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 inline void __bch2_sb_layout_size_assert(void)
 265 {
 266         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 267 }
 268
 269 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
 270 {
 271         u64 offset, prev_offset, max_sectors;
 272         unsigned i;
 273
 274         if (!uuid_equal(&layout->magic, &BCACHE_MAGIC) &&
 275             !uuid_equal(&layout->magic, &BCHFS_MAGIC)) {
 276                 prt_printf(out, "Not a bcachefs superblock layout");
 277                 return -BCH_ERR_invalid_sb_layout;
 278         }
 279
 280         if (layout->layout_type != 0) {
 281                 prt_printf(out, "Invalid superblock layout type %u",
 282                        layout->layout_type);
 283                 return -BCH_ERR_invalid_sb_layout_type;
 284         }
 285
 286         if (!layout->nr_superblocks) {
 287                 prt_printf(out, "Invalid superblock layout: no superblocks");
 288                 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 289         }
 290
 291         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
 292                 prt_printf(out, "Invalid superblock layout: too many superblocks");
 293                 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 294         }
 295
 296         max_sectors = 1 << layout->sb_max_size_bits;
 297
 298         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 299
 300         for (i = 1; i < layout->nr_superblocks; i++) {
 301                 offset = le64_to_cpu(layout->sb_offset[i]);
 302
 303                 if (offset < prev_offset + max_sectors) {
 304                         prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
 305                                "  (sb %u ends at %llu next starts at %llu",
 306                                i - 1, prev_offset + max_sectors, offset);
 307                         return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
 308                 }
 309                 prev_offset = offset;
 310         }
 311
 312         return 0;
 313 }
 314
 315 static int bch2_sb_compatible(struct bch_sb *sb, struct printbuf *out)
 316 {
 317         u16 version             = le16_to_cpu(sb->version);
 318         u16 version_min         = le16_to_cpu(sb->version_min);
 319
 320         if (!bch2_version_compatible(version)) {
 321                 prt_str(out, "Unsupported superblock version ");
 322                 bch2_version_to_text(out, version);
 323                 prt_str(out, " (min ");
 324                 bch2_version_to_text(out, bcachefs_metadata_version_min);
 325                 prt_str(out, ", max ");
 326                 bch2_version_to_text(out, bcachefs_metadata_version_current);
 327                 prt_str(out, ")");
 328                 return -BCH_ERR_invalid_sb_version;
 329         }
 330
 331         if (!bch2_version_compatible(version_min)) {
 332                 prt_str(out, "Unsupported superblock version_min ");
 333                 bch2_version_to_text(out, version_min);
 334                 prt_str(out, " (min ");
 335                 bch2_version_to_text(out, bcachefs_metadata_version_min);
 336                 prt_str(out, ", max ");
 337                 bch2_version_to_text(out, bcachefs_metadata_version_current);
 338                 prt_str(out, ")");
 339                 return -BCH_ERR_invalid_sb_version;
 340         }
 341
 342         if (version_min > version) {
 343                 prt_str(out, "Bad minimum version ");
 344                 bch2_version_to_text(out, version_min);
 345                 prt_str(out, ", greater than version field ");
 346                 bch2_version_to_text(out, version);
 347                 return -BCH_ERR_invalid_sb_version;
 348         }
 349
 350         return 0;
 351 }
 352
 353 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 354                             int rw)
 355 {
 356         struct bch_sb *sb = disk_sb->sb;
 357         struct bch_sb_field *f;
 358         struct bch_sb_field_members *mi;
 359         enum bch_opt_id opt_id;
 360         u16 block_size;
 361         int ret;
 362
 363         ret = bch2_sb_compatible(sb, out);
 364         if (ret)
 365                 return ret;
 366
 367         if (sb->features[1] ||
 368             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
 369                 prt_printf(out, "Filesystem has incompatible features");
 370                 return -BCH_ERR_invalid_sb_features;
 371         }
 372
 373         block_size = le16_to_cpu(sb->block_size);
 374
 375         if (block_size > PAGE_SECTORS) {
 376                 prt_printf(out, "Block size too big (got %u, max %u)",
 377                        block_size, PAGE_SECTORS);
 378                 return -BCH_ERR_invalid_sb_block_size;
 379         }
 380
 381         if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid))) {
 382                 prt_printf(out, "Bad user UUID (got zeroes)");
 383                 return -BCH_ERR_invalid_sb_uuid;
 384         }
 385
 386         if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid))) {
 387                 prt_printf(out, "Bad intenal UUID (got zeroes)");
 388                 return -BCH_ERR_invalid_sb_uuid;
 389         }
 390
 391         if (!sb->nr_devices ||
 392             sb->nr_devices > BCH_SB_MEMBERS_MAX) {
 393                 prt_printf(out, "Bad number of member devices %u (max %u)",
 394                        sb->nr_devices, BCH_SB_MEMBERS_MAX);
 395                 return -BCH_ERR_invalid_sb_too_many_members;
 396         }
 397
 398         if (sb->dev_idx >= sb->nr_devices) {
 399                 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
 400                        sb->dev_idx, sb->nr_devices);
 401                 return -BCH_ERR_invalid_sb_dev_idx;
 402         }
 403
 404         if (!sb->time_precision ||
 405             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
 406                 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
 407                        le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
 408                 return -BCH_ERR_invalid_sb_time_precision;
 409         }
 410
 411         if (rw == READ) {
 412                 /*
 413                  * Been seeing a bug where these are getting inexplicably
 414                  * zeroed, so we're now validating them, but we have to be
 415                  * careful not to preven people's filesystems from mounting:
 416                  */
 417                 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 418                         SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 419                 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 420                         SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
