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 #undef SCRAMBLE_DELAYED_REFS
41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
42 struct btrfs_delayed_ref_node *node, u64 parent,
43 u64 root_objectid, u64 owner_objectid,
44 u64 owner_offset, int refs_to_drop,
45 struct btrfs_delayed_extent_op *extra_op);
46 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
47 struct extent_buffer *leaf,
48 struct btrfs_extent_item *ei);
49 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
50 u64 parent, u64 root_objectid,
51 u64 flags, u64 owner, u64 offset,
52 struct btrfs_key *ins, int ref_mod);
53 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
54 struct btrfs_delayed_ref_node *node,
55 struct btrfs_delayed_extent_op *extent_op);
56 static int find_next_key(struct btrfs_path *path, int level,
57 struct btrfs_key *key);
59 static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
61 return (cache->flags & bits) == bits;
64 int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info,
65 u64 start, u64 num_bytes)
67 u64 end = start + num_bytes - 1;
68 set_extent_bits(&fs_info->excluded_extents, start, end,
73 void btrfs_free_excluded_extents(struct btrfs_block_group *cache)
75 struct btrfs_fs_info *fs_info = cache->fs_info;
79 end = start + cache->length - 1;
81 clear_extent_bits(&fs_info->excluded_extents, start, end,
85 static u64 generic_ref_to_space_flags(struct btrfs_ref *ref)
87 if (ref->type == BTRFS_REF_METADATA) {
88 if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID)
89 return BTRFS_BLOCK_GROUP_SYSTEM;
91 return BTRFS_BLOCK_GROUP_METADATA;
93 return BTRFS_BLOCK_GROUP_DATA;
96 static void add_pinned_bytes(struct btrfs_fs_info *fs_info,
97 struct btrfs_ref *ref)
99 struct btrfs_space_info *space_info;
100 u64 flags = generic_ref_to_space_flags(ref);
102 space_info = btrfs_find_space_info(fs_info, flags);
104 percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len,
105 BTRFS_TOTAL_BYTES_PINNED_BATCH);
108 static void sub_pinned_bytes(struct btrfs_fs_info *fs_info,
109 struct btrfs_ref *ref)
111 struct btrfs_space_info *space_info;
112 u64 flags = generic_ref_to_space_flags(ref);
114 space_info = btrfs_find_space_info(fs_info, flags);
116 percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len,
117 BTRFS_TOTAL_BYTES_PINNED_BATCH);
120 /* simple helper to search for an existing data extent at a given offset */
121 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
124 struct btrfs_key key;
125 struct btrfs_path *path;
127 path = btrfs_alloc_path();
131 key.objectid = start;
133 key.type = BTRFS_EXTENT_ITEM_KEY;
134 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
135 btrfs_free_path(path);
140 * helper function to lookup reference count and flags of a tree block.
142 * the head node for delayed ref is used to store the sum of all the
143 * reference count modifications queued up in the rbtree. the head
144 * node may also store the extent flags to set. This way you can check
145 * to see what the reference count and extent flags would be if all of
146 * the delayed refs are not processed.
148 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
149 struct btrfs_fs_info *fs_info, u64 bytenr,
150 u64 offset, int metadata, u64 *refs, u64 *flags)
152 struct btrfs_delayed_ref_head *head;
153 struct btrfs_delayed_ref_root *delayed_refs;
154 struct btrfs_path *path;
155 struct btrfs_extent_item *ei;
156 struct extent_buffer *leaf;
157 struct btrfs_key key;
164 * If we don't have skinny metadata, don't bother doing anything
167 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
168 offset = fs_info->nodesize;
172 path = btrfs_alloc_path();
177 path->skip_locking = 1;
178 path->search_commit_root = 1;
182 key.objectid = bytenr;
185 key.type = BTRFS_METADATA_ITEM_KEY;
187 key.type = BTRFS_EXTENT_ITEM_KEY;
189 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
193 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
194 if (path->slots[0]) {
196 btrfs_item_key_to_cpu(path->nodes[0], &key,
198 if (key.objectid == bytenr &&
199 key.type == BTRFS_EXTENT_ITEM_KEY &&
200 key.offset == fs_info->nodesize)
206 leaf = path->nodes[0];
207 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
208 if (item_size >= sizeof(*ei)) {
209 ei = btrfs_item_ptr(leaf, path->slots[0],
210 struct btrfs_extent_item);
211 num_refs = btrfs_extent_refs(leaf, ei);
212 extent_flags = btrfs_extent_flags(leaf, ei);
215 btrfs_print_v0_err(fs_info);
217 btrfs_abort_transaction(trans, ret);
219 btrfs_handle_fs_error(fs_info, ret, NULL);
224 BUG_ON(num_refs == 0);
234 delayed_refs = &trans->transaction->delayed_refs;
235 spin_lock(&delayed_refs->lock);
236 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
238 if (!mutex_trylock(&head->mutex)) {
239 refcount_inc(&head->refs);
240 spin_unlock(&delayed_refs->lock);
242 btrfs_release_path(path);
245 * Mutex was contended, block until it's released and try
248 mutex_lock(&head->mutex);
249 mutex_unlock(&head->mutex);
250 btrfs_put_delayed_ref_head(head);
253 spin_lock(&head->lock);
254 if (head->extent_op && head->extent_op->update_flags)
255 extent_flags |= head->extent_op->flags_to_set;
257 BUG_ON(num_refs == 0);
259 num_refs += head->ref_mod;
260 spin_unlock(&head->lock);
261 mutex_unlock(&head->mutex);
263 spin_unlock(&delayed_refs->lock);
265 WARN_ON(num_refs == 0);
269 *flags = extent_flags;
271 btrfs_free_path(path);
276 * Back reference rules. Back refs have three main goals:
278 * 1) differentiate between all holders of references to an extent so that
279 * when a reference is dropped we can make sure it was a valid reference
280 * before freeing the extent.
282 * 2) Provide enough information to quickly find the holders of an extent
283 * if we notice a given block is corrupted or bad.
285 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
286 * maintenance. This is actually the same as #2, but with a slightly
287 * different use case.
289 * There are two kinds of back refs. The implicit back refs is optimized
290 * for pointers in non-shared tree blocks. For a given pointer in a block,
291 * back refs of this kind provide information about the block's owner tree
292 * and the pointer's key. These information allow us to find the block by
293 * b-tree searching. The full back refs is for pointers in tree blocks not
294 * referenced by their owner trees. The location of tree block is recorded
295 * in the back refs. Actually the full back refs is generic, and can be
296 * used in all cases the implicit back refs is used. The major shortcoming
297 * of the full back refs is its overhead. Every time a tree block gets
298 * COWed, we have to update back refs entry for all pointers in it.
300 * For a newly allocated tree block, we use implicit back refs for
301 * pointers in it. This means most tree related operations only involve
302 * implicit back refs. For a tree block created in old transaction, the
303 * only way to drop a reference to it is COW it. So we can detect the
304 * event that tree block loses its owner tree's reference and do the
305 * back refs conversion.
307 * When a tree block is COWed through a tree, there are four cases:
309 * The reference count of the block is one and the tree is the block's
310 * owner tree. Nothing to do in this case.
312 * The reference count of the block is one and the tree is not the
313 * block's owner tree. In this case, full back refs is used for pointers
314 * in the block. Remove these full back refs, add implicit back refs for
315 * every pointers in the new block.
317 * The reference count of the block is greater than one and the tree is
318 * the block's owner tree. In this case, implicit back refs is used for
319 * pointers in the block. Add full back refs for every pointers in the
320 * block, increase lower level extents' reference counts. The original
321 * implicit back refs are entailed to the new block.
323 * The reference count of the block is greater than one and the tree is
324 * not the block's owner tree. Add implicit back refs for every pointer in
325 * the new block, increase lower level extents' reference count.
327 * Back Reference Key composing:
329 * The key objectid corresponds to the first byte in the extent,
330 * The key type is used to differentiate between types of back refs.
331 * There are different meanings of the key offset for different types
334 * File extents can be referenced by:
336 * - multiple snapshots, subvolumes, or different generations in one subvol
337 * - different files inside a single subvolume
338 * - different offsets inside a file (bookend extents in file.c)
340 * The extent ref structure for the implicit back refs has fields for:
342 * - Objectid of the subvolume root
343 * - objectid of the file holding the reference
344 * - original offset in the file
345 * - how many bookend extents
347 * The key offset for the implicit back refs is hash of the first
350 * The extent ref structure for the full back refs has field for:
352 * - number of pointers in the tree leaf
354 * The key offset for the implicit back refs is the first byte of
357 * When a file extent is allocated, The implicit back refs is used.
358 * the fields are filled in:
360 * (root_key.objectid, inode objectid, offset in file, 1)
362 * When a file extent is removed file truncation, we find the
363 * corresponding implicit back refs and check the following fields:
365 * (btrfs_header_owner(leaf), inode objectid, offset in file)
367 * Btree extents can be referenced by:
369 * - Different subvolumes
371 * Both the implicit back refs and the full back refs for tree blocks
372 * only consist of key. The key offset for the implicit back refs is
373 * objectid of block's owner tree. The key offset for the full back refs
374 * is the first byte of parent block.
376 * When implicit back refs is used, information about the lowest key and
377 * level of the tree block are required. These information are stored in
378 * tree block info structure.
382 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
383 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
384 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
386 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
387 struct btrfs_extent_inline_ref *iref,
388 enum btrfs_inline_ref_type is_data)
390 int type = btrfs_extent_inline_ref_type(eb, iref);
391 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
393 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
394 type == BTRFS_SHARED_BLOCK_REF_KEY ||
395 type == BTRFS_SHARED_DATA_REF_KEY ||
396 type == BTRFS_EXTENT_DATA_REF_KEY) {
397 if (is_data == BTRFS_REF_TYPE_BLOCK) {
398 if (type == BTRFS_TREE_BLOCK_REF_KEY)
400 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
403 * Every shared one has parent tree block,
404 * which must be aligned to sector size.
407 IS_ALIGNED(offset, eb->fs_info->sectorsize))
410 } else if (is_data == BTRFS_REF_TYPE_DATA) {
411 if (type == BTRFS_EXTENT_DATA_REF_KEY)
413 if (type == BTRFS_SHARED_DATA_REF_KEY) {
416 * Every shared one has parent tree block,
417 * which must be aligned to sector size.
420 IS_ALIGNED(offset, eb->fs_info->sectorsize))
424 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
429 btrfs_print_leaf((struct extent_buffer *)eb);
430 btrfs_err(eb->fs_info,
431 "eb %llu iref 0x%lx invalid extent inline ref type %d",
432 eb->start, (unsigned long)iref, type);
435 return BTRFS_REF_TYPE_INVALID;
438 u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
440 u32 high_crc = ~(u32)0;
441 u32 low_crc = ~(u32)0;
444 lenum = cpu_to_le64(root_objectid);
445 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
446 lenum = cpu_to_le64(owner);
447 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
448 lenum = cpu_to_le64(offset);
449 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
451 return ((u64)high_crc << 31) ^ (u64)low_crc;
454 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
455 struct btrfs_extent_data_ref *ref)
457 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
458 btrfs_extent_data_ref_objectid(leaf, ref),
459 btrfs_extent_data_ref_offset(leaf, ref));
462 static int match_extent_data_ref(struct extent_buffer *leaf,
463 struct btrfs_extent_data_ref *ref,
464 u64 root_objectid, u64 owner, u64 offset)
466 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
467 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
468 btrfs_extent_data_ref_offset(leaf, ref) != offset)
473 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
474 struct btrfs_path *path,
475 u64 bytenr, u64 parent,
477 u64 owner, u64 offset)
479 struct btrfs_root *root = trans->fs_info->extent_root;
480 struct btrfs_key key;
481 struct btrfs_extent_data_ref *ref;
482 struct extent_buffer *leaf;
488 key.objectid = bytenr;
490 key.type = BTRFS_SHARED_DATA_REF_KEY;
493 key.type = BTRFS_EXTENT_DATA_REF_KEY;
494 key.offset = hash_extent_data_ref(root_objectid,
499 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
511 leaf = path->nodes[0];
512 nritems = btrfs_header_nritems(leaf);
514 if (path->slots[0] >= nritems) {
515 ret = btrfs_next_leaf(root, path);
521 leaf = path->nodes[0];
522 nritems = btrfs_header_nritems(leaf);
526 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
527 if (key.objectid != bytenr ||
528 key.type != BTRFS_EXTENT_DATA_REF_KEY)
531 ref = btrfs_item_ptr(leaf, path->slots[0],
532 struct btrfs_extent_data_ref);
534 if (match_extent_data_ref(leaf, ref, root_objectid,
537 btrfs_release_path(path);
549 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
550 struct btrfs_path *path,
551 u64 bytenr, u64 parent,
552 u64 root_objectid, u64 owner,
553 u64 offset, int refs_to_add)
555 struct btrfs_root *root = trans->fs_info->extent_root;
556 struct btrfs_key key;
557 struct extent_buffer *leaf;
562 key.objectid = bytenr;
564 key.type = BTRFS_SHARED_DATA_REF_KEY;
566 size = sizeof(struct btrfs_shared_data_ref);
568 key.type = BTRFS_EXTENT_DATA_REF_KEY;
569 key.offset = hash_extent_data_ref(root_objectid,
571 size = sizeof(struct btrfs_extent_data_ref);
574 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
575 if (ret && ret != -EEXIST)
578 leaf = path->nodes[0];
580 struct btrfs_shared_data_ref *ref;
581 ref = btrfs_item_ptr(leaf, path->slots[0],
582 struct btrfs_shared_data_ref);
584 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
586 num_refs = btrfs_shared_data_ref_count(leaf, ref);
587 num_refs += refs_to_add;
588 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
591 struct btrfs_extent_data_ref *ref;
592 while (ret == -EEXIST) {
593 ref = btrfs_item_ptr(leaf, path->slots[0],
594 struct btrfs_extent_data_ref);
595 if (match_extent_data_ref(leaf, ref, root_objectid,
598 btrfs_release_path(path);
600 ret = btrfs_insert_empty_item(trans, root, path, &key,
602 if (ret && ret != -EEXIST)
605 leaf = path->nodes[0];
607 ref = btrfs_item_ptr(leaf, path->slots[0],
608 struct btrfs_extent_data_ref);
610 btrfs_set_extent_data_ref_root(leaf, ref,
612 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
613 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
614 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
616 num_refs = btrfs_extent_data_ref_count(leaf, ref);
617 num_refs += refs_to_add;
618 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
621 btrfs_mark_buffer_dirty(leaf);
624 btrfs_release_path(path);
628 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
629 struct btrfs_path *path,
630 int refs_to_drop, int *last_ref)
632 struct btrfs_key key;
633 struct btrfs_extent_data_ref *ref1 = NULL;
634 struct btrfs_shared_data_ref *ref2 = NULL;
635 struct extent_buffer *leaf;
639 leaf = path->nodes[0];
640 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
642 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
643 ref1 = btrfs_item_ptr(leaf, path->slots[0],
644 struct btrfs_extent_data_ref);
645 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
646 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
647 ref2 = btrfs_item_ptr(leaf, path->slots[0],
648 struct btrfs_shared_data_ref);
649 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
650 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
651 btrfs_print_v0_err(trans->fs_info);
652 btrfs_abort_transaction(trans, -EINVAL);
658 BUG_ON(num_refs < refs_to_drop);
659 num_refs -= refs_to_drop;
662 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
665 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
666 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
667 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
668 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
669 btrfs_mark_buffer_dirty(leaf);
674 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
675 struct btrfs_extent_inline_ref *iref)
677 struct btrfs_key key;
678 struct extent_buffer *leaf;
679 struct btrfs_extent_data_ref *ref1;
680 struct btrfs_shared_data_ref *ref2;
684 leaf = path->nodes[0];
685 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
687 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
690 * If type is invalid, we should have bailed out earlier than
693 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
694 ASSERT(type != BTRFS_REF_TYPE_INVALID);
695 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
696 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
697 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
699 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
700 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
702 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
703 ref1 = btrfs_item_ptr(leaf, path->slots[0],
704 struct btrfs_extent_data_ref);
705 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
706 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
707 ref2 = btrfs_item_ptr(leaf, path->slots[0],
708 struct btrfs_shared_data_ref);
709 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
716 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
717 struct btrfs_path *path,
718 u64 bytenr, u64 parent,
721 struct btrfs_root *root = trans->fs_info->extent_root;
722 struct btrfs_key key;
725 key.objectid = bytenr;
727 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
730 key.type = BTRFS_TREE_BLOCK_REF_KEY;
731 key.offset = root_objectid;
734 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
740 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
741 struct btrfs_path *path,
742 u64 bytenr, u64 parent,
745 struct btrfs_key key;
748 key.objectid = bytenr;
750 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
753 key.type = BTRFS_TREE_BLOCK_REF_KEY;
754 key.offset = root_objectid;
757 ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
759 btrfs_release_path(path);
763 static inline int extent_ref_type(u64 parent, u64 owner)
766 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
768 type = BTRFS_SHARED_BLOCK_REF_KEY;
770 type = BTRFS_TREE_BLOCK_REF_KEY;
773 type = BTRFS_SHARED_DATA_REF_KEY;
775 type = BTRFS_EXTENT_DATA_REF_KEY;
780 static int find_next_key(struct btrfs_path *path, int level,
781 struct btrfs_key *key)
784 for (; level < BTRFS_MAX_LEVEL; level++) {
785 if (!path->nodes[level])
787 if (path->slots[level] + 1 >=
788 btrfs_header_nritems(path->nodes[level]))
791 btrfs_item_key_to_cpu(path->nodes[level], key,
792 path->slots[level] + 1);
794 btrfs_node_key_to_cpu(path->nodes[level], key,
795 path->slots[level] + 1);
802 * look for inline back ref. if back ref is found, *ref_ret is set
803 * to the address of inline back ref, and 0 is returned.
