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 #include "counters.h"
  21
  22 #include <linux/backing-dev.h>
  23 #include <linux/pretty-printers.h>
  24 #include <linux/sort.h>
  25
  26 #include <trace/events/bcachefs.h>
  27
  28 const char * const bch2_sb_fields[] = {
  29 #define x(name, nr)     #name,
  30         BCH_SB_FIELDS()
  31 #undef x
  32         NULL
  33 };
  34
  35 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
  36                                   struct printbuf *);
  37
  38 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  39                                       enum bch_sb_field_type type)
  40 {
  41         struct bch_sb_field *f;
  42
  43         /* XXX: need locking around superblock to access optional fields */
  44
  45         vstruct_for_each(sb, f)
  46                 if (le32_to_cpu(f->type) == type)
  47                         return f;
  48         return NULL;
  49 }
  50
  51 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  52                                                    struct bch_sb_field *f,
  53                                                    unsigned u64s)
  54 {
  55         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  56         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  57
  58         BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
  59
  60         if (!f && !u64s) {
  61                 /* nothing to do: */
  62         } else if (!f) {
  63                 f = vstruct_last(sb->sb);
  64                 memset(f, 0, sizeof(u64) * u64s);
  65                 f->u64s = cpu_to_le32(u64s);
  66                 f->type = 0;
  67         } else {
  68                 void *src, *dst;
  69
  70                 src = vstruct_end(f);
  71
  72                 if (u64s) {
  73                         f->u64s = cpu_to_le32(u64s);
  74                         dst = vstruct_end(f);
  75                 } else {
  76                         dst = f;
  77                 }
  78
  79                 memmove(dst, src, vstruct_end(sb->sb) - src);
  80
  81                 if (dst > src)
  82                         memset(src, 0, dst - src);
  83         }
  84
  85         sb->sb->u64s = cpu_to_le32(sb_u64s);
  86
  87         return u64s ? f : NULL;
  88 }
  89
  90 void bch2_sb_field_delete(struct bch_sb_handle *sb,
  91                           enum bch_sb_field_type type)
  92 {
  93         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
  94
  95         if (f)
  96                 __bch2_sb_field_resize(sb, f, 0);
  97 }
  98
  99 /* Superblock realloc/free: */
 100
 101 void bch2_free_super(struct bch_sb_handle *sb)
 102 {
 103         if (sb->bio)
 104                 kfree(sb->bio);
 105         if (!IS_ERR_OR_NULL(sb->bdev))
 106                 blkdev_put(sb->bdev, sb->mode);
 107
 108         kfree(sb->sb);
 109         memset(sb, 0, sizeof(*sb));
 110 }
 111
 112 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 113 {
 114         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 115         size_t new_buffer_size;
 116         struct bch_sb *new_sb;
 117         struct bio *bio;
 118
 119         if (sb->bdev)
 120                 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
 121
 122         new_buffer_size = roundup_pow_of_two(new_bytes);
 123
 124         if (sb->sb && sb->buffer_size >= new_buffer_size)
 125                 return 0;
 126
 127         if (sb->have_layout) {
 128                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 129
 130                 if (new_bytes > max_bytes) {
 131                         pr_err("%pg: superblock too big: want %zu but have %llu",
 132                                sb->bdev, new_bytes, max_bytes);
 133                         return -BCH_ERR_ENOSPC_sb;
 134                 }
 135         }
 136
 137         if (sb->buffer_size >= new_buffer_size && sb->sb)
 138                 return 0;
 139
 140         if (dynamic_fault("bcachefs:add:super_realloc"))
 141                 return -ENOMEM;
 142
 143         if (sb->have_bio) {
 144                 unsigned nr_bvecs = DIV_ROUND_UP(new_buffer_size, PAGE_SIZE);
 145
 146                 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 147                 if (!bio)
 148                         return -ENOMEM;
 149
 150                 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
 151
 152                 if (sb->bio)
 153                         kfree(sb->bio);
 154                 sb->bio = bio;
 155         }
 156
 157         new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 158         if (!new_sb)
 159                 return -ENOMEM;
 160
 161         sb->sb = new_sb;
 162         sb->buffer_size = new_buffer_size;
 163
 164         return 0;
 165 }
 166
 167 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 168                                           enum bch_sb_field_type type,
 169                                           unsigned u64s)
 170 {
 171         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 172         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 173         ssize_t d = -old_u64s + u64s;
 174
 175         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 176                 return NULL;
 177
 178         if (sb->fs_sb) {
 179                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 180                 struct bch_dev *ca;
 181                 unsigned i;
 182
 183                 lockdep_assert_held(&c->sb_lock);
 184
 185                 /* XXX: we're not checking that offline device have enough space */
 186
 187                 for_each_online_member(ca, c, i) {
 188                         struct bch_sb_handle *sb = &ca->disk_sb;
 189
 190                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 191                                 percpu_ref_put(&ca->ref);
 192                                 return NULL;
 193                         }
 194                 }
 195         }
 196
 197         f = bch2_sb_field_get(sb->sb, type);
 198         f = __bch2_sb_field_resize(sb, f, u64s);
 199         if (f)
 200                 f->type = cpu_to_le32(type);
 201         return f;
 202 }
 203
 204 /* Superblock validate: */
 205
 206 static inline void __bch2_sb_layout_size_assert(void)
 207 {
 208         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 209 }
 210
 211 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
 212 {
 213         u64 offset, prev_offset, max_sectors;
 214         unsigned i;
 215
 216         if (uuid_le_cmp(layout->magic, BCACHE_MAGIC)) {
 217                 prt_printf(out, "Not a bcachefs superblock layout");
 218                 return -EINVAL;
 219         }
 220
 221         if (layout->layout_type != 0) {
 222                 prt_printf(out, "Invalid superblock layout type %u",
 223                        layout->layout_type);
 224                 return -EINVAL;
 225         }
 226
 227         if (!layout->nr_superblocks) {
 228                 prt_printf(out, "Invalid superblock layout: no superblocks");
 229                 return -EINVAL;
 230         }
 231
 232         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
 233                 prt_printf(out, "Invalid superblock layout: too many superblocks");
 234                 return -EINVAL;
