1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007,2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/rbtree.h>
12 #include "transaction.h"
13 #include "print-tree.h"
17 #include "tree-mod-log.h"
19 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
20 *root, struct btrfs_path *path, int level);
21 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root,
22 const struct btrfs_key *ins_key, struct btrfs_path *path,
23 int data_size, int extend);
24 static int push_node_left(struct btrfs_trans_handle *trans,
25 struct extent_buffer *dst,
26 struct extent_buffer *src, int empty);
27 static int balance_node_right(struct btrfs_trans_handle *trans,
28 struct extent_buffer *dst_buf,
29 struct extent_buffer *src_buf);
30 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
33 static const struct btrfs_csums {
36 const char driver[12];
38 [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" },
39 [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" },
40 [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" },
41 [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b",
42 .driver = "blake2b-256" },
45 int btrfs_super_csum_size(const struct btrfs_super_block *s)
47 u16 t = btrfs_super_csum_type(s);
49 * csum type is validated at mount time
51 return btrfs_csums[t].size;
54 const char *btrfs_super_csum_name(u16 csum_type)
56 /* csum type is validated at mount time */
57 return btrfs_csums[csum_type].name;
61 * Return driver name if defined, otherwise the name that's also a valid driver
64 const char *btrfs_super_csum_driver(u16 csum_type)
66 /* csum type is validated at mount time */
67 return btrfs_csums[csum_type].driver[0] ?
68 btrfs_csums[csum_type].driver :
69 btrfs_csums[csum_type].name;
72 size_t __attribute_const__ btrfs_get_num_csums(void)
74 return ARRAY_SIZE(btrfs_csums);
77 struct btrfs_path *btrfs_alloc_path(void)
79 return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
82 /* this also releases the path */
83 void btrfs_free_path(struct btrfs_path *p)
87 btrfs_release_path(p);
88 kmem_cache_free(btrfs_path_cachep, p);
92 * path release drops references on the extent buffers in the path
93 * and it drops any locks held by this path
95 * It is safe to call this on paths that no locks or extent buffers held.
97 noinline void btrfs_release_path(struct btrfs_path *p)
101 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
106 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
109 free_extent_buffer(p->nodes[i]);
115 * safely gets a reference on the root node of a tree. A lock
116 * is not taken, so a concurrent writer may put a different node
117 * at the root of the tree. See btrfs_lock_root_node for the
120 * The extent buffer returned by this has a reference taken, so
121 * it won't disappear. It may stop being the root of the tree
122 * at any time because there are no locks held.
124 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
126 struct extent_buffer *eb;
130 eb = rcu_dereference(root->node);
133 * RCU really hurts here, we could free up the root node because
134 * it was COWed but we may not get the new root node yet so do
135 * the inc_not_zero dance and if it doesn't work then
136 * synchronize_rcu and try again.
138 if (atomic_inc_not_zero(&eb->refs)) {
149 * Cowonly root (not-shareable trees, everything not subvolume or reloc roots),
150 * just get put onto a simple dirty list. Transaction walks this list to make
151 * sure they get properly updated on disk.
153 static void add_root_to_dirty_list(struct btrfs_root *root)
155 struct btrfs_fs_info *fs_info = root->fs_info;
157 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) ||
158 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state))
161 spin_lock(&fs_info->trans_lock);
162 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) {
163 /* Want the extent tree to be the last on the list */
164 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID)
165 list_move_tail(&root->dirty_list,
166 &fs_info->dirty_cowonly_roots);
168 list_move(&root->dirty_list,
169 &fs_info->dirty_cowonly_roots);
171 spin_unlock(&fs_info->trans_lock);
175 * used by snapshot creation to make a copy of a root for a tree with
176 * a given objectid. The buffer with the new root node is returned in
177 * cow_ret, and this func returns zero on success or a negative error code.
179 int btrfs_copy_root(struct btrfs_trans_handle *trans,
180 struct btrfs_root *root,
181 struct extent_buffer *buf,
182 struct extent_buffer **cow_ret, u64 new_root_objectid)
184 struct btrfs_fs_info *fs_info = root->fs_info;
185 struct extent_buffer *cow;
188 struct btrfs_disk_key disk_key;
190 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
191 trans->transid != fs_info->running_transaction->transid);
192 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
193 trans->transid != root->last_trans);
195 level = btrfs_header_level(buf);
197 btrfs_item_key(buf, &disk_key, 0);
199 btrfs_node_key(buf, &disk_key, 0);
201 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid,
202 &disk_key, level, buf->start, 0,
203 BTRFS_NESTING_NEW_ROOT);
207 copy_extent_buffer_full(cow, buf);
208 btrfs_set_header_bytenr(cow, cow->start);
209 btrfs_set_header_generation(cow, trans->transid);
210 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
211 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
212 BTRFS_HEADER_FLAG_RELOC);
213 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
214 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
216 btrfs_set_header_owner(cow, new_root_objectid);
218 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
220 WARN_ON(btrfs_header_generation(buf) > trans->transid);
221 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
222 ret = btrfs_inc_ref(trans, root, cow, 1);
224 ret = btrfs_inc_ref(trans, root, cow, 0);
226 btrfs_tree_unlock(cow);
227 free_extent_buffer(cow);
228 btrfs_abort_transaction(trans, ret);
232 btrfs_mark_buffer_dirty(cow);
238 * check if the tree block can be shared by multiple trees
240 int btrfs_block_can_be_shared(struct btrfs_root *root,
241 struct extent_buffer *buf)
244 * Tree blocks not in shareable trees and tree roots are never shared.
245 * If a block was allocated after the last snapshot and the block was
246 * not allocated by tree relocation, we know the block is not shared.
248 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
249 buf != root->node && buf != root->commit_root &&
250 (btrfs_header_generation(buf) <=
251 btrfs_root_last_snapshot(&root->root_item) ||
252 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
258 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
259 struct btrfs_root *root,
260 struct extent_buffer *buf,
261 struct extent_buffer *cow,
264 struct btrfs_fs_info *fs_info = root->fs_info;
272 * Backrefs update rules:
274 * Always use full backrefs for extent pointers in tree block
275 * allocated by tree relocation.
277 * If a shared tree block is no longer referenced by its owner
278 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
279 * use full backrefs for extent pointers in tree block.
281 * If a tree block is been relocating
282 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
283 * use full backrefs for extent pointers in tree block.
284 * The reason for this is some operations (such as drop tree)
285 * are only allowed for blocks use full backrefs.
288 if (btrfs_block_can_be_shared(root, buf)) {
289 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start,
290 btrfs_header_level(buf), 1,
296 btrfs_handle_fs_error(fs_info, ret, NULL);
301 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
302 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
303 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
308 owner = btrfs_header_owner(buf);
309 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
310 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
313 if ((owner == root->root_key.objectid ||
314 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
315 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
316 ret = btrfs_inc_ref(trans, root, buf, 1);
320 if (root->root_key.objectid ==
321 BTRFS_TREE_RELOC_OBJECTID) {
322 ret = btrfs_dec_ref(trans, root, buf, 0);
325 ret = btrfs_inc_ref(trans, root, cow, 1);
329 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
332 if (root->root_key.objectid ==
333 BTRFS_TREE_RELOC_OBJECTID)
334 ret = btrfs_inc_ref(trans, root, cow, 1);
336 ret = btrfs_inc_ref(trans, root, cow, 0);
340 if (new_flags != 0) {
341 int level = btrfs_header_level(buf);
343 ret = btrfs_set_disk_extent_flags(trans, buf,
344 new_flags, level, 0);
349 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
350 if (root->root_key.objectid ==
351 BTRFS_TREE_RELOC_OBJECTID)
352 ret = btrfs_inc_ref(trans, root, cow, 1);
354 ret = btrfs_inc_ref(trans, root, cow, 0);
357 ret = btrfs_dec_ref(trans, root, buf, 1);
361 btrfs_clean_tree_block(buf);
368 * does the dirty work in cow of a single block. The parent block (if
369 * supplied) is updated to point to the new cow copy. The new buffer is marked
370 * dirty and returned locked. If you modify the block it needs to be marked
373 * search_start -- an allocation hint for the new block
375 * empty_size -- a hint that you plan on doing more cow. This is the size in
376 * bytes the allocator should try to find free next to the block it returns.
377 * This is just a hint and may be ignored by the allocator.
379 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
380 struct btrfs_root *root,
381 struct extent_buffer *buf,
382 struct extent_buffer *parent, int parent_slot,
383 struct extent_buffer **cow_ret,
384 u64 search_start, u64 empty_size,
385 enum btrfs_lock_nesting nest)
387 struct btrfs_fs_info *fs_info = root->fs_info;
388 struct btrfs_disk_key disk_key;
389 struct extent_buffer *cow;
393 u64 parent_start = 0;
398 btrfs_assert_tree_locked(buf);
400 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
401 trans->transid != fs_info->running_transaction->transid);
402 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) &&
403 trans->transid != root->last_trans);
405 level = btrfs_header_level(buf);
408 btrfs_item_key(buf, &disk_key, 0);
410 btrfs_node_key(buf, &disk_key, 0);
412 if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent)
413 parent_start = parent->start;
415 cow = btrfs_alloc_tree_block(trans, root, parent_start,
416 root->root_key.objectid, &disk_key, level,
417 search_start, empty_size, nest);
421 /* cow is set to blocking by btrfs_init_new_buffer */
423 copy_extent_buffer_full(cow, buf);
424 btrfs_set_header_bytenr(cow, cow->start);
425 btrfs_set_header_generation(cow, trans->transid);
426 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
427 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
428 BTRFS_HEADER_FLAG_RELOC);
429 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
430 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
432 btrfs_set_header_owner(cow, root->root_key.objectid);
434 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid);
436 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
438 btrfs_tree_unlock(cow);
439 free_extent_buffer(cow);
440 btrfs_abort_transaction(trans, ret);
444 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
445 ret = btrfs_reloc_cow_block(trans, root, buf, cow);
447 btrfs_tree_unlock(cow);
448 free_extent_buffer(cow);
449 btrfs_abort_transaction(trans, ret);
454 if (buf == root->node) {
455 WARN_ON(parent && parent != buf);
456 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
457 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
458 parent_start = buf->start;
460 atomic_inc(&cow->refs);
461 ret = btrfs_tree_mod_log_insert_root(root->node, cow, true);
463 rcu_assign_pointer(root->node, cow);
465 btrfs_free_tree_block(trans, root, buf, parent_start,
467 free_extent_buffer(buf);
468 add_root_to_dirty_list(root);
470 WARN_ON(trans->transid != btrfs_header_generation(parent));
471 btrfs_tree_mod_log_insert_key(parent, parent_slot,
472 BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS);
473 btrfs_set_node_blockptr(parent, parent_slot,
475 btrfs_set_node_ptr_generation(parent, parent_slot,
477 btrfs_mark_buffer_dirty(parent);
479 ret = btrfs_tree_mod_log_free_eb(buf);
481 btrfs_tree_unlock(cow);
482 free_extent_buffer(cow);
483 btrfs_abort_transaction(trans, ret);
487 btrfs_free_tree_block(trans, root, buf, parent_start,
491 btrfs_tree_unlock(buf);
492 free_extent_buffer_stale(buf);
493 btrfs_mark_buffer_dirty(cow);
498 static inline int should_cow_block(struct btrfs_trans_handle *trans,
499 struct btrfs_root *root,
500 struct extent_buffer *buf)
502 if (btrfs_is_testing(root->fs_info))
505 /* Ensure we can see the FORCE_COW bit */
506 smp_mb__before_atomic();
509 * We do not need to cow a block if
510 * 1) this block is not created or changed in this transaction;
511 * 2) this block does not belong to TREE_RELOC tree;
512 * 3) the root is not forced COW.
514 * What is forced COW:
515 * when we create snapshot during committing the transaction,
516 * after we've finished copying src root, we must COW the shared
517 * block to ensure the metadata consistency.
519 if (btrfs_header_generation(buf) == trans->transid &&
520 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
521 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
522 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
523 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state))
529 * cows a single block, see __btrfs_cow_block for the real work.
530 * This version of it has extra checks so that a block isn't COWed more than
531 * once per transaction, as long as it hasn't been written yet
533 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
534 struct btrfs_root *root, struct extent_buffer *buf,
535 struct extent_buffer *parent, int parent_slot,
536 struct extent_buffer **cow_ret,
537 enum btrfs_lock_nesting nest)
539 struct btrfs_fs_info *fs_info = root->fs_info;
543 if (test_bit(BTRFS_ROOT_DELETING, &root->state))
545 "COW'ing blocks on a fs root that's being dropped");
547 if (trans->transaction != fs_info->running_transaction)
548 WARN(1, KERN_CRIT "trans %llu running %llu\n",
550 fs_info->running_transaction->transid);
552 if (trans->transid != fs_info->generation)
553 WARN(1, KERN_CRIT "trans %llu running %llu\n",
554 trans->transid, fs_info->generation);
556 if (!should_cow_block(trans, root, buf)) {
561 search_start = buf->start & ~((u64)SZ_1G - 1);
564 * Before CoWing this block for later modification, check if it's
565 * the subtree root and do the delayed subtree trace if needed.
567 * Also We don't care about the error, as it's handled internally.
569 btrfs_qgroup_trace_subtree_after_cow(trans, root, buf);
570 ret = __btrfs_cow_block(trans, root, buf, parent,
571 parent_slot, cow_ret, search_start, 0, nest);
573 trace_btrfs_cow_block(root, buf, *cow_ret);
577 ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO);
580 * helper function for defrag to decide if two blocks pointed to by a
581 * node are actually close by
583 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
585 if (blocknr < other && other - (blocknr + blocksize) < 32768)
587 if (blocknr > other && blocknr - (other + blocksize) < 32768)
592 #ifdef __LITTLE_ENDIAN
595 * Compare two keys, on little-endian the disk order is same as CPU order and
596 * we can avoid the conversion.