805 * if back ref isn't found, *ref_ret is set to the address where it
806 * should be inserted, and -ENOENT is returned.
808 * if insert is true and there are too many inline back refs, the path
809 * points to the extent item, and -EAGAIN is returned.
811 * NOTE: inline back refs are ordered in the same way that back ref
812 * items in the tree are ordered.
814 static noinline_for_stack
815 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
816 struct btrfs_path *path,
817 struct btrfs_extent_inline_ref **ref_ret,
818 u64 bytenr, u64 num_bytes,
819 u64 parent, u64 root_objectid,
820 u64 owner, u64 offset, int insert)
822 struct btrfs_fs_info *fs_info = trans->fs_info;
823 struct btrfs_root *root = fs_info->extent_root;
824 struct btrfs_key key;
825 struct extent_buffer *leaf;
826 struct btrfs_extent_item *ei;
827 struct btrfs_extent_inline_ref *iref;
837 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
840 key.objectid = bytenr;
841 key.type = BTRFS_EXTENT_ITEM_KEY;
842 key.offset = num_bytes;
844 want = extent_ref_type(parent, owner);
846 extra_size = btrfs_extent_inline_ref_size(want);
847 path->search_for_extension = 1;
848 path->keep_locks = 1;
853 * Owner is our level, so we can just add one to get the level for the
854 * block we are interested in.
856 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
857 key.type = BTRFS_METADATA_ITEM_KEY;
862 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
869 * We may be a newly converted file system which still has the old fat
870 * extent entries for metadata, so try and see if we have one of those.
872 if (ret > 0 && skinny_metadata) {
873 skinny_metadata = false;
874 if (path->slots[0]) {
876 btrfs_item_key_to_cpu(path->nodes[0], &key,
878 if (key.objectid == bytenr &&
879 key.type == BTRFS_EXTENT_ITEM_KEY &&
880 key.offset == num_bytes)
884 key.objectid = bytenr;
885 key.type = BTRFS_EXTENT_ITEM_KEY;
886 key.offset = num_bytes;
887 btrfs_release_path(path);
892 if (ret && !insert) {
895 } else if (WARN_ON(ret)) {
900 leaf = path->nodes[0];
901 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
902 if (unlikely(item_size < sizeof(*ei))) {
904 btrfs_print_v0_err(fs_info);
905 btrfs_abort_transaction(trans, err);
909 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
910 flags = btrfs_extent_flags(leaf, ei);
912 ptr = (unsigned long)(ei + 1);
913 end = (unsigned long)ei + item_size;
915 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
916 ptr += sizeof(struct btrfs_tree_block_info);
920 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
921 needed = BTRFS_REF_TYPE_DATA;
923 needed = BTRFS_REF_TYPE_BLOCK;
931 iref = (struct btrfs_extent_inline_ref *)ptr;
932 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
933 if (type == BTRFS_REF_TYPE_INVALID) {
941 ptr += btrfs_extent_inline_ref_size(type);
945 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
946 struct btrfs_extent_data_ref *dref;
947 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
948 if (match_extent_data_ref(leaf, dref, root_objectid,
953 if (hash_extent_data_ref_item(leaf, dref) <
954 hash_extent_data_ref(root_objectid, owner, offset))
958 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
960 if (parent == ref_offset) {
964 if (ref_offset < parent)
967 if (root_objectid == ref_offset) {
971 if (ref_offset < root_objectid)
975 ptr += btrfs_extent_inline_ref_size(type);
977 if (err == -ENOENT && insert) {
978 if (item_size + extra_size >=
979 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
984 * To add new inline back ref, we have to make sure
985 * there is no corresponding back ref item.
986 * For simplicity, we just do not add new inline back
987 * ref if there is any kind of item for this block
989 if (find_next_key(path, 0, &key) == 0 &&
990 key.objectid == bytenr &&
991 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
996 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
999 path->keep_locks = 0;
1000 path->search_for_extension = 0;
1001 btrfs_unlock_up_safe(path, 1);
1007 * helper to add new inline back ref
1009 static noinline_for_stack
1010 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1011 struct btrfs_path *path,
1012 struct btrfs_extent_inline_ref *iref,
1013 u64 parent, u64 root_objectid,
1014 u64 owner, u64 offset, int refs_to_add,
1015 struct btrfs_delayed_extent_op *extent_op)
1017 struct extent_buffer *leaf;
1018 struct btrfs_extent_item *ei;
1021 unsigned long item_offset;
1026 leaf = path->nodes[0];
1027 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1028 item_offset = (unsigned long)iref - (unsigned long)ei;
1030 type = extent_ref_type(parent, owner);
1031 size = btrfs_extent_inline_ref_size(type);
1033 btrfs_extend_item(path, size);
1035 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1036 refs = btrfs_extent_refs(leaf, ei);
1037 refs += refs_to_add;
1038 btrfs_set_extent_refs(leaf, ei, refs);
1040 __run_delayed_extent_op(extent_op, leaf, ei);
1042 ptr = (unsigned long)ei + item_offset;
1043 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1044 if (ptr < end - size)
1045 memmove_extent_buffer(leaf, ptr + size, ptr,
1048 iref = (struct btrfs_extent_inline_ref *)ptr;
1049 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1050 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1051 struct btrfs_extent_data_ref *dref;
1052 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1053 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1054 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1055 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1056 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1057 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1058 struct btrfs_shared_data_ref *sref;
1059 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1060 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1061 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1062 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1063 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1065 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1067 btrfs_mark_buffer_dirty(leaf);
1070 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1071 struct btrfs_path *path,
1072 struct btrfs_extent_inline_ref **ref_ret,
1073 u64 bytenr, u64 num_bytes, u64 parent,
1074 u64 root_objectid, u64 owner, u64 offset)
1078 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1079 num_bytes, parent, root_objectid,
1084 btrfs_release_path(path);
1087 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1088 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1091 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1092 root_objectid, owner, offset);
1098 * helper to update/remove inline back ref
1100 static noinline_for_stack
1101 void update_inline_extent_backref(struct btrfs_path *path,
1102 struct btrfs_extent_inline_ref *iref,
1104 struct btrfs_delayed_extent_op *extent_op,
1107 struct extent_buffer *leaf = path->nodes[0];
1108 struct btrfs_extent_item *ei;
1109 struct btrfs_extent_data_ref *dref = NULL;
1110 struct btrfs_shared_data_ref *sref = NULL;
1118 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1119 refs = btrfs_extent_refs(leaf, ei);
1120 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1121 refs += refs_to_mod;
1122 btrfs_set_extent_refs(leaf, ei, refs);
1124 __run_delayed_extent_op(extent_op, leaf, ei);
1127 * If type is invalid, we should have bailed out after
1128 * lookup_inline_extent_backref().
1130 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1131 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1133 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1134 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1135 refs = btrfs_extent_data_ref_count(leaf, dref);
1136 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1137 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1138 refs = btrfs_shared_data_ref_count(leaf, sref);
1141 BUG_ON(refs_to_mod != -1);
1144 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1145 refs += refs_to_mod;
1148 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1149 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1151 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1154 size = btrfs_extent_inline_ref_size(type);
1155 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1156 ptr = (unsigned long)iref;
1157 end = (unsigned long)ei + item_size;
1158 if (ptr + size < end)
1159 memmove_extent_buffer(leaf, ptr, ptr + size,
1162 btrfs_truncate_item(path, item_size, 1);
1164 btrfs_mark_buffer_dirty(leaf);
1167 static noinline_for_stack
1168 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1169 struct btrfs_path *path,
1170 u64 bytenr, u64 num_bytes, u64 parent,
1171 u64 root_objectid, u64 owner,
1172 u64 offset, int refs_to_add,
1173 struct btrfs_delayed_extent_op *extent_op)
1175 struct btrfs_extent_inline_ref *iref;
1178 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1179 num_bytes, parent, root_objectid,
1183 * We're adding refs to a tree block we already own, this
1184 * should not happen at all.
1186 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1187 btrfs_crit(trans->fs_info,
1188 "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu",
1189 bytenr, num_bytes, root_objectid);
1190 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) {
1192 btrfs_crit(trans->fs_info,
1193 "path->slots[0]=%d path->nodes[0]:", path->slots[0]);
1194 btrfs_print_leaf(path->nodes[0]);
1198 update_inline_extent_backref(path, iref, refs_to_add,
1200 } else if (ret == -ENOENT) {
1201 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1202 root_objectid, owner, offset,
1203 refs_to_add, extent_op);
1209 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1210 struct btrfs_path *path,
1211 struct btrfs_extent_inline_ref *iref,
1212 int refs_to_drop, int is_data, int *last_ref)
1216 BUG_ON(!is_data && refs_to_drop != 1);
1218 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1220 } else if (is_data) {
1221 ret = remove_extent_data_ref(trans, path, refs_to_drop,
1225 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1230 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1231 u64 *discarded_bytes)
1234 u64 bytes_left, end;
1235 u64 aligned_start = ALIGN(start, 1 << 9);
1237 if (WARN_ON(start != aligned_start)) {
1238 len -= aligned_start - start;
1239 len = round_down(len, 1 << 9);
1240 start = aligned_start;
1243 *discarded_bytes = 0;
1251 /* Skip any superblocks on this device. */
1252 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1253 u64 sb_start = btrfs_sb_offset(j);
1254 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1255 u64 size = sb_start - start;
1257 if (!in_range(sb_start, start, bytes_left) &&
1258 !in_range(sb_end, start, bytes_left) &&
1259 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1263 * Superblock spans beginning of range. Adjust start and
1266 if (sb_start <= start) {
1267 start += sb_end - start;
1272 bytes_left = end - start;
1277 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1280 *discarded_bytes += size;
1281 else if (ret != -EOPNOTSUPP)
1290 bytes_left = end - start;
1294 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1297 *discarded_bytes += bytes_left;
1302 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1303 u64 num_bytes, u64 *actual_bytes)
1306 u64 discarded_bytes = 0;
1307 u64 end = bytenr + num_bytes;
1309 struct btrfs_bio *bbio = NULL;
1313 * Avoid races with device replace and make sure our bbio has devices
1314 * associated to its stripes that don't go away while we are discarding.
1316 btrfs_bio_counter_inc_blocked(fs_info);
1318 struct btrfs_bio_stripe *stripe;
1321 num_bytes = end - cur;
1322 /* Tell the block device(s) that the sectors can be discarded */
1323 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur,
1324 &num_bytes, &bbio, 0);
1326 * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or
1327 * -EOPNOTSUPP. For any such error, @num_bytes is not updated,
1328 * thus we can't continue anyway.
1333 stripe = bbio->stripes;
1334 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1336 struct request_queue *req_q;
1338 if (!stripe->dev->bdev) {
1339 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1342 req_q = bdev_get_queue(stripe->dev->bdev);
1343 if (!blk_queue_discard(req_q))
1346 ret = btrfs_issue_discard(stripe->dev->bdev,
1351 discarded_bytes += bytes;
1352 } else if (ret != -EOPNOTSUPP) {
1354 * Logic errors or -ENOMEM, or -EIO, but
1355 * unlikely to happen.
1357 * And since there are two loops, explicitly
1358 * go to out to avoid confusion.
1360 btrfs_put_bbio(bbio);
1365 * Just in case we get back EOPNOTSUPP for some reason,
1366 * just ignore the return value so we don't screw up
1367 * people calling discard_extent.
1371 btrfs_put_bbio(bbio);
1375 btrfs_bio_counter_dec(fs_info);
1378 *actual_bytes = discarded_bytes;
1381 if (ret == -EOPNOTSUPP)
1386 /* Can return -ENOMEM */
1387 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1388 struct btrfs_ref *generic_ref)
1390 struct btrfs_fs_info *fs_info = trans->fs_info;
1391 int old_ref_mod, new_ref_mod;
1394 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1395 generic_ref->action);
1396 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1397 generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID);
1399 if (generic_ref->type == BTRFS_REF_METADATA)
1400 ret = btrfs_add_delayed_tree_ref(trans, generic_ref,
1401 NULL, &old_ref_mod, &new_ref_mod);
1403 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0,
1404 &old_ref_mod, &new_ref_mod);
1406 btrfs_ref_tree_mod(fs_info, generic_ref);
1408 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
1409 sub_pinned_bytes(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. Otheriwse, 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_key key;
1579 struct btrfs_path *path;
1580 struct btrfs_extent_item *ei;
1581 struct extent_buffer *leaf;
1585 int metadata = !extent_op->is_data;
1587 if (TRANS_ABORTED(trans))
1590 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1593 path = btrfs_alloc_path();
1597 key.objectid = head->bytenr;
1600 key.type = BTRFS_METADATA_ITEM_KEY;
1601 key.offset = extent_op->level;
1603 key.type = BTRFS_EXTENT_ITEM_KEY;
1604 key.offset = head->num_bytes;
1608 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
1615 if (path->slots[0] > 0) {
1617 btrfs_item_key_to_cpu(path->nodes[0], &key,
1619 if (key.objectid == head->bytenr &&
1620 key.type == BTRFS_EXTENT_ITEM_KEY &&
1621 key.offset == head->num_bytes)
1625 btrfs_release_path(path);
1628 key.objectid = head->bytenr;
1629 key.offset = head->num_bytes;
1630 key.type = BTRFS_EXTENT_ITEM_KEY;
1639 leaf = path->nodes[0];
1640 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1642 if (unlikely(item_size < sizeof(*ei))) {
1644 btrfs_print_v0_err(fs_info);
1645 btrfs_abort_transaction(trans, err);
1649 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1650 __run_delayed_extent_op(extent_op, leaf, ei);
1652 btrfs_mark_buffer_dirty(leaf);
1654 btrfs_free_path(path);
1658 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1659 struct btrfs_delayed_ref_node *node,
1660 struct btrfs_delayed_extent_op *extent_op,
1661 int insert_reserved)
1664 struct btrfs_delayed_tree_ref *ref;
1668 ref = btrfs_delayed_node_to_tree_ref(node);
1669 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1671 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1672 parent = ref->parent;
1673 ref_root = ref->root;
1675 if (node->ref_mod != 1) {
1676 btrfs_err(trans->fs_info,
1677 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
1678 node->bytenr, node->ref_mod, node->action, ref_root,
1682 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1683 BUG_ON(!extent_op || !extent_op->update_flags);
1684 ret = alloc_reserved_tree_block(trans, node, extent_op);
1685 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1686 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1687 ref->level, 0, 1, extent_op);
1688 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1689 ret = __btrfs_free_extent(trans, node, parent, ref_root,
1690 ref->level, 0, 1, extent_op);
1697 /* helper function to actually process a single delayed ref entry */
1698 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1699 struct btrfs_delayed_ref_node *node,
1700 struct btrfs_delayed_extent_op *extent_op,
1701 int insert_reserved)
1705 if (TRANS_ABORTED(trans)) {
1706 if (insert_reserved)
1707 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1711 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1712 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1713 ret = run_delayed_tree_ref(trans, node, extent_op,
1715 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1716 node->type == BTRFS_SHARED_DATA_REF_KEY)
1717 ret = run_delayed_data_ref(trans, node, extent_op,
1721 if (ret && insert_reserved)
1722 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1726 static inline struct btrfs_delayed_ref_node *
1727 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1729 struct btrfs_delayed_ref_node *ref;
1731 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1735 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1736 * This is to prevent a ref count from going down to zero, which deletes
1737 * the extent item from the extent tree, when there still are references
1738 * to add, which would fail because they would not find the extent item.
