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)) {
870 leaf = path->nodes[0];
871 item_size = btrfs_item_size(leaf, path->slots[0]);
872 if (unlikely(item_size < sizeof(*ei))) {
874 btrfs_print_v0_err(fs_info);
875 btrfs_abort_transaction(trans, err);
879 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
880 flags = btrfs_extent_flags(leaf, ei);
882 ptr = (unsigned long)(ei + 1);
883 end = (unsigned long)ei + item_size;
885 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
886 ptr += sizeof(struct btrfs_tree_block_info);
890 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
891 needed = BTRFS_REF_TYPE_DATA;
893 needed = BTRFS_REF_TYPE_BLOCK;
900 btrfs_print_leaf(path->nodes[0]);
902 "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu",
903 path->slots[0], root_objectid, owner, offset, parent);
907 iref = (struct btrfs_extent_inline_ref *)ptr;
908 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
909 if (type == BTRFS_REF_TYPE_INVALID) {
917 ptr += btrfs_extent_inline_ref_size(type);
921 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
922 struct btrfs_extent_data_ref *dref;
923 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
924 if (match_extent_data_ref(leaf, dref, root_objectid,
929 if (hash_extent_data_ref_item(leaf, dref) <
930 hash_extent_data_ref(root_objectid, owner, offset))
934 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
936 if (parent == ref_offset) {
940 if (ref_offset < parent)
943 if (root_objectid == ref_offset) {
947 if (ref_offset < root_objectid)
951 ptr += btrfs_extent_inline_ref_size(type);
953 if (err == -ENOENT && insert) {
954 if (item_size + extra_size >=
955 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
960 * To add new inline back ref, we have to make sure
961 * there is no corresponding back ref item.
962 * For simplicity, we just do not add new inline back
963 * ref if there is any kind of item for this block
965 if (find_next_key(path, 0, &key) == 0 &&
966 key.objectid == bytenr &&
967 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
972 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
975 path->keep_locks = 0;
976 path->search_for_extension = 0;
977 btrfs_unlock_up_safe(path, 1);
983 * helper to add new inline back ref
985 static noinline_for_stack
986 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
987 struct btrfs_path *path,
988 struct btrfs_extent_inline_ref *iref,
989 u64 parent, u64 root_objectid,
990 u64 owner, u64 offset, int refs_to_add,
991 struct btrfs_delayed_extent_op *extent_op)
993 struct extent_buffer *leaf;
994 struct btrfs_extent_item *ei;
997 unsigned long item_offset;
1002 leaf = path->nodes[0];
1003 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1004 item_offset = (unsigned long)iref - (unsigned long)ei;
1006 type = extent_ref_type(parent, owner);
1007 size = btrfs_extent_inline_ref_size(type);
1009 btrfs_extend_item(path, size);
1011 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1012 refs = btrfs_extent_refs(leaf, ei);
1013 refs += refs_to_add;
1014 btrfs_set_extent_refs(leaf, ei, refs);
1016 __run_delayed_extent_op(extent_op, leaf, ei);
1018 ptr = (unsigned long)ei + item_offset;
1019 end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]);
1020 if (ptr < end - size)
1021 memmove_extent_buffer(leaf, ptr + size, ptr,
1024 iref = (struct btrfs_extent_inline_ref *)ptr;
1025 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1026 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1027 struct btrfs_extent_data_ref *dref;
1028 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1029 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1030 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1031 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1032 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1033 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1034 struct btrfs_shared_data_ref *sref;
1035 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1036 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1037 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1038 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1039 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1041 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1043 btrfs_mark_buffer_dirty(leaf);
1046 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1047 struct btrfs_path *path,
1048 struct btrfs_extent_inline_ref **ref_ret,
1049 u64 bytenr, u64 num_bytes, u64 parent,
1050 u64 root_objectid, u64 owner, u64 offset)
1054 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1055 num_bytes, parent, root_objectid,
1060 btrfs_release_path(path);
1063 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1064 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1067 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1068 root_objectid, owner, offset);
1074 * helper to update/remove inline back ref
1076 static noinline_for_stack
1077 void update_inline_extent_backref(struct btrfs_path *path,
1078 struct btrfs_extent_inline_ref *iref,
1080 struct btrfs_delayed_extent_op *extent_op)
1082 struct extent_buffer *leaf = path->nodes[0];
1083 struct btrfs_extent_item *ei;
1084 struct btrfs_extent_data_ref *dref = NULL;
1085 struct btrfs_shared_data_ref *sref = NULL;
1093 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1094 refs = btrfs_extent_refs(leaf, ei);
1095 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1096 refs += refs_to_mod;
1097 btrfs_set_extent_refs(leaf, ei, refs);
1099 __run_delayed_extent_op(extent_op, leaf, ei);
1102 * If type is invalid, we should have bailed out after
1103 * lookup_inline_extent_backref().
1105 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1106 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1108 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1109 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1110 refs = btrfs_extent_data_ref_count(leaf, dref);
1111 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1112 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1113 refs = btrfs_shared_data_ref_count(leaf, sref);
1116 BUG_ON(refs_to_mod != -1);
1119 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1120 refs += refs_to_mod;
1123 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1124 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1126 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1128 size = btrfs_extent_inline_ref_size(type);
1129 item_size = btrfs_item_size(leaf, path->slots[0]);
1130 ptr = (unsigned long)iref;
1131 end = (unsigned long)ei + item_size;
1132 if (ptr + size < end)
1133 memmove_extent_buffer(leaf, ptr, ptr + size,
1136 btrfs_truncate_item(path, item_size, 1);
1138 btrfs_mark_buffer_dirty(leaf);
1141 static noinline_for_stack
1142 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1143 struct btrfs_path *path,
1144 u64 bytenr, u64 num_bytes, u64 parent,
1145 u64 root_objectid, u64 owner,
1146 u64 offset, int refs_to_add,
1147 struct btrfs_delayed_extent_op *extent_op)
1149 struct btrfs_extent_inline_ref *iref;
1152 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1153 num_bytes, parent, root_objectid,
1157 * We're adding refs to a tree block we already own, this
1158 * should not happen at all.
1160 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1161 btrfs_crit(trans->fs_info,
1162 "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu",
1163 bytenr, num_bytes, root_objectid);
1164 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) {
1166 btrfs_crit(trans->fs_info,
1167 "path->slots[0]=%d path->nodes[0]:", path->slots[0]);
1168 btrfs_print_leaf(path->nodes[0]);
1172 update_inline_extent_backref(path, iref, refs_to_add, extent_op);
1173 } else if (ret == -ENOENT) {
1174 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1175 root_objectid, owner, offset,
1176 refs_to_add, extent_op);
1182 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1183 struct btrfs_root *root,
1184 struct btrfs_path *path,
1185 struct btrfs_extent_inline_ref *iref,
1186 int refs_to_drop, int is_data)
1190 BUG_ON(!is_data && refs_to_drop != 1);
1192 update_inline_extent_backref(path, iref, -refs_to_drop, NULL);
1194 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1196 ret = btrfs_del_item(trans, root, path);
1200 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1201 u64 *discarded_bytes)
1204 u64 bytes_left, end;
1205 u64 aligned_start = ALIGN(start, 1 << 9);
1207 if (WARN_ON(start != aligned_start)) {
1208 len -= aligned_start - start;
1209 len = round_down(len, 1 << 9);
1210 start = aligned_start;
1213 *discarded_bytes = 0;
1221 /* Skip any superblocks on this device. */
1222 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1223 u64 sb_start = btrfs_sb_offset(j);
1224 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1225 u64 size = sb_start - start;
1227 if (!in_range(sb_start, start, bytes_left) &&
1228 !in_range(sb_end, start, bytes_left) &&
1229 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1233 * Superblock spans beginning of range. Adjust start and
1236 if (sb_start <= start) {
1237 start += sb_end - start;
1242 bytes_left = end - start;
1247 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1250 *discarded_bytes += size;
1251 else if (ret != -EOPNOTSUPP)
1260 bytes_left = end - start;
1264 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1267 *discarded_bytes += bytes_left;
1272 static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes)
1274 struct btrfs_device *dev = stripe->dev;
1275 struct btrfs_fs_info *fs_info = dev->fs_info;
1276 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1277 u64 phys = stripe->physical;
1278 u64 len = stripe->length;
1282 /* Zone reset on a zoned filesystem */
1283 if (btrfs_can_zone_reset(dev, phys, len)) {
1286 ret = btrfs_reset_device_zone(dev, phys, len, &discarded);
1290 if (!btrfs_dev_replace_is_ongoing(dev_replace) ||
1291 dev != dev_replace->srcdev)
1294 src_disc = discarded;
1296 /* Send to replace target as well */
1297 ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len,
1299 discarded += src_disc;
1300 } else if (bdev_max_discard_sectors(stripe->dev->bdev)) {
1301 ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded);
1312 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1313 u64 num_bytes, u64 *actual_bytes)
1316 u64 discarded_bytes = 0;
1317 u64 end = bytenr + num_bytes;
1321 * Avoid races with device replace and make sure the devices in the
1322 * stripes don't go away while we are discarding.
1324 btrfs_bio_counter_inc_blocked(fs_info);
1326 struct btrfs_discard_stripe *stripes;
1327 unsigned int num_stripes;
1330 num_bytes = end - cur;
1331 stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes);
1332 if (IS_ERR(stripes)) {
1333 ret = PTR_ERR(stripes);
1334 if (ret == -EOPNOTSUPP)
1339 for (i = 0; i < num_stripes; i++) {
1340 struct btrfs_discard_stripe *stripe = stripes + i;
1343 if (!stripe->dev->bdev) {
1344 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1348 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
1349 &stripe->dev->dev_state))
1352 ret = do_discard_extent(stripe, &bytes);
1355 * Keep going if discard is not supported by the
1358 if (ret != -EOPNOTSUPP)
1362 discarded_bytes += bytes;
1370 btrfs_bio_counter_dec(fs_info);
1372 *actual_bytes = discarded_bytes;
1376 /* Can return -ENOMEM */
1377 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1378 struct btrfs_ref *generic_ref)
1380 struct btrfs_fs_info *fs_info = trans->fs_info;
1383 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1384 generic_ref->action);
1385 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1386 generic_ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID);
1388 if (generic_ref->type == BTRFS_REF_METADATA)
1389 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1391 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1393 btrfs_ref_tree_mod(fs_info, generic_ref);
1399 * __btrfs_inc_extent_ref - insert backreference for a given extent
1401 * The counterpart is in __btrfs_free_extent(), with examples and more details
1404 * @trans: Handle of transaction
1406 * @node: The delayed ref node used to get the bytenr/length for
1407 * extent whose references are incremented.
1409 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1410 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1411 * bytenr of the parent block. Since new extents are always
1412 * created with indirect references, this will only be the case
1413 * when relocating a shared extent. In that case, root_objectid
1414 * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1417 * @root_objectid: The id of the root where this modification has originated,
1418 * this can be either one of the well-known metadata trees or
1419 * the subvolume id which references this extent.
1421 * @owner: For data extents it is the inode number of the owning file.
1422 * For metadata extents this parameter holds the level in the
1423 * tree of the extent.
1425 * @offset: For metadata extents the offset is ignored and is currently
1426 * always passed as 0. For data extents it is the fileoffset
1427 * this extent belongs to.
1429 * @refs_to_add Number of references to add
1431 * @extent_op Pointer to a structure, holding information necessary when
1432 * updating a tree block's flags
1435 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1436 struct btrfs_delayed_ref_node *node,
1437 u64 parent, u64 root_objectid,
1438 u64 owner, u64 offset, int refs_to_add,
1439 struct btrfs_delayed_extent_op *extent_op)
1441 struct btrfs_path *path;
1442 struct extent_buffer *leaf;
1443 struct btrfs_extent_item *item;
1444 struct btrfs_key key;
1445 u64 bytenr = node->bytenr;
1446 u64 num_bytes = node->num_bytes;
1450 path = btrfs_alloc_path();
1454 /* this will setup the path even if it fails to insert the back ref */
1455 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1456 parent, root_objectid, owner,
1457 offset, refs_to_add, extent_op);
1458 if ((ret < 0 && ret != -EAGAIN) || !ret)
1462 * Ok we had -EAGAIN which means we didn't have space to insert and
1463 * inline extent ref, so just update the reference count and add a
1466 leaf = path->nodes[0];
1467 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1468 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1469 refs = btrfs_extent_refs(leaf, item);
1470 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1472 __run_delayed_extent_op(extent_op, leaf, item);
1474 btrfs_mark_buffer_dirty(leaf);
1475 btrfs_release_path(path);
1477 /* now insert the actual backref */
1478 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1479 BUG_ON(refs_to_add != 1);
1480 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1483 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1484 root_objectid, owner, offset,
1488 btrfs_abort_transaction(trans, ret);
1490 btrfs_free_path(path);
1494 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1495 struct btrfs_delayed_ref_node *node,
1496 struct btrfs_delayed_extent_op *extent_op,
1497 int insert_reserved)
1500 struct btrfs_delayed_data_ref *ref;
1501 struct btrfs_key ins;
1506 ins.objectid = node->bytenr;
1507 ins.offset = node->num_bytes;
1508 ins.type = BTRFS_EXTENT_ITEM_KEY;
1510 ref = btrfs_delayed_node_to_data_ref(node);
1511 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1513 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1514 parent = ref->parent;
1515 ref_root = ref->root;
1517 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1519 flags |= extent_op->flags_to_set;
1520 ret = alloc_reserved_file_extent(trans, parent, ref_root,
1521 flags, ref->objectid,
1524 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1525 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1526 ref->objectid, ref->offset,
1527 node->ref_mod, extent_op);
1528 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1529 ret = __btrfs_free_extent(trans, node, parent,
1530 ref_root, ref->objectid,
1531 ref->offset, node->ref_mod,
1539 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1540 struct extent_buffer *leaf,
1541 struct btrfs_extent_item *ei)
1543 u64 flags = btrfs_extent_flags(leaf, ei);
1544 if (extent_op->update_flags) {
1545 flags |= extent_op->flags_to_set;
1546 btrfs_set_extent_flags(leaf, ei, flags);
1549 if (extent_op->update_key) {
1550 struct btrfs_tree_block_info *bi;
1551 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1552 bi = (struct btrfs_tree_block_info *)(ei + 1);
1553 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1557 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1558 struct btrfs_delayed_ref_head *head,
1559 struct btrfs_delayed_extent_op *extent_op)
1561 struct btrfs_fs_info *fs_info = trans->fs_info;
1562 struct btrfs_root *root;
1563 struct btrfs_key key;
1564 struct btrfs_path *path;
1565 struct btrfs_extent_item *ei;
1566 struct extent_buffer *leaf;
1572 if (TRANS_ABORTED(trans))
1575 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1578 path = btrfs_alloc_path();
1582 key.objectid = head->bytenr;
1585 key.type = BTRFS_METADATA_ITEM_KEY;
1586 key.offset = extent_op->level;
1588 key.type = BTRFS_EXTENT_ITEM_KEY;
1589 key.offset = head->num_bytes;
1592 root = btrfs_extent_root(fs_info, key.objectid);
1594 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1601 if (path->slots[0] > 0) {
1603 btrfs_item_key_to_cpu(path->nodes[0], &key,
1605 if (key.objectid == head->bytenr &&
1606 key.type == BTRFS_EXTENT_ITEM_KEY &&
1607 key.offset == head->num_bytes)
1611 btrfs_release_path(path);
1614 key.objectid = head->bytenr;
1615 key.offset = head->num_bytes;
1616 key.type = BTRFS_EXTENT_ITEM_KEY;
1625 leaf = path->nodes[0];
1626 item_size = btrfs_item_size(leaf, path->slots[0]);
1628 if (unlikely(item_size < sizeof(*ei))) {
1630 btrfs_print_v0_err(fs_info);
1631 btrfs_abort_transaction(trans, err);
1635 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1636 __run_delayed_extent_op(extent_op, leaf, ei);
1638 btrfs_mark_buffer_dirty(leaf);
1640 btrfs_free_path(path);
1644 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1645 struct btrfs_delayed_ref_node *node,
1646 struct btrfs_delayed_extent_op *extent_op,
1647 int insert_reserved)
1650 struct btrfs_delayed_tree_ref *ref;
1654 ref = btrfs_delayed_node_to_tree_ref(node);
1655 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1657 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1658 parent = ref->parent;
1659 ref_root = ref->root;
1661 if (node->ref_mod != 1) {
1662 btrfs_err(trans->fs_info,
1663 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
1664 node->bytenr, node->ref_mod, node->action, ref_root,
1668 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1669 BUG_ON(!extent_op || !extent_op->update_flags);
1670 ret = alloc_reserved_tree_block(trans, node, extent_op);
1671 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1672 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1673 ref->level, 0, 1, extent_op);
1674 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1675 ret = __btrfs_free_extent(trans, node, parent, ref_root,
1676 ref->level, 0, 1, extent_op);
1683 /* helper function to actually process a single delayed ref entry */
1684 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1685 struct btrfs_delayed_ref_node *node,
1686 struct btrfs_delayed_extent_op *extent_op,
1687 int insert_reserved)
1691 if (TRANS_ABORTED(trans)) {
1692 if (insert_reserved)
1693 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1697 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1698 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1699 ret = run_delayed_tree_ref(trans, node, extent_op,
1701 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1702 node->type == BTRFS_SHARED_DATA_REF_KEY)
1703 ret = run_delayed_data_ref(trans, node, extent_op,
1707 if (ret && insert_reserved)
1708 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1712 static inline struct btrfs_delayed_ref_node *
1713 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1715 struct btrfs_delayed_ref_node *ref;
1717 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1721 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1722 * This is to prevent a ref count from going down to zero, which deletes
1723 * the extent item from the extent tree, when there still are references
1724 * to add, which would fail because they would not find the extent item.
