]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0 | |
2 | /* | |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
4 | */ | |
5 | ||
6 | #include <linux/fs.h> | |
7 | #include <linux/blkdev.h> | |
8 | #include <linux/radix-tree.h> | |
9 | #include <linux/writeback.h> | |
10 | #include <linux/workqueue.h> | |
11 | #include <linux/kthread.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/migrate.h> | |
14 | #include <linux/ratelimit.h> | |
15 | #include <linux/uuid.h> | |
16 | #include <linux/semaphore.h> | |
17 | #include <linux/error-injection.h> | |
18 | #include <linux/crc32c.h> | |
19 | #include <linux/sched/mm.h> | |
20 | #include <asm/unaligned.h> | |
21 | #include <crypto/hash.h> | |
22 | #include "ctree.h" | |
23 | #include "disk-io.h" | |
24 | #include "transaction.h" | |
25 | #include "btrfs_inode.h" | |
26 | #include "volumes.h" | |
27 | #include "print-tree.h" | |
28 | #include "locking.h" | |
29 | #include "tree-log.h" | |
30 | #include "free-space-cache.h" | |
31 | #include "free-space-tree.h" | |
32 | #include "check-integrity.h" | |
33 | #include "rcu-string.h" | |
34 | #include "dev-replace.h" | |
35 | #include "raid56.h" | |
36 | #include "sysfs.h" | |
37 | #include "qgroup.h" | |
38 | #include "compression.h" | |
39 | #include "tree-checker.h" | |
40 | #include "ref-verify.h" | |
41 | #include "block-group.h" | |
42 | #include "discard.h" | |
43 | #include "space-info.h" | |
44 | #include "zoned.h" | |
45 | ||
46 | #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\ | |
47 | BTRFS_HEADER_FLAG_RELOC |\ | |
48 | BTRFS_SUPER_FLAG_ERROR |\ | |
49 | BTRFS_SUPER_FLAG_SEEDING |\ | |
50 | BTRFS_SUPER_FLAG_METADUMP |\ | |
51 | BTRFS_SUPER_FLAG_METADUMP_V2) | |
52 | ||
53 | static void end_workqueue_fn(struct btrfs_work *work); | |
54 | static void btrfs_destroy_ordered_extents(struct btrfs_root *root); | |
55 | static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, | |
56 | struct btrfs_fs_info *fs_info); | |
57 | static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); | |
58 | static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, | |
59 | struct extent_io_tree *dirty_pages, | |
60 | int mark); | |
61 | static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, | |
62 | struct extent_io_tree *pinned_extents); | |
63 | static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info); | |
64 | static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info); | |
65 | ||
66 | /* | |
67 | * btrfs_end_io_wq structs are used to do processing in task context when an IO | |
68 | * is complete. This is used during reads to verify checksums, and it is used | |
69 | * by writes to insert metadata for new file extents after IO is complete. | |
70 | */ | |
71 | struct btrfs_end_io_wq { | |
72 | struct bio *bio; | |
73 | bio_end_io_t *end_io; | |
74 | void *private; | |
75 | struct btrfs_fs_info *info; | |
76 | blk_status_t status; | |
77 | enum btrfs_wq_endio_type metadata; | |
78 | struct btrfs_work work; | |
79 | }; | |
80 | ||
81 | static struct kmem_cache *btrfs_end_io_wq_cache; | |
82 | ||
83 | int __init btrfs_end_io_wq_init(void) | |
84 | { | |
85 | btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq", | |
86 | sizeof(struct btrfs_end_io_wq), | |
87 | 0, | |
88 | SLAB_MEM_SPREAD, | |
89 | NULL); | |
90 | if (!btrfs_end_io_wq_cache) | |
91 | return -ENOMEM; | |
92 | return 0; | |
93 | } | |
94 | ||
95 | void __cold btrfs_end_io_wq_exit(void) | |
96 | { | |
97 | kmem_cache_destroy(btrfs_end_io_wq_cache); | |
98 | } | |
99 | ||
100 | static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info) | |
101 | { | |
102 | if (fs_info->csum_shash) | |
103 | crypto_free_shash(fs_info->csum_shash); | |
104 | } | |
105 | ||
106 | /* | |
107 | * async submit bios are used to offload expensive checksumming | |
108 | * onto the worker threads. They checksum file and metadata bios | |
109 | * just before they are sent down the IO stack. | |
110 | */ | |
111 | struct async_submit_bio { | |
112 | struct inode *inode; | |
113 | struct bio *bio; | |
114 | extent_submit_bio_start_t *submit_bio_start; | |
115 | int mirror_num; | |
116 | ||
117 | /* Optional parameter for submit_bio_start used by direct io */ | |
118 | u64 dio_file_offset; | |
119 | struct btrfs_work work; | |
120 | blk_status_t status; | |
121 | }; | |
122 | ||
123 | /* | |
124 | * Lockdep class keys for extent_buffer->lock's in this root. For a given | |
125 | * eb, the lockdep key is determined by the btrfs_root it belongs to and | |
126 | * the level the eb occupies in the tree. | |
127 | * | |
128 | * Different roots are used for different purposes and may nest inside each | |
129 | * other and they require separate keysets. As lockdep keys should be | |
130 | * static, assign keysets according to the purpose of the root as indicated | |
131 | * by btrfs_root->root_key.objectid. This ensures that all special purpose | |
132 | * roots have separate keysets. | |
133 | * | |
134 | * Lock-nesting across peer nodes is always done with the immediate parent | |
135 | * node locked thus preventing deadlock. As lockdep doesn't know this, use | |
136 | * subclass to avoid triggering lockdep warning in such cases. | |
137 | * | |
138 | * The key is set by the readpage_end_io_hook after the buffer has passed | |
139 | * csum validation but before the pages are unlocked. It is also set by | |
140 | * btrfs_init_new_buffer on freshly allocated blocks. | |
141 | * | |
142 | * We also add a check to make sure the highest level of the tree is the | |
143 | * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code | |
144 | * needs update as well. | |
145 | */ | |
146 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
147 | # if BTRFS_MAX_LEVEL != 8 | |
148 | # error | |
149 | # endif | |
150 | ||
151 | #define DEFINE_LEVEL(stem, level) \ | |
152 | .names[level] = "btrfs-" stem "-0" #level, | |
153 | ||
154 | #define DEFINE_NAME(stem) \ | |
155 | DEFINE_LEVEL(stem, 0) \ | |
156 | DEFINE_LEVEL(stem, 1) \ | |
157 | DEFINE_LEVEL(stem, 2) \ | |
158 | DEFINE_LEVEL(stem, 3) \ | |
159 | DEFINE_LEVEL(stem, 4) \ | |
160 | DEFINE_LEVEL(stem, 5) \ | |
161 | DEFINE_LEVEL(stem, 6) \ | |
162 | DEFINE_LEVEL(stem, 7) | |
163 | ||
164 | static struct btrfs_lockdep_keyset { | |
165 | u64 id; /* root objectid */ | |
166 | /* Longest entry: btrfs-free-space-00 */ | |
167 | char names[BTRFS_MAX_LEVEL][20]; | |
168 | struct lock_class_key keys[BTRFS_MAX_LEVEL]; | |
169 | } btrfs_lockdep_keysets[] = { | |
170 | { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") }, | |
171 | { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") }, | |
172 | { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") }, | |
173 | { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") }, | |
174 | { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") }, | |
175 | { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") }, | |
176 | { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") }, | |
177 | { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") }, | |
178 | { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") }, | |
179 | { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") }, | |
180 | { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") }, | |
181 | { .id = 0, DEFINE_NAME("tree") }, | |
182 | }; | |
183 | ||
184 | #undef DEFINE_LEVEL | |
185 | #undef DEFINE_NAME | |
186 | ||
187 | void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, | |
188 | int level) | |
189 | { | |
190 | struct btrfs_lockdep_keyset *ks; | |
191 | ||
192 | BUG_ON(level >= ARRAY_SIZE(ks->keys)); | |
193 | ||
194 | /* find the matching keyset, id 0 is the default entry */ | |
195 | for (ks = btrfs_lockdep_keysets; ks->id; ks++) | |
196 | if (ks->id == objectid) | |
197 | break; | |
198 | ||
199 | lockdep_set_class_and_name(&eb->lock, | |
200 | &ks->keys[level], ks->names[level]); | |
201 | } | |
202 | ||
203 | #endif | |
204 | ||
205 | /* | |
206 | * Compute the csum of a btree block and store the result to provided buffer. | |
207 | */ | |
208 | static void csum_tree_block(struct extent_buffer *buf, u8 *result) | |
209 | { | |
210 | struct btrfs_fs_info *fs_info = buf->fs_info; | |
211 | const int num_pages = fs_info->nodesize >> PAGE_SHIFT; | |
212 | const int first_page_part = min_t(u32, PAGE_SIZE, fs_info->nodesize); | |
213 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
214 | char *kaddr; | |
215 | int i; | |
216 | ||
217 | shash->tfm = fs_info->csum_shash; | |
218 | crypto_shash_init(shash); | |
219 | kaddr = page_address(buf->pages[0]) + offset_in_page(buf->start); | |
220 | crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, | |
221 | first_page_part - BTRFS_CSUM_SIZE); | |
222 | ||
223 | for (i = 1; i < num_pages; i++) { | |
224 | kaddr = page_address(buf->pages[i]); | |
225 | crypto_shash_update(shash, kaddr, PAGE_SIZE); | |
226 | } | |
227 | memset(result, 0, BTRFS_CSUM_SIZE); | |
228 | crypto_shash_final(shash, result); | |
229 | } | |
230 | ||
231 | /* | |
232 | * we can't consider a given block up to date unless the transid of the | |
233 | * block matches the transid in the parent node's pointer. This is how we | |
234 | * detect blocks that either didn't get written at all or got written | |
235 | * in the wrong place. | |
236 | */ | |
237 | static int verify_parent_transid(struct extent_io_tree *io_tree, | |
238 | struct extent_buffer *eb, u64 parent_transid, | |
239 | int atomic) | |
240 | { | |
241 | struct extent_state *cached_state = NULL; | |
242 | int ret; | |
243 | bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); | |
244 | ||
245 | if (!parent_transid || btrfs_header_generation(eb) == parent_transid) | |
246 | return 0; | |
247 | ||
248 | if (atomic) | |
249 | return -EAGAIN; | |
250 | ||
251 | if (need_lock) | |
252 | btrfs_tree_read_lock(eb); | |
253 | ||
254 | lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, | |
255 | &cached_state); | |
256 | if (extent_buffer_uptodate(eb) && | |
257 | btrfs_header_generation(eb) == parent_transid) { | |
258 | ret = 0; | |
259 | goto out; | |
260 | } | |
261 | btrfs_err_rl(eb->fs_info, | |
262 | "parent transid verify failed on %llu wanted %llu found %llu", | |
263 | eb->start, | |
264 | parent_transid, btrfs_header_generation(eb)); | |
265 | ret = 1; | |
266 | ||
267 | /* | |
268 | * Things reading via commit roots that don't have normal protection, | |
269 | * like send, can have a really old block in cache that may point at a | |
270 | * block that has been freed and re-allocated. So don't clear uptodate | |
271 | * if we find an eb that is under IO (dirty/writeback) because we could | |
272 | * end up reading in the stale data and then writing it back out and | |
273 | * making everybody very sad. | |
274 | */ | |
275 | if (!extent_buffer_under_io(eb)) | |
276 | clear_extent_buffer_uptodate(eb); | |
277 | out: | |
278 | unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, | |
279 | &cached_state); | |
280 | if (need_lock) | |
281 | btrfs_tree_read_unlock(eb); | |
282 | return ret; | |
283 | } | |
284 | ||
285 | static bool btrfs_supported_super_csum(u16 csum_type) | |
286 | { | |
287 | switch (csum_type) { | |
288 | case BTRFS_CSUM_TYPE_CRC32: | |
289 | case BTRFS_CSUM_TYPE_XXHASH: | |
290 | case BTRFS_CSUM_TYPE_SHA256: | |
291 | case BTRFS_CSUM_TYPE_BLAKE2: | |
292 | return true; | |
293 | default: | |
294 | return false; | |
295 | } | |
296 | } | |
297 | ||
298 | /* | |
299 | * Return 0 if the superblock checksum type matches the checksum value of that | |
300 | * algorithm. Pass the raw disk superblock data. | |
301 | */ | |
302 | static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, | |
303 | char *raw_disk_sb) | |
304 | { | |
305 | struct btrfs_super_block *disk_sb = | |
306 | (struct btrfs_super_block *)raw_disk_sb; | |
307 | char result[BTRFS_CSUM_SIZE]; | |
308 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
309 | ||
310 | shash->tfm = fs_info->csum_shash; | |
311 | ||
312 | /* | |
313 | * The super_block structure does not span the whole | |
314 | * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is | |
315 | * filled with zeros and is included in the checksum. | |
316 | */ | |
317 | crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE, | |
318 | BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); | |
319 | ||
320 | if (memcmp(disk_sb->csum, result, fs_info->csum_size)) | |
321 | return 1; | |
322 | ||
323 | return 0; | |
324 | } | |
325 | ||
326 | int btrfs_verify_level_key(struct extent_buffer *eb, int level, | |
327 | struct btrfs_key *first_key, u64 parent_transid) | |
328 | { | |
329 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
330 | int found_level; | |
331 | struct btrfs_key found_key; | |
332 | int ret; | |
333 | ||
334 | found_level = btrfs_header_level(eb); | |
335 | if (found_level != level) { | |
336 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), | |
337 | KERN_ERR "BTRFS: tree level check failed\n"); | |
338 | btrfs_err(fs_info, | |
339 | "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", | |
340 | eb->start, level, found_level); | |
341 | return -EIO; | |
342 | } | |
343 | ||
344 | if (!first_key) | |
345 | return 0; | |
346 | ||
347 | /* | |
348 | * For live tree block (new tree blocks in current transaction), | |
349 | * we need proper lock context to avoid race, which is impossible here. | |
350 | * So we only checks tree blocks which is read from disk, whose | |
351 | * generation <= fs_info->last_trans_committed. | |
352 | */ | |
353 | if (btrfs_header_generation(eb) > fs_info->last_trans_committed) | |
354 | return 0; | |
355 | ||
356 | /* We have @first_key, so this @eb must have at least one item */ | |
357 | if (btrfs_header_nritems(eb) == 0) { | |
358 | btrfs_err(fs_info, | |
359 | "invalid tree nritems, bytenr=%llu nritems=0 expect >0", | |
360 | eb->start); | |
361 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
362 | return -EUCLEAN; | |
363 | } | |
364 | ||
365 | if (found_level) | |
366 | btrfs_node_key_to_cpu(eb, &found_key, 0); | |
367 | else | |
368 | btrfs_item_key_to_cpu(eb, &found_key, 0); | |
369 | ret = btrfs_comp_cpu_keys(first_key, &found_key); | |
370 | ||
371 | if (ret) { | |
372 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), | |
373 | KERN_ERR "BTRFS: tree first key check failed\n"); | |
374 | btrfs_err(fs_info, | |
375 | "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", | |
376 | eb->start, parent_transid, first_key->objectid, | |
377 | first_key->type, first_key->offset, | |
378 | found_key.objectid, found_key.type, | |
379 | found_key.offset); | |
380 | } | |
381 | return ret; | |
382 | } | |
383 | ||
384 | /* | |
385 | * helper to read a given tree block, doing retries as required when | |
386 | * the checksums don't match and we have alternate mirrors to try. | |
387 | * | |
388 | * @parent_transid: expected transid, skip check if 0 | |
389 | * @level: expected level, mandatory check | |
390 | * @first_key: expected key of first slot, skip check if NULL | |
391 | */ | |
392 | static int btree_read_extent_buffer_pages(struct extent_buffer *eb, | |
393 | u64 parent_transid, int level, | |
394 | struct btrfs_key *first_key) | |
395 | { | |
396 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
397 | struct extent_io_tree *io_tree; | |
398 | int failed = 0; | |
399 | int ret; | |
400 | int num_copies = 0; | |
401 | int mirror_num = 0; | |
402 | int failed_mirror = 0; | |
403 | ||
404 | io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; | |
405 | while (1) { | |
406 | clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); | |
407 | ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); | |
408 | if (!ret) { | |
409 | if (verify_parent_transid(io_tree, eb, | |
410 | parent_transid, 0)) | |
411 | ret = -EIO; | |
412 | else if (btrfs_verify_level_key(eb, level, | |
413 | first_key, parent_transid)) | |
414 | ret = -EUCLEAN; | |
415 | else | |
416 | break; | |
417 | } | |
418 | ||
419 | num_copies = btrfs_num_copies(fs_info, | |
420 | eb->start, eb->len); | |
421 | if (num_copies == 1) | |
422 | break; | |
423 | ||
424 | if (!failed_mirror) { | |
425 | failed = 1; | |
426 | failed_mirror = eb->read_mirror; | |
427 | } | |
428 | ||
429 | mirror_num++; | |
430 | if (mirror_num == failed_mirror) | |
431 | mirror_num++; | |
432 | ||
433 | if (mirror_num > num_copies) | |
434 | break; | |
435 | } | |
436 | ||
437 | if (failed && !ret && failed_mirror) | |
438 | btrfs_repair_eb_io_failure(eb, failed_mirror); | |
439 | ||
440 | return ret; | |
441 | } | |
442 | ||
443 | /* | |
444 | * Checksum a dirty tree block before IO. This has extra checks to make sure | |
445 | * we only fill in the checksum field in the first page of a multi-page block. | |
446 | * For subpage extent buffers we need bvec to also read the offset in the page. | |
447 | */ | |
448 | static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct bio_vec *bvec) | |
449 | { | |
450 | struct page *page = bvec->bv_page; | |
451 | u64 start = page_offset(page); | |
452 | u64 found_start; | |
453 | u8 result[BTRFS_CSUM_SIZE]; | |
454 | struct extent_buffer *eb; | |
455 | int ret; | |
456 | ||
457 | eb = (struct extent_buffer *)page->private; | |
458 | if (page != eb->pages[0]) | |
459 | return 0; | |
460 | ||
461 | found_start = btrfs_header_bytenr(eb); | |
462 | /* | |
463 | * Please do not consolidate these warnings into a single if. | |
464 | * It is useful to know what went wrong. | |
465 | */ | |
466 | if (WARN_ON(found_start != start)) | |
467 | return -EUCLEAN; | |
468 | if (WARN_ON(!PageUptodate(page))) | |
469 | return -EUCLEAN; | |
470 | ||
471 | ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, | |
472 | offsetof(struct btrfs_header, fsid), | |
473 | BTRFS_FSID_SIZE) == 0); | |
474 | ||
475 | csum_tree_block(eb, result); | |
476 | ||
477 | if (btrfs_header_level(eb)) | |
478 | ret = btrfs_check_node(eb); | |
479 | else | |
480 | ret = btrfs_check_leaf_full(eb); | |
481 | ||
482 | if (ret < 0) { | |
483 | btrfs_print_tree(eb, 0); | |
484 | btrfs_err(fs_info, | |
485 | "block=%llu write time tree block corruption detected", | |
486 | eb->start); | |
487 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
488 | return ret; | |
489 | } | |
490 | write_extent_buffer(eb, result, 0, fs_info->csum_size); | |
491 | ||
492 | return 0; | |
493 | } | |
494 | ||
495 | static int check_tree_block_fsid(struct extent_buffer *eb) | |
496 | { | |
497 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
498 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; | |
499 | u8 fsid[BTRFS_FSID_SIZE]; | |
500 | u8 *metadata_uuid; | |
501 | ||
502 | read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), | |
503 | BTRFS_FSID_SIZE); | |
504 | /* | |
505 | * Checking the incompat flag is only valid for the current fs. For | |
506 | * seed devices it's forbidden to have their uuid changed so reading | |
507 | * ->fsid in this case is fine | |
508 | */ | |
509 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) | |
510 | metadata_uuid = fs_devices->metadata_uuid; | |
511 | else | |
512 | metadata_uuid = fs_devices->fsid; | |
513 | ||
514 | if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) | |
515 | return 0; | |
516 | ||
517 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) | |
518 | if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) | |
519 | return 0; | |
520 | ||
521 | return 1; | |
522 | } | |
523 | ||
524 | /* Do basic extent buffer checks at read time */ | |
525 | static int validate_extent_buffer(struct extent_buffer *eb) | |
526 | { | |
527 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
528 | u64 found_start; | |
529 | const u32 csum_size = fs_info->csum_size; | |
530 | u8 found_level; | |
531 | u8 result[BTRFS_CSUM_SIZE]; | |
532 | int ret = 0; | |
533 | ||
534 | found_start = btrfs_header_bytenr(eb); | |
535 | if (found_start != eb->start) { | |
536 | btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", | |
537 | eb->start, found_start); | |
538 | ret = -EIO; | |
539 | goto out; | |
540 | } | |
541 | if (check_tree_block_fsid(eb)) { | |
542 | btrfs_err_rl(fs_info, "bad fsid on block %llu", | |
543 | eb->start); | |
544 | ret = -EIO; | |
545 | goto out; | |
546 | } | |
547 | found_level = btrfs_header_level(eb); | |
548 | if (found_level >= BTRFS_MAX_LEVEL) { | |
549 | btrfs_err(fs_info, "bad tree block level %d on %llu", | |
550 | (int)btrfs_header_level(eb), eb->start); | |
551 | ret = -EIO; | |
552 | goto out; | |
553 | } | |
554 | ||
555 | csum_tree_block(eb, result); | |
556 | ||
557 | if (memcmp_extent_buffer(eb, result, 0, csum_size)) { | |
558 | u8 val[BTRFS_CSUM_SIZE] = { 0 }; | |
559 | ||
560 | read_extent_buffer(eb, &val, 0, csum_size); | |
561 | btrfs_warn_rl(fs_info, | |
562 | "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d", | |
563 | fs_info->sb->s_id, eb->start, | |
564 | CSUM_FMT_VALUE(csum_size, val), | |
565 | CSUM_FMT_VALUE(csum_size, result), | |
566 | btrfs_header_level(eb)); | |
567 | ret = -EUCLEAN; | |
568 | goto out; | |
569 | } | |
570 | ||
571 | /* | |
572 | * If this is a leaf block and it is corrupt, set the corrupt bit so | |
573 | * that we don't try and read the other copies of this block, just | |
574 | * return -EIO. | |
575 | */ | |
576 | if (found_level == 0 && btrfs_check_leaf_full(eb)) { | |
577 | set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); | |
578 | ret = -EIO; | |
579 | } | |
580 | ||
581 | if (found_level > 0 && btrfs_check_node(eb)) | |
582 | ret = -EIO; | |
583 | ||
584 | if (!ret) | |
585 | set_extent_buffer_uptodate(eb); | |
586 | else | |
587 | btrfs_err(fs_info, | |
588 | "block=%llu read time tree block corruption detected", | |
589 | eb->start); | |
590 | out: | |
591 | return ret; | |
592 | } | |
593 | ||
594 | int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio, | |
595 | struct page *page, u64 start, u64 end, | |
596 | int mirror) | |
597 | { | |
598 | struct extent_buffer *eb; | |
599 | int ret = 0; | |
600 | int reads_done; | |
601 | ||
602 | ASSERT(page->private); | |
603 | eb = (struct extent_buffer *)page->private; | |
604 | ||
605 | /* | |
606 | * The pending IO might have been the only thing that kept this buffer | |
