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/sort.h>
  24
  25 #include <trace/events/bcachefs.h>
  26
  27 const char * const bch2_sb_fields[] = {
  28 #define x(name, nr)     #name,
  29         BCH_SB_FIELDS()
  30 #undef x
  31         NULL
  32 };
  33
  34 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
  35                                   struct printbuf *);
  36
  37 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  38                                       enum bch_sb_field_type type)
  39 {
  40         struct bch_sb_field *f;
  41
  42         /* XXX: need locking around superblock to access optional fields */
  43
  44         vstruct_for_each(sb, f)
  45                 if (le32_to_cpu(f->type) == type)
  46                         return f;
  47         return NULL;
  48 }
  49
  50 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  51                                                    struct bch_sb_field *f,
  52                                                    unsigned u64s)
  53 {
  54         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  55         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  56
  57         BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
  58
  59         if (!f && !u64s) {
  60                 /* nothing to do: */
  61         } else if (!f) {
  62                 f = vstruct_last(sb->sb);
  63                 memset(f, 0, sizeof(u64) * u64s);
  64                 f->u64s = cpu_to_le32(u64s);
  65                 f->type = 0;
  66         } else {
  67                 void *src, *dst;
  68
  69                 src = vstruct_end(f);
  70
  71                 if (u64s) {
  72                         f->u64s = cpu_to_le32(u64s);
  73                         dst = vstruct_end(f);
  74                 } else {
  75                         dst = f;
  76                 }
  77
  78                 memmove(dst, src, vstruct_end(sb->sb) - src);
  79
  80                 if (dst > src)
  81                         memset(src, 0, dst - src);
  82         }
  83
  84         sb->sb->u64s = cpu_to_le32(sb_u64s);
  85
  86         return u64s ? f : NULL;
  87 }
  88
  89 void bch2_sb_field_delete(struct bch_sb_handle *sb,
  90                           enum bch_sb_field_type type)
  91 {
  92         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
  93
  94         if (f)
  95                 __bch2_sb_field_resize(sb, f, 0);
  96 }
  97
  98 /* Superblock realloc/free: */
  99
 100 void bch2_free_super(struct bch_sb_handle *sb)
 101 {
 102         kfree(sb->bio);
 103         if (!IS_ERR_OR_NULL(sb->bdev))
 104                 blkdev_put(sb->bdev, sb->mode);
 105
 106         kfree(sb->sb);
 107         memset(sb, 0, sizeof(*sb));
 108 }
 109
 110 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 111 {
 112         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 113         size_t new_buffer_size;
 114         struct bch_sb *new_sb;
 115         struct bio *bio;
 116
 117         if (sb->bdev)
 118                 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
 119
 120         new_buffer_size = roundup_pow_of_two(new_bytes);
 121
 122         if (sb->sb && sb->buffer_size >= new_buffer_size)
 123                 return 0;
 124
 125         if (sb->have_layout) {
 126                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 127
 128                 if (new_bytes > max_bytes) {
 129                         pr_err("%pg: superblock too big: want %zu but have %llu",
 130                                sb->bdev, new_bytes, max_bytes);
 131                         return -BCH_ERR_ENOSPC_sb;
 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                 unsigned nr_bvecs = DIV_ROUND_UP(new_buffer_size, PAGE_SIZE);
 143
 144                 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 145                 if (!bio)
 146                         return -ENOMEM;
 147
 148                 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
 149
 150                 kfree(sb->bio);
 151                 sb->bio = bio;
 152         }
 153
 154         new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 155         if (!new_sb)
 156                 return -ENOMEM;
 157
 158         sb->sb = new_sb;
 159         sb->buffer_size = new_buffer_size;
 160
 161         return 0;
 162 }
 163
 164 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 165                                           enum bch_sb_field_type type,
 166                                           unsigned u64s)
 167 {
 168         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 169         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 170         ssize_t d = -old_u64s + u64s;
 171
 172         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 173                 return NULL;
 174
 175         if (sb->fs_sb) {
 176                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 177                 struct bch_dev *ca;
 178                 unsigned i;
 179
 180                 lockdep_assert_held(&c->sb_lock);
 181
 182                 /* XXX: we're not checking that offline device have enough space */
 183
 184                 for_each_online_member(ca, c, i) {
 185                         struct bch_sb_handle *sb = &ca->disk_sb;
 186
 187                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 188                                 percpu_ref_put(&ca->ref);
 189                                 return NULL;
 190                         }
 191                 }
 192         }
 193
 194         f = bch2_sb_field_get(sb->sb, type);
 195         f = __bch2_sb_field_resize(sb, f, u64s);
 196         if (f)
 197                 f->type = cpu_to_le32(type);
 198         return f;
 199 }
 200
 201 /* Superblock validate: */
 202
 203 static inline void __bch2_sb_layout_size_assert(void)
 204 {
 205         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 206 }
 207
 208 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
 209 {
 210         u64 offset, prev_offset, max_sectors;
 211         unsigned i;
 212
 213         if (uuid_le_cmp(layout->magic, BCACHE_MAGIC) &&
