fs/bcachefs/super-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "checksum.h"
   5 #include "disk_groups.h"
   6 #include "ec.h"
   7 #include "error.h"
   8 #include "io.h"
   9 #include "journal.h"
  10 #include "replicas.h"
  11 #include "quota.h"
  12 #include "super-io.h"
  13 #include "super.h"
  14 #include "vstructs.h"
  15
  16 #include <linux/backing-dev.h>
  17 #include <linux/sort.h>
  18
  19 static const struct blk_holder_ops bch2_sb_handle_bdev_ops = {
  20 };
  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) {
  56                 f = vstruct_last(sb->sb);
  57                 memset(f, 0, sizeof(u64) * u64s);
  58                 f->u64s = cpu_to_le32(u64s);
  59                 f->type = 0;
  60         } else {
  61                 void *src, *dst;
  62
  63                 src = vstruct_end(f);
  64
  65                 if (u64s) {
  66                         f->u64s = cpu_to_le32(u64s);
  67                         dst = vstruct_end(f);
  68                 } else {
  69                         dst = f;
  70                 }
  71
  72                 memmove(dst, src, vstruct_end(sb->sb) - src);
  73
  74                 if (dst > src)
  75                         memset(src, 0, dst - src);
  76         }
  77
  78         sb->sb->u64s = cpu_to_le32(sb_u64s);
  79
  80         return u64s ? f : NULL;
  81 }
  82
  83 void bch2_sb_field_delete(struct bch_sb_handle *sb,
  84                           enum bch_sb_field_type type)
  85 {
  86         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
  87
  88         if (f)
  89                 __bch2_sb_field_resize(sb, f, 0);
  90 }
  91
  92 /* Superblock realloc/free: */
  93
  94 void bch2_free_super(struct bch_sb_handle *sb)
  95 {
  96         if (sb->bio)
  97                 kfree(sb->bio);
  98         if (!IS_ERR_OR_NULL(sb->bdev))
  99                 blkdev_put(sb->bdev, sb->holder);
 100         kfree(sb->holder);
 101
 102         free_pages((unsigned long) sb->sb, sb->page_order);
 103         memset(sb, 0, sizeof(*sb));
 104 }
 105
 106 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 107 {
 108         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 109         unsigned order = get_order(new_bytes);
 110         struct bch_sb *new_sb;
 111         struct bio *bio;
 112
 113         if (sb->sb && sb->page_order >= order)
 114                 return 0;
 115
 116         if (sb->have_layout) {
 117                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 118
 119                 if (new_bytes > max_bytes) {
 120                         pr_err("%pg: superblock too big: want %zu but have %llu",
 121                                sb->bdev, 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                 unsigned nr_bvecs = 1 << order;
 134
 135                 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 136                 if (!bio)
 137                         return -ENOMEM;
 138
 139                 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
 140
 141                 if (sb->bio)
 142                         kfree(sb->bio);
 143                 sb->bio = bio;
 144         }
 145
 146         new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
 147         if (!new_sb)
 148                 return -ENOMEM;
 149
 150         if (sb->sb)
 151                 memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
 152
 153         free_pages((unsigned long) sb->sb, sb->page_order);
 154         sb->sb = new_sb;
 155
 156         sb->page_order = order;
 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_equal(&layout->magic, &BCACHE_MAGIC) &&
 211             !uuid_equal(&layout->magic, &BCHFS_MAGIC))
 212                 return "Not a bcachefs superblock layout";
 213
 214         if (layout->layout_type != 0)
 215                 return "Invalid superblock layout type";
 216
 217         if (!layout->nr_superblocks)
 218                 return "Invalid superblock layout: no superblocks";
 219
 220         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
 221                 return "Invalid superblock layout: too many superblocks";
 222
 223         max_sectors = 1 << layout->sb_max_size_bits;
 224
 225         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 226
 227         for (i = 1; i < layout->nr_superblocks; i++) {
 228                 offset = le64_to_cpu(layout->sb_offset[i]);
 229
 230                 if (offset < prev_offset + max_sectors)
