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 "disk_groups.h"
   8 #include "ec.h"
   9 #include "error.h"
  10 #include "io.h"
  11 #include "journal.h"
  12 #include "journal_io.h"
  13 #include "journal_seq_blacklist.h"
  14 #include "replicas.h"
  15 #include "quota.h"
  16 #include "super-io.h"
  17 #include "super.h"
  18 #include "vstructs.h"
  19
  20 #include <linux/backing-dev.h>
  21 #include <linux/sort.h>
  22
  23 const char * const bch2_sb_fields[] = {
  24 #define x(name, nr)     #name,
  25         BCH_SB_FIELDS()
  26 #undef x
  27         NULL
  28 };
  29
  30 static const char *bch2_sb_field_validate(struct bch_sb *,
  31                                           struct bch_sb_field *);
  32
  33 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  34                                       enum bch_sb_field_type type)
  35 {
  36         struct bch_sb_field *f;
  37
  38         /* XXX: need locking around superblock to access optional fields */
  39
  40         vstruct_for_each(sb, f)
  41                 if (le32_to_cpu(f->type) == type)
  42                         return f;
  43         return NULL;
  44 }
  45
  46 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  47                                                    struct bch_sb_field *f,
  48                                                    unsigned u64s)
  49 {
  50         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  51         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  52
  53         BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
  54
  55         if (!f && !u64s) {
  56                 /* nothing to do: */
  57         } else if (!f) {
  58                 f = vstruct_last(sb->sb);
  59                 memset(f, 0, sizeof(u64) * u64s);
  60                 f->u64s = cpu_to_le32(u64s);
  61                 f->type = 0;
  62         } else {
  63                 void *src, *dst;
  64
  65                 src = vstruct_end(f);
  66
  67                 if (u64s) {
  68                         f->u64s = cpu_to_le32(u64s);
  69                         dst = vstruct_end(f);
  70                 } else {
  71                         dst = f;
  72                 }
  73
  74                 memmove(dst, src, vstruct_end(sb->sb) - src);
  75
  76                 if (dst > src)
  77                         memset(src, 0, dst - src);
  78         }
  79
  80         sb->sb->u64s = cpu_to_le32(sb_u64s);
  81
  82         return u64s ? f : NULL;
  83 }
  84
  85 void bch2_sb_field_delete(struct bch_sb_handle *sb,
  86                           enum bch_sb_field_type type)
  87 {
  88         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
  89
  90         if (f)
  91                 __bch2_sb_field_resize(sb, f, 0);
  92 }
  93
  94 /* Superblock realloc/free: */
  95
  96 void bch2_free_super(struct bch_sb_handle *sb)
  97 {
  98         if (sb->bio)
  99                 bio_put(sb->bio);
 100         if (!IS_ERR_OR_NULL(sb->bdev))
 101                 blkdev_put(sb->bdev, sb->mode);
 102
 103         kfree(sb->sb);
 104         memset(sb, 0, sizeof(*sb));
 105 }
 106
 107 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 108 {
 109         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 110         size_t new_buffer_size;
 111         struct bch_sb *new_sb;
 112         struct bio *bio;
 113
 114         if (sb->bdev)
 115                 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
 116
 117         new_buffer_size = roundup_pow_of_two(new_bytes);
 118
 119         if (sb->sb && sb->buffer_size >= new_buffer_size)
 120                 return 0;
 121
 122         if (sb->have_layout) {
 123                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 124
 125                 if (new_bytes > max_bytes) {
 126                         char buf[BDEVNAME_SIZE];
 127
 128                         pr_err("%s: superblock too big: want %zu but have %llu",
 129                                bdevname(sb->bdev, buf), new_bytes, max_bytes);
 130                         return -ENOSPC;
 131                 }
 132         }
 133
 134         if (sb->buffer_size >= new_buffer_size && sb->sb)
 135                 return 0;
 136
 137         if (dynamic_fault("bcachefs:add:super_realloc"))
 138                 return -ENOMEM;
 139
 140         if (sb->have_bio) {
 141                 bio = bio_kmalloc(GFP_KERNEL,
 142                         DIV_ROUND_UP(new_buffer_size, PAGE_SIZE));
 143                 if (!bio)
 144                         return -ENOMEM;
 145
 146                 if (sb->bio)
 147                         bio_put(sb->bio);
 148                 sb->bio = bio;
 149         }
 150
 151         new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 152         if (!new_sb)
 153                 return -ENOMEM;
 154
 155         sb->sb = new_sb;
 156         sb->buffer_size = new_buffer_size;
 157
 158         return 0;
 159 }
 160
 161 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 162                                           enum bch_sb_field_type type,
 163                                           unsigned u64s)
 164 {
 165         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 166         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 167         ssize_t d = -old_u64s + u64s;
 168
 169         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 170                 return NULL;
 171
 172         if (sb->fs_sb) {
 173                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 174                 struct bch_dev *ca;
