Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Copyright (C) 2007 Oracle. All rights reserved.
4 : */
5 :
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>
19 : #include "ctree.h"
20 : #include "extent-tree.h"
21 : #include "tree-log.h"
22 : #include "disk-io.h"
23 : #include "print-tree.h"
24 : #include "volumes.h"
25 : #include "raid56.h"
26 : #include "locking.h"
27 : #include "free-space-cache.h"
28 : #include "free-space-tree.h"
29 : #include "sysfs.h"
30 : #include "qgroup.h"
31 : #include "ref-verify.h"
32 : #include "space-info.h"
33 : #include "block-rsv.h"
34 : #include "delalloc-space.h"
35 : #include "discard.h"
36 : #include "rcu-string.h"
37 : #include "zoned.h"
38 : #include "dev-replace.h"
39 : #include "fs.h"
40 : #include "accessors.h"
41 : #include "root-tree.h"
42 : #include "file-item.h"
43 : #include "orphan.h"
44 : #include "tree-checker.h"
45 :
46 : #undef SCRAMBLE_DELAYED_REFS
47 :
48 :
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);
66 :
67 : static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
68 : {
69 23685218 : return (cache->flags & bits) == bits;
70 : }
71 :
72 3240 : int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info,
73 : u64 start, u64 num_bytes)
74 : {
75 3240 : u64 end = start + num_bytes - 1;
76 3240 : set_extent_bit(&fs_info->excluded_extents, start, end,
77 : EXTENT_UPTODATE, NULL);
78 3240 : return 0;
79 : }
80 :
81 22711 : void btrfs_free_excluded_extents(struct btrfs_block_group *cache)
82 : {
83 22711 : struct btrfs_fs_info *fs_info = cache->fs_info;
84 22711 : u64 start, end;
85 :
86 22711 : start = cache->start;
87 22711 : end = start + cache->length - 1;
88 :
89 22711 : clear_extent_bits(&fs_info->excluded_extents, start, end,
90 : EXTENT_UPTODATE);
91 22711 : }
92 :
93 : /* simple helper to search for an existing data extent at a given offset */
94 87872 : int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
95 : {
96 87872 : struct btrfs_root *root = btrfs_extent_root(fs_info, start);
97 87872 : int ret;
98 87872 : struct btrfs_key key;
99 87872 : struct btrfs_path *path;
100 :
101 87872 : path = btrfs_alloc_path();
102 87872 : if (!path)
103 : return -ENOMEM;
104 :
105 87872 : key.objectid = start;
106 87872 : key.offset = len;
107 87872 : key.type = BTRFS_EXTENT_ITEM_KEY;
108 87872 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
109 87872 : btrfs_free_path(path);
110 87872 : return ret;
111 : }
112 :
113 : /*
114 : * helper function to lookup reference count and flags of a tree block.
115 : *
116 : * the head node for delayed ref is used to store the sum of all the
117 : * reference count modifications queued up in the rbtree. the head
118 : * node may also store the extent flags to set. This way you can check
119 : * to see what the reference count and extent flags would be if all of
120 : * the delayed refs are not processed.
121 : */
122 8717771 : int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
123 : struct btrfs_fs_info *fs_info, u64 bytenr,
124 : u64 offset, int metadata, u64 *refs, u64 *flags)
125 : {
126 8717771 : struct btrfs_root *extent_root;
127 8717771 : struct btrfs_delayed_ref_head *head;
128 8717771 : struct btrfs_delayed_ref_root *delayed_refs;
129 8717771 : struct btrfs_path *path;
130 8717771 : struct btrfs_extent_item *ei;
131 8717771 : struct extent_buffer *leaf;
132 8717771 : struct btrfs_key key;
133 8717771 : u32 item_size;
134 8717771 : u64 num_refs;
135 8717771 : u64 extent_flags;
136 8717771 : int ret;
137 :
138 : /*
139 : * If we don't have skinny metadata, don't bother doing anything
140 : * different
141 : */
142 8717771 : if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
143 0 : offset = fs_info->nodesize;
144 0 : metadata = 0;
145 : }
146 :
147 8717771 : path = btrfs_alloc_path();
148 8717772 : if (!path)
149 : return -ENOMEM;
150 :
151 8717772 : if (!trans) {
152 0 : path->skip_locking = 1;
153 0 : path->search_commit_root = 1;
154 : }
155 :
156 8717772 : search_again:
157 8717822 : key.objectid = bytenr;
158 8717822 : key.offset = offset;
159 8717822 : if (metadata)
160 8717822 : key.type = BTRFS_METADATA_ITEM_KEY;
161 : else
162 0 : key.type = BTRFS_EXTENT_ITEM_KEY;
163 :
164 8717822 : extent_root = btrfs_extent_root(fs_info, bytenr);
165 8717822 : ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
166 8717821 : if (ret < 0)
167 0 : goto out_free;
168 :
169 8717821 : if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
170 721 : if (path->slots[0]) {
171 721 : path->slots[0]--;
172 721 : btrfs_item_key_to_cpu(path->nodes[0], &key,
173 : path->slots[0]);
174 721 : if (key.objectid == bytenr &&
175 0 : key.type == BTRFS_EXTENT_ITEM_KEY &&
176 0 : key.offset == fs_info->nodesize)
177 : ret = 0;
178 : }
179 : }
180 :
181 8717821 : if (ret == 0) {
182 8717100 : leaf = path->nodes[0];
183 8717100 : item_size = btrfs_item_size(leaf, path->slots[0]);
184 8717100 : if (item_size >= sizeof(*ei)) {
185 8717100 : ei = btrfs_item_ptr(leaf, path->slots[0],
186 : struct btrfs_extent_item);
187 8717100 : num_refs = btrfs_extent_refs(leaf, ei);
188 8717101 : extent_flags = btrfs_extent_flags(leaf, ei);
189 : } else {
190 0 : ret = -EINVAL;
191 0 : btrfs_print_v0_err(fs_info);
192 0 : if (trans)
193 0 : btrfs_abort_transaction(trans, ret);
194 : else
195 0 : btrfs_handle_fs_error(fs_info, ret, NULL);
196 :
197 0 : goto out_free;
198 : }
199 :
200 8717101 : BUG_ON(num_refs == 0);
201 : } else {
202 : num_refs = 0;
203 : extent_flags = 0;
204 : ret = 0;
205 : }
206 :
207 8717822 : if (!trans)
208 0 : goto out;
209 :
210 8717822 : delayed_refs = &trans->transaction->delayed_refs;
211 8717822 : spin_lock(&delayed_refs->lock);
212 8717822 : head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
213 8717822 : if (head) {
214 2520512 : if (!mutex_trylock(&head->mutex)) {
215 50 : refcount_inc(&head->refs);
216 50 : spin_unlock(&delayed_refs->lock);
217 :
218 50 : btrfs_release_path(path);
219 :
220 : /*
221 : * Mutex was contended, block until it's released and try
222 : * again
223 : */
224 50 : mutex_lock(&head->mutex);
225 50 : mutex_unlock(&head->mutex);
226 50 : btrfs_put_delayed_ref_head(head);
227 50 : goto search_again;
228 : }
229 2520462 : spin_lock(&head->lock);
230 2520462 : if (head->extent_op && head->extent_op->update_flags)
231 2506 : extent_flags |= head->extent_op->flags_to_set;
232 : else
233 2517956 : BUG_ON(num_refs == 0);
234 :
235 2520462 : num_refs += head->ref_mod;
236 2520462 : spin_unlock(&head->lock);
237 2520462 : mutex_unlock(&head->mutex);
238 : }
239 8717772 : spin_unlock(&delayed_refs->lock);
240 8717772 : out:
241 8717772 : WARN_ON(num_refs == 0);
242 8717772 : if (refs)
243 8717772 : *refs = num_refs;
244 8717772 : if (flags)
245 8717772 : *flags = extent_flags;
246 0 : out_free:
247 8717772 : btrfs_free_path(path);
248 8717772 : return ret;
249 : }
250 :
251 : /*
252 : * Back reference rules. Back refs have three main goals:
253 : *
254 : * 1) differentiate between all holders of references to an extent so that
255 : * when a reference is dropped we can make sure it was a valid reference
256 : * before freeing the extent.
257 : *
258 : * 2) Provide enough information to quickly find the holders of an extent
259 : * if we notice a given block is corrupted or bad.
260 : *
261 : * 3) Make it easy to migrate blocks for FS shrinking or storage pool
262 : * maintenance. This is actually the same as #2, but with a slightly
263 : * different use case.
264 : *
265 : * There are two kinds of back refs. The implicit back refs is optimized
266 : * for pointers in non-shared tree blocks. For a given pointer in a block,
267 : * back refs of this kind provide information about the block's owner tree
268 : * and the pointer's key. These information allow us to find the block by
269 : * b-tree searching. The full back refs is for pointers in tree blocks not
270 : * referenced by their owner trees. The location of tree block is recorded
271 : * in the back refs. Actually the full back refs is generic, and can be
272 : * used in all cases the implicit back refs is used. The major shortcoming
273 : * of the full back refs is its overhead. Every time a tree block gets
274 : * COWed, we have to update back refs entry for all pointers in it.
275 : *
276 : * For a newly allocated tree block, we use implicit back refs for
277 : * pointers in it. This means most tree related operations only involve
278 : * implicit back refs. For a tree block created in old transaction, the
279 : * only way to drop a reference to it is COW it. So we can detect the
280 : * event that tree block loses its owner tree's reference and do the
281 : * back refs conversion.
282 : *
283 : * When a tree block is COWed through a tree, there are four cases:
284 : *
285 : * The reference count of the block is one and the tree is the block's
286 : * owner tree. Nothing to do in this case.
287 : *
288 : * The reference count of the block is one and the tree is not the
289 : * block's owner tree. In this case, full back refs is used for pointers
290 : * in the block. Remove these full back refs, add implicit back refs for
291 : * every pointers in the new block.
292 : *
293 : * The reference count of the block is greater than one and the tree is
294 : * the block's owner tree. In this case, implicit back refs is used for
295 : * pointers in the block. Add full back refs for every pointers in the
296 : * block, increase lower level extents' reference counts. The original
297 : * implicit back refs are entailed to the new block.
298 : *
299 : * The reference count of the block is greater than one and the tree is
300 : * not the block's owner tree. Add implicit back refs for every pointer in
301 : * the new block, increase lower level extents' reference count.
302 : *
303 : * Back Reference Key composing:
304 : *
305 : * The key objectid corresponds to the first byte in the extent,
306 : * The key type is used to differentiate between types of back refs.
307 : * There are different meanings of the key offset for different types
308 : * of back refs.
309 : *
310 : * File extents can be referenced by:
311 : *
312 : * - multiple snapshots, subvolumes, or different generations in one subvol
313 : * - different files inside a single subvolume
314 : * - different offsets inside a file (bookend extents in file.c)
315 : *
316 : * The extent ref structure for the implicit back refs has fields for:
317 : *
318 : * - Objectid of the subvolume root
319 : * - objectid of the file holding the reference
320 : * - original offset in the file
321 : * - how many bookend extents
322 : *
323 : * The key offset for the implicit back refs is hash of the first
324 : * three fields.
325 : *
326 : * The extent ref structure for the full back refs has field for:
327 : *
328 : * - number of pointers in the tree leaf
329 : *
330 : * The key offset for the implicit back refs is the first byte of
331 : * the tree leaf
332 : *
333 : * When a file extent is allocated, The implicit back refs is used.
334 : * the fields are filled in:
335 : *
336 : * (root_key.objectid, inode objectid, offset in file, 1)
337 : *
338 : * When a file extent is removed file truncation, we find the
339 : * corresponding implicit back refs and check the following fields:
340 : *
341 : * (btrfs_header_owner(leaf), inode objectid, offset in file)
342 : *
343 : * Btree extents can be referenced by:
344 : *
345 : * - Different subvolumes
346 : *
347 : * Both the implicit back refs and the full back refs for tree blocks
348 : * only consist of key. The key offset for the implicit back refs is
349 : * objectid of block's owner tree. The key offset for the full back refs
350 : * is the first byte of parent block.
351 : *
352 : * When implicit back refs is used, information about the lowest key and
353 : * level of the tree block are required. These information are stored in
354 : * tree block info structure.
355 : */
356 :
357 : /*
358 : * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
359 : * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
360 : * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
361 : */
362 344209226 : int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
363 : struct btrfs_extent_inline_ref *iref,
364 : enum btrfs_inline_ref_type is_data)
365 : {
366 344209226 : int type = btrfs_extent_inline_ref_type(eb, iref);
367 344179502 : u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
368 :
369 344180061 : if (type == BTRFS_TREE_BLOCK_REF_KEY ||
370 344180061 : type == BTRFS_SHARED_BLOCK_REF_KEY ||
371 291346511 : type == BTRFS_SHARED_DATA_REF_KEY ||
372 291346511 : type == BTRFS_EXTENT_DATA_REF_KEY) {
373 344183489 : if (is_data == BTRFS_REF_TYPE_BLOCK) {
374 29797488 : if (type == BTRFS_TREE_BLOCK_REF_KEY)
375 : return type;
376 23562031 : if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
377 23562036 : ASSERT(eb->fs_info);
378 : /*
379 : * Every shared one has parent tree block,
380 : * which must be aligned to sector size.
381 : */
382 23562036 : if (offset &&
383 23562021 : IS_ALIGNED(offset, eb->fs_info->sectorsize))
384 : return type;
385 : }
386 314386001 : } else if (is_data == BTRFS_REF_TYPE_DATA) {
387 281806178 : if (type == BTRFS_EXTENT_DATA_REF_KEY)
388 : return type;
389 40113018 : if (type == BTRFS_SHARED_DATA_REF_KEY) {
390 40113087 : ASSERT(eb->fs_info);
391 : /*
392 : * Every shared one has parent tree block,
393 : * which must be aligned to sector size.
394 : */
395 40113087 : if (offset &&
396 40113152 : IS_ALIGNED(offset, eb->fs_info->sectorsize))
397 : return type;
398 : }
399 : } else {
400 : ASSERT(is_data == BTRFS_REF_TYPE_ANY);
401 : return type;
402 : }
403 : }
404 :
405 0 : btrfs_print_leaf(eb);
406 0 : btrfs_err(eb->fs_info,
407 : "eb %llu iref 0x%lx invalid extent inline ref type %d",
408 : eb->start, (unsigned long)iref, type);
409 0 : WARN_ON(1);
410 :
411 : return BTRFS_REF_TYPE_INVALID;
412 : }
413 :
414 481407809 : u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
415 : {
416 481407809 : u32 high_crc = ~(u32)0;
417 481407809 : u32 low_crc = ~(u32)0;
418 481407809 : __le64 lenum;
419 :
420 481407809 : lenum = cpu_to_le64(root_objectid);
421 481407809 : high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
422 481466592 : lenum = cpu_to_le64(owner);
423 481466592 : low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
424 481523809 : lenum = cpu_to_le64(offset);
425 481523809 : low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
426 :
427 481556783 : return ((u64)high_crc << 31) ^ (u64)low_crc;
428 : }
429 :
430 234209774 : static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
431 : struct btrfs_extent_data_ref *ref)
432 : {
433 468414881 : return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
434 : btrfs_extent_data_ref_objectid(leaf, ref),
435 : btrfs_extent_data_ref_offset(leaf, ref));
436 : }
437 :
438 235655287 : static int match_extent_data_ref(struct extent_buffer *leaf,
439 : struct btrfs_extent_data_ref *ref,
440 : u64 root_objectid, u64 owner, u64 offset)
441 : {
442 471158707 : if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
443 100344708 : btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
444 : btrfs_extent_data_ref_offset(leaf, ref) != offset)
445 234207874 : return 0;
446 : return 1;
447 : }
448 :
449 223934 : static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
450 : struct btrfs_path *path,
451 : u64 bytenr, u64 parent,
452 : u64 root_objectid,
453 : u64 owner, u64 offset)
454 : {
455 223934 : struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
456 223934 : struct btrfs_key key;
457 223934 : struct btrfs_extent_data_ref *ref;
458 223934 : struct extent_buffer *leaf;
459 223934 : u32 nritems;
460 223934 : int ret;
461 223934 : int recow;
462 223934 : int err = -ENOENT;
463 :
464 223934 : key.objectid = bytenr;
465 223934 : if (parent) {
466 86511 : key.type = BTRFS_SHARED_DATA_REF_KEY;
467 86511 : key.offset = parent;
468 : } else {
469 137423 : key.type = BTRFS_EXTENT_DATA_REF_KEY;
470 137423 : key.offset = hash_extent_data_ref(root_objectid,
471 : owner, offset);
472 : }
473 : again:
474 223934 : recow = 0;
475 223934 : ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
476 223934 : if (ret < 0) {
477 0 : err = ret;
478 0 : goto fail;
479 : }
480 :
481 223934 : if (parent) {
482 86511 : if (!ret)
483 : return 0;
484 0 : goto fail;
485 : }
486 :
487 137423 : leaf = path->nodes[0];
488 137423 : nritems = btrfs_header_nritems(leaf);
489 137423 : while (1) {
490 137423 : if (path->slots[0] >= nritems) {
491 0 : ret = btrfs_next_leaf(root, path);
492 0 : if (ret < 0)
493 0 : err = ret;
494 0 : if (ret)
495 0 : goto fail;
496 :
497 0 : leaf = path->nodes[0];
498 0 : nritems = btrfs_header_nritems(leaf);
499 0 : recow = 1;
500 : }
501 :
502 137423 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
503 137423 : if (key.objectid != bytenr ||
504 137423 : key.type != BTRFS_EXTENT_DATA_REF_KEY)
505 0 : goto fail;
506 :
507 137423 : ref = btrfs_item_ptr(leaf, path->slots[0],
508 : struct btrfs_extent_data_ref);
509 :
510 137423 : if (match_extent_data_ref(leaf, ref, root_objectid,
511 : owner, offset)) {
512 137423 : if (recow) {
513 0 : btrfs_release_path(path);
514 0 : goto again;
515 : }
516 : err = 0;
517 : break;
518 : }
519 0 : path->slots[0]++;
520 : }
521 : fail:
522 : return err;
523 : }
524 :
525 13662563 : static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
526 : struct btrfs_path *path,
527 : u64 bytenr, u64 parent,
528 : u64 root_objectid, u64 owner,
529 : u64 offset, int refs_to_add)
530 : {
531 13662563 : struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
532 13662501 : struct btrfs_key key;
533 13662501 : struct extent_buffer *leaf;
534 13662501 : u32 size;
535 13662501 : u32 num_refs;
536 13662501 : int ret;
537 :
538 13662501 : key.objectid = bytenr;
539 13662501 : if (parent) {
540 658303 : key.type = BTRFS_SHARED_DATA_REF_KEY;
541 658303 : key.offset = parent;
542 658303 : size = sizeof(struct btrfs_shared_data_ref);
543 : } else {
544 13004198 : key.type = BTRFS_EXTENT_DATA_REF_KEY;
545 13004198 : key.offset = hash_extent_data_ref(root_objectid,
546 : owner, offset);
547 13004222 : size = sizeof(struct btrfs_extent_data_ref);
548 : }
549 :
550 13662525 : ret = btrfs_insert_empty_item(trans, root, path, &key, size);
551 13661917 : if (ret && ret != -EEXIST)
552 0 : goto fail;
553 :
554 13661917 : leaf = path->nodes[0];
555 13661917 : if (parent) {
556 658277 : struct btrfs_shared_data_ref *ref;
557 658277 : ref = btrfs_item_ptr(leaf, path->slots[0],
558 : struct btrfs_shared_data_ref);
559 658278 : if (ret == 0) {
560 658246 : btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
561 : } else {
562 32 : num_refs = btrfs_shared_data_ref_count(leaf, ref);
563 32 : num_refs += refs_to_add;
564 32 : btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
565 : }
566 : } else {
567 : struct btrfs_extent_data_ref *ref;
568 13003640 : while (ret == -EEXIST) {
569 3344 : ref = btrfs_item_ptr(leaf, path->slots[0],
570 : struct btrfs_extent_data_ref);
571 3301 : if (match_extent_data_ref(leaf, ref, root_objectid,
572 : owner, offset))
573 : break;
574 0 : btrfs_release_path(path);
575 0 : key.offset++;
576 0 : ret = btrfs_insert_empty_item(trans, root, path, &key,
577 : size);
578 0 : if (ret && ret != -EEXIST)
579 0 : goto fail;
580 :
581 0 : leaf = path->nodes[0];
582 : }
583 13003597 : ref = btrfs_item_ptr(leaf, path->slots[0],
584 : struct btrfs_extent_data_ref);
585 13003893 : if (ret == 0) {
586 13000592 : btrfs_set_extent_data_ref_root(leaf, ref,
587 : root_objectid);
588 13000065 : btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
589 13000085 : btrfs_set_extent_data_ref_offset(leaf, ref, offset);
590 13000062 : btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
591 : } else {
592 3301 : num_refs = btrfs_extent_data_ref_count(leaf, ref);
593 3301 : num_refs += refs_to_add;
594 3301 : btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
595 : }
596 : }
597 13661849 : btrfs_mark_buffer_dirty(leaf);
598 13661849 : ret = 0;
599 13662301 : fail:
600 13662301 : btrfs_release_path(path);
601 13662572 : return ret;
602 : }
603 :
604 223662 : static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
605 : struct btrfs_root *root,
606 : struct btrfs_path *path,
607 : int refs_to_drop)
608 : {
609 223662 : struct btrfs_key key;
610 223662 : struct btrfs_extent_data_ref *ref1 = NULL;
611 223662 : struct btrfs_shared_data_ref *ref2 = NULL;
612 223662 : struct extent_buffer *leaf;
613 223662 : u32 num_refs = 0;
614 223662 : int ret = 0;
615 :
616 223662 : leaf = path->nodes[0];
617 223662 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
618 :
619 223662 : if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
620 137268 : ref1 = btrfs_item_ptr(leaf, path->slots[0],
621 : struct btrfs_extent_data_ref);
622 137268 : num_refs = btrfs_extent_data_ref_count(leaf, ref1);
623 86394 : } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
624 86394 : ref2 = btrfs_item_ptr(leaf, path->slots[0],
625 : struct btrfs_shared_data_ref);
626 86394 : num_refs = btrfs_shared_data_ref_count(leaf, ref2);
627 0 : } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
628 0 : btrfs_print_v0_err(trans->fs_info);
629 0 : btrfs_abort_transaction(trans, -EINVAL);
630 0 : return -EINVAL;
631 : } else {
632 0 : BUG();
633 : }
634 :
635 223662 : BUG_ON(num_refs < refs_to_drop);
636 223662 : num_refs -= refs_to_drop;
637 :
638 223662 : if (num_refs == 0) {
639 220396 : ret = btrfs_del_item(trans, root, path);
640 : } else {
641 3266 : if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
642 3255 : btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
643 11 : else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
644 11 : btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
645 3266 : btrfs_mark_buffer_dirty(leaf);
646 : }
647 : return ret;
648 : }
649 :
650 1015682 : static noinline u32 extent_data_ref_count(struct btrfs_path *path,
651 : struct btrfs_extent_inline_ref *iref)
652 : {
653 1015682 : struct btrfs_key key;
654 1015682 : struct extent_buffer *leaf;
655 1015682 : struct btrfs_extent_data_ref *ref1;
656 1015682 : struct btrfs_shared_data_ref *ref2;
657 1015682 : u32 num_refs = 0;
658 1015682 : int type;
659 :
660 1015682 : leaf = path->nodes[0];
661 1015682 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
662 :
663 1015682 : BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
664 1015682 : if (iref) {
665 : /*
666 : * If type is invalid, we should have bailed out earlier than
667 : * this call.
668 : */
669 1015410 : type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
670 1015410 : ASSERT(type != BTRFS_REF_TYPE_INVALID);
671 1015410 : if (type == BTRFS_EXTENT_DATA_REF_KEY) {
672 991703 : ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
673 991703 : num_refs = btrfs_extent_data_ref_count(leaf, ref1);
674 : } else {
675 23707 : ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
676 23707 : num_refs = btrfs_shared_data_ref_count(leaf, ref2);
677 : }
678 272 : } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
679 155 : ref1 = btrfs_item_ptr(leaf, path->slots[0],
680 : struct btrfs_extent_data_ref);
681 155 : num_refs = btrfs_extent_data_ref_count(leaf, ref1);
682 117 : } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
683 117 : ref2 = btrfs_item_ptr(leaf, path->slots[0],
684 : struct btrfs_shared_data_ref);
685 117 : num_refs = btrfs_shared_data_ref_count(leaf, ref2);
686 : } else {
687 0 : WARN_ON(1);
688 : }
689 1015682 : return num_refs;
690 : }
691 :
692 0 : static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
693 : struct btrfs_path *path,
694 : u64 bytenr, u64 parent,
695 : u64 root_objectid)
696 : {
697 0 : struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
698 0 : struct btrfs_key key;
699 0 : int ret;
700 :
701 0 : key.objectid = bytenr;
702 0 : if (parent) {
703 0 : key.type = BTRFS_SHARED_BLOCK_REF_KEY;
704 0 : key.offset = parent;
705 : } else {
706 0 : key.type = BTRFS_TREE_BLOCK_REF_KEY;
707 0 : key.offset = root_objectid;
708 : }
709 :
710 0 : ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
711 0 : if (ret > 0)
712 0 : ret = -ENOENT;
713 0 : return ret;
714 : }
715 :
716 0 : static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
717 : struct btrfs_path *path,
718 : u64 bytenr, u64 parent,
719 : u64 root_objectid)
720 : {
721 0 : struct btrfs_root *root = btrfs_extent_root(trans->fs_info, bytenr);
722 0 : struct btrfs_key key;
723 0 : int ret;
724 :
725 0 : key.objectid = bytenr;
726 0 : if (parent) {
727 0 : key.type = BTRFS_SHARED_BLOCK_REF_KEY;
728 0 : key.offset = parent;
729 : } else {
730 0 : key.type = BTRFS_TREE_BLOCK_REF_KEY;
731 0 : key.offset = root_objectid;
732 : }
733 :
734 0 : ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
735 0 : btrfs_release_path(path);
736 0 : return ret;
737 : }
738 :
739 : static inline int extent_ref_type(u64 parent, u64 owner)
740 : {
741 65705321 : int type;
742 65705321 : if (owner < BTRFS_FIRST_FREE_OBJECTID) {
743 26245367 : if (parent > 0)
744 : type = BTRFS_SHARED_BLOCK_REF_KEY;
745 : else
746 5625412 : type = BTRFS_TREE_BLOCK_REF_KEY;
747 : } else {
748 39459954 : if (parent > 0)
749 : type = BTRFS_SHARED_DATA_REF_KEY;
750 : else
751 16066948 : type = BTRFS_EXTENT_DATA_REF_KEY;
752 : }
753 65705321 : return type;
754 : }
755 :
756 19924655 : static int find_next_key(struct btrfs_path *path, int level,
757 : struct btrfs_key *key)
758 :
759 : {
760 21409883 : for (; level < BTRFS_MAX_LEVEL; level++) {
761 21409883 : if (!path->nodes[level])
762 : break;
763 20975515 : if (path->slots[level] + 1 >=
764 : btrfs_header_nritems(path->nodes[level]))
765 1485228 : continue;
766 19490287 : if (level == 0)
767 14856565 : btrfs_item_key_to_cpu(path->nodes[level], key,
768 : path->slots[level] + 1);
769 : else
770 4633722 : btrfs_node_key_to_cpu(path->nodes[level], key,
771 : path->slots[level] + 1);
772 : return 0;
773 : }
774 : return 1;
775 : }
776 :
777 : /*
778 : * look for inline back ref. if back ref is found, *ref_ret is set
779 : * to the address of inline back ref, and 0 is returned.
780 : *
781 : * if back ref isn't found, *ref_ret is set to the address where it
782 : * should be inserted, and -ENOENT is returned.
783 : *
784 : * if insert is true and there are too many inline back refs, the path
785 : * points to the extent item, and -EAGAIN is returned.
786 : *
787 : * NOTE: inline back refs are ordered in the same way that back ref
788 : * items in the tree are ordered.
789 : */
790 : static noinline_for_stack
791 50673653 : int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
792 : struct btrfs_path *path,
793 : struct btrfs_extent_inline_ref **ref_ret,
794 : u64 bytenr, u64 num_bytes,
795 : u64 parent, u64 root_objectid,
796 : u64 owner, u64 offset, int insert)
797 : {
798 50673653 : struct btrfs_fs_info *fs_info = trans->fs_info;
799 50673653 : struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
800 50673494 : struct btrfs_key key;
801 50673494 : struct extent_buffer *leaf;
802 50673494 : struct btrfs_extent_item *ei;
803 50673494 : struct btrfs_extent_inline_ref *iref;
804 50673494 : u64 flags;
805 50673494 : u64 item_size;
806 50673494 : unsigned long ptr;
807 50673494 : unsigned long end;
808 50673494 : int extra_size;
809 50673494 : int type;
810 50673494 : int want;
811 50673494 : int ret;
812 50673494 : int err = 0;
813 50673494 : bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
814 50673494 : int needed;
815 :
816 50673494 : key.objectid = bytenr;
817 50673494 : key.type = BTRFS_EXTENT_ITEM_KEY;
818 50673494 : key.offset = num_bytes;
819 :
820 50673494 : want = extent_ref_type(parent, owner);
821 50673494 : if (insert) {
822 28764274 : extra_size = btrfs_extent_inline_ref_size(want);
823 28764274 : path->search_for_extension = 1;
824 28764274 : path->keep_locks = 1;
825 : } else
826 : extra_size = -1;
827 :
828 : /*
829 : * Owner is our level, so we can just add one to get the level for the
830 : * block we are interested in.
831 : */
832 50673494 : if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
833 19819096 : key.type = BTRFS_METADATA_ITEM_KEY;
834 19819096 : key.offset = owner;
835 : }
836 :
837 30854398 : again:
838 50673494 : ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
839 50674130 : if (ret < 0) {
840 0 : err = ret;
841 0 : goto out;
842 : }
843 :
844 : /*
845 : * We may be a newly converted file system which still has the old fat
846 : * extent entries for metadata, so try and see if we have one of those.
