1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
20 #include "extent-tree.h"
23 #include "print-tree.h"
27 #include "free-space-cache.h"
28 #include "free-space-tree.h"
31 #include "ref-verify.h"
32 #include "space-info.h"
33 #include "block-rsv.h"
34 #include "delalloc-space.h"
36 #include "rcu-string.h"
38 #include "dev-replace.h"
40 #include "accessors.h"
41 #include "root-tree.h"
42 #include "file-item.h"
44 #include "tree-checker.h"
45 #include "raid-stripe-tree.h"
47 #undef SCRAMBLE_DELAYED_REFS
50 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
51 struct btrfs_delayed_ref_head *href,
52 struct btrfs_delayed_ref_node *node, u64 parent,
53 u64 root_objectid, u64 owner_objectid,
55 struct btrfs_delayed_extent_op *extra_op);
56 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
57 struct extent_buffer *leaf,
58 struct btrfs_extent_item *ei);
59 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
60 u64 parent, u64 root_objectid,
61 u64 flags, u64 owner, u64 offset,
62 struct btrfs_key *ins, int ref_mod, u64 oref_root);
63 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
64 struct btrfs_delayed_ref_node *node,
65 struct btrfs_delayed_extent_op *extent_op);
66 static int find_next_key(struct btrfs_path *path, int level,
67 struct btrfs_key *key);
69 static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
71 return (cache->flags & bits) == bits;
74 /* simple helper to search for an existing data extent at a given offset */
75 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
77 struct btrfs_root *root = btrfs_extent_root(fs_info, start);
80 struct btrfs_path *path;
82 path = btrfs_alloc_path();
88 key.type = BTRFS_EXTENT_ITEM_KEY;
89 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
90 btrfs_free_path(path);
95 * helper function to lookup reference count and flags of a tree block.
97 * the head node for delayed ref is used to store the sum of all the
98 * reference count modifications queued up in the rbtree. the head
99 * node may also store the extent flags to set. This way you can check
100 * to see what the reference count and extent flags would be if all of
101 * the delayed refs are not processed.
103 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
104 struct btrfs_fs_info *fs_info, u64 bytenr,
105 u64 offset, int metadata, u64 *refs, u64 *flags,
108 struct btrfs_root *extent_root;
109 struct btrfs_delayed_ref_head *head;
110 struct btrfs_delayed_ref_root *delayed_refs;
111 struct btrfs_path *path;
112 struct btrfs_extent_item *ei;
113 struct extent_buffer *leaf;
114 struct btrfs_key key;
122 * If we don't have skinny metadata, don't bother doing anything
125 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
126 offset = fs_info->nodesize;
130 path = btrfs_alloc_path();
135 path->skip_locking = 1;
136 path->search_commit_root = 1;
140 key.objectid = bytenr;
143 key.type = BTRFS_METADATA_ITEM_KEY;
145 key.type = BTRFS_EXTENT_ITEM_KEY;
147 extent_root = btrfs_extent_root(fs_info, bytenr);
148 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
152 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
153 if (path->slots[0]) {
155 btrfs_item_key_to_cpu(path->nodes[0], &key,
157 if (key.objectid == bytenr &&
158 key.type == BTRFS_EXTENT_ITEM_KEY &&
159 key.offset == fs_info->nodesize)
165 leaf = path->nodes[0];
166 item_size = btrfs_item_size(leaf, path->slots[0]);
167 if (item_size >= sizeof(*ei)) {
168 ei = btrfs_item_ptr(leaf, path->slots[0],
169 struct btrfs_extent_item);
170 num_refs = btrfs_extent_refs(leaf, ei);
171 extent_flags = btrfs_extent_flags(leaf, ei);
172 owner = btrfs_get_extent_owner_root(fs_info, leaf,
177 "unexpected extent item size, has %u expect >= %zu",
178 item_size, sizeof(*ei));
180 btrfs_abort_transaction(trans, ret);
182 btrfs_handle_fs_error(fs_info, ret, NULL);
187 BUG_ON(num_refs == 0);
197 delayed_refs = &trans->transaction->delayed_refs;
198 spin_lock(&delayed_refs->lock);
199 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
201 if (!mutex_trylock(&head->mutex)) {
202 refcount_inc(&head->refs);
203 spin_unlock(&delayed_refs->lock);
205 btrfs_release_path(path);
208 * Mutex was contended, block until it's released and try
211 mutex_lock(&head->mutex);
212 mutex_unlock(&head->mutex);
213 btrfs_put_delayed_ref_head(head);
216 spin_lock(&head->lock);
217 if (head->extent_op && head->extent_op->update_flags)
218 extent_flags |= head->extent_op->flags_to_set;
220 BUG_ON(num_refs == 0);
222 num_refs += head->ref_mod;
223 spin_unlock(&head->lock);
224 mutex_unlock(&head->mutex);
226 spin_unlock(&delayed_refs->lock);
228 WARN_ON(num_refs == 0);
232 *flags = extent_flags;
234 *owning_root = owner;
236 btrfs_free_path(path);
241 * Back reference rules. Back refs have three main goals:
243 * 1) differentiate between all holders of references to an extent so that
244 * when a reference is dropped we can make sure it was a valid reference
245 * before freeing the extent.
247 * 2) Provide enough information to quickly find the holders of an extent
248 * if we notice a given block is corrupted or bad.
250 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
251 * maintenance. This is actually the same as #2, but with a slightly
252 * different use case.
254 * There are two kinds of back refs. The implicit back refs is optimized
255 * for pointers in non-shared tree blocks. For a given pointer in a block,
256 * back refs of this kind provide information about the block's owner tree
257 * and the pointer's key. These information allow us to find the block by
258 * b-tree searching. The full back refs is for pointers in tree blocks not
259 * referenced by their owner trees. The location of tree block is recorded
260 * in the back refs. Actually the full back refs is generic, and can be
261 * used in all cases the implicit back refs is used. The major shortcoming
262 * of the full back refs is its overhead. Every time a tree block gets
263 * COWed, we have to update back refs entry for all pointers in it.
265 * For a newly allocated tree block, we use implicit back refs for
266 * pointers in it. This means most tree related operations only involve
267 * implicit back refs. For a tree block created in old transaction, the
268 * only way to drop a reference to it is COW it. So we can detect the
269 * event that tree block loses its owner tree's reference and do the
270 * back refs conversion.
272 * When a tree block is COWed through a tree, there are four cases:
274 * The reference count of the block is one and the tree is the block's
275 * owner tree. Nothing to do in this case.
277 * The reference count of the block is one and the tree is not the
278 * block's owner tree. In this case, full back refs is used for pointers
279 * in the block. Remove these full back refs, add implicit back refs for
280 * every pointers in the new block.
282 * The reference count of the block is greater than one and the tree is
283 * the block's owner tree. In this case, implicit back refs is used for
284 * pointers in the block. Add full back refs for every pointers in the
285 * block, increase lower level extents' reference counts. The original
286 * implicit back refs are entailed to the new block.
288 * The reference count of the block is greater than one and the tree is
289 * not the block's owner tree. Add implicit back refs for every pointer in
290 * the new block, increase lower level extents' reference count.
292 * Back Reference Key composing:
294 * The key objectid corresponds to the first byte in the extent,
295 * The key type is used to differentiate between types of back refs.
296 * There are different meanings of the key offset for different types
299 * File extents can be referenced by:
301 * - multiple snapshots, subvolumes, or different generations in one subvol
302 * - different files inside a single subvolume
303 * - different offsets inside a file (bookend extents in file.c)
305 * The extent ref structure for the implicit back refs has fields for:
307 * - Objectid of the subvolume root
308 * - objectid of the file holding the reference
309 * - original offset in the file
310 * - how many bookend extents
312 * The key offset for the implicit back refs is hash of the first
315 * The extent ref structure for the full back refs has field for:
317 * - number of pointers in the tree leaf
319 * The key offset for the implicit back refs is the first byte of
322 * When a file extent is allocated, The implicit back refs is used.
323 * the fields are filled in:
325 * (root_key.objectid, inode objectid, offset in file, 1)
327 * When a file extent is removed file truncation, we find the
328 * corresponding implicit back refs and check the following fields:
330 * (btrfs_header_owner(leaf), inode objectid, offset in file)
332 * Btree extents can be referenced by:
334 * - Different subvolumes
336 * Both the implicit back refs and the full back refs for tree blocks
337 * only consist of key. The key offset for the implicit back refs is
338 * objectid of block's owner tree. The key offset for the full back refs
339 * is the first byte of parent block.
341 * When implicit back refs is used, information about the lowest key and
342 * level of the tree block are required. These information are stored in
343 * tree block info structure.
347 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
348 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
349 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
351 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
352 struct btrfs_extent_inline_ref *iref,
353 enum btrfs_inline_ref_type is_data)
355 struct btrfs_fs_info *fs_info = eb->fs_info;
356 int type = btrfs_extent_inline_ref_type(eb, iref);
357 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
359 if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
360 ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
364 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
365 type == BTRFS_SHARED_BLOCK_REF_KEY ||
366 type == BTRFS_SHARED_DATA_REF_KEY ||
367 type == BTRFS_EXTENT_DATA_REF_KEY) {
368 if (is_data == BTRFS_REF_TYPE_BLOCK) {
369 if (type == BTRFS_TREE_BLOCK_REF_KEY)
371 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
374 * Every shared one has parent tree block,
375 * which must be aligned to sector size.
377 if (offset && IS_ALIGNED(offset, fs_info->sectorsize))
380 } else if (is_data == BTRFS_REF_TYPE_DATA) {
381 if (type == BTRFS_EXTENT_DATA_REF_KEY)
383 if (type == BTRFS_SHARED_DATA_REF_KEY) {
386 * Every shared one has parent tree block,
387 * which must be aligned to sector size.
390 IS_ALIGNED(offset, fs_info->sectorsize))
394 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
400 btrfs_print_leaf(eb);
402 "eb %llu iref 0x%lx invalid extent inline ref type %d",
403 eb->start, (unsigned long)iref, type);
405 return BTRFS_REF_TYPE_INVALID;
408 u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
410 u32 high_crc = ~(u32)0;
411 u32 low_crc = ~(u32)0;
414 lenum = cpu_to_le64(root_objectid);
415 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
416 lenum = cpu_to_le64(owner);
417 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
418 lenum = cpu_to_le64(offset);
419 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
421 return ((u64)high_crc << 31) ^ (u64)low_crc;
424 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
425 struct btrfs_extent_data_ref *ref)
427 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
428 btrfs_extent_data_ref_objectid(leaf, ref),
429 btrfs_extent_data_ref_offset(leaf, ref));
432 static int match_extent_data_ref(struct extent_buffer *leaf,
433 struct btrfs_extent_data_ref *ref,
434 u64 root_objectid, u64 owner, u64 offset)
436 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
437 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
438 btrfs_extent_data_ref_offset(leaf, ref) != offset)
443 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
444 struct btrfs_path *path,
445 u64 bytenr, u64 parent,
447 u64 owner, u64 offset)
449 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
450 struct btrfs_key key;
451 struct btrfs_extent_data_ref *ref;
452 struct extent_buffer *leaf;
458 key.objectid = bytenr;
460 key.type = BTRFS_SHARED_DATA_REF_KEY;
463 key.type = BTRFS_EXTENT_DATA_REF_KEY;
464 key.offset = hash_extent_data_ref(root_objectid,
469 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
481 leaf = path->nodes[0];
482 nritems = btrfs_header_nritems(leaf);
484 if (path->slots[0] >= nritems) {
485 ret = btrfs_next_leaf(root, path);
491 leaf = path->nodes[0];
492 nritems = btrfs_header_nritems(leaf);
496 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
497 if (key.objectid != bytenr ||
498 key.type != BTRFS_EXTENT_DATA_REF_KEY)
501 ref = btrfs_item_ptr(leaf, path->slots[0],
502 struct btrfs_extent_data_ref);
504 if (match_extent_data_ref(leaf, ref, root_objectid,
507 btrfs_release_path(path);
519 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
520 struct btrfs_path *path,
521 u64 bytenr, u64 parent,
522 u64 root_objectid, u64 owner,
523 u64 offset, int refs_to_add)
525 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
526 struct btrfs_key key;
527 struct extent_buffer *leaf;
532 key.objectid = bytenr;
534 key.type = BTRFS_SHARED_DATA_REF_KEY;
536 size = sizeof(struct btrfs_shared_data_ref);
538 key.type = BTRFS_EXTENT_DATA_REF_KEY;
539 key.offset = hash_extent_data_ref(root_objectid,
541 size = sizeof(struct btrfs_extent_data_ref);
544 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
545 if (ret && ret != -EEXIST)
548 leaf = path->nodes[0];
550 struct btrfs_shared_data_ref *ref;
551 ref = btrfs_item_ptr(leaf, path->slots[0],
552 struct btrfs_shared_data_ref);
554 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
556 num_refs = btrfs_shared_data_ref_count(leaf, ref);
557 num_refs += refs_to_add;
558 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
561 struct btrfs_extent_data_ref *ref;
562 while (ret == -EEXIST) {
563 ref = btrfs_item_ptr(leaf, path->slots[0],
564 struct btrfs_extent_data_ref);
565 if (match_extent_data_ref(leaf, ref, root_objectid,
568 btrfs_release_path(path);
570 ret = btrfs_insert_empty_item(trans, root, path, &key,
572 if (ret && ret != -EEXIST)
575 leaf = path->nodes[0];
577 ref = btrfs_item_ptr(leaf, path->slots[0],
578 struct btrfs_extent_data_ref);
580 btrfs_set_extent_data_ref_root(leaf, ref,
582 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
583 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
584 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
586 num_refs = btrfs_extent_data_ref_count(leaf, ref);
587 num_refs += refs_to_add;
588 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
591 btrfs_mark_buffer_dirty(trans, leaf);
594 btrfs_release_path(path);
598 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
599 struct btrfs_root *root,
600 struct btrfs_path *path,
603 struct btrfs_key key;
604 struct btrfs_extent_data_ref *ref1 = NULL;
605 struct btrfs_shared_data_ref *ref2 = NULL;
606 struct extent_buffer *leaf;
610 leaf = path->nodes[0];
611 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
613 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
614 ref1 = btrfs_item_ptr(leaf, path->slots[0],
615 struct btrfs_extent_data_ref);
616 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
617 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
618 ref2 = btrfs_item_ptr(leaf, path->slots[0],
619 struct btrfs_shared_data_ref);
620 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
622 btrfs_err(trans->fs_info,
623 "unrecognized backref key (%llu %u %llu)",
624 key.objectid, key.type, key.offset);
625 btrfs_abort_transaction(trans, -EUCLEAN);
629 BUG_ON(num_refs < refs_to_drop);
630 num_refs -= refs_to_drop;
633 ret = btrfs_del_item(trans, root, path);
635 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
636 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
637 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
638 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
639 btrfs_mark_buffer_dirty(trans, leaf);
644 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
645 struct btrfs_extent_inline_ref *iref)
647 struct btrfs_key key;
648 struct extent_buffer *leaf;
649 struct btrfs_extent_data_ref *ref1;
650 struct btrfs_shared_data_ref *ref2;
654 leaf = path->nodes[0];
655 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
659 * If type is invalid, we should have bailed out earlier than
662 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
663 ASSERT(type != BTRFS_REF_TYPE_INVALID);
664 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
665 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
666 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
668 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
669 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
671 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
672 ref1 = btrfs_item_ptr(leaf, path->slots[0],
673 struct btrfs_extent_data_ref);
674 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
675 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
676 ref2 = btrfs_item_ptr(leaf, path->slots[0],
677 struct btrfs_shared_data_ref);
678 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
685 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
686 struct btrfs_path *path,
687 u64 bytenr, u64 parent,
690 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
691 struct btrfs_key key;
694 key.objectid = bytenr;
696 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
699 key.type = BTRFS_TREE_BLOCK_REF_KEY;
700 key.offset = root_objectid;
703 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
709 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
710 struct btrfs_path *path,
711 u64 bytenr, u64 parent,
714 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
715 struct btrfs_key key;
718 key.objectid = bytenr;
720 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
723 key.type = BTRFS_TREE_BLOCK_REF_KEY;
724 key.offset = root_objectid;
727 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
728 btrfs_release_path(path);
732 static inline int extent_ref_type(u64 parent, u64 owner)
735 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
737 type = BTRFS_SHARED_BLOCK_REF_KEY;
739 type = BTRFS_TREE_BLOCK_REF_KEY;
742 type = BTRFS_SHARED_DATA_REF_KEY;
744 type = BTRFS_EXTENT_DATA_REF_KEY;
749 static int find_next_key(struct btrfs_path *path, int level,
750 struct btrfs_key *key)
753 for (; level < BTRFS_MAX_LEVEL; level++) {
754 if (!path->nodes[level])
756 if (path->slots[level] + 1 >=
757 btrfs_header_nritems(path->nodes[level]))
760 btrfs_item_key_to_cpu(path->nodes[level], key,
761 path->slots[level] + 1);
763 btrfs_node_key_to_cpu(path->nodes[level], key,
764 path->slots[level] + 1);
771 * look for inline back ref. if back ref is found, *ref_ret is set
772 * to the address of inline back ref, and 0 is returned.
774 * if back ref isn't found, *ref_ret is set to the address where it
775 * should be inserted, and -ENOENT is returned.
777 * if insert is true and there are too many inline back refs, the path
778 * points to the extent item, and -EAGAIN is returned.
780 * NOTE: inline back refs are ordered in the same way that back ref
781 * items in the tree are ordered.
783 static noinline_for_stack
784 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
785 struct btrfs_path *path,
786 struct btrfs_extent_inline_ref **ref_ret,
787 u64 bytenr, u64 num_bytes,
788 u64 parent, u64 root_objectid,
789 u64 owner, u64 offset, int insert)
791 struct btrfs_fs_info *fs_info = trans->fs_info;
792 struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
793 struct btrfs_key key;
794 struct extent_buffer *leaf;
795 struct btrfs_extent_item *ei;
796 struct btrfs_extent_inline_ref *iref;
805 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
808 key.objectid = bytenr;
809 key.type = BTRFS_EXTENT_ITEM_KEY;
810 key.offset = num_bytes;
812 want = extent_ref_type(parent, owner);
814 extra_size = btrfs_extent_inline_ref_size(want);
815 path->search_for_extension = 1;
816 path->keep_locks = 1;
821 * Owner is our level, so we can just add one to get the level for the
822 * block we are interested in.
824 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
825 key.type = BTRFS_METADATA_ITEM_KEY;
830 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
835 * We may be a newly converted file system which still has the old fat
836 * extent entries for metadata, so try and see if we have one of those.