 421
 422                 if (!BCH_SB_VERSION_UPGRADE_COMPLETE(sb))
 423                         SET_BCH_SB_VERSION_UPGRADE_COMPLETE(sb, le16_to_cpu(sb->version));
 424         }
 425
 426         for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
 427                 const struct bch_option *opt = bch2_opt_table + opt_id;
 428
 429                 if (opt->get_sb != BCH2_NO_SB_OPT) {
 430                         u64 v = bch2_opt_from_sb(sb, opt_id);
 431
 432                         prt_printf(out, "Invalid option ");
 433                         ret = bch2_opt_validate(opt, v, out);
 434                         if (ret)
 435                                 return ret;
 436
 437                         printbuf_reset(out);
 438                 }
 439         }
 440
 441         /* validate layout */
 442         ret = validate_sb_layout(&sb->layout, out);
 443         if (ret)
 444                 return ret;
 445
 446         vstruct_for_each(sb, f) {
 447                 if (!f->u64s) {
 448                         prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
 449                                le32_to_cpu(f->type));
 450                         return -BCH_ERR_invalid_sb_field_size;
 451                 }
 452
 453                 if (vstruct_next(f) > vstruct_last(sb)) {
 454                         prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
 455                                le32_to_cpu(f->type));
 456                         return -BCH_ERR_invalid_sb_field_size;
 457                 }
 458         }
 459
 460         /* members must be validated first: */
 461         mi = bch2_sb_get_members(sb);
 462         if (!mi) {
 463                 prt_printf(out, "Invalid superblock: member info area missing");
 464                 return -BCH_ERR_invalid_sb_members_missing;
 465         }
 466
 467         ret = bch2_sb_field_validate(sb, &mi->field, out);
 468         if (ret)
 469                 return ret;
 470
 471         vstruct_for_each(sb, f) {
 472                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 473                         continue;
 474
 475                 ret = bch2_sb_field_validate(sb, f, out);
 476                 if (ret)
 477                         return ret;
 478         }
 479
 480         return 0;
 481 }
 482
 483 /* device open: */
 484
 485 static void bch2_sb_update(struct bch_fs *c)
 486 {
 487         struct bch_sb *src = c->disk_sb.sb;
 488         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 489         struct bch_dev *ca;
 490         unsigned i;
 491
 492         lockdep_assert_held(&c->sb_lock);
 493
 494         c->sb.uuid              = src->uuid;
 495         c->sb.user_uuid         = src->user_uuid;
 496         c->sb.version           = le16_to_cpu(src->version);
 497         c->sb.version_min       = le16_to_cpu(src->version_min);
 498         c->sb.version_upgrade_complete = BCH_SB_VERSION_UPGRADE_COMPLETE(src);
 499         c->sb.nr_devices        = src->nr_devices;
 500         c->sb.clean             = BCH_SB_CLEAN(src);
 501         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 502
 503         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 504         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 505
 506         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 507         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 508                                           c->sb.nsec_per_time_unit);
 509         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 510
 511         c->sb.features          = le64_to_cpu(src->features[0]);
 512         c->sb.compat            = le64_to_cpu(src->compat[0]);
 513
 514         for_each_member_device(ca, c, i)
 515                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 516 }
 517
 518 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 519 {
 520         struct bch_sb_field *src_f, *dst_f;
 521         struct bch_sb *dst = dst_handle->sb;
 522         unsigned i;
 523
 524         dst->version            = src->version;
 525         dst->version_min        = src->version_min;
 526         dst->seq                = src->seq;
 527         dst->uuid               = src->uuid;
 528         dst->user_uuid          = src->user_uuid;
 529         memcpy(dst->label,      src->label, sizeof(dst->label));
 530
 531         dst->block_size         = src->block_size;
 532         dst->nr_devices         = src->nr_devices;
 533
 534         dst->time_base_lo       = src->time_base_lo;
 535         dst->time_base_hi       = src->time_base_hi;
 536         dst->time_precision     = src->time_precision;
 537
 538         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 539         memcpy(dst->features,   src->features,  sizeof(dst->features));
 540         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 541
 542         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 543                 int d;
 544
 545                 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 546                         continue;
 547
 548                 src_f = bch2_sb_field_get(src, i);
 549                 dst_f = bch2_sb_field_get(dst, i);
 550
 551                 d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
 552                     (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
 553                 if (d > 0) {
 554                         int ret = bch2_sb_realloc(dst_handle, le32_to_cpu(dst_handle->sb->u64s) + d);
 555                         if (ret)
 556                                 return ret;
 557
 558                         dst = dst_handle->sb;
 559                         dst_f = bch2_sb_field_get(dst, i);
 560                 }
 561
 562                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 563                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 564
 565                 if (src_f)
 566                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 567         }
 568
 569         return 0;
 570 }
 571
 572 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 573 {
 574         int ret;
 575
 576         lockdep_assert_held(&c->sb_lock);
 577
 578         ret =   bch2_sb_realloc(&c->disk_sb, 0) ?:
 579                 __copy_super(&c->disk_sb, src) ?:
 580                 bch2_sb_replicas_to_cpu_replicas(c) ?:
 581                 bch2_sb_disk_groups_to_cpu(c);
 582         if (ret)
 583                 return ret;
 584
 585         bch2_sb_update(c);
 586         return 0;
 587 }
 588
 589 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 590 {
