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 (block_size > PAGE_SECTORS)
 265                 return "Bad block size";
 266
 267         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
 268                 return "Bad user UUID";
 269
 270         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le)))
 271                 return "Bad internal UUID";
 272
 273         if (!sb->nr_devices ||
 274             sb->nr_devices <= sb->dev_idx ||
 275             sb->nr_devices > BCH_SB_MEMBERS_MAX)
 276                 return "Bad number of member devices";
 277
 278         if (!BCH_SB_META_REPLICAS_WANT(sb) ||
 279             BCH_SB_META_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
 280                 return "Invalid number of metadata replicas";
 281
 282         if (!BCH_SB_META_REPLICAS_REQ(sb) ||
 283             BCH_SB_META_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
 284                 return "Invalid number of metadata replicas";
 285
 286         if (!BCH_SB_DATA_REPLICAS_WANT(sb) ||
 287             BCH_SB_DATA_REPLICAS_WANT(sb) > BCH_REPLICAS_MAX)
 288                 return "Invalid number of data replicas";
 289
 290         if (!BCH_SB_DATA_REPLICAS_REQ(sb) ||
 291             BCH_SB_DATA_REPLICAS_REQ(sb) > BCH_REPLICAS_MAX)
 292                 return "Invalid number of data replicas";
 293
 294         if (BCH_SB_META_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 295                 return "Invalid metadata checksum type";
 296
 297         if (BCH_SB_DATA_CSUM_TYPE(sb) >= BCH_CSUM_OPT_NR)
 298                 return "Invalid metadata checksum type";
 299
 300         if (BCH_SB_COMPRESSION_TYPE(sb) >= BCH_COMPRESSION_OPT_NR)
 301                 return "Invalid compression type";
 302
 303         if (!BCH_SB_BTREE_NODE_SIZE(sb))
 304                 return "Btree node size not set";
 305
 306         if (BCH_SB_GC_RESERVE(sb) < 5)
 307                 return "gc reserve percentage too small";
 308
 309         if (!sb->time_precision ||
 310             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC)
 311                 return "invalid time precision";
 312
 313         /* validate layout */
 314         err = validate_sb_layout(&sb->layout);
 315         if (err)
 316                 return err;
 317
 318         vstruct_for_each(sb, f) {
 319                 if (!f->u64s)
 320                         return "Invalid superblock: invalid optional field";
 321
 322                 if (vstruct_next(f) > vstruct_last(sb))
 323                         return "Invalid superblock: invalid optional field";
 324         }
 325
 326         /* members must be validated first: */
 327         mi = bch2_sb_get_members(sb);
 328         if (!mi)
 329                 return "Invalid superblock: member info area missing";
 330
 331         err = bch2_sb_field_validate(sb, &mi->field);
 332         if (err)
 333                 return err;
 334
 335         vstruct_for_each(sb, f) {
 336                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 337                         continue;
 338
 339                 err = bch2_sb_field_validate(sb, f);
 340                 if (err)
 341                         return err;
 342         }
 343
 344         return NULL;
 345 }
 346
 347 /* device open: */
 348
 349 static void bch2_sb_update(struct bch_fs *c)
 350 {
 351         struct bch_sb *src = c->disk_sb.sb;
 352         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 353         struct bch_dev *ca;
 354         unsigned i;
 355
 356         lockdep_assert_held(&c->sb_lock);
 357
 358         c->sb.uuid              = src->uuid;
 359         c->sb.user_uuid         = src->user_uuid;
 360         c->sb.version           = le16_to_cpu(src->version);
 361         c->sb.version_min       = le16_to_cpu(src->version_min);
 362         c->sb.nr_devices        = src->nr_devices;
 363         c->sb.clean             = BCH_SB_CLEAN(src);
 364         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 365
 366         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 367         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 368
 369         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 370         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 371                                           c->sb.nsec_per_time_unit);
 372         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 373
 374         c->sb.features          = le64_to_cpu(src->features[0]);
 375         c->sb.compat            = le64_to_cpu(src->compat[0]);
 376
 377         for_each_member_device(ca, c, i)
 378                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 379 }
