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 void __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                 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 461                         continue;
 462
 463                 src_f = bch2_sb_field_get(src, i);
 464                 dst_f = bch2_sb_field_get(dst, i);
 465                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 466                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 467
 468                 if (src_f)
 469                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 470         }
 471 }
 472
 473 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 474 {
 475         struct bch_sb_field_journal *journal_buckets =
 476                 bch2_sb_get_journal(src);
 477         unsigned journal_u64s = journal_buckets
 478                 ? le32_to_cpu(journal_buckets->field.u64s)
 479                 : 0;
 480         int ret;
 481
 482         lockdep_assert_held(&c->sb_lock);
 483
 484         ret = bch2_sb_realloc(&c->disk_sb,
 485                               le32_to_cpu(src->u64s) - journal_u64s);
 486         if (ret)
 487                 return ret;
 488
 489         __copy_super(&c->disk_sb, src);
 490
 491         ret = bch2_sb_replicas_to_cpu_replicas(c);
 492         if (ret)
 493                 return ret;
 494
 495         ret = bch2_sb_disk_groups_to_cpu(c);
 496         if (ret)
 497                 return ret;
 498
 499         bch2_sb_update(c);
 500         return 0;
 501 }
 502
 503 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 504 {
 505         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 506         struct bch_sb_field_journal *journal_buckets =
 507                 bch2_sb_get_journal(dst);
 508         unsigned journal_u64s = journal_buckets
 509                 ? le32_to_cpu(journal_buckets->field.u64s)
 510                 : 0;
 511         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 512         int ret;
 513
 514         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 515         if (ret)
 516                 return ret;
 517
 518         __copy_super(&ca->disk_sb, src);
 519         return 0;
 520 }
 521
 522 /* read superblock: */
 523
 524 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 525 {
 526         struct bch_csum csum;
 527         u32 version, version_min;
 528         size_t bytes;
 529         int ret;
 530 reread:
 531         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 532         sb->bio->bi_iter.bi_sector = offset;
 533         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 534
 535         ret = submit_bio_wait(sb->bio);
 536         if (ret) {
 537                 prt_printf(err, "IO error: %i", ret);
 538                 return ret;
 539         }
 540
 541         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC) &&
 542             uuid_le_cmp(sb->sb->magic, BCHFS_MAGIC)) {
 543                 prt_printf(err, "Not a bcachefs superblock");
 544                 return -BCH_ERR_invalid_sb_magic;
 545         }
 546
 547         version         = le16_to_cpu(sb->sb->version);
 548         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 549                 ? le16_to_cpu(sb->sb->version_min)
 550                 : version;
 551
 552         if (version    >= bcachefs_metadata_version_max) {
 553                 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
 554                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 555                 return -BCH_ERR_invalid_sb_version;
 556         }
 557
 558         if (version_min < bcachefs_metadata_version_min) {
 559                 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
 560                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 561                 return -BCH_ERR_invalid_sb_version;
 562         }
 563
 564         bytes = vstruct_bytes(sb->sb);
 565
 566         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 567                 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 568                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 569                 return -BCH_ERR_invalid_sb_too_big;
 570         }
 571
 572         if (bytes > sb->buffer_size) {
 573                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 574                         return -ENOMEM;
 575                 goto reread;
 576         }
 577
 578         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
 579                 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 580                 return -BCH_ERR_invalid_sb_csum_type;
 581         }
 582
 583         /* XXX: verify MACs */
 584         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 585                             null_nonce(), sb->sb);
 586
 587         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 588                 prt_printf(err, "bad checksum");
 589                 return -BCH_ERR_invalid_sb_csum;
 590         }
 591
 592         sb->seq = le64_to_cpu(sb->sb->seq);
 593
 594         return 0;
 595 }
 596
 597 int bch2_read_super(const char *path, struct bch_opts *opts,
 598                     struct bch_sb_handle *sb)
 599 {
 600         u64 offset = opt_get(*opts, sb);
 601         struct bch_sb_layout layout;
 602         struct printbuf err = PRINTBUF;
 603         __le64 *i;
 604         int ret;
 605
 606         pr_verbose_init(*opts, "");
 607
 608         memset(sb, 0, sizeof(*sb));
 609         sb->mode        = FMODE_READ;
 610         sb->have_bio    = true;
 611
 612         if (!opt_get(*opts, noexcl))
 613                 sb->mode |= FMODE_EXCL;
