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_seq_blacklist.h"
  13 #include "replicas.h"
  14 #include "quota.h"
  15 #include "super-io.h"
  16 #include "super.h"
  17 #include "vstructs.h"
  18
  19 #include <linux/backing-dev.h>
  20 #include <linux/sort.h>
  21
  22 const char * const bch2_sb_fields[] = {
  23 #define x(name, nr)     #name,
  24         BCH_SB_FIELDS()
  25 #undef x
  26         NULL
  27 };
  28
  29 static const char *bch2_sb_field_validate(struct bch_sb *,
  30                                           struct bch_sb_field *);
  31
  32 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  33                                       enum bch_sb_field_type type)
  34 {
  35         struct bch_sb_field *f;
  36
  37         /* XXX: need locking around superblock to access optional fields */
  38
  39         vstruct_for_each(sb, f)
  40                 if (le32_to_cpu(f->type) == type)
  41                         return f;
  42         return NULL;
  43 }
  44
  45 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  46                                                    struct bch_sb_field *f,
  47                                                    unsigned u64s)
  48 {
  49         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  50         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  51
  52         BUG_ON(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
  53                sb->page_order);
  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         free_pages((unsigned long) sb->sb, sb->page_order);
 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         unsigned order = get_order(new_bytes);
 111         struct bch_sb *new_sb;
 112         struct bio *bio;
 113
 114         if (sb->sb && sb->page_order >= order)
 115                 return 0;
 116
 117         if (sb->have_layout) {
 118                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 119
 120                 if (new_bytes > max_bytes) {
 121                         char buf[BDEVNAME_SIZE];
 122
 123                         pr_err("%s: superblock too big: want %zu but have %llu",
 124                                bdevname(sb->bdev, buf), new_bytes, max_bytes);
 125                         return -ENOSPC;
 126                 }
 127         }
 128
 129         if (sb->page_order >= order && sb->sb)
 130                 return 0;
 131
 132         if (dynamic_fault("bcachefs:add:super_realloc"))
 133                 return -ENOMEM;
 134
 135         if (sb->have_bio) {
 136                 bio = bio_kmalloc(GFP_KERNEL, 1 << order);
 137                 if (!bio)
 138                         return -ENOMEM;
 139
 140                 if (sb->bio)
 141                         bio_put(sb->bio);
 142                 sb->bio = bio;
 143         }
 144
 145         new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
 146         if (!new_sb)
 147                 return -ENOMEM;
 148
 149         if (sb->sb)
 150                 memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
 151
 152         free_pages((unsigned long) sb->sb, sb->page_order);
 153         sb->sb = new_sb;
 154
 155         sb->page_order = order;
 156
 157         return 0;
 158 }
 159
 160 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 161                                           enum bch_sb_field_type type,
 162                                           unsigned u64s)
 163 {
 164         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 165         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 166         ssize_t d = -old_u64s + u64s;
 167
 168         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 169                 return NULL;
 170
 171         if (sb->fs_sb) {
 172                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 173                 struct bch_dev *ca;
 174                 unsigned i;
 175
 176                 lockdep_assert_held(&c->sb_lock);
 177
 178                 /* XXX: we're not checking that offline device have enough space */
 179
 180                 for_each_online_member(ca, c, i) {
 181                         struct bch_sb_handle *sb = &ca->disk_sb;
 182
 183                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 184                                 percpu_ref_put(&ca->ref);
 185                                 return NULL;
 186                         }
 187                 }
 188         }
 189
 190         f = bch2_sb_field_get(sb->sb, type);
 191         f = __bch2_sb_field_resize(sb, f, u64s);
 192         if (f)
 193                 f->type = cpu_to_le32(type);
 194         return f;
 195 }
 196
 197 /* Superblock validate: */
 198
 199 static inline void __bch2_sb_layout_size_assert(void)
 200 {
 201         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 202 }
 203
 204 static const char *validate_sb_layout(struct bch_sb_layout *layout)
 205 {
 206         u64 offset, prev_offset, max_sectors;
 207         unsigned i;
 208
