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 "btree_update_interior.h"
   5 #include "buckets.h"
   6 #include "checksum.h"
   7 #include "counters.h"
   8 #include "disk_groups.h"
   9 #include "ec.h"
  10 #include "error.h"
  11 #include "io.h"
  12 #include "journal.h"
  13 #include "journal_io.h"
  14 #include "journal_sb.h"
  15 #include "journal_seq_blacklist.h"
  16 #include "recovery.h"
  17 #include "replicas.h"
  18 #include "quota.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 struct bch2_metadata_version {
  28         u16             version;
  29         const char      *name;
  30         u64             recovery_passes;
  31 };
  32
  33 static const struct bch2_metadata_version bch2_metadata_versions[] = {
  34 #define x(n, v, _recovery_passes) {             \
  35         .version = v,                           \
  36         .name = #n,                             \
  37         .recovery_passes = _recovery_passes,    \
  38 },
  39         BCH_METADATA_VERSIONS()
  40 #undef x
  41 };
  42
  43 void bch2_version_to_text(struct printbuf *out, unsigned v)
  44 {
  45         const char *str = "(unknown version)";
  46
  47         for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
  48                 if (bch2_metadata_versions[i].version == v) {
  49                         str = bch2_metadata_versions[i].name;
  50                         break;
  51                 }
  52
  53         prt_printf(out, "%u.%u: %s", BCH_VERSION_MAJOR(v), BCH_VERSION_MINOR(v), str);
  54 }
  55
  56 unsigned bch2_latest_compatible_version(unsigned v)
  57 {
  58         if (!BCH_VERSION_MAJOR(v))
  59                 return v;
  60
  61         for (unsigned i = 0; i < ARRAY_SIZE(bch2_metadata_versions); i++)
  62                 if (bch2_metadata_versions[i].version > v &&
  63                     BCH_VERSION_MAJOR(bch2_metadata_versions[i].version) ==
  64                     BCH_VERSION_MAJOR(v))
  65                         v = bch2_metadata_versions[i].version;
  66
  67         return v;
  68 }
  69
  70 u64 bch2_upgrade_recovery_passes(struct bch_fs *c,
  71                                  unsigned old_version,
  72                                  unsigned new_version)
  73 {
  74         u64 ret = 0;
  75
  76         for (const struct bch2_metadata_version *i = bch2_metadata_versions;
  77              i < bch2_metadata_versions + ARRAY_SIZE(bch2_metadata_versions);
  78              i++)
  79                 if (i->version > old_version && i->version <= new_version) {
  80                         if (i->recovery_passes & RECOVERY_PASS_ALL_FSCK)
  81                                 ret |= bch2_fsck_recovery_passes();
  82                         ret |= i->recovery_passes;
  83                 }
  84
  85         return ret &= ~RECOVERY_PASS_ALL_FSCK;
  86 }
  87
  88 const char * const bch2_sb_fields[] = {
  89 #define x(name, nr)     #name,
  90         BCH_SB_FIELDS()
  91 #undef x
  92         NULL
  93 };
  94
  95 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
  96                                   struct printbuf *);
  97
  98 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  99                                       enum bch_sb_field_type type)
 100 {
 101         struct bch_sb_field *f;
 102
 103         /* XXX: need locking around superblock to access optional fields */
 104
 105         vstruct_for_each(sb, f)
 106                 if (le32_to_cpu(f->type) == type)
 107                         return f;
 108         return NULL;
 109 }
 110
 111 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
 112                                                    struct bch_sb_field *f,
 113                                                    unsigned u64s)
 114 {
 115         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 116         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
 117
 118         BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
 119
 120         if (!f && !u64s) {
 121                 /* nothing to do: */
 122         } else if (!f) {
 123                 f = vstruct_last(sb->sb);
 124                 memset(f, 0, sizeof(u64) * u64s);
 125                 f->u64s = cpu_to_le32(u64s);
 126                 f->type = 0;
 127         } else {
 128                 void *src, *dst;
 129
 130                 src = vstruct_end(f);
 131
 132                 if (u64s) {
 133                         f->u64s = cpu_to_le32(u64s);
 134                         dst = vstruct_end(f);
 135                 } else {
 136                         dst = f;
 137                 }
 138
 139                 memmove(dst, src, vstruct_end(sb->sb) - src);
 140
 141                 if (dst > src)
 142                         memset(src, 0, dst - src);
 143         }
 144
 145         sb->sb->u64s = cpu_to_le32(sb_u64s);
 146
 147         return u64s ? f : NULL;
 148 }
 149
 150 void bch2_sb_field_delete(struct bch_sb_handle *sb,
 151                           enum bch_sb_field_type type)
 152 {
 153         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 154
 155         if (f)
 156                 __bch2_sb_field_resize(sb, f, 0);
 157 }
 158
 159 /* Superblock realloc/free: */
 160
 161 void bch2_free_super(struct bch_sb_handle *sb)
 162 {
 163         kfree(sb->bio);
 164         if (!IS_ERR_OR_NULL(sb->bdev))
 165                 blkdev_put(sb->bdev, sb->mode);
 166
 167         kfree(sb->sb);
 168         memset(sb, 0, sizeof(*sb));
 169 }
 170
 171 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 172 {
 173         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 174         size_t new_buffer_size;
 175         struct bch_sb *new_sb;
 176         struct bio *bio;
 177
 178         if (sb->bdev)
 179                 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
 180
 181         new_buffer_size = roundup_pow_of_two(new_bytes);
 182
 183         if (sb->sb && sb->buffer_size >= new_buffer_size)
 184                 return 0;
 185
 186         if (sb->have_layout) {
 187                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 188
 189                 if (new_bytes > max_bytes) {
 190                         pr_err("%pg: superblock too big: want %zu but have %llu",
 191                                sb->bdev, new_bytes, max_bytes);
 192                         return -BCH_ERR_ENOSPC_sb;
 193                 }
 194         }
 195
 196         if (sb->buffer_size >= new_buffer_size && sb->sb)
 197                 return 0;
 198
 199         if (dynamic_fault("bcachefs:add:super_realloc"))
 200                 return -BCH_ERR_ENOMEM_sb_realloc_injected;
 201
 202         if (sb->have_bio) {
 203                 unsigned nr_bvecs = DIV_ROUND_UP(new_buffer_size, PAGE_SIZE);
 204
 205                 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 206                 if (!bio)
 207                         return -BCH_ERR_ENOMEM_sb_bio_realloc;
 208
 209                 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
 210
 211                 kfree(sb->bio);
 212                 sb->bio = bio;
 213         }
 214
 215         new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 216         if (!new_sb)
 217                 return -BCH_ERR_ENOMEM_sb_buf_realloc;
 218
 219         sb->sb = new_sb;
 220         sb->buffer_size = new_buffer_size;
 221
 222         return 0;
 223 }
 224
 225 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 226                                           enum bch_sb_field_type type,
 227                                           unsigned u64s)
 228 {
 229         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 230         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 231         ssize_t d = -old_u64s + u64s;
 232
 233         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 234                 return NULL;
 235
 236         if (sb->fs_sb) {
 237                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 238                 struct bch_dev *ca;
 239                 unsigned i;
 240
 241                 lockdep_assert_held(&c->sb_lock);
 242
