git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/super-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "btree_update_interior.h"
   5 #include "buckets.h"
   6 #include "checksum.h"
   7 #include "disk_groups.h"
   8 #include "ec.h"
   9 #include "error.h"
  10 #include "io.h"
  11 #include "journal.h"
  12 #include "journal_io.h"
  13 #include "journal_sb.h"
  14 #include "journal_seq_blacklist.h"
  15 #include "replicas.h"
  16 #include "quota.h"
  17 #include "super-io.h"
  18 #include "super.h"
  19 #include "vstructs.h"
  20
  21 #include <linux/backing-dev.h>
  22 #include <linux/sort.h>
  23
  24 const char * const bch2_sb_fields[] = {
  25 #define x(name, nr)     #name,
  26         BCH_SB_FIELDS()
  27 #undef x
  28         NULL
  29 };
  30
  31 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
  32                                   struct printbuf *);
  33
  34 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  35                                       enum bch_sb_field_type type)
  36 {
  37         struct bch_sb_field *f;
  38
  39         /* XXX: need locking around superblock to access optional fields */
  40
  41         vstruct_for_each(sb, f)
  42                 if (le32_to_cpu(f->type) == type)
  43                         return f;
  44         return NULL;
  45 }
  46
  47 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  48                                                    struct bch_sb_field *f,
  49                                                    unsigned u64s)
  50 {
  51         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  52         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  53
  54         BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
  55
  56         if (!f && !u64s) {
  57                 /* nothing to do: */
  58         } else if (!f) {
  59                 f = vstruct_last(sb->sb);
  60                 memset(f, 0, sizeof(u64) * u64s);
  61                 f->u64s = cpu_to_le32(u64s);
  62                 f->type = 0;
  63         } else {
  64                 void *src, *dst;
  65
  66                 src = vstruct_end(f);
  67
  68                 if (u64s) {
  69                         f->u64s = cpu_to_le32(u64s);
  70                         dst = vstruct_end(f);
  71                 } else {
  72                         dst = f;
  73                 }
  74
  75                 memmove(dst, src, vstruct_end(sb->sb) - src);
  76
  77                 if (dst > src)
  78                         memset(src, 0, dst - src);
  79         }
  80
  81         sb->sb->u64s = cpu_to_le32(sb_u64s);
  82
  83         return u64s ? f : NULL;
  84 }
  85
  86 void bch2_sb_field_delete(struct bch_sb_handle *sb,
  87                           enum bch_sb_field_type type)
  88 {
  89         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
  90
  91         if (f)
  92                 __bch2_sb_field_resize(sb, f, 0);
  93 }
  94
  95 /* Superblock realloc/free: */
  96
  97 void bch2_free_super(struct bch_sb_handle *sb)
  98 {
  99         if (sb->bio)
 100                 bio_put(sb->bio);
 101         if (!IS_ERR_OR_NULL(sb->bdev))
 102                 blkdev_put(sb->bdev, sb->mode);
 103
 104         kfree(sb->sb);
 105         memset(sb, 0, sizeof(*sb));
 106 }
 107
 108 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 109 {
 110         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 111         size_t new_buffer_size;
 112         struct bch_sb *new_sb;
 113         struct bio *bio;
 114
 115         if (sb->bdev)
 116                 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
 117
 118         new_buffer_size = roundup_pow_of_two(new_bytes);
 119
 120         if (sb->sb && sb->buffer_size >= new_buffer_size)
 121                 return 0;
 122
 123         if (sb->have_layout) {
 124                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 125
 126                 if (new_bytes > max_bytes) {
 127                         char buf[BDEVNAME_SIZE];
 128
 129                         pr_err("%s: superblock too big: want %zu but have %llu",
 130                                bdevname(sb->bdev, buf), new_bytes, max_bytes);
 131                         return -ENOSPC;
 132                 }
 133         }
 134
 135         if (sb->buffer_size >= new_buffer_size && sb->sb)
 136                 return 0;
 137
 138         if (dynamic_fault("bcachefs:add:super_realloc"))
 139                 return -ENOMEM;
 140
 141         if (sb->have_bio) {
 142                 bio = bio_kmalloc(GFP_KERNEL,
 143                         DIV_ROUND_UP(new_buffer_size, PAGE_SIZE));
 144                 if (!bio)
 145                         return -ENOMEM;
 146
 147                 if (sb->bio)
 148                         bio_put(sb->bio);
 149                 sb->bio = bio;
 150         }
 151
 152         new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 153         if (!new_sb)
 154                 return -ENOMEM;
 155
 156         sb->sb = new_sb;
 157         sb->buffer_size = new_buffer_size;
 158
 159         return 0;
 160 }
 161
 162 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 163                                           enum bch_sb_field_type type,
 164                                           unsigned u64s)
 165 {
 166         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 167         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 168         ssize_t d = -old_u64s + u64s;
 169
 170         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 171                 return NULL;
 172
 173         if (sb->fs_sb) {
 174                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 175                 struct bch_dev *ca;
 176                 unsigned i;
 177
 178                 lockdep_assert_held(&c->sb_lock);
 179
 180                 /* XXX: we're not checking that offline device have enough space */
 181
 182                 for_each_online_member(ca, c, i) {
 183                         struct bch_sb_handle *sb = &ca->disk_sb;
 184
 185                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 186                                 percpu_ref_put(&ca->ref);
 187                                 return NULL;
 188                         }
 189                 }
 190         }
 191
 192         f = bch2_sb_field_get(sb->sb, type);
 193         f = __bch2_sb_field_resize(sb, f, u64s);
 194         if (f)
 195                 f->type = cpu_to_le32(type);
 196         return f;
 197 }
 198
 199 /* Superblock validate: */
 200
 201 static inline void __bch2_sb_layout_size_assert(void)
 202 {
 203         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 204 }
 205
 206 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
 207 {
 208         u64 offset, prev_offset, max_sectors;
 209         unsigned i;
 210
 211         if (uuid_le_cmp(layout->magic, BCACHE_MAGIC)) {
 212                 pr_buf(out, "Not a bcachefs superblock layout");
 213                 return -EINVAL;
 214         }
 215
 216         if (layout->layout_type != 0) {
 217                 pr_buf(out, "Invalid superblock layout type %u",
 218                        layout->layout_type);
 219                 return -EINVAL;