 209         if (uuid_le_cmp(layout->magic, BCACHE_MAGIC))
 210                 return "Not a bcachefs superblock layout";
 211
 212         if (layout->layout_type != 0)
 213                 return "Invalid superblock layout type";
 214
 215         if (!layout->nr_superblocks)
 216                 return "Invalid superblock layout: no superblocks";
 217
 218         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
 219                 return "Invalid superblock layout: too many superblocks";
 220
 221         max_sectors = 1 << layout->sb_max_size_bits;
 222
 223         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 224
 225         for (i = 1; i < layout->nr_superblocks; i++) {
 226                 offset = le64_to_cpu(layout->sb_offset[i]);
 227
 228                 if (offset < prev_offset + max_sectors)
 229                         return "Invalid superblock layout: superblocks overlap";
 230                 prev_offset = offset;
 231         }
 232
 233         return NULL;
 234 }
 235
 236 const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
 237 {
 238         struct bch_sb *sb = disk_sb->sb;
 239         struct bch_sb_field *f;
 240         struct bch_sb_field_members *mi;
 241         const char *err;
 242         u32 version, version_min;
 243         u16 block_size;
 244
 245         version         = le16_to_cpu(sb->version);
 246         version_min     = version >= bcachefs_metadata_version_new_versioning
 247                 ? le16_to_cpu(sb->version_min)
 248                 : version;
 249
 250         if (version    >= bcachefs_metadata_version_max ||
 251             version_min < bcachefs_metadata_version_min)
 252                 return "Unsupported superblock version";
 253
 254         if (version_min > version)
 255                 return "Bad minimum version";
 256
 257         if (sb->features[1] ||
 258             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
 259                 return "Filesystem has incompatible features";
 260
 261         block_size = le16_to_cpu(sb->block_size);
 262
 263         if (!is_power_of_2(block_size) ||
 264             block_size > PAGE_SECTORS)
 265                 return "Bad block size";
 266
 267         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
 268                 return "Bad user UUID";
 269
 270         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
 271                 return "Bad internal UUID";
 272
 273         if (!sb->nr_devices ||
 274             sb->nr_devices <= sb->dev_idx ||
 275             sb->nr_devices > BCH_SB_MEMBERS_MAX)
 276                 return "Bad number of member devices";
 277
 278         if (!BCH_SB_META_REPLICAS_WANT(sb) ||
 279             BCH_SB_META_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
 280                 return "Invalid number of metadata replicas";
 281
 282         if (!BCH_SB_META_REPLICAS_REQ(sb) ||
 283             BCH_SB_META_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
 284                 return "Invalid number of metadata replicas";
 285
 286         if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
 287             BCH_SB_DATA_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
 288                 return "Invalid number of data replicas";
 289
 290         if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
 291             BCH_SB_DATA_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
 292                 return "Invalid number of data replicas";
 293
 294         if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 295                 return "Invalid metadata checksum type";
 296
 297         if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 298                 return "Invalid metadata checksum type";
 299
 300         if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
 301                 return "Invalid compression type";
 302
 303         if (!BCH_SB_BTREE_NODE_SIZE(sb))
 304                 return "Btree node size not set";
 305
 306         if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
 307                 return "Btree node size not a power of two";
 308
 309         if (BCH_SB_GC_RESERVE(sb) < 5)
 310                 return "gc reserve percentage too small";
 311
 312         if (!sb->time_precision ||
 313             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
 314                 return "invalid time precision";
 315
 316         /* validate layout */
 317         err = validate_sb_layout(&sb->layout);
 318         if (err)
 319                 return err;
 320
 321         vstruct_for_each(sb, f) {
 322                 if (!f->u64s)
 323                         return "Invalid superblock: invalid optional field";
 324
 325                 if (vstruct_next(f) > vstruct_last(sb))
 326                         return "Invalid superblock: invalid optional field";
 327         }
 328
 329         /* members must be validated first: */
 330         mi = bch2_sb_get_members(sb);
 331         if (!mi)
