git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/super-io.c

   1
   2 #include "bcachefs.h"
   3 #include "checksum.h"
   4 #include "disk_groups.h"
   5 #include "ec.h"
   6 #include "error.h"
   7 #include "io.h"
   8 #include "journal.h"
   9 #include "journal_seq_blacklist.h"
  10 #include "replicas.h"
  11 #include "quota.h"
  12 #include "super-io.h"
  13 #include "super.h"
  14 #include "vstructs.h"
  15
  16 #include <linux/backing-dev.h>
  17 #include <linux/sort.h>
  18
  19 const char * const bch2_sb_fields[] = {
  20 #define x(name, nr)     #name,
  21         BCH_SB_FIELDS()
  22 #undef x
  23         NULL
  24 };
  25
  26 static const char *bch2_sb_field_validate(struct bch_sb *,
  27                                           struct bch_sb_field *);
  28
  29 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  30                                       enum bch_sb_field_type type)
  31 {
  32         struct bch_sb_field *f;
  33
  34         /* XXX: need locking around superblock to access optional fields */
  35
  36         vstruct_for_each(sb, f)
  37                 if (le32_to_cpu(f->type) == type)
  38                         return f;
  39         return NULL;
  40 }
  41
  42 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  43                                                    struct bch_sb_field *f,
  44                                                    unsigned u64s)
  45 {
  46         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  47         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  48
  49         BUG_ON(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
  50                sb->page_order);
  51
  52         if (!f) {
  53                 f = vstruct_last(sb->sb);
  54                 memset(f, 0, sizeof(u64) * u64s);
  55                 f->u64s = cpu_to_le32(u64s);
  56                 f->type = 0;
  57         } else {
  58                 void *src, *dst;
  59
  60                 src = vstruct_end(f);
  61
  62                 if (u64s) {
  63                         f->u64s = cpu_to_le32(u64s);
  64                         dst = vstruct_end(f);
  65                 } else {
  66                         dst = f;
  67                 }
  68
  69                 memmove(dst, src, vstruct_end(sb->sb) - src);
  70
  71                 if (dst > src)
  72                         memset(src, 0, dst - src);
  73         }
  74
  75         sb->sb->u64s = cpu_to_le32(sb_u64s);
  76
  77         return u64s ? f : NULL;
  78 }
  79
  80 void bch2_sb_field_delete(struct bch_sb_handle *sb,
  81                           enum bch_sb_field_type type)
  82 {
  83         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
  84
  85         if (f)
  86                 __bch2_sb_field_resize(sb, f, 0);
  87 }
  88
  89 /* Superblock realloc/free: */
  90
  91 void bch2_free_super(struct bch_sb_handle *sb)
  92 {
  93         if (sb->bio)
  94                 bio_put(sb->bio);
  95         if (!IS_ERR_OR_NULL(sb->bdev))
  96                 blkdev_put(sb->bdev, sb->mode);
  97
  98         free_pages((unsigned long) sb->sb, sb->page_order);
  99         memset(sb, 0, sizeof(*sb));
 100 }
 101
 102 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 103 {
 104         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 105         unsigned order = get_order(new_bytes);
 106         struct bch_sb *new_sb;
 107         struct bio *bio;
 108
 109         if (sb->sb && sb->page_order >= order)
 110                 return 0;
 111
 112         if (sb->have_layout) {
 113                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 114
 115                 if (new_bytes > max_bytes) {
 116                         char buf[BDEVNAME_SIZE];
 117
 118                         pr_err("%s: superblock too big: want %zu but have %llu",
 119                                bdevname(sb->bdev, buf), new_bytes, max_bytes);
 120                         return -ENOSPC;
 121                 }
 122         }
 123
 124         if (sb->page_order >= order && sb->sb)
 125                 return 0;
 126
 127         if (dynamic_fault("bcachefs:add:super_realloc"))
 128                 return -ENOMEM;
 129
 130         if (sb->have_bio) {
 131                 bio = bio_kmalloc(GFP_KERNEL, 1 << order);
 132                 if (!bio)
 133                         return -ENOMEM;
 134
 135                 if (sb->bio)
 136                         bio_put(sb->bio);
 137                 sb->bio = bio;
 138         }
 139
 140         new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
 141         if (!new_sb)
 142                 return -ENOMEM;
 143
 144         if (sb->sb)
 145                 memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
 146
 147         free_pages((unsigned long) sb->sb, sb->page_order);
 148         sb->sb = new_sb;
 149
 150         sb->page_order = order;
 151
 152         return 0;
 153 }
 154
 155 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 156                                           enum bch_sb_field_type type,
 157                                           unsigned u64s)
 158 {
