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 const char *bch2_sb_field_validate(struct bch_sb *,
  31                                           struct bch_sb_field *);
  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 const char *validate_sb_layout(struct bch_sb_layout *layout)
 206 {
 207         u64 offset, prev_offset, max_sectors;
 208         unsigned i;
 209
 210         if (uuid_le_cmp(layout->magic, BCACHE_MAGIC))
 211                 return "Not a bcachefs superblock layout";
 212
 213         if (layout->layout_type != 0)
 214                 return "Invalid superblock layout type";
 215
 216         if (!layout->nr_superblocks)
 217                 return "Invalid superblock layout: no superblocks";
 218
 219         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset))
 220                 return "Invalid superblock layout: too many superblocks";
 221
 222         max_sectors = 1 << layout->sb_max_size_bits;
 223
 224         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 225
 226         for (i = 1; i < layout->nr_superblocks; i++) {
 227                 offset = le64_to_cpu(layout->sb_offset[i]);
 228
 229                 if (offset < prev_offset + max_sectors)
 230                         return "Invalid superblock layout: superblocks overlap";
 231                 prev_offset = offset;
 232         }
 233
 234         return NULL;
 235 }
 236
 237 const char *bch2_sb_validate(struct bch_sb_handle *disk_sb)
 238 {
 239         struct bch_sb *sb = disk_sb->sb;
 240         struct bch_sb_field *f;
 241         struct bch_sb_field_members *mi;
 242         const char *err;
 243         u32 version, version_min;
 244         u16 block_size;
 245
 246         version         = le16_to_cpu(sb->version);
 247         version_min     = version >= bcachefs_metadata_version_new_versioning
 248                 ? le16_to_cpu(sb->version_min)
 249                 : version;
 250
 251         if (version    >= bcachefs_metadata_version_max ||
 252             version_min < bcachefs_metadata_version_min)
 253                 return "Unsupported superblock version";
 254
 255         if (version_min > version)
 256                 return "Bad minimum version";
 257
 258         if (sb->features[1] ||
 259             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR)))
 260                 return "Filesystem has incompatible features";
 261
 262         block_size = le16_to_cpu(sb->block_size);
 263
 264         if (!is_power_of_2(block_size) ||
 265             block_size > PAGE_SECTORS)
 266                 return "Bad block size";
 267
 268         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
 269                 return "Bad user UUID";
 270
 271         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
 272                 return "Bad internal UUID";
 273
 274         if (!sb->nr_devices ||
 275             sb->nr_devices <= sb->dev_idx ||
 276             sb->nr_devices > BCH_SB_MEMBERS_MAX)
 277                 return "Bad number of member devices";
 278
 279         if (!BCH_SB_META_REPLICAS_WANT(sb) ||
 280             BCH_SB_META_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
 281                 return "Invalid number of metadata replicas";
 282
 283         if (!BCH_SB_META_REPLICAS_REQ(sb) ||
 284             BCH_SB_META_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
 285                 return "Invalid number of metadata replicas";
 286
 287         if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
 288             BCH_SB_DATA_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
 289                 return "Invalid number of data replicas";
 290
 291         if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
 292             BCH_SB_DATA_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
 293                 return "Invalid number of data replicas";
 294
 295         if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 296                 return "Invalid metadata checksum type";
 297
 298         if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 299                 return "Invalid metadata checksum type";
 300
 301         if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
 302                 return "Invalid compression type";
 303
 304         if (!BCH_SB_BTREE_NODE_SIZE(sb))
 305                 return "Btree node size not set";
 306
 307         if (!is_power_of_2(BCH_SB_BTREE_NODE_SIZE(sb)))
 308                 return "Btree node size not a power of two";
 309
 310         if (BCH_SB_GC_RESERVE(sb) < 5)
 311                 return "gc reserve percentage too small";
 312
 313         if (!sb->time_precision ||
 314             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
 315                 return "invalid time precision";
 316
 317         /* validate layout */
 318         err = validate_sb_layout(&sb->layout);
 319         if (err)
 320                 return err;
 321
 322         vstruct_for_each(sb, f) {
 323                 if (!f->u64s)
 324                         return "Invalid superblock: invalid optional field";
 325
 326                 if (vstruct_next(f) > vstruct_last(sb))
 327                         return "Invalid superblock: invalid optional field";
 328         }
 329
 330         /* members must be validated first: */
