projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bcachefs: Refactor filesystem usage accounting
[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"
1dd7f9d9 12#include "journal_seq_blacklist.h"
1c6fdbd8
KO		 13#include "replicas.h"
14#include "quota.h"
15#include "super-io.h"
16#include "super.h"
17#include "vstructs.h"
18
19#include <linux/backing-dev.h>
20#include <linux/sort.h>
21
22static const struct blk_holder_ops bch2_sb_handle_bdev_ops = {
23};
24
25const char * const bch2_sb_fields[] = {
26#define x(name, nr) #name,
27 BCH_SB_FIELDS()
28#undef x
29 NULL
30};
31
32static const char *bch2_sb_field_validate(struct bch_sb *,
33 struct bch_sb_field *);
34
35struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
36 enum bch_sb_field_type type)
37{
38 struct bch_sb_field *f;
39
40 /* XXX: need locking around superblock to access optional fields */
41
42 vstruct_for_each(sb, f)
43 if (le32_to_cpu(f->type) == type)
44 return f;
45 return NULL;
46}
47
48static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
49 struct bch_sb_field *f,
50 unsigned u64s)
51{
52 unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
53 unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
54
55 BUG_ON(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
56 sb->page_order);
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
107 free_pages((unsigned long) sb->sb, sb->page_order);
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);
114 unsigned order = get_order(new_bytes);
115 struct bch_sb *new_sb;
116 struct bio *bio;
117
118 if (sb->sb && sb->page_order >= order)
119 return 0;
120
121 if (sb->have_layout) {
122 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
123
124 if (new_bytes > max_bytes) {
125 pr_err("%pg: superblock too big: want %zu but have %llu",
126 sb->bdev, new_bytes, max_bytes);
127 return -ENOSPC;
128 }
129 }
130
131 if (sb->page_order >= order && sb->sb)
132 return 0;
133
134 if (dynamic_fault("bcachefs:add:super_realloc"))
135 return -ENOMEM;
136
137 if (sb->have_bio) {
138 unsigned nr_bvecs = 1 << order;
139
140 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
141 if (!bio)
142 return -ENOMEM;
143
144 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
145
146 if (sb->bio)
147 kfree(sb->bio);
148 sb->bio = bio;
149 }
150
66393392 151 new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
1c6fdbd8
KO		 152 if (!new_sb)
153 return -ENOMEM;
154
155 if (sb->sb)
156 memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
157
158 free_pages((unsigned long) sb->sb, sb->page_order);
159 sb->sb = new_sb;
160
161 sb->page_order = order;
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) ||
286 BCH_SB_META_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
287 return "Invalid number of metadata replicas";
288
289 if (!BCH_SB_META_REPLICAS_REQ(sb) ||
290 BCH_SB_META_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
291 return "Invalid number of metadata replicas";
292
293 if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
294 BCH_SB_DATA_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
295 return "Invalid number of data replicas";
296
297 if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
298 BCH_SB_DATA_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
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);
1c6fdbd8
KO		 371 c->sb.nr_devices = src->nr_devices;
372 c->sb.clean = BCH_SB_CLEAN(src);
373 c->sb.encryption_type = BCH_SB_ENCRYPTION_TYPE(src);
374 c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
375 c->sb.time_base_lo = le64_to_cpu(src->time_base_lo);
376 c->sb.time_base_hi = le32_to_cpu(src->time_base_hi);
377 c->sb.time_precision = le32_to_cpu(src->time_precision);
378 c->sb.features = le64_to_cpu(src->features[0]);
1df42b57 379 c->sb.compat = le64_to_cpu(src->compat[0]);
1c6fdbd8
KO		 380
381 for_each_member_device(ca, c, i)
382 ca->mi = bch2_mi_to_cpu(mi->members + i);
383}
384
385/* doesn't copy member info */
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;