 175                 unsigned i;
 176
 177                 lockdep_assert_held(&c->sb_lock);
 178
 179                 /* XXX: we're not checking that offline device have enough space */
 180
 181                 for_each_online_member(ca, c, i) {
 182                         struct bch_sb_handle *sb = &ca->disk_sb;
 183
 184                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 185                                 percpu_ref_put(&ca->ref);
 186                                 return NULL;
 187                         }
 188                 }
 189         }
 190
 191         f = bch2_sb_field_get(sb->sb, type);
 192         f = __bch2_sb_field_resize(sb, f, u64s);
 193         if (f)
 194                 f->type = cpu_to_le32(type);
 195         return f;
 196 }
 197
 198 /* Superblock validate: */
 199
 200 static inline void __bch2_sb_layout_size_assert(void)
 201 {
 202         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 203 }
 204
 205 static const char *validate_sb_layout(struct bch_sb_layout *layout)
 206 {
 207         u64 offset, prev_offset, max_sectors;
 208         unsigned i;
 209
 210         if (uuid_le_cmp(layout->magic, BCACHE_MAGIC))
 211                 return "Not a bcachefs superblock layout";
 212
 213         if (layout->layout_type != 0)
 214                 return "Invalid superblock layout type";
 215
 216         if (!layout->nr_superblocks)
 217                 return "Invalid superblock layout: no superblocks";
 218
 219         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
 220                 return "Invalid superblock layout: too many superblocks";
 221
 222         max_sectors = 1 << layout->sb_max_size_bits;
 223
 224         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 225
 226         for (i = 1; i < layout->nr_superblocks; i++) {
 227                 offset = le64_to_cpu(layout->sb_offset[i]);
 228
 229                 if (offset < prev_offset + max_sectors)
 230                         return "Invalid superblock layout: superblocks overlap";
 231                 prev_offset = offset;
 232         }
 233
 234         return NULL;
 235 }
 236
 237 const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
 238 {
 239         struct bch_sb *sb = disk_sb->sb;
 240         struct bch_sb_field *f;
 241         struct bch_sb_field_members *mi;
 242         const char *err;
 243         u32 version, version_min;
 244         u16 block_size;
 245
 246         version         = le16_to_cpu(sb->version);
 247         version_min     = version >= bcachefs_metadata_version_new_versioning
 248                 ? le16_to_cpu(sb->version_min)
 249                 : version;
 250
 251         if (version    >= bcachefs_metadata_version_max ||
 252             version_min < bcachefs_metadata_version_min)
 253                 return "Unsupported superblock version";
 254
 255         if (version_min > version)
 256                 return "Bad minimum version";
 257
 258         if (sb->features[1] ||
 259             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
 260                 return "Filesystem has incompatible features";
 261
 262         block_size = le16_to_cpu(sb->block_size);
 263
 264         if (!is_power_of_2(block_size) ||
 265             block_size > PAGE_SECTORS)
 266                 return "Bad block size";
 267
 268         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
 269                 return "Bad user UUID";
 270
 271         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
 272                 return "Bad internal UUID";
 273
 274         if (!sb->nr_devices ||
 275             sb->nr_devices <= sb->dev_idx ||
 276             sb->nr_devices > BCH_SB_MEMBERS_MAX)
 277                 return "Bad number of member devices";
 278
 279         if (!BCH_SB_META_REPLICAS_WANT(sb) ||
 280             BCH_SB_META_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
 281                 return "Invalid number of metadata replicas";
 282
 283         if (!BCH_SB_META_REPLICAS_REQ(sb) ||
 284             BCH_SB_META_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
 285                 return "Invalid number of metadata replicas";
 286
 287         if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
 288             BCH_SB_DATA_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
 289                 return "Invalid number of data replicas";
 290
 291         if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
 292             BCH_SB_DATA_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
 293                 return "Invalid number of data replicas";
 294
 295         if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 296                 return "Invalid metadata checksum type";
 297
 298         if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 299                 return "Invalid metadata checksum type";
 300
 301         if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
 302                 return "Invalid compression type";
 303
 304         if (!BCH_SB_BTREE_NODE_SIZE(sb))
 305                 return "Btree node size not set";
 306
 307         if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
 308                 return "Btree node size not a power of two";
 309
 310         if (BCH_SB_GC_RESERVE(sb) < 5)
 311                 return "gc reserve percentage too small";
 312
 313         if (!sb->time_precision ||
 314             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
