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
 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) ?: c->sb.version;
 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
 662         memset(sb, 0, sizeof(*sb));
 663         sb->mode        = FMODE_READ;
 664         sb->have_bio    = true;
 665
 666         if (!opt_get(*opts, noexcl))
 667                 sb->mode |= FMODE_EXCL;
 668
 669         if (!opt_get(*opts, nochanges))
 670                 sb->mode |= FMODE_WRITE;
 671
 672         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 673         if (IS_ERR(sb->bdev) &&
 674             PTR_ERR(sb->bdev) == -EACCES &&
 675             opt_get(*opts, read_only)) {
 676                 sb->mode &= ~FMODE_WRITE;
 677
 678                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 679                 if (!IS_ERR(sb->bdev))
 680                         opt_set(*opts, nochanges, true);
 681         }
 682
 683         if (IS_ERR(sb->bdev)) {
 684                 ret = PTR_ERR(sb->bdev);
 685                 goto out;
 686         }
 687
 688         ret = bch2_sb_realloc(sb, 0);
 689         if (ret) {
 690                 prt_printf(&err, "error allocating memory for superblock");
 691                 goto err;
 692         }
 693
 694         if (bch2_fs_init_fault("read_super")) {
 695                 prt_printf(&err, "dynamic fault");
 696                 ret = -EFAULT;
 697                 goto err;
 698         }
 699
 700         ret = read_one_super(sb, offset, &err);
 701         if (!ret)
 702                 goto got_super;
 703
 704         if (opt_defined(*opts, sb))
 705                 goto err;
 706
 707         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 708                path, err.buf);
 709         printbuf_reset(&err);
 710
 711         /*
 712          * Error reading primary superblock - read location of backup
 713          * superblocks:
 714          */
 715         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 716         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 717         /*
 718          * use sb buffer to read layout, since sb buffer is page aligned but
 719          * layout won't be:
 720          */
 721         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 722
 723         ret = submit_bio_wait(sb->bio);
 724         if (ret) {
 725                 prt_printf(&err, "IO error: %i", ret);
 726                 goto err;
 727         }
 728
 729         memcpy(&layout, sb->sb, sizeof(layout));
 730         ret = validate_sb_layout(&layout, &err);
 731         if (ret)
 732                 goto err;
 733
 734         for (i = layout.sb_offset;
 735              i < layout.sb_offset + layout.nr_superblocks; i++) {
 736                 offset = le64_to_cpu(*i);
 737
 738                 if (offset == opt_get(*opts, sb))
 739                         continue;
 740
 741                 ret = read_one_super(sb, offset, &err);
 742                 if (!ret)
 743                         goto got_super;
 744         }
 745
 746         goto err;
 747
 748 got_super:
 749         if (le16_to_cpu(sb->sb->block_size) << 9 <
 750             bdev_logical_block_size(sb->bdev)) {
 751                 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
 752                        le16_to_cpu(sb->sb->block_size) << 9,
 753                        bdev_logical_block_size(sb->bdev));
 754                 ret = -BCH_ERR_block_size_too_small;
 755                 goto err;
 756         }
 757
 758         ret = 0;
 759         sb->have_layout = true;
 760
 761         ret = bch2_sb_validate(sb, &err, READ);
 762         if (ret) {
 763                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 764                        path, err.buf);
 765                 goto err_no_print;
 766         }
 767 out:
 768         printbuf_exit(&err);
 769         return ret;
 770 err:
 771         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 772                path, err.buf);
 773 err_no_print:
 774         bch2_free_super(sb);
 775         goto out;
 776 }
 777
 778 /* write superblock: */
 779
 780 static void write_super_endio(struct bio *bio)
 781 {
 782         struct bch_dev *ca = bio->bi_private;
 783
 784         /* XXX: return errors directly */
 785
 786         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 787                                bch2_blk_status_to_str(bio->bi_status)))
 788                 ca->sb_write_error = 1;
 789
 790         closure_put(&ca->fs->sb_write);
 791         percpu_ref_put(&ca->io_ref);
 792 }
 793
 794 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 795 {
 796         struct bch_sb *sb = ca->disk_sb.sb;
 797         struct bio *bio = ca->disk_sb.bio;
 798
