Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 CM |
4 | */ |
5 | ||
e20d96d6 | 6 | #include <linux/fs.h> |
d98237b3 | 7 | #include <linux/blkdev.h> |
0f7d52f4 | 8 | #include <linux/radix-tree.h> |
35b7e476 | 9 | #include <linux/writeback.h> |
ce9adaa5 | 10 | #include <linux/workqueue.h> |
a74a4b97 | 11 | #include <linux/kthread.h> |
5a0e3ad6 | 12 | #include <linux/slab.h> |
784b4e29 | 13 | #include <linux/migrate.h> |
7a36ddec | 14 | #include <linux/ratelimit.h> |
6463fe58 | 15 | #include <linux/uuid.h> |
803b2f54 | 16 | #include <linux/semaphore.h> |
540adea3 | 17 | #include <linux/error-injection.h> |
9678c543 | 18 | #include <linux/crc32c.h> |
b89f6d1f | 19 | #include <linux/sched/mm.h> |
7e75bf3f | 20 | #include <asm/unaligned.h> |
6d97c6e3 | 21 | #include <crypto/hash.h> |
eb60ceac CM |
22 | #include "ctree.h" |
23 | #include "disk-io.h" | |
e089f05c | 24 | #include "transaction.h" |
0f7d52f4 | 25 | #include "btrfs_inode.h" |
0b86a832 | 26 | #include "volumes.h" |
db94535d | 27 | #include "print-tree.h" |
925baedd | 28 | #include "locking.h" |
e02119d5 | 29 | #include "tree-log.h" |
fa9c0d79 | 30 | #include "free-space-cache.h" |
70f6d82e | 31 | #include "free-space-tree.h" |
581bb050 | 32 | #include "inode-map.h" |
21adbd5c | 33 | #include "check-integrity.h" |
606686ee | 34 | #include "rcu-string.h" |
8dabb742 | 35 | #include "dev-replace.h" |
53b381b3 | 36 | #include "raid56.h" |
5ac1d209 | 37 | #include "sysfs.h" |
fcebe456 | 38 | #include "qgroup.h" |
ebb8765b | 39 | #include "compression.h" |
557ea5dd | 40 | #include "tree-checker.h" |
fd708b81 | 41 | #include "ref-verify.h" |
aac0023c | 42 | #include "block-group.h" |
b0643e59 | 43 | #include "discard.h" |
f603bb94 | 44 | #include "space-info.h" |
eb60ceac | 45 | |
319e4d06 QW |
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 |\ | |
e2731e55 AJ |
50 | BTRFS_SUPER_FLAG_METADUMP |\ |
51 | BTRFS_SUPER_FLAG_METADUMP_V2) | |
319e4d06 | 52 | |
8b712842 | 53 | static void end_workqueue_fn(struct btrfs_work *work); |
143bede5 | 54 | static void btrfs_destroy_ordered_extents(struct btrfs_root *root); |
acce952b | 55 | static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, |
2ff7e61e | 56 | struct btrfs_fs_info *fs_info); |
143bede5 | 57 | static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); |
2ff7e61e | 58 | static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, |
acce952b | 59 | struct extent_io_tree *dirty_pages, |
60 | int mark); | |
2ff7e61e | 61 | static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, |
acce952b | 62 | struct extent_io_tree *pinned_extents); |
2ff7e61e JM |
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); | |
ce9adaa5 | 65 | |
d352ac68 | 66 | /* |
97eb6b69 DS |
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 | |
d352ac68 CM |
69 | * by writes to insert metadata for new file extents after IO is complete. |
70 | */ | |
97eb6b69 | 71 | struct btrfs_end_io_wq { |
ce9adaa5 CM |
72 | struct bio *bio; |
73 | bio_end_io_t *end_io; | |
74 | void *private; | |
75 | struct btrfs_fs_info *info; | |
4e4cbee9 | 76 | blk_status_t status; |
bfebd8b5 | 77 | enum btrfs_wq_endio_type metadata; |
8b712842 | 78 | struct btrfs_work work; |
ce9adaa5 | 79 | }; |
0da5468f | 80 | |
97eb6b69 DS |
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, | |
fba4b697 | 88 | SLAB_MEM_SPREAD, |
97eb6b69 DS |
89 | NULL); |
90 | if (!btrfs_end_io_wq_cache) | |
91 | return -ENOMEM; | |
92 | return 0; | |
93 | } | |
94 | ||
e67c718b | 95 | void __cold btrfs_end_io_wq_exit(void) |
97eb6b69 | 96 | { |
5598e900 | 97 | kmem_cache_destroy(btrfs_end_io_wq_cache); |
97eb6b69 DS |
98 | } |
99 | ||
141386e1 JB |
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 | ||
d352ac68 CM |
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 | */ | |
44b8bd7e | 111 | struct async_submit_bio { |
8896a08d | 112 | struct inode *inode; |
44b8bd7e | 113 | struct bio *bio; |
a758781d | 114 | extent_submit_bio_start_t *submit_bio_start; |
44b8bd7e | 115 | int mirror_num; |
eaf25d93 CM |
116 | /* |
117 | * bio_offset is optional, can be used if the pages in the bio | |
118 | * can't tell us where in the file the bio should go | |
119 | */ | |
120 | u64 bio_offset; | |
8b712842 | 121 | struct btrfs_work work; |
4e4cbee9 | 122 | blk_status_t status; |
44b8bd7e CM |
123 | }; |
124 | ||
85d4e461 CM |
125 | /* |
126 | * Lockdep class keys for extent_buffer->lock's in this root. For a given | |
127 | * eb, the lockdep key is determined by the btrfs_root it belongs to and | |
128 | * the level the eb occupies in the tree. | |
129 | * | |
130 | * Different roots are used for different purposes and may nest inside each | |
131 | * other and they require separate keysets. As lockdep keys should be | |
132 | * static, assign keysets according to the purpose of the root as indicated | |
4fd786e6 MT |
133 | * by btrfs_root->root_key.objectid. This ensures that all special purpose |
134 | * roots have separate keysets. | |
4008c04a | 135 | * |
85d4e461 CM |
136 | * Lock-nesting across peer nodes is always done with the immediate parent |
137 | * node locked thus preventing deadlock. As lockdep doesn't know this, use | |
138 | * subclass to avoid triggering lockdep warning in such cases. | |
4008c04a | 139 | * |
85d4e461 CM |
140 | * The key is set by the readpage_end_io_hook after the buffer has passed |
141 | * csum validation but before the pages are unlocked. It is also set by | |
142 | * btrfs_init_new_buffer on freshly allocated blocks. | |
4008c04a | 143 | * |
85d4e461 CM |
144 | * We also add a check to make sure the highest level of the tree is the |
145 | * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code | |
146 | * needs update as well. | |
4008c04a CM |
147 | */ |
148 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
149 | # if BTRFS_MAX_LEVEL != 8 | |
150 | # error | |
151 | # endif | |
85d4e461 | 152 | |
ab1405aa DS |
153 | #define DEFINE_LEVEL(stem, level) \ |
154 | .names[level] = "btrfs-" stem "-0" #level, | |
155 | ||
156 | #define DEFINE_NAME(stem) \ | |
157 | DEFINE_LEVEL(stem, 0) \ | |
158 | DEFINE_LEVEL(stem, 1) \ | |
159 | DEFINE_LEVEL(stem, 2) \ | |
160 | DEFINE_LEVEL(stem, 3) \ | |
161 | DEFINE_LEVEL(stem, 4) \ | |
162 | DEFINE_LEVEL(stem, 5) \ | |
163 | DEFINE_LEVEL(stem, 6) \ | |
164 | DEFINE_LEVEL(stem, 7) | |
165 | ||
85d4e461 CM |
166 | static struct btrfs_lockdep_keyset { |
167 | u64 id; /* root objectid */ | |
ab1405aa | 168 | /* Longest entry: btrfs-free-space-00 */ |
387824af DS |
169 | char names[BTRFS_MAX_LEVEL][20]; |
170 | struct lock_class_key keys[BTRFS_MAX_LEVEL]; | |
85d4e461 | 171 | } btrfs_lockdep_keysets[] = { |
ab1405aa DS |
172 | { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") }, |
173 | { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") }, | |
174 | { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") }, | |
175 | { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") }, | |
176 | { .id = BTRFS_FS_TREE_OBJECTID, DEFINE_NAME("fs") }, | |
177 | { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") }, | |
178 | { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") }, | |
179 | { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") }, | |
180 | { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") }, | |
181 | { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") }, | |
182 | { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") }, | |
183 | { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") }, | |
184 | { .id = 0, DEFINE_NAME("tree") }, | |
4008c04a | 185 | }; |
85d4e461 | 186 | |
ab1405aa DS |
187 | #undef DEFINE_LEVEL |
188 | #undef DEFINE_NAME | |
85d4e461 CM |
189 | |
190 | void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, | |
191 | int level) | |
192 | { | |
193 | struct btrfs_lockdep_keyset *ks; | |
194 | ||
195 | BUG_ON(level >= ARRAY_SIZE(ks->keys)); | |
196 | ||
197 | /* find the matching keyset, id 0 is the default entry */ | |
198 | for (ks = btrfs_lockdep_keysets; ks->id; ks++) | |
199 | if (ks->id == objectid) | |
200 | break; | |
201 | ||
202 | lockdep_set_class_and_name(&eb->lock, | |
203 | &ks->keys[level], ks->names[level]); | |
204 | } | |
205 | ||
4008c04a CM |
206 | #endif |
207 | ||
d352ac68 | 208 | /* |
2996e1f8 | 209 | * Compute the csum of a btree block and store the result to provided buffer. |
d352ac68 | 210 | */ |
c67b3892 | 211 | static void csum_tree_block(struct extent_buffer *buf, u8 *result) |
19c00ddc | 212 | { |
d5178578 | 213 | struct btrfs_fs_info *fs_info = buf->fs_info; |
e9be5a30 | 214 | const int num_pages = fs_info->nodesize >> PAGE_SHIFT; |
d5178578 | 215 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); |
19c00ddc | 216 | char *kaddr; |
e9be5a30 | 217 | int i; |
d5178578 JT |
218 | |
219 | shash->tfm = fs_info->csum_shash; | |
220 | crypto_shash_init(shash); | |
e9be5a30 DS |
221 | kaddr = page_address(buf->pages[0]); |
222 | crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE, | |
223 | PAGE_SIZE - BTRFS_CSUM_SIZE); | |
19c00ddc | 224 | |
e9be5a30 DS |
225 | for (i = 1; i < num_pages; i++) { |
226 | kaddr = page_address(buf->pages[i]); | |
227 | crypto_shash_update(shash, kaddr, PAGE_SIZE); | |
19c00ddc | 228 | } |
71a63551 | 229 | memset(result, 0, BTRFS_CSUM_SIZE); |
d5178578 | 230 | crypto_shash_final(shash, result); |
19c00ddc CM |
231 | } |
232 | ||
d352ac68 CM |
233 | /* |
234 | * we can't consider a given block up to date unless the transid of the | |
235 | * block matches the transid in the parent node's pointer. This is how we | |
236 | * detect blocks that either didn't get written at all or got written | |
237 | * in the wrong place. | |
238 | */ | |
1259ab75 | 239 | static int verify_parent_transid(struct extent_io_tree *io_tree, |
b9fab919 CM |
240 | struct extent_buffer *eb, u64 parent_transid, |
241 | int atomic) | |
1259ab75 | 242 | { |
2ac55d41 | 243 | struct extent_state *cached_state = NULL; |
1259ab75 | 244 | int ret; |
2755a0de | 245 | bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB); |
1259ab75 CM |
246 | |
247 | if (!parent_transid || btrfs_header_generation(eb) == parent_transid) | |
248 | return 0; | |
249 | ||
b9fab919 CM |
250 | if (atomic) |
251 | return -EAGAIN; | |
252 | ||
ac5887c8 | 253 | if (need_lock) |
a26e8c9f | 254 | btrfs_tree_read_lock(eb); |
a26e8c9f | 255 | |
2ac55d41 | 256 | lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, |
ff13db41 | 257 | &cached_state); |
0b32f4bb | 258 | if (extent_buffer_uptodate(eb) && |
1259ab75 CM |
259 | btrfs_header_generation(eb) == parent_transid) { |
260 | ret = 0; | |
261 | goto out; | |
262 | } | |
94647322 DS |
263 | btrfs_err_rl(eb->fs_info, |
264 | "parent transid verify failed on %llu wanted %llu found %llu", | |
265 | eb->start, | |
29549aec | 266 | parent_transid, btrfs_header_generation(eb)); |
1259ab75 | 267 | ret = 1; |
a26e8c9f JB |
268 | |
269 | /* | |
270 | * Things reading via commit roots that don't have normal protection, | |
271 | * like send, can have a really old block in cache that may point at a | |
01327610 | 272 | * block that has been freed and re-allocated. So don't clear uptodate |
a26e8c9f JB |
273 | * if we find an eb that is under IO (dirty/writeback) because we could |
274 | * end up reading in the stale data and then writing it back out and | |
275 | * making everybody very sad. | |
276 | */ | |
277 | if (!extent_buffer_under_io(eb)) | |
278 | clear_extent_buffer_uptodate(eb); | |
33958dc6 | 279 | out: |
2ac55d41 | 280 | unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, |
e43bbe5e | 281 | &cached_state); |
472b909f | 282 | if (need_lock) |
ac5887c8 | 283 | btrfs_tree_read_unlock(eb); |
1259ab75 | 284 | return ret; |
1259ab75 CM |
285 | } |
286 | ||
e7e16f48 JT |
287 | static bool btrfs_supported_super_csum(u16 csum_type) |
288 | { | |
289 | switch (csum_type) { | |
290 | case BTRFS_CSUM_TYPE_CRC32: | |
3951e7f0 | 291 | case BTRFS_CSUM_TYPE_XXHASH: |
3831bf00 | 292 | case BTRFS_CSUM_TYPE_SHA256: |
352ae07b | 293 | case BTRFS_CSUM_TYPE_BLAKE2: |
e7e16f48 JT |
294 | return true; |
295 | default: | |
296 | return false; | |
297 | } | |
298 | } | |
299 | ||
1104a885 DS |
300 | /* |
301 | * Return 0 if the superblock checksum type matches the checksum value of that | |
302 | * algorithm. Pass the raw disk superblock data. | |
303 | */ | |
ab8d0fc4 JM |
304 | static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info, |
305 | char *raw_disk_sb) | |
1104a885 DS |
306 | { |
307 | struct btrfs_super_block *disk_sb = | |
308 | (struct btrfs_super_block *)raw_disk_sb; | |
51bce6c9 | 309 | char result[BTRFS_CSUM_SIZE]; |
d5178578 JT |
310 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); |
311 | ||
312 | shash->tfm = fs_info->csum_shash; | |
1104a885 | 313 | |
51bce6c9 JT |
314 | /* |
315 | * The super_block structure does not span the whole | |
316 | * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is | |
317 | * filled with zeros and is included in the checksum. | |
318 | */ | |
fd08001f EB |
319 | crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE, |
320 | BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result); | |
1104a885 | 321 | |
55fc29be | 322 | if (memcmp(disk_sb->csum, result, fs_info->csum_size)) |
51bce6c9 | 323 | return 1; |
1104a885 | 324 | |
e7e16f48 | 325 | return 0; |
1104a885 DS |
326 | } |
327 | ||
e064d5e9 | 328 | int btrfs_verify_level_key(struct extent_buffer *eb, int level, |
448de471 | 329 | struct btrfs_key *first_key, u64 parent_transid) |
581c1760 | 330 | { |
e064d5e9 | 331 | struct btrfs_fs_info *fs_info = eb->fs_info; |
581c1760 QW |
332 | int found_level; |
333 | struct btrfs_key found_key; | |
334 | int ret; | |
335 | ||
336 | found_level = btrfs_header_level(eb); | |
337 | if (found_level != level) { | |
63489055 QW |
338 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), |
339 | KERN_ERR "BTRFS: tree level check failed\n"); | |
581c1760 QW |
340 | btrfs_err(fs_info, |
341 | "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", | |
342 | eb->start, level, found_level); | |
581c1760 QW |
343 | return -EIO; |
344 | } | |
345 | ||
346 | if (!first_key) | |
347 | return 0; | |
348 | ||
5d41be6f QW |
349 | /* |
350 | * For live tree block (new tree blocks in current transaction), | |
351 | * we need proper lock context to avoid race, which is impossible here. | |
352 | * So we only checks tree blocks which is read from disk, whose | |
353 | * generation <= fs_info->last_trans_committed. | |
354 | */ | |
355 | if (btrfs_header_generation(eb) > fs_info->last_trans_committed) | |
356 | return 0; | |
62fdaa52 QW |
357 | |
358 | /* We have @first_key, so this @eb must have at least one item */ | |
359 | if (btrfs_header_nritems(eb) == 0) { | |
360 | btrfs_err(fs_info, | |
361 | "invalid tree nritems, bytenr=%llu nritems=0 expect >0", | |
362 | eb->start); | |
363 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
364 | return -EUCLEAN; | |
365 | } | |
366 | ||
581c1760 QW |
367 | if (found_level) |
368 | btrfs_node_key_to_cpu(eb, &found_key, 0); | |
369 | else | |
370 | btrfs_item_key_to_cpu(eb, &found_key, 0); | |
371 | ret = btrfs_comp_cpu_keys(first_key, &found_key); | |
372 | ||
581c1760 | 373 | if (ret) { |
63489055 QW |
374 | WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG), |
375 | KERN_ERR "BTRFS: tree first key check failed\n"); | |
581c1760 | 376 | btrfs_err(fs_info, |
ff76a864 LB |
377 | "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", |
378 | eb->start, parent_transid, first_key->objectid, | |
379 | first_key->type, first_key->offset, | |
380 | found_key.objectid, found_key.type, | |
381 | found_key.offset); | |
581c1760 | 382 | } |
581c1760 QW |
383 | return ret; |
384 | } | |
385 | ||
d352ac68 CM |
386 | /* |
387 | * helper to read a given tree block, doing retries as required when | |
388 | * the checksums don't match and we have alternate mirrors to try. | |
581c1760 QW |
389 | * |
390 | * @parent_transid: expected transid, skip check if 0 | |
391 | * @level: expected level, mandatory check | |
392 | * @first_key: expected key of first slot, skip check if NULL | |
d352ac68 | 393 | */ |
5ab12d1f | 394 | static int btree_read_extent_buffer_pages(struct extent_buffer *eb, |
581c1760 QW |
395 | u64 parent_transid, int level, |
396 | struct btrfs_key *first_key) | |
f188591e | 397 | { |
5ab12d1f | 398 | struct btrfs_fs_info *fs_info = eb->fs_info; |
f188591e | 399 | struct extent_io_tree *io_tree; |
ea466794 | 400 | int failed = 0; |
f188591e CM |
401 | int ret; |
402 | int num_copies = 0; | |
403 | int mirror_num = 0; | |
ea466794 | 404 | int failed_mirror = 0; |
f188591e | 405 | |
0b246afa | 406 | io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
f188591e | 407 | while (1) { |
f8397d69 | 408 | clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); |
c2ccfbc6 | 409 | ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num); |
256dd1bb | 410 | if (!ret) { |
581c1760 | 411 | if (verify_parent_transid(io_tree, eb, |
b9fab919 | 412 | parent_transid, 0)) |
256dd1bb | 413 | ret = -EIO; |
e064d5e9 | 414 | else if (btrfs_verify_level_key(eb, level, |
448de471 | 415 | first_key, parent_transid)) |
581c1760 QW |
416 | ret = -EUCLEAN; |
417 | else | |
418 | break; | |
256dd1bb | 419 | } |
d397712b | 420 | |
0b246afa | 421 | num_copies = btrfs_num_copies(fs_info, |
f188591e | 422 | eb->start, eb->len); |
4235298e | 423 | if (num_copies == 1) |
ea466794 | 424 | break; |
4235298e | 425 | |
5cf1ab56 JB |
426 | if (!failed_mirror) { |
427 | failed = 1; | |
428 | failed_mirror = eb->read_mirror; | |
429 | } | |
430 | ||
f188591e | 431 | mirror_num++; |
ea466794 JB |
432 | if (mirror_num == failed_mirror) |
433 | mirror_num++; | |
434 | ||
4235298e | 435 | if (mirror_num > num_copies) |
ea466794 | 436 | break; |
f188591e | 437 | } |
ea466794 | 438 | |
c0901581 | 439 | if (failed && !ret && failed_mirror) |
20a1fbf9 | 440 | btrfs_repair_eb_io_failure(eb, failed_mirror); |
ea466794 JB |
441 | |
442 | return ret; | |
f188591e | 443 | } |
19c00ddc | 444 | |
d352ac68 | 445 | /* |
d397712b CM |
446 | * checksum a dirty tree block before IO. This has extra checks to make sure |
447 | * we only fill in the checksum field in the first page of a multi-page block | |
d352ac68 | 448 | */ |
d397712b | 449 | |
01d58472 | 450 | static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page) |
19c00ddc | 451 | { |
4eee4fa4 | 452 | u64 start = page_offset(page); |
19c00ddc | 453 | u64 found_start; |
2996e1f8 | 454 | u8 result[BTRFS_CSUM_SIZE]; |
19c00ddc | 455 | struct extent_buffer *eb; |
8d47a0d8 | 456 | int ret; |
f188591e | 457 | |
4f2de97a JB |
458 | eb = (struct extent_buffer *)page->private; |
459 | if (page != eb->pages[0]) | |
460 | return 0; | |
0f805531 | 461 | |
19c00ddc | 462 | found_start = btrfs_header_bytenr(eb); |
0f805531 AL |
463 | /* |
464 | * Please do not consolidate these warnings into a single if. | |
465 | * It is useful to know what went wrong. | |
466 | */ | |
467 | if (WARN_ON(found_start != start)) | |
468 | return -EUCLEAN; | |
469 | if (WARN_ON(!PageUptodate(page))) | |
470 | return -EUCLEAN; | |
471 | ||
de37aa51 | 472 | ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid, |
9a8658e3 DS |
473 | offsetof(struct btrfs_header, fsid), |
474 | BTRFS_FSID_SIZE) == 0); | |
0f805531 | 475 | |
c67b3892 | 476 | csum_tree_block(eb, result); |
2996e1f8 | 477 | |
8d47a0d8 QW |
478 | if (btrfs_header_level(eb)) |
479 | ret = btrfs_check_node(eb); | |
480 | else | |
481 | ret = btrfs_check_leaf_full(eb); | |
482 | ||
483 | if (ret < 0) { | |
c06631b0 | 484 | btrfs_print_tree(eb, 0); |
8d47a0d8 QW |
485 | btrfs_err(fs_info, |
486 | "block=%llu write time tree block corruption detected", | |
487 | eb->start); | |
c06631b0 | 488 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); |
8d47a0d8 QW |
489 | return ret; |
490 | } | |
713cebfb | 491 | write_extent_buffer(eb, result, 0, fs_info->csum_size); |
8d47a0d8 | 492 | |
2996e1f8 | 493 | return 0; |
19c00ddc CM |
494 | } |
495 | ||
b0c9b3b0 | 496 | static int check_tree_block_fsid(struct extent_buffer *eb) |
2b82032c | 497 | { |
b0c9b3b0 | 498 | struct btrfs_fs_info *fs_info = eb->fs_info; |
944d3f9f | 499 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; |
44880fdc | 500 | u8 fsid[BTRFS_FSID_SIZE]; |
944d3f9f | 501 | u8 *metadata_uuid; |
2b82032c | 502 | |
9a8658e3 DS |
503 | read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid), |
504 | BTRFS_FSID_SIZE); | |
944d3f9f NB |
505 | /* |
506 | * Checking the incompat flag is only valid for the current fs. For | |
507 | * seed devices it's forbidden to have their uuid changed so reading | |
508 | * ->fsid in this case is fine | |
509 | */ | |
510 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) | |
511 | metadata_uuid = fs_devices->metadata_uuid; | |
512 | else | |
513 | metadata_uuid = fs_devices->fsid; | |
514 | ||
515 | if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) | |
516 | return 0; | |
517 | ||
518 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) | |
519 | if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE)) | |
520 | return 0; | |
521 | ||
522 | return 1; | |
2b82032c YZ |
523 | } |
524 | ||
9a446d6a NB |
525 | int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio, u64 phy_offset, |
526 | struct page *page, u64 start, u64 end, | |
527 | int mirror) | |
ce9adaa5 | 528 | { |
ce9adaa5 CM |
529 | u64 found_start; |
530 | int found_level; | |
ce9adaa5 | 531 | struct extent_buffer *eb; |
15b6e8a8 DS |
532 | struct btrfs_fs_info *fs_info; |
533 | u16 csum_size; | |
f188591e | 534 | int ret = 0; |
2996e1f8 | 535 | u8 result[BTRFS_CSUM_SIZE]; |
727011e0 | 536 | int reads_done; |
ce9adaa5 | 537 | |
ce9adaa5 CM |
538 | if (!page->private) |
539 | goto out; | |
d397712b | 540 | |
4f2de97a | 541 | eb = (struct extent_buffer *)page->private; |
15b6e8a8 | 542 | fs_info = eb->fs_info; |
55fc29be | 543 | csum_size = fs_info->csum_size; |
d397712b | 544 | |
0b32f4bb JB |
545 | /* the pending IO might have been the only thing that kept this buffer |
546 | * in memory. Make sure we have a ref for all this other checks | |
547 | */ | |
67439dad | 548 | atomic_inc(&eb->refs); |
0b32f4bb JB |
549 | |
550 | reads_done = atomic_dec_and_test(&eb->io_pages); | |
727011e0 CM |
551 | if (!reads_done) |
552 | goto err; | |
f188591e | 553 | |
5cf1ab56 | 554 | eb->read_mirror = mirror; |
656f30db | 555 | if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) { |
ea466794 JB |
556 | ret = -EIO; |
557 | goto err; | |
558 | } | |
559 | ||
ce9adaa5 | 560 | found_start = btrfs_header_bytenr(eb); |
727011e0 | 561 | if (found_start != eb->start) { |
893bf4b1 SY |
562 | btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu", |
563 | eb->start, found_start); | |
f188591e | 564 | ret = -EIO; |
ce9adaa5 CM |
565 | goto err; |
566 | } | |
b0c9b3b0 | 567 | if (check_tree_block_fsid(eb)) { |
02873e43 ZL |
568 | btrfs_err_rl(fs_info, "bad fsid on block %llu", |
569 | eb->start); | |
1259ab75 CM |
570 | ret = -EIO; |
571 | goto err; | |
572 | } | |
ce9adaa5 | 573 | found_level = btrfs_header_level(eb); |
1c24c3ce | 574 | if (found_level >= BTRFS_MAX_LEVEL) { |
893bf4b1 SY |
575 | btrfs_err(fs_info, "bad tree block level %d on %llu", |
576 | (int)btrfs_header_level(eb), eb->start); | |
1c24c3ce JB |
577 | ret = -EIO; |
578 | goto err; | |
579 | } | |
ce9adaa5 | 580 | |
85d4e461 CM |
581 | btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), |
582 | eb, found_level); | |
4008c04a | 583 | |
c67b3892 | 584 | csum_tree_block(eb, result); |
a826d6dc | 585 | |
2996e1f8 | 586 | if (memcmp_extent_buffer(eb, result, 0, csum_size)) { |
35be8851 | 587 | u8 val[BTRFS_CSUM_SIZE] = { 0 }; |
2996e1f8 JT |
588 | |
589 | read_extent_buffer(eb, &val, 0, csum_size); | |
590 | btrfs_warn_rl(fs_info, | |
35be8851 | 591 | "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d", |
2996e1f8 | 592 | fs_info->sb->s_id, eb->start, |
35be8851 JT |
593 | CSUM_FMT_VALUE(csum_size, val), |
594 | CSUM_FMT_VALUE(csum_size, result), | |
595 | btrfs_header_level(eb)); | |
2996e1f8 JT |
596 | ret = -EUCLEAN; |
597 | goto err; | |
598 | } | |
599 | ||
a826d6dc JB |
600 | /* |
601 | * If this is a leaf block and it is corrupt, set the corrupt bit so | |
602 | * that we don't try and read the other copies of this block, just | |
603 | * return -EIO. | |
604 | */ | |
1c4360ee | 605 | if (found_level == 0 && btrfs_check_leaf_full(eb)) { |
a826d6dc JB |
606 | set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); |
607 | ret = -EIO; | |
608 | } | |
ce9adaa5 | 609 | |
813fd1dc | 610 | if (found_level > 0 && btrfs_check_node(eb)) |
053ab70f LB |
611 | ret = -EIO; |
612 | ||
