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 "buckets.h"
   5 #include "checksum.h"
   6 #include "disk_groups.h"
   7 #include "ec.h"
   8 #include "error.h"
   9 #include "io.h"
  10 #include "journal.h"
  11 #include "journal_seq_blacklist.h"
  12 #include "replicas.h"
  13 #include "quota.h"
  14 #include "super-io.h"
  15 #include "super.h"
  16 #include "vstructs.h"
  17
  18 #include <linux/backing-dev.h>
  19 #include <linux/sort.h>
  20
  21 const char * const bch2_sb_fields[] = {
  22 #define x(name, nr)     #name,
  23         BCH_SB_FIELDS()
  24 #undef x
  25         NULL
  26 };
  27
  28 static const char *bch2_sb_field_validate(struct bch_sb *,
  29                                           struct bch_sb_field *);
  30
  31 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  32                                       enum bch_sb_field_type type)
  33 {
  34         struct bch_sb_field *f;
  35
  36         /* XXX: need locking around superblock to access optional fields */
  37
  38         vstruct_for_each(sb, f)
  39                 if (le32_to_cpu(f->type) == type)
  40                         return f;
  41         return NULL;
  42 }
  43
  44 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  45                                                    struct bch_sb_field *f,
  46                                                    unsigned u64s)
  47 {
  48         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  49         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  50
  51         BUG_ON(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
  52                sb->page_order);
  53
  54         if (!f) {
  55                 f = vstruct_last(sb->sb);
  56                 memset(f, 0, sizeof(u64) * u64s);
  57                 f->u64s = cpu_to_le32(u64s);
  58                 f->type = 0;
  59         } else {
  60                 void *src, *dst;
  61
  62                 src = vstruct_end(f);
  63
  64                 if (u64s) {
  65                         f->u64s = cpu_to_le32(u64s);
  66                         dst = vstruct_end(f);
  67                 } else {
  68                         dst = f;
  69                 }
  70
  71                 memmove(dst, src, vstruct_end(sb->sb) - src);
  72
  73                 if (dst > src)
  74                         memset(src, 0, dst - src);
  75         }
  76
  77         sb->sb->u64s = cpu_to_le32(sb_u64s);
  78
  79         return u64s ? f : NULL;
  80 }
  81
  82 void bch2_sb_field_delete(struct bch_sb_handle *sb,
  83                           enum bch_sb_field_type type)
  84 {
  85         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
  86
  87         if (f)
  88                 __bch2_sb_field_resize(sb, f, 0);
  89 }
  90
  91 /* Superblock realloc/free: */
  92
  93 void bch2_free_super(struct bch_sb_handle *sb)
  94 {
  95         if (sb->bio)
  96                 bio_put(sb->bio);
  97         if (!IS_ERR_OR_NULL(sb->bdev))
  98                 blkdev_put(sb->bdev, sb->mode);
  99
 100         free_pages((unsigned long) sb->sb, sb->page_order);
 101         memset(sb, 0, sizeof(*sb));
 102 }
 103
 104 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 105 {
 106         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 107         unsigned order = get_order(new_bytes);
 108         struct bch_sb *new_sb;
 109         struct bio *bio;
 110
 111         if (sb->sb && sb->page_order >= order)
 112                 return 0;
 113
 114         if (sb->have_layout) {
 115                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 116
 117                 if (new_bytes > max_bytes) {
 118                         char buf[BDEVNAME_SIZE];
 119
 120                         pr_err("%s: superblock too big: want %zu but have %llu",
 121                                bdevname(sb->bdev, buf), new_bytes, max_bytes);
 122                         return -ENOSPC;
 123                 }
 124         }
 125
 126         if (sb->page_order >= order && sb->sb)
 127                 return 0;
 128
 129         if (dynamic_fault("bcachefs:add:super_realloc"))
 130                 return -ENOMEM;
 131
 132         if (sb->have_bio) {
 133                 bio = bio_kmalloc(GFP_KERNEL, 1 << order);
 134                 if (!bio)
 135                         return -ENOMEM;
 136
 137                 if (sb->bio)
 138                         bio_put(sb->bio);
 139                 sb->bio = bio;
 140         }
 141
 142         new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
 143         if (!new_sb)
 144                 return -ENOMEM;
 145
 146         if (sb->sb)
 147                 memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
 148
 149         free_pages((unsigned long) sb->sb, sb->page_order);
 150         sb->sb = new_sb;
 151
 152         sb->page_order = order;
 153
 154         return 0;
 155 }
 156
 157 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 158                                           enum bch_sb_field_type type,
 159                                           unsigned u64s)
 160 {
 161         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 162         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 163         ssize_t d = -old_u64s + u64s;