 799         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 800         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 801         bio->bi_end_io          = write_super_endio;
 802         bio->bi_private         = ca;
 803         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 804
 805         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 806                      bio_sectors(bio));
 807
 808         percpu_ref_get(&ca->io_ref);
 809         closure_bio_submit(bio, &c->sb_write);
 810 }
 811
 812 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 813 {
 814         struct bch_sb *sb = ca->disk_sb.sb;
 815         struct bio *bio = ca->disk_sb.bio;
 816
 817         sb->offset = sb->layout.sb_offset[idx];
 818
 819         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 820         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 821                                 null_nonce(), sb);
 822
 823         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 824         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 825         bio->bi_end_io          = write_super_endio;
 826         bio->bi_private         = ca;
 827         bch2_bio_map(bio, sb,
 828                      roundup((size_t) vstruct_bytes(sb),
 829                              bdev_logical_block_size(ca->disk_sb.bdev)));
 830
 831         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 832                      bio_sectors(bio));
 833
 834         percpu_ref_get(&ca->io_ref);
 835         closure_bio_submit(bio, &c->sb_write);
 836 }
 837
 838 int bch2_write_super(struct bch_fs *c)
 839 {
 840         struct closure *cl = &c->sb_write;
 841         struct bch_dev *ca;
 842         struct printbuf err = PRINTBUF;
 843         unsigned i, sb = 0, nr_wrote;
 844         struct bch_devs_mask sb_written;
 845         bool wrote, can_mount_without_written, can_mount_with_written;
 846         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 847         int ret = 0;
 848
 849         trace_and_count(c, write_super, c, _RET_IP_);
 850
 851         if (c->opts.very_degraded)
 852                 degraded_flags |= BCH_FORCE_IF_LOST;
 853
 854         lockdep_assert_held(&c->sb_lock);
 855
 856         closure_init_stack(cl);
 857         memset(&sb_written, 0, sizeof(sb_written));
 858
 859         /* Make sure we're using the new magic numbers: */
 860         c->disk_sb.sb->magic = BCHFS_MAGIC;
 861         c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
 862
 863         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 864
 865         if (test_bit(BCH_FS_ERROR, &c->flags))
 866                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 867         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 868                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 869
 870         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 871
 872         bch2_sb_counters_from_cpu(c);
 873
 874         for_each_online_member(ca, c, i)
 875                 bch2_sb_from_fs(c, ca);
 876
 877         for_each_online_member(ca, c, i) {
 878                 printbuf_reset(&err);
 879
 880                 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
 881                 if (ret) {
 882                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 883                         percpu_ref_put(&ca->io_ref);
 884                         goto out;
 885                 }
 886         }
 887
 888         if (c->opts.nochanges)
 889                 goto out;
 890
 891         /*
 892          * Defer writing the superblock until filesystem initialization is
 893          * complete - don't write out a partly initialized superblock:
 894          */
 895         if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
 896                 goto out;
 897
 898         for_each_online_member(ca, c, i) {
 899                 __set_bit(ca->dev_idx, sb_written.d);
 900                 ca->sb_write_error = 0;
 901         }
 902
 903         for_each_online_member(ca, c, i)
 904                 read_back_super(c, ca);
 905         closure_sync(cl);
 906
 907         for_each_online_member(ca, c, i) {
 908                 if (ca->sb_write_error)
 909                         continue;
 910
 911                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 912                         bch2_fs_fatal_error(c,
 913                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 914                                 le64_to_cpu(ca->sb_read_scratch->seq),
 915                                 ca->disk_sb.seq);
 916                         percpu_ref_put(&ca->io_ref);
 917                         ret = -BCH_ERR_erofs_sb_err;
 918                         goto out;
 919                 }
 920
 921                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 922                         bch2_fs_fatal_error(c,