 243                 /* XXX: we're not checking that offline device have enough space */
 244
 245                 for_each_online_member(ca, c, i) {
 246                         struct bch_sb_handle *sb = &ca->disk_sb;
 247
 248                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 249                                 percpu_ref_put(&ca->ref);
 250                                 return NULL;
 251                         }
 252                 }
 253         }
 254
 255         f = bch2_sb_field_get(sb->sb, type);
 256         f = __bch2_sb_field_resize(sb, f, u64s);
 257         if (f)
 258                 f->type = cpu_to_le32(type);
 259         return f;
 260 }
 261
 262 /* Superblock validate: */
 263
 264 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
 265 {
 266         u64 offset, prev_offset, max_sectors;
 267         unsigned i;
 268
 269         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 270
 271         if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
 272             !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
 273                 prt_printf(out, "Not a bcachefs superblock layout");
 274                 return -BCH_ERR_invalid_sb_layout;
 275         }
 276
 277         if (layout->layout_type != 0) {
 278                 prt_printf(out, "Invalid superblock layout type %u",
 279                        layout->layout_type);
 280                 return -BCH_ERR_invalid_sb_layout_type;
 281         }
 282
 283         if (!layout->nr_superblocks) {
 284                 prt_printf(out, "Invalid superblock layout: no superblocks");
 285                 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 286         }
 287
 288         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
 289                 prt_printf(out, "Invalid superblock layout: too many superblocks");
 290                 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 291         }
 292
 293         max_sectors = 1 << layout->sb_max_size_bits;
 294
 295         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 296
 297         for (i = 1; i < layout->nr_superblocks; i++) {
 298                 offset = le64_to_cpu(layout->sb_offset[i]);
 299
 300                 if (offset < prev_offset + max_sectors) {
 301                         prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
 302                                "  (sb %u ends at %llu next starts at %llu",
 303                                i - 1, prev_offset + max_sectors, offset);
 304                         return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
 305                 }
 306                 prev_offset = offset;
 307         }
 308
 309         return 0;
 310 }
 311
 312 static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
 313 {
 314         u16 version             = le16_to_cpu(sb->version);
 315         u16 version_min         = le16_to_cpu(sb->version_min);
 316
 317         if (!bch2_version_compatible(version)) {
 318                 prt_str(out, "Unsupported superblock version ");
 319                 bch2_version_to_text(out, version);
 320                 prt_str(out, " (min ");
 321                 bch2_version_to_text(out, bcachefs_metadata_version_min);
 322                 prt_str(out, ", max ");
 323                 bch2_version_to_text(out, bcachefs_metadata_version_current);
 324                 prt_str(out, ")");
 325                 return -BCH_ERR_invalid_sb_version;
 326         }
 327
 328         if (!bch2_version_compatible(version_min)) {
 329                 prt_str(out, "Unsupported superblock version_min ");
 330                 bch2_version_to_text(out, version_min);
 331                 prt_str(out, " (min ");
 332                 bch2_version_to_text(out, bcachefs_metadata_version_min);
 333                 prt_str(out, ", max ");
 334                 bch2_version_to_text(out, bcachefs_metadata_version_current);
 335                 prt_str(out, ")");
 336                 return -BCH_ERR_invalid_sb_version;
 337         }
 338
 339         if (version_min > version) {
 340                 prt_str(out, "Bad minimum version ");
 341                 bch2_version_to_text(out, version_min);
 342                 prt_str(out, ", greater than version field ");
 343                 bch2_version_to_text(out, version);
 344                 return -BCH_ERR_invalid_sb_version;
 345         }
 346
 347         return 0;
 348 }
 349
 350 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 351                             int rw)
 352 {
 353         struct bch_sb *sb = disk_sb->sb;
 354         struct bch_sb_field *f;
 355         struct bch_sb_field_members *mi;
 356         enum bch_opt_id opt_id;
 357         u16 block_size;
 358         int ret;
 359
 360         ret = bch2_sb_compatible(sb, out);
 361         if (ret)
 362                 return ret;
 363
 364         if (sb->features[1] ||
 365             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
 366                 prt_printf(out, "Filesystem has incompatible features");
 367                 return -BCH_ERR_invalid_sb_features;
 368         }
 369
 370         block_size = le16_to_cpu(sb->block_size);
 371
 372         if (block_size > PAGE_SECTORS) {
 373                 prt_printf(out, "Block size too big (got %u, max %u)",
 374                        block_size, PAGE_SECTORS);
 375                 return -BCH_ERR_invalid_sb_block_size;
 376         }
 377
 378         if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
 379                 prt_printf(out, "Bad user UUID (got zeroes)");
 380                 return -BCH_ERR_invalid_sb_uuid;
 381         }
 382
 383         if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
 384                 prt_printf(out, "Bad intenal UUID (got zeroes)");
 385                 return -BCH_ERR_invalid_sb_uuid;
 386         }
 387
 388         if (!sb->nr_devices ||
 389             sb->nr_devices > BCH_SB_MEMBERS_MAX) {
 390                 prt_printf(out, "Bad number of member devices %u (max %u)",
 391                        sb->nr_devices, BCH_SB_MEMBERS_MAX);
 392                 return -BCH_ERR_invalid_sb_too_many_members;
 393         }
 394
 395         if (sb->dev_idx >= sb->nr_devices) {
 396                 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
 397                        sb->dev_idx, sb->nr_devices);
 398                 return -BCH_ERR_invalid_sb_dev_idx;
 399         }
 400
 401         if (!sb->time_precision ||
 402             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
 403                 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
 404                        le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
 405                 return -BCH_ERR_invalid_sb_time_precision;
 406         }
 407
 408         if (rw == READ) {
 409                 /*
 410                  * Been seeing a bug where these are getting inexplicably
 411                  * zeroed, so we're now validating them, but we have to be
 412                  * careful not to preven people's filesystems from mounting:
 413                  */
 414                 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 415                         SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 416                 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 417                         SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
 418
 419                 if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
 420                         SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
 421         }
 422
 423         for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
 424                 const struct bch_option *opt = bch2_opt_table + opt_id;
 425
 426                 if (opt->get_sb != BCH2_NO_SB_OPT) {
 427                         u64 v = bch2_opt_from_sb(sb, opt_id);
 428
 429                         prt_printf(out, "Invalid option ");
 430                         ret = bch2_opt_validate(opt, v, out);
 431                         if (ret)
 432                                 return ret;
 433
 434                         printbuf_reset(out);
 435                 }
