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_sb.h"
  14 #include "journal_seq_blacklist.h"
  15 #include "replicas.h"
  16 #include "quota.h"
  17 #include "super-io.h"
  18 #include "super.h"
  19 #include "vstructs.h"
  20
  21 #include <linux/backing-dev.h>
  22 #include <linux/sort.h>
  23
  24 const char * const bch2_sb_fields[] = {
  25 #define x(name, nr)     #name,
  26         BCH_SB_FIELDS()
  27 #undef x
  28         NULL
  29 };
  30
  31 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
  32                                   struct printbuf *);
  33
  34 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  35                                       enum bch_sb_field_type type)
  36 {
  37         struct bch_sb_field *f;
  38
  39         /* XXX: need locking around superblock to access optional fields */
  40
  41         vstruct_for_each(sb, f)
  42                 if (le32_to_cpu(f->type) == type)
  43                         return f;
  44         return NULL;
  45 }
  46
  47 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  48                                                    struct bch_sb_field *f,
  49                                                    unsigned u64s)
  50 {
  51         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  52         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  53
  54         BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
  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         kfree(sb->sb);
 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         size_t new_buffer_size;
 112         struct bch_sb *new_sb;
 113         struct bio *bio;
 114
 115         if (sb->bdev)
 116                 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
 117
 118         new_buffer_size = roundup_pow_of_two(new_bytes);
 119
 120         if (sb->sb && sb->buffer_size >= new_buffer_size)
 121                 return 0;
 122
 123         if (sb->have_layout) {
 124                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 125
 126                 if (new_bytes > max_bytes) {
 127                         char buf[BDEVNAME_SIZE];
 128
 129                         pr_err("%s: superblock too big: want %zu but have %llu",
 130                                bdevname(sb->bdev, buf), new_bytes, max_bytes);
 131                         return -ENOSPC;
 132                 }
 133         }
 134
 135         if (sb->buffer_size >= new_buffer_size && sb->sb)
 136                 return 0;
 137
 138         if (dynamic_fault("bcachefs:add:super_realloc"))
 139                 return -ENOMEM;
 140
 141         if (sb->have_bio) {
 142                 bio = bio_kmalloc(GFP_KERNEL,
 143                         DIV_ROUND_UP(new_buffer_size, PAGE_SIZE));
 144                 if (!bio)
 145                         return -ENOMEM;
 146
 147                 if (sb->bio)
 148                         bio_put(sb->bio);
 149                 sb->bio = bio;
 150         }
 151
 152         new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 153         if (!new_sb)
 154                 return -ENOMEM;
 155
 156         sb->sb = new_sb;
 157         sb->buffer_size = new_buffer_size;
 158
 159         return 0;
 160 }
 161
 162 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 163                                           enum bch_sb_field_type type,
 164                                           unsigned u64s)
 165 {
 166         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 167         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 168         ssize_t d = -old_u64s + u64s;
 169
 170         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 171                 return NULL;
 172
 173         if (sb->fs_sb) {
 174                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 175                 struct bch_dev *ca;
 176                 unsigned i;
 177
 178                 lockdep_assert_held(&c->sb_lock);
 179
 180                 /* XXX: we're not checking that offline device have enough space */
 181
 182                 for_each_online_member(ca, c, i) {
 183                         struct bch_sb_handle *sb = &ca->disk_sb;
 184
 185                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 186                                 percpu_ref_put(&ca->ref);
 187                                 return NULL;
 188                         }
 189                 }
 190         }
 191
 192         f = bch2_sb_field_get(sb->sb, type);
 193         f = __bch2_sb_field_resize(sb, f, u64s);
 194         if (f)
 195                 f->type = cpu_to_le32(type);
 196         return f;
 197 }
 198
 199 /* Superblock validate: */
 200
 201 static inline void __bch2_sb_layout_size_assert(void)
 202 {
 203         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 204 }
 205
 206 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
 207 {
 208         u64 offset, prev_offset, max_sectors;
 209         unsigned i;
 210
 211         if (uuid_le_cmp(layout->magic, BCACHE_MAGIC)) {
 212                 pr_buf(out, "Not a bcachefs superblock layout");
 213                 return -EINVAL;
 214         }
 215
 216         if (layout->layout_type != 0) {
 217                 pr_buf(out, "Invalid superblock layout type %u",
 218                        layout->layout_type);
 219                 return -EINVAL;
 220         }
 221
 222         if (!layout->nr_superblocks) {
 223                 pr_buf(out, "Invalid superblock layout: no superblocks");
 224                 return -EINVAL;
 225         }
 226
 227         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
 228                 pr_buf(out, "Invalid superblock layout: too many superblocks");
 229                 return -EINVAL;
 230         }
 231
 232         max_sectors = 1 << layout->sb_max_size_bits;
 233
 234         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 235
 236         for (i = 1; i < layout->nr_superblocks; i++) {
 237                 offset = le64_to_cpu(layout->sb_offset[i]);
 238
 239                 if (offset < prev_offset + max_sectors) {
 240                         pr_buf(out, "Invalid superblock layout: superblocks overlap\n"
 241                                "  (sb %u ends at %llu next starts at %llu",
 242                                i - 1, prev_offset + max_sectors, offset);
 243                         return -EINVAL;
 244                 }
 245                 prev_offset = offset;
 246         }
 247
 248         return 0;
 249 }
 250
 251 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 252                             int rw)
 253 {
 254         struct bch_sb *sb = disk_sb->sb;
 255         struct bch_sb_field *f;
 256         struct bch_sb_field_members *mi;
 257         enum bch_opt_id opt_id;
 258         u32 version, version_min;
 259         u16 block_size;
 260         int ret;
 261
 262         version         = le16_to_cpu(sb->version);
 263         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 264                 ? le16_to_cpu(sb->version_min)