 591         return __copy_super(&ca->disk_sb, c->disk_sb.sb);
 592 }
 593
 594 /* read superblock: */
 595
 596 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 597 {
 598         struct bch_csum csum;
 599         size_t bytes;
 600         int ret;
 601 reread:
 602         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 603         sb->bio->bi_iter.bi_sector = offset;
 604         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 605
 606         ret = submit_bio_wait(sb->bio);
 607         if (ret) {
 608                 prt_printf(err, "IO error: %i", ret);
 609                 return ret;
 610         }
 611
 612         if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
 613             !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC)) {
 614                 prt_printf(err, "Not a bcachefs superblock");
 615                 return -BCH_ERR_invalid_sb_magic;
 616         }
 617
 618         ret = bch2_sb_compatible(sb->sb, err);
 619         if (ret)
 620                 return ret;
 621
 622         bytes = vstruct_bytes(sb->sb);
 623
 624         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 625                 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 626                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 627                 return -BCH_ERR_invalid_sb_too_big;
 628         }
 629
 630         if (bytes > sb->buffer_size) {
 631                 ret = bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s));
 632                 if (ret)
 633                         return ret;
 634                 goto reread;
 635         }
 636
 637         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
 638                 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 639                 return -BCH_ERR_invalid_sb_csum_type;
 640         }
 641
 642         /* XXX: verify MACs */
 643         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 644                             null_nonce(), sb->sb);
 645
 646         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 647                 prt_printf(err, "bad checksum");
 648                 return -BCH_ERR_invalid_sb_csum;
 649         }
 650
 651         sb->seq = le64_to_cpu(sb->sb->seq);
 652
 653         return 0;
 654 }
 655
 656 int bch2_read_super(const char *path, struct bch_opts *opts,
 657                     struct bch_sb_handle *sb)
 658 {
 659         u64 offset = opt_get(*opts, sb);
 660         struct bch_sb_layout layout;
 661         struct printbuf err = PRINTBUF;
 662         __le64 *i;
 663         int ret;
 664 #ifndef __KERNEL__
 665 retry:
 666 #endif
 667         memset(sb, 0, sizeof(*sb));
 668         sb->mode        = FMODE_READ;
 669         sb->have_bio    = true;
 670
 671 #ifndef __KERNEL__
 672         if (opt_get(*opts, direct_io) == false)
 673                 sb->mode |= FMODE_BUFFERED;
 674 #endif
 675
 676         if (!opt_get(*opts, noexcl))
 677                 sb->mode |= FMODE_EXCL;
 678
 679         if (!opt_get(*opts, nochanges))
 680                 sb->mode |= FMODE_WRITE;
 681
 682         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 683         if (IS_ERR(sb->bdev) &&
 684             PTR_ERR(sb->bdev) == -EACCES &&
 685             opt_get(*opts, read_only)) {
 686                 sb->mode &= ~FMODE_WRITE;
 687
 688                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 689                 if (!IS_ERR(sb->bdev))
 690                         opt_set(*opts, nochanges, true);
 691         }
 692
 693         if (IS_ERR(sb->bdev)) {
 694                 ret = PTR_ERR(sb->bdev);
 695                 goto out;
 696         }
 697
 698         ret = bch2_sb_realloc(sb, 0);
 699         if (ret) {
 700                 prt_printf(&err, "error allocating memory for superblock");
 701                 goto err;
 702         }
 703
 704         if (bch2_fs_init_fault("read_super")) {
 705                 prt_printf(&err, "dynamic fault");
 706                 ret = -EFAULT;
 707                 goto err;
 708         }
 709
 710         ret = read_one_super(sb, offset, &err);
 711         if (!ret)
 712                 goto got_super;
 713
 714         if (opt_defined(*opts, sb))
 715                 goto err;
 716
 717         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 718                path, err.buf);
 719         printbuf_reset(&err);
 720
 721         /*
 722          * Error reading primary superblock - read location of backup
 723          * superblocks:
 724          */
 725         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 726         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 727         /*
 728          * use sb buffer to read layout, since sb buffer is page aligned but
 729          * layout won't be:
 730          */
 731         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 732
 733         ret = submit_bio_wait(sb->bio);
 734         if (ret) {
 735                 prt_printf(&err, "IO error: %i", ret);
 736                 goto err;
 737         }
 738
 739         memcpy(&layout, sb->sb, sizeof(layout));
 740         ret = validate_sb_layout(&layout, &err);
 741         if (ret)
 742                 goto err;
 743
 744         for (i = layout.sb_offset;
 745              i < layout.sb_offset + layout.nr_superblocks; i++) {
 746                 offset = le64_to_cpu(*i);
 747
 748                 if (offset == opt_get(*opts, sb))
 749                         continue;
 750
 751                 ret = read_one_super(sb, offset, &err);
 752                 if (!ret)
 753                         goto got_super;
 754         }
 755
 756         goto err;
 757
 758 got_super:
 759         if (le16_to_cpu(sb->sb->block_size) << 9 <
 760             bdev_logical_block_size(sb->bdev) &&
 761             opt_get(*opts, direct_io)) {
 762 #ifndef __KERNEL__
 763                 opt_set(*opts, direct_io, false);
 764                 bch2_free_super(sb);
 765                 goto retry;
 766 #endif
 767                 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
 768                        le16_to_cpu(sb->sb->block_size) << 9,
 769                        bdev_logical_block_size(sb->bdev));
 770                 ret = -BCH_ERR_block_size_too_small;
 771                 goto err;
 772         }
 773
 774         ret = 0;
 775         sb->have_layout = true;
 776
 777         ret = bch2_sb_validate(sb, &err, READ);
 778         if (ret) {
 779                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 780                        path, err.buf);
 781                 goto err_no_print;
 782         }
 783 out:
 784         printbuf_exit(&err);