 159         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 160         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 161         ssize_t d = -old_u64s + u64s;
 162
 163         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 164                 return NULL;
 165
 166         if (sb->fs_sb) {
 167                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 168                 struct bch_dev *ca;
 169                 unsigned i;
 170
 171                 lockdep_assert_held(&c->sb_lock);
 172
 173                 /* XXX: we're not checking that offline device have enough space */
 174
 175                 for_each_online_member(ca, c, i) {
 176                         struct bch_sb_handle *sb = &ca->disk_sb;
 177
 178                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 179                                 percpu_ref_put(&ca->ref);
 180                                 return NULL;
 181                         }
 182                 }
 183         }
 184
 185         f = bch2_sb_field_get(sb->sb, type);
 186         f = __bch2_sb_field_resize(sb, f, u64s);
 187         if (f)
 188                 f->type = cpu_to_le32(type);
 189         return f;
 190 }
 191
 192 /* Superblock validate: */
 193
 194 static inline void __bch2_sb_layout_size_assert(void)
 195 {
 196         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 197 }
 198
 199 static const char *validate_sb_layout(struct bch_sb_layout *layout)
 200 {
 201         u64 offset, prev_offset, max_sectors;
 202         unsigned i;
 203
 204         if (uuid_le_cmp(layout->magic, BCACHE_MAGIC))
 205                 return "Not a bcachefs superblock layout";
 206
 207         if (layout->layout_type != 0)
 208                 return "Invalid superblock layout type";
 209
 210         if (!layout->nr_superblocks)
 211                 return "Invalid superblock layout: no superblocks";
 212
 213         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
 214                 return "Invalid superblock layout: too many superblocks";
 215
 216         max_sectors = 1 << layout->sb_max_size_bits;
 217
 218         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 219
 220         for (i = 1; i < layout->nr_superblocks; i++) {
 221                 offset = le64_to_cpu(layout->sb_offset[i]);
 222
 223                 if (offset < prev_offset + max_sectors)
 224                         return "Invalid superblock layout: superblocks overlap";
 225                 prev_offset = offset;
 226         }
 227
 228         return NULL;
 229 }
 230
 231 const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
 232 {
 233         struct bch_sb *sb = disk_sb->sb;
 234         struct bch_sb_field *f;
 235         struct bch_sb_field_members *mi;
 236         const char *err;
 237         u32 version, version_min;
 238         u16 block_size;
 239
 240         version         = le16_to_cpu(sb->version);
 241         version_min     = version >= bcachefs_metadata_version_new_versioning
 242                 ? le16_to_cpu(sb->version_min)
 243                 : version;
 244
 245         if (version    >= bcachefs_metadata_version_max ||
 246             version_min < bcachefs_metadata_version_min)
 247                 return "Unsupported superblock version";
 248
 249         if (version_min > version)
 250                 return "Bad minimum version";
 251
 252         if (sb->features[1] ||
 253             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
 254                 return "Filesystem has incompatible features";
 255
 256         block_size = le16_to_cpu(sb->block_size);
 257
 258         if (!is_power_of_2(block_size) ||
 259             block_size > PAGE_SECTORS)
 260                 return "Bad block size";
 261
 262         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
 263                 return "Bad user UUID";
 264
 265         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
 266                 return "Bad internal UUID";
 267
 268         if (!sb->nr_devices ||
 269             sb->nr_devices <= sb->dev_idx ||
 270             sb->nr_devices > BCH_SB_MEMBERS_MAX)
 271                 return "Bad number of member devices";
 272
 273         if (!BCH_SB_META_REPLICAS_WANT(sb) ||
 274             BCH_SB_META_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
 275                 return "Invalid number of metadata replicas";
 276
 277         if (!BCH_SB_META_REPLICAS_REQ(sb) ||
 278             BCH_SB_META_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
 279                 return "Invalid number of metadata replicas";
 280
 281         if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
 282             BCH_SB_DATA_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
 283                 return "Invalid number of data replicas";
 284
 285         if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
 286             BCH_SB_DATA_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
 287                 return "Invalid number of data replicas";
 288
 289         if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 290                 return "Invalid metadata checksum type";
 291
