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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "btree_update_interior.h"
   5 #include "buckets.h"
   6 #include "checksum.h"
   7 #include "disk_groups.h"
   8 #include "ec.h"
   9 #include "error.h"
  10 #include "io.h"
  11 #include "journal.h"
  12 #include "journal_io.h"
  13 #include "journal_seq_blacklist.h"
  14 #include "replicas.h"
  15 #include "quota.h"
  16 #include "super-io.h"
  17 #include "super.h"
  18 #include "vstructs.h"
  19
  20 #include <linux/backing-dev.h>
  21 #include <linux/sort.h>
  22
  23 const char * const bch2_sb_fields[] = {
  24 #define x(name, nr)     #name,
  25         BCH_SB_FIELDS()
  26 #undef x
  27         NULL
  28 };
  29
  30 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
  31                                   struct printbuf *);
  32
  33 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  34                                       enum bch_sb_field_type type)
  35 {
  36         struct bch_sb_field *f;
  37
  38         /* XXX: need locking around superblock to access optional fields */
  39
  40         vstruct_for_each(sb, f)
  41                 if (le32_to_cpu(f->type) == type)
  42                         return f;
  43         return NULL;
  44 }
  45
  46 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  47                                                    struct bch_sb_field *f,
  48                                                    unsigned u64s)
  49 {
  50         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  51         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  52
  53         BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
  54
  55         if (!f && !u64s) {
  56                 /* nothing to do: */
  57         } else if (!f) {
  58                 f = vstruct_last(sb->sb);
  59                 memset(f, 0, sizeof(u64) * u64s);
  60                 f->u64s = cpu_to_le32(u64s);
  61                 f->type = 0;
  62         } else {
  63                 void *src, *dst;
  64
  65                 src = vstruct_end(f);
  66
  67                 if (u64s) {
  68                         f->u64s = cpu_to_le32(u64s);
  69                         dst = vstruct_end(f);
  70                 } else {
  71                         dst = f;
  72                 }
  73
  74                 memmove(dst, src, vstruct_end(sb->sb) - src);
  75
  76                 if (dst > src)
  77                         memset(src, 0, dst - src);
  78         }
  79
  80         sb->sb->u64s = cpu_to_le32(sb_u64s);
  81
  82         return u64s ? f : NULL;
  83 }
  84
  85 void bch2_sb_field_delete(struct bch_sb_handle *sb,
  86                           enum bch_sb_field_type type)
  87 {
  88         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
  89
  90         if (f)
  91                 __bch2_sb_field_resize(sb, f, 0);
  92 }
  93
  94 /* Superblock realloc/free: */
  95
  96 void bch2_free_super(struct bch_sb_handle *sb)
  97 {
  98         if (sb->bio)
  99                 bio_put(sb->bio);
 100         if (!IS_ERR_OR_NULL(sb->bdev))
 101                 blkdev_put(sb->bdev, sb->mode);
 102
 103         kfree(sb->sb);
 104         memset(sb, 0, sizeof(*sb));
 105 }
 106
 107 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 108 {
 109         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 110         size_t new_buffer_size;
 111         struct bch_sb *new_sb;
 112         struct bio *bio;
 113
 114         if (sb->bdev)
 115                 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
 116
 117         new_buffer_size = roundup_pow_of_two(new_bytes);
 118
 119         if (sb->sb && sb->buffer_size >= new_buffer_size)
 120                 return 0;
 121
 122         if (sb->have_layout) {
 123                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 124
 125                 if (new_bytes > max_bytes) {
 126                         char buf[BDEVNAME_SIZE];
 127
 128                         pr_err("%s: superblock too big: want %zu but have %llu",
 129                                bdevname(sb->bdev, buf), new_bytes, max_bytes);
 130                         return -ENOSPC;
 131                 }
 132         }
 133
 134         if (sb->buffer_size >= new_buffer_size && sb->sb)
 135                 return 0;
 136
 137         if (dynamic_fault("bcachefs:add:super_realloc"))
 138                 return -ENOMEM;
 139
 140         if (sb->have_bio) {
 141                 bio = bio_kmalloc(GFP_KERNEL,
 142                         DIV_ROUND_UP(new_buffer_size, PAGE_SIZE));
 143                 if (!bio)
 144                         return -ENOMEM;
 145
 146                 if (sb->bio)
 147                         bio_put(sb->bio);
 148                 sb->bio = bio;
 149         }
 150
 151         new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 152         if (!new_sb)
 153                 return -ENOMEM;
 154
 155         sb->sb = new_sb;
 156         sb->buffer_size = new_buffer_size;
 157
 158         return 0;
 159 }
 160
 161 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 162                                           enum bch_sb_field_type type,
 163                                           unsigned u64s)
 164 {
 165         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 166         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 167         ssize_t d = -old_u64s + u64s;
 168
 169         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 170                 return NULL;
 171
 172         if (sb->fs_sb) {
 173                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 174                 struct bch_dev *ca;
 175                 unsigned i;
 176
 177                 lockdep_assert_held(&c->sb_lock);
 178
 179                 /* XXX: we're not checking that offline device have enough space */
 180
 181                 for_each_online_member(ca, c, i) {
 182                         struct bch_sb_handle *sb = &ca->disk_sb;
 183
 184                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 185                                 percpu_ref_put(&ca->ref);
 186                                 return NULL;
 187                         }
 188                 }
 189         }
 190
 191         f = bch2_sb_field_get(sb->sb, type);
 192         f = __bch2_sb_field_resize(sb, f, u64s);
 193         if (f)
 194                 f->type = cpu_to_le32(type);
 195         return f;
 196 }
 197
 198 /* Superblock validate: */
 199
 200 static inline void __bch2_sb_layout_size_assert(void)
 201 {
 202         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 203 }
 204
 205 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
 206 {
 207         u64 offset, prev_offset, max_sectors;
 208         unsigned i;
 209
 210         if (uuid_le_cmp(layout->magic, BCACHE_MAGIC)) {
 211                 pr_buf(out, "Not a bcachefs superblock layout");
 212                 return -EINVAL;
 213         }
 214
 215         if (layout->layout_type != 0) {
 216                 pr_buf(out, "Invalid superblock layout type %u",
 217                        layout->layout_type);
 218                 return -EINVAL;
 219         }
 220
 221         if (!layout->nr_superblocks) {
 222                 pr_buf(out, "Invalid superblock layout: no superblocks");
 223                 return -EINVAL;
 224         }
 225
 226         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
 227                 pr_buf(out, "Invalid superblock layout: too many superblocks");
 228                 return -EINVAL;
 229         }
 230
 231         max_sectors = 1 << layout->sb_max_size_bits;
 232
 233         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 234
 235         for (i = 1; i < layout->nr_superblocks; i++) {