1726 if (!list_empty(&head->ref_add_list))
1727 return list_first_entry(&head->ref_add_list,
1728 struct btrfs_delayed_ref_node, add_list);
1730 ref = rb_entry(rb_first_cached(&head->ref_tree),
1731 struct btrfs_delayed_ref_node, ref_node);
1732 ASSERT(list_empty(&ref->add_list));
1736 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1737 struct btrfs_delayed_ref_head *head)
1739 spin_lock(&delayed_refs->lock);
1740 head->processing = 0;
1741 delayed_refs->num_heads_ready++;
1742 spin_unlock(&delayed_refs->lock);
1743 btrfs_delayed_ref_unlock(head);
1746 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1747 struct btrfs_delayed_ref_head *head)
1749 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1754 if (head->must_insert_reserved) {
1755 head->extent_op = NULL;
1756 btrfs_free_delayed_extent_op(extent_op);
1762 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1763 struct btrfs_delayed_ref_head *head)
1765 struct btrfs_delayed_extent_op *extent_op;
1768 extent_op = cleanup_extent_op(head);
1771 head->extent_op = NULL;
1772 spin_unlock(&head->lock);
1773 ret = run_delayed_extent_op(trans, head, extent_op);
1774 btrfs_free_delayed_extent_op(extent_op);
1775 return ret ? ret : 1;
1778 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1779 struct btrfs_delayed_ref_root *delayed_refs,
1780 struct btrfs_delayed_ref_head *head)
1782 int nr_items = 1; /* Dropping this ref head update. */
1785 * We had csum deletions accounted for in our delayed refs rsv, we need
1786 * to drop the csum leaves for this update from our delayed_refs_rsv.
1788 if (head->total_ref_mod < 0 && head->is_data) {
1789 spin_lock(&delayed_refs->lock);
1790 delayed_refs->pending_csums -= head->num_bytes;
1791 spin_unlock(&delayed_refs->lock);
1792 nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1795 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1798 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1799 struct btrfs_delayed_ref_head *head)
1802 struct btrfs_fs_info *fs_info = trans->fs_info;
1803 struct btrfs_delayed_ref_root *delayed_refs;
1806 delayed_refs = &trans->transaction->delayed_refs;
1808 ret = run_and_cleanup_extent_op(trans, head);
1810 unselect_delayed_ref_head(delayed_refs, head);
1811 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1818 * Need to drop our head ref lock and re-acquire the delayed ref lock
1819 * and then re-check to make sure nobody got added.
1821 spin_unlock(&head->lock);
1822 spin_lock(&delayed_refs->lock);
1823 spin_lock(&head->lock);
1824 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1825 spin_unlock(&head->lock);
1826 spin_unlock(&delayed_refs->lock);
1829 btrfs_delete_ref_head(delayed_refs, head);
1830 spin_unlock(&head->lock);
1831 spin_unlock(&delayed_refs->lock);
1833 if (head->must_insert_reserved) {
1834 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1835 if (head->is_data) {
1836 struct btrfs_root *csum_root;
1838 csum_root = btrfs_csum_root(fs_info, head->bytenr);
1839 ret = btrfs_del_csums(trans, csum_root, head->bytenr,
1844 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1846 trace_run_delayed_ref_head(fs_info, head, 0);
1847 btrfs_delayed_ref_unlock(head);
1848 btrfs_put_delayed_ref_head(head);
1852 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1853 struct btrfs_trans_handle *trans)
1855 struct btrfs_delayed_ref_root *delayed_refs =
1856 &trans->transaction->delayed_refs;
1857 struct btrfs_delayed_ref_head *head = NULL;
1860 spin_lock(&delayed_refs->lock);
1861 head = btrfs_select_ref_head(delayed_refs);
1863 spin_unlock(&delayed_refs->lock);
1868 * Grab the lock that says we are going to process all the refs for
1871 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1872 spin_unlock(&delayed_refs->lock);
1875 * We may have dropped the spin lock to get the head mutex lock, and
1876 * that might have given someone else time to free the head. If that's
1877 * true, it has been removed from our list and we can move on.
1880 head = ERR_PTR(-EAGAIN);
1885 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1886 struct btrfs_delayed_ref_head *locked_ref,
1887 unsigned long *run_refs)
1889 struct btrfs_fs_info *fs_info = trans->fs_info;
1890 struct btrfs_delayed_ref_root *delayed_refs;
1891 struct btrfs_delayed_extent_op *extent_op;
1892 struct btrfs_delayed_ref_node *ref;
1893 int must_insert_reserved = 0;
1896 delayed_refs = &trans->transaction->delayed_refs;
1898 lockdep_assert_held(&locked_ref->mutex);
1899 lockdep_assert_held(&locked_ref->lock);
1901 while ((ref = select_delayed_ref(locked_ref))) {
1903 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1904 spin_unlock(&locked_ref->lock);
1905 unselect_delayed_ref_head(delayed_refs, locked_ref);
1911 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1912 RB_CLEAR_NODE(&ref->ref_node);
1913 if (!list_empty(&ref->add_list))
1914 list_del(&ref->add_list);
1916 * When we play the delayed ref, also correct the ref_mod on
1919 switch (ref->action) {
1920 case BTRFS_ADD_DELAYED_REF:
1921 case BTRFS_ADD_DELAYED_EXTENT:
1922 locked_ref->ref_mod -= ref->ref_mod;
1924 case BTRFS_DROP_DELAYED_REF:
1925 locked_ref->ref_mod += ref->ref_mod;
1930 atomic_dec(&delayed_refs->num_entries);
1933 * Record the must_insert_reserved flag before we drop the
1936 must_insert_reserved = locked_ref->must_insert_reserved;
1937 locked_ref->must_insert_reserved = 0;
1939 extent_op = locked_ref->extent_op;
1940 locked_ref->extent_op = NULL;
1941 spin_unlock(&locked_ref->lock);
1943 ret = run_one_delayed_ref(trans, ref, extent_op,
1944 must_insert_reserved);
1946 btrfs_free_delayed_extent_op(extent_op);
1948 unselect_delayed_ref_head(delayed_refs, locked_ref);
1949 btrfs_put_delayed_ref(ref);
1950 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
1955 btrfs_put_delayed_ref(ref);
1958 spin_lock(&locked_ref->lock);
1959 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
1966 * Returns 0 on success or if called with an already aborted transaction.
1967 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
1969 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1972 struct btrfs_fs_info *fs_info = trans->fs_info;
1973 struct btrfs_delayed_ref_root *delayed_refs;
1974 struct btrfs_delayed_ref_head *locked_ref = NULL;
1975 ktime_t start = ktime_get();
1977 unsigned long count = 0;
1978 unsigned long actual_count = 0;
1980 delayed_refs = &trans->transaction->delayed_refs;
1983 locked_ref = btrfs_obtain_ref_head(trans);
1984 if (IS_ERR_OR_NULL(locked_ref)) {
1985 if (PTR_ERR(locked_ref) == -EAGAIN) {
1994 * We need to try and merge add/drops of the same ref since we
1995 * can run into issues with relocate dropping the implicit ref
1996 * and then it being added back again before the drop can
1997 * finish. If we merged anything we need to re-loop so we can
1999 * Or we can get node references of the same type that weren't
2000 * merged when created due to bumps in the tree mod seq, and
2001 * we need to merge them to prevent adding an inline extent
2002 * backref before dropping it (triggering a BUG_ON at
2003 * insert_inline_extent_backref()).
2005 spin_lock(&locked_ref->lock);
2006 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2008 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2010 if (ret < 0 && ret != -EAGAIN) {
2012 * Error, btrfs_run_delayed_refs_for_head already
2013 * unlocked everything so just bail out
2018 * Success, perform the usual cleanup of a processed
2021 ret = cleanup_ref_head(trans, locked_ref);
2023 /* We dropped our lock, we need to loop. */
2032 * Either success case or btrfs_run_delayed_refs_for_head
2033 * returned -EAGAIN, meaning we need to select another head
2038 } while ((nr != -1 && count < nr) || locked_ref);
2041 * We don't want to include ref heads since we can have empty ref heads
2042 * and those will drastically skew our runtime down since we just do
2043 * accounting, no actual extent tree updates.
2045 if (actual_count > 0) {
2046 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2050 * We weigh the current average higher than our current runtime
2051 * to avoid large swings in the average.
2053 spin_lock(&delayed_refs->lock);
2054 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2055 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2056 spin_unlock(&delayed_refs->lock);
2061 #ifdef SCRAMBLE_DELAYED_REFS
2063 * Normally delayed refs get processed in ascending bytenr order. This
2064 * correlates in most cases to the order added. To expose dependencies on this
2065 * order, we start to process the tree in the middle instead of the beginning
2067 static u64 find_middle(struct rb_root *root)
2069 struct rb_node *n = root->rb_node;
2070 struct btrfs_delayed_ref_node *entry;
2073 u64 first = 0, last = 0;
2077 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2078 first = entry->bytenr;
2082 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2083 last = entry->bytenr;
2088 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2089 WARN_ON(!entry->in_tree);
2091 middle = entry->bytenr;
2105 * this starts processing the delayed reference count updates and
2106 * extent insertions we have queued up so far. count can be
2107 * 0, which means to process everything in the tree at the start
2108 * of the run (but not newly added entries), or it can be some target
2109 * number you'd like to process.
2111 * Returns 0 on success or if called with an aborted transaction
2112 * Returns <0 on error and aborts the transaction
2114 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2115 unsigned long count)
2117 struct btrfs_fs_info *fs_info = trans->fs_info;
2118 struct rb_node *node;
2119 struct btrfs_delayed_ref_root *delayed_refs;
2120 struct btrfs_delayed_ref_head *head;
2122 int run_all = count == (unsigned long)-1;
2124 /* We'll clean this up in btrfs_cleanup_transaction */
2125 if (TRANS_ABORTED(trans))
2128 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2131 delayed_refs = &trans->transaction->delayed_refs;
2133 count = delayed_refs->num_heads_ready;
2136 #ifdef SCRAMBLE_DELAYED_REFS
2137 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2139 ret = __btrfs_run_delayed_refs(trans, count);
2141 btrfs_abort_transaction(trans, ret);
2146 btrfs_create_pending_block_groups(trans);
2148 spin_lock(&delayed_refs->lock);
2149 node = rb_first_cached(&delayed_refs->href_root);
2151 spin_unlock(&delayed_refs->lock);
2154 head = rb_entry(node, struct btrfs_delayed_ref_head,
2156 refcount_inc(&head->refs);
2157 spin_unlock(&delayed_refs->lock);
2159 /* Mutex was contended, block until it's released and retry. */
2160 mutex_lock(&head->mutex);
2161 mutex_unlock(&head->mutex);
2163 btrfs_put_delayed_ref_head(head);
2171 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2172 struct extent_buffer *eb, u64 flags,
2175 struct btrfs_delayed_extent_op *extent_op;
2178 extent_op = btrfs_alloc_delayed_extent_op();
2182 extent_op->flags_to_set = flags;
2183 extent_op->update_flags = true;
2184 extent_op->update_key = false;
2185 extent_op->level = level;
2187 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2189 btrfs_free_delayed_extent_op(extent_op);
2193 static noinline int check_delayed_ref(struct btrfs_root *root,
2194 struct btrfs_path *path,
2195 u64 objectid, u64 offset, u64 bytenr)
2197 struct btrfs_delayed_ref_head *head;
2198 struct btrfs_delayed_ref_node *ref;
2199 struct btrfs_delayed_data_ref *data_ref;
2200 struct btrfs_delayed_ref_root *delayed_refs;
2201 struct btrfs_transaction *cur_trans;
2202 struct rb_node *node;
2205 spin_lock(&root->fs_info->trans_lock);
2206 cur_trans = root->fs_info->running_transaction;
2208 refcount_inc(&cur_trans->use_count);
2209 spin_unlock(&root->fs_info->trans_lock);
2213 delayed_refs = &cur_trans->delayed_refs;
2214 spin_lock(&delayed_refs->lock);
2215 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2217 spin_unlock(&delayed_refs->lock);
2218 btrfs_put_transaction(cur_trans);
2222 if (!mutex_trylock(&head->mutex)) {
2223 refcount_inc(&head->refs);
2224 spin_unlock(&delayed_refs->lock);
2226 btrfs_release_path(path);
2229 * Mutex was contended, block until it's released and let
2232 mutex_lock(&head->mutex);
2233 mutex_unlock(&head->mutex);
2234 btrfs_put_delayed_ref_head(head);
2235 btrfs_put_transaction(cur_trans);
2238 spin_unlock(&delayed_refs->lock);
2240 spin_lock(&head->lock);
2242 * XXX: We should replace this with a proper search function in the
2245 for (node = rb_first_cached(&head->ref_tree); node;
2246 node = rb_next(node)) {
2247 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2248 /* If it's a shared ref we know a cross reference exists */
2249 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2254 data_ref = btrfs_delayed_node_to_data_ref(ref);
2257 * If our ref doesn't match the one we're currently looking at
2258 * then we have a cross reference.
