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

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include "bcachefs.h"
   3 #include "alloc_background.h"
   4 #include "alloc_foreground.h"
   5 #include "btree_io.h"
   6 #include "btree_update_interior.h"
   7 #include "btree_write_buffer.h"
   8 #include "buckets.h"
   9 #include "checksum.h"
  10 #include "disk_groups.h"
  11 #include "error.h"
  12 #include "journal.h"
  13 #include "journal_io.h"
  14 #include "journal_reclaim.h"
  15 #include "journal_seq_blacklist.h"
  16 #include "replicas.h"
  17 #include "sb-clean.h"
  18 #include "trace.h"
  19
  20 void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
  21                                struct journal_replay *j)
  22 {
  23         darray_for_each(j->ptrs, i) {
  24                 struct bch_dev *ca = bch_dev_bkey_exists(c, i->dev);
  25                 u64 offset;
  26
  27                 div64_u64_rem(i->sector, ca->mi.bucket_size, &offset);
  28
  29                 if (i != j->ptrs.data)
  30                         prt_printf(out, " ");
  31                 prt_printf(out, "%u:%u:%u (sector %llu)",
  32                            i->dev, i->bucket, i->bucket_offset, i->sector);
  33         }
  34 }
  35
  36 static void bch2_journal_replay_to_text(struct printbuf *out, struct bch_fs *c,
  37                                         struct journal_replay *j)
  38 {
  39         prt_printf(out, "seq %llu ", le64_to_cpu(j->j.seq));
  40
  41         bch2_journal_ptrs_to_text(out, c, j);
  42
  43         struct jset_entry *entry;
  44         for_each_jset_entry_type(entry, &j->j, BCH_JSET_ENTRY_datetime) {
  45                 struct jset_entry_datetime *datetime =
  46                         container_of(entry, struct jset_entry_datetime, entry);
  47                 bch2_prt_datetime(out, le64_to_cpu(datetime->seconds));
  48                 break;
  49         }
  50 }
  51
  52 static struct nonce journal_nonce(const struct jset *jset)
  53 {
  54         return (struct nonce) {{
  55                 [0] = 0,
  56                 [1] = ((__le32 *) &jset->seq)[0],
  57                 [2] = ((__le32 *) &jset->seq)[1],
  58                 [3] = BCH_NONCE_JOURNAL,
  59         }};
  60 }
  61
  62 static bool jset_csum_good(struct bch_fs *c, struct jset *j, struct bch_csum *csum)
  63 {
  64         if (!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(j))) {
  65                 *csum = (struct bch_csum) {};
  66                 return false;
  67         }
  68
  69         *csum = csum_vstruct(c, JSET_CSUM_TYPE(j), journal_nonce(j), j);
  70         return !bch2_crc_cmp(j->csum, *csum);
  71 }
  72
  73 static inline u32 journal_entry_radix_idx(struct bch_fs *c, u64 seq)
  74 {
  75         return (seq - c->journal_entries_base_seq) & (~0U >> 1);
  76 }
  77
  78 static void __journal_replay_free(struct bch_fs *c,
  79                                   struct journal_replay *i)
  80 {
  81         struct journal_replay **p =
  82                 genradix_ptr(&c->journal_entries,
  83                              journal_entry_radix_idx(c, le64_to_cpu(i->j.seq)));
  84
  85         BUG_ON(*p != i);
  86         *p = NULL;
  87         kvfree(i);
  88 }
  89
  90 static void journal_replay_free(struct bch_fs *c, struct journal_replay *i)
  91 {
  92         i->ignore = true;
  93
  94         if (!c->opts.read_entire_journal)
  95                 __journal_replay_free(c, i);
  96 }
  97
  98 struct journal_list {
  99         struct closure          cl;
 100         u64                     last_seq;
 101         struct mutex            lock;
 102         int                     ret;
 103 };
 104
 105 #define JOURNAL_ENTRY_ADD_OK            0
 106 #define JOURNAL_ENTRY_ADD_OUT_OF_RANGE  5
 107
 108 /*
 109  * Given a journal entry we just read, add it to the list of journal entries to
 110  * be replayed:
 111  */
 112 static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca,
 113                              struct journal_ptr entry_ptr,
 114                              struct journal_list *jlist, struct jset *j)
 115 {
 116         struct genradix_iter iter;
 117         struct journal_replay **_i, *i, *dup;
 118         size_t bytes = vstruct_bytes(j);
 119         u64 last_seq = !JSET_NO_FLUSH(j) ? le64_to_cpu(j->last_seq) : 0;
 120         struct printbuf buf = PRINTBUF;
 121         int ret = JOURNAL_ENTRY_ADD_OK;
 122
 123         /* Is this entry older than the range we need? */
 124         if (!c->opts.read_entire_journal &&
 125             le64_to_cpu(j->seq) < jlist->last_seq)
 126                 return JOURNAL_ENTRY_ADD_OUT_OF_RANGE;
 127
 128         /*
 129          * genradixes are indexed by a ulong, not a u64, so we can't index them
 130          * by sequence number directly: Assume instead that they will all fall
 131          * within the range of +-2billion of the filrst one we find.
 132          */
 133         if (!c->journal_entries_base_seq)
 134                 c->journal_entries_base_seq = max_t(s64, 1, le64_to_cpu(j->seq) - S32_MAX);
 135
 136         /* Drop entries we don't need anymore */
 137         if (last_seq > jlist->last_seq && !c->opts.read_entire_journal) {
 138                 genradix_for_each_from(&c->journal_entries, iter, _i,
 139                                        journal_entry_radix_idx(c, jlist->last_seq)) {
 140                         i = *_i;
 141
 142                         if (!i || i->ignore)
 143                                 continue;
 144
 145                         if (le64_to_cpu(i->j.seq) >= last_seq)
 146                                 break;
 147                         journal_replay_free(c, i);
 148                 }
 149         }
 150
 151         jlist->last_seq = max(jlist->last_seq, last_seq);
 152
 153         _i = genradix_ptr_alloc(&c->journal_entries,
 154                                 journal_entry_radix_idx(c, le64_to_cpu(j->seq)),
 155                                 GFP_KERNEL);
 156         if (!_i)
 157                 return -BCH_ERR_ENOMEM_journal_entry_add;
 158
 159         /*
 160          * Duplicate journal entries? If so we want the one that didn't have a
 161          * checksum error:
 162          */
 163         dup = *_i;
 164         if (dup) {
 165                 bool identical = bytes == vstruct_bytes(&dup->j) &&
 166                         !memcmp(j, &dup->j, bytes);
 167                 bool not_identical = !identical &&
 168                         entry_ptr.csum_good &&
 169                         dup->csum_good;
 170
 171                 bool same_device = false;
 172                 darray_for_each(dup->ptrs, ptr)
 173                         if (ptr->dev == ca->dev_idx)
 174                                 same_device = true;
 175
 176                 ret = darray_push(&dup->ptrs, entry_ptr);
 177                 if (ret)
 178                         goto out;
 179
 180                 bch2_journal_replay_to_text(&buf, c, dup);
 181
 182                 fsck_err_on(same_device,
 183                             c, journal_entry_dup_same_device,
 184                             "duplicate journal entry on same device\n  %s",
 185                             buf.buf);
 186
 187                 fsck_err_on(not_identical,
 188                             c, journal_entry_replicas_data_mismatch,
 189                             "found duplicate but non identical journal entries\n  %s",
 190                             buf.buf);
 191
 192                 if (entry_ptr.csum_good && !identical)
 193                         goto replace;
 194
 195                 goto out;
 196         }
 197 replace:
 198         i = kvmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL);
 199         if (!i)
 200                 return -BCH_ERR_ENOMEM_journal_entry_add;
 201
 202         darray_init(&i->ptrs);
 203         i->csum_good    = entry_ptr.csum_good;
 204         i->ignore       = false;
 205         unsafe_memcpy(&i->j, j, bytes, "embedded variable length struct");
 206
 207         if (dup) {
 208                 /* The first ptr should represent the jset we kept: */
 209                 darray_for_each(dup->ptrs, ptr)
 210                         darray_push(&i->ptrs, *ptr);
 211                 __journal_replay_free(c, dup);
 212         } else {
 213                 darray_push(&i->ptrs, entry_ptr);
 214         }
 215
 216         *_i = i;
 217 out:
 218 fsck_err:
 219         printbuf_exit(&buf);
 220         return ret;
 221 }
 222
 223 /* this fills in a range with empty jset_entries: */
 224 static void journal_entry_null_range(void *start, void *end)
 225 {
 226         struct jset_entry *entry;
 227
 228         for (entry = start; entry != end; entry = vstruct_next(entry))
 229                 memset(entry, 0, sizeof(*entry));
 230 }
 231
 232 #define JOURNAL_ENTRY_REREAD    5
 233 #define JOURNAL_ENTRY_NONE      6
 234 #define JOURNAL_ENTRY_BAD       7
 235
 236 static void journal_entry_err_msg(struct printbuf *out,
 237                                   u32 version,
 238                                   struct jset *jset,
 239                                   struct jset_entry *entry)
 240 {
 241         prt_str(out, "invalid journal entry, version=");
 242         bch2_version_to_text(out, version);
 243
 244         if (entry) {
 245                 prt_str(out, " type=");
 246                 prt_str(out, bch2_jset_entry_types[entry->type]);
 247         }
 248
 249         if (!jset) {
 250                 prt_printf(out, " in superblock");
 251         } else {
 252
 253                 prt_printf(out, " seq=%llu", le64_to_cpu(jset->seq));
 254
 255                 if (entry)
 256                         prt_printf(out, " offset=%zi/%u",
 257                                    (u64 *) entry - jset->_data,
 258                                    le32_to_cpu(jset->u64s));
 259         }
 260
 261         prt_str(out, ": ");
 262 }
 263
 264 #define journal_entry_err(c, version, jset, entry, _err, msg, ...)      \
 265 ({                                                                      \
 266         struct printbuf _buf = PRINTBUF;                                \
 267                                                                         \
 268         journal_entry_err_msg(&_buf, version, jset, entry);             \
 269         prt_printf(&_buf, msg, ##__VA_ARGS__);                          \
 270                                                                         \
 271         switch (flags & BKEY_INVALID_WRITE) {                           \
 272         case READ:                                                      \
 273                 mustfix_fsck_err(c, _err, "%s", _buf.buf);              \
 274                 break;                                                  \