 436         }
 437
 438         /* validate layout */
 439         ret = validate_sb_layout(&sb->layout, out);
 440         if (ret)
 441                 return ret;
 442
 443         vstruct_for_each(sb, f) {
 444                 if (!f->u64s) {
 445                         prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
 446                                le32_to_cpu(f->type));
 447                         return -BCH_ERR_invalid_sb_field_size;
 448                 }
 449
 450                 if (vstruct_next(f) > vstruct_last(sb)) {
 451                         prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
 452                                le32_to_cpu(f->type));
 453                         return -BCH_ERR_invalid_sb_field_size;
 454                 }
 455         }
 456
 457         /* members must be validated first: */
 458         mi = bch2_sb_get_members(sb);
 459         if (!mi) {
 460                 prt_printf(out, "Invalid superblock: member info area missing");
 461                 return -BCH_ERR_invalid_sb_members_missing;
 462         }
 463
 464         ret = bch2_sb_field_validate(sb, &mi->field, out);
 465         if (ret)
 466                 return ret;
 467
 468         vstruct_for_each(sb, f) {
 469                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 470                         continue;
 471
 472                 ret = bch2_sb_field_validate(sb, f, out);
 473                 if (ret)
 474                         return ret;
 475         }
 476
 477         return 0;
 478 }
 479
 480 /* device open: */
 481
 482 static void bch2_sb_update(struct bch_fs *c)
 483 {
 484         struct bch_sb *src = c->disk_sb.sb;
 485         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 486         struct bch_dev *ca;
 487         unsigned i;
 488
 489         lockdep_assert_held(&c->sb_lock);
 490
 491         c->sb.uuid              = src->uuid;
 492         c->sb.user_uuid         = src->user_uuid;
 493         c->sb.version           = le16_to_cpu(src->version);
 494         c->sb.version_min       = le16_to_cpu(src->version_min);
 495         c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
 496         c->sb.nr_devices        = src->nr_devices;
 497         c->sb.clean             = BCH_SB_CLEAN(src);
 498         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 499
 500         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 501         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 502
 503         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 504         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 505                                           c->sb.nsec_per_time_unit);
 506         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 507
 508         c->sb.features          = le64_to_cpu(src->features[0]);
 509         c->sb.compat            = le64_to_cpu(src->compat[0]);
 510
 511         for_each_member_device(ca, c, i)
 512                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 513 }
 514
 515 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 516 {
 517         struct bch_sb_field *src_f, *dst_f;
 518         struct bch_sb *dst = dst_handle->sb;
 519         unsigned i;
 520
 521         dst->version            = src->version;
 522         dst->version_min        = src->version_min;
 523         dst->seq                = src->seq;
 524         dst->uuid               = src->uuid;
 525         dst->user_uuid          = src->user_uuid;
 526         memcpy(dst->label,      src->label, sizeof(dst->label));
 527
 528         dst->block_size         = src->block_size;
 529         dst->nr_devices         = src->nr_devices;
 530
 531         dst->time_base_lo       = src->time_base_lo;
 532         dst->time_base_hi       = src->time_base_hi;
 533         dst->time_precision     = src->time_precision;
 534
 535         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 536         memcpy(dst->features,   src->features,  sizeof(dst->features));
 537         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 538
 539         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 540                 int d;
 541
 542                 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 543                         continue;
 544
 545                 src_f = bch2_sb_field_get(src, i);
 546                 dst_f = bch2_sb_field_get(dst, i);
 547
 548                 d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
 549                     (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
 550                 if (d > 0) {
 551                         int ret = bch2_sb_realloc(dst_handle, le32_to_cpu(dst_handle->sb->u64s) + d);
 552                         if (ret)
 553                                 return ret;
 554
 555                         dst = dst_handle->sb;
 556                         dst_f = bch2_sb_field_get(dst, i);
 557                 }
 558
 559                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 560                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 561
 562                 if (src_f)
 563                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 564         }
 565
 566         return 0;
 567 }
 568
 569 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 570 {
 571         int ret;
 572
 573         lockdep_assert_held(&c->sb_lock);
 574
 575         ret =   bch2_sb_realloc(&c->disk_sb, 0) ?:
 576                 __copy_super(&c->disk_sb, src) ?:
 577                 bch2_sb_replicas_to_cpu_replicas(c) ?:
 578                 bch2_sb_disk_groups_to_cpu(c);
 579         if (ret)
 580                 return ret;
 581
 582         bch2_sb_update(c);
 583         return 0;
 584 }
 585
 586 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 587 {
 588         return __copy_super(&ca->disk_sb, c->disk_sb.sb);
 589 }
 590
 591 /* read superblock: */
 592
 593 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 594 {
 595         struct bch_csum csum;
 596         size_t bytes;
 597         int ret;
 598 reread:
 599         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 600         sb->bio->bi_iter.bi_sector = offset;
 601         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 602
 603         ret = submit_bio_wait(sb->bio);
 604         if (ret) {
 605                 prt_printf(err, "IO error: %i", ret);
 606                 return ret;
 607         }
 608
 609         if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
 610             !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
 611                 prt_printf(err, "Not a bcachefs superblock");
 612                 return -BCH_ERR_invalid_sb_magic;
 613         }
 614
 615         ret = bch2_sb_compatible(sb->sb, err);
 616         if (ret)
 617                 return ret;
 618
 619         bytes = vstruct_bytes(sb->sb);
 620
 621         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 622                 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 623                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 624                 return -BCH_ERR_invalid_sb_too_big;
 625         }
 626
 627         if (bytes > sb->buffer_size) {
 628                 ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
 629                 if (ret)
 630                         return ret;
 631                 goto reread;
 632         }
 633
 634         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
 635                 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 636                 return -BCH_ERR_invalid_sb_csum_type;
 637         }
 638
 639         /* XXX: verify MACs */
 640         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 641                             null_nonce(), sb->sb);
 642
 643         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 644                 prt_printf(err, "bad checksum");
 645                 return -BCH_ERR_invalid_sb_csum;
 646         }
 647
 648         sb->seq = le64_to_cpu(sb->sb->seq);
 649
 650         return 0;
 651 }
 652
 653 int bch2_read_super(const char *path, struct bch_opts *opts,
