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             uuid_le_cmp(layout->magic, BCHFS_MAGIC)) {
 216                 prt_printf(out, "Not a bcachefs superblock layout");
 217                 return -BCH_ERR_invalid_sb_layout;
 218         }
 219
 220         if (layout->layout_type != 0) {
 221                 prt_printf(out, "Invalid superblock layout type %u",
 222                        layout->layout_type);
 223                 return -BCH_ERR_invalid_sb_layout_type;
 224         }
 225
 226         if (!layout->nr_superblocks) {
 227                 prt_printf(out, "Invalid superblock layout: no superblocks");
 228                 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 229         }
 230
 231         if (layout->nr_superblocks > ARRAY_SIZE(layout->sb_offset)) {
 232                 prt_printf(out, "Invalid superblock layout: too many superblocks");
 233                 return -BCH_ERR_invalid_sb_layout_nr_superblocks;
 234         }
 235
 236         max_sectors = 1 << layout->sb_max_size_bits;
 237
 238         prev_offset = le64_to_cpu(layout->sb_offset[0]);
 239
 240         for (i = 1; i < layout->nr_superblocks; i++) {
 241                 offset = le64_to_cpu(layout->sb_offset[i]);
 242
 243                 if (offset < prev_offset + max_sectors) {
 244                         prt_printf(out, "Invalid superblock layout: superblocks overlap\n"
 245                                "  (sb %u ends at %llu next starts at %llu",
 246                                i - 1, prev_offset + max_sectors, offset);
 247                         return -BCH_ERR_invalid_sb_layout_superblocks_overlap;
 248                 }
 249                 prev_offset = offset;
 250         }
 251
 252         return 0;
 253 }
 254
 255 static int bch2_sb_validate(struct bch_sb_handle *disk_sb, struct printbuf *out,
 256                             int rw)
 257 {
 258         struct bch_sb *sb = disk_sb->sb;
 259         struct bch_sb_field *f;
 260         struct bch_sb_field_members *mi;
 261         enum bch_opt_id opt_id;
 262         u32 version, version_min;
 263         u16 block_size;
 264         int ret;
 265
 266         version         = le16_to_cpu(sb->version);
 267         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 268                 ? le16_to_cpu(sb->version_min)
 269                 : version;
 270
 271         if (version    >= bcachefs_metadata_version_max) {
 272                 prt_printf(out, "Unsupported superblock version %u (min %u, max %u)",
 273                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 274                 return -BCH_ERR_invalid_sb_version;
 275         }
 276
 277         if (version_min < bcachefs_metadata_version_min) {
 278                 prt_printf(out, "Unsupported superblock version %u (min %u, max %u)",
 279                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 280                 return -BCH_ERR_invalid_sb_version;
 281         }
 282
 283         if (version_min > version) {
 284                 prt_printf(out, "Bad minimum version %u, greater than version field %u",
 285                        version_min, version);
 286                 return -BCH_ERR_invalid_sb_version;
 287         }
 288
 289         if (sb->features[1] ||
 290             (le64_to_cpu(sb->features[0]) & (~0ULL << BCH_FEATURE_NR))) {
 291                 prt_printf(out, "Filesystem has incompatible features");
 292                 return -BCH_ERR_invalid_sb_features;
 293         }
 294
 295         block_size = le16_to_cpu(sb->block_size);
 296
 297         if (block_size > PAGE_SECTORS) {
 298                 prt_printf(out, "Block size too big (got %u, max %u)",
 299                        block_size, PAGE_SECTORS);
 300                 return -BCH_ERR_invalid_sb_block_size;
 301         }
 302
 303         if (bch2_is_zero(sb->user_uuid.b, sizeof(uuid_le))) {
 304                 prt_printf(out, "Bad user UUID (got zeroes)");
 305                 return -BCH_ERR_invalid_sb_uuid;
 306         }
 307
 308         if (bch2_is_zero(sb->uuid.b, sizeof(uuid_le))) {
 309                 prt_printf(out, "Bad intenal UUID (got zeroes)");
 310                 return -BCH_ERR_invalid_sb_uuid;
 311         }
 312
 313         if (!sb->nr_devices ||
 314             sb->nr_devices > BCH_SB_MEMBERS_MAX) {
 315                 prt_printf(out, "Bad number of member devices %u (max %u)",
 316                        sb->nr_devices, BCH_SB_MEMBERS_MAX);
 317                 return -BCH_ERR_invalid_sb_too_many_members;
 318         }
 319
 320         if (sb->dev_idx >= sb->nr_devices) {
 321                 prt_printf(out, "Bad dev_idx (got %u, nr_devices %u)",
 322                        sb->dev_idx, sb->nr_devices);
 323                 return -BCH_ERR_invalid_sb_dev_idx;
 324         }
 325
 326         if (!sb->time_precision ||
 327             le32_to_cpu(sb->time_precision) > NSEC_PER_SEC) {
 328                 prt_printf(out, "Invalid time precision: %u (min 1, max %lu)",
 329                        le32_to_cpu(sb->time_precision), NSEC_PER_SEC);
 330                 return -BCH_ERR_invalid_sb_time_precision;
 331         }
 332
 333         if (rw == READ) {
 334                 /*
 335                  * Been seeing a bug where these are getting inexplicably