 785         return ret;
 786 err:
 787         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 788                path, err.buf);
 789 err_no_print:
 790         bch2_free_super(sb);
 791         goto out;
 792 }
 793
 794 /* write superblock: */
 795
 796 static void write_super_endio(struct bio *bio)
 797 {
 798         struct bch_dev *ca = bio->bi_private;
 799
 800         /* XXX: return errors directly */
 801
 802         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 803                                bch2_blk_status_to_str(bio->bi_status)))
 804                 ca->sb_write_error = 1;
 805
 806         closure_put(&ca->fs->sb_write);
 807         percpu_ref_put(&ca->io_ref);
 808 }
 809
 810 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 811 {
 812         struct bch_sb *sb = ca->disk_sb.sb;
 813         struct bio *bio = ca->disk_sb.bio;
 814
 815         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 816         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 817         bio->bi_end_io          = write_super_endio;
 818         bio->bi_private         = ca;
 819         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 820
 821         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 822                      bio_sectors(bio));
 823
 824         percpu_ref_get(&ca->io_ref);
 825         closure_bio_submit(bio, &c->sb_write);
 826 }
 827
 828 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 829 {
 830         struct bch_sb *sb = ca->disk_sb.sb;
 831         struct bio *bio = ca->disk_sb.bio;
 832
 833         sb->offset = sb->layout.sb_offset[idx];
 834
 835         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 836         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 837                                 null_nonce(), sb);
 838
 839         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 840         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 841         bio->bi_end_io          = write_super_endio;
 842         bio->bi_private         = ca;
 843         bch2_bio_map(bio, sb,
 844                      roundup((size_t) vstruct_bytes(sb),
 845                              bdev_logical_block_size(ca->disk_sb.bdev)));
 846
 847         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 848                      bio_sectors(bio));
 849
 850         percpu_ref_get(&ca->io_ref);
 851         closure_bio_submit(bio, &c->sb_write);
 852 }
 853
 854 int bch2_write_super(struct bch_fs *c)
 855 {
 856         struct closure *cl = &c->sb_write;
 857         struct bch_dev *ca;
 858         struct printbuf err = PRINTBUF;
 859         unsigned i, sb = 0, nr_wrote;
 860         struct bch_devs_mask sb_written;
 861         bool wrote, can_mount_without_written, can_mount_with_written;
 862         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 863         int ret = 0;
 864
 865         trace_and_count(c, write_super, c, _RET_IP_);
 866
 867         if (c->opts.very_degraded)
 868                 degraded_flags |= BCH_FORCE_IF_LOST;
 869
 870         lockdep_assert_held(&c->sb_lock);
 871
 872         closure_init_stack(cl);
 873         memset(&sb_written, 0, sizeof(sb_written));
 874
 875         /* Make sure we're using the new magic numbers: */
 876         c->disk_sb.sb->magic = BCHFS_MAGIC;
 877         c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
 878
 879         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 880
 881         if (test_bit(BCH_FS_ERROR, &c->flags))
 882                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 883         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 884                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 885
 886         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 887
 888         bch2_sb_counters_from_cpu(c);
 889
 890         for_each_online_member(ca, c, i)
 891                 bch2_sb_from_fs(c, ca);
 892
 893         for_each_online_member(ca, c, i) {
 894                 printbuf_reset(&err);
 895
 896                 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
 897                 if (ret) {
 898                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 899                         percpu_ref_put(&ca->io_ref);
 900                         goto out;
 901                 }
 902         }
 903
 904         if (c->opts.nochanges)
 905                 goto out;
 906
 907         /*
 908          * Defer writing the superblock until filesystem initialization is
 909          * complete - don't write out a partly initialized superblock:
 910          */
 911         if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
 912                 goto out;
 913
 914         for_each_online_member(ca, c, i) {
 915                 __set_bit(ca->dev_idx, sb_written.d);
 916                 ca->sb_write_error = 0;
 917         }
 918
 919         for_each_online_member(ca, c, i)
 920                 read_back_super(c, ca);
 921         closure_sync(cl);
 922
 923         for_each_online_member(ca, c, i) {
 924                 if (ca->sb_write_error)
 925                         continue;
 926
 927                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 928                         bch2_fs_fatal_error(c,
 929                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 930                                 le64_to_cpu(ca->sb_read_scratch->seq),
 931                                 ca->disk_sb.seq);
 932                         percpu_ref_put(&ca->io_ref);
 933                         ret = -BCH_ERR_erofs_sb_err;
 934                         goto out;
 935                 }
 936
 937                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 938                         bch2_fs_fatal_error(c,
 939                                 "Superblock modified by another process (seq %llu expected %llu)",
 940                                 le64_to_cpu(ca->sb_read_scratch->seq),
 941                                 ca->disk_sb.seq);
 942                         percpu_ref_put(&ca->io_ref);
 943                         ret = -BCH_ERR_erofs_sb_err;
 944                         goto out;
 945                 }
 946         }
 947
 948         do {
 949                 wrote = false;
 950                 for_each_online_member(ca, c, i)
 951                         if (!ca->sb_write_error &&
 952                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 953                                 write_one_super(c, ca, sb);
 954                                 wrote = true;
 955                         }
 956                 closure_sync(cl);
 957                 sb++;
 958         } while (wrote);
 959