838 if (ret > 0 && skinny_metadata) {
839 skinny_metadata = false;
840 if (path->slots[0]) {
842 btrfs_item_key_to_cpu(path->nodes[0], &key,
844 if (key.objectid == bytenr &&
845 key.type == BTRFS_EXTENT_ITEM_KEY &&
846 key.offset == num_bytes)
850 key.objectid = bytenr;
851 key.type = BTRFS_EXTENT_ITEM_KEY;
852 key.offset = num_bytes;
853 btrfs_release_path(path);
858 if (ret && !insert) {
861 } else if (WARN_ON(ret)) {
862 btrfs_print_leaf(path->nodes[0]);
864 "extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu",
865 bytenr, num_bytes, parent, root_objectid, owner,
871 leaf = path->nodes[0];
872 item_size = btrfs_item_size(leaf, path->slots[0]);
873 if (unlikely(item_size < sizeof(*ei))) {
876 "unexpected extent item size, has %llu expect >= %zu",
877 item_size, sizeof(*ei));
878 btrfs_abort_transaction(trans, ret);
882 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
883 flags = btrfs_extent_flags(leaf, ei);
885 ptr = (unsigned long)(ei + 1);
886 end = (unsigned long)ei + item_size;
888 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
889 ptr += sizeof(struct btrfs_tree_block_info);
893 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
894 needed = BTRFS_REF_TYPE_DATA;
896 needed = BTRFS_REF_TYPE_BLOCK;
900 iref = (struct btrfs_extent_inline_ref *)ptr;
901 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
902 if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
903 ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
904 ptr += btrfs_extent_inline_ref_size(type);
907 if (type == BTRFS_REF_TYPE_INVALID) {
915 ptr += btrfs_extent_inline_ref_size(type);
919 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
920 struct btrfs_extent_data_ref *dref;
921 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
922 if (match_extent_data_ref(leaf, dref, root_objectid,
927 if (hash_extent_data_ref_item(leaf, dref) <
928 hash_extent_data_ref(root_objectid, owner, offset))
932 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
934 if (parent == ref_offset) {
938 if (ref_offset < parent)
941 if (root_objectid == ref_offset) {
945 if (ref_offset < root_objectid)
949 ptr += btrfs_extent_inline_ref_size(type);
952 if (unlikely(ptr > end)) {
954 btrfs_print_leaf(path->nodes[0]);
956 "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu",
957 path->slots[0], root_objectid, owner, offset, parent);
961 if (ret == -ENOENT && insert) {
962 if (item_size + extra_size >=
963 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
968 * To add new inline back ref, we have to make sure
969 * there is no corresponding back ref item.
970 * For simplicity, we just do not add new inline back
971 * ref if there is any kind of item for this block
973 if (find_next_key(path, 0, &key) == 0 &&
974 key.objectid == bytenr &&
975 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
980 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
983 path->keep_locks = 0;
984 path->search_for_extension = 0;
985 btrfs_unlock_up_safe(path, 1);
991 * helper to add new inline back ref
993 static noinline_for_stack
994 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
995 struct btrfs_path *path,
996 struct btrfs_extent_inline_ref *iref,
997 u64 parent, u64 root_objectid,
998 u64 owner, u64 offset, int refs_to_add,
999 struct btrfs_delayed_extent_op *extent_op)
1001 struct extent_buffer *leaf;
1002 struct btrfs_extent_item *ei;
1005 unsigned long item_offset;
1010 leaf = path->nodes[0];
1011 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1012 item_offset = (unsigned long)iref - (unsigned long)ei;
1014 type = extent_ref_type(parent, owner);
1015 size = btrfs_extent_inline_ref_size(type);
1017 btrfs_extend_item(trans, path, size);
1019 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1020 refs = btrfs_extent_refs(leaf, ei);
1021 refs += refs_to_add;
1022 btrfs_set_extent_refs(leaf, ei, refs);
1024 __run_delayed_extent_op(extent_op, leaf, ei);
1026 ptr = (unsigned long)ei + item_offset;
1027 end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]);
1028 if (ptr < end - size)
1029 memmove_extent_buffer(leaf, ptr + size, ptr,
1032 iref = (struct btrfs_extent_inline_ref *)ptr;
1033 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1034 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1035 struct btrfs_extent_data_ref *dref;
1036 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1037 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1038 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1039 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1040 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1041 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1042 struct btrfs_shared_data_ref *sref;
1043 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1044 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1045 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1046 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1047 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1049 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1051 btrfs_mark_buffer_dirty(trans, leaf);
1054 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1055 struct btrfs_path *path,
1056 struct btrfs_extent_inline_ref **ref_ret,
1057 u64 bytenr, u64 num_bytes, u64 parent,
1058 u64 root_objectid, u64 owner, u64 offset)
1062 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1063 num_bytes, parent, root_objectid,
1068 btrfs_release_path(path);
1071 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1072 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1075 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1076 root_objectid, owner, offset);
1082 * helper to update/remove inline back ref
1084 static noinline_for_stack int update_inline_extent_backref(
1085 struct btrfs_trans_handle *trans,
1086 struct btrfs_path *path,
1087 struct btrfs_extent_inline_ref *iref,
1089 struct btrfs_delayed_extent_op *extent_op)
1091 struct extent_buffer *leaf = path->nodes[0];
1092 struct btrfs_fs_info *fs_info = leaf->fs_info;
1093 struct btrfs_extent_item *ei;
1094 struct btrfs_extent_data_ref *dref = NULL;
1095 struct btrfs_shared_data_ref *sref = NULL;
1103 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1104 refs = btrfs_extent_refs(leaf, ei);
1105 if (unlikely(refs_to_mod < 0 && refs + refs_to_mod <= 0)) {
1106 struct btrfs_key key;
1109 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1110 if (key.type == BTRFS_METADATA_ITEM_KEY)
1111 extent_size = fs_info->nodesize;
1113 extent_size = key.offset;
1114 btrfs_print_leaf(leaf);
1116 "invalid refs_to_mod for extent %llu num_bytes %u, has %d expect >= -%llu",
1117 key.objectid, extent_size, refs_to_mod, refs);
1120 refs += refs_to_mod;
1121 btrfs_set_extent_refs(leaf, ei, refs);
1123 __run_delayed_extent_op(extent_op, leaf, ei);
1125 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1127 * Function btrfs_get_extent_inline_ref_type() has already printed
1130 if (unlikely(type == BTRFS_REF_TYPE_INVALID))
1133 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1134 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1135 refs = btrfs_extent_data_ref_count(leaf, dref);
1136 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1137 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1138 refs = btrfs_shared_data_ref_count(leaf, sref);
1142 * For tree blocks we can only drop one ref for it, and tree
1143 * blocks should not have refs > 1.
1145 * Furthermore if we're inserting a new inline backref, we
1146 * won't reach this path either. That would be
1147 * setup_inline_extent_backref().
1149 if (unlikely(refs_to_mod != -1)) {
1150 struct btrfs_key key;
1152 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1154 btrfs_print_leaf(leaf);
1156 "invalid refs_to_mod for tree block %llu, has %d expect -1",
1157 key.objectid, refs_to_mod);
1162 if (unlikely(refs_to_mod < 0 && refs < -refs_to_mod)) {
1163 struct btrfs_key key;
1166 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1167 if (key.type == BTRFS_METADATA_ITEM_KEY)
1168 extent_size = fs_info->nodesize;
1170 extent_size = key.offset;
1171 btrfs_print_leaf(leaf);
1173 "invalid refs_to_mod for backref entry, iref %lu extent %llu num_bytes %u, has %d expect >= -%llu",
1174 (unsigned long)iref, key.objectid, extent_size,
1178 refs += refs_to_mod;
1181 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1182 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1184 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1186 size = btrfs_extent_inline_ref_size(type);
1187 item_size = btrfs_item_size(leaf, path->slots[0]);
1188 ptr = (unsigned long)iref;
1189 end = (unsigned long)ei + item_size;
1190 if (ptr + size < end)
1191 memmove_extent_buffer(leaf, ptr, ptr + size,
1194 btrfs_truncate_item(trans, path, item_size, 1);
1196 btrfs_mark_buffer_dirty(trans, leaf);
1200 static noinline_for_stack
1201 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1202 struct btrfs_path *path,
1203 u64 bytenr, u64 num_bytes, u64 parent,
1204 u64 root_objectid, u64 owner,
1205 u64 offset, int refs_to_add,
1206 struct btrfs_delayed_extent_op *extent_op)
1208 struct btrfs_extent_inline_ref *iref;
1211 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1212 num_bytes, parent, root_objectid,
1216 * We're adding refs to a tree block we already own, this
1217 * should not happen at all.
1219 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1220 btrfs_print_leaf(path->nodes[0]);
1221 btrfs_crit(trans->fs_info,
1222 "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu slot %u",
1223 bytenr, num_bytes, root_objectid, path->slots[0]);
1226 ret = update_inline_extent_backref(trans, path, iref,
1227 refs_to_add, extent_op);
1228 } else if (ret == -ENOENT) {
1229 setup_inline_extent_backref(trans, path, iref, parent,
1230 root_objectid, owner, offset,
1231 refs_to_add, extent_op);
1237 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1238 struct btrfs_root *root,
1239 struct btrfs_path *path,
1240 struct btrfs_extent_inline_ref *iref,
1241 int refs_to_drop, int is_data)
1245 BUG_ON(!is_data && refs_to_drop != 1);
1247 ret = update_inline_extent_backref(trans, path, iref,
1248 -refs_to_drop, NULL);
1250 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1252 ret = btrfs_del_item(trans, root, path);
1256 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1257 u64 *discarded_bytes)
1260 u64 bytes_left, end;
1261 u64 aligned_start = ALIGN(start, 1 << SECTOR_SHIFT);
1263 if (WARN_ON(start != aligned_start)) {
1264 len -= aligned_start - start;
1265 len = round_down(len, 1 << SECTOR_SHIFT);
1266 start = aligned_start;
1269 *discarded_bytes = 0;
1277 /* Skip any superblocks on this device. */
1278 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1279 u64 sb_start = btrfs_sb_offset(j);
1280 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1281 u64 size = sb_start - start;
1283 if (!in_range(sb_start, start, bytes_left) &&
1284 !in_range(sb_end, start, bytes_left) &&
1285 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1289 * Superblock spans beginning of range. Adjust start and
1292 if (sb_start <= start) {
1293 start += sb_end - start;
1298 bytes_left = end - start;
1303 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1304 size >> SECTOR_SHIFT,
1307 *discarded_bytes += size;
1308 else if (ret != -EOPNOTSUPP)
1317 bytes_left = end - start;
1321 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1322 bytes_left >> SECTOR_SHIFT,
1325 *discarded_bytes += bytes_left;
1330 static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes)
1332 struct btrfs_device *dev = stripe->dev;
1333 struct btrfs_fs_info *fs_info = dev->fs_info;
1334 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1335 u64 phys = stripe->physical;
1336 u64 len = stripe->length;
1340 /* Zone reset on a zoned filesystem */
1341 if (btrfs_can_zone_reset(dev, phys, len)) {
1344 ret = btrfs_reset_device_zone(dev, phys, len, &discarded);
1348 if (!btrfs_dev_replace_is_ongoing(dev_replace) ||
1349 dev != dev_replace->srcdev)
1352 src_disc = discarded;
1354 /* Send to replace target as well */
1355 ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len,
1357 discarded += src_disc;
1358 } else if (bdev_max_discard_sectors(stripe->dev->bdev)) {
1359 ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded);
1370 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1371 u64 num_bytes, u64 *actual_bytes)
1374 u64 discarded_bytes = 0;
1375 u64 end = bytenr + num_bytes;
1379 * Avoid races with device replace and make sure the devices in the
1380 * stripes don't go away while we are discarding.
1382 btrfs_bio_counter_inc_blocked(fs_info);
1384 struct btrfs_discard_stripe *stripes;
1385 unsigned int num_stripes;
1388 num_bytes = end - cur;
1389 stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes);
1390 if (IS_ERR(stripes)) {
1391 ret = PTR_ERR(stripes);
1392 if (ret == -EOPNOTSUPP)
1397 for (i = 0; i < num_stripes; i++) {
1398 struct btrfs_discard_stripe *stripe = stripes + i;
1401 if (!stripe->dev->bdev) {
1402 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1406 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
1407 &stripe->dev->dev_state))
1410 ret = do_discard_extent(stripe, &bytes);
1413 * Keep going if discard is not supported by the
1416 if (ret != -EOPNOTSUPP)
1420 discarded_bytes += bytes;
1428 btrfs_bio_counter_dec(fs_info);
1430 *actual_bytes = discarded_bytes;
1434 /* Can return -ENOMEM */
1435 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1436 struct btrfs_ref *generic_ref)
1438 struct btrfs_fs_info *fs_info = trans->fs_info;
1441 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1442 generic_ref->action);
1443 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1444 generic_ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID);
1446 if (generic_ref->type == BTRFS_REF_METADATA)
1447 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1449 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1451 btrfs_ref_tree_mod(fs_info, generic_ref);
1457 * Insert backreference for a given extent.
1459 * The counterpart is in __btrfs_free_extent(), with examples and more details
1462 * @trans: Handle of transaction
1464 * @node: The delayed ref node used to get the bytenr/length for
1465 * extent whose references are incremented.
1467 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1468 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1469 * bytenr of the parent block. Since new extents are always
1470 * created with indirect references, this will only be the case
1471 * when relocating a shared extent. In that case, root_objectid
1472 * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1475 * @root_objectid: The id of the root where this modification has originated,
1476 * this can be either one of the well-known metadata trees or
1477 * the subvolume id which references this extent.
1479 * @owner: For data extents it is the inode number of the owning file.
1480 * For metadata extents this parameter holds the level in the
1481 * tree of the extent.
1483 * @offset: For metadata extents the offset is ignored and is currently
1484 * always passed as 0. For data extents it is the fileoffset
1485 * this extent belongs to.
1487 * @extent_op Pointer to a structure, holding information necessary when
1488 * updating a tree block's flags
1491 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1492 struct btrfs_delayed_ref_node *node,
1493 u64 parent, u64 root_objectid,
1494 u64 owner, u64 offset,
1495 struct btrfs_delayed_extent_op *extent_op)
1497 struct btrfs_path *path;
1498 struct extent_buffer *leaf;
1499 struct btrfs_extent_item *item;
1500 struct btrfs_key key;
1501 u64 bytenr = node->bytenr;
1502 u64 num_bytes = node->num_bytes;
1504 int refs_to_add = node->ref_mod;
1507 path = btrfs_alloc_path();
1511 /* this will setup the path even if it fails to insert the back ref */
1512 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1513 parent, root_objectid, owner,
1514 offset, refs_to_add, extent_op);
1515 if ((ret < 0 && ret != -EAGAIN) || !ret)
1519 * Ok we had -EAGAIN which means we didn't have space to insert and
1520 * inline extent ref, so just update the reference count and add a
1523 leaf = path->nodes[0];
1524 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1525 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1526 refs = btrfs_extent_refs(leaf, item);
1527 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1529 __run_delayed_extent_op(extent_op, leaf, item);
1531 btrfs_mark_buffer_dirty(trans, leaf);
1532 btrfs_release_path(path);
1534 /* now insert the actual backref */
1535 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1536 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1539 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1540 root_objectid, owner, offset,
1544 btrfs_abort_transaction(trans, ret);
1546 btrfs_free_path(path);
1550 static void free_head_ref_squota_rsv(struct btrfs_fs_info *fs_info,
1551 struct btrfs_delayed_ref_head *href)
1553 u64 root = href->owning_root;
1556 * Don't check must_insert_reserved, as this is called from contexts
1557 * where it has already been unset.
