1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
20 #include "extent-tree.h"
23 #include "print-tree.h"
27 #include "free-space-cache.h"
28 #include "free-space-tree.h"
31 #include "ref-verify.h"
32 #include "space-info.h"
33 #include "block-rsv.h"
34 #include "delalloc-space.h"
36 #include "rcu-string.h"
38 #include "dev-replace.h"
40 #include "accessors.h"
41 #include "root-tree.h"
42 #include "file-item.h"
44 #include "tree-checker.h"
46 #undef SCRAMBLE_DELAYED_REFS
49 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
50 struct btrfs_delayed_ref_node *node, u64 parent,
51 u64 root_objectid, u64 owner_objectid,
52 u64 owner_offset, int refs_to_drop,
53 struct btrfs_delayed_extent_op *extra_op);
54 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
55 struct extent_buffer *leaf,
56 struct btrfs_extent_item *ei);
57 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
58 u64 parent, u64 root_objectid,
59 u64 flags, u64 owner, u64 offset,
60 struct btrfs_key *ins, int ref_mod);
61 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
62 struct btrfs_delayed_ref_node *node,
63 struct btrfs_delayed_extent_op *extent_op);
64 static int find_next_key(struct btrfs_path *path, int level,
65 struct btrfs_key *key);
67 static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
69 return (cache->flags & bits) == bits;
72 /* simple helper to search for an existing data extent at a given offset */
73 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
75 struct btrfs_root *root = btrfs_extent_root(fs_info, start);
78 struct btrfs_path *path;
80 path = btrfs_alloc_path();
86 key.type = BTRFS_EXTENT_ITEM_KEY;
87 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
88 btrfs_free_path(path);
93 * helper function to lookup reference count and flags of a tree block.
95 * the head node for delayed ref is used to store the sum of all the
96 * reference count modifications queued up in the rbtree. the head
97 * node may also store the extent flags to set. This way you can check
98 * to see what the reference count and extent flags would be if all of
99 * the delayed refs are not processed.
101 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
102 struct btrfs_fs_info *fs_info, u64 bytenr,
103 u64 offset, int metadata, u64 *refs, u64 *flags)
105 struct btrfs_root *extent_root;
106 struct btrfs_delayed_ref_head *head;
107 struct btrfs_delayed_ref_root *delayed_refs;
108 struct btrfs_path *path;
109 struct btrfs_extent_item *ei;
110 struct extent_buffer *leaf;
111 struct btrfs_key key;
118 * If we don't have skinny metadata, don't bother doing anything
121 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
122 offset = fs_info->nodesize;
126 path = btrfs_alloc_path();
131 path->skip_locking = 1;
132 path->search_commit_root = 1;
136 key.objectid = bytenr;
139 key.type = BTRFS_METADATA_ITEM_KEY;
141 key.type = BTRFS_EXTENT_ITEM_KEY;
143 extent_root = btrfs_extent_root(fs_info, bytenr);
144 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
148 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
149 if (path->slots[0]) {
151 btrfs_item_key_to_cpu(path->nodes[0], &key,
153 if (key.objectid == bytenr &&
154 key.type == BTRFS_EXTENT_ITEM_KEY &&
155 key.offset == fs_info->nodesize)
161 leaf = path->nodes[0];
162 item_size = btrfs_item_size(leaf, path->slots[0]);
163 if (item_size >= sizeof(*ei)) {
164 ei = btrfs_item_ptr(leaf, path->slots[0],
165 struct btrfs_extent_item);
166 num_refs = btrfs_extent_refs(leaf, ei);
167 extent_flags = btrfs_extent_flags(leaf, ei);
171 "unexpected extent item size, has %u expect >= %zu",
172 item_size, sizeof(*ei));
174 btrfs_abort_transaction(trans, ret);
176 btrfs_handle_fs_error(fs_info, ret, NULL);
181 BUG_ON(num_refs == 0);
191 delayed_refs = &trans->transaction->delayed_refs;
192 spin_lock(&delayed_refs->lock);
193 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
195 if (!mutex_trylock(&head->mutex)) {
196 refcount_inc(&head->refs);
197 spin_unlock(&delayed_refs->lock);
199 btrfs_release_path(path);
202 * Mutex was contended, block until it's released and try
205 mutex_lock(&head->mutex);
206 mutex_unlock(&head->mutex);
207 btrfs_put_delayed_ref_head(head);
210 spin_lock(&head->lock);
211 if (head->extent_op && head->extent_op->update_flags)
212 extent_flags |= head->extent_op->flags_to_set;
214 BUG_ON(num_refs == 0);
216 num_refs += head->ref_mod;
217 spin_unlock(&head->lock);
218 mutex_unlock(&head->mutex);
220 spin_unlock(&delayed_refs->lock);
222 WARN_ON(num_refs == 0);
226 *flags = extent_flags;
228 btrfs_free_path(path);
233 * Back reference rules. Back refs have three main goals:
235 * 1) differentiate between all holders of references to an extent so that
236 * when a reference is dropped we can make sure it was a valid reference
237 * before freeing the extent.
239 * 2) Provide enough information to quickly find the holders of an extent
240 * if we notice a given block is corrupted or bad.
242 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
243 * maintenance. This is actually the same as #2, but with a slightly
244 * different use case.
246 * There are two kinds of back refs. The implicit back refs is optimized
247 * for pointers in non-shared tree blocks. For a given pointer in a block,
248 * back refs of this kind provide information about the block's owner tree
249 * and the pointer's key. These information allow us to find the block by
250 * b-tree searching. The full back refs is for pointers in tree blocks not
251 * referenced by their owner trees. The location of tree block is recorded
252 * in the back refs. Actually the full back refs is generic, and can be
253 * used in all cases the implicit back refs is used. The major shortcoming
254 * of the full back refs is its overhead. Every time a tree block gets
255 * COWed, we have to update back refs entry for all pointers in it.
257 * For a newly allocated tree block, we use implicit back refs for
258 * pointers in it. This means most tree related operations only involve
259 * implicit back refs. For a tree block created in old transaction, the
260 * only way to drop a reference to it is COW it. So we can detect the
261 * event that tree block loses its owner tree's reference and do the
262 * back refs conversion.
264 * When a tree block is COWed through a tree, there are four cases:
266 * The reference count of the block is one and the tree is the block's
267 * owner tree. Nothing to do in this case.
269 * The reference count of the block is one and the tree is not the
270 * block's owner tree. In this case, full back refs is used for pointers
271 * in the block. Remove these full back refs, add implicit back refs for
272 * every pointers in the new block.
274 * The reference count of the block is greater than one and the tree is
275 * the block's owner tree. In this case, implicit back refs is used for
276 * pointers in the block. Add full back refs for every pointers in the
277 * block, increase lower level extents' reference counts. The original
278 * implicit back refs are entailed to the new block.
280 * The reference count of the block is greater than one and the tree is
281 * not the block's owner tree. Add implicit back refs for every pointer in
282 * the new block, increase lower level extents' reference count.
284 * Back Reference Key composing:
286 * The key objectid corresponds to the first byte in the extent,
287 * The key type is used to differentiate between types of back refs.
288 * There are different meanings of the key offset for different types
291 * File extents can be referenced by:
293 * - multiple snapshots, subvolumes, or different generations in one subvol
294 * - different files inside a single subvolume
295 * - different offsets inside a file (bookend extents in file.c)
297 * The extent ref structure for the implicit back refs has fields for:
299 * - Objectid of the subvolume root
300 * - objectid of the file holding the reference
301 * - original offset in the file
302 * - how many bookend extents
304 * The key offset for the implicit back refs is hash of the first
307 * The extent ref structure for the full back refs has field for:
309 * - number of pointers in the tree leaf
311 * The key offset for the implicit back refs is the first byte of
314 * When a file extent is allocated, The implicit back refs is used.
315 * the fields are filled in:
317 * (root_key.objectid, inode objectid, offset in file, 1)
319 * When a file extent is removed file truncation, we find the
320 * corresponding implicit back refs and check the following fields:
322 * (btrfs_header_owner(leaf), inode objectid, offset in file)
324 * Btree extents can be referenced by:
326 * - Different subvolumes
328 * Both the implicit back refs and the full back refs for tree blocks
329 * only consist of key. The key offset for the implicit back refs is
330 * objectid of block's owner tree. The key offset for the full back refs
331 * is the first byte of parent block.
333 * When implicit back refs is used, information about the lowest key and
334 * level of the tree block are required. These information are stored in
335 * tree block info structure.
339 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
340 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
341 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
343 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
344 struct btrfs_extent_inline_ref *iref,
345 enum btrfs_inline_ref_type is_data)
347 int type = btrfs_extent_inline_ref_type(eb, iref);
348 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
350 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
351 type == BTRFS_SHARED_BLOCK_REF_KEY ||
352 type == BTRFS_SHARED_DATA_REF_KEY ||
353 type == BTRFS_EXTENT_DATA_REF_KEY) {
354 if (is_data == BTRFS_REF_TYPE_BLOCK) {
355 if (type == BTRFS_TREE_BLOCK_REF_KEY)
357 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
360 * Every shared one has parent tree block,
361 * which must be aligned to sector size.
364 IS_ALIGNED(offset, eb->fs_info->sectorsize))
367 } else if (is_data == BTRFS_REF_TYPE_DATA) {
368 if (type == BTRFS_EXTENT_DATA_REF_KEY)
370 if (type == BTRFS_SHARED_DATA_REF_KEY) {
373 * Every shared one has parent tree block,
374 * which must be aligned to sector size.
377 IS_ALIGNED(offset, eb->fs_info->sectorsize))
381 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
387 btrfs_print_leaf(eb);
388 btrfs_err(eb->fs_info,
389 "eb %llu iref 0x%lx invalid extent inline ref type %d",
390 eb->start, (unsigned long)iref, type);
392 return BTRFS_REF_TYPE_INVALID;
395 u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
397 u32 high_crc = ~(u32)0;
398 u32 low_crc = ~(u32)0;
401 lenum = cpu_to_le64(root_objectid);
402 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
403 lenum = cpu_to_le64(owner);
404 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
405 lenum = cpu_to_le64(offset);
406 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
408 return ((u64)high_crc << 31) ^ (u64)low_crc;
411 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
412 struct btrfs_extent_data_ref *ref)
414 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
415 btrfs_extent_data_ref_objectid(leaf, ref),
416 btrfs_extent_data_ref_offset(leaf, ref));
419 static int match_extent_data_ref(struct extent_buffer *leaf,
420 struct btrfs_extent_data_ref *ref,
421 u64 root_objectid, u64 owner, u64 offset)
423 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
424 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
425 btrfs_extent_data_ref_offset(leaf, ref) != offset)
430 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
431 struct btrfs_path *path,
432 u64 bytenr, u64 parent,
434 u64 owner, u64 offset)
436 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
437 struct btrfs_key key;
438 struct btrfs_extent_data_ref *ref;
439 struct extent_buffer *leaf;
445 key.objectid = bytenr;
447 key.type = BTRFS_SHARED_DATA_REF_KEY;
450 key.type = BTRFS_EXTENT_DATA_REF_KEY;
451 key.offset = hash_extent_data_ref(root_objectid,
456 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
468 leaf = path->nodes[0];
469 nritems = btrfs_header_nritems(leaf);
471 if (path->slots[0] >= nritems) {
472 ret = btrfs_next_leaf(root, path);
478 leaf = path->nodes[0];
479 nritems = btrfs_header_nritems(leaf);
483 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
484 if (key.objectid != bytenr ||
485 key.type != BTRFS_EXTENT_DATA_REF_KEY)
488 ref = btrfs_item_ptr(leaf, path->slots[0],
489 struct btrfs_extent_data_ref);
491 if (match_extent_data_ref(leaf, ref, root_objectid,
494 btrfs_release_path(path);
506 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
507 struct btrfs_path *path,
508 u64 bytenr, u64 parent,
509 u64 root_objectid, u64 owner,
510 u64 offset, int refs_to_add)
512 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
513 struct btrfs_key key;
514 struct extent_buffer *leaf;
519 key.objectid = bytenr;
521 key.type = BTRFS_SHARED_DATA_REF_KEY;
523 size = sizeof(struct btrfs_shared_data_ref);
525 key.type = BTRFS_EXTENT_DATA_REF_KEY;
526 key.offset = hash_extent_data_ref(root_objectid,
528 size = sizeof(struct btrfs_extent_data_ref);
531 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
532 if (ret && ret != -EEXIST)
535 leaf = path->nodes[0];
537 struct btrfs_shared_data_ref *ref;
538 ref = btrfs_item_ptr(leaf, path->slots[0],
539 struct btrfs_shared_data_ref);
541 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
543 num_refs = btrfs_shared_data_ref_count(leaf, ref);
544 num_refs += refs_to_add;
545 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
548 struct btrfs_extent_data_ref *ref;
549 while (ret == -EEXIST) {
550 ref = btrfs_item_ptr(leaf, path->slots[0],
551 struct btrfs_extent_data_ref);
552 if (match_extent_data_ref(leaf, ref, root_objectid,
555 btrfs_release_path(path);
557 ret = btrfs_insert_empty_item(trans, root, path, &key,
559 if (ret && ret != -EEXIST)
562 leaf = path->nodes[0];
564 ref = btrfs_item_ptr(leaf, path->slots[0],
565 struct btrfs_extent_data_ref);
567 btrfs_set_extent_data_ref_root(leaf, ref,
569 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
570 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
571 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
573 num_refs = btrfs_extent_data_ref_count(leaf, ref);
574 num_refs += refs_to_add;
575 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
578 btrfs_mark_buffer_dirty(trans, leaf);
581 btrfs_release_path(path);
585 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
586 struct btrfs_root *root,
587 struct btrfs_path *path,
590 struct btrfs_key key;
591 struct btrfs_extent_data_ref *ref1 = NULL;
592 struct btrfs_shared_data_ref *ref2 = NULL;
593 struct extent_buffer *leaf;
597 leaf = path->nodes[0];
598 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
600 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
601 ref1 = btrfs_item_ptr(leaf, path->slots[0],
602 struct btrfs_extent_data_ref);
603 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
604 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
605 ref2 = btrfs_item_ptr(leaf, path->slots[0],
606 struct btrfs_shared_data_ref);
607 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
609 btrfs_err(trans->fs_info,
610 "unrecognized backref key (%llu %u %llu)",
611 key.objectid, key.type, key.offset);
612 btrfs_abort_transaction(trans, -EUCLEAN);
616 BUG_ON(num_refs < refs_to_drop);
617 num_refs -= refs_to_drop;
620 ret = btrfs_del_item(trans, root, path);
622 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
623 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
624 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
625 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
626 btrfs_mark_buffer_dirty(trans, leaf);
631 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
632 struct btrfs_extent_inline_ref *iref)
634 struct btrfs_key key;
635 struct extent_buffer *leaf;
636 struct btrfs_extent_data_ref *ref1;
637 struct btrfs_shared_data_ref *ref2;
641 leaf = path->nodes[0];
642 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
646 * If type is invalid, we should have bailed out earlier than
649 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
650 ASSERT(type != BTRFS_REF_TYPE_INVALID);
651 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
652 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
653 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
655 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
656 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
658 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
659 ref1 = btrfs_item_ptr(leaf, path->slots[0],
660 struct btrfs_extent_data_ref);
661 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
662 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
663 ref2 = btrfs_item_ptr(leaf, path->slots[0],
664 struct btrfs_shared_data_ref);
665 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
672 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
673 struct btrfs_path *path,
674 u64 bytenr, u64 parent,
677 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
678 struct btrfs_key key;
681 key.objectid = bytenr;
683 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
686 key.type = BTRFS_TREE_BLOCK_REF_KEY;
687 key.offset = root_objectid;
690 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
696 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
697 struct btrfs_path *path,
698 u64 bytenr, u64 parent,
701 struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
702 struct btrfs_key key;
705 key.objectid = bytenr;
707 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
710 key.type = BTRFS_TREE_BLOCK_REF_KEY;
711 key.offset = root_objectid;
714 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
715 btrfs_release_path(path);
719 static inline int extent_ref_type(u64 parent, u64 owner)
722 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
724 type = BTRFS_SHARED_BLOCK_REF_KEY;
726 type = BTRFS_TREE_BLOCK_REF_KEY;
729 type = BTRFS_SHARED_DATA_REF_KEY;
731 type = BTRFS_EXTENT_DATA_REF_KEY;
736 static int find_next_key(struct btrfs_path *path, int level,
737 struct btrfs_key *key)
740 for (; level < BTRFS_MAX_LEVEL; level++) {
741 if (!path->nodes[level])
743 if (path->slots[level] + 1 >=
744 btrfs_header_nritems(path->nodes[level]))
747 btrfs_item_key_to_cpu(path->nodes[level], key,
748 path->slots[level] + 1);
750 btrfs_node_key_to_cpu(path->nodes[level], key,
751 path->slots[level] + 1);
758 * look for inline back ref. if back ref is found, *ref_ret is set
759 * to the address of inline back ref, and 0 is returned.
761 * if back ref isn't found, *ref_ret is set to the address where it
762 * should be inserted, and -ENOENT is returned.
764 * if insert is true and there are too many inline back refs, the path
765 * points to the extent item, and -EAGAIN is returned.
767 * NOTE: inline back refs are ordered in the same way that back ref
768 * items in the tree are ordered.
770 static noinline_for_stack
771 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
772 struct btrfs_path *path,
773 struct btrfs_extent_inline_ref **ref_ret,
774 u64 bytenr, u64 num_bytes,
775 u64 parent, u64 root_objectid,
776 u64 owner, u64 offset, int insert)
778 struct btrfs_fs_info *fs_info = trans->fs_info;
779 struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
780 struct btrfs_key key;
781 struct extent_buffer *leaf;
782 struct btrfs_extent_item *ei;
783 struct btrfs_extent_inline_ref *iref;
793 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
796 key.objectid = bytenr;
797 key.type = BTRFS_EXTENT_ITEM_KEY;
798 key.offset = num_bytes;
800 want = extent_ref_type(parent, owner);
802 extra_size = btrfs_extent_inline_ref_size(want);
803 path->search_for_extension = 1;
804 path->keep_locks = 1;
809 * Owner is our level, so we can just add one to get the level for the
810 * block we are interested in.
