projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bcachefs: Simplify hash table checks
[linux-block.git] / fs / bcachefs / super-io.c
CommitLineData
1c6fdbd8
KO		 1// SPDX-License-Identifier: GPL-2.0
2
3#include "bcachefs.h"
00b8ccf7 4#include "btree_update_interior.h"
5e82a9a1 5#include "buckets.h"
1c6fdbd8
KO		 6#include "checksum.h"
7#include "disk_groups.h"
cd575ddf 8#include "ec.h"
1c6fdbd8
KO		 9#include "error.h"
10#include "io.h"
11#include "journal.h"
7d6f07ed 12#include "journal_io.h"
1dd7f9d9 13#include "journal_seq_blacklist.h"
1c6fdbd8
KO		 14#include "replicas.h"
15#include "quota.h"
16#include "super-io.h"
17#include "super.h"
18#include "vstructs.h"
19
20#include <linux/backing-dev.h>
21#include <linux/sort.h>
22
23static const struct blk_holder_ops bch2_sb_handle_bdev_ops = {
24};
25
26const char * const bch2_sb_fields[] = {
27#define x(name, nr) #name,
28 BCH_SB_FIELDS()
29#undef x
30 NULL
31};
32
33static const char *bch2_sb_field_validate(struct bch_sb *,
34 struct bch_sb_field *);
35
36struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
37 enum bch_sb_field_type type)
38{
39 struct bch_sb_field *f;
40
41 /* XXX: need locking around superblock to access optional fields */
42
43 vstruct_for_each(sb, f)
44 if (le32_to_cpu(f->type) == type)
45 return f;
46 return NULL;
47}
48
49static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
50 struct bch_sb_field *f,
51 unsigned u64s)
52{
53 unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
54 unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
55
9d8022db 56 BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
1c6fdbd8 57
187c71f6
KO		 58 if (!f && !u64s) {
59 /* nothing to do: */
60 } else if (!f) {
1c6fdbd8
KO		 61 f = vstruct_last(sb->sb);
62 memset(f, 0, sizeof(u64) * u64s);
63 f->u64s = cpu_to_le32(u64s);
64 f->type = 0;
65 } else {
66 void *src, *dst;
67
68 src = vstruct_end(f);
af9d3bc2
KO		 69
70 if (u64s) {
71 f->u64s = cpu_to_le32(u64s);
72 dst = vstruct_end(f);
73 } else {
74 dst = f;
75 }
1c6fdbd8
KO		 76
77 memmove(dst, src, vstruct_end(sb->sb) - src);
78
79 if (dst > src)
80 memset(src, 0, dst - src);
81 }
82
83 sb->sb->u64s = cpu_to_le32(sb_u64s);
84
af9d3bc2
KO		 85 return u64s ? f : NULL;
86}
87
88void bch2_sb_field_delete(struct bch_sb_handle *sb,
89 enum bch_sb_field_type type)
90{
91 struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
92
93 if (f)
94 __bch2_sb_field_resize(sb, f, 0);
1c6fdbd8
KO		 95}
96
97/* Superblock realloc/free: */
98
99void bch2_free_super(struct bch_sb_handle *sb)
100{
101 if (sb->bio)
102 kfree(sb->bio);
103 if (!IS_ERR_OR_NULL(sb->bdev))
104 blkdev_put(sb->bdev, sb->holder);
105 kfree(sb->holder);
106
9d8022db 107 kfree(sb->sb);
1c6fdbd8
KO		 108 memset(sb, 0, sizeof(*sb));
109}
110
111int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
112{
113 size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
9d8022db 114 size_t new_buffer_size;
1c6fdbd8
KO		 115 struct bch_sb *new_sb;
116 struct bio *bio;
117
9d8022db
KO		 118 if (sb->bdev)
119 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
120
121 new_buffer_size = roundup_pow_of_two(new_bytes);
122
123 if (sb->sb && sb->buffer_size >= new_buffer_size)
1c6fdbd8
KO		 124 return 0;
125
126 if (sb->have_layout) {
127 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
128
129 if (new_bytes > max_bytes) {
130 pr_err("%pg: superblock too big: want %zu but have %llu",
131 sb->bdev, new_bytes, max_bytes);
132 return -ENOSPC;
133 }
134 }
135
9d8022db 136 if (sb->buffer_size >= new_buffer_size && sb->sb)
1c6fdbd8
KO		 137 return 0;
138
139 if (dynamic_fault("bcachefs:add:super_realloc"))
140 return -ENOMEM;
141
142 if (sb->have_bio) {
9d8022db 143 unsigned nr_bvecs = DIV_ROUND_UP(new_buffer_size, PAGE_SIZE);
1c6fdbd8
KO		 144
145 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
146 if (!bio)
147 return -ENOMEM;
148
149 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
150
151 if (sb->bio)
152 kfree(sb->bio);
153 sb->bio = bio;
154 }
155
9d8022db 156 new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
1c6fdbd8
KO		 157 if (!new_sb)
158 return -ENOMEM;
159
1c6fdbd8 160 sb->sb = new_sb;
9d8022db 161 sb->buffer_size = new_buffer_size;
1c6fdbd8
KO		 162
163 return 0;
164}
165
166struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
167 enum bch_sb_field_type type,
168 unsigned u64s)
169{
170 struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
