fs/bcachefs/super-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "checksum.h"
   5 #include "disk_groups.h"
   6 #include "ec.h"
   7 #include "error.h"
   8 #include "journal.h"
   9 #include "journal_sb.h"
  10 #include "journal_seq_blacklist.h"
  11 #include "recovery_passes.h"
  12 #include "replicas.h"
  13 #include "quota.h"
  14 #include "sb-clean.h"
  15 #include "sb-counters.h"
  16 #include "sb-downgrade.h"
  17 #include "sb-errors.h"
  18 #include "sb-members.h"
  19 #include "super-io.h"
  20 #include "super.h"
  21 #include "trace.h"
  22 #include "vstructs.h"
  23
  24 #include <linux/backing-dev.h>
  25 #include <linux/sort.h>
  26
  27 static const struct blk_holder_ops bch2_sb_handle_bdev_ops = {
  28 };
  29
  30 struct bch2_metadata_version {
  31         u16             version;
  32         const char      *name;
  33 };
  34
  35 static const struct bch2_metadata_version bch2_metadata_versions[] = {
  36 #define x(n, v) {               \
  37         .version = v,                           \
  38         .name = #n,                             \
  39 },
  40         BCH_METADATA_VERSIONS()
  41 #undef x
  42 };
  43
  44 void bch2_version_to_text(struct printbuf *out, unsigned v)
  45 {
  46         const char *str = "(unknown version)";
  47
  48         for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
  49                 if (bch2_metadata_versions[i].version == v) {
  50                         str = bch2_metadata_versions[i].name;
  51                         break;
  52                 }
  53
  54         prt_printf(out, "%u.%u: %s", BCH_VERSION_MAJOR(v), BCH_VERSION_MINOR(v), str);
  55 }
  56
  57 unsigned bch2_latest_compatible_version(unsigned v)
  58 {
  59         if (!BCH_VERSION_MAJOR(v))
  60                 return v;
  61
  62         for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
  63                 if (bch2_metadata_versions[i].version > v &&
  64                     BCH_VERSION_MAJOR(bch2_metadata_versions[i].version) ==
  65                     BCH_VERSION_MAJOR(v))
  66                         v = bch2_metadata_versions[i].version;
  67
  68         return v;
  69 }
  70
  71 const char * const bch2_sb_fields[] = {
  72 #define x(name, nr)     #name,
  73         BCH_SB_FIELDS()
  74 #undef x
  75         NULL
  76 };
  77
  78 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
  79                                   enum bch_validate_flags, struct printbuf *);
  80
  81 struct bch_sb_field *bch2_sb_field_get_id(struct bch_sb *sb,
  82                                       enum bch_sb_field_type type)
  83 {
  84         /* XXX: need locking around superblock to access optional fields */
  85
  86         vstruct_for_each(sb, f)
  87                 if (le32_to_cpu(f->type) == type)
  88                         return f;
  89         return NULL;
  90 }
  91
  92 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  93                                                    struct bch_sb_field *f,
  94                                                    unsigned u64s)
  95 {
  96         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  97         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  98
  99         BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
 100
 101         if (!f && !u64s) {
 102                 /* nothing to do: */
 103         } else if (!f) {
 104                 f = vstruct_last(sb->sb);
 105                 memset(f, 0, sizeof(u64) * u64s);
 106                 f->u64s = cpu_to_le32(u64s);
 107                 f->type = 0;
 108         } else {
 109                 void *src, *dst;
 110
 111                 src = vstruct_end(f);
 112
 113                 if (u64s) {
 114                         f->u64s = cpu_to_le32(u64s);
 115                         dst = vstruct_end(f);
 116                 } else {
 117                         dst = f;
 118                 }
 119
 120                 memmove(dst, src, vstruct_end(sb->sb) - src);
 121
 122                 if (dst > src)
 123                         memset(src, 0, dst - src);
 124         }
 125
 126         sb->sb->u64s = cpu_to_le32(sb_u64s);
 127
 128         return u64s ? f : NULL;
 129 }
 130
 131 void bch2_sb_field_delete(struct bch_sb_handle *sb,
 132                           enum bch_sb_field_type type)
 133 {
 134         struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
 135
 136         if (f)
 137                 __bch2_sb_field_resize(sb, f, 0);
 138 }
 139
 140 /* Superblock realloc/free: */
 141
 142 void bch2_free_super(struct bch_sb_handle *sb)
 143 {
 144         kfree(sb->bio);
 145         if (!IS_ERR_OR_NULL(sb->s_bdev_file))
 146                 bdev_fput(sb->s_bdev_file);
 147         kfree(sb->holder);
 148         kfree(sb->sb_name);
 149
 150         kfree(sb->sb);
 151         memset(sb, 0, sizeof(*sb));
 152 }
 153
 154 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 155 {
 156         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 157         size_t new_buffer_size;
 158         struct bch_sb *new_sb;
 159         struct bio *bio;
 160
 161         if (sb->bdev)
 162                 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
 163
 164         new_buffer_size = roundup_pow_of_two(new_bytes);
 165
 166         if (sb->sb && sb->buffer_size >= new_buffer_size)
 167                 return 0;
 168
 169         if (sb->sb && sb->have_layout) {
 170                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 171
 172                 if (new_bytes > max_bytes) {
 173                         struct printbuf buf = PRINTBUF;
 174
 175                         prt_bdevname(&buf, sb->bdev);
 176                         prt_printf(&buf, ": superblock too big: want %zu but have %llu", new_bytes, max_bytes);
 177                         pr_err("%s", buf.buf);
 178                         printbuf_exit(&buf);
 179                         return -BCH_ERR_ENOSPC_sb;
 180                 }
 181         }
 182
 183         if (sb->buffer_size >= new_buffer_size && sb->sb)
 184                 return 0;
 185
 186         if (dynamic_fault("bcachefs:add:super_realloc"))
 187                 return -BCH_ERR_ENOMEM_sb_realloc_injected;
 188
 189         new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 190         if (!new_sb)
 191                 return -BCH_ERR_ENOMEM_sb_buf_realloc;
 192
 193         sb->sb = new_sb;
 194
 195         if (sb->have_bio) {
 196                 unsigned nr_bvecs = buf_pages(sb->sb, new_buffer_size);
 197
 198                 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 199                 if (!bio)
 200                         return -BCH_ERR_ENOMEM_sb_bio_realloc;
 201
 202                 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
 203
 204                 kfree(sb->bio);
 205                 sb->bio = bio;
 206         }
 207