 654                     struct bch_sb_handle *sb)
 655 {
 656         u64 offset = opt_get(*opts, sb);
 657         struct bch_sb_layout layout;
 658         struct printbuf err = PRINTBUF;
 659         __le64 *i;
 660         int ret;
 661 #ifndef __KERNEL__
 662 retry:
 663 #endif
 664         memset(sb, 0, sizeof(*sb));
 665         sb->mode        = FMODE_READ;
 666         sb->have_bio    = true;
 667
 668 #ifndef __KERNEL__
 669         if (opt_get(*opts, direct_io) == false)
 670                 sb->mode |= FMODE_BUFFERED;
 671 #endif
 672
 673         if (!opt_get(*opts, noexcl))
 674                 sb->mode |= FMODE_EXCL;
 675
 676         if (!opt_get(*opts, nochanges))
 677                 sb->mode |= FMODE_WRITE;
 678
 679         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 680         if (IS_ERR(sb->bdev) &&
 681             PTR_ERR(sb->bdev) == -EACCES &&
 682             opt_get(*opts, read_only)) {
 683                 sb->mode &= ~FMODE_WRITE;
 684
 685                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 686                 if (!IS_ERR(sb->bdev))
 687                         opt_set(*opts, nochanges, true);
 688         }
 689
 690         if (IS_ERR(sb->bdev)) {
 691                 ret = PTR_ERR(sb->bdev);
 692                 goto out;
 693         }
 694
 695         ret = bch2_sb_realloc(sb, 0);
 696         if (ret) {
 697                 prt_printf(&err, "error allocating memory for superblock");
 698                 goto err;
 699         }
 700
 701         if (bch2_fs_init_fault("read_super")) {
 702                 prt_printf(&err, "dynamic fault");
 703                 ret = -EFAULT;
 704                 goto err;
 705         }
 706
 707         ret = read_one_super(sb, offset, &err);
 708         if (!ret)
 709                 goto got_super;
 710
 711         if (opt_defined(*opts, sb))
 712                 goto err;
 713
 714         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 715                path, err.buf);
 716         printbuf_reset(&err);
 717
 718         /*
 719          * Error reading primary superblock - read location of backup
 720          * superblocks:
 721          */
 722         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 723         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 724         /*
 725          * use sb buffer to read layout, since sb buffer is page aligned but
 726          * layout won't be:
 727          */
 728         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 729
 730         ret = submit_bio_wait(sb->bio);
 731         if (ret) {
 732                 prt_printf(&err, "IO error: %i", ret);
 733                 goto err;
 734         }
 735
 736         memcpy(&layout, sb->sb, sizeof(layout));
 737         ret = validate_sb_layout(&layout, &err);
 738         if (ret)
 739                 goto err;
 740
 741         for (i = layout.sb_offset;
 742              i < layout.sb_offset + layout.nr_superblocks; i++) {
 743                 offset = le64_to_cpu(*i);
 744
 745                 if (offset == opt_get(*opts, sb))
 746                         continue;
 747
 748                 ret = read_one_super(sb, offset, &err);
 749                 if (!ret)
 750                         goto got_super;
 751         }
 752
 753         goto err;
 754
 755 got_super:
 756         if (le16_to_cpu(sb->sb->block_size) << 9 <
 757             bdev_logical_block_size(sb->bdev) &&
 758             opt_get(*opts, direct_io)) {
 759 #ifndef __KERNEL__
 760                 opt_set(*opts, direct_io, false);
 761                 bch2_free_super(sb);
 762                 goto retry;
 763 #endif
 764                 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
 765                        le16_to_cpu(sb->sb->block_size) << 9,
 766                        bdev_logical_block_size(sb->bdev));
 767                 ret = -BCH_ERR_block_size_too_small;
 768                 goto err;
 769         }
 770
 771         ret = 0;
 772         sb->have_layout = true;
 773
 774         ret = bch2_sb_validate(sb, &err, READ);
 775         if (ret) {
 776                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 777                        path, err.buf);
 778                 goto err_no_print;
 779         }
 780 out:
 781         printbuf_exit(&err);
 782         return ret;
 783 err:
 784         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 785                path, err.buf);
 786 err_no_print:
 787         bch2_free_super(sb);
 788         goto out;
 789 }
 790
 791 /* write superblock: */
 792
 793 static void write_super_endio(struct bio *bio)
 794 {
 795         struct bch_dev *ca = bio->bi_private;
 796
 797         /* XXX: return errors directly */
 798
 799         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 800                                bch2_blk_status_to_str(bio->bi_status)))
 801                 ca->sb_write_error = 1;
 802
 803         closure_put(&ca->fs->sb_write);
 804         percpu_ref_put(&ca->io_ref);
 805 }
 806
 807 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 808 {
 809         struct bch_sb *sb = ca->disk_sb.sb;
 810         struct bio *bio = ca->disk_sb.bio;
 811
 812         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 813         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 814         bio->bi_end_io          = write_super_endio;
 815         bio->bi_private         = ca;
 816         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 817
 818         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 819                      bio_sectors(bio));
 820
 821         percpu_ref_get(&ca->io_ref);
 822         closure_bio_submit(bio, &c->sb_write);
 823 }
 824
 825 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 826 {
 827         struct bch_sb *sb = ca->disk_sb.sb;
 828         struct bio *bio = ca->disk_sb.bio;
 829
 830         sb->offset = sb->layout.sb_offset[idx];
 831
 832         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 833         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 834                                 null_nonce(), sb);
 835
 836         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 837         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 838         bio->bi_end_io          = write_super_endio;
 839         bio->bi_private         = ca;
 840         bch2_bio_map(bio, sb,
 841                      roundup((size_t) vstruct_bytes(sb),
 842                              bdev_logical_block_size(ca->disk_sb.bdev)));
 843
 844         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 845                      bio_sectors(bio));
 846
 847         percpu_ref_get(&ca->io_ref);
 848         closure_bio_submit(bio, &c->sb_write);
 849 }
 850
 851 int bch2_write_super(struct bch_fs *c)
 852 {
 853         struct closure *cl = &c->sb_write;
 854         struct bch_dev *ca;
 855         struct printbuf err = PRINTBUF;
 856         unsigned i, sb = 0, nr_wrote;
 857         struct bch_devs_mask sb_written;
 858         bool wrote, can_mount_without_written, can_mount_with_written;
 859         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 860         int ret = 0;
 861
 862         trace_and_count(c, write_super, c, _RET_IP_);
 863
 864         if (c->opts.very_degraded)
 865                 degraded_flags |= BCH_FORCE_IF_LOST;
 866
 867         lockdep_assert_held(&c->sb_lock);
 868
 869         closure_init_stack(cl);
 870         memset(&sb_written, 0, sizeof(sb_written));
 871
 872         /* Make sure we're using the new magic numbers: */
 873         c->disk_sb.sb->magic = BCHFS_MAGIC;
 874         c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
 875
 876         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 877
 878         if (test_bit(BCH_FS_ERROR, &c->flags))
 879                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 880         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 881                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 882
 883         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 884