171 ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
172 ssize_t d = -old_u64s + u64s;
173
174 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
175 return NULL;
176
177 if (sb->fs_sb) {
178 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
179 struct bch_dev *ca;
180 unsigned i;
181
182 lockdep_assert_held(&c->sb_lock);
183
184 /* XXX: we're not checking that offline device have enough space */
185
186 for_each_online_member(ca, c, i) {
187 struct bch_sb_handle *sb = &ca->disk_sb;
188
189 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
190 percpu_ref_put(&ca->ref);
191 return NULL;
192 }
193 }
194 }
195
24547d09 196 f = bch2_sb_field_get(sb->sb, type);
1c6fdbd8 197 f = __bch2_sb_field_resize(sb, f, u64s);
af9d3bc2
KO		 198 if (f)
199 f->type = cpu_to_le32(type);
1c6fdbd8
KO		 200 return f;
201}
202
203/* Superblock validate: */
204
205static inline void __bch2_sb_layout_size_assert(void)
206{
207 BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
208}
209
210static const char *validate_sb_layout(struct bch_sb_layout *layout)
211{
212 u64 offset, prev_offset, max_sectors;
213 unsigned i;
214
215 if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
216 !uuid_equal(&layout->magic, &BCHFS_MAGIC))
217 return "Not a bcachefs superblock layout";
218
219 if (layout->layout_type != 0)
220 return "Invalid superblock layout type";
221
222 if (!layout->nr_superblocks)
223 return "Invalid superblock layout: no superblocks";
224
225 if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
226 return "Invalid superblock layout: too many superblocks";
227
228 max_sectors = 1 << layout->sb_max_size_bits;
229
230 prev_offset = le64_to_cpu(layout->sb_offset[0]);
231
232 for (i = 1; i < layout->nr_superblocks; i++) {
233 offset = le64_to_cpu(layout->sb_offset[i]);
234
235 if (offset < prev_offset + max_sectors)
236 return "Invalid superblock layout: superblocks overlap";
237 prev_offset = offset;
238 }
239
240 return NULL;
241}
242
243const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
244{
245 struct bch_sb *sb = disk_sb->sb;
246 struct bch_sb_field *f;
247 struct bch_sb_field_members *mi;
248 const char *err;
26609b61 249 u32 version, version_min;
1c6fdbd8
KO		 250 u16 block_size;
251
26609b61
KO		 252 version = le16_to_cpu(sb->version);
253 version_min = version >= bcachefs_metadata_version_new_versioning
254 ? le16_to_cpu(sb->version_min)
255 : version;
256
257 if (version >= bcachefs_metadata_version_max ||
258 version_min < bcachefs_metadata_version_min)
1c6fdbd8
KO		 259 return "Unsupported superblock version";
260
26609b61
KO		 261 if (version_min > version)
262 return "Bad minimum version";
263
c258f28e
KO		 264 if (sb->features[1] ||
265 (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
266 return "Filesystem has incompatible features";
267
1c6fdbd8
KO		 268 block_size = le16_to_cpu(sb->block_size);
269
270 if (!is_power_of_2(block_size) ||
271 block_size > PAGE_SECTORS)
272 return "Bad block size";
273
274 if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid)))
275 return "Bad user UUID";
276
277 if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid)))
278 return "Bad internal UUID";
279
280 if (!sb->nr_devices ||
281 sb->nr_devices <= sb->dev_idx ||
282 sb->nr_devices > BCH_SB_MEMBERS_MAX)
283 return "Bad number of member devices";
284
285 if (!BCH_SB_META_REPLICAS_WANT(sb) ||
522c25f0 286 BCH_SB_META_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
1c6fdbd8
KO		 287 return "Invalid number of metadata replicas";
288
289 if (!BCH_SB_META_REPLICAS_REQ(sb) ||
522c25f0 290 BCH_SB_META_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
1c6fdbd8
KO		 291 return "Invalid number of metadata replicas";
292
293 if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
522c25f0 294 BCH_SB_DATA_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
1c6fdbd8
KO		 295 return "Invalid number of data replicas";
296
297 if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
522c25f0 298 BCH_SB_DATA_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
1c6fdbd8
KO		 299 return "Invalid number of data replicas";
300
301 if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
302 return "Invalid metadata checksum type";
303
304 if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
305 return "Invalid metadata checksum type";
306
307 if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
