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(get_order(__vstruct_bytes(struct bch_sb, sb_u64s)) >
  54                sb->page_order);
  55
  56         if (!f && !u64s) {
  57                 /* nothing to do: */
  58         } else if (!f) {
  59                 f = vstruct_last(sb->sb);
  60                 memset(f, 0, sizeof(u64) * u64s);
  61                 f->u64s = cpu_to_le32(u64s);
  62                 f->type = 0;
  63         } else {
  64                 void *src, *dst;
  65
  66                 src = vstruct_end(f);
  67
  68                 if (u64s) {
  69                         f->u64s = cpu_to_le32(u64s);
  70                         dst = vstruct_end(f);
  71                 } else {
  72                         dst = f;
  73                 }
  74
  75                 memmove(dst, src, vstruct_end(sb->sb) - src);
  76
  77                 if (dst > src)
  78                         memset(src, 0, dst - src);
  79         }
  80
  81         sb->sb->u64s = cpu_to_le32(sb_u64s);
  82
  83         return u64s ? f : NULL;
  84 }
  85
  86 void bch2_sb_field_delete(struct bch_sb_handle *sb,
  87                           enum bch_sb_field_type type)
  88 {
  89         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
  90
  91         if (f)
  92                 __bch2_sb_field_resize(sb, f, 0);
  93 }
  94
  95 /* Superblock realloc/free: */
  96
  97 void bch2_free_super(struct bch_sb_handle *sb)
  98 {
  99         if (sb->bio)
 100                 bio_put(sb->bio);
 101         if (!IS_ERR_OR_NULL(sb->bdev))
 102                 blkdev_put(sb->bdev, sb->mode);
 103
 104         free_pages((unsigned long) sb->sb, sb->page_order);
 105         memset(sb, 0, sizeof(*sb));
 106 }
 107
 108 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 109 {
 110         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 111         unsigned order = get_order(new_bytes);
 112         struct bch_sb *new_sb;
 113         struct bio *bio;
 114
 115         if (sb->sb && sb->page_order >= order)
 116                 return 0;
 117
 118         if (sb->have_layout) {
 119                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 120
 121                 if (new_bytes > max_bytes) {
 122                         char buf[BDEVNAME_SIZE];
 123
 124                         pr_err("%s: superblock too big: want %zu but have %llu",
 125                                bdevname(sb->bdev, buf), new_bytes, max_bytes);
 126                         return -ENOSPC;
 127                 }
 128         }
 129
 130         if (sb->page_order >= order && sb->sb)
 131                 return 0;
 132
 133         if (dynamic_fault("bcachefs:add:super_realloc"))
 134                 return -ENOMEM;
 135
 136         if (sb->have_bio) {
 137                 bio = bio_kmalloc(GFP_KERNEL, 1 << order);
 138                 if (!bio)
 139                         return -ENOMEM;
 140
 141                 if (sb->bio)
 142                         bio_put(sb->bio);
 143                 sb->bio = bio;
 144         }
 145
 146         new_sb = (void *) __get_free_pages(GFP_NOFS|__GFP_ZERO, order);
 147         if (!new_sb)
 148                 return -ENOMEM;
 149
 150         if (sb->sb)
 151                 memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
 152
 153         free_pages((unsigned long) sb->sb, sb->page_order);
 154         sb->sb = new_sb;
 155
 156         sb->page_order = order;
 157
 158         return 0;
 159 }
 160
 161 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 162                                           enum bch_sb_field_type type,
 163                                           unsigned u64s)
 164 {
 165         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 166         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 167         ssize_t d = -old_u64s + u64s;
 168
 169         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 170                 return NULL;
 171
 172         if (sb->fs_sb) {
 173                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 174                 struct bch_dev *ca;
 175                 unsigned i;
 176
 177                 lockdep_assert_held(&c->sb_lock);
 178
 179                 /* XXX: we're not checking that offline device have enough space */
 180
 181                 for_each_online_member(ca, c, i) {
 182                         struct bch_sb_handle *sb = &ca->disk_sb;
 183
 184                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 185                                 percpu_ref_put(&ca->ref);
 186                                 return NULL;
 187                         }
 188                 }
 189         }
 190
 191         f = bch2_sb_field_get(sb->sb, type);
 192         f = __bch2_sb_field_resize(sb, f, u64s);