2260 if (data_ref->root != root->root_key.objectid ||
2261 data_ref->objectid != objectid ||
2262 data_ref->offset != offset) {
2267 spin_unlock(&head->lock);
2268 mutex_unlock(&head->mutex);
2269 btrfs_put_transaction(cur_trans);
2273 static noinline int check_committed_ref(struct btrfs_root *root,
2274 struct btrfs_path *path,
2275 u64 objectid, u64 offset, u64 bytenr,
2278 struct btrfs_fs_info *fs_info = root->fs_info;
2279 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2280 struct extent_buffer *leaf;
2281 struct btrfs_extent_data_ref *ref;
2282 struct btrfs_extent_inline_ref *iref;
2283 struct btrfs_extent_item *ei;
2284 struct btrfs_key key;
2289 key.objectid = bytenr;
2290 key.offset = (u64)-1;
2291 key.type = BTRFS_EXTENT_ITEM_KEY;
2293 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2296 BUG_ON(ret == 0); /* Corruption */
2299 if (path->slots[0] == 0)
2303 leaf = path->nodes[0];
2304 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2306 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2310 item_size = btrfs_item_size(leaf, path->slots[0]);
2311 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2313 /* If extent item has more than 1 inline ref then it's shared */
2314 if (item_size != sizeof(*ei) +
2315 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2319 * If extent created before last snapshot => it's shared unless the
2320 * snapshot has been deleted. Use the heuristic if strict is false.
2323 (btrfs_extent_generation(leaf, ei) <=
2324 btrfs_root_last_snapshot(&root->root_item)))
2327 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2329 /* If this extent has SHARED_DATA_REF then it's shared */
2330 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2331 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2334 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2335 if (btrfs_extent_refs(leaf, ei) !=
2336 btrfs_extent_data_ref_count(leaf, ref) ||
2337 btrfs_extent_data_ref_root(leaf, ref) !=
2338 root->root_key.objectid ||
2339 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2340 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2348 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2349 u64 bytenr, bool strict, struct btrfs_path *path)
2354 ret = check_committed_ref(root, path, objectid,
2355 offset, bytenr, strict);
2356 if (ret && ret != -ENOENT)
2359 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2360 } while (ret == -EAGAIN);
2363 btrfs_release_path(path);
2364 if (btrfs_is_data_reloc_root(root))
2369 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2370 struct btrfs_root *root,
2371 struct extent_buffer *buf,
2372 int full_backref, int inc)
2374 struct btrfs_fs_info *fs_info = root->fs_info;
2380 struct btrfs_key key;
2381 struct btrfs_file_extent_item *fi;
2382 struct btrfs_ref generic_ref = { 0 };
2383 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2389 if (btrfs_is_testing(fs_info))
2392 ref_root = btrfs_header_owner(buf);
2393 nritems = btrfs_header_nritems(buf);
2394 level = btrfs_header_level(buf);
2396 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2400 parent = buf->start;
2404 action = BTRFS_ADD_DELAYED_REF;
2406 action = BTRFS_DROP_DELAYED_REF;
2408 for (i = 0; i < nritems; i++) {
2410 btrfs_item_key_to_cpu(buf, &key, i);
2411 if (key.type != BTRFS_EXTENT_DATA_KEY)
2413 fi = btrfs_item_ptr(buf, i,
2414 struct btrfs_file_extent_item);
2415 if (btrfs_file_extent_type(buf, fi) ==
2416 BTRFS_FILE_EXTENT_INLINE)
2418 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2422 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2423 key.offset -= btrfs_file_extent_offset(buf, fi);
2424 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2426 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2427 key.offset, root->root_key.objectid,
2430 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2432 ret = btrfs_free_extent(trans, &generic_ref);
2436 bytenr = btrfs_node_blockptr(buf, i);
2437 num_bytes = fs_info->nodesize;
2438 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2440 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root,
2441 root->root_key.objectid, for_reloc);
2443 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2445 ret = btrfs_free_extent(trans, &generic_ref);
2455 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2456 struct extent_buffer *buf, int full_backref)
2458 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2461 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2462 struct extent_buffer *buf, int full_backref)
2464 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2467 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2469 struct btrfs_fs_info *fs_info = root->fs_info;
2474 flags = BTRFS_BLOCK_GROUP_DATA;
2475 else if (root == fs_info->chunk_root)
2476 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2478 flags = BTRFS_BLOCK_GROUP_METADATA;
2480 ret = btrfs_get_alloc_profile(fs_info, flags);
2484 static u64 first_logical_byte(struct btrfs_fs_info *fs_info)
2486 struct rb_node *leftmost;
2489 read_lock(&fs_info->block_group_cache_lock);
2490 /* Get the block group with the lowest logical start address. */
2491 leftmost = rb_first_cached(&fs_info->block_group_cache_tree);
2493 struct btrfs_block_group *bg;
2495 bg = rb_entry(leftmost, struct btrfs_block_group, cache_node);
2498 read_unlock(&fs_info->block_group_cache_lock);
2503 static int pin_down_extent(struct btrfs_trans_handle *trans,
2504 struct btrfs_block_group *cache,
2505 u64 bytenr, u64 num_bytes, int reserved)
2507 struct btrfs_fs_info *fs_info = cache->fs_info;
2509 spin_lock(&cache->space_info->lock);
2510 spin_lock(&cache->lock);
2511 cache->pinned += num_bytes;
2512 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2515 cache->reserved -= num_bytes;
2516 cache->space_info->bytes_reserved -= num_bytes;
2518 spin_unlock(&cache->lock);
2519 spin_unlock(&cache->space_info->lock);
2521 set_extent_dirty(&trans->transaction->pinned_extents, bytenr,
2522 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
2526 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2527 u64 bytenr, u64 num_bytes, int reserved)
2529 struct btrfs_block_group *cache;
2531 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2532 BUG_ON(!cache); /* Logic error */
2534 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2536 btrfs_put_block_group(cache);
2541 * this function must be called within transaction
2543 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2544 u64 bytenr, u64 num_bytes)
2546 struct btrfs_block_group *cache;
2549 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2554 * Fully cache the free space first so that our pin removes the free space
2557 ret = btrfs_cache_block_group(cache, true);
2561 pin_down_extent(trans, cache, bytenr, num_bytes, 0);
2563 /* remove us from the free space cache (if we're there at all) */
2564 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2566 btrfs_put_block_group(cache);
2570 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2571 u64 start, u64 num_bytes)
2574 struct btrfs_block_group *block_group;
2576 block_group = btrfs_lookup_block_group(fs_info, start);
2580 ret = btrfs_cache_block_group(block_group, true);
2584 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2586 btrfs_put_block_group(block_group);
2590 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2592 struct btrfs_fs_info *fs_info = eb->fs_info;
2593 struct btrfs_file_extent_item *item;
2594 struct btrfs_key key;
2599 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2602 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2603 btrfs_item_key_to_cpu(eb, &key, i);
2604 if (key.type != BTRFS_EXTENT_DATA_KEY)
2606 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2607 found_type = btrfs_file_extent_type(eb, item);
2608 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2610 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2612 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2613 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2614 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2623 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2625 atomic_inc(&bg->reservations);
2629 * Returns the free cluster for the given space info and sets empty_cluster to
2630 * what it should be based on the mount options.
2632 static struct btrfs_free_cluster *
2633 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2634 struct btrfs_space_info *space_info, u64 *empty_cluster)
2636 struct btrfs_free_cluster *ret = NULL;
2639 if (btrfs_mixed_space_info(space_info))
2642 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2643 ret = &fs_info->meta_alloc_cluster;
2644 if (btrfs_test_opt(fs_info, SSD))
2645 *empty_cluster = SZ_2M;
2647 *empty_cluster = SZ_64K;
2648 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2649 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2650 *empty_cluster = SZ_2M;
2651 ret = &fs_info->data_alloc_cluster;
2657 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2659 const bool return_free_space)
2661 struct btrfs_block_group *cache = NULL;
2662 struct btrfs_space_info *space_info;
2663 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2664 struct btrfs_free_cluster *cluster = NULL;
2666 u64 total_unpinned = 0;
2667 u64 empty_cluster = 0;
2670 while (start <= end) {
2673 start >= cache->start + cache->length) {
2675 btrfs_put_block_group(cache);
2677 cache = btrfs_lookup_block_group(fs_info, start);
2678 BUG_ON(!cache); /* Logic error */
2680 cluster = fetch_cluster_info(fs_info,
2683 empty_cluster <<= 1;
2686 len = cache->start + cache->length - start;
2687 len = min(len, end + 1 - start);
2689 down_read(&fs_info->commit_root_sem);
2690 if (start < cache->last_byte_to_unpin && return_free_space) {
2691 u64 add_len = min(len, cache->last_byte_to_unpin - start);
2693 btrfs_add_free_space(cache, start, add_len);
2695 up_read(&fs_info->commit_root_sem);
2698 total_unpinned += len;
2699 space_info = cache->space_info;
2702 * If this space cluster has been marked as fragmented and we've
2703 * unpinned enough in this block group to potentially allow a
2704 * cluster to be created inside of it go ahead and clear the
2707 if (cluster && cluster->fragmented &&
2708 total_unpinned > empty_cluster) {
2709 spin_lock(&cluster->lock);
2710 cluster->fragmented = 0;
2711 spin_unlock(&cluster->lock);
2714 spin_lock(&space_info->lock);
2715 spin_lock(&cache->lock);
2716 cache->pinned -= len;
2717 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2718 space_info->max_extent_size = 0;
2720 space_info->bytes_readonly += len;
2722 } else if (btrfs_is_zoned(fs_info)) {
2723 /* Need reset before reusing in a zoned block group */
2724 space_info->bytes_zone_unusable += len;
2727 spin_unlock(&cache->lock);
2728 if (!readonly && return_free_space &&
2729 global_rsv->space_info == space_info) {
2730 spin_lock(&global_rsv->lock);
2731 if (!global_rsv->full) {
2732 u64 to_add = min(len, global_rsv->size -
2733 global_rsv->reserved);
2735 global_rsv->reserved += to_add;
2736 btrfs_space_info_update_bytes_may_use(fs_info,
2737 space_info, to_add);
2738 if (global_rsv->reserved >= global_rsv->size)
2739 global_rsv->full = 1;
2742 spin_unlock(&global_rsv->lock);
2744 /* Add to any tickets we may have */
2745 if (!readonly && return_free_space && len)
2746 btrfs_try_granting_tickets(fs_info, space_info);
2747 spin_unlock(&space_info->lock);
2751 btrfs_put_block_group(cache);
2755 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2757 struct btrfs_fs_info *fs_info = trans->fs_info;
2758 struct btrfs_block_group *block_group, *tmp;
2759 struct list_head *deleted_bgs;
2760 struct extent_io_tree *unpin;
2765 unpin = &trans->transaction->pinned_extents;
2767 while (!TRANS_ABORTED(trans)) {
2768 struct extent_state *cached_state = NULL;
2770 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2771 ret = find_first_extent_bit(unpin, 0, &start, &end,
2772 EXTENT_DIRTY, &cached_state);
2774 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2778 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2779 ret = btrfs_discard_extent(fs_info, start,
2780 end + 1 - start, NULL);
2782 clear_extent_dirty(unpin, start, end, &cached_state);
2783 unpin_extent_range(fs_info, start, end, true);
2784 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2785 free_extent_state(cached_state);
2789 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2790 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2791 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2795 * Transaction is finished. We don't need the lock anymore. We
2796 * do need to clean up the block groups in case of a transaction
2799 deleted_bgs = &trans->transaction->deleted_bgs;
2800 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2804 if (!TRANS_ABORTED(trans))
2805 ret = btrfs_discard_extent(fs_info,
2807 block_group->length,
2810 list_del_init(&block_group->bg_list);
2811 btrfs_unfreeze_block_group(block_group);
2812 btrfs_put_block_group(block_group);
2815 const char *errstr = btrfs_decode_error(ret);
2817 "discard failed while removing blockgroup: errno=%d %s",
2825 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
2826 u64 bytenr, u64 num_bytes, bool is_data)
2831 struct btrfs_root *csum_root;
2833 csum_root = btrfs_csum_root(trans->fs_info, bytenr);
2834 ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes);
2836 btrfs_abort_transaction(trans, ret);
2841 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
2843 btrfs_abort_transaction(trans, ret);
2847 ret = btrfs_update_block_group(trans, bytenr, num_bytes, false);
2849 btrfs_abort_transaction(trans, ret);
2855 * Drop one or more refs of @node.
2857 * 1. Locate the extent refs.
2858 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2859 * Locate it, then reduce the refs number or remove the ref line completely.
2861 * 2. Update the refs count in EXTENT/METADATA_ITEM
2863 * Inline backref case:
2865 * in extent tree we have:
2867 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2868 * refs 2 gen 6 flags DATA
2869 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2870 * extent data backref root FS_TREE objectid 257 offset 0 count 1
2872 * This function gets called with:
2874 * node->bytenr = 13631488
2875 * node->num_bytes = 1048576
2876 * root_objectid = FS_TREE
2877 * owner_objectid = 257
2881 * Then we should get some like:
2883 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2884 * refs 1 gen 6 flags DATA
2885 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2887 * Keyed backref case:
2889 * in extent tree we have:
2891 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2892 * refs 754 gen 6 flags DATA
2894 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2895 * extent data backref root FS_TREE objectid 866 offset 0 count 1
2897 * This function get called with:
2899 * node->bytenr = 13631488
2900 * node->num_bytes = 1048576
2901 * root_objectid = FS_TREE
2902 * owner_objectid = 866
2906 * Then we should get some like:
2908 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2909 * refs 753 gen 6 flags DATA
2911 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2913 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2914 struct btrfs_delayed_ref_node *node, u64 parent,
2915 u64 root_objectid, u64 owner_objectid,
2916 u64 owner_offset, int refs_to_drop,
2917 struct btrfs_delayed_extent_op *extent_op)
2919 struct btrfs_fs_info *info = trans->fs_info;
2920 struct btrfs_key key;
2921 struct btrfs_path *path;
2922 struct btrfs_root *extent_root;
2923 struct extent_buffer *leaf;
2924 struct btrfs_extent_item *ei;
2925 struct btrfs_extent_inline_ref *iref;
2928 int extent_slot = 0;
2929 int found_extent = 0;
2933 u64 bytenr = node->bytenr;
2934 u64 num_bytes = node->num_bytes;
2935 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2937 extent_root = btrfs_extent_root(info, bytenr);
2938 ASSERT(extent_root);
2940 path = btrfs_alloc_path();
2944 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2946 if (!is_data && refs_to_drop != 1) {
2948 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2949 node->bytenr, refs_to_drop);
2951 btrfs_abort_transaction(trans, ret);
2956 skinny_metadata = false;
2958 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2959 parent, root_objectid, owner_objectid,
2963 * Either the inline backref or the SHARED_DATA_REF/
2964 * SHARED_BLOCK_REF is found
2966 * Here is a quick path to locate EXTENT/METADATA_ITEM.