 220         }
 221
 222         if (!layout->nr_superblocks) {
 223                 pr_buf(out, "Invalid superblock layout: no superblocks");
 224                 return -EINVAL;
 225         }
 226
 227         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
 228                 pr_buf(out, "Invalid superblock layout: too many superblocks");
 229                 return -EINVAL;
 230         }
 231
 232         max_sectors = 1 << layout->sb_max_size_bits;
 233
 234         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 235
 236         for (i = 1; i < layout->nr_superblocks; i++) {
 237                 offset = le64_to_cpu(layout->sb_offset[i]);
 238
 239                 if (offset < prev_offset + max_sectors) {
 240                         pr_buf(out, "Invalid superblock layout: superblocks overlap\n"
 241                                "  (sb %u ends at %llu next starts at %llu",
 242                                i - 1, prev_offset + max_sectors, offset);
 243                         return -EINVAL;
 244                 }
 245                 prev_offset = offset;
 246         }
 247
 248         return 0;
 249 }
 250
 251 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out)
 252 {
 253         struct bch_sb *sb = disk_sb->sb;
 254         struct bch_sb_field *f;
 255         struct bch_sb_field_members *mi;
 256         enum bch_opt_id opt_id;
 257         u32 version, version_min;
 258         u16 block_size;
 259         int ret;
 260
 261         version         = le16_to_cpu(sb->version);
 262         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 263                 ? le16_to_cpu(sb->version_min)
 264                 : version;
 265
 266         if (version    >= bcachefs_metadata_version_max) {
 267                 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
 268                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 269                 return -EINVAL;
 270         }
 271
 272         if (version_min < bcachefs_metadata_version_min) {
 273                 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
 274                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 275                 return -EINVAL;
 276         }
 277
 278         if (version_min > version) {
 279                 pr_buf(out, "Bad minimum version %u, greater than version field %u",
 280                        version_min, version);
 281                 return -EINVAL;
 282         }
 283
 284         if (sb->features[1] ||
 285             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
 286                 pr_buf(out, "Filesystem has incompatible features");
 287                 return -EINVAL;
 288         }
 289
 290         block_size = le16_to_cpu(sb->block_size);
 291
 292         if (block_size > PAGE_SECTORS) {
 293                 pr_buf(out, "Block size too big (got %u, max %u)",
 294                        block_size, PAGE_SECTORS);
 295                 return -EINVAL;
 296         }
 297
 298         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
 299                 pr_buf(out, "Bad user UUID (got zeroes)");
 300                 return -EINVAL;
 301         }
 302
 303         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
 304                 pr_buf(out, "Bad intenal UUID (got zeroes)");
 305                 return -EINVAL;
 306         }
 307
 308         if (!sb->nr_devices ||
 309             sb->nr_devices > BCH_SB_MEMBERS_MAX) {
 310                 pr_buf(out, "Bad number of member devices %u (max %u)",
 311                        sb->nr_devices, BCH_SB_MEMBERS_MAX);
 312                 return -EINVAL;
 313         }
 314
 315         if (sb->dev_idx >= sb->nr_devices) {
 316                 pr_buf(out, "Bad dev_idx (got %u, nr_devices %u)",
 317                        sb->dev_idx, sb->nr_devices);
 318                 return -EINVAL;
 319         }
 320
 321         if (!sb->time_precision ||
 322             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
 323                 pr_buf(out, "Invalid time precision: %u (min 1, max %lu)",
 324                        le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
 325                 return -EINVAL;
 326         }
 327
 328         for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
 329                 const struct bch_option *opt = bch2_opt_table + opt_id;
 330
 331                 if (opt->get_sb != BCH2_NO_SB_OPT) {
 332                         u64 v = bch2_opt_from_sb(sb, opt_id);
 333
 334                         pr_buf(out, "Invalid option ");
 335                         ret = bch2_opt_validate(opt, v, out);
 336                         if (ret)
 337                                 return ret;
 338
 339                         printbuf_reset(out);
 340                 }
 341         }
 342
 343         /* validate layout */
 344         ret = validate_sb_layout(&sb->layout, out);
 345         if (ret)
 346                 return ret;
 347
 348         vstruct_for_each(sb, f) {
 349                 if (!f->u64s) {
 350                         pr_buf(out, "Invalid superblock: optional with size 0 (type %u)",
 351                                le32_to_cpu(f->type));
 352                         return -EINVAL;
 353                 }
 354
 355                 if (vstruct_next(f) > vstruct_last(sb)) {
 356                         pr_buf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
 357                                le32_to_cpu(f->type));
 358                         return -EINVAL;
 359                 }
 360         }
 361
 362         /* members must be validated first: */
 363         mi = bch2_sb_get_members(sb);
 364         if (!mi) {
 365                 pr_buf(out, "Invalid superblock: member info area missing");
 366                 return -EINVAL;
 367         }
 368
 369         ret = bch2_sb_field_validate(sb, &mi->field, out);
 370         if (ret)
 371                 return ret;
 372
 373         vstruct_for_each(sb, f) {
 374                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 375                         continue;
 376
 377                 ret = bch2_sb_field_validate(sb, f, out);
 378                 if (ret)
 379                         return ret;
 380         }
 381
 382         return 0;
 383 }
 384
 385 /* device open: */
 386
 387 static void bch2_sb_update(struct bch_fs *c)
 388 {
 389         struct bch_sb *src = c->disk_sb.sb;
 390         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 391         struct bch_dev *ca;
 392         unsigned i;
 393
 394         lockdep_assert_held(&c->sb_lock);
 395
 396         c->sb.uuid              = src->uuid;
 397         c->sb.user_uuid         = src->user_uuid;
 398         c->sb.version           = le16_to_cpu(src->version);
 399         c->sb.version_min       = le16_to_cpu(src->version_min);
 400         c->sb.nr_devices        = src->nr_devices;
 401         c->sb.clean             = BCH_SB_CLEAN(src);
 402         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 403
 404         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 405         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 406