 265                 : version;
 266
 267         if (version    >= bcachefs_metadata_version_max) {
 268                 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
 269                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 270                 return -EINVAL;
 271         }
 272
 273         if (version_min < bcachefs_metadata_version_min) {
 274                 pr_buf(out, "Unsupported superblock version %u (min %u, max %u)",
 275                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 276                 return -EINVAL;
 277         }
 278
 279         if (version_min > version) {
 280                 pr_buf(out, "Bad minimum version %u, greater than version field %u",
 281                        version_min, version);
 282                 return -EINVAL;
 283         }
 284
 285         if (sb->features[1] ||
 286             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
 287                 pr_buf(out, "Filesystem has incompatible features");
 288                 return -EINVAL;
 289         }
 290
 291         block_size = le16_to_cpu(sb->block_size);
 292
 293         if (block_size > PAGE_SECTORS) {
 294                 pr_buf(out, "Block size too big (got %u, max %u)",
 295                        block_size, PAGE_SECTORS);
 296                 return -EINVAL;
 297         }
 298
 299         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
 300                 pr_buf(out, "Bad user UUID (got zeroes)");
 301                 return -EINVAL;
 302         }
 303
 304         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
 305                 pr_buf(out, "Bad intenal UUID (got zeroes)");
 306                 return -EINVAL;
 307         }
 308
 309         if (!sb->nr_devices ||
 310             sb->nr_devices > BCH_SB_MEMBERS_MAX) {
 311                 pr_buf(out, "Bad number of member devices %u (max %u)",
 312                        sb->nr_devices, BCH_SB_MEMBERS_MAX);
 313                 return -EINVAL;
 314         }
 315
 316         if (sb->dev_idx >= sb->nr_devices) {
 317                 pr_buf(out, "Bad dev_idx (got %u, nr_devices %u)",
 318                        sb->dev_idx, sb->nr_devices);
 319                 return -EINVAL;
 320         }
 321
 322         if (!sb->time_precision ||
 323             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
 324                 pr_buf(out, "Invalid time precision: %u (min 1, max %lu)",
 325                        le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
 326                 return -EINVAL;
 327         }
 328
 329         if (rw == READ) {
 330                 /*
 331                  * Been seeing a bug where these are getting inexplicably
 332                  * zeroed, so we'r now validating them, but we have to be
 333                  * careful not to preven people's filesystems from mounting:
 334                  */
 335                 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 336                         SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 337                 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 338                         SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
 339         }
 340
 341         for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
 342                 const struct bch_option *opt = bch2_opt_table + opt_id;
 343
 344                 if (opt->get_sb != BCH2_NO_SB_OPT) {
 345                         u64 v = bch2_opt_from_sb(sb, opt_id);
 346
 347                         pr_buf(out, "Invalid option ");
 348                         ret = bch2_opt_validate(opt, v, out);
 349                         if (ret)
 350                                 return ret;
 351
 352                         printbuf_reset(out);
 353                 }
 354         }
 355
 356         /* validate layout */
 357         ret = validate_sb_layout(&sb->layout, out);
 358         if (ret)
 359                 return ret;
 360
 361         vstruct_for_each(sb, f) {
 362                 if (!f->u64s) {
 363                         pr_buf(out, "Invalid superblock: optional with size 0 (type %u)",
 364                                le32_to_cpu(f->type));
 365                         return -EINVAL;
 366                 }
 367
 368                 if (vstruct_next(f) > vstruct_last(sb)) {
 369                         pr_buf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
 370                                le32_to_cpu(f->type));
 371                         return -EINVAL;
 372                 }
 373         }
 374
 375         /* members must be validated first: */
 376         mi = bch2_sb_get_members(sb);
 377         if (!mi) {
 378                 pr_buf(out, "Invalid superblock: member info area missing");
 379                 return -EINVAL;
 380         }
 381
 382         ret = bch2_sb_field_validate(sb, &mi->field, out);
 383         if (ret)
 384                 return ret;
 385
 386         vstruct_for_each(sb, f) {
 387                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 388                         continue;
 389
 390                 ret = bch2_sb_field_validate(sb, f, out);
 391                 if (ret)
 392                         return ret;
 393         }
 394
 395         return 0;
 396 }
 397
 398 /* device open: */
 399
 400 static void bch2_sb_update(struct bch_fs *c)
 401 {
 402         struct bch_sb *src = c->disk_sb.sb;
 403         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 404         struct bch_dev *ca;
 405         unsigned i;
 406
 407         lockdep_assert_held(&c->sb_lock);
 408
 409         c->sb.uuid              = src->uuid;
 410         c->sb.user_uuid         = src->user_uuid;
 411         c->sb.version           = le16_to_cpu(src->version);
 412         c->sb.version_min       = le16_to_cpu(src->version_min);
 413         c->sb.nr_devices        = src->nr_devices;
 414         c->sb.clean             = BCH_SB_CLEAN(src);
 415         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 416
 417         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 418         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 419
 420         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 421         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 422                                           c->sb.nsec_per_time_unit);
 423         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 424
 425         c->sb.features          = le64_to_cpu(src->features[0]);
 426         c->sb.compat            = le64_to_cpu(src->compat[0]);
 427
 428         for_each_member_device(ca, c, i)
 429                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 430 }
 431
 432 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 433 {
 434         struct bch_sb_field *src_f, *dst_f;
 435         struct bch_sb *dst = dst_handle->sb;
 436         unsigned i;
 437
 438         dst->version            = src->version;
 439         dst->version_min        = src->version_min;