812 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
813 key.type = BTRFS_METADATA_ITEM_KEY;
818 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
825 * We may be a newly converted file system which still has the old fat
826 * extent entries for metadata, so try and see if we have one of those.
828 if (ret > 0 && skinny_metadata) {
829 skinny_metadata = false;
830 if (path->slots[0]) {
832 btrfs_item_key_to_cpu(path->nodes[0], &key,
834 if (key.objectid == bytenr &&
835 key.type == BTRFS_EXTENT_ITEM_KEY &&
836 key.offset == num_bytes)
840 key.objectid = bytenr;
841 key.type = BTRFS_EXTENT_ITEM_KEY;
842 key.offset = num_bytes;
843 btrfs_release_path(path);
848 if (ret && !insert) {
851 } else if (WARN_ON(ret)) {
852 btrfs_print_leaf(path->nodes[0]);
854 "extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu",
855 bytenr, num_bytes, parent, root_objectid, owner,
861 leaf = path->nodes[0];
862 item_size = btrfs_item_size(leaf, path->slots[0]);
863 if (unlikely(item_size < sizeof(*ei))) {
866 "unexpected extent item size, has %llu expect >= %zu",
867 item_size, sizeof(*ei));
868 btrfs_abort_transaction(trans, err);
872 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
873 flags = btrfs_extent_flags(leaf, ei);
875 ptr = (unsigned long)(ei + 1);
876 end = (unsigned long)ei + item_size;
878 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
879 ptr += sizeof(struct btrfs_tree_block_info);
883 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
884 needed = BTRFS_REF_TYPE_DATA;
886 needed = BTRFS_REF_TYPE_BLOCK;
893 btrfs_print_leaf(path->nodes[0]);
895 "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu",
896 path->slots[0], root_objectid, owner, offset, parent);
900 iref = (struct btrfs_extent_inline_ref *)ptr;
901 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
902 if (type == BTRFS_REF_TYPE_INVALID) {
910 ptr += btrfs_extent_inline_ref_size(type);
914 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
915 struct btrfs_extent_data_ref *dref;
916 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
917 if (match_extent_data_ref(leaf, dref, root_objectid,
922 if (hash_extent_data_ref_item(leaf, dref) <
923 hash_extent_data_ref(root_objectid, owner, offset))
927 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
929 if (parent == ref_offset) {
933 if (ref_offset < parent)
936 if (root_objectid == ref_offset) {
940 if (ref_offset < root_objectid)
944 ptr += btrfs_extent_inline_ref_size(type);
946 if (err == -ENOENT && insert) {
947 if (item_size + extra_size >=
948 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
953 * To add new inline back ref, we have to make sure
954 * there is no corresponding back ref item.
955 * For simplicity, we just do not add new inline back
956 * ref if there is any kind of item for this block
958 if (find_next_key(path, 0, &key) == 0 &&
959 key.objectid == bytenr &&
960 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
965 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
968 path->keep_locks = 0;
969 path->search_for_extension = 0;
970 btrfs_unlock_up_safe(path, 1);
976 * helper to add new inline back ref
978 static noinline_for_stack
979 void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
980 struct btrfs_path *path,
981 struct btrfs_extent_inline_ref *iref,
982 u64 parent, u64 root_objectid,
983 u64 owner, u64 offset, int refs_to_add,
984 struct btrfs_delayed_extent_op *extent_op)
986 struct extent_buffer *leaf;
987 struct btrfs_extent_item *ei;
990 unsigned long item_offset;
995 leaf = path->nodes[0];
996 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
997 item_offset = (unsigned long)iref - (unsigned long)ei;
999 type = extent_ref_type(parent, owner);
1000 size = btrfs_extent_inline_ref_size(type);
1002 btrfs_extend_item(trans, path, size);
1004 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1005 refs = btrfs_extent_refs(leaf, ei);
1006 refs += refs_to_add;
1007 btrfs_set_extent_refs(leaf, ei, refs);
1009 __run_delayed_extent_op(extent_op, leaf, ei);
1011 ptr = (unsigned long)ei + item_offset;
1012 end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]);
1013 if (ptr < end - size)
1014 memmove_extent_buffer(leaf, ptr + size, ptr,
1017 iref = (struct btrfs_extent_inline_ref *)ptr;
1018 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1019 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1020 struct btrfs_extent_data_ref *dref;
1021 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1022 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1023 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1024 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1025 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1026 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1027 struct btrfs_shared_data_ref *sref;
1028 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1029 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1030 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1031 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1032 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1034 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1036 btrfs_mark_buffer_dirty(trans, leaf);
1039 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1040 struct btrfs_path *path,
1041 struct btrfs_extent_inline_ref **ref_ret,
1042 u64 bytenr, u64 num_bytes, u64 parent,
1043 u64 root_objectid, u64 owner, u64 offset)
1047 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1048 num_bytes, parent, root_objectid,
1053 btrfs_release_path(path);
1056 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1057 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1060 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1061 root_objectid, owner, offset);
1067 * helper to update/remove inline back ref
1069 static noinline_for_stack int update_inline_extent_backref(
1070 struct btrfs_trans_handle *trans,
1071 struct btrfs_path *path,
1072 struct btrfs_extent_inline_ref *iref,
1074 struct btrfs_delayed_extent_op *extent_op)
1076 struct extent_buffer *leaf = path->nodes[0];
1077 struct btrfs_fs_info *fs_info = leaf->fs_info;
1078 struct btrfs_extent_item *ei;
1079 struct btrfs_extent_data_ref *dref = NULL;
1080 struct btrfs_shared_data_ref *sref = NULL;
1088 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1089 refs = btrfs_extent_refs(leaf, ei);
1090 if (unlikely(refs_to_mod < 0 && refs + refs_to_mod <= 0)) {
1091 struct btrfs_key key;
1094 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1095 if (key.type == BTRFS_METADATA_ITEM_KEY)
1096 extent_size = fs_info->nodesize;
1098 extent_size = key.offset;
1099 btrfs_print_leaf(leaf);
1101 "invalid refs_to_mod for extent %llu num_bytes %u, has %d expect >= -%llu",
1102 key.objectid, extent_size, refs_to_mod, refs);
1105 refs += refs_to_mod;
1106 btrfs_set_extent_refs(leaf, ei, refs);
1108 __run_delayed_extent_op(extent_op, leaf, ei);
1110 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1112 * Function btrfs_get_extent_inline_ref_type() has already printed
1115 if (unlikely(type == BTRFS_REF_TYPE_INVALID))
1118 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1119 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1120 refs = btrfs_extent_data_ref_count(leaf, dref);
1121 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1122 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1123 refs = btrfs_shared_data_ref_count(leaf, sref);
1127 * For tree blocks we can only drop one ref for it, and tree
1128 * blocks should not have refs > 1.
1130 * Furthermore if we're inserting a new inline backref, we
1131 * won't reach this path either. That would be
1132 * setup_inline_extent_backref().
1134 if (unlikely(refs_to_mod != -1)) {
1135 struct btrfs_key key;
1137 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1139 btrfs_print_leaf(leaf);
1141 "invalid refs_to_mod for tree block %llu, has %d expect -1",
1142 key.objectid, refs_to_mod);
1147 if (unlikely(refs_to_mod < 0 && refs < -refs_to_mod)) {
1148 struct btrfs_key key;
1151 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1152 if (key.type == BTRFS_METADATA_ITEM_KEY)
1153 extent_size = fs_info->nodesize;
1155 extent_size = key.offset;
1156 btrfs_print_leaf(leaf);
1158 "invalid refs_to_mod for backref entry, iref %lu extent %llu num_bytes %u, has %d expect >= -%llu",
1159 (unsigned long)iref, key.objectid, extent_size,
1163 refs += refs_to_mod;
1166 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1167 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1169 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1171 size = btrfs_extent_inline_ref_size(type);
1172 item_size = btrfs_item_size(leaf, path->slots[0]);
1173 ptr = (unsigned long)iref;
1174 end = (unsigned long)ei + item_size;
1175 if (ptr + size < end)
1176 memmove_extent_buffer(leaf, ptr, ptr + size,
1179 btrfs_truncate_item(trans, path, item_size, 1);
1181 btrfs_mark_buffer_dirty(trans, leaf);
1185 static noinline_for_stack
1186 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1187 struct btrfs_path *path,
1188 u64 bytenr, u64 num_bytes, u64 parent,
1189 u64 root_objectid, u64 owner,
1190 u64 offset, int refs_to_add,
1191 struct btrfs_delayed_extent_op *extent_op)
1193 struct btrfs_extent_inline_ref *iref;
1196 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1197 num_bytes, parent, root_objectid,
1201 * We're adding refs to a tree block we already own, this
1202 * should not happen at all.
1204 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1205 btrfs_print_leaf(path->nodes[0]);
1206 btrfs_crit(trans->fs_info,
1207 "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu slot %u",
1208 bytenr, num_bytes, root_objectid, path->slots[0]);
1211 ret = update_inline_extent_backref(trans, path, iref,
1212 refs_to_add, extent_op);
1213 } else if (ret == -ENOENT) {
1214 setup_inline_extent_backref(trans, path, iref, parent,
1215 root_objectid, owner, offset,
1216 refs_to_add, extent_op);
1222 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1223 struct btrfs_root *root,
1224 struct btrfs_path *path,
1225 struct btrfs_extent_inline_ref *iref,
1226 int refs_to_drop, int is_data)
1230 BUG_ON(!is_data && refs_to_drop != 1);
1232 ret = update_inline_extent_backref(trans, path, iref,
1233 -refs_to_drop, NULL);
1235 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1237 ret = btrfs_del_item(trans, root, path);
1241 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1242 u64 *discarded_bytes)
1245 u64 bytes_left, end;
1246 u64 aligned_start = ALIGN(start, 1 << SECTOR_SHIFT);
1248 /* Adjust the range to be aligned to 512B sectors if necessary. */
1249 if (start != aligned_start) {
1250 len -= aligned_start - start;
1251 len = round_down(len, 1 << SECTOR_SHIFT);
1252 start = aligned_start;
1255 *discarded_bytes = 0;
1263 /* Skip any superblocks on this device. */
1264 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1265 u64 sb_start = btrfs_sb_offset(j);
1266 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1267 u64 size = sb_start - start;
1269 if (!in_range(sb_start, start, bytes_left) &&
1270 !in_range(sb_end, start, bytes_left) &&
1271 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1275 * Superblock spans beginning of range. Adjust start and
1278 if (sb_start <= start) {
1279 start += sb_end - start;
1284 bytes_left = end - start;
1289 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1290 size >> SECTOR_SHIFT,
1293 *discarded_bytes += size;
1294 else if (ret != -EOPNOTSUPP)
1303 bytes_left = end - start;
1307 ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1308 bytes_left >> SECTOR_SHIFT,
1311 *discarded_bytes += bytes_left;
1316 static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes)
1318 struct btrfs_device *dev = stripe->dev;
1319 struct btrfs_fs_info *fs_info = dev->fs_info;
1320 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1321 u64 phys = stripe->physical;
1322 u64 len = stripe->length;
1326 /* Zone reset on a zoned filesystem */
1327 if (btrfs_can_zone_reset(dev, phys, len)) {
1330 ret = btrfs_reset_device_zone(dev, phys, len, &discarded);
1334 if (!btrfs_dev_replace_is_ongoing(dev_replace) ||
1335 dev != dev_replace->srcdev)
1338 src_disc = discarded;
1340 /* Send to replace target as well */
1341 ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len,
1343 discarded += src_disc;
1344 } else if (bdev_max_discard_sectors(stripe->dev->bdev)) {
1345 ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded);
1356 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1357 u64 num_bytes, u64 *actual_bytes)
1360 u64 discarded_bytes = 0;
1361 u64 end = bytenr + num_bytes;
1365 * Avoid races with device replace and make sure the devices in the
1366 * stripes don't go away while we are discarding.
1368 btrfs_bio_counter_inc_blocked(fs_info);
1370 struct btrfs_discard_stripe *stripes;
1371 unsigned int num_stripes;
1374 num_bytes = end - cur;
1375 stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes);
1376 if (IS_ERR(stripes)) {
1377 ret = PTR_ERR(stripes);
1378 if (ret == -EOPNOTSUPP)
1383 for (i = 0; i < num_stripes; i++) {
1384 struct btrfs_discard_stripe *stripe = stripes + i;
1387 if (!stripe->dev->bdev) {
1388 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1392 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
1393 &stripe->dev->dev_state))
1396 ret = do_discard_extent(stripe, &bytes);
1399 * Keep going if discard is not supported by the
1402 if (ret != -EOPNOTSUPP)
1406 discarded_bytes += bytes;
1414 btrfs_bio_counter_dec(fs_info);
1416 *actual_bytes = discarded_bytes;
1420 /* Can return -ENOMEM */
1421 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1422 struct btrfs_ref *generic_ref)
1424 struct btrfs_fs_info *fs_info = trans->fs_info;
1427 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1428 generic_ref->action);
1429 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1430 generic_ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID);
1432 if (generic_ref->type == BTRFS_REF_METADATA)
1433 ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1435 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1437 btrfs_ref_tree_mod(fs_info, generic_ref);
1443 * __btrfs_inc_extent_ref - insert backreference for a given extent
1445 * The counterpart is in __btrfs_free_extent(), with examples and more details
1448 * @trans: Handle of transaction
1450 * @node: The delayed ref node used to get the bytenr/length for
1451 * extent whose references are incremented.
1453 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1454 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1455 * bytenr of the parent block. Since new extents are always
1456 * created with indirect references, this will only be the case
1457 * when relocating a shared extent. In that case, root_objectid
1458 * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1461 * @root_objectid: The id of the root where this modification has originated,
1462 * this can be either one of the well-known metadata trees or
1463 * the subvolume id which references this extent.
1465 * @owner: For data extents it is the inode number of the owning file.
1466 * For metadata extents this parameter holds the level in the
1467 * tree of the extent.
1469 * @offset: For metadata extents the offset is ignored and is currently
1470 * always passed as 0. For data extents it is the fileoffset
1471 * this extent belongs to.
1473 * @refs_to_add Number of references to add
1475 * @extent_op Pointer to a structure, holding information necessary when
1476 * updating a tree block's flags
1479 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1480 struct btrfs_delayed_ref_node *node,
1481 u64 parent, u64 root_objectid,
1482 u64 owner, u64 offset, int refs_to_add,
1483 struct btrfs_delayed_extent_op *extent_op)
1485 struct btrfs_path *path;
1486 struct extent_buffer *leaf;
1487 struct btrfs_extent_item *item;
1488 struct btrfs_key key;
1489 u64 bytenr = node->bytenr;
1490 u64 num_bytes = node->num_bytes;
1494 path = btrfs_alloc_path();
1498 /* this will setup the path even if it fails to insert the back ref */
1499 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1500 parent, root_objectid, owner,
1501 offset, refs_to_add, extent_op);
1502 if ((ret < 0 && ret != -EAGAIN) || !ret)
1506 * Ok we had -EAGAIN which means we didn't have space to insert and
1507 * inline extent ref, so just update the reference count and add a
1510 leaf = path->nodes[0];
1511 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1512 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1513 refs = btrfs_extent_refs(leaf, item);
1514 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1516 __run_delayed_extent_op(extent_op, leaf, item);
1518 btrfs_mark_buffer_dirty(trans, leaf);
1519 btrfs_release_path(path);
1521 /* now insert the actual backref */
1522 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1523 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1526 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1527 root_objectid, owner, offset,
1531 btrfs_abort_transaction(trans, ret);
1533 btrfs_free_path(path);
1537 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1538 struct btrfs_delayed_ref_node *node,
1539 struct btrfs_delayed_extent_op *extent_op,
1540 bool insert_reserved)
1543 struct btrfs_delayed_data_ref *ref;
1544 struct btrfs_key ins;
1549 ins.objectid = node->bytenr;
1550 ins.offset = node->num_bytes;
1551 ins.type = BTRFS_EXTENT_ITEM_KEY;
1553 ref = btrfs_delayed_node_to_data_ref(node);
1554 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1556 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1557 parent = ref->parent;
1558 ref_root = ref->root;
1560 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1562 flags |= extent_op->flags_to_set;
1563 ret = alloc_reserved_file_extent(trans, parent, ref_root,
1564 flags, ref->objectid,
1567 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1568 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1569 ref->objectid, ref->offset,
1570 node->ref_mod, extent_op);
1571 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1572 ret = __btrfs_free_extent(trans, node, parent,
1573 ref_root, ref->objectid,
1574 ref->offset, node->ref_mod,
1582 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1583 struct extent_buffer *leaf,
1584 struct btrfs_extent_item *ei)
1586 u64 flags = btrfs_extent_flags(leaf, ei);
1587 if (extent_op->update_flags) {
1588 flags |= extent_op->flags_to_set;
1589 btrfs_set_extent_flags(leaf, ei, flags);
1592 if (extent_op->update_key) {
1593 struct btrfs_tree_block_info *bi;
1594 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1595 bi = (struct btrfs_tree_block_info *)(ei + 1);
1596 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1600 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1601 struct btrfs_delayed_ref_head *head,
1602 struct btrfs_delayed_extent_op *extent_op)
1604 struct btrfs_fs_info *fs_info = trans->fs_info;
1605 struct btrfs_root *root;
1606 struct btrfs_key key;
1607 struct btrfs_path *path;
1608 struct btrfs_extent_item *ei;
1609 struct extent_buffer *leaf;
1615 if (TRANS_ABORTED(trans))
1618 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1621 path = btrfs_alloc_path();
1625 key.objectid = head->bytenr;
1628 key.type = BTRFS_METADATA_ITEM_KEY;
1629 key.offset = extent_op->level;
1631 key.type = BTRFS_EXTENT_ITEM_KEY;
1632 key.offset = head->num_bytes;
1635 root = btrfs_extent_root(fs_info, key.objectid);
1637 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1644 if (path->slots[0] > 0) {
1646 btrfs_item_key_to_cpu(path->nodes[0], &key,
1648 if (key.objectid == head->bytenr &&
1649 key.type == BTRFS_EXTENT_ITEM_KEY &&
1650 key.offset == head->num_bytes)
1654 btrfs_release_path(path);
1657 key.objectid = head->bytenr;
1658 key.offset = head->num_bytes;
1659 key.type = BTRFS_EXTENT_ITEM_KEY;
1665 "missing extent item for extent %llu num_bytes %llu level %d",
1666 head->bytenr, head->num_bytes, extent_op->level);
1671 leaf = path->nodes[0];
1672 item_size = btrfs_item_size(leaf, path->slots[0]);
1674 if (unlikely(item_size < sizeof(*ei))) {
1677 "unexpected extent item size, has %u expect >= %zu",
1678 item_size, sizeof(*ei));
1679 btrfs_abort_transaction(trans, err);
1683 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1684 __run_delayed_extent_op(extent_op, leaf, ei);
1686 btrfs_mark_buffer_dirty(trans, leaf);
1688 btrfs_free_path(path);
1692 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1693 struct btrfs_delayed_ref_node *node,
1694 struct btrfs_delayed_extent_op *extent_op,
1695 bool insert_reserved)
1698 struct btrfs_delayed_tree_ref *ref;
1702 ref = btrfs_delayed_node_to_tree_ref(node);
1703 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1705 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1706 parent = ref->parent;
1707 ref_root = ref->root;
1709 if (unlikely(node->ref_mod != 1)) {
1710 btrfs_err(trans->fs_info,
1711 "btree block %llu has %d references rather than 1: action %d ref_root %llu parent %llu",
1712 node->bytenr, node->ref_mod, node->action, ref_root,
1716 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1717 BUG_ON(!extent_op || !extent_op->update_flags);
1718 ret = alloc_reserved_tree_block(trans, node, extent_op);
1719 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1720 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1721 ref->level, 0, 1, extent_op);
1722 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1723 ret = __btrfs_free_extent(trans, node, parent, ref_root,
1724 ref->level, 0, 1, extent_op);
1731 /* helper function to actually process a single delayed ref entry */
1732 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1733 struct btrfs_delayed_ref_node *node,
1734 struct btrfs_delayed_extent_op *extent_op,
1735 bool insert_reserved)
1739 if (TRANS_ABORTED(trans)) {
1740 if (insert_reserved)
1741 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1745 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1746 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1747 ret = run_delayed_tree_ref(trans, node, extent_op,
1749 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1750 node->type == BTRFS_SHARED_DATA_REF_KEY)
1751 ret = run_delayed_data_ref(trans, node, extent_op,
1755 if (ret && insert_reserved)
1756 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1758 btrfs_err(trans->fs_info,
1759 "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d",
1760 node->bytenr, node->num_bytes, node->type,
1761 node->action, node->ref_mod, ret);
1765 static inline struct btrfs_delayed_ref_node *
1766 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1768 struct btrfs_delayed_ref_node *ref;
1770 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1774 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1775 * This is to prevent a ref count from going down to zero, which deletes
1776 * the extent item from the extent tree, when there still are references
1777 * to add, which would fail because they would not find the extent item.