885678f6 482 bch2_bio_map(sb->bio, sb->sb, PAGE_SIZE << sb->page_order);
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
500 if (get_order(bytes) > sb->page_order) {
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
306d40df 643 if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write: %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;
03e183cb 703 int ret = 0;
1c6fdbd8
KO		 704
705 lockdep_assert_held(&c->sb_lock);
706
707 closure_init_stack(cl);
708 memset(&sb_written, 0, sizeof(sb_written));
709
710 le64_add_cpu(&c->disk_sb.sb->seq, 1);
711
0bc166ff
KO		 712 if (test_bit(BCH_FS_ERROR, &c->flags))
713 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
714
1c6fdbd8
KO		 715 for_each_online_member(ca, c, i)
716 bch2_sb_from_fs(c, ca);
717
718 for_each_online_member(ca, c, i) {
719 err = bch2_sb_validate(&ca->disk_sb);
720 if (err) {
721 bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
03e183cb 722 ret = -1;
1c6fdbd8
KO		 723 goto out;
724 }
725 }
726
0bc166ff 727 if (c->opts.nochanges)
1c6fdbd8
KO		 728 goto out;
729
730 for_each_online_member(ca, c, i) {
731 __set_bit(ca->dev_idx, sb_written.d);
732 ca->sb_write_error = 0;
733 }
734
03e183cb
KO		 735 for_each_online_member(ca, c, i)
736 read_back_super(c, ca);
737 closure_sync(cl);
738
739 for_each_online_member(ca, c, i) {
740 if (!ca->sb_write_error &&
741 ca->disk_sb.seq !=
742 le64_to_cpu(ca->sb_read_scratch->seq)) {
743 bch2_fs_fatal_error(c,
744 "Superblock modified by another process");
745 percpu_ref_put(&ca->io_ref);
746 ret = -EROFS;
747 goto out;
748 }
749 }
750
1c6fdbd8
KO		 751 do {
752 wrote = false;
753 for_each_online_member(ca, c, i)
03e183cb
KO		 754 if (!ca->sb_write_error &&
755 sb < ca->disk_sb.sb->layout.nr_superblocks) {
1c6fdbd8
KO		 756 write_one_super(c, ca, sb);
757 wrote = true;
758 }
759 closure_sync(cl);
760 sb++;
761 } while (wrote);
762
03e183cb 763 for_each_online_member(ca, c, i) {
1c6fdbd8
KO		 764 if (ca->sb_write_error)
765 __clear_bit(ca->dev_idx, sb_written.d);
03e183cb
KO		 766 else
767 ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
768 }
1c6fdbd8
KO		 769
770 nr_wrote = dev_mask_nr(&sb_written);
771
772 can_mount_with_written =
773 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
774 BCH_FORCE_IF_DEGRADED);
775
776 for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
777 sb_written.d[i] = ~sb_written.d[i];
778
779 can_mount_without_written =
780 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
781 BCH_FORCE_IF_DEGRADED);
782
783 /*
784 * If we would be able to mount _without_ the devices we successfully
785 * wrote superblocks to, we weren't able to write to enough devices:
786 *
787 * Exception: if we can mount without the successes because we haven't
788 * written anything (new filesystem), we continue if we'd be able to
789 * mount with the devices we did successfully write to:
790 */
03e183cb
KO		 791 if (bch2_fs_fatal_err_on(!nr_wrote ||
792 (can_mount_without_written &&
793 !can_mount_with_written), c,
794 "Unable to write superblock to sufficient devices"))
795 ret = -1;
1c6fdbd8
KO		 796out:
797 /* Make new options visible after they're persistent: */
798 bch2_sb_update(c);
03e183cb 799 return ret;
1c6fdbd8
KO		 800}
801
ba239c95
KO		 802void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
803{
804 mutex_lock(&c->sb_lock);
805 if (!(c->sb.features & (1ULL << feat))) {
806 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
807
808 bch2_write_super(c);
809 }
810 mutex_unlock(&c->sb_lock);
811}
812
1c6fdbd8
KO		 813/* BCH_SB_FIELD_journal: */
814
815static int u64_cmp(const void *_l, const void *_r)
816{
817 u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
818
819 return l < r ? -1 : l > r ? 1 : 0;
820}
821
822static const char *bch2_sb_validate_journal(struct bch_sb *sb,
823 struct bch_sb_field *f)
824{
825 struct bch_sb_field_journal *journal = field_to_type(f, journal);
826 struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
827 const char *err;
828 unsigned nr;
829 unsigned i;
