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>
22 #include "print-tree.h"
26 #include "free-space-cache.h"
27 #include "free-space-tree.h"
30 #include "ref-verify.h"
31 #include "space-info.h"
32 #include "block-rsv.h"
33 #include "delalloc-space.h"
34 #include "block-group.h"
36 #include "rcu-string.h"
38 #include "dev-replace.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
44 struct btrfs_delayed_ref_node *node, u64 parent,
45 u64 root_objectid, u64 owner_objectid,
46 u64 owner_offset, int refs_to_drop,
47 struct btrfs_delayed_extent_op *extra_op);
48 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
49 struct extent_buffer *leaf,
50 struct btrfs_extent_item *ei);
51 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
52 u64 parent, u64 root_objectid,
53 u64 flags, u64 owner, u64 offset,
54 struct btrfs_key *ins, int ref_mod);
55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
56 struct btrfs_delayed_ref_node *node,
57 struct btrfs_delayed_extent_op *extent_op);
58 static int find_next_key(struct btrfs_path *path, int level,
59 struct btrfs_key *key);
61 static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
63 return (cache->flags & bits) == bits;
66 int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info,
67 u64 start, u64 num_bytes)
69 u64 end = start + num_bytes - 1;
70 set_extent_bits(&fs_info->excluded_extents, start, end,
75 void btrfs_free_excluded_extents(struct btrfs_block_group *cache)
77 struct btrfs_fs_info *fs_info = cache->fs_info;
81 end = start + cache->length - 1;
83 clear_extent_bits(&fs_info->excluded_extents, start, end,
87 /* simple helper to search for an existing data extent at a given offset */
88 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
90 struct btrfs_root *root = btrfs_extent_root(fs_info, start);
93 struct btrfs_path *path;
95 path = btrfs_alloc_path();
101 key.type = BTRFS_EXTENT_ITEM_KEY;
102 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
103 btrfs_free_path(path);
108 * helper function to lookup reference count and flags of a tree block.
110 * the head node for delayed ref is used to store the sum of all the
111 * reference count modifications queued up in the rbtree. the head
112 * node may also store the extent flags to set. This way you can check
113 * to see what the reference count and extent flags would be if all of
114 * the delayed refs are not processed.
116 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
117 struct btrfs_fs_info *fs_info, u64 bytenr,
118 u64 offset, int metadata, u64 *refs, u64 *flags)
120 struct btrfs_root *extent_root;
121 struct btrfs_delayed_ref_head *head;
122 struct btrfs_delayed_ref_root *delayed_refs;
123 struct btrfs_path *path;
124 struct btrfs_extent_item *ei;
125 struct extent_buffer *leaf;
126 struct btrfs_key key;
133 * If we don't have skinny metadata, don't bother doing anything
136 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
137 offset = fs_info->nodesize;
141 path = btrfs_alloc_path();
146 path->skip_locking = 1;
147 path->search_commit_root = 1;
151 key.objectid = bytenr;
154 key.type = BTRFS_METADATA_ITEM_KEY;
156 key.type = BTRFS_EXTENT_ITEM_KEY;
158 extent_root = btrfs_extent_root(fs_info, bytenr);
159 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
163 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
164 if (path->slots[0]) {
166 btrfs_item_key_to_cpu(path->nodes[0], &key,
168 if (key.objectid == bytenr &&
169 key.type == BTRFS_EXTENT_ITEM_KEY &&
170 key.offset == fs_info->nodesize)
176 leaf = path->nodes[0];
177 item_size = btrfs_item_size(leaf, path->slots[0]);
178 if (item_size >= sizeof(*ei)) {
179 ei = btrfs_item_ptr(leaf, path->slots[0],
180 struct btrfs_extent_item);
181 num_refs = btrfs_extent_refs(leaf, ei);
182 extent_flags = btrfs_extent_flags(leaf, ei);
185 btrfs_print_v0_err(fs_info);
187 btrfs_abort_transaction(trans, ret);
189 btrfs_handle_fs_error(fs_info, ret, NULL);
194 BUG_ON(num_refs == 0);
204 delayed_refs = &trans->transaction->delayed_refs;
205 spin_lock(&delayed_refs->lock);
206 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
208 if (!mutex_trylock(&head->mutex)) {
209 refcount_inc(&head->refs);
210 spin_unlock(&delayed_refs->lock);
212 btrfs_release_path(path);
215 * Mutex was contended, block until it's released and try
218 mutex_lock(&head->mutex);
219 mutex_unlock(&head->mutex);
220 btrfs_put_delayed_ref_head(head);
223 spin_lock(&head->lock);
224 if (head->extent_op && head->extent_op->update_flags)
225 extent_flags |= head->extent_op->flags_to_set;
227 BUG_ON(num_refs == 0);
229 num_refs += head->ref_mod;
230 spin_unlock(&head->lock);
231 mutex_unlock(&head->mutex);
233 spin_unlock(&delayed_refs->lock);
235 WARN_ON(num_refs == 0);
239 *flags = extent_flags;
241 btrfs_free_path(path);
246 * Back reference rules. Back refs have three main goals:
248 * 1) differentiate between all holders of references to an extent so that
249 * when a reference is dropped we can make sure it was a valid reference
250 * before freeing the extent.
252 * 2) Provide enough information to quickly find the holders of an extent
253 * if we notice a given block is corrupted or bad.
255 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
256 * maintenance. This is actually the same as #2, but with a slightly
257 * different use case.
259 * There are two kinds of back refs. The implicit back refs is optimized
260 * for pointers in non-shared tree blocks. For a given pointer in a block,
261 * back refs of this kind provide information about the block's owner tree
262 * and the pointer's key. These information allow us to find the block by
263 * b-tree searching. The full back refs is for pointers in tree blocks not
264 * referenced by their owner trees. The location of tree block is recorded
265 * in the back refs. Actually the full back refs is generic, and can be
266 * used in all cases the implicit back refs is used. The major shortcoming
267 * of the full back refs is its overhead. Every time a tree block gets
268 * COWed, we have to update back refs entry for all pointers in it.
270 * For a newly allocated tree block, we use implicit back refs for
271 * pointers in it. This means most tree related operations only involve
272 * implicit back refs. For a tree block created in old transaction, the
273 * only way to drop a reference to it is COW it. So we can detect the
274 * event that tree block loses its owner tree's reference and do the
275 * back refs conversion.
277 * When a tree block is COWed through a tree, there are four cases:
279 * The reference count of the block is one and the tree is the block's
280 * owner tree. Nothing to do in this case.
282 * The reference count of the block is one and the tree is not the
283 * block's owner tree. In this case, full back refs is used for pointers
284 * in the block. Remove these full back refs, add implicit back refs for
285 * every pointers in the new block.
287 * The reference count of the block is greater than one and the tree is
288 * the block's owner tree. In this case, implicit back refs is used for
289 * pointers in the block. Add full back refs for every pointers in the
290 * block, increase lower level extents' reference counts. The original
291 * implicit back refs are entailed to the new block.
293 * The reference count of the block is greater than one and the tree is
294 * not the block's owner tree. Add implicit back refs for every pointer in
295 * the new block, increase lower level extents' reference count.
297 * Back Reference Key composing:
299 * The key objectid corresponds to the first byte in the extent,
300 * The key type is used to differentiate between types of back refs.
301 * There are different meanings of the key offset for different types
304 * File extents can be referenced by:
306 * - multiple snapshots, subvolumes, or different generations in one subvol
307 * - different files inside a single subvolume
308 * - different offsets inside a file (bookend extents in file.c)
310 * The extent ref structure for the implicit back refs has fields for:
312 * - Objectid of the subvolume root
313 * - objectid of the file holding the reference
314 * - original offset in the file
315 * - how many bookend extents
317 * The key offset for the implicit back refs is hash of the first
320 * The extent ref structure for the full back refs has field for:
322 * - number of pointers in the tree leaf
324 * The key offset for the implicit back refs is the first byte of
327 * When a file extent is allocated, The implicit back refs is used.
328 * the fields are filled in:
330 * (root_key.objectid, inode objectid, offset in file, 1)
332 * When a file extent is removed file truncation, we find the
333 * corresponding implicit back refs and check the following fields:
335 * (btrfs_header_owner(leaf), inode objectid, offset in file)
337 * Btree extents can be referenced by:
339 * - Different subvolumes
341 * Both the implicit back refs and the full back refs for tree blocks
342 * only consist of key. The key offset for the implicit back refs is
343 * objectid of block's owner tree. The key offset for the full back refs
344 * is the first byte of parent block.
346 * When implicit back refs is used, information about the lowest key and
347 * level of the tree block are required. These information are stored in
348 * tree block info structure.
352 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
353 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
354 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
356 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
357 struct btrfs_extent_inline_ref *iref,
358 enum btrfs_inline_ref_type is_data)
360 int type = btrfs_extent_inline_ref_type(eb, iref);
361 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
363 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
364 type == BTRFS_SHARED_BLOCK_REF_KEY ||
365 type == BTRFS_SHARED_DATA_REF_KEY ||
366 type == BTRFS_EXTENT_DATA_REF_KEY) {
367 if (is_data == BTRFS_REF_TYPE_BLOCK) {
368 if (type == BTRFS_TREE_BLOCK_REF_KEY)
370 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
373 * Every shared one has parent tree block,
374 * which must be aligned to sector size.
377 IS_ALIGNED(offset, eb->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, eb->fs_info->sectorsize))
394 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
399 btrfs_print_leaf((struct extent_buffer *)eb);
400 btrfs_err(eb->fs_info,
401 "eb %llu iref 0x%lx invalid extent inline ref type %d",
402 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 = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
416 lenum = cpu_to_le64(owner);
417 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
418 lenum = cpu_to_le64(offset);
419 low_crc = btrfs_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(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);
621 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
622 btrfs_print_v0_err(trans->fs_info);
623 btrfs_abort_transaction(trans, -EINVAL);
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(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]);
657 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
660 * If type is invalid, we should have bailed out earlier than
663 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
664 ASSERT(type != BTRFS_REF_TYPE_INVALID);
665 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
666 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
667 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
669 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
670 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
672 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
673 ref1 = btrfs_item_ptr(leaf, path->slots[0],
674 struct btrfs_extent_data_ref);
675 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
676 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
677 ref2 = btrfs_item_ptr(leaf, path->slots[0],
678 struct btrfs_shared_data_ref);
679 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
686 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
687 struct btrfs_path *path,
688 u64 bytenr, u64 parent,
691 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
692 struct btrfs_key key;
695 key.objectid = bytenr;
697 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
700 key.type = BTRFS_TREE_BLOCK_REF_KEY;
701 key.offset = root_objectid;
704 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
710 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
711 struct btrfs_path *path,
712 u64 bytenr, u64 parent,
715 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
716 struct btrfs_key key;
719 key.objectid = bytenr;
721 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
724 key.type = BTRFS_TREE_BLOCK_REF_KEY;
725 key.offset = root_objectid;
728 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
729 btrfs_release_path(path);
733 static inline int extent_ref_type(u64 parent, u64 owner)
736 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
738 type = BTRFS_SHARED_BLOCK_REF_KEY;
740 type = BTRFS_TREE_BLOCK_REF_KEY;
743 type = BTRFS_SHARED_DATA_REF_KEY;
745 type = BTRFS_EXTENT_DATA_REF_KEY;
750 static int find_next_key(struct btrfs_path *path, int level,
751 struct btrfs_key *key)
754 for (; level < BTRFS_MAX_LEVEL; level++) {
755 if (!path->nodes[level])
757 if (path->slots[level] + 1 >=
758 btrfs_header_nritems(path->nodes[level]))
761 btrfs_item_key_to_cpu(path->nodes[level], key,
762 path->slots[level] + 1);
764 btrfs_node_key_to_cpu(path->nodes[level], key,
765 path->slots[level] + 1);
772 * look for inline back ref. if back ref is found, *ref_ret is set
773 * to the address of inline back ref, and 0 is returned.
775 * if back ref isn't found, *ref_ret is set to the address where it
776 * should be inserted, and -ENOENT is returned.
778 * if insert is true and there are too many inline back refs, the path
779 * points to the extent item, and -EAGAIN is returned.
781 * NOTE: inline back refs are ordered in the same way that back ref
782 * items in the tree are ordered.
784 static noinline_for_stack
785 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
786 struct btrfs_path *path,
787 struct btrfs_extent_inline_ref **ref_ret,
788 u64 bytenr, u64 num_bytes,
789 u64 parent, u64 root_objectid,
790 u64 owner, u64 offset, int insert)
792 struct btrfs_fs_info *fs_info = trans->fs_info;
793 struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
794 struct btrfs_key key;
795 struct extent_buffer *leaf;
796 struct btrfs_extent_item *ei;
797 struct btrfs_extent_inline_ref *iref;
807 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
810 key.objectid = bytenr;
811 key.type = BTRFS_EXTENT_ITEM_KEY;
812 key.offset = num_bytes;
814 want = extent_ref_type(parent, owner);
816 extra_size = btrfs_extent_inline_ref_size(want);
817 path->search_for_extension = 1;
818 path->keep_locks = 1;
823 * Owner is our level, so we can just add one to get the level for the
824 * block we are interested in.
826 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
827 key.type = BTRFS_METADATA_ITEM_KEY;
832 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
839 * We may be a newly converted file system which still has the old fat
840 * extent entries for metadata, so try and see if we have one of those.
842 if (ret > 0 && skinny_metadata) {
843 skinny_metadata = false;
844 if (path->slots[0]) {
846 btrfs_item_key_to_cpu(path->nodes[0], &key,
848 if (key.objectid == bytenr &&
849 key.type == BTRFS_EXTENT_ITEM_KEY &&
850 key.offset == num_bytes)
854 key.objectid = bytenr;
855 key.type = BTRFS_EXTENT_ITEM_KEY;
856 key.offset = num_bytes;
857 btrfs_release_path(path);
862 if (ret && !insert) {
865 } else if (WARN_ON(ret)) {
866 btrfs_print_leaf(path->nodes[0]);
868 "extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu",
869 bytenr, num_bytes, parent, root_objectid, owner,
875 leaf = path->nodes[0];
876 item_size = btrfs_item_size(leaf, path->slots[0]);
877 if (unlikely(item_size < sizeof(*ei))) {
879 btrfs_print_v0_err(fs_info);
880 btrfs_abort_transaction(trans, err);
884 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
885 flags = btrfs_extent_flags(leaf, ei);
887 ptr = (unsigned long)(ei + 1);
888 end = (unsigned long)ei + item_size;
890 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
891 ptr += sizeof(struct btrfs_tree_block_info);
895 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
896 needed = BTRFS_REF_TYPE_DATA;
898 needed = BTRFS_REF_TYPE_BLOCK;
905 btrfs_print_leaf(path->nodes[0]);
907 "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu",
908 path->slots[0], root_objectid, owner, offset, parent);
912 iref = (struct btrfs_extent_inline_ref *)ptr;
913 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
914 if (type == BTRFS_REF_TYPE_INVALID) {
922 ptr += btrfs_extent_inline_ref_size(type);
926 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
927 struct btrfs_extent_data_ref *dref;
928 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
929 if (match_extent_data_ref(leaf, dref, root_objectid,
934 if (hash_extent_data_ref_item(leaf, dref) <
935 hash_extent_data_ref(root_objectid, owner, offset))
939 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
941 if (parent == ref_offset) {
945 if (ref_offset < parent)
948 if (root_objectid == ref_offset) {
952 if (ref_offset < root_objectid)
956 ptr += btrfs_extent_inline_ref_size(type);
958 if (err == -ENOENT && insert) {
959 if (item_size + extra_size >=
960 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
965 * To add new inline back ref, we have to make sure
966 * there is no corresponding back ref item.