2967 * It's possible the EXTENT/METADATA_ITEM is near current slot.
2969 extent_slot = path->slots[0];
2970 while (extent_slot >= 0) {
2971 btrfs_item_key_to_cpu(path->nodes[0], &key,
2973 if (key.objectid != bytenr)
2975 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2976 key.offset == num_bytes) {
2980 if (key.type == BTRFS_METADATA_ITEM_KEY &&
2981 key.offset == owner_objectid) {
2986 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
2987 if (path->slots[0] - extent_slot > 5)
2992 if (!found_extent) {
2995 "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref");
2996 btrfs_abort_transaction(trans, -EUCLEAN);
2999 /* Must be SHARED_* item, remove the backref first */
3000 ret = remove_extent_backref(trans, extent_root, path,
3001 NULL, refs_to_drop, is_data);
3003 btrfs_abort_transaction(trans, ret);
3006 btrfs_release_path(path);
3008 /* Slow path to locate EXTENT/METADATA_ITEM */
3009 key.objectid = bytenr;
3010 key.type = BTRFS_EXTENT_ITEM_KEY;
3011 key.offset = num_bytes;
3013 if (!is_data && skinny_metadata) {
3014 key.type = BTRFS_METADATA_ITEM_KEY;
3015 key.offset = owner_objectid;
3018 ret = btrfs_search_slot(trans, extent_root,
3020 if (ret > 0 && skinny_metadata && path->slots[0]) {
3022 * Couldn't find our skinny metadata item,
3023 * see if we have ye olde extent item.
3026 btrfs_item_key_to_cpu(path->nodes[0], &key,
3028 if (key.objectid == bytenr &&
3029 key.type == BTRFS_EXTENT_ITEM_KEY &&
3030 key.offset == num_bytes)
3034 if (ret > 0 && skinny_metadata) {
3035 skinny_metadata = false;
3036 key.objectid = bytenr;
3037 key.type = BTRFS_EXTENT_ITEM_KEY;
3038 key.offset = num_bytes;
3039 btrfs_release_path(path);
3040 ret = btrfs_search_slot(trans, extent_root,
3046 "umm, got %d back from search, was looking for %llu",
3049 btrfs_print_leaf(path->nodes[0]);
3052 btrfs_abort_transaction(trans, ret);
3055 extent_slot = path->slots[0];
3057 } else if (WARN_ON(ret == -ENOENT)) {
3058 btrfs_print_leaf(path->nodes[0]);
3060 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
3061 bytenr, parent, root_objectid, owner_objectid,
3063 btrfs_abort_transaction(trans, ret);
3066 btrfs_abort_transaction(trans, ret);
3070 leaf = path->nodes[0];
3071 item_size = btrfs_item_size(leaf, extent_slot);
3072 if (unlikely(item_size < sizeof(*ei))) {
3074 btrfs_print_v0_err(info);
3075 btrfs_abort_transaction(trans, ret);
3078 ei = btrfs_item_ptr(leaf, extent_slot,
3079 struct btrfs_extent_item);
3080 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3081 key.type == BTRFS_EXTENT_ITEM_KEY) {
3082 struct btrfs_tree_block_info *bi;
3083 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3085 "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu",
3086 key.objectid, key.type, key.offset,
3087 owner_objectid, item_size,
3088 sizeof(*ei) + sizeof(*bi));
3089 btrfs_abort_transaction(trans, -EUCLEAN);
3092 bi = (struct btrfs_tree_block_info *)(ei + 1);
3093 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3096 refs = btrfs_extent_refs(leaf, ei);
3097 if (refs < refs_to_drop) {
3099 "trying to drop %d refs but we only have %llu for bytenr %llu",
3100 refs_to_drop, refs, bytenr);
3101 btrfs_abort_transaction(trans, -EUCLEAN);
3104 refs -= refs_to_drop;
3108 __run_delayed_extent_op(extent_op, leaf, ei);
3110 * In the case of inline back ref, reference count will
3111 * be updated by remove_extent_backref
3114 if (!found_extent) {
3116 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found");
3117 btrfs_abort_transaction(trans, -EUCLEAN);
3121 btrfs_set_extent_refs(leaf, ei, refs);
3122 btrfs_mark_buffer_dirty(leaf);
3125 ret = remove_extent_backref(trans, extent_root, path,
3126 iref, refs_to_drop, is_data);
3128 btrfs_abort_transaction(trans, ret);
3133 /* In this branch refs == 1 */
3135 if (is_data && refs_to_drop !=
3136 extent_data_ref_count(path, iref)) {
3138 "invalid refs_to_drop, current refs %u refs_to_drop %u",
3139 extent_data_ref_count(path, iref),
3141 btrfs_abort_transaction(trans, -EUCLEAN);
3145 if (path->slots[0] != extent_slot) {
3147 "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref",
3148 key.objectid, key.type,
3150 btrfs_abort_transaction(trans, -EUCLEAN);
3155 * No inline ref, we must be at SHARED_* item,
3156 * And it's single ref, it must be:
3157 * | extent_slot ||extent_slot + 1|
3158 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3160 if (path->slots[0] != extent_slot + 1) {
3162 "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM");
3163 btrfs_abort_transaction(trans, -EUCLEAN);
3166 path->slots[0] = extent_slot;
3171 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3174 btrfs_abort_transaction(trans, ret);
3177 btrfs_release_path(path);
3179 ret = do_free_extent_accounting(trans, bytenr, num_bytes, is_data);
3181 btrfs_release_path(path);
3184 btrfs_free_path(path);
3188 * Leaf dump can take up a lot of log buffer, so we only do full leaf
3189 * dump for debug build.
3191 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) {
3192 btrfs_crit(info, "path->slots[0]=%d extent_slot=%d",
3193 path->slots[0], extent_slot);
3194 btrfs_print_leaf(path->nodes[0]);
3197 btrfs_free_path(path);
3202 * when we free an block, it is possible (and likely) that we free the last
3203 * delayed ref for that extent as well. This searches the delayed ref tree for
3204 * a given extent, and if there are no other delayed refs to be processed, it
3205 * removes it from the tree.
3207 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3210 struct btrfs_delayed_ref_head *head;
3211 struct btrfs_delayed_ref_root *delayed_refs;
3214 delayed_refs = &trans->transaction->delayed_refs;
3215 spin_lock(&delayed_refs->lock);
3216 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3218 goto out_delayed_unlock;
3220 spin_lock(&head->lock);
3221 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3224 if (cleanup_extent_op(head) != NULL)
3228 * waiting for the lock here would deadlock. If someone else has it
3229 * locked they are already in the process of dropping it anyway
3231 if (!mutex_trylock(&head->mutex))
3234 btrfs_delete_ref_head(delayed_refs, head);
3235 head->processing = 0;
3237 spin_unlock(&head->lock);
3238 spin_unlock(&delayed_refs->lock);
3240 BUG_ON(head->extent_op);
3241 if (head->must_insert_reserved)
3244 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3245 mutex_unlock(&head->mutex);
3246 btrfs_put_delayed_ref_head(head);
3249 spin_unlock(&head->lock);
3252 spin_unlock(&delayed_refs->lock);
3256 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3258 struct extent_buffer *buf,
3259 u64 parent, int last_ref)
3261 struct btrfs_fs_info *fs_info = trans->fs_info;
3262 struct btrfs_ref generic_ref = { 0 };
3265 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3266 buf->start, buf->len, parent);
3267 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3270 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3271 btrfs_ref_tree_mod(fs_info, &generic_ref);
3272 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3273 BUG_ON(ret); /* -ENOMEM */
3276 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3277 struct btrfs_block_group *cache;
3278 bool must_pin = false;
3280 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3281 ret = check_ref_cleanup(trans, buf->start);
3283 btrfs_redirty_list_add(trans->transaction, buf);
3288 cache = btrfs_lookup_block_group(fs_info, buf->start);
3290 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3291 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3292 btrfs_put_block_group(cache);
3297 * If this is a leaf and there are tree mod log users, we may
3298 * have recorded mod log operations that point to this leaf.
3299 * So we must make sure no one reuses this leaf's extent before
3300 * mod log operations are applied to a node, otherwise after
3301 * rewinding a node using the mod log operations we get an
3302 * inconsistent btree, as the leaf's extent may now be used as
3303 * a node or leaf for another different btree.
3304 * We are safe from races here because at this point no other
3305 * node or root points to this extent buffer, so if after this
3306 * check a new tree mod log user joins, it will not be able to
3307 * find a node pointing to this leaf and record operations that
3308 * point to this leaf.
3310 if (btrfs_header_level(buf) == 0 &&
3311 test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
3314 if (must_pin || btrfs_is_zoned(fs_info)) {
3315 btrfs_redirty_list_add(trans->transaction, buf);
3316 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3317 btrfs_put_block_group(cache);
3321 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3323 btrfs_add_free_space(cache, buf->start, buf->len);
3324 btrfs_free_reserved_bytes(cache, buf->len, 0);
3325 btrfs_put_block_group(cache);
3326 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3331 * Deleting the buffer, clear the corrupt flag since it doesn't
3334 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3338 /* Can return -ENOMEM */
3339 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3341 struct btrfs_fs_info *fs_info = trans->fs_info;
3344 if (btrfs_is_testing(fs_info))
3348 * tree log blocks never actually go into the extent allocation
3349 * tree, just update pinning info and exit early.
3351 if ((ref->type == BTRFS_REF_METADATA &&
3352 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3353 (ref->type == BTRFS_REF_DATA &&
3354 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)) {
3355 /* unlocks the pinned mutex */
3356 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3358 } else if (ref->type == BTRFS_REF_METADATA) {
3359 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3361 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3364 if (!((ref->type == BTRFS_REF_METADATA &&
3365 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3366 (ref->type == BTRFS_REF_DATA &&
3367 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)))
3368 btrfs_ref_tree_mod(fs_info, ref);
3373 enum btrfs_loop_type {
3374 LOOP_CACHING_NOWAIT,
3381 btrfs_lock_block_group(struct btrfs_block_group *cache,
3385 down_read(&cache->data_rwsem);
3388 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3391 btrfs_get_block_group(cache);
3393 down_read(&cache->data_rwsem);
3396 static struct btrfs_block_group *btrfs_lock_cluster(
3397 struct btrfs_block_group *block_group,
3398 struct btrfs_free_cluster *cluster,
3400 __acquires(&cluster->refill_lock)
3402 struct btrfs_block_group *used_bg = NULL;
3404 spin_lock(&cluster->refill_lock);
3406 used_bg = cluster->block_group;
3410 if (used_bg == block_group)
3413 btrfs_get_block_group(used_bg);
3418 if (down_read_trylock(&used_bg->data_rwsem))
3421 spin_unlock(&cluster->refill_lock);
3423 /* We should only have one-level nested. */
3424 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3426 spin_lock(&cluster->refill_lock);
3427 if (used_bg == cluster->block_group)
3430 up_read(&used_bg->data_rwsem);
3431 btrfs_put_block_group(used_bg);
3436 btrfs_release_block_group(struct btrfs_block_group *cache,
3440 up_read(&cache->data_rwsem);
3441 btrfs_put_block_group(cache);
3444 enum btrfs_extent_allocation_policy {
3445 BTRFS_EXTENT_ALLOC_CLUSTERED,
3446 BTRFS_EXTENT_ALLOC_ZONED,
3450 * Structure used internally for find_free_extent() function. Wraps needed
3453 struct find_free_extent_ctl {
3454 /* Basic allocation info */
3462 /* Where to start the search inside the bg */
3465 /* For clustered allocation */
3467 struct btrfs_free_cluster *last_ptr;
3470 bool have_caching_bg;
3471 bool orig_have_caching_bg;
3473 /* Allocation is called for tree-log */
3476 /* Allocation is called for data relocation */
3477 bool for_data_reloc;
3479 /* RAID index, converted from flags */
3483 * Current loop number, check find_free_extent_update_loop() for details
3488 * Whether we're refilling a cluster, if true we need to re-search
3489 * current block group but don't try to refill the cluster again.
3491 bool retry_clustered;
3494 * Whether we're updating free space cache, if true we need to re-search
3495 * current block group but don't try updating free space cache again.
3497 bool retry_unclustered;
3499 /* If current block group is cached */
3502 /* Max contiguous hole found */
3503 u64 max_extent_size;
3505 /* Total free space from free space cache, not always contiguous */
3506 u64 total_free_space;
3511 /* Hint where to start looking for an empty space */
3514 /* Allocation policy */
3515 enum btrfs_extent_allocation_policy policy;
3520 * Helper function for find_free_extent().
3522 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3523 * Return -EAGAIN to inform caller that we need to re-search this block group
3524 * Return >0 to inform caller that we find nothing
3525 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3527 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3528 struct find_free_extent_ctl *ffe_ctl,
3529 struct btrfs_block_group **cluster_bg_ret)
3531 struct btrfs_block_group *cluster_bg;
3532 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3533 u64 aligned_cluster;
3537 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3539 goto refill_cluster;
3540 if (cluster_bg != bg && (cluster_bg->ro ||
3541 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3542 goto release_cluster;
3544 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3545 ffe_ctl->num_bytes, cluster_bg->start,
3546 &ffe_ctl->max_extent_size);
3548 /* We have a block, we're done */
3549 spin_unlock(&last_ptr->refill_lock);
3550 trace_btrfs_reserve_extent_cluster(cluster_bg,
3551 ffe_ctl->search_start, ffe_ctl->num_bytes);
3552 *cluster_bg_ret = cluster_bg;
3553 ffe_ctl->found_offset = offset;
3556 WARN_ON(last_ptr->block_group != cluster_bg);
3560 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3561 * lets just skip it and let the allocator find whatever block it can
3562 * find. If we reach this point, we will have tried the cluster
3563 * allocator plenty of times and not have found anything, so we are
3564 * likely way too fragmented for the clustering stuff to find anything.
3566 * However, if the cluster is taken from the current block group,
3567 * release the cluster first, so that we stand a better chance of
3568 * succeeding in the unclustered allocation.
3570 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3571 spin_unlock(&last_ptr->refill_lock);
3572 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3576 /* This cluster didn't work out, free it and start over */
3577 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3579 if (cluster_bg != bg)
3580 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3583 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3584 spin_unlock(&last_ptr->refill_lock);
3588 aligned_cluster = max_t(u64,
3589 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3590 bg->full_stripe_len);
3591 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3592 ffe_ctl->num_bytes, aligned_cluster);
3594 /* Now pull our allocation out of this cluster */
3595 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3596 ffe_ctl->num_bytes, ffe_ctl->search_start,
3597 &ffe_ctl->max_extent_size);
3599 /* We found one, proceed */
3600 spin_unlock(&last_ptr->refill_lock);
3601 trace_btrfs_reserve_extent_cluster(bg,
3602 ffe_ctl->search_start,
3603 ffe_ctl->num_bytes);
3604 ffe_ctl->found_offset = offset;
3607 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3608 !ffe_ctl->retry_clustered) {
3609 spin_unlock(&last_ptr->refill_lock);
3611 ffe_ctl->retry_clustered = true;
3612 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3613 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3617 * At this point we either didn't find a cluster or we weren't able to
3618 * allocate a block from our cluster. Free the cluster we've been
3619 * trying to use, and go to the next block group.