 332                 return "Invalid superblock: member info area missing";
 333
 334         err = bch2_sb_field_validate(sb, &mi->field);
 335         if (err)
 336                 return err;
 337
 338         vstruct_for_each(sb, f) {
 339                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 340                         continue;
 341
 342                 err = bch2_sb_field_validate(sb, f);
 343                 if (err)
 344                         return err;
 345         }
 346
 347         return NULL;
 348 }
 349
 350 /* device open: */
 351
 352 static void bch2_sb_update(struct bch_fs *c)
 353 {
 354         struct bch_sb *src = c->disk_sb.sb;
 355         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 356         struct bch_dev *ca;
 357         unsigned i;
 358
 359         lockdep_assert_held(&c->sb_lock);
 360
 361         c->sb.uuid              = src->uuid;
 362         c->sb.user_uuid         = src->user_uuid;
 363         c->sb.version           = le16_to_cpu(src->version);
 364         c->sb.nr_devices        = src->nr_devices;
 365         c->sb.clean             = BCH_SB_CLEAN(src);
 366         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 367         c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
 368         c->sb.time_base_lo      = le64_to_cpu(src->time_base_lo);
 369         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 370         c->sb.time_precision    = le32_to_cpu(src->time_precision);
 371         c->sb.features          = le64_to_cpu(src->features[0]);
 372         c->sb.compat            = le64_to_cpu(src->compat[0]);
 373
 374         for_each_member_device(ca, c, i)
 375                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 376 }
 377
 378 /* doesn't copy member info */
 379 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 380 {
 381         struct bch_sb_field *src_f, *dst_f;
 382         struct bch_sb *dst = dst_handle->sb;
 383         unsigned i;
 384
 385         dst->version            = src->version;
 386         dst->version_min        = src->version_min;
 387         dst->seq                = src->seq;
 388         dst->uuid               = src->uuid;
 389         dst->user_uuid          = src->user_uuid;
 390         memcpy(dst->label,      src->label, sizeof(dst->label));
 391
 392         dst->block_size         = src->block_size;
 393         dst->nr_devices         = src->nr_devices;
 394
 395         dst->time_base_lo       = src->time_base_lo;
 396         dst->time_base_hi       = src->time_base_hi;
 397         dst->time_precision     = src->time_precision;
 398
 399         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 400         memcpy(dst->features,   src->features,  sizeof(dst->features));
 401         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 402
 403         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 404                 if (i == BCH_SB_FIELD_journal)
 405                         continue;
 406
 407                 src_f = bch2_sb_field_get(src, i);
 408                 dst_f = bch2_sb_field_get(dst, i);
 409                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 410                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 411
 412                 if (src_f)
 413                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 414         }
 415 }
 416
 417 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 418 {
 419         struct bch_sb_field_journal *journal_buckets =
 420                 bch2_sb_get_journal(src);
 421         unsigned journal_u64s = journal_buckets
 422                 ? le32_to_cpu(journal_buckets->field.u64s)
 423                 : 0;
 424         int ret;
 425
 426         lockdep_assert_held(&c->sb_lock);
 427
 428         ret = bch2_sb_realloc(&c->disk_sb,
 429                               le32_to_cpu(src->u64s) - journal_u64s);
 430         if (ret)
 431                 return ret;
 432
 433         __copy_super(&c->disk_sb, src);
 434
 435         ret = bch2_sb_replicas_to_cpu_replicas(c);
 436         if (ret)
 437                 return ret;
 438
 439         ret = bch2_sb_disk_groups_to_cpu(c);
 440         if (ret)
 441                 return ret;
 442
 443         bch2_sb_update(c);
 444         return 0;
 445 }
 446
 447 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 448 {
 449         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 450         struct bch_sb_field_journal *journal_buckets =
 451                 bch2_sb_get_journal(dst);
 452         unsigned journal_u64s = journal_buckets
 453                 ? le32_to_cpu(journal_buckets->field.u64s)
 454                 : 0;
 455         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 456         int ret;
 457
 458         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 459         if (ret)
 460                 return ret;
 461
 462         __copy_super(&ca->disk_sb, src);
 463         return 0;
 464 }
 465
 466 /* read superblock: */
 467