967 * For simplicity, we just do not add new inline back
968 * ref if there is any kind of item for this block
970 if (find_next_key(path, 0, &key) == 0 &&
971 key.objectid == bytenr &&
972 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
977 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
980 path->keep_locks = 0;
981 path->search_for_extension = 0;
982 btrfs_unlock_up_safe(path, 1);
988 * helper to add new inline back ref
990 static noinline_for_stack
991 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
992 struct btrfs_path *path,
993 struct btrfs_extent_inline_ref *iref,
994 u64 parent, u64 root_objectid,
995 u64 owner, u64 offset, int refs_to_add,
996 struct btrfs_delayed_extent_op *extent_op)
998 struct extent_buffer *leaf;
999 struct btrfs_extent_item *ei;
1002 unsigned long item_offset;
1007 leaf = path->nodes[0];
1008 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1009 item_offset = (unsigned long)iref - (unsigned long)ei;
1011 type = extent_ref_type(parent, owner);
1012 size = btrfs_extent_inline_ref_size(type);
1014 btrfs_extend_item(path, size);
1016 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1017 refs = btrfs_extent_refs(leaf, ei);
1018 refs += refs_to_add;
1019 btrfs_set_extent_refs(leaf, ei, refs);
1021 __run_delayed_extent_op(extent_op, leaf, ei);
1023 ptr = (unsigned long)ei + item_offset;
1024 end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]);
1025 if (ptr < end - size)
1026 memmove_extent_buffer(leaf, ptr + size, ptr,
1029 iref = (struct btrfs_extent_inline_ref *)ptr;
1030 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1031 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1032 struct btrfs_extent_data_ref *dref;
1033 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1034 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1035 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1036 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1037 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1038 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1039 struct btrfs_shared_data_ref *sref;
1040 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1041 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1042 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1043 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1044 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1046 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1048 btrfs_mark_buffer_dirty(leaf);
1051 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1052 struct btrfs_path *path,
1053 struct btrfs_extent_inline_ref **ref_ret,
1054 u64 bytenr, u64 num_bytes, u64 parent,
1055 u64 root_objectid, u64 owner, u64 offset)
1059 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1060 num_bytes, parent, root_objectid,
1065 btrfs_release_path(path);
1068 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1069 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1072 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1073 root_objectid, owner, offset);
1079 * helper to update/remove inline back ref
1081 static noinline_for_stack
1082 void update_inline_extent_backref(struct btrfs_path *path,
1083 struct btrfs_extent_inline_ref *iref,
1085 struct btrfs_delayed_extent_op *extent_op)
1087 struct extent_buffer *leaf = path->nodes[0];
1088 struct btrfs_extent_item *ei;
1089 struct btrfs_extent_data_ref *dref = NULL;
1090 struct btrfs_shared_data_ref *sref = NULL;
1098 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1099 refs = btrfs_extent_refs(leaf, ei);
1100 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1101 refs += refs_to_mod;
1102 btrfs_set_extent_refs(leaf, ei, refs);
1104 __run_delayed_extent_op(extent_op, leaf, ei);
1107 * If type is invalid, we should have bailed out after
1108 * lookup_inline_extent_backref().
1110 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1111 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1113 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1114 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1115 refs = btrfs_extent_data_ref_count(leaf, dref);
1116 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1117 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1118 refs = btrfs_shared_data_ref_count(leaf, sref);
1121 BUG_ON(refs_to_mod != -1);
1124 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1125 refs += refs_to_mod;
1128 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1129 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1131 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1133 size = btrfs_extent_inline_ref_size(type);
1134 item_size = btrfs_item_size(leaf, path->slots[0]);
1135 ptr = (unsigned long)iref;
1136 end = (unsigned long)ei + item_size;
1137 if (ptr + size < end)
1138 memmove_extent_buffer(leaf, ptr, ptr + size,
1141 btrfs_truncate_item(path, item_size, 1);
1143 btrfs_mark_buffer_dirty(leaf);
1146 static noinline_for_stack
1147 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1148 struct btrfs_path *path,
1149 u64 bytenr, u64 num_bytes, u64 parent,
1150 u64 root_objectid, u64 owner,
1151 u64 offset, int refs_to_add,
1152 struct btrfs_delayed_extent_op *extent_op)
1154 struct btrfs_extent_inline_ref *iref;
1157 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1158 num_bytes, parent, root_objectid,
1162 * We're adding refs to a tree block we already own, this
1163 * should not happen at all.
1165 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1166 btrfs_crit(trans->fs_info,
1167 "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu",
1168 bytenr, num_bytes, root_objectid);
1169 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) {
1171 btrfs_crit(trans->fs_info,
1172 "path->slots[0]=%d path->nodes[0]:", path->slots[0]);
1173 btrfs_print_leaf(path->nodes[0]);
1177 update_inline_extent_backref(path, iref, refs_to_add, extent_op);
1178 } else if (ret == -ENOENT) {
1179 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1180 root_objectid, owner, offset,
1181 refs_to_add, extent_op);
1187 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1188 struct btrfs_root *root,
1189 struct btrfs_path *path,
1190 struct btrfs_extent_inline_ref *iref,
1191 int refs_to_drop, int is_data)
1195 BUG_ON(!is_data && refs_to_drop != 1);
1197 update_inline_extent_backref(path, iref, -refs_to_drop, NULL);
1199 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1201 ret = btrfs_del_item(trans, root, path);
1205 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1206 u64 *discarded_bytes)
1209 u64 bytes_left, end;
1210 u64 aligned_start = ALIGN(start, 1 << 9);
1212 /* Adjust the range to be aligned to 512B sectors if necessary. */
1213 if (start != aligned_start) {
1214 len -= aligned_start - start;
1215 len = round_down(len, 1 << 9);
1216 start = aligned_start;
1219 *discarded_bytes = 0;
1227 /* Skip any superblocks on this device. */
1228 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1229 u64 sb_start = btrfs_sb_offset(j);
1230 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1231 u64 size = sb_start - start;
1233 if (!in_range(sb_start, start, bytes_left) &&
1234 !in_range(sb_end, start, bytes_left) &&
1235 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1239 * Superblock spans beginning of range. Adjust start and
1242 if (sb_start <= start) {
1243 start += sb_end - start;
1248 bytes_left = end - start;
1253 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1256 *discarded_bytes += size;
1257 else if (ret != -EOPNOTSUPP)
1266 bytes_left = end - start;
1270 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1273 *discarded_bytes += bytes_left;
1278 static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes)
1280 struct btrfs_device *dev = stripe->dev;
1281 struct btrfs_fs_info *fs_info = dev->fs_info;
1282 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1283 u64 phys = stripe->physical;
1284 u64 len = stripe->length;
1288 /* Zone reset on a zoned filesystem */
1289 if (btrfs_can_zone_reset(dev, phys, len)) {
1292 ret = btrfs_reset_device_zone(dev, phys, len, &discarded);
1296 if (!btrfs_dev_replace_is_ongoing(dev_replace) ||
1297 dev != dev_replace->srcdev)
1300 src_disc = discarded;
1302 /* Send to replace target as well */
1303 ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len,
1305 discarded += src_disc;
1306 } else if (bdev_max_discard_sectors(stripe->dev->bdev)) {
1307 ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded);
1318 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1319 u64 num_bytes, u64 *actual_bytes)
1322 u64 discarded_bytes = 0;
1323 u64 end = bytenr + num_bytes;
1327 * Avoid races with device replace and make sure the devices in the
1328 * stripes don't go away while we are discarding.
1330 btrfs_bio_counter_inc_blocked(fs_info);
1332 struct btrfs_discard_stripe *stripes;
1333 unsigned int num_stripes;
1336 num_bytes = end - cur;
1337 stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes);
1338 if (IS_ERR(stripes)) {
1339 ret = PTR_ERR(stripes);
1340 if (ret == -EOPNOTSUPP)
1345 for (i = 0; i < num_stripes; i++) {
1346 struct btrfs_discard_stripe *stripe = stripes + i;
1349 if (!stripe->dev->bdev) {
1350 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1354 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
1355 &stripe->dev->dev_state))
1358 ret = do_discard_extent(stripe, &bytes);
1361 * Keep going if discard is not supported by the
1364 if (ret != -EOPNOTSUPP)
1368 discarded_bytes += bytes;
1376 btrfs_bio_counter_dec(fs_info);
1378 *actual_bytes = discarded_bytes;
1382 /* Can return -ENOMEM */
1383 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1384 struct btrfs_ref *generic_ref)
1386 struct btrfs_fs_info *fs_info = trans->fs_info;
1389 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1390 generic_ref->action);
1391 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1392 generic_ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID);
1394 if (generic_ref->type == BTRFS_REF_METADATA)
1395 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1397 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1399 btrfs_ref_tree_mod(fs_info, generic_ref);
1405 * __btrfs_inc_extent_ref - insert backreference for a given extent
1407 * The counterpart is in __btrfs_free_extent(), with examples and more details
1410 * @trans: Handle of transaction
1412 * @node: The delayed ref node used to get the bytenr/length for
1413 * extent whose references are incremented.
1415 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1416 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1417 * bytenr of the parent block. Since new extents are always
1418 * created with indirect references, this will only be the case
1419 * when relocating a shared extent. In that case, root_objectid
1420 * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1423 * @root_objectid: The id of the root where this modification has originated,
1424 * this can be either one of the well-known metadata trees or
1425 * the subvolume id which references this extent.
1427 * @owner: For data extents it is the inode number of the owning file.
1428 * For metadata extents this parameter holds the level in the
1429 * tree of the extent.
1431 * @offset: For metadata extents the offset is ignored and is currently
1432 * always passed as 0. For data extents it is the fileoffset
1433 * this extent belongs to.
1435 * @refs_to_add Number of references to add
1437 * @extent_op Pointer to a structure, holding information necessary when
1438 * updating a tree block's flags
1441 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1442 struct btrfs_delayed_ref_node *node,
1443 u64 parent, u64 root_objectid,
1444 u64 owner, u64 offset, int refs_to_add,
1445 struct btrfs_delayed_extent_op *extent_op)
1447 struct btrfs_path *path;
1448 struct extent_buffer *leaf;
1449 struct btrfs_extent_item *item;
1450 struct btrfs_key key;
1451 u64 bytenr = node->bytenr;
1452 u64 num_bytes = node->num_bytes;
1456 path = btrfs_alloc_path();
1460 /* this will setup the path even if it fails to insert the back ref */
1461 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1462 parent, root_objectid, owner,
1463 offset, refs_to_add, extent_op);
1464 if ((ret < 0 && ret != -EAGAIN) || !ret)
1468 * Ok we had -EAGAIN which means we didn't have space to insert and
1469 * inline extent ref, so just update the reference count and add a
1472 leaf = path->nodes[0];
1473 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1474 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1475 refs = btrfs_extent_refs(leaf, item);
1476 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1478 __run_delayed_extent_op(extent_op, leaf, item);
1480 btrfs_mark_buffer_dirty(leaf);
1481 btrfs_release_path(path);
1483 /* now insert the actual backref */
1484 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1485 BUG_ON(refs_to_add != 1);
1486 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1489 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1490 root_objectid, owner, offset,
1494 btrfs_abort_transaction(trans, ret);
1496 btrfs_free_path(path);
1500 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1501 struct btrfs_delayed_ref_node *node,
1502 struct btrfs_delayed_extent_op *extent_op,
1503 int insert_reserved)
1506 struct btrfs_delayed_data_ref *ref;
1507 struct btrfs_key ins;
1512 ins.objectid = node->bytenr;
1513 ins.offset = node->num_bytes;
1514 ins.type = BTRFS_EXTENT_ITEM_KEY;
1516 ref = btrfs_delayed_node_to_data_ref(node);
1517 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1519 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1520 parent = ref->parent;
1521 ref_root = ref->root;
1523 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1525 flags |= extent_op->flags_to_set;
1526 ret = alloc_reserved_file_extent(trans, parent, ref_root,
1527 flags, ref->objectid,
1530 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1531 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1532 ref->objectid, ref->offset,
1533 node->ref_mod, extent_op);
1534 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1535 ret = __btrfs_free_extent(trans, node, parent,
1536 ref_root, ref->objectid,
1537 ref->offset, node->ref_mod,
1545 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1546 struct extent_buffer *leaf,
1547 struct btrfs_extent_item *ei)
1549 u64 flags = btrfs_extent_flags(leaf, ei);
1550 if (extent_op->update_flags) {
1551 flags |= extent_op->flags_to_set;
1552 btrfs_set_extent_flags(leaf, ei, flags);
1555 if (extent_op->update_key) {
1556 struct btrfs_tree_block_info *bi;
1557 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1558 bi = (struct btrfs_tree_block_info *)(ei + 1);
1559 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1563 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1564 struct btrfs_delayed_ref_head *head,
1565 struct btrfs_delayed_extent_op *extent_op)
1567 struct btrfs_fs_info *fs_info = trans->fs_info;
1568 struct btrfs_root *root;
1569 struct btrfs_key key;
1570 struct btrfs_path *path;
1571 struct btrfs_extent_item *ei;
1572 struct extent_buffer *leaf;
1578 if (TRANS_ABORTED(trans))
1581 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1584 path = btrfs_alloc_path();
1588 key.objectid = head->bytenr;
1591 key.type = BTRFS_METADATA_ITEM_KEY;
1592 key.offset = extent_op->level;
1594 key.type = BTRFS_EXTENT_ITEM_KEY;
1595 key.offset = head->num_bytes;
1598 root = btrfs_extent_root(fs_info, key.objectid);
1600 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1607 if (path->slots[0] > 0) {
1609 btrfs_item_key_to_cpu(path->nodes[0], &key,
1611 if (key.objectid == head->bytenr &&
1612 key.type == BTRFS_EXTENT_ITEM_KEY &&
1613 key.offset == head->num_bytes)
1617 btrfs_release_path(path);
1620 key.objectid = head->bytenr;
1621 key.offset = head->num_bytes;
1622 key.type = BTRFS_EXTENT_ITEM_KEY;
1631 leaf = path->nodes[0];
1632 item_size = btrfs_item_size(leaf, path->slots[0]);
1634 if (unlikely(item_size < sizeof(*ei))) {
1636 btrfs_print_v0_err(fs_info);
1637 btrfs_abort_transaction(trans, err);
1641 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1642 __run_delayed_extent_op(extent_op, leaf, ei);
1644 btrfs_mark_buffer_dirty(leaf);
1646 btrfs_free_path(path);
1650 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1651 struct btrfs_delayed_ref_node *node,
1652 struct btrfs_delayed_extent_op *extent_op,
1653 int insert_reserved)
1656 struct btrfs_delayed_tree_ref *ref;
1660 ref = btrfs_delayed_node_to_tree_ref(node);
1661 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1663 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1664 parent = ref->parent;
1665 ref_root = ref->root;
1667 if (unlikely(node->ref_mod != 1)) {
1668 btrfs_err(trans->fs_info,
1669 "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu",
1670 node->bytenr, node->ref_mod, node->action, ref_root,
1674 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1675 BUG_ON(!extent_op || !extent_op->update_flags);
1676 ret = alloc_reserved_tree_block(trans, node, extent_op);
1677 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1678 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1679 ref->level, 0, 1, extent_op);
1680 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1681 ret = __btrfs_free_extent(trans, node, parent, ref_root,
1682 ref->level, 0, 1, extent_op);
1689 /* helper function to actually process a single delayed ref entry */
1690 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1691 struct btrfs_delayed_ref_node *node,
1692 struct btrfs_delayed_extent_op *extent_op,
1693 int insert_reserved)
1697 if (TRANS_ABORTED(trans)) {
1698 if (insert_reserved)
1699 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1703 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1704 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1705 ret = run_delayed_tree_ref(trans, node, extent_op,
1707 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1708 node->type == BTRFS_SHARED_DATA_REF_KEY)
1709 ret = run_delayed_data_ref(trans, node, extent_op,
1713 if (ret && insert_reserved)
1714 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1716 btrfs_err(trans->fs_info,
1717 "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d",
1718 node->bytenr, node->num_bytes, node->type,
1719 node->action, node->ref_mod, ret);
1723 static inline struct btrfs_delayed_ref_node *
1724 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1726 struct btrfs_delayed_ref_node *ref;
1728 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1732 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1733 * This is to prevent a ref count from going down to zero, which deletes
1734 * the extent item from the extent tree, when there still are references
1735 * to add, which would fail because they would not find the extent item.
1737 if (!list_empty(&head->ref_add_list))
1738 return list_first_entry(&head->ref_add_list,
1739 struct btrfs_delayed_ref_node, add_list);
1741 ref = rb_entry(rb_first_cached(&head->ref_tree),
1742 struct btrfs_delayed_ref_node, ref_node);
1743 ASSERT(list_empty(&ref->add_list));
1747 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1748 struct btrfs_delayed_ref_head *head)
1750 spin_lock(&delayed_refs->lock);
1751 head->processing = 0;
1752 delayed_refs->num_heads_ready++;
1753 spin_unlock(&delayed_refs->lock);
1754 btrfs_delayed_ref_unlock(head);
1757 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1758 struct btrfs_delayed_ref_head *head)
1760 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1765 if (head->must_insert_reserved) {
1766 head->extent_op = NULL;
1767 btrfs_free_delayed_extent_op(extent_op);
1773 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1774 struct btrfs_delayed_ref_head *head)
1776 struct btrfs_delayed_extent_op *extent_op;
1779 extent_op = cleanup_extent_op(head);
1782 head->extent_op = NULL;
1783 spin_unlock(&head->lock);
1784 ret = run_delayed_extent_op(trans, head, extent_op);
1785 btrfs_free_delayed_extent_op(extent_op);
1786 return ret ? ret : 1;
1789 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1790 struct btrfs_delayed_ref_root *delayed_refs,
1791 struct btrfs_delayed_ref_head *head)
1793 int nr_items = 1; /* Dropping this ref head update. */
1796 * We had csum deletions accounted for in our delayed refs rsv, we need
1797 * to drop the csum leaves for this update from our delayed_refs_rsv.
1799 if (head->total_ref_mod < 0 && head->is_data) {
1800 spin_lock(&delayed_refs->lock);
1801 delayed_refs->pending_csums -= head->num_bytes;
1802 spin_unlock(&delayed_refs->lock);
1803 nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1806 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1809 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1810 struct btrfs_delayed_ref_head *head)
1813 struct btrfs_fs_info *fs_info = trans->fs_info;
1814 struct btrfs_delayed_ref_root *delayed_refs;
1817 delayed_refs = &trans->transaction->delayed_refs;
1819 ret = run_and_cleanup_extent_op(trans, head);
1821 unselect_delayed_ref_head(delayed_refs, head);
1822 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1829 * Need to drop our head ref lock and re-acquire the delayed ref lock
1830 * and then re-check to make sure nobody got added.