 614
 615         if (!opt_get(*opts, nochanges))
 616                 sb->mode |= FMODE_WRITE;
 617
 618         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 619         if (IS_ERR(sb->bdev) &&
 620             PTR_ERR(sb->bdev) == -EACCES &&
 621             opt_get(*opts, read_only)) {
 622                 sb->mode &= ~FMODE_WRITE;
 623
 624                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 625                 if (!IS_ERR(sb->bdev))
 626                         opt_set(*opts, nochanges, true);
 627         }
 628
 629         if (IS_ERR(sb->bdev)) {
 630                 ret = PTR_ERR(sb->bdev);
 631                 goto out;
 632         }
 633
 634         ret = bch2_sb_realloc(sb, 0);
 635         if (ret) {
 636                 prt_printf(&err, "error allocating memory for superblock");
 637                 goto err;
 638         }
 639
 640         if (bch2_fs_init_fault("read_super")) {
 641                 prt_printf(&err, "dynamic fault");
 642                 ret = -EFAULT;
 643                 goto err;
 644         }
 645
 646         ret = read_one_super(sb, offset, &err);
 647         if (!ret)
 648                 goto got_super;
 649
 650         if (opt_defined(*opts, sb))
 651                 goto err;
 652
 653         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 654                path, err.buf);
 655         printbuf_reset(&err);
 656
 657         /*
 658          * Error reading primary superblock - read location of backup
 659          * superblocks:
 660          */
 661         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 662         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 663         /*
 664          * use sb buffer to read layout, since sb buffer is page aligned but
 665          * layout won't be:
 666          */
 667         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 668
 669         ret = submit_bio_wait(sb->bio);
 670         if (ret) {
 671                 prt_printf(&err, "IO error: %i", ret);
 672                 goto err;
 673         }
 674
 675         memcpy(&layout, sb->sb, sizeof(layout));
 676         ret = validate_sb_layout(&layout, &err);
 677         if (ret)
 678                 goto err;
 679
 680         for (i = layout.sb_offset;
 681              i < layout.sb_offset + layout.nr_superblocks; i++) {
 682                 offset = le64_to_cpu(*i);
 683
 684                 if (offset == opt_get(*opts, sb))
 685                         continue;
 686
 687                 ret = read_one_super(sb, offset, &err);
 688                 if (!ret)
 689                         goto got_super;
 690         }
 691
 692         goto err;
 693
 694 got_super:
 695         if (le16_to_cpu(sb->sb->block_size) << 9 <
 696             bdev_logical_block_size(sb->bdev)) {
 697                 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
 698                        le16_to_cpu(sb->sb->block_size) << 9,
 699                        bdev_logical_block_size(sb->bdev));
 700                 ret = -BCH_ERR_block_size_too_small;
 701                 goto err;
 702         }
 703
 704         ret = 0;
 705         sb->have_layout = true;
 706
 707         ret = bch2_sb_validate(sb, &err, READ);
 708         if (ret) {
 709                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 710                        path, err.buf);
 711                 goto err_no_print;
 712         }
 713 out:
 714         pr_verbose_init(*opts, "ret %i", ret);
 715         printbuf_exit(&err);
 716         return ret;
 717 err:
 718         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 719                path, err.buf);
 720 err_no_print:
 721         bch2_free_super(sb);
 722         goto out;
 723 }
 724
 725 /* write superblock: */
 726
 727 static void write_super_endio(struct bio *bio)
 728 {
 729         struct bch_dev *ca = bio->bi_private;
 730
 731         /* XXX: return errors directly */
 732
 733         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 734                                bch2_blk_status_to_str(bio->bi_status)))
 735                 ca->sb_write_error = 1;
 736
 737         closure_put(&ca->fs->sb_write);
 738         percpu_ref_put(&ca->io_ref);
 739 }
 740
 741 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 742 {
 743         struct bch_sb *sb = ca->disk_sb.sb;
 744         struct bio *bio = ca->disk_sb.bio;
 745
 746         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 747         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 748         bio->bi_end_io          = write_super_endio;
 749         bio->bi_private         = ca;
 750         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 751
 752         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 753                      bio_sectors(bio));
 754
 755         percpu_ref_get(&ca->io_ref);
 756         closure_bio_submit(bio, &c->sb_write);
 757 }
 758
 759 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 760 {
 761         struct bch_sb *sb = ca->disk_sb.sb;
 762         struct bio *bio = ca->disk_sb.bio;
 763
 764         sb->offset = sb->layout.sb_offset[idx];
 765
 766         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 767         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 768                                 null_nonce(), sb);
 769
 770         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 771         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 772         bio->bi_end_io          = write_super_endio;
 773         bio->bi_private         = ca;
 774         bch2_bio_map(bio, sb,