 468 static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
 469 {
 470         struct bch_csum csum;
 471         size_t bytes;
 472 reread:
 473         bio_reset(sb->bio);
 474         bio_set_dev(sb->bio, sb->bdev);
 475         sb->bio->bi_iter.bi_sector = offset;
 476         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 477         bch2_bio_map(sb->bio, sb->sb, PAGE_SIZE << sb->page_order);
 478
 479         if (submit_bio_wait(sb->bio))
 480                 return "IO error";
 481
 482         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
 483                 return "Not a bcachefs superblock";
 484
 485         if (le16_to_cpu(sb->sb->version) <  bcachefs_metadata_version_min ||
 486             le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
 487                 return "Unsupported superblock version";
 488
 489         bytes = vstruct_bytes(sb->sb);
 490
 491         if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
 492                 return "Bad superblock: too big";
 493
 494         if (get_order(bytes) > sb->page_order) {
 495                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 496                         return "cannot allocate memory";
 497                 goto reread;
 498         }
 499
 500         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
 501                 return "unknown csum type";
 502
 503         /* XXX: verify MACs */
 504         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 505                             null_nonce(), sb->sb);
 506
 507         if (bch2_crc_cmp(csum, sb->sb->csum))
 508                 return "bad checksum reading superblock";
 509
 510         sb->seq = le64_to_cpu(sb->sb->seq);
 511
 512         return NULL;
 513 }
 514
 515 int bch2_read_super(const char *path, struct bch_opts *opts,
 516                     struct bch_sb_handle *sb)
 517 {
 518         u64 offset = opt_get(*opts, sb);
 519         struct bch_sb_layout layout;
 520         const char *err;
 521         __le64 *i;
 522         int ret;
 523
 524         pr_verbose_init(*opts, "");
 525
 526         memset(sb, 0, sizeof(*sb));
 527         sb->mode        = FMODE_READ;
 528         sb->have_bio    = true;
 529
 530         if (!opt_get(*opts, noexcl))
 531                 sb->mode |= FMODE_EXCL;
 532
 533         if (!opt_get(*opts, nochanges))
 534                 sb->mode |= FMODE_WRITE;
 535
 536         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 537         if (IS_ERR(sb->bdev) &&
 538             PTR_ERR(sb->bdev) == -EACCES &&
 539             opt_get(*opts, read_only)) {
 540                 sb->mode &= ~FMODE_WRITE;
 541
 542                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 543                 if (!IS_ERR(sb->bdev))
 544                         opt_set(*opts, nochanges, true);
 545         }
 546
 547         if (IS_ERR(sb->bdev)) {
 548                 ret = PTR_ERR(sb->bdev);
 549                 goto out;
 550         }
 551
 552         err = "cannot allocate memory";
 553         ret = bch2_sb_realloc(sb, 0);
 554         if (ret)
 555                 goto err;
 556
 557         ret = -EFAULT;
 558         err = "dynamic fault";
 559         if (bch2_fs_init_fault("read_super"))
 560                 goto err;
 561
 562         ret = -EINVAL;
 563         err = read_one_super(sb, offset);
 564         if (!err)
 565                 goto got_super;
 566
 567         if (opt_defined(*opts, sb))
 568                 goto err;
 569
 570         pr_err("error reading default superblock: %s", err);
 571
 572         /*
 573          * Error reading primary superblock - read location of backup
 574          * superblocks:
 575          */
 576         bio_reset(sb->bio);
 577         bio_set_dev(sb->bio, sb->bdev);
 578         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 579         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 580         /*
 581          * use sb buffer to read layout, since sb buffer is page aligned but
 582          * layout won't be:
 583          */
 584         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 585
 586         err = "IO error";
 587         if (submit_bio_wait(sb->bio))
 588                 goto err;
 589
 590         memcpy(&layout, sb->sb, sizeof(layout));
 591         err = validate_sb_layout(&layout);
 592         if (err)
 593                 goto err;
 594
 595         for (i = layout.sb_offset;
 596              i < layout.sb_offset + layout.nr_superblocks; i++) {
 597                 offset = le64_to_cpu(*i);
 598
 599                 if (offset == opt_get(*opts, sb))
 600                         continue;
 601
 602                 err = read_one_super(sb, offset);
 603                 if (!err)
 604                         goto got_super;
 605         }
 606
 607         ret = -EINVAL;
 608         goto err;
 609
 610 got_super:
 611         err = "Superblock block size smaller than device block size";
 612         ret = -EINVAL;