 214             uuid_le_cmp(layout->magic, BCHFS_MAGIC)) {
 215                 prt_printf(out, "Not a bcachefs superblock layout");
 216                 return -BCH_ERR_invalid_sb_layout;
 217         }
 218
 219         if (layout->layout_type != 0) {
 220                 prt_printf(out, "Invalid superblock layout type %u",
 221                        layout->layout_type);
 222                 return -BCH_ERR_invalid_sb_layout_type;
 223         }
 224
 225         if (!layout->nr_superblocks) {
 226                 prt_printf(out, "Invalid superblock layout: no superblocks");
 227                 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 228         }
 229
 230         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
 231                 prt_printf(out, "Invalid superblock layout: too many superblocks");
 232                 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 233         }
 234
 235         max_sectors = 1 << layout->sb_max_size_bits;
 236
 237         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 238
 239         for (i = 1; i < layout->nr_superblocks; i++) {
 240                 offset = le64_to_cpu(layout->sb_offset[i]);
 241
 242                 if (offset < prev_offset + max_sectors) {
 243                         prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
 244                                "  (sb %u ends at %llu next starts at %llu",
 245                                i - 1, prev_offset + max_sectors, offset);
 246                         return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
 247                 }
 248                 prev_offset = offset;
 249         }
 250
 251         return 0;
 252 }
 253
 254 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 255                             int rw)
 256 {
 257         struct bch_sb *sb = disk_sb->sb;
 258         struct bch_sb_field *f;
 259         struct bch_sb_field_members *mi;
 260         enum bch_opt_id opt_id;
 261         u32 version, version_min;
 262         u16 block_size;
 263         int ret;
 264
 265         version         = le16_to_cpu(sb->version);
 266         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 267                 ? le16_to_cpu(sb->version_min)
 268                 : version;
 269
 270         if (version    >= bcachefs_metadata_version_max) {
 271                 prt_printf(out, "Unsupported superblock version %u (min %u, max %u)",
 272                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 273                 return -BCH_ERR_invalid_sb_version;
 274         }
 275
 276         if (version_min < bcachefs_metadata_version_min) {
 277                 prt_printf(out, "Unsupported superblock version %u (min %u, max %u)",
 278                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 279                 return -BCH_ERR_invalid_sb_version;
 280         }
 281
 282         if (version_min > version) {
 283                 prt_printf(out, "Bad minimum version %u, greater than version field %u",
 284                        version_min, version);
 285                 return -BCH_ERR_invalid_sb_version;
 286         }
 287
 288         if (sb->features[1] ||
 289             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
 290                 prt_printf(out, "Filesystem has incompatible features");
 291                 return -BCH_ERR_invalid_sb_features;
 292         }
 293
 294         block_size = le16_to_cpu(sb->block_size);
 295
 296         if (block_size > PAGE_SECTORS) {
 297                 prt_printf(out, "Block size too big (got %u, max %u)",
 298                        block_size, PAGE_SECTORS);
 299                 return -BCH_ERR_invalid_sb_block_size;
 300         }
 301
 302         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
 303                 prt_printf(out, "Bad user UUID (got zeroes)");
 304                 return -BCH_ERR_invalid_sb_uuid;
 305         }
 306
 307         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
 308                 prt_printf(out, "Bad intenal UUID (got zeroes)");
 309                 return -BCH_ERR_invalid_sb_uuid;
 310         }
 311
 312         if (!sb->nr_devices ||
 313             sb->nr_devices > BCH_SB_MEMBERS_MAX) {
 314                 prt_printf(out, "Bad number of member devices %u (max %u)",
 315                        sb->nr_devices, BCH_SB_MEMBERS_MAX);
 316                 return -BCH_ERR_invalid_sb_too_many_members;
 317         }
 318
 319         if (sb->dev_idx >= sb->nr_devices) {
 320                 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
 321                        sb->dev_idx, sb->nr_devices);
 322                 return -BCH_ERR_invalid_sb_dev_idx;
 323         }
 324
 325         if (!sb->time_precision ||
 326             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
 327                 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
 328                        le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
 329                 return -BCH_ERR_invalid_sb_time_precision;
 330         }
 331
 332         if (rw == READ) {
 333                 /*
 334                  * Been seeing a bug where these are getting inexplicably
 335                  * zeroed, so we'r now validating them, but we have to be
 336                  * careful not to preven people's filesystems from mounting:
 337                  */
 338                 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 339                         SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 340                 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 341                         SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
 342         }
 343
 344         for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
 345                 const struct bch_option *opt = bch2_opt_table + opt_id;
 346
 347                 if (opt->get_sb != BCH2_NO_SB_OPT) {
 348                         u64 v = bch2_opt_from_sb(sb, opt_id);
 349
 350                         prt_printf(out, "Invalid option ");
 351                         ret = bch2_opt_validate(opt, v, out);
 352                         if (ret)