 440         dst->seq                = src->seq;
 441         dst->uuid               = src->uuid;
 442         dst->user_uuid          = src->user_uuid;
 443         memcpy(dst->label,      src->label, sizeof(dst->label));
 444
 445         dst->block_size         = src->block_size;
 446         dst->nr_devices         = src->nr_devices;
 447
 448         dst->time_base_lo       = src->time_base_lo;
 449         dst->time_base_hi       = src->time_base_hi;
 450         dst->time_precision     = src->time_precision;
 451
 452         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 453         memcpy(dst->features,   src->features,  sizeof(dst->features));
 454         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 455
 456         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 457                 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 458                         continue;
 459
 460                 src_f = bch2_sb_field_get(src, i);
 461                 dst_f = bch2_sb_field_get(dst, i);
 462                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 463                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 464
 465                 if (src_f)
 466                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 467         }
 468 }
 469
 470 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 471 {
 472         struct bch_sb_field_journal *journal_buckets =
 473                 bch2_sb_get_journal(src);
 474         unsigned journal_u64s = journal_buckets
 475                 ? le32_to_cpu(journal_buckets->field.u64s)
 476                 : 0;
 477         int ret;
 478
 479         lockdep_assert_held(&c->sb_lock);
 480
 481         ret = bch2_sb_realloc(&c->disk_sb,
 482                               le32_to_cpu(src->u64s) - journal_u64s);
 483         if (ret)
 484                 return ret;
 485
 486         __copy_super(&c->disk_sb, src);
 487
 488         ret = bch2_sb_replicas_to_cpu_replicas(c);
 489         if (ret)
 490                 return ret;
 491
 492         ret = bch2_sb_disk_groups_to_cpu(c);
 493         if (ret)
 494                 return ret;
 495
 496         bch2_sb_update(c);
 497         return 0;
 498 }
 499
 500 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 501 {
 502         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 503         struct bch_sb_field_journal *journal_buckets =
 504                 bch2_sb_get_journal(dst);
 505         unsigned journal_u64s = journal_buckets
 506                 ? le32_to_cpu(journal_buckets->field.u64s)
 507                 : 0;
 508         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 509         int ret;
 510
 511         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 512         if (ret)
 513                 return ret;
 514
 515         __copy_super(&ca->disk_sb, src);
 516         return 0;
 517 }
 518
 519 /* read superblock: */
 520
 521 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 522 {
 523         struct bch_csum csum;
 524         u32 version, version_min;
 525         size_t bytes;
 526         int ret;
 527 reread:
 528         bio_reset(sb->bio);
 529         bio_set_dev(sb->bio, sb->bdev);
 530         sb->bio->bi_iter.bi_sector = offset;
 531         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 532         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 533
 534         ret = submit_bio_wait(sb->bio);
 535         if (ret) {
 536                 pr_buf(err, "IO error: %i", ret);
 537                 return ret;
 538         }
 539
 540         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC)) {
 541                 pr_buf(err, "Not a bcachefs superblock");
 542                 return -EINVAL;
 543         }
 544
 545         version         = le16_to_cpu(sb->sb->version);
 546         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 547                 ? le16_to_cpu(sb->sb->version_min)
 548                 : version;
 549
 550         if (version    >= bcachefs_metadata_version_max) {
 551                 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
 552                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 553                 return -EINVAL;
 554         }
 555
 556         if (version_min < bcachefs_metadata_version_min) {
 557                 pr_buf(err, "Unsupported superblock version %u (min %u, max %u)",
 558                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 559                 return -EINVAL;
 560         }
 561
 562         bytes = vstruct_bytes(sb->sb);
 563
 564         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 565                 pr_buf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 566                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 567                 return -EINVAL;
 568         }
 569
 570         if (bytes > sb->buffer_size) {
 571                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 572                         return -ENOMEM;
 573                 goto reread;
 574         }
 575
 576         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
 577                 pr_buf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 578                 return -EINVAL;
 579         }
 580
 581         /* XXX: verify MACs */
 582         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 583                             null_nonce(), sb->sb);
 584
 585         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 586                 pr_buf(err, "bad checksum");
 587                 return -EINVAL;
 588         }
 589
 590         sb->seq = le64_to_cpu(sb->sb->seq);
 591
 592         return 0;
 593 }
 594
 595 int bch2_read_super(const char *path, struct bch_opts *opts,
 596                     struct bch_sb_handle *sb)
 597 {
 598         u64 offset = opt_get(*opts, sb);
 599         struct bch_sb_layout layout;
 600         struct printbuf err = PRINTBUF;
 601         __le64 *i;
 602         int ret;
 603
 604         pr_verbose_init(*opts, "");
 605
 606         memset(sb, 0, sizeof(*sb));
 607         sb->mode        = FMODE_READ;
 608         sb->have_bio    = true;
 609
 610         if (!opt_get(*opts, noexcl))
 611                 sb->mode |= FMODE_EXCL;
 612
 613         if (!opt_get(*opts, nochanges))
 614                 sb->mode |= FMODE_WRITE;
 615
 616         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 617         if (IS_ERR(sb->bdev) &&
 618             PTR_ERR(sb->bdev) == -EACCES &&
 619             opt_get(*opts, read_only)) {
 620                 sb->mode &= ~FMODE_WRITE;
 621
 622                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 623                 if (!IS_ERR(sb->bdev))
 624                         opt_set(*opts, nochanges, true);
 625         }
 626
 627         if (IS_ERR(sb->bdev)) {
 628                 ret = PTR_ERR(sb->bdev);
 629                 goto out;
 630         }
 631
 632         ret = bch2_sb_realloc(sb, 0);
 633         if (ret) {
 634                 pr_buf(&err, "error allocating memory for superblock");