 613         if (le16_to_cpu(sb->sb->block_size) << 9 <
 614             bdev_logical_block_size(sb->bdev))
 615                 goto err;
 616
 617         if (sb->mode & FMODE_WRITE)
 618                 bdev_get_queue(sb->bdev)->backing_dev_info->capabilities
 619                         |= BDI_CAP_STABLE_WRITES;
 620         ret = 0;
 621         sb->have_layout = true;
 622 out:
 623         pr_verbose_init(*opts, "ret %i", ret);
 624         return ret;
 625 err:
 626         bch2_free_super(sb);
 627         pr_err("error reading superblock: %s", err);
 628         goto out;
 629 }
 630
 631 /* write superblock: */
 632
 633 static void write_super_endio(struct bio *bio)
 634 {
 635         struct bch_dev *ca = bio->bi_private;
 636
 637         /* XXX: return errors directly */
 638
 639         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 640                                bch2_blk_status_to_str(bio->bi_status)))
 641                 ca->sb_write_error = 1;
 642
 643         closure_put(&ca->fs->sb_write);
 644         percpu_ref_put(&ca->io_ref);
 645 }
 646
 647 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 648 {
 649         struct bch_sb *sb = ca->disk_sb.sb;
 650         struct bio *bio = ca->disk_sb.bio;
 651
 652         bio_reset(bio);
 653         bio_set_dev(bio, ca->disk_sb.bdev);
 654         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 655         bio->bi_end_io          = write_super_endio;
 656         bio->bi_private         = ca;
 657         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 658         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 659
 660         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 661                      bio_sectors(bio));
 662
 663         percpu_ref_get(&ca->io_ref);
 664         closure_bio_submit(bio, &c->sb_write);
 665 }
 666
 667 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 668 {
 669         struct bch_sb *sb = ca->disk_sb.sb;
 670         struct bio *bio = ca->disk_sb.bio;
 671
 672         sb->offset = sb->layout.sb_offset[idx];
 673
 674         SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
 675         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 676                                 null_nonce(), sb);
 677
 678         bio_reset(bio);
 679         bio_set_dev(bio, ca->disk_sb.bdev);
 680         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 681         bio->bi_end_io          = write_super_endio;
 682         bio->bi_private         = ca;
 683         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 684         bch2_bio_map(bio, sb,
 685                      roundup((size_t) vstruct_bytes(sb),
 686                              bdev_logical_block_size(ca->disk_sb.bdev)));
 687
 688         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 689                      bio_sectors(bio));
 690
 691         percpu_ref_get(&ca->io_ref);
 692         closure_bio_submit(bio, &c->sb_write);
 693 }
 694
 695 int bch2_write_super(struct bch_fs *c)
 696 {
 697         struct closure *cl = &c->sb_write;
 698         struct bch_dev *ca;
 699         unsigned i, sb = 0, nr_wrote;
 700         const char *err;
 701         struct bch_devs_mask sb_written;
 702         bool wrote, can_mount_without_written, can_mount_with_written;
 703         int ret = 0;
 704
 705         lockdep_assert_held(&c->sb_lock);
 706
 707         closure_init_stack(cl);
 708         memset(&sb_written, 0, sizeof(sb_written));
 709
 710         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 711
 712         if (test_bit(BCH_FS_ERROR, &c->flags))
 713                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 714
 715         for_each_online_member(ca, c, i)
 716                 bch2_sb_from_fs(c, ca);
 717
 718         for_each_online_member(ca, c, i) {
 719                 err = bch2_sb_validate(&ca->disk_sb);
 720                 if (err) {
 721                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
 722                         ret = -1;
 723                         goto out;
 724                 }
 725         }
 726
 727         if (c->opts.nochanges)
 728                 goto out;
 729
 730         for_each_online_member(ca, c, i) {
 731                 __set_bit(ca->dev_idx, sb_written.d);
 732                 ca->sb_write_error = 0;
 733         }
 734
 735         for_each_online_member(ca, c, i)
 736                 read_back_super(c, ca);
 737         closure_sync(cl);
 738
 739         for_each_online_member(ca, c, i) {
 740                 if (!ca->sb_write_error &&
 741                     ca->disk_sb.seq !=
 742                     le64_to_cpu(ca->sb_read_scratch->seq)) {
 743                         bch2_fs_fatal_error(c,
 744                                 "Superblock modified by another process");
 745                         percpu_ref_put(&ca->io_ref);
 746                         ret = -EROFS;
 747                         goto out;
 748                 }
 749         }
 750
 751         do {