 231                         return "Invalid superblock layout: superblocks overlap";
 232                 prev_offset = offset;
 233         }
 234
 235         return NULL;
 236 }
 237
 238 const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
 239 {
 240         struct bch_sb *sb = disk_sb->sb;
 241         struct bch_sb_field *f;
 242         struct bch_sb_field_members *mi;
 243         const char *err;
 244         u32 version, version_min;
 245         u16 block_size;
 246
 247         version         = le16_to_cpu(sb->version);
 248         version_min     = version >= bcachefs_metadata_version_new_versioning
 249                 ? le16_to_cpu(sb->version_min)
 250                 : version;
 251
 252         if (version    >= bcachefs_metadata_version_max ||
 253             version_min < bcachefs_metadata_version_min)
 254                 return "Unsupported superblock version";
 255
 256         if (version_min > version)
 257                 return "Bad minimum version";
 258
 259         if (sb->features[1] ||
 260             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
 261                 return "Filesystem has incompatible features";
 262
 263         block_size = le16_to_cpu(sb->block_size);
 264
 265         if (!is_power_of_2(block_size) ||
 266             block_size > PAGE_SECTORS)
 267                 return "Bad block size";
 268
 269         if (bch2_is_zero(sb->user_uuid.b, sizeof(sb->user_uuid)))
 270                 return "Bad user UUID";
 271
 272         if (bch2_is_zero(sb->uuid.b, sizeof(sb->uuid)))
 273                 return "Bad internal UUID";
 274
 275         if (!sb->nr_devices ||
 276             sb->nr_devices <= sb->dev_idx ||
 277             sb->nr_devices > BCH_SB_MEMBERS_MAX)
 278                 return "Bad number of member devices";
 279
 280         if (!BCH_SB_META_REPLICAS_WANT(sb) ||
 281             BCH_SB_META_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
 282                 return "Invalid number of metadata replicas";
 283
 284         if (!BCH_SB_META_REPLICAS_REQ(sb) ||
 285             BCH_SB_META_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
 286                 return "Invalid number of metadata replicas";
 287
 288         if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
 289             BCH_SB_DATA_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
 290                 return "Invalid number of data replicas";
 291
 292         if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
 293             BCH_SB_DATA_REPLICAS_REQ(sb) >= BCH_REPLICAS_MAX)
 294                 return "Invalid number of data replicas";
 295
 296         if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 297                 return "Invalid metadata checksum type";
 298
 299         if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 300                 return "Invalid metadata checksum type";
 301
 302         if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
 303                 return "Invalid compression type";
 304
 305         if (!BCH_SB_BTREE_NODE_SIZE(sb))
 306                 return "Btree node size not set";
 307
 308         if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
 309                 return "Btree node size not a power of two";
 310
 311         if (BCH_SB_GC_RESERVE(sb) < 5)
 312                 return "gc reserve percentage too small";
 313
 314         if (!sb->time_precision ||
 315             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
 316                 return "invalid time precision";
 317
 318         /* validate layout */
 319         err = validate_sb_layout(&sb->layout);
 320         if (err)
 321                 return err;
 322
 323         vstruct_for_each(sb, f) {
 324                 if (!f->u64s)
 325                         return "Invalid superblock: invalid optional field";
 326
 327                 if (vstruct_next(f) > vstruct_last(sb))
 328                         return "Invalid superblock: invalid optional field";
 329         }
 330
 331         /* members must be validated first: */
 332         mi = bch2_sb_get_members(sb);
 333         if (!mi)
 334                 return "Invalid superblock: member info area missing";
 335
 336         err = bch2_sb_field_validate(sb, &mi->field);
 337         if (err)
 338                 return err;
 339
 340         vstruct_for_each(sb, f) {
 341                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 342                         continue;
 343
 344                 err = bch2_sb_field_validate(sb, f);