 775                      roundup((size_t) vstruct_bytes(sb),
 776                              bdev_logical_block_size(ca->disk_sb.bdev)));
 777
 778         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 779                      bio_sectors(bio));
 780
 781         percpu_ref_get(&ca->io_ref);
 782         closure_bio_submit(bio, &c->sb_write);
 783 }
 784
 785 int bch2_write_super(struct bch_fs *c)
 786 {
 787         struct closure *cl = &c->sb_write;
 788         struct bch_dev *ca;
 789         struct printbuf err = PRINTBUF;
 790         unsigned i, sb = 0, nr_wrote;
 791         struct bch_devs_mask sb_written;
 792         bool wrote, can_mount_without_written, can_mount_with_written;
 793         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 794         int ret = 0;
 795
 796         trace_and_count(c, write_super, c, _RET_IP_);
 797
 798         if (c->opts.very_degraded)
 799                 degraded_flags |= BCH_FORCE_IF_LOST;
 800
 801         lockdep_assert_held(&c->sb_lock);
 802
 803         closure_init_stack(cl);
 804         memset(&sb_written, 0, sizeof(sb_written));
 805
 806         if (c->opts.version_upgrade) {
 807                 c->disk_sb.sb->magic = BCHFS_MAGIC;
 808                 c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
 809         }
 810
 811         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 812
 813         if (test_bit(BCH_FS_ERROR, &c->flags))
 814                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 815         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 816                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 817
 818         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 819
 820         bch2_sb_counters_from_cpu(c);
 821
 822         for_each_online_member(ca, c, i)
 823                 bch2_sb_from_fs(c, ca);
 824
 825         for_each_online_member(ca, c, i) {
 826                 printbuf_reset(&err);
 827
 828                 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
 829                 if (ret) {
 830                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 831                         percpu_ref_put(&ca->io_ref);
 832                         goto out;
 833                 }
 834         }
 835
 836         if (c->opts.nochanges)
 837                 goto out;
 838
 839         /*
 840          * Defer writing the superblock until filesystem initialization is
 841          * complete - don't write out a partly initialized superblock:
 842          */
 843         if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
 844                 goto out;
 845
 846         for_each_online_member(ca, c, i) {
 847                 __set_bit(ca->dev_idx, sb_written.d);
 848                 ca->sb_write_error = 0;
 849         }
 850
 851         for_each_online_member(ca, c, i)
 852                 read_back_super(c, ca);
 853         closure_sync(cl);
 854
 855         for_each_online_member(ca, c, i) {
 856                 if (ca->sb_write_error)
 857                         continue;
 858
 859                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 860                         bch2_fs_fatal_error(c,
 861                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 862                                 le64_to_cpu(ca->sb_read_scratch->seq),
 863                                 ca->disk_sb.seq);
 864                         percpu_ref_put(&ca->io_ref);
 865                         ret = -BCH_ERR_erofs_sb_err;
 866                         goto out;
 867                 }
 868
 869                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 870                         bch2_fs_fatal_error(c,
 871                                 "Superblock modified by another process (seq %llu expected %llu)",
 872                                 le64_to_cpu(ca->sb_read_scratch->seq),
 873                                 ca->disk_sb.seq);
 874                         percpu_ref_put(&ca->io_ref);
 875                         ret = -BCH_ERR_erofs_sb_err;
 876                         goto out;
 877                 }
 878         }
 879
 880         do {
 881                 wrote = false;
 882                 for_each_online_member(ca, c, i)
 883                         if (!ca->sb_write_error &&
 884                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 885                                 write_one_super(c, ca, sb);
 886                                 wrote = true;
 887                         }
 888                 closure_sync(cl);
 889                 sb++;
 890         } while (wrote);
 891
 892         for_each_online_member(ca, c, i) {
 893                 if (ca->sb_write_error)
 894                         __clear_bit(ca->dev_idx, sb_written.d);
 895                 else
 896                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 897         }
 898
 899         nr_wrote = dev_mask_nr(&sb_written);
 900
 901         can_mount_with_written =
 902                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 903
 904         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 905                 sb_written.d[i] = ~sb_written.d[i];
 906
 907         can_mount_without_written =
 908                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 909
 910         /*
 911          * If we would be able to mount _without_ the devices we successfully
 912          * wrote superblocks to, we weren't able to write to enough devices:
 913          *
 914          * Exception: if we can mount without the successes because we haven't
 915          * written anything (new filesystem), we continue if we'd be able to