847 : */
848 50674130 : if (ret > 0 && skinny_metadata) {
849 0 : skinny_metadata = false;
850 0 : if (path->slots[0]) {
851 0 : path->slots[0]--;
852 0 : btrfs_item_key_to_cpu(path->nodes[0], &key,
853 : path->slots[0]);
854 0 : if (key.objectid == bytenr &&
855 0 : key.type == BTRFS_EXTENT_ITEM_KEY &&
856 0 : key.offset == num_bytes)
857 : ret = 0;
858 : }
859 0 : if (ret) {
860 0 : key.objectid = bytenr;
861 0 : key.type = BTRFS_EXTENT_ITEM_KEY;
862 0 : key.offset = num_bytes;
863 0 : btrfs_release_path(path);
864 0 : goto again;
865 : }
866 : }
867 :
868 50674130 : if (ret && !insert) {
869 0 : err = -ENOENT;
870 0 : goto out;
871 50674130 : } else if (WARN_ON(ret)) {
872 0 : err = -EIO;
873 0 : goto out;
874 : }
875 :
876 50674130 : leaf = path->nodes[0];
877 50674130 : item_size = btrfs_item_size(leaf, path->slots[0]);
878 50674125 : if (unlikely(item_size < sizeof(*ei))) {
879 0 : err = -EINVAL;
880 0 : btrfs_print_v0_err(fs_info);
881 0 : btrfs_abort_transaction(trans, err);
882 0 : goto out;
883 : }
884 :
885 50674125 : ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
886 50674126 : flags = btrfs_extent_flags(leaf, ei);
887 :
888 50674127 : ptr = (unsigned long)(ei + 1);
889 50674127 : end = (unsigned long)ei + item_size;
890 :
891 50674127 : if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
892 0 : ptr += sizeof(struct btrfs_tree_block_info);
893 0 : BUG_ON(ptr > end);
894 : }
895 :
896 50674127 : if (owner >= BTRFS_FIRST_FREE_OBJECTID)
897 : needed = BTRFS_REF_TYPE_DATA;
898 : else
899 19819153 : needed = BTRFS_REF_TYPE_BLOCK;
900 :
901 50674127 : err = -ENOENT;
902 310181380 : while (1) {
903 310181380 : if (ptr >= end) {
904 2163760 : if (ptr > end) {
905 0 : err = -EUCLEAN;
906 0 : btrfs_print_leaf(path->nodes[0]);
907 0 : btrfs_crit(fs_info,
908 : "overrun extent record at slot %d while looking for inline extent for root %llu owner %llu offset %llu parent %llu",
909 : path->slots[0], root_objectid, owner, offset, parent);
910 : }
911 : break;
912 : }
913 308017620 : iref = (struct btrfs_extent_inline_ref *)ptr;
914 308017620 : type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
915 307978301 : if (type == BTRFS_REF_TYPE_INVALID) {
916 0 : err = -EUCLEAN;
917 0 : goto out;
918 : }
919 :
920 307978301 : if (want < type)
921 : break;
922 307900827 : if (want > type) {
923 5598734 : ptr += btrfs_extent_inline_ref_size(type);
924 5598734 : continue;
925 : }
926 :
927 302302093 : if (type == BTRFS_EXTENT_DATA_REF_KEY) {
928 235517461 : struct btrfs_extent_data_ref *dref;
929 235517461 : dref = (struct btrfs_extent_data_ref *)(&iref->offset);
930 235517461 : if (match_extent_data_ref(leaf, dref, root_objectid,
931 : owner, offset)) {
932 : err = 0;
933 : break;
934 : }
935 468488810 : if (hash_extent_data_ref_item(leaf, dref) <
936 234273108 : hash_extent_data_ref(root_objectid, owner, offset))
937 : break;
938 : } else {
939 66784632 : u64 ref_offset;
940 66784632 : ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
941 66784372 : if (parent > 0) {
942 61262104 : if (parent == ref_offset) {
943 : err = 0;
944 : break;
945 : }
946 46133926 : if (ref_offset < parent)
947 : break;
948 : } else {
949 5522268 : if (root_objectid == ref_offset) {
950 : err = 0;
951 : break;
952 : }
953 175977 : if (ref_offset < root_objectid)
954 : break;
955 : }
956 : }
957 499481773 : ptr += btrfs_extent_inline_ref_size(type);
958 : }
959 50673829 : if (err == -ENOENT && insert) {
960 28694069 : if (item_size + extra_size >=
961 28694069 : BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
962 13229742 : err = -EAGAIN;
963 13229742 : goto out;
964 : }
965 : /*
966 : * To add new inline back ref, we have to make sure
967 : * there is no corresponding back ref item.
968 : * For simplicity, we just do not add new inline back
969 : * ref if there is any kind of item for this block
970 : */
971 15464327 : if (find_next_key(path, 0, &key) == 0 &&
972 15030503 : key.objectid == bytenr &&
973 436318 : key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
974 432512 : err = -EAGAIN;
975 432512 : goto out;
976 : }
977 : }
978 37011574 : *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
979 50673828 : out:
980 50673828 : if (insert) {
981 28764398 : path->keep_locks = 0;
982 28764398 : path->search_for_extension = 0;
983 28764398 : btrfs_unlock_up_safe(path, 1);
984 : }
985 50673866 : return err;
986 : }
987 :
988 : /*
989 : * helper to add new inline back ref
990 : */
991 : static noinline_for_stack
992 15031827 : void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
993 : struct btrfs_path *path,
994 : struct btrfs_extent_inline_ref *iref,
995 : u64 parent, u64 root_objectid,
996 : u64 owner, u64 offset, int refs_to_add,
997 : struct btrfs_delayed_extent_op *extent_op)
998 : {
999 15031827 : struct extent_buffer *leaf;
1000 15031827 : struct btrfs_extent_item *ei;
1001 15031827 : unsigned long ptr;
1002 15031827 : unsigned long end;
1003 15031827 : unsigned long item_offset;
1004 15031827 : u64 refs;
1005 15031827 : int size;
1006 15031827 : int type;
1007 :
1008 15031827 : leaf = path->nodes[0];
1009 15031827 : ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1010 15031827 : item_offset = (unsigned long)iref - (unsigned long)ei;
1011 :
1012 15031827 : type = extent_ref_type(parent, owner);
1013 15031827 : size = btrfs_extent_inline_ref_size(type);
1014 :
1015 15031827 : btrfs_extend_item(path, size);
1016 :
1017 15031804 : ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1018 15031799 : refs = btrfs_extent_refs(leaf, ei);
1019 15031793 : refs += refs_to_add;
1020 15031793 : btrfs_set_extent_refs(leaf, ei, refs);
1021 15031794 : if (extent_op)
1022 22148 : __run_delayed_extent_op(extent_op, leaf, ei);
1023 :
1024 15031794 : ptr = (unsigned long)ei + item_offset;
1025 15031794 : end = (unsigned long)ei + btrfs_item_size(leaf, path->slots[0]);
1026 15031804 : if (ptr < end - size)
1027 13203789 : memmove_extent_buffer(leaf, ptr + size, ptr,
1028 : end - size - ptr);
1029 :
1030 15031798 : iref = (struct btrfs_extent_inline_ref *)ptr;
1031 15031798 : btrfs_set_extent_inline_ref_type(leaf, iref, type);
1032 15031797 : if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1033 822193 : struct btrfs_extent_data_ref *dref;
1034 822193 : dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1035 822193 : btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1036 822192 : btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1037 822193 : btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1038 822189 : btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1039 14209604 : } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1040 7783335 : struct btrfs_shared_data_ref *sref;
1041 7783335 : sref = (struct btrfs_shared_data_ref *)(iref + 1);
1042 7783335 : btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1043 7783333 : btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1044 6426269 : } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1045 6286705 : btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1046 : } else {
1047 139564 : btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1048 : }
1049 15031801 : btrfs_mark_buffer_dirty(leaf);
1050 15031814 : }
1051 :
1052 21909246 : static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1053 : struct btrfs_path *path,
1054 : struct btrfs_extent_inline_ref **ref_ret,
1055 : u64 bytenr, u64 num_bytes, u64 parent,
1056 : u64 root_objectid, u64 owner, u64 offset)
1057 : {
1058 21909246 : int ret;
1059 :
1060 21909246 : ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1061 : num_bytes, parent, root_objectid,
1062 : owner, offset, 0);
1063 21909286 : if (ret != -ENOENT)
1064 : return ret;
1065 :
1066 223934 : btrfs_release_path(path);
1067 223934 : *ref_ret = NULL;
1068 :
1069 223934 : if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1070 0 : ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1071 : root_objectid);
1072 : } else {
1073 223934 : ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1074 : root_objectid, owner, offset);
1075 : }
1076 : return ret;
1077 : }
1078 :
1079 : /*
1080 : * helper to update/remove inline back ref
1081 : */
1082 : static noinline_for_stack
1083 13682819 : void update_inline_extent_backref(struct btrfs_path *path,
1084 : struct btrfs_extent_inline_ref *iref,
1085 : int refs_to_mod,
1086 : struct btrfs_delayed_extent_op *extent_op)
1087 : {
1088 13682819 : struct extent_buffer *leaf = path->nodes[0];
1089 13682819 : struct btrfs_extent_item *ei;
1090 13682819 : struct btrfs_extent_data_ref *dref = NULL;
1091 13682819 : struct btrfs_shared_data_ref *sref = NULL;
1092 13682819 : unsigned long ptr;
1093 13682819 : unsigned long end;
1094 13682819 : u32 item_size;
1095 13682819 : int size;
1096 13682819 : int type;
1097 13682819 : u64 refs;
1098 :
1099 13682819 : ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1100 13682819 : refs = btrfs_extent_refs(leaf, ei);
1101 27365636 : WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1102 13682818 : refs += refs_to_mod;
1103 13682818 : btrfs_set_extent_refs(leaf, ei, refs);
1104 13682818 : if (extent_op)
1105 0 : __run_delayed_extent_op(extent_op, leaf, ei);
1106 :
1107 : /*
1108 : * If type is invalid, we should have bailed out after
1109 : * lookup_inline_extent_backref().
1110 : */
1111 13682818 : type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1112 13682818 : ASSERT(type != BTRFS_REF_TYPE_INVALID);
1113 :
1114 13682818 : if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1115 289610 : dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1116 289610 : refs = btrfs_extent_data_ref_count(leaf, dref);
1117 13393208 : } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1118 7057879 : sref = (struct btrfs_shared_data_ref *)(iref + 1);
1119 7057879 : refs = btrfs_shared_data_ref_count(leaf, sref);
1120 : } else {
1121 6335329 : refs = 1;
1122 6335329 : BUG_ON(refs_to_mod != -1);
1123 : }
1124 :
1125 13682819 : BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1126 13682819 : refs += refs_to_mod;
1127 :
1128 13682819 : if (refs > 0) {
1129 89086 : if (type == BTRFS_EXTENT_DATA_REF_KEY)
1130 89085 : btrfs_set_extent_data_ref_count(leaf, dref, refs);
1131 : else
1132 1 : btrfs_set_shared_data_ref_count(leaf, sref, refs);
1133 : } else {
1134 13593733 : size = btrfs_extent_inline_ref_size(type);
1135 13593733 : item_size = btrfs_item_size(leaf, path->slots[0]);
1136 13593733 : ptr = (unsigned long)iref;
1137 13593733 : end = (unsigned long)ei + item_size;
1138 13593733 : if (ptr + size < end)
1139 5019234 : memmove_extent_buffer(leaf, ptr, ptr + size,
1140 5019234 : end - ptr - size);
1141 13593733 : item_size -= size;
1142 13593733 : btrfs_truncate_item(path, item_size, 1);
1143 : }
1144 13682819 : btrfs_mark_buffer_dirty(leaf);
1145 13682819 : }
1146 :
1147 : static noinline_for_stack
1148 28764275 : int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1149 : struct btrfs_path *path,
1150 : u64 bytenr, u64 num_bytes, u64 parent,
1151 : u64 root_objectid, u64 owner,
1152 : u64 offset, int refs_to_add,
1153 : struct btrfs_delayed_extent_op *extent_op)
1154 : {
1155 28764275 : struct btrfs_extent_inline_ref *iref;
1156 28764275 : int ret;
1157 :
1158 28764275 : ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1159 : num_bytes, parent, root_objectid,
1160 : owner, offset, 1);
1161 28764548 : if (ret == 0) {
1162 : /*
1163 : * We're adding refs to a tree block we already own, this
1164 : * should not happen at all.
1165 : */
1166 70431 : if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1167 0 : btrfs_print_leaf(path->nodes[0]);
1168 0 : btrfs_crit(trans->fs_info,
1169 : "adding refs to an existing tree ref, bytenr %llu num_bytes %llu root_objectid %llu slot %u",
1170 : bytenr, num_bytes, root_objectid, path->slots[0]);
1171 0 : return -EUCLEAN;
1172 : }
1173 70431 : update_inline_extent_backref(path, iref, refs_to_add, extent_op);
1174 28694117 : } else if (ret == -ENOENT) {
1175 15031829 : setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1176 : root_objectid, owner, offset,
1177 : refs_to_add, extent_op);
1178 15031829 : ret = 0;
1179 : }
1180 : return ret;
1181 : }
1182 :
1183 13836050 : static int remove_extent_backref(struct btrfs_trans_handle *trans,
1184 : struct btrfs_root *root,
1185 : struct btrfs_path *path,
1186 : struct btrfs_extent_inline_ref *iref,
1187 : int refs_to_drop, int is_data)
1188 : {
1189 13836050 : int ret = 0;
1190 :
1191 13836050 : BUG_ON(!is_data && refs_to_drop != 1);
1192 13836050 : if (iref)
1193 13612388 : update_inline_extent_backref(path, iref, -refs_to_drop, NULL);
1194 223662 : else if (is_data)
1195 223662 : ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1196 : else
1197 0 : ret = btrfs_del_item(trans, root, path);
1198 13836050 : return ret;
1199 : }
1200 :
1201 36347 : static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1202 : u64 *discarded_bytes)
1203 : {
1204 36347 : int j, ret = 0;
1205 36347 : u64 bytes_left, end;
1206 36347 : u64 aligned_start = ALIGN(start, 1 << SECTOR_SHIFT);
1207 :
1208 36347 : if (WARN_ON(start != aligned_start)) {
1209 0 : len -= aligned_start - start;
1210 0 : len = round_down(len, 1 << SECTOR_SHIFT);
1211 0 : start = aligned_start;
1212 : }
1213 :
1214 36347 : *discarded_bytes = 0;
1215 :
1216 36347 : if (!len)
1217 : return 0;
1218 :
1219 36347 : end = start + len;
1220 36347 : bytes_left = len;
1221 :
1222 : /* Skip any superblocks on this device. */
1223 145388 : for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1224 109041 : u64 sb_start = btrfs_sb_offset(j);
1225 109041 : u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1226 109041 : u64 size = sb_start - start;
1227 :
1228 109041 : if (!in_range(sb_start, start, bytes_left) &&
1229 109040 : !in_range(sb_end, start, bytes_left) &&
1230 71189 : !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1231 109040 : continue;
1232 :
1233 : /*
1234 : * Superblock spans beginning of range. Adjust start and
1235 : * try again.
1236 : */
1237 1 : if (sb_start <= start) {
1238 0 : start += sb_end - start;
1239 0 : if (start > end) {
1240 : bytes_left = 0;
1241 : break;
1242 : }
1243 0 : bytes_left = end - start;
1244 0 : continue;
1245 : }
1246 :
1247 1 : if (size) {
1248 1 : ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1249 : size >> SECTOR_SHIFT,
1250 : GFP_NOFS);
1251 1 : if (!ret)
1252 1 : *discarded_bytes += size;
1253 0 : else if (ret != -EOPNOTSUPP)
1254 0 : return ret;
1255 : }
1256 :
1257 1 : start = sb_end;
1258 1 : if (start > end) {
1259 : bytes_left = 0;
1260 : break;
1261 : }
1262 1 : bytes_left = end - start;
1263 : }
1264 :
1265 36347 : if (bytes_left) {
1266 36347 : ret = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT,
1267 : bytes_left >> SECTOR_SHIFT,
1268 : GFP_NOFS);
1269 36347 : if (!ret)
1270 36345 : *discarded_bytes += bytes_left;
1271 : }
1272 : return ret;
1273 : }
1274 :
1275 36316 : static int do_discard_extent(struct btrfs_discard_stripe *stripe, u64 *bytes)
1276 : {
1277 36316 : struct btrfs_device *dev = stripe->dev;
1278 36316 : struct btrfs_fs_info *fs_info = dev->fs_info;
1279 36316 : struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1280 36316 : u64 phys = stripe->physical;
1281 36316 : u64 len = stripe->length;
1282 36316 : u64 discarded = 0;
1283 36316 : int ret = 0;
1284 :
1285 : /* Zone reset on a zoned filesystem */
1286 36316 : if (btrfs_can_zone_reset(dev, phys, len)) {
1287 0 : u64 src_disc;
1288 :
1289 0 : ret = btrfs_reset_device_zone(dev, phys, len, &discarded);
1290 0 : if (ret)
1291 : goto out;
1292 :
1293 0 : if (!btrfs_dev_replace_is_ongoing(dev_replace) ||
1294 0 : dev != dev_replace->srcdev)
1295 0 : goto out;
1296 :
1297 : src_disc = discarded;
1298 :
1299 : /* Send to replace target as well */
1300 : ret = btrfs_reset_device_zone(dev_replace->tgtdev, phys, len,
1301 : &discarded);
1302 : discarded += src_disc;
1303 36316 : } else if (bdev_max_discard_sectors(stripe->dev->bdev)) {
1304 36316 : ret = btrfs_issue_discard(dev->bdev, phys, len, &discarded);
1305 : } else {
1306 : ret = 0;
1307 : *bytes = 0;
1308 : }
1309 :
1310 36316 : out:
1311 36316 : *bytes = discarded;
1312 36316 : return ret;
1313 : }
1314 :
1315 29039 : int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1316 : u64 num_bytes, u64 *actual_bytes)
1317 : {
1318 29039 : int ret = 0;
1319 29039 : u64 discarded_bytes = 0;
1320 29039 : u64 end = bytenr + num_bytes;
1321 29039 : u64 cur = bytenr;
1322 :
1323 : /*
1324 : * Avoid races with device replace and make sure the devices in the
1325 : * stripes don't go away while we are discarding.
1326 : */
1327 29039 : btrfs_bio_counter_inc_blocked(fs_info);
1328 58077 : while (cur < end) {
1329 29040 : struct btrfs_discard_stripe *stripes;
1330 29040 : unsigned int num_stripes;
1331 29040 : int i;
1332 :
1333 29040 : num_bytes = end - cur;
1334 29040 : stripes = btrfs_map_discard(fs_info, cur, &num_bytes, &num_stripes);
1335 29040 : if (IS_ERR(stripes)) {
1336 0 : ret = PTR_ERR(stripes);
1337 0 : if (ret == -EOPNOTSUPP)
1338 0 : ret = 0;
1339 2 : break;
1340 : }
1341 :
1342 65354 : for (i = 0; i < num_stripes; i++) {
1343 36316 : struct btrfs_discard_stripe *stripe = stripes + i;
1344 36316 : u64 bytes;
1345 :
1346 36316 : if (!stripe->dev->bdev) {
1347 0 : ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1348 0 : continue;
1349 : }
1350 :
1351 36316 : if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
1352 : &stripe->dev->dev_state))
1353 0 : continue;
1354 :
1355 36316 : ret = do_discard_extent(stripe, &bytes);
1356 36316 : if (ret) {
1357 : /*
1358 : * Keep going if discard is not supported by the
1359 : * device.
1360 : */
1361 2 : if (ret != -EOPNOTSUPP)
1362 : break;
1363 : ret = 0;
1364 : } else {
1365 36314 : discarded_bytes += bytes;
1366 : }
1367 : }
1368 29040 : kfree(stripes);
1369 29040 : if (ret)
1370 : break;
1371 29038 : cur += num_bytes;
1372 : }
1373 29039 : btrfs_bio_counter_dec(fs_info);
1374 29039 : if (actual_bytes)
1375 29001 : *actual_bytes = discarded_bytes;
1376 29039 : return ret;
1377 : }
1378 :
1379 : /* Can return -ENOMEM */
1380 34293764 : int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1381 : struct btrfs_ref *generic_ref)
1382 : {
1383 34293764 : struct btrfs_fs_info *fs_info = trans->fs_info;
1384 34293764 : int ret;
1385 :
1386 34293764 : ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1387 : generic_ref->action);
1388 34293764 : BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1389 : generic_ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID);
1390 :
1391 34293764 : if (generic_ref->type == BTRFS_REF_METADATA)
1392 7175378 : ret = btrfs_add_delayed_tree_ref(trans, generic_ref, NULL);
1393 : else
1394 27118386 : ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0);
1395 :
1396 34293810 : btrfs_ref_tree_mod(fs_info, generic_ref);
1397 :
1398 34293810 : return ret;
1399 : }
1400 :
1401 : /*
1402 : * __btrfs_inc_extent_ref - insert backreference for a given extent
1403 : *
1404 : * The counterpart is in __btrfs_free_extent(), with examples and more details
1405 : * how it works.
1406 : *
1407 : * @trans: Handle of transaction
1408 : *
1409 : * @node: The delayed ref node used to get the bytenr/length for
1410 : * extent whose references are incremented.
1411 : *
1412 : * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1413 : * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1414 : * bytenr of the parent block. Since new extents are always
1415 : * created with indirect references, this will only be the case
1416 : * when relocating a shared extent. In that case, root_objectid
1417 : * will be BTRFS_TREE_RELOC_OBJECTID. Otherwise, parent must
1418 : * be 0
1419 : *
1420 : * @root_objectid: The id of the root where this modification has originated,
1421 : * this can be either one of the well-known metadata trees or
1422 : * the subvolume id which references this extent.
1423 : *
1424 : * @owner: For data extents it is the inode number of the owning file.
1425 : * For metadata extents this parameter holds the level in the
1426 : * tree of the extent.
1427 : *
1428 : * @offset: For metadata extents the offset is ignored and is currently
1429 : * always passed as 0. For data extents it is the fileoffset
1430 : * this extent belongs to.
1431 : *
1432 : * @refs_to_add Number of references to add
1433 : *
1434 : * @extent_op Pointer to a structure, holding information necessary when
1435 : * updating a tree block's flags
1436 : *
1437 : */
1438 28764648 : static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1439 : struct btrfs_delayed_ref_node *node,
1440 : u64 parent, u64 root_objectid,
1441 : u64 owner, u64 offset, int refs_to_add,
1442 : struct btrfs_delayed_extent_op *extent_op)
1443 : {
1444 28764648 : struct btrfs_path *path;
1445 28764648 : struct extent_buffer *leaf;
1446 28764648 : struct btrfs_extent_item *item;
1447 28764648 : struct btrfs_key key;
1448 28764648 : u64 bytenr = node->bytenr;
1449 28764648 : u64 num_bytes = node->num_bytes;
1450 28764648 : u64 refs;
1451 28764648 : int ret;
1452 :
1453 28764648 : path = btrfs_alloc_path();
1454 28764324 : if (!path)
1455 : return -ENOMEM;
1456 :
1457 : /* this will setup the path even if it fails to insert the back ref */
1458 28764324 : ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1459 : parent, root_objectid, owner,
1460 : offset, refs_to_add, extent_op);
1461 28763695 : if ((ret < 0 && ret != -EAGAIN) || !ret)
1462 15102242 : goto out;
1463 :
1464 : /*
1465 : * Ok we had -EAGAIN which means we didn't have space to insert and
1466 : * inline extent ref, so just update the reference count and add a
1467 : * normal backref.
1468 : */
1469 13661453 : leaf = path->nodes[0];
1470 13661453 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1471 13661656 : item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1472 13661692 : refs = btrfs_extent_refs(leaf, item);
1473 13661608 : btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1474 13661405 : if (extent_op)
1475 0 : __run_delayed_extent_op(extent_op, leaf, item);
1476 :
1477 13661405 : btrfs_mark_buffer_dirty(leaf);
1478 13662033 : btrfs_release_path(path);
1479 :
1480 : /* now insert the actual backref */
1481 13662558 : if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1482 0 : BUG_ON(refs_to_add != 1);
1483 0 : ret = insert_tree_block_ref(trans, path, bytenr, parent,
1484 : root_objectid);
1485 : } else {
1486 13662558 : ret = insert_extent_data_ref(trans, path, bytenr, parent,
1487 : root_objectid, owner, offset,
1488 : refs_to_add);
1489 : }
1490 13662575 : if (ret)
1491 0 : btrfs_abort_transaction(trans, ret);
1492 13662575 : out:
1493 28764817 : btrfs_free_path(path);
1494 28764817 : return ret;
1495 : }
1496 :
1497 32774020 : static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1498 : struct btrfs_delayed_ref_node *node,
1499 : struct btrfs_delayed_extent_op *extent_op,
1500 : bool insert_reserved)
1501 : {
1502 32774020 : int ret = 0;
1503 32774020 : struct btrfs_delayed_data_ref *ref;
1504 32774020 : struct btrfs_key ins;
1505 32774020 : u64 parent = 0;
1506 32774020 : u64 ref_root = 0;
1507 32774020 : u64 flags = 0;
1508 :
1509 32774020 : ins.objectid = node->bytenr;
1510 32774020 : ins.offset = node->num_bytes;
1511 32774020 : ins.type = BTRFS_EXTENT_ITEM_KEY;
1512 :
1513 32774020 : ref = btrfs_delayed_node_to_data_ref(node);
1514 32774020 : trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
1515 :
1516 32773972 : if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1517 15609668 : parent = ref->parent;
1518 32773972 : ref_root = ref->root;
1519 :
1520 32773972 : if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1521 1919198 : if (extent_op)
1522 0 : flags |= extent_op->flags_to_set;
1523 1919198 : ret = alloc_reserved_file_extent(trans, parent, ref_root,
1524 : flags, ref->objectid,
1525 : ref->offset, &ins,
1526 : node->ref_mod);
1527 30854774 : } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1528 22338393 : ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1529 : ref->objectid, ref->offset,
1530 : node->ref_mod, extent_op);
1531 8516381 : } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1532 8516381 : ret = __btrfs_free_extent(trans, node, parent,
1533 : ref_root, ref->objectid,
1534 : ref->offset, node->ref_mod,
1535 : extent_op);
1536 : } else {
1537 0 : BUG();
1538 : }
1539 32774042 : return ret;
1540 : }
1541 :
1542 71472 : static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1543 : struct extent_buffer *leaf,
1544 : struct btrfs_extent_item *ei)
1545 : {
1546 71472 : u64 flags = btrfs_extent_flags(leaf, ei);
1547 71472 : if (extent_op->update_flags) {
1548 71472 : flags |= extent_op->flags_to_set;
1549 71472 : btrfs_set_extent_flags(leaf, ei, flags);
1550 : }
1551 :
1552 71472 : if (extent_op->update_key) {
1553 0 : struct btrfs_tree_block_info *bi;
1554 0 : BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1555 0 : bi = (struct btrfs_tree_block_info *)(ei + 1);
1556 0 : btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1557 : }
1558 71472 : }
1559 :
1560 7234 : static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1561 : struct btrfs_delayed_ref_head *head,
1562 : struct btrfs_delayed_extent_op *extent_op)
1563 : {
1564 7234 : struct btrfs_fs_info *fs_info = trans->fs_info;
1565 7234 : struct btrfs_root *root;
1566 7234 : struct btrfs_key key;
1567 7234 : struct btrfs_path *path;
1568 7234 : struct btrfs_extent_item *ei;
1569 7234 : struct extent_buffer *leaf;
1570 7234 : u32 item_size;
1571 7234 : int ret;
1572 7234 : int err = 0;
1573 7234 : int metadata = 1;
1574 :
1575 7234 : if (TRANS_ABORTED(trans))
1576 : return 0;
1577 :
1578 7234 : if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1579 0 : metadata = 0;
1580 :
1581 7234 : path = btrfs_alloc_path();
1582 7234 : if (!path)
1583 : return -ENOMEM;
1584 :
1585 7234 : key.objectid = head->bytenr;
1586 :
1587 7234 : if (metadata) {
1588 7234 : key.type = BTRFS_METADATA_ITEM_KEY;
1589 7234 : key.offset = extent_op->level;
1590 : } else {
1591 0 : key.type = BTRFS_EXTENT_ITEM_KEY;
1592 0 : key.offset = head->num_bytes;
1593 : }
1594 :
1595 7234 : root = btrfs_extent_root(fs_info, key.objectid);
1596 7234 : again:
1597 7234 : ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1598 7234 : if (ret < 0) {
1599 0 : err = ret;
1600 0 : goto out;
1601 : }
1602 7234 : if (ret > 0) {
1603 0 : if (metadata) {
1604 0 : if (path->slots[0] > 0) {
1605 0 : path->slots[0]--;
1606 0 : btrfs_item_key_to_cpu(path->nodes[0], &key,
1607 : path->slots[0]);
1608 0 : if (key.objectid == head->bytenr &&
1609 0 : key.type == BTRFS_EXTENT_ITEM_KEY &&
1610 0 : key.offset == head->num_bytes)
1611 : ret = 0;
1612 : }
1613 0 : if (ret > 0) {
1614 0 : btrfs_release_path(path);
1615 0 : metadata = 0;
1616 :
1617 0 : key.objectid = head->bytenr;
1618 0 : key.offset = head->num_bytes;
1619 0 : key.type = BTRFS_EXTENT_ITEM_KEY;
1620 0 : goto again;
1621 : }
1622 : } else {
1623 0 : err = -EIO;
1624 0 : goto out;
1625 : }
1626 : }
1627 :
1628 7234 : leaf = path->nodes[0];
1629 7234 : item_size = btrfs_item_size(leaf, path->slots[0]);
1630 :
1631 7234 : if (unlikely(item_size < sizeof(*ei))) {
1632 0 : err = -EINVAL;
1633 0 : btrfs_print_v0_err(fs_info);
1634 0 : btrfs_abort_transaction(trans, err);
1635 0 : goto out;
1636 : }
1637 :
1638 7234 : ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1639 7234 : __run_delayed_extent_op(extent_op, leaf, ei);
1640 :
1641 7234 : btrfs_mark_buffer_dirty(leaf);
1642 7234 : out:
1643 7234 : btrfs_free_path(path);
1644 7234 : return err;
1645 : }
1646 :
1647 27150216 : static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1648 : struct btrfs_delayed_ref_node *node,
1649 : struct btrfs_delayed_extent_op *extent_op,
1650 : bool insert_reserved)
1651 : {
1652 27150216 : int ret = 0;
1653 27150216 : struct btrfs_delayed_tree_ref *ref;
1654 27150216 : u64 parent = 0;
1655 27150216 : u64 ref_root = 0;
1656 :
1657 27150216 : ref = btrfs_delayed_node_to_tree_ref(node);
1658 27150216 : trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
1659 :
1660 27150247 : if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1661 16188055 : parent = ref->parent;
1662 27150247 : ref_root = ref->root;
1663 :
1664 27150247 : if (node->ref_mod != 1) {
1665 0 : btrfs_err(trans->fs_info,
1666 : "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
1667 : node->bytenr, node->ref_mod, node->action, ref_root,
1668 : parent);
1669 0 : return -EIO;
1670 : }
1671 27150247 : if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1672 7331161 : BUG_ON(!extent_op || !extent_op->update_flags);
1673 7331161 : ret = alloc_reserved_tree_block(trans, node, extent_op);
1674 19819086 : } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1675 6426212 : ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1676 6426212 : ref->level, 0, 1, extent_op);
1677 13392874 : } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1678 13392874 : ret = __btrfs_free_extent(trans, node, parent, ref_root,
1679 13392874 : ref->level, 0, 1, extent_op);
1680 : } else {
1681 0 : BUG();
1682 : }
1683 : return ret;
1684 : }
1685 :
1686 : /* helper function to actually process a single delayed ref entry */
1687 59924256 : static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1688 : struct btrfs_delayed_ref_node *node,
1689 : struct btrfs_delayed_extent_op *extent_op,
1690 : bool insert_reserved)
1691 : {
1692 59924256 : int ret = 0;
1693 :
1694 59924256 : if (TRANS_ABORTED(trans)) {
1695 0 : if (insert_reserved)
1696 0 : btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1697 0 : return 0;
1698 : }
1699 :
1700 59924256 : if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1701 : node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1702 27150228 : ret = run_delayed_tree_ref(trans, node, extent_op,
1703 : insert_reserved);
1704 32774028 : else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1705 : node->type == BTRFS_SHARED_DATA_REF_KEY)
1706 32774028 : ret = run_delayed_data_ref(trans, node, extent_op,
1707 : insert_reserved);
1708 : else
1709 0 : BUG();
1710 59924300 : if (ret && insert_reserved)
1711 1 : btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
1712 59924300 : if (ret < 0)
1713 1 : btrfs_err(trans->fs_info,
1714 : "failed to run delayed ref for logical %llu num_bytes %llu type %u action %u ref_mod %d: %d",
1715 : node->bytenr, node->num_bytes, node->type,
1716 : node->action, node->ref_mod, ret);
1717 : return ret;
1718 : }
1719 :
1720 : static inline struct btrfs_delayed_ref_node *
1721 : select_delayed_ref(struct btrfs_delayed_ref_head *head)
1722 : {
1723 108189323 : struct btrfs_delayed_ref_node *ref;
1724 :
1725 108189323 : if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
1726 : return NULL;
1727 :
1728 : /*
1729 : * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1730 : * This is to prevent a ref count from going down to zero, which deletes
1731 : * the extent item from the extent tree, when there still are references
1732 : * to add, which would fail because they would not find the extent item.