 235         }
 236
 237         max_sectors = 1 << layout->sb_max_size_bits;
 238
 239         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 240
 241         for (i = 1; i < layout->nr_superblocks; i++) {
 242                 offset = le64_to_cpu(layout->sb_offset[i]);
 243
 244                 if (offset < prev_offset + max_sectors) {
 245                         prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
 246                                "  (sb %u ends at %llu next starts at %llu",
 247                                i - 1, prev_offset + max_sectors, offset);
 248                         return -EINVAL;
 249                 }
 250                 prev_offset = offset;
 251         }
 252
 253         return 0;
 254 }
 255
 256 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 257                             int rw)
 258 {
 259         struct bch_sb *sb = disk_sb->sb;
 260         struct bch_sb_field *f;
 261         struct bch_sb_field_members *mi;
 262         enum bch_opt_id opt_id;
 263         u32 version, version_min;
 264         u16 block_size;
 265         int ret;
 266
 267         version         = le16_to_cpu(sb->version);
 268         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 269                 ? le16_to_cpu(sb->version_min)
 270                 : version;
 271
 272         if (version    >= bcachefs_metadata_version_max) {
 273                 prt_printf(out, "Unsupported superblock version %u (min %u, max %u)",
 274                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 275                 return -EINVAL;
 276         }
 277
 278         if (version_min < bcachefs_metadata_version_min) {
 279                 prt_printf(out, "Unsupported superblock version %u (min %u, max %u)",
 280                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 281                 return -EINVAL;
 282         }
 283
 284         if (version_min > version) {
 285                 prt_printf(out, "Bad minimum version %u, greater than version field %u",
 286                        version_min, version);
 287                 return -EINVAL;
 288         }
 289
 290         if (sb->features[1] ||
 291             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
 292                 prt_printf(out, "Filesystem has incompatible features");
 293                 return -EINVAL;
 294         }
 295
 296         block_size = le16_to_cpu(sb->block_size);
 297
 298         if (block_size > PAGE_SECTORS) {
 299                 prt_printf(out, "Block size too big (got %u, max %u)",
 300                        block_size, PAGE_SECTORS);
 301                 return -EINVAL;
 302         }
 303
 304         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
 305                 prt_printf(out, "Bad user UUID (got zeroes)");
 306                 return -EINVAL;
 307         }
 308
 309         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
 310                 prt_printf(out, "Bad intenal UUID (got zeroes)");
 311                 return -EINVAL;
 312         }
 313
 314         if (!sb->nr_devices ||
 315             sb->nr_devices > BCH_SB_MEMBERS_MAX) {
 316                 prt_printf(out, "Bad number of member devices %u (max %u)",
 317                        sb->nr_devices, BCH_SB_MEMBERS_MAX);
 318                 return -EINVAL;
 319         }
 320
 321         if (sb->dev_idx >= sb->nr_devices) {
 322                 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
 323                        sb->dev_idx, sb->nr_devices);
 324                 return -EINVAL;
 325         }
 326
 327         if (!sb->time_precision ||
 328             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
 329                 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
 330                        le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
 331                 return -EINVAL;
 332         }
 333
 334         if (rw == READ) {
 335                 /*
 336                  * Been seeing a bug where these are getting inexplicably
 337                  * zeroed, so we'r now validating them, but we have to be
 338                  * careful not to preven people's filesystems from mounting:
 339                  */
 340                 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 341                         SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 342                 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 343                         SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
 344         }
 345
 346         for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
 347                 const struct bch_option *opt = bch2_opt_table + opt_id;
 348
 349                 if (opt->get_sb != BCH2_NO_SB_OPT) {
 350                         u64 v = bch2_opt_from_sb(sb, opt_id);
 351
 352                         prt_printf(out, "Invalid option ");
 353                         ret = bch2_opt_validate(opt, v, out);
 354                         if (ret)
 355                                 return ret;
 356
 357                         printbuf_reset(out);
 358                 }
 359         }
 360
 361         /* validate layout */
 362         ret = validate_sb_layout(&sb->layout, out);
 363         if (ret)
 364                 return ret;
 365
 366         vstruct_for_each(sb, f) {
 367                 if (!f->u64s) {
 368                         prt_printf(out, "Invalid superblock: optional with size 0 (type %u)",
 369                                le32_to_cpu(f->type));
 370                         return -EINVAL;
 371                 }
 372
 373                 if (vstruct_next(f) > vstruct_last(sb)) {
 374                         prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
 375                                le32_to_cpu(f->type));
 376                         return -EINVAL;
 377                 }
 378         }
 379
 380         /* members must be validated first: */
 381         mi = bch2_sb_get_members(sb);
 382         if (!mi) {
 383                 prt_printf(out, "Invalid superblock: member info area missing");
 384                 return -EINVAL;
 385         }
 386
 387         ret = bch2_sb_field_validate(sb, &mi->field, out);
 388         if (ret)
 389                 return ret;
 390
 391         vstruct_for_each(sb, f) {
 392                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 393                         continue;
 394
 395                 ret = bch2_sb_field_validate(sb, f, out);
 396                 if (ret)
 397                         return ret;
 398         }
 399
 400         return 0;
 401 }
 402
 403 /* device open: */
 404
 405 static void bch2_sb_update(struct bch_fs *c)
 406 {
 407         struct bch_sb *src = c->disk_sb.sb;
 408         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 409         struct bch_dev *ca;
 410         unsigned i;
 411
 412         lockdep_assert_held(&c->sb_lock);
 413
 414         c->sb.uuid              = src->uuid;
 415         c->sb.user_uuid         = src->user_uuid;
 416         c->sb.version           = le16_to_cpu(src->version);
 417         c->sb.version_min       = le16_to_cpu(src->version_min);
 418         c->sb.nr_devices        = src->nr_devices;
 419         c->sb.clean             = BCH_SB_CLEAN(src);
