git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/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.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                                   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->bdev_handle))
 146                 bdev_release(sb->bdev_handle);
 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->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         max_sectors = 1 << layout->sb_max_size_bits;
 291
 292         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 293
 294         for (i = 1; i < layout->nr_superblocks; i++) {
 295                 offset = le64_to_cpu(layout->sb_offset[i]);
 296
 297                 if (offset < prev_offset + max_sectors) {
 298                         prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
 299                                "  (sb %u ends at %llu next starts at %llu",
 300                                i - 1, prev_offset + max_sectors, offset);
 301                         return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
 302                 }
 303                 prev_offset = offset;
 304         }
 305
 306         return 0;
 307 }
 308
 309 static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
 310 {
 311         u16 version             = le16_to_cpu(sb->version);
 312         u16 version_min         = le16_to_cpu(sb->version_min);
 313
 314         if (!bch2_version_compatible(version)) {
 315                 prt_str(out, "Unsupported superblock version ");
 316                 bch2_version_to_text(out, version);
 317                 prt_str(out, " (min ");
 318                 bch2_version_to_text(out, bcachefs_metadata_version_min);
 319                 prt_str(out, ", max ");
 320                 bch2_version_to_text(out, bcachefs_metadata_version_current);
 321                 prt_str(out, ")");
 322                 return -BCH_ERR_invalid_sb_version;
 323         }
 324
 325         if (!bch2_version_compatible(version_min)) {
 326                 prt_str(out, "Unsupported superblock version_min ");
 327                 bch2_version_to_text(out, version_min);
 328                 prt_str(out, " (min ");
 329                 bch2_version_to_text(out, bcachefs_metadata_version_min);
 330                 prt_str(out, ", max ");
 331                 bch2_version_to_text(out, bcachefs_metadata_version_current);
 332                 prt_str(out, ")");
 333                 return -BCH_ERR_invalid_sb_version;
 334         }
 335
 336         if (version_min > version) {
 337                 prt_str(out, "Bad minimum version ");
 338                 bch2_version_to_text(out, version_min);
 339                 prt_str(out, ", greater than version field ");
 340                 bch2_version_to_text(out, version);
 341                 return -BCH_ERR_invalid_sb_version;
 342         }
 343
 344         return 0;
 345 }
 346
 347 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 348                             int rw)
 349 {
 350         struct bch_sb *sb = disk_sb->sb;
 351         struct bch_sb_field_members_v1 *mi;
 352         enum bch_opt_id opt_id;
 353         u16 block_size;
 354         int ret;
 355
 356         ret = bch2_sb_compatible(sb, out);
 357         if (ret)
 358                 return ret;
 359
 360         if (sb->features[1] ||
 361             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
 362                 prt_printf(out, "Filesystem has incompatible features");
 363                 return -BCH_ERR_invalid_sb_features;
 364         }
 365
 366         block_size = le16_to_cpu(sb->block_size);
 367
 368         if (block_size > PAGE_SECTORS) {
 369                 prt_printf(out, "Block size too big (got %u, max %u)",
 370                        block_size, PAGE_SECTORS);
 371                 return -BCH_ERR_invalid_sb_block_size;
 372         }
 373
 374         if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
 375                 prt_printf(out, "Bad user UUID (got zeroes)");
 376                 return -BCH_ERR_invalid_sb_uuid;
 377         }
 378
 379         if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
 380                 prt_printf(out, "Bad internal UUID (got zeroes)");
 381                 return -BCH_ERR_invalid_sb_uuid;
 382         }
 383
 384         if (!sb->nr_devices ||
 385             sb->nr_devices > BCH_SB_MEMBERS_MAX) {
 386                 prt_printf(out, "Bad number of member devices %u (max %u)",
 387                        sb->nr_devices, BCH_SB_MEMBERS_MAX);
 388                 return -BCH_ERR_invalid_sb_too_many_members;
 389         }
 390
 391         if (sb->dev_idx >= sb->nr_devices) {
 392                 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
 393                        sb->dev_idx, sb->nr_devices);
 394                 return -BCH_ERR_invalid_sb_dev_idx;
 395         }
 396
 397         if (!sb->time_precision ||
 398             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
 399                 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
 400                        le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
 401                 return -BCH_ERR_invalid_sb_time_precision;
 402         }
 403
 404         if (rw == READ) {
 405                 /*
 406                  * Been seeing a bug where these are getting inexplicably
 407                  * zeroed, so we're now validating them, but we have to be