1733 : */
1734 59924215 : if (!list_empty(&head->ref_add_list))
1735 38014914 : return list_first_entry(&head->ref_add_list,
1736 : struct btrfs_delayed_ref_node, add_list);
1737 :
1738 21909301 : ref = rb_entry(rb_first_cached(&head->ref_tree),
1739 : struct btrfs_delayed_ref_node, ref_node);
1740 21909301 : ASSERT(list_empty(&ref->add_list));
1741 21909301 : return ref;
1742 : }
1743 :
1744 80 : static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1745 : struct btrfs_delayed_ref_head *head)
1746 : {
1747 80 : spin_lock(&delayed_refs->lock);
1748 80 : head->processing = false;
1749 80 : delayed_refs->num_heads_ready++;
1750 80 : spin_unlock(&delayed_refs->lock);
1751 80 : btrfs_delayed_ref_unlock(head);
1752 80 : }
1753 :
1754 48603924 : static struct btrfs_delayed_extent_op *cleanup_extent_op(
1755 : struct btrfs_delayed_ref_head *head)
1756 : {
1757 48603924 : struct btrfs_delayed_extent_op *extent_op = head->extent_op;
1758 :
1759 48603924 : if (!extent_op)
1760 : return NULL;
1761 :
1762 378391 : if (head->must_insert_reserved) {
1763 371157 : head->extent_op = NULL;
1764 371157 : btrfs_free_delayed_extent_op(extent_op);
1765 371157 : return NULL;
1766 : }
1767 : return extent_op;
1768 : }
1769 :
1770 48265107 : static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1771 : struct btrfs_delayed_ref_head *head)
1772 : {
1773 48265107 : struct btrfs_delayed_extent_op *extent_op;
1774 48265107 : int ret;
1775 :
1776 48265107 : extent_op = cleanup_extent_op(head);
1777 48265081 : if (!extent_op)
1778 : return 0;
1779 7234 : head->extent_op = NULL;
1780 7234 : spin_unlock(&head->lock);
1781 7234 : ret = run_delayed_extent_op(trans, head, extent_op);
1782 7234 : btrfs_free_delayed_extent_op(extent_op);
1783 7234 : return ret ? ret : 1;
1784 : }
1785 :
1786 48597070 : void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1787 : struct btrfs_delayed_ref_root *delayed_refs,
1788 : struct btrfs_delayed_ref_head *head)
1789 : {
1790 48597070 : int nr_items = 1; /* Dropping this ref head update. */
1791 :
1792 : /*
1793 : * We had csum deletions accounted for in our delayed refs rsv, we need
1794 : * to drop the csum leaves for this update from our delayed_refs_rsv.
1795 : */
1796 48597070 : if (head->total_ref_mod < 0 && head->is_data) {
1797 6806704 : spin_lock(&delayed_refs->lock);
1798 6806704 : delayed_refs->pending_csums -= head->num_bytes;
1799 6806704 : spin_unlock(&delayed_refs->lock);
1800 6806704 : nr_items += btrfs_csum_bytes_to_leaves(fs_info, head->num_bytes);
1801 : }
1802 :
1803 48597070 : btrfs_delayed_refs_rsv_release(fs_info, nr_items);
1804 48597016 : }
1805 :
1806 48265138 : static int cleanup_ref_head(struct btrfs_trans_handle *trans,
1807 : struct btrfs_delayed_ref_head *head)
1808 : {
1809 :
1810 48265138 : struct btrfs_fs_info *fs_info = trans->fs_info;
1811 48265138 : struct btrfs_delayed_ref_root *delayed_refs;
1812 48265138 : int ret;
1813 :
1814 48265138 : delayed_refs = &trans->transaction->delayed_refs;
1815 :
1816 48265138 : ret = run_and_cleanup_extent_op(trans, head);
1817 48265071 : if (ret < 0) {
1818 0 : unselect_delayed_ref_head(delayed_refs, head);
1819 0 : btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1820 0 : return ret;
1821 48265071 : } else if (ret) {
1822 : return ret;
1823 : }
1824 :
1825 : /*
1826 : * Need to drop our head ref lock and re-acquire the delayed ref lock
1827 : * and then re-check to make sure nobody got added.
1828 : */
1829 48257820 : spin_unlock(&head->lock);
1830 48258090 : spin_lock(&delayed_refs->lock);
1831 48258264 : spin_lock(&head->lock);
1832 48258264 : if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
1833 13 : spin_unlock(&head->lock);
1834 13 : spin_unlock(&delayed_refs->lock);
1835 13 : return 1;
1836 : }
1837 48258251 : btrfs_delete_ref_head(delayed_refs, head);
1838 48258251 : spin_unlock(&head->lock);
1839 48258250 : spin_unlock(&delayed_refs->lock);
1840 :
1841 48258251 : if (head->must_insert_reserved) {
1842 1673293 : btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
1843 1673293 : if (head->is_data) {
1844 1640931 : struct btrfs_root *csum_root;
1845 :
1846 1640931 : csum_root = btrfs_csum_root(fs_info, head->bytenr);
1847 1640931 : ret = btrfs_del_csums(trans, csum_root, head->bytenr,
1848 : head->num_bytes);
1849 : }
1850 : }
1851 :
1852 48258251 : btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
1853 :
1854 48258165 : trace_run_delayed_ref_head(fs_info, head, 0);
1855 48258127 : btrfs_delayed_ref_unlock(head);
1856 48258207 : btrfs_put_delayed_ref_head(head);
1857 48258207 : return ret;
1858 : }
1859 :
1860 49264078 : static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1861 : struct btrfs_trans_handle *trans)
1862 : {
1863 49264078 : struct btrfs_delayed_ref_root *delayed_refs =
1864 49264078 : &trans->transaction->delayed_refs;
1865 49264078 : struct btrfs_delayed_ref_head *head = NULL;
1866 49264078 : int ret;
1867 :
1868 49264078 : spin_lock(&delayed_refs->lock);
1869 49265018 : head = btrfs_select_ref_head(delayed_refs);
1870 49265017 : if (!head) {
1871 1006687 : spin_unlock(&delayed_refs->lock);
1872 1006687 : return head;
1873 : }
1874 :
1875 : /*
1876 : * Grab the lock that says we are going to process all the refs for
1877 : * this head
1878 : */
1879 48258330 : ret = btrfs_delayed_ref_lock(delayed_refs, head);
1880 48258330 : spin_unlock(&delayed_refs->lock);
1881 :
1882 : /*
1883 : * We may have dropped the spin lock to get the head mutex lock, and
1884 : * that might have given someone else time to free the head. If that's
1885 : * true, it has been removed from our list and we can move on.
1886 : */
1887 48258329 : if (ret == -EAGAIN)
1888 0 : head = ERR_PTR(-EAGAIN);
1889 :
1890 : return head;
1891 : }
1892 :
1893 48265471 : static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1894 : struct btrfs_delayed_ref_head *locked_ref)
1895 : {
1896 48265471 : struct btrfs_fs_info *fs_info = trans->fs_info;
1897 48265471 : struct btrfs_delayed_ref_root *delayed_refs;
1898 48265471 : struct btrfs_delayed_extent_op *extent_op;
1899 48265471 : struct btrfs_delayed_ref_node *ref;
1900 48265471 : bool must_insert_reserved;
1901 48265471 : int ret;
1902 :
1903 48265471 : delayed_refs = &trans->transaction->delayed_refs;
1904 :
1905 48265471 : lockdep_assert_held(&locked_ref->mutex);
1906 48265471 : lockdep_assert_held(&locked_ref->lock);
1907 :
1908 168113538 : while ((ref = select_delayed_ref(locked_ref))) {
1909 60868627 : if (ref->seq &&
1910 944411 : btrfs_check_delayed_seq(fs_info, ref->seq)) {
1911 79 : spin_unlock(&locked_ref->lock);
1912 79 : unselect_delayed_ref_head(delayed_refs, locked_ref);
1913 79 : return -EAGAIN;
1914 : }
1915 :
1916 59924137 : rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1917 59923949 : RB_CLEAR_NODE(&ref->ref_node);
1918 59923949 : if (!list_empty(&ref->add_list))
1919 38014724 : list_del(&ref->add_list);
1920 : /*
1921 : * When we play the delayed ref, also correct the ref_mod on
1922 : * head
1923 : */
1924 59923925 : switch (ref->action) {
1925 38014695 : case BTRFS_ADD_DELAYED_REF:
1926 : case BTRFS_ADD_DELAYED_EXTENT:
1927 38014695 : locked_ref->ref_mod -= ref->ref_mod;
1928 38014695 : break;
1929 21909230 : case BTRFS_DROP_DELAYED_REF:
1930 21909230 : locked_ref->ref_mod += ref->ref_mod;
1931 21909230 : break;
1932 : default:
1933 0 : WARN_ON(1);
1934 : }
1935 59923925 : atomic_dec(&delayed_refs->num_entries);
1936 :
1937 : /*
1938 : * Record the must_insert_reserved flag before we drop the
1939 : * spin lock.
1940 : */
1941 59924229 : must_insert_reserved = locked_ref->must_insert_reserved;
1942 59924229 : locked_ref->must_insert_reserved = false;
1943 :
1944 59924229 : extent_op = locked_ref->extent_op;
1945 59924229 : locked_ref->extent_op = NULL;
1946 59924229 : spin_unlock(&locked_ref->lock);
1947 :
1948 59924284 : ret = run_one_delayed_ref(trans, ref, extent_op,
1949 : must_insert_reserved);
1950 :
1951 59924204 : btrfs_free_delayed_extent_op(extent_op);
1952 59924178 : if (ret) {
1953 1 : unselect_delayed_ref_head(delayed_refs, locked_ref);
1954 1 : btrfs_put_delayed_ref(ref);
1955 1 : return ret;
1956 : }
1957 :
1958 59924177 : btrfs_put_delayed_ref(ref);
1959 59924251 : cond_resched();
1960 :
1961 59924198 : spin_lock(&locked_ref->lock);
1962 59924078 : btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
1963 : }
1964 :
1965 : return 0;
1966 : }
1967 :
1968 : /*
1969 : * Returns 0 on success or if called with an already aborted transaction.
1970 : * Returns -ENOMEM or -EIO on failure and will abort the transaction.
1971 : */
1972 4375260 : static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1973 : unsigned long nr)
1974 : {
1975 4375260 : struct btrfs_fs_info *fs_info = trans->fs_info;
1976 4375260 : struct btrfs_delayed_ref_root *delayed_refs;
1977 4375260 : struct btrfs_delayed_ref_head *locked_ref = NULL;
1978 4375260 : int ret;
1979 4375260 : unsigned long count = 0;
1980 :
1981 4375260 : delayed_refs = &trans->transaction->delayed_refs;
1982 49271607 : do {
1983 49271607 : if (!locked_ref) {
1984 49264337 : locked_ref = btrfs_obtain_ref_head(trans);
1985 97523345 : if (IS_ERR_OR_NULL(locked_ref)) {
1986 1006687 : if (PTR_ERR(locked_ref) == -EAGAIN) {
1987 0 : continue;
1988 : } else {
1989 : break;
1990 : }
1991 : }
1992 48258329 : count++;
1993 : }
1994 : /*
1995 : * We need to try and merge add/drops of the same ref since we
1996 : * can run into issues with relocate dropping the implicit ref
1997 : * and then it being added back again before the drop can
1998 : * finish. If we merged anything we need to re-loop so we can
1999 : * get a good ref.
2000 : * Or we can get node references of the same type that weren't
2001 : * merged when created due to bumps in the tree mod seq, and
2002 : * we need to merge them to prevent adding an inline extent
2003 : * backref before dropping it (triggering a BUG_ON at
2004 : * insert_inline_extent_backref()).
2005 : */
2006 48265599 : spin_lock(&locked_ref->lock);
2007 48265554 : btrfs_merge_delayed_refs(fs_info, delayed_refs, locked_ref);
2008 :
2009 48265469 : ret = btrfs_run_delayed_refs_for_head(trans, locked_ref);
2010 48265211 : if (ret < 0 && ret != -EAGAIN) {
2011 : /*
2012 : * Error, btrfs_run_delayed_refs_for_head already
2013 : * unlocked everything so just bail out
2014 : */
2015 1 : return ret;
2016 48265210 : } else if (!ret) {
2017 : /*
2018 : * Success, perform the usual cleanup of a processed
2019 : * head
2020 : */
2021 48265131 : ret = cleanup_ref_head(trans, locked_ref);
2022 48264764 : if (ret > 0 ) {
2023 : /* We dropped our lock, we need to loop. */
2024 7247 : ret = 0;
2025 7247 : continue;
2026 48257517 : } else if (ret) {
2027 0 : return ret;
2028 : }
2029 : }
2030 :
2031 : /*
2032 : * Either success case or btrfs_run_delayed_refs_for_head
2033 : * returned -EAGAIN, meaning we need to select another head
2034 : */
2035 :
2036 48257596 : locked_ref = NULL;
2037 48257596 : cond_resched();
2038 48264922 : } while ((nr != -1 && count < nr) || locked_ref);
2039 :
2040 : return 0;
2041 : }
2042 :
2043 : #ifdef SCRAMBLE_DELAYED_REFS
2044 : /*
2045 : * Normally delayed refs get processed in ascending bytenr order. This
2046 : * correlates in most cases to the order added. To expose dependencies on this
2047 : * order, we start to process the tree in the middle instead of the beginning
2048 : */
2049 : static u64 find_middle(struct rb_root *root)
2050 : {
2051 : struct rb_node *n = root->rb_node;
2052 : struct btrfs_delayed_ref_node *entry;
2053 : int alt = 1;
2054 : u64 middle;
2055 : u64 first = 0, last = 0;
2056 :
2057 : n = rb_first(root);
2058 : if (n) {
2059 : entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2060 : first = entry->bytenr;
2061 : }
2062 : n = rb_last(root);
2063 : if (n) {
2064 : entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2065 : last = entry->bytenr;
2066 : }
2067 : n = root->rb_node;
2068 :
2069 : while (n) {
2070 : entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2071 : WARN_ON(!entry->in_tree);
2072 :
2073 : middle = entry->bytenr;
2074 :
2075 : if (alt)
2076 : n = n->rb_left;
2077 : else
2078 : n = n->rb_right;
2079 :
2080 : alt = 1 - alt;
2081 : }
2082 : return middle;
2083 : }
2084 : #endif
2085 :
2086 : /*
2087 : * this starts processing the delayed reference count updates and
2088 : * extent insertions we have queued up so far. count can be
2089 : * 0, which means to process everything in the tree at the start
2090 : * of the run (but not newly added entries), or it can be some target
2091 : * number you'd like to process.
2092 : *
2093 : * Returns 0 on success or if called with an aborted transaction
2094 : * Returns <0 on error and aborts the transaction
2095 : */
2096 2400106 : int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2097 : unsigned long count)
2098 : {
2099 2400106 : struct btrfs_fs_info *fs_info = trans->fs_info;
2100 2400106 : struct rb_node *node;
2101 2400106 : struct btrfs_delayed_ref_root *delayed_refs;
2102 2400106 : struct btrfs_delayed_ref_head *head;
2103 2400106 : int ret;
2104 2400106 : int run_all = count == (unsigned long)-1;
2105 :
2106 : /* We'll clean this up in btrfs_cleanup_transaction */
2107 2400106 : if (TRANS_ABORTED(trans))
2108 : return 0;
2109 :
2110 4800210 : if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2111 : return 0;
2112 :
2113 2400105 : delayed_refs = &trans->transaction->delayed_refs;
2114 2400105 : if (count == 0)
2115 591385 : count = delayed_refs->num_heads_ready;
2116 :
2117 2400105 : again:
2118 : #ifdef SCRAMBLE_DELAYED_REFS
2119 : delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2120 : #endif
2121 4375259 : ret = __btrfs_run_delayed_refs(trans, count);
2122 4375263 : if (ret < 0) {
2123 1 : btrfs_abort_transaction(trans, ret);
2124 1 : return ret;
2125 : }
2126 :
2127 4375262 : if (run_all) {
2128 2868867 : btrfs_create_pending_block_groups(trans);
2129 :
2130 2868867 : spin_lock(&delayed_refs->lock);
2131 2868867 : node = rb_first_cached(&delayed_refs->href_root);
2132 2868867 : if (!node) {
2133 893713 : spin_unlock(&delayed_refs->lock);
2134 893713 : goto out;
2135 : }
2136 1975154 : head = rb_entry(node, struct btrfs_delayed_ref_head,
2137 : href_node);
2138 1975154 : refcount_inc(&head->refs);
2139 1975154 : spin_unlock(&delayed_refs->lock);
2140 :
2141 : /* Mutex was contended, block until it's released and retry. */
2142 1975154 : mutex_lock(&head->mutex);
2143 1975154 : mutex_unlock(&head->mutex);
2144 :
2145 1975154 : btrfs_put_delayed_ref_head(head);
2146 1975154 : cond_resched();
2147 1975154 : goto again;
2148 : }
2149 1506395 : out:
2150 : return 0;
2151 : }
2152 :
2153 71699 : int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2154 : struct extent_buffer *eb, u64 flags)
2155 : {
2156 71699 : struct btrfs_delayed_extent_op *extent_op;
2157 71699 : int level = btrfs_header_level(eb);
2158 71699 : int ret;
2159 :
2160 71699 : extent_op = btrfs_alloc_delayed_extent_op();
2161 71699 : if (!extent_op)
2162 : return -ENOMEM;
2163 :
2164 71699 : extent_op->flags_to_set = flags;
2165 71699 : extent_op->update_flags = true;
2166 71699 : extent_op->update_key = false;
2167 71699 : extent_op->level = level;
2168 :
2169 71699 : ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
2170 71699 : if (ret)
2171 0 : btrfs_free_delayed_extent_op(extent_op);
2172 : return ret;
2173 : }
2174 :
2175 355897 : static noinline int check_delayed_ref(struct btrfs_root *root,
2176 : struct btrfs_path *path,
2177 : u64 objectid, u64 offset, u64 bytenr)
2178 : {
2179 355897 : struct btrfs_delayed_ref_head *head;
2180 355897 : struct btrfs_delayed_ref_node *ref;
2181 355897 : struct btrfs_delayed_data_ref *data_ref;
2182 355897 : struct btrfs_delayed_ref_root *delayed_refs;
2183 355897 : struct btrfs_transaction *cur_trans;
2184 355897 : struct rb_node *node;
2185 355897 : int ret = 0;
2186 :
2187 355897 : spin_lock(&root->fs_info->trans_lock);
2188 355899 : cur_trans = root->fs_info->running_transaction;
2189 355899 : if (cur_trans)
2190 354217 : refcount_inc(&cur_trans->use_count);
2191 355899 : spin_unlock(&root->fs_info->trans_lock);
2192 355899 : if (!cur_trans)
2193 : return 0;
2194 :
2195 354217 : delayed_refs = &cur_trans->delayed_refs;
2196 354217 : spin_lock(&delayed_refs->lock);
2197 354217 : head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
2198 354217 : if (!head) {
2199 60860 : spin_unlock(&delayed_refs->lock);
2200 60860 : btrfs_put_transaction(cur_trans);
2201 60860 : return 0;
2202 : }
2203 :
2204 293357 : if (!mutex_trylock(&head->mutex)) {
2205 0 : if (path->nowait) {
2206 0 : spin_unlock(&delayed_refs->lock);
2207 0 : btrfs_put_transaction(cur_trans);
2208 0 : return -EAGAIN;
2209 : }
2210 :
2211 0 : refcount_inc(&head->refs);
2212 0 : spin_unlock(&delayed_refs->lock);
2213 :
2214 0 : btrfs_release_path(path);
2215 :
2216 : /*
2217 : * Mutex was contended, block until it's released and let
2218 : * caller try again
2219 : */
2220 0 : mutex_lock(&head->mutex);
2221 0 : mutex_unlock(&head->mutex);
2222 0 : btrfs_put_delayed_ref_head(head);
2223 0 : btrfs_put_transaction(cur_trans);
2224 0 : return -EAGAIN;
2225 : }
2226 293357 : spin_unlock(&delayed_refs->lock);
2227 :
2228 293357 : spin_lock(&head->lock);
2229 : /*
2230 : * XXX: We should replace this with a proper search function in the
2231 : * future.
2232 : */
2233 549106 : for (node = rb_first_cached(&head->ref_tree); node;
2234 255749 : node = rb_next(node)) {
2235 319980 : ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
2236 : /* If it's a shared ref we know a cross reference exists */
2237 319980 : if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2238 : ret = 1;
2239 : break;
2240 : }
2241 :
2242 319980 : data_ref = btrfs_delayed_node_to_data_ref(ref);
2243 :
2244 : /*
2245 : * If our ref doesn't match the one we're currently looking at
2246 : * then we have a cross reference.
2247 : */
2248 319980 : if (data_ref->root != root->root_key.objectid ||
2249 319980 : data_ref->objectid != objectid ||
2250 319905 : data_ref->offset != offset) {
2251 : ret = 1;
2252 : break;
2253 : }
2254 : }
2255 293357 : spin_unlock(&head->lock);
2256 293357 : mutex_unlock(&head->mutex);
2257 293357 : btrfs_put_transaction(cur_trans);
2258 293357 : return ret;
2259 : }
2260 :
2261 361234 : static noinline int check_committed_ref(struct btrfs_root *root,
2262 : struct btrfs_path *path,
2263 : u64 objectid, u64 offset, u64 bytenr,
2264 : bool strict)
2265 : {
2266 361234 : struct btrfs_fs_info *fs_info = root->fs_info;
2267 361234 : struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2268 361236 : struct extent_buffer *leaf;
2269 361236 : struct btrfs_extent_data_ref *ref;
2270 361236 : struct btrfs_extent_inline_ref *iref;
2271 361236 : struct btrfs_extent_item *ei;
2272 361236 : struct btrfs_key key;
2273 361236 : u32 item_size;
2274 361236 : int type;
2275 361236 : int ret;
2276 :
2277 361236 : key.objectid = bytenr;
2278 361236 : key.offset = (u64)-1;
2279 361236 : key.type = BTRFS_EXTENT_ITEM_KEY;
2280 :
2281 361236 : ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2282 361236 : if (ret < 0)
2283 0 : goto out;
2284 361236 : BUG_ON(ret == 0); /* Corruption */
2285 :
2286 361236 : ret = -ENOENT;
2287 361236 : if (path->slots[0] == 0)
2288 2220 : goto out;
2289 :
2290 359016 : path->slots[0]--;
2291 359016 : leaf = path->nodes[0];
2292 359016 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2293 :
2294 359017 : if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2295 259917 : goto out;
2296 :
2297 99100 : ret = 1;
2298 99100 : item_size = btrfs_item_size(leaf, path->slots[0]);
2299 99100 : ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2300 :
2301 : /* If extent item has more than 1 inline ref then it's shared */
2302 99099 : if (item_size != sizeof(*ei) +
2303 : btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2304 5227 : goto out;
2305 :
2306 : /*
2307 : * If extent created before last snapshot => it's shared unless the
2308 : * snapshot has been deleted. Use the heuristic if strict is false.