 236                 offset = le64_to_cpu(layout->sb_offset[i]);
 237
 238                 if (offset < prev_offset + max_sectors) {
 239                         pr_buf(out, "Invalid superblock layout: superblocks overlap\n"
 240                                "  (sb %u ends at %llu next starts at %llu",
 241                                i - 1, prev_offset + max_sectors, offset);
 242                         return -EINVAL;
 243                 }
 244                 prev_offset = offset;
 245         }
 246
 247         return 0;
 248 }
 249
 250 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out)
 251 {
 252         struct bch_sb *sb = disk_sb->sb;
 253         struct bch_sb_field *f;
 254         struct bch_sb_field_members *mi;
 255         u32 version, version_min;
 256         u16 block_size;
 257         int ret;
 258
 259         version         = le16_to_cpu(sb->version);
 260         version_min     = version >= bcachefs_metadata_version_new_versioning
 261                 ? le16_to_cpu(sb->version_min)
 262                 : version;
 263
 264         if (version    >= bcachefs_metadata_version_max) {
 265                 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
 266                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 267                 return -EINVAL;
 268         }
 269
 270         if (version_min < bcachefs_metadata_version_min) {
 271                 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
 272                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 273                 return -EINVAL;
 274         }
 275
 276         if (version_min > version) {
 277                 pr_buf(out, "Bad minimum version %u, greater than version field %u",
 278                        version_min, version);
 279                 return -EINVAL;
 280         }
 281
 282         if (sb->features[1] ||
 283             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
 284                 pr_buf(out, "Filesystem has incompatible features");
 285                 return -EINVAL;
 286         }
 287
 288         block_size = le16_to_cpu(sb->block_size);
 289
 290         if (block_size > PAGE_SECTORS) {
 291                 pr_buf(out, "Block size too big (got %u, max %u)",
 292                        block_size, PAGE_SECTORS);
 293                 return -EINVAL;
 294         }
 295
 296         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
 297                 pr_buf(out, "Bad user UUID (got zeroes)");
 298                 return -EINVAL;
 299         }
 300
 301         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
 302                 pr_buf(out, "Bad intenal UUID (got zeroes)");
 303                 return -EINVAL;
 304         }
 305
 306         if (!sb->nr_devices ||
 307             sb->nr_devices > BCH_SB_MEMBERS_MAX) {
 308                 pr_buf(out, "Bad number of member devices %u (max %u)",
 309                        sb->nr_devices, BCH_SB_MEMBERS_MAX);
 310                 return -EINVAL;
 311         }
 312
 313         if (sb->dev_idx >= sb->nr_devices) {
 314                 pr_buf(out, "Bad dev_idx (got %u, nr_devices %u)",
 315                        sb->dev_idx, sb->nr_devices);
 316                 return -EINVAL;
 317         }
 318
 319         if (!sb->time_precision ||
 320             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
 321                 pr_buf(out, "Invalid time precision: %u (min 1, max %lu)",
 322                        le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
 323                 return -EINVAL;
 324         }
 325
 326         /* validate layout */
 327         ret = validate_sb_layout(&sb->layout, out);
 328         if (ret)
 329                 return ret;
 330
 331         vstruct_for_each(sb, f) {
 332                 if (!f->u64s) {
 333                         pr_buf(out, "Invalid superblock: optional with size 0 (type %u)",
 334                                le32_to_cpu(f->type));
 335                         return -EINVAL;
 336                 }
 337
 338                 if (vstruct_next(f) > vstruct_last(sb)) {
 339                         pr_buf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
 340                                le32_to_cpu(f->type));
 341                         return -EINVAL;
 342                 }
 343         }
 344
 345         /* members must be validated first: */
 346         mi = bch2_sb_get_members(sb);
 347         if (!mi) {
 348                 pr_buf(out, "Invalid superblock: member info area missing");
 349                 return -EINVAL;
 350         }
 351
 352         ret = bch2_sb_field_validate(sb, &mi->field, out);
 353         if (ret)
 354                 return ret;
 355
 356         vstruct_for_each(sb, f) {
 357                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 358                         continue;
 359
 360                 ret = bch2_sb_field_validate(sb, f, out);
 361                 if (ret)
 362                         return ret;
 363         }
 364
 365         return 0;
 366 }
 367
 368 /* device open: */
 369
 370 static void bch2_sb_update(struct bch_fs *c)
 371 {
 372         struct bch_sb *src = c->disk_sb.sb;
 373         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 374         struct bch_dev *ca;
 375         unsigned i;
 376
 377         lockdep_assert_held(&c->sb_lock);
 378
 379         c->sb.uuid              = src->uuid;
 380         c->sb.user_uuid         = src->user_uuid;
 381         c->sb.version           = le16_to_cpu(src->version);
 382         c->sb.version_min       = le16_to_cpu(src->version_min);
 383         c->sb.nr_devices        = src->nr_devices;
 384         c->sb.clean             = BCH_SB_CLEAN(src);
 385         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 386
 387         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 388         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 389
 390         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 391         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 392                                           c->sb.nsec_per_time_unit);
 393         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 394
 395         c->sb.features          = le64_to_cpu(src->features[0]);
 396         c->sb.compat            = le64_to_cpu(src->compat[0]);
 397
 398         for_each_member_device(ca, c, i)
 399                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 400 }
 401
 402 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 403 {
 404         struct bch_sb_field *src_f, *dst_f;
 405         struct bch_sb *dst = dst_handle->sb;
 406         unsigned i;
 407
 408         dst->version            = src->version;
 409         dst->version_min        = src->version_min;
 410         dst->seq                = src->seq;
 411         dst->uuid               = src->uuid;
 412         dst->user_uuid          = src->user_uuid;
 413         memcpy(dst->label,      src->label, sizeof(dst->label));
 414
 415         dst->block_size         = src->block_size;
 416         dst->nr_devices         = src->nr_devices;
 417
 418         dst->time_base_lo       = src->time_base_lo;
 419         dst->time_base_hi       = src->time_base_hi;
 420         dst->time_precision     = src->time_precision;
 421
 422         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 423         memcpy(dst->features,   src->features,  sizeof(dst->features));
 424         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 425
 426         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 427                 if (i == BCH_SB_FIELD_journal)
 428                         continue;
 429
 430                 src_f = bch2_sb_field_get(src, i);
 431                 dst_f = bch2_sb_field_get(dst, i);