 408                  * careful not to preven people's filesystems from mounting:
 409                  */
 410                 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 411                         SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 412                 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 413                         SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
 414
 415                 if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
 416                         SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
 417         }
 418
 419         for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
 420                 const struct bch_option *opt = bch2_opt_table + opt_id;
 421
 422                 if (opt->get_sb != BCH2_NO_SB_OPT) {
 423                         u64 v = bch2_opt_from_sb(sb, opt_id);
 424
 425                         prt_printf(out, "Invalid option ");
 426                         ret = bch2_opt_validate(opt, v, out);
 427                         if (ret)
 428                                 return ret;
 429
 430                         printbuf_reset(out);
 431                 }
 432         }
 433
 434         /* validate layout */
 435         ret = validate_sb_layout(&sb->layout, out);
 436         if (ret)
 437                 return ret;
 438
 439         vstruct_for_each(sb, f) {
 440                 if (!f->u64s) {
 441                         prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
 442                                le32_to_cpu(f->type));
 443                         return -BCH_ERR_invalid_sb_field_size;
 444                 }
 445
 446                 if (vstruct_next(f) > vstruct_last(sb)) {
 447                         prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
 448                                le32_to_cpu(f->type));
 449                         return -BCH_ERR_invalid_sb_field_size;
 450                 }
 451         }
 452
 453         /* members must be validated first: */
 454         mi = bch2_sb_field_get(sb, members_v1);
 455         if (!mi) {
 456                 prt_printf(out, "Invalid superblock: member info area missing");
 457                 return -BCH_ERR_invalid_sb_members_missing;
 458         }
 459
 460         ret = bch2_sb_field_validate(sb, &mi->field, out);
 461         if (ret)
 462                 return ret;
 463
 464         vstruct_for_each(sb, f) {
 465                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members_v1)
 466                         continue;
 467
 468                 ret = bch2_sb_field_validate(sb, f, out);
 469                 if (ret)
 470                         return ret;
 471         }
 472
 473         return 0;
 474 }
 475
 476 /* device open: */
 477
 478 static unsigned long le_ulong_to_cpu(unsigned long v)
 479 {
 480         return sizeof(unsigned long) == 8
 481                 ? le64_to_cpu(v)
 482                 : le32_to_cpu(v);
 483 }
 484
 485 static void le_bitvector_to_cpu(unsigned long *dst, unsigned long *src, unsigned nr)
 486 {
 487         BUG_ON(nr & (BITS_PER_TYPE(long) - 1));
 488
 489         for (unsigned i = 0; i < BITS_TO_LONGS(nr); i++)
 490                 dst[i] = le_ulong_to_cpu(src[i]);
 491 }
 492
 493 static void bch2_sb_update(struct bch_fs *c)
 494 {
 495         struct bch_sb *src = c->disk_sb.sb;
 496
 497         lockdep_assert_held(&c->sb_lock);
 498
 499         c->sb.uuid              = src->uuid;
 500         c->sb.user_uuid         = src->user_uuid;
 501         c->sb.version           = le16_to_cpu(src->version);
 502         c->sb.version_min       = le16_to_cpu(src->version_min);
 503         c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
 504         c->sb.nr_devices        = src->nr_devices;
 505         c->sb.clean             = BCH_SB_CLEAN(src);
 506         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 507
 508         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 509         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 510
 511         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 512         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 513                                           c->sb.nsec_per_time_unit);
 514         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 515
 516         c->sb.features          = le64_to_cpu(src->features[0]);
 517         c->sb.compat            = le64_to_cpu(src->compat[0]);
 518
 519         memset(c->sb.errors_silent, 0, sizeof(c->sb.errors_silent));
 520
 521         struct bch_sb_field_ext *ext = bch2_sb_field_get(src, ext);
 522         if (ext)
 523                 le_bitvector_to_cpu(c->sb.errors_silent, (void *) ext->errors_silent,
 524                                     sizeof(c->sb.errors_silent) * 8);
 525
 526         for_each_member_device(c, ca) {
 527                 struct bch_member m = bch2_sb_member_get(src, ca->dev_idx);
 528                 ca->mi = bch2_mi_to_cpu(&m);
 529         }
 530 }
 531
 532 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 533 {
 534         struct bch_sb_field *src_f, *dst_f;
 535         struct bch_sb *dst = dst_handle->sb;
 536         unsigned i;
 537
 538         dst->version            = src->version;
 539         dst->version_min        = src->version_min;
 540         dst->seq                = src->seq;