 923                                 "Superblock modified by another process (seq %llu expected %llu)",
 924                                 le64_to_cpu(ca->sb_read_scratch->seq),
 925                                 ca->disk_sb.seq);
 926                         percpu_ref_put(&ca->io_ref);
 927                         ret = -BCH_ERR_erofs_sb_err;
 928                         goto out;
 929                 }
 930         }
 931
 932         do {
 933                 wrote = false;
 934                 for_each_online_member(ca, c, i)
 935                         if (!ca->sb_write_error &&
 936                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 937                                 write_one_super(c, ca, sb);
 938                                 wrote = true;
 939                         }
 940                 closure_sync(cl);
 941                 sb++;
 942         } while (wrote);
 943
 944         for_each_online_member(ca, c, i) {
 945                 if (ca->sb_write_error)
 946                         __clear_bit(ca->dev_idx, sb_written.d);
 947                 else
 948                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 949         }
 950
 951         nr_wrote = dev_mask_nr(&sb_written);
 952
 953         can_mount_with_written =
 954                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 955
 956         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 957                 sb_written.d[i] = ~sb_written.d[i];
 958
 959         can_mount_without_written =
 960                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 961
 962         /*
 963          * If we would be able to mount _without_ the devices we successfully
 964          * wrote superblocks to, we weren't able to write to enough devices:
 965          *
 966          * Exception: if we can mount without the successes because we haven't
 967          * written anything (new filesystem), we continue if we'd be able to
 968          * mount with the devices we did successfully write to:
 969          */
 970         if (bch2_fs_fatal_err_on(!nr_wrote ||
 971                                  !can_mount_with_written ||
 972                                  (can_mount_without_written &&
 973                                   !can_mount_with_written), c,
 974                 "Unable to write superblock to sufficient devices (from %ps)",
 975                 (void *) _RET_IP_))
 976                 ret = -1;
 977 out:
 978         /* Make new options visible after they're persistent: */
 979         bch2_sb_update(c);
 980         printbuf_exit(&err);
 981         return ret;
 982 }
 983
 984 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 985 {
 986         mutex_lock(&c->sb_lock);
 987         if (!(c->sb.features & (1ULL << feat))) {
 988                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 989
 990                 bch2_write_super(c);
 991         }
 992         mutex_unlock(&c->sb_lock);
 993 }
 994
 995 /* BCH_SB_FIELD_members: */
 996
 997 static int bch2_sb_members_validate(struct bch_sb *sb,
 998                                     struct bch_sb_field *f,
 999                                     struct printbuf *err)
1000 {
1001         struct bch_sb_field_members *mi = field_to_type(f, members);
1002         unsigned i;
1003
1004         if ((void *) (mi->members + sb->nr_devices) >
1005             vstruct_end(&mi->field)) {
1006                 prt_printf(err, "too many devices for section size");
1007                 return -BCH_ERR_invalid_sb_members;
1008         }
1009
1010         for (i = 0; i < sb->nr_devices; i++) {
1011                 struct bch_member *m = mi->members + i;
1012
1013                 if (!bch2_member_exists(m))
1014                         continue;
1015
1016                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
1017                         prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
1018                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
1019                         return -BCH_ERR_invalid_sb_members;
1020                 }
1021
1022                 if (le64_to_cpu(m->nbuckets) -
1023                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
1024                         prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
1025                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
1026                         return -BCH_ERR_invalid_sb_members;
1027                 }
1028
1029                 if (le16_to_cpu(m->bucket_size) <
1030                     le16_to_cpu(sb->block_size)) {
1031                         prt_printf(err, "device %u: bucket size %u smaller than block size %u",
1032                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
1033                         return -BCH_ERR_invalid_sb_members;