3621 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3622 spin_unlock(&last_ptr->refill_lock);
3627 * Return >0 to inform caller that we find nothing
3628 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3629 * Return -EAGAIN to inform caller that we need to re-search this block group
3631 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3632 struct find_free_extent_ctl *ffe_ctl)
3634 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3638 * We are doing an unclustered allocation, set the fragmented flag so
3639 * we don't bother trying to setup a cluster again until we get more
3642 if (unlikely(last_ptr)) {
3643 spin_lock(&last_ptr->lock);
3644 last_ptr->fragmented = 1;
3645 spin_unlock(&last_ptr->lock);
3647 if (ffe_ctl->cached) {
3648 struct btrfs_free_space_ctl *free_space_ctl;
3650 free_space_ctl = bg->free_space_ctl;
3651 spin_lock(&free_space_ctl->tree_lock);
3652 if (free_space_ctl->free_space <
3653 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3654 ffe_ctl->empty_size) {
3655 ffe_ctl->total_free_space = max_t(u64,
3656 ffe_ctl->total_free_space,
3657 free_space_ctl->free_space);
3658 spin_unlock(&free_space_ctl->tree_lock);
3661 spin_unlock(&free_space_ctl->tree_lock);
3664 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3665 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3666 &ffe_ctl->max_extent_size);
3669 * If we didn't find a chunk, and we haven't failed on this block group
3670 * before, and this block group is in the middle of caching and we are
3671 * ok with waiting, then go ahead and wait for progress to be made, and
3672 * set @retry_unclustered to true.
3674 * If @retry_unclustered is true then we've already waited on this
3675 * block group once and should move on to the next block group.
3677 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3678 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
3679 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3680 ffe_ctl->empty_size);
3681 ffe_ctl->retry_unclustered = true;
3683 } else if (!offset) {
3686 ffe_ctl->found_offset = offset;
3690 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3691 struct find_free_extent_ctl *ffe_ctl,
3692 struct btrfs_block_group **bg_ret)
3696 /* We want to try and use the cluster allocator, so lets look there */
3697 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3698 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3699 if (ret >= 0 || ret == -EAGAIN)
3701 /* ret == -ENOENT case falls through */
3704 return find_free_extent_unclustered(block_group, ffe_ctl);
3708 * Tree-log block group locking
3709 * ============================
3711 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3712 * indicates the starting address of a block group, which is reserved only
3713 * for tree-log metadata.
3720 * fs_info::treelog_bg_lock
3724 * Simple allocator for sequential-only block group. It only allows sequential
3725 * allocation. No need to play with trees. This function also reserves the
3726 * bytes as in btrfs_add_reserved_bytes.
3728 static int do_allocation_zoned(struct btrfs_block_group *block_group,
3729 struct find_free_extent_ctl *ffe_ctl,
3730 struct btrfs_block_group **bg_ret)
3732 struct btrfs_fs_info *fs_info = block_group->fs_info;
3733 struct btrfs_space_info *space_info = block_group->space_info;
3734 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3735 u64 start = block_group->start;
3736 u64 num_bytes = ffe_ctl->num_bytes;
3738 u64 bytenr = block_group->start;
3740 u64 data_reloc_bytenr;
3744 ASSERT(btrfs_is_zoned(block_group->fs_info));
3747 * Do not allow non-tree-log blocks in the dedicated tree-log block
3748 * group, and vice versa.
3750 spin_lock(&fs_info->treelog_bg_lock);
3751 log_bytenr = fs_info->treelog_bg;
3752 if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3753 (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3755 spin_unlock(&fs_info->treelog_bg_lock);
3760 * Do not allow non-relocation blocks in the dedicated relocation block
3761 * group, and vice versa.
3763 spin_lock(&fs_info->relocation_bg_lock);
3764 data_reloc_bytenr = fs_info->data_reloc_bg;
3765 if (data_reloc_bytenr &&
3766 ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3767 (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3769 spin_unlock(&fs_info->relocation_bg_lock);
3773 /* Check RO and no space case before trying to activate it */
3774 spin_lock(&block_group->lock);
3775 if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) {
3778 * May need to clear fs_info->{treelog,data_reloc}_bg.
3779 * Return the error after taking the locks.
3782 spin_unlock(&block_group->lock);
3784 if (!ret && !btrfs_zone_activate(block_group)) {
3787 * May need to clear fs_info->{treelog,data_reloc}_bg.
3788 * Return the error after taking the locks.
3792 spin_lock(&space_info->lock);
3793 spin_lock(&block_group->lock);
3794 spin_lock(&fs_info->treelog_bg_lock);
3795 spin_lock(&fs_info->relocation_bg_lock);
3800 ASSERT(!ffe_ctl->for_treelog ||
3801 block_group->start == fs_info->treelog_bg ||
3802 fs_info->treelog_bg == 0);
3803 ASSERT(!ffe_ctl->for_data_reloc ||
3804 block_group->start == fs_info->data_reloc_bg ||
3805 fs_info->data_reloc_bg == 0);
3807 if (block_group->ro || block_group->zoned_data_reloc_ongoing) {
3813 * Do not allow currently using block group to be tree-log dedicated
3816 if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3817 (block_group->used || block_group->reserved)) {
3823 * Do not allow currently used block group to be the data relocation
3824 * dedicated block group.
3826 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3827 (block_group->used || block_group->reserved)) {
3832 WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3833 avail = block_group->zone_capacity - block_group->alloc_offset;
3834 if (avail < num_bytes) {
3835 if (ffe_ctl->max_extent_size < avail) {
3837 * With sequential allocator, free space is always
3840 ffe_ctl->max_extent_size = avail;
3841 ffe_ctl->total_free_space = avail;
3847 if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3848 fs_info->treelog_bg = block_group->start;
3850 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg)
3851 fs_info->data_reloc_bg = block_group->start;
3853 ffe_ctl->found_offset = start + block_group->alloc_offset;
3854 block_group->alloc_offset += num_bytes;
3855 spin_lock(&ctl->tree_lock);
3856 ctl->free_space -= num_bytes;
3857 spin_unlock(&ctl->tree_lock);
3860 * We do not check if found_offset is aligned to stripesize. The
3861 * address is anyway rewritten when using zone append writing.
3864 ffe_ctl->search_start = ffe_ctl->found_offset;
3867 if (ret && ffe_ctl->for_treelog)
3868 fs_info->treelog_bg = 0;
3869 if (ret && ffe_ctl->for_data_reloc &&
3870 fs_info->data_reloc_bg == block_group->start) {
3872 * Do not allow further allocations from this block group.
3873 * Compared to increasing the ->ro, setting the
3874 * ->zoned_data_reloc_ongoing flag still allows nocow
3875 * writers to come in. See btrfs_inc_nocow_writers().
3877 * We need to disable an allocation to avoid an allocation of
3878 * regular (non-relocation data) extent. With mix of relocation
3879 * extents and regular extents, we can dispatch WRITE commands
3880 * (for relocation extents) and ZONE APPEND commands (for
3881 * regular extents) at the same time to the same zone, which
3882 * easily break the write pointer.
3884 block_group->zoned_data_reloc_ongoing = 1;
3885 fs_info->data_reloc_bg = 0;
3887 spin_unlock(&fs_info->relocation_bg_lock);
3888 spin_unlock(&fs_info->treelog_bg_lock);
3889 spin_unlock(&block_group->lock);
3890 spin_unlock(&space_info->lock);
3894 static int do_allocation(struct btrfs_block_group *block_group,
3895 struct find_free_extent_ctl *ffe_ctl,
3896 struct btrfs_block_group **bg_ret)
3898 switch (ffe_ctl->policy) {
3899 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3900 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3901 case BTRFS_EXTENT_ALLOC_ZONED:
3902 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
3908 static void release_block_group(struct btrfs_block_group *block_group,
3909 struct find_free_extent_ctl *ffe_ctl,
3912 switch (ffe_ctl->policy) {
3913 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3914 ffe_ctl->retry_clustered = false;
3915 ffe_ctl->retry_unclustered = false;
3917 case BTRFS_EXTENT_ALLOC_ZONED:
3924 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
3926 btrfs_release_block_group(block_group, delalloc);
3929 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
3930 struct btrfs_key *ins)
3932 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3934 if (!ffe_ctl->use_cluster && last_ptr) {
3935 spin_lock(&last_ptr->lock);
3936 last_ptr->window_start = ins->objectid;
3937 spin_unlock(&last_ptr->lock);
3941 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
3942 struct btrfs_key *ins)
3944 switch (ffe_ctl->policy) {
3945 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3946 found_extent_clustered(ffe_ctl, ins);
3948 case BTRFS_EXTENT_ALLOC_ZONED:
3956 static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info,
3957 struct find_free_extent_ctl *ffe_ctl)
3959 /* If we can activate new zone, just allocate a chunk and use it */
3960 if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags))
3964 * We already reached the max active zones. Try to finish one block
3965 * group to make a room for a new block group. This is only possible
3966 * for a data block group because btrfs_zone_finish() may need to wait
3967 * for a running transaction which can cause a deadlock for metadata
3970 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
3971 int ret = btrfs_zone_finish_one_bg(fs_info);
3980 * If we have enough free space left in an already active block group
3981 * and we can't activate any other zone now, do not allow allocating a
3982 * new chunk and let find_free_extent() retry with a smaller size.
3984 if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size)
3988 * Even min_alloc_size is not left in any block groups. Since we cannot
3989 * activate a new block group, allocating it may not help. Let's tell a
3990 * caller to try again and hope it progress something by writing some
3991 * parts of the region. That is only possible for data block groups,
3992 * where a part of the region can be written.
3994 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)
3998 * We cannot activate a new block group and no enough space left in any
3999 * block groups. So, allocating a new block group may not help. But,
4000 * there is nothing to do anyway, so let's go with it.
4005 static int can_allocate_chunk(struct btrfs_fs_info *fs_info,
4006 struct find_free_extent_ctl *ffe_ctl)
4008 switch (ffe_ctl->policy) {
4009 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4011 case BTRFS_EXTENT_ALLOC_ZONED:
4012 return can_allocate_chunk_zoned(fs_info, ffe_ctl);
4018 static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl)
4020 switch (ffe_ctl->policy) {
4021 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4023 * If we can't allocate a new chunk we've already looped through
4024 * at least once, move on to the NO_EMPTY_SIZE case.
4026 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
4028 case BTRFS_EXTENT_ALLOC_ZONED:
4037 * Return >0 means caller needs to re-search for free extent
4038 * Return 0 means we have the needed free extent.
4039 * Return <0 means we failed to locate any free extent.
4041 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
4042 struct btrfs_key *ins,
4043 struct find_free_extent_ctl *ffe_ctl,
4046 struct btrfs_root *root = fs_info->chunk_root;
4049 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
4050 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
4051 ffe_ctl->orig_have_caching_bg = true;
4053 if (ins->objectid) {
4054 found_extent(ffe_ctl, ins);
4058 if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
4062 if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
4066 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4067 * caching kthreads as we move along
4068 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4069 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4070 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4073 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
4075 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
4077 * We want to skip the LOOP_CACHING_WAIT step if we
4078 * don't have any uncached bgs and we've already done a
4079 * full search through.
4081 if (ffe_ctl->orig_have_caching_bg || !full_search)
4082 ffe_ctl->loop = LOOP_CACHING_WAIT;
4084 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
4089 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
4090 struct btrfs_trans_handle *trans;
4093 /*Check if allocation policy allows to create a new chunk */
4094 ret = can_allocate_chunk(fs_info, ffe_ctl);
4098 trans = current->journal_info;
4102 trans = btrfs_join_transaction(root);
4104 if (IS_ERR(trans)) {
4105 ret = PTR_ERR(trans);
4109 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
4110 CHUNK_ALLOC_FORCE_FOR_EXTENT);
4112 /* Do not bail out on ENOSPC since we can do more. */
4114 ret = chunk_allocation_failed(ffe_ctl);
4116 btrfs_abort_transaction(trans, ret);
4120 btrfs_end_transaction(trans);
4125 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4126 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4130 * Don't loop again if we already have no empty_size and
4133 if (ffe_ctl->empty_size == 0 &&
4134 ffe_ctl->empty_cluster == 0)
4136 ffe_ctl->empty_size = 0;
4137 ffe_ctl->empty_cluster = 0;
4144 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4145 struct find_free_extent_ctl *ffe_ctl,
4146 struct btrfs_space_info *space_info,
4147 struct btrfs_key *ins)
4150 * If our free space is heavily fragmented we may not be able to make
4151 * big contiguous allocations, so instead of doing the expensive search
4152 * for free space, simply return ENOSPC with our max_extent_size so we
4153 * can go ahead and search for a more manageable chunk.
4155 * If our max_extent_size is large enough for our allocation simply
4156 * disable clustering since we will likely not be able to find enough
4157 * space to create a cluster and induce latency trying.
4159 if (space_info->max_extent_size) {
4160 spin_lock(&space_info->lock);
4161 if (space_info->max_extent_size &&
4162 ffe_ctl->num_bytes > space_info->max_extent_size) {
4163 ins->offset = space_info->max_extent_size;
4164 spin_unlock(&space_info->lock);
4166 } else if (space_info->max_extent_size) {
4167 ffe_ctl->use_cluster = false;
4169 spin_unlock(&space_info->lock);
4172 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4173 &ffe_ctl->empty_cluster);
4174 if (ffe_ctl->last_ptr) {
4175 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4177 spin_lock(&last_ptr->lock);
4178 if (last_ptr->block_group)
4179 ffe_ctl->hint_byte = last_ptr->window_start;
4180 if (last_ptr->fragmented) {
4182 * We still set window_start so we can keep track of the
4183 * last place we found an allocation to try and save
4186 ffe_ctl->hint_byte = last_ptr->window_start;
4187 ffe_ctl->use_cluster = false;
4189 spin_unlock(&last_ptr->lock);
4195 static int prepare_allocation(struct btrfs_fs_info *fs_info,
4196 struct find_free_extent_ctl *ffe_ctl,
4197 struct btrfs_space_info *space_info,
4198 struct btrfs_key *ins)
4200 switch (ffe_ctl->policy) {
4201 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4202 return prepare_allocation_clustered(fs_info, ffe_ctl,
4204 case BTRFS_EXTENT_ALLOC_ZONED:
4205 if (ffe_ctl->for_treelog) {
4206 spin_lock(&fs_info->treelog_bg_lock);
4207 if (fs_info->treelog_bg)
4208 ffe_ctl->hint_byte = fs_info->treelog_bg;
4209 spin_unlock(&fs_info->treelog_bg_lock);
4211 if (ffe_ctl->for_data_reloc) {
4212 spin_lock(&fs_info->relocation_bg_lock);
4213 if (fs_info->data_reloc_bg)
4214 ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4215 spin_unlock(&fs_info->relocation_bg_lock);
4224 * walks the btree of allocated extents and find a hole of a given size.
4225 * The key ins is changed to record the hole:
4226 * ins->objectid == start position
4227 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4228 * ins->offset == the size of the hole.
4229 * Any available blocks before search_start are skipped.
4231 * If there is no suitable free space, we will record the max size of
4232 * the free space extent currently.