 541         dst->uuid               = src->uuid;
 542         dst->user_uuid          = src->user_uuid;
 543         memcpy(dst->label,      src->label, sizeof(dst->label));
 544
 545         dst->block_size         = src->block_size;
 546         dst->nr_devices         = src->nr_devices;
 547
 548         dst->time_base_lo       = src->time_base_lo;
 549         dst->time_base_hi       = src->time_base_hi;
 550         dst->time_precision     = src->time_precision;
 551         dst->write_time         = src->write_time;
 552
 553         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 554         memcpy(dst->features,   src->features,  sizeof(dst->features));
 555         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 556
 557         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 558                 int d;
 559
 560                 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 561                         continue;
 562
 563                 src_f = bch2_sb_field_get_id(src, i);
 564                 dst_f = bch2_sb_field_get_id(dst, i);
 565
 566                 d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
 567                     (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
 568                 if (d > 0) {
 569                         int ret = bch2_sb_realloc(dst_handle,
 570                                         le32_to_cpu(dst_handle->sb->u64s) + d);
 571
 572                         if (ret)
 573                                 return ret;
 574
 575                         dst = dst_handle->sb;
 576                         dst_f = bch2_sb_field_get_id(dst, i);
 577                 }
 578
 579                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 580                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 581
 582                 if (src_f)
 583                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 584         }
 585
 586         return 0;
 587 }
 588
 589 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 590 {
 591         int ret;
 592
 593         lockdep_assert_held(&c->sb_lock);
 594
 595         ret =   bch2_sb_realloc(&c->disk_sb, 0) ?:
 596                 __copy_super(&c->disk_sb, src) ?:
 597                 bch2_sb_replicas_to_cpu_replicas(c) ?:
 598                 bch2_sb_disk_groups_to_cpu(c);
 599         if (ret)
 600                 return ret;
 601
 602         bch2_sb_update(c);
 603         return 0;
 604 }
 605
 606 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 607 {
 608         return __copy_super(&ca->disk_sb, c->disk_sb.sb);
 609 }
 610
 611 /* read superblock: */
 612
 613 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 614 {
 615         size_t bytes;
 616         int ret;
 617 reread:
 618         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 619         sb->bio->bi_iter.bi_sector = offset;
 620         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 621
 622         ret = submit_bio_wait(sb->bio);
 623         if (ret) {
 624                 prt_printf(err, "IO error: %i", ret);
 625                 return ret;
 626         }
 627
 628         if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
 629             !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
 630                 prt_str(err, "Not a bcachefs superblock (got magic ");
 631                 pr_uuid(err, sb->sb->magic.b);
 632                 prt_str(err, ")");
 633                 return -BCH_ERR_invalid_sb_magic;
 634         }
 635
 636         ret = bch2_sb_compatible(sb->sb, err);
 637         if (ret)
 638                 return ret;
 639
 640         bytes = vstruct_bytes(sb->sb);
 641
 642         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 643                 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 644                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 645                 return -BCH_ERR_invalid_sb_too_big;
 646         }
 647
 648         if (bytes > sb->buffer_size) {
 649                 ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
 650                 if (ret)
 651                         return ret;
 652                 goto reread;
 653         }
 654
 655         enum bch_csum_type csum_type = BCH_SB_CSUM_TYPE(sb->sb);
 656         if (csum_type >= BCH_CSUM_NR) {
 657                 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 658                 return -BCH_ERR_invalid_sb_csum_type;
 659         }
 660
 661         /* XXX: verify MACs */
 662         struct bch_csum csum = csum_vstruct(NULL, csum_type, null_nonce(), sb->sb);
 663         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 664                 bch2_csum_err_msg(err, csum_type, sb->sb->csum, csum);
 665                 return -BCH_ERR_invalid_sb_csum;
 666         }
 667
 668         sb->seq = le64_to_cpu(sb->sb->seq);
 669
 670         return 0;
 671 }
 672
 673 static int __bch2_read_super(const char *path, struct bch_opts *opts,
 674                     struct bch_sb_handle *sb, bool ignore_notbchfs_msg)
 675 {
 676         u64 offset = opt_get(*opts, sb);
 677         struct bch_sb_layout layout;
 678         struct printbuf err = PRINTBUF;
 679         struct printbuf err2 = PRINTBUF;
 680         __le64 *i;
 681         int ret;
 682 #ifndef __KERNEL__
 683 retry:
 684 #endif
 685         memset(sb, 0, sizeof(*sb));
 686         sb->mode        = BLK_OPEN_READ;
 687         sb->have_bio    = true;