0b32f4bb JB |
613 | if (!ret) |
614 | set_extent_buffer_uptodate(eb); | |
75391f0d QW |
615 | else |
616 | btrfs_err(fs_info, | |
617 | "block=%llu read time tree block corruption detected", | |
618 | eb->start); | |
ce9adaa5 | 619 | err: |
79fb65a1 JB |
620 | if (reads_done && |
621 | test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) | |
d48d71aa | 622 | btree_readahead_hook(eb, ret); |
4bb31e92 | 623 | |
53b381b3 DW |
624 | if (ret) { |
625 | /* | |
626 | * our io error hook is going to dec the io pages | |
627 | * again, we have to make sure it has something | |
628 | * to decrement | |
629 | */ | |
630 | atomic_inc(&eb->io_pages); | |
0b32f4bb | 631 | clear_extent_buffer_uptodate(eb); |
53b381b3 | 632 | } |
0b32f4bb | 633 | free_extent_buffer(eb); |
ce9adaa5 | 634 | out: |
f188591e | 635 | return ret; |
ce9adaa5 CM |
636 | } |
637 | ||
4246a0b6 | 638 | static void end_workqueue_bio(struct bio *bio) |
ce9adaa5 | 639 | { |
97eb6b69 | 640 | struct btrfs_end_io_wq *end_io_wq = bio->bi_private; |
ce9adaa5 | 641 | struct btrfs_fs_info *fs_info; |
9e0af237 | 642 | struct btrfs_workqueue *wq; |
ce9adaa5 | 643 | |
ce9adaa5 | 644 | fs_info = end_io_wq->info; |
4e4cbee9 | 645 | end_io_wq->status = bio->bi_status; |
d20f7043 | 646 | |
37226b21 | 647 | if (bio_op(bio) == REQ_OP_WRITE) { |
a0cac0ec | 648 | if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) |
9e0af237 | 649 | wq = fs_info->endio_meta_write_workers; |
a0cac0ec | 650 | else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) |
9e0af237 | 651 | wq = fs_info->endio_freespace_worker; |
a0cac0ec | 652 | else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) |
9e0af237 | 653 | wq = fs_info->endio_raid56_workers; |
a0cac0ec | 654 | else |
9e0af237 | 655 | wq = fs_info->endio_write_workers; |
d20f7043 | 656 | } else { |
5c047a69 | 657 | if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) |
9e0af237 | 658 | wq = fs_info->endio_raid56_workers; |
a0cac0ec | 659 | else if (end_io_wq->metadata) |
9e0af237 | 660 | wq = fs_info->endio_meta_workers; |
a0cac0ec | 661 | else |
9e0af237 | 662 | wq = fs_info->endio_workers; |
d20f7043 | 663 | } |
9e0af237 | 664 | |
a0cac0ec | 665 | btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); |
9e0af237 | 666 | btrfs_queue_work(wq, &end_io_wq->work); |
ce9adaa5 CM |
667 | } |
668 | ||
4e4cbee9 | 669 | blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, |
bfebd8b5 | 670 | enum btrfs_wq_endio_type metadata) |
0b86a832 | 671 | { |
97eb6b69 | 672 | struct btrfs_end_io_wq *end_io_wq; |
8b110e39 | 673 | |
97eb6b69 | 674 | end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS); |
ce9adaa5 | 675 | if (!end_io_wq) |
4e4cbee9 | 676 | return BLK_STS_RESOURCE; |
ce9adaa5 CM |
677 | |
678 | end_io_wq->private = bio->bi_private; | |
679 | end_io_wq->end_io = bio->bi_end_io; | |
22c59948 | 680 | end_io_wq->info = info; |
4e4cbee9 | 681 | end_io_wq->status = 0; |
ce9adaa5 | 682 | end_io_wq->bio = bio; |
22c59948 | 683 | end_io_wq->metadata = metadata; |
ce9adaa5 CM |
684 | |
685 | bio->bi_private = end_io_wq; | |
686 | bio->bi_end_io = end_workqueue_bio; | |
22c59948 CM |
687 | return 0; |
688 | } | |
689 | ||
4a69a410 CM |
690 | static void run_one_async_start(struct btrfs_work *work) |
691 | { | |
4a69a410 | 692 | struct async_submit_bio *async; |
4e4cbee9 | 693 | blk_status_t ret; |
4a69a410 CM |
694 | |
695 | async = container_of(work, struct async_submit_bio, work); | |
8896a08d | 696 | ret = async->submit_bio_start(async->inode, async->bio, async->bio_offset); |
79787eaa | 697 | if (ret) |
4e4cbee9 | 698 | async->status = ret; |
4a69a410 CM |
699 | } |
700 | ||
06ea01b1 DS |
701 | /* |
702 | * In order to insert checksums into the metadata in large chunks, we wait | |
703 | * until bio submission time. All the pages in the bio are checksummed and | |
704 | * sums are attached onto the ordered extent record. | |
705 | * | |
706 | * At IO completion time the csums attached on the ordered extent record are | |
707 | * inserted into the tree. | |
708 | */ | |
4a69a410 | 709 | static void run_one_async_done(struct btrfs_work *work) |
8b712842 | 710 | { |
8b712842 | 711 | struct async_submit_bio *async; |
06ea01b1 DS |
712 | struct inode *inode; |
713 | blk_status_t ret; | |
8b712842 CM |
714 | |
715 | async = container_of(work, struct async_submit_bio, work); | |
8896a08d | 716 | inode = async->inode; |
4854ddd0 | 717 | |
bb7ab3b9 | 718 | /* If an error occurred we just want to clean up the bio and move on */ |
4e4cbee9 CH |
719 | if (async->status) { |
720 | async->bio->bi_status = async->status; | |
4246a0b6 | 721 | bio_endio(async->bio); |
79787eaa JM |
722 | return; |
723 | } | |
724 | ||
ec39f769 CM |
725 | /* |
726 | * All of the bios that pass through here are from async helpers. | |
727 | * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context. | |
728 | * This changes nothing when cgroups aren't in use. | |
729 | */ | |
730 | async->bio->bi_opf |= REQ_CGROUP_PUNT; | |
08635bae | 731 | ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num); |
06ea01b1 DS |
732 | if (ret) { |
733 | async->bio->bi_status = ret; | |
734 | bio_endio(async->bio); | |
735 | } | |
4a69a410 CM |
736 | } |
737 | ||
738 | static void run_one_async_free(struct btrfs_work *work) | |
739 | { | |
740 | struct async_submit_bio *async; | |
741 | ||
742 | async = container_of(work, struct async_submit_bio, work); | |
8b712842 CM |
743 | kfree(async); |
744 | } | |
745 | ||
8896a08d | 746 | blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio, |
8c27cb35 | 747 | int mirror_num, unsigned long bio_flags, |
8896a08d | 748 | u64 bio_offset, |
e288c080 | 749 | extent_submit_bio_start_t *submit_bio_start) |
44b8bd7e | 750 | { |
8896a08d | 751 | struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; |
44b8bd7e CM |
752 | struct async_submit_bio *async; |
753 | ||
754 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
755 | if (!async) | |
4e4cbee9 | 756 | return BLK_STS_RESOURCE; |
44b8bd7e | 757 | |
8896a08d | 758 | async->inode = inode; |
44b8bd7e CM |
759 | async->bio = bio; |
760 | async->mirror_num = mirror_num; | |
4a69a410 | 761 | async->submit_bio_start = submit_bio_start; |
4a69a410 | 762 | |
a0cac0ec OS |
763 | btrfs_init_work(&async->work, run_one_async_start, run_one_async_done, |
764 | run_one_async_free); | |
4a69a410 | 765 | |
eaf25d93 | 766 | async->bio_offset = bio_offset; |
8c8bee1d | 767 | |
4e4cbee9 | 768 | async->status = 0; |
79787eaa | 769 | |
67f055c7 | 770 | if (op_is_sync(bio->bi_opf)) |
5cdc7ad3 | 771 | btrfs_set_work_high_priority(&async->work); |
d313d7a3 | 772 | |
5cdc7ad3 | 773 | btrfs_queue_work(fs_info->workers, &async->work); |
44b8bd7e CM |
774 | return 0; |
775 | } | |
776 | ||
4e4cbee9 | 777 | static blk_status_t btree_csum_one_bio(struct bio *bio) |
ce3ed71a | 778 | { |
2c30c71b | 779 | struct bio_vec *bvec; |
ce3ed71a | 780 | struct btrfs_root *root; |
2b070cfe | 781 | int ret = 0; |
6dc4f100 | 782 | struct bvec_iter_all iter_all; |
ce3ed71a | 783 | |
c09abff8 | 784 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 785 | bio_for_each_segment_all(bvec, bio, iter_all) { |
ce3ed71a | 786 | root = BTRFS_I(bvec->bv_page->mapping->host)->root; |
01d58472 | 787 | ret = csum_dirty_buffer(root->fs_info, bvec->bv_page); |
79787eaa JM |
788 | if (ret) |
789 | break; | |
ce3ed71a | 790 | } |
2c30c71b | 791 | |
4e4cbee9 | 792 | return errno_to_blk_status(ret); |
ce3ed71a CM |
793 | } |
794 | ||
8896a08d QW |
795 | static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio, |
796 | u64 bio_offset) | |
22c59948 | 797 | { |
8b712842 CM |
798 | /* |
799 | * when we're called for a write, we're already in the async | |
5443be45 | 800 | * submission context. Just jump into btrfs_map_bio |
8b712842 | 801 | */ |
79787eaa | 802 | return btree_csum_one_bio(bio); |
4a69a410 | 803 | } |
22c59948 | 804 | |
9b4e675a DS |
805 | static int check_async_write(struct btrfs_fs_info *fs_info, |
806 | struct btrfs_inode *bi) | |
de0022b9 | 807 | { |
6300463b LB |
808 | if (atomic_read(&bi->sync_writers)) |
809 | return 0; | |
9b4e675a | 810 | if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags)) |
de0022b9 | 811 | return 0; |
de0022b9 JB |
812 | return 1; |
813 | } | |
814 | ||
1b36294a NB |
815 | blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio, |
816 | int mirror_num, unsigned long bio_flags) | |
44b8bd7e | 817 | { |
0b246afa | 818 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9b4e675a | 819 | int async = check_async_write(fs_info, BTRFS_I(inode)); |
4e4cbee9 | 820 | blk_status_t ret; |
cad321ad | 821 | |
37226b21 | 822 | if (bio_op(bio) != REQ_OP_WRITE) { |
4a69a410 CM |
823 | /* |
824 | * called for a read, do the setup so that checksum validation | |
825 | * can happen in the async kernel threads | |
826 | */ | |
0b246afa JM |
827 | ret = btrfs_bio_wq_end_io(fs_info, bio, |
828 | BTRFS_WQ_ENDIO_METADATA); | |
1d4284bd | 829 | if (ret) |
61891923 | 830 | goto out_w_error; |
08635bae | 831 | ret = btrfs_map_bio(fs_info, bio, mirror_num); |
de0022b9 JB |
832 | } else if (!async) { |
833 | ret = btree_csum_one_bio(bio); | |
834 | if (ret) | |
61891923 | 835 | goto out_w_error; |
08635bae | 836 | ret = btrfs_map_bio(fs_info, bio, mirror_num); |
61891923 SB |
837 | } else { |
838 | /* | |
839 | * kthread helpers are used to submit writes so that | |
840 | * checksumming can happen in parallel across all CPUs | |
841 | */ | |
8896a08d QW |
842 | ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0, |
843 | 0, btree_submit_bio_start); | |
44b8bd7e | 844 | } |
d313d7a3 | 845 | |
4246a0b6 CH |
846 | if (ret) |
847 | goto out_w_error; | |
848 | return 0; | |
849 | ||
61891923 | 850 | out_w_error: |
4e4cbee9 | 851 | bio->bi_status = ret; |
4246a0b6 | 852 | bio_endio(bio); |
61891923 | 853 | return ret; |
44b8bd7e CM |
854 | } |
855 | ||
3dd1462e | 856 | #ifdef CONFIG_MIGRATION |
784b4e29 | 857 | static int btree_migratepage(struct address_space *mapping, |
a6bc32b8 MG |
858 | struct page *newpage, struct page *page, |
859 | enum migrate_mode mode) | |
784b4e29 CM |
860 | { |
861 | /* | |
862 | * we can't safely write a btree page from here, | |
863 | * we haven't done the locking hook | |
864 | */ | |
865 | if (PageDirty(page)) | |
866 | return -EAGAIN; | |
867 | /* | |
868 | * Buffers may be managed in a filesystem specific way. | |
869 | * We must have no buffers or drop them. | |
870 | */ | |
871 | if (page_has_private(page) && | |
872 | !try_to_release_page(page, GFP_KERNEL)) | |
873 | return -EAGAIN; | |
a6bc32b8 | 874 | return migrate_page(mapping, newpage, page, mode); |
784b4e29 | 875 | } |
3dd1462e | 876 | #endif |
784b4e29 | 877 | |
0da5468f CM |
878 | |
879 | static int btree_writepages(struct address_space *mapping, | |
880 | struct writeback_control *wbc) | |
881 | { | |
e2d84521 MX |
882 | struct btrfs_fs_info *fs_info; |
883 | int ret; | |
884 | ||
d8d5f3e1 | 885 | if (wbc->sync_mode == WB_SYNC_NONE) { |
448d640b CM |
886 | |
887 | if (wbc->for_kupdate) | |
888 | return 0; | |
889 | ||
e2d84521 | 890 | fs_info = BTRFS_I(mapping->host)->root->fs_info; |
b9473439 | 891 | /* this is a bit racy, but that's ok */ |
d814a491 EL |
892 | ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, |
893 | BTRFS_DIRTY_METADATA_THRESH, | |
894 | fs_info->dirty_metadata_batch); | |
e2d84521 | 895 | if (ret < 0) |
793955bc | 896 | return 0; |
793955bc | 897 | } |
0b32f4bb | 898 | return btree_write_cache_pages(mapping, wbc); |
0da5468f CM |
899 | } |
900 | ||
70dec807 | 901 | static int btree_releasepage(struct page *page, gfp_t gfp_flags) |
5f39d397 | 902 | { |
98509cfc | 903 | if (PageWriteback(page) || PageDirty(page)) |
d397712b | 904 | return 0; |
0c4e538b | 905 | |
f7a52a40 | 906 | return try_release_extent_buffer(page); |
d98237b3 CM |
907 | } |
908 | ||
d47992f8 LC |
909 | static void btree_invalidatepage(struct page *page, unsigned int offset, |
910 | unsigned int length) | |
d98237b3 | 911 | { |
d1310b2e CM |
912 | struct extent_io_tree *tree; |
913 | tree = &BTRFS_I(page->mapping->host)->io_tree; | |
5f39d397 CM |
914 | extent_invalidatepage(tree, page, offset); |
915 | btree_releasepage(page, GFP_NOFS); | |
9ad6b7bc | 916 | if (PagePrivate(page)) { |
efe120a0 FH |
917 | btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, |
918 | "page private not zero on page %llu", | |
919 | (unsigned long long)page_offset(page)); | |
d1b89bc0 | 920 | detach_page_private(page); |
9ad6b7bc | 921 | } |
d98237b3 CM |
922 | } |
923 | ||
0b32f4bb JB |
924 | static int btree_set_page_dirty(struct page *page) |
925 | { | |
bb146eb2 | 926 | #ifdef DEBUG |
0b32f4bb JB |
927 | struct extent_buffer *eb; |
928 | ||
929 | BUG_ON(!PagePrivate(page)); | |
930 | eb = (struct extent_buffer *)page->private; | |
931 | BUG_ON(!eb); | |
932 | BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
933 | BUG_ON(!atomic_read(&eb->refs)); | |
934 | btrfs_assert_tree_locked(eb); | |
bb146eb2 | 935 | #endif |
0b32f4bb JB |
936 | return __set_page_dirty_nobuffers(page); |
937 | } | |
938 | ||
7f09410b | 939 | static const struct address_space_operations btree_aops = { |
0da5468f | 940 | .writepages = btree_writepages, |
5f39d397 CM |
941 | .releasepage = btree_releasepage, |
942 | .invalidatepage = btree_invalidatepage, | |
5a92bc88 | 943 | #ifdef CONFIG_MIGRATION |
784b4e29 | 944 | .migratepage = btree_migratepage, |
5a92bc88 | 945 | #endif |
0b32f4bb | 946 | .set_page_dirty = btree_set_page_dirty, |
d98237b3 CM |
947 | }; |
948 | ||
2ff7e61e | 949 | void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr) |
090d1875 | 950 | { |
5f39d397 | 951 | struct extent_buffer *buf = NULL; |
537f38f0 | 952 | int ret; |
090d1875 | 953 | |
2ff7e61e | 954 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 | 955 | if (IS_ERR(buf)) |
6197d86e | 956 | return; |
537f38f0 | 957 | |
c2ccfbc6 | 958 | ret = read_extent_buffer_pages(buf, WAIT_NONE, 0); |
537f38f0 NB |
959 | if (ret < 0) |
960 | free_extent_buffer_stale(buf); | |
961 | else | |
962 | free_extent_buffer(buf); | |
090d1875 CM |
963 | } |
964 | ||
2ff7e61e JM |
965 | struct extent_buffer *btrfs_find_create_tree_block( |
966 | struct btrfs_fs_info *fs_info, | |
967 | u64 bytenr) | |
0999df54 | 968 | { |
0b246afa JM |
969 | if (btrfs_is_testing(fs_info)) |
970 | return alloc_test_extent_buffer(fs_info, bytenr); | |
971 | return alloc_extent_buffer(fs_info, bytenr); | |
0999df54 CM |
972 | } |
973 | ||
581c1760 QW |
974 | /* |
975 | * Read tree block at logical address @bytenr and do variant basic but critical | |
976 | * verification. | |
977 | * | |
978 | * @parent_transid: expected transid of this tree block, skip check if 0 | |
979 | * @level: expected level, mandatory check | |
980 | * @first_key: expected key in slot 0, skip check if NULL | |
981 | */ | |
2ff7e61e | 982 | struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr, |
581c1760 QW |
983 | u64 parent_transid, int level, |
984 | struct btrfs_key *first_key) | |
0999df54 CM |
985 | { |
986 | struct extent_buffer *buf = NULL; | |
0999df54 CM |
987 | int ret; |
988 | ||
2ff7e61e | 989 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
990 | if (IS_ERR(buf)) |
991 | return buf; | |
0999df54 | 992 | |
5ab12d1f | 993 | ret = btree_read_extent_buffer_pages(buf, parent_transid, |
581c1760 | 994 | level, first_key); |
0f0fe8f7 | 995 | if (ret) { |
537f38f0 | 996 | free_extent_buffer_stale(buf); |
64c043de | 997 | return ERR_PTR(ret); |
0f0fe8f7 | 998 | } |
5f39d397 | 999 | return buf; |
ce9adaa5 | 1000 | |
eb60ceac CM |
1001 | } |
1002 | ||
6a884d7d | 1003 | void btrfs_clean_tree_block(struct extent_buffer *buf) |
ed2ff2cb | 1004 | { |
6a884d7d | 1005 | struct btrfs_fs_info *fs_info = buf->fs_info; |
55c69072 | 1006 | if (btrfs_header_generation(buf) == |
e2d84521 | 1007 | fs_info->running_transaction->transid) { |
b9447ef8 | 1008 | btrfs_assert_tree_locked(buf); |
b4ce94de | 1009 | |
b9473439 | 1010 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { |
104b4e51 NB |
1011 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
1012 | -buf->len, | |
1013 | fs_info->dirty_metadata_batch); | |
ed7b63eb JB |
1014 | clear_extent_buffer_dirty(buf); |
1015 | } | |
925baedd | 1016 | } |
5f39d397 CM |
1017 | } |
1018 | ||
da17066c | 1019 | static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info, |
143bede5 | 1020 | u64 objectid) |
d97e63b6 | 1021 | { |
7c0260ee | 1022 | bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); |
96dfcb46 | 1023 | root->fs_info = fs_info; |
cfaa7295 | 1024 | root->node = NULL; |
a28ec197 | 1025 | root->commit_root = NULL; |
27cdeb70 | 1026 | root->state = 0; |
d68fc57b | 1027 | root->orphan_cleanup_state = 0; |
0b86a832 | 1028 | |
0f7d52f4 | 1029 | root->last_trans = 0; |
13a8a7c8 | 1030 | root->highest_objectid = 0; |
eb73c1b7 | 1031 | root->nr_delalloc_inodes = 0; |
199c2a9c | 1032 | root->nr_ordered_extents = 0; |
6bef4d31 | 1033 | root->inode_tree = RB_ROOT; |
16cdcec7 | 1034 | INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); |
f0486c68 | 1035 | root->block_rsv = NULL; |
0b86a832 CM |
1036 | |
1037 | INIT_LIST_HEAD(&root->dirty_list); | |
5d4f98a2 | 1038 | INIT_LIST_HEAD(&root->root_list); |
eb73c1b7 MX |
1039 | INIT_LIST_HEAD(&root->delalloc_inodes); |
1040 | INIT_LIST_HEAD(&root->delalloc_root); | |
199c2a9c MX |
1041 | INIT_LIST_HEAD(&root->ordered_extents); |
1042 | INIT_LIST_HEAD(&root->ordered_root); | |
d2311e69 | 1043 | INIT_LIST_HEAD(&root->reloc_dirty_list); |
2ab28f32 JB |
1044 | INIT_LIST_HEAD(&root->logged_list[0]); |
1045 | INIT_LIST_HEAD(&root->logged_list[1]); | |
5d4f98a2 | 1046 | spin_lock_init(&root->inode_lock); |
eb73c1b7 | 1047 | spin_lock_init(&root->delalloc_lock); |
199c2a9c | 1048 | spin_lock_init(&root->ordered_extent_lock); |
f0486c68 | 1049 | spin_lock_init(&root->accounting_lock); |
2ab28f32 JB |
1050 | spin_lock_init(&root->log_extents_lock[0]); |
1051 | spin_lock_init(&root->log_extents_lock[1]); | |
8287475a | 1052 | spin_lock_init(&root->qgroup_meta_rsv_lock); |
a2135011 | 1053 | mutex_init(&root->objectid_mutex); |
e02119d5 | 1054 | mutex_init(&root->log_mutex); |
31f3d255 | 1055 | mutex_init(&root->ordered_extent_mutex); |
573bfb72 | 1056 | mutex_init(&root->delalloc_mutex); |
c53e9653 | 1057 | init_waitqueue_head(&root->qgroup_flush_wait); |
7237f183 YZ |
1058 | init_waitqueue_head(&root->log_writer_wait); |
1059 | init_waitqueue_head(&root->log_commit_wait[0]); | |
1060 | init_waitqueue_head(&root->log_commit_wait[1]); | |
8b050d35 MX |
1061 | INIT_LIST_HEAD(&root->log_ctxs[0]); |
1062 | INIT_LIST_HEAD(&root->log_ctxs[1]); | |
7237f183 YZ |
1063 | atomic_set(&root->log_commit[0], 0); |
1064 | atomic_set(&root->log_commit[1], 0); | |
1065 | atomic_set(&root->log_writers, 0); | |
2ecb7923 | 1066 | atomic_set(&root->log_batch, 0); |
0700cea7 | 1067 | refcount_set(&root->refs, 1); |
8ecebf4d | 1068 | atomic_set(&root->snapshot_force_cow, 0); |
eede2bf3 | 1069 | atomic_set(&root->nr_swapfiles, 0); |
7237f183 | 1070 | root->log_transid = 0; |
d1433deb | 1071 | root->log_transid_committed = -1; |
257c62e1 | 1072 | root->last_log_commit = 0; |
e289f03e | 1073 | if (!dummy) { |
43eb5f29 QW |
1074 | extent_io_tree_init(fs_info, &root->dirty_log_pages, |
1075 | IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL); | |
e289f03e FM |
1076 | extent_io_tree_init(fs_info, &root->log_csum_range, |
1077 | IO_TREE_LOG_CSUM_RANGE, NULL); | |
1078 | } | |
017e5369 | 1079 | |
3768f368 CM |
1080 | memset(&root->root_key, 0, sizeof(root->root_key)); |
1081 | memset(&root->root_item, 0, sizeof(root->root_item)); | |
6702ed49 | 1082 | memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); |
4d775673 | 1083 | root->root_key.objectid = objectid; |
0ee5dc67 | 1084 | root->anon_dev = 0; |
8ea05e3a | 1085 | |
5f3ab90a | 1086 | spin_lock_init(&root->root_item_lock); |
370a11b8 | 1087 | btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks); |
bd647ce3 JB |
1088 | #ifdef CONFIG_BTRFS_DEBUG |
1089 | INIT_LIST_HEAD(&root->leak_list); | |
1090 | spin_lock(&fs_info->fs_roots_radix_lock); | |
1091 | list_add_tail(&root->leak_list, &fs_info->allocated_roots); | |
1092 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
1093 | #endif | |
3768f368 CM |
1094 | } |
1095 | ||
74e4d827 | 1096 | static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info, |
96dfcb46 | 1097 | u64 objectid, gfp_t flags) |
6f07e42e | 1098 | { |
74e4d827 | 1099 | struct btrfs_root *root = kzalloc(sizeof(*root), flags); |
6f07e42e | 1100 | if (root) |
96dfcb46 | 1101 | __setup_root(root, fs_info, objectid); |
6f07e42e AV |
1102 | return root; |
1103 | } | |
1104 | ||
06ea65a3 JB |
1105 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
1106 | /* Should only be used by the testing infrastructure */ | |
da17066c | 1107 | struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info) |
06ea65a3 JB |
1108 | { |
1109 | struct btrfs_root *root; | |
1110 | ||
7c0260ee JM |
1111 | if (!fs_info) |
1112 | return ERR_PTR(-EINVAL); | |
1113 | ||
96dfcb46 | 1114 | root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL); |
06ea65a3 JB |
1115 | if (!root) |
1116 | return ERR_PTR(-ENOMEM); | |
da17066c | 1117 | |
b9ef22de | 1118 | /* We don't use the stripesize in selftest, set it as sectorsize */ |
faa2dbf0 | 1119 | root->alloc_bytenr = 0; |
06ea65a3 JB |
1120 | |
1121 | return root; | |
1122 | } | |
1123 | #endif | |
1124 | ||
20897f5c | 1125 | struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, |
20897f5c AJ |
1126 | u64 objectid) |
1127 | { | |
9b7a2440 | 1128 | struct btrfs_fs_info *fs_info = trans->fs_info; |
20897f5c AJ |
1129 | struct extent_buffer *leaf; |
1130 | struct btrfs_root *tree_root = fs_info->tree_root; | |
1131 | struct btrfs_root *root; | |
1132 | struct btrfs_key key; | |
b89f6d1f | 1133 | unsigned int nofs_flag; |
20897f5c | 1134 | int ret = 0; |
20897f5c | 1135 | |
b89f6d1f FM |
1136 | /* |
1137 | * We're holding a transaction handle, so use a NOFS memory allocation | |
1138 | * context to avoid deadlock if reclaim happens. | |
1139 | */ | |
1140 | nofs_flag = memalloc_nofs_save(); | |
96dfcb46 | 1141 | root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL); |
b89f6d1f | 1142 | memalloc_nofs_restore(nofs_flag); |
20897f5c AJ |
1143 | if (!root) |
1144 | return ERR_PTR(-ENOMEM); | |
1145 | ||
20897f5c AJ |
1146 | root->root_key.objectid = objectid; |
1147 | root->root_key.type = BTRFS_ROOT_ITEM_KEY; | |
1148 | root->root_key.offset = 0; | |
1149 | ||
9631e4cc JB |
1150 | leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, |
1151 | BTRFS_NESTING_NORMAL); | |
20897f5c AJ |
1152 | if (IS_ERR(leaf)) { |
1153 | ret = PTR_ERR(leaf); | |
1dd05682 | 1154 | leaf = NULL; |
20897f5c AJ |
1155 | goto fail; |
1156 | } | |
1157 | ||
20897f5c | 1158 | root->node = leaf; |
20897f5c AJ |
1159 | btrfs_mark_buffer_dirty(leaf); |
1160 | ||
1161 | root->commit_root = btrfs_root_node(root); | |
27cdeb70 | 1162 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); |
20897f5c | 1163 | |
f944d2cb DS |
1164 | btrfs_set_root_flags(&root->root_item, 0); |
1165 | btrfs_set_root_limit(&root->root_item, 0); | |
20897f5c AJ |
1166 | btrfs_set_root_bytenr(&root->root_item, leaf->start); |
1167 | btrfs_set_root_generation(&root->root_item, trans->transid); | |
1168 | btrfs_set_root_level(&root->root_item, 0); | |
1169 | btrfs_set_root_refs(&root->root_item, 1); | |
1170 | btrfs_set_root_used(&root->root_item, leaf->len); | |
1171 | btrfs_set_root_last_snapshot(&root->root_item, 0); | |
1172 | btrfs_set_root_dirid(&root->root_item, 0); | |
33d85fda | 1173 | if (is_fstree(objectid)) |
807fc790 AS |
1174 | generate_random_guid(root->root_item.uuid); |
1175 | else | |
1176 | export_guid(root->root_item.uuid, &guid_null); | |
c8422684 | 1177 | btrfs_set_root_drop_level(&root->root_item, 0); |
20897f5c AJ |
1178 | |
1179 | key.objectid = objectid; | |
1180 | key.type = BTRFS_ROOT_ITEM_KEY; | |
1181 | key.offset = 0; | |
1182 | ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); | |
1183 | if (ret) | |
1184 | goto fail; | |
1185 | ||
1186 | btrfs_tree_unlock(leaf); | |
1187 | ||
1dd05682 TI |
1188 | return root; |
1189 | ||
20897f5c | 1190 | fail: |
8c38938c | 1191 | if (leaf) |
1dd05682 | 1192 | btrfs_tree_unlock(leaf); |
00246528 | 1193 | btrfs_put_root(root); |
20897f5c | 1194 | |
1dd05682 | 1195 | return ERR_PTR(ret); |
20897f5c AJ |
1196 | } |
1197 | ||
7237f183 YZ |
1198 | static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, |
1199 | struct btrfs_fs_info *fs_info) | |
0f7d52f4 CM |
1200 | { |
1201 | struct btrfs_root *root; | |
7237f183 | 1202 | struct extent_buffer *leaf; |
e02119d5 | 1203 | |