 432                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 433                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 434
 435                 if (src_f)
 436                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 437         }
 438 }
 439
 440 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 441 {
 442         struct bch_sb_field_journal *journal_buckets =
 443                 bch2_sb_get_journal(src);
 444         unsigned journal_u64s = journal_buckets
 445                 ? le32_to_cpu(journal_buckets->field.u64s)
 446                 : 0;
 447         int ret;
 448
 449         lockdep_assert_held(&c->sb_lock);
 450
 451         ret = bch2_sb_realloc(&c->disk_sb,
 452                               le32_to_cpu(src->u64s) - journal_u64s);
 453         if (ret)
 454                 return ret;
 455
 456         __copy_super(&c->disk_sb, src);
 457
 458         ret = bch2_sb_replicas_to_cpu_replicas(c);
 459         if (ret)
 460                 return ret;
 461
 462         ret = bch2_sb_disk_groups_to_cpu(c);
 463         if (ret)
 464                 return ret;
 465
 466         bch2_sb_update(c);
 467         return 0;
 468 }
 469
 470 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 471 {
 472         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 473         struct bch_sb_field_journal *journal_buckets =
 474                 bch2_sb_get_journal(dst);
 475         unsigned journal_u64s = journal_buckets
 476                 ? le32_to_cpu(journal_buckets->field.u64s)
 477                 : 0;
 478         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 479         int ret;
 480
 481         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 482         if (ret)
 483                 return ret;
 484
 485         __copy_super(&ca->disk_sb, src);
 486         return 0;
 487 }
 488
 489 /* read superblock: */
 490
 491 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 492 {
 493         struct bch_csum csum;
 494         u32 version, version_min;
 495         size_t bytes;
 496         int ret;
 497 reread:
 498         bio_reset(sb->bio);
 499         bio_set_dev(sb->bio, sb->bdev);
 500         sb->bio->bi_iter.bi_sector = offset;
 501         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 502         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 503
 504         ret = submit_bio_wait(sb->bio);
 505         if (ret) {
 506                 pr_buf(err, "IO error: %i", ret);
 507                 return ret;
 508         }
 509
 510         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC)) {
 511                 pr_buf(err, "Not a bcachefs superblock");
 512                 return -EINVAL;
 513         }
 514
 515         version         = le16_to_cpu(sb->sb->version);
 516         version_min     = version >= bcachefs_metadata_version_new_versioning
 517                 ? le16_to_cpu(sb->sb->version_min)
 518                 : version;
 519
 520         if (version    >= bcachefs_metadata_version_max) {
 521                 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
 522                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 523                 return -EINVAL;
 524         }
 525
 526         if (version_min < bcachefs_metadata_version_min) {
 527                 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
 528                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 529                 return -EINVAL;
 530         }
 531
 532         bytes = vstruct_bytes(sb->sb);
 533
 534         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 535                 pr_buf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 536                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 537                 return -EINVAL;
 538         }
 539
 540         if (bytes > sb->buffer_size) {
 541                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 542                         return -ENOMEM;
 543                 goto reread;
 544         }
 545
 546         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
 547                 pr_buf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 548                 return -EINVAL;
 549         }
 550
 551         /* XXX: verify MACs */
 552         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 553                             null_nonce(), sb->sb);
 554
 555         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 556                 pr_buf(err, "bad checksum");
 557                 return -EINVAL;
 558         }
 559
 560         sb->seq = le64_to_cpu(sb->sb->seq);
 561
 562         return 0;
 563 }
 564
 565 int bch2_read_super(const char *path, struct bch_opts *opts,
 566                     struct bch_sb_handle *sb)
 567 {
 568         u64 offset = opt_get(*opts, sb);
 569         struct bch_sb_layout layout;
 570         char *_err;
 571         struct printbuf err;
 572         __le64 *i;
 573         int ret;
 574
 575         _err = kmalloc(4096, GFP_KERNEL);
 576         if (!_err)
 577                 return -ENOMEM;
 578         err = _PBUF(_err, 4096);
 579
 580         pr_verbose_init(*opts, "");
 581
 582         memset(sb, 0, sizeof(*sb));
 583         sb->mode        = FMODE_READ;
 584         sb->have_bio    = true;
 585
 586         if (!opt_get(*opts, noexcl))
 587                 sb->mode |= FMODE_EXCL;
 588
 589         if (!opt_get(*opts, nochanges))
 590                 sb->mode |= FMODE_WRITE;
 591
 592         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 593         if (IS_ERR(sb->bdev) &&
 594             PTR_ERR(sb->bdev) == -EACCES &&
 595             opt_get(*opts, read_only)) {
 596                 sb->mode &= ~FMODE_WRITE;
 597
 598                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 599                 if (!IS_ERR(sb->bdev))
 600                         opt_set(*opts, nochanges, true);
 601         }
 602
 603         if (IS_ERR(sb->bdev)) {
 604                 ret = PTR_ERR(sb->bdev);
 605                 goto out;
 606         }
 607
 608         ret = bch2_sb_realloc(sb, 0);
 609         if (ret) {
 610                 pr_buf(&err, "error allocating memory for superblock");
 611                 goto err;
 612         }
 613
 614         if (bch2_fs_init_fault("read_super")) {
 615                 pr_buf(&err, "dynamic fault");
 616                 ret = -EFAULT;
 617                 goto err;
 618         }
 619
 620         ret = read_one_super(sb, offset, &err);
 621         if (!ret)
 622                 goto got_super;
 623
 624         if (opt_defined(*opts, sb))
 625                 goto err;
 626
 627         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 628                path, _err);
 629         err = _PBUF(_err, 4096);
 630
 631         /*
 632          * Error reading primary superblock - read location of backup
 633          * superblocks:
 634          */
 635         bio_reset(sb->bio);
 636         bio_set_dev(sb->bio, sb->bdev);
 637         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 638         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 639         /*
 640          * use sb buffer to read layout, since sb buffer is page aligned but
 641          * layout won't be:
 642          */
 643         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 644
 645         ret = submit_bio_wait(sb->bio);
 646         if (ret) {
 647                 pr_buf(&err, "IO error: %i", ret);
 648                 goto err;
 649         }
 650
 651         memcpy(&layout, sb->sb, sizeof(layout));
 652         ret = validate_sb_layout(&layout, &err);
 653         if (ret)
 654                 goto err;
 655
 656         for (i = layout.sb_offset;
 657              i < layout.sb_offset + layout.nr_superblocks; i++) {