4234 * The overall logic and call chain:
4236 * find_free_extent()
4237 * |- Iterate through all block groups
4238 * | |- Get a valid block group
4239 * | |- Try to do clustered allocation in that block group
4240 * | |- Try to do unclustered allocation in that block group
4241 * | |- Check if the result is valid
4242 * | | |- If valid, then exit
4243 * | |- Jump to next block group
4245 * |- Push harder to find free extents
4246 * |- If not found, re-iterate all block groups
4248 static noinline int find_free_extent(struct btrfs_root *root,
4249 struct btrfs_key *ins,
4250 struct find_free_extent_ctl *ffe_ctl)
4252 struct btrfs_fs_info *fs_info = root->fs_info;
4254 int cache_block_group_error = 0;
4255 struct btrfs_block_group *block_group = NULL;
4256 struct btrfs_space_info *space_info;
4257 bool full_search = false;
4259 WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4261 ffe_ctl->search_start = 0;
4262 /* For clustered allocation */
4263 ffe_ctl->empty_cluster = 0;
4264 ffe_ctl->last_ptr = NULL;
4265 ffe_ctl->use_cluster = true;
4266 ffe_ctl->have_caching_bg = false;
4267 ffe_ctl->orig_have_caching_bg = false;
4268 ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4270 /* For clustered allocation */
4271 ffe_ctl->retry_clustered = false;
4272 ffe_ctl->retry_unclustered = false;
4273 ffe_ctl->cached = 0;
4274 ffe_ctl->max_extent_size = 0;
4275 ffe_ctl->total_free_space = 0;
4276 ffe_ctl->found_offset = 0;
4277 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4279 if (btrfs_is_zoned(fs_info))
4280 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4282 ins->type = BTRFS_EXTENT_ITEM_KEY;
4286 trace_find_free_extent(root, ffe_ctl->num_bytes, ffe_ctl->empty_size,
4289 space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4291 btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4295 ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4299 ffe_ctl->search_start = max(ffe_ctl->search_start,
4300 first_logical_byte(fs_info));
4301 ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4302 if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4303 block_group = btrfs_lookup_block_group(fs_info,
4304 ffe_ctl->search_start);
4306 * we don't want to use the block group if it doesn't match our
4307 * allocation bits, or if its not cached.
4309 * However if we are re-searching with an ideal block group
4310 * picked out then we don't care that the block group is cached.
4312 if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4313 block_group->cached != BTRFS_CACHE_NO) {
4314 down_read(&space_info->groups_sem);
4315 if (list_empty(&block_group->list) ||
4318 * someone is removing this block group,
4319 * we can't jump into the have_block_group
4320 * target because our list pointers are not
4323 btrfs_put_block_group(block_group);
4324 up_read(&space_info->groups_sem);
4326 ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4327 block_group->flags);
4328 btrfs_lock_block_group(block_group,
4330 goto have_block_group;
4332 } else if (block_group) {
4333 btrfs_put_block_group(block_group);
4337 ffe_ctl->have_caching_bg = false;
4338 if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4339 ffe_ctl->index == 0)
4341 down_read(&space_info->groups_sem);
4342 list_for_each_entry(block_group,
4343 &space_info->block_groups[ffe_ctl->index], list) {
4344 struct btrfs_block_group *bg_ret;
4346 /* If the block group is read-only, we can skip it entirely. */
4347 if (unlikely(block_group->ro)) {
4348 if (ffe_ctl->for_treelog)
4349 btrfs_clear_treelog_bg(block_group);
4350 if (ffe_ctl->for_data_reloc)
4351 btrfs_clear_data_reloc_bg(block_group);
4355 btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4356 ffe_ctl->search_start = block_group->start;
4359 * this can happen if we end up cycling through all the
4360 * raid types, but we want to make sure we only allocate
4361 * for the proper type.
4363 if (!block_group_bits(block_group, ffe_ctl->flags)) {
4364 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4365 BTRFS_BLOCK_GROUP_RAID1_MASK |
4366 BTRFS_BLOCK_GROUP_RAID56_MASK |
4367 BTRFS_BLOCK_GROUP_RAID10;
4370 * if they asked for extra copies and this block group
4371 * doesn't provide them, bail. This does allow us to
4372 * fill raid0 from raid1.
4374 if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4378 * This block group has different flags than we want.
4379 * It's possible that we have MIXED_GROUP flag but no
4380 * block group is mixed. Just skip such block group.
4382 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4387 ffe_ctl->cached = btrfs_block_group_done(block_group);
4388 if (unlikely(!ffe_ctl->cached)) {
4389 ffe_ctl->have_caching_bg = true;
4390 ret = btrfs_cache_block_group(block_group, false);
4393 * If we get ENOMEM here or something else we want to
4394 * try other block groups, because it may not be fatal.
4395 * However if we can't find anything else we need to
4396 * save our return here so that we return the actual
4397 * error that caused problems, not ENOSPC.
4400 if (!cache_block_group_error)
4401 cache_block_group_error = ret;
4408 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
4412 ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4414 if (bg_ret && bg_ret != block_group) {
4415 btrfs_release_block_group(block_group,
4417 block_group = bg_ret;
4419 } else if (ret == -EAGAIN) {
4420 goto have_block_group;
4421 } else if (ret > 0) {
4426 ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4427 fs_info->stripesize);
4429 /* move on to the next group */
4430 if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4431 block_group->start + block_group->length) {
4432 btrfs_add_free_space_unused(block_group,
4433 ffe_ctl->found_offset,
4434 ffe_ctl->num_bytes);
4438 if (ffe_ctl->found_offset < ffe_ctl->search_start)
4439 btrfs_add_free_space_unused(block_group,
4440 ffe_ctl->found_offset,
4441 ffe_ctl->search_start - ffe_ctl->found_offset);
4443 ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4446 if (ret == -EAGAIN) {
4447 btrfs_add_free_space_unused(block_group,
4448 ffe_ctl->found_offset,
4449 ffe_ctl->num_bytes);
4452 btrfs_inc_block_group_reservations(block_group);
4454 /* we are all good, lets return */
4455 ins->objectid = ffe_ctl->search_start;
4456 ins->offset = ffe_ctl->num_bytes;
4458 trace_btrfs_reserve_extent(block_group, ffe_ctl->search_start,
4459 ffe_ctl->num_bytes);
4460 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4463 release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4466 up_read(&space_info->groups_sem);
4468 ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4472 if (ret == -ENOSPC && !cache_block_group_error) {
4474 * Use ffe_ctl->total_free_space as fallback if we can't find
4475 * any contiguous hole.
4477 if (!ffe_ctl->max_extent_size)
4478 ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4479 spin_lock(&space_info->lock);
4480 space_info->max_extent_size = ffe_ctl->max_extent_size;
4481 spin_unlock(&space_info->lock);
4482 ins->offset = ffe_ctl->max_extent_size;
4483 } else if (ret == -ENOSPC) {
4484 ret = cache_block_group_error;
4490 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4491 * hole that is at least as big as @num_bytes.
4493 * @root - The root that will contain this extent
4495 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4496 * is used for accounting purposes. This value differs
4497 * from @num_bytes only in the case of compressed extents.
4499 * @num_bytes - Number of bytes to allocate on-disk.
4501 * @min_alloc_size - Indicates the minimum amount of space that the
4502 * allocator should try to satisfy. In some cases
4503 * @num_bytes may be larger than what is required and if
4504 * the filesystem is fragmented then allocation fails.
4505 * However, the presence of @min_alloc_size gives a
4506 * chance to try and satisfy the smaller allocation.
4508 * @empty_size - A hint that you plan on doing more COW. This is the
4509 * size in bytes the allocator should try to find free
4510 * next to the block it returns. This is just a hint and
4511 * may be ignored by the allocator.
4513 * @hint_byte - Hint to the allocator to start searching above the byte
4514 * address passed. It might be ignored.
4516 * @ins - This key is modified to record the found hole. It will
4517 * have the following values:
4518 * ins->objectid == start position
4519 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4520 * ins->offset == the size of the hole.
4522 * @is_data - Boolean flag indicating whether an extent is
4523 * allocated for data (true) or metadata (false)
4525 * @delalloc - Boolean flag indicating whether this allocation is for
4526 * delalloc or not. If 'true' data_rwsem of block groups
4527 * is going to be acquired.
4530 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4531 * case -ENOSPC is returned then @ins->offset will contain the size of the
4532 * largest available hole the allocator managed to find.
4534 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4535 u64 num_bytes, u64 min_alloc_size,
4536 u64 empty_size, u64 hint_byte,
4537 struct btrfs_key *ins, int is_data, int delalloc)
4539 struct btrfs_fs_info *fs_info = root->fs_info;
4540 struct find_free_extent_ctl ffe_ctl = {};
4541 bool final_tried = num_bytes == min_alloc_size;
4544 bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4545 bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4547 flags = get_alloc_profile_by_root(root, is_data);
4549 WARN_ON(num_bytes < fs_info->sectorsize);
4551 ffe_ctl.ram_bytes = ram_bytes;
4552 ffe_ctl.num_bytes = num_bytes;
4553 ffe_ctl.min_alloc_size = min_alloc_size;
4554 ffe_ctl.empty_size = empty_size;
4555 ffe_ctl.flags = flags;
4556 ffe_ctl.delalloc = delalloc;
4557 ffe_ctl.hint_byte = hint_byte;
4558 ffe_ctl.for_treelog = for_treelog;
4559 ffe_ctl.for_data_reloc = for_data_reloc;
4561 ret = find_free_extent(root, ins, &ffe_ctl);
4562 if (!ret && !is_data) {
4563 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4564 } else if (ret == -ENOSPC) {
4565 if (!final_tried && ins->offset) {
4566 num_bytes = min(num_bytes >> 1, ins->offset);
4567 num_bytes = round_down(num_bytes,
4568 fs_info->sectorsize);
4569 num_bytes = max(num_bytes, min_alloc_size);
4570 ram_bytes = num_bytes;
4571 if (num_bytes == min_alloc_size)
4574 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4575 struct btrfs_space_info *sinfo;
4577 sinfo = btrfs_find_space_info(fs_info, flags);
4579 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4580 flags, num_bytes, for_treelog, for_data_reloc);
4582 btrfs_dump_space_info(fs_info, sinfo,
4590 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4591 u64 start, u64 len, int delalloc)
4593 struct btrfs_block_group *cache;
4595 cache = btrfs_lookup_block_group(fs_info, start);
4597 btrfs_err(fs_info, "Unable to find block group for %llu",
4602 btrfs_add_free_space(cache, start, len);
4603 btrfs_free_reserved_bytes(cache, len, delalloc);
4604 trace_btrfs_reserved_extent_free(fs_info, start, len);
4606 btrfs_put_block_group(cache);
4610 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start,
4613 struct btrfs_block_group *cache;
4616 cache = btrfs_lookup_block_group(trans->fs_info, start);
4618 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4623 ret = pin_down_extent(trans, cache, start, len, 1);
4624 btrfs_put_block_group(cache);
4628 static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
4631 struct btrfs_fs_info *fs_info = trans->fs_info;
4634 ret = remove_from_free_space_tree(trans, bytenr, num_bytes);
4638 ret = btrfs_update_block_group(trans, bytenr, num_bytes, true);
4641 btrfs_err(fs_info, "update block group failed for %llu %llu",
4646 trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes);
4650 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4651 u64 parent, u64 root_objectid,
4652 u64 flags, u64 owner, u64 offset,
4653 struct btrfs_key *ins, int ref_mod)
4655 struct btrfs_fs_info *fs_info = trans->fs_info;
4656 struct btrfs_root *extent_root;
4658 struct btrfs_extent_item *extent_item;
4659 struct btrfs_extent_inline_ref *iref;
4660 struct btrfs_path *path;
4661 struct extent_buffer *leaf;
4666 type = BTRFS_SHARED_DATA_REF_KEY;
4668 type = BTRFS_EXTENT_DATA_REF_KEY;
4670 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4672 path = btrfs_alloc_path();
4676 extent_root = btrfs_extent_root(fs_info, ins->objectid);
4677 ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size);
4679 btrfs_free_path(path);
4683 leaf = path->nodes[0];
4684 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4685 struct btrfs_extent_item);
4686 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4687 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4688 btrfs_set_extent_flags(leaf, extent_item,
4689 flags | BTRFS_EXTENT_FLAG_DATA);
4691 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4692 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4694 struct btrfs_shared_data_ref *ref;
4695 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4696 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4697 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4699 struct btrfs_extent_data_ref *ref;
4700 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4701 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4702 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4703 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4704 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4707 btrfs_mark_buffer_dirty(path->nodes[0]);
4708 btrfs_free_path(path);
4710 return alloc_reserved_extent(trans, ins->objectid, ins->offset);
4713 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4714 struct btrfs_delayed_ref_node *node,
4715 struct btrfs_delayed_extent_op *extent_op)
4717 struct btrfs_fs_info *fs_info = trans->fs_info;
4718 struct btrfs_root *extent_root;
4720 struct btrfs_extent_item *extent_item;
4721 struct btrfs_key extent_key;
4722 struct btrfs_tree_block_info *block_info;
4723 struct btrfs_extent_inline_ref *iref;
4724 struct btrfs_path *path;
4725 struct extent_buffer *leaf;
4726 struct btrfs_delayed_tree_ref *ref;
4727 u32 size = sizeof(*extent_item) + sizeof(*iref);
4728 u64 flags = extent_op->flags_to_set;
4729 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4731 ref = btrfs_delayed_node_to_tree_ref(node);
4733 extent_key.objectid = node->bytenr;
4734 if (skinny_metadata) {
4735 extent_key.offset = ref->level;
4736 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4738 extent_key.offset = node->num_bytes;
4739 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4740 size += sizeof(*block_info);
4743 path = btrfs_alloc_path();
4747 extent_root = btrfs_extent_root(fs_info, extent_key.objectid);
4748 ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key,
4751 btrfs_free_path(path);
4755 leaf = path->nodes[0];
4756 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4757 struct btrfs_extent_item);
4758 btrfs_set_extent_refs(leaf, extent_item, 1);
4759 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4760 btrfs_set_extent_flags(leaf, extent_item,
4761 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4763 if (skinny_metadata) {
4764 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4766 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4767 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4768 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4769 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4772 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4773 btrfs_set_extent_inline_ref_type(leaf, iref,
4774 BTRFS_SHARED_BLOCK_REF_KEY);
4775 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4777 btrfs_set_extent_inline_ref_type(leaf, iref,
4778 BTRFS_TREE_BLOCK_REF_KEY);
4779 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4782 btrfs_mark_buffer_dirty(leaf);
4783 btrfs_free_path(path);
4785 return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize);
4788 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4789 struct btrfs_root *root, u64 owner,
4790 u64 offset, u64 ram_bytes,
4791 struct btrfs_key *ins)
4793 struct btrfs_ref generic_ref = { 0 };
4795 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4797 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4798 ins->objectid, ins->offset, 0);
4799 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner,
4801 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4803 return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4807 * this is used by the tree logging recovery code. It records that
4808 * an extent has been allocated and makes sure to clear the free
4809 * space cache bits as well
4811 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4812 u64 root_objectid, u64 owner, u64 offset,
4813 struct btrfs_key *ins)
4815 struct btrfs_fs_info *fs_info = trans->fs_info;
4817 struct btrfs_block_group *block_group;
4818 struct btrfs_space_info *space_info;
4821 * Mixed block groups will exclude before processing the log so we only
4822 * need to do the exclude dance if this fs isn't mixed.