 380
 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         if (BCH_SB_INITIALIZED(c->disk_sb.sb))
 438                 set_bit(BCH_FS_INITIALIZED, &c->flags);
 439
 440         ret = bch2_sb_replicas_to_cpu_replicas(c);
 441         if (ret)
 442                 return ret;
 443
 444         ret = bch2_sb_disk_groups_to_cpu(c);
 445         if (ret)
 446                 return ret;
 447
 448         bch2_sb_update(c);
 449         return 0;
 450 }
 451
 452 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 453 {
 454         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 455         struct bch_sb_field_journal *journal_buckets =
 456                 bch2_sb_get_journal(dst);
 457         unsigned journal_u64s = journal_buckets
 458                 ? le32_to_cpu(journal_buckets->field.u64s)
 459                 : 0;
 460         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 461         int ret;
 462
 463         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 464         if (ret)
 465                 return ret;
 466
 467         __copy_super(&ca->disk_sb, src);
 468         return 0;
 469 }
 470
 471 /* read superblock: */
 472
 473 static const char *read_one_super(struct bch_sb_handle *sb, u64 offset)
 474 {
 475         struct bch_csum csum;
 476         size_t bytes;
 477 reread:
 478         bio_reset(sb->bio);
 479         bio_set_dev(sb->bio, sb->bdev);
 480         sb->bio->bi_iter.bi_sector = offset;
 481         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 482         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 483
 484         if (submit_bio_wait(sb->bio))
 485                 return "IO error";
 486
 487         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
 488                 return "Not a bcachefs superblock";
 489
 490         if (le16_to_cpu(sb->sb->version) <  bcachefs_metadata_version_min ||
 491             le16_to_cpu(sb->sb->version) >= bcachefs_metadata_version_max)
 492                 return "Unsupported superblock version";
 493
 494         bytes = vstruct_bytes(sb->sb);
 495
 496         if (bytes > 512 << sb->sb->layout.sb_max_size_bits)
 497                 return "Bad superblock: too big";
 498
 499         if (bytes > sb->buffer_size) {
 500                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 501                         return "cannot allocate memory";
 502                 goto reread;
 503         }
 504
 505         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR)
 506                 return "unknown csum type";
 507
 508         /* XXX: verify MACs */
 509         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 510                             null_nonce(), sb->sb);
 511
 512         if (bch2_crc_cmp(csum, sb->sb->csum))
 513                 return "bad checksum reading superblock";
 514
 515         sb->seq = le64_to_cpu(sb->sb->seq);
 516
 517         return NULL;
 518 }
 519
 520 int bch2_read_super(const char *path, struct bch_opts *opts,
 521                     struct bch_sb_handle *sb)
 522 {
 523         u64 offset = opt_get(*opts, sb);
 524         struct bch_sb_layout layout;
 525         const char *err;
 526         __le64 *i;
 527         int ret;
 528
 529         pr_verbose_init(*opts, "");
 530
 531         memset(sb, 0, sizeof(*sb));
 532         sb->mode        = FMODE_READ;
 533         sb->have_bio    = true;
 534
 535         if (!opt_get(*opts, noexcl))
 536                 sb->mode |= FMODE_EXCL;
 537
 538         if (!opt_get(*opts, nochanges))
 539                 sb->mode |= FMODE_WRITE;
 540
 541         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 542         if (IS_ERR(sb->bdev) &&
 543             PTR_ERR(sb->bdev) == -EACCES &&
 544             opt_get(*opts, read_only)) {
 545                 sb->mode &= ~FMODE_WRITE;
 546
 547                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 548                 if (!IS_ERR(sb->bdev))
 549                         opt_set(*opts, nochanges, true);
 550         }
 551
 552         if (IS_ERR(sb->bdev)) {
 553                 ret = PTR_ERR(sb->bdev);
 554                 goto out;
 555         }
 556
 557         err = "cannot allocate memory";
 558         ret = bch2_sb_realloc(sb, 0);
 559         if (ret)
 560                 goto err;
 561
 562         ret = -EFAULT;
 563         err = "dynamic fault";
 564         if (bch2_fs_init_fault("read_super"))
 565                 goto err;
 566
 567         ret = -EINVAL;
 568         err = read_one_super(sb, offset);
 569         if (!err)
 570                 goto got_super;
 571
 572         if (opt_defined(*opts, sb))
 573                 goto err;
 574