 331         mi = bch2_sb_get_members(sb);
 332         if (!mi)
 333                 return "Invalid superblock: member info area missing";
 334
 335         err = bch2_sb_field_validate(sb, &mi->field);
 336         if (err)
 337                 return err;
 338
 339         vstruct_for_each(sb, f) {
 340                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 341                         continue;
 342
 343                 err = bch2_sb_field_validate(sb, f);
 344                 if (err)
 345                         return err;
 346         }
 347
 348         return NULL;
 349 }
 350
 351 /* device open: */
 352
 353 static void bch2_sb_update(struct bch_fs *c)
 354 {
 355         struct bch_sb *src = c->disk_sb.sb;
 356         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 357         struct bch_dev *ca;
 358         unsigned i;
 359
 360         lockdep_assert_held(&c->sb_lock);
 361
 362         c->sb.uuid              = src->uuid;
 363         c->sb.user_uuid         = src->user_uuid;
 364         c->sb.version           = le16_to_cpu(src->version);
 365         c->sb.version_min       = le16_to_cpu(src->version_min);
 366         c->sb.nr_devices        = src->nr_devices;
 367         c->sb.clean             = BCH_SB_CLEAN(src);
 368         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 369         c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
 370         c->sb.time_base_lo      = le64_to_cpu(src->time_base_lo);
 371         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 372         c->sb.time_precision    = le32_to_cpu(src->time_precision);
 373         c->sb.features          = le64_to_cpu(src->features[0]);
 374         c->sb.compat            = le64_to_cpu(src->compat[0]);
 375
 376         for_each_member_device(ca, c, i)
 377                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 378 }
 379
 380 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 381 {
 382         struct bch_sb_field *src_f, *dst_f;
 383         struct bch_sb *dst = dst_handle->sb;
 384         unsigned i;
 385
 386         dst->version            = src->version;
 387         dst->version_min        = src->version_min;
 388         dst->seq                = src->seq;
 389         dst->uuid               = src->uuid;
 390         dst->user_uuid          = src->user_uuid;
 391         memcpy(dst->label,      src->label, sizeof(dst->label));
 392
 393         dst->block_size         = src->block_size;
 394         dst->nr_devices         = src->nr_devices;
 395
 396         dst->time_base_lo       = src->time_base_lo;
 397         dst->time_base_hi       = src->time_base_hi;
 398         dst->time_precision     = src->time_precision;
 399
 400         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 401         memcpy(dst->features,   src->features,  sizeof(dst->features));
 402         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 403
 404         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 405                 if (i == BCH_SB_FIELD_journal)
 406                         continue;
 407
 408                 src_f = bch2_sb_field_get(src, i);
 409                 dst_f = bch2_sb_field_get(dst, i);
 410                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 411                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 412
 413                 if (src_f)
 414                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 415         }
 416 }
 417
 418 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 419 {
 420         struct bch_sb_field_journal *journal_buckets =
 421                 bch2_sb_get_journal(src);
 422         unsigned journal_u64s = journal_buckets
 423                 ? le32_to_cpu(journal_buckets->field.u64s)
 424                 : 0;
 425         int ret;
 426
 427         lockdep_assert_held(&c->sb_lock);
 428
 429         ret = bch2_sb_realloc(&c->disk_sb,
 430                               le32_to_cpu(src->u64s) - journal_u64s);
 431         if (ret)
 432                 return ret;
 433
 434         __copy_super(&c->disk_sb, src);
 435
 436         if (BCH_SB_HAS_ERRORS(c->disk_sb.sb))
 437                 set_bit(BCH_FS_ERROR, &c->flags);
 438         if (BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb))
 439                 set_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags);
 440
 441         ret = bch2_sb_replicas_to_cpu_replicas(c);
 442         if (ret)
 443                 return ret;
 444
 445         ret = bch2_sb_disk_groups_to_cpu(c);
 446         if (ret)
 447                 return ret;
 448
 449         bch2_sb_update(c);
 450         return 0;
 451 }
 452
 453 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 454 {
 455         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 456         struct bch_sb_field_journal *journal_buckets =
 457                 bch2_sb_get_journal(dst);
 458         unsigned journal_u64s = journal_buckets
 459                 ? le32_to_cpu(journal_buckets->field.u64s)
 460                 : 0;
 461         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 462         int ret;
 463
 464         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 465         if (ret)
 466                 return ret;
 467
 468         __copy_super(&ca->disk_sb, src);
 469         return 0;
 470 }
 471