 208         sb->buffer_size = new_buffer_size;
 209
 210         return 0;
 211 }
 212
 213 struct bch_sb_field *bch2_sb_field_resize_id(struct bch_sb_handle *sb,
 214                                           enum bch_sb_field_type type,
 215                                           unsigned u64s)
 216 {
 217         struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
 218         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 219         ssize_t d = -old_u64s + u64s;
 220
 221         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 222                 return NULL;
 223
 224         if (sb->fs_sb) {
 225                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 226
 227                 lockdep_assert_held(&c->sb_lock);
 228
 229                 /* XXX: we're not checking that offline device have enough space */
 230
 231                 for_each_online_member(c, ca) {
 232                         struct bch_sb_handle *dev_sb = &ca->disk_sb;
 233
 234                         if (bch2_sb_realloc(dev_sb, le32_to_cpu(dev_sb->sb->u64s) + d)) {
 235                                 percpu_ref_put(&ca->io_ref);
 236                                 return NULL;
 237                         }
 238                 }
 239         }
 240
 241         f = bch2_sb_field_get_id(sb->sb, type);
 242         f = __bch2_sb_field_resize(sb, f, u64s);
 243         if (f)
 244                 f->type = cpu_to_le32(type);
 245         return f;
 246 }
 247
 248 struct bch_sb_field *bch2_sb_field_get_minsize_id(struct bch_sb_handle *sb,
 249                                                   enum bch_sb_field_type type,
 250                                                   unsigned u64s)
 251 {
 252         struct bch_sb_field *f = bch2_sb_field_get_id(sb->sb, type);
 253
 254         if (!f || le32_to_cpu(f->u64s) < u64s)
 255                 f = bch2_sb_field_resize_id(sb, type, u64s);
 256         return f;
 257 }
 258
 259 /* Superblock validate: */
 260
 261 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
 262 {
 263         u64 offset, prev_offset, max_sectors;
 264         unsigned i;
 265
 266         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 267
 268         if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
 269             !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
 270                 prt_printf(out, "Not a bcachefs superblock layout");
 271                 return -BCH_ERR_invalid_sb_layout;
 272         }
 273
 274         if (layout->layout_type != 0) {
 275                 prt_printf(out, "Invalid superblock layout type %u",
 276                        layout->layout_type);
 277                 return -BCH_ERR_invalid_sb_layout_type;
 278         }
 279
 280         if (!layout->nr_superblocks) {
 281                 prt_printf(out, "Invalid superblock layout: no superblocks");
 282                 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 283         }
 284
 285         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
 286                 prt_printf(out, "Invalid superblock layout: too many superblocks");
 287                 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 288         }
 289
 290         if (layout->sb_max_size_bits > BCH_SB_LAYOUT_SIZE_BITS_MAX) {
 291                 prt_printf(out, "Invalid superblock layout: max_size_bits too high");
 292                 return -BCH_ERR_invalid_sb_layout_sb_max_size_bits;
 293         }
 294
 295         max_sectors = 1 << layout->sb_max_size_bits;
 296
 297         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 298
 299         for (i = 1; i < layout->nr_superblocks; i++) {
 300                 offset = le64_to_cpu(layout->sb_offset[i]);
 301
 302                 if (offset < prev_offset + max_sectors) {
 303                         prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
 304                                "  (sb %u ends at %llu next starts at %llu",
 305                                i - 1, prev_offset + max_sectors, offset);
 306                         return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
 307                 }
 308                 prev_offset = offset;
 309         }
 310
 311         return 0;
 312 }
 313
 314 static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
 315 {
 316         u16 version             = le16_to_cpu(sb->version);
 317         u16 version_min         = le16_to_cpu(sb->version_min);
 318
 319         if (!bch2_version_compatible(version)) {
 320                 prt_str(out, "Unsupported superblock version ");
 321                 bch2_version_to_text(out, version);
 322                 prt_str(out, " (min ");
 323                 bch2_version_to_text(out, bcachefs_metadata_version_min);
 324                 prt_str(out, ", max ");
 325                 bch2_version_to_text(out, bcachefs_metadata_version_current);
 326                 prt_str(out, ")");
 327                 return -BCH_ERR_invalid_sb_version;
 328         }
 329
 330         if (!bch2_version_compatible(version_min)) {
 331                 prt_str(out, "Unsupported superblock version_min ");
 332                 bch2_version_to_text(out, version_min);
 333                 prt_str(out, " (min ");
 334                 bch2_version_to_text(out, bcachefs_metadata_version_min);
 335                 prt_str(out, ", max ");
 336                 bch2_version_to_text(out, bcachefs_metadata_version_current);
 337                 prt_str(out, ")");
 338                 return -BCH_ERR_invalid_sb_version;
 339         }
 340
 341         if (version_min > version) {
 342                 prt_str(out, "Bad minimum version ");
 343                 bch2_version_to_text(out, version_min);
 344                 prt_str(out, ", greater than version field ");
 345                 bch2_version_to_text(out, version);
 346                 return -BCH_ERR_invalid_sb_version;
 347         }
 348
 349         return 0;
 350 }
 351
 352 static int bch2_sb_validate(struct bch_sb_handle *disk_sb,
 353                             enum bch_validate_flags flags, struct printbuf *out)
 354 {
 355         struct bch_sb *sb = disk_sb->sb;
 356         struct bch_sb_field_members_v1 *mi;
 357         enum bch_opt_id opt_id;
 358         u16 block_size;
 359         int ret;
 360
 361         ret = bch2_sb_compatible(sb, out);
 362         if (ret)
 363                 return ret;
 364
 365         if (sb->features[1] ||
 366             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
 367                 prt_printf(out, "Filesystem has incompatible features");
 368                 return -BCH_ERR_invalid_sb_features;
 369         }
 370
 371         block_size = le16_to_cpu(sb->block_size);
 372
 373         if (block_size > PAGE_SECTORS) {
 374                 prt_printf(out, "Block size too big (got %u, max %u)",
 375                        block_size, PAGE_SECTORS);