308 return "Invalid compression type";
309
310 if (!BCH_SB_BTREE_NODE_SIZE(sb))
311 return "Btree node size not set";
312
313 if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
314 return "Btree node size not a power of two";
315
316 if (BCH_SB_GC_RESERVE(sb) < 5)
317 return "gc reserve percentage too small";
318
319 if (!sb->time_precision ||
320 le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
321 return "invalid time precision";
322
323 /* validate layout */
324 err = validate_sb_layout(&sb->layout);
325 if (err)
326 return err;
327
328 vstruct_for_each(sb, f) {
329 if (!f->u64s)
330 return "Invalid superblock: invalid optional field";
331
332 if (vstruct_next(f) > vstruct_last(sb))
333 return "Invalid superblock: invalid optional field";
334 }
335
336 /* members must be validated first: */
337 mi = bch2_sb_get_members(sb);
338 if (!mi)
339 return "Invalid superblock: member info area missing";
340
341 err = bch2_sb_field_validate(sb, &mi->field);
342 if (err)
343 return err;
344
345 vstruct_for_each(sb, f) {
346 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
347 continue;
348
349 err = bch2_sb_field_validate(sb, f);
350 if (err)
351 return err;
352 }
353
1c6fdbd8
KO		 354 return NULL;
355}
356
357/* device open: */
358
359static void bch2_sb_update(struct bch_fs *c)
360{
361 struct bch_sb *src = c->disk_sb.sb;
362 struct bch_sb_field_members *mi = bch2_sb_get_members(src);
363 struct bch_dev *ca;
364 unsigned i;
365
366 lockdep_assert_held(&c->sb_lock);
367
368 c->sb.uuid = src->uuid;
369 c->sb.user_uuid = src->user_uuid;
26609b61 370 c->sb.version = le16_to_cpu(src->version);
84cc758d 371 c->sb.version_min = le16_to_cpu(src->version_min);
1c6fdbd8
KO		 372 c->sb.nr_devices = src->nr_devices;
373 c->sb.clean = BCH_SB_CLEAN(src);
374 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
375 c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
376 c->sb.time_base_lo = le64_to_cpu(src->time_base_lo);
377 c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
378 c->sb.time_precision = le32_to_cpu(src->time_precision);
379 c->sb.features = le64_to_cpu(src->features[0]);
1df42b57 380 c->sb.compat = le64_to_cpu(src->compat[0]);
1c6fdbd8
KO		 381
382 for_each_member_device(ca, c, i)
383 ca->mi = bch2_mi_to_cpu(mi->members + i);
384}
385
1c6fdbd8
KO		 386static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
387{
388 struct bch_sb_field *src_f, *dst_f;
389 struct bch_sb *dst = dst_handle->sb;
af9d3bc2 390 unsigned i;
1c6fdbd8
KO		 391
392 dst->version = src->version;
26609b61 393 dst->version_min = src->version_min;
1c6fdbd8
KO		 394 dst->seq = src->seq;
395 dst->uuid = src->uuid;
396 dst->user_uuid = src->user_uuid;
397 memcpy(dst->label, src->label, sizeof(dst->label));
398
399 dst->block_size = src->block_size;
400 dst->nr_devices = src->nr_devices;
401
402 dst->time_base_lo = src->time_base_lo;
403 dst->time_base_hi = src->time_base_hi;
404 dst->time_precision = src->time_precision;
405
406 memcpy(dst->flags, src->flags, sizeof(dst->flags));
407 memcpy(dst->features, src->features, sizeof(dst->features));
408 memcpy(dst->compat, src->compat, sizeof(dst->compat));
409
af9d3bc2
KO		 410 for (i = 0; i < BCH_SB_FIELD_NR; i++) {
411 if (i == BCH_SB_FIELD_journal)
1c6fdbd8
KO		 412 continue;
413
af9d3bc2
KO		 414 src_f = bch2_sb_field_get(src, i);
415 dst_f = bch2_sb_field_get(dst, i);
1c6fdbd8 416 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
af9d3bc2 417 src_f ? le32_to_cpu(src_f->u64s) : 0);
1c6fdbd8 418
af9d3bc2
KO		 419 if (src_f)
420 memcpy(dst_f, src_f, vstruct_bytes(src_f));
1c6fdbd8
KO		 421 }
422}
423
424int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
425{
426 struct bch_sb_field_journal *journal_buckets =
427 bch2_sb_get_journal(src);
428 unsigned journal_u64s = journal_buckets
429 ? le32_to_cpu(journal_buckets->field.u64s)
430 : 0;
431 int ret;
432
433 lockdep_assert_held(&c->sb_lock);
434
435 ret = bch2_sb_realloc(&c->disk_sb,
436 le32_to_cpu(src->u64s) - journal_u64s);
437 if (ret)
438 return ret;
439
440 __copy_super(&c->disk_sb, src);
441
442 ret = bch2_sb_replicas_to_cpu_replicas(c);
443 if (ret)
444 return ret;
445
446 ret = bch2_sb_disk_groups_to_cpu(c);
447 if (ret)
448 return ret;
449
450 bch2_sb_update(c);
451 return 0;
452}
453
454int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
455{
456 struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
457 struct bch_sb_field_journal *journal_buckets =