 345                 if (err)
 346                         return err;
 347         }
 348
 349         return NULL;
 350 }
 351
 352 /* device open: */
 353
 354 static void bch2_sb_update(struct bch_fs *c)
 355 {
 356         struct bch_sb *src = c->disk_sb.sb;
 357         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 358         struct bch_dev *ca;
 359         unsigned i;
 360
 361         lockdep_assert_held(&c->sb_lock);
 362
 363         c->sb.uuid              = src->uuid;
 364         c->sb.user_uuid         = src->user_uuid;
 365         c->sb.version           = le16_to_cpu(src->version);
 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         c->sb.time_base_lo      = le64_to_cpu(src->time_base_lo);
 371         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 372         c->sb.time_precision    = le32_to_cpu(src->time_precision);
 373         c->sb.features          = le64_to_cpu(src->features[0]);
 374         c->sb.compat            = le64_to_cpu(src->compat[0]);
 375
 376         for_each_member_device(ca, c, i)
 377                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 378 }
 379
 380 /* doesn't copy member info */
 381 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 382 {
 383         struct bch_sb_field *src_f, *dst_f;
 384         struct bch_sb *dst = dst_handle->sb;
 385         unsigned i;
 386
 387         dst->version            = src->version;
 388         dst->version_min        = src->version_min;
 389         dst->seq                = src->seq;
 390         dst->uuid               = src->uuid;
 391         dst->user_uuid          = src->user_uuid;
 392         memcpy(dst->label,      src->label, sizeof(dst->label));
 393
 394         dst->block_size         = src->block_size;
 395         dst->nr_devices         = src->nr_devices;
 396
 397         dst->time_base_lo       = src->time_base_lo;
 398         dst->time_base_hi       = src->time_base_hi;
 399         dst->time_precision     = src->time_precision;
 400
 401         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 402         memcpy(dst->features,   src->features,  sizeof(dst->features));
 403         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 404
 405         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 406                 if (i == BCH_SB_FIELD_journal)
 407                         continue;
 408
 409                 src_f = bch2_sb_field_get(src, i);
 410                 dst_f = bch2_sb_field_get(dst, i);
 411                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 412                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 413
 414                 if (src_f)
 415                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 416         }
 417 }
 418
 419 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 420 {
 421         struct bch_sb_field_journal *journal_buckets =
 422                 bch2_sb_get_journal(src);
 423         unsigned journal_u64s = journal_buckets
 424                 ? le32_to_cpu(journal_buckets->field.u64s)
 425                 : 0;
 426         int ret;
 427
 428         lockdep_assert_held(&c->sb_lock);
 429
 430         ret = bch2_sb_realloc(&c->disk_sb,
 431                               le32_to_cpu(src->u64s) - journal_u64s);
 432         if (ret)
 433                 return ret;
 434
 435         __copy_super(&c->disk_sb, src);
 436
 437         ret = bch2_sb_replicas_to_cpu_replicas(c);
 438         if (ret)
 439                 return ret;
 440
 441         ret = bch2_sb_disk_groups_to_cpu(c);
 442         if (ret)
 443                 return ret;
 444
 445         bch2_sb_update(c);
 446         return 0;
 447 }
 448
 449 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 450 {
 451         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 452         struct bch_sb_field_journal *journal_buckets =
 453                 bch2_sb_get_journal(dst);
 454         unsigned journal_u64s = journal_buckets
 455                 ? le32_to_cpu(journal_buckets->field.u64s)
 456                 : 0;
 457         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 458         int ret;
 459
 460         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 461         if (ret)
 462                 return ret;
 463
 464         __copy_super(&ca->disk_sb, src);
 465         return 0;
 466 }
 467
 468 /* read superblock: */
 469
 470 static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
 471 {
 472         struct bch_csum csum;