 164
 165         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 166                 return NULL;
 167
 168         if (sb->fs_sb) {
 169                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 170                 struct bch_dev *ca;
 171                 unsigned i;
 172
 173                 lockdep_assert_held(&c->sb_lock);
 174
 175                 /* XXX: we're not checking that offline device have enough space */
 176
 177                 for_each_online_member(ca, c, i) {
 178                         struct bch_sb_handle *sb = &ca->disk_sb;
 179
 180                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 181                                 percpu_ref_put(&ca->ref);
 182                                 return NULL;
 183                         }
 184                 }
 185         }
 186
 187         f = bch2_sb_field_get(sb->sb, type);
 188         f = __bch2_sb_field_resize(sb, f, u64s);
 189         if (f)
 190                 f->type = cpu_to_le32(type);
 191         return f;
 192 }
 193
 194 /* Superblock validate: */
 195
 196 static inline void __bch2_sb_layout_size_assert(void)
 197 {
 198         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 199 }
 200
 201 static const char *validate_sb_layout(struct bch_sb_layout *layout)
 202 {
 203         u64 offset, prev_offset, max_sectors;
 204         unsigned i;
 205
 206         if (uuid_le_cmp(layout->magic, BCACHE_MAGIC))
 207                 return "Not a bcachefs superblock layout";
 208
 209         if (layout->layout_type != 0)
 210                 return "Invalid superblock layout type";
 211
 212         if (!layout->nr_superblocks)
 213                 return "Invalid superblock layout: no superblocks";
 214
 215         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
 216                 return "Invalid superblock layout: too many superblocks";
 217
 218         max_sectors = 1 << layout->sb_max_size_bits;
 219
 220         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 221
 222         for (i = 1; i < layout->nr_superblocks; i++) {
 223                 offset = le64_to_cpu(layout->sb_offset[i]);
 224
 225                 if (offset < prev_offset + max_sectors)
 226                         return "Invalid superblock layout: superblocks overlap";
 227                 prev_offset = offset;
 228         }
 229
 230         return NULL;
 231 }
 232
 233 const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
 234 {
 235         struct bch_sb *sb = disk_sb->sb;
 236         struct bch_sb_field *f;
 237         struct bch_sb_field_members *mi;
 238         const char *err;
 239         u32 version, version_min;
 240         u16 block_size;
 241
 242         version         = le16_to_cpu(sb->version);
 243         version_min     = version >= bcachefs_metadata_version_new_versioning
 244                 ? le16_to_cpu(sb->version_min)
 245                 : version;
 246
 247         if (version    >= bcachefs_metadata_version_max ||
 248             version_min < bcachefs_metadata_version_min)
 249                 return "Unsupported superblock version";
 250
 251         if (version_min > version)
 252                 return "Bad minimum version";
 253
 254         if (sb->features[1] ||
 255             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
 256                 return "Filesystem has incompatible features";
 257
 258         block_size = le16_to_cpu(sb->block_size);
 259
 260         if (!is_power_of_2(block_size) ||
 261             block_size > PAGE_SECTORS)
 262                 return "Bad block size";
 263
 264         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
 265                 return "Bad user UUID";
 266
 267         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
 268                 return "Bad internal UUID";
 269
 270         if (!sb->nr_devices ||
 271             sb->nr_devices <= sb->dev_idx ||
 272             sb->nr_devices > BCH_SB_MEMBERS_MAX)
 273                 return "Bad number of member devices";
 274
 275         if (!BCH_SB_META_REPLICAS_WANT(sb) ||
 276             BCH_SB_META_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
 277                 return "Invalid number of metadata replicas";
 278
 279         if (!BCH_SB_META_REPLICAS_REQ(sb) ||
 280             BCH_SB_META_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
 281                 return "Invalid number of metadata replicas";
 282
 283         if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
 284             BCH_SB_DATA_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
 285                 return "Invalid number of data replicas";
 286
 287         if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
 288             BCH_SB_DATA_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
 289                 return "Invalid number of data replicas";
 290
 291         if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 292                 return "Invalid metadata checksum type";
 293
 294         if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 295                 return "Invalid metadata checksum type";
 296
 297         if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