1559 if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE ||
1560 !href->is_data || !is_fstree(root))
1563 btrfs_qgroup_free_refroot(fs_info, root, href->reserved_bytes,
1564 BTRFS_QGROUP_RSV_DATA);
1567 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1568 struct btrfs_delayed_ref_head *href,
1569 struct btrfs_delayed_ref_node *node,
1570 struct btrfs_delayed_extent_op *extent_op,
1571 bool insert_reserved)
1574 struct btrfs_delayed_data_ref *ref;
1578 ref = btrfs_delayed_node_to_data_ref(node);
1579 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1581 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1582 parent = ref->parent;
1584 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1585 struct btrfs_key key;
1586 struct btrfs_squota_delta delta = {
1587 .root = href->owning_root,
1588 .num_bytes = node->num_bytes,
1591 .generation = trans->transid,
1595 flags |= extent_op->flags_to_set;
1597 key.objectid = node->bytenr;
1598 key.type = BTRFS_EXTENT_ITEM_KEY;
1599 key.offset = node->num_bytes;
1601 ret = alloc_reserved_file_extent(trans, parent, ref->root,
1602 flags, ref->objectid,
1604 node->ref_mod, href->owning_root);
1605 free_head_ref_squota_rsv(trans->fs_info, href);
1607 ret = btrfs_record_squota_delta(trans->fs_info, &delta);
1608 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1609 ret = __btrfs_inc_extent_ref(trans, node, parent, ref->root,
1610 ref->objectid, ref->offset,
1612 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1613 ret = __btrfs_free_extent(trans, href, node, parent,
1614 ref->root, ref->objectid,
1615 ref->offset, extent_op);
1622 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1623 struct extent_buffer *leaf,
1624 struct btrfs_extent_item *ei)
1626 u64 flags = btrfs_extent_flags(leaf, ei);
1627 if (extent_op->update_flags) {
1628 flags |= extent_op->flags_to_set;
1629 btrfs_set_extent_flags(leaf, ei, flags);
1632 if (extent_op->update_key) {
1633 struct btrfs_tree_block_info *bi;
1634 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1635 bi = (struct btrfs_tree_block_info *)(ei + 1);
1636 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1640 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1641 struct btrfs_delayed_ref_head *head,
1642 struct btrfs_delayed_extent_op *extent_op)
1644 struct btrfs_fs_info *fs_info = trans->fs_info;
1645 struct btrfs_root *root;
1646 struct btrfs_key key;
1647 struct btrfs_path *path;
1648 struct btrfs_extent_item *ei;
1649 struct extent_buffer *leaf;
1654 if (TRANS_ABORTED(trans))
1657 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1660 path = btrfs_alloc_path();
1664 key.objectid = head->bytenr;
1667 key.type = BTRFS_METADATA_ITEM_KEY;
1668 key.offset = extent_op->level;
1670 key.type = BTRFS_EXTENT_ITEM_KEY;
1671 key.offset = head->num_bytes;
1674 root = btrfs_extent_root(fs_info, key.objectid);
1676 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1679 } else if (ret > 0) {
1681 if (path->slots[0] > 0) {
1683 btrfs_item_key_to_cpu(path->nodes[0], &key,
1685 if (key.objectid == head->bytenr &&
1686 key.type == BTRFS_EXTENT_ITEM_KEY &&
1687 key.offset == head->num_bytes)
1691 btrfs_release_path(path);
1694 key.objectid = head->bytenr;
1695 key.offset = head->num_bytes;
1696 key.type = BTRFS_EXTENT_ITEM_KEY;
1702 "missing extent item for extent %llu num_bytes %llu level %d",
1703 head->bytenr, head->num_bytes, extent_op->level);
1708 leaf = path->nodes[0];
1709 item_size = btrfs_item_size(leaf, path->slots[0]);
1711 if (unlikely(item_size < sizeof(*ei))) {
1714 "unexpected extent item size, has %u expect >= %zu",
1715 item_size, sizeof(*ei));
1716 btrfs_abort_transaction(trans, ret);
1720 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1721 __run_delayed_extent_op(extent_op, leaf, ei);
1723 btrfs_mark_buffer_dirty(trans, leaf);
1725 btrfs_free_path(path);
1729 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1730 struct btrfs_delayed_ref_head *href,
1731 struct btrfs_delayed_ref_node *node,
1732 struct btrfs_delayed_extent_op *extent_op,
1733 bool insert_reserved)
1736 struct btrfs_fs_info *fs_info = trans->fs_info;
1737 struct btrfs_delayed_tree_ref *ref;
1741 ref = btrfs_delayed_node_to_tree_ref(node);
1742 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1744 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1745 parent = ref->parent;
1746 ref_root = ref->root;
1748 if (unlikely(node->ref_mod != 1)) {
1749 btrfs_err(trans->fs_info,
1750 "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu",
1751 node->bytenr, node->ref_mod, node->action, ref_root,
1755 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1756 struct btrfs_squota_delta delta = {
1757 .root = href->owning_root,
1758 .num_bytes = fs_info->nodesize,
1761 .generation = trans->transid,
1764 BUG_ON(!extent_op || !extent_op->update_flags);
1765 ret = alloc_reserved_tree_block(trans, node, extent_op);
1767 btrfs_record_squota_delta(fs_info, &delta);
1768 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1769 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1770 ref->level, 0, extent_op);
1771 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1772 ret = __btrfs_free_extent(trans, href, node, parent, ref_root,
1773 ref->level, 0, extent_op);
1780 /* helper function to actually process a single delayed ref entry */
1781 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1782 struct btrfs_delayed_ref_head *href,
1783 struct btrfs_delayed_ref_node *node,
1784 struct btrfs_delayed_extent_op *extent_op,
1785 bool insert_reserved)
1789 if (TRANS_ABORTED(trans)) {
1790 if (insert_reserved) {
1791 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1792 free_head_ref_squota_rsv(trans->fs_info, href);
1797 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1798 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1799 ret = run_delayed_tree_ref(trans, href, node, extent_op,
1801 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1802 node->type == BTRFS_SHARED_DATA_REF_KEY)
1803 ret = run_delayed_data_ref(trans, href, node, extent_op,
1805 else if (node->type == BTRFS_EXTENT_OWNER_REF_KEY)
1809 if (ret && insert_reserved)
1810 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1812 btrfs_err(trans->fs_info,
1813 "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d",
1814 node->bytenr, node->num_bytes, node->type,
1815 node->action, node->ref_mod, ret);
1819 static inline struct btrfs_delayed_ref_node *
1820 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1822 struct btrfs_delayed_ref_node *ref;
1824 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1828 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1829 * This is to prevent a ref count from going down to zero, which deletes
1830 * the extent item from the extent tree, when there still are references
1831 * to add, which would fail because they would not find the extent item.
1833 if (!list_empty(&head->ref_add_list))
1834 return list_first_entry(&head->ref_add_list,
1835 struct btrfs_delayed_ref_node, add_list);
1837 ref = rb_entry(rb_first_cached(&head->ref_tree),
1838 struct btrfs_delayed_ref_node, ref_node);
1839 ASSERT(list_empty(&ref->add_list));
1843 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1844 struct btrfs_delayed_ref_head *head)
1846 spin_lock(&delayed_refs->lock);
1847 head->processing = false;
1848 delayed_refs->num_heads_ready++;
1849 spin_unlock(&delayed_refs->lock);
1850 btrfs_delayed_ref_unlock(head);
1853 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1854 struct btrfs_delayed_ref_head *head)
1856 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1861 if (head->must_insert_reserved) {
1862 head->extent_op = NULL;
1863 btrfs_free_delayed_extent_op(extent_op);
1869 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1870 struct btrfs_delayed_ref_head *head)
1872 struct btrfs_delayed_extent_op *extent_op;
1875 extent_op = cleanup_extent_op(head);
1878 head->extent_op = NULL;
1879 spin_unlock(&head->lock);
1880 ret = run_delayed_extent_op(trans, head, extent_op);
1881 btrfs_free_delayed_extent_op(extent_op);
1882 return ret ? ret : 1;
1885 u64 btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1886 struct btrfs_delayed_ref_root *delayed_refs,
1887 struct btrfs_delayed_ref_head *head)
1892 * We had csum deletions accounted for in our delayed refs rsv, we need
1893 * to drop the csum leaves for this update from our delayed_refs_rsv.
1895 if (head->total_ref_mod < 0 && head->is_data) {
1898 spin_lock(&delayed_refs->lock);
1899 delayed_refs->pending_csums -= head->num_bytes;
1900 spin_unlock(&delayed_refs->lock);
1901 nr_csums = btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1903 btrfs_delayed_refs_rsv_release(fs_info, 0, nr_csums);
1905 ret = btrfs_calc_delayed_ref_csum_bytes(fs_info, nr_csums);
1907 /* must_insert_reserved can be set only if we didn't run the head ref. */
1908 if (head->must_insert_reserved)
1909 free_head_ref_squota_rsv(fs_info, head);
1914 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1915 struct btrfs_delayed_ref_head *head,
1916 u64 *bytes_released)
1919 struct btrfs_fs_info *fs_info = trans->fs_info;
1920 struct btrfs_delayed_ref_root *delayed_refs;
1923 delayed_refs = &trans->transaction->delayed_refs;
1925 ret = run_and_cleanup_extent_op(trans, head);
1927 unselect_delayed_ref_head(delayed_refs, head);
1928 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1935 * Need to drop our head ref lock and re-acquire the delayed ref lock
1936 * and then re-check to make sure nobody got added.
1938 spin_unlock(&head->lock);
1939 spin_lock(&delayed_refs->lock);
1940 spin_lock(&head->lock);
1941 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1942 spin_unlock(&head->lock);
1943 spin_unlock(&delayed_refs->lock);
1946 btrfs_delete_ref_head(delayed_refs, head);
1947 spin_unlock(&head->lock);
1948 spin_unlock(&delayed_refs->lock);
1950 if (head->must_insert_reserved) {
1951 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1952 if (head->is_data) {
1953 struct btrfs_root *csum_root;
1955 csum_root = btrfs_csum_root(fs_info, head->bytenr);
1956 ret = btrfs_del_csums(trans, csum_root, head->bytenr,
1961 *bytes_released += btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1963 trace_run_delayed_ref_head(fs_info, head, 0);
1964 btrfs_delayed_ref_unlock(head);
1965 btrfs_put_delayed_ref_head(head);
1969 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1970 struct btrfs_trans_handle *trans)
1972 struct btrfs_delayed_ref_root *delayed_refs =
1973 &trans->transaction->delayed_refs;
1974 struct btrfs_delayed_ref_head *head = NULL;
1977 spin_lock(&delayed_refs->lock);
1978 head = btrfs_select_ref_head(delayed_refs);
1980 spin_unlock(&delayed_refs->lock);
1985 * Grab the lock that says we are going to process all the refs for
1988 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1989 spin_unlock(&delayed_refs->lock);
1992 * We may have dropped the spin lock to get the head mutex lock, and
1993 * that might have given someone else time to free the head. If that's
1994 * true, it has been removed from our list and we can move on.
1997 head = ERR_PTR(-EAGAIN);
2002 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
2003 struct btrfs_delayed_ref_head *locked_ref,
2004 u64 *bytes_released)
2006 struct btrfs_fs_info *fs_info = trans->fs_info;
2007 struct btrfs_delayed_ref_root *delayed_refs;
2008 struct btrfs_delayed_extent_op *extent_op;
2009 struct btrfs_delayed_ref_node *ref;
2010 bool must_insert_reserved;
2013 delayed_refs = &trans->transaction->delayed_refs;
2015 lockdep_assert_held(&locked_ref->mutex);
2016 lockdep_assert_held(&locked_ref->lock);
2018 while ((ref = select_delayed_ref(locked_ref))) {
2020 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2021 spin_unlock(&locked_ref->lock);
2022 unselect_delayed_ref_head(delayed_refs, locked_ref);
2026 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
2027 RB_CLEAR_NODE(&ref->ref_node);
2028 if (!list_empty(&ref->add_list))
2029 list_del(&ref->add_list);
2031 * When we play the delayed ref, also correct the ref_mod on
2034 switch (ref->action) {
2035 case BTRFS_ADD_DELAYED_REF:
2036 case BTRFS_ADD_DELAYED_EXTENT:
2037 locked_ref->ref_mod -= ref->ref_mod;
2039 case BTRFS_DROP_DELAYED_REF:
2040 locked_ref->ref_mod += ref->ref_mod;
2045 atomic_dec(&delayed_refs->num_entries);
2048 * Record the must_insert_reserved flag before we drop the
2051 must_insert_reserved = locked_ref->must_insert_reserved;
2053 * Unsetting this on the head ref relinquishes ownership of
2054 * the rsv_bytes, so it is critical that every possible code
2055 * path from here forward frees all reserves including qgroup
2058 locked_ref->must_insert_reserved = false;
2060 extent_op = locked_ref->extent_op;
2061 locked_ref->extent_op = NULL;
2062 spin_unlock(&locked_ref->lock);
2064 ret = run_one_delayed_ref(trans, locked_ref, ref, extent_op,
2065 must_insert_reserved);
2066 btrfs_delayed_refs_rsv_release(fs_info, 1, 0);
2067 *bytes_released += btrfs_calc_delayed_ref_bytes(fs_info, 1);
2069 btrfs_free_delayed_extent_op(extent_op);
2071 unselect_delayed_ref_head(delayed_refs, locked_ref);
2072 btrfs_put_delayed_ref(ref);
2076 btrfs_put_delayed_ref(ref);
2079 spin_lock(&locked_ref->lock);
2080 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2087 * Returns 0 on success or if called with an already aborted transaction.
2088 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2090 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2093 struct btrfs_fs_info *fs_info = trans->fs_info;
2094 struct btrfs_delayed_ref_root *delayed_refs;
2095 struct btrfs_delayed_ref_head *locked_ref = NULL;
2097 unsigned long count = 0;
2098 unsigned long max_count = 0;
2099 u64 bytes_processed = 0;
2101 delayed_refs = &trans->transaction->delayed_refs;
2102 if (min_bytes == 0) {
2103 max_count = delayed_refs->num_heads_ready;
2104 min_bytes = U64_MAX;
2109 locked_ref = btrfs_obtain_ref_head(trans);
2110 if (IS_ERR_OR_NULL(locked_ref)) {
2111 if (PTR_ERR(locked_ref) == -EAGAIN) {
2120 * We need to try and merge add/drops of the same ref since we
2121 * can run into issues with relocate dropping the implicit ref
2122 * and then it being added back again before the drop can
2123 * finish. If we merged anything we need to re-loop so we can
2125 * Or we can get node references of the same type that weren't
2126 * merged when created due to bumps in the tree mod seq, and
2127 * we need to merge them to prevent adding an inline extent
2128 * backref before dropping it (triggering a BUG_ON at
2129 * insert_inline_extent_backref()).
2131 spin_lock(&locked_ref->lock);
2132 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2134 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, &bytes_processed);
2135 if (ret < 0 && ret != -EAGAIN) {
2137 * Error, btrfs_run_delayed_refs_for_head already
2138 * unlocked everything so just bail out
2143 * Success, perform the usual cleanup of a processed
2146 ret = cleanup_ref_head(trans, locked_ref, &bytes_processed);
2148 /* We dropped our lock, we need to loop. */
2157 * Either success case or btrfs_run_delayed_refs_for_head
2158 * returned -EAGAIN, meaning we need to select another head
2163 } while ((min_bytes != U64_MAX && bytes_processed < min_bytes) ||
2164 (max_count > 0 && count < max_count) ||
2170 #ifdef SCRAMBLE_DELAYED_REFS
2172 * Normally delayed refs get processed in ascending bytenr order. This
2173 * correlates in most cases to the order added. To expose dependencies on this
2174 * order, we start to process the tree in the middle instead of the beginning
2176 static u64 find_middle(struct rb_root *root)
2178 struct rb_node *n = root->rb_node;
2179 struct btrfs_delayed_ref_node *entry;
2182 u64 first = 0, last = 0;
2186 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2187 first = entry->bytenr;
2191 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2192 last = entry->bytenr;
2197 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2198 WARN_ON(!entry->in_tree);
2200 middle = entry->bytenr;
2214 * Start processing the delayed reference count updates and extent insertions
2215 * we have queued up so far.
2217 * @trans: Transaction handle.
2218 * @min_bytes: How many bytes of delayed references to process. After this
2219 * many bytes we stop processing delayed references if there are
2220 * any more. If 0 it means to run all existing delayed references,
2221 * but not new ones added after running all existing ones.
2222 * Use (u64)-1 (U64_MAX) to run all existing delayed references
2223 * plus any new ones that are added.
2225 * Returns 0 on success or if called with an aborted transaction
2226 * Returns <0 on error and aborts the transaction
2228 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, u64 min_bytes)
2230 struct btrfs_fs_info *fs_info = trans->fs_info;
2231 struct btrfs_delayed_ref_root *delayed_refs;
2234 /* We'll clean this up in btrfs_cleanup_transaction */
2235 if (TRANS_ABORTED(trans))
2238 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2241 delayed_refs = &trans->transaction->delayed_refs;
2243 #ifdef SCRAMBLE_DELAYED_REFS
2244 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2246 ret = __btrfs_run_delayed_refs(trans, min_bytes);
2248 btrfs_abort_transaction(trans, ret);
2252 if (min_bytes == U64_MAX) {
2253 btrfs_create_pending_block_groups(trans);
2255 spin_lock(&delayed_refs->lock);
2256 if (RB_EMPTY_ROOT(&delayed_refs->href_root.rb_root)) {
2257 spin_unlock(&delayed_refs->lock);
2260 spin_unlock(&delayed_refs->lock);
2269 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2270 struct extent_buffer *eb, u64 flags)
2272 struct btrfs_delayed_extent_op *extent_op;
2273 int level = btrfs_header_level(eb);
2276 extent_op = btrfs_alloc_delayed_extent_op();
2280 extent_op->flags_to_set = flags;
2281 extent_op->update_flags = true;
2282 extent_op->update_key = false;
2283 extent_op->level = level;
2285 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2287 btrfs_free_delayed_extent_op(extent_op);
2291 static noinline int check_delayed_ref(struct btrfs_root *root,
2292 struct btrfs_path *path,
2293 u64 objectid, u64 offset, u64 bytenr)
2295 struct btrfs_delayed_ref_head *head;
2296 struct btrfs_delayed_ref_node *ref;
2297 struct btrfs_delayed_data_ref *data_ref;
2298 struct btrfs_delayed_ref_root *delayed_refs;
2299 struct btrfs_transaction *cur_trans;
2300 struct rb_node *node;
2303 spin_lock(&root->fs_info->trans_lock);
2304 cur_trans = root->fs_info->running_transaction;
2306 refcount_inc(&cur_trans->use_count);
2307 spin_unlock(&root->fs_info->trans_lock);
2311 delayed_refs = &cur_trans->delayed_refs;
2312 spin_lock(&delayed_refs->lock);
2313 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2315 spin_unlock(&delayed_refs->lock);
2316 btrfs_put_transaction(cur_trans);
2320 if (!mutex_trylock(&head->mutex)) {
2322 spin_unlock(&delayed_refs->lock);
2323 btrfs_put_transaction(cur_trans);
2327 refcount_inc(&head->refs);
2328 spin_unlock(&delayed_refs->lock);
2330 btrfs_release_path(path);
2333 * Mutex was contended, block until it's released and let
2336 mutex_lock(&head->mutex);
2337 mutex_unlock(&head->mutex);
2338 btrfs_put_delayed_ref_head(head);
2339 btrfs_put_transaction(cur_trans);
2342 spin_unlock(&delayed_refs->lock);
2344 spin_lock(&head->lock);
2346 * XXX: We should replace this with a proper search function in the
2349 for (node = rb_first_cached(&head->ref_tree); node;
2350 node = rb_next(node)) {
2351 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2352 /* If it's a shared ref we know a cross reference exists */
2353 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2358 data_ref = btrfs_delayed_node_to_data_ref(ref);
2361 * If our ref doesn't match the one we're currently looking at
2362 * then we have a cross reference.
2364 if (data_ref->root != root->root_key.objectid ||
2365 data_ref->objectid != objectid ||
2366 data_ref->offset != offset) {
2371 spin_unlock(&head->lock);
2372 mutex_unlock(&head->mutex);
2373 btrfs_put_transaction(cur_trans);
2377 static noinline int check_committed_ref(struct btrfs_root *root,
2378 struct btrfs_path *path,
2379 u64 objectid, u64 offset, u64 bytenr,
2382 struct btrfs_fs_info *fs_info = root->fs_info;
2383 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2384 struct extent_buffer *leaf;
2385 struct btrfs_extent_data_ref *ref;
2386 struct btrfs_extent_inline_ref *iref;
2387 struct btrfs_extent_item *ei;
2388 struct btrfs_key key;
2394 key.objectid = bytenr;
2395 key.offset = (u64)-1;
2396 key.type = BTRFS_EXTENT_ITEM_KEY;
2398 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2401 BUG_ON(ret == 0); /* Corruption */
2404 if (path->slots[0] == 0)
2408 leaf = path->nodes[0];
2409 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2411 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2415 item_size = btrfs_item_size(leaf, path->slots[0]);
2416 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2417 expected_size = sizeof(*ei) + btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY);
2419 /* No inline refs; we need to bail before checking for owner ref. */
2420 if (item_size == sizeof(*ei))
2423 /* Check for an owner ref; skip over it to the real inline refs. */
2424 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2425 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2426 if (btrfs_fs_incompat(fs_info, SIMPLE_QUOTA) && type == BTRFS_EXTENT_OWNER_REF_KEY) {
2427 expected_size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY);
2428 iref = (struct btrfs_extent_inline_ref *)(iref + 1);
2431 /* If extent item has more than 1 inline ref then it's shared */
2432 if (item_size != expected_size)
2436 * If extent created before last snapshot => it's shared unless the
2437 * snapshot has been deleted. Use the heuristic if strict is false.