 376                 return -BCH_ERR_invalid_sb_block_size;
 377         }
 378
 379         if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
 380                 prt_printf(out, "Bad user UUID (got zeroes)");
 381                 return -BCH_ERR_invalid_sb_uuid;
 382         }
 383
 384         if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
 385                 prt_printf(out, "Bad internal UUID (got zeroes)");
 386                 return -BCH_ERR_invalid_sb_uuid;
 387         }
 388
 389         if (!sb->nr_devices ||
 390             sb->nr_devices > BCH_SB_MEMBERS_MAX) {
 391                 prt_printf(out, "Bad number of member devices %u (max %u)",
 392                        sb->nr_devices, BCH_SB_MEMBERS_MAX);
 393                 return -BCH_ERR_invalid_sb_too_many_members;
 394         }
 395
 396         if (sb->dev_idx >= sb->nr_devices) {
 397                 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
 398                        sb->dev_idx, sb->nr_devices);
 399                 return -BCH_ERR_invalid_sb_dev_idx;
 400         }
 401
 402         if (!sb->time_precision ||
 403             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
 404                 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
 405                        le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
 406                 return -BCH_ERR_invalid_sb_time_precision;
 407         }
 408
 409         if (!flags) {
 410                 /*
 411                  * Been seeing a bug where these are getting inexplicably
 412                  * zeroed, so we're now validating them, but we have to be
 413                  * careful not to preven people's filesystems from mounting:
 414                  */
 415                 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 416                         SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 417                 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 418                         SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
 419
 420                 if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
 421                         SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
 422
 423                 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_disk_accounting_v2 &&
 424                     !BCH_SB_ALLOCATOR_STUCK_TIMEOUT(sb))
 425                         SET_BCH_SB_ALLOCATOR_STUCK_TIMEOUT(sb, 30);
 426
 427                 if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_disk_accounting_v2)
 428                         SET_BCH_SB_PROMOTE_WHOLE_EXTENTS(sb, true);
 429         }
 430
 431         for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
 432                 const struct bch_option *opt = bch2_opt_table + opt_id;
 433
 434                 if (opt->get_sb != BCH2_NO_SB_OPT) {
 435                         u64 v = bch2_opt_from_sb(sb, opt_id);
 436
 437                         prt_printf(out, "Invalid option ");
 438                         ret = bch2_opt_validate(opt, v, out);
 439                         if (ret)
 440                                 return ret;
 441
 442                         printbuf_reset(out);
 443                 }
 444         }
 445
 446         /* validate layout */
 447         ret = validate_sb_layout(&sb->layout, out);
 448         if (ret)
 449                 return ret;
 450
 451         vstruct_for_each(sb, f) {
 452                 if (!f->u64s) {
 453                         prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
 454                                le32_to_cpu(f->type));
 455                         return -BCH_ERR_invalid_sb_field_size;
 456                 }
 457
 458                 if (vstruct_next(f) > vstruct_last(sb)) {
 459                         prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
 460                                le32_to_cpu(f->type));
 461                         return -BCH_ERR_invalid_sb_field_size;
 462                 }
 463         }
 464
 465         /* members must be validated first: */
 466         mi = bch2_sb_field_get(sb, members_v1);
 467         if (!mi) {
 468                 prt_printf(out, "Invalid superblock: member info area missing");
 469                 return -BCH_ERR_invalid_sb_members_missing;
 470         }
 471
 472         ret = bch2_sb_field_validate(sb, &mi->field, flags, out);
 473         if (ret)
 474                 return ret;
 475
 476         vstruct_for_each(sb, f) {
 477                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members_v1)
 478                         continue;
 479
 480                 ret = bch2_sb_field_validate(sb, f, flags, out);
 481                 if (ret)
 482                         return ret;
 483         }
 484
 485         if ((flags & BCH_VALIDATE_write) &&
 486             bch2_sb_member_get(sb, sb->dev_idx).seq != sb->seq) {
 487                 prt_printf(out, "Invalid superblock: member seq %llu != sb seq %llu",
 488                            le64_to_cpu(bch2_sb_member_get(sb, sb->dev_idx).seq),
 489                            le64_to_cpu(sb->seq));
 490                 return -BCH_ERR_invalid_sb_members_missing;
 491         }
 492
 493         return 0;
 494 }
 495
 496 /* device open: */
 497
 498 static unsigned long le_ulong_to_cpu(unsigned long v)
 499 {
 500         return sizeof(unsigned long) == 8
 501                 ? le64_to_cpu(v)
 502                 : le32_to_cpu(v);
 503 }
 504
 505 static void le_bitvector_to_cpu(unsigned long *dst, unsigned long *src, unsigned nr)
 506 {
 507         BUG_ON(nr & (BITS_PER_TYPE(long) - 1));
 508
 509         for (unsigned i = 0; i < BITS_TO_LONGS(nr); i++)
 510                 dst[i] = le_ulong_to_cpu(src[i]);
 511 }
 512
 513 static void bch2_sb_update(struct bch_fs *c)
 514 {
 515         struct bch_sb *src = c->disk_sb.sb;
 516
 517         lockdep_assert_held(&c->sb_lock);
 518
 519         c->sb.uuid              = src->uuid;
 520         c->sb.user_uuid         = src->user_uuid;
 521         c->sb.version           = le16_to_cpu(src->version);
 522         c->sb.version_min       = le16_to_cpu(src->version_min);
 523         c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
 524         c->sb.nr_devices        = src->nr_devices;
 525         c->sb.clean             = BCH_SB_CLEAN(src);
 526         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 527
 528         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 529         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 530
 531         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 532         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 533                                           c->sb.nsec_per_time_unit);
 534         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 535