1740 if (!list_empty(&head->ref_add_list))
1741 return list_first_entry(&head->ref_add_list,
1742 struct btrfs_delayed_ref_node, add_list);
1744 ref = rb_entry(rb_first_cached(&head->ref_tree),
1745 struct btrfs_delayed_ref_node, ref_node);
1746 ASSERT(list_empty(&ref->add_list));
1750 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1751 struct btrfs_delayed_ref_head *head)
1753 spin_lock(&delayed_refs->lock);
1754 head->processing = 0;
1755 delayed_refs->num_heads_ready++;
1756 spin_unlock(&delayed_refs->lock);
1757 btrfs_delayed_ref_unlock(head);
1760 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1761 struct btrfs_delayed_ref_head *head)
1763 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1768 if (head->must_insert_reserved) {
1769 head->extent_op = NULL;
1770 btrfs_free_delayed_extent_op(extent_op);
1776 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1777 struct btrfs_delayed_ref_head *head)
1779 struct btrfs_delayed_extent_op *extent_op;
1782 extent_op = cleanup_extent_op(head);
1785 head->extent_op = NULL;
1786 spin_unlock(&head->lock);
1787 ret = run_delayed_extent_op(trans, head, extent_op);
1788 btrfs_free_delayed_extent_op(extent_op);
1789 return ret ? ret : 1;
1792 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1793 struct btrfs_delayed_ref_root *delayed_refs,
1794 struct btrfs_delayed_ref_head *head)
1796 int nr_items = 1; /* Dropping this ref head update. */
1798 if (head->total_ref_mod < 0) {
1799 struct btrfs_space_info *space_info;
1803 flags = BTRFS_BLOCK_GROUP_DATA;
1804 else if (head->is_system)
1805 flags = BTRFS_BLOCK_GROUP_SYSTEM;
1807 flags = BTRFS_BLOCK_GROUP_METADATA;
1808 space_info = btrfs_find_space_info(fs_info, flags);
1810 percpu_counter_add_batch(&space_info->total_bytes_pinned,
1812 BTRFS_TOTAL_BYTES_PINNED_BATCH);
1815 * We had csum deletions accounted for in our delayed refs rsv,
1816 * we need to drop the csum leaves for this update from our
1819 if (head->is_data) {
1820 spin_lock(&delayed_refs->lock);
1821 delayed_refs->pending_csums -= head->num_bytes;
1822 spin_unlock(&delayed_refs->lock);
1823 nr_items += btrfs_csum_bytes_to_leaves(fs_info,
1828 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1831 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1832 struct btrfs_delayed_ref_head *head)
1835 struct btrfs_fs_info *fs_info = trans->fs_info;
1836 struct btrfs_delayed_ref_root *delayed_refs;
1839 delayed_refs = &trans->transaction->delayed_refs;
1841 ret = run_and_cleanup_extent_op(trans, head);
1843 unselect_delayed_ref_head(delayed_refs, head);
1844 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1851 * Need to drop our head ref lock and re-acquire the delayed ref lock
1852 * and then re-check to make sure nobody got added.
1854 spin_unlock(&head->lock);
1855 spin_lock(&delayed_refs->lock);
1856 spin_lock(&head->lock);
1857 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1858 spin_unlock(&head->lock);
1859 spin_unlock(&delayed_refs->lock);
1862 btrfs_delete_ref_head(delayed_refs, head);
1863 spin_unlock(&head->lock);
1864 spin_unlock(&delayed_refs->lock);
1866 if (head->must_insert_reserved) {
1867 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1868 if (head->is_data) {
1869 ret = btrfs_del_csums(trans, fs_info->csum_root,
1870 head->bytenr, head->num_bytes);
1874 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1876 trace_run_delayed_ref_head(fs_info, head, 0);
1877 btrfs_delayed_ref_unlock(head);
1878 btrfs_put_delayed_ref_head(head);
1882 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1883 struct btrfs_trans_handle *trans)
1885 struct btrfs_delayed_ref_root *delayed_refs =
1886 &trans->transaction->delayed_refs;
1887 struct btrfs_delayed_ref_head *head = NULL;
1890 spin_lock(&delayed_refs->lock);
1891 head = btrfs_select_ref_head(delayed_refs);
1893 spin_unlock(&delayed_refs->lock);
1898 * Grab the lock that says we are going to process all the refs for
1901 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1902 spin_unlock(&delayed_refs->lock);
1905 * We may have dropped the spin lock to get the head mutex lock, and
1906 * that might have given someone else time to free the head. If that's
1907 * true, it has been removed from our list and we can move on.
1910 head = ERR_PTR(-EAGAIN);
1915 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1916 struct btrfs_delayed_ref_head *locked_ref,
1917 unsigned long *run_refs)
1919 struct btrfs_fs_info *fs_info = trans->fs_info;
1920 struct btrfs_delayed_ref_root *delayed_refs;
1921 struct btrfs_delayed_extent_op *extent_op;
1922 struct btrfs_delayed_ref_node *ref;
1923 int must_insert_reserved = 0;
1926 delayed_refs = &trans->transaction->delayed_refs;
1928 lockdep_assert_held(&locked_ref->mutex);
1929 lockdep_assert_held(&locked_ref->lock);
1931 while ((ref = select_delayed_ref(locked_ref))) {
1933 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1934 spin_unlock(&locked_ref->lock);
1935 unselect_delayed_ref_head(delayed_refs, locked_ref);
1941 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1942 RB_CLEAR_NODE(&ref->ref_node);
1943 if (!list_empty(&ref->add_list))
1944 list_del(&ref->add_list);
1946 * When we play the delayed ref, also correct the ref_mod on
1949 switch (ref->action) {
1950 case BTRFS_ADD_DELAYED_REF:
1951 case BTRFS_ADD_DELAYED_EXTENT:
1952 locked_ref->ref_mod -= ref->ref_mod;
1954 case BTRFS_DROP_DELAYED_REF:
1955 locked_ref->ref_mod += ref->ref_mod;
1960 atomic_dec(&delayed_refs->num_entries);
1963 * Record the must_insert_reserved flag before we drop the
1966 must_insert_reserved = locked_ref->must_insert_reserved;
1967 locked_ref->must_insert_reserved = 0;
1969 extent_op = locked_ref->extent_op;
1970 locked_ref->extent_op = NULL;
1971 spin_unlock(&locked_ref->lock);
1973 ret = run_one_delayed_ref(trans, ref, extent_op,
1974 must_insert_reserved);
1976 btrfs_free_delayed_extent_op(extent_op);
1978 unselect_delayed_ref_head(delayed_refs, locked_ref);
1979 btrfs_put_delayed_ref(ref);
1980 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
1985 btrfs_put_delayed_ref(ref);
1988 spin_lock(&locked_ref->lock);
1989 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
1996 * Returns 0 on success or if called with an already aborted transaction.
1997 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
1999 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2002 struct btrfs_fs_info *fs_info = trans->fs_info;
2003 struct btrfs_delayed_ref_root *delayed_refs;
2004 struct btrfs_delayed_ref_head *locked_ref = NULL;
2005 ktime_t start = ktime_get();
2007 unsigned long count = 0;
2008 unsigned long actual_count = 0;
2010 delayed_refs = &trans->transaction->delayed_refs;
2013 locked_ref = btrfs_obtain_ref_head(trans);
2014 if (IS_ERR_OR_NULL(locked_ref)) {
2015 if (PTR_ERR(locked_ref) == -EAGAIN) {
2024 * We need to try and merge add/drops of the same ref since we
2025 * can run into issues with relocate dropping the implicit ref
2026 * and then it being added back again before the drop can
2027 * finish. If we merged anything we need to re-loop so we can
2029 * Or we can get node references of the same type that weren't
2030 * merged when created due to bumps in the tree mod seq, and
2031 * we need to merge them to prevent adding an inline extent
2032 * backref before dropping it (triggering a BUG_ON at
2033 * insert_inline_extent_backref()).
2035 spin_lock(&locked_ref->lock);
2036 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2038 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2040 if (ret < 0 && ret != -EAGAIN) {
2042 * Error, btrfs_run_delayed_refs_for_head already
2043 * unlocked everything so just bail out
2048 * Success, perform the usual cleanup of a processed
2051 ret = cleanup_ref_head(trans, locked_ref);
2053 /* We dropped our lock, we need to loop. */
2062 * Either success case or btrfs_run_delayed_refs_for_head
2063 * returned -EAGAIN, meaning we need to select another head
2068 } while ((nr != -1 && count < nr) || locked_ref);
2071 * We don't want to include ref heads since we can have empty ref heads
2072 * and those will drastically skew our runtime down since we just do
2073 * accounting, no actual extent tree updates.
2075 if (actual_count > 0) {
2076 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2080 * We weigh the current average higher than our current runtime
2081 * to avoid large swings in the average.
2083 spin_lock(&delayed_refs->lock);
2084 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2085 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2086 spin_unlock(&delayed_refs->lock);
2091 #ifdef SCRAMBLE_DELAYED_REFS
2093 * Normally delayed refs get processed in ascending bytenr order. This
2094 * correlates in most cases to the order added. To expose dependencies on this
2095 * order, we start to process the tree in the middle instead of the beginning
2097 static u64 find_middle(struct rb_root *root)
2099 struct rb_node *n = root->rb_node;
2100 struct btrfs_delayed_ref_node *entry;
2103 u64 first = 0, last = 0;
2107 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2108 first = entry->bytenr;
2112 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2113 last = entry->bytenr;
2118 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2119 WARN_ON(!entry->in_tree);
2121 middle = entry->bytenr;
2135 * this starts processing the delayed reference count updates and
2136 * extent insertions we have queued up so far. count can be
2137 * 0, which means to process everything in the tree at the start
2138 * of the run (but not newly added entries), or it can be some target
2139 * number you'd like to process.
2141 * Returns 0 on success or if called with an aborted transaction
2142 * Returns <0 on error and aborts the transaction
2144 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2145 unsigned long count)
2147 struct btrfs_fs_info *fs_info = trans->fs_info;
2148 struct rb_node *node;
2149 struct btrfs_delayed_ref_root *delayed_refs;
2150 struct btrfs_delayed_ref_head *head;
2152 int run_all = count == (unsigned long)-1;
2154 /* We'll clean this up in btrfs_cleanup_transaction */
2155 if (TRANS_ABORTED(trans))
2158 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2161 delayed_refs = &trans->transaction->delayed_refs;
2163 count = atomic_read(&delayed_refs->num_entries) * 2;
2166 #ifdef SCRAMBLE_DELAYED_REFS
2167 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2169 ret = __btrfs_run_delayed_refs(trans, count);
2171 btrfs_abort_transaction(trans, ret);
2176 btrfs_create_pending_block_groups(trans);
2178 spin_lock(&delayed_refs->lock);
2179 node = rb_first_cached(&delayed_refs->href_root);
2181 spin_unlock(&delayed_refs->lock);
2184 head = rb_entry(node, struct btrfs_delayed_ref_head,
2186 refcount_inc(&head->refs);
2187 spin_unlock(&delayed_refs->lock);
2189 /* Mutex was contended, block until it's released and retry. */
2190 mutex_lock(&head->mutex);
2191 mutex_unlock(&head->mutex);
2193 btrfs_put_delayed_ref_head(head);
2201 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2202 struct extent_buffer *eb, u64 flags,
2203 int level, int is_data)
2205 struct btrfs_delayed_extent_op *extent_op;
2208 extent_op = btrfs_alloc_delayed_extent_op();
2212 extent_op->flags_to_set = flags;
2213 extent_op->update_flags = true;
2214 extent_op->update_key = false;
2215 extent_op->is_data = is_data ? true : false;
2216 extent_op->level = level;
2218 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2220 btrfs_free_delayed_extent_op(extent_op);
2224 static noinline int check_delayed_ref(struct btrfs_root *root,
2225 struct btrfs_path *path,
2226 u64 objectid, u64 offset, u64 bytenr)
2228 struct btrfs_delayed_ref_head *head;
2229 struct btrfs_delayed_ref_node *ref;
2230 struct btrfs_delayed_data_ref *data_ref;
2231 struct btrfs_delayed_ref_root *delayed_refs;
2232 struct btrfs_transaction *cur_trans;
2233 struct rb_node *node;
2236 spin_lock(&root->fs_info->trans_lock);
2237 cur_trans = root->fs_info->running_transaction;
2239 refcount_inc(&cur_trans->use_count);
2240 spin_unlock(&root->fs_info->trans_lock);
2244 delayed_refs = &cur_trans->delayed_refs;
2245 spin_lock(&delayed_refs->lock);
2246 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2248 spin_unlock(&delayed_refs->lock);
2249 btrfs_put_transaction(cur_trans);
2253 if (!mutex_trylock(&head->mutex)) {
2254 refcount_inc(&head->refs);
2255 spin_unlock(&delayed_refs->lock);
2257 btrfs_release_path(path);
2260 * Mutex was contended, block until it's released and let
2263 mutex_lock(&head->mutex);
2264 mutex_unlock(&head->mutex);
2265 btrfs_put_delayed_ref_head(head);
2266 btrfs_put_transaction(cur_trans);
2269 spin_unlock(&delayed_refs->lock);
2271 spin_lock(&head->lock);
2273 * XXX: We should replace this with a proper search function in the
2276 for (node = rb_first_cached(&head->ref_tree); node;
2277 node = rb_next(node)) {
2278 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2279 /* If it's a shared ref we know a cross reference exists */
2280 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2285 data_ref = btrfs_delayed_node_to_data_ref(ref);
2288 * If our ref doesn't match the one we're currently looking at
2289 * then we have a cross reference.
2291 if (data_ref->root != root->root_key.objectid ||
2292 data_ref->objectid != objectid ||
2293 data_ref->offset != offset) {
2298 spin_unlock(&head->lock);
2299 mutex_unlock(&head->mutex);
2300 btrfs_put_transaction(cur_trans);
2304 static noinline int check_committed_ref(struct btrfs_root *root,
2305 struct btrfs_path *path,
2306 u64 objectid, u64 offset, u64 bytenr,
2309 struct btrfs_fs_info *fs_info = root->fs_info;
2310 struct btrfs_root *extent_root = fs_info->extent_root;
2311 struct extent_buffer *leaf;
2312 struct btrfs_extent_data_ref *ref;
2313 struct btrfs_extent_inline_ref *iref;
2314 struct btrfs_extent_item *ei;
2315 struct btrfs_key key;
2320 key.objectid = bytenr;
2321 key.offset = (u64)-1;
2322 key.type = BTRFS_EXTENT_ITEM_KEY;
2324 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2327 BUG_ON(ret == 0); /* Corruption */
2330 if (path->slots[0] == 0)
2334 leaf = path->nodes[0];
2335 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2337 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2341 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2342 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2344 /* If extent item has more than 1 inline ref then it's shared */
2345 if (item_size != sizeof(*ei) +
2346 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2350 * If extent created before last snapshot => it's shared unless the
2351 * snapshot has been deleted. Use the heuristic if strict is false.