2440 (btrfs_extent_generation(leaf, ei) <=
2441 btrfs_root_last_snapshot(&root->root_item)))
2444 /* If this extent has SHARED_DATA_REF then it's shared */
2445 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2446 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2449 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2450 if (btrfs_extent_refs(leaf, ei) !=
2451 btrfs_extent_data_ref_count(leaf, ref) ||
2452 btrfs_extent_data_ref_root(leaf, ref) !=
2453 root->root_key.objectid ||
2454 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2455 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2463 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2464 u64 bytenr, bool strict, struct btrfs_path *path)
2469 ret = check_committed_ref(root, path, objectid,
2470 offset, bytenr, strict);
2471 if (ret && ret != -ENOENT)
2474 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2475 } while (ret == -EAGAIN);
2478 btrfs_release_path(path);
2479 if (btrfs_is_data_reloc_root(root))
2484 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2485 struct btrfs_root *root,
2486 struct extent_buffer *buf,
2487 int full_backref, int inc)
2489 struct btrfs_fs_info *fs_info = root->fs_info;
2495 struct btrfs_key key;
2496 struct btrfs_file_extent_item *fi;
2497 struct btrfs_ref generic_ref = { 0 };
2498 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2504 if (btrfs_is_testing(fs_info))
2507 ref_root = btrfs_header_owner(buf);
2508 nritems = btrfs_header_nritems(buf);
2509 level = btrfs_header_level(buf);
2511 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2515 parent = buf->start;
2519 action = BTRFS_ADD_DELAYED_REF;
2521 action = BTRFS_DROP_DELAYED_REF;
2523 for (i = 0; i < nritems; i++) {
2525 btrfs_item_key_to_cpu(buf, &key, i);
2526 if (key.type != BTRFS_EXTENT_DATA_KEY)
2528 fi = btrfs_item_ptr(buf, i,
2529 struct btrfs_file_extent_item);
2530 if (btrfs_file_extent_type(buf, fi) ==
2531 BTRFS_FILE_EXTENT_INLINE)
2533 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2537 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2538 key.offset -= btrfs_file_extent_offset(buf, fi);
2539 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2540 num_bytes, parent, ref_root);
2541 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2542 key.offset, root->root_key.objectid,
2545 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2547 ret = btrfs_free_extent(trans, &generic_ref);
2551 bytenr = btrfs_node_blockptr(buf, i);
2552 num_bytes = fs_info->nodesize;
2553 /* We don't know the owning_root, use 0. */
2554 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2555 num_bytes, parent, 0);
2556 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root,
2557 root->root_key.objectid, for_reloc);
2559 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2561 ret = btrfs_free_extent(trans, &generic_ref);
2571 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2572 struct extent_buffer *buf, int full_backref)
2574 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2577 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2578 struct extent_buffer *buf, int full_backref)
2580 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2583 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2585 struct btrfs_fs_info *fs_info = root->fs_info;
2590 flags = BTRFS_BLOCK_GROUP_DATA;
2591 else if (root == fs_info->chunk_root)
2592 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2594 flags = BTRFS_BLOCK_GROUP_METADATA;
2596 ret = btrfs_get_alloc_profile(fs_info, flags);
2600 static u64 first_logical_byte(struct btrfs_fs_info *fs_info)
2602 struct rb_node *leftmost;
2605 read_lock(&fs_info->block_group_cache_lock);
2606 /* Get the block group with the lowest logical start address. */
2607 leftmost = rb_first_cached(&fs_info->block_group_cache_tree);
2609 struct btrfs_block_group *bg;
2611 bg = rb_entry(leftmost, struct btrfs_block_group, cache_node);
2614 read_unlock(&fs_info->block_group_cache_lock);
2619 static int pin_down_extent(struct btrfs_trans_handle *trans,
2620 struct btrfs_block_group *cache,
2621 u64 bytenr, u64 num_bytes, int reserved)
2623 struct btrfs_fs_info *fs_info = cache->fs_info;
2625 spin_lock(&cache->space_info->lock);
2626 spin_lock(&cache->lock);
2627 cache->pinned += num_bytes;
2628 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2631 cache->reserved -= num_bytes;
2632 cache->space_info->bytes_reserved -= num_bytes;
2634 spin_unlock(&cache->lock);
2635 spin_unlock(&cache->space_info->lock);
2637 set_extent_bit(&trans->transaction->pinned_extents, bytenr,
2638 bytenr + num_bytes - 1, EXTENT_DIRTY, NULL);
2642 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2643 u64 bytenr, u64 num_bytes, int reserved)
2645 struct btrfs_block_group *cache;
2647 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2648 BUG_ON(!cache); /* Logic error */
2650 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2652 btrfs_put_block_group(cache);
2656 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2657 const struct extent_buffer *eb)
2659 struct btrfs_block_group *cache;
2662 cache = btrfs_lookup_block_group(trans->fs_info, eb->start);
2667 * Fully cache the free space first so that our pin removes the free space
2670 ret = btrfs_cache_block_group(cache, true);
2674 pin_down_extent(trans, cache, eb->start, eb->len, 0);
2676 /* remove us from the free space cache (if we're there at all) */
2677 ret = btrfs_remove_free_space(cache, eb->start, eb->len);
2679 btrfs_put_block_group(cache);
2683 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2684 u64 start, u64 num_bytes)
2687 struct btrfs_block_group *block_group;
2689 block_group = btrfs_lookup_block_group(fs_info, start);
2693 ret = btrfs_cache_block_group(block_group, true);
2697 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2699 btrfs_put_block_group(block_group);
2703 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2705 struct btrfs_fs_info *fs_info = eb->fs_info;
2706 struct btrfs_file_extent_item *item;
2707 struct btrfs_key key;
2712 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2715 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2716 btrfs_item_key_to_cpu(eb, &key, i);
2717 if (key.type != BTRFS_EXTENT_DATA_KEY)
2719 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2720 found_type = btrfs_file_extent_type(eb, item);
2721 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2723 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2725 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2726 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2727 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2736 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2738 atomic_inc(&bg->reservations);
2742 * Returns the free cluster for the given space info and sets empty_cluster to
2743 * what it should be based on the mount options.
2745 static struct btrfs_free_cluster *
2746 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2747 struct btrfs_space_info *space_info, u64 *empty_cluster)
2749 struct btrfs_free_cluster *ret = NULL;
2752 if (btrfs_mixed_space_info(space_info))
2755 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2756 ret = &fs_info->meta_alloc_cluster;
2757 if (btrfs_test_opt(fs_info, SSD))
2758 *empty_cluster = SZ_2M;
2760 *empty_cluster = SZ_64K;
2761 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2762 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2763 *empty_cluster = SZ_2M;
2764 ret = &fs_info->data_alloc_cluster;
2770 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2772 const bool return_free_space)
2774 struct btrfs_block_group *cache = NULL;
2775 struct btrfs_space_info *space_info;
2776 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2777 struct btrfs_free_cluster *cluster = NULL;
2779 u64 total_unpinned = 0;
2780 u64 empty_cluster = 0;
2783 while (start <= end) {
2786 start >= cache->start + cache->length) {
2788 btrfs_put_block_group(cache);
2790 cache = btrfs_lookup_block_group(fs_info, start);
2791 BUG_ON(!cache); /* Logic error */
2793 cluster = fetch_cluster_info(fs_info,
2796 empty_cluster <<= 1;
2799 len = cache->start + cache->length - start;
2800 len = min(len, end + 1 - start);
2802 if (return_free_space)
2803 btrfs_add_free_space(cache, start, len);
2806 total_unpinned += len;
2807 space_info = cache->space_info;
2810 * If this space cluster has been marked as fragmented and we've
2811 * unpinned enough in this block group to potentially allow a
2812 * cluster to be created inside of it go ahead and clear the
2815 if (cluster && cluster->fragmented &&
2816 total_unpinned > empty_cluster) {
2817 spin_lock(&cluster->lock);
2818 cluster->fragmented = 0;
2819 spin_unlock(&cluster->lock);
2822 spin_lock(&space_info->lock);
2823 spin_lock(&cache->lock);
2824 cache->pinned -= len;
2825 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2826 space_info->max_extent_size = 0;
2828 space_info->bytes_readonly += len;
2830 } else if (btrfs_is_zoned(fs_info)) {
2831 /* Need reset before reusing in a zoned block group */
2832 space_info->bytes_zone_unusable += len;
2835 spin_unlock(&cache->lock);
2836 if (!readonly && return_free_space &&
2837 global_rsv->space_info == space_info) {
2838 spin_lock(&global_rsv->lock);
2839 if (!global_rsv->full) {
2840 u64 to_add = min(len, global_rsv->size -
2841 global_rsv->reserved);
2843 global_rsv->reserved += to_add;
2844 btrfs_space_info_update_bytes_may_use(fs_info,
2845 space_info, to_add);
2846 if (global_rsv->reserved >= global_rsv->size)
2847 global_rsv->full = 1;
2850 spin_unlock(&global_rsv->lock);
2852 /* Add to any tickets we may have */
2853 if (!readonly && return_free_space && len)
2854 btrfs_try_granting_tickets(fs_info, space_info);
2855 spin_unlock(&space_info->lock);
2859 btrfs_put_block_group(cache);
2863 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2865 struct btrfs_fs_info *fs_info = trans->fs_info;
2866 struct btrfs_block_group *block_group, *tmp;
2867 struct list_head *deleted_bgs;
2868 struct extent_io_tree *unpin;
2873 unpin = &trans->transaction->pinned_extents;
2875 while (!TRANS_ABORTED(trans)) {
2876 struct extent_state *cached_state = NULL;
2878 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2879 if (!find_first_extent_bit(unpin, 0, &start, &end,
2880 EXTENT_DIRTY, &cached_state)) {
2881 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2885 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2886 ret = btrfs_discard_extent(fs_info, start,
2887 end + 1 - start, NULL);
2889 clear_extent_dirty(unpin, start, end, &cached_state);
2890 unpin_extent_range(fs_info, start, end, true);
2891 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2892 free_extent_state(cached_state);
2896 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2897 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2898 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2902 * Transaction is finished. We don't need the lock anymore. We
2903 * do need to clean up the block groups in case of a transaction
2906 deleted_bgs = &trans->transaction->deleted_bgs;
2907 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2911 if (!TRANS_ABORTED(trans))
2912 ret = btrfs_discard_extent(fs_info,
2914 block_group->length,
2917 list_del_init(&block_group->bg_list);
2918 btrfs_unfreeze_block_group(block_group);
2919 btrfs_put_block_group(block_group);
2922 const char *errstr = btrfs_decode_error(ret);
2924 "discard failed while removing blockgroup: errno=%d %s",
2933 * Parse an extent item's inline extents looking for a simple quotas owner ref.
2935 * @fs_info: the btrfs_fs_info for this mount
2936 * @leaf: a leaf in the extent tree containing the extent item
2937 * @slot: the slot in the leaf where the extent item is found
2939 * Returns the objectid of the root that originally allocated the extent item
2940 * if the inline owner ref is expected and present, otherwise 0.
2942 * If an extent item has an owner ref item, it will be the first inline ref
2943 * item. Therefore the logic is to check whether there are any inline ref
2944 * items, then check the type of the first one.
2946 u64 btrfs_get_extent_owner_root(struct btrfs_fs_info *fs_info,
2947 struct extent_buffer *leaf, int slot)
2949 struct btrfs_extent_item *ei;
2950 struct btrfs_extent_inline_ref *iref;
2951 struct btrfs_extent_owner_ref *oref;
2956 if (!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA))
2959 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
2960 ptr = (unsigned long)(ei + 1);
2961 end = (unsigned long)ei + btrfs_item_size(leaf, slot);
2963 /* No inline ref items of any kind, can't check type. */
2967 iref = (struct btrfs_extent_inline_ref *)ptr;
2968 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
2970 /* We found an owner ref, get the root out of it. */
2971 if (type == BTRFS_EXTENT_OWNER_REF_KEY) {
2972 oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
2973 return btrfs_extent_owner_ref_root_id(leaf, oref);
2976 /* We have inline refs, but not an owner ref. */
2980 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
2981 u64 bytenr, struct btrfs_squota_delta *delta)
2984 u64 num_bytes = delta->num_bytes;
2986 if (delta->is_data) {
2987 struct btrfs_root *csum_root;
2989 csum_root = btrfs_csum_root(trans->fs_info, bytenr);
2990 ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes);
2992 btrfs_abort_transaction(trans, ret);
2996 ret = btrfs_delete_raid_extent(trans, bytenr, num_bytes);
2998 btrfs_abort_transaction(trans, ret);
3003 ret = btrfs_record_squota_delta(trans->fs_info, delta);
3005 btrfs_abort_transaction(trans, ret);
3009 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
3011 btrfs_abort_transaction(trans, ret);
3015 ret = btrfs_update_block_group(trans, bytenr, num_bytes, false);
3017 btrfs_abort_transaction(trans, ret);
3022 #define abort_and_dump(trans, path, fmt, args...) \
3024 btrfs_abort_transaction(trans, -EUCLEAN); \
3025 btrfs_print_leaf(path->nodes[0]); \
3026 btrfs_crit(trans->fs_info, fmt, ##args); \
3030 * Drop one or more refs of @node.
3032 * 1. Locate the extent refs.
3033 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
3034 * Locate it, then reduce the refs number or remove the ref line completely.
3036 * 2. Update the refs count in EXTENT/METADATA_ITEM
3038 * Inline backref case:
3040 * in extent tree we have:
3042 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
3043 * refs 2 gen 6 flags DATA
3044 * extent data backref root FS_TREE objectid 258 offset 0 count 1
3045 * extent data backref root FS_TREE objectid 257 offset 0 count 1
3047 * This function gets called with:
3049 * node->bytenr = 13631488
3050 * node->num_bytes = 1048576
3051 * root_objectid = FS_TREE
3052 * owner_objectid = 257
3056 * Then we should get some like:
3058 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
3059 * refs 1 gen 6 flags DATA
3060 * extent data backref root FS_TREE objectid 258 offset 0 count 1
3062 * Keyed backref case:
3064 * in extent tree we have:
3066 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
3067 * refs 754 gen 6 flags DATA
3069 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
3070 * extent data backref root FS_TREE objectid 866 offset 0 count 1
3072 * This function get called with:
3074 * node->bytenr = 13631488
3075 * node->num_bytes = 1048576
3076 * root_objectid = FS_TREE
3077 * owner_objectid = 866
3081 * Then we should get some like:
3083 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
3084 * refs 753 gen 6 flags DATA
3086 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
3088 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3089 struct btrfs_delayed_ref_head *href,
3090 struct btrfs_delayed_ref_node *node, u64 parent,
3091 u64 root_objectid, u64 owner_objectid,
3093 struct btrfs_delayed_extent_op *extent_op)
3095 struct btrfs_fs_info *info = trans->fs_info;
3096 struct btrfs_key key;
3097 struct btrfs_path *path;
3098 struct btrfs_root *extent_root;
3099 struct extent_buffer *leaf;
3100 struct btrfs_extent_item *ei;
3101 struct btrfs_extent_inline_ref *iref;
3104 int extent_slot = 0;
3105 int found_extent = 0;
3107 int refs_to_drop = node->ref_mod;
3110 u64 bytenr = node->bytenr;
3111 u64 num_bytes = node->num_bytes;
3112 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
3113 u64 delayed_ref_root = href->owning_root;
3115 extent_root = btrfs_extent_root(info, bytenr);
3116 ASSERT(extent_root);
3118 path = btrfs_alloc_path();
3122 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3124 if (!is_data && refs_to_drop != 1) {
3126 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
3127 node->bytenr, refs_to_drop);
3129 btrfs_abort_transaction(trans, ret);
3134 skinny_metadata = false;
3136 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
3137 parent, root_objectid, owner_objectid,
3141 * Either the inline backref or the SHARED_DATA_REF/
3142 * SHARED_BLOCK_REF is found
3144 * Here is a quick path to locate EXTENT/METADATA_ITEM.
3145 * It's possible the EXTENT/METADATA_ITEM is near current slot.
3147 extent_slot = path->slots[0];
3148 while (extent_slot >= 0) {
3149 btrfs_item_key_to_cpu(path->nodes[0], &key,
3151 if (key.objectid != bytenr)
3153 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3154 key.offset == num_bytes) {
3158 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3159 key.offset == owner_objectid) {
3164 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
3165 if (path->slots[0] - extent_slot > 5)
3170 if (!found_extent) {
3172 abort_and_dump(trans, path,
3173 "invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref",
3178 /* Must be SHARED_* item, remove the backref first */
3179 ret = remove_extent_backref(trans, extent_root, path,
3180 NULL, refs_to_drop, is_data);
3182 btrfs_abort_transaction(trans, ret);
3185 btrfs_release_path(path);
3187 /* Slow path to locate EXTENT/METADATA_ITEM */
3188 key.objectid = bytenr;
3189 key.type = BTRFS_EXTENT_ITEM_KEY;
3190 key.offset = num_bytes;
3192 if (!is_data && skinny_metadata) {
3193 key.type = BTRFS_METADATA_ITEM_KEY;
3194 key.offset = owner_objectid;
3197 ret = btrfs_search_slot(trans, extent_root,
3199 if (ret > 0 && skinny_metadata && path->slots[0]) {
3201 * Couldn't find our skinny metadata item,
3202 * see if we have ye olde extent item.