 885         bch2_sb_counters_from_cpu(c);
 886
 887         for_each_online_member(ca, c, i)
 888                 bch2_sb_from_fs(c, ca);
 889
 890         for_each_online_member(ca, c, i) {
 891                 printbuf_reset(&err);
 892
 893                 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
 894                 if (ret) {
 895                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 896                         percpu_ref_put(&ca->io_ref);
 897                         goto out;
 898                 }
 899         }
 900
 901         if (c->opts.nochanges)
 902                 goto out;
 903
 904         /*
 905          * Defer writing the superblock until filesystem initialization is
 906          * complete - don't write out a partly initialized superblock:
 907          */
 908         if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
 909                 goto out;
 910
 911         for_each_online_member(ca, c, i) {
 912                 __set_bit(ca->dev_idx, sb_written.d);
 913                 ca->sb_write_error = 0;
 914         }
 915
 916         for_each_online_member(ca, c, i)
 917                 read_back_super(c, ca);
 918         closure_sync(cl);
 919
 920         for_each_online_member(ca, c, i) {
 921                 if (ca->sb_write_error)
 922                         continue;
 923
 924                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 925                         bch2_fs_fatal_error(c,
 926                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 927                                 le64_to_cpu(ca->sb_read_scratch->seq),
 928                                 ca->disk_sb.seq);
 929                         percpu_ref_put(&ca->io_ref);
 930                         ret = -BCH_ERR_erofs_sb_err;
 931                         goto out;
 932                 }
 933
 934                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 935                         bch2_fs_fatal_error(c,
 936                                 "Superblock modified by another process (seq %llu expected %llu)",
 937                                 le64_to_cpu(ca->sb_read_scratch->seq),
 938                                 ca->disk_sb.seq);
 939                         percpu_ref_put(&ca->io_ref);
 940                         ret = -BCH_ERR_erofs_sb_err;
 941                         goto out;
 942                 }
 943         }
 944
 945         do {
 946                 wrote = false;
 947                 for_each_online_member(ca, c, i)
 948                         if (!ca->sb_write_error &&
 949                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 950                                 write_one_super(c, ca, sb);
 951                                 wrote = true;
 952                         }
 953                 closure_sync(cl);
 954                 sb++;
 955         } while (wrote);
 956
 957         for_each_online_member(ca, c, i) {
 958                 if (ca->sb_write_error)
 959                         __clear_bit(ca->dev_idx, sb_written.d);
 960                 else
 961                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 962         }
 963
 964         nr_wrote = dev_mask_nr(&sb_written);
 965
 966         can_mount_with_written =
 967                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 968
 969         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 970                 sb_written.d[i] = ~sb_written.d[i];
 971
 972         can_mount_without_written =
 973                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 974
 975         /*
 976          * If we would be able to mount _without_ the devices we successfully
 977          * wrote superblocks to, we weren't able to write to enough devices:
 978          *
 979          * Exception: if we can mount without the successes because we haven't
 980          * written anything (new filesystem), we continue if we'd be able to
 981          * mount with the devices we did successfully write to:
 982          */
 983         if (bch2_fs_fatal_err_on(!nr_wrote ||
 984                                  !can_mount_with_written ||
 985                                  (can_mount_without_written &&
 986                                   !can_mount_with_written), c,
 987                 "Unable to write superblock to sufficient devices (from %ps)",
 988                 (void *) _RET_IP_))
 989                 ret = -1;
 990 out:
 991         /* Make new options visible after they're persistent: */
 992         bch2_sb_update(c);
 993         printbuf_exit(&err);
 994         return ret;
 995 }
 996
 997 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 998 {
 999         mutex_lock(&c->sb_lock);
1000         if (!(c->sb.features & (1ULL << feat))) {
1001                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
1002
1003                 bch2_write_super(c);
1004         }
1005         mutex_unlock(&c->sb_lock);
1006 }
1007
1008 /* BCH_SB_FIELD_members: */
1009
1010 static int bch2_sb_members_validate(struct bch_sb *sb,
1011                                     struct bch_sb_field *f,
1012                                     struct printbuf *err)
1013 {
1014         struct bch_sb_field_members *mi = field_to_type(f, members);
1015         unsigned i;
1016
1017         if ((void *) (mi->members + sb->nr_devices) >
1018             vstruct_end(&mi->field)) {
1019                 prt_printf(err, "too many devices for section size");
1020                 return -BCH_ERR_invalid_sb_members;
1021         }
1022
1023         for (i = 0; i < sb->nr_devices; i++) {
1024                 struct bch_member *m = mi->members + i;
1025
1026                 if (!bch2_member_exists(m))
1027                         continue;
1028
1029                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
1030                         prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
1031                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
1032                         return -BCH_ERR_invalid_sb_members;
1033                 }
1034
1035                 if (le64_to_cpu(m->nbuckets) -
1036                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
1037                         prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
1038                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
1039                         return -BCH_ERR_invalid_sb_members;
1040                 }
1041
1042                 if (le16_to_cpu(m->bucket_size) <
1043                     le16_to_cpu(sb->block_size)) {
1044                         prt_printf(err, "device %u: bucket size %u smaller than block size %u",
1045                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
1046                         return -BCH_ERR_invalid_sb_members;
1047                 }
1048
1049                 if (le16_to_cpu(m->bucket_size) <
1050                     BCH_SB_BTREE_NODE_SIZE(sb)) {
1051                         prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
1052                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
1053                         return -BCH_ERR_invalid_sb_members;
1054                 }
1055         }
1056
1057         return 0;
1058 }
1059
1060 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
1061                                     struct bch_sb_field *f)
1062 {
1063         struct bch_sb_field_members *mi = field_to_type(f, members);
1064         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
1065         unsigned i;
1066
1067         for (i = 0; i < sb->nr_devices; i++) {
1068                 struct bch_member *m = mi->members + i;
1069                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
1070                 u64 bucket_size = le16_to_cpu(m->bucket_size);
1071                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1072
1073                 if (!bch2_member_exists(m))
1074                         continue;
1075
1076                 prt_printf(out, "Device:");
1077                 prt_tab(out);
1078                 prt_printf(out, "%u", i);
1079                 prt_newline(out);
1080