 407         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 408         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 409                                           c->sb.nsec_per_time_unit);
 410         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 411
 412         c->sb.features          = le64_to_cpu(src->features[0]);
 413         c->sb.compat            = le64_to_cpu(src->compat[0]);
 414
 415         for_each_member_device(ca, c, i)
 416                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 417 }
 418
 419 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 420 {
 421         struct bch_sb_field *src_f, *dst_f;
 422         struct bch_sb *dst = dst_handle->sb;
 423         unsigned i;
 424
 425         dst->version            = src->version;
 426         dst->version_min        = src->version_min;
 427         dst->seq                = src->seq;
 428         dst->uuid               = src->uuid;
 429         dst->user_uuid          = src->user_uuid;
 430         memcpy(dst->label,      src->label, sizeof(dst->label));
 431
 432         dst->block_size         = src->block_size;
 433         dst->nr_devices         = src->nr_devices;
 434
 435         dst->time_base_lo       = src->time_base_lo;
 436         dst->time_base_hi       = src->time_base_hi;
 437         dst->time_precision     = src->time_precision;
 438
 439         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 440         memcpy(dst->features,   src->features,  sizeof(dst->features));
 441         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 442
 443         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 444                 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 445                         continue;
 446
 447                 src_f = bch2_sb_field_get(src, i);
 448                 dst_f = bch2_sb_field_get(dst, i);
 449                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 450                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 451
 452                 if (src_f)
 453                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 454         }
 455 }
 456
 457 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 458 {
 459         struct bch_sb_field_journal *journal_buckets =
 460                 bch2_sb_get_journal(src);
 461         unsigned journal_u64s = journal_buckets
 462                 ? le32_to_cpu(journal_buckets->field.u64s)
 463                 : 0;
 464         int ret;
 465
 466         lockdep_assert_held(&c->sb_lock);
 467
 468         ret = bch2_sb_realloc(&c->disk_sb,
 469                               le32_to_cpu(src->u64s) - journal_u64s);
 470         if (ret)
 471                 return ret;
 472
 473         __copy_super(&c->disk_sb, src);
 474
 475         ret = bch2_sb_replicas_to_cpu_replicas(c);
 476         if (ret)
 477                 return ret;
 478
 479         ret = bch2_sb_disk_groups_to_cpu(c);
 480         if (ret)
 481                 return ret;
 482
 483         bch2_sb_update(c);
 484         return 0;
 485 }
 486
 487 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 488 {
 489         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 490         struct bch_sb_field_journal *journal_buckets =
 491                 bch2_sb_get_journal(dst);
 492         unsigned journal_u64s = journal_buckets
 493                 ? le32_to_cpu(journal_buckets->field.u64s)
 494                 : 0;
 495         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 496         int ret;
 497
 498         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 499         if (ret)
 500                 return ret;
 501
 502         __copy_super(&ca->disk_sb, src);
 503         return 0;
 504 }
 505
 506 /* read superblock: */
 507
 508 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 509 {
 510         struct bch_csum csum;
 511         u32 version, version_min;
 512         size_t bytes;
 513         int ret;
 514 reread:
 515         bio_reset(sb->bio);
 516         bio_set_dev(sb->bio, sb->bdev);
 517         sb->bio->bi_iter.bi_sector = offset;
 518         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 519         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 520
 521         ret = submit_bio_wait(sb->bio);
 522         if (ret) {
 523                 pr_buf(err, "IO error: %i", ret);
 524                 return ret;
 525         }
 526
 527         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC)) {
 528                 pr_buf(err, "Not a bcachefs superblock");
 529                 return -EINVAL;
 530         }
 531
 532         version         = le16_to_cpu(sb->sb->version);
 533         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 534                 ? le16_to_cpu(sb->sb->version_min)
 535                 : version;
 536
 537         if (version    >= bcachefs_metadata_version_max) {
 538                 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
 539                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 540                 return -EINVAL;
 541         }
 542
 543         if (version_min < bcachefs_metadata_version_min) {
 544                 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
 545                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 546                 return -EINVAL;
 547         }
 548
 549         bytes = vstruct_bytes(sb->sb);
 550
 551         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 552                 pr_buf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 553                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 554                 return -EINVAL;
 555         }
 556
 557         if (bytes > sb->buffer_size) {
 558                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 559                         return -ENOMEM;
 560                 goto reread;
 561         }
 562
 563         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
 564                 pr_buf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 565                 return -EINVAL;
 566         }
 567
 568         /* XXX: verify MACs */
 569         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 570                             null_nonce(), sb->sb);
 571
 572         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 573                 pr_buf(err, "bad checksum");
 574                 return -EINVAL;
 575         }
 576
 577         sb->seq = le64_to_cpu(sb->sb->seq);
 578
 579         return 0;
 580 }
 581
 582 int bch2_read_super(const char *path, struct bch_opts *opts,
 583                     struct bch_sb_handle *sb)
 584 {
 585         u64 offset = opt_get(*opts, sb);
 586         struct bch_sb_layout layout;
 587         struct printbuf err = PRINTBUF;
 588         __le64 *i;
 589         int ret;
 590
 591         pr_verbose_init(*opts, "");
 592
 593         memset(sb, 0, sizeof(*sb));
 594         sb->mode        = FMODE_READ;
 595         sb->have_bio    = true;
 596
 597         if (!opt_get(*opts, noexcl))
 598                 sb->mode |= FMODE_EXCL;
 599
 600         if (!opt_get(*opts, nochanges))
 601                 sb->mode |= FMODE_WRITE;
 602
 603         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 604         if (IS_ERR(sb->bdev) &&
