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_io_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;
1319 struct btrfs_io_context *bioc = NULL;
1322 * Avoid races with device replace and make sure our bioc has devices
1323 * associated to its stripes that don't go away while we are discarding.
1325 btrfs_bio_counter_inc_blocked(fs_info);
1327 struct btrfs_io_stripe *stripe;
1330 num_bytes = end - cur;
1331 /* Tell the block device(s) that the sectors can be discarded */
1332 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur,
1333 &num_bytes, &bioc, 0);
1335 * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or
1336 * -EOPNOTSUPP. For any such error, @num_bytes is not updated,
1337 * thus we can't continue anyway.
1342 stripe = bioc->stripes;
1343 for (i = 0; i < bioc->num_stripes; i++, stripe++) {
1345 struct btrfs_device *device = stripe->dev;
1347 if (!device->bdev) {
1348 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1352 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
1355 ret = do_discard_extent(stripe, &bytes);
1357 discarded_bytes += bytes;
1358 } else if (ret != -EOPNOTSUPP) {
1360 * Logic errors or -ENOMEM, or -EIO, but
1361 * unlikely to happen.
1363 * And since there are two loops, explicitly
1364 * go to out to avoid confusion.
1366 btrfs_put_bioc(bioc);
1371 * Just in case we get back EOPNOTSUPP for some reason,
1372 * just ignore the return value so we don't screw up
1373 * people calling discard_extent.
1377 btrfs_put_bioc(bioc);
1381 btrfs_bio_counter_dec(fs_info);
1384 *actual_bytes = discarded_bytes;
1387 if (ret == -EOPNOTSUPP)
1392 /* Can return -ENOMEM */
1393 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1394 struct btrfs_ref *generic_ref)
1396 struct btrfs_fs_info *fs_info = trans->fs_info;
1399 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1400 generic_ref->action);
1401 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1402 generic_ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID);
1404 if (generic_ref->type == BTRFS_REF_METADATA)
1405 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1407 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1409 btrfs_ref_tree_mod(fs_info, generic_ref);
1415 * __btrfs_inc_extent_ref - insert backreference for a given extent
1417 * The counterpart is in __btrfs_free_extent(), with examples and more details
1420 * @trans: Handle of transaction
1422 * @node: The delayed ref node used to get the bytenr/length for
1423 * extent whose references are incremented.
1425 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1426 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1427 * bytenr of the parent block. Since new extents are always
1428 * created with indirect references, this will only be the case
1429 * when relocating a shared extent. In that case, root_objectid
1430 * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1433 * @root_objectid: The id of the root where this modification has originated,
1434 * this can be either one of the well-known metadata trees or
1435 * the subvolume id which references this extent.
1437 * @owner: For data extents it is the inode number of the owning file.
1438 * For metadata extents this parameter holds the level in the
1439 * tree of the extent.
1441 * @offset: For metadata extents the offset is ignored and is currently
1442 * always passed as 0. For data extents it is the fileoffset
1443 * this extent belongs to.
1445 * @refs_to_add Number of references to add
1447 * @extent_op Pointer to a structure, holding information necessary when
1448 * updating a tree block's flags
1451 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1452 struct btrfs_delayed_ref_node *node,
1453 u64 parent, u64 root_objectid,
1454 u64 owner, u64 offset, int refs_to_add,
1455 struct btrfs_delayed_extent_op *extent_op)
1457 struct btrfs_path *path;
1458 struct extent_buffer *leaf;
1459 struct btrfs_extent_item *item;
1460 struct btrfs_key key;
1461 u64 bytenr = node->bytenr;
1462 u64 num_bytes = node->num_bytes;
1466 path = btrfs_alloc_path();
1470 /* this will setup the path even if it fails to insert the back ref */
1471 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1472 parent, root_objectid, owner,
1473 offset, refs_to_add, extent_op);
1474 if ((ret < 0 && ret != -EAGAIN) || !ret)
1478 * Ok we had -EAGAIN which means we didn't have space to insert and
1479 * inline extent ref, so just update the reference count and add a
1482 leaf = path->nodes[0];
1483 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1484 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1485 refs = btrfs_extent_refs(leaf, item);
1486 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1488 __run_delayed_extent_op(extent_op, leaf, item);
1490 btrfs_mark_buffer_dirty(leaf);
1491 btrfs_release_path(path);
1493 /* now insert the actual backref */
1494 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1495 BUG_ON(refs_to_add != 1);
1496 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1499 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1500 root_objectid, owner, offset,
1504 btrfs_abort_transaction(trans, ret);
1506 btrfs_free_path(path);
1510 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1511 struct btrfs_delayed_ref_node *node,
1512 struct btrfs_delayed_extent_op *extent_op,
1513 int insert_reserved)
1516 struct btrfs_delayed_data_ref *ref;
1517 struct btrfs_key ins;
1522 ins.objectid = node->bytenr;
1523 ins.offset = node->num_bytes;
1524 ins.type = BTRFS_EXTENT_ITEM_KEY;
1526 ref = btrfs_delayed_node_to_data_ref(node);
1527 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1529 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1530 parent = ref->parent;
1531 ref_root = ref->root;
1533 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1535 flags |= extent_op->flags_to_set;
1536 ret = alloc_reserved_file_extent(trans, parent, ref_root,
1537 flags, ref->objectid,
1540 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1541 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1542 ref->objectid, ref->offset,
1543 node->ref_mod, extent_op);
1544 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1545 ret = __btrfs_free_extent(trans, node, parent,
1546 ref_root, ref->objectid,
1547 ref->offset, node->ref_mod,
1555 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1556 struct extent_buffer *leaf,
1557 struct btrfs_extent_item *ei)
1559 u64 flags = btrfs_extent_flags(leaf, ei);
1560 if (extent_op->update_flags) {
1561 flags |= extent_op->flags_to_set;
1562 btrfs_set_extent_flags(leaf, ei, flags);
1565 if (extent_op->update_key) {
1566 struct btrfs_tree_block_info *bi;
1567 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1568 bi = (struct btrfs_tree_block_info *)(ei + 1);
1569 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1573 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1574 struct btrfs_delayed_ref_head *head,
1575 struct btrfs_delayed_extent_op *extent_op)
1577 struct btrfs_fs_info *fs_info = trans->fs_info;
1578 struct btrfs_root *root;
1579 struct btrfs_key key;
1580 struct btrfs_path *path;
1581 struct btrfs_extent_item *ei;
1582 struct extent_buffer *leaf;
1588 if (TRANS_ABORTED(trans))
1591 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1594 path = btrfs_alloc_path();
1598 key.objectid = head->bytenr;
1601 key.type = BTRFS_METADATA_ITEM_KEY;
1602 key.offset = extent_op->level;
1604 key.type = BTRFS_EXTENT_ITEM_KEY;
1605 key.offset = head->num_bytes;
1608 root = btrfs_extent_root(fs_info, key.objectid);
1610 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1617 if (path->slots[0] > 0) {
1619 btrfs_item_key_to_cpu(path->nodes[0], &key,
1621 if (key.objectid == head->bytenr &&
1622 key.type == BTRFS_EXTENT_ITEM_KEY &&
1623 key.offset == head->num_bytes)
1627 btrfs_release_path(path);
1630 key.objectid = head->bytenr;
1631 key.offset = head->num_bytes;
1632 key.type = BTRFS_EXTENT_ITEM_KEY;
1641 leaf = path->nodes[0];
1642 item_size = btrfs_item_size(leaf, path->slots[0]);
1644 if (unlikely(item_size < sizeof(*ei))) {
1646 btrfs_print_v0_err(fs_info);
1647 btrfs_abort_transaction(trans, err);
1651 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1652 __run_delayed_extent_op(extent_op, leaf, ei);
1654 btrfs_mark_buffer_dirty(leaf);
1656 btrfs_free_path(path);
1660 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1661 struct btrfs_delayed_ref_node *node,
1662 struct btrfs_delayed_extent_op *extent_op,
1663 int insert_reserved)
1666 struct btrfs_delayed_tree_ref *ref;
1670 ref = btrfs_delayed_node_to_tree_ref(node);
1671 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1673 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1674 parent = ref->parent;
1675 ref_root = ref->root;
1677 if (node->ref_mod != 1) {
1678 btrfs_err(trans->fs_info,
1679 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
1680 node->bytenr, node->ref_mod, node->action, ref_root,
1684 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1685 BUG_ON(!extent_op || !extent_op->update_flags);
1686 ret = alloc_reserved_tree_block(trans, node, extent_op);
1687 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1688 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1689 ref->level, 0, 1, extent_op);
1690 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1691 ret = __btrfs_free_extent(trans, node, parent, ref_root,
1692 ref->level, 0, 1, extent_op);
1699 /* helper function to actually process a single delayed ref entry */
1700 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1701 struct btrfs_delayed_ref_node *node,
1702 struct btrfs_delayed_extent_op *extent_op,
1703 int insert_reserved)
1707 if (TRANS_ABORTED(trans)) {
1708 if (insert_reserved)
1709 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1713 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1714 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1715 ret = run_delayed_tree_ref(trans, node, extent_op,
1717 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1718 node->type == BTRFS_SHARED_DATA_REF_KEY)
1719 ret = run_delayed_data_ref(trans, node, extent_op,
1723 if (ret && insert_reserved)
1724 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1728 static inline struct btrfs_delayed_ref_node *
1729 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1731 struct btrfs_delayed_ref_node *ref;
1733 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1737 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1738 * This is to prevent a ref count from going down to zero, which deletes
1739 * the extent item from the extent tree, when there still are references
1740 * to add, which would fail because they would not find the extent item.
1742 if (!list_empty(&head->ref_add_list))
1743 return list_first_entry(&head->ref_add_list,
1744 struct btrfs_delayed_ref_node, add_list);
1746 ref = rb_entry(rb_first_cached(&head->ref_tree),
1747 struct btrfs_delayed_ref_node, ref_node);
1748 ASSERT(list_empty(&ref->add_list));
1752 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1753 struct btrfs_delayed_ref_head *head)
1755 spin_lock(&delayed_refs->lock);
1756 head->processing = 0;
1757 delayed_refs->num_heads_ready++;
1758 spin_unlock(&delayed_refs->lock);
1759 btrfs_delayed_ref_unlock(head);
1762 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1763 struct btrfs_delayed_ref_head *head)
1765 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1770 if (head->must_insert_reserved) {
1771 head->extent_op = NULL;
1772 btrfs_free_delayed_extent_op(extent_op);
1778 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1779 struct btrfs_delayed_ref_head *head)
1781 struct btrfs_delayed_extent_op *extent_op;
1784 extent_op = cleanup_extent_op(head);
1787 head->extent_op = NULL;
1788 spin_unlock(&head->lock);
1789 ret = run_delayed_extent_op(trans, head, extent_op);
1790 btrfs_free_delayed_extent_op(extent_op);
1791 return ret ? ret : 1;
1794 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1795 struct btrfs_delayed_ref_root *delayed_refs,
1796 struct btrfs_delayed_ref_head *head)
1798 int nr_items = 1; /* Dropping this ref head update. */
1801 * We had csum deletions accounted for in our delayed refs rsv, we need
1802 * to drop the csum leaves for this update from our delayed_refs_rsv.
1804 if (head->total_ref_mod < 0 && head->is_data) {
1805 spin_lock(&delayed_refs->lock);
1806 delayed_refs->pending_csums -= head->num_bytes;
1807 spin_unlock(&delayed_refs->lock);
1808 nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1811 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1814 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1815 struct btrfs_delayed_ref_head *head)
1818 struct btrfs_fs_info *fs_info = trans->fs_info;
1819 struct btrfs_delayed_ref_root *delayed_refs;
1822 delayed_refs = &trans->transaction->delayed_refs;
1824 ret = run_and_cleanup_extent_op(trans, head);
1826 unselect_delayed_ref_head(delayed_refs, head);
1827 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1834 * Need to drop our head ref lock and re-acquire the delayed ref lock
1835 * and then re-check to make sure nobody got added.
1837 spin_unlock(&head->lock);
1838 spin_lock(&delayed_refs->lock);
1839 spin_lock(&head->lock);
1840 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1841 spin_unlock(&head->lock);
1842 spin_unlock(&delayed_refs->lock);
1845 btrfs_delete_ref_head(delayed_refs, head);
1846 spin_unlock(&head->lock);
1847 spin_unlock(&delayed_refs->lock);
1849 if (head->must_insert_reserved) {
1850 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1851 if (head->is_data) {
1852 struct btrfs_root *csum_root;
1854 csum_root = btrfs_csum_root(fs_info, head->bytenr);
1855 ret = btrfs_del_csums(trans, csum_root, head->bytenr,
1860 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1862 trace_run_delayed_ref_head(fs_info, head, 0);
1863 btrfs_delayed_ref_unlock(head);
1864 btrfs_put_delayed_ref_head(head);
1868 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1869 struct btrfs_trans_handle *trans)
1871 struct btrfs_delayed_ref_root *delayed_refs =
1872 &trans->transaction->delayed_refs;
1873 struct btrfs_delayed_ref_head *head = NULL;
1876 spin_lock(&delayed_refs->lock);
1877 head = btrfs_select_ref_head(delayed_refs);
1879 spin_unlock(&delayed_refs->lock);
1884 * Grab the lock that says we are going to process all the refs for
1887 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1888 spin_unlock(&delayed_refs->lock);
1891 * We may have dropped the spin lock to get the head mutex lock, and
1892 * that might have given someone else time to free the head. If that's
1893 * true, it has been removed from our list and we can move on.