 916          * mount with the devices we did successfully write to:
 917          */
 918         if (bch2_fs_fatal_err_on(!nr_wrote ||
 919                                  !can_mount_with_written ||
 920                                  (can_mount_without_written &&
 921                                   !can_mount_with_written), c,
 922                 "Unable to write superblock to sufficient devices (from %ps)",
 923                 (void *) _RET_IP_))
 924                 ret = -1;
 925 out:
 926         /* Make new options visible after they're persistent: */
 927         bch2_sb_update(c);
 928         printbuf_exit(&err);
 929         return ret;
 930 }
 931
 932 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 933 {
 934         mutex_lock(&c->sb_lock);
 935         if (!(c->sb.features & (1ULL << feat))) {
 936                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 937
 938                 bch2_write_super(c);
 939         }
 940         mutex_unlock(&c->sb_lock);
 941 }
 942
 943 /* BCH_SB_FIELD_members: */
 944
 945 static int bch2_sb_members_validate(struct bch_sb *sb,
 946                                     struct bch_sb_field *f,
 947                                     struct printbuf *err)
 948 {
 949         struct bch_sb_field_members *mi = field_to_type(f, members);
 950         unsigned i;
 951
 952         if ((void *) (mi->members + sb->nr_devices) >
 953             vstruct_end(&mi->field)) {
 954                 prt_printf(err, "too many devices for section size");
 955                 return -BCH_ERR_invalid_sb_members;
 956         }
 957
 958         for (i = 0; i < sb->nr_devices; i++) {
 959                 struct bch_member *m = mi->members + i;
 960
 961                 if (!bch2_member_exists(m))
 962                         continue;
 963
 964                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
 965                         prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
 966                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
 967                         return -BCH_ERR_invalid_sb_members;
 968                 }
 969
 970                 if (le64_to_cpu(m->nbuckets) -
 971                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
 972                         prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
 973                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
 974                         return -BCH_ERR_invalid_sb_members;
 975                 }
 976
 977                 if (le16_to_cpu(m->bucket_size) <
 978                     le16_to_cpu(sb->block_size)) {
 979                         prt_printf(err, "device %u: bucket size %u smaller than block size %u",
 980                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
 981                         return -BCH_ERR_invalid_sb_members;
 982                 }
 983
 984                 if (le16_to_cpu(m->bucket_size) <
 985                     BCH_SB_BTREE_NODE_SIZE(sb)) {
 986                         prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
 987                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
 988                         return -BCH_ERR_invalid_sb_members;
 989                 }
 990         }
 991
 992         return 0;
 993 }
 994
 995 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
 996                                     struct bch_sb_field *f)
 997 {
 998         struct bch_sb_field_members *mi = field_to_type(f, members);
 999         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
1000         unsigned i;
1001
1002         for (i = 0; i < sb->nr_devices; i++) {
1003                 struct bch_member *m = mi->members + i;
1004                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
1005                 u64 bucket_size = le16_to_cpu(m->bucket_size);
1006                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1007
1008                 if (!bch2_member_exists(m))
1009                         continue;
1010
1011                 prt_printf(out, "Device:");
1012                 prt_tab(out);
1013                 prt_printf(out, "%u", i);
1014                 prt_newline(out);
1015
1016                 printbuf_indent_add(out, 2);
1017
1018                 prt_printf(out, "UUID:");
1019                 prt_tab(out);
1020                 pr_uuid(out, m->uuid.b);
1021                 prt_newline(out);
1022
1023                 prt_printf(out, "Size:");
1024                 prt_tab(out);
1025                 prt_units_u64(out, device_size << 9);
1026                 prt_newline(out);
1027
1028                 prt_printf(out, "Bucket size:");
1029                 prt_tab(out);
1030                 prt_units_u64(out, bucket_size << 9);
1031                 prt_newline(out);
1032
1033                 prt_printf(out, "First bucket:");
1034                 prt_tab(out);
1035                 prt_printf(out, "%u", le16_to_cpu(m->first_bucket));
1036                 prt_newline(out);
1037
1038                 prt_printf(out, "Buckets:");
1039                 prt_tab(out);
1040                 prt_printf(out, "%llu", le64_to_cpu(m->nbuckets));
1041                 prt_newline(out);
1042
1043                 prt_printf(out, "Last mount:");
1044                 prt_tab(out);
1045                 if (m->last_mount)
1046                         pr_time(out, le64_to_cpu(m->last_mount));
1047                 else
1048                         prt_printf(out, "(never)");
1049                 prt_newline(out);
1050
1051                 prt_printf(out, "State:");
1052                 prt_tab(out);