 575         pr_err("error reading default superblock: %s", err);
 576
 577         /*
 578          * Error reading primary superblock - read location of backup
 579          * superblocks:
 580          */
 581         bio_reset(sb->bio);
 582         bio_set_dev(sb->bio, sb->bdev);
 583         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 584         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 585         /*
 586          * use sb buffer to read layout, since sb buffer is page aligned but
 587          * layout won't be:
 588          */
 589         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 590
 591         err = "IO error";
 592         if (submit_bio_wait(sb->bio))
 593                 goto err;
 594
 595         memcpy(&layout, sb->sb, sizeof(layout));
 596         err = validate_sb_layout(&layout);
 597         if (err)
 598                 goto err;
 599
 600         for (i = layout.sb_offset;
 601              i < layout.sb_offset + layout.nr_superblocks; i++) {
 602                 offset = le64_to_cpu(*i);
 603
 604                 if (offset == opt_get(*opts, sb))
 605                         continue;
 606
 607                 err = read_one_super(sb, offset);
 608                 if (!err)
 609                         goto got_super;
 610         }
 611
 612         ret = -EINVAL;
 613         goto err;
 614
 615 got_super:
 616         err = "Superblock block size smaller than device block size";
 617         ret = -EINVAL;
 618         if (le16_to_cpu(sb->sb->block_size) << 9 <
 619             bdev_logical_block_size(sb->bdev)) {
 620                 pr_err("error reading superblock: Superblock block size (%u) smaller than device block size (%u)",
 621                        le16_to_cpu(sb->sb->block_size) << 9,
 622                        bdev_logical_block_size(sb->bdev));
 623                 goto err_no_print;
 624         }
 625
 626         ret = 0;
 627         sb->have_layout = true;
 628 out:
 629         pr_verbose_init(*opts, "ret %i", ret);
 630         return ret;
 631 err:
 632         pr_err("error reading superblock: %s", err);
 633 err_no_print:
 634         bch2_free_super(sb);
 635         goto out;
 636 }
 637
 638 /* write superblock: */
 639
 640 static void write_super_endio(struct bio *bio)
 641 {
 642         struct bch_dev *ca = bio->bi_private;
 643
 644         /* XXX: return errors directly */
 645
 646         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 647                                bch2_blk_status_to_str(bio->bi_status)))
 648                 ca->sb_write_error = 1;
 649
 650         closure_put(&ca->fs->sb_write);
 651         percpu_ref_put(&ca->io_ref);
 652 }
 653
 654 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 655 {
 656         struct bch_sb *sb = ca->disk_sb.sb;
 657         struct bio *bio = ca->disk_sb.bio;
 658
 659         bio_reset(bio);
 660         bio_set_dev(bio, ca->disk_sb.bdev);
 661         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 662         bio->bi_end_io          = write_super_endio;
 663         bio->bi_private         = ca;
 664         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 665         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 666
 667         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 668                      bio_sectors(bio));
 669
 670         percpu_ref_get(&ca->io_ref);
 671         closure_bio_submit(bio, &c->sb_write);
 672 }
 673
 674 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 675 {
 676         struct bch_sb *sb = ca->disk_sb.sb;
 677         struct bio *bio = ca->disk_sb.bio;
 678
 679         sb->offset = sb->layout.sb_offset[idx];
 680
 681         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 682         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 683                                 null_nonce(), sb);
 684
 685         bio_reset(bio);
 686         bio_set_dev(bio, ca->disk_sb.bdev);
 687         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 688         bio->bi_end_io          = write_super_endio;
 689         bio->bi_private         = ca;
 690         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 691         bch2_bio_map(bio, sb,
 692                      roundup((size_t) vstruct_bytes(sb),
 693                              bdev_logical_block_size(ca->disk_sb.bdev)));
 694
 695         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 696                      bio_sectors(bio));
 697
 698         percpu_ref_get(&ca->io_ref);
 699         closure_bio_submit(bio, &c->sb_write);
 700 }
 701
 702 int bch2_write_super(struct bch_fs *c)
 703 {
 704         struct closure *cl = &c->sb_write;
 705         struct bch_dev *ca;