96dfcb46 | 1204 | root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS); |
e02119d5 | 1205 | if (!root) |
7237f183 | 1206 | return ERR_PTR(-ENOMEM); |
e02119d5 | 1207 | |
e02119d5 CM |
1208 | root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; |
1209 | root->root_key.type = BTRFS_ROOT_ITEM_KEY; | |
1210 | root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; | |
27cdeb70 | 1211 | |
7237f183 | 1212 | /* |
92a7cc42 | 1213 | * DON'T set SHAREABLE bit for log trees. |
27cdeb70 | 1214 | * |
92a7cc42 QW |
1215 | * Log trees are not exposed to user space thus can't be snapshotted, |
1216 | * and they go away before a real commit is actually done. | |
1217 | * | |
1218 | * They do store pointers to file data extents, and those reference | |
1219 | * counts still get updated (along with back refs to the log tree). | |
7237f183 | 1220 | */ |
e02119d5 | 1221 | |
4d75f8a9 | 1222 | leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID, |
9631e4cc | 1223 | NULL, 0, 0, 0, BTRFS_NESTING_NORMAL); |
7237f183 | 1224 | if (IS_ERR(leaf)) { |
00246528 | 1225 | btrfs_put_root(root); |
7237f183 YZ |
1226 | return ERR_CAST(leaf); |
1227 | } | |
e02119d5 | 1228 | |
7237f183 | 1229 | root->node = leaf; |
e02119d5 | 1230 | |
e02119d5 CM |
1231 | btrfs_mark_buffer_dirty(root->node); |
1232 | btrfs_tree_unlock(root->node); | |
7237f183 YZ |
1233 | return root; |
1234 | } | |
1235 | ||
1236 | int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, | |
1237 | struct btrfs_fs_info *fs_info) | |
1238 | { | |
1239 | struct btrfs_root *log_root; | |
1240 | ||
1241 | log_root = alloc_log_tree(trans, fs_info); | |
1242 | if (IS_ERR(log_root)) | |
1243 | return PTR_ERR(log_root); | |
1244 | WARN_ON(fs_info->log_root_tree); | |
1245 | fs_info->log_root_tree = log_root; | |
1246 | return 0; | |
1247 | } | |
1248 | ||
1249 | int btrfs_add_log_tree(struct btrfs_trans_handle *trans, | |
1250 | struct btrfs_root *root) | |
1251 | { | |
0b246afa | 1252 | struct btrfs_fs_info *fs_info = root->fs_info; |
7237f183 YZ |
1253 | struct btrfs_root *log_root; |
1254 | struct btrfs_inode_item *inode_item; | |
1255 | ||
0b246afa | 1256 | log_root = alloc_log_tree(trans, fs_info); |
7237f183 YZ |
1257 | if (IS_ERR(log_root)) |
1258 | return PTR_ERR(log_root); | |
1259 | ||
1260 | log_root->last_trans = trans->transid; | |
1261 | log_root->root_key.offset = root->root_key.objectid; | |
1262 | ||
1263 | inode_item = &log_root->root_item.inode; | |
3cae210f QW |
1264 | btrfs_set_stack_inode_generation(inode_item, 1); |
1265 | btrfs_set_stack_inode_size(inode_item, 3); | |
1266 | btrfs_set_stack_inode_nlink(inode_item, 1); | |
da17066c | 1267 | btrfs_set_stack_inode_nbytes(inode_item, |
0b246afa | 1268 | fs_info->nodesize); |
3cae210f | 1269 | btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); |
7237f183 | 1270 | |
5d4f98a2 | 1271 | btrfs_set_root_node(&log_root->root_item, log_root->node); |
7237f183 YZ |
1272 | |
1273 | WARN_ON(root->log_root); | |
1274 | root->log_root = log_root; | |
1275 | root->log_transid = 0; | |
d1433deb | 1276 | root->log_transid_committed = -1; |
257c62e1 | 1277 | root->last_log_commit = 0; |
e02119d5 CM |
1278 | return 0; |
1279 | } | |
1280 | ||
49d11bea JB |
1281 | static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root, |
1282 | struct btrfs_path *path, | |
1283 | struct btrfs_key *key) | |
e02119d5 CM |
1284 | { |
1285 | struct btrfs_root *root; | |
1286 | struct btrfs_fs_info *fs_info = tree_root->fs_info; | |
84234f3a | 1287 | u64 generation; |
cb517eab | 1288 | int ret; |
581c1760 | 1289 | int level; |
0f7d52f4 | 1290 | |
96dfcb46 | 1291 | root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS); |
49d11bea JB |
1292 | if (!root) |
1293 | return ERR_PTR(-ENOMEM); | |
0f7d52f4 | 1294 | |
cb517eab MX |
1295 | ret = btrfs_find_root(tree_root, key, path, |
1296 | &root->root_item, &root->root_key); | |
0f7d52f4 | 1297 | if (ret) { |
13a8a7c8 YZ |
1298 | if (ret > 0) |
1299 | ret = -ENOENT; | |
49d11bea | 1300 | goto fail; |
0f7d52f4 | 1301 | } |
13a8a7c8 | 1302 | |
84234f3a | 1303 | generation = btrfs_root_generation(&root->root_item); |
581c1760 | 1304 | level = btrfs_root_level(&root->root_item); |
2ff7e61e JM |
1305 | root->node = read_tree_block(fs_info, |
1306 | btrfs_root_bytenr(&root->root_item), | |
581c1760 | 1307 | generation, level, NULL); |
64c043de LB |
1308 | if (IS_ERR(root->node)) { |
1309 | ret = PTR_ERR(root->node); | |
8c38938c | 1310 | root->node = NULL; |
49d11bea | 1311 | goto fail; |
cb517eab MX |
1312 | } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { |
1313 | ret = -EIO; | |
49d11bea | 1314 | goto fail; |
416bc658 | 1315 | } |
5d4f98a2 | 1316 | root->commit_root = btrfs_root_node(root); |
cb517eab | 1317 | return root; |
49d11bea | 1318 | fail: |
00246528 | 1319 | btrfs_put_root(root); |
49d11bea JB |
1320 | return ERR_PTR(ret); |
1321 | } | |
1322 | ||
1323 | struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, | |
1324 | struct btrfs_key *key) | |
1325 | { | |
1326 | struct btrfs_root *root; | |
1327 | struct btrfs_path *path; | |
1328 | ||
1329 | path = btrfs_alloc_path(); | |
1330 | if (!path) | |
1331 | return ERR_PTR(-ENOMEM); | |
1332 | root = read_tree_root_path(tree_root, path, key); | |
1333 | btrfs_free_path(path); | |
1334 | ||
1335 | return root; | |
cb517eab MX |
1336 | } |
1337 | ||
2dfb1e43 QW |
1338 | /* |
1339 | * Initialize subvolume root in-memory structure | |
1340 | * | |
1341 | * @anon_dev: anonymous device to attach to the root, if zero, allocate new | |
1342 | */ | |
1343 | static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev) | |
cb517eab MX |
1344 | { |
1345 | int ret; | |
dcc3eb96 | 1346 | unsigned int nofs_flag; |
cb517eab MX |
1347 | |
1348 | root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); | |
1349 | root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), | |
1350 | GFP_NOFS); | |
1351 | if (!root->free_ino_pinned || !root->free_ino_ctl) { | |
1352 | ret = -ENOMEM; | |
1353 | goto fail; | |
1354 | } | |
1355 | ||
dcc3eb96 NB |
1356 | /* |
1357 | * We might be called under a transaction (e.g. indirect backref | |
1358 | * resolution) which could deadlock if it triggers memory reclaim | |
1359 | */ | |
1360 | nofs_flag = memalloc_nofs_save(); | |
1361 | ret = btrfs_drew_lock_init(&root->snapshot_lock); | |
1362 | memalloc_nofs_restore(nofs_flag); | |
1363 | if (ret) | |
8257b2dc | 1364 | goto fail; |
8257b2dc | 1365 | |
aeb935a4 QW |
1366 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID && |
1367 | root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) { | |
92a7cc42 | 1368 | set_bit(BTRFS_ROOT_SHAREABLE, &root->state); |
f39e4571 JB |
1369 | btrfs_check_and_init_root_item(&root->root_item); |
1370 | } | |
1371 | ||
cb517eab | 1372 | btrfs_init_free_ino_ctl(root); |
57cdc8db DS |
1373 | spin_lock_init(&root->ino_cache_lock); |
1374 | init_waitqueue_head(&root->ino_cache_wait); | |
cb517eab | 1375 | |
851fd730 QW |
1376 | /* |
1377 | * Don't assign anonymous block device to roots that are not exposed to | |
1378 | * userspace, the id pool is limited to 1M | |
1379 | */ | |
1380 | if (is_fstree(root->root_key.objectid) && | |
1381 | btrfs_root_refs(&root->root_item) > 0) { | |
2dfb1e43 QW |
1382 | if (!anon_dev) { |
1383 | ret = get_anon_bdev(&root->anon_dev); | |
1384 | if (ret) | |
1385 | goto fail; | |
1386 | } else { | |
1387 | root->anon_dev = anon_dev; | |
1388 | } | |
851fd730 | 1389 | } |
f32e48e9 CR |
1390 | |
1391 | mutex_lock(&root->objectid_mutex); | |
1392 | ret = btrfs_find_highest_objectid(root, | |
1393 | &root->highest_objectid); | |
1394 | if (ret) { | |
1395 | mutex_unlock(&root->objectid_mutex); | |
876d2cf1 | 1396 | goto fail; |
f32e48e9 CR |
1397 | } |
1398 | ||
1399 | ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); | |
1400 | ||
1401 | mutex_unlock(&root->objectid_mutex); | |
1402 | ||
cb517eab MX |
1403 | return 0; |
1404 | fail: | |
84db5ccf | 1405 | /* The caller is responsible to call btrfs_free_fs_root */ |
cb517eab MX |
1406 | return ret; |
1407 | } | |
1408 | ||
a98db0f3 JB |
1409 | static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, |
1410 | u64 root_id) | |
cb517eab MX |
1411 | { |
1412 | struct btrfs_root *root; | |
1413 | ||
1414 | spin_lock(&fs_info->fs_roots_radix_lock); | |
1415 | root = radix_tree_lookup(&fs_info->fs_roots_radix, | |
1416 | (unsigned long)root_id); | |
bc44d7c4 | 1417 | if (root) |
00246528 | 1418 | root = btrfs_grab_root(root); |
cb517eab MX |
1419 | spin_unlock(&fs_info->fs_roots_radix_lock); |
1420 | return root; | |
1421 | } | |
1422 | ||
49d11bea JB |
1423 | static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info, |
1424 | u64 objectid) | |
1425 | { | |
1426 | if (objectid == BTRFS_ROOT_TREE_OBJECTID) | |
1427 | return btrfs_grab_root(fs_info->tree_root); | |
1428 | if (objectid == BTRFS_EXTENT_TREE_OBJECTID) | |
1429 | return btrfs_grab_root(fs_info->extent_root); | |
1430 | if (objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
1431 | return btrfs_grab_root(fs_info->chunk_root); | |
1432 | if (objectid == BTRFS_DEV_TREE_OBJECTID) | |
1433 | return btrfs_grab_root(fs_info->dev_root); | |
1434 | if (objectid == BTRFS_CSUM_TREE_OBJECTID) | |
1435 | return btrfs_grab_root(fs_info->csum_root); | |
1436 | if (objectid == BTRFS_QUOTA_TREE_OBJECTID) | |
1437 | return btrfs_grab_root(fs_info->quota_root) ? | |
1438 | fs_info->quota_root : ERR_PTR(-ENOENT); | |
1439 | if (objectid == BTRFS_UUID_TREE_OBJECTID) | |
1440 | return btrfs_grab_root(fs_info->uuid_root) ? | |
1441 | fs_info->uuid_root : ERR_PTR(-ENOENT); | |
1442 | if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) | |
1443 | return btrfs_grab_root(fs_info->free_space_root) ? | |
1444 | fs_info->free_space_root : ERR_PTR(-ENOENT); | |
1445 | return NULL; | |
1446 | } | |
1447 | ||
cb517eab MX |
1448 | int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, |
1449 | struct btrfs_root *root) | |
1450 | { | |
1451 | int ret; | |
1452 | ||
e1860a77 | 1453 | ret = radix_tree_preload(GFP_NOFS); |
cb517eab MX |
1454 | if (ret) |
1455 | return ret; | |
1456 | ||
1457 | spin_lock(&fs_info->fs_roots_radix_lock); | |
1458 | ret = radix_tree_insert(&fs_info->fs_roots_radix, | |
1459 | (unsigned long)root->root_key.objectid, | |
1460 | root); | |
af01d2e5 | 1461 | if (ret == 0) { |
00246528 | 1462 | btrfs_grab_root(root); |
27cdeb70 | 1463 | set_bit(BTRFS_ROOT_IN_RADIX, &root->state); |
af01d2e5 | 1464 | } |
cb517eab MX |
1465 | spin_unlock(&fs_info->fs_roots_radix_lock); |
1466 | radix_tree_preload_end(); | |
1467 | ||
1468 | return ret; | |
1469 | } | |
1470 | ||
bd647ce3 JB |
1471 | void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info) |
1472 | { | |
1473 | #ifdef CONFIG_BTRFS_DEBUG | |
1474 | struct btrfs_root *root; | |
1475 | ||
1476 | while (!list_empty(&fs_info->allocated_roots)) { | |
457f1864 JB |
1477 | char buf[BTRFS_ROOT_NAME_BUF_LEN]; |
1478 | ||
bd647ce3 JB |
1479 | root = list_first_entry(&fs_info->allocated_roots, |
1480 | struct btrfs_root, leak_list); | |
457f1864 JB |
1481 | btrfs_err(fs_info, "leaked root %s refcount %d", |
1482 | btrfs_root_name(root->root_key.objectid, buf), | |
bd647ce3 JB |
1483 | refcount_read(&root->refs)); |
1484 | while (refcount_read(&root->refs) > 1) | |
00246528 JB |
1485 | btrfs_put_root(root); |
1486 | btrfs_put_root(root); | |
bd647ce3 JB |
1487 | } |
1488 | #endif | |
1489 | } | |
1490 | ||
0d4b0463 JB |
1491 | void btrfs_free_fs_info(struct btrfs_fs_info *fs_info) |
1492 | { | |
141386e1 JB |
1493 | percpu_counter_destroy(&fs_info->dirty_metadata_bytes); |
1494 | percpu_counter_destroy(&fs_info->delalloc_bytes); | |
1495 | percpu_counter_destroy(&fs_info->dio_bytes); | |
1496 | percpu_counter_destroy(&fs_info->dev_replace.bio_counter); | |
1497 | btrfs_free_csum_hash(fs_info); | |
1498 | btrfs_free_stripe_hash_table(fs_info); | |
1499 | btrfs_free_ref_cache(fs_info); | |
0d4b0463 JB |
1500 | kfree(fs_info->balance_ctl); |
1501 | kfree(fs_info->delayed_root); | |
00246528 JB |
1502 | btrfs_put_root(fs_info->extent_root); |
1503 | btrfs_put_root(fs_info->tree_root); | |
1504 | btrfs_put_root(fs_info->chunk_root); | |
1505 | btrfs_put_root(fs_info->dev_root); | |
1506 | btrfs_put_root(fs_info->csum_root); | |
1507 | btrfs_put_root(fs_info->quota_root); | |
1508 | btrfs_put_root(fs_info->uuid_root); | |
1509 | btrfs_put_root(fs_info->free_space_root); | |
1510 | btrfs_put_root(fs_info->fs_root); | |
aeb935a4 | 1511 | btrfs_put_root(fs_info->data_reloc_root); |
bd647ce3 | 1512 | btrfs_check_leaked_roots(fs_info); |
3fd63727 | 1513 | btrfs_extent_buffer_leak_debug_check(fs_info); |
0d4b0463 JB |
1514 | kfree(fs_info->super_copy); |
1515 | kfree(fs_info->super_for_commit); | |
1516 | kvfree(fs_info); | |
1517 | } | |
1518 | ||
1519 | ||
2dfb1e43 QW |
1520 | /* |
1521 | * Get an in-memory reference of a root structure. | |
1522 | * | |
1523 | * For essential trees like root/extent tree, we grab it from fs_info directly. | |
1524 | * For subvolume trees, we check the cached filesystem roots first. If not | |
1525 | * found, then read it from disk and add it to cached fs roots. | |
1526 | * | |
1527 | * Caller should release the root by calling btrfs_put_root() after the usage. | |
1528 | * | |
1529 | * NOTE: Reloc and log trees can't be read by this function as they share the | |
1530 | * same root objectid. | |
1531 | * | |
1532 | * @objectid: root id | |
1533 | * @anon_dev: preallocated anonymous block device number for new roots, | |
1534 | * pass 0 for new allocation. | |
1535 | * @check_ref: whether to check root item references, If true, return -ENOENT | |
1536 | * for orphan roots | |
1537 | */ | |
1538 | static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info, | |
1539 | u64 objectid, dev_t anon_dev, | |
1540 | bool check_ref) | |
5eda7b5e CM |
1541 | { |
1542 | struct btrfs_root *root; | |
381cf658 | 1543 | struct btrfs_path *path; |
1d4c08e0 | 1544 | struct btrfs_key key; |
5eda7b5e CM |
1545 | int ret; |
1546 | ||
49d11bea JB |
1547 | root = btrfs_get_global_root(fs_info, objectid); |
1548 | if (root) | |
1549 | return root; | |
4df27c4d | 1550 | again: |
56e9357a | 1551 | root = btrfs_lookup_fs_root(fs_info, objectid); |
48475471 | 1552 | if (root) { |
2dfb1e43 QW |
1553 | /* Shouldn't get preallocated anon_dev for cached roots */ |
1554 | ASSERT(!anon_dev); | |
bc44d7c4 | 1555 | if (check_ref && btrfs_root_refs(&root->root_item) == 0) { |
00246528 | 1556 | btrfs_put_root(root); |
48475471 | 1557 | return ERR_PTR(-ENOENT); |
bc44d7c4 | 1558 | } |
5eda7b5e | 1559 | return root; |
48475471 | 1560 | } |
5eda7b5e | 1561 | |
56e9357a DS |
1562 | key.objectid = objectid; |
1563 | key.type = BTRFS_ROOT_ITEM_KEY; | |
1564 | key.offset = (u64)-1; | |
1565 | root = btrfs_read_tree_root(fs_info->tree_root, &key); | |
5eda7b5e CM |
1566 | if (IS_ERR(root)) |
1567 | return root; | |
3394e160 | 1568 | |
c00869f1 | 1569 | if (check_ref && btrfs_root_refs(&root->root_item) == 0) { |
cb517eab | 1570 | ret = -ENOENT; |
581bb050 | 1571 | goto fail; |
35a30d7c | 1572 | } |
581bb050 | 1573 | |
2dfb1e43 | 1574 | ret = btrfs_init_fs_root(root, anon_dev); |
ac08aedf CM |
1575 | if (ret) |
1576 | goto fail; | |
3394e160 | 1577 | |
381cf658 DS |
1578 | path = btrfs_alloc_path(); |
1579 | if (!path) { | |
1580 | ret = -ENOMEM; | |
1581 | goto fail; | |
1582 | } | |
1d4c08e0 DS |
1583 | key.objectid = BTRFS_ORPHAN_OBJECTID; |
1584 | key.type = BTRFS_ORPHAN_ITEM_KEY; | |
56e9357a | 1585 | key.offset = objectid; |
1d4c08e0 DS |
1586 | |
1587 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
381cf658 | 1588 | btrfs_free_path(path); |
d68fc57b YZ |
1589 | if (ret < 0) |
1590 | goto fail; | |
1591 | if (ret == 0) | |
27cdeb70 | 1592 | set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); |
d68fc57b | 1593 | |
cb517eab | 1594 | ret = btrfs_insert_fs_root(fs_info, root); |
0f7d52f4 | 1595 | if (ret) { |
00246528 | 1596 | btrfs_put_root(root); |
4785e24f | 1597 | if (ret == -EEXIST) |
4df27c4d | 1598 | goto again; |
4df27c4d | 1599 | goto fail; |
0f7d52f4 | 1600 | } |
edbd8d4e | 1601 | return root; |
4df27c4d | 1602 | fail: |
8c38938c | 1603 | btrfs_put_root(root); |
4df27c4d | 1604 | return ERR_PTR(ret); |
edbd8d4e CM |
1605 | } |
1606 | ||
2dfb1e43 QW |
1607 | /* |
1608 | * Get in-memory reference of a root structure | |
1609 | * | |
1610 | * @objectid: tree objectid | |
1611 | * @check_ref: if set, verify that the tree exists and the item has at least | |
1612 | * one reference | |
1613 | */ | |
1614 | struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, | |
1615 | u64 objectid, bool check_ref) | |
1616 | { | |
1617 | return btrfs_get_root_ref(fs_info, objectid, 0, check_ref); | |
1618 | } | |
1619 | ||
1620 | /* | |
1621 | * Get in-memory reference of a root structure, created as new, optionally pass | |
1622 | * the anonymous block device id | |
1623 | * | |
1624 | * @objectid: tree objectid | |
1625 | * @anon_dev: if zero, allocate a new anonymous block device or use the | |
1626 | * parameter value | |
1627 | */ | |
1628 | struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info, | |
1629 | u64 objectid, dev_t anon_dev) | |
1630 | { | |
1631 | return btrfs_get_root_ref(fs_info, objectid, anon_dev, true); | |
1632 | } | |
1633 | ||
49d11bea JB |
1634 | /* |
1635 | * btrfs_get_fs_root_commit_root - return a root for the given objectid | |
1636 | * @fs_info: the fs_info | |
1637 | * @objectid: the objectid we need to lookup | |
1638 | * | |
1639 | * This is exclusively used for backref walking, and exists specifically because | |
1640 | * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref | |
1641 | * creation time, which means we may have to read the tree_root in order to look | |
1642 | * up a fs root that is not in memory. If the root is not in memory we will | |
1643 | * read the tree root commit root and look up the fs root from there. This is a | |
1644 | * temporary root, it will not be inserted into the radix tree as it doesn't | |
1645 | * have the most uptodate information, it'll simply be discarded once the | |
1646 | * backref code is finished using the root. | |
1647 | */ | |
1648 | struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info, | |
1649 | struct btrfs_path *path, | |
1650 | u64 objectid) | |
1651 | { | |
1652 | struct btrfs_root *root; | |
1653 | struct btrfs_key key; | |
1654 | ||
1655 | ASSERT(path->search_commit_root && path->skip_locking); | |
1656 | ||
1657 | /* | |
1658 | * This can return -ENOENT if we ask for a root that doesn't exist, but | |
1659 | * since this is called via the backref walking code we won't be looking | |
1660 | * up a root that doesn't exist, unless there's corruption. So if root | |
1661 | * != NULL just return it. | |
1662 | */ | |
1663 | root = btrfs_get_global_root(fs_info, objectid); | |
1664 | if (root) | |
1665 | return root; | |
1666 | ||
1667 | root = btrfs_lookup_fs_root(fs_info, objectid); | |
1668 | if (root) | |
1669 | return root; | |
1670 | ||
1671 | key.objectid = objectid; | |
1672 | key.type = BTRFS_ROOT_ITEM_KEY; | |
1673 | key.offset = (u64)-1; | |
1674 | root = read_tree_root_path(fs_info->tree_root, path, &key); | |
1675 | btrfs_release_path(path); | |
1676 | ||
1677 | return root; | |
1678 | } | |
1679 | ||
8b712842 CM |
1680 | /* |
1681 | * called by the kthread helper functions to finally call the bio end_io | |
1682 | * functions. This is where read checksum verification actually happens | |
1683 | */ | |
1684 | static void end_workqueue_fn(struct btrfs_work *work) | |
ce9adaa5 | 1685 | { |
ce9adaa5 | 1686 | struct bio *bio; |
97eb6b69 | 1687 | struct btrfs_end_io_wq *end_io_wq; |
ce9adaa5 | 1688 | |
97eb6b69 | 1689 | end_io_wq = container_of(work, struct btrfs_end_io_wq, work); |
8b712842 | 1690 | bio = end_io_wq->bio; |
ce9adaa5 | 1691 | |
4e4cbee9 | 1692 | bio->bi_status = end_io_wq->status; |
8b712842 CM |
1693 | bio->bi_private = end_io_wq->private; |
1694 | bio->bi_end_io = end_io_wq->end_io; | |
4246a0b6 | 1695 | bio_endio(bio); |
9be490f1 | 1696 | kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq); |
44b8bd7e CM |
1697 | } |
1698 | ||
a74a4b97 CM |
1699 | static int cleaner_kthread(void *arg) |
1700 | { | |
1701 | struct btrfs_root *root = arg; | |
0b246afa | 1702 | struct btrfs_fs_info *fs_info = root->fs_info; |
d0278245 | 1703 | int again; |
a74a4b97 | 1704 | |
d6fd0ae2 | 1705 | while (1) { |
d0278245 | 1706 | again = 0; |
a74a4b97 | 1707 | |
fd340d0f JB |
1708 | set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); |
1709 | ||
d0278245 | 1710 | /* Make the cleaner go to sleep early. */ |
2ff7e61e | 1711 | if (btrfs_need_cleaner_sleep(fs_info)) |
d0278245 MX |
1712 | goto sleep; |
1713 | ||
90c711ab ZB |
1714 | /* |
1715 | * Do not do anything if we might cause open_ctree() to block | |
1716 | * before we have finished mounting the filesystem. | |
1717 | */ | |
0b246afa | 1718 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
90c711ab ZB |
1719 | goto sleep; |
1720 | ||
0b246afa | 1721 | if (!mutex_trylock(&fs_info->cleaner_mutex)) |
d0278245 MX |
1722 | goto sleep; |
1723 | ||
dc7f370c MX |
1724 | /* |
1725 | * Avoid the problem that we change the status of the fs | |
1726 | * during the above check and trylock. | |
1727 | */ | |
2ff7e61e | 1728 | if (btrfs_need_cleaner_sleep(fs_info)) { |
0b246afa | 1729 | mutex_unlock(&fs_info->cleaner_mutex); |
dc7f370c | 1730 | goto sleep; |
76dda93c | 1731 | } |
a74a4b97 | 1732 | |
2ff7e61e | 1733 | btrfs_run_delayed_iputs(fs_info); |
c2d6cb16 | 1734 | |
d0278245 | 1735 | again = btrfs_clean_one_deleted_snapshot(root); |
0b246afa | 1736 | mutex_unlock(&fs_info->cleaner_mutex); |
d0278245 MX |
1737 | |
1738 | /* | |
05323cd1 MX |
1739 | * The defragger has dealt with the R/O remount and umount, |
1740 | * needn't do anything special here. | |
d0278245 | 1741 | */ |
0b246afa | 1742 | btrfs_run_defrag_inodes(fs_info); |
67c5e7d4 FM |
1743 | |
1744 | /* | |
1745 | * Acquires fs_info->delete_unused_bgs_mutex to avoid racing | |
1746 | * with relocation (btrfs_relocate_chunk) and relocation | |
1747 | * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group) | |
1748 | * after acquiring fs_info->delete_unused_bgs_mutex. So we | |
1749 | * can't hold, nor need to, fs_info->cleaner_mutex when deleting | |
1750 | * unused block groups. | |
1751 | */ | |
0b246afa | 1752 | btrfs_delete_unused_bgs(fs_info); |
d0278245 | 1753 | sleep: |
fd340d0f | 1754 | clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags); |
d6fd0ae2 OS |
1755 | if (kthread_should_park()) |
1756 | kthread_parkme(); | |
1757 | if (kthread_should_stop()) | |
1758 | return 0; | |
838fe188 | 1759 | if (!again) { |
a74a4b97 | 1760 | set_current_state(TASK_INTERRUPTIBLE); |
d6fd0ae2 | 1761 | schedule(); |
a74a4b97 CM |
1762 | __set_current_state(TASK_RUNNING); |
1763 | } | |
da288d28 | 1764 | } |
a74a4b97 CM |
1765 | } |
1766 | ||
1767 | static int transaction_kthread(void *arg) | |
1768 | { | |
1769 | struct btrfs_root *root = arg; | |
0b246afa | 1770 | struct btrfs_fs_info *fs_info = root->fs_info; |
a74a4b97 CM |
1771 | struct btrfs_trans_handle *trans; |
1772 | struct btrfs_transaction *cur; | |
8929ecfa | 1773 | u64 transid; |
643900be | 1774 | time64_t delta; |
a74a4b97 | 1775 | unsigned long delay; |
914b2007 | 1776 | bool cannot_commit; |
a74a4b97 CM |
1777 | |
1778 | do { | |
914b2007 | 1779 | cannot_commit = false; |
ba1bc00f | 1780 | delay = msecs_to_jiffies(fs_info->commit_interval * 1000); |
0b246afa | 1781 | mutex_lock(&fs_info->transaction_kthread_mutex); |
a74a4b97 | 1782 | |
0b246afa JM |
1783 | spin_lock(&fs_info->trans_lock); |
1784 | cur = fs_info->running_transaction; | |
a74a4b97 | 1785 | if (!cur) { |
0b246afa | 1786 | spin_unlock(&fs_info->trans_lock); |
a74a4b97 CM |
1787 | goto sleep; |
1788 | } | |
31153d81 | 1789 | |
643900be | 1790 | delta = ktime_get_seconds() - cur->start_time; |
3296bf56 | 1791 | if (cur->state < TRANS_STATE_COMMIT_START && |
643900be | 1792 | delta < fs_info->commit_interval) { |
0b246afa | 1793 | spin_unlock(&fs_info->trans_lock); |
fb8a7e94 NB |
1794 | delay -= msecs_to_jiffies((delta - 1) * 1000); |
1795 | delay = min(delay, | |
1796 | msecs_to_jiffies(fs_info->commit_interval * 1000)); | |
a74a4b97 CM |
1797 | goto sleep; |
1798 | } | |
8929ecfa | 1799 | transid = cur->transid; |
0b246afa | 1800 | spin_unlock(&fs_info->trans_lock); |
56bec294 | 1801 | |
79787eaa | 1802 | /* If the file system is aborted, this will always fail. */ |
354aa0fb | 1803 | trans = btrfs_attach_transaction(root); |
914b2007 | 1804 | if (IS_ERR(trans)) { |
354aa0fb MX |
1805 | if (PTR_ERR(trans) != -ENOENT) |
1806 | cannot_commit = true; | |
