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_NOFS|__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         sb->seq = le64_to_cpu(sb->sb->seq);
 506
 507         return NULL;
 508 }
 509
 510 int bch2_read_super(const char *path, struct bch_opts *opts,
 511                     struct bch_sb_handle *sb)
 512 {
 513         u64 offset = opt_get(*opts, sb);
 514         struct bch_sb_layout layout;
 515         const char *err;
 516         __le64 *i;
 517         int ret;
 518
 519         pr_verbose_init(*opts, "");
 520
 521         memset(sb, 0, sizeof(*sb));
 522         sb->mode        = FMODE_READ;
 523         sb->have_bio    = true;
 524
 525         if (!opt_get(*opts, noexcl))
 526                 sb->mode |= FMODE_EXCL;
 527
 528         if (!opt_get(*opts, nochanges))
 529                 sb->mode |= FMODE_WRITE;
 530
 531         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 532         if (IS_ERR(sb->bdev) &&
 533             PTR_ERR(sb->bdev) == -EACCES &&
 534             opt_get(*opts, read_only)) {
 535                 sb->mode &= ~FMODE_WRITE;
 536
 537                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 538                 if (!IS_ERR(sb->bdev))
 539                         opt_set(*opts, nochanges, true);
 540         }
 541
 542         if (IS_ERR(sb->bdev)) {
 543                 ret = PTR_ERR(sb->bdev);
 544                 goto out;
 545         }
 546
 547         err = "cannot allocate memory";
 548         ret = bch2_sb_realloc(sb, 0);
 549         if (ret)
 550                 goto err;
 551
 552         ret = -EFAULT;
 553         err = "dynamic fault";
 554         if (bch2_fs_init_fault("read_super"))
 555                 goto err;
 556
 557         ret = -EINVAL;
 558         err = read_one_super(sb, offset);
 559         if (!err)
 560                 goto got_super;
 561
 562         if (opt_defined(*opts, sb))
 563                 goto err;
 564
 565         pr_err("error reading default superblock: %s", err);
 566
 567         /*
 568          * Error reading primary superblock - read location of backup
 569          * superblocks:
 570          */
 571         bio_reset(sb->bio);
 572         bio_set_dev(sb->bio, sb->bdev);
 573         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 574         sb->bio->bi_iter.bi_size = sizeof(struct bch_sb_layout);
 575         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 576         /*
 577          * use sb buffer to read layout, since sb buffer is page aligned but
 578          * layout won't be:
 579          */
 580         bch2_bio_map(sb->bio, sb->sb);
 581
 582         err = "IO error";
 583         if (submit_bio_wait(sb->bio))
 584                 goto err;
 585
 586         memcpy(&layout, sb->sb, sizeof(layout));
 587         err = validate_sb_layout(&layout);
 588         if (err)
 589                 goto err;
 590
 591         for (i = layout.sb_offset;
 592              i < layout.sb_offset + layout.nr_superblocks; i++) {
 593                 offset = le64_to_cpu(*i);
 594
 595                 if (offset == opt_get(*opts, sb))
 596                         continue;
 597
 598                 err = read_one_super(sb, offset);
 599                 if (!err)
 600                         goto got_super;
 601         }
 602
 603         ret = -EINVAL;
 604         goto err;
 605
 606 got_super:
 607         err = "Superblock block size smaller than device block size";
 608         ret = -EINVAL;
 609         if (le16_to_cpu(sb->sb->block_size) << 9 <
 610             bdev_logical_block_size(sb->bdev))
 611                 goto err;
 612
 613         if (sb->mode & FMODE_WRITE)
 614                 bdev_get_queue(sb->bdev)->backing_dev_info->capabilities
 615                         |= BDI_CAP_STABLE_WRITES;
 616         ret = 0;
 617         sb->have_layout = true;
 618 out:
 619         pr_verbose_init(*opts, "ret %i", ret);
 620         return ret;
 621 err:
 622         bch2_free_super(sb);
 623         pr_err("error reading superblock: %s", err);
 624         goto out;
 625 }
 626
 627 /* write superblock: */
 628
 629 static void write_super_endio(struct bio *bio)
 630 {
 631         struct bch_dev *ca = bio->bi_private;
 632
 633         /* XXX: return errors directly */
 634
 635         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write"))
 636                 ca->sb_write_error = 1;
 637
 638         closure_put(&ca->fs->sb_write);
 639         percpu_ref_put(&ca->io_ref);
 640 }
 641
 642 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 643 {
 644         struct bch_sb *sb = ca->disk_sb.sb;
