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