79787eaa | 1807 | goto sleep; |
914b2007 | 1808 | } |
8929ecfa | 1809 | if (transid == trans->transid) { |
3a45bb20 | 1810 | btrfs_commit_transaction(trans); |
8929ecfa | 1811 | } else { |
3a45bb20 | 1812 | btrfs_end_transaction(trans); |
8929ecfa | 1813 | } |
a74a4b97 | 1814 | sleep: |
0b246afa JM |
1815 | wake_up_process(fs_info->cleaner_kthread); |
1816 | mutex_unlock(&fs_info->transaction_kthread_mutex); | |
a74a4b97 | 1817 | |
4e121c06 | 1818 | if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, |
0b246afa | 1819 | &fs_info->fs_state))) |
2ff7e61e | 1820 | btrfs_cleanup_transaction(fs_info); |
ce63f891 | 1821 | if (!kthread_should_stop() && |
0b246afa | 1822 | (!btrfs_transaction_blocked(fs_info) || |
ce63f891 | 1823 | cannot_commit)) |
bc5511d0 | 1824 | schedule_timeout_interruptible(delay); |
a74a4b97 CM |
1825 | } while (!kthread_should_stop()); |
1826 | return 0; | |
1827 | } | |
1828 | ||
af31f5e5 | 1829 | /* |
01f0f9da NB |
1830 | * This will find the highest generation in the array of root backups. The |
1831 | * index of the highest array is returned, or -EINVAL if we can't find | |
1832 | * anything. | |
af31f5e5 CM |
1833 | * |
1834 | * We check to make sure the array is valid by comparing the | |
1835 | * generation of the latest root in the array with the generation | |
1836 | * in the super block. If they don't match we pitch it. | |
1837 | */ | |
01f0f9da | 1838 | static int find_newest_super_backup(struct btrfs_fs_info *info) |
af31f5e5 | 1839 | { |
01f0f9da | 1840 | const u64 newest_gen = btrfs_super_generation(info->super_copy); |
af31f5e5 | 1841 | u64 cur; |
af31f5e5 CM |
1842 | struct btrfs_root_backup *root_backup; |
1843 | int i; | |
1844 | ||
1845 | for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { | |
1846 | root_backup = info->super_copy->super_roots + i; | |
1847 | cur = btrfs_backup_tree_root_gen(root_backup); | |
1848 | if (cur == newest_gen) | |
01f0f9da | 1849 | return i; |
af31f5e5 CM |
1850 | } |
1851 | ||
01f0f9da | 1852 | return -EINVAL; |
af31f5e5 CM |
1853 | } |
1854 | ||
af31f5e5 CM |
1855 | /* |
1856 | * copy all the root pointers into the super backup array. | |
1857 | * this will bump the backup pointer by one when it is | |
1858 | * done | |
1859 | */ | |
1860 | static void backup_super_roots(struct btrfs_fs_info *info) | |
1861 | { | |
6ef108dd | 1862 | const int next_backup = info->backup_root_index; |
af31f5e5 | 1863 | struct btrfs_root_backup *root_backup; |
af31f5e5 CM |
1864 | |
1865 | root_backup = info->super_for_commit->super_roots + next_backup; | |
1866 | ||
1867 | /* | |
1868 | * make sure all of our padding and empty slots get zero filled | |
1869 | * regardless of which ones we use today | |
1870 | */ | |
1871 | memset(root_backup, 0, sizeof(*root_backup)); | |
1872 | ||
1873 | info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; | |
1874 | ||
1875 | btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); | |
1876 | btrfs_set_backup_tree_root_gen(root_backup, | |
1877 | btrfs_header_generation(info->tree_root->node)); | |
1878 | ||
1879 | btrfs_set_backup_tree_root_level(root_backup, | |
1880 | btrfs_header_level(info->tree_root->node)); | |
1881 | ||
1882 | btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); | |
1883 | btrfs_set_backup_chunk_root_gen(root_backup, | |
1884 | btrfs_header_generation(info->chunk_root->node)); | |
1885 | btrfs_set_backup_chunk_root_level(root_backup, | |
1886 | btrfs_header_level(info->chunk_root->node)); | |
1887 | ||
1888 | btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); | |
1889 | btrfs_set_backup_extent_root_gen(root_backup, | |
1890 | btrfs_header_generation(info->extent_root->node)); | |
1891 | btrfs_set_backup_extent_root_level(root_backup, | |
1892 | btrfs_header_level(info->extent_root->node)); | |
1893 | ||
7c7e82a7 CM |
1894 | /* |
1895 | * we might commit during log recovery, which happens before we set | |
1896 | * the fs_root. Make sure it is valid before we fill it in. | |
1897 | */ | |
1898 | if (info->fs_root && info->fs_root->node) { | |
1899 | btrfs_set_backup_fs_root(root_backup, | |
1900 | info->fs_root->node->start); | |
1901 | btrfs_set_backup_fs_root_gen(root_backup, | |
af31f5e5 | 1902 | btrfs_header_generation(info->fs_root->node)); |
7c7e82a7 | 1903 | btrfs_set_backup_fs_root_level(root_backup, |
af31f5e5 | 1904 | btrfs_header_level(info->fs_root->node)); |
7c7e82a7 | 1905 | } |
af31f5e5 CM |
1906 | |
1907 | btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); | |
1908 | btrfs_set_backup_dev_root_gen(root_backup, | |
1909 | btrfs_header_generation(info->dev_root->node)); | |
1910 | btrfs_set_backup_dev_root_level(root_backup, | |
1911 | btrfs_header_level(info->dev_root->node)); | |
1912 | ||
1913 | btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); | |
1914 | btrfs_set_backup_csum_root_gen(root_backup, | |
1915 | btrfs_header_generation(info->csum_root->node)); | |
1916 | btrfs_set_backup_csum_root_level(root_backup, | |
1917 | btrfs_header_level(info->csum_root->node)); | |
1918 | ||
1919 | btrfs_set_backup_total_bytes(root_backup, | |
1920 | btrfs_super_total_bytes(info->super_copy)); | |
1921 | btrfs_set_backup_bytes_used(root_backup, | |
1922 | btrfs_super_bytes_used(info->super_copy)); | |
1923 | btrfs_set_backup_num_devices(root_backup, | |
1924 | btrfs_super_num_devices(info->super_copy)); | |
1925 | ||
1926 | /* | |
1927 | * if we don't copy this out to the super_copy, it won't get remembered | |
1928 | * for the next commit | |
1929 | */ | |
1930 | memcpy(&info->super_copy->super_roots, | |
1931 | &info->super_for_commit->super_roots, | |
1932 | sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); | |
1933 | } | |
1934 | ||
bd2336b2 NB |
1935 | /* |
1936 | * read_backup_root - Reads a backup root based on the passed priority. Prio 0 | |
1937 | * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots | |
1938 | * | |
1939 | * fs_info - filesystem whose backup roots need to be read | |
1940 | * priority - priority of backup root required | |
1941 | * | |
1942 | * Returns backup root index on success and -EINVAL otherwise. | |
1943 | */ | |
1944 | static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority) | |
1945 | { | |
1946 | int backup_index = find_newest_super_backup(fs_info); | |
1947 | struct btrfs_super_block *super = fs_info->super_copy; | |
1948 | struct btrfs_root_backup *root_backup; | |
1949 | ||
1950 | if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) { | |
1951 | if (priority == 0) | |
1952 | return backup_index; | |
1953 | ||
1954 | backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority; | |
1955 | backup_index %= BTRFS_NUM_BACKUP_ROOTS; | |
1956 | } else { | |
1957 | return -EINVAL; | |
1958 | } | |
1959 | ||
1960 | root_backup = super->super_roots + backup_index; | |
1961 | ||
1962 | btrfs_set_super_generation(super, | |
1963 | btrfs_backup_tree_root_gen(root_backup)); | |
1964 | btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); | |
1965 | btrfs_set_super_root_level(super, | |
1966 | btrfs_backup_tree_root_level(root_backup)); | |
1967 | btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); | |
1968 | ||
1969 | /* | |
1970 | * Fixme: the total bytes and num_devices need to match or we should | |
1971 | * need a fsck | |
1972 | */ | |
1973 | btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); | |
1974 | btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); | |
1975 | ||
1976 | return backup_index; | |
1977 | } | |
1978 | ||
7abadb64 LB |
1979 | /* helper to cleanup workers */ |
1980 | static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) | |
1981 | { | |
dc6e3209 | 1982 | btrfs_destroy_workqueue(fs_info->fixup_workers); |
afe3d242 | 1983 | btrfs_destroy_workqueue(fs_info->delalloc_workers); |
5cdc7ad3 | 1984 | btrfs_destroy_workqueue(fs_info->workers); |
fccb5d86 | 1985 | btrfs_destroy_workqueue(fs_info->endio_workers); |
fccb5d86 | 1986 | btrfs_destroy_workqueue(fs_info->endio_raid56_workers); |
d05a33ac | 1987 | btrfs_destroy_workqueue(fs_info->rmw_workers); |
fccb5d86 QW |
1988 | btrfs_destroy_workqueue(fs_info->endio_write_workers); |
1989 | btrfs_destroy_workqueue(fs_info->endio_freespace_worker); | |
5b3bc44e | 1990 | btrfs_destroy_workqueue(fs_info->delayed_workers); |
e66f0bb1 | 1991 | btrfs_destroy_workqueue(fs_info->caching_workers); |
736cfa15 | 1992 | btrfs_destroy_workqueue(fs_info->readahead_workers); |
a44903ab | 1993 | btrfs_destroy_workqueue(fs_info->flush_workers); |
fc97fab0 | 1994 | btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); |
b0643e59 DZ |
1995 | if (fs_info->discard_ctl.discard_workers) |
1996 | destroy_workqueue(fs_info->discard_ctl.discard_workers); | |
a9b9477d FM |
1997 | /* |
1998 | * Now that all other work queues are destroyed, we can safely destroy | |
1999 | * the queues used for metadata I/O, since tasks from those other work | |
2000 | * queues can do metadata I/O operations. | |
2001 | */ | |
2002 | btrfs_destroy_workqueue(fs_info->endio_meta_workers); | |
2003 | btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); | |
7abadb64 LB |
2004 | } |
2005 | ||
2e9f5954 R |
2006 | static void free_root_extent_buffers(struct btrfs_root *root) |
2007 | { | |
2008 | if (root) { | |
2009 | free_extent_buffer(root->node); | |
2010 | free_extent_buffer(root->commit_root); | |
2011 | root->node = NULL; | |
2012 | root->commit_root = NULL; | |
2013 | } | |
2014 | } | |
2015 | ||
af31f5e5 | 2016 | /* helper to cleanup tree roots */ |
4273eaff | 2017 | static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root) |
af31f5e5 | 2018 | { |
2e9f5954 | 2019 | free_root_extent_buffers(info->tree_root); |
655b09fe | 2020 | |
2e9f5954 R |
2021 | free_root_extent_buffers(info->dev_root); |
2022 | free_root_extent_buffers(info->extent_root); | |
2023 | free_root_extent_buffers(info->csum_root); | |
2024 | free_root_extent_buffers(info->quota_root); | |
2025 | free_root_extent_buffers(info->uuid_root); | |
8c38938c | 2026 | free_root_extent_buffers(info->fs_root); |
aeb935a4 | 2027 | free_root_extent_buffers(info->data_reloc_root); |
4273eaff | 2028 | if (free_chunk_root) |
2e9f5954 | 2029 | free_root_extent_buffers(info->chunk_root); |
70f6d82e | 2030 | free_root_extent_buffers(info->free_space_root); |
af31f5e5 CM |
2031 | } |
2032 | ||
8c38938c JB |
2033 | void btrfs_put_root(struct btrfs_root *root) |
2034 | { | |
2035 | if (!root) | |
2036 | return; | |
2037 | ||
2038 | if (refcount_dec_and_test(&root->refs)) { | |
2039 | WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); | |
1dae7e0e | 2040 | WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)); |
8c38938c JB |
2041 | if (root->anon_dev) |
2042 | free_anon_bdev(root->anon_dev); | |
2043 | btrfs_drew_lock_destroy(&root->snapshot_lock); | |
923eb523 | 2044 | free_root_extent_buffers(root); |
8c38938c JB |
2045 | kfree(root->free_ino_ctl); |
2046 | kfree(root->free_ino_pinned); | |
2047 | #ifdef CONFIG_BTRFS_DEBUG | |
2048 | spin_lock(&root->fs_info->fs_roots_radix_lock); | |
2049 | list_del_init(&root->leak_list); | |
2050 | spin_unlock(&root->fs_info->fs_roots_radix_lock); | |
2051 | #endif | |
2052 | kfree(root); | |
2053 | } | |
2054 | } | |
2055 | ||
faa2dbf0 | 2056 | void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) |
171f6537 JB |
2057 | { |
2058 | int ret; | |
2059 | struct btrfs_root *gang[8]; | |
2060 | int i; | |
2061 | ||
2062 | while (!list_empty(&fs_info->dead_roots)) { | |
2063 | gang[0] = list_entry(fs_info->dead_roots.next, | |
2064 | struct btrfs_root, root_list); | |
2065 | list_del(&gang[0]->root_list); | |
2066 | ||
8c38938c | 2067 | if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) |
cb517eab | 2068 | btrfs_drop_and_free_fs_root(fs_info, gang[0]); |
dc9492c1 | 2069 | btrfs_put_root(gang[0]); |
171f6537 JB |
2070 | } |
2071 | ||
2072 | while (1) { | |
2073 | ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, | |
2074 | (void **)gang, 0, | |
2075 | ARRAY_SIZE(gang)); | |
2076 | if (!ret) | |
2077 | break; | |
2078 | for (i = 0; i < ret; i++) | |
cb517eab | 2079 | btrfs_drop_and_free_fs_root(fs_info, gang[i]); |
171f6537 JB |
2080 | } |
2081 | } | |
af31f5e5 | 2082 | |
638aa7ed ES |
2083 | static void btrfs_init_scrub(struct btrfs_fs_info *fs_info) |
2084 | { | |
2085 | mutex_init(&fs_info->scrub_lock); | |
2086 | atomic_set(&fs_info->scrubs_running, 0); | |
2087 | atomic_set(&fs_info->scrub_pause_req, 0); | |
2088 | atomic_set(&fs_info->scrubs_paused, 0); | |
2089 | atomic_set(&fs_info->scrub_cancel_req, 0); | |
2090 | init_waitqueue_head(&fs_info->scrub_pause_wait); | |
ff09c4ca | 2091 | refcount_set(&fs_info->scrub_workers_refcnt, 0); |
638aa7ed ES |
2092 | } |
2093 | ||
779a65a4 ES |
2094 | static void btrfs_init_balance(struct btrfs_fs_info *fs_info) |
2095 | { | |
2096 | spin_lock_init(&fs_info->balance_lock); | |
2097 | mutex_init(&fs_info->balance_mutex); | |
779a65a4 ES |
2098 | atomic_set(&fs_info->balance_pause_req, 0); |
2099 | atomic_set(&fs_info->balance_cancel_req, 0); | |
2100 | fs_info->balance_ctl = NULL; | |
2101 | init_waitqueue_head(&fs_info->balance_wait_q); | |
2102 | } | |
2103 | ||
6bccf3ab | 2104 | static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info) |
f37938e0 | 2105 | { |
2ff7e61e JM |
2106 | struct inode *inode = fs_info->btree_inode; |
2107 | ||
2108 | inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; | |
2109 | set_nlink(inode, 1); | |
f37938e0 ES |
2110 | /* |
2111 | * we set the i_size on the btree inode to the max possible int. | |
2112 | * the real end of the address space is determined by all of | |
2113 | * the devices in the system | |
2114 | */ | |
2ff7e61e JM |
2115 | inode->i_size = OFFSET_MAX; |
2116 | inode->i_mapping->a_ops = &btree_aops; | |
f37938e0 | 2117 | |
2ff7e61e | 2118 | RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node); |
43eb5f29 | 2119 | extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree, |
2c53a14d | 2120 | IO_TREE_BTREE_INODE_IO, inode); |
7b439738 | 2121 | BTRFS_I(inode)->io_tree.track_uptodate = false; |
2ff7e61e | 2122 | extent_map_tree_init(&BTRFS_I(inode)->extent_tree); |
f37938e0 | 2123 | |
5c8fd99f | 2124 | BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root); |
2ff7e61e JM |
2125 | memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key)); |
2126 | set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags); | |
2127 | btrfs_insert_inode_hash(inode); | |
f37938e0 ES |
2128 | } |
2129 | ||
ad618368 ES |
2130 | static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info) |
2131 | { | |
ad618368 | 2132 | mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); |
129827e3 | 2133 | init_rwsem(&fs_info->dev_replace.rwsem); |
7f8d236a | 2134 | init_waitqueue_head(&fs_info->dev_replace.replace_wait); |
ad618368 ES |
2135 | } |
2136 | ||
f9e92e40 ES |
2137 | static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info) |
2138 | { | |
2139 | spin_lock_init(&fs_info->qgroup_lock); | |
2140 | mutex_init(&fs_info->qgroup_ioctl_lock); | |
2141 | fs_info->qgroup_tree = RB_ROOT; | |
f9e92e40 ES |
2142 | INIT_LIST_HEAD(&fs_info->dirty_qgroups); |
2143 | fs_info->qgroup_seq = 1; | |
f9e92e40 | 2144 | fs_info->qgroup_ulist = NULL; |
d2c609b8 | 2145 | fs_info->qgroup_rescan_running = false; |
f9e92e40 ES |
2146 | mutex_init(&fs_info->qgroup_rescan_lock); |
2147 | } | |
2148 | ||
2a458198 ES |
2149 | static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info, |
2150 | struct btrfs_fs_devices *fs_devices) | |
2151 | { | |
f7b885be | 2152 | u32 max_active = fs_info->thread_pool_size; |
6f011058 | 2153 | unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; |
2a458198 ES |
2154 | |
2155 | fs_info->workers = | |
cb001095 JM |
2156 | btrfs_alloc_workqueue(fs_info, "worker", |
2157 | flags | WQ_HIGHPRI, max_active, 16); | |
2a458198 ES |
2158 | |
2159 | fs_info->delalloc_workers = | |
cb001095 JM |
2160 | btrfs_alloc_workqueue(fs_info, "delalloc", |
2161 | flags, max_active, 2); | |
2a458198 ES |
2162 | |
2163 | fs_info->flush_workers = | |
cb001095 JM |
2164 | btrfs_alloc_workqueue(fs_info, "flush_delalloc", |
2165 | flags, max_active, 0); | |
2a458198 ES |
2166 | |
2167 | fs_info->caching_workers = | |
cb001095 | 2168 | btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0); |
2a458198 | 2169 | |
2a458198 | 2170 | fs_info->fixup_workers = |
cb001095 | 2171 | btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0); |
2a458198 ES |
2172 | |
2173 | /* | |
2174 | * endios are largely parallel and should have a very | |
2175 | * low idle thresh | |
2176 | */ | |
2177 | fs_info->endio_workers = | |
cb001095 | 2178 | btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4); |
2a458198 | 2179 | fs_info->endio_meta_workers = |
cb001095 JM |
2180 | btrfs_alloc_workqueue(fs_info, "endio-meta", flags, |
2181 | max_active, 4); | |
2a458198 | 2182 | fs_info->endio_meta_write_workers = |
cb001095 JM |
2183 | btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags, |
2184 | max_active, 2); | |
2a458198 | 2185 | fs_info->endio_raid56_workers = |
cb001095 JM |
2186 | btrfs_alloc_workqueue(fs_info, "endio-raid56", flags, |
2187 | max_active, 4); | |
2a458198 | 2188 | fs_info->rmw_workers = |
cb001095 | 2189 | btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2); |
2a458198 | 2190 | fs_info->endio_write_workers = |
cb001095 JM |
2191 | btrfs_alloc_workqueue(fs_info, "endio-write", flags, |
2192 | max_active, 2); | |
2a458198 | 2193 | fs_info->endio_freespace_worker = |
cb001095 JM |
2194 | btrfs_alloc_workqueue(fs_info, "freespace-write", flags, |
2195 | max_active, 0); | |
2a458198 | 2196 | fs_info->delayed_workers = |
cb001095 JM |
2197 | btrfs_alloc_workqueue(fs_info, "delayed-meta", flags, |
2198 | max_active, 0); | |
2a458198 | 2199 | fs_info->readahead_workers = |
cb001095 JM |
2200 | btrfs_alloc_workqueue(fs_info, "readahead", flags, |
2201 | max_active, 2); | |
2a458198 | 2202 | fs_info->qgroup_rescan_workers = |
cb001095 | 2203 | btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0); |
b0643e59 DZ |
2204 | fs_info->discard_ctl.discard_workers = |
2205 | alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1); | |
2a458198 ES |
2206 | |
2207 | if (!(fs_info->workers && fs_info->delalloc_workers && | |
ba8a9d07 | 2208 | fs_info->flush_workers && |
2a458198 ES |
2209 | fs_info->endio_workers && fs_info->endio_meta_workers && |
2210 | fs_info->endio_meta_write_workers && | |
2a458198 ES |
2211 | fs_info->endio_write_workers && fs_info->endio_raid56_workers && |
2212 | fs_info->endio_freespace_worker && fs_info->rmw_workers && | |
2213 | fs_info->caching_workers && fs_info->readahead_workers && | |
2214 | fs_info->fixup_workers && fs_info->delayed_workers && | |
b0643e59 DZ |
2215 | fs_info->qgroup_rescan_workers && |
2216 | fs_info->discard_ctl.discard_workers)) { | |
2a458198 ES |
2217 | return -ENOMEM; |
2218 | } | |
2219 | ||
2220 | return 0; | |
2221 | } | |
2222 | ||
6d97c6e3 JT |
2223 | static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type) |
2224 | { | |
2225 | struct crypto_shash *csum_shash; | |
b4e967be | 2226 | const char *csum_driver = btrfs_super_csum_driver(csum_type); |
6d97c6e3 | 2227 | |
b4e967be | 2228 | csum_shash = crypto_alloc_shash(csum_driver, 0, 0); |
6d97c6e3 JT |
2229 | |
2230 | if (IS_ERR(csum_shash)) { | |
2231 | btrfs_err(fs_info, "error allocating %s hash for checksum", | |
b4e967be | 2232 | csum_driver); |
6d97c6e3 JT |
2233 | return PTR_ERR(csum_shash); |
2234 | } | |
2235 | ||
2236 | fs_info->csum_shash = csum_shash; | |
2237 | ||
2238 | return 0; | |
2239 | } | |
2240 | ||
63443bf5 ES |
2241 | static int btrfs_replay_log(struct btrfs_fs_info *fs_info, |
2242 | struct btrfs_fs_devices *fs_devices) | |
2243 | { | |
2244 | int ret; | |
63443bf5 ES |
2245 | struct btrfs_root *log_tree_root; |
2246 | struct btrfs_super_block *disk_super = fs_info->super_copy; | |
2247 | u64 bytenr = btrfs_super_log_root(disk_super); | |
581c1760 | 2248 | int level = btrfs_super_log_root_level(disk_super); |
63443bf5 ES |
2249 | |
2250 | if (fs_devices->rw_devices == 0) { | |
f14d104d | 2251 | btrfs_warn(fs_info, "log replay required on RO media"); |
63443bf5 ES |
2252 | return -EIO; |
2253 | } | |
2254 | ||
96dfcb46 JB |
2255 | log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, |
2256 | GFP_KERNEL); | |
63443bf5 ES |
2257 | if (!log_tree_root) |
2258 | return -ENOMEM; | |
2259 | ||
2ff7e61e | 2260 | log_tree_root->node = read_tree_block(fs_info, bytenr, |
581c1760 QW |
2261 | fs_info->generation + 1, |
2262 | level, NULL); | |
64c043de | 2263 | if (IS_ERR(log_tree_root->node)) { |
f14d104d | 2264 | btrfs_warn(fs_info, "failed to read log tree"); |
0eeff236 | 2265 | ret = PTR_ERR(log_tree_root->node); |
8c38938c | 2266 | log_tree_root->node = NULL; |
00246528 | 2267 | btrfs_put_root(log_tree_root); |
0eeff236 | 2268 | return ret; |
64c043de | 2269 | } else if (!extent_buffer_uptodate(log_tree_root->node)) { |
f14d104d | 2270 | btrfs_err(fs_info, "failed to read log tree"); |
00246528 | 2271 | btrfs_put_root(log_tree_root); |
63443bf5 ES |
2272 | return -EIO; |
2273 | } | |
2274 | /* returns with log_tree_root freed on success */ | |
2275 | ret = btrfs_recover_log_trees(log_tree_root); | |
2276 | if (ret) { | |
0b246afa JM |
2277 | btrfs_handle_fs_error(fs_info, ret, |
2278 | "Failed to recover log tree"); | |
00246528 | 2279 | btrfs_put_root(log_tree_root); |
63443bf5 ES |
2280 | return ret; |
2281 | } | |
2282 | ||
bc98a42c | 2283 | if (sb_rdonly(fs_info->sb)) { |
6bccf3ab | 2284 | ret = btrfs_commit_super(fs_info); |
63443bf5 ES |
2285 | if (ret) |
2286 | return ret; | |
2287 | } | |
2288 | ||
2289 | return 0; | |
2290 | } | |
2291 | ||
6bccf3ab | 2292 | static int btrfs_read_roots(struct btrfs_fs_info *fs_info) |
4bbcaa64 | 2293 | { |
6bccf3ab | 2294 | struct btrfs_root *tree_root = fs_info->tree_root; |
a4f3d2c4 | 2295 | struct btrfs_root *root; |
4bbcaa64 ES |
2296 | struct btrfs_key location; |
2297 | int ret; | |
2298 | ||
6bccf3ab JM |
2299 | BUG_ON(!fs_info->tree_root); |
2300 | ||
4bbcaa64 ES |
2301 | location.objectid = BTRFS_EXTENT_TREE_OBJECTID; |
2302 | location.type = BTRFS_ROOT_ITEM_KEY; | |
2303 | location.offset = 0; | |
2304 | ||
a4f3d2c4 | 2305 | root = btrfs_read_tree_root(tree_root, &location); |
f50f4353 | 2306 | if (IS_ERR(root)) { |
42437a63 JB |
2307 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { |
2308 | ret = PTR_ERR(root); | |
2309 | goto out; | |
2310 | } | |
2311 | } else { | |
2312 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2313 | fs_info->extent_root = root; | |
f50f4353 | 2314 | } |
4bbcaa64 ES |
2315 | |
2316 | location.objectid = BTRFS_DEV_TREE_OBJECTID; | |
a4f3d2c4 | 2317 | root = btrfs_read_tree_root(tree_root, &location); |
f50f4353 | 2318 | if (IS_ERR(root)) { |
42437a63 JB |
2319 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { |
2320 | ret = PTR_ERR(root); | |
2321 | goto out; | |
2322 | } | |
2323 | } else { | |
2324 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2325 | fs_info->dev_root = root; | |
2326 | btrfs_init_devices_late(fs_info); | |
f50f4353 | 2327 | } |
4bbcaa64 | 2328 | |
882dbe0c JB |
2329 | /* If IGNOREDATACSUMS is set don't bother reading the csum root. */ |
2330 | if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) { | |
2331 | location.objectid = BTRFS_CSUM_TREE_OBJECTID; | |
2332 | root = btrfs_read_tree_root(tree_root, &location); | |
2333 | if (IS_ERR(root)) { | |
2334 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { | |
2335 | ret = PTR_ERR(root); | |
2336 | goto out; | |
2337 | } | |
2338 | } else { | |
2339 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2340 | fs_info->csum_root = root; | |
42437a63 | 2341 | } |
f50f4353 | 2342 | } |
4bbcaa64 | 2343 | |
aeb935a4 QW |
2344 | /* |
2345 | * This tree can share blocks with some other fs tree during relocation | |
2346 | * and we need a proper setup by btrfs_get_fs_root | |
2347 | */ | |
56e9357a DS |
2348 | root = btrfs_get_fs_root(tree_root->fs_info, |
2349 | BTRFS_DATA_RELOC_TREE_OBJECTID, true); | |
aeb935a4 | 2350 | if (IS_ERR(root)) { |
42437a63 JB |
2351 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { |
2352 | ret = PTR_ERR(root); | |
2353 | goto out; | |
2354 | } | |
2355 | } else { | |
2356 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2357 | fs_info->data_reloc_root = root; | |
aeb935a4 | 2358 | } |
aeb935a4 | 2359 | |
4bbcaa64 | 2360 | location.objectid = BTRFS_QUOTA_TREE_OBJECTID; |
a4f3d2c4 DS |
2361 | root = btrfs_read_tree_root(tree_root, &location); |
2362 | if (!IS_ERR(root)) { | |
2363 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
