git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/super-io.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "btree_update_interior.h"
   5 #include "buckets.h"
   6 #include "checksum.h"
   7 #include "disk_groups.h"
   8 #include "ec.h"
   9 #include "error.h"
  10 #include "io.h"
  11 #include "journal.h"
  12 #include "journal_io.h"
  13 #include "journal_sb.h"
  14 #include "journal_seq_blacklist.h"
  15 #include "replicas.h"
  16 #include "quota.h"
  17 #include "super-io.h"
  18 #include "super.h"
  19 #include "vstructs.h"
  20 #include "counters.h"
  21
  22 #include <linux/backing-dev.h>
  23 #include <linux/pretty-printers.h>
  24 #include <linux/sort.h>
  25
  26 #include <trace/events/bcachefs.h>
  27
  28 const char * const bch2_sb_fields[] = {
  29 #define x(name, nr)     #name,
  30         BCH_SB_FIELDS()
  31 #undef x
  32         NULL
  33 };
  34
  35 static int bch2_sb_field_validate(struct bch_sb *, struct bch_sb_field *,
  36                                   struct printbuf *);
  37
  38 struct bch_sb_field *bch2_sb_field_get(struct bch_sb *sb,
  39                                       enum bch_sb_field_type type)
  40 {
  41         struct bch_sb_field *f;
  42
  43         /* XXX: need locking around superblock to access optional fields */
  44
  45         vstruct_for_each(sb, f)
  46                 if (le32_to_cpu(f->type) == type)
  47                         return f;
  48         return NULL;
  49 }
  50
  51 static struct bch_sb_field *__bch2_sb_field_resize(struct bch_sb_handle *sb,
  52                                                    struct bch_sb_field *f,
  53                                                    unsigned u64s)
  54 {
  55         unsigned old_u64s = f ? le32_to_cpu(f->u64s) : 0;
  56         unsigned sb_u64s = le32_to_cpu(sb->sb->u64s) + u64s - old_u64s;
  57
  58         BUG_ON(__vstruct_bytes(struct bch_sb, sb_u64s) > sb->buffer_size);
  59
  60         if (!f && !u64s) {
  61                 /* nothing to do: */
  62         } else if (!f) {
  63                 f = vstruct_last(sb->sb);
  64                 memset(f, 0, sizeof(u64) * u64s);
  65                 f->u64s = cpu_to_le32(u64s);
  66                 f->type = 0;
  67         } else {
  68                 void *src, *dst;
  69
  70                 src = vstruct_end(f);
  71
  72                 if (u64s) {
  73                         f->u64s = cpu_to_le32(u64s);
  74                         dst = vstruct_end(f);
  75                 } else {
  76                         dst = f;
  77                 }
  78
  79                 memmove(dst, src, vstruct_end(sb->sb) - src);
  80
  81                 if (dst > src)
  82                         memset(src, 0, dst - src);
  83         }
  84
  85         sb->sb->u64s = cpu_to_le32(sb_u64s);
  86
  87         return u64s ? f : NULL;
  88 }
  89
  90 void bch2_sb_field_delete(struct bch_sb_handle *sb,
  91                           enum bch_sb_field_type type)
  92 {
  93         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
  94
  95         if (f)
  96                 __bch2_sb_field_resize(sb, f, 0);
  97 }
  98
  99 /* Superblock realloc/free: */
 100
 101 void bch2_free_super(struct bch_sb_handle *sb)
 102 {
 103         kfree(sb->bio);
 104         if (!IS_ERR_OR_NULL(sb->bdev))
 105                 blkdev_put(sb->bdev, sb->mode);
 106
 107         kfree(sb->sb);
 108         memset(sb, 0, sizeof(*sb));
 109 }
 110
 111 int bch2_sb_realloc(struct bch_sb_handle *sb, unsigned u64s)
 112 {
 113         size_t new_bytes = __vstruct_bytes(struct bch_sb, u64s);
 114         size_t new_buffer_size;
 115         struct bch_sb *new_sb;
 116         struct bio *bio;
 117
 118         if (sb->bdev)
 119                 new_bytes = max_t(size_t, new_bytes, bdev_logical_block_size(sb->bdev));
 120
 121         new_buffer_size = roundup_pow_of_two(new_bytes);
 122
 123         if (sb->sb && sb->buffer_size >= new_buffer_size)
 124                 return 0;
 125
 126         if (sb->have_layout) {
 127                 u64 max_bytes = 512 << sb->sb->layout.sb_max_size_bits;
 128
 129                 if (new_bytes > max_bytes) {
 130                         pr_err("%pg: superblock too big: want %zu but have %llu",
 131                                sb->bdev, new_bytes, max_bytes);
 132                         return -BCH_ERR_ENOSPC_sb;
 133                 }
 134         }
 135
 136         if (sb->buffer_size >= new_buffer_size && sb->sb)
 137                 return 0;
 138
 139         if (dynamic_fault("bcachefs:add:super_realloc"))
 140                 return -ENOMEM;
 141
 142         if (sb->have_bio) {
 143                 unsigned nr_bvecs = DIV_ROUND_UP(new_buffer_size, PAGE_SIZE);
 144
 145                 bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 146                 if (!bio)
 147                         return -ENOMEM;
 148
 149                 bio_init(bio, NULL, bio->bi_inline_vecs, nr_bvecs, 0);
 150
 151                 kfree(sb->bio);
 152                 sb->bio = bio;
 153         }
 154
 155         new_sb = krealloc(sb->sb, new_buffer_size, GFP_NOFS|__GFP_ZERO);
 156         if (!new_sb)
 157                 return -ENOMEM;
 158
 159         sb->sb = new_sb;
 160         sb->buffer_size = new_buffer_size;
 161
 162         return 0;
 163 }
 164
 165 struct bch_sb_field *bch2_sb_field_resize(struct bch_sb_handle *sb,
 166                                           enum bch_sb_field_type type,
 167                                           unsigned u64s)
 168 {
 169         struct bch_sb_field *f = bch2_sb_field_get(sb->sb, type);
 170         ssize_t old_u64s = f ? le32_to_cpu(f->u64s) : 0;
 171         ssize_t d = -old_u64s + u64s;
 172
 173         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d))
 174                 return NULL;
 175
 176         if (sb->fs_sb) {
 177                 struct bch_fs *c = container_of(sb, struct bch_fs, disk_sb);
 178                 struct bch_dev *ca;
 179                 unsigned i;
 180
 181                 lockdep_assert_held(&c->sb_lock);
 182
 183                 /* XXX: we're not checking that offline device have enough space */
 184
 185                 for_each_online_member(ca, c, i) {
 186                         struct bch_sb_handle *sb = &ca->disk_sb;
 187
 188                         if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s) + d)) {
 189                                 percpu_ref_put(&ca->ref);