 635                 goto err;
 636         }
 637
 638         if (bch2_fs_init_fault("read_super")) {
 639                 pr_buf(&err, "dynamic fault");
 640                 ret = -EFAULT;
 641                 goto err;
 642         }
 643
 644         ret = read_one_super(sb, offset, &err);
 645         if (!ret)
 646                 goto got_super;
 647
 648         if (opt_defined(*opts, sb))
 649                 goto err;
 650
 651         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 652                path, err.buf);
 653         printbuf_reset(&err);
 654
 655         /*
 656          * Error reading primary superblock - read location of backup
 657          * superblocks:
 658          */
 659         bio_reset(sb->bio);
 660         bio_set_dev(sb->bio, sb->bdev);
 661         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 662         bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 663         /*
 664          * use sb buffer to read layout, since sb buffer is page aligned but
 665          * layout won't be:
 666          */
 667         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 668
 669         ret = submit_bio_wait(sb->bio);
 670         if (ret) {
 671                 pr_buf(&err, "IO error: %i", ret);
 672                 goto err;
 673         }
 674
 675         memcpy(&layout, sb->sb, sizeof(layout));
 676         ret = validate_sb_layout(&layout, &err);
 677         if (ret)
 678                 goto err;
 679
 680         for (i = layout.sb_offset;
 681              i < layout.sb_offset + layout.nr_superblocks; i++) {
 682                 offset = le64_to_cpu(*i);
 683
 684                 if (offset == opt_get(*opts, sb))
 685                         continue;
 686
 687                 ret = read_one_super(sb, offset, &err);
 688                 if (!ret)
 689                         goto got_super;
 690         }
 691
 692         goto err;
 693
 694 got_super:
 695         if (le16_to_cpu(sb->sb->block_size) << 9 <
 696             bdev_logical_block_size(sb->bdev)) {
 697                 pr_buf(&err, "block size (%u) smaller than device block size (%u)",
 698                        le16_to_cpu(sb->sb->block_size) << 9,
 699                        bdev_logical_block_size(sb->bdev));
 700                 ret = -EINVAL;
 701                 goto err;
 702         }
 703
 704         ret = 0;
 705         sb->have_layout = true;
 706
 707         ret = bch2_sb_validate(sb, &err, READ);
 708         if (ret) {
 709                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 710                        path, err.buf);
 711                 goto err_no_print;
 712         }
 713 out:
 714         pr_verbose_init(*opts, "ret %i", ret);
 715         printbuf_exit(&err);
 716         return ret;
 717 err:
 718         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 719                path, err.buf);
 720 err_no_print:
 721         bch2_free_super(sb);
 722         goto out;
 723 }
 724
 725 /* write superblock: */
 726
 727 static void write_super_endio(struct bio *bio)
 728 {
 729         struct bch_dev *ca = bio->bi_private;
 730
 731         /* XXX: return errors directly */
 732
 733         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 734                                bch2_blk_status_to_str(bio->bi_status)))
 735                 ca->sb_write_error = 1;
 736
 737         closure_put(&ca->fs->sb_write);
 738         percpu_ref_put(&ca->io_ref);
 739 }
 740
 741 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 742 {
 743         struct bch_sb *sb = ca->disk_sb.sb;
 744         struct bio *bio = ca->disk_sb.bio;
 745
 746         bio_reset(bio);
 747         bio_set_dev(bio, ca->disk_sb.bdev);
 748         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 749         bio->bi_end_io          = write_super_endio;
 750         bio->bi_private         = ca;
 751         bio_set_op_attrs(bio, REQ_OP_READ, REQ_SYNC|REQ_META);
 752         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 753
 754         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 755                      bio_sectors(bio));
 756
 757         percpu_ref_get(&ca->io_ref);
 758         closure_bio_submit(bio, &c->sb_write);
 759 }
 760
 761 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 762 {
 763         struct bch_sb *sb = ca->disk_sb.sb;
 764         struct bio *bio = ca->disk_sb.bio;
 765
 766         sb->offset = sb->layout.sb_offset[idx];
 767
 768         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 769         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 770                                 null_nonce(), sb);
 771
 772         bio_reset(bio);
 773         bio_set_dev(bio, ca->disk_sb.bdev);
 774         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 775         bio->bi_end_io          = write_super_endio;
 776         bio->bi_private         = ca;
 777         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
 778         bch2_bio_map(bio, sb,
 779                      roundup((size_t) vstruct_bytes(sb),
 780                              bdev_logical_block_size(ca->disk_sb.bdev)));
 781
 782         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 783                      bio_sectors(bio));
 784
 785         percpu_ref_get(&ca->io_ref);
 786         closure_bio_submit(bio, &c->sb_write);
 787 }
 788
 789 int bch2_write_super(struct bch_fs *c)
 790 {
 791         struct closure *cl = &c->sb_write;
 792         struct bch_dev *ca;
 793         struct printbuf err = PRINTBUF;
 794         unsigned i, sb = 0, nr_wrote;
 795         struct bch_devs_mask sb_written;
 796         bool wrote, can_mount_without_written, can_mount_with_written;
 797         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 798         int ret = 0;
 799
 800         if (c->opts.very_degraded)
 801                 degraded_flags |= BCH_FORCE_IF_LOST;
 802
 803         lockdep_assert_held(&c->sb_lock);
 804
 805         closure_init_stack(cl);
 806         memset(&sb_written, 0, sizeof(sb_written));
 807
 808         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 809
 810         if (test_bit(BCH_FS_ERROR, &c->flags))
 811                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 812         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 813                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 814
 815         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 816
 817         for_each_online_member(ca, c, i)
 818                 bch2_sb_from_fs(c, ca);
 819
 820         for_each_online_member(ca, c, i) {
 821                 printbuf_reset(&err);
 822
 823                 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
 824                 if (ret) {
 825                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 826                         percpu_ref_put(&ca->io_ref);
 827                         goto out;
 828                 }
 829         }
 830
 831         if (c->opts.nochanges)
 832                 goto out;