607 | * in memory. Make sure we have a ref for all this other checks | |
608 | */ | |
609 | atomic_inc(&eb->refs); | |
610 | ||
611 | reads_done = atomic_dec_and_test(&eb->io_pages); | |
612 | if (!reads_done) | |
613 | goto err; | |
614 | ||
615 | eb->read_mirror = mirror; | |
616 | if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { | |
617 | ret = -EIO; | |
618 | goto err; | |
619 | } | |
620 | ret = validate_extent_buffer(eb); | |
621 | err: | |
622 | if (reads_done && | |
623 | test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) | |
624 | btree_readahead_hook(eb, ret); | |
625 | ||
626 | if (ret) { | |
627 | /* | |
628 | * our io error hook is going to dec the io pages | |
629 | * again, we have to make sure it has something | |
630 | * to decrement | |
631 | */ | |
632 | atomic_inc(&eb->io_pages); | |
633 | clear_extent_buffer_uptodate(eb); | |
634 | } | |
635 | free_extent_buffer(eb); | |
636 | ||
637 | return ret; | |
638 | } | |
639 | ||
640 | static void end_workqueue_bio(struct bio *bio) | |
641 | { | |
642 | struct btrfs_end_io_wq *end_io_wq = bio->bi_private; | |
643 | struct btrfs_fs_info *fs_info; | |
644 | struct btrfs_workqueue *wq; | |
645 | ||
646 | fs_info = end_io_wq->info; | |
647 | end_io_wq->status = bio->bi_status; | |
648 | ||
649 | if (bio_op(bio) == REQ_OP_WRITE) { | |
650 | if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) | |
651 | wq = fs_info->endio_meta_write_workers; | |
652 | else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) | |
653 | wq = fs_info->endio_freespace_worker; | |
654 | else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) | |
655 | wq = fs_info->endio_raid56_workers; | |
656 | else | |
657 | wq = fs_info->endio_write_workers; | |
658 | } else { | |
659 | if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) | |
660 | wq = fs_info->endio_raid56_workers; | |
661 | else if (end_io_wq->metadata) | |
662 | wq = fs_info->endio_meta_workers; | |
663 | else | |
664 | wq = fs_info->endio_workers; | |
665 | } | |
666 | ||
667 | btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); | |
668 | btrfs_queue_work(wq, &end_io_wq->work); | |
669 | } | |
670 | ||
671 | blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, | |
672 | enum btrfs_wq_endio_type metadata) | |
673 | { | |
674 | struct btrfs_end_io_wq *end_io_wq; | |
675 | ||
676 | end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); | |
677 | if (!end_io_wq) | |
678 | return BLK_STS_RESOURCE; | |
679 | ||
680 | end_io_wq->private = bio->bi_private; | |
681 | end_io_wq->end_io = bio->bi_end_io; | |
682 | end_io_wq->info = info; | |
683 | end_io_wq->status = 0; | |
684 | end_io_wq->bio = bio; | |
685 | end_io_wq->metadata = metadata; | |
686 | ||
687 | bio->bi_private = end_io_wq; | |
688 | bio->bi_end_io = end_workqueue_bio; | |
689 | return 0; | |
690 | } | |
691 | ||
692 | static void run_one_async_start(struct btrfs_work *work) | |
693 | { | |
694 | struct async_submit_bio *async; | |
695 | blk_status_t ret; | |
696 | ||
697 | async = container_of(work, struct async_submit_bio, work); | |
698 | ret = async->submit_bio_start(async->inode, async->bio, | |
699 | async->dio_file_offset); | |
700 | if (ret) | |
701 | async->status = ret; | |
702 | } | |
703 | ||
704 | /* | |
705 | * In order to insert checksums into the metadata in large chunks, we wait | |
706 | * until bio submission time. All the pages in the bio are checksummed and | |
707 | * sums are attached onto the ordered extent record. | |
708 | * | |
709 | * At IO completion time the csums attached on the ordered extent record are | |
710 | * inserted into the tree. | |
711 | */ | |
712 | static void run_one_async_done(struct btrfs_work *work) | |
713 | { | |
714 | struct async_submit_bio *async; | |
715 | struct inode *inode; | |
716 | blk_status_t ret; | |
717 | ||
718 | async = container_of(work, struct async_submit_bio, work); | |
719 | inode = async->inode; | |
720 | ||
721 | /* If an error occurred we just want to clean up the bio and move on */ | |
722 | if (async->status) { | |
723 | async->bio->bi_status = async->status; | |
724 | bio_endio(async->bio); | |
725 | return; | |
726 | } | |
727 | ||
728 | /* | |
729 | * All of the bios that pass through here are from async helpers. | |
730 | * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. | |
731 | * This changes nothing when cgroups aren't in use. | |
732 | */ | |
733 | async->bio->bi_opf |= REQ_CGROUP_PUNT; | |
734 | ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); | |
735 | if (ret) { | |
736 | async->bio->bi_status = ret; | |
737 | bio_endio(async->bio); | |
738 | } | |
739 | } | |
740 | ||
741 | static void run_one_async_free(struct btrfs_work *work) | |
742 | { | |
743 | struct async_submit_bio *async; | |
744 | ||
745 | async = container_of(work, struct async_submit_bio, work); | |
746 | kfree(async); | |
747 | } | |
748 | ||
749 | blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, | |
750 | int mirror_num, unsigned long bio_flags, | |
751 | u64 dio_file_offset, | |
752 | extent_submit_bio_start_t *submit_bio_start) | |
753 | { | |
754 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; | |
755 | struct async_submit_bio *async; | |
756 | ||
757 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
758 | if (!async) | |
759 | return BLK_STS_RESOURCE; | |
760 | ||
761 | async->inode = inode; | |
762 | async->bio = bio; | |
763 | async->mirror_num = mirror_num; | |
764 | async->submit_bio_start = submit_bio_start; | |
765 | ||
766 | btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, | |
767 | run_one_async_free); | |
768 | ||
769 | async->dio_file_offset = dio_file_offset; | |
770 | ||
771 | async->status = 0; | |
772 | ||
773 | if (op_is_sync(bio->bi_opf)) | |
774 | btrfs_set_work_high_priority(&async->work); | |
775 | ||
776 | btrfs_queue_work(fs_info->workers, &async->work); | |
777 | return 0; | |
778 | } | |
779 | ||
780 | static blk_status_t btree_csum_one_bio(struct bio *bio) | |
781 | { | |
782 | struct bio_vec *bvec; | |
783 | struct btrfs_root *root; | |
784 | int ret = 0; | |
785 | struct bvec_iter_all iter_all; | |
786 | ||
787 | ASSERT(!bio_flagged(bio, BIO_CLONED)); | |
788 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
789 | root = BTRFS_I(bvec->bv_page->mapping->host)->root; | |
790 | ret = csum_dirty_buffer(root->fs_info, bvec); | |
791 | if (ret) | |
792 | break; | |
793 | } | |
794 | ||
795 | return errno_to_blk_status(ret); | |
796 | } | |
797 | ||
798 | static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio, | |
799 | u64 dio_file_offset) | |
800 | { | |
801 | /* | |
802 | * when we're called for a write, we're already in the async | |
803 | * submission context. Just jump into btrfs_map_bio | |
804 | */ | |
805 | return btree_csum_one_bio(bio); | |
806 | } | |
807 | ||
808 | static int check_async_write(struct btrfs_fs_info *fs_info, | |
809 | struct btrfs_inode *bi) | |
810 | { | |
811 | if (atomic_read(&bi->sync_writers)) | |
812 | return 0; | |
813 | if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) | |
814 | return 0; | |
815 | return 1; | |
816 | } | |
817 | ||
818 | blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, | |
819 | int mirror_num, unsigned long bio_flags) | |
820 | { | |
821 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
822 | int async = check_async_write(fs_info, BTRFS_I(inode)); | |
823 | blk_status_t ret; | |
824 | ||
825 | if (bio_op(bio) != REQ_OP_WRITE) { | |
826 | /* | |
827 | * called for a read, do the setup so that checksum validation | |
828 | * can happen in the async kernel threads | |
829 | */ | |
830 | ret = btrfs_bio_wq_end_io(fs_info, bio, | |
831 | BTRFS_WQ_ENDIO_METADATA); | |
832 | if (ret) | |
833 | goto out_w_error; | |
834 | ret = btrfs_map_bio(fs_info, bio, mirror_num); | |
835 | } else if (!async) { | |
836 | ret = btree_csum_one_bio(bio); | |
837 | if (ret) | |
838 | goto out_w_error; | |
839 | ret = btrfs_map_bio(fs_info, bio, mirror_num); | |
840 | } else { | |
841 | /* | |
842 | * kthread helpers are used to submit writes so that | |
843 | * checksumming can happen in parallel across all CPUs | |
844 | */ | |
845 | ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0, | |
846 | 0, btree_submit_bio_start); | |
847 | } | |
848 | ||
849 | if (ret) | |
850 | goto out_w_error; | |
851 | return 0; | |
852 | ||
853 | out_w_error: | |
854 | bio->bi_status = ret; | |
855 | bio_endio(bio); | |
856 | return ret; | |
857 | } | |
858 | ||
859 | #ifdef CONFIG_MIGRATION | |
860 | static int btree_migratepage(struct address_space *mapping, | |
861 | struct page *newpage, struct page *page, | |
862 | enum migrate_mode mode) | |
863 | { | |
864 | /* | |
865 | * we can't safely write a btree page from here, | |
866 | * we haven't done the locking hook | |
867 | */ | |
868 | if (PageDirty(page)) | |
869 | return -EAGAIN; | |
870 | /* | |
871 | * Buffers may be managed in a filesystem specific way. | |
872 | * We must have no buffers or drop them. | |
873 | */ | |
874 | if (page_has_private(page) && | |
875 | !try_to_release_page(page, GFP_KERNEL)) | |
876 | return -EAGAIN; | |
877 | return migrate_page(mapping, newpage, page, mode); | |
878 | } | |
879 | #endif | |
880 | ||
881 | ||
882 | static int btree_writepages(struct address_space *mapping, | |
883 | struct writeback_control *wbc) | |
884 | { | |
885 | struct btrfs_fs_info *fs_info; | |
886 | int ret; | |
887 | ||
888 | if (wbc->sync_mode == WB_SYNC_NONE) { | |
889 | ||
890 | if (wbc->for_kupdate) | |
891 | return 0; | |
892 | ||
893 | fs_info = BTRFS_I(mapping->host)->root->fs_info; | |
894 | /* this is a bit racy, but that's ok */ | |
895 | ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, | |
896 | BTRFS_DIRTY_METADATA_THRESH, | |
897 | fs_info->dirty_metadata_batch); | |
898 | if (ret < 0) | |
899 | return 0; | |
900 | } | |
901 | return btree_write_cache_pages(mapping, wbc); | |
902 | } | |
903 | ||
904 | static int btree_releasepage(struct page *page, gfp_t gfp_flags) | |
905 | { | |
906 | if (PageWriteback(page) || PageDirty(page)) | |
907 | return 0; | |
908 | ||
909 | return try_release_extent_buffer(page); | |
910 | } | |
911 | ||
912 | static void btree_invalidatepage(struct page *page, unsigned int offset, | |
913 | unsigned int length) | |
914 | { | |
915 | struct extent_io_tree *tree; | |
916 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
917 | extent_invalidatepage(tree, page, offset); | |
918 | btree_releasepage(page, GFP_NOFS); | |
919 | if (PagePrivate(page)) { | |
920 | btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, | |
921 | "page private not zero on page %llu", | |
922 | (unsigned long long)page_offset(page)); | |
923 | detach_page_private(page); | |
924 | } | |
925 | } | |
926 | ||
927 | static int btree_set_page_dirty(struct page *page) | |
928 | { | |
929 | #ifdef DEBUG | |
930 | struct extent_buffer *eb; | |
931 | ||
932 | BUG_ON(!PagePrivate(page)); | |
933 | eb = (struct extent_buffer *)page->private; | |
934 | BUG_ON(!eb); | |
935 | BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
936 | BUG_ON(!atomic_read(&eb->refs)); | |
937 | btrfs_assert_tree_locked(eb); | |
938 | #endif | |
939 | return __set_page_dirty_nobuffers(page); | |
940 | } | |
941 | ||
942 | static const struct address_space_operations btree_aops = { | |
943 | .writepages = btree_writepages, | |
944 | .releasepage = btree_releasepage, | |
945 | .invalidatepage = btree_invalidatepage, | |
946 | #ifdef CONFIG_MIGRATION | |
947 | .migratepage = btree_migratepage, | |
948 | #endif | |
949 | .set_page_dirty = btree_set_page_dirty, | |
950 | }; | |
951 | ||
952 | struct extent_buffer *btrfs_find_create_tree_block( | |
953 | struct btrfs_fs_info *fs_info, | |
954 | u64 bytenr, u64 owner_root, | |
955 | int level) | |
956 | { | |
957 | if (btrfs_is_testing(fs_info)) | |
958 | return alloc_test_extent_buffer(fs_info, bytenr); | |
959 | return alloc_extent_buffer(fs_info, bytenr, owner_root, level); | |
960 | } | |
961 | ||
962 | /* | |
963 | * Read tree block at logical address @bytenr and do variant basic but critical | |
964 | * verification. | |
965 | * | |
966 | * @owner_root: the objectid of the root owner for this block. | |
967 | * @parent_transid: expected transid of this tree block, skip check if 0 | |
968 | * @level: expected level, mandatory check | |
969 | * @first_key: expected key in slot 0, skip check if NULL | |
970 | */ | |
971 | struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, | |
972 | u64 owner_root, u64 parent_transid, | |
973 | int level, struct btrfs_key *first_key) | |
974 | { | |
975 | struct extent_buffer *buf = NULL; | |
976 | int ret; | |
977 | ||
978 | buf = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); | |
979 | if (IS_ERR(buf)) | |
980 | return buf; | |
981 | ||
982 | ret = btree_read_extent_buffer_pages(buf, parent_transid, | |
983 | level, first_key); | |
984 | if (ret) { | |
985 | free_extent_buffer_stale(buf); | |
986 | return ERR_PTR(ret); | |
987 | } | |
988 | return buf; | |
989 | ||
990 | } | |
991 | ||
992 | void btrfs_clean_tree_block(struct extent_buffer *buf) | |
993 | { | |
994 | struct btrfs_fs_info *fs_info = buf->fs_info; | |
995 | if (btrfs_header_generation(buf) == | |
996 | fs_info->running_transaction->transid) { | |
997 | btrfs_assert_tree_locked(buf); | |
998 | ||
999 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { | |
1000 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, | |
1001 | -buf->len, | |
1002 | fs_info->dirty_metadata_batch); | |
1003 | clear_extent_buffer_dirty(buf); | |
1004 | } | |
1005 | } | |
1006 | } | |
1007 | ||
1008 | static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, | |
1009 | u64 objectid) | |
1010 | { | |
1011 | bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); | |
1012 | root->fs_info = fs_info; | |
1013 | root->node = NULL; | |
1014 | root->commit_root = NULL; | |
1015 | root->state = 0; | |
1016 | root->orphan_cleanup_state = 0; | |
1017 | ||
1018 | root->last_trans = 0; | |
1019 | root->highest_objectid = 0; | |
1020 | root->nr_delalloc_inodes = 0; | |
1021 | root->nr_ordered_extents = 0; | |
1022 | root->inode_tree = RB_ROOT; | |
1023 | INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); | |
1024 | root->block_rsv = NULL; | |
1025 | ||
1026 | INIT_LIST_HEAD(&root->dirty_list); | |
1027 | INIT_LIST_HEAD(&root->root_list); | |
1028 | INIT_LIST_HEAD(&root->delalloc_inodes); | |
1029 | INIT_LIST_HEAD(&root->delalloc_root); | |
1030 | INIT_LIST_HEAD(&root->ordered_extents); | |
1031 | INIT_LIST_HEAD(&root->ordered_root); | |
1032 | INIT_LIST_HEAD(&root->reloc_dirty_list); | |
1033 | INIT_LIST_HEAD(&root->logged_list[0]); | |
1034 | INIT_LIST_HEAD(&root->logged_list[1]); | |
1035 | spin_lock_init(&root->inode_lock); | |
1036 | spin_lock_init(&root->delalloc_lock); | |
1037 | spin_lock_init(&root->ordered_extent_lock); | |
1038 | spin_lock_init(&root->accounting_lock); | |
1039 | spin_lock_init(&root->log_extents_lock[0]); | |
1040 | spin_lock_init(&root->log_extents_lock[1]); | |
1041 | spin_lock_init(&root->qgroup_meta_rsv_lock); | |
1042 | mutex_init(&root->objectid_mutex); | |
1043 | mutex_init(&root->log_mutex); | |
1044 | mutex_init(&root->ordered_extent_mutex); | |
1045 | mutex_init(&root->delalloc_mutex); | |
1046 | init_waitqueue_head(&root->qgroup_flush_wait); | |
1047 | init_waitqueue_head(&root->log_writer_wait); | |
1048 | init_waitqueue_head(&root->log_commit_wait[0]); | |
1049 | init_waitqueue_head(&root->log_commit_wait[1]); | |
1050 | INIT_LIST_HEAD(&root->log_ctxs[0]); | |
1051 | INIT_LIST_HEAD(&root->log_ctxs[1]); | |
1052 | atomic_set(&root->log_commit[0], 0); | |
1053 | atomic_set(&root->log_commit[1], 0); | |
1054 | atomic_set(&root->log_writers, 0); | |
1055 | atomic_set(&root->log_batch, 0); | |
1056 | refcount_set(&root->refs, 1); | |
1057 | atomic_set(&root->snapshot_force_cow, 0); | |
1058 | atomic_set(&root->nr_swapfiles, 0); | |
1059 | root->log_transid = 0; | |
1060 | root->log_transid_committed = -1; | |
1061 | root->last_log_commit = 0; | |
1062 | if (!dummy) { | |
1063 | extent_io_tree_init(fs_info, &root->dirty_log_pages, | |
1064 | IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); | |
1065 | extent_io_tree_init(fs_info, &root->log_csum_range, | |
1066 | IO_TREE_LOG_CSUM_RANGE, NULL); | |
1067 | } | |
1068 | ||
1069 | memset(&root->root_key, 0, sizeof(root->root_key)); | |
1070 | memset(&root->root_item, 0, sizeof(root->root_item)); | |
1071 | memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); | |
1072 | root->root_key.objectid = objectid; | |
1073 | root->anon_dev = 0; | |
1074 | ||
1075 | spin_lock_init(&root->root_item_lock); | |
1076 | btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); | |
1077 | #ifdef CONFIG_BTRFS_DEBUG | |
1078 | INIT_LIST_HEAD(&root->leak_list); | |
1079 | spin_lock(&fs_info->fs_roots_radix_lock); | |
1080 | list_add_tail(&root->leak_list, &fs_info->allocated_roots); | |
1081 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
1082 | #endif | |
1083 | } | |
1084 | ||
1085 | static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, | |
1086 | u64 objectid, gfp_t flags) | |
1087 | { | |
1088 | struct btrfs_root *root = kzalloc(sizeof(*root), flags); | |
1089 | if (root) | |
1090 | __setup_root(root, fs_info, objectid); | |
1091 | return root; | |
1092 | } | |
1093 | ||
1094 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
1095 | /* Should only be used by the testing infrastructure */ | |
1096 | struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) | |
1097 | { | |
1098 | struct btrfs_root *root; | |
1099 | ||
1100 | if (!fs_info) | |
1101 | return ERR_PTR(-EINVAL); | |
1102 | ||
1103 | root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); | |
1104 | if (!root) | |
1105 | return ERR_PTR(-ENOMEM); | |
1106 | ||
1107 | /* We don't use the stripesize in selftest, set it as sectorsize */ | |
1108 | root->alloc_bytenr = 0; | |
1109 | ||
1110 | return root; | |
1111 | } | |
1112 | #endif | |
1113 | ||
1114 | struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, | |
1115 | u64 objectid) | |
1116 | { | |
1117 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
1118 | struct extent_buffer *leaf; | |
1119 | struct btrfs_root *tree_root = fs_info->tree_root; | |
1120 | struct btrfs_root *root; | |
1121 | struct btrfs_key key; | |
1122 | unsigned int nofs_flag; | |
1123 | int ret = 0; | |
1124 | ||
1125 | /* | |
1126 | * We're holding a transaction handle, so use a NOFS memory allocation | |
1127 | * context to avoid deadlock if reclaim happens. | |
1128 | */ | |
1129 | nofs_flag = memalloc_nofs_save(); | |
1130 | root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); | |
1131 | memalloc_nofs_restore(nofs_flag); | |
1132 | if (!root) | |
1133 | return ERR_PTR(-ENOMEM); | |
1134 | ||
1135 | root->root_key.objectid = objectid; | |
1136 | root->root_key.type = BTRFS_ROOT_ITEM_KEY; | |
1137 | root->root_key.offset = 0; | |
1138 | ||
1139 | leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, | |
1140 | BTRFS_NESTING_NORMAL); | |
1141 | if (IS_ERR(leaf)) { | |
1142 | ret = PTR_ERR(leaf); | |
1143 | leaf = NULL; | |
1144 | goto fail_unlock; | |
1145 | } | |
1146 | ||
1147 | root->node = leaf; | |
1148 | btrfs_mark_buffer_dirty(leaf); | |
1149 | ||
1150 | root->commit_root = btrfs_root_node(root); | |
1151 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
1152 | ||
1153 | btrfs_set_root_flags(&root->root_item, 0); | |
1154 | btrfs_set_root_limit(&root->root_item, 0); | |
1155 | btrfs_set_root_bytenr(&root->root_item, leaf->start); | |
1156 | btrfs_set_root_generation(&root->root_item, trans->transid); | |
1157 | btrfs_set_root_level(&root->root_item, 0); | |
1158 | btrfs_set_root_refs(&root->root_item, 1); | |
1159 | btrfs_set_root_used(&root->root_item, leaf->len); | |
1160 | btrfs_set_root_last_snapshot(&root->root_item, 0); | |
1161 | btrfs_set_root_dirid(&root->root_item, 0); | |
1162 | if (is_fstree(objectid)) | |
1163 | generate_random_guid(root->root_item.uuid); | |
1164 | else | |
1165 | export_guid(root->root_item.uuid, &guid_null); | |
1166 | btrfs_set_root_drop_level(&root->root_item, 0); | |
1167 | ||
1168 | btrfs_tree_unlock(leaf); | |
1169 | ||
1170 | key.objectid = objectid; | |
1171 | key.type = BTRFS_ROOT_ITEM_KEY; | |
1172 | key.offset = 0; | |
1173 | ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); | |
1174 | if (ret) | |
1175 | goto fail; | |
1176 | ||
1177 | return root; | |
1178 | ||
1179 | fail_unlock: | |
1180 | if (leaf) | |
1181 | btrfs_tree_unlock(leaf); | |
1182 | fail: | |
1183 | btrfs_put_root(root); | |
1184 | ||
1185 | return ERR_PTR(ret); | |
1186 | } | |
1187 | ||
1188 | static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, | |
1189 | struct btrfs_fs_info *fs_info) | |
1190 | { | |
1191 | struct btrfs_root *root; | |
1192 | struct extent_buffer *leaf; | |
1193 | ||
1194 | root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); | |
1195 | if (!root) | |
1196 | return ERR_PTR(-ENOMEM); | |
1197 | ||
1198 | root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; | |
1199 | root->root_key.type = BTRFS_ROOT_ITEM_KEY; | |
1200 | root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; | |
1201 | ||
1202 | /* | |
1203 | * DON'T set SHAREABLE bit for log trees. | |
1204 | * | |
1205 | * Log trees are not exposed to user space thus can't be snapshotted, | |
1206 | * and they go away before a real commit is actually done. | |
1207 | * | |
1208 | * They do store pointers to file data extents, and those reference | |
1209 | * counts still get updated (along with back refs to the log tree). | |
1210 | */ | |
1211 | ||
1212 | leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, | |
1213 | NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); | |
1214 | if (IS_ERR(leaf)) { | |
1215 | btrfs_put_root(root); | |
1216 | return ERR_CAST(leaf); | |
1217 | } | |
1218 | ||
1219 | root->node = leaf; | |
1220 | ||
1221 | btrfs_mark_buffer_dirty(root->node); | |
1222 | btrfs_tree_unlock(root->node); | |
1223 | return root; | |
1224 | } | |
1225 | ||
1226 | int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, | |
1227 | struct btrfs_fs_info *fs_info) | |
1228 | { | |
1229 | struct btrfs_root *log_root; | |