 536         c->sb.features          = le64_to_cpu(src->features[0]);
 537         c->sb.compat            = le64_to_cpu(src->compat[0]);
 538
 539         memset(c->sb.errors_silent, 0, sizeof(c->sb.errors_silent));
 540
 541         struct bch_sb_field_ext *ext = bch2_sb_field_get(src, ext);
 542         if (ext) {
 543                 le_bitvector_to_cpu(c->sb.errors_silent, (void *) ext->errors_silent,
 544                                     sizeof(c->sb.errors_silent) * 8);
 545                 c->sb.btrees_lost_data = le64_to_cpu(ext->btrees_lost_data);
 546         }
 547
 548         for_each_member_device(c, ca) {
 549                 struct bch_member m = bch2_sb_member_get(src, ca->dev_idx);
 550                 ca->mi = bch2_mi_to_cpu(&m);
 551         }
 552 }
 553
 554 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 555 {
 556         struct bch_sb_field *src_f, *dst_f;
 557         struct bch_sb *dst = dst_handle->sb;
 558         unsigned i;
 559
 560         dst->version            = src->version;
 561         dst->version_min        = src->version_min;
 562         dst->seq                = src->seq;
 563         dst->uuid               = src->uuid;
 564         dst->user_uuid          = src->user_uuid;
 565         memcpy(dst->label,      src->label, sizeof(dst->label));
 566
 567         dst->block_size         = src->block_size;
 568         dst->nr_devices         = src->nr_devices;
 569
 570         dst->time_base_lo       = src->time_base_lo;
 571         dst->time_base_hi       = src->time_base_hi;
 572         dst->time_precision     = src->time_precision;
 573         dst->write_time         = src->write_time;
 574
 575         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 576         memcpy(dst->features,   src->features,  sizeof(dst->features));
 577         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 578
 579         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 580                 int d;
 581
 582                 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 583                         continue;
 584
 585                 src_f = bch2_sb_field_get_id(src, i);
 586                 dst_f = bch2_sb_field_get_id(dst, i);
 587
 588                 d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
 589                     (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
 590                 if (d > 0) {
 591                         int ret = bch2_sb_realloc(dst_handle,
 592                                         le32_to_cpu(dst_handle->sb->u64s) + d);
 593
 594                         if (ret)
 595                                 return ret;
 596
 597                         dst = dst_handle->sb;
 598                         dst_f = bch2_sb_field_get_id(dst, i);
 599                 }
 600
 601                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 602                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 603
 604                 if (src_f)
 605                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 606         }
 607
 608         return 0;
 609 }
 610
 611 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 612 {
 613         int ret;
 614
 615         lockdep_assert_held(&c->sb_lock);
 616
 617         ret =   bch2_sb_realloc(&c->disk_sb, 0) ?:
 618                 __copy_super(&c->disk_sb, src) ?:
 619                 bch2_sb_replicas_to_cpu_replicas(c) ?:
 620                 bch2_sb_disk_groups_to_cpu(c);
 621         if (ret)
 622                 return ret;
 623
 624         bch2_sb_update(c);
 625         return 0;
 626 }
 627
 628 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 629 {
 630         return __copy_super(&ca->disk_sb, c->disk_sb.sb);
 631 }
 632
 633 /* read superblock: */
 634
 635 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 636 {
 637         size_t bytes;
 638         int ret;
 639 reread:
 640         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 641         sb->bio->bi_iter.bi_sector = offset;
 642         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 643
 644         ret = submit_bio_wait(sb->bio);
 645         if (ret) {
 646                 prt_printf(err, "IO error: %i", ret);
 647                 return ret;
 648         }
 649
 650         if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
 651             !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
 652                 prt_str(err, "Not a bcachefs superblock (got magic ");
 653                 pr_uuid(err, sb->sb->magic.b);
 654                 prt_str(err, ")");
 655                 return -BCH_ERR_invalid_sb_magic;
 656         }
 657
 658         ret = bch2_sb_compatible(sb->sb, err);
 659         if (ret)
 660                 return ret;
 661
 662         bytes = vstruct_bytes(sb->sb);
 663
 664         u64 sb_size = 512ULL << min(BCH_SB_LAYOUT_SIZE_BITS_MAX, sb->sb->layout.sb_max_size_bits);
 665         if (bytes > sb_size) {
 666                 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %llu)",
 667                            bytes, sb_size);
 668                 return -BCH_ERR_invalid_sb_too_big;
 669         }
 670
 671         if (bytes > sb->buffer_size) {
 672                 ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
 673                 if (ret)
 674                         return ret;
 675                 goto reread;
 676         }
 677
 678         enum bch_csum_type csum_type = BCH_SB_CSUM_TYPE(sb->sb);
 679         if (csum_type >= BCH_CSUM_NR) {
 680                 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 681                 return -BCH_ERR_invalid_sb_csum_type;
 682         }
 683
 684         /* XXX: verify MACs */
 685         struct bch_csum csum = csum_vstruct(NULL, csum_type, null_nonce(), sb->sb);
 686         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 687                 bch2_csum_err_msg(err, csum_type, sb->sb->csum, csum);
 688                 return -BCH_ERR_invalid_sb_csum;
 689         }
 690
 691         sb->seq = le64_to_cpu(sb->sb->seq);
 692
 693         return 0;
 694 }
 695
 696 static int __bch2_read_super(const char *path, struct bch_opts *opts,
 697                     struct bch_sb_handle *sb, bool ignore_notbchfs_msg)
 698 {
 699         u64 offset = opt_get(*opts, sb);
 700         struct bch_sb_layout layout;
 701         struct printbuf err = PRINTBUF;
 702         struct printbuf err2 = PRINTBUF;
 703         __le64 *i;
 704         int ret;
 705 #ifndef __KERNEL__
 706 retry:
 707 #endif
 708         memset(sb, 0, sizeof(*sb));
 709         sb->mode        = BLK_OPEN_READ;
 710         sb->have_bio    = true;
 711         sb->holder      = kmalloc(1, GFP_KERNEL);
 712         if (!sb->holder)
 713                 return -ENOMEM;
 714
 715         sb->sb_name = kstrdup(path, GFP_KERNEL);
 716         if (!sb->sb_name) {
 717                 ret = -ENOMEM;