 298                 return "Invalid compression type";
 299
 300         if (!BCH_SB_BTREE_NODE_SIZE(sb))
 301                 return "Btree node size not set";
 302
 303         if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
 304                 return "Btree node size not a power of two";
 305
 306         if (BCH_SB_GC_RESERVE(sb) < 5)
 307                 return "gc reserve percentage too small";
 308
 309         if (!sb->time_precision ||
 310             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
 311                 return "invalid time precision";
 312
 313         /* validate layout */
 314         err = validate_sb_layout(&sb->layout);
 315         if (err)
 316                 return err;
 317
 318         vstruct_for_each(sb, f) {
 319                 if (!f->u64s)
 320                         return "Invalid superblock: invalid optional field";
 321
 322                 if (vstruct_next(f) > vstruct_last(sb))
 323                         return "Invalid superblock: invalid optional field";
 324         }
 325
 326         /* members must be validated first: */
 327         mi = bch2_sb_get_members(sb);
 328         if (!mi)
 329                 return "Invalid superblock: member info area missing";
 330
 331         err = bch2_sb_field_validate(sb, &mi->field);
 332         if (err)
 333                 return err;
 334
 335         vstruct_for_each(sb, f) {
 336                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 337                         continue;
 338
 339                 err = bch2_sb_field_validate(sb, f);
 340                 if (err)
 341                         return err;
 342         }
 343
 344         return NULL;
 345 }
 346
 347 /* device open: */
 348
 349 static void bch2_sb_update(struct bch_fs *c)
 350 {
 351         struct bch_sb *src = c->disk_sb.sb;
 352         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 353         struct bch_dev *ca;
 354         unsigned i;
 355
 356         lockdep_assert_held(&c->sb_lock);
 357
 358         c->sb.uuid              = src->uuid;
 359         c->sb.user_uuid         = src->user_uuid;
 360         c->sb.version           = le16_to_cpu(src->version);
 361         c->sb.nr_devices        = src->nr_devices;
 362         c->sb.clean             = BCH_SB_CLEAN(src);
 363         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 364         c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
 365         c->sb.time_base_lo      = le64_to_cpu(src->time_base_lo);
 366         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 367         c->sb.time_precision    = le32_to_cpu(src->time_precision);
 368         c->sb.features          = le64_to_cpu(src->features[0]);
 369         c->sb.compat            = le64_to_cpu(src->compat[0]);
 370
 371         for_each_member_device(ca, c, i)
 372                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 373 }
 374
 375 /* doesn't copy member info */
 376 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 377 {
 378         struct bch_sb_field *src_f, *dst_f;
 379         struct bch_sb *dst = dst_handle->sb;
 380         unsigned i;
 381
 382         dst->version            = src->version;
 383         dst->version_min        = src->version_min;
 384         dst->seq                = src->seq;
 385         dst->uuid               = src->uuid;
 386         dst->user_uuid          = src->user_uuid;
 387         memcpy(dst->label,      src->label, sizeof(dst->label));
 388
 389         dst->block_size         = src->block_size;
 390         dst->nr_devices         = src->nr_devices;
 391
 392         dst->time_base_lo       = src->time_base_lo;
 393         dst->time_base_hi       = src->time_base_hi;
 394         dst->time_precision     = src->time_precision;
 395
 396         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 397         memcpy(dst->features,   src->features,  sizeof(dst->features));
 398         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 399
 400         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 401                 if (i == BCH_SB_FIELD_journal)
 402                         continue;
 403
 404                 src_f = bch2_sb_field_get(src, i);
 405                 dst_f = bch2_sb_field_get(dst, i);
 406                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 407                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 408
 409                 if (src_f)
 410                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 411         }
 412 }
 413
 414 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 415 {
 416         struct bch_sb_field_journal *journal_buckets =
 417                 bch2_sb_get_journal(src);
 418         unsigned journal_u64s = journal_buckets
 419                 ? le32_to_cpu(journal_buckets->field.u64s)
 420                 : 0;
 421         int ret;
 422
 423         lockdep_assert_held(&c->sb_lock);
 424
 425         ret = bch2_sb_realloc(&c->disk_sb,
 426                               le32_to_cpu(src->u64s) - journal_u64s);
 427         if (ret)
 428                 return ret;
 429
 430         __copy_super(&c->disk_sb, src);
 431
 432         ret = bch2_sb_replicas_to_cpu_replicas(c);