2354 (btrfs_extent_generation(leaf, ei) <=
2355 btrfs_root_last_snapshot(&root->root_item)))
2358 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2360 /* If this extent has SHARED_DATA_REF then it's shared */
2361 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2362 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2365 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2366 if (btrfs_extent_refs(leaf, ei) !=
2367 btrfs_extent_data_ref_count(leaf, ref) ||
2368 btrfs_extent_data_ref_root(leaf, ref) !=
2369 root->root_key.objectid ||
2370 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2371 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2379 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2380 u64 bytenr, bool strict)
2382 struct btrfs_path *path;
2385 path = btrfs_alloc_path();
2390 ret = check_committed_ref(root, path, objectid,
2391 offset, bytenr, strict);
2392 if (ret && ret != -ENOENT)
2395 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2396 } while (ret == -EAGAIN);
2399 btrfs_free_path(path);
2400 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2405 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2406 struct btrfs_root *root,
2407 struct extent_buffer *buf,
2408 int full_backref, int inc)
2410 struct btrfs_fs_info *fs_info = root->fs_info;
2416 struct btrfs_key key;
2417 struct btrfs_file_extent_item *fi;
2418 struct btrfs_ref generic_ref = { 0 };
2419 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2425 if (btrfs_is_testing(fs_info))
2428 ref_root = btrfs_header_owner(buf);
2429 nritems = btrfs_header_nritems(buf);
2430 level = btrfs_header_level(buf);
2432 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2436 parent = buf->start;
2440 action = BTRFS_ADD_DELAYED_REF;
2442 action = BTRFS_DROP_DELAYED_REF;
2444 for (i = 0; i < nritems; i++) {
2446 btrfs_item_key_to_cpu(buf, &key, i);
2447 if (key.type != BTRFS_EXTENT_DATA_KEY)
2449 fi = btrfs_item_ptr(buf, i,
2450 struct btrfs_file_extent_item);
2451 if (btrfs_file_extent_type(buf, fi) ==
2452 BTRFS_FILE_EXTENT_INLINE)
2454 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2458 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2459 key.offset -= btrfs_file_extent_offset(buf, fi);
2460 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2462 generic_ref.real_root = root->root_key.objectid;
2463 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2465 generic_ref.skip_qgroup = for_reloc;
2467 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2469 ret = btrfs_free_extent(trans, &generic_ref);
2473 bytenr = btrfs_node_blockptr(buf, i);
2474 num_bytes = fs_info->nodesize;
2475 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2477 generic_ref.real_root = root->root_key.objectid;
2478 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root);
2479 generic_ref.skip_qgroup = for_reloc;
2481 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2483 ret = btrfs_free_extent(trans, &generic_ref);
2493 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2494 struct extent_buffer *buf, int full_backref)
2496 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2499 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2500 struct extent_buffer *buf, int full_backref)
2502 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2505 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
2507 struct btrfs_block_group *block_group;
2510 block_group = btrfs_lookup_block_group(fs_info, bytenr);
2511 if (!block_group || block_group->ro)
2514 btrfs_put_block_group(block_group);
2518 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2520 struct btrfs_fs_info *fs_info = root->fs_info;
2525 flags = BTRFS_BLOCK_GROUP_DATA;
2526 else if (root == fs_info->chunk_root)
2527 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2529 flags = BTRFS_BLOCK_GROUP_METADATA;
2531 ret = btrfs_get_alloc_profile(fs_info, flags);
2535 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
2537 struct btrfs_block_group *cache;
2540 spin_lock(&fs_info->block_group_cache_lock);
2541 bytenr = fs_info->first_logical_byte;
2542 spin_unlock(&fs_info->block_group_cache_lock);
2544 if (bytenr < (u64)-1)
2547 cache = btrfs_lookup_first_block_group(fs_info, search_start);
2551 bytenr = cache->start;
2552 btrfs_put_block_group(cache);
2557 static int pin_down_extent(struct btrfs_trans_handle *trans,
2558 struct btrfs_block_group *cache,
2559 u64 bytenr, u64 num_bytes, int reserved)
2561 struct btrfs_fs_info *fs_info = cache->fs_info;
2563 spin_lock(&cache->space_info->lock);
2564 spin_lock(&cache->lock);
2565 cache->pinned += num_bytes;
2566 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2569 cache->reserved -= num_bytes;
2570 cache->space_info->bytes_reserved -= num_bytes;
2572 spin_unlock(&cache->lock);
2573 spin_unlock(&cache->space_info->lock);
2575 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
2576 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
2577 set_extent_dirty(&trans->transaction->pinned_extents, bytenr,
2578 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
2582 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2583 u64 bytenr, u64 num_bytes, int reserved)
2585 struct btrfs_block_group *cache;
2587 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2588 BUG_ON(!cache); /* Logic error */
2590 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2592 btrfs_put_block_group(cache);
2597 * this function must be called within transaction
2599 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2600 u64 bytenr, u64 num_bytes)
2602 struct btrfs_block_group *cache;
2605 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2610 * pull in the free space cache (if any) so that our pin
2611 * removes the free space from the cache. We have load_only set
2612 * to one because the slow code to read in the free extents does check
2613 * the pinned extents.
2615 btrfs_cache_block_group(cache, 1);
2617 * Make sure we wait until the cache is completely built in case it is
2618 * missing or is invalid and therefore needs to be rebuilt.
2620 ret = btrfs_wait_block_group_cache_done(cache);
2624 pin_down_extent(trans, cache, bytenr, num_bytes, 0);
2626 /* remove us from the free space cache (if we're there at all) */
2627 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2629 btrfs_put_block_group(cache);
2633 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2634 u64 start, u64 num_bytes)
2637 struct btrfs_block_group *block_group;
2639 block_group = btrfs_lookup_block_group(fs_info, start);
2643 btrfs_cache_block_group(block_group, 1);
2645 * Make sure we wait until the cache is completely built in case it is
2646 * missing or is invalid and therefore needs to be rebuilt.
2648 ret = btrfs_wait_block_group_cache_done(block_group);
2652 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2654 btrfs_put_block_group(block_group);
2658 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2660 struct btrfs_fs_info *fs_info = eb->fs_info;
2661 struct btrfs_file_extent_item *item;
2662 struct btrfs_key key;
2667 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2670 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2671 btrfs_item_key_to_cpu(eb, &key, i);
2672 if (key.type != BTRFS_EXTENT_DATA_KEY)
2674 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2675 found_type = btrfs_file_extent_type(eb, item);
2676 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2678 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2680 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2681 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2682 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2691 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2693 atomic_inc(&bg->reservations);
2697 * Returns the free cluster for the given space info and sets empty_cluster to
2698 * what it should be based on the mount options.
2700 static struct btrfs_free_cluster *
2701 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2702 struct btrfs_space_info *space_info, u64 *empty_cluster)
2704 struct btrfs_free_cluster *ret = NULL;
2707 if (btrfs_mixed_space_info(space_info))
2710 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2711 ret = &fs_info->meta_alloc_cluster;
2712 if (btrfs_test_opt(fs_info, SSD))
2713 *empty_cluster = SZ_2M;
2715 *empty_cluster = SZ_64K;
2716 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2717 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2718 *empty_cluster = SZ_2M;
2719 ret = &fs_info->data_alloc_cluster;
2725 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2727 const bool return_free_space)
2729 struct btrfs_block_group *cache = NULL;
2730 struct btrfs_space_info *space_info;
2731 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2732 struct btrfs_free_cluster *cluster = NULL;
2734 u64 total_unpinned = 0;
2735 u64 empty_cluster = 0;
2738 while (start <= end) {
2741 start >= cache->start + cache->length) {
2743 btrfs_put_block_group(cache);
2745 cache = btrfs_lookup_block_group(fs_info, start);
2746 BUG_ON(!cache); /* Logic error */
2748 cluster = fetch_cluster_info(fs_info,
2751 empty_cluster <<= 1;
2754 len = cache->start + cache->length - start;
2755 len = min(len, end + 1 - start);
2757 down_read(&fs_info->commit_root_sem);
2758 if (start < cache->last_byte_to_unpin && return_free_space) {
2759 u64 add_len = min(len, cache->last_byte_to_unpin - start);
2761 btrfs_add_free_space(cache, start, add_len);
2763 up_read(&fs_info->commit_root_sem);
2766 total_unpinned += len;
2767 space_info = cache->space_info;
2770 * If this space cluster has been marked as fragmented and we've
2771 * unpinned enough in this block group to potentially allow a
2772 * cluster to be created inside of it go ahead and clear the
2775 if (cluster && cluster->fragmented &&
2776 total_unpinned > empty_cluster) {
2777 spin_lock(&cluster->lock);
2778 cluster->fragmented = 0;
2779 spin_unlock(&cluster->lock);
2782 spin_lock(&space_info->lock);
2783 spin_lock(&cache->lock);
2784 cache->pinned -= len;
2785 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2786 space_info->max_extent_size = 0;
2787 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2788 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
2790 space_info->bytes_readonly += len;
2793 spin_unlock(&cache->lock);
2794 if (!readonly && return_free_space &&
2795 global_rsv->space_info == space_info) {
2798 spin_lock(&global_rsv->lock);
2799 if (!global_rsv->full) {
2800 to_add = min(len, global_rsv->size -
2801 global_rsv->reserved);
2802 global_rsv->reserved += to_add;
2803 btrfs_space_info_update_bytes_may_use(fs_info,
2804 space_info, to_add);
2805 if (global_rsv->reserved >= global_rsv->size)
2806 global_rsv->full = 1;
2809 spin_unlock(&global_rsv->lock);
2811 /* Add to any tickets we may have */
2812 if (!readonly && return_free_space && len)
2813 btrfs_try_granting_tickets(fs_info, space_info);
2814 spin_unlock(&space_info->lock);
2818 btrfs_put_block_group(cache);
2822 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2824 struct btrfs_fs_info *fs_info = trans->fs_info;
2825 struct btrfs_block_group *block_group, *tmp;
2826 struct list_head *deleted_bgs;
2827 struct extent_io_tree *unpin;
2832 unpin = &trans->transaction->pinned_extents;
2834 while (!TRANS_ABORTED(trans)) {
2835 struct extent_state *cached_state = NULL;
2837 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2838 ret = find_first_extent_bit(unpin, 0, &start, &end,
2839 EXTENT_DIRTY, &cached_state);
2841 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2845 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2846 ret = btrfs_discard_extent(fs_info, start,
2847 end + 1 - start, NULL);
2849 clear_extent_dirty(unpin, start, end, &cached_state);
2850 unpin_extent_range(fs_info, start, end, true);
2851 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2852 free_extent_state(cached_state);
2856 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2857 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2858 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2862 * Transaction is finished. We don't need the lock anymore. We
2863 * do need to clean up the block groups in case of a transaction
2866 deleted_bgs = &trans->transaction->deleted_bgs;
2867 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2871 if (!TRANS_ABORTED(trans))
2872 ret = btrfs_discard_extent(fs_info,
2874 block_group->length,
2877 list_del_init(&block_group->bg_list);
2878 btrfs_unfreeze_block_group(block_group);
2879 btrfs_put_block_group(block_group);
2882 const char *errstr = btrfs_decode_error(ret);
2884 "discard failed while removing blockgroup: errno=%d %s",
2893 * Drop one or more refs of @node.
2895 * 1. Locate the extent refs.
2896 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2897 * Locate it, then reduce the refs number or remove the ref line completely.
2899 * 2. Update the refs count in EXTENT/METADATA_ITEM
2901 * Inline backref case:
2903 * in extent tree we have:
2905 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2906 * refs 2 gen 6 flags DATA
2907 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2908 * extent data backref root FS_TREE objectid 257 offset 0 count 1
2910 * This function gets called with:
2912 * node->bytenr = 13631488
2913 * node->num_bytes = 1048576
2914 * root_objectid = FS_TREE
2915 * owner_objectid = 257
2919 * Then we should get some like:
2921 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2922 * refs 1 gen 6 flags DATA
2923 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2925 * Keyed backref case:
2927 * in extent tree we have:
2929 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2930 * refs 754 gen 6 flags DATA
2932 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2933 * extent data backref root FS_TREE objectid 866 offset 0 count 1
2935 * This function get called with:
2937 * node->bytenr = 13631488
2938 * node->num_bytes = 1048576
2939 * root_objectid = FS_TREE
2940 * owner_objectid = 866
2944 * Then we should get some like:
2946 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2947 * refs 753 gen 6 flags DATA
2949 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2951 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2952 struct btrfs_delayed_ref_node *node, u64 parent,
2953 u64 root_objectid, u64 owner_objectid,
2954 u64 owner_offset, int refs_to_drop,
2955 struct btrfs_delayed_extent_op *extent_op)
2957 struct btrfs_fs_info *info = trans->fs_info;
2958 struct btrfs_key key;
2959 struct btrfs_path *path;
2960 struct btrfs_root *extent_root = info->extent_root;
2961 struct extent_buffer *leaf;
2962 struct btrfs_extent_item *ei;
2963 struct btrfs_extent_inline_ref *iref;
2966 int extent_slot = 0;
2967 int found_extent = 0;
2971 u64 bytenr = node->bytenr;
2972 u64 num_bytes = node->num_bytes;
2974 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2976 path = btrfs_alloc_path();
2980 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2982 if (!is_data && refs_to_drop != 1) {
2984 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2985 node->bytenr, refs_to_drop);
2987 btrfs_abort_transaction(trans, ret);
2992 skinny_metadata = false;
2994 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2995 parent, root_objectid, owner_objectid,
2999 * Either the inline backref or the SHARED_DATA_REF/
3000 * SHARED_BLOCK_REF is found
3002 * Here is a quick path to locate EXTENT/METADATA_ITEM.
3003 * It's possible the EXTENT/METADATA_ITEM is near current slot.
3005 extent_slot = path->slots[0];
3006 while (extent_slot >= 0) {
3007 btrfs_item_key_to_cpu(path->nodes[0], &key,
3009 if (key.objectid != bytenr)
3011 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3012 key.offset == num_bytes) {
3016 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3017 key.offset == owner_objectid) {
3022 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
3023 if (path->slots[0] - extent_slot > 5)
3028 if (!found_extent) {
3031 "invalid iref, no EXTENT/METADATA_ITEM found but has inline extent ref");
3032 btrfs_abort_transaction(trans, -EUCLEAN);
3035 /* Must be SHARED_* item, remove the backref first */
3036 ret = remove_extent_backref(trans, path, NULL,
3038 is_data, &last_ref);
3040 btrfs_abort_transaction(trans, ret);
3043 btrfs_release_path(path);
3045 /* Slow path to locate EXTENT/METADATA_ITEM */
3046 key.objectid = bytenr;
3047 key.type = BTRFS_EXTENT_ITEM_KEY;
3048 key.offset = num_bytes;
3050 if (!is_data && skinny_metadata) {
3051 key.type = BTRFS_METADATA_ITEM_KEY;
3052 key.offset = owner_objectid;
3055 ret = btrfs_search_slot(trans, extent_root,
3057 if (ret > 0 && skinny_metadata && path->slots[0]) {
3059 * Couldn't find our skinny metadata item,
3060 * see if we have ye olde extent item.