 718                 prt_printf(&err, "error allocating memory for sb_name");
 719                 goto err;
 720         }
 721
 722 #ifndef __KERNEL__
 723         if (opt_get(*opts, direct_io) == false)
 724                 sb->mode |= BLK_OPEN_BUFFERED;
 725 #endif
 726
 727         if (!opt_get(*opts, noexcl))
 728                 sb->mode |= BLK_OPEN_EXCL;
 729
 730         if (!opt_get(*opts, nochanges))
 731                 sb->mode |= BLK_OPEN_WRITE;
 732
 733         sb->s_bdev_file = bdev_file_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
 734         if (IS_ERR(sb->s_bdev_file) &&
 735             PTR_ERR(sb->s_bdev_file) == -EACCES &&
 736             opt_get(*opts, read_only)) {
 737                 sb->mode &= ~BLK_OPEN_WRITE;
 738
 739                 sb->s_bdev_file = bdev_file_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
 740                 if (!IS_ERR(sb->s_bdev_file))
 741                         opt_set(*opts, nochanges, true);
 742         }
 743
 744         if (IS_ERR(sb->s_bdev_file)) {
 745                 ret = PTR_ERR(sb->s_bdev_file);
 746                 prt_printf(&err, "error opening %s: %s", path, bch2_err_str(ret));
 747                 goto err;
 748         }
 749         sb->bdev = file_bdev(sb->s_bdev_file);
 750
 751         ret = bch2_sb_realloc(sb, 0);
 752         if (ret) {
 753                 prt_printf(&err, "error allocating memory for superblock");
 754                 goto err;
 755         }
 756
 757         if (bch2_fs_init_fault("read_super")) {
 758                 prt_printf(&err, "dynamic fault");
 759                 ret = -EFAULT;
 760                 goto err;
 761         }
 762
 763         ret = read_one_super(sb, offset, &err);
 764         if (!ret)
 765                 goto got_super;
 766
 767         if (opt_defined(*opts, sb))
 768                 goto err;
 769
 770         prt_printf(&err2, "bcachefs (%s): error reading default superblock: %s\n",
 771                path, err.buf);
 772         if (ret == -BCH_ERR_invalid_sb_magic && ignore_notbchfs_msg)
 773                 bch2_print_opts(opts, KERN_INFO "%s", err2.buf);
 774         else
 775                 bch2_print_opts(opts, KERN_ERR "%s", err2.buf);
 776
 777         printbuf_exit(&err2);
 778         printbuf_reset(&err);
 779
 780         /*
 781          * Error reading primary superblock - read location of backup
 782          * superblocks:
 783          */
 784         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 785         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 786         /*
 787          * use sb buffer to read layout, since sb buffer is page aligned but
 788          * layout won't be:
 789          */
 790         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 791
 792         ret = submit_bio_wait(sb->bio);
 793         if (ret) {
 794                 prt_printf(&err, "IO error: %i", ret);
 795                 goto err;
 796         }
 797
 798         memcpy(&layout, sb->sb, sizeof(layout));
 799         ret = validate_sb_layout(&layout, &err);
 800         if (ret)
 801                 goto err;
 802
 803         for (i = layout.sb_offset;
 804              i < layout.sb_offset + layout.nr_superblocks; i++) {
 805                 offset = le64_to_cpu(*i);
 806
 807                 if (offset == opt_get(*opts, sb)) {
 808                         ret = -BCH_ERR_invalid;
 809                         continue;
 810                 }
 811
 812                 ret = read_one_super(sb, offset, &err);
 813                 if (!ret)
 814                         goto got_super;
 815         }
 816
 817         goto err;
 818
 819 got_super:
 820         if (le16_to_cpu(sb->sb->block_size) << 9 <
 821             bdev_logical_block_size(sb->bdev) &&
 822             opt_get(*opts, direct_io)) {
 823 #ifndef __KERNEL__
 824                 opt_set(*opts, direct_io, false);
 825                 bch2_free_super(sb);
 826                 goto retry;
 827 #endif
 828                 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
 829                        le16_to_cpu(sb->sb->block_size) << 9,
 830                        bdev_logical_block_size(sb->bdev));
 831                 ret = -BCH_ERR_block_size_too_small;
 832                 goto err;
 833         }
 834
 835         sb->have_layout = true;
 836
 837         ret = bch2_sb_validate(sb, 0, &err);
 838         if (ret) {
 839                 bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error validating superblock: %s\n",
 840                                 path, err.buf);
 841                 goto err_no_print;
 842         }
 843 out:
 844         printbuf_exit(&err);
 845         return ret;
 846 err:
 847         bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error reading superblock: %s\n",
 848                         path, err.buf);
 849 err_no_print:
 850         bch2_free_super(sb);
 851         goto out;
 852 }
 853
 854 int bch2_read_super(const char *path, struct bch_opts *opts,
 855                     struct bch_sb_handle *sb)
 856 {
 857         return __bch2_read_super(path, opts, sb, false);
 858 }
 859
 860 /* provide a silenced version for mount.bcachefs */
 861
 862 int bch2_read_super_silent(const char *path, struct bch_opts *opts,
 863                     struct bch_sb_handle *sb)
 864 {
 865         return __bch2_read_super(path, opts, sb, true);
 866 }
 867
 868 /* write superblock: */
 869
 870 static void write_super_endio(struct bio *bio)
 871 {
 872         struct bch_dev *ca = bio->bi_private;
 873
 874         /* XXX: return errors directly */
 875
 876         if (bch2_dev_io_err_on(bio->bi_status, ca,
 877                                bio_data_dir(bio)
 878                                ? BCH_MEMBER_ERROR_write
 879                                : BCH_MEMBER_ERROR_read,
 880                                "superblock %s error: %s",
 881                                bio_data_dir(bio) ? "write" : "read",
 882                                bch2_blk_status_to_str(bio->bi_status)))
 883                 ca->sb_write_error = 1;
 884
 885         closure_put(&ca->fs->sb_write);
 886         percpu_ref_put(&ca->io_ref);
 887 }
 888
 889 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 890 {
 891         struct bch_sb *sb = ca->disk_sb.sb;
 892         struct bio *bio = ca->disk_sb.bio;
 893
 894         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 895         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 896         bio->bi_end_io          = write_super_endio;
 897         bio->bi_private         = ca;
 898         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 899
 900         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 901                      bio_sectors(bio));
 902
 903         percpu_ref_get(&ca->io_ref);
 904         closure_bio_submit(bio, &c->sb_write);
 905 }
 906