 353                                 return ret;
 354
 355                         printbuf_reset(out);
 356                 }
 357         }
 358
 359         /* validate layout */
 360         ret = validate_sb_layout(&sb->layout, out);
 361         if (ret)
 362                 return ret;
 363
 364         vstruct_for_each(sb, f) {
 365                 if (!f->u64s) {
 366                         prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
 367                                le32_to_cpu(f->type));
 368                         return -BCH_ERR_invalid_sb_field_size;
 369                 }
 370
 371                 if (vstruct_next(f) > vstruct_last(sb)) {
 372                         prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
 373                                le32_to_cpu(f->type));
 374                         return -BCH_ERR_invalid_sb_field_size;
 375                 }
 376         }
 377
 378         /* members must be validated first: */
 379         mi = bch2_sb_get_members(sb);
 380         if (!mi) {
 381                 prt_printf(out, "Invalid superblock: member info area missing");
 382                 return -BCH_ERR_invalid_sb_members_missing;
 383         }
 384
 385         ret = bch2_sb_field_validate(sb, &mi->field, out);
 386         if (ret)
 387                 return ret;
 388
 389         vstruct_for_each(sb, f) {
 390                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 391                         continue;
 392
 393                 ret = bch2_sb_field_validate(sb, f, out);
 394                 if (ret)
 395                         return ret;
 396         }
 397
 398         return 0;
 399 }
 400
 401 /* device open: */
 402
 403 static void bch2_sb_update(struct bch_fs *c)
 404 {
 405         struct bch_sb *src = c->disk_sb.sb;
 406         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 407         struct bch_dev *ca;
 408         unsigned i;
 409
 410         lockdep_assert_held(&c->sb_lock);
 411
 412         c->sb.uuid              = src->uuid;
 413         c->sb.user_uuid         = src->user_uuid;
 414         c->sb.version           = le16_to_cpu(src->version);
 415         c->sb.version_min       = le16_to_cpu(src->version_min);
 416         c->sb.nr_devices        = src->nr_devices;
 417         c->sb.clean             = BCH_SB_CLEAN(src);
 418         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 419
 420         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 421         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 422
 423         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 424         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 425                                           c->sb.nsec_per_time_unit);
 426         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 427
 428         c->sb.features          = le64_to_cpu(src->features[0]);
 429         c->sb.compat            = le64_to_cpu(src->compat[0]);
 430
 431         for_each_member_device(ca, c, i)
 432                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 433 }
 434
 435 static int __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 436 {
 437         struct bch_sb_field *src_f, *dst_f;
 438         struct bch_sb *dst = dst_handle->sb;
 439         unsigned i;
 440
 441         dst->version            = src->version;
 442         dst->version_min        = src->version_min;
 443         dst->seq                = src->seq;
 444         dst->uuid               = src->uuid;
 445         dst->user_uuid          = src->user_uuid;
 446         memcpy(dst->label,      src->label, sizeof(dst->label));
 447
 448         dst->block_size         = src->block_size;
 449         dst->nr_devices         = src->nr_devices;
 450
 451         dst->time_base_lo       = src->time_base_lo;
 452         dst->time_base_hi       = src->time_base_hi;
 453         dst->time_precision     = src->time_precision;
 454
 455         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 456         memcpy(dst->features,   src->features,  sizeof(dst->features));
 457         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 458
 459         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 460                 int d;
 461
 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
 468                 d = (src_f ? le32_to_cpu(src_f->u64s) : 0) -
 469                     (dst_f ? le32_to_cpu(dst_f->u64s) : 0);
 470                 if (d > 0) {
 471                         int ret = bch2_sb_realloc(dst_handle, le32_to_cpu(dst_handle->sb->u64s) + d);
 472                         if (ret)
 473                                 return ret;
 474
 475                         dst = dst_handle->sb;
 476                         dst_f = bch2_sb_field_get(dst, i);
 477                 }
 478
 479                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 480                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 481
 482                 if (src_f)
 483                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 484         }
 485
 486         return 0;
 487 }
 488
 489 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 490 {
 491         int ret;
 492
 493         lockdep_assert_held(&c->sb_lock);
 494
 495         ret =   bch2_sb_realloc(&c->disk_sb, 0) ?:
 496                 __copy_super(&c->disk_sb, src) ?:
 497                 bch2_sb_replicas_to_cpu_replicas(c) ?:
 498                 bch2_sb_disk_groups_to_cpu(c);
 499         if (ret)
 500                 return ret;
 501
 502         bch2_sb_update(c);
 503         return 0;
 504 }
 505
 506 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 507 {
 508         return __copy_super(&ca->disk_sb, c->disk_sb.sb);
 509 }
 510
 511 /* read superblock: */
 512
 513 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 514 {
 515         struct bch_csum csum;
 516         u32 version, version_min;
 517         size_t bytes;
 518         int ret;
 519 reread:
 520         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 521         sb->bio->bi_iter.bi_sector = offset;
 522         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 523