 420         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 421
 422         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 423         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 424
 425         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 426         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 427                                           c->sb.nsec_per_time_unit);
 428         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 429
 430         c->sb.features          = le64_to_cpu(src->features[0]);
 431         c->sb.compat            = le64_to_cpu(src->compat[0]);
 432
 433         for_each_member_device(ca, c, i)
 434                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 435 }
 436
 437 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 438 {
 439         struct bch_sb_field *src_f, *dst_f;
 440         struct bch_sb *dst = dst_handle->sb;
 441         unsigned i;
 442
 443         dst->version            = src->version;
 444         dst->version_min        = src->version_min;
 445         dst->seq                = src->seq;
 446         dst->uuid               = src->uuid;
 447         dst->user_uuid          = src->user_uuid;
 448         memcpy(dst->label,      src->label, sizeof(dst->label));
 449
 450         dst->block_size         = src->block_size;
 451         dst->nr_devices         = src->nr_devices;
 452
 453         dst->time_base_lo       = src->time_base_lo;
 454         dst->time_base_hi       = src->time_base_hi;
 455         dst->time_precision     = src->time_precision;
 456
 457         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 458         memcpy(dst->features,   src->features,  sizeof(dst->features));
 459         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 460
 461         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 462                 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 463                         continue;
 464
 465                 src_f = bch2_sb_field_get(src, i);
 466                 dst_f = bch2_sb_field_get(dst, i);
 467                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 468                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 469
 470                 if (src_f)
 471                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 472         }
 473 }
 474
 475 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 476 {
 477         struct bch_sb_field_journal *journal_buckets =
 478                 bch2_sb_get_journal(src);
 479         unsigned journal_u64s = journal_buckets
 480                 ? le32_to_cpu(journal_buckets->field.u64s)
 481                 : 0;
 482         int ret;
 483
 484         lockdep_assert_held(&c->sb_lock);
 485
 486         ret = bch2_sb_realloc(&c->disk_sb,
 487                               le32_to_cpu(src->u64s) - journal_u64s);
 488         if (ret)
 489                 return ret;
 490
 491         __copy_super(&c->disk_sb, src);
 492
 493         ret = bch2_sb_replicas_to_cpu_replicas(c);
 494         if (ret)
 495                 return ret;
 496
 497         ret = bch2_sb_disk_groups_to_cpu(c);
 498         if (ret)
 499                 return ret;
 500
 501         bch2_sb_update(c);
 502         return 0;
 503 }
 504
 505 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 506 {
 507         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 508         struct bch_sb_field_journal *journal_buckets =
 509                 bch2_sb_get_journal(dst);
 510         unsigned journal_u64s = journal_buckets
 511                 ? le32_to_cpu(journal_buckets->field.u64s)
 512                 : 0;
 513         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 514         int ret;
 515
 516         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 517         if (ret)
 518                 return ret;
 519
 520         __copy_super(&ca->disk_sb, src);
 521         return 0;
 522 }
 523
 524 /* read superblock: */
 525
 526 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 527 {
 528         struct bch_csum csum;
 529         u32 version, version_min;
 530         size_t bytes;
 531         int ret;
 532 reread:
 533         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 534         sb->bio->bi_iter.bi_sector = offset;
 535         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 536
 537         ret = submit_bio_wait(sb->bio);
 538         if (ret) {
 539                 prt_printf(err, "IO error: %i", ret);
 540                 return ret;
 541         }
 542
 543         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC)) {
 544                 prt_printf(err, "Not a bcachefs superblock");
 545                 return -EINVAL;
 546         }
 547
 548         version         = le16_to_cpu(sb->sb->version);
 549         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 550                 ? le16_to_cpu(sb->sb->version_min)
 551                 : version;
 552
 553         if (version    >= bcachefs_metadata_version_max) {
 554                 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
 555                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 556                 return -EINVAL;
 557         }
 558
 559         if (version_min < bcachefs_metadata_version_min) {
 560                 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
 561                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 562                 return -EINVAL;
 563         }
 564
 565         bytes = vstruct_bytes(sb->sb);
 566
 567         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 568                 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 569                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 570                 return -EINVAL;
 571         }
 572
 573         if (bytes > sb->buffer_size) {
 574                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 575                         return -ENOMEM;
 576                 goto reread;
 577         }
 578
 579         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
 580                 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 581                 return -EINVAL;
 582         }
 583
 584         /* XXX: verify MACs */
 585         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 586                             null_nonce(), sb->sb);
 587
 588         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 589                 prt_printf(err, "bad checksum");
 590                 return -EINVAL;
 591         }
 592
 593         sb->seq = le64_to_cpu(sb->sb->seq);
 594
 595         return 0;
 596 }
 597
 598 int bch2_read_super(const char *path, struct bch_opts *opts,
 599                     struct bch_sb_handle *sb)
 600 {
 601         u64 offset = opt_get(*opts, sb);
 602         struct bch_sb_layout layout;
 603         struct printbuf err = PRINTBUF;
 604         __le64 *i;
 605         int ret;
 606
 607         pr_verbose_init(*opts, "");
 608
 609         memset(sb, 0, sizeof(*sb));
 610         sb->mode        = FMODE_READ;
 611         sb->have_bio    = true;
 612
 613         if (!opt_get(*opts, noexcl))
 614                 sb->mode |= FMODE_EXCL;
 615
 616         if (!opt_get(*opts, nochanges))
 617                 sb->mode |= FMODE_WRITE;
 618