 336                  * zeroed, so we'r now validating them, but we have to be
 337                  * careful not to preven people's filesystems from mounting:
 338                  */
 339                 if (!BCH_SB_JOURNAL_FLUSH_DELAY(sb))
 340                         SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
 341                 if (!BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
 342                         SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 1000);
 343         }
 344
 345         for (opt_id = 0; opt_id < bch2_opts_nr; opt_id++) {
 346                 const struct bch_option *opt = bch2_opt_table + opt_id;
 347
 348                 if (opt->get_sb != BCH2_NO_SB_OPT) {
 349                         u64 v = bch2_opt_from_sb(sb, opt_id);
 350
 351                         prt_printf(out, "Invalid option ");
 352                         ret = bch2_opt_validate(opt, v, out);
 353                         if (ret)
 354                                 return ret;
 355
 356                         printbuf_reset(out);
 357                 }
 358         }
 359
 360         /* validate layout */
 361         ret = validate_sb_layout(&sb->layout, out);
 362         if (ret)
 363                 return ret;
 364
 365         vstruct_for_each(sb, f) {
 366                 if (!f->u64s) {
 367                         prt_printf(out, "Invalid superblock: optional field with size 0 (type %u)",
 368                                le32_to_cpu(f->type));
 369                         return -BCH_ERR_invalid_sb_field_size;
 370                 }
 371
 372                 if (vstruct_next(f) > vstruct_last(sb)) {
 373                         prt_printf(out, "Invalid superblock: optional field extends past end of superblock (type %u)",
 374                                le32_to_cpu(f->type));
 375                         return -BCH_ERR_invalid_sb_field_size;
 376                 }
 377         }
 378
 379         /* members must be validated first: */
 380         mi = bch2_sb_get_members(sb);
 381         if (!mi) {
 382                 prt_printf(out, "Invalid superblock: member info area missing");
 383                 return -BCH_ERR_invalid_sb_members_missing;
 384         }
 385
 386         ret = bch2_sb_field_validate(sb, &mi->field, out);
 387         if (ret)
 388                 return ret;
 389
 390         vstruct_for_each(sb, f) {
 391                 if (le32_to_cpu(f->type) == BCH_SB_FIELD_members)
 392                         continue;
 393
 394                 ret = bch2_sb_field_validate(sb, f, out);
 395                 if (ret)
 396                         return ret;
 397         }
 398
 399         return 0;
 400 }
 401
 402 /* device open: */
 403
 404 static void bch2_sb_update(struct bch_fs *c)
 405 {
 406         struct bch_sb *src = c->disk_sb.sb;
 407         struct bch_sb_field_members *mi = bch2_sb_get_members(src);
 408         struct bch_dev *ca;
 409         unsigned i;
 410
 411         lockdep_assert_held(&c->sb_lock);
 412
 413         c->sb.uuid              = src->uuid;
 414         c->sb.user_uuid         = src->user_uuid;
 415         c->sb.version           = le16_to_cpu(src->version);
 416         c->sb.version_min       = le16_to_cpu(src->version_min);
 417         c->sb.nr_devices        = src->nr_devices;
 418         c->sb.clean             = BCH_SB_CLEAN(src);
 419         c->sb.encryption_type   = BCH_SB_ENCRYPTION_TYPE(src);
 420
 421         c->sb.nsec_per_time_unit = le32_to_cpu(src->time_precision);
 422         c->sb.time_units_per_sec = NSEC_PER_SEC / c->sb.nsec_per_time_unit;
 423
 424         /* XXX this is wrong, we need a 96 or 128 bit integer type */
 425         c->sb.time_base_lo      = div_u64(le64_to_cpu(src->time_base_lo),
 426                                           c->sb.nsec_per_time_unit);
 427         c->sb.time_base_hi      = le32_to_cpu(src->time_base_hi);
 428
 429         c->sb.features          = le64_to_cpu(src->features[0]);
 430         c->sb.compat            = le64_to_cpu(src->compat[0]);
 431
 432         for_each_member_device(ca, c, i)
 433                 ca->mi = bch2_mi_to_cpu(mi->members + i);
 434 }
 435
 436 static void __copy_super(struct bch_sb_handle *dst_handle, struct bch_sb *src)
 437 {
 438         struct bch_sb_field *src_f, *dst_f;
 439         struct bch_sb *dst = dst_handle->sb;
 440         unsigned i;
 441
 442         dst->version            = src->version;
 443         dst->version_min        = src->version_min;
 444         dst->seq                = src->seq;
 445         dst->uuid               = src->uuid;
 446         dst->user_uuid          = src->user_uuid;
 447         memcpy(dst->label,      src->label, sizeof(dst->label));
 448
 449         dst->block_size         = src->block_size;
 450         dst->nr_devices         = src->nr_devices;
 451
 452         dst->time_base_lo       = src->time_base_lo;
 453         dst->time_base_hi       = src->time_base_hi;
 454         dst->time_precision     = src->time_precision;
 455
 456         memcpy(dst->flags,      src->flags,     sizeof(dst->flags));
 457         memcpy(dst->features,   src->features,  sizeof(dst->features));
 458         memcpy(dst->compat,     src->compat,    sizeof(dst->compat));
 459