1053                 prt_printf(out, "%s",
1054                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1055                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1056                        : "unknown");
1057                 prt_newline(out);
1058
1059                 prt_printf(out, "Label:");
1060                 prt_tab(out);
1061                 if (BCH_MEMBER_GROUP(m)) {
1062                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1063
1064                         if (idx < disk_groups_nr(gi))
1065                                 prt_printf(out, "%s (%u)",
1066                                        gi->entries[idx].label, idx);
1067                         else
1068                                 prt_printf(out, "(bad disk labels section)");
1069                 } else {
1070                         prt_printf(out, "(none)");
1071                 }
1072                 prt_newline(out);
1073
1074                 prt_printf(out, "Data allowed:");
1075                 prt_tab(out);
1076                 if (BCH_MEMBER_DATA_ALLOWED(m))
1077                         prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(m));
1078                 else
1079                         prt_printf(out, "(none)");
1080                 prt_newline(out);
1081
1082                 prt_printf(out, "Has data:");
1083                 prt_tab(out);
1084                 if (data_have)
1085                         prt_bitflags(out, bch2_data_types, data_have);
1086                 else
1087                         prt_printf(out, "(none)");
1088                 prt_newline(out);
1089
1090                 prt_printf(out, "Discard:");
1091                 prt_tab(out);
1092                 prt_printf(out, "%llu", BCH_MEMBER_DISCARD(m));
1093                 prt_newline(out);
1094
1095                 prt_printf(out, "Freespace initialized:");
1096                 prt_tab(out);
1097                 prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1098                 prt_newline(out);
1099
1100                 printbuf_indent_sub(out, 2);
1101         }
1102 }
1103
1104 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1105         .validate       = bch2_sb_members_validate,
1106         .to_text        = bch2_sb_members_to_text,
1107 };
1108
1109 /* BCH_SB_FIELD_crypt: */
1110
1111 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1112                                   struct bch_sb_field *f,
1113                                   struct printbuf *err)
1114 {
1115         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1116
1117         if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1118                 prt_printf(err, "wrong size (got %zu should be %zu)",
1119                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1120                 return -BCH_ERR_invalid_sb_crypt;
1121         }
1122
1123         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1124                 prt_printf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1125                 return -BCH_ERR_invalid_sb_crypt;
1126         }
1127
1128         return 0;
1129 }
1130
1131 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1132                                   struct bch_sb_field *f)
1133 {
1134         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1135
1136         prt_printf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1137         prt_newline(out);
1138         prt_printf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1139         prt_newline(out);
1140         prt_printf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1141         prt_newline(out);
1142         prt_printf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1143         prt_newline(out);
1144 }
1145
1146 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1147         .validate       = bch2_sb_crypt_validate,
1148         .to_text        = bch2_sb_crypt_to_text,
1149 };
1150
1151 /* BCH_SB_FIELD_clean: */
1152
1153 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1154 {
1155         struct jset_entry *entry;
1156         int ret;
1157
1158         for (entry = clean->start;
1159              entry < (struct jset_entry *) vstruct_end(&clean->field);
1160              entry = vstruct_next(entry)) {
1161                 ret = bch2_journal_entry_validate(c, NULL, entry,
1162                                                   le16_to_cpu(c->disk_sb.sb->version),
1163                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1164                                                   write);
1165                 if (ret)
1166                         return ret;
1167         }
1168
1169         return 0;
1170 }
1171
1172 int bch2_fs_mark_dirty(struct bch_fs *c)
1173 {
1174         int ret;
1175
1176         /*
1177          * Unconditionally write superblock, to verify it hasn't changed before
1178          * we go rw:
1179          */
1180
1181         mutex_lock(&c->sb_lock);
1182         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1183         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1184         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1185         ret = bch2_write_super(c);
1186         mutex_unlock(&c->sb_lock);
1187
1188         return ret;
1189 }
1190
1191 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1192 {
1193         struct jset_entry *entry = *end;
1194         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1195
1196         memset(entry, 0, u64s * sizeof(u64));
1197         /*
1198          * The u64s field counts from the start of data, ignoring the shared
1199          * fields.