3205 btrfs_item_key_to_cpu(path->nodes[0], &key,
3207 if (key.objectid == bytenr &&
3208 key.type == BTRFS_EXTENT_ITEM_KEY &&
3209 key.offset == num_bytes)
3213 if (ret > 0 && skinny_metadata) {
3214 skinny_metadata = false;
3215 key.objectid = bytenr;
3216 key.type = BTRFS_EXTENT_ITEM_KEY;
3217 key.offset = num_bytes;
3218 btrfs_release_path(path);
3219 ret = btrfs_search_slot(trans, extent_root,
3225 btrfs_print_leaf(path->nodes[0]);
3227 "umm, got %d back from search, was looking for %llu, slot %d",
3228 ret, bytenr, path->slots[0]);
3231 btrfs_abort_transaction(trans, ret);
3234 extent_slot = path->slots[0];
3236 } else if (WARN_ON(ret == -ENOENT)) {
3237 abort_and_dump(trans, path,
3238 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d",
3239 bytenr, parent, root_objectid, owner_objectid,
3240 owner_offset, path->slots[0]);
3243 btrfs_abort_transaction(trans, ret);
3247 leaf = path->nodes[0];
3248 item_size = btrfs_item_size(leaf, extent_slot);
3249 if (unlikely(item_size < sizeof(*ei))) {
3251 btrfs_err(trans->fs_info,
3252 "unexpected extent item size, has %u expect >= %zu",
3253 item_size, sizeof(*ei));
3254 btrfs_abort_transaction(trans, ret);
3257 ei = btrfs_item_ptr(leaf, extent_slot,
3258 struct btrfs_extent_item);
3259 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3260 key.type == BTRFS_EXTENT_ITEM_KEY) {
3261 struct btrfs_tree_block_info *bi;
3263 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3264 abort_and_dump(trans, path,
3265 "invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu",
3266 key.objectid, key.type, key.offset,
3267 path->slots[0], owner_objectid, item_size,
3268 sizeof(*ei) + sizeof(*bi));
3272 bi = (struct btrfs_tree_block_info *)(ei + 1);
3273 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3276 refs = btrfs_extent_refs(leaf, ei);
3277 if (refs < refs_to_drop) {
3278 abort_and_dump(trans, path,
3279 "trying to drop %d refs but we only have %llu for bytenr %llu slot %u",
3280 refs_to_drop, refs, bytenr, path->slots[0]);
3284 refs -= refs_to_drop;
3288 __run_delayed_extent_op(extent_op, leaf, ei);
3290 * In the case of inline back ref, reference count will
3291 * be updated by remove_extent_backref
3294 if (!found_extent) {
3295 abort_and_dump(trans, path,
3296 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u",
3302 btrfs_set_extent_refs(leaf, ei, refs);
3303 btrfs_mark_buffer_dirty(trans, leaf);
3306 ret = remove_extent_backref(trans, extent_root, path,
3307 iref, refs_to_drop, is_data);
3309 btrfs_abort_transaction(trans, ret);
3314 struct btrfs_squota_delta delta = {
3315 .root = delayed_ref_root,
3316 .num_bytes = num_bytes,
3319 .generation = btrfs_extent_generation(leaf, ei),
3322 /* In this branch refs == 1 */
3324 if (is_data && refs_to_drop !=
3325 extent_data_ref_count(path, iref)) {
3326 abort_and_dump(trans, path,
3327 "invalid refs_to_drop, current refs %u refs_to_drop %u slot %u",
3328 extent_data_ref_count(path, iref),
3329 refs_to_drop, path->slots[0]);
3334 if (path->slots[0] != extent_slot) {
3335 abort_and_dump(trans, path,
3336 "invalid iref, extent item key (%llu %u %llu) slot %u doesn't have wanted iref",
3337 key.objectid, key.type,
3338 key.offset, path->slots[0]);
3344 * No inline ref, we must be at SHARED_* item,
3345 * And it's single ref, it must be:
3346 * | extent_slot ||extent_slot + 1|
3347 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3349 if (path->slots[0] != extent_slot + 1) {
3350 abort_and_dump(trans, path,
3351 "invalid SHARED_* item slot %u, previous item is not EXTENT/METADATA_ITEM",
3356 path->slots[0] = extent_slot;
3361 * We can't infer the data owner from the delayed ref, so we need
3362 * to try to get it from the owning ref item.
3364 * If it is not present, then that extent was not written under
3365 * simple quotas mode, so we don't need to account for its deletion.
3368 delta.root = btrfs_get_extent_owner_root(trans->fs_info,
3371 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3374 btrfs_abort_transaction(trans, ret);
3377 btrfs_release_path(path);
3379 ret = do_free_extent_accounting(trans, bytenr, &delta);
3381 btrfs_release_path(path);
3384 btrfs_free_path(path);
3389 * when we free an block, it is possible (and likely) that we free the last
3390 * delayed ref for that extent as well. This searches the delayed ref tree for
3391 * a given extent, and if there are no other delayed refs to be processed, it
3392 * removes it from the tree.
3394 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3397 struct btrfs_delayed_ref_head *head;
3398 struct btrfs_delayed_ref_root *delayed_refs;
3401 delayed_refs = &trans->transaction->delayed_refs;
3402 spin_lock(&delayed_refs->lock);
3403 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3405 goto out_delayed_unlock;
3407 spin_lock(&head->lock);
3408 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3411 if (cleanup_extent_op(head) != NULL)
3415 * waiting for the lock here would deadlock. If someone else has it
3416 * locked they are already in the process of dropping it anyway
3418 if (!mutex_trylock(&head->mutex))
3421 btrfs_delete_ref_head(delayed_refs, head);
3422 head->processing = false;
3424 spin_unlock(&head->lock);
3425 spin_unlock(&delayed_refs->lock);
3427 BUG_ON(head->extent_op);
3428 if (head->must_insert_reserved)
3431 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3432 mutex_unlock(&head->mutex);
3433 btrfs_put_delayed_ref_head(head);
3436 spin_unlock(&head->lock);
3439 spin_unlock(&delayed_refs->lock);
3443 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3445 struct extent_buffer *buf,
3446 u64 parent, int last_ref)
3448 struct btrfs_fs_info *fs_info = trans->fs_info;
3449 struct btrfs_ref generic_ref = { 0 };
3452 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3453 buf->start, buf->len, parent, btrfs_header_owner(buf));
3454 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3457 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3458 btrfs_ref_tree_mod(fs_info, &generic_ref);
3459 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3460 BUG_ON(ret); /* -ENOMEM */
3463 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3464 struct btrfs_block_group *cache;
3465 bool must_pin = false;
3467 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3468 ret = check_ref_cleanup(trans, buf->start);
3470 btrfs_redirty_list_add(trans->transaction, buf);
3475 cache = btrfs_lookup_block_group(fs_info, buf->start);
3477 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3478 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3479 btrfs_put_block_group(cache);
3484 * If there are tree mod log users we may have recorded mod log
3485 * operations for this node. If we re-allocate this node we
3486 * could replay operations on this node that happened when it
3487 * existed in a completely different root. For example if it
3488 * was part of root A, then was reallocated to root B, and we
3489 * are doing a btrfs_old_search_slot(root b), we could replay
3490 * operations that happened when the block was part of root A,
3491 * giving us an inconsistent view of the btree.
3493 * We are safe from races here because at this point no other
3494 * node or root points to this extent buffer, so if after this
3495 * check a new tree mod log user joins we will not have an
3496 * existing log of operations on this node that we have to
3499 if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
3502 if (must_pin || btrfs_is_zoned(fs_info)) {
3503 btrfs_redirty_list_add(trans->transaction, buf);
3504 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3505 btrfs_put_block_group(cache);
3509 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3511 btrfs_add_free_space(cache, buf->start, buf->len);
3512 btrfs_free_reserved_bytes(cache, buf->len, 0);
3513 btrfs_put_block_group(cache);
3514 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3519 * Deleting the buffer, clear the corrupt flag since it doesn't
3522 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3526 /* Can return -ENOMEM */
3527 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3529 struct btrfs_fs_info *fs_info = trans->fs_info;
3532 if (btrfs_is_testing(fs_info))
3536 * tree log blocks never actually go into the extent allocation
3537 * tree, just update pinning info and exit early.
3539 if ((ref->type == BTRFS_REF_METADATA &&
3540 ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID) ||
3541 (ref->type == BTRFS_REF_DATA &&
3542 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) {
3543 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3545 } else if (ref->type == BTRFS_REF_METADATA) {
3546 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3548 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3551 if (!((ref->type == BTRFS_REF_METADATA &&
3552 ref->tree_ref.ref_root == BTRFS_TREE_LOG_OBJECTID) ||
3553 (ref->type == BTRFS_REF_DATA &&
3554 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)))
3555 btrfs_ref_tree_mod(fs_info, ref);
3560 enum btrfs_loop_type {
3562 * Start caching block groups but do not wait for progress or for them
3565 LOOP_CACHING_NOWAIT,
3568 * Wait for the block group free_space >= the space we're waiting for if
3569 * the block group isn't cached.
3574 * Allow allocations to happen from block groups that do not yet have a
3575 * size classification.
3577 LOOP_UNSET_SIZE_CLASS,
3580 * Allocate a chunk and then retry the allocation.
3585 * Ignore the size class restrictions for this allocation.
3587 LOOP_WRONG_SIZE_CLASS,
3590 * Ignore the empty size, only try to allocate the number of bytes
3591 * needed for this allocation.
3597 btrfs_lock_block_group(struct btrfs_block_group *cache,
3601 down_read(&cache->data_rwsem);
3604 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3607 btrfs_get_block_group(cache);
3609 down_read(&cache->data_rwsem);
3612 static struct btrfs_block_group *btrfs_lock_cluster(
3613 struct btrfs_block_group *block_group,
3614 struct btrfs_free_cluster *cluster,
3616 __acquires(&cluster->refill_lock)
3618 struct btrfs_block_group *used_bg = NULL;
3620 spin_lock(&cluster->refill_lock);
3622 used_bg = cluster->block_group;
3626 if (used_bg == block_group)
3629 btrfs_get_block_group(used_bg);
3634 if (down_read_trylock(&used_bg->data_rwsem))
3637 spin_unlock(&cluster->refill_lock);
3639 /* We should only have one-level nested. */
3640 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3642 spin_lock(&cluster->refill_lock);
3643 if (used_bg == cluster->block_group)
3646 up_read(&used_bg->data_rwsem);
3647 btrfs_put_block_group(used_bg);
3652 btrfs_release_block_group(struct btrfs_block_group *cache,
3656 up_read(&cache->data_rwsem);
3657 btrfs_put_block_group(cache);
3661 * Helper function for find_free_extent().
3663 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3664 * Return >0 to inform caller that we find nothing
3665 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3667 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3668 struct find_free_extent_ctl *ffe_ctl,
3669 struct btrfs_block_group **cluster_bg_ret)
3671 struct btrfs_block_group *cluster_bg;
3672 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3673 u64 aligned_cluster;
3677 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3679 goto refill_cluster;
3680 if (cluster_bg != bg && (cluster_bg->ro ||
3681 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3682 goto release_cluster;
3684 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3685 ffe_ctl->num_bytes, cluster_bg->start,
3686 &ffe_ctl->max_extent_size);
3688 /* We have a block, we're done */
3689 spin_unlock(&last_ptr->refill_lock);
3690 trace_btrfs_reserve_extent_cluster(cluster_bg, ffe_ctl);
3691 *cluster_bg_ret = cluster_bg;
3692 ffe_ctl->found_offset = offset;
3695 WARN_ON(last_ptr->block_group != cluster_bg);
3699 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3700 * lets just skip it and let the allocator find whatever block it can
3701 * find. If we reach this point, we will have tried the cluster
3702 * allocator plenty of times and not have found anything, so we are
3703 * likely way too fragmented for the clustering stuff to find anything.
3705 * However, if the cluster is taken from the current block group,
3706 * release the cluster first, so that we stand a better chance of
3707 * succeeding in the unclustered allocation.
3709 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3710 spin_unlock(&last_ptr->refill_lock);
3711 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3715 /* This cluster didn't work out, free it and start over */
3716 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3718 if (cluster_bg != bg)
3719 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3722 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3723 spin_unlock(&last_ptr->refill_lock);
3727 aligned_cluster = max_t(u64,
3728 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3729 bg->full_stripe_len);
3730 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3731 ffe_ctl->num_bytes, aligned_cluster);
3733 /* Now pull our allocation out of this cluster */
3734 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3735 ffe_ctl->num_bytes, ffe_ctl->search_start,
3736 &ffe_ctl->max_extent_size);
3738 /* We found one, proceed */
3739 spin_unlock(&last_ptr->refill_lock);
3740 ffe_ctl->found_offset = offset;
3741 trace_btrfs_reserve_extent_cluster(bg, ffe_ctl);
3746 * At this point we either didn't find a cluster or we weren't able to
3747 * allocate a block from our cluster. Free the cluster we've been
3748 * trying to use, and go to the next block group.
3750 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3751 spin_unlock(&last_ptr->refill_lock);
3756 * Return >0 to inform caller that we find nothing
3757 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3759 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3760 struct find_free_extent_ctl *ffe_ctl)
3762 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3766 * We are doing an unclustered allocation, set the fragmented flag so
3767 * we don't bother trying to setup a cluster again until we get more
3770 if (unlikely(last_ptr)) {
3771 spin_lock(&last_ptr->lock);
3772 last_ptr->fragmented = 1;
3773 spin_unlock(&last_ptr->lock);
3775 if (ffe_ctl->cached) {
3776 struct btrfs_free_space_ctl *free_space_ctl;
3778 free_space_ctl = bg->free_space_ctl;
3779 spin_lock(&free_space_ctl->tree_lock);
3780 if (free_space_ctl->free_space <
3781 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3782 ffe_ctl->empty_size) {
3783 ffe_ctl->total_free_space = max_t(u64,
3784 ffe_ctl->total_free_space,
3785 free_space_ctl->free_space);
3786 spin_unlock(&free_space_ctl->tree_lock);
3789 spin_unlock(&free_space_ctl->tree_lock);
3792 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3793 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3794 &ffe_ctl->max_extent_size);
3797 ffe_ctl->found_offset = offset;
3801 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3802 struct find_free_extent_ctl *ffe_ctl,
3803 struct btrfs_block_group **bg_ret)
3807 /* We want to try and use the cluster allocator, so lets look there */
3808 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3809 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3812 /* ret == -ENOENT case falls through */
3815 return find_free_extent_unclustered(block_group, ffe_ctl);
3819 * Tree-log block group locking
3820 * ============================
3822 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3823 * indicates the starting address of a block group, which is reserved only
3824 * for tree-log metadata.
3831 * fs_info::treelog_bg_lock
3835 * Simple allocator for sequential-only block group. It only allows sequential
3836 * allocation. No need to play with trees. This function also reserves the
3837 * bytes as in btrfs_add_reserved_bytes.
3839 static int do_allocation_zoned(struct btrfs_block_group *block_group,
3840 struct find_free_extent_ctl *ffe_ctl,
3841 struct btrfs_block_group **bg_ret)
3843 struct btrfs_fs_info *fs_info = block_group->fs_info;
3844 struct btrfs_space_info *space_info = block_group->space_info;
3845 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3846 u64 start = block_group->start;
3847 u64 num_bytes = ffe_ctl->num_bytes;
3849 u64 bytenr = block_group->start;
3851 u64 data_reloc_bytenr;
3855 ASSERT(btrfs_is_zoned(block_group->fs_info));
3858 * Do not allow non-tree-log blocks in the dedicated tree-log block
3859 * group, and vice versa.
3861 spin_lock(&fs_info->treelog_bg_lock);
3862 log_bytenr = fs_info->treelog_bg;
3863 if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3864 (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3866 spin_unlock(&fs_info->treelog_bg_lock);
3871 * Do not allow non-relocation blocks in the dedicated relocation block
3872 * group, and vice versa.
3874 spin_lock(&fs_info->relocation_bg_lock);
3875 data_reloc_bytenr = fs_info->data_reloc_bg;
3876 if (data_reloc_bytenr &&
3877 ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3878 (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3880 spin_unlock(&fs_info->relocation_bg_lock);
3884 /* Check RO and no space case before trying to activate it */
3885 spin_lock(&block_group->lock);
3886 if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) {
3889 * May need to clear fs_info->{treelog,data_reloc}_bg.
3890 * Return the error after taking the locks.
3893 spin_unlock(&block_group->lock);
3895 /* Metadata block group is activated at write time. */
3896 if (!ret && (block_group->flags & BTRFS_BLOCK_GROUP_DATA) &&
3897 !btrfs_zone_activate(block_group)) {
3900 * May need to clear fs_info->{treelog,data_reloc}_bg.
3901 * Return the error after taking the locks.
3905 spin_lock(&space_info->lock);
3906 spin_lock(&block_group->lock);
3907 spin_lock(&fs_info->treelog_bg_lock);
3908 spin_lock(&fs_info->relocation_bg_lock);
3913 ASSERT(!ffe_ctl->for_treelog ||
3914 block_group->start == fs_info->treelog_bg ||
3915 fs_info->treelog_bg == 0);
3916 ASSERT(!ffe_ctl->for_data_reloc ||
3917 block_group->start == fs_info->data_reloc_bg ||
3918 fs_info->data_reloc_bg == 0);
3920 if (block_group->ro ||
3921 (!ffe_ctl->for_data_reloc &&
3922 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))) {
3928 * Do not allow currently using block group to be tree-log dedicated
3931 if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3932 (block_group->used || block_group->reserved)) {
3938 * Do not allow currently used block group to be the data relocation
3939 * dedicated block group.
3941 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3942 (block_group->used || block_group->reserved)) {
3947 WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3948 avail = block_group->zone_capacity - block_group->alloc_offset;
3949 if (avail < num_bytes) {
3950 if (ffe_ctl->max_extent_size < avail) {
3952 * With sequential allocator, free space is always
3955 ffe_ctl->max_extent_size = avail;
3956 ffe_ctl->total_free_space = avail;
3962 if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3963 fs_info->treelog_bg = block_group->start;
3965 if (ffe_ctl->for_data_reloc) {
3966 if (!fs_info->data_reloc_bg)
3967 fs_info->data_reloc_bg = block_group->start;
3969 * Do not allow allocations from this block group, unless it is
3970 * for data relocation. Compared to increasing the ->ro, setting
3971 * the ->zoned_data_reloc_ongoing flag still allows nocow
3972 * writers to come in. See btrfs_inc_nocow_writers().
3974 * We need to disable an allocation to avoid an allocation of
3975 * regular (non-relocation data) extent. With mix of relocation
3976 * extents and regular extents, we can dispatch WRITE commands
3977 * (for relocation extents) and ZONE APPEND commands (for
3978 * regular extents) at the same time to the same zone, which
3979 * easily break the write pointer.
3981 * Also, this flag avoids this block group to be zone finished.
3983 set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags);
3986 ffe_ctl->found_offset = start + block_group->alloc_offset;
3987 block_group->alloc_offset += num_bytes;
3988 spin_lock(&ctl->tree_lock);
3989 ctl->free_space -= num_bytes;
3990 spin_unlock(&ctl->tree_lock);
3993 * We do not check if found_offset is aligned to stripesize. The
3994 * address is anyway rewritten when using zone append writing.