 275         case WRITE:                                                     \
 276                 bch2_sb_error_count(c, BCH_FSCK_ERR_##_err);            \
 277                 bch_err(c, "corrupt metadata before write: %s\n", _buf.buf);\
 278                 if (bch2_fs_inconsistent(c)) {                          \
 279                         ret = -BCH_ERR_fsck_errors_not_fixed;           \
 280                         goto fsck_err;                                  \
 281                 }                                                       \
 282                 break;                                                  \
 283         }                                                               \
 284                                                                         \
 285         printbuf_exit(&_buf);                                           \
 286         true;                                                           \
 287 })
 288
 289 #define journal_entry_err_on(cond, ...)                                 \
 290         ((cond) ? journal_entry_err(__VA_ARGS__) : false)
 291
 292 #define FSCK_DELETED_KEY        5
 293
 294 static int journal_validate_key(struct bch_fs *c,
 295                                 struct jset *jset,
 296                                 struct jset_entry *entry,
 297                                 unsigned level, enum btree_id btree_id,
 298                                 struct bkey_i *k,
 299                                 unsigned version, int big_endian,
 300                                 enum bkey_invalid_flags flags)
 301 {
 302         int write = flags & BKEY_INVALID_WRITE;
 303         void *next = vstruct_next(entry);
 304         struct printbuf buf = PRINTBUF;
 305         int ret = 0;
 306
 307         if (journal_entry_err_on(!k->k.u64s,
 308                                  c, version, jset, entry,
 309                                  journal_entry_bkey_u64s_0,
 310                                  "k->u64s 0")) {
 311                 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
 312                 journal_entry_null_range(vstruct_next(entry), next);
 313                 return FSCK_DELETED_KEY;
 314         }
 315
 316         if (journal_entry_err_on((void *) bkey_next(k) >
 317                                  (void *) vstruct_next(entry),
 318                                  c, version, jset, entry,
 319                                  journal_entry_bkey_past_end,
 320                                  "extends past end of journal entry")) {
 321                 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
 322                 journal_entry_null_range(vstruct_next(entry), next);
 323                 return FSCK_DELETED_KEY;
 324         }
 325
 326         if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT,
 327                                  c, version, jset, entry,
 328                                  journal_entry_bkey_bad_format,
 329                                  "bad format %u", k->k.format)) {
 330                 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
 331                 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
 332                 journal_entry_null_range(vstruct_next(entry), next);
 333                 return FSCK_DELETED_KEY;
 334         }
 335
 336         if (!write)
 337                 bch2_bkey_compat(level, btree_id, version, big_endian,
 338                                  write, NULL, bkey_to_packed(k));
 339
 340         if (bch2_bkey_invalid(c, bkey_i_to_s_c(k),
 341                               __btree_node_type(level, btree_id), write, &buf)) {
 342                 printbuf_reset(&buf);
 343                 journal_entry_err_msg(&buf, version, jset, entry);
 344                 prt_newline(&buf);
 345                 printbuf_indent_add(&buf, 2);
 346
 347                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
 348                 prt_newline(&buf);
 349                 bch2_bkey_invalid(c, bkey_i_to_s_c(k),
 350                                   __btree_node_type(level, btree_id), write, &buf);
 351
 352                 mustfix_fsck_err(c, journal_entry_bkey_invalid,
 353                                  "%s", buf.buf);
 354
 355                 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
 356                 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
 357                 journal_entry_null_range(vstruct_next(entry), next);
 358
 359                 printbuf_exit(&buf);
 360                 return FSCK_DELETED_KEY;
 361         }
 362
 363         if (write)
 364                 bch2_bkey_compat(level, btree_id, version, big_endian,
 365                                  write, NULL, bkey_to_packed(k));
 366 fsck_err:
 367         printbuf_exit(&buf);
 368         return ret;
 369 }
 370
 371 static int journal_entry_btree_keys_validate(struct bch_fs *c,
 372                                 struct jset *jset,
 373                                 struct jset_entry *entry,
 374                                 unsigned version, int big_endian,
 375                                 enum bkey_invalid_flags flags)
 376 {
 377         struct bkey_i *k = entry->start;
 378
 379         while (k != vstruct_last(entry)) {
 380                 int ret = journal_validate_key(c, jset, entry,
 381                                                entry->level,
 382                                                entry->btree_id,
 383                                                k, version, big_endian,
 384                                                flags|BKEY_INVALID_JOURNAL);
 385                 if (ret == FSCK_DELETED_KEY)
 386                         continue;
 387
 388                 k = bkey_next(k);
 389         }
 390
 391         return 0;
 392 }
 393
 394 static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs *c,
 395                                              struct jset_entry *entry)
 396 {
 397         struct bkey_i *k;
 398         bool first = true;
 399
 400         jset_entry_for_each_key(entry, k) {
 401                 if (!first) {
 402                         prt_newline(out);
 403                         prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
 404                 }
 405                 prt_printf(out, "btree=%s l=%u ", bch2_btree_id_str(entry->btree_id), entry->level);
 406                 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k));
 407                 first = false;
 408         }
 409 }
 410
 411 static int journal_entry_btree_root_validate(struct bch_fs *c,
 412                                 struct jset *jset,
 413                                 struct jset_entry *entry,
 414                                 unsigned version, int big_endian,
 415                                 enum bkey_invalid_flags flags)
 416 {
 417         struct bkey_i *k = entry->start;
 418         int ret = 0;
 419
 420         if (journal_entry_err_on(!entry->u64s ||
 421                                  le16_to_cpu(entry->u64s) != k->k.u64s,
 422                                  c, version, jset, entry,
 423                                  journal_entry_btree_root_bad_size,
 424                                  "invalid btree root journal entry: wrong number of keys")) {
 425                 void *next = vstruct_next(entry);
 426                 /*
 427                  * we don't want to null out this jset_entry,
 428                  * just the contents, so that later we can tell
 429                  * we were _supposed_ to have a btree root
 430                  */
 431                 entry->u64s = 0;
 432                 journal_entry_null_range(vstruct_next(entry), next);
 433                 return 0;
 434         }
 435
 436         ret = journal_validate_key(c, jset, entry, 1, entry->btree_id, k,
 437                                    version, big_endian, flags);
 438         if (ret == FSCK_DELETED_KEY)
 439                 ret = 0;
 440 fsck_err:
 441         return ret;
 442 }
 443
 444 static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs *c,
 445                                              struct jset_entry *entry)
 446 {
 447         journal_entry_btree_keys_to_text(out, c, entry);
 448 }
 449
 450 static int journal_entry_prio_ptrs_validate(struct bch_fs *c,
 451                                 struct jset *jset,
 452                                 struct jset_entry *entry,
 453                                 unsigned version, int big_endian,
 454                                 enum bkey_invalid_flags flags)
 455 {
 456         /* obsolete, don't care: */
 457         return 0;
 458 }
 459
 460 static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 461                                             struct jset_entry *entry)
 462 {
 463 }
 464
 465 static int journal_entry_blacklist_validate(struct bch_fs *c,
 466                                 struct jset *jset,
 467                                 struct jset_entry *entry,
 468                                 unsigned version, int big_endian,
 469                                 enum bkey_invalid_flags flags)
 470 {
 471         int ret = 0;
 472
 473         if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1,
 474                                  c, version, jset, entry,
 475                                  journal_entry_blacklist_bad_size,
 476                 "invalid journal seq blacklist entry: bad size")) {
 477                 journal_entry_null_range(entry, vstruct_next(entry));
 478         }
 479 fsck_err:
 480         return ret;
 481 }
 482
 483 static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs *c,
 484                                             struct jset_entry *entry)
 485 {
 486         struct jset_entry_blacklist *bl =
 487                 container_of(entry, struct jset_entry_blacklist, entry);
 488
 489         prt_printf(out, "seq=%llu", le64_to_cpu(bl->seq));
 490 }
 491
 492 static int journal_entry_blacklist_v2_validate(struct bch_fs *c,
 493                                 struct jset *jset,
 494                                 struct jset_entry *entry,
 495                                 unsigned version, int big_endian,
 496                                 enum bkey_invalid_flags flags)
 497 {
 498         struct jset_entry_blacklist_v2 *bl_entry;
 499         int ret = 0;
 500
 501         if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2,
 502                                  c, version, jset, entry,
 503                                  journal_entry_blacklist_v2_bad_size,
 504                 "invalid journal seq blacklist entry: bad size")) {
 505                 journal_entry_null_range(entry, vstruct_next(entry));
 506                 goto out;