1779 if (!list_empty(&head->ref_add_list))
1780 return list_first_entry(&head->ref_add_list,
1781 struct btrfs_delayed_ref_node, add_list);
1783 ref = rb_entry(rb_first_cached(&head->ref_tree),
1784 struct btrfs_delayed_ref_node, ref_node);
1785 ASSERT(list_empty(&ref->add_list));
1789 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1790 struct btrfs_delayed_ref_head *head)
1792 spin_lock(&delayed_refs->lock);
1793 head->processing = false;
1794 delayed_refs->num_heads_ready++;
1795 spin_unlock(&delayed_refs->lock);
1796 btrfs_delayed_ref_unlock(head);
1799 static struct btrfs_delayed_extent_op *cleanup_extent_op(
1800 struct btrfs_delayed_ref_head *head)
1802 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1807 if (head->must_insert_reserved) {
1808 head->extent_op = NULL;
1809 btrfs_free_delayed_extent_op(extent_op);
1815 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1816 struct btrfs_delayed_ref_head *head)
1818 struct btrfs_delayed_extent_op *extent_op;
1821 extent_op = cleanup_extent_op(head);
1824 head->extent_op = NULL;
1825 spin_unlock(&head->lock);
1826 ret = run_delayed_extent_op(trans, head, extent_op);
1827 btrfs_free_delayed_extent_op(extent_op);
1828 return ret ? ret : 1;
1831 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1832 struct btrfs_delayed_ref_root *delayed_refs,
1833 struct btrfs_delayed_ref_head *head)
1835 int nr_items = 1; /* Dropping this ref head update. */
1838 * We had csum deletions accounted for in our delayed refs rsv, we need
1839 * to drop the csum leaves for this update from our delayed_refs_rsv.
1841 if (head->total_ref_mod < 0 && head->is_data) {
1842 spin_lock(&delayed_refs->lock);
1843 delayed_refs->pending_csums -= head->num_bytes;
1844 spin_unlock(&delayed_refs->lock);
1845 nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1848 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1851 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1852 struct btrfs_delayed_ref_head *head)
1855 struct btrfs_fs_info *fs_info = trans->fs_info;
1856 struct btrfs_delayed_ref_root *delayed_refs;
1859 delayed_refs = &trans->transaction->delayed_refs;
1861 ret = run_and_cleanup_extent_op(trans, head);
1863 unselect_delayed_ref_head(delayed_refs, head);
1864 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1871 * Need to drop our head ref lock and re-acquire the delayed ref lock
1872 * and then re-check to make sure nobody got added.
1874 spin_unlock(&head->lock);
1875 spin_lock(&delayed_refs->lock);
1876 spin_lock(&head->lock);
1877 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1878 spin_unlock(&head->lock);
1879 spin_unlock(&delayed_refs->lock);
1882 btrfs_delete_ref_head(delayed_refs, head);
1883 spin_unlock(&head->lock);
1884 spin_unlock(&delayed_refs->lock);
1886 if (head->must_insert_reserved) {
1887 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1888 if (head->is_data) {
1889 struct btrfs_root *csum_root;
1891 csum_root = btrfs_csum_root(fs_info, head->bytenr);
1892 ret = btrfs_del_csums(trans, csum_root, head->bytenr,
1897 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1899 trace_run_delayed_ref_head(fs_info, head, 0);
1900 btrfs_delayed_ref_unlock(head);
1901 btrfs_put_delayed_ref_head(head);
1905 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1906 struct btrfs_trans_handle *trans)
1908 struct btrfs_delayed_ref_root *delayed_refs =
1909 &trans->transaction->delayed_refs;
1910 struct btrfs_delayed_ref_head *head = NULL;
1913 spin_lock(&delayed_refs->lock);
1914 head = btrfs_select_ref_head(delayed_refs);
1916 spin_unlock(&delayed_refs->lock);
1921 * Grab the lock that says we are going to process all the refs for
1924 ret = btrfs_delayed_ref_lock(delayed_refs, head);
1925 spin_unlock(&delayed_refs->lock);
1928 * We may have dropped the spin lock to get the head mutex lock, and
1929 * that might have given someone else time to free the head. If that's
1930 * true, it has been removed from our list and we can move on.
1933 head = ERR_PTR(-EAGAIN);
1938 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1939 struct btrfs_delayed_ref_head *locked_ref)
1941 struct btrfs_fs_info *fs_info = trans->fs_info;
1942 struct btrfs_delayed_ref_root *delayed_refs;
1943 struct btrfs_delayed_extent_op *extent_op;
1944 struct btrfs_delayed_ref_node *ref;
1945 bool must_insert_reserved;
1948 delayed_refs = &trans->transaction->delayed_refs;
1950 lockdep_assert_held(&locked_ref->mutex);
1951 lockdep_assert_held(&locked_ref->lock);
1953 while ((ref = select_delayed_ref(locked_ref))) {
1955 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1956 spin_unlock(&locked_ref->lock);
1957 unselect_delayed_ref_head(delayed_refs, locked_ref);
1961 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1962 RB_CLEAR_NODE(&ref->ref_node);
1963 if (!list_empty(&ref->add_list))
1964 list_del(&ref->add_list);
1966 * When we play the delayed ref, also correct the ref_mod on
1969 switch (ref->action) {
1970 case BTRFS_ADD_DELAYED_REF:
1971 case BTRFS_ADD_DELAYED_EXTENT:
1972 locked_ref->ref_mod -= ref->ref_mod;
1974 case BTRFS_DROP_DELAYED_REF:
1975 locked_ref->ref_mod += ref->ref_mod;
1980 atomic_dec(&delayed_refs->num_entries);
1983 * Record the must_insert_reserved flag before we drop the
1986 must_insert_reserved = locked_ref->must_insert_reserved;
1987 locked_ref->must_insert_reserved = false;
1989 extent_op = locked_ref->extent_op;
1990 locked_ref->extent_op = NULL;
1991 spin_unlock(&locked_ref->lock);
1993 ret = run_one_delayed_ref(trans, ref, extent_op,
1994 must_insert_reserved);
1996 btrfs_free_delayed_extent_op(extent_op);
1998 unselect_delayed_ref_head(delayed_refs, locked_ref);
1999 btrfs_put_delayed_ref(ref);
2003 btrfs_put_delayed_ref(ref);
2006 spin_lock(&locked_ref->lock);
2007 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2014 * Returns 0 on success or if called with an already aborted transaction.
2015 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2017 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2020 struct btrfs_fs_info *fs_info = trans->fs_info;
2021 struct btrfs_delayed_ref_root *delayed_refs;
2022 struct btrfs_delayed_ref_head *locked_ref = NULL;
2024 unsigned long count = 0;
2026 delayed_refs = &trans->transaction->delayed_refs;
2029 locked_ref = btrfs_obtain_ref_head(trans);
2030 if (IS_ERR_OR_NULL(locked_ref)) {
2031 if (PTR_ERR(locked_ref) == -EAGAIN) {
2040 * We need to try and merge add/drops of the same ref since we
2041 * can run into issues with relocate dropping the implicit ref
2042 * and then it being added back again before the drop can
2043 * finish. If we merged anything we need to re-loop so we can
2045 * Or we can get node references of the same type that weren't
2046 * merged when created due to bumps in the tree mod seq, and
2047 * we need to merge them to prevent adding an inline extent
2048 * backref before dropping it (triggering a BUG_ON at
2049 * insert_inline_extent_backref()).
2051 spin_lock(&locked_ref->lock);
2052 btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2054 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref);
2055 if (ret < 0 && ret != -EAGAIN) {
2057 * Error, btrfs_run_delayed_refs_for_head already
2058 * unlocked everything so just bail out
2063 * Success, perform the usual cleanup of a processed
2066 ret = cleanup_ref_head(trans, locked_ref);
2068 /* We dropped our lock, we need to loop. */
2077 * Either success case or btrfs_run_delayed_refs_for_head
2078 * returned -EAGAIN, meaning we need to select another head
2083 } while ((nr != -1 && count < nr) || locked_ref);
2088 #ifdef SCRAMBLE_DELAYED_REFS
2090 * Normally delayed refs get processed in ascending bytenr order. This
2091 * correlates in most cases to the order added. To expose dependencies on this
2092 * order, we start to process the tree in the middle instead of the beginning
2094 static u64 find_middle(struct rb_root *root)
2096 struct rb_node *n = root->rb_node;
2097 struct btrfs_delayed_ref_node *entry;
2100 u64 first = 0, last = 0;
2104 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2105 first = entry->bytenr;
2109 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2110 last = entry->bytenr;
2115 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2116 WARN_ON(!entry->in_tree);
2118 middle = entry->bytenr;
2132 * this starts processing the delayed reference count updates and
2133 * extent insertions we have queued up so far. count can be
2134 * 0, which means to process everything in the tree at the start
2135 * of the run (but not newly added entries), or it can be some target
2136 * number you'd like to process.
2138 * Returns 0 on success or if called with an aborted transaction
2139 * Returns <0 on error and aborts the transaction
2141 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2142 unsigned long count)
2144 struct btrfs_fs_info *fs_info = trans->fs_info;
2145 struct rb_node *node;
2146 struct btrfs_delayed_ref_root *delayed_refs;
2147 struct btrfs_delayed_ref_head *head;
2149 int run_all = count == (unsigned long)-1;
2151 /* We'll clean this up in btrfs_cleanup_transaction */
2152 if (TRANS_ABORTED(trans))
2155 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2158 delayed_refs = &trans->transaction->delayed_refs;
2160 count = delayed_refs->num_heads_ready;
2163 #ifdef SCRAMBLE_DELAYED_REFS
2164 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2166 ret = __btrfs_run_delayed_refs(trans, count);
2168 btrfs_abort_transaction(trans, ret);
2173 btrfs_create_pending_block_groups(trans);
2175 spin_lock(&delayed_refs->lock);
2176 node = rb_first_cached(&delayed_refs->href_root);
2178 spin_unlock(&delayed_refs->lock);
2181 head = rb_entry(node, struct btrfs_delayed_ref_head,
2183 refcount_inc(&head->refs);
2184 spin_unlock(&delayed_refs->lock);
2186 /* Mutex was contended, block until it's released and retry. */
2187 mutex_lock(&head->mutex);
2188 mutex_unlock(&head->mutex);
2190 btrfs_put_delayed_ref_head(head);
2198 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2199 struct extent_buffer *eb, u64 flags)
2201 struct btrfs_delayed_extent_op *extent_op;
2202 int level = btrfs_header_level(eb);
2205 extent_op = btrfs_alloc_delayed_extent_op();
2209 extent_op->flags_to_set = flags;
2210 extent_op->update_flags = true;
2211 extent_op->update_key = false;
2212 extent_op->level = level;
2214 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2216 btrfs_free_delayed_extent_op(extent_op);
2220 static noinline int check_delayed_ref(struct btrfs_root *root,
2221 struct btrfs_path *path,
2222 u64 objectid, u64 offset, u64 bytenr)
2224 struct btrfs_delayed_ref_head *head;
2225 struct btrfs_delayed_ref_node *ref;
2226 struct btrfs_delayed_data_ref *data_ref;
2227 struct btrfs_delayed_ref_root *delayed_refs;
2228 struct btrfs_transaction *cur_trans;
2229 struct rb_node *node;
2232 spin_lock(&root->fs_info->trans_lock);
2233 cur_trans = root->fs_info->running_transaction;
2235 refcount_inc(&cur_trans->use_count);
2236 spin_unlock(&root->fs_info->trans_lock);
2240 delayed_refs = &cur_trans->delayed_refs;
2241 spin_lock(&delayed_refs->lock);
2242 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2244 spin_unlock(&delayed_refs->lock);
2245 btrfs_put_transaction(cur_trans);
2249 if (!mutex_trylock(&head->mutex)) {
2251 spin_unlock(&delayed_refs->lock);
2252 btrfs_put_transaction(cur_trans);
2256 refcount_inc(&head->refs);
2257 spin_unlock(&delayed_refs->lock);
2259 btrfs_release_path(path);
2262 * Mutex was contended, block until it's released and let
2265 mutex_lock(&head->mutex);
2266 mutex_unlock(&head->mutex);
2267 btrfs_put_delayed_ref_head(head);
2268 btrfs_put_transaction(cur_trans);
2271 spin_unlock(&delayed_refs->lock);
2273 spin_lock(&head->lock);
2275 * XXX: We should replace this with a proper search function in the
2278 for (node = rb_first_cached(&head->ref_tree); node;
2279 node = rb_next(node)) {
2280 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2281 /* If it's a shared ref we know a cross reference exists */
2282 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2287 data_ref = btrfs_delayed_node_to_data_ref(ref);
2290 * If our ref doesn't match the one we're currently looking at
2291 * then we have a cross reference.
2293 if (data_ref->root != root->root_key.objectid ||
2294 data_ref->objectid != objectid ||
2295 data_ref->offset != offset) {
2300 spin_unlock(&head->lock);
2301 mutex_unlock(&head->mutex);
2302 btrfs_put_transaction(cur_trans);
2306 static noinline int check_committed_ref(struct btrfs_root *root,
2307 struct btrfs_path *path,
2308 u64 objectid, u64 offset, u64 bytenr,
2311 struct btrfs_fs_info *fs_info = root->fs_info;
2312 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2313 struct extent_buffer *leaf;
2314 struct btrfs_extent_data_ref *ref;
2315 struct btrfs_extent_inline_ref *iref;
2316 struct btrfs_extent_item *ei;
2317 struct btrfs_key key;
2322 key.objectid = bytenr;
2323 key.offset = (u64)-1;
2324 key.type = BTRFS_EXTENT_ITEM_KEY;
2326 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2329 BUG_ON(ret == 0); /* Corruption */
2332 if (path->slots[0] == 0)
2336 leaf = path->nodes[0];
2337 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2339 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2343 item_size = btrfs_item_size(leaf, path->slots[0]);
2344 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2346 /* If extent item has more than 1 inline ref then it's shared */
2347 if (item_size != sizeof(*ei) +
2348 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2352 * If extent created before last snapshot => it's shared unless the
2353 * snapshot has been deleted. Use the heuristic if strict is false.