 833
 834         for_each_online_member(ca, c, i) {
 835                 __set_bit(ca->dev_idx, sb_written.d);
 836                 ca->sb_write_error = 0;
 837         }
 838
 839         for_each_online_member(ca, c, i)
 840                 read_back_super(c, ca);
 841         closure_sync(cl);
 842
 843         for_each_online_member(ca, c, i) {
 844                 if (ca->sb_write_error)
 845                         continue;
 846
 847                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 848                         bch2_fs_fatal_error(c,
 849                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 850                                 le64_to_cpu(ca->sb_read_scratch->seq),
 851                                 ca->disk_sb.seq);
 852                         percpu_ref_put(&ca->io_ref);
 853                         ret = -EROFS;
 854                         goto out;
 855                 }
 856
 857                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 858                         bch2_fs_fatal_error(c,
 859                                 "Superblock modified by another process (seq %llu expected %llu)",
 860                                 le64_to_cpu(ca->sb_read_scratch->seq),
 861                                 ca->disk_sb.seq);
 862                         percpu_ref_put(&ca->io_ref);
 863                         ret = -EROFS;
 864                         goto out;
 865                 }
 866         }
 867
 868         do {
 869                 wrote = false;
 870                 for_each_online_member(ca, c, i)
 871                         if (!ca->sb_write_error &&
 872                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 873                                 write_one_super(c, ca, sb);
 874                                 wrote = true;
 875                         }
 876                 closure_sync(cl);
 877                 sb++;
 878         } while (wrote);
 879
 880         for_each_online_member(ca, c, i) {
 881                 if (ca->sb_write_error)
 882                         __clear_bit(ca->dev_idx, sb_written.d);
 883                 else
 884                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 885         }
 886
 887         nr_wrote = dev_mask_nr(&sb_written);
 888
 889         can_mount_with_written =
 890                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 891
 892         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 893                 sb_written.d[i] = ~sb_written.d[i];
 894
 895         can_mount_without_written =
 896                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 897
 898         /*
 899          * If we would be able to mount _without_ the devices we successfully
 900          * wrote superblocks to, we weren't able to write to enough devices:
 901          *
 902          * Exception: if we can mount without the successes because we haven't
 903          * written anything (new filesystem), we continue if we'd be able to
 904          * mount with the devices we did successfully write to:
 905          */
 906         if (bch2_fs_fatal_err_on(!nr_wrote ||
 907                                  !can_mount_with_written ||
 908                                  (can_mount_without_written &&
 909                                   !can_mount_with_written), c,
 910                 "Unable to write superblock to sufficient devices (from %ps)",
 911                 (void *) _RET_IP_))
 912                 ret = -1;
 913 out:
 914         /* Make new options visible after they're persistent: */
 915         bch2_sb_update(c);
 916         printbuf_exit(&err);
 917         return ret;
 918 }
 919
 920 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 921 {
 922         mutex_lock(&c->sb_lock);
 923         if (!(c->sb.features & (1ULL << feat))) {
 924                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 925
 926                 bch2_write_super(c);
 927         }
 928         mutex_unlock(&c->sb_lock);
 929 }
 930
 931 /* BCH_SB_FIELD_members: */
 932
 933 static int bch2_sb_members_validate(struct bch_sb *sb,
 934                                     struct bch_sb_field *f,
 935                                     struct printbuf *err)
 936 {
 937         struct bch_sb_field_members *mi = field_to_type(f, members);
 938         unsigned i;
 939
 940         if ((void *) (mi->members + sb->nr_devices) >
 941             vstruct_end(&mi->field)) {
 942                 pr_buf(err, "too many devices for section size");
 943                 return -EINVAL;
 944         }
 945
 946         for (i = 0; i < sb->nr_devices; i++) {
 947                 struct bch_member *m = mi->members + i;
 948
 949                 if (!bch2_member_exists(m))
 950                         continue;
 951
 952                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
 953                         pr_buf(err, "device %u: too many buckets (got %llu, max %lu)",
 954                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
 955                         return -EINVAL;
 956                 }
 957
 958                 if (le64_to_cpu(m->nbuckets) -
 959                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
 960                         pr_buf(err, "device %u: not enough buckets (got %llu, max %u)",
 961                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
 962                         return -EINVAL;
 963                 }
 964
 965                 if (le16_to_cpu(m->bucket_size) <
 966                     le16_to_cpu(sb->block_size)) {
 967                         pr_buf(err, "device %u: bucket size %u smaller than block size %u",
 968                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
 969                         return -EINVAL;
 970                 }
 971
 972                 if (le16_to_cpu(m->bucket_size) <
 973                     BCH_SB_BTREE_NODE_SIZE(sb)) {
 974                         pr_buf(err, "device %u: bucket size %u smaller than btree node size %llu",
 975                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
 976                         return -EINVAL;
 977                 }
 978         }
 979
 980         return 0;
 981 }
 982
 983 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
 984                                     struct bch_sb_field *f)
 985 {
 986         struct bch_sb_field_members *mi = field_to_type(f, members);
 987         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
 988         unsigned i;
 989
 990         for (i = 0; i < sb->nr_devices; i++) {
 991                 struct bch_member *m = mi->members + i;
 992                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
 993                 u64 bucket_size = le16_to_cpu(m->bucket_size);
 994                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
 995
 996                 if (!bch2_member_exists(m))
 997                         continue;
 998
 999                 pr_buf(out, "Device:");
1000                 pr_tab(out);