 292         if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 293                 return "Invalid metadata checksum type";
 294
 295         if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
 296                 return "Invalid compression type";
 297
 298         if (!BCH_SB_BTREE_NODE_SIZE(sb))
 299                 return "Btree node size not set";
 300
 301         if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
 302                 return "Btree node size not a power of two";
 303
 304         if (BCH_SB_GC_RESERVE(sb) < 5)
 305                 return "gc reserve percentage too small";
 306
 307         if (!sb->time_precision ||
 308             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
 309                 return "invalid time precision";
 310
 311         /* validate layout */
 312         err = validate_sb_layout(&sb->layout);
 313         if (err)
 314                 return err;
 315
 316         vstruct_for_each(sb, f) {
 317                 if (!f->u64s)
 318                         return "Invalid superblock: invalid optional field";
 319
 320                 if (vstruct_next(f) > vstruct_last(sb))
 321                         return "Invalid superblock: invalid optional field";
 322         }
 323
 324         /* members must be validated first: */
 325         mi = bch2_sb_get_members(sb);
 326         if (!mi)
 327                 return "Invalid superblock: member info area missing";
 328
 329         err = bch2_sb_field_validate(sb, &mi->field);
 330         if (err)
 331                 return err;
 332
 333         vstruct_for_each(sb, f) {
 334                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 335                         continue;
 336
 337                 err = bch2_sb_field_validate(sb, f);
 338                 if (err)
 339                         return err;
 340         }
 341
 342         return NULL;
 343 }
 344
 345 /* device open: */
 346
 347 static void bch2_sb_update(struct bch_fs *c)
 348 {
 349         struct bch_sb *src = c->disk_sb.sb;
 350         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 351         struct bch_dev *ca;
 352         unsigned i;
 353
 354         lockdep_assert_held(&c->sb_lock);
 355
 356         c->sb.uuid              = src->uuid;
 357         c->sb.user_uuid         = src->user_uuid;
 358         c->sb.version           = le16_to_cpu(src->version);
 359         c->sb.nr_devices        = src->nr_devices;
 360         c->sb.clean             = BCH_SB_CLEAN(src);
 361         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 362         c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
 363         c->sb.time_base_lo      = le64_to_cpu(src->time_base_lo);
 364         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 365         c->sb.time_precision    = le32_to_cpu(src->time_precision);
 366         c->sb.features          = le64_to_cpu(src->features[0]);
 367         c->sb.compat            = le64_to_cpu(src->compat[0]);
 368
 369         for_each_member_device(ca, c, i)
 370                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 371 }
 372
 373 /* doesn't copy member info */
 374 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 375 {
 376         struct bch_sb_field *src_f, *dst_f;
 377         struct bch_sb *dst = dst_handle->sb;
 378         unsigned i;
 379
 380         dst->version            = src->version;
 381         dst->version_min        = src->version_min;
 382         dst->seq                = src->seq;
 383         dst->uuid               = src->uuid;
 384         dst->user_uuid          = src->user_uuid;
 385         memcpy(dst->label,      src->label, sizeof(dst->label));
 386
 387         dst->block_size         = src->block_size;
 388         dst->nr_devices         = src->nr_devices;
 389
 390         dst->time_base_lo       = src->time_base_lo;
 391         dst->time_base_hi       = src->time_base_hi;
 392         dst->time_precision     = src->time_precision;
 393
 394         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 395         memcpy(dst->features,   src->features,  sizeof(dst->features));
 396         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 397
 398         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 399                 if (i == BCH_SB_FIELD_journal)
 400                         continue;
 401
 402                 src_f = bch2_sb_field_get(src, i);
 403                 dst_f = bch2_sb_field_get(dst, i);
 404                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 405                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 406
 407                 if (src_f)
 408                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 409         }
 410 }
 411
 412 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 413 {
 414         struct bch_sb_field_journal *journal_buckets =
 415                 bch2_sb_get_journal(src);
 416         unsigned journal_u64s = journal_buckets
 417                 ? le32_to_cpu(journal_buckets->field.u64s)
 418                 : 0;
 419         int ret;
 420
 421         lockdep_assert_held(&c->sb_lock);
 422
 423         ret = bch2_sb_realloc(&c->disk_sb,
 424                               le32_to_cpu(src->u64s) - journal_u64s);