2309 : */
2310 187730 : if (!strict &&
2311 : (btrfs_extent_generation(leaf, ei) <=
2312 : btrfs_root_last_snapshot(&root->root_item)))
2313 0 : goto out;
2314 :
2315 93872 : iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2316 :
2317 : /* If this extent has SHARED_DATA_REF then it's shared */
2318 93872 : type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2319 93871 : if (type != BTRFS_EXTENT_DATA_REF_KEY)
2320 0 : goto out;
2321 :
2322 93871 : ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2323 93871 : if (btrfs_extent_refs(leaf, ei) !=
2324 93870 : btrfs_extent_data_ref_count(leaf, ref) ||
2325 : btrfs_extent_data_ref_root(leaf, ref) !=
2326 187742 : root->root_key.objectid ||
2327 93849 : btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2328 : btrfs_extent_data_ref_offset(leaf, ref) != offset)
2329 111 : goto out;
2330 :
2331 : ret = 0;
2332 361236 : out:
2333 361236 : return ret;
2334 : }
2335 :
2336 361234 : int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2337 : u64 bytenr, bool strict, struct btrfs_path *path)
2338 : {
2339 361234 : int ret;
2340 :
2341 361234 : do {
2342 361234 : ret = check_committed_ref(root, path, objectid,
2343 : offset, bytenr, strict);
2344 361236 : if (ret && ret != -ENOENT)
2345 5338 : goto out;
2346 :
2347 355898 : ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2348 355899 : } while (ret == -EAGAIN);
2349 :
2350 355899 : out:
2351 361237 : btrfs_release_path(path);
2352 361237 : if (btrfs_is_data_reloc_root(root))
2353 22660 : WARN_ON(ret > 0);
2354 361237 : return ret;
2355 : }
2356 :
2357 3801480 : static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2358 : struct btrfs_root *root,
2359 : struct extent_buffer *buf,
2360 : int full_backref, int inc)
2361 : {
2362 3801480 : struct btrfs_fs_info *fs_info = root->fs_info;
2363 3801480 : u64 bytenr;
2364 3801480 : u64 num_bytes;
2365 3801480 : u64 parent;
2366 3801480 : u64 ref_root;
2367 3801480 : u32 nritems;
2368 3801480 : struct btrfs_key key;
2369 3801480 : struct btrfs_file_extent_item *fi;
2370 3801480 : struct btrfs_ref generic_ref = { 0 };
2371 3801480 : bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
2372 3801480 : int i;
2373 3801480 : int action;
2374 3801480 : int level;
2375 3801480 : int ret = 0;
2376 :
2377 3801480 : if (btrfs_is_testing(fs_info))
2378 : return 0;
2379 :
2380 3801480 : ref_root = btrfs_header_owner(buf);
2381 3801480 : nritems = btrfs_header_nritems(buf);
2382 3801480 : level = btrfs_header_level(buf);
2383 :
2384 3801480 : if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
2385 : return 0;
2386 :
2387 3801480 : if (full_backref)
2388 3766605 : parent = buf->start;
2389 : else
2390 : parent = 0;
2391 3801480 : if (inc)
2392 : action = BTRFS_ADD_DELAYED_REF;
2393 : else
2394 1872726 : action = BTRFS_DROP_DELAYED_REF;
2395 :
2396 91714074 : for (i = 0; i < nritems; i++) {
2397 87912594 : if (level == 0) {
2398 81188489 : btrfs_item_key_to_cpu(buf, &key, i);
2399 81188586 : if (key.type != BTRFS_EXTENT_DATA_KEY)
2400 64564499 : continue;
2401 16624087 : fi = btrfs_item_ptr(buf, i,
2402 : struct btrfs_file_extent_item);
2403 16624080 : if (btrfs_file_extent_type(buf, fi) ==
2404 : BTRFS_FILE_EXTENT_INLINE)
2405 130282 : continue;
2406 16493799 : bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2407 16493797 : if (bytenr == 0)
2408 3195 : continue;
2409 :
2410 16490602 : num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2411 16490606 : key.offset -= btrfs_file_extent_offset(buf, fi);
2412 16490606 : btrfs_init_generic_ref(&generic_ref, action, bytenr,
2413 : num_bytes, parent);
2414 16490606 : btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2415 : key.offset, root->root_key.objectid,
2416 : for_reloc);
2417 16490613 : if (inc)
2418 9023206 : ret = btrfs_inc_extent_ref(trans, &generic_ref);
2419 : else
2420 7467407 : ret = btrfs_free_extent(trans, &generic_ref);
2421 16490621 : if (ret)
2422 0 : goto fail;
2423 : } else {
2424 6724105 : bytenr = btrfs_node_blockptr(buf, i);
2425 6724105 : num_bytes = fs_info->nodesize;
2426 6724105 : btrfs_init_generic_ref(&generic_ref, action, bytenr,
2427 : num_bytes, parent);
2428 6724105 : btrfs_init_tree_ref(&generic_ref, level - 1, ref_root,
2429 : root->root_key.objectid, for_reloc);
2430 6724105 : if (inc)
2431 6583821 : ret = btrfs_inc_extent_ref(trans, &generic_ref);
2432 : else
2433 140284 : ret = btrfs_free_extent(trans, &generic_ref);
2434 6723997 : if (ret)
2435 0 : goto fail;
2436 : }
2437 : }
2438 : return 0;
2439 : fail:
2440 : return ret;
2441 : }
2442 :
2443 1928754 : int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2444 : struct extent_buffer *buf, int full_backref)
2445 : {
2446 1928754 : return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2447 : }
2448 :
2449 33901 : int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2450 : struct extent_buffer *buf, int full_backref)
2451 : {
2452 33901 : return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2453 : }
2454 :
2455 12354834 : static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
2456 : {
2457 12354834 : struct btrfs_fs_info *fs_info = root->fs_info;
2458 12354834 : u64 flags;
2459 12354834 : u64 ret;
2460 :
2461 12354834 : if (data)
2462 : flags = BTRFS_BLOCK_GROUP_DATA;
2463 8795172 : else if (root == fs_info->chunk_root)
2464 : flags = BTRFS_BLOCK_GROUP_SYSTEM;
2465 : else
2466 8793761 : flags = BTRFS_BLOCK_GROUP_METADATA;
2467 :
2468 12354834 : ret = btrfs_get_alloc_profile(fs_info, flags);
2469 12355313 : return ret;
2470 : }
2471 :
2472 12358356 : static u64 first_logical_byte(struct btrfs_fs_info *fs_info)
2473 : {
2474 12358356 : struct rb_node *leftmost;
2475 12358356 : u64 bytenr = 0;
2476 :
2477 12358356 : read_lock(&fs_info->block_group_cache_lock);
2478 : /* Get the block group with the lowest logical start address. */
2479 12358675 : leftmost = rb_first_cached(&fs_info->block_group_cache_tree);
2480 12358675 : if (leftmost) {
2481 12358675 : struct btrfs_block_group *bg;
2482 :
2483 12358675 : bg = rb_entry(leftmost, struct btrfs_block_group, cache_node);
2484 12358675 : bytenr = bg->start;
2485 : }
2486 12358675 : read_unlock(&fs_info->block_group_cache_lock);
2487 :
2488 12358958 : return bytenr;
2489 : }
2490 :
2491 3069205 : static int pin_down_extent(struct btrfs_trans_handle *trans,
2492 : struct btrfs_block_group *cache,
2493 : u64 bytenr, u64 num_bytes, int reserved)
2494 : {
2495 3069205 : struct btrfs_fs_info *fs_info = cache->fs_info;
2496 :
2497 3069205 : spin_lock(&cache->space_info->lock);
2498 3069230 : spin_lock(&cache->lock);
2499 3069229 : cache->pinned += num_bytes;
2500 3069229 : btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2501 : num_bytes);
2502 3069223 : if (reserved) {
2503 3064252 : cache->reserved -= num_bytes;
2504 3064252 : cache->space_info->bytes_reserved -= num_bytes;
2505 : }
2506 3069223 : spin_unlock(&cache->lock);
2507 3069234 : spin_unlock(&cache->space_info->lock);
2508 :
2509 3069233 : set_extent_bit(&trans->transaction->pinned_extents, bytenr,
2510 3069233 : bytenr + num_bytes - 1, EXTENT_DIRTY, NULL);
2511 3069236 : return 0;
2512 : }
2513 :
2514 1673294 : int btrfs_pin_extent(struct btrfs_trans_handle *trans,
2515 : u64 bytenr, u64 num_bytes, int reserved)
2516 : {
2517 1673294 : struct btrfs_block_group *cache;
2518 :
2519 1673294 : cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2520 1673294 : BUG_ON(!cache); /* Logic error */
2521 :
2522 1673294 : pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
2523 :
2524 1673294 : btrfs_put_block_group(cache);
2525 1673294 : return 0;
2526 : }
2527 :
2528 : /*
2529 : * this function must be called within transaction
2530 : */
2531 4971 : int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
2532 : u64 bytenr, u64 num_bytes)
2533 : {
2534 4971 : struct btrfs_block_group *cache;
2535 4971 : int ret;
2536 :
2537 4971 : cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
2538 4971 : if (!cache)
2539 : return -EINVAL;
2540 :
2541 : /*
2542 : * Fully cache the free space first so that our pin removes the free space
2543 : * from the cache.
2544 : */
2545 4971 : ret = btrfs_cache_block_group(cache, true);
2546 4971 : if (ret)
2547 0 : goto out;
2548 :
2549 4971 : pin_down_extent(trans, cache, bytenr, num_bytes, 0);
2550 :
2551 : /* remove us from the free space cache (if we're there at all) */
2552 4971 : ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
2553 4971 : out:
2554 4971 : btrfs_put_block_group(cache);
2555 4971 : return ret;
2556 : }
2557 :
2558 76673 : static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2559 : u64 start, u64 num_bytes)
2560 : {
2561 76673 : int ret;
2562 76673 : struct btrfs_block_group *block_group;
2563 :
2564 76673 : block_group = btrfs_lookup_block_group(fs_info, start);
2565 76673 : if (!block_group)
2566 : return -EINVAL;
2567 :
2568 76673 : ret = btrfs_cache_block_group(block_group, true);
2569 76673 : if (ret)
2570 0 : goto out;
2571 :
2572 76673 : ret = btrfs_remove_free_space(block_group, start, num_bytes);
2573 76673 : out:
2574 76673 : btrfs_put_block_group(block_group);
2575 76673 : return ret;
2576 : }
2577 :
2578 4431 : int btrfs_exclude_logged_extents(struct extent_buffer *eb)
2579 : {
2580 4431 : struct btrfs_fs_info *fs_info = eb->fs_info;
2581 4431 : struct btrfs_file_extent_item *item;
2582 4431 : struct btrfs_key key;
2583 4431 : int found_type;
2584 4431 : int i;
2585 4431 : int ret = 0;
2586 :
2587 4431 : if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
2588 : return 0;
2589 :
2590 0 : for (i = 0; i < btrfs_header_nritems(eb); i++) {
2591 0 : btrfs_item_key_to_cpu(eb, &key, i);
2592 0 : if (key.type != BTRFS_EXTENT_DATA_KEY)
2593 0 : continue;
2594 0 : item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2595 0 : found_type = btrfs_file_extent_type(eb, item);
2596 0 : if (found_type == BTRFS_FILE_EXTENT_INLINE)
2597 0 : continue;
2598 0 : if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2599 0 : continue;
2600 0 : key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2601 0 : key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
2602 0 : ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2603 0 : if (ret)
2604 : break;
2605 : }
2606 :
2607 : return ret;
2608 : }
2609 :
2610 : static void
2611 : btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
2612 : {
2613 12355409 : atomic_inc(&bg->reservations);
2614 : }
2615 :
2616 : /*
2617 : * Returns the free cluster for the given space info and sets empty_cluster to
2618 : * what it should be based on the mount options.
2619 : */
2620 : static struct btrfs_free_cluster *
2621 16012921 : fetch_cluster_info(struct btrfs_fs_info *fs_info,
2622 : struct btrfs_space_info *space_info, u64 *empty_cluster)
2623 : {
2624 16012921 : struct btrfs_free_cluster *ret = NULL;
2625 :
2626 16012921 : *empty_cluster = 0;
2627 16012921 : if (btrfs_mixed_space_info(space_info))
2628 : return ret;
2629 :
2630 15422031 : if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
2631 11779375 : ret = &fs_info->meta_alloc_cluster;
2632 11779375 : if (btrfs_test_opt(fs_info, SSD))
2633 6579 : *empty_cluster = SZ_2M;
2634 : else
2635 11772796 : *empty_cluster = SZ_64K;
2636 3642656 : } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2637 3639837 : btrfs_test_opt(fs_info, SSD_SPREAD)) {
2638 0 : *empty_cluster = SZ_2M;
2639 0 : ret = &fs_info->data_alloc_cluster;
2640 : }
2641 :
2642 : return ret;
2643 : }
2644 :
2645 3654276 : static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2646 : u64 start, u64 end,
2647 : const bool return_free_space)
2648 : {
2649 3654276 : struct btrfs_block_group *cache = NULL;
2650 3654276 : struct btrfs_space_info *space_info;
2651 3654276 : struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2652 3654276 : struct btrfs_free_cluster *cluster = NULL;
2653 3654276 : u64 len;
2654 3654276 : u64 total_unpinned = 0;
2655 3654276 : u64 empty_cluster = 0;
2656 3654276 : bool readonly;
2657 :
2658 7309036 : while (start <= end) {
2659 3654760 : readonly = false;
2660 3654760 : if (!cache ||
2661 484 : start >= cache->start + cache->length) {
2662 3654760 : if (cache)
2663 484 : btrfs_put_block_group(cache);
2664 3654760 : total_unpinned = 0;
2665 3654760 : cache = btrfs_lookup_block_group(fs_info, start);
2666 3654760 : BUG_ON(!cache); /* Logic error */
2667 :
2668 3654760 : cluster = fetch_cluster_info(fs_info,
2669 : cache->space_info,
2670 : &empty_cluster);
2671 3654760 : empty_cluster <<= 1;
2672 : }
2673 :
2674 3654760 : len = cache->start + cache->length - start;
2675 3654760 : len = min(len, end + 1 - start);
2676 :
2677 3654760 : if (return_free_space)
2678 3654616 : btrfs_add_free_space(cache, start, len);
2679 :
2680 3654760 : start += len;
2681 3654760 : total_unpinned += len;
2682 3654760 : space_info = cache->space_info;
2683 :
2684 : /*
2685 : * If this space cluster has been marked as fragmented and we've
2686 : * unpinned enough in this block group to potentially allow a
2687 : * cluster to be created inside of it go ahead and clear the
2688 : * fragmented check.
2689 : */
2690 3654760 : if (cluster && cluster->fragmented &&
2691 0 : total_unpinned > empty_cluster) {
2692 0 : spin_lock(&cluster->lock);
2693 0 : cluster->fragmented = 0;
2694 0 : spin_unlock(&cluster->lock);
2695 : }
2696 :
2697 3654760 : spin_lock(&space_info->lock);
2698 3654760 : spin_lock(&cache->lock);
2699 3654760 : cache->pinned -= len;
2700 3654760 : btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
2701 3654760 : space_info->max_extent_size = 0;
2702 3654760 : if (cache->ro) {
2703 55494 : space_info->bytes_readonly += len;
2704 55494 : readonly = true;
2705 : } else if (btrfs_is_zoned(fs_info)) {
2706 : /* Need reset before reusing in a zoned block group */
2707 : space_info->bytes_zone_unusable += len;
2708 : readonly = true;
2709 : }
2710 3654760 : spin_unlock(&cache->lock);
2711 3654760 : if (!readonly && return_free_space &&
2712 3599122 : global_rsv->space_info == space_info) {
2713 3438369 : spin_lock(&global_rsv->lock);
2714 3438369 : if (!global_rsv->full) {
2715 291299 : u64 to_add = min(len, global_rsv->size -
2716 : global_rsv->reserved);
2717 :
2718 291299 : global_rsv->reserved += to_add;
2719 291299 : btrfs_space_info_update_bytes_may_use(fs_info,
2720 : space_info, to_add);
2721 291299 : if (global_rsv->reserved >= global_rsv->size)
2722 142810 : global_rsv->full = 1;
2723 291299 : len -= to_add;
2724 : }
2725 3438369 : spin_unlock(&global_rsv->lock);
2726 : }
2727 : /* Add to any tickets we may have */
2728 3654760 : if (!readonly && return_free_space && len)
2729 3347676 : btrfs_try_granting_tickets(fs_info, space_info);
2730 3654760 : spin_unlock(&space_info->lock);
2731 : }
2732 :
2733 3654276 : if (cache)
2734 3654276 : btrfs_put_block_group(cache);
2735 3654276 : return 0;
2736 : }
2737 :
2738 203256 : int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
2739 : {
2740 203256 : struct btrfs_fs_info *fs_info = trans->fs_info;
2741 203256 : struct btrfs_block_group *block_group, *tmp;
2742 203256 : struct list_head *deleted_bgs;
2743 203256 : struct extent_io_tree *unpin;
2744 203256 : u64 start;
2745 203256 : u64 end;
2746 203256 : int ret;
2747 :
2748 203256 : unpin = &trans->transaction->pinned_extents;
2749 :
2750 3857388 : while (!TRANS_ABORTED(trans)) {
2751 3857388 : struct extent_state *cached_state = NULL;
2752 :
2753 3857388 : mutex_lock(&fs_info->unused_bg_unpin_mutex);
2754 3857388 : ret = find_first_extent_bit(unpin, 0, &start, &end,
2755 : EXTENT_DIRTY, &cached_state);
2756 3857388 : if (ret) {
2757 203256 : mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2758 203256 : break;
2759 : }
2760 :
2761 3654132 : if (btrfs_test_opt(fs_info, DISCARD_SYNC))
2762 36 : ret = btrfs_discard_extent(fs_info, start,
2763 36 : end + 1 - start, NULL);
2764 :
2765 3654132 : clear_extent_dirty(unpin, start, end, &cached_state);
2766 3654132 : unpin_extent_range(fs_info, start, end, true);
2767 3654132 : mutex_unlock(&fs_info->unused_bg_unpin_mutex);
2768 3654132 : free_extent_state(cached_state);
2769 3654132 : cond_resched();
2770 : }
2771 :
2772 203256 : if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2773 203227 : btrfs_discard_calc_delay(&fs_info->discard_ctl);
2774 203227 : btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
2775 : }
2776 :
2777 : /*
2778 : * Transaction is finished. We don't need the lock anymore. We
2779 : * do need to clean up the block groups in case of a transaction
2780 : * abort.
2781 : */
2782 203256 : deleted_bgs = &trans->transaction->deleted_bgs;
2783 203256 : list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2784 0 : u64 trimmed = 0;
2785 :
2786 0 : ret = -EROFS;
2787 0 : if (!TRANS_ABORTED(trans))
2788 0 : ret = btrfs_discard_extent(fs_info,
2789 : block_group->start,
2790 : block_group->length,
2791 : &trimmed);
2792 :
2793 0 : list_del_init(&block_group->bg_list);
2794 0 : btrfs_unfreeze_block_group(block_group);
2795 0 : btrfs_put_block_group(block_group);
2796 :
2797 0 : if (ret) {
2798 0 : const char *errstr = btrfs_decode_error(ret);
2799 0 : btrfs_warn(fs_info,
2800 : "discard failed while removing blockgroup: errno=%d %s",
2801 : ret, errstr);
2802 : }
2803 : }
2804 :
2805 203256 : return 0;
2806 : }
2807 :
2808 8073269 : static int do_free_extent_accounting(struct btrfs_trans_handle *trans,
2809 : u64 bytenr, u64 num_bytes, bool is_data)
2810 : {
2811 8073269 : int ret;
2812 :
2813 8073269 : if (is_data) {
2814 1015715 : struct btrfs_root *csum_root;
2815 :
2816 1015715 : csum_root = btrfs_csum_root(trans->fs_info, bytenr);
2817 1015714 : ret = btrfs_del_csums(trans, csum_root, bytenr, num_bytes);
2818 1015715 : if (ret) {
2819 0 : btrfs_abort_transaction(trans, ret);
2820 0 : return ret;
2821 : }
2822 : }
2823 :
2824 8073269 : ret = add_to_free_space_tree(trans, bytenr, num_bytes);
2825 8073268 : if (ret) {
2826 0 : btrfs_abort_transaction(trans, ret);
2827 0 : return ret;
2828 : }
2829 :
2830 8073268 : ret = btrfs_update_block_group(trans, bytenr, num_bytes, false);
2831 8073269 : if (ret)
2832 0 : btrfs_abort_transaction(trans, ret);
2833 :
2834 : return ret;
2835 : }
2836 :
2837 : #define abort_and_dump(trans, path, fmt, args...) \
2838 : ({ \
2839 : btrfs_abort_transaction(trans, -EUCLEAN); \
2840 : btrfs_print_leaf(path->nodes[0]); \
2841 : btrfs_crit(trans->fs_info, fmt, ##args); \
2842 : })
2843 :
2844 : /*
2845 : * Drop one or more refs of @node.
2846 : *
2847 : * 1. Locate the extent refs.
2848 : * It's either inline in EXTENT/METADATA_ITEM or in keyed SHARED_* item.
2849 : * Locate it, then reduce the refs number or remove the ref line completely.
2850 : *
2851 : * 2. Update the refs count in EXTENT/METADATA_ITEM
2852 : *
2853 : * Inline backref case:
2854 : *
2855 : * in extent tree we have:
2856 : *
2857 : * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2858 : * refs 2 gen 6 flags DATA
2859 : * extent data backref root FS_TREE objectid 258 offset 0 count 1
2860 : * extent data backref root FS_TREE objectid 257 offset 0 count 1
2861 : *
2862 : * This function gets called with:
2863 : *
2864 : * node->bytenr = 13631488
2865 : * node->num_bytes = 1048576
2866 : * root_objectid = FS_TREE
2867 : * owner_objectid = 257
2868 : * owner_offset = 0
2869 : * refs_to_drop = 1
2870 : *
2871 : * Then we should get some like:
2872 : *
2873 : * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 16201 itemsize 82
2874 : * refs 1 gen 6 flags DATA
2875 : * extent data backref root FS_TREE objectid 258 offset 0 count 1
2876 : *
2877 : * Keyed backref case:
2878 : *
2879 : * in extent tree we have:
2880 : *
2881 : * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2882 : * refs 754 gen 6 flags DATA
2883 : * [...]
2884 : * item 2 key (13631488 EXTENT_DATA_REF <HASH>) itemoff 3915 itemsize 28
2885 : * extent data backref root FS_TREE objectid 866 offset 0 count 1
2886 : *
2887 : * This function get called with:
2888 : *
2889 : * node->bytenr = 13631488
2890 : * node->num_bytes = 1048576
2891 : * root_objectid = FS_TREE
2892 : * owner_objectid = 866
2893 : * owner_offset = 0
2894 : * refs_to_drop = 1
2895 : *
2896 : * Then we should get some like:
2897 : *
2898 : * item 0 key (13631488 EXTENT_ITEM 1048576) itemoff 3971 itemsize 24
2899 : * refs 753 gen 6 flags DATA
2900 : *
2901 : * And that (13631488 EXTENT_DATA_REF <HASH>) gets removed.
2902 : */
2903 21909262 : static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2904 : struct btrfs_delayed_ref_node *node, u64 parent,
2905 : u64 root_objectid, u64 owner_objectid,
2906 : u64 owner_offset, int refs_to_drop,
2907 : struct btrfs_delayed_extent_op *extent_op)
2908 : {
2909 21909262 : struct btrfs_fs_info *info = trans->fs_info;
2910 21909262 : struct btrfs_key key;
2911 21909262 : struct btrfs_path *path;
2912 21909262 : struct btrfs_root *extent_root;
2913 21909262 : struct extent_buffer *leaf;
2914 21909262 : struct btrfs_extent_item *ei;
2915 21909262 : struct btrfs_extent_inline_ref *iref;
2916 21909262 : int ret;
2917 21909262 : int is_data;
2918 21909262 : int extent_slot = 0;
2919 21909262 : int found_extent = 0;
2920 21909262 : int num_to_del = 1;
2921 21909262 : u32 item_size;
2922 21909262 : u64 refs;
2923 21909262 : u64 bytenr = node->bytenr;
2924 21909262 : u64 num_bytes = node->num_bytes;
2925 21909262 : bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
2926 :
2927 21909262 : extent_root = btrfs_extent_root(info, bytenr);
2928 21909242 : ASSERT(extent_root);
2929 :
2930 21909242 : path = btrfs_alloc_path();
2931 21909255 : if (!path)
2932 : return -ENOMEM;
2933 :
2934 21909255 : is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2935 :
2936 21909255 : if (!is_data && refs_to_drop != 1) {
2937 0 : btrfs_crit(info,
2938 : "invalid refs_to_drop, dropping more than 1 refs for tree block %llu refs_to_drop %u",
2939 : node->bytenr, refs_to_drop);
2940 0 : ret = -EINVAL;
2941 0 : btrfs_abort_transaction(trans, ret);
2942 0 : goto out;
2943 : }
2944 :
2945 21909255 : if (is_data)
2946 8516373 : skinny_metadata = false;
2947 :
2948 21909255 : ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2949 : parent, root_objectid, owner_objectid,
2950 : owner_offset);
2951 21909286 : if (ret == 0) {
2952 : /*
2953 : * Either the inline backref or the SHARED_DATA_REF/
2954 : * SHARED_BLOCK_REF is found
2955 : *
2956 : * Here is a quick path to locate EXTENT/METADATA_ITEM.
2957 : * It's possible the EXTENT/METADATA_ITEM is near current slot.
2958 : */
2959 21909286 : extent_slot = path->slots[0];
2960 22875572 : while (extent_slot >= 0) {
2961 22849000 : btrfs_item_key_to_cpu(path->nodes[0], &key,
2962 : extent_slot);
2963 22849000 : if (key.objectid != bytenr)
2964 : break;
2965 22849000 : if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2966 8351722 : key.offset == num_bytes) {
2967 : found_extent = 1;
2968 : break;
2969 : }
2970 14497278 : if (key.type == BTRFS_METADATA_ITEM_KEY &&
2971 13392883 : key.offset == owner_objectid) {
2972 : found_extent = 1;
2973 : break;
2974 : }
2975 :
2976 : /* Quick path didn't find the EXTEMT/METADATA_ITEM */
2977 1104395 : if (path->slots[0] - extent_slot > 5)
2978 : break;
2979 966286 : extent_slot--;
2980 : }
2981 :
2982 21909286 : if (!found_extent) {
2983 164681 : if (iref) {
2984 0 : abort_and_dump(trans, path,
2985 : "invalid iref slot %u, no EXTENT/METADATA_ITEM found but has inline extent ref",
2986 : path->slots[0]);
2987 0 : ret = -EUCLEAN;
2988 0 : goto out;
2989 : }
2990 : /* Must be SHARED_* item, remove the backref first */
2991 164681 : ret = remove_extent_backref(trans, extent_root, path,
2992 : NULL, refs_to_drop, is_data);
2993 164681 : if (ret) {
2994 0 : btrfs_abort_transaction(trans, ret);
2995 0 : goto out;
2996 : }
2997 164681 : btrfs_release_path(path);
2998 :
2999 : /* Slow path to locate EXTENT/METADATA_ITEM */
3000 164681 : key.objectid = bytenr;
3001 164681 : key.type = BTRFS_EXTENT_ITEM_KEY;
3002 164681 : key.offset = num_bytes;
3003 :
3004 164681 : if (!is_data && skinny_metadata) {
3005 0 : key.type = BTRFS_METADATA_ITEM_KEY;
3006 0 : key.offset = owner_objectid;
3007 : }
3008 :
3009 164681 : ret = btrfs_search_slot(trans, extent_root,
3010 : &key, path, -1, 1);
3011 164681 : if (ret > 0 && skinny_metadata && path->slots[0]) {
3012 : /*
3013 : * Couldn't find our skinny metadata item,
3014 : * see if we have ye olde extent item.
3015 : */
3016 0 : path->slots[0]--;
3017 0 : btrfs_item_key_to_cpu(path->nodes[0], &key,
3018 : path->slots[0]);
3019 0 : if (key.objectid == bytenr &&
3020 0 : key.type == BTRFS_EXTENT_ITEM_KEY &&
3021 0 : key.offset == num_bytes)
3022 0 : ret = 0;
3023 : }
3024 :
3025 164681 : if (ret > 0 && skinny_metadata) {
3026 0 : skinny_metadata = false;
3027 0 : key.objectid = bytenr;
3028 0 : key.type = BTRFS_EXTENT_ITEM_KEY;
3029 0 : key.offset = num_bytes;
3030 0 : btrfs_release_path(path);
3031 0 : ret = btrfs_search_slot(trans, extent_root,
3032 : &key, path, -1, 1);
3033 : }
3034 :
3035 164681 : if (ret) {
3036 0 : if (ret > 0)
3037 0 : btrfs_print_leaf(path->nodes[0]);
3038 0 : btrfs_err(info,
3039 : "umm, got %d back from search, was looking for %llu, slot %d",
3040 : ret, bytenr, path->slots[0]);
3041 : }
3042 164681 : if (ret < 0) {
3043 0 : btrfs_abort_transaction(trans, ret);
3044 0 : goto out;
3045 : }
3046 164681 : extent_slot = path->slots[0];
3047 : }
3048 0 : } else if (WARN_ON(ret == -ENOENT)) {
3049 0 : abort_and_dump(trans, path,
3050 : "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu slot %d",
3051 : bytenr, parent, root_objectid, owner_objectid,
3052 : owner_offset, path->slots[0]);
3053 0 : goto out;
3054 : } else {
3055 0 : btrfs_abort_transaction(trans, ret);
3056 0 : goto out;
3057 : }
3058 :
3059 21909286 : leaf = path->nodes[0];
3060 21909286 : item_size = btrfs_item_size(leaf, extent_slot);
3061 21909286 : if (unlikely(item_size < sizeof(*ei))) {
3062 0 : ret = -EINVAL;
3063 0 : btrfs_print_v0_err(info);
3064 0 : btrfs_abort_transaction(trans, ret);
3065 0 : goto out;
3066 : }
3067 21909286 : ei = btrfs_item_ptr(leaf, extent_slot,
3068 : struct btrfs_extent_item);
3069 21909286 : if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3070 13392883 : key.type == BTRFS_EXTENT_ITEM_KEY) {
3071 0 : struct btrfs_tree_block_info *bi;
3072 :
3073 0 : if (item_size < sizeof(*ei) + sizeof(*bi)) {
3074 0 : abort_and_dump(trans, path,
3075 : "invalid extent item size for key (%llu, %u, %llu) slot %u owner %llu, has %u expect >= %zu",
3076 : key.objectid, key.type, key.offset,
3077 : path->slots[0], owner_objectid, item_size,
3078 : sizeof(*ei) + sizeof(*bi));
3079 0 : ret = -EUCLEAN;
3080 0 : goto out;
3081 : }
3082 0 : bi = (struct btrfs_tree_block_info *)(ei + 1);
3083 0 : WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3084 : }
3085 :
3086 21909286 : refs = btrfs_extent_refs(leaf, ei);
3087 21909286 : if (refs < refs_to_drop) {
3088 0 : abort_and_dump(trans, path,
3089 : "trying to drop %d refs but we only have %llu for bytenr %llu slot %u",
3090 : refs_to_drop, refs, bytenr, path->slots[0]);
3091 0 : ret = -EUCLEAN;
3092 0 : goto out;
3093 : }
3094 21909286 : refs -= refs_to_drop;
3095 :
3096 21909286 : if (refs > 0) {
3097 13836017 : if (extent_op)
3098 42090 : __run_delayed_extent_op(extent_op, leaf, ei);
3099 : /*
3100 : * In the case of inline back ref, reference count will
3101 : * be updated by remove_extent_backref
3102 : */
3103 13836017 : if (iref) {
3104 13612388 : if (!found_extent) {
3105 0 : abort_and_dump(trans, path,
3106 : "invalid iref, got inlined extent ref but no EXTENT/METADATA_ITEM found, slot %u",
3107 : path->slots[0]);
3108 0 : ret = -EUCLEAN;
3109 0 : goto out;
3110 : }
3111 : } else {
3112 223629 : btrfs_set_extent_refs(leaf, ei, refs);
3113 223629 : btrfs_mark_buffer_dirty(leaf);
3114 : }
3115 13836017 : if (found_extent) {
3116 13671369 : ret = remove_extent_backref(trans, extent_root, path,
3117 : iref, refs_to_drop, is_data);
3118 13671369 : if (ret) {
3119 0 : btrfs_abort_transaction(trans, ret);
3120 0 : goto out;
3121 : }
3122 : }
3123 : } else {
3124 : /* In this branch refs == 1 */
3125 8073269 : if (found_extent) {
3126 9088918 : if (is_data && refs_to_drop !=
3127 1015682 : extent_data_ref_count(path, iref)) {
3128 0 : abort_and_dump(trans, path,
3129 : "invalid refs_to_drop, current refs %u refs_to_drop %u slot %u",
3130 : extent_data_ref_count(path, iref),
3131 : refs_to_drop, path->slots[0]);
3132 0 : ret = -EUCLEAN;
3133 0 : goto out;
3134 : }
3135 8073236 : if (iref) {
3136 8072964 : if (path->slots[0] != extent_slot) {
3137 0 : abort_and_dump(trans, path,
3138 : "invalid iref, extent item key (%llu %u %llu) slot %u doesn't have wanted iref",
3139 : key.objectid, key.type,
3140 : key.offset, path->slots[0]);
3141 0 : ret = -EUCLEAN;
3142 0 : goto out;
3143 : }
3144 : } else {
3145 : /*
3146 : * No inline ref, we must be at SHARED_* item,
3147 : * And it's single ref, it must be:
3148 : * | extent_slot ||extent_slot + 1|
3149 : * [ EXTENT/METADATA_ITEM ][ SHARED_* ITEM ]
3150 : */
3151 272 : if (path->slots[0] != extent_slot + 1) {
3152 0 : abort_and_dump(trans, path,
3153 : "invalid SHARED_* item slot %u, previous item is not EXTENT/METADATA_ITEM",
3154 : path->slots[0]);
3155 0 : ret = -EUCLEAN;
3156 0 : goto out;
3157 : }
3158 272 : path->slots[0] = extent_slot;
3159 272 : num_to_del = 2;
3160 : }
3161 : }
3162 :
3163 8073269 : ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3164 : num_to_del);
3165 8073268 : if (ret) {
3166 0 : btrfs_abort_transaction(trans, ret);
3167 0 : goto out;
3168 : }
3169 8073268 : btrfs_release_path(path);
3170 :
3171 8073269 : ret = do_free_extent_accounting(trans, bytenr, num_bytes, is_data);
3172 : }
3173 21909286 : btrfs_release_path(path);
3174 :
3175 21909270 : out:
3176 21909270 : btrfs_free_path(path);
3177 21909270 : return ret;
3178 : }
3179 :
3180 : /*
3181 : * when we free an block, it is possible (and likely) that we free the last
3182 : * delayed ref for that extent as well. This searches the delayed ref tree for
3183 : * a given extent, and if there are no other delayed refs to be processed, it
3184 : * removes it from the tree.