1832 spin_unlock(&head->lock);
1833 spin_lock(&delayed_refs->lock);
1834 spin_lock(&head->lock);
1835 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1836 spin_unlock(&head->lock);
1837 spin_unlock(&delayed_refs->lock);
1840 btrfs_delete_ref_head(delayed_refs, head);
1841 spin_unlock(&head->lock);
1842 spin_unlock(&delayed_refs->lock);
1844 if (head->must_insert_reserved) {
1845 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1846 if (head->is_data) {
1847 struct btrfs_root *csum_root;
1849 csum_root = btrfs_csum_root(fs_info, head->bytenr);
1850 ret = btrfs_del_csums(trans, csum_root, head->bytenr,
1855 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1857 trace_run_delayed_ref_head(fs_info, head, 0);
1858 btrfs_delayed_ref_unlock(head);
1859 btrfs_put_delayed_ref_head(head);
1863 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1864 struct btrfs_trans_handle *trans)
1866 struct btrfs_delayed_ref_root *delayed_refs =
1867 &trans->transaction->delayed_refs;
1868 struct btrfs_delayed_ref_head *head = NULL;
1871 spin_lock(&delayed_refs->lock);
1872 head = btrfs_select_ref_head(delayed_refs);
1874 spin_unlock(&delayed_refs->lock);
1879 * Grab the lock that says we are going to process all the refs for
1882 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1883 spin_unlock(&delayed_refs->lock);
1886 * We may have dropped the spin lock to get the head mutex lock, and
1887 * that might have given someone else time to free the head. If that's
1888 * true, it has been removed from our list and we can move on.
1891 head = ERR_PTR(-EAGAIN);
1896 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1897 struct btrfs_delayed_ref_head *locked_ref,
1898 unsigned long *run_refs)
1900 struct btrfs_fs_info *fs_info = trans->fs_info;
1901 struct btrfs_delayed_ref_root *delayed_refs;
1902 struct btrfs_delayed_extent_op *extent_op;
1903 struct btrfs_delayed_ref_node *ref;
1904 int must_insert_reserved = 0;
1907 delayed_refs = &trans->transaction->delayed_refs;
1909 lockdep_assert_held(&locked_ref->mutex);
1910 lockdep_assert_held(&locked_ref->lock);
1912 while ((ref = select_delayed_ref(locked_ref))) {
1914 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1915 spin_unlock(&locked_ref->lock);
1916 unselect_delayed_ref_head(delayed_refs, locked_ref);
1922 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1923 RB_CLEAR_NODE(&ref->ref_node);
1924 if (!list_empty(&ref->add_list))
1925 list_del(&ref->add_list);
1927 * When we play the delayed ref, also correct the ref_mod on
1930 switch (ref->action) {
1931 case BTRFS_ADD_DELAYED_REF:
1932 case BTRFS_ADD_DELAYED_EXTENT:
1933 locked_ref->ref_mod -= ref->ref_mod;
1935 case BTRFS_DROP_DELAYED_REF:
1936 locked_ref->ref_mod += ref->ref_mod;
1941 atomic_dec(&delayed_refs->num_entries);
1944 * Record the must_insert_reserved flag before we drop the
1947 must_insert_reserved = locked_ref->must_insert_reserved;
1948 locked_ref->must_insert_reserved = 0;
1950 extent_op = locked_ref->extent_op;
1951 locked_ref->extent_op = NULL;
1952 spin_unlock(&locked_ref->lock);
1954 ret = run_one_delayed_ref(trans, ref, extent_op,
1955 must_insert_reserved);
1957 btrfs_free_delayed_extent_op(extent_op);
1959 unselect_delayed_ref_head(delayed_refs, locked_ref);
1960 btrfs_put_delayed_ref(ref);
1964 btrfs_put_delayed_ref(ref);
1967 spin_lock(&locked_ref->lock);
1968 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
1975 * Returns 0 on success or if called with an already aborted transaction.
1976 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
1978 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1981 struct btrfs_fs_info *fs_info = trans->fs_info;
1982 struct btrfs_delayed_ref_root *delayed_refs;
1983 struct btrfs_delayed_ref_head *locked_ref = NULL;
1984 ktime_t start = ktime_get();
1986 unsigned long count = 0;
1987 unsigned long actual_count = 0;
1989 delayed_refs = &trans->transaction->delayed_refs;
1992 locked_ref = btrfs_obtain_ref_head(trans);
1993 if (IS_ERR_OR_NULL(locked_ref)) {
1994 if (PTR_ERR(locked_ref) == -EAGAIN) {
2003 * We need to try and merge add/drops of the same ref since we
2004 * can run into issues with relocate dropping the implicit ref
2005 * and then it being added back again before the drop can
2006 * finish. If we merged anything we need to re-loop so we can
2008 * Or we can get node references of the same type that weren't
2009 * merged when created due to bumps in the tree mod seq, and
2010 * we need to merge them to prevent adding an inline extent
2011 * backref before dropping it (triggering a BUG_ON at
2012 * insert_inline_extent_backref()).
2014 spin_lock(&locked_ref->lock);
2015 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2017 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2019 if (ret < 0 && ret != -EAGAIN) {
2021 * Error, btrfs_run_delayed_refs_for_head already
2022 * unlocked everything so just bail out
2027 * Success, perform the usual cleanup of a processed
2030 ret = cleanup_ref_head(trans, locked_ref);
2032 /* We dropped our lock, we need to loop. */
2041 * Either success case or btrfs_run_delayed_refs_for_head
2042 * returned -EAGAIN, meaning we need to select another head
2047 } while ((nr != -1 && count < nr) || locked_ref);
2050 * We don't want to include ref heads since we can have empty ref heads
2051 * and those will drastically skew our runtime down since we just do
2052 * accounting, no actual extent tree updates.
2054 if (actual_count > 0) {
2055 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2059 * We weigh the current average higher than our current runtime
2060 * to avoid large swings in the average.
2062 spin_lock(&delayed_refs->lock);
2063 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2064 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2065 spin_unlock(&delayed_refs->lock);
2070 #ifdef SCRAMBLE_DELAYED_REFS
2072 * Normally delayed refs get processed in ascending bytenr order. This
2073 * correlates in most cases to the order added. To expose dependencies on this
2074 * order, we start to process the tree in the middle instead of the beginning
2076 static u64 find_middle(struct rb_root *root)
2078 struct rb_node *n = root->rb_node;
2079 struct btrfs_delayed_ref_node *entry;
2082 u64 first = 0, last = 0;
2086 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2087 first = entry->bytenr;
2091 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2092 last = entry->bytenr;
2097 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2098 WARN_ON(!entry->in_tree);
2100 middle = entry->bytenr;
2114 * this starts processing the delayed reference count updates and
2115 * extent insertions we have queued up so far. count can be
2116 * 0, which means to process everything in the tree at the start
2117 * of the run (but not newly added entries), or it can be some target
2118 * number you'd like to process.
2120 * Returns 0 on success or if called with an aborted transaction
2121 * Returns <0 on error and aborts the transaction
2123 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2124 unsigned long count)
2126 struct btrfs_fs_info *fs_info = trans->fs_info;
2127 struct rb_node *node;
2128 struct btrfs_delayed_ref_root *delayed_refs;
2129 struct btrfs_delayed_ref_head *head;
2131 int run_all = count == (unsigned long)-1;
2133 /* We'll clean this up in btrfs_cleanup_transaction */
2134 if (TRANS_ABORTED(trans))
2137 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2140 delayed_refs = &trans->transaction->delayed_refs;
2142 count = delayed_refs->num_heads_ready;
2145 #ifdef SCRAMBLE_DELAYED_REFS
2146 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2148 ret = __btrfs_run_delayed_refs(trans, count);
2150 btrfs_abort_transaction(trans, ret);
2155 btrfs_create_pending_block_groups(trans);
2157 spin_lock(&delayed_refs->lock);
2158 node = rb_first_cached(&delayed_refs->href_root);
2160 spin_unlock(&delayed_refs->lock);
2163 head = rb_entry(node, struct btrfs_delayed_ref_head,
2165 refcount_inc(&head->refs);
2166 spin_unlock(&delayed_refs->lock);
2168 /* Mutex was contended, block until it's released and retry. */
2169 mutex_lock(&head->mutex);
2170 mutex_unlock(&head->mutex);
2172 btrfs_put_delayed_ref_head(head);
2180 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2181 struct extent_buffer *eb, u64 flags,
2184 struct btrfs_delayed_extent_op *extent_op;
2187 extent_op = btrfs_alloc_delayed_extent_op();
2191 extent_op->flags_to_set = flags;
2192 extent_op->update_flags = true;
2193 extent_op->update_key = false;
2194 extent_op->level = level;
2196 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2198 btrfs_free_delayed_extent_op(extent_op);
2202 static noinline int check_delayed_ref(struct btrfs_root *root,
2203 struct btrfs_path *path,
2204 u64 objectid, u64 offset, u64 bytenr)
2206 struct btrfs_delayed_ref_head *head;
2207 struct btrfs_delayed_ref_node *ref;
2208 struct btrfs_delayed_data_ref *data_ref;
2209 struct btrfs_delayed_ref_root *delayed_refs;
2210 struct btrfs_transaction *cur_trans;
2211 struct rb_node *node;
2214 spin_lock(&root->fs_info->trans_lock);
2215 cur_trans = root->fs_info->running_transaction;
2217 refcount_inc(&cur_trans->use_count);
2218 spin_unlock(&root->fs_info->trans_lock);
2222 delayed_refs = &cur_trans->delayed_refs;
2223 spin_lock(&delayed_refs->lock);
2224 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2226 spin_unlock(&delayed_refs->lock);
2227 btrfs_put_transaction(cur_trans);
2231 if (!mutex_trylock(&head->mutex)) {
2233 spin_unlock(&delayed_refs->lock);
2234 btrfs_put_transaction(cur_trans);
2238 refcount_inc(&head->refs);
2239 spin_unlock(&delayed_refs->lock);
2241 btrfs_release_path(path);
2244 * Mutex was contended, block until it's released and let
2247 mutex_lock(&head->mutex);
2248 mutex_unlock(&head->mutex);
2249 btrfs_put_delayed_ref_head(head);
2250 btrfs_put_transaction(cur_trans);
2253 spin_unlock(&delayed_refs->lock);
2255 spin_lock(&head->lock);
2257 * XXX: We should replace this with a proper search function in the
2260 for (node = rb_first_cached(&head->ref_tree); node;
2261 node = rb_next(node)) {
2262 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2263 /* If it's a shared ref we know a cross reference exists */
2264 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2269 data_ref = btrfs_delayed_node_to_data_ref(ref);
2272 * If our ref doesn't match the one we're currently looking at
2273 * then we have a cross reference.
2275 if (data_ref->root != root->root_key.objectid ||
2276 data_ref->objectid != objectid ||
2277 data_ref->offset != offset) {
2282 spin_unlock(&head->lock);
2283 mutex_unlock(&head->mutex);
2284 btrfs_put_transaction(cur_trans);
2288 static noinline int check_committed_ref(struct btrfs_root *root,
2289 struct btrfs_path *path,
2290 u64 objectid, u64 offset, u64 bytenr,
2293 struct btrfs_fs_info *fs_info = root->fs_info;
2294 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2295 struct extent_buffer *leaf;
2296 struct btrfs_extent_data_ref *ref;
2297 struct btrfs_extent_inline_ref *iref;
2298 struct btrfs_extent_item *ei;
2299 struct btrfs_key key;
2304 key.objectid = bytenr;
2305 key.offset = (u64)-1;
2306 key.type = BTRFS_EXTENT_ITEM_KEY;
2308 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2311 BUG_ON(ret == 0); /* Corruption */
2314 if (path->slots[0] == 0)
2318 leaf = path->nodes[0];
2319 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2321 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2325 item_size = btrfs_item_size(leaf, path->slots[0]);
2326 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2328 /* If extent item has more than 1 inline ref then it's shared */
2329 if (item_size != sizeof(*ei) +
2330 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2334 * If extent created before last snapshot => it's shared unless the
2335 * snapshot has been deleted. Use the heuristic if strict is false.
2338 (btrfs_extent_generation(leaf, ei) <=
2339 btrfs_root_last_snapshot(&root->root_item)))
2342 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2344 /* If this extent has SHARED_DATA_REF then it's shared */
2345 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2346 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2349 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2350 if (btrfs_extent_refs(leaf, ei) !=
2351 btrfs_extent_data_ref_count(leaf, ref) ||
2352 btrfs_extent_data_ref_root(leaf, ref) !=
2353 root->root_key.objectid ||
2354 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2355 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2363 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2364 u64 bytenr, bool strict, struct btrfs_path *path)
2369 ret = check_committed_ref(root, path, objectid,
2370 offset, bytenr, strict);
2371 if (ret && ret != -ENOENT)
2374 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2375 } while (ret == -EAGAIN);
2378 btrfs_release_path(path);
2379 if (btrfs_is_data_reloc_root(root))
2384 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2385 struct btrfs_root *root,
2386 struct extent_buffer *buf,
2387 int full_backref, int inc)
2389 struct btrfs_fs_info *fs_info = root->fs_info;
2395 struct btrfs_key key;
2396 struct btrfs_file_extent_item *fi;
2397 struct btrfs_ref generic_ref = { 0 };
2398 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2404 if (btrfs_is_testing(fs_info))
2407 ref_root = btrfs_header_owner(buf);
2408 nritems = btrfs_header_nritems(buf);
2409 level = btrfs_header_level(buf);
2411 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2415 parent = buf->start;
2419 action = BTRFS_ADD_DELAYED_REF;
2421 action = BTRFS_DROP_DELAYED_REF;
2423 for (i = 0; i < nritems; i++) {
2425 btrfs_item_key_to_cpu(buf, &key, i);
2426 if (key.type != BTRFS_EXTENT_DATA_KEY)
2428 fi = btrfs_item_ptr(buf, i,
2429 struct btrfs_file_extent_item);
2430 if (btrfs_file_extent_type(buf, fi) ==
2431 BTRFS_FILE_EXTENT_INLINE)
2433 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2437 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2438 key.offset -= btrfs_file_extent_offset(buf, fi);
2439 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2441 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2442 key.offset, root->root_key.objectid,
2445 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2447 ret = btrfs_free_extent(trans, &generic_ref);
2451 bytenr = btrfs_node_blockptr(buf, i);
2452 num_bytes = fs_info->nodesize;
2453 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2455 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root,
2456 root->root_key.objectid, for_reloc);
2458 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2460 ret = btrfs_free_extent(trans, &generic_ref);
2470 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2471 struct extent_buffer *buf, int full_backref)
2473 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2476 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2477 struct extent_buffer *buf, int full_backref)
2479 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2482 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2484 struct btrfs_fs_info *fs_info = root->fs_info;
2489 flags = BTRFS_BLOCK_GROUP_DATA;
2490 else if (root == fs_info->chunk_root)
2491 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2493 flags = BTRFS_BLOCK_GROUP_METADATA;
2495 ret = btrfs_get_alloc_profile(fs_info, flags);
2499 static u64 first_logical_byte(struct btrfs_fs_info *fs_info)
2501 struct rb_node *leftmost;
2504 read_lock(&fs_info->block_group_cache_lock);
2505 /* Get the block group with the lowest logical start address. */
2506 leftmost = rb_first_cached(&fs_info->block_group_cache_tree);
2508 struct btrfs_block_group *bg;
2510 bg = rb_entry(leftmost, struct btrfs_block_group, cache_node);
2513 read_unlock(&fs_info->block_group_cache_lock);
2518 static int pin_down_extent(struct btrfs_trans_handle *trans,
2519 struct btrfs_block_group *cache,
2520 u64 bytenr, u64 num_bytes, int reserved)
2522 struct btrfs_fs_info *fs_info = cache->fs_info;
2524 spin_lock(&cache->space_info->lock);
2525 spin_lock(&cache->lock);
2526 cache->pinned += num_bytes;
2527 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2530 cache->reserved -= num_bytes;
2531 cache->space_info->bytes_reserved -= num_bytes;
2533 spin_unlock(&cache->lock);
2534 spin_unlock(&cache->space_info->lock);
2536 set_extent_dirty(&trans->transaction->pinned_extents, bytenr,
2537 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
2541 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2542 u64 bytenr, u64 num_bytes, int reserved)
2544 struct btrfs_block_group *cache;
2546 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2547 BUG_ON(!cache); /* Logic error */
2549 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2551 btrfs_put_block_group(cache);
2556 * this function must be called within transaction
2558 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2559 u64 bytenr, u64 num_bytes)
2561 struct btrfs_block_group *cache;
2564 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2569 * Fully cache the free space first so that our pin removes the free space
2572 ret = btrfs_cache_block_group(cache, true);
2576 pin_down_extent(trans, cache, bytenr, num_bytes, 0);
2578 /* remove us from the free space cache (if we're there at all) */
2579 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2581 btrfs_put_block_group(cache);
2585 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2586 u64 start, u64 num_bytes)
2589 struct btrfs_block_group *block_group;
2591 block_group = btrfs_lookup_block_group(fs_info, start);
2595 ret = btrfs_cache_block_group(block_group, true);
2599 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2601 btrfs_put_block_group(block_group);
2605 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2607 struct btrfs_fs_info *fs_info = eb->fs_info;
2608 struct btrfs_file_extent_item *item;
2609 struct btrfs_key key;
2614 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2617 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2618 btrfs_item_key_to_cpu(eb, &key, i);
2619 if (key.type != BTRFS_EXTENT_DATA_KEY)
2621 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2622 found_type = btrfs_file_extent_type(eb, item);
2623 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2625 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2627 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2628 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2629 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2638 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2640 atomic_inc(&bg->reservations);
2644 * Returns the free cluster for the given space info and sets empty_cluster to
2645 * what it should be based on the mount options.