1081                 printbuf_indent_add(out, 2);
1082
1083                 prt_printf(out, "UUID:");
1084                 prt_tab(out);
1085                 pr_uuid(out, m->uuid.b);
1086                 prt_newline(out);
1087
1088                 prt_printf(out, "Size:");
1089                 prt_tab(out);
1090                 prt_units_u64(out, device_size << 9);
1091                 prt_newline(out);
1092
1093                 prt_printf(out, "Bucket size:");
1094                 prt_tab(out);
1095                 prt_units_u64(out, bucket_size << 9);
1096                 prt_newline(out);
1097
1098                 prt_printf(out, "First bucket:");
1099                 prt_tab(out);
1100                 prt_printf(out, "%u", le16_to_cpu(m->first_bucket));
1101                 prt_newline(out);
1102
1103                 prt_printf(out, "Buckets:");
1104                 prt_tab(out);
1105                 prt_printf(out, "%llu", le64_to_cpu(m->nbuckets));
1106                 prt_newline(out);
1107
1108                 prt_printf(out, "Last mount:");
1109                 prt_tab(out);
1110                 if (m->last_mount)
1111                         pr_time(out, le64_to_cpu(m->last_mount));
1112                 else
1113                         prt_printf(out, "(never)");
1114                 prt_newline(out);
1115
1116                 prt_printf(out, "State:");
1117                 prt_tab(out);
1118                 prt_printf(out, "%s",
1119                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1120                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1121                        : "unknown");
1122                 prt_newline(out);
1123
1124                 prt_printf(out, "Label:");
1125                 prt_tab(out);
1126                 if (BCH_MEMBER_GROUP(m)) {
1127                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1128
1129                         if (idx < disk_groups_nr(gi))
1130                                 prt_printf(out, "%s (%u)",
1131                                        gi->entries[idx].label, idx);
1132                         else
1133                                 prt_printf(out, "(bad disk labels section)");
1134                 } else {
1135                         prt_printf(out, "(none)");
1136                 }
1137                 prt_newline(out);
1138
1139                 prt_printf(out, "Data allowed:");
1140                 prt_tab(out);
1141                 if (BCH_MEMBER_DATA_ALLOWED(m))
1142                         prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(m));
1143                 else
1144                         prt_printf(out, "(none)");
1145                 prt_newline(out);
1146
1147                 prt_printf(out, "Has data:");
1148                 prt_tab(out);
1149                 if (data_have)
1150                         prt_bitflags(out, bch2_data_types, data_have);
1151                 else
1152                         prt_printf(out, "(none)");
1153                 prt_newline(out);
1154
1155                 prt_printf(out, "Discard:");
1156                 prt_tab(out);
1157                 prt_printf(out, "%llu", BCH_MEMBER_DISCARD(m));
1158                 prt_newline(out);
1159
1160                 prt_printf(out, "Freespace initialized:");
1161                 prt_tab(out);
1162                 prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1163                 prt_newline(out);
1164
1165                 printbuf_indent_sub(out, 2);
1166         }
1167 }
1168
1169 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1170         .validate       = bch2_sb_members_validate,
1171         .to_text        = bch2_sb_members_to_text,
1172 };
1173
1174 /* BCH_SB_FIELD_crypt: */
1175
1176 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1177                                   struct bch_sb_field *f,
1178                                   struct printbuf *err)
1179 {
1180         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1181
1182         if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1183                 prt_printf(err, "wrong size (got %zu should be %zu)",
1184                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1185                 return -BCH_ERR_invalid_sb_crypt;
1186         }
1187
1188         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1189                 prt_printf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1190                 return -BCH_ERR_invalid_sb_crypt;
1191         }
1192
1193         return 0;
1194 }
1195
1196 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1197                                   struct bch_sb_field *f)
1198 {
1199         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1200
1201         prt_printf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1202         prt_newline(out);
1203         prt_printf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1204         prt_newline(out);
1205         prt_printf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1206         prt_newline(out);
1207         prt_printf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1208         prt_newline(out);
1209 }
1210
1211 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1212         .validate       = bch2_sb_crypt_validate,
1213         .to_text        = bch2_sb_crypt_to_text,
1214 };
1215
1216 /* BCH_SB_FIELD_clean: */
1217
1218 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1219 {
1220         struct jset_entry *entry;
1221         int ret;
1222
1223         for (entry = clean->start;
1224              entry < (struct jset_entry *) vstruct_end(&clean->field);
1225              entry = vstruct_next(entry)) {
1226                 ret = bch2_journal_entry_validate(c, NULL, entry,
1227                                                   le16_to_cpu(c->disk_sb.sb->version),
1228                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1229                                                   write);
1230                 if (ret)
1231                         return ret;
1232         }
1233
1234         return 0;
1235 }
1236
1237 /* Downgrade if superblock is at a higher version than currently supported: */
1238 void bch2_sb_maybe_downgrade(struct bch_fs *c)
1239 {
1240         lockdep_assert_held(&c->sb_lock);
1241
1242         /*
1243          * Downgrade, if superblock is at a higher version than currently
1244          * supported:
1245          */
1246         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
1247                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1248         if (c->sb.version > bcachefs_metadata_version_current)
1249                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1250         if (c->sb.version_min > bcachefs_metadata_version_current)
1251                 c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
1252         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1253 }
1254
1255 void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version)
1256 {
1257         lockdep_assert_held(&c->sb_lock);
1258
1259         c->disk_sb.sb->version = cpu_to_le16(new_version);
1260         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1261 }
1262
1263 int bch2_fs_mark_dirty(struct bch_fs *c)
1264 {
1265         int ret;
1266
1267         /*
1268          * Unconditionally write superblock, to verify it hasn't changed before
1269          * we go rw:
1270          */
1271
1272         mutex_lock(&c->sb_lock);
1273         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1274
1275         bch2_sb_maybe_downgrade(c);
1276         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1277
1278         ret = bch2_write_super(c);
1279         mutex_unlock(&c->sb_lock);
1280
1281         return ret;
1282 }
1283
1284 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1285 {
1286         struct jset_entry *entry = *end;
1287         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1288
1289         memset(entry, 0, u64s * sizeof(u64));
1290         /*
1291          * The u64s field counts from the start of data, ignoring the shared
1292          * fields.
