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 "buckets.h"
   8 #include "checksum.h"
   9 #include "disk_groups.h"
  10 #include "error.h"
  11 #include "journal.h"
  12 #include "journal_io.h"
  13 #include "journal_reclaim.h"
  14 #include "journal_seq_blacklist.h"
  15 #include "replicas.h"
  16 #include "sb-clean.h"
  17 #include "trace.h"
  18
  19 static struct nonce journal_nonce(const struct jset *jset)
  20 {
  21         return (struct nonce) {{
  22                 [0] = 0,
  23                 [1] = ((__le32 *) &jset->seq)[0],
  24                 [2] = ((__le32 *) &jset->seq)[1],
  25                 [3] = BCH_NONCE_JOURNAL,
  26         }};
  27 }
  28
  29 static bool jset_csum_good(struct bch_fs *c, struct jset *j)
  30 {
  31         return bch2_checksum_type_valid(c, JSET_CSUM_TYPE(j)) &&
  32                 !bch2_crc_cmp(j->csum,
  33                               csum_vstruct(c, JSET_CSUM_TYPE(j), journal_nonce(j), j));
  34 }
  35
  36 static inline u32 journal_entry_radix_idx(struct bch_fs *c, u64 seq)
  37 {
  38         return (seq - c->journal_entries_base_seq) & (~0U >> 1);
  39 }
  40
  41 static void __journal_replay_free(struct bch_fs *c,
  42                                   struct journal_replay *i)
  43 {
  44         struct journal_replay **p =
  45                 genradix_ptr(&c->journal_entries,
  46                              journal_entry_radix_idx(c, le64_to_cpu(i->j.seq)));
  47
  48         BUG_ON(*p != i);
  49         *p = NULL;
  50         kvpfree(i, offsetof(struct journal_replay, j) +
  51                 vstruct_bytes(&i->j));
  52 }
  53
  54 static void journal_replay_free(struct bch_fs *c, struct journal_replay *i)
  55 {
  56         i->ignore = true;
  57
  58         if (!c->opts.read_entire_journal)
  59                 __journal_replay_free(c, i);
  60 }
  61
  62 struct journal_list {
  63         struct closure          cl;
  64         u64                     last_seq;
  65         struct mutex            lock;
  66         int                     ret;
  67 };
  68
  69 #define JOURNAL_ENTRY_ADD_OK            0
  70 #define JOURNAL_ENTRY_ADD_OUT_OF_RANGE  5
  71
  72 /*
  73  * Given a journal entry we just read, add it to the list of journal entries to
  74  * be replayed:
  75  */
  76 static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca,
  77                              struct journal_ptr entry_ptr,
  78                              struct journal_list *jlist, struct jset *j)
  79 {
  80         struct genradix_iter iter;
  81         struct journal_replay **_i, *i, *dup;
  82         struct journal_ptr *ptr;
  83         size_t bytes = vstruct_bytes(j);
  84         u64 last_seq = !JSET_NO_FLUSH(j) ? le64_to_cpu(j->last_seq) : 0;
  85         int ret = JOURNAL_ENTRY_ADD_OK;
  86
  87         /* Is this entry older than the range we need? */
  88         if (!c->opts.read_entire_journal &&
  89             le64_to_cpu(j->seq) < jlist->last_seq)
  90                 return JOURNAL_ENTRY_ADD_OUT_OF_RANGE;
  91
  92         /*
  93          * genradixes are indexed by a ulong, not a u64, so we can't index them
  94          * by sequence number directly: Assume instead that they will all fall
  95          * within the range of +-2billion of the filrst one we find.