1034                 }
1035
1036                 if (le16_to_cpu(m->bucket_size) <
1037                     BCH_SB_BTREE_NODE_SIZE(sb)) {
1038                         prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
1039                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
1040                         return -BCH_ERR_invalid_sb_members;
1041                 }
1042         }
1043
1044         return 0;
1045 }
1046
1047 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
1048                                     struct bch_sb_field *f)
1049 {
1050         struct bch_sb_field_members *mi = field_to_type(f, members);
1051         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
1052         unsigned i;
1053
1054         for (i = 0; i < sb->nr_devices; i++) {
1055                 struct bch_member *m = mi->members + i;
1056                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
1057                 u64 bucket_size = le16_to_cpu(m->bucket_size);
1058                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1059
1060                 if (!bch2_member_exists(m))
1061                         continue;
1062
1063                 prt_printf(out, "Device:");
1064                 prt_tab(out);
1065                 prt_printf(out, "%u", i);
1066                 prt_newline(out);
1067
1068                 printbuf_indent_add(out, 2);
1069
1070                 prt_printf(out, "UUID:");
1071                 prt_tab(out);
1072                 pr_uuid(out, m->uuid.b);
1073                 prt_newline(out);
1074
1075                 prt_printf(out, "Size:");
1076                 prt_tab(out);
1077                 prt_units_u64(out, device_size << 9);
1078                 prt_newline(out);
1079
1080                 prt_printf(out, "Bucket size:");
1081                 prt_tab(out);
1082                 prt_units_u64(out, bucket_size << 9);
1083                 prt_newline(out);
1084
1085                 prt_printf(out, "First bucket:");
1086                 prt_tab(out);
1087                 prt_printf(out, "%u", le16_to_cpu(m->first_bucket));
1088                 prt_newline(out);
1089
1090                 prt_printf(out, "Buckets:");
1091                 prt_tab(out);
1092                 prt_printf(out, "%llu", le64_to_cpu(m->nbuckets));
1093                 prt_newline(out);
1094
1095                 prt_printf(out, "Last mount:");
1096                 prt_tab(out);
1097                 if (m->last_mount)
1098                         pr_time(out, le64_to_cpu(m->last_mount));
1099                 else
1100                         prt_printf(out, "(never)");
1101                 prt_newline(out);
1102
1103                 prt_printf(out, "State:");
1104                 prt_tab(out);
1105                 prt_printf(out, "%s",
1106                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1107                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1108                        : "unknown");
1109                 prt_newline(out);
1110
1111                 prt_printf(out, "Label:");
1112                 prt_tab(out);
1113                 if (BCH_MEMBER_GROUP(m)) {
1114                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1115
1116                         if (idx < disk_groups_nr(gi))
1117                                 prt_printf(out, "%s (%u)",
1118                                        gi->entries[idx].label, idx);
1119                         else
1120                                 prt_printf(out, "(bad disk labels section)");
1121                 } else {
1122                         prt_printf(out, "(none)");
1123                 }
1124                 prt_newline(out);
1125
1126                 prt_printf(out, "Data allowed:");
1127                 prt_tab(out);
1128                 if (BCH_MEMBER_DATA_ALLOWED(m))
1129                         prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(m));
1130                 else
1131                         prt_printf(out, "(none)");
1132                 prt_newline(out);
1133
1134                 prt_printf(out, "Has data:");
1135                 prt_tab(out);
1136                 if (data_have)
1137                         prt_bitflags(out, bch2_data_types, data_have);
1138                 else
1139                         prt_printf(out, "(none)");
1140                 prt_newline(out);
1141
1142                 prt_printf(out, "Discard:");
1143                 prt_tab(out);
1144                 prt_printf(out, "%llu", BCH_MEMBER_DISCARD(m));
1145                 prt_newline(out);
1146
1147                 prt_printf(out, "Freespace initialized:");
1148                 prt_tab(out);
1149                 prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1150                 prt_newline(out);
1151