2647 static struct btrfs_free_cluster *
2648 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2649 struct btrfs_space_info *space_info, u64 *empty_cluster)
2651 struct btrfs_free_cluster *ret = NULL;
2654 if (btrfs_mixed_space_info(space_info))
2657 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2658 ret = &fs_info->meta_alloc_cluster;
2659 if (btrfs_test_opt(fs_info, SSD))
2660 *empty_cluster = SZ_2M;
2662 *empty_cluster = SZ_64K;
2663 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2664 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2665 *empty_cluster = SZ_2M;
2666 ret = &fs_info->data_alloc_cluster;
2672 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2674 const bool return_free_space)
2676 struct btrfs_block_group *cache = NULL;
2677 struct btrfs_space_info *space_info;
2678 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2679 struct btrfs_free_cluster *cluster = NULL;
2681 u64 total_unpinned = 0;
2682 u64 empty_cluster = 0;
2685 while (start <= end) {
2688 start >= cache->start + cache->length) {
2690 btrfs_put_block_group(cache);
2692 cache = btrfs_lookup_block_group(fs_info, start);
2693 BUG_ON(!cache); /* Logic error */
2695 cluster = fetch_cluster_info(fs_info,
2698 empty_cluster <<= 1;
2701 len = cache->start + cache->length - start;
2702 len = min(len, end + 1 - start);
2704 if (return_free_space)
2705 btrfs_add_free_space(cache, start, len);
2708 total_unpinned += len;
2709 space_info = cache->space_info;
2712 * If this space cluster has been marked as fragmented and we've
2713 * unpinned enough in this block group to potentially allow a
2714 * cluster to be created inside of it go ahead and clear the
2717 if (cluster && cluster->fragmented &&
2718 total_unpinned > empty_cluster) {
2719 spin_lock(&cluster->lock);
2720 cluster->fragmented = 0;
2721 spin_unlock(&cluster->lock);
2724 spin_lock(&space_info->lock);
2725 spin_lock(&cache->lock);
2726 cache->pinned -= len;
2727 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2728 space_info->max_extent_size = 0;
2730 space_info->bytes_readonly += len;
2732 } else if (btrfs_is_zoned(fs_info)) {
2733 /* Need reset before reusing in a zoned block group */
2734 space_info->bytes_zone_unusable += len;
2737 spin_unlock(&cache->lock);
2738 if (!readonly && return_free_space &&
2739 global_rsv->space_info == space_info) {
2740 spin_lock(&global_rsv->lock);
2741 if (!global_rsv->full) {
2742 u64 to_add = min(len, global_rsv->size -
2743 global_rsv->reserved);
2745 global_rsv->reserved += to_add;
2746 btrfs_space_info_update_bytes_may_use(fs_info,
2747 space_info, to_add);
2748 if (global_rsv->reserved >= global_rsv->size)
2749 global_rsv->full = 1;
2752 spin_unlock(&global_rsv->lock);
2754 /* Add to any tickets we may have */
2755 if (!readonly && return_free_space && len)
2756 btrfs_try_granting_tickets(fs_info, space_info);
2757 spin_unlock(&space_info->lock);
2761 btrfs_put_block_group(cache);
2765 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2767 struct btrfs_fs_info *fs_info = trans->fs_info;
2768 struct btrfs_block_group *block_group, *tmp;
2769 struct list_head *deleted_bgs;
2770 struct extent_io_tree *unpin;
2775 unpin = &trans->transaction->pinned_extents;
2777 while (!TRANS_ABORTED(trans)) {
2778 struct extent_state *cached_state = NULL;
2780 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2781 ret = find_first_extent_bit(unpin, 0, &start, &end,
2782 EXTENT_DIRTY, &cached_state);
2784 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2788 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2789 ret = btrfs_discard_extent(fs_info, start,
2790 end + 1 - start, NULL);
2792 clear_extent_dirty(unpin, start, end, &cached_state);
2793 unpin_extent_range(fs_info, start, end, true);
2794 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2795 free_extent_state(cached_state);
2799 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2800 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2801 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2805 * Transaction is finished. We don't need the lock anymore. We
2806 * do need to clean up the block groups in case of a transaction
2809 deleted_bgs = &trans->transaction->deleted_bgs;
2810 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2814 if (!TRANS_ABORTED(trans))
2815 ret = btrfs_discard_extent(fs_info,
2817 block_group->length,
2820 list_del_init(&block_group->bg_list);
2821 btrfs_unfreeze_block_group(block_group);
2822 btrfs_put_block_group(block_group);
2825 const char *errstr = btrfs_decode_error(ret);
2827 "discard failed while removing blockgroup: errno=%d %s",
2835 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
2836 u64 bytenr, u64 num_bytes, bool is_data)
2841 struct btrfs_root *csum_root;
2843 csum_root = btrfs_csum_root(trans->fs_info, bytenr);
2844 ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes);
2846 btrfs_abort_transaction(trans, ret);
2851 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
2853 btrfs_abort_transaction(trans, ret);
2857 ret = btrfs_update_block_group(trans, bytenr, num_bytes, false);
2859 btrfs_abort_transaction(trans, ret);
2865 * Drop one or more refs of @node.
2867 * 1. Locate the extent refs.
2868 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2869 * Locate it, then reduce the refs number or remove the ref line completely.
2871 * 2. Update the refs count in EXTENT/METADATA_ITEM
2873 * Inline backref case:
2875 * in extent tree we have:
2877 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2878 * refs 2 gen 6 flags DATA
2879 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2880 * extent data backref root FS_TREE objectid 257 offset 0 count 1
2882 * This function gets called with:
2884 * node->bytenr = 13631488
2885 * node->num_bytes = 1048576
2886 * root_objectid = FS_TREE
2887 * owner_objectid = 257
2891 * Then we should get some like:
2893 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2894 * refs 1 gen 6 flags DATA
2895 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2897 * Keyed backref case:
2899 * in extent tree we have:
2901 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2902 * refs 754 gen 6 flags DATA
2904 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2905 * extent data backref root FS_TREE objectid 866 offset 0 count 1
2907 * This function get called with:
2909 * node->bytenr = 13631488
2910 * node->num_bytes = 1048576
2911 * root_objectid = FS_TREE
2912 * owner_objectid = 866
2916 * Then we should get some like:
2918 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2919 * refs 753 gen 6 flags DATA
2921 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2923 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2924 struct btrfs_delayed_ref_node *node, u64 parent,
2925 u64 root_objectid, u64 owner_objectid,
2926 u64 owner_offset, int refs_to_drop,
2927 struct btrfs_delayed_extent_op *extent_op)
2929 struct btrfs_fs_info *info = trans->fs_info;
2930 struct btrfs_key key;
2931 struct btrfs_path *path;
2932 struct btrfs_root *extent_root;
2933 struct extent_buffer *leaf;
2934 struct btrfs_extent_item *ei;
2935 struct btrfs_extent_inline_ref *iref;
2938 int extent_slot = 0;
2939 int found_extent = 0;
2943 u64 bytenr = node->bytenr;
2944 u64 num_bytes = node->num_bytes;
2945 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2947 extent_root = btrfs_extent_root(info, bytenr);
2948 ASSERT(extent_root);
2950 path = btrfs_alloc_path();
2954 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2956 if (!is_data && refs_to_drop != 1) {
2958 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2959 node->bytenr, refs_to_drop);
2961 btrfs_abort_transaction(trans, ret);
2966 skinny_metadata = false;
2968 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2969 parent, root_objectid, owner_objectid,
2973 * Either the inline backref or the SHARED_DATA_REF/
2974 * SHARED_BLOCK_REF is found
2976 * Here is a quick path to locate EXTENT/METADATA_ITEM.
2977 * It's possible the EXTENT/METADATA_ITEM is near current slot.
2979 extent_slot = path->slots[0];
2980 while (extent_slot >= 0) {
2981 btrfs_item_key_to_cpu(path->nodes[0], &key,
2983 if (key.objectid != bytenr)
2985 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2986 key.offset == num_bytes) {
2990 if (key.type == BTRFS_METADATA_ITEM_KEY &&
2991 key.offset == owner_objectid) {
2996 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
2997 if (path->slots[0] - extent_slot > 5)
3002 if (!found_extent) {
3005 "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref");
3006 btrfs_abort_transaction(trans, -EUCLEAN);
3009 /* Must be SHARED_* item, remove the backref first */
3010 ret = remove_extent_backref(trans, extent_root, path,
3011 NULL, refs_to_drop, is_data);
3013 btrfs_abort_transaction(trans, ret);
3016 btrfs_release_path(path);
3018 /* Slow path to locate EXTENT/METADATA_ITEM */
3019 key.objectid = bytenr;
3020 key.type = BTRFS_EXTENT_ITEM_KEY;
3021 key.offset = num_bytes;
3023 if (!is_data && skinny_metadata) {
3024 key.type = BTRFS_METADATA_ITEM_KEY;
3025 key.offset = owner_objectid;
3028 ret = btrfs_search_slot(trans, extent_root,
3030 if (ret > 0 && skinny_metadata && path->slots[0]) {
3032 * Couldn't find our skinny metadata item,
3033 * see if we have ye olde extent item.
3036 btrfs_item_key_to_cpu(path->nodes[0], &key,
3038 if (key.objectid == bytenr &&
3039 key.type == BTRFS_EXTENT_ITEM_KEY &&
3040 key.offset == num_bytes)
3044 if (ret > 0 && skinny_metadata) {
3045 skinny_metadata = false;
3046 key.objectid = bytenr;
3047 key.type = BTRFS_EXTENT_ITEM_KEY;
3048 key.offset = num_bytes;
3049 btrfs_release_path(path);
3050 ret = btrfs_search_slot(trans, extent_root,
3056 "umm, got %d back from search, was looking for %llu",
3059 btrfs_print_leaf(path->nodes[0]);
3062 btrfs_abort_transaction(trans, ret);
3065 extent_slot = path->slots[0];
3067 } else if (WARN_ON(ret == -ENOENT)) {
3068 btrfs_print_leaf(path->nodes[0]);
3070 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
3071 bytenr, parent, root_objectid, owner_objectid,
3073 btrfs_abort_transaction(trans, ret);
3076 btrfs_abort_transaction(trans, ret);
3080 leaf = path->nodes[0];
3081 item_size = btrfs_item_size(leaf, extent_slot);
3082 if (unlikely(item_size < sizeof(*ei))) {
3084 btrfs_print_v0_err(info);
3085 btrfs_abort_transaction(trans, ret);
3088 ei = btrfs_item_ptr(leaf, extent_slot,
3089 struct btrfs_extent_item);
3090 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3091 key.type == BTRFS_EXTENT_ITEM_KEY) {
3092 struct btrfs_tree_block_info *bi;
3093 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3095 "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu",
3096 key.objectid, key.type, key.offset,
3097 owner_objectid, item_size,
3098 sizeof(*ei) + sizeof(*bi));
3099 btrfs_abort_transaction(trans, -EUCLEAN);
3102 bi = (struct btrfs_tree_block_info *)(ei + 1);
3103 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3106 refs = btrfs_extent_refs(leaf, ei);
3107 if (refs < refs_to_drop) {
3109 "trying to drop %d refs but we only have %llu for bytenr %llu",
3110 refs_to_drop, refs, bytenr);
3111 btrfs_abort_transaction(trans, -EUCLEAN);
3114 refs -= refs_to_drop;
3118 __run_delayed_extent_op(extent_op, leaf, ei);
3120 * In the case of inline back ref, reference count will
3121 * be updated by remove_extent_backref
3124 if (!found_extent) {
3126 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found");
3127 btrfs_abort_transaction(trans, -EUCLEAN);
3131 btrfs_set_extent_refs(leaf, ei, refs);
3132 btrfs_mark_buffer_dirty(leaf);
3135 ret = remove_extent_backref(trans, extent_root, path,
3136 iref, refs_to_drop, is_data);
3138 btrfs_abort_transaction(trans, ret);
3143 /* In this branch refs == 1 */
3145 if (is_data && refs_to_drop !=
3146 extent_data_ref_count(path, iref)) {
3148 "invalid refs_to_drop, current refs %u refs_to_drop %u",
3149 extent_data_ref_count(path, iref),
3151 btrfs_abort_transaction(trans, -EUCLEAN);
3155 if (path->slots[0] != extent_slot) {
3157 "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref",
3158 key.objectid, key.type,
3160 btrfs_abort_transaction(trans, -EUCLEAN);
3165 * No inline ref, we must be at SHARED_* item,
3166 * And it's single ref, it must be:
3167 * | extent_slot ||extent_slot + 1|
3168 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3170 if (path->slots[0] != extent_slot + 1) {
3172 "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM");
3173 btrfs_abort_transaction(trans, -EUCLEAN);
3176 path->slots[0] = extent_slot;
3181 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3184 btrfs_abort_transaction(trans, ret);
3187 btrfs_release_path(path);
3189 ret = do_free_extent_accounting(trans, bytenr, num_bytes, is_data);
3191 btrfs_release_path(path);
3194 btrfs_free_path(path);
3198 * Leaf dump can take up a lot of log buffer, so we only do full leaf
3199 * dump for debug build.
3201 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) {
3202 btrfs_crit(info, "path->slots[0]=%d extent_slot=%d",
3203 path->slots[0], extent_slot);
3204 btrfs_print_leaf(path->nodes[0]);
3207 btrfs_free_path(path);
3212 * when we free an block, it is possible (and likely) that we free the last
3213 * delayed ref for that extent as well. This searches the delayed ref tree for
3214 * a given extent, and if there are no other delayed refs to be processed, it
3215 * removes it from the tree.
3217 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3220 struct btrfs_delayed_ref_head *head;
3221 struct btrfs_delayed_ref_root *delayed_refs;
3224 delayed_refs = &trans->transaction->delayed_refs;
3225 spin_lock(&delayed_refs->lock);
3226 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3228 goto out_delayed_unlock;
3230 spin_lock(&head->lock);
3231 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3234 if (cleanup_extent_op(head) != NULL)
3238 * waiting for the lock here would deadlock. If someone else has it
3239 * locked they are already in the process of dropping it anyway
3241 if (!mutex_trylock(&head->mutex))
3244 btrfs_delete_ref_head(delayed_refs, head);
3245 head->processing = 0;
3247 spin_unlock(&head->lock);
3248 spin_unlock(&delayed_refs->lock);
3250 BUG_ON(head->extent_op);
3251 if (head->must_insert_reserved)
3254 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3255 mutex_unlock(&head->mutex);
3256 btrfs_put_delayed_ref_head(head);
3259 spin_unlock(&head->lock);
3262 spin_unlock(&delayed_refs->lock);
3266 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3268 struct extent_buffer *buf,
3269 u64 parent, int last_ref)
3271 struct btrfs_fs_info *fs_info = trans->fs_info;
3272 struct btrfs_ref generic_ref = { 0 };
3275 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3276 buf->start, buf->len, parent);
3277 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3280 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3281 btrfs_ref_tree_mod(fs_info, &generic_ref);
3282 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3283 BUG_ON(ret); /* -ENOMEM */
3286 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3287 struct btrfs_block_group *cache;
3288 bool must_pin = false;
3290 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3291 ret = check_ref_cleanup(trans, buf->start);
3293 btrfs_redirty_list_add(trans->transaction, buf);
3298 cache = btrfs_lookup_block_group(fs_info, buf->start);
3300 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3301 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3302 btrfs_put_block_group(cache);
3307 * If there are tree mod log users we may have recorded mod log
3308 * operations for this node. If we re-allocate this node we
3309 * could replay operations on this node that happened when it
3310 * existed in a completely different root. For example if it
3311 * was part of root A, then was reallocated to root B, and we
3312 * are doing a btrfs_old_search_slot(root b), we could replay
3313 * operations that happened when the block was part of root A,
3314 * giving us an inconsistent view of the btree.
3316 * We are safe from races here because at this point no other
3317 * node or root points to this extent buffer, so if after this
3318 * check a new tree mod log user joins we will not have an
3319 * existing log of operations on this node that we have to
3322 if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
3325 if (must_pin || btrfs_is_zoned(fs_info)) {
3326 btrfs_redirty_list_add(trans->transaction, buf);
3327 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3328 btrfs_put_block_group(cache);
3332 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3334 btrfs_add_free_space(cache, buf->start, buf->len);
3335 btrfs_free_reserved_bytes(cache, buf->len, 0);
3336 btrfs_put_block_group(cache);
3337 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3342 * Deleting the buffer, clear the corrupt flag since it doesn't
3345 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3349 /* Can return -ENOMEM */
3350 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3352 struct btrfs_fs_info *fs_info = trans->fs_info;
3355 if (btrfs_is_testing(fs_info))
3359 * tree log blocks never actually go into the extent allocation
3360 * tree, just update pinning info and exit early.