1200          */
1201         entry->u64s = cpu_to_le16(u64s - 1);
1202
1203         *end = vstruct_next(*end);
1204         return entry;
1205 }
1206
1207 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1208                                            struct jset_entry **end,
1209                                            u64 journal_seq)
1210 {
1211         struct bch_dev *ca;
1212         unsigned i, dev;
1213
1214         percpu_down_read(&c->mark_lock);
1215
1216         if (!journal_seq) {
1217                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1218                         bch2_fs_usage_acc_to_base(c, i);
1219         } else {
1220                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1221         }
1222
1223         {
1224                 struct jset_entry_usage *u =
1225                         container_of(jset_entry_init(end, sizeof(*u)),
1226                                      struct jset_entry_usage, entry);
1227
1228                 u->entry.type   = BCH_JSET_ENTRY_usage;
1229                 u->entry.btree_id = BCH_FS_USAGE_inodes;
1230                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1231         }
1232
1233         {
1234                 struct jset_entry_usage *u =
1235                         container_of(jset_entry_init(end, sizeof(*u)),
1236                                      struct jset_entry_usage, entry);
1237
1238                 u->entry.type   = BCH_JSET_ENTRY_usage;
1239                 u->entry.btree_id = BCH_FS_USAGE_key_version;
1240                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1241         }
1242
1243         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1244                 struct jset_entry_usage *u =
1245                         container_of(jset_entry_init(end, sizeof(*u)),
1246                                      struct jset_entry_usage, entry);
1247
1248                 u->entry.type   = BCH_JSET_ENTRY_usage;
1249                 u->entry.btree_id = BCH_FS_USAGE_reserved;
1250                 u->entry.level  = i;
1251                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1252         }
1253
1254         for (i = 0; i < c->replicas.nr; i++) {
1255                 struct bch_replicas_entry *e =
1256                         cpu_replicas_entry(&c->replicas, i);
1257                 struct jset_entry_data_usage *u =
1258                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1259                                      struct jset_entry_data_usage, entry);
1260
1261                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1262                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1263                 unsafe_memcpy(&u->r, e, replicas_entry_bytes(e),
1264                               "embedded variable length struct");
1265         }
1266
1267         for_each_member_device(ca, c, dev) {
1268                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1269                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1270                 struct jset_entry_dev_usage *u =
1271                         container_of(jset_entry_init(end, b),
1272                                      struct jset_entry_dev_usage, entry);
1273
1274                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1275                 u->dev = cpu_to_le32(dev);
1276                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1277
1278                 for (i = 0; i < BCH_DATA_NR; i++) {
1279                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1280                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1281                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1282                 }
1283         }
1284
1285         percpu_up_read(&c->mark_lock);
1286
1287         for (i = 0; i < 2; i++) {
1288                 struct jset_entry_clock *clock =
1289                         container_of(jset_entry_init(end, sizeof(*clock)),
1290                                      struct jset_entry_clock, entry);
1291
1292                 clock->entry.type = BCH_JSET_ENTRY_clock;
1293                 clock->rw       = i;
1294                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1295         }
1296 }
1297
1298 void bch2_fs_mark_clean(struct bch_fs *c)
1299 {
1300         struct bch_sb_field_clean *sb_clean;
1301         struct jset_entry *entry;
1302         unsigned u64s;
1303         int ret;
1304
1305         mutex_lock(&c->sb_lock);
1306         if (BCH_SB_CLEAN(c->disk_sb.sb))
1307                 goto out;
1308
1309         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1310
1311         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1312         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1313         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1314         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1315
1316         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1317
1318         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1319         if (!sb_clean) {
1320                 bch_err(c, "error resizing superblock while setting filesystem clean");