  96          */
  97         if (!c->journal_entries_base_seq)
  98                 c->journal_entries_base_seq = max_t(s64, 1, le64_to_cpu(j->seq) - S32_MAX);
  99
 100         /* Drop entries we don't need anymore */
 101         if (last_seq > jlist->last_seq && !c->opts.read_entire_journal) {
 102                 genradix_for_each_from(&c->journal_entries, iter, _i,
 103                                        journal_entry_radix_idx(c, jlist->last_seq)) {
 104                         i = *_i;
 105
 106                         if (!i || i->ignore)
 107                                 continue;
 108
 109                         if (le64_to_cpu(i->j.seq) >= last_seq)
 110                                 break;
 111                         journal_replay_free(c, i);
 112                 }
 113         }
 114
 115         jlist->last_seq = max(jlist->last_seq, last_seq);
 116
 117         _i = genradix_ptr_alloc(&c->journal_entries,
 118                                 journal_entry_radix_idx(c, le64_to_cpu(j->seq)),
 119                                 GFP_KERNEL);
 120         if (!_i)
 121                 return -BCH_ERR_ENOMEM_journal_entry_add;
 122
 123         /*
 124          * Duplicate journal entries? If so we want the one that didn't have a
 125          * checksum error:
 126          */
 127         dup = *_i;
 128         if (dup) {
 129                 if (bytes == vstruct_bytes(&dup->j) &&
 130                     !memcmp(j, &dup->j, bytes)) {
 131                         i = dup;
 132                         goto found;
 133                 }
 134
 135                 if (!entry_ptr.csum_good) {
 136                         i = dup;
 137                         goto found;
 138                 }
 139
 140                 if (!dup->csum_good)
 141                         goto replace;
 142
 143                 fsck_err(c, journal_entry_replicas_data_mismatch,
 144                          "found duplicate but non identical journal entries (seq %llu)",
 145                          le64_to_cpu(j->seq));
 146                 i = dup;
 147                 goto found;
 148         }
 149 replace:
 150         i = kvpmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL);
 151         if (!i)
 152                 return -BCH_ERR_ENOMEM_journal_entry_add;
 153
 154         i->nr_ptrs      = 0;
 155         i->csum_good    = entry_ptr.csum_good;
 156         i->ignore       = false;
 157         unsafe_memcpy(&i->j, j, bytes, "embedded variable length struct");
 158         i->ptrs[i->nr_ptrs++] = entry_ptr;
 159
 160         if (dup) {
 161                 if (dup->nr_ptrs >= ARRAY_SIZE(dup->ptrs)) {
 162                         bch_err(c, "found too many copies of journal entry %llu",
 163                                 le64_to_cpu(i->j.seq));
 164                         dup->nr_ptrs = ARRAY_SIZE(dup->ptrs) - 1;
 165                 }
 166
 167                 /* The first ptr should represent the jset we kept: */
 168                 memcpy(i->ptrs + i->nr_ptrs,
 169                        dup->ptrs,
 170                        sizeof(dup->ptrs[0]) * dup->nr_ptrs);
 171                 i->nr_ptrs += dup->nr_ptrs;
 172                 __journal_replay_free(c, dup);
 173         }
 174
 175         *_i = i;
 176         return 0;
 177 found:
 178         for (ptr = i->ptrs; ptr < i->ptrs + i->nr_ptrs; ptr++) {
 179                 if (ptr->dev == ca->dev_idx) {
 180                         bch_err(c, "duplicate journal entry %llu on same device",
 181                                 le64_to_cpu(i->j.seq));
 182                         goto out;
 183                 }
 184         }
 185
 186         if (i->nr_ptrs >= ARRAY_SIZE(i->ptrs)) {
 187                 bch_err(c, "found too many copies of journal entry %llu",
 188                         le64_to_cpu(i->j.seq));
 189                 goto out;
 190         }
 191
 192         i->ptrs[i->nr_ptrs++] = entry_ptr;
 193 out:
 194 fsck_err:
 195         return ret;
 196 }
 197
 198 /* this fills in a range with empty jset_entries: */
 199 static void journal_entry_null_range(void *start, void *end)
 200 {
 201         struct jset_entry *entry;
 202
 203         for (entry = start; entry != end; entry = vstruct_next(entry))
 204                 memset(entry, 0, sizeof(*entry));
 205 }
 206
 207 #define JOURNAL_ENTRY_REREAD    5
 208 #define JOURNAL_ENTRY_NONE      6
 209 #define JOURNAL_ENTRY_BAD       7
 210
 211 static void journal_entry_err_msg(struct printbuf *out,
 212                                   u32 version,
 213                                   struct jset *jset,
 214                                   struct jset_entry *entry)
 215 {
 216         prt_str(out, "invalid journal entry, version=");
 217         bch2_version_to_text(out, version);
 218
 219         if (entry) {
 220                 prt_str(out, " type=");
 221                 prt_str(out, bch2_jset_entry_types[entry->type]);
 222         }
 223
 224         if (!jset) {
 225                 prt_printf(out, " in superblock");
 226         } else {
 227
 228                 prt_printf(out, " seq=%llu", le64_to_cpu(jset->seq));
 229