1896 head = ERR_PTR(-EAGAIN);
1901 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1902 struct btrfs_delayed_ref_head *locked_ref,
1903 unsigned long *run_refs)
1905 struct btrfs_fs_info *fs_info = trans->fs_info;
1906 struct btrfs_delayed_ref_root *delayed_refs;
1907 struct btrfs_delayed_extent_op *extent_op;
1908 struct btrfs_delayed_ref_node *ref;
1909 int must_insert_reserved = 0;
1912 delayed_refs = &trans->transaction->delayed_refs;
1914 lockdep_assert_held(&locked_ref->mutex);
1915 lockdep_assert_held(&locked_ref->lock);
1917 while ((ref = select_delayed_ref(locked_ref))) {
1919 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1920 spin_unlock(&locked_ref->lock);
1921 unselect_delayed_ref_head(delayed_refs, locked_ref);
1927 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1928 RB_CLEAR_NODE(&ref->ref_node);
1929 if (!list_empty(&ref->add_list))
1930 list_del(&ref->add_list);
1932 * When we play the delayed ref, also correct the ref_mod on
1935 switch (ref->action) {
1936 case BTRFS_ADD_DELAYED_REF:
1937 case BTRFS_ADD_DELAYED_EXTENT:
1938 locked_ref->ref_mod -= ref->ref_mod;
1940 case BTRFS_DROP_DELAYED_REF:
1941 locked_ref->ref_mod += ref->ref_mod;
1946 atomic_dec(&delayed_refs->num_entries);
1949 * Record the must_insert_reserved flag before we drop the
1952 must_insert_reserved = locked_ref->must_insert_reserved;
1953 locked_ref->must_insert_reserved = 0;
1955 extent_op = locked_ref->extent_op;
1956 locked_ref->extent_op = NULL;
1957 spin_unlock(&locked_ref->lock);
1959 ret = run_one_delayed_ref(trans, ref, extent_op,
1960 must_insert_reserved);
1962 btrfs_free_delayed_extent_op(extent_op);
1964 unselect_delayed_ref_head(delayed_refs, locked_ref);
1965 btrfs_put_delayed_ref(ref);
1966 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
1971 btrfs_put_delayed_ref(ref);
1974 spin_lock(&locked_ref->lock);
1975 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
1982 * Returns 0 on success or if called with an already aborted transaction.
1983 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
1985 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1988 struct btrfs_fs_info *fs_info = trans->fs_info;
1989 struct btrfs_delayed_ref_root *delayed_refs;
1990 struct btrfs_delayed_ref_head *locked_ref = NULL;
1991 ktime_t start = ktime_get();
1993 unsigned long count = 0;
1994 unsigned long actual_count = 0;
1996 delayed_refs = &trans->transaction->delayed_refs;
1999 locked_ref = btrfs_obtain_ref_head(trans);
2000 if (IS_ERR_OR_NULL(locked_ref)) {
2001 if (PTR_ERR(locked_ref) == -EAGAIN) {
2010 * We need to try and merge add/drops of the same ref since we
2011 * can run into issues with relocate dropping the implicit ref
2012 * and then it being added back again before the drop can
2013 * finish. If we merged anything we need to re-loop so we can
2015 * Or we can get node references of the same type that weren't
2016 * merged when created due to bumps in the tree mod seq, and
2017 * we need to merge them to prevent adding an inline extent
2018 * backref before dropping it (triggering a BUG_ON at
2019 * insert_inline_extent_backref()).
2021 spin_lock(&locked_ref->lock);
2022 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2024 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2026 if (ret < 0 && ret != -EAGAIN) {
2028 * Error, btrfs_run_delayed_refs_for_head already
2029 * unlocked everything so just bail out
2034 * Success, perform the usual cleanup of a processed
2037 ret = cleanup_ref_head(trans, locked_ref);
2039 /* We dropped our lock, we need to loop. */
2048 * Either success case or btrfs_run_delayed_refs_for_head
2049 * returned -EAGAIN, meaning we need to select another head
2054 } while ((nr != -1 && count < nr) || locked_ref);
2057 * We don't want to include ref heads since we can have empty ref heads
2058 * and those will drastically skew our runtime down since we just do
2059 * accounting, no actual extent tree updates.
2061 if (actual_count > 0) {
2062 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2066 * We weigh the current average higher than our current runtime
2067 * to avoid large swings in the average.
2069 spin_lock(&delayed_refs->lock);
2070 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2071 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2072 spin_unlock(&delayed_refs->lock);
2077 #ifdef SCRAMBLE_DELAYED_REFS
2079 * Normally delayed refs get processed in ascending bytenr order. This
2080 * correlates in most cases to the order added. To expose dependencies on this
2081 * order, we start to process the tree in the middle instead of the beginning
2083 static u64 find_middle(struct rb_root *root)
2085 struct rb_node *n = root->rb_node;
2086 struct btrfs_delayed_ref_node *entry;
2089 u64 first = 0, last = 0;
2093 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2094 first = entry->bytenr;
2098 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2099 last = entry->bytenr;
2104 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2105 WARN_ON(!entry->in_tree);
2107 middle = entry->bytenr;
2121 * this starts processing the delayed reference count updates and
2122 * extent insertions we have queued up so far. count can be
2123 * 0, which means to process everything in the tree at the start
2124 * of the run (but not newly added entries), or it can be some target
2125 * number you'd like to process.
2127 * Returns 0 on success or if called with an aborted transaction
2128 * Returns <0 on error and aborts the transaction
2130 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2131 unsigned long count)
2133 struct btrfs_fs_info *fs_info = trans->fs_info;
2134 struct rb_node *node;
2135 struct btrfs_delayed_ref_root *delayed_refs;
2136 struct btrfs_delayed_ref_head *head;
2138 int run_all = count == (unsigned long)-1;
2140 /* We'll clean this up in btrfs_cleanup_transaction */
2141 if (TRANS_ABORTED(trans))
2144 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2147 delayed_refs = &trans->transaction->delayed_refs;
2149 count = delayed_refs->num_heads_ready;
2152 #ifdef SCRAMBLE_DELAYED_REFS
2153 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2155 ret = __btrfs_run_delayed_refs(trans, count);
2157 btrfs_abort_transaction(trans, ret);
2162 btrfs_create_pending_block_groups(trans);
2164 spin_lock(&delayed_refs->lock);
2165 node = rb_first_cached(&delayed_refs->href_root);
2167 spin_unlock(&delayed_refs->lock);
2170 head = rb_entry(node, struct btrfs_delayed_ref_head,
2172 refcount_inc(&head->refs);
2173 spin_unlock(&delayed_refs->lock);
2175 /* Mutex was contended, block until it's released and retry. */
2176 mutex_lock(&head->mutex);
2177 mutex_unlock(&head->mutex);
2179 btrfs_put_delayed_ref_head(head);
2187 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2188 struct extent_buffer *eb, u64 flags,
2191 struct btrfs_delayed_extent_op *extent_op;
2194 extent_op = btrfs_alloc_delayed_extent_op();
2198 extent_op->flags_to_set = flags;
2199 extent_op->update_flags = true;
2200 extent_op->update_key = false;
2201 extent_op->level = level;
2203 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2205 btrfs_free_delayed_extent_op(extent_op);
2209 static noinline int check_delayed_ref(struct btrfs_root *root,
2210 struct btrfs_path *path,
2211 u64 objectid, u64 offset, u64 bytenr)
2213 struct btrfs_delayed_ref_head *head;
2214 struct btrfs_delayed_ref_node *ref;
2215 struct btrfs_delayed_data_ref *data_ref;
2216 struct btrfs_delayed_ref_root *delayed_refs;
2217 struct btrfs_transaction *cur_trans;
2218 struct rb_node *node;
2221 spin_lock(&root->fs_info->trans_lock);
2222 cur_trans = root->fs_info->running_transaction;
2224 refcount_inc(&cur_trans->use_count);
2225 spin_unlock(&root->fs_info->trans_lock);
2229 delayed_refs = &cur_trans->delayed_refs;
2230 spin_lock(&delayed_refs->lock);
2231 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2233 spin_unlock(&delayed_refs->lock);
2234 btrfs_put_transaction(cur_trans);
2238 if (!mutex_trylock(&head->mutex)) {
2239 refcount_inc(&head->refs);
2240 spin_unlock(&delayed_refs->lock);
2242 btrfs_release_path(path);
2245 * Mutex was contended, block until it's released and let
2248 mutex_lock(&head->mutex);
2249 mutex_unlock(&head->mutex);
2250 btrfs_put_delayed_ref_head(head);
2251 btrfs_put_transaction(cur_trans);
2254 spin_unlock(&delayed_refs->lock);
2256 spin_lock(&head->lock);
2258 * XXX: We should replace this with a proper search function in the
2261 for (node = rb_first_cached(&head->ref_tree); node;
2262 node = rb_next(node)) {
2263 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2264 /* If it's a shared ref we know a cross reference exists */
2265 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2270 data_ref = btrfs_delayed_node_to_data_ref(ref);
2273 * If our ref doesn't match the one we're currently looking at
2274 * then we have a cross reference.
2276 if (data_ref->root != root->root_key.objectid ||
2277 data_ref->objectid != objectid ||
2278 data_ref->offset != offset) {
2283 spin_unlock(&head->lock);
2284 mutex_unlock(&head->mutex);
2285 btrfs_put_transaction(cur_trans);
2289 static noinline int check_committed_ref(struct btrfs_root *root,
2290 struct btrfs_path *path,
2291 u64 objectid, u64 offset, u64 bytenr,
2294 struct btrfs_fs_info *fs_info = root->fs_info;
2295 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2296 struct extent_buffer *leaf;
2297 struct btrfs_extent_data_ref *ref;
2298 struct btrfs_extent_inline_ref *iref;
2299 struct btrfs_extent_item *ei;
2300 struct btrfs_key key;
2305 key.objectid = bytenr;
2306 key.offset = (u64)-1;
2307 key.type = BTRFS_EXTENT_ITEM_KEY;
2309 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2312 BUG_ON(ret == 0); /* Corruption */
2315 if (path->slots[0] == 0)
2319 leaf = path->nodes[0];
2320 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2322 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2326 item_size = btrfs_item_size(leaf, path->slots[0]);
2327 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2329 /* If extent item has more than 1 inline ref then it's shared */
2330 if (item_size != sizeof(*ei) +
2331 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2335 * If extent created before last snapshot => it's shared unless the
2336 * snapshot has been deleted. Use the heuristic if strict is false.
2339 (btrfs_extent_generation(leaf, ei) <=
2340 btrfs_root_last_snapshot(&root->root_item)))
2343 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2345 /* If this extent has SHARED_DATA_REF then it's shared */
2346 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2347 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2350 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2351 if (btrfs_extent_refs(leaf, ei) !=
2352 btrfs_extent_data_ref_count(leaf, ref) ||
2353 btrfs_extent_data_ref_root(leaf, ref) !=
2354 root->root_key.objectid ||
2355 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2356 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2364 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2365 u64 bytenr, bool strict, struct btrfs_path *path)
2370 ret = check_committed_ref(root, path, objectid,
2371 offset, bytenr, strict);
2372 if (ret && ret != -ENOENT)
2375 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2376 } while (ret == -EAGAIN);
2379 btrfs_release_path(path);
2380 if (btrfs_is_data_reloc_root(root))
2385 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2386 struct btrfs_root *root,
2387 struct extent_buffer *buf,
2388 int full_backref, int inc)
2390 struct btrfs_fs_info *fs_info = root->fs_info;
2396 struct btrfs_key key;
2397 struct btrfs_file_extent_item *fi;
2398 struct btrfs_ref generic_ref = { 0 };
2399 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2405 if (btrfs_is_testing(fs_info))
2408 ref_root = btrfs_header_owner(buf);
2409 nritems = btrfs_header_nritems(buf);
2410 level = btrfs_header_level(buf);
2412 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2416 parent = buf->start;
2420 action = BTRFS_ADD_DELAYED_REF;
2422 action = BTRFS_DROP_DELAYED_REF;
2424 for (i = 0; i < nritems; i++) {
2426 btrfs_item_key_to_cpu(buf, &key, i);
2427 if (key.type != BTRFS_EXTENT_DATA_KEY)
2429 fi = btrfs_item_ptr(buf, i,
2430 struct btrfs_file_extent_item);
2431 if (btrfs_file_extent_type(buf, fi) ==
2432 BTRFS_FILE_EXTENT_INLINE)
2434 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2438 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2439 key.offset -= btrfs_file_extent_offset(buf, fi);
2440 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2442 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2443 key.offset, root->root_key.objectid,
2446 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2448 ret = btrfs_free_extent(trans, &generic_ref);
2452 bytenr = btrfs_node_blockptr(buf, i);
2453 num_bytes = fs_info->nodesize;
2454 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2456 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root,
2457 root->root_key.objectid, for_reloc);
2459 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2461 ret = btrfs_free_extent(trans, &generic_ref);
2471 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2472 struct extent_buffer *buf, int full_backref)
2474 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2477 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2478 struct extent_buffer *buf, int full_backref)
2480 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2483 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2485 struct btrfs_fs_info *fs_info = root->fs_info;
2490 flags = BTRFS_BLOCK_GROUP_DATA;
2491 else if (root == fs_info->chunk_root)
2492 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2494 flags = BTRFS_BLOCK_GROUP_METADATA;
2496 ret = btrfs_get_alloc_profile(fs_info, flags);
2500 static u64 first_logical_byte(struct btrfs_fs_info *fs_info)
2502 struct rb_node *leftmost;
2505 read_lock(&fs_info->block_group_cache_lock);
2506 /* Get the block group with the lowest logical start address. */
2507 leftmost = rb_first_cached(&fs_info->block_group_cache_tree);
2509 struct btrfs_block_group *bg;
2511 bg = rb_entry(leftmost, struct btrfs_block_group, cache_node);
2514 read_unlock(&fs_info->block_group_cache_lock);
2519 static int pin_down_extent(struct btrfs_trans_handle *trans,
2520 struct btrfs_block_group *cache,
2521 u64 bytenr, u64 num_bytes, int reserved)
2523 struct btrfs_fs_info *fs_info = cache->fs_info;
2525 spin_lock(&cache->space_info->lock);
2526 spin_lock(&cache->lock);
2527 cache->pinned += num_bytes;
2528 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2531 cache->reserved -= num_bytes;
2532 cache->space_info->bytes_reserved -= num_bytes;
2534 spin_unlock(&cache->lock);
2535 spin_unlock(&cache->space_info->lock);
2537 set_extent_dirty(&trans->transaction->pinned_extents, bytenr,
2538 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
2542 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2543 u64 bytenr, u64 num_bytes, int reserved)
2545 struct btrfs_block_group *cache;
2547 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2548 BUG_ON(!cache); /* Logic error */
2550 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2552 btrfs_put_block_group(cache);
2557 * this function must be called within transaction
2559 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2560 u64 bytenr, u64 num_bytes)
2562 struct btrfs_block_group *cache;
2565 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2570 * pull in the free space cache (if any) so that our pin
2571 * removes the free space from the cache. We have load_only set
2572 * to one because the slow code to read in the free extents does check
2573 * the pinned extents.