 658                 offset = le64_to_cpu(*i);
 659
 660                 if (offset == opt_get(*opts, sb))
 661                         continue;
 662
 663                 ret = read_one_super(sb, offset, &err);
 664                 if (!ret)
 665                         goto got_super;
 666         }
 667
 668         goto err;
 669
 670 got_super:
 671         if (le16_to_cpu(sb->sb->block_size) << 9 <
 672             bdev_logical_block_size(sb->bdev)) {
 673                 pr_buf(&err, "block size (%u) smaller than device block size (%u)",
 674                        le16_to_cpu(sb->sb->block_size) << 9,
 675                        bdev_logical_block_size(sb->bdev));
 676                 ret = -EINVAL;
 677                 goto err;
 678         }
 679
 680         ret = 0;
 681         sb->have_layout = true;
 682
 683         ret = bch2_sb_validate(sb, &err);
 684         if (ret) {
 685                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 686                        path, _err);
 687                 goto err_no_print;
 688         }
 689 out:
 690         pr_verbose_init(*opts, "ret %i", ret);
 691         kfree(_err);
 692         return ret;
 693 err:
 694         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 695                path, _err);
 696 err_no_print:
 697         bch2_free_super(sb);
 698         goto out;
 699 }
 700
 701 /* write superblock: */
 702
 703 static void write_super_endio(struct bio *bio)
 704 {
 705         struct bch_dev *ca = bio->bi_private;
 706
 707         /* XXX: return errors directly */
 708
 709         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 710                                bch2_blk_status_to_str(bio->bi_status)))
 711                 ca->sb_write_error = 1;
 712
 713         closure_put(&ca->fs->sb_write);
 714         percpu_ref_put(&ca->io_ref);
 715 }
 716
 717 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 718 {
 719         struct bch_sb *sb = ca->disk_sb.sb;
 720         struct bio *bio = ca->disk_sb.bio;
 721
 722         bio_reset(bio);
 723         bio_set_dev(bio, ca->disk_sb.bdev);
 724         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 725         bio->bi_end_io          = write_super_endio;
 726         bio->bi_private         = ca;
 727         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 728         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 729
 730         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 731                      bio_sectors(bio));
 732
 733         percpu_ref_get(&ca->io_ref);
 734         closure_bio_submit(bio, &c->sb_write);
 735 }
 736
 737 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 738 {
 739         struct bch_sb *sb = ca->disk_sb.sb;
 740         struct bio *bio = ca->disk_sb.bio;
 741
 742         sb->offset = sb->layout.sb_offset[idx];
 743
 744         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 745         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 746                                 null_nonce(), sb);
 747
 748         bio_reset(bio);
 749         bio_set_dev(bio, ca->disk_sb.bdev);
 750         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 751         bio->bi_end_io          = write_super_endio;
 752         bio->bi_private         = ca;
 753         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 754         bch2_bio_map(bio, sb,
 755                      roundup((size_t) vstruct_bytes(sb),
 756                              bdev_logical_block_size(ca->disk_sb.bdev)));
 757
 758         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 759                      bio_sectors(bio));
 760
 761         percpu_ref_get(&ca->io_ref);
 762         closure_bio_submit(bio, &c->sb_write);
 763 }
 764
 765 int bch2_write_super(struct bch_fs *c)
 766 {
 767         struct closure *cl = &c->sb_write;
 768         struct bch_dev *ca;
 769         unsigned i, sb = 0, nr_wrote;
 770         struct bch_devs_mask sb_written;
 771         bool wrote, can_mount_without_written, can_mount_with_written;
 772         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 773         int ret = 0;
 774
 775         if (c->opts.very_degraded)
 776                 degraded_flags |= BCH_FORCE_IF_LOST;
 777
 778         lockdep_assert_held(&c->sb_lock);
 779
 780         closure_init_stack(cl);
 781         memset(&sb_written, 0, sizeof(sb_written));
 782
 783         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 784
 785         if (test_bit(BCH_FS_ERROR, &c->flags))
 786                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 787         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 788                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 789
 790         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 791
 792         for_each_online_member(ca, c, i)
 793                 bch2_sb_from_fs(c, ca);
 794
 795         for_each_online_member(ca, c, i) {
 796                 struct printbuf buf = { NULL, NULL };
 797
 798                 ret = bch2_sb_validate(&ca->disk_sb, &buf);
 799                 if (ret) {
 800                         char *_buf = kmalloc(4096, GFP_NOFS);
 801                         if (_buf) {
 802                                 buf = _PBUF(_buf, 4096);
 803                                 bch2_sb_validate(&ca->disk_sb, &buf);
 804                         }
 805
 806                         bch2_fs_inconsistent(c, "sb invalid before write: %s", _buf);
 807                         kfree(_buf);
 808                         percpu_ref_put(&ca->io_ref);
 809                         goto out;
 810                 }
 811         }
 812
 813         if (c->opts.nochanges)
 814                 goto out;
 815
 816         for_each_online_member(ca, c, i) {
 817                 __set_bit(ca->dev_idx, sb_written.d);
 818                 ca->sb_write_error = 0;
 819         }
 820
 821         for_each_online_member(ca, c, i)
 822                 read_back_super(c, ca);
 823         closure_sync(cl);
 824
 825         for_each_online_member(ca, c, i) {
 826                 if (ca->sb_write_error)
 827                         continue;
 828
 829                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 830                         bch2_fs_fatal_error(c,
 831                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 832                                 le64_to_cpu(ca->sb_read_scratch->seq),
 833                                 ca->disk_sb.seq);
 834                         percpu_ref_put(&ca->io_ref);
 835                         ret = -EROFS;
 836                         goto out;
 837                 }
 838
 839                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 840                         bch2_fs_fatal_error(c,
 841                                 "Superblock modified by another process (seq %llu expected %llu)",
 842                                 le64_to_cpu(ca->sb_read_scratch->seq),
 843                                 ca->disk_sb.seq);
 844                         percpu_ref_put(&ca->io_ref);
 845                         ret = -EROFS;
 846                         goto out;
 847                 }
 848         }
 849
 850         do {
 851                 wrote = false;
 852                 for_each_online_member(ca, c, i)
 853                         if (!ca->sb_write_error &&
 854                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 855                                 write_one_super(c, ca, sb);
 856                                 wrote = true;
 857                         }
 858                 closure_sync(cl);
 859                 sb++;
 860         } while (wrote);
 861
 862         for_each_online_member(ca, c, i) {
 863                 if (ca->sb_write_error)
 864                         __clear_bit(ca->dev_idx, sb_written.d);