 425         if (ret)
 426                 return ret;
 427
 428         __copy_super(&c->disk_sb, src);
 429
 430         ret = bch2_sb_replicas_to_cpu_replicas(c);
 431         if (ret)
 432                 return ret;
 433
 434         ret = bch2_sb_disk_groups_to_cpu(c);
 435         if (ret)
 436                 return ret;
 437
 438         bch2_sb_update(c);
 439         return 0;
 440 }
 441
 442 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 443 {
 444         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 445         struct bch_sb_field_journal *journal_buckets =
 446                 bch2_sb_get_journal(dst);
 447         unsigned journal_u64s = journal_buckets
 448                 ? le32_to_cpu(journal_buckets->field.u64s)
 449                 : 0;
 450         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 451         int ret;
 452
 453         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 454         if (ret)
 455                 return ret;
 456
 457         __copy_super(&ca->disk_sb, src);
 458         return 0;
 459 }
 460
 461 /* read superblock: */
 462
 463 static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
 464 {
 465         struct bch_csum csum;
 466         size_t bytes;
 467 reread:
 468         bio_reset(sb->bio);
 469         bio_set_dev(sb->bio, sb->bdev);
 470         sb->bio->bi_iter.bi_sector = offset;
 471         sb->bio->bi_iter.bi_size = PAGE_SIZE << sb->page_order;
 472         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 473         bch2_bio_map(sb->bio, sb->sb);
 474
 475         if (submit_bio_wait(sb->bio))
 476                 return "IO error";
 477
 478         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
 479                 return "Not a bcachefs superblock";
 480
 481         if (le16_to_cpu(sb->sb->version) <  bcachefs_metadata_version_min ||
 482             le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
 483                 return "Unsupported superblock version";
 484
 485         bytes = vstruct_bytes(sb->sb);
 486
 487         if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
 488                 return "Bad superblock: too big";
 489
 490         if (get_order(bytes) > sb->page_order) {
 491                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 492                         return "cannot allocate memory";
 493                 goto reread;
 494         }
 495
 496         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
 497                 return "unknown csum type";
 498
 499         /* XXX: verify MACs */
 500         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 501                             null_nonce(), sb->sb);
 502
 503         if (bch2_crc_cmp(csum, sb->sb->csum))
 504                 return "bad checksum reading superblock";
 505
 506         sb->seq = le64_to_cpu(sb->sb->seq);
 507
 508         return NULL;
 509 }
 510
 511 int bch2_read_super(const char *path, struct bch_opts *opts,
 512                     struct bch_sb_handle *sb)
 513 {
 514         u64 offset = opt_get(*opts, sb);
 515         struct bch_sb_layout layout;
 516         const char *err;
 517         __le64 *i;
 518         int ret;
 519
 520         pr_verbose_init(*opts, "");
 521
 522         memset(sb, 0, sizeof(*sb));
 523         sb->mode        = FMODE_READ;
 524         sb->have_bio    = true;
 525
 526         if (!opt_get(*opts, noexcl))
 527                 sb->mode |= FMODE_EXCL;
 528
 529         if (!opt_get(*opts, nochanges))
 530                 sb->mode |= FMODE_WRITE;
 531
 532         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 533         if (IS_ERR(sb->bdev) &&
 534             PTR_ERR(sb->bdev) == -EACCES &&
 535             opt_get(*opts, read_only)) {
 536                 sb->mode &= ~FMODE_WRITE;
 537
 538                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 539                 if (!IS_ERR(sb->bdev))
 540                         opt_set(*opts, nochanges, true);
 541         }
 542
 543         if (IS_ERR(sb->bdev)) {
 544                 ret = PTR_ERR(sb->bdev);
 545                 goto out;
 546         }
 547
 548         err = "cannot allocate memory";
 549         ret = bch2_sb_realloc(sb, 0);
 550         if (ret)
 551                 goto err;
 552
 553         ret = -EFAULT;
 554         err = "dynamic fault";
 555         if (bch2_fs_init_fault("read_super"))
 556                 goto err;
 557
 558         ret = -EINVAL;
 559         err = read_one_super(sb, offset);
 560         if (!err)
 561                 goto got_super;
 562
 563         if (opt_defined(*opts, sb))
 564                 goto err;
 565
 566         pr_err("error reading default superblock: %s", err);
 567
 568         /*
 569          * Error reading primary superblock - read location of backup
 570          * superblocks:
 571          */
 572         bio_reset(sb->bio);
 573         bio_set_dev(sb->bio, sb->bdev);
 574         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 575         sb->bio->bi_iter.bi_size = sizeof(struct bch_sb_layout);
 576         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 577         /*