458 bch2_sb_get_journal(dst);
459 unsigned journal_u64s = journal_buckets
460 ? le32_to_cpu(journal_buckets->field.u64s)
461 : 0;
462 unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
463 int ret;
464
465 ret = bch2_sb_realloc(&ca->disk_sb, u64s);
466 if (ret)
467 return ret;
468
469 __copy_super(&ca->disk_sb, src);
470 return 0;
471}
472
473/* read superblock: */
474
475static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
476{
477 struct bch_csum csum;
478 size_t bytes;
479reread:
480 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
481 sb->bio->bi_iter.bi_sector = offset;
9d8022db 482 bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
1c6fdbd8
KO		 483
484 if (submit_bio_wait(sb->bio))
485 return "IO error";
486
487 if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
488 !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC))
489 return "Not a bcachefs superblock";
490
26609b61
KO		 491 if (le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_min ||
492 le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
1c6fdbd8
KO		 493 return "Unsupported superblock version";
494
495 bytes = vstruct_bytes(sb->sb);
496
497 if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
498 return "Bad superblock: too big";
499
9d8022db 500 if (bytes > sb->buffer_size) {
1c6fdbd8
KO		 501 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
502 return "cannot allocate memory";
503 goto reread;
504 }
505
506 if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
507 return "unknown csum type";
508
509 /* XXX: verify MACs */
510 csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
511 null_nonce(), sb->sb);
512
513 if (bch2_crc_cmp(csum, sb->sb->csum))
514 return "bad checksum reading superblock";
515
03e183cb
KO		 516 sb->seq = le64_to_cpu(sb->sb->seq);
517
1c6fdbd8
KO		 518 return NULL;
519}
520
521int bch2_read_super(const char *path, struct bch_opts *opts,
522 struct bch_sb_handle *sb)
523{
524 u64 offset = opt_get(*opts, sb);
525 struct bch_sb_layout layout;
526 const char *err;
527 __le64 *i;
528 int ret;
529
530 pr_verbose_init(*opts, "");
531
532 memset(sb, 0, sizeof(*sb));
533 sb->mode = BLK_OPEN_READ;
534 sb->have_bio = true;
535 sb->holder = kmalloc(1, GFP_KERNEL);
536 if (!sb->holder)
537 return -ENOMEM;
538
539 if (!opt_get(*opts, noexcl))
540 sb->mode |= BLK_OPEN_EXCL;
541
542 if (!opt_get(*opts, nochanges))
543 sb->mode |= BLK_OPEN_WRITE;
544
545 sb->bdev = blkdev_get_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
546 if (IS_ERR(sb->bdev) &&
547 PTR_ERR(sb->bdev) == -EACCES &&
548 opt_get(*opts, read_only)) {
549 sb->mode &= ~BLK_OPEN_WRITE;
550
551 sb->bdev = blkdev_get_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
552 if (!IS_ERR(sb->bdev))
553 opt_set(*opts, nochanges, true);
554 }
555
556 if (IS_ERR(sb->bdev)) {
557 ret = PTR_ERR(sb->bdev);
558 goto out;
559 }
560
561 err = "cannot allocate memory";
562 ret = bch2_sb_realloc(sb, 0);
563 if (ret)
564 goto err;
565
566 ret = -EFAULT;
567 err = "dynamic fault";
568 if (bch2_fs_init_fault("read_super"))
569 goto err;
570
571 ret = -EINVAL;
572 err = read_one_super(sb, offset);
573 if (!err)
574 goto got_super;
575
576 if (opt_defined(*opts, sb))
577 goto err;
578
579 pr_err("error reading default superblock: %s", err);
580
581 /*
582 * Error reading primary superblock - read location of backup
583 * superblocks:
584 */
585 bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
586 sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
1c6fdbd8
KO		 587 /*
588 * use sb buffer to read layout, since sb buffer is page aligned but
589 * layout won't be:
590 */
885678f6 591 bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
1c6fdbd8
KO		 592
593 err = "IO error";
594 if (submit_bio_wait(sb->bio))
595 goto err;
596
597 memcpy(&layout, sb->sb, sizeof(layout));
598 err = validate_sb_layout(&layout);
599 if (err)
600 goto err;
601
602 for (i = layout.sb_offset;
603 i < layout.sb_offset + layout.nr_superblocks; i++) {
604 offset = le64_to_cpu(*i);
605
606 if (offset == opt_get(*opts, sb))
607 continue;
608
609 err = read_one_super(sb, offset);
610 if (!err)
611 goto got_super;
612 }
613
614 ret = -EINVAL;
615 goto err;
616
617got_super:
618 err = "Superblock block size smaller than device block size";
619 ret = -EINVAL;
620 if (le16_to_cpu(sb->sb->block_size) << 9 <
621 bdev_logical_block_size(sb->bdev))
622 goto err;
623
624 ret = 0;