 907 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 908 {
 909         struct bch_sb *sb = ca->disk_sb.sb;
 910         struct bio *bio = ca->disk_sb.bio;
 911
 912         sb->offset = sb->layout.sb_offset[idx];
 913
 914         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 915         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 916                                 null_nonce(), sb);
 917
 918         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 919         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 920         bio->bi_end_io          = write_super_endio;
 921         bio->bi_private         = ca;
 922         bch2_bio_map(bio, sb,
 923                      roundup((size_t) vstruct_bytes(sb),
 924                              bdev_logical_block_size(ca->disk_sb.bdev)));
 925
 926         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 927                      bio_sectors(bio));
 928
 929         percpu_ref_get(&ca->io_ref);
 930         closure_bio_submit(bio, &c->sb_write);
 931 }
 932
 933 int bch2_write_super(struct bch_fs *c)
 934 {
 935         struct closure *cl = &c->sb_write;
 936         struct printbuf err = PRINTBUF;
 937         unsigned sb = 0, nr_wrote;
 938         struct bch_devs_mask sb_written;
 939         bool wrote, can_mount_without_written, can_mount_with_written;
 940         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 941         DARRAY(struct bch_dev *) online_devices = {};
 942         int ret = 0;
 943
 944         trace_and_count(c, write_super, c, _RET_IP_);
 945
 946         if (c->opts.very_degraded)
 947                 degraded_flags |= BCH_FORCE_IF_LOST;
 948
 949         lockdep_assert_held(&c->sb_lock);
 950
 951         closure_init_stack(cl);
 952         memset(&sb_written, 0, sizeof(sb_written));
 953
 954         for_each_online_member(c, ca) {
 955                 ret = darray_push(&online_devices, ca);
 956                 if (bch2_fs_fatal_err_on(ret, c, "%s: error allocating online devices", __func__)) {
 957                         percpu_ref_put(&ca->io_ref);
 958                         goto out;
 959                 }
 960                 percpu_ref_get(&ca->io_ref);
 961         }
 962
 963         /* Make sure we're using the new magic numbers: */
 964         c->disk_sb.sb->magic = BCHFS_MAGIC;
 965         c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
 966
 967         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 968
 969         struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
 970         darray_for_each(online_devices, ca)
 971                 __bch2_members_v2_get_mut(mi, (*ca)->dev_idx)->seq = c->disk_sb.sb->seq;
 972         c->disk_sb.sb->write_time = cpu_to_le64(ktime_get_real_seconds());
 973
 974         if (test_bit(BCH_FS_error, &c->flags))
 975                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 976         if (test_bit(BCH_FS_topology_error, &c->flags))
 977                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 978
 979         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 980
 981         bch2_sb_counters_from_cpu(c);
 982         bch2_sb_members_from_cpu(c);
 983         bch2_sb_members_cpy_v2_v1(&c->disk_sb);
 984         bch2_sb_errors_from_cpu(c);
 985         bch2_sb_downgrade_update(c);
 986
 987         darray_for_each(online_devices, ca)
 988                 bch2_sb_from_fs(c, (*ca));
 989
 990         darray_for_each(online_devices, ca) {
 991                 printbuf_reset(&err);
 992
 993                 ret = bch2_sb_validate(&(*ca)->disk_sb, BCH_VALIDATE_write, &err);
 994                 if (ret) {
 995                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 996                         goto out;
 997                 }
 998         }
 999
1000         if (c->opts.nochanges)
1001                 goto out;
1002
1003         /*
1004          * Defer writing the superblock until filesystem initialization is
1005          * complete - don't write out a partly initialized superblock:
1006          */
1007         if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
1008                 goto out;
1009
1010         if (le16_to_cpu(c->disk_sb.sb->version) > bcachefs_metadata_version_current) {
1011                 struct printbuf buf = PRINTBUF;
1012                 prt_printf(&buf, "attempting to write superblock that wasn't version downgraded (");
1013                 bch2_version_to_text(&buf, le16_to_cpu(c->disk_sb.sb->version));
1014                 prt_str(&buf, " > ");
1015                 bch2_version_to_text(&buf, bcachefs_metadata_version_current);
1016                 prt_str(&buf, ")");
1017                 bch2_fs_fatal_error(c, ": %s", buf.buf);
1018                 printbuf_exit(&buf);
1019                 return -BCH_ERR_sb_not_downgraded;
1020         }
1021
1022         darray_for_each(online_devices, ca) {
1023                 __set_bit((*ca)->dev_idx, sb_written.d);
1024                 (*ca)->sb_write_error = 0;
1025         }
1026
1027         darray_for_each(online_devices, ca)
1028                 read_back_super(c, *ca);
1029         closure_sync(cl);
1030
1031         darray_for_each(online_devices, cap) {
1032                 struct bch_dev *ca = *cap;
1033
1034                 if (ca->sb_write_error)
1035                         continue;
1036
1037                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
1038                         struct printbuf buf = PRINTBUF;
1039                         prt_char(&buf, ' ');
1040                         prt_bdevname(&buf, ca->disk_sb.bdev);
1041                         prt_printf(&buf,
1042                                 ": Superblock write was silently dropped! (seq %llu expected %llu)",
1043                                 le64_to_cpu(ca->sb_read_scratch->seq),
1044                                 ca->disk_sb.seq);
1045                         bch2_fs_fatal_error(c, "%s", buf.buf);
1046                         printbuf_exit(&buf);
1047                         ret = -BCH_ERR_erofs_sb_err;
1048                 }
1049
1050                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
1051                         struct printbuf buf = PRINTBUF;
1052                         prt_char(&buf, ' ');
1053                         prt_bdevname(&buf, ca->disk_sb.bdev);
1054                         prt_printf(&buf,
1055                                 ": Superblock modified by another process (seq %llu expected %llu)",
1056                                 le64_to_cpu(ca->sb_read_scratch->seq),
1057                                 ca->disk_sb.seq);
1058                         bch2_fs_fatal_error(c, "%s", buf.buf);
1059                         printbuf_exit(&buf);
1060                         ret = -BCH_ERR_erofs_sb_err;
1061                 }
1062         }
1063
1064         if (ret)
1065                 goto out;