 193         if (f)
 194                 f->type = cpu_to_le32(type);
 195         return f;
 196 }
 197
 198 /* Superblock validate: */
 199
 200 static inline void __bch2_sb_layout_size_assert(void)
 201 {
 202         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 203 }
 204
 205 static const char *validate_sb_layout(struct bch_sb_layout *layout)
 206 {
 207         u64 offset, prev_offset, max_sectors;
 208         unsigned i;
 209
 210         if (uuid_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.nr_devices        = src->nr_devices;
 366         c->sb.clean             = BCH_SB_CLEAN(src);
 367         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 368         c->sb.encoded_extent_max= 1 << BCH_SB_ENCODED_EXTENT_MAX_BITS(src);
 369         c->sb.time_base_lo      = le64_to_cpu(src->time_base_lo);
 370         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 371         c->sb.time_precision    = le32_to_cpu(src->time_precision);
 372         c->sb.features          = le64_to_cpu(src->features[0]);
 373         c->sb.compat            = le64_to_cpu(src->compat[0]);
 374
 375         for_each_member_device(ca, c, i)
 376                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 377 }
 378
 379 /* doesn't copy member info */
 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         ret = bch2_sb_replicas_to_cpu_replicas(c);
 437         if (ret)
 438                 return ret;
 439
 440         ret = bch2_sb_disk_groups_to_cpu(c);
 441         if (ret)
 442                 return ret;
 443
 444         bch2_sb_update(c);
 445         return 0;
 446 }
 447
 448 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 449 {
 450         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 451         struct bch_sb_field_journal *journal_buckets =
 452                 bch2_sb_get_journal(dst);
 453         unsigned journal_u64s = journal_buckets
 454                 ? le32_to_cpu(journal_buckets->field.u64s)
 455                 : 0;
 456         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 457         int ret;
 458
 459         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 460         if (ret)
 461                 return ret;
 462
 463         __copy_super(&ca->disk_sb, src);
 464         return 0;
 465 }
 466
 467 /* read superblock: */
 468
 469 static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
 470 {
 471         struct bch_csum csum;
 472         size_t bytes;
 473 reread:
 474         bio_reset(sb->bio);
 475         bio_set_dev(sb->bio, sb->bdev);
 476         sb->bio->bi_iter.bi_sector = offset;
 477         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 478         bch2_bio_map(sb->bio, sb->sb, PAGE_SIZE << sb->page_order);
 479
 480         if (submit_bio_wait(sb->bio))
 481                 return "IO error";
 482
 483         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
 484                 return "Not a bcachefs superblock";
 485
 486         if (le16_to_cpu(sb->sb->version) <  bcachefs_metadata_version_min ||
 487             le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
 488                 return "Unsupported superblock version";
 489
 490         bytes = vstruct_bytes(sb->sb);
 491
 492         if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
 493                 return "Bad superblock: too big";
 494
 495         if (get_order(bytes) > sb->page_order) {
 496                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 497                         return "cannot allocate memory";
 498                 goto reread;
 499         }
 500
 501         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
 502                 return "unknown csum type";
 503
 504         /* XXX: verify MACs */
 505         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 506                             null_nonce(), sb->sb);
 507
 508         if (bch2_crc_cmp(csum, sb->sb->csum))
 509                 return "bad checksum reading superblock";
 510
 511         sb->seq = le64_to_cpu(sb->sb->seq);
 512
 513         return NULL;
 514 }
 515
 516 int bch2_read_super(const char *path, struct bch_opts *opts,
 517                     struct bch_sb_handle *sb)
 518 {
 519         u64 offset = opt_get(*opts, sb);
 520         struct bch_sb_layout layout;
 521         const char *err;
 522         __le64 *i;
 523         int ret;
 524
 525         pr_verbose_init(*opts, "");
 526
 527         memset(sb, 0, sizeof(*sb));
 528         sb->mode        = FMODE_READ;
 529         sb->have_bio    = true;
 530
 531         if (!opt_get(*opts, noexcl))
 532                 sb->mode |= FMODE_EXCL;
 533
 534         if (!opt_get(*opts, nochanges))
 535                 sb->mode |= FMODE_WRITE;
 536