 190                                 return NULL;
 191                         }
 192                 }
 193         }
 194
 195         f = bch2_sb_field_get(sb->sb, type);
 196         f = __bch2_sb_field_resize(sb, f, u64s);
 197         if (f)
 198                 f->type = cpu_to_le32(type);
 199         return f;
 200 }
 201
 202 /* Superblock validate: */
 203
 204 static inline void __bch2_sb_layout_size_assert(void)
 205 {
 206         BUILD_BUG_ON(sizeof(struct bch_sb_layout) != 512);
 207 }
 208
 209 static int validate_sb_layout(struct bch_sb_layout *layout, struct printbuf *out)
 210 {
 211         u64 offset, prev_offset, max_sectors;
 212         unsigned i;
 213
 214         if (uuid_le_cmp(layout->magic, BCACHE_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                 prt_printf(err, "Not a bcachefs superblock");
 543                 return -BCH_ERR_invalid_sb_magic;
 544         }
 545
 546         version         = le16_to_cpu(sb->sb->version);
 547         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 548                 ? le16_to_cpu(sb->sb->version_min)
 549                 : version;
 550
 551         if (version    >= bcachefs_metadata_version_max) {
 552                 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
 553                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 554                 return -BCH_ERR_invalid_sb_version;
 555         }
 556
 557         if (version_min < bcachefs_metadata_version_min) {
 558                 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
 559                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 560                 return -BCH_ERR_invalid_sb_version;
 561         }
 562
 563         bytes = vstruct_bytes(sb->sb);
 564
 565         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 566                 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 567                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 568                 return -BCH_ERR_invalid_sb_too_big;
 569         }
 570
 571         if (bytes > sb->buffer_size) {
 572                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 573                         return -ENOMEM;
 574                 goto reread;
 575         }
 576
 577         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
 578                 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 579                 return -BCH_ERR_invalid_sb_csum_type;
 580         }
 581
 582         /* XXX: verify MACs */
 583         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 584                             null_nonce(), sb->sb);
 585
 586         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 587                 prt_printf(err, "bad checksum");
 588                 return -BCH_ERR_invalid_sb_csum;
 589         }
 590
 591         sb->seq = le64_to_cpu(sb->sb->seq);
 592
 593         return 0;
 594 }
 595
 596 int bch2_read_super(const char *path, struct bch_opts *opts,
 597                     struct bch_sb_handle *sb)
 598 {
 599         u64 offset = opt_get(*opts, sb);
 600         struct bch_sb_layout layout;
 601         struct printbuf err = PRINTBUF;
 602         __le64 *i;
 603         int ret;
 604
 605         pr_verbose_init(*opts, "");
 606
 607         memset(sb, 0, sizeof(*sb));
 608         sb->mode        = FMODE_READ;
 609         sb->have_bio    = true;
 610
 611         if (!opt_get(*opts, noexcl))
 612                 sb->mode |= FMODE_EXCL;
 613
 614         if (!opt_get(*opts, nochanges))
 615                 sb->mode |= FMODE_WRITE;
 616
 617         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 618         if (IS_ERR(sb->bdev) &&
 619             PTR_ERR(sb->bdev) == -EACCES &&
 620             opt_get(*opts, read_only)) {
 621                 sb->mode &= ~FMODE_WRITE;
 622
 623                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 624                 if (!IS_ERR(sb->bdev))
 625                         opt_set(*opts, nochanges, true);
 626         }
 627
 628         if (IS_ERR(sb->bdev)) {
 629                 ret = PTR_ERR(sb->bdev);
 630                 goto out;
 631         }
 632
 633         ret = bch2_sb_realloc(sb, 0);
 634         if (ret) {
 635                 prt_printf(&err, "error allocating memory for superblock");
 636                 goto err;
 637         }
 638
 639         if (bch2_fs_init_fault("read_super")) {
 640                 prt_printf(&err, "dynamic fault");
 641                 ret = -EFAULT;
 642                 goto err;
 643         }
 644
 645         ret = read_one_super(sb, offset, &err);
 646         if (!ret)
 647                 goto got_super;
 648
 649         if (opt_defined(*opts, sb))
 650                 goto err;
 651
 652         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 653                path, err.buf);
 654         printbuf_reset(&err);
 655
 656         /*
 657          * Error reading primary superblock - read location of backup
 658          * superblocks:
 659          */
 660         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 661         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 662         /*
 663          * use sb buffer to read layout, since sb buffer is page aligned but
 664          * layout won't be:
 665          */
 666         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 667
 668         ret = submit_bio_wait(sb->bio);
 669         if (ret) {
 670                 prt_printf(&err, "IO error: %i", ret);
 671                 goto err;
 672         }
 673
 674         memcpy(&layout, sb->sb, sizeof(layout));
 675         ret = validate_sb_layout(&layout, &err);
 676         if (ret)
 677                 goto err;
 678
 679         for (i = layout.sb_offset;
 680              i < layout.sb_offset + layout.nr_superblocks; i++) {
 681                 offset = le64_to_cpu(*i);
 682
 683                 if (offset == opt_get(*opts, sb))
 684                         continue;
 685
 686                 ret = read_one_super(sb, offset, &err);
 687                 if (!ret)
 688                         goto got_super;
 689         }
 690
 691         goto err;
 692
 693 got_super:
 694         if (le16_to_cpu(sb->sb->block_size) << 9 <
 695             bdev_logical_block_size(sb->bdev)) {
 696                 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
 697                        le16_to_cpu(sb->sb->block_size) << 9,