 315                 return "invalid time precision";
 316
 317         /* validate layout */
 318         err = validate_sb_layout(&sb->layout);
 319         if (err)
 320                 return err;
 321
 322         vstruct_for_each(sb, f) {
 323                 if (!f->u64s)
 324                         return "Invalid superblock: invalid optional field";
 325
 326                 if (vstruct_next(f) > vstruct_last(sb))
 327                         return "Invalid superblock: invalid optional field";
 328         }
 329
 330         /* members must be validated first: */
 331         mi = bch2_sb_get_members(sb);
 332         if (!mi)
 333                 return "Invalid superblock: member info area missing";
 334
 335         err = bch2_sb_field_validate(sb, &mi->field);
 336         if (err)
 337                 return err;
 338
 339         vstruct_for_each(sb, f) {
 340                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 341                         continue;
 342
 343                 err = bch2_sb_field_validate(sb, f);
 344                 if (err)
 345                         return err;
 346         }
 347
 348         return NULL;
 349 }
 350
 351 /* device open: */
 352
 353 static void bch2_sb_update(struct bch_fs *c)
 354 {
 355         struct bch_sb *src = c->disk_sb.sb;
 356         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 357         struct bch_dev *ca;
 358         unsigned i;
 359
 360         lockdep_assert_held(&c->sb_lock);
 361
 362         c->sb.uuid              = src->uuid;
 363         c->sb.user_uuid         = src->user_uuid;
 364         c->sb.version           = le16_to_cpu(src->version);
 365         c->sb.version_min       = le16_to_cpu(src->version_min);
 366         c->sb.nr_devices        = src->nr_devices;
 367         c->sb.clean             = BCH_SB_CLEAN(src);
 368         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 369         c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
 370
 371         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 372         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 373
 374         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 375         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 376                                           c->sb.nsec_per_time_unit);
 377         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 378
 379         c->sb.features          = le64_to_cpu(src->features[0]);
 380         c->sb.compat            = le64_to_cpu(src->compat[0]);
 381
 382         for_each_member_device(ca, c, i)
 383                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 384 }
 385
 386 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 387 {
 388         struct bch_sb_field *src_f, *dst_f;
 389         struct bch_sb *dst = dst_handle->sb;
 390         unsigned i;
 391
 392         dst->version            = src->version;
 393         dst->version_min        = src->version_min;
 394         dst->seq                = src->seq;
 395         dst->uuid               = src->uuid;
 396         dst->user_uuid          = src->user_uuid;
 397         memcpy(dst->label,      src->label, sizeof(dst->label));
 398
 399         dst->block_size         = src->block_size;
 400         dst->nr_devices         = src->nr_devices;
 401
 402         dst->time_base_lo       = src->time_base_lo;
 403         dst->time_base_hi       = src->time_base_hi;
 404         dst->time_precision     = src->time_precision;
 405
 406         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 407         memcpy(dst->features,   src->features,  sizeof(dst->features));
 408         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 409
 410         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 411                 if (i == BCH_SB_FIELD_journal)
 412                         continue;
 413
 414                 src_f = bch2_sb_field_get(src, i);
 415                 dst_f = bch2_sb_field_get(dst, i);
 416                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 417                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 418
 419                 if (src_f)
 420                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 421         }
 422 }
 423
 424 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 425 {
 426         struct bch_sb_field_journal *journal_buckets =
 427                 bch2_sb_get_journal(src);
 428         unsigned journal_u64s = journal_buckets
 429                 ? le32_to_cpu(journal_buckets->field.u64s)
 430                 : 0;
 431         int ret;
 432
 433         lockdep_assert_held(&c->sb_lock);
 434
 435         ret = bch2_sb_realloc(&c->disk_sb,
 436                               le32_to_cpu(src->u64s) - journal_u64s);
 437         if (ret)
 438                 return ret;
 439
 440         __copy_super(&c->disk_sb, src);
 441
 442         if (BCH_SB_HAS_ERRORS(c->disk_sb.sb))
 443                 set_bit(BCH_FS_ERROR, &c->flags);
 444         else
 445                 clear_bit(BCH_FS_ERROR, &c->flags);
 446
 447         if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb))
 448                 set_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags);
 449         else
 450                 clear_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags);
 451
 452         if (BCH_SB_INITIALIZED(c->disk_sb.sb))
 453                 set_bit(BCH_FS_INITIALIZED, &c->flags);
 454
 455         ret = bch2_sb_replicas_to_cpu_replicas(c);