afcdd129 | 2364 | set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags); |
a4f3d2c4 | 2365 | fs_info->quota_root = root; |
4bbcaa64 ES |
2366 | } |
2367 | ||
2368 | location.objectid = BTRFS_UUID_TREE_OBJECTID; | |
a4f3d2c4 DS |
2369 | root = btrfs_read_tree_root(tree_root, &location); |
2370 | if (IS_ERR(root)) { | |
42437a63 JB |
2371 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { |
2372 | ret = PTR_ERR(root); | |
2373 | if (ret != -ENOENT) | |
2374 | goto out; | |
2375 | } | |
4bbcaa64 | 2376 | } else { |
a4f3d2c4 DS |
2377 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); |
2378 | fs_info->uuid_root = root; | |
4bbcaa64 ES |
2379 | } |
2380 | ||
70f6d82e OS |
2381 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { |
2382 | location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID; | |
2383 | root = btrfs_read_tree_root(tree_root, &location); | |
f50f4353 | 2384 | if (IS_ERR(root)) { |
42437a63 JB |
2385 | if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) { |
2386 | ret = PTR_ERR(root); | |
2387 | goto out; | |
2388 | } | |
2389 | } else { | |
2390 | set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); | |
2391 | fs_info->free_space_root = root; | |
f50f4353 | 2392 | } |
70f6d82e OS |
2393 | } |
2394 | ||
4bbcaa64 | 2395 | return 0; |
f50f4353 LB |
2396 | out: |
2397 | btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d", | |
2398 | location.objectid, ret); | |
2399 | return ret; | |
4bbcaa64 ES |
2400 | } |
2401 | ||
069ec957 QW |
2402 | /* |
2403 | * Real super block validation | |
2404 | * NOTE: super csum type and incompat features will not be checked here. | |
2405 | * | |
2406 | * @sb: super block to check | |
2407 | * @mirror_num: the super block number to check its bytenr: | |
2408 | * 0 the primary (1st) sb | |
2409 | * 1, 2 2nd and 3rd backup copy | |
2410 | * -1 skip bytenr check | |
2411 | */ | |
2412 | static int validate_super(struct btrfs_fs_info *fs_info, | |
2413 | struct btrfs_super_block *sb, int mirror_num) | |
21a852b0 | 2414 | { |
21a852b0 QW |
2415 | u64 nodesize = btrfs_super_nodesize(sb); |
2416 | u64 sectorsize = btrfs_super_sectorsize(sb); | |
2417 | int ret = 0; | |
2418 | ||
2419 | if (btrfs_super_magic(sb) != BTRFS_MAGIC) { | |
2420 | btrfs_err(fs_info, "no valid FS found"); | |
2421 | ret = -EINVAL; | |
2422 | } | |
2423 | if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) { | |
2424 | btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu", | |
2425 | btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP); | |
2426 | ret = -EINVAL; | |
2427 | } | |
2428 | if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) { | |
2429 | btrfs_err(fs_info, "tree_root level too big: %d >= %d", | |
2430 | btrfs_super_root_level(sb), BTRFS_MAX_LEVEL); | |
2431 | ret = -EINVAL; | |
2432 | } | |
2433 | if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) { | |
2434 | btrfs_err(fs_info, "chunk_root level too big: %d >= %d", | |
2435 | btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL); | |
2436 | ret = -EINVAL; | |
2437 | } | |
2438 | if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) { | |
2439 | btrfs_err(fs_info, "log_root level too big: %d >= %d", | |
2440 | btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL); | |
2441 | ret = -EINVAL; | |
2442 | } | |
2443 | ||
2444 | /* | |
2445 | * Check sectorsize and nodesize first, other check will need it. | |
2446 | * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here. | |
2447 | */ | |
2448 | if (!is_power_of_2(sectorsize) || sectorsize < 4096 || | |
2449 | sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) { | |
2450 | btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize); | |
2451 | ret = -EINVAL; | |
2452 | } | |
2453 | /* Only PAGE SIZE is supported yet */ | |
2454 | if (sectorsize != PAGE_SIZE) { | |
2455 | btrfs_err(fs_info, | |
2456 | "sectorsize %llu not supported yet, only support %lu", | |
2457 | sectorsize, PAGE_SIZE); | |
2458 | ret = -EINVAL; | |
2459 | } | |
2460 | if (!is_power_of_2(nodesize) || nodesize < sectorsize || | |
2461 | nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) { | |
2462 | btrfs_err(fs_info, "invalid nodesize %llu", nodesize); | |
2463 | ret = -EINVAL; | |
2464 | } | |
2465 | if (nodesize != le32_to_cpu(sb->__unused_leafsize)) { | |
2466 | btrfs_err(fs_info, "invalid leafsize %u, should be %llu", | |
2467 | le32_to_cpu(sb->__unused_leafsize), nodesize); | |
2468 | ret = -EINVAL; | |
2469 | } | |
2470 | ||
2471 | /* Root alignment check */ | |
2472 | if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) { | |
2473 | btrfs_warn(fs_info, "tree_root block unaligned: %llu", | |
2474 | btrfs_super_root(sb)); | |
2475 | ret = -EINVAL; | |
2476 | } | |
2477 | if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) { | |
2478 | btrfs_warn(fs_info, "chunk_root block unaligned: %llu", | |
2479 | btrfs_super_chunk_root(sb)); | |
2480 | ret = -EINVAL; | |
2481 | } | |
2482 | if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) { | |
2483 | btrfs_warn(fs_info, "log_root block unaligned: %llu", | |
2484 | btrfs_super_log_root(sb)); | |
2485 | ret = -EINVAL; | |
2486 | } | |
2487 | ||
de37aa51 | 2488 | if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid, |
7239ff4b | 2489 | BTRFS_FSID_SIZE) != 0) { |
21a852b0 | 2490 | btrfs_err(fs_info, |
7239ff4b | 2491 | "dev_item UUID does not match metadata fsid: %pU != %pU", |
de37aa51 | 2492 | fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid); |
21a852b0 QW |
2493 | ret = -EINVAL; |
2494 | } | |
2495 | ||
2496 | /* | |
2497 | * Hint to catch really bogus numbers, bitflips or so, more exact checks are | |
2498 | * done later | |
2499 | */ | |
2500 | if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) { | |
2501 | btrfs_err(fs_info, "bytes_used is too small %llu", | |
2502 | btrfs_super_bytes_used(sb)); | |
2503 | ret = -EINVAL; | |
2504 | } | |
2505 | if (!is_power_of_2(btrfs_super_stripesize(sb))) { | |
2506 | btrfs_err(fs_info, "invalid stripesize %u", | |
2507 | btrfs_super_stripesize(sb)); | |
2508 | ret = -EINVAL; | |
2509 | } | |
2510 | if (btrfs_super_num_devices(sb) > (1UL << 31)) | |
2511 | btrfs_warn(fs_info, "suspicious number of devices: %llu", | |
2512 | btrfs_super_num_devices(sb)); | |
2513 | if (btrfs_super_num_devices(sb) == 0) { | |
2514 | btrfs_err(fs_info, "number of devices is 0"); | |
2515 | ret = -EINVAL; | |
2516 | } | |
2517 | ||
069ec957 QW |
2518 | if (mirror_num >= 0 && |
2519 | btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) { | |
21a852b0 QW |
2520 | btrfs_err(fs_info, "super offset mismatch %llu != %u", |
2521 | btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET); | |
2522 | ret = -EINVAL; | |
2523 | } | |
2524 | ||
2525 | /* | |
2526 | * Obvious sys_chunk_array corruptions, it must hold at least one key | |
2527 | * and one chunk | |
2528 | */ | |
2529 | if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { | |
2530 | btrfs_err(fs_info, "system chunk array too big %u > %u", | |
2531 | btrfs_super_sys_array_size(sb), | |
2532 | BTRFS_SYSTEM_CHUNK_ARRAY_SIZE); | |
2533 | ret = -EINVAL; | |
2534 | } | |
2535 | if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key) | |
2536 | + sizeof(struct btrfs_chunk)) { | |
2537 | btrfs_err(fs_info, "system chunk array too small %u < %zu", | |
2538 | btrfs_super_sys_array_size(sb), | |
2539 | sizeof(struct btrfs_disk_key) | |
2540 | + sizeof(struct btrfs_chunk)); | |
2541 | ret = -EINVAL; | |
2542 | } | |
2543 | ||
2544 | /* | |
2545 | * The generation is a global counter, we'll trust it more than the others | |
2546 | * but it's still possible that it's the one that's wrong. | |
2547 | */ | |
2548 | if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb)) | |
2549 | btrfs_warn(fs_info, | |
2550 | "suspicious: generation < chunk_root_generation: %llu < %llu", | |
2551 | btrfs_super_generation(sb), | |
2552 | btrfs_super_chunk_root_generation(sb)); | |
2553 | if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb) | |
2554 | && btrfs_super_cache_generation(sb) != (u64)-1) | |
2555 | btrfs_warn(fs_info, | |
2556 | "suspicious: generation < cache_generation: %llu < %llu", | |
2557 | btrfs_super_generation(sb), | |
2558 | btrfs_super_cache_generation(sb)); | |
2559 | ||
2560 | return ret; | |
2561 | } | |
2562 | ||
069ec957 QW |
2563 | /* |
2564 | * Validation of super block at mount time. | |
2565 | * Some checks already done early at mount time, like csum type and incompat | |
2566 | * flags will be skipped. | |
2567 | */ | |
2568 | static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info) | |
2569 | { | |
2570 | return validate_super(fs_info, fs_info->super_copy, 0); | |
2571 | } | |
2572 | ||
75cb857d QW |
2573 | /* |
2574 | * Validation of super block at write time. | |
2575 | * Some checks like bytenr check will be skipped as their values will be | |
2576 | * overwritten soon. | |
2577 | * Extra checks like csum type and incompat flags will be done here. | |
2578 | */ | |
2579 | static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info, | |
2580 | struct btrfs_super_block *sb) | |
2581 | { | |
2582 | int ret; | |
2583 | ||
2584 | ret = validate_super(fs_info, sb, -1); | |
2585 | if (ret < 0) | |
2586 | goto out; | |
e7e16f48 | 2587 | if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) { |
75cb857d QW |
2588 | ret = -EUCLEAN; |
2589 | btrfs_err(fs_info, "invalid csum type, has %u want %u", | |
2590 | btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32); | |
2591 | goto out; | |
2592 | } | |
2593 | if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) { | |
2594 | ret = -EUCLEAN; | |
2595 | btrfs_err(fs_info, | |
2596 | "invalid incompat flags, has 0x%llx valid mask 0x%llx", | |
2597 | btrfs_super_incompat_flags(sb), | |
2598 | (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP); | |
2599 | goto out; | |
2600 | } | |
2601 | out: | |
2602 | if (ret < 0) | |
2603 | btrfs_err(fs_info, | |
2604 | "super block corruption detected before writing it to disk"); | |
2605 | return ret; | |
2606 | } | |
2607 | ||
6ef108dd | 2608 | static int __cold init_tree_roots(struct btrfs_fs_info *fs_info) |
b8522a1e | 2609 | { |
6ef108dd | 2610 | int backup_index = find_newest_super_backup(fs_info); |
b8522a1e NB |
2611 | struct btrfs_super_block *sb = fs_info->super_copy; |
2612 | struct btrfs_root *tree_root = fs_info->tree_root; | |
2613 | bool handle_error = false; | |
2614 | int ret = 0; | |
2615 | int i; | |
2616 | ||
2617 | for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { | |
2618 | u64 generation; | |
2619 | int level; | |
2620 | ||
2621 | if (handle_error) { | |
2622 | if (!IS_ERR(tree_root->node)) | |
2623 | free_extent_buffer(tree_root->node); | |
2624 | tree_root->node = NULL; | |
2625 | ||
2626 | if (!btrfs_test_opt(fs_info, USEBACKUPROOT)) | |
2627 | break; | |
2628 | ||
2629 | free_root_pointers(fs_info, 0); | |
2630 | ||
2631 | /* | |
2632 | * Don't use the log in recovery mode, it won't be | |
2633 | * valid | |
2634 | */ | |
2635 | btrfs_set_super_log_root(sb, 0); | |
2636 | ||
2637 | /* We can't trust the free space cache either */ | |
2638 | btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); | |
2639 | ||
2640 | ret = read_backup_root(fs_info, i); | |
6ef108dd | 2641 | backup_index = ret; |
b8522a1e NB |
2642 | if (ret < 0) |
2643 | return ret; | |
2644 | } | |
2645 | generation = btrfs_super_generation(sb); | |
2646 | level = btrfs_super_root_level(sb); | |
2647 | tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb), | |
2648 | generation, level, NULL); | |
217f5004 | 2649 | if (IS_ERR(tree_root->node)) { |
b8522a1e | 2650 | handle_error = true; |
217f5004 NB |
2651 | ret = PTR_ERR(tree_root->node); |
2652 | tree_root->node = NULL; | |
2653 | btrfs_warn(fs_info, "couldn't read tree root"); | |
2654 | continue; | |
b8522a1e | 2655 | |
217f5004 NB |
2656 | } else if (!extent_buffer_uptodate(tree_root->node)) { |
2657 | handle_error = true; | |
2658 | ret = -EIO; | |
2659 | btrfs_warn(fs_info, "error while reading tree root"); | |
b8522a1e NB |
2660 | continue; |
2661 | } | |
2662 | ||
2663 | btrfs_set_root_node(&tree_root->root_item, tree_root->node); | |
2664 | tree_root->commit_root = btrfs_root_node(tree_root); | |
2665 | btrfs_set_root_refs(&tree_root->root_item, 1); | |
2666 | ||
336a0d8d NB |
2667 | /* |
2668 | * No need to hold btrfs_root::objectid_mutex since the fs | |
2669 | * hasn't been fully initialised and we are the only user | |
2670 | */ | |
b8522a1e NB |
2671 | ret = btrfs_find_highest_objectid(tree_root, |
2672 | &tree_root->highest_objectid); | |
2673 | if (ret < 0) { | |
b8522a1e NB |
2674 | handle_error = true; |
2675 | continue; | |
2676 | } | |
2677 | ||
2678 | ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID); | |
b8522a1e NB |
2679 | |
2680 | ret = btrfs_read_roots(fs_info); | |
2681 | if (ret < 0) { | |
2682 | handle_error = true; | |
2683 | continue; | |
2684 | } | |
2685 | ||
2686 | /* All successful */ | |
2687 | fs_info->generation = generation; | |
2688 | fs_info->last_trans_committed = generation; | |
6ef108dd NB |
2689 | |
2690 | /* Always begin writing backup roots after the one being used */ | |
2691 | if (backup_index < 0) { | |
2692 | fs_info->backup_root_index = 0; | |
2693 | } else { | |
2694 | fs_info->backup_root_index = backup_index + 1; | |
2695 | fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS; | |
2696 | } | |
b8522a1e NB |
2697 | break; |
2698 | } | |
2699 | ||
2700 | return ret; | |
2701 | } | |
2702 | ||
8260edba | 2703 | void btrfs_init_fs_info(struct btrfs_fs_info *fs_info) |
2e635a27 | 2704 | { |
76dda93c | 2705 | INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); |
f28491e0 | 2706 | INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); |
8fd17795 | 2707 | INIT_LIST_HEAD(&fs_info->trans_list); |
facda1e7 | 2708 | INIT_LIST_HEAD(&fs_info->dead_roots); |
24bbcf04 | 2709 | INIT_LIST_HEAD(&fs_info->delayed_iputs); |
eb73c1b7 | 2710 | INIT_LIST_HEAD(&fs_info->delalloc_roots); |
11833d66 | 2711 | INIT_LIST_HEAD(&fs_info->caching_block_groups); |
eb73c1b7 | 2712 | spin_lock_init(&fs_info->delalloc_root_lock); |
a4abeea4 | 2713 | spin_lock_init(&fs_info->trans_lock); |
76dda93c | 2714 | spin_lock_init(&fs_info->fs_roots_radix_lock); |
24bbcf04 | 2715 | spin_lock_init(&fs_info->delayed_iput_lock); |
4cb5300b | 2716 | spin_lock_init(&fs_info->defrag_inodes_lock); |
ceda0864 | 2717 | spin_lock_init(&fs_info->super_lock); |
f28491e0 | 2718 | spin_lock_init(&fs_info->buffer_lock); |
47ab2a6c | 2719 | spin_lock_init(&fs_info->unused_bgs_lock); |
f29021b2 | 2720 | rwlock_init(&fs_info->tree_mod_log_lock); |
d7c15171 | 2721 | mutex_init(&fs_info->unused_bg_unpin_mutex); |
67c5e7d4 | 2722 | mutex_init(&fs_info->delete_unused_bgs_mutex); |
7585717f | 2723 | mutex_init(&fs_info->reloc_mutex); |
573bfb72 | 2724 | mutex_init(&fs_info->delalloc_root_mutex); |
de98ced9 | 2725 | seqlock_init(&fs_info->profiles_lock); |
19c00ddc | 2726 | |
0b86a832 | 2727 | INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); |
6324fbf3 | 2728 | INIT_LIST_HEAD(&fs_info->space_info); |
f29021b2 | 2729 | INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); |
47ab2a6c | 2730 | INIT_LIST_HEAD(&fs_info->unused_bgs); |
bd647ce3 JB |
2731 | #ifdef CONFIG_BTRFS_DEBUG |
2732 | INIT_LIST_HEAD(&fs_info->allocated_roots); | |
3fd63727 JB |
2733 | INIT_LIST_HEAD(&fs_info->allocated_ebs); |
2734 | spin_lock_init(&fs_info->eb_leak_lock); | |
bd647ce3 | 2735 | #endif |
c8bf1b67 | 2736 | extent_map_tree_init(&fs_info->mapping_tree); |
66d8f3dd MX |
2737 | btrfs_init_block_rsv(&fs_info->global_block_rsv, |
2738 | BTRFS_BLOCK_RSV_GLOBAL); | |
66d8f3dd MX |
2739 | btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); |
2740 | btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); | |
2741 | btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); | |
2742 | btrfs_init_block_rsv(&fs_info->delayed_block_rsv, | |
2743 | BTRFS_BLOCK_RSV_DELOPS); | |
ba2c4d4e JB |
2744 | btrfs_init_block_rsv(&fs_info->delayed_refs_rsv, |
2745 | BTRFS_BLOCK_RSV_DELREFS); | |
2746 | ||
771ed689 | 2747 | atomic_set(&fs_info->async_delalloc_pages, 0); |
4cb5300b | 2748 | atomic_set(&fs_info->defrag_running, 0); |
2fefd558 | 2749 | atomic_set(&fs_info->reada_works_cnt, 0); |
034f784d | 2750 | atomic_set(&fs_info->nr_delayed_iputs, 0); |
fc36ed7e | 2751 | atomic64_set(&fs_info->tree_mod_seq, 0); |
95ac567a | 2752 | fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE; |
9ed74f2d | 2753 | fs_info->metadata_ratio = 0; |
4cb5300b | 2754 | fs_info->defrag_inodes = RB_ROOT; |
a5ed45f8 | 2755 | atomic64_set(&fs_info->free_chunk_space, 0); |
f29021b2 | 2756 | fs_info->tree_mod_log = RB_ROOT; |
8b87dc17 | 2757 | fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; |
f8c269d7 | 2758 | fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */ |
90519d66 | 2759 | /* readahead state */ |
d0164adc | 2760 | INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
90519d66 | 2761 | spin_lock_init(&fs_info->reada_lock); |
fd708b81 | 2762 | btrfs_init_ref_verify(fs_info); |
c8b97818 | 2763 | |
b34b086c CM |
2764 | fs_info->thread_pool_size = min_t(unsigned long, |
2765 | num_online_cpus() + 2, 8); | |
0afbaf8c | 2766 | |
199c2a9c MX |
2767 | INIT_LIST_HEAD(&fs_info->ordered_roots); |
2768 | spin_lock_init(&fs_info->ordered_root_lock); | |
69fe2d75 | 2769 | |
638aa7ed | 2770 | btrfs_init_scrub(fs_info); |
21adbd5c SB |
2771 | #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY |
2772 | fs_info->check_integrity_print_mask = 0; | |
2773 | #endif | |
779a65a4 | 2774 | btrfs_init_balance(fs_info); |
57056740 | 2775 | btrfs_init_async_reclaim_work(fs_info); |
a2de733c | 2776 | |
0f9dd46c | 2777 | spin_lock_init(&fs_info->block_group_cache_lock); |
6bef4d31 | 2778 | fs_info->block_group_cache_tree = RB_ROOT; |
a1897fdd | 2779 | fs_info->first_logical_byte = (u64)-1; |
0f9dd46c | 2780 | |
fe119a6e NB |
2781 | extent_io_tree_init(fs_info, &fs_info->excluded_extents, |
2782 | IO_TREE_FS_EXCLUDED_EXTENTS, NULL); | |
afcdd129 | 2783 | set_bit(BTRFS_FS_BARRIER, &fs_info->flags); |
39279cc3 | 2784 | |
5a3f23d5 | 2785 | mutex_init(&fs_info->ordered_operations_mutex); |
e02119d5 | 2786 | mutex_init(&fs_info->tree_log_mutex); |
925baedd | 2787 | mutex_init(&fs_info->chunk_mutex); |
a74a4b97 CM |
2788 | mutex_init(&fs_info->transaction_kthread_mutex); |
2789 | mutex_init(&fs_info->cleaner_mutex); | |
1bbc621e | 2790 | mutex_init(&fs_info->ro_block_group_mutex); |
9e351cc8 | 2791 | init_rwsem(&fs_info->commit_root_sem); |
c71bf099 | 2792 | init_rwsem(&fs_info->cleanup_work_sem); |
76dda93c | 2793 | init_rwsem(&fs_info->subvol_sem); |
803b2f54 | 2794 | sema_init(&fs_info->uuid_tree_rescan_sem, 1); |
fa9c0d79 | 2795 | |
ad618368 | 2796 | btrfs_init_dev_replace_locks(fs_info); |
f9e92e40 | 2797 | btrfs_init_qgroup(fs_info); |
b0643e59 | 2798 | btrfs_discard_init(fs_info); |
416ac51d | 2799 | |
fa9c0d79 CM |
2800 | btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); |
2801 | btrfs_init_free_cluster(&fs_info->data_alloc_cluster); | |
2802 | ||
e6dcd2dc | 2803 | init_waitqueue_head(&fs_info->transaction_throttle); |
f9295749 | 2804 | init_waitqueue_head(&fs_info->transaction_wait); |
bb9c12c9 | 2805 | init_waitqueue_head(&fs_info->transaction_blocked_wait); |
4854ddd0 | 2806 | init_waitqueue_head(&fs_info->async_submit_wait); |
034f784d | 2807 | init_waitqueue_head(&fs_info->delayed_iputs_wait); |
3768f368 | 2808 | |
da17066c JM |
2809 | /* Usable values until the real ones are cached from the superblock */ |
2810 | fs_info->nodesize = 4096; | |
2811 | fs_info->sectorsize = 4096; | |
ab108d99 | 2812 | fs_info->sectorsize_bits = ilog2(4096); |
da17066c JM |
2813 | fs_info->stripesize = 4096; |
2814 | ||
eede2bf3 OS |
2815 | spin_lock_init(&fs_info->swapfile_pins_lock); |
2816 | fs_info->swapfile_pins = RB_ROOT; | |
2817 | ||
9e967495 | 2818 | fs_info->send_in_progress = 0; |
8260edba JB |
2819 | } |
2820 | ||
2821 | static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb) | |
2822 | { | |
2823 | int ret; | |
2824 | ||
2825 | fs_info->sb = sb; | |
2826 | sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE; | |
2827 | sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE); | |
9e967495 | 2828 | |
ae18c37a JB |
2829 | ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL); |
2830 | if (ret) | |
c75e8394 | 2831 | return ret; |
ae18c37a JB |
2832 | |
2833 | ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL); | |
2834 | if (ret) | |
c75e8394 | 2835 | return ret; |
ae18c37a JB |
2836 | |
2837 | fs_info->dirty_metadata_batch = PAGE_SIZE * | |
2838 | (1 + ilog2(nr_cpu_ids)); | |
2839 | ||
2840 | ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL); | |
2841 | if (ret) | |
c75e8394 | 2842 | return ret; |
ae18c37a JB |
2843 | |
2844 | ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0, | |
2845 | GFP_KERNEL); | |
2846 | if (ret) | |
c75e8394 | 2847 | return ret; |
ae18c37a JB |
2848 | |
2849 | fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), | |
2850 | GFP_KERNEL); | |
c75e8394 JB |
2851 | if (!fs_info->delayed_root) |
2852 | return -ENOMEM; | |
ae18c37a JB |
2853 | btrfs_init_delayed_root(fs_info->delayed_root); |
2854 | ||
c75e8394 | 2855 | return btrfs_alloc_stripe_hash_table(fs_info); |
ae18c37a JB |
2856 | } |
2857 | ||
97f4dd09 NB |
2858 | static int btrfs_uuid_rescan_kthread(void *data) |
2859 | { | |
2860 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
2861 | int ret; | |
2862 | ||
2863 | /* | |
2864 | * 1st step is to iterate through the existing UUID tree and | |
2865 | * to delete all entries that contain outdated data. | |
2866 | * 2nd step is to add all missing entries to the UUID tree. | |
2867 | */ | |
2868 | ret = btrfs_uuid_tree_iterate(fs_info); | |
2869 | if (ret < 0) { | |
c94bec2c JB |
2870 | if (ret != -EINTR) |
2871 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", | |
2872 | ret); | |
97f4dd09 NB |
2873 | up(&fs_info->uuid_tree_rescan_sem); |
2874 | return ret; | |
2875 | } | |
2876 | return btrfs_uuid_scan_kthread(data); | |
2877 | } | |
2878 | ||
2879 | static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) | |
2880 | { | |
2881 | struct task_struct *task; | |
2882 | ||
2883 | down(&fs_info->uuid_tree_rescan_sem); | |
2884 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
2885 | if (IS_ERR(task)) { | |
2886 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
2887 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); | |
2888 | up(&fs_info->uuid_tree_rescan_sem); | |
2889 | return PTR_ERR(task); | |
2890 | } | |
2891 | ||
2892 | return 0; | |
2893 | } | |
2894 | ||
ae18c37a JB |
2895 | int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices, |
2896 | char *options) | |
2897 | { | |
2898 | u32 sectorsize; | |
2899 | u32 nodesize; | |
2900 | u32 stripesize; | |
2901 | u64 generation; | |
2902 | u64 features; | |
2903 | u16 csum_type; | |
ae18c37a JB |
2904 | struct btrfs_super_block *disk_super; |
2905 | struct btrfs_fs_info *fs_info = btrfs_sb(sb); | |
2906 | struct btrfs_root *tree_root; | |
2907 | struct btrfs_root *chunk_root; | |
2908 | int ret; | |
2909 | int err = -EINVAL; | |
2910 | int clear_free_space_tree = 0; | |
2911 | int level; | |
2912 | ||
8260edba | 2913 | ret = init_mount_fs_info(fs_info, sb); |
53b381b3 | 2914 | if (ret) { |
83c8266a | 2915 | err = ret; |
ae18c37a | 2916 | goto fail; |
53b381b3 DW |
2917 | } |
2918 | ||
ae18c37a JB |
2919 | /* These need to be init'ed before we start creating inodes and such. */ |
2920 | tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, | |
2921 | GFP_KERNEL); | |
2922 | fs_info->tree_root = tree_root; | |
2923 | chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID, | |
2924 | GFP_KERNEL); | |
2925 | fs_info->chunk_root = chunk_root; | |
2926 | if (!tree_root || !chunk_root) { | |
2927 | err = -ENOMEM; | |
c75e8394 | 2928 | goto fail; |
ae18c37a JB |
2929 | } |
2930 | ||
2931 | fs_info->btree_inode = new_inode(sb); | |
2932 | if (!fs_info->btree_inode) { | |
2933 | err = -ENOMEM; | |
c75e8394 | 2934 | goto fail; |
ae18c37a JB |
2935 | } |
2936 | mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); | |
2937 | btrfs_init_btree_inode(fs_info); | |
2938 | ||
3c4bb26b | 2939 | invalidate_bdev(fs_devices->latest_bdev); |
1104a885 DS |
2940 | |
2941 | /* | |
2942 | * Read super block and check the signature bytes only | |
2943 | */ | |
8f32380d JT |
2944 | disk_super = btrfs_read_dev_super(fs_devices->latest_bdev); |
2945 | if (IS_ERR(disk_super)) { | |
2946 | err = PTR_ERR(disk_super); | |
16cdcec7 | 2947 | goto fail_alloc; |
20b45077 | 2948 | } |
39279cc3 | 2949 | |
8dc3f22c | 2950 | /* |