 605             PTR_ERR(sb->bdev) == -EACCES &&
 606             opt_get(*opts, read_only)) {
 607                 sb->mode &= ~FMODE_WRITE;
 608
 609                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 610                 if (!IS_ERR(sb->bdev))
 611                         opt_set(*opts, nochanges, true);
 612         }
 613
 614         if (IS_ERR(sb->bdev)) {
 615                 ret = PTR_ERR(sb->bdev);
 616                 goto out;
 617         }
 618
 619         ret = bch2_sb_realloc(sb, 0);
 620         if (ret) {
 621                 pr_buf(&err, "error allocating memory for superblock");
 622                 goto err;
 623         }
 624
 625         if (bch2_fs_init_fault("read_super")) {
 626                 pr_buf(&err, "dynamic fault");
 627                 ret = -EFAULT;
 628                 goto err;
 629         }
 630
 631         ret = read_one_super(sb, offset, &err);
 632         if (!ret)
 633                 goto got_super;
 634
 635         if (opt_defined(*opts, sb))
 636                 goto err;
 637
 638         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 639                path, err.buf);
 640         printbuf_reset(&err);
 641
 642         /*
 643          * Error reading primary superblock - read location of backup
 644          * superblocks:
 645          */
 646         bio_reset(sb->bio);
 647         bio_set_dev(sb->bio, sb->bdev);
 648         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 649         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 650         /*
 651          * use sb buffer to read layout, since sb buffer is page aligned but
 652          * layout won't be:
 653          */
 654         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 655
 656         ret = submit_bio_wait(sb->bio);
 657         if (ret) {
 658                 pr_buf(&err, "IO error: %i", ret);
 659                 goto err;
 660         }
 661
 662         memcpy(&layout, sb->sb, sizeof(layout));
 663         ret = validate_sb_layout(&layout, &err);
 664         if (ret)
 665                 goto err;
 666
 667         for (i = layout.sb_offset;
 668              i < layout.sb_offset + layout.nr_superblocks; i++) {
 669                 offset = le64_to_cpu(*i);
 670
 671                 if (offset == opt_get(*opts, sb))
 672                         continue;
 673
 674                 ret = read_one_super(sb, offset, &err);
 675                 if (!ret)
 676                         goto got_super;
 677         }
 678
 679         goto err;
 680
 681 got_super:
 682         if (le16_to_cpu(sb->sb->block_size) << 9 <
 683             bdev_logical_block_size(sb->bdev)) {
 684                 pr_buf(&err, "block size (%u) smaller than device block size (%u)",
 685                        le16_to_cpu(sb->sb->block_size) << 9,
 686                        bdev_logical_block_size(sb->bdev));
 687                 ret = -EINVAL;
 688                 goto err;
 689         }
 690
 691         ret = 0;
 692         sb->have_layout = true;
 693
 694         ret = bch2_sb_validate(sb, &err);
 695         if (ret) {
 696                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 697                        path, err.buf);
 698                 goto err_no_print;
 699         }
 700 out:
 701         pr_verbose_init(*opts, "ret %i", ret);
 702         printbuf_exit(&err);
 703         return ret;
 704 err:
 705         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 706                path, err.buf);
 707 err_no_print:
 708         bch2_free_super(sb);
 709         goto out;
 710 }
 711
 712 /* write superblock: */
 713
 714 static void write_super_endio(struct bio *bio)
 715 {
 716         struct bch_dev *ca = bio->bi_private;
 717
 718         /* XXX: return errors directly */
 719
 720         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 721                                bch2_blk_status_to_str(bio->bi_status)))
 722                 ca->sb_write_error = 1;
 723
 724         closure_put(&ca->fs->sb_write);
 725         percpu_ref_put(&ca->io_ref);
 726 }
 727
 728 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 729 {
 730         struct bch_sb *sb = ca->disk_sb.sb;
 731         struct bio *bio = ca->disk_sb.bio;
 732
 733         bio_reset(bio);
 734         bio_set_dev(bio, ca->disk_sb.bdev);
 735         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 736         bio->bi_end_io          = write_super_endio;
 737         bio->bi_private         = ca;
 738         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 739         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 740
 741         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 742                      bio_sectors(bio));
 743
 744         percpu_ref_get(&ca->io_ref);
 745         closure_bio_submit(bio, &c->sb_write);
 746 }
 747
 748 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 749 {
 750         struct bch_sb *sb = ca->disk_sb.sb;
 751         struct bio *bio = ca->disk_sb.bio;
 752
 753         sb->offset = sb->layout.sb_offset[idx];
 754
 755         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 756         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 757                                 null_nonce(), sb);
 758
 759         bio_reset(bio);
 760         bio_set_dev(bio, ca->disk_sb.bdev);
 761         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 762         bio->bi_end_io          = write_super_endio;
 763         bio->bi_private         = ca;
 764         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 765         bch2_bio_map(bio, sb,
 766                      roundup((size_t) vstruct_bytes(sb),
 767                              bdev_logical_block_size(ca->disk_sb.bdev)));
 768
 769         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 770                      bio_sectors(bio));
 771
 772         percpu_ref_get(&ca->io_ref);
 773         closure_bio_submit(bio, &c->sb_write);
 774 }
 775
 776 int bch2_write_super(struct bch_fs *c)
 777 {
 778         struct closure *cl = &c->sb_write;
 779         struct bch_dev *ca;
 780         struct printbuf err = PRINTBUF;
 781         unsigned i, sb = 0, nr_wrote;
 782         struct bch_devs_mask sb_written;
 783         bool wrote, can_mount_without_written, can_mount_with_written;
 784         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 785         int ret = 0;
 786
 787         if (c->opts.very_degraded)
 788                 degraded_flags |= BCH_FORCE_IF_LOST;
 789
 790         lockdep_assert_held(&c->sb_lock);
 791
 792         closure_init_stack(cl);
 793         memset(&sb_written, 0, sizeof(sb_written));
 794
 795         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 796
 797         if (test_bit(BCH_FS_ERROR, &c->flags))
 798                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 799         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 800                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 801
 802         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 803
 804         for_each_online_member(ca, c, i)
 805                 bch2_sb_from_fs(c, ca);
 806
 807         for_each_online_member(ca, c, i) {
 808                 printbuf_reset(&err);