625 sb->have_layout = true;
626out:
627 pr_verbose_init(*opts, "ret %i", ret);
628 return ret;
629err:
630 bch2_free_super(sb);
631 pr_err("error reading superblock: %s", err);
632 goto out;
633}
634
635/* write superblock: */
636
637static void write_super_endio(struct bio *bio)
638{
639 struct bch_dev *ca = bio->bi_private;
640
641 /* XXX: return errors directly */
642
0fefe8d8 643 if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
63b214e7 644 bch2_blk_status_to_str(bio->bi_status)))
1c6fdbd8
KO		 645 ca->sb_write_error = 1;
646
647 closure_put(&ca->fs->sb_write);
648 percpu_ref_put(&ca->io_ref);
649}
650
03e183cb
KO		 651static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
652{
653 struct bch_sb *sb = ca->disk_sb.sb;
654 struct bio *bio = ca->disk_sb.bio;
655
656 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
657 bio->bi_iter.bi_sector = le64_to_cpu(sb->layout.sb_offset[0]);
03e183cb
KO		 658 bio->bi_end_io = write_super_endio;
659 bio->bi_private = ca;
885678f6 660 bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
03e183cb 661
89fd25be 662 this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
03e183cb
KO		 663 bio_sectors(bio));
664
665 percpu_ref_get(&ca->io_ref);
666 closure_bio_submit(bio, &c->sb_write);
667}
668
1c6fdbd8
KO		 669static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
670{
671 struct bch_sb *sb = ca->disk_sb.sb;
672 struct bio *bio = ca->disk_sb.bio;
673
674 sb->offset = sb->layout.sb_offset[idx];
675
676 SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
677 sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
678 null_nonce(), sb);
679
680 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
681 bio->bi_iter.bi_sector = le64_to_cpu(sb->offset);
1c6fdbd8
KO		 682 bio->bi_end_io = write_super_endio;
683 bio->bi_private = ca;
885678f6
KO		 684 bch2_bio_map(bio, sb,
685 roundup((size_t) vstruct_bytes(sb),
686 bdev_logical_block_size(ca->disk_sb.bdev)));
1c6fdbd8 687
89fd25be 688 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
1c6fdbd8
KO		 689 bio_sectors(bio));
690
691 percpu_ref_get(&ca->io_ref);
692 closure_bio_submit(bio, &c->sb_write);
693}
694
03e183cb 695int bch2_write_super(struct bch_fs *c)
1c6fdbd8
KO		 696{
697 struct closure *cl = &c->sb_write;
698 struct bch_dev *ca;
699 unsigned i, sb = 0, nr_wrote;
700 const char *err;
701 struct bch_devs_mask sb_written;
702 bool wrote, can_mount_without_written, can_mount_with_written;
98f2197d 703 unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
03e183cb 704 int ret = 0;
1c6fdbd8 705
98f2197d
KO		 706 if (c->opts.very_degraded)
707 degraded_flags |= BCH_FORCE_IF_LOST;
708
1c6fdbd8
KO		 709 lockdep_assert_held(&c->sb_lock);
710
711 closure_init_stack(cl);
712 memset(&sb_written, 0, sizeof(sb_written));
713
714 le64_add_cpu(&c->disk_sb.sb->seq, 1);
715
0bc166ff
KO		 716 if (test_bit(BCH_FS_ERROR, &c->flags))
717 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
718
7d6f07ed
KO		 719 SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
720
1c6fdbd8
KO		 721 for_each_online_member(ca, c, i)
722 bch2_sb_from_fs(c, ca);
723
724 for_each_online_member(ca, c, i) {
725 err = bch2_sb_validate(&ca->disk_sb);
726 if (err) {
727 bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
03e183cb 728 ret = -1;
1c6fdbd8
KO		 729 goto out;
730 }
731 }
732
0bc166ff 733 if (c->opts.nochanges)
1c6fdbd8
KO		 734 goto out;
735
736 for_each_online_member(ca, c, i) {
737 __set_bit(ca->dev_idx, sb_written.d);
738 ca->sb_write_error = 0;
739 }
740
03e183cb
KO		 741 for_each_online_member(ca, c, i)
742 read_back_super(c, ca);
743 closure_sync(cl);
744
745 for_each_online_member(ca, c, i) {
746 if (!ca->sb_write_error &&
747 ca->disk_sb.seq !=
748 le64_to_cpu(ca->sb_read_scratch->seq)) {
749 bch2_fs_fatal_error(c,
750 "Superblock modified by another process");
751 percpu_ref_put(&ca->io_ref);
752 ret = -EROFS;
753 goto out;
754 }
755 }
756
1c6fdbd8
KO		 757 do {
758 wrote = false;
759 for_each_online_member(ca, c, i)
03e183cb
KO		 760 if (!ca->sb_write_error &&
761 sb < ca->disk_sb.sb->layout.nr_superblocks) {
1c6fdbd8
KO		 762 write_one_super(c, ca, sb);
763 wrote = true;
764 }
765 closure_sync(cl);
766 sb++;
767 } while (wrote);
768
03e183cb 769 for_each_online_member(ca, c, i) {
1c6fdbd8
KO		 770 if (ca->sb_write_error)
771 __clear_bit(ca->dev_idx, sb_written.d);
03e183cb
KO		 772 else
773 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