3063 btrfs_item_key_to_cpu(path->nodes[0], &key,
3065 if (key.objectid == bytenr &&
3066 key.type == BTRFS_EXTENT_ITEM_KEY &&
3067 key.offset == num_bytes)
3071 if (ret > 0 && skinny_metadata) {
3072 skinny_metadata = false;
3073 key.objectid = bytenr;
3074 key.type = BTRFS_EXTENT_ITEM_KEY;
3075 key.offset = num_bytes;
3076 btrfs_release_path(path);
3077 ret = btrfs_search_slot(trans, extent_root,
3083 "umm, got %d back from search, was looking for %llu",
3086 btrfs_print_leaf(path->nodes[0]);
3089 btrfs_abort_transaction(trans, ret);
3092 extent_slot = path->slots[0];
3094 } else if (WARN_ON(ret == -ENOENT)) {
3095 btrfs_print_leaf(path->nodes[0]);
3097 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
3098 bytenr, parent, root_objectid, owner_objectid,
3100 btrfs_abort_transaction(trans, ret);
3103 btrfs_abort_transaction(trans, ret);
3107 leaf = path->nodes[0];
3108 item_size = btrfs_item_size_nr(leaf, extent_slot);
3109 if (unlikely(item_size < sizeof(*ei))) {
3111 btrfs_print_v0_err(info);
3112 btrfs_abort_transaction(trans, ret);
3115 ei = btrfs_item_ptr(leaf, extent_slot,
3116 struct btrfs_extent_item);
3117 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3118 key.type == BTRFS_EXTENT_ITEM_KEY) {
3119 struct btrfs_tree_block_info *bi;
3120 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3122 "invalid extent item size for key (%llu, %u, %llu) owner %llu, has %u expect >= %zu",
3123 key.objectid, key.type, key.offset,
3124 owner_objectid, item_size,
3125 sizeof(*ei) + sizeof(*bi));
3126 btrfs_abort_transaction(trans, -EUCLEAN);
3129 bi = (struct btrfs_tree_block_info *)(ei + 1);
3130 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3133 refs = btrfs_extent_refs(leaf, ei);
3134 if (refs < refs_to_drop) {
3136 "trying to drop %d refs but we only have %llu for bytenr %llu",
3137 refs_to_drop, refs, bytenr);
3138 btrfs_abort_transaction(trans, -EUCLEAN);
3141 refs -= refs_to_drop;
3145 __run_delayed_extent_op(extent_op, leaf, ei);
3147 * In the case of inline back ref, reference count will
3148 * be updated by remove_extent_backref
3151 if (!found_extent) {
3153 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found");
3154 btrfs_abort_transaction(trans, -EUCLEAN);
3158 btrfs_set_extent_refs(leaf, ei, refs);
3159 btrfs_mark_buffer_dirty(leaf);
3162 ret = remove_extent_backref(trans, path, iref,
3163 refs_to_drop, is_data,
3166 btrfs_abort_transaction(trans, ret);
3171 /* In this branch refs == 1 */
3173 if (is_data && refs_to_drop !=
3174 extent_data_ref_count(path, iref)) {
3176 "invalid refs_to_drop, current refs %u refs_to_drop %u",
3177 extent_data_ref_count(path, iref),
3179 btrfs_abort_transaction(trans, -EUCLEAN);
3183 if (path->slots[0] != extent_slot) {
3185 "invalid iref, extent item key (%llu %u %llu) doesn't have wanted iref",
3186 key.objectid, key.type,
3188 btrfs_abort_transaction(trans, -EUCLEAN);
3193 * No inline ref, we must be at SHARED_* item,
3194 * And it's single ref, it must be:
3195 * | extent_slot ||extent_slot + 1|
3196 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3198 if (path->slots[0] != extent_slot + 1) {
3200 "invalid SHARED_* item, previous item is not EXTENT/METADATA_ITEM");
3201 btrfs_abort_transaction(trans, -EUCLEAN);
3204 path->slots[0] = extent_slot;
3210 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3213 btrfs_abort_transaction(trans, ret);
3216 btrfs_release_path(path);
3219 ret = btrfs_del_csums(trans, info->csum_root, bytenr,
3222 btrfs_abort_transaction(trans, ret);
3227 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
3229 btrfs_abort_transaction(trans, ret);
3233 ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0);
3235 btrfs_abort_transaction(trans, ret);
3239 btrfs_release_path(path);
3242 btrfs_free_path(path);
3246 * Leaf dump can take up a lot of log buffer, so we only do full leaf
3247 * dump for debug build.
3249 if (IS_ENABLED(CONFIG_BTRFS_DEBUG)) {
3250 btrfs_crit(info, "path->slots[0]=%d extent_slot=%d",
3251 path->slots[0], extent_slot);
3252 btrfs_print_leaf(path->nodes[0]);
3255 btrfs_free_path(path);
3260 * when we free an block, it is possible (and likely) that we free the last
3261 * delayed ref for that extent as well. This searches the delayed ref tree for
3262 * a given extent, and if there are no other delayed refs to be processed, it
3263 * removes it from the tree.
3265 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3268 struct btrfs_delayed_ref_head *head;
3269 struct btrfs_delayed_ref_root *delayed_refs;
3272 delayed_refs = &trans->transaction->delayed_refs;
3273 spin_lock(&delayed_refs->lock);
3274 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3276 goto out_delayed_unlock;
3278 spin_lock(&head->lock);
3279 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3282 if (cleanup_extent_op(head) != NULL)
3286 * waiting for the lock here would deadlock. If someone else has it
3287 * locked they are already in the process of dropping it anyway
3289 if (!mutex_trylock(&head->mutex))
3292 btrfs_delete_ref_head(delayed_refs, head);
3293 head->processing = 0;
3295 spin_unlock(&head->lock);
3296 spin_unlock(&delayed_refs->lock);
3298 BUG_ON(head->extent_op);
3299 if (head->must_insert_reserved)
3302 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3303 mutex_unlock(&head->mutex);
3304 btrfs_put_delayed_ref_head(head);
3307 spin_unlock(&head->lock);
3310 spin_unlock(&delayed_refs->lock);
3314 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3315 struct btrfs_root *root,
3316 struct extent_buffer *buf,
3317 u64 parent, int last_ref)
3319 struct btrfs_fs_info *fs_info = root->fs_info;
3320 struct btrfs_ref generic_ref = { 0 };
3324 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3325 buf->start, buf->len, parent);
3326 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3327 root->root_key.objectid);
3329 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3330 int old_ref_mod, new_ref_mod;
3332 btrfs_ref_tree_mod(fs_info, &generic_ref);
3333 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL,
3334 &old_ref_mod, &new_ref_mod);
3335 BUG_ON(ret); /* -ENOMEM */
3336 pin = old_ref_mod >= 0 && new_ref_mod < 0;
3339 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3340 struct btrfs_block_group *cache;
3342 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
3343 ret = check_ref_cleanup(trans, buf->start);
3349 cache = btrfs_lookup_block_group(fs_info, buf->start);
3351 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3352 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3353 btrfs_put_block_group(cache);
3357 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3359 btrfs_add_free_space(cache, buf->start, buf->len);
3360 btrfs_free_reserved_bytes(cache, buf->len, 0);
3361 btrfs_put_block_group(cache);
3362 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3366 add_pinned_bytes(fs_info, &generic_ref);
3370 * Deleting the buffer, clear the corrupt flag since it doesn't
3373 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3377 /* Can return -ENOMEM */
3378 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3380 struct btrfs_fs_info *fs_info = trans->fs_info;
3381 int old_ref_mod, new_ref_mod;
3384 if (btrfs_is_testing(fs_info))
3388 * tree log blocks never actually go into the extent allocation
3389 * tree, just update pinning info and exit early.
3391 if ((ref->type == BTRFS_REF_METADATA &&
3392 ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3393 (ref->type == BTRFS_REF_DATA &&
3394 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) {
3395 /* unlocks the pinned mutex */
3396 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3397 old_ref_mod = new_ref_mod = 0;
3399 } else if (ref->type == BTRFS_REF_METADATA) {
3400 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL,
3401 &old_ref_mod, &new_ref_mod);
3403 ret = btrfs_add_delayed_data_ref(trans, ref, 0,
3404 &old_ref_mod, &new_ref_mod);
3407 if (!((ref->type == BTRFS_REF_METADATA &&
3408 ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3409 (ref->type == BTRFS_REF_DATA &&
3410 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)))
3411 btrfs_ref_tree_mod(fs_info, ref);
3413 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
3414 add_pinned_bytes(fs_info, ref);
3419 enum btrfs_loop_type {
3420 LOOP_CACHING_NOWAIT,
3427 btrfs_lock_block_group(struct btrfs_block_group *cache,
3431 down_read(&cache->data_rwsem);
3434 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3437 btrfs_get_block_group(cache);
3439 down_read(&cache->data_rwsem);
3442 static struct btrfs_block_group *btrfs_lock_cluster(
3443 struct btrfs_block_group *block_group,
3444 struct btrfs_free_cluster *cluster,
3446 __acquires(&cluster->refill_lock)
3448 struct btrfs_block_group *used_bg = NULL;
3450 spin_lock(&cluster->refill_lock);
3452 used_bg = cluster->block_group;
3456 if (used_bg == block_group)
3459 btrfs_get_block_group(used_bg);
3464 if (down_read_trylock(&used_bg->data_rwsem))
3467 spin_unlock(&cluster->refill_lock);
3469 /* We should only have one-level nested. */
3470 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3472 spin_lock(&cluster->refill_lock);
3473 if (used_bg == cluster->block_group)
3476 up_read(&used_bg->data_rwsem);
3477 btrfs_put_block_group(used_bg);
3482 btrfs_release_block_group(struct btrfs_block_group *cache,
3486 up_read(&cache->data_rwsem);
3487 btrfs_put_block_group(cache);
3490 enum btrfs_extent_allocation_policy {
3491 BTRFS_EXTENT_ALLOC_CLUSTERED,
3495 * Structure used internally for find_free_extent() function. Wraps needed
3498 struct find_free_extent_ctl {
3499 /* Basic allocation info */
3505 /* Where to start the search inside the bg */
3508 /* For clustered allocation */
3510 struct btrfs_free_cluster *last_ptr;
3513 bool have_caching_bg;
3514 bool orig_have_caching_bg;
3516 /* RAID index, converted from flags */
3520 * Current loop number, check find_free_extent_update_loop() for details
3525 * Whether we're refilling a cluster, if true we need to re-search
3526 * current block group but don't try to refill the cluster again.
3528 bool retry_clustered;
3531 * Whether we're updating free space cache, if true we need to re-search
3532 * current block group but don't try updating free space cache again.
3534 bool retry_unclustered;
3536 /* If current block group is cached */
3539 /* Max contiguous hole found */
3540 u64 max_extent_size;
3542 /* Total free space from free space cache, not always contiguous */
3543 u64 total_free_space;
3548 /* Hint where to start looking for an empty space */
3551 /* Allocation policy */
3552 enum btrfs_extent_allocation_policy policy;
3557 * Helper function for find_free_extent().
3559 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3560 * Return -EAGAIN to inform caller that we need to re-search this block group
3561 * Return >0 to inform caller that we find nothing
3562 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3564 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3565 struct find_free_extent_ctl *ffe_ctl,
3566 struct btrfs_block_group **cluster_bg_ret)
3568 struct btrfs_block_group *cluster_bg;
3569 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3570 u64 aligned_cluster;
3574 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3576 goto refill_cluster;
3577 if (cluster_bg != bg && (cluster_bg->ro ||
3578 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3579 goto release_cluster;
3581 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3582 ffe_ctl->num_bytes, cluster_bg->start,
3583 &ffe_ctl->max_extent_size);
3585 /* We have a block, we're done */
3586 spin_unlock(&last_ptr->refill_lock);
3587 trace_btrfs_reserve_extent_cluster(cluster_bg,
3588 ffe_ctl->search_start, ffe_ctl->num_bytes);
3589 *cluster_bg_ret = cluster_bg;
3590 ffe_ctl->found_offset = offset;
3593 WARN_ON(last_ptr->block_group != cluster_bg);
3597 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3598 * lets just skip it and let the allocator find whatever block it can
3599 * find. If we reach this point, we will have tried the cluster
3600 * allocator plenty of times and not have found anything, so we are
3601 * likely way too fragmented for the clustering stuff to find anything.
3603 * However, if the cluster is taken from the current block group,
3604 * release the cluster first, so that we stand a better chance of
3605 * succeeding in the unclustered allocation.
3607 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3608 spin_unlock(&last_ptr->refill_lock);
3609 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3613 /* This cluster didn't work out, free it and start over */
3614 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3616 if (cluster_bg != bg)
3617 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3620 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3621 spin_unlock(&last_ptr->refill_lock);
3625 aligned_cluster = max_t(u64,
3626 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3627 bg->full_stripe_len);
3628 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3629 ffe_ctl->num_bytes, aligned_cluster);
3631 /* Now pull our allocation out of this cluster */
3632 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3633 ffe_ctl->num_bytes, ffe_ctl->search_start,
3634 &ffe_ctl->max_extent_size);
3636 /* We found one, proceed */
3637 spin_unlock(&last_ptr->refill_lock);
3638 trace_btrfs_reserve_extent_cluster(bg,
3639 ffe_ctl->search_start,
3640 ffe_ctl->num_bytes);
3641 ffe_ctl->found_offset = offset;
3644 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3645 !ffe_ctl->retry_clustered) {
3646 spin_unlock(&last_ptr->refill_lock);
3648 ffe_ctl->retry_clustered = true;
3649 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3650 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3654 * At this point we either didn't find a cluster or we weren't able to
3655 * allocate a block from our cluster. Free the cluster we've been
3656 * trying to use, and go to the next block group.
3658 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3659 spin_unlock(&last_ptr->refill_lock);
3664 * Return >0 to inform caller that we find nothing
3665 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3666 * Return -EAGAIN to inform caller that we need to re-search this block group
3668 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3669 struct find_free_extent_ctl *ffe_ctl)
3671 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3675 * We are doing an unclustered allocation, set the fragmented flag so
3676 * we don't bother trying to setup a cluster again until we get more
3679 if (unlikely(last_ptr)) {
3680 spin_lock(&last_ptr->lock);
3681 last_ptr->fragmented = 1;
3682 spin_unlock(&last_ptr->lock);
3684 if (ffe_ctl->cached) {
3685 struct btrfs_free_space_ctl *free_space_ctl;
3687 free_space_ctl = bg->free_space_ctl;
3688 spin_lock(&free_space_ctl->tree_lock);
3689 if (free_space_ctl->free_space <
3690 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3691 ffe_ctl->empty_size) {
3692 ffe_ctl->total_free_space = max_t(u64,
3693 ffe_ctl->total_free_space,
3694 free_space_ctl->free_space);
3695 spin_unlock(&free_space_ctl->tree_lock);
3698 spin_unlock(&free_space_ctl->tree_lock);
3701 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3702 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3703 &ffe_ctl->max_extent_size);
3706 * If we didn't find a chunk, and we haven't failed on this block group
3707 * before, and this block group is in the middle of caching and we are
3708 * ok with waiting, then go ahead and wait for progress to be made, and
3709 * set @retry_unclustered to true.
3711 * If @retry_unclustered is true then we've already waited on this
3712 * block group once and should move on to the next block group.
3714 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3715 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
3716 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3717 ffe_ctl->empty_size);
3718 ffe_ctl->retry_unclustered = true;
3720 } else if (!offset) {
3723 ffe_ctl->found_offset = offset;
3727 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3728 struct find_free_extent_ctl *ffe_ctl,
3729 struct btrfs_block_group **bg_ret)
3733 /* We want to try and use the cluster allocator, so lets look there */
3734 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3735 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3736 if (ret >= 0 || ret == -EAGAIN)
3738 /* ret == -ENOENT case falls through */
3741 return find_free_extent_unclustered(block_group, ffe_ctl);
3744 static int do_allocation(struct btrfs_block_group *block_group,
3745 struct find_free_extent_ctl *ffe_ctl,
3746 struct btrfs_block_group **bg_ret)
3748 switch (ffe_ctl->policy) {
3749 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3750 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3756 static void release_block_group(struct btrfs_block_group *block_group,
3757 struct find_free_extent_ctl *ffe_ctl,
3760 switch (ffe_ctl->policy) {
3761 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3762 ffe_ctl->retry_clustered = false;
3763 ffe_ctl->retry_unclustered = false;
3769 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
3771 btrfs_release_block_group(block_group, delalloc);
3774 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
3775 struct btrfs_key *ins)
3777 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3779 if (!ffe_ctl->use_cluster && last_ptr) {
3780 spin_lock(&last_ptr->lock);
3781 last_ptr->window_start = ins->objectid;
3782 spin_unlock(&last_ptr->lock);
3786 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
3787 struct btrfs_key *ins)
3789 switch (ffe_ctl->policy) {
3790 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3791 found_extent_clustered(ffe_ctl, ins);
3798 static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl)
3800 switch (ffe_ctl->policy) {
3801 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3803 * If we can't allocate a new chunk we've already looped through
3804 * at least once, move on to the NO_EMPTY_SIZE case.
3806 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
3814 * Return >0 means caller needs to re-search for free extent
3815 * Return 0 means we have the needed free extent.
3816 * Return <0 means we failed to locate any free extent.