 507         }
 508
 509         bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry);
 510
 511         if (journal_entry_err_on(le64_to_cpu(bl_entry->start) >
 512                                  le64_to_cpu(bl_entry->end),
 513                                  c, version, jset, entry,
 514                                  journal_entry_blacklist_v2_start_past_end,
 515                 "invalid journal seq blacklist entry: start > end")) {
 516                 journal_entry_null_range(entry, vstruct_next(entry));
 517         }
 518 out:
 519 fsck_err:
 520         return ret;
 521 }
 522
 523 static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_fs *c,
 524                                                struct jset_entry *entry)
 525 {
 526         struct jset_entry_blacklist_v2 *bl =
 527                 container_of(entry, struct jset_entry_blacklist_v2, entry);
 528
 529         prt_printf(out, "start=%llu end=%llu",
 530                le64_to_cpu(bl->start),
 531                le64_to_cpu(bl->end));
 532 }
 533
 534 static int journal_entry_usage_validate(struct bch_fs *c,
 535                                 struct jset *jset,
 536                                 struct jset_entry *entry,
 537                                 unsigned version, int big_endian,
 538                                 enum bkey_invalid_flags flags)
 539 {
 540         struct jset_entry_usage *u =
 541                 container_of(entry, struct jset_entry_usage, entry);
 542         unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 543         int ret = 0;
 544
 545         if (journal_entry_err_on(bytes < sizeof(*u),
 546                                  c, version, jset, entry,
 547                                  journal_entry_usage_bad_size,
 548                                  "invalid journal entry usage: bad size")) {
 549                 journal_entry_null_range(entry, vstruct_next(entry));
 550                 return ret;
 551         }
 552
 553 fsck_err:
 554         return ret;
 555 }
 556
 557 static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c,
 558                                         struct jset_entry *entry)
 559 {
 560         struct jset_entry_usage *u =
 561                 container_of(entry, struct jset_entry_usage, entry);
 562
 563         prt_printf(out, "type=%s v=%llu",
 564                bch2_fs_usage_types[u->entry.btree_id],
 565                le64_to_cpu(u->v));
 566 }
 567
 568 static int journal_entry_data_usage_validate(struct bch_fs *c,
 569                                 struct jset *jset,
 570                                 struct jset_entry *entry,
 571                                 unsigned version, int big_endian,
 572                                 enum bkey_invalid_flags flags)
 573 {
 574         struct jset_entry_data_usage *u =
 575                 container_of(entry, struct jset_entry_data_usage, entry);
 576         unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 577         struct printbuf err = PRINTBUF;
 578         int ret = 0;
 579
 580         if (journal_entry_err_on(bytes < sizeof(*u) ||
 581                                  bytes < sizeof(*u) + u->r.nr_devs,
 582                                  c, version, jset, entry,
 583                                  journal_entry_data_usage_bad_size,
 584                                  "invalid journal entry usage: bad size")) {
 585                 journal_entry_null_range(entry, vstruct_next(entry));
 586                 goto out;
 587         }
 588
 589         if (journal_entry_err_on(bch2_replicas_entry_validate(&u->r, c->disk_sb.sb, &err),
 590                                  c, version, jset, entry,
 591                                  journal_entry_data_usage_bad_size,
 592                                  "invalid journal entry usage: %s", err.buf)) {
 593                 journal_entry_null_range(entry, vstruct_next(entry));
 594                 goto out;
 595         }
 596 out:
 597 fsck_err:
 598         printbuf_exit(&err);
 599         return ret;
 600 }
 601
 602 static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs *c,
 603                                              struct jset_entry *entry)
 604 {
 605         struct jset_entry_data_usage *u =
 606                 container_of(entry, struct jset_entry_data_usage, entry);
 607
 608         bch2_replicas_entry_to_text(out, &u->r);
 609         prt_printf(out, "=%llu", le64_to_cpu(u->v));
 610 }
 611
 612 static int journal_entry_clock_validate(struct bch_fs *c,
 613                                 struct jset *jset,
 614                                 struct jset_entry *entry,
 615                                 unsigned version, int big_endian,
 616                                 enum bkey_invalid_flags flags)
 617 {
 618         struct jset_entry_clock *clock =
 619                 container_of(entry, struct jset_entry_clock, entry);
 620         unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 621         int ret = 0;
 622
 623         if (journal_entry_err_on(bytes != sizeof(*clock),
 624                                  c, version, jset, entry,
 625                                  journal_entry_clock_bad_size,
 626                                  "bad size")) {
 627                 journal_entry_null_range(entry, vstruct_next(entry));
 628                 return ret;
 629         }
 630
 631         if (journal_entry_err_on(clock->rw > 1,
 632                                  c, version, jset, entry,
 633                                  journal_entry_clock_bad_rw,
 634                                  "bad rw")) {
 635                 journal_entry_null_range(entry, vstruct_next(entry));
 636                 return ret;
 637         }
 638
 639 fsck_err:
 640         return ret;
 641 }
 642
 643 static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c,
 644                                         struct jset_entry *entry)
 645 {
 646         struct jset_entry_clock *clock =
 647                 container_of(entry, struct jset_entry_clock, entry);
 648
 649         prt_printf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time));
 650 }
 651
 652 static int journal_entry_dev_usage_validate(struct bch_fs *c,
 653                                 struct jset *jset,
 654                                 struct jset_entry *entry,
 655                                 unsigned version, int big_endian,
 656                                 enum bkey_invalid_flags flags)
 657 {
 658         struct jset_entry_dev_usage *u =
 659                 container_of(entry, struct jset_entry_dev_usage, entry);
 660         unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 661         unsigned expected = sizeof(*u);
 662         unsigned dev;
 663         int ret = 0;
 664
 665         if (journal_entry_err_on(bytes < expected,
 666                                  c, version, jset, entry,
 667                                  journal_entry_dev_usage_bad_size,
 668                                  "bad size (%u < %u)",
 669                                  bytes, expected)) {
 670                 journal_entry_null_range(entry, vstruct_next(entry));
 671                 return ret;
 672         }
 673
 674         dev = le32_to_cpu(u->dev);
 675
 676         if (journal_entry_err_on(!bch2_dev_exists2(c, dev),
 677                                  c, version, jset, entry,
 678                                  journal_entry_dev_usage_bad_dev,
 679                                  "bad dev")) {
 680                 journal_entry_null_range(entry, vstruct_next(entry));
 681                 return ret;
 682         }
 683
 684         if (journal_entry_err_on(u->pad,
 685                                  c, version, jset, entry,
 686                                  journal_entry_dev_usage_bad_pad,
 687                                  "bad pad")) {
 688                 journal_entry_null_range(entry, vstruct_next(entry));
 689                 return ret;
 690         }
 691
 692 fsck_err:
 693         return ret;
 694 }
 695
 696 static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs *c,
 697                                             struct jset_entry *entry)
 698 {
 699         struct jset_entry_dev_usage *u =
 700                 container_of(entry, struct jset_entry_dev_usage, entry);
 701         unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 702
 703         prt_printf(out, "dev=%u", le32_to_cpu(u->dev));
 704
 705         for (i = 0; i < nr_types; i++) {
 706                 bch2_prt_data_type(out, i);
 707                 prt_printf(out, ": buckets=%llu sectors=%llu fragmented=%llu",
 708                        le64_to_cpu(u->d[i].buckets),
 709                        le64_to_cpu(u->d[i].sectors),
 710                        le64_to_cpu(u->d[i].fragmented));
 711         }
 712 }
 713
 714 static int journal_entry_log_validate(struct bch_fs *c,
 715                                 struct jset *jset,
 716                                 struct jset_entry *entry,
 717                                 unsigned version, int big_endian,
 718                                 enum bkey_invalid_flags flags)
 719 {
 720         return 0;
 721 }
 722
 723 static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c,
 724                                       struct jset_entry *entry)
 725 {
 726         struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry);
 727         unsigned bytes = vstruct_bytes(entry) - offsetof(struct jset_entry_log, d);
 728
 729         prt_printf(out, "%.*s", bytes, l->d);
 730 }
 731
 732 static int journal_entry_overwrite_validate(struct bch_fs *c,
 733                                 struct jset *jset,
 734                                 struct jset_entry *entry,
 735                                 unsigned version, int big_endian,
 736                                 enum bkey_invalid_flags flags)
 737 {
 738         return journal_entry_btree_keys_validate(c, jset, entry,
 739                                 version, big_endian, READ);
 740 }
 741
 742 static void journal_entry_overwrite_to_text(struct printbuf *out, struct bch_fs *c,
 743                                             struct jset_entry *entry)
 744 {
 745         journal_entry_btree_keys_to_text(out, c, entry);
 746 }
 747
 748 static int journal_entry_write_buffer_keys_validate(struct bch_fs *c,
 749                                 struct jset *jset,
 750                                 struct jset_entry *entry,
 751                                 unsigned version, int big_endian,
 752                                 enum bkey_invalid_flags flags)
 753 {
 754         return journal_entry_btree_keys_validate(c, jset, entry,
 755                                 version, big_endian, READ);
 756 }