774 }
1c6fdbd8
KO		 775
776 nr_wrote = dev_mask_nr(&sb_written);
777
778 can_mount_with_written =
98f2197d 779 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1c6fdbd8
KO		 780
781 for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
782 sb_written.d[i] = ~sb_written.d[i];
783
784 can_mount_without_written =
98f2197d 785 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1c6fdbd8
KO		 786
787 /*
788 * If we would be able to mount _without_ the devices we successfully
789 * wrote superblocks to, we weren't able to write to enough devices:
790 *
791 * Exception: if we can mount without the successes because we haven't
792 * written anything (new filesystem), we continue if we'd be able to
793 * mount with the devices we did successfully write to:
794 */
03e183cb 795 if (bch2_fs_fatal_err_on(!nr_wrote ||
fcb3431b 796 !can_mount_with_written ||
03e183cb
KO		 797 (can_mount_without_written &&
798 !can_mount_with_written), c,
799 "Unable to write superblock to sufficient devices"))
800 ret = -1;
1c6fdbd8
KO		 801out:
802 /* Make new options visible after they're persistent: */
803 bch2_sb_update(c);
03e183cb 804 return ret;
1c6fdbd8
KO		 805}
806
ba239c95
KO		 807void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
808{
809 mutex_lock(&c->sb_lock);
810 if (!(c->sb.features & (1ULL << feat))) {
811 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
812
813 bch2_write_super(c);
814 }
815 mutex_unlock(&c->sb_lock);
816}
817
1c6fdbd8
KO		 818/* BCH_SB_FIELD_journal: */
819
820static int u64_cmp(const void *_l, const void *_r)
821{
822 u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
823
824 return l < r ? -1 : l > r ? 1 : 0;
825}
826
827static const char *bch2_sb_validate_journal(struct bch_sb *sb,
828 struct bch_sb_field *f)
829{
830 struct bch_sb_field_journal *journal = field_to_type(f, journal);
831 struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
832 const char *err;
833 unsigned nr;
834 unsigned i;
835 u64 *b;
836
837 journal = bch2_sb_get_journal(sb);
838 if (!journal)
839 return NULL;
840
841 nr = bch2_nr_journal_buckets(journal);
842 if (!nr)
843 return NULL;
844
845 b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
846 if (!b)
847 return "cannot allocate memory";
848
849 for (i = 0; i < nr; i++)
850 b[i] = le64_to_cpu(journal->buckets[i]);
851
852 sort(b, nr, sizeof(u64), u64_cmp, NULL);
853
854 err = "journal bucket at sector 0";
855 if (!b[0])
856 goto err;
857
858 err = "journal bucket before first bucket";
859 if (m && b[0] < le16_to_cpu(m->first_bucket))
860 goto err;
861
862 err = "journal bucket past end of device";
863 if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
864 goto err;
865
866 err = "duplicate journal buckets";
867 for (i = 0; i + 1 < nr; i++)
868 if (b[i] == b[i + 1])
869 goto err;
870
871 err = NULL;
872err:
873 kfree(b);
874 return err;
875}
876
877static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
878 .validate = bch2_sb_validate_journal,
879};
880
881/* BCH_SB_FIELD_members: */
882
883static const char *bch2_sb_validate_members(struct bch_sb *sb,
884 struct bch_sb_field *f)
885{
886 struct bch_sb_field_members *mi = field_to_type(f, members);
887 struct bch_member *m;
888
889 if ((void *) (mi->members + sb->nr_devices) >
890 vstruct_end(&mi->field))
891 return "Invalid superblock: bad member info";
892
893 for (m = mi->members;
894 m < mi->members + sb->nr_devices;
895 m++) {
896 if (!bch2_member_exists(m))
897 continue;
898
899 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
900 return "Too many buckets";
901
902 if (le64_to_cpu(m->nbuckets) -
8b335bae 903 le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
1c6fdbd8
KO		 904 return "Not enough buckets";
905
906 if (le16_to_cpu(m->bucket_size) <
907 le16_to_cpu(sb->block_size))
908 return "bucket size smaller than block size";
909
910 if (le16_to_cpu(m->bucket_size) <
911 BCH_SB_BTREE_NODE_SIZE(sb))
912 return "bucket size smaller than btree node size";
913 }
914
1c6fdbd8
KO		 915 return NULL;
916}
917
918static const struct bch_sb_field_ops bch_sb_field_ops_members = {
919 .validate = bch2_sb_validate_members,
920};
921
922/* BCH_SB_FIELD_crypt: */
923
924static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
925 struct bch_sb_field *f)
926{
927 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
928
929 if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