260db43c | 2951 | * Verify the type first, if that or the checksum value are |
8dc3f22c JT |
2952 | * corrupted, we'll find out |
2953 | */ | |
8f32380d | 2954 | csum_type = btrfs_super_csum_type(disk_super); |
51bce6c9 | 2955 | if (!btrfs_supported_super_csum(csum_type)) { |
8dc3f22c | 2956 | btrfs_err(fs_info, "unsupported checksum algorithm: %u", |
51bce6c9 | 2957 | csum_type); |
8dc3f22c | 2958 | err = -EINVAL; |
8f32380d | 2959 | btrfs_release_disk_super(disk_super); |
8dc3f22c JT |
2960 | goto fail_alloc; |
2961 | } | |
2962 | ||
6d97c6e3 JT |
2963 | ret = btrfs_init_csum_hash(fs_info, csum_type); |
2964 | if (ret) { | |
2965 | err = ret; | |
8f32380d | 2966 | btrfs_release_disk_super(disk_super); |
6d97c6e3 JT |
2967 | goto fail_alloc; |
2968 | } | |
2969 | ||
1104a885 DS |
2970 | /* |
2971 | * We want to check superblock checksum, the type is stored inside. | |
2972 | * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). | |
2973 | */ | |
8f32380d | 2974 | if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) { |
05135f59 | 2975 | btrfs_err(fs_info, "superblock checksum mismatch"); |
1104a885 | 2976 | err = -EINVAL; |
8f32380d | 2977 | btrfs_release_disk_super(disk_super); |
141386e1 | 2978 | goto fail_alloc; |
1104a885 DS |
2979 | } |
2980 | ||
2981 | /* | |
2982 | * super_copy is zeroed at allocation time and we never touch the | |
2983 | * following bytes up to INFO_SIZE, the checksum is calculated from | |
2984 | * the whole block of INFO_SIZE | |
2985 | */ | |
8f32380d JT |
2986 | memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy)); |
2987 | btrfs_release_disk_super(disk_super); | |
5f39d397 | 2988 | |
fbc6feae NB |
2989 | disk_super = fs_info->super_copy; |
2990 | ||
de37aa51 NB |
2991 | ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid, |
2992 | BTRFS_FSID_SIZE)); | |
2993 | ||
7239ff4b | 2994 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) { |
de37aa51 NB |
2995 | ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid, |
2996 | fs_info->super_copy->metadata_uuid, | |
2997 | BTRFS_FSID_SIZE)); | |
7239ff4b | 2998 | } |
0b86a832 | 2999 | |
fbc6feae NB |
3000 | features = btrfs_super_flags(disk_super); |
3001 | if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { | |
3002 | features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2; | |
3003 | btrfs_set_super_flags(disk_super, features); | |
3004 | btrfs_info(fs_info, | |
3005 | "found metadata UUID change in progress flag, clearing"); | |
3006 | } | |
3007 | ||
3008 | memcpy(fs_info->super_for_commit, fs_info->super_copy, | |
3009 | sizeof(*fs_info->super_for_commit)); | |
de37aa51 | 3010 | |
069ec957 | 3011 | ret = btrfs_validate_mount_super(fs_info); |
1104a885 | 3012 | if (ret) { |
05135f59 | 3013 | btrfs_err(fs_info, "superblock contains fatal errors"); |
1104a885 | 3014 | err = -EINVAL; |
141386e1 | 3015 | goto fail_alloc; |
1104a885 DS |
3016 | } |
3017 | ||
0f7d52f4 | 3018 | if (!btrfs_super_root(disk_super)) |
141386e1 | 3019 | goto fail_alloc; |
0f7d52f4 | 3020 | |
acce952b | 3021 | /* check FS state, whether FS is broken. */ |
87533c47 MX |
3022 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) |
3023 | set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); | |
acce952b | 3024 | |
75e7cb7f LB |
3025 | /* |
3026 | * In the long term, we'll store the compression type in the super | |
3027 | * block, and it'll be used for per file compression control. | |
3028 | */ | |
3029 | fs_info->compress_type = BTRFS_COMPRESS_ZLIB; | |
3030 | ||
2ff7e61e | 3031 | ret = btrfs_parse_options(fs_info, options, sb->s_flags); |
2b82032c YZ |
3032 | if (ret) { |
3033 | err = ret; | |
141386e1 | 3034 | goto fail_alloc; |
2b82032c | 3035 | } |
dfe25020 | 3036 | |
f2b636e8 JB |
3037 | features = btrfs_super_incompat_flags(disk_super) & |
3038 | ~BTRFS_FEATURE_INCOMPAT_SUPP; | |
3039 | if (features) { | |
05135f59 DS |
3040 | btrfs_err(fs_info, |
3041 | "cannot mount because of unsupported optional features (%llx)", | |
3042 | features); | |
f2b636e8 | 3043 | err = -EINVAL; |
141386e1 | 3044 | goto fail_alloc; |
f2b636e8 JB |
3045 | } |
3046 | ||
5d4f98a2 | 3047 | features = btrfs_super_incompat_flags(disk_super); |
a6fa6fae | 3048 | features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; |
0b246afa | 3049 | if (fs_info->compress_type == BTRFS_COMPRESS_LZO) |
a6fa6fae | 3050 | features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; |
5c1aab1d NT |
3051 | else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD) |
3052 | features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD; | |
727011e0 | 3053 | |
3173a18f | 3054 | if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) |
05135f59 | 3055 | btrfs_info(fs_info, "has skinny extents"); |
3173a18f | 3056 | |
727011e0 CM |
3057 | /* |
3058 | * flag our filesystem as having big metadata blocks if | |
3059 | * they are bigger than the page size | |
3060 | */ | |
09cbfeaf | 3061 | if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) { |
727011e0 | 3062 | if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) |
05135f59 DS |
3063 | btrfs_info(fs_info, |
3064 | "flagging fs with big metadata feature"); | |
727011e0 CM |
3065 | features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; |
3066 | } | |
3067 | ||
bc3f116f | 3068 | nodesize = btrfs_super_nodesize(disk_super); |
bc3f116f | 3069 | sectorsize = btrfs_super_sectorsize(disk_super); |
b7f67055 | 3070 | stripesize = sectorsize; |
707e8a07 | 3071 | fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids)); |
963d678b | 3072 | fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); |
bc3f116f | 3073 | |
da17066c JM |
3074 | /* Cache block sizes */ |
3075 | fs_info->nodesize = nodesize; | |
3076 | fs_info->sectorsize = sectorsize; | |
ab108d99 | 3077 | fs_info->sectorsize_bits = ilog2(sectorsize); |
22b6331d | 3078 | fs_info->csum_size = btrfs_super_csum_size(disk_super); |
fe5ecbe8 | 3079 | fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size; |
da17066c JM |
3080 | fs_info->stripesize = stripesize; |
3081 | ||
bc3f116f CM |
3082 | /* |
3083 | * mixed block groups end up with duplicate but slightly offset | |
3084 | * extent buffers for the same range. It leads to corruptions | |
3085 | */ | |
3086 | if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && | |
707e8a07 | 3087 | (sectorsize != nodesize)) { |
05135f59 DS |
3088 | btrfs_err(fs_info, |
3089 | "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups", | |
3090 | nodesize, sectorsize); | |
141386e1 | 3091 | goto fail_alloc; |
bc3f116f CM |
3092 | } |
3093 | ||
ceda0864 MX |
3094 | /* |
3095 | * Needn't use the lock because there is no other task which will | |
3096 | * update the flag. | |
3097 | */ | |
a6fa6fae | 3098 | btrfs_set_super_incompat_flags(disk_super, features); |
5d4f98a2 | 3099 | |
f2b636e8 JB |
3100 | features = btrfs_super_compat_ro_flags(disk_super) & |
3101 | ~BTRFS_FEATURE_COMPAT_RO_SUPP; | |
bc98a42c | 3102 | if (!sb_rdonly(sb) && features) { |
05135f59 DS |
3103 | btrfs_err(fs_info, |
3104 | "cannot mount read-write because of unsupported optional features (%llx)", | |
c1c9ff7c | 3105 | features); |
f2b636e8 | 3106 | err = -EINVAL; |
141386e1 | 3107 | goto fail_alloc; |
f2b636e8 | 3108 | } |
61d92c32 | 3109 | |
2a458198 ES |
3110 | ret = btrfs_init_workqueues(fs_info, fs_devices); |
3111 | if (ret) { | |
3112 | err = ret; | |
0dc3b84a JB |
3113 | goto fail_sb_buffer; |
3114 | } | |
4543df7e | 3115 | |
9e11ceee JK |
3116 | sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super); |
3117 | sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE); | |
4575c9cc | 3118 | |
a061fc8d CM |
3119 | sb->s_blocksize = sectorsize; |
3120 | sb->s_blocksize_bits = blksize_bits(sectorsize); | |
de37aa51 | 3121 | memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE); |
db94535d | 3122 | |
925baedd | 3123 | mutex_lock(&fs_info->chunk_mutex); |
6bccf3ab | 3124 | ret = btrfs_read_sys_array(fs_info); |
925baedd | 3125 | mutex_unlock(&fs_info->chunk_mutex); |
84eed90f | 3126 | if (ret) { |
05135f59 | 3127 | btrfs_err(fs_info, "failed to read the system array: %d", ret); |
5d4f98a2 | 3128 | goto fail_sb_buffer; |
84eed90f | 3129 | } |
0b86a832 | 3130 | |
84234f3a | 3131 | generation = btrfs_super_chunk_root_generation(disk_super); |
581c1760 | 3132 | level = btrfs_super_chunk_root_level(disk_super); |
0b86a832 | 3133 | |
2ff7e61e | 3134 | chunk_root->node = read_tree_block(fs_info, |
0b86a832 | 3135 | btrfs_super_chunk_root(disk_super), |
581c1760 | 3136 | generation, level, NULL); |
64c043de LB |
3137 | if (IS_ERR(chunk_root->node) || |
3138 | !extent_buffer_uptodate(chunk_root->node)) { | |
05135f59 | 3139 | btrfs_err(fs_info, "failed to read chunk root"); |
e5fffbac | 3140 | if (!IS_ERR(chunk_root->node)) |
3141 | free_extent_buffer(chunk_root->node); | |
95ab1f64 | 3142 | chunk_root->node = NULL; |
af31f5e5 | 3143 | goto fail_tree_roots; |
83121942 | 3144 | } |
5d4f98a2 YZ |
3145 | btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); |
3146 | chunk_root->commit_root = btrfs_root_node(chunk_root); | |
0b86a832 | 3147 | |
e17cade2 | 3148 | read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, |
c4ac7541 DS |
3149 | offsetof(struct btrfs_header, chunk_tree_uuid), |
3150 | BTRFS_UUID_SIZE); | |
e17cade2 | 3151 | |
5b4aacef | 3152 | ret = btrfs_read_chunk_tree(fs_info); |
2b82032c | 3153 | if (ret) { |
05135f59 | 3154 | btrfs_err(fs_info, "failed to read chunk tree: %d", ret); |
af31f5e5 | 3155 | goto fail_tree_roots; |
2b82032c | 3156 | } |
0b86a832 | 3157 | |
8dabb742 | 3158 | /* |
9b99b115 AJ |
3159 | * Keep the devid that is marked to be the target device for the |
3160 | * device replace procedure | |
8dabb742 | 3161 | */ |
9b99b115 | 3162 | btrfs_free_extra_devids(fs_devices, 0); |
dfe25020 | 3163 | |
a6b0d5c8 | 3164 | if (!fs_devices->latest_bdev) { |
05135f59 | 3165 | btrfs_err(fs_info, "failed to read devices"); |
a6b0d5c8 CM |
3166 | goto fail_tree_roots; |
3167 | } | |
3168 | ||
b8522a1e | 3169 | ret = init_tree_roots(fs_info); |
4bbcaa64 | 3170 | if (ret) |
b8522a1e | 3171 | goto fail_tree_roots; |
8929ecfa | 3172 | |
75ec1db8 JB |
3173 | /* |
3174 | * If we have a uuid root and we're not being told to rescan we need to | |
3175 | * check the generation here so we can set the | |
3176 | * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the | |
3177 | * transaction during a balance or the log replay without updating the | |
3178 | * uuid generation, and then if we crash we would rescan the uuid tree, | |
3179 | * even though it was perfectly fine. | |
3180 | */ | |
3181 | if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) && | |
3182 | fs_info->generation == btrfs_super_uuid_tree_generation(disk_super)) | |
3183 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); | |
3184 | ||
cf90d884 QW |
3185 | ret = btrfs_verify_dev_extents(fs_info); |
3186 | if (ret) { | |
3187 | btrfs_err(fs_info, | |
3188 | "failed to verify dev extents against chunks: %d", | |
3189 | ret); | |
3190 | goto fail_block_groups; | |
3191 | } | |
68310a5e ID |
3192 | ret = btrfs_recover_balance(fs_info); |
3193 | if (ret) { | |
05135f59 | 3194 | btrfs_err(fs_info, "failed to recover balance: %d", ret); |
68310a5e ID |
3195 | goto fail_block_groups; |
3196 | } | |
3197 | ||
733f4fbb SB |
3198 | ret = btrfs_init_dev_stats(fs_info); |
3199 | if (ret) { | |
05135f59 | 3200 | btrfs_err(fs_info, "failed to init dev_stats: %d", ret); |
733f4fbb SB |
3201 | goto fail_block_groups; |
3202 | } | |
3203 | ||
8dabb742 SB |
3204 | ret = btrfs_init_dev_replace(fs_info); |
3205 | if (ret) { | |
05135f59 | 3206 | btrfs_err(fs_info, "failed to init dev_replace: %d", ret); |
8dabb742 SB |
3207 | goto fail_block_groups; |
3208 | } | |
3209 | ||
9b99b115 | 3210 | btrfs_free_extra_devids(fs_devices, 1); |
8dabb742 | 3211 | |
c6761a9e | 3212 | ret = btrfs_sysfs_add_fsid(fs_devices); |
b7c35e81 | 3213 | if (ret) { |
05135f59 DS |
3214 | btrfs_err(fs_info, "failed to init sysfs fsid interface: %d", |
3215 | ret); | |
b7c35e81 AJ |
3216 | goto fail_block_groups; |
3217 | } | |
3218 | ||
96f3136e | 3219 | ret = btrfs_sysfs_add_mounted(fs_info); |
c59021f8 | 3220 | if (ret) { |
05135f59 | 3221 | btrfs_err(fs_info, "failed to init sysfs interface: %d", ret); |
b7c35e81 | 3222 | goto fail_fsdev_sysfs; |
c59021f8 | 3223 | } |
3224 | ||
c59021f8 | 3225 | ret = btrfs_init_space_info(fs_info); |
3226 | if (ret) { | |
05135f59 | 3227 | btrfs_err(fs_info, "failed to initialize space info: %d", ret); |
2365dd3c | 3228 | goto fail_sysfs; |
c59021f8 | 3229 | } |
3230 | ||
5b4aacef | 3231 | ret = btrfs_read_block_groups(fs_info); |
1b1d1f66 | 3232 | if (ret) { |
05135f59 | 3233 | btrfs_err(fs_info, "failed to read block groups: %d", ret); |
2365dd3c | 3234 | goto fail_sysfs; |
1b1d1f66 | 3235 | } |
4330e183 | 3236 | |
6528b99d | 3237 | if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) { |
05135f59 | 3238 | btrfs_warn(fs_info, |
52042d8e | 3239 | "writable mount is not allowed due to too many missing devices"); |
2365dd3c | 3240 | goto fail_sysfs; |
292fd7fc | 3241 | } |
9078a3e1 | 3242 | |
a74a4b97 CM |
3243 | fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, |
3244 | "btrfs-cleaner"); | |
57506d50 | 3245 | if (IS_ERR(fs_info->cleaner_kthread)) |
2365dd3c | 3246 | goto fail_sysfs; |
a74a4b97 CM |
3247 | |
3248 | fs_info->transaction_kthread = kthread_run(transaction_kthread, | |
3249 | tree_root, | |
3250 | "btrfs-transaction"); | |
57506d50 | 3251 | if (IS_ERR(fs_info->transaction_kthread)) |
3f157a2f | 3252 | goto fail_cleaner; |
a74a4b97 | 3253 | |
583b7231 | 3254 | if (!btrfs_test_opt(fs_info, NOSSD) && |
c289811c | 3255 | !fs_info->fs_devices->rotating) { |
583b7231 | 3256 | btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations"); |
c289811c CM |
3257 | } |
3258 | ||
572d9ab7 | 3259 | /* |
01327610 | 3260 | * Mount does not set all options immediately, we can do it now and do |
572d9ab7 DS |
3261 | * not have to wait for transaction commit |
3262 | */ | |
3263 | btrfs_apply_pending_changes(fs_info); | |
3818aea2 | 3264 | |
21adbd5c | 3265 | #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY |
0b246afa | 3266 | if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) { |
2ff7e61e | 3267 | ret = btrfsic_mount(fs_info, fs_devices, |
0b246afa | 3268 | btrfs_test_opt(fs_info, |
21adbd5c SB |
3269 | CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? |
3270 | 1 : 0, | |
3271 | fs_info->check_integrity_print_mask); | |
3272 | if (ret) | |
05135f59 DS |
3273 | btrfs_warn(fs_info, |
3274 | "failed to initialize integrity check module: %d", | |
3275 | ret); | |
21adbd5c SB |
3276 | } |
3277 | #endif | |
bcef60f2 AJ |
3278 | ret = btrfs_read_qgroup_config(fs_info); |
3279 | if (ret) | |
3280 | goto fail_trans_kthread; | |
21adbd5c | 3281 | |
fd708b81 JB |
3282 | if (btrfs_build_ref_tree(fs_info)) |
3283 | btrfs_err(fs_info, "couldn't build ref tree"); | |
3284 | ||
96da0919 QW |
3285 | /* do not make disk changes in broken FS or nologreplay is given */ |
3286 | if (btrfs_super_log_root(disk_super) != 0 && | |
0b246afa | 3287 | !btrfs_test_opt(fs_info, NOLOGREPLAY)) { |
e8294f2f | 3288 | btrfs_info(fs_info, "start tree-log replay"); |
63443bf5 | 3289 | ret = btrfs_replay_log(fs_info, fs_devices); |
79787eaa | 3290 | if (ret) { |
63443bf5 | 3291 | err = ret; |
28c16cbb | 3292 | goto fail_qgroup; |
79787eaa | 3293 | } |
e02119d5 | 3294 | } |
1a40e23b | 3295 | |
6bccf3ab | 3296 | ret = btrfs_find_orphan_roots(fs_info); |
79787eaa | 3297 | if (ret) |
28c16cbb | 3298 | goto fail_qgroup; |
76dda93c | 3299 | |
bc98a42c | 3300 | if (!sb_rdonly(sb)) { |
d68fc57b | 3301 | ret = btrfs_cleanup_fs_roots(fs_info); |
44c44af2 | 3302 | if (ret) |
28c16cbb | 3303 | goto fail_qgroup; |
90c711ab ZB |
3304 | |
3305 | mutex_lock(&fs_info->cleaner_mutex); | |
5d4f98a2 | 3306 | ret = btrfs_recover_relocation(tree_root); |
90c711ab | 3307 | mutex_unlock(&fs_info->cleaner_mutex); |
d7ce5843 | 3308 | if (ret < 0) { |
05135f59 DS |
3309 | btrfs_warn(fs_info, "failed to recover relocation: %d", |
3310 | ret); | |
d7ce5843 | 3311 | err = -EINVAL; |
bcef60f2 | 3312 | goto fail_qgroup; |
d7ce5843 | 3313 | } |
7c2ca468 | 3314 | } |
1a40e23b | 3315 | |
56e9357a | 3316 | fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true); |
3140c9a3 DC |
3317 | if (IS_ERR(fs_info->fs_root)) { |
3318 | err = PTR_ERR(fs_info->fs_root); | |
f50f4353 | 3319 | btrfs_warn(fs_info, "failed to read fs tree: %d", err); |
315bf8ef | 3320 | fs_info->fs_root = NULL; |
bcef60f2 | 3321 | goto fail_qgroup; |
3140c9a3 | 3322 | } |
c289811c | 3323 | |
bc98a42c | 3324 | if (sb_rdonly(sb)) |
2b6ba629 | 3325 | return 0; |
59641015 | 3326 | |
f8d468a1 OS |
3327 | if (btrfs_test_opt(fs_info, CLEAR_CACHE) && |
3328 | btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { | |
6675df31 OS |
3329 | clear_free_space_tree = 1; |
3330 | } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) && | |
3331 | !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) { | |
3332 | btrfs_warn(fs_info, "free space tree is invalid"); | |
3333 | clear_free_space_tree = 1; | |
3334 | } | |
3335 | ||
3336 | if (clear_free_space_tree) { | |
f8d468a1 OS |
3337 | btrfs_info(fs_info, "clearing free space tree"); |
3338 | ret = btrfs_clear_free_space_tree(fs_info); | |
3339 | if (ret) { | |
3340 | btrfs_warn(fs_info, | |
3341 | "failed to clear free space tree: %d", ret); | |
6bccf3ab | 3342 | close_ctree(fs_info); |
f8d468a1 OS |
3343 | return ret; |
3344 | } | |
3345 | } | |
3346 | ||
0b246afa | 3347 | if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) && |
511711af | 3348 | !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) { |
05135f59 | 3349 | btrfs_info(fs_info, "creating free space tree"); |
511711af CM |
3350 | ret = btrfs_create_free_space_tree(fs_info); |
3351 | if (ret) { | |
05135f59 DS |
3352 | btrfs_warn(fs_info, |
3353 | "failed to create free space tree: %d", ret); | |
6bccf3ab | 3354 | close_ctree(fs_info); |
511711af CM |
3355 | return ret; |
3356 | } | |
3357 | } | |
3358 | ||
2b6ba629 ID |
3359 | down_read(&fs_info->cleanup_work_sem); |
3360 | if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || | |
3361 | (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { | |
e3acc2a6 | 3362 | up_read(&fs_info->cleanup_work_sem); |
6bccf3ab | 3363 | close_ctree(fs_info); |
2b6ba629 ID |
3364 | return ret; |
3365 | } | |
3366 | up_read(&fs_info->cleanup_work_sem); | |
59641015 | 3367 | |
2b6ba629 ID |
3368 | ret = btrfs_resume_balance_async(fs_info); |
3369 | if (ret) { | |
05135f59 | 3370 | btrfs_warn(fs_info, "failed to resume balance: %d", ret); |
6bccf3ab | 3371 | close_ctree(fs_info); |
2b6ba629 | 3372 | return ret; |
e3acc2a6 JB |
3373 | } |
3374 | ||
8dabb742 SB |
3375 | ret = btrfs_resume_dev_replace_async(fs_info); |
3376 | if (ret) { | |
05135f59 | 3377 | btrfs_warn(fs_info, "failed to resume device replace: %d", ret); |
6bccf3ab | 3378 | close_ctree(fs_info); |
8dabb742 SB |
3379 | return ret; |
3380 | } | |
3381 | ||
b382a324 | 3382 | btrfs_qgroup_rescan_resume(fs_info); |
b0643e59 | 3383 | btrfs_discard_resume(fs_info); |
b382a324 | 3384 | |
4bbcaa64 | 3385 | if (!fs_info->uuid_root) { |
05135f59 | 3386 | btrfs_info(fs_info, "creating UUID tree"); |
f7a81ea4 SB |
3387 | ret = btrfs_create_uuid_tree(fs_info); |
3388 | if (ret) { | |
05135f59 DS |
3389 | btrfs_warn(fs_info, |
3390 | "failed to create the UUID tree: %d", ret); | |
6bccf3ab | 3391 | close_ctree(fs_info); |
f7a81ea4 SB |
3392 | return ret; |
3393 | } | |
0b246afa | 3394 | } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) || |
4bbcaa64 ES |
3395 | fs_info->generation != |
3396 | btrfs_super_uuid_tree_generation(disk_super)) { | |
05135f59 | 3397 | btrfs_info(fs_info, "checking UUID tree"); |
70f80175 SB |
3398 | ret = btrfs_check_uuid_tree(fs_info); |
3399 | if (ret) { | |
05135f59 DS |
3400 | btrfs_warn(fs_info, |
3401 | "failed to check the UUID tree: %d", ret); | |
6bccf3ab | 3402 | close_ctree(fs_info); |
70f80175 SB |
3403 | return ret; |
3404 | } | |
f7a81ea4 | 3405 | } |
afcdd129 | 3406 | set_bit(BTRFS_FS_OPEN, &fs_info->flags); |
47ab2a6c | 3407 | |
ad2b2c80 | 3408 | return 0; |
39279cc3 | 3409 | |
bcef60f2 AJ |
3410 | fail_qgroup: |
3411 | btrfs_free_qgroup_config(fs_info); | |
7c2ca468 CM |
3412 | fail_trans_kthread: |
3413 | kthread_stop(fs_info->transaction_kthread); | |
2ff7e61e | 3414 | btrfs_cleanup_transaction(fs_info); |
faa2dbf0 | 3415 | btrfs_free_fs_roots(fs_info); |
3f157a2f | 3416 | fail_cleaner: |
a74a4b97 | 3417 | kthread_stop(fs_info->cleaner_kthread); |
7c2ca468 CM |
3418 | |
3419 | /* | |
3420 | * make sure we're done with the btree inode before we stop our | |
3421 | * kthreads | |
3422 | */ | |
3423 | filemap_write_and_wait(fs_info->btree_inode->i_mapping); | |
7c2ca468 | 3424 | |
2365dd3c | 3425 | fail_sysfs: |
6618a59b | 3426 | btrfs_sysfs_remove_mounted(fs_info); |
2365dd3c | 3427 | |
b7c35e81 AJ |
3428 | fail_fsdev_sysfs: |
3429 | btrfs_sysfs_remove_fsid(fs_info->fs_devices); | |
3430 | ||
1b1d1f66 | 3431 | fail_block_groups: |
54067ae9 | 3432 | btrfs_put_block_group_cache(fs_info); |
af31f5e5 CM |
3433 | |
3434 | fail_tree_roots: | |
9e3aa805 JB |
3435 | if (fs_info->data_reloc_root) |
3436 | btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root); | |
4273eaff | 3437 | free_root_pointers(fs_info, true); |
2b8195bb | 3438 | invalidate_inode_pages2(fs_info->btree_inode->i_mapping); |
af31f5e5 | 3439 | |
39279cc3 | 3440 | fail_sb_buffer: |
7abadb64 | 3441 | btrfs_stop_all_workers(fs_info); |
5cdd7db6 | 3442 | btrfs_free_block_groups(fs_info); |
16cdcec7 | 3443 | fail_alloc: |
586e46e2 ID |
3444 | btrfs_mapping_tree_free(&fs_info->mapping_tree); |
3445 | ||
4543df7e | 3446 | iput(fs_info->btree_inode); |
7e662854 | 3447 | fail: |
586e46e2 | 3448 | btrfs_close_devices(fs_info->fs_devices); |
ad2b2c80 | 3449 | return err; |
eb60ceac | 3450 | } |
663faf9f | 3451 | ALLOW_ERROR_INJECTION(open_ctree, ERRNO); |
eb60ceac | 3452 | |
314b6dd0 | 3453 | static void btrfs_end_super_write(struct bio *bio) |
f2984462 | 3454 | { |
314b6dd0 JT |
3455 | struct btrfs_device *device = bio->bi_private; |
3456 | struct bio_vec *bvec; | |
3457 | struct bvec_iter_all iter_all; | |
3458 | struct page *page; | |
3459 | ||
3460 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
3461 | page = bvec->bv_page; | |
3462 | ||
3463 | if (bio->bi_status) { | |
3464 | btrfs_warn_rl_in_rcu(device->fs_info, | |
3465 | "lost page write due to IO error on %s (%d)", | |
3466 | rcu_str_deref(device->name), | |
3467 | blk_status_to_errno(bio->bi_status)); | |
3468 | ClearPageUptodate(page); | |
3469 | SetPageError(page); | |
3470 | btrfs_dev_stat_inc_and_print(device, | |
3471 | BTRFS_DEV_STAT_WRITE_ERRS); | |
3472 | } else { | |
3473 | SetPageUptodate(page); | |
3474 | } | |
3475 | ||
3476 | put_page(page); | |
3477 | unlock_page(page); | |
f2984462 | 3478 | } |
314b6dd0 JT |
3479 | |
3480 | bio_put(bio); | |
f2984462 CM |
3481 | } |
3482 | ||
8f32380d JT |
3483 | struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev, |
3484 | int copy_num) | |
29c36d72 | 3485 | { |
29c36d72 | 3486 | struct btrfs_super_block *super; |
8f32380d | 3487 | struct page *page; |
29c36d72 | 3488 | u64 bytenr; |
8f32380d | 3489 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
29c36d72 AJ |
3490 | |
3491 | bytenr = btrfs_sb_offset(copy_num); | |
3492 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode)) | |
8f32380d | 3493 | return ERR_PTR(-EINVAL); |
29c36d72 | 3494 | |
8f32380d JT |
3495 | page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS); |
3496 | if (IS_ERR(page)) | |
3497 | return ERR_CAST(page); | |
29c36d72 | 3498 | |
8f32380d | 3499 | super = page_address(page); |
96c2e067 AJ |
3500 | if (btrfs_super_magic(super) != BTRFS_MAGIC) { |
3501 | btrfs_release_disk_super(super); | |
3502 | return ERR_PTR(-ENODATA); | |
3503 | } | |
3504 | ||
3505 | if (btrfs_super_bytenr(super) != bytenr) { | |
8f32380d JT |
3506 | btrfs_release_disk_super(super); |
3507 | return ERR_PTR(-EINVAL); | |
29c36d72 AJ |
3508 | } |
3509 | ||
8f32380d | 3510 | return super; |
29c36d72 AJ |
3511 | } |
3512 | ||
3513 | ||
8f32380d | 3514 | struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev) |
a512bbf8 | 3515 | { |
8f32380d | 3516 | struct btrfs_super_block *super, *latest = NULL; |
a512bbf8 YZ |
3517 | int i; |
3518 | u64 transid = 0; | |
a512bbf8 YZ |
3519 | |
3520 | /* we would like to check all the supers, but that would make | |
3521 | * a btrfs mount succeed after a mkfs from a different FS. | |
3522 | * So, we need to add a special mount option to scan for | |