 960         for_each_online_member(ca, c, i) {
 961                 if (ca->sb_write_error)
 962                         __clear_bit(ca->dev_idx, sb_written.d);
 963                 else
 964                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 965         }
 966
 967         nr_wrote = dev_mask_nr(&sb_written);
 968
 969         can_mount_with_written =
 970                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 971
 972         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 973                 sb_written.d[i] = ~sb_written.d[i];
 974
 975         can_mount_without_written =
 976                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 977
 978         /*
 979          * If we would be able to mount _without_ the devices we successfully
 980          * wrote superblocks to, we weren't able to write to enough devices:
 981          *
 982          * Exception: if we can mount without the successes because we haven't
 983          * written anything (new filesystem), we continue if we'd be able to
 984          * mount with the devices we did successfully write to:
 985          */
 986         if (bch2_fs_fatal_err_on(!nr_wrote ||
 987                                  !can_mount_with_written ||
 988                                  (can_mount_without_written &&
 989                                   !can_mount_with_written), c,
 990                 "Unable to write superblock to sufficient devices (from %ps)",
 991                 (void *) _RET_IP_))
 992                 ret = -1;
 993 out:
 994         /* Make new options visible after they're persistent: */
 995         bch2_sb_update(c);
 996         printbuf_exit(&err);
 997         return ret;
 998 }
 999
1000 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
1001 {
1002         mutex_lock(&c->sb_lock);
1003         if (!(c->sb.features & (1ULL << feat))) {
1004                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
1005
1006                 bch2_write_super(c);
1007         }
1008         mutex_unlock(&c->sb_lock);
1009 }
1010
1011 /* BCH_SB_FIELD_members: */
1012
1013 static int bch2_sb_members_validate(struct bch_sb *sb,
1014                                     struct bch_sb_field *f,
1015                                     struct printbuf *err)
1016 {
1017         struct bch_sb_field_members *mi = field_to_type(f, members);
1018         unsigned i;
1019
1020         if ((void *) (mi->members + sb->nr_devices) >
1021             vstruct_end(&mi->field)) {
1022                 prt_printf(err, "too many devices for section size");
1023                 return -BCH_ERR_invalid_sb_members;
1024         }
1025
1026         for (i = 0; i < sb->nr_devices; i++) {
1027                 struct bch_member *m = mi->members + i;
1028
1029                 if (!bch2_member_exists(m))
1030                         continue;
1031
1032                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
1033                         prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
1034                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
1035                         return -BCH_ERR_invalid_sb_members;
1036                 }
1037
1038                 if (le64_to_cpu(m->nbuckets) -
1039                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
1040                         prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
1041                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
1042                         return -BCH_ERR_invalid_sb_members;
1043                 }
1044
1045                 if (le16_to_cpu(m->bucket_size) <
1046                     le16_to_cpu(sb->block_size)) {
1047                         prt_printf(err, "device %u: bucket size %u smaller than block size %u",
1048                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
1049                         return -BCH_ERR_invalid_sb_members;
1050                 }
1051
1052                 if (le16_to_cpu(m->bucket_size) <
1053                     BCH_SB_BTREE_NODE_SIZE(sb)) {
1054                         prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
1055                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
1056                         return -BCH_ERR_invalid_sb_members;
1057                 }
1058         }
1059
1060         return 0;
1061 }
1062
1063 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
1064                                     struct bch_sb_field *f)
1065 {
1066         struct bch_sb_field_members *mi = field_to_type(f, members);
1067         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
1068         unsigned i;
1069
1070         for (i = 0; i < sb->nr_devices; i++) {
1071                 struct bch_member *m = mi->members + i;
1072                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
1073                 u64 bucket_size = le16_to_cpu(m->bucket_size);
1074                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1075
1076                 if (!bch2_member_exists(m))
1077                         continue;
1078
1079                 prt_printf(out, "Device:");
1080                 prt_tab(out);
1081                 prt_printf(out, "%u", i);
1082                 prt_newline(out);
1083
1084                 printbuf_indent_add(out, 2);
1085
1086                 prt_printf(out, "UUID:");
1087                 prt_tab(out);
1088                 pr_uuid(out, m->uuid.b);
1089                 prt_newline(out);
1090
1091                 prt_printf(out, "Size:");
1092                 prt_tab(out);
1093                 prt_units_u64(out, device_size << 9);
1094                 prt_newline(out);
1095
1096                 prt_printf(out, "Bucket size:");
1097                 prt_tab(out);
1098                 prt_units_u64(out, bucket_size << 9);
1099                 prt_newline(out);
1100
1101                 prt_printf(out, "First bucket:");
1102                 prt_tab(out);
1103                 prt_printf(out, "%u", le16_to_cpu(m->first_bucket));
1104                 prt_newline(out);
1105
1106                 prt_printf(out, "Buckets:");
1107                 prt_tab(out);
1108                 prt_printf(out, "%llu", le64_to_cpu(m->nbuckets));
1109                 prt_newline(out);
1110
1111                 prt_printf(out, "Last mount:");
1112                 prt_tab(out);
1113                 if (m->last_mount)
1114                         pr_time(out, le64_to_cpu(m->last_mount));
1115                 else
1116                         prt_printf(out, "(never)");
1117                 prt_newline(out);
1118
1119                 prt_printf(out, "State:");
1120                 prt_tab(out);
1121                 prt_printf(out, "%s",