 619         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 620         if (IS_ERR(sb->bdev) &&
 621             PTR_ERR(sb->bdev) == -EACCES &&
 622             opt_get(*opts, read_only)) {
 623                 sb->mode &= ~FMODE_WRITE;
 624
 625                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 626                 if (!IS_ERR(sb->bdev))
 627                         opt_set(*opts, nochanges, true);
 628         }
 629
 630         if (IS_ERR(sb->bdev)) {
 631                 ret = PTR_ERR(sb->bdev);
 632                 goto out;
 633         }
 634
 635         ret = bch2_sb_realloc(sb, 0);
 636         if (ret) {
 637                 prt_printf(&err, "error allocating memory for superblock");
 638                 goto err;
 639         }
 640
 641         if (bch2_fs_init_fault("read_super")) {
 642                 prt_printf(&err, "dynamic fault");
 643                 ret = -EFAULT;
 644                 goto err;
 645         }
 646
 647         ret = read_one_super(sb, offset, &err);
 648         if (!ret)
 649                 goto got_super;
 650
 651         if (opt_defined(*opts, sb))
 652                 goto err;
 653
 654         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 655                path, err.buf);
 656         printbuf_reset(&err);
 657
 658         /*
 659          * Error reading primary superblock - read location of backup
 660          * superblocks:
 661          */
 662         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 663         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 664         /*
 665          * use sb buffer to read layout, since sb buffer is page aligned but
 666          * layout won't be:
 667          */
 668         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 669
 670         ret = submit_bio_wait(sb->bio);
 671         if (ret) {
 672                 prt_printf(&err, "IO error: %i", ret);
 673                 goto err;
 674         }
 675
 676         memcpy(&layout, sb->sb, sizeof(layout));
 677         ret = validate_sb_layout(&layout, &err);
 678         if (ret)
 679                 goto err;
 680
 681         for (i = layout.sb_offset;
 682              i < layout.sb_offset + layout.nr_superblocks; i++) {
 683                 offset = le64_to_cpu(*i);
 684
 685                 if (offset == opt_get(*opts, sb))
 686                         continue;
 687
 688                 ret = read_one_super(sb, offset, &err);
 689                 if (!ret)
 690                         goto got_super;
 691         }
 692
 693         goto err;
 694
 695 got_super:
 696         if (le16_to_cpu(sb->sb->block_size) << 9 <
 697             bdev_logical_block_size(sb->bdev)) {
 698                 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
 699                        le16_to_cpu(sb->sb->block_size) << 9,
 700                        bdev_logical_block_size(sb->bdev));
 701                 ret = -EINVAL;
 702                 goto err;
 703         }
 704
 705         ret = 0;
 706         sb->have_layout = true;
 707
 708         ret = bch2_sb_validate(sb, &err, READ);
 709         if (ret) {
 710                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 711                        path, err.buf);
 712                 goto err_no_print;
 713         }
 714 out:
 715         pr_verbose_init(*opts, "ret %i", ret);
 716         printbuf_exit(&err);
 717         return ret;
 718 err:
 719         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 720                path, err.buf);
 721 err_no_print:
 722         bch2_free_super(sb);
 723         goto out;
 724 }
 725
 726 /* write superblock: */
 727
 728 static void write_super_endio(struct bio *bio)
 729 {
 730         struct bch_dev *ca = bio->bi_private;
 731
 732         /* XXX: return errors directly */
 733
 734         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 735                                bch2_blk_status_to_str(bio->bi_status)))
 736                 ca->sb_write_error = 1;
 737
 738         closure_put(&ca->fs->sb_write);
 739         percpu_ref_put(&ca->io_ref);
 740 }
 741
 742 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 743 {
 744         struct bch_sb *sb = ca->disk_sb.sb;
 745         struct bio *bio = ca->disk_sb.bio;
 746
 747         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 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         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 752
 753         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 754                      bio_sectors(bio));
 755
 756         percpu_ref_get(&ca->io_ref);
 757         closure_bio_submit(bio, &c->sb_write);
 758 }
 759
 760 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 761 {
 762         struct bch_sb *sb = ca->disk_sb.sb;
 763         struct bio *bio = ca->disk_sb.bio;
 764
 765         sb->offset = sb->layout.sb_offset[idx];
 766
 767         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 768         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 769                                 null_nonce(), sb);
 770
 771         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 772         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 773         bio->bi_end_io          = write_super_endio;
 774         bio->bi_private         = ca;
 775         bch2_bio_map(bio, sb,
 776                      roundup((size_t) vstruct_bytes(sb),
 777                              bdev_logical_block_size(ca->disk_sb.bdev)));
 778
 779         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 780                      bio_sectors(bio));
 781
 782         percpu_ref_get(&ca->io_ref);
 783         closure_bio_submit(bio, &c->sb_write);
 784 }
 785
 786 int bch2_write_super(struct bch_fs *c)
 787 {
 788         struct closure *cl = &c->sb_write;
 789         struct bch_dev *ca;
 790         struct printbuf err = PRINTBUF;
 791         unsigned i, sb = 0, nr_wrote;
 792         struct bch_devs_mask sb_written;
 793         bool wrote, can_mount_without_written, can_mount_with_written;
 794         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 795         int ret = 0;
 796
 797         trace_and_count(c, write_super, c, _RET_IP_);
 798
 799         if (c->opts.very_degraded)
 800                 degraded_flags |= BCH_FORCE_IF_LOST;
 801
 802         lockdep_assert_held(&c->sb_lock);
 803
 804         closure_init_stack(cl);
 805         memset(&sb_written, 0, sizeof(sb_written));
 806
 807         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 808
 809         if (test_bit(BCH_FS_ERROR, &c->flags))
 810                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 811         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 812                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 813
 814         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 815
 816         bch2_sb_counters_from_cpu(c);
 817
 818         for_each_online_member(ca, c, i)
 819                 bch2_sb_from_fs(c, ca);
 820
 821         for_each_online_member(ca, c, i) {
 822                 printbuf_reset(&err);
 823
 824                 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
 825                 if (ret) {
 826                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 827                         percpu_ref_put(&ca->io_ref);