598 static int comp_keys(const struct btrfs_disk_key *disk_key,
599 const struct btrfs_key *k2)
601 const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key;
603 return btrfs_comp_cpu_keys(k1, k2);
609 * compare two keys in a memcmp fashion
611 static int comp_keys(const struct btrfs_disk_key *disk,
612 const struct btrfs_key *k2)
616 btrfs_disk_key_to_cpu(&k1, disk);
618 return btrfs_comp_cpu_keys(&k1, k2);
623 * same as comp_keys only with two btrfs_key's
625 int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2)
627 if (k1->objectid > k2->objectid)
629 if (k1->objectid < k2->objectid)
631 if (k1->type > k2->type)
633 if (k1->type < k2->type)
635 if (k1->offset > k2->offset)
637 if (k1->offset < k2->offset)
643 * this is used by the defrag code to go through all the
644 * leaves pointed to by a node and reallocate them so that
645 * disk order is close to key order
647 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
648 struct btrfs_root *root, struct extent_buffer *parent,
649 int start_slot, u64 *last_ret,
650 struct btrfs_key *progress)
652 struct btrfs_fs_info *fs_info = root->fs_info;
653 struct extent_buffer *cur;
655 u64 search_start = *last_ret;
663 int progress_passed = 0;
664 struct btrfs_disk_key disk_key;
666 WARN_ON(trans->transaction != fs_info->running_transaction);
667 WARN_ON(trans->transid != fs_info->generation);
669 parent_nritems = btrfs_header_nritems(parent);
670 blocksize = fs_info->nodesize;
671 end_slot = parent_nritems - 1;
673 if (parent_nritems <= 1)
676 for (i = start_slot; i <= end_slot; i++) {
679 btrfs_node_key(parent, &disk_key, i);
680 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
684 blocknr = btrfs_node_blockptr(parent, i);
686 last_block = blocknr;
689 other = btrfs_node_blockptr(parent, i - 1);
690 close = close_blocks(blocknr, other, blocksize);
692 if (!close && i < end_slot) {
693 other = btrfs_node_blockptr(parent, i + 1);
694 close = close_blocks(blocknr, other, blocksize);
697 last_block = blocknr;
701 cur = btrfs_read_node_slot(parent, i);
704 if (search_start == 0)
705 search_start = last_block;
707 btrfs_tree_lock(cur);
708 err = __btrfs_cow_block(trans, root, cur, parent, i,
711 (end_slot - i) * blocksize),
714 btrfs_tree_unlock(cur);
715 free_extent_buffer(cur);
718 search_start = cur->start;
719 last_block = cur->start;
720 *last_ret = search_start;
721 btrfs_tree_unlock(cur);
722 free_extent_buffer(cur);
728 * search for key in the extent_buffer. The items start at offset p,
729 * and they are item_size apart.
731 * the slot in the array is returned via slot, and it points to
732 * the place where you would insert key if it is not found in
735 * Slot may point to total number of items if the key is bigger than
738 static noinline int generic_bin_search(struct extent_buffer *eb,
739 unsigned long p, int item_size,
740 const struct btrfs_key *key, int *slot)
743 int high = btrfs_header_nritems(eb);
745 const int key_size = sizeof(struct btrfs_disk_key);
748 btrfs_err(eb->fs_info,
749 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
750 __func__, low, high, eb->start,
751 btrfs_header_owner(eb), btrfs_header_level(eb));
757 unsigned long offset;
758 struct btrfs_disk_key *tmp;
759 struct btrfs_disk_key unaligned;
762 mid = (low + high) / 2;
763 offset = p + mid * item_size;
764 oip = offset_in_page(offset);
766 if (oip + key_size <= PAGE_SIZE) {
767 const unsigned long idx = get_eb_page_index(offset);
768 char *kaddr = page_address(eb->pages[idx]);
770 oip = get_eb_offset_in_page(eb, offset);
771 tmp = (struct btrfs_disk_key *)(kaddr + oip);
773 read_extent_buffer(eb, &unaligned, offset, key_size);
777 ret = comp_keys(tmp, key);
793 * simple bin_search frontend that does the right thing for
796 int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
799 if (btrfs_header_level(eb) == 0)
800 return generic_bin_search(eb,
801 offsetof(struct btrfs_leaf, items),
802 sizeof(struct btrfs_item), key, slot);
804 return generic_bin_search(eb,
805 offsetof(struct btrfs_node, ptrs),
806 sizeof(struct btrfs_key_ptr), key, slot);
809 static void root_add_used(struct btrfs_root *root, u32 size)
811 spin_lock(&root->accounting_lock);
812 btrfs_set_root_used(&root->root_item,
813 btrfs_root_used(&root->root_item) + size);
814 spin_unlock(&root->accounting_lock);
817 static void root_sub_used(struct btrfs_root *root, u32 size)
819 spin_lock(&root->accounting_lock);
820 btrfs_set_root_used(&root->root_item,
821 btrfs_root_used(&root->root_item) - size);
822 spin_unlock(&root->accounting_lock);
825 /* given a node and slot number, this reads the blocks it points to. The
826 * extent buffer is returned with a reference taken (but unlocked).
828 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
831 int level = btrfs_header_level(parent);
832 struct extent_buffer *eb;
833 struct btrfs_key first_key;
835 if (slot < 0 || slot >= btrfs_header_nritems(parent))
836 return ERR_PTR(-ENOENT);
840 btrfs_node_key_to_cpu(parent, &first_key, slot);
841 eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot),
842 btrfs_header_owner(parent),
843 btrfs_node_ptr_generation(parent, slot),
844 level - 1, &first_key);
845 if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) {
846 free_extent_buffer(eb);
854 * node level balancing, used to make sure nodes are in proper order for
855 * item deletion. We balance from the top down, so we have to make sure
856 * that a deletion won't leave an node completely empty later on.
858 static noinline int balance_level(struct btrfs_trans_handle *trans,
859 struct btrfs_root *root,
860 struct btrfs_path *path, int level)
862 struct btrfs_fs_info *fs_info = root->fs_info;
863 struct extent_buffer *right = NULL;
864 struct extent_buffer *mid;
865 struct extent_buffer *left = NULL;
866 struct extent_buffer *parent = NULL;
870 int orig_slot = path->slots[level];
875 mid = path->nodes[level];
877 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK);
878 WARN_ON(btrfs_header_generation(mid) != trans->transid);
880 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
882 if (level < BTRFS_MAX_LEVEL - 1) {
883 parent = path->nodes[level + 1];
884 pslot = path->slots[level + 1];
888 * deal with the case where there is only one pointer in the root
889 * by promoting the node below to a root
892 struct extent_buffer *child;
894 if (btrfs_header_nritems(mid) != 1)
897 /* promote the child to a root */
898 child = btrfs_read_node_slot(mid, 0);
900 ret = PTR_ERR(child);
901 btrfs_handle_fs_error(fs_info, ret, NULL);
905 btrfs_tree_lock(child);
906 ret = btrfs_cow_block(trans, root, child, mid, 0, &child,
909 btrfs_tree_unlock(child);
910 free_extent_buffer(child);
914 ret = btrfs_tree_mod_log_insert_root(root->node, child, true);
916 rcu_assign_pointer(root->node, child);
918 add_root_to_dirty_list(root);
919 btrfs_tree_unlock(child);
921 path->locks[level] = 0;
922 path->nodes[level] = NULL;
923 btrfs_clean_tree_block(mid);
924 btrfs_tree_unlock(mid);
925 /* once for the path */
926 free_extent_buffer(mid);
928 root_sub_used(root, mid->len);
929 btrfs_free_tree_block(trans, root, mid, 0, 1);
930 /* once for the root ptr */
931 free_extent_buffer_stale(mid);
934 if (btrfs_header_nritems(mid) >
935 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4)
938 left = btrfs_read_node_slot(parent, pslot - 1);
943 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
944 wret = btrfs_cow_block(trans, root, left,
945 parent, pslot - 1, &left,
946 BTRFS_NESTING_LEFT_COW);
953 right = btrfs_read_node_slot(parent, pslot + 1);
958 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
959 wret = btrfs_cow_block(trans, root, right,
960 parent, pslot + 1, &right,
961 BTRFS_NESTING_RIGHT_COW);
968 /* first, try to make some room in the middle buffer */
970 orig_slot += btrfs_header_nritems(left);
971 wret = push_node_left(trans, left, mid, 1);
977 * then try to empty the right most buffer into the middle
980 wret = push_node_left(trans, mid, right, 1);
981 if (wret < 0 && wret != -ENOSPC)
983 if (btrfs_header_nritems(right) == 0) {
984 btrfs_clean_tree_block(right);
985 btrfs_tree_unlock(right);
986 del_ptr(root, path, level + 1, pslot + 1);
987 root_sub_used(root, right->len);
988 btrfs_free_tree_block(trans, root, right, 0, 1);
989 free_extent_buffer_stale(right);
992 struct btrfs_disk_key right_key;
993 btrfs_node_key(right, &right_key, 0);
994 ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1,
995 BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS);
997 btrfs_set_node_key(parent, &right_key, pslot + 1);
998 btrfs_mark_buffer_dirty(parent);
1001 if (btrfs_header_nritems(mid) == 1) {
1003 * we're not allowed to leave a node with one item in the
1004 * tree during a delete. A deletion from lower in the tree
1005 * could try to delete the only pointer in this node.
1006 * So, pull some keys from the left.
1007 * There has to be a left pointer at this point because
1008 * otherwise we would have pulled some pointers from the
1013 btrfs_handle_fs_error(fs_info, ret, NULL);
1016 wret = balance_node_right(trans, mid, left);
1022 wret = push_node_left(trans, left, mid, 1);
1028 if (btrfs_header_nritems(mid) == 0) {
1029 btrfs_clean_tree_block(mid);
1030 btrfs_tree_unlock(mid);
1031 del_ptr(root, path, level + 1, pslot);
1032 root_sub_used(root, mid->len);
1033 btrfs_free_tree_block(trans, root, mid, 0, 1);
1034 free_extent_buffer_stale(mid);
1037 /* update the parent key to reflect our changes */
1038 struct btrfs_disk_key mid_key;
1039 btrfs_node_key(mid, &mid_key, 0);
1040 ret = btrfs_tree_mod_log_insert_key(parent, pslot,
1041 BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS);
1043 btrfs_set_node_key(parent, &mid_key, pslot);
1044 btrfs_mark_buffer_dirty(parent);
1047 /* update the path */
1049 if (btrfs_header_nritems(left) > orig_slot) {
1050 atomic_inc(&left->refs);
1051 /* left was locked after cow */
1052 path->nodes[level] = left;
1053 path->slots[level + 1] -= 1;
1054 path->slots[level] = orig_slot;
1056 btrfs_tree_unlock(mid);
1057 free_extent_buffer(mid);
1060 orig_slot -= btrfs_header_nritems(left);
1061 path->slots[level] = orig_slot;
1064 /* double check we haven't messed things up */
1066 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1070 btrfs_tree_unlock(right);
1071 free_extent_buffer(right);
1074 if (path->nodes[level] != left)
1075 btrfs_tree_unlock(left);
1076 free_extent_buffer(left);
1081 /* Node balancing for insertion. Here we only split or push nodes around
1082 * when they are completely full. This is also done top down, so we
1083 * have to be pessimistic.
1085 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1086 struct btrfs_root *root,
1087 struct btrfs_path *path, int level)
1089 struct btrfs_fs_info *fs_info = root->fs_info;
1090 struct extent_buffer *right = NULL;
1091 struct extent_buffer *mid;
1092 struct extent_buffer *left = NULL;
1093 struct extent_buffer *parent = NULL;
1097 int orig_slot = path->slots[level];
1102 mid = path->nodes[level];
1103 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1105 if (level < BTRFS_MAX_LEVEL - 1) {
1106 parent = path->nodes[level + 1];
1107 pslot = path->slots[level + 1];
1113 left = btrfs_read_node_slot(parent, pslot - 1);
1117 /* first, try to make some room in the middle buffer */
1121 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
1123 left_nr = btrfs_header_nritems(left);
1124 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
1127 ret = btrfs_cow_block(trans, root, left, parent,
1129 BTRFS_NESTING_LEFT_COW);
1133 wret = push_node_left(trans, left, mid, 0);
1139 struct btrfs_disk_key disk_key;
1140 orig_slot += left_nr;
1141 btrfs_node_key(mid, &disk_key, 0);
1142 ret = btrfs_tree_mod_log_insert_key(parent, pslot,
1143 BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS);
1145 btrfs_set_node_key(parent, &disk_key, pslot);
1146 btrfs_mark_buffer_dirty(parent);
1147 if (btrfs_header_nritems(left) > orig_slot) {
1148 path->nodes[level] = left;
1149 path->slots[level + 1] -= 1;
1150 path->slots[level] = orig_slot;
1151 btrfs_tree_unlock(mid);
1152 free_extent_buffer(mid);
1155 btrfs_header_nritems(left);
1156 path->slots[level] = orig_slot;
1157 btrfs_tree_unlock(left);
1158 free_extent_buffer(left);
1162 btrfs_tree_unlock(left);
1163 free_extent_buffer(left);
1165 right = btrfs_read_node_slot(parent, pslot + 1);
1170 * then try to empty the right most buffer into the middle
1175 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
1177 right_nr = btrfs_header_nritems(right);
1178 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) {
1181 ret = btrfs_cow_block(trans, root, right,
1183 &right, BTRFS_NESTING_RIGHT_COW);
1187 wret = balance_node_right(trans, right, mid);
1193 struct btrfs_disk_key disk_key;
1195 btrfs_node_key(right, &disk_key, 0);
1196 ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1,
1197 BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS);
1199 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1200 btrfs_mark_buffer_dirty(parent);
1202 if (btrfs_header_nritems(mid) <= orig_slot) {
1203 path->nodes[level] = right;
1204 path->slots[level + 1] += 1;
1205 path->slots[level] = orig_slot -
1206 btrfs_header_nritems(mid);
1207 btrfs_tree_unlock(mid);
1208 free_extent_buffer(mid);
1210 btrfs_tree_unlock(right);
1211 free_extent_buffer(right);
1215 btrfs_tree_unlock(right);
1216 free_extent_buffer(right);
1222 * readahead one full node of leaves, finding things that are close
1223 * to the block in 'slot', and triggering ra on them.