 698                        bdev_logical_block_size(sb->bdev));
 699                 ret = -BCH_ERR_block_size_too_small;
 700                 goto err;
 701         }
 702
 703         ret = 0;
 704         sb->have_layout = true;
 705
 706         ret = bch2_sb_validate(sb, &err, READ);
 707         if (ret) {
 708                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 709                        path, err.buf);
 710                 goto err_no_print;
 711         }
 712 out:
 713         pr_verbose_init(*opts, "ret %i", ret);
 714         printbuf_exit(&err);
 715         return ret;
 716 err:
 717         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 718                path, err.buf);
 719 err_no_print:
 720         bch2_free_super(sb);
 721         goto out;
 722 }
 723
 724 /* write superblock: */
 725
 726 static void write_super_endio(struct bio *bio)
 727 {
 728         struct bch_dev *ca = bio->bi_private;
 729
 730         /* XXX: return errors directly */
 731
 732         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 733                                bch2_blk_status_to_str(bio->bi_status)))
 734                 ca->sb_write_error = 1;
 735
 736         closure_put(&ca->fs->sb_write);
 737         percpu_ref_put(&ca->io_ref);
 738 }
 739
 740 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 741 {
 742         struct bch_sb *sb = ca->disk_sb.sb;
 743         struct bio *bio = ca->disk_sb.bio;
 744
 745         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 746         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 747         bio->bi_end_io          = write_super_endio;
 748         bio->bi_private         = ca;
 749         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 750
 751         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 752                      bio_sectors(bio));
 753
 754         percpu_ref_get(&ca->io_ref);
 755         closure_bio_submit(bio, &c->sb_write);
 756 }
 757
 758 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 759 {
 760         struct bch_sb *sb = ca->disk_sb.sb;
 761         struct bio *bio = ca->disk_sb.bio;
 762
 763         sb->offset = sb->layout.sb_offset[idx];
 764
 765         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 766         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 767                                 null_nonce(), sb);
 768
 769         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 770         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 771         bio->bi_end_io          = write_super_endio;
 772         bio->bi_private         = ca;
 773         bch2_bio_map(bio, sb,
 774                      roundup((size_t) vstruct_bytes(sb),
 775                              bdev_logical_block_size(ca->disk_sb.bdev)));
 776
 777         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 778                      bio_sectors(bio));
 779
 780         percpu_ref_get(&ca->io_ref);
 781         closure_bio_submit(bio, &c->sb_write);
 782 }
 783
 784 int bch2_write_super(struct bch_fs *c)
 785 {
 786         struct closure *cl = &c->sb_write;
 787         struct bch_dev *ca;
 788         struct printbuf err = PRINTBUF;
 789         unsigned i, sb = 0, nr_wrote;
 790         struct bch_devs_mask sb_written;
 791         bool wrote, can_mount_without_written, can_mount_with_written;
 792         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 793         int ret = 0;
 794
 795         trace_and_count(c, write_super, c, _RET_IP_);
 796
 797         if (c->opts.very_degraded)
 798                 degraded_flags |= BCH_FORCE_IF_LOST;
 799
 800         lockdep_assert_held(&c->sb_lock);
 801
 802         closure_init_stack(cl);
 803         memset(&sb_written, 0, sizeof(sb_written));
 804
 805         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 806
 807         if (test_bit(BCH_FS_ERROR, &c->flags))
 808                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 809         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 810                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 811
 812         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 813
 814         bch2_sb_counters_from_cpu(c);
 815
 816         for_each_online_member(ca, c, i)
 817                 bch2_sb_from_fs(c, ca);
 818
 819         for_each_online_member(ca, c, i) {
 820                 printbuf_reset(&err);
 821
 822                 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
 823                 if (ret) {
 824                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 825                         percpu_ref_put(&ca->io_ref);
 826                         goto out;
 827                 }
 828         }
 829
 830         if (c->opts.nochanges)
 831                 goto out;
 832
 833         /*
 834          * Defer writing the superblock until filesystem initialization is
 835          * complete - don't write out a partly initialized superblock:
 836          */
 837         if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
 838                 goto out;
 839
 840         for_each_online_member(ca, c, i) {
 841                 __set_bit(ca->dev_idx, sb_written.d);
 842                 ca->sb_write_error = 0;
 843         }
 844
 845         for_each_online_member(ca, c, i)
 846                 read_back_super(c, ca);
 847         closure_sync(cl);
 848
 849         for_each_online_member(ca, c, i) {
 850                 if (ca->sb_write_error)
 851                         continue;
 852
 853                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 854                         bch2_fs_fatal_error(c,
 855                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 856                                 le64_to_cpu(ca->sb_read_scratch->seq),
 857                                 ca->disk_sb.seq);
 858                         percpu_ref_put(&ca->io_ref);
 859                         ret = -EROFS;
 860                         goto out;
 861                 }
 862
 863                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 864                         bch2_fs_fatal_error(c,
 865                                 "Superblock modified by another process (seq %llu expected %llu)",
 866                                 le64_to_cpu(ca->sb_read_scratch->seq),
 867                                 ca->disk_sb.seq);
 868                         percpu_ref_put(&ca->io_ref);