 472 /* read superblock: */
 473
 474 static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
 475 {
 476         struct bch_csum csum;
 477         size_t bytes;
 478 reread:
 479         bio_reset(sb->bio);
 480         bio_set_dev(sb->bio, sb->bdev);
 481         sb->bio->bi_iter.bi_sector = offset;
 482         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 483         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 484
 485         if (submit_bio_wait(sb->bio))
 486                 return "IO error";
 487
 488         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
 489                 return "Not a bcachefs superblock";
 490
 491         if (le16_to_cpu(sb->sb->version) <  bcachefs_metadata_version_min ||
 492             le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
 493                 return "Unsupported superblock version";
 494
 495         bytes = vstruct_bytes(sb->sb);
 496
 497         if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
 498                 return "Bad superblock: too big";
 499
 500         if (bytes > sb->buffer_size) {
 501                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 502                         return "cannot allocate memory";
 503                 goto reread;
 504         }
 505
 506         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
 507                 return "unknown csum type";
 508
 509         /* XXX: verify MACs */
 510         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 511                             null_nonce(), sb->sb);
 512
 513         if (bch2_crc_cmp(csum, sb->sb->csum))
 514                 return "bad checksum reading superblock";
 515
 516         sb->seq = le64_to_cpu(sb->sb->seq);
 517
 518         return NULL;
 519 }
 520
 521 int bch2_read_super(const char *path, struct bch_opts *opts,
 522                     struct bch_sb_handle *sb)
 523 {
 524         u64 offset = opt_get(*opts, sb);
 525         struct bch_sb_layout layout;
 526         const char *err;
 527         __le64 *i;
 528         int ret;
 529
 530         pr_verbose_init(*opts, "");
 531
 532         memset(sb, 0, sizeof(*sb));
 533         sb->mode        = FMODE_READ;
 534         sb->have_bio    = true;
 535
 536         if (!opt_get(*opts, noexcl))
 537                 sb->mode |= FMODE_EXCL;
 538
 539         if (!opt_get(*opts, nochanges))
 540                 sb->mode |= FMODE_WRITE;
 541
 542         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 543         if (IS_ERR(sb->bdev) &&
 544             PTR_ERR(sb->bdev) == -EACCES &&
 545             opt_get(*opts, read_only)) {
 546                 sb->mode &= ~FMODE_WRITE;
 547
 548                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 549                 if (!IS_ERR(sb->bdev))
 550                         opt_set(*opts, nochanges, true);
 551         }
 552
 553         if (IS_ERR(sb->bdev)) {
 554                 ret = PTR_ERR(sb->bdev);
 555                 goto out;
 556         }
 557
 558         err = "cannot allocate memory";
 559         ret = bch2_sb_realloc(sb, 0);
 560         if (ret)
 561                 goto err;
 562
 563         ret = -EFAULT;
 564         err = "dynamic fault";
 565         if (bch2_fs_init_fault("read_super"))
 566                 goto err;
 567
 568         ret = -EINVAL;
 569         err = read_one_super(sb, offset);
 570         if (!err)
 571                 goto got_super;
 572
 573         if (opt_defined(*opts, sb))
 574                 goto err;
 575
 576         pr_err("error reading default superblock: %s", err);
 577
 578         /*
 579          * Error reading primary superblock - read location of backup
 580          * superblocks:
 581          */
 582         bio_reset(sb->bio);
 583         bio_set_dev(sb->bio, sb->bdev);
 584         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 585         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 586         /*
 587          * use sb buffer to read layout, since sb buffer is page aligned but
 588          * layout won't be:
 589          */
 590         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 591
 592         err = "IO error";
 593         if (submit_bio_wait(sb->bio))
 594                 goto err;
 595
 596         memcpy(&layout, sb->sb, sizeof(layout));
 597         err = validate_sb_layout(&layout);
 598         if (err)
 599                 goto err;
 600
 601         for (i = layout.sb_offset;
 602              i < layout.sb_offset + layout.nr_superblocks; i++) {
 603                 offset = le64_to_cpu(*i);
 604
 605                 if (offset == opt_get(*opts, sb))
 606                         continue;
 607
 608                 err = read_one_super(sb, offset);
 609                 if (!err)
 610                         goto got_super;
 611         }
 612
 613         ret = -EINVAL;
 614         goto err;
 615
 616 got_super:
 617         err = "Superblock block size smaller than device block size";
 618         ret = -EINVAL;
 619         if (le16_to_cpu(sb->sb->block_size) << 9 <
 620             bdev_logical_block_size(sb->bdev))
 621                 goto err;
 622
 623         ret = 0;
 624         sb->have_layout = true;
 625 out:
 626         pr_verbose_init(*opts, "ret %i", ret);
 627         return ret;
 628 err:
 629         bch2_free_super(sb);