1066
1067         do {
1068                 wrote = false;
1069                 darray_for_each(online_devices, cap) {
1070                         struct bch_dev *ca = *cap;
1071                         if (!ca->sb_write_error &&
1072                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
1073                                 write_one_super(c, ca, sb);
1074                                 wrote = true;
1075                         }
1076                 }
1077                 closure_sync(cl);
1078                 sb++;
1079         } while (wrote);
1080
1081         darray_for_each(online_devices, cap) {
1082                 struct bch_dev *ca = *cap;
1083                 if (ca->sb_write_error)
1084                         __clear_bit(ca->dev_idx, sb_written.d);
1085                 else
1086                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
1087         }
1088
1089         nr_wrote = dev_mask_nr(&sb_written);
1090
1091         can_mount_with_written =
1092                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1093
1094         for (unsigned i = 0; i < ARRAY_SIZE(sb_written.d); i++)
1095                 sb_written.d[i] = ~sb_written.d[i];
1096
1097         can_mount_without_written =
1098                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1099
1100         /*
1101          * If we would be able to mount _without_ the devices we successfully
1102          * wrote superblocks to, we weren't able to write to enough devices:
1103          *
1104          * Exception: if we can mount without the successes because we haven't
1105          * written anything (new filesystem), we continue if we'd be able to
1106          * mount with the devices we did successfully write to:
1107          */
1108         if (bch2_fs_fatal_err_on(!nr_wrote ||
1109                                  !can_mount_with_written ||
1110                                  (can_mount_without_written &&
1111                                   !can_mount_with_written), c,
1112                 ": Unable to write superblock to sufficient devices (from %ps)",
1113                 (void *) _RET_IP_))
1114                 ret = -1;
1115 out:
1116         /* Make new options visible after they're persistent: */
1117         bch2_sb_update(c);
1118         darray_for_each(online_devices, ca)
1119                 percpu_ref_put(&(*ca)->io_ref);
1120         darray_exit(&online_devices);
1121         printbuf_exit(&err);
1122         return ret;
1123 }
1124
1125 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
1126 {
1127         mutex_lock(&c->sb_lock);
1128         if (!(c->sb.features & (1ULL << feat))) {
1129                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
1130
1131                 bch2_write_super(c);
1132         }
1133         mutex_unlock(&c->sb_lock);
1134 }
1135
1136 /* Downgrade if superblock is at a higher version than currently supported: */
1137 bool bch2_check_version_downgrade(struct bch_fs *c)
1138 {
1139         bool ret = bcachefs_metadata_version_current < c->sb.version;
1140
1141         lockdep_assert_held(&c->sb_lock);
1142
1143         /*
1144          * Downgrade, if superblock is at a higher version than currently
1145          * supported:
1146          *
1147          * c->sb will be checked before we write the superblock, so update it as
1148          * well:
1149          */
1150         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
1151                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1152         if (c->sb.version > bcachefs_metadata_version_current)
1153                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1154         if (c->sb.version_min > bcachefs_metadata_version_current)
1155                 c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
1156         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1157         return ret;
1158 }
1159
1160 void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version)
1161 {
1162         lockdep_assert_held(&c->sb_lock);
1163
1164         if (BCH_VERSION_MAJOR(new_version) >
1165             BCH_VERSION_MAJOR(le16_to_cpu(c->disk_sb.sb->version)))
1166                 bch2_sb_field_resize(&c->disk_sb, downgrade, 0);
1167
1168         c->disk_sb.sb->version = cpu_to_le16(new_version);
1169         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1170 }
1171
1172 static int bch2_sb_ext_validate(struct bch_sb *sb, struct bch_sb_field *f,
1173                                 enum bch_validate_flags flags, struct printbuf *err)
1174 {
1175         if (vstruct_bytes(f) < 88) {
1176                 prt_printf(err, "field too small (%zu < %u)", vstruct_bytes(f), 88);
1177                 return -BCH_ERR_invalid_sb_ext;
1178         }
1179
1180         return 0;
1181 }
1182
1183 static void bch2_sb_ext_to_text(struct printbuf *out, struct bch_sb *sb,
1184                                 struct bch_sb_field *f)
1185 {
1186         struct bch_sb_field_ext *e = field_to_type(f, ext);
1187
1188         prt_printf(out, "Recovery passes required:\t");
1189         prt_bitflags(out, bch2_recovery_passes,
1190                      bch2_recovery_passes_from_stable(le64_to_cpu(e->recovery_passes_required[0])));
1191         prt_newline(out);
1192
1193         unsigned long *errors_silent = kmalloc(sizeof(e->errors_silent), GFP_KERNEL);
1194         if (errors_silent) {
1195                 le_bitvector_to_cpu(errors_silent, (void *) e->errors_silent, sizeof(e->errors_silent) * 8);
1196
1197                 prt_printf(out, "Errors to silently fix:\t");
1198                 prt_bitflags_vector(out, bch2_sb_error_strs, errors_silent,
1199                                     min(BCH_FSCK_ERR_MAX, sizeof(e->errors_silent) * 8));
1200                 prt_newline(out);
1201
1202                 kfree(errors_silent);
1203         }
1204
1205         prt_printf(out, "Btrees with missing data:\t");
1206         prt_bitflags(out, __bch2_btree_ids, le64_to_cpu(e->btrees_lost_data));
1207         prt_newline(out);
1208 }
1209
1210 static const struct bch_sb_field_ops bch_sb_field_ops_ext = {
1211         .validate       = bch2_sb_ext_validate,
1212         .to_text        = bch2_sb_ext_to_text,
1213 };
1214
1215 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1216 #define x(f, nr)                                        \
1217         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1218         BCH_SB_FIELDS()
1219 #undef x
1220 };
1221
1222 static const struct bch_sb_field_ops bch2_sb_field_null_ops;
1223
1224 static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
1225 {
1226         return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
1227                 ? bch2_sb_field_ops[type]
1228                 : &bch2_sb_field_null_ops;
1229 }
1230
1231 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1232                                   enum bch_validate_flags flags, struct printbuf *err)