1321                 goto out;
1322         }
1323
1324         sb_clean->flags         = 0;
1325         sb_clean->journal_seq   = cpu_to_le64(atomic64_read(&c->journal.seq));
1326
1327         /* Trying to catch outstanding bug: */
1328         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1329
1330         entry = sb_clean->start;
1331         bch2_journal_super_entries_add_common(c, &entry, 0);
1332         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1333         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1334
1335         memset(entry, 0,
1336                vstruct_end(&sb_clean->field) - (void *) entry);
1337
1338         /*
1339          * this should be in the write path, and we should be validating every
1340          * superblock section:
1341          */
1342         ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1343         if (ret) {
1344                 bch_err(c, "error writing marking filesystem clean: validate error");
1345                 goto out;
1346         }
1347
1348         bch2_write_super(c);
1349 out:
1350         mutex_unlock(&c->sb_lock);
1351 }
1352
1353 static int bch2_sb_clean_validate(struct bch_sb *sb,
1354                                   struct bch_sb_field *f,
1355                                   struct printbuf *err)
1356 {
1357         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1358
1359         if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1360                 prt_printf(err, "wrong size (got %zu should be %zu)",
1361                        vstruct_bytes(&clean->field), sizeof(*clean));
1362                 return -BCH_ERR_invalid_sb_clean;
1363         }
1364
1365         return 0;
1366 }
1367
1368 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1369                                   struct bch_sb_field *f)
1370 {
1371         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1372         struct jset_entry *entry;
1373
1374         prt_printf(out, "flags:          %x",   le32_to_cpu(clean->flags));
1375         prt_newline(out);
1376         prt_printf(out, "journal_seq:    %llu", le64_to_cpu(clean->journal_seq));
1377         prt_newline(out);
1378
1379         for (entry = clean->start;
1380              entry != vstruct_end(&clean->field);
1381              entry = vstruct_next(entry)) {
1382                 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1383                     !entry->u64s)
1384                         continue;
1385
1386                 bch2_journal_entry_to_text(out, NULL, entry);
1387                 prt_newline(out);
1388         }
1389 }
1390
1391 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1392         .validate       = bch2_sb_clean_validate,
1393         .to_text        = bch2_sb_clean_to_text,
1394 };
1395
1396 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1397 #define x(f, nr)                                        \
1398         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1399         BCH_SB_FIELDS()
1400 #undef x
1401 };
1402
1403 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1404                                   struct printbuf *err)
1405 {
1406         unsigned type = le32_to_cpu(f->type);
1407         struct printbuf field_err = PRINTBUF;
1408         int ret;
1409
1410         if (type >= BCH_SB_FIELD_NR)
1411                 return 0;
1412
1413         ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
1414         if (ret) {
1415                 prt_printf(err, "Invalid superblock section %s: %s",
1416                        bch2_sb_fields[type],
1417                        field_err.buf);
1418                 prt_newline(err);
1419                 bch2_sb_field_to_text(err, sb, f);
1420         }
1421
1422         printbuf_exit(&field_err);
1423         return ret;
1424 }
1425
1426 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1427                            struct bch_sb_field *f)
1428 {
1429         unsigned type = le32_to_cpu(f->type);
1430         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1431                 ? bch2_sb_field_ops[type] : NULL;
1432
1433         if (!out->nr_tabstops)
1434                 printbuf_tabstop_push(out, 32);
1435
1436         if (ops)
1437                 prt_printf(out, "%s", bch2_sb_fields[type]);
1438         else
1439                 prt_printf(out, "(unknown field %u)", type);
1440
1441         prt_printf(out, " (size %zu):", vstruct_bytes(f));
1442         prt_newline(out);
1443
1444         if (ops && ops->to_text) {
1445                 printbuf_indent_add(out, 2);
1446                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1447                 printbuf_indent_sub(out, 2);
1448         }
1449 }
1450
1451 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1452 {
1453         unsigned i;
1454
1455         prt_printf(out, "Type:                    %u", l->layout_type);
1456         prt_newline(out);
1457
1458         prt_str(out, "Superblock max size:     ");
1459         prt_units_u64(out, 512 << l->sb_max_size_bits);
1460         prt_newline(out);
1461
1462         prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1463         prt_newline(out);