 752                 wrote = false;
 753                 for_each_online_member(ca, c, i)
 754                         if (!ca->sb_write_error &&
 755                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 756                                 write_one_super(c, ca, sb);
 757                                 wrote = true;
 758                         }
 759                 closure_sync(cl);
 760                 sb++;
 761         } while (wrote);
 762
 763         for_each_online_member(ca, c, i) {
 764                 if (ca->sb_write_error)
 765                         __clear_bit(ca->dev_idx, sb_written.d);
 766                 else
 767                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 768         }
 769
 770         nr_wrote = dev_mask_nr(&sb_written);
 771
 772         can_mount_with_written =
 773                 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
 774                                       BCH_FORCE_IF_DEGRADED);
 775
 776         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 777                 sb_written.d[i] = ~sb_written.d[i];
 778
 779         can_mount_without_written =
 780                 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
 781                                       BCH_FORCE_IF_DEGRADED);
 782
 783         /*
 784          * If we would be able to mount _without_ the devices we successfully
 785          * wrote superblocks to, we weren't able to write to enough devices:
 786          *
 787          * Exception: if we can mount without the successes because we haven't
 788          * written anything (new filesystem), we continue if we'd be able to
 789          * mount with the devices we did successfully write to:
 790          */
 791         if (bch2_fs_fatal_err_on(!nr_wrote ||
 792                                  (can_mount_without_written &&
 793                                   !can_mount_with_written), c,
 794                 "Unable to write superblock to sufficient devices"))
 795                 ret = -1;
 796 out:
 797         /* Make new options visible after they're persistent: */
 798         bch2_sb_update(c);
 799         return ret;
 800 }
 801
 802 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 803 {
 804         mutex_lock(&c->sb_lock);
 805         if (!(c->sb.features & (1ULL << feat))) {
 806                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 807
 808                 bch2_write_super(c);
 809         }
 810         mutex_unlock(&c->sb_lock);
 811 }
 812
 813 /* BCH_SB_FIELD_journal: */
 814
 815 static int u64_cmp(const void *_l, const void *_r)
 816 {
 817         u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
 818
 819         return l < r ? -1 : l > r ? 1 : 0;
 820 }
 821
 822 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
 823                                             struct bch_sb_field *f)
 824 {
 825         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 826         struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
 827         const char *err;
 828         unsigned nr;
 829         unsigned i;
 830         u64 *b;
 831
 832         journal = bch2_sb_get_journal(sb);
 833         if (!journal)
 834                 return NULL;
 835
 836         nr = bch2_nr_journal_buckets(journal);
 837         if (!nr)
 838                 return NULL;
 839
 840         b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
 841         if (!b)
 842                 return "cannot allocate memory";
 843
 844         for (i = 0; i < nr; i++)
 845                 b[i] = le64_to_cpu(journal->buckets[i]);
 846
 847         sort(b, nr, sizeof(u64), u64_cmp, NULL);
 848
 849         err = "journal bucket at sector 0";
 850         if (!b[0])
 851                 goto err;
 852
 853         err = "journal bucket before first bucket";
 854         if (m && b[0] < le16_to_cpu(m->first_bucket))
 855                 goto err;
 856
 857         err = "journal bucket past end of device";
 858         if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
 859                 goto err;
 860
 861         err = "duplicate journal buckets";
 862         for (i = 0; i + 1 < nr; i++)
 863                 if (b[i] == b[i + 1])
 864                         goto err;
 865
 866         err = NULL;
 867 err:
 868         kfree(b);
 869         return err;
 870 }
 871
 872 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
 873         .validate       = bch2_sb_validate_journal,
 874 };
 875
 876 /* BCH_SB_FIELD_members: */
 877
 878 static const char *bch2_sb_validate_members(struct bch_sb *sb,
 879                                             struct bch_sb_field *f)
 880 {
 881         struct bch_sb_field_members *mi = field_to_type(f, members);
 882         struct bch_member *m;
 883
 884         if ((void *) (mi->members + sb->nr_devices) >
 885             vstruct_end(&mi->field))
 886                 return "Invalid superblock: bad member info";
 887
 888         for (m = mi->members;