 230                 if (entry)
 231                         prt_printf(out, " offset=%zi/%u",
 232                                    (u64 *) entry - jset->_data,
 233                                    le32_to_cpu(jset->u64s));
 234         }
 235
 236         prt_str(out, ": ");
 237 }
 238
 239 #define journal_entry_err(c, version, jset, entry, _err, msg, ...)      \
 240 ({                                                                      \
 241         struct printbuf _buf = PRINTBUF;                                \
 242                                                                         \
 243         journal_entry_err_msg(&_buf, version, jset, entry);             \
 244         prt_printf(&_buf, msg, ##__VA_ARGS__);                          \
 245                                                                         \
 246         switch (flags & BKEY_INVALID_WRITE) {                           \
 247         case READ:                                                      \
 248                 mustfix_fsck_err(c, _err, "%s", _buf.buf);              \
 249                 break;                                                  \
 250         case WRITE:                                                     \
 251                 bch2_sb_error_count(c, BCH_FSCK_ERR_##_err);            \
 252                 bch_err(c, "corrupt metadata before write: %s\n", _buf.buf);\
 253                 if (bch2_fs_inconsistent(c)) {                          \
 254                         ret = -BCH_ERR_fsck_errors_not_fixed;           \
 255                         goto fsck_err;                                  \
 256                 }                                                       \
 257                 break;                                                  \
 258         }                                                               \
 259                                                                         \
 260         printbuf_exit(&_buf);                                           \
 261         true;                                                           \
 262 })
 263
 264 #define journal_entry_err_on(cond, ...)                                 \
 265         ((cond) ? journal_entry_err(__VA_ARGS__) : false)
 266
 267 #define FSCK_DELETED_KEY        5
 268
 269 static int journal_validate_key(struct bch_fs *c,
 270                                 struct jset *jset,
 271                                 struct jset_entry *entry,
 272                                 unsigned level, enum btree_id btree_id,
 273                                 struct bkey_i *k,
 274                                 unsigned version, int big_endian,
 275                                 enum bkey_invalid_flags flags)
 276 {
 277         int write = flags & BKEY_INVALID_WRITE;
 278         void *next = vstruct_next(entry);
 279         struct printbuf buf = PRINTBUF;
 280         int ret = 0;
 281
 282         if (journal_entry_err_on(!k->k.u64s,
 283                                  c, version, jset, entry,
 284                                  journal_entry_bkey_u64s_0,
 285                                  "k->u64s 0")) {
 286                 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
 287                 journal_entry_null_range(vstruct_next(entry), next);
 288                 return FSCK_DELETED_KEY;
 289         }
 290
 291         if (journal_entry_err_on((void *) bkey_next(k) >
 292                                  (void *) vstruct_next(entry),
 293                                  c, version, jset, entry,
 294                                  journal_entry_bkey_past_end,
 295                                  "extends past end of journal entry")) {
 296                 entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
 297                 journal_entry_null_range(vstruct_next(entry), next);
 298                 return FSCK_DELETED_KEY;
 299         }
 300
 301         if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT,
 302                                  c, version, jset, entry,
 303                                  journal_entry_bkey_bad_format,
 304                                  "bad format %u", k->k.format)) {
 305                 le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
 306                 memmove(k, bkey_next(k), next - (void *) bkey_next(k));
 307                 journal_entry_null_range(vstruct_next(entry), next);
 308                 return FSCK_DELETED_KEY;
 309         }
 310
 311         if (!write)
 312                 bch2_bkey_compat(level, btree_id, version, big_endian,
 313                                  write, NULL, bkey_to_packed(k));
 314
 315         if (bch2_bkey_invalid(c, bkey_i_to_s_c(k),
 316                               __btree_node_type(level, btree_id), write, &buf)) {
 317                 printbuf_reset(&buf);
 318                 journal_entry_err_msg(&buf, version, jset, entry);
 319                 prt_newline(&buf);
 320                 printbuf_indent_add(&buf, 2);
 321
 322                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
 323                 prt_newline(&buf);