3362 if ((ref->type == BTRFS_REF_METADATA &&
3363 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3364 (ref->type == BTRFS_REF_DATA &&
3365 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)) {
3366 /* unlocks the pinned mutex */
3367 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3369 } else if (ref->type == BTRFS_REF_METADATA) {
3370 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3372 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3375 if (!((ref->type == BTRFS_REF_METADATA &&
3376 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3377 (ref->type == BTRFS_REF_DATA &&
3378 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)))
3379 btrfs_ref_tree_mod(fs_info, ref);
3384 enum btrfs_loop_type {
3385 LOOP_CACHING_NOWAIT,
3392 btrfs_lock_block_group(struct btrfs_block_group *cache,
3396 down_read(&cache->data_rwsem);
3399 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3402 btrfs_get_block_group(cache);
3404 down_read(&cache->data_rwsem);
3407 static struct btrfs_block_group *btrfs_lock_cluster(
3408 struct btrfs_block_group *block_group,
3409 struct btrfs_free_cluster *cluster,
3411 __acquires(&cluster->refill_lock)
3413 struct btrfs_block_group *used_bg = NULL;
3415 spin_lock(&cluster->refill_lock);
3417 used_bg = cluster->block_group;
3421 if (used_bg == block_group)
3424 btrfs_get_block_group(used_bg);
3429 if (down_read_trylock(&used_bg->data_rwsem))
3432 spin_unlock(&cluster->refill_lock);
3434 /* We should only have one-level nested. */
3435 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3437 spin_lock(&cluster->refill_lock);
3438 if (used_bg == cluster->block_group)
3441 up_read(&used_bg->data_rwsem);
3442 btrfs_put_block_group(used_bg);
3447 btrfs_release_block_group(struct btrfs_block_group *cache,
3451 up_read(&cache->data_rwsem);
3452 btrfs_put_block_group(cache);
3455 enum btrfs_extent_allocation_policy {
3456 BTRFS_EXTENT_ALLOC_CLUSTERED,
3457 BTRFS_EXTENT_ALLOC_ZONED,
3461 * Structure used internally for find_free_extent() function. Wraps needed
3464 struct find_free_extent_ctl {
3465 /* Basic allocation info */
3473 /* Where to start the search inside the bg */
3476 /* For clustered allocation */
3478 struct btrfs_free_cluster *last_ptr;
3481 bool have_caching_bg;
3482 bool orig_have_caching_bg;
3484 /* Allocation is called for tree-log */
3487 /* Allocation is called for data relocation */
3488 bool for_data_reloc;
3490 /* RAID index, converted from flags */
3494 * Current loop number, check find_free_extent_update_loop() for details
3499 * Whether we're refilling a cluster, if true we need to re-search
3500 * current block group but don't try to refill the cluster again.
3502 bool retry_clustered;
3505 * Whether we're updating free space cache, if true we need to re-search
3506 * current block group but don't try updating free space cache again.
3508 bool retry_unclustered;
3510 /* If current block group is cached */
3513 /* Max contiguous hole found */
3514 u64 max_extent_size;
3516 /* Total free space from free space cache, not always contiguous */
3517 u64 total_free_space;
3522 /* Hint where to start looking for an empty space */
3525 /* Allocation policy */
3526 enum btrfs_extent_allocation_policy policy;
3531 * Helper function for find_free_extent().
3533 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3534 * Return -EAGAIN to inform caller that we need to re-search this block group
3535 * Return >0 to inform caller that we find nothing
3536 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3538 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3539 struct find_free_extent_ctl *ffe_ctl,
3540 struct btrfs_block_group **cluster_bg_ret)
3542 struct btrfs_block_group *cluster_bg;
3543 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3544 u64 aligned_cluster;
3548 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3550 goto refill_cluster;
3551 if (cluster_bg != bg && (cluster_bg->ro ||
3552 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3553 goto release_cluster;
3555 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3556 ffe_ctl->num_bytes, cluster_bg->start,
3557 &ffe_ctl->max_extent_size);
3559 /* We have a block, we're done */
3560 spin_unlock(&last_ptr->refill_lock);
3561 trace_btrfs_reserve_extent_cluster(cluster_bg,
3562 ffe_ctl->search_start, ffe_ctl->num_bytes);
3563 *cluster_bg_ret = cluster_bg;
3564 ffe_ctl->found_offset = offset;
3567 WARN_ON(last_ptr->block_group != cluster_bg);
3571 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3572 * lets just skip it and let the allocator find whatever block it can
3573 * find. If we reach this point, we will have tried the cluster
3574 * allocator plenty of times and not have found anything, so we are
3575 * likely way too fragmented for the clustering stuff to find anything.
3577 * However, if the cluster is taken from the current block group,
3578 * release the cluster first, so that we stand a better chance of
3579 * succeeding in the unclustered allocation.
3581 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3582 spin_unlock(&last_ptr->refill_lock);
3583 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3587 /* This cluster didn't work out, free it and start over */
3588 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3590 if (cluster_bg != bg)
3591 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3594 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3595 spin_unlock(&last_ptr->refill_lock);
3599 aligned_cluster = max_t(u64,
3600 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3601 bg->full_stripe_len);
3602 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3603 ffe_ctl->num_bytes, aligned_cluster);
3605 /* Now pull our allocation out of this cluster */
3606 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3607 ffe_ctl->num_bytes, ffe_ctl->search_start,
3608 &ffe_ctl->max_extent_size);
3610 /* We found one, proceed */
3611 spin_unlock(&last_ptr->refill_lock);
3612 trace_btrfs_reserve_extent_cluster(bg,
3613 ffe_ctl->search_start,
3614 ffe_ctl->num_bytes);
3615 ffe_ctl->found_offset = offset;
3618 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3619 !ffe_ctl->retry_clustered) {
3620 spin_unlock(&last_ptr->refill_lock);
3622 ffe_ctl->retry_clustered = true;
3623 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3624 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3628 * At this point we either didn't find a cluster or we weren't able to
3629 * allocate a block from our cluster. Free the cluster we've been
3630 * trying to use, and go to the next block group.
3632 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3633 spin_unlock(&last_ptr->refill_lock);
3638 * Return >0 to inform caller that we find nothing
3639 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3640 * Return -EAGAIN to inform caller that we need to re-search this block group
3642 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3643 struct find_free_extent_ctl *ffe_ctl)
3645 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3649 * We are doing an unclustered allocation, set the fragmented flag so
3650 * we don't bother trying to setup a cluster again until we get more
3653 if (unlikely(last_ptr)) {
3654 spin_lock(&last_ptr->lock);
3655 last_ptr->fragmented = 1;
3656 spin_unlock(&last_ptr->lock);
3658 if (ffe_ctl->cached) {
3659 struct btrfs_free_space_ctl *free_space_ctl;
3661 free_space_ctl = bg->free_space_ctl;
3662 spin_lock(&free_space_ctl->tree_lock);
3663 if (free_space_ctl->free_space <
3664 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3665 ffe_ctl->empty_size) {
3666 ffe_ctl->total_free_space = max_t(u64,
3667 ffe_ctl->total_free_space,
3668 free_space_ctl->free_space);
3669 spin_unlock(&free_space_ctl->tree_lock);
3672 spin_unlock(&free_space_ctl->tree_lock);
3675 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3676 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3677 &ffe_ctl->max_extent_size);
3680 * If we didn't find a chunk, and we haven't failed on this block group
3681 * before, and this block group is in the middle of caching and we are
3682 * ok with waiting, then go ahead and wait for progress to be made, and
3683 * set @retry_unclustered to true.
3685 * If @retry_unclustered is true then we've already waited on this
3686 * block group once and should move on to the next block group.
3688 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3689 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
3690 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3691 ffe_ctl->empty_size);
3692 ffe_ctl->retry_unclustered = true;
3694 } else if (!offset) {
3697 ffe_ctl->found_offset = offset;
3701 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3702 struct find_free_extent_ctl *ffe_ctl,
3703 struct btrfs_block_group **bg_ret)
3707 /* We want to try and use the cluster allocator, so lets look there */
3708 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3709 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3710 if (ret >= 0 || ret == -EAGAIN)
3712 /* ret == -ENOENT case falls through */
3715 return find_free_extent_unclustered(block_group, ffe_ctl);
3719 * Tree-log block group locking
3720 * ============================
3722 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3723 * indicates the starting address of a block group, which is reserved only
3724 * for tree-log metadata.
3731 * fs_info::treelog_bg_lock
3735 * Simple allocator for sequential-only block group. It only allows sequential
3736 * allocation. No need to play with trees. This function also reserves the
3737 * bytes as in btrfs_add_reserved_bytes.
3739 static int do_allocation_zoned(struct btrfs_block_group *block_group,
3740 struct find_free_extent_ctl *ffe_ctl,
3741 struct btrfs_block_group **bg_ret)
3743 struct btrfs_fs_info *fs_info = block_group->fs_info;
3744 struct btrfs_space_info *space_info = block_group->space_info;
3745 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3746 u64 start = block_group->start;
3747 u64 num_bytes = ffe_ctl->num_bytes;
3749 u64 bytenr = block_group->start;
3751 u64 data_reloc_bytenr;
3755 ASSERT(btrfs_is_zoned(block_group->fs_info));
3758 * Do not allow non-tree-log blocks in the dedicated tree-log block
3759 * group, and vice versa.
3761 spin_lock(&fs_info->treelog_bg_lock);
3762 log_bytenr = fs_info->treelog_bg;
3763 if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3764 (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3766 spin_unlock(&fs_info->treelog_bg_lock);
3771 * Do not allow non-relocation blocks in the dedicated relocation block
3772 * group, and vice versa.
3774 spin_lock(&fs_info->relocation_bg_lock);
3775 data_reloc_bytenr = fs_info->data_reloc_bg;
3776 if (data_reloc_bytenr &&
3777 ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3778 (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3780 spin_unlock(&fs_info->relocation_bg_lock);
3784 /* Check RO and no space case before trying to activate it */
3785 spin_lock(&block_group->lock);
3786 if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) {
3789 * May need to clear fs_info->{treelog,data_reloc}_bg.
3790 * Return the error after taking the locks.
3793 spin_unlock(&block_group->lock);
3795 if (!ret && !btrfs_zone_activate(block_group)) {
3798 * May need to clear fs_info->{treelog,data_reloc}_bg.
3799 * Return the error after taking the locks.
3803 spin_lock(&space_info->lock);
3804 spin_lock(&block_group->lock);
3805 spin_lock(&fs_info->treelog_bg_lock);
3806 spin_lock(&fs_info->relocation_bg_lock);
3811 ASSERT(!ffe_ctl->for_treelog ||
3812 block_group->start == fs_info->treelog_bg ||
3813 fs_info->treelog_bg == 0);
3814 ASSERT(!ffe_ctl->for_data_reloc ||
3815 block_group->start == fs_info->data_reloc_bg ||
3816 fs_info->data_reloc_bg == 0);
3818 if (block_group->ro ||
3819 (!ffe_ctl->for_data_reloc &&
3820 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))) {
3826 * Do not allow currently using block group to be tree-log dedicated
3829 if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3830 (block_group->used || block_group->reserved)) {
3836 * Do not allow currently used block group to be the data relocation
3837 * dedicated block group.
3839 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3840 (block_group->used || block_group->reserved)) {
3845 WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3846 avail = block_group->zone_capacity - block_group->alloc_offset;
3847 if (avail < num_bytes) {
3848 if (ffe_ctl->max_extent_size < avail) {
3850 * With sequential allocator, free space is always
3853 ffe_ctl->max_extent_size = avail;
3854 ffe_ctl->total_free_space = avail;
3860 if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3861 fs_info->treelog_bg = block_group->start;
3863 if (ffe_ctl->for_data_reloc) {
3864 if (!fs_info->data_reloc_bg)
3865 fs_info->data_reloc_bg = block_group->start;
3867 * Do not allow allocations from this block group, unless it is
3868 * for data relocation. Compared to increasing the ->ro, setting
3869 * the ->zoned_data_reloc_ongoing flag still allows nocow
3870 * writers to come in. See btrfs_inc_nocow_writers().
3872 * We need to disable an allocation to avoid an allocation of
3873 * regular (non-relocation data) extent. With mix of relocation
3874 * extents and regular extents, we can dispatch WRITE commands
3875 * (for relocation extents) and ZONE APPEND commands (for
3876 * regular extents) at the same time to the same zone, which
3877 * easily break the write pointer.
3879 * Also, this flag avoids this block group to be zone finished.
3881 set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags);
3884 ffe_ctl->found_offset = start + block_group->alloc_offset;
3885 block_group->alloc_offset += num_bytes;
3886 spin_lock(&ctl->tree_lock);
3887 ctl->free_space -= num_bytes;
3888 spin_unlock(&ctl->tree_lock);
3891 * We do not check if found_offset is aligned to stripesize. The
3892 * address is anyway rewritten when using zone append writing.
3895 ffe_ctl->search_start = ffe_ctl->found_offset;
3898 if (ret && ffe_ctl->for_treelog)
3899 fs_info->treelog_bg = 0;
3900 if (ret && ffe_ctl->for_data_reloc)
3901 fs_info->data_reloc_bg = 0;
3902 spin_unlock(&fs_info->relocation_bg_lock);
3903 spin_unlock(&fs_info->treelog_bg_lock);
3904 spin_unlock(&block_group->lock);
3905 spin_unlock(&space_info->lock);
3909 static int do_allocation(struct btrfs_block_group *block_group,
3910 struct find_free_extent_ctl *ffe_ctl,
3911 struct btrfs_block_group **bg_ret)
3913 switch (ffe_ctl->policy) {
3914 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3915 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3916 case BTRFS_EXTENT_ALLOC_ZONED:
3917 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
3923 static void release_block_group(struct btrfs_block_group *block_group,
3924 struct find_free_extent_ctl *ffe_ctl,
3927 switch (ffe_ctl->policy) {
3928 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3929 ffe_ctl->retry_clustered = false;
3930 ffe_ctl->retry_unclustered = false;
3932 case BTRFS_EXTENT_ALLOC_ZONED:
3939 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
3941 btrfs_release_block_group(block_group, delalloc);
3944 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
3945 struct btrfs_key *ins)
3947 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3949 if (!ffe_ctl->use_cluster && last_ptr) {
3950 spin_lock(&last_ptr->lock);
3951 last_ptr->window_start = ins->objectid;
3952 spin_unlock(&last_ptr->lock);
3956 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
3957 struct btrfs_key *ins)
3959 switch (ffe_ctl->policy) {
3960 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3961 found_extent_clustered(ffe_ctl, ins);
3963 case BTRFS_EXTENT_ALLOC_ZONED:
3971 static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info,
3972 struct find_free_extent_ctl *ffe_ctl)
3974 /* If we can activate new zone, just allocate a chunk and use it */
3975 if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags))
3979 * We already reached the max active zones. Try to finish one block
3980 * group to make a room for a new block group. This is only possible
3981 * for a data block group because btrfs_zone_finish() may need to wait
3982 * for a running transaction which can cause a deadlock for metadata
3985 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
3986 int ret = btrfs_zone_finish_one_bg(fs_info);
3995 * If we have enough free space left in an already active block group
3996 * and we can't activate any other zone now, do not allow allocating a
3997 * new chunk and let find_free_extent() retry with a smaller size.
3999 if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size)
4003 * Even min_alloc_size is not left in any block groups. Since we cannot
4004 * activate a new block group, allocating it may not help. Let's tell a
4005 * caller to try again and hope it progress something by writing some
4006 * parts of the region. That is only possible for data block groups,
4007 * where a part of the region can be written.
4009 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)
4013 * We cannot activate a new block group and no enough space left in any
4014 * block groups. So, allocating a new block group may not help. But,
4015 * there is nothing to do anyway, so let's go with it.
4020 static int can_allocate_chunk(struct btrfs_fs_info *fs_info,
4021 struct find_free_extent_ctl *ffe_ctl)
4023 switch (ffe_ctl->policy) {
4024 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4026 case BTRFS_EXTENT_ALLOC_ZONED:
4027 return can_allocate_chunk_zoned(fs_info, ffe_ctl);
4033 static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl)
4035 switch (ffe_ctl->policy) {
4036 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4038 * If we can't allocate a new chunk we've already looped through
4039 * at least once, move on to the NO_EMPTY_SIZE case.
4041 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
4043 case BTRFS_EXTENT_ALLOC_ZONED:
4052 * Return >0 means caller needs to re-search for free extent
4053 * Return 0 means we have the needed free extent.
4054 * Return <0 means we failed to locate any free extent.