2575 btrfs_cache_block_group(cache, 1);
2577 * Make sure we wait until the cache is completely built in case it is
2578 * missing or is invalid and therefore needs to be rebuilt.
2580 ret = btrfs_wait_block_group_cache_done(cache);
2584 pin_down_extent(trans, cache, bytenr, num_bytes, 0);
2586 /* remove us from the free space cache (if we're there at all) */
2587 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2589 btrfs_put_block_group(cache);
2593 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2594 u64 start, u64 num_bytes)
2597 struct btrfs_block_group *block_group;
2599 block_group = btrfs_lookup_block_group(fs_info, start);
2603 btrfs_cache_block_group(block_group, 1);
2605 * Make sure we wait until the cache is completely built in case it is
2606 * missing or is invalid and therefore needs to be rebuilt.
2608 ret = btrfs_wait_block_group_cache_done(block_group);
2612 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2614 btrfs_put_block_group(block_group);
2618 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2620 struct btrfs_fs_info *fs_info = eb->fs_info;
2621 struct btrfs_file_extent_item *item;
2622 struct btrfs_key key;
2627 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2630 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2631 btrfs_item_key_to_cpu(eb, &key, i);
2632 if (key.type != BTRFS_EXTENT_DATA_KEY)
2634 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2635 found_type = btrfs_file_extent_type(eb, item);
2636 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2638 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2640 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2641 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2642 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2651 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2653 atomic_inc(&bg->reservations);
2657 * Returns the free cluster for the given space info and sets empty_cluster to
2658 * what it should be based on the mount options.
2660 static struct btrfs_free_cluster *
2661 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2662 struct btrfs_space_info *space_info, u64 *empty_cluster)
2664 struct btrfs_free_cluster *ret = NULL;
2667 if (btrfs_mixed_space_info(space_info))
2670 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2671 ret = &fs_info->meta_alloc_cluster;
2672 if (btrfs_test_opt(fs_info, SSD))
2673 *empty_cluster = SZ_2M;
2675 *empty_cluster = SZ_64K;
2676 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2677 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2678 *empty_cluster = SZ_2M;
2679 ret = &fs_info->data_alloc_cluster;
2685 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2687 const bool return_free_space)
2689 struct btrfs_block_group *cache = NULL;
2690 struct btrfs_space_info *space_info;
2691 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2692 struct btrfs_free_cluster *cluster = NULL;
2694 u64 total_unpinned = 0;
2695 u64 empty_cluster = 0;
2698 while (start <= end) {
2701 start >= cache->start + cache->length) {
2703 btrfs_put_block_group(cache);
2705 cache = btrfs_lookup_block_group(fs_info, start);
2706 BUG_ON(!cache); /* Logic error */
2708 cluster = fetch_cluster_info(fs_info,
2711 empty_cluster <<= 1;
2714 len = cache->start + cache->length - start;
2715 len = min(len, end + 1 - start);
2717 down_read(&fs_info->commit_root_sem);
2718 if (start < cache->last_byte_to_unpin && return_free_space) {
2719 u64 add_len = min(len, cache->last_byte_to_unpin - start);
2721 btrfs_add_free_space(cache, start, add_len);
2723 up_read(&fs_info->commit_root_sem);
2726 total_unpinned += len;
2727 space_info = cache->space_info;
2730 * If this space cluster has been marked as fragmented and we've
2731 * unpinned enough in this block group to potentially allow a
2732 * cluster to be created inside of it go ahead and clear the
2735 if (cluster && cluster->fragmented &&
2736 total_unpinned > empty_cluster) {
2737 spin_lock(&cluster->lock);
2738 cluster->fragmented = 0;
2739 spin_unlock(&cluster->lock);
2742 spin_lock(&space_info->lock);
2743 spin_lock(&cache->lock);
2744 cache->pinned -= len;
2745 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2746 space_info->max_extent_size = 0;
2748 space_info->bytes_readonly += len;
2750 } else if (btrfs_is_zoned(fs_info)) {
2751 /* Need reset before reusing in a zoned block group */
2752 space_info->bytes_zone_unusable += len;
2755 spin_unlock(&cache->lock);
2756 if (!readonly && return_free_space &&
2757 global_rsv->space_info == space_info) {
2758 spin_lock(&global_rsv->lock);
2759 if (!global_rsv->full) {
2760 u64 to_add = min(len, global_rsv->size -
2761 global_rsv->reserved);
2763 global_rsv->reserved += to_add;
2764 btrfs_space_info_update_bytes_may_use(fs_info,
2765 space_info, to_add);
2766 if (global_rsv->reserved >= global_rsv->size)
2767 global_rsv->full = 1;
2770 spin_unlock(&global_rsv->lock);
2772 /* Add to any tickets we may have */
2773 if (!readonly && return_free_space && len)
2774 btrfs_try_granting_tickets(fs_info, space_info);
2775 spin_unlock(&space_info->lock);
2779 btrfs_put_block_group(cache);
2783 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2785 struct btrfs_fs_info *fs_info = trans->fs_info;
2786 struct btrfs_block_group *block_group, *tmp;
2787 struct list_head *deleted_bgs;
2788 struct extent_io_tree *unpin;
2793 unpin = &trans->transaction->pinned_extents;
2795 while (!TRANS_ABORTED(trans)) {
2796 struct extent_state *cached_state = NULL;
2798 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2799 ret = find_first_extent_bit(unpin, 0, &start, &end,
2800 EXTENT_DIRTY, &cached_state);
2802 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2806 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2807 ret = btrfs_discard_extent(fs_info, start,
2808 end + 1 - start, NULL);
2810 clear_extent_dirty(unpin, start, end, &cached_state);
2811 unpin_extent_range(fs_info, start, end, true);
2812 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2813 free_extent_state(cached_state);
2817 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2818 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2819 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2823 * Transaction is finished. We don't need the lock anymore. We
2824 * do need to clean up the block groups in case of a transaction
2827 deleted_bgs = &trans->transaction->deleted_bgs;
2828 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2832 if (!TRANS_ABORTED(trans))
2833 ret = btrfs_discard_extent(fs_info,
2835 block_group->length,
2838 list_del_init(&block_group->bg_list);
2839 btrfs_unfreeze_block_group(block_group);
2840 btrfs_put_block_group(block_group);
2843 const char *errstr = btrfs_decode_error(ret);
2845 "discard failed while removing blockgroup: errno=%d %s",
2853 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
2854 u64 bytenr, u64 num_bytes, bool is_data)
2859 struct btrfs_root *csum_root;
2861 csum_root = btrfs_csum_root(trans->fs_info, bytenr);
2862 ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes);
2864 btrfs_abort_transaction(trans, ret);
2869 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
2871 btrfs_abort_transaction(trans, ret);
2875 ret = btrfs_update_block_group(trans, bytenr, num_bytes, false);
2877 btrfs_abort_transaction(trans, ret);
2883 * Drop one or more refs of @node.
2885 * 1. Locate the extent refs.
2886 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2887 * Locate it, then reduce the refs number or remove the ref line completely.
2889 * 2. Update the refs count in EXTENT/METADATA_ITEM
2891 * Inline backref case:
2893 * in extent tree we have:
2895 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2896 * refs 2 gen 6 flags DATA
2897 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2898 * extent data backref root FS_TREE objectid 257 offset 0 count 1
2900 * This function gets called with:
2902 * node->bytenr = 13631488
2903 * node->num_bytes = 1048576
2904 * root_objectid = FS_TREE
2905 * owner_objectid = 257
2909 * Then we should get some like:
2911 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2912 * refs 1 gen 6 flags DATA
2913 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2915 * Keyed backref case:
2917 * in extent tree we have:
2919 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2920 * refs 754 gen 6 flags DATA
2922 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2923 * extent data backref root FS_TREE objectid 866 offset 0 count 1
2925 * This function get called with:
2927 * node->bytenr = 13631488
2928 * node->num_bytes = 1048576
2929 * root_objectid = FS_TREE
2930 * owner_objectid = 866
2934 * Then we should get some like:
2936 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2937 * refs 753 gen 6 flags DATA
2939 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2941 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2942 struct btrfs_delayed_ref_node *node, u64 parent,
2943 u64 root_objectid, u64 owner_objectid,
2944 u64 owner_offset, int refs_to_drop,
2945 struct btrfs_delayed_extent_op *extent_op)
2947 struct btrfs_fs_info *info = trans->fs_info;
2948 struct btrfs_key key;
2949 struct btrfs_path *path;
2950 struct btrfs_root *extent_root;
2951 struct extent_buffer *leaf;
2952 struct btrfs_extent_item *ei;
2953 struct btrfs_extent_inline_ref *iref;
2956 int extent_slot = 0;
2957 int found_extent = 0;
2961 u64 bytenr = node->bytenr;
2962 u64 num_bytes = node->num_bytes;
2963 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2965 extent_root = btrfs_extent_root(info, bytenr);
2966 ASSERT(extent_root);
2968 path = btrfs_alloc_path();
2972 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2974 if (!is_data && refs_to_drop != 1) {
2976 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2977 node->bytenr, refs_to_drop);
2979 btrfs_abort_transaction(trans, ret);
2984 skinny_metadata = false;
2986 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2987 parent, root_objectid, owner_objectid,
2991 * Either the inline backref or the SHARED_DATA_REF/
2992 * SHARED_BLOCK_REF is found
2994 * Here is a quick path to locate EXTENT/METADATA_ITEM.
2995 * It's possible the EXTENT/METADATA_ITEM is near current slot.
2997 extent_slot = path->slots[0];
2998 while (extent_slot >= 0) {
2999 btrfs_item_key_to_cpu(path->nodes[0], &key,
3001 if (key.objectid != bytenr)
3003 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3004 key.offset == num_bytes) {
3008 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3009 key.offset == owner_objectid) {
3014 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
3015 if (path->slots[0] - extent_slot > 5)
3020 if (!found_extent) {
3023 "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref");
3024 btrfs_abort_transaction(trans, -EUCLEAN);
3027 /* Must be SHARED_* item, remove the backref first */
3028 ret = remove_extent_backref(trans, extent_root, path,
3029 NULL, refs_to_drop, is_data);
3031 btrfs_abort_transaction(trans, ret);
3034 btrfs_release_path(path);
3036 /* Slow path to locate EXTENT/METADATA_ITEM */
3037 key.objectid = bytenr;
3038 key.type = BTRFS_EXTENT_ITEM_KEY;
3039 key.offset = num_bytes;
3041 if (!is_data && skinny_metadata) {
3042 key.type = BTRFS_METADATA_ITEM_KEY;
3043 key.offset = owner_objectid;
3046 ret = btrfs_search_slot(trans, extent_root,
3048 if (ret > 0 && skinny_metadata && path->slots[0]) {
3050 * Couldn't find our skinny metadata item,
3051 * see if we have ye olde extent item.
3054 btrfs_item_key_to_cpu(path->nodes[0], &key,
3056 if (key.objectid == bytenr &&
3057 key.type == BTRFS_EXTENT_ITEM_KEY &&
3058 key.offset == num_bytes)
3062 if (ret > 0 && skinny_metadata) {
3063 skinny_metadata = false;
3064 key.objectid = bytenr;
3065 key.type = BTRFS_EXTENT_ITEM_KEY;
3066 key.offset = num_bytes;
3067 btrfs_release_path(path);
3068 ret = btrfs_search_slot(trans, extent_root,
3074 "umm, got %d back from search, was looking for %llu",
3077 btrfs_print_leaf(path->nodes[0]);
3080 btrfs_abort_transaction(trans, ret);
3083 extent_slot = path->slots[0];
3085 } else if (WARN_ON(ret == -ENOENT)) {
3086 btrfs_print_leaf(path->nodes[0]);
3088 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
3089 bytenr, parent, root_objectid, owner_objectid,
3091 btrfs_abort_transaction(trans, ret);
3094 btrfs_abort_transaction(trans, ret);
3098 leaf = path->nodes[0];
3099 item_size = btrfs_item_size(leaf, extent_slot);
3100 if (unlikely(item_size < sizeof(*ei))) {
3102 btrfs_print_v0_err(info);
3103 btrfs_abort_transaction(trans, ret);
3106 ei = btrfs_item_ptr(leaf, extent_slot,
3107 struct btrfs_extent_item);
3108 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3109 key.type == BTRFS_EXTENT_ITEM_KEY) {
3110 struct btrfs_tree_block_info *bi;
3111 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3113 "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu",
3114 key.objectid, key.type, key.offset,
3115 owner_objectid, item_size,
3116 sizeof(*ei) + sizeof(*bi));
3117 btrfs_abort_transaction(trans, -EUCLEAN);
3120 bi = (struct btrfs_tree_block_info *)(ei + 1);
3121 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3124 refs = btrfs_extent_refs(leaf, ei);
3125 if (refs < refs_to_drop) {
3127 "trying to drop %d refs but we only have %llu for bytenr %llu",
3128 refs_to_drop, refs, bytenr);
3129 btrfs_abort_transaction(trans, -EUCLEAN);
3132 refs -= refs_to_drop;
3136 __run_delayed_extent_op(extent_op, leaf, ei);
3138 * In the case of inline back ref, reference count will
3139 * be updated by remove_extent_backref
3142 if (!found_extent) {
3144 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found");
3145 btrfs_abort_transaction(trans, -EUCLEAN);
3149 btrfs_set_extent_refs(leaf, ei, refs);
3150 btrfs_mark_buffer_dirty(leaf);
3153 ret = remove_extent_backref(trans, extent_root, path,
3154 iref, refs_to_drop, is_data);
3156 btrfs_abort_transaction(trans, ret);
3161 /* In this branch refs == 1 */
3163 if (is_data && refs_to_drop !=
3164 extent_data_ref_count(path, iref)) {
3166 "invalid refs_to_drop, current refs %u refs_to_drop %u",
3167 extent_data_ref_count(path, iref),
3169 btrfs_abort_transaction(trans, -EUCLEAN);
3173 if (path->slots[0] != extent_slot) {
3175 "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref",
3176 key.objectid, key.type,
3178 btrfs_abort_transaction(trans, -EUCLEAN);
3183 * No inline ref, we must be at SHARED_* item,
3184 * And it's single ref, it must be:
3185 * | extent_slot ||extent_slot + 1|
3186 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3188 if (path->slots[0] != extent_slot + 1) {
3190 "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM");
3191 btrfs_abort_transaction(trans, -EUCLEAN);
3194 path->slots[0] = extent_slot;
3199 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3202 btrfs_abort_transaction(trans, ret);
3205 btrfs_release_path(path);
3207 ret = do_free_extent_accounting(trans, bytenr, num_bytes, is_data);
3209 btrfs_release_path(path);
3212 btrfs_free_path(path);
3216 * Leaf dump can take up a lot of log buffer, so we only do full leaf
3217 * dump for debug build.