 869                         ret = -EROFS;
 870                         goto out;
 871                 }
 872         }
 873
 874         do {
 875                 wrote = false;
 876                 for_each_online_member(ca, c, i)
 877                         if (!ca->sb_write_error &&
 878                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 879                                 write_one_super(c, ca, sb);
 880                                 wrote = true;
 881                         }
 882                 closure_sync(cl);
 883                 sb++;
 884         } while (wrote);
 885
 886         for_each_online_member(ca, c, i) {
 887                 if (ca->sb_write_error)
 888                         __clear_bit(ca->dev_idx, sb_written.d);
 889                 else
 890                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 891         }
 892
 893         nr_wrote = dev_mask_nr(&sb_written);
 894
 895         can_mount_with_written =
 896                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 897
 898         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 899                 sb_written.d[i] = ~sb_written.d[i];
 900
 901         can_mount_without_written =
 902                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 903
 904         /*
 905          * If we would be able to mount _without_ the devices we successfully
 906          * wrote superblocks to, we weren't able to write to enough devices:
 907          *
 908          * Exception: if we can mount without the successes because we haven't
 909          * written anything (new filesystem), we continue if we'd be able to
 910          * mount with the devices we did successfully write to:
 911          */
 912         if (bch2_fs_fatal_err_on(!nr_wrote ||
 913                                  !can_mount_with_written ||
 914                                  (can_mount_without_written &&
 915                                   !can_mount_with_written), c,
 916                 "Unable to write superblock to sufficient devices (from %ps)",
 917                 (void *) _RET_IP_))
 918                 ret = -1;
 919 out:
 920         /* Make new options visible after they're persistent: */
 921         bch2_sb_update(c);
 922         printbuf_exit(&err);
 923         return ret;
 924 }
 925
 926 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 927 {
 928         mutex_lock(&c->sb_lock);
 929         if (!(c->sb.features & (1ULL << feat))) {
 930                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 931
 932                 bch2_write_super(c);
 933         }
 934         mutex_unlock(&c->sb_lock);
 935 }
 936
 937 /* BCH_SB_FIELD_members: */
 938
 939 static int bch2_sb_members_validate(struct bch_sb *sb,
 940                                     struct bch_sb_field *f,
 941                                     struct printbuf *err)
 942 {
 943         struct bch_sb_field_members *mi = field_to_type(f, members);
 944         unsigned i;
 945
 946         if ((void *) (mi->members + sb->nr_devices) >
 947             vstruct_end(&mi->field)) {
 948                 prt_printf(err, "too many devices for section size");
 949                 return -BCH_ERR_invalid_sb_members;
 950         }
 951
 952         for (i = 0; i < sb->nr_devices; i++) {
 953                 struct bch_member *m = mi->members + i;
 954
 955                 if (!bch2_member_exists(m))
 956                         continue;
 957
 958                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
 959                         prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
 960                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
 961                         return -BCH_ERR_invalid_sb_members;
 962                 }
 963
 964                 if (le64_to_cpu(m->nbuckets) -
 965                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
 966                         prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
 967                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
 968                         return -BCH_ERR_invalid_sb_members;
 969                 }
 970
 971                 if (le16_to_cpu(m->bucket_size) <
 972                     le16_to_cpu(sb->block_size)) {
 973                         prt_printf(err, "device %u: bucket size %u smaller than block size %u",
 974                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
 975                         return -BCH_ERR_invalid_sb_members;
 976                 }
 977
 978                 if (le16_to_cpu(m->bucket_size) <
 979                     BCH_SB_BTREE_NODE_SIZE(sb)) {
 980                         prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
 981                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
 982                         return -BCH_ERR_invalid_sb_members;
 983                 }
 984         }
 985
 986         return 0;
 987 }
 988
 989 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
 990                                     struct bch_sb_field *f)
 991 {
 992         struct bch_sb_field_members *mi = field_to_type(f, members);
 993         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
 994         unsigned i;
 995
 996         for (i = 0; i < sb->nr_devices; i++) {
 997                 struct bch_member *m = mi->members + i;
 998                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
 999                 u64 bucket_size = le16_to_cpu(m->bucket_size);
1000                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1001
1002                 if (!bch2_member_exists(m))
1003                         continue;
1004
1005                 prt_printf(out, "Device:");
1006                 prt_tab(out);
1007                 prt_printf(out, "%u", i);
1008                 prt_newline(out);
1009
1010                 printbuf_indent_add(out, 2);
1011
1012                 prt_printf(out, "UUID:");
1013                 prt_tab(out);
1014                 pr_uuid(out, m->uuid.b);
1015                 prt_newline(out);
1016
1017                 prt_printf(out, "Size:");
1018                 prt_tab(out);
1019                 prt_units_u64(out, device_size << 9);
1020                 prt_newline(out);
1021
1022                 prt_printf(out, "Bucket size:");
1023                 prt_tab(out);
1024                 prt_units_u64(out, bucket_size << 9);
1025                 prt_newline(out);
1026