 809
 810                 ret = bch2_sb_validate(&ca->disk_sb, &err);
 811                 if (ret) {
 812                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 813                         percpu_ref_put(&ca->io_ref);
 814                         goto out;
 815                 }
 816         }
 817
 818         if (c->opts.nochanges)
 819                 goto out;
 820
 821         for_each_online_member(ca, c, i) {
 822                 __set_bit(ca->dev_idx, sb_written.d);
 823                 ca->sb_write_error = 0;
 824         }
 825
 826         for_each_online_member(ca, c, i)
 827                 read_back_super(c, ca);
 828         closure_sync(cl);
 829
 830         for_each_online_member(ca, c, i) {
 831                 if (ca->sb_write_error)
 832                         continue;
 833
 834                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 835                         bch2_fs_fatal_error(c,
 836                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 837                                 le64_to_cpu(ca->sb_read_scratch->seq),
 838                                 ca->disk_sb.seq);
 839                         percpu_ref_put(&ca->io_ref);
 840                         ret = -EROFS;
 841                         goto out;
 842                 }
 843
 844                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 845                         bch2_fs_fatal_error(c,
 846                                 "Superblock modified by another process (seq %llu expected %llu)",
 847                                 le64_to_cpu(ca->sb_read_scratch->seq),
 848                                 ca->disk_sb.seq);
 849                         percpu_ref_put(&ca->io_ref);
 850                         ret = -EROFS;
 851                         goto out;
 852                 }
 853         }
 854
 855         do {
 856                 wrote = false;
 857                 for_each_online_member(ca, c, i)
 858                         if (!ca->sb_write_error &&
 859                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 860                                 write_one_super(c, ca, sb);
 861                                 wrote = true;
 862                         }
 863                 closure_sync(cl);
 864                 sb++;
 865         } while (wrote);
 866
 867         for_each_online_member(ca, c, i) {
 868                 if (ca->sb_write_error)
 869                         __clear_bit(ca->dev_idx, sb_written.d);
 870                 else
 871                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 872         }
 873
 874         nr_wrote = dev_mask_nr(&sb_written);
 875
 876         can_mount_with_written =
 877                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 878
 879         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 880                 sb_written.d[i] = ~sb_written.d[i];
 881
 882         can_mount_without_written =
 883                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 884
 885         /*
 886          * If we would be able to mount _without_ the devices we successfully
 887          * wrote superblocks to, we weren't able to write to enough devices:
 888          *
 889          * Exception: if we can mount without the successes because we haven't
 890          * written anything (new filesystem), we continue if we'd be able to
 891          * mount with the devices we did successfully write to:
 892          */
 893         if (bch2_fs_fatal_err_on(!nr_wrote ||
 894                                  !can_mount_with_written ||
 895                                  (can_mount_without_written &&
 896                                   !can_mount_with_written), c,
 897                 "Unable to write superblock to sufficient devices (from %ps)",
 898                 (void *) _RET_IP_))
 899                 ret = -1;
 900 out:
 901         /* Make new options visible after they're persistent: */
 902         bch2_sb_update(c);
 903         printbuf_exit(&err);
 904         return ret;
 905 }
 906
 907 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 908 {
 909         mutex_lock(&c->sb_lock);
 910         if (!(c->sb.features & (1ULL << feat))) {
 911                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 912
 913                 bch2_write_super(c);
 914         }
 915         mutex_unlock(&c->sb_lock);
 916 }
 917
 918 /* BCH_SB_FIELD_members: */
 919
 920 static int bch2_sb_members_validate(struct bch_sb *sb,
 921                                     struct bch_sb_field *f,
 922                                     struct printbuf *err)
 923 {
 924         struct bch_sb_field_members *mi = field_to_type(f, members);
 925         unsigned i;
 926
 927         if ((void *) (mi->members + sb->nr_devices) >
 928             vstruct_end(&mi->field)) {
 929                 pr_buf(err, "too many devices for section size");
 930                 return -EINVAL;
 931         }
 932
 933         for (i = 0; i < sb->nr_devices; i++) {
 934                 struct bch_member *m = mi->members + i;
 935
 936                 if (!bch2_member_exists(m))
 937                         continue;
 938
 939                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
 940                         pr_buf(err, "device %u: too many buckets (got %llu, max %lu)",
 941                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
 942                         return -EINVAL;
 943                 }
 944
 945                 if (le64_to_cpu(m->nbuckets) -
 946                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
 947                         pr_buf(err, "device %u: not enough buckets (got %llu, max %u)",
 948                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
 949                         return -EINVAL;
 950                 }
 951
 952                 if (le16_to_cpu(m->bucket_size) <
 953                     le16_to_cpu(sb->block_size)) {
 954                         pr_buf(err, "device %u: bucket size %u smaller than block size %u",
 955                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
 956                         return -EINVAL;
 957                 }
 958
 959                 if (le16_to_cpu(m->bucket_size) <
 960                     BCH_SB_BTREE_NODE_SIZE(sb)) {
 961                         pr_buf(err, "device %u: bucket size %u smaller than btree node size %llu",
 962                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
 963                         return -EINVAL;
 964                 }
 965         }
 966
 967         return 0;
 968 }
 969
 970 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
 971                                     struct bch_sb_field *f)
 972 {
 973         struct bch_sb_field_members *mi = field_to_type(f, members);
 974         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
 975         unsigned i;
 976
 977         for (i = 0; i < sb->nr_devices; i++) {
 978                 struct bch_member *m = mi->members + i;
 979                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
 980                 u64 bucket_size = le16_to_cpu(m->bucket_size);
 981                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