1230 | ||
1231 | log_root = alloc_log_tree(trans, fs_info); | |
1232 | if (IS_ERR(log_root)) | |
1233 | return PTR_ERR(log_root); | |
1234 | WARN_ON(fs_info->log_root_tree); | |
1235 | fs_info->log_root_tree = log_root; | |
1236 | return 0; | |
1237 | } | |
1238 | ||
1239 | int btrfs_add_log_tree(struct btrfs_trans_handle *trans, | |
1240 | struct btrfs_root *root) | |
1241 | { | |
1242 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1243 | struct btrfs_root *log_root; | |
1244 | struct btrfs_inode_item *inode_item; | |
1245 | ||
1246 | log_root = alloc_log_tree(trans, fs_info); | |
1247 | if (IS_ERR(log_root)) | |
1248 | return PTR_ERR(log_root); | |
1249 | ||
1250 | log_root->last_trans = trans->transid; | |
1251 | log_root->root_key.offset = root->root_key.objectid; | |
1252 | ||
1253 | inode_item = &log_root->root_item.inode; | |
1254 | btrfs_set_stack_inode_generation(inode_item, 1); | |
1255 | btrfs_set_stack_inode_size(inode_item, 3); | |
1256 | btrfs_set_stack_inode_nlink(inode_item, 1); | |
1257 | btrfs_set_stack_inode_nbytes(inode_item, | |
1258 | fs_info->nodesize); | |
1259 | btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); | |
1260 | ||
1261 | btrfs_set_root_node(&log_root->root_item, log_root->node); | |
1262 | ||
1263 | WARN_ON(root->log_root); | |
1264 | root->log_root = log_root; | |
1265 | root->log_transid = 0; | |
1266 | root->log_transid_committed = -1; | |
1267 | root->last_log_commit = 0; | |
1268 | return 0; | |
1269 | } | |
1270 | ||
1271 | static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, | |
1272 | struct btrfs_path *path, | |
1273 | struct btrfs_key *key) | |
1274 | { | |
1275 | struct btrfs_root *root; | |
1276 | struct btrfs_fs_info *fs_info = tree_root->fs_info; | |
1277 | u64 generation; | |
1278 | int ret; | |
1279 | int level; | |
1280 | ||
1281 | root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); | |
1282 | if (!root) | |
1283 | return ERR_PTR(-ENOMEM); | |
1284 | ||
1285 | ret = btrfs_find_root(tree_root, key, path, | |
1286 | &root->root_item, &root->root_key); | |
1287 | if (ret) { | |
1288 | if (ret > 0) | |
1289 | ret = -ENOENT; | |
1290 | goto fail; | |
1291 | } | |
1292 | ||
1293 | generation = btrfs_root_generation(&root->root_item); | |
1294 | level = btrfs_root_level(&root->root_item); | |
1295 | root->node = read_tree_block(fs_info, | |
1296 | btrfs_root_bytenr(&root->root_item), | |
1297 | key->objectid, generation, level, NULL); | |
1298 | if (IS_ERR(root->node)) { | |
1299 | ret = PTR_ERR(root->node); | |
1300 | root->node = NULL; | |
1301 | goto fail; | |
1302 | } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { | |
1303 | ret = -EIO; | |
1304 | goto fail; | |
1305 | } | |
1306 | root->commit_root = btrfs_root_node(root); | |
1307 | return root; | |
1308 | fail: | |
1309 | btrfs_put_root(root); | |
1310 | return ERR_PTR(ret); | |
1311 | } | |
1312 | ||
1313 | struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, | |
1314 | struct btrfs_key *key) | |
1315 | { | |
1316 | struct btrfs_root *root; | |
1317 | struct btrfs_path *path; | |
1318 | ||
1319 | path = btrfs_alloc_path(); | |
1320 | if (!path) | |
1321 | return ERR_PTR(-ENOMEM); | |
1322 | root = read_tree_root_path(tree_root, path, key); | |
1323 | btrfs_free_path(path); | |
1324 | ||
1325 | return root; | |
1326 | } | |
1327 | ||
1328 | /* | |
1329 | * Initialize subvolume root in-memory structure | |
1330 | * | |
1331 | * @anon_dev: anonymous device to attach to the root, if zero, allocate new | |
1332 | */ | |
1333 | static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) | |
1334 | { | |
1335 | int ret; | |
1336 | unsigned int nofs_flag; | |
1337 | ||
1338 | /* | |
1339 | * We might be called under a transaction (e.g. indirect backref | |
1340 | * resolution) which could deadlock if it triggers memory reclaim | |
1341 | */ | |
1342 | nofs_flag = memalloc_nofs_save(); | |
1343 | ret = btrfs_drew_lock_init(&root->snapshot_lock); | |
1344 | memalloc_nofs_restore(nofs_flag); | |
1345 | if (ret) | |
1346 | goto fail; | |
1347 | ||
1348 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && | |
1349 | root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
1350 | set_bit(BTRFS_ROOT_SHAREABLE, &root->state); | |
1351 | btrfs_check_and_init_root_item(&root->root_item); | |
1352 | } | |
1353 | ||
1354 | /* | |
1355 | * Don't assign anonymous block device to roots that are not exposed to | |
1356 | * userspace, the id pool is limited to 1M | |
1357 | */ | |
1358 | if (is_fstree(root->root_key.objectid) && | |
1359 | btrfs_root_refs(&root->root_item) > 0) { | |
1360 | if (!anon_dev) { | |
1361 | ret = get_anon_bdev(&root->anon_dev); | |
1362 | if (ret) | |
1363 | goto fail; | |
1364 | } else { | |
1365 | root->anon_dev = anon_dev; | |
1366 | } | |
1367 | } | |
1368 | ||
1369 | mutex_lock(&root->objectid_mutex); | |
1370 | ret = btrfs_find_highest_objectid(root, | |
1371 | &root->highest_objectid); | |
1372 | if (ret) { | |
1373 | mutex_unlock(&root->objectid_mutex); | |
1374 | goto fail; | |
1375 | } | |
1376 | ||
1377 | ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); | |
1378 | ||
1379 | mutex_unlock(&root->objectid_mutex); | |
1380 | ||
1381 | return 0; | |
1382 | fail: | |
1383 | /* The caller is responsible to call btrfs_free_fs_root */ | |
1384 | return ret; | |
1385 | } | |
1386 | ||
1387 | static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, | |
1388 | u64 root_id) | |
1389 | { | |
1390 | struct btrfs_root *root; | |
1391 | ||
1392 | spin_lock(&fs_info->fs_roots_radix_lock); | |
1393 | root = radix_tree_lookup(&fs_info->fs_roots_radix, | |
1394 | (unsigned long)root_id); | |
1395 | if (root) | |
1396 | root = btrfs_grab_root(root); | |
1397 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
1398 | return root; | |
1399 | } | |
1400 | ||
1401 | static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, | |
1402 | u64 objectid) | |
1403 | { | |
1404 | if (objectid == BTRFS_ROOT_TREE_OBJECTID) | |
1405 | return btrfs_grab_root(fs_info->tree_root); | |
1406 | if (objectid == BTRFS_EXTENT_TREE_OBJECTID) | |
1407 | return btrfs_grab_root(fs_info->extent_root); | |
1408 | if (objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
1409 | return btrfs_grab_root(fs_info->chunk_root); | |
1410 | if (objectid == BTRFS_DEV_TREE_OBJECTID) | |
1411 | return btrfs_grab_root(fs_info->dev_root); | |
1412 | if (objectid == BTRFS_CSUM_TREE_OBJECTID) | |
1413 | return btrfs_grab_root(fs_info->csum_root); | |
1414 | if (objectid == BTRFS_QUOTA_TREE_OBJECTID) | |
1415 | return btrfs_grab_root(fs_info->quota_root) ? | |
1416 | fs_info->quota_root : ERR_PTR(-ENOENT); | |
1417 | if (objectid == BTRFS_UUID_TREE_OBJECTID) | |
1418 | return btrfs_grab_root(fs_info->uuid_root) ? | |
1419 | fs_info->uuid_root : ERR_PTR(-ENOENT); | |
1420 | if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) | |
1421 | return btrfs_grab_root(fs_info->free_space_root) ? | |
1422 | fs_info->free_space_root : ERR_PTR(-ENOENT); | |
1423 | return NULL; | |
1424 | } | |
1425 | ||
1426 | int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, | |
1427 | struct btrfs_root *root) | |
1428 | { | |
1429 | int ret; | |
1430 | ||
1431 | ret = radix_tree_preload(GFP_NOFS); | |
1432 | if (ret) | |
1433 | return ret; | |
1434 | ||
1435 | spin_lock(&fs_info->fs_roots_radix_lock); | |
1436 | ret = radix_tree_insert(&fs_info->fs_roots_radix, | |
1437 | (unsigned long)root->root_key.objectid, | |
1438 | root); | |
1439 | if (ret == 0) { | |
1440 | btrfs_grab_root(root); | |
1441 | set_bit(BTRFS_ROOT_IN_RADIX, &root->state); | |
1442 | } | |
1443 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
1444 | radix_tree_preload_end(); | |
1445 | ||
1446 | return ret; | |
1447 | } | |
1448 | ||
1449 | void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) | |
1450 | { | |
1451 | #ifdef CONFIG_BTRFS_DEBUG | |
1452 | struct btrfs_root *root; | |
1453 | ||
1454 | while (!list_empty(&fs_info->allocated_roots)) { | |
1455 | char buf[BTRFS_ROOT_NAME_BUF_LEN]; | |
1456 | ||
1457 | root = list_first_entry(&fs_info->allocated_roots, | |
1458 | struct btrfs_root, leak_list); | |
1459 | btrfs_err(fs_info, "leaked root %s refcount %d", | |
1460 | btrfs_root_name(&root->root_key, buf), | |
1461 | refcount_read(&root->refs)); | |
1462 | while (refcount_read(&root->refs) > 1) | |
1463 | btrfs_put_root(root); | |
1464 | btrfs_put_root(root); | |
1465 | } | |
1466 | #endif | |
1467 | } | |
1468 | ||
1469 | void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) | |
1470 | { | |
1471 | percpu_counter_destroy(&fs_info->dirty_metadata_bytes); | |
1472 | percpu_counter_destroy(&fs_info->delalloc_bytes); | |
1473 | percpu_counter_destroy(&fs_info->dio_bytes); | |
1474 | percpu_counter_destroy(&fs_info->dev_replace.bio_counter); | |
1475 | btrfs_free_csum_hash(fs_info); | |
1476 | btrfs_free_stripe_hash_table(fs_info); | |
1477 | btrfs_free_ref_cache(fs_info); | |
1478 | kfree(fs_info->balance_ctl); | |
1479 | kfree(fs_info->delayed_root); | |
1480 | btrfs_put_root(fs_info->extent_root); | |
1481 | btrfs_put_root(fs_info->tree_root); | |
1482 | btrfs_put_root(fs_info->chunk_root); | |
1483 | btrfs_put_root(fs_info->dev_root); | |
1484 | btrfs_put_root(fs_info->csum_root); | |
1485 | btrfs_put_root(fs_info->quota_root); | |
1486 | btrfs_put_root(fs_info->uuid_root); | |
1487 | btrfs_put_root(fs_info->free_space_root); | |
1488 | btrfs_put_root(fs_info->fs_root); | |
1489 | btrfs_put_root(fs_info->data_reloc_root); | |
1490 | btrfs_check_leaked_roots(fs_info); | |
1491 | btrfs_extent_buffer_leak_debug_check(fs_info); | |
1492 | kfree(fs_info->super_copy); | |
1493 | kfree(fs_info->super_for_commit); | |
1494 | kvfree(fs_info); | |
1495 | } | |
1496 | ||
1497 | ||
1498 | /* | |
1499 | * Get an in-memory reference of a root structure. | |
1500 | * | |
1501 | * For essential trees like root/extent tree, we grab it from fs_info directly. | |
1502 | * For subvolume trees, we check the cached filesystem roots first. If not | |
1503 | * found, then read it from disk and add it to cached fs roots. | |
1504 | * | |
1505 | * Caller should release the root by calling btrfs_put_root() after the usage. | |
1506 | * | |
1507 | * NOTE: Reloc and log trees can't be read by this function as they share the | |
1508 | * same root objectid. | |
1509 | * | |
1510 | * @objectid: root id | |
1511 | * @anon_dev: preallocated anonymous block device number for new roots, | |
1512 | * pass 0 for new allocation. | |
1513 | * @check_ref: whether to check root item references, If true, return -ENOENT | |
1514 | * for orphan roots | |
1515 | */ | |
1516 | static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, | |
1517 | u64 objectid, dev_t anon_dev, | |
1518 | bool check_ref) | |
1519 | { | |
1520 | struct btrfs_root *root; | |
1521 | struct btrfs_path *path; | |
1522 | struct btrfs_key key; | |
1523 | int ret; | |
1524 | ||
1525 | root = btrfs_get_global_root(fs_info, objectid); | |
1526 | if (root) | |
1527 | return root; | |
1528 | again: | |
1529 | root = btrfs_lookup_fs_root(fs_info, objectid); | |
1530 | if (root) { | |
1531 | /* Shouldn't get preallocated anon_dev for cached roots */ | |
1532 | ASSERT(!anon_dev); | |
1533 | if (check_ref && btrfs_root_refs(&root->root_item) == 0) { | |
1534 | btrfs_put_root(root); | |
1535 | return ERR_PTR(-ENOENT); | |
1536 | } | |
1537 | return root; | |
1538 | } | |
1539 | ||
1540 | key.objectid = objectid; | |
1541 | key.type = BTRFS_ROOT_ITEM_KEY; | |
1542 | key.offset = (u64)-1; | |
1543 | root = btrfs_read_tree_root(fs_info->tree_root, &key); | |
1544 | if (IS_ERR(root)) | |
1545 | return root; | |
1546 | ||
1547 | if (check_ref && btrfs_root_refs(&root->root_item) == 0) { | |
1548 | ret = -ENOENT; | |
1549 | goto fail; | |
1550 | } | |
1551 | ||
1552 | ret = btrfs_init_fs_root(root, anon_dev); | |
1553 | if (ret) | |
1554 | goto fail; | |
1555 | ||
1556 | path = btrfs_alloc_path(); | |
1557 | if (!path) { | |
1558 | ret = -ENOMEM; | |
1559 | goto fail; | |
1560 | } | |
1561 | key.objectid = BTRFS_ORPHAN_OBJECTID; | |
1562 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
1563 | key.offset = objectid; | |
1564 | ||
1565 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
1566 | btrfs_free_path(path); | |
1567 | if (ret < 0) | |
1568 | goto fail; | |
1569 | if (ret == 0) | |
1570 | set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); | |
1571 | ||
1572 | ret = btrfs_insert_fs_root(fs_info, root); | |
1573 | if (ret) { | |
1574 | btrfs_put_root(root); | |
1575 | if (ret == -EEXIST) | |
1576 | goto again; | |
1577 | goto fail; | |
1578 | } | |
1579 | return root; | |
1580 | fail: | |
1581 | btrfs_put_root(root); | |
1582 | return ERR_PTR(ret); | |
1583 | } | |
1584 | ||
1585 | /* | |
1586 | * Get in-memory reference of a root structure | |
1587 | * | |
1588 | * @objectid: tree objectid | |
1589 | * @check_ref: if set, verify that the tree exists and the item has at least | |
1590 | * one reference | |
1591 | */ | |
1592 | struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, | |
1593 | u64 objectid, bool check_ref) | |
1594 | { | |
1595 | return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); | |
1596 | } | |
1597 | ||
1598 | /* | |
1599 | * Get in-memory reference of a root structure, created as new, optionally pass | |
1600 | * the anonymous block device id | |
1601 | * | |
1602 | * @objectid: tree objectid | |
1603 | * @anon_dev: if zero, allocate a new anonymous block device or use the | |
1604 | * parameter value | |
1605 | */ | |
1606 | struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, | |
1607 | u64 objectid, dev_t anon_dev) | |
1608 | { | |
1609 | return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); | |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * btrfs_get_fs_root_commit_root - return a root for the given objectid | |
1614 | * @fs_info: the fs_info | |
1615 | * @objectid: the objectid we need to lookup | |
1616 | * | |
1617 | * This is exclusively used for backref walking, and exists specifically because | |
1618 | * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref | |
1619 | * creation time, which means we may have to read the tree_root in order to look | |
1620 | * up a fs root that is not in memory. If the root is not in memory we will | |
1621 | * read the tree root commit root and look up the fs root from there. This is a | |
1622 | * temporary root, it will not be inserted into the radix tree as it doesn't | |
1623 | * have the most uptodate information, it'll simply be discarded once the | |
1624 | * backref code is finished using the root. | |
1625 | */ | |
1626 | struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, | |
1627 | struct btrfs_path *path, | |
1628 | u64 objectid) | |
1629 | { | |
1630 | struct btrfs_root *root; | |
1631 | struct btrfs_key key; | |
1632 | ||
1633 | ASSERT(path->search_commit_root && path->skip_locking); | |
1634 | ||
1635 | /* | |
1636 | * This can return -ENOENT if we ask for a root that doesn't exist, but | |
1637 | * since this is called via the backref walking code we won't be looking | |
1638 | * up a root that doesn't exist, unless there's corruption. So if root | |
1639 | * != NULL just return it. | |
1640 | */ | |
1641 | root = btrfs_get_global_root(fs_info, objectid); | |
1642 | if (root) | |
1643 | return root; | |
1644 | ||
1645 | root = btrfs_lookup_fs_root(fs_info, objectid); | |
1646 | if (root) | |
1647 | return root; | |
1648 | ||
1649 | key.objectid = objectid; | |
1650 | key.type = BTRFS_ROOT_ITEM_KEY; | |
1651 | key.offset = (u64)-1; | |
1652 | root = read_tree_root_path(fs_info->tree_root, path, &key); | |
1653 | btrfs_release_path(path); | |
1654 | ||
1655 | return root; | |
1656 | } | |
1657 | ||
1658 | /* | |
1659 | * called by the kthread helper functions to finally call the bio end_io | |
1660 | * functions. This is where read checksum verification actually happens | |
1661 | */ | |
1662 | static void end_workqueue_fn(struct btrfs_work *work) | |
1663 | { | |
1664 | struct bio *bio; | |
1665 | struct btrfs_end_io_wq *end_io_wq; | |
1666 | ||
1667 | end_io_wq = container_of(work, struct btrfs_end_io_wq, work); | |
1668 | bio = end_io_wq->bio; | |
1669 | ||
1670 | bio->bi_status = end_io_wq->status; | |
1671 | bio->bi_private = end_io_wq->private; | |
1672 | bio->bi_end_io = end_io_wq->end_io; | |
1673 | bio_endio(bio); | |
1674 | kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); | |
1675 | } | |
1676 | ||
1677 | static int cleaner_kthread(void *arg) | |
1678 | { | |
1679 | struct btrfs_root *root = arg; | |
1680 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1681 | int again; | |
1682 | ||
1683 | while (1) { | |
1684 | again = 0; | |
1685 | ||
1686 | set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); | |
1687 | ||
1688 | /* Make the cleaner go to sleep early. */ | |
1689 | if (btrfs_need_cleaner_sleep(fs_info)) | |
1690 | goto sleep; | |
1691 | ||
1692 | /* | |
1693 | * Do not do anything if we might cause open_ctree() to block | |
1694 | * before we have finished mounting the filesystem. | |
1695 | */ | |
1696 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) | |
1697 | goto sleep; | |
1698 | ||
1699 | if (!mutex_trylock(&fs_info->cleaner_mutex)) | |
1700 | goto sleep; | |
1701 | ||
1702 | /* | |
1703 | * Avoid the problem that we change the status of the fs | |
1704 | * during the above check and trylock. | |
1705 | */ | |
1706 | if (btrfs_need_cleaner_sleep(fs_info)) { | |
1707 | mutex_unlock(&fs_info->cleaner_mutex); | |
1708 | goto sleep; | |
1709 | } | |
1710 | ||
1711 | btrfs_run_delayed_iputs(fs_info); | |
1712 | ||
1713 | again = btrfs_clean_one_deleted_snapshot(root); | |
1714 | mutex_unlock(&fs_info->cleaner_mutex); | |
1715 | ||
1716 | /* | |
1717 | * The defragger has dealt with the R/O remount and umount, | |
1718 | * needn't do anything special here. | |
1719 | */ | |
1720 | btrfs_run_defrag_inodes(fs_info); | |
1721 | ||
1722 | /* | |
1723 | * Acquires fs_info->delete_unused_bgs_mutex to avoid racing | |
1724 | * with relocation (btrfs_relocate_chunk) and relocation | |
1725 | * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) | |
1726 | * after acquiring fs_info->delete_unused_bgs_mutex. So we | |
1727 | * can't hold, nor need to, fs_info->cleaner_mutex when deleting | |
1728 | * unused block groups. | |
1729 | */ | |
1730 | btrfs_delete_unused_bgs(fs_info); | |
1731 | sleep: | |
1732 | clear_and_wake_up_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); | |
1733 | if (kthread_should_park()) | |
1734 | kthread_parkme(); | |
1735 | if (kthread_should_stop()) | |
1736 | return 0; | |
1737 | if (!again) { | |
1738 | set_current_state(TASK_INTERRUPTIBLE); | |
1739 | schedule(); | |
1740 | __set_current_state(TASK_RUNNING); | |
1741 | } | |
1742 | } | |
1743 | } | |
1744 | ||
1745 | static int transaction_kthread(void *arg) | |
1746 | { | |
1747 | struct btrfs_root *root = arg; | |
1748 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1749 | struct btrfs_trans_handle *trans; | |
1750 | struct btrfs_transaction *cur; | |
1751 | u64 transid; | |
1752 | time64_t delta; | |
1753 | unsigned long delay; | |
1754 | bool cannot_commit; | |
1755 | ||
1756 | do { | |
1757 | cannot_commit = false; | |
1758 | delay = msecs_to_jiffies(fs_info->commit_interval * 1000); | |
1759 | mutex_lock(&fs_info->transaction_kthread_mutex); | |
1760 | ||
1761 | spin_lock(&fs_info->trans_lock); | |
1762 | cur = fs_info->running_transaction; | |
1763 | if (!cur) { | |
1764 | spin_unlock(&fs_info->trans_lock); | |
1765 | goto sleep; | |
1766 | } | |
1767 | ||
1768 | delta = ktime_get_seconds() - cur->start_time; | |
1769 | if (cur->state < TRANS_STATE_COMMIT_START && | |
1770 | delta < fs_info->commit_interval) { | |
1771 | spin_unlock(&fs_info->trans_lock); | |
1772 | delay -= msecs_to_jiffies((delta - 1) * 1000); | |
1773 | delay = min(delay, | |
1774 | msecs_to_jiffies(fs_info->commit_interval * 1000)); | |
1775 | goto sleep; | |
1776 | } | |
1777 | transid = cur->transid; | |
1778 | spin_unlock(&fs_info->trans_lock); | |
1779 | ||
1780 | /* If the file system is aborted, this will always fail. */ | |
1781 | trans = btrfs_attach_transaction(root); | |
1782 | if (IS_ERR(trans)) { | |
1783 | if (PTR_ERR(trans) != -ENOENT) | |
1784 | cannot_commit = true; | |
1785 | goto sleep; | |
1786 | } | |
1787 | if (transid == trans->transid) { | |
1788 | btrfs_commit_transaction(trans); | |
1789 | } else { | |
1790 | btrfs_end_transaction(trans); | |
1791 | } | |
1792 | sleep: | |
1793 | wake_up_process(fs_info->cleaner_kthread); | |
1794 | mutex_unlock(&fs_info->transaction_kthread_mutex); | |
1795 | ||
1796 | if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, | |
1797 | &fs_info->fs_state))) | |
1798 | btrfs_cleanup_transaction(fs_info); | |
1799 | if (!kthread_should_stop() && | |
1800 | (!btrfs_transaction_blocked(fs_info) || | |
1801 | cannot_commit)) | |
1802 | schedule_timeout_interruptible(delay); | |
1803 | } while (!kthread_should_stop()); | |
1804 | return 0; | |
1805 | } | |
1806 | ||
1807 | /* | |
1808 | * This will find the highest generation in the array of root backups. The | |
1809 | * index of the highest array is returned, or -EINVAL if we can't find | |
1810 | * anything. | |
1811 | * | |
1812 | * We check to make sure the array is valid by comparing the | |
1813 | * generation of the latest root in the array with the generation | |
1814 | * in the super block. If they don't match we pitch it. | |
1815 | */ | |
1816 | static int find_newest_super_backup(struct btrfs_fs_info *info) | |
1817 | { | |
1818 | const u64 newest_gen = btrfs_super_generation(info->super_copy); | |
1819 | u64 cur; | |
1820 | struct btrfs_root_backup *root_backup; | |
1821 | int i; | |
1822 | ||
1823 | for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { | |
1824 | root_backup = info->super_copy->super_roots + i; | |
1825 | cur = btrfs_backup_tree_root_gen(root_backup); | |
1826 | if (cur == newest_gen) | |
1827 | return i; | |
1828 | } | |
1829 | ||
1830 | return -EINVAL; | |
1831 | } | |
1832 | ||
1833 | /* | |
1834 | * copy all the root pointers into the super backup array. | |
1835 | * this will bump the backup pointer by one when it is | |