 688         sb->holder      = kmalloc(1, GFP_KERNEL);
 689         if (!sb->holder)
 690                 return -ENOMEM;
 691
 692         sb->sb_name = kstrdup(path, GFP_KERNEL);
 693         if (!sb->sb_name)
 694                 return -ENOMEM;
 695
 696 #ifndef __KERNEL__
 697         if (opt_get(*opts, direct_io) == false)
 698                 sb->mode |= BLK_OPEN_BUFFERED;
 699 #endif
 700
 701         if (!opt_get(*opts, noexcl))
 702                 sb->mode |= BLK_OPEN_EXCL;
 703
 704         if (!opt_get(*opts, nochanges))
 705                 sb->mode |= BLK_OPEN_WRITE;
 706
 707         sb->bdev_handle = bdev_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
 708         if (IS_ERR(sb->bdev_handle) &&
 709             PTR_ERR(sb->bdev_handle) == -EACCES &&
 710             opt_get(*opts, read_only)) {
 711                 sb->mode &= ~BLK_OPEN_WRITE;
 712
 713                 sb->bdev_handle = bdev_open_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
 714                 if (!IS_ERR(sb->bdev_handle))
 715                         opt_set(*opts, nochanges, true);
 716         }
 717
 718         if (IS_ERR(sb->bdev_handle)) {
 719                 ret = PTR_ERR(sb->bdev_handle);
 720                 prt_printf(&err, "error opening %s: %s", path, bch2_err_str(ret));
 721                 goto err;
 722         }
 723         sb->bdev = sb->bdev_handle->bdev;
 724
 725         ret = bch2_sb_realloc(sb, 0);
 726         if (ret) {
 727                 prt_printf(&err, "error allocating memory for superblock");
 728                 goto err;
 729         }
 730
 731         if (bch2_fs_init_fault("read_super")) {
 732                 prt_printf(&err, "dynamic fault");
 733                 ret = -EFAULT;
 734                 goto err;
 735         }
 736
 737         ret = read_one_super(sb, offset, &err);
 738         if (!ret)
 739                 goto got_super;
 740
 741         if (opt_defined(*opts, sb))
 742                 goto err;
 743
 744         prt_printf(&err2, "bcachefs (%s): error reading default superblock: %s\n",
 745                path, err.buf);
 746         if (ret == -BCH_ERR_invalid_sb_magic && ignore_notbchfs_msg)
 747                 bch2_print_opts(opts, KERN_INFO "%s", err2.buf);
 748         else
 749                 bch2_print_opts(opts, KERN_ERR "%s", err2.buf);
 750
 751         printbuf_exit(&err2);
 752         printbuf_reset(&err);
 753
 754         /*
 755          * Error reading primary superblock - read location of backup
 756          * superblocks:
 757          */
 758         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 759         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 760         /*
 761          * use sb buffer to read layout, since sb buffer is page aligned but
 762          * layout won't be:
 763          */
 764         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 765
 766         ret = submit_bio_wait(sb->bio);
 767         if (ret) {
 768                 prt_printf(&err, "IO error: %i", ret);
 769                 goto err;
 770         }
 771
 772         memcpy(&layout, sb->sb, sizeof(layout));
 773         ret = validate_sb_layout(&layout, &err);
 774         if (ret)
 775                 goto err;
 776
 777         for (i = layout.sb_offset;
 778              i < layout.sb_offset + layout.nr_superblocks; i++) {
 779                 offset = le64_to_cpu(*i);
 780
 781                 if (offset == opt_get(*opts, sb))
 782                         continue;
 783
 784                 ret = read_one_super(sb, offset, &err);
 785                 if (!ret)
 786                         goto got_super;
 787         }
 788
 789         goto err;
 790
 791 got_super:
 792         if (le16_to_cpu(sb->sb->block_size) << 9 <
 793             bdev_logical_block_size(sb->bdev) &&
 794             opt_get(*opts, direct_io)) {
 795 #ifndef __KERNEL__
 796                 opt_set(*opts, direct_io, false);
 797                 bch2_free_super(sb);
 798                 goto retry;
 799 #endif
 800                 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
 801                        le16_to_cpu(sb->sb->block_size) << 9,
 802                        bdev_logical_block_size(sb->bdev));
 803                 ret = -BCH_ERR_block_size_too_small;
 804                 goto err;
 805         }
 806
 807         ret = 0;
 808         sb->have_layout = true;
 809
 810         ret = bch2_sb_validate(sb, &err, READ);
 811         if (ret) {
 812                 bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error validating superblock: %s\n",
 813                                 path, err.buf);
 814                 goto err_no_print;
 815         }
 816 out:
 817         printbuf_exit(&err);
 818         return ret;
 819 err:
 820         bch2_print_opts(opts, KERN_ERR "bcachefs (%s): error reading superblock: %s\n",
 821                         path, err.buf);
 822 err_no_print:
 823         bch2_free_super(sb);
 824         goto out;
 825 }
 826
 827 int bch2_read_super(const char *path, struct bch_opts *opts,
 828                     struct bch_sb_handle *sb)
 829 {
 830         return __bch2_read_super(path, opts, sb, false);
 831 }
 832
 833 /* provide a silenced version for mount.bcachefs */
 834
 835 int bch2_read_super_silent(const char *path, struct bch_opts *opts,
 836                     struct bch_sb_handle *sb)
 837 {