2356 (btrfs_extent_generation(leaf, ei) <=
2357 btrfs_root_last_snapshot(&root->root_item)))
2360 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2362 /* If this extent has SHARED_DATA_REF then it's shared */
2363 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2364 if (type != BTRFS_EXTENT_DATA_REF_KEY)
2367 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2368 if (btrfs_extent_refs(leaf, ei) !=
2369 btrfs_extent_data_ref_count(leaf, ref) ||
2370 btrfs_extent_data_ref_root(leaf, ref) !=
2371 root->root_key.objectid ||
2372 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2373 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2381 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2382 u64 bytenr, bool strict, struct btrfs_path *path)
2387 ret = check_committed_ref(root, path, objectid,
2388 offset, bytenr, strict);
2389 if (ret && ret != -ENOENT)
2392 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2393 } while (ret == -EAGAIN);
2396 btrfs_release_path(path);
2397 if (btrfs_is_data_reloc_root(root))
2402 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2403 struct btrfs_root *root,
2404 struct extent_buffer *buf,
2405 int full_backref, int inc)
2407 struct btrfs_fs_info *fs_info = root->fs_info;
2413 struct btrfs_key key;
2414 struct btrfs_file_extent_item *fi;
2415 struct btrfs_ref generic_ref = { 0 };
2416 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2422 if (btrfs_is_testing(fs_info))
2425 ref_root = btrfs_header_owner(buf);
2426 nritems = btrfs_header_nritems(buf);
2427 level = btrfs_header_level(buf);
2429 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2433 parent = buf->start;
2437 action = BTRFS_ADD_DELAYED_REF;
2439 action = BTRFS_DROP_DELAYED_REF;
2441 for (i = 0; i < nritems; i++) {
2443 btrfs_item_key_to_cpu(buf, &key, i);
2444 if (key.type != BTRFS_EXTENT_DATA_KEY)
2446 fi = btrfs_item_ptr(buf, i,
2447 struct btrfs_file_extent_item);
2448 if (btrfs_file_extent_type(buf, fi) ==
2449 BTRFS_FILE_EXTENT_INLINE)
2451 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2455 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2456 key.offset -= btrfs_file_extent_offset(buf, fi);
2457 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2459 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2460 key.offset, root->root_key.objectid,
2463 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2465 ret = btrfs_free_extent(trans, &generic_ref);
2469 bytenr = btrfs_node_blockptr(buf, i);
2470 num_bytes = fs_info->nodesize;
2471 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2473 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root,
2474 root->root_key.objectid, for_reloc);
2476 ret = btrfs_inc_extent_ref(trans, &generic_ref);
2478 ret = btrfs_free_extent(trans, &generic_ref);
2488 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2489 struct extent_buffer *buf, int full_backref)
2491 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2494 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2495 struct extent_buffer *buf, int full_backref)
2497 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2500 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2502 struct btrfs_fs_info *fs_info = root->fs_info;
2507 flags = BTRFS_BLOCK_GROUP_DATA;
2508 else if (root == fs_info->chunk_root)
2509 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2511 flags = BTRFS_BLOCK_GROUP_METADATA;
2513 ret = btrfs_get_alloc_profile(fs_info, flags);
2517 static u64 first_logical_byte(struct btrfs_fs_info *fs_info)
2519 struct rb_node *leftmost;
2522 read_lock(&fs_info->block_group_cache_lock);
2523 /* Get the block group with the lowest logical start address. */
2524 leftmost = rb_first_cached(&fs_info->block_group_cache_tree);
2526 struct btrfs_block_group *bg;
2528 bg = rb_entry(leftmost, struct btrfs_block_group, cache_node);
2531 read_unlock(&fs_info->block_group_cache_lock);
2536 static int pin_down_extent(struct btrfs_trans_handle *trans,
2537 struct btrfs_block_group *cache,
2538 u64 bytenr, u64 num_bytes, int reserved)
2540 struct btrfs_fs_info *fs_info = cache->fs_info;
2542 spin_lock(&cache->space_info->lock);
2543 spin_lock(&cache->lock);
2544 cache->pinned += num_bytes;
2545 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2548 cache->reserved -= num_bytes;
2549 cache->space_info->bytes_reserved -= num_bytes;
2551 spin_unlock(&cache->lock);
2552 spin_unlock(&cache->space_info->lock);
2554 set_extent_bit(&trans->transaction->pinned_extents, bytenr,
2555 bytenr + num_bytes - 1, EXTENT_DIRTY, NULL);
2559 int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2560 u64 bytenr, u64 num_bytes, int reserved)
2562 struct btrfs_block_group *cache;
2564 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2565 BUG_ON(!cache); /* Logic error */
2567 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2569 btrfs_put_block_group(cache);
2574 * this function must be called within transaction
2576 int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2577 u64 bytenr, u64 num_bytes)
2579 struct btrfs_block_group *cache;
2582 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2587 * Fully cache the free space first so that our pin removes the free space
2590 ret = btrfs_cache_block_group(cache, true);
2594 pin_down_extent(trans, cache, bytenr, num_bytes, 0);
2596 /* remove us from the free space cache (if we're there at all) */
2597 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2599 btrfs_put_block_group(cache);
2603 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2604 u64 start, u64 num_bytes)
2607 struct btrfs_block_group *block_group;
2609 block_group = btrfs_lookup_block_group(fs_info, start);
2613 ret = btrfs_cache_block_group(block_group, true);
2617 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2619 btrfs_put_block_group(block_group);
2623 int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2625 struct btrfs_fs_info *fs_info = eb->fs_info;
2626 struct btrfs_file_extent_item *item;
2627 struct btrfs_key key;
2632 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2635 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2636 btrfs_item_key_to_cpu(eb, &key, i);
2637 if (key.type != BTRFS_EXTENT_DATA_KEY)
2639 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2640 found_type = btrfs_file_extent_type(eb, item);
2641 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2643 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2645 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2646 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2647 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2656 btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2658 atomic_inc(&bg->reservations);
2662 * Returns the free cluster for the given space info and sets empty_cluster to
2663 * what it should be based on the mount options.
2665 static struct btrfs_free_cluster *
2666 fetch_cluster_info(struct btrfs_fs_info *fs_info,
2667 struct btrfs_space_info *space_info, u64 *empty_cluster)
2669 struct btrfs_free_cluster *ret = NULL;
2672 if (btrfs_mixed_space_info(space_info))
2675 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2676 ret = &fs_info->meta_alloc_cluster;
2677 if (btrfs_test_opt(fs_info, SSD))
2678 *empty_cluster = SZ_2M;
2680 *empty_cluster = SZ_64K;
2681 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2682 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2683 *empty_cluster = SZ_2M;
2684 ret = &fs_info->data_alloc_cluster;
2690 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2692 const bool return_free_space)
2694 struct btrfs_block_group *cache = NULL;
2695 struct btrfs_space_info *space_info;
2696 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2697 struct btrfs_free_cluster *cluster = NULL;
2699 u64 total_unpinned = 0;
2700 u64 empty_cluster = 0;
2703 while (start <= end) {
2706 start >= cache->start + cache->length) {
2708 btrfs_put_block_group(cache);
2710 cache = btrfs_lookup_block_group(fs_info, start);
2711 BUG_ON(!cache); /* Logic error */
2713 cluster = fetch_cluster_info(fs_info,
2716 empty_cluster <<= 1;
2719 len = cache->start + cache->length - start;
2720 len = min(len, end + 1 - start);
2722 if (return_free_space)
2723 btrfs_add_free_space(cache, start, len);
2726 total_unpinned += len;
2727 space_info = cache->space_info;
2730 * If this space cluster has been marked as fragmented and we've
2731 * unpinned enough in this block group to potentially allow a
2732 * cluster to be created inside of it go ahead and clear the
2735 if (cluster && cluster->fragmented &&
2736 total_unpinned > empty_cluster) {
2737 spin_lock(&cluster->lock);
2738 cluster->fragmented = 0;
2739 spin_unlock(&cluster->lock);
2742 spin_lock(&space_info->lock);
2743 spin_lock(&cache->lock);
2744 cache->pinned -= len;
2745 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2746 space_info->max_extent_size = 0;
2748 space_info->bytes_readonly += len;
2750 } else if (btrfs_is_zoned(fs_info)) {
2751 /* Need reset before reusing in a zoned block group */
2752 space_info->bytes_zone_unusable += len;
2755 spin_unlock(&cache->lock);
2756 if (!readonly && return_free_space &&
2757 global_rsv->space_info == space_info) {
2758 spin_lock(&global_rsv->lock);
2759 if (!global_rsv->full) {
2760 u64 to_add = min(len, global_rsv->size -
2761 global_rsv->reserved);
2763 global_rsv->reserved += to_add;
2764 btrfs_space_info_update_bytes_may_use(fs_info,
2765 space_info, to_add);
2766 if (global_rsv->reserved >= global_rsv->size)
2767 global_rsv->full = 1;
2770 spin_unlock(&global_rsv->lock);
2772 /* Add to any tickets we may have */
2773 if (!readonly && return_free_space && len)
2774 btrfs_try_granting_tickets(fs_info, space_info);
2775 spin_unlock(&space_info->lock);
2779 btrfs_put_block_group(cache);
2783 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2785 struct btrfs_fs_info *fs_info = trans->fs_info;
2786 struct btrfs_block_group *block_group, *tmp;
2787 struct list_head *deleted_bgs;
2788 struct extent_io_tree *unpin;
2793 unpin = &trans->transaction->pinned_extents;
2795 while (!TRANS_ABORTED(trans)) {
2796 struct extent_state *cached_state = NULL;
2798 mutex_lock(&fs_info->unused_bg_unpin_mutex);
2799 if (!find_first_extent_bit(unpin, 0, &start, &end,
2800 EXTENT_DIRTY, &cached_state)) {
2801 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2805 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2806 ret = btrfs_discard_extent(fs_info, start,
2807 end + 1 - start, NULL);
2809 clear_extent_dirty(unpin, start, end, &cached_state);
2810 unpin_extent_range(fs_info, start, end, true);
2811 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2812 free_extent_state(cached_state);
2816 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2817 btrfs_discard_calc_delay(&fs_info->discard_ctl);
2818 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2822 * Transaction is finished. We don't need the lock anymore. We
2823 * do need to clean up the block groups in case of a transaction
2826 deleted_bgs = &trans->transaction->deleted_bgs;
2827 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2831 if (!TRANS_ABORTED(trans))
2832 ret = btrfs_discard_extent(fs_info,
2834 block_group->length,
2837 list_del_init(&block_group->bg_list);
2838 btrfs_unfreeze_block_group(block_group);
2839 btrfs_put_block_group(block_group);
2842 const char *errstr = btrfs_decode_error(ret);
2844 "discard failed while removing blockgroup: errno=%d %s",
2852 static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
2853 u64 bytenr, u64 num_bytes, bool is_data)
2858 struct btrfs_root *csum_root;
2860 csum_root = btrfs_csum_root(trans->fs_info, bytenr);
2861 ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes);
2863 btrfs_abort_transaction(trans, ret);
2868 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
2870 btrfs_abort_transaction(trans, ret);
2874 ret = btrfs_update_block_group(trans, bytenr, num_bytes, false);
2876 btrfs_abort_transaction(trans, ret);
2881 #define abort_and_dump(trans, path, fmt, args...) \
2883 btrfs_abort_transaction(trans, -EUCLEAN); \
2884 btrfs_print_leaf(path->nodes[0]); \
2885 btrfs_crit(trans->fs_info, fmt, ##args); \
2889 * Drop one or more refs of @node.
2891 * 1. Locate the extent refs.
2892 * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2893 * Locate it, then reduce the refs number or remove the ref line completely.
2895 * 2. Update the refs count in EXTENT/METADATA_ITEM
2897 * Inline backref case:
2899 * in extent tree we have:
2901 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2902 * refs 2 gen 6 flags DATA
2903 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2904 * extent data backref root FS_TREE objectid 257 offset 0 count 1
2906 * This function gets called with:
2908 * node->bytenr = 13631488
2909 * node->num_bytes = 1048576
2910 * root_objectid = FS_TREE
2911 * owner_objectid = 257
2915 * Then we should get some like:
2917 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2918 * refs 1 gen 6 flags DATA
2919 * extent data backref root FS_TREE objectid 258 offset 0 count 1
2921 * Keyed backref case:
2923 * in extent tree we have:
2925 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2926 * refs 754 gen 6 flags DATA
2928 * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2929 * extent data backref root FS_TREE objectid 866 offset 0 count 1
2931 * This function get called with:
2933 * node->bytenr = 13631488
2934 * node->num_bytes = 1048576
2935 * root_objectid = FS_TREE
2936 * owner_objectid = 866
2940 * Then we should get some like:
2942 * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2943 * refs 753 gen 6 flags DATA
2945 * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2947 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2948 struct btrfs_delayed_ref_node *node, u64 parent,
2949 u64 root_objectid, u64 owner_objectid,
2950 u64 owner_offset, int refs_to_drop,
2951 struct btrfs_delayed_extent_op *extent_op)
2953 struct btrfs_fs_info *info = trans->fs_info;
2954 struct btrfs_key key;
2955 struct btrfs_path *path;
2956 struct btrfs_root *extent_root;
2957 struct extent_buffer *leaf;
2958 struct btrfs_extent_item *ei;
2959 struct btrfs_extent_inline_ref *iref;
2962 int extent_slot = 0;
2963 int found_extent = 0;
2967 u64 bytenr = node->bytenr;
2968 u64 num_bytes = node->num_bytes;
2969 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2971 extent_root = btrfs_extent_root(info, bytenr);
2972 ASSERT(extent_root);
2974 path = btrfs_alloc_path();
2978 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2980 if (!is_data && refs_to_drop != 1) {
2982 "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2983 node->bytenr, refs_to_drop);
2985 btrfs_abort_transaction(trans, ret);
2990 skinny_metadata = false;
2992 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2993 parent, root_objectid, owner_objectid,
2997 * Either the inline backref or the SHARED_DATA_REF/
2998 * SHARED_BLOCK_REF is found
3000 * Here is a quick path to locate EXTENT/METADATA_ITEM.
3001 * It's possible the EXTENT/METADATA_ITEM is near current slot.
3003 extent_slot = path->slots[0];
3004 while (extent_slot >= 0) {
3005 btrfs_item_key_to_cpu(path->nodes[0], &key,
3007 if (key.objectid != bytenr)
3009 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3010 key.offset == num_bytes) {
3014 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3015 key.offset == owner_objectid) {
3020 /* Quick path didn't find the EXTEMT/METADATA_ITEM */
3021 if (path->slots[0] - extent_slot > 5)
3026 if (!found_extent) {
3028 abort_and_dump(trans, path,
3029 "invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref",
3034 /* Must be SHARED_* item, remove the backref first */
3035 ret = remove_extent_backref(trans, extent_root, path,
3036 NULL, refs_to_drop, is_data);
3038 btrfs_abort_transaction(trans, ret);
3041 btrfs_release_path(path);
3043 /* Slow path to locate EXTENT/METADATA_ITEM */
3044 key.objectid = bytenr;
3045 key.type = BTRFS_EXTENT_ITEM_KEY;
3046 key.offset = num_bytes;
3048 if (!is_data && skinny_metadata) {
3049 key.type = BTRFS_METADATA_ITEM_KEY;
3050 key.offset = owner_objectid;
3053 ret = btrfs_search_slot(trans, extent_root,
3055 if (ret > 0 && skinny_metadata && path->slots[0]) {
3057 * Couldn't find our skinny metadata item,
3058 * see if we have ye olde extent item.
3061 btrfs_item_key_to_cpu(path->nodes[0], &key,
3063 if (key.objectid == bytenr &&
3064 key.type == BTRFS_EXTENT_ITEM_KEY &&
3065 key.offset == num_bytes)
3069 if (ret > 0 && skinny_metadata) {
3070 skinny_metadata = false;
3071 key.objectid = bytenr;
3072 key.type = BTRFS_EXTENT_ITEM_KEY;
3073 key.offset = num_bytes;
3074 btrfs_release_path(path);
3075 ret = btrfs_search_slot(trans, extent_root,
3081 btrfs_print_leaf(path->nodes[0]);
3083 "umm, got %d back from search, was looking for %llu, slot %d",
3084 ret, bytenr, path->slots[0]);
3087 btrfs_abort_transaction(trans, ret);
3090 extent_slot = path->slots[0];
3092 } else if (WARN_ON(ret == -ENOENT)) {
3093 abort_and_dump(trans, path,
3094 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d",
3095 bytenr, parent, root_objectid, owner_objectid,
3096 owner_offset, path->slots[0]);
3099 btrfs_abort_transaction(trans, ret);
3103 leaf = path->nodes[0];
3104 item_size = btrfs_item_size(leaf, extent_slot);
3105 if (unlikely(item_size < sizeof(*ei))) {
3107 btrfs_err(trans->fs_info,
3108 "unexpected extent item size, has %u expect >= %zu",
3109 item_size, sizeof(*ei));
3110 btrfs_abort_transaction(trans, ret);
3113 ei = btrfs_item_ptr(leaf, extent_slot,
3114 struct btrfs_extent_item);
3115 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3116 key.type == BTRFS_EXTENT_ITEM_KEY) {
3117 struct btrfs_tree_block_info *bi;
3119 if (item_size < sizeof(*ei) + sizeof(*bi)) {
3120 abort_and_dump(trans, path,
3121 "invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu",
3122 key.objectid, key.type, key.offset,
3123 path->slots[0], owner_objectid, item_size,
3124 sizeof(*ei) + sizeof(*bi));
3128 bi = (struct btrfs_tree_block_info *)(ei + 1);
3129 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3132 refs = btrfs_extent_refs(leaf, ei);
3133 if (refs < refs_to_drop) {
3134 abort_and_dump(trans, path,
3135 "trying to drop %d refs but we only have %llu for bytenr %llu slot %u",
3136 refs_to_drop, refs, bytenr, path->slots[0]);
3140 refs -= refs_to_drop;
3144 __run_delayed_extent_op(extent_op, leaf, ei);
3146 * In the case of inline back ref, reference count will
3147 * be updated by remove_extent_backref
3150 if (!found_extent) {
3151 abort_and_dump(trans, path,
3152 "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u",
3158 btrfs_set_extent_refs(leaf, ei, refs);
3159 btrfs_mark_buffer_dirty(trans, leaf);
3162 ret = remove_extent_backref(trans, extent_root, path,
3163 iref, refs_to_drop, is_data);
3165 btrfs_abort_transaction(trans, ret);
3170 /* In this branch refs == 1 */
3172 if (is_data && refs_to_drop !=
3173 extent_data_ref_count(path, iref)) {
3174 abort_and_dump(trans, path,
3175 "invalid refs_to_drop, current refs %u refs_to_drop %u slot %u",
3176 extent_data_ref_count(path, iref),
3177 refs_to_drop, path->slots[0]);
3182 if (path->slots[0] != extent_slot) {
3183 abort_and_dump(trans, path,
3184 "invalid iref, extent item key (%llu %u %llu) slot %u doesn't have wanted iref",
3185 key.objectid, key.type,
3186 key.offset, path->slots[0]);
3192 * No inline ref, we must be at SHARED_* item,
3193 * And it's single ref, it must be:
3194 * | extent_slot ||extent_slot + 1|
3195 * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3197 if (path->slots[0] != extent_slot + 1) {
3198 abort_and_dump(trans, path,
3199 "invalid SHARED_* item slot %u, previous item is not EXTENT/METADATA_ITEM",
3204 path->slots[0] = extent_slot;
3209 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3212 btrfs_abort_transaction(trans, ret);
3215 btrfs_release_path(path);
3217 ret = do_free_extent_accounting(trans, bytenr, num_bytes, is_data);
3219 btrfs_release_path(path);
3222 btrfs_free_path(path);
3227 * when we free an block, it is possible (and likely) that we free the last
3228 * delayed ref for that extent as well. This searches the delayed ref tree for
3229 * a given extent, and if there are no other delayed refs to be processed, it
3230 * removes it from the tree.