 524         ret = submit_bio_wait(sb->bio);
 525         if (ret) {
 526                 prt_printf(err, "IO error: %i", ret);
 527                 return ret;
 528         }
 529
 530         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC) &&
 531             uuid_le_cmp(sb->sb->magic, BCHFS_MAGIC)) {
 532                 prt_printf(err, "Not a bcachefs superblock");
 533                 return -BCH_ERR_invalid_sb_magic;
 534         }
 535
 536         version         = le16_to_cpu(sb->sb->version);
 537         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 538                 ? le16_to_cpu(sb->sb->version_min)
 539                 : version;
 540
 541         if (version    >= bcachefs_metadata_version_max) {
 542                 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
 543                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 544                 return -BCH_ERR_invalid_sb_version;
 545         }
 546
 547         if (version_min < bcachefs_metadata_version_min) {
 548                 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
 549                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 550                 return -BCH_ERR_invalid_sb_version;
 551         }
 552
 553         bytes = vstruct_bytes(sb->sb);
 554
 555         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 556                 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 557                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 558                 return -BCH_ERR_invalid_sb_too_big;
 559         }
 560
 561         if (bytes > sb->buffer_size) {
 562                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 563                         return -ENOMEM;
 564                 goto reread;
 565         }
 566
 567         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
 568                 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 569                 return -BCH_ERR_invalid_sb_csum_type;
 570         }
 571
 572         /* XXX: verify MACs */
 573         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 574                             null_nonce(), sb->sb);
 575
 576         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 577                 prt_printf(err, "bad checksum");
 578                 return -BCH_ERR_invalid_sb_csum;
 579         }
 580
 581         sb->seq = le64_to_cpu(sb->sb->seq);
 582
 583         return 0;
 584 }
 585
 586 int bch2_read_super(const char *path, struct bch_opts *opts,
 587                     struct bch_sb_handle *sb)
 588 {
 589         u64 offset = opt_get(*opts, sb);
 590         struct bch_sb_layout layout;
 591         struct printbuf err = PRINTBUF;
 592         __le64 *i;
 593         int ret;
 594
 595         pr_verbose_init(*opts, "");
 596
 597         memset(sb, 0, sizeof(*sb));
 598         sb->mode        = FMODE_READ;
 599         sb->have_bio    = true;
 600
 601         if (!opt_get(*opts, noexcl))
 602                 sb->mode |= FMODE_EXCL;
 603
 604         if (!opt_get(*opts, nochanges))
 605                 sb->mode |= FMODE_WRITE;
 606
 607         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 608         if (IS_ERR(sb->bdev) &&
 609             PTR_ERR(sb->bdev) == -EACCES &&
 610             opt_get(*opts, read_only)) {
 611                 sb->mode &= ~FMODE_WRITE;
 612
 613                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 614                 if (!IS_ERR(sb->bdev))
 615                         opt_set(*opts, nochanges, true);
 616         }
 617
 618         if (IS_ERR(sb->bdev)) {
 619                 ret = PTR_ERR(sb->bdev);
 620                 goto out;
 621         }
 622
 623         ret = bch2_sb_realloc(sb, 0);
 624         if (ret) {
 625                 prt_printf(&err, "error allocating memory for superblock");
 626                 goto err;
 627         }
 628
 629         if (bch2_fs_init_fault("read_super")) {
 630                 prt_printf(&err, "dynamic fault");
 631                 ret = -EFAULT;
 632                 goto err;
 633         }
 634
 635         ret = read_one_super(sb, offset, &err);
 636         if (!ret)
 637                 goto got_super;
 638
 639         if (opt_defined(*opts, sb))
 640                 goto err;
 641
 642         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 643                path, err.buf);
 644         printbuf_reset(&err);
 645
 646         /*
 647          * Error reading primary superblock - read location of backup
 648          * superblocks:
 649          */
 650         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 651         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 652         /*
 653          * use sb buffer to read layout, since sb buffer is page aligned but
 654          * layout won't be:
 655          */
 656         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 657
 658         ret = submit_bio_wait(sb->bio);
 659         if (ret) {
 660                 prt_printf(&err, "IO error: %i", ret);
 661                 goto err;
 662         }
 663
 664         memcpy(&layout, sb->sb, sizeof(layout));
 665         ret = validate_sb_layout(&layout, &err);
 666         if (ret)
 667                 goto err;
 668
 669         for (i = layout.sb_offset;
 670              i < layout.sb_offset + layout.nr_superblocks; i++) {
 671                 offset = le64_to_cpu(*i);
 672
 673                 if (offset == opt_get(*opts, sb))
 674                         continue;
 675
 676                 ret = read_one_super(sb, offset, &err);
 677                 if (!ret)
 678                         goto got_super;
 679         }
 680
 681         goto err;
 682
 683 got_super:
 684         if (le16_to_cpu(sb->sb->block_size) << 9 <
 685             bdev_logical_block_size(sb->bdev)) {
 686                 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
 687                        le16_to_cpu(sb->sb->block_size) << 9,
 688                        bdev_logical_block_size(sb->bdev));
 689                 ret = -BCH_ERR_block_size_too_small;
 690                 goto err;
 691         }