 537         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 538         if (IS_ERR(sb->bdev) &&
 539             PTR_ERR(sb->bdev) == -EACCES &&
 540             opt_get(*opts, read_only)) {
 541                 sb->mode &= ~FMODE_WRITE;
 542
 543                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 544                 if (!IS_ERR(sb->bdev))
 545                         opt_set(*opts, nochanges, true);
 546         }
 547
 548         if (IS_ERR(sb->bdev)) {
 549                 ret = PTR_ERR(sb->bdev);
 550                 goto out;
 551         }
 552
 553         err = "cannot allocate memory";
 554         ret = bch2_sb_realloc(sb, 0);
 555         if (ret)
 556                 goto err;
 557
 558         ret = -EFAULT;
 559         err = "dynamic fault";
 560         if (bch2_fs_init_fault("read_super"))
 561                 goto err;
 562
 563         ret = -EINVAL;
 564         err = read_one_super(sb, offset);
 565         if (!err)
 566                 goto got_super;
 567
 568         if (opt_defined(*opts, sb))
 569                 goto err;
 570
 571         pr_err("error reading default superblock: %s", err);
 572
 573         /*
 574          * Error reading primary superblock - read location of backup
 575          * superblocks:
 576          */
 577         bio_reset(sb->bio);
 578         bio_set_dev(sb->bio, sb->bdev);
 579         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 580         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 581         /*
 582          * use sb buffer to read layout, since sb buffer is page aligned but
 583          * layout won't be:
 584          */
 585         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 586
 587         err = "IO error";
 588         if (submit_bio_wait(sb->bio))
 589                 goto err;
 590
 591         memcpy(&layout, sb->sb, sizeof(layout));
 592         err = validate_sb_layout(&layout);
 593         if (err)
 594                 goto err;
 595
 596         for (i = layout.sb_offset;
 597              i < layout.sb_offset + layout.nr_superblocks; i++) {
 598                 offset = le64_to_cpu(*i);
 599
 600                 if (offset == opt_get(*opts, sb))
 601                         continue;
 602
 603                 err = read_one_super(sb, offset);
 604                 if (!err)
 605                         goto got_super;
 606         }
 607
 608         ret = -EINVAL;
 609         goto err;
 610
 611 got_super:
 612         err = "Superblock block size smaller than device block size";
 613         ret = -EINVAL;
 614         if (le16_to_cpu(sb->sb->block_size) << 9 <
 615             bdev_logical_block_size(sb->bdev))
 616                 goto err;
 617
 618         ret = 0;
 619         sb->have_layout = true;
 620 out:
 621         pr_verbose_init(*opts, "ret %i", ret);
 622         return ret;
 623 err:
 624         bch2_free_super(sb);
 625         pr_err("error reading superblock: %s", err);
 626         goto out;
 627 }
 628
 629 /* write superblock: */
 630
 631 static void write_super_endio(struct bio *bio)
 632 {
 633         struct bch_dev *ca = bio->bi_private;
 634
 635         /* XXX: return errors directly */
 636
 637         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 638                                bch2_blk_status_to_str(bio->bi_status)))
 639                 ca->sb_write_error = 1;
 640
 641         closure_put(&ca->fs->sb_write);
 642         percpu_ref_put(&ca->io_ref);
 643 }
 644
 645 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 646 {
 647         struct bch_sb *sb = ca->disk_sb.sb;
 648         struct bio *bio = ca->disk_sb.bio;
 649
 650         bio_reset(bio);
 651         bio_set_dev(bio, ca->disk_sb.bdev);
 652         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 653         bio->bi_end_io          = write_super_endio;
 654         bio->bi_private         = ca;
 655         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 656         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 657
 658         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 659                      bio_sectors(bio));
 660
 661         percpu_ref_get(&ca->io_ref);
 662         closure_bio_submit(bio, &c->sb_write);
 663 }
 664
 665 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 666 {
 667         struct bch_sb *sb = ca->disk_sb.sb;
 668         struct bio *bio = ca->disk_sb.bio;
 669
 670         sb->offset = sb->layout.sb_offset[idx];
 671
 672         SET_BCH_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
 673         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 674                                 null_nonce(), sb);
 675
 676         bio_reset(bio);
 677         bio_set_dev(bio, ca->disk_sb.bdev);