1464
1465         prt_str(out, "Offsets:                 ");
1466         for (i = 0; i < l->nr_superblocks; i++) {
1467                 if (i)
1468                         prt_str(out, ", ");
1469                 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1470         }
1471         prt_newline(out);
1472 }
1473
1474 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1475                      bool print_layout, unsigned fields)
1476 {
1477         struct bch_sb_field_members *mi;
1478         struct bch_sb_field *f;
1479         u64 fields_have = 0;
1480         unsigned nr_devices = 0;
1481
1482         if (!out->nr_tabstops)
1483                 printbuf_tabstop_push(out, 44);
1484
1485         mi = bch2_sb_get_members(sb);
1486         if (mi) {
1487                 struct bch_member *m;
1488
1489                 for (m = mi->members;
1490                      m < mi->members + sb->nr_devices;
1491                      m++)
1492                         nr_devices += bch2_member_exists(m);
1493         }
1494
1495         prt_printf(out, "External UUID:");
1496         prt_tab(out);
1497         pr_uuid(out, sb->user_uuid.b);
1498         prt_newline(out);
1499
1500         prt_printf(out, "Internal UUID:");
1501         prt_tab(out);
1502         pr_uuid(out, sb->uuid.b);
1503         prt_newline(out);
1504
1505         prt_str(out, "Device index:");
1506         prt_tab(out);
1507         prt_printf(out, "%u", sb->dev_idx);
1508         prt_newline(out);
1509
1510         prt_str(out, "Label:");
1511         prt_tab(out);
1512         prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1513         prt_newline(out);
1514
1515         prt_str(out, "Version:");
1516         prt_tab(out);
1517         prt_printf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
1518         prt_newline(out);
1519
1520         prt_printf(out, "Oldest version on disk:");
1521         prt_tab(out);
1522         prt_printf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
1523         prt_newline(out);
1524
1525         prt_printf(out, "Created:");
1526         prt_tab(out);
1527         if (sb->time_base_lo)
1528                 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1529         else
1530                 prt_printf(out, "(not set)");
1531         prt_newline(out);
1532
1533         prt_printf(out, "Sequence number:");
1534         prt_tab(out);
1535         prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1536         prt_newline(out);
1537
1538         prt_printf(out, "Superblock size:");
1539         prt_tab(out);
1540         prt_printf(out, "%zu", vstruct_bytes(sb));
1541         prt_newline(out);
1542
1543         prt_printf(out, "Clean:");
1544         prt_tab(out);
1545         prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1546         prt_newline(out);
1547
1548         prt_printf(out, "Devices:");
1549         prt_tab(out);
1550         prt_printf(out, "%u", nr_devices);
1551         prt_newline(out);
1552
1553         prt_printf(out, "Sections:");
1554         vstruct_for_each(sb, f)
1555                 fields_have |= 1 << le32_to_cpu(f->type);
1556         prt_tab(out);
1557         prt_bitflags(out, bch2_sb_fields, fields_have);
1558         prt_newline(out);
1559
1560         prt_printf(out, "Features:");
1561         prt_tab(out);
1562         prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1563         prt_newline(out);
1564
1565         prt_printf(out, "Compat features:");
1566         prt_tab(out);
1567         prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1568         prt_newline(out);
1569
1570         prt_newline(out);
1571         prt_printf(out, "Options:");
1572         prt_newline(out);
1573         printbuf_indent_add(out, 2);
1574         {
1575                 enum bch_opt_id id;
1576
1577                 for (id = 0; id < bch2_opts_nr; id++) {
1578                         const struct bch_option *opt = bch2_opt_table + id;
1579
1580                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1581                                 u64 v = bch2_opt_from_sb(sb, id);
1582
1583                                 prt_printf(out, "%s:", opt->attr.name);
1584                                 prt_tab(out);
1585                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1586                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1587                                 prt_newline(out);
1588                         }
1589                 }
1590         }
1591
1592         printbuf_indent_sub(out, 2);
1593
1594         if (print_layout) {
1595                 prt_newline(out);
1596                 prt_printf(out, "layout:");
1597                 prt_newline(out);
1598                 printbuf_indent_add(out, 2);
1599                 bch2_sb_layout_to_text(out, &sb->layout);
1600                 printbuf_indent_sub(out, 2);
1601         }
1602
1603         vstruct_for_each(sb, f)
1604                 if (fields & (1 << le32_to_cpu(f->type))) {
1605                         prt_newline(out);
1606                         bch2_sb_field_to_text(out, sb, f);
1607                 }
1608 }