1225 static void reada_for_search(struct btrfs_fs_info *fs_info,
1226 struct btrfs_path *path,
1227 int level, int slot, u64 objectid)
1229 struct extent_buffer *node;
1230 struct btrfs_disk_key disk_key;
1236 struct extent_buffer *eb;
1241 if (level != 1 && path->reada != READA_FORWARD_ALWAYS)
1244 if (!path->nodes[level])
1247 node = path->nodes[level];
1250 * Since the time between visiting leaves is much shorter than the time
1251 * between visiting nodes, limit read ahead of nodes to 1, to avoid too
1252 * much IO at once (possibly random).
1254 if (path->reada == READA_FORWARD_ALWAYS) {
1256 nread_max = node->fs_info->nodesize;
1258 nread_max = SZ_128K;
1263 search = btrfs_node_blockptr(node, slot);
1264 blocksize = fs_info->nodesize;
1265 eb = find_extent_buffer(fs_info, search);
1267 free_extent_buffer(eb);
1273 nritems = btrfs_header_nritems(node);
1277 if (path->reada == READA_BACK) {
1281 } else if (path->reada == READA_FORWARD ||
1282 path->reada == READA_FORWARD_ALWAYS) {
1287 if (path->reada == READA_BACK && objectid) {
1288 btrfs_node_key(node, &disk_key, nr);
1289 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1292 search = btrfs_node_blockptr(node, nr);
1293 if (path->reada == READA_FORWARD_ALWAYS ||
1294 (search <= target && target - search <= 65536) ||
1295 (search > target && search - target <= 65536)) {
1296 btrfs_readahead_node_child(node, nr);
1300 if (nread > nread_max || nscan > 32)
1305 static noinline void reada_for_balance(struct btrfs_path *path, int level)
1307 struct extent_buffer *parent;
1311 parent = path->nodes[level + 1];
1315 nritems = btrfs_header_nritems(parent);
1316 slot = path->slots[level + 1];
1319 btrfs_readahead_node_child(parent, slot - 1);
1320 if (slot + 1 < nritems)
1321 btrfs_readahead_node_child(parent, slot + 1);
1326 * when we walk down the tree, it is usually safe to unlock the higher layers
1327 * in the tree. The exceptions are when our path goes through slot 0, because
1328 * operations on the tree might require changing key pointers higher up in the
1331 * callers might also have set path->keep_locks, which tells this code to keep
1332 * the lock if the path points to the last slot in the block. This is part of
1333 * walking through the tree, and selecting the next slot in the higher block.
1335 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1336 * if lowest_unlock is 1, level 0 won't be unlocked
1338 static noinline void unlock_up(struct btrfs_path *path, int level,
1339 int lowest_unlock, int min_write_lock_level,
1340 int *write_lock_level)
1343 int skip_level = level;
1345 struct extent_buffer *t;
1347 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1348 if (!path->nodes[i])
1350 if (!path->locks[i])
1352 if (!no_skips && path->slots[i] == 0) {
1356 if (!no_skips && path->keep_locks) {
1359 nritems = btrfs_header_nritems(t);
1360 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1365 if (skip_level < i && i >= lowest_unlock)
1369 if (i >= lowest_unlock && i > skip_level) {
1370 btrfs_tree_unlock_rw(t, path->locks[i]);
1372 if (write_lock_level &&
1373 i > min_write_lock_level &&
1374 i <= *write_lock_level) {
1375 *write_lock_level = i - 1;
1382 * helper function for btrfs_search_slot. The goal is to find a block
1383 * in cache without setting the path to blocking. If we find the block
1384 * we return zero and the path is unchanged.
1386 * If we can't find the block, we set the path blocking and do some
1387 * reada. -EAGAIN is returned and the search must be repeated.
1390 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p,
1391 struct extent_buffer **eb_ret, int level, int slot,
1392 const struct btrfs_key *key)
1394 struct btrfs_fs_info *fs_info = root->fs_info;
1397 struct extent_buffer *tmp;
1398 struct btrfs_key first_key;
1402 blocknr = btrfs_node_blockptr(*eb_ret, slot);
1403 gen = btrfs_node_ptr_generation(*eb_ret, slot);
1404 parent_level = btrfs_header_level(*eb_ret);
1405 btrfs_node_key_to_cpu(*eb_ret, &first_key, slot);
1407 tmp = find_extent_buffer(fs_info, blocknr);
1409 if (p->reada == READA_FORWARD_ALWAYS)
1410 reada_for_search(fs_info, p, level, slot, key->objectid);
1412 /* first we do an atomic uptodate check */
1413 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
1415 * Do extra check for first_key, eb can be stale due to
1416 * being cached, read from scrub, or have multiple
1417 * parents (shared tree blocks).
1419 if (btrfs_verify_level_key(tmp,
1420 parent_level - 1, &first_key, gen)) {
1421 free_extent_buffer(tmp);
1428 /* now we're allowed to do a blocking uptodate check */
1429 ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key);
1434 free_extent_buffer(tmp);
1435 btrfs_release_path(p);
1440 * reduce lock contention at high levels
1441 * of the btree by dropping locks before
1442 * we read. Don't release the lock on the current
1443 * level because we need to walk this node to figure
1444 * out which blocks to read.
1446 btrfs_unlock_up_safe(p, level + 1);
1448 if (p->reada != READA_NONE)
1449 reada_for_search(fs_info, p, level, slot, key->objectid);
1452 tmp = read_tree_block(fs_info, blocknr, root->root_key.objectid,
1453 gen, parent_level - 1, &first_key);
1456 * If the read above didn't mark this buffer up to date,
1457 * it will never end up being up to date. Set ret to EIO now
1458 * and give up so that our caller doesn't loop forever
1461 if (!extent_buffer_uptodate(tmp))
1463 free_extent_buffer(tmp);
1468 btrfs_release_path(p);
1473 * helper function for btrfs_search_slot. This does all of the checks
1474 * for node-level blocks and does any balancing required based on
1477 * If no extra work was required, zero is returned. If we had to
1478 * drop the path, -EAGAIN is returned and btrfs_search_slot must
1482 setup_nodes_for_search(struct btrfs_trans_handle *trans,
1483 struct btrfs_root *root, struct btrfs_path *p,
1484 struct extent_buffer *b, int level, int ins_len,
1485 int *write_lock_level)
1487 struct btrfs_fs_info *fs_info = root->fs_info;
1490 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
1491 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) {
1493 if (*write_lock_level < level + 1) {
1494 *write_lock_level = level + 1;
1495 btrfs_release_path(p);
1499 reada_for_balance(p, level);
1500 ret = split_node(trans, root, p, level);
1502 b = p->nodes[level];
1503 } else if (ins_len < 0 && btrfs_header_nritems(b) <
1504 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) {
1506 if (*write_lock_level < level + 1) {
1507 *write_lock_level = level + 1;
1508 btrfs_release_path(p);
1512 reada_for_balance(p, level);
1513 ret = balance_level(trans, root, p, level);
1517 b = p->nodes[level];
1519 btrfs_release_path(p);
1522 BUG_ON(btrfs_header_nritems(b) == 1);
1527 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
1528 u64 iobjectid, u64 ioff, u8 key_type,
1529 struct btrfs_key *found_key)
1532 struct btrfs_key key;
1533 struct extent_buffer *eb;
1538 key.type = key_type;
1539 key.objectid = iobjectid;
1542 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1546 eb = path->nodes[0];
1547 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1548 ret = btrfs_next_leaf(fs_root, path);
1551 eb = path->nodes[0];
1554 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1555 if (found_key->type != key.type ||
1556 found_key->objectid != key.objectid)
1562 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root,
1563 struct btrfs_path *p,
1564 int write_lock_level)
1566 struct btrfs_fs_info *fs_info = root->fs_info;
1567 struct extent_buffer *b;
1571 /* We try very hard to do read locks on the root */
1572 root_lock = BTRFS_READ_LOCK;
1574 if (p->search_commit_root) {
1576 * The commit roots are read only so we always do read locks,
1577 * and we always must hold the commit_root_sem when doing
1578 * searches on them, the only exception is send where we don't
1579 * want to block transaction commits for a long time, so
1580 * we need to clone the commit root in order to avoid races
1581 * with transaction commits that create a snapshot of one of
1582 * the roots used by a send operation.
1584 if (p->need_commit_sem) {
1585 down_read(&fs_info->commit_root_sem);
1586 b = btrfs_clone_extent_buffer(root->commit_root);
1587 up_read(&fs_info->commit_root_sem);
1589 return ERR_PTR(-ENOMEM);
1592 b = root->commit_root;
1593 atomic_inc(&b->refs);
1595 level = btrfs_header_level(b);
1597 * Ensure that all callers have set skip_locking when
1598 * p->search_commit_root = 1.
1600 ASSERT(p->skip_locking == 1);
1605 if (p->skip_locking) {
1606 b = btrfs_root_node(root);
1607 level = btrfs_header_level(b);
1612 * If the level is set to maximum, we can skip trying to get the read
1615 if (write_lock_level < BTRFS_MAX_LEVEL) {
1617 * We don't know the level of the root node until we actually
1618 * have it read locked
1620 b = btrfs_read_lock_root_node(root);
1621 level = btrfs_header_level(b);
1622 if (level > write_lock_level)
1625 /* Whoops, must trade for write lock */
1626 btrfs_tree_read_unlock(b);
1627 free_extent_buffer(b);
1630 b = btrfs_lock_root_node(root);
1631 root_lock = BTRFS_WRITE_LOCK;
1633 /* The level might have changed, check again */
1634 level = btrfs_header_level(b);
1637 p->nodes[level] = b;
1638 if (!p->skip_locking)
1639 p->locks[level] = root_lock;
1641 * Callers are responsible for dropping b's references.
1648 * btrfs_search_slot - look for a key in a tree and perform necessary
1649 * modifications to preserve tree invariants.
1651 * @trans: Handle of transaction, used when modifying the tree
1652 * @p: Holds all btree nodes along the search path
1653 * @root: The root node of the tree
1654 * @key: The key we are looking for
1655 * @ins_len: Indicates purpose of search:
1656 * >0 for inserts it's size of item inserted (*)
1658 * 0 for plain searches, not modifying the tree
1660 * (*) If size of item inserted doesn't include
1661 * sizeof(struct btrfs_item), then p->search_for_extension must
1663 * @cow: boolean should CoW operations be performed. Must always be 1
1664 * when modifying the tree.
1666 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
1667 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
1669 * If @key is found, 0 is returned and you can find the item in the leaf level
1670 * of the path (level 0)
1672 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
1673 * points to the slot where it should be inserted
1675 * If an error is encountered while searching the tree a negative error number
1678 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1679 const struct btrfs_key *key, struct btrfs_path *p,
1680 int ins_len, int cow)
1682 struct extent_buffer *b;
1687 int lowest_unlock = 1;
1688 /* everything at write_lock_level or lower must be write locked */
1689 int write_lock_level = 0;
1690 u8 lowest_level = 0;
1691 int min_write_lock_level;
1694 lowest_level = p->lowest_level;
1695 WARN_ON(lowest_level && ins_len > 0);
1696 WARN_ON(p->nodes[0] != NULL);
1697 BUG_ON(!cow && ins_len);
1702 /* when we are removing items, we might have to go up to level
1703 * two as we update tree pointers Make sure we keep write
1704 * for those levels as well
1706 write_lock_level = 2;
1707 } else if (ins_len > 0) {
1709 * for inserting items, make sure we have a write lock on
1710 * level 1 so we can update keys
1712 write_lock_level = 1;
1716 write_lock_level = -1;
1718 if (cow && (p->keep_locks || p->lowest_level))
1719 write_lock_level = BTRFS_MAX_LEVEL;
1721 min_write_lock_level = write_lock_level;
1725 b = btrfs_search_slot_get_root(root, p, write_lock_level);
1734 level = btrfs_header_level(b);
1737 bool last_level = (level == (BTRFS_MAX_LEVEL - 1));
1740 * if we don't really need to cow this block
1741 * then we don't want to set the path blocking,
1742 * so we test it here
1744 if (!should_cow_block(trans, root, b))
1748 * must have write locks on this node and the
1751 if (level > write_lock_level ||
1752 (level + 1 > write_lock_level &&
1753 level + 1 < BTRFS_MAX_LEVEL &&
1754 p->nodes[level + 1])) {
1755 write_lock_level = level + 1;
1756 btrfs_release_path(p);
1761 err = btrfs_cow_block(trans, root, b, NULL, 0,
1765 err = btrfs_cow_block(trans, root, b,
1766 p->nodes[level + 1],
1767 p->slots[level + 1], &b,
1775 p->nodes[level] = b;
1777 * Leave path with blocking locks to avoid massive
1778 * lock context switch, this is made on purpose.
1782 * we have a lock on b and as long as we aren't changing
1783 * the tree, there is no way to for the items in b to change.
1784 * It is safe to drop the lock on our parent before we
1785 * go through the expensive btree search on b.
1787 * If we're inserting or deleting (ins_len != 0), then we might
1788 * be changing slot zero, which may require changing the parent.