3185 : */
3186 737141 : static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3187 : u64 bytenr)
3188 : {
3189 737141 : struct btrfs_delayed_ref_head *head;
3190 737141 : struct btrfs_delayed_ref_root *delayed_refs;
3191 737141 : int ret = 0;
3192 :
3193 737141 : delayed_refs = &trans->transaction->delayed_refs;
3194 737141 : spin_lock(&delayed_refs->lock);
3195 737141 : head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3196 737141 : if (!head)
3197 1 : goto out_delayed_unlock;
3198 :
3199 737140 : spin_lock(&head->lock);
3200 737140 : if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
3201 398321 : goto out;
3202 :
3203 338819 : if (cleanup_extent_op(head) != NULL)
3204 0 : goto out;
3205 :
3206 : /*
3207 : * waiting for the lock here would deadlock. If someone else has it
3208 : * locked they are already in the process of dropping it anyway
3209 : */
3210 338819 : if (!mutex_trylock(&head->mutex))
3211 24 : goto out;
3212 :
3213 338795 : btrfs_delete_ref_head(delayed_refs, head);
3214 338795 : head->processing = false;
3215 :
3216 338795 : spin_unlock(&head->lock);
3217 338795 : spin_unlock(&delayed_refs->lock);
3218 :
3219 338795 : BUG_ON(head->extent_op);
3220 338795 : if (head->must_insert_reserved)
3221 338795 : ret = 1;
3222 :
3223 338795 : btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
3224 338795 : mutex_unlock(&head->mutex);
3225 338793 : btrfs_put_delayed_ref_head(head);
3226 338793 : return ret;
3227 398345 : out:
3228 398345 : spin_unlock(&head->lock);
3229 :
3230 398346 : out_delayed_unlock:
3231 398346 : spin_unlock(&delayed_refs->lock);
3232 398346 : return 0;
3233 : }
3234 :
3235 10895459 : void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3236 : u64 root_id,
3237 : struct extent_buffer *buf,
3238 : u64 parent, int last_ref)
3239 : {
3240 10895459 : struct btrfs_fs_info *fs_info = trans->fs_info;
3241 10895459 : struct btrfs_ref generic_ref = { 0 };
3242 10895459 : int ret;
3243 :
3244 10895459 : btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3245 10895459 : buf->start, buf->len, parent);
3246 10895459 : btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3247 : root_id, 0, false);
3248 :
3249 10895459 : if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3250 9825529 : btrfs_ref_tree_mod(fs_info, &generic_ref);
3251 9825529 : ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL);
3252 9825887 : BUG_ON(ret); /* -ENOMEM */
3253 : }
3254 :
3255 10895817 : if (last_ref && btrfs_header_generation(buf) == trans->transid) {
3256 1807073 : struct btrfs_block_group *cache;
3257 1807073 : bool must_pin = false;
3258 :
3259 1807073 : if (root_id != BTRFS_TREE_LOG_OBJECTID) {
3260 737141 : ret = check_ref_cleanup(trans, buf->start);
3261 737136 : if (!ret) {
3262 398346 : btrfs_redirty_list_add(trans->transaction, buf);
3263 398346 : goto out;
3264 : }
3265 : }
3266 :
3267 1408722 : cache = btrfs_lookup_block_group(fs_info, buf->start);
3268 :
3269 1408741 : if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3270 1367909 : pin_down_extent(trans, cache, buf->start, buf->len, 1);
3271 1367917 : btrfs_put_block_group(cache);
3272 1367916 : goto out;
3273 : }
3274 :
3275 : /*
3276 : * If there are tree mod log users we may have recorded mod log
3277 : * operations for this node. If we re-allocate this node we
3278 : * could replay operations on this node that happened when it
3279 : * existed in a completely different root. For example if it
3280 : * was part of root A, then was reallocated to root B, and we
3281 : * are doing a btrfs_old_search_slot(root b), we could replay
3282 : * operations that happened when the block was part of root A,
3283 : * giving us an inconsistent view of the btree.
3284 : *
3285 : * We are safe from races here because at this point no other
3286 : * node or root points to this extent buffer, so if after this
3287 : * check a new tree mod log user joins we will not have an
3288 : * existing log of operations on this node that we have to
3289 : * contend with.
3290 : */
3291 81664 : if (test_bit(BTRFS_FS_TREE_MOD_LOG_USERS, &fs_info->flags))
3292 2 : must_pin = true;
3293 :
3294 40832 : if (must_pin || btrfs_is_zoned(fs_info)) {
3295 2 : btrfs_redirty_list_add(trans->transaction, buf);
3296 2 : pin_down_extent(trans, cache, buf->start, buf->len, 1);
3297 2 : btrfs_put_block_group(cache);
3298 2 : goto out;
3299 : }
3300 :
3301 40830 : WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3302 :
3303 40830 : btrfs_add_free_space(cache, buf->start, buf->len);
3304 40830 : btrfs_free_reserved_bytes(cache, buf->len, 0);
3305 40830 : btrfs_put_block_group(cache);
3306 40830 : trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
3307 : }
3308 9088744 : out:
3309 10895838 : if (last_ref) {
3310 : /*
3311 : * Deleting the buffer, clear the corrupt flag since it doesn't
3312 : * matter anymore.
3313 : */
3314 8498640 : clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3315 : }
3316 10895854 : }
3317 :
3318 : /* Can return -ENOMEM */
3319 16044375 : int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
3320 : {
3321 16044375 : struct btrfs_fs_info *fs_info = trans->fs_info;
3322 16044375 : int ret;
3323 :
3324 16044375 : if (btrfs_is_testing(fs_info))
3325 : return 0;
3326 :
3327 : /*
3328 : * tree log blocks never actually go into the extent allocation
3329 : * tree, just update pinning info and exit early.
3330 : */
3331 16044375 : if ((ref->type == BTRFS_REF_METADATA &&
3332 16044375 : ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3333 11357119 : (ref->type == BTRFS_REF_DATA &&
3334 11357119 : ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)) {
3335 : /* unlocks the pinned mutex */
3336 0 : btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
3337 0 : ret = 0;
3338 16044375 : } else if (ref->type == BTRFS_REF_METADATA) {
3339 4687258 : ret = btrfs_add_delayed_tree_ref(trans, ref, NULL);
3340 : } else {
3341 11357117 : ret = btrfs_add_delayed_data_ref(trans, ref, 0);
3342 : }
3343 :
3344 16044382 : if (!((ref->type == BTRFS_REF_METADATA &&
3345 : ref->tree_ref.owning_root == BTRFS_TREE_LOG_OBJECTID) ||
3346 : (ref->type == BTRFS_REF_DATA &&
3347 : ref->data_ref.owning_root == BTRFS_TREE_LOG_OBJECTID)))
3348 : btrfs_ref_tree_mod(fs_info, ref);
3349 :
3350 16044382 : return ret;
3351 : }
3352 :
3353 : enum btrfs_loop_type {
3354 : LOOP_CACHING_NOWAIT,
3355 : LOOP_CACHING_WAIT,
3356 : LOOP_UNSET_SIZE_CLASS,
3357 : LOOP_ALLOC_CHUNK,
3358 : LOOP_WRONG_SIZE_CLASS,
3359 : LOOP_NO_EMPTY_SIZE,
3360 : };
3361 :
3362 : static inline void
3363 : btrfs_lock_block_group(struct btrfs_block_group *cache,
3364 : int delalloc)
3365 : {
3366 10621478 : if (delalloc)
3367 2016844 : down_read(&cache->data_rwsem);
3368 : }
3369 :
3370 13062784 : static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
3371 : int delalloc)
3372 : {
3373 13062784 : btrfs_get_block_group(cache);
3374 13062298 : if (delalloc)
3375 8996044 : down_read(&cache->data_rwsem);
3376 13062738 : }
3377 :
3378 8519835 : static struct btrfs_block_group *btrfs_lock_cluster(
3379 : struct btrfs_block_group *block_group,
3380 : struct btrfs_free_cluster *cluster,
3381 : int delalloc)
3382 : __acquires(&cluster->refill_lock)
3383 : {
3384 8519835 : struct btrfs_block_group *used_bg = NULL;
3385 :
3386 8519835 : spin_lock(&cluster->refill_lock);
3387 8520473 : while (1) {
3388 8520473 : used_bg = cluster->block_group;
3389 8520473 : if (!used_bg)
3390 : return NULL;
3391 :
3392 8510333 : if (used_bg == block_group)
3393 8510247 : return used_bg;
3394 :
3395 86 : btrfs_get_block_group(used_bg);
3396 :
3397 86 : if (!delalloc)
3398 86 : return used_bg;
3399 :
3400 0 : if (down_read_trylock(&used_bg->data_rwsem))
3401 0 : return used_bg;
3402 :
3403 0 : spin_unlock(&cluster->refill_lock);
3404 :
3405 : /* We should only have one-level nested. */
3406 0 : down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
3407 :
3408 0 : spin_lock(&cluster->refill_lock);
3409 0 : if (used_bg == cluster->block_group)
3410 0 : return used_bg;
3411 :
3412 0 : up_read(&used_bg->data_rwsem);
3413 0 : btrfs_put_block_group(used_bg);
3414 : }
3415 : }
3416 :
3417 : static inline void
3418 23684797 : btrfs_release_block_group(struct btrfs_block_group *cache,
3419 : int delalloc)
3420 : {
3421 23684797 : if (delalloc)
3422 11013762 : up_read(&cache->data_rwsem);
3423 23685090 : btrfs_put_block_group(cache);
3424 23685463 : }
3425 :
3426 : /*
3427 : * Helper function for find_free_extent().
3428 : *
3429 : * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3430 : * Return -EAGAIN to inform caller that we need to re-search this block group
3431 : * Return >0 to inform caller that we find nothing
3432 : * Return 0 means we have found a location and set ffe_ctl->found_offset.
3433 : */
3434 8519919 : static int find_free_extent_clustered(struct btrfs_block_group *bg,
3435 : struct find_free_extent_ctl *ffe_ctl,
3436 : struct btrfs_block_group **cluster_bg_ret)
3437 : {
3438 8519919 : struct btrfs_block_group *cluster_bg;
3439 8519919 : struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3440 8519919 : u64 aligned_cluster;
3441 8519919 : u64 offset;
3442 8519919 : int ret;
3443 :
3444 8519919 : cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3445 8520473 : if (!cluster_bg)
3446 10140 : goto refill_cluster;
3447 8510333 : if (cluster_bg != bg && (cluster_bg->ro ||
3448 0 : !block_group_bits(cluster_bg, ffe_ctl->flags)))
3449 86 : goto release_cluster;
3450 :
3451 8510247 : offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
3452 : ffe_ctl->num_bytes, cluster_bg->start,
3453 : &ffe_ctl->max_extent_size);
3454 8510247 : if (offset) {
3455 : /* We have a block, we're done */
3456 8503940 : spin_unlock(&last_ptr->refill_lock);
3457 8503940 : trace_btrfs_reserve_extent_cluster(cluster_bg, ffe_ctl);
3458 8503940 : *cluster_bg_ret = cluster_bg;
3459 8503940 : ffe_ctl->found_offset = offset;
3460 8503940 : return 0;
3461 : }
3462 6307 : WARN_ON(last_ptr->block_group != cluster_bg);
3463 :
3464 6393 : release_cluster:
3465 : /*
3466 : * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3467 : * lets just skip it and let the allocator find whatever block it can
3468 : * find. If we reach this point, we will have tried the cluster
3469 : * allocator plenty of times and not have found anything, so we are
3470 : * likely way too fragmented for the clustering stuff to find anything.
3471 : *
3472 : * However, if the cluster is taken from the current block group,
3473 : * release the cluster first, so that we stand a better chance of
3474 : * succeeding in the unclustered allocation.
3475 : */
3476 6393 : if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3477 0 : spin_unlock(&last_ptr->refill_lock);
3478 0 : btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3479 0 : return -ENOENT;
3480 : }
3481 :
3482 : /* This cluster didn't work out, free it and start over */
3483 6393 : btrfs_return_cluster_to_free_space(NULL, last_ptr);
3484 :
3485 6393 : if (cluster_bg != bg)
3486 86 : btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3487 :
3488 6307 : refill_cluster:
3489 16533 : if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3490 0 : spin_unlock(&last_ptr->refill_lock);
3491 0 : return -ENOENT;
3492 : }
3493 :
3494 16533 : aligned_cluster = max_t(u64,
3495 : ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3496 : bg->full_stripe_len);
3497 16533 : ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3498 : ffe_ctl->num_bytes, aligned_cluster);
3499 16533 : if (ret == 0) {
3500 : /* Now pull our allocation out of this cluster */
3501 9539 : offset = btrfs_alloc_from_cluster(bg, last_ptr,
3502 : ffe_ctl->num_bytes, ffe_ctl->search_start,
3503 : &ffe_ctl->max_extent_size);
3504 9539 : if (offset) {
3505 : /* We found one, proceed */
3506 9539 : spin_unlock(&last_ptr->refill_lock);
3507 9539 : ffe_ctl->found_offset = offset;
3508 9539 : trace_btrfs_reserve_extent_cluster(bg, ffe_ctl);
3509 9539 : return 0;
3510 : }
3511 6994 : } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3512 2254 : !ffe_ctl->retry_clustered) {
3513 2254 : spin_unlock(&last_ptr->refill_lock);
3514 :
3515 2254 : ffe_ctl->retry_clustered = true;
3516 2254 : btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3517 2254 : ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3518 2254 : return -EAGAIN;
3519 : }
3520 : /*
3521 : * At this point we either didn't find a cluster or we weren't able to
3522 : * allocate a block from our cluster. Free the cluster we've been
3523 : * trying to use, and go to the next block group.
3524 : */
3525 4740 : btrfs_return_cluster_to_free_space(NULL, last_ptr);
3526 4740 : spin_unlock(&last_ptr->refill_lock);
3527 4740 : return 1;
3528 : }
3529 :
3530 : /*
3531 : * Return >0 to inform caller that we find nothing
3532 : * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3533 : * Return -EAGAIN to inform caller that we need to re-search this block group
3534 : */
3535 7721784 : static int find_free_extent_unclustered(struct btrfs_block_group *bg,
3536 : struct find_free_extent_ctl *ffe_ctl)
3537 : {
3538 7721784 : struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3539 7721784 : u64 offset;
3540 :
3541 : /*
3542 : * We are doing an unclustered allocation, set the fragmented flag so
3543 : * we don't bother trying to setup a cluster again until we get more
3544 : * space.
3545 : */
3546 7721784 : if (unlikely(last_ptr)) {
3547 0 : spin_lock(&last_ptr->lock);
3548 0 : last_ptr->fragmented = 1;
3549 0 : spin_unlock(&last_ptr->lock);
3550 : }
3551 7721784 : if (ffe_ctl->cached) {
3552 7720358 : struct btrfs_free_space_ctl *free_space_ctl;
3553 :
3554 7720358 : free_space_ctl = bg->free_space_ctl;
3555 7720358 : spin_lock(&free_space_ctl->tree_lock);
3556 7721339 : if (free_space_ctl->free_space <
3557 7721339 : ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3558 7721339 : ffe_ctl->empty_size) {
3559 3608667 : ffe_ctl->total_free_space = max_t(u64,
3560 : ffe_ctl->total_free_space,
3561 : free_space_ctl->free_space);
3562 3608667 : spin_unlock(&free_space_ctl->tree_lock);
3563 3608667 : return 1;
3564 : }
3565 4112672 : spin_unlock(&free_space_ctl->tree_lock);
3566 : }
3567 :
3568 4114037 : offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3569 : ffe_ctl->num_bytes, ffe_ctl->empty_size,
3570 : &ffe_ctl->max_extent_size);
3571 :
3572 : /*
3573 : * If we didn't find a chunk, and we haven't failed on this block group
3574 : * before, and this block group is in the middle of caching and we are
3575 : * ok with waiting, then go ahead and wait for progress to be made, and
3576 : * set @retry_unclustered to true.
3577 : *
3578 : * If @retry_unclustered is true then we've already waited on this
3579 : * block group once and should move on to the next block group.
3580 : */
3581 4114112 : if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3582 1127 : ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
3583 741 : btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3584 741 : ffe_ctl->empty_size);
3585 741 : ffe_ctl->retry_unclustered = true;
3586 741 : return -EAGAIN;
3587 4113371 : } else if (!offset) {
3588 : return 1;
3589 : }
3590 3842004 : ffe_ctl->found_offset = offset;
3591 3842004 : return 0;
3592 : }
3593 :
3594 16241909 : static int do_allocation_clustered(struct btrfs_block_group *block_group,
3595 : struct find_free_extent_ctl *ffe_ctl,
3596 : struct btrfs_block_group **bg_ret)
3597 : {
3598 16241909 : int ret;
3599 :
3600 : /* We want to try and use the cluster allocator, so lets look there */
3601 16241909 : if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
3602 8519931 : ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
3603 8520473 : if (ret >= 0 || ret == -EAGAIN)
3604 : return ret;
3605 : /* ret == -ENOENT case falls through */
3606 : }
3607 :
3608 7721979 : return find_free_extent_unclustered(block_group, ffe_ctl);
3609 : }
3610 :
3611 : /*
3612 : * Tree-log block group locking
3613 : * ============================
3614 : *
3615 : * fs_info::treelog_bg_lock protects the fs_info::treelog_bg which
3616 : * indicates the starting address of a block group, which is reserved only
3617 : * for tree-log metadata.
3618 : *
3619 : * Lock nesting
3620 : * ============
3621 : *
3622 : * space_info::lock
3623 : * block_group::lock
3624 : * fs_info::treelog_bg_lock
3625 : */
3626 :
3627 : /*
3628 : * Simple allocator for sequential-only block group. It only allows sequential
3629 : * allocation. No need to play with trees. This function also reserves the
3630 : * bytes as in btrfs_add_reserved_bytes.
3631 : */
3632 0 : static int do_allocation_zoned(struct btrfs_block_group *block_group,
3633 : struct find_free_extent_ctl *ffe_ctl,
3634 : struct btrfs_block_group **bg_ret)
3635 : {
3636 0 : struct btrfs_fs_info *fs_info = block_group->fs_info;
3637 0 : struct btrfs_space_info *space_info = block_group->space_info;
3638 0 : struct btrfs_free_space_ctl *ctl = block_group->free_space_ctl;
3639 0 : u64 start = block_group->start;
3640 0 : u64 num_bytes = ffe_ctl->num_bytes;
3641 0 : u64 avail;
3642 0 : u64 bytenr = block_group->start;
3643 0 : u64 log_bytenr;
3644 0 : u64 data_reloc_bytenr;
3645 0 : int ret = 0;
3646 0 : bool skip = false;
3647 :
3648 0 : ASSERT(btrfs_is_zoned(block_group->fs_info));
3649 :
3650 : /*
3651 : * Do not allow non-tree-log blocks in the dedicated tree-log block
3652 : * group, and vice versa.
3653 : */
3654 0 : spin_lock(&fs_info->treelog_bg_lock);
3655 0 : log_bytenr = fs_info->treelog_bg;
3656 0 : if (log_bytenr && ((ffe_ctl->for_treelog && bytenr != log_bytenr) ||
3657 0 : (!ffe_ctl->for_treelog && bytenr == log_bytenr)))
3658 0 : skip = true;
3659 0 : spin_unlock(&fs_info->treelog_bg_lock);
3660 0 : if (skip)
3661 : return 1;
3662 :
3663 : /*
3664 : * Do not allow non-relocation blocks in the dedicated relocation block
3665 : * group, and vice versa.
3666 : */
3667 0 : spin_lock(&fs_info->relocation_bg_lock);
3668 0 : data_reloc_bytenr = fs_info->data_reloc_bg;
3669 0 : if (data_reloc_bytenr &&
3670 0 : ((ffe_ctl->for_data_reloc && bytenr != data_reloc_bytenr) ||
3671 0 : (!ffe_ctl->for_data_reloc && bytenr == data_reloc_bytenr)))
3672 0 : skip = true;
3673 0 : spin_unlock(&fs_info->relocation_bg_lock);
3674 0 : if (skip)
3675 : return 1;
3676 :
3677 : /* Check RO and no space case before trying to activate it */
3678 0 : spin_lock(&block_group->lock);
3679 0 : if (block_group->ro || btrfs_zoned_bg_is_full(block_group)) {
3680 : ret = 1;
3681 : /*
3682 : * May need to clear fs_info->{treelog,data_reloc}_bg.
3683 : * Return the error after taking the locks.
3684 : */
3685 : }
3686 0 : spin_unlock(&block_group->lock);
3687 :
3688 0 : if (!ret && !btrfs_zone_activate(block_group)) {
3689 : ret = 1;
3690 : /*
3691 : * May need to clear fs_info->{treelog,data_reloc}_bg.
3692 : * Return the error after taking the locks.
3693 : */
3694 : }
3695 :
3696 0 : spin_lock(&space_info->lock);
3697 0 : spin_lock(&block_group->lock);
3698 0 : spin_lock(&fs_info->treelog_bg_lock);
3699 0 : spin_lock(&fs_info->relocation_bg_lock);
3700 :
3701 0 : if (ret)
3702 0 : goto out;
3703 :
3704 0 : ASSERT(!ffe_ctl->for_treelog ||
3705 : block_group->start == fs_info->treelog_bg ||
3706 : fs_info->treelog_bg == 0);
3707 0 : ASSERT(!ffe_ctl->for_data_reloc ||
3708 : block_group->start == fs_info->data_reloc_bg ||
3709 : fs_info->data_reloc_bg == 0);
3710 :
3711 0 : if (block_group->ro ||
3712 0 : test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags)) {
3713 0 : ret = 1;
3714 0 : goto out;
3715 : }
3716 :
3717 : /*
3718 : * Do not allow currently using block group to be tree-log dedicated
3719 : * block group.
3720 : */
3721 0 : if (ffe_ctl->for_treelog && !fs_info->treelog_bg &&
3722 0 : (block_group->used || block_group->reserved)) {
3723 0 : ret = 1;
3724 0 : goto out;
3725 : }
3726 :
3727 : /*
3728 : * Do not allow currently used block group to be the data relocation
3729 : * dedicated block group.
3730 : */
3731 0 : if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg &&
3732 0 : (block_group->used || block_group->reserved)) {
3733 0 : ret = 1;
3734 0 : goto out;
3735 : }
3736 :
3737 0 : WARN_ON_ONCE(block_group->alloc_offset > block_group->zone_capacity);
3738 0 : avail = block_group->zone_capacity - block_group->alloc_offset;
3739 0 : if (avail < num_bytes) {
3740 0 : if (ffe_ctl->max_extent_size < avail) {
3741 : /*
3742 : * With sequential allocator, free space is always
3743 : * contiguous
3744 : */
3745 0 : ffe_ctl->max_extent_size = avail;
3746 0 : ffe_ctl->total_free_space = avail;
3747 : }
3748 0 : ret = 1;
3749 0 : goto out;
3750 : }
3751 :
3752 0 : if (ffe_ctl->for_treelog && !fs_info->treelog_bg)
3753 0 : fs_info->treelog_bg = block_group->start;
3754 :
3755 0 : if (ffe_ctl->for_data_reloc && !fs_info->data_reloc_bg)
3756 0 : fs_info->data_reloc_bg = block_group->start;
3757 :
3758 0 : ffe_ctl->found_offset = start + block_group->alloc_offset;
3759 0 : block_group->alloc_offset += num_bytes;
3760 0 : spin_lock(&ctl->tree_lock);
3761 0 : ctl->free_space -= num_bytes;
3762 0 : spin_unlock(&ctl->tree_lock);
3763 :
3764 : /*
3765 : * We do not check if found_offset is aligned to stripesize. The
3766 : * address is anyway rewritten when using zone append writing.
3767 : */
3768 :
3769 0 : ffe_ctl->search_start = ffe_ctl->found_offset;
3770 :
3771 0 : out:
3772 0 : if (ret && ffe_ctl->for_treelog)
3773 0 : fs_info->treelog_bg = 0;
3774 0 : if (ret && ffe_ctl->for_data_reloc &&
3775 0 : fs_info->data_reloc_bg == block_group->start) {
3776 : /*
3777 : * Do not allow further allocations from this block group.
3778 : * Compared to increasing the ->ro, setting the
3779 : * ->zoned_data_reloc_ongoing flag still allows nocow
3780 : * writers to come in. See btrfs_inc_nocow_writers().
3781 : *
3782 : * We need to disable an allocation to avoid an allocation of
3783 : * regular (non-relocation data) extent. With mix of relocation
3784 : * extents and regular extents, we can dispatch WRITE commands
3785 : * (for relocation extents) and ZONE APPEND commands (for
3786 : * regular extents) at the same time to the same zone, which
3787 : * easily break the write pointer.
3788 : */
3789 0 : set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags);
3790 0 : fs_info->data_reloc_bg = 0;
3791 : }
3792 0 : spin_unlock(&fs_info->relocation_bg_lock);
3793 0 : spin_unlock(&fs_info->treelog_bg_lock);
3794 0 : spin_unlock(&block_group->lock);
3795 0 : spin_unlock(&space_info->lock);
3796 0 : return ret;
3797 : }
3798 :
3799 16242004 : static int do_allocation(struct btrfs_block_group *block_group,
3800 : struct find_free_extent_ctl *ffe_ctl,
3801 : struct btrfs_block_group **bg_ret)
3802 : {
3803 16242004 : switch (ffe_ctl->policy) {
3804 16242004 : case BTRFS_EXTENT_ALLOC_CLUSTERED:
3805 16242004 : return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3806 0 : case BTRFS_EXTENT_ALLOC_ZONED:
3807 0 : return do_allocation_zoned(block_group, ffe_ctl, bg_ret);
3808 0 : default:
3809 0 : BUG();
3810 : }
3811 : }
3812 :
3813 11329727 : static void release_block_group(struct btrfs_block_group *block_group,
3814 : struct find_free_extent_ctl *ffe_ctl,
3815 : int delalloc)
3816 : {
3817 11329727 : switch (ffe_ctl->policy) {
3818 11329916 : case BTRFS_EXTENT_ALLOC_CLUSTERED:
3819 11329916 : ffe_ctl->retry_clustered = false;
3820 11329916 : ffe_ctl->retry_unclustered = false;
3821 11329916 : break;
3822 : case BTRFS_EXTENT_ALLOC_ZONED:
3823 : /* Nothing to do */
3824 : break;
3825 0 : default:
3826 0 : BUG();
3827 : }
3828 :
3829 11329727 : BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
3830 : ffe_ctl->index);
3831 11329727 : btrfs_release_block_group(block_group, delalloc);
3832 11330061 : }
3833 :
3834 12355281 : static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
3835 : struct btrfs_key *ins)
3836 : {
3837 12355281 : struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3838 :
3839 12355281 : if (!ffe_ctl->use_cluster && last_ptr) {
3840 0 : spin_lock(&last_ptr->lock);
3841 0 : last_ptr->window_start = ins->objectid;
3842 0 : spin_unlock(&last_ptr->lock);
3843 : }
3844 12355281 : }
3845 :
3846 12355304 : static void found_extent(struct find_free_extent_ctl *ffe_ctl,
3847 : struct btrfs_key *ins)
3848 : {
3849 12355304 : switch (ffe_ctl->policy) {
3850 12355304 : case BTRFS_EXTENT_ALLOC_CLUSTERED:
3851 12355304 : found_extent_clustered(ffe_ctl, ins);
3852 12355304 : break;
3853 : case BTRFS_EXTENT_ALLOC_ZONED:
3854 : /* Nothing to do */
3855 : break;
3856 0 : default:
3857 0 : BUG();
3858 : }
3859 12355263 : }
3860 :
3861 : static int can_allocate_chunk_zoned(struct btrfs_fs_info *fs_info,
3862 : struct find_free_extent_ctl *ffe_ctl)
3863 : {
3864 : /* If we can activate new zone, just allocate a chunk and use it */
3865 : if (btrfs_can_activate_zone(fs_info->fs_devices, ffe_ctl->flags))
3866 : return 0;
3867 :
3868 : /*
3869 : * We already reached the max active zones. Try to finish one block
3870 : * group to make a room for a new block group. This is only possible
3871 : * for a data block group because btrfs_zone_finish() may need to wait
3872 : * for a running transaction which can cause a deadlock for metadata
3873 : * allocation.