 460         for (i = 0; i < BCH_SB_FIELD_NR; i++) {
 461                 if ((1U << i) & BCH_SINGLE_DEVICE_SB_FIELDS)
 462                         continue;
 463
 464                 src_f = bch2_sb_field_get(src, i);
 465                 dst_f = bch2_sb_field_get(dst, i);
 466                 dst_f = __bch2_sb_field_resize(dst_handle, dst_f,
 467                                 src_f ? le32_to_cpu(src_f->u64s) : 0);
 468
 469                 if (src_f)
 470                         memcpy(dst_f, src_f, vstruct_bytes(src_f));
 471         }
 472 }
 473
 474 int bch2_sb_to_fs(struct bch_fs *c, struct bch_sb *src)
 475 {
 476         struct bch_sb_field_journal *journal_buckets =
 477                 bch2_sb_get_journal(src);
 478         unsigned journal_u64s = journal_buckets
 479                 ? le32_to_cpu(journal_buckets->field.u64s)
 480                 : 0;
 481         int ret;
 482
 483         lockdep_assert_held(&c->sb_lock);
 484
 485         ret = bch2_sb_realloc(&c->disk_sb,
 486                               le32_to_cpu(src->u64s) - journal_u64s);
 487         if (ret)
 488                 return ret;
 489
 490         __copy_super(&c->disk_sb, src);
 491
 492         ret = bch2_sb_replicas_to_cpu_replicas(c);
 493         if (ret)
 494                 return ret;
 495
 496         ret = bch2_sb_disk_groups_to_cpu(c);
 497         if (ret)
 498                 return ret;
 499
 500         bch2_sb_update(c);
 501         return 0;
 502 }
 503
 504 int bch2_sb_from_fs(struct bch_fs *c, struct bch_dev *ca)
 505 {
 506         struct bch_sb *src = c->disk_sb.sb, *dst = ca->disk_sb.sb;
 507         struct bch_sb_field_journal *journal_buckets =
 508                 bch2_sb_get_journal(dst);
 509         unsigned journal_u64s = journal_buckets
 510                 ? le32_to_cpu(journal_buckets->field.u64s)
 511                 : 0;
 512         unsigned u64s = le32_to_cpu(src->u64s) + journal_u64s;
 513         int ret;
 514
 515         ret = bch2_sb_realloc(&ca->disk_sb, u64s);
 516         if (ret)
 517                 return ret;
 518
 519         __copy_super(&ca->disk_sb, src);
 520         return 0;
 521 }
 522
 523 /* read superblock: */
 524
 525 static int read_one_super(struct bch_sb_handle *sb, u64 offset, struct printbuf *err)
 526 {
 527         struct bch_csum csum;
 528         u32 version, version_min;
 529         size_t bytes;
 530         int ret;
 531 reread:
 532         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 533         sb->bio->bi_iter.bi_sector = offset;
 534         bch2_bio_map(sb->bio, sb->sb, sb->buffer_size);
 535
 536         ret = submit_bio_wait(sb->bio);
 537         if (ret) {
 538                 prt_printf(err, "IO error: %i", ret);
 539                 return ret;
 540         }
 541
 542         if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC) &&
 543             uuid_le_cmp(sb->sb->magic, BCHFS_MAGIC)) {
 544                 prt_printf(err, "Not a bcachefs superblock");
 545                 return -BCH_ERR_invalid_sb_magic;
 546         }
 547
 548         version         = le16_to_cpu(sb->sb->version);
 549         version_min     = version >= bcachefs_metadata_version_bkey_renumber
 550                 ? le16_to_cpu(sb->sb->version_min)
 551                 : version;
 552
 553         if (version    >= bcachefs_metadata_version_max) {
 554                 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
 555                        version, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 556                 return -BCH_ERR_invalid_sb_version;
 557         }
 558
 559         if (version_min < bcachefs_metadata_version_min) {
 560                 prt_printf(err, "Unsupported superblock version %u (min %u, max %u)",
 561                        version_min, bcachefs_metadata_version_min, bcachefs_metadata_version_max);
 562                 return -BCH_ERR_invalid_sb_version;
 563         }
 564
 565         bytes = vstruct_bytes(sb->sb);
 566
 567         if (bytes > 512 << sb->sb->layout.sb_max_size_bits) {
 568                 prt_printf(err, "Invalid superblock: too big (got %zu bytes, layout max %lu)",
 569                        bytes, 512UL << sb->sb->layout.sb_max_size_bits);
 570                 return -BCH_ERR_invalid_sb_too_big;
 571         }
 572
 573         if (bytes > sb->buffer_size) {
 574                 if (bch2_sb_realloc(sb, le32_to_cpu(sb->sb->u64s)))
 575                         return -ENOMEM;
 576                 goto reread;
 577         }
 578
 579         if (BCH_SB_CSUM_TYPE(sb->sb) >= BCH_CSUM_NR) {
 580                 prt_printf(err, "unknown checksum type %llu", BCH_SB_CSUM_TYPE(sb->sb));
 581                 return -BCH_ERR_invalid_sb_csum_type;
 582         }
 583
 584         /* XXX: verify MACs */
 585         csum = csum_vstruct(NULL, BCH_SB_CSUM_TYPE(sb->sb),
 586                             null_nonce(), sb->sb);
 587
 588         if (bch2_crc_cmp(csum, sb->sb->csum)) {
 589                 prt_printf(err, "bad checksum");
 590                 return -BCH_ERR_invalid_sb_csum;
 591         }
 592
 593         sb->seq = le64_to_cpu(sb->sb->seq);
 594
 595         return 0;
 596 }
 597
 598 int bch2_read_super(const char *path, struct bch_opts *opts,
 599                     struct bch_sb_handle *sb)