1001                 pr_buf(out, "%u", i);
1002                 pr_newline(out);
1003
1004                 pr_indent_push(out, 2);
1005
1006                 pr_buf(out, "UUID:");
1007                 pr_tab(out);
1008                 pr_uuid(out, m->uuid.b);
1009                 pr_newline(out);
1010
1011                 pr_buf(out, "Size:");
1012                 pr_tab(out);
1013                 pr_units(out, device_size, device_size << 9);
1014                 pr_newline(out);
1015
1016                 pr_buf(out, "Bucket size:");
1017                 pr_tab(out);
1018                 pr_units(out, bucket_size, bucket_size << 9);
1019                 pr_newline(out);
1020
1021                 pr_buf(out, "First bucket:");
1022                 pr_tab(out);
1023                 pr_buf(out, "%u", le16_to_cpu(m->first_bucket));
1024                 pr_newline(out);
1025
1026                 pr_buf(out, "Buckets:");
1027                 pr_tab(out);
1028                 pr_buf(out, "%llu", le64_to_cpu(m->nbuckets));
1029                 pr_newline(out);
1030
1031                 pr_buf(out, "Last mount:");
1032                 pr_tab(out);
1033                 if (m->last_mount)
1034                         pr_time(out, le64_to_cpu(m->last_mount));
1035                 else
1036                         pr_buf(out, "(never)");
1037                 pr_newline(out);
1038
1039                 pr_buf(out, "State:");
1040                 pr_tab(out);
1041                 pr_buf(out, "%s",
1042                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1043                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1044                        : "unknown");
1045                 pr_newline(out);
1046
1047                 pr_buf(out, "Group:");
1048                 pr_tab(out);
1049                 if (BCH_MEMBER_GROUP(m)) {
1050                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1051
1052                         if (idx < disk_groups_nr(gi))
1053                                 pr_buf(out, "%s (%u)",
1054                                        gi->entries[idx].label, idx);
1055                         else
1056                                 pr_buf(out, "(bad disk labels section)");
1057                 } else {
1058                         pr_buf(out, "(none)");
1059                 }
1060                 pr_newline(out);
1061
1062                 pr_buf(out, "Data allowed:");
1063                 pr_tab(out);
1064                 if (BCH_MEMBER_DATA_ALLOWED(m))
1065                         bch2_flags_to_text(out, bch2_data_types,
1066                                            BCH_MEMBER_DATA_ALLOWED(m));
1067                 else
1068                         pr_buf(out, "(none)");
1069                 pr_newline(out);
1070
1071                 pr_buf(out, "Has data:");
1072                 pr_tab(out);
1073                 if (data_have)
1074                         bch2_flags_to_text(out, bch2_data_types, data_have);
1075                 else
1076                         pr_buf(out, "(none)");
1077                 pr_newline(out);
1078
1079                 pr_buf(out, "Discard:");
1080                 pr_tab(out);
1081                 pr_buf(out, "%llu", BCH_MEMBER_DISCARD(m));
1082                 pr_newline(out);
1083
1084                 pr_buf(out, "Freespace initialized:");
1085                 pr_tab(out);
1086                 pr_buf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1087                 pr_newline(out);
1088
1089                 pr_indent_pop(out, 2);
1090         }
1091 }
1092
1093 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1094         .validate       = bch2_sb_members_validate,
1095         .to_text        = bch2_sb_members_to_text,
1096 };
1097
1098 /* BCH_SB_FIELD_crypt: */
1099
1100 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1101                                   struct bch_sb_field *f,
1102                                   struct printbuf *err)
1103 {
1104         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1105
1106         if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1107                 pr_buf(err, "wrong size (got %zu should be %zu)",
1108                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1109                 return -EINVAL;
1110         }
1111
1112         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1113                 pr_buf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1114                 return -EINVAL;
1115         }
1116
1117         return 0;
1118 }
1119
1120 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1121                                   struct bch_sb_field *f)
1122 {
1123         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1124
1125         pr_buf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1126         pr_newline(out);
1127         pr_buf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1128         pr_newline(out);
1129         pr_buf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1130         pr_newline(out);
1131         pr_buf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1132         pr_newline(out);
1133 }
1134
1135 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1136         .validate       = bch2_sb_crypt_validate,
1137         .to_text        = bch2_sb_crypt_to_text,
1138 };
1139
1140 /* BCH_SB_FIELD_clean: */
1141
1142 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1143 {
1144         struct jset_entry *entry;
1145         int ret;
1146
1147         for (entry = clean->start;
1148              entry < (struct jset_entry *) vstruct_end(&clean->field);
1149              entry = vstruct_next(entry)) {
1150                 ret = bch2_journal_entry_validate(c, "superblock", entry,
1151                                                   le16_to_cpu(c->disk_sb.sb->version),
1152                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1153                                                   write);
1154                 if (ret)
1155                         return ret;
1156         }
1157
1158         return 0;
1159 }
1160
1161 int bch2_fs_mark_dirty(struct bch_fs *c)
1162 {
1163         int ret;
1164
1165         /*
1166          * Unconditionally write superblock, to verify it hasn't changed before
1167          * we go rw:
1168          */
1169
1170         mutex_lock(&c->sb_lock);
1171         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1172         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1173         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1174         ret = bch2_write_super(c);
1175         mutex_unlock(&c->sb_lock);
1176
1177         return ret;
1178 }
1179
1180 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1181 {
1182         struct jset_entry *entry = *end;
1183         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1184
1185         memset(entry, 0, u64s * sizeof(u64));
1186         /*
1187          * The u64s field counts from the start of data, ignoring the shared
1188          * fields.