1122                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1123                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1124                        : "unknown");
1125                 prt_newline(out);
1126
1127                 prt_printf(out, "Label:");
1128                 prt_tab(out);
1129                 if (BCH_MEMBER_GROUP(m)) {
1130                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1131
1132                         if (idx < disk_groups_nr(gi))
1133                                 prt_printf(out, "%s (%u)",
1134                                        gi->entries[idx].label, idx);
1135                         else
1136                                 prt_printf(out, "(bad disk labels section)");
1137                 } else {
1138                         prt_printf(out, "(none)");
1139                 }
1140                 prt_newline(out);
1141
1142                 prt_printf(out, "Data allowed:");
1143                 prt_tab(out);
1144                 if (BCH_MEMBER_DATA_ALLOWED(m))
1145                         prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(m));
1146                 else
1147                         prt_printf(out, "(none)");
1148                 prt_newline(out);
1149
1150                 prt_printf(out, "Has data:");
1151                 prt_tab(out);
1152                 if (data_have)
1153                         prt_bitflags(out, bch2_data_types, data_have);
1154                 else
1155                         prt_printf(out, "(none)");
1156                 prt_newline(out);
1157
1158                 prt_printf(out, "Discard:");
1159                 prt_tab(out);
1160                 prt_printf(out, "%llu", BCH_MEMBER_DISCARD(m));
1161                 prt_newline(out);
1162
1163                 prt_printf(out, "Freespace initialized:");
1164                 prt_tab(out);
1165                 prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1166                 prt_newline(out);
1167
1168                 printbuf_indent_sub(out, 2);
1169         }
1170 }
1171
1172 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1173         .validate       = bch2_sb_members_validate,
1174         .to_text        = bch2_sb_members_to_text,
1175 };
1176
1177 /* BCH_SB_FIELD_crypt: */
1178
1179 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1180                                   struct bch_sb_field *f,
1181                                   struct printbuf *err)
1182 {
1183         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1184
1185         if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1186                 prt_printf(err, "wrong size (got %zu should be %zu)",
1187                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1188                 return -BCH_ERR_invalid_sb_crypt;
1189         }
1190
1191         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1192                 prt_printf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1193                 return -BCH_ERR_invalid_sb_crypt;
1194         }
1195
1196         return 0;
1197 }
1198
1199 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1200                                   struct bch_sb_field *f)
1201 {
1202         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1203
1204         prt_printf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1205         prt_newline(out);
1206         prt_printf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1207         prt_newline(out);
1208         prt_printf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1209         prt_newline(out);
1210         prt_printf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1211         prt_newline(out);
1212 }
1213
1214 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1215         .validate       = bch2_sb_crypt_validate,
1216         .to_text        = bch2_sb_crypt_to_text,
1217 };
1218
1219 /* BCH_SB_FIELD_clean: */
1220
1221 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1222 {
1223         struct jset_entry *entry;
1224         int ret;
1225
1226         for (entry = clean->start;
1227              entry < (struct jset_entry *) vstruct_end(&clean->field);
1228              entry = vstruct_next(entry)) {
1229                 ret = bch2_journal_entry_validate(c, NULL, entry,
1230                                                   le16_to_cpu(c->disk_sb.sb->version),
1231                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1232                                                   write);
1233                 if (ret)
1234                         return ret;
1235         }
1236
1237         return 0;
1238 }
1239
1240 /* Downgrade if superblock is at a higher version than currently supported: */
1241 void bch2_sb_maybe_downgrade(struct bch_fs *c)
1242 {
1243         lockdep_assert_held(&c->sb_lock);
1244
1245         /*
1246          * Downgrade, if superblock is at a higher version than currently
1247          * supported:
1248          */
1249         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
1250                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1251         if (c->sb.version > bcachefs_metadata_version_current)
1252                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1253         if (c->sb.version_min > bcachefs_metadata_version_current)
1254                 c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
1255         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1256 }
1257
1258 void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version)
1259 {
1260         lockdep_assert_held(&c->sb_lock);
1261
1262         c->disk_sb.sb->version = cpu_to_le16(new_version);
1263         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1264 }
1265
1266 int bch2_fs_mark_dirty(struct bch_fs *c)
1267 {
1268         int ret;
1269
1270         /*
1271          * Unconditionally write superblock, to verify it hasn't changed before
1272          * we go rw:
1273          */
1274
1275         mutex_lock(&c->sb_lock);
1276         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1277
1278         bch2_sb_maybe_downgrade(c);
1279         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1280
1281         ret = bch2_write_super(c);
1282         mutex_unlock(&c->sb_lock);
1283
1284         return ret;
1285 }
1286
1287 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1288 {
1289         struct jset_entry *entry = *end;
1290         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1291
1292         memset(entry, 0, u64s * sizeof(u64));
1293         /*
1294          * The u64s field counts from the start of data, ignoring the shared
1295          * fields.