 838         return __bch2_read_super(path, opts, sb, true);
 839 }
 840
 841 /* write superblock: */
 842
 843 static void write_super_endio(struct bio *bio)
 844 {
 845         struct bch_dev *ca = bio->bi_private;
 846
 847         /* XXX: return errors directly */
 848
 849         if (bch2_dev_io_err_on(bio->bi_status, ca,
 850                                bio_data_dir(bio)
 851                                ? BCH_MEMBER_ERROR_write
 852                                : BCH_MEMBER_ERROR_read,
 853                                "superblock %s error: %s",
 854                                bio_data_dir(bio) ? "write" : "read",
 855                                bch2_blk_status_to_str(bio->bi_status)))
 856                 ca->sb_write_error = 1;
 857
 858         closure_put(&ca->fs->sb_write);
 859         percpu_ref_put(&ca->io_ref);
 860 }
 861
 862 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 863 {
 864         struct bch_sb *sb = ca->disk_sb.sb;
 865         struct bio *bio = ca->disk_sb.bio;
 866
 867         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 868         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 869         bio->bi_end_io          = write_super_endio;
 870         bio->bi_private         = ca;
 871         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 872
 873         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 874                      bio_sectors(bio));
 875
 876         percpu_ref_get(&ca->io_ref);
 877         closure_bio_submit(bio, &c->sb_write);
 878 }
 879
 880 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 881 {
 882         struct bch_sb *sb = ca->disk_sb.sb;
 883         struct bio *bio = ca->disk_sb.bio;
 884
 885         sb->offset = sb->layout.sb_offset[idx];
 886
 887         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 888         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 889                                 null_nonce(), sb);
 890
 891         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 892         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 893         bio->bi_end_io          = write_super_endio;
 894         bio->bi_private         = ca;
 895         bch2_bio_map(bio, sb,
 896                      roundup((size_t) vstruct_bytes(sb),
 897                              bdev_logical_block_size(ca->disk_sb.bdev)));
 898
 899         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 900                      bio_sectors(bio));
 901
 902         percpu_ref_get(&ca->io_ref);
 903         closure_bio_submit(bio, &c->sb_write);
 904 }
 905
 906 int bch2_write_super(struct bch_fs *c)
 907 {
 908         struct closure *cl = &c->sb_write;
 909         struct printbuf err = PRINTBUF;
 910         unsigned sb = 0, nr_wrote;
 911         struct bch_devs_mask sb_written;
 912         bool wrote, can_mount_without_written, can_mount_with_written;
 913         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 914         int ret = 0;
 915
 916         trace_and_count(c, write_super, c, _RET_IP_);
 917
 918         if (c->opts.very_degraded)
 919                 degraded_flags |= BCH_FORCE_IF_LOST;
 920
 921         lockdep_assert_held(&c->sb_lock);
 922
 923         closure_init_stack(cl);
 924         memset(&sb_written, 0, sizeof(sb_written));
 925
 926         /* Make sure we're using the new magic numbers: */
 927         c->disk_sb.sb->magic = BCHFS_MAGIC;
 928         c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
 929
 930         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 931
 932         struct bch_sb_field_members_v2 *mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
 933         for_each_online_member(c, ca)
 934                 __bch2_members_v2_get_mut(mi, ca->dev_idx)->seq = c->disk_sb.sb->seq;
 935         c->disk_sb.sb->write_time = cpu_to_le64(ktime_get_real_seconds());
 936
 937         if (test_bit(BCH_FS_error, &c->flags))
 938                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 939         if (test_bit(BCH_FS_topology_error, &c->flags))
 940                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 941
 942         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 943
 944         bch2_sb_counters_from_cpu(c);
 945         bch2_sb_members_from_cpu(c);
 946         bch2_sb_members_cpy_v2_v1(&c->disk_sb);
 947         bch2_sb_errors_from_cpu(c);
 948         bch2_sb_downgrade_update(c);
 949
 950         for_each_online_member(c, ca)
 951                 bch2_sb_from_fs(c, ca);
 952
 953         for_each_online_member(c, ca) {
 954                 printbuf_reset(&err);
 955
 956                 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
 957                 if (ret) {
 958                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 959                         percpu_ref_put(&ca->io_ref);
 960                         goto out;
 961                 }
 962         }
 963
 964         if (c->opts.nochanges)
 965                 goto out;
 966
 967         /*
 968          * Defer writing the superblock until filesystem initialization is
 969          * complete - don't write out a partly initialized superblock:
 970          */
 971         if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
 972                 goto out;
 973
 974         if (le16_to_cpu(c->disk_sb.sb->version) > bcachefs_metadata_version_current) {
 975                 struct printbuf buf = PRINTBUF;