1189          */
1190         entry->u64s = cpu_to_le16(u64s - 1);
1191
1192         *end = vstruct_next(*end);
1193         return entry;
1194 }
1195
1196 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1197                                            struct jset_entry **end,
1198                                            u64 journal_seq)
1199 {
1200         struct bch_dev *ca;
1201         unsigned i, dev;
1202
1203         percpu_down_read(&c->mark_lock);
1204
1205         if (!journal_seq) {
1206                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1207                         bch2_fs_usage_acc_to_base(c, i);
1208         } else {
1209                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1210         }
1211
1212         {
1213                 struct jset_entry_usage *u =
1214                         container_of(jset_entry_init(end, sizeof(*u)),
1215                                      struct jset_entry_usage, entry);
1216
1217                 u->entry.type   = BCH_JSET_ENTRY_usage;
1218                 u->entry.btree_id = BCH_FS_USAGE_inodes;
1219                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1220         }
1221
1222         {
1223                 struct jset_entry_usage *u =
1224                         container_of(jset_entry_init(end, sizeof(*u)),
1225                                      struct jset_entry_usage, entry);
1226
1227                 u->entry.type   = BCH_JSET_ENTRY_usage;
1228                 u->entry.btree_id = BCH_FS_USAGE_key_version;
1229                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1230         }
1231
1232         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1233                 struct jset_entry_usage *u =
1234                         container_of(jset_entry_init(end, sizeof(*u)),
1235                                      struct jset_entry_usage, entry);
1236
1237                 u->entry.type   = BCH_JSET_ENTRY_usage;
1238                 u->entry.btree_id = BCH_FS_USAGE_reserved;
1239                 u->entry.level  = i;
1240                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1241         }
1242
1243         for (i = 0; i < c->replicas.nr; i++) {
1244                 struct bch_replicas_entry *e =
1245                         cpu_replicas_entry(&c->replicas, i);
1246                 struct jset_entry_data_usage *u =
1247                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1248                                      struct jset_entry_data_usage, entry);
1249
1250                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1251                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1252                 memcpy(&u->r, e, replicas_entry_bytes(e));
1253         }
1254
1255         for_each_member_device(ca, c, dev) {
1256                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1257                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1258                 struct jset_entry_dev_usage *u =
1259                         container_of(jset_entry_init(end, b),
1260                                      struct jset_entry_dev_usage, entry);
1261
1262                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1263                 u->dev = cpu_to_le32(dev);
1264                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1265                 u->buckets_unavailable  = cpu_to_le64(ca->usage_base->buckets_unavailable);
1266
1267                 for (i = 0; i < BCH_DATA_NR; i++) {
1268                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1269                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1270                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1271                 }
1272         }
1273
1274         percpu_up_read(&c->mark_lock);
1275
1276         for (i = 0; i < 2; i++) {
1277                 struct jset_entry_clock *clock =
1278                         container_of(jset_entry_init(end, sizeof(*clock)),
1279                                      struct jset_entry_clock, entry);
1280
1281                 clock->entry.type = BCH_JSET_ENTRY_clock;
1282                 clock->rw       = i;
1283                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1284         }
1285 }
1286
1287 void bch2_fs_mark_clean(struct bch_fs *c)
1288 {
1289         struct bch_sb_field_clean *sb_clean;
1290         struct jset_entry *entry;
1291         unsigned u64s;
1292         int ret;
1293
1294         mutex_lock(&c->sb_lock);
1295         if (BCH_SB_CLEAN(c->disk_sb.sb))
1296                 goto out;
1297
1298         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1299
1300         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1301         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1302         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1303         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1304
1305         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1306
1307         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1308         if (!sb_clean) {
1309                 bch_err(c, "error resizing superblock while setting filesystem clean");
1310                 goto out;
1311         }
1312
1313         sb_clean->flags         = 0;
1314         sb_clean->journal_seq   = cpu_to_le64(atomic64_read(&c->journal.seq));
1315
1316         /* Trying to catch outstanding bug: */
1317         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1318
1319         entry = sb_clean->start;
1320         bch2_journal_super_entries_add_common(c, &entry, 0);
1321         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1322         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1323
1324         memset(entry, 0,
1325                vstruct_end(&sb_clean->field) - (void *) entry);
1326
1327         /*
1328          * this should be in the write path, and we should be validating every
1329          * superblock section:
1330          */
1331         ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1332         if (ret) {
1333                 bch_err(c, "error writing marking filesystem clean: validate error");
1334                 goto out;
1335         }
1336
1337         bch2_write_super(c);
1338 out:
1339         mutex_unlock(&c->sb_lock);
1340 }
1341
1342 static int bch2_sb_clean_validate(struct bch_sb *sb,
1343                                   struct bch_sb_field *f,
1344                                   struct printbuf *err)
1345 {
1346         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1347
1348         if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1349                 pr_buf(err, "wrong size (got %zu should be %zu)",
1350                        vstruct_bytes(&clean->field), sizeof(*clean));
1351                 return -EINVAL;
1352         }
1353
1354         return 0;
1355 }
1356
1357 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1358                                   struct bch_sb_field *f)
1359 {
1360         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1361         struct jset_entry *entry;
1362
1363         pr_buf(out, "flags:          %x",       le32_to_cpu(clean->flags));
1364         pr_newline(out);
1365         pr_buf(out, "journal_seq:    %llu",     le64_to_cpu(clean->journal_seq));
1366         pr_newline(out);
1367
1368         for (entry = clean->start;
1369              entry != vstruct_end(&clean->field);
1370              entry = vstruct_next(entry)) {
1371                 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1372                     !entry->u64s)
1373                         continue;
1374
1375                 bch2_journal_entry_to_text(out, NULL, entry);