 982
 983                 if (!bch2_member_exists(m))
 984                         continue;
 985
 986                 pr_buf(out, "Device:");
 987                 pr_tab(out);
 988                 pr_buf(out, "%u", i);
 989                 pr_newline(out);
 990
 991                 pr_indent_push(out, 2);
 992
 993                 pr_buf(out, "UUID:");
 994                 pr_tab(out);
 995                 pr_uuid(out, m->uuid.b);
 996                 pr_newline(out);
 997
 998                 pr_buf(out, "Size:");
 999                 pr_tab(out);
1000                 pr_units(out, device_size, device_size << 9);
1001                 pr_newline(out);
1002
1003                 pr_buf(out, "Bucket size:");
1004                 pr_tab(out);
1005                 pr_units(out, bucket_size, bucket_size << 9);
1006                 pr_newline(out);
1007
1008                 pr_buf(out, "First bucket:");
1009                 pr_tab(out);
1010                 pr_buf(out, "%u", le16_to_cpu(m->first_bucket));
1011                 pr_newline(out);
1012
1013                 pr_buf(out, "Buckets:");
1014                 pr_tab(out);
1015                 pr_buf(out, "%llu", le64_to_cpu(m->nbuckets));
1016                 pr_newline(out);
1017
1018                 pr_buf(out, "Last mount:");
1019                 pr_tab(out);
1020                 if (m->last_mount)
1021                         pr_time(out, le64_to_cpu(m->last_mount));
1022                 else
1023                         pr_buf(out, "(never)");
1024                 pr_newline(out);
1025
1026                 pr_buf(out, "State:");
1027                 pr_tab(out);
1028                 pr_buf(out, "%s",
1029                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1030                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1031                        : "unknown");
1032                 pr_newline(out);
1033
1034                 pr_buf(out, "Group:");
1035                 pr_tab(out);
1036                 if (BCH_MEMBER_GROUP(m)) {
1037                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1038
1039                         if (idx < disk_groups_nr(gi))
1040                                 pr_buf(out, "%s (%u)",
1041                                        gi->entries[idx].label, idx);
1042                         else
1043                                 pr_buf(out, "(bad disk labels section)");
1044                 } else {
1045                         pr_buf(out, "(none)");
1046                 }
1047                 pr_newline(out);
1048
1049                 pr_buf(out, "Data allowed:");
1050                 pr_tab(out);
1051                 if (BCH_MEMBER_DATA_ALLOWED(m))
1052                         bch2_flags_to_text(out, bch2_data_types,
1053                                            BCH_MEMBER_DATA_ALLOWED(m));
1054                 else
1055                         pr_buf(out, "(none)");
1056                 pr_newline(out);
1057
1058                 pr_buf(out, "Has data:");
1059                 pr_tab(out);
1060                 if (data_have)
1061                         bch2_flags_to_text(out, bch2_data_types, data_have);
1062                 else
1063                         pr_buf(out, "(none)");
1064                 pr_newline(out);
1065
1066                 pr_buf(out, "Discard:");
1067                 pr_tab(out);
1068                 pr_buf(out, "%llu", BCH_MEMBER_DISCARD(m));
1069                 pr_newline(out);
1070
1071                 pr_buf(out, "Freespace initialized:");
1072                 pr_tab(out);
1073                 pr_buf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1074                 pr_newline(out);
1075
1076                 pr_indent_pop(out, 2);
1077         }
1078 }
1079
1080 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1081         .validate       = bch2_sb_members_validate,
1082         .to_text        = bch2_sb_members_to_text,
1083 };
1084
1085 /* BCH_SB_FIELD_crypt: */
1086
1087 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1088                                   struct bch_sb_field *f,
1089                                   struct printbuf *err)
1090 {
1091         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1092
1093         if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1094                 pr_buf(err, "wrong size (got %zu should be %zu)",
1095                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1096                 return -EINVAL;
1097         }
1098
1099         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1100                 pr_buf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1101                 return -EINVAL;
1102         }
1103
1104         return 0;
1105 }
1106
1107 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1108                                   struct bch_sb_field *f)
1109 {
1110         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1111
1112         pr_buf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1113         pr_newline(out);
1114         pr_buf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1115         pr_newline(out);
1116         pr_buf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1117         pr_newline(out);
1118         pr_buf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1119         pr_newline(out);
1120 }
1121
1122 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1123         .validate       = bch2_sb_crypt_validate,
1124         .to_text        = bch2_sb_crypt_to_text,
1125 };
1126
1127 /* BCH_SB_FIELD_clean: */
1128
1129 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1130 {
1131         struct jset_entry *entry;
1132         int ret;
1133
1134         for (entry = clean->start;
1135              entry < (struct jset_entry *) vstruct_end(&clean->field);
1136              entry = vstruct_next(entry)) {
1137                 ret = bch2_journal_entry_validate(c, "superblock", entry,
1138                                                   le16_to_cpu(c->disk_sb.sb->version),
1139                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1140                                                   write);
1141                 if (ret)
1142                         return ret;
1143         }
1144
1145         return 0;
1146 }
1147
1148 int bch2_fs_mark_dirty(struct bch_fs *c)
1149 {
1150         int ret;
1151
1152         /*
1153          * Unconditionally write superblock, to verify it hasn't changed before
1154          * we go rw:
1155          */
1156
1157         mutex_lock(&c->sb_lock);
1158         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1159         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1160         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1161         ret = bch2_write_super(c);
1162         mutex_unlock(&c->sb_lock);
1163
1164         return ret;
1165 }
1166
1167 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1168 {
1169         struct jset_entry *entry = *end;
1170         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1171
1172         memset(entry, 0, u64s * sizeof(u64));
1173         /*
1174          * The u64s field counts from the start of data, ignoring the shared
1175          * fields.