 828                         goto out;
 829                 }
 830         }
 831
 832         if (c->opts.nochanges)
 833                 goto out;
 834
 835         /*
 836          * Defer writing the superblock until filesystem initialization is
 837          * complete - don't write out a partly initialized superblock:
 838          */
 839         if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
 840                 goto out;
 841
 842         for_each_online_member(ca, c, i) {
 843                 __set_bit(ca->dev_idx, sb_written.d);
 844                 ca->sb_write_error = 0;
 845         }
 846
 847         for_each_online_member(ca, c, i)
 848                 read_back_super(c, ca);
 849         closure_sync(cl);
 850
 851         for_each_online_member(ca, c, i) {
 852                 if (ca->sb_write_error)
 853                         continue;
 854
 855                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 856                         bch2_fs_fatal_error(c,
 857                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 858                                 le64_to_cpu(ca->sb_read_scratch->seq),
 859                                 ca->disk_sb.seq);
 860                         percpu_ref_put(&ca->io_ref);
 861                         ret = -EROFS;
 862                         goto out;
 863                 }
 864
 865                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 866                         bch2_fs_fatal_error(c,
 867                                 "Superblock modified by another process (seq %llu expected %llu)",
 868                                 le64_to_cpu(ca->sb_read_scratch->seq),
 869                                 ca->disk_sb.seq);
 870                         percpu_ref_put(&ca->io_ref);
 871                         ret = -EROFS;
 872                         goto out;
 873                 }
 874         }
 875
 876         do {
 877                 wrote = false;
 878                 for_each_online_member(ca, c, i)
 879                         if (!ca->sb_write_error &&
 880                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 881                                 write_one_super(c, ca, sb);
 882                                 wrote = true;
 883                         }
 884                 closure_sync(cl);
 885                 sb++;
 886         } while (wrote);
 887
 888         for_each_online_member(ca, c, i) {
 889                 if (ca->sb_write_error)
 890                         __clear_bit(ca->dev_idx, sb_written.d);
 891                 else
 892                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 893         }
 894
 895         nr_wrote = dev_mask_nr(&sb_written);
 896
 897         can_mount_with_written =
 898                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 899
 900         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 901                 sb_written.d[i] = ~sb_written.d[i];
 902
 903         can_mount_without_written =
 904                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 905
 906         /*
 907          * If we would be able to mount _without_ the devices we successfully
 908          * wrote superblocks to, we weren't able to write to enough devices:
 909          *
 910          * Exception: if we can mount without the successes because we haven't
 911          * written anything (new filesystem), we continue if we'd be able to
 912          * mount with the devices we did successfully write to:
 913          */
 914         if (bch2_fs_fatal_err_on(!nr_wrote ||
 915                                  !can_mount_with_written ||
 916                                  (can_mount_without_written &&
 917                                   !can_mount_with_written), c,
 918                 "Unable to write superblock to sufficient devices (from %ps)",
 919                 (void *) _RET_IP_))
 920                 ret = -1;
 921 out:
 922         /* Make new options visible after they're persistent: */
 923         bch2_sb_update(c);
 924         printbuf_exit(&err);
 925         return ret;
 926 }
 927
 928 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 929 {
 930         mutex_lock(&c->sb_lock);
 931         if (!(c->sb.features & (1ULL << feat))) {
 932                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 933
 934                 bch2_write_super(c);
 935         }
 936         mutex_unlock(&c->sb_lock);
 937 }
 938
 939 /* BCH_SB_FIELD_members: */
 940
 941 static int bch2_sb_members_validate(struct bch_sb *sb,
 942                                     struct bch_sb_field *f,
 943                                     struct printbuf *err)
 944 {
 945         struct bch_sb_field_members *mi = field_to_type(f, members);
 946         unsigned i;
 947
 948         if ((void *) (mi->members + sb->nr_devices) >
 949             vstruct_end(&mi->field)) {
 950                 prt_printf(err, "too many devices for section size");
 951                 return -EINVAL;
 952         }
 953
 954         for (i = 0; i < sb->nr_devices; i++) {
 955                 struct bch_member *m = mi->members + i;
 956
 957                 if (!bch2_member_exists(m))
 958                         continue;
 959
 960                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
 961                         prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
 962                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
 963                         return -EINVAL;
 964                 }
 965
 966                 if (le64_to_cpu(m->nbuckets) -
 967                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
 968                         prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
 969                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
 970                         return -EINVAL;
 971                 }
 972
 973                 if (le16_to_cpu(m->bucket_size) <
 974                     le16_to_cpu(sb->block_size)) {
 975                         prt_printf(err, "device %u: bucket size %u smaller than block size %u",
 976                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
 977                         return -EINVAL;
 978                 }
 979
 980                 if (le16_to_cpu(m->bucket_size) <
 981                     BCH_SB_BTREE_NODE_SIZE(sb)) {
 982                         prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
 983                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
 984                         return -EINVAL;
 985                 }
 986         }
 987
 988         return 0;
 989 }
 990
 991 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
 992                                     struct bch_sb_field *f)
 993 {
 994         struct bch_sb_field_members *mi = field_to_type(f, members);
 995         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
 996         unsigned i;
 997
 998         for (i = 0; i < sb->nr_devices; i++) {
 999                 struct bch_member *m = mi->members + i;
1000                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
1001                 u64 bucket_size = le16_to_cpu(m->bucket_size);
1002                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1003