3232 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3235 struct btrfs_delayed_ref_head *head;
3236 struct btrfs_delayed_ref_root *delayed_refs;
3239 delayed_refs = &trans->transaction->delayed_refs;
3240 spin_lock(&delayed_refs->lock);
3241 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3243 goto out_delayed_unlock;
3245 spin_lock(&head->lock);
3246 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3249 if (cleanup_extent_op(head) != NULL)
3253 * waiting for the lock here would deadlock. If someone else has it
3254 * locked they are already in the process of dropping it anyway
3256 if (!mutex_trylock(&head->mutex))
3259 btrfs_delete_ref_head(delayed_refs, head);
3260 head->processing = false;
3262 spin_unlock(&head->lock);
3263 spin_unlock(&delayed_refs->lock);
3265 BUG_ON(head->extent_op);
3266 if (head->must_insert_reserved)
3269 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3270 mutex_unlock(&head->mutex);
3271 btrfs_put_delayed_ref_head(head);
3274 spin_unlock(&head->lock);
3277 spin_unlock(&delayed_refs->lock);
3281 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3283 struct extent_buffer *buf,
3284 u64 parent, int last_ref)
3286 struct btrfs_fs_info *fs_info = trans->fs_info;
3287 struct btrfs_ref generic_ref = { 0 };
3290 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3291 buf->start, buf->len, parent);
3292 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3295 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3296 btrfs_ref_tree_mod(fs_info, &generic_ref);
3297 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3298 BUG_ON(ret); /* -ENOMEM */
3301 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3302 struct btrfs_block_group *cache;
3303 bool must_pin = false;
3305 if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3306 ret = check_ref_cleanup(trans, buf->start);
3308 btrfs_redirty_list_add(trans->transaction, buf);
3313 cache = btrfs_lookup_block_group(fs_info, buf->start);
3315 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3316 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3317 btrfs_put_block_group(cache);
3322 * If there are tree mod log users we may have recorded mod log
3323 * operations for this node. If we re-allocate this node we
3324 * could replay operations on this node that happened when it
3325 * existed in a completely different root. For example if it
3326 * was part of root A, then was reallocated to root B, and we
3327 * are doing a btrfs_old_search_slot(root b), we could replay
3328 * operations that happened when the block was part of root A,
3329 * giving us an inconsistent view of the btree.
3331 * We are safe from races here because at this point no other
3332 * node or root points to this extent buffer, so if after this
3333 * check a new tree mod log user joins we will not have an
3334 * existing log of operations on this node that we have to
3337 if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
3340 if (must_pin || btrfs_is_zoned(fs_info)) {
3341 btrfs_redirty_list_add(trans->transaction, buf);
3342 pin_down_extent(trans, cache, buf->start, buf->len, 1);
3343 btrfs_put_block_group(cache);
3347 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3349 btrfs_add_free_space(cache, buf->start, buf->len);
3350 btrfs_free_reserved_bytes(cache, buf->len, 0);
3351 btrfs_put_block_group(cache);
3352 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3357 * Deleting the buffer, clear the corrupt flag since it doesn't
3360 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3364 /* Can return -ENOMEM */
3365 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3367 struct btrfs_fs_info *fs_info = trans->fs_info;
3370 if (btrfs_is_testing(fs_info))
3374 * tree log blocks never actually go into the extent allocation
3375 * tree, just update pinning info and exit early.
3377 if ((ref->type == BTRFS_REF_METADATA &&
3378 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3379 (ref->type == BTRFS_REF_DATA &&
3380 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)) {
3381 /* unlocks the pinned mutex */
3382 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3384 } else if (ref->type == BTRFS_REF_METADATA) {
3385 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3387 ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3390 if (!((ref->type == BTRFS_REF_METADATA &&
3391 ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3392 (ref->type == BTRFS_REF_DATA &&
3393 ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)))
3394 btrfs_ref_tree_mod(fs_info, ref);
3399 enum btrfs_loop_type {
3401 * Start caching block groups but do not wait for progress or for them
3404 LOOP_CACHING_NOWAIT,
3407 * Wait for the block group free_space >= the space we're waiting for if
3408 * the block group isn't cached.
3413 * Allow allocations to happen from block groups that do not yet have a
3414 * size classification.
3416 LOOP_UNSET_SIZE_CLASS,
3419 * Allocate a chunk and then retry the allocation.
3424 * Ignore the size class restrictions for this allocation.
3426 LOOP_WRONG_SIZE_CLASS,
3429 * Ignore the empty size, only try to allocate the number of bytes
3430 * needed for this allocation.
3436 btrfs_lock_block_group(struct btrfs_block_group *cache,
3440 down_read(&cache->data_rwsem);
3443 static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3446 btrfs_get_block_group(cache);
3448 down_read(&cache->data_rwsem);
3451 static struct btrfs_block_group *btrfs_lock_cluster(
3452 struct btrfs_block_group *block_group,
3453 struct btrfs_free_cluster *cluster,
3455 __acquires(&cluster->refill_lock)
3457 struct btrfs_block_group *used_bg = NULL;
3459 spin_lock(&cluster->refill_lock);
3461 used_bg = cluster->block_group;
3465 if (used_bg == block_group)
3468 btrfs_get_block_group(used_bg);
3473 if (down_read_trylock(&used_bg->data_rwsem))
3476 spin_unlock(&cluster->refill_lock);
3478 /* We should only have one-level nested. */
3479 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3481 spin_lock(&cluster->refill_lock);
3482 if (used_bg == cluster->block_group)
3485 up_read(&used_bg->data_rwsem);
3486 btrfs_put_block_group(used_bg);
3491 btrfs_release_block_group(struct btrfs_block_group *cache,
3495 up_read(&cache->data_rwsem);
3496 btrfs_put_block_group(cache);
3500 * Helper function for find_free_extent().
3502 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3503 * Return >0 to inform caller that we find nothing
3504 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3506 static int find_free_extent_clustered(struct btrfs_block_group *bg,
3507 struct find_free_extent_ctl *ffe_ctl,
3508 struct btrfs_block_group **cluster_bg_ret)
3510 struct btrfs_block_group *cluster_bg;
3511 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3512 u64 aligned_cluster;
3516 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3518 goto refill_cluster;
3519 if (cluster_bg != bg && (cluster_bg->ro ||
3520 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3521 goto release_cluster;
3523 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3524 ffe_ctl->num_bytes, cluster_bg->start,
3525 &ffe_ctl->max_extent_size);
3527 /* We have a block, we're done */
3528 spin_unlock(&last_ptr->refill_lock);
3529 trace_btrfs_reserve_extent_cluster(cluster_bg, ffe_ctl);
3530 *cluster_bg_ret = cluster_bg;
3531 ffe_ctl->found_offset = offset;
3534 WARN_ON(last_ptr->block_group != cluster_bg);
3538 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3539 * lets just skip it and let the allocator find whatever block it can
3540 * find. If we reach this point, we will have tried the cluster
3541 * allocator plenty of times and not have found anything, so we are
3542 * likely way too fragmented for the clustering stuff to find anything.
3544 * However, if the cluster is taken from the current block group,
3545 * release the cluster first, so that we stand a better chance of
3546 * succeeding in the unclustered allocation.
3548 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3549 spin_unlock(&last_ptr->refill_lock);
3550 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3554 /* This cluster didn't work out, free it and start over */
3555 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3557 if (cluster_bg != bg)
3558 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3561 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3562 spin_unlock(&last_ptr->refill_lock);
3566 aligned_cluster = max_t(u64,
3567 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3568 bg->full_stripe_len);
3569 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3570 ffe_ctl->num_bytes, aligned_cluster);
3572 /* Now pull our allocation out of this cluster */
3573 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3574 ffe_ctl->num_bytes, ffe_ctl->search_start,
3575 &ffe_ctl->max_extent_size);
3577 /* We found one, proceed */
3578 spin_unlock(&last_ptr->refill_lock);
3579 ffe_ctl->found_offset = offset;
3580 trace_btrfs_reserve_extent_cluster(bg, ffe_ctl);
3585 * At this point we either didn't find a cluster or we weren't able to
3586 * allocate a block from our cluster. Free the cluster we've been
3587 * trying to use, and go to the next block group.
3589 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3590 spin_unlock(&last_ptr->refill_lock);
3595 * Return >0 to inform caller that we find nothing
3596 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3598 static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3599 struct find_free_extent_ctl *ffe_ctl)
3601 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3605 * We are doing an unclustered allocation, set the fragmented flag so
3606 * we don't bother trying to setup a cluster again until we get more
3609 if (unlikely(last_ptr)) {
3610 spin_lock(&last_ptr->lock);
3611 last_ptr->fragmented = 1;
3612 spin_unlock(&last_ptr->lock);
3614 if (ffe_ctl->cached) {
3615 struct btrfs_free_space_ctl *free_space_ctl;
3617 free_space_ctl = bg->free_space_ctl;
3618 spin_lock(&free_space_ctl->tree_lock);
3619 if (free_space_ctl->free_space <
3620 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3621 ffe_ctl->empty_size) {
3622 ffe_ctl->total_free_space = max_t(u64,
3623 ffe_ctl->total_free_space,
3624 free_space_ctl->free_space);
3625 spin_unlock(&free_space_ctl->tree_lock);
3628 spin_unlock(&free_space_ctl->tree_lock);
3631 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3632 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3633 &ffe_ctl->max_extent_size);
3636 ffe_ctl->found_offset = offset;
3640 static int do_allocation_clustered(struct btrfs_block_group *block_group,
3641 struct find_free_extent_ctl *ffe_ctl,
3642 struct btrfs_block_group **bg_ret)
3646 /* We want to try and use the cluster allocator, so lets look there */
3647 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3648 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3651 /* ret == -ENOENT case falls through */
3654 return find_free_extent_unclustered(block_group, ffe_ctl);
3658 * Tree-log block group locking
3659 * ============================
3661 * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3662 * indicates the starting address of a block group, which is reserved only
3663 * for tree-log metadata.
3670 * fs_info::treelog_bg_lock
3674 * Simple allocator for sequential-only block group. It only allows sequential
3675 * allocation. No need to play with trees. This function also reserves the
3676 * bytes as in btrfs_add_reserved_bytes.
3678 static int do_allocation_zoned(struct btrfs_block_group *block_group,
3679 struct find_free_extent_ctl *ffe_ctl,
3680 struct btrfs_block_group **bg_ret)
3682 struct btrfs_fs_info *fs_info = block_group->fs_info;
3683 struct btrfs_space_info *space_info = block_group->space_info;
3684 struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3685 u64 start = block_group->start;
3686 u64 num_bytes = ffe_ctl->num_bytes;
3688 u64 bytenr = block_group->start;
3690 u64 data_reloc_bytenr;
3694 ASSERT(btrfs_is_zoned(block_group->fs_info));
3697 * Do not allow non-tree-log blocks in the dedicated tree-log block
3698 * group, and vice versa.
3700 spin_lock(&fs_info->treelog_bg_lock);
3701 log_bytenr = fs_info->treelog_bg;
3702 if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3703 (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3705 spin_unlock(&fs_info->treelog_bg_lock);
3710 * Do not allow non-relocation blocks in the dedicated relocation block
3711 * group, and vice versa.
3713 spin_lock(&fs_info->relocation_bg_lock);
3714 data_reloc_bytenr = fs_info->data_reloc_bg;
3715 if (data_reloc_bytenr &&
3716 ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3717 (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3719 spin_unlock(&fs_info->relocation_bg_lock);
3723 /* Check RO and no space case before trying to activate it */
3724 spin_lock(&block_group->lock);
3725 if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) {
3728 * May need to clear fs_info->{treelog,data_reloc}_bg.
3729 * Return the error after taking the locks.
3732 spin_unlock(&block_group->lock);
3734 /* Metadata block group is activated at write time. */
3735 if (!ret && (block_group->flags & BTRFS_BLOCK_GROUP_DATA) &&
3736 !btrfs_zone_activate(block_group)) {
3739 * May need to clear fs_info->{treelog,data_reloc}_bg.
3740 * Return the error after taking the locks.
3744 spin_lock(&space_info->lock);
3745 spin_lock(&block_group->lock);
3746 spin_lock(&fs_info->treelog_bg_lock);
3747 spin_lock(&fs_info->relocation_bg_lock);
3752 ASSERT(!ffe_ctl->for_treelog ||
3753 block_group->start == fs_info->treelog_bg ||
3754 fs_info->treelog_bg == 0);
3755 ASSERT(!ffe_ctl->for_data_reloc ||
3756 block_group->start == fs_info->data_reloc_bg ||
3757 fs_info->data_reloc_bg == 0);
3759 if (block_group->ro ||
3760 (!ffe_ctl->for_data_reloc &&
3761 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))) {
3767 * Do not allow currently using block group to be tree-log dedicated
3770 if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3771 (block_group->used || block_group->reserved)) {
3777 * Do not allow currently used block group to be the data relocation
3778 * dedicated block group.
3780 if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3781 (block_group->used || block_group->reserved)) {
3786 WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3787 avail = block_group->zone_capacity - block_group->alloc_offset;
3788 if (avail < num_bytes) {
3789 if (ffe_ctl->max_extent_size < avail) {
3791 * With sequential allocator, free space is always
3794 ffe_ctl->max_extent_size = avail;
3795 ffe_ctl->total_free_space = avail;
3801 if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3802 fs_info->treelog_bg = block_group->start;
3804 if (ffe_ctl->for_data_reloc) {
3805 if (!fs_info->data_reloc_bg)
3806 fs_info->data_reloc_bg = block_group->start;
3808 * Do not allow allocations from this block group, unless it is
3809 * for data relocation. Compared to increasing the ->ro, setting
3810 * the ->zoned_data_reloc_ongoing flag still allows nocow
3811 * writers to come in. See btrfs_inc_nocow_writers().
3813 * We need to disable an allocation to avoid an allocation of
3814 * regular (non-relocation data) extent. With mix of relocation
3815 * extents and regular extents, we can dispatch WRITE commands
3816 * (for relocation extents) and ZONE APPEND commands (for
3817 * regular extents) at the same time to the same zone, which
3818 * easily break the write pointer.
3820 * Also, this flag avoids this block group to be zone finished.
3822 set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags);
3825 ffe_ctl->found_offset = start + block_group->alloc_offset;
3826 block_group->alloc_offset += num_bytes;
3827 spin_lock(&ctl->tree_lock);
3828 ctl->free_space -= num_bytes;
3829 spin_unlock(&ctl->tree_lock);
3832 * We do not check if found_offset is aligned to stripesize. The
3833 * address is anyway rewritten when using zone append writing.
3836 ffe_ctl->search_start = ffe_ctl->found_offset;
3839 if (ret && ffe_ctl->for_treelog)
3840 fs_info->treelog_bg = 0;
3841 if (ret && ffe_ctl->for_data_reloc)
3842 fs_info->data_reloc_bg = 0;
3843 spin_unlock(&fs_info->relocation_bg_lock);
3844 spin_unlock(&fs_info->treelog_bg_lock);
3845 spin_unlock(&block_group->lock);
3846 spin_unlock(&space_info->lock);
3850 static int do_allocation(struct btrfs_block_group *block_group,
3851 struct find_free_extent_ctl *ffe_ctl,
3852 struct btrfs_block_group **bg_ret)
3854 switch (ffe_ctl->policy) {
3855 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3856 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3857 case BTRFS_EXTENT_ALLOC_ZONED:
3858 return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
3864 static void release_block_group(struct btrfs_block_group *block_group,
3865 struct find_free_extent_ctl *ffe_ctl,
3868 switch (ffe_ctl->policy) {
3869 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3870 ffe_ctl->retry_uncached = false;
3872 case BTRFS_EXTENT_ALLOC_ZONED:
3879 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
3881 btrfs_release_block_group(block_group, delalloc);
3884 static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
3885 struct btrfs_key *ins)
3887 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3889 if (!ffe_ctl->use_cluster && last_ptr) {
3890 spin_lock(&last_ptr->lock);
3891 last_ptr->window_start = ins->objectid;
3892 spin_unlock(&last_ptr->lock);
3896 static void found_extent(struct find_free_extent_ctl *ffe_ctl,
3897 struct btrfs_key *ins)
3899 switch (ffe_ctl->policy) {
3900 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3901 found_extent_clustered(ffe_ctl, ins);
3903 case BTRFS_EXTENT_ALLOC_ZONED:
3911 static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info,
3912 struct find_free_extent_ctl *ffe_ctl)
3914 /* Block group's activeness is not a requirement for METADATA block groups. */
3915 if (!(ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA))
3918 /* If we can activate new zone, just allocate a chunk and use it */
3919 if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags))
3923 * We already reached the max active zones. Try to finish one block
3924 * group to make a room for a new block group. This is only possible
3925 * for a data block group because btrfs_zone_finish() may need to wait
3926 * for a running transaction which can cause a deadlock for metadata
3929 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
3930 int ret = btrfs_zone_finish_one_bg(fs_info);
3939 * If we have enough free space left in an already active block group
3940 * and we can't activate any other zone now, do not allow allocating a
3941 * new chunk and let find_free_extent() retry with a smaller size.