3874 : */
3875 : if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA) {
3876 : int ret = btrfs_zone_finish_one_bg(fs_info);
3877 :
3878 : if (ret == 1)
3879 : return 0;
3880 : else if (ret < 0)
3881 : return ret;
3882 : }
3883 :
3884 : /*
3885 : * If we have enough free space left in an already active block group
3886 : * and we can't activate any other zone now, do not allow allocating a
3887 : * new chunk and let find_free_extent() retry with a smaller size.
3888 : */
3889 : if (ffe_ctl->max_extent_size >= ffe_ctl->min_alloc_size)
3890 : return -ENOSPC;
3891 :
3892 : /*
3893 : * Even min_alloc_size is not left in any block groups. Since we cannot
3894 : * activate a new block group, allocating it may not help. Let's tell a
3895 : * caller to try again and hope it progress something by writing some
3896 : * parts of the region. That is only possible for data block groups,
3897 : * where a part of the region can be written.
3898 : */
3899 : if (ffe_ctl->flags & BTRFS_BLOCK_GROUP_DATA)
3900 : return -EAGAIN;
3901 :
3902 : /*
3903 : * We cannot activate a new block group and no enough space left in any
3904 : * block groups. So, allocating a new block group may not help. But,
3905 : * there is nothing to do anyway, so let's go with it.
3906 : */
3907 : return 0;
3908 : }
3909 :
3910 : static int can_allocate_chunk(struct btrfs_fs_info *fs_info,
3911 : struct find_free_extent_ctl *ffe_ctl)
3912 : {
3913 42389 : switch (ffe_ctl->policy) {
3914 : case BTRFS_EXTENT_ALLOC_CLUSTERED:
3915 : return 0;
3916 : case BTRFS_EXTENT_ALLOC_ZONED:
3917 : return can_allocate_chunk_zoned(fs_info, ffe_ctl);
3918 0 : default:
3919 0 : BUG();
3920 : }
3921 : }
3922 :
3923 : /*
3924 : * Return >0 means caller needs to re-search for free extent
3925 : * Return 0 means we have the needed free extent.
3926 : * Return <0 means we failed to locate any free extent.
3927 : */
3928 14767068 : static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
3929 : struct btrfs_key *ins,
3930 : struct find_free_extent_ctl *ffe_ctl,
3931 : bool full_search)
3932 : {
3933 14767068 : struct btrfs_root *root = fs_info->chunk_root;
3934 14767068 : int ret;
3935 :
3936 14767068 : if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
3937 13187071 : ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
3938 4179 : ffe_ctl->orig_have_caching_bg = true;
3939 :
3940 14767068 : if (ins->objectid) {
3941 12355348 : found_extent(ffe_ctl, ins);
3942 12355348 : return 0;
3943 : }
3944 :
3945 2411720 : if (ffe_ctl->loop >= LOOP_CACHING_WAIT && ffe_ctl->have_caching_bg)
3946 : return 1;
3947 :
3948 1707841 : ffe_ctl->index++;
3949 1707841 : if (ffe_ctl->index < BTRFS_NR_RAID_TYPES)
3950 : return 1;
3951 :
3952 : /*
3953 : * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
3954 : * caching kthreads as we move along
3955 : * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3956 : * LOOP_UNSET_SIZE_CLASS, allow unset size class
3957 : * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3958 : * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3959 : * again
3960 : */
3961 189764 : if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
3962 189764 : ffe_ctl->index = 0;
3963 : /*
3964 : * We want to skip the LOOP_CACHING_WAIT step if we don't have
3965 : * any uncached bgs and we've already done a full search
3966 : * through.
3967 : */
3968 189764 : if (ffe_ctl->loop == LOOP_CACHING_NOWAIT &&
3969 104843 : (!ffe_ctl->orig_have_caching_bg && full_search))
3970 5066 : ffe_ctl->loop++;
3971 189764 : ffe_ctl->loop++;
3972 :
3973 189764 : if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
3974 42389 : struct btrfs_trans_handle *trans;
3975 42389 : int exist = 0;
3976 :
3977 : /* Check if allocation policy allows to create a new chunk */
3978 42389 : ret = can_allocate_chunk(fs_info, ffe_ctl);
3979 42389 : if (ret)
3980 : return ret;
3981 :
3982 42389 : trans = current->journal_info;
3983 42389 : if (trans)
3984 : exist = 1;
3985 : else
3986 42175 : trans = btrfs_join_transaction(root);
3987 :
3988 42390 : if (IS_ERR(trans)) {
3989 0 : ret = PTR_ERR(trans);
3990 0 : return ret;
3991 : }
3992 :
3993 42390 : ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
3994 : CHUNK_ALLOC_FORCE_FOR_EXTENT);
3995 :
3996 : /* Do not bail out on ENOSPC since we can do more. */
3997 42390 : if (ret == -ENOSPC) {
3998 41805 : ret = 0;
3999 41805 : ffe_ctl->loop++;
4000 : }
4001 585 : else if (ret < 0)
4002 0 : btrfs_abort_transaction(trans, ret);
4003 : else
4004 : ret = 0;
4005 42390 : if (!exist)
4006 42175 : btrfs_end_transaction(trans);
4007 42388 : if (ret)
4008 : return ret;
4009 : }
4010 :
4011 189763 : if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
4012 3947 : if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
4013 : return -ENOSPC;
4014 :
4015 : /*
4016 : * Don't loop again if we already have no empty_size and
4017 : * no empty_cluster.
4018 : */
4019 3947 : if (ffe_ctl->empty_size == 0 &&
4020 3947 : ffe_ctl->empty_cluster == 0)
4021 : return -ENOSPC;
4022 0 : ffe_ctl->empty_size = 0;
4023 0 : ffe_ctl->empty_cluster = 0;
4024 : }
4025 185816 : return 1;
4026 : }
4027 : return -ENOSPC;
4028 : }
4029 :
4030 23687982 : static bool find_free_extent_check_size_class(struct find_free_extent_ctl *ffe_ctl,
4031 : struct btrfs_block_group *bg)
4032 : {
4033 23687982 : if (ffe_ctl->policy == BTRFS_EXTENT_ALLOC_ZONED)
4034 : return true;
4035 23687982 : if (!btrfs_block_group_should_use_size_class(bg))
4036 : return true;
4037 13503883 : if (ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS)
4038 : return true;
4039 13246791 : if (ffe_ctl->loop >= LOOP_UNSET_SIZE_CLASS &&
4040 391110 : bg->size_class == BTRFS_BG_SZ_NONE)
4041 : return true;
4042 13244458 : return ffe_ctl->size_class == bg->size_class;
4043 : }
4044 :
4045 12364385 : static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
4046 : struct find_free_extent_ctl *ffe_ctl,
4047 : struct btrfs_space_info *space_info,
4048 : struct btrfs_key *ins)
4049 : {
4050 : /*
4051 : * If our free space is heavily fragmented we may not be able to make
4052 : * big contiguous allocations, so instead of doing the expensive search
4053 : * for free space, simply return ENOSPC with our max_extent_size so we
4054 : * can go ahead and search for a more manageable chunk.
4055 : *
4056 : * If our max_extent_size is large enough for our allocation simply
4057 : * disable clustering since we will likely not be able to find enough
4058 : * space to create a cluster and induce latency trying.
4059 : */
4060 12364385 : if (space_info->max_extent_size) {
4061 22511 : spin_lock(&space_info->lock);
4062 22512 : if (space_info->max_extent_size &&
4063 22510 : ffe_ctl->num_bytes > space_info->max_extent_size) {
4064 5981 : ins->offset = space_info->max_extent_size;
4065 5981 : spin_unlock(&space_info->lock);
4066 5981 : return -ENOSPC;
4067 16531 : } else if (space_info->max_extent_size) {
4068 16529 : ffe_ctl->use_cluster = false;
4069 : }
4070 16531 : spin_unlock(&space_info->lock);
4071 : }
4072 :
4073 12358405 : ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
4074 : &ffe_ctl->empty_cluster);
4075 12358180 : if (ffe_ctl->last_ptr) {
4076 8512982 : struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
4077 :
4078 8512982 : spin_lock(&last_ptr->lock);
4079 8513479 : if (last_ptr->block_group)
4080 8510320 : ffe_ctl->hint_byte = last_ptr->window_start;
4081 8513479 : if (last_ptr->fragmented) {
4082 : /*
4083 : * We still set window_start so we can keep track of the
4084 : * last place we found an allocation to try and save
4085 : * some time.
4086 : */
4087 0 : ffe_ctl->hint_byte = last_ptr->window_start;
4088 0 : ffe_ctl->use_cluster = false;
4089 : }
4090 8513479 : spin_unlock(&last_ptr->lock);
4091 : }
4092 :
4093 : return 0;
4094 : }
4095 :
4096 12364443 : static int prepare_allocation(struct btrfs_fs_info *fs_info,
4097 : struct find_free_extent_ctl *ffe_ctl,
4098 : struct btrfs_space_info *space_info,
4099 : struct btrfs_key *ins)
4100 : {
4101 12364443 : switch (ffe_ctl->policy) {
4102 12364443 : case BTRFS_EXTENT_ALLOC_CLUSTERED:
4103 12364443 : return prepare_allocation_clustered(fs_info, ffe_ctl,
4104 : space_info, ins);
4105 0 : case BTRFS_EXTENT_ALLOC_ZONED:
4106 0 : if (ffe_ctl->for_treelog) {
4107 0 : spin_lock(&fs_info->treelog_bg_lock);
4108 0 : if (fs_info->treelog_bg)
4109 0 : ffe_ctl->hint_byte = fs_info->treelog_bg;
4110 0 : spin_unlock(&fs_info->treelog_bg_lock);
4111 : }
4112 0 : if (ffe_ctl->for_data_reloc) {
4113 0 : spin_lock(&fs_info->relocation_bg_lock);
4114 0 : if (fs_info->data_reloc_bg)
4115 0 : ffe_ctl->hint_byte = fs_info->data_reloc_bg;
4116 0 : spin_unlock(&fs_info->relocation_bg_lock);
4117 : }
4118 : return 0;
4119 0 : default:
4120 0 : BUG();
4121 : }
4122 : }
4123 :
4124 : /*
4125 : * walks the btree of allocated extents and find a hole of a given size.
4126 : * The key ins is changed to record the hole:
4127 : * ins->objectid == start position
4128 : * ins->flags = BTRFS_EXTENT_ITEM_KEY
4129 : * ins->offset == the size of the hole.
4130 : * Any available blocks before search_start are skipped.
4131 : *
4132 : * If there is no suitable free space, we will record the max size of
4133 : * the free space extent currently.
4134 : *
4135 : * The overall logic and call chain:
4136 : *
4137 : * find_free_extent()
4138 : * |- Iterate through all block groups
4139 : * | |- Get a valid block group
4140 : * | |- Try to do clustered allocation in that block group
4141 : * | |- Try to do unclustered allocation in that block group
4142 : * | |- Check if the result is valid
4143 : * | | |- If valid, then exit
4144 : * | |- Jump to next block group
4145 : * |
4146 : * |- Push harder to find free extents
4147 : * |- If not found, re-iterate all block groups
4148 : */
4149 12365114 : static noinline int find_free_extent(struct btrfs_root *root,
4150 : struct btrfs_key *ins,
4151 : struct find_free_extent_ctl *ffe_ctl)
4152 : {
4153 12365114 : struct btrfs_fs_info *fs_info = root->fs_info;
4154 12365114 : int ret = 0;
4155 12365114 : int cache_block_group_error = 0;
4156 12365114 : struct btrfs_block_group *block_group = NULL;
4157 12365114 : struct btrfs_space_info *space_info;
4158 12365114 : bool full_search = false;
4159 :
4160 12365114 : WARN_ON(ffe_ctl->num_bytes < fs_info->sectorsize);
4161 :
4162 12365114 : ffe_ctl->search_start = 0;
4163 : /* For clustered allocation */
4164 12365114 : ffe_ctl->empty_cluster = 0;
4165 12365114 : ffe_ctl->last_ptr = NULL;
4166 12365114 : ffe_ctl->use_cluster = true;
4167 12365114 : ffe_ctl->have_caching_bg = false;
4168 12365114 : ffe_ctl->orig_have_caching_bg = false;
4169 12365114 : ffe_ctl->index = btrfs_bg_flags_to_raid_index(ffe_ctl->flags);
4170 12365114 : ffe_ctl->loop = 0;
4171 : /* For clustered allocation */
4172 12365114 : ffe_ctl->retry_clustered = false;
4173 12365114 : ffe_ctl->retry_unclustered = false;
4174 12365114 : ffe_ctl->cached = 0;
4175 12365114 : ffe_ctl->max_extent_size = 0;
4176 12365114 : ffe_ctl->total_free_space = 0;
4177 12365114 : ffe_ctl->found_offset = 0;
4178 12365114 : ffe_ctl->policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
4179 12365114 : ffe_ctl->size_class = btrfs_calc_block_group_size_class(ffe_ctl->num_bytes);
4180 :
4181 12364539 : if (btrfs_is_zoned(fs_info))
4182 : ffe_ctl->policy = BTRFS_EXTENT_ALLOC_ZONED;
4183 :
4184 12364539 : ins->type = BTRFS_EXTENT_ITEM_KEY;
4185 12364539 : ins->objectid = 0;
4186 12364539 : ins->offset = 0;
4187 :
4188 12364539 : trace_find_free_extent(root, ffe_ctl);
4189 :
4190 12364422 : space_info = btrfs_find_space_info(fs_info, ffe_ctl->flags);
4191 12364469 : if (!space_info) {
4192 0 : btrfs_err(fs_info, "No space info for %llu", ffe_ctl->flags);
4193 0 : return -ENOSPC;
4194 : }
4195 :
4196 12364469 : ret = prepare_allocation(fs_info, ffe_ctl, space_info, ins);
4197 12364328 : if (ret < 0)
4198 : return ret;
4199 :
4200 12358347 : ffe_ctl->search_start = max(ffe_ctl->search_start,
4201 : first_logical_byte(fs_info));
4202 12358932 : ffe_ctl->search_start = max(ffe_ctl->search_start, ffe_ctl->hint_byte);
4203 12358932 : if (ffe_ctl->search_start == ffe_ctl->hint_byte) {
4204 10623361 : block_group = btrfs_lookup_block_group(fs_info,
4205 : ffe_ctl->search_start);
4206 : /*
4207 : * we don't want to use the block group if it doesn't match our
4208 : * allocation bits, or if its not cached.
4209 : *
4210 : * However if we are re-searching with an ideal block group
4211 : * picked out then we don't care that the block group is cached.
4212 : */
4213 10623027 : if (block_group && block_group_bits(block_group, ffe_ctl->flags) &&
4214 10622293 : block_group->cached != BTRFS_CACHE_NO) {
4215 10622236 : down_read(&space_info->groups_sem);
4216 10622266 : if (list_empty(&block_group->list) ||
4217 10622266 : block_group->ro) {
4218 : /*
4219 : * someone is removing this block group,
4220 : * we can't jump into the have_block_group
4221 : * target because our list pointers are not
4222 : * valid
4223 : */
4224 788 : btrfs_put_block_group(block_group);
4225 616 : up_read(&space_info->groups_sem);
4226 : } else {
4227 10621478 : ffe_ctl->index = btrfs_bg_flags_to_raid_index(
4228 : block_group->flags);
4229 10621478 : btrfs_lock_block_group(block_group,
4230 : ffe_ctl->delalloc);
4231 10621518 : ffe_ctl->hinted = true;
4232 10621518 : goto have_block_group;
4233 : }
4234 791 : } else if (block_group) {
4235 403 : btrfs_put_block_group(block_group);
4236 : }
4237 : }
4238 1735959 : search:
4239 4144734 : trace_find_free_extent_search_loop(root, ffe_ctl);
4240 4144770 : ffe_ctl->have_caching_bg = false;
4241 4144770 : if (ffe_ctl->index == btrfs_bg_flags_to_raid_index(ffe_ctl->flags) ||
4242 : ffe_ctl->index == 0)
4243 2630715 : full_search = true;
4244 4144770 : down_read(&space_info->groups_sem);
4245 15482121 : list_for_each_entry(block_group,
4246 : &space_info->block_groups[ffe_ctl->index], list) {
4247 13070376 : struct btrfs_block_group *bg_ret;
4248 :
4249 13070376 : ffe_ctl->hinted = false;
4250 : /* If the block group is read-only, we can skip it entirely. */
4251 13070376 : if (unlikely(block_group->ro)) {
4252 7476 : if (ffe_ctl->for_treelog)
4253 : btrfs_clear_treelog_bg(block_group);
4254 7476 : if (ffe_ctl->for_data_reloc)
4255 : btrfs_clear_data_reloc_bg(block_group);
4256 7476 : continue;
4257 : }
4258 :
4259 13062900 : btrfs_grab_block_group(block_group, ffe_ctl->delalloc);
4260 13062579 : ffe_ctl->search_start = block_group->start;
4261 :
4262 : /*
4263 : * this can happen if we end up cycling through all the
4264 : * raid types, but we want to make sure we only allocate
4265 : * for the proper type.
4266 : */
4267 13062579 : if (!block_group_bits(block_group, ffe_ctl->flags)) {
4268 0 : u64 extra = BTRFS_BLOCK_GROUP_DUP |
4269 : BTRFS_BLOCK_GROUP_RAID1_MASK |
4270 : BTRFS_BLOCK_GROUP_RAID56_MASK |
4271 : BTRFS_BLOCK_GROUP_RAID10;
4272 :
4273 : /*
4274 : * if they asked for extra copies and this block group
4275 : * doesn't provide them, bail. This does allow us to
4276 : * fill raid0 from raid1.
4277 : */
4278 0 : if ((ffe_ctl->flags & extra) && !(block_group->flags & extra))
4279 0 : goto loop;
4280 :
4281 : /*
4282 : * This block group has different flags than we want.
4283 : * It's possible that we have MIXED_GROUP flag but no
4284 : * block group is mixed. Just skip such block group.
4285 : */
4286 0 : btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4287 0 : continue;
4288 : }
4289 :
4290 23684097 : have_block_group:
4291 23687092 : trace_find_free_extent_have_block_group(root, ffe_ctl, block_group);
4292 23685966 : ffe_ctl->cached = btrfs_block_group_done(block_group);
4293 23688180 : if (unlikely(!ffe_ctl->cached)) {
4294 1515640 : ffe_ctl->have_caching_bg = true;
4295 1515640 : ret = btrfs_cache_block_group(block_group, false);
4296 :
4297 : /*
4298 : * If we get ENOMEM here or something else we want to
4299 : * try other block groups, because it may not be fatal.
4300 : * However if we can't find anything else we need to
4301 : * save our return here so that we return the actual
4302 : * error that caused problems, not ENOSPC.
4303 : */
4304 1515636 : if (ret < 0) {
4305 0 : if (!cache_block_group_error)
4306 0 : cache_block_group_error = ret;
4307 0 : ret = 0;
4308 0 : goto loop;
4309 : }
4310 : ret = 0;
4311 : }
4312 :
4313 23688176 : if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
4314 0 : goto loop;
4315 :
4316 23688176 : if (!find_free_extent_check_size_class(ffe_ctl, block_group))
4317 7445433 : goto loop;
4318 :
4319 16242122 : bg_ret = NULL;
4320 16242122 : ret = do_allocation(block_group, ffe_ctl, &bg_ret);
4321 16243075 : if (ret == 0) {
4322 12355430 : if (bg_ret && bg_ret != block_group) {
4323 0 : btrfs_release_block_group(block_group,
4324 : ffe_ctl->delalloc);
4325 0 : block_group = bg_ret;
4326 : }
4327 3887645 : } else if (ret == -EAGAIN) {
4328 2995 : goto have_block_group;
4329 3884650 : } else if (ret > 0) {
4330 3884634 : goto loop;
4331 : }
4332 :
4333 : /* Checks */
4334 12355446 : ffe_ctl->search_start = round_up(ffe_ctl->found_offset,
4335 : fs_info->stripesize);
4336 :
4337 : /* move on to the next group */
4338 12355446 : if (ffe_ctl->search_start + ffe_ctl->num_bytes >
4339 12355446 : block_group->start + block_group->length) {
4340 0 : btrfs_add_free_space_unused(block_group,
4341 : ffe_ctl->found_offset,
4342 : ffe_ctl->num_bytes);
4343 0 : goto loop;
4344 : }
4345 :
4346 12355446 : if (ffe_ctl->found_offset < ffe_ctl->search_start)
4347 0 : btrfs_add_free_space_unused(block_group,
4348 : ffe_ctl->found_offset,
4349 : ffe_ctl->search_start - ffe_ctl->found_offset);
4350 :
4351 12355446 : ret = btrfs_add_reserved_bytes(block_group, ffe_ctl->ram_bytes,
4352 : ffe_ctl->num_bytes,
4353 : ffe_ctl->delalloc,
4354 12355446 : ffe_ctl->loop >= LOOP_WRONG_SIZE_CLASS);
4355 12355409 : if (ret == -EAGAIN) {
4356 0 : btrfs_add_free_space_unused(block_group,
4357 : ffe_ctl->found_offset,
4358 : ffe_ctl->num_bytes);
4359 0 : goto loop;
4360 : }
4361 12355409 : btrfs_inc_block_group_reservations(block_group);
4362 :
4363 : /* we are all good, lets return */
4364 12355470 : ins->objectid = ffe_ctl->search_start;
4365 12355470 : ins->offset = ffe_ctl->num_bytes;
4366 :
4367 12355470 : trace_btrfs_reserve_extent(block_group, ffe_ctl);
4368 12355449 : btrfs_release_block_group(block_group, ffe_ctl->delalloc);
4369 12355415 : break;
4370 11330067 : loop:
4371 11330067 : release_block_group(block_group, ffe_ctl, ffe_ctl->delalloc);
4372 11329967 : cond_resched();
4373 : }
4374 14767160 : up_read(&space_info->groups_sem);
4375 :
4376 14767060 : ret = find_free_extent_update_loop(fs_info, ins, ffe_ctl, full_search);
4377 14766954 : if (ret > 0)
4378 2407756 : goto search;
4379 :
4380 12359198 : if (ret == -ENOSPC && !cache_block_group_error) {
4381 : /*
4382 : * Use ffe_ctl->total_free_space as fallback if we can't find
4383 : * any contiguous hole.
4384 : */
4385 3947 : if (!ffe_ctl->max_extent_size)
4386 1539 : ffe_ctl->max_extent_size = ffe_ctl->total_free_space;
4387 3947 : spin_lock(&space_info->lock);
4388 3947 : space_info->max_extent_size = ffe_ctl->max_extent_size;
4389 3947 : spin_unlock(&space_info->lock);
4390 3947 : ins->offset = ffe_ctl->max_extent_size;
4391 12355251 : } else if (ret == -ENOSPC) {
4392 0 : ret = cache_block_group_error;
4393 : }
4394 : return ret;
4395 : }
4396 :
4397 : /*
4398 : * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4399 : * hole that is at least as big as @num_bytes.
4400 : *
4401 : * @root - The root that will contain this extent
4402 : *
4403 : * @ram_bytes - The amount of space in ram that @num_bytes take. This
4404 : * is used for accounting purposes. This value differs
4405 : * from @num_bytes only in the case of compressed extents.
4406 : *
4407 : * @num_bytes - Number of bytes to allocate on-disk.
4408 : *
4409 : * @min_alloc_size - Indicates the minimum amount of space that the
4410 : * allocator should try to satisfy. In some cases
4411 : * @num_bytes may be larger than what is required and if
4412 : * the filesystem is fragmented then allocation fails.
4413 : * However, the presence of @min_alloc_size gives a
4414 : * chance to try and satisfy the smaller allocation.
4415 : *
4416 : * @empty_size - A hint that you plan on doing more COW. This is the
4417 : * size in bytes the allocator should try to find free
4418 : * next to the block it returns. This is just a hint and
4419 : * may be ignored by the allocator.
4420 : *
4421 : * @hint_byte - Hint to the allocator to start searching above the byte
4422 : * address passed. It might be ignored.
4423 : *
4424 : * @ins - This key is modified to record the found hole. It will
4425 : * have the following values:
4426 : * ins->objectid == start position
4427 : * ins->flags = BTRFS_EXTENT_ITEM_KEY
4428 : * ins->offset == the size of the hole.
4429 : *
4430 : * @is_data - Boolean flag indicating whether an extent is
4431 : * allocated for data (true) or metadata (false)
4432 : *
4433 : * @delalloc - Boolean flag indicating whether this allocation is for
4434 : * delalloc or not. If 'true' data_rwsem of block groups
4435 : * is going to be acquired.
4436 : *
4437 : *
4438 : * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4439 : * case -ENOSPC is returned then @ins->offset will contain the size of the
4440 : * largest available hole the allocator managed to find.