1836 | * done | |
1837 | */ | |
1838 | static void backup_super_roots(struct btrfs_fs_info *info) | |
1839 | { | |
1840 | const int next_backup = info->backup_root_index; | |
1841 | struct btrfs_root_backup *root_backup; | |
1842 | ||
1843 | root_backup = info->super_for_commit->super_roots + next_backup; | |
1844 | ||
1845 | /* | |
1846 | * make sure all of our padding and empty slots get zero filled | |
1847 | * regardless of which ones we use today | |
1848 | */ | |
1849 | memset(root_backup, 0, sizeof(*root_backup)); | |
1850 | ||
1851 | info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; | |
1852 | ||
1853 | btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); | |
1854 | btrfs_set_backup_tree_root_gen(root_backup, | |
1855 | btrfs_header_generation(info->tree_root->node)); | |
1856 | ||
1857 | btrfs_set_backup_tree_root_level(root_backup, | |
1858 | btrfs_header_level(info->tree_root->node)); | |
1859 | ||
1860 | btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); | |
1861 | btrfs_set_backup_chunk_root_gen(root_backup, | |
1862 | btrfs_header_generation(info->chunk_root->node)); | |
1863 | btrfs_set_backup_chunk_root_level(root_backup, | |
1864 | btrfs_header_level(info->chunk_root->node)); | |
1865 | ||
1866 | btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); | |
1867 | btrfs_set_backup_extent_root_gen(root_backup, | |
1868 | btrfs_header_generation(info->extent_root->node)); | |
1869 | btrfs_set_backup_extent_root_level(root_backup, | |
1870 | btrfs_header_level(info->extent_root->node)); | |
1871 | ||
1872 | /* | |
1873 | * we might commit during log recovery, which happens before we set | |
1874 | * the fs_root. Make sure it is valid before we fill it in. | |
1875 | */ | |
1876 | if (info->fs_root && info->fs_root->node) { | |
1877 | btrfs_set_backup_fs_root(root_backup, | |
1878 | info->fs_root->node->start); | |
1879 | btrfs_set_backup_fs_root_gen(root_backup, | |
1880 | btrfs_header_generation(info->fs_root->node)); | |
1881 | btrfs_set_backup_fs_root_level(root_backup, | |
1882 | btrfs_header_level(info->fs_root->node)); | |
1883 | } | |
1884 | ||
1885 | btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); | |
1886 | btrfs_set_backup_dev_root_gen(root_backup, | |
1887 | btrfs_header_generation(info->dev_root->node)); | |
1888 | btrfs_set_backup_dev_root_level(root_backup, | |
1889 | btrfs_header_level(info->dev_root->node)); | |
1890 | ||
1891 | btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); | |
1892 | btrfs_set_backup_csum_root_gen(root_backup, | |
1893 | btrfs_header_generation(info->csum_root->node)); | |
1894 | btrfs_set_backup_csum_root_level(root_backup, | |
1895 | btrfs_header_level(info->csum_root->node)); | |
1896 | ||
1897 | btrfs_set_backup_total_bytes(root_backup, | |
1898 | btrfs_super_total_bytes(info->super_copy)); | |
1899 | btrfs_set_backup_bytes_used(root_backup, | |
1900 | btrfs_super_bytes_used(info->super_copy)); | |
1901 | btrfs_set_backup_num_devices(root_backup, | |
1902 | btrfs_super_num_devices(info->super_copy)); | |
1903 | ||
1904 | /* | |
1905 | * if we don't copy this out to the super_copy, it won't get remembered | |
1906 | * for the next commit | |
1907 | */ | |
1908 | memcpy(&info->super_copy->super_roots, | |
1909 | &info->super_for_commit->super_roots, | |
1910 | sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); | |
1911 | } | |
1912 | ||
1913 | /* | |
1914 | * read_backup_root - Reads a backup root based on the passed priority. Prio 0 | |
1915 | * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots | |
1916 | * | |
1917 | * fs_info - filesystem whose backup roots need to be read | |
1918 | * priority - priority of backup root required | |
1919 | * | |
1920 | * Returns backup root index on success and -EINVAL otherwise. | |
1921 | */ | |
1922 | static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) | |
1923 | { | |
1924 | int backup_index = find_newest_super_backup(fs_info); | |
1925 | struct btrfs_super_block *super = fs_info->super_copy; | |
1926 | struct btrfs_root_backup *root_backup; | |
1927 | ||
1928 | if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { | |
1929 | if (priority == 0) | |
1930 | return backup_index; | |
1931 | ||
1932 | backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; | |
1933 | backup_index %= BTRFS_NUM_BACKUP_ROOTS; | |
1934 | } else { | |
1935 | return -EINVAL; | |
1936 | } | |
1937 | ||
1938 | root_backup = super->super_roots + backup_index; | |
1939 | ||
1940 | btrfs_set_super_generation(super, | |
1941 | btrfs_backup_tree_root_gen(root_backup)); | |
1942 | btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); | |
1943 | btrfs_set_super_root_level(super, | |
1944 | btrfs_backup_tree_root_level(root_backup)); | |
1945 | btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); | |
1946 | ||
1947 | /* | |
1948 | * Fixme: the total bytes and num_devices need to match or we should | |
1949 | * need a fsck | |
1950 | */ | |
1951 | btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); | |
1952 | btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); | |
1953 | ||
1954 | return backup_index; | |
1955 | } | |
1956 | ||
1957 | /* helper to cleanup workers */ | |
1958 | static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) | |
1959 | { | |
1960 | btrfs_destroy_workqueue(fs_info->fixup_workers); | |
1961 | btrfs_destroy_workqueue(fs_info->delalloc_workers); | |
1962 | btrfs_destroy_workqueue(fs_info->workers); | |
1963 | btrfs_destroy_workqueue(fs_info->endio_workers); | |
1964 | btrfs_destroy_workqueue(fs_info->endio_raid56_workers); | |
1965 | btrfs_destroy_workqueue(fs_info->rmw_workers); | |
1966 | btrfs_destroy_workqueue(fs_info->endio_write_workers); | |
1967 | btrfs_destroy_workqueue(fs_info->endio_freespace_worker); | |
1968 | btrfs_destroy_workqueue(fs_info->delayed_workers); | |
1969 | btrfs_destroy_workqueue(fs_info->caching_workers); | |
1970 | btrfs_destroy_workqueue(fs_info->readahead_workers); | |
1971 | btrfs_destroy_workqueue(fs_info->flush_workers); | |
1972 | btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); | |
1973 | if (fs_info->discard_ctl.discard_workers) | |
1974 | destroy_workqueue(fs_info->discard_ctl.discard_workers); | |
1975 | /* | |
1976 | * Now that all other work queues are destroyed, we can safely destroy | |
1977 | * the queues used for metadata I/O, since tasks from those other work | |
1978 | * queues can do metadata I/O operations. | |
1979 | */ | |
1980 | btrfs_destroy_workqueue(fs_info->endio_meta_workers); | |
1981 | btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); | |
1982 | } | |
1983 | ||
1984 | static void free_root_extent_buffers(struct btrfs_root *root) | |
1985 | { | |
1986 | if (root) { | |
1987 | free_extent_buffer(root->node); | |
1988 | free_extent_buffer(root->commit_root); | |
1989 | root->node = NULL; | |
1990 | root->commit_root = NULL; | |
1991 | } | |
1992 | } | |
1993 | ||
1994 | /* helper to cleanup tree roots */ | |
1995 | static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) | |
1996 | { | |
1997 | free_root_extent_buffers(info->tree_root); | |
1998 | ||
1999 | free_root_extent_buffers(info->dev_root); | |
2000 | free_root_extent_buffers(info->extent_root); | |
2001 | free_root_extent_buffers(info->csum_root); | |
2002 | free_root_extent_buffers(info->quota_root); | |
2003 | free_root_extent_buffers(info->uuid_root); | |
2004 | free_root_extent_buffers(info->fs_root); | |
2005 | free_root_extent_buffers(info->data_reloc_root); | |
2006 | if (free_chunk_root) | |
2007 | free_root_extent_buffers(info->chunk_root); | |
2008 | free_root_extent_buffers(info->free_space_root); | |
2009 | } | |
2010 | ||
2011 | void btrfs_put_root(struct btrfs_root *root) | |
2012 | { | |
2013 | if (!root) | |
2014 | return; | |
2015 | ||
2016 | if (refcount_dec_and_test(&root->refs)) { | |
2017 | WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); | |
2018 | WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); | |
2019 | if (root->anon_dev) | |
2020 | free_anon_bdev(root->anon_dev); | |
2021 | btrfs_drew_lock_destroy(&root->snapshot_lock); | |
2022 | free_root_extent_buffers(root); | |
2023 | #ifdef CONFIG_BTRFS_DEBUG | |
2024 | spin_lock(&root->fs_info->fs_roots_radix_lock); | |
2025 | list_del_init(&root->leak_list); | |
2026 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
2027 | #endif | |
2028 | kfree(root); | |
2029 | } | |
2030 | } | |
2031 | ||
2032 | void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) | |
2033 | { | |
2034 | int ret; | |
2035 | struct btrfs_root *gang[8]; | |
2036 | int i; | |
2037 | ||
2038 | while (!list_empty(&fs_info->dead_roots)) { | |
2039 | gang[0] = list_entry(fs_info->dead_roots.next, | |
2040 | struct btrfs_root, root_list); | |
2041 | list_del(&gang[0]->root_list); | |
2042 | ||
2043 | if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) | |
2044 | btrfs_drop_and_free_fs_root(fs_info, gang[0]); | |
2045 | btrfs_put_root(gang[0]); | |
2046 | } | |
2047 | ||
2048 | while (1) { | |
2049 | ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, | |
2050 | (void **)gang, 0, | |
2051 | ARRAY_SIZE(gang)); | |
2052 | if (!ret) | |
2053 | break; | |
2054 | for (i = 0; i < ret; i++) | |
2055 | btrfs_drop_and_free_fs_root(fs_info, gang[i]); | |
2056 | } | |
2057 | } | |
2058 | ||
2059 | static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) | |
2060 | { | |
2061 | mutex_init(&fs_info->scrub_lock); | |
2062 | atomic_set(&fs_info->scrubs_running, 0); | |
2063 | atomic_set(&fs_info->scrub_pause_req, 0); | |
2064 | atomic_set(&fs_info->scrubs_paused, 0); | |
2065 | atomic_set(&fs_info->scrub_cancel_req, 0); | |
2066 | init_waitqueue_head(&fs_info->scrub_pause_wait); | |
2067 | refcount_set(&fs_info->scrub_workers_refcnt, 0); | |
2068 | } | |
2069 | ||
2070 | static void btrfs_init_balance(struct btrfs_fs_info *fs_info) | |
2071 | { | |
2072 | spin_lock_init(&fs_info->balance_lock); | |
2073 | mutex_init(&fs_info->balance_mutex); | |
2074 | atomic_set(&fs_info->balance_pause_req, 0); | |
2075 | atomic_set(&fs_info->balance_cancel_req, 0); | |
2076 | fs_info->balance_ctl = NULL; | |
2077 | init_waitqueue_head(&fs_info->balance_wait_q); | |
2078 | } | |
2079 | ||
2080 | static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) | |
2081 | { | |
2082 | struct inode *inode = fs_info->btree_inode; | |
2083 | ||
2084 | inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; | |
2085 | set_nlink(inode, 1); | |
2086 | /* | |
2087 | * we set the i_size on the btree inode to the max possible int. | |
2088 | * the real end of the address space is determined by all of | |
2089 | * the devices in the system | |
2090 | */ | |
2091 | inode->i_size = OFFSET_MAX; | |
2092 | inode->i_mapping->a_ops = &btree_aops; | |
2093 | ||
2094 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); | |
2095 | extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, | |
2096 | IO_TREE_BTREE_INODE_IO, inode); | |
2097 | BTRFS_I(inode)->io_tree.track_uptodate = false; | |
2098 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree); | |
2099 | ||
2100 | BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); | |
2101 | memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); | |
2102 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); | |
2103 | btrfs_insert_inode_hash(inode); | |
2104 | } | |
2105 | ||
2106 | static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) | |
2107 | { | |
2108 | mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); | |
2109 | init_rwsem(&fs_info->dev_replace.rwsem); | |
2110 | init_waitqueue_head(&fs_info->dev_replace.replace_wait); | |
2111 | } | |
2112 | ||
2113 | static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) | |
2114 | { | |
2115 | spin_lock_init(&fs_info->qgroup_lock); | |
2116 | mutex_init(&fs_info->qgroup_ioctl_lock); | |
2117 | fs_info->qgroup_tree = RB_ROOT; | |
2118 | INIT_LIST_HEAD(&fs_info->dirty_qgroups); | |
2119 | fs_info->qgroup_seq = 1; | |
2120 | fs_info->qgroup_ulist = NULL; | |
2121 | fs_info->qgroup_rescan_running = false; | |
2122 | mutex_init(&fs_info->qgroup_rescan_lock); | |
2123 | } | |
2124 | ||
2125 | static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, | |
2126 | struct btrfs_fs_devices *fs_devices) | |
2127 | { | |
2128 | u32 max_active = fs_info->thread_pool_size; | |
2129 | unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; | |
2130 | ||
2131 | fs_info->workers = | |
2132 | btrfs_alloc_workqueue(fs_info, "worker", | |
2133 | flags | WQ_HIGHPRI, max_active, 16); | |
2134 | ||
2135 | fs_info->delalloc_workers = | |
2136 | btrfs_alloc_workqueue(fs_info, "delalloc", | |
2137 | flags, max_active, 2); | |
2138 | ||
2139 | fs_info->flush_workers = | |
2140 | btrfs_alloc_workqueue(fs_info, "flush_delalloc", | |
2141 | flags, max_active, 0); | |
2142 | ||
2143 | fs_info->caching_workers = | |
2144 | btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); | |
2145 | ||
2146 | fs_info->fixup_workers = | |
2147 | btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); | |
2148 | ||
2149 | /* | |
2150 | * endios are largely parallel and should have a very | |
2151 | * low idle thresh | |
2152 | */ | |
2153 | fs_info->endio_workers = | |
2154 | btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); | |
2155 | fs_info->endio_meta_workers = | |
2156 | btrfs_alloc_workqueue(fs_info, "endio-meta", flags, | |
2157 | max_active, 4); | |
2158 | fs_info->endio_meta_write_workers = | |
2159 | btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, | |
2160 | max_active, 2); | |
2161 | fs_info->endio_raid56_workers = | |
2162 | btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, | |
2163 | max_active, 4); | |
2164 | fs_info->rmw_workers = | |
2165 | btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); | |
2166 | fs_info->endio_write_workers = | |
2167 | btrfs_alloc_workqueue(fs_info, "endio-write", flags, | |
2168 | max_active, 2); | |
2169 | fs_info->endio_freespace_worker = | |
2170 | btrfs_alloc_workqueue(fs_info, "freespace-write", flags, | |
2171 | max_active, 0); | |
2172 | fs_info->delayed_workers = | |
2173 | btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, | |
2174 | max_active, 0); | |
2175 | fs_info->readahead_workers = | |
2176 | btrfs_alloc_workqueue(fs_info, "readahead", flags, | |
2177 | max_active, 2); | |
2178 | fs_info->qgroup_rescan_workers = | |
2179 | btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); | |
2180 | fs_info->discard_ctl.discard_workers = | |
2181 | alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); | |
2182 | ||
2183 | if (!(fs_info->workers && fs_info->delalloc_workers && | |
2184 | fs_info->flush_workers && | |
2185 | fs_info->endio_workers && fs_info->endio_meta_workers && | |
2186 | fs_info->endio_meta_write_workers && | |
2187 | fs_info->endio_write_workers && fs_info->endio_raid56_workers && | |
2188 | fs_info->endio_freespace_worker && fs_info->rmw_workers && | |
2189 | fs_info->caching_workers && fs_info->readahead_workers && | |
2190 | fs_info->fixup_workers && fs_info->delayed_workers && | |
2191 | fs_info->qgroup_rescan_workers && | |
2192 | fs_info->discard_ctl.discard_workers)) { | |
2193 | return -ENOMEM; | |
2194 | } | |
2195 | ||
2196 | return 0; | |
2197 | } | |
2198 | ||
2199 | static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) | |
2200 | { | |
2201 | struct crypto_shash *csum_shash; | |
2202 | const char *csum_driver = btrfs_super_csum_driver(csum_type); | |
2203 | ||
2204 | csum_shash = crypto_alloc_shash(csum_driver, 0, 0); | |
2205 | ||
2206 | if (IS_ERR(csum_shash)) { | |
2207 | btrfs_err(fs_info, "error allocating %s hash for checksum", | |
2208 | csum_driver); | |
2209 | return PTR_ERR(csum_shash); | |
2210 | } | |
2211 | ||
2212 | fs_info->csum_shash = csum_shash; | |
2213 | ||
2214 | return 0; | |
2215 | } | |
2216 | ||
2217 | static int btrfs_replay_log(struct btrfs_fs_info *fs_info, | |
2218 | struct btrfs_fs_devices *fs_devices) | |
2219 | { | |
2220 | int ret; | |
2221 | struct btrfs_root *log_tree_root; | |
2222 | struct btrfs_super_block *disk_super = fs_info->super_copy; | |
2223 | u64 bytenr = btrfs_super_log_root(disk_super); | |
2224 | int level = btrfs_super_log_root_level(disk_super); | |
2225 | ||
2226 | if (fs_devices->rw_devices == 0) { | |
2227 | btrfs_warn(fs_info, "log replay required on RO media"); | |
2228 | return -EIO; | |
2229 | } | |
2230 | ||
2231 | log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, | |
2232 | GFP_KERNEL); | |
2233 | if (!log_tree_root) | |
2234 | return -ENOMEM; | |
2235 | ||
2236 | log_tree_root->node = read_tree_block(fs_info, bytenr, | |
2237 | BTRFS_TREE_LOG_OBJECTID, | |
2238 | fs_info->generation + 1, level, | |
2239 | NULL); | |
2240 | if (IS_ERR(log_tree_root->node)) { | |
2241 | btrfs_warn(fs_info, "failed to read log tree"); | |
2242 | ret = PTR_ERR(log_tree_root->node); | |
2243 | log_tree_root->node = NULL; | |
2244 | btrfs_put_root(log_tree_root); | |
2245 | return ret; | |
2246 | } else if (!extent_buffer_uptodate(log_tree_root->node)) { | |
2247 | btrfs_err(fs_info, "failed to read log tree"); | |
2248 | btrfs_put_root(log_tree_root); | |
2249 | return -EIO; | |
2250 | } | |
2251 | /* returns with log_tree_root freed on success */ | |
2252 | ret = btrfs_recover_log_trees(log_tree_root); | |
2253 | if (ret) { | |
2254 | btrfs_handle_fs_error(fs_info, ret, | |
2255 | "Failed to recover log tree"); | |
2256 | btrfs_put_root(log_tree_root); | |
2257 | return ret; | |
2258 | } | |
2259 | ||
2260 | if (sb_rdonly(fs_info->sb)) { | |
2261 | ret = btrfs_commit_super(fs_info); | |
2262 | if (ret) | |
2263 | return ret; | |
2264 | } | |
2265 | ||
2266 | return 0; | |
2267 | } | |
2268 | ||
2269 | static int btrfs_read_roots(struct btrfs_fs_info *fs_info) | |
2270 | { | |
2271 | struct btrfs_root *tree_root = fs_info->tree_root; | |
2272 | struct btrfs_root *root; | |
2273 | struct btrfs_key location; | |
2274 | int ret; | |
2275 | ||
2276 | BUG_ON(!fs_info->tree_root); | |
2277 | ||
2278 | location.objectid = BTRFS_EXTENT_TREE_OBJECTID; | |
2279 | location.type = BTRFS_ROOT_ITEM_KEY; | |
2280 | location.offset = 0; | |
2281 | ||
2282 | root = btrfs_read_tree_root(tree_root, &location); | |
2283 | if (IS_ERR(root)) { | |
2284 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { | |
2285 | ret = PTR_ERR(root); | |
2286 | goto out; | |
2287 | } | |
2288 | } else { | |
2289 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2290 | fs_info->extent_root = root; | |
2291 | } | |
2292 | ||
2293 | location.objectid = BTRFS_DEV_TREE_OBJECTID; | |
2294 | root = btrfs_read_tree_root(tree_root, &location); | |
2295 | if (IS_ERR(root)) { | |
2296 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { | |
2297 | ret = PTR_ERR(root); | |
2298 | goto out; | |
2299 | } | |
2300 | } else { | |
2301 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2302 | fs_info->dev_root = root; | |
2303 | btrfs_init_devices_late(fs_info); | |
2304 | } | |
2305 | ||
2306 | /* If IGNOREDATACSUMS is set don't bother reading the csum root. */ | |
2307 | if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { | |
2308 | location.objectid = BTRFS_CSUM_TREE_OBJECTID; | |
2309 | root = btrfs_read_tree_root(tree_root, &location); | |
2310 | if (IS_ERR(root)) { | |
2311 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { | |
2312 | ret = PTR_ERR(root); | |
2313 | goto out; | |
2314 | } | |
2315 | } else { | |
2316 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2317 | fs_info->csum_root = root; | |
2318 | } | |
2319 | } | |
2320 | ||
2321 | /* | |
2322 | * This tree can share blocks with some other fs tree during relocation | |
2323 | * and we need a proper setup by btrfs_get_fs_root | |
2324 | */ | |
2325 | root = btrfs_get_fs_root(tree_root->fs_info, | |
2326 | BTRFS_DATA_RELOC_TREE_OBJECTID, true); | |
2327 | if (IS_ERR(root)) { | |
2328 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { | |
2329 | ret = PTR_ERR(root); | |
2330 | goto out; | |
2331 | } | |
2332 | } else { | |
2333 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2334 | fs_info->data_reloc_root = root; | |
2335 | } | |
2336 | ||
2337 | location.objectid = BTRFS_QUOTA_TREE_OBJECTID; | |
2338 | root = btrfs_read_tree_root(tree_root, &location); | |
2339 | if (!IS_ERR(root)) { | |
2340 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2341 | set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); | |
2342 | fs_info->quota_root = root; | |
2343 | } | |
2344 | ||
2345 | location.objectid = BTRFS_UUID_TREE_OBJECTID; | |
2346 | root = btrfs_read_tree_root(tree_root, &location); | |
2347 | if (IS_ERR(root)) { | |
2348 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { | |
2349 | ret = PTR_ERR(root); | |
2350 | if (ret != -ENOENT) | |
2351 | goto out; | |
2352 | } | |
2353 | } else { | |
2354 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2355 | fs_info->uuid_root = root; | |
2356 | } | |
2357 | ||
2358 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { | |
2359 | location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; | |
2360 | root = btrfs_read_tree_root(tree_root, &location); | |
2361 | if (IS_ERR(root)) { | |
2362 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { | |
2363 | ret = PTR_ERR(root); | |
2364 | goto out; | |
2365 | } | |
2366 | } else { | |
2367 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2368 | fs_info->free_space_root = root; | |
2369 | } | |
2370 | } | |
2371 | ||
2372 | return 0; | |
2373 | out: | |
2374 | btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", | |
2375 | location.objectid, ret); | |
2376 | return ret; | |
2377 | } | |
2378 | ||
2379 | /* | |
2380 | * Real super block validation | |
2381 | * NOTE: super csum type and incompat features will not be checked here. | |
2382 | * | |
2383 | * @sb: super block to check | |
2384 | * @mirror_num: the super block number to check its bytenr: | |
2385 | * 0 the primary (1st) sb | |
2386 | * 1, 2 2nd and 3rd backup copy | |
2387 | * -1 skip bytenr check | |
2388 | */ | |
2389 | static int validate_super(struct btrfs_fs_info *fs_info, | |
2390 | struct btrfs_super_block *sb, int mirror_num) | |
2391 | { | |