3523 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
3524 | */ | |
3525 | for (i = 0; i < 1; i++) { | |
8f32380d JT |
3526 | super = btrfs_read_dev_one_super(bdev, i); |
3527 | if (IS_ERR(super)) | |
a512bbf8 YZ |
3528 | continue; |
3529 | ||
a512bbf8 | 3530 | if (!latest || btrfs_super_generation(super) > transid) { |
8f32380d JT |
3531 | if (latest) |
3532 | btrfs_release_disk_super(super); | |
3533 | ||
3534 | latest = super; | |
a512bbf8 | 3535 | transid = btrfs_super_generation(super); |
a512bbf8 YZ |
3536 | } |
3537 | } | |
92fc03fb | 3538 | |
8f32380d | 3539 | return super; |
a512bbf8 YZ |
3540 | } |
3541 | ||
4eedeb75 | 3542 | /* |
abbb3b8e | 3543 | * Write superblock @sb to the @device. Do not wait for completion, all the |
314b6dd0 | 3544 | * pages we use for writing are locked. |
4eedeb75 | 3545 | * |
abbb3b8e DS |
3546 | * Write @max_mirrors copies of the superblock, where 0 means default that fit |
3547 | * the expected device size at commit time. Note that max_mirrors must be | |
3548 | * same for write and wait phases. | |
4eedeb75 | 3549 | * |
314b6dd0 | 3550 | * Return number of errors when page is not found or submission fails. |
4eedeb75 | 3551 | */ |
a512bbf8 | 3552 | static int write_dev_supers(struct btrfs_device *device, |
abbb3b8e | 3553 | struct btrfs_super_block *sb, int max_mirrors) |
a512bbf8 | 3554 | { |
d5178578 | 3555 | struct btrfs_fs_info *fs_info = device->fs_info; |
314b6dd0 | 3556 | struct address_space *mapping = device->bdev->bd_inode->i_mapping; |
d5178578 | 3557 | SHASH_DESC_ON_STACK(shash, fs_info->csum_shash); |
a512bbf8 | 3558 | int i; |
a512bbf8 | 3559 | int errors = 0; |
a512bbf8 | 3560 | u64 bytenr; |
a512bbf8 YZ |
3561 | |
3562 | if (max_mirrors == 0) | |
3563 | max_mirrors = BTRFS_SUPER_MIRROR_MAX; | |
3564 | ||
d5178578 JT |
3565 | shash->tfm = fs_info->csum_shash; |
3566 | ||
a512bbf8 | 3567 | for (i = 0; i < max_mirrors; i++) { |
314b6dd0 JT |
3568 | struct page *page; |
3569 | struct bio *bio; | |
3570 | struct btrfs_super_block *disk_super; | |
3571 | ||
a512bbf8 | 3572 | bytenr = btrfs_sb_offset(i); |
935e5cc9 MX |
3573 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= |
3574 | device->commit_total_bytes) | |
a512bbf8 YZ |
3575 | break; |
3576 | ||
abbb3b8e | 3577 | btrfs_set_super_bytenr(sb, bytenr); |
4eedeb75 | 3578 | |
fd08001f EB |
3579 | crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE, |
3580 | BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, | |
3581 | sb->csum); | |
4eedeb75 | 3582 | |
314b6dd0 JT |
3583 | page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT, |
3584 | GFP_NOFS); | |
3585 | if (!page) { | |
abbb3b8e | 3586 | btrfs_err(device->fs_info, |
314b6dd0 | 3587 | "couldn't get super block page for bytenr %llu", |
abbb3b8e DS |
3588 | bytenr); |
3589 | errors++; | |
4eedeb75 | 3590 | continue; |
abbb3b8e | 3591 | } |
634554dc | 3592 | |
314b6dd0 JT |
3593 | /* Bump the refcount for wait_dev_supers() */ |
3594 | get_page(page); | |
a512bbf8 | 3595 | |
314b6dd0 JT |
3596 | disk_super = page_address(page); |
3597 | memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE); | |
4eedeb75 | 3598 | |
314b6dd0 JT |
3599 | /* |
3600 | * Directly use bios here instead of relying on the page cache | |
3601 | * to do I/O, so we don't lose the ability to do integrity | |
3602 | * checking. | |
3603 | */ | |
3604 | bio = bio_alloc(GFP_NOFS, 1); | |
3605 | bio_set_dev(bio, device->bdev); | |
3606 | bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT; | |
3607 | bio->bi_private = device; | |
3608 | bio->bi_end_io = btrfs_end_super_write; | |
3609 | __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE, | |
3610 | offset_in_page(bytenr)); | |
a512bbf8 | 3611 | |
387125fc | 3612 | /* |
314b6dd0 JT |
3613 | * We FUA only the first super block. The others we allow to |
3614 | * go down lazy and there's a short window where the on-disk | |
3615 | * copies might still contain the older version. | |
387125fc | 3616 | */ |
314b6dd0 | 3617 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO; |
1b9e619c | 3618 | if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER)) |
314b6dd0 JT |
3619 | bio->bi_opf |= REQ_FUA; |
3620 | ||
3621 | btrfsic_submit_bio(bio); | |
a512bbf8 YZ |
3622 | } |
3623 | return errors < i ? 0 : -1; | |
3624 | } | |
3625 | ||
abbb3b8e DS |
3626 | /* |
3627 | * Wait for write completion of superblocks done by write_dev_supers, | |
3628 | * @max_mirrors same for write and wait phases. | |
3629 | * | |
314b6dd0 | 3630 | * Return number of errors when page is not found or not marked up to |
abbb3b8e DS |
3631 | * date. |
3632 | */ | |
3633 | static int wait_dev_supers(struct btrfs_device *device, int max_mirrors) | |
3634 | { | |
abbb3b8e DS |
3635 | int i; |
3636 | int errors = 0; | |
b6a535fa | 3637 | bool primary_failed = false; |
abbb3b8e DS |
3638 | u64 bytenr; |
3639 | ||
3640 | if (max_mirrors == 0) | |
3641 | max_mirrors = BTRFS_SUPER_MIRROR_MAX; | |
3642 | ||
3643 | for (i = 0; i < max_mirrors; i++) { | |
314b6dd0 JT |
3644 | struct page *page; |
3645 | ||
abbb3b8e DS |
3646 | bytenr = btrfs_sb_offset(i); |
3647 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= | |
3648 | device->commit_total_bytes) | |
3649 | break; | |
3650 | ||
314b6dd0 JT |
3651 | page = find_get_page(device->bdev->bd_inode->i_mapping, |
3652 | bytenr >> PAGE_SHIFT); | |
3653 | if (!page) { | |
abbb3b8e | 3654 | errors++; |
b6a535fa HM |
3655 | if (i == 0) |
3656 | primary_failed = true; | |
abbb3b8e DS |
3657 | continue; |
3658 | } | |
314b6dd0 JT |
3659 | /* Page is submitted locked and unlocked once the IO completes */ |
3660 | wait_on_page_locked(page); | |
3661 | if (PageError(page)) { | |
abbb3b8e | 3662 | errors++; |
b6a535fa HM |
3663 | if (i == 0) |
3664 | primary_failed = true; | |
3665 | } | |
abbb3b8e | 3666 | |
314b6dd0 JT |
3667 | /* Drop our reference */ |
3668 | put_page(page); | |
abbb3b8e | 3669 | |
314b6dd0 JT |
3670 | /* Drop the reference from the writing run */ |
3671 | put_page(page); | |
abbb3b8e DS |
3672 | } |
3673 | ||
b6a535fa HM |
3674 | /* log error, force error return */ |
3675 | if (primary_failed) { | |
3676 | btrfs_err(device->fs_info, "error writing primary super block to device %llu", | |
3677 | device->devid); | |
3678 | return -1; | |
3679 | } | |
3680 | ||
abbb3b8e DS |
3681 | return errors < i ? 0 : -1; |
3682 | } | |
3683 | ||
387125fc CM |
3684 | /* |
3685 | * endio for the write_dev_flush, this will wake anyone waiting | |
3686 | * for the barrier when it is done | |
3687 | */ | |
4246a0b6 | 3688 | static void btrfs_end_empty_barrier(struct bio *bio) |
387125fc | 3689 | { |
e0ae9994 | 3690 | complete(bio->bi_private); |
387125fc CM |
3691 | } |
3692 | ||
3693 | /* | |
4fc6441a AJ |
3694 | * Submit a flush request to the device if it supports it. Error handling is |
3695 | * done in the waiting counterpart. | |
387125fc | 3696 | */ |
4fc6441a | 3697 | static void write_dev_flush(struct btrfs_device *device) |
387125fc | 3698 | { |
c2a9c7ab | 3699 | struct request_queue *q = bdev_get_queue(device->bdev); |
e0ae9994 | 3700 | struct bio *bio = device->flush_bio; |
387125fc | 3701 | |
c2a9c7ab | 3702 | if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) |
4fc6441a | 3703 | return; |
387125fc | 3704 | |
e0ae9994 | 3705 | bio_reset(bio); |
387125fc | 3706 | bio->bi_end_io = btrfs_end_empty_barrier; |
74d46992 | 3707 | bio_set_dev(bio, device->bdev); |
8d910125 | 3708 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH; |
387125fc CM |
3709 | init_completion(&device->flush_wait); |
3710 | bio->bi_private = &device->flush_wait; | |
387125fc | 3711 | |
43a01111 | 3712 | btrfsic_submit_bio(bio); |
1c3063b6 | 3713 | set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); |
4fc6441a | 3714 | } |
387125fc | 3715 | |
4fc6441a AJ |
3716 | /* |
3717 | * If the flush bio has been submitted by write_dev_flush, wait for it. | |
3718 | */ | |
8c27cb35 | 3719 | static blk_status_t wait_dev_flush(struct btrfs_device *device) |
4fc6441a | 3720 | { |
4fc6441a | 3721 | struct bio *bio = device->flush_bio; |
387125fc | 3722 | |
1c3063b6 | 3723 | if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)) |
58efbc9f | 3724 | return BLK_STS_OK; |
387125fc | 3725 | |
1c3063b6 | 3726 | clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state); |
2980d574 | 3727 | wait_for_completion_io(&device->flush_wait); |
387125fc | 3728 | |
8c27cb35 | 3729 | return bio->bi_status; |
387125fc | 3730 | } |
387125fc | 3731 | |
d10b82fe | 3732 | static int check_barrier_error(struct btrfs_fs_info *fs_info) |
401b41e5 | 3733 | { |
6528b99d | 3734 | if (!btrfs_check_rw_degradable(fs_info, NULL)) |
401b41e5 | 3735 | return -EIO; |
387125fc CM |
3736 | return 0; |
3737 | } | |
3738 | ||
3739 | /* | |
3740 | * send an empty flush down to each device in parallel, | |
3741 | * then wait for them | |
3742 | */ | |
3743 | static int barrier_all_devices(struct btrfs_fs_info *info) | |
3744 | { | |
3745 | struct list_head *head; | |
3746 | struct btrfs_device *dev; | |
5af3e8cc | 3747 | int errors_wait = 0; |
4e4cbee9 | 3748 | blk_status_t ret; |
387125fc | 3749 | |
1538e6c5 | 3750 | lockdep_assert_held(&info->fs_devices->device_list_mutex); |
387125fc CM |
3751 | /* send down all the barriers */ |
3752 | head = &info->fs_devices->devices; | |
1538e6c5 | 3753 | list_for_each_entry(dev, head, dev_list) { |
e6e674bd | 3754 | if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) |
f88ba6a2 | 3755 | continue; |
cea7c8bf | 3756 | if (!dev->bdev) |
387125fc | 3757 | continue; |
e12c9621 | 3758 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || |
ebbede42 | 3759 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) |
387125fc CM |
3760 | continue; |
3761 | ||
4fc6441a | 3762 | write_dev_flush(dev); |
58efbc9f | 3763 | dev->last_flush_error = BLK_STS_OK; |
387125fc CM |
3764 | } |
3765 | ||
3766 | /* wait for all the barriers */ | |
1538e6c5 | 3767 | list_for_each_entry(dev, head, dev_list) { |
e6e674bd | 3768 | if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state)) |
f88ba6a2 | 3769 | continue; |
387125fc | 3770 | if (!dev->bdev) { |
5af3e8cc | 3771 | errors_wait++; |
387125fc CM |
3772 | continue; |
3773 | } | |
e12c9621 | 3774 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || |
ebbede42 | 3775 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) |
387125fc CM |
3776 | continue; |
3777 | ||
4fc6441a | 3778 | ret = wait_dev_flush(dev); |
401b41e5 AJ |
3779 | if (ret) { |
3780 | dev->last_flush_error = ret; | |
66b4993e DS |
3781 | btrfs_dev_stat_inc_and_print(dev, |
3782 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
5af3e8cc | 3783 | errors_wait++; |
401b41e5 AJ |
3784 | } |
3785 | } | |
3786 | ||
cea7c8bf | 3787 | if (errors_wait) { |
401b41e5 AJ |
3788 | /* |
3789 | * At some point we need the status of all disks | |
3790 | * to arrive at the volume status. So error checking | |
3791 | * is being pushed to a separate loop. | |
3792 | */ | |
d10b82fe | 3793 | return check_barrier_error(info); |
387125fc | 3794 | } |
387125fc CM |
3795 | return 0; |
3796 | } | |
3797 | ||
943c6e99 ZL |
3798 | int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags) |
3799 | { | |
8789f4fe ZL |
3800 | int raid_type; |
3801 | int min_tolerated = INT_MAX; | |
943c6e99 | 3802 | |
8789f4fe ZL |
3803 | if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 || |
3804 | (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE)) | |
8c3e3582 | 3805 | min_tolerated = min_t(int, min_tolerated, |
8789f4fe ZL |
3806 | btrfs_raid_array[BTRFS_RAID_SINGLE]. |
3807 | tolerated_failures); | |
943c6e99 | 3808 | |
8789f4fe ZL |
3809 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
3810 | if (raid_type == BTRFS_RAID_SINGLE) | |
3811 | continue; | |
41a6e891 | 3812 | if (!(flags & btrfs_raid_array[raid_type].bg_flag)) |
8789f4fe | 3813 | continue; |
8c3e3582 | 3814 | min_tolerated = min_t(int, min_tolerated, |
8789f4fe ZL |
3815 | btrfs_raid_array[raid_type]. |
3816 | tolerated_failures); | |
3817 | } | |
943c6e99 | 3818 | |
8789f4fe | 3819 | if (min_tolerated == INT_MAX) { |
ab8d0fc4 | 3820 | pr_warn("BTRFS: unknown raid flag: %llu", flags); |
8789f4fe ZL |
3821 | min_tolerated = 0; |
3822 | } | |
3823 | ||
3824 | return min_tolerated; | |
943c6e99 ZL |
3825 | } |
3826 | ||
eece6a9c | 3827 | int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors) |
f2984462 | 3828 | { |
e5e9a520 | 3829 | struct list_head *head; |
f2984462 | 3830 | struct btrfs_device *dev; |
a061fc8d | 3831 | struct btrfs_super_block *sb; |
f2984462 | 3832 | struct btrfs_dev_item *dev_item; |
f2984462 CM |
3833 | int ret; |
3834 | int do_barriers; | |
a236aed1 CM |
3835 | int max_errors; |
3836 | int total_errors = 0; | |
a061fc8d | 3837 | u64 flags; |
f2984462 | 3838 | |
0b246afa | 3839 | do_barriers = !btrfs_test_opt(fs_info, NOBARRIER); |
fed3b381 LB |
3840 | |
3841 | /* | |
3842 | * max_mirrors == 0 indicates we're from commit_transaction, | |
3843 | * not from fsync where the tree roots in fs_info have not | |
3844 | * been consistent on disk. | |
3845 | */ | |
3846 | if (max_mirrors == 0) | |
3847 | backup_super_roots(fs_info); | |
f2984462 | 3848 | |
0b246afa | 3849 | sb = fs_info->super_for_commit; |
a061fc8d | 3850 | dev_item = &sb->dev_item; |
e5e9a520 | 3851 | |
0b246afa JM |
3852 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
3853 | head = &fs_info->fs_devices->devices; | |
3854 | max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1; | |
387125fc | 3855 | |
5af3e8cc | 3856 | if (do_barriers) { |
0b246afa | 3857 | ret = barrier_all_devices(fs_info); |
5af3e8cc SB |
3858 | if (ret) { |
3859 | mutex_unlock( | |
0b246afa JM |
3860 | &fs_info->fs_devices->device_list_mutex); |
3861 | btrfs_handle_fs_error(fs_info, ret, | |
3862 | "errors while submitting device barriers."); | |
5af3e8cc SB |
3863 | return ret; |
3864 | } | |
3865 | } | |
387125fc | 3866 | |
1538e6c5 | 3867 | list_for_each_entry(dev, head, dev_list) { |
dfe25020 CM |
3868 | if (!dev->bdev) { |
3869 | total_errors++; | |
3870 | continue; | |
3871 | } | |
e12c9621 | 3872 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || |
ebbede42 | 3873 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) |
dfe25020 CM |
3874 | continue; |
3875 | ||
2b82032c | 3876 | btrfs_set_stack_device_generation(dev_item, 0); |
a061fc8d CM |
3877 | btrfs_set_stack_device_type(dev_item, dev->type); |
3878 | btrfs_set_stack_device_id(dev_item, dev->devid); | |
7df69d3e | 3879 | btrfs_set_stack_device_total_bytes(dev_item, |
935e5cc9 | 3880 | dev->commit_total_bytes); |
ce7213c7 MX |
3881 | btrfs_set_stack_device_bytes_used(dev_item, |
3882 | dev->commit_bytes_used); | |
a061fc8d CM |
3883 | btrfs_set_stack_device_io_align(dev_item, dev->io_align); |
3884 | btrfs_set_stack_device_io_width(dev_item, dev->io_width); | |
3885 | btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); | |
3886 | memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); | |
7239ff4b NB |
3887 | memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid, |
3888 | BTRFS_FSID_SIZE); | |
a512bbf8 | 3889 | |
a061fc8d CM |
3890 | flags = btrfs_super_flags(sb); |
3891 | btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); | |
3892 | ||
75cb857d QW |
3893 | ret = btrfs_validate_write_super(fs_info, sb); |
3894 | if (ret < 0) { | |
3895 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
3896 | btrfs_handle_fs_error(fs_info, -EUCLEAN, | |
3897 | "unexpected superblock corruption detected"); | |
3898 | return -EUCLEAN; | |
3899 | } | |
3900 | ||
abbb3b8e | 3901 | ret = write_dev_supers(dev, sb, max_mirrors); |
a236aed1 CM |
3902 | if (ret) |
3903 | total_errors++; | |
f2984462 | 3904 | } |
a236aed1 | 3905 | if (total_errors > max_errors) { |
0b246afa JM |
3906 | btrfs_err(fs_info, "%d errors while writing supers", |
3907 | total_errors); | |
3908 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 3909 | |
9d565ba4 | 3910 | /* FUA is masked off if unsupported and can't be the reason */ |
0b246afa JM |
3911 | btrfs_handle_fs_error(fs_info, -EIO, |
3912 | "%d errors while writing supers", | |
3913 | total_errors); | |
9d565ba4 | 3914 | return -EIO; |
a236aed1 | 3915 | } |
f2984462 | 3916 | |
a512bbf8 | 3917 | total_errors = 0; |
1538e6c5 | 3918 | list_for_each_entry(dev, head, dev_list) { |
dfe25020 CM |
3919 | if (!dev->bdev) |
3920 | continue; | |
e12c9621 | 3921 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) || |
ebbede42 | 3922 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) |
dfe25020 CM |
3923 | continue; |
3924 | ||
abbb3b8e | 3925 | ret = wait_dev_supers(dev, max_mirrors); |
a512bbf8 YZ |
3926 | if (ret) |
3927 | total_errors++; | |
f2984462 | 3928 | } |
0b246afa | 3929 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
a236aed1 | 3930 | if (total_errors > max_errors) { |
0b246afa JM |
3931 | btrfs_handle_fs_error(fs_info, -EIO, |
3932 | "%d errors while writing supers", | |
3933 | total_errors); | |
79787eaa | 3934 | return -EIO; |
a236aed1 | 3935 | } |
f2984462 CM |
3936 | return 0; |
3937 | } | |
3938 | ||
cb517eab MX |
3939 | /* Drop a fs root from the radix tree and free it. */ |
3940 | void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, | |
3941 | struct btrfs_root *root) | |
2619ba1f | 3942 | { |
4785e24f JB |
3943 | bool drop_ref = false; |
3944 | ||
4df27c4d | 3945 | spin_lock(&fs_info->fs_roots_radix_lock); |
2619ba1f CM |
3946 | radix_tree_delete(&fs_info->fs_roots_radix, |
3947 | (unsigned long)root->root_key.objectid); | |
af01d2e5 | 3948 | if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state)) |
4785e24f | 3949 | drop_ref = true; |
4df27c4d | 3950 | spin_unlock(&fs_info->fs_roots_radix_lock); |
76dda93c | 3951 | |
1c1ea4f7 | 3952 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { |
ef67963d | 3953 | ASSERT(root->log_root == NULL); |
1c1ea4f7 | 3954 | if (root->reloc_root) { |
00246528 | 3955 | btrfs_put_root(root->reloc_root); |
1c1ea4f7 LB |
3956 | root->reloc_root = NULL; |
3957 | } | |
3958 | } | |
3321719e | 3959 | |
faa2dbf0 JB |
3960 | if (root->free_ino_pinned) |
3961 | __btrfs_remove_free_space_cache(root->free_ino_pinned); | |
3962 | if (root->free_ino_ctl) | |
3963 | __btrfs_remove_free_space_cache(root->free_ino_ctl); | |
0e996e7f JB |
3964 | if (root->ino_cache_inode) { |
3965 | iput(root->ino_cache_inode); | |
3966 | root->ino_cache_inode = NULL; | |
3967 | } | |
4785e24f JB |
3968 | if (drop_ref) |
3969 | btrfs_put_root(root); | |
2619ba1f CM |
3970 | } |
3971 | ||
c146afad | 3972 | int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) |
cfaa7295 | 3973 | { |
c146afad YZ |
3974 | u64 root_objectid = 0; |
3975 | struct btrfs_root *gang[8]; | |
65d33fd7 QW |
3976 | int i = 0; |
3977 | int err = 0; | |
3978 | unsigned int ret = 0; | |
e089f05c | 3979 | |
c146afad | 3980 | while (1) { |
efc34534 | 3981 | spin_lock(&fs_info->fs_roots_radix_lock); |
c146afad YZ |
3982 | ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, |
3983 | (void **)gang, root_objectid, | |
3984 | ARRAY_SIZE(gang)); | |
65d33fd7 | 3985 | if (!ret) { |
efc34534 | 3986 | spin_unlock(&fs_info->fs_roots_radix_lock); |
c146afad | 3987 | break; |
65d33fd7 | 3988 | } |
5d4f98a2 | 3989 | root_objectid = gang[ret - 1]->root_key.objectid + 1; |
65d33fd7 | 3990 | |
c146afad | 3991 | for (i = 0; i < ret; i++) { |
65d33fd7 QW |
3992 | /* Avoid to grab roots in dead_roots */ |
3993 | if (btrfs_root_refs(&gang[i]->root_item) == 0) { | |
3994 | gang[i] = NULL; | |
3995 | continue; | |
3996 | } | |
3997 | /* grab all the search result for later use */ | |
00246528 | 3998 | gang[i] = btrfs_grab_root(gang[i]); |
65d33fd7 | 3999 | } |
efc34534 | 4000 | spin_unlock(&fs_info->fs_roots_radix_lock); |
66b4ffd1 | 4001 | |
65d33fd7 QW |
4002 | for (i = 0; i < ret; i++) { |
4003 | if (!gang[i]) | |
4004 | continue; | |
c146afad | 4005 | root_objectid = gang[i]->root_key.objectid; |
66b4ffd1 JB |
4006 | err = btrfs_orphan_cleanup(gang[i]); |
4007 | if (err) | |
65d33fd7 | 4008 | break; |
00246528 | 4009 | btrfs_put_root(gang[i]); |
c146afad YZ |
4010 | } |
4011 | root_objectid++; | |
4012 | } | |
65d33fd7 QW |
4013 | |
4014 | /* release the uncleaned roots due to error */ | |
4015 | for (; i < ret; i++) { | |
4016 | if (gang[i]) | |
00246528 | 4017 | btrfs_put_root(gang[i]); |
65d33fd7 QW |
4018 | } |
4019 | return err; | |
c146afad | 4020 | } |
a2135011 | 4021 | |
6bccf3ab | 4022 | int btrfs_commit_super(struct btrfs_fs_info *fs_info) |
c146afad | 4023 | { |
6bccf3ab | 4024 | struct btrfs_root *root = fs_info->tree_root; |
c146afad | 4025 | struct btrfs_trans_handle *trans; |
a74a4b97 | 4026 | |
0b246afa | 4027 | mutex_lock(&fs_info->cleaner_mutex); |
2ff7e61e | 4028 | btrfs_run_delayed_iputs(fs_info); |
0b246afa JM |
4029 | mutex_unlock(&fs_info->cleaner_mutex); |
4030 | wake_up_process(fs_info->cleaner_kthread); | |
c71bf099 YZ |
4031 | |
4032 | /* wait until ongoing cleanup work done */ | |
0b246afa JM |
4033 | down_write(&fs_info->cleanup_work_sem); |
4034 | up_write(&fs_info->cleanup_work_sem); | |
c71bf099 | 4035 | |
7a7eaa40 | 4036 | trans = btrfs_join_transaction(root); |
3612b495 TI |
4037 | if (IS_ERR(trans)) |
4038 | return PTR_ERR(trans); | |
3a45bb20 | 4039 | return btrfs_commit_transaction(trans); |
c146afad YZ |
4040 | } |
4041 | ||
b105e927 | 4042 | void __cold close_ctree(struct btrfs_fs_info *fs_info) |
c146afad | 4043 | { |
c146afad YZ |
4044 | int ret; |
4045 | ||
afcdd129 | 4046 | set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags); |
d6fd0ae2 OS |
4047 | /* |
4048 | * We don't want the cleaner to start new transactions, add more delayed | |
4049 | * iputs, etc. while we're closing. We can't use kthread_stop() yet | |
4050 | * because that frees the task_struct, and the transaction kthread might | |
4051 | * still try to wake up the cleaner. | |
4052 | */ | |
4053 | kthread_park(fs_info->cleaner_kthread); | |
c146afad | 4054 | |
7343dd61 | 4055 | /* wait for the qgroup rescan worker to stop */ |
d06f23d6 | 4056 | btrfs_qgroup_wait_for_completion(fs_info, false); |
7343dd61 | 4057 | |
803b2f54 SB |
4058 | /* wait for the uuid_scan task to finish */ |
4059 | down(&fs_info->uuid_tree_rescan_sem); | |
4060 | /* avoid complains from lockdep et al., set sem back to initial state */ | |
4061 | up(&fs_info->uuid_tree_rescan_sem); | |
4062 | ||
837d5b6e | 4063 | /* pause restriper - we want to resume on mount */ |
aa1b8cd4 | 4064 | btrfs_pause_balance(fs_info); |
837d5b6e | 4065 | |
8dabb742 SB |
4066 | btrfs_dev_replace_suspend_for_unmount(fs_info); |
4067 | ||
aa1b8cd4 | 4068 | btrfs_scrub_cancel(fs_info); |
4cb5300b CM |
4069 | |
4070 | /* wait for any defraggers to finish */ | |
4071 | wait_event(fs_info->transaction_wait, | |
4072 | (atomic_read(&fs_info->defrag_running) == 0)); | |
4073 | ||
4074 | /* clear out the rbtree of defraggable inodes */ | |
26176e7c | 4075 | btrfs_cleanup_defrag_inodes(fs_info); |
4cb5300b | 4076 | |
21c7e756 | 4077 | cancel_work_sync(&fs_info->async_reclaim_work); |
57056740 | 4078 | cancel_work_sync(&fs_info->async_data_reclaim_work); |
21c7e756 | 4079 | |
b0643e59 DZ |
4080 | /* Cancel or finish ongoing discard work */ |
4081 | btrfs_discard_cleanup(fs_info); | |
4082 | ||
bc98a42c | 4083 | if (!sb_rdonly(fs_info->sb)) { |
e44163e1 | 4084 | /* |
d6fd0ae2 OS |
4085 | * The cleaner kthread is stopped, so do one final pass over |
4086 | * unused block groups. | |
e44163e1 | 4087 | */ |
0b246afa | 4088 | btrfs_delete_unused_bgs(fs_info); |
e44163e1 | 4089 | |
f0cc2cd7 FM |
4090 | /* |
4091 | * There might be existing delayed inode workers still running | |
4092 | * and holding an empty delayed inode item. We must wait for | |
4093 | * them to complete first because they can create a transaction. | |
4094 | * This happens when someone calls btrfs_balance_delayed_items() | |
4095 | * and then a transaction commit runs the same delayed nodes | |
4096 | * before any delayed worker has done something with the nodes. | |
4097 | * We must wait for any worker here and not at transaction | |
4098 | * commit time since that could cause a deadlock. | |
4099 | * This is a very rare case. | |
4100 | */ | |
4101 | btrfs_flush_workqueue(fs_info->delayed_workers); | |
4102 | ||
6bccf3ab | 4103 | ret = btrfs_commit_super(fs_info); |
acce952b | 4104 | if (ret) |
04892340 | 4105 | btrfs_err(fs_info, "commit super ret %d", ret); |
acce952b | 4106 | } |
4107 | ||
af722733 LB |
4108 | if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) || |
4109 | test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) | |
2ff7e61e | 4110 | btrfs_error_commit_super(fs_info); |
0f7d52f4 | 4111 | |
e3029d9f AV |
4112 | kthread_stop(fs_info->transaction_kthread); |