 706         unsigned i, sb = 0, nr_wrote;
 707         const char *err;
 708         struct bch_devs_mask sb_written;
 709         bool wrote, can_mount_without_written, can_mount_with_written;
 710         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 711         int ret = 0;
 712
 713         if (c->opts.very_degraded)
 714                 degraded_flags |= BCH_FORCE_IF_LOST;
 715
 716         lockdep_assert_held(&c->sb_lock);
 717
 718         closure_init_stack(cl);
 719         memset(&sb_written, 0, sizeof(sb_written));
 720
 721         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 722
 723         if (test_bit(BCH_FS_ERROR, &c->flags))
 724                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 725         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 726                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 727
 728         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 729
 730         for_each_online_member(ca, c, i)
 731                 bch2_sb_from_fs(c, ca);
 732
 733         for_each_online_member(ca, c, i) {
 734                 err = bch2_sb_validate(&ca->disk_sb);
 735                 if (err) {
 736                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err);
 737                         ret = -1;
 738                         goto out;
 739                 }
 740         }
 741
 742         if (c->opts.nochanges)
 743                 goto out;
 744
 745         for_each_online_member(ca, c, i) {
 746                 __set_bit(ca->dev_idx, sb_written.d);
 747                 ca->sb_write_error = 0;
 748         }
 749
 750         for_each_online_member(ca, c, i)
 751                 read_back_super(c, ca);
 752         closure_sync(cl);
 753
 754         for_each_online_member(ca, c, i) {
 755                 if (!ca->sb_write_error &&
 756                     ca->disk_sb.seq !=
 757                     le64_to_cpu(ca->sb_read_scratch->seq)) {
 758                         bch2_fs_fatal_error(c,
 759                                 "Superblock modified by another process");
 760                         percpu_ref_put(&ca->io_ref);
 761                         ret = -EROFS;
 762                         goto out;
 763                 }
 764         }
 765
 766         do {
 767                 wrote = false;
 768                 for_each_online_member(ca, c, i)
 769                         if (!ca->sb_write_error &&
 770                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 771                                 write_one_super(c, ca, sb);
 772                                 wrote = true;
 773                         }
 774                 closure_sync(cl);
 775                 sb++;
 776         } while (wrote);
 777
 778         for_each_online_member(ca, c, i) {
 779                 if (ca->sb_write_error)
 780                         __clear_bit(ca->dev_idx, sb_written.d);
 781                 else
 782                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 783         }
 784
 785         nr_wrote = dev_mask_nr(&sb_written);
 786
 787         can_mount_with_written =
 788                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 789
 790         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 791                 sb_written.d[i] = ~sb_written.d[i];
 792
 793         can_mount_without_written =
 794                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 795
 796         /*
 797          * If we would be able to mount _without_ the devices we successfully
 798          * wrote superblocks to, we weren't able to write to enough devices:
 799          *
 800          * Exception: if we can mount without the successes because we haven't
 801          * written anything (new filesystem), we continue if we'd be able to
 802          * mount with the devices we did successfully write to:
 803          */
 804         if (bch2_fs_fatal_err_on(!nr_wrote ||
 805                                  !can_mount_with_written ||
 806                                  (can_mount_without_written &&
 807                                   !can_mount_with_written), c,
 808                 "Unable to write superblock to sufficient devices (from %ps)",
 809                 (void *) _RET_IP_))
 810                 ret = -1;
 811 out:
 812         /* Make new options visible after they're persistent: */
 813         bch2_sb_update(c);
 814         return ret;
 815 }
 816
 817 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 818 {
 819         mutex_lock(&c->sb_lock);
 820         if (!(c->sb.features & (1ULL << feat))) {
 821                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 822