1027                 prt_printf(out, "First bucket:");
1028                 prt_tab(out);
1029                 prt_printf(out, "%u", le16_to_cpu(m->first_bucket));
1030                 prt_newline(out);
1031
1032                 prt_printf(out, "Buckets:");
1033                 prt_tab(out);
1034                 prt_printf(out, "%llu", le64_to_cpu(m->nbuckets));
1035                 prt_newline(out);
1036
1037                 prt_printf(out, "Last mount:");
1038                 prt_tab(out);
1039                 if (m->last_mount)
1040                         pr_time(out, le64_to_cpu(m->last_mount));
1041                 else
1042                         prt_printf(out, "(never)");
1043                 prt_newline(out);
1044
1045                 prt_printf(out, "State:");
1046                 prt_tab(out);
1047                 prt_printf(out, "%s",
1048                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1049                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1050                        : "unknown");
1051                 prt_newline(out);
1052
1053                 prt_printf(out, "Label:");
1054                 prt_tab(out);
1055                 if (BCH_MEMBER_GROUP(m)) {
1056                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1057
1058                         if (idx < disk_groups_nr(gi))
1059                                 prt_printf(out, "%s (%u)",
1060                                        gi->entries[idx].label, idx);
1061                         else
1062                                 prt_printf(out, "(bad disk labels section)");
1063                 } else {
1064                         prt_printf(out, "(none)");
1065                 }
1066                 prt_newline(out);
1067
1068                 prt_printf(out, "Data allowed:");
1069                 prt_tab(out);
1070                 if (BCH_MEMBER_DATA_ALLOWED(m))
1071                         prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(m));
1072                 else
1073                         prt_printf(out, "(none)");
1074                 prt_newline(out);
1075
1076                 prt_printf(out, "Has data:");
1077                 prt_tab(out);
1078                 if (data_have)
1079                         prt_bitflags(out, bch2_data_types, data_have);
1080                 else
1081                         prt_printf(out, "(none)");
1082                 prt_newline(out);
1083
1084                 prt_printf(out, "Discard:");
1085                 prt_tab(out);
1086                 prt_printf(out, "%llu", BCH_MEMBER_DISCARD(m));
1087                 prt_newline(out);
1088
1089                 prt_printf(out, "Freespace initialized:");
1090                 prt_tab(out);
1091                 prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1092                 prt_newline(out);
1093
1094                 printbuf_indent_sub(out, 2);
1095         }
1096 }
1097
1098 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1099         .validate       = bch2_sb_members_validate,
1100         .to_text        = bch2_sb_members_to_text,
1101 };
1102
1103 /* BCH_SB_FIELD_crypt: */
1104
1105 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1106                                   struct bch_sb_field *f,
1107                                   struct printbuf *err)
1108 {
1109         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1110
1111         if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1112                 prt_printf(err, "wrong size (got %zu should be %zu)",
1113                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1114                 return -BCH_ERR_invalid_sb_crypt;
1115         }
1116
1117         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1118                 prt_printf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1119                 return -BCH_ERR_invalid_sb_crypt;
1120         }
1121
1122         return 0;
1123 }
1124
1125 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1126                                   struct bch_sb_field *f)
1127 {
1128         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1129
1130         prt_printf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1131         prt_newline(out);
1132         prt_printf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1133         prt_newline(out);
1134         prt_printf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1135         prt_newline(out);
1136         prt_printf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1137         prt_newline(out);
1138 }
1139
1140 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1141         .validate       = bch2_sb_crypt_validate,
1142         .to_text        = bch2_sb_crypt_to_text,
1143 };
1144
1145 /* BCH_SB_FIELD_clean: */
1146
1147 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1148 {
1149         struct jset_entry *entry;
1150         int ret;
1151
1152         for (entry = clean->start;
1153              entry < (struct jset_entry *) vstruct_end(&clean->field);
1154              entry = vstruct_next(entry)) {
1155                 ret = bch2_journal_entry_validate(c, NULL, entry,
1156                                                   le16_to_cpu(c->disk_sb.sb->version),
1157                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1158                                                   write);
1159                 if (ret)
1160                         return ret;
1161         }
1162
1163         return 0;
1164 }
1165
1166 int bch2_fs_mark_dirty(struct bch_fs *c)
1167 {
1168         int ret;
1169
1170         /*
1171          * Unconditionally write superblock, to verify it hasn't changed before
1172          * we go rw:
1173          */
1174
1175         mutex_lock(&c->sb_lock);
1176         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1177         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1178         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1179         ret = bch2_write_super(c);
1180         mutex_unlock(&c->sb_lock);
1181
1182         return ret;
1183 }
1184
1185 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1186 {
1187         struct jset_entry *entry = *end;
1188         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1189
1190         memset(entry, 0, u64s * sizeof(u64));
1191         /*
1192          * The u64s field counts from the start of data, ignoring the shared
1193          * fields.