1004                 if (!bch2_member_exists(m))
1005                         continue;
1006
1007                 prt_printf(out, "Device:");
1008                 prt_tab(out);
1009                 prt_printf(out, "%u", i);
1010                 prt_newline(out);
1011
1012                 printbuf_indent_add(out, 2);
1013
1014                 prt_printf(out, "UUID:");
1015                 prt_tab(out);
1016                 pr_uuid(out, m->uuid.b);
1017                 prt_newline(out);
1018
1019                 prt_printf(out, "Size:");
1020                 prt_tab(out);
1021                 prt_units_u64(out, device_size << 9);
1022                 prt_newline(out);
1023
1024                 prt_printf(out, "Bucket size:");
1025                 prt_tab(out);
1026                 prt_units_u64(out, bucket_size << 9);
1027                 prt_newline(out);
1028
1029                 prt_printf(out, "First bucket:");
1030                 prt_tab(out);
1031                 prt_printf(out, "%u", le16_to_cpu(m->first_bucket));
1032                 prt_newline(out);
1033
1034                 prt_printf(out, "Buckets:");
1035                 prt_tab(out);
1036                 prt_printf(out, "%llu", le64_to_cpu(m->nbuckets));
1037                 prt_newline(out);
1038
1039                 prt_printf(out, "Last mount:");
1040                 prt_tab(out);
1041                 if (m->last_mount)
1042                         pr_time(out, le64_to_cpu(m->last_mount));
1043                 else
1044                         prt_printf(out, "(never)");
1045                 prt_newline(out);
1046
1047                 prt_printf(out, "State:");
1048                 prt_tab(out);
1049                 prt_printf(out, "%s",
1050                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1051                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1052                        : "unknown");
1053                 prt_newline(out);
1054
1055                 prt_printf(out, "Label:");
1056                 prt_tab(out);
1057                 if (BCH_MEMBER_GROUP(m)) {
1058                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1059
1060                         if (idx < disk_groups_nr(gi))
1061                                 prt_printf(out, "%s (%u)",
1062                                        gi->entries[idx].label, idx);
1063                         else
1064                                 prt_printf(out, "(bad disk labels section)");
1065                 } else {
1066                         prt_printf(out, "(none)");
1067                 }
1068                 prt_newline(out);
1069
1070                 prt_printf(out, "Data allowed:");
1071                 prt_tab(out);
1072                 if (BCH_MEMBER_DATA_ALLOWED(m))
1073                         prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(m));
1074                 else
1075                         prt_printf(out, "(none)");
1076                 prt_newline(out);
1077
1078                 prt_printf(out, "Has data:");
1079                 prt_tab(out);
1080                 if (data_have)
1081                         prt_bitflags(out, bch2_data_types, data_have);
1082                 else
1083                         prt_printf(out, "(none)");
1084                 prt_newline(out);
1085
1086                 prt_printf(out, "Discard:");
1087                 prt_tab(out);
1088                 prt_printf(out, "%llu", BCH_MEMBER_DISCARD(m));
1089                 prt_newline(out);
1090
1091                 prt_printf(out, "Freespace initialized:");
1092                 prt_tab(out);
1093                 prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1094                 prt_newline(out);
1095
1096                 printbuf_indent_sub(out, 2);
1097         }
1098 }
1099
1100 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1101         .validate       = bch2_sb_members_validate,
1102         .to_text        = bch2_sb_members_to_text,
1103 };
1104
1105 /* BCH_SB_FIELD_crypt: */
1106
1107 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1108                                   struct bch_sb_field *f,
1109                                   struct printbuf *err)
1110 {
1111         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1112
1113         if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1114                 prt_printf(err, "wrong size (got %zu should be %zu)",
1115                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1116                 return -EINVAL;
1117         }
1118
1119         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1120                 prt_printf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1121                 return -EINVAL;
1122         }
1123
1124         return 0;
1125 }
1126
1127 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1128                                   struct bch_sb_field *f)
1129 {
1130         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1131
1132         prt_printf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1133         prt_newline(out);
1134         prt_printf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1135         prt_newline(out);
1136         prt_printf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1137         prt_newline(out);
1138         prt_printf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1139         prt_newline(out);
1140 }
1141
1142 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1143         .validate       = bch2_sb_crypt_validate,
1144         .to_text        = bch2_sb_crypt_to_text,
1145 };
1146
1147 /* BCH_SB_FIELD_clean: */
1148
1149 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1150 {
1151         struct jset_entry *entry;
1152         int ret;
1153
1154         for (entry = clean->start;
1155              entry < (struct jset_entry *) vstruct_end(&clean->field);
1156              entry = vstruct_next(entry)) {
1157                 ret = bch2_journal_entry_validate(c, NULL, entry,
1158                                                   le16_to_cpu(c->disk_sb.sb->version),
1159                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1160                                                   write);
1161                 if (ret)
1162                         return ret;
1163         }
1164
1165         return 0;
1166 }
1167
1168 int bch2_fs_mark_dirty(struct bch_fs *c)
1169 {
1170         int ret;
1171
1172         /*
1173          * Unconditionally write superblock, to verify it hasn't changed before
1174          * we go rw:
1175          */
1176
1177         mutex_lock(&c->sb_lock);
1178         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1179         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1180         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1181         ret = bch2_write_super(c);
1182         mutex_unlock(&c->sb_lock);
1183
1184         return ret;
1185 }
1186
1187 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1188 {
1189         struct jset_entry *entry = *end;
1190         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1191
1192         memset(entry, 0, u64s * sizeof(u64));
1193         /*
1194          * The u64s field counts from the start of data, ignoring the shared
1195          * fields.