 692
 693         ret = 0;
 694         sb->have_layout = true;
 695
 696         ret = bch2_sb_validate(sb, &err, READ);
 697         if (ret) {
 698                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 699                        path, err.buf);
 700                 goto err_no_print;
 701         }
 702 out:
 703         pr_verbose_init(*opts, "ret %i", ret);
 704         printbuf_exit(&err);
 705         return ret;
 706 err:
 707         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 708                path, err.buf);
 709 err_no_print:
 710         bch2_free_super(sb);
 711         goto out;
 712 }
 713
 714 /* write superblock: */
 715
 716 static void write_super_endio(struct bio *bio)
 717 {
 718         struct bch_dev *ca = bio->bi_private;
 719
 720         /* XXX: return errors directly */
 721
 722         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 723                                bch2_blk_status_to_str(bio->bi_status)))
 724                 ca->sb_write_error = 1;
 725
 726         closure_put(&ca->fs->sb_write);
 727         percpu_ref_put(&ca->io_ref);
 728 }
 729
 730 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 731 {
 732         struct bch_sb *sb = ca->disk_sb.sb;
 733         struct bio *bio = ca->disk_sb.bio;
 734
 735         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 736         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 737         bio->bi_end_io          = write_super_endio;
 738         bio->bi_private         = ca;
 739         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 740
 741         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 742                      bio_sectors(bio));
 743
 744         percpu_ref_get(&ca->io_ref);
 745         closure_bio_submit(bio, &c->sb_write);
 746 }
 747
 748 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 749 {
 750         struct bch_sb *sb = ca->disk_sb.sb;
 751         struct bio *bio = ca->disk_sb.bio;
 752
 753         sb->offset = sb->layout.sb_offset[idx];
 754
 755         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 756         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 757                                 null_nonce(), sb);
 758
 759         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 760         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 761         bio->bi_end_io          = write_super_endio;
 762         bio->bi_private         = ca;
 763         bch2_bio_map(bio, sb,
 764                      roundup((size_t) vstruct_bytes(sb),
 765                              bdev_logical_block_size(ca->disk_sb.bdev)));
 766
 767         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 768                      bio_sectors(bio));
 769
 770         percpu_ref_get(&ca->io_ref);
 771         closure_bio_submit(bio, &c->sb_write);
 772 }
 773
 774 int bch2_write_super(struct bch_fs *c)
 775 {
 776         struct closure *cl = &c->sb_write;
 777         struct bch_dev *ca;
 778         struct printbuf err = PRINTBUF;
 779         unsigned i, sb = 0, nr_wrote;
 780         struct bch_devs_mask sb_written;
 781         bool wrote, can_mount_without_written, can_mount_with_written;
 782         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 783         int ret = 0;
 784
 785         trace_and_count(c, write_super, c, _RET_IP_);
 786
 787         if (c->opts.very_degraded)
 788                 degraded_flags |= BCH_FORCE_IF_LOST;
 789
 790         lockdep_assert_held(&c->sb_lock);
 791
 792         closure_init_stack(cl);
 793         memset(&sb_written, 0, sizeof(sb_written));
 794
 795         if (c->opts.version_upgrade) {
 796                 c->disk_sb.sb->magic = BCHFS_MAGIC;
 797                 c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
 798         }
 799
 800         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 801
 802         if (test_bit(BCH_FS_ERROR, &c->flags))
 803                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 804         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 805                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 806
 807         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 808
 809         bch2_sb_counters_from_cpu(c);
 810
 811         for_each_online_member(ca, c, i)
 812                 bch2_sb_from_fs(c, ca);
 813
 814         for_each_online_member(ca, c, i) {
 815                 printbuf_reset(&err);
 816
 817                 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
 818                 if (ret) {
 819                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 820                         percpu_ref_put(&ca->io_ref);
 821                         goto out;
 822                 }
 823         }
 824
 825         if (c->opts.nochanges)
 826                 goto out;
 827
 828         /*
 829          * Defer writing the superblock until filesystem initialization is
 830          * complete - don't write out a partly initialized superblock:
 831          */
 832         if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
 833                 goto out;
 834
 835         for_each_online_member(ca, c, i) {
 836                 __set_bit(ca->dev_idx, sb_written.d);
 837                 ca->sb_write_error = 0;
 838         }
 839
 840         for_each_online_member(ca, c, i)
 841                 read_back_super(c, ca);
 842         closure_sync(cl);
 843
 844         for_each_online_member(ca, c, i) {
 845                 if (ca->sb_write_error)
 846                         continue;
 847
 848                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 849                         bch2_fs_fatal_error(c,
 850                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 851                                 le64_to_cpu(ca->sb_read_scratch->seq),
 852                                 ca->disk_sb.seq);
 853                         percpu_ref_put(&ca->io_ref);
 854                         ret = -BCH_ERR_erofs_sb_err;
 855                         goto out;
 856                 }
 857
 858                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 859                         bch2_fs_fatal_error(c,