 678         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 679         bio->bi_end_io          = write_super_endio;
 680         bio->bi_private         = ca;
 681         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 682         bch2_bio_map(bio, sb,
 683                      roundup((size_t) vstruct_bytes(sb),
 684                              bdev_logical_block_size(ca->disk_sb.bdev)));
 685
 686         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 687                      bio_sectors(bio));
 688
 689         percpu_ref_get(&ca->io_ref);
 690         closure_bio_submit(bio, &c->sb_write);
 691 }
 692
 693 int bch2_write_super(struct bch_fs *c)
 694 {
 695         struct closure *cl = &c->sb_write;
 696         struct bch_dev *ca;
 697         unsigned i, sb = 0, nr_wrote;
 698         const char *err;
 699         struct bch_devs_mask sb_written;
 700         bool wrote, can_mount_without_written, can_mount_with_written;
 701         int ret = 0;
 702
 703         lockdep_assert_held(&c->sb_lock);
 704
 705         closure_init_stack(cl);
 706         memset(&sb_written, 0, sizeof(sb_written));
 707
 708         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 709
 710         if (test_bit(BCH_FS_ERROR, &c->flags))
 711                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 712
 713         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 714
 715         for_each_online_member(ca, c, i)
 716                 bch2_sb_from_fs(c, ca);
 717
 718         for_each_online_member(ca, c, i) {
 719                 err = bch2_sb_validate(&ca->disk_sb);
 720                 if (err) {
 721                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
 722                         ret = -1;
 723                         goto out;
 724                 }
 725         }
 726
 727         if (c->opts.nochanges)
 728                 goto out;
 729
 730         for_each_online_member(ca, c, i) {
 731                 __set_bit(ca->dev_idx, sb_written.d);
 732                 ca->sb_write_error = 0;
 733         }
 734
 735         for_each_online_member(ca, c, i)
 736                 read_back_super(c, ca);
 737         closure_sync(cl);
 738
 739         for_each_online_member(ca, c, i) {
 740                 if (!ca->sb_write_error &&
 741                     ca->disk_sb.seq !=
 742                     le64_to_cpu(ca->sb_read_scratch->seq)) {
 743                         bch2_fs_fatal_error(c,
 744                                 "Superblock modified by another process");
 745                         percpu_ref_put(&ca->io_ref);
 746                         ret = -EROFS;
 747                         goto out;
 748                 }
 749         }
 750
 751         do {
 752                 wrote = false;
 753                 for_each_online_member(ca, c, i)
 754                         if (!ca->sb_write_error &&
 755                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 756                                 write_one_super(c, ca, sb);
 757                                 wrote = true;
 758                         }
 759                 closure_sync(cl);
 760                 sb++;
 761         } while (wrote);
 762
 763         for_each_online_member(ca, c, i) {
 764                 if (ca->sb_write_error)
 765                         __clear_bit(ca->dev_idx, sb_written.d);
 766                 else
 767                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 768         }
 769
 770         nr_wrote = dev_mask_nr(&sb_written);
 771
 772         can_mount_with_written =
 773                 bch2_have_enough_devs(c, sb_written, BCH_FORCE_IF_DEGRADED, false);
 774
 775         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 776                 sb_written.d[i] = ~sb_written.d[i];
 777
 778         can_mount_without_written =
 779                 bch2_have_enough_devs(c, sb_written, BCH_FORCE_IF_DEGRADED, false);
 780
 781         /*
 782          * If we would be able to mount _without_ the devices we successfully
 783          * wrote superblocks to, we weren't able to write to enough devices:
 784          *
 785          * Exception: if we can mount without the successes because we haven't
 786          * written anything (new filesystem), we continue if we'd be able to
 787          * mount with the devices we did successfully write to:
 788          */
 789         if (bch2_fs_fatal_err_on(!nr_wrote ||
 790                                  !can_mount_with_written ||
 791                                  (can_mount_without_written &&
 792                                   !can_mount_with_written), c,
 793                 "Unable to write superblock to sufficient devices"))
 794                 ret = -1;
 795 out:
 796         /* Make new options visible after they're persistent: */
 797         bch2_sb_update(c);