3219 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) {
3220 btrfs_crit(info, "path->slots[0]=%d extent_slot=%d",
3221 path->slots[0], extent_slot);
3222 btrfs_print_leaf(path->nodes[0]);
3225 btrfs_free_path(path);
3230 * when we free an block, it is possible (and likely) that we free the last
3231 * delayed ref for that extent as well. This searches the delayed ref tree for
3232 * a given extent, and if there are no other delayed refs to be processed, it
3233 * removes it from the tree.
3235 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3238 struct btrfs_delayed_ref_head *head;
3239 struct btrfs_delayed_ref_root *delayed_refs;
3242 delayed_refs = &trans->transaction->delayed_refs;
3243 spin_lock(&delayed_refs->lock);
3244 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3246 goto out_delayed_unlock;
3248 spin_lock(&head->lock);
3249 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3252 if (cleanup_extent_op(head) != NULL)
3256 * waiting for the lock here would deadlock. If someone else has it
3257 * locked they are already in the process of dropping it anyway
3259 if (!mutex_trylock(&head->mutex))
3262 btrfs_delete_ref_head(delayed_refs, head);
3263 head->processing = 0;
3265 spin_unlock(&head->lock);
3266 spin_unlock(&delayed_refs->lock);
3268 BUG_ON(head->extent_op);
3269 if (head->must_insert_reserved)
3272 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3273 mutex_unlock(&head->mutex);
3274 btrfs_put_delayed_ref_head(head);
3277 spin_unlock(&head->lock);
3280 spin_unlock(&delayed_refs->lock);
3284 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3286 struct extent_buffer *buf,
3287 u64 parent, int last_ref)
3289 struct btrfs_fs_info *fs_info = trans->fs_info;
3290 struct btrfs_ref generic_ref = { 0 };
3293 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3294 buf->start, buf->len, parent);
3295 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3298 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3299 btrfs_ref_tree_mod(fs_info, &generic_ref);
3300 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3301 BUG_ON(ret); /* -ENOMEM */
3304 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3305 struct btrfs_block_group *cache;
3306 bool must_pin = false;
3308 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3309 ret = check_ref_cleanup(trans, buf->start);
3311 btrfs_redirty_list_add(trans->transaction, buf);
3316 cache = btrfs_lookup_block_group(fs_info, buf->start);
3318 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3319 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3320 btrfs_put_block_group(cache);
3325 * If this is a leaf and there are tree mod log users, we may
3326 * have recorded mod log operations that point to this leaf.
3327 * So we must make sure no one reuses this leaf's extent before
3328 * mod log operations are applied to a node, otherwise after
3329 * rewinding a node using the mod log operations we get an
3330 * inconsistent btree, as the leaf's extent may now be used as
3331 * a node or leaf for another different btree.
3332 * We are safe from races here because at this point no other
3333 * node or root points to this extent buffer, so if after this
3334 * check a new tree mod log user joins, it will not be able to
3335 * find a node pointing to this leaf and record operations that
3336 * point to this leaf.
3338 if (btrfs_header_level(buf) == 0 &&
3339 test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
3342 if (must_pin || btrfs_is_zoned(fs_info)) {
3343 btrfs_redirty_list_add(trans->transaction, buf);
3344 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3345 btrfs_put_block_group(cache);
3349 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3351 btrfs_add_free_space(cache, buf->start, buf->len);
3352 btrfs_free_reserved_bytes(cache, buf->len, 0);
3353 btrfs_put_block_group(cache);
3354 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3359 * Deleting the buffer, clear the corrupt flag since it doesn't
3362 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3366 /* Can return -ENOMEM */
3367 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3369 struct btrfs_fs_info *fs_info = trans->fs_info;
3372 if (btrfs_is_testing(fs_info))
3376 * tree log blocks never actually go into the extent allocation
3377 * tree, just update pinning info and exit early.
3379 if ((ref->type == BTRFS_REF_METADATA &&
3380 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3381 (ref->type == BTRFS_REF_DATA &&
3382 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)) {
3383 /* unlocks the pinned mutex */
3384 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3386 } else if (ref->type == BTRFS_REF_METADATA) {
3387 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3389 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3392 if (!((ref->type == BTRFS_REF_METADATA &&
3393 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3394 (ref->type == BTRFS_REF_DATA &&
3395 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)))
3396 btrfs_ref_tree_mod(fs_info, ref);
3401 enum btrfs_loop_type {
3402 LOOP_CACHING_NOWAIT,
3409 btrfs_lock_block_group(struct btrfs_block_group *cache,
3413 down_read(&cache->data_rwsem);
3416 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3419 btrfs_get_block_group(cache);
3421 down_read(&cache->data_rwsem);
3424 static struct btrfs_block_group *btrfs_lock_cluster(
3425 struct btrfs_block_group *block_group,
3426 struct btrfs_free_cluster *cluster,
3428 __acquires(&cluster->refill_lock)
3430 struct btrfs_block_group *used_bg = NULL;
3432 spin_lock(&cluster->refill_lock);
3434 used_bg = cluster->block_group;
3438 if (used_bg == block_group)
3441 btrfs_get_block_group(used_bg);
3446 if (down_read_trylock(&used_bg->data_rwsem))
3449 spin_unlock(&cluster->refill_lock);
3451 /* We should only have one-level nested. */
3452 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3454 spin_lock(&cluster->refill_lock);
3455 if (used_bg == cluster->block_group)
3458 up_read(&used_bg->data_rwsem);
3459 btrfs_put_block_group(used_bg);
3464 btrfs_release_block_group(struct btrfs_block_group *cache,
3468 up_read(&cache->data_rwsem);
3469 btrfs_put_block_group(cache);
3472 enum btrfs_extent_allocation_policy {
3473 BTRFS_EXTENT_ALLOC_CLUSTERED,
3474 BTRFS_EXTENT_ALLOC_ZONED,
3478 * Structure used internally for find_free_extent() function. Wraps needed
3481 struct find_free_extent_ctl {
3482 /* Basic allocation info */
3490 /* Where to start the search inside the bg */
3493 /* For clustered allocation */
3495 struct btrfs_free_cluster *last_ptr;
3498 bool have_caching_bg;
3499 bool orig_have_caching_bg;
3501 /* Allocation is called for tree-log */
3504 /* Allocation is called for data relocation */
3505 bool for_data_reloc;
3507 /* RAID index, converted from flags */
3511 * Current loop number, check find_free_extent_update_loop() for details
3516 * Whether we're refilling a cluster, if true we need to re-search
3517 * current block group but don't try to refill the cluster again.
3519 bool retry_clustered;
3522 * Whether we're updating free space cache, if true we need to re-search
3523 * current block group but don't try updating free space cache again.
3525 bool retry_unclustered;
3527 /* If current block group is cached */
3530 /* Max contiguous hole found */
3531 u64 max_extent_size;
3533 /* Total free space from free space cache, not always contiguous */
3534 u64 total_free_space;
3539 /* Hint where to start looking for an empty space */
3542 /* Allocation policy */
3543 enum btrfs_extent_allocation_policy policy;
3548 * Helper function for find_free_extent().
3550 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3551 * Return -EAGAIN to inform caller that we need to re-search this block group
3552 * Return >0 to inform caller that we find nothing
3553 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3555 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3556 struct find_free_extent_ctl *ffe_ctl,
3557 struct btrfs_block_group **cluster_bg_ret)
3559 struct btrfs_block_group *cluster_bg;
3560 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3561 u64 aligned_cluster;
3565 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3567 goto refill_cluster;
3568 if (cluster_bg != bg && (cluster_bg->ro ||
3569 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3570 goto release_cluster;
3572 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3573 ffe_ctl->num_bytes, cluster_bg->start,
3574 &ffe_ctl->max_extent_size);
3576 /* We have a block, we're done */
3577 spin_unlock(&last_ptr->refill_lock);
3578 trace_btrfs_reserve_extent_cluster(cluster_bg,
3579 ffe_ctl->search_start, ffe_ctl->num_bytes);
3580 *cluster_bg_ret = cluster_bg;
3581 ffe_ctl->found_offset = offset;
3584 WARN_ON(last_ptr->block_group != cluster_bg);
3588 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3589 * lets just skip it and let the allocator find whatever block it can
3590 * find. If we reach this point, we will have tried the cluster
3591 * allocator plenty of times and not have found anything, so we are
3592 * likely way too fragmented for the clustering stuff to find anything.
3594 * However, if the cluster is taken from the current block group,
3595 * release the cluster first, so that we stand a better chance of
3596 * succeeding in the unclustered allocation.
3598 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3599 spin_unlock(&last_ptr->refill_lock);
3600 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3604 /* This cluster didn't work out, free it and start over */
3605 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3607 if (cluster_bg != bg)
3608 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3611 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3612 spin_unlock(&last_ptr->refill_lock);
3616 aligned_cluster = max_t(u64,
3617 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3618 bg->full_stripe_len);
3619 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3620 ffe_ctl->num_bytes, aligned_cluster);
3622 /* Now pull our allocation out of this cluster */
3623 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3624 ffe_ctl->num_bytes, ffe_ctl->search_start,
3625 &ffe_ctl->max_extent_size);
3627 /* We found one, proceed */
3628 spin_unlock(&last_ptr->refill_lock);
3629 trace_btrfs_reserve_extent_cluster(bg,
3630 ffe_ctl->search_start,
3631 ffe_ctl->num_bytes);
3632 ffe_ctl->found_offset = offset;
3635 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3636 !ffe_ctl->retry_clustered) {
3637 spin_unlock(&last_ptr->refill_lock);
3639 ffe_ctl->retry_clustered = true;
3640 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3641 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3645 * At this point we either didn't find a cluster or we weren't able to
3646 * allocate a block from our cluster. Free the cluster we've been
3647 * trying to use, and go to the next block group.
3649 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3650 spin_unlock(&last_ptr->refill_lock);
3655 * Return >0 to inform caller that we find nothing
3656 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3657 * Return -EAGAIN to inform caller that we need to re-search this block group
3659 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3660 struct find_free_extent_ctl *ffe_ctl)
3662 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3666 * We are doing an unclustered allocation, set the fragmented flag so
3667 * we don't bother trying to setup a cluster again until we get more
3670 if (unlikely(last_ptr)) {
3671 spin_lock(&last_ptr->lock);
3672 last_ptr->fragmented = 1;
3673 spin_unlock(&last_ptr->lock);
3675 if (ffe_ctl->cached) {
3676 struct btrfs_free_space_ctl *free_space_ctl;
3678 free_space_ctl = bg->free_space_ctl;
3679 spin_lock(&free_space_ctl->tree_lock);
3680 if (free_space_ctl->free_space <
3681 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3682 ffe_ctl->empty_size) {
3683 ffe_ctl->total_free_space = max_t(u64,
3684 ffe_ctl->total_free_space,
3685 free_space_ctl->free_space);
3686 spin_unlock(&free_space_ctl->tree_lock);
3689 spin_unlock(&free_space_ctl->tree_lock);
3692 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3693 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3694 &ffe_ctl->max_extent_size);
3697 * If we didn't find a chunk, and we haven't failed on this block group
3698 * before, and this block group is in the middle of caching and we are
3699 * ok with waiting, then go ahead and wait for progress to be made, and
3700 * set @retry_unclustered to true.
3702 * If @retry_unclustered is true then we've already waited on this
3703 * block group once and should move on to the next block group.
3705 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3706 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
3707 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3708 ffe_ctl->empty_size);
3709 ffe_ctl->retry_unclustered = true;
3711 } else if (!offset) {
3714 ffe_ctl->found_offset = offset;
3718 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3719 struct find_free_extent_ctl *ffe_ctl,
3720 struct btrfs_block_group **bg_ret)
3724 /* We want to try and use the cluster allocator, so lets look there */
3725 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3726 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3727 if (ret >= 0 || ret == -EAGAIN)
3729 /* ret == -ENOENT case falls through */
3732 return find_free_extent_unclustered(block_group, ffe_ctl);
3736 * Tree-log block group locking
3737 * ============================
3739 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3740 * indicates the starting address of a block group, which is reserved only
3741 * for tree-log metadata.
3748 * fs_info::treelog_bg_lock
3752 * Simple allocator for sequential-only block group. It only allows sequential
3753 * allocation. No need to play with trees. This function also reserves the
3754 * bytes as in btrfs_add_reserved_bytes.
3756 static int do_allocation_zoned(struct btrfs_block_group *block_group,
3757 struct find_free_extent_ctl *ffe_ctl,
3758 struct btrfs_block_group **bg_ret)
3760 struct btrfs_fs_info *fs_info = block_group->fs_info;
3761 struct btrfs_space_info *space_info = block_group->space_info;
3762 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3763 u64 start = block_group->start;
3764 u64 num_bytes = ffe_ctl->num_bytes;
3766 u64 bytenr = block_group->start;
3768 u64 data_reloc_bytenr;
3772 ASSERT(btrfs_is_zoned(block_group->fs_info));
3775 * Do not allow non-tree-log blocks in the dedicated tree-log block
3776 * group, and vice versa.