 976                 prt_printf(&buf, "attempting to write superblock that wasn't version downgraded (");
 977                 bch2_version_to_text(&buf, le16_to_cpu(c->disk_sb.sb->version));
 978                 prt_str(&buf, " > ");
 979                 bch2_version_to_text(&buf, bcachefs_metadata_version_current);
 980                 prt_str(&buf, ")");
 981                 bch2_fs_fatal_error(c, "%s", buf.buf);
 982                 printbuf_exit(&buf);
 983                 return -BCH_ERR_sb_not_downgraded;
 984         }
 985
 986         for_each_online_member(c, ca) {
 987                 __set_bit(ca->dev_idx, sb_written.d);
 988                 ca->sb_write_error = 0;
 989         }
 990
 991         for_each_online_member(c, ca)
 992                 read_back_super(c, ca);
 993         closure_sync(cl);
 994
 995         for_each_online_member(c, ca) {
 996                 if (ca->sb_write_error)
 997                         continue;
 998
 999                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
1000                         bch2_fs_fatal_error(c,
1001                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
1002                                 le64_to_cpu(ca->sb_read_scratch->seq),
1003                                 ca->disk_sb.seq);
1004                         percpu_ref_put(&ca->io_ref);
1005                         ret = -BCH_ERR_erofs_sb_err;
1006                         goto out;
1007                 }
1008
1009                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
1010                         bch2_fs_fatal_error(c,
1011                                 "Superblock modified by another process (seq %llu expected %llu)",
1012                                 le64_to_cpu(ca->sb_read_scratch->seq),
1013                                 ca->disk_sb.seq);
1014                         percpu_ref_put(&ca->io_ref);
1015                         ret = -BCH_ERR_erofs_sb_err;
1016                         goto out;
1017                 }
1018         }
1019
1020         do {
1021                 wrote = false;
1022                 for_each_online_member(c, ca)
1023                         if (!ca->sb_write_error &&
1024                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
1025                                 write_one_super(c, ca, sb);
1026                                 wrote = true;
1027                         }
1028                 closure_sync(cl);
1029                 sb++;
1030         } while (wrote);
1031
1032         for_each_online_member(c, ca) {
1033                 if (ca->sb_write_error)
1034                         __clear_bit(ca->dev_idx, sb_written.d);
1035                 else
1036                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
1037         }
1038
1039         nr_wrote = dev_mask_nr(&sb_written);
1040
1041         can_mount_with_written =
1042                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1043
1044         for (unsigned i = 0; i < ARRAY_SIZE(sb_written.d); i++)
1045                 sb_written.d[i] = ~sb_written.d[i];
1046
1047         can_mount_without_written =
1048                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
1049
1050         /*
1051          * If we would be able to mount _without_ the devices we successfully
1052          * wrote superblocks to, we weren't able to write to enough devices:
1053          *
1054          * Exception: if we can mount without the successes because we haven't
1055          * written anything (new filesystem), we continue if we'd be able to
1056          * mount with the devices we did successfully write to:
1057          */
1058         if (bch2_fs_fatal_err_on(!nr_wrote ||
1059                                  !can_mount_with_written ||
1060                                  (can_mount_without_written &&
1061                                   !can_mount_with_written), c,
1062                 "Unable to write superblock to sufficient devices (from %ps)",
1063                 (void *) _RET_IP_))
1064                 ret = -1;
1065 out:
1066         /* Make new options visible after they're persistent: */
1067         bch2_sb_update(c);
1068         printbuf_exit(&err);
1069         return ret;
1070 }
1071
1072 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
1073 {
1074         mutex_lock(&c->sb_lock);
1075         if (!(c->sb.features & (1ULL << feat))) {
1076                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
1077
1078                 bch2_write_super(c);
1079         }
1080         mutex_unlock(&c->sb_lock);
1081 }
1082
1083 /* Downgrade if superblock is at a higher version than currently supported: */
1084 bool bch2_check_version_downgrade(struct bch_fs *c)
1085 {
1086         bool ret = bcachefs_metadata_version_current < c->sb.version;
1087
1088         lockdep_assert_held(&c->sb_lock);
1089
1090         /*
1091          * Downgrade, if superblock is at a higher version than currently
1092          * supported:
1093          *
1094          * c->sb will be checked before we write the superblock, so update it as
1095          * well:
1096          */
1097         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current) {
1098                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1099                 c->sb.version_upgrade_complete = bcachefs_metadata_version_current;
1100         }