4824 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4825 ret = __exclude_logged_extent(fs_info, ins->objectid,
4831 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4835 space_info = block_group->space_info;
4836 spin_lock(&space_info->lock);
4837 spin_lock(&block_group->lock);
4838 space_info->bytes_reserved += ins->offset;
4839 block_group->reserved += ins->offset;
4840 spin_unlock(&block_group->lock);
4841 spin_unlock(&space_info->lock);
4843 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4846 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4847 btrfs_put_block_group(block_group);
4851 static struct extent_buffer *
4852 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4853 u64 bytenr, int level, u64 owner,
4854 enum btrfs_lock_nesting nest)
4856 struct btrfs_fs_info *fs_info = root->fs_info;
4857 struct extent_buffer *buf;
4858 u64 lockdep_owner = owner;
4860 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
4865 * Extra safety check in case the extent tree is corrupted and extent
4866 * allocator chooses to use a tree block which is already used and
4869 if (buf->lock_owner == current->pid) {
4870 btrfs_err_rl(fs_info,
4871 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4872 buf->start, btrfs_header_owner(buf), current->pid);
4873 free_extent_buffer(buf);
4874 return ERR_PTR(-EUCLEAN);
4878 * The reloc trees are just snapshots, so we need them to appear to be
4879 * just like any other fs tree WRT lockdep.
4881 * The exception however is in replace_path() in relocation, where we
4882 * hold the lock on the original fs root and then search for the reloc
4883 * root. At that point we need to make sure any reloc root buffers are
4884 * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
4887 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID &&
4888 !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
4889 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
4892 * This needs to stay, because we could allocate a freed block from an
4893 * old tree into a new tree, so we need to make sure this new block is
4894 * set to the appropriate level and owner.
4896 btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level);
4898 __btrfs_tree_lock(buf, nest);
4899 btrfs_clean_tree_block(buf);
4900 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4901 clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags);
4903 set_extent_buffer_uptodate(buf);
4905 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4906 btrfs_set_header_level(buf, level);
4907 btrfs_set_header_bytenr(buf, buf->start);
4908 btrfs_set_header_generation(buf, trans->transid);
4909 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4910 btrfs_set_header_owner(buf, owner);
4911 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4912 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4913 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4914 buf->log_index = root->log_transid % 2;
4916 * we allow two log transactions at a time, use different
4917 * EXTENT bit to differentiate dirty pages.
4919 if (buf->log_index == 0)
4920 set_extent_dirty(&root->dirty_log_pages, buf->start,
4921 buf->start + buf->len - 1, GFP_NOFS);
4923 set_extent_new(&root->dirty_log_pages, buf->start,
4924 buf->start + buf->len - 1);
4926 buf->log_index = -1;
4927 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4928 buf->start + buf->len - 1, GFP_NOFS);
4930 /* this returns a buffer locked for blocking */
4935 * finds a free extent and does all the dirty work required for allocation
4936 * returns the tree buffer or an ERR_PTR on error.
4938 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4939 struct btrfs_root *root,
4940 u64 parent, u64 root_objectid,
4941 const struct btrfs_disk_key *key,
4942 int level, u64 hint,
4944 enum btrfs_lock_nesting nest)
4946 struct btrfs_fs_info *fs_info = root->fs_info;
4947 struct btrfs_key ins;
4948 struct btrfs_block_rsv *block_rsv;
4949 struct extent_buffer *buf;
4950 struct btrfs_delayed_extent_op *extent_op;
4951 struct btrfs_ref generic_ref = { 0 };
4954 u32 blocksize = fs_info->nodesize;
4955 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4957 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4958 if (btrfs_is_testing(fs_info)) {
4959 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4960 level, root_objectid, nest);
4962 root->alloc_bytenr += blocksize;
4967 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4968 if (IS_ERR(block_rsv))
4969 return ERR_CAST(block_rsv);
4971 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4972 empty_size, hint, &ins, 0, 0);
4976 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4977 root_objectid, nest);
4980 goto out_free_reserved;
4983 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4985 parent = ins.objectid;
4986 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4990 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4991 extent_op = btrfs_alloc_delayed_extent_op();
4997 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4999 memset(&extent_op->key, 0, sizeof(extent_op->key));
5000 extent_op->flags_to_set = flags;
5001 extent_op->update_key = skinny_metadata ? false : true;
5002 extent_op->update_flags = true;
5003 extent_op->level = level;
5005 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
5006 ins.objectid, ins.offset, parent);
5007 btrfs_init_tree_ref(&generic_ref, level, root_objectid,
5008 root->root_key.objectid, false);
5009 btrfs_ref_tree_mod(fs_info, &generic_ref);
5010 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
5012 goto out_free_delayed;
5017 btrfs_free_delayed_extent_op(extent_op);
5019 btrfs_tree_unlock(buf);
5020 free_extent_buffer(buf);
5022 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
5024 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
5025 return ERR_PTR(ret);
5028 struct walk_control {
5029 u64 refs[BTRFS_MAX_LEVEL];
5030 u64 flags[BTRFS_MAX_LEVEL];
5031 struct btrfs_key update_progress;
5032 struct btrfs_key drop_progress;
5044 #define DROP_REFERENCE 1
5045 #define UPDATE_BACKREF 2
5047 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5048 struct btrfs_root *root,
5049 struct walk_control *wc,
5050 struct btrfs_path *path)
5052 struct btrfs_fs_info *fs_info = root->fs_info;
5058 struct btrfs_key key;
5059 struct extent_buffer *eb;
5064 if (path->slots[wc->level] < wc->reada_slot) {
5065 wc->reada_count = wc->reada_count * 2 / 3;
5066 wc->reada_count = max(wc->reada_count, 2);
5068 wc->reada_count = wc->reada_count * 3 / 2;
5069 wc->reada_count = min_t(int, wc->reada_count,
5070 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
5073 eb = path->nodes[wc->level];
5074 nritems = btrfs_header_nritems(eb);
5076 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5077 if (nread >= wc->reada_count)
5081 bytenr = btrfs_node_blockptr(eb, slot);
5082 generation = btrfs_node_ptr_generation(eb, slot);
5084 if (slot == path->slots[wc->level])
5087 if (wc->stage == UPDATE_BACKREF &&
5088 generation <= root->root_key.offset)
5091 /* We don't lock the tree block, it's OK to be racy here */
5092 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5093 wc->level - 1, 1, &refs,
5095 /* We don't care about errors in readahead. */
5100 if (wc->stage == DROP_REFERENCE) {
5104 if (wc->level == 1 &&
5105 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5107 if (!wc->update_ref ||
5108 generation <= root->root_key.offset)
5110 btrfs_node_key_to_cpu(eb, &key, slot);
5111 ret = btrfs_comp_cpu_keys(&key,
5112 &wc->update_progress);
5116 if (wc->level == 1 &&
5117 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5121 btrfs_readahead_node_child(eb, slot);
5124 wc->reada_slot = slot;
5128 * helper to process tree block while walking down the tree.
5130 * when wc->stage == UPDATE_BACKREF, this function updates
5131 * back refs for pointers in the block.
5133 * NOTE: return value 1 means we should stop walking down.
5135 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5136 struct btrfs_root *root,
5137 struct btrfs_path *path,
5138 struct walk_control *wc, int lookup_info)
5140 struct btrfs_fs_info *fs_info = root->fs_info;
5141 int level = wc->level;
5142 struct extent_buffer *eb = path->nodes[level];
5143 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5146 if (wc->stage == UPDATE_BACKREF &&
5147 btrfs_header_owner(eb) != root->root_key.objectid)
5151 * when reference count of tree block is 1, it won't increase
5152 * again. once full backref flag is set, we never clear it.
5155 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5156 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5157 BUG_ON(!path->locks[level]);
5158 ret = btrfs_lookup_extent_info(trans, fs_info,
5159 eb->start, level, 1,
5162 BUG_ON(ret == -ENOMEM);
5165 BUG_ON(wc->refs[level] == 0);
5168 if (wc->stage == DROP_REFERENCE) {
5169 if (wc->refs[level] > 1)
5172 if (path->locks[level] && !wc->keep_locks) {
5173 btrfs_tree_unlock_rw(eb, path->locks[level]);
5174 path->locks[level] = 0;
5179 /* wc->stage == UPDATE_BACKREF */
5180 if (!(wc->flags[level] & flag)) {
5181 BUG_ON(!path->locks[level]);
5182 ret = btrfs_inc_ref(trans, root, eb, 1);
5183 BUG_ON(ret); /* -ENOMEM */
5184 ret = btrfs_dec_ref(trans, root, eb, 0);
5185 BUG_ON(ret); /* -ENOMEM */
5186 ret = btrfs_set_disk_extent_flags(trans, eb, flag,
5187 btrfs_header_level(eb));
5188 BUG_ON(ret); /* -ENOMEM */
5189 wc->flags[level] |= flag;
5193 * the block is shared by multiple trees, so it's not good to
5194 * keep the tree lock
5196 if (path->locks[level] && level > 0) {
5197 btrfs_tree_unlock_rw(eb, path->locks[level]);
5198 path->locks[level] = 0;
5204 * This is used to verify a ref exists for this root to deal with a bug where we
5205 * would have a drop_progress key that hadn't been updated properly.
5207 static int check_ref_exists(struct btrfs_trans_handle *trans,
5208 struct btrfs_root *root, u64 bytenr, u64 parent,
5211 struct btrfs_path *path;
5212 struct btrfs_extent_inline_ref *iref;
5215 path = btrfs_alloc_path();
5219 ret = lookup_extent_backref(trans, path, &iref, bytenr,
5220 root->fs_info->nodesize, parent,
5221 root->root_key.objectid, level, 0);
5222 btrfs_free_path(path);
5231 * helper to process tree block pointer.
5233 * when wc->stage == DROP_REFERENCE, this function checks
5234 * reference count of the block pointed to. if the block
5235 * is shared and we need update back refs for the subtree
5236 * rooted at the block, this function changes wc->stage to
5237 * UPDATE_BACKREF. if the block is shared and there is no
5238 * need to update back, this function drops the reference
5241 * NOTE: return value 1 means we should stop walking down.
5243 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5244 struct btrfs_root *root,
5245 struct btrfs_path *path,
5246 struct walk_control *wc, int *lookup_info)
5248 struct btrfs_fs_info *fs_info = root->fs_info;
5252 struct btrfs_key key;
5253 struct btrfs_key first_key;
5254 struct btrfs_ref ref = { 0 };
5255 struct extent_buffer *next;
5256 int level = wc->level;
5259 bool need_account = false;
5261 generation = btrfs_node_ptr_generation(path->nodes[level],
5262 path->slots[level]);
5264 * if the lower level block was created before the snapshot
5265 * was created, we know there is no need to update back refs
5268 if (wc->stage == UPDATE_BACKREF &&
5269 generation <= root->root_key.offset) {
5274 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5275 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
5276 path->slots[level]);
5278 next = find_extent_buffer(fs_info, bytenr);
5280 next = btrfs_find_create_tree_block(fs_info, bytenr,
5281 root->root_key.objectid, level - 1);
5283 return PTR_ERR(next);
5286 btrfs_tree_lock(next);
5288 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5289 &wc->refs[level - 1],
5290 &wc->flags[level - 1]);
5294 if (unlikely(wc->refs[level - 1] == 0)) {
5295 btrfs_err(fs_info, "Missing references.");
5301 if (wc->stage == DROP_REFERENCE) {
5302 if (wc->refs[level - 1] > 1) {
5303 need_account = true;
5305 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5308 if (!wc->update_ref ||
5309 generation <= root->root_key.offset)
5312 btrfs_node_key_to_cpu(path->nodes[level], &key,
5313 path->slots[level]);
5314 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5318 wc->stage = UPDATE_BACKREF;
5319 wc->shared_level = level - 1;
5323 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5327 if (!btrfs_buffer_uptodate(next, generation, 0)) {
5328 btrfs_tree_unlock(next);
5329 free_extent_buffer(next);
5335 if (reada && level == 1)
5336 reada_walk_down(trans, root, wc, path);
5337 next = read_tree_block(fs_info, bytenr, root->root_key.objectid,
5338 generation, level - 1, &first_key);
5340 return PTR_ERR(next);
5341 } else if (!extent_buffer_uptodate(next)) {
5342 free_extent_buffer(next);
5345 btrfs_tree_lock(next);
5349 ASSERT(level == btrfs_header_level(next));
5350 if (level != btrfs_header_level(next)) {
5351 btrfs_err(root->fs_info, "mismatched level");
5355 path->nodes[level] = next;
5356 path->slots[level] = 0;
5357 path->locks[level] = BTRFS_WRITE_LOCK;
5363 wc->refs[level - 1] = 0;
5364 wc->flags[level - 1] = 0;
5365 if (wc->stage == DROP_REFERENCE) {
5366 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5367 parent = path->nodes[level]->start;
5369 ASSERT(root->root_key.objectid ==
5370 btrfs_header_owner(path->nodes[level]));
5371 if (root->root_key.objectid !=
5372 btrfs_header_owner(path->nodes[level])) {
5373 btrfs_err(root->fs_info,
5374 "mismatched block owner");
5382 * If we had a drop_progress we need to verify the refs are set
5383 * as expected. If we find our ref then we know that from here
5384 * on out everything should be correct, and we can clear the
5387 if (wc->restarted) {
5388 ret = check_ref_exists(trans, root, bytenr, parent,
5399 * Reloc tree doesn't contribute to qgroup numbers, and we have
5400 * already accounted them at merge time (replace_path),
5401 * thus we could skip expensive subtree trace here.
5403 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5405 ret = btrfs_qgroup_trace_subtree(trans, next,
5406 generation, level - 1);
5408 btrfs_err_rl(fs_info,
5409 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5415 * We need to update the next key in our walk control so we can
5416 * update the drop_progress key accordingly. We don't care if
5417 * find_next_key doesn't find a key because that means we're at
5418 * the end and are going to clean up now.
5420 wc->drop_level = level;
5421 find_next_key(path, level, &wc->drop_progress);
5423 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5424 fs_info->nodesize, parent);
5425 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid,
5427 ret = btrfs_free_extent(trans, &ref);
5436 btrfs_tree_unlock(next);
5437 free_extent_buffer(next);
5443 * helper to process tree block while walking up the tree.
5445 * when wc->stage == DROP_REFERENCE, this function drops
5446 * reference count on the block.
5448 * when wc->stage == UPDATE_BACKREF, this function changes
5449 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5450 * to UPDATE_BACKREF previously while processing the block.
5452 * NOTE: return value 1 means we should stop walking up.
5454 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5455 struct btrfs_root *root,
5456 struct btrfs_path *path,
5457 struct walk_control *wc)
5459 struct btrfs_fs_info *fs_info = root->fs_info;
5461 int level = wc->level;
5462 struct extent_buffer *eb = path->nodes[level];
5465 if (wc->stage == UPDATE_BACKREF) {
5466 BUG_ON(wc->shared_level < level);
5467 if (level < wc->shared_level)
5470 ret = find_next_key(path, level + 1, &wc->update_progress);
5474 wc->stage = DROP_REFERENCE;
5475 wc->shared_level = -1;
5476 path->slots[level] = 0;
5479 * check reference count again if the block isn't locked.