 473         size_t bytes;
 474 reread:
 475         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 476         sb->bio->bi_iter.bi_sector = offset;
 477         sb->bio->bi_iter.bi_size = PAGE_SIZE << sb->page_order;
 478         bch2_bio_map(sb->bio, sb->sb);
 479
 480         if (submit_bio_wait(sb->bio))
 481                 return "IO error";
 482
 483         if (!uuid_equal(&sb->sb->magic, &BCACHE_MAGIC) &&
 484             !uuid_equal(&sb->sb->magic, &BCHFS_MAGIC))
 485                 return "Not a bcachefs superblock";
 486
 487         if (le16_to_cpu(sb->sb->version) <  bcachefs_metadata_version_min ||
 488             le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
 489                 return "Unsupported superblock version";
 490
 491         bytes = vstruct_bytes(sb->sb);
 492
 493         if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
 494                 return "Bad superblock: too big";
 495
 496         if (get_order(bytes) > sb->page_order) {
 497                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 498                         return "cannot allocate memory";
 499                 goto reread;
 500         }
 501
 502         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
 503                 return "unknown csum type";
 504
 505         /* XXX: verify MACs */
 506         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 507                             null_nonce(), sb->sb);
 508
 509         if (bch2_crc_cmp(csum, sb->sb->csum))
 510                 return "bad checksum reading superblock";
 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        = BLK_OPEN_READ;
 528         sb->have_bio    = true;
 529         sb->holder      = kmalloc(1, GFP_KERNEL);
 530         if (!sb->holder)
 531                 return -ENOMEM;
 532
 533         if (!opt_get(*opts, noexcl))
 534                 sb->mode |= BLK_OPEN_EXCL;
 535
 536         if (!opt_get(*opts, nochanges))
 537                 sb->mode |= BLK_OPEN_WRITE;
 538
 539         sb->bdev = blkdev_get_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
 540         if (IS_ERR(sb->bdev) &&
 541             PTR_ERR(sb->bdev) == -EACCES &&
 542             opt_get(*opts, read_only)) {
 543                 sb->mode &= ~BLK_OPEN_WRITE;
 544
 545                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb->holder, &bch2_sb_handle_bdev_ops);
 546                 if (!IS_ERR(sb->bdev))
 547                         opt_set(*opts, nochanges, true);
 548         }
 549
 550         if (IS_ERR(sb->bdev)) {
 551                 ret = PTR_ERR(sb->bdev);
 552                 goto out;
 553         }
 554
 555         err = "cannot allocate memory";
 556         ret = bch2_sb_realloc(sb, 0);
 557         if (ret)
 558                 goto err;
 559
 560         ret = -EFAULT;
 561         err = "dynamic fault";
 562         if (bch2_fs_init_fault("read_super"))
 563                 goto err;
 564
 565         ret = -EINVAL;
 566         err = read_one_super(sb, offset);
 567         if (!err)
 568                 goto got_super;
 569
 570         if (opt_defined(*opts, sb))
 571                 goto err;
 572
 573         pr_err("error reading default superblock: %s", err);
 574
 575         /*
 576          * Error reading primary superblock - read location of backup
 577          * superblocks:
 578          */
 579         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 580         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 581         sb->bio->bi_iter.bi_size = sizeof(struct bch_sb_layout);
 582         /*
 583          * use sb buffer to read layout, since sb buffer is page aligned but
 584          * layout won't be:
 585          */
 586         bch2_bio_map(sb->bio, sb->sb);
 587
 588         err = "IO error";
 589         if (submit_bio_wait(sb->bio))
 590                 goto err;
 591
 592         memcpy(&layout, sb->sb, sizeof(layout));
 593         err = validate_sb_layout(&layout);
 594         if (err)
 595                 goto err;
 596
 597         for (i = layout.sb_offset;
 598              i < layout.sb_offset + layout.nr_superblocks; i++) {
 599                 offset = le64_to_cpu(*i);
 600
 601                 if (offset == opt_get(*opts, sb))
 602                         continue;
 603
 604                 err = read_one_super(sb, offset);
 605                 if (!err)
 606                         goto got_super;
 607         }
 608
 609         ret = -EINVAL;
 610         goto err;