 757
 758 static void journal_entry_write_buffer_keys_to_text(struct printbuf *out, struct bch_fs *c,
 759                                             struct jset_entry *entry)
 760 {
 761         journal_entry_btree_keys_to_text(out, c, entry);
 762 }
 763
 764 static int journal_entry_datetime_validate(struct bch_fs *c,
 765                                 struct jset *jset,
 766                                 struct jset_entry *entry,
 767                                 unsigned version, int big_endian,
 768                                 enum bkey_invalid_flags flags)
 769 {
 770         unsigned bytes = vstruct_bytes(entry);
 771         unsigned expected = 16;
 772         int ret = 0;
 773
 774         if (journal_entry_err_on(vstruct_bytes(entry) < expected,
 775                                  c, version, jset, entry,
 776                                  journal_entry_dev_usage_bad_size,
 777                                  "bad size (%u < %u)",
 778                                  bytes, expected)) {
 779                 journal_entry_null_range(entry, vstruct_next(entry));
 780                 return ret;
 781         }
 782 fsck_err:
 783         return ret;
 784 }
 785
 786 static void journal_entry_datetime_to_text(struct printbuf *out, struct bch_fs *c,
 787                                             struct jset_entry *entry)
 788 {
 789         struct jset_entry_datetime *datetime =
 790                 container_of(entry, struct jset_entry_datetime, entry);
 791
 792         bch2_prt_datetime(out, le64_to_cpu(datetime->seconds));
 793 }
 794
 795 struct jset_entry_ops {
 796         int (*validate)(struct bch_fs *, struct jset *,
 797                         struct jset_entry *, unsigned, int,
 798                         enum bkey_invalid_flags);
 799         void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *);
 800 };
 801
 802 static const struct jset_entry_ops bch2_jset_entry_ops[] = {
 803 #define x(f, nr)                                                \
 804         [BCH_JSET_ENTRY_##f]    = (struct jset_entry_ops) {     \
 805                 .validate       = journal_entry_##f##_validate, \
 806                 .to_text        = journal_entry_##f##_to_text,  \
 807         },
 808         BCH_JSET_ENTRY_TYPES()
 809 #undef x
 810 };
 811
 812 int bch2_journal_entry_validate(struct bch_fs *c,
 813                                 struct jset *jset,
 814                                 struct jset_entry *entry,
 815                                 unsigned version, int big_endian,
 816                                 enum bkey_invalid_flags flags)
 817 {
 818         return entry->type < BCH_JSET_ENTRY_NR
 819                 ? bch2_jset_entry_ops[entry->type].validate(c, jset, entry,
 820                                 version, big_endian, flags)
 821                 : 0;
 822 }
 823
 824 void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c,
 825                                 struct jset_entry *entry)
 826 {
 827         if (entry->type < BCH_JSET_ENTRY_NR) {
 828                 prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
 829                 bch2_jset_entry_ops[entry->type].to_text(out, c, entry);
 830         } else {
 831                 prt_printf(out, "(unknown type %u)", entry->type);
 832         }
 833 }
 834
 835 static int jset_validate_entries(struct bch_fs *c, struct jset *jset,
 836                                  enum bkey_invalid_flags flags)
 837 {
 838         unsigned version = le32_to_cpu(jset->version);
 839         int ret = 0;
 840
 841         vstruct_for_each(jset, entry) {
 842                 if (journal_entry_err_on(vstruct_next(entry) > vstruct_last(jset),
 843                                 c, version, jset, entry,
 844                                 journal_entry_past_jset_end,
 845                                 "journal entry extends past end of jset")) {
 846                         jset->u64s = cpu_to_le32((u64 *) entry - jset->_data);
 847                         break;
 848                 }
 849
 850                 ret = bch2_journal_entry_validate(c, jset, entry,
 851                                         version, JSET_BIG_ENDIAN(jset), flags);
 852                 if (ret)
 853                         break;
 854         }
 855 fsck_err:
 856         return ret;
 857 }
 858
 859 static int jset_validate(struct bch_fs *c,
 860                          struct bch_dev *ca,
 861                          struct jset *jset, u64 sector,
 862                          enum bkey_invalid_flags flags)
 863 {
 864         unsigned version;
 865         int ret = 0;
 866
 867         if (le64_to_cpu(jset->magic) != jset_magic(c))
 868                 return JOURNAL_ENTRY_NONE;
 869
 870         version = le32_to_cpu(jset->version);
 871         if (journal_entry_err_on(!bch2_version_compatible(version),
 872                         c, version, jset, NULL,
 873                         jset_unsupported_version,
 874                         "%s sector %llu seq %llu: incompatible journal entry version %u.%u",
 875                         ca ? ca->name : c->name,
 876                         sector, le64_to_cpu(jset->seq),
 877                         BCH_VERSION_MAJOR(version),
 878                         BCH_VERSION_MINOR(version))) {
 879                 /* don't try to continue: */
 880                 return -EINVAL;
 881         }
 882
 883         if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)),
 884                         c, version, jset, NULL,
 885                         jset_unknown_csum,
 886                         "%s sector %llu seq %llu: journal entry with unknown csum type %llu",
 887                         ca ? ca->name : c->name,
 888                         sector, le64_to_cpu(jset->seq),
 889                         JSET_CSUM_TYPE(jset)))
 890                 ret = JOURNAL_ENTRY_BAD;
 891
 892         /* last_seq is ignored when JSET_NO_FLUSH is true */
 893         if (journal_entry_err_on(!JSET_NO_FLUSH(jset) &&
 894                                  le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq),
 895                                  c, version, jset, NULL,
 896                                  jset_last_seq_newer_than_seq,
 897                                  "invalid journal entry: last_seq > seq (%llu > %llu)",
 898                                  le64_to_cpu(jset->last_seq),
 899                                  le64_to_cpu(jset->seq))) {
 900                 jset->last_seq = jset->seq;
 901                 return JOURNAL_ENTRY_BAD;
 902         }
 903
 904         ret = jset_validate_entries(c, jset, flags);
 905 fsck_err:
 906         return ret;
 907 }
 908
 909 static int jset_validate_early(struct bch_fs *c,
 910                          struct bch_dev *ca,
 911                          struct jset *jset, u64 sector,
 912                          unsigned bucket_sectors_left,
 913                          unsigned sectors_read)
 914 {
 915         size_t bytes = vstruct_bytes(jset);
 916         unsigned version;
 917         enum bkey_invalid_flags flags = BKEY_INVALID_JOURNAL;
 918         int ret = 0;
 919
 920         if (le64_to_cpu(jset->magic) != jset_magic(c))
 921                 return JOURNAL_ENTRY_NONE;
 922
 923         version = le32_to_cpu(jset->version);
 924         if (journal_entry_err_on(!bch2_version_compatible(version),
 925                         c, version, jset, NULL,
 926                         jset_unsupported_version,
 927                         "%s sector %llu seq %llu: unknown journal entry version %u.%u",
 928                         ca ? ca->name : c->name,
 929                         sector, le64_to_cpu(jset->seq),
 930                         BCH_VERSION_MAJOR(version),
 931                         BCH_VERSION_MINOR(version))) {
 932                 /* don't try to continue: */
 933                 return -EINVAL;
 934         }
 935
 936         if (bytes > (sectors_read << 9) &&
 937             sectors_read < bucket_sectors_left)
 938                 return JOURNAL_ENTRY_REREAD;
 939
 940         if (journal_entry_err_on(bytes > bucket_sectors_left << 9,
 941                         c, version, jset, NULL,
 942                         jset_past_bucket_end,
 943                         "%s sector %llu seq %llu: journal entry too big (%zu bytes)",
 944                         ca ? ca->name : c->name,
 945                         sector, le64_to_cpu(jset->seq), bytes))
 946                 le32_add_cpu(&jset->u64s,
 947                              -((bytes - (bucket_sectors_left << 9)) / 8));
 948 fsck_err:
 949         return ret;
 950 }
 951
 952 struct journal_read_buf {
 953         void            *data;
 954         size_t          size;
 955 };
 956
 957 static int journal_read_buf_realloc(struct journal_read_buf *b,
 958                                     size_t new_size)
 959 {
 960         void *n;
 961
 962         /* the bios are sized for this many pages, max: */
 963         if (new_size > JOURNAL_ENTRY_SIZE_MAX)
 964                 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
 965
 966         new_size = roundup_pow_of_two(new_size);
 967         n = kvmalloc(new_size, GFP_KERNEL);
 968         if (!n)
 969                 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
 970
 971         kvfree(b->data);
 972         b->data = n;
 973         b->size = new_size;
 974         return 0;
 975 }
 976
 977 static int journal_read_bucket(struct bch_dev *ca,
 978                                struct journal_read_buf *buf,
 979                                struct journal_list *jlist,
 980                                unsigned bucket)
 981 {
 982         struct bch_fs *c = ca->fs;
 983         struct journal_device *ja = &ca->journal;
 984         struct jset *j = NULL;
 985         unsigned sectors, sectors_read = 0;
 986         u64 offset = bucket_to_sector(ca, ja->buckets[bucket]),
 987             end = offset + ca->mi.bucket_size;
 988         bool saw_bad = false, csum_good;
 989         struct printbuf err = PRINTBUF;
 990         int ret = 0;
 991
 992         pr_debug("reading %u", bucket);
 993
 994         while (offset < end) {
 995                 if (!sectors_read) {
 996                         struct bio *bio;
 997                         unsigned nr_bvecs;
 998 reread:
 999                         sectors_read = min_t(unsigned,
1000                                 end - offset, buf->size >> 9);
1001                         nr_bvecs = buf_pages(buf->data, sectors_read << 9);
1002
1003                         bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1004                         bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ);
1005
1006                         bio->bi_iter.bi_sector = offset;