 433         if (ret)
 434                 return ret;
 435
 436         ret = bch2_sb_disk_groups_to_cpu(c);
 437         if (ret)
 438                 return ret;
 439
 440         bch2_sb_update(c);
 441         return 0;
 442 }
 443
 444 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 445 {
 446         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 447         struct bch_sb_field_journal *journal_buckets =
 448                 bch2_sb_get_journal(dst);
 449         unsigned journal_u64s = journal_buckets
 450                 ? le32_to_cpu(journal_buckets->field.u64s)
 451                 : 0;
 452         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 453         int ret;
 454
 455         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 456         if (ret)
 457                 return ret;
 458
 459         __copy_super(&ca->disk_sb, src);
 460         return 0;
 461 }
 462
 463 /* read superblock: */
 464
 465 static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
 466 {
 467         struct bch_csum csum;
 468         size_t bytes;
 469 reread:
 470         bio_reset(sb->bio);
 471         bio_set_dev(sb->bio, sb->bdev);
 472         sb->bio->bi_iter.bi_sector = offset;
 473         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 474         bch2_bio_map(sb->bio, sb->sb, PAGE_SIZE << sb->page_order);
 475
 476         if (submit_bio_wait(sb->bio))
 477                 return "IO error";
 478
 479         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
 480                 return "Not a bcachefs superblock";
 481
 482         if (le16_to_cpu(sb->sb->version) <  bcachefs_metadata_version_min ||
 483             le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
 484                 return "Unsupported superblock version";
 485
 486         bytes = vstruct_bytes(sb->sb);
 487
 488         if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
 489                 return "Bad superblock: too big";
 490
 491         if (get_order(bytes) > sb->page_order) {
 492                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 493                         return "cannot allocate memory";
 494                 goto reread;
 495         }
 496
 497         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
 498                 return "unknown csum type";
 499
 500         /* XXX: verify MACs */
 501         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 502                             null_nonce(), sb->sb);
 503
 504         if (bch2_crc_cmp(csum, sb->sb->csum))
 505                 return "bad checksum reading superblock";
 506
 507         sb->seq = le64_to_cpu(sb->sb->seq);
 508
 509         return NULL;
 510 }
 511
 512 int bch2_read_super(const char *path, struct bch_opts *opts,
 513                     struct bch_sb_handle *sb)
 514 {
 515         u64 offset = opt_get(*opts, sb);
 516         struct bch_sb_layout layout;
 517         const char *err;
 518         __le64 *i;
 519         int ret;
 520
 521         pr_verbose_init(*opts, "");
 522
 523         memset(sb, 0, sizeof(*sb));
 524         sb->mode        = FMODE_READ;
 525         sb->have_bio    = true;
 526
 527         if (!opt_get(*opts, noexcl))
 528                 sb->mode |= FMODE_EXCL;
 529
 530         if (!opt_get(*opts, nochanges))
 531                 sb->mode |= FMODE_WRITE;
 532
 533         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 534         if (IS_ERR(sb->bdev) &&
 535             PTR_ERR(sb->bdev) == -EACCES &&
 536             opt_get(*opts, read_only)) {
 537                 sb->mode &= ~FMODE_WRITE;
 538
 539                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 540                 if (!IS_ERR(sb->bdev))
 541                         opt_set(*opts, nochanges, true);
 542         }
 543
 544         if (IS_ERR(sb->bdev)) {
 545                 ret = PTR_ERR(sb->bdev);
 546                 goto out;
 547         }
 548
 549         err = "cannot allocate memory";
 550         ret = bch2_sb_realloc(sb, 0);
 551         if (ret)
 552                 goto err;
 553
 554         ret = -EFAULT;
 555         err = "dynamic fault";
 556         if (bch2_fs_init_fault("read_super"))
 557                 goto err;
 558
 559         ret = -EINVAL;
 560         err = read_one_super(sb, offset);
 561         if (!err)
 562                 goto got_super;
 563
 564         if (opt_defined(*opts, sb))
 565                 goto err;
 566
 567         pr_err("error reading default superblock: %s", err);
 568
 569         /*
 570          * Error reading primary superblock - read location of backup
 571          * superblocks:
 572          */
 573         bio_reset(sb->bio);
 574         bio_set_dev(sb->bio, sb->bdev);
 575         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 576         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 577         /*
 578          * use sb buffer to read layout, since sb buffer is page aligned but
 579          * layout won't be:
 580          */
 581         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 582
 583         err = "IO error";
 584         if (submit_bio_wait(sb->bio))
 585                 goto err;
 586
 587         memcpy(&layout, sb->sb, sizeof(layout));