1789 * So, we can't drop the lock until after we know which slot
1790 * we're operating on.
1792 if (!ins_len && !p->keep_locks) {
1795 if (u < BTRFS_MAX_LEVEL && p->locks[u]) {
1796 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]);
1802 * If btrfs_bin_search returns an exact match (prev_cmp == 0)
1803 * we can safely assume the target key will always be in slot 0
1804 * on lower levels due to the invariants BTRFS' btree provides,
1805 * namely that a btrfs_key_ptr entry always points to the
1806 * lowest key in the child node, thus we can skip searching
1809 if (prev_cmp == 0) {
1813 ret = btrfs_bin_search(b, key, &slot);
1820 p->slots[level] = slot;
1822 * Item key already exists. In this case, if we are
1823 * allowed to insert the item (for example, in dir_item
1824 * case, item key collision is allowed), it will be
1825 * merged with the original item. Only the item size
1826 * grows, no new btrfs item will be added. If
1827 * search_for_extension is not set, ins_len already
1828 * accounts the size btrfs_item, deduct it here so leaf
1829 * space check will be correct.
1831 if (ret == 0 && ins_len > 0 && !p->search_for_extension) {
1832 ASSERT(ins_len >= sizeof(struct btrfs_item));
1833 ins_len -= sizeof(struct btrfs_item);
1836 btrfs_leaf_free_space(b) < ins_len) {
1837 if (write_lock_level < 1) {
1838 write_lock_level = 1;
1839 btrfs_release_path(p);
1843 err = split_leaf(trans, root, key,
1844 p, ins_len, ret == 0);
1852 if (!p->search_for_split)
1853 unlock_up(p, level, lowest_unlock,
1854 min_write_lock_level, NULL);
1857 if (ret && slot > 0) {
1861 p->slots[level] = slot;
1862 err = setup_nodes_for_search(trans, root, p, b, level, ins_len,
1870 b = p->nodes[level];
1871 slot = p->slots[level];
1874 * Slot 0 is special, if we change the key we have to update
1875 * the parent pointer which means we must have a write lock on
1878 if (slot == 0 && ins_len && write_lock_level < level + 1) {
1879 write_lock_level = level + 1;
1880 btrfs_release_path(p);
1884 unlock_up(p, level, lowest_unlock, min_write_lock_level,
1887 if (level == lowest_level) {
1893 err = read_block_for_search(root, p, &b, level, slot, key);
1901 if (!p->skip_locking) {
1902 level = btrfs_header_level(b);
1903 if (level <= write_lock_level) {
1905 p->locks[level] = BTRFS_WRITE_LOCK;
1907 btrfs_tree_read_lock(b);
1908 p->locks[level] = BTRFS_READ_LOCK;
1910 p->nodes[level] = b;
1915 if (ret < 0 && !p->skip_release_on_error)
1916 btrfs_release_path(p);
1919 ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO);
1922 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
1923 * current state of the tree together with the operations recorded in the tree
1924 * modification log to search for the key in a previous version of this tree, as
1925 * denoted by the time_seq parameter.
1927 * Naturally, there is no support for insert, delete or cow operations.
1929 * The resulting path and return value will be set up as if we called
1930 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
1932 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
1933 struct btrfs_path *p, u64 time_seq)
1935 struct btrfs_fs_info *fs_info = root->fs_info;
1936 struct extent_buffer *b;
1941 int lowest_unlock = 1;
1942 u8 lowest_level = 0;
1944 lowest_level = p->lowest_level;
1945 WARN_ON(p->nodes[0] != NULL);
1947 if (p->search_commit_root) {
1949 return btrfs_search_slot(NULL, root, key, p, 0, 0);
1953 b = btrfs_get_old_root(root, time_seq);
1958 level = btrfs_header_level(b);
1959 p->locks[level] = BTRFS_READ_LOCK;
1964 level = btrfs_header_level(b);
1965 p->nodes[level] = b;
1968 * we have a lock on b and as long as we aren't changing
1969 * the tree, there is no way to for the items in b to change.
1970 * It is safe to drop the lock on our parent before we
1971 * go through the expensive btree search on b.
1973 btrfs_unlock_up_safe(p, level + 1);
1975 ret = btrfs_bin_search(b, key, &slot);
1980 p->slots[level] = slot;
1981 unlock_up(p, level, lowest_unlock, 0, NULL);
1985 if (ret && slot > 0) {
1989 p->slots[level] = slot;
1990 unlock_up(p, level, lowest_unlock, 0, NULL);
1992 if (level == lowest_level) {
1998 err = read_block_for_search(root, p, &b, level, slot, key);
2006 level = btrfs_header_level(b);
2007 btrfs_tree_read_lock(b);
2008 b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq);
2013 p->locks[level] = BTRFS_READ_LOCK;
2014 p->nodes[level] = b;
2019 btrfs_release_path(p);
2025 * helper to use instead of search slot if no exact match is needed but
2026 * instead the next or previous item should be returned.
2027 * When find_higher is true, the next higher item is returned, the next lower
2029 * When return_any and find_higher are both true, and no higher item is found,
2030 * return the next lower instead.
2031 * When return_any is true and find_higher is false, and no lower item is found,
2032 * return the next higher instead.
2033 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2036 int btrfs_search_slot_for_read(struct btrfs_root *root,
2037 const struct btrfs_key *key,
2038 struct btrfs_path *p, int find_higher,
2042 struct extent_buffer *leaf;
2045 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
2049 * a return value of 1 means the path is at the position where the
2050 * item should be inserted. Normally this is the next bigger item,
2051 * but in case the previous item is the last in a leaf, path points
2052 * to the first free slot in the previous leaf, i.e. at an invalid
2058 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
2059 ret = btrfs_next_leaf(root, p);
2065 * no higher item found, return the next
2070 btrfs_release_path(p);
2074 if (p->slots[0] == 0) {
2075 ret = btrfs_prev_leaf(root, p);
2080 if (p->slots[0] == btrfs_header_nritems(leaf))
2087 * no lower item found, return the next
2092 btrfs_release_path(p);
2102 * adjust the pointers going up the tree, starting at level
2103 * making sure the right key of each node is points to 'key'.
2104 * This is used after shifting pointers to the left, so it stops
2105 * fixing up pointers when a given leaf/node is not in slot 0 of the
2109 static void fixup_low_keys(struct btrfs_path *path,
2110 struct btrfs_disk_key *key, int level)
2113 struct extent_buffer *t;
2116 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2117 int tslot = path->slots[i];
2119 if (!path->nodes[i])
2122 ret = btrfs_tree_mod_log_insert_key(t, tslot,
2123 BTRFS_MOD_LOG_KEY_REPLACE, GFP_ATOMIC);
2125 btrfs_set_node_key(t, key, tslot);
2126 btrfs_mark_buffer_dirty(path->nodes[i]);
2135 * This function isn't completely safe. It's the caller's responsibility
2136 * that the new key won't break the order
2138 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
2139 struct btrfs_path *path,
2140 const struct btrfs_key *new_key)
2142 struct btrfs_disk_key disk_key;
2143 struct extent_buffer *eb;
2146 eb = path->nodes[0];
2147 slot = path->slots[0];
2149 btrfs_item_key(eb, &disk_key, slot - 1);
2150 if (unlikely(comp_keys(&disk_key, new_key) >= 0)) {
2152 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
2153 slot, btrfs_disk_key_objectid(&disk_key),
2154 btrfs_disk_key_type(&disk_key),
2155 btrfs_disk_key_offset(&disk_key),
2156 new_key->objectid, new_key->type,
2158 btrfs_print_leaf(eb);
2162 if (slot < btrfs_header_nritems(eb) - 1) {
2163 btrfs_item_key(eb, &disk_key, slot + 1);
2164 if (unlikely(comp_keys(&disk_key, new_key) <= 0)) {
2166 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
2167 slot, btrfs_disk_key_objectid(&disk_key),
2168 btrfs_disk_key_type(&disk_key),
2169 btrfs_disk_key_offset(&disk_key),
2170 new_key->objectid, new_key->type,
2172 btrfs_print_leaf(eb);
2177 btrfs_cpu_key_to_disk(&disk_key, new_key);
2178 btrfs_set_item_key(eb, &disk_key, slot);
2179 btrfs_mark_buffer_dirty(eb);
2181 fixup_low_keys(path, &disk_key, 1);
2185 * Check key order of two sibling extent buffers.
2187 * Return true if something is wrong.
2188 * Return false if everything is fine.
2190 * Tree-checker only works inside one tree block, thus the following
2191 * corruption can not be detected by tree-checker:
2193 * Leaf @left | Leaf @right
2194 * --------------------------------------------------------------
2195 * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 |
2197 * Key f6 in leaf @left itself is valid, but not valid when the next
2198 * key in leaf @right is 7.
2199 * This can only be checked at tree block merge time.
2200 * And since tree checker has ensured all key order in each tree block
2201 * is correct, we only need to bother the last key of @left and the first
2204 static bool check_sibling_keys(struct extent_buffer *left,
2205 struct extent_buffer *right)
2207 struct btrfs_key left_last;
2208 struct btrfs_key right_first;
2209 int level = btrfs_header_level(left);
2210 int nr_left = btrfs_header_nritems(left);
2211 int nr_right = btrfs_header_nritems(right);
2213 /* No key to check in one of the tree blocks */
2214 if (!nr_left || !nr_right)
2218 btrfs_node_key_to_cpu(left, &left_last, nr_left - 1);
2219 btrfs_node_key_to_cpu(right, &right_first, 0);
2221 btrfs_item_key_to_cpu(left, &left_last, nr_left - 1);
2222 btrfs_item_key_to_cpu(right, &right_first, 0);
2225 if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) {
2226 btrfs_crit(left->fs_info,
2227 "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)",
2228 left_last.objectid, left_last.type,
2229 left_last.offset, right_first.objectid,
2230 right_first.type, right_first.offset);
2237 * try to push data from one node into the next node left in the
2240 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2241 * error, and > 0 if there was no room in the left hand block.
2243 static int push_node_left(struct btrfs_trans_handle *trans,
2244 struct extent_buffer *dst,
2245 struct extent_buffer *src, int empty)
2247 struct btrfs_fs_info *fs_info = trans->fs_info;
2253 src_nritems = btrfs_header_nritems(src);
2254 dst_nritems = btrfs_header_nritems(dst);
2255 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
2256 WARN_ON(btrfs_header_generation(src) != trans->transid);
2257 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2259 if (!empty && src_nritems <= 8)
2262 if (push_items <= 0)
2266 push_items = min(src_nritems, push_items);
2267 if (push_items < src_nritems) {
2268 /* leave at least 8 pointers in the node if
2269 * we aren't going to empty it
2271 if (src_nritems - push_items < 8) {
2272 if (push_items <= 8)
2278 push_items = min(src_nritems - 8, push_items);
2280 /* dst is the left eb, src is the middle eb */
2281 if (check_sibling_keys(dst, src)) {
2283 btrfs_abort_transaction(trans, ret);
2286 ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items);
2288 btrfs_abort_transaction(trans, ret);
2291 copy_extent_buffer(dst, src,
2292 btrfs_node_key_ptr_offset(dst_nritems),
2293 btrfs_node_key_ptr_offset(0),
2294 push_items * sizeof(struct btrfs_key_ptr));
2296 if (push_items < src_nritems) {
2298 * Don't call btrfs_tree_mod_log_insert_move() here, key removal
2299 * was already fully logged by btrfs_tree_mod_log_eb_copy() above.
2301 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
2302 btrfs_node_key_ptr_offset(push_items),
2303 (src_nritems - push_items) *
2304 sizeof(struct btrfs_key_ptr));
2306 btrfs_set_header_nritems(src, src_nritems - push_items);
2307 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2308 btrfs_mark_buffer_dirty(src);
2309 btrfs_mark_buffer_dirty(dst);
2315 * try to push data from one node into the next node right in the
2318 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
2319 * error, and > 0 if there was no room in the right hand block.
2321 * this will only push up to 1/2 the contents of the left node over
2323 static int balance_node_right(struct btrfs_trans_handle *trans,
2324 struct extent_buffer *dst,
2325 struct extent_buffer *src)
2327 struct btrfs_fs_info *fs_info = trans->fs_info;
2334 WARN_ON(btrfs_header_generation(src) != trans->transid);
2335 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2337 src_nritems = btrfs_header_nritems(src);
2338 dst_nritems = btrfs_header_nritems(dst);
2339 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems;
2340 if (push_items <= 0)
2343 if (src_nritems < 4)
2346 max_push = src_nritems / 2 + 1;
2347 /* don't try to empty the node */
2348 if (max_push >= src_nritems)
2351 if (max_push < push_items)
2352 push_items = max_push;
2354 /* dst is the right eb, src is the middle eb */
2355 if (check_sibling_keys(src, dst)) {
2357 btrfs_abort_transaction(trans, ret);
2360 ret = btrfs_tree_mod_log_insert_move(dst, push_items, 0, dst_nritems);
2362 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2363 btrfs_node_key_ptr_offset(0),
2365 sizeof(struct btrfs_key_ptr));
2367 ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items,
2370 btrfs_abort_transaction(trans, ret);
2373 copy_extent_buffer(dst, src,
2374 btrfs_node_key_ptr_offset(0),
2375 btrfs_node_key_ptr_offset(src_nritems - push_items),
2376 push_items * sizeof(struct btrfs_key_ptr));
2378 btrfs_set_header_nritems(src, src_nritems - push_items);
2379 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2381 btrfs_mark_buffer_dirty(src);
2382 btrfs_mark_buffer_dirty(dst);
2388 * helper function to insert a new root level in the tree.
2389 * A new node is allocated, and a single item is inserted to
2390 * point to the existing root
2392 * returns zero on success or < 0 on failure.