1007                         bch2_bio_map(bio, buf->data, sectors_read << 9);
1008
1009                         ret = submit_bio_wait(bio);
1010                         kfree(bio);
1011
1012                         if (bch2_dev_io_err_on(ret, ca, BCH_MEMBER_ERROR_read,
1013                                                "journal read error: sector %llu",
1014                                                offset) ||
1015                             bch2_meta_read_fault("journal")) {
1016                                 /*
1017                                  * We don't error out of the recovery process
1018                                  * here, since the relevant journal entry may be
1019                                  * found on a different device, and missing or
1020                                  * no journal entries will be handled later
1021                                  */
1022                                 goto out;
1023                         }
1024
1025                         j = buf->data;
1026                 }
1027
1028                 ret = jset_validate_early(c, ca, j, offset,
1029                                     end - offset, sectors_read);
1030                 switch (ret) {
1031                 case 0:
1032                         sectors = vstruct_sectors(j, c->block_bits);
1033                         break;
1034                 case JOURNAL_ENTRY_REREAD:
1035                         if (vstruct_bytes(j) > buf->size) {
1036                                 ret = journal_read_buf_realloc(buf,
1037                                                         vstruct_bytes(j));
1038                                 if (ret)
1039                                         goto err;
1040                         }
1041                         goto reread;
1042                 case JOURNAL_ENTRY_NONE:
1043                         if (!saw_bad)
1044                                 goto out;
1045                         /*
1046                          * On checksum error we don't really trust the size
1047                          * field of the journal entry we read, so try reading
1048                          * again at next block boundary:
1049                          */
1050                         sectors = block_sectors(c);
1051                         goto next_block;
1052                 default:
1053                         goto err;
1054                 }
1055
1056                 /*
1057                  * This happens sometimes if we don't have discards on -
1058                  * when we've partially overwritten a bucket with new
1059                  * journal entries. We don't need the rest of the
1060                  * bucket:
1061                  */
1062                 if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket])
1063                         goto out;
1064
1065                 ja->bucket_seq[bucket] = le64_to_cpu(j->seq);
1066
1067                 enum bch_csum_type csum_type = JSET_CSUM_TYPE(j);
1068                 struct bch_csum csum;
1069                 csum_good = jset_csum_good(c, j, &csum);
1070
1071                 if (bch2_dev_io_err_on(!csum_good, ca, BCH_MEMBER_ERROR_checksum,
1072                                        "%s",
1073                                        (printbuf_reset(&err),
1074                                         prt_str(&err, "journal "),
1075                                         bch2_csum_err_msg(&err, csum_type, j->csum, csum),
1076                                         err.buf)))
1077                         saw_bad = true;
1078
1079                 ret = bch2_encrypt(c, JSET_CSUM_TYPE(j), journal_nonce(j),
1080                              j->encrypted_start,
1081                              vstruct_end(j) - (void *) j->encrypted_start);
1082                 bch2_fs_fatal_err_on(ret, c,
1083                                 "error decrypting journal entry: %s",
1084                                 bch2_err_str(ret));
1085
1086                 mutex_lock(&jlist->lock);
1087                 ret = journal_entry_add(c, ca, (struct journal_ptr) {
1088                                         .csum_good      = csum_good,
1089                                         .dev            = ca->dev_idx,
1090                                         .bucket         = bucket,
1091                                         .bucket_offset  = offset -
1092                                                 bucket_to_sector(ca, ja->buckets[bucket]),
1093                                         .sector         = offset,
1094                                         }, jlist, j);
1095                 mutex_unlock(&jlist->lock);
1096
1097                 switch (ret) {
1098                 case JOURNAL_ENTRY_ADD_OK:
1099                         break;
1100                 case JOURNAL_ENTRY_ADD_OUT_OF_RANGE:
1101                         break;
1102                 default:
1103                         goto err;
1104                 }
1105 next_block:
1106                 pr_debug("next");
1107                 offset          += sectors;
1108                 sectors_read    -= sectors;
1109                 j = ((void *) j) + (sectors << 9);
1110         }
1111
1112 out:
1113         ret = 0;
1114 err:
1115         printbuf_exit(&err);
1116         return ret;
1117 }
1118
1119 static CLOSURE_CALLBACK(bch2_journal_read_device)
1120 {
1121         closure_type(ja, struct journal_device, read);
1122         struct bch_dev *ca = container_of(ja, struct bch_dev, journal);
1123         struct bch_fs *c = ca->fs;
1124         struct journal_list *jlist =
1125                 container_of(cl->parent, struct journal_list, cl);
1126         struct journal_replay *r, **_r;
1127         struct genradix_iter iter;
1128         struct journal_read_buf buf = { NULL, 0 };
1129         unsigned i;
1130         int ret = 0;
1131
1132         if (!ja->nr)
1133                 goto out;
1134
1135         ret = journal_read_buf_realloc(&buf, PAGE_SIZE);
1136         if (ret)
1137                 goto err;
1138
1139         pr_debug("%u journal buckets", ja->nr);
1140
1141         for (i = 0; i < ja->nr; i++) {
1142                 ret = journal_read_bucket(ca, &buf, jlist, i);
1143                 if (ret)
1144                         goto err;
1145         }
1146
1147         ja->sectors_free = ca->mi.bucket_size;
1148
1149         mutex_lock(&jlist->lock);
1150         genradix_for_each_reverse(&c->journal_entries, iter, _r) {
1151                 r = *_r;
1152
1153                 if (!r)
1154                         continue;
1155
1156                 darray_for_each(r->ptrs, i)
1157                         if (i->dev == ca->dev_idx) {
1158                                 unsigned wrote = bucket_remainder(ca, i->sector) +
1159                                         vstruct_sectors(&r->j, c->block_bits);
1160
1161                                 ja->cur_idx = i->bucket;
1162                                 ja->sectors_free = ca->mi.bucket_size - wrote;
1163                                 goto found;
1164                         }
1165         }
1166 found:
1167         mutex_unlock(&jlist->lock);
1168
1169         if (ja->bucket_seq[ja->cur_idx] &&
1170             ja->sectors_free == ca->mi.bucket_size) {
1171 #if 0
1172                 /*
1173                  * Debug code for ZNS support, where we (probably) want to be
1174                  * correlated where we stopped in the journal to the zone write
1175                  * points:
1176                  */
1177                 bch_err(c, "ja->sectors_free == ca->mi.bucket_size");
1178                 bch_err(c, "cur_idx %u/%u", ja->cur_idx, ja->nr);
1179                 for (i = 0; i < 3; i++) {
1180                         unsigned idx = (ja->cur_idx + ja->nr - 1 + i) % ja->nr;
1181
1182                         bch_err(c, "bucket_seq[%u] = %llu", idx, ja->bucket_seq[idx]);
1183                 }
1184 #endif
1185                 ja->sectors_free = 0;
1186         }
1187
1188         /*
1189          * Set dirty_idx to indicate the entire journal is full and needs to be
1190          * reclaimed - journal reclaim will immediately reclaim whatever isn't
1191          * pinned when it first runs:
1192          */
1193         ja->discard_idx = ja->dirty_idx_ondisk =
1194                 ja->dirty_idx = (ja->cur_idx + 1) % ja->nr;
1195 out:
1196         bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret);
1197         kvfree(buf.data);
1198         percpu_ref_put(&ca->io_ref);
1199         closure_return(cl);
1200         return;
1201 err:
1202         mutex_lock(&jlist->lock);
1203         jlist->ret = ret;
1204         mutex_unlock(&jlist->lock);
1205         goto out;
1206 }
1207
1208 int bch2_journal_read(struct bch_fs *c,
1209                       u64 *last_seq,
1210                       u64 *blacklist_seq,
1211                       u64 *start_seq)
1212 {
1213         struct journal_list jlist;
1214         struct journal_replay *i, **_i, *prev = NULL;
1215         struct genradix_iter radix_iter;
1216         struct printbuf buf = PRINTBUF;
1217         bool degraded = false, last_write_torn = false;
1218         u64 seq;
1219         int ret = 0;
1220
1221         closure_init_stack(&jlist.cl);
1222         mutex_init(&jlist.lock);
1223         jlist.last_seq = 0;
1224         jlist.ret = 0;
1225
1226         for_each_member_device(c, ca) {
1227                 if (!c->opts.fsck &&
1228                     !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal)))
1229                         continue;
1230
1231                 if ((ca->mi.state == BCH_MEMBER_STATE_rw ||
1232                      ca->mi.state == BCH_MEMBER_STATE_ro) &&
1233                     percpu_ref_tryget(&ca->io_ref))
1234                         closure_call(&ca->journal.read,
1235                                      bch2_journal_read_device,
1236                                      system_unbound_wq,
1237                                      &jlist.cl);
1238                 else
1239                         degraded = true;
1240         }
1241
1242         closure_sync(&jlist.cl);
1243
1244         if (jlist.ret)
1245                 return jlist.ret;
1246
1247         *last_seq       = 0;
1248         *start_seq      = 0;
1249         *blacklist_seq  = 0;
1250
1251         /*
1252          * Find most recent flush entry, and ignore newer non flush entries -
1253          * those entries will be blacklisted:
1254          */
1255         genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) {
1256                 enum bkey_invalid_flags flags = BKEY_INVALID_JOURNAL;
1257
1258                 i = *_i;
1259
1260                 if (!i || i->ignore)
1261                         continue;
1262
1263                 if (!*start_seq)