 578          * use sb buffer to read layout, since sb buffer is page aligned but
 579          * layout won't be:
 580          */
 581         bch2_bio_map(sb->bio, sb->sb);
 582
 583         err = "IO error";
 584         if (submit_bio_wait(sb->bio))
 585                 goto err;
 586
 587         memcpy(&layout, sb->sb, sizeof(layout));
 588         err = validate_sb_layout(&layout);
 589         if (err)
 590                 goto err;
 591
 592         for (i = layout.sb_offset;
 593              i < layout.sb_offset + layout.nr_superblocks; i++) {
 594                 offset = le64_to_cpu(*i);
 595
 596                 if (offset == opt_get(*opts, sb))
 597                         continue;
 598
 599                 err = read_one_super(sb, offset);
 600                 if (!err)
 601                         goto got_super;
 602         }
 603
 604         ret = -EINVAL;
 605         goto err;
 606
 607 got_super:
 608         err = "Superblock block size smaller than device block size";
 609         ret = -EINVAL;
 610         if (le16_to_cpu(sb->sb->block_size) << 9 <
 611             bdev_logical_block_size(sb->bdev))
 612                 goto err;
 613
 614         if (sb->mode & FMODE_WRITE)
 615                 bdev_get_queue(sb->bdev)->backing_dev_info->capabilities
 616                         |= BDI_CAP_STABLE_WRITES;
 617         ret = 0;
 618         sb->have_layout = true;
 619 out:
 620         pr_verbose_init(*opts, "ret %i", ret);
 621         return ret;
 622 err:
 623         bch2_free_super(sb);
 624         pr_err("error reading superblock: %s", err);
 625         goto out;
 626 }
 627
 628 /* write superblock: */
 629
 630 static void write_super_endio(struct bio *bio)
 631 {
 632         struct bch_dev *ca = bio->bi_private;
 633
 634         /* XXX: return errors directly */
 635
 636         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write"))
 637                 ca->sb_write_error = 1;
 638
 639         closure_put(&ca->fs->sb_write);
 640         percpu_ref_put(&ca->io_ref);
 641 }
 642
 643 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 644 {
 645         struct bch_sb *sb = ca->disk_sb.sb;
 646         struct bio *bio = ca->disk_sb.bio;
 647
 648         bio_reset(bio);
 649         bio_set_dev(bio, ca->disk_sb.bdev);
 650         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 651         bio->bi_iter.bi_size    = 4096;
 652         bio->bi_end_io          = write_super_endio;
 653         bio->bi_private         = ca;
 654         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 655         bch2_bio_map(bio, ca->sb_read_scratch);
 656
 657         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_SB],
 658                      bio_sectors(bio));
 659
 660         percpu_ref_get(&ca->io_ref);
 661         closure_bio_submit(bio, &c->sb_write);
 662 }
 663
 664 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 665 {
 666         struct bch_sb *sb = ca->disk_sb.sb;
 667         struct bio *bio = ca->disk_sb.bio;
 668
 669         sb->offset = sb->layout.sb_offset[idx];
 670
 671         SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
 672         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 673                                 null_nonce(), sb);
 674
 675         bio_reset(bio);
 676         bio_set_dev(bio, ca->disk_sb.bdev);
 677         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 678         bio->bi_iter.bi_size    =
 679                 roundup((size_t) vstruct_bytes(sb),
 680                         bdev_logical_block_size(ca->disk_sb.bdev));
 681         bio->bi_end_io          = write_super_endio;
 682         bio->bi_private         = ca;
 683         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 684         bch2_bio_map(bio, sb);
 685
 686         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_SB],
 687                      bio_sectors(bio));
 688
 689         percpu_ref_get(&ca->io_ref);
 690         closure_bio_submit(bio, &c->sb_write);
 691 }
 692
 693 int bch2_write_super(struct bch_fs *c)
 694 {
 695         struct closure *cl = &c->sb_write;
 696         struct bch_dev *ca;
 697         unsigned i, sb = 0, nr_wrote;
 698         const char *err;
 699         struct bch_devs_mask sb_written;
 700         bool wrote, can_mount_without_written, can_mount_with_written;
 701         int ret = 0;
 702
 703         lockdep_assert_held(&c->sb_lock);
 704
 705         closure_init_stack(cl);
 706         memset(&sb_written, 0, sizeof(sb_written));
 707
 708         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 709
 710         if (test_bit(BCH_FS_ERROR, &c->flags))
 711                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 712
 713         for_each_online_member(ca, c, i)
 714                 bch2_sb_from_fs(c, ca);
 715
 716         for_each_online_member(ca, c, i) {
 717                 err = bch2_sb_validate(&ca->disk_sb);
 718                 if (err) {
 719                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
 720                         ret = -1;
 721                         goto out;
 722                 }
 723         }
 724
 725         if (c->opts.nochanges)