 600 {
 601         u64 offset = opt_get(*opts, sb);
 602         struct bch_sb_layout layout;
 603         struct printbuf err = PRINTBUF;
 604         __le64 *i;
 605         int ret;
 606
 607         pr_verbose_init(*opts, "");
 608
 609         memset(sb, 0, sizeof(*sb));
 610         sb->mode        = FMODE_READ;
 611         sb->have_bio    = true;
 612
 613         if (!opt_get(*opts, noexcl))
 614                 sb->mode |= FMODE_EXCL;
 615
 616         if (!opt_get(*opts, nochanges))
 617                 sb->mode |= FMODE_WRITE;
 618
 619         sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 620         if (IS_ERR(sb->bdev) &&
 621             PTR_ERR(sb->bdev) == -EACCES &&
 622             opt_get(*opts, read_only)) {
 623                 sb->mode &= ~FMODE_WRITE;
 624
 625                 sb->bdev = blkdev_get_by_path(path, sb->mode, sb);
 626                 if (!IS_ERR(sb->bdev))
 627                         opt_set(*opts, nochanges, true);
 628         }
 629
 630         if (IS_ERR(sb->bdev)) {
 631                 ret = PTR_ERR(sb->bdev);
 632                 goto out;
 633         }
 634
 635         ret = bch2_sb_realloc(sb, 0);
 636         if (ret) {
 637                 prt_printf(&err, "error allocating memory for superblock");
 638                 goto err;
 639         }
 640
 641         if (bch2_fs_init_fault("read_super")) {
 642                 prt_printf(&err, "dynamic fault");
 643                 ret = -EFAULT;
 644                 goto err;
 645         }
 646
 647         ret = read_one_super(sb, offset, &err);
 648         if (!ret)
 649                 goto got_super;
 650
 651         if (opt_defined(*opts, sb))
 652                 goto err;
 653
 654         printk(KERN_ERR "bcachefs (%s): error reading default superblock: %s",
 655                path, err.buf);
 656         printbuf_reset(&err);
 657
 658         /*
 659          * Error reading primary superblock - read location of backup
 660          * superblocks:
 661          */
 662         bio_reset(sb->bio, sb->bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 663         sb->bio->bi_iter.bi_sector = BCH_SB_LAYOUT_SECTOR;
 664         /*
 665          * use sb buffer to read layout, since sb buffer is page aligned but
 666          * layout won't be:
 667          */
 668         bch2_bio_map(sb->bio, sb->sb, sizeof(struct bch_sb_layout));
 669
 670         ret = submit_bio_wait(sb->bio);
 671         if (ret) {
 672                 prt_printf(&err, "IO error: %i", ret);
 673                 goto err;
 674         }
 675
 676         memcpy(&layout, sb->sb, sizeof(layout));
 677         ret = validate_sb_layout(&layout, &err);
 678         if (ret)
 679                 goto err;
 680
 681         for (i = layout.sb_offset;
 682              i < layout.sb_offset + layout.nr_superblocks; i++) {
 683                 offset = le64_to_cpu(*i);
 684
 685                 if (offset == opt_get(*opts, sb))
 686                         continue;
 687
 688                 ret = read_one_super(sb, offset, &err);
 689                 if (!ret)
 690                         goto got_super;
 691         }
 692
 693         goto err;
 694
 695 got_super:
 696         if (le16_to_cpu(sb->sb->block_size) << 9 <
 697             bdev_logical_block_size(sb->bdev)) {
 698                 prt_printf(&err, "block size (%u) smaller than device block size (%u)",
 699                        le16_to_cpu(sb->sb->block_size) << 9,
 700                        bdev_logical_block_size(sb->bdev));
 701                 ret = -BCH_ERR_block_size_too_small;
 702                 goto err;
 703         }
 704
 705         ret = 0;
 706         sb->have_layout = true;
 707
 708         ret = bch2_sb_validate(sb, &err, READ);
 709         if (ret) {
 710                 printk(KERN_ERR "bcachefs (%s): error validating superblock: %s",
 711                        path, err.buf);
 712                 goto err_no_print;
 713         }
 714 out:
 715         pr_verbose_init(*opts, "ret %i", ret);
 716         printbuf_exit(&err);
 717         return ret;
 718 err:
 719         printk(KERN_ERR "bcachefs (%s): error reading superblock: %s",
 720                path, err.buf);
 721 err_no_print:
 722         bch2_free_super(sb);
 723         goto out;
 724 }
 725
 726 /* write superblock: */
 727
 728 static void write_super_endio(struct bio *bio)
 729 {
 730         struct bch_dev *ca = bio->bi_private;
 731
 732         /* XXX: return errors directly */
 733
 734         if (bch2_dev_io_err_on(bio->bi_status, ca, "superblock write error: %s",
 735                                bch2_blk_status_to_str(bio->bi_status)))
 736                 ca->sb_write_error = 1;
 737
 738         closure_put(&ca->fs->sb_write);
 739         percpu_ref_put(&ca->io_ref);
 740 }
 741
 742 static void read_back_super(struct bch_fs *c, struct bch_dev *ca)
 743 {
 744         struct bch_sb *sb = ca->disk_sb.sb;
 745         struct bio *bio = ca->disk_sb.bio;
 746
 747         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_READ|REQ_SYNC|REQ_META);
 748         bio->bi_iter.bi_sector  = le64_to_cpu(sb->layout.sb_offset[0]);
 749         bio->bi_end_io          = write_super_endio;
 750         bio->bi_private         = ca;
 751         bch2_bio_map(bio, ca->sb_read_scratch, PAGE_SIZE);