 860                                 "Superblock modified by another process (seq %llu expected %llu)",
 861                                 le64_to_cpu(ca->sb_read_scratch->seq),
 862                                 ca->disk_sb.seq);
 863                         percpu_ref_put(&ca->io_ref);
 864                         ret = -BCH_ERR_erofs_sb_err;
 865                         goto out;
 866                 }
 867         }
 868
 869         do {
 870                 wrote = false;
 871                 for_each_online_member(ca, c, i)
 872                         if (!ca->sb_write_error &&
 873                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 874                                 write_one_super(c, ca, sb);
 875                                 wrote = true;
 876                         }
 877                 closure_sync(cl);
 878                 sb++;
 879         } while (wrote);
 880
 881         for_each_online_member(ca, c, i) {
 882                 if (ca->sb_write_error)
 883                         __clear_bit(ca->dev_idx, sb_written.d);
 884                 else
 885                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 886         }
 887
 888         nr_wrote = dev_mask_nr(&sb_written);
 889
 890         can_mount_with_written =
 891                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 892
 893         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 894                 sb_written.d[i] = ~sb_written.d[i];
 895
 896         can_mount_without_written =
 897                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 898
 899         /*
 900          * If we would be able to mount _without_ the devices we successfully
 901          * wrote superblocks to, we weren't able to write to enough devices:
 902          *
 903          * Exception: if we can mount without the successes because we haven't
 904          * written anything (new filesystem), we continue if we'd be able to
 905          * mount with the devices we did successfully write to:
 906          */
 907         if (bch2_fs_fatal_err_on(!nr_wrote ||
 908                                  !can_mount_with_written ||
 909                                  (can_mount_without_written &&
 910                                   !can_mount_with_written), c,
 911                 "Unable to write superblock to sufficient devices (from %ps)",
 912                 (void *) _RET_IP_))
 913                 ret = -1;
 914 out:
 915         /* Make new options visible after they're persistent: */
 916         bch2_sb_update(c);
 917         printbuf_exit(&err);
 918         return ret;
 919 }
 920
 921 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 922 {
 923         mutex_lock(&c->sb_lock);
 924         if (!(c->sb.features & (1ULL << feat))) {
 925                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 926
 927                 bch2_write_super(c);
 928         }
 929         mutex_unlock(&c->sb_lock);
 930 }
 931
 932 /* BCH_SB_FIELD_members: */
 933
 934 static int bch2_sb_members_validate(struct bch_sb *sb,
 935                                     struct bch_sb_field *f,
 936                                     struct printbuf *err)
 937 {
 938         struct bch_sb_field_members *mi = field_to_type(f, members);
 939         unsigned i;
 940
 941         if ((void *) (mi->members + sb->nr_devices) >
 942             vstruct_end(&mi->field)) {
 943                 prt_printf(err, "too many devices for section size");
 944                 return -BCH_ERR_invalid_sb_members;
 945         }
 946
 947         for (i = 0; i < sb->nr_devices; i++) {
 948                 struct bch_member *m = mi->members + i;
 949
 950                 if (!bch2_member_exists(m))
 951                         continue;
 952
 953                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
 954                         prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
 955                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
 956                         return -BCH_ERR_invalid_sb_members;
 957                 }
 958
 959                 if (le64_to_cpu(m->nbuckets) -
 960                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
 961                         prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
 962                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
 963                         return -BCH_ERR_invalid_sb_members;
 964                 }
 965
 966                 if (le16_to_cpu(m->bucket_size) <
 967                     le16_to_cpu(sb->block_size)) {
 968                         prt_printf(err, "device %u: bucket size %u smaller than block size %u",
 969                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
 970                         return -BCH_ERR_invalid_sb_members;
 971                 }
 972
 973                 if (le16_to_cpu(m->bucket_size) <
 974                     BCH_SB_BTREE_NODE_SIZE(sb)) {
 975                         prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
 976                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
 977                         return -BCH_ERR_invalid_sb_members;
 978                 }
 979         }
 980
 981         return 0;
 982 }
 983
 984 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
 985                                     struct bch_sb_field *f)
 986 {
 987         struct bch_sb_field_members *mi = field_to_type(f, members);
 988         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
 989         unsigned i;
 990
 991         for (i = 0; i < sb->nr_devices; i++) {
 992                 struct bch_member *m = mi->members + i;
 993                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
 994                 u64 bucket_size = le16_to_cpu(m->bucket_size);
 995                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
 996
 997                 if (!bch2_member_exists(m))
 998                         continue;
 999
1000                 prt_printf(out, "Device:");
1001                 prt_tab(out);
1002                 prt_printf(out, "%u", i);