1194          */
1195         entry->u64s = cpu_to_le16(u64s - 1);
1196
1197         *end = vstruct_next(*end);
1198         return entry;
1199 }
1200
1201 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1202                                            struct jset_entry **end,
1203                                            u64 journal_seq)
1204 {
1205         struct bch_dev *ca;
1206         unsigned i, dev;
1207
1208         percpu_down_read(&c->mark_lock);
1209
1210         if (!journal_seq) {
1211                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1212                         bch2_fs_usage_acc_to_base(c, i);
1213         } else {
1214                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1215         }
1216
1217         {
1218                 struct jset_entry_usage *u =
1219                         container_of(jset_entry_init(end, sizeof(*u)),
1220                                      struct jset_entry_usage, entry);
1221
1222                 u->entry.type   = BCH_JSET_ENTRY_usage;
1223                 u->entry.btree_id = BCH_FS_USAGE_inodes;
1224                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1225         }
1226
1227         {
1228                 struct jset_entry_usage *u =
1229                         container_of(jset_entry_init(end, sizeof(*u)),
1230                                      struct jset_entry_usage, entry);
1231
1232                 u->entry.type   = BCH_JSET_ENTRY_usage;
1233                 u->entry.btree_id = BCH_FS_USAGE_key_version;
1234                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1235         }
1236
1237         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1238                 struct jset_entry_usage *u =
1239                         container_of(jset_entry_init(end, sizeof(*u)),
1240                                      struct jset_entry_usage, entry);
1241
1242                 u->entry.type   = BCH_JSET_ENTRY_usage;
1243                 u->entry.btree_id = BCH_FS_USAGE_reserved;
1244                 u->entry.level  = i;
1245                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1246         }
1247
1248         for (i = 0; i < c->replicas.nr; i++) {
1249                 struct bch_replicas_entry *e =
1250                         cpu_replicas_entry(&c->replicas, i);
1251                 struct jset_entry_data_usage *u =
1252                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1253                                      struct jset_entry_data_usage, entry);
1254
1255                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1256                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1257                 memcpy(&u->r, e, replicas_entry_bytes(e));
1258         }
1259
1260         for_each_member_device(ca, c, dev) {
1261                 unsigned b = sizeof(struct jset_entry_dev_usage) +
1262                         sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR;
1263                 struct jset_entry_dev_usage *u =
1264                         container_of(jset_entry_init(end, b),
1265                                      struct jset_entry_dev_usage, entry);
1266
1267                 u->entry.type = BCH_JSET_ENTRY_dev_usage;
1268                 u->dev = cpu_to_le32(dev);
1269                 u->buckets_ec           = cpu_to_le64(ca->usage_base->buckets_ec);
1270
1271                 for (i = 0; i < BCH_DATA_NR; i++) {
1272                         u->d[i].buckets = cpu_to_le64(ca->usage_base->d[i].buckets);
1273                         u->d[i].sectors = cpu_to_le64(ca->usage_base->d[i].sectors);
1274                         u->d[i].fragmented = cpu_to_le64(ca->usage_base->d[i].fragmented);
1275                 }
1276         }
1277
1278         percpu_up_read(&c->mark_lock);
1279
1280         for (i = 0; i < 2; i++) {
1281                 struct jset_entry_clock *clock =
1282                         container_of(jset_entry_init(end, sizeof(*clock)),
1283                                      struct jset_entry_clock, entry);
1284
1285                 clock->entry.type = BCH_JSET_ENTRY_clock;
1286                 clock->rw       = i;
1287                 clock->time     = cpu_to_le64(atomic64_read(&c->io_clock[i].now));
1288         }
1289 }
1290
1291 void bch2_fs_mark_clean(struct bch_fs *c)
1292 {
1293         struct bch_sb_field_clean *sb_clean;
1294         struct jset_entry *entry;
1295         unsigned u64s;
1296         int ret;
1297
1298         mutex_lock(&c->sb_lock);
1299         if (BCH_SB_CLEAN(c->disk_sb.sb))
1300                 goto out;
1301
1302         SET_BCH_SB_CLEAN(c->disk_sb.sb, true);
1303
1304         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_info);
1305         c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_alloc_metadata);
1306         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_extents_above_btree_updates));
1307         c->disk_sb.sb->features[0] &= cpu_to_le64(~(1ULL << BCH_FEATURE_btree_updates_journalled));
1308
1309         u64s = sizeof(*sb_clean) / sizeof(u64) + c->journal.entry_u64s_reserved;
1310
1311         sb_clean = bch2_sb_resize_clean(&c->disk_sb, u64s);
1312         if (!sb_clean) {
1313                 bch_err(c, "error resizing superblock while setting filesystem clean");