 588         err = validate_sb_layout(&layout);
 589         if (err)
 590                 goto err;
 591
 592         for (i = layout.sb_offset;
 593              i < layout.sb_offset + layout.nr_superblocks; i++) {
 594                 offset = le64_to_cpu(*i);
 595
 596                 if (offset == opt_get(*opts, sb))
 597                         continue;
 598
 599                 err = read_one_super(sb, offset);
 600                 if (!err)
 601                         goto got_super;
 602         }
 603
 604         ret = -EINVAL;
 605         goto err;
 606
 607 got_super:
 608         err = "Superblock block size smaller than device block size";
 609         ret = -EINVAL;
 610         if (le16_to_cpu(sb->sb->block_size) << 9 <
 611             bdev_logical_block_size(sb->bdev))
 612                 goto err;
 613
 614         if (sb->mode & FMODE_WRITE)
 615                 bdev_get_queue(sb->bdev)->backing_dev_info->capabilities
 616                         |= BDI_CAP_STABLE_WRITES;
 617         ret = 0;
 618         sb->have_layout = true;
 619 out:
 620         pr_verbose_init(*opts, "ret %i", ret);
 621         return ret;
 622 err:
 623         bch2_free_super(sb);
 624         pr_err("error reading superblock: %s", err);
 625         goto out;
 626 }
 627
 628 /* write superblock: */
 629
 630 static void write_super_endio(struct bio *bio)
 631 {
 632         struct bch_dev *ca = bio->bi_private;
 633
 634         /* XXX: return errors directly */
 635
 636         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write"))
 637                 ca->sb_write_error = 1;
 638
 639         closure_put(&ca->fs->sb_write);
 640         percpu_ref_put(&ca->io_ref);
 641 }
 642
 643 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 644 {
 645         struct bch_sb *sb = ca->disk_sb.sb;
 646         struct bio *bio = ca->disk_sb.bio;
 647
 648         bio_reset(bio);
 649         bio_set_dev(bio, ca->disk_sb.bdev);
 650         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 651         bio->bi_end_io          = write_super_endio;
 652         bio->bi_private         = ca;
 653         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 654         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 655
 656         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_SB],
 657                      bio_sectors(bio));
 658
 659         percpu_ref_get(&ca->io_ref);
 660         closure_bio_submit(bio, &c->sb_write);
 661 }
 662
 663 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 664 {
 665         struct bch_sb *sb = ca->disk_sb.sb;
 666         struct bio *bio = ca->disk_sb.bio;
 667
 668         sb->offset = sb->layout.sb_offset[idx];
 669
 670         SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
 671         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 672                                 null_nonce(), sb);
 673
 674         bio_reset(bio);
 675         bio_set_dev(bio, ca->disk_sb.bdev);
 676         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 677         bio->bi_end_io          = write_super_endio;
 678         bio->bi_private         = ca;
 679         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 680         bch2_bio_map(bio, sb,
 681                      roundup((size_t) vstruct_bytes(sb),
 682                              bdev_logical_block_size(ca->disk_sb.bdev)));
 683
 684         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_SB],
 685                      bio_sectors(bio));
 686
 687         percpu_ref_get(&ca->io_ref);
 688         closure_bio_submit(bio, &c->sb_write);
 689 }
 690
 691 int bch2_write_super(struct bch_fs *c)
 692 {
 693         struct closure *cl = &c->sb_write;
 694         struct bch_dev *ca;
 695         unsigned i, sb = 0, nr_wrote;
 696         const char *err;
 697         struct bch_devs_mask sb_written;
 698         bool wrote, can_mount_without_written, can_mount_with_written;
 699         int ret = 0;
 700
 701         lockdep_assert_held(&c->sb_lock);
 702
 703         closure_init_stack(cl);
 704         memset(&sb_written, 0, sizeof(sb_written));
 705
 706         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 707
 708         if (test_bit(BCH_FS_ERROR, &c->flags))
 709                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 710
 711         for_each_online_member(ca, c, i)
 712                 bch2_sb_from_fs(c, ca);
 713
 714         for_each_online_member(ca, c, i) {
 715                 err = bch2_sb_validate(&ca->disk_sb);
 716                 if (err) {
 717                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
 718                         ret = -1;
 719                         goto out;
 720                 }
 721         }
 722
 723         if (c->opts.nochanges)
 724                 goto out;
 725
 726         for_each_online_member(ca, c, i) {
 727                 __set_bit(ca->dev_idx, sb_written.d);
 728                 ca->sb_write_error = 0;
 729         }
 730
 731         for_each_online_member(ca, c, i)
 732                 read_back_super(c, ca);
 733         closure_sync(cl);
 734
 735         for_each_online_member(ca, c, i) {
 736                 if (!ca->sb_write_error &&
 737                     ca->disk_sb.seq !=