 645         struct bio *bio = ca->disk_sb.bio;
 646
 647         bio_reset(bio);
 648         bio_set_dev(bio, ca->disk_sb.bdev);
 649         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 650         bio->bi_iter.bi_size    = 4096;
 651         bio->bi_end_io          = write_super_endio;
 652         bio->bi_private         = ca;
 653         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 654         bch2_bio_map(bio, ca->sb_read_scratch);
 655
 656         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_SB],
 657                      bio_sectors(bio));
 658
 659         percpu_ref_get(&ca->io_ref);
 660         closure_bio_submit(bio, &c->sb_write);
 661 }
 662
 663 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 664 {
 665         struct bch_sb *sb = ca->disk_sb.sb;
 666         struct bio *bio = ca->disk_sb.bio;
 667
 668         sb->offset = sb->layout.sb_offset[idx];
 669
 670         SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
 671         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 672                                 null_nonce(), sb);
 673
 674         bio_reset(bio);
 675         bio_set_dev(bio, ca->disk_sb.bdev);
 676         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 677         bio->bi_iter.bi_size    =
 678                 roundup((size_t) vstruct_bytes(sb),
 679                         bdev_logical_block_size(ca->disk_sb.bdev));
 680         bio->bi_end_io          = write_super_endio;
 681         bio->bi_private         = ca;
 682         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 683         bch2_bio_map(bio, sb);
 684
 685         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_SB],
 686                      bio_sectors(bio));
 687
 688         percpu_ref_get(&ca->io_ref);
 689         closure_bio_submit(bio, &c->sb_write);
 690 }
 691
 692 int bch2_write_super(struct bch_fs *c)
 693 {
 694         struct closure *cl = &c->sb_write;
 695         struct bch_dev *ca;
 696         unsigned i, sb = 0, nr_wrote;
 697         const char *err;
 698         struct bch_devs_mask sb_written;
 699         bool wrote, can_mount_without_written, can_mount_with_written;
 700         int ret = 0;
 701
 702         lockdep_assert_held(&c->sb_lock);
 703
 704         closure_init_stack(cl);
 705         memset(&sb_written, 0, sizeof(sb_written));
 706
 707         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 708
 709         for_each_online_member(ca, c, i)
 710                 bch2_sb_from_fs(c, ca);
 711
 712         for_each_online_member(ca, c, i) {
 713                 err = bch2_sb_validate(&ca->disk_sb);
 714                 if (err) {
 715                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
 716                         ret = -1;
 717                         goto out;
 718                 }
 719         }
 720
 721         if (c->opts.nochanges ||
 722             test_bit(BCH_FS_ERROR, &c->flags))
 723                 goto out;
 724
 725         for_each_online_member(ca, c, i) {
 726                 __set_bit(ca->dev_idx, sb_written.d);
 727                 ca->sb_write_error = 0;
 728         }
 729
 730         for_each_online_member(ca, c, i)
 731                 read_back_super(c, ca);
 732         closure_sync(cl);
 733
 734         for_each_online_member(ca, c, i) {
 735                 if (!ca->sb_write_error &&
 736                     ca->disk_sb.seq !=
 737                     le64_to_cpu(ca->sb_read_scratch->seq)) {
 738                         bch2_fs_fatal_error(c,
 739                                 "Superblock modified by another process");
 740                         percpu_ref_put(&ca->io_ref);
 741                         ret = -EROFS;
 742                         goto out;
 743                 }
 744         }
 745
 746         do {
 747                 wrote = false;
 748                 for_each_online_member(ca, c, i)
 749                         if (!ca->sb_write_error &&
 750                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 751                                 write_one_super(c, ca, sb);
 752                                 wrote = true;
 753                         }
 754                 closure_sync(cl);
 755                 sb++;
 756         } while (wrote);
 757
 758         for_each_online_member(ca, c, i) {
 759                 if (ca->sb_write_error)
 760                         __clear_bit(ca->dev_idx, sb_written.d);
 761                 else
 762                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 763         }
 764
 765         nr_wrote = dev_mask_nr(&sb_written);
 766