3943 if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size)
3947 * Even min_alloc_size is not left in any block groups. Since we cannot
3948 * activate a new block group, allocating it may not help. Let's tell a
3949 * caller to try again and hope it progress something by writing some
3950 * parts of the region. That is only possible for data block groups,
3951 * where a part of the region can be written.
3953 if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)
3957 * We cannot activate a new block group and no enough space left in any
3958 * block groups. So, allocating a new block group may not help. But,
3959 * there is nothing to do anyway, so let's go with it.
3964 static int can_allocate_chunk(struct btrfs_fs_info *fs_info,
3965 struct find_free_extent_ctl *ffe_ctl)
3967 switch (ffe_ctl->policy) {
3968 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3970 case BTRFS_EXTENT_ALLOC_ZONED:
3971 return can_allocate_chunk_zoned(fs_info, ffe_ctl);
3978 * Return >0 means caller needs to re-search for free extent
3979 * Return 0 means we have the needed free extent.
3980 * Return <0 means we failed to locate any free extent.
3982 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
3983 struct btrfs_key *ins,
3984 struct find_free_extent_ctl *ffe_ctl,
3987 struct btrfs_root *root = fs_info->chunk_root;
3990 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
3991 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
3992 ffe_ctl->orig_have_caching_bg = true;
3994 if (ins->objectid) {
3995 found_extent(ffe_ctl, ins);
3999 if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
4003 if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
4006 /* See the comments for btrfs_loop_type for an explanation of the phases. */
4007 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
4010 * We want to skip the LOOP_CACHING_WAIT step if we don't have
4011 * any uncached bgs and we've already done a full search
4014 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT &&
4015 (!ffe_ctl->orig_have_caching_bg && full_search))
4019 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
4020 struct btrfs_trans_handle *trans;
4023 /* Check if allocation policy allows to create a new chunk */
4024 ret = can_allocate_chunk(fs_info, ffe_ctl);
4028 trans = current->journal_info;
4032 trans = btrfs_join_transaction(root);
4034 if (IS_ERR(trans)) {
4035 ret = PTR_ERR(trans);
4039 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
4040 CHUNK_ALLOC_FORCE_FOR_EXTENT);
4042 /* Do not bail out on ENOSPC since we can do more. */
4043 if (ret == -ENOSPC) {
4048 btrfs_abort_transaction(trans, ret);
4052 btrfs_end_transaction(trans);
4057 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4058 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4062 * Don't loop again if we already have no empty_size and
4065 if (ffe_ctl->empty_size == 0 &&
4066 ffe_ctl->empty_cluster == 0)
4068 ffe_ctl->empty_size = 0;
4069 ffe_ctl->empty_cluster = 0;
4076 static bool find_free_extent_check_size_class(struct find_free_extent_ctl *ffe_ctl,
4077 struct btrfs_block_group *bg)
4079 if (ffe_ctl->policy == BTRFS_EXTENT_ALLOC_ZONED)
4081 if (!btrfs_block_group_should_use_size_class(bg))
4083 if (ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS)
4085 if (ffe_ctl->loop >= LOOP_UNSET_SIZE_CLASS &&
4086 bg->size_class == BTRFS_BG_SZ_NONE)
4088 return ffe_ctl->size_class == bg->size_class;
4091 static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4092 struct find_free_extent_ctl *ffe_ctl,
4093 struct btrfs_space_info *space_info,
4094 struct btrfs_key *ins)
4097 * If our free space is heavily fragmented we may not be able to make
4098 * big contiguous allocations, so instead of doing the expensive search
4099 * for free space, simply return ENOSPC with our max_extent_size so we
4100 * can go ahead and search for a more manageable chunk.
4102 * If our max_extent_size is large enough for our allocation simply
4103 * disable clustering since we will likely not be able to find enough
4104 * space to create a cluster and induce latency trying.
4106 if (space_info->max_extent_size) {
4107 spin_lock(&space_info->lock);
4108 if (space_info->max_extent_size &&
4109 ffe_ctl->num_bytes > space_info->max_extent_size) {
4110 ins->offset = space_info->max_extent_size;
4111 spin_unlock(&space_info->lock);
4113 } else if (space_info->max_extent_size) {
4114 ffe_ctl->use_cluster = false;
4116 spin_unlock(&space_info->lock);
4119 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4120 &ffe_ctl->empty_cluster);
4121 if (ffe_ctl->last_ptr) {
4122 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4124 spin_lock(&last_ptr->lock);
4125 if (last_ptr->block_group)
4126 ffe_ctl->hint_byte = last_ptr->window_start;
4127 if (last_ptr->fragmented) {
4129 * We still set window_start so we can keep track of the
4130 * last place we found an allocation to try and save
4133 ffe_ctl->hint_byte = last_ptr->window_start;
4134 ffe_ctl->use_cluster = false;
4136 spin_unlock(&last_ptr->lock);
4142 static int prepare_allocation(struct btrfs_fs_info *fs_info,
4143 struct find_free_extent_ctl *ffe_ctl,
4144 struct btrfs_space_info *space_info,
4145 struct btrfs_key *ins)
4147 switch (ffe_ctl->policy) {
4148 case BTRFS_EXTENT_ALLOC_CLUSTERED:
4149 return prepare_allocation_clustered(fs_info, ffe_ctl,
4151 case BTRFS_EXTENT_ALLOC_ZONED:
4152 if (ffe_ctl->for_treelog) {
4153 spin_lock(&fs_info->treelog_bg_lock);
4154 if (fs_info->treelog_bg)
4155 ffe_ctl->hint_byte = fs_info->treelog_bg;
4156 spin_unlock(&fs_info->treelog_bg_lock);
4158 if (ffe_ctl->for_data_reloc) {
4159 spin_lock(&fs_info->relocation_bg_lock);
4160 if (fs_info->data_reloc_bg)
4161 ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4162 spin_unlock(&fs_info->relocation_bg_lock);
4171 * walks the btree of allocated extents and find a hole of a given size.
4172 * The key ins is changed to record the hole:
4173 * ins->objectid == start position
4174 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4175 * ins->offset == the size of the hole.
4176 * Any available blocks before search_start are skipped.
4178 * If there is no suitable free space, we will record the max size of
4179 * the free space extent currently.
4181 * The overall logic and call chain:
4183 * find_free_extent()
4184 * |- Iterate through all block groups
4185 * | |- Get a valid block group
4186 * | |- Try to do clustered allocation in that block group
4187 * | |- Try to do unclustered allocation in that block group
4188 * | |- Check if the result is valid
4189 * | | |- If valid, then exit
4190 * | |- Jump to next block group
4192 * |- Push harder to find free extents
4193 * |- If not found, re-iterate all block groups
4195 static noinline int find_free_extent(struct btrfs_root *root,
4196 struct btrfs_key *ins,
4197 struct find_free_extent_ctl *ffe_ctl)
4199 struct btrfs_fs_info *fs_info = root->fs_info;
4201 int cache_block_group_error = 0;
4202 struct btrfs_block_group *block_group = NULL;
4203 struct btrfs_space_info *space_info;
4204 bool full_search = false;
4206 WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4208 ffe_ctl->search_start = 0;
4209 /* For clustered allocation */
4210 ffe_ctl->empty_cluster = 0;
4211 ffe_ctl->last_ptr = NULL;
4212 ffe_ctl->use_cluster = true;
4213 ffe_ctl->have_caching_bg = false;
4214 ffe_ctl->orig_have_caching_bg = false;
4215 ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4217 ffe_ctl->retry_uncached = false;
4218 ffe_ctl->cached = 0;
4219 ffe_ctl->max_extent_size = 0;
4220 ffe_ctl->total_free_space = 0;
4221 ffe_ctl->found_offset = 0;
4222 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4223 ffe_ctl->size_class = btrfs_calc_block_group_size_class(ffe_ctl->num_bytes);
4225 if (btrfs_is_zoned(fs_info))
4226 ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4228 ins->type = BTRFS_EXTENT_ITEM_KEY;
4232 trace_find_free_extent(root, ffe_ctl);
4234 space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4236 btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4240 ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4244 ffe_ctl->search_start = max(ffe_ctl->search_start,
4245 first_logical_byte(fs_info));
4246 ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4247 if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4248 block_group = btrfs_lookup_block_group(fs_info,
4249 ffe_ctl->search_start);
4251 * we don't want to use the block group if it doesn't match our
4252 * allocation bits, or if its not cached.
4254 * However if we are re-searching with an ideal block group
4255 * picked out then we don't care that the block group is cached.
4257 if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4258 block_group->cached != BTRFS_CACHE_NO) {
4259 down_read(&space_info->groups_sem);
4260 if (list_empty(&block_group->list) ||
4263 * someone is removing this block group,
4264 * we can't jump into the have_block_group
4265 * target because our list pointers are not
4268 btrfs_put_block_group(block_group);
4269 up_read(&space_info->groups_sem);
4271 ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4272 block_group->flags);
4273 btrfs_lock_block_group(block_group,
4275 ffe_ctl->hinted = true;
4276 goto have_block_group;
4278 } else if (block_group) {
4279 btrfs_put_block_group(block_group);
4283 trace_find_free_extent_search_loop(root, ffe_ctl);
4284 ffe_ctl->have_caching_bg = false;
4285 if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4286 ffe_ctl->index == 0)
4288 down_read(&space_info->groups_sem);
4289 list_for_each_entry(block_group,
4290 &space_info->block_groups[ffe_ctl->index], list) {
4291 struct btrfs_block_group *bg_ret;
4293 ffe_ctl->hinted = false;
4294 /* If the block group is read-only, we can skip it entirely. */
4295 if (unlikely(block_group->ro)) {
4296 if (ffe_ctl->for_treelog)
4297 btrfs_clear_treelog_bg(block_group);
4298 if (ffe_ctl->for_data_reloc)
4299 btrfs_clear_data_reloc_bg(block_group);
4303 btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4304 ffe_ctl->search_start = block_group->start;
4307 * this can happen if we end up cycling through all the
4308 * raid types, but we want to make sure we only allocate
4309 * for the proper type.
4311 if (!block_group_bits(block_group, ffe_ctl->flags)) {
4312 u64 extra = BTRFS_BLOCK_GROUP_DUP |
4313 BTRFS_BLOCK_GROUP_RAID1_MASK |
4314 BTRFS_BLOCK_GROUP_RAID56_MASK |
4315 BTRFS_BLOCK_GROUP_RAID10;
4318 * if they asked for extra copies and this block group
4319 * doesn't provide them, bail. This does allow us to
4320 * fill raid0 from raid1.
4322 if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4326 * This block group has different flags than we want.
4327 * It's possible that we have MIXED_GROUP flag but no
4328 * block group is mixed. Just skip such block group.
4330 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4335 trace_find_free_extent_have_block_group(root, ffe_ctl, block_group);
4336 ffe_ctl->cached = btrfs_block_group_done(block_group);
4337 if (unlikely(!ffe_ctl->cached)) {
4338 ffe_ctl->have_caching_bg = true;
4339 ret = btrfs_cache_block_group(block_group, false);
4342 * If we get ENOMEM here or something else we want to
4343 * try other block groups, because it may not be fatal.
4344 * However if we can't find anything else we need to
4345 * save our return here so that we return the actual
4346 * error that caused problems, not ENOSPC.
4349 if (!cache_block_group_error)
4350 cache_block_group_error = ret;
4357 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) {
4358 if (!cache_block_group_error)
4359 cache_block_group_error = -EIO;
4363 if (!find_free_extent_check_size_class(ffe_ctl, block_group))
4367 ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4371 if (bg_ret && bg_ret != block_group) {
4372 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4373 block_group = bg_ret;
4377 ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4378 fs_info->stripesize);
4380 /* move on to the next group */
4381 if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4382 block_group->start + block_group->length) {
4383 btrfs_add_free_space_unused(block_group,
4384 ffe_ctl->found_offset,
4385 ffe_ctl->num_bytes);
4389 if (ffe_ctl->found_offset < ffe_ctl->search_start)
4390 btrfs_add_free_space_unused(block_group,
4391 ffe_ctl->found_offset,
4392 ffe_ctl->search_start - ffe_ctl->found_offset);
4394 ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4397 ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS);
4398 if (ret == -EAGAIN) {
4399 btrfs_add_free_space_unused(block_group,
4400 ffe_ctl->found_offset,
4401 ffe_ctl->num_bytes);
4404 btrfs_inc_block_group_reservations(block_group);
4406 /* we are all good, lets return */
4407 ins->objectid = ffe_ctl->search_start;
4408 ins->offset = ffe_ctl->num_bytes;
4410 trace_btrfs_reserve_extent(block_group, ffe_ctl);
4411 btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4414 if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
4415 !ffe_ctl->retry_uncached) {
4416 ffe_ctl->retry_uncached = true;
4417 btrfs_wait_block_group_cache_progress(block_group,
4418 ffe_ctl->num_bytes +
4419 ffe_ctl->empty_cluster +
4420 ffe_ctl->empty_size);
4421 goto have_block_group;
4423 release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4426 up_read(&space_info->groups_sem);
4428 ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4432 if (ret == -ENOSPC && !cache_block_group_error) {
4434 * Use ffe_ctl->total_free_space as fallback if we can't find
4435 * any contiguous hole.
4437 if (!ffe_ctl->max_extent_size)
4438 ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4439 spin_lock(&space_info->lock);
4440 space_info->max_extent_size = ffe_ctl->max_extent_size;
4441 spin_unlock(&space_info->lock);
4442 ins->offset = ffe_ctl->max_extent_size;
4443 } else if (ret == -ENOSPC) {
4444 ret = cache_block_group_error;
4450 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4451 * hole that is at least as big as @num_bytes.
4453 * @root - The root that will contain this extent
4455 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4456 * is used for accounting purposes. This value differs
4457 * from @num_bytes only in the case of compressed extents.
4459 * @num_bytes - Number of bytes to allocate on-disk.
4461 * @min_alloc_size - Indicates the minimum amount of space that the
4462 * allocator should try to satisfy. In some cases
4463 * @num_bytes may be larger than what is required and if
4464 * the filesystem is fragmented then allocation fails.
4465 * However, the presence of @min_alloc_size gives a
4466 * chance to try and satisfy the smaller allocation.
4468 * @empty_size - A hint that you plan on doing more COW. This is the
4469 * size in bytes the allocator should try to find free
4470 * next to the block it returns. This is just a hint and
4471 * may be ignored by the allocator.
4473 * @hint_byte - Hint to the allocator to start searching above the byte
4474 * address passed. It might be ignored.
4476 * @ins - This key is modified to record the found hole. It will
4477 * have the following values:
4478 * ins->objectid == start position
4479 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4480 * ins->offset == the size of the hole.
4482 * @is_data - Boolean flag indicating whether an extent is
4483 * allocated for data (true) or metadata (false)
4485 * @delalloc - Boolean flag indicating whether this allocation is for
4486 * delalloc or not. If 'true' data_rwsem of block groups
4487 * is going to be acquired.
4490 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4491 * case -ENOSPC is returned then @ins->offset will contain the size of the
4492 * largest available hole the allocator managed to find.