1376                 pr_newline(out);
1377         }
1378 }
1379
1380 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1381         .validate       = bch2_sb_clean_validate,
1382         .to_text        = bch2_sb_clean_to_text,
1383 };
1384
1385 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1386 #define x(f, nr)                                        \
1387         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1388         BCH_SB_FIELDS()
1389 #undef x
1390 };
1391
1392 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1393                                   struct printbuf *err)
1394 {
1395         unsigned type = le32_to_cpu(f->type);
1396         struct printbuf field_err = PRINTBUF;
1397         int ret;
1398
1399         if (type >= BCH_SB_FIELD_NR)
1400                 return 0;
1401
1402         ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
1403         if (ret) {
1404                 pr_buf(err, "Invalid superblock section %s: %s",
1405                        bch2_sb_fields[type],
1406                        field_err.buf);
1407                 pr_newline(err);
1408                 bch2_sb_field_to_text(err, sb, f);
1409         }
1410
1411         printbuf_exit(&field_err);
1412         return ret;
1413 }
1414
1415 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1416                            struct bch_sb_field *f)
1417 {
1418         unsigned type = le32_to_cpu(f->type);
1419         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1420                 ? bch2_sb_field_ops[type] : NULL;
1421
1422         if (!out->tabstops[0])
1423                 out->tabstops[0] = 32;
1424
1425         if (ops)
1426                 pr_buf(out, "%s", bch2_sb_fields[type]);
1427         else
1428                 pr_buf(out, "(unknown field %u)", type);
1429
1430         pr_buf(out, " (size %zu):", vstruct_bytes(f));
1431         pr_newline(out);
1432
1433         if (ops && ops->to_text) {
1434                 pr_indent_push(out, 2);
1435                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1436                 pr_indent_pop(out, 2);
1437         }
1438 }
1439
1440 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1441 {
1442         unsigned i;
1443
1444         pr_buf(out, "Type:                    %u", l->layout_type);
1445         pr_newline(out);
1446
1447         pr_buf(out, "Superblock max size:     ");
1448         pr_units(out,
1449                  1 << l->sb_max_size_bits,
1450                  512 << l->sb_max_size_bits);
1451         pr_newline(out);
1452
1453         pr_buf(out, "Nr superblocks:          %u", l->nr_superblocks);
1454         pr_newline(out);
1455
1456         pr_buf(out, "Offsets:                 ");
1457         for (i = 0; i < l->nr_superblocks; i++) {
1458                 if (i)
1459                         pr_buf(out, ", ");
1460                 pr_buf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1461         }
1462         pr_newline(out);
1463 }
1464
1465 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1466                      bool print_layout, unsigned fields)
1467 {
1468         struct bch_sb_field_members *mi;
1469         struct bch_sb_field *f;
1470         u64 fields_have = 0;
1471         unsigned nr_devices = 0;
1472
1473         if (!out->tabstops[0])
1474                 out->tabstops[0] = 32;
1475
1476         mi = bch2_sb_get_members(sb);
1477         if (mi) {
1478                 struct bch_member *m;
1479
1480                 for (m = mi->members;
1481                      m < mi->members + sb->nr_devices;
1482                      m++)
1483                         nr_devices += bch2_member_exists(m);
1484         }
1485
1486         pr_buf(out, "External UUID:");
1487         pr_tab(out);
1488         pr_uuid(out, sb->user_uuid.b);
1489         pr_newline(out);
1490
1491         pr_buf(out, "Internal UUID:");
1492         pr_tab(out);
1493         pr_uuid(out, sb->uuid.b);
1494         pr_newline(out);
1495
1496         pr_buf(out, "Device index:");
1497         pr_tab(out);
1498         pr_buf(out, "%u", sb->dev_idx);
1499         pr_newline(out);
1500
1501         pr_buf(out, "Label:");
1502         pr_tab(out);
1503         pr_buf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1504         pr_newline(out);
1505
1506         pr_buf(out, "Version:");
1507         pr_tab(out);
1508         pr_buf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
1509         pr_newline(out);
1510
1511         pr_buf(out, "Oldest version on disk:");
1512         pr_tab(out);
1513         pr_buf(out, "%u", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
1514         pr_newline(out);
1515
1516         pr_buf(out, "Created:");
1517         pr_tab(out);
1518         if (sb->time_base_lo)
1519                 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1520         else
1521                 pr_buf(out, "(not set)");
1522         pr_newline(out);
1523
1524         pr_buf(out, "Sequence number:");
1525         pr_tab(out);
1526         pr_buf(out, "%llu", le64_to_cpu(sb->seq));
1527         pr_newline(out);
1528
1529         pr_buf(out, "Superblock size:");
1530         pr_tab(out);
1531         pr_buf(out, "%zu", vstruct_bytes(sb));
1532         pr_newline(out);
1533
1534         pr_buf(out, "Clean:");
1535         pr_tab(out);
1536         pr_buf(out, "%llu", BCH_SB_CLEAN(sb));
1537         pr_newline(out);
1538
1539         pr_buf(out, "Devices:");
1540         pr_tab(out);
1541         pr_buf(out, "%u", nr_devices);
1542         pr_newline(out);
1543
1544         pr_buf(out, "Sections:");
1545         vstruct_for_each(sb, f)
1546                 fields_have |= 1 << le32_to_cpu(f->type);
1547         pr_tab(out);
1548         bch2_flags_to_text(out, bch2_sb_fields, fields_have);
1549         pr_newline(out);
1550
1551         pr_buf(out, "Features:");
1552         pr_tab(out);
1553         bch2_flags_to_text(out, bch2_sb_features,
1554                            le64_to_cpu(sb->features[0]));
1555         pr_newline(out);
1556
1557         pr_buf(out, "Compat features:");
1558         pr_tab(out);
1559         bch2_flags_to_text(out, bch2_sb_compat,
1560                            le64_to_cpu(sb->compat[0]));
1561         pr_newline(out);
1562
1563         pr_newline(out);
1564         pr_buf(out, "Options:");
1565         pr_newline(out);
1566         pr_indent_push(out, 2);
1567         {
1568                 enum bch_opt_id id;
1569
1570                 for (id = 0; id < bch2_opts_nr; id++) {
1571                         const struct bch_option *opt = bch2_opt_table + id;
1572
1573                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1574                                 u64 v = bch2_opt_from_sb(sb, id);
1575
1576                                 pr_buf(out, "%s:", opt->attr.name);
1577                                 pr_tab(out);
1578                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1579                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1580                                 pr_newline(out);
1581                         }
1582                 }
1583         }
1584
1585         pr_indent_pop(out, 2);
1586
1587         if (print_layout) {
1588                 pr_newline(out);
1589                 pr_buf(out, "layout:");
1590                 pr_newline(out);
1591                 pr_indent_push(out, 2);
1592                 bch2_sb_layout_to_text(out, &sb->layout);
1593                 pr_indent_pop(out, 2);
1594         }
1595
1596         vstruct_for_each(sb, f)
1597                 if (fields & (1 << le32_to_cpu(f->type))) {
1598                         pr_newline(out);
1599                         bch2_sb_field_to_text(out, sb, f);
1600                 }
1601 }