1176          */
1177         entry->u64s = cpu_to_le16(u64s - 1);
1178
1179         *end = vstruct_next(*end);
1180         return entry;
1181 }
1182
1183 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1184                                            struct jset_entry **end,
1185                                            u64 journal_seq)
1186 {
1187         struct bch_dev *ca;
1188         unsigned i, dev;
1189
1190         percpu_down_read(&c->mark_lock);
1191
1192         if (!journal_seq) {
1193                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1194                         bch2_fs_usage_acc_to_base(c, i);
1195         } else {
1196                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1197         }
1198
1199         {
1200                 struct jset_entry_usage *u =
1201                         container_of(jset_entry_init(end, sizeof(*u)),
1202                                      struct jset_entry_usage, entry);
1203
1204                 u->entry.type   = BCH_JSET_ENTRY_usage;
1205                 u->entry.btree_id = BCH_FS_USAGE_inodes;
1206                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1207         }
1208
1209         {
1210                 struct jset_entry_usage *u =
1211                         container_of(jset_entry_init(end, sizeof(*u)),
1212                                      struct jset_entry_usage, entry);
1213
1214                 u->entry.type   = BCH_JSET_ENTRY_usage;
1215                 u->entry.btree_id = BCH_FS_USAGE_key_version;
1216                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1217         }
1218
1219         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1220                 struct jset_entry_usage *u =
1221                         container_of(jset_entry_init(end, sizeof(*u)),
1222                                      struct jset_entry_usage, entry);
1223
1224                 u->entry.type   = BCH_JSET_ENTRY_usage;
1225                 u->entry.btree_id = BCH_FS_USAGE_reserved;
1226                 u->entry.level  = i;
1227                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1228         }
1229
1230         for (i = 0; i < c->replicas.nr; i++) {
1231                 struct bch_replicas_entry *e =
1232                         cpu_replicas_entry(&c->replicas, i);
1233                 struct jset_entry_data_usage *u =
1234                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1235                                      struct jset_entry_data_usage, entry);
1236
1237                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1238                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1239                 memcpy(&u->r, e, replicas_entry_bytes(e));
1240         }
1241
1242         for_each_member_device(ca, c, dev) {
1243                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1244                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1245                 struct jset_entry_dev_usage *u =
1246                         container_of(jset_entry_init(end, b),
1247                                      struct jset_entry_dev_usage, entry);
1248
1249                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1250                 u->dev = cpu_to_le32(dev);
1251                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1252                 u->buckets_unavailable  = cpu_to_le64(ca->usage_base->buckets_unavailable);
1253
1254                 for (i = 0; i < BCH_DATA_NR; i++) {
1255                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1256                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1257                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1258                 }
1259         }
1260
1261         percpu_up_read(&c->mark_lock);
1262
1263         for (i = 0; i < 2; i++) {
1264                 struct jset_entry_clock *clock =
1265                         container_of(jset_entry_init(end, sizeof(*clock)),
1266                                      struct jset_entry_clock, entry);
1267
1268                 clock->entry.type = BCH_JSET_ENTRY_clock;
1269                 clock->rw       = i;
1270                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1271         }
1272 }
1273
1274 void bch2_fs_mark_clean(struct bch_fs *c)
1275 {
1276         struct bch_sb_field_clean *sb_clean;
1277         struct jset_entry *entry;
1278         unsigned u64s;
1279         int ret;
1280
1281         mutex_lock(&c->sb_lock);
1282         if (BCH_SB_CLEAN(c->disk_sb.sb))
1283                 goto out;
1284
1285         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1286
1287         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1288         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1289         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1290         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1291
1292         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1293
1294         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1295         if (!sb_clean) {
1296                 bch_err(c, "error resizing superblock while setting filesystem clean");
1297                 goto out;
1298         }
1299
1300         sb_clean->flags         = 0;
1301         sb_clean->journal_seq   = cpu_to_le64(atomic64_read(&c->journal.seq));
1302
1303         /* Trying to catch outstanding bug: */
1304         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1305
1306         entry = sb_clean->start;
1307         bch2_journal_super_entries_add_common(c, &entry, 0);
1308         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1309         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1310
1311         memset(entry, 0,
1312                vstruct_end(&sb_clean->field) - (void *) entry);
1313
1314         /*
1315          * this should be in the write path, and we should be validating every
1316          * superblock section:
1317          */
1318         ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1319         if (ret) {
1320                 bch_err(c, "error writing marking filesystem clean: validate error");
1321                 goto out;
1322         }
1323
1324         bch2_write_super(c);
1325 out:
1326         mutex_unlock(&c->sb_lock);
1327 }
1328
1329 static int bch2_sb_clean_validate(struct bch_sb *sb,
1330                                   struct bch_sb_field *f,
1331                                   struct printbuf *err)
1332 {
1333         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1334
1335         if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1336                 pr_buf(err, "wrong size (got %zu should be %zu)",
1337                        vstruct_bytes(&clean->field), sizeof(*clean));
1338                 return -EINVAL;
1339         }
1340
1341         return 0;
1342 }
1343
1344 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1345                                   struct bch_sb_field *f)
1346 {
1347         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1348         struct jset_entry *entry;
1349
1350         pr_buf(out, "flags:          %x",       le32_to_cpu(clean->flags));
1351         pr_newline(out);
1352         pr_buf(out, "journal_seq:    %llu",     le64_to_cpu(clean->journal_seq));
1353         pr_newline(out);
1354
1355         for (entry = clean->start;
1356              entry != vstruct_end(&clean->field);
1357              entry = vstruct_next(entry)) {
1358                 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1359                     !entry->u64s)
1360                         continue;
1361
1362                 bch2_journal_entry_to_text(out, NULL, entry);