1152                 printbuf_indent_sub(out, 2);
1153         }
1154 }
1155
1156 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1157         .validate       = bch2_sb_members_validate,
1158         .to_text        = bch2_sb_members_to_text,
1159 };
1160
1161 /* BCH_SB_FIELD_crypt: */
1162
1163 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1164                                   struct bch_sb_field *f,
1165                                   struct printbuf *err)
1166 {
1167         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1168
1169         if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1170                 prt_printf(err, "wrong size (got %zu should be %zu)",
1171                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1172                 return -BCH_ERR_invalid_sb_crypt;
1173         }
1174
1175         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1176                 prt_printf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1177                 return -BCH_ERR_invalid_sb_crypt;
1178         }
1179
1180         return 0;
1181 }
1182
1183 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1184                                   struct bch_sb_field *f)
1185 {
1186         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1187
1188         prt_printf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1189         prt_newline(out);
1190         prt_printf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1191         prt_newline(out);
1192         prt_printf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1193         prt_newline(out);
1194         prt_printf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1195         prt_newline(out);
1196 }
1197
1198 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1199         .validate       = bch2_sb_crypt_validate,
1200         .to_text        = bch2_sb_crypt_to_text,
1201 };
1202
1203 /* BCH_SB_FIELD_clean: */
1204
1205 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1206 {
1207         struct jset_entry *entry;
1208         int ret;
1209
1210         for (entry = clean->start;
1211              entry < (struct jset_entry *) vstruct_end(&clean->field);
1212              entry = vstruct_next(entry)) {
1213                 ret = bch2_journal_entry_validate(c, NULL, entry,
1214                                                   le16_to_cpu(c->disk_sb.sb->version),
1215                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1216                                                   write);
1217                 if (ret)
1218                         return ret;
1219         }
1220
1221         return 0;
1222 }
1223
1224 /* Downgrade if superblock is at a higher version than currently supported: */
1225 void bch2_sb_maybe_downgrade(struct bch_fs *c)
1226 {
1227         lockdep_assert_held(&c->sb_lock);
1228
1229         /*
1230          * Downgrade, if superblock is at a higher version than currently
1231          * supported:
1232          */
1233         if (BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb) > bcachefs_metadata_version_current)
1234                 SET_BCH_SB_VERSION_UPGRADE_COMPLETE(c->disk_sb.sb, bcachefs_metadata_version_current);
1235         if (c->sb.version > bcachefs_metadata_version_current)
1236                 c->disk_sb.sb->version = cpu_to_le16(bcachefs_metadata_version_current);
1237         if (c->sb.version_min > bcachefs_metadata_version_current)
1238                 c->disk_sb.sb->version_min = cpu_to_le16(bcachefs_metadata_version_current);
1239         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1240 }
1241
1242 void bch2_sb_upgrade(struct bch_fs *c, unsigned new_version)
1243 {
1244         lockdep_assert_held(&c->sb_lock);
1245
1246         c->disk_sb.sb->version = cpu_to_le16(new_version);
1247         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALL);
1248 }
1249
1250 int bch2_fs_mark_dirty(struct bch_fs *c)
1251 {
1252         int ret;
1253
1254         /*
1255          * Unconditionally write superblock, to verify it hasn't changed before
1256          * we go rw:
1257          */
1258
1259         mutex_lock(&c->sb_lock);
1260         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1261
1262         bch2_sb_maybe_downgrade(c);
1263         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1264
1265         ret = bch2_write_super(c);
1266         mutex_unlock(&c->sb_lock);
1267
1268         return ret;
1269 }
1270
1271 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1272 {
1273         struct jset_entry *entry = *end;
1274         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1275
1276         memset(entry, 0, u64s * sizeof(u64));
1277         /*
1278          * The u64s field counts from the start of data, ignoring the shared
1279          * fields.