3818 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
3819 struct btrfs_key *ins,
3820 struct find_free_extent_ctl *ffe_ctl,
3823 struct btrfs_root *root = fs_info->extent_root;
3826 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
3827 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
3828 ffe_ctl->orig_have_caching_bg = true;
3830 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
3831 ffe_ctl->have_caching_bg)
3834 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
3837 if (ins->objectid) {
3838 found_extent(ffe_ctl, ins);
3843 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
3844 * caching kthreads as we move along
3845 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3846 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3847 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3850 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
3852 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
3854 * We want to skip the LOOP_CACHING_WAIT step if we
3855 * don't have any uncached bgs and we've already done a
3856 * full search through.
3858 if (ffe_ctl->orig_have_caching_bg || !full_search)
3859 ffe_ctl->loop = LOOP_CACHING_WAIT;
3861 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
3866 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
3867 struct btrfs_trans_handle *trans;
3870 trans = current->journal_info;
3874 trans = btrfs_join_transaction(root);
3876 if (IS_ERR(trans)) {
3877 ret = PTR_ERR(trans);
3881 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
3884 /* Do not bail out on ENOSPC since we can do more. */
3886 ret = chunk_allocation_failed(ffe_ctl);
3888 btrfs_abort_transaction(trans, ret);
3892 btrfs_end_transaction(trans);
3897 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
3898 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
3902 * Don't loop again if we already have no empty_size and
3905 if (ffe_ctl->empty_size == 0 &&
3906 ffe_ctl->empty_cluster == 0)
3908 ffe_ctl->empty_size = 0;
3909 ffe_ctl->empty_cluster = 0;
3916 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
3917 struct find_free_extent_ctl *ffe_ctl,
3918 struct btrfs_space_info *space_info,
3919 struct btrfs_key *ins)
3922 * If our free space is heavily fragmented we may not be able to make
3923 * big contiguous allocations, so instead of doing the expensive search
3924 * for free space, simply return ENOSPC with our max_extent_size so we
3925 * can go ahead and search for a more manageable chunk.
3927 * If our max_extent_size is large enough for our allocation simply
3928 * disable clustering since we will likely not be able to find enough
3929 * space to create a cluster and induce latency trying.
3931 if (space_info->max_extent_size) {
3932 spin_lock(&space_info->lock);
3933 if (space_info->max_extent_size &&
3934 ffe_ctl->num_bytes > space_info->max_extent_size) {
3935 ins->offset = space_info->max_extent_size;
3936 spin_unlock(&space_info->lock);
3938 } else if (space_info->max_extent_size) {
3939 ffe_ctl->use_cluster = false;
3941 spin_unlock(&space_info->lock);
3944 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
3945 &ffe_ctl->empty_cluster);
3946 if (ffe_ctl->last_ptr) {
3947 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3949 spin_lock(&last_ptr->lock);
3950 if (last_ptr->block_group)
3951 ffe_ctl->hint_byte = last_ptr->window_start;
3952 if (last_ptr->fragmented) {
3954 * We still set window_start so we can keep track of the
3955 * last place we found an allocation to try and save
3958 ffe_ctl->hint_byte = last_ptr->window_start;
3959 ffe_ctl->use_cluster = false;
3961 spin_unlock(&last_ptr->lock);
3967 static int prepare_allocation(struct btrfs_fs_info *fs_info,
3968 struct find_free_extent_ctl *ffe_ctl,
3969 struct btrfs_space_info *space_info,
3970 struct btrfs_key *ins)
3972 switch (ffe_ctl->policy) {
3973 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3974 return prepare_allocation_clustered(fs_info, ffe_ctl,
3982 * walks the btree of allocated extents and find a hole of a given size.
3983 * The key ins is changed to record the hole:
3984 * ins->objectid == start position
3985 * ins->flags = BTRFS_EXTENT_ITEM_KEY
3986 * ins->offset == the size of the hole.
3987 * Any available blocks before search_start are skipped.
3989 * If there is no suitable free space, we will record the max size of
3990 * the free space extent currently.
3992 * The overall logic and call chain:
3994 * find_free_extent()
3995 * |- Iterate through all block groups
3996 * | |- Get a valid block group
3997 * | |- Try to do clustered allocation in that block group
3998 * | |- Try to do unclustered allocation in that block group
3999 * | |- Check if the result is valid
4000 * | | |- If valid, then exit
4001 * | |- Jump to next block group
4003 * |- Push harder to find free extents
4004 * |- If not found, re-iterate all block groups
4006 static noinline int find_free_extent(struct btrfs_root *root,
4007 u64 ram_bytes, u64 num_bytes, u64 empty_size,
4008 u64 hint_byte_orig, struct btrfs_key *ins,
4009 u64 flags, int delalloc)
4011 struct btrfs_fs_info *fs_info = root->fs_info;
4013 int cache_block_group_error = 0;
4014 struct btrfs_block_group *block_group = NULL;
4015 struct find_free_extent_ctl ffe_ctl = {0};
4016 struct btrfs_space_info *space_info;
4017 bool full_search = false;
4019 WARN_ON(num_bytes < fs_info->sectorsize);
4021 ffe_ctl.num_bytes = num_bytes;
4022 ffe_ctl.empty_size = empty_size;
4023 ffe_ctl.flags = flags;
4024 ffe_ctl.search_start = 0;
4025 ffe_ctl.delalloc = delalloc;
4026 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
4027 ffe_ctl.have_caching_bg = false;
4028 ffe_ctl.orig_have_caching_bg = false;
4029 ffe_ctl.found_offset = 0;
4030 ffe_ctl.hint_byte = hint_byte_orig;
4031 ffe_ctl.policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4033 /* For clustered allocation */
4034 ffe_ctl.retry_clustered = false;
4035 ffe_ctl.retry_unclustered = false;
4036 ffe_ctl.last_ptr = NULL;
4037 ffe_ctl.use_cluster = true;
4039 ins->type = BTRFS_EXTENT_ITEM_KEY;
4043 trace_find_free_extent(root, num_bytes, empty_size, flags);
4045 space_info = btrfs_find_space_info(fs_info, flags);
4047 btrfs_err(fs_info, "No space info for %llu", flags);
4051 ret = prepare_allocation(fs_info, &ffe_ctl, space_info, ins);
4055 ffe_ctl.search_start = max(ffe_ctl.search_start,
4056 first_logical_byte(fs_info, 0));
4057 ffe_ctl.search_start = max(ffe_ctl.search_start, ffe_ctl.hint_byte);
4058 if (ffe_ctl.search_start == ffe_ctl.hint_byte) {
4059 block_group = btrfs_lookup_block_group(fs_info,
4060 ffe_ctl.search_start);
4062 * we don't want to use the block group if it doesn't match our
4063 * allocation bits, or if its not cached.
4065 * However if we are re-searching with an ideal block group
4066 * picked out then we don't care that the block group is cached.
4068 if (block_group && block_group_bits(block_group, flags) &&
4069 block_group->cached != BTRFS_CACHE_NO) {
4070 down_read(&space_info->groups_sem);
4071 if (list_empty(&block_group->list) ||
4074 * someone is removing this block group,
4075 * we can't jump into the have_block_group
4076 * target because our list pointers are not
4079 btrfs_put_block_group(block_group);
4080 up_read(&space_info->groups_sem);
4082 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
4083 block_group->flags);
4084 btrfs_lock_block_group(block_group, delalloc);
4085 goto have_block_group;
4087 } else if (block_group) {
4088 btrfs_put_block_group(block_group);
4092 ffe_ctl.have_caching_bg = false;
4093 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
4096 down_read(&space_info->groups_sem);
4097 list_for_each_entry(block_group,
4098 &space_info->block_groups[ffe_ctl.index], list) {
4099 struct btrfs_block_group *bg_ret;
4101 /* If the block group is read-only, we can skip it entirely. */
4102 if (unlikely(block_group->ro))
4105 btrfs_grab_block_group(block_group, delalloc);
4106 ffe_ctl.search_start = block_group->start;
4109 * this can happen if we end up cycling through all the
4110 * raid types, but we want to make sure we only allocate
4111 * for the proper type.
4113 if (!block_group_bits(block_group, flags)) {
4114 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4115 BTRFS_BLOCK_GROUP_RAID1_MASK |
4116 BTRFS_BLOCK_GROUP_RAID56_MASK |
4117 BTRFS_BLOCK_GROUP_RAID10;
4120 * if they asked for extra copies and this block group
4121 * doesn't provide them, bail. This does allow us to
4122 * fill raid0 from raid1.
4124 if ((flags & extra) && !(block_group->flags & extra))
4128 * This block group has different flags than we want.
4129 * It's possible that we have MIXED_GROUP flag but no
4130 * block group is mixed. Just skip such block group.
4132 btrfs_release_block_group(block_group, delalloc);
4137 ffe_ctl.cached = btrfs_block_group_done(block_group);
4138 if (unlikely(!ffe_ctl.cached)) {
4139 ffe_ctl.have_caching_bg = true;
4140 ret = btrfs_cache_block_group(block_group, 0);
4143 * If we get ENOMEM here or something else we want to
4144 * try other block groups, because it may not be fatal.
4145 * However if we can't find anything else we need to
4146 * save our return here so that we return the actual
4147 * error that caused problems, not ENOSPC.
4150 if (!cache_block_group_error)
4151 cache_block_group_error = ret;
4158 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
4162 ret = do_allocation(block_group, &ffe_ctl, &bg_ret);
4164 if (bg_ret && bg_ret != block_group) {
4165 btrfs_release_block_group(block_group, delalloc);
4166 block_group = bg_ret;
4168 } else if (ret == -EAGAIN) {
4169 goto have_block_group;
4170 } else if (ret > 0) {
4175 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
4176 fs_info->stripesize);
4178 /* move on to the next group */
4179 if (ffe_ctl.search_start + num_bytes >
4180 block_group->start + block_group->length) {
4181 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4186 if (ffe_ctl.found_offset < ffe_ctl.search_start)
4187 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4188 ffe_ctl.search_start - ffe_ctl.found_offset);
4190 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
4191 num_bytes, delalloc);
4192 if (ret == -EAGAIN) {
4193 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4197 btrfs_inc_block_group_reservations(block_group);
4199 /* we are all good, lets return */
4200 ins->objectid = ffe_ctl.search_start;
4201 ins->offset = num_bytes;
4203 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
4205 btrfs_release_block_group(block_group, delalloc);
4208 release_block_group(block_group, &ffe_ctl, delalloc);
4211 up_read(&space_info->groups_sem);
4213 ret = find_free_extent_update_loop(fs_info, ins, &ffe_ctl, full_search);
4217 if (ret == -ENOSPC && !cache_block_group_error) {
4219 * Use ffe_ctl->total_free_space as fallback if we can't find
4220 * any contiguous hole.
4222 if (!ffe_ctl.max_extent_size)
4223 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
4224 spin_lock(&space_info->lock);
4225 space_info->max_extent_size = ffe_ctl.max_extent_size;
4226 spin_unlock(&space_info->lock);
4227 ins->offset = ffe_ctl.max_extent_size;
4228 } else if (ret == -ENOSPC) {
4229 ret = cache_block_group_error;
4235 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4236 * hole that is at least as big as @num_bytes.
4238 * @root - The root that will contain this extent
4240 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4241 * is used for accounting purposes. This value differs
4242 * from @num_bytes only in the case of compressed extents.
4244 * @num_bytes - Number of bytes to allocate on-disk.
4246 * @min_alloc_size - Indicates the minimum amount of space that the
4247 * allocator should try to satisfy. In some cases
4248 * @num_bytes may be larger than what is required and if
4249 * the filesystem is fragmented then allocation fails.
4250 * However, the presence of @min_alloc_size gives a
4251 * chance to try and satisfy the smaller allocation.
4253 * @empty_size - A hint that you plan on doing more COW. This is the
4254 * size in bytes the allocator should try to find free
4255 * next to the block it returns. This is just a hint and
4256 * may be ignored by the allocator.
4258 * @hint_byte - Hint to the allocator to start searching above the byte
4259 * address passed. It might be ignored.
4261 * @ins - This key is modified to record the found hole. It will
4262 * have the following values:
4263 * ins->objectid == start position
4264 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4265 * ins->offset == the size of the hole.
4267 * @is_data - Boolean flag indicating whether an extent is
4268 * allocated for data (true) or metadata (false)
4270 * @delalloc - Boolean flag indicating whether this allocation is for
4271 * delalloc or not. If 'true' data_rwsem of block groups
4272 * is going to be acquired.
4275 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4276 * case -ENOSPC is returned then @ins->offset will contain the size of the
4277 * largest available hole the allocator managed to find.
4279 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4280 u64 num_bytes, u64 min_alloc_size,
4281 u64 empty_size, u64 hint_byte,
4282 struct btrfs_key *ins, int is_data, int delalloc)
4284 struct btrfs_fs_info *fs_info = root->fs_info;
4285 bool final_tried = num_bytes == min_alloc_size;
4289 flags = get_alloc_profile_by_root(root, is_data);
4291 WARN_ON(num_bytes < fs_info->sectorsize);
4292 ret = find_free_extent(root, ram_bytes, num_bytes, empty_size,
4293 hint_byte, ins, flags, delalloc);
4294 if (!ret && !is_data) {
4295 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4296 } else if (ret == -ENOSPC) {
4297 if (!final_tried && ins->offset) {
4298 num_bytes = min(num_bytes >> 1, ins->offset);
4299 num_bytes = round_down(num_bytes,
4300 fs_info->sectorsize);
4301 num_bytes = max(num_bytes, min_alloc_size);
4302 ram_bytes = num_bytes;
4303 if (num_bytes == min_alloc_size)
4306 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4307 struct btrfs_space_info *sinfo;
4309 sinfo = btrfs_find_space_info(fs_info, flags);
4311 "allocation failed flags %llu, wanted %llu",
4314 btrfs_dump_space_info(fs_info, sinfo,
4322 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4323 u64 start, u64 len, int delalloc)
4325 struct btrfs_block_group *cache;
4327 cache = btrfs_lookup_block_group(fs_info, start);
4329 btrfs_err(fs_info, "Unable to find block group for %llu",
4334 btrfs_add_free_space(cache, start, len);
4335 btrfs_free_reserved_bytes(cache, len, delalloc);
4336 trace_btrfs_reserved_extent_free(fs_info, start, len);
4338 btrfs_put_block_group(cache);
4342 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start,
4345 struct btrfs_block_group *cache;
4348 cache = btrfs_lookup_block_group(trans->fs_info, start);
4350 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4355 ret = pin_down_extent(trans, cache, start, len, 1);
4356 btrfs_put_block_group(cache);
4360 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4361 u64 parent, u64 root_objectid,
4362 u64 flags, u64 owner, u64 offset,
4363 struct btrfs_key *ins, int ref_mod)
4365 struct btrfs_fs_info *fs_info = trans->fs_info;
4367 struct btrfs_extent_item *extent_item;
4368 struct btrfs_extent_inline_ref *iref;
4369 struct btrfs_path *path;
4370 struct extent_buffer *leaf;
4375 type = BTRFS_SHARED_DATA_REF_KEY;
4377 type = BTRFS_EXTENT_DATA_REF_KEY;
4379 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4381 path = btrfs_alloc_path();
4385 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4388 btrfs_free_path(path);
4392 leaf = path->nodes[0];
4393 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4394 struct btrfs_extent_item);
4395 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4396 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4397 btrfs_set_extent_flags(leaf, extent_item,
4398 flags | BTRFS_EXTENT_FLAG_DATA);
4400 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4401 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4403 struct btrfs_shared_data_ref *ref;
4404 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4405 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4406 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4408 struct btrfs_extent_data_ref *ref;
4409 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4410 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4411 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4412 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4413 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4416 btrfs_mark_buffer_dirty(path->nodes[0]);
4417 btrfs_free_path(path);
4419 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
4423 ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1);
4424 if (ret) { /* -ENOENT, logic error */
4425 btrfs_err(fs_info, "update block group failed for %llu %llu",
4426 ins->objectid, ins->offset);
4429 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
4433 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4434 struct btrfs_delayed_ref_node *node,
4435 struct btrfs_delayed_extent_op *extent_op)
4437 struct btrfs_fs_info *fs_info = trans->fs_info;
4439 struct btrfs_extent_item *extent_item;
4440 struct btrfs_key extent_key;
4441 struct btrfs_tree_block_info *block_info;
4442 struct btrfs_extent_inline_ref *iref;
4443 struct btrfs_path *path;
4444 struct extent_buffer *leaf;
4445 struct btrfs_delayed_tree_ref *ref;
4446 u32 size = sizeof(*extent_item) + sizeof(*iref);
4448 u64 flags = extent_op->flags_to_set;
4449 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4451 ref = btrfs_delayed_node_to_tree_ref(node);
4453 extent_key.objectid = node->bytenr;
4454 if (skinny_metadata) {
4455 extent_key.offset = ref->level;
4456 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4457 num_bytes = fs_info->nodesize;
4459 extent_key.offset = node->num_bytes;
4460 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4461 size += sizeof(*block_info);
4462 num_bytes = node->num_bytes;
4465 path = btrfs_alloc_path();
4469 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4472 btrfs_free_path(path);
4476 leaf = path->nodes[0];
4477 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4478 struct btrfs_extent_item);
4479 btrfs_set_extent_refs(leaf, extent_item, 1);
4480 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4481 btrfs_set_extent_flags(leaf, extent_item,
4482 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4484 if (skinny_metadata) {
4485 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4487 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4488 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4489 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4490 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4493 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4494 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4495 btrfs_set_extent_inline_ref_type(leaf, iref,
4496 BTRFS_SHARED_BLOCK_REF_KEY);
4497 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4499 btrfs_set_extent_inline_ref_type(leaf, iref,
4500 BTRFS_TREE_BLOCK_REF_KEY);
4501 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4504 btrfs_mark_buffer_dirty(leaf);
4505 btrfs_free_path(path);
4507 ret = remove_from_free_space_tree(trans, extent_key.objectid,
4512 ret = btrfs_update_block_group(trans, extent_key.objectid,
4513 fs_info->nodesize, 1);
4514 if (ret) { /* -ENOENT, logic error */
4515 btrfs_err(fs_info, "update block group failed for %llu %llu",
4516 extent_key.objectid, extent_key.offset);
4520 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
4525 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4526 struct btrfs_root *root, u64 owner,
4527 u64 offset, u64 ram_bytes,
4528 struct btrfs_key *ins)
4530 struct btrfs_ref generic_ref = { 0 };
4533 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4535 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4536 ins->objectid, ins->offset, 0);
4537 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset);
4538 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4539 ret = btrfs_add_delayed_data_ref(trans, &generic_ref,
4540 ram_bytes, NULL, NULL);
4545 * this is used by the tree logging recovery code. It records that
4546 * an extent has been allocated and makes sure to clear the free
4547 * space cache bits as well
4549 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4550 u64 root_objectid, u64 owner, u64 offset,
4551 struct btrfs_key *ins)
4553 struct btrfs_fs_info *fs_info = trans->fs_info;
4555 struct btrfs_block_group *block_group;
4556 struct btrfs_space_info *space_info;
4559 * Mixed block groups will exclude before processing the log so we only
4560 * need to do the exclude dance if this fs isn't mixed.