 823                 bch2_write_super(c);
 824         }
 825         mutex_unlock(&c->sb_lock);
 826 }
 827
 828 /* BCH_SB_FIELD_journal: */
 829
 830 static int u64_cmp(const void *_l, const void *_r)
 831 {
 832         u64 l = *((const u64 *) _l), r = *((const u64 *) _r);
 833
 834         return l < r ? -1 : l > r ? 1 : 0;
 835 }
 836
 837 static const char *bch2_sb_validate_journal(struct bch_sb *sb,
 838                                             struct bch_sb_field *f)
 839 {
 840         struct bch_sb_field_journal *journal = field_to_type(f, journal);
 841         struct bch_member *m = bch2_sb_get_members(sb)->members + sb->dev_idx;
 842         const char *err;
 843         unsigned nr;
 844         unsigned i;
 845         u64 *b;
 846
 847         journal = bch2_sb_get_journal(sb);
 848         if (!journal)
 849                 return NULL;
 850
 851         nr = bch2_nr_journal_buckets(journal);
 852         if (!nr)
 853                 return NULL;
 854
 855         b = kmalloc_array(sizeof(u64), nr, GFP_KERNEL);
 856         if (!b)
 857                 return "cannot allocate memory";
 858
 859         for (i = 0; i < nr; i++)
 860                 b[i] = le64_to_cpu(journal->buckets[i]);
 861
 862         sort(b, nr, sizeof(u64), u64_cmp, NULL);
 863
 864         err = "journal bucket at sector 0";
 865         if (!b[0])
 866                 goto err;
 867
 868         err = "journal bucket before first bucket";
 869         if (m && b[0] < le16_to_cpu(m->first_bucket))
 870                 goto err;
 871
 872         err = "journal bucket past end of device";
 873         if (m && b[nr - 1] >= le64_to_cpu(m->nbuckets))
 874                 goto err;
 875
 876         err = "duplicate journal buckets";
 877         for (i = 0; i + 1 < nr; i++)
 878                 if (b[i] == b[i + 1])
 879                         goto err;
 880
 881         err = NULL;
 882 err:
 883         kfree(b);
 884         return err;
 885 }
 886
 887 static const struct bch_sb_field_ops bch_sb_field_ops_journal = {
 888         .validate       = bch2_sb_validate_journal,
 889 };
 890
 891 /* BCH_SB_FIELD_members: */
 892
 893 static const char *bch2_sb_validate_members(struct bch_sb *sb,
 894                                             struct bch_sb_field *f)
 895 {
 896         struct bch_sb_field_members *mi = field_to_type(f, members);
 897         struct bch_member *m;
 898
 899         if ((void *) (mi->members + sb->nr_devices) >
 900             vstruct_end(&mi->field))
 901                 return "Invalid superblock: bad member info";
 902
 903         for (m = mi->members;
 904              m < mi->members + sb->nr_devices;
 905              m++) {
 906                 if (!bch2_member_exists(m))
 907                         continue;
 908
 909                 if (le64_to_cpu(m->nbuckets) > LONG_MAX)
 910                         return "Too many buckets";
 911
 912                 if (le64_to_cpu(m->nbuckets) -
 913                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS)
 914                         return "Not enough buckets";
 915
 916                 if (le16_to_cpu(m->bucket_size) <
 917                     le16_to_cpu(sb->block_size))
 918                         return "bucket size smaller than block size";
 919
 920                 if (le16_to_cpu(m->bucket_size) <
 921                     BCH_SB_BTREE_NODE_SIZE(sb))
 922                         return "bucket size smaller than btree node size";
 923         }
 924
 925         return NULL;
 926 }
 927
 928 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
 929         .validate       = bch2_sb_validate_members,
 930 };
 931
 932 /* BCH_SB_FIELD_crypt: */
 933
 934 static const char *bch2_sb_validate_crypt(struct bch_sb *sb,
 935                                           struct bch_sb_field *f)
 936 {
 937         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
 938
 939         if (vstruct_bytes(&crypt->field) != sizeof(*crypt))
 940                 return "invalid field crypt: wrong size";
 941
 942         if (BCH_CRYPT_KDF_TYPE(crypt))
 943                 return "invalid field crypt: bad kdf type";
 944
 945         return NULL;
 946 }
 947
 948 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
 949         .validate       = bch2_sb_validate_crypt,
 950 };
 951
 952 /* BCH_SB_FIELD_clean: */
 953
 954 int bch2_sb_clean_validate(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
 955 {
 956         struct jset_entry *entry;
 957         int ret;
 958
 959         for (entry = clean->start;
 960              entry < (struct jset_entry *) vstruct_end(&clean->field);
 961              entry = vstruct_next(entry)) {