4056 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
4057 struct btrfs_key *ins,
4058 struct find_free_extent_ctl *ffe_ctl,
4061 struct btrfs_root *root = fs_info->chunk_root;
4064 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
4065 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
4066 ffe_ctl->orig_have_caching_bg = true;
4068 if (ins->objectid) {
4069 found_extent(ffe_ctl, ins);
4073 if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
4077 if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
4081 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4082 * caching kthreads as we move along
4083 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4084 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4085 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4088 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
4090 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
4092 * We want to skip the LOOP_CACHING_WAIT step if we
4093 * don't have any uncached bgs and we've already done a
4094 * full search through.
4096 if (ffe_ctl->orig_have_caching_bg || !full_search)
4097 ffe_ctl->loop = LOOP_CACHING_WAIT;
4099 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
4104 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
4105 struct btrfs_trans_handle *trans;
4108 /*Check if allocation policy allows to create a new chunk */
4109 ret = can_allocate_chunk(fs_info, ffe_ctl);
4113 trans = current->journal_info;
4117 trans = btrfs_join_transaction(root);
4119 if (IS_ERR(trans)) {
4120 ret = PTR_ERR(trans);
4124 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
4125 CHUNK_ALLOC_FORCE_FOR_EXTENT);
4127 /* Do not bail out on ENOSPC since we can do more. */
4129 ret = chunk_allocation_failed(ffe_ctl);
4131 btrfs_abort_transaction(trans, ret);
4135 btrfs_end_transaction(trans);
4140 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4141 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4145 * Don't loop again if we already have no empty_size and
4148 if (ffe_ctl->empty_size == 0 &&
4149 ffe_ctl->empty_cluster == 0)
4151 ffe_ctl->empty_size = 0;
4152 ffe_ctl->empty_cluster = 0;
4159 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4160 struct find_free_extent_ctl *ffe_ctl,
4161 struct btrfs_space_info *space_info,
4162 struct btrfs_key *ins)
4165 * If our free space is heavily fragmented we may not be able to make
4166 * big contiguous allocations, so instead of doing the expensive search
4167 * for free space, simply return ENOSPC with our max_extent_size so we
4168 * can go ahead and search for a more manageable chunk.
4170 * If our max_extent_size is large enough for our allocation simply
4171 * disable clustering since we will likely not be able to find enough
4172 * space to create a cluster and induce latency trying.
4174 if (space_info->max_extent_size) {
4175 spin_lock(&space_info->lock);
4176 if (space_info->max_extent_size &&
4177 ffe_ctl->num_bytes > space_info->max_extent_size) {
4178 ins->offset = space_info->max_extent_size;
4179 spin_unlock(&space_info->lock);
4181 } else if (space_info->max_extent_size) {
4182 ffe_ctl->use_cluster = false;
4184 spin_unlock(&space_info->lock);
4187 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4188 &ffe_ctl->empty_cluster);
4189 if (ffe_ctl->last_ptr) {
4190 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4192 spin_lock(&last_ptr->lock);
4193 if (last_ptr->block_group)
4194 ffe_ctl->hint_byte = last_ptr->window_start;
4195 if (last_ptr->fragmented) {
4197 * We still set window_start so we can keep track of the
4198 * last place we found an allocation to try and save
4201 ffe_ctl->hint_byte = last_ptr->window_start;
4202 ffe_ctl->use_cluster = false;
4204 spin_unlock(&last_ptr->lock);
4210 static int prepare_allocation_zoned(struct btrfs_fs_info *fs_info,
4211 struct find_free_extent_ctl *ffe_ctl)
4213 if (ffe_ctl->for_treelog) {
4214 spin_lock(&fs_info->treelog_bg_lock);
4215 if (fs_info->treelog_bg)
4216 ffe_ctl->hint_byte = fs_info->treelog_bg;
4217 spin_unlock(&fs_info->treelog_bg_lock);
4218 } else if (ffe_ctl->for_data_reloc) {
4219 spin_lock(&fs_info->relocation_bg_lock);
4220 if (fs_info->data_reloc_bg)
4221 ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4222 spin_unlock(&fs_info->relocation_bg_lock);
4223 } else if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
4224 struct btrfs_block_group *block_group;
4226 spin_lock(&fs_info->zone_active_bgs_lock);
4227 list_for_each_entry(block_group, &fs_info->zone_active_bgs, active_bg_list) {
4229 * No lock is OK here because avail is monotinically
4230 * decreasing, and this is just a hint.
4232 u64 avail = block_group->zone_capacity - block_group->alloc_offset;
4234 if (block_group_bits(block_group, ffe_ctl->flags) &&
4235 avail >= ffe_ctl->num_bytes) {
4236 ffe_ctl->hint_byte = block_group->start;
4240 spin_unlock(&fs_info->zone_active_bgs_lock);
4246 static int prepare_allocation(struct btrfs_fs_info *fs_info,
4247 struct find_free_extent_ctl *ffe_ctl,
4248 struct btrfs_space_info *space_info,
4249 struct btrfs_key *ins)
4251 switch (ffe_ctl->policy) {
4252 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4253 return prepare_allocation_clustered(fs_info, ffe_ctl,
4255 case BTRFS_EXTENT_ALLOC_ZONED:
4256 return prepare_allocation_zoned(fs_info, ffe_ctl);
4263 * walks the btree of allocated extents and find a hole of a given size.
4264 * The key ins is changed to record the hole:
4265 * ins->objectid == start position
4266 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4267 * ins->offset == the size of the hole.
4268 * Any available blocks before search_start are skipped.
4270 * If there is no suitable free space, we will record the max size of
4271 * the free space extent currently.
4273 * The overall logic and call chain:
4275 * find_free_extent()
4276 * |- Iterate through all block groups
4277 * | |- Get a valid block group
4278 * | |- Try to do clustered allocation in that block group
4279 * | |- Try to do unclustered allocation in that block group
4280 * | |- Check if the result is valid
4281 * | | |- If valid, then exit
4282 * | |- Jump to next block group
4284 * |- Push harder to find free extents
4285 * |- If not found, re-iterate all block groups
4287 static noinline int find_free_extent(struct btrfs_root *root,
4288 struct btrfs_key *ins,
4289 struct find_free_extent_ctl *ffe_ctl)
4291 struct btrfs_fs_info *fs_info = root->fs_info;
4293 int cache_block_group_error = 0;
4294 struct btrfs_block_group *block_group = NULL;
4295 struct btrfs_space_info *space_info;
4296 bool full_search = false;
4298 WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4300 ffe_ctl->search_start = 0;
4301 /* For clustered allocation */
4302 ffe_ctl->empty_cluster = 0;
4303 ffe_ctl->last_ptr = NULL;
4304 ffe_ctl->use_cluster = true;
4305 ffe_ctl->have_caching_bg = false;
4306 ffe_ctl->orig_have_caching_bg = false;
4307 ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4309 /* For clustered allocation */
4310 ffe_ctl->retry_clustered = false;
4311 ffe_ctl->retry_unclustered = false;
4312 ffe_ctl->cached = 0;
4313 ffe_ctl->max_extent_size = 0;
4314 ffe_ctl->total_free_space = 0;
4315 ffe_ctl->found_offset = 0;
4316 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4318 if (btrfs_is_zoned(fs_info))
4319 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4321 ins->type = BTRFS_EXTENT_ITEM_KEY;
4325 trace_find_free_extent(root, ffe_ctl->num_bytes, ffe_ctl->empty_size,
4328 space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4330 btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4334 ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4338 ffe_ctl->search_start = max(ffe_ctl->search_start,
4339 first_logical_byte(fs_info));
4340 ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4341 if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4342 block_group = btrfs_lookup_block_group(fs_info,
4343 ffe_ctl->search_start);
4345 * we don't want to use the block group if it doesn't match our
4346 * allocation bits, or if its not cached.
4348 * However if we are re-searching with an ideal block group
4349 * picked out then we don't care that the block group is cached.
4351 if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4352 block_group->cached != BTRFS_CACHE_NO) {
4353 down_read(&space_info->groups_sem);
4354 if (list_empty(&block_group->list) ||
4357 * someone is removing this block group,
4358 * we can't jump into the have_block_group
4359 * target because our list pointers are not
4362 btrfs_put_block_group(block_group);
4363 up_read(&space_info->groups_sem);
4365 ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4366 block_group->flags);
4367 btrfs_lock_block_group(block_group,
4369 goto have_block_group;
4371 } else if (block_group) {
4372 btrfs_put_block_group(block_group);
4376 ffe_ctl->have_caching_bg = false;
4377 if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4378 ffe_ctl->index == 0)
4380 down_read(&space_info->groups_sem);
4381 list_for_each_entry(block_group,
4382 &space_info->block_groups[ffe_ctl->index], list) {
4383 struct btrfs_block_group *bg_ret;
4385 /* If the block group is read-only, we can skip it entirely. */
4386 if (unlikely(block_group->ro)) {
4387 if (ffe_ctl->for_treelog)
4388 btrfs_clear_treelog_bg(block_group);
4389 if (ffe_ctl->for_data_reloc)
4390 btrfs_clear_data_reloc_bg(block_group);
4394 btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4395 ffe_ctl->search_start = block_group->start;
4398 * this can happen if we end up cycling through all the
4399 * raid types, but we want to make sure we only allocate
4400 * for the proper type.
4402 if (!block_group_bits(block_group, ffe_ctl->flags)) {
4403 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4404 BTRFS_BLOCK_GROUP_RAID1_MASK |
4405 BTRFS_BLOCK_GROUP_RAID56_MASK |
4406 BTRFS_BLOCK_GROUP_RAID10;
4409 * if they asked for extra copies and this block group
4410 * doesn't provide them, bail. This does allow us to
4411 * fill raid0 from raid1.
4413 if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4417 * This block group has different flags than we want.
4418 * It's possible that we have MIXED_GROUP flag but no
4419 * block group is mixed. Just skip such block group.
4421 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4426 ffe_ctl->cached = btrfs_block_group_done(block_group);
4427 if (unlikely(!ffe_ctl->cached)) {
4428 ffe_ctl->have_caching_bg = true;
4429 ret = btrfs_cache_block_group(block_group, false);
4432 * If we get ENOMEM here or something else we want to
4433 * try other block groups, because it may not be fatal.
4434 * However if we can't find anything else we need to
4435 * save our return here so that we return the actual
4436 * error that caused problems, not ENOSPC.
4439 if (!cache_block_group_error)
4440 cache_block_group_error = ret;
4447 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) {
4448 if (!cache_block_group_error)
4449 cache_block_group_error = -EIO;
4454 ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4456 if (bg_ret && bg_ret != block_group) {
4457 btrfs_release_block_group(block_group,
4459 block_group = bg_ret;
4461 } else if (ret == -EAGAIN) {
4462 goto have_block_group;
4463 } else if (ret > 0) {
4468 ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4469 fs_info->stripesize);
4471 /* move on to the next group */
4472 if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4473 block_group->start + block_group->length) {
4474 btrfs_add_free_space_unused(block_group,
4475 ffe_ctl->found_offset,
4476 ffe_ctl->num_bytes);
4480 if (ffe_ctl->found_offset < ffe_ctl->search_start)
4481 btrfs_add_free_space_unused(block_group,
4482 ffe_ctl->found_offset,
4483 ffe_ctl->search_start - ffe_ctl->found_offset);
4485 ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4488 if (ret == -EAGAIN) {
4489 btrfs_add_free_space_unused(block_group,
4490 ffe_ctl->found_offset,
4491 ffe_ctl->num_bytes);
4494 btrfs_inc_block_group_reservations(block_group);
4496 /* we are all good, lets return */
4497 ins->objectid = ffe_ctl->search_start;
4498 ins->offset = ffe_ctl->num_bytes;
4500 trace_btrfs_reserve_extent(block_group, ffe_ctl->search_start,
4501 ffe_ctl->num_bytes);
4502 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4505 release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4508 up_read(&space_info->groups_sem);
4510 ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4514 if (ret == -ENOSPC && !cache_block_group_error) {
4516 * Use ffe_ctl->total_free_space as fallback if we can't find
4517 * any contiguous hole.
4519 if (!ffe_ctl->max_extent_size)
4520 ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4521 spin_lock(&space_info->lock);
4522 space_info->max_extent_size = ffe_ctl->max_extent_size;
4523 spin_unlock(&space_info->lock);
4524 ins->offset = ffe_ctl->max_extent_size;
4525 } else if (ret == -ENOSPC) {
4526 ret = cache_block_group_error;
4532 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4533 * hole that is at least as big as @num_bytes.
4535 * @root - The root that will contain this extent
4537 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4538 * is used for accounting purposes. This value differs
4539 * from @num_bytes only in the case of compressed extents.
4541 * @num_bytes - Number of bytes to allocate on-disk.
4543 * @min_alloc_size - Indicates the minimum amount of space that the
4544 * allocator should try to satisfy. In some cases
4545 * @num_bytes may be larger than what is required and if
4546 * the filesystem is fragmented then allocation fails.
4547 * However, the presence of @min_alloc_size gives a
4548 * chance to try and satisfy the smaller allocation.
4550 * @empty_size - A hint that you plan on doing more COW. This is the
4551 * size in bytes the allocator should try to find free
4552 * next to the block it returns. This is just a hint and
4553 * may be ignored by the allocator.
4555 * @hint_byte - Hint to the allocator to start searching above the byte
4556 * address passed. It might be ignored.
4558 * @ins - This key is modified to record the found hole. It will
4559 * have the following values:
4560 * ins->objectid == start position
4561 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4562 * ins->offset == the size of the hole.
4564 * @is_data - Boolean flag indicating whether an extent is
4565 * allocated for data (true) or metadata (false)
4567 * @delalloc - Boolean flag indicating whether this allocation is for
4568 * delalloc or not. If 'true' data_rwsem of block groups
4569 * is going to be acquired.
4572 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4573 * case -ENOSPC is returned then @ins->offset will contain the size of the
4574 * largest available hole the allocator managed to find.