 611
 612 got_super:
 613         err = "Superblock block size smaller than device block size";
 614         ret = -EINVAL;
 615         if (le16_to_cpu(sb->sb->block_size) << 9 <
 616             bdev_logical_block_size(sb->bdev))
 617                 goto err;
 618
 619         ret = 0;
 620         sb->have_layout = true;
 621 out:
 622         pr_verbose_init(*opts, "ret %i", ret);
 623         return ret;
 624 err:
 625         bch2_free_super(sb);
 626         pr_err("error reading superblock: %s", err);
 627         goto out;
 628 }
 629
 630 /* write superblock: */
 631
 632 static void write_super_endio(struct bio *bio)
 633 {
 634         struct bch_dev *ca = bio->bi_private;
 635
 636         /* XXX: return errors directly */
 637
 638         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write"))
 639                 ca->sb_write_error = 1;
 640
 641         closure_put(&ca->fs->sb_write);
 642         percpu_ref_put(&ca->io_ref);
 643 }
 644
 645 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 646 {
 647         struct bch_sb *sb = ca->disk_sb.sb;
 648         struct bio *bio = ca->disk_sb.bio;
 649
 650         sb->offset = sb->layout.sb_offset[idx];
 651
 652         SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
 653         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 654                                 null_nonce(), sb);
 655
 656         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 657         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 658         bio->bi_iter.bi_size    =
 659                 roundup((size_t) vstruct_bytes(sb),
 660                         bdev_logical_block_size(ca->disk_sb.bdev));
 661         bio->bi_end_io          = write_super_endio;
 662         bio->bi_private         = ca;
 663         bch2_bio_map(bio, sb);
 664
 665         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_SB],
 666                      bio_sectors(bio));
 667
 668         percpu_ref_get(&ca->io_ref);
 669         closure_bio_submit(bio, &c->sb_write);
 670 }
 671
 672 void bch2_write_super(struct bch_fs *c)
 673 {
 674         struct closure *cl = &c->sb_write;
 675         struct bch_dev *ca;
 676         unsigned i, sb = 0, nr_wrote;
 677         const char *err;
 678         struct bch_devs_mask sb_written;
 679         bool wrote, can_mount_without_written, can_mount_with_written;
 680
 681         lockdep_assert_held(&c->sb_lock);
 682
 683         closure_init_stack(cl);
 684         memset(&sb_written, 0, sizeof(sb_written));
 685
 686         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 687
 688         for_each_online_member(ca, c, i)
 689                 bch2_sb_from_fs(c, ca);
 690
 691         for_each_online_member(ca, c, i) {
 692                 err = bch2_sb_validate(&ca->disk_sb);
 693                 if (err) {
 694                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
 695                         goto out;
 696                 }
 697         }
 698
 699         if (c->opts.nochanges ||
 700             test_bit(BCH_FS_ERROR, &c->flags))
 701                 goto out;
 702
 703         for_each_online_member(ca, c, i) {
 704                 __set_bit(ca->dev_idx, sb_written.d);
 705                 ca->sb_write_error = 0;
 706         }
 707
 708         do {
 709                 wrote = false;
 710                 for_each_online_member(ca, c, i)
 711                         if (sb < ca->disk_sb.sb->layout.nr_superblocks) {
 712                                 write_one_super(c, ca, sb);
 713                                 wrote = true;
 714                         }
 715                 closure_sync(cl);
 716                 sb++;
 717         } while (wrote);
 718
 719         for_each_online_member(ca, c, i)
 720                 if (ca->sb_write_error)
 721                         __clear_bit(ca->dev_idx, sb_written.d);
 722
 723         nr_wrote = dev_mask_nr(&sb_written);
 724
 725         can_mount_with_written =
 726                 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
 727                                       BCH_FORCE_IF_DEGRADED);
 728
 729         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 730                 sb_written.d[i] = ~sb_written.d[i];
 731
 732         can_mount_without_written =
 733                 bch2_have_enough_devs(__bch2_replicas_status(c, sb_written),
 734                                       BCH_FORCE_IF_DEGRADED);
 735
 736         /*
 737          * If we would be able to mount _without_ the devices we successfully