 962                 ret = bch2_journal_entry_validate(c, "superblock", entry,
 963                                                   le16_to_cpu(c->disk_sb.sb->version),
 964                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
 965                                                   write);
 966                 if (ret)
 967                         return ret;
 968         }
 969
 970         return 0;
 971 }
 972
 973 int bch2_fs_mark_dirty(struct bch_fs *c)
 974 {
 975         int ret;
 976
 977         /*
 978          * Unconditionally write superblock, to verify it hasn't changed before
 979          * we go rw:
 980          */
 981
 982         mutex_lock(&c->sb_lock);
 983         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
 984         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
 985         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
 986         ret = bch2_write_super(c);
 987         mutex_unlock(&c->sb_lock);
 988
 989         return ret;
 990 }
 991
 992 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
 993 {
 994         struct jset_entry *entry = *end;
 995         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
 996
 997         memset(entry, 0, u64s * sizeof(u64));
 998         /*
 999          * The u64s field counts from the start of data, ignoring the shared
1000          * fields.
1001          */
1002         entry->u64s = cpu_to_le16(u64s - 1);
1003
1004         *end = vstruct_next(*end);
1005         return entry;
1006 }
1007
1008 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1009                                            struct jset_entry **end,
1010                                            u64 journal_seq)
1011 {
1012         struct bch_dev *ca;
1013         unsigned i, dev;
1014
1015         percpu_down_read(&c->mark_lock);
1016
1017         if (!journal_seq) {
1018                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1019                         bch2_fs_usage_acc_to_base(c, i);
1020         } else {
1021                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1022         }
1023
1024         {
1025                 struct jset_entry_usage *u =
1026                         container_of(jset_entry_init(end, sizeof(*u)),
1027                                      struct jset_entry_usage, entry);
1028
1029                 u->entry.type   = BCH_JSET_ENTRY_usage;
1030                 u->entry.btree_id = FS_USAGE_INODES;
1031                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1032         }
1033
1034         {
1035                 struct jset_entry_usage *u =
1036                         container_of(jset_entry_init(end, sizeof(*u)),
1037                                      struct jset_entry_usage, entry);
1038
1039                 u->entry.type   = BCH_JSET_ENTRY_usage;
1040                 u->entry.btree_id = FS_USAGE_KEY_VERSION;
1041                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1042         }
1043
1044         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1045                 struct jset_entry_usage *u =
1046                         container_of(jset_entry_init(end, sizeof(*u)),
1047                                      struct jset_entry_usage, entry);
1048
1049                 u->entry.type   = BCH_JSET_ENTRY_usage;
1050                 u->entry.btree_id = FS_USAGE_RESERVED;
1051                 u->entry.level  = i;
1052                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1053         }
1054
1055         for (i = 0; i < c->replicas.nr; i++) {
1056                 struct bch_replicas_entry *e =
1057                         cpu_replicas_entry(&c->replicas, i);
1058                 struct jset_entry_data_usage *u =
1059                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1060                                      struct jset_entry_data_usage, entry);
1061
1062                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1063                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1064                 memcpy(&u->r, e, replicas_entry_bytes(e));
1065         }
1066
1067         for_each_member_device(ca, c, dev) {
1068                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1069                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1070                 struct jset_entry_dev_usage *u =
1071                         container_of(jset_entry_init(end, b),
1072                                      struct jset_entry_dev_usage, entry);
1073
1074                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1075                 u->dev = cpu_to_le32(dev);