 456         if (ret)
 457                 return ret;
 458
 459         ret = bch2_sb_disk_groups_to_cpu(c);
 460         if (ret)
 461                 return ret;
 462
 463         bch2_sb_update(c);
 464         return 0;
 465 }
 466
 467 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 468 {
 469         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 470         struct bch_sb_field_journal *journal_buckets =
 471                 bch2_sb_get_journal(dst);
 472         unsigned journal_u64s = journal_buckets
 473                 ? le32_to_cpu(journal_buckets->field.u64s)
 474                 : 0;
 475         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 476         int ret;
 477
 478         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 479         if (ret)
 480                 return ret;
 481
 482         __copy_super(&ca->disk_sb, src);
 483         return 0;
 484 }
 485
 486 /* read superblock: */
 487
 488 static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
 489 {
 490         struct bch_csum csum;
 491         size_t bytes;
 492 reread:
 493         bio_reset(sb->bio);
 494         bio_set_dev(sb->bio, sb->bdev);
 495         sb->bio->bi_iter.bi_sector = offset;
 496         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 497         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 498
 499         if (submit_bio_wait(sb->bio))
 500                 return "IO error";
 501
 502         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
 503                 return "Not a bcachefs superblock";
 504
 505         if (le16_to_cpu(sb->sb->version) <  bcachefs_metadata_version_min ||
 506             le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
 507                 return "Unsupported superblock version";
 508
 509         bytes = vstruct_bytes(sb->sb);
 510
 511         if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
 512                 return "Bad superblock: too big";
 513
 514         if (bytes > sb->buffer_size) {
 515                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 516                         return "cannot allocate memory";
 517                 goto reread;
 518         }
 519
 520         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
 521                 return "unknown csum type";
 522
 523         /* XXX: verify MACs */
 524         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 525                             null_nonce(), sb->sb);
 526
 527         if (bch2_crc_cmp(csum, sb->sb->csum))
 528                 return "bad checksum reading superblock";
 529
 530         sb->seq = le64_to_cpu(sb->sb->seq);
 531
 532         return NULL;
 533 }
 534
 535 int bch2_read_super(const char *path, struct bch_opts *opts,
 536                     struct bch_sb_handle *sb)
 537 {
 538         u64 offset = opt_get(*opts, sb);
 539         struct bch_sb_layout layout;
 540         const char *err;
 541         __le64 *i;
 542         int ret;
 543
 544         pr_verbose_init(*opts, "");
 545
 546         memset(sb, 0, sizeof(*sb));
 547         sb->mode        = FMODE_READ;
 548         sb->have_bio    = true;
 549
 550         if (!opt_get(*opts, noexcl))
 551                 sb->mode |= FMODE_EXCL;
 552
 553         if (!opt_get(*opts, nochanges))
 554                 sb->mode |= FMODE_WRITE;
 555
 556         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 557         if (IS_ERR(sb->bdev) &&
 558             PTR_ERR(sb->bdev) == -EACCES &&
 559             opt_get(*opts, read_only)) {
 560                 sb->mode &= ~FMODE_WRITE;
 561
 562                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 563                 if (!IS_ERR(sb->bdev))
 564                         opt_set(*opts, nochanges, true);
 565         }
 566
 567         if (IS_ERR(sb->bdev)) {
 568                 ret = PTR_ERR(sb->bdev);
 569                 goto out;
 570         }
 571
 572         err = "cannot allocate memory";
 573         ret = bch2_sb_realloc(sb, 0);
 574         if (ret)
 575                 goto err;
 576
 577         ret = -EFAULT;
 578         err = "dynamic fault";
 579         if (bch2_fs_init_fault("read_super"))
 580                 goto err;
 581
 582         ret = -EINVAL;
 583         err = read_one_super(sb, offset);
 584         if (!err)
 585                 goto got_super;
 586
 587         if (opt_defined(*opts, sb))
 588                 goto err;
 589
 590         pr_err("error reading default superblock: %s", err);
 591
 592         /*
 593          * Error reading primary superblock - read location of backup
 594          * superblocks:
 595          */
 596         bio_reset(sb->bio);
 597         bio_set_dev(sb->bio, sb->bdev);
 598         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 599         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 600         /*
 601          * use sb buffer to read layout, since sb buffer is page aligned but
 602          * layout won't be:
 603          */
 604         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 605
 606         err = "IO error";
 607         if (submit_bio_wait(sb->bio))
 608                 goto err;
 609
 610         memcpy(&layout, sb->sb, sizeof(layout));
 611         err = validate_sb_layout(&layout);
 612         if (err)
 613                 goto err;
 614
 615         for (i = layout.sb_offset;
 616              i < layout.sb_offset + layout.nr_superblocks; i++) {
 617                 offset = le64_to_cpu(*i);
 618