930 return "invalid field crypt: wrong size";
931
932 if (BCH_CRYPT_KDF_TYPE(crypt))
933 return "invalid field crypt: bad kdf type";
934
935 return NULL;
936}
937
938static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
939 .validate = bch2_sb_validate_crypt,
940};
941
942/* BCH_SB_FIELD_clean: */
943
7d6f07ed 944int bch2_sb_clean_validate(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
26609b61
KO		 945{
946 struct jset_entry *entry;
7d6f07ed 947 int ret;
26609b61
KO		 948
949 for (entry = clean->start;
950 entry < (struct jset_entry *) vstruct_end(&clean->field);
7d6f07ed
KO		 951 entry = vstruct_next(entry)) {
952 ret = bch2_journal_entry_validate(c, "superblock", entry,
953 le16_to_cpu(c->disk_sb.sb->version),
954 BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
955 write);
956 if (ret)
957 return ret;
958 }
959
960 return 0;
26609b61
KO		 961}
962
134915f3 963int bch2_fs_mark_dirty(struct bch_fs *c)
1c6fdbd8 964{
03e183cb
KO		 965 int ret;
966
967 /*
968 * Unconditionally write superblock, to verify it hasn't changed before
969 * we go rw:
970 */
971
1c6fdbd8 972 mutex_lock(&c->sb_lock);
03e183cb 973 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
8042b5b7 974 c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALWAYS;
03e183cb 975 ret = bch2_write_super(c);
3ccc5c50 976 mutex_unlock(&c->sb_lock);
134915f3 977
03e183cb 978 return ret;
3ccc5c50 979}
1c6fdbd8 980
2abe5420 981static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
43cfbad6 982{
2abe5420
KO		 983 struct jset_entry *entry = *end;
984 unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
43cfbad6 985
2abe5420 986 memset(entry, 0, u64s * sizeof(u64));
43cfbad6
JH		 987 /*
988 * The u64s field counts from the start of data, ignoring the shared
989 * fields.
990 */
991 entry->u64s = u64s - 1;
43cfbad6 992
2abe5420
KO		 993 *end = vstruct_next(*end);
994 return entry;
43cfbad6
JH		 995}
996
2abe5420
KO		 997void bch2_journal_super_entries_add_common(struct bch_fs *c,
998 struct jset_entry **end,
999 u64 journal_seq)
3ccc5c50 1000{
180fb49d
KO		 1001 struct bch_dev *ca;
1002 unsigned i, dev;
1c6fdbd8 1003
5e82a9a1 1004 percpu_down_read(&c->mark_lock);
3ccc5c50 1005
5e82a9a1
KO		 1006 if (!journal_seq) {
1007 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1008 bch2_fs_usage_acc_to_base(c, i);
1009 } else {
f299d573 1010 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
5e82a9a1 1011 }
3ccc5c50
KO		 1012
1013 {
3577df5f 1014 struct jset_entry_usage *u =
2abe5420
KO		 1015 container_of(jset_entry_init(end, sizeof(*u)),
1016 struct jset_entry_usage, entry);
3ccc5c50 1017
3ccc5c50 1018 u->entry.type = BCH_JSET_ENTRY_usage;
3577df5f 1019 u->entry.btree_id = FS_USAGE_INODES;
5e82a9a1 1020 u->v = cpu_to_le64(c->usage_base->nr_inodes);
3ccc5c50
KO		 1021 }
1022
1023 {
3577df5f 1024 struct jset_entry_usage *u =
2abe5420
KO		 1025 container_of(jset_entry_init(end, sizeof(*u)),
1026 struct jset_entry_usage, entry);
3577df5f 1027
3ccc5c50 1028 u->entry.type = BCH_JSET_ENTRY_usage;
3577df5f
KO		 1029 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1030 u->v = cpu_to_le64(atomic64_read(&c->key_version));
3577df5f
KO		 1031 }
1032
1033 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1034 struct jset_entry_usage *u =
2abe5420
KO		 1035 container_of(jset_entry_init(end, sizeof(*u)),
1036 struct jset_entry_usage, entry);
3577df5f 1037
3577df5f
KO		 1038 u->entry.type = BCH_JSET_ENTRY_usage;
1039 u->entry.btree_id = FS_USAGE_RESERVED;
1040 u->entry.level = i;
5e82a9a1 1041 u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
3ccc5c50
KO		 1042 }
1043
1044 for (i = 0; i < c->replicas.nr; i++) {
1045 struct bch_replicas_entry *e =
1046 cpu_replicas_entry(&c->replicas, i);
3577df5f 1047 struct jset_entry_data_usage *u =
2abe5420
KO		 1048 container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1049 struct jset_entry_data_usage, entry);
3ccc5c50 1050
3577df5f 1051 u->entry.type = BCH_JSET_ENTRY_data_usage;
5e82a9a1 1052 u->v = cpu_to_le64(c->usage_base->replicas[i]);
3ccc5c50
KO		 1053 unsafe_memcpy(&u->r, e, replicas_entry_bytes(e),
1054 "embedded variable length struct");
3ccc5c50
KO		 1055 }
1056
180fb49d
KO		 1057 for_each_member_device(ca, c, dev) {
1058 unsigned b = sizeof(struct jset_entry_dev_usage) +
1059 sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1060 struct jset_entry_dev_usage *u =
1061 container_of(jset_entry_init(end, b),