4494 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4495 u64 num_bytes, u64 min_alloc_size,
4496 u64 empty_size, u64 hint_byte,
4497 struct btrfs_key *ins, int is_data, int delalloc)
4499 struct btrfs_fs_info *fs_info = root->fs_info;
4500 struct find_free_extent_ctl ffe_ctl = {};
4501 bool final_tried = num_bytes == min_alloc_size;
4504 bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4505 bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4507 flags = get_alloc_profile_by_root(root, is_data);
4509 WARN_ON(num_bytes < fs_info->sectorsize);
4511 ffe_ctl.ram_bytes = ram_bytes;
4512 ffe_ctl.num_bytes = num_bytes;
4513 ffe_ctl.min_alloc_size = min_alloc_size;
4514 ffe_ctl.empty_size = empty_size;
4515 ffe_ctl.flags = flags;
4516 ffe_ctl.delalloc = delalloc;
4517 ffe_ctl.hint_byte = hint_byte;
4518 ffe_ctl.for_treelog = for_treelog;
4519 ffe_ctl.for_data_reloc = for_data_reloc;
4521 ret = find_free_extent(root, ins, &ffe_ctl);
4522 if (!ret && !is_data) {
4523 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4524 } else if (ret == -ENOSPC) {
4525 if (!final_tried && ins->offset) {
4526 num_bytes = min(num_bytes >> 1, ins->offset);
4527 num_bytes = round_down(num_bytes,
4528 fs_info->sectorsize);
4529 num_bytes = max(num_bytes, min_alloc_size);
4530 ram_bytes = num_bytes;
4531 if (num_bytes == min_alloc_size)
4534 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4535 struct btrfs_space_info *sinfo;
4537 sinfo = btrfs_find_space_info(fs_info, flags);
4539 "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4540 flags, num_bytes, for_treelog, for_data_reloc);
4542 btrfs_dump_space_info(fs_info, sinfo,
4550 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4551 u64 start, u64 len, int delalloc)
4553 struct btrfs_block_group *cache;
4555 cache = btrfs_lookup_block_group(fs_info, start);
4557 btrfs_err(fs_info, "Unable to find block group for %llu",
4562 btrfs_add_free_space(cache, start, len);
4563 btrfs_free_reserved_bytes(cache, len, delalloc);
4564 trace_btrfs_reserved_extent_free(fs_info, start, len);
4566 btrfs_put_block_group(cache);
4570 int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start,
4573 struct btrfs_block_group *cache;
4576 cache = btrfs_lookup_block_group(trans->fs_info, start);
4578 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4583 ret = pin_down_extent(trans, cache, start, len, 1);
4584 btrfs_put_block_group(cache);
4588 static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
4591 struct btrfs_fs_info *fs_info = trans->fs_info;
4594 ret = remove_from_free_space_tree(trans, bytenr, num_bytes);
4598 ret = btrfs_update_block_group(trans, bytenr, num_bytes, true);
4601 btrfs_err(fs_info, "update block group failed for %llu %llu",
4606 trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes);
4610 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4611 u64 parent, u64 root_objectid,
4612 u64 flags, u64 owner, u64 offset,
4613 struct btrfs_key *ins, int ref_mod)
4615 struct btrfs_fs_info *fs_info = trans->fs_info;
4616 struct btrfs_root *extent_root;
4618 struct btrfs_extent_item *extent_item;
4619 struct btrfs_extent_inline_ref *iref;
4620 struct btrfs_path *path;
4621 struct extent_buffer *leaf;
4626 type = BTRFS_SHARED_DATA_REF_KEY;
4628 type = BTRFS_EXTENT_DATA_REF_KEY;
4630 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4632 path = btrfs_alloc_path();
4636 extent_root = btrfs_extent_root(fs_info, ins->objectid);
4637 ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size);
4639 btrfs_free_path(path);
4643 leaf = path->nodes[0];
4644 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4645 struct btrfs_extent_item);
4646 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4647 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4648 btrfs_set_extent_flags(leaf, extent_item,
4649 flags | BTRFS_EXTENT_FLAG_DATA);
4651 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4652 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4654 struct btrfs_shared_data_ref *ref;
4655 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4656 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4657 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4659 struct btrfs_extent_data_ref *ref;
4660 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4661 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4662 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4663 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4664 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4667 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
4668 btrfs_free_path(path);
4670 return alloc_reserved_extent(trans, ins->objectid, ins->offset);
4673 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4674 struct btrfs_delayed_ref_node *node,
4675 struct btrfs_delayed_extent_op *extent_op)
4677 struct btrfs_fs_info *fs_info = trans->fs_info;
4678 struct btrfs_root *extent_root;
4680 struct btrfs_extent_item *extent_item;
4681 struct btrfs_key extent_key;
4682 struct btrfs_tree_block_info *block_info;
4683 struct btrfs_extent_inline_ref *iref;
4684 struct btrfs_path *path;
4685 struct extent_buffer *leaf;
4686 struct btrfs_delayed_tree_ref *ref;
4687 u32 size = sizeof(*extent_item) + sizeof(*iref);
4688 u64 flags = extent_op->flags_to_set;
4689 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4691 ref = btrfs_delayed_node_to_tree_ref(node);
4693 extent_key.objectid = node->bytenr;
4694 if (skinny_metadata) {
4695 extent_key.offset = ref->level;
4696 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4698 extent_key.offset = node->num_bytes;
4699 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4700 size += sizeof(*block_info);
4703 path = btrfs_alloc_path();
4707 extent_root = btrfs_extent_root(fs_info, extent_key.objectid);
4708 ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key,
4711 btrfs_free_path(path);
4715 leaf = path->nodes[0];
4716 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4717 struct btrfs_extent_item);
4718 btrfs_set_extent_refs(leaf, extent_item, 1);
4719 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4720 btrfs_set_extent_flags(leaf, extent_item,
4721 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4723 if (skinny_metadata) {
4724 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4726 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4727 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4728 btrfs_set_tree_block_level(leaf, block_info, ref->level);
4729 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4732 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4733 btrfs_set_extent_inline_ref_type(leaf, iref,
4734 BTRFS_SHARED_BLOCK_REF_KEY);
4735 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4737 btrfs_set_extent_inline_ref_type(leaf, iref,
4738 BTRFS_TREE_BLOCK_REF_KEY);
4739 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4742 btrfs_mark_buffer_dirty(trans, leaf);
4743 btrfs_free_path(path);
4745 return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize);
4748 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4749 struct btrfs_root *root, u64 owner,
4750 u64 offset, u64 ram_bytes,
4751 struct btrfs_key *ins)
4753 struct btrfs_ref generic_ref = { 0 };
4755 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4757 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4758 ins->objectid, ins->offset, 0);
4759 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner,
4761 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4763 return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4767 * this is used by the tree logging recovery code. It records that
4768 * an extent has been allocated and makes sure to clear the free
4769 * space cache bits as well
4771 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4772 u64 root_objectid, u64 owner, u64 offset,
4773 struct btrfs_key *ins)
4775 struct btrfs_fs_info *fs_info = trans->fs_info;
4777 struct btrfs_block_group *block_group;
4778 struct btrfs_space_info *space_info;
4781 * Mixed block groups will exclude before processing the log so we only
4782 * need to do the exclude dance if this fs isn't mixed.
4784 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4785 ret = __exclude_logged_extent(fs_info, ins->objectid,
4791 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4795 space_info = block_group->space_info;
4796 spin_lock(&space_info->lock);
4797 spin_lock(&block_group->lock);
4798 space_info->bytes_reserved += ins->offset;
4799 block_group->reserved += ins->offset;
4800 spin_unlock(&block_group->lock);
4801 spin_unlock(&space_info->lock);
4803 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4806 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4807 btrfs_put_block_group(block_group);
4811 static struct extent_buffer *
4812 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4813 u64 bytenr, int level, u64 owner,
4814 enum btrfs_lock_nesting nest)
4816 struct btrfs_fs_info *fs_info = root->fs_info;
4817 struct extent_buffer *buf;
4818 u64 lockdep_owner = owner;
4820 buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
4825 * Extra safety check in case the extent tree is corrupted and extent
4826 * allocator chooses to use a tree block which is already used and
4829 if (buf->lock_owner == current->pid) {
4830 btrfs_err_rl(fs_info,
4831 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4832 buf->start, btrfs_header_owner(buf), current->pid);
4833 free_extent_buffer(buf);
4834 return ERR_PTR(-EUCLEAN);
4838 * The reloc trees are just snapshots, so we need them to appear to be
4839 * just like any other fs tree WRT lockdep.
4841 * The exception however is in replace_path() in relocation, where we
4842 * hold the lock on the original fs root and then search for the reloc
4843 * root. At that point we need to make sure any reloc root buffers are
4844 * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
4847 if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID &&
4848 !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
4849 lockdep_owner = BTRFS_FS_TREE_OBJECTID;
4851 /* btrfs_clear_buffer_dirty() accesses generation field. */
4852 btrfs_set_header_generation(buf, trans->transid);
4855 * This needs to stay, because we could allocate a freed block from an
4856 * old tree into a new tree, so we need to make sure this new block is
4857 * set to the appropriate level and owner.
4859 btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level);
4861 __btrfs_tree_lock(buf, nest);
4862 btrfs_clear_buffer_dirty(trans, buf);
4863 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4864 clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags);
4866 set_extent_buffer_uptodate(buf);
4868 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4869 btrfs_set_header_level(buf, level);
4870 btrfs_set_header_bytenr(buf, buf->start);
4871 btrfs_set_header_generation(buf, trans->transid);
4872 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4873 btrfs_set_header_owner(buf, owner);
4874 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4875 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4876 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4877 buf->log_index = root->log_transid % 2;
4879 * we allow two log transactions at a time, use different
4880 * EXTENT bit to differentiate dirty pages.
4882 if (buf->log_index == 0)
4883 set_extent_bit(&root->dirty_log_pages, buf->start,
4884 buf->start + buf->len - 1,
4885 EXTENT_DIRTY, NULL);
4887 set_extent_bit(&root->dirty_log_pages, buf->start,
4888 buf->start + buf->len - 1,
4891 buf->log_index = -1;
4892 set_extent_bit(&trans->transaction->dirty_pages, buf->start,
4893 buf->start + buf->len - 1, EXTENT_DIRTY, NULL);
4895 /* this returns a buffer locked for blocking */
4900 * finds a free extent and does all the dirty work required for allocation
4901 * returns the tree buffer or an ERR_PTR on error.
4903 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4904 struct btrfs_root *root,
4905 u64 parent, u64 root_objectid,
4906 const struct btrfs_disk_key *key,
4907 int level, u64 hint,
4909 enum btrfs_lock_nesting nest)
4911 struct btrfs_fs_info *fs_info = root->fs_info;
4912 struct btrfs_key ins;
4913 struct btrfs_block_rsv *block_rsv;
4914 struct extent_buffer *buf;
4915 struct btrfs_delayed_extent_op *extent_op;
4916 struct btrfs_ref generic_ref = { 0 };
4919 u32 blocksize = fs_info->nodesize;
4920 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4922 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4923 if (btrfs_is_testing(fs_info)) {
4924 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4925 level, root_objectid, nest);
4927 root->alloc_bytenr += blocksize;
4932 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4933 if (IS_ERR(block_rsv))
4934 return ERR_CAST(block_rsv);
4936 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4937 empty_size, hint, &ins, 0, 0);
4941 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4942 root_objectid, nest);
4945 goto out_free_reserved;
4948 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4950 parent = ins.objectid;
4951 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4955 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4956 extent_op = btrfs_alloc_delayed_extent_op();
4962 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4964 memset(&extent_op->key, 0, sizeof(extent_op->key));
4965 extent_op->flags_to_set = flags;
4966 extent_op->update_key = skinny_metadata ? false : true;
4967 extent_op->update_flags = true;
4968 extent_op->level = level;
4970 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4971 ins.objectid, ins.offset, parent);
4972 btrfs_init_tree_ref(&generic_ref, level, root_objectid,
4973 root->root_key.objectid, false);
4974 btrfs_ref_tree_mod(fs_info, &generic_ref);
4975 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
4977 goto out_free_delayed;
4982 btrfs_free_delayed_extent_op(extent_op);
4984 btrfs_tree_unlock(buf);
4985 free_extent_buffer(buf);
4987 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
4989 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
4990 return ERR_PTR(ret);
4993 struct walk_control {
4994 u64 refs[BTRFS_MAX_LEVEL];
4995 u64 flags[BTRFS_MAX_LEVEL];
4996 struct btrfs_key update_progress;
4997 struct btrfs_key drop_progress;
5009 #define DROP_REFERENCE 1
5010 #define UPDATE_BACKREF 2
5012 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5013 struct btrfs_root *root,
5014 struct walk_control *wc,
5015 struct btrfs_path *path)
5017 struct btrfs_fs_info *fs_info = root->fs_info;
5023 struct btrfs_key key;
5024 struct extent_buffer *eb;
5029 if (path->slots[wc->level] < wc->reada_slot) {
5030 wc->reada_count = wc->reada_count * 2 / 3;
5031 wc->reada_count = max(wc->reada_count, 2);
5033 wc->reada_count = wc->reada_count * 3 / 2;
5034 wc->reada_count = min_t(int, wc->reada_count,
5035 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
5038 eb = path->nodes[wc->level];
5039 nritems = btrfs_header_nritems(eb);
5041 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5042 if (nread >= wc->reada_count)
5046 bytenr = btrfs_node_blockptr(eb, slot);
5047 generation = btrfs_node_ptr_generation(eb, slot);
5049 if (slot == path->slots[wc->level])
5052 if (wc->stage == UPDATE_BACKREF &&
5053 generation <= root->root_key.offset)
5056 /* We don't lock the tree block, it's OK to be racy here */
5057 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5058 wc->level - 1, 1, &refs,
5060 /* We don't care about errors in readahead. */
5065 if (wc->stage == DROP_REFERENCE) {
5069 if (wc->level == 1 &&
5070 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5072 if (!wc->update_ref ||
5073 generation <= root->root_key.offset)
5075 btrfs_node_key_to_cpu(eb, &key, slot);
5076 ret = btrfs_comp_cpu_keys(&key,
5077 &wc->update_progress);
5081 if (wc->level == 1 &&
5082 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5086 btrfs_readahead_node_child(eb, slot);
5089 wc->reada_slot = slot;
5093 * helper to process tree block while walking down the tree.
5095 * when wc->stage == UPDATE_BACKREF, this function updates
5096 * back refs for pointers in the block.
5098 * NOTE: return value 1 means we should stop walking down.
5100 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5101 struct btrfs_root *root,
5102 struct btrfs_path *path,
5103 struct walk_control *wc, int lookup_info)
5105 struct btrfs_fs_info *fs_info = root->fs_info;
5106 int level = wc->level;
5107 struct extent_buffer *eb = path->nodes[level];
5108 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5111 if (wc->stage == UPDATE_BACKREF &&
5112 btrfs_header_owner(eb) != root->root_key.objectid)
5116 * when reference count of tree block is 1, it won't increase
5117 * again. once full backref flag is set, we never clear it.
5120 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5121 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5122 BUG_ON(!path->locks[level]);
5123 ret = btrfs_lookup_extent_info(trans, fs_info,
5124 eb->start, level, 1,
5127 BUG_ON(ret == -ENOMEM);
5130 BUG_ON(wc->refs[level] == 0);
5133 if (wc->stage == DROP_REFERENCE) {
5134 if (wc->refs[level] > 1)
5137 if (path->locks[level] && !wc->keep_locks) {
5138 btrfs_tree_unlock_rw(eb, path->locks[level]);
5139 path->locks[level] = 0;
5144 /* wc->stage == UPDATE_BACKREF */
5145 if (!(wc->flags[level] & flag)) {
5146 BUG_ON(!path->locks[level]);
5147 ret = btrfs_inc_ref(trans, root, eb, 1);
5148 BUG_ON(ret); /* -ENOMEM */
5149 ret = btrfs_dec_ref(trans, root, eb, 0);
5150 BUG_ON(ret); /* -ENOMEM */
5151 ret = btrfs_set_disk_extent_flags(trans, eb, flag);
5152 BUG_ON(ret); /* -ENOMEM */
5153 wc->flags[level] |= flag;
5157 * the block is shared by multiple trees, so it's not good to
5158 * keep the tree lock
5160 if (path->locks[level] && level > 0) {
5161 btrfs_tree_unlock_rw(eb, path->locks[level]);
5162 path->locks[level] = 0;
5168 * This is used to verify a ref exists for this root to deal with a bug where we
5169 * would have a drop_progress key that hadn't been updated properly.
5171 static int check_ref_exists(struct btrfs_trans_handle *trans,
5172 struct btrfs_root *root, u64 bytenr, u64 parent,
5175 struct btrfs_path *path;
5176 struct btrfs_extent_inline_ref *iref;
5179 path = btrfs_alloc_path();
5183 ret = lookup_extent_backref(trans, path, &iref, bytenr,
5184 root->fs_info->nodesize, parent,
5185 root->root_key.objectid, level, 0);
5186 btrfs_free_path(path);
5195 * helper to process tree block pointer.
5197 * when wc->stage == DROP_REFERENCE, this function checks
5198 * reference count of the block pointed to. if the block
5199 * is shared and we need update back refs for the subtree
5200 * rooted at the block, this function changes wc->stage to
5201 * UPDATE_BACKREF. if the block is shared and there is no
5202 * need to update back, this function drops the reference
5205 * NOTE: return value 1 means we should stop walking down.
5207 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5208 struct btrfs_root *root,
5209 struct btrfs_path *path,
5210 struct walk_control *wc, int *lookup_info)
5212 struct btrfs_fs_info *fs_info = root->fs_info;
5216 struct btrfs_tree_parent_check check = { 0 };
5217 struct btrfs_key key;
5218 struct btrfs_ref ref = { 0 };
5219 struct extent_buffer *next;
5220 int level = wc->level;
5223 bool need_account = false;
5225 generation = btrfs_node_ptr_generation(path->nodes[level],
5226 path->slots[level]);
5228 * if the lower level block was created before the snapshot
5229 * was created, we know there is no need to update back refs
5232 if (wc->stage == UPDATE_BACKREF &&
5233 generation <= root->root_key.offset) {
5238 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5240 check.level = level - 1;
5241 check.transid = generation;
5242 check.owner_root = root->root_key.objectid;
5243 check.has_first_key = true;
5244 btrfs_node_key_to_cpu(path->nodes[level], &check.first_key,
5245 path->slots[level]);
5247 next = find_extent_buffer(fs_info, bytenr);
5249 next = btrfs_find_create_tree_block(fs_info, bytenr,
5250 root->root_key.objectid, level - 1);
5252 return PTR_ERR(next);
5255 btrfs_tree_lock(next);
5257 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5258 &wc->refs[level - 1],
5259 &wc->flags[level - 1]);
5263 if (unlikely(wc->refs[level - 1] == 0)) {
5264 btrfs_err(fs_info, "Missing references.");
5270 if (wc->stage == DROP_REFERENCE) {
5271 if (wc->refs[level - 1] > 1) {
5272 need_account = true;
5274 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5277 if (!wc->update_ref ||
5278 generation <= root->root_key.offset)
5281 btrfs_node_key_to_cpu(path->nodes[level], &key,
5282 path->slots[level]);
5283 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5287 wc->stage = UPDATE_BACKREF;
5288 wc->shared_level = level - 1;
5292 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5296 if (!btrfs_buffer_uptodate(next, generation, 0)) {
5297 btrfs_tree_unlock(next);
5298 free_extent_buffer(next);
5304 if (reada && level == 1)
5305 reada_walk_down(trans, root, wc, path);
5306 next = read_tree_block(fs_info, bytenr, &check);
5308 return PTR_ERR(next);
5309 } else if (!extent_buffer_uptodate(next)) {
5310 free_extent_buffer(next);
5313 btrfs_tree_lock(next);
5317 ASSERT(level == btrfs_header_level(next));
5318 if (level != btrfs_header_level(next)) {
5319 btrfs_err(root->fs_info, "mismatched level");
5323 path->nodes[level] = next;
5324 path->slots[level] = 0;
5325 path->locks[level] = BTRFS_WRITE_LOCK;
5331 wc->refs[level - 1] = 0;
5332 wc->flags[level - 1] = 0;
5333 if (wc->stage == DROP_REFERENCE) {
5334 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5335 parent = path->nodes[level]->start;
5337 ASSERT(root->root_key.objectid ==
5338 btrfs_header_owner(path->nodes[level]));
5339 if (root->root_key.objectid !=
5340 btrfs_header_owner(path->nodes[level])) {
5341 btrfs_err(root->fs_info,
5342 "mismatched block owner");
5350 * If we had a drop_progress we need to verify the refs are set
5351 * as expected. If we find our ref then we know that from here
5352 * on out everything should be correct, and we can clear the
5355 if (wc->restarted) {
5356 ret = check_ref_exists(trans, root, bytenr, parent,
5367 * Reloc tree doesn't contribute to qgroup numbers, and we have
5368 * already accounted them at merge time (replace_path),
5369 * thus we could skip expensive subtree trace here.