2394 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2395 struct btrfs_root *root,
2396 struct btrfs_path *path, int level)
2398 struct btrfs_fs_info *fs_info = root->fs_info;
2400 struct extent_buffer *lower;
2401 struct extent_buffer *c;
2402 struct extent_buffer *old;
2403 struct btrfs_disk_key lower_key;
2406 BUG_ON(path->nodes[level]);
2407 BUG_ON(path->nodes[level-1] != root->node);
2409 lower = path->nodes[level-1];
2411 btrfs_item_key(lower, &lower_key, 0);
2413 btrfs_node_key(lower, &lower_key, 0);
2415 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
2416 &lower_key, level, root->node->start, 0,
2417 BTRFS_NESTING_NEW_ROOT);
2421 root_add_used(root, fs_info->nodesize);
2423 btrfs_set_header_nritems(c, 1);
2424 btrfs_set_node_key(c, &lower_key, 0);
2425 btrfs_set_node_blockptr(c, 0, lower->start);
2426 lower_gen = btrfs_header_generation(lower);
2427 WARN_ON(lower_gen != trans->transid);
2429 btrfs_set_node_ptr_generation(c, 0, lower_gen);
2431 btrfs_mark_buffer_dirty(c);
2434 ret = btrfs_tree_mod_log_insert_root(root->node, c, false);
2436 rcu_assign_pointer(root->node, c);
2438 /* the super has an extra ref to root->node */
2439 free_extent_buffer(old);
2441 add_root_to_dirty_list(root);
2442 atomic_inc(&c->refs);
2443 path->nodes[level] = c;
2444 path->locks[level] = BTRFS_WRITE_LOCK;
2445 path->slots[level] = 0;
2450 * worker function to insert a single pointer in a node.
2451 * the node should have enough room for the pointer already
2453 * slot and level indicate where you want the key to go, and
2454 * blocknr is the block the key points to.
2456 static void insert_ptr(struct btrfs_trans_handle *trans,
2457 struct btrfs_path *path,
2458 struct btrfs_disk_key *key, u64 bytenr,
2459 int slot, int level)
2461 struct extent_buffer *lower;
2465 BUG_ON(!path->nodes[level]);
2466 btrfs_assert_tree_locked(path->nodes[level]);
2467 lower = path->nodes[level];
2468 nritems = btrfs_header_nritems(lower);
2469 BUG_ON(slot > nritems);
2470 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info));
2471 if (slot != nritems) {
2473 ret = btrfs_tree_mod_log_insert_move(lower, slot + 1,
2474 slot, nritems - slot);
2477 memmove_extent_buffer(lower,
2478 btrfs_node_key_ptr_offset(slot + 1),
2479 btrfs_node_key_ptr_offset(slot),
2480 (nritems - slot) * sizeof(struct btrfs_key_ptr));
2483 ret = btrfs_tree_mod_log_insert_key(lower, slot,
2484 BTRFS_MOD_LOG_KEY_ADD, GFP_NOFS);
2487 btrfs_set_node_key(lower, key, slot);
2488 btrfs_set_node_blockptr(lower, slot, bytenr);
2489 WARN_ON(trans->transid == 0);
2490 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2491 btrfs_set_header_nritems(lower, nritems + 1);
2492 btrfs_mark_buffer_dirty(lower);
2496 * split the node at the specified level in path in two.
2497 * The path is corrected to point to the appropriate node after the split
2499 * Before splitting this tries to make some room in the node by pushing
2500 * left and right, if either one works, it returns right away.
2502 * returns 0 on success and < 0 on failure
2504 static noinline int split_node(struct btrfs_trans_handle *trans,
2505 struct btrfs_root *root,
2506 struct btrfs_path *path, int level)
2508 struct btrfs_fs_info *fs_info = root->fs_info;
2509 struct extent_buffer *c;
2510 struct extent_buffer *split;
2511 struct btrfs_disk_key disk_key;
2516 c = path->nodes[level];
2517 WARN_ON(btrfs_header_generation(c) != trans->transid);
2518 if (c == root->node) {
2520 * trying to split the root, lets make a new one
2522 * tree mod log: We don't log_removal old root in
2523 * insert_new_root, because that root buffer will be kept as a
2524 * normal node. We are going to log removal of half of the
2525 * elements below with btrfs_tree_mod_log_eb_copy(). We're
2526 * holding a tree lock on the buffer, which is why we cannot
2527 * race with other tree_mod_log users.
2529 ret = insert_new_root(trans, root, path, level + 1);
2533 ret = push_nodes_for_insert(trans, root, path, level);
2534 c = path->nodes[level];
2535 if (!ret && btrfs_header_nritems(c) <
2536 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3)
2542 c_nritems = btrfs_header_nritems(c);
2543 mid = (c_nritems + 1) / 2;
2544 btrfs_node_key(c, &disk_key, mid);
2546 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
2547 &disk_key, level, c->start, 0,
2548 BTRFS_NESTING_SPLIT);
2550 return PTR_ERR(split);
2552 root_add_used(root, fs_info->nodesize);
2553 ASSERT(btrfs_header_level(c) == level);
2555 ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid);
2557 btrfs_abort_transaction(trans, ret);
2560 copy_extent_buffer(split, c,
2561 btrfs_node_key_ptr_offset(0),
2562 btrfs_node_key_ptr_offset(mid),
2563 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2564 btrfs_set_header_nritems(split, c_nritems - mid);
2565 btrfs_set_header_nritems(c, mid);
2567 btrfs_mark_buffer_dirty(c);
2568 btrfs_mark_buffer_dirty(split);
2570 insert_ptr(trans, path, &disk_key, split->start,
2571 path->slots[level + 1] + 1, level + 1);
2573 if (path->slots[level] >= mid) {
2574 path->slots[level] -= mid;
2575 btrfs_tree_unlock(c);
2576 free_extent_buffer(c);
2577 path->nodes[level] = split;
2578 path->slots[level + 1] += 1;
2580 btrfs_tree_unlock(split);
2581 free_extent_buffer(split);
2587 * how many bytes are required to store the items in a leaf. start
2588 * and nr indicate which items in the leaf to check. This totals up the
2589 * space used both by the item structs and the item data
2591 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2593 struct btrfs_item *start_item;
2594 struct btrfs_item *end_item;
2596 int nritems = btrfs_header_nritems(l);
2597 int end = min(nritems, start + nr) - 1;
2601 start_item = btrfs_item_nr(start);
2602 end_item = btrfs_item_nr(end);
2603 data_len = btrfs_item_offset(l, start_item) +
2604 btrfs_item_size(l, start_item);
2605 data_len = data_len - btrfs_item_offset(l, end_item);
2606 data_len += sizeof(struct btrfs_item) * nr;
2607 WARN_ON(data_len < 0);
2612 * The space between the end of the leaf items and
2613 * the start of the leaf data. IOW, how much room
2614 * the leaf has left for both items and data
2616 noinline int btrfs_leaf_free_space(struct extent_buffer *leaf)
2618 struct btrfs_fs_info *fs_info = leaf->fs_info;
2619 int nritems = btrfs_header_nritems(leaf);
2622 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems);
2625 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
2627 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info),
2628 leaf_space_used(leaf, 0, nritems), nritems);
2634 * min slot controls the lowest index we're willing to push to the
2635 * right. We'll push up to and including min_slot, but no lower
2637 static noinline int __push_leaf_right(struct btrfs_path *path,
2638 int data_size, int empty,
2639 struct extent_buffer *right,
2640 int free_space, u32 left_nritems,
2643 struct btrfs_fs_info *fs_info = right->fs_info;
2644 struct extent_buffer *left = path->nodes[0];
2645 struct extent_buffer *upper = path->nodes[1];
2646 struct btrfs_map_token token;
2647 struct btrfs_disk_key disk_key;
2652 struct btrfs_item *item;
2661 nr = max_t(u32, 1, min_slot);
2663 if (path->slots[0] >= left_nritems)
2664 push_space += data_size;
2666 slot = path->slots[1];
2667 i = left_nritems - 1;
2669 item = btrfs_item_nr(i);
2671 if (!empty && push_items > 0) {
2672 if (path->slots[0] > i)
2674 if (path->slots[0] == i) {
2675 int space = btrfs_leaf_free_space(left);
2677 if (space + push_space * 2 > free_space)
2682 if (path->slots[0] == i)
2683 push_space += data_size;
2685 this_item_size = btrfs_item_size(left, item);
2686 if (this_item_size + sizeof(*item) + push_space > free_space)
2690 push_space += this_item_size + sizeof(*item);
2696 if (push_items == 0)
2699 WARN_ON(!empty && push_items == left_nritems);
2701 /* push left to right */
2702 right_nritems = btrfs_header_nritems(right);
2704 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2705 push_space -= leaf_data_end(left);
2707 /* make room in the right data area */
2708 data_end = leaf_data_end(right);
2709 memmove_extent_buffer(right,
2710 BTRFS_LEAF_DATA_OFFSET + data_end - push_space,
2711 BTRFS_LEAF_DATA_OFFSET + data_end,
2712 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end);
2714 /* copy from the left data area */
2715 copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET +
2716 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
2717 BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left),
2720 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2721 btrfs_item_nr_offset(0),
2722 right_nritems * sizeof(struct btrfs_item));
2724 /* copy the items from left to right */
2725 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2726 btrfs_item_nr_offset(left_nritems - push_items),
2727 push_items * sizeof(struct btrfs_item));
2729 /* update the item pointers */
2730 btrfs_init_map_token(&token, right);
2731 right_nritems += push_items;
2732 btrfs_set_header_nritems(right, right_nritems);
2733 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
2734 for (i = 0; i < right_nritems; i++) {
2735 item = btrfs_item_nr(i);
2736 push_space -= btrfs_token_item_size(&token, item);
2737 btrfs_set_token_item_offset(&token, item, push_space);
2740 left_nritems -= push_items;
2741 btrfs_set_header_nritems(left, left_nritems);
2744 btrfs_mark_buffer_dirty(left);
2746 btrfs_clean_tree_block(left);
2748 btrfs_mark_buffer_dirty(right);
2750 btrfs_item_key(right, &disk_key, 0);
2751 btrfs_set_node_key(upper, &disk_key, slot + 1);
2752 btrfs_mark_buffer_dirty(upper);
2754 /* then fixup the leaf pointer in the path */
2755 if (path->slots[0] >= left_nritems) {
2756 path->slots[0] -= left_nritems;
2757 if (btrfs_header_nritems(path->nodes[0]) == 0)
2758 btrfs_clean_tree_block(path->nodes[0]);
2759 btrfs_tree_unlock(path->nodes[0]);
2760 free_extent_buffer(path->nodes[0]);
2761 path->nodes[0] = right;
2762 path->slots[1] += 1;
2764 btrfs_tree_unlock(right);
2765 free_extent_buffer(right);
2770 btrfs_tree_unlock(right);
2771 free_extent_buffer(right);
2776 * push some data in the path leaf to the right, trying to free up at
2777 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2779 * returns 1 if the push failed because the other node didn't have enough
2780 * room, 0 if everything worked out and < 0 if there were major errors.
2782 * this will push starting from min_slot to the end of the leaf. It won't
2783 * push any slot lower than min_slot
2785 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2786 *root, struct btrfs_path *path,
2787 int min_data_size, int data_size,
2788 int empty, u32 min_slot)
2790 struct extent_buffer *left = path->nodes[0];
2791 struct extent_buffer *right;
2792 struct extent_buffer *upper;
2798 if (!path->nodes[1])
2801 slot = path->slots[1];
2802 upper = path->nodes[1];
2803 if (slot >= btrfs_header_nritems(upper) - 1)
2806 btrfs_assert_tree_locked(path->nodes[1]);
2808 right = btrfs_read_node_slot(upper, slot + 1);
2810 * slot + 1 is not valid or we fail to read the right node,
2811 * no big deal, just return.