1264                         *blacklist_seq = *start_seq = le64_to_cpu(i->j.seq) + 1;
1265
1266                 if (JSET_NO_FLUSH(&i->j)) {
1267                         i->ignore = true;
1268                         continue;
1269                 }
1270
1271                 if (!last_write_torn && !i->csum_good) {
1272                         last_write_torn = true;
1273                         i->ignore = true;
1274                         continue;
1275                 }
1276
1277                 if (journal_entry_err_on(le64_to_cpu(i->j.last_seq) > le64_to_cpu(i->j.seq),
1278                                          c, le32_to_cpu(i->j.version), &i->j, NULL,
1279                                          jset_last_seq_newer_than_seq,
1280                                          "invalid journal entry: last_seq > seq (%llu > %llu)",
1281                                          le64_to_cpu(i->j.last_seq),
1282                                          le64_to_cpu(i->j.seq)))
1283                         i->j.last_seq = i->j.seq;
1284
1285                 *last_seq       = le64_to_cpu(i->j.last_seq);
1286                 *blacklist_seq  = le64_to_cpu(i->j.seq) + 1;
1287                 break;
1288         }
1289
1290         if (!*start_seq) {
1291                 bch_info(c, "journal read done, but no entries found");
1292                 return 0;
1293         }
1294
1295         if (!*last_seq) {
1296                 fsck_err(c, dirty_but_no_journal_entries_post_drop_nonflushes,
1297                          "journal read done, but no entries found after dropping non-flushes");
1298                 return 0;
1299         }
1300
1301         bch_info(c, "journal read done, replaying entries %llu-%llu",
1302                  *last_seq, *blacklist_seq - 1);
1303
1304         if (*start_seq != *blacklist_seq)
1305                 bch_info(c, "dropped unflushed entries %llu-%llu",
1306                          *blacklist_seq, *start_seq - 1);
1307
1308         /* Drop blacklisted entries and entries older than last_seq: */
1309         genradix_for_each(&c->journal_entries, radix_iter, _i) {
1310                 i = *_i;
1311
1312                 if (!i || i->ignore)
1313                         continue;
1314
1315                 seq = le64_to_cpu(i->j.seq);
1316                 if (seq < *last_seq) {
1317                         journal_replay_free(c, i);
1318                         continue;
1319                 }
1320
1321                 if (bch2_journal_seq_is_blacklisted(c, seq, true)) {
1322                         fsck_err_on(!JSET_NO_FLUSH(&i->j), c,
1323                                     jset_seq_blacklisted,
1324                                     "found blacklisted journal entry %llu", seq);
1325                         i->ignore = true;
1326                 }
1327         }
1328
1329         /* Check for missing entries: */
1330         seq = *last_seq;
1331         genradix_for_each(&c->journal_entries, radix_iter, _i) {
1332                 i = *_i;
1333
1334                 if (!i || i->ignore)
1335                         continue;
1336
1337                 BUG_ON(seq > le64_to_cpu(i->j.seq));
1338
1339                 while (seq < le64_to_cpu(i->j.seq)) {
1340                         u64 missing_start, missing_end;
1341                         struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
1342
1343                         while (seq < le64_to_cpu(i->j.seq) &&
1344                                bch2_journal_seq_is_blacklisted(c, seq, false))
1345                                 seq++;
1346
1347                         if (seq == le64_to_cpu(i->j.seq))
1348                                 break;
1349
1350                         missing_start = seq;
1351
1352                         while (seq < le64_to_cpu(i->j.seq) &&
1353                                !bch2_journal_seq_is_blacklisted(c, seq, false))
1354                                 seq++;
1355
1356                         if (prev) {
1357                                 bch2_journal_ptrs_to_text(&buf1, c, prev);
1358                                 prt_printf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits));
1359                         } else
1360                                 prt_printf(&buf1, "(none)");
1361                         bch2_journal_ptrs_to_text(&buf2, c, i);
1362
1363                         missing_end = seq - 1;
1364                         fsck_err(c, journal_entries_missing,
1365                                  "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
1366                                  "  prev at %s\n"
1367                                  "  next at %s",
1368                                  missing_start, missing_end,
1369                                  *last_seq, *blacklist_seq - 1,
1370                                  buf1.buf, buf2.buf);
1371
1372                         printbuf_exit(&buf1);
1373                         printbuf_exit(&buf2);
1374                 }
1375
1376                 prev = i;
1377                 seq++;
1378         }
1379
1380         genradix_for_each(&c->journal_entries, radix_iter, _i) {
1381                 struct bch_replicas_padded replicas = {
1382                         .e.data_type = BCH_DATA_journal,
1383                         .e.nr_required = 1,
1384                 };
1385
1386                 i = *_i;
1387                 if (!i || i->ignore)
1388                         continue;
1389
1390                 darray_for_each(i->ptrs, ptr) {
1391                         struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
1392
1393                         if (!ptr->csum_good)
1394                                 bch_err_dev_offset(ca, ptr->sector,
1395                                                    "invalid journal checksum, seq %llu%s",
1396                                                    le64_to_cpu(i->j.seq),
1397                                                    i->csum_good ? " (had good copy on another device)" : "");
1398                 }
1399
1400                 ret = jset_validate(c,
1401                                     bch_dev_bkey_exists(c, i->ptrs.data[0].dev),
1402                                     &i->j,
1403                                     i->ptrs.data[0].sector,
1404                                     READ);
1405                 if (ret)
1406                         goto err;
1407
1408                 darray_for_each(i->ptrs, ptr)
1409                         replicas.e.devs[replicas.e.nr_devs++] = ptr->dev;
1410
1411                 bch2_replicas_entry_sort(&replicas.e);
1412
1413                 printbuf_reset(&buf);
1414                 bch2_replicas_entry_to_text(&buf, &replicas.e);
1415
1416                 if (!degraded &&
1417                     !bch2_replicas_marked(c, &replicas.e) &&
1418                     (le64_to_cpu(i->j.seq) == *last_seq ||
1419                      fsck_err(c, journal_entry_replicas_not_marked,
1420                               "superblock not marked as containing replicas for journal entry %llu\n  %s",
1421                               le64_to_cpu(i->j.seq), buf.buf))) {
1422                         ret = bch2_mark_replicas(c, &replicas.e);
1423                         if (ret)
1424                                 goto err;
1425                 }
1426         }
1427 err:
1428 fsck_err:
1429         printbuf_exit(&buf);
1430         return ret;
1431 }
1432
1433 /* journal write: */
1434
1435 static void __journal_write_alloc(struct journal *j,
1436                                   struct journal_buf *w,
1437                                   struct dev_alloc_list *devs_sorted,
1438                                   unsigned sectors,
1439                                   unsigned *replicas,
1440                                   unsigned replicas_want)
1441 {
1442         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1443         struct journal_device *ja;
1444         struct bch_dev *ca;
1445         unsigned i;
1446
1447         if (*replicas >= replicas_want)
1448                 return;
1449
1450         for (i = 0; i < devs_sorted->nr; i++) {
1451                 ca = rcu_dereference(c->devs[devs_sorted->devs[i]]);
1452                 if (!ca)
1453                         continue;
1454
1455                 ja = &ca->journal;
1456
1457                 /*
1458                  * Check that we can use this device, and aren't already using
1459                  * it:
1460                  */
1461                 if (!ca->mi.durability ||
1462                     ca->mi.state != BCH_MEMBER_STATE_rw ||
1463                     !ja->nr ||
1464                     bch2_bkey_has_device_c(bkey_i_to_s_c(&w->key), ca->dev_idx) ||
1465                     sectors > ja->sectors_free)
1466                         continue;
1467
1468                 bch2_dev_stripe_increment(ca, &j->wp.stripe);
1469
1470                 bch2_bkey_append_ptr(&w->key,
1471                         (struct bch_extent_ptr) {
1472                                   .offset = bucket_to_sector(ca,
1473                                         ja->buckets[ja->cur_idx]) +
1474                                         ca->mi.bucket_size -
1475                                         ja->sectors_free,
1476                                   .dev = ca->dev_idx,
1477                 });
1478
1479                 ja->sectors_free -= sectors;
1480                 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1481
1482                 *replicas += ca->mi.durability;
1483
1484                 if (*replicas >= replicas_want)
1485                         break;
1486         }
1487 }
1488
1489 /**
1490  * journal_write_alloc - decide where to write next journal entry
1491  *
1492  * @j:          journal object
1493  * @w:          journal buf (entry to be written)
1494  *
1495  * Returns: 0 on success, or -EROFS on failure
1496  */
1497 static int journal_write_alloc(struct journal *j, struct journal_buf *w)
1498 {
1499         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1500         struct bch_devs_mask devs;
1501         struct journal_device *ja;
1502         struct bch_dev *ca;
1503         struct dev_alloc_list devs_sorted;
1504         unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1505         unsigned target = c->opts.metadata_target ?:
1506                 c->opts.foreground_target;
1507         unsigned i, replicas = 0, replicas_want =
1508                 READ_ONCE(c->opts.metadata_replicas);
1509         unsigned replicas_need = min(replicas_want,
1510                                      READ_ONCE(c->opts.metadata_replicas_required));
1511
1512         rcu_read_lock();
1513 retry:
1514         devs = target_rw_devs(c, BCH_DATA_journal, target);
1515