1101         if (c->sb.version > bcachefs_metadata_version_current) {
1102                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1103                 c->sb.version = bcachefs_metadata_version_current;
1104         }
1105         if (c->sb.version_min > bcachefs_metadata_version_current) {
1106                 c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
1107                 c->sb.version_min = bcachefs_metadata_version_current;
1108         }
1109         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1110         return ret;
1111 }
1112
1113 void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version)
1114 {
1115         lockdep_assert_held(&c->sb_lock);
1116
1117         if (BCH_VERSION_MAJOR(new_version) >
1118             BCH_VERSION_MAJOR(le16_to_cpu(c->disk_sb.sb->version)))
1119                 bch2_sb_field_resize(&c->disk_sb, downgrade, 0);
1120
1121         c->disk_sb.sb->version = cpu_to_le16(new_version);
1122         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1123 }
1124
1125 static int bch2_sb_ext_validate(struct bch_sb *sb, struct bch_sb_field *f,
1126                                 struct printbuf *err)
1127 {
1128         if (vstruct_bytes(f) < 88) {
1129                 prt_printf(err, "field too small (%zu < %u)", vstruct_bytes(f), 88);
1130                 return -BCH_ERR_invalid_sb_ext;
1131         }
1132
1133         return 0;
1134 }
1135
1136 static void bch2_sb_ext_to_text(struct printbuf *out, struct bch_sb *sb,
1137                                 struct bch_sb_field *f)
1138 {
1139         struct bch_sb_field_ext *e = field_to_type(f, ext);
1140
1141         prt_printf(out, "Recovery passes required:");
1142         prt_tab(out);
1143         prt_bitflags(out, bch2_recovery_passes,
1144                      bch2_recovery_passes_from_stable(le64_to_cpu(e->recovery_passes_required[0])));
1145         prt_newline(out);
1146
1147         unsigned long *errors_silent = kmalloc(sizeof(e->errors_silent), GFP_KERNEL);
1148         if (errors_silent) {
1149                 le_bitvector_to_cpu(errors_silent, (void *) e->errors_silent, sizeof(e->errors_silent) * 8);
1150
1151                 prt_printf(out, "Errors to silently fix:");
1152                 prt_tab(out);
1153                 prt_bitflags_vector(out, bch2_sb_error_strs, errors_silent, sizeof(e->errors_silent) * 8);
1154                 prt_newline(out);
1155
1156                 kfree(errors_silent);
1157         }
1158 }
1159
1160 static const struct bch_sb_field_ops bch_sb_field_ops_ext = {
1161         .validate       = bch2_sb_ext_validate,
1162         .to_text        = bch2_sb_ext_to_text,
1163 };
1164
1165 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1166 #define x(f, nr)                                        \
1167         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1168         BCH_SB_FIELDS()
1169 #undef x
1170 };
1171
1172 static const struct bch_sb_field_ops bch2_sb_field_null_ops;
1173
1174 static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
1175 {
1176         return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
1177                 ? bch2_sb_field_ops[type]
1178                 : &bch2_sb_field_null_ops;
1179 }
1180
1181 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1182                                   struct printbuf *err)
1183 {
1184         unsigned type = le32_to_cpu(f->type);
1185         struct printbuf field_err = PRINTBUF;
1186         const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1187         int ret;
1188
1189         ret = ops->validate ? ops->validate(sb, f, &field_err) : 0;
1190         if (ret) {
1191                 prt_printf(err, "Invalid superblock section %s: %s",
1192                            bch2_sb_fields[type], field_err.buf);
1193                 prt_newline(err);
1194                 bch2_sb_field_to_text(err, sb, f);
1195         }
1196
1197         printbuf_exit(&field_err);
1198         return ret;
1199 }
1200
1201 void __bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1202                              struct bch_sb_field *f)
1203 {
1204         unsigned type = le32_to_cpu(f->type);
1205         const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1206
1207         if (!out->nr_tabstops)
1208                 printbuf_tabstop_push(out, 32);
1209
1210         if (ops->to_text)
1211                 ops->to_text(out, sb, f);
1212 }
1213
1214 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1215                            struct bch_sb_field *f)
1216 {
1217         unsigned type = le32_to_cpu(f->type);
1218
1219         if (type < BCH_SB_FIELD_NR)
1220                 prt_printf(out, "%s", bch2_sb_fields[type]);
1221         else
1222                 prt_printf(out, "(unknown field %u)", type);
1223
1224         prt_printf(out, " (size %zu):", vstruct_bytes(f));
1225         prt_newline(out);
1226
1227         __bch2_sb_field_to_text(out, sb, f);
1228 }
1229
1230 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1231 {
1232         unsigned i;
1233
1234         prt_printf(out, "Type:                    %u", l->layout_type);
1235         prt_newline(out);
1236
1237         prt_str(out, "Superblock max size:     ");
1238         prt_units_u64(out, 512 << l->sb_max_size_bits);
1239         prt_newline(out);
1240
1241         prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1242         prt_newline(out);
1243