1003                 prt_newline(out);
1004
1005                 printbuf_indent_add(out, 2);
1006
1007                 prt_printf(out, "UUID:");
1008                 prt_tab(out);
1009                 pr_uuid(out, m->uuid.b);
1010                 prt_newline(out);
1011
1012                 prt_printf(out, "Size:");
1013                 prt_tab(out);
1014                 prt_units_u64(out, device_size << 9);
1015                 prt_newline(out);
1016
1017                 prt_printf(out, "Bucket size:");
1018                 prt_tab(out);
1019                 prt_units_u64(out, bucket_size << 9);
1020                 prt_newline(out);
1021
1022                 prt_printf(out, "First bucket:");
1023                 prt_tab(out);
1024                 prt_printf(out, "%u", le16_to_cpu(m->first_bucket));
1025                 prt_newline(out);
1026
1027                 prt_printf(out, "Buckets:");
1028                 prt_tab(out);
1029                 prt_printf(out, "%llu", le64_to_cpu(m->nbuckets));
1030                 prt_newline(out);
1031
1032                 prt_printf(out, "Last mount:");
1033                 prt_tab(out);
1034                 if (m->last_mount)
1035                         pr_time(out, le64_to_cpu(m->last_mount));
1036                 else
1037                         prt_printf(out, "(never)");
1038                 prt_newline(out);
1039
1040                 prt_printf(out, "State:");
1041                 prt_tab(out);
1042                 prt_printf(out, "%s",
1043                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1044                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1045                        : "unknown");
1046                 prt_newline(out);
1047
1048                 prt_printf(out, "Label:");
1049                 prt_tab(out);
1050                 if (BCH_MEMBER_GROUP(m)) {
1051                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1052
1053                         if (idx < disk_groups_nr(gi))
1054                                 prt_printf(out, "%s (%u)",
1055                                        gi->entries[idx].label, idx);
1056                         else
1057                                 prt_printf(out, "(bad disk labels section)");
1058                 } else {
1059                         prt_printf(out, "(none)");
1060                 }
1061                 prt_newline(out);
1062
1063                 prt_printf(out, "Data allowed:");
1064                 prt_tab(out);
1065                 if (BCH_MEMBER_DATA_ALLOWED(m))
1066                         prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(m));
1067                 else
1068                         prt_printf(out, "(none)");
1069                 prt_newline(out);
1070
1071                 prt_printf(out, "Has data:");
1072                 prt_tab(out);
1073                 if (data_have)
1074                         prt_bitflags(out, bch2_data_types, data_have);
1075                 else
1076                         prt_printf(out, "(none)");
1077                 prt_newline(out);
1078
1079                 prt_printf(out, "Discard:");
1080                 prt_tab(out);
1081                 prt_printf(out, "%llu", BCH_MEMBER_DISCARD(m));
1082                 prt_newline(out);
1083
1084                 prt_printf(out, "Freespace initialized:");
1085                 prt_tab(out);
1086                 prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1087                 prt_newline(out);
1088
1089                 printbuf_indent_sub(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                 prt_printf(err, "wrong size (got %zu should be %zu)",
1108                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1109                 return -BCH_ERR_invalid_sb_crypt;
1110         }
1111
1112         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1113                 prt_printf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1114                 return -BCH_ERR_invalid_sb_crypt;
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         prt_printf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1126         prt_newline(out);
1127         prt_printf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1128         prt_newline(out);
1129         prt_printf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1130         prt_newline(out);
1131         prt_printf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1132         prt_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, NULL, 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                 unsafe_memcpy(&u->r, e, replicas_entry_bytes(e),
1253                               "embedded variable length struct");
1254         }
1255
1256         for_each_member_device(ca, c, dev) {
1257                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1258                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1259                 struct jset_entry_dev_usage *u =
1260                         container_of(jset_entry_init(end, b),
1261                                      struct jset_entry_dev_usage, entry);
1262
1263                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1264                 u->dev = cpu_to_le32(dev);
1265                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
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                 prt_printf(err, "wrong size (got %zu should be %zu)",
1350                        vstruct_bytes(&clean->field), sizeof(*clean));
1351                 return -BCH_ERR_invalid_sb_clean;
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         prt_printf(out, "flags:          %x",   le32_to_cpu(clean->flags));
1364         prt_newline(out);
1365         prt_printf(out, "journal_seq:    %llu", le64_to_cpu(clean->journal_seq));
1366         prt_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                 prt_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                 prt_printf(err, "Invalid superblock section %s: %s",
1405                        bch2_sb_fields[type],
1406                        field_err.buf);