1062 struct jset_entry_dev_usage, entry);
1063
1064 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1065 u->dev = cpu_to_le32(dev);
1066 u->buckets_ec = cpu_to_le64(ca->usage_base->buckets_ec);
1067 u->buckets_unavailable = cpu_to_le64(ca->usage_base->buckets_unavailable);
1068
1069 for (i = 0; i < BCH_DATA_NR; i++) {
1070 u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1071 u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1072 u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1073 }
1074 }
1075
5e82a9a1 1076 percpu_up_read(&c->mark_lock);
3ccc5c50 1077
2abe5420
KO		 1078 for (i = 0; i < 2; i++) {
1079 struct jset_entry_clock *clock =
1080 container_of(jset_entry_init(end, sizeof(*clock)),
1081 struct jset_entry_clock, entry);
1082
1083 clock->entry.type = BCH_JSET_ENTRY_clock;
1084 clock->rw = i;
1085 clock->time = atomic64_read(&c->io_clock[i].now);
1086 }
3ccc5c50
KO		 1087}
1088
134915f3 1089void bch2_fs_mark_clean(struct bch_fs *c)
3ccc5c50
KO		 1090{
1091 struct bch_sb_field_clean *sb_clean;
1092 struct jset_entry *entry;
1093 unsigned u64s;
7d6f07ed 1094 int ret;
3ccc5c50 1095
3ccc5c50
KO		 1096 mutex_lock(&c->sb_lock);
1097 if (BCH_SB_CLEAN(c->disk_sb.sb))
1098 goto out;
1099
1100 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1101
19dd3172
KO		 1102 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_alloc_info;
1103 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_alloc_metadata;
5d548743 1104 c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_extents_above_btree_updates);
6357d607 1105 c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_btree_updates_journalled);
1df42b57 1106
3ccc5c50 1107 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1c6fdbd8
KO		 1108
1109 sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1110 if (!sb_clean) {
1111 bch_err(c, "error resizing superblock while setting filesystem clean");
1112 goto out;
1113 }
1114
1115 sb_clean->flags = 0;
812af308
KO		 1116 sb_clean->journal_seq = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1117
1118 /* Trying to catch outstanding bug: */
1119 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1c6fdbd8
KO		 1120
1121 entry = sb_clean->start;
2abe5420 1122 bch2_journal_super_entries_add_common(c, &entry, 0);
00b8ccf7 1123 entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
3ccc5c50
KO		 1124 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1125
1c6fdbd8
KO		 1126 memset(entry, 0,
1127 vstruct_end(&sb_clean->field) - (void *) entry);
1128
7d6f07ed
KO		 1129 /*
1130 * this should be in the write path, and we should be validating every
1131 * superblock section:
1132 */
1133 ret = bch2_sb_clean_validate(c, sb_clean, WRITE);
1134 if (ret) {
1135 bch_err(c, "error writing marking filesystem clean: validate error");
1136 goto out;
1137 }
26609b61 1138
1c6fdbd8
KO		 1139 bch2_write_super(c);
1140out:
1141 mutex_unlock(&c->sb_lock);
1142}
1143
1144static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1145 struct bch_sb_field *f)
1146{
1147 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1148
1149 if (vstruct_bytes(&clean->field) < sizeof(*clean))
1150 return "invalid field crypt: wrong size";
1151
1152 return NULL;
1153}
1154
1155static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1156 .validate = bch2_sb_validate_clean,
1157};
1158
1159static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1160#define x(f, nr) \
1161 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1162 BCH_SB_FIELDS()
1163#undef x
1164};
1165
1166static const char *bch2_sb_field_validate(struct bch_sb *sb,
1167 struct bch_sb_field *f)
1168{
1169 unsigned type = le32_to_cpu(f->type);
1170
1171 return type < BCH_SB_FIELD_NR
1172 ? bch2_sb_field_ops[type]->validate(sb, f)
1173 : NULL;
1174}
1175
319f9ac3
KO		 1176void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1177 struct bch_sb_field *f)
1c6fdbd8
KO		 1178{
1179 unsigned type = le32_to_cpu(f->type);
319f9ac3
KO		 1180 const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1181 ? bch2_sb_field_ops[type] : NULL;
1c6fdbd8 1182
319f9ac3
KO		 1183 if (ops)
1184 pr_buf(out, "%s", bch2_sb_fields[type]);
1185 else
1186 pr_buf(out, "(unknown field %u)", type);
1187
1188 pr_buf(out, " (size %llu):", vstruct_bytes(f));
1c6fdbd8 1189
319f9ac3
KO		 1190 if (ops && ops->to_text)
1191 bch2_sb_field_ops[type]->to_text(out, sb, f);
1c6fdbd8 1192}
Linux 6.x block layer and io_uring tree(s)