3997 ffe_ctl->search_start = ffe_ctl->found_offset;
4000 if (ret && ffe_ctl->for_treelog)
4001 fs_info->treelog_bg = 0;
4002 if (ret && ffe_ctl->for_data_reloc)
4003 fs_info->data_reloc_bg = 0;
4004 spin_unlock(&fs_info->relocation_bg_lock);
4005 spin_unlock(&fs_info->treelog_bg_lock);
4006 spin_unlock(&block_group->lock);
4007 spin_unlock(&space_info->lock);
4011 static int do_allocation(struct btrfs_block_group *block_group,
4012 struct find_free_extent_ctl *ffe_ctl,
4013 struct btrfs_block_group **bg_ret)
4015 switch (ffe_ctl->policy) {
4016 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4017 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
4018 case BTRFS_EXTENT_ALLOC_ZONED:
4019 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
4025 static void release_block_group(struct btrfs_block_group *block_group,
4026 struct find_free_extent_ctl *ffe_ctl,
4029 switch (ffe_ctl->policy) {
4030 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4031 ffe_ctl->retry_uncached = false;
4033 case BTRFS_EXTENT_ALLOC_ZONED:
4040 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
4042 btrfs_release_block_group(block_group, delalloc);
4045 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
4046 struct btrfs_key *ins)
4048 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4050 if (!ffe_ctl->use_cluster && last_ptr) {
4051 spin_lock(&last_ptr->lock);
4052 last_ptr->window_start = ins->objectid;
4053 spin_unlock(&last_ptr->lock);
4057 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
4058 struct btrfs_key *ins)
4060 switch (ffe_ctl->policy) {
4061 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4062 found_extent_clustered(ffe_ctl, ins);
4064 case BTRFS_EXTENT_ALLOC_ZONED:
4072 static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info,
4073 struct find_free_extent_ctl *ffe_ctl)
4075 /* Block group's activeness is not a requirement for METADATA block groups. */
4076 if (!(ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA))
4079 /* If we can activate new zone, just allocate a chunk and use it */
4080 if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags))
4084 * We already reached the max active zones. Try to finish one block
4085 * group to make a room for a new block group. This is only possible
4086 * for a data block group because btrfs_zone_finish() may need to wait
4087 * for a running transaction which can cause a deadlock for metadata
4090 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
4091 int ret = btrfs_zone_finish_one_bg(fs_info);
4100 * If we have enough free space left in an already active block group
4101 * and we can't activate any other zone now, do not allow allocating a
4102 * new chunk and let find_free_extent() retry with a smaller size.
4104 if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size)
4108 * Even min_alloc_size is not left in any block groups. Since we cannot
4109 * activate a new block group, allocating it may not help. Let's tell a
4110 * caller to try again and hope it progress something by writing some
4111 * parts of the region. That is only possible for data block groups,
4112 * where a part of the region can be written.
4114 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)
4118 * We cannot activate a new block group and no enough space left in any
4119 * block groups. So, allocating a new block group may not help. But,
4120 * there is nothing to do anyway, so let's go with it.
4125 static int can_allocate_chunk(struct btrfs_fs_info *fs_info,
4126 struct find_free_extent_ctl *ffe_ctl)
4128 switch (ffe_ctl->policy) {
4129 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4131 case BTRFS_EXTENT_ALLOC_ZONED:
4132 return can_allocate_chunk_zoned(fs_info, ffe_ctl);
4139 * Return >0 means caller needs to re-search for free extent
4140 * Return 0 means we have the needed free extent.
4141 * Return <0 means we failed to locate any free extent.
4143 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
4144 struct btrfs_key *ins,
4145 struct find_free_extent_ctl *ffe_ctl,
4148 struct btrfs_root *root = fs_info->chunk_root;
4151 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
4152 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
4153 ffe_ctl->orig_have_caching_bg = true;
4155 if (ins->objectid) {
4156 found_extent(ffe_ctl, ins);
4160 if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
4164 if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
4167 /* See the comments for btrfs_loop_type for an explanation of the phases. */
4168 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
4171 * We want to skip the LOOP_CACHING_WAIT step if we don't have
4172 * any uncached bgs and we've already done a full search
4175 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT &&
4176 (!ffe_ctl->orig_have_caching_bg && full_search))
4180 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
4181 struct btrfs_trans_handle *trans;
4184 /* Check if allocation policy allows to create a new chunk */
4185 ret = can_allocate_chunk(fs_info, ffe_ctl);
4189 trans = current->journal_info;
4193 trans = btrfs_join_transaction(root);
4195 if (IS_ERR(trans)) {
4196 ret = PTR_ERR(trans);
4200 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
4201 CHUNK_ALLOC_FORCE_FOR_EXTENT);
4203 /* Do not bail out on ENOSPC since we can do more. */
4204 if (ret == -ENOSPC) {
4209 btrfs_abort_transaction(trans, ret);
4213 btrfs_end_transaction(trans);
4218 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4219 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4223 * Don't loop again if we already have no empty_size and
4226 if (ffe_ctl->empty_size == 0 &&
4227 ffe_ctl->empty_cluster == 0)
4229 ffe_ctl->empty_size = 0;
4230 ffe_ctl->empty_cluster = 0;
4237 static bool find_free_extent_check_size_class(struct find_free_extent_ctl *ffe_ctl,
4238 struct btrfs_block_group *bg)
4240 if (ffe_ctl->policy == BTRFS_EXTENT_ALLOC_ZONED)
4242 if (!btrfs_block_group_should_use_size_class(bg))
4244 if (ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS)
4246 if (ffe_ctl->loop >= LOOP_UNSET_SIZE_CLASS &&
4247 bg->size_class == BTRFS_BG_SZ_NONE)
4249 return ffe_ctl->size_class == bg->size_class;
4252 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4253 struct find_free_extent_ctl *ffe_ctl,
4254 struct btrfs_space_info *space_info,
4255 struct btrfs_key *ins)
4258 * If our free space is heavily fragmented we may not be able to make
4259 * big contiguous allocations, so instead of doing the expensive search
4260 * for free space, simply return ENOSPC with our max_extent_size so we
4261 * can go ahead and search for a more manageable chunk.
4263 * If our max_extent_size is large enough for our allocation simply
4264 * disable clustering since we will likely not be able to find enough
4265 * space to create a cluster and induce latency trying.
4267 if (space_info->max_extent_size) {
4268 spin_lock(&space_info->lock);
4269 if (space_info->max_extent_size &&
4270 ffe_ctl->num_bytes > space_info->max_extent_size) {
4271 ins->offset = space_info->max_extent_size;
4272 spin_unlock(&space_info->lock);
4274 } else if (space_info->max_extent_size) {
4275 ffe_ctl->use_cluster = false;
4277 spin_unlock(&space_info->lock);
4280 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4281 &ffe_ctl->empty_cluster);
4282 if (ffe_ctl->last_ptr) {
4283 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4285 spin_lock(&last_ptr->lock);
4286 if (last_ptr->block_group)
4287 ffe_ctl->hint_byte = last_ptr->window_start;
4288 if (last_ptr->fragmented) {
4290 * We still set window_start so we can keep track of the
4291 * last place we found an allocation to try and save
4294 ffe_ctl->hint_byte = last_ptr->window_start;
4295 ffe_ctl->use_cluster = false;
4297 spin_unlock(&last_ptr->lock);
4303 static int prepare_allocation(struct btrfs_fs_info *fs_info,
4304 struct find_free_extent_ctl *ffe_ctl,
4305 struct btrfs_space_info *space_info,
4306 struct btrfs_key *ins)
4308 switch (ffe_ctl->policy) {
4309 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4310 return prepare_allocation_clustered(fs_info, ffe_ctl,
4312 case BTRFS_EXTENT_ALLOC_ZONED:
4313 if (ffe_ctl->for_treelog) {
4314 spin_lock(&fs_info->treelog_bg_lock);
4315 if (fs_info->treelog_bg)
4316 ffe_ctl->hint_byte = fs_info->treelog_bg;
4317 spin_unlock(&fs_info->treelog_bg_lock);
4319 if (ffe_ctl->for_data_reloc) {
4320 spin_lock(&fs_info->relocation_bg_lock);
4321 if (fs_info->data_reloc_bg)
4322 ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4323 spin_unlock(&fs_info->relocation_bg_lock);
4332 * walks the btree of allocated extents and find a hole of a given size.
4333 * The key ins is changed to record the hole:
4334 * ins->objectid == start position
4335 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4336 * ins->offset == the size of the hole.
4337 * Any available blocks before search_start are skipped.
4339 * If there is no suitable free space, we will record the max size of
4340 * the free space extent currently.
4342 * The overall logic and call chain:
4344 * find_free_extent()
4345 * |- Iterate through all block groups
4346 * | |- Get a valid block group
4347 * | |- Try to do clustered allocation in that block group
4348 * | |- Try to do unclustered allocation in that block group
4349 * | |- Check if the result is valid
4350 * | | |- If valid, then exit
4351 * | |- Jump to next block group
4353 * |- Push harder to find free extents
4354 * |- If not found, re-iterate all block groups
4356 static noinline int find_free_extent(struct btrfs_root *root,
4357 struct btrfs_key *ins,
4358 struct find_free_extent_ctl *ffe_ctl)
4360 struct btrfs_fs_info *fs_info = root->fs_info;
4362 int cache_block_group_error = 0;
4363 struct btrfs_block_group *block_group = NULL;
4364 struct btrfs_space_info *space_info;
4365 bool full_search = false;
4367 WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4369 ffe_ctl->search_start = 0;
4370 /* For clustered allocation */
4371 ffe_ctl->empty_cluster = 0;
4372 ffe_ctl->last_ptr = NULL;
4373 ffe_ctl->use_cluster = true;
4374 ffe_ctl->have_caching_bg = false;
4375 ffe_ctl->orig_have_caching_bg = false;
4376 ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4378 ffe_ctl->retry_uncached = false;
4379 ffe_ctl->cached = 0;
4380 ffe_ctl->max_extent_size = 0;
4381 ffe_ctl->total_free_space = 0;
4382 ffe_ctl->found_offset = 0;
4383 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4384 ffe_ctl->size_class = btrfs_calc_block_group_size_class(ffe_ctl->num_bytes);
4386 if (btrfs_is_zoned(fs_info))
4387 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4389 ins->type = BTRFS_EXTENT_ITEM_KEY;
4393 trace_find_free_extent(root, ffe_ctl);
4395 space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4397 btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4401 ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4405 ffe_ctl->search_start = max(ffe_ctl->search_start,
4406 first_logical_byte(fs_info));
4407 ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4408 if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4409 block_group = btrfs_lookup_block_group(fs_info,
4410 ffe_ctl->search_start);
4412 * we don't want to use the block group if it doesn't match our
4413 * allocation bits, or if its not cached.
4415 * However if we are re-searching with an ideal block group
4416 * picked out then we don't care that the block group is cached.
4418 if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4419 block_group->cached != BTRFS_CACHE_NO) {
4420 down_read(&space_info->groups_sem);
4421 if (list_empty(&block_group->list) ||
4424 * someone is removing this block group,
4425 * we can't jump into the have_block_group
4426 * target because our list pointers are not
4429 btrfs_put_block_group(block_group);
4430 up_read(&space_info->groups_sem);
4432 ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4433 block_group->flags);
4434 btrfs_lock_block_group(block_group,
4436 ffe_ctl->hinted = true;
4437 goto have_block_group;
4439 } else if (block_group) {
4440 btrfs_put_block_group(block_group);
4444 trace_find_free_extent_search_loop(root, ffe_ctl);
4445 ffe_ctl->have_caching_bg = false;
4446 if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4447 ffe_ctl->index == 0)
4449 down_read(&space_info->groups_sem);
4450 list_for_each_entry(block_group,
4451 &space_info->block_groups[ffe_ctl->index], list) {
4452 struct btrfs_block_group *bg_ret;
4454 ffe_ctl->hinted = false;
4455 /* If the block group is read-only, we can skip it entirely. */
4456 if (unlikely(block_group->ro)) {
4457 if (ffe_ctl->for_treelog)
4458 btrfs_clear_treelog_bg(block_group);
4459 if (ffe_ctl->for_data_reloc)
4460 btrfs_clear_data_reloc_bg(block_group);
4464 btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4465 ffe_ctl->search_start = block_group->start;
4468 * this can happen if we end up cycling through all the
4469 * raid types, but we want to make sure we only allocate
4470 * for the proper type.
4472 if (!block_group_bits(block_group, ffe_ctl->flags)) {
4473 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4474 BTRFS_BLOCK_GROUP_RAID1_MASK |
4475 BTRFS_BLOCK_GROUP_RAID56_MASK |
4476 BTRFS_BLOCK_GROUP_RAID10;
4479 * if they asked for extra copies and this block group
4480 * doesn't provide them, bail. This does allow us to
4481 * fill raid0 from raid1.
4483 if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4487 * This block group has different flags than we want.
4488 * It's possible that we have MIXED_GROUP flag but no
4489 * block group is mixed. Just skip such block group.
4491 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4496 trace_find_free_extent_have_block_group(root, ffe_ctl, block_group);
4497 ffe_ctl->cached = btrfs_block_group_done(block_group);
4498 if (unlikely(!ffe_ctl->cached)) {
4499 ffe_ctl->have_caching_bg = true;
4500 ret = btrfs_cache_block_group(block_group, false);
4503 * If we get ENOMEM here or something else we want to
4504 * try other block groups, because it may not be fatal.
4505 * However if we can't find anything else we need to
4506 * save our return here so that we return the actual
4507 * error that caused problems, not ENOSPC.
4510 if (!cache_block_group_error)
4511 cache_block_group_error = ret;
4518 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) {
4519 if (!cache_block_group_error)
4520 cache_block_group_error = -EIO;
4524 if (!find_free_extent_check_size_class(ffe_ctl, block_group))
4528 ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4532 if (bg_ret && bg_ret != block_group) {
4533 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4534 block_group = bg_ret;
4538 ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4539 fs_info->stripesize);
4541 /* move on to the next group */
4542 if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4543 block_group->start + block_group->length) {
4544 btrfs_add_free_space_unused(block_group,
4545 ffe_ctl->found_offset,
4546 ffe_ctl->num_bytes);
4550 if (ffe_ctl->found_offset < ffe_ctl->search_start)
4551 btrfs_add_free_space_unused(block_group,
4552 ffe_ctl->found_offset,
4553 ffe_ctl->search_start - ffe_ctl->found_offset);
4555 ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4558 ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS);
4559 if (ret == -EAGAIN) {
4560 btrfs_add_free_space_unused(block_group,
4561 ffe_ctl->found_offset,
4562 ffe_ctl->num_bytes);
4565 btrfs_inc_block_group_reservations(block_group);
4567 /* we are all good, lets return */
4568 ins->objectid = ffe_ctl->search_start;
4569 ins->offset = ffe_ctl->num_bytes;
4571 trace_btrfs_reserve_extent(block_group, ffe_ctl);
4572 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4575 if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
4576 !ffe_ctl->retry_uncached) {
4577 ffe_ctl->retry_uncached = true;
4578 btrfs_wait_block_group_cache_progress(block_group,
4579 ffe_ctl->num_bytes +
4580 ffe_ctl->empty_cluster +
4581 ffe_ctl->empty_size);
4582 goto have_block_group;
4584 release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4587 up_read(&space_info->groups_sem);
4589 ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4593 if (ret == -ENOSPC && !cache_block_group_error) {
4595 * Use ffe_ctl->total_free_space as fallback if we can't find
4596 * any contiguous hole.
4598 if (!ffe_ctl->max_extent_size)
4599 ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4600 spin_lock(&space_info->lock);
4601 space_info->max_extent_size = ffe_ctl->max_extent_size;
4602 spin_unlock(&space_info->lock);
4603 ins->offset = ffe_ctl->max_extent_size;
4604 } else if (ret == -ENOSPC) {
4605 ret = cache_block_group_error;
4611 * Entry point to the extent allocator. Tries to find a hole that is at least
4612 * as big as @num_bytes.
4614 * @root - The root that will contain this extent
4616 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4617 * is used for accounting purposes. This value differs
4618 * from @num_bytes only in the case of compressed extents.
4620 * @num_bytes - Number of bytes to allocate on-disk.
4622 * @min_alloc_size - Indicates the minimum amount of space that the
4623 * allocator should try to satisfy. In some cases
4624 * @num_bytes may be larger than what is required and if
4625 * the filesystem is fragmented then allocation fails.
4626 * However, the presence of @min_alloc_size gives a
4627 * chance to try and satisfy the smaller allocation.
4629 * @empty_size - A hint that you plan on doing more COW. This is the
4630 * size in bytes the allocator should try to find free
4631 * next to the block it returns. This is just a hint and
4632 * may be ignored by the allocator.
4634 * @hint_byte - Hint to the allocator to start searching above the byte
4635 * address passed. It might be ignored.
4637 * @ins - This key is modified to record the found hole. It will
4638 * have the following values:
4639 * ins->objectid == start position
4640 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4641 * ins->offset == the size of the hole.
4643 * @is_data - Boolean flag indicating whether an extent is
4644 * allocated for data (true) or metadata (false)
4646 * @delalloc - Boolean flag indicating whether this allocation is for
4647 * delalloc or not. If 'true' data_rwsem of block groups
4648 * is going to be acquired.
4651 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4652 * case -ENOSPC is returned then @ins->offset will contain the size of the
4653 * largest available hole the allocator managed to find.
