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