 726                 goto out;
 727
 728         for_each_online_member(ca, c, i) {
 729                 __set_bit(ca->dev_idx, sb_written.d);
 730                 ca->sb_write_error = 0;
 731         }
 732
 733         for_each_online_member(ca, c, i)
 734                 read_back_super(c, ca);
 735         closure_sync(cl);
 736
 737         for_each_online_member(ca, c, i) {
 738                 if (!ca->sb_write_error &&
 739                     ca->disk_sb.seq !=
 740                     le64_to_cpu(ca->sb_read_scratch->seq)) {
 741                         bch2_fs_fatal_error(c,
 742                                 "Superblock modified by another process");
 743                         percpu_ref_put(&ca->io_ref);
 744                         ret = -EROFS;
 745                         goto out;
 746                 }
 747         }
 748
 749         do {
 750                 wrote = false;
 751                 for_each_online_member(ca, c, i)
 752                         if (!ca->sb_write_error &&
 753                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 754                                 write_one_super(c, ca, sb);
 755                                 wrote = true;
 756                         }
 757                 closure_sync(cl);
 758                 sb++;
 759         } while (wrote);
 760
 761         for_each_online_member(ca, c, i) {
 762                 if (ca->sb_write_error)
 763                         __clear_bit(ca->dev_idx, sb_written.d);
 764                 else
 765                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 766         }
 767
 768         nr_wrote = dev_mask_nr(&sb_written);
 769
 770         can_mount_with_written =
 771                 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
 772                                       BCH_FORCE_IF_DEGRADED);
 773
 774         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 775                 sb_written.d[i] = ~sb_written.d[i];
 776
 777         can_mount_without_written =
 778                 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
 779                                       BCH_FORCE_IF_DEGRADED);
 780
 781         /*
 782          * If we would be able to mount _without_ the devices we successfully
 783          * wrote superblocks to, we weren't able to write to enough devices:
 784          *
 785          * Exception: if we can mount without the successes because we haven't
 786          * written anything (new filesystem), we continue if we'd be able to
 787          * mount with the devices we did successfully write to:
 788          */
 789         if (bch2_fs_fatal_err_on(!nr_wrote ||
 790                                  (can_mount_without_written &&
 791                                   !can_mount_with_written), c,
 792                 "Unable to write superblock to sufficient devices"))
 793                 ret = -1;
 794 out:
 795         /* Make new options visible after they're persistent: */
 796         bch2_sb_update(c);
 797         return ret;
 798 }
 799
 800 /* BCH_SB_FIELD_journal: */
 801
 802 static int u64_cmp(const void *_l, const void *_r)
 803 {
 804         u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
 805
 806         return l < r ? -1 : l > r ? 1 : 0;
 807 }
 808
 809 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
 810                                             struct bch_sb_field *f)
 811 {
 812         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 813         struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
 814         const char *err;
 815         unsigned nr;
 816         unsigned i;
 817         u64 *b;
 818
 819         journal = bch2_sb_get_journal(sb);
 820         if (!journal)
 821                 return NULL;
 822
 823         nr = bch2_nr_journal_buckets(journal);
 824         if (!nr)
 825                 return NULL;
 826
 827         b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
 828         if (!b)
 829                 return "cannot allocate memory";
 830
 831         for (i = 0; i < nr; i++)
 832                 b[i] = le64_to_cpu(journal->buckets[i]);
 833
 834         sort(b, nr, sizeof(u64), u64_cmp, NULL);
 835
 836         err = "journal bucket at sector 0";
 837         if (!b[0])
 838                 goto err;
 839
 840         err = "journal bucket before first bucket";
 841         if (m && b[0] < le16_to_cpu(m->first_bucket))
 842                 goto err;
 843
 844         err = "journal bucket past end of device";
 845         if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
 846                 goto err;
 847
 848         err = "duplicate journal buckets";
 849         for (i = 0; i + 1 < nr; i++)
 850                 if (b[i] == b[i + 1])
 851                         goto err;
 852
 853         err = NULL;
 854 err:
 855         kfree(b);
 856         return err;
 857 }
 858
 859 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
 860         .validate       = bch2_sb_validate_journal,
 861 };
 862
 863 /* BCH_SB_FIELD_members: */
 864
 865 static const char *bch2_sb_validate_members(struct bch_sb *sb,
 866                                             struct bch_sb_field *f)