1293          */
1294         entry->u64s = cpu_to_le16(u64s - 1);
1295
1296         *end = vstruct_next(*end);
1297         return entry;
1298 }
1299
1300 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1301                                            struct jset_entry **end,
1302                                            u64 journal_seq)
1303 {
1304         struct bch_dev *ca;
1305         unsigned i, dev;
1306
1307         percpu_down_read(&c->mark_lock);
1308
1309         if (!journal_seq) {
1310                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1311                         bch2_fs_usage_acc_to_base(c, i);
1312         } else {
1313                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1314         }
1315
1316         {
1317                 struct jset_entry_usage *u =
1318                         container_of(jset_entry_init(end, sizeof(*u)),
1319                                      struct jset_entry_usage, entry);
1320
1321                 u->entry.type   = BCH_JSET_ENTRY_usage;
1322                 u->entry.btree_id = BCH_FS_USAGE_inodes;
1323                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1324         }
1325
1326         {
1327                 struct jset_entry_usage *u =
1328                         container_of(jset_entry_init(end, sizeof(*u)),
1329                                      struct jset_entry_usage, entry);
1330
1331                 u->entry.type   = BCH_JSET_ENTRY_usage;
1332                 u->entry.btree_id = BCH_FS_USAGE_key_version;
1333                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1334         }
1335
1336         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1337                 struct jset_entry_usage *u =
1338                         container_of(jset_entry_init(end, sizeof(*u)),
1339                                      struct jset_entry_usage, entry);
1340
1341                 u->entry.type   = BCH_JSET_ENTRY_usage;
1342                 u->entry.btree_id = BCH_FS_USAGE_reserved;
1343                 u->entry.level  = i;
1344                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1345         }
1346
1347         for (i = 0; i < c->replicas.nr; i++) {
1348                 struct bch_replicas_entry *e =
1349                         cpu_replicas_entry(&c->replicas, i);
1350                 struct jset_entry_data_usage *u =
1351                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1352                                      struct jset_entry_data_usage, entry);
1353
1354                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1355                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1356                 unsafe_memcpy(&u->r, e, replicas_entry_bytes(e),
1357                               "embedded variable length struct");
1358         }
1359
1360         for_each_member_device(ca, c, dev) {
1361                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1362                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1363                 struct jset_entry_dev_usage *u =
1364                         container_of(jset_entry_init(end, b),
1365                                      struct jset_entry_dev_usage, entry);
1366
1367                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1368                 u->dev = cpu_to_le32(dev);
1369                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1370
1371                 for (i = 0; i < BCH_DATA_NR; i++) {
1372                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1373                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1374                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1375                 }
1376         }
1377
1378         percpu_up_read(&c->mark_lock);
1379
1380         for (i = 0; i < 2; i++) {
1381                 struct jset_entry_clock *clock =
1382                         container_of(jset_entry_init(end, sizeof(*clock)),
1383                                      struct jset_entry_clock, entry);
1384
1385                 clock->entry.type = BCH_JSET_ENTRY_clock;
1386                 clock->rw       = i;
1387                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1388         }
1389 }
1390
1391 void bch2_fs_mark_clean(struct bch_fs *c)
1392 {
1393         struct bch_sb_field_clean *sb_clean;
1394         struct jset_entry *entry;
1395         unsigned u64s;
1396         int ret;
1397
1398         mutex_lock(&c->sb_lock);
1399         if (BCH_SB_CLEAN(c->disk_sb.sb))
1400                 goto out;
1401
1402         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1403
1404         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1405         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1406         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1407         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1408
1409         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1410
1411         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1412         if (!sb_clean) {
1413                 bch_err(c, "error resizing superblock while setting filesystem clean");
1414                 goto out;
1415         }
1416
1417         sb_clean->flags         = 0;
1418         sb_clean->journal_seq   = cpu_to_le64(atomic64_read(&c->journal.seq));
1419
1420         /* Trying to catch outstanding bug: */
1421         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1422
1423         entry = sb_clean->start;
1424         bch2_journal_super_entries_add_common(c, &entry, 0);
1425         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1426         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1427
1428         memset(entry, 0,
1429                vstruct_end(&sb_clean->field) - (void *) entry);
1430
1431         /*
1432          * this should be in the write path, and we should be validating every
1433          * superblock section:
1434          */
1435         ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1436         if (ret) {
1437                 bch_err(c, "error writing marking filesystem clean: validate error");
1438                 goto out;
1439         }
1440
1441         bch2_write_super(c);
1442 out:
1443         mutex_unlock(&c->sb_lock);
1444 }
1445
1446 static int bch2_sb_clean_validate(struct bch_sb *sb,
1447                                   struct bch_sb_field *f,
1448                                   struct printbuf *err)
1449 {
1450         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1451
1452         if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1453                 prt_printf(err, "wrong size (got %zu should be %zu)",
1454                        vstruct_bytes(&clean->field), sizeof(*clean));
1455                 return -BCH_ERR_invalid_sb_clean;
1456         }
1457
1458         return 0;
1459 }
1460
1461 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1462                                   struct bch_sb_field *f)
1463 {
1464         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1465         struct jset_entry *entry;
1466
1467         prt_printf(out, "flags:          %x",   le32_to_cpu(clean->flags));
1468         prt_newline(out);
1469         prt_printf(out, "journal_seq:    %llu", le64_to_cpu(clean->journal_seq));
1470         prt_newline(out);
1471
1472         for (entry = clean->start;
1473              entry != vstruct_end(&clean->field);
1474              entry = vstruct_next(entry)) {
1475                 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1476                     !entry->u64s)
1477                         continue;
1478
1479                 bch2_journal_entry_to_text(out, NULL, entry);
1480                 prt_newline(out);
1481         }
1482 }
1483
1484 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1485         .validate       = bch2_sb_clean_validate,
1486         .to_text        = bch2_sb_clean_to_text,