 619                 if (offset == opt_get(*opts, sb))
 620                         continue;
 621
 622                 err = read_one_super(sb, offset);
 623                 if (!err)
 624                         goto got_super;
 625         }
 626
 627         ret = -EINVAL;
 628         goto err;
 629
 630 got_super:
 631         err = "Superblock block size smaller than device block size";
 632         ret = -EINVAL;
 633         if (le16_to_cpu(sb->sb->block_size) << 9 <
 634             bdev_logical_block_size(sb->bdev))
 635                 goto err;
 636
 637         ret = 0;
 638         sb->have_layout = true;
 639 out:
 640         pr_verbose_init(*opts, "ret %i", ret);
 641         return ret;
 642 err:
 643         bch2_free_super(sb);
 644         pr_err("error reading superblock: %s", err);
 645         goto out;
 646 }
 647
 648 /* write superblock: */
 649
 650 static void write_super_endio(struct bio *bio)
 651 {
 652         struct bch_dev *ca = bio->bi_private;
 653
 654         /* XXX: return errors directly */
 655
 656         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 657                                bch2_blk_status_to_str(bio->bi_status)))
 658                 ca->sb_write_error = 1;
 659
 660         closure_put(&ca->fs->sb_write);
 661         percpu_ref_put(&ca->io_ref);
 662 }
 663
 664 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 665 {
 666         struct bch_sb *sb = ca->disk_sb.sb;
 667         struct bio *bio = ca->disk_sb.bio;
 668
 669         bio_reset(bio);
 670         bio_set_dev(bio, ca->disk_sb.bdev);
 671         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 672         bio->bi_end_io          = write_super_endio;
 673         bio->bi_private         = ca;
 674         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 675         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 676
 677         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 678                      bio_sectors(bio));
 679
 680         percpu_ref_get(&ca->io_ref);
 681         closure_bio_submit(bio, &c->sb_write);
 682 }
 683
 684 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 685 {
 686         struct bch_sb *sb = ca->disk_sb.sb;
 687         struct bio *bio = ca->disk_sb.bio;
 688
 689         sb->offset = sb->layout.sb_offset[idx];
 690
 691         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 692         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 693                                 null_nonce(), sb);
 694
 695         bio_reset(bio);
 696         bio_set_dev(bio, ca->disk_sb.bdev);
 697         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 698         bio->bi_end_io          = write_super_endio;
 699         bio->bi_private         = ca;
 700         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 701         bch2_bio_map(bio, sb,
 702                      roundup((size_t) vstruct_bytes(sb),
 703                              bdev_logical_block_size(ca->disk_sb.bdev)));
 704
 705         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 706                      bio_sectors(bio));
 707
 708         percpu_ref_get(&ca->io_ref);
 709         closure_bio_submit(bio, &c->sb_write);
 710 }
 711
 712 int bch2_write_super(struct bch_fs *c)
 713 {
 714         struct closure *cl = &c->sb_write;
 715         struct bch_dev *ca;
 716         unsigned i, sb = 0, nr_wrote;
 717         const char *err;
 718         struct bch_devs_mask sb_written;
 719         bool wrote, can_mount_without_written, can_mount_with_written;
 720         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 721         int ret = 0;
 722
 723         if (c->opts.very_degraded)
 724                 degraded_flags |= BCH_FORCE_IF_LOST;
 725
 726         lockdep_assert_held(&c->sb_lock);
 727
 728         closure_init_stack(cl);
 729         memset(&sb_written, 0, sizeof(sb_written));
 730
 731         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 732
 733         if (test_bit(BCH_FS_ERROR, &c->flags))
 734                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 735         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 736                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 737
 738         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 739
 740         for_each_online_member(ca, c, i)
 741                 bch2_sb_from_fs(c, ca);
 742
 743         for_each_online_member(ca, c, i) {
 744                 err = bch2_sb_validate(&ca->disk_sb);
 745                 if (err) {
 746                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
 747                         ret = -1;
 748                         goto out;
 749                 }
 750         }
 751
 752         if (c->opts.nochanges)
 753                 goto out;
 754
 755         for_each_online_member(ca, c, i) {
 756                 __set_bit(ca->dev_idx, sb_written.d);
 757                 ca->sb_write_error = 0;
 758         }
 759
 760         for_each_online_member(ca, c, i)
 761                 read_back_super(c, ca);
 762         closure_sync(cl);
 763
 764         for_each_online_member(ca, c, i) {
 765                 if (!ca->sb_write_error &&
 766                     ca->disk_sb.seq !=
 767                     le64_to_cpu(ca->sb_read_scratch->seq)) {
 768                         bch2_fs_fatal_error(c,
 769                                 "Superblock modified by another process");
 770                         percpu_ref_put(&ca->io_ref);
 771                         ret = -EROFS;