 867 {
 868         struct bch_sb_field_members *mi = field_to_type(f, members);
 869         struct bch_member *m;
 870
 871         if ((void *) (mi->members + sb->nr_devices) >
 872             vstruct_end(&mi->field))
 873                 return "Invalid superblock: bad member info";
 874
 875         for (m = mi->members;
 876              m < mi->members + sb->nr_devices;
 877              m++) {
 878                 if (!bch2_member_exists(m))
 879                         continue;
 880
 881                 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
 882                         return "Too many buckets";
 883
 884                 if (le64_to_cpu(m->nbuckets) -
 885                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
 886                         return "Not enough buckets";
 887
 888                 if (le16_to_cpu(m->bucket_size) <
 889                     le16_to_cpu(sb->block_size))
 890                         return "bucket size smaller than block size";
 891
 892                 if (le16_to_cpu(m->bucket_size) <
 893                     BCH_SB_BTREE_NODE_SIZE(sb))
 894                         return "bucket size smaller than btree node size";
 895         }
 896
 897         return NULL;
 898 }
 899
 900 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
 901         .validate       = bch2_sb_validate_members,
 902 };
 903
 904 /* BCH_SB_FIELD_crypt: */
 905
 906 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
 907                                           struct bch_sb_field *f)
 908 {
 909         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
 910
 911         if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
 912                 return "invalid field crypt: wrong size";
 913
 914         if (BCH_CRYPT_KDF_TYPE(crypt))
 915                 return "invalid field crypt: bad kdf type";
 916
 917         return NULL;
 918 }
 919
 920 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
 921         .validate       = bch2_sb_validate_crypt,
 922 };
 923
 924 /* BCH_SB_FIELD_clean: */
 925
 926 void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
 927 {
 928         struct jset_entry *entry;
 929
 930         for (entry = clean->start;
 931              entry < (struct jset_entry *) vstruct_end(&clean->field);
 932              entry = vstruct_next(entry))
 933                 bch2_bkey_renumber(BKEY_TYPE_BTREE, bkey_to_packed(entry->start), write);
 934 }
 935
 936 int bch2_fs_mark_dirty(struct bch_fs *c)
 937 {
 938         int ret;
 939
 940         /*
 941          * Unconditionally write superblock, to verify it hasn't changed before
 942          * we go rw:
 943          */
 944
 945         mutex_lock(&c->sb_lock);
 946         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 947         c->disk_sb.sb->compat[0] &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 948         ret = bch2_write_super(c);
 949         mutex_unlock(&c->sb_lock);
 950
 951         return ret;
 952 }
 953
 954 struct jset_entry *
 955 bch2_journal_super_entries_add_common(struct bch_fs *c,
 956                                       struct jset_entry *entry)
 957 {
 958         struct btree_root *r;
 959         unsigned i;
 960
 961         mutex_lock(&c->btree_root_lock);
 962
 963         for (r = c->btree_roots;
 964              r < c->btree_roots + BTREE_ID_NR;
 965              r++)
 966                 if (r->alive) {
 967                         entry->u64s     = r->key.u64s;
 968                         entry->btree_id = r - c->btree_roots;
 969                         entry->level    = r->level;
 970                         entry->type     = BCH_JSET_ENTRY_btree_root;
 971                         bkey_copy(&entry->start[0], &r->key);
 972
 973                         entry = vstruct_next(entry);
 974                 }
 975         c->btree_roots_dirty = false;
 976
 977         mutex_unlock(&c->btree_root_lock);
 978
 979         percpu_down_read_preempt_disable(&c->mark_lock);
 980
 981         {
 982                 u64 nr_inodes = percpu_u64_get(&c->usage[0]->nr_inodes);
 983                 struct jset_entry_usage *u =
 984                         container_of(entry, struct jset_entry_usage, entry);
 985
 986                 memset(u, 0, sizeof(*u));
 987                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
 988                 u->entry.type   = BCH_JSET_ENTRY_usage;
 989                 u->entry.btree_id = FS_USAGE_INODES;
 990                 u->v            = cpu_to_le64(nr_inodes);
 991
 992                 entry = vstruct_next(entry);
 993         }
 994
 995         {
 996                 struct jset_entry_usage *u =
 997                         container_of(entry, struct jset_entry_usage, entry);
 998
 999                 memset(u, 0, sizeof(*u));
1000                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
1001                 u->entry.type   = BCH_JSET_ENTRY_usage;
1002                 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1003                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1004
1005                 entry = vstruct_next(entry);
1006         }
1007