 865                 else
 866                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 867         }
 868
 869         nr_wrote = dev_mask_nr(&sb_written);
 870
 871         can_mount_with_written =
 872                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 873
 874         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 875                 sb_written.d[i] = ~sb_written.d[i];
 876
 877         can_mount_without_written =
 878                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 879
 880         /*
 881          * If we would be able to mount _without_ the devices we successfully
 882          * wrote superblocks to, we weren't able to write to enough devices:
 883          *
 884          * Exception: if we can mount without the successes because we haven't
 885          * written anything (new filesystem), we continue if we'd be able to
 886          * mount with the devices we did successfully write to:
 887          */
 888         if (bch2_fs_fatal_err_on(!nr_wrote ||
 889                                  !can_mount_with_written ||
 890                                  (can_mount_without_written &&
 891                                   !can_mount_with_written), c,
 892                 "Unable to write superblock to sufficient devices (from %ps)",
 893                 (void *) _RET_IP_))
 894                 ret = -1;
 895 out:
 896         /* Make new options visible after they're persistent: */
 897         bch2_sb_update(c);
 898         return ret;
 899 }
 900
 901 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 902 {
 903         mutex_lock(&c->sb_lock);
 904         if (!(c->sb.features & (1ULL << feat))) {
 905                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 906
 907                 bch2_write_super(c);
 908         }
 909         mutex_unlock(&c->sb_lock);
 910 }
 911
 912 /* BCH_SB_FIELD_journal: */
 913
 914 static int u64_cmp(const void *_l, const void *_r)
 915 {
 916         u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
 917
 918         return l < r ? -1 : l > r ? 1 : 0;
 919 }
 920
 921 static int bch2_sb_journal_validate(struct bch_sb *sb,
 922                                     struct bch_sb_field *f,
 923                                     struct printbuf *err)
 924 {
 925         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 926         struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
 927         int ret = -EINVAL;
 928         unsigned nr;
 929         unsigned i;
 930         u64 *b;
 931
 932         nr = bch2_nr_journal_buckets(journal);
 933         if (!nr)
 934                 return 0;
 935
 936         b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
 937         if (!b)
 938                 return -ENOMEM;
 939
 940         for (i = 0; i < nr; i++)
 941                 b[i] = le64_to_cpu(journal->buckets[i]);
 942
 943         sort(b, nr, sizeof(u64), u64_cmp, NULL);
 944
 945         if (!b[0]) {
 946                 pr_buf(err, "journal bucket at sector 0");
 947                 goto err;
 948         }
 949
 950         if (b[0] < le16_to_cpu(m->first_bucket)) {
 951                 pr_buf(err, "journal bucket %llu before first bucket %u",
 952                        b[0], le16_to_cpu(m->first_bucket));
 953                 goto err;
 954         }
 955
 956         if (b[nr - 1] >= le64_to_cpu(m->nbuckets)) {
 957                 pr_buf(err, "journal bucket %llu past end of device (nbuckets %llu)",
 958                        b[nr - 1], le64_to_cpu(m->nbuckets));
 959                 goto err;
 960         }
 961
 962         for (i = 0; i + 1 < nr; i++)
 963                 if (b[i] == b[i + 1]) {
 964                         pr_buf(err, "duplicate journal buckets %llu", b[i]);
 965                         goto err;
 966                 }
 967
 968         ret = 0;
 969 err:
 970         kfree(b);
 971         return ret;
 972 }
 973
 974 static void bch2_sb_journal_to_text(struct printbuf *out, struct bch_sb *sb,
 975                                     struct bch_sb_field *f)
 976 {
 977         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 978         unsigned i, nr = bch2_nr_journal_buckets(journal);
 979
 980         pr_buf(out, "Buckets: ");
 981         for (i = 0; i < nr; i++)
 982                 pr_buf(out, " %llu", le64_to_cpu(journal->buckets[i]));
 983         pr_newline(out);
 984 }
 985
 986 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
 987         .validate       = bch2_sb_journal_validate,
 988         .to_text        = bch2_sb_journal_to_text,
 989 };
 990
 991 /* BCH_SB_FIELD_members: */
 992
 993 static int bch2_sb_members_validate(struct bch_sb *sb,
 994                                     struct bch_sb_field *f,
 995                                     struct printbuf *err)
 996 {
 997         struct bch_sb_field_members *mi = field_to_type(f, members);
 998         unsigned i;
 999
1000         if ((void *) (mi->members + sb->nr_devices) >
1001             vstruct_end(&mi->field)) {
1002                 pr_buf(err, "too many devices for section size");
1003                 return -EINVAL;
1004         }
1005
1006         for (i = 0; i < sb->nr_devices; i++) {
1007                 struct bch_member *m = mi->members + i;
1008
1009                 if (!bch2_member_exists(m))
1010                         continue;
1011
1012                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
1013                         pr_buf(err, "device %u: too many buckets (got %llu, max %lu)",
1014                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
1015                         return -EINVAL;
1016                 }
1017
1018                 if (le64_to_cpu(m->nbuckets) -
1019                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
1020                         pr_buf(err, "device %u: not enough buckets (got %llu, max %u)",
1021                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
1022                         return -EINVAL;
1023                 }
1024
1025                 if (le16_to_cpu(m->bucket_size) <
1026                     le16_to_cpu(sb->block_size)) {
1027                         pr_buf(err, "device %u: bucket size %u smaller than block size %u",
1028                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
1029                         return -EINVAL;
1030                 }
1031
1032                 if (le16_to_cpu(m->bucket_size) <
1033                     BCH_SB_BTREE_NODE_SIZE(sb)) {
1034                         pr_buf(err, "device %u: bucket size %u smaller than btree node size %llu",
1035                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
1036                         return -EINVAL;
1037                 }
1038         }
1039
1040         return 0;
1041 }
1042
1043 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
1044                                     struct bch_sb_field *f)
1045 {
1046         struct bch_sb_field_members *mi = field_to_type(f, members);
1047         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
1048         unsigned i;
1049
1050         for (i = 0; i < sb->nr_devices; i++) {
1051                 struct bch_member *m = mi->members + i;
1052                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
1053                 u64 bucket_size = le16_to_cpu(m->bucket_size);
1054                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1055
1056                 if (!bch2_member_exists(m))
1057                         continue;
1058
1059                 pr_buf(out, "Device:                  %u", i);
1060                 pr_newline(out);
1061
1062                 pr_indent_push(out, 2);
1063
1064                 pr_buf(out, "UUID:                  ");
1065                 pr_uuid(out, m->uuid.b);