2816 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT);
2818 free_space = btrfs_leaf_free_space(right);
2819 if (free_space < data_size)
2822 /* cow and double check */
2823 ret = btrfs_cow_block(trans, root, right, upper,
2824 slot + 1, &right, BTRFS_NESTING_RIGHT_COW);
2828 free_space = btrfs_leaf_free_space(right);
2829 if (free_space < data_size)
2832 left_nritems = btrfs_header_nritems(left);
2833 if (left_nritems == 0)
2836 if (check_sibling_keys(left, right)) {
2838 btrfs_tree_unlock(right);
2839 free_extent_buffer(right);
2842 if (path->slots[0] == left_nritems && !empty) {
2843 /* Key greater than all keys in the leaf, right neighbor has
2844 * enough room for it and we're not emptying our leaf to delete
2845 * it, therefore use right neighbor to insert the new item and
2846 * no need to touch/dirty our left leaf. */
2847 btrfs_tree_unlock(left);
2848 free_extent_buffer(left);
2849 path->nodes[0] = right;
2855 return __push_leaf_right(path, min_data_size, empty,
2856 right, free_space, left_nritems, min_slot);
2858 btrfs_tree_unlock(right);
2859 free_extent_buffer(right);
2864 * push some data in the path leaf to the left, trying to free up at
2865 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2867 * max_slot can put a limit on how far into the leaf we'll push items. The
2868 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
2871 static noinline int __push_leaf_left(struct btrfs_path *path, int data_size,
2872 int empty, struct extent_buffer *left,
2873 int free_space, u32 right_nritems,
2876 struct btrfs_fs_info *fs_info = left->fs_info;
2877 struct btrfs_disk_key disk_key;
2878 struct extent_buffer *right = path->nodes[0];
2882 struct btrfs_item *item;
2883 u32 old_left_nritems;
2887 u32 old_left_item_size;
2888 struct btrfs_map_token token;
2891 nr = min(right_nritems, max_slot);
2893 nr = min(right_nritems - 1, max_slot);
2895 for (i = 0; i < nr; i++) {
2896 item = btrfs_item_nr(i);
2898 if (!empty && push_items > 0) {
2899 if (path->slots[0] < i)
2901 if (path->slots[0] == i) {
2902 int space = btrfs_leaf_free_space(right);
2904 if (space + push_space * 2 > free_space)
2909 if (path->slots[0] == i)
2910 push_space += data_size;
2912 this_item_size = btrfs_item_size(right, item);
2913 if (this_item_size + sizeof(*item) + push_space > free_space)
2917 push_space += this_item_size + sizeof(*item);
2920 if (push_items == 0) {
2924 WARN_ON(!empty && push_items == btrfs_header_nritems(right));
2926 /* push data from right to left */
2927 copy_extent_buffer(left, right,
2928 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2929 btrfs_item_nr_offset(0),
2930 push_items * sizeof(struct btrfs_item));
2932 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) -
2933 btrfs_item_offset_nr(right, push_items - 1);
2935 copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET +
2936 leaf_data_end(left) - push_space,
2937 BTRFS_LEAF_DATA_OFFSET +
2938 btrfs_item_offset_nr(right, push_items - 1),
2940 old_left_nritems = btrfs_header_nritems(left);
2941 BUG_ON(old_left_nritems <= 0);
2943 btrfs_init_map_token(&token, left);
2944 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2945 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2948 item = btrfs_item_nr(i);
2950 ioff = btrfs_token_item_offset(&token, item);
2951 btrfs_set_token_item_offset(&token, item,
2952 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size));
2954 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2956 /* fixup right node */
2957 if (push_items > right_nritems)
2958 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
2961 if (push_items < right_nritems) {
2962 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2963 leaf_data_end(right);
2964 memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET +
2965 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space,
2966 BTRFS_LEAF_DATA_OFFSET +
2967 leaf_data_end(right), push_space);
2969 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2970 btrfs_item_nr_offset(push_items),
2971 (btrfs_header_nritems(right) - push_items) *
2972 sizeof(struct btrfs_item));
2975 btrfs_init_map_token(&token, right);
2976 right_nritems -= push_items;
2977 btrfs_set_header_nritems(right, right_nritems);
2978 push_space = BTRFS_LEAF_DATA_SIZE(fs_info);
2979 for (i = 0; i < right_nritems; i++) {
2980 item = btrfs_item_nr(i);
2982 push_space = push_space - btrfs_token_item_size(&token, item);
2983 btrfs_set_token_item_offset(&token, item, push_space);
2986 btrfs_mark_buffer_dirty(left);
2988 btrfs_mark_buffer_dirty(right);
2990 btrfs_clean_tree_block(right);
2992 btrfs_item_key(right, &disk_key, 0);
2993 fixup_low_keys(path, &disk_key, 1);
2995 /* then fixup the leaf pointer in the path */
2996 if (path->slots[0] < push_items) {
2997 path->slots[0] += old_left_nritems;
2998 btrfs_tree_unlock(path->nodes[0]);
2999 free_extent_buffer(path->nodes[0]);
3000 path->nodes[0] = left;
3001 path->slots[1] -= 1;
3003 btrfs_tree_unlock(left);
3004 free_extent_buffer(left);
3005 path->slots[0] -= push_items;
3007 BUG_ON(path->slots[0] < 0);
3010 btrfs_tree_unlock(left);
3011 free_extent_buffer(left);
3016 * push some data in the path leaf to the left, trying to free up at
3017 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3019 * max_slot can put a limit on how far into the leaf we'll push items. The
3020 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3023 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3024 *root, struct btrfs_path *path, int min_data_size,
3025 int data_size, int empty, u32 max_slot)
3027 struct extent_buffer *right = path->nodes[0];
3028 struct extent_buffer *left;
3034 slot = path->slots[1];
3037 if (!path->nodes[1])
3040 right_nritems = btrfs_header_nritems(right);
3041 if (right_nritems == 0)
3044 btrfs_assert_tree_locked(path->nodes[1]);
3046 left = btrfs_read_node_slot(path->nodes[1], slot - 1);
3048 * slot - 1 is not valid or we fail to read the left node,
3049 * no big deal, just return.
3054 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT);
3056 free_space = btrfs_leaf_free_space(left);
3057 if (free_space < data_size) {
3062 /* cow and double check */
3063 ret = btrfs_cow_block(trans, root, left,
3064 path->nodes[1], slot - 1, &left,
3065 BTRFS_NESTING_LEFT_COW);
3067 /* we hit -ENOSPC, but it isn't fatal here */
3073 free_space = btrfs_leaf_free_space(left);
3074 if (free_space < data_size) {
3079 if (check_sibling_keys(left, right)) {
3083 return __push_leaf_left(path, min_data_size,
3084 empty, left, free_space, right_nritems,
3087 btrfs_tree_unlock(left);
3088 free_extent_buffer(left);
3093 * split the path's leaf in two, making sure there is at least data_size
3094 * available for the resulting leaf level of the path.
3096 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
3097 struct btrfs_path *path,
3098 struct extent_buffer *l,
3099 struct extent_buffer *right,
3100 int slot, int mid, int nritems)
3102 struct btrfs_fs_info *fs_info = trans->fs_info;
3106 struct btrfs_disk_key disk_key;
3107 struct btrfs_map_token token;
3109 nritems = nritems - mid;
3110 btrfs_set_header_nritems(right, nritems);
3111 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l);
3113 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
3114 btrfs_item_nr_offset(mid),
3115 nritems * sizeof(struct btrfs_item));
3117 copy_extent_buffer(right, l,
3118 BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) -
3119 data_copy_size, BTRFS_LEAF_DATA_OFFSET +
3120 leaf_data_end(l), data_copy_size);
3122 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid);
3124 btrfs_init_map_token(&token, right);
3125 for (i = 0; i < nritems; i++) {
3126 struct btrfs_item *item = btrfs_item_nr(i);
3129 ioff = btrfs_token_item_offset(&token, item);
3130 btrfs_set_token_item_offset(&token, item, ioff + rt_data_off);
3133 btrfs_set_header_nritems(l, mid);
3134 btrfs_item_key(right, &disk_key, 0);
3135 insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1);
3137 btrfs_mark_buffer_dirty(right);
3138 btrfs_mark_buffer_dirty(l);
3139 BUG_ON(path->slots[0] != slot);
3142 btrfs_tree_unlock(path->nodes[0]);
3143 free_extent_buffer(path->nodes[0]);
3144 path->nodes[0] = right;
3145 path->slots[0] -= mid;
3146 path->slots[1] += 1;
3148 btrfs_tree_unlock(right);
3149 free_extent_buffer(right);
3152 BUG_ON(path->slots[0] < 0);
3156 * double splits happen when we need to insert a big item in the middle
3157 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3158 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3161 * We avoid this by trying to push the items on either side of our target
3162 * into the adjacent leaves. If all goes well we can avoid the double split
3165 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
3166 struct btrfs_root *root,
3167 struct btrfs_path *path,
3174 int space_needed = data_size;
3176 slot = path->slots[0];
3177 if (slot < btrfs_header_nritems(path->nodes[0]))
3178 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
3181 * try to push all the items after our slot into the
3184 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot);
3191 nritems = btrfs_header_nritems(path->nodes[0]);
3193 * our goal is to get our slot at the start or end of a leaf. If
3194 * we've done so we're done
3196 if (path->slots[0] == 0 || path->slots[0] == nritems)
3199 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
3202 /* try to push all the items before our slot into the next leaf */
3203 slot = path->slots[0];
3204 space_needed = data_size;
3206 space_needed -= btrfs_leaf_free_space(path->nodes[0]);
3207 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot);
3220 * split the path's leaf in two, making sure there is at least data_size
3221 * available for the resulting leaf level of the path.
3223 * returns 0 if all went well and < 0 on failure.
3225 static noinline int split_leaf(struct btrfs_trans_handle *trans,
3226 struct btrfs_root *root,
3227 const struct btrfs_key *ins_key,
3228 struct btrfs_path *path, int data_size,
3231 struct btrfs_disk_key disk_key;
3232 struct extent_buffer *l;
3236 struct extent_buffer *right;
3237 struct btrfs_fs_info *fs_info = root->fs_info;
3241 int num_doubles = 0;
3242 int tried_avoid_double = 0;
3245 slot = path->slots[0];
3246 if (extend && data_size + btrfs_item_size_nr(l, slot) +
3247 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info))
3250 /* first try to make some room by pushing left and right */
3251 if (data_size && path->nodes[1]) {
3252 int space_needed = data_size;
3254 if (slot < btrfs_header_nritems(l))
3255 space_needed -= btrfs_leaf_free_space(l);
3257 wret = push_leaf_right(trans, root, path, space_needed,
3258 space_needed, 0, 0);
3262 space_needed = data_size;
3264 space_needed -= btrfs_leaf_free_space(l);
3265 wret = push_leaf_left(trans, root, path, space_needed,
3266 space_needed, 0, (u32)-1);
3272 /* did the pushes work? */
3273 if (btrfs_leaf_free_space(l) >= data_size)
3277 if (!path->nodes[1]) {
3278 ret = insert_new_root(trans, root, path, 1);
3285 slot = path->slots[0];
3286 nritems = btrfs_header_nritems(l);
3287 mid = (nritems + 1) / 2;
3291 leaf_space_used(l, mid, nritems - mid) + data_size >
3292 BTRFS_LEAF_DATA_SIZE(fs_info)) {
3293 if (slot >= nritems) {
3297 if (mid != nritems &&
3298 leaf_space_used(l, mid, nritems - mid) +
3299 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
3300 if (data_size && !tried_avoid_double)
3301 goto push_for_double;
3307 if (leaf_space_used(l, 0, mid) + data_size >
3308 BTRFS_LEAF_DATA_SIZE(fs_info)) {
3309 if (!extend && data_size && slot == 0) {
3311 } else if ((extend || !data_size) && slot == 0) {
3315 if (mid != nritems &&
3316 leaf_space_used(l, mid, nritems - mid) +
3317 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) {
3318 if (data_size && !tried_avoid_double)
3319 goto push_for_double;
3327 btrfs_cpu_key_to_disk(&disk_key, ins_key);
3329 btrfs_item_key(l, &disk_key, mid);
3332 * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double
3333 * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES
3334 * subclasses, which is 8 at the time of this patch, and we've maxed it
3335 * out. In the future we could add a
3336 * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just
3337 * use BTRFS_NESTING_NEW_ROOT.
3339 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid,
3340 &disk_key, 0, l->start, 0,
3341 num_doubles ? BTRFS_NESTING_NEW_ROOT :
3342 BTRFS_NESTING_SPLIT);
3344 return PTR_ERR(right);
3346 root_add_used(root, fs_info->nodesize);
3350 btrfs_set_header_nritems(right, 0);
3351 insert_ptr(trans, path, &disk_key,
3352 right->start, path->slots[1] + 1, 1);
3353 btrfs_tree_unlock(path->nodes[0]);
3354 free_extent_buffer(path->nodes[0]);
3355 path->nodes[0] = right;
3357 path->slots[1] += 1;
3359 btrfs_set_header_nritems(right, 0);
3360 insert_ptr(trans, path, &disk_key,
3361 right->start, path->slots[1], 1);
3362 btrfs_tree_unlock(path->nodes[0]);
3363 free_extent_buffer(path->nodes[0]);
3364 path->nodes[0] = right;
3366 if (path->slots[1] == 0)
3367 fixup_low_keys(path, &disk_key, 1);
3370 * We create a new leaf 'right' for the required ins_len and
3371 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
3372 * the content of ins_len to 'right'.
3377 copy_for_split(trans, path, l, right, slot, mid, nritems);
3380 BUG_ON(num_doubles != 0);
3388 push_for_double_split(trans, root, path, data_size);
3389 tried_avoid_double = 1;
3390 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size)
3395 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
3396 struct btrfs_root *root,
3397 struct btrfs_path *path, int ins_len)
3399 struct btrfs_key key;
3400 struct extent_buffer *leaf;
3401 struct btrfs_file_extent_item *fi;
3406 leaf = path->nodes[0];
3407 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3409 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
3410 key.type != BTRFS_EXTENT_CSUM_KEY);
3412 if (btrfs_leaf_free_space(leaf) >= ins_len)
3415 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3416 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3417 fi = btrfs_item_ptr(leaf, path->slots[0],
3418 struct btrfs_file_extent_item);
3419 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
3421 btrfs_release_path(path);
3423 path->keep_locks = 1;
3424 path->search_for_split = 1;
3425 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3426 path->search_for_split = 0;
3433 leaf = path->nodes[0];
3434 /* if our item isn't there, return now */
3435 if (item_size != btrfs_item_size_nr(leaf, path->slots[0]))
3438 /* the leaf has changed, it now has room. return now */
3439 if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len)
3442 if (key.type == BTRFS_EXTENT_DATA_KEY) {
3443 fi = btrfs_item_ptr(leaf, path->slots[0],
3444 struct btrfs_file_extent_item);
3445 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
3449 ret = split_leaf(trans, root, &key, path, ins_len, 1);
3453 path->keep_locks = 0;
3454 btrfs_unlock_up_safe(path, 1);
3457 path->keep_locks = 0;
3461 static noinline int split_item(struct btrfs_path *path,
3462 const struct btrfs_key *new_key,
3463 unsigned long split_offset)
3465 struct extent_buffer *leaf;
3466 struct btrfs_item *item;
3467 struct btrfs_item *new_item;
3473 struct btrfs_disk_key disk_key;
3475 leaf = path->nodes[0];
3476 BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item));
3478 item = btrfs_item_nr(path->slots[0]);
3479 orig_offset = btrfs_item_offset(leaf, item);
3480 item_size = btrfs_item_size(leaf, item);
3482 buf = kmalloc(item_size, GFP_NOFS);
3486 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3487 path->slots[0]), item_size);
3489 slot = path->slots[0] + 1;
3490 nritems = btrfs_header_nritems(leaf);
3491 if (slot != nritems) {
3492 /* shift the items */
3493 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3494 btrfs_item_nr_offset(slot),
3495 (nritems - slot) * sizeof(struct btrfs_item));
3498 btrfs_cpu_key_to_disk(&disk_key, new_key);
3499 btrfs_set_item_key(leaf, &disk_key, slot);
3501 new_item = btrfs_item_nr(slot);
3503 btrfs_set_item_offset(leaf, new_item, orig_offset);
3504 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3506 btrfs_set_item_offset(leaf, item,
3507 orig_offset + item_size - split_offset);
3508 btrfs_set_item_size(leaf, item, split_offset);
3510 btrfs_set_header_nritems(leaf, nritems + 1);
3512 /* write the data for the start of the original item */
3513 write_extent_buffer(leaf, buf,
3514 btrfs_item_ptr_offset(leaf, path->slots[0]),
3517 /* write the data for the new item */
3518 write_extent_buffer(leaf, buf + split_offset,
3519 btrfs_item_ptr_offset(leaf, slot),
3520 item_size - split_offset);
3521 btrfs_mark_buffer_dirty(leaf);
3523 BUG_ON(btrfs_leaf_free_space(leaf) < 0);
3529 * This function splits a single item into two items,
3530 * giving 'new_key' to the new item and splitting the
3531 * old one at split_offset (from the start of the item).