3778 spin_lock(&fs_info->treelog_bg_lock);
3779 log_bytenr = fs_info->treelog_bg;
3780 if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3781 (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3783 spin_unlock(&fs_info->treelog_bg_lock);
3788 * Do not allow non-relocation blocks in the dedicated relocation block
3789 * group, and vice versa.
3791 spin_lock(&fs_info->relocation_bg_lock);
3792 data_reloc_bytenr = fs_info->data_reloc_bg;
3793 if (data_reloc_bytenr &&
3794 ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3795 (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3797 spin_unlock(&fs_info->relocation_bg_lock);
3801 /* Check RO and no space case before trying to activate it */
3802 spin_lock(&block_group->lock);
3803 if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) {
3806 * May need to clear fs_info->{treelog,data_reloc}_bg.
3807 * Return the error after taking the locks.
3810 spin_unlock(&block_group->lock);
3812 if (!ret && !btrfs_zone_activate(block_group)) {
3815 * May need to clear fs_info->{treelog,data_reloc}_bg.
3816 * Return the error after taking the locks.
3820 spin_lock(&space_info->lock);
3821 spin_lock(&block_group->lock);
3822 spin_lock(&fs_info->treelog_bg_lock);
3823 spin_lock(&fs_info->relocation_bg_lock);
3828 ASSERT(!ffe_ctl->for_treelog ||
3829 block_group->start == fs_info->treelog_bg ||
3830 fs_info->treelog_bg == 0);
3831 ASSERT(!ffe_ctl->for_data_reloc ||
3832 block_group->start == fs_info->data_reloc_bg ||
3833 fs_info->data_reloc_bg == 0);
3835 if (block_group->ro) {
3841 * Do not allow currently using block group to be tree-log dedicated
3844 if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3845 (block_group->used || block_group->reserved)) {
3851 * Do not allow currently used block group to be the data relocation
3852 * dedicated block group.
3854 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3855 (block_group->used || block_group->reserved)) {
3860 WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3861 avail = block_group->zone_capacity - block_group->alloc_offset;
3862 if (avail < num_bytes) {
3863 if (ffe_ctl->max_extent_size < avail) {
3865 * With sequential allocator, free space is always
3868 ffe_ctl->max_extent_size = avail;
3869 ffe_ctl->total_free_space = avail;
3875 if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3876 fs_info->treelog_bg = block_group->start;
3878 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg)
3879 fs_info->data_reloc_bg = block_group->start;
3881 ffe_ctl->found_offset = start + block_group->alloc_offset;
3882 block_group->alloc_offset += num_bytes;
3883 spin_lock(&ctl->tree_lock);
3884 ctl->free_space -= num_bytes;
3885 spin_unlock(&ctl->tree_lock);
3888 * We do not check if found_offset is aligned to stripesize. The
3889 * address is anyway rewritten when using zone append writing.
3892 ffe_ctl->search_start = ffe_ctl->found_offset;
3895 if (ret && ffe_ctl->for_treelog)
3896 fs_info->treelog_bg = 0;
3897 if (ret && ffe_ctl->for_data_reloc)
3898 fs_info->data_reloc_bg = 0;
3899 spin_unlock(&fs_info->relocation_bg_lock);
3900 spin_unlock(&fs_info->treelog_bg_lock);
3901 spin_unlock(&block_group->lock);
3902 spin_unlock(&space_info->lock);
3906 static int do_allocation(struct btrfs_block_group *block_group,
3907 struct find_free_extent_ctl *ffe_ctl,
3908 struct btrfs_block_group **bg_ret)
3910 switch (ffe_ctl->policy) {
3911 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3912 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3913 case BTRFS_EXTENT_ALLOC_ZONED:
3914 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
3920 static void release_block_group(struct btrfs_block_group *block_group,
3921 struct find_free_extent_ctl *ffe_ctl,
3924 switch (ffe_ctl->policy) {
3925 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3926 ffe_ctl->retry_clustered = false;
3927 ffe_ctl->retry_unclustered = false;
3929 case BTRFS_EXTENT_ALLOC_ZONED:
3936 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
3938 btrfs_release_block_group(block_group, delalloc);
3941 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
3942 struct btrfs_key *ins)
3944 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3946 if (!ffe_ctl->use_cluster && last_ptr) {
3947 spin_lock(&last_ptr->lock);
3948 last_ptr->window_start = ins->objectid;
3949 spin_unlock(&last_ptr->lock);
3953 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
3954 struct btrfs_key *ins)
3956 switch (ffe_ctl->policy) {
3957 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3958 found_extent_clustered(ffe_ctl, ins);
3960 case BTRFS_EXTENT_ALLOC_ZONED:
3968 static bool can_allocate_chunk(struct btrfs_fs_info *fs_info,
3969 struct find_free_extent_ctl *ffe_ctl)
3971 switch (ffe_ctl->policy) {
3972 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3974 case BTRFS_EXTENT_ALLOC_ZONED:
3976 * If we have enough free space left in an already
3977 * active block group and we can't activate any other
3978 * zone now, do not allow allocating a new chunk and
3979 * let find_free_extent() retry with a smaller size.
3981 if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size &&
3982 !btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags))
3990 static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl)
3992 switch (ffe_ctl->policy) {
3993 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3995 * If we can't allocate a new chunk we've already looped through
3996 * at least once, move on to the NO_EMPTY_SIZE case.
3998 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
4000 case BTRFS_EXTENT_ALLOC_ZONED:
4009 * Return >0 means caller needs to re-search for free extent
4010 * Return 0 means we have the needed free extent.
4011 * Return <0 means we failed to locate any free extent.
4013 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
4014 struct btrfs_key *ins,
4015 struct find_free_extent_ctl *ffe_ctl,
4018 struct btrfs_root *root = fs_info->chunk_root;
4021 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
4022 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
4023 ffe_ctl->orig_have_caching_bg = true;
4025 if (ins->objectid) {
4026 found_extent(ffe_ctl, ins);
4030 if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
4034 if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
4038 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4039 * caching kthreads as we move along
4040 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4041 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4042 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4045 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
4047 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
4049 * We want to skip the LOOP_CACHING_WAIT step if we
4050 * don't have any uncached bgs and we've already done a
4051 * full search through.
4053 if (ffe_ctl->orig_have_caching_bg || !full_search)
4054 ffe_ctl->loop = LOOP_CACHING_WAIT;
4056 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
4061 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
4062 struct btrfs_trans_handle *trans;
4065 /*Check if allocation policy allows to create a new chunk */
4066 if (!can_allocate_chunk(fs_info, ffe_ctl))
4069 trans = current->journal_info;
4073 trans = btrfs_join_transaction(root);
4075 if (IS_ERR(trans)) {
4076 ret = PTR_ERR(trans);
4080 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
4081 CHUNK_ALLOC_FORCE_FOR_EXTENT);
4083 /* Do not bail out on ENOSPC since we can do more. */
4085 ret = chunk_allocation_failed(ffe_ctl);
4087 btrfs_abort_transaction(trans, ret);
4091 btrfs_end_transaction(trans);
4096 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4097 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4101 * Don't loop again if we already have no empty_size and
4104 if (ffe_ctl->empty_size == 0 &&
4105 ffe_ctl->empty_cluster == 0)
4107 ffe_ctl->empty_size = 0;
4108 ffe_ctl->empty_cluster = 0;
4115 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4116 struct find_free_extent_ctl *ffe_ctl,
4117 struct btrfs_space_info *space_info,
4118 struct btrfs_key *ins)
4121 * If our free space is heavily fragmented we may not be able to make
4122 * big contiguous allocations, so instead of doing the expensive search
4123 * for free space, simply return ENOSPC with our max_extent_size so we
4124 * can go ahead and search for a more manageable chunk.
4126 * If our max_extent_size is large enough for our allocation simply
4127 * disable clustering since we will likely not be able to find enough
4128 * space to create a cluster and induce latency trying.
4130 if (space_info->max_extent_size) {
4131 spin_lock(&space_info->lock);
4132 if (space_info->max_extent_size &&
4133 ffe_ctl->num_bytes > space_info->max_extent_size) {
4134 ins->offset = space_info->max_extent_size;
4135 spin_unlock(&space_info->lock);
4137 } else if (space_info->max_extent_size) {
4138 ffe_ctl->use_cluster = false;
4140 spin_unlock(&space_info->lock);
4143 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4144 &ffe_ctl->empty_cluster);
4145 if (ffe_ctl->last_ptr) {
4146 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4148 spin_lock(&last_ptr->lock);
4149 if (last_ptr->block_group)
4150 ffe_ctl->hint_byte = last_ptr->window_start;
4151 if (last_ptr->fragmented) {
4153 * We still set window_start so we can keep track of the
4154 * last place we found an allocation to try and save
4157 ffe_ctl->hint_byte = last_ptr->window_start;
4158 ffe_ctl->use_cluster = false;
4160 spin_unlock(&last_ptr->lock);
4166 static int prepare_allocation(struct btrfs_fs_info *fs_info,
4167 struct find_free_extent_ctl *ffe_ctl,
4168 struct btrfs_space_info *space_info,
4169 struct btrfs_key *ins)
4171 switch (ffe_ctl->policy) {
4172 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4173 return prepare_allocation_clustered(fs_info, ffe_ctl,
4175 case BTRFS_EXTENT_ALLOC_ZONED:
4176 if (ffe_ctl->for_treelog) {
4177 spin_lock(&fs_info->treelog_bg_lock);
4178 if (fs_info->treelog_bg)
4179 ffe_ctl->hint_byte = fs_info->treelog_bg;
4180 spin_unlock(&fs_info->treelog_bg_lock);
4182 if (ffe_ctl->for_data_reloc) {
4183 spin_lock(&fs_info->relocation_bg_lock);
4184 if (fs_info->data_reloc_bg)
4185 ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4186 spin_unlock(&fs_info->relocation_bg_lock);
4195 * walks the btree of allocated extents and find a hole of a given size.
4196 * The key ins is changed to record the hole:
4197 * ins->objectid == start position
4198 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4199 * ins->offset == the size of the hole.
4200 * Any available blocks before search_start are skipped.
4202 * If there is no suitable free space, we will record the max size of
4203 * the free space extent currently.
4205 * The overall logic and call chain:
4207 * find_free_extent()
4208 * |- Iterate through all block groups
4209 * | |- Get a valid block group
4210 * | |- Try to do clustered allocation in that block group
4211 * | |- Try to do unclustered allocation in that block group
4212 * | |- Check if the result is valid
4213 * | | |- If valid, then exit
4214 * | |- Jump to next block group
4216 * |- Push harder to find free extents
4217 * |- If not found, re-iterate all block groups
4219 static noinline int find_free_extent(struct btrfs_root *root,
4220 struct btrfs_key *ins,
4221 struct find_free_extent_ctl *ffe_ctl)
4223 struct btrfs_fs_info *fs_info = root->fs_info;
4225 int cache_block_group_error = 0;
4226 struct btrfs_block_group *block_group = NULL;
4227 struct btrfs_space_info *space_info;
4228 bool full_search = false;
4230 WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4232 ffe_ctl->search_start = 0;
4233 /* For clustered allocation */
4234 ffe_ctl->empty_cluster = 0;
4235 ffe_ctl->last_ptr = NULL;
4236 ffe_ctl->use_cluster = true;
4237 ffe_ctl->have_caching_bg = false;
4238 ffe_ctl->orig_have_caching_bg = false;
4239 ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4241 /* For clustered allocation */
4242 ffe_ctl->retry_clustered = false;
4243 ffe_ctl->retry_unclustered = false;
4244 ffe_ctl->cached = 0;
4245 ffe_ctl->max_extent_size = 0;
4246 ffe_ctl->total_free_space = 0;
4247 ffe_ctl->found_offset = 0;
4248 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4250 if (btrfs_is_zoned(fs_info))
4251 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4253 ins->type = BTRFS_EXTENT_ITEM_KEY;
4257 trace_find_free_extent(root, ffe_ctl->num_bytes, ffe_ctl->empty_size,
4260 space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4262 btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4266 ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4270 ffe_ctl->search_start = max(ffe_ctl->search_start,
4271 first_logical_byte(fs_info));
4272 ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4273 if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4274 block_group = btrfs_lookup_block_group(fs_info,
4275 ffe_ctl->search_start);
4277 * we don't want to use the block group if it doesn't match our
4278 * allocation bits, or if its not cached.
4280 * However if we are re-searching with an ideal block group
4281 * picked out then we don't care that the block group is cached.
4283 if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4284 block_group->cached != BTRFS_CACHE_NO) {
4285 down_read(&space_info->groups_sem);
4286 if (list_empty(&block_group->list) ||
4289 * someone is removing this block group,
4290 * we can't jump into the have_block_group
4291 * target because our list pointers are not
4294 btrfs_put_block_group(block_group);
4295 up_read(&space_info->groups_sem);
4297 ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4298 block_group->flags);
4299 btrfs_lock_block_group(block_group,
4301 goto have_block_group;
4303 } else if (block_group) {
4304 btrfs_put_block_group(block_group);
4308 ffe_ctl->have_caching_bg = false;
4309 if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4310 ffe_ctl->index == 0)
4312 down_read(&space_info->groups_sem);
4313 list_for_each_entry(block_group,
4314 &space_info->block_groups[ffe_ctl->index], list) {
4315 struct btrfs_block_group *bg_ret;
4317 /* If the block group is read-only, we can skip it entirely. */
4318 if (unlikely(block_group->ro)) {
4319 if (ffe_ctl->for_treelog)
4320 btrfs_clear_treelog_bg(block_group);
4321 if (ffe_ctl->for_data_reloc)
4322 btrfs_clear_data_reloc_bg(block_group);
4326 btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4327 ffe_ctl->search_start = block_group->start;
4330 * this can happen if we end up cycling through all the
4331 * raid types, but we want to make sure we only allocate
4332 * for the proper type.