 752
 753         this_cpu_add(ca->io_done->sectors[READ][BCH_DATA_sb],
 754                      bio_sectors(bio));
 755
 756         percpu_ref_get(&ca->io_ref);
 757         closure_bio_submit(bio, &c->sb_write);
 758 }
 759
 760 static void write_one_super(struct bch_fs *c, struct bch_dev *ca, unsigned idx)
 761 {
 762         struct bch_sb *sb = ca->disk_sb.sb;
 763         struct bio *bio = ca->disk_sb.bio;
 764
 765         sb->offset = sb->layout.sb_offset[idx];
 766
 767         SET_BCH_SB_CSUM_TYPE(sb, bch2_csum_opt_to_type(c->opts.metadata_checksum, false));
 768         sb->csum = csum_vstruct(c, BCH_SB_CSUM_TYPE(sb),
 769                                 null_nonce(), sb);
 770
 771         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
 772         bio->bi_iter.bi_sector  = le64_to_cpu(sb->offset);
 773         bio->bi_end_io          = write_super_endio;
 774         bio->bi_private         = ca;
 775         bch2_bio_map(bio, sb,
 776                      roundup((size_t) vstruct_bytes(sb),
 777                              bdev_logical_block_size(ca->disk_sb.bdev)));
 778
 779         this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_sb],
 780                      bio_sectors(bio));
 781
 782         percpu_ref_get(&ca->io_ref);
 783         closure_bio_submit(bio, &c->sb_write);
 784 }
 785
 786 int bch2_write_super(struct bch_fs *c)
 787 {
 788         struct closure *cl = &c->sb_write;
 789         struct bch_dev *ca;
 790         struct printbuf err = PRINTBUF;
 791         unsigned i, sb = 0, nr_wrote;
 792         struct bch_devs_mask sb_written;
 793         bool wrote, can_mount_without_written, can_mount_with_written;
 794         unsigned degraded_flags = BCH_FORCE_IF_DEGRADED;
 795         int ret = 0;
 796
 797         trace_and_count(c, write_super, c, _RET_IP_);
 798
 799         if (c->opts.very_degraded)
 800                 degraded_flags |= BCH_FORCE_IF_LOST;
 801
 802         lockdep_assert_held(&c->sb_lock);
 803
 804         closure_init_stack(cl);
 805         memset(&sb_written, 0, sizeof(sb_written));
 806
 807         if (c->opts.version_upgrade) {
 808                 c->disk_sb.sb->magic = BCHFS_MAGIC;
 809                 c->disk_sb.sb->layout.magic = BCHFS_MAGIC;
 810         }
 811
 812         le64_add_cpu(&c->disk_sb.sb->seq, 1);
 813
 814         if (test_bit(BCH_FS_ERROR, &c->flags))
 815                 SET_BCH_SB_HAS_ERRORS(c->disk_sb.sb, 1);
 816         if (test_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags))
 817                 SET_BCH_SB_HAS_TOPOLOGY_ERRORS(c->disk_sb.sb, 1);
 818
 819         SET_BCH_SB_BIG_ENDIAN(c->disk_sb.sb, CPU_BIG_ENDIAN);
 820
 821         bch2_sb_counters_from_cpu(c);
 822
 823         for_each_online_member(ca, c, i)
 824                 bch2_sb_from_fs(c, ca);
 825
 826         for_each_online_member(ca, c, i) {
 827                 printbuf_reset(&err);
 828
 829                 ret = bch2_sb_validate(&ca->disk_sb, &err, WRITE);
 830                 if (ret) {
 831                         bch2_fs_inconsistent(c, "sb invalid before write: %s", err.buf);
 832                         percpu_ref_put(&ca->io_ref);
 833                         goto out;
 834                 }
 835         }
 836
 837         if (c->opts.nochanges)
 838                 goto out;
 839
 840         /*
 841          * Defer writing the superblock until filesystem initialization is
 842          * complete - don't write out a partly initialized superblock:
 843          */
 844         if (!BCH_SB_INITIALIZED(c->disk_sb.sb))
 845                 goto out;
 846
 847         for_each_online_member(ca, c, i) {
 848                 __set_bit(ca->dev_idx, sb_written.d);
 849                 ca->sb_write_error = 0;
 850         }
 851
 852         for_each_online_member(ca, c, i)
 853                 read_back_super(c, ca);
 854         closure_sync(cl);
 855
 856         for_each_online_member(ca, c, i) {
 857                 if (ca->sb_write_error)
 858                         continue;
 859
 860                 if (le64_to_cpu(ca->sb_read_scratch->seq) < ca->disk_sb.seq) {
 861                         bch2_fs_fatal_error(c,
 862                                 "Superblock write was silently dropped! (seq %llu expected %llu)",
 863                                 le64_to_cpu(ca->sb_read_scratch->seq),
 864                                 ca->disk_sb.seq);
 865                         percpu_ref_put(&ca->io_ref);
 866                         ret = -EROFS;
 867                         goto out;
 868                 }
 869
 870                 if (le64_to_cpu(ca->sb_read_scratch->seq) > ca->disk_sb.seq) {
 871                         bch2_fs_fatal_error(c,
 872                                 "Superblock modified by another process (seq %llu expected %llu)",
 873                                 le64_to_cpu(ca->sb_read_scratch->seq),
 874                                 ca->disk_sb.seq);
 875                         percpu_ref_put(&ca->io_ref);
 876                         ret = -EROFS;
 877                         goto out;
 878                 }
 879         }
 880
 881         do {
 882                 wrote = false;
 883                 for_each_online_member(ca, c, i)
 884                         if (!ca->sb_write_error &&