4655 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4656 u64 num_bytes, u64 min_alloc_size,
4657 u64 empty_size, u64 hint_byte,
4658 struct btrfs_key *ins, int is_data, int delalloc)
4660 struct btrfs_fs_info *fs_info = root->fs_info;
4661 struct find_free_extent_ctl ffe_ctl = {};
4662 bool final_tried = num_bytes == min_alloc_size;
4665 bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4666 bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4668 flags = get_alloc_profile_by_root(root, is_data);
4670 WARN_ON(num_bytes < fs_info->sectorsize);
4672 ffe_ctl.ram_bytes = ram_bytes;
4673 ffe_ctl.num_bytes = num_bytes;
4674 ffe_ctl.min_alloc_size = min_alloc_size;
4675 ffe_ctl.empty_size = empty_size;
4676 ffe_ctl.flags = flags;
4677 ffe_ctl.delalloc = delalloc;
4678 ffe_ctl.hint_byte = hint_byte;
4679 ffe_ctl.for_treelog = for_treelog;
4680 ffe_ctl.for_data_reloc = for_data_reloc;
4682 ret = find_free_extent(root, ins, &ffe_ctl);
4683 if (!ret && !is_data) {
4684 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4685 } else if (ret == -ENOSPC) {
4686 if (!final_tried && ins->offset) {
4687 num_bytes = min(num_bytes >> 1, ins->offset);
4688 num_bytes = round_down(num_bytes,
4689 fs_info->sectorsize);
4690 num_bytes = max(num_bytes, min_alloc_size);
4691 ram_bytes = num_bytes;
4692 if (num_bytes == min_alloc_size)
4695 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4696 struct btrfs_space_info *sinfo;
4698 sinfo = btrfs_find_space_info(fs_info, flags);
4700 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4701 flags, num_bytes, for_treelog, for_data_reloc);
4703 btrfs_dump_space_info(fs_info, sinfo,
4711 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4712 u64 start, u64 len, int delalloc)
4714 struct btrfs_block_group *cache;
4716 cache = btrfs_lookup_block_group(fs_info, start);
4718 btrfs_err(fs_info, "Unable to find block group for %llu",
4723 btrfs_add_free_space(cache, start, len);
4724 btrfs_free_reserved_bytes(cache, len, delalloc);
4725 trace_btrfs_reserved_extent_free(fs_info, start, len);
4727 btrfs_put_block_group(cache);
4731 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans,
4732 const struct extent_buffer *eb)
4734 struct btrfs_block_group *cache;
4737 cache = btrfs_lookup_block_group(trans->fs_info, eb->start);
4739 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4744 ret = pin_down_extent(trans, cache, eb->start, eb->len, 1);
4745 btrfs_put_block_group(cache);
4749 static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
4752 struct btrfs_fs_info *fs_info = trans->fs_info;
4755 ret = remove_from_free_space_tree(trans, bytenr, num_bytes);
4759 ret = btrfs_update_block_group(trans, bytenr, num_bytes, true);
4762 btrfs_err(fs_info, "update block group failed for %llu %llu",
4767 trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes);
4771 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4772 u64 parent, u64 root_objectid,
4773 u64 flags, u64 owner, u64 offset,
4774 struct btrfs_key *ins, int ref_mod, u64 oref_root)
4776 struct btrfs_fs_info *fs_info = trans->fs_info;
4777 struct btrfs_root *extent_root;
4779 struct btrfs_extent_item *extent_item;
4780 struct btrfs_extent_owner_ref *oref;
4781 struct btrfs_extent_inline_ref *iref;
4782 struct btrfs_path *path;
4783 struct extent_buffer *leaf;
4786 const bool simple_quota = (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE);
4789 type = BTRFS_SHARED_DATA_REF_KEY;
4791 type = BTRFS_EXTENT_DATA_REF_KEY;
4793 size = sizeof(*extent_item);
4795 size += btrfs_extent_inline_ref_size(BTRFS_EXTENT_OWNER_REF_KEY);
4796 size += btrfs_extent_inline_ref_size(type);
4798 path = btrfs_alloc_path();
4802 extent_root = btrfs_extent_root(fs_info, ins->objectid);
4803 ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size);
4805 btrfs_free_path(path);
4809 leaf = path->nodes[0];
4810 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4811 struct btrfs_extent_item);
4812 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4813 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4814 btrfs_set_extent_flags(leaf, extent_item,
4815 flags | BTRFS_EXTENT_FLAG_DATA);
4817 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4819 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_EXTENT_OWNER_REF_KEY);
4820 oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
4821 btrfs_set_extent_owner_ref_root_id(leaf, oref, oref_root);
4822 iref = (struct btrfs_extent_inline_ref *)(oref + 1);
4824 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4827 struct btrfs_shared_data_ref *ref;
4828 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4829 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4830 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4832 struct btrfs_extent_data_ref *ref;
4833 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4834 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4835 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4836 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4837 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4840 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
4841 btrfs_free_path(path);
4843 return alloc_reserved_extent(trans, ins->objectid, ins->offset);
4846 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4847 struct btrfs_delayed_ref_node *node,
4848 struct btrfs_delayed_extent_op *extent_op)
4850 struct btrfs_fs_info *fs_info = trans->fs_info;
4851 struct btrfs_root *extent_root;
4853 struct btrfs_extent_item *extent_item;
4854 struct btrfs_key extent_key;
4855 struct btrfs_tree_block_info *block_info;
4856 struct btrfs_extent_inline_ref *iref;
4857 struct btrfs_path *path;
4858 struct extent_buffer *leaf;
4859 struct btrfs_delayed_tree_ref *ref;
4860 u32 size = sizeof(*extent_item) + sizeof(*iref);
4861 u64 flags = extent_op->flags_to_set;
4862 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4864 ref = btrfs_delayed_node_to_tree_ref(node);
4866 extent_key.objectid = node->bytenr;
4867 if (skinny_metadata) {
4868 extent_key.offset = ref->level;
4869 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4871 extent_key.offset = node->num_bytes;
4872 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4873 size += sizeof(*block_info);
4876 path = btrfs_alloc_path();
4880 extent_root = btrfs_extent_root(fs_info, extent_key.objectid);
4881 ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key,
4884 btrfs_free_path(path);
4888 leaf = path->nodes[0];
4889 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4890 struct btrfs_extent_item);
4891 btrfs_set_extent_refs(leaf, extent_item, 1);
4892 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4893 btrfs_set_extent_flags(leaf, extent_item,
4894 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4896 if (skinny_metadata) {
4897 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4899 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4900 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4901 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4902 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4905 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4906 btrfs_set_extent_inline_ref_type(leaf, iref,
4907 BTRFS_SHARED_BLOCK_REF_KEY);
4908 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4910 btrfs_set_extent_inline_ref_type(leaf, iref,
4911 BTRFS_TREE_BLOCK_REF_KEY);
4912 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4915 btrfs_mark_buffer_dirty(trans, leaf);
4916 btrfs_free_path(path);
4918 return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize);
4921 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4922 struct btrfs_root *root, u64 owner,
4923 u64 offset, u64 ram_bytes,
4924 struct btrfs_key *ins)
4926 struct btrfs_ref generic_ref = { 0 };
4927 u64 root_objectid = root->root_key.objectid;
4928 u64 owning_root = root_objectid;
4930 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4932 if (btrfs_is_data_reloc_root(root) && is_fstree(root->relocation_src_root))
4933 owning_root = root->relocation_src_root;
4935 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4936 ins->objectid, ins->offset, 0, owning_root);
4937 btrfs_init_data_ref(&generic_ref, root_objectid, owner,
4939 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4941 return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4945 * this is used by the tree logging recovery code. It records that
4946 * an extent has been allocated and makes sure to clear the free
4947 * space cache bits as well
4949 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4950 u64 root_objectid, u64 owner, u64 offset,
4951 struct btrfs_key *ins)
4953 struct btrfs_fs_info *fs_info = trans->fs_info;
4955 struct btrfs_block_group *block_group;
4956 struct btrfs_space_info *space_info;
4957 struct btrfs_squota_delta delta = {
4958 .root = root_objectid,
4959 .num_bytes = ins->offset,
4960 .generation = trans->transid,
4966 * Mixed block groups will exclude before processing the log so we only
4967 * need to do the exclude dance if this fs isn't mixed.
4969 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4970 ret = __exclude_logged_extent(fs_info, ins->objectid,
4976 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4980 space_info = block_group->space_info;
4981 spin_lock(&space_info->lock);
4982 spin_lock(&block_group->lock);
4983 space_info->bytes_reserved += ins->offset;
4984 block_group->reserved += ins->offset;
4985 spin_unlock(&block_group->lock);
4986 spin_unlock(&space_info->lock);
4988 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4989 offset, ins, 1, root_objectid);
4991 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4992 ret = btrfs_record_squota_delta(fs_info, &delta);
4993 btrfs_put_block_group(block_group);
4997 #ifdef CONFIG_BTRFS_DEBUG
4999 * Extra safety check in case the extent tree is corrupted and extent allocator
5000 * chooses to use a tree block which is already used and locked.
5002 static bool check_eb_lock_owner(const struct extent_buffer *eb)
5004 if (eb->lock_owner == current->pid) {
5005 btrfs_err_rl(eb->fs_info,
5006 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
5007 eb->start, btrfs_header_owner(eb), current->pid);
5013 static bool check_eb_lock_owner(struct extent_buffer *eb)
5019 static struct extent_buffer *
5020 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5021 u64 bytenr, int level, u64 owner,
5022 enum btrfs_lock_nesting nest)
5024 struct btrfs_fs_info *fs_info = root->fs_info;
5025 struct extent_buffer *buf;
5026 u64 lockdep_owner = owner;
5028 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
5032 if (check_eb_lock_owner(buf)) {
5033 free_extent_buffer(buf);
5034 return ERR_PTR(-EUCLEAN);
5038 * The reloc trees are just snapshots, so we need them to appear to be
5039 * just like any other fs tree WRT lockdep.
5041 * The exception however is in replace_path() in relocation, where we
5042 * hold the lock on the original fs root and then search for the reloc
5043 * root. At that point we need to make sure any reloc root buffers are
5044 * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
5047 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID &&
5048 !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
5049 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
5051 /* btrfs_clear_buffer_dirty() accesses generation field. */
5052 btrfs_set_header_generation(buf, trans->transid);
5055 * This needs to stay, because we could allocate a freed block from an
5056 * old tree into a new tree, so we need to make sure this new block is
5057 * set to the appropriate level and owner.
5059 btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level);
5061 __btrfs_tree_lock(buf, nest);
5062 btrfs_clear_buffer_dirty(trans, buf);
5063 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
5064 clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags);
5066 set_extent_buffer_uptodate(buf);
5068 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
5069 btrfs_set_header_level(buf, level);
5070 btrfs_set_header_bytenr(buf, buf->start);
5071 btrfs_set_header_generation(buf, trans->transid);
5072 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
5073 btrfs_set_header_owner(buf, owner);
5074 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
5075 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
5076 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5077 buf->log_index = root->log_transid % 2;
5079 * we allow two log transactions at a time, use different
5080 * EXTENT bit to differentiate dirty pages.
5082 if (buf->log_index == 0)
5083 set_extent_bit(&root->dirty_log_pages, buf->start,
5084 buf->start + buf->len - 1,
5085 EXTENT_DIRTY, NULL);
5087 set_extent_bit(&root->dirty_log_pages, buf->start,
5088 buf->start + buf->len - 1,
5091 buf->log_index = -1;
5092 set_extent_bit(&trans->transaction->dirty_pages, buf->start,
5093 buf->start + buf->len - 1, EXTENT_DIRTY, NULL);
5095 /* this returns a buffer locked for blocking */
5100 * finds a free extent and does all the dirty work required for allocation
5101 * returns the tree buffer or an ERR_PTR on error.
5103 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
5104 struct btrfs_root *root,
5105 u64 parent, u64 root_objectid,
5106 const struct btrfs_disk_key *key,
5107 int level, u64 hint,
5110 enum btrfs_lock_nesting nest)
5112 struct btrfs_fs_info *fs_info = root->fs_info;
5113 struct btrfs_key ins;
5114 struct btrfs_block_rsv *block_rsv;
5115 struct extent_buffer *buf;
5116 struct btrfs_delayed_extent_op *extent_op;
5117 struct btrfs_ref generic_ref = { 0 };
5120 u32 blocksize = fs_info->nodesize;
5121 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
5124 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5125 if (btrfs_is_testing(fs_info)) {
5126 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
5127 level, root_objectid, nest);
5129 root->alloc_bytenr += blocksize;
5134 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
5135 if (IS_ERR(block_rsv))
5136 return ERR_CAST(block_rsv);
5138 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
5139 empty_size, hint, &ins, 0, 0);
5143 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
5144 root_objectid, nest);
5147 goto out_free_reserved;
5149 owning_root = btrfs_header_owner(buf);
5151 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5153 parent = ins.objectid;
5154 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5155 owning_root = reloc_src_root;
5159 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5160 extent_op = btrfs_alloc_delayed_extent_op();
5166 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5168 memset(&extent_op->key, 0, sizeof(extent_op->key));
5169 extent_op->flags_to_set = flags;
5170 extent_op->update_key = skinny_metadata ? false : true;
5171 extent_op->update_flags = true;
5172 extent_op->level = level;
5174 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
5175 ins.objectid, ins.offset, parent, owning_root);
5176 btrfs_init_tree_ref(&generic_ref, level, root_objectid,
5177 root->root_key.objectid, false);
5178 btrfs_ref_tree_mod(fs_info, &generic_ref);
5179 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
5181 goto out_free_delayed;
5186 btrfs_free_delayed_extent_op(extent_op);
5188 btrfs_tree_unlock(buf);
5189 free_extent_buffer(buf);
5191 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
5193 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
5194 return ERR_PTR(ret);
5197 struct walk_control {
5198 u64 refs[BTRFS_MAX_LEVEL];
5199 u64 flags[BTRFS_MAX_LEVEL];
5200 struct btrfs_key update_progress;
5201 struct btrfs_key drop_progress;
5213 #define DROP_REFERENCE 1
5214 #define UPDATE_BACKREF 2
5216 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5217 struct btrfs_root *root,
5218 struct walk_control *wc,
5219 struct btrfs_path *path)
5221 struct btrfs_fs_info *fs_info = root->fs_info;
5227 struct btrfs_key key;
5228 struct extent_buffer *eb;
5233 if (path->slots[wc->level] < wc->reada_slot) {
5234 wc->reada_count = wc->reada_count * 2 / 3;
5235 wc->reada_count = max(wc->reada_count, 2);
5237 wc->reada_count = wc->reada_count * 3 / 2;
5238 wc->reada_count = min_t(int, wc->reada_count,
5239 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
5242 eb = path->nodes[wc->level];
5243 nritems = btrfs_header_nritems(eb);
5245 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5246 if (nread >= wc->reada_count)
5250 bytenr = btrfs_node_blockptr(eb, slot);
5251 generation = btrfs_node_ptr_generation(eb, slot);
5253 if (slot == path->slots[wc->level])
5256 if (wc->stage == UPDATE_BACKREF &&
5257 generation <= root->root_key.offset)
5260 /* We don't lock the tree block, it's OK to be racy here */
5261 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5262 wc->level - 1, 1, &refs,
5264 /* We don't care about errors in readahead. */
5269 if (wc->stage == DROP_REFERENCE) {
5273 if (wc->level == 1 &&
5274 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5276 if (!wc->update_ref ||
5277 generation <= root->root_key.offset)
5279 btrfs_node_key_to_cpu(eb, &key, slot);
5280 ret = btrfs_comp_cpu_keys(&key,
5281 &wc->update_progress);
5285 if (wc->level == 1 &&
5286 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5290 btrfs_readahead_node_child(eb, slot);
5293 wc->reada_slot = slot;
5297 * helper to process tree block while walking down the tree.
5299 * when wc->stage == UPDATE_BACKREF, this function updates
5300 * back refs for pointers in the block.
5302 * NOTE: return value 1 means we should stop walking down.
5304 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5305 struct btrfs_root *root,
5306 struct btrfs_path *path,
5307 struct walk_control *wc, int lookup_info)
5309 struct btrfs_fs_info *fs_info = root->fs_info;
5310 int level = wc->level;
5311 struct extent_buffer *eb = path->nodes[level];
5312 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5315 if (wc->stage == UPDATE_BACKREF &&
5316 btrfs_header_owner(eb) != root->root_key.objectid)
5320 * when reference count of tree block is 1, it won't increase
5321 * again. once full backref flag is set, we never clear it.
5324 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5325 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5326 BUG_ON(!path->locks[level]);
5327 ret = btrfs_lookup_extent_info(trans, fs_info,
5328 eb->start, level, 1,
5332 BUG_ON(ret == -ENOMEM);
5335 BUG_ON(wc->refs[level] == 0);
5338 if (wc->stage == DROP_REFERENCE) {
5339 if (wc->refs[level] > 1)
5342 if (path->locks[level] && !wc->keep_locks) {
5343 btrfs_tree_unlock_rw(eb, path->locks[level]);
5344 path->locks[level] = 0;
5349 /* wc->stage == UPDATE_BACKREF */
5350 if (!(wc->flags[level] & flag)) {
5351 BUG_ON(!path->locks[level]);
5352 ret = btrfs_inc_ref(trans, root, eb, 1);
5353 BUG_ON(ret); /* -ENOMEM */
5354 ret = btrfs_dec_ref(trans, root, eb, 0);
5355 BUG_ON(ret); /* -ENOMEM */
5356 ret = btrfs_set_disk_extent_flags(trans, eb, flag);
5357 BUG_ON(ret); /* -ENOMEM */
5358 wc->flags[level] |= flag;
5362 * the block is shared by multiple trees, so it's not good to
5363 * keep the tree lock
5365 if (path->locks[level] && level > 0) {
5366 btrfs_tree_unlock_rw(eb, path->locks[level]);
5367 path->locks[level] = 0;
5373 * This is used to verify a ref exists for this root to deal with a bug where we
5374 * would have a drop_progress key that hadn't been updated properly.
5376 static int check_ref_exists(struct btrfs_trans_handle *trans,
5377 struct btrfs_root *root, u64 bytenr, u64 parent,
5380 struct btrfs_path *path;
5381 struct btrfs_extent_inline_ref *iref;
5384 path = btrfs_alloc_path();
5388 ret = lookup_extent_backref(trans, path, &iref, bytenr,
5389 root->fs_info->nodesize, parent,
5390 root->root_key.objectid, level, 0);
5391 btrfs_free_path(path);
5400 * helper to process tree block pointer.
5402 * when wc->stage == DROP_REFERENCE, this function checks
5403 * reference count of the block pointed to. if the block
5404 * is shared and we need update back refs for the subtree
5405 * rooted at the block, this function changes wc->stage to
5406 * UPDATE_BACKREF. if the block is shared and there is no
5407 * need to update back, this function drops the reference
5410 * NOTE: return value 1 means we should stop walking down.