 772                         goto out;
 773                 }
 774         }
 775
 776         do {
 777                 wrote = false;
 778                 for_each_online_member(ca, c, i)
 779                         if (!ca->sb_write_error &&
 780                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 781                                 write_one_super(c, ca, sb);
 782                                 wrote = true;
 783                         }
 784                 closure_sync(cl);
 785                 sb++;
 786         } while (wrote);
 787
 788         for_each_online_member(ca, c, i) {
 789                 if (ca->sb_write_error)
 790                         __clear_bit(ca->dev_idx, sb_written.d);
 791                 else
 792                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 793         }
 794
 795         nr_wrote = dev_mask_nr(&sb_written);
 796
 797         can_mount_with_written =
 798                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 799
 800         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 801                 sb_written.d[i] = ~sb_written.d[i];
 802
 803         can_mount_without_written =
 804                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 805
 806         /*
 807          * If we would be able to mount _without_ the devices we successfully
 808          * wrote superblocks to, we weren't able to write to enough devices:
 809          *
 810          * Exception: if we can mount without the successes because we haven't
 811          * written anything (new filesystem), we continue if we'd be able to
 812          * mount with the devices we did successfully write to:
 813          */
 814         if (bch2_fs_fatal_err_on(!nr_wrote ||
 815                                  !can_mount_with_written ||
 816                                  (can_mount_without_written &&
 817                                   !can_mount_with_written), c,
 818                 "Unable to write superblock to sufficient devices (from %ps)",
 819                 (void *) _RET_IP_))
 820                 ret = -1;
 821 out:
 822         /* Make new options visible after they're persistent: */
 823         bch2_sb_update(c);
 824         return ret;
 825 }
 826
 827 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 828 {
 829         mutex_lock(&c->sb_lock);
 830         if (!(c->sb.features & (1ULL << feat))) {
 831                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 832
 833                 bch2_write_super(c);
 834         }
 835         mutex_unlock(&c->sb_lock);
 836 }
 837
 838 /* BCH_SB_FIELD_journal: */
 839
 840 static int u64_cmp(const void *_l, const void *_r)
 841 {
 842         u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
 843
 844         return l < r ? -1 : l > r ? 1 : 0;
 845 }
 846
 847 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
 848                                             struct bch_sb_field *f)
 849 {
 850         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 851         struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
 852         const char *err;
 853         unsigned nr;
 854         unsigned i;
 855         u64 *b;
 856
 857         journal = bch2_sb_get_journal(sb);
 858         if (!journal)
 859                 return NULL;
 860
 861         nr = bch2_nr_journal_buckets(journal);
 862         if (!nr)
 863                 return NULL;
 864
 865         b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
 866         if (!b)
 867                 return "cannot allocate memory";
 868
 869         for (i = 0; i < nr; i++)
 870                 b[i] = le64_to_cpu(journal->buckets[i]);
 871
 872         sort(b, nr, sizeof(u64), u64_cmp, NULL);
 873
 874         err = "journal bucket at sector 0";
 875         if (!b[0])
 876                 goto err;
 877
 878         err = "journal bucket before first bucket";
 879         if (m && b[0] < le16_to_cpu(m->first_bucket))
 880                 goto err;
 881
 882         err = "journal bucket past end of device";
 883         if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
 884                 goto err;
 885
 886         err = "duplicate journal buckets";
 887         for (i = 0; i + 1 < nr; i++)
 888                 if (b[i] == b[i + 1])
 889                         goto err;
 890
 891         err = NULL;
 892 err:
 893         kfree(b);
 894         return err;
 895 }
 896
 897 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
 898         .validate       = bch2_sb_validate_journal,
 899 };
 900
 901 /* BCH_SB_FIELD_members: */
 902
 903 static const char *bch2_sb_validate_members(struct bch_sb *sb,
 904                                             struct bch_sb_field *f)
 905 {
 906         struct bch_sb_field_members *mi = field_to_type(f, members);
 907         struct bch_member *m;
 908
 909         if ((void *) (mi->members + sb->nr_devices) >
 910             vstruct_end(&mi->field))
 911                 return "Invalid superblock: bad member info";
 912
 913         for (m = mi->members;
 914              m < mi->members + sb->nr_devices;
 915              m++) {
 916                 if (!bch2_member_exists(m))
 917                         continue;
 918
 919                 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
 920                         return "Too many buckets";
 921
 922                 if (le64_to_cpu(m->nbuckets) -
 923                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