4113 | kthread_stop(fs_info->cleaner_kthread); | |
8929ecfa | 4114 | |
e187831e | 4115 | ASSERT(list_empty(&fs_info->delayed_iputs)); |
afcdd129 | 4116 | set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags); |
f25784b3 | 4117 | |
5958253c QW |
4118 | if (btrfs_check_quota_leak(fs_info)) { |
4119 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
4120 | btrfs_err(fs_info, "qgroup reserved space leaked"); | |
4121 | } | |
4122 | ||
04892340 | 4123 | btrfs_free_qgroup_config(fs_info); |
fe816d0f | 4124 | ASSERT(list_empty(&fs_info->delalloc_roots)); |
bcef60f2 | 4125 | |
963d678b | 4126 | if (percpu_counter_sum(&fs_info->delalloc_bytes)) { |
04892340 | 4127 | btrfs_info(fs_info, "at unmount delalloc count %lld", |
963d678b | 4128 | percpu_counter_sum(&fs_info->delalloc_bytes)); |
b0c68f8b | 4129 | } |
bcc63abb | 4130 | |
4297ff84 JB |
4131 | if (percpu_counter_sum(&fs_info->dio_bytes)) |
4132 | btrfs_info(fs_info, "at unmount dio bytes count %lld", | |
4133 | percpu_counter_sum(&fs_info->dio_bytes)); | |
4134 | ||
6618a59b | 4135 | btrfs_sysfs_remove_mounted(fs_info); |
b7c35e81 | 4136 | btrfs_sysfs_remove_fsid(fs_info->fs_devices); |
5ac1d209 | 4137 | |
1a4319cc LB |
4138 | btrfs_put_block_group_cache(fs_info); |
4139 | ||
de348ee0 WS |
4140 | /* |
4141 | * we must make sure there is not any read request to | |
4142 | * submit after we stopping all workers. | |
4143 | */ | |
4144 | invalidate_inode_pages2(fs_info->btree_inode->i_mapping); | |
96192499 JB |
4145 | btrfs_stop_all_workers(fs_info); |
4146 | ||
afcdd129 | 4147 | clear_bit(BTRFS_FS_OPEN, &fs_info->flags); |
4273eaff | 4148 | free_root_pointers(fs_info, true); |
8c38938c | 4149 | btrfs_free_fs_roots(fs_info); |
9ad6b7bc | 4150 | |
4e19443d JB |
4151 | /* |
4152 | * We must free the block groups after dropping the fs_roots as we could | |
4153 | * have had an IO error and have left over tree log blocks that aren't | |
4154 | * cleaned up until the fs roots are freed. This makes the block group | |
4155 | * accounting appear to be wrong because there's pending reserved bytes, | |
4156 | * so make sure we do the block group cleanup afterwards. | |
4157 | */ | |
4158 | btrfs_free_block_groups(fs_info); | |
4159 | ||
13e6c37b | 4160 | iput(fs_info->btree_inode); |
d6bfde87 | 4161 | |
21adbd5c | 4162 | #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY |
0b246afa | 4163 | if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) |
2ff7e61e | 4164 | btrfsic_unmount(fs_info->fs_devices); |
21adbd5c SB |
4165 | #endif |
4166 | ||
0b86a832 | 4167 | btrfs_mapping_tree_free(&fs_info->mapping_tree); |
68c94e55 | 4168 | btrfs_close_devices(fs_info->fs_devices); |
eb60ceac CM |
4169 | } |
4170 | ||
b9fab919 CM |
4171 | int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, |
4172 | int atomic) | |
5f39d397 | 4173 | { |
1259ab75 | 4174 | int ret; |
727011e0 | 4175 | struct inode *btree_inode = buf->pages[0]->mapping->host; |
1259ab75 | 4176 | |
0b32f4bb | 4177 | ret = extent_buffer_uptodate(buf); |
1259ab75 CM |
4178 | if (!ret) |
4179 | return ret; | |
4180 | ||
4181 | ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, | |
b9fab919 CM |
4182 | parent_transid, atomic); |
4183 | if (ret == -EAGAIN) | |
4184 | return ret; | |
1259ab75 | 4185 | return !ret; |
5f39d397 CM |
4186 | } |
4187 | ||
5f39d397 CM |
4188 | void btrfs_mark_buffer_dirty(struct extent_buffer *buf) |
4189 | { | |
0b246afa | 4190 | struct btrfs_fs_info *fs_info; |
06ea65a3 | 4191 | struct btrfs_root *root; |
5f39d397 | 4192 | u64 transid = btrfs_header_generation(buf); |
b9473439 | 4193 | int was_dirty; |
b4ce94de | 4194 | |
06ea65a3 JB |
4195 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
4196 | /* | |
4197 | * This is a fast path so only do this check if we have sanity tests | |
52042d8e | 4198 | * enabled. Normal people shouldn't be using unmapped buffers as dirty |
06ea65a3 JB |
4199 | * outside of the sanity tests. |
4200 | */ | |
b0132a3b | 4201 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags))) |
06ea65a3 JB |
4202 | return; |
4203 | #endif | |
4204 | root = BTRFS_I(buf->pages[0]->mapping->host)->root; | |
0b246afa | 4205 | fs_info = root->fs_info; |
b9447ef8 | 4206 | btrfs_assert_tree_locked(buf); |
0b246afa | 4207 | if (transid != fs_info->generation) |
5d163e0e | 4208 | WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n", |
0b246afa | 4209 | buf->start, transid, fs_info->generation); |
0b32f4bb | 4210 | was_dirty = set_extent_buffer_dirty(buf); |
e2d84521 | 4211 | if (!was_dirty) |
104b4e51 NB |
4212 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
4213 | buf->len, | |
4214 | fs_info->dirty_metadata_batch); | |
1f21ef0a | 4215 | #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY |
69fc6cbb QW |
4216 | /* |
4217 | * Since btrfs_mark_buffer_dirty() can be called with item pointer set | |
4218 | * but item data not updated. | |
4219 | * So here we should only check item pointers, not item data. | |
4220 | */ | |
4221 | if (btrfs_header_level(buf) == 0 && | |
cfdaad5e | 4222 | btrfs_check_leaf_relaxed(buf)) { |
a4f78750 | 4223 | btrfs_print_leaf(buf); |
1f21ef0a FM |
4224 | ASSERT(0); |
4225 | } | |
4226 | #endif | |
eb60ceac CM |
4227 | } |
4228 | ||
2ff7e61e | 4229 | static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info, |
b53d3f5d | 4230 | int flush_delayed) |
16cdcec7 MX |
4231 | { |
4232 | /* | |
4233 | * looks as though older kernels can get into trouble with | |
4234 | * this code, they end up stuck in balance_dirty_pages forever | |
4235 | */ | |
e2d84521 | 4236 | int ret; |
16cdcec7 MX |
4237 | |
4238 | if (current->flags & PF_MEMALLOC) | |
4239 | return; | |
4240 | ||
b53d3f5d | 4241 | if (flush_delayed) |
2ff7e61e | 4242 | btrfs_balance_delayed_items(fs_info); |
16cdcec7 | 4243 | |
d814a491 EL |
4244 | ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes, |
4245 | BTRFS_DIRTY_METADATA_THRESH, | |
4246 | fs_info->dirty_metadata_batch); | |
e2d84521 | 4247 | if (ret > 0) { |
0b246afa | 4248 | balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping); |
16cdcec7 | 4249 | } |
16cdcec7 MX |
4250 | } |
4251 | ||
2ff7e61e | 4252 | void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info) |
35b7e476 | 4253 | { |
2ff7e61e | 4254 | __btrfs_btree_balance_dirty(fs_info, 1); |
b53d3f5d | 4255 | } |
585ad2c3 | 4256 | |
2ff7e61e | 4257 | void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info) |
b53d3f5d | 4258 | { |
2ff7e61e | 4259 | __btrfs_btree_balance_dirty(fs_info, 0); |
35b7e476 | 4260 | } |
6b80053d | 4261 | |
581c1760 QW |
4262 | int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level, |
4263 | struct btrfs_key *first_key) | |
6b80053d | 4264 | { |
5ab12d1f | 4265 | return btree_read_extent_buffer_pages(buf, parent_transid, |
581c1760 | 4266 | level, first_key); |
6b80053d | 4267 | } |
0da5468f | 4268 | |
2ff7e61e | 4269 | static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info) |
acce952b | 4270 | { |
fe816d0f NB |
4271 | /* cleanup FS via transaction */ |
4272 | btrfs_cleanup_transaction(fs_info); | |
4273 | ||
0b246afa | 4274 | mutex_lock(&fs_info->cleaner_mutex); |
2ff7e61e | 4275 | btrfs_run_delayed_iputs(fs_info); |
0b246afa | 4276 | mutex_unlock(&fs_info->cleaner_mutex); |
acce952b | 4277 | |
0b246afa JM |
4278 | down_write(&fs_info->cleanup_work_sem); |
4279 | up_write(&fs_info->cleanup_work_sem); | |
acce952b | 4280 | } |
4281 | ||
ef67963d JB |
4282 | static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info) |
4283 | { | |
4284 | struct btrfs_root *gang[8]; | |
4285 | u64 root_objectid = 0; | |
4286 | int ret; | |
4287 | ||
4288 | spin_lock(&fs_info->fs_roots_radix_lock); | |
4289 | while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, | |
4290 | (void **)gang, root_objectid, | |
4291 | ARRAY_SIZE(gang))) != 0) { | |
4292 | int i; | |
4293 | ||
4294 | for (i = 0; i < ret; i++) | |
4295 | gang[i] = btrfs_grab_root(gang[i]); | |
4296 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
4297 | ||
4298 | for (i = 0; i < ret; i++) { | |
4299 | if (!gang[i]) | |
4300 | continue; | |
4301 | root_objectid = gang[i]->root_key.objectid; | |
4302 | btrfs_free_log(NULL, gang[i]); | |
4303 | btrfs_put_root(gang[i]); | |
4304 | } | |
4305 | root_objectid++; | |
4306 | spin_lock(&fs_info->fs_roots_radix_lock); | |
4307 | } | |
4308 | spin_unlock(&fs_info->fs_roots_radix_lock); | |
4309 | btrfs_free_log_root_tree(NULL, fs_info); | |
4310 | } | |
4311 | ||
143bede5 | 4312 | static void btrfs_destroy_ordered_extents(struct btrfs_root *root) |
acce952b | 4313 | { |
acce952b | 4314 | struct btrfs_ordered_extent *ordered; |
acce952b | 4315 | |
199c2a9c | 4316 | spin_lock(&root->ordered_extent_lock); |
779880ef JB |
4317 | /* |
4318 | * This will just short circuit the ordered completion stuff which will | |
4319 | * make sure the ordered extent gets properly cleaned up. | |
4320 | */ | |
199c2a9c | 4321 | list_for_each_entry(ordered, &root->ordered_extents, |
779880ef JB |
4322 | root_extent_list) |
4323 | set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); | |
199c2a9c MX |
4324 | spin_unlock(&root->ordered_extent_lock); |
4325 | } | |
4326 | ||
4327 | static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) | |
4328 | { | |
4329 | struct btrfs_root *root; | |
4330 | struct list_head splice; | |
4331 | ||
4332 | INIT_LIST_HEAD(&splice); | |
4333 | ||
4334 | spin_lock(&fs_info->ordered_root_lock); | |
4335 | list_splice_init(&fs_info->ordered_roots, &splice); | |
4336 | while (!list_empty(&splice)) { | |
4337 | root = list_first_entry(&splice, struct btrfs_root, | |
4338 | ordered_root); | |
1de2cfde JB |
4339 | list_move_tail(&root->ordered_root, |
4340 | &fs_info->ordered_roots); | |
199c2a9c | 4341 | |
2a85d9ca | 4342 | spin_unlock(&fs_info->ordered_root_lock); |
199c2a9c MX |
4343 | btrfs_destroy_ordered_extents(root); |
4344 | ||
2a85d9ca LB |
4345 | cond_resched(); |
4346 | spin_lock(&fs_info->ordered_root_lock); | |
199c2a9c MX |
4347 | } |
4348 | spin_unlock(&fs_info->ordered_root_lock); | |
74d5d229 JB |
4349 | |
4350 | /* | |
4351 | * We need this here because if we've been flipped read-only we won't | |
4352 | * get sync() from the umount, so we need to make sure any ordered | |
4353 | * extents that haven't had their dirty pages IO start writeout yet | |
4354 | * actually get run and error out properly. | |
4355 | */ | |
4356 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); | |
acce952b | 4357 | } |
4358 | ||
35a3621b | 4359 | static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, |
2ff7e61e | 4360 | struct btrfs_fs_info *fs_info) |
acce952b | 4361 | { |
4362 | struct rb_node *node; | |
4363 | struct btrfs_delayed_ref_root *delayed_refs; | |
4364 | struct btrfs_delayed_ref_node *ref; | |
4365 | int ret = 0; | |
4366 | ||
4367 | delayed_refs = &trans->delayed_refs; | |
4368 | ||
4369 | spin_lock(&delayed_refs->lock); | |
d7df2c79 | 4370 | if (atomic_read(&delayed_refs->num_entries) == 0) { |
cfece4db | 4371 | spin_unlock(&delayed_refs->lock); |
b79ce3dd | 4372 | btrfs_debug(fs_info, "delayed_refs has NO entry"); |
acce952b | 4373 | return ret; |
4374 | } | |
4375 | ||
5c9d028b | 4376 | while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) { |
d7df2c79 | 4377 | struct btrfs_delayed_ref_head *head; |
0e0adbcf | 4378 | struct rb_node *n; |
e78417d1 | 4379 | bool pin_bytes = false; |
acce952b | 4380 | |
d7df2c79 JB |
4381 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
4382 | href_node); | |
3069bd26 | 4383 | if (btrfs_delayed_ref_lock(delayed_refs, head)) |
d7df2c79 | 4384 | continue; |
3069bd26 | 4385 | |
d7df2c79 | 4386 | spin_lock(&head->lock); |
e3d03965 | 4387 | while ((n = rb_first_cached(&head->ref_tree)) != NULL) { |
0e0adbcf JB |
4388 | ref = rb_entry(n, struct btrfs_delayed_ref_node, |
4389 | ref_node); | |
d7df2c79 | 4390 | ref->in_tree = 0; |
e3d03965 | 4391 | rb_erase_cached(&ref->ref_node, &head->ref_tree); |
0e0adbcf | 4392 | RB_CLEAR_NODE(&ref->ref_node); |
1d57ee94 WX |
4393 | if (!list_empty(&ref->add_list)) |
4394 | list_del(&ref->add_list); | |
d7df2c79 JB |
4395 | atomic_dec(&delayed_refs->num_entries); |
4396 | btrfs_put_delayed_ref(ref); | |
e78417d1 | 4397 | } |
d7df2c79 JB |
4398 | if (head->must_insert_reserved) |
4399 | pin_bytes = true; | |
4400 | btrfs_free_delayed_extent_op(head->extent_op); | |
fa781cea | 4401 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 JB |
4402 | spin_unlock(&head->lock); |
4403 | spin_unlock(&delayed_refs->lock); | |
4404 | mutex_unlock(&head->mutex); | |
acce952b | 4405 | |
f603bb94 NB |
4406 | if (pin_bytes) { |
4407 | struct btrfs_block_group *cache; | |
4408 | ||
4409 | cache = btrfs_lookup_block_group(fs_info, head->bytenr); | |
4410 | BUG_ON(!cache); | |
4411 | ||
4412 | spin_lock(&cache->space_info->lock); | |
4413 | spin_lock(&cache->lock); | |
4414 | cache->pinned += head->num_bytes; | |
4415 | btrfs_space_info_update_bytes_pinned(fs_info, | |
4416 | cache->space_info, head->num_bytes); | |
4417 | cache->reserved -= head->num_bytes; | |
4418 | cache->space_info->bytes_reserved -= head->num_bytes; | |
4419 | spin_unlock(&cache->lock); | |
4420 | spin_unlock(&cache->space_info->lock); | |
4421 | percpu_counter_add_batch( | |
4422 | &cache->space_info->total_bytes_pinned, | |
4423 | head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
4424 | ||
4425 | btrfs_put_block_group(cache); | |
4426 | ||
4427 | btrfs_error_unpin_extent_range(fs_info, head->bytenr, | |
4428 | head->bytenr + head->num_bytes - 1); | |
4429 | } | |
31890da0 | 4430 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
d278850e | 4431 | btrfs_put_delayed_ref_head(head); |
acce952b | 4432 | cond_resched(); |
4433 | spin_lock(&delayed_refs->lock); | |
4434 | } | |
81f7eb00 | 4435 | btrfs_qgroup_destroy_extent_records(trans); |
acce952b | 4436 | |
4437 | spin_unlock(&delayed_refs->lock); | |
4438 | ||
4439 | return ret; | |
4440 | } | |
4441 | ||
143bede5 | 4442 | static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) |
acce952b | 4443 | { |
4444 | struct btrfs_inode *btrfs_inode; | |
4445 | struct list_head splice; | |
4446 | ||
4447 | INIT_LIST_HEAD(&splice); | |
4448 | ||
eb73c1b7 MX |
4449 | spin_lock(&root->delalloc_lock); |
4450 | list_splice_init(&root->delalloc_inodes, &splice); | |
acce952b | 4451 | |
4452 | while (!list_empty(&splice)) { | |
fe816d0f | 4453 | struct inode *inode = NULL; |
eb73c1b7 MX |
4454 | btrfs_inode = list_first_entry(&splice, struct btrfs_inode, |
4455 | delalloc_inodes); | |
fe816d0f | 4456 | __btrfs_del_delalloc_inode(root, btrfs_inode); |
eb73c1b7 | 4457 | spin_unlock(&root->delalloc_lock); |
acce952b | 4458 | |
fe816d0f NB |
4459 | /* |
4460 | * Make sure we get a live inode and that it'll not disappear | |
4461 | * meanwhile. | |
4462 | */ | |
4463 | inode = igrab(&btrfs_inode->vfs_inode); | |
4464 | if (inode) { | |
4465 | invalidate_inode_pages2(inode->i_mapping); | |
4466 | iput(inode); | |
4467 | } | |
eb73c1b7 | 4468 | spin_lock(&root->delalloc_lock); |
acce952b | 4469 | } |
eb73c1b7 MX |
4470 | spin_unlock(&root->delalloc_lock); |
4471 | } | |
4472 | ||
4473 | static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) | |
4474 | { | |
4475 | struct btrfs_root *root; | |
4476 | struct list_head splice; | |
4477 | ||
4478 | INIT_LIST_HEAD(&splice); | |
4479 | ||
4480 | spin_lock(&fs_info->delalloc_root_lock); | |
4481 | list_splice_init(&fs_info->delalloc_roots, &splice); | |
4482 | while (!list_empty(&splice)) { | |
4483 | root = list_first_entry(&splice, struct btrfs_root, | |
4484 | delalloc_root); | |
00246528 | 4485 | root = btrfs_grab_root(root); |
eb73c1b7 MX |
4486 | BUG_ON(!root); |
4487 | spin_unlock(&fs_info->delalloc_root_lock); | |
4488 | ||
4489 | btrfs_destroy_delalloc_inodes(root); | |
00246528 | 4490 | btrfs_put_root(root); |
eb73c1b7 MX |
4491 | |
4492 | spin_lock(&fs_info->delalloc_root_lock); | |
4493 | } | |
4494 | spin_unlock(&fs_info->delalloc_root_lock); | |
acce952b | 4495 | } |
4496 | ||
2ff7e61e | 4497 | static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info, |
acce952b | 4498 | struct extent_io_tree *dirty_pages, |
4499 | int mark) | |
4500 | { | |
4501 | int ret; | |
acce952b | 4502 | struct extent_buffer *eb; |
4503 | u64 start = 0; | |
4504 | u64 end; | |
acce952b | 4505 | |
4506 | while (1) { | |
4507 | ret = find_first_extent_bit(dirty_pages, start, &start, &end, | |
e6138876 | 4508 | mark, NULL); |
acce952b | 4509 | if (ret) |
4510 | break; | |
4511 | ||
91166212 | 4512 | clear_extent_bits(dirty_pages, start, end, mark); |
acce952b | 4513 | while (start <= end) { |
0b246afa JM |
4514 | eb = find_extent_buffer(fs_info, start); |
4515 | start += fs_info->nodesize; | |
fd8b2b61 | 4516 | if (!eb) |
acce952b | 4517 | continue; |
fd8b2b61 | 4518 | wait_on_extent_buffer_writeback(eb); |
acce952b | 4519 | |
fd8b2b61 JB |
4520 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, |
4521 | &eb->bflags)) | |
4522 | clear_extent_buffer_dirty(eb); | |
4523 | free_extent_buffer_stale(eb); | |
acce952b | 4524 | } |
4525 | } | |
4526 | ||
4527 | return ret; | |
4528 | } | |
4529 | ||
2ff7e61e | 4530 | static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info, |
fe119a6e | 4531 | struct extent_io_tree *unpin) |
acce952b | 4532 | { |
acce952b | 4533 | u64 start; |
4534 | u64 end; | |
4535 | int ret; | |
4536 | ||
acce952b | 4537 | while (1) { |
0e6ec385 FM |
4538 | struct extent_state *cached_state = NULL; |
4539 | ||
fcd5e742 LF |
4540 | /* |
4541 | * The btrfs_finish_extent_commit() may get the same range as | |
4542 | * ours between find_first_extent_bit and clear_extent_dirty. | |
4543 | * Hence, hold the unused_bg_unpin_mutex to avoid double unpin | |
4544 | * the same extent range. | |
4545 | */ | |
4546 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
acce952b | 4547 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 4548 | EXTENT_DIRTY, &cached_state); |
fcd5e742 LF |
4549 | if (ret) { |
4550 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
acce952b | 4551 | break; |
fcd5e742 | 4552 | } |
acce952b | 4553 | |
0e6ec385 FM |
4554 | clear_extent_dirty(unpin, start, end, &cached_state); |
4555 | free_extent_state(cached_state); | |
2ff7e61e | 4556 | btrfs_error_unpin_extent_range(fs_info, start, end); |
fcd5e742 | 4557 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
acce952b | 4558 | cond_resched(); |
4559 | } | |
4560 | ||
4561 | return 0; | |
4562 | } | |
4563 | ||
32da5386 | 4564 | static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache) |
c79a1751 LB |
4565 | { |
4566 | struct inode *inode; | |
4567 | ||
4568 | inode = cache->io_ctl.inode; | |
4569 | if (inode) { | |
4570 | invalidate_inode_pages2(inode->i_mapping); | |
4571 | BTRFS_I(inode)->generation = 0; | |
4572 | cache->io_ctl.inode = NULL; | |
4573 | iput(inode); | |
4574 | } | |
bbc37d6e | 4575 | ASSERT(cache->io_ctl.pages == NULL); |
c79a1751 LB |
4576 | btrfs_put_block_group(cache); |
4577 | } | |
4578 | ||
4579 | void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans, | |
2ff7e61e | 4580 | struct btrfs_fs_info *fs_info) |
c79a1751 | 4581 | { |
32da5386 | 4582 | struct btrfs_block_group *cache; |
c79a1751 LB |
4583 | |
4584 | spin_lock(&cur_trans->dirty_bgs_lock); | |
4585 | while (!list_empty(&cur_trans->dirty_bgs)) { | |
4586 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
32da5386 | 4587 | struct btrfs_block_group, |
c79a1751 | 4588 | dirty_list); |
c79a1751 LB |
4589 | |
4590 | if (!list_empty(&cache->io_list)) { | |
4591 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
4592 | list_del_init(&cache->io_list); | |
4593 | btrfs_cleanup_bg_io(cache); | |
4594 | spin_lock(&cur_trans->dirty_bgs_lock); | |
4595 | } | |
4596 | ||
4597 | list_del_init(&cache->dirty_list); | |
4598 | spin_lock(&cache->lock); | |
4599 | cache->disk_cache_state = BTRFS_DC_ERROR; | |
4600 | spin_unlock(&cache->lock); | |
4601 | ||
4602 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
4603 | btrfs_put_block_group(cache); | |
ba2c4d4e | 4604 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
c79a1751 LB |
4605 | spin_lock(&cur_trans->dirty_bgs_lock); |
4606 | } | |
4607 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
4608 | ||
45ae2c18 NB |
4609 | /* |
4610 | * Refer to the definition of io_bgs member for details why it's safe | |
4611 | * to use it without any locking | |
4612 | */ | |
c79a1751 LB |
4613 | while (!list_empty(&cur_trans->io_bgs)) { |
4614 | cache = list_first_entry(&cur_trans->io_bgs, | |
32da5386 | 4615 | struct btrfs_block_group, |
c79a1751 | 4616 | io_list); |
c79a1751 LB |
4617 | |
4618 | list_del_init(&cache->io_list); | |
4619 | spin_lock(&cache->lock); | |
4620 | cache->disk_cache_state = BTRFS_DC_ERROR; | |
4621 | spin_unlock(&cache->lock); | |
4622 | btrfs_cleanup_bg_io(cache); | |
4623 | } | |
4624 | } | |
4625 | ||
49b25e05 | 4626 | void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, |
2ff7e61e | 4627 | struct btrfs_fs_info *fs_info) |
49b25e05 | 4628 | { |
bbbf7243 NB |
4629 | struct btrfs_device *dev, *tmp; |
4630 | ||
2ff7e61e | 4631 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
c79a1751 LB |
4632 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
4633 | ASSERT(list_empty(&cur_trans->io_bgs)); | |
4634 | ||
bbbf7243 NB |
4635 | list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list, |
4636 | post_commit_list) { | |
4637 | list_del_init(&dev->post_commit_list); | |
4638 | } | |
4639 | ||
2ff7e61e | 4640 | btrfs_destroy_delayed_refs(cur_trans, fs_info); |
49b25e05 | 4641 | |
4a9d8bde | 4642 | cur_trans->state = TRANS_STATE_COMMIT_START; |
0b246afa | 4643 | wake_up(&fs_info->transaction_blocked_wait); |
49b25e05 | 4644 | |
4a9d8bde | 4645 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
0b246afa | 4646 | wake_up(&fs_info->transaction_wait); |
49b25e05 | 4647 | |
ccdf9b30 | 4648 | btrfs_destroy_delayed_inodes(fs_info); |
49b25e05 | 4649 | |
2ff7e61e | 4650 | btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages, |
49b25e05 | 4651 | EXTENT_DIRTY); |
fe119a6e | 4652 | btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents); |
49b25e05 | 4653 | |
4a9d8bde MX |
4654 | cur_trans->state =TRANS_STATE_COMPLETED; |
4655 | wake_up(&cur_trans->commit_wait); | |
49b25e05 JM |
4656 | } |
4657 | ||
2ff7e61e | 4658 | static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info) |
acce952b | 4659 | { |
4660 | struct btrfs_transaction *t; | |
acce952b | 4661 | |
0b246afa | 4662 | mutex_lock(&fs_info->transaction_kthread_mutex); |
acce952b | 4663 | |
0b246afa JM |
4664 | spin_lock(&fs_info->trans_lock); |
4665 | while (!list_empty(&fs_info->trans_list)) { | |
4666 | t = list_first_entry(&fs_info->trans_list, | |
724e2315 JB |
4667 | struct btrfs_transaction, list); |
4668 | if (t->state >= TRANS_STATE_COMMIT_START) { | |
9b64f57d | 4669 | refcount_inc(&t->use_count); |
0b246afa | 4670 | spin_unlock(&fs_info->trans_lock); |
2ff7e61e | 4671 | btrfs_wait_for_commit(fs_info, t->transid); |
724e2315 | 4672 | btrfs_put_transaction(t); |
0b246afa | 4673 | spin_lock(&fs_info->trans_lock); |
724e2315 JB |
4674 | continue; |
4675 | } | |
0b246afa | 4676 | if (t == fs_info->running_transaction) { |
724e2315 | 4677 | t->state = TRANS_STATE_COMMIT_DOING; |
0b246afa | 4678 | spin_unlock(&fs_info->trans_lock); |
724e2315 JB |
4679 | /* |
4680 | * We wait for 0 num_writers since we don't hold a trans | |
4681 | * handle open currently for this transaction. | |
4682 | */ | |
4683 | wait_event(t->writer_wait, | |
4684 | atomic_read(&t->num_writers) == 0); | |
4685 | } else { | |
0b246afa | 4686 | spin_unlock(&fs_info->trans_lock); |
724e2315 | 4687 | } |
2ff7e61e | 4688 | btrfs_cleanup_one_transaction(t, fs_info); |
4a9d8bde | 4689 | |
0b246afa JM |
4690 | spin_lock(&fs_info->trans_lock); |
4691 | if (t == fs_info->running_transaction) | |
4692 | fs_info->running_transaction = NULL; | |
acce952b | 4693 | list_del_init(&t->list); |
0b246afa | 4694 | spin_unlock(&fs_info->trans_lock); |
acce952b | 4695 | |
724e2315 | 4696 | btrfs_put_transaction(t); |
2ff7e61e | 4697 | trace_btrfs_transaction_commit(fs_info->tree_root); |
0b246afa | 4698 | spin_lock(&fs_info->trans_lock); |
724e2315 | 4699 | } |
0b246afa JM |
4700 | spin_unlock(&fs_info->trans_lock); |
4701 | btrfs_destroy_all_ordered_extents(fs_info); | |
ccdf9b30 JM |
4702 | btrfs_destroy_delayed_inodes(fs_info); |
4703 | btrfs_assert_delayed_root_empty(fs_info); | |
0b246afa | 4704 | btrfs_destroy_all_delalloc_inodes(fs_info); |
ef67963d | 4705 | btrfs_drop_all_logs(fs_info); |
0b246afa | 4706 | mutex_unlock(&fs_info->transaction_kthread_mutex); |
acce952b | 4707 | |
4708 | return 0; | |
4709 | } |