1008         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1009                 struct jset_entry_usage *u =
1010                         container_of(entry, struct jset_entry_usage, entry);
1011                 u64 sectors = percpu_u64_get(&c->usage[0]->persistent_reserved[i]);
1012
1013                 if (!sectors)
1014                         continue;
1015
1016                 memset(u, 0, sizeof(*u));
1017                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
1018                 u->entry.type   = BCH_JSET_ENTRY_usage;
1019                 u->entry.btree_id = FS_USAGE_RESERVED;
1020                 u->entry.level  = i;
1021                 u->v            = sectors;
1022
1023                 entry = vstruct_next(entry);
1024         }
1025
1026         for (i = 0; i < c->replicas.nr; i++) {
1027                 struct bch_replicas_entry *e =
1028                         cpu_replicas_entry(&c->replicas, i);
1029                 u64 sectors = percpu_u64_get(&c->usage[0]->replicas[i]);
1030                 struct jset_entry_data_usage *u =
1031                         container_of(entry, struct jset_entry_data_usage, entry);
1032
1033                 memset(u, 0, sizeof(*u));
1034                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u) + e->nr_devs,
1035                                                sizeof(u64)) - 1;
1036                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1037                 u->v            = cpu_to_le64(sectors);
1038                 memcpy(&u->r, e, replicas_entry_bytes(e));
1039
1040                 entry = vstruct_next(entry);
1041         }
1042
1043         percpu_up_read_preempt_enable(&c->mark_lock);
1044
1045         return entry;
1046 }
1047
1048 void bch2_fs_mark_clean(struct bch_fs *c)
1049 {
1050         struct bch_sb_field_clean *sb_clean;
1051         struct jset_entry *entry;
1052         unsigned u64s;
1053
1054         mutex_lock(&c->sb_lock);
1055         if (BCH_SB_CLEAN(c->disk_sb.sb))
1056                 goto out;
1057
1058         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1059
1060         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1061
1062         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1063
1064         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1065         if (!sb_clean) {
1066                 bch_err(c, "error resizing superblock while setting filesystem clean");
1067                 goto out;
1068         }
1069
1070         sb_clean->flags         = 0;
1071         sb_clean->read_clock    = cpu_to_le16(c->bucket_clock[READ].hand);
1072         sb_clean->write_clock   = cpu_to_le16(c->bucket_clock[WRITE].hand);
1073         sb_clean->journal_seq   = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1074
1075         /* Trying to catch outstanding bug: */
1076         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1077
1078         entry = sb_clean->start;
1079         entry = bch2_journal_super_entries_add_common(c, entry);
1080         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1081
1082         memset(entry, 0,
1083                vstruct_end(&sb_clean->field) - (void *) entry);
1084
1085         if (le16_to_cpu(c->disk_sb.sb->version) <
1086             bcachefs_metadata_version_bkey_renumber)
1087                 bch2_sb_clean_renumber(sb_clean, WRITE);
1088
1089         bch2_write_super(c);
1090 out:
1091         mutex_unlock(&c->sb_lock);
1092 }
1093
1094 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1095                                           struct bch_sb_field *f)
1096 {
1097         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1098
1099         if (vstruct_bytes(&clean->field) < sizeof(*clean))
1100                 return "invalid field crypt: wrong size";
1101
1102         return NULL;
1103 }
1104
1105 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1106         .validate       = bch2_sb_validate_clean,
1107 };
1108
1109 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1110 #define x(f, nr)                                        \
1111         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1112         BCH_SB_FIELDS()
1113 #undef x
1114 };
1115
1116 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1117                                           struct bch_sb_field *f)
1118 {
1119         unsigned type = le32_to_cpu(f->type);
1120
1121         return type < BCH_SB_FIELD_NR
1122                 ? bch2_sb_field_ops[type]->validate(sb, f)
1123                 : NULL;
1124 }
1125
1126 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1127                            struct bch_sb_field *f)
1128 {
1129         unsigned type = le32_to_cpu(f->type);
1130         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1131                 ? bch2_sb_field_ops[type] : NULL;
1132
1133         if (ops)
1134                 pr_buf(out, "%s", bch2_sb_fields[type]);
1135         else
1136                 pr_buf(out, "(unknown field %u)", type);
1137
1138         pr_buf(out, " (size %llu):", vstruct_bytes(f));
1139
1140         if (ops && ops->to_text)
1141                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1142 }