1233 {
1234         unsigned type = le32_to_cpu(f->type);
1235         struct printbuf field_err = PRINTBUF;
1236         const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1237         int ret;
1238
1239         ret = ops->validate ? ops->validate(sb, f, flags, &field_err) : 0;
1240         if (ret) {
1241                 prt_printf(err, "Invalid superblock section %s: %s",
1242                            bch2_sb_fields[type], field_err.buf);
1243                 prt_newline(err);
1244                 bch2_sb_field_to_text(err, sb, f);
1245         }
1246
1247         printbuf_exit(&field_err);
1248         return ret;
1249 }
1250
1251 void __bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1252                              struct bch_sb_field *f)
1253 {
1254         unsigned type = le32_to_cpu(f->type);
1255         const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1256
1257         if (!out->nr_tabstops)
1258                 printbuf_tabstop_push(out, 32);
1259
1260         if (ops->to_text)
1261                 ops->to_text(out, sb, f);
1262 }
1263
1264 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1265                            struct bch_sb_field *f)
1266 {
1267         unsigned type = le32_to_cpu(f->type);
1268
1269         if (type < BCH_SB_FIELD_NR)
1270                 prt_printf(out, "%s", bch2_sb_fields[type]);
1271         else
1272                 prt_printf(out, "(unknown field %u)", type);
1273
1274         prt_printf(out, " (size %zu):", vstruct_bytes(f));
1275         prt_newline(out);
1276
1277         __bch2_sb_field_to_text(out, sb, f);
1278 }
1279
1280 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1281 {
1282         unsigned i;
1283
1284         prt_printf(out, "Type:                    %u", l->layout_type);
1285         prt_newline(out);
1286
1287         prt_str(out, "Superblock max size:     ");
1288         prt_units_u64(out, 512 << l->sb_max_size_bits);
1289         prt_newline(out);
1290
1291         prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1292         prt_newline(out);
1293
1294         prt_str(out, "Offsets:                 ");
1295         for (i = 0; i < l->nr_superblocks; i++) {
1296                 if (i)
1297                         prt_str(out, ", ");
1298                 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1299         }
1300         prt_newline(out);
1301 }
1302
1303 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1304                      bool print_layout, unsigned fields)
1305 {
1306         if (!out->nr_tabstops)
1307                 printbuf_tabstop_push(out, 44);
1308
1309         prt_printf(out, "External UUID:\t");
1310         pr_uuid(out, sb->user_uuid.b);
1311         prt_newline(out);
1312
1313         prt_printf(out, "Internal UUID:\t");
1314         pr_uuid(out, sb->uuid.b);
1315         prt_newline(out);
1316
1317         prt_printf(out, "Magic number:\t");
1318         pr_uuid(out, sb->magic.b);
1319         prt_newline(out);
1320
1321         prt_printf(out, "Device index:\t%u\n", sb->dev_idx);
1322
1323         prt_printf(out, "Label:\t");
1324         if (!strlen(sb->label))
1325                 prt_printf(out, "(none)");
1326         else
1327                 prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1328         prt_newline(out);
1329
1330         prt_printf(out, "Version:\t");
1331         bch2_version_to_text(out, le16_to_cpu(sb->version));
1332         prt_newline(out);
1333
1334         prt_printf(out, "Version upgrade complete:\t");
1335         bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
1336         prt_newline(out);
1337
1338         prt_printf(out, "Oldest version on disk:\t");
1339         bch2_version_to_text(out, le16_to_cpu(sb->version_min));
1340         prt_newline(out);
1341
1342         prt_printf(out, "Created:\t");
1343         if (sb->time_base_lo)
1344                 bch2_prt_datetime(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1345         else
1346                 prt_printf(out, "(not set)");
1347         prt_newline(out);
1348
1349         prt_printf(out, "Sequence number:\t");
1350         prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1351         prt_newline(out);
1352
1353         prt_printf(out, "Time of last write:\t");
1354         bch2_prt_datetime(out, le64_to_cpu(sb->write_time));
1355         prt_newline(out);
1356
1357         prt_printf(out, "Superblock size:\t");
1358         prt_units_u64(out, vstruct_bytes(sb));
1359         prt_str(out, "/");
1360         prt_units_u64(out, 512ULL << sb->layout.sb_max_size_bits);
1361         prt_newline(out);
1362
1363         prt_printf(out, "Clean:\t%llu\n", BCH_SB_CLEAN(sb));
1364         prt_printf(out, "Devices:\t%u\n", bch2_sb_nr_devices(sb));
1365
1366         prt_printf(out, "Sections:\t");
1367         u64 fields_have = 0;
1368         vstruct_for_each(sb, f)
1369                 fields_have |= 1 << le32_to_cpu(f->type);
1370         prt_bitflags(out, bch2_sb_fields, fields_have);
1371         prt_newline(out);
1372
1373         prt_printf(out, "Features:\t");
1374         prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1375         prt_newline(out);
1376
1377         prt_printf(out, "Compat features:\t");
1378         prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1379         prt_newline(out);
1380
1381         prt_newline(out);
1382         prt_printf(out, "Options:");
1383         prt_newline(out);
1384         printbuf_indent_add(out, 2);
1385         {
1386                 enum bch_opt_id id;
1387
1388                 for (id = 0; id < bch2_opts_nr; id++) {
1389                         const struct bch_option *opt = bch2_opt_table + id;
1390
1391                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1392                                 u64 v = bch2_opt_from_sb(sb, id);
1393
1394                                 prt_printf(out, "%s:\t", opt->attr.name);
1395                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1396                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1397                                 prt_newline(out);
1398                         }
1399                 }
1400         }
1401
1402         printbuf_indent_sub(out, 2);
1403
1404         if (print_layout) {
1405                 prt_newline(out);
1406                 prt_printf(out, "layout:");
1407                 prt_newline(out);
1408                 printbuf_indent_add(out, 2);
1409                 bch2_sb_layout_to_text(out, &sb->layout);
1410                 printbuf_indent_sub(out, 2);
1411         }
1412
1413         vstruct_for_each(sb, f)
1414                 if (fields & (1 << le32_to_cpu(f->type))) {
1415                         prt_newline(out);
1416                         bch2_sb_field_to_text(out, sb, f);
1417                 }
1418 }