830 u64 *b;
831
832 journal = bch2_sb_get_journal(sb);
833 if (!journal)
834 return NULL;
835
836 nr = bch2_nr_journal_buckets(journal);
837 if (!nr)
838 return NULL;
839
840 b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
841 if (!b)
842 return "cannot allocate memory";
843
844 for (i = 0; i < nr; i++)
845 b[i] = le64_to_cpu(journal->buckets[i]);
846
847 sort(b, nr, sizeof(u64), u64_cmp, NULL);
848
849 err = "journal bucket at sector 0";
850 if (!b[0])
851 goto err;
852
853 err = "journal bucket before first bucket";
854 if (m && b[0] < le16_to_cpu(m->first_bucket))
855 goto err;
856
857 err = "journal bucket past end of device";
858 if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
859 goto err;
860
861 err = "duplicate journal buckets";
862 for (i = 0; i + 1 < nr; i++)
863 if (b[i] == b[i + 1])
864 goto err;
865
866 err = NULL;
867err:
868 kfree(b);
869 return err;
870}
871
872static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
873 .validate = bch2_sb_validate_journal,
874};
875
876/* BCH_SB_FIELD_members: */
877
878static const char *bch2_sb_validate_members(struct bch_sb *sb,
879 struct bch_sb_field *f)
880{
881 struct bch_sb_field_members *mi = field_to_type(f, members);
882 struct bch_member *m;
883
884 if ((void *) (mi->members + sb->nr_devices) >
885 vstruct_end(&mi->field))
886 return "Invalid superblock: bad member info";
887
888 for (m = mi->members;
889 m < mi->members + sb->nr_devices;
890 m++) {
891 if (!bch2_member_exists(m))
892 continue;
893
894 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
895 return "Too many buckets";
896
897 if (le64_to_cpu(m->nbuckets) -
8b335bae 898 le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
1c6fdbd8
KO		 899 return "Not enough buckets";
900
901 if (le16_to_cpu(m->bucket_size) <
902 le16_to_cpu(sb->block_size))
903 return "bucket size smaller than block size";
904
905 if (le16_to_cpu(m->bucket_size) <
906 BCH_SB_BTREE_NODE_SIZE(sb))
907 return "bucket size smaller than btree node size";
908 }
909
1c6fdbd8
KO		 910 return NULL;
911}
912
913static const struct bch_sb_field_ops bch_sb_field_ops_members = {
914 .validate = bch2_sb_validate_members,
915};
916
917/* BCH_SB_FIELD_crypt: */
918
919static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
920 struct bch_sb_field *f)
921{
922 struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
923
924 if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
925 return "invalid field crypt: wrong size";
926
927 if (BCH_CRYPT_KDF_TYPE(crypt))
928 return "invalid field crypt: bad kdf type";
929
930 return NULL;
931}
932
933static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
934 .validate = bch2_sb_validate_crypt,
935};
936
937/* BCH_SB_FIELD_clean: */
938
26609b61
KO		 939void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
940{
941 struct jset_entry *entry;
942
943 for (entry = clean->start;
944 entry < (struct jset_entry *) vstruct_end(&clean->field);
945 entry = vstruct_next(entry))
946 bch2_bkey_renumber(BKEY_TYPE_BTREE, bkey_to_packed(entry->start), write);
947}
948
134915f3 949int bch2_fs_mark_dirty(struct bch_fs *c)
1c6fdbd8 950{
03e183cb
KO		 951 int ret;
952
953 /*
954 * Unconditionally write superblock, to verify it hasn't changed before
955 * we go rw:
956 */
957
1c6fdbd8 958 mutex_lock(&c->sb_lock);
03e183cb 959 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
97328a1a 960 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
5d548743 961 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_extents_above_btree_updates;
6357d607 962 c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_btree_updates_journalled;
03e183cb 963 ret = bch2_write_super(c);
3ccc5c50 964 mutex_unlock(&c->sb_lock);
134915f3 965
03e183cb 966 return ret;
3ccc5c50 967}
1c6fdbd8 968
43cfbad6
JH		 969static void
970entry_init_u64s(struct jset_entry *entry, unsigned u64s)
971{
972 memset(entry, 0, u64s * sizeof(u64));
973
974 /*
975 * The u64s field counts from the start of data, ignoring the shared
976 * fields.