3533 * The path may be released by this operation. After
3534 * the split, the path is pointing to the old item. The
3535 * new item is going to be in the same node as the old one.
3537 * Note, the item being split must be smaller enough to live alone on
3538 * a tree block with room for one extra struct btrfs_item
3540 * This allows us to split the item in place, keeping a lock on the
3541 * leaf the entire time.
3543 int btrfs_split_item(struct btrfs_trans_handle *trans,
3544 struct btrfs_root *root,
3545 struct btrfs_path *path,
3546 const struct btrfs_key *new_key,
3547 unsigned long split_offset)
3550 ret = setup_leaf_for_split(trans, root, path,
3551 sizeof(struct btrfs_item));
3555 ret = split_item(path, new_key, split_offset);
3560 * This function duplicate a item, giving 'new_key' to the new item.
3561 * It guarantees both items live in the same tree leaf and the new item
3562 * is contiguous with the original item.
3564 * This allows us to split file extent in place, keeping a lock on the
3565 * leaf the entire time.
3567 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
3568 struct btrfs_root *root,
3569 struct btrfs_path *path,
3570 const struct btrfs_key *new_key)
3572 struct extent_buffer *leaf;
3576 leaf = path->nodes[0];
3577 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3578 ret = setup_leaf_for_split(trans, root, path,
3579 item_size + sizeof(struct btrfs_item));
3584 setup_items_for_insert(root, path, new_key, &item_size, 1);
3585 leaf = path->nodes[0];
3586 memcpy_extent_buffer(leaf,
3587 btrfs_item_ptr_offset(leaf, path->slots[0]),
3588 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
3594 * make the item pointed to by the path smaller. new_size indicates
3595 * how small to make it, and from_end tells us if we just chop bytes
3596 * off the end of the item or if we shift the item to chop bytes off
3599 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end)
3602 struct extent_buffer *leaf;
3603 struct btrfs_item *item;
3605 unsigned int data_end;
3606 unsigned int old_data_start;
3607 unsigned int old_size;
3608 unsigned int size_diff;
3610 struct btrfs_map_token token;
3612 leaf = path->nodes[0];
3613 slot = path->slots[0];
3615 old_size = btrfs_item_size_nr(leaf, slot);
3616 if (old_size == new_size)
3619 nritems = btrfs_header_nritems(leaf);
3620 data_end = leaf_data_end(leaf);
3622 old_data_start = btrfs_item_offset_nr(leaf, slot);
3624 size_diff = old_size - new_size;
3627 BUG_ON(slot >= nritems);
3630 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3632 /* first correct the data pointers */
3633 btrfs_init_map_token(&token, leaf);
3634 for (i = slot; i < nritems; i++) {
3636 item = btrfs_item_nr(i);
3638 ioff = btrfs_token_item_offset(&token, item);
3639 btrfs_set_token_item_offset(&token, item, ioff + size_diff);
3642 /* shift the data */
3644 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
3645 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
3646 data_end, old_data_start + new_size - data_end);
3648 struct btrfs_disk_key disk_key;
3651 btrfs_item_key(leaf, &disk_key, slot);
3653 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3655 struct btrfs_file_extent_item *fi;
3657 fi = btrfs_item_ptr(leaf, slot,
3658 struct btrfs_file_extent_item);
3659 fi = (struct btrfs_file_extent_item *)(
3660 (unsigned long)fi - size_diff);
3662 if (btrfs_file_extent_type(leaf, fi) ==
3663 BTRFS_FILE_EXTENT_INLINE) {
3664 ptr = btrfs_item_ptr_offset(leaf, slot);
3665 memmove_extent_buffer(leaf, ptr,
3667 BTRFS_FILE_EXTENT_INLINE_DATA_START);
3671 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
3672 data_end + size_diff, BTRFS_LEAF_DATA_OFFSET +
3673 data_end, old_data_start - data_end);
3675 offset = btrfs_disk_key_offset(&disk_key);
3676 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3677 btrfs_set_item_key(leaf, &disk_key, slot);
3679 fixup_low_keys(path, &disk_key, 1);
3682 item = btrfs_item_nr(slot);
3683 btrfs_set_item_size(leaf, item, new_size);
3684 btrfs_mark_buffer_dirty(leaf);
3686 if (btrfs_leaf_free_space(leaf) < 0) {
3687 btrfs_print_leaf(leaf);
3693 * make the item pointed to by the path bigger, data_size is the added size.
3695 void btrfs_extend_item(struct btrfs_path *path, u32 data_size)
3698 struct extent_buffer *leaf;
3699 struct btrfs_item *item;
3701 unsigned int data_end;
3702 unsigned int old_data;
3703 unsigned int old_size;
3705 struct btrfs_map_token token;
3707 leaf = path->nodes[0];
3709 nritems = btrfs_header_nritems(leaf);
3710 data_end = leaf_data_end(leaf);
3712 if (btrfs_leaf_free_space(leaf) < data_size) {
3713 btrfs_print_leaf(leaf);
3716 slot = path->slots[0];
3717 old_data = btrfs_item_end_nr(leaf, slot);
3720 if (slot >= nritems) {
3721 btrfs_print_leaf(leaf);
3722 btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d",
3728 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3730 /* first correct the data pointers */
3731 btrfs_init_map_token(&token, leaf);
3732 for (i = slot; i < nritems; i++) {
3734 item = btrfs_item_nr(i);
3736 ioff = btrfs_token_item_offset(&token, item);
3737 btrfs_set_token_item_offset(&token, item, ioff - data_size);
3740 /* shift the data */
3741 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
3742 data_end - data_size, BTRFS_LEAF_DATA_OFFSET +
3743 data_end, old_data - data_end);
3745 data_end = old_data;
3746 old_size = btrfs_item_size_nr(leaf, slot);
3747 item = btrfs_item_nr(slot);
3748 btrfs_set_item_size(leaf, item, old_size + data_size);
3749 btrfs_mark_buffer_dirty(leaf);
3751 if (btrfs_leaf_free_space(leaf) < 0) {
3752 btrfs_print_leaf(leaf);
3758 * setup_items_for_insert - Helper called before inserting one or more items
3759 * to a leaf. Main purpose is to save stack depth by doing the bulk of the work
3760 * in a function that doesn't call btrfs_search_slot
3762 * @root: root we are inserting items to
3763 * @path: points to the leaf/slot where we are going to insert new items
3764 * @cpu_key: array of keys for items to be inserted
3765 * @data_size: size of the body of each item we are going to insert
3766 * @nr: size of @cpu_key/@data_size arrays
3768 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
3769 const struct btrfs_key *cpu_key, u32 *data_size,
3772 struct btrfs_fs_info *fs_info = root->fs_info;
3773 struct btrfs_item *item;
3776 unsigned int data_end;
3777 struct btrfs_disk_key disk_key;
3778 struct extent_buffer *leaf;
3780 struct btrfs_map_token token;
3784 for (i = 0; i < nr; i++)
3785 total_data += data_size[i];
3786 total_size = total_data + (nr * sizeof(struct btrfs_item));
3788 if (path->slots[0] == 0) {
3789 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3790 fixup_low_keys(path, &disk_key, 1);
3792 btrfs_unlock_up_safe(path, 1);
3794 leaf = path->nodes[0];
3795 slot = path->slots[0];
3797 nritems = btrfs_header_nritems(leaf);
3798 data_end = leaf_data_end(leaf);
3800 if (btrfs_leaf_free_space(leaf) < total_size) {
3801 btrfs_print_leaf(leaf);
3802 btrfs_crit(fs_info, "not enough freespace need %u have %d",
3803 total_size, btrfs_leaf_free_space(leaf));
3807 btrfs_init_map_token(&token, leaf);
3808 if (slot != nritems) {
3809 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3811 if (old_data < data_end) {
3812 btrfs_print_leaf(leaf);
3814 "item at slot %d with data offset %u beyond data end of leaf %u",
3815 slot, old_data, data_end);
3819 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3821 /* first correct the data pointers */
3822 for (i = slot; i < nritems; i++) {
3825 item = btrfs_item_nr(i);
3826 ioff = btrfs_token_item_offset(&token, item);
3827 btrfs_set_token_item_offset(&token, item,
3830 /* shift the items */
3831 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3832 btrfs_item_nr_offset(slot),
3833 (nritems - slot) * sizeof(struct btrfs_item));
3835 /* shift the data */
3836 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
3837 data_end - total_data, BTRFS_LEAF_DATA_OFFSET +
3838 data_end, old_data - data_end);
3839 data_end = old_data;
3842 /* setup the item for the new data */
3843 for (i = 0; i < nr; i++) {
3844 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3845 btrfs_set_item_key(leaf, &disk_key, slot + i);
3846 item = btrfs_item_nr(slot + i);
3847 data_end -= data_size[i];
3848 btrfs_set_token_item_offset(&token, item, data_end);
3849 btrfs_set_token_item_size(&token, item, data_size[i]);
3852 btrfs_set_header_nritems(leaf, nritems + nr);
3853 btrfs_mark_buffer_dirty(leaf);
3855 if (btrfs_leaf_free_space(leaf) < 0) {
3856 btrfs_print_leaf(leaf);
3862 * Given a key and some data, insert items into the tree.
3863 * This does all the path init required, making room in the tree if needed.
3865 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3866 struct btrfs_root *root,
3867 struct btrfs_path *path,
3868 const struct btrfs_key *cpu_key, u32 *data_size,
3877 for (i = 0; i < nr; i++)
3878 total_data += data_size[i];
3880 total_size = total_data + (nr * sizeof(struct btrfs_item));
3881 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3887 slot = path->slots[0];
3890 setup_items_for_insert(root, path, cpu_key, data_size, nr);
3895 * Given a key and some data, insert an item into the tree.
3896 * This does all the path init required, making room in the tree if needed.
3898 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3899 const struct btrfs_key *cpu_key, void *data,
3903 struct btrfs_path *path;
3904 struct extent_buffer *leaf;
3907 path = btrfs_alloc_path();
3910 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3912 leaf = path->nodes[0];
3913 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3914 write_extent_buffer(leaf, data, ptr, data_size);
3915 btrfs_mark_buffer_dirty(leaf);
3917 btrfs_free_path(path);
3922 * delete the pointer from a given node.
3924 * the tree should have been previously balanced so the deletion does not
3927 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
3928 int level, int slot)
3930 struct extent_buffer *parent = path->nodes[level];
3934 nritems = btrfs_header_nritems(parent);
3935 if (slot != nritems - 1) {
3937 ret = btrfs_tree_mod_log_insert_move(parent, slot,
3938 slot + 1, nritems - slot - 1);
3941 memmove_extent_buffer(parent,
3942 btrfs_node_key_ptr_offset(slot),
3943 btrfs_node_key_ptr_offset(slot + 1),
3944 sizeof(struct btrfs_key_ptr) *
3945 (nritems - slot - 1));
3947 ret = btrfs_tree_mod_log_insert_key(parent, slot,
3948 BTRFS_MOD_LOG_KEY_REMOVE, GFP_NOFS);
3953 btrfs_set_header_nritems(parent, nritems);
3954 if (nritems == 0 && parent == root->node) {
3955 BUG_ON(btrfs_header_level(root->node) != 1);
3956 /* just turn the root into a leaf and break */
3957 btrfs_set_header_level(root->node, 0);
3958 } else if (slot == 0) {
3959 struct btrfs_disk_key disk_key;
3961 btrfs_node_key(parent, &disk_key, 0);
3962 fixup_low_keys(path, &disk_key, level + 1);
3964 btrfs_mark_buffer_dirty(parent);
3968 * a helper function to delete the leaf pointed to by path->slots[1] and
3971 * This deletes the pointer in path->nodes[1] and frees the leaf
3972 * block extent. zero is returned if it all worked out, < 0 otherwise.
3974 * The path must have already been setup for deleting the leaf, including
3975 * all the proper balancing. path->nodes[1] must be locked.