1407                 prt_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->nr_tabstops)
1423                 printbuf_tabstop_push(out, 32);
1424
1425         if (ops)
1426                 prt_printf(out, "%s", bch2_sb_fields[type]);
1427         else
1428                 prt_printf(out, "(unknown field %u)", type);
1429
1430         prt_printf(out, " (size %zu):", vstruct_bytes(f));
1431         prt_newline(out);
1432
1433         if (ops && ops->to_text) {
1434                 printbuf_indent_add(out, 2);
1435                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1436                 printbuf_indent_sub(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         prt_printf(out, "Type:                    %u", l->layout_type);
1445         prt_newline(out);
1446
1447         prt_str(out, "Superblock max size:     ");
1448         prt_units_u64(out, 512 << l->sb_max_size_bits);
1449         prt_newline(out);
1450
1451         prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1452         prt_newline(out);
1453
1454         prt_str(out, "Offsets:                 ");
1455         for (i = 0; i < l->nr_superblocks; i++) {
1456                 if (i)
1457                         prt_str(out, ", ");
1458                 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1459         }
1460         prt_newline(out);
1461 }
1462
1463 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1464                      bool print_layout, unsigned fields)
1465 {
1466         struct bch_sb_field_members *mi;
1467         struct bch_sb_field *f;
1468         u64 fields_have = 0;
1469         unsigned nr_devices = 0;
1470
1471         if (!out->nr_tabstops)
1472                 printbuf_tabstop_push(out, 44);
1473
1474         mi = bch2_sb_get_members(sb);
1475         if (mi) {
1476                 struct bch_member *m;
1477
1478                 for (m = mi->members;
1479                      m < mi->members + sb->nr_devices;
1480                      m++)
1481                         nr_devices += bch2_member_exists(m);
1482         }
1483
1484         prt_printf(out, "External UUID:");
1485         prt_tab(out);
1486         pr_uuid(out, sb->user_uuid.b);
1487         prt_newline(out);
1488
1489         prt_printf(out, "Internal UUID:");
1490         prt_tab(out);
1491         pr_uuid(out, sb->uuid.b);
1492         prt_newline(out);
1493
1494         prt_str(out, "Device index:");
1495         prt_tab(out);
1496         prt_printf(out, "%u", sb->dev_idx);
1497         prt_newline(out);
1498
1499         prt_str(out, "Label:");
1500         prt_tab(out);
1501         prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1502         prt_newline(out);
1503
1504         prt_str(out, "Version:");
1505         prt_tab(out);
1506         prt_printf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
1507         prt_newline(out);
1508
1509         prt_printf(out, "Oldest version on disk:");
1510         prt_tab(out);
1511         prt_printf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
1512         prt_newline(out);
1513
1514         prt_printf(out, "Created:");
1515         prt_tab(out);
1516         if (sb->time_base_lo)
1517                 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1518         else
1519                 prt_printf(out, "(not set)");
1520         prt_newline(out);
1521
1522         prt_printf(out, "Sequence number:");
1523         prt_tab(out);
1524         prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1525         prt_newline(out);
1526
1527         prt_printf(out, "Superblock size:");
1528         prt_tab(out);
1529         prt_printf(out, "%zu", vstruct_bytes(sb));
1530         prt_newline(out);
1531
1532         prt_printf(out, "Clean:");
1533         prt_tab(out);
1534         prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1535         prt_newline(out);
1536
1537         prt_printf(out, "Devices:");
1538         prt_tab(out);
1539         prt_printf(out, "%u", nr_devices);
1540         prt_newline(out);
1541
1542         prt_printf(out, "Sections:");
1543         vstruct_for_each(sb, f)
1544                 fields_have |= 1 << le32_to_cpu(f->type);
1545         prt_tab(out);
1546         prt_bitflags(out, bch2_sb_fields, fields_have);
1547         prt_newline(out);
1548
1549         prt_printf(out, "Features:");
1550         prt_tab(out);
1551         prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1552         prt_newline(out);
1553
1554         prt_printf(out, "Compat features:");
1555         prt_tab(out);
1556         prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1557         prt_newline(out);
1558
1559         prt_newline(out);
1560         prt_printf(out, "Options:");
1561         prt_newline(out);
1562         printbuf_indent_add(out, 2);
1563         {
1564                 enum bch_opt_id id;
1565
1566                 for (id = 0; id < bch2_opts_nr; id++) {
1567                         const struct bch_option *opt = bch2_opt_table + id;
1568
1569                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1570                                 u64 v = bch2_opt_from_sb(sb, id);
1571
1572                                 prt_printf(out, "%s:", opt->attr.name);
1573                                 prt_tab(out);
1574                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1575                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1576                                 prt_newline(out);
1577                         }
1578                 }
1579         }
1580
1581         printbuf_indent_sub(out, 2);
1582
1583         if (print_layout) {
1584                 prt_newline(out);
1585                 prt_printf(out, "layout:");
1586                 prt_newline(out);
1587                 printbuf_indent_add(out, 2);
1588                 bch2_sb_layout_to_text(out, &sb->layout);
1589                 printbuf_indent_sub(out, 2);
1590         }
1591
1592         vstruct_for_each(sb, f)
1593                 if (fields & (1 << le32_to_cpu(f->type))) {
1594                         prt_newline(out);
1595                         bch2_sb_field_to_text(out, sb, f);
1596                 }
1597 }