1076                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1077                 u->buckets_unavailable  = cpu_to_le64(ca->usage_base->buckets_unavailable);
1078
1079                 for (i = 0; i < BCH_DATA_NR; i++) {
1080                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1081                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1082                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1083                 }
1084         }
1085
1086         percpu_up_read(&c->mark_lock);
1087
1088         for (i = 0; i < 2; i++) {
1089                 struct jset_entry_clock *clock =
1090                         container_of(jset_entry_init(end, sizeof(*clock)),
1091                                      struct jset_entry_clock, entry);
1092
1093                 clock->entry.type = BCH_JSET_ENTRY_clock;
1094                 clock->rw       = i;
1095                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1096         }
1097 }
1098
1099 void bch2_fs_mark_clean(struct bch_fs *c)
1100 {
1101         struct bch_sb_field_clean *sb_clean;
1102         struct jset_entry *entry;
1103         unsigned u64s;
1104         int ret;
1105
1106         mutex_lock(&c->sb_lock);
1107         if (BCH_SB_CLEAN(c->disk_sb.sb))
1108                 goto out;
1109
1110         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1111
1112         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1113         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1114         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1115         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1116
1117         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1118
1119         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1120         if (!sb_clean) {
1121                 bch_err(c, "error resizing superblock while setting filesystem clean");
1122                 goto out;
1123         }
1124
1125         sb_clean->flags         = 0;
1126         sb_clean->journal_seq   = cpu_to_le64(journal_cur_seq(&c->journal) - 1);
1127
1128         /* Trying to catch outstanding bug: */
1129         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1130
1131         entry = sb_clean->start;
1132         bch2_journal_super_entries_add_common(c, &entry, 0);
1133         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1134         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1135
1136         memset(entry, 0,
1137                vstruct_end(&sb_clean->field) - (void *) entry);
1138
1139         /*
1140          * this should be in the write path, and we should be validating every
1141          * superblock section:
1142          */
1143         ret = bch2_sb_clean_validate(c, sb_clean, WRITE);
1144         if (ret) {
1145                 bch_err(c, "error writing marking filesystem clean: validate error");
1146                 goto out;
1147         }
1148
1149         bch2_write_super(c);
1150 out:
1151         mutex_unlock(&c->sb_lock);
1152 }
1153
1154 static const char *bch2_sb_validate_clean(struct bch_sb *sb,
1155                                           struct bch_sb_field *f)
1156 {
1157         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1158
1159         if (vstruct_bytes(&clean->field) < sizeof(*clean))
1160                 return "invalid field crypt: wrong size";
1161
1162         return NULL;
1163 }
1164
1165 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1166         .validate       = bch2_sb_validate_clean,
1167 };
1168
1169 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1170 #define x(f, nr)                                        \
1171         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1172         BCH_SB_FIELDS()
1173 #undef x
1174 };
1175
1176 static const char *bch2_sb_field_validate(struct bch_sb *sb,
1177                                           struct bch_sb_field *f)
1178 {
1179         unsigned type = le32_to_cpu(f->type);
1180
1181         return type < BCH_SB_FIELD_NR
1182                 ? bch2_sb_field_ops[type]->validate(sb, f)
1183                 : NULL;
1184 }
1185
1186 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1187                            struct bch_sb_field *f)
1188 {
1189         unsigned type = le32_to_cpu(f->type);
1190         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1191                 ? bch2_sb_field_ops[type] : NULL;
1192
1193         if (ops)
1194                 pr_buf(out, "%s", bch2_sb_fields[type]);
1195         else
1196                 pr_buf(out, "(unknown field %u)", type);
1197
1198         pr_buf(out, " (size %llu):", vstruct_bytes(f));
1199
1200         if (ops && ops->to_text)
1201                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1202 }