1363                 pr_newline(out);
1364         }
1365 }
1366
1367 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1368         .validate       = bch2_sb_clean_validate,
1369         .to_text        = bch2_sb_clean_to_text,
1370 };
1371
1372 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1373 #define x(f, nr)                                        \
1374         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1375         BCH_SB_FIELDS()
1376 #undef x
1377 };
1378
1379 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1380                                   struct printbuf *err)
1381 {
1382         unsigned type = le32_to_cpu(f->type);
1383         struct printbuf field_err = PRINTBUF;
1384         int ret;
1385
1386         if (type >= BCH_SB_FIELD_NR)
1387                 return 0;
1388
1389         ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
1390         if (ret) {
1391                 pr_buf(err, "Invalid superblock section %s: %s",
1392                        bch2_sb_fields[type],
1393                        field_err.buf);
1394                 pr_newline(err);
1395                 bch2_sb_field_to_text(err, sb, f);
1396         }
1397
1398         printbuf_exit(&field_err);
1399         return ret;
1400 }
1401
1402 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1403                            struct bch_sb_field *f)
1404 {
1405         unsigned type = le32_to_cpu(f->type);
1406         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1407                 ? bch2_sb_field_ops[type] : NULL;
1408
1409         if (!out->tabstops[0])
1410                 out->tabstops[0] = 32;
1411
1412         if (ops)
1413                 pr_buf(out, "%s", bch2_sb_fields[type]);
1414         else
1415                 pr_buf(out, "(unknown field %u)", type);
1416
1417         pr_buf(out, " (size %zu):", vstruct_bytes(f));
1418         pr_newline(out);
1419
1420         if (ops && ops->to_text) {
1421                 pr_indent_push(out, 2);
1422                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1423                 pr_indent_pop(out, 2);
1424         }
1425 }
1426
1427 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1428 {
1429         unsigned i;
1430
1431         pr_buf(out, "Type:                    %u", l->layout_type);
1432         pr_newline(out);
1433
1434         pr_buf(out, "Superblock max size:     ");
1435         pr_units(out,
1436                  1 << l->sb_max_size_bits,
1437                  512 << l->sb_max_size_bits);
1438         pr_newline(out);
1439
1440         pr_buf(out, "Nr superblocks:          %u", l->nr_superblocks);
1441         pr_newline(out);
1442
1443         pr_buf(out, "Offsets:                 ");
1444         for (i = 0; i < l->nr_superblocks; i++) {
1445                 if (i)
1446                         pr_buf(out, ", ");
1447                 pr_buf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1448         }
1449         pr_newline(out);
1450 }
1451
1452 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1453                      bool print_layout, unsigned fields)
1454 {
1455         struct bch_sb_field_members *mi;
1456         struct bch_sb_field *f;
1457         u64 fields_have = 0;
1458         unsigned nr_devices = 0;
1459
1460         if (!out->tabstops[0])
1461                 out->tabstops[0] = 32;
1462
1463         mi = bch2_sb_get_members(sb);
1464         if (mi) {
1465                 struct bch_member *m;
1466
1467                 for (m = mi->members;
1468                      m < mi->members + sb->nr_devices;
1469                      m++)
1470                         nr_devices += bch2_member_exists(m);
1471         }
1472
1473         pr_buf(out, "External UUID:");
1474         pr_tab(out);
1475         pr_uuid(out, sb->user_uuid.b);
1476         pr_newline(out);
1477
1478         pr_buf(out, "Internal UUID:");
1479         pr_tab(out);
1480         pr_uuid(out, sb->uuid.b);
1481         pr_newline(out);
1482
1483         pr_buf(out, "Device index:");
1484         pr_tab(out);
1485         pr_buf(out, "%u", sb->dev_idx);
1486         pr_newline(out);
1487
1488         pr_buf(out, "Label:");
1489         pr_tab(out);
1490         pr_buf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1491         pr_newline(out);
1492
1493         pr_buf(out, "Version:");
1494         pr_tab(out);
1495         pr_buf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
1496         pr_newline(out);
1497
1498         pr_buf(out, "Oldest version on disk:");
1499         pr_tab(out);
1500         pr_buf(out, "%u", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
1501         pr_newline(out);
1502
1503         pr_buf(out, "Created:");
1504         pr_tab(out);
1505         if (sb->time_base_lo)
1506                 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1507         else
1508                 pr_buf(out, "(not set)");
1509         pr_newline(out);
1510
1511         pr_buf(out, "Sequence number:");
1512         pr_tab(out);
1513         pr_buf(out, "%llu", le64_to_cpu(sb->seq));
1514         pr_newline(out);
1515
1516         pr_buf(out, "Superblock size:");
1517         pr_tab(out);
1518         pr_buf(out, "%zu", vstruct_bytes(sb));
1519         pr_newline(out);
1520
1521         pr_buf(out, "Clean:");
1522         pr_tab(out);
1523         pr_buf(out, "%llu", BCH_SB_CLEAN(sb));
1524         pr_newline(out);
1525
1526         pr_buf(out, "Devices:");
1527         pr_tab(out);
1528         pr_buf(out, "%u", nr_devices);
1529         pr_newline(out);
1530
1531         pr_buf(out, "Sections:");
1532         vstruct_for_each(sb, f)
1533                 fields_have |= 1 << le32_to_cpu(f->type);
1534         pr_tab(out);
1535         bch2_flags_to_text(out, bch2_sb_fields, fields_have);
1536         pr_newline(out);
1537
1538         pr_buf(out, "Features:");
1539         pr_tab(out);
1540         bch2_flags_to_text(out, bch2_sb_features,
1541                            le64_to_cpu(sb->features[0]));
1542         pr_newline(out);
1543
1544         pr_buf(out, "Compat features:");
1545         pr_tab(out);
1546         bch2_flags_to_text(out, bch2_sb_compat,
1547                            le64_to_cpu(sb->compat[0]));
1548         pr_newline(out);
1549
1550         pr_newline(out);
1551         pr_buf(out, "Options:");
1552         pr_newline(out);
1553         pr_indent_push(out, 2);
1554         {
1555                 enum bch_opt_id id;
1556
1557                 for (id = 0; id < bch2_opts_nr; id++) {
1558                         const struct bch_option *opt = bch2_opt_table + id;
1559
1560                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1561                                 u64 v = bch2_opt_from_sb(sb, id);
1562
1563                                 pr_buf(out, "%s:", opt->attr.name);
1564                                 pr_tab(out);
1565                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1566                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1567                                 pr_newline(out);
1568                         }
1569                 }
1570         }
1571
1572         pr_indent_pop(out, 2);
1573
1574         if (print_layout) {
1575                 pr_newline(out);
1576                 pr_buf(out, "layout:");
1577                 pr_newline(out);
1578                 pr_indent_push(out, 2);
1579                 bch2_sb_layout_to_text(out, &sb->layout);
1580                 pr_indent_pop(out, 2);
1581         }
1582
1583         vstruct_for_each(sb, f)
1584                 if (fields & (1 << le32_to_cpu(f->type))) {
1585                         pr_newline(out);
1586                         bch2_sb_field_to_text(out, sb, f);
1587                 }
1588 }