 738          * wrote superblocks to, we weren't able to write to enough devices:
 739          *
 740          * Exception: if we can mount without the successes because we haven't
 741          * written anything (new filesystem), we continue if we'd be able to
 742          * mount with the devices we did successfully write to:
 743          */
 744         bch2_fs_fatal_err_on(!nr_wrote ||
 745                              (can_mount_without_written &&
 746                               !can_mount_with_written), c,
 747                 "Unable to write superblock to sufficient devices");
 748 out:
 749         /* Make new options visible after they're persistent: */
 750         bch2_sb_update(c);
 751 }
 752
 753 /* BCH_SB_FIELD_journal: */
 754
 755 static int u64_cmp(const void *_l, const void *_r)
 756 {
 757         u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
 758
 759         return l < r ? -1 : l > r ? 1 : 0;
 760 }
 761
 762 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
 763                                             struct bch_sb_field *f)
 764 {
 765         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 766         struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
 767         const char *err;
 768         unsigned nr;
 769         unsigned i;
 770         u64 *b;
 771
 772         journal = bch2_sb_get_journal(sb);
 773         if (!journal)
 774                 return NULL;
 775
 776         nr = bch2_nr_journal_buckets(journal);
 777         if (!nr)
 778                 return NULL;
 779
 780         b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
 781         if (!b)
 782                 return "cannot allocate memory";
 783
 784         for (i = 0; i < nr; i++)
 785                 b[i] = le64_to_cpu(journal->buckets[i]);
 786
 787         sort(b, nr, sizeof(u64), u64_cmp, NULL);
 788
 789         err = "journal bucket at sector 0";
 790         if (!b[0])
 791                 goto err;
 792
 793         err = "journal bucket before first bucket";
 794         if (m && b[0] < le16_to_cpu(m->first_bucket))
 795                 goto err;
 796
 797         err = "journal bucket past end of device";
 798         if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
 799                 goto err;
 800
 801         err = "duplicate journal buckets";
 802         for (i = 0; i + 1 < nr; i++)
 803                 if (b[i] == b[i + 1])
 804                         goto err;
 805
 806         err = NULL;
 807 err:
 808         kfree(b);
 809         return err;
 810 }
 811
 812 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
 813         .validate       = bch2_sb_validate_journal,
 814 };
 815
 816 /* BCH_SB_FIELD_members: */
 817
 818 static const char *bch2_sb_validate_members(struct bch_sb *sb,
 819                                             struct bch_sb_field *f)
 820 {
 821         struct bch_sb_field_members *mi = field_to_type(f, members);
 822         struct bch_member *m;
 823
 824         if ((void *) (mi->members + sb->nr_devices) >
 825             vstruct_end(&mi->field))
 826                 return "Invalid superblock: bad member info";
 827
 828         for (m = mi->members;
 829              m < mi->members + sb->nr_devices;
 830              m++) {
 831                 if (!bch2_member_exists(m))
 832                         continue;
 833
 834                 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
 835                         return "Too many buckets";
 836
 837                 if (le64_to_cpu(m->nbuckets) -
 838                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
 839                         return "Not enough buckets";
 840
 841                 if (le16_to_cpu(m->bucket_size) <
 842                     le16_to_cpu(sb->block_size))
 843                         return "bucket size smaller than block size";
 844
 845                 if (le16_to_cpu(m->bucket_size) <
 846                     BCH_SB_BTREE_NODE_SIZE(sb))
 847                         return "bucket size smaller than btree node size";
 848         }
 849
 850         return NULL;
 851 }
 852
 853 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
 854         .validate       = bch2_sb_validate_members,
 855 };
 856
 857 /* BCH_SB_FIELD_crypt: */
 858
 859 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
 860                                           struct bch_sb_field *f)
 861 {
 862         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