 738                     le64_to_cpu(ca->sb_read_scratch->seq)) {
 739                         bch2_fs_fatal_error(c,
 740                                 "Superblock modified by another process");
 741                         percpu_ref_put(&ca->io_ref);
 742                         ret = -EROFS;
 743                         goto out;
 744                 }
 745         }
 746
 747         do {
 748                 wrote = false;
 749                 for_each_online_member(ca, c, i)
 750                         if (!ca->sb_write_error &&
 751                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 752                                 write_one_super(c, ca, sb);
 753                                 wrote = true;
 754                         }
 755                 closure_sync(cl);
 756                 sb++;
 757         } while (wrote);
 758
 759         for_each_online_member(ca, c, i) {
 760                 if (ca->sb_write_error)
 761                         __clear_bit(ca->dev_idx, sb_written.d);
 762                 else
 763                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 764         }
 765
 766         nr_wrote = dev_mask_nr(&sb_written);
 767
 768         can_mount_with_written =
 769                 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
 770                                       BCH_FORCE_IF_DEGRADED);
 771
 772         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 773                 sb_written.d[i] = ~sb_written.d[i];
 774
 775         can_mount_without_written =
 776                 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
 777                                       BCH_FORCE_IF_DEGRADED);
 778
 779         /*
 780          * If we would be able to mount _without_ the devices we successfully
 781          * wrote superblocks to, we weren't able to write to enough devices:
 782          *
 783          * Exception: if we can mount without the successes because we haven't
 784          * written anything (new filesystem), we continue if we'd be able to
 785          * mount with the devices we did successfully write to:
 786          */
 787         if (bch2_fs_fatal_err_on(!nr_wrote ||
 788                                  (can_mount_without_written &&
 789                                   !can_mount_with_written), c,
 790                 "Unable to write superblock to sufficient devices"))
 791                 ret = -1;
 792 out:
 793         /* Make new options visible after they're persistent: */
 794         bch2_sb_update(c);
 795         return ret;
 796 }
 797
 798 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 799 {
 800         mutex_lock(&c->sb_lock);
 801         if (!(c->sb.features & (1ULL << feat))) {
 802                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 803
 804                 bch2_write_super(c);
 805         }
 806         mutex_unlock(&c->sb_lock);
 807 }
 808
 809 /* BCH_SB_FIELD_journal: */
 810
 811 static int u64_cmp(const void *_l, const void *_r)
 812 {
 813         u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
 814
 815         return l < r ? -1 : l > r ? 1 : 0;
 816 }
 817
 818 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
 819                                             struct bch_sb_field *f)
 820 {
 821         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 822         struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
 823         const char *err;
 824         unsigned nr;
 825         unsigned i;
 826         u64 *b;
 827
 828         journal = bch2_sb_get_journal(sb);
 829         if (!journal)
 830                 return NULL;
 831
 832         nr = bch2_nr_journal_buckets(journal);
 833         if (!nr)
 834                 return NULL;
 835
 836         b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
 837         if (!b)
 838                 return "cannot allocate memory";
 839
 840         for (i = 0; i < nr; i++)
 841                 b[i] = le64_to_cpu(journal->buckets[i]);
 842
 843         sort(b, nr, sizeof(u64), u64_cmp, NULL);
 844
 845         err = "journal bucket at sector 0";
 846         if (!b[0])
 847                 goto err;
 848
 849         err = "journal bucket before first bucket";
 850         if (m && b[0] < le16_to_cpu(m->first_bucket))
 851                 goto err;
 852
 853         err = "journal bucket past end of device";
 854         if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
 855                 goto err;
 856
 857         err = "duplicate journal buckets";
 858         for (i = 0; i + 1 < nr; i++)
 859                 if (b[i] == b[i + 1])
 860                         goto err;
 861
 862         err = NULL;
 863 err:
 864         kfree(b);
 865         return err;
 866 }
 867
 868 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
 869         .validate       = bch2_sb_validate_journal,
 870 };
 871
 872 /* BCH_SB_FIELD_members: */
 873
 874 static const char *bch2_sb_validate_members(struct bch_sb *sb,
 875                                             struct bch_sb_field *f)
 876 {
 877         struct bch_sb_field_members *mi = field_to_type(f, members);
 878         struct bch_member *m;
 879
 880         if ((void *) (mi->members + sb->nr_devices) >