1066                 pr_newline(out);
1067
1068                 pr_buf(out, "Size:                  ");
1069                 pr_units(out, device_size, device_size << 9);
1070                 pr_newline(out);
1071
1072                 pr_buf(out, "Bucket size:           ");
1073                 pr_units(out, bucket_size, bucket_size << 9);
1074                 pr_newline(out);
1075
1076                 pr_buf(out, "First bucket:          %u",
1077                        le16_to_cpu(m->first_bucket));
1078                 pr_newline(out);
1079
1080                 pr_buf(out, "Buckets:               %llu",
1081                        le64_to_cpu(m->nbuckets));
1082                 pr_newline(out);
1083
1084                 pr_buf(out, "Last mount:            ");
1085                 if (m->last_mount)
1086                         pr_time(out, le64_to_cpu(m->last_mount));
1087                 else
1088                         pr_buf(out, "(never)");
1089                 pr_newline(out);
1090
1091                 pr_buf(out, "State:                 %s",
1092                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1093                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1094                        : "unknown");
1095                 pr_newline(out);
1096
1097                 pr_buf(out, "Group:                 ");
1098                 if (BCH_MEMBER_GROUP(m)) {
1099                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1100
1101                         if (idx < disk_groups_nr(gi))
1102                                 pr_buf(out, "%s (%u)",
1103                                        gi->entries[idx].label, idx);
1104                         else
1105                                 pr_buf(out, "(bad disk labels section)");
1106                 } else {
1107                         pr_buf(out, "(none)");
1108                 }
1109                 pr_newline(out);
1110
1111                 pr_buf(out, "Data allowed:          ");
1112                 if (BCH_MEMBER_DATA_ALLOWED(m))
1113                         bch2_flags_to_text(out, bch2_data_types,
1114                                            BCH_MEMBER_DATA_ALLOWED(m));
1115                 else
1116                         pr_buf(out, "(none)");
1117                 pr_newline(out);
1118
1119                 pr_buf(out, "Has data:              ");
1120                 if (data_have)
1121                         bch2_flags_to_text(out, bch2_data_types, data_have);
1122                 else
1123                         pr_buf(out, "(none)");
1124                 pr_newline(out);
1125
1126                 pr_buf(out, "Discard:               %llu",
1127                        BCH_MEMBER_DISCARD(m));
1128                 pr_newline(out);
1129
1130                 pr_indent_pop(out, 2);
1131         }
1132 }
1133
1134 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1135         .validate       = bch2_sb_members_validate,
1136         .to_text        = bch2_sb_members_to_text,
1137 };
1138
1139 /* BCH_SB_FIELD_crypt: */
1140
1141 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1142                                   struct bch_sb_field *f,
1143                                   struct printbuf *err)
1144 {
1145         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1146
1147         if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1148                 pr_buf(err, "wrong size (got %llu should be %zu)",
1149                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1150                 return -EINVAL;
1151         }
1152
1153         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1154                 pr_buf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1155                 return -EINVAL;
1156         }
1157
1158         return 0;
1159 }
1160
1161 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1162                                   struct bch_sb_field *f)
1163 {
1164         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1165
1166         pr_buf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1167         pr_newline(out);
1168         pr_buf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1169         pr_newline(out);
1170         pr_buf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1171         pr_newline(out);
1172         pr_buf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1173         pr_newline(out);
1174 }
1175
1176 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1177         .validate       = bch2_sb_crypt_validate,
1178         .to_text        = bch2_sb_crypt_to_text,
1179 };
1180
1181 /* BCH_SB_FIELD_clean: */
1182
1183 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1184 {
1185         struct jset_entry *entry;
1186         int ret;
1187
1188         for (entry = clean->start;
1189              entry < (struct jset_entry *) vstruct_end(&clean->field);
1190              entry = vstruct_next(entry)) {
1191                 ret = bch2_journal_entry_validate(c, "superblock", entry,
1192                                                   le16_to_cpu(c->disk_sb.sb->version),
1193                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1194                                                   write);
1195                 if (ret)
1196                         return ret;
1197         }
1198
1199         return 0;
1200 }
1201
1202 int bch2_fs_mark_dirty(struct bch_fs *c)
1203 {
1204         int ret;
1205
1206         /*
1207          * Unconditionally write superblock, to verify it hasn't changed before
1208          * we go rw:
1209          */
1210
1211         mutex_lock(&c->sb_lock);
1212         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1213         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1214         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1215         ret = bch2_write_super(c);
1216         mutex_unlock(&c->sb_lock);
1217
1218         return ret;
1219 }
1220
1221 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1222 {
1223         struct jset_entry *entry = *end;
1224         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1225
1226         memset(entry, 0, u64s * sizeof(u64));
1227         /*
1228          * The u64s field counts from the start of data, ignoring the shared
1229          * fields.
1230          */
1231         entry->u64s = cpu_to_le16(u64s - 1);
1232
1233         *end = vstruct_next(*end);
1234         return entry;
1235 }
1236
1237 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1238                                            struct jset_entry **end,
1239                                            u64 journal_seq)
1240 {
1241         struct bch_dev *ca;
1242         unsigned i, dev;
1243
1244         percpu_down_read(&c->mark_lock);
1245
1246         if (!journal_seq) {
1247                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1248                         bch2_fs_usage_acc_to_base(c, i);
1249         } else {
1250                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1251         }
1252
1253         {
1254                 struct jset_entry_usage *u =
1255                         container_of(jset_entry_init(end, sizeof(*u)),
1256                                      struct jset_entry_usage, entry);
1257
1258                 u->entry.type   = BCH_JSET_ENTRY_usage;
1259                 u->entry.btree_id = BCH_FS_USAGE_inodes;
1260                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1261         }
1262
1263         {
1264                 struct jset_entry_usage *u =
1265                         container_of(jset_entry_init(end, sizeof(*u)),
1266                                      struct jset_entry_usage, entry);
1267
1268                 u->entry.type   = BCH_JSET_ENTRY_usage;
1269                 u->entry.btree_id = BCH_FS_USAGE_key_version;