4334 if (!block_group_bits(block_group, ffe_ctl->flags)) {
4335 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4336 BTRFS_BLOCK_GROUP_RAID1_MASK |
4337 BTRFS_BLOCK_GROUP_RAID56_MASK |
4338 BTRFS_BLOCK_GROUP_RAID10;
4341 * if they asked for extra copies and this block group
4342 * doesn't provide them, bail. This does allow us to
4343 * fill raid0 from raid1.
4345 if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4349 * This block group has different flags than we want.
4350 * It's possible that we have MIXED_GROUP flag but no
4351 * block group is mixed. Just skip such block group.
4353 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4358 ffe_ctl->cached = btrfs_block_group_done(block_group);
4359 if (unlikely(!ffe_ctl->cached)) {
4360 ffe_ctl->have_caching_bg = true;
4361 ret = btrfs_cache_block_group(block_group, 0);
4364 * If we get ENOMEM here or something else we want to
4365 * try other block groups, because it may not be fatal.
4366 * However if we can't find anything else we need to
4367 * save our return here so that we return the actual
4368 * error that caused problems, not ENOSPC.
4371 if (!cache_block_group_error)
4372 cache_block_group_error = ret;
4379 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
4383 ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4385 if (bg_ret && bg_ret != block_group) {
4386 btrfs_release_block_group(block_group,
4388 block_group = bg_ret;
4390 } else if (ret == -EAGAIN) {
4391 goto have_block_group;
4392 } else if (ret > 0) {
4397 ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4398 fs_info->stripesize);
4400 /* move on to the next group */
4401 if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4402 block_group->start + block_group->length) {
4403 btrfs_add_free_space_unused(block_group,
4404 ffe_ctl->found_offset,
4405 ffe_ctl->num_bytes);
4409 if (ffe_ctl->found_offset < ffe_ctl->search_start)
4410 btrfs_add_free_space_unused(block_group,
4411 ffe_ctl->found_offset,
4412 ffe_ctl->search_start - ffe_ctl->found_offset);
4414 ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4417 if (ret == -EAGAIN) {
4418 btrfs_add_free_space_unused(block_group,
4419 ffe_ctl->found_offset,
4420 ffe_ctl->num_bytes);
4423 btrfs_inc_block_group_reservations(block_group);
4425 /* we are all good, lets return */
4426 ins->objectid = ffe_ctl->search_start;
4427 ins->offset = ffe_ctl->num_bytes;
4429 trace_btrfs_reserve_extent(block_group, ffe_ctl->search_start,
4430 ffe_ctl->num_bytes);
4431 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4434 release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4437 up_read(&space_info->groups_sem);
4439 ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4443 if (ret == -ENOSPC && !cache_block_group_error) {
4445 * Use ffe_ctl->total_free_space as fallback if we can't find
4446 * any contiguous hole.
4448 if (!ffe_ctl->max_extent_size)
4449 ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4450 spin_lock(&space_info->lock);
4451 space_info->max_extent_size = ffe_ctl->max_extent_size;
4452 spin_unlock(&space_info->lock);
4453 ins->offset = ffe_ctl->max_extent_size;
4454 } else if (ret == -ENOSPC) {
4455 ret = cache_block_group_error;
4461 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4462 * hole that is at least as big as @num_bytes.
4464 * @root - The root that will contain this extent
4466 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4467 * is used for accounting purposes. This value differs
4468 * from @num_bytes only in the case of compressed extents.
4470 * @num_bytes - Number of bytes to allocate on-disk.
4472 * @min_alloc_size - Indicates the minimum amount of space that the
4473 * allocator should try to satisfy. In some cases
4474 * @num_bytes may be larger than what is required and if
4475 * the filesystem is fragmented then allocation fails.
4476 * However, the presence of @min_alloc_size gives a
4477 * chance to try and satisfy the smaller allocation.
4479 * @empty_size - A hint that you plan on doing more COW. This is the
4480 * size in bytes the allocator should try to find free
4481 * next to the block it returns. This is just a hint and
4482 * may be ignored by the allocator.
4484 * @hint_byte - Hint to the allocator to start searching above the byte
4485 * address passed. It might be ignored.
4487 * @ins - This key is modified to record the found hole. It will
4488 * have the following values:
4489 * ins->objectid == start position
4490 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4491 * ins->offset == the size of the hole.
4493 * @is_data - Boolean flag indicating whether an extent is
4494 * allocated for data (true) or metadata (false)
4496 * @delalloc - Boolean flag indicating whether this allocation is for
4497 * delalloc or not. If 'true' data_rwsem of block groups
4498 * is going to be acquired.
4501 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4502 * case -ENOSPC is returned then @ins->offset will contain the size of the
4503 * largest available hole the allocator managed to find.
4505 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4506 u64 num_bytes, u64 min_alloc_size,
4507 u64 empty_size, u64 hint_byte,
4508 struct btrfs_key *ins, int is_data, int delalloc)
4510 struct btrfs_fs_info *fs_info = root->fs_info;
4511 struct find_free_extent_ctl ffe_ctl = {};
4512 bool final_tried = num_bytes == min_alloc_size;
4515 bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4516 bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4518 flags = get_alloc_profile_by_root(root, is_data);
4520 WARN_ON(num_bytes < fs_info->sectorsize);
4522 ffe_ctl.ram_bytes = ram_bytes;
4523 ffe_ctl.num_bytes = num_bytes;
4524 ffe_ctl.min_alloc_size = min_alloc_size;
4525 ffe_ctl.empty_size = empty_size;
4526 ffe_ctl.flags = flags;
4527 ffe_ctl.delalloc = delalloc;
4528 ffe_ctl.hint_byte = hint_byte;
4529 ffe_ctl.for_treelog = for_treelog;
4530 ffe_ctl.for_data_reloc = for_data_reloc;
4532 ret = find_free_extent(root, ins, &ffe_ctl);
4533 if (!ret && !is_data) {
4534 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4535 } else if (ret == -ENOSPC) {
4536 if (!final_tried && ins->offset) {
4537 num_bytes = min(num_bytes >> 1, ins->offset);
4538 num_bytes = round_down(num_bytes,
4539 fs_info->sectorsize);
4540 num_bytes = max(num_bytes, min_alloc_size);
4541 ram_bytes = num_bytes;
4542 if (num_bytes == min_alloc_size)
4545 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4546 struct btrfs_space_info *sinfo;
4548 sinfo = btrfs_find_space_info(fs_info, flags);
4550 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4551 flags, num_bytes, for_treelog, for_data_reloc);
4553 btrfs_dump_space_info(fs_info, sinfo,
4561 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4562 u64 start, u64 len, int delalloc)
4564 struct btrfs_block_group *cache;
4566 cache = btrfs_lookup_block_group(fs_info, start);
4568 btrfs_err(fs_info, "Unable to find block group for %llu",
4573 btrfs_add_free_space(cache, start, len);
4574 btrfs_free_reserved_bytes(cache, len, delalloc);
4575 trace_btrfs_reserved_extent_free(fs_info, start, len);
4577 btrfs_put_block_group(cache);
4581 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start,
4584 struct btrfs_block_group *cache;
4587 cache = btrfs_lookup_block_group(trans->fs_info, start);
4589 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4594 ret = pin_down_extent(trans, cache, start, len, 1);
4595 btrfs_put_block_group(cache);
4599 static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
4602 struct btrfs_fs_info *fs_info = trans->fs_info;
4605 ret = remove_from_free_space_tree(trans, bytenr, num_bytes);
4609 ret = btrfs_update_block_group(trans, bytenr, num_bytes, true);
4612 btrfs_err(fs_info, "update block group failed for %llu %llu",
4617 trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes);
4621 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4622 u64 parent, u64 root_objectid,
4623 u64 flags, u64 owner, u64 offset,
4624 struct btrfs_key *ins, int ref_mod)
4626 struct btrfs_fs_info *fs_info = trans->fs_info;
4627 struct btrfs_root *extent_root;
4629 struct btrfs_extent_item *extent_item;
4630 struct btrfs_extent_inline_ref *iref;
4631 struct btrfs_path *path;
4632 struct extent_buffer *leaf;
4637 type = BTRFS_SHARED_DATA_REF_KEY;
4639 type = BTRFS_EXTENT_DATA_REF_KEY;
4641 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4643 path = btrfs_alloc_path();
4647 extent_root = btrfs_extent_root(fs_info, ins->objectid);
4648 ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size);
4650 btrfs_free_path(path);
4654 leaf = path->nodes[0];
4655 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4656 struct btrfs_extent_item);
4657 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4658 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4659 btrfs_set_extent_flags(leaf, extent_item,
4660 flags | BTRFS_EXTENT_FLAG_DATA);
4662 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4663 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4665 struct btrfs_shared_data_ref *ref;
4666 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4667 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4668 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4670 struct btrfs_extent_data_ref *ref;
4671 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4672 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4673 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4674 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4675 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4678 btrfs_mark_buffer_dirty(path->nodes[0]);
4679 btrfs_free_path(path);
4681 return alloc_reserved_extent(trans, ins->objectid, ins->offset);
4684 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4685 struct btrfs_delayed_ref_node *node,
4686 struct btrfs_delayed_extent_op *extent_op)
4688 struct btrfs_fs_info *fs_info = trans->fs_info;
4689 struct btrfs_root *extent_root;
4691 struct btrfs_extent_item *extent_item;
4692 struct btrfs_key extent_key;
4693 struct btrfs_tree_block_info *block_info;
4694 struct btrfs_extent_inline_ref *iref;
4695 struct btrfs_path *path;
4696 struct extent_buffer *leaf;
4697 struct btrfs_delayed_tree_ref *ref;
4698 u32 size = sizeof(*extent_item) + sizeof(*iref);
4699 u64 flags = extent_op->flags_to_set;
4700 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4702 ref = btrfs_delayed_node_to_tree_ref(node);
4704 extent_key.objectid = node->bytenr;
4705 if (skinny_metadata) {
4706 extent_key.offset = ref->level;
4707 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4709 extent_key.offset = node->num_bytes;
4710 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4711 size += sizeof(*block_info);
4714 path = btrfs_alloc_path();
4718 extent_root = btrfs_extent_root(fs_info, extent_key.objectid);
4719 ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key,
4722 btrfs_free_path(path);
4726 leaf = path->nodes[0];
4727 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4728 struct btrfs_extent_item);
4729 btrfs_set_extent_refs(leaf, extent_item, 1);
4730 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4731 btrfs_set_extent_flags(leaf, extent_item,
4732 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4734 if (skinny_metadata) {
4735 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4737 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4738 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4739 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4740 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4743 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4744 btrfs_set_extent_inline_ref_type(leaf, iref,
4745 BTRFS_SHARED_BLOCK_REF_KEY);
4746 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4748 btrfs_set_extent_inline_ref_type(leaf, iref,
4749 BTRFS_TREE_BLOCK_REF_KEY);
4750 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4753 btrfs_mark_buffer_dirty(leaf);
4754 btrfs_free_path(path);
4756 return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize);
4759 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4760 struct btrfs_root *root, u64 owner,
4761 u64 offset, u64 ram_bytes,
4762 struct btrfs_key *ins)
4764 struct btrfs_ref generic_ref = { 0 };
4766 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4768 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4769 ins->objectid, ins->offset, 0);
4770 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner,
4772 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4774 return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4778 * this is used by the tree logging recovery code. It records that
4779 * an extent has been allocated and makes sure to clear the free
4780 * space cache bits as well
4782 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4783 u64 root_objectid, u64 owner, u64 offset,
4784 struct btrfs_key *ins)
4786 struct btrfs_fs_info *fs_info = trans->fs_info;
4788 struct btrfs_block_group *block_group;
4789 struct btrfs_space_info *space_info;
4792 * Mixed block groups will exclude before processing the log so we only
4793 * need to do the exclude dance if this fs isn't mixed.
4795 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4796 ret = __exclude_logged_extent(fs_info, ins->objectid,
4802 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4806 space_info = block_group->space_info;
4807 spin_lock(&space_info->lock);
4808 spin_lock(&block_group->lock);
4809 space_info->bytes_reserved += ins->offset;
4810 block_group->reserved += ins->offset;
4811 spin_unlock(&block_group->lock);
4812 spin_unlock(&space_info->lock);
4814 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4817 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4818 btrfs_put_block_group(block_group);
4822 static struct extent_buffer *
4823 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4824 u64 bytenr, int level, u64 owner,
4825 enum btrfs_lock_nesting nest)
4827 struct btrfs_fs_info *fs_info = root->fs_info;
4828 struct extent_buffer *buf;
4830 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
4835 * Extra safety check in case the extent tree is corrupted and extent
4836 * allocator chooses to use a tree block which is already used and
4839 if (buf->lock_owner == current->pid) {
4840 btrfs_err_rl(fs_info,
4841 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4842 buf->start, btrfs_header_owner(buf), current->pid);
4843 free_extent_buffer(buf);
4844 return ERR_PTR(-EUCLEAN);
4848 * This needs to stay, because we could allocate a freed block from an
4849 * old tree into a new tree, so we need to make sure this new block is
4850 * set to the appropriate level and owner.
4852 btrfs_set_buffer_lockdep_class(owner, buf, level);
4853 __btrfs_tree_lock(buf, nest);
4854 btrfs_clean_tree_block(buf);
4855 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4856 clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags);
4858 set_extent_buffer_uptodate(buf);
4860 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4861 btrfs_set_header_level(buf, level);
4862 btrfs_set_header_bytenr(buf, buf->start);
4863 btrfs_set_header_generation(buf, trans->transid);
4864 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4865 btrfs_set_header_owner(buf, owner);
4866 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4867 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4868 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4869 buf->log_index = root->log_transid % 2;
4871 * we allow two log transactions at a time, use different
4872 * EXTENT bit to differentiate dirty pages.