 630         pr_err("error reading superblock: %s", err);
 631         goto out;
 632 }
 633
 634 /* write superblock: */
 635
 636 static void write_super_endio(struct bio *bio)
 637 {
 638         struct bch_dev *ca = bio->bi_private;
 639
 640         /* XXX: return errors directly */
 641
 642         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 643                                bch2_blk_status_to_str(bio->bi_status)))
 644                 ca->sb_write_error = 1;
 645
 646         closure_put(&ca->fs->sb_write);
 647         percpu_ref_put(&ca->io_ref);
 648 }
 649
 650 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 651 {
 652         struct bch_sb *sb = ca->disk_sb.sb;
 653         struct bio *bio = ca->disk_sb.bio;
 654
 655         bio_reset(bio);
 656         bio_set_dev(bio, ca->disk_sb.bdev);
 657         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 658         bio->bi_end_io          = write_super_endio;
 659         bio->bi_private         = ca;
 660         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 661         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 662
 663         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 664                      bio_sectors(bio));
 665
 666         percpu_ref_get(&ca->io_ref);
 667         closure_bio_submit(bio, &c->sb_write);
 668 }
 669
 670 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 671 {
 672         struct bch_sb *sb = ca->disk_sb.sb;
 673         struct bio *bio = ca->disk_sb.bio;
 674
 675         sb->offset = sb->layout.sb_offset[idx];
 676
 677         SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
 678         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 679                                 null_nonce(), sb);
 680
 681         bio_reset(bio);
 682         bio_set_dev(bio, ca->disk_sb.bdev);
 683         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 684         bio->bi_end_io          = write_super_endio;
 685         bio->bi_private         = ca;
 686         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 687         bch2_bio_map(bio, sb,
 688                      roundup((size_t) vstruct_bytes(sb),
 689                              bdev_logical_block_size(ca->disk_sb.bdev)));
 690
 691         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 692                      bio_sectors(bio));
 693
 694         percpu_ref_get(&ca->io_ref);
 695         closure_bio_submit(bio, &c->sb_write);
 696 }
 697
 698 int bch2_write_super(struct bch_fs *c)
 699 {
 700         struct closure *cl = &c->sb_write;
 701         struct bch_dev *ca;
 702         unsigned i, sb = 0, nr_wrote;
 703         const char *err;
 704         struct bch_devs_mask sb_written;
 705         bool wrote, can_mount_without_written, can_mount_with_written;
 706         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 707         int ret = 0;
 708
 709         if (c->opts.very_degraded)
 710                 degraded_flags |= BCH_FORCE_IF_LOST;
 711
 712         lockdep_assert_held(&c->sb_lock);
 713
 714         closure_init_stack(cl);
 715         memset(&sb_written, 0, sizeof(sb_written));
 716
 717         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 718
 719         if (test_bit(BCH_FS_ERROR, &c->flags))
 720                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 721         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 722                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 723
 724         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 725
 726         for_each_online_member(ca, c, i)
 727                 bch2_sb_from_fs(c, ca);
 728
 729         for_each_online_member(ca, c, i) {
 730                 err = bch2_sb_validate(&ca->disk_sb);
 731                 if (err) {
 732                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
 733                         ret = -1;
 734                         goto out;
 735                 }
 736         }
 737
 738         if (c->opts.nochanges)
 739                 goto out;
 740
 741         for_each_online_member(ca, c, i) {
 742                 __set_bit(ca->dev_idx, sb_written.d);
 743                 ca->sb_write_error = 0;
 744         }
 745
 746         for_each_online_member(ca, c, i)
 747                 read_back_super(c, ca);
 748         closure_sync(cl);
 749
 750         for_each_online_member(ca, c, i) {
 751                 if (!ca->sb_write_error &&
 752                     ca->disk_sb.seq !=
 753                     le64_to_cpu(ca->sb_read_scratch->seq)) {
 754                         bch2_fs_fatal_error(c,
 755                                 "Superblock modified by another process");
 756                         percpu_ref_put(&ca->io_ref);
 757                         ret = -EROFS;
 758                         goto out;
 759                 }
 760         }
 761
 762         do {
 763                 wrote = false;
 764                 for_each_online_member(ca, c, i)
 765                         if (!ca->sb_write_error &&
 766                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 767                                 write_one_super(c, ca, sb);
 768                                 wrote = true;
 769                         }
 770                 closure_sync(cl);
 771                 sb++;