 924                         return "Not enough buckets";
 925
 926                 if (le16_to_cpu(m->bucket_size) <
 927                     le16_to_cpu(sb->block_size))
 928                         return "bucket size smaller than block size";
 929
 930                 if (le16_to_cpu(m->bucket_size) <
 931                     BCH_SB_BTREE_NODE_SIZE(sb))
 932                         return "bucket size smaller than btree node size";
 933         }
 934
 935         return NULL;
 936 }
 937
 938 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
 939         .validate       = bch2_sb_validate_members,
 940 };
 941
 942 /* BCH_SB_FIELD_crypt: */
 943
 944 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
 945                                           struct bch_sb_field *f)
 946 {
 947         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
 948
 949         if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
 950                 return "invalid field crypt: wrong size";
 951
 952         if (BCH_CRYPT_KDF_TYPE(crypt))
 953                 return "invalid field crypt: bad kdf type";
 954
 955         return NULL;
 956 }
 957
 958 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
 959         .validate       = bch2_sb_validate_crypt,
 960 };
 961
 962 /* BCH_SB_FIELD_clean: */
 963
 964 int bch2_sb_clean_validate(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
 965 {
 966         struct jset_entry *entry;
 967         int ret;
 968
 969         for (entry = clean->start;
 970              entry < (struct jset_entry *) vstruct_end(&clean->field);
 971              entry = vstruct_next(entry)) {
 972                 ret = bch2_journal_entry_validate(c, "superblock", entry,
 973                                                   le16_to_cpu(c->disk_sb.sb->version),
 974                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
 975                                                   write);
 976                 if (ret)
 977                         return ret;
 978         }
 979
 980         return 0;
 981 }
 982
 983 int bch2_fs_mark_dirty(struct bch_fs *c)
 984 {
 985         int ret;
 986
 987         /*
 988          * Unconditionally write superblock, to verify it hasn't changed before
 989          * we go rw:
 990          */
 991
 992         mutex_lock(&c->sb_lock);
 993         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 994         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
 995         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
 996         ret = bch2_write_super(c);
 997         mutex_unlock(&c->sb_lock);
 998
 999         return ret;
1000 }
1001
1002 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1003 {
1004         struct jset_entry *entry = *end;
1005         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1006
1007         memset(entry, 0, u64s * sizeof(u64));
1008         /*
1009          * The u64s field counts from the start of data, ignoring the shared
1010          * fields.
1011          */
1012         entry->u64s = cpu_to_le16(u64s - 1);
1013
1014         *end = vstruct_next(*end);
1015         return entry;
1016 }
1017
1018 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1019                                            struct jset_entry **end,
1020                                            u64 journal_seq)
1021 {
1022         struct bch_dev *ca;
1023         unsigned i, dev;
1024
1025         percpu_down_read(&c->mark_lock);
1026
1027         if (!journal_seq) {
1028                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1029                         bch2_fs_usage_acc_to_base(c, i);
1030         } else {
1031                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1032         }
1033
1034         {
1035                 struct jset_entry_usage *u =
1036                         container_of(jset_entry_init(end, sizeof(*u)),
1037                                      struct jset_entry_usage, entry);
1038
1039                 u->entry.type   = BCH_JSET_ENTRY_usage;
1040                 u->entry.btree_id = FS_USAGE_INODES;
1041                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1042         }
1043
1044         {
1045                 struct jset_entry_usage *u =
1046                         container_of(jset_entry_init(end, sizeof(*u)),
1047                                      struct jset_entry_usage, entry);
1048
1049                 u->entry.type   = BCH_JSET_ENTRY_usage;
1050                 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1051                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1052         }
1053
1054         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1055                 struct jset_entry_usage *u =
1056                         container_of(jset_entry_init(end, sizeof(*u)),
1057                                      struct jset_entry_usage, entry);
1058
1059                 u->entry.type   = BCH_JSET_ENTRY_usage;
1060                 u->entry.btree_id = FS_USAGE_RESERVED;
1061                 u->entry.level  = i;
1062                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1063         }
1064
1065         for (i = 0; i < c->replicas.nr; i++) {
1066                 struct bch_replicas_entry *e =
1067                         cpu_replicas_entry(&c->replicas, i);
1068                 struct jset_entry_data_usage *u =
1069                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1070                                      struct jset_entry_data_usage, entry);