 798         return ret;
 799 }
 800
 801 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 802 {
 803         mutex_lock(&c->sb_lock);
 804         if (!(c->sb.features & (1ULL << feat))) {
 805                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 806
 807                 bch2_write_super(c);
 808         }
 809         mutex_unlock(&c->sb_lock);
 810 }
 811
 812 /* BCH_SB_FIELD_journal: */
 813
 814 static int u64_cmp(const void *_l, const void *_r)
 815 {
 816         u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
 817
 818         return l < r ? -1 : l > r ? 1 : 0;
 819 }
 820
 821 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
 822                                             struct bch_sb_field *f)
 823 {
 824         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 825         struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
 826         const char *err;
 827         unsigned nr;
 828         unsigned i;
 829         u64 *b;
 830
 831         journal = bch2_sb_get_journal(sb);
 832         if (!journal)
 833                 return NULL;
 834
 835         nr = bch2_nr_journal_buckets(journal);
 836         if (!nr)
 837                 return NULL;
 838
 839         b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
 840         if (!b)
 841                 return "cannot allocate memory";
 842
 843         for (i = 0; i < nr; i++)
 844                 b[i] = le64_to_cpu(journal->buckets[i]);
 845
 846         sort(b, nr, sizeof(u64), u64_cmp, NULL);
 847
 848         err = "journal bucket at sector 0";
 849         if (!b[0])
 850                 goto err;
 851
 852         err = "journal bucket before first bucket";
 853         if (m && b[0] < le16_to_cpu(m->first_bucket))
 854                 goto err;
 855
 856         err = "journal bucket past end of device";
 857         if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
 858                 goto err;
 859
 860         err = "duplicate journal buckets";
 861         for (i = 0; i + 1 < nr; i++)
 862                 if (b[i] == b[i + 1])
 863                         goto err;
 864
 865         err = NULL;
 866 err:
 867         kfree(b);
 868         return err;
 869 }
 870
 871 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
 872         .validate       = bch2_sb_validate_journal,
 873 };
 874
 875 /* BCH_SB_FIELD_members: */
 876
 877 static const char *bch2_sb_validate_members(struct bch_sb *sb,
 878                                             struct bch_sb_field *f)
 879 {
 880         struct bch_sb_field_members *mi = field_to_type(f, members);
 881         struct bch_member *m;
 882
 883         if ((void *) (mi->members + sb->nr_devices) >
 884             vstruct_end(&mi->field))
 885                 return "Invalid superblock: bad member info";
 886
 887         for (m = mi->members;
 888              m < mi->members + sb->nr_devices;
 889              m++) {
 890                 if (!bch2_member_exists(m))
 891                         continue;
 892
 893                 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
 894                         return "Too many buckets";
 895
 896                 if (le64_to_cpu(m->nbuckets) -
 897                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
 898                         return "Not enough buckets";
 899
 900                 if (le16_to_cpu(m->bucket_size) <
 901                     le16_to_cpu(sb->block_size))
 902                         return "bucket size smaller than block size";
 903
 904                 if (le16_to_cpu(m->bucket_size) <
 905                     BCH_SB_BTREE_NODE_SIZE(sb))
 906                         return "bucket size smaller than btree node size";
 907         }
 908
 909         return NULL;
 910 }
 911
 912 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
 913         .validate       = bch2_sb_validate_members,
 914 };
 915
 916 /* BCH_SB_FIELD_crypt: */
 917
 918 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
 919                                           struct bch_sb_field *f)
 920 {
 921         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
 922
 923         if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
 924                 return "invalid field crypt: wrong size";
 925
 926         if (BCH_CRYPT_KDF_TYPE(crypt))
 927                 return "invalid field crypt: bad kdf type";
 928
 929         return NULL;
 930 }
 931
 932 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
 933         .validate       = bch2_sb_validate_crypt,
 934 };
 935
 936 /* BCH_SB_FIELD_clean: */
 937
 938 int bch2_sb_clean_validate(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
 939 {
 940         struct jset_entry *entry;
 941         int ret;
 942
 943         for (entry = clean->start;