1314                 goto out;
1315         }
1316
1317         sb_clean->flags         = 0;
1318         sb_clean->journal_seq   = cpu_to_le64(atomic64_read(&c->journal.seq));
1319
1320         /* Trying to catch outstanding bug: */
1321         BUG_ON(le64_to_cpu(sb_clean->journal_seq) > S64_MAX);
1322
1323         entry = sb_clean->start;
1324         bch2_journal_super_entries_add_common(c, &entry, 0);
1325         entry = bch2_btree_roots_to_journal_entries(c, entry, entry);
1326         BUG_ON((void *) entry > vstruct_end(&sb_clean->field));
1327
1328         memset(entry, 0,
1329                vstruct_end(&sb_clean->field) - (void *) entry);
1330
1331         /*
1332          * this should be in the write path, and we should be validating every
1333          * superblock section:
1334          */
1335         ret = bch2_sb_clean_validate_late(c, sb_clean, WRITE);
1336         if (ret) {
1337                 bch_err(c, "error writing marking filesystem clean: validate error");
1338                 goto out;
1339         }
1340
1341         bch2_write_super(c);
1342 out:
1343         mutex_unlock(&c->sb_lock);
1344 }
1345
1346 static int bch2_sb_clean_validate(struct bch_sb *sb,
1347                                   struct bch_sb_field *f,
1348                                   struct printbuf *err)
1349 {
1350         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1351
1352         if (vstruct_bytes(&clean->field) < sizeof(*clean)) {
1353                 prt_printf(err, "wrong size (got %zu should be %zu)",
1354                        vstruct_bytes(&clean->field), sizeof(*clean));
1355                 return -BCH_ERR_invalid_sb_clean;
1356         }
1357
1358         return 0;
1359 }
1360
1361 static void bch2_sb_clean_to_text(struct printbuf *out, struct bch_sb *sb,
1362                                   struct bch_sb_field *f)
1363 {
1364         struct bch_sb_field_clean *clean = field_to_type(f, clean);
1365         struct jset_entry *entry;
1366
1367         prt_printf(out, "flags:          %x",   le32_to_cpu(clean->flags));
1368         prt_newline(out);
1369         prt_printf(out, "journal_seq:    %llu", le64_to_cpu(clean->journal_seq));
1370         prt_newline(out);
1371
1372         for (entry = clean->start;
1373              entry != vstruct_end(&clean->field);
1374              entry = vstruct_next(entry)) {
1375                 if (entry->type == BCH_JSET_ENTRY_btree_keys &&
1376                     !entry->u64s)
1377                         continue;
1378
1379                 bch2_journal_entry_to_text(out, NULL, entry);
1380                 prt_newline(out);
1381         }
1382 }
1383
1384 static const struct bch_sb_field_ops bch_sb_field_ops_clean = {
1385         .validate       = bch2_sb_clean_validate,
1386         .to_text        = bch2_sb_clean_to_text,
1387 };
1388
1389 static const struct bch_sb_field_ops *bch2_sb_field_ops[] = {
1390 #define x(f, nr)                                        \
1391         [BCH_SB_FIELD_##f] = &bch_sb_field_ops_##f,
1392         BCH_SB_FIELDS()
1393 #undef x
1394 };
1395
1396 static int bch2_sb_field_validate(struct bch_sb *sb, struct bch_sb_field *f,
1397                                   struct printbuf *err)
1398 {
1399         unsigned type = le32_to_cpu(f->type);
1400         struct printbuf field_err = PRINTBUF;
1401         int ret;
1402
1403         if (type >= BCH_SB_FIELD_NR)
1404                 return 0;
1405
1406         ret = bch2_sb_field_ops[type]->validate(sb, f, &field_err);
1407         if (ret) {
1408                 prt_printf(err, "Invalid superblock section %s: %s",
1409                        bch2_sb_fields[type],
1410                        field_err.buf);
1411                 prt_newline(err);
1412                 bch2_sb_field_to_text(err, sb, f);
1413         }
1414
1415         printbuf_exit(&field_err);
1416         return ret;
1417 }
1418
1419 void bch2_sb_field_to_text(struct printbuf *out, struct bch_sb *sb,
1420                            struct bch_sb_field *f)
1421 {
1422         unsigned type = le32_to_cpu(f->type);
1423         const struct bch_sb_field_ops *ops = type < BCH_SB_FIELD_NR
1424                 ? bch2_sb_field_ops[type] : NULL;
1425
1426         if (!out->nr_tabstops)
1427                 printbuf_tabstop_push(out, 32);
1428
1429         if (ops)
1430                 prt_printf(out, "%s", bch2_sb_fields[type]);
1431         else
1432                 prt_printf(out, "(unknown field %u)", type);
1433
1434         prt_printf(out, " (size %zu):", vstruct_bytes(f));
1435         prt_newline(out);
1436
1437         if (ops && ops->to_text) {
1438                 printbuf_indent_add(out, 2);
1439                 bch2_sb_field_ops[type]->to_text(out, sb, f);
1440                 printbuf_indent_sub(out, 2);
1441         }
1442 }
1443
1444 void bch2_sb_layout_to_text(struct printbuf *out, struct bch_sb_layout *l)
1445 {
1446         unsigned i;
1447
1448         prt_printf(out, "Type:                    %u", l->layout_type);
1449         prt_newline(out);
1450
1451         prt_str(out, "Superblock max size:     ");
1452         prt_units_u64(out, 512 << l->sb_max_size_bits);
1453         prt_newline(out);
1454
1455         prt_printf(out, "Nr superblocks:          %u", l->nr_superblocks);