 885                             sb < ca->disk_sb.sb->layout.nr_superblocks) {
 886                                 write_one_super(c, ca, sb);
 887                                 wrote = true;
 888                         }
 889                 closure_sync(cl);
 890                 sb++;
 891         } while (wrote);
 892
 893         for_each_online_member(ca, c, i) {
 894                 if (ca->sb_write_error)
 895                         __clear_bit(ca->dev_idx, sb_written.d);
 896                 else
 897                         ca->disk_sb.seq = le64_to_cpu(ca->disk_sb.sb->seq);
 898         }
 899
 900         nr_wrote = dev_mask_nr(&sb_written);
 901
 902         can_mount_with_written =
 903                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 904
 905         for (i = 0; i < ARRAY_SIZE(sb_written.d); i++)
 906                 sb_written.d[i] = ~sb_written.d[i];
 907
 908         can_mount_without_written =
 909                 bch2_have_enough_devs(c, sb_written, degraded_flags, false);
 910
 911         /*
 912          * If we would be able to mount _without_ the devices we successfully
 913          * wrote superblocks to, we weren't able to write to enough devices:
 914          *
 915          * Exception: if we can mount without the successes because we haven't
 916          * written anything (new filesystem), we continue if we'd be able to
 917          * mount with the devices we did successfully write to:
 918          */
 919         if (bch2_fs_fatal_err_on(!nr_wrote ||
 920                                  !can_mount_with_written ||
 921                                  (can_mount_without_written &&
 922                                   !can_mount_with_written), c,
 923                 "Unable to write superblock to sufficient devices (from %ps)",
 924                 (void *) _RET_IP_))
 925                 ret = -1;
 926 out:
 927         /* Make new options visible after they're persistent: */
 928         bch2_sb_update(c);
 929         printbuf_exit(&err);
 930         return ret;
 931 }
 932
 933 void __bch2_check_set_feature(struct bch_fs *c, unsigned feat)
 934 {
 935         mutex_lock(&c->sb_lock);
 936         if (!(c->sb.features & (1ULL << feat))) {
 937                 c->disk_sb.sb->features[0] |= cpu_to_le64(1ULL << feat);
 938
 939                 bch2_write_super(c);
 940         }
 941         mutex_unlock(&c->sb_lock);
 942 }
 943
 944 /* BCH_SB_FIELD_members: */
 945
 946 static int bch2_sb_members_validate(struct bch_sb *sb,
 947                                     struct bch_sb_field *f,
 948                                     struct printbuf *err)
 949 {
 950         struct bch_sb_field_members *mi = field_to_type(f, members);
 951         unsigned i;
 952
 953         if ((void *) (mi->members + sb->nr_devices) >
 954             vstruct_end(&mi->field)) {
 955                 prt_printf(err, "too many devices for section size");
 956                 return -BCH_ERR_invalid_sb_members;
 957         }
 958
 959         for (i = 0; i < sb->nr_devices; i++) {
 960                 struct bch_member *m = mi->members + i;
 961
 962                 if (!bch2_member_exists(m))
 963                         continue;
 964
 965                 if (le64_to_cpu(m->nbuckets) > LONG_MAX) {
 966                         prt_printf(err, "device %u: too many buckets (got %llu, max %lu)",
 967                                i, le64_to_cpu(m->nbuckets), LONG_MAX);
 968                         return -BCH_ERR_invalid_sb_members;
 969                 }
 970
 971                 if (le64_to_cpu(m->nbuckets) -
 972                     le16_to_cpu(m->first_bucket) < BCH_MIN_NR_NBUCKETS) {
 973                         prt_printf(err, "device %u: not enough buckets (got %llu, max %u)",
 974                                i, le64_to_cpu(m->nbuckets), BCH_MIN_NR_NBUCKETS);
 975                         return -BCH_ERR_invalid_sb_members;
 976                 }
 977
 978                 if (le16_to_cpu(m->bucket_size) <
 979                     le16_to_cpu(sb->block_size)) {
 980                         prt_printf(err, "device %u: bucket size %u smaller than block size %u",
 981                                i, le16_to_cpu(m->bucket_size), le16_to_cpu(sb->block_size));
 982                         return -BCH_ERR_invalid_sb_members;
 983                 }
 984
 985                 if (le16_to_cpu(m->bucket_size) <
 986                     BCH_SB_BTREE_NODE_SIZE(sb)) {
 987                         prt_printf(err, "device %u: bucket size %u smaller than btree node size %llu",
 988                                i, le16_to_cpu(m->bucket_size), BCH_SB_BTREE_NODE_SIZE(sb));
 989                         return -BCH_ERR_invalid_sb_members;
 990                 }
 991         }
 992
 993         return 0;
 994 }
 995
 996 static void bch2_sb_members_to_text(struct printbuf *out, struct bch_sb *sb,
 997                                     struct bch_sb_field *f)
 998 {
 999         struct bch_sb_field_members *mi = field_to_type(f, members);
1000         struct bch_sb_field_disk_groups *gi = bch2_sb_get_disk_groups(sb);
1001         unsigned i;
1002
1003         for (i = 0; i < sb->nr_devices; i++) {
1004                 struct bch_member *m = mi->members + i;