1270                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1271         }
1272
1273         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1274                 struct jset_entry_usage *u =
1275                         container_of(jset_entry_init(end, sizeof(*u)),
1276                                      struct jset_entry_usage, entry);
1277
1278                 u->entry.type   = BCH_JSET_ENTRY_usage;
1279                 u->entry.btree_id = BCH_FS_USAGE_reserved;
1280                 u->entry.level  = i;
1281                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1282         }
1283
1284         for (i = 0; i < c->replicas.nr; i++) {
1285                 struct bch_replicas_entry *e =
1286                         cpu_replicas_entry(&c->replicas, i);
1287                 struct jset_entry_data_usage *u =
1288                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1289                                      struct jset_entry_data_usage, entry);
1290
1291                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1292                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1293                 memcpy(&u->r, e, replicas_entry_bytes(e));
1294         }
1295
1296         for_each_member_device(ca, c, dev) {
1297                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1298                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1299                 struct jset_entry_dev_usage *u =
1300                         container_of(jset_entry_init(end, b),
1301                                      struct jset_entry_dev_usage, entry);
1302
1303                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1304                 u->dev = cpu_to_le32(dev);
1305                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1306                 u->buckets_unavailable  = cpu_to_le64(ca->usage_base->buckets_unavailable);
1307
1308                 for (i = 0; i < BCH_DATA_NR; i++) {
1309                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1310                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1311                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1312                 }
1313         }
1314
1315         percpu_up_read(&c->mark_lock);
1316
1317         for (i = 0; i < 2; i++) {
1318                 struct jset_entry_clock *clock =
1319                         container_of(jset_entry_init(end, sizeof(*clock)),
1320                                      struct jset_entry_clock, entry);
1321
1322                 clock->entry.type = BCH_JSET_ENTRY_clock;
1323                 clock->rw       = i;
1324                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1325         }
1326 }
1327
1328 void bch2_fs_mark_clean(struct bch_fs *c)
1329 {
1330         struct bch_sb_field_clean *sb_clean;
1331         struct jset_entry *entry;
1332         unsigned u64s;
1333         int ret;
1334
1335         mutex_lock(&c->sb_lock);
1336         if (BCH_SB_CLEAN(c->disk_sb.sb))
1337                 goto out;
1338
1339         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1340
1341         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1342         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1343         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1344         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1345
1346         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1347
1348         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1349         if (!sb_clean) {
1350                 bch_err(c, "error resizing superblock while setting filesystem clean");
1351                 goto out;
1352         }
1353
1354         sb_clean->flags         = 0;
1355         sb_clean->journal_seq   = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1356
1357         /* Trying to catch outstanding bug: */
1358         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1359
1360         entry = sb_clean->start;
1361         bch2_journal_super_entries_add_common(c, &entry, 0);
1362         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1363         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1364
1365         memset(entry, 0,
1366                vstruct_end(&sb_clean->field) - (void *) entry);
1367
1368         /*
1369          * this should be in the write path, and we should be validating every
1370          * superblock section:
1371          */
1372         ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1373         if (ret) {
1374                 bch_err(c, "error writing marking filesystem clean: validate error");
1375                 goto out;
1376         }
1377
1378         bch2_write_super(c);
1379 out:
1380         mutex_unlock(&c->sb_lock);
1381 }
1382
1383 static int bch2_sb_clean_validate(struct bch_sb *sb,
1384                                   struct bch_sb_field *f,
1385                                   struct printbuf *err)
1386 {
1387         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1388
1389         if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1390                 pr_buf(err, "wrong size (got %llu should be %zu)",
1391                        vstruct_bytes(&clean->field), sizeof(*clean));
1392                 return -EINVAL;
1393         }
1394
1395         return 0;
1396 }
1397
1398 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1399                                   struct bch_sb_field *f)
1400 {
1401         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1402         struct jset_entry *entry;
1403
1404         pr_buf(out, "flags:          %x",       le32_to_cpu(clean->flags));
1405         pr_newline(out);
1406         pr_buf(out, "journal_seq:    %llu",     le64_to_cpu(clean->journal_seq));
1407         pr_newline(out);
1408
1409         for (entry = clean->start;
1410              entry != vstruct_end(&clean->field);
1411              entry = vstruct_next(entry)) {
1412                 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1413                     !entry->u64s)
1414                         continue;
1415
1416                 bch2_journal_entry_to_text(out, NULL, entry);
1417                 pr_newline(out);
1418         }
1419 }
1420
1421 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1422         .validate       = bch2_sb_clean_validate,
1423         .to_text        = bch2_sb_clean_to_text,
1424 };
1425
1426 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1427 #define x(f, nr)                                        \
1428         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1429         BCH_SB_FIELDS()
1430 #undef x
1431 };
1432
1433 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1434                                   struct printbuf *orig_err)
1435 {
1436         unsigned type = le32_to_cpu(f->type);
1437         struct printbuf err = *orig_err;
1438         int ret;
1439
1440         if (type >= BCH_SB_FIELD_NR)
1441                 return 0;
1442
1443         pr_buf(&err, "Invalid superblock section %s: ", bch2_sb_fields[type]);
1444
1445         ret = bch2_sb_field_ops[type]->validate(sb, f, &err);
1446         if (ret) {
1447                 pr_newline(&err);
1448                 bch2_sb_field_to_text(&err, sb, f);
1449                 *orig_err = err;
1450         }
1451
1452         return ret;
1453 }
1454
1455 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1456                            struct bch_sb_field *f)
1457 {
1458         unsigned type = le32_to_cpu(f->type);
1459         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1460                 ? bch2_sb_field_ops[type] : NULL;
1461
1462         if (ops)
1463                 pr_buf(out, "%s", bch2_sb_fields[type]);
1464         else