 889              m < mi->members + sb->nr_devices;
 890              m++) {
 891                 if (!bch2_member_exists(m))
 892                         continue;
 893
 894                 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
 895                         return "Too many buckets";
 896
 897                 if (le64_to_cpu(m->nbuckets) -
 898                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
 899                         return "Not enough buckets";
 900
 901                 if (le16_to_cpu(m->bucket_size) <
 902                     le16_to_cpu(sb->block_size))
 903                         return "bucket size smaller than block size";
 904
 905                 if (le16_to_cpu(m->bucket_size) <
 906                     BCH_SB_BTREE_NODE_SIZE(sb))
 907                         return "bucket size smaller than btree node size";
 908         }
 909
 910         return NULL;
 911 }
 912
 913 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
 914         .validate       = bch2_sb_validate_members,
 915 };
 916
 917 /* BCH_SB_FIELD_crypt: */
 918
 919 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
 920                                           struct bch_sb_field *f)
 921 {
 922         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
 923
 924         if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
 925                 return "invalid field crypt: wrong size";
 926
 927         if (BCH_CRYPT_KDF_TYPE(crypt))
 928                 return "invalid field crypt: bad kdf type";
 929
 930         return NULL;
 931 }
 932
 933 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
 934         .validate       = bch2_sb_validate_crypt,
 935 };
 936
 937 /* BCH_SB_FIELD_clean: */
 938
 939 void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
 940 {
 941         struct jset_entry *entry;
 942
 943         for (entry = clean->start;
 944              entry < (struct jset_entry *) vstruct_end(&clean->field);
 945              entry = vstruct_next(entry))
 946                 bch2_bkey_renumber(BKEY_TYPE_BTREE, bkey_to_packed(entry->start), write);
 947 }
 948
 949 int bch2_fs_mark_dirty(struct bch_fs *c)
 950 {
 951         int ret;
 952
 953         /*
 954          * Unconditionally write superblock, to verify it hasn't changed before
 955          * we go rw:
 956          */
 957
 958         mutex_lock(&c->sb_lock);
 959         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 960         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_new_extent_overwrite;
 961         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_extents_above_btree_updates;
 962         c->disk_sb.sb->features[0] |= 1ULL << BCH_FEATURE_btree_updates_journalled;
 963         ret = bch2_write_super(c);
 964         mutex_unlock(&c->sb_lock);
 965
 966         return ret;
 967 }
 968
 969 static void
 970 entry_init_u64s(struct jset_entry *entry, unsigned u64s)
 971 {
 972         memset(entry, 0, u64s * sizeof(u64));
 973
 974         /*
 975          * The u64s field counts from the start of data, ignoring the shared
 976          * fields.
 977          */
 978         entry->u64s = u64s - 1;
 979 }
 980
 981 static void
 982 entry_init_size(struct jset_entry *entry, size_t size)
 983 {
 984         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
 985         entry_init_u64s(entry, u64s);
 986 }
 987
 988 struct jset_entry *
 989 bch2_journal_super_entries_add_common(struct bch_fs *c,
 990                                       struct jset_entry *entry,
 991                                       u64 journal_seq)
 992 {
 993         unsigned i;
 994
 995         percpu_down_write(&c->mark_lock);
 996
 997         if (!journal_seq) {
 998                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
 999                         bch2_fs_usage_acc_to_base(c, i);
1000         } else {
1001                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1002         }
1003
1004         {
1005                 struct jset_entry_usage *u =
1006                         container_of(entry, struct jset_entry_usage, entry);
1007
1008                 entry_init_size(entry, sizeof(*u));
1009                 u->entry.type   = BCH_JSET_ENTRY_usage;
1010                 u->entry.btree_id = FS_USAGE_INODES;
1011                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1012
1013                 entry = vstruct_next(entry);
1014         }
1015
1016         {
1017                 struct jset_entry_usage *u =
1018                         container_of(entry, struct jset_entry_usage, entry);
1019
1020                 entry_init_size(entry, sizeof(*u));
1021                 u->entry.type   = BCH_JSET_ENTRY_usage;
1022                 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1023                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1024
1025                 entry = vstruct_next(entry);
1026         }
1027
1028         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1029                 struct jset_entry_usage *u =