1516         devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs);
1517
1518         __journal_write_alloc(j, w, &devs_sorted,
1519                               sectors, &replicas, replicas_want);
1520
1521         if (replicas >= replicas_want)
1522                 goto done;
1523
1524         for (i = 0; i < devs_sorted.nr; i++) {
1525                 ca = rcu_dereference(c->devs[devs_sorted.devs[i]]);
1526                 if (!ca)
1527                         continue;
1528
1529                 ja = &ca->journal;
1530
1531                 if (sectors > ja->sectors_free &&
1532                     sectors <= ca->mi.bucket_size &&
1533                     bch2_journal_dev_buckets_available(j, ja,
1534                                         journal_space_discarded)) {
1535                         ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
1536                         ja->sectors_free = ca->mi.bucket_size;
1537
1538                         /*
1539                          * ja->bucket_seq[ja->cur_idx] must always have
1540                          * something sensible:
1541                          */
1542                         ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1543                 }
1544         }
1545
1546         __journal_write_alloc(j, w, &devs_sorted,
1547                               sectors, &replicas, replicas_want);
1548
1549         if (replicas < replicas_want && target) {
1550                 /* Retry from all devices: */
1551                 target = 0;
1552                 goto retry;
1553         }
1554 done:
1555         rcu_read_unlock();
1556
1557         BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX);
1558
1559         return replicas >= replicas_need ? 0 : -EROFS;
1560 }
1561
1562 static void journal_buf_realloc(struct journal *j, struct journal_buf *buf)
1563 {
1564         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1565
1566         /* we aren't holding j->lock: */
1567         unsigned new_size = READ_ONCE(j->buf_size_want);
1568         void *new_buf;
1569
1570         if (buf->buf_size >= new_size)
1571                 return;
1572
1573         size_t btree_write_buffer_size = new_size / 64;
1574
1575         if (bch2_btree_write_buffer_resize(c, btree_write_buffer_size))
1576                 return;
1577
1578         new_buf = kvmalloc(new_size, GFP_NOFS|__GFP_NOWARN);
1579         if (!new_buf)
1580                 return;
1581
1582         memcpy(new_buf, buf->data, buf->buf_size);
1583
1584         spin_lock(&j->lock);
1585         swap(buf->data,         new_buf);
1586         swap(buf->buf_size,     new_size);
1587         spin_unlock(&j->lock);
1588
1589         kvfree(new_buf);
1590 }
1591
1592 static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j)
1593 {
1594         return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK);
1595 }
1596
1597 static CLOSURE_CALLBACK(journal_write_done)
1598 {
1599         closure_type(w, struct journal_buf, io);
1600         struct journal *j = container_of(w, struct journal, buf[w->idx]);
1601         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1602         struct bch_replicas_padded replicas;
1603         union journal_res_state old, new;
1604         u64 v, seq = le64_to_cpu(w->data->seq);
1605         int err = 0;
1606
1607         time_stats_update(!JSET_NO_FLUSH(w->data)
1608                           ? j->flush_write_time
1609                           : j->noflush_write_time, j->write_start_time);
1610
1611         if (!w->devs_written.nr) {
1612                 bch_err(c, "unable to write journal to sufficient devices");
1613                 err = -EIO;
1614         } else {
1615                 bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
1616                                          w->devs_written);
1617                 if (bch2_mark_replicas(c, &replicas.e))
1618                         err = -EIO;
1619         }
1620
1621         if (err)
1622                 bch2_fatal_error(c);
1623
1624         closure_debug_destroy(cl);
1625
1626         spin_lock(&j->lock);
1627         if (seq >= j->pin.front)
1628                 journal_seq_pin(j, seq)->devs = w->devs_written;
1629         if (err && (!j->err_seq || seq < j->err_seq))
1630                 j->err_seq      = seq;
1631         w->write_done = true;
1632
1633         bool completed = false;
1634
1635         for (seq = journal_last_unwritten_seq(j);
1636              seq <= journal_cur_seq(j);
1637              seq++) {
1638                 w = j->buf + (seq & JOURNAL_BUF_MASK);
1639                 if (!w->write_done)
1640                         break;
1641
1642                 if (!j->err_seq && !JSET_NO_FLUSH(w->data)) {
1643                         j->flushed_seq_ondisk = seq;
1644                         j->last_seq_ondisk = w->last_seq;
1645
1646                         bch2_do_discards(c);
1647                         closure_wake_up(&c->freelist_wait);
1648                         bch2_reset_alloc_cursors(c);
1649                 }
1650
1651                 j->seq_ondisk = seq;
1652
1653                 /*
1654                  * Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard
1655                  * more buckets:
1656                  *
1657                  * Must come before signaling write completion, for
1658                  * bch2_fs_journal_stop():
1659                  */
1660                 if (j->watermark != BCH_WATERMARK_stripe)
1661                         journal_reclaim_kick(&c->journal);
1662
1663                 v = atomic64_read(&j->reservations.counter);
1664                 do {
1665                         old.v = new.v = v;
1666                         BUG_ON(journal_state_count(new, new.unwritten_idx));
1667                         BUG_ON(new.unwritten_idx != (seq & JOURNAL_BUF_MASK));
1668
1669                         new.unwritten_idx++;
1670                 } while ((v = atomic64_cmpxchg(&j->reservations.counter, old.v, new.v)) != old.v);
1671
1672                 completed = true;
1673         }
1674
1675         if (completed) {
1676                 bch2_journal_reclaim_fast(j);
1677                 bch2_journal_space_available(j);
1678
1679                 track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight], false);
1680
1681                 closure_wake_up(&w->wait);
1682                 journal_wake(j);
1683         }
1684
1685         if (journal_last_unwritten_seq(j) == journal_cur_seq(j) &&
1686                    new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) {
1687                 struct journal_buf *buf = journal_cur_buf(j);
1688                 long delta = buf->expires - jiffies;
1689
1690                 /*
1691                  * We don't close a journal entry to write it while there's
1692                  * previous entries still in flight - the current journal entry
1693                  * might want to be written now:
1694                  */
1695                 mod_delayed_work(j->wq, &j->write_work, max(0L, delta));
1696         }
1697
1698         spin_unlock(&j->lock);
1699 }
1700
1701 static void journal_write_endio(struct bio *bio)
1702 {
1703         struct journal_bio *jbio = container_of(bio, struct journal_bio, bio);
1704         struct bch_dev *ca = jbio->ca;
1705         struct journal *j = &ca->fs->journal;
1706         struct journal_buf *w = j->buf + jbio->buf_idx;
1707
1708         if (bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_write,
1709                                "error writing journal entry %llu: %s",
1710                                le64_to_cpu(w->data->seq),
1711                                bch2_blk_status_to_str(bio->bi_status)) ||
1712             bch2_meta_write_fault("journal")) {
1713                 unsigned long flags;
1714
1715                 spin_lock_irqsave(&j->err_lock, flags);
1716                 bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx);
1717                 spin_unlock_irqrestore(&j->err_lock, flags);
1718         }
1719
1720         closure_put(&w->io);
1721         percpu_ref_put(&ca->io_ref);
1722 }
1723
1724 static CLOSURE_CALLBACK(do_journal_write)
1725 {
1726         closure_type(w, struct journal_buf, io);
1727         struct journal *j = container_of(w, struct journal, buf[w->idx]);
1728         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1729         unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1730
1731         extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) {
1732                 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
1733                 struct journal_device *ja = &ca->journal;
1734
1735                 if (!percpu_ref_tryget(&ca->io_ref)) {
1736                         /* XXX: fix this */
1737                         bch_err(c, "missing device for journal write\n");
1738                         continue;
1739                 }
1740
1741                 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal],
1742                              sectors);
1743
1744                 struct bio *bio = &ja->bio[w->idx]->bio;
1745                 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
1746                 bio->bi_iter.bi_sector  = ptr->offset;
1747                 bio->bi_end_io          = journal_write_endio;
1748                 bio->bi_private         = ca;
1749
1750                 BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector);
1751                 ca->prev_journal_sector = bio->bi_iter.bi_sector;
1752
1753                 if (!JSET_NO_FLUSH(w->data))
1754                         bio->bi_opf    |= REQ_FUA;
1755                 if (!JSET_NO_FLUSH(w->data) && !w->separate_flush)
1756                         bio->bi_opf    |= REQ_PREFLUSH;
1757
1758                 bch2_bio_map(bio, w->data, sectors << 9);
1759
1760                 trace_and_count(c, journal_write, bio);
1761                 closure_bio_submit(bio, cl);
1762
1763                 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1764         }
1765
1766         continue_at(cl, journal_write_done, j->wq);
1767 }
1768
1769 static int bch2_journal_write_prep(struct journal *j, struct journal_buf *w)
1770 {
1771         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1772         struct jset_entry *start, *end;
1773         struct jset *jset = w->data;
1774         struct journal_keys_to_wb wb = { NULL };
1775         unsigned sectors, bytes, u64s;
1776         unsigned long btree_roots_have = 0;
1777         bool validate_before_checksum = false;
1778         u64 seq = le64_to_cpu(jset->seq);
1779         int ret;
1780
1781         /*
1782          * Simple compaction, dropping empty jset_entries (from journal
1783          * reservations that weren't fully used) and merging jset_entries that
1784          * can be.