4441 : */
4442 12355062 : int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
4443 : u64 num_bytes, u64 min_alloc_size,
4444 : u64 empty_size, u64 hint_byte,
4445 : struct btrfs_key *ins, int is_data, int delalloc)
4446 : {
4447 12355062 : struct btrfs_fs_info *fs_info = root->fs_info;
4448 12355062 : struct find_free_extent_ctl ffe_ctl = {};
4449 12355062 : bool final_tried = num_bytes == min_alloc_size;
4450 12355062 : u64 flags;
4451 12355062 : int ret;
4452 12355062 : bool for_treelog = (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4453 12355062 : bool for_data_reloc = (btrfs_is_data_reloc_root(root) && is_data);
4454 :
4455 12355062 : flags = get_alloc_profile_by_root(root, is_data);
4456 12365242 : again:
4457 12365242 : WARN_ON(num_bytes < fs_info->sectorsize);
4458 :
4459 12365242 : ffe_ctl.ram_bytes = ram_bytes;
4460 12365242 : ffe_ctl.num_bytes = num_bytes;
4461 12365242 : ffe_ctl.min_alloc_size = min_alloc_size;
4462 12365242 : ffe_ctl.empty_size = empty_size;
4463 12365242 : ffe_ctl.flags = flags;
4464 12365242 : ffe_ctl.delalloc = delalloc;
4465 12365242 : ffe_ctl.hint_byte = hint_byte;
4466 12365242 : ffe_ctl.for_treelog = for_treelog;
4467 12365242 : ffe_ctl.for_data_reloc = for_data_reloc;
4468 :
4469 12365242 : ret = find_free_extent(root, ins, &ffe_ctl);
4470 12365188 : if (!ret && !is_data) {
4471 8795280 : btrfs_dec_block_group_reservations(fs_info, ins->objectid);
4472 3569908 : } else if (ret == -ENOSPC) {
4473 9928 : if (!final_tried && ins->offset) {
4474 9928 : num_bytes = min(num_bytes >> 1, ins->offset);
4475 9928 : num_bytes = round_down(num_bytes,
4476 : fs_info->sectorsize);
4477 9928 : num_bytes = max(num_bytes, min_alloc_size);
4478 9928 : ram_bytes = num_bytes;
4479 9928 : if (num_bytes == min_alloc_size)
4480 1338 : final_tried = true;
4481 9928 : goto again;
4482 0 : } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4483 0 : struct btrfs_space_info *sinfo;
4484 :
4485 0 : sinfo = btrfs_find_space_info(fs_info, flags);
4486 0 : btrfs_err(fs_info,
4487 : "allocation failed flags %llu, wanted %llu tree-log %d, relocation: %d",
4488 : flags, num_bytes, for_treelog, for_data_reloc);
4489 0 : if (sinfo)
4490 0 : btrfs_dump_space_info(fs_info, sinfo,
4491 : num_bytes, 1);
4492 : }
4493 : }
4494 :
4495 12355330 : return ret;
4496 : }
4497 :
4498 174 : int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
4499 : u64 start, u64 len, int delalloc)
4500 : {
4501 174 : struct btrfs_block_group *cache;
4502 :
4503 174 : cache = btrfs_lookup_block_group(fs_info, start);
4504 174 : if (!cache) {
4505 0 : btrfs_err(fs_info, "Unable to find block group for %llu",
4506 : start);
4507 0 : return -ENOSPC;
4508 : }
4509 :
4510 174 : btrfs_add_free_space(cache, start, len);
4511 174 : btrfs_free_reserved_bytes(cache, len, delalloc);
4512 174 : trace_btrfs_reserved_extent_free(fs_info, start, len);
4513 :
4514 174 : btrfs_put_block_group(cache);
4515 174 : return 0;
4516 : }
4517 :
4518 23052 : int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start,
4519 : u64 len)
4520 : {
4521 23052 : struct btrfs_block_group *cache;
4522 23052 : int ret = 0;
4523 :
4524 23052 : cache = btrfs_lookup_block_group(trans->fs_info, start);
4525 23052 : if (!cache) {
4526 0 : btrfs_err(trans->fs_info, "unable to find block group for %llu",
4527 : start);
4528 0 : return -ENOSPC;
4529 : }
4530 :
4531 23052 : ret = pin_down_extent(trans, cache, start, len, 1);
4532 23052 : btrfs_put_block_group(cache);
4533 23052 : return ret;
4534 : }
4535 :
4536 9327087 : static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
4537 : u64 num_bytes)
4538 : {
4539 9327087 : struct btrfs_fs_info *fs_info = trans->fs_info;
4540 9327087 : int ret;
4541 :
4542 9327087 : ret = remove_from_free_space_tree(trans, bytenr, num_bytes);
4543 9327087 : if (ret)
4544 : return ret;
4545 :
4546 9327087 : ret = btrfs_update_block_group(trans, bytenr, num_bytes, true);
4547 9327084 : if (ret) {
4548 0 : ASSERT(!ret);
4549 0 : btrfs_err(fs_info, "update block group failed for %llu %llu",
4550 : bytenr, num_bytes);
4551 0 : return ret;
4552 : }
4553 :
4554 9327084 : trace_btrfs_reserved_extent_alloc(fs_info, bytenr, num_bytes);
4555 9327084 : return 0;
4556 : }
4557 :
4558 1995870 : static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4559 : u64 parent, u64 root_objectid,
4560 : u64 flags, u64 owner, u64 offset,
4561 : struct btrfs_key *ins, int ref_mod)
4562 : {
4563 1995870 : struct btrfs_fs_info *fs_info = trans->fs_info;
4564 1995870 : struct btrfs_root *extent_root;
4565 1995870 : int ret;
4566 1995870 : struct btrfs_extent_item *extent_item;
4567 1995870 : struct btrfs_extent_inline_ref *iref;
4568 1995870 : struct btrfs_path *path;
4569 1995870 : struct extent_buffer *leaf;
4570 1995870 : int type;
4571 1995870 : u32 size;
4572 :
4573 1995870 : if (parent > 0)
4574 : type = BTRFS_SHARED_DATA_REF_KEY;
4575 : else
4576 1995870 : type = BTRFS_EXTENT_DATA_REF_KEY;
4577 :
4578 1995870 : size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4579 :
4580 1995870 : path = btrfs_alloc_path();
4581 1995866 : if (!path)
4582 : return -ENOMEM;
4583 :
4584 1995866 : extent_root = btrfs_extent_root(fs_info, ins->objectid);
4585 1995868 : ret = btrfs_insert_empty_item(trans, extent_root, path, ins, size);
4586 1995872 : if (ret) {
4587 1 : btrfs_free_path(path);
4588 1 : return ret;
4589 : }
4590 :
4591 1995871 : leaf = path->nodes[0];
4592 1995871 : extent_item = btrfs_item_ptr(leaf, path->slots[0],
4593 : struct btrfs_extent_item);
4594 1995871 : btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4595 1995870 : btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4596 1995871 : btrfs_set_extent_flags(leaf, extent_item,
4597 : flags | BTRFS_EXTENT_FLAG_DATA);
4598 :
4599 1995871 : iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4600 1995871 : btrfs_set_extent_inline_ref_type(leaf, iref, type);
4601 1995871 : if (parent > 0) {
4602 0 : struct btrfs_shared_data_ref *ref;
4603 0 : ref = (struct btrfs_shared_data_ref *)(iref + 1);
4604 0 : btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4605 0 : btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4606 : } else {
4607 1995871 : struct btrfs_extent_data_ref *ref;
4608 1995871 : ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4609 1995871 : btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4610 1995870 : btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4611 1995871 : btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4612 1995870 : btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4613 : }
4614 :
4615 1995871 : btrfs_mark_buffer_dirty(path->nodes[0]);
4616 1995871 : btrfs_free_path(path);
4617 :
4618 1995871 : return alloc_reserved_extent(trans, ins->objectid, ins->offset);
4619 : }
4620 :
4621 7331147 : static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4622 : struct btrfs_delayed_ref_node *node,
4623 : struct btrfs_delayed_extent_op *extent_op)
4624 : {
4625 7331147 : struct btrfs_fs_info *fs_info = trans->fs_info;
4626 7331147 : struct btrfs_root *extent_root;
4627 7331147 : int ret;
4628 7331147 : struct btrfs_extent_item *extent_item;
4629 7331147 : struct btrfs_key extent_key;
4630 7331147 : struct btrfs_tree_block_info *block_info;
4631 7331147 : struct btrfs_extent_inline_ref *iref;
4632 7331147 : struct btrfs_path *path;
4633 7331147 : struct extent_buffer *leaf;
4634 7331147 : struct btrfs_delayed_tree_ref *ref;
4635 7331147 : u32 size = sizeof(*extent_item) + sizeof(*iref);
4636 7331147 : u64 flags = extent_op->flags_to_set;
4637 7331147 : bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4638 :
4639 7331147 : ref = btrfs_delayed_node_to_tree_ref(node);
4640 :
4641 7331147 : extent_key.objectid = node->bytenr;
4642 7331147 : if (skinny_metadata) {
4643 7331147 : extent_key.offset = ref->level;
4644 7331147 : extent_key.type = BTRFS_METADATA_ITEM_KEY;
4645 : } else {
4646 0 : extent_key.offset = node->num_bytes;
4647 0 : extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4648 0 : size += sizeof(*block_info);
4649 : }
4650 :
4651 7331147 : path = btrfs_alloc_path();
4652 7331171 : if (!path)
4653 : return -ENOMEM;
4654 :
4655 7331171 : extent_root = btrfs_extent_root(fs_info, extent_key.objectid);
4656 7331121 : ret = btrfs_insert_empty_item(trans, extent_root, path, &extent_key,
4657 : size);
4658 7331216 : if (ret) {
4659 0 : btrfs_free_path(path);
4660 0 : return ret;
4661 : }
4662 :
4663 7331216 : leaf = path->nodes[0];
4664 7331216 : extent_item = btrfs_item_ptr(leaf, path->slots[0],
4665 : struct btrfs_extent_item);
4666 7331216 : btrfs_set_extent_refs(leaf, extent_item, 1);
4667 7331216 : btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4668 7331216 : btrfs_set_extent_flags(leaf, extent_item,
4669 : flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4670 :
4671 7331216 : if (skinny_metadata) {
4672 7331216 : iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4673 : } else {
4674 0 : block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4675 0 : btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
4676 0 : btrfs_set_tree_block_level(leaf, block_info, ref->level);
4677 0 : iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4678 : }
4679 :
4680 7331216 : if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
4681 1854913 : btrfs_set_extent_inline_ref_type(leaf, iref,
4682 : BTRFS_SHARED_BLOCK_REF_KEY);
4683 1854913 : btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
4684 : } else {
4685 5476303 : btrfs_set_extent_inline_ref_type(leaf, iref,
4686 : BTRFS_TREE_BLOCK_REF_KEY);
4687 5476303 : btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
4688 : }
4689 :
4690 7331216 : btrfs_mark_buffer_dirty(leaf);
4691 7331216 : btrfs_free_path(path);
4692 :
4693 7331216 : return alloc_reserved_extent(trans, node->bytenr, fs_info->nodesize);
4694 : }
4695 :
4696 3560088 : int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4697 : struct btrfs_root *root, u64 owner,
4698 : u64 offset, u64 ram_bytes,
4699 : struct btrfs_key *ins)
4700 : {
4701 3560088 : struct btrfs_ref generic_ref = { 0 };
4702 :
4703 3560088 : BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
4704 :
4705 3560088 : btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4706 : ins->objectid, ins->offset, 0);
4707 3560088 : btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner,
4708 : offset, 0, false);
4709 3560088 : btrfs_ref_tree_mod(root->fs_info, &generic_ref);
4710 :
4711 3560088 : return btrfs_add_delayed_data_ref(trans, &generic_ref, ram_bytes);
4712 : }
4713 :
4714 : /*
4715 : * this is used by the tree logging recovery code. It records that
4716 : * an extent has been allocated and makes sure to clear the free
4717 : * space cache bits as well
4718 : */
4719 76673 : int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4720 : u64 root_objectid, u64 owner, u64 offset,
4721 : struct btrfs_key *ins)
4722 : {
4723 76673 : struct btrfs_fs_info *fs_info = trans->fs_info;
4724 76673 : int ret;
4725 76673 : struct btrfs_block_group *block_group;
4726 76673 : struct btrfs_space_info *space_info;
4727 :
4728 : /*
4729 : * Mixed block groups will exclude before processing the log so we only
4730 : * need to do the exclude dance if this fs isn't mixed.
4731 : */
4732 76673 : if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
4733 76673 : ret = __exclude_logged_extent(fs_info, ins->objectid,
4734 : ins->offset);
4735 76673 : if (ret)
4736 : return ret;
4737 : }
4738 :
4739 76673 : block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
4740 76673 : if (!block_group)
4741 : return -EINVAL;
4742 :
4743 76673 : space_info = block_group->space_info;
4744 76673 : spin_lock(&space_info->lock);
4745 76673 : spin_lock(&block_group->lock);
4746 76673 : space_info->bytes_reserved += ins->offset;
4747 76673 : block_group->reserved += ins->offset;
4748 76673 : spin_unlock(&block_group->lock);
4749 76673 : spin_unlock(&space_info->lock);
4750 :
4751 76673 : ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4752 : offset, ins, 1);
4753 76673 : if (ret)
4754 0 : btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
4755 76673 : btrfs_put_block_group(block_group);
4756 76673 : return ret;
4757 : }
4758 :
4759 : static struct extent_buffer *
4760 8795353 : btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4761 : u64 bytenr, int level, u64 owner,
4762 : enum btrfs_lock_nesting nest)
4763 : {
4764 8795353 : struct btrfs_fs_info *fs_info = root->fs_info;
4765 8795353 : struct extent_buffer *buf;
4766 8795353 : u64 lockdep_owner = owner;
4767 :
4768 8795353 : buf = btrfs_find_create_tree_block(fs_info, bytenr, owner, level);
4769 8795287 : if (IS_ERR(buf))
4770 : return buf;
4771 :
4772 : /*
4773 : * Extra safety check in case the extent tree is corrupted and extent
4774 : * allocator chooses to use a tree block which is already used and
4775 : * locked.
4776 : */
4777 8795287 : if (buf->lock_owner == current->pid) {
4778 0 : btrfs_err_rl(fs_info,
4779 : "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4780 : buf->start, btrfs_header_owner(buf), current->pid);
4781 0 : free_extent_buffer(buf);
4782 0 : return ERR_PTR(-EUCLEAN);
4783 : }
4784 :
4785 : /*
4786 : * The reloc trees are just snapshots, so we need them to appear to be
4787 : * just like any other fs tree WRT lockdep.
4788 : *
4789 : * The exception however is in replace_path() in relocation, where we
4790 : * hold the lock on the original fs root and then search for the reloc
4791 : * root. At that point we need to make sure any reloc root buffers are
4792 : * set to the BTRFS_TREE_RELOC_OBJECTID lockdep class in order to make
4793 : * lockdep happy.
4794 : */
4795 8795287 : if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID &&
4796 1854913 : !test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
4797 : lockdep_owner = BTRFS_FS_TREE_OBJECTID;
4798 :
4799 : /* btrfs_clear_buffer_dirty() accesses generation field. */
4800 8795287 : btrfs_set_header_generation(buf, trans->transid);
4801 :
4802 : /*
4803 : * This needs to stay, because we could allocate a freed block from an
4804 : * old tree into a new tree, so we need to make sure this new block is
4805 : * set to the appropriate level and owner.
4806 : */
4807 8795287 : btrfs_set_buffer_lockdep_class(lockdep_owner, buf, level);
4808 :
4809 8795287 : __btrfs_tree_lock(buf, nest);
4810 8794619 : btrfs_clear_buffer_dirty(trans, buf);
4811 8795094 : clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
4812 8795266 : clear_bit(EXTENT_BUFFER_NO_CHECK, &buf->bflags);
4813 :
4814 8795160 : set_extent_buffer_uptodate(buf);
4815 :
4816 8795086 : memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4817 8793684 : btrfs_set_header_level(buf, level);
4818 8793684 : btrfs_set_header_bytenr(buf, buf->start);
4819 8793684 : btrfs_set_header_generation(buf, trans->transid);
4820 8793684 : btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4821 8793930 : btrfs_set_header_owner(buf, owner);
4822 8793930 : write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
4823 8794508 : write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
4824 8794418 : if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4825 1093045 : buf->log_index = root->log_transid % 2;
4826 : /*
4827 : * we allow two log transactions at a time, use different
4828 : * EXTENT bit to differentiate dirty pages.
4829 : */
4830 1093045 : if (buf->log_index == 0)
4831 555296 : set_extent_bit(&root->dirty_log_pages, buf->start,
4832 555296 : buf->start + buf->len - 1,
4833 : EXTENT_DIRTY, NULL);
4834 : else
4835 537749 : set_extent_bit(&root->dirty_log_pages, buf->start,
4836 537749 : buf->start + buf->len - 1,
4837 : EXTENT_NEW, NULL);
4838 : } else {
4839 7701373 : buf->log_index = -1;
4840 7701373 : set_extent_bit(&trans->transaction->dirty_pages, buf->start,
4841 7701373 : buf->start + buf->len - 1, EXTENT_DIRTY, NULL);
4842 : }
4843 : /* this returns a buffer locked for blocking */
4844 : return buf;
4845 : }
4846 :
4847 : /*
4848 : * finds a free extent and does all the dirty work required for allocation
4849 : * returns the tree buffer or an ERR_PTR on error.
4850 : */
4851 8797528 : struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
4852 : struct btrfs_root *root,
4853 : u64 parent, u64 root_objectid,
4854 : const struct btrfs_disk_key *key,
4855 : int level, u64 hint,
4856 : u64 empty_size,
4857 : enum btrfs_lock_nesting nest)
4858 : {
4859 8797528 : struct btrfs_fs_info *fs_info = root->fs_info;
4860 8797528 : struct btrfs_key ins;
4861 8797528 : struct btrfs_block_rsv *block_rsv;
4862 8797528 : struct extent_buffer *buf;
4863 8797528 : struct btrfs_delayed_extent_op *extent_op;
4864 8797528 : struct btrfs_ref generic_ref = { 0 };
4865 8797528 : u64 flags = 0;
4866 8797528 : int ret;
4867 8797528 : u32 blocksize = fs_info->nodesize;
4868 8797528 : bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
4869 :
4870 : #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4871 : if (btrfs_is_testing(fs_info)) {
4872 : buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
4873 : level, root_objectid, nest);
4874 : if (!IS_ERR(buf))
4875 : root->alloc_bytenr += blocksize;
4876 : return buf;
4877 : }
4878 : #endif
4879 :
4880 8797528 : block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
4881 8797496 : if (IS_ERR(block_rsv))
4882 : return ERR_CAST(block_rsv);
4883 :
4884 8795239 : ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
4885 : empty_size, hint, &ins, 0, 0);
4886 8795356 : if (ret)
4887 0 : goto out_unuse;
4888 :
4889 8795356 : buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4890 : root_objectid, nest);
4891 8795387 : if (IS_ERR(buf)) {
4892 0 : ret = PTR_ERR(buf);
4893 0 : goto out_free_reserved;
4894 : }
4895 :
4896 8795387 : if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4897 1854913 : if (parent == 0)
4898 2165 : parent = ins.objectid;
4899 : flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4900 : } else
4901 6940474 : BUG_ON(parent > 0);
4902 :
4903 8795387 : if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4904 7702373 : extent_op = btrfs_alloc_delayed_extent_op();
4905 7702364 : if (!extent_op) {
4906 0 : ret = -ENOMEM;
4907 0 : goto out_free_buf;
4908 : }
4909 7702364 : if (key)
4910 15403898 : memcpy(&extent_op->key, key, sizeof(extent_op->key));
4911 : else
4912 830 : memset(&extent_op->key, 0, sizeof(extent_op->key));
4913 7702364 : extent_op->flags_to_set = flags;
4914 7702364 : extent_op->update_key = skinny_metadata ? false : true;
4915 7702364 : extent_op->update_flags = true;
4916 7702364 : extent_op->level = level;
4917 :
4918 7702364 : btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4919 : ins.objectid, ins.offset, parent);
4920 7702364 : btrfs_init_tree_ref(&generic_ref, level, root_objectid,
4921 : root->root_key.objectid, false);
4922 7702354 : btrfs_ref_tree_mod(fs_info, &generic_ref);
4923 7702354 : ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, extent_op);
4924 7702367 : if (ret)
4925 0 : goto out_free_delayed;
4926 : }
4927 : return buf;
4928 :
4929 : out_free_delayed:
4930 0 : btrfs_free_delayed_extent_op(extent_op);
4931 0 : out_free_buf:
4932 0 : btrfs_tree_unlock(buf);
4933 0 : free_extent_buffer(buf);
4934 0 : out_free_reserved:
4935 0 : btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
4936 0 : out_unuse:
4937 0 : btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
4938 0 : return ERR_PTR(ret);
4939 : }
4940 :
4941 : struct walk_control {
4942 : u64 refs[BTRFS_MAX_LEVEL];
4943 : u64 flags[BTRFS_MAX_LEVEL];
4944 : struct btrfs_key update_progress;
4945 : struct btrfs_key drop_progress;
4946 : int drop_level;
4947 : int stage;
4948 : int level;
4949 : int shared_level;
4950 : int update_ref;
4951 : int keep_locks;
4952 : int reada_slot;
4953 : int reada_count;
4954 : int restarted;
4955 : };
4956 :
4957 : #define DROP_REFERENCE 1
4958 : #define UPDATE_BACKREF 2
4959 :
4960 0 : static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4961 : struct btrfs_root *root,
4962 : struct walk_control *wc,
4963 : struct btrfs_path *path)
4964 : {
4965 0 : struct btrfs_fs_info *fs_info = root->fs_info;
4966 0 : u64 bytenr;
4967 0 : u64 generation;
4968 0 : u64 refs;
4969 0 : u64 flags;
4970 0 : u32 nritems;
4971 0 : struct btrfs_key key;
4972 0 : struct extent_buffer *eb;
4973 0 : int ret;
4974 0 : int slot;
4975 0 : int nread = 0;
4976 :
4977 0 : if (path->slots[wc->level] < wc->reada_slot) {
4978 0 : wc->reada_count = wc->reada_count * 2 / 3;
4979 0 : wc->reada_count = max(wc->reada_count, 2);
4980 : } else {
4981 0 : wc->reada_count = wc->reada_count * 3 / 2;
4982 0 : wc->reada_count = min_t(int, wc->reada_count,
4983 : BTRFS_NODEPTRS_PER_BLOCK(fs_info));
4984 : }
4985 :
4986 0 : eb = path->nodes[wc->level];
4987 0 : nritems = btrfs_header_nritems(eb);
4988 :
4989 0 : for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4990 0 : if (nread >= wc->reada_count)
4991 : break;
4992 :
4993 0 : cond_resched();
4994 0 : bytenr = btrfs_node_blockptr(eb, slot);
4995 0 : generation = btrfs_node_ptr_generation(eb, slot);
4996 :
4997 0 : if (slot == path->slots[wc->level])
4998 0 : goto reada;
4999 :
5000 0 : if (wc->stage == UPDATE_BACKREF &&
5001 0 : generation <= root->root_key.offset)
5002 0 : continue;
5003 :
5004 : /* We don't lock the tree block, it's OK to be racy here */
5005 0 : ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
5006 0 : wc->level - 1, 1, &refs,
5007 : &flags);
5008 : /* We don't care about errors in readahead. */
5009 0 : if (ret < 0)
5010 0 : continue;
5011 0 : BUG_ON(refs == 0);
5012 :
5013 0 : if (wc->stage == DROP_REFERENCE) {
5014 0 : if (refs == 1)
5015 0 : goto reada;
5016 :
5017 0 : if (wc->level == 1 &&
5018 0 : (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5019 0 : continue;
5020 0 : if (!wc->update_ref ||
5021 0 : generation <= root->root_key.offset)
5022 0 : continue;
5023 0 : btrfs_node_key_to_cpu(eb, &key, slot);
5024 0 : ret = btrfs_comp_cpu_keys(&key,
5025 0 : &wc->update_progress);
5026 0 : if (ret < 0)
5027 0 : continue;
5028 : } else {
5029 0 : if (wc->level == 1 &&
5030 0 : (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5031 0 : continue;
5032 : }
5033 0 : reada:
5034 0 : btrfs_readahead_node_child(eb, slot);
5035 0 : nread++;
5036 : }
5037 0 : wc->reada_slot = slot;
5038 0 : }
5039 :
5040 : /*
5041 : * helper to process tree block while walking down the tree.
5042 : *
5043 : * when wc->stage == UPDATE_BACKREF, this function updates
5044 : * back refs for pointers in the block.
5045 : *
5046 : * NOTE: return value 1 means we should stop walking down.
5047 : */
5048 8674504 : static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5049 : struct btrfs_root *root,
5050 : struct btrfs_path *path,
5051 : struct walk_control *wc, int lookup_info)
5052 : {
5053 8674504 : struct btrfs_fs_info *fs_info = root->fs_info;
5054 8674504 : int level = wc->level;
5055 8674504 : struct extent_buffer *eb = path->nodes[level];
5056 8674504 : u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5057 8674504 : int ret;
5058 :
5059 8674504 : if (wc->stage == UPDATE_BACKREF &&
5060 0 : btrfs_header_owner(eb) != root->root_key.objectid)
5061 : return 1;
5062 :
5063 : /*
5064 : * when reference count of tree block is 1, it won't increase
5065 : * again. once full backref flag is set, we never clear it.
5066 : */
5067 8674504 : if (lookup_info &&
5068 6817545 : ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5069 0 : (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5070 506527 : BUG_ON(!path->locks[level]);
5071 506527 : ret = btrfs_lookup_extent_info(trans, fs_info,
5072 : eb->start, level, 1,
5073 : &wc->refs[level],
5074 : &wc->flags[level]);
5075 506527 : BUG_ON(ret == -ENOMEM);
5076 506527 : if (ret)
5077 : return ret;
5078 506527 : BUG_ON(wc->refs[level] == 0);
5079 : }
5080 :
5081 8674504 : if (wc->stage == DROP_REFERENCE) {
5082 8674504 : if (wc->refs[level] > 1)
5083 : return 1;
5084 :
5085 8169650 : if (path->locks[level] && !wc->keep_locks) {
5086 1858577 : btrfs_tree_unlock_rw(eb, path->locks[level]);
5087 1858577 : path->locks[level] = 0;
5088 : }
5089 8169650 : return 0;
5090 : }
5091 :
5092 : /* wc->stage == UPDATE_BACKREF */
5093 0 : if (!(wc->flags[level] & flag)) {
5094 0 : BUG_ON(!path->locks[level]);
5095 0 : ret = btrfs_inc_ref(trans, root, eb, 1);
5096 0 : BUG_ON(ret); /* -ENOMEM */
5097 0 : ret = btrfs_dec_ref(trans, root, eb, 0);
5098 0 : BUG_ON(ret); /* -ENOMEM */
5099 0 : ret = btrfs_set_disk_extent_flags(trans, eb, flag);
5100 0 : BUG_ON(ret); /* -ENOMEM */
5101 0 : wc->flags[level] |= flag;
5102 : }
5103 :
5104 : /*
5105 : * the block is shared by multiple trees, so it's not good to
5106 : * keep the tree lock
5107 : */
5108 0 : if (path->locks[level] && level > 0) {
5109 0 : btrfs_tree_unlock_rw(eb, path->locks[level]);
5110 0 : path->locks[level] = 0;
5111 : }
5112 : return 0;
5113 : }
5114 :
5115 : /*
5116 : * This is used to verify a ref exists for this root to deal with a bug where we
5117 : * would have a drop_progress key that hadn't been updated properly.
5118 : */
5119 0 : static int check_ref_exists(struct btrfs_trans_handle *trans,
5120 : struct btrfs_root *root, u64 bytenr, u64 parent,
5121 : int level)
5122 : {
5123 0 : struct btrfs_path *path;
5124 0 : struct btrfs_extent_inline_ref *iref;
5125 0 : int ret;
5126 :
5127 0 : path = btrfs_alloc_path();
5128 0 : if (!path)
5129 : return -ENOMEM;
5130 :
5131 0 : ret = lookup_extent_backref(trans, path, &iref, bytenr,
5132 0 : root->fs_info->nodesize, parent,
5133 : root->root_key.objectid, level, 0);
5134 0 : btrfs_free_path(path);
5135 0 : if (ret == -ENOENT)
5136 : return 0;
5137 0 : if (ret < 0)
5138 0 : return ret;
5139 : return 1;
5140 : }
5141 :
5142 : /*
5143 : * helper to process tree block pointer.
5144 : *
5145 : * when wc->stage == DROP_REFERENCE, this function checks
5146 : * reference count of the block pointed to. if the block
5147 : * is shared and we need update back refs for the subtree
5148 : * rooted at the block, this function changes wc->stage to
5149 : * UPDATE_BACKREF. if the block is shared and there is no
5150 : * need to update back, this function drops the reference
5151 : * to the block.
5152 : *
5153 : * NOTE: return value 1 means we should stop walking down.
5154 : */
5155 6317285 : static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5156 : struct btrfs_root *root,
5157 : struct btrfs_path *path,
5158 : struct walk_control *wc, int *lookup_info)
5159 : {
5160 6317285 : struct btrfs_fs_info *fs_info = root->fs_info;
5161 6317285 : u64 bytenr;
5162 6317285 : u64 generation;
5163 6317285 : u64 parent;
5164 6317285 : struct btrfs_tree_parent_check check = { 0 };
5165 6317285 : struct btrfs_key key;
5166 6317285 : struct btrfs_ref ref = { 0 };
5167 6317285 : struct extent_buffer *next;
5168 6317285 : int level = wc->level;
5169 6317285 : int reada = 0;
5170 6317285 : int ret = 0;
5171 6317285 : bool need_account = false;
5172 :
5173 6317285 : generation = btrfs_node_ptr_generation(path->nodes[level],
5174 : path->slots[level]);
5175 : /*
5176 : * if the lower level block was created before the snapshot
5177 : * was created, we know there is no need to update back refs
5178 : * for the subtree
5179 : */
5180 6317285 : if (wc->stage == UPDATE_BACKREF &&
5181 0 : generation <= root->root_key.offset) {
5182 0 : *lookup_info = 1;
5183 0 : return 1;
5184 : }
5185 :
5186 6317285 : bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5187 :
5188 6317285 : check.level = level - 1;
5189 6317285 : check.transid = generation;
5190 6317285 : check.owner_root = root->root_key.objectid;
5191 6317285 : check.has_first_key = true;
5192 6317285 : btrfs_node_key_to_cpu(path->nodes[level], &check.first_key,
5193 : path->slots[level]);
5194 :
5195 6317285 : next = find_extent_buffer(fs_info, bytenr);
5196 6317285 : if (!next) {
5197 1244265 : next = btrfs_find_create_tree_block(fs_info, bytenr,
5198 : root->root_key.objectid, level - 1);
5199 1244265 : if (IS_ERR(next))
5200 0 : return PTR_ERR(next);
5201 : reada = 1;
5202 : }
5203 6317285 : btrfs_tree_lock(next);
5204 :
5205 6317285 : ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
5206 : &wc->refs[level - 1],
5207 6317285 : &wc->flags[level - 1]);
5208 6317285 : if (ret < 0)
5209 0 : goto out_unlock;
5210 :
5211 6317285 : if (unlikely(wc->refs[level - 1] == 0)) {
5212 0 : btrfs_err(fs_info, "Missing references.");
5213 0 : ret = -EIO;
5214 0 : goto out_unlock;
5215 : }
5216 6317285 : *lookup_info = 0;
5217 :
5218 6317285 : if (wc->stage == DROP_REFERENCE) {
5219 6317285 : if (wc->refs[level - 1] > 1) {
5220 4460326 : need_account = true;
5221 4460326 : if (level == 1 &&
5222 4448408 : (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5223 4416352 : goto skip;
5224 :
5225 43974 : if (!wc->update_ref ||
5226 21681 : generation <= root->root_key.offset)
5227 43974 : goto skip;
5228 :
5229 0 : btrfs_node_key_to_cpu(path->nodes[level], &key,
5230 : path->slots[level]);
5231 0 : ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5232 0 : if (ret < 0)
5233 0 : goto skip;
5234 :
5235 0 : wc->stage = UPDATE_BACKREF;
5236 0 : wc->shared_level = level - 1;
5237 : }
5238 : } else {
5239 0 : if (level == 1 &&
5240 0 : (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5241 0 : goto skip;
5242 : }
5243 :
5244 1856959 : if (!btrfs_buffer_uptodate(next, generation, 0)) {
5245 0 : btrfs_tree_unlock(next);
5246 0 : free_extent_buffer(next);
5247 0 : next = NULL;
5248 0 : *lookup_info = 1;
5249 : }
5250 :
5251 1856959 : if (!next) {
5252 0 : if (reada && level == 1)
5253 0 : reada_walk_down(trans, root, wc, path);
5254 0 : next = read_tree_block(fs_info, bytenr, &check);
5255 0 : if (IS_ERR(next)) {
5256 0 : return PTR_ERR(next);
5257 0 : } else if (!extent_buffer_uptodate(next)) {
5258 0 : free_extent_buffer(next);
5259 0 : return -EIO;
5260 : }
5261 0 : btrfs_tree_lock(next);
5262 : }
5263 :
5264 1856959 : level--;
5265 1856959 : ASSERT(level == btrfs_header_level(next));
5266 1856959 : if (level != btrfs_header_level(next)) {
5267 0 : btrfs_err(root->fs_info, "mismatched level");
5268 0 : ret = -EIO;
5269 0 : goto out_unlock;
5270 : }
5271 1856959 : path->nodes[level] = next;
5272 1856959 : path->slots[level] = 0;
5273 1856959 : path->locks[level] = BTRFS_WRITE_LOCK;
5274 1856959 : wc->level = level;
5275 1856959 : if (wc->level == 1)
5276 18673 : wc->reada_slot = 0;
5277 : return 0;
5278 4460326 : skip:
5279 4460326 : wc->refs[level - 1] = 0;
5280 4460326 : wc->flags[level - 1] = 0;
5281 4460326 : if (wc->stage == DROP_REFERENCE) {
5282 4460326 : if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5283 4454128 : parent = path->nodes[level]->start;
5284 : } else {
5285 6198 : ASSERT(root->root_key.objectid ==
5286 : btrfs_header_owner(path->nodes[level]));
5287 6198 : if (root->root_key.objectid !=
5288 : btrfs_header_owner(path->nodes[level])) {
5289 0 : btrfs_err(root->fs_info,
5290 : "mismatched block owner");
5291 0 : ret = -EIO;
5292 0 : goto out_unlock;
5293 : }
5294 : parent = 0;
5295 : }
5296 :
5297 : /*
5298 : * If we had a drop_progress we need to verify the refs are set
5299 : * as expected. If we find our ref then we know that from here
5300 : * on out everything should be correct, and we can clear the
5301 : * ->restarted flag.