3977 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
3978 struct btrfs_root *root,
3979 struct btrfs_path *path,
3980 struct extent_buffer *leaf)
3982 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
3983 del_ptr(root, path, 1, path->slots[1]);
3986 * btrfs_free_extent is expensive, we want to make sure we
3987 * aren't holding any locks when we call it
3989 btrfs_unlock_up_safe(path, 0);
3991 root_sub_used(root, leaf->len);
3993 atomic_inc(&leaf->refs);
3994 btrfs_free_tree_block(trans, root, leaf, 0, 1);
3995 free_extent_buffer_stale(leaf);
3998 * delete the item at the leaf level in path. If that empties
3999 * the leaf, remove it from the tree
4001 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4002 struct btrfs_path *path, int slot, int nr)
4004 struct btrfs_fs_info *fs_info = root->fs_info;
4005 struct extent_buffer *leaf;
4006 struct btrfs_item *item;
4014 leaf = path->nodes[0];
4015 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4017 for (i = 0; i < nr; i++)
4018 dsize += btrfs_item_size_nr(leaf, slot + i);
4020 nritems = btrfs_header_nritems(leaf);
4022 if (slot + nr != nritems) {
4023 int data_end = leaf_data_end(leaf);
4024 struct btrfs_map_token token;
4026 memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET +
4028 BTRFS_LEAF_DATA_OFFSET + data_end,
4029 last_off - data_end);
4031 btrfs_init_map_token(&token, leaf);
4032 for (i = slot + nr; i < nritems; i++) {
4035 item = btrfs_item_nr(i);
4036 ioff = btrfs_token_item_offset(&token, item);
4037 btrfs_set_token_item_offset(&token, item, ioff + dsize);
4040 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4041 btrfs_item_nr_offset(slot + nr),
4042 sizeof(struct btrfs_item) *
4043 (nritems - slot - nr));
4045 btrfs_set_header_nritems(leaf, nritems - nr);
4048 /* delete the leaf if we've emptied it */
4050 if (leaf == root->node) {
4051 btrfs_set_header_level(leaf, 0);
4053 btrfs_clean_tree_block(leaf);
4054 btrfs_del_leaf(trans, root, path, leaf);
4057 int used = leaf_space_used(leaf, 0, nritems);
4059 struct btrfs_disk_key disk_key;
4061 btrfs_item_key(leaf, &disk_key, 0);
4062 fixup_low_keys(path, &disk_key, 1);
4065 /* delete the leaf if it is mostly empty */
4066 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) {
4067 /* push_leaf_left fixes the path.
4068 * make sure the path still points to our leaf
4069 * for possible call to del_ptr below
4071 slot = path->slots[1];
4072 atomic_inc(&leaf->refs);
4074 wret = push_leaf_left(trans, root, path, 1, 1,
4076 if (wret < 0 && wret != -ENOSPC)
4079 if (path->nodes[0] == leaf &&
4080 btrfs_header_nritems(leaf)) {
4081 wret = push_leaf_right(trans, root, path, 1,
4083 if (wret < 0 && wret != -ENOSPC)
4087 if (btrfs_header_nritems(leaf) == 0) {
4088 path->slots[1] = slot;
4089 btrfs_del_leaf(trans, root, path, leaf);
4090 free_extent_buffer(leaf);
4093 /* if we're still in the path, make sure
4094 * we're dirty. Otherwise, one of the
4095 * push_leaf functions must have already
4096 * dirtied this buffer
4098 if (path->nodes[0] == leaf)
4099 btrfs_mark_buffer_dirty(leaf);
4100 free_extent_buffer(leaf);
4103 btrfs_mark_buffer_dirty(leaf);
4110 * search the tree again to find a leaf with lesser keys
4111 * returns 0 if it found something or 1 if there are no lesser leaves.
4112 * returns < 0 on io errors.
4114 * This may release the path, and so you may lose any locks held at the
4117 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
4119 struct btrfs_key key;
4120 struct btrfs_disk_key found_key;
4123 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
4125 if (key.offset > 0) {
4127 } else if (key.type > 0) {
4129 key.offset = (u64)-1;
4130 } else if (key.objectid > 0) {
4133 key.offset = (u64)-1;
4138 btrfs_release_path(path);
4139 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4142 btrfs_item_key(path->nodes[0], &found_key, 0);
4143 ret = comp_keys(&found_key, &key);
4145 * We might have had an item with the previous key in the tree right
4146 * before we released our path. And after we released our path, that
4147 * item might have been pushed to the first slot (0) of the leaf we
4148 * were holding due to a tree balance. Alternatively, an item with the
4149 * previous key can exist as the only element of a leaf (big fat item).
4150 * Therefore account for these 2 cases, so that our callers (like
4151 * btrfs_previous_item) don't miss an existing item with a key matching
4152 * the previous key we computed above.
4160 * A helper function to walk down the tree starting at min_key, and looking
4161 * for nodes or leaves that are have a minimum transaction id.
4162 * This is used by the btree defrag code, and tree logging
4164 * This does not cow, but it does stuff the starting key it finds back
4165 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4166 * key and get a writable path.
4168 * This honors path->lowest_level to prevent descent past a given level
4171 * min_trans indicates the oldest transaction that you are interested
4172 * in walking through. Any nodes or leaves older than min_trans are
4173 * skipped over (without reading them).
4175 * returns zero if something useful was found, < 0 on error and 1 if there
4176 * was nothing in the tree that matched the search criteria.
4178 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
4179 struct btrfs_path *path,
4182 struct extent_buffer *cur;
4183 struct btrfs_key found_key;
4189 int keep_locks = path->keep_locks;
4191 path->keep_locks = 1;
4193 cur = btrfs_read_lock_root_node(root);
4194 level = btrfs_header_level(cur);
4195 WARN_ON(path->nodes[level]);
4196 path->nodes[level] = cur;
4197 path->locks[level] = BTRFS_READ_LOCK;
4199 if (btrfs_header_generation(cur) < min_trans) {
4204 nritems = btrfs_header_nritems(cur);
4205 level = btrfs_header_level(cur);
4206 sret = btrfs_bin_search(cur, min_key, &slot);
4212 /* at the lowest level, we're done, setup the path and exit */
4213 if (level == path->lowest_level) {
4214 if (slot >= nritems)
4217 path->slots[level] = slot;
4218 btrfs_item_key_to_cpu(cur, &found_key, slot);
4221 if (sret && slot > 0)
4224 * check this node pointer against the min_trans parameters.
4225 * If it is too old, skip to the next one.
4227 while (slot < nritems) {
4230 gen = btrfs_node_ptr_generation(cur, slot);
4231 if (gen < min_trans) {
4239 * we didn't find a candidate key in this node, walk forward
4240 * and find another one
4242 if (slot >= nritems) {
4243 path->slots[level] = slot;
4244 sret = btrfs_find_next_key(root, path, min_key, level,
4247 btrfs_release_path(path);
4253 /* save our key for returning back */
4254 btrfs_node_key_to_cpu(cur, &found_key, slot);
4255 path->slots[level] = slot;
4256 if (level == path->lowest_level) {
4260 cur = btrfs_read_node_slot(cur, slot);
4266 btrfs_tree_read_lock(cur);
4268 path->locks[level - 1] = BTRFS_READ_LOCK;
4269 path->nodes[level - 1] = cur;
4270 unlock_up(path, level, 1, 0, NULL);
4273 path->keep_locks = keep_locks;
4275 btrfs_unlock_up_safe(path, path->lowest_level + 1);
4276 memcpy(min_key, &found_key, sizeof(found_key));
4282 * this is similar to btrfs_next_leaf, but does not try to preserve
4283 * and fixup the path. It looks for and returns the next key in the
4284 * tree based on the current path and the min_trans parameters.
4286 * 0 is returned if another key is found, < 0 if there are any errors
4287 * and 1 is returned if there are no higher keys in the tree
4289 * path->keep_locks should be set to 1 on the search made before
4290 * calling this function.
4292 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
4293 struct btrfs_key *key, int level, u64 min_trans)
4296 struct extent_buffer *c;
4298 WARN_ON(!path->keep_locks && !path->skip_locking);
4299 while (level < BTRFS_MAX_LEVEL) {
4300 if (!path->nodes[level])
4303 slot = path->slots[level] + 1;
4304 c = path->nodes[level];
4306 if (slot >= btrfs_header_nritems(c)) {
4309 struct btrfs_key cur_key;
4310 if (level + 1 >= BTRFS_MAX_LEVEL ||
4311 !path->nodes[level + 1])
4314 if (path->locks[level + 1] || path->skip_locking) {
4319 slot = btrfs_header_nritems(c) - 1;
4321 btrfs_item_key_to_cpu(c, &cur_key, slot);
4323 btrfs_node_key_to_cpu(c, &cur_key, slot);
4325 orig_lowest = path->lowest_level;
4326 btrfs_release_path(path);
4327 path->lowest_level = level;
4328 ret = btrfs_search_slot(NULL, root, &cur_key, path,
4330 path->lowest_level = orig_lowest;
4334 c = path->nodes[level];
4335 slot = path->slots[level];
4342 btrfs_item_key_to_cpu(c, key, slot);
4344 u64 gen = btrfs_node_ptr_generation(c, slot);
4346 if (gen < min_trans) {
4350 btrfs_node_key_to_cpu(c, key, slot);
4358 * search the tree again to find a leaf with greater keys
4359 * returns 0 if it found something or 1 if there are no greater leaves.
4360 * returns < 0 on io errors.
4362 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4364 return btrfs_next_old_leaf(root, path, 0);
4367 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
4372 struct extent_buffer *c;
4373 struct extent_buffer *next;
4374 struct btrfs_key key;
4379 nritems = btrfs_header_nritems(path->nodes[0]);
4383 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4387 btrfs_release_path(path);
4389 path->keep_locks = 1;
4392 ret = btrfs_search_old_slot(root, &key, path, time_seq);
4394 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4395 path->keep_locks = 0;
4400 nritems = btrfs_header_nritems(path->nodes[0]);
4402 * by releasing the path above we dropped all our locks. A balance
4403 * could have added more items next to the key that used to be
4404 * at the very end of the block. So, check again here and
4405 * advance the path if there are now more items available.
4407 if (nritems > 0 && path->slots[0] < nritems - 1) {
4414 * So the above check misses one case:
4415 * - after releasing the path above, someone has removed the item that
4416 * used to be at the very end of the block, and balance between leafs
4417 * gets another one with bigger key.offset to replace it.
4419 * This one should be returned as well, or we can get leaf corruption
4420 * later(esp. in __btrfs_drop_extents()).
4422 * And a bit more explanation about this check,
4423 * with ret > 0, the key isn't found, the path points to the slot
4424 * where it should be inserted, so the path->slots[0] item must be the
4427 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) {
4432 while (level < BTRFS_MAX_LEVEL) {
4433 if (!path->nodes[level]) {
4438 slot = path->slots[level] + 1;
4439 c = path->nodes[level];
4440 if (slot >= btrfs_header_nritems(c)) {
4442 if (level == BTRFS_MAX_LEVEL) {
4451 * Our current level is where we're going to start from, and to
4452 * make sure lockdep doesn't complain we need to drop our locks
4453 * and nodes from 0 to our current level.
4455 for (i = 0; i < level; i++) {
4456 if (path->locks[level]) {
4457 btrfs_tree_read_unlock(path->nodes[i]);
4460 free_extent_buffer(path->nodes[i]);
4461 path->nodes[i] = NULL;
4465 ret = read_block_for_search(root, path, &next, level,
4471 btrfs_release_path(path);
4475 if (!path->skip_locking) {
4476 ret = btrfs_try_tree_read_lock(next);
4477 if (!ret && time_seq) {
4479 * If we don't get the lock, we may be racing
4480 * with push_leaf_left, holding that lock while
4481 * itself waiting for the leaf we've currently
4482 * locked. To solve this situation, we give up
4483 * on our lock and cycle.
4485 free_extent_buffer(next);
4486 btrfs_release_path(path);
4491 btrfs_tree_read_lock(next);
4495 path->slots[level] = slot;
4498 path->nodes[level] = next;
4499 path->slots[level] = 0;
4500 if (!path->skip_locking)
4501 path->locks[level] = BTRFS_READ_LOCK;
4505 ret = read_block_for_search(root, path, &next, level,
4511 btrfs_release_path(path);
4515 if (!path->skip_locking)
4516 btrfs_tree_read_lock(next);
4520 unlock_up(path, 0, 1, 0, NULL);
4526 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4527 * searching until it gets past min_objectid or finds an item of 'type'
4529 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4531 int btrfs_previous_item(struct btrfs_root *root,
4532 struct btrfs_path *path, u64 min_objectid,
4535 struct btrfs_key found_key;
4536 struct extent_buffer *leaf;
4541 if (path->slots[0] == 0) {
4542 ret = btrfs_prev_leaf(root, path);
4548 leaf = path->nodes[0];
4549 nritems = btrfs_header_nritems(leaf);
4552 if (path->slots[0] == nritems)
4555 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4556 if (found_key.objectid < min_objectid)
4558 if (found_key.type == type)
4560 if (found_key.objectid == min_objectid &&
4561 found_key.type < type)
4568 * search in extent tree to find a previous Metadata/Data extent item with
4571 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4573 int btrfs_previous_extent_item(struct btrfs_root *root,
4574 struct btrfs_path *path, u64 min_objectid)
4576 struct btrfs_key found_key;
4577 struct extent_buffer *leaf;
4582 if (path->slots[0] == 0) {
4583 ret = btrfs_prev_leaf(root, path);
4589 leaf = path->nodes[0];
4590 nritems = btrfs_header_nritems(leaf);
4593 if (path->slots[0] == nritems)
4596 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4597 if (found_key.objectid < min_objectid)
4599 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
4600 found_key.type == BTRFS_METADATA_ITEM_KEY)
4602 if (found_key.objectid == min_objectid &&
4603 found_key.type < BTRFS_EXTENT_ITEM_KEY)