 772         } while (wrote);
 773
 774         for_each_online_member(ca, c, i) {
 775                 if (ca->sb_write_error)
 776                         __clear_bit(ca->dev_idx, sb_written.d);
 777                 else
 778                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 779         }
 780
 781         nr_wrote = dev_mask_nr(&sb_written);
 782
 783         can_mount_with_written =
 784                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 785
 786         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 787                 sb_written.d[i] = ~sb_written.d[i];
 788
 789         can_mount_without_written =
 790                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 791
 792         /*
 793          * If we would be able to mount _without_ the devices we successfully
 794          * wrote superblocks to, we weren't able to write to enough devices:
 795          *
 796          * Exception: if we can mount without the successes because we haven't
 797          * written anything (new filesystem), we continue if we'd be able to
 798          * mount with the devices we did successfully write to:
 799          */
 800         if (bch2_fs_fatal_err_on(!nr_wrote ||
 801                                  !can_mount_with_written ||
 802                                  (can_mount_without_written &&
 803                                   !can_mount_with_written), c,
 804                 "Unable to write superblock to sufficient devices"))
 805                 ret = -1;
 806 out:
 807         /* Make new options visible after they're persistent: */
 808         bch2_sb_update(c);
 809         return ret;
 810 }
 811
 812 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 813 {
 814         mutex_lock(&c->sb_lock);
 815         if (!(c->sb.features & (1ULL << feat))) {
 816                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 817
 818                 bch2_write_super(c);
 819         }
 820         mutex_unlock(&c->sb_lock);
 821 }
 822
 823 /* BCH_SB_FIELD_journal: */
 824
 825 static int u64_cmp(const void *_l, const void *_r)
 826 {
 827         u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
 828
 829         return l < r ? -1 : l > r ? 1 : 0;
 830 }
 831
 832 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
 833                                             struct bch_sb_field *f)
 834 {
 835         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 836         struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
 837         const char *err;
 838         unsigned nr;
 839         unsigned i;
 840         u64 *b;
 841
 842         journal = bch2_sb_get_journal(sb);
 843         if (!journal)
 844                 return NULL;
 845
 846         nr = bch2_nr_journal_buckets(journal);
 847         if (!nr)
 848                 return NULL;
 849
 850         b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
 851         if (!b)
 852                 return "cannot allocate memory";
 853
 854         for (i = 0; i < nr; i++)
 855                 b[i] = le64_to_cpu(journal->buckets[i]);
 856
 857         sort(b, nr, sizeof(u64), u64_cmp, NULL);
 858
 859         err = "journal bucket at sector 0";
 860         if (!b[0])
 861                 goto err;
 862
 863         err = "journal bucket before first bucket";
 864         if (m && b[0] < le16_to_cpu(m->first_bucket))
 865                 goto err;
 866
 867         err = "journal bucket past end of device";
 868         if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
 869                 goto err;
 870
 871         err = "duplicate journal buckets";
 872         for (i = 0; i + 1 < nr; i++)
 873                 if (b[i] == b[i + 1])
 874                         goto err;
 875
 876         err = NULL;
 877 err:
 878         kfree(b);
 879         return err;
 880 }
 881
 882 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
 883         .validate       = bch2_sb_validate_journal,
 884 };
 885
 886 /* BCH_SB_FIELD_members: */
 887
 888 static const char *bch2_sb_validate_members(struct bch_sb *sb,
 889                                             struct bch_sb_field *f)
 890 {
 891         struct bch_sb_field_members *mi = field_to_type(f, members);
 892         struct bch_member *m;
 893
 894         if ((void *) (mi->members + sb->nr_devices) >
 895             vstruct_end(&mi->field))
 896                 return "Invalid superblock: bad member info";
 897
 898         for (m = mi->members;
 899              m < mi->members + sb->nr_devices;
 900              m++) {
 901                 if (!bch2_member_exists(m))
 902                         continue;
 903
 904                 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
 905                         return "Too many buckets";
 906
 907                 if (le64_to_cpu(m->nbuckets) -
 908                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
 909                         return "Not enough buckets";
 910
 911                 if (le16_to_cpu(m->bucket_size) <
 912                     le16_to_cpu(sb->block_size))
 913                         return "bucket size smaller than block size";
 914
 915                 if (le16_to_cpu(m->bucket_size) <
 916                     BCH_SB_BTREE_NODE_SIZE(sb))
 917                         return "bucket size smaller than btree node size";