 881             vstruct_end(&mi->field))
 882                 return "Invalid superblock: bad member info";
 883
 884         for (m = mi->members;
 885              m < mi->members + sb->nr_devices;
 886              m++) {
 887                 if (!bch2_member_exists(m))
 888                         continue;
 889
 890                 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
 891                         return "Too many buckets";
 892
 893                 if (le64_to_cpu(m->nbuckets) -
 894                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
 895                         return "Not enough buckets";
 896
 897                 if (le16_to_cpu(m->bucket_size) <
 898                     le16_to_cpu(sb->block_size))
 899                         return "bucket size smaller than block size";
 900
 901                 if (le16_to_cpu(m->bucket_size) <
 902                     BCH_SB_BTREE_NODE_SIZE(sb))
 903                         return "bucket size smaller than btree node size";
 904         }
 905
 906         return NULL;
 907 }
 908
 909 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
 910         .validate       = bch2_sb_validate_members,
 911 };
 912
 913 /* BCH_SB_FIELD_crypt: */
 914
 915 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
 916                                           struct bch_sb_field *f)
 917 {
 918         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
 919
 920         if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
 921                 return "invalid field crypt: wrong size";
 922
 923         if (BCH_CRYPT_KDF_TYPE(crypt))
 924                 return "invalid field crypt: bad kdf type";
 925
 926         return NULL;
 927 }
 928
 929 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
 930         .validate       = bch2_sb_validate_crypt,
 931 };
 932
 933 /* BCH_SB_FIELD_clean: */
 934
 935 void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
 936 {
 937         struct jset_entry *entry;
 938
 939         for (entry = clean->start;
 940              entry < (struct jset_entry *) vstruct_end(&clean->field);
 941              entry = vstruct_next(entry))
 942                 bch2_bkey_renumber(BKEY_TYPE_BTREE, bkey_to_packed(entry->start), write);
 943 }
 944
 945 int bch2_fs_mark_dirty(struct bch_fs *c)
 946 {
 947         int ret;
 948
 949         /*
 950          * Unconditionally write superblock, to verify it hasn't changed before
 951          * we go rw:
 952          */
 953
 954         mutex_lock(&c->sb_lock);
 955         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 956         c->disk_sb.sb->compat[0] &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA);
 957         ret = bch2_write_super(c);
 958         mutex_unlock(&c->sb_lock);
 959
 960         return ret;
 961 }
 962
 963 static void
 964 entry_init_u64s(struct jset_entry *entry, unsigned u64s)
 965 {
 966         memset(entry, 0, u64s * sizeof(u64));
 967
 968         /*
 969          * The u64s field counts from the start of data, ignoring the shared
 970          * fields.
 971          */
 972         entry->u64s = u64s - 1;
 973 }
 974
 975 static void
 976 entry_init_size(struct jset_entry *entry, size_t size)
 977 {
 978         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
 979         entry_init_u64s(entry, u64s);
 980 }
 981
 982 struct jset_entry *
 983 bch2_journal_super_entries_add_common(struct bch_fs *c,
 984                                       struct jset_entry *entry,
 985                                       u64 journal_seq)
 986 {
 987         struct btree_root *r;
 988         unsigned i;
 989
 990         mutex_lock(&c->btree_root_lock);
 991
 992         for (r = c->btree_roots;
 993              r < c->btree_roots + BTREE_ID_NR;
 994              r++)
 995                 if (r->alive) {
 996                         entry_init_u64s(entry, r->key.u64s + 1);
 997                         entry->btree_id = r - c->btree_roots;
 998                         entry->level    = r->level;
 999                         entry->type     = BCH_JSET_ENTRY_btree_root;
1000                         bkey_copy(&entry->start[0], &r->key);
1001
1002                         entry = vstruct_next(entry);
1003                 }
1004         c->btree_roots_dirty = false;
1005
1006         mutex_unlock(&c->btree_root_lock);
1007
1008         percpu_down_write(&c->mark_lock);
1009
1010         if (!journal_seq) {
1011                 bch2_fs_usage_acc_to_base(c, 0);
1012                 bch2_fs_usage_acc_to_base(c, 1);
1013         } else {
1014                 bch2_fs_usage_acc_to_base(c, journal_seq & 1);
1015         }
1016
1017         {
1018                 struct jset_entry_usage *u =
1019                         container_of(entry, struct jset_entry_usage, entry);
1020
1021                 entry_init_size(entry, sizeof(*u));
1022                 u->entry.type   = BCH_JSET_ENTRY_usage;
1023                 u->entry.btree_id = FS_USAGE_INODES;
1024                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1025
1026                 entry = vstruct_next(entry);
1027         }
1028
1029         {
1030                 struct jset_entry_usage *u =