 944              entry < (struct jset_entry *) vstruct_end(&clean->field);
 945              entry = vstruct_next(entry)) {
 946                 ret = bch2_journal_entry_validate(c, "superblock", entry,
 947                                                   le16_to_cpu(c->disk_sb.sb->version),
 948                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
 949                                                   write);
 950                 if (ret)
 951                         return ret;
 952         }
 953
 954         return 0;
 955 }
 956
 957 int bch2_fs_mark_dirty(struct bch_fs *c)
 958 {
 959         int ret;
 960
 961         /*
 962          * Unconditionally write superblock, to verify it hasn't changed before
 963          * we go rw:
 964          */
 965
 966         mutex_lock(&c->sb_lock);
 967         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 968         c->disk_sb.sb->features[0] |= BCH_SB_FEATURES_ALWAYS;
 969         ret = bch2_write_super(c);
 970         mutex_unlock(&c->sb_lock);
 971
 972         return ret;
 973 }
 974
 975 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
 976 {
 977         struct jset_entry *entry = *end;
 978         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
 979
 980         memset(entry, 0, u64s * sizeof(u64));
 981         /*
 982          * The u64s field counts from the start of data, ignoring the shared
 983          * fields.
 984          */
 985         entry->u64s = u64s - 1;
 986
 987         *end = vstruct_next(*end);
 988         return entry;
 989 }
 990
 991 void bch2_journal_super_entries_add_common(struct bch_fs *c,
 992                                            struct jset_entry **end,
 993                                            u64 journal_seq)
 994 {
 995         struct bch_dev *ca;
 996         unsigned i, dev;
 997
 998         percpu_down_write(&c->mark_lock);
 999
1000         if (!journal_seq) {
1001                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1002                         bch2_fs_usage_acc_to_base(c, i);
1003         } else {
1004                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1005         }
1006
1007         {
1008                 struct jset_entry_usage *u =
1009                         container_of(jset_entry_init(end, sizeof(*u)),
1010                                      struct jset_entry_usage, entry);
1011
1012                 u->entry.type   = BCH_JSET_ENTRY_usage;
1013                 u->entry.btree_id = FS_USAGE_INODES;
1014                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1015         }
1016
1017         {
1018                 struct jset_entry_usage *u =
1019                         container_of(jset_entry_init(end, sizeof(*u)),
1020                                      struct jset_entry_usage, entry);
1021
1022                 u->entry.type   = BCH_JSET_ENTRY_usage;
1023                 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1024                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1025         }
1026
1027         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1028                 struct jset_entry_usage *u =
1029                         container_of(jset_entry_init(end, sizeof(*u)),
1030                                      struct jset_entry_usage, entry);
1031
1032                 u->entry.type   = BCH_JSET_ENTRY_usage;
1033                 u->entry.btree_id = FS_USAGE_RESERVED;
1034                 u->entry.level  = i;
1035                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1036         }
1037
1038         for (i = 0; i < c->replicas.nr; i++) {
1039                 struct bch_replicas_entry *e =
1040                         cpu_replicas_entry(&c->replicas, i);
1041                 struct jset_entry_data_usage *u =
1042                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1043                                      struct jset_entry_data_usage, entry);
1044
1045                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1046                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1047                 memcpy(&u->r, e, replicas_entry_bytes(e));
1048         }
1049
1050         for_each_member_device(ca, c, dev) {
1051                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1052                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1053                 struct jset_entry_dev_usage *u =
1054                         container_of(jset_entry_init(end, b),
1055                                      struct jset_entry_dev_usage, entry);
1056
1057                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1058                 u->dev = cpu_to_le32(dev);