1487 };
1488
1489 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1490 #define x(f, nr)                                        \
1491         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1492         BCH_SB_FIELDS()
1493 #undef x
1494 };
1495
1496 static const struct bch_sb_field_ops bch2_sb_field_null_ops;
1497
1498 static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
1499 {
1500         return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
1501                 ? bch2_sb_field_ops[type]
1502                 : &bch2_sb_field_null_ops;
1503 }
1504
1505 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1506                                   struct printbuf *err)
1507 {
1508         unsigned type = le32_to_cpu(f->type);
1509         struct printbuf field_err = PRINTBUF;
1510         const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1511         int ret;
1512
1513         ret = ops->validate ? ops->validate(sb, f, &field_err) : 0;
1514         if (ret) {
1515                 prt_printf(err, "Invalid superblock section %s: %s",
1516                            bch2_sb_fields[type], field_err.buf);
1517                 prt_newline(err);
1518                 bch2_sb_field_to_text(err, sb, f);
1519         }
1520
1521         printbuf_exit(&field_err);
1522         return ret;
1523 }
1524
1525 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1526                            struct bch_sb_field *f)
1527 {
1528         unsigned type = le32_to_cpu(f->type);
1529         const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1530
1531         if (!out->nr_tabstops)
1532                 printbuf_tabstop_push(out, 32);
1533
1534         if (type < BCH_SB_FIELD_NR)
1535                 prt_printf(out, "%s", bch2_sb_fields[type]);
1536         else
1537                 prt_printf(out, "(unknown field %u)", type);
1538
1539         prt_printf(out, " (size %zu):", vstruct_bytes(f));
1540         prt_newline(out);
1541
1542         if (ops->to_text) {
1543                 printbuf_indent_add(out, 2);
1544                 ops->to_text(out, sb, f);
1545                 printbuf_indent_sub(out, 2);
1546         }
1547 }
1548
1549 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1550 {
1551         unsigned i;
1552
1553         prt_printf(out, "Type:                    %u", l->layout_type);
1554         prt_newline(out);
1555
1556         prt_str(out, "Superblock max size:     ");
1557         prt_units_u64(out, 512 << l->sb_max_size_bits);
1558         prt_newline(out);
1559
1560         prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1561         prt_newline(out);
1562
1563         prt_str(out, "Offsets:                 ");
1564         for (i = 0; i < l->nr_superblocks; i++) {
1565                 if (i)
1566                         prt_str(out, ", ");
1567                 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1568         }
1569         prt_newline(out);
1570 }
1571
1572 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1573                      bool print_layout, unsigned fields)
1574 {
1575         struct bch_sb_field_members *mi;
1576         struct bch_sb_field *f;
1577         u64 fields_have = 0;
1578         unsigned nr_devices = 0;
1579
1580         if (!out->nr_tabstops)
1581                 printbuf_tabstop_push(out, 44);
1582
1583         mi = bch2_sb_get_members(sb);
1584         if (mi) {
1585                 struct bch_member *m;
1586
1587                 for (m = mi->members;
1588                      m < mi->members + sb->nr_devices;
1589                      m++)
1590                         nr_devices += bch2_member_exists(m);
1591         }
1592
1593         prt_printf(out, "External UUID:");
1594         prt_tab(out);
1595         pr_uuid(out, sb->user_uuid.b);
1596         prt_newline(out);
1597
1598         prt_printf(out, "Internal UUID:");
1599         prt_tab(out);
1600         pr_uuid(out, sb->uuid.b);
1601         prt_newline(out);
1602
1603         prt_str(out, "Device index:");
1604         prt_tab(out);
1605         prt_printf(out, "%u", sb->dev_idx);
1606         prt_newline(out);
1607
1608         prt_str(out, "Label:");
1609         prt_tab(out);
1610         prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1611         prt_newline(out);
1612
1613         prt_str(out, "Version:");
1614         prt_tab(out);
1615         bch2_version_to_text(out, le16_to_cpu(sb->version));
1616         prt_newline(out);
1617
1618         prt_str(out, "Version upgrade complete:");
1619         prt_tab(out);
1620         bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
1621         prt_newline(out);
1622
1623         prt_printf(out, "Oldest version on disk:");
1624         prt_tab(out);
1625         bch2_version_to_text(out, le16_to_cpu(sb->version_min));
1626         prt_newline(out);
1627
1628         prt_printf(out, "Created:");
1629         prt_tab(out);
1630         if (sb->time_base_lo)
1631                 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1632         else
1633                 prt_printf(out, "(not set)");
1634         prt_newline(out);
1635
1636         prt_printf(out, "Sequence number:");
1637         prt_tab(out);
1638         prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1639         prt_newline(out);
1640
1641         prt_printf(out, "Superblock size:");
1642         prt_tab(out);
1643         prt_printf(out, "%zu", vstruct_bytes(sb));
1644         prt_newline(out);
1645
1646         prt_printf(out, "Clean:");
1647         prt_tab(out);
1648         prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1649         prt_newline(out);
1650
1651         prt_printf(out, "Devices:");
1652         prt_tab(out);
1653         prt_printf(out, "%u", nr_devices);
1654         prt_newline(out);
1655
1656         prt_printf(out, "Sections:");
1657         vstruct_for_each(sb, f)
1658                 fields_have |= 1 << le32_to_cpu(f->type);
1659         prt_tab(out);
1660         prt_bitflags(out, bch2_sb_fields, fields_have);
1661         prt_newline(out);
1662
1663         prt_printf(out, "Features:");
1664         prt_tab(out);
1665         prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1666         prt_newline(out);
1667
1668         prt_printf(out, "Compat features:");
1669         prt_tab(out);
1670         prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1671         prt_newline(out);
1672
1673         prt_newline(out);
1674         prt_printf(out, "Options:");
1675         prt_newline(out);
1676         printbuf_indent_add(out, 2);
1677         {
1678                 enum bch_opt_id id;
1679
1680                 for (id = 0; id < bch2_opts_nr; id++) {
1681                         const struct bch_option *opt = bch2_opt_table + id;
1682
1683                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1684                                 u64 v = bch2_opt_from_sb(sb, id);
1685
1686                                 prt_printf(out, "%s:", opt->attr.name);
1687                                 prt_tab(out);
1688                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1689                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1690                                 prt_newline(out);
1691                         }
1692                 }
1693         }
1694
1695         printbuf_indent_sub(out, 2);
1696
1697         if (print_layout) {
1698                 prt_newline(out);
1699                 prt_printf(out, "layout:");
1700                 prt_newline(out);
1701                 printbuf_indent_add(out, 2);
1702                 bch2_sb_layout_to_text(out, &sb->layout);
1703                 printbuf_indent_sub(out, 2);
1704         }
1705
1706         vstruct_for_each(sb, f)
1707                 if (fields & (1 << le32_to_cpu(f->type))) {
1708                         prt_newline(out);
1709                         bch2_sb_field_to_text(out, sb, f);
1710                 }
1711 }