1280          */
1281         entry->u64s = cpu_to_le16(u64s - 1);
1282
1283         *end = vstruct_next(*end);
1284         return entry;
1285 }
1286
1287 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1288                                            struct jset_entry **end,
1289                                            u64 journal_seq)
1290 {
1291         struct bch_dev *ca;
1292         unsigned i, dev;
1293
1294         percpu_down_read(&c->mark_lock);
1295
1296         if (!journal_seq) {
1297                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1298                         bch2_fs_usage_acc_to_base(c, i);
1299         } else {
1300                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1301         }
1302
1303         {
1304                 struct jset_entry_usage *u =
1305                         container_of(jset_entry_init(end, sizeof(*u)),
1306                                      struct jset_entry_usage, entry);
1307
1308                 u->entry.type   = BCH_JSET_ENTRY_usage;
1309                 u->entry.btree_id = BCH_FS_USAGE_inodes;
1310                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1311         }
1312
1313         {
1314                 struct jset_entry_usage *u =
1315                         container_of(jset_entry_init(end, sizeof(*u)),
1316                                      struct jset_entry_usage, entry);
1317
1318                 u->entry.type   = BCH_JSET_ENTRY_usage;
1319                 u->entry.btree_id = BCH_FS_USAGE_key_version;
1320                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1321         }
1322
1323         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1324                 struct jset_entry_usage *u =
1325                         container_of(jset_entry_init(end, sizeof(*u)),
1326                                      struct jset_entry_usage, entry);
1327
1328                 u->entry.type   = BCH_JSET_ENTRY_usage;
1329                 u->entry.btree_id = BCH_FS_USAGE_reserved;
1330                 u->entry.level  = i;
1331                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1332         }
1333
1334         for (i = 0; i < c->replicas.nr; i++) {
1335                 struct bch_replicas_entry *e =
1336                         cpu_replicas_entry(&c->replicas, i);
1337                 struct jset_entry_data_usage *u =
1338                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1339                                      struct jset_entry_data_usage, entry);
1340
1341                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1342                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1343                 unsafe_memcpy(&u->r, e, replicas_entry_bytes(e),
1344                               "embedded variable length struct");
1345         }
1346
1347         for_each_member_device(ca, c, dev) {
1348                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1349                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1350                 struct jset_entry_dev_usage *u =
1351                         container_of(jset_entry_init(end, b),
1352                                      struct jset_entry_dev_usage, entry);
1353
1354                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1355                 u->dev = cpu_to_le32(dev);
1356                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1357
1358                 for (i = 0; i < BCH_DATA_NR; i++) {
1359                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1360                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1361                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1362                 }
1363         }
1364
1365         percpu_up_read(&c->mark_lock);
1366
1367         for (i = 0; i < 2; i++) {
1368                 struct jset_entry_clock *clock =
1369                         container_of(jset_entry_init(end, sizeof(*clock)),
1370                                      struct jset_entry_clock, entry);
1371
1372                 clock->entry.type = BCH_JSET_ENTRY_clock;
1373                 clock->rw       = i;
1374                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1375         }
1376 }
1377
1378 void bch2_fs_mark_clean(struct bch_fs *c)
1379 {
1380         struct bch_sb_field_clean *sb_clean;
1381         struct jset_entry *entry;
1382         unsigned u64s;
1383         int ret;
1384
1385         mutex_lock(&c->sb_lock);
1386         if (BCH_SB_CLEAN(c->disk_sb.sb))
1387                 goto out;
1388
1389         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1390
1391         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1392         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1393         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1394         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1395
1396         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1397
1398         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1399         if (!sb_clean) {
1400                 bch_err(c, "error resizing superblock while setting filesystem clean");
1401                 goto out;
1402         }
1403
1404         sb_clean->flags         = 0;
1405         sb_clean->journal_seq   = cpu_to_le64(atomic64_read(&c->journal.seq));