1785          *
1786          * If we wanted to be really fancy here, we could sort all the keys in
1787          * the jset and drop keys that were overwritten - probably not worth it:
1788          */
1789         vstruct_for_each(jset, i) {
1790                 unsigned u64s = le16_to_cpu(i->u64s);
1791
1792                 /* Empty entry: */
1793                 if (!u64s)
1794                         continue;
1795
1796                 /*
1797                  * New btree roots are set by journalling them; when the journal
1798                  * entry gets written we have to propagate them to
1799                  * c->btree_roots
1800                  *
1801                  * But, every journal entry we write has to contain all the
1802                  * btree roots (at least for now); so after we copy btree roots
1803                  * to c->btree_roots we have to get any missing btree roots and
1804                  * add them to this journal entry:
1805                  */
1806                 switch (i->type) {
1807                 case BCH_JSET_ENTRY_btree_root:
1808                         bch2_journal_entry_to_btree_root(c, i);
1809                         __set_bit(i->btree_id, &btree_roots_have);
1810                         break;
1811                 case BCH_JSET_ENTRY_write_buffer_keys:
1812                         EBUG_ON(!w->need_flush_to_write_buffer);
1813
1814                         if (!wb.wb)
1815                                 bch2_journal_keys_to_write_buffer_start(c, &wb, seq);
1816
1817                         struct bkey_i *k;
1818                         jset_entry_for_each_key(i, k) {
1819                                 ret = bch2_journal_key_to_wb(c, &wb, i->btree_id, k);
1820                                 if (ret) {
1821                                         bch2_fs_fatal_error(c, "-ENOMEM flushing journal keys to btree write buffer");
1822                                         bch2_journal_keys_to_write_buffer_end(c, &wb);
1823                                         return ret;
1824                                 }
1825                         }
1826                         i->type = BCH_JSET_ENTRY_btree_keys;
1827                         break;
1828                 }
1829         }
1830
1831         if (wb.wb)
1832                 bch2_journal_keys_to_write_buffer_end(c, &wb);
1833         w->need_flush_to_write_buffer = false;
1834
1835         start = end = vstruct_last(jset);
1836
1837         end     = bch2_btree_roots_to_journal_entries(c, end, btree_roots_have);
1838
1839         struct jset_entry_datetime *d =
1840                 container_of(jset_entry_init(&end, sizeof(*d)), struct jset_entry_datetime, entry);
1841         d->entry.type   = BCH_JSET_ENTRY_datetime;
1842         d->seconds      = cpu_to_le64(ktime_get_real_seconds());
1843
1844         bch2_journal_super_entries_add_common(c, &end, seq);
1845         u64s    = (u64 *) end - (u64 *) start;
1846         BUG_ON(u64s > j->entry_u64s_reserved);
1847
1848         le32_add_cpu(&jset->u64s, u64s);
1849
1850         sectors = vstruct_sectors(jset, c->block_bits);
1851         bytes   = vstruct_bytes(jset);
1852
1853         if (sectors > w->sectors) {
1854                 bch2_fs_fatal_error(c, "aieeee! journal write overran available space, %zu > %u (extra %u reserved %u/%u)",
1855                                     vstruct_bytes(jset), w->sectors << 9,
1856                                     u64s, w->u64s_reserved, j->entry_u64s_reserved);
1857                 return -EINVAL;
1858         }
1859
1860         jset->magic             = cpu_to_le64(jset_magic(c));
1861         jset->version           = cpu_to_le32(c->sb.version);
1862
1863         SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN);
1864         SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c));
1865
1866         if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset))
1867                 j->last_empty_seq = seq;
1868
1869         if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset)))
1870                 validate_before_checksum = true;
1871
1872         if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current)
1873                 validate_before_checksum = true;
1874
1875         if (validate_before_checksum &&
1876             (ret = jset_validate(c, NULL, jset, 0, WRITE)))
1877                 return ret;
1878
1879         ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
1880                     jset->encrypted_start,
1881                     vstruct_end(jset) - (void *) jset->encrypted_start);
1882         if (bch2_fs_fatal_err_on(ret, c,
1883                         "error decrypting journal entry: %i", ret))
1884                 return ret;
1885
1886         jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset),
1887                                   journal_nonce(jset), jset);
1888
1889         if (!validate_before_checksum &&
1890             (ret = jset_validate(c, NULL, jset, 0, WRITE)))
1891                 return ret;
1892
1893         memset((void *) jset + bytes, 0, (sectors << 9) - bytes);
1894         return 0;
1895 }
1896
1897 static int bch2_journal_write_pick_flush(struct journal *j, struct journal_buf *w)
1898 {
1899         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1900         int error = bch2_journal_error(j);
1901
1902         /*
1903          * If the journal is in an error state - we did an emergency shutdown -
1904          * we prefer to continue doing journal writes. We just mark them as
1905          * noflush so they'll never be used, but they'll still be visible by the
1906          * list_journal tool - this helps in debugging.
1907          *
1908          * There's a caveat: the first journal write after marking the
1909          * superblock dirty must always be a flush write, because on startup
1910          * from a clean shutdown we didn't necessarily read the journal and the
1911          * new journal write might overwrite whatever was in the journal
1912          * previously - we can't leave the journal without any flush writes in
1913          * it.
1914          *
1915          * So if we're in an error state, and we're still starting up, we don't
1916          * write anything at all.
1917          */
1918         if (error && test_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags))
1919                 return -EIO;
1920
1921         if (error ||
1922             w->noflush ||
1923             (!w->must_flush &&
1924              (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) &&
1925              test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags))) {
1926                 w->noflush = true;
1927                 SET_JSET_NO_FLUSH(w->data, true);
1928                 w->data->last_seq       = 0;
1929                 w->last_seq             = 0;
1930
1931                 j->nr_noflush_writes++;
1932         } else {
1933                 j->last_flush_write = jiffies;
1934                 j->nr_flush_writes++;
1935                 clear_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags);
1936         }
1937
1938         return 0;
1939 }
1940
1941 CLOSURE_CALLBACK(bch2_journal_write)
1942 {
1943         closure_type(w, struct journal_buf, io);
1944         struct journal *j = container_of(w, struct journal, buf[w->idx]);
1945         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1946         struct bch_replicas_padded replicas;
1947         struct printbuf journal_debug_buf = PRINTBUF;
1948         unsigned nr_rw_members = 0;
1949         int ret;
1950
1951         BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
1952         BUG_ON(w->write_allocated);
1953
1954         j->write_start_time = local_clock();
1955
1956         spin_lock(&j->lock);
1957         ret = bch2_journal_write_pick_flush(j, w);
1958         spin_unlock(&j->lock);
1959         if (ret)
1960                 goto err;
1961
1962         mutex_lock(&j->buf_lock);
1963         journal_buf_realloc(j, w);
1964
1965         ret = bch2_journal_write_prep(j, w);
1966         mutex_unlock(&j->buf_lock);
1967         if (ret)
1968                 goto err;
1969
1970         j->entry_bytes_written += vstruct_bytes(w->data);
1971
1972         while (1) {
1973                 spin_lock(&j->lock);
1974                 ret = journal_write_alloc(j, w);
1975                 if (!ret || !j->can_discard)
1976                         break;
1977
1978                 spin_unlock(&j->lock);
1979                 bch2_journal_do_discards(j);
1980         }
1981
1982         if (ret) {
1983                 __bch2_journal_debug_to_text(&journal_debug_buf, j);
1984                 spin_unlock(&j->lock);
1985                 bch_err(c, "Unable to allocate journal write:\n%s",
1986                         journal_debug_buf.buf);
1987                 printbuf_exit(&journal_debug_buf);
1988                 goto err;
1989         }
1990
1991         /*
1992          * write is allocated, no longer need to account for it in
1993          * bch2_journal_space_available():
1994          */
1995         w->sectors = 0;
1996         w->write_allocated = true;
1997
1998         /*
1999          * journal entry has been compacted and allocated, recalculate space
2000          * available:
2001          */
2002         bch2_journal_space_available(j);
2003         bch2_journal_do_writes(j);
2004         spin_unlock(&j->lock);
2005
2006         w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key));
2007
2008         if (c->opts.nochanges)
2009                 goto no_io;
2010
2011         for_each_rw_member(c, ca)
2012                 nr_rw_members++;
2013
2014         if (nr_rw_members > 1)
2015                 w->separate_flush = true;
2016
2017         /*
2018          * Mark journal replicas before we submit the write to guarantee
2019          * recovery will find the journal entries after a crash.
2020          */
2021         bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
2022                                  w->devs_written);
2023         ret = bch2_mark_replicas(c, &replicas.e);
2024         if (ret)
2025                 goto err;
2026
2027         if (!JSET_NO_FLUSH(w->data))
2028                 closure_wait_event(&j->async_wait, j->seq_ondisk + 1 == le64_to_cpu(w->data->seq));
2029
2030         if (!JSET_NO_FLUSH(w->data) && w->separate_flush) {
2031                 for_each_rw_member(c, ca) {
2032                         percpu_ref_get(&ca->io_ref);
2033
2034                         struct journal_device *ja = &ca->journal;
2035                         struct bio *bio = &ja->bio[w->idx]->bio;
2036                         bio_reset(bio, ca->disk_sb.bdev,
2037                                   REQ_OP_WRITE|REQ_SYNC|REQ_META|REQ_PREFLUSH);
2038                         bio->bi_end_io          = journal_write_endio;
2039                         bio->bi_private         = ca;
2040                         closure_bio_submit(bio, cl);
2041                 }
2042         }
2043
2044         continue_at(cl, do_journal_write, j->wq);
2045         return;
2046 no_io:
2047         continue_at(cl, journal_write_done, j->wq);
2048         return;
2049 err:
2050         bch2_fatal_error(c);
2051         continue_at(cl, journal_write_done, j->wq);
2052 }