4562 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4563 ret = __exclude_logged_extent(fs_info, ins->objectid,
4569 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4573 space_info = block_group->space_info;
4574 spin_lock(&space_info->lock);
4575 spin_lock(&block_group->lock);
4576 space_info->bytes_reserved += ins->offset;
4577 block_group->reserved += ins->offset;
4578 spin_unlock(&block_group->lock);
4579 spin_unlock(&space_info->lock);
4581 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4584 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4585 btrfs_put_block_group(block_group);
4589 static struct extent_buffer *
4590 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4591 u64 bytenr, int level, u64 owner,
4592 enum btrfs_lock_nesting nest)
4594 struct btrfs_fs_info *fs_info = root->fs_info;
4595 struct extent_buffer *buf;
4597 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
4602 * Extra safety check in case the extent tree is corrupted and extent
4603 * allocator chooses to use a tree block which is already used and
4606 if (buf->lock_owner == current->pid) {
4607 btrfs_err_rl(fs_info,
4608 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4609 buf->start, btrfs_header_owner(buf), current->pid);
4610 free_extent_buffer(buf);
4611 return ERR_PTR(-EUCLEAN);
4615 * This needs to stay, because we could allocate a freed block from an
4616 * old tree into a new tree, so we need to make sure this new block is
4617 * set to the appropriate level and owner.
4619 btrfs_set_buffer_lockdep_class(owner, buf, level);
4620 __btrfs_tree_lock(buf, nest);
4621 btrfs_clean_tree_block(buf);
4622 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4624 set_extent_buffer_uptodate(buf);
4626 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4627 btrfs_set_header_level(buf, level);
4628 btrfs_set_header_bytenr(buf, buf->start);
4629 btrfs_set_header_generation(buf, trans->transid);
4630 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4631 btrfs_set_header_owner(buf, owner);
4632 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4633 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4634 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4635 buf->log_index = root->log_transid % 2;
4637 * we allow two log transactions at a time, use different
4638 * EXTENT bit to differentiate dirty pages.
4640 if (buf->log_index == 0)
4641 set_extent_dirty(&root->dirty_log_pages, buf->start,
4642 buf->start + buf->len - 1, GFP_NOFS);
4644 set_extent_new(&root->dirty_log_pages, buf->start,
4645 buf->start + buf->len - 1);
4647 buf->log_index = -1;
4648 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4649 buf->start + buf->len - 1, GFP_NOFS);
4651 trans->dirty = true;
4652 /* this returns a buffer locked for blocking */
4657 * finds a free extent and does all the dirty work required for allocation
4658 * returns the tree buffer or an ERR_PTR on error.
4660 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4661 struct btrfs_root *root,
4662 u64 parent, u64 root_objectid,
4663 const struct btrfs_disk_key *key,
4664 int level, u64 hint,
4666 enum btrfs_lock_nesting nest)
4668 struct btrfs_fs_info *fs_info = root->fs_info;
4669 struct btrfs_key ins;
4670 struct btrfs_block_rsv *block_rsv;
4671 struct extent_buffer *buf;
4672 struct btrfs_delayed_extent_op *extent_op;
4673 struct btrfs_ref generic_ref = { 0 };
4676 u32 blocksize = fs_info->nodesize;
4677 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4679 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4680 if (btrfs_is_testing(fs_info)) {
4681 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4682 level, root_objectid, nest);
4684 root->alloc_bytenr += blocksize;
4689 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4690 if (IS_ERR(block_rsv))
4691 return ERR_CAST(block_rsv);
4693 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4694 empty_size, hint, &ins, 0, 0);
4698 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4699 root_objectid, nest);
4702 goto out_free_reserved;
4705 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4707 parent = ins.objectid;
4708 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4712 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4713 extent_op = btrfs_alloc_delayed_extent_op();
4719 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4721 memset(&extent_op->key, 0, sizeof(extent_op->key));
4722 extent_op->flags_to_set = flags;
4723 extent_op->update_key = skinny_metadata ? false : true;
4724 extent_op->update_flags = true;
4725 extent_op->is_data = false;
4726 extent_op->level = level;
4728 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4729 ins.objectid, ins.offset, parent);
4730 generic_ref.real_root = root->root_key.objectid;
4731 btrfs_init_tree_ref(&generic_ref, level, root_objectid);
4732 btrfs_ref_tree_mod(fs_info, &generic_ref);
4733 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref,
4734 extent_op, NULL, NULL);
4736 goto out_free_delayed;
4741 btrfs_free_delayed_extent_op(extent_op);
4743 free_extent_buffer(buf);
4745 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
4747 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
4748 return ERR_PTR(ret);
4751 struct walk_control {
4752 u64 refs[BTRFS_MAX_LEVEL];
4753 u64 flags[BTRFS_MAX_LEVEL];
4754 struct btrfs_key update_progress;
4755 struct btrfs_key drop_progress;
4767 #define DROP_REFERENCE 1
4768 #define UPDATE_BACKREF 2
4770 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4771 struct btrfs_root *root,
4772 struct walk_control *wc,
4773 struct btrfs_path *path)
4775 struct btrfs_fs_info *fs_info = root->fs_info;
4781 struct btrfs_key key;
4782 struct extent_buffer *eb;
4787 if (path->slots[wc->level] < wc->reada_slot) {
4788 wc->reada_count = wc->reada_count * 2 / 3;
4789 wc->reada_count = max(wc->reada_count, 2);
4791 wc->reada_count = wc->reada_count * 3 / 2;
4792 wc->reada_count = min_t(int, wc->reada_count,
4793 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
4796 eb = path->nodes[wc->level];
4797 nritems = btrfs_header_nritems(eb);
4799 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4800 if (nread >= wc->reada_count)
4804 bytenr = btrfs_node_blockptr(eb, slot);
4805 generation = btrfs_node_ptr_generation(eb, slot);
4807 if (slot == path->slots[wc->level])
4810 if (wc->stage == UPDATE_BACKREF &&
4811 generation <= root->root_key.offset)
4814 /* We don't lock the tree block, it's OK to be racy here */
4815 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
4816 wc->level - 1, 1, &refs,
4818 /* We don't care about errors in readahead. */
4823 if (wc->stage == DROP_REFERENCE) {
4827 if (wc->level == 1 &&
4828 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4830 if (!wc->update_ref ||
4831 generation <= root->root_key.offset)
4833 btrfs_node_key_to_cpu(eb, &key, slot);
4834 ret = btrfs_comp_cpu_keys(&key,
4835 &wc->update_progress);
4839 if (wc->level == 1 &&
4840 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4844 btrfs_readahead_node_child(eb, slot);
4847 wc->reada_slot = slot;
4851 * helper to process tree block while walking down the tree.
4853 * when wc->stage == UPDATE_BACKREF, this function updates
4854 * back refs for pointers in the block.
4856 * NOTE: return value 1 means we should stop walking down.
4858 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4859 struct btrfs_root *root,
4860 struct btrfs_path *path,
4861 struct walk_control *wc, int lookup_info)
4863 struct btrfs_fs_info *fs_info = root->fs_info;
4864 int level = wc->level;
4865 struct extent_buffer *eb = path->nodes[level];
4866 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4869 if (wc->stage == UPDATE_BACKREF &&
4870 btrfs_header_owner(eb) != root->root_key.objectid)
4874 * when reference count of tree block is 1, it won't increase
4875 * again. once full backref flag is set, we never clear it.
4878 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4879 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
4880 BUG_ON(!path->locks[level]);
4881 ret = btrfs_lookup_extent_info(trans, fs_info,
4882 eb->start, level, 1,
4885 BUG_ON(ret == -ENOMEM);
4888 BUG_ON(wc->refs[level] == 0);
4891 if (wc->stage == DROP_REFERENCE) {
4892 if (wc->refs[level] > 1)
4895 if (path->locks[level] && !wc->keep_locks) {
4896 btrfs_tree_unlock_rw(eb, path->locks[level]);
4897 path->locks[level] = 0;
4902 /* wc->stage == UPDATE_BACKREF */
4903 if (!(wc->flags[level] & flag)) {
4904 BUG_ON(!path->locks[level]);
4905 ret = btrfs_inc_ref(trans, root, eb, 1);
4906 BUG_ON(ret); /* -ENOMEM */
4907 ret = btrfs_dec_ref(trans, root, eb, 0);
4908 BUG_ON(ret); /* -ENOMEM */
4909 ret = btrfs_set_disk_extent_flags(trans, eb, flag,
4910 btrfs_header_level(eb), 0);
4911 BUG_ON(ret); /* -ENOMEM */
4912 wc->flags[level] |= flag;
4916 * the block is shared by multiple trees, so it's not good to
4917 * keep the tree lock
4919 if (path->locks[level] && level > 0) {
4920 btrfs_tree_unlock_rw(eb, path->locks[level]);
4921 path->locks[level] = 0;
4927 * This is used to verify a ref exists for this root to deal with a bug where we
4928 * would have a drop_progress key that hadn't been updated properly.
4930 static int check_ref_exists(struct btrfs_trans_handle *trans,
4931 struct btrfs_root *root, u64 bytenr, u64 parent,
4934 struct btrfs_path *path;
4935 struct btrfs_extent_inline_ref *iref;
4938 path = btrfs_alloc_path();
4942 ret = lookup_extent_backref(trans, path, &iref, bytenr,
4943 root->fs_info->nodesize, parent,
4944 root->root_key.objectid, level, 0);
4945 btrfs_free_path(path);
4954 * helper to process tree block pointer.
4956 * when wc->stage == DROP_REFERENCE, this function checks
4957 * reference count of the block pointed to. if the block
4958 * is shared and we need update back refs for the subtree
4959 * rooted at the block, this function changes wc->stage to
4960 * UPDATE_BACKREF. if the block is shared and there is no
4961 * need to update back, this function drops the reference
4964 * NOTE: return value 1 means we should stop walking down.
4966 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4967 struct btrfs_root *root,
4968 struct btrfs_path *path,
4969 struct walk_control *wc, int *lookup_info)
4971 struct btrfs_fs_info *fs_info = root->fs_info;
4975 struct btrfs_key key;
4976 struct btrfs_key first_key;
4977 struct btrfs_ref ref = { 0 };
4978 struct extent_buffer *next;
4979 int level = wc->level;
4982 bool need_account = false;
4984 generation = btrfs_node_ptr_generation(path->nodes[level],
4985 path->slots[level]);
4987 * if the lower level block was created before the snapshot
4988 * was created, we know there is no need to update back refs
4991 if (wc->stage == UPDATE_BACKREF &&
4992 generation <= root->root_key.offset) {
4997 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
4998 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
4999 path->slots[level]);
5001 next = find_extent_buffer(fs_info, bytenr);
5003 next = btrfs_find_create_tree_block(fs_info, bytenr,
5004 root->root_key.objectid, level - 1);
5006 return PTR_ERR(next);
5009 btrfs_tree_lock(next);
5011 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5012 &wc->refs[level - 1],
5013 &wc->flags[level - 1]);
5017 if (unlikely(wc->refs[level - 1] == 0)) {
5018 btrfs_err(fs_info, "Missing references.");
5024 if (wc->stage == DROP_REFERENCE) {
5025 if (wc->refs[level - 1] > 1) {
5026 need_account = true;
5028 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5031 if (!wc->update_ref ||
5032 generation <= root->root_key.offset)
5035 btrfs_node_key_to_cpu(path->nodes[level], &key,
5036 path->slots[level]);
5037 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5041 wc->stage = UPDATE_BACKREF;
5042 wc->shared_level = level - 1;
5046 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5050 if (!btrfs_buffer_uptodate(next, generation, 0)) {
5051 btrfs_tree_unlock(next);
5052 free_extent_buffer(next);
5058 if (reada && level == 1)
5059 reada_walk_down(trans, root, wc, path);
5060 next = read_tree_block(fs_info, bytenr, root->root_key.objectid,
5061 generation, level - 1, &first_key);
5063 return PTR_ERR(next);
5064 } else if (!extent_buffer_uptodate(next)) {
5065 free_extent_buffer(next);
5068 btrfs_tree_lock(next);
5072 ASSERT(level == btrfs_header_level(next));
5073 if (level != btrfs_header_level(next)) {
5074 btrfs_err(root->fs_info, "mismatched level");
5078 path->nodes[level] = next;
5079 path->slots[level] = 0;
5080 path->locks[level] = BTRFS_WRITE_LOCK;
5086 wc->refs[level - 1] = 0;
5087 wc->flags[level - 1] = 0;
5088 if (wc->stage == DROP_REFERENCE) {
5089 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5090 parent = path->nodes[level]->start;
5092 ASSERT(root->root_key.objectid ==
5093 btrfs_header_owner(path->nodes[level]));
5094 if (root->root_key.objectid !=
5095 btrfs_header_owner(path->nodes[level])) {
5096 btrfs_err(root->fs_info,
5097 "mismatched block owner");
5105 * If we had a drop_progress we need to verify the refs are set
5106 * as expected. If we find our ref then we know that from here
5107 * on out everything should be correct, and we can clear the
5110 if (wc->restarted) {
5111 ret = check_ref_exists(trans, root, bytenr, parent,
5122 * Reloc tree doesn't contribute to qgroup numbers, and we have
5123 * already accounted them at merge time (replace_path),
5124 * thus we could skip expensive subtree trace here.
5126 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5128 ret = btrfs_qgroup_trace_subtree(trans, next,
5129 generation, level - 1);
5131 btrfs_err_rl(fs_info,
5132 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5138 * We need to update the next key in our walk control so we can
5139 * update the drop_progress key accordingly. We don't care if
5140 * find_next_key doesn't find a key because that means we're at
5141 * the end and are going to clean up now.