5302 : */
5303 4460326 : if (wc->restarted) {
5304 0 : ret = check_ref_exists(trans, root, bytenr, parent,
5305 : level - 1);
5306 0 : if (ret < 0)
5307 0 : goto out_unlock;
5308 0 : if (ret == 0)
5309 0 : goto no_delete;
5310 0 : ret = 0;
5311 0 : wc->restarted = 0;
5312 : }
5313 :
5314 : /*
5315 : * Reloc tree doesn't contribute to qgroup numbers, and we have
5316 : * already accounted them at merge time (replace_path),
5317 : * thus we could skip expensive subtree trace here.
5318 : */
5319 4460326 : if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5320 : need_account) {
5321 33729 : ret = btrfs_qgroup_trace_subtree(trans, next,
5322 : generation, level - 1);
5323 33729 : if (ret) {
5324 0 : btrfs_err_rl(fs_info,
5325 : "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5326 : ret);
5327 : }
5328 : }
5329 :
5330 : /*
5331 : * We need to update the next key in our walk control so we can
5332 : * update the drop_progress key accordingly. We don't care if
5333 : * find_next_key doesn't find a key because that means we're at
5334 : * the end and are going to clean up now.
5335 : */
5336 4460326 : wc->drop_level = level;
5337 4460326 : find_next_key(path, level, &wc->drop_progress);
5338 :
5339 4460326 : btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5340 4460326 : fs_info->nodesize, parent);
5341 4460326 : btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid,
5342 : 0, false);
5343 4460326 : ret = btrfs_free_extent(trans, &ref);
5344 4460326 : if (ret)
5345 0 : goto out_unlock;
5346 : }
5347 4460326 : no_delete:
5348 4460326 : *lookup_info = 1;
5349 4460326 : ret = 1;
5350 :
5351 4460326 : out_unlock:
5352 4460326 : btrfs_tree_unlock(next);
5353 4460326 : free_extent_buffer(next);
5354 :
5355 4460326 : return ret;
5356 : }
5357 :
5358 : /*
5359 : * helper to process tree block while walking up the tree.
5360 : *
5361 : * when wc->stage == DROP_REFERENCE, this function drops
5362 : * reference count on the block.
5363 : *
5364 : * when wc->stage == UPDATE_BACKREF, this function changes
5365 : * wc->stage back to DROP_REFERENCE if we changed wc->stage
5366 : * to UPDATE_BACKREF previously while processing the block.
5367 : *
5368 : * NOTE: return value 1 means we should stop walking up.
5369 : */
5370 2363486 : static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5371 : struct btrfs_root *root,
5372 : struct btrfs_path *path,
5373 : struct walk_control *wc)
5374 : {
5375 2363486 : struct btrfs_fs_info *fs_info = root->fs_info;
5376 2363486 : int ret;
5377 2363486 : int level = wc->level;
5378 2363486 : struct extent_buffer *eb = path->nodes[level];
5379 2363486 : u64 parent = 0;
5380 :
5381 2363486 : if (wc->stage == UPDATE_BACKREF) {
5382 0 : BUG_ON(wc->shared_level < level);
5383 0 : if (level < wc->shared_level)
5384 0 : goto out;
5385 :
5386 0 : ret = find_next_key(path, level + 1, &wc->update_progress);
5387 0 : if (ret > 0)
5388 0 : wc->update_ref = 0;
5389 :
5390 0 : wc->stage = DROP_REFERENCE;
5391 0 : wc->shared_level = -1;
5392 0 : path->slots[level] = 0;
5393 :
5394 : /*
5395 : * check reference count again if the block isn't locked.
5396 : * we should start walking down the tree again if reference
5397 : * count is one.
5398 : */
5399 0 : if (!path->locks[level]) {
5400 0 : BUG_ON(level == 0);
5401 0 : btrfs_tree_lock(eb);
5402 0 : path->locks[level] = BTRFS_WRITE_LOCK;
5403 :
5404 0 : ret = btrfs_lookup_extent_info(trans, fs_info,
5405 : eb->start, level, 1,
5406 : &wc->refs[level],
5407 : &wc->flags[level]);
5408 0 : if (ret < 0) {
5409 0 : btrfs_tree_unlock_rw(eb, path->locks[level]);
5410 0 : path->locks[level] = 0;
5411 0 : return ret;
5412 : }
5413 0 : BUG_ON(wc->refs[level] == 0);
5414 0 : if (wc->refs[level] == 1) {
5415 0 : btrfs_tree_unlock_rw(eb, path->locks[level]);
5416 0 : path->locks[level] = 0;
5417 0 : return 1;
5418 : }
5419 : }
5420 : }
5421 :
5422 : /* wc->stage == DROP_REFERENCE */
5423 2363486 : BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5424 :
5425 2363486 : if (wc->refs[level] == 1) {
5426 1858632 : if (level == 0) {
5427 1838825 : if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5428 1838776 : ret = btrfs_dec_ref(trans, root, eb, 1);
5429 : else
5430 49 : ret = btrfs_dec_ref(trans, root, eb, 0);
5431 1838825 : BUG_ON(ret); /* -ENOMEM */
5432 1838825 : if (is_fstree(root->root_key.objectid)) {
5433 3313 : ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5434 3313 : if (ret) {
5435 0 : btrfs_err_rl(fs_info,
5436 : "error %d accounting leaf items, quota is out of sync, rescan required",
5437 : ret);
5438 : }
5439 : }
5440 : }
5441 : /* Make block locked assertion in btrfs_clear_buffer_dirty happy. */
5442 1858632 : if (!path->locks[level]) {
5443 1858577 : btrfs_tree_lock(eb);
5444 1858577 : path->locks[level] = BTRFS_WRITE_LOCK;
5445 : }
5446 1858632 : btrfs_clear_buffer_dirty(trans, eb);
5447 : }
5448 :
5449 2363486 : if (eb == root->node) {
5450 1618 : if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5451 1500 : parent = eb->start;
5452 118 : else if (root->root_key.objectid != btrfs_header_owner(eb))
5453 0 : goto owner_mismatch;
5454 : } else {
5455 2361868 : if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5456 2360097 : parent = path->nodes[level + 1]->start;
5457 1771 : else if (root->root_key.objectid !=
5458 1771 : btrfs_header_owner(path->nodes[level + 1]))
5459 0 : goto owner_mismatch;
5460 : }
5461 :
5462 2363486 : btrfs_free_tree_block(trans, btrfs_root_id(root), eb, parent,
5463 2363486 : wc->refs[level] == 1);
5464 2363486 : out:
5465 2363486 : wc->refs[level] = 0;
5466 2363486 : wc->flags[level] = 0;
5467 2363486 : return 0;
5468 :
5469 0 : owner_mismatch:
5470 0 : btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5471 : btrfs_header_owner(eb), root->root_key.objectid);
5472 : return -EUCLEAN;
5473 : }
5474 :
5475 2357219 : static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5476 : struct btrfs_root *root,
5477 : struct btrfs_path *path,
5478 : struct walk_control *wc)
5479 : {
5480 2357219 : int level = wc->level;
5481 2357219 : int lookup_info = 1;
5482 2357219 : int ret = 0;
5483 :
5484 8674504 : while (level >= 0) {
5485 8674504 : ret = walk_down_proc(trans, root, path, wc, lookup_info);
5486 8674504 : if (ret)
5487 : break;
5488 :
5489 8169650 : if (level == 0)
5490 : break;
5491 :
5492 6330825 : if (path->slots[level] >=
5493 6330825 : btrfs_header_nritems(path->nodes[level]))
5494 : break;
5495 :
5496 6317285 : ret = do_walk_down(trans, root, path, wc, &lookup_info);
5497 6317285 : if (ret > 0) {
5498 4460326 : path->slots[level]++;
5499 4460326 : continue;
5500 1856959 : } else if (ret < 0)
5501 : break;
5502 1856959 : level = wc->level;
5503 : }
5504 2357219 : return (ret == 1) ? 0 : ret;
5505 : }
5506 :
5507 2357219 : static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5508 : struct btrfs_root *root,
5509 : struct btrfs_path *path,
5510 : struct walk_control *wc, int max_level)
5511 : {
5512 2357219 : int level = wc->level;
5513 2357219 : int ret;
5514 :
5515 2357219 : path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5516 4720705 : while (level < max_level && path->nodes[level]) {
5517 4214178 : wc->level = level;
5518 4214178 : if (path->slots[level] + 1 <
5519 4214178 : btrfs_header_nritems(path->nodes[level])) {
5520 1850692 : path->slots[level]++;
5521 1850692 : return 0;
5522 : } else {
5523 2363486 : ret = walk_up_proc(trans, root, path, wc);
5524 2363486 : if (ret > 0)
5525 : return 0;
5526 2363486 : if (ret < 0)
5527 0 : return ret;
5528 :
5529 2363486 : if (path->locks[level]) {
5530 2363486 : btrfs_tree_unlock_rw(path->nodes[level],
5531 : path->locks[level]);
5532 2363486 : path->locks[level] = 0;
5533 : }
5534 2363486 : free_extent_buffer(path->nodes[level]);
5535 2363486 : path->nodes[level] = NULL;
5536 2363486 : level++;
5537 : }
5538 : }
5539 : return 1;
5540 : }
5541 :
5542 : /*
5543 : * drop a subvolume tree.
5544 : *
5545 : * this function traverses the tree freeing any blocks that only
5546 : * referenced by the tree.
5547 : *
5548 : * when a shared tree block is found. this function decreases its
5549 : * reference count by one. if update_ref is true, this function
5550 : * also make sure backrefs for the shared block and all lower level
5551 : * blocks are properly updated.
5552 : *
5553 : * If called with for_reloc == 0, may exit early with -EAGAIN
5554 : */
5555 1618 : int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
5556 : {
5557 1618 : const bool is_reloc_root = (root->root_key.objectid ==
5558 : BTRFS_TREE_RELOC_OBJECTID);
5559 1618 : struct btrfs_fs_info *fs_info = root->fs_info;
5560 1618 : struct btrfs_path *path;
5561 1618 : struct btrfs_trans_handle *trans;
5562 1618 : struct btrfs_root *tree_root = fs_info->tree_root;
5563 1618 : struct btrfs_root_item *root_item = &root->root_item;
5564 1618 : struct walk_control *wc;
5565 1618 : struct btrfs_key key;
5566 1618 : int err = 0;
5567 1618 : int ret;
5568 1618 : int level;
5569 1618 : bool root_dropped = false;
5570 1618 : bool unfinished_drop = false;
5571 :
5572 1618 : btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
5573 :
5574 1618 : path = btrfs_alloc_path();
5575 1618 : if (!path) {
5576 0 : err = -ENOMEM;
5577 0 : goto out;
5578 : }
5579 :
5580 1618 : wc = kzalloc(sizeof(*wc), GFP_NOFS);
5581 1618 : if (!wc) {
5582 0 : btrfs_free_path(path);
5583 0 : err = -ENOMEM;
5584 0 : goto out;
5585 : }
5586 :
5587 : /*
5588 : * Use join to avoid potential EINTR from transaction start. See
5589 : * wait_reserve_ticket and the whole reservation callchain.
5590 : */
5591 1618 : if (for_reloc)
5592 1500 : trans = btrfs_join_transaction(tree_root);
5593 : else
5594 118 : trans = btrfs_start_transaction(tree_root, 0);
5595 1618 : if (IS_ERR(trans)) {
5596 0 : err = PTR_ERR(trans);
5597 0 : goto out_free;
5598 : }
5599 :
5600 1618 : err = btrfs_run_delayed_items(trans);
5601 1618 : if (err)
5602 0 : goto out_end_trans;
5603 :
5604 : /*
5605 : * This will help us catch people modifying the fs tree while we're
5606 : * dropping it. It is unsafe to mess with the fs tree while it's being
5607 : * dropped as we unlock the root node and parent nodes as we walk down
5608 : * the tree, assuming nothing will change. If something does change
5609 : * then we'll have stale information and drop references to blocks we've
5610 : * already dropped.
5611 : */
5612 1618 : set_bit(BTRFS_ROOT_DELETING, &root->state);
5613 1618 : unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
5614 :
5615 1618 : if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5616 1618 : level = btrfs_header_level(root->node);
5617 1618 : path->nodes[level] = btrfs_lock_root_node(root);
5618 1618 : path->slots[level] = 0;
5619 1618 : path->locks[level] = BTRFS_WRITE_LOCK;
5620 3236 : memset(&wc->update_progress, 0,
5621 : sizeof(wc->update_progress));
5622 : } else {
5623 0 : btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5624 0 : memcpy(&wc->update_progress, &key,
5625 : sizeof(wc->update_progress));
5626 :
5627 0 : level = btrfs_root_drop_level(root_item);
5628 0 : BUG_ON(level == 0);
5629 0 : path->lowest_level = level;
5630 0 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5631 0 : path->lowest_level = 0;
5632 0 : if (ret < 0) {
5633 0 : err = ret;
5634 0 : goto out_end_trans;
5635 : }
5636 0 : WARN_ON(ret > 0);
5637 :
5638 : /*
5639 : * unlock our path, this is safe because only this
5640 : * function is allowed to delete this snapshot
5641 : */
5642 0 : btrfs_unlock_up_safe(path, 0);
5643 :
5644 0 : level = btrfs_header_level(root->node);
5645 0 : while (1) {
5646 0 : btrfs_tree_lock(path->nodes[level]);
5647 0 : path->locks[level] = BTRFS_WRITE_LOCK;
5648 :
5649 0 : ret = btrfs_lookup_extent_info(trans, fs_info,
5650 0 : path->nodes[level]->start,
5651 : level, 1, &wc->refs[level],
5652 : &wc->flags[level]);
5653 0 : if (ret < 0) {
5654 0 : err = ret;
5655 0 : goto out_end_trans;
5656 : }
5657 0 : BUG_ON(wc->refs[level] == 0);
5658 :
5659 0 : if (level == btrfs_root_drop_level(root_item))
5660 : break;
5661 :
5662 0 : btrfs_tree_unlock(path->nodes[level]);
5663 0 : path->locks[level] = 0;
5664 0 : WARN_ON(wc->refs[level] != 1);
5665 0 : level--;
5666 : }
5667 : }
5668 :
5669 1618 : wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
5670 1618 : wc->level = level;
5671 1618 : wc->shared_level = -1;
5672 1618 : wc->stage = DROP_REFERENCE;
5673 1618 : wc->update_ref = update_ref;
5674 1618 : wc->keep_locks = 0;
5675 1618 : wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5676 :
5677 1852310 : while (1) {
5678 :
5679 1852310 : ret = walk_down_tree(trans, root, path, wc);
5680 1852310 : if (ret < 0) {
5681 0 : btrfs_abort_transaction(trans, ret);
5682 0 : err = ret;
5683 0 : break;
5684 : }
5685 :
5686 1852310 : ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5687 1852310 : if (ret < 0) {
5688 0 : btrfs_abort_transaction(trans, ret);
5689 0 : err = ret;
5690 0 : break;
5691 : }
5692 :
5693 1852310 : if (ret > 0) {
5694 1618 : BUG_ON(wc->stage != DROP_REFERENCE);
5695 : break;
5696 : }
5697 :
5698 1850692 : if (wc->stage == DROP_REFERENCE) {
5699 1850692 : wc->drop_level = wc->level;
5700 1850692 : btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5701 : &wc->drop_progress,
5702 : path->slots[wc->drop_level]);
5703 : }
5704 1850692 : btrfs_cpu_key_to_disk(&root_item->drop_progress,
5705 1850692 : &wc->drop_progress);
5706 1850692 : btrfs_set_root_drop_level(root_item, wc->drop_level);
5707 :
5708 1850692 : BUG_ON(wc->level == 0);
5709 1850692 : if (btrfs_should_end_transaction(trans) ||
5710 0 : (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
5711 1848545 : ret = btrfs_update_root(trans, tree_root,
5712 : &root->root_key,
5713 : root_item);
5714 1848545 : if (ret) {
5715 0 : btrfs_abort_transaction(trans, ret);
5716 0 : err = ret;
5717 0 : goto out_end_trans;
5718 : }
5719 :
5720 1848545 : if (!is_reloc_root)
5721 3242 : btrfs_set_last_root_drop_gen(fs_info, trans->transid);
5722 :
5723 1848545 : btrfs_end_transaction_throttle(trans);
5724 1848545 : if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
5725 0 : btrfs_debug(fs_info,
5726 : "drop snapshot early exit");
5727 0 : err = -EAGAIN;
5728 0 : goto out_free;
5729 : }
5730 :
5731 : /*
5732 : * Use join to avoid potential EINTR from transaction
5733 : * start. See wait_reserve_ticket and the whole
5734 : * reservation callchain.
5735 : */
5736 1848545 : if (for_reloc)
5737 1845303 : trans = btrfs_join_transaction(tree_root);
5738 : else
5739 3242 : trans = btrfs_start_transaction(tree_root, 0);
5740 1848545 : if (IS_ERR(trans)) {
5741 0 : err = PTR_ERR(trans);
5742 0 : goto out_free;
5743 : }
5744 : }
5745 : }
5746 1618 : btrfs_release_path(path);
5747 1618 : if (err)
5748 0 : goto out_end_trans;
5749 :
5750 1618 : ret = btrfs_del_root(trans, &root->root_key);
5751 1618 : if (ret) {
5752 0 : btrfs_abort_transaction(trans, ret);
5753 0 : err = ret;
5754 0 : goto out_end_trans;
5755 : }
5756 :
5757 1618 : if (!is_reloc_root) {
5758 118 : ret = btrfs_find_root(tree_root, &root->root_key, path,
5759 : NULL, NULL);
5760 118 : if (ret < 0) {
5761 0 : btrfs_abort_transaction(trans, ret);
5762 0 : err = ret;
5763 0 : goto out_end_trans;
5764 118 : } else if (ret > 0) {
5765 : /* if we fail to delete the orphan item this time
5766 : * around, it'll get picked up the next time.
5767 : *
5768 : * The most common failure here is just -ENOENT.
5769 : */
5770 118 : btrfs_del_orphan_item(trans, tree_root,
5771 : root->root_key.objectid);
5772 : }
5773 : }
5774 :
5775 : /*
5776 : * This subvolume is going to be completely dropped, and won't be
5777 : * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5778 : * commit transaction time. So free it here manually.
5779 : */
5780 1618 : btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
5781 1618 : btrfs_qgroup_free_meta_all_pertrans(root);
5782 :
5783 3236 : if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
5784 118 : btrfs_add_dropped_root(trans, root);
5785 : else
5786 1500 : btrfs_put_root(root);
5787 : root_dropped = true;
5788 1618 : out_end_trans:
5789 1618 : if (!is_reloc_root)
5790 118 : btrfs_set_last_root_drop_gen(fs_info, trans->transid);
5791 :
5792 1618 : btrfs_end_transaction_throttle(trans);
5793 1618 : out_free:
5794 1618 : kfree(wc);
5795 1618 : btrfs_free_path(path);
5796 1618 : out:
5797 : /*
5798 : * We were an unfinished drop root, check to see if there are any
5799 : * pending, and if not clear and wake up any waiters.
5800 : */
5801 1618 : if (!err && unfinished_drop)
5802 0 : btrfs_maybe_wake_unfinished_drop(fs_info);
5803 :
5804 : /*
5805 : * So if we need to stop dropping the snapshot for whatever reason we
5806 : * need to make sure to add it back to the dead root list so that we
5807 : * keep trying to do the work later. This also cleans up roots if we
5808 : * don't have it in the radix (like when we recover after a power fail
5809 : * or unmount) so we don't leak memory.
5810 : */
5811 1618 : if (!for_reloc && !root_dropped)
5812 0 : btrfs_add_dead_root(root);
5813 1618 : return err;
5814 : }
5815 :
5816 : /*
5817 : * drop subtree rooted at tree block 'node'.
5818 : *
5819 : * NOTE: this function will unlock and release tree block 'node'
5820 : * only used by relocation code
5821 : */
5822 504909 : int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5823 : struct btrfs_root *root,
5824 : struct extent_buffer *node,
5825 : struct extent_buffer *parent)
5826 : {
5827 504909 : struct btrfs_fs_info *fs_info = root->fs_info;
5828 504909 : struct btrfs_path *path;
5829 504909 : struct walk_control *wc;
5830 504909 : int level;
5831 504909 : int parent_level;
5832 504909 : int ret = 0;
5833 504909 : int wret;
5834 :
5835 504909 : BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5836 :
5837 504909 : path = btrfs_alloc_path();
5838 504909 : if (!path)
5839 : return -ENOMEM;
5840 :
5841 504909 : wc = kzalloc(sizeof(*wc), GFP_NOFS);
5842 504909 : if (!wc) {
5843 0 : btrfs_free_path(path);
5844 0 : return -ENOMEM;
5845 : }
5846 :
5847 504909 : btrfs_assert_tree_write_locked(parent);
5848 504909 : parent_level = btrfs_header_level(parent);
5849 504909 : atomic_inc(&parent->refs);
5850 504909 : path->nodes[parent_level] = parent;
5851 504909 : path->slots[parent_level] = btrfs_header_nritems(parent);
5852 :
5853 504909 : btrfs_assert_tree_write_locked(node);
5854 504909 : level = btrfs_header_level(node);
5855 504909 : path->nodes[level] = node;
5856 504909 : path->slots[level] = 0;
5857 504909 : path->locks[level] = BTRFS_WRITE_LOCK;
5858 :
5859 504909 : wc->refs[parent_level] = 1;
5860 504909 : wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5861 504909 : wc->level = level;
5862 504909 : wc->shared_level = -1;
5863 504909 : wc->stage = DROP_REFERENCE;
5864 504909 : wc->update_ref = 0;
5865 504909 : wc->keep_locks = 1;
5866 504909 : wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
5867 :
5868 504909 : while (1) {
5869 504909 : wret = walk_down_tree(trans, root, path, wc);
5870 504909 : if (wret < 0) {
5871 : ret = wret;
5872 : break;
5873 : }
5874 :
5875 504909 : wret = walk_up_tree(trans, root, path, wc, parent_level);
5876 504909 : if (wret < 0)
5877 0 : ret = wret;
5878 504909 : if (wret != 0)
5879 : break;
5880 : }
5881 :
5882 504909 : kfree(wc);
5883 504909 : btrfs_free_path(path);
5884 504909 : return ret;
5885 : }
5886 :
5887 144 : int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5888 : u64 start, u64 end)
5889 : {
5890 144 : return unpin_extent_range(fs_info, start, end, false);
5891 : }
5892 :
5893 : /*
5894 : * It used to be that old block groups would be left around forever.
5895 : * Iterating over them would be enough to trim unused space. Since we
5896 : * now automatically remove them, we also need to iterate over unallocated
5897 : * space.
5898 : *
5899 : * We don't want a transaction for this since the discard may take a
5900 : * substantial amount of time. We don't require that a transaction be
5901 : * running, but we do need to take a running transaction into account
5902 : * to ensure that we're not discarding chunks that were released or
5903 : * allocated in the current transaction.
5904 : *
5905 : * Holding the chunks lock will prevent other threads from allocating
5906 : * or releasing chunks, but it won't prevent a running transaction
5907 : * from committing and releasing the memory that the pending chunks
5908 : * list head uses. For that, we need to take a reference to the
5909 : * transaction and hold the commit root sem. We only need to hold
5910 : * it while performing the free space search since we have already
5911 : * held back allocations.
5912 : */
5913 589 : static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
5914 : {
5915 589 : u64 start = BTRFS_DEVICE_RANGE_RESERVED, len = 0, end = 0;
5916 589 : int ret;
5917 :
5918 589 : *trimmed = 0;
5919 :
5920 : /* Discard not supported = nothing to do. */
5921 589 : if (!bdev_max_discard_sectors(device->bdev))
5922 : return 0;
5923 :
5924 : /* Not writable = nothing to do. */
5925 589 : if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5926 : return 0;
5927 :
5928 : /* No free space = nothing to do. */
5929 589 : if (device->total_bytes <= device->bytes_used)
5930 : return 0;
5931 :
5932 : ret = 0;
5933 :
5934 31 : while (1) {
5935 620 : struct btrfs_fs_info *fs_info = device->fs_info;
5936 620 : u64 bytes;
5937 :
5938 620 : ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
5939 620 : if (ret)
5940 : break;
5941 :
5942 620 : find_first_clear_extent_bit(&device->alloc_state, start,
5943 : &start, &end,
5944 : CHUNK_TRIMMED | CHUNK_ALLOCATED);
5945 :
5946 : /* Check if there are any CHUNK_* bits left */
5947 620 : if (start > device->total_bytes) {
5948 0 : WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
5949 0 : btrfs_warn_in_rcu(fs_info,
5950 : "ignoring attempt to trim beyond device size: offset %llu length %llu device %s device size %llu",
5951 : start, end - start + 1,
5952 : btrfs_dev_name(device),
5953 : device->total_bytes);
5954 0 : mutex_unlock(&fs_info->chunk_mutex);
5955 0 : ret = 0;
5956 0 : break;
5957 : }
5958 :
5959 : /* Ensure we skip the reserved space on each device. */
5960 620 : start = max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
5961 :
5962 : /*
5963 : * If find_first_clear_extent_bit find a range that spans the
5964 : * end of the device it will set end to -1, in this case it's up
5965 : * to the caller to trim the value to the size of the device.
5966 : */
5967 620 : end = min(end, device->total_bytes - 1);
5968 :
5969 620 : len = end - start + 1;
5970 :
5971 : /* We didn't find any extents */
5972 620 : if (!len) {
5973 589 : mutex_unlock(&fs_info->chunk_mutex);
5974 589 : ret = 0;
5975 589 : break;
5976 : }
5977 :
5978 31 : ret = btrfs_issue_discard(device->bdev, start, len,
5979 : &bytes);
5980 31 : if (!ret)
5981 31 : set_extent_bit(&device->alloc_state, start,
5982 31 : start + bytes - 1, CHUNK_TRIMMED, NULL);
5983 31 : mutex_unlock(&fs_info->chunk_mutex);
5984 :
5985 31 : if (ret)
5986 : break;
5987 :
5988 31 : start += len;
5989 31 : *trimmed += bytes;
5990 :
5991 31 : if (fatal_signal_pending(current)) {
5992 : ret = -ERESTARTSYS;
5993 : break;
5994 : }
5995 :
5996 31 : cond_resched();
5997 : }
5998 :
5999 589 : return ret;
6000 : }
6001 :
6002 : /*
6003 : * Trim the whole filesystem by:
6004 : * 1) trimming the free space in each block group
6005 : * 2) trimming the unallocated space on each device
6006 : *
6007 : * This will also continue trimming even if a block group or device encounters
6008 : * an error. The return value will be the last error, or 0 if nothing bad
6009 : * happens.
6010 : */
6011 594 : int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
6012 : {
6013 594 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6014 594 : struct btrfs_block_group *cache = NULL;
6015 594 : struct btrfs_device *device;
6016 594 : u64 group_trimmed;
6017 594 : u64 range_end = U64_MAX;
6018 594 : u64 start;
6019 594 : u64 end;
6020 594 : u64 trimmed = 0;
6021 594 : u64 bg_failed = 0;
6022 594 : u64 dev_failed = 0;
6023 594 : int bg_ret = 0;
6024 594 : int dev_ret = 0;
6025 594 : int ret = 0;
6026 :
6027 594 : if (range->start == U64_MAX)
6028 : return -EINVAL;
6029 :
6030 : /*
6031 : * Check range overflow if range->len is set.
6032 : * The default range->len is U64_MAX.
6033 : */
6034 589 : if (range->len != U64_MAX &&
6035 449 : check_add_overflow(range->start, range->len, &range_end))
6036 : return -EINVAL;
6037 :
6038 589 : cache = btrfs_lookup_first_block_group(fs_info, range->start);
6039 3335 : for (; cache; cache = btrfs_next_block_group(cache)) {
6040 2160 : if (cache->start >= range_end) {
6041 3 : btrfs_put_block_group(cache);
6042 3 : break;
6043 : }
6044 :
6045 2157 : start = max(range->start, cache->start);
6046 2157 : end = min(range_end, cache->start + cache->length);
6047 :
6048 2157 : if (end - start >= range->minlen) {
6049 1980 : if (!btrfs_block_group_done(cache)) {
6050 27 : ret = btrfs_cache_block_group(cache, true);
6051 27 : if (ret) {
6052 0 : bg_failed++;
6053 0 : bg_ret = ret;
6054 0 : continue;
6055 : }
6056 : }
6057 1980 : ret = btrfs_trim_block_group(cache,
6058 : &group_trimmed,
6059 : start,
6060 : end,
6061 : range->minlen);
6062 :
6063 1980 : trimmed += group_trimmed;
6064 1980 : if (ret) {
6065 5 : bg_failed++;
6066 5 : bg_ret = ret;
6067 5 : continue;
6068 : }
6069 : }
6070 : }
6071 :
6072 589 : if (bg_failed)
6073 1 : btrfs_warn(fs_info,
6074 : "failed to trim %llu block group(s), last error %d",
6075 : bg_failed, bg_ret);
6076 :
6077 589 : mutex_lock(&fs_devices->device_list_mutex);
6078 1178 : list_for_each_entry(device, &fs_devices->devices, dev_list) {
6079 1178 : if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
6080 0 : continue;
6081 :
6082 589 : ret = btrfs_trim_free_extents(device, &group_trimmed);
6083 589 : if (ret) {
6084 : dev_failed++;
6085 : dev_ret = ret;
6086 : break;
6087 : }
6088 :
6089 589 : trimmed += group_trimmed;
6090 : }
6091 589 : mutex_unlock(&fs_devices->device_list_mutex);
6092 :
6093 589 : if (dev_failed)
6094 0 : btrfs_warn(fs_info,
6095 : "failed to trim %llu device(s), last error %d",
6096 : dev_failed, dev_ret);
6097 589 : range->len = trimmed;
6098 589 : if (bg_ret)
6099 1 : return bg_ret;
6100 : return dev_ret;
6101 : }
|