 863
 864         if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
 865                 return "invalid field crypt: wrong size";
 866
 867         if (BCH_CRYPT_KDF_TYPE(crypt))
 868                 return "invalid field crypt: bad kdf type";
 869
 870         return NULL;
 871 }
 872
 873 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
 874         .validate       = bch2_sb_validate_crypt,
 875 };
 876
 877 /* BCH_SB_FIELD_clean: */
 878
 879 void bch2_sb_clean_renumber(struct bch_sb_field_clean *clean, int write)
 880 {
 881         struct jset_entry *entry;
 882
 883         for (entry = clean->start;
 884              entry < (struct jset_entry *) vstruct_end(&clean->field);
 885              entry = vstruct_next(entry))
 886                 bch2_bkey_renumber(BKEY_TYPE_BTREE, bkey_to_packed(entry->start), write);
 887 }
 888
 889 static void bch2_fs_mark_dirty(struct bch_fs *c)
 890 {
 891         mutex_lock(&c->sb_lock);
 892         if (BCH_SB_CLEAN(c->disk_sb.sb) ||
 893             (c->disk_sb.sb->compat[0] & (1ULL << BCH_COMPAT_FEAT_ALLOC_INFO))) {
 894                 SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 895                 c->disk_sb.sb->compat[0] &= ~(1ULL << BCH_COMPAT_FEAT_ALLOC_INFO);
 896                 bch2_write_super(c);
 897         }
 898         mutex_unlock(&c->sb_lock);
 899 }
 900
 901 struct jset_entry *
 902 bch2_journal_super_entries_add_common(struct bch_fs *c,
 903                                       struct jset_entry *entry,
 904                                       u64 journal_seq)
 905 {
 906         struct btree_root *r;
 907         unsigned i;
 908
 909         mutex_lock(&c->btree_root_lock);
 910
 911         for (r = c->btree_roots;
 912              r < c->btree_roots + BTREE_ID_NR;
 913              r++)
 914                 if (r->alive) {
 915                         entry->u64s     = r->key.u64s;
 916                         entry->btree_id = r - c->btree_roots;
 917                         entry->level    = r->level;
 918                         entry->type     = BCH_JSET_ENTRY_btree_root;
 919                         bkey_copy(&entry->start[0], &r->key);
 920
 921                         entry = vstruct_next(entry);
 922                 }
 923         c->btree_roots_dirty = false;
 924
 925         mutex_unlock(&c->btree_root_lock);
 926
 927         if (journal_seq)
 928                 return entry;
 929
 930         percpu_down_write(&c->mark_lock);
 931
 932         {
 933                 u64 nr_inodes = percpu_u64_get(&c->usage[0]->nr_inodes);
 934                 struct jset_entry_usage *u =
 935                         container_of(entry, struct jset_entry_usage, entry);
 936
 937                 memset(u, 0, sizeof(*u));
 938                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
 939                 u->entry.type   = BCH_JSET_ENTRY_usage;
 940                 u->entry.btree_id = FS_USAGE_INODES;
 941                 u->v            = cpu_to_le64(nr_inodes);
 942
 943                 entry = vstruct_next(entry);
 944         }
 945
 946         {
 947                 struct jset_entry_usage *u =
 948                         container_of(entry, struct jset_entry_usage, entry);
 949
 950                 memset(u, 0, sizeof(*u));
 951                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
 952                 u->entry.type   = BCH_JSET_ENTRY_usage;
 953                 u->entry.btree_id = FS_USAGE_KEY_VERSION;
 954                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
 955
 956                 entry = vstruct_next(entry);
 957         }
 958
 959         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
 960                 struct jset_entry_usage *u =
 961                         container_of(entry, struct jset_entry_usage, entry);
 962                 u64 sectors = percpu_u64_get(&c->usage[0]->persistent_reserved[i]);
 963
 964                 if (!sectors)
 965                         continue;
 966
 967                 memset(u, 0, sizeof(*u));
 968                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u), sizeof(u64)) - 1;
 969                 u->entry.type   = BCH_JSET_ENTRY_usage;
 970                 u->entry.btree_id = FS_USAGE_RESERVED;
 971                 u->entry.level  = i;
 972                 u->v            = sectors;
 973
 974                 entry = vstruct_next(entry);
 975         }
 976
 977         for (i = 0; i < c->replicas.nr; i++) {
 978                 struct bch_replicas_entry *e =