1465                 pr_buf(out, "(unknown field %u)", type);
1466
1467         pr_buf(out, " (size %llu):", vstruct_bytes(f));
1468         pr_newline(out);
1469
1470         if (ops && ops->to_text) {
1471                 pr_indent_push(out, 2);
1472                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1473                 pr_indent_pop(out, 2);
1474         }
1475 }
1476
1477 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1478 {
1479         unsigned i;
1480
1481         pr_buf(out, "Type:                    %u", l->layout_type);
1482         pr_newline(out);
1483
1484         pr_buf(out, "Superblock max size:     ");
1485         pr_units(out,
1486                  1 << l->sb_max_size_bits,
1487                  512 << l->sb_max_size_bits);
1488         pr_newline(out);
1489
1490         pr_buf(out, "Nr superblocks:          %u", l->nr_superblocks);
1491         pr_newline(out);
1492
1493         pr_buf(out, "Offsets:                 ");
1494         for (i = 0; i < l->nr_superblocks; i++) {
1495                 if (i)
1496                         pr_buf(out, ", ");
1497                 pr_buf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1498         }
1499         pr_newline(out);
1500 }
1501
1502 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1503                      bool print_layout, unsigned fields)
1504 {
1505         struct bch_sb_field_members *mi;
1506         struct bch_sb_field *f;
1507         u64 fields_have = 0;
1508         unsigned nr_devices = 0;
1509
1510         mi = bch2_sb_get_members(sb);
1511         if (mi) {
1512                 struct bch_member *m;
1513
1514                 for (m = mi->members;
1515                      m < mi->members + sb->nr_devices;
1516                      m++)
1517                         nr_devices += bch2_member_exists(m);
1518         }
1519
1520         pr_buf(out, "External UUID:             ");
1521         pr_uuid(out, sb->user_uuid.b);
1522         pr_newline(out);
1523
1524         pr_buf(out, "Internal UUID:             ");
1525         pr_uuid(out, sb->uuid.b);
1526         pr_newline(out);
1527
1528         pr_buf(out, "Device index:              %u", sb->dev_idx);
1529         pr_newline(out);
1530
1531         pr_buf(out, "Label:                     ");
1532         pr_buf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1533         pr_newline(out);
1534
1535         pr_buf(out, "Version:                   %u", le16_to_cpu(sb->version));
1536         pr_newline(out);
1537
1538         pr_buf(out, "Oldest version on disk:    %u", le16_to_cpu(sb->version_min));
1539         pr_newline(out);
1540
1541         pr_buf(out, "Created:                   ");
1542         if (sb->time_base_lo)
1543                 pr_time(out, le64_to_cpu(sb->time_base_lo) / NSEC_PER_SEC);
1544         else
1545                 pr_buf(out, "(not set)");
1546         pr_newline(out);
1547
1548         pr_buf(out, "Squence number:            %llu", le64_to_cpu(sb->seq));
1549         pr_newline(out);
1550
1551         pr_buf(out, "Block_size:                ");
1552         pr_units(out, le16_to_cpu(sb->block_size),
1553                  (u32) le16_to_cpu(sb->block_size) << 9);
1554         pr_newline(out);
1555
1556         pr_buf(out, "Btree node size:           ");
1557         pr_units(out, BCH_SB_BTREE_NODE_SIZE(sb),
1558                  BCH_SB_BTREE_NODE_SIZE(sb) << 9);
1559         pr_newline(out);
1560
1561         pr_buf(out, "Error action:              %s",
1562                BCH_SB_ERROR_ACTION(sb) < BCH_ON_ERROR_NR
1563                ? bch2_error_actions[BCH_SB_ERROR_ACTION(sb)]
1564                : "unknown");
1565         pr_newline(out);
1566
1567         pr_buf(out, "Clean:                     %llu", BCH_SB_CLEAN(sb));
1568         pr_newline(out);
1569
1570         pr_buf(out, "Features:                  ");
1571         bch2_flags_to_text(out, bch2_sb_features,
1572                            le64_to_cpu(sb->features[0]));
1573         pr_newline(out);
1574
1575         pr_buf(out, "Compat features:           ");
1576         bch2_flags_to_text(out, bch2_sb_compat,
1577                            le64_to_cpu(sb->compat[0]));
1578         pr_newline(out);
1579
1580         pr_buf(out, "Metadata replicas:         %llu", BCH_SB_META_REPLICAS_WANT(sb));
1581         pr_newline(out);
1582
1583         pr_buf(out, "Data replicas:             %llu", BCH_SB_DATA_REPLICAS_WANT(sb));
1584         pr_newline(out);
1585
1586         pr_buf(out, "Metadata checksum type:    %s (%llu)",
1587                BCH_SB_META_CSUM_TYPE(sb) < BCH_CSUM_OPT_NR
1588                ? bch2_csum_opts[BCH_SB_META_CSUM_TYPE(sb)]
1589                : "unknown",
1590                BCH_SB_META_CSUM_TYPE(sb));
1591         pr_newline(out);
1592
1593         pr_buf(out, "Data checksum type:        %s (%llu)",
1594                BCH_SB_DATA_CSUM_TYPE(sb) < BCH_CSUM_OPT_NR
1595                ? bch2_csum_opts[BCH_SB_DATA_CSUM_TYPE(sb)]
1596                : "unknown",
1597                BCH_SB_DATA_CSUM_TYPE(sb));
1598         pr_newline(out);
1599
1600         pr_buf(out, "Compression type:          %s (%llu)",
1601                BCH_SB_COMPRESSION_TYPE(sb) < BCH_COMPRESSION_OPT_NR
1602                ? bch2_compression_opts[BCH_SB_COMPRESSION_TYPE(sb)]
1603                : "unknown",
1604                BCH_SB_COMPRESSION_TYPE(sb));
1605         pr_newline(out);
1606
1607         pr_buf(out, "Foreground write target:   ");
1608         bch2_sb_target_to_text(out, sb, BCH_SB_FOREGROUND_TARGET(sb));
1609         pr_newline(out);
1610
1611         pr_buf(out, "Background write target:   ");
1612         bch2_sb_target_to_text(out, sb, BCH_SB_BACKGROUND_TARGET(sb));
1613         pr_newline(out);
1614
1615         pr_buf(out, "Promote target:            ");
1616         bch2_sb_target_to_text(out, sb, BCH_SB_PROMOTE_TARGET(sb));
1617         pr_newline(out);
1618
1619         pr_buf(out, "Metadata target:           ");
1620         bch2_sb_target_to_text(out, sb, BCH_SB_METADATA_TARGET(sb));
1621         pr_newline(out);
1622
1623         pr_buf(out, "String hash type:          %s (%llu)",
1624                BCH_SB_STR_HASH_TYPE(sb) < BCH_STR_HASH_NR
1625                ? bch2_str_hash_types[BCH_SB_STR_HASH_TYPE(sb)]
1626                : "unknown",
1627                BCH_SB_STR_HASH_TYPE(sb));
1628         pr_newline(out);
1629
1630         pr_buf(out, "32 bit inodes:             %llu", BCH_SB_INODE_32BIT(sb));
1631         pr_newline(out);
1632
1633         pr_buf(out, "GC reserve percentage:     %llu%%", BCH_SB_GC_RESERVE(sb));
1634         pr_newline(out);
1635
1636         pr_buf(out, "Root reserve percentage:   %llu%%", BCH_SB_ROOT_RESERVE(sb));
1637         pr_newline(out);
1638
1639         pr_buf(out, "Devices:                   %u live, %u total",
1640                nr_devices, sb->nr_devices);
1641         pr_newline(out);
1642
1643         pr_buf(out, "Sections:                  ");
1644         vstruct_for_each(sb, f)
1645                 fields_have |= 1 << le32_to_cpu(f->type);
1646         bch2_flags_to_text(out, bch2_sb_fields, fields_have);
1647         pr_newline(out);
1648
1649         pr_buf(out, "Superblock size:           %llu", vstruct_bytes(sb));
1650         pr_newline(out);
1651
1652         if (print_layout) {
1653                 pr_newline(out);
1654                 pr_buf(out, "layout:");
1655                 pr_newline(out);
1656                 pr_indent_push(out, 2);
1657                 bch2_sb_layout_to_text(out, &sb->layout);
1658                 pr_indent_pop(out, 2);
1659         }
1660
1661         vstruct_for_each(sb, f)
1662                 if (fields & (1 << le32_to_cpu(f->type))) {
1663                         pr_newline(out);
1664                         bch2_sb_field_to_text(out, sb, f);
1665                 }
1666 }