2392 | u64 nodesize = btrfs_super_nodesize(sb); | |
2393 | u64 sectorsize = btrfs_super_sectorsize(sb); | |
2394 | int ret = 0; | |
2395 | ||
2396 | if (btrfs_super_magic(sb) != BTRFS_MAGIC) { | |
2397 | btrfs_err(fs_info, "no valid FS found"); | |
2398 | ret = -EINVAL; | |
2399 | } | |
2400 | if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { | |
2401 | btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", | |
2402 | btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); | |
2403 | ret = -EINVAL; | |
2404 | } | |
2405 | if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { | |
2406 | btrfs_err(fs_info, "tree_root level too big: %d >= %d", | |
2407 | btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); | |
2408 | ret = -EINVAL; | |
2409 | } | |
2410 | if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { | |
2411 | btrfs_err(fs_info, "chunk_root level too big: %d >= %d", | |
2412 | btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); | |
2413 | ret = -EINVAL; | |
2414 | } | |
2415 | if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { | |
2416 | btrfs_err(fs_info, "log_root level too big: %d >= %d", | |
2417 | btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); | |
2418 | ret = -EINVAL; | |
2419 | } | |
2420 | ||
2421 | /* | |
2422 | * Check sectorsize and nodesize first, other check will need it. | |
2423 | * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. | |
2424 | */ | |
2425 | if (!is_power_of_2(sectorsize) || sectorsize < 4096 || | |
2426 | sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { | |
2427 | btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); | |
2428 | ret = -EINVAL; | |
2429 | } | |
2430 | /* Only PAGE SIZE is supported yet */ | |
2431 | if (sectorsize != PAGE_SIZE) { | |
2432 | btrfs_err(fs_info, | |
2433 | "sectorsize %llu not supported yet, only support %lu", | |
2434 | sectorsize, PAGE_SIZE); | |
2435 | ret = -EINVAL; | |
2436 | } | |
2437 | if (!is_power_of_2(nodesize) || nodesize < sectorsize || | |
2438 | nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { | |
2439 | btrfs_err(fs_info, "invalid nodesize %llu", nodesize); | |
2440 | ret = -EINVAL; | |
2441 | } | |
2442 | if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { | |
2443 | btrfs_err(fs_info, "invalid leafsize %u, should be %llu", | |
2444 | le32_to_cpu(sb->__unused_leafsize), nodesize); | |
2445 | ret = -EINVAL; | |
2446 | } | |
2447 | ||
2448 | /* Root alignment check */ | |
2449 | if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { | |
2450 | btrfs_warn(fs_info, "tree_root block unaligned: %llu", | |
2451 | btrfs_super_root(sb)); | |
2452 | ret = -EINVAL; | |
2453 | } | |
2454 | if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { | |
2455 | btrfs_warn(fs_info, "chunk_root block unaligned: %llu", | |
2456 | btrfs_super_chunk_root(sb)); | |
2457 | ret = -EINVAL; | |
2458 | } | |
2459 | if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { | |
2460 | btrfs_warn(fs_info, "log_root block unaligned: %llu", | |
2461 | btrfs_super_log_root(sb)); | |
2462 | ret = -EINVAL; | |
2463 | } | |
2464 | ||
2465 | if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, | |
2466 | BTRFS_FSID_SIZE) != 0) { | |
2467 | btrfs_err(fs_info, | |
2468 | "dev_item UUID does not match metadata fsid: %pU != %pU", | |
2469 | fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); | |
2470 | ret = -EINVAL; | |
2471 | } | |
2472 | ||
2473 | /* | |
2474 | * Hint to catch really bogus numbers, bitflips or so, more exact checks are | |
2475 | * done later | |
2476 | */ | |
2477 | if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { | |
2478 | btrfs_err(fs_info, "bytes_used is too small %llu", | |
2479 | btrfs_super_bytes_used(sb)); | |
2480 | ret = -EINVAL; | |
2481 | } | |
2482 | if (!is_power_of_2(btrfs_super_stripesize(sb))) { | |
2483 | btrfs_err(fs_info, "invalid stripesize %u", | |
2484 | btrfs_super_stripesize(sb)); | |
2485 | ret = -EINVAL; | |
2486 | } | |
2487 | if (btrfs_super_num_devices(sb) > (1UL << 31)) | |
2488 | btrfs_warn(fs_info, "suspicious number of devices: %llu", | |
2489 | btrfs_super_num_devices(sb)); | |
2490 | if (btrfs_super_num_devices(sb) == 0) { | |
2491 | btrfs_err(fs_info, "number of devices is 0"); | |
2492 | ret = -EINVAL; | |
2493 | } | |
2494 | ||
2495 | if (mirror_num >= 0 && | |
2496 | btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { | |
2497 | btrfs_err(fs_info, "super offset mismatch %llu != %u", | |
2498 | btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); | |
2499 | ret = -EINVAL; | |
2500 | } | |
2501 | ||
2502 | /* | |
2503 | * Obvious sys_chunk_array corruptions, it must hold at least one key | |
2504 | * and one chunk | |
2505 | */ | |
2506 | if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { | |
2507 | btrfs_err(fs_info, "system chunk array too big %u > %u", | |
2508 | btrfs_super_sys_array_size(sb), | |
2509 | BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); | |
2510 | ret = -EINVAL; | |
2511 | } | |
2512 | if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) | |
2513 | + sizeof(struct btrfs_chunk)) { | |
2514 | btrfs_err(fs_info, "system chunk array too small %u < %zu", | |
2515 | btrfs_super_sys_array_size(sb), | |
2516 | sizeof(struct btrfs_disk_key) | |
2517 | + sizeof(struct btrfs_chunk)); | |
2518 | ret = -EINVAL; | |
2519 | } | |
2520 | ||
2521 | /* | |
2522 | * The generation is a global counter, we'll trust it more than the others | |
2523 | * but it's still possible that it's the one that's wrong. | |
2524 | */ | |
2525 | if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) | |
2526 | btrfs_warn(fs_info, | |
2527 | "suspicious: generation < chunk_root_generation: %llu < %llu", | |
2528 | btrfs_super_generation(sb), | |
2529 | btrfs_super_chunk_root_generation(sb)); | |
2530 | if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) | |
2531 | && btrfs_super_cache_generation(sb) != (u64)-1) | |
2532 | btrfs_warn(fs_info, | |
2533 | "suspicious: generation < cache_generation: %llu < %llu", | |
2534 | btrfs_super_generation(sb), | |
2535 | btrfs_super_cache_generation(sb)); | |
2536 | ||
2537 | return ret; | |
2538 | } | |
2539 | ||
2540 | /* | |
2541 | * Validation of super block at mount time. | |
2542 | * Some checks already done early at mount time, like csum type and incompat | |
2543 | * flags will be skipped. | |
2544 | */ | |
2545 | static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) | |
2546 | { | |
2547 | return validate_super(fs_info, fs_info->super_copy, 0); | |
2548 | } | |
2549 | ||
2550 | /* | |
2551 | * Validation of super block at write time. | |
2552 | * Some checks like bytenr check will be skipped as their values will be | |
2553 | * overwritten soon. | |
2554 | * Extra checks like csum type and incompat flags will be done here. | |
2555 | */ | |
2556 | static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, | |
2557 | struct btrfs_super_block *sb) | |
2558 | { | |
2559 | int ret; | |
2560 | ||
2561 | ret = validate_super(fs_info, sb, -1); | |
2562 | if (ret < 0) | |
2563 | goto out; | |
2564 | if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { | |
2565 | ret = -EUCLEAN; | |
2566 | btrfs_err(fs_info, "invalid csum type, has %u want %u", | |
2567 | btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); | |
2568 | goto out; | |
2569 | } | |
2570 | if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { | |
2571 | ret = -EUCLEAN; | |
2572 | btrfs_err(fs_info, | |
2573 | "invalid incompat flags, has 0x%llx valid mask 0x%llx", | |
2574 | btrfs_super_incompat_flags(sb), | |
2575 | (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); | |
2576 | goto out; | |
2577 | } | |
2578 | out: | |
2579 | if (ret < 0) | |
2580 | btrfs_err(fs_info, | |
2581 | "super block corruption detected before writing it to disk"); | |
2582 | return ret; | |
2583 | } | |
2584 | ||
2585 | static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) | |
2586 | { | |
2587 | int backup_index = find_newest_super_backup(fs_info); | |
2588 | struct btrfs_super_block *sb = fs_info->super_copy; | |
2589 | struct btrfs_root *tree_root = fs_info->tree_root; | |
2590 | bool handle_error = false; | |
2591 | int ret = 0; | |
2592 | int i; | |
2593 | ||
2594 | for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { | |
2595 | u64 generation; | |
2596 | int level; | |
2597 | ||
2598 | if (handle_error) { | |
2599 | if (!IS_ERR(tree_root->node)) | |
2600 | free_extent_buffer(tree_root->node); | |
2601 | tree_root->node = NULL; | |
2602 | ||
2603 | if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) | |
2604 | break; | |
2605 | ||
2606 | free_root_pointers(fs_info, 0); | |
2607 | ||
2608 | /* | |
2609 | * Don't use the log in recovery mode, it won't be | |
2610 | * valid | |
2611 | */ | |
2612 | btrfs_set_super_log_root(sb, 0); | |
2613 | ||
2614 | /* We can't trust the free space cache either */ | |
2615 | btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); | |
2616 | ||
2617 | ret = read_backup_root(fs_info, i); | |
2618 | backup_index = ret; | |
2619 | if (ret < 0) | |
2620 | return ret; | |
2621 | } | |
2622 | generation = btrfs_super_generation(sb); | |
2623 | level = btrfs_super_root_level(sb); | |
2624 | tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb), | |
2625 | BTRFS_ROOT_TREE_OBJECTID, | |
2626 | generation, level, NULL); | |
2627 | if (IS_ERR(tree_root->node)) { | |
2628 | handle_error = true; | |
2629 | ret = PTR_ERR(tree_root->node); | |
2630 | tree_root->node = NULL; | |
2631 | btrfs_warn(fs_info, "couldn't read tree root"); | |
2632 | continue; | |
2633 | ||
2634 | } else if (!extent_buffer_uptodate(tree_root->node)) { | |
2635 | handle_error = true; | |
2636 | ret = -EIO; | |
2637 | btrfs_warn(fs_info, "error while reading tree root"); | |
2638 | continue; | |
2639 | } | |
2640 | ||
2641 | btrfs_set_root_node(&tree_root->root_item, tree_root->node); | |
2642 | tree_root->commit_root = btrfs_root_node(tree_root); | |
2643 | btrfs_set_root_refs(&tree_root->root_item, 1); | |
2644 | ||
2645 | /* | |
2646 | * No need to hold btrfs_root::objectid_mutex since the fs | |
2647 | * hasn't been fully initialised and we are the only user | |
2648 | */ | |
2649 | ret = btrfs_find_highest_objectid(tree_root, | |
2650 | &tree_root->highest_objectid); | |
2651 | if (ret < 0) { | |
2652 | handle_error = true; | |
2653 | continue; | |
2654 | } | |
2655 | ||
2656 | ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); | |
2657 | ||
2658 | ret = btrfs_read_roots(fs_info); | |
2659 | if (ret < 0) { | |
2660 | handle_error = true; | |
2661 | continue; | |
2662 | } | |
2663 | ||
2664 | /* All successful */ | |
2665 | fs_info->generation = generation; | |
2666 | fs_info->last_trans_committed = generation; | |
2667 | ||
2668 | /* Always begin writing backup roots after the one being used */ | |
2669 | if (backup_index < 0) { | |
2670 | fs_info->backup_root_index = 0; | |
2671 | } else { | |
2672 | fs_info->backup_root_index = backup_index + 1; | |
2673 | fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; | |
2674 | } | |
2675 | break; | |
2676 | } | |
2677 | ||
2678 | return ret; | |
2679 | } | |
2680 | ||
2681 | void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) | |
2682 | { | |
2683 | INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); | |
2684 | INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); | |
2685 | INIT_LIST_HEAD(&fs_info->trans_list); | |
2686 | INIT_LIST_HEAD(&fs_info->dead_roots); | |
2687 | INIT_LIST_HEAD(&fs_info->delayed_iputs); | |
2688 | INIT_LIST_HEAD(&fs_info->delalloc_roots); | |
2689 | INIT_LIST_HEAD(&fs_info->caching_block_groups); | |
2690 | spin_lock_init(&fs_info->delalloc_root_lock); | |
2691 | spin_lock_init(&fs_info->trans_lock); | |
2692 | spin_lock_init(&fs_info->fs_roots_radix_lock); | |
2693 | spin_lock_init(&fs_info->delayed_iput_lock); | |
2694 | spin_lock_init(&fs_info->defrag_inodes_lock); | |
2695 | spin_lock_init(&fs_info->super_lock); | |
2696 | spin_lock_init(&fs_info->buffer_lock); | |
2697 | spin_lock_init(&fs_info->unused_bgs_lock); | |
2698 | rwlock_init(&fs_info->tree_mod_log_lock); | |
2699 | mutex_init(&fs_info->unused_bg_unpin_mutex); | |
2700 | mutex_init(&fs_info->delete_unused_bgs_mutex); | |
2701 | mutex_init(&fs_info->reloc_mutex); | |
2702 | mutex_init(&fs_info->delalloc_root_mutex); | |
2703 | seqlock_init(&fs_info->profiles_lock); | |
2704 | ||
2705 | INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); | |
2706 | INIT_LIST_HEAD(&fs_info->space_info); | |
2707 | INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); | |
2708 | INIT_LIST_HEAD(&fs_info->unused_bgs); | |
2709 | #ifdef CONFIG_BTRFS_DEBUG | |
2710 | INIT_LIST_HEAD(&fs_info->allocated_roots); | |
2711 | INIT_LIST_HEAD(&fs_info->allocated_ebs); | |
2712 | spin_lock_init(&fs_info->eb_leak_lock); | |
2713 | #endif | |
2714 | extent_map_tree_init(&fs_info->mapping_tree); | |
2715 | btrfs_init_block_rsv(&fs_info->global_block_rsv, | |
2716 | BTRFS_BLOCK_RSV_GLOBAL); | |
2717 | btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); | |
2718 | btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); | |
2719 | btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); | |
2720 | btrfs_init_block_rsv(&fs_info->delayed_block_rsv, | |
2721 | BTRFS_BLOCK_RSV_DELOPS); | |
2722 | btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, | |
2723 | BTRFS_BLOCK_RSV_DELREFS); | |
2724 | ||
2725 | atomic_set(&fs_info->async_delalloc_pages, 0); | |
2726 | atomic_set(&fs_info->defrag_running, 0); | |
2727 | atomic_set(&fs_info->reada_works_cnt, 0); | |
2728 | atomic_set(&fs_info->nr_delayed_iputs, 0); | |
2729 | atomic64_set(&fs_info->tree_mod_seq, 0); | |
2730 | fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; | |
2731 | fs_info->metadata_ratio = 0; | |
2732 | fs_info->defrag_inodes = RB_ROOT; | |
2733 | atomic64_set(&fs_info->free_chunk_space, 0); | |
2734 | fs_info->tree_mod_log = RB_ROOT; | |
2735 | fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; | |
2736 | fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ | |
2737 | /* readahead state */ | |
2738 | INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); | |
2739 | spin_lock_init(&fs_info->reada_lock); | |
2740 | btrfs_init_ref_verify(fs_info); | |
2741 | ||
2742 | fs_info->thread_pool_size = min_t(unsigned long, | |
2743 | num_online_cpus() + 2, 8); | |
2744 | ||
2745 | INIT_LIST_HEAD(&fs_info->ordered_roots); | |
2746 | spin_lock_init(&fs_info->ordered_root_lock); | |
2747 | ||
2748 | btrfs_init_scrub(fs_info); | |
2749 | #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY | |
2750 | fs_info->check_integrity_print_mask = 0; | |
2751 | #endif | |
2752 | btrfs_init_balance(fs_info); | |
2753 | btrfs_init_async_reclaim_work(fs_info); | |
2754 | ||
2755 | spin_lock_init(&fs_info->block_group_cache_lock); | |
2756 | fs_info->block_group_cache_tree = RB_ROOT; | |
2757 | fs_info->first_logical_byte = (u64)-1; | |
2758 | ||
2759 | extent_io_tree_init(fs_info, &fs_info->excluded_extents, | |
2760 | IO_TREE_FS_EXCLUDED_EXTENTS, NULL); | |
2761 | set_bit(BTRFS_FS_BARRIER, &fs_info->flags); | |
2762 | ||
2763 | mutex_init(&fs_info->ordered_operations_mutex); | |
2764 | mutex_init(&fs_info->tree_log_mutex); | |
2765 | mutex_init(&fs_info->chunk_mutex); | |
2766 | mutex_init(&fs_info->transaction_kthread_mutex); | |
2767 | mutex_init(&fs_info->cleaner_mutex); | |
2768 | mutex_init(&fs_info->ro_block_group_mutex); | |
2769 | init_rwsem(&fs_info->commit_root_sem); | |
2770 | init_rwsem(&fs_info->cleanup_work_sem); | |
2771 | init_rwsem(&fs_info->subvol_sem); | |
2772 | sema_init(&fs_info->uuid_tree_rescan_sem, 1); | |
2773 | ||
2774 | btrfs_init_dev_replace_locks(fs_info); | |
2775 | btrfs_init_qgroup(fs_info); | |
2776 | btrfs_discard_init(fs_info); | |
2777 | ||
2778 | btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); | |
2779 | btrfs_init_free_cluster(&fs_info->data_alloc_cluster); | |
2780 | ||
2781 | init_waitqueue_head(&fs_info->transaction_throttle); | |
2782 | init_waitqueue_head(&fs_info->transaction_wait); | |
2783 | init_waitqueue_head(&fs_info->transaction_blocked_wait); | |
2784 | init_waitqueue_head(&fs_info->async_submit_wait); | |
2785 | init_waitqueue_head(&fs_info->delayed_iputs_wait); | |
2786 | ||
2787 | /* Usable values until the real ones are cached from the superblock */ | |
2788 | fs_info->nodesize = 4096; | |
2789 | fs_info->sectorsize = 4096; | |
2790 | fs_info->sectorsize_bits = ilog2(4096); | |
2791 | fs_info->stripesize = 4096; | |
2792 | ||
2793 | spin_lock_init(&fs_info->swapfile_pins_lock); | |
2794 | fs_info->swapfile_pins = RB_ROOT; | |
2795 | ||
2796 | fs_info->send_in_progress = 0; | |
2797 | } | |
2798 | ||
2799 | static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) | |
2800 | { | |
2801 | int ret; | |
2802 | ||
2803 | fs_info->sb = sb; | |
2804 | sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; | |
2805 | sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); | |
2806 | ||
2807 | ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); | |
2808 | if (ret) | |
2809 | return ret; | |
2810 | ||
2811 | ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); | |
2812 | if (ret) | |
2813 | return ret; | |
2814 | ||
2815 | fs_info->dirty_metadata_batch = PAGE_SIZE * | |
2816 | (1 + ilog2(nr_cpu_ids)); | |
2817 | ||
2818 | ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); | |
2819 | if (ret) | |
2820 | return ret; | |
2821 | ||
2822 | ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, | |
2823 | GFP_KERNEL); | |
2824 | if (ret) | |
2825 | return ret; | |
2826 | ||
2827 | fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), | |
2828 | GFP_KERNEL); | |
2829 | if (!fs_info->delayed_root) | |
2830 | return -ENOMEM; | |
2831 | btrfs_init_delayed_root(fs_info->delayed_root); | |
2832 | ||
2833 | if (sb_rdonly(sb)) | |
2834 | set_bit(BTRFS_FS_STATE_RO, &fs_info->fs_state); | |
2835 | ||
2836 | return btrfs_alloc_stripe_hash_table(fs_info); | |
2837 | } | |
2838 | ||
2839 | static int btrfs_uuid_rescan_kthread(void *data) | |
2840 | { | |
2841 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
2842 | int ret; | |
2843 | ||
2844 | /* | |
2845 | * 1st step is to iterate through the existing UUID tree and | |
2846 | * to delete all entries that contain outdated data. | |
2847 | * 2nd step is to add all missing entries to the UUID tree. | |
2848 | */ | |
2849 | ret = btrfs_uuid_tree_iterate(fs_info); | |
2850 | if (ret < 0) { | |
2851 | if (ret != -EINTR) | |
2852 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", | |
2853 | ret); | |
2854 | up(&fs_info->uuid_tree_rescan_sem); | |
2855 | return ret; | |
2856 | } | |
2857 | return btrfs_uuid_scan_kthread(data); | |
2858 | } | |
2859 | ||
2860 | static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) | |
2861 | { | |
2862 | struct task_struct *task; | |
2863 | ||
2864 | down(&fs_info->uuid_tree_rescan_sem); | |
2865 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
2866 | if (IS_ERR(task)) { | |
2867 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
2868 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); | |
2869 | up(&fs_info->uuid_tree_rescan_sem); | |
2870 | return PTR_ERR(task); | |
2871 | } | |
2872 | ||
2873 | return 0; | |
2874 | } | |
2875 | ||
2876 | /* | |
2877 | * Some options only have meaning at mount time and shouldn't persist across | |
2878 | * remounts, or be displayed. Clear these at the end of mount and remount | |
2879 | * code paths. | |
2880 | */ | |
2881 | void btrfs_clear_oneshot_options(struct btrfs_fs_info *fs_info) | |
2882 | { | |
2883 | btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT); | |
2884 | btrfs_clear_opt(fs_info->mount_opt, CLEAR_CACHE); | |
2885 | } | |
2886 | ||
2887 | /* | |
2888 | * Mounting logic specific to read-write file systems. Shared by open_ctree | |
2889 | * and btrfs_remount when remounting from read-only to read-write. | |
2890 | */ | |
2891 | int btrfs_start_pre_rw_mount(struct btrfs_fs_info *fs_info) | |
2892 | { | |
2893 | int ret; | |
2894 | const bool cache_opt = btrfs_test_opt(fs_info, SPACE_CACHE); | |
2895 | bool clear_free_space_tree = false; | |
2896 | ||
2897 | if (btrfs_test_opt(fs_info, CLEAR_CACHE) && | |
2898 | btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { | |
2899 | clear_free_space_tree = true; | |
2900 | } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && | |
2901 | !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { | |
2902 | btrfs_warn(fs_info, "free space tree is invalid"); | |
2903 | clear_free_space_tree = true; | |
2904 | } | |
2905 | ||
2906 | if (clear_free_space_tree) { | |
2907 | btrfs_info(fs_info, "clearing free space tree"); | |
2908 | ret = btrfs_clear_free_space_tree(fs_info); | |
2909 | if (ret) { | |
2910 | btrfs_warn(fs_info, | |
2911 | "failed to clear free space tree: %d", ret); | |
2912 | goto out; | |
2913 | } | |
2914 | } | |
2915 | ||
2916 | ret = btrfs_cleanup_fs_roots(fs_info); | |
2917 | if (ret) | |
2918 | goto out; | |
2919 | ||
2920 | down_read(&fs_info->cleanup_work_sem); | |
2921 | if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || | |
2922 | (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { | |
2923 | up_read(&fs_info->cleanup_work_sem); | |
2924 | goto out; | |
2925 | } | |
2926 | up_read(&fs_info->cleanup_work_sem); | |
2927 | ||
2928 | mutex_lock(&fs_info->cleaner_mutex); | |
2929 | ret = btrfs_recover_relocation(fs_info->tree_root); | |
2930 | mutex_unlock(&fs_info->cleaner_mutex); | |
2931 | if (ret < 0) { | |
2932 | btrfs_warn(fs_info, "failed to recover relocation: %d", ret); | |
2933 | goto out; | |
2934 | } | |
2935 | ||
2936 | if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && | |
2937 | !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { | |
2938 | btrfs_info(fs_info, "creating free space tree"); | |
2939 | ret = btrfs_create_free_space_tree(fs_info); | |
2940 | if (ret) { | |
2941 | btrfs_warn(fs_info, | |
2942 | "failed to create free space tree: %d", ret); | |
2943 | goto out; | |
2944 | } | |
2945 | } | |
2946 | ||
2947 | if (cache_opt != btrfs_free_space_cache_v1_active(fs_info)) { | |
2948 | ret = btrfs_set_free_space_cache_v1_active(fs_info, cache_opt); | |