1030                         container_of(entry, struct jset_entry_usage, entry);
1031
1032                 entry_init_size(entry, sizeof(*u));
1033                 u->entry.type   = BCH_JSET_ENTRY_usage;
1034                 u->entry.btree_id = FS_USAGE_RESERVED;
1035                 u->entry.level  = i;
1036                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1037
1038                 entry = vstruct_next(entry);
1039         }
1040
1041         for (i = 0; i < c->replicas.nr; i++) {
1042                 struct bch_replicas_entry *e =
1043                         cpu_replicas_entry(&c->replicas, i);
1044                 struct jset_entry_data_usage *u =
1045                         container_of(entry, struct jset_entry_data_usage, entry);
1046
1047                 entry_init_size(entry, sizeof(*u) + e->nr_devs);
1048                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1049                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1050                 memcpy(&u->r, e, replicas_entry_bytes(e));
1051
1052                 entry = vstruct_next(entry);
1053         }
1054
1055         percpu_up_write(&c->mark_lock);
1056
1057         return entry;
1058 }
1059
1060 void bch2_fs_mark_clean(struct bch_fs *c)
1061 {
1062         struct bch_sb_field_clean *sb_clean;
1063         struct jset_entry *entry;
1064         unsigned u64s;
1065
1066         mutex_lock(&c->sb_lock);
1067         if (BCH_SB_CLEAN(c->disk_sb.sb))
1068                 goto out;
1069
1070         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1071
1072         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1073         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA;
1074         c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_extents_above_btree_updates);
1075         c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_btree_updates_journalled);
1076
1077         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1078
1079         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1080         if (!sb_clean) {
1081                 bch_err(c, "error resizing superblock while setting filesystem clean");
1082                 goto out;
1083         }
1084
1085         sb_clean->flags         = 0;
1086         sb_clean->read_clock    = cpu_to_le16(c->bucket_clock[READ].hand);
1087         sb_clean->write_clock   = cpu_to_le16(c->bucket_clock[WRITE].hand);
1088         sb_clean->journal_seq   = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1089
1090         /* Trying to catch outstanding bug: */
1091         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1092
1093         entry = sb_clean->start;
1094         entry = bch2_journal_super_entries_add_common(c, entry, 0);
1095         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1096         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1097
1098         memset(entry, 0,
1099                vstruct_end(&sb_clean->field) - (void *) entry);
1100
1101         if (le16_to_cpu(c->disk_sb.sb->version) <
1102             bcachefs_metadata_version_bkey_renumber)
1103                 bch2_sb_clean_renumber(sb_clean, WRITE);
1104
1105         bch2_write_super(c);
1106 out:
1107         mutex_unlock(&c->sb_lock);
1108 }
1109
1110 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1111                                           struct bch_sb_field *f)
1112 {
1113         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1114
1115         if (vstruct_bytes(&clean->field) < sizeof(*clean))
1116                 return "invalid field crypt: wrong size";
1117
1118         return NULL;
1119 }
1120
1121 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1122         .validate       = bch2_sb_validate_clean,
1123 };
1124
1125 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1126 #define x(f, nr)                                        \
1127         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1128         BCH_SB_FIELDS()
1129 #undef x
1130 };
1131
1132 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1133                                           struct bch_sb_field *f)
1134 {
1135         unsigned type = le32_to_cpu(f->type);
1136
1137         return type < BCH_SB_FIELD_NR
1138                 ? bch2_sb_field_ops[type]->validate(sb, f)
1139                 : NULL;
1140 }
1141
1142 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1143                            struct bch_sb_field *f)
1144 {
1145         unsigned type = le32_to_cpu(f->type);
1146         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1147                 ? bch2_sb_field_ops[type] : NULL;
1148
1149         if (ops)
1150                 pr_buf(out, "%s", bch2_sb_fields[type]);
1151         else
1152                 pr_buf(out, "(unknown field %u)", type);
1153
1154         pr_buf(out, " (size %llu):", vstruct_bytes(f));
1155
1156         if (ops && ops->to_text)
1157                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1158 }