5143 wc->drop_level = level;
5144 find_next_key(path, level, &wc->drop_progress);
5146 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5147 fs_info->nodesize, parent);
5148 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid);
5149 ret = btrfs_free_extent(trans, &ref);
5158 btrfs_tree_unlock(next);
5159 free_extent_buffer(next);
5165 * helper to process tree block while walking up the tree.
5167 * when wc->stage == DROP_REFERENCE, this function drops
5168 * reference count on the block.
5170 * when wc->stage == UPDATE_BACKREF, this function changes
5171 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5172 * to UPDATE_BACKREF previously while processing the block.
5174 * NOTE: return value 1 means we should stop walking up.
5176 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5177 struct btrfs_root *root,
5178 struct btrfs_path *path,
5179 struct walk_control *wc)
5181 struct btrfs_fs_info *fs_info = root->fs_info;
5183 int level = wc->level;
5184 struct extent_buffer *eb = path->nodes[level];
5187 if (wc->stage == UPDATE_BACKREF) {
5188 BUG_ON(wc->shared_level < level);
5189 if (level < wc->shared_level)
5192 ret = find_next_key(path, level + 1, &wc->update_progress);
5196 wc->stage = DROP_REFERENCE;
5197 wc->shared_level = -1;
5198 path->slots[level] = 0;
5201 * check reference count again if the block isn't locked.
5202 * we should start walking down the tree again if reference
5205 if (!path->locks[level]) {
5207 btrfs_tree_lock(eb);
5208 path->locks[level] = BTRFS_WRITE_LOCK;
5210 ret = btrfs_lookup_extent_info(trans, fs_info,
5211 eb->start, level, 1,
5215 btrfs_tree_unlock_rw(eb, path->locks[level]);
5216 path->locks[level] = 0;
5219 BUG_ON(wc->refs[level] == 0);
5220 if (wc->refs[level] == 1) {
5221 btrfs_tree_unlock_rw(eb, path->locks[level]);
5222 path->locks[level] = 0;
5228 /* wc->stage == DROP_REFERENCE */
5229 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5231 if (wc->refs[level] == 1) {
5233 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5234 ret = btrfs_dec_ref(trans, root, eb, 1);
5236 ret = btrfs_dec_ref(trans, root, eb, 0);
5237 BUG_ON(ret); /* -ENOMEM */
5238 if (is_fstree(root->root_key.objectid)) {
5239 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5241 btrfs_err_rl(fs_info,
5242 "error %d accounting leaf items, quota is out of sync, rescan required",
5247 /* make block locked assertion in btrfs_clean_tree_block happy */
5248 if (!path->locks[level] &&
5249 btrfs_header_generation(eb) == trans->transid) {
5250 btrfs_tree_lock(eb);
5251 path->locks[level] = BTRFS_WRITE_LOCK;
5253 btrfs_clean_tree_block(eb);
5256 if (eb == root->node) {
5257 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5259 else if (root->root_key.objectid != btrfs_header_owner(eb))
5260 goto owner_mismatch;
5262 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5263 parent = path->nodes[level + 1]->start;
5264 else if (root->root_key.objectid !=
5265 btrfs_header_owner(path->nodes[level + 1]))
5266 goto owner_mismatch;
5269 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5271 wc->refs[level] = 0;
5272 wc->flags[level] = 0;
5276 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5277 btrfs_header_owner(eb), root->root_key.objectid);
5281 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5282 struct btrfs_root *root,
5283 struct btrfs_path *path,
5284 struct walk_control *wc)
5286 int level = wc->level;
5287 int lookup_info = 1;
5290 while (level >= 0) {
5291 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5298 if (path->slots[level] >=
5299 btrfs_header_nritems(path->nodes[level]))
5302 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5304 path->slots[level]++;
5313 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5314 struct btrfs_root *root,
5315 struct btrfs_path *path,
5316 struct walk_control *wc, int max_level)
5318 int level = wc->level;
5321 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5322 while (level < max_level && path->nodes[level]) {
5324 if (path->slots[level] + 1 <
5325 btrfs_header_nritems(path->nodes[level])) {
5326 path->slots[level]++;
5329 ret = walk_up_proc(trans, root, path, wc);
5335 if (path->locks[level]) {
5336 btrfs_tree_unlock_rw(path->nodes[level],
5337 path->locks[level]);
5338 path->locks[level] = 0;
5340 free_extent_buffer(path->nodes[level]);
5341 path->nodes[level] = NULL;
5349 * drop a subvolume tree.
5351 * this function traverses the tree freeing any blocks that only
5352 * referenced by the tree.
5354 * when a shared tree block is found. this function decreases its
5355 * reference count by one. if update_ref is true, this function
5356 * also make sure backrefs for the shared block and all lower level
5357 * blocks are properly updated.
5359 * If called with for_reloc == 0, may exit early with -EAGAIN
5361 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5363 struct btrfs_fs_info *fs_info = root->fs_info;
5364 struct btrfs_path *path;
5365 struct btrfs_trans_handle *trans;
5366 struct btrfs_root *tree_root = fs_info->tree_root;
5367 struct btrfs_root_item *root_item = &root->root_item;
5368 struct walk_control *wc;
5369 struct btrfs_key key;
5373 bool root_dropped = false;
5375 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5377 path = btrfs_alloc_path();
5383 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5385 btrfs_free_path(path);
5391 * Use join to avoid potential EINTR from transaction start. See
5392 * wait_reserve_ticket and the whole reservation callchain.
5395 trans = btrfs_join_transaction(tree_root);
5397 trans = btrfs_start_transaction(tree_root, 0);
5398 if (IS_ERR(trans)) {
5399 err = PTR_ERR(trans);
5403 err = btrfs_run_delayed_items(trans);
5408 * This will help us catch people modifying the fs tree while we're
5409 * dropping it. It is unsafe to mess with the fs tree while it's being
5410 * dropped as we unlock the root node and parent nodes as we walk down
5411 * the tree, assuming nothing will change. If something does change
5412 * then we'll have stale information and drop references to blocks we've
5415 set_bit(BTRFS_ROOT_DELETING, &root->state);
5416 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5417 level = btrfs_header_level(root->node);
5418 path->nodes[level] = btrfs_lock_root_node(root);
5419 path->slots[level] = 0;
5420 path->locks[level] = BTRFS_WRITE_LOCK;
5421 memset(&wc->update_progress, 0,
5422 sizeof(wc->update_progress));
5424 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5425 memcpy(&wc->update_progress, &key,
5426 sizeof(wc->update_progress));
5428 level = btrfs_root_drop_level(root_item);
5430 path->lowest_level = level;
5431 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5432 path->lowest_level = 0;
5440 * unlock our path, this is safe because only this
5441 * function is allowed to delete this snapshot
5443 btrfs_unlock_up_safe(path, 0);
5445 level = btrfs_header_level(root->node);
5447 btrfs_tree_lock(path->nodes[level]);
5448 path->locks[level] = BTRFS_WRITE_LOCK;
5450 ret = btrfs_lookup_extent_info(trans, fs_info,
5451 path->nodes[level]->start,
5452 level, 1, &wc->refs[level],
5458 BUG_ON(wc->refs[level] == 0);
5460 if (level == btrfs_root_drop_level(root_item))
5463 btrfs_tree_unlock(path->nodes[level]);
5464 path->locks[level] = 0;
5465 WARN_ON(wc->refs[level] != 1);
5470 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5472 wc->shared_level = -1;
5473 wc->stage = DROP_REFERENCE;
5474 wc->update_ref = update_ref;
5476 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5480 ret = walk_down_tree(trans, root, path, wc);
5486 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5493 BUG_ON(wc->stage != DROP_REFERENCE);
5497 if (wc->stage == DROP_REFERENCE) {
5498 wc->drop_level = wc->level;
5499 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5501 path->slots[wc->drop_level]);
5503 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5504 &wc->drop_progress);
5505 btrfs_set_root_drop_level(root_item, wc->drop_level);
5507 BUG_ON(wc->level == 0);
5508 if (btrfs_should_end_transaction(trans) ||
5509 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5510 ret = btrfs_update_root(trans, tree_root,
5514 btrfs_abort_transaction(trans, ret);
5519 btrfs_end_transaction_throttle(trans);
5520 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5521 btrfs_debug(fs_info,
5522 "drop snapshot early exit");
5528 * Use join to avoid potential EINTR from transaction
5529 * start. See wait_reserve_ticket and the whole
5530 * reservation callchain.
5533 trans = btrfs_join_transaction(tree_root);
5535 trans = btrfs_start_transaction(tree_root, 0);
5536 if (IS_ERR(trans)) {
5537 err = PTR_ERR(trans);
5542 btrfs_release_path(path);
5546 ret = btrfs_del_root(trans, &root->root_key);
5548 btrfs_abort_transaction(trans, ret);
5553 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5554 ret = btrfs_find_root(tree_root, &root->root_key, path,
5557 btrfs_abort_transaction(trans, ret);
5560 } else if (ret > 0) {
5561 /* if we fail to delete the orphan item this time
5562 * around, it'll get picked up the next time.
5564 * The most common failure here is just -ENOENT.
5566 btrfs_del_orphan_item(trans, tree_root,
5567 root->root_key.objectid);
5572 * This subvolume is going to be completely dropped, and won't be
5573 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5574 * commit transaction time. So free it here manually.
5576 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
5577 btrfs_qgroup_free_meta_all_pertrans(root);
5579 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
5580 btrfs_add_dropped_root(trans, root);
5582 btrfs_put_root(root);
5583 root_dropped = true;
5585 btrfs_end_transaction_throttle(trans);
5588 btrfs_free_path(path);
5591 * So if we need to stop dropping the snapshot for whatever reason we
5592 * need to make sure to add it back to the dead root list so that we
5593 * keep trying to do the work later. This also cleans up roots if we
5594 * don't have it in the radix (like when we recover after a power fail
5595 * or unmount) so we don't leak memory.
5597 if (!for_reloc && !root_dropped)
5598 btrfs_add_dead_root(root);
5603 * drop subtree rooted at tree block 'node'.
5605 * NOTE: this function will unlock and release tree block 'node'
5606 * only used by relocation code
5608 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5609 struct btrfs_root *root,
5610 struct extent_buffer *node,
5611 struct extent_buffer *parent)
5613 struct btrfs_fs_info *fs_info = root->fs_info;
5614 struct btrfs_path *path;
5615 struct walk_control *wc;
5621 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5623 path = btrfs_alloc_path();
5627 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5629 btrfs_free_path(path);
5633 btrfs_assert_tree_locked(parent);
5634 parent_level = btrfs_header_level(parent);
5635 atomic_inc(&parent->refs);
5636 path->nodes[parent_level] = parent;
5637 path->slots[parent_level] = btrfs_header_nritems(parent);
5639 btrfs_assert_tree_locked(node);
5640 level = btrfs_header_level(node);
5641 path->nodes[level] = node;
5642 path->slots[level] = 0;
5643 path->locks[level] = BTRFS_WRITE_LOCK;
5645 wc->refs[parent_level] = 1;
5646 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5648 wc->shared_level = -1;
5649 wc->stage = DROP_REFERENCE;
5652 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5655 wret = walk_down_tree(trans, root, path, wc);
5661 wret = walk_up_tree(trans, root, path, wc, parent_level);
5669 btrfs_free_path(path);
5674 * helper to account the unused space of all the readonly block group in the
5675 * space_info. takes mirrors into account.
5677 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
5679 struct btrfs_block_group *block_group;
5683 /* It's df, we don't care if it's racy */
5684 if (list_empty(&sinfo->ro_bgs))
5687 spin_lock(&sinfo->lock);
5688 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
5689 spin_lock(&block_group->lock);
5691 if (!block_group->ro) {
5692 spin_unlock(&block_group->lock);
5696 factor = btrfs_bg_type_to_factor(block_group->flags);
5697 free_bytes += (block_group->length -
5698 block_group->used) * factor;
5700 spin_unlock(&block_group->lock);
5702 spin_unlock(&sinfo->lock);
5707 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5710 return unpin_extent_range(fs_info, start, end, false);
5714 * It used to be that old block groups would be left around forever.
5715 * Iterating over them would be enough to trim unused space. Since we
5716 * now automatically remove them, we also need to iterate over unallocated
5719 * We don't want a transaction for this since the discard may take a
5720 * substantial amount of time. We don't require that a transaction be
5721 * running, but we do need to take a running transaction into account
5722 * to ensure that we're not discarding chunks that were released or
5723 * allocated in the current transaction.
5725 * Holding the chunks lock will prevent other threads from allocating
5726 * or releasing chunks, but it won't prevent a running transaction
5727 * from committing and releasing the memory that the pending chunks
5728 * list head uses. For that, we need to take a reference to the
5729 * transaction and hold the commit root sem. We only need to hold
5730 * it while performing the free space search since we have already
5731 * held back allocations.
5733 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
5735 u64 start = SZ_1M, len = 0, end = 0;
5740 /* Discard not supported = nothing to do. */
5741 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
5744 /* Not writable = nothing to do. */
5745 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5748 /* No free space = nothing to do. */
5749 if (device->total_bytes <= device->bytes_used)
5755 struct btrfs_fs_info *fs_info = device->fs_info;
5758 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
5762 find_first_clear_extent_bit(&device->alloc_state, start,
5764 CHUNK_TRIMMED | CHUNK_ALLOCATED);
5766 /* Check if there are any CHUNK_* bits left */
5767 if (start > device->total_bytes) {
5768 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
5769 btrfs_warn_in_rcu(fs_info,
5770 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
5771 start, end - start + 1,
5772 rcu_str_deref(device->name),
5773 device->total_bytes);
5774 mutex_unlock(&fs_info->chunk_mutex);
5779 /* Ensure we skip the reserved area in the first 1M */
5780 start = max_t(u64, start, SZ_1M);
5783 * If find_first_clear_extent_bit find a range that spans the
5784 * end of the device it will set end to -1, in this case it's up
5785 * to the caller to trim the value to the size of the device.
5787 end = min(end, device->total_bytes - 1);
5789 len = end - start + 1;
5791 /* We didn't find any extents */
5793 mutex_unlock(&fs_info->chunk_mutex);
5798 ret = btrfs_issue_discard(device->bdev, start, len,
5801 set_extent_bits(&device->alloc_state, start,
5804 mutex_unlock(&fs_info->chunk_mutex);
5812 if (fatal_signal_pending(current)) {
5824 * Trim the whole filesystem by:
5825 * 1) trimming the free space in each block group
5826 * 2) trimming the unallocated space on each device
5828 * This will also continue trimming even if a block group or device encounters
5829 * an error. The return value will be the last error, or 0 if nothing bad
5832 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
5834 struct btrfs_block_group *cache = NULL;
5835 struct btrfs_device *device;
5836 struct list_head *devices;
5838 u64 range_end = U64_MAX;
5849 * Check range overflow if range->len is set.
5850 * The default range->len is U64_MAX.
5852 if (range->len != U64_MAX &&
5853 check_add_overflow(range->start, range->len, &range_end))
5856 cache = btrfs_lookup_first_block_group(fs_info, range->start);
5857 for (; cache; cache = btrfs_next_block_group(cache)) {
5858 if (cache->start >= range_end) {
5859 btrfs_put_block_group(cache);
5863 start = max(range->start, cache->start);
5864 end = min(range_end, cache->start + cache->length);
5866 if (end - start >= range->minlen) {
5867 if (!btrfs_block_group_done(cache)) {
5868 ret = btrfs_cache_block_group(cache, 0);
5874 ret = btrfs_wait_block_group_cache_done(cache);
5881 ret = btrfs_trim_block_group(cache,
5887 trimmed += group_trimmed;
5898 "failed to trim %llu block group(s), last error %d",
5900 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5901 devices = &fs_info->fs_devices->devices;
5902 list_for_each_entry(device, devices, dev_list) {
5903 ret = btrfs_trim_free_extents(device, &group_trimmed);
5910 trimmed += group_trimmed;
5912 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5916 "failed to trim %llu device(s), last error %d",
5917 dev_failed, dev_ret);
5918 range->len = trimmed;