1005                 unsigned data_have = bch2_sb_dev_has_data(sb, i);
1006                 u64 bucket_size = le16_to_cpu(m->bucket_size);
1007                 u64 device_size = le64_to_cpu(m->nbuckets) * bucket_size;
1008
1009                 if (!bch2_member_exists(m))
1010                         continue;
1011
1012                 prt_printf(out, "Device:");
1013                 prt_tab(out);
1014                 prt_printf(out, "%u", i);
1015                 prt_newline(out);
1016
1017                 printbuf_indent_add(out, 2);
1018
1019                 prt_printf(out, "UUID:");
1020                 prt_tab(out);
1021                 pr_uuid(out, m->uuid.b);
1022                 prt_newline(out);
1023
1024                 prt_printf(out, "Size:");
1025                 prt_tab(out);
1026                 prt_units_u64(out, device_size << 9);
1027                 prt_newline(out);
1028
1029                 prt_printf(out, "Bucket size:");
1030                 prt_tab(out);
1031                 prt_units_u64(out, bucket_size << 9);
1032                 prt_newline(out);
1033
1034                 prt_printf(out, "First bucket:");
1035                 prt_tab(out);
1036                 prt_printf(out, "%u", le16_to_cpu(m->first_bucket));
1037                 prt_newline(out);
1038
1039                 prt_printf(out, "Buckets:");
1040                 prt_tab(out);
1041                 prt_printf(out, "%llu", le64_to_cpu(m->nbuckets));
1042                 prt_newline(out);
1043
1044                 prt_printf(out, "Last mount:");
1045                 prt_tab(out);
1046                 if (m->last_mount)
1047                         pr_time(out, le64_to_cpu(m->last_mount));
1048                 else
1049                         prt_printf(out, "(never)");
1050                 prt_newline(out);
1051
1052                 prt_printf(out, "State:");
1053                 prt_tab(out);
1054                 prt_printf(out, "%s",
1055                        BCH_MEMBER_STATE(m) < BCH_MEMBER_STATE_NR
1056                        ? bch2_member_states[BCH_MEMBER_STATE(m)]
1057                        : "unknown");
1058                 prt_newline(out);
1059
1060                 prt_printf(out, "Label:");
1061                 prt_tab(out);
1062                 if (BCH_MEMBER_GROUP(m)) {
1063                         unsigned idx = BCH_MEMBER_GROUP(m) - 1;
1064
1065                         if (idx < disk_groups_nr(gi))
1066                                 prt_printf(out, "%s (%u)",
1067                                        gi->entries[idx].label, idx);
1068                         else
1069                                 prt_printf(out, "(bad disk labels section)");
1070                 } else {
1071                         prt_printf(out, "(none)");
1072                 }
1073                 prt_newline(out);
1074
1075                 prt_printf(out, "Data allowed:");
1076                 prt_tab(out);
1077                 if (BCH_MEMBER_DATA_ALLOWED(m))
1078                         prt_bitflags(out, bch2_data_types, BCH_MEMBER_DATA_ALLOWED(m));
1079                 else
1080                         prt_printf(out, "(none)");
1081                 prt_newline(out);
1082
1083                 prt_printf(out, "Has data:");
1084                 prt_tab(out);
1085                 if (data_have)
1086                         prt_bitflags(out, bch2_data_types, data_have);
1087                 else
1088                         prt_printf(out, "(none)");
1089                 prt_newline(out);
1090
1091                 prt_printf(out, "Discard:");
1092                 prt_tab(out);
1093                 prt_printf(out, "%llu", BCH_MEMBER_DISCARD(m));
1094                 prt_newline(out);
1095
1096                 prt_printf(out, "Freespace initialized:");
1097                 prt_tab(out);
1098                 prt_printf(out, "%llu", BCH_MEMBER_FREESPACE_INITIALIZED(m));
1099                 prt_newline(out);
1100
1101                 printbuf_indent_sub(out, 2);
1102         }
1103 }
1104
1105 static const struct bch_sb_field_ops bch_sb_field_ops_members = {
1106         .validate       = bch2_sb_members_validate,
1107         .to_text        = bch2_sb_members_to_text,
1108 };
1109
1110 /* BCH_SB_FIELD_crypt: */
1111
1112 static int bch2_sb_crypt_validate(struct bch_sb *sb,
1113                                   struct bch_sb_field *f,
1114                                   struct printbuf *err)
1115 {
1116         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1117
1118         if (vstruct_bytes(&crypt->field) < sizeof(*crypt)) {
1119                 prt_printf(err, "wrong size (got %zu should be %zu)",
1120                        vstruct_bytes(&crypt->field), sizeof(*crypt));
1121                 return -BCH_ERR_invalid_sb_crypt;
1122         }
1123
1124         if (BCH_CRYPT_KDF_TYPE(crypt)) {
1125                 prt_printf(err, "bad kdf type %llu", BCH_CRYPT_KDF_TYPE(crypt));
1126                 return -BCH_ERR_invalid_sb_crypt;
1127         }
1128
1129         return 0;
1130 }
1131
1132 static void bch2_sb_crypt_to_text(struct printbuf *out, struct bch_sb *sb,
1133                                   struct bch_sb_field *f)
1134 {
1135         struct bch_sb_field_crypt *crypt = field_to_type(f, crypt);
1136