1296          */
1297         entry->u64s = cpu_to_le16(u64s - 1);
1298
1299         *end = vstruct_next(*end);
1300         return entry;
1301 }
1302
1303 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1304                                            struct jset_entry **end,
1305                                            u64 journal_seq)
1306 {
1307         struct bch_dev *ca;
1308         unsigned i, dev;
1309
1310         percpu_down_read(&c->mark_lock);
1311
1312         if (!journal_seq) {
1313                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1314                         bch2_fs_usage_acc_to_base(c, i);
1315         } else {
1316                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1317         }
1318
1319         {
1320                 struct jset_entry_usage *u =
1321                         container_of(jset_entry_init(end, sizeof(*u)),
1322                                      struct jset_entry_usage, entry);
1323
1324                 u->entry.type   = BCH_JSET_ENTRY_usage;
1325                 u->entry.btree_id = BCH_FS_USAGE_inodes;
1326                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1327         }
1328
1329         {
1330                 struct jset_entry_usage *u =
1331                         container_of(jset_entry_init(end, sizeof(*u)),
1332                                      struct jset_entry_usage, entry);
1333
1334                 u->entry.type   = BCH_JSET_ENTRY_usage;
1335                 u->entry.btree_id = BCH_FS_USAGE_key_version;
1336                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1337         }
1338
1339         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1340                 struct jset_entry_usage *u =
1341                         container_of(jset_entry_init(end, sizeof(*u)),
1342                                      struct jset_entry_usage, entry);
1343
1344                 u->entry.type   = BCH_JSET_ENTRY_usage;
1345                 u->entry.btree_id = BCH_FS_USAGE_reserved;
1346                 u->entry.level  = i;
1347                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1348         }
1349
1350         for (i = 0; i < c->replicas.nr; i++) {
1351                 struct bch_replicas_entry *e =
1352                         cpu_replicas_entry(&c->replicas, i);
1353                 struct jset_entry_data_usage *u =
1354                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1355                                      struct jset_entry_data_usage, entry);
1356
1357                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1358                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1359                 unsafe_memcpy(&u->r, e, replicas_entry_bytes(e),
1360                               "embedded variable length struct");
1361         }
1362
1363         for_each_member_device(ca, c, dev) {
1364                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1365                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1366                 struct jset_entry_dev_usage *u =
1367                         container_of(jset_entry_init(end, b),
1368                                      struct jset_entry_dev_usage, entry);
1369
1370                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1371                 u->dev = cpu_to_le32(dev);
1372                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1373
1374                 for (i = 0; i < BCH_DATA_NR; i++) {
1375                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1376                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1377                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1378                 }
1379         }
1380
1381         percpu_up_read(&c->mark_lock);
1382
1383         for (i = 0; i < 2; i++) {
1384                 struct jset_entry_clock *clock =
1385                         container_of(jset_entry_init(end, sizeof(*clock)),
1386                                      struct jset_entry_clock, entry);
1387
1388                 clock->entry.type = BCH_JSET_ENTRY_clock;
1389                 clock->rw       = i;
1390                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1391         }
1392 }
1393
1394 void bch2_fs_mark_clean(struct bch_fs *c)
1395 {
1396         struct bch_sb_field_clean *sb_clean;
1397         struct jset_entry *entry;
1398         unsigned u64s;
1399         int ret;
1400
1401         mutex_lock(&c->sb_lock);
1402         if (BCH_SB_CLEAN(c->disk_sb.sb))
1403                 goto out;
1404
1405         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1406
1407         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1408         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1409         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1410         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1411
1412         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1413
1414         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1415         if (!sb_clean) {
1416                 bch_err(c, "error resizing superblock while setting filesystem clean");
1417                 goto out;
1418         }
1419
1420         sb_clean->flags         = 0;
1421         sb_clean->journal_seq   = cpu_to_le64(atomic64_read(&c->journal.seq));
1422
1423         /* Trying to catch outstanding bug: */
1424         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1425
1426         entry = sb_clean->start;
1427         bch2_journal_super_entries_add_common(c, &entry, 0);
1428         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1429         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1430
1431         memset(entry, 0,
1432                vstruct_end(&sb_clean->field) - (void *) entry);
1433
1434         /*
1435          * this should be in the write path, and we should be validating every
1436          * superblock section:
1437          */
1438         ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1439         if (ret) {
1440                 bch_err(c, "error writing marking filesystem clean: validate error");
1441                 goto out;
1442         }
1443
1444         bch2_write_super(c);
1445 out:
1446         mutex_unlock(&c->sb_lock);
1447 }
1448
1449 static int bch2_sb_clean_validate(struct bch_sb *sb,
1450                                   struct bch_sb_field *f,
1451                                   struct printbuf *err)
1452 {
1453         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1454
1455         if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1456                 prt_printf(err, "wrong size (got %zu should be %zu)",
1457                        vstruct_bytes(&clean->field), sizeof(*clean));
1458                 return -BCH_ERR_invalid_sb_clean;
1459         }
1460
1461         return 0;
1462 }
1463
1464 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1465                                   struct bch_sb_field *f)
1466 {
1467         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1468         struct jset_entry *entry;
1469
1470         prt_printf(out, "flags:          %x",   le32_to_cpu(clean->flags));
1471         prt_newline(out);
1472         prt_printf(out, "journal_seq:    %llu", le64_to_cpu(clean->journal_seq));
1473         prt_newline(out);
1474
1475         for (entry = clean->start;
1476              entry != vstruct_end(&clean->field);
1477              entry = vstruct_next(entry)) {
1478                 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1479                     !entry->u64s)
1480                         continue;
1481
1482                 bch2_journal_entry_to_text(out, NULL, entry);
1483                 prt_newline(out);
1484         }
1485 }
1486
1487 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1488         .validate       = bch2_sb_clean_validate,
1489         .to_text        = bch2_sb_clean_to_text,