2949 | if (ret) | |
2950 | goto out; | |
2951 | } | |
2952 | ||
2953 | ret = btrfs_resume_balance_async(fs_info); | |
2954 | if (ret) | |
2955 | goto out; | |
2956 | ||
2957 | ret = btrfs_resume_dev_replace_async(fs_info); | |
2958 | if (ret) { | |
2959 | btrfs_warn(fs_info, "failed to resume dev_replace"); | |
2960 | goto out; | |
2961 | } | |
2962 | ||
2963 | btrfs_qgroup_rescan_resume(fs_info); | |
2964 | ||
2965 | if (!fs_info->uuid_root) { | |
2966 | btrfs_info(fs_info, "creating UUID tree"); | |
2967 | ret = btrfs_create_uuid_tree(fs_info); | |
2968 | if (ret) { | |
2969 | btrfs_warn(fs_info, | |
2970 | "failed to create the UUID tree %d", ret); | |
2971 | goto out; | |
2972 | } | |
2973 | } | |
2974 | ||
2975 | ret = btrfs_find_orphan_roots(fs_info); | |
2976 | out: | |
2977 | return ret; | |
2978 | } | |
2979 | ||
2980 | int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, | |
2981 | char *options) | |
2982 | { | |
2983 | u32 sectorsize; | |
2984 | u32 nodesize; | |
2985 | u32 stripesize; | |
2986 | u64 generation; | |
2987 | u64 features; | |
2988 | u16 csum_type; | |
2989 | struct btrfs_super_block *disk_super; | |
2990 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); | |
2991 | struct btrfs_root *tree_root; | |
2992 | struct btrfs_root *chunk_root; | |
2993 | int ret; | |
2994 | int err = -EINVAL; | |
2995 | int level; | |
2996 | ||
2997 | ret = init_mount_fs_info(fs_info, sb); | |
2998 | if (ret) { | |
2999 | err = ret; | |
3000 | goto fail; | |
3001 | } | |
3002 | ||
3003 | /* These need to be init'ed before we start creating inodes and such. */ | |
3004 | tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, | |
3005 | GFP_KERNEL); | |
3006 | fs_info->tree_root = tree_root; | |
3007 | chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, | |
3008 | GFP_KERNEL); | |
3009 | fs_info->chunk_root = chunk_root; | |
3010 | if (!tree_root || !chunk_root) { | |
3011 | err = -ENOMEM; | |
3012 | goto fail; | |
3013 | } | |
3014 | ||
3015 | fs_info->btree_inode = new_inode(sb); | |
3016 | if (!fs_info->btree_inode) { | |
3017 | err = -ENOMEM; | |
3018 | goto fail; | |
3019 | } | |
3020 | mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); | |
3021 | btrfs_init_btree_inode(fs_info); | |
3022 | ||
3023 | invalidate_bdev(fs_devices->latest_bdev); | |
3024 | ||
3025 | /* | |
3026 | * Read super block and check the signature bytes only | |
3027 | */ | |
3028 | disk_super = btrfs_read_dev_super(fs_devices->latest_bdev); | |
3029 | if (IS_ERR(disk_super)) { | |
3030 | err = PTR_ERR(disk_super); | |
3031 | goto fail_alloc; | |
3032 | } | |
3033 | ||
3034 | /* | |
3035 | * Verify the type first, if that or the checksum value are | |
3036 | * corrupted, we'll find out | |
3037 | */ | |
3038 | csum_type = btrfs_super_csum_type(disk_super); | |
3039 | if (!btrfs_supported_super_csum(csum_type)) { | |
3040 | btrfs_err(fs_info, "unsupported checksum algorithm: %u", | |
3041 | csum_type); | |
3042 | err = -EINVAL; | |
3043 | btrfs_release_disk_super(disk_super); | |
3044 | goto fail_alloc; | |
3045 | } | |
3046 | ||
3047 | ret = btrfs_init_csum_hash(fs_info, csum_type); | |
3048 | if (ret) { | |
3049 | err = ret; | |
3050 | btrfs_release_disk_super(disk_super); | |
3051 | goto fail_alloc; | |
3052 | } | |
3053 | ||
3054 | /* | |
3055 | * We want to check superblock checksum, the type is stored inside. | |
3056 | * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). | |
3057 | */ | |
3058 | if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { | |
3059 | btrfs_err(fs_info, "superblock checksum mismatch"); | |
3060 | err = -EINVAL; | |
3061 | btrfs_release_disk_super(disk_super); | |
3062 | goto fail_alloc; | |
3063 | } | |
3064 | ||
3065 | /* | |
3066 | * super_copy is zeroed at allocation time and we never touch the | |
3067 | * following bytes up to INFO_SIZE, the checksum is calculated from | |
3068 | * the whole block of INFO_SIZE | |
3069 | */ | |
3070 | memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); | |
3071 | btrfs_release_disk_super(disk_super); | |
3072 | ||
3073 | disk_super = fs_info->super_copy; | |
3074 | ||
3075 | ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, | |
3076 | BTRFS_FSID_SIZE)); | |
3077 | ||
3078 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { | |
3079 | ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, | |
3080 | fs_info->super_copy->metadata_uuid, | |
3081 | BTRFS_FSID_SIZE)); | |
3082 | } | |
3083 | ||
3084 | features = btrfs_super_flags(disk_super); | |
3085 | if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { | |
3086 | features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; | |
3087 | btrfs_set_super_flags(disk_super, features); | |
3088 | btrfs_info(fs_info, | |
3089 | "found metadata UUID change in progress flag, clearing"); | |
3090 | } | |
3091 | ||
3092 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
3093 | sizeof(*fs_info->super_for_commit)); | |
3094 | ||
3095 | ret = btrfs_validate_mount_super(fs_info); | |
3096 | if (ret) { | |
3097 | btrfs_err(fs_info, "superblock contains fatal errors"); | |
3098 | err = -EINVAL; | |
3099 | goto fail_alloc; | |
3100 | } | |
3101 | ||
3102 | if (!btrfs_super_root(disk_super)) | |
3103 | goto fail_alloc; | |
3104 | ||
3105 | /* check FS state, whether FS is broken. */ | |
3106 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) | |
3107 | set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); | |
3108 | ||
3109 | /* | |
3110 | * In the long term, we'll store the compression type in the super | |
3111 | * block, and it'll be used for per file compression control. | |
3112 | */ | |
3113 | fs_info->compress_type = BTRFS_COMPRESS_ZLIB; | |
3114 | ||
3115 | ret = btrfs_parse_options(fs_info, options, sb->s_flags); | |
3116 | if (ret) { | |
3117 | err = ret; | |
3118 | goto fail_alloc; | |
3119 | } | |
3120 | ||
3121 | features = btrfs_super_incompat_flags(disk_super) & | |
3122 | ~BTRFS_FEATURE_INCOMPAT_SUPP; | |
3123 | if (features) { | |
3124 | btrfs_err(fs_info, | |
3125 | "cannot mount because of unsupported optional features (%llx)", | |
3126 | features); | |
3127 | err = -EINVAL; | |
3128 | goto fail_alloc; | |
3129 | } | |
3130 | ||
3131 | features = btrfs_super_incompat_flags(disk_super); | |
3132 | features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; | |
3133 | if (fs_info->compress_type == BTRFS_COMPRESS_LZO) | |
3134 | features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; | |
3135 | else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) | |
3136 | features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; | |
3137 | ||
3138 | if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) | |
3139 | btrfs_info(fs_info, "has skinny extents"); | |
3140 | ||
3141 | fs_info->zoned = (features & BTRFS_FEATURE_INCOMPAT_ZONED); | |
3142 | ||
3143 | /* | |
3144 | * flag our filesystem as having big metadata blocks if | |
3145 | * they are bigger than the page size | |
3146 | */ | |
3147 | if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { | |
3148 | if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) | |
3149 | btrfs_info(fs_info, | |
3150 | "flagging fs with big metadata feature"); | |
3151 | features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; | |
3152 | } | |
3153 | ||
3154 | nodesize = btrfs_super_nodesize(disk_super); | |
3155 | sectorsize = btrfs_super_sectorsize(disk_super); | |
3156 | stripesize = sectorsize; | |
3157 | fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); | |
3158 | fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); | |
3159 | ||
3160 | /* Cache block sizes */ | |
3161 | fs_info->nodesize = nodesize; | |
3162 | fs_info->sectorsize = sectorsize; | |
3163 | fs_info->sectorsize_bits = ilog2(sectorsize); | |
3164 | fs_info->csum_size = btrfs_super_csum_size(disk_super); | |
3165 | fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; | |
3166 | fs_info->stripesize = stripesize; | |
3167 | ||
3168 | /* | |
3169 | * mixed block groups end up with duplicate but slightly offset | |
3170 | * extent buffers for the same range. It leads to corruptions | |
3171 | */ | |
3172 | if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && | |
3173 | (sectorsize != nodesize)) { | |
3174 | btrfs_err(fs_info, | |
3175 | "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", | |
3176 | nodesize, sectorsize); | |
3177 | goto fail_alloc; | |
3178 | } | |
3179 | ||
3180 | /* | |
3181 | * Needn't use the lock because there is no other task which will | |
3182 | * update the flag. | |
3183 | */ | |
3184 | btrfs_set_super_incompat_flags(disk_super, features); | |
3185 | ||
3186 | features = btrfs_super_compat_ro_flags(disk_super) & | |
3187 | ~BTRFS_FEATURE_COMPAT_RO_SUPP; | |
3188 | if (!sb_rdonly(sb) && features) { | |
3189 | btrfs_err(fs_info, | |
3190 | "cannot mount read-write because of unsupported optional features (%llx)", | |
3191 | features); | |
3192 | err = -EINVAL; | |
3193 | goto fail_alloc; | |
3194 | } | |
3195 | ||
3196 | ret = btrfs_init_workqueues(fs_info, fs_devices); | |
3197 | if (ret) { | |
3198 | err = ret; | |
3199 | goto fail_sb_buffer; | |
3200 | } | |
3201 | ||
3202 | sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); | |
3203 | sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); | |
3204 | ||
3205 | sb->s_blocksize = sectorsize; | |
3206 | sb->s_blocksize_bits = blksize_bits(sectorsize); | |
3207 | memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); | |
3208 | ||
3209 | mutex_lock(&fs_info->chunk_mutex); | |
3210 | ret = btrfs_read_sys_array(fs_info); | |
3211 | mutex_unlock(&fs_info->chunk_mutex); | |
3212 | if (ret) { | |
3213 | btrfs_err(fs_info, "failed to read the system array: %d", ret); | |
3214 | goto fail_sb_buffer; | |
3215 | } | |
3216 | ||
3217 | generation = btrfs_super_chunk_root_generation(disk_super); | |
3218 | level = btrfs_super_chunk_root_level(disk_super); | |
3219 | ||
3220 | chunk_root->node = read_tree_block(fs_info, | |
3221 | btrfs_super_chunk_root(disk_super), | |
3222 | BTRFS_CHUNK_TREE_OBJECTID, | |
3223 | generation, level, NULL); | |
3224 | if (IS_ERR(chunk_root->node) || | |
3225 | !extent_buffer_uptodate(chunk_root->node)) { | |
3226 | btrfs_err(fs_info, "failed to read chunk root"); | |
3227 | if (!IS_ERR(chunk_root->node)) | |
3228 | free_extent_buffer(chunk_root->node); | |
3229 | chunk_root->node = NULL; | |
3230 | goto fail_tree_roots; | |
3231 | } | |
3232 | btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); | |
3233 | chunk_root->commit_root = btrfs_root_node(chunk_root); | |
3234 | ||
3235 | read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, | |
3236 | offsetof(struct btrfs_header, chunk_tree_uuid), | |
3237 | BTRFS_UUID_SIZE); | |
3238 | ||
3239 | ret = btrfs_read_chunk_tree(fs_info); | |
3240 | if (ret) { | |
3241 | btrfs_err(fs_info, "failed to read chunk tree: %d", ret); | |
3242 | goto fail_tree_roots; | |
3243 | } | |
3244 | ||
3245 | /* | |
3246 | * At this point we know all the devices that make this filesystem, | |
3247 | * including the seed devices but we don't know yet if the replace | |
3248 | * target is required. So free devices that are not part of this | |
3249 | * filesystem but skip the replace traget device which is checked | |
3250 | * below in btrfs_init_dev_replace(). | |
3251 | */ | |
3252 | btrfs_free_extra_devids(fs_devices); | |
3253 | if (!fs_devices->latest_bdev) { | |
3254 | btrfs_err(fs_info, "failed to read devices"); | |
3255 | goto fail_tree_roots; | |
3256 | } | |
3257 | ||
3258 | ret = init_tree_roots(fs_info); | |
3259 | if (ret) | |
3260 | goto fail_tree_roots; | |
3261 | ||
3262 | /* | |
3263 | * If we have a uuid root and we're not being told to rescan we need to | |
3264 | * check the generation here so we can set the | |
3265 | * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the | |
3266 | * transaction during a balance or the log replay without updating the | |
3267 | * uuid generation, and then if we crash we would rescan the uuid tree, | |
3268 | * even though it was perfectly fine. | |
3269 | */ | |
3270 | if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && | |
3271 | fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) | |
3272 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); | |
3273 | ||
3274 | ret = btrfs_verify_dev_extents(fs_info); | |
3275 | if (ret) { | |
3276 | btrfs_err(fs_info, | |
3277 | "failed to verify dev extents against chunks: %d", | |
3278 | ret); | |
3279 | goto fail_block_groups; | |
3280 | } | |
3281 | ret = btrfs_recover_balance(fs_info); | |
3282 | if (ret) { | |
3283 | btrfs_err(fs_info, "failed to recover balance: %d", ret); | |
3284 | goto fail_block_groups; | |
3285 | } | |
3286 | ||
3287 | ret = btrfs_init_dev_stats(fs_info); | |
3288 | if (ret) { | |
3289 | btrfs_err(fs_info, "failed to init dev_stats: %d", ret); | |
3290 | goto fail_block_groups; | |
3291 | } | |
3292 | ||
3293 | ret = btrfs_init_dev_replace(fs_info); | |
3294 | if (ret) { | |
3295 | btrfs_err(fs_info, "failed to init dev_replace: %d", ret); | |
3296 | goto fail_block_groups; | |
3297 | } | |
3298 | ||
3299 | ret = btrfs_check_zoned_mode(fs_info); | |
3300 | if (ret) { | |
3301 | btrfs_err(fs_info, "failed to initialize zoned mode: %d", | |
3302 | ret); | |
3303 | goto fail_block_groups; | |
3304 | } | |
3305 | ||
3306 | ret = btrfs_sysfs_add_fsid(fs_devices); | |
3307 | if (ret) { | |
3308 | btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", | |
3309 | ret); | |
3310 | goto fail_block_groups; | |
3311 | } | |
3312 | ||
3313 | ret = btrfs_sysfs_add_mounted(fs_info); | |
3314 | if (ret) { | |
3315 | btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); | |
3316 | goto fail_fsdev_sysfs; | |
3317 | } | |
3318 | ||
3319 | ret = btrfs_init_space_info(fs_info); | |
3320 | if (ret) { | |
3321 | btrfs_err(fs_info, "failed to initialize space info: %d", ret); | |
3322 | goto fail_sysfs; | |
3323 | } | |
3324 | ||
3325 | ret = btrfs_read_block_groups(fs_info); | |
3326 | if (ret) { | |
3327 | btrfs_err(fs_info, "failed to read block groups: %d", ret); | |
3328 | goto fail_sysfs; | |
3329 | } | |
3330 | ||
3331 | if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { | |
3332 | btrfs_warn(fs_info, | |
3333 | "writable mount is not allowed due to too many missing devices"); | |
3334 | goto fail_sysfs; | |
3335 | } | |
3336 | ||
3337 | fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, | |
3338 | "btrfs-cleaner"); | |
3339 | if (IS_ERR(fs_info->cleaner_kthread)) | |
3340 | goto fail_sysfs; | |
3341 | ||
3342 | fs_info->transaction_kthread = kthread_run(transaction_kthread, | |
3343 | tree_root, | |
3344 | "btrfs-transaction"); | |
3345 | if (IS_ERR(fs_info->transaction_kthread)) | |
3346 | goto fail_cleaner; | |
3347 | ||
3348 | if (!btrfs_test_opt(fs_info, NOSSD) && | |
3349 | !fs_info->fs_devices->rotating) { | |
3350 | btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); | |
3351 | } | |
3352 | ||
3353 | /* | |
3354 | * Mount does not set all options immediately, we can do it now and do | |
3355 | * not have to wait for transaction commit | |
3356 | */ | |
3357 | btrfs_apply_pending_changes(fs_info); | |
3358 | ||
3359 | #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY | |
3360 | if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { | |
3361 | ret = btrfsic_mount(fs_info, fs_devices, | |
3362 | btrfs_test_opt(fs_info, | |
3363 | CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? | |
3364 | 1 : 0, | |
3365 | fs_info->check_integrity_print_mask); | |
3366 | if (ret) | |
3367 | btrfs_warn(fs_info, | |
3368 | "failed to initialize integrity check module: %d", | |
3369 | ret); | |
3370 | } | |
3371 | #endif | |
3372 | ret = btrfs_read_qgroup_config(fs_info); | |
3373 | if (ret) | |
3374 | goto fail_trans_kthread; | |
3375 | ||
3376 | if (btrfs_build_ref_tree(fs_info)) | |
3377 | btrfs_err(fs_info, "couldn't build ref tree"); | |
3378 | ||
3379 | /* do not make disk changes in broken FS or nologreplay is given */ | |
3380 | if (btrfs_super_log_root(disk_super) != 0 && | |
3381 | !btrfs_test_opt(fs_info, NOLOGREPLAY)) { | |
3382 | btrfs_info(fs_info, "start tree-log replay"); | |
3383 | ret = btrfs_replay_log(fs_info, fs_devices); | |
3384 | if (ret) { | |
3385 | err = ret; | |
3386 | goto fail_qgroup; | |
3387 | } | |
3388 | } | |
3389 | ||
3390 | fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); | |
3391 | if (IS_ERR(fs_info->fs_root)) { | |
3392 | err = PTR_ERR(fs_info->fs_root); | |
3393 | btrfs_warn(fs_info, "failed to read fs tree: %d", err); | |
3394 | fs_info->fs_root = NULL; | |
3395 | goto fail_qgroup; | |
3396 | } | |
3397 | ||
3398 | if (sb_rdonly(sb)) | |
3399 | goto clear_oneshot; | |
3400 | ||
3401 | ret = btrfs_start_pre_rw_mount(fs_info); | |
3402 | if (ret) { | |
3403 | close_ctree(fs_info); | |
3404 | return ret; | |
3405 | } | |
3406 | btrfs_discard_resume(fs_info); | |
3407 | ||
3408 | if (fs_info->uuid_root && | |
3409 | (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || | |
3410 | fs_info->generation != btrfs_super_uuid_tree_generation(disk_super))) { | |
3411 | btrfs_info(fs_info, "checking UUID tree"); | |
3412 | ret = btrfs_check_uuid_tree(fs_info); | |
3413 | if (ret) { | |
3414 | btrfs_warn(fs_info, | |
3415 | "failed to check the UUID tree: %d", ret); | |
3416 | close_ctree(fs_info); | |
3417 | return ret; | |
3418 | } | |
3419 | } | |
3420 | ||
3421 | set_bit(BTRFS_FS_OPEN, &fs_info->flags); | |
3422 | ||
3423 | clear_oneshot: | |
3424 | btrfs_clear_oneshot_options(fs_info); | |
3425 | return 0; | |
3426 | ||
3427 | fail_qgroup: | |
3428 | btrfs_free_qgroup_config(fs_info); | |
3429 | fail_trans_kthread: | |
3430 | kthread_stop(fs_info->transaction_kthread); | |
3431 | btrfs_cleanup_transaction(fs_info); | |
3432 | btrfs_free_fs_roots(fs_info); | |
3433 | fail_cleaner: | |
3434 | kthread_stop(fs_info->cleaner_kthread); | |
3435 | ||
3436 | /* | |
3437 | * make sure we're done with the btree inode before we stop our | |
3438 | * kthreads | |
3439 | */ | |
3440 | filemap_write_and_wait(fs_info->btree_inode->i_mapping); | |
3441 | ||
3442 | fail_sysfs: | |
3443 | btrfs_sysfs_remove_mounted(fs_info); | |
3444 | ||
3445 | fail_fsdev_sysfs: | |
3446 | btrfs_sysfs_remove_fsid(fs_info->fs_devices); | |
3447 | ||
3448 | fail_block_groups: | |
3449 | btrfs_put_block_group_cache(fs_info); | |
3450 | ||
3451 | fail_tree_roots: | |
3452 | if (fs_info->data_reloc_root) | |
3453 | btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); | |
3454 | free_root_pointers(fs_info, true); | |
3455 | invalidate_inode_pages2(fs_info->btree_inode->i_mapping); | |
3456 | ||
3457 | fail_sb_buffer: | |
3458 | btrfs_stop_all_workers(fs_info); | |
3459 | btrfs_free_block_groups(fs_info); | |
3460 | fail_alloc: | |
3461 | btrfs_mapping_tree_free(&fs_info->mapping_tree); | |
3462 | ||
3463 | iput(fs_info->btree_inode); | |
3464 | fail: | |
3465 | btrfs_close_devices(fs_info->fs_devices); | |
3466 | return err; | |
3467 | } | |
3468 | ALLOW_ERROR_INJECTION(open_ctree, ERRNO); | |
3469 | ||
3470 | static void btrfs_end_super_write(struct bio *bio) | |
3471 | { | |
3472 | struct btrfs_device *device = bio->bi_private; | |
3473 | struct bio_vec *bvec; | |
3474 | struct bvec_iter_all iter_all; | |
3475 | struct page *page; | |
3476 | ||
3477 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
3478 | page = bvec->bv_page; | |
3479 | ||
3480 | if (bio->bi_status) { | |
3481 | btrfs_warn_rl_in_rcu(device->fs_info, | |
3482 | "lost page write due to IO error on %s (%d)", | |
3483 | rcu_str_deref(device->name), | |
3484 | blk_status_to_errno(bio->bi_status)); | |
3485 | ClearPageUptodate(page); | |
3486 | SetPageError(page); | |
3487 | btrfs_dev_stat_inc_and_print(device, | |
3488 | BTRFS_DEV_STAT_WRITE_ERRS); | |
3489 | } else { | |
3490 | SetPageUptodate(page); | |
3491 | } | |
3492 | ||
3493 | put_page(page); | |
3494 | unlock_page(page); | |
3495 | } | |
3496 | ||
3497 | bio_put(bio); | |
3498 | } | |
3499 | ||
3500 | struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, | |
3501 | int copy_num) | |
3502 | { | |
3503 | struct btrfs_super_block *super; | |
3504 | struct page *page; | |
3505 | u64 bytenr, bytenr_orig; | |
3506 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
3507 | int ret; | |
3508 | ||
3509 | bytenr_orig = btrfs_sb_offset(copy_num); | |
3510 | ret = btrfs_sb_log_location_bdev(bdev, copy_num, READ, &bytenr); | |
3511 | if (ret == -ENOENT) | |
3512 | return ERR_PTR(-EINVAL); | |
3513 | else if (ret) | |
3514 | return ERR_PTR(ret); | |
3515 | ||
3516 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) | |
3517 | return ERR_PTR(-EINVAL); | |
3518 | ||
3519 | page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); | |
3520 | if (IS_ERR(page)) | |
3521 | return ERR_CAST(page); | |
3522 | ||
3523 | super = page_address(page); | |
3524 | if (btrfs_super_magic(super) != BTRFS_MAGIC) { | |
3525 | btrfs_release_disk_super(super); | |
3526 | return ERR_PTR(-ENODATA); | |
3527 | } | |
3528 | ||
3529 | if (btrfs_super_bytenr(super) != bytenr_orig) { | |
3530 | btrfs_release_disk_super(super); | |
3531 | return ERR_PTR(-EINVAL); | |
3532 | } | |
3533 | ||
3534 | return super; | |
3535 | } | |
3536 | ||
3537 | ||
3538 | struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) | |
3539 | { | |
3540 | struct btrfs_super_block *super, *latest = NULL; | |
3541 | int i; | |
3542 | u64 transid = 0; | |
3543 | ||
3544 | /* we would like to check all the supers, but that would make | |
3545 | * a btrfs mount succeed after a mkfs from a different FS. | |
3546 | * So, we need to add a special mount option to scan for | |
3547 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
3548 | */ | |
3549 | for (i = 0; i < 1; i++) { | |
3550 | super = btrfs_read_dev_one_super(bdev, i); | |
3551 | if (IS_ERR(super)) | |
3552 | continue; | |
3553 | ||
3554 | if (!latest || btrfs_super_generation(super) > transid) { | |
3555 | if (latest) | |
3556 | btrfs_release_disk_super(super); | |
3557 | ||
3558 | latest = super; | |
3559 | transid = btrfs_super_generation(super); | |
3560 | } | |
3561 | } | |
3562 | ||
3563 | return super; | |
3564 | } | |
3565 | ||
3566 | /* | |