 979                         cpu_replicas_entry(&c->replicas, i);
 980                 u64 sectors = percpu_u64_get(&c->usage[0]->replicas[i]);
 981                 struct jset_entry_data_usage *u =
 982                         container_of(entry, struct jset_entry_data_usage, entry);
 983
 984                 memset(u, 0, sizeof(*u));
 985                 u->entry.u64s   = DIV_ROUND_UP(sizeof(*u) + e->nr_devs,
 986                                                sizeof(u64)) - 1;
 987                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
 988                 u->v            = cpu_to_le64(sectors);
 989                 unsafe_memcpy(&u->r, e, replicas_entry_bytes(e),
 990                               "embedded variable length struct");
 991
 992                 entry = vstruct_next(entry);
 993         }
 994
 995         percpu_up_write(&c->mark_lock);
 996
 997         return entry;
 998 }
 999
1000 void bch2_fs_mark_clean(struct bch_fs *c, bool clean)
1001 {
1002         struct bch_sb_field_clean *sb_clean;
1003         struct jset_entry *entry;
1004         unsigned u64s;
1005
1006         if (!clean) {
1007                 bch2_fs_mark_dirty(c);
1008                 return;
1009         }
1010
1011         mutex_lock(&c->sb_lock);
1012         if (BCH_SB_CLEAN(c->disk_sb.sb))
1013                 goto out;
1014
1015         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1016
1017         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1018
1019         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1020
1021         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1022         if (!sb_clean) {
1023                 bch_err(c, "error resizing superblock while setting filesystem clean");
1024                 goto out;
1025         }
1026
1027         sb_clean->flags         = 0;
1028         sb_clean->read_clock    = cpu_to_le16(c->bucket_clock[READ].hand);
1029         sb_clean->write_clock   = cpu_to_le16(c->bucket_clock[WRITE].hand);
1030         sb_clean->journal_seq   = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1031
1032         /* Trying to catch outstanding bug: */
1033         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1034
1035         entry = sb_clean->start;
1036         entry = bch2_journal_super_entries_add_common(c, entry, 0);
1037         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1038
1039         memset(entry, 0,
1040                vstruct_end(&sb_clean->field) - (void *) entry);
1041
1042         if (le16_to_cpu(c->disk_sb.sb->version) <
1043             bcachefs_metadata_version_bkey_renumber)
1044                 bch2_sb_clean_renumber(sb_clean, WRITE);
1045
1046         bch2_write_super(c);
1047 out:
1048         mutex_unlock(&c->sb_lock);
1049 }
1050
1051 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1052                                           struct bch_sb_field *f)
1053 {
1054         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1055
1056         if (vstruct_bytes(&clean->field) < sizeof(*clean))
1057                 return "invalid field crypt: wrong size";
1058
1059         return NULL;
1060 }
1061
1062 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1063         .validate       = bch2_sb_validate_clean,
1064 };
1065
1066 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1067 #define x(f, nr)                                        \
1068         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1069         BCH_SB_FIELDS()
1070 #undef x
1071 };
1072
1073 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1074                                           struct bch_sb_field *f)
1075 {
1076         unsigned type = le32_to_cpu(f->type);
1077
1078         return type < BCH_SB_FIELD_NR
1079                 ? bch2_sb_field_ops[type]->validate(sb, f)
1080                 : NULL;
1081 }
1082
1083 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1084                            struct bch_sb_field *f)
1085 {
1086         unsigned type = le32_to_cpu(f->type);
1087         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1088                 ? bch2_sb_field_ops[type] : NULL;
1089
1090         if (ops)
1091                 pr_buf(out, "%s", bch2_sb_fields[type]);
1092         else
1093                 pr_buf(out, "(unknown field %u)", type);
1094
1095         pr_buf(out, " (size %llu):", vstruct_bytes(f));
1096
1097         if (ops && ops->to_text)
1098                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1099 }