1490 };
1491
1492 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1493 #define x(f, nr)                                        \
1494         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1495         BCH_SB_FIELDS()
1496 #undef x
1497 };
1498
1499 static const struct bch_sb_field_ops bch2_sb_field_null_ops;
1500
1501 static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
1502 {
1503         return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
1504                 ? bch2_sb_field_ops[type]
1505                 : &bch2_sb_field_null_ops;
1506 }
1507
1508 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1509                                   struct printbuf *err)
1510 {
1511         unsigned type = le32_to_cpu(f->type);
1512         struct printbuf field_err = PRINTBUF;
1513         const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1514         int ret;
1515
1516         ret = ops->validate ? ops->validate(sb, f, &field_err) : 0;
1517         if (ret) {
1518                 prt_printf(err, "Invalid superblock section %s: %s",
1519                            bch2_sb_fields[type], field_err.buf);
1520                 prt_newline(err);
1521                 bch2_sb_field_to_text(err, sb, f);
1522         }
1523
1524         printbuf_exit(&field_err);
1525         return ret;
1526 }
1527
1528 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1529                            struct bch_sb_field *f)
1530 {
1531         unsigned type = le32_to_cpu(f->type);
1532         const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1533
1534         if (!out->nr_tabstops)
1535                 printbuf_tabstop_push(out, 32);
1536
1537         if (type < BCH_SB_FIELD_NR)
1538                 prt_printf(out, "%s", bch2_sb_fields[type]);
1539         else
1540                 prt_printf(out, "(unknown field %u)", type);
1541
1542         prt_printf(out, " (size %zu):", vstruct_bytes(f));
1543         prt_newline(out);
1544
1545         if (ops->to_text) {
1546                 printbuf_indent_add(out, 2);
1547                 ops->to_text(out, sb, f);
1548                 printbuf_indent_sub(out, 2);
1549         }
1550 }
1551
1552 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1553 {
1554         unsigned i;
1555
1556         prt_printf(out, "Type:                    %u", l->layout_type);
1557         prt_newline(out);
1558
1559         prt_str(out, "Superblock max size:     ");
1560         prt_units_u64(out, 512 << l->sb_max_size_bits);
1561         prt_newline(out);
1562
1563         prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1564         prt_newline(out);
1565
1566         prt_str(out, "Offsets:                 ");
1567         for (i = 0; i < l->nr_superblocks; i++) {
1568                 if (i)
1569                         prt_str(out, ", ");
1570                 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1571         }
1572         prt_newline(out);
1573 }
1574
1575 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1576                      bool print_layout, unsigned fields)
1577 {
1578         struct bch_sb_field_members *mi;
1579         struct bch_sb_field *f;
1580         u64 fields_have = 0;
1581         unsigned nr_devices = 0;
1582
1583         if (!out->nr_tabstops)
1584                 printbuf_tabstop_push(out, 44);
1585
1586         mi = bch2_sb_get_members(sb);
1587         if (mi) {
1588                 struct bch_member *m;
1589
1590                 for (m = mi->members;
1591                      m < mi->members + sb->nr_devices;
1592                      m++)
1593                         nr_devices += bch2_member_exists(m);
1594         }
1595
1596         prt_printf(out, "External UUID:");
1597         prt_tab(out);
1598         pr_uuid(out, sb->user_uuid.b);
1599         prt_newline(out);
1600
1601         prt_printf(out, "Internal UUID:");
1602         prt_tab(out);
1603         pr_uuid(out, sb->uuid.b);
1604         prt_newline(out);
1605
1606         prt_str(out, "Device index:");
1607         prt_tab(out);
1608         prt_printf(out, "%u", sb->dev_idx);
1609         prt_newline(out);
1610
1611         prt_str(out, "Label:");
1612         prt_tab(out);
1613         prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1614         prt_newline(out);
1615
1616         prt_str(out, "Version:");
1617         prt_tab(out);
1618         bch2_version_to_text(out, le16_to_cpu(sb->version));
1619         prt_newline(out);
1620
1621         prt_str(out, "Version upgrade complete:");
1622         prt_tab(out);
1623         bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
1624         prt_newline(out);
1625
1626         prt_printf(out, "Oldest version on disk:");
1627         prt_tab(out);
1628         bch2_version_to_text(out, le16_to_cpu(sb->version_min));
1629         prt_newline(out);
1630
1631         prt_printf(out, "Created:");
1632         prt_tab(out);
1633         if (sb->time_base_lo)
1634                 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1635         else
1636                 prt_printf(out, "(not set)");
1637         prt_newline(out);
1638
1639         prt_printf(out, "Sequence number:");
1640         prt_tab(out);
1641         prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1642         prt_newline(out);
1643
1644         prt_printf(out, "Superblock size:");
1645         prt_tab(out);
1646         prt_printf(out, "%zu", vstruct_bytes(sb));
1647         prt_newline(out);
1648
1649         prt_printf(out, "Clean:");
1650         prt_tab(out);
1651         prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1652         prt_newline(out);
1653
1654         prt_printf(out, "Devices:");
1655         prt_tab(out);
1656         prt_printf(out, "%u", nr_devices);
1657         prt_newline(out);
1658
1659         prt_printf(out, "Sections:");
1660         vstruct_for_each(sb, f)
1661                 fields_have |= 1 << le32_to_cpu(f->type);
1662         prt_tab(out);
1663         prt_bitflags(out, bch2_sb_fields, fields_have);
1664         prt_newline(out);
1665
1666         prt_printf(out, "Features:");
1667         prt_tab(out);
1668         prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1669         prt_newline(out);
1670
1671         prt_printf(out, "Compat features:");
1672         prt_tab(out);
1673         prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1674         prt_newline(out);
1675
1676         prt_newline(out);
1677         prt_printf(out, "Options:");
1678         prt_newline(out);
1679         printbuf_indent_add(out, 2);
1680         {
1681                 enum bch_opt_id id;
1682
1683                 for (id = 0; id < bch2_opts_nr; id++) {
1684                         const struct bch_option *opt = bch2_opt_table + id;
1685
1686                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1687                                 u64 v = bch2_opt_from_sb(sb, id);
1688
1689                                 prt_printf(out, "%s:", opt->attr.name);
1690                                 prt_tab(out);
1691                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1692                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1693                                 prt_newline(out);
1694                         }
1695                 }
1696         }
1697
1698         printbuf_indent_sub(out, 2);
1699
1700         if (print_layout) {
1701                 prt_newline(out);
1702                 prt_printf(out, "layout:");
1703                 prt_newline(out);
1704                 printbuf_indent_add(out, 2);
1705                 bch2_sb_layout_to_text(out, &sb->layout);
1706                 printbuf_indent_sub(out, 2);
1707         }
1708
1709         vstruct_for_each(sb, f)
1710                 if (fields & (1 << le32_to_cpu(f->type))) {
1711                         prt_newline(out);
1712                         bch2_sb_field_to_text(out, sb, f);
1713                 }
1714 }