1196          */
1197         entry->u64s = cpu_to_le16(u64s - 1);
1198
1199         *end = vstruct_next(*end);
1200         return entry;
1201 }
1202
1203 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1204                                            struct jset_entry **end,
1205                                            u64 journal_seq)
1206 {
1207         struct bch_dev *ca;
1208         unsigned i, dev;
1209
1210         percpu_down_read(&c->mark_lock);
1211
1212         if (!journal_seq) {
1213                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1214                         bch2_fs_usage_acc_to_base(c, i);
1215         } else {
1216                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1217         }
1218
1219         {
1220                 struct jset_entry_usage *u =
1221                         container_of(jset_entry_init(end, sizeof(*u)),
1222                                      struct jset_entry_usage, entry);
1223
1224                 u->entry.type   = BCH_JSET_ENTRY_usage;
1225                 u->entry.btree_id = BCH_FS_USAGE_inodes;
1226                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1227         }
1228
1229         {
1230                 struct jset_entry_usage *u =
1231                         container_of(jset_entry_init(end, sizeof(*u)),
1232                                      struct jset_entry_usage, entry);
1233
1234                 u->entry.type   = BCH_JSET_ENTRY_usage;
1235                 u->entry.btree_id = BCH_FS_USAGE_key_version;
1236                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1237         }
1238
1239         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1240                 struct jset_entry_usage *u =
1241                         container_of(jset_entry_init(end, sizeof(*u)),
1242                                      struct jset_entry_usage, entry);
1243
1244                 u->entry.type   = BCH_JSET_ENTRY_usage;
1245                 u->entry.btree_id = BCH_FS_USAGE_reserved;
1246                 u->entry.level  = i;
1247                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1248         }
1249
1250         for (i = 0; i < c->replicas.nr; i++) {
1251                 struct bch_replicas_entry *e =
1252                         cpu_replicas_entry(&c->replicas, i);
1253                 struct jset_entry_data_usage *u =
1254                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1255                                      struct jset_entry_data_usage, entry);
1256
1257                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1258                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1259                 memcpy(&u->r, e, replicas_entry_bytes(e));
1260         }
1261
1262         for_each_member_device(ca, c, dev) {
1263                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1264                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1265                 struct jset_entry_dev_usage *u =
1266                         container_of(jset_entry_init(end, b),
1267                                      struct jset_entry_dev_usage, entry);
1268
1269                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1270                 u->dev = cpu_to_le32(dev);
1271                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1272
1273                 for (i = 0; i < BCH_DATA_NR; i++) {
1274                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1275                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1276                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1277                 }
1278         }
1279
1280         percpu_up_read(&c->mark_lock);
1281
1282         for (i = 0; i < 2; i++) {
1283                 struct jset_entry_clock *clock =
1284                         container_of(jset_entry_init(end, sizeof(*clock)),
1285                                      struct jset_entry_clock, entry);
1286
1287                 clock->entry.type = BCH_JSET_ENTRY_clock;
1288                 clock->rw       = i;
1289                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1290         }
1291 }
1292
1293 void bch2_fs_mark_clean(struct bch_fs *c)
1294 {
1295         struct bch_sb_field_clean *sb_clean;
1296         struct jset_entry *entry;
1297         unsigned u64s;
1298         int ret;
1299
1300         mutex_lock(&c->sb_lock);
1301         if (BCH_SB_CLEAN(c->disk_sb.sb))
1302                 goto out;
1303
1304         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1305
1306         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1307         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1308         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1309         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1310
1311         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1312
1313         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1314         if (!sb_clean) {
1315                 bch_err(c, "error resizing superblock while setting filesystem clean");
1316                 goto out;
1317         }
1318
1319         sb_clean->flags         = 0;
1320         sb_clean->journal_seq   = cpu_to_le64(atomic64_read(&c->journal.seq));
1321
1322         /* Trying to catch outstanding bug: */
1323         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1324
1325         entry = sb_clean->start;
1326         bch2_journal_super_entries_add_common(c, &entry, 0);
1327         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1328         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1329
1330         memset(entry, 0,
1331                vstruct_end(&sb_clean->field) - (void *) entry);
1332
1333         /*
1334          * this should be in the write path, and we should be validating every
1335          * superblock section:
1336          */
1337         ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1338         if (ret) {
1339                 bch_err(c, "error writing marking filesystem clean: validate error");
1340                 goto out;
1341         }
1342
1343         bch2_write_super(c);
1344 out:
1345         mutex_unlock(&c->sb_lock);
1346 }
1347
1348 static int bch2_sb_clean_validate(struct bch_sb *sb,
1349                                   struct bch_sb_field *f,
1350                                   struct printbuf *err)
1351 {
1352         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1353
1354         if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1355                 prt_printf(err, "wrong size (got %zu should be %zu)",
1356                        vstruct_bytes(&clean->field), sizeof(*clean));
1357                 return -EINVAL;
1358         }
1359
1360         return 0;
1361 }
1362
1363 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1364                                   struct bch_sb_field *f)
1365 {
1366         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1367         struct jset_entry *entry;
1368
1369         prt_printf(out, "flags:          %x",   le32_to_cpu(clean->flags));
1370         prt_newline(out);
1371         prt_printf(out, "journal_seq:    %llu", le64_to_cpu(clean->journal_seq));
1372         prt_newline(out);
1373
1374         for (entry = clean->start;
1375              entry != vstruct_end(&clean->field);
1376              entry = vstruct_next(entry)) {
1377                 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1378                     !entry->u64s)
1379                         continue;
1380
1381                 bch2_journal_entry_to_text(out, NULL, entry);
1382                 prt_newline(out);
1383         }
1384 }
1385