4576 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4577 u64 num_bytes, u64 min_alloc_size,
4578 u64 empty_size, u64 hint_byte,
4579 struct btrfs_key *ins, int is_data, int delalloc)
4581 struct btrfs_fs_info *fs_info = root->fs_info;
4582 struct find_free_extent_ctl ffe_ctl = {};
4583 bool final_tried = num_bytes == min_alloc_size;
4586 bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4587 bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4589 flags = get_alloc_profile_by_root(root, is_data);
4591 WARN_ON(num_bytes < fs_info->sectorsize);
4593 ffe_ctl.ram_bytes = ram_bytes;
4594 ffe_ctl.num_bytes = num_bytes;
4595 ffe_ctl.min_alloc_size = min_alloc_size;
4596 ffe_ctl.empty_size = empty_size;
4597 ffe_ctl.flags = flags;
4598 ffe_ctl.delalloc = delalloc;
4599 ffe_ctl.hint_byte = hint_byte;
4600 ffe_ctl.for_treelog = for_treelog;
4601 ffe_ctl.for_data_reloc = for_data_reloc;
4603 ret = find_free_extent(root, ins, &ffe_ctl);
4604 if (!ret && !is_data) {
4605 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4606 } else if (ret == -ENOSPC) {
4607 if (!final_tried && ins->offset) {
4608 num_bytes = min(num_bytes >> 1, ins->offset);
4609 num_bytes = round_down(num_bytes,
4610 fs_info->sectorsize);
4611 num_bytes = max(num_bytes, min_alloc_size);
4612 ram_bytes = num_bytes;
4613 if (num_bytes == min_alloc_size)
4616 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4617 struct btrfs_space_info *sinfo;
4619 sinfo = btrfs_find_space_info(fs_info, flags);
4621 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4622 flags, num_bytes, for_treelog, for_data_reloc);
4624 btrfs_dump_space_info(fs_info, sinfo,
4632 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4633 u64 start, u64 len, int delalloc)
4635 struct btrfs_block_group *cache;
4637 cache = btrfs_lookup_block_group(fs_info, start);
4639 btrfs_err(fs_info, "Unable to find block group for %llu",
4644 btrfs_add_free_space(cache, start, len);
4645 btrfs_free_reserved_bytes(cache, len, delalloc);
4646 trace_btrfs_reserved_extent_free(fs_info, start, len);
4648 btrfs_put_block_group(cache);
4652 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start,
4655 struct btrfs_block_group *cache;
4658 cache = btrfs_lookup_block_group(trans->fs_info, start);
4660 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4665 ret = pin_down_extent(trans, cache, start, len, 1);
4666 btrfs_put_block_group(cache);
4670 static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
4673 struct btrfs_fs_info *fs_info = trans->fs_info;
4676 ret = remove_from_free_space_tree(trans, bytenr, num_bytes);
4680 ret = btrfs_update_block_group(trans, bytenr, num_bytes, true);
4683 btrfs_err(fs_info, "update block group failed for %llu %llu",
4688 trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes);
4692 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4693 u64 parent, u64 root_objectid,
4694 u64 flags, u64 owner, u64 offset,
4695 struct btrfs_key *ins, int ref_mod)
4697 struct btrfs_fs_info *fs_info = trans->fs_info;
4698 struct btrfs_root *extent_root;
4700 struct btrfs_extent_item *extent_item;
4701 struct btrfs_extent_inline_ref *iref;
4702 struct btrfs_path *path;
4703 struct extent_buffer *leaf;
4708 type = BTRFS_SHARED_DATA_REF_KEY;
4710 type = BTRFS_EXTENT_DATA_REF_KEY;
4712 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4714 path = btrfs_alloc_path();
4718 extent_root = btrfs_extent_root(fs_info, ins->objectid);
4719 ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size);
4721 btrfs_free_path(path);
4725 leaf = path->nodes[0];
4726 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4727 struct btrfs_extent_item);
4728 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4729 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4730 btrfs_set_extent_flags(leaf, extent_item,
4731 flags | BTRFS_EXTENT_FLAG_DATA);
4733 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4734 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4736 struct btrfs_shared_data_ref *ref;
4737 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4738 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4739 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4741 struct btrfs_extent_data_ref *ref;
4742 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4743 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4744 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4745 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4746 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4749 btrfs_mark_buffer_dirty(path->nodes[0]);
4750 btrfs_free_path(path);
4752 return alloc_reserved_extent(trans, ins->objectid, ins->offset);
4755 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4756 struct btrfs_delayed_ref_node *node,
4757 struct btrfs_delayed_extent_op *extent_op)
4759 struct btrfs_fs_info *fs_info = trans->fs_info;
4760 struct btrfs_root *extent_root;
4762 struct btrfs_extent_item *extent_item;
4763 struct btrfs_key extent_key;
4764 struct btrfs_tree_block_info *block_info;
4765 struct btrfs_extent_inline_ref *iref;
4766 struct btrfs_path *path;
4767 struct extent_buffer *leaf;
4768 struct btrfs_delayed_tree_ref *ref;
4769 u32 size = sizeof(*extent_item) + sizeof(*iref);
4770 u64 flags = extent_op->flags_to_set;
4771 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4773 ref = btrfs_delayed_node_to_tree_ref(node);
4775 extent_key.objectid = node->bytenr;
4776 if (skinny_metadata) {
4777 extent_key.offset = ref->level;
4778 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4780 extent_key.offset = node->num_bytes;
4781 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4782 size += sizeof(*block_info);
4785 path = btrfs_alloc_path();
4789 extent_root = btrfs_extent_root(fs_info, extent_key.objectid);
4790 ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key,
4793 btrfs_free_path(path);
4797 leaf = path->nodes[0];
4798 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4799 struct btrfs_extent_item);
4800 btrfs_set_extent_refs(leaf, extent_item, 1);
4801 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4802 btrfs_set_extent_flags(leaf, extent_item,
4803 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4805 if (skinny_metadata) {
4806 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4808 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4809 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4810 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4811 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4814 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4815 btrfs_set_extent_inline_ref_type(leaf, iref,
4816 BTRFS_SHARED_BLOCK_REF_KEY);
4817 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4819 btrfs_set_extent_inline_ref_type(leaf, iref,
4820 BTRFS_TREE_BLOCK_REF_KEY);
4821 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4824 btrfs_mark_buffer_dirty(leaf);
4825 btrfs_free_path(path);
4827 return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize);
4830 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4831 struct btrfs_root *root, u64 owner,
4832 u64 offset, u64 ram_bytes,
4833 struct btrfs_key *ins)
4835 struct btrfs_ref generic_ref = { 0 };
4837 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4839 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4840 ins->objectid, ins->offset, 0);
4841 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner,
4843 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4845 return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4849 * this is used by the tree logging recovery code. It records that
4850 * an extent has been allocated and makes sure to clear the free
4851 * space cache bits as well
4853 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4854 u64 root_objectid, u64 owner, u64 offset,
4855 struct btrfs_key *ins)
4857 struct btrfs_fs_info *fs_info = trans->fs_info;
4859 struct btrfs_block_group *block_group;
4860 struct btrfs_space_info *space_info;
4863 * Mixed block groups will exclude before processing the log so we only
4864 * need to do the exclude dance if this fs isn't mixed.
4866 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4867 ret = __exclude_logged_extent(fs_info, ins->objectid,
4873 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4877 space_info = block_group->space_info;
4878 spin_lock(&space_info->lock);
4879 spin_lock(&block_group->lock);
4880 space_info->bytes_reserved += ins->offset;
4881 block_group->reserved += ins->offset;
4882 spin_unlock(&block_group->lock);
4883 spin_unlock(&space_info->lock);
4885 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4888 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4889 btrfs_put_block_group(block_group);
4893 static struct extent_buffer *
4894 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4895 u64 bytenr, int level, u64 owner,
4896 enum btrfs_lock_nesting nest)
4898 struct btrfs_fs_info *fs_info = root->fs_info;
4899 struct extent_buffer *buf;
4900 u64 lockdep_owner = owner;
4902 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
4907 * Extra safety check in case the extent tree is corrupted and extent
4908 * allocator chooses to use a tree block which is already used and
4911 if (buf->lock_owner == current->pid) {
4912 btrfs_err_rl(fs_info,
4913 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4914 buf->start, btrfs_header_owner(buf), current->pid);
4915 free_extent_buffer(buf);
4916 return ERR_PTR(-EUCLEAN);
4920 * The reloc trees are just snapshots, so we need them to appear to be
4921 * just like any other fs tree WRT lockdep.
4923 * The exception however is in replace_path() in relocation, where we
4924 * hold the lock on the original fs root and then search for the reloc
4925 * root. At that point we need to make sure any reloc root buffers are
4926 * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
4929 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID &&
4930 !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
4931 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
4933 /* btrfs_clean_tree_block() accesses generation field. */
4934 btrfs_set_header_generation(buf, trans->transid);
4937 * This needs to stay, because we could allocate a freed block from an
4938 * old tree into a new tree, so we need to make sure this new block is
4939 * set to the appropriate level and owner.
4941 btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level);
4943 __btrfs_tree_lock(buf, nest);
4944 btrfs_clean_tree_block(buf);
4945 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4946 clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags);
4948 set_extent_buffer_uptodate(buf);
4950 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4951 btrfs_set_header_level(buf, level);
4952 btrfs_set_header_bytenr(buf, buf->start);
4953 btrfs_set_header_generation(buf, trans->transid);
4954 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4955 btrfs_set_header_owner(buf, owner);
4956 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4957 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4958 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4959 buf->log_index = root->log_transid % 2;
4961 * we allow two log transactions at a time, use different
4962 * EXTENT bit to differentiate dirty pages.
4964 if (buf->log_index == 0)
4965 set_extent_dirty(&root->dirty_log_pages, buf->start,
4966 buf->start + buf->len - 1, GFP_NOFS);
4968 set_extent_new(&root->dirty_log_pages, buf->start,
4969 buf->start + buf->len - 1);
4971 buf->log_index = -1;
4972 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4973 buf->start + buf->len - 1, GFP_NOFS);
4975 /* this returns a buffer locked for blocking */
4980 * finds a free extent and does all the dirty work required for allocation
4981 * returns the tree buffer or an ERR_PTR on error.
4983 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4984 struct btrfs_root *root,
4985 u64 parent, u64 root_objectid,
4986 const struct btrfs_disk_key *key,
4987 int level, u64 hint,
4989 enum btrfs_lock_nesting nest)
4991 struct btrfs_fs_info *fs_info = root->fs_info;
4992 struct btrfs_key ins;
4993 struct btrfs_block_rsv *block_rsv;
4994 struct extent_buffer *buf;
4995 struct btrfs_delayed_extent_op *extent_op;
4996 struct btrfs_ref generic_ref = { 0 };
4999 u32 blocksize = fs_info->nodesize;
5000 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
5002 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
5003 if (btrfs_is_testing(fs_info)) {
5004 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
5005 level, root_objectid, nest);
5007 root->alloc_bytenr += blocksize;
5012 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
5013 if (IS_ERR(block_rsv))
5014 return ERR_CAST(block_rsv);
5016 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
5017 empty_size, hint, &ins, 0, 0);
5021 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
5022 root_objectid, nest);
5025 goto out_free_reserved;
5028 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5030 parent = ins.objectid;
5031 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5035 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5036 extent_op = btrfs_alloc_delayed_extent_op();
5042 memcpy(&extent_op->key, key, sizeof(extent_op->key));
5044 memset(&extent_op->key, 0, sizeof(extent_op->key));
5045 extent_op->flags_to_set = flags;
5046 extent_op->update_key = skinny_metadata ? false : true;
5047 extent_op->update_flags = true;
5048 extent_op->level = level;
5050 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
5051 ins.objectid, ins.offset, parent);
5052 btrfs_init_tree_ref(&generic_ref, level, root_objectid,
5053 root->root_key.objectid, false);
5054 btrfs_ref_tree_mod(fs_info, &generic_ref);
5055 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
5057 goto out_free_delayed;
5062 btrfs_free_delayed_extent_op(extent_op);
5064 btrfs_tree_unlock(buf);
5065 free_extent_buffer(buf);
5067 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
5069 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
5070 return ERR_PTR(ret);
5073 struct walk_control {
5074 u64 refs[BTRFS_MAX_LEVEL];
5075 u64 flags[BTRFS_MAX_LEVEL];
5076 struct btrfs_key update_progress;
5077 struct btrfs_key drop_progress;
5089 #define DROP_REFERENCE 1
5090 #define UPDATE_BACKREF 2
5092 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5093 struct btrfs_root *root,
5094 struct walk_control *wc,
5095 struct btrfs_path *path)
5097 struct btrfs_fs_info *fs_info = root->fs_info;
5103 struct btrfs_key key;
5104 struct extent_buffer *eb;
5109 if (path->slots[wc->level] < wc->reada_slot) {
5110 wc->reada_count = wc->reada_count * 2 / 3;
5111 wc->reada_count = max(wc->reada_count, 2);
5113 wc->reada_count = wc->reada_count * 3 / 2;
5114 wc->reada_count = min_t(int, wc->reada_count,
5115 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
5118 eb = path->nodes[wc->level];
5119 nritems = btrfs_header_nritems(eb);
5121 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5122 if (nread >= wc->reada_count)
5126 bytenr = btrfs_node_blockptr(eb, slot);
5127 generation = btrfs_node_ptr_generation(eb, slot);
5129 if (slot == path->slots[wc->level])
5132 if (wc->stage == UPDATE_BACKREF &&
5133 generation <= root->root_key.offset)
5136 /* We don't lock the tree block, it's OK to be racy here */
5137 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5138 wc->level - 1, 1, &refs,
5140 /* We don't care about errors in readahead. */
5145 if (wc->stage == DROP_REFERENCE) {
5149 if (wc->level == 1 &&
5150 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5152 if (!wc->update_ref ||
5153 generation <= root->root_key.offset)
5155 btrfs_node_key_to_cpu(eb, &key, slot);
5156 ret = btrfs_comp_cpu_keys(&key,
5157 &wc->update_progress);
5161 if (wc->level == 1 &&
5162 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5166 btrfs_readahead_node_child(eb, slot);
5169 wc->reada_slot = slot;
5173 * helper to process tree block while walking down the tree.
5175 * when wc->stage == UPDATE_BACKREF, this function updates
5176 * back refs for pointers in the block.
5178 * NOTE: return value 1 means we should stop walking down.
5180 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5181 struct btrfs_root *root,
5182 struct btrfs_path *path,
5183 struct walk_control *wc, int lookup_info)
5185 struct btrfs_fs_info *fs_info = root->fs_info;
5186 int level = wc->level;
5187 struct extent_buffer *eb = path->nodes[level];
5188 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5191 if (wc->stage == UPDATE_BACKREF &&
5192 btrfs_header_owner(eb) != root->root_key.objectid)
5196 * when reference count of tree block is 1, it won't increase
5197 * again. once full backref flag is set, we never clear it.
5200 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5201 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5202 BUG_ON(!path->locks[level]);
5203 ret = btrfs_lookup_extent_info(trans, fs_info,
5204 eb->start, level, 1,
5207 BUG_ON(ret == -ENOMEM);
5210 BUG_ON(wc->refs[level] == 0);
5213 if (wc->stage == DROP_REFERENCE) {
5214 if (wc->refs[level] > 1)
5217 if (path->locks[level] && !wc->keep_locks) {
5218 btrfs_tree_unlock_rw(eb, path->locks[level]);
5219 path->locks[level] = 0;
5224 /* wc->stage == UPDATE_BACKREF */
5225 if (!(wc->flags[level] & flag)) {
5226 BUG_ON(!path->locks[level]);
5227 ret = btrfs_inc_ref(trans, root, eb, 1);
5228 BUG_ON(ret); /* -ENOMEM */
5229 ret = btrfs_dec_ref(trans, root, eb, 0);
5230 BUG_ON(ret); /* -ENOMEM */
5231 ret = btrfs_set_disk_extent_flags(trans, eb, flag,
5232 btrfs_header_level(eb));
5233 BUG_ON(ret); /* -ENOMEM */
5234 wc->flags[level] |= flag;
5238 * the block is shared by multiple trees, so it's not good to
5239 * keep the tree lock
5241 if (path->locks[level] && level > 0) {
5242 btrfs_tree_unlock_rw(eb, path->locks[level]);
5243 path->locks[level] = 0;
5249 * This is used to verify a ref exists for this root to deal with a bug where we
5250 * would have a drop_progress key that hadn't been updated properly.
5252 static int check_ref_exists(struct btrfs_trans_handle *trans,
5253 struct btrfs_root *root, u64 bytenr, u64 parent,
5256 struct btrfs_path *path;
5257 struct btrfs_extent_inline_ref *iref;
5260 path = btrfs_alloc_path();
5264 ret = lookup_extent_backref(trans, path, &iref, bytenr,
5265 root->fs_info->nodesize, parent,
5266 root->root_key.objectid, level, 0);
5267 btrfs_free_path(path);
5276 * helper to process tree block pointer.
5278 * when wc->stage == DROP_REFERENCE, this function checks
5279 * reference count of the block pointed to. if the block
5280 * is shared and we need update back refs for the subtree
5281 * rooted at the block, this function changes wc->stage to
5282 * UPDATE_BACKREF. if the block is shared and there is no
5283 * need to update back, this function drops the reference
5286 * NOTE: return value 1 means we should stop walking down.
5288 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5289 struct btrfs_root *root,
5290 struct btrfs_path *path,
5291 struct walk_control *wc, int *lookup_info)
5293 struct btrfs_fs_info *fs_info = root->fs_info;
5297 struct btrfs_key key;
5298 struct btrfs_key first_key;
5299 struct btrfs_ref ref = { 0 };
5300 struct extent_buffer *next;
5301 int level = wc->level;
5304 bool need_account = false;
5306 generation = btrfs_node_ptr_generation(path->nodes[level],
5307 path->slots[level]);
5309 * if the lower level block was created before the snapshot
5310 * was created, we know there is no need to update back refs
5313 if (wc->stage == UPDATE_BACKREF &&
5314 generation <= root->root_key.offset) {
5319 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5320 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
5321 path->slots[level]);
5323 next = find_extent_buffer(fs_info, bytenr);
5325 next = btrfs_find_create_tree_block(fs_info, bytenr,
5326 root->root_key.objectid, level - 1);
5328 return PTR_ERR(next);
5331 btrfs_tree_lock(next);
5333 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5334 &wc->refs[level - 1],
5335 &wc->flags[level - 1]);
5339 if (unlikely(wc->refs[level - 1] == 0)) {
5340 btrfs_err(fs_info, "Missing references.");
5346 if (wc->stage == DROP_REFERENCE) {
5347 if (wc->refs[level - 1] > 1) {
5348 need_account = true;
5350 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5353 if (!wc->update_ref ||
5354 generation <= root->root_key.offset)
5357 btrfs_node_key_to_cpu(path->nodes[level], &key,
5358 path->slots[level]);
5359 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5363 wc->stage = UPDATE_BACKREF;
5364 wc->shared_level = level - 1;
5368 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5372 if (!btrfs_buffer_uptodate(next, generation, 0)) {
5373 btrfs_tree_unlock(next);
5374 free_extent_buffer(next);
5380 if (reada && level == 1)
5381 reada_walk_down(trans, root, wc, path);
5382 next = read_tree_block(fs_info, bytenr, root->root_key.objectid,
5383 generation, level - 1, &first_key);
5385 return PTR_ERR(next);
5386 } else if (!extent_buffer_uptodate(next)) {
5387 free_extent_buffer(next);
5390 btrfs_tree_lock(next);
5394 ASSERT(level == btrfs_header_level(next));
5395 if (level != btrfs_header_level(next)) {
5396 btrfs_err(root->fs_info, "mismatched level");
5400 path->nodes[level] = next;
5401 path->slots[level] = 0;
5402 path->locks[level] = BTRFS_WRITE_LOCK;
5408 wc->refs[level - 1] = 0;
5409 wc->flags[level - 1] = 0;
5410 if (wc->stage == DROP_REFERENCE) {
5411 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5412 parent = path->nodes[level]->start;
5414 ASSERT(root->root_key.objectid ==
5415 btrfs_header_owner(path->nodes[level]));
5416 if (root->root_key.objectid !=
5417 btrfs_header_owner(path->nodes[level])) {
5418 btrfs_err(root->fs_info,
5419 "mismatched block owner");
5427 * If we had a drop_progress we need to verify the refs are set
5428 * as expected. If we find our ref then we know that from here
5429 * on out everything should be correct, and we can clear the
5432 if (wc->restarted) {
5433 ret = check_ref_exists(trans, root, bytenr, parent,
5444 * Reloc tree doesn't contribute to qgroup numbers, and we have
5445 * already accounted them at merge time (replace_path),
5446 * thus we could skip expensive subtree trace here.