4874 if (buf->log_index == 0)
4875 set_extent_dirty(&root->dirty_log_pages, buf->start,
4876 buf->start + buf->len - 1, GFP_NOFS);
4878 set_extent_new(&root->dirty_log_pages, buf->start,
4879 buf->start + buf->len - 1);
4881 buf->log_index = -1;
4882 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4883 buf->start + buf->len - 1, GFP_NOFS);
4885 /* this returns a buffer locked for blocking */
4890 * finds a free extent and does all the dirty work required for allocation
4891 * returns the tree buffer or an ERR_PTR on error.
4893 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4894 struct btrfs_root *root,
4895 u64 parent, u64 root_objectid,
4896 const struct btrfs_disk_key *key,
4897 int level, u64 hint,
4899 enum btrfs_lock_nesting nest)
4901 struct btrfs_fs_info *fs_info = root->fs_info;
4902 struct btrfs_key ins;
4903 struct btrfs_block_rsv *block_rsv;
4904 struct extent_buffer *buf;
4905 struct btrfs_delayed_extent_op *extent_op;
4906 struct btrfs_ref generic_ref = { 0 };
4909 u32 blocksize = fs_info->nodesize;
4910 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4912 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4913 if (btrfs_is_testing(fs_info)) {
4914 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4915 level, root_objectid, nest);
4917 root->alloc_bytenr += blocksize;
4922 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4923 if (IS_ERR(block_rsv))
4924 return ERR_CAST(block_rsv);
4926 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4927 empty_size, hint, &ins, 0, 0);
4931 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4932 root_objectid, nest);
4935 goto out_free_reserved;
4938 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4940 parent = ins.objectid;
4941 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4945 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4946 extent_op = btrfs_alloc_delayed_extent_op();
4952 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4954 memset(&extent_op->key, 0, sizeof(extent_op->key));
4955 extent_op->flags_to_set = flags;
4956 extent_op->update_key = skinny_metadata ? false : true;
4957 extent_op->update_flags = true;
4958 extent_op->level = level;
4960 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4961 ins.objectid, ins.offset, parent);
4962 btrfs_init_tree_ref(&generic_ref, level, root_objectid,
4963 root->root_key.objectid, false);
4964 btrfs_ref_tree_mod(fs_info, &generic_ref);
4965 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
4967 goto out_free_delayed;
4972 btrfs_free_delayed_extent_op(extent_op);
4974 btrfs_tree_unlock(buf);
4975 free_extent_buffer(buf);
4977 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
4979 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
4980 return ERR_PTR(ret);
4983 struct walk_control {
4984 u64 refs[BTRFS_MAX_LEVEL];
4985 u64 flags[BTRFS_MAX_LEVEL];
4986 struct btrfs_key update_progress;
4987 struct btrfs_key drop_progress;
4999 #define DROP_REFERENCE 1
5000 #define UPDATE_BACKREF 2
5002 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5003 struct btrfs_root *root,
5004 struct walk_control *wc,
5005 struct btrfs_path *path)
5007 struct btrfs_fs_info *fs_info = root->fs_info;
5013 struct btrfs_key key;
5014 struct extent_buffer *eb;
5019 if (path->slots[wc->level] < wc->reada_slot) {
5020 wc->reada_count = wc->reada_count * 2 / 3;
5021 wc->reada_count = max(wc->reada_count, 2);
5023 wc->reada_count = wc->reada_count * 3 / 2;
5024 wc->reada_count = min_t(int, wc->reada_count,
5025 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
5028 eb = path->nodes[wc->level];
5029 nritems = btrfs_header_nritems(eb);
5031 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5032 if (nread >= wc->reada_count)
5036 bytenr = btrfs_node_blockptr(eb, slot);
5037 generation = btrfs_node_ptr_generation(eb, slot);
5039 if (slot == path->slots[wc->level])
5042 if (wc->stage == UPDATE_BACKREF &&
5043 generation <= root->root_key.offset)
5046 /* We don't lock the tree block, it's OK to be racy here */
5047 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5048 wc->level - 1, 1, &refs,
5050 /* We don't care about errors in readahead. */
5055 if (wc->stage == DROP_REFERENCE) {
5059 if (wc->level == 1 &&
5060 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5062 if (!wc->update_ref ||
5063 generation <= root->root_key.offset)
5065 btrfs_node_key_to_cpu(eb, &key, slot);
5066 ret = btrfs_comp_cpu_keys(&key,
5067 &wc->update_progress);
5071 if (wc->level == 1 &&
5072 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5076 btrfs_readahead_node_child(eb, slot);
5079 wc->reada_slot = slot;
5083 * helper to process tree block while walking down the tree.
5085 * when wc->stage == UPDATE_BACKREF, this function updates
5086 * back refs for pointers in the block.
5088 * NOTE: return value 1 means we should stop walking down.
5090 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5091 struct btrfs_root *root,
5092 struct btrfs_path *path,
5093 struct walk_control *wc, int lookup_info)
5095 struct btrfs_fs_info *fs_info = root->fs_info;
5096 int level = wc->level;
5097 struct extent_buffer *eb = path->nodes[level];
5098 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5101 if (wc->stage == UPDATE_BACKREF &&
5102 btrfs_header_owner(eb) != root->root_key.objectid)
5106 * when reference count of tree block is 1, it won't increase
5107 * again. once full backref flag is set, we never clear it.
5110 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5111 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5112 BUG_ON(!path->locks[level]);
5113 ret = btrfs_lookup_extent_info(trans, fs_info,
5114 eb->start, level, 1,
5117 BUG_ON(ret == -ENOMEM);
5120 BUG_ON(wc->refs[level] == 0);
5123 if (wc->stage == DROP_REFERENCE) {
5124 if (wc->refs[level] > 1)
5127 if (path->locks[level] && !wc->keep_locks) {
5128 btrfs_tree_unlock_rw(eb, path->locks[level]);
5129 path->locks[level] = 0;
5134 /* wc->stage == UPDATE_BACKREF */
5135 if (!(wc->flags[level] & flag)) {
5136 BUG_ON(!path->locks[level]);
5137 ret = btrfs_inc_ref(trans, root, eb, 1);
5138 BUG_ON(ret); /* -ENOMEM */
5139 ret = btrfs_dec_ref(trans, root, eb, 0);
5140 BUG_ON(ret); /* -ENOMEM */
5141 ret = btrfs_set_disk_extent_flags(trans, eb, flag,
5142 btrfs_header_level(eb));
5143 BUG_ON(ret); /* -ENOMEM */
5144 wc->flags[level] |= flag;
5148 * the block is shared by multiple trees, so it's not good to
5149 * keep the tree lock
5151 if (path->locks[level] && level > 0) {
5152 btrfs_tree_unlock_rw(eb, path->locks[level]);
5153 path->locks[level] = 0;
5159 * This is used to verify a ref exists for this root to deal with a bug where we
5160 * would have a drop_progress key that hadn't been updated properly.
5162 static int check_ref_exists(struct btrfs_trans_handle *trans,
5163 struct btrfs_root *root, u64 bytenr, u64 parent,
5166 struct btrfs_path *path;
5167 struct btrfs_extent_inline_ref *iref;
5170 path = btrfs_alloc_path();
5174 ret = lookup_extent_backref(trans, path, &iref, bytenr,
5175 root->fs_info->nodesize, parent,
5176 root->root_key.objectid, level, 0);
5177 btrfs_free_path(path);
5186 * helper to process tree block pointer.
5188 * when wc->stage == DROP_REFERENCE, this function checks
5189 * reference count of the block pointed to. if the block
5190 * is shared and we need update back refs for the subtree
5191 * rooted at the block, this function changes wc->stage to
5192 * UPDATE_BACKREF. if the block is shared and there is no
5193 * need to update back, this function drops the reference
5196 * NOTE: return value 1 means we should stop walking down.
5198 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5199 struct btrfs_root *root,
5200 struct btrfs_path *path,
5201 struct walk_control *wc, int *lookup_info)
5203 struct btrfs_fs_info *fs_info = root->fs_info;
5207 struct btrfs_key key;
5208 struct btrfs_key first_key;
5209 struct btrfs_ref ref = { 0 };
5210 struct extent_buffer *next;
5211 int level = wc->level;
5214 bool need_account = false;
5216 generation = btrfs_node_ptr_generation(path->nodes[level],
5217 path->slots[level]);
5219 * if the lower level block was created before the snapshot
5220 * was created, we know there is no need to update back refs
5223 if (wc->stage == UPDATE_BACKREF &&
5224 generation <= root->root_key.offset) {
5229 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5230 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
5231 path->slots[level]);
5233 next = find_extent_buffer(fs_info, bytenr);
5235 next = btrfs_find_create_tree_block(fs_info, bytenr,
5236 root->root_key.objectid, level - 1);
5238 return PTR_ERR(next);
5241 btrfs_tree_lock(next);
5243 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5244 &wc->refs[level - 1],
5245 &wc->flags[level - 1]);
5249 if (unlikely(wc->refs[level - 1] == 0)) {
5250 btrfs_err(fs_info, "Missing references.");
5256 if (wc->stage == DROP_REFERENCE) {
5257 if (wc->refs[level - 1] > 1) {
5258 need_account = true;
5260 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5263 if (!wc->update_ref ||
5264 generation <= root->root_key.offset)
5267 btrfs_node_key_to_cpu(path->nodes[level], &key,
5268 path->slots[level]);
5269 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5273 wc->stage = UPDATE_BACKREF;
5274 wc->shared_level = level - 1;
5278 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5282 if (!btrfs_buffer_uptodate(next, generation, 0)) {
5283 btrfs_tree_unlock(next);
5284 free_extent_buffer(next);
5290 if (reada && level == 1)
5291 reada_walk_down(trans, root, wc, path);
5292 next = read_tree_block(fs_info, bytenr, root->root_key.objectid,
5293 generation, level - 1, &first_key);
5295 return PTR_ERR(next);
5296 } else if (!extent_buffer_uptodate(next)) {
5297 free_extent_buffer(next);
5300 btrfs_tree_lock(next);
5304 ASSERT(level == btrfs_header_level(next));
5305 if (level != btrfs_header_level(next)) {
5306 btrfs_err(root->fs_info, "mismatched level");
5310 path->nodes[level] = next;
5311 path->slots[level] = 0;
5312 path->locks[level] = BTRFS_WRITE_LOCK;
5318 wc->refs[level - 1] = 0;
5319 wc->flags[level - 1] = 0;
5320 if (wc->stage == DROP_REFERENCE) {
5321 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5322 parent = path->nodes[level]->start;
5324 ASSERT(root->root_key.objectid ==
5325 btrfs_header_owner(path->nodes[level]));
5326 if (root->root_key.objectid !=
5327 btrfs_header_owner(path->nodes[level])) {
5328 btrfs_err(root->fs_info,
5329 "mismatched block owner");
5337 * If we had a drop_progress we need to verify the refs are set
5338 * as expected. If we find our ref then we know that from here
5339 * on out everything should be correct, and we can clear the
5342 if (wc->restarted) {
5343 ret = check_ref_exists(trans, root, bytenr, parent,
5354 * Reloc tree doesn't contribute to qgroup numbers, and we have
5355 * already accounted them at merge time (replace_path),
5356 * thus we could skip expensive subtree trace here.
5358 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5360 ret = btrfs_qgroup_trace_subtree(trans, next,
5361 generation, level - 1);
5363 btrfs_err_rl(fs_info,
5364 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5370 * We need to update the next key in our walk control so we can
5371 * update the drop_progress key accordingly. We don't care if
5372 * find_next_key doesn't find a key because that means we're at
5373 * the end and are going to clean up now.
5375 wc->drop_level = level;
5376 find_next_key(path, level, &wc->drop_progress);
5378 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5379 fs_info->nodesize, parent);
5380 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid,
5382 ret = btrfs_free_extent(trans, &ref);
5391 btrfs_tree_unlock(next);
5392 free_extent_buffer(next);
5398 * helper to process tree block while walking up the tree.
5400 * when wc->stage == DROP_REFERENCE, this function drops
5401 * reference count on the block.
5403 * when wc->stage == UPDATE_BACKREF, this function changes
5404 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5405 * to UPDATE_BACKREF previously while processing the block.
5407 * NOTE: return value 1 means we should stop walking up.
5409 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5410 struct btrfs_root *root,
5411 struct btrfs_path *path,
5412 struct walk_control *wc)
5414 struct btrfs_fs_info *fs_info = root->fs_info;
5416 int level = wc->level;
5417 struct extent_buffer *eb = path->nodes[level];
5420 if (wc->stage == UPDATE_BACKREF) {
5421 BUG_ON(wc->shared_level < level);
5422 if (level < wc->shared_level)
5425 ret = find_next_key(path, level + 1, &wc->update_progress);
5429 wc->stage = DROP_REFERENCE;
5430 wc->shared_level = -1;
5431 path->slots[level] = 0;
5434 * check reference count again if the block isn't locked.