3567 | * Write superblock @sb to the @device. Do not wait for completion, all the | |
3568 | * pages we use for writing are locked. | |
3569 | * | |
3570 | * Write @max_mirrors copies of the superblock, where 0 means default that fit | |
3571 | * the expected device size at commit time. Note that max_mirrors must be | |
3572 | * same for write and wait phases. | |
3573 | * | |
3574 | * Return number of errors when page is not found or submission fails. | |
3575 | */ | |
3576 | static int write_dev_supers(struct btrfs_device *device, | |
3577 | struct btrfs_super_block *sb, int max_mirrors) | |
3578 | { | |
3579 | struct btrfs_fs_info *fs_info = device->fs_info; | |
3580 | struct address_space *mapping = device->bdev->bd_inode->i_mapping; | |
3581 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); | |
3582 | int i; | |
3583 | int errors = 0; | |
3584 | int ret; | |
3585 | u64 bytenr, bytenr_orig; | |
3586 | ||
3587 | if (max_mirrors == 0) | |
3588 | max_mirrors = BTRFS_SUPER_MIRROR_MAX; | |
3589 | ||
3590 | shash->tfm = fs_info->csum_shash; | |
3591 | ||
3592 | for (i = 0; i < max_mirrors; i++) { | |
3593 | struct page *page; | |
3594 | struct bio *bio; | |
3595 | struct btrfs_super_block *disk_super; | |
3596 | ||
3597 | bytenr_orig = btrfs_sb_offset(i); | |
3598 | ret = btrfs_sb_log_location(device, i, WRITE, &bytenr); | |
3599 | if (ret == -ENOENT) { | |
3600 | continue; | |
3601 | } else if (ret < 0) { | |
3602 | btrfs_err(device->fs_info, | |
3603 | "couldn't get super block location for mirror %d", | |
3604 | i); | |
3605 | errors++; | |
3606 | continue; | |
3607 | } | |
3608 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= | |
3609 | device->commit_total_bytes) | |
3610 | break; | |
3611 | ||
3612 | btrfs_set_super_bytenr(sb, bytenr_orig); | |
3613 | ||
3614 | crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, | |
3615 | BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, | |
3616 | sb->csum); | |
3617 | ||
3618 | page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, | |
3619 | GFP_NOFS); | |
3620 | if (!page) { | |
3621 | btrfs_err(device->fs_info, | |
3622 | "couldn't get super block page for bytenr %llu", | |
3623 | bytenr); | |
3624 | errors++; | |
3625 | continue; | |
3626 | } | |
3627 | ||
3628 | /* Bump the refcount for wait_dev_supers() */ | |
3629 | get_page(page); | |
3630 | ||
3631 | disk_super = page_address(page); | |
3632 | memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); | |
3633 | ||
3634 | /* | |
3635 | * Directly use bios here instead of relying on the page cache | |
3636 | * to do I/O, so we don't lose the ability to do integrity | |
3637 | * checking. | |
3638 | */ | |
3639 | bio = bio_alloc(GFP_NOFS, 1); | |
3640 | bio_set_dev(bio, device->bdev); | |
3641 | bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; | |
3642 | bio->bi_private = device; | |
3643 | bio->bi_end_io = btrfs_end_super_write; | |
3644 | __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, | |
3645 | offset_in_page(bytenr)); | |
3646 | ||
3647 | /* | |
3648 | * We FUA only the first super block. The others we allow to | |
3649 | * go down lazy and there's a short window where the on-disk | |
3650 | * copies might still contain the older version. | |
3651 | */ | |
3652 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO; | |
3653 | if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) | |
3654 | bio->bi_opf |= REQ_FUA; | |
3655 | ||
3656 | btrfsic_submit_bio(bio); | |
3657 | btrfs_advance_sb_log(device, i); | |
3658 | } | |
3659 | return errors < i ? 0 : -1; | |
3660 | } | |
3661 | ||
3662 | /* | |
3663 | * Wait for write completion of superblocks done by write_dev_supers, | |
3664 | * @max_mirrors same for write and wait phases. | |
3665 | * | |
3666 | * Return number of errors when page is not found or not marked up to | |
3667 | * date. | |
3668 | */ | |
3669 | static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) | |
3670 | { | |
3671 | int i; | |
3672 | int errors = 0; | |
3673 | bool primary_failed = false; | |
3674 | int ret; | |
3675 | u64 bytenr; | |
3676 | ||
3677 | if (max_mirrors == 0) | |
3678 | max_mirrors = BTRFS_SUPER_MIRROR_MAX; | |
3679 | ||
3680 | for (i = 0; i < max_mirrors; i++) { | |
3681 | struct page *page; | |
3682 | ||
3683 | ret = btrfs_sb_log_location(device, i, READ, &bytenr); | |
3684 | if (ret == -ENOENT) { | |
3685 | break; | |
3686 | } else if (ret < 0) { | |
3687 | errors++; | |
3688 | if (i == 0) | |
3689 | primary_failed = true; | |
3690 | continue; | |
3691 | } | |
3692 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= | |
3693 | device->commit_total_bytes) | |
3694 | break; | |
3695 | ||
3696 | page = find_get_page(device->bdev->bd_inode->i_mapping, | |
3697 | bytenr >> PAGE_SHIFT); | |
3698 | if (!page) { | |
3699 | errors++; | |
3700 | if (i == 0) | |
3701 | primary_failed = true; | |
3702 | continue; | |
3703 | } | |
3704 | /* Page is submitted locked and unlocked once the IO completes */ | |
3705 | wait_on_page_locked(page); | |
3706 | if (PageError(page)) { | |
3707 | errors++; | |
3708 | if (i == 0) | |
3709 | primary_failed = true; | |
3710 | } | |
3711 | ||
3712 | /* Drop our reference */ | |
3713 | put_page(page); | |
3714 | ||
3715 | /* Drop the reference from the writing run */ | |
3716 | put_page(page); | |
3717 | } | |
3718 | ||
3719 | /* log error, force error return */ | |
3720 | if (primary_failed) { | |
3721 | btrfs_err(device->fs_info, "error writing primary super block to device %llu", | |
3722 | device->devid); | |
3723 | return -1; | |
3724 | } | |
3725 | ||
3726 | return errors < i ? 0 : -1; | |
3727 | } | |
3728 | ||
3729 | /* | |
3730 | * endio for the write_dev_flush, this will wake anyone waiting | |
3731 | * for the barrier when it is done | |
3732 | */ | |
3733 | static void btrfs_end_empty_barrier(struct bio *bio) | |
3734 | { | |
3735 | complete(bio->bi_private); | |
3736 | } | |
3737 | ||
3738 | /* | |
3739 | * Submit a flush request to the device if it supports it. Error handling is | |
3740 | * done in the waiting counterpart. | |
3741 | */ | |
3742 | static void write_dev_flush(struct btrfs_device *device) | |
3743 | { | |
3744 | struct request_queue *q = bdev_get_queue(device->bdev); | |
3745 | struct bio *bio = device->flush_bio; | |
3746 | ||
3747 | if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) | |
3748 | return; | |
3749 | ||
3750 | bio_reset(bio); | |
3751 | bio->bi_end_io = btrfs_end_empty_barrier; | |
3752 | bio_set_dev(bio, device->bdev); | |
3753 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; | |
3754 | init_completion(&device->flush_wait); | |
3755 | bio->bi_private = &device->flush_wait; | |
3756 | ||
3757 | btrfsic_submit_bio(bio); | |
3758 | set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); | |
3759 | } | |
3760 | ||
3761 | /* | |
3762 | * If the flush bio has been submitted by write_dev_flush, wait for it. | |
3763 | */ | |
3764 | static blk_status_t wait_dev_flush(struct btrfs_device *device) | |
3765 | { | |
3766 | struct bio *bio = device->flush_bio; | |
3767 | ||
3768 | if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) | |
3769 | return BLK_STS_OK; | |
3770 | ||
3771 | clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); | |
3772 | wait_for_completion_io(&device->flush_wait); | |
3773 | ||
3774 | return bio->bi_status; | |
3775 | } | |
3776 | ||
3777 | static int check_barrier_error(struct btrfs_fs_info *fs_info) | |
3778 | { | |
3779 | if (!btrfs_check_rw_degradable(fs_info, NULL)) | |
3780 | return -EIO; | |
3781 | return 0; | |
3782 | } | |
3783 | ||
3784 | /* | |
3785 | * send an empty flush down to each device in parallel, | |
3786 | * then wait for them | |
3787 | */ | |
3788 | static int barrier_all_devices(struct btrfs_fs_info *info) | |
3789 | { | |
3790 | struct list_head *head; | |
3791 | struct btrfs_device *dev; | |
3792 | int errors_wait = 0; | |
3793 | blk_status_t ret; | |
3794 | ||
3795 | lockdep_assert_held(&info->fs_devices->device_list_mutex); | |
3796 | /* send down all the barriers */ | |
3797 | head = &info->fs_devices->devices; | |
3798 | list_for_each_entry(dev, head, dev_list) { | |
3799 | if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) | |
3800 | continue; | |
3801 | if (!dev->bdev) | |
3802 | continue; | |
3803 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || | |
3804 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) | |
3805 | continue; | |
3806 | ||
3807 | write_dev_flush(dev); | |
3808 | dev->last_flush_error = BLK_STS_OK; | |
3809 | } | |
3810 | ||
3811 | /* wait for all the barriers */ | |
3812 | list_for_each_entry(dev, head, dev_list) { | |
3813 | if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) | |
3814 | continue; | |
3815 | if (!dev->bdev) { | |
3816 | errors_wait++; | |
3817 | continue; | |
3818 | } | |
3819 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || | |
3820 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) | |
3821 | continue; | |
3822 | ||
3823 | ret = wait_dev_flush(dev); | |
3824 | if (ret) { | |
3825 | dev->last_flush_error = ret; | |
3826 | btrfs_dev_stat_inc_and_print(dev, | |
3827 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
3828 | errors_wait++; | |
3829 | } | |
3830 | } | |
3831 | ||
3832 | if (errors_wait) { | |
3833 | /* | |
3834 | * At some point we need the status of all disks | |
3835 | * to arrive at the volume status. So error checking | |
3836 | * is being pushed to a separate loop. | |
3837 | */ | |
3838 | return check_barrier_error(info); | |
3839 | } | |
3840 | return 0; | |
3841 | } | |
3842 | ||
3843 | int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) | |
3844 | { | |
3845 | int raid_type; | |
3846 | int min_tolerated = INT_MAX; | |
3847 | ||
3848 | if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || | |
3849 | (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) | |
3850 | min_tolerated = min_t(int, min_tolerated, | |
3851 | btrfs_raid_array[BTRFS_RAID_SINGLE]. | |
3852 | tolerated_failures); | |
3853 | ||
3854 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { | |
3855 | if (raid_type == BTRFS_RAID_SINGLE) | |
3856 | continue; | |
3857 | if (!(flags & btrfs_raid_array[raid_type].bg_flag)) | |
3858 | continue; | |
3859 | min_tolerated = min_t(int, min_tolerated, | |
3860 | btrfs_raid_array[raid_type]. | |
3861 | tolerated_failures); | |
3862 | } | |
3863 | ||
3864 | if (min_tolerated == INT_MAX) { | |
3865 | pr_warn("BTRFS: unknown raid flag: %llu", flags); | |
3866 | min_tolerated = 0; | |
3867 | } | |
3868 | ||
3869 | return min_tolerated; | |
3870 | } | |
3871 | ||
3872 | int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) | |
3873 | { | |
3874 | struct list_head *head; | |
3875 | struct btrfs_device *dev; | |
3876 | struct btrfs_super_block *sb; | |
3877 | struct btrfs_dev_item *dev_item; | |
3878 | int ret; | |
3879 | int do_barriers; | |
3880 | int max_errors; | |
3881 | int total_errors = 0; | |
3882 | u64 flags; | |
3883 | ||
3884 | do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); | |
3885 | ||
3886 | /* | |
3887 | * max_mirrors == 0 indicates we're from commit_transaction, | |
3888 | * not from fsync where the tree roots in fs_info have not | |
3889 | * been consistent on disk. | |
3890 | */ | |
3891 | if (max_mirrors == 0) | |
3892 | backup_super_roots(fs_info); | |
3893 | ||
3894 | sb = fs_info->super_for_commit; | |
3895 | dev_item = &sb->dev_item; | |
3896 | ||
3897 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
3898 | head = &fs_info->fs_devices->devices; | |
3899 | max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; | |
3900 | ||
3901 | if (do_barriers) { | |
3902 | ret = barrier_all_devices(fs_info); | |
3903 | if (ret) { | |
3904 | mutex_unlock( | |
3905 | &fs_info->fs_devices->device_list_mutex); | |
3906 | btrfs_handle_fs_error(fs_info, ret, | |
3907 | "errors while submitting device barriers."); | |
3908 | return ret; | |
3909 | } | |
3910 | } | |
3911 | ||
3912 | list_for_each_entry(dev, head, dev_list) { | |
3913 | if (!dev->bdev) { | |
3914 | total_errors++; | |
3915 | continue; | |
3916 | } | |
3917 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || | |
3918 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) | |
3919 | continue; | |
3920 | ||
3921 | btrfs_set_stack_device_generation(dev_item, 0); | |
3922 | btrfs_set_stack_device_type(dev_item, dev->type); | |
3923 | btrfs_set_stack_device_id(dev_item, dev->devid); | |
3924 | btrfs_set_stack_device_total_bytes(dev_item, | |
3925 | dev->commit_total_bytes); | |
3926 | btrfs_set_stack_device_bytes_used(dev_item, | |
3927 | dev->commit_bytes_used); | |
3928 | btrfs_set_stack_device_io_align(dev_item, dev->io_align); | |
3929 | btrfs_set_stack_device_io_width(dev_item, dev->io_width); | |
3930 | btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); | |
3931 | memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); | |
3932 | memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, | |
3933 | BTRFS_FSID_SIZE); | |
3934 | ||
3935 | flags = btrfs_super_flags(sb); | |
3936 | btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); | |
3937 | ||
3938 | ret = btrfs_validate_write_super(fs_info, sb); | |
3939 | if (ret < 0) { | |
3940 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
3941 | btrfs_handle_fs_error(fs_info, -EUCLEAN, | |
3942 | "unexpected superblock corruption detected"); | |
3943 | return -EUCLEAN; | |
3944 | } | |
3945 | ||
3946 | ret = write_dev_supers(dev, sb, max_mirrors); | |
3947 | if (ret) | |
3948 | total_errors++; | |
3949 | } | |
3950 | if (total_errors > max_errors) { | |
3951 | btrfs_err(fs_info, "%d errors while writing supers", | |
3952 | total_errors); | |
3953 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
3954 | ||
3955 | /* FUA is masked off if unsupported and can't be the reason */ | |
3956 | btrfs_handle_fs_error(fs_info, -EIO, | |
3957 | "%d errors while writing supers", | |
3958 | total_errors); | |
3959 | return -EIO; | |
3960 | } | |
3961 | ||
3962 | total_errors = 0; | |
3963 | list_for_each_entry(dev, head, dev_list) { | |
3964 | if (!dev->bdev) | |
3965 | continue; | |
3966 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || | |
3967 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) | |
3968 | continue; | |
3969 | ||
3970 | ret = wait_dev_supers(dev, max_mirrors); | |
3971 | if (ret) | |
3972 | total_errors++; | |
3973 | } | |
3974 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
3975 | if (total_errors > max_errors) { | |
3976 | btrfs_handle_fs_error(fs_info, -EIO, | |
3977 | "%d errors while writing supers", | |
3978 | total_errors); | |
3979 | return -EIO; | |
3980 | } | |
3981 | return 0; | |
3982 | } | |
3983 | ||
3984 | /* Drop a fs root from the radix tree and free it. */ | |
3985 | void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, | |
3986 | struct btrfs_root *root) | |
3987 | { | |
3988 | bool drop_ref = false; | |
3989 | ||
3990 | spin_lock(&fs_info->fs_roots_radix_lock); | |
3991 | radix_tree_delete(&fs_info->fs_roots_radix, | |
3992 | (unsigned long)root->root_key.objectid); | |
3993 | if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) | |
3994 | drop_ref = true; | |
3995 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
3996 | ||
3997 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { | |
3998 | ASSERT(root->log_root == NULL); | |
3999 | if (root->reloc_root) { | |
4000 | btrfs_put_root(root->reloc_root); | |
4001 | root->reloc_root = NULL; | |
4002 | } | |
4003 | } | |
4004 | ||
4005 | if (drop_ref) | |
4006 | btrfs_put_root(root); | |
4007 | } | |
4008 | ||
4009 | int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) | |
4010 | { | |
4011 | u64 root_objectid = 0; | |
4012 | struct btrfs_root *gang[8]; | |
4013 | int i = 0; | |
4014 | int err = 0; | |
4015 | unsigned int ret = 0; | |
4016 | ||
4017 | while (1) { | |
4018 | spin_lock(&fs_info->fs_roots_radix_lock); | |
4019 | ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, | |
4020 | (void **)gang, root_objectid, | |
4021 | ARRAY_SIZE(gang)); | |
4022 | if (!ret) { | |
4023 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
4024 | break; | |
4025 | } | |
4026 | root_objectid = gang[ret - 1]->root_key.objectid + 1; | |
4027 | ||
4028 | for (i = 0; i < ret; i++) { | |
4029 | /* Avoid to grab roots in dead_roots */ | |
4030 | if (btrfs_root_refs(&gang[i]->root_item) == 0) { | |
4031 | gang[i] = NULL; | |
4032 | continue; | |
4033 | } | |
4034 | /* grab all the search result for later use */ | |
4035 | gang[i] = btrfs_grab_root(gang[i]); | |
4036 | } | |
4037 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
4038 | ||
4039 | for (i = 0; i < ret; i++) { | |
4040 | if (!gang[i]) | |
4041 | continue; | |
4042 | root_objectid = gang[i]->root_key.objectid; | |
4043 | err = btrfs_orphan_cleanup(gang[i]); | |
4044 | if (err) | |
4045 | break; | |
4046 | btrfs_put_root(gang[i]); | |
4047 | } | |
4048 | root_objectid++; | |
4049 | } | |
4050 | ||
4051 | /* release the uncleaned roots due to error */ | |
4052 | for (; i < ret; i++) { | |
4053 | if (gang[i]) | |
4054 | btrfs_put_root(gang[i]); | |
4055 | } | |
4056 | return err; | |
4057 | } | |
4058 | ||
4059 | int btrfs_commit_super(struct btrfs_fs_info *fs_info) | |
4060 | { | |
4061 | struct btrfs_root *root = fs_info->tree_root; | |
4062 | struct btrfs_trans_handle *trans; | |
4063 | ||
4064 | mutex_lock(&fs_info->cleaner_mutex); | |
4065 | btrfs_run_delayed_iputs(fs_info); | |
4066 | mutex_unlock(&fs_info->cleaner_mutex); | |
4067 | wake_up_process(fs_info->cleaner_kthread); | |
4068 | ||
4069 | /* wait until ongoing cleanup work done */ | |
4070 | down_write(&fs_info->cleanup_work_sem); | |
4071 | up_write(&fs_info->cleanup_work_sem); | |
4072 | ||
4073 | trans = btrfs_join_transaction(root); | |
4074 | if (IS_ERR(trans)) | |
4075 | return PTR_ERR(trans); | |
4076 | return btrfs_commit_transaction(trans); | |
4077 | } | |
4078 | ||
4079 | void __cold close_ctree(struct btrfs_fs_info *fs_info) | |
4080 | { | |
4081 | int ret; | |
4082 | ||
4083 | set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); | |
4084 | /* | |
4085 | * We don't want the cleaner to start new transactions, add more delayed | |
4086 | * iputs, etc. while we're closing. We can't use kthread_stop() yet | |
4087 | * because that frees the task_struct, and the transaction kthread might | |
4088 | * still try to wake up the cleaner. | |
4089 | */ | |
4090 | kthread_park(fs_info->cleaner_kthread); | |
4091 | ||
4092 | /* wait for the qgroup rescan worker to stop */ | |
4093 | btrfs_qgroup_wait_for_completion(fs_info, false); | |
4094 | ||
4095 | /* wait for the uuid_scan task to finish */ | |
4096 | down(&fs_info->uuid_tree_rescan_sem); | |
4097 | /* avoid complains from lockdep et al., set sem back to initial state */ | |
4098 | up(&fs_info->uuid_tree_rescan_sem); | |
4099 | ||
4100 | /* pause restriper - we want to resume on mount */ | |
4101 | btrfs_pause_balance(fs_info); | |
4102 | ||
4103 | btrfs_dev_replace_suspend_for_unmount(fs_info); | |
4104 | ||
4105 | btrfs_scrub_cancel(fs_info); | |
4106 | ||
4107 | /* wait for any defraggers to finish */ | |
4108 | wait_event(fs_info->transaction_wait, | |
4109 | (atomic_read(&fs_info->defrag_running) == 0)); | |
4110 | ||
4111 | /* clear out the rbtree of defraggable inodes */ | |
4112 | btrfs_cleanup_defrag_inodes(fs_info); | |
4113 | ||
4114 | cancel_work_sync(&fs_info->async_reclaim_work); | |
4115 | cancel_work_sync(&fs_info->async_data_reclaim_work); | |
4116 | ||
4117 | /* Cancel or finish ongoing discard work */ | |
4118 | btrfs_discard_cleanup(fs_info); | |
4119 | ||
4120 | if (!sb_rdonly(fs_info->sb)) { | |
4121 | /* | |
4122 | * The cleaner kthread is stopped, so do one final pass over | |
4123 | * unused block groups. | |
4124 | */ | |
4125 | btrfs_delete_unused_bgs(fs_info); | |
4126 | ||
4127 | /* | |
4128 | * There might be existing delayed inode workers still running | |
4129 | * and holding an empty delayed inode item. We must wait for | |
4130 | * them to complete first because they can create a transaction. | |
4131 | * This happens when someone calls btrfs_balance_delayed_items() | |
4132 | * and then a transaction commit runs the same delayed nodes | |
4133 | * before any delayed worker has done something with the nodes. | |
4134 | * We must wait for any worker here and not at transaction | |
4135 | * commit time since that could cause a deadlock. | |
4136 | * This is a very rare case. | |
4137 | */ | |
4138 | btrfs_flush_workqueue(fs_info->delayed_workers); | |
4139 | ||
4140 | ret = btrfs_commit_super(fs_info); | |
4141 | if (ret) | |
4142 | btrfs_err(fs_info, "commit super ret %d", ret); | |
4143 | } | |
4144 | ||
4145 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || | |
4146 | test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) | |
4147 | btrfs_error_commit_super(fs_info); | |
4148 | ||
4149 | kthread_stop(fs_info->transaction_kthread); | |
4150 | kthread_stop(fs_info->cleaner_kthread); | |
4151 | ||
4152 | ASSERT(list_empty(&fs_info->delayed_iputs)); | |
4153 | set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); | |
4154 | ||
4155 | if (btrfs_check_quota_leak(fs_info)) { | |
4156 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
4157 | btrfs_err(fs_info, "qgroup reserved space leaked"); | |
4158 | } | |
4159 | ||
4160 | btrfs_free_qgroup_config(fs_info); | |
4161 | ASSERT(list_empty(&fs_info->delalloc_roots)); | |
4162 | ||
4163 | if (percpu_counter_sum(&fs_info->delalloc_bytes)) { | |
4164 | btrfs_info(fs_info, "at unmount delalloc count %lld", | |
4165 | percpu_counter_sum(&fs_info->delalloc_bytes)); | |
4166 | } | |
4167 | ||
4168 | if (percpu_counter_sum(&fs_info->dio_bytes)) | |
4169 | btrfs_info(fs_info, "at unmount dio bytes count %lld", | |
4170 | percpu_counter_sum(&fs_info->dio_bytes)); | |
4171 | ||
4172 | btrfs_sysfs_remove_mounted(fs_info); | |
4173 | btrfs_sysfs_remove_fsid(fs_info->fs_devices); | |
4174 | ||
4175 | btrfs_put_block_group_cache(fs_info); | |
4176 | ||
4177 | /* | |
4178 | * we must make sure there is not any read request to | |
4179 | * submit after we stopping all workers. | |
4180 | */ | |
4181 | invalidate_inode_pages2(fs_info->btree_inode->i_mapping); | |
4182 | btrfs_stop_all_workers(fs_info); | |