1456         prt_newline(out);
1457
1458         prt_str(out, "Offsets:                 ");
1459         for (i = 0; i < l->nr_superblocks; i++) {
1460                 if (i)
1461                         prt_str(out, ", ");
1462                 prt_printf(out, "%llu", le64_to_cpu(l->sb_offset[i]));
1463         }
1464         prt_newline(out);
1465 }
1466
1467 void bch2_sb_to_text(struct printbuf *out, struct bch_sb *sb,
1468                      bool print_layout, unsigned fields)
1469 {
1470         struct bch_sb_field_members *mi;
1471         struct bch_sb_field *f;
1472         u64 fields_have = 0;
1473         unsigned nr_devices = 0;
1474
1475         if (!out->nr_tabstops)
1476                 printbuf_tabstop_push(out, 44);
1477
1478         mi = bch2_sb_get_members(sb);
1479         if (mi) {
1480                 struct bch_member *m;
1481
1482                 for (m = mi->members;
1483                      m < mi->members + sb->nr_devices;
1484                      m++)
1485                         nr_devices += bch2_member_exists(m);
1486         }
1487
1488         prt_printf(out, "External UUID:");
1489         prt_tab(out);
1490         pr_uuid(out, sb->user_uuid.b);
1491         prt_newline(out);
1492
1493         prt_printf(out, "Internal UUID:");
1494         prt_tab(out);
1495         pr_uuid(out, sb->uuid.b);
1496         prt_newline(out);
1497
1498         prt_str(out, "Device index:");
1499         prt_tab(out);
1500         prt_printf(out, "%u", sb->dev_idx);
1501         prt_newline(out);
1502
1503         prt_str(out, "Label:");
1504         prt_tab(out);
1505         prt_printf(out, "%.*s", (int) sizeof(sb->label), sb->label);
1506         prt_newline(out);
1507
1508         prt_str(out, "Version:");
1509         prt_tab(out);
1510         prt_printf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version)]);
1511         prt_newline(out);
1512
1513         prt_printf(out, "Oldest version on disk:");
1514         prt_tab(out);
1515         prt_printf(out, "%s", bch2_metadata_versions[le16_to_cpu(sb->version_min)]);
1516         prt_newline(out);
1517
1518         prt_printf(out, "Created:");
1519         prt_tab(out);
1520         if (sb->time_base_lo)
1521                 pr_time(out, div_u64(le64_to_cpu(sb->time_base_lo), NSEC_PER_SEC));
1522         else
1523                 prt_printf(out, "(not set)");
1524         prt_newline(out);
1525
1526         prt_printf(out, "Sequence number:");
1527         prt_tab(out);
1528         prt_printf(out, "%llu", le64_to_cpu(sb->seq));
1529         prt_newline(out);
1530
1531         prt_printf(out, "Superblock size:");
1532         prt_tab(out);
1533         prt_printf(out, "%zu", vstruct_bytes(sb));
1534         prt_newline(out);
1535
1536         prt_printf(out, "Clean:");
1537         prt_tab(out);
1538         prt_printf(out, "%llu", BCH_SB_CLEAN(sb));
1539         prt_newline(out);
1540
1541         prt_printf(out, "Devices:");
1542         prt_tab(out);
1543         prt_printf(out, "%u", nr_devices);
1544         prt_newline(out);
1545
1546         prt_printf(out, "Sections:");
1547         vstruct_for_each(sb, f)
1548                 fields_have |= 1 << le32_to_cpu(f->type);
1549         prt_tab(out);
1550         prt_bitflags(out, bch2_sb_fields, fields_have);
1551         prt_newline(out);
1552
1553         prt_printf(out, "Features:");
1554         prt_tab(out);
1555         prt_bitflags(out, bch2_sb_features, le64_to_cpu(sb->features[0]));
1556         prt_newline(out);
1557
1558         prt_printf(out, "Compat features:");
1559         prt_tab(out);
1560         prt_bitflags(out, bch2_sb_compat, le64_to_cpu(sb->compat[0]));
1561         prt_newline(out);
1562
1563         prt_newline(out);
1564         prt_printf(out, "Options:");
1565         prt_newline(out);
1566         printbuf_indent_add(out, 2);
1567         {
1568                 enum bch_opt_id id;
1569
1570                 for (id = 0; id < bch2_opts_nr; id++) {
1571                         const struct bch_option *opt = bch2_opt_table + id;
1572
1573                         if (opt->get_sb != BCH2_NO_SB_OPT) {
1574                                 u64 v = bch2_opt_from_sb(sb, id);
1575
1576                                 prt_printf(out, "%s:", opt->attr.name);
1577                                 prt_tab(out);
1578                                 bch2_opt_to_text(out, NULL, sb, opt, v,
1579                                                  OPT_HUMAN_READABLE|OPT_SHOW_FULL_LIST);
1580                                 prt_newline(out);
1581                         }
1582                 }
1583         }
1584
1585         printbuf_indent_sub(out, 2);
1586
1587         if (print_layout) {
1588                 prt_newline(out);
1589                 prt_printf(out, "layout:");
1590                 prt_newline(out);
1591                 printbuf_indent_add(out, 2);
1592                 bch2_sb_layout_to_text(out, &sb->layout);
1593                 printbuf_indent_sub(out, 2);
1594         }
1595
1596         vstruct_for_each(sb, f)
1597                 if (fields & (1 << le32_to_cpu(f->type))) {
1598                         prt_newline(out);
1599                         bch2_sb_field_to_text(out, sb, f);
1600                 }
1601 }