1031                         container_of(entry, struct jset_entry_usage, entry);
1032
1033                 entry_init_size(entry, sizeof(*u));
1034                 u->entry.type   = BCH_JSET_ENTRY_usage;
1035                 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1036                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1037
1038                 entry = vstruct_next(entry);
1039         }
1040
1041         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1042                 struct jset_entry_usage *u =
1043                         container_of(entry, struct jset_entry_usage, entry);
1044
1045                 entry_init_size(entry, sizeof(*u));
1046                 u->entry.type   = BCH_JSET_ENTRY_usage;
1047                 u->entry.btree_id = FS_USAGE_RESERVED;
1048                 u->entry.level  = i;
1049                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1050
1051                 entry = vstruct_next(entry);
1052         }
1053
1054         for (i = 0; i < c->replicas.nr; i++) {
1055                 struct bch_replicas_entry *e =
1056                         cpu_replicas_entry(&c->replicas, i);
1057                 struct jset_entry_data_usage *u =
1058                         container_of(entry, struct jset_entry_data_usage, entry);
1059
1060                 entry_init_size(entry, sizeof(*u) + e->nr_devs);
1061                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1062                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1063                 memcpy(&u->r, e, replicas_entry_bytes(e));
1064
1065                 entry = vstruct_next(entry);
1066         }
1067
1068         percpu_up_write(&c->mark_lock);
1069
1070         return entry;
1071 }
1072
1073 void bch2_fs_mark_clean(struct bch_fs *c)
1074 {
1075         struct bch_sb_field_clean *sb_clean;
1076         struct jset_entry *entry;
1077         unsigned u64s;
1078
1079         mutex_lock(&c->sb_lock);
1080         if (BCH_SB_CLEAN(c->disk_sb.sb))
1081                 goto out;
1082
1083         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1084
1085         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1086         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA;
1087
1088         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1089
1090         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1091         if (!sb_clean) {
1092                 bch_err(c, "error resizing superblock while setting filesystem clean");
1093                 goto out;
1094         }
1095
1096         sb_clean->flags         = 0;
1097         sb_clean->read_clock    = cpu_to_le16(c->bucket_clock[READ].hand);
1098         sb_clean->write_clock   = cpu_to_le16(c->bucket_clock[WRITE].hand);
1099         sb_clean->journal_seq   = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1100
1101         /* Trying to catch outstanding bug: */
1102         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1103
1104         entry = sb_clean->start;
1105         entry = bch2_journal_super_entries_add_common(c, entry, 0);
1106         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1107
1108         memset(entry, 0,
1109                vstruct_end(&sb_clean->field) - (void *) entry);
1110
1111         if (le16_to_cpu(c->disk_sb.sb->version) <
1112             bcachefs_metadata_version_bkey_renumber)
1113                 bch2_sb_clean_renumber(sb_clean, WRITE);
1114
1115         bch2_write_super(c);
1116 out:
1117         mutex_unlock(&c->sb_lock);
1118 }
1119
1120 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1121                                           struct bch_sb_field *f)
1122 {
1123         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1124
1125         if (vstruct_bytes(&clean->field) < sizeof(*clean))
1126                 return "invalid field crypt: wrong size";
1127
1128         return NULL;
1129 }
1130
1131 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1132         .validate       = bch2_sb_validate_clean,
1133 };
1134
1135 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1136 #define x(f, nr)                                        \
1137         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1138         BCH_SB_FIELDS()
1139 #undef x
1140 };
1141
1142 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1143                                           struct bch_sb_field *f)
1144 {
1145         unsigned type = le32_to_cpu(f->type);
1146
1147         return type < BCH_SB_FIELD_NR
1148                 ? bch2_sb_field_ops[type]->validate(sb, f)
1149                 : NULL;
1150 }
1151
1152 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1153                            struct bch_sb_field *f)
1154 {
1155         unsigned type = le32_to_cpu(f->type);
1156         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1157                 ? bch2_sb_field_ops[type] : NULL;
1158
1159         if (ops)
1160                 pr_buf(out, "%s", bch2_sb_fields[type]);
1161         else
1162                 pr_buf(out, "(unknown field %u)", type);
1163
1164         pr_buf(out, " (size %llu):", vstruct_bytes(f));
1165
1166         if (ops && ops->to_text)
1167                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1168 }