5371 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5373 ret = btrfs_qgroup_trace_subtree(trans, next,
5374 generation, level - 1);
5376 btrfs_err_rl(fs_info,
5377 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5383 * We need to update the next key in our walk control so we can
5384 * update the drop_progress key accordingly. We don't care if
5385 * find_next_key doesn't find a key because that means we're at
5386 * the end and are going to clean up now.
5388 wc->drop_level = level;
5389 find_next_key(path, level, &wc->drop_progress);
5391 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5392 fs_info->nodesize, parent);
5393 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid,
5395 ret = btrfs_free_extent(trans, &ref);
5404 btrfs_tree_unlock(next);
5405 free_extent_buffer(next);
5411 * helper to process tree block while walking up the tree.
5413 * when wc->stage == DROP_REFERENCE, this function drops
5414 * reference count on the block.
5416 * when wc->stage == UPDATE_BACKREF, this function changes
5417 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5418 * to UPDATE_BACKREF previously while processing the block.
5420 * NOTE: return value 1 means we should stop walking up.
5422 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5423 struct btrfs_root *root,
5424 struct btrfs_path *path,
5425 struct walk_control *wc)
5427 struct btrfs_fs_info *fs_info = root->fs_info;
5429 int level = wc->level;
5430 struct extent_buffer *eb = path->nodes[level];
5433 if (wc->stage == UPDATE_BACKREF) {
5434 BUG_ON(wc->shared_level < level);
5435 if (level < wc->shared_level)
5438 ret = find_next_key(path, level + 1, &wc->update_progress);
5442 wc->stage = DROP_REFERENCE;
5443 wc->shared_level = -1;
5444 path->slots[level] = 0;
5447 * check reference count again if the block isn't locked.
5448 * we should start walking down the tree again if reference
5451 if (!path->locks[level]) {
5453 btrfs_tree_lock(eb);
5454 path->locks[level] = BTRFS_WRITE_LOCK;
5456 ret = btrfs_lookup_extent_info(trans, fs_info,
5457 eb->start, level, 1,
5461 btrfs_tree_unlock_rw(eb, path->locks[level]);
5462 path->locks[level] = 0;
5465 BUG_ON(wc->refs[level] == 0);
5466 if (wc->refs[level] == 1) {
5467 btrfs_tree_unlock_rw(eb, path->locks[level]);
5468 path->locks[level] = 0;
5474 /* wc->stage == DROP_REFERENCE */
5475 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5477 if (wc->refs[level] == 1) {
5479 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5480 ret = btrfs_dec_ref(trans, root, eb, 1);
5482 ret = btrfs_dec_ref(trans, root, eb, 0);
5483 BUG_ON(ret); /* -ENOMEM */
5484 if (is_fstree(root->root_key.objectid)) {
5485 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5487 btrfs_err_rl(fs_info,
5488 "error %d accounting leaf items, quota is out of sync, rescan required",
5493 /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */
5494 if (!path->locks[level]) {
5495 btrfs_tree_lock(eb);
5496 path->locks[level] = BTRFS_WRITE_LOCK;
5498 btrfs_clear_buffer_dirty(trans, eb);
5501 if (eb == root->node) {
5502 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5504 else if (root->root_key.objectid != btrfs_header_owner(eb))
5505 goto owner_mismatch;
5507 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5508 parent = path->nodes[level + 1]->start;
5509 else if (root->root_key.objectid !=
5510 btrfs_header_owner(path->nodes[level + 1]))
5511 goto owner_mismatch;
5514 btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent,
5515 wc->refs[level] == 1);
5517 wc->refs[level] = 0;
5518 wc->flags[level] = 0;
5522 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5523 btrfs_header_owner(eb), root->root_key.objectid);
5527 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5528 struct btrfs_root *root,
5529 struct btrfs_path *path,
5530 struct walk_control *wc)
5532 int level = wc->level;
5533 int lookup_info = 1;
5536 while (level >= 0) {
5537 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5544 if (path->slots[level] >=
5545 btrfs_header_nritems(path->nodes[level]))
5548 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5550 path->slots[level]++;
5556 return (ret == 1) ? 0 : ret;
5559 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5560 struct btrfs_root *root,
5561 struct btrfs_path *path,
5562 struct walk_control *wc, int max_level)
5564 int level = wc->level;
5567 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5568 while (level < max_level && path->nodes[level]) {
5570 if (path->slots[level] + 1 <
5571 btrfs_header_nritems(path->nodes[level])) {
5572 path->slots[level]++;
5575 ret = walk_up_proc(trans, root, path, wc);
5581 if (path->locks[level]) {
5582 btrfs_tree_unlock_rw(path->nodes[level],
5583 path->locks[level]);
5584 path->locks[level] = 0;
5586 free_extent_buffer(path->nodes[level]);
5587 path->nodes[level] = NULL;
5595 * drop a subvolume tree.
5597 * this function traverses the tree freeing any blocks that only
5598 * referenced by the tree.
5600 * when a shared tree block is found. this function decreases its
5601 * reference count by one. if update_ref is true, this function
5602 * also make sure backrefs for the shared block and all lower level
5603 * blocks are properly updated.
5605 * If called with for_reloc == 0, may exit early with -EAGAIN
5607 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5609 const bool is_reloc_root = (root->root_key.objectid ==
5610 BTRFS_TREE_RELOC_OBJECTID);
5611 struct btrfs_fs_info *fs_info = root->fs_info;
5612 struct btrfs_path *path;
5613 struct btrfs_trans_handle *trans;
5614 struct btrfs_root *tree_root = fs_info->tree_root;
5615 struct btrfs_root_item *root_item = &root->root_item;
5616 struct walk_control *wc;
5617 struct btrfs_key key;
5621 bool root_dropped = false;
5622 bool unfinished_drop = false;
5624 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5626 path = btrfs_alloc_path();
5632 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5634 btrfs_free_path(path);
5640 * Use join to avoid potential EINTR from transaction start. See
5641 * wait_reserve_ticket and the whole reservation callchain.
5644 trans = btrfs_join_transaction(tree_root);
5646 trans = btrfs_start_transaction(tree_root, 0);
5647 if (IS_ERR(trans)) {
5648 err = PTR_ERR(trans);
5652 err = btrfs_run_delayed_items(trans);
5657 * This will help us catch people modifying the fs tree while we're
5658 * dropping it. It is unsafe to mess with the fs tree while it's being
5659 * dropped as we unlock the root node and parent nodes as we walk down
5660 * the tree, assuming nothing will change. If something does change
5661 * then we'll have stale information and drop references to blocks we've
5664 set_bit(BTRFS_ROOT_DELETING, &root->state);
5665 unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
5667 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5668 level = btrfs_header_level(root->node);
5669 path->nodes[level] = btrfs_lock_root_node(root);
5670 path->slots[level] = 0;
5671 path->locks[level] = BTRFS_WRITE_LOCK;
5672 memset(&wc->update_progress, 0,
5673 sizeof(wc->update_progress));
5675 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5676 memcpy(&wc->update_progress, &key,
5677 sizeof(wc->update_progress));
5679 level = btrfs_root_drop_level(root_item);
5681 path->lowest_level = level;
5682 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5683 path->lowest_level = 0;
5691 * unlock our path, this is safe because only this
5692 * function is allowed to delete this snapshot
5694 btrfs_unlock_up_safe(path, 0);
5696 level = btrfs_header_level(root->node);
5698 btrfs_tree_lock(path->nodes[level]);
5699 path->locks[level] = BTRFS_WRITE_LOCK;
5701 ret = btrfs_lookup_extent_info(trans, fs_info,
5702 path->nodes[level]->start,
5703 level, 1, &wc->refs[level],
5709 BUG_ON(wc->refs[level] == 0);
5711 if (level == btrfs_root_drop_level(root_item))
5714 btrfs_tree_unlock(path->nodes[level]);
5715 path->locks[level] = 0;
5716 WARN_ON(wc->refs[level] != 1);
5721 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5723 wc->shared_level = -1;
5724 wc->stage = DROP_REFERENCE;
5725 wc->update_ref = update_ref;
5727 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5731 ret = walk_down_tree(trans, root, path, wc);
5733 btrfs_abort_transaction(trans, ret);
5738 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5740 btrfs_abort_transaction(trans, ret);
5746 BUG_ON(wc->stage != DROP_REFERENCE);
5750 if (wc->stage == DROP_REFERENCE) {
5751 wc->drop_level = wc->level;
5752 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5754 path->slots[wc->drop_level]);
5756 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5757 &wc->drop_progress);
5758 btrfs_set_root_drop_level(root_item, wc->drop_level);
5760 BUG_ON(wc->level == 0);
5761 if (btrfs_should_end_transaction(trans) ||
5762 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5763 ret = btrfs_update_root(trans, tree_root,
5767 btrfs_abort_transaction(trans, ret);
5773 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
5775 btrfs_end_transaction_throttle(trans);
5776 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5777 btrfs_debug(fs_info,
5778 "drop snapshot early exit");
5784 * Use join to avoid potential EINTR from transaction
5785 * start. See wait_reserve_ticket and the whole
5786 * reservation callchain.
5789 trans = btrfs_join_transaction(tree_root);
5791 trans = btrfs_start_transaction(tree_root, 0);
5792 if (IS_ERR(trans)) {
5793 err = PTR_ERR(trans);
5798 btrfs_release_path(path);
5802 ret = btrfs_del_root(trans, &root->root_key);
5804 btrfs_abort_transaction(trans, ret);
5809 if (!is_reloc_root) {
5810 ret = btrfs_find_root(tree_root, &root->root_key, path,
5813 btrfs_abort_transaction(trans, ret);
5816 } else if (ret > 0) {
5817 /* if we fail to delete the orphan item this time
5818 * around, it'll get picked up the next time.
5820 * The most common failure here is just -ENOENT.
5822 btrfs_del_orphan_item(trans, tree_root,
5823 root->root_key.objectid);
5828 * This subvolume is going to be completely dropped, and won't be
5829 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5830 * commit transaction time. So free it here manually.
5832 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
5833 btrfs_qgroup_free_meta_all_pertrans(root);
5835 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
5836 btrfs_add_dropped_root(trans, root);
5838 btrfs_put_root(root);
5839 root_dropped = true;
5842 btrfs_set_last_root_drop_gen(fs_info, trans->transid);
5844 btrfs_end_transaction_throttle(trans);
5847 btrfs_free_path(path);
5850 * We were an unfinished drop root, check to see if there are any
5851 * pending, and if not clear and wake up any waiters.
5853 if (!err && unfinished_drop)
5854 btrfs_maybe_wake_unfinished_drop(fs_info);
5857 * So if we need to stop dropping the snapshot for whatever reason we
5858 * need to make sure to add it back to the dead root list so that we
5859 * keep trying to do the work later. This also cleans up roots if we
5860 * don't have it in the radix (like when we recover after a power fail
5861 * or unmount) so we don't leak memory.
5863 if (!for_reloc && !root_dropped)
5864 btrfs_add_dead_root(root);
5869 * drop subtree rooted at tree block 'node'.
5871 * NOTE: this function will unlock and release tree block 'node'
5872 * only used by relocation code
5874 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5875 struct btrfs_root *root,
5876 struct extent_buffer *node,
5877 struct extent_buffer *parent)
5879 struct btrfs_fs_info *fs_info = root->fs_info;
5880 struct btrfs_path *path;
5881 struct walk_control *wc;
5887 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5889 path = btrfs_alloc_path();
5893 wc = kzalloc(sizeof(*wc), GFP_NOFS);
5895 btrfs_free_path(path);
5899 btrfs_assert_tree_write_locked(parent);
5900 parent_level = btrfs_header_level(parent);
5901 atomic_inc(&parent->refs);
5902 path->nodes[parent_level] = parent;
5903 path->slots[parent_level] = btrfs_header_nritems(parent);
5905 btrfs_assert_tree_write_locked(node);
5906 level = btrfs_header_level(node);
5907 path->nodes[level] = node;
5908 path->slots[level] = 0;
5909 path->locks[level] = BTRFS_WRITE_LOCK;
5911 wc->refs[parent_level] = 1;
5912 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5914 wc->shared_level = -1;
5915 wc->stage = DROP_REFERENCE;
5918 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5921 wret = walk_down_tree(trans, root, path, wc);
5927 wret = walk_up_tree(trans, root, path, wc, parent_level);
5935 btrfs_free_path(path);
5939 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5942 return unpin_extent_range(fs_info, start, end, false);
5946 * It used to be that old block groups would be left around forever.
5947 * Iterating over them would be enough to trim unused space. Since we
5948 * now automatically remove them, we also need to iterate over unallocated
5951 * We don't want a transaction for this since the discard may take a
5952 * substantial amount of time. We don't require that a transaction be
5953 * running, but we do need to take a running transaction into account
5954 * to ensure that we're not discarding chunks that were released or
5955 * allocated in the current transaction.
5957 * Holding the chunks lock will prevent other threads from allocating
5958 * or releasing chunks, but it won't prevent a running transaction
5959 * from committing and releasing the memory that the pending chunks
5960 * list head uses. For that, we need to take a reference to the
5961 * transaction and hold the commit root sem. We only need to hold
5962 * it while performing the free space search since we have already
5963 * held back allocations.
5965 static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
5967 u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0;
5972 /* Discard not supported = nothing to do. */
5973 if (!bdev_max_discard_sectors(device->bdev))
5976 /* Not writable = nothing to do. */
5977 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5980 /* No free space = nothing to do. */
5981 if (device->total_bytes <= device->bytes_used)
5987 struct btrfs_fs_info *fs_info = device->fs_info;
5990 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
5994 find_first_clear_extent_bit(&device->alloc_state, start,
5996 CHUNK_TRIMMED | CHUNK_ALLOCATED);
5998 /* Check if there are any CHUNK_* bits left */
5999 if (start > device->total_bytes) {
6000 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
6001 btrfs_warn_in_rcu(fs_info,
6002 "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
6003 start, end - start + 1,
6004 btrfs_dev_name(device),
6005 device->total_bytes);
6006 mutex_unlock(&fs_info->chunk_mutex);
6011 /* Ensure we skip the reserved space on each device. */
6012 start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
6015 * If find_first_clear_extent_bit find a range that spans the
6016 * end of the device it will set end to -1, in this case it's up
6017 * to the caller to trim the value to the size of the device.
6019 end = min(end, device->total_bytes - 1);
6021 len = end - start + 1;
6023 /* We didn't find any extents */
6025 mutex_unlock(&fs_info->chunk_mutex);
6030 ret = btrfs_issue_discard(device->bdev, start, len,
6033 set_extent_bit(&device->alloc_state, start,
6034 start + bytes - 1, CHUNK_TRIMMED, NULL);
6035 mutex_unlock(&fs_info->chunk_mutex);
6043 if (fatal_signal_pending(current)) {
6055 * Trim the whole filesystem by:
6056 * 1) trimming the free space in each block group
6057 * 2) trimming the unallocated space on each device
6059 * This will also continue trimming even if a block group or device encounters
6060 * an error. The return value will be the last error, or 0 if nothing bad
6063 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6065 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6066 struct btrfs_block_group *cache = NULL;
6067 struct btrfs_device *device;
6069 u64 range_end = U64_MAX;
6079 if (range->start == U64_MAX)
6083 * Check range overflow if range->len is set.
6084 * The default range->len is U64_MAX.
6086 if (range->len != U64_MAX &&
6087 check_add_overflow(range->start, range->len, &range_end))
6090 cache = btrfs_lookup_first_block_group(fs_info, range->start);
6091 for (; cache; cache = btrfs_next_block_group(cache)) {
6092 if (cache->start >= range_end) {
6093 btrfs_put_block_group(cache);
6097 start = max(range->start, cache->start);
6098 end = min(range_end, cache->start + cache->length);
6100 if (end - start >= range->minlen) {
6101 if (!btrfs_block_group_done(cache)) {
6102 ret = btrfs_cache_block_group(cache, true);
6109 ret = btrfs_trim_block_group(cache,
6115 trimmed += group_trimmed;
6126 "failed to trim %llu block group(s), last error %d",
6129 mutex_lock(&fs_devices->device_list_mutex);
6130 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6131 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6134 ret = btrfs_trim_free_extents(device, &group_trimmed);
6141 trimmed += group_trimmed;
6143 mutex_unlock(&fs_devices->device_list_mutex);
6147 "failed to trim %llu device(s), last error %d",
6148 dev_failed, dev_ret);
6149 range->len = trimmed;