 767         can_mount_with_written =
 768                 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
 769                                       BCH_FORCE_IF_DEGRADED);
 770
 771         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 772                 sb_written.d[i] = ~sb_written.d[i];
 773
 774         can_mount_without_written =
 775                 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
 776                                       BCH_FORCE_IF_DEGRADED);
 777
 778         /*
 779          * If we would be able to mount _without_ the devices we successfully
 780          * wrote superblocks to, we weren't able to write to enough devices:
 781          *
 782          * Exception: if we can mount without the successes because we haven't
 783          * written anything (new filesystem), we continue if we'd be able to
 784          * mount with the devices we did successfully write to:
 785          */
 786         if (bch2_fs_fatal_err_on(!nr_wrote ||
 787                                  (can_mount_without_written &&
 788                                   !can_mount_with_written), c,
 789                 "Unable to write superblock to sufficient devices"))
 790                 ret = -1;
 791 out:
 792         /* Make new options visible after they're persistent: */
 793         bch2_sb_update(c);
 794         return ret;
 795 }
 796
 797 /* BCH_SB_FIELD_journal: */
 798
 799 static int u64_cmp(const void *_l, const void *_r)
 800 {
 801         u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
 802
 803         return l < r ? -1 : l > r ? 1 : 0;
 804 }
 805
 806 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
 807                                             struct bch_sb_field *f)
 808 {
 809         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 810         struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
 811         const char *err;
 812         unsigned nr;
 813         unsigned i;
 814         u64 *b;
 815
 816         journal = bch2_sb_get_journal(sb);
 817         if (!journal)
 818                 return NULL;
 819
 820         nr = bch2_nr_journal_buckets(journal);
 821         if (!nr)
 822                 return NULL;
 823
 824         b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
 825         if (!b)
 826                 return "cannot allocate memory";
 827
 828         for (i = 0; i < nr; i++)
 829                 b[i] = le64_to_cpu(journal->buckets[i]);
 830
 831         sort(b, nr, sizeof(u64), u64_cmp, NULL);
 832
 833         err = "journal bucket at sector 0";
 834         if (!b[0])
 835                 goto err;
 836
 837         err = "journal bucket before first bucket";
 838         if (m && b[0] < le16_to_cpu(m->first_bucket))
 839                 goto err;
 840
 841         err = "journal bucket past end of device";
 842         if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
 843                 goto err;
 844
 845         err = "duplicate journal buckets";
 846         for (i = 0; i + 1 < nr; i++)
 847                 if (b[i] == b[i + 1])
 848                         goto err;
 849
 850         err = NULL;
 851 err:
 852         kfree(b);
 853         return err;
 854 }
 855
 856 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
 857         .validate       = bch2_sb_validate_journal,
 858 };
 859
 860 /* BCH_SB_FIELD_members: */
 861
 862 static const char *bch2_sb_validate_members(struct bch_sb *sb,
 863                                             struct bch_sb_field *f)
 864 {
 865         struct bch_sb_field_members *mi = field_to_type(f, members);
 866         struct bch_member *m;
 867
 868         if ((void *) (mi->members + sb->nr_devices) >
 869             vstruct_end(&mi->field))
 870                 return "Invalid superblock: bad member info";
 871
 872         for (m = mi->members;
 873              m < mi->members + sb->nr_devices;
 874              m++) {
 875                 if (!bch2_member_exists(m))
 876                         continue;
 877
 878                 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
 879                         return "Too many buckets";
 880
 881                 if (le64_to_cpu(m->nbuckets) -
 882                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
 883                         return "Not enough buckets";
 884
 885                 if (le16_to_cpu(m->bucket_size) <
 886                     le16_to_cpu(sb->block_size))
 887                         return "bucket size smaller than block size";
 888
 889                 if (le16_to_cpu(m->bucket_size) <
 890                     BCH_SB_BTREE_NODE_SIZE(sb))