1244         prt_str(out, "Offsets:                 ");
1245         for (i = 0; i < l->nr_superblocks; i++) {
1246                 if (i)
1247                         prt_str(out, ", ");
1248                 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1249         }
1250         prt_newline(out);
1251 }
1252
1253 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1254                      bool print_layout, unsigned fields)
1255 {
1256         u64 fields_have = 0;
1257         unsigned nr_devices = 0;
1258
1259         if (!out->nr_tabstops)
1260                 printbuf_tabstop_push(out, 44);
1261
1262         for (int i = 0; i < sb->nr_devices; i++)
1263                 nr_devices += bch2_dev_exists(sb, i);
1264
1265         prt_printf(out, "External UUID:");
1266         prt_tab(out);
1267         pr_uuid(out, sb->user_uuid.b);
1268         prt_newline(out);
1269
1270         prt_printf(out, "Internal UUID:");
1271         prt_tab(out);
1272         pr_uuid(out, sb->uuid.b);
1273         prt_newline(out);
1274
1275         prt_printf(out, "Magic number:");
1276         prt_tab(out);
1277         pr_uuid(out, sb->magic.b);
1278         prt_newline(out);
1279
1280         prt_str(out, "Device index:");
1281         prt_tab(out);
1282         prt_printf(out, "%u", sb->dev_idx);
1283         prt_newline(out);
1284
1285         prt_str(out, "Label:");
1286         prt_tab(out);
1287         prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1288         prt_newline(out);
1289
1290         prt_str(out, "Version:");
1291         prt_tab(out);
1292         bch2_version_to_text(out, le16_to_cpu(sb->version));
1293         prt_newline(out);
1294
1295         prt_str(out, "Version upgrade complete:");
1296         prt_tab(out);
1297         bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
1298         prt_newline(out);
1299
1300         prt_printf(out, "Oldest version on disk:");
1301         prt_tab(out);
1302         bch2_version_to_text(out, le16_to_cpu(sb->version_min));
1303         prt_newline(out);
1304
1305         prt_printf(out, "Created:");
1306         prt_tab(out);
1307         if (sb->time_base_lo)
1308                 bch2_prt_datetime(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1309         else
1310                 prt_printf(out, "(not set)");
1311         prt_newline(out);
1312
1313         prt_printf(out, "Sequence number:");
1314         prt_tab(out);
1315         prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1316         prt_newline(out);
1317
1318         prt_printf(out, "Time of last write:");
1319         prt_tab(out);
1320         bch2_prt_datetime(out, le64_to_cpu(sb->write_time));
1321         prt_newline(out);
1322
1323         prt_printf(out, "Superblock size:");
1324         prt_tab(out);
1325         prt_units_u64(out, vstruct_bytes(sb));
1326         prt_str(out, "/");
1327         prt_units_u64(out, 512ULL << sb->layout.sb_max_size_bits);
1328         prt_newline(out);
1329
1330         prt_printf(out, "Clean:");
1331         prt_tab(out);
1332         prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1333         prt_newline(out);
1334
1335         prt_printf(out, "Devices:");
1336         prt_tab(out);
1337         prt_printf(out, "%u", nr_devices);
1338         prt_newline(out);
1339
1340         prt_printf(out, "Sections:");
1341         vstruct_for_each(sb, f)
1342                 fields_have |= 1 << le32_to_cpu(f->type);
1343         prt_tab(out);
1344         prt_bitflags(out, bch2_sb_fields, fields_have);
1345         prt_newline(out);
1346
1347         prt_printf(out, "Features:");
1348         prt_tab(out);
1349         prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1350         prt_newline(out);
1351
1352         prt_printf(out, "Compat features:");
1353         prt_tab(out);
1354         prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1355         prt_newline(out);
1356
1357         prt_newline(out);
1358         prt_printf(out, "Options:");
1359         prt_newline(out);
1360         printbuf_indent_add(out, 2);
1361         {
1362                 enum bch_opt_id id;
1363
1364                 for (id = 0; id < bch2_opts_nr; id++) {
1365                         const struct bch_option *opt = bch2_opt_table + id;
1366
1367                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1368                                 u64 v = bch2_opt_from_sb(sb, id);
1369
1370                                 prt_printf(out, "%s:", opt->attr.name);
1371                                 prt_tab(out);
1372                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1373                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1374                                 prt_newline(out);
1375                         }
1376                 }
1377         }
1378
1379         printbuf_indent_sub(out, 2);
1380
1381         if (print_layout) {
1382                 prt_newline(out);
1383                 prt_printf(out, "layout:");
1384                 prt_newline(out);
1385                 printbuf_indent_add(out, 2);
1386                 bch2_sb_layout_to_text(out, &sb->layout);
1387                 printbuf_indent_sub(out, 2);
1388         }
1389
1390         vstruct_for_each(sb, f)
1391                 if (fields & (1 << le32_to_cpu(f->type))) {
1392                         prt_newline(out);
1393                         bch2_sb_field_to_text(out, sb, f);
1394                 }
1395 }