5480 * we should start walking down the tree again if reference
5483 if (!path->locks[level]) {
5485 btrfs_tree_lock(eb);
5486 path->locks[level] = BTRFS_WRITE_LOCK;
5488 ret = btrfs_lookup_extent_info(trans, fs_info,
5489 eb->start, level, 1,
5493 btrfs_tree_unlock_rw(eb, path->locks[level]);
5494 path->locks[level] = 0;
5497 BUG_ON(wc->refs[level] == 0);
5498 if (wc->refs[level] == 1) {
5499 btrfs_tree_unlock_rw(eb, path->locks[level]);
5500 path->locks[level] = 0;
5506 /* wc->stage == DROP_REFERENCE */
5507 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5509 if (wc->refs[level] == 1) {
5511 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5512 ret = btrfs_dec_ref(trans, root, eb, 1);
5514 ret = btrfs_dec_ref(trans, root, eb, 0);
5515 BUG_ON(ret); /* -ENOMEM */
5516 if (is_fstree(root->root_key.objectid)) {
5517 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5519 btrfs_err_rl(fs_info,
5520 "error %d accounting leaf items, quota is out of sync, rescan required",
5525 /* make block locked assertion in btrfs_clean_tree_block happy */
5526 if (!path->locks[level] &&
5527 btrfs_header_generation(eb) == trans->transid) {
5528 btrfs_tree_lock(eb);
5529 path->locks[level] = BTRFS_WRITE_LOCK;
5531 btrfs_clean_tree_block(eb);
5534 if (eb == root->node) {
5535 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5537 else if (root->root_key.objectid != btrfs_header_owner(eb))
5538 goto owner_mismatch;
5540 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5541 parent = path->nodes[level + 1]->start;
5542 else if (root->root_key.objectid !=
5543 btrfs_header_owner(path->nodes[level + 1]))
5544 goto owner_mismatch;
5547 btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent,
5548 wc->refs[level] == 1);
5550 wc->refs[level] = 0;
5551 wc->flags[level] = 0;
5555 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5556 btrfs_header_owner(eb), root->root_key.objectid);
5560 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5561 struct btrfs_root *root,
5562 struct btrfs_path *path,
5563 struct walk_control *wc)
5565 int level = wc->level;
5566 int lookup_info = 1;
5569 while (level >= 0) {
5570 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5577 if (path->slots[level] >=
5578 btrfs_header_nritems(path->nodes[level]))
5581 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5583 path->slots[level]++;
5592 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5593 struct btrfs_root *root,
5594 struct btrfs_path *path,
5595 struct walk_control *wc, int max_level)
5597 int level = wc->level;
5600 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5601 while (level < max_level && path->nodes[level]) {
5603 if (path->slots[level] + 1 <
5604 btrfs_header_nritems(path->nodes[level])) {
5605 path->slots[level]++;
5608 ret = walk_up_proc(trans, root, path, wc);
5614 if (path->locks[level]) {
5615 btrfs_tree_unlock_rw(path->nodes[level],
5616 path->locks[level]);
5617 path->locks[level] = 0;
5619 free_extent_buffer(path->nodes[level]);
5620 path->nodes[level] = NULL;
5628 * drop a subvolume tree.
5630 * this function traverses the tree freeing any blocks that only
5631 * referenced by the tree.
5633 * when a shared tree block is found. this function decreases its
5634 * reference count by one. if update_ref is true, this function
5635 * also make sure backrefs for the shared block and all lower level
5636 * blocks are properly updated.
5638 * If called with for_reloc == 0, may exit early with -EAGAIN
5640 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5642 struct btrfs_fs_info *fs_info = root->fs_info;
5643 struct btrfs_path *path;
5644 struct btrfs_trans_handle *trans;
5645 struct btrfs_root *tree_root = fs_info->tree_root;
5646 struct btrfs_root_item *root_item = &root->root_item;
5647 struct walk_control *wc;
5648 struct btrfs_key key;
5652 bool root_dropped = false;
5653 bool unfinished_drop = false;
5655 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5657 path = btrfs_alloc_path();
5663 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5665 btrfs_free_path(path);
5671 * Use join to avoid potential EINTR from transaction start. See
5672 * wait_reserve_ticket and the whole reservation callchain.
5675 trans = btrfs_join_transaction(tree_root);
5677 trans = btrfs_start_transaction(tree_root, 0);
5678 if (IS_ERR(trans)) {
5679 err = PTR_ERR(trans);
5683 err = btrfs_run_delayed_items(trans);
5688 * This will help us catch people modifying the fs tree while we're
5689 * dropping it. It is unsafe to mess with the fs tree while it's being
5690 * dropped as we unlock the root node and parent nodes as we walk down
5691 * the tree, assuming nothing will change. If something does change
5692 * then we'll have stale information and drop references to blocks we've
5695 set_bit(BTRFS_ROOT_DELETING, &root->state);
5696 unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
5698 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5699 level = btrfs_header_level(root->node);
5700 path->nodes[level] = btrfs_lock_root_node(root);
5701 path->slots[level] = 0;
5702 path->locks[level] = BTRFS_WRITE_LOCK;
5703 memset(&wc->update_progress, 0,
5704 sizeof(wc->update_progress));
5706 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5707 memcpy(&wc->update_progress, &key,
5708 sizeof(wc->update_progress));
5710 level = btrfs_root_drop_level(root_item);
5712 path->lowest_level = level;
5713 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5714 path->lowest_level = 0;
5722 * unlock our path, this is safe because only this
5723 * function is allowed to delete this snapshot
5725 btrfs_unlock_up_safe(path, 0);
5727 level = btrfs_header_level(root->node);
5729 btrfs_tree_lock(path->nodes[level]);
5730 path->locks[level] = BTRFS_WRITE_LOCK;
5732 ret = btrfs_lookup_extent_info(trans, fs_info,
5733 path->nodes[level]->start,
5734 level, 1, &wc->refs[level],
5740 BUG_ON(wc->refs[level] == 0);
5742 if (level == btrfs_root_drop_level(root_item))
5745 btrfs_tree_unlock(path->nodes[level]);
5746 path->locks[level] = 0;
5747 WARN_ON(wc->refs[level] != 1);
5752 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5754 wc->shared_level = -1;
5755 wc->stage = DROP_REFERENCE;
5756 wc->update_ref = update_ref;
5758 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5762 ret = walk_down_tree(trans, root, path, wc);
5768 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5775 BUG_ON(wc->stage != DROP_REFERENCE);
5779 if (wc->stage == DROP_REFERENCE) {
5780 wc->drop_level = wc->level;
5781 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5783 path->slots[wc->drop_level]);
5785 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5786 &wc->drop_progress);
5787 btrfs_set_root_drop_level(root_item, wc->drop_level);
5789 BUG_ON(wc->level == 0);
5790 if (btrfs_should_end_transaction(trans) ||
5791 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5792 ret = btrfs_update_root(trans, tree_root,
5796 btrfs_abort_transaction(trans, ret);
5801 btrfs_end_transaction_throttle(trans);
5802 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5803 btrfs_debug(fs_info,
5804 "drop snapshot early exit");
5810 * Use join to avoid potential EINTR from transaction
5811 * start. See wait_reserve_ticket and the whole
5812 * reservation callchain.
5815 trans = btrfs_join_transaction(tree_root);
5817 trans = btrfs_start_transaction(tree_root, 0);
5818 if (IS_ERR(trans)) {
5819 err = PTR_ERR(trans);
5824 btrfs_release_path(path);
5828 ret = btrfs_del_root(trans, &root->root_key);
5830 btrfs_abort_transaction(trans, ret);
5835 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5836 ret = btrfs_find_root(tree_root, &root->root_key, path,
5839 btrfs_abort_transaction(trans, ret);
5842 } else if (ret > 0) {
5843 /* if we fail to delete the orphan item this time
5844 * around, it'll get picked up the next time.
5846 * The most common failure here is just -ENOENT.
5848 btrfs_del_orphan_item(trans, tree_root,
5849 root->root_key.objectid);
5854 * This subvolume is going to be completely dropped, and won't be
5855 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5856 * commit transaction time. So free it here manually.
5858 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
5859 btrfs_qgroup_free_meta_all_pertrans(root);
5861 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
5862 btrfs_add_dropped_root(trans, root);
5864 btrfs_put_root(root);
5865 root_dropped = true;
5867 btrfs_end_transaction_throttle(trans);
5870 btrfs_free_path(path);
5873 * We were an unfinished drop root, check to see if there are any
5874 * pending, and if not clear and wake up any waiters.
5876 if (!err && unfinished_drop)
5877 btrfs_maybe_wake_unfinished_drop(fs_info);
5880 * So if we need to stop dropping the snapshot for whatever reason we
5881 * need to make sure to add it back to the dead root list so that we
5882 * keep trying to do the work later. This also cleans up roots if we
5883 * don't have it in the radix (like when we recover after a power fail
5884 * or unmount) so we don't leak memory.
5886 if (!for_reloc && !root_dropped)
5887 btrfs_add_dead_root(root);
5892 * drop subtree rooted at tree block 'node'.
5894 * NOTE: this function will unlock and release tree block 'node'
5895 * only used by relocation code
5897 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5898 struct btrfs_root *root,
5899 struct extent_buffer *node,
5900 struct extent_buffer *parent)
5902 struct btrfs_fs_info *fs_info = root->fs_info;
5903 struct btrfs_path *path;
5904 struct walk_control *wc;
5910 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5912 path = btrfs_alloc_path();
5916 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5918 btrfs_free_path(path);
5922 btrfs_assert_tree_write_locked(parent);
5923 parent_level = btrfs_header_level(parent);
5924 atomic_inc(&parent->refs);
5925 path->nodes[parent_level] = parent;
5926 path->slots[parent_level] = btrfs_header_nritems(parent);
5928 btrfs_assert_tree_write_locked(node);
5929 level = btrfs_header_level(node);
5930 path->nodes[level] = node;
5931 path->slots[level] = 0;
5932 path->locks[level] = BTRFS_WRITE_LOCK;
5934 wc->refs[parent_level] = 1;
5935 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5937 wc->shared_level = -1;
5938 wc->stage = DROP_REFERENCE;
5941 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5944 wret = walk_down_tree(trans, root, path, wc);
5950 wret = walk_up_tree(trans, root, path, wc, parent_level);
5958 btrfs_free_path(path);
5963 * helper to account the unused space of all the readonly block group in the
5964 * space_info. takes mirrors into account.
5966 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
5968 struct btrfs_block_group *block_group;
5972 /* It's df, we don't care if it's racy */
5973 if (list_empty(&sinfo->ro_bgs))
5976 spin_lock(&sinfo->lock);
5977 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
5978 spin_lock(&block_group->lock);
5980 if (!block_group->ro) {
5981 spin_unlock(&block_group->lock);
5985 factor = btrfs_bg_type_to_factor(block_group->flags);
5986 free_bytes += (block_group->length -
5987 block_group->used) * factor;
5989 spin_unlock(&block_group->lock);
5991 spin_unlock(&sinfo->lock);
5996 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5999 return unpin_extent_range(fs_info, start, end, false);
6003 * It used to be that old block groups would be left around forever.
6004 * Iterating over them would be enough to trim unused space. Since we
6005 * now automatically remove them, we also need to iterate over unallocated
6008 * We don't want a transaction for this since the discard may take a
6009 * substantial amount of time. We don't require that a transaction be
6010 * running, but we do need to take a running transaction into account
6011 * to ensure that we're not discarding chunks that were released or
6012 * allocated in the current transaction.
6014 * Holding the chunks lock will prevent other threads from allocating
6015 * or releasing chunks, but it won't prevent a running transaction
6016 * from committing and releasing the memory that the pending chunks
6017 * list head uses. For that, we need to take a reference to the
6018 * transaction and hold the commit root sem. We only need to hold
6019 * it while performing the free space search since we have already
6020 * held back allocations.
6022 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
6024 u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0;
6029 /* Discard not supported = nothing to do. */
6030 if (!bdev_max_discard_sectors(device->bdev))
6033 /* Not writable = nothing to do. */
6034 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
6037 /* No free space = nothing to do. */
6038 if (device->total_bytes <= device->bytes_used)
6044 struct btrfs_fs_info *fs_info = device->fs_info;
6047 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
6051 find_first_clear_extent_bit(&device->alloc_state, start,
6053 CHUNK_TRIMMED | CHUNK_ALLOCATED);
6055 /* Check if there are any CHUNK_* bits left */
6056 if (start > device->total_bytes) {
6057 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6058 btrfs_warn_in_rcu(fs_info,
6059 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
6060 start, end - start + 1,
6061 rcu_str_deref(device->name),
6062 device->total_bytes);
6063 mutex_unlock(&fs_info->chunk_mutex);
6068 /* Ensure we skip the reserved space on each device. */
6069 start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
6072 * If find_first_clear_extent_bit find a range that spans the
6073 * end of the device it will set end to -1, in this case it's up
6074 * to the caller to trim the value to the size of the device.
6076 end = min(end, device->total_bytes - 1);
6078 len = end - start + 1;
6080 /* We didn't find any extents */
6082 mutex_unlock(&fs_info->chunk_mutex);
6087 ret = btrfs_issue_discard(device->bdev, start, len,
6090 set_extent_bits(&device->alloc_state, start,
6093 mutex_unlock(&fs_info->chunk_mutex);
6101 if (fatal_signal_pending(current)) {
6113 * Trim the whole filesystem by:
6114 * 1) trimming the free space in each block group
6115 * 2) trimming the unallocated space on each device
6117 * This will also continue trimming even if a block group or device encounters
6118 * an error. The return value will be the last error, or 0 if nothing bad
6121 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6123 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6124 struct btrfs_block_group *cache = NULL;
6125 struct btrfs_device *device;
6127 u64 range_end = U64_MAX;
6137 if (range->start == U64_MAX)
6141 * Check range overflow if range->len is set.
6142 * The default range->len is U64_MAX.
6144 if (range->len != U64_MAX &&
6145 check_add_overflow(range->start, range->len, &range_end))
6148 cache = btrfs_lookup_first_block_group(fs_info, range->start);
6149 for (; cache; cache = btrfs_next_block_group(cache)) {
6150 if (cache->start >= range_end) {
6151 btrfs_put_block_group(cache);
6155 start = max(range->start, cache->start);
6156 end = min(range_end, cache->start + cache->length);
6158 if (end - start >= range->minlen) {
6159 if (!btrfs_block_group_done(cache)) {
6160 ret = btrfs_cache_block_group(cache, true);
6167 ret = btrfs_trim_block_group(cache,
6173 trimmed += group_trimmed;
6184 "failed to trim %llu block group(s), last error %d",
6187 mutex_lock(&fs_devices->device_list_mutex);
6188 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6189 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6192 ret = btrfs_trim_free_extents(device, &group_trimmed);
6199 trimmed += group_trimmed;
6201 mutex_unlock(&fs_devices->device_list_mutex);
6205 "failed to trim %llu device(s), last error %d",
6206 dev_failed, dev_ret);
6207 range->len = trimmed;