1406
1407         /* Trying to catch outstanding bug: */
1408         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1409
1410         entry = sb_clean->start;
1411         bch2_journal_super_entries_add_common(c, &entry, 0);
1412         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1413         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1414
1415         memset(entry, 0,
1416                vstruct_end(&sb_clean->field) - (void *) entry);
1417
1418         /*
1419          * this should be in the write path, and we should be validating every
1420          * superblock section:
1421          */
1422         ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1423         if (ret) {
1424                 bch_err(c, "error writing marking filesystem clean: validate error");
1425                 goto out;
1426         }
1427
1428         bch2_write_super(c);
1429 out:
1430         mutex_unlock(&c->sb_lock);
1431 }
1432
1433 static int bch2_sb_clean_validate(struct bch_sb *sb,
1434                                   struct bch_sb_field *f,
1435                                   struct printbuf *err)
1436 {
1437         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1438
1439         if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1440                 prt_printf(err, "wrong size (got %zu should be %zu)",
1441                        vstruct_bytes(&clean->field), sizeof(*clean));
1442                 return -BCH_ERR_invalid_sb_clean;
1443         }
1444
1445         return 0;
1446 }
1447
1448 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1449                                   struct bch_sb_field *f)
1450 {
1451         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1452         struct jset_entry *entry;
1453
1454         prt_printf(out, "flags:          %x",   le32_to_cpu(clean->flags));
1455         prt_newline(out);
1456         prt_printf(out, "journal_seq:    %llu", le64_to_cpu(clean->journal_seq));
1457         prt_newline(out);
1458
1459         for (entry = clean->start;
1460              entry != vstruct_end(&clean->field);
1461              entry = vstruct_next(entry)) {
1462                 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1463                     !entry->u64s)
1464                         continue;
1465
1466                 bch2_journal_entry_to_text(out, NULL, entry);
1467                 prt_newline(out);
1468         }
1469 }
1470
1471 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1472         .validate       = bch2_sb_clean_validate,
1473         .to_text        = bch2_sb_clean_to_text,
1474 };
1475
1476 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1477 #define x(f, nr)                                        \
1478         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1479         BCH_SB_FIELDS()
1480 #undef x
1481 };
1482
1483 static const struct bch_sb_field_ops bch2_sb_field_null_ops = {
1484         NULL
1485 };
1486
1487 static const struct bch_sb_field_ops *bch2_sb_field_type_ops(unsigned type)
1488 {
1489         return likely(type < ARRAY_SIZE(bch2_sb_field_ops))
1490                 ? bch2_sb_field_ops[type]
1491                 : &bch2_sb_field_null_ops;
1492 }
1493
1494 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1495                                   struct printbuf *err)
1496 {
1497         unsigned type = le32_to_cpu(f->type);
1498         struct printbuf field_err = PRINTBUF;
1499         const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1500         int ret;
1501
1502         ret = ops->validate ? ops->validate(sb, f, &field_err) : 0;
1503         if (ret) {
1504                 prt_printf(err, "Invalid superblock section %s: %s",
1505                            bch2_sb_fields[type], field_err.buf);
1506                 prt_newline(err);
1507                 bch2_sb_field_to_text(err, sb, f);
1508         }
1509
1510         printbuf_exit(&field_err);
1511         return ret;
1512 }
1513
1514 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1515                            struct bch_sb_field *f)
1516 {
1517         unsigned type = le32_to_cpu(f->type);
1518         const struct bch_sb_field_ops *ops = bch2_sb_field_type_ops(type);
1519
1520         if (!out->nr_tabstops)
1521                 printbuf_tabstop_push(out, 32);
1522
1523         if (type < BCH_SB_FIELD_NR)
1524                 prt_printf(out, "%s", bch2_sb_fields[type]);
1525         else
1526                 prt_printf(out, "(unknown field %u)", type);
1527
1528         prt_printf(out, " (size %zu):", vstruct_bytes(f));
1529         prt_newline(out);
1530
1531         if (ops->to_text) {
1532                 printbuf_indent_add(out, 2);
1533                 ops->to_text(out, sb, f);
1534                 printbuf_indent_sub(out, 2);
1535         }
1536 }
1537
1538 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1539 {
1540         unsigned i;
1541
1542         prt_printf(out, "Type:                    %u", l->layout_type);
1543         prt_newline(out);
1544
1545         prt_str(out, "Superblock max size:     ");
1546         prt_units_u64(out, 512 << l->sb_max_size_bits);
1547         prt_newline(out);
1548
1549         prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1550         prt_newline(out);