5448 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5450 ret = btrfs_qgroup_trace_subtree(trans, next,
5451 generation, level - 1);
5453 btrfs_err_rl(fs_info,
5454 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5460 * We need to update the next key in our walk control so we can
5461 * update the drop_progress key accordingly. We don't care if
5462 * find_next_key doesn't find a key because that means we're at
5463 * the end and are going to clean up now.
5465 wc->drop_level = level;
5466 find_next_key(path, level, &wc->drop_progress);
5468 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5469 fs_info->nodesize, parent);
5470 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid,
5472 ret = btrfs_free_extent(trans, &ref);
5481 btrfs_tree_unlock(next);
5482 free_extent_buffer(next);
5488 * helper to process tree block while walking up the tree.
5490 * when wc->stage == DROP_REFERENCE, this function drops
5491 * reference count on the block.
5493 * when wc->stage == UPDATE_BACKREF, this function changes
5494 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5495 * to UPDATE_BACKREF previously while processing the block.
5497 * NOTE: return value 1 means we should stop walking up.
5499 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5500 struct btrfs_root *root,
5501 struct btrfs_path *path,
5502 struct walk_control *wc)
5504 struct btrfs_fs_info *fs_info = root->fs_info;
5506 int level = wc->level;
5507 struct extent_buffer *eb = path->nodes[level];
5510 if (wc->stage == UPDATE_BACKREF) {
5511 BUG_ON(wc->shared_level < level);
5512 if (level < wc->shared_level)
5515 ret = find_next_key(path, level + 1, &wc->update_progress);
5519 wc->stage = DROP_REFERENCE;
5520 wc->shared_level = -1;
5521 path->slots[level] = 0;
5524 * check reference count again if the block isn't locked.
5525 * we should start walking down the tree again if reference
5528 if (!path->locks[level]) {
5530 btrfs_tree_lock(eb);
5531 path->locks[level] = BTRFS_WRITE_LOCK;
5533 ret = btrfs_lookup_extent_info(trans, fs_info,
5534 eb->start, level, 1,
5538 btrfs_tree_unlock_rw(eb, path->locks[level]);
5539 path->locks[level] = 0;
5542 BUG_ON(wc->refs[level] == 0);
5543 if (wc->refs[level] == 1) {
5544 btrfs_tree_unlock_rw(eb, path->locks[level]);
5545 path->locks[level] = 0;
5551 /* wc->stage == DROP_REFERENCE */
5552 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5554 if (wc->refs[level] == 1) {
5556 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5557 ret = btrfs_dec_ref(trans, root, eb, 1);
5559 ret = btrfs_dec_ref(trans, root, eb, 0);
5560 BUG_ON(ret); /* -ENOMEM */
5561 if (is_fstree(root->root_key.objectid)) {
5562 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5564 btrfs_err_rl(fs_info,
5565 "error %d accounting leaf items, quota is out of sync, rescan required",
5570 /* make block locked assertion in btrfs_clean_tree_block happy */
5571 if (!path->locks[level] &&
5572 btrfs_header_generation(eb) == trans->transid) {
5573 btrfs_tree_lock(eb);
5574 path->locks[level] = BTRFS_WRITE_LOCK;
5576 btrfs_clean_tree_block(eb);
5579 if (eb == root->node) {
5580 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5582 else if (root->root_key.objectid != btrfs_header_owner(eb))
5583 goto owner_mismatch;
5585 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5586 parent = path->nodes[level + 1]->start;
5587 else if (root->root_key.objectid !=
5588 btrfs_header_owner(path->nodes[level + 1]))
5589 goto owner_mismatch;
5592 btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent,
5593 wc->refs[level] == 1);
5595 wc->refs[level] = 0;
5596 wc->flags[level] = 0;
5600 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5601 btrfs_header_owner(eb), root->root_key.objectid);
5605 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5606 struct btrfs_root *root,
5607 struct btrfs_path *path,
5608 struct walk_control *wc)
5610 int level = wc->level;
5611 int lookup_info = 1;
5614 while (level >= 0) {
5615 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5622 if (path->slots[level] >=
5623 btrfs_header_nritems(path->nodes[level]))
5626 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5628 path->slots[level]++;
5637 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5638 struct btrfs_root *root,
5639 struct btrfs_path *path,
5640 struct walk_control *wc, int max_level)
5642 int level = wc->level;
5645 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5646 while (level < max_level && path->nodes[level]) {
5648 if (path->slots[level] + 1 <
5649 btrfs_header_nritems(path->nodes[level])) {
5650 path->slots[level]++;
5653 ret = walk_up_proc(trans, root, path, wc);
5659 if (path->locks[level]) {
5660 btrfs_tree_unlock_rw(path->nodes[level],
5661 path->locks[level]);
5662 path->locks[level] = 0;
5664 free_extent_buffer(path->nodes[level]);
5665 path->nodes[level] = NULL;
5673 * drop a subvolume tree.
5675 * this function traverses the tree freeing any blocks that only
5676 * referenced by the tree.
5678 * when a shared tree block is found. this function decreases its
5679 * reference count by one. if update_ref is true, this function
5680 * also make sure backrefs for the shared block and all lower level
5681 * blocks are properly updated.
5683 * If called with for_reloc == 0, may exit early with -EAGAIN
5685 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5687 const bool is_reloc_root = (root->root_key.objectid ==
5688 BTRFS_TREE_RELOC_OBJECTID);
5689 struct btrfs_fs_info *fs_info = root->fs_info;
5690 struct btrfs_path *path;
5691 struct btrfs_trans_handle *trans;
5692 struct btrfs_root *tree_root = fs_info->tree_root;
5693 struct btrfs_root_item *root_item = &root->root_item;
5694 struct walk_control *wc;
5695 struct btrfs_key key;
5699 bool root_dropped = false;
5700 bool unfinished_drop = false;
5702 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5704 path = btrfs_alloc_path();
5710 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5712 btrfs_free_path(path);
5718 * Use join to avoid potential EINTR from transaction start. See
5719 * wait_reserve_ticket and the whole reservation callchain.
5722 trans = btrfs_join_transaction(tree_root);
5724 trans = btrfs_start_transaction(tree_root, 0);
5725 if (IS_ERR(trans)) {
5726 err = PTR_ERR(trans);
5730 err = btrfs_run_delayed_items(trans);
5735 * This will help us catch people modifying the fs tree while we're
5736 * dropping it. It is unsafe to mess with the fs tree while it's being
5737 * dropped as we unlock the root node and parent nodes as we walk down
5738 * the tree, assuming nothing will change. If something does change
5739 * then we'll have stale information and drop references to blocks we've
5742 set_bit(BTRFS_ROOT_DELETING, &root->state);
5743 unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
5745 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5746 level = btrfs_header_level(root->node);
5747 path->nodes[level] = btrfs_lock_root_node(root);
5748 path->slots[level] = 0;
5749 path->locks[level] = BTRFS_WRITE_LOCK;
5750 memset(&wc->update_progress, 0,
5751 sizeof(wc->update_progress));
5753 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5754 memcpy(&wc->update_progress, &key,
5755 sizeof(wc->update_progress));
5757 level = btrfs_root_drop_level(root_item);
5759 path->lowest_level = level;
5760 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5761 path->lowest_level = 0;
5769 * unlock our path, this is safe because only this
5770 * function is allowed to delete this snapshot
5772 btrfs_unlock_up_safe(path, 0);
5774 level = btrfs_header_level(root->node);
5776 btrfs_tree_lock(path->nodes[level]);
5777 path->locks[level] = BTRFS_WRITE_LOCK;
5779 ret = btrfs_lookup_extent_info(trans, fs_info,
5780 path->nodes[level]->start,
5781 level, 1, &wc->refs[level],
5787 BUG_ON(wc->refs[level] == 0);
5789 if (level == btrfs_root_drop_level(root_item))
5792 btrfs_tree_unlock(path->nodes[level]);
5793 path->locks[level] = 0;
5794 WARN_ON(wc->refs[level] != 1);
5799 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5801 wc->shared_level = -1;
5802 wc->stage = DROP_REFERENCE;
5803 wc->update_ref = update_ref;
5805 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5809 ret = walk_down_tree(trans, root, path, wc);
5815 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5822 BUG_ON(wc->stage != DROP_REFERENCE);
5826 if (wc->stage == DROP_REFERENCE) {
5827 wc->drop_level = wc->level;
5828 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5830 path->slots[wc->drop_level]);
5832 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5833 &wc->drop_progress);
5834 btrfs_set_root_drop_level(root_item, wc->drop_level);
5836 BUG_ON(wc->level == 0);
5837 if (btrfs_should_end_transaction(trans) ||
5838 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5839 ret = btrfs_update_root(trans, tree_root,
5843 btrfs_abort_transaction(trans, ret);
5849 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
5851 btrfs_end_transaction_throttle(trans);
5852 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5853 btrfs_debug(fs_info,
5854 "drop snapshot early exit");
5860 * Use join to avoid potential EINTR from transaction
5861 * start. See wait_reserve_ticket and the whole
5862 * reservation callchain.
5865 trans = btrfs_join_transaction(tree_root);
5867 trans = btrfs_start_transaction(tree_root, 0);
5868 if (IS_ERR(trans)) {
5869 err = PTR_ERR(trans);
5874 btrfs_release_path(path);
5878 ret = btrfs_del_root(trans, &root->root_key);
5880 btrfs_abort_transaction(trans, ret);
5885 if (!is_reloc_root) {
5886 ret = btrfs_find_root(tree_root, &root->root_key, path,
5889 btrfs_abort_transaction(trans, ret);
5892 } else if (ret > 0) {
5893 /* if we fail to delete the orphan item this time
5894 * around, it'll get picked up the next time.
5896 * The most common failure here is just -ENOENT.
5898 btrfs_del_orphan_item(trans, tree_root,
5899 root->root_key.objectid);
5904 * This subvolume is going to be completely dropped, and won't be
5905 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5906 * commit transaction time. So free it here manually.
5908 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
5909 btrfs_qgroup_free_meta_all_pertrans(root);
5911 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
5912 btrfs_add_dropped_root(trans, root);
5914 btrfs_put_root(root);
5915 root_dropped = true;
5918 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
5920 btrfs_end_transaction_throttle(trans);
5923 btrfs_free_path(path);
5926 * We were an unfinished drop root, check to see if there are any
5927 * pending, and if not clear and wake up any waiters.
5929 if (!err && unfinished_drop)
5930 btrfs_maybe_wake_unfinished_drop(fs_info);
5933 * So if we need to stop dropping the snapshot for whatever reason we
5934 * need to make sure to add it back to the dead root list so that we
5935 * keep trying to do the work later. This also cleans up roots if we
5936 * don't have it in the radix (like when we recover after a power fail
5937 * or unmount) so we don't leak memory.
5939 if (!for_reloc && !root_dropped)
5940 btrfs_add_dead_root(root);
5945 * drop subtree rooted at tree block 'node'.
5947 * NOTE: this function will unlock and release tree block 'node'
5948 * only used by relocation code
5950 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5951 struct btrfs_root *root,
5952 struct extent_buffer *node,
5953 struct extent_buffer *parent)
5955 struct btrfs_fs_info *fs_info = root->fs_info;
5956 struct btrfs_path *path;
5957 struct walk_control *wc;
5963 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5965 path = btrfs_alloc_path();
5969 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5971 btrfs_free_path(path);
5975 btrfs_assert_tree_write_locked(parent);
5976 parent_level = btrfs_header_level(parent);
5977 atomic_inc(&parent->refs);
5978 path->nodes[parent_level] = parent;
5979 path->slots[parent_level] = btrfs_header_nritems(parent);
5981 btrfs_assert_tree_write_locked(node);
5982 level = btrfs_header_level(node);
5983 path->nodes[level] = node;
5984 path->slots[level] = 0;
5985 path->locks[level] = BTRFS_WRITE_LOCK;
5987 wc->refs[parent_level] = 1;
5988 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5990 wc->shared_level = -1;
5991 wc->stage = DROP_REFERENCE;
5994 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5997 wret = walk_down_tree(trans, root, path, wc);
6003 wret = walk_up_tree(trans, root, path, wc, parent_level);
6011 btrfs_free_path(path);
6016 * helper to account the unused space of all the readonly block group in the
6017 * space_info. takes mirrors into account.
6019 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
6021 struct btrfs_block_group *block_group;
6025 /* It's df, we don't care if it's racy */
6026 if (list_empty(&sinfo->ro_bgs))
6029 spin_lock(&sinfo->lock);
6030 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
6031 spin_lock(&block_group->lock);
6033 if (!block_group->ro) {
6034 spin_unlock(&block_group->lock);
6038 factor = btrfs_bg_type_to_factor(block_group->flags);
6039 free_bytes += (block_group->length -
6040 block_group->used) * factor;
6042 spin_unlock(&block_group->lock);
6044 spin_unlock(&sinfo->lock);
6049 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
6052 return unpin_extent_range(fs_info, start, end, false);
6056 * It used to be that old block groups would be left around forever.
6057 * Iterating over them would be enough to trim unused space. Since we
6058 * now automatically remove them, we also need to iterate over unallocated
6061 * We don't want a transaction for this since the discard may take a
6062 * substantial amount of time. We don't require that a transaction be
6063 * running, but we do need to take a running transaction into account
6064 * to ensure that we're not discarding chunks that were released or
6065 * allocated in the current transaction.
6067 * Holding the chunks lock will prevent other threads from allocating
6068 * or releasing chunks, but it won't prevent a running transaction
6069 * from committing and releasing the memory that the pending chunks
6070 * list head uses. For that, we need to take a reference to the
6071 * transaction and hold the commit root sem. We only need to hold
6072 * it while performing the free space search since we have already
6073 * held back allocations.
6075 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
6077 u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0;
6082 /* Discard not supported = nothing to do. */
6083 if (!bdev_max_discard_sectors(device->bdev))
6086 /* Not writable = nothing to do. */
6087 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
6090 /* No free space = nothing to do. */
6091 if (device->total_bytes <= device->bytes_used)
6097 struct btrfs_fs_info *fs_info = device->fs_info;
6100 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
6104 find_first_clear_extent_bit(&device->alloc_state, start,
6106 CHUNK_TRIMMED | CHUNK_ALLOCATED);
6108 /* Check if there are any CHUNK_* bits left */
6109 if (start > device->total_bytes) {
6110 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6111 btrfs_warn_in_rcu(fs_info,
6112 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
6113 start, end - start + 1,
6114 rcu_str_deref(device->name),
6115 device->total_bytes);
6116 mutex_unlock(&fs_info->chunk_mutex);
6121 /* Ensure we skip the reserved space on each device. */
6122 start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
6125 * If find_first_clear_extent_bit find a range that spans the
6126 * end of the device it will set end to -1, in this case it's up
6127 * to the caller to trim the value to the size of the device.
6129 end = min(end, device->total_bytes - 1);
6131 len = end - start + 1;
6133 /* We didn't find any extents */
6135 mutex_unlock(&fs_info->chunk_mutex);
6140 ret = btrfs_issue_discard(device->bdev, start, len,
6143 set_extent_bits(&device->alloc_state, start,
6146 mutex_unlock(&fs_info->chunk_mutex);
6154 if (fatal_signal_pending(current)) {
6166 * Trim the whole filesystem by:
6167 * 1) trimming the free space in each block group
6168 * 2) trimming the unallocated space on each device
6170 * This will also continue trimming even if a block group or device encounters
6171 * an error. The return value will be the last error, or 0 if nothing bad
6174 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6176 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6177 struct btrfs_block_group *cache = NULL;
6178 struct btrfs_device *device;
6180 u64 range_end = U64_MAX;
6190 if (range->start == U64_MAX)
6194 * Check range overflow if range->len is set.
6195 * The default range->len is U64_MAX.
6197 if (range->len != U64_MAX &&
6198 check_add_overflow(range->start, range->len, &range_end))
6201 cache = btrfs_lookup_first_block_group(fs_info, range->start);
6202 for (; cache; cache = btrfs_next_block_group(cache)) {
6203 if (cache->start >= range_end) {
6204 btrfs_put_block_group(cache);
6208 start = max(range->start, cache->start);
6209 end = min(range_end, cache->start + cache->length);
6211 if (end - start >= range->minlen) {
6212 if (!btrfs_block_group_done(cache)) {
6213 ret = btrfs_cache_block_group(cache, true);
6220 ret = btrfs_trim_block_group(cache,
6226 trimmed += group_trimmed;
6237 "failed to trim %llu block group(s), last error %d",
6240 mutex_lock(&fs_devices->device_list_mutex);
6241 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6242 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6245 ret = btrfs_trim_free_extents(device, &group_trimmed);
6252 trimmed += group_trimmed;
6254 mutex_unlock(&fs_devices->device_list_mutex);
6258 "failed to trim %llu device(s), last error %d",
6259 dev_failed, dev_ret);
6260 range->len = trimmed;