 918         }
 919
 920         return NULL;
 921 }
 922
 923 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
 924         .validate       = bch2_sb_validate_members,
 925 };
 926
 927 /* BCH_SB_FIELD_crypt: */
 928
 929 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
 930                                           struct bch_sb_field *f)
 931 {
 932         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
 933
 934         if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
 935                 return "invalid field crypt: wrong size";
 936
 937         if (BCH_CRYPT_KDF_TYPE(crypt))
 938                 return "invalid field crypt: bad kdf type";
 939
 940         return NULL;
 941 }
 942
 943 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
 944         .validate       = bch2_sb_validate_crypt,
 945 };
 946
 947 /* BCH_SB_FIELD_clean: */
 948
 949 int bch2_sb_clean_validate(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
 950 {
 951         struct jset_entry *entry;
 952         int ret;
 953
 954         for (entry = clean->start;
 955              entry < (struct jset_entry *) vstruct_end(&clean->field);
 956              entry = vstruct_next(entry)) {
 957                 ret = bch2_journal_entry_validate(c, "superblock", entry,
 958                                                   le16_to_cpu(c->disk_sb.sb->version),
 959                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
 960                                                   write);
 961                 if (ret)
 962                         return ret;
 963         }
 964
 965         return 0;
 966 }
 967
 968 int bch2_fs_mark_dirty(struct bch_fs *c)
 969 {
 970         int ret;
 971
 972         /*
 973          * Unconditionally write superblock, to verify it hasn't changed before
 974          * we go rw:
 975          */
 976
 977         mutex_lock(&c->sb_lock);
 978         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 979         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALWAYS;
 980         ret = bch2_write_super(c);
 981         mutex_unlock(&c->sb_lock);
 982
 983         return ret;
 984 }
 985
 986 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
 987 {
 988         struct jset_entry *entry = *end;
 989         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
 990
 991         memset(entry, 0, u64s * sizeof(u64));
 992         /*
 993          * The u64s field counts from the start of data, ignoring the shared
 994          * fields.
 995          */
 996         entry->u64s = u64s - 1;
 997
 998         *end = vstruct_next(*end);
 999         return entry;
1000 }
1001
1002 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1003                                            struct jset_entry **end,
1004                                            u64 journal_seq)
1005 {
1006         struct bch_dev *ca;
1007         unsigned i, dev;
1008
1009         percpu_down_read(&c->mark_lock);
1010
1011         if (!journal_seq) {
1012                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1013                         bch2_fs_usage_acc_to_base(c, i);
1014         } else {
1015                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1016         }
1017
1018         {
1019                 struct jset_entry_usage *u =
1020                         container_of(jset_entry_init(end, sizeof(*u)),
1021                                      struct jset_entry_usage, entry);
1022
1023                 u->entry.type   = BCH_JSET_ENTRY_usage;
1024                 u->entry.btree_id = FS_USAGE_INODES;
1025                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1026         }
1027
1028         {
1029                 struct jset_entry_usage *u =
1030                         container_of(jset_entry_init(end, sizeof(*u)),
1031                                      struct jset_entry_usage, entry);
1032
1033                 u->entry.type   = BCH_JSET_ENTRY_usage;
1034                 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1035                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1036         }
1037
1038         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1039                 struct jset_entry_usage *u =
1040                         container_of(jset_entry_init(end, sizeof(*u)),
1041                                      struct jset_entry_usage, entry);
1042
1043                 u->entry.type   = BCH_JSET_ENTRY_usage;
1044                 u->entry.btree_id = FS_USAGE_RESERVED;
1045                 u->entry.level  = i;
1046                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1047         }
1048
1049         for (i = 0; i < c->replicas.nr; i++) {
1050                 struct bch_replicas_entry *e =
1051                         cpu_replicas_entry(&c->replicas, i);
1052                 struct jset_entry_data_usage *u =
1053                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1054                                      struct jset_entry_data_usage, entry);
1055
1056                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1057                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1058                 memcpy(&u->r, e, replicas_entry_bytes(e));
1059         }
1060
1061         for_each_member_device(ca, c, dev) {