5412 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5413 struct btrfs_root *root,
5414 struct btrfs_path *path,
5415 struct walk_control *wc, int *lookup_info)
5417 struct btrfs_fs_info *fs_info = root->fs_info;
5422 struct btrfs_tree_parent_check check = { 0 };
5423 struct btrfs_key key;
5424 struct btrfs_ref ref = { 0 };
5425 struct extent_buffer *next;
5426 int level = wc->level;
5429 bool need_account = false;
5431 generation = btrfs_node_ptr_generation(path->nodes[level],
5432 path->slots[level]);
5434 * if the lower level block was created before the snapshot
5435 * was created, we know there is no need to update back refs
5438 if (wc->stage == UPDATE_BACKREF &&
5439 generation <= root->root_key.offset) {
5444 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5446 check.level = level - 1;
5447 check.transid = generation;
5448 check.owner_root = root->root_key.objectid;
5449 check.has_first_key = true;
5450 btrfs_node_key_to_cpu(path->nodes[level], &check.first_key,
5451 path->slots[level]);
5453 next = find_extent_buffer(fs_info, bytenr);
5455 next = btrfs_find_create_tree_block(fs_info, bytenr,
5456 root->root_key.objectid, level - 1);
5458 return PTR_ERR(next);
5461 btrfs_tree_lock(next);
5463 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5464 &wc->refs[level - 1],
5465 &wc->flags[level - 1],
5470 if (unlikely(wc->refs[level - 1] == 0)) {
5471 btrfs_err(fs_info, "Missing references.");
5477 if (wc->stage == DROP_REFERENCE) {
5478 if (wc->refs[level - 1] > 1) {
5479 need_account = true;
5481 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5484 if (!wc->update_ref ||
5485 generation <= root->root_key.offset)
5488 btrfs_node_key_to_cpu(path->nodes[level], &key,
5489 path->slots[level]);
5490 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5494 wc->stage = UPDATE_BACKREF;
5495 wc->shared_level = level - 1;
5499 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5503 if (!btrfs_buffer_uptodate(next, generation, 0)) {
5504 btrfs_tree_unlock(next);
5505 free_extent_buffer(next);
5511 if (reada && level == 1)
5512 reada_walk_down(trans, root, wc, path);
5513 next = read_tree_block(fs_info, bytenr, &check);
5515 return PTR_ERR(next);
5516 } else if (!extent_buffer_uptodate(next)) {
5517 free_extent_buffer(next);
5520 btrfs_tree_lock(next);
5524 ASSERT(level == btrfs_header_level(next));
5525 if (level != btrfs_header_level(next)) {
5526 btrfs_err(root->fs_info, "mismatched level");
5530 path->nodes[level] = next;
5531 path->slots[level] = 0;
5532 path->locks[level] = BTRFS_WRITE_LOCK;
5538 wc->refs[level - 1] = 0;
5539 wc->flags[level - 1] = 0;
5540 if (wc->stage == DROP_REFERENCE) {
5541 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5542 parent = path->nodes[level]->start;
5544 ASSERT(root->root_key.objectid ==
5545 btrfs_header_owner(path->nodes[level]));
5546 if (root->root_key.objectid !=
5547 btrfs_header_owner(path->nodes[level])) {
5548 btrfs_err(root->fs_info,
5549 "mismatched block owner");
5557 * If we had a drop_progress we need to verify the refs are set
5558 * as expected. If we find our ref then we know that from here
5559 * on out everything should be correct, and we can clear the
5562 if (wc->restarted) {
5563 ret = check_ref_exists(trans, root, bytenr, parent,
5574 * Reloc tree doesn't contribute to qgroup numbers, and we have
5575 * already accounted them at merge time (replace_path),
5576 * thus we could skip expensive subtree trace here.
5578 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5580 ret = btrfs_qgroup_trace_subtree(trans, next,
5581 generation, level - 1);
5583 btrfs_err_rl(fs_info,
5584 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5590 * We need to update the next key in our walk control so we can
5591 * update the drop_progress key accordingly. We don't care if
5592 * find_next_key doesn't find a key because that means we're at
5593 * the end and are going to clean up now.
5595 wc->drop_level = level;
5596 find_next_key(path, level, &wc->drop_progress);
5598 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5599 fs_info->nodesize, parent, owner_root);
5600 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid,
5602 ret = btrfs_free_extent(trans, &ref);
5611 btrfs_tree_unlock(next);
5612 free_extent_buffer(next);
5618 * helper to process tree block while walking up the tree.
5620 * when wc->stage == DROP_REFERENCE, this function drops
5621 * reference count on the block.
5623 * when wc->stage == UPDATE_BACKREF, this function changes
5624 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5625 * to UPDATE_BACKREF previously while processing the block.
5627 * NOTE: return value 1 means we should stop walking up.
5629 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5630 struct btrfs_root *root,
5631 struct btrfs_path *path,
5632 struct walk_control *wc)
5634 struct btrfs_fs_info *fs_info = root->fs_info;
5636 int level = wc->level;
5637 struct extent_buffer *eb = path->nodes[level];
5640 if (wc->stage == UPDATE_BACKREF) {
5641 BUG_ON(wc->shared_level < level);
5642 if (level < wc->shared_level)
5645 ret = find_next_key(path, level + 1, &wc->update_progress);
5649 wc->stage = DROP_REFERENCE;
5650 wc->shared_level = -1;
5651 path->slots[level] = 0;
5654 * check reference count again if the block isn't locked.
5655 * we should start walking down the tree again if reference
5658 if (!path->locks[level]) {
5660 btrfs_tree_lock(eb);
5661 path->locks[level] = BTRFS_WRITE_LOCK;
5663 ret = btrfs_lookup_extent_info(trans, fs_info,
5664 eb->start, level, 1,
5669 btrfs_tree_unlock_rw(eb, path->locks[level]);
5670 path->locks[level] = 0;
5673 BUG_ON(wc->refs[level] == 0);
5674 if (wc->refs[level] == 1) {
5675 btrfs_tree_unlock_rw(eb, path->locks[level]);
5676 path->locks[level] = 0;
5682 /* wc->stage == DROP_REFERENCE */
5683 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5685 if (wc->refs[level] == 1) {
5687 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5688 ret = btrfs_dec_ref(trans, root, eb, 1);
5690 ret = btrfs_dec_ref(trans, root, eb, 0);
5691 BUG_ON(ret); /* -ENOMEM */
5692 if (is_fstree(root->root_key.objectid)) {
5693 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5695 btrfs_err_rl(fs_info,
5696 "error %d accounting leaf items, quota is out of sync, rescan required",
5701 /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */
5702 if (!path->locks[level]) {
5703 btrfs_tree_lock(eb);
5704 path->locks[level] = BTRFS_WRITE_LOCK;
5706 btrfs_clear_buffer_dirty(trans, eb);
5709 if (eb == root->node) {
5710 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5712 else if (root->root_key.objectid != btrfs_header_owner(eb))
5713 goto owner_mismatch;
5715 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5716 parent = path->nodes[level + 1]->start;
5717 else if (root->root_key.objectid !=
5718 btrfs_header_owner(path->nodes[level + 1]))
5719 goto owner_mismatch;
5722 btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent,
5723 wc->refs[level] == 1);
5725 wc->refs[level] = 0;
5726 wc->flags[level] = 0;
5730 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5731 btrfs_header_owner(eb), root->root_key.objectid);
5735 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5736 struct btrfs_root *root,
5737 struct btrfs_path *path,
5738 struct walk_control *wc)
5740 int level = wc->level;
5741 int lookup_info = 1;
5744 while (level >= 0) {
5745 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5752 if (path->slots[level] >=
5753 btrfs_header_nritems(path->nodes[level]))
5756 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5758 path->slots[level]++;
5764 return (ret == 1) ? 0 : ret;
5767 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5768 struct btrfs_root *root,
5769 struct btrfs_path *path,
5770 struct walk_control *wc, int max_level)
5772 int level = wc->level;
5775 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5776 while (level < max_level && path->nodes[level]) {
5778 if (path->slots[level] + 1 <
5779 btrfs_header_nritems(path->nodes[level])) {
5780 path->slots[level]++;
5783 ret = walk_up_proc(trans, root, path, wc);
5789 if (path->locks[level]) {
5790 btrfs_tree_unlock_rw(path->nodes[level],
5791 path->locks[level]);
5792 path->locks[level] = 0;
5794 free_extent_buffer(path->nodes[level]);
5795 path->nodes[level] = NULL;
5803 * drop a subvolume tree.
5805 * this function traverses the tree freeing any blocks that only
5806 * referenced by the tree.
5808 * when a shared tree block is found. this function decreases its
5809 * reference count by one. if update_ref is true, this function
5810 * also make sure backrefs for the shared block and all lower level
5811 * blocks are properly updated.
5813 * If called with for_reloc == 0, may exit early with -EAGAIN
5815 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5817 const bool is_reloc_root = (root->root_key.objectid ==
5818 BTRFS_TREE_RELOC_OBJECTID);
5819 struct btrfs_fs_info *fs_info = root->fs_info;
5820 struct btrfs_path *path;
5821 struct btrfs_trans_handle *trans;
5822 struct btrfs_root *tree_root = fs_info->tree_root;
5823 struct btrfs_root_item *root_item = &root->root_item;
5824 struct walk_control *wc;
5825 struct btrfs_key key;
5829 bool root_dropped = false;
5830 bool unfinished_drop = false;
5832 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5834 path = btrfs_alloc_path();
5840 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5842 btrfs_free_path(path);
5848 * Use join to avoid potential EINTR from transaction start. See
5849 * wait_reserve_ticket and the whole reservation callchain.
5852 trans = btrfs_join_transaction(tree_root);
5854 trans = btrfs_start_transaction(tree_root, 0);
5855 if (IS_ERR(trans)) {
5856 err = PTR_ERR(trans);
5860 err = btrfs_run_delayed_items(trans);
5865 * This will help us catch people modifying the fs tree while we're
5866 * dropping it. It is unsafe to mess with the fs tree while it's being
5867 * dropped as we unlock the root node and parent nodes as we walk down
5868 * the tree, assuming nothing will change. If something does change
5869 * then we'll have stale information and drop references to blocks we've
5872 set_bit(BTRFS_ROOT_DELETING, &root->state);
5873 unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
5875 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5876 level = btrfs_header_level(root->node);
5877 path->nodes[level] = btrfs_lock_root_node(root);
5878 path->slots[level] = 0;
5879 path->locks[level] = BTRFS_WRITE_LOCK;
5880 memset(&wc->update_progress, 0,
5881 sizeof(wc->update_progress));
5883 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5884 memcpy(&wc->update_progress, &key,
5885 sizeof(wc->update_progress));
5887 level = btrfs_root_drop_level(root_item);
5889 path->lowest_level = level;
5890 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5891 path->lowest_level = 0;
5899 * unlock our path, this is safe because only this
5900 * function is allowed to delete this snapshot
5902 btrfs_unlock_up_safe(path, 0);
5904 level = btrfs_header_level(root->node);
5906 btrfs_tree_lock(path->nodes[level]);
5907 path->locks[level] = BTRFS_WRITE_LOCK;
5909 ret = btrfs_lookup_extent_info(trans, fs_info,
5910 path->nodes[level]->start,
5911 level, 1, &wc->refs[level],
5912 &wc->flags[level], NULL);
5917 BUG_ON(wc->refs[level] == 0);
5919 if (level == btrfs_root_drop_level(root_item))
5922 btrfs_tree_unlock(path->nodes[level]);
5923 path->locks[level] = 0;
5924 WARN_ON(wc->refs[level] != 1);
5929 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5931 wc->shared_level = -1;
5932 wc->stage = DROP_REFERENCE;
5933 wc->update_ref = update_ref;
5935 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5939 ret = walk_down_tree(trans, root, path, wc);
5941 btrfs_abort_transaction(trans, ret);
5946 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5948 btrfs_abort_transaction(trans, ret);
5954 BUG_ON(wc->stage != DROP_REFERENCE);
5958 if (wc->stage == DROP_REFERENCE) {
5959 wc->drop_level = wc->level;
5960 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5962 path->slots[wc->drop_level]);
5964 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5965 &wc->drop_progress);
5966 btrfs_set_root_drop_level(root_item, wc->drop_level);
5968 BUG_ON(wc->level == 0);
5969 if (btrfs_should_end_transaction(trans) ||
5970 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5971 ret = btrfs_update_root(trans, tree_root,
5975 btrfs_abort_transaction(trans, ret);
5981 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
5983 btrfs_end_transaction_throttle(trans);
5984 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5985 btrfs_debug(fs_info,
5986 "drop snapshot early exit");
5992 * Use join to avoid potential EINTR from transaction
5993 * start. See wait_reserve_ticket and the whole
5994 * reservation callchain.
5997 trans = btrfs_join_transaction(tree_root);
5999 trans = btrfs_start_transaction(tree_root, 0);
6000 if (IS_ERR(trans)) {
6001 err = PTR_ERR(trans);
6006 btrfs_release_path(path);
6010 ret = btrfs_del_root(trans, &root->root_key);
6012 btrfs_abort_transaction(trans, ret);
6017 if (!is_reloc_root) {
6018 ret = btrfs_find_root(tree_root, &root->root_key, path,
6021 btrfs_abort_transaction(trans, ret);
6024 } else if (ret > 0) {
6025 /* if we fail to delete the orphan item this time
6026 * around, it'll get picked up the next time.
6028 * The most common failure here is just -ENOENT.
6030 btrfs_del_orphan_item(trans, tree_root,
6031 root->root_key.objectid);
6036 * This subvolume is going to be completely dropped, and won't be
6037 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
6038 * commit transaction time. So free it here manually.
6040 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
6041 btrfs_qgroup_free_meta_all_pertrans(root);
6043 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
6044 btrfs_add_dropped_root(trans, root);
6046 btrfs_put_root(root);
6047 root_dropped = true;
6050 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
6052 btrfs_end_transaction_throttle(trans);
6055 btrfs_free_path(path);
6058 * We were an unfinished drop root, check to see if there are any
6059 * pending, and if not clear and wake up any waiters.
6061 if (!err && unfinished_drop)
6062 btrfs_maybe_wake_unfinished_drop(fs_info);
6065 * So if we need to stop dropping the snapshot for whatever reason we
6066 * need to make sure to add it back to the dead root list so that we
6067 * keep trying to do the work later. This also cleans up roots if we
6068 * don't have it in the radix (like when we recover after a power fail
6069 * or unmount) so we don't leak memory.
6071 if (!for_reloc && !root_dropped)
6072 btrfs_add_dead_root(root);
6077 * drop subtree rooted at tree block 'node'.
6079 * NOTE: this function will unlock and release tree block 'node'
6080 * only used by relocation code
6082 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6083 struct btrfs_root *root,
6084 struct extent_buffer *node,
6085 struct extent_buffer *parent)
6087 struct btrfs_fs_info *fs_info = root->fs_info;
6088 struct btrfs_path *path;
6089 struct walk_control *wc;
6095 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6097 path = btrfs_alloc_path();
6101 wc = kzalloc(sizeof(*wc), GFP_NOFS);
6103 btrfs_free_path(path);
6107 btrfs_assert_tree_write_locked(parent);
6108 parent_level = btrfs_header_level(parent);
6109 atomic_inc(&parent->refs);
6110 path->nodes[parent_level] = parent;
6111 path->slots[parent_level] = btrfs_header_nritems(parent);
6113 btrfs_assert_tree_write_locked(node);
6114 level = btrfs_header_level(node);
6115 path->nodes[level] = node;
6116 path->slots[level] = 0;
6117 path->locks[level] = BTRFS_WRITE_LOCK;
6119 wc->refs[parent_level] = 1;
6120 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6122 wc->shared_level = -1;
6123 wc->stage = DROP_REFERENCE;
6126 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
6129 wret = walk_down_tree(trans, root, path, wc);
6135 wret = walk_up_tree(trans, root, path, wc, parent_level);
6143 btrfs_free_path(path);
6147 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
6150 return unpin_extent_range(fs_info, start, end, false);
6154 * It used to be that old block groups would be left around forever.
6155 * Iterating over them would be enough to trim unused space. Since we
6156 * now automatically remove them, we also need to iterate over unallocated
6159 * We don't want a transaction for this since the discard may take a
6160 * substantial amount of time. We don't require that a transaction be
6161 * running, but we do need to take a running transaction into account
6162 * to ensure that we're not discarding chunks that were released or
6163 * allocated in the current transaction.
6165 * Holding the chunks lock will prevent other threads from allocating
6166 * or releasing chunks, but it won't prevent a running transaction
6167 * from committing and releasing the memory that the pending chunks
6168 * list head uses. For that, we need to take a reference to the
6169 * transaction and hold the commit root sem. We only need to hold
6170 * it while performing the free space search since we have already
6171 * held back allocations.
6173 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
6175 u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0;
6180 /* Discard not supported = nothing to do. */
6181 if (!bdev_max_discard_sectors(device->bdev))
6184 /* Not writable = nothing to do. */
6185 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
6188 /* No free space = nothing to do. */
6189 if (device->total_bytes <= device->bytes_used)
6195 struct btrfs_fs_info *fs_info = device->fs_info;
6198 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
6202 find_first_clear_extent_bit(&device->alloc_state, start,
6204 CHUNK_TRIMMED | CHUNK_ALLOCATED);
6206 /* Check if there are any CHUNK_* bits left */
6207 if (start > device->total_bytes) {
6208 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6209 btrfs_warn_in_rcu(fs_info,
6210 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
6211 start, end - start + 1,
6212 btrfs_dev_name(device),
6213 device->total_bytes);
6214 mutex_unlock(&fs_info->chunk_mutex);
6219 /* Ensure we skip the reserved space on each device. */
6220 start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
6223 * If find_first_clear_extent_bit find a range that spans the
6224 * end of the device it will set end to -1, in this case it's up
6225 * to the caller to trim the value to the size of the device.
6227 end = min(end, device->total_bytes - 1);
6229 len = end - start + 1;
6231 /* We didn't find any extents */
6233 mutex_unlock(&fs_info->chunk_mutex);
6238 ret = btrfs_issue_discard(device->bdev, start, len,
6241 set_extent_bit(&device->alloc_state, start,
6242 start + bytes - 1, CHUNK_TRIMMED, NULL);
6243 mutex_unlock(&fs_info->chunk_mutex);
6251 if (fatal_signal_pending(current)) {
6263 * Trim the whole filesystem by:
6264 * 1) trimming the free space in each block group
6265 * 2) trimming the unallocated space on each device
6267 * This will also continue trimming even if a block group or device encounters
6268 * an error. The return value will be the last error, or 0 if nothing bad
6271 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6273 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6274 struct btrfs_block_group *cache = NULL;
6275 struct btrfs_device *device;
6277 u64 range_end = U64_MAX;
6287 if (range->start == U64_MAX)
6291 * Check range overflow if range->len is set.
6292 * The default range->len is U64_MAX.
6294 if (range->len != U64_MAX &&
6295 check_add_overflow(range->start, range->len, &range_end))
6298 cache = btrfs_lookup_first_block_group(fs_info, range->start);
6299 for (; cache; cache = btrfs_next_block_group(cache)) {
6300 if (cache->start >= range_end) {
6301 btrfs_put_block_group(cache);
6305 start = max(range->start, cache->start);
6306 end = min(range_end, cache->start + cache->length);
6308 if (end - start >= range->minlen) {
6309 if (!btrfs_block_group_done(cache)) {
6310 ret = btrfs_cache_block_group(cache, true);
6317 ret = btrfs_trim_block_group(cache,
6323 trimmed += group_trimmed;
6334 "failed to trim %llu block group(s), last error %d",
6337 mutex_lock(&fs_devices->device_list_mutex);
6338 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6339 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6342 ret = btrfs_trim_free_extents(device, &group_trimmed);
6349 trimmed += group_trimmed;
6351 mutex_unlock(&fs_devices->device_list_mutex);
6355 "failed to trim %llu device(s), last error %d",
6356 dev_failed, dev_ret);
6357 range->len = trimmed;