1386 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1387         .validate       = bch2_sb_clean_validate,
1388         .to_text        = bch2_sb_clean_to_text,
1389 };
1390
1391 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1392 #define x(f, nr)                                        \
1393         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1394         BCH_SB_FIELDS()
1395 #undef x
1396 };
1397
1398 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1399                                   struct printbuf *err)
1400 {
1401         unsigned type = le32_to_cpu(f->type);
1402         struct printbuf field_err = PRINTBUF;
1403         int ret;
1404
1405         if (type >= BCH_SB_FIELD_NR)
1406                 return 0;
1407
1408         ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
1409         if (ret) {
1410                 prt_printf(err, "Invalid superblock section %s: %s",
1411                        bch2_sb_fields[type],
1412                        field_err.buf);
1413                 prt_newline(err);
1414                 bch2_sb_field_to_text(err, sb, f);
1415         }
1416
1417         printbuf_exit(&field_err);
1418         return ret;
1419 }
1420
1421 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1422                            struct bch_sb_field *f)
1423 {
1424         unsigned type = le32_to_cpu(f->type);
1425         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1426                 ? bch2_sb_field_ops[type] : NULL;
1427
1428         if (!out->nr_tabstops)
1429                 printbuf_tabstop_push(out, 32);
1430
1431         if (ops)
1432                 prt_printf(out, "%s", bch2_sb_fields[type]);
1433         else
1434                 prt_printf(out, "(unknown field %u)", type);
1435
1436         prt_printf(out, " (size %zu):", vstruct_bytes(f));
1437         prt_newline(out);
1438
1439         if (ops && ops->to_text) {
1440                 printbuf_indent_add(out, 2);
1441                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1442                 printbuf_indent_sub(out, 2);
1443         }
1444 }
1445
1446 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1447 {
1448         unsigned i;
1449
1450         prt_printf(out, "Type:                    %u", l->layout_type);
1451         prt_newline(out);
1452
1453         prt_str(out, "Superblock max size:     ");
1454         prt_units_u64(out, 512 << l->sb_max_size_bits);
1455         prt_newline(out);
1456
1457         prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1458         prt_newline(out);
1459
1460         prt_str(out, "Offsets:                 ");
1461         for (i = 0; i < l->nr_superblocks; i++) {
1462                 if (i)
1463                         prt_str(out, ", ");
1464                 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1465         }
1466         prt_newline(out);
1467 }
1468
1469 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1470                      bool print_layout, unsigned fields)
1471 {
1472         struct bch_sb_field_members *mi;
1473         struct bch_sb_field *f;
1474         u64 fields_have = 0;
1475         unsigned nr_devices = 0;
1476
1477         if (!out->nr_tabstops)
1478                 printbuf_tabstop_push(out, 44);
1479
1480         mi = bch2_sb_get_members(sb);
1481         if (mi) {
1482                 struct bch_member *m;
1483
1484                 for (m = mi->members;
1485                      m < mi->members + sb->nr_devices;
1486                      m++)
1487                         nr_devices += bch2_member_exists(m);
1488         }
1489
1490         prt_printf(out, "External UUID:");
1491         prt_tab(out);
1492         pr_uuid(out, sb->user_uuid.b);
1493         prt_newline(out);
1494
1495         prt_printf(out, "Internal UUID:");
1496         prt_tab(out);
1497         pr_uuid(out, sb->uuid.b);
1498         prt_newline(out);
1499
1500         prt_str(out, "Device index:");
1501         prt_tab(out);
1502         prt_printf(out, "%u", sb->dev_idx);
1503         prt_newline(out);
1504
1505         prt_str(out, "Label:");
1506         prt_tab(out);
1507         prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1508         prt_newline(out);
1509
1510         prt_str(out, "Version:");
1511         prt_tab(out);
1512         prt_printf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
1513         prt_newline(out);
1514
1515         prt_printf(out, "Oldest version on disk:");
1516         prt_tab(out);
1517         prt_printf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
1518         prt_newline(out);
1519
1520         prt_printf(out, "Created:");
1521         prt_tab(out);
1522         if (sb->time_base_lo)
1523                 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1524         else
1525                 prt_printf(out, "(not set)");
1526         prt_newline(out);
1527
1528         prt_printf(out, "Sequence number:");
1529         prt_tab(out);
1530         prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1531         prt_newline(out);
1532
1533         prt_printf(out, "Superblock size:");
1534         prt_tab(out);
1535         prt_printf(out, "%zu", vstruct_bytes(sb));
1536         prt_newline(out);
1537
1538         prt_printf(out, "Clean:");
1539         prt_tab(out);
1540         prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1541         prt_newline(out);
1542
1543         prt_printf(out, "Devices:");
1544         prt_tab(out);
1545         prt_printf(out, "%u", nr_devices);
1546         prt_newline(out);
1547
1548         prt_printf(out, "Sections:");
1549         vstruct_for_each(sb, f)
1550                 fields_have |= 1 << le32_to_cpu(f->type);
1551         prt_tab(out);
1552         prt_bitflags(out, bch2_sb_fields, fields_have);
1553         prt_newline(out);
1554
1555         prt_printf(out, "Features:");
1556         prt_tab(out);
1557         prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1558         prt_newline(out);
1559
1560         prt_printf(out, "Compat features:");
1561         prt_tab(out);
1562         prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1563         prt_newline(out);
1564
1565         prt_newline(out);
1566         prt_printf(out, "Options:");
1567         prt_newline(out);
1568         printbuf_indent_add(out, 2);
1569         {
1570                 enum bch_opt_id id;
1571
1572                 for (id = 0; id < bch2_opts_nr; id++) {
1573                         const struct bch_option *opt = bch2_opt_table + id;
1574
1575                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1576                                 u64 v = bch2_opt_from_sb(sb, id);
1577
1578                                 prt_printf(out, "%s:", opt->attr.name);
1579                                 prt_tab(out);
1580                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1581                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1582                                 prt_newline(out);
1583                         }
1584                 }
1585         }
1586
1587         printbuf_indent_sub(out, 2);
1588
1589         if (print_layout) {
1590                 prt_newline(out);
1591                 prt_printf(out, "layout:");
1592                 prt_newline(out);
1593                 printbuf_indent_add(out, 2);
1594                 bch2_sb_layout_to_text(out, &sb->layout);
1595                 printbuf_indent_sub(out, 2);
1596         }
1597
1598         vstruct_for_each(sb, f)
1599                 if (fields & (1 << le32_to_cpu(f->type))) {
1600                         prt_newline(out);
1601                         bch2_sb_field_to_text(out, sb, f);
1602                 }
1603 }