977 */
978 entry->u64s = u64s - 1;
979}
980
981static void
982entry_init_size(struct jset_entry *entry, size_t size)
983{
984 unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
985 entry_init_u64s(entry, u64s);
986}
987
3ccc5c50
KO		 988struct jset_entry *
989bch2_journal_super_entries_add_common(struct bch_fs *c,
990 struct jset_entry *entry,
991 u64 journal_seq)
992{
3ccc5c50 993 unsigned i;
1c6fdbd8 994
5e82a9a1 995 percpu_down_read(&c->mark_lock);
3ccc5c50 996
5e82a9a1
KO		 997 if (!journal_seq) {
998 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
999 bch2_fs_usage_acc_to_base(c, i);
1000 } else {
f299d573 1001 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
5e82a9a1 1002 }
3ccc5c50
KO		 1003
1004 {
3577df5f
KO		 1005 struct jset_entry_usage *u =
1006 container_of(entry, struct jset_entry_usage, entry);
3ccc5c50 1007
43cfbad6 1008 entry_init_size(entry, sizeof(*u));
3ccc5c50 1009 u->entry.type = BCH_JSET_ENTRY_usage;
3577df5f 1010 u->entry.btree_id = FS_USAGE_INODES;
5e82a9a1 1011 u->v = cpu_to_le64(c->usage_base->nr_inodes);
3ccc5c50
KO		 1012
1013 entry = vstruct_next(entry);
1014 }
1015
1016 {
3577df5f
KO		 1017 struct jset_entry_usage *u =
1018 container_of(entry, struct jset_entry_usage, entry);
1019
43cfbad6 1020 entry_init_size(entry, sizeof(*u));
3ccc5c50 1021 u->entry.type = BCH_JSET_ENTRY_usage;
3577df5f
KO		 1022 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1023 u->v = cpu_to_le64(atomic64_read(&c->key_version));
1024
1025 entry = vstruct_next(entry);
1026 }
1027
1028 for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1029 struct jset_entry_usage *u =
1030 container_of(entry, struct jset_entry_usage, entry);
3577df5f 1031
43cfbad6 1032 entry_init_size(entry, sizeof(*u));
3577df5f
KO		 1033 u->entry.type = BCH_JSET_ENTRY_usage;
1034 u->entry.btree_id = FS_USAGE_RESERVED;
1035 u->entry.level = i;
5e82a9a1 1036 u->v = cpu_to_le64(c->usage_base->persistent_reserved[i]);
3ccc5c50
KO		 1037
1038 entry = vstruct_next(entry);
1039 }
1040
1041 for (i = 0; i < c->replicas.nr; i++) {
1042 struct bch_replicas_entry *e =
1043 cpu_replicas_entry(&c->replicas, i);
3577df5f
KO		 1044 struct jset_entry_data_usage *u =
1045 container_of(entry, struct jset_entry_data_usage, entry);
3ccc5c50 1046
43cfbad6 1047 entry_init_size(entry, sizeof(*u) + e->nr_devs);
3577df5f 1048 u->entry.type = BCH_JSET_ENTRY_data_usage;
5e82a9a1 1049 u->v = cpu_to_le64(c->usage_base->replicas[i]);
3ccc5c50
KO		 1050 unsafe_memcpy(&u->r, e, replicas_entry_bytes(e),
1051 "embedded variable length struct");
1052
1053 entry = vstruct_next(entry);
1054 }
1055
5e82a9a1 1056 percpu_up_read(&c->mark_lock);
3ccc5c50
KO		 1057
1058 return entry;
1059}
1060
134915f3 1061void bch2_fs_mark_clean(struct bch_fs *c)
3ccc5c50
KO		 1062{
1063 struct bch_sb_field_clean *sb_clean;
1064 struct jset_entry *entry;
1065 unsigned u64s;
1066
3ccc5c50
KO		 1067 mutex_lock(&c->sb_lock);
1068 if (BCH_SB_CLEAN(c->disk_sb.sb))
1069 goto out;
1070
1071 SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1072
1df42b57 1073 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