1071
1072                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1073                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1074                 memcpy(&u->r, e, replicas_entry_bytes(e));
1075         }
1076
1077         for_each_member_device(ca, c, dev) {
1078                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1079                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1080                 struct jset_entry_dev_usage *u =
1081                         container_of(jset_entry_init(end, b),
1082                                      struct jset_entry_dev_usage, entry);
1083
1084                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1085                 u->dev = cpu_to_le32(dev);
1086                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1087                 u->buckets_unavailable  = cpu_to_le64(ca->usage_base->buckets_unavailable);
1088
1089                 for (i = 0; i < BCH_DATA_NR; i++) {
1090                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1091                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1092                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1093                 }
1094         }
1095
1096         percpu_up_read(&c->mark_lock);
1097
1098         for (i = 0; i < 2; i++) {
1099                 struct jset_entry_clock *clock =
1100                         container_of(jset_entry_init(end, sizeof(*clock)),
1101                                      struct jset_entry_clock, entry);
1102
1103                 clock->entry.type = BCH_JSET_ENTRY_clock;
1104                 clock->rw       = i;
1105                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1106         }
1107 }
1108
1109 void bch2_fs_mark_clean(struct bch_fs *c)
1110 {
1111         struct bch_sb_field_clean *sb_clean;
1112         struct jset_entry *entry;
1113         unsigned u64s;
1114         int ret;
1115
1116         mutex_lock(&c->sb_lock);
1117         if (BCH_SB_CLEAN(c->disk_sb.sb))
1118                 goto out;
1119
1120         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1121
1122         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1123         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1124         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1125         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1126
1127         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1128
1129         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1130         if (!sb_clean) {
1131                 bch_err(c, "error resizing superblock while setting filesystem clean");
1132                 goto out;
1133         }
1134
1135         sb_clean->flags         = 0;
1136         sb_clean->journal_seq   = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1137
1138         /* Trying to catch outstanding bug: */
1139         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1140
1141         entry = sb_clean->start;
1142         bch2_journal_super_entries_add_common(c, &entry, 0);
1143         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1144         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1145
1146         memset(entry, 0,
1147                vstruct_end(&sb_clean->field) - (void *) entry);
1148
1149         /*
1150          * this should be in the write path, and we should be validating every
1151          * superblock section:
1152          */
1153         ret = bch2_sb_clean_validate(c, sb_clean, WRITE);
1154         if (ret) {
1155                 bch_err(c, "error writing marking filesystem clean: validate error");
1156                 goto out;
1157         }
1158
1159         bch2_write_super(c);
1160 out:
1161         mutex_unlock(&c->sb_lock);
1162 }
1163
1164 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1165                                           struct bch_sb_field *f)
1166 {
1167         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1168
1169         if (vstruct_bytes(&clean->field) < sizeof(*clean))
1170                 return "invalid field crypt: wrong size";
1171
1172         return NULL;
1173 }
1174
1175 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1176         .validate       = bch2_sb_validate_clean,
1177 };
1178
1179 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1180 #define x(f, nr)                                        \
1181         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1182         BCH_SB_FIELDS()
1183 #undef x
1184 };
1185
1186 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1187                                           struct bch_sb_field *f)
1188 {
1189         unsigned type = le32_to_cpu(f->type);
1190
1191         return type < BCH_SB_FIELD_NR
1192                 ? bch2_sb_field_ops[type]->validate(sb, f)
1193                 : NULL;
1194 }
1195
1196 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1197                            struct bch_sb_field *f)
1198 {
1199         unsigned type = le32_to_cpu(f->type);
1200         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1201                 ? bch2_sb_field_ops[type] : NULL;
1202
1203         if (ops)
1204                 pr_buf(out, "%s", bch2_sb_fields[type]);
1205         else
1206                 pr_buf(out, "(unknown field %u)", type);
1207
1208         pr_buf(out, " (size %llu):", vstruct_bytes(f));
1209
1210         if (ops && ops->to_text)
1211                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1212 }