1059                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1060                 u->buckets_unavailable  = cpu_to_le64(ca->usage_base->buckets_unavailable);
1061
1062                 for (i = 0; i < BCH_DATA_NR; i++) {
1063                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1064                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1065                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1066                 }
1067         }
1068
1069         percpu_up_write(&c->mark_lock);
1070
1071         for (i = 0; i < 2; i++) {
1072                 struct jset_entry_clock *clock =
1073                         container_of(jset_entry_init(end, sizeof(*clock)),
1074                                      struct jset_entry_clock, entry);
1075
1076                 clock->entry.type = BCH_JSET_ENTRY_clock;
1077                 clock->rw       = i;
1078                 clock->time     = atomic64_read(&c->io_clock[i].now);
1079         }
1080 }
1081
1082 void bch2_fs_mark_clean(struct bch_fs *c)
1083 {
1084         struct bch_sb_field_clean *sb_clean;
1085         struct jset_entry *entry;
1086         unsigned u64s;
1087         int ret;
1088
1089         mutex_lock(&c->sb_lock);
1090         if (BCH_SB_CLEAN(c->disk_sb.sb))
1091                 goto out;
1092
1093         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1094
1095         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_INFO;
1096         c->disk_sb.sb->compat[0] |= 1ULL << BCH_COMPAT_FEAT_ALLOC_METADATA;
1097         c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_extents_above_btree_updates);
1098         c->disk_sb.sb->features[0] &= ~(1ULL << BCH_FEATURE_btree_updates_journalled);
1099
1100         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1101
1102         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1103         if (!sb_clean) {
1104                 bch_err(c, "error resizing superblock while setting filesystem clean");
1105                 goto out;
1106         }
1107
1108         sb_clean->flags         = 0;
1109         sb_clean->journal_seq   = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1110
1111         /* Trying to catch outstanding bug: */
1112         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1113
1114         entry = sb_clean->start;
1115         bch2_journal_super_entries_add_common(c, &entry, 0);
1116         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1117         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1118
1119         memset(entry, 0,
1120                vstruct_end(&sb_clean->field) - (void *) entry);
1121
1122         /*
1123          * this should be in the write path, and we should be validating every
1124          * superblock section:
1125          */
1126         ret = bch2_sb_clean_validate(c, sb_clean, WRITE);
1127         if (ret) {
1128                 bch_err(c, "error writing marking filesystem clean: validate error");
1129                 goto out;
1130         }
1131
1132         bch2_write_super(c);
1133 out:
1134         mutex_unlock(&c->sb_lock);
1135 }
1136
1137 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1138                                           struct bch_sb_field *f)
1139 {
1140         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1141
1142         if (vstruct_bytes(&clean->field) < sizeof(*clean))
1143                 return "invalid field crypt: wrong size";
1144
1145         return NULL;
1146 }
1147
1148 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1149         .validate       = bch2_sb_validate_clean,
1150 };
1151
1152 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1153 #define x(f, nr)                                        \
1154         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1155         BCH_SB_FIELDS()
1156 #undef x
1157 };
1158
1159 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1160                                           struct bch_sb_field *f)
1161 {
1162         unsigned type = le32_to_cpu(f->type);
1163
1164         return type < BCH_SB_FIELD_NR
1165                 ? bch2_sb_field_ops[type]->validate(sb, f)
1166                 : NULL;
1167 }
1168
1169 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1170                            struct bch_sb_field *f)
1171 {
1172         unsigned type = le32_to_cpu(f->type);
1173         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1174                 ? bch2_sb_field_ops[type] : NULL;
1175
1176         if (ops)
1177                 pr_buf(out, "%s", bch2_sb_fields[type]);
1178         else
1179                 pr_buf(out, "(unknown field %u)", type);
1180
1181         pr_buf(out, " (size %llu):", vstruct_bytes(f));
1182
1183         if (ops && ops->to_text)
1184                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1185 }