1551
1552         prt_str(out, "Offsets:                 ");
1553         for (i = 0; i < l->nr_superblocks; i++) {
1554                 if (i)
1555                         prt_str(out, ", ");
1556                 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1557         }
1558         prt_newline(out);
1559 }
1560
1561 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1562                      bool print_layout, unsigned fields)
1563 {
1564         struct bch_sb_field_members *mi;
1565         struct bch_sb_field *f;
1566         u64 fields_have = 0;
1567         unsigned nr_devices = 0;
1568
1569         if (!out->nr_tabstops)
1570                 printbuf_tabstop_push(out, 44);
1571
1572         mi = bch2_sb_get_members(sb);
1573         if (mi) {
1574                 struct bch_member *m;
1575
1576                 for (m = mi->members;
1577                      m < mi->members + sb->nr_devices;
1578                      m++)
1579                         nr_devices += bch2_member_exists(m);
1580         }
1581
1582         prt_printf(out, "External UUID:");
1583         prt_tab(out);
1584         pr_uuid(out, sb->user_uuid.b);
1585         prt_newline(out);
1586
1587         prt_printf(out, "Internal UUID:");
1588         prt_tab(out);
1589         pr_uuid(out, sb->uuid.b);
1590         prt_newline(out);
1591
1592         prt_str(out, "Device index:");
1593         prt_tab(out);
1594         prt_printf(out, "%u", sb->dev_idx);
1595         prt_newline(out);
1596
1597         prt_str(out, "Label:");
1598         prt_tab(out);
1599         prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1600         prt_newline(out);
1601
1602         prt_str(out, "Version:");
1603         prt_tab(out);
1604         bch2_version_to_text(out, le16_to_cpu(sb->version));
1605         prt_newline(out);
1606
1607         prt_str(out, "Version upgrade complete:");
1608         prt_tab(out);
1609         bch2_version_to_text(out, BCH_SB_VERSION_UPGRADE_COMPLETE(sb));
1610         prt_newline(out);
1611
1612         prt_printf(out, "Oldest version on disk:");
1613         prt_tab(out);
1614         bch2_version_to_text(out, le16_to_cpu(sb->version_min));
1615         prt_newline(out);
1616
1617         prt_printf(out, "Created:");
1618         prt_tab(out);
1619         if (sb->time_base_lo)
1620                 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1621         else
1622                 prt_printf(out, "(not set)");
1623         prt_newline(out);
1624
1625         prt_printf(out, "Sequence number:");
1626         prt_tab(out);
1627         prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1628         prt_newline(out);
1629
1630         prt_printf(out, "Superblock size:");
1631         prt_tab(out);
1632         prt_printf(out, "%zu", vstruct_bytes(sb));
1633         prt_newline(out);
1634
1635         prt_printf(out, "Clean:");
1636         prt_tab(out);
1637         prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1638         prt_newline(out);
1639
1640         prt_printf(out, "Devices:");
1641         prt_tab(out);
1642         prt_printf(out, "%u", nr_devices);
1643         prt_newline(out);
1644
1645         prt_printf(out, "Sections:");
1646         vstruct_for_each(sb, f)
1647                 fields_have |= 1 << le32_to_cpu(f->type);
1648         prt_tab(out);
1649         prt_bitflags(out, bch2_sb_fields, fields_have);
1650         prt_newline(out);
1651
1652         prt_printf(out, "Features:");
1653         prt_tab(out);
1654         prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1655         prt_newline(out);
1656
1657         prt_printf(out, "Compat features:");
1658         prt_tab(out);
1659         prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1660         prt_newline(out);
1661
1662         prt_newline(out);
1663         prt_printf(out, "Options:");
1664         prt_newline(out);
1665         printbuf_indent_add(out, 2);
1666         {
1667                 enum bch_opt_id id;
1668
1669                 for (id = 0; id < bch2_opts_nr; id++) {
1670                         const struct bch_option *opt = bch2_opt_table + id;
1671
1672                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1673                                 u64 v = bch2_opt_from_sb(sb, id);
1674
1675                                 prt_printf(out, "%s:", opt->attr.name);
1676                                 prt_tab(out);
1677                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1678                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1679                                 prt_newline(out);
1680                         }
1681                 }
1682         }
1683
1684         printbuf_indent_sub(out, 2);
1685
1686         if (print_layout) {
1687                 prt_newline(out);
1688                 prt_printf(out, "layout:");
1689                 prt_newline(out);
1690                 printbuf_indent_add(out, 2);
1691                 bch2_sb_layout_to_text(out, &sb->layout);
1692                 printbuf_indent_sub(out, 2);
1693         }
1694
1695         vstruct_for_each(sb, f)
1696                 if (fields & (1 << le32_to_cpu(f->type))) {
1697                         prt_newline(out);
1698                         bch2_sb_field_to_text(out, sb, f);
1699                 }
1700 }