 891                         return "bucket size smaller than btree node size";
 892         }
 893
 894         return NULL;
 895 }
 896
 897 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
 898         .validate       = bch2_sb_validate_members,
 899 };
 900
 901 /* BCH_SB_FIELD_crypt: */
 902
 903 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
 904                                           struct bch_sb_field *f)
 905 {
 906         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
 907
 908         if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
 909                 return "invalid field crypt: wrong size";
 910
 911         if (BCH_CRYPT_KDF_TYPE(crypt))
 912                 return "invalid field crypt: bad kdf type";
 913
 914         return NULL;
 915 }
 916
 917 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
 918         .validate       = bch2_sb_validate_crypt,
 919 };
 920
 921 /* BCH_SB_FIELD_clean: */
 922
 923 void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
 924 {
 925         struct jset_entry *entry;
 926
 927         for (entry = clean->start;
 928              entry < (struct jset_entry *) vstruct_end(&clean->field);
 929              entry = vstruct_next(entry))
 930                 bch2_bkey_renumber(BKEY_TYPE_BTREE, bkey_to_packed(entry->start), write);
 931 }
 932
 933 int bch2_fs_mark_dirty(struct bch_fs *c)
 934 {
 935         int ret;
 936
 937         /*
 938          * Unconditionally write superblock, to verify it hasn't changed before
 939          * we go rw:
 940          */
 941
 942         mutex_lock(&c->sb_lock);
 943         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 944         c->disk_sb.sb->compat[0] &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 945         ret = bch2_write_super(c);
 946         mutex_unlock(&c->sb_lock);
 947
 948         return ret;
 949 }
 950
 951 struct jset_entry *
 952 bch2_journal_super_entries_add_common(struct bch_fs *c,
 953                                       struct jset_entry *entry)
 954 {
 955         struct btree_root *r;
 956         unsigned i;
 957
 958         mutex_lock(&c->btree_root_lock);
 959
 960         for (r = c->btree_roots;
 961              r < c->btree_roots + BTREE_ID_NR;
 962              r++)
 963                 if (r->alive) {
 964                         entry->u64s     = r->key.u64s;
 965                         entry->btree_id = r - c->btree_roots;
 966                         entry->level    = r->level;
 967                         entry->type     = BCH_JSET_ENTRY_btree_root;
 968                         bkey_copy(&entry->start[0], &r->key);
 969
 970                         entry = vstruct_next(entry);
 971                 }
 972         c->btree_roots_dirty = false;
 973
 974         mutex_unlock(&c->btree_root_lock);
 975
 976         percpu_down_read_preempt_disable(&c->mark_lock);
 977
 978         {
 979                 u64 nr_inodes = percpu_u64_get(&c->usage[0]->nr_inodes);
 980                 struct jset_entry_usage *u =
 981                         container_of(entry, struct jset_entry_usage, entry);
 982
 983                 memset(u, 0, sizeof(*u));
 984                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
 985                 u->entry.type   = BCH_JSET_ENTRY_usage;
 986                 u->entry.btree_id = FS_USAGE_INODES;
 987                 u->v            = cpu_to_le64(nr_inodes);
 988
 989                 entry = vstruct_next(entry);
 990         }
 991
 992         {
 993                 struct jset_entry_usage *u =
 994                         container_of(entry, struct jset_entry_usage, entry);
 995
 996                 memset(u, 0, sizeof(*u));
 997                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
 998                 u->entry.type   = BCH_JSET_ENTRY_usage;
 999                 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1000                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1001
1002                 entry = vstruct_next(entry);
1003         }
1004
1005         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1006                 struct jset_entry_usage *u =
1007                         container_of(entry, struct jset_entry_usage, entry);
1008                 u64 sectors = percpu_u64_get(&c->usage[0]->persistent_reserved[i]);
1009
1010                 if (!sectors)
1011                         continue;
1012
1013                 memset(u, 0, sizeof(*u));
1014                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
1015                 u->entry.type   = BCH_JSET_ENTRY_usage;