932aa837 1074 c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA;
5d548743 1075 c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_extents_above_btree_updates);
6357d607 1076 c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_btree_updates_journalled);
1df42b57 1077
3ccc5c50 1078 u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1c6fdbd8
KO		 1079
1080 sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1081 if (!sb_clean) {
1082 bch_err(c, "error resizing superblock while setting filesystem clean");
1083 goto out;
1084 }
1085
1086 sb_clean->flags = 0;
1087 sb_clean->read_clock = cpu_to_le16(c->bucket_clock[READ].hand);
1088 sb_clean->write_clock = cpu_to_le16(c->bucket_clock[WRITE].hand);
812af308
KO		 1089 sb_clean->journal_seq = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1090
1091 /* Trying to catch outstanding bug: */
1092 BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1c6fdbd8
KO		 1093
1094 entry = sb_clean->start;
3ccc5c50 1095 entry = bch2_journal_super_entries_add_common(c, entry, 0);
00b8ccf7 1096 entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
3ccc5c50
KO		 1097 BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1098
1c6fdbd8
KO		 1099 memset(entry, 0,
1100 vstruct_end(&sb_clean->field) - (void *) entry);
1101
26609b61
KO		 1102 if (le16_to_cpu(c->disk_sb.sb->version) <
1103 bcachefs_metadata_version_bkey_renumber)
1104 bch2_sb_clean_renumber(sb_clean, WRITE);
1105
1c6fdbd8
KO		 1106 bch2_write_super(c);
1107out:
1108 mutex_unlock(&c->sb_lock);
1109}
1110
1111static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1112 struct bch_sb_field *f)
1113{
1114 struct bch_sb_field_clean *clean = field_to_type(f, clean);
1115
1116 if (vstruct_bytes(&clean->field) < sizeof(*clean))
1117 return "invalid field crypt: wrong size";
1118
1119 return NULL;
1120}
1121
1122static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1123 .validate = bch2_sb_validate_clean,
1124};
1125
1126static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1127#define x(f, nr) \
1128 [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1129 BCH_SB_FIELDS()
1130#undef x
1131};
1132
1133static const char *bch2_sb_field_validate(struct bch_sb *sb,
1134 struct bch_sb_field *f)
1135{
1136 unsigned type = le32_to_cpu(f->type);
1137
1138 return type < BCH_SB_FIELD_NR
1139 ? bch2_sb_field_ops[type]->validate(sb, f)
1140 : NULL;
1141}
1142
319f9ac3
KO		 1143void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1144 struct bch_sb_field *f)
1c6fdbd8
KO		 1145{
1146 unsigned type = le32_to_cpu(f->type);
319f9ac3
KO		 1147 const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1148 ? bch2_sb_field_ops[type] : NULL;
1c6fdbd8 1149
319f9ac3
KO		 1150 if (ops)
1151 pr_buf(out, "%s", bch2_sb_fields[type]);
1152 else
1153 pr_buf(out, "(unknown field %u)", type);
1154
1155 pr_buf(out, " (size %llu):", vstruct_bytes(f));
1c6fdbd8 1156
319f9ac3
KO		 1157 if (ops && ops->to_text)
1158 bch2_sb_field_ops[type]->to_text(out, sb, f);
1c6fdbd8 1159}
Linux 6.x block layer and io_uring tree(s)