == nritems)
5548 path->slots[0]--;
5550 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5551 if (found_key.objectid < min_objectid)
5552 break;
5553 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5554 found_key.type == BTRFS_METADATA_ITEM_KEY)
5555 return 0;
5556 if (found_key.objectid == min_objectid &&
5557 found_key.type < BTRFS_EXTENT_ITEM_KEY)
5558 break;
5560 return 1;