1062                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1063                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1064                 struct jset_entry_dev_usage *u =
1065                         container_of(jset_entry_init(end, b),
1066                                      struct jset_entry_dev_usage, entry);
1067
1068                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1069                 u->dev = cpu_to_le32(dev);
1070                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1071                 u->buckets_unavailable  = cpu_to_le64(ca->usage_base->buckets_unavailable);
1072
1073                 for (i = 0; i < BCH_DATA_NR; i++) {
1074                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1075                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1076                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1077                 }
1078         }
1079
1080         percpu_up_read(&c->mark_lock);
1081
1082         for (i = 0; i < 2; i++) {
1083                 struct jset_entry_clock *clock =
1084                         container_of(jset_entry_init(end, sizeof(*clock)),
1085                                      struct jset_entry_clock, entry);
1086
1087                 clock->entry.type = BCH_JSET_ENTRY_clock;
1088                 clock->rw       = i;
1089                 clock->time     = atomic64_read(&c->io_clock[i].now);
1090         }
1091 }
1092
1093 void bch2_fs_mark_clean(struct bch_fs *c)
1094 {
1095         struct bch_sb_field_clean *sb_clean;
1096         struct jset_entry *entry;
1097         unsigned u64s;
1098         int ret;
1099
1100         mutex_lock(&c->sb_lock);
1101         if (BCH_SB_CLEAN(c->disk_sb.sb))
1102                 goto out;
1103
1104         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1105
1106         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_alloc_info;
1107         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_alloc_metadata;
1108         c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_extents_above_btree_updates);
1109         c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_btree_updates_journalled);
1110
1111         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1112
1113         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1114         if (!sb_clean) {
1115                 bch_err(c, "error resizing superblock while setting filesystem clean");
1116                 goto out;
1117         }
1118
1119         sb_clean->flags         = 0;
1120         sb_clean->journal_seq   = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1121
1122         /* Trying to catch outstanding bug: */
1123         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1124
1125         entry = sb_clean->start;
1126         bch2_journal_super_entries_add_common(c, &entry, 0);
1127         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1128         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1129
1130         memset(entry, 0,
1131                vstruct_end(&sb_clean->field) - (void *) entry);
1132
1133         /*
1134          * this should be in the write path, and we should be validating every
1135          * superblock section:
1136          */
1137         ret = bch2_sb_clean_validate(c, sb_clean, WRITE);
1138         if (ret) {
1139                 bch_err(c, "error writing marking filesystem clean: validate error");
1140                 goto out;
1141         }
1142
1143         bch2_write_super(c);
1144 out:
1145         mutex_unlock(&c->sb_lock);
1146 }
1147
1148 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1149                                           struct bch_sb_field *f)
1150 {
1151         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1152
1153         if (vstruct_bytes(&clean->field) < sizeof(*clean))
1154                 return "invalid field crypt: wrong size";
1155
1156         return NULL;
1157 }
1158
1159 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1160         .validate       = bch2_sb_validate_clean,
1161 };
1162
1163 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1164 #define x(f, nr)                                        \
1165         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1166         BCH_SB_FIELDS()
1167 #undef x
1168 };
1169
1170 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1171                                           struct bch_sb_field *f)
1172 {
1173         unsigned type = le32_to_cpu(f->type);
1174
1175         return type < BCH_SB_FIELD_NR
1176                 ? bch2_sb_field_ops[type]->validate(sb, f)
1177                 : NULL;
1178 }
1179
1180 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1181                            struct bch_sb_field *f)
1182 {
1183         unsigned type = le32_to_cpu(f->type);
1184         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1185                 ? bch2_sb_field_ops[type] : NULL;
1186
1187         if (ops)
1188                 pr_buf(out, "%s", bch2_sb_fields[type]);
1189         else
1190                 pr_buf(out, "(unknown field %u)", type);
1191
1192         pr_buf(out, " (size %llu):", vstruct_bytes(f));
1193
1194         if (ops && ops->to_text)
1195                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1196 }