1137         prt_printf(out, "KFD:               %llu", BCH_CRYPT_KDF_TYPE(crypt));
1138         prt_newline(out);
1139         prt_printf(out, "scrypt n:          %llu", BCH_KDF_SCRYPT_N(crypt));
1140         prt_newline(out);
1141         prt_printf(out, "scrypt r:          %llu", BCH_KDF_SCRYPT_R(crypt));
1142         prt_newline(out);
1143         prt_printf(out, "scrypt p:          %llu", BCH_KDF_SCRYPT_P(crypt));
1144         prt_newline(out);
1145 }
1146
1147 static const struct bch_sb_field_ops bch_sb_field_ops_crypt = {
1148         .validate       = bch2_sb_crypt_validate,
1149         .to_text        = bch2_sb_crypt_to_text,
1150 };
1151
1152 /* BCH_SB_FIELD_clean: */
1153
1154 int bch2_sb_clean_validate_late(struct bch_fs *c, struct bch_sb_field_clean *clean, int write)
1155 {
1156         struct jset_entry *entry;
1157         int ret;
1158
1159         for (entry = clean->start;
1160              entry < (struct jset_entry *) vstruct_end(&clean->field);
1161              entry = vstruct_next(entry)) {
1162                 ret = bch2_journal_entry_validate(c, NULL, entry,
1163                                                   le16_to_cpu(c->disk_sb.sb->version),
1164                                                   BCH_SB_BIG_ENDIAN(c->disk_sb.sb),
1165                                                   write);
1166                 if (ret)
1167                         return ret;
1168         }
1169
1170         return 0;
1171 }
1172
1173 int bch2_fs_mark_dirty(struct bch_fs *c)
1174 {
1175         int ret;
1176
1177         /*
1178          * Unconditionally write superblock, to verify it hasn't changed before
1179          * we go rw:
1180          */
1181
1182         mutex_lock(&c->sb_lock);
1183         SET_BCH_SB_CLEAN(c->disk_sb.sb, false);
1184         c->disk_sb.sb->features[0] |= cpu_to_le64(BCH_SB_FEATURES_ALWAYS);
1185         c->disk_sb.sb->compat[0] &= cpu_to_le64((1ULL << BCH_COMPAT_NR) - 1);
1186         ret = bch2_write_super(c);
1187         mutex_unlock(&c->sb_lock);
1188
1189         return ret;
1190 }
1191
1192 static struct jset_entry *jset_entry_init(struct jset_entry **end, size_t size)
1193 {
1194         struct jset_entry *entry = *end;
1195         unsigned u64s = DIV_ROUND_UP(size, sizeof(u64));
1196
1197         memset(entry, 0, u64s * sizeof(u64));
1198         /*
1199          * The u64s field counts from the start of data, ignoring the shared
1200          * fields.
1201          */
1202         entry->u64s = cpu_to_le16(u64s - 1);
1203
1204         *end = vstruct_next(*end);
1205         return entry;
1206 }
1207
1208 void bch2_journal_super_entries_add_common(struct bch_fs *c,
1209                                            struct jset_entry **end,
1210                                            u64 journal_seq)
1211 {
1212         struct bch_dev *ca;
1213         unsigned i, dev;
1214
1215         percpu_down_read(&c->mark_lock);
1216
1217         if (!journal_seq) {
1218                 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1219                         bch2_fs_usage_acc_to_base(c, i);
1220         } else {
1221                 bch2_fs_usage_acc_to_base(c, journal_seq & JOURNAL_BUF_MASK);
1222         }
1223
1224         {
1225                 struct jset_entry_usage *u =
1226                         container_of(jset_entry_init(end, sizeof(*u)),
1227                                      struct jset_entry_usage, entry);
1228
1229                 u->entry.type   = BCH_JSET_ENTRY_usage;
1230                 u->entry.btree_id = BCH_FS_USAGE_inodes;
1231                 u->v            = cpu_to_le64(c->usage_base->nr_inodes);
1232         }
1233
1234         {
1235                 struct jset_entry_usage *u =
1236                         container_of(jset_entry_init(end, sizeof(*u)),
1237                                      struct jset_entry_usage, entry);
1238
1239                 u->entry.type   = BCH_JSET_ENTRY_usage;
1240                 u->entry.btree_id = BCH_FS_USAGE_key_version;
1241                 u->v            = cpu_to_le64(atomic64_read(&c->key_version));
1242         }
1243
1244         for (i = 0; i < BCH_REPLICAS_MAX; i++) {
1245                 struct jset_entry_usage *u =
1246                         container_of(jset_entry_init(end, sizeof(*u)),
1247                                      struct jset_entry_usage, entry);
1248
1249                 u->entry.type   = BCH_JSET_ENTRY_usage;
1250                 u->entry.btree_id = BCH_FS_USAGE_reserved;
1251                 u->entry.level  = i;
1252                 u->v            = cpu_to_le64(c->usage_base->persistent_reserved[i]);
1253         }
1254
1255         for (i = 0; i < c->replicas.nr; i++) {
1256                 struct bch_replicas_entry *e =
1257                         cpu_replicas_entry(&c->replicas, i);
1258                 struct jset_entry_data_usage *u =
1259                         container_of(jset_entry_init(end, sizeof(*u) + e->nr_devs),
1260                                      struct jset_entry_data_usage, entry);
1261
1262                 u->entry.type   = BCH_JSET_ENTRY_data_usage;
1263                 u->v            = cpu_to_le64(c->usage_base->replicas[i]);
1264                 memcpy(&u->r, e, replicas_entry_bytes(e));
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 }