5435 * we should start walking down the tree again if reference
5438 if (!path->locks[level]) {
5440 btrfs_tree_lock(eb);
5441 path->locks[level] = BTRFS_WRITE_LOCK;
5443 ret = btrfs_lookup_extent_info(trans, fs_info,
5444 eb->start, level, 1,
5448 btrfs_tree_unlock_rw(eb, path->locks[level]);
5449 path->locks[level] = 0;
5452 BUG_ON(wc->refs[level] == 0);
5453 if (wc->refs[level] == 1) {
5454 btrfs_tree_unlock_rw(eb, path->locks[level]);
5455 path->locks[level] = 0;
5461 /* wc->stage == DROP_REFERENCE */
5462 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5464 if (wc->refs[level] == 1) {
5466 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5467 ret = btrfs_dec_ref(trans, root, eb, 1);
5469 ret = btrfs_dec_ref(trans, root, eb, 0);
5470 BUG_ON(ret); /* -ENOMEM */
5471 if (is_fstree(root->root_key.objectid)) {
5472 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5474 btrfs_err_rl(fs_info,
5475 "error %d accounting leaf items, quota is out of sync, rescan required",
5480 /* make block locked assertion in btrfs_clean_tree_block happy */
5481 if (!path->locks[level] &&
5482 btrfs_header_generation(eb) == trans->transid) {
5483 btrfs_tree_lock(eb);
5484 path->locks[level] = BTRFS_WRITE_LOCK;
5486 btrfs_clean_tree_block(eb);
5489 if (eb == root->node) {
5490 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5492 else if (root->root_key.objectid != btrfs_header_owner(eb))
5493 goto owner_mismatch;
5495 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5496 parent = path->nodes[level + 1]->start;
5497 else if (root->root_key.objectid !=
5498 btrfs_header_owner(path->nodes[level + 1]))
5499 goto owner_mismatch;
5502 btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent,
5503 wc->refs[level] == 1);
5505 wc->refs[level] = 0;
5506 wc->flags[level] = 0;
5510 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5511 btrfs_header_owner(eb), root->root_key.objectid);
5515 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5516 struct btrfs_root *root,
5517 struct btrfs_path *path,
5518 struct walk_control *wc)
5520 int level = wc->level;
5521 int lookup_info = 1;
5524 while (level >= 0) {
5525 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5532 if (path->slots[level] >=
5533 btrfs_header_nritems(path->nodes[level]))
5536 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5538 path->slots[level]++;
5547 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5548 struct btrfs_root *root,
5549 struct btrfs_path *path,
5550 struct walk_control *wc, int max_level)
5552 int level = wc->level;
5555 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5556 while (level < max_level && path->nodes[level]) {
5558 if (path->slots[level] + 1 <
5559 btrfs_header_nritems(path->nodes[level])) {
5560 path->slots[level]++;
5563 ret = walk_up_proc(trans, root, path, wc);
5569 if (path->locks[level]) {
5570 btrfs_tree_unlock_rw(path->nodes[level],
5571 path->locks[level]);
5572 path->locks[level] = 0;
5574 free_extent_buffer(path->nodes[level]);
5575 path->nodes[level] = NULL;
5583 * drop a subvolume tree.
5585 * this function traverses the tree freeing any blocks that only
5586 * referenced by the tree.
5588 * when a shared tree block is found. this function decreases its
5589 * reference count by one. if update_ref is true, this function
5590 * also make sure backrefs for the shared block and all lower level
5591 * blocks are properly updated.
5593 * If called with for_reloc == 0, may exit early with -EAGAIN
5595 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5597 struct btrfs_fs_info *fs_info = root->fs_info;
5598 struct btrfs_path *path;
5599 struct btrfs_trans_handle *trans;
5600 struct btrfs_root *tree_root = fs_info->tree_root;
5601 struct btrfs_root_item *root_item = &root->root_item;
5602 struct walk_control *wc;
5603 struct btrfs_key key;
5607 bool root_dropped = false;
5608 bool unfinished_drop = false;
5610 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5612 path = btrfs_alloc_path();
5618 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5620 btrfs_free_path(path);
5626 * Use join to avoid potential EINTR from transaction start. See
5627 * wait_reserve_ticket and the whole reservation callchain.
5630 trans = btrfs_join_transaction(tree_root);
5632 trans = btrfs_start_transaction(tree_root, 0);
5633 if (IS_ERR(trans)) {
5634 err = PTR_ERR(trans);
5638 err = btrfs_run_delayed_items(trans);
5643 * This will help us catch people modifying the fs tree while we're
5644 * dropping it. It is unsafe to mess with the fs tree while it's being
5645 * dropped as we unlock the root node and parent nodes as we walk down
5646 * the tree, assuming nothing will change. If something does change
5647 * then we'll have stale information and drop references to blocks we've
5650 set_bit(BTRFS_ROOT_DELETING, &root->state);
5651 unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
5653 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5654 level = btrfs_header_level(root->node);
5655 path->nodes[level] = btrfs_lock_root_node(root);
5656 path->slots[level] = 0;
5657 path->locks[level] = BTRFS_WRITE_LOCK;
5658 memset(&wc->update_progress, 0,
5659 sizeof(wc->update_progress));
5661 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5662 memcpy(&wc->update_progress, &key,
5663 sizeof(wc->update_progress));
5665 level = btrfs_root_drop_level(root_item);
5667 path->lowest_level = level;
5668 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5669 path->lowest_level = 0;
5677 * unlock our path, this is safe because only this
5678 * function is allowed to delete this snapshot
5680 btrfs_unlock_up_safe(path, 0);
5682 level = btrfs_header_level(root->node);
5684 btrfs_tree_lock(path->nodes[level]);
5685 path->locks[level] = BTRFS_WRITE_LOCK;
5687 ret = btrfs_lookup_extent_info(trans, fs_info,
5688 path->nodes[level]->start,
5689 level, 1, &wc->refs[level],
5695 BUG_ON(wc->refs[level] == 0);
5697 if (level == btrfs_root_drop_level(root_item))
5700 btrfs_tree_unlock(path->nodes[level]);
5701 path->locks[level] = 0;
5702 WARN_ON(wc->refs[level] != 1);
5707 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5709 wc->shared_level = -1;
5710 wc->stage = DROP_REFERENCE;
5711 wc->update_ref = update_ref;
5713 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5717 ret = walk_down_tree(trans, root, path, wc);
5723 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5730 BUG_ON(wc->stage != DROP_REFERENCE);
5734 if (wc->stage == DROP_REFERENCE) {
5735 wc->drop_level = wc->level;
5736 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5738 path->slots[wc->drop_level]);
5740 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5741 &wc->drop_progress);
5742 btrfs_set_root_drop_level(root_item, wc->drop_level);
5744 BUG_ON(wc->level == 0);
5745 if (btrfs_should_end_transaction(trans) ||
5746 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5747 ret = btrfs_update_root(trans, tree_root,
5751 btrfs_abort_transaction(trans, ret);
5756 btrfs_end_transaction_throttle(trans);
5757 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5758 btrfs_debug(fs_info,
5759 "drop snapshot early exit");
5765 * Use join to avoid potential EINTR from transaction
5766 * start. See wait_reserve_ticket and the whole
5767 * reservation callchain.
5770 trans = btrfs_join_transaction(tree_root);
5772 trans = btrfs_start_transaction(tree_root, 0);
5773 if (IS_ERR(trans)) {
5774 err = PTR_ERR(trans);
5779 btrfs_release_path(path);
5783 ret = btrfs_del_root(trans, &root->root_key);
5785 btrfs_abort_transaction(trans, ret);
5790 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5791 ret = btrfs_find_root(tree_root, &root->root_key, path,
5794 btrfs_abort_transaction(trans, ret);
5797 } else if (ret > 0) {
5798 /* if we fail to delete the orphan item this time
5799 * around, it'll get picked up the next time.
5801 * The most common failure here is just -ENOENT.
5803 btrfs_del_orphan_item(trans, tree_root,
5804 root->root_key.objectid);
5809 * This subvolume is going to be completely dropped, and won't be
5810 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5811 * commit transaction time. So free it here manually.
5813 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
5814 btrfs_qgroup_free_meta_all_pertrans(root);
5816 if (test_bit(BTRFS_ROOT_REGISTERED, &root->state))
5817 btrfs_add_dropped_root(trans, root);
5819 btrfs_put_root(root);
5820 root_dropped = true;
5822 btrfs_end_transaction_throttle(trans);
5825 btrfs_free_path(path);
5828 * We were an unfinished drop root, check to see if there are any
5829 * pending, and if not clear and wake up any waiters.
5831 if (!err && unfinished_drop)
5832 btrfs_maybe_wake_unfinished_drop(fs_info);
5835 * So if we need to stop dropping the snapshot for whatever reason we
5836 * need to make sure to add it back to the dead root list so that we
5837 * keep trying to do the work later. This also cleans up roots if we
5838 * don't have it in the radix (like when we recover after a power fail
5839 * or unmount) so we don't leak memory.
5841 if (!for_reloc && !root_dropped)
5842 btrfs_add_dead_root(root);
5847 * drop subtree rooted at tree block 'node'.
5849 * NOTE: this function will unlock and release tree block 'node'
5850 * only used by relocation code
5852 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5853 struct btrfs_root *root,
5854 struct extent_buffer *node,
5855 struct extent_buffer *parent)
5857 struct btrfs_fs_info *fs_info = root->fs_info;
5858 struct btrfs_path *path;
5859 struct walk_control *wc;
5865 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5867 path = btrfs_alloc_path();
5871 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5873 btrfs_free_path(path);
5877 btrfs_assert_tree_write_locked(parent);
5878 parent_level = btrfs_header_level(parent);
5879 atomic_inc(&parent->refs);
5880 path->nodes[parent_level] = parent;
5881 path->slots[parent_level] = btrfs_header_nritems(parent);
5883 btrfs_assert_tree_write_locked(node);
5884 level = btrfs_header_level(node);
5885 path->nodes[level] = node;
5886 path->slots[level] = 0;
5887 path->locks[level] = BTRFS_WRITE_LOCK;
5889 wc->refs[parent_level] = 1;
5890 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5892 wc->shared_level = -1;
5893 wc->stage = DROP_REFERENCE;
5896 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5899 wret = walk_down_tree(trans, root, path, wc);
5905 wret = walk_up_tree(trans, root, path, wc, parent_level);
5913 btrfs_free_path(path);
5918 * helper to account the unused space of all the readonly block group in the
5919 * space_info. takes mirrors into account.
5921 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
5923 struct btrfs_block_group *block_group;
5927 /* It's df, we don't care if it's racy */
5928 if (list_empty(&sinfo->ro_bgs))
5931 spin_lock(&sinfo->lock);
5932 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
5933 spin_lock(&block_group->lock);
5935 if (!block_group->ro) {
5936 spin_unlock(&block_group->lock);
5940 factor = btrfs_bg_type_to_factor(block_group->flags);
5941 free_bytes += (block_group->length -
5942 block_group->used) * factor;
5944 spin_unlock(&block_group->lock);
5946 spin_unlock(&sinfo->lock);
5951 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5954 return unpin_extent_range(fs_info, start, end, false);
5958 * It used to be that old block groups would be left around forever.
5959 * Iterating over them would be enough to trim unused space. Since we
5960 * now automatically remove them, we also need to iterate over unallocated
5963 * We don't want a transaction for this since the discard may take a
5964 * substantial amount of time. We don't require that a transaction be
5965 * running, but we do need to take a running transaction into account
5966 * to ensure that we're not discarding chunks that were released or
5967 * allocated in the current transaction.
5969 * Holding the chunks lock will prevent other threads from allocating
5970 * or releasing chunks, but it won't prevent a running transaction
5971 * from committing and releasing the memory that the pending chunks
5972 * list head uses. For that, we need to take a reference to the
5973 * transaction and hold the commit root sem. We only need to hold
5974 * it while performing the free space search since we have already
5975 * held back allocations.
5977 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
5979 u64 start = SZ_1M, len = 0, end = 0;
5984 /* Discard not supported = nothing to do. */
5985 if (!bdev_max_discard_sectors(device->bdev))
5988 /* Not writable = nothing to do. */
5989 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5992 /* No free space = nothing to do. */
5993 if (device->total_bytes <= device->bytes_used)
5999 struct btrfs_fs_info *fs_info = device->fs_info;
6002 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
6006 find_first_clear_extent_bit(&device->alloc_state, start,
6008 CHUNK_TRIMMED | CHUNK_ALLOCATED);
6010 /* Check if there are any CHUNK_* bits left */
6011 if (start > device->total_bytes) {
6012 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6013 btrfs_warn_in_rcu(fs_info,
6014 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
6015 start, end - start + 1,
6016 rcu_str_deref(device->name),
6017 device->total_bytes);
6018 mutex_unlock(&fs_info->chunk_mutex);
6023 /* Ensure we skip the reserved area in the first 1M */
6024 start = max_t(u64, start, SZ_1M);
6027 * If find_first_clear_extent_bit find a range that spans the
6028 * end of the device it will set end to -1, in this case it's up
6029 * to the caller to trim the value to the size of the device.
6031 end = min(end, device->total_bytes - 1);
6033 len = end - start + 1;
6035 /* We didn't find any extents */
6037 mutex_unlock(&fs_info->chunk_mutex);
6042 ret = btrfs_issue_discard(device->bdev, start, len,
6045 set_extent_bits(&device->alloc_state, start,
6048 mutex_unlock(&fs_info->chunk_mutex);
6056 if (fatal_signal_pending(current)) {
6068 * Trim the whole filesystem by:
6069 * 1) trimming the free space in each block group
6070 * 2) trimming the unallocated space on each device
6072 * This will also continue trimming even if a block group or device encounters
6073 * an error. The return value will be the last error, or 0 if nothing bad
6076 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6078 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6079 struct btrfs_block_group *cache = NULL;
6080 struct btrfs_device *device;
6082 u64 range_end = U64_MAX;
6092 if (range->start == U64_MAX)
6096 * Check range overflow if range->len is set.
6097 * The default range->len is U64_MAX.
6099 if (range->len != U64_MAX &&
6100 check_add_overflow(range->start, range->len, &range_end))
6103 cache = btrfs_lookup_first_block_group(fs_info, range->start);
6104 for (; cache; cache = btrfs_next_block_group(cache)) {
6105 if (cache->start >= range_end) {
6106 btrfs_put_block_group(cache);
6110 start = max(range->start, cache->start);
6111 end = min(range_end, cache->start + cache->length);
6113 if (end - start >= range->minlen) {
6114 if (!btrfs_block_group_done(cache)) {
6115 ret = btrfs_cache_block_group(cache, 0);
6121 ret = btrfs_wait_block_group_cache_done(cache);
6128 ret = btrfs_trim_block_group(cache,
6134 trimmed += group_trimmed;
6145 "failed to trim %llu block group(s), last error %d",
6148 mutex_lock(&fs_devices->device_list_mutex);
6149 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6150 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6153 ret = btrfs_trim_free_extents(device, &group_trimmed);
6160 trimmed += group_trimmed;
6162 mutex_unlock(&fs_devices->device_list_mutex);
6166 "failed to trim %llu device(s), last error %d",
6167 dev_failed, dev_ret);
6168 range->len = trimmed;