1016                 u->entry.btree_id = FS_USAGE_RESERVED;
1017                 u->entry.level  = i;
1018                 u->v            = sectors;
1019
1020                 entry = vstruct_next(entry);
1021         }
1022
1023         for (i = 0; i < c->replicas.nr; i++) {
1024                 struct bch_replicas_entry *e =
1025                         cpu_replicas_entry(&c->replicas, i);
1026                 u64 sectors = percpu_u64_get(&c->usage[0]->replicas[i]);
1027                 struct jset_entry_data_usage *u =
1028                         container_of(entry, struct jset_entry_data_usage, entry);
1029
1030                 memset(u, 0, sizeof(*u));
1031                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u) + e->nr_devs,
1032                                                sizeof(u64)) - 1;
1033                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1034                 u->v            = cpu_to_le64(sectors);
1035                 memcpy(&u->r, e, replicas_entry_bytes(e));
1036
1037                 entry = vstruct_next(entry);
1038         }
1039
1040         percpu_up_read_preempt_enable(&c->mark_lock);
1041
1042         return entry;
1043 }
1044
1045 void bch2_fs_mark_clean(struct bch_fs *c)
1046 {
1047         struct bch_sb_field_clean *sb_clean;
1048         struct jset_entry *entry;
1049         unsigned u64s;
1050
1051         mutex_lock(&c->sb_lock);
1052         if (BCH_SB_CLEAN(c->disk_sb.sb))
1053                 goto out;
1054
1055         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1056
1057         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1058
1059         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1060
1061         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1062         if (!sb_clean) {
1063                 bch_err(c, "error resizing superblock while setting filesystem clean");
1064                 goto out;
1065         }
1066
1067         sb_clean->flags         = 0;
1068         sb_clean->read_clock    = cpu_to_le16(c->bucket_clock[READ].hand);
1069         sb_clean->write_clock   = cpu_to_le16(c->bucket_clock[WRITE].hand);
1070         sb_clean->journal_seq   = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1071
1072         /* Trying to catch outstanding bug: */
1073         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1074
1075         entry = sb_clean->start;
1076         entry = bch2_journal_super_entries_add_common(c, entry);
1077         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1078
1079         memset(entry, 0,
1080                vstruct_end(&sb_clean->field) - (void *) entry);
1081
1082         if (le16_to_cpu(c->disk_sb.sb->version) <
1083             bcachefs_metadata_version_bkey_renumber)
1084                 bch2_sb_clean_renumber(sb_clean, WRITE);
1085
1086         bch2_write_super(c);
1087 out:
1088         mutex_unlock(&c->sb_lock);
1089 }
1090
1091 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1092                                           struct bch_sb_field *f)
1093 {
1094         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1095
1096         if (vstruct_bytes(&clean->field) < sizeof(*clean))
1097                 return "invalid field crypt: wrong size";
1098
1099         return NULL;
1100 }
1101
1102 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1103         .validate       = bch2_sb_validate_clean,
1104 };
1105
1106 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1107 #define x(f, nr)                                        \
1108         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1109         BCH_SB_FIELDS()
1110 #undef x
1111 };
1112
1113 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1114                                           struct bch_sb_field *f)
1115 {
1116         unsigned type = le32_to_cpu(f->type);
1117
1118         return type < BCH_SB_FIELD_NR
1119                 ? bch2_sb_field_ops[type]->validate(sb, f)
1120                 : NULL;
1121 }
1122
1123 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1124                            struct bch_sb_field *f)
1125 {
1126         unsigned type = le32_to_cpu(f->type);
1127         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1128                 ? bch2_sb_field_ops[type] : NULL;
1129
1130         if (ops)
1131                 pr_buf(out, "%s", bch2_sb_fields[type]);
1132         else
1133                 pr_buf(out, "(unknown field %u)", type);
1134
1135         pr_buf(out, " (size %llu):", vstruct_bytes(f));
1136
1137         if (ops && ops->to_text)
1138                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1139 }