4183 | ||
4184 | /* We shouldn't have any transaction open at this point */ | |
4185 | ASSERT(list_empty(&fs_info->trans_list)); | |
4186 | ||
4187 | clear_bit(BTRFS_FS_OPEN, &fs_info->flags); | |
4188 | free_root_pointers(fs_info, true); | |
4189 | btrfs_free_fs_roots(fs_info); | |
4190 | ||
4191 | /* | |
4192 | * We must free the block groups after dropping the fs_roots as we could | |
4193 | * have had an IO error and have left over tree log blocks that aren't | |
4194 | * cleaned up until the fs roots are freed. This makes the block group | |
4195 | * accounting appear to be wrong because there's pending reserved bytes, | |
4196 | * so make sure we do the block group cleanup afterwards. | |
4197 | */ | |
4198 | btrfs_free_block_groups(fs_info); | |
4199 | ||
4200 | iput(fs_info->btree_inode); | |
4201 | ||
4202 | #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY | |
4203 | if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) | |
4204 | btrfsic_unmount(fs_info->fs_devices); | |
4205 | #endif | |
4206 | ||
4207 | btrfs_mapping_tree_free(&fs_info->mapping_tree); | |
4208 | btrfs_close_devices(fs_info->fs_devices); | |
4209 | } | |
4210 | ||
4211 | int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, | |
4212 | int atomic) | |
4213 | { | |
4214 | int ret; | |
4215 | struct inode *btree_inode = buf->pages[0]->mapping->host; | |
4216 | ||
4217 | ret = extent_buffer_uptodate(buf); | |
4218 | if (!ret) | |
4219 | return ret; | |
4220 | ||
4221 | ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, | |
4222 | parent_transid, atomic); | |
4223 | if (ret == -EAGAIN) | |
4224 | return ret; | |
4225 | return !ret; | |
4226 | } | |
4227 | ||
4228 | void btrfs_mark_buffer_dirty(struct extent_buffer *buf) | |
4229 | { | |
4230 | struct btrfs_fs_info *fs_info = buf->fs_info; | |
4231 | u64 transid = btrfs_header_generation(buf); | |
4232 | int was_dirty; | |
4233 | ||
4234 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS | |
4235 | /* | |
4236 | * This is a fast path so only do this check if we have sanity tests | |
4237 | * enabled. Normal people shouldn't be using unmapped buffers as dirty | |
4238 | * outside of the sanity tests. | |
4239 | */ | |
4240 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) | |
4241 | return; | |
4242 | #endif | |
4243 | btrfs_assert_tree_locked(buf); | |
4244 | if (transid != fs_info->generation) | |
4245 | WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", | |
4246 | buf->start, transid, fs_info->generation); | |
4247 | was_dirty = set_extent_buffer_dirty(buf); | |
4248 | if (!was_dirty) | |
4249 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, | |
4250 | buf->len, | |
4251 | fs_info->dirty_metadata_batch); | |
4252 | #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY | |
4253 | /* | |
4254 | * Since btrfs_mark_buffer_dirty() can be called with item pointer set | |
4255 | * but item data not updated. | |
4256 | * So here we should only check item pointers, not item data. | |
4257 | */ | |
4258 | if (btrfs_header_level(buf) == 0 && | |
4259 | btrfs_check_leaf_relaxed(buf)) { | |
4260 | btrfs_print_leaf(buf); | |
4261 | ASSERT(0); | |
4262 | } | |
4263 | #endif | |
4264 | } | |
4265 | ||
4266 | static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, | |
4267 | int flush_delayed) | |
4268 | { | |
4269 | /* | |
4270 | * looks as though older kernels can get into trouble with | |
4271 | * this code, they end up stuck in balance_dirty_pages forever | |
4272 | */ | |
4273 | int ret; | |
4274 | ||
4275 | if (current->flags & PF_MEMALLOC) | |
4276 | return; | |
4277 | ||
4278 | if (flush_delayed) | |
4279 | btrfs_balance_delayed_items(fs_info); | |
4280 | ||
4281 | ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, | |
4282 | BTRFS_DIRTY_METADATA_THRESH, | |
4283 | fs_info->dirty_metadata_batch); | |
4284 | if (ret > 0) { | |
4285 | balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); | |
4286 | } | |
4287 | } | |
4288 | ||
4289 | void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) | |
4290 | { | |
4291 | __btrfs_btree_balance_dirty(fs_info, 1); | |
4292 | } | |
4293 | ||
4294 | void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) | |
4295 | { | |
4296 | __btrfs_btree_balance_dirty(fs_info, 0); | |
4297 | } | |
4298 | ||
4299 | int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, | |
4300 | struct btrfs_key *first_key) | |
4301 | { | |
4302 | return btree_read_extent_buffer_pages(buf, parent_transid, | |
4303 | level, first_key); | |
4304 | } | |
4305 | ||
4306 | static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) | |
4307 | { | |
4308 | /* cleanup FS via transaction */ | |
4309 | btrfs_cleanup_transaction(fs_info); | |
4310 | ||
4311 | mutex_lock(&fs_info->cleaner_mutex); | |
4312 | btrfs_run_delayed_iputs(fs_info); | |
4313 | mutex_unlock(&fs_info->cleaner_mutex); | |
4314 | ||
4315 | down_write(&fs_info->cleanup_work_sem); | |
4316 | up_write(&fs_info->cleanup_work_sem); | |
4317 | } | |
4318 | ||
4319 | static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) | |
4320 | { | |
4321 | struct btrfs_root *gang[8]; | |
4322 | u64 root_objectid = 0; | |
4323 | int ret; | |
4324 | ||
4325 | spin_lock(&fs_info->fs_roots_radix_lock); | |
4326 | while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, | |
4327 | (void **)gang, root_objectid, | |
4328 | ARRAY_SIZE(gang))) != 0) { | |
4329 | int i; | |
4330 | ||
4331 | for (i = 0; i < ret; i++) | |
4332 | gang[i] = btrfs_grab_root(gang[i]); | |
4333 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
4334 | ||
4335 | for (i = 0; i < ret; i++) { | |
4336 | if (!gang[i]) | |
4337 | continue; | |
4338 | root_objectid = gang[i]->root_key.objectid; | |
4339 | btrfs_free_log(NULL, gang[i]); | |
4340 | btrfs_put_root(gang[i]); | |
4341 | } | |
4342 | root_objectid++; | |
4343 | spin_lock(&fs_info->fs_roots_radix_lock); | |
4344 | } | |
4345 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
4346 | btrfs_free_log_root_tree(NULL, fs_info); | |
4347 | } | |
4348 | ||
4349 | static void btrfs_destroy_ordered_extents(struct btrfs_root *root) | |
4350 | { | |
4351 | struct btrfs_ordered_extent *ordered; | |
4352 | ||
4353 | spin_lock(&root->ordered_extent_lock); | |
4354 | /* | |
4355 | * This will just short circuit the ordered completion stuff which will | |
4356 | * make sure the ordered extent gets properly cleaned up. | |
4357 | */ | |
4358 | list_for_each_entry(ordered, &root->ordered_extents, | |
4359 | root_extent_list) | |
4360 | set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); | |
4361 | spin_unlock(&root->ordered_extent_lock); | |
4362 | } | |
4363 | ||
4364 | static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) | |
4365 | { | |
4366 | struct btrfs_root *root; | |
4367 | struct list_head splice; | |
4368 | ||
4369 | INIT_LIST_HEAD(&splice); | |
4370 | ||
4371 | spin_lock(&fs_info->ordered_root_lock); | |
4372 | list_splice_init(&fs_info->ordered_roots, &splice); | |
4373 | while (!list_empty(&splice)) { | |
4374 | root = list_first_entry(&splice, struct btrfs_root, | |
4375 | ordered_root); | |
4376 | list_move_tail(&root->ordered_root, | |
4377 | &fs_info->ordered_roots); | |
4378 | ||
4379 | spin_unlock(&fs_info->ordered_root_lock); | |
4380 | btrfs_destroy_ordered_extents(root); | |
4381 | ||
4382 | cond_resched(); | |
4383 | spin_lock(&fs_info->ordered_root_lock); | |
4384 | } | |
4385 | spin_unlock(&fs_info->ordered_root_lock); | |
4386 | ||
4387 | /* | |
4388 | * We need this here because if we've been flipped read-only we won't | |
4389 | * get sync() from the umount, so we need to make sure any ordered | |
4390 | * extents that haven't had their dirty pages IO start writeout yet | |
4391 | * actually get run and error out properly. | |
4392 | */ | |
4393 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); | |
4394 | } | |
4395 | ||
4396 | static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, | |
4397 | struct btrfs_fs_info *fs_info) | |
4398 | { | |
4399 | struct rb_node *node; | |
4400 | struct btrfs_delayed_ref_root *delayed_refs; | |
4401 | struct btrfs_delayed_ref_node *ref; | |
4402 | int ret = 0; | |
4403 | ||
4404 | delayed_refs = &trans->delayed_refs; | |
4405 | ||
4406 | spin_lock(&delayed_refs->lock); | |
4407 | if (atomic_read(&delayed_refs->num_entries) == 0) { | |
4408 | spin_unlock(&delayed_refs->lock); | |
4409 | btrfs_debug(fs_info, "delayed_refs has NO entry"); | |
4410 | return ret; | |
4411 | } | |
4412 | ||
4413 | while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { | |
4414 | struct btrfs_delayed_ref_head *head; | |
4415 | struct rb_node *n; | |
4416 | bool pin_bytes = false; | |
4417 | ||
4418 | head = rb_entry(node, struct btrfs_delayed_ref_head, | |
4419 | href_node); | |
4420 | if (btrfs_delayed_ref_lock(delayed_refs, head)) | |
4421 | continue; | |
4422 | ||
4423 | spin_lock(&head->lock); | |
4424 | while ((n = rb_first_cached(&head->ref_tree)) != NULL) { | |
4425 | ref = rb_entry(n, struct btrfs_delayed_ref_node, | |
4426 | ref_node); | |
4427 | ref->in_tree = 0; | |
4428 | rb_erase_cached(&ref->ref_node, &head->ref_tree); | |
4429 | RB_CLEAR_NODE(&ref->ref_node); | |
4430 | if (!list_empty(&ref->add_list)) | |
4431 | list_del(&ref->add_list); | |
4432 | atomic_dec(&delayed_refs->num_entries); | |
4433 | btrfs_put_delayed_ref(ref); | |
4434 | } | |
4435 | if (head->must_insert_reserved) | |
4436 | pin_bytes = true; | |
4437 | btrfs_free_delayed_extent_op(head->extent_op); | |
4438 | btrfs_delete_ref_head(delayed_refs, head); | |
4439 | spin_unlock(&head->lock); | |
4440 | spin_unlock(&delayed_refs->lock); | |
4441 | mutex_unlock(&head->mutex); | |
4442 | ||
4443 | if (pin_bytes) { | |
4444 | struct btrfs_block_group *cache; | |
4445 | ||
4446 | cache = btrfs_lookup_block_group(fs_info, head->bytenr); | |
4447 | BUG_ON(!cache); | |
4448 | ||
4449 | spin_lock(&cache->space_info->lock); | |
4450 | spin_lock(&cache->lock); | |
4451 | cache->pinned += head->num_bytes; | |
4452 | btrfs_space_info_update_bytes_pinned(fs_info, | |
4453 | cache->space_info, head->num_bytes); | |
4454 | cache->reserved -= head->num_bytes; | |
4455 | cache->space_info->bytes_reserved -= head->num_bytes; | |
4456 | spin_unlock(&cache->lock); | |
4457 | spin_unlock(&cache->space_info->lock); | |
4458 | percpu_counter_add_batch( | |
4459 | &cache->space_info->total_bytes_pinned, | |
4460 | head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
4461 | ||
4462 | btrfs_put_block_group(cache); | |
4463 | ||
4464 | btrfs_error_unpin_extent_range(fs_info, head->bytenr, | |
4465 | head->bytenr + head->num_bytes - 1); | |
4466 | } | |
4467 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); | |
4468 | btrfs_put_delayed_ref_head(head); | |
4469 | cond_resched(); | |
4470 | spin_lock(&delayed_refs->lock); | |
4471 | } | |
4472 | btrfs_qgroup_destroy_extent_records(trans); | |
4473 | ||
4474 | spin_unlock(&delayed_refs->lock); | |
4475 | ||
4476 | return ret; | |
4477 | } | |
4478 | ||
4479 | static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) | |
4480 | { | |
4481 | struct btrfs_inode *btrfs_inode; | |
4482 | struct list_head splice; | |
4483 | ||
4484 | INIT_LIST_HEAD(&splice); | |
4485 | ||
4486 | spin_lock(&root->delalloc_lock); | |
4487 | list_splice_init(&root->delalloc_inodes, &splice); | |
4488 | ||
4489 | while (!list_empty(&splice)) { | |
4490 | struct inode *inode = NULL; | |
4491 | btrfs_inode = list_first_entry(&splice, struct btrfs_inode, | |
4492 | delalloc_inodes); | |
4493 | __btrfs_del_delalloc_inode(root, btrfs_inode); | |
4494 | spin_unlock(&root->delalloc_lock); | |
4495 | ||
4496 | /* | |
4497 | * Make sure we get a live inode and that it'll not disappear | |
4498 | * meanwhile. | |
4499 | */ | |
4500 | inode = igrab(&btrfs_inode->vfs_inode); | |
4501 | if (inode) { | |
4502 | invalidate_inode_pages2(inode->i_mapping); | |
4503 | iput(inode); | |
4504 | } | |
4505 | spin_lock(&root->delalloc_lock); | |
4506 | } | |
4507 | spin_unlock(&root->delalloc_lock); | |
4508 | } | |
4509 | ||
4510 | static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) | |
4511 | { | |
4512 | struct btrfs_root *root; | |
4513 | struct list_head splice; | |
4514 | ||
4515 | INIT_LIST_HEAD(&splice); | |
4516 | ||
4517 | spin_lock(&fs_info->delalloc_root_lock); | |
4518 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
4519 | while (!list_empty(&splice)) { | |
4520 | root = list_first_entry(&splice, struct btrfs_root, | |
4521 | delalloc_root); | |
4522 | root = btrfs_grab_root(root); | |
4523 | BUG_ON(!root); | |
4524 | spin_unlock(&fs_info->delalloc_root_lock); | |
4525 | ||
4526 | btrfs_destroy_delalloc_inodes(root); | |
4527 | btrfs_put_root(root); | |
4528 | ||
4529 | spin_lock(&fs_info->delalloc_root_lock); | |
4530 | } | |
4531 | spin_unlock(&fs_info->delalloc_root_lock); | |
4532 | } | |
4533 | ||
4534 | static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, | |
4535 | struct extent_io_tree *dirty_pages, | |
4536 | int mark) | |
4537 | { | |
4538 | int ret; | |
4539 | struct extent_buffer *eb; | |
4540 | u64 start = 0; | |
4541 | u64 end; | |
4542 | ||
4543 | while (1) { | |
4544 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, | |
4545 | mark, NULL); | |
4546 | if (ret) | |
4547 | break; | |
4548 | ||
4549 | clear_extent_bits(dirty_pages, start, end, mark); | |
4550 | while (start <= end) { | |
4551 | eb = find_extent_buffer(fs_info, start); | |
4552 | start += fs_info->nodesize; | |
4553 | if (!eb) | |
4554 | continue; | |
4555 | wait_on_extent_buffer_writeback(eb); | |
4556 | ||
4557 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, | |
4558 | &eb->bflags)) | |
4559 | clear_extent_buffer_dirty(eb); | |
4560 | free_extent_buffer_stale(eb); | |
4561 | } | |
4562 | } | |
4563 | ||
4564 | return ret; | |
4565 | } | |
4566 | ||
4567 | static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, | |
4568 | struct extent_io_tree *unpin) | |
4569 | { | |
4570 | u64 start; | |
4571 | u64 end; | |
4572 | int ret; | |
4573 | ||
4574 | while (1) { | |
4575 | struct extent_state *cached_state = NULL; | |
4576 | ||
4577 | /* | |
4578 | * The btrfs_finish_extent_commit() may get the same range as | |
4579 | * ours between find_first_extent_bit and clear_extent_dirty. | |
4580 | * Hence, hold the unused_bg_unpin_mutex to avoid double unpin | |
4581 | * the same extent range. | |
4582 | */ | |
4583 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
4584 | ret = find_first_extent_bit(unpin, 0, &start, &end, | |
4585 | EXTENT_DIRTY, &cached_state); | |
4586 | if (ret) { | |
4587 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
4588 | break; | |
4589 | } | |
4590 | ||
4591 | clear_extent_dirty(unpin, start, end, &cached_state); | |
4592 | free_extent_state(cached_state); | |
4593 | btrfs_error_unpin_extent_range(fs_info, start, end); | |
4594 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
4595 | cond_resched(); | |
4596 | } | |
4597 | ||
4598 | return 0; | |
4599 | } | |
4600 | ||
4601 | static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) | |
4602 | { | |
4603 | struct inode *inode; | |
4604 | ||
4605 | inode = cache->io_ctl.inode; | |
4606 | if (inode) { | |
4607 | invalidate_inode_pages2(inode->i_mapping); | |
4608 | BTRFS_I(inode)->generation = 0; | |
4609 | cache->io_ctl.inode = NULL; | |
4610 | iput(inode); | |
4611 | } | |
4612 | ASSERT(cache->io_ctl.pages == NULL); | |
4613 | btrfs_put_block_group(cache); | |
4614 | } | |
4615 | ||
4616 | void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, | |
4617 | struct btrfs_fs_info *fs_info) | |
4618 | { | |
4619 | struct btrfs_block_group *cache; | |
4620 | ||
4621 | spin_lock(&cur_trans->dirty_bgs_lock); | |
4622 | while (!list_empty(&cur_trans->dirty_bgs)) { | |
4623 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
4624 | struct btrfs_block_group, | |
4625 | dirty_list); | |
4626 | ||
4627 | if (!list_empty(&cache->io_list)) { | |
4628 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
4629 | list_del_init(&cache->io_list); | |
4630 | btrfs_cleanup_bg_io(cache); | |
4631 | spin_lock(&cur_trans->dirty_bgs_lock); | |
4632 | } | |
4633 | ||
4634 | list_del_init(&cache->dirty_list); | |
4635 | spin_lock(&cache->lock); | |
4636 | cache->disk_cache_state = BTRFS_DC_ERROR; | |
4637 | spin_unlock(&cache->lock); | |
4638 | ||
4639 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
4640 | btrfs_put_block_group(cache); | |
4641 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
4642 | spin_lock(&cur_trans->dirty_bgs_lock); | |
4643 | } | |
4644 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
4645 | ||
4646 | /* | |
4647 | * Refer to the definition of io_bgs member for details why it's safe | |
4648 | * to use it without any locking | |
4649 | */ | |
4650 | while (!list_empty(&cur_trans->io_bgs)) { | |
4651 | cache = list_first_entry(&cur_trans->io_bgs, | |
4652 | struct btrfs_block_group, | |
4653 | io_list); | |
4654 | ||
4655 | list_del_init(&cache->io_list); | |
4656 | spin_lock(&cache->lock); | |
4657 | cache->disk_cache_state = BTRFS_DC_ERROR; | |
4658 | spin_unlock(&cache->lock); | |
4659 | btrfs_cleanup_bg_io(cache); | |
4660 | } | |
4661 | } | |
4662 | ||
4663 | void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, | |
4664 | struct btrfs_fs_info *fs_info) | |
4665 | { | |
4666 | struct btrfs_device *dev, *tmp; | |
4667 | ||
4668 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); | |
4669 | ASSERT(list_empty(&cur_trans->dirty_bgs)); | |
4670 | ASSERT(list_empty(&cur_trans->io_bgs)); | |
4671 | ||
4672 | list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, | |
4673 | post_commit_list) { | |
4674 | list_del_init(&dev->post_commit_list); | |
4675 | } | |
4676 | ||
4677 | btrfs_destroy_delayed_refs(cur_trans, fs_info); | |
4678 | ||
4679 | cur_trans->state = TRANS_STATE_COMMIT_START; | |
4680 | wake_up(&fs_info->transaction_blocked_wait); | |
4681 | ||
4682 | cur_trans->state = TRANS_STATE_UNBLOCKED; | |
4683 | wake_up(&fs_info->transaction_wait); | |
4684 | ||
4685 | btrfs_destroy_delayed_inodes(fs_info); | |
4686 | ||
4687 | btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, | |
4688 | EXTENT_DIRTY); | |
4689 | btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); | |
4690 | ||
4691 | cur_trans->state =TRANS_STATE_COMPLETED; | |
4692 | wake_up(&cur_trans->commit_wait); | |
4693 | } | |
4694 | ||
4695 | static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) | |
4696 | { | |
4697 | struct btrfs_transaction *t; | |
4698 | ||
4699 | mutex_lock(&fs_info->transaction_kthread_mutex); | |
4700 | ||
4701 | spin_lock(&fs_info->trans_lock); | |
4702 | while (!list_empty(&fs_info->trans_list)) { | |
4703 | t = list_first_entry(&fs_info->trans_list, | |
4704 | struct btrfs_transaction, list); | |
4705 | if (t->state >= TRANS_STATE_COMMIT_START) { | |
4706 | refcount_inc(&t->use_count); | |
4707 | spin_unlock(&fs_info->trans_lock); | |
4708 | btrfs_wait_for_commit(fs_info, t->transid); | |
4709 | btrfs_put_transaction(t); | |
4710 | spin_lock(&fs_info->trans_lock); | |
4711 | continue; | |
4712 | } | |
4713 | if (t == fs_info->running_transaction) { | |
4714 | t->state = TRANS_STATE_COMMIT_DOING; | |
4715 | spin_unlock(&fs_info->trans_lock); | |
4716 | /* | |
4717 | * We wait for 0 num_writers since we don't hold a trans | |
4718 | * handle open currently for this transaction. | |
4719 | */ | |
4720 | wait_event(t->writer_wait, | |
4721 | atomic_read(&t->num_writers) == 0); | |
4722 | } else { | |
4723 | spin_unlock(&fs_info->trans_lock); | |
4724 | } | |
4725 | btrfs_cleanup_one_transaction(t, fs_info); | |
4726 | ||
4727 | spin_lock(&fs_info->trans_lock); | |
4728 | if (t == fs_info->running_transaction) | |
4729 | fs_info->running_transaction = NULL; | |
4730 | list_del_init(&t->list); | |
4731 | spin_unlock(&fs_info->trans_lock); | |
4732 | ||
4733 | btrfs_put_transaction(t); | |
4734 | trace_btrfs_transaction_commit(fs_info->tree_root); | |
4735 | spin_lock(&fs_info->trans_lock); | |
4736 | } | |
4737 | spin_unlock(&fs_info->trans_lock); | |
4738 | btrfs_destroy_all_ordered_extents(fs_info); | |
4739 | btrfs_destroy_delayed_inodes(fs_info); | |
4740 | btrfs_assert_delayed_root_empty(fs_info); | |
4741 | btrfs_destroy_all_delalloc_inodes(fs_info); | |
4742 | btrfs_drop_all_logs(fs_info); | |
4743 | mutex_unlock(&fs_info->transaction_kthread_mutex); | |
4744 | ||
4745 | return 0; | |
4746 | } | |
4747 | ||
4748 | int btrfs_find_highest_objectid(struct btrfs_root *root, u64 *objectid) | |
4749 | { | |
4750 | struct btrfs_path *path; | |
4751 | int ret; | |
4752 | struct extent_buffer *l; | |
4753 | struct btrfs_key search_key; | |
4754 | struct btrfs_key found_key; | |
4755 | int slot; | |
4756 | ||
4757 | path = btrfs_alloc_path(); | |
4758 | if (!path) | |
4759 | return -ENOMEM; | |
4760 | ||
4761 | search_key.objectid = BTRFS_LAST_FREE_OBJECTID; | |
4762 | search_key.type = -1; | |
4763 | search_key.offset = (u64)-1; | |
4764 | ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); | |
4765 | if (ret < 0) | |
4766 | goto error; | |
4767 | BUG_ON(ret == 0); /* Corruption */ | |
4768 | if (path->slots[0] > 0) { | |
4769 | slot = path->slots[0] - 1; | |
4770 | l = path->nodes[0]; | |
4771 | btrfs_item_key_to_cpu(l, &found_key, slot); | |
4772 | *objectid = max_t(u64, found_key.objectid, | |
4773 | BTRFS_FIRST_FREE_OBJECTID - 1); | |
4774 | } else { | |
4775 | *objectid = BTRFS_FIRST_FREE_OBJECTID - 1; | |
4776 | } | |
4777 | ret = 0; | |
4778 | error: | |
4779 | btrfs_free_path(path); | |
4780 | return ret; | |
4781 | } | |
4782 | ||
4783 | int btrfs_find_free_objectid(struct btrfs_root *root, u64 *objectid) | |
4784 | { | |
4785 | int ret; | |
4786 | mutex_lock(&root->objectid_mutex); | |
4787 | ||
4788 | if (unlikely(root->highest_objectid >= BTRFS_LAST_FREE_OBJECTID)) { | |
4789 | btrfs_warn(root->fs_info, | |
4790 | "the objectid of root %llu reaches its highest value", | |
4791 | root->root_key.objectid); | |
4792 | ret = -ENOSPC; | |
4793 | goto out; | |
4794 | } | |
4795 | ||
4796 | *objectid = ++root->highest_objectid; | |
4797 | ret = 0; | |
4798 | out: | |
4799 | mutex_unlock(&root->objectid_mutex); | |
4800 | return ret; | |
4801 | } |