 324                 bch2_bkey_invalid(c, bkey_i_to_s_c(k),
 325                                   __btree_node_type(level, btree_id), write, &buf);
 326
 327                 mustfix_fsck_err(c, journal_entry_bkey_invalid,
 328                                  "%s", buf.buf);
 329
 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
 334                 printbuf_exit(&buf);
 335                 return FSCK_DELETED_KEY;
 336         }
 337
 338         if (write)
 339                 bch2_bkey_compat(level, btree_id, version, big_endian,
 340                                  write, NULL, bkey_to_packed(k));
 341 fsck_err:
 342         printbuf_exit(&buf);
 343         return ret;
 344 }
 345
 346 static int journal_entry_btree_keys_validate(struct bch_fs *c,
 347                                 struct jset *jset,
 348                                 struct jset_entry *entry,
 349                                 unsigned version, int big_endian,
 350                                 enum bkey_invalid_flags flags)
 351 {
 352         struct bkey_i *k = entry->start;
 353
 354         while (k != vstruct_last(entry)) {
 355                 int ret = journal_validate_key(c, jset, entry,
 356                                                entry->level,
 357                                                entry->btree_id,
 358                                                k, version, big_endian,
 359                                                flags|BKEY_INVALID_JOURNAL);
 360                 if (ret == FSCK_DELETED_KEY)
 361                         continue;
 362
 363                 k = bkey_next(k);
 364         }
 365
 366         return 0;
 367 }
 368
 369 static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs *c,
 370                                              struct jset_entry *entry)
 371 {
 372         struct bkey_i *k;
 373         bool first = true;
 374
 375         jset_entry_for_each_key(entry, k) {
 376                 if (!first) {
 377                         prt_newline(out);
 378                         prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
 379                 }
 380                 prt_printf(out, "btree=%s l=%u ", bch2_btree_id_str(entry->btree_id), entry->level);
 381                 bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k));
 382                 first = false;
 383         }
 384 }
 385
 386 static int journal_entry_btree_root_validate(struct bch_fs *c,
 387                                 struct jset *jset,
 388                                 struct jset_entry *entry,
 389                                 unsigned version, int big_endian,
 390                                 enum bkey_invalid_flags flags)
 391 {
 392         struct bkey_i *k = entry->start;
 393         int ret = 0;
 394
 395         if (journal_entry_err_on(!entry->u64s ||
 396                                  le16_to_cpu(entry->u64s) != k->k.u64s,
 397                                  c, version, jset, entry,
 398                                  journal_entry_btree_root_bad_size,
 399                                  "invalid btree root journal entry: wrong number of keys")) {
 400                 void *next = vstruct_next(entry);
 401                 /*
 402                  * we don't want to null out this jset_entry,
 403                  * just the contents, so that later we can tell
 404                  * we were _supposed_ to have a btree root
 405                  */
 406                 entry->u64s = 0;
 407                 journal_entry_null_range(vstruct_next(entry), next);
 408                 return 0;
 409         }
 410
 411         return journal_validate_key(c, jset, entry, 1, entry->btree_id, k,
 412                                     version, big_endian, flags);
 413 fsck_err:
 414         return ret;
 415 }
 416
 417 static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs *c,
 418                                              struct jset_entry *entry)
 419 {
 420         journal_entry_btree_keys_to_text(out, c, entry);
 421 }
 422
 423 static int journal_entry_prio_ptrs_validate(struct bch_fs *c,
 424                                 struct jset *jset,
 425                                 struct jset_entry *entry,
 426                                 unsigned version, int big_endian,
 427                                 enum bkey_invalid_flags flags)
 428 {
 429         /* obsolete, don't care: */
 430         return 0;
 431 }
 432
 433 static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
 434                                             struct jset_entry *entry)
 435 {
 436 }
 437
 438 static int journal_entry_blacklist_validate(struct bch_fs *c,
 439                                 struct jset *jset,
 440                                 struct jset_entry *entry,
 441                                 unsigned version, int big_endian,
 442                                 enum bkey_invalid_flags flags)
 443 {
 444         int ret = 0;
 445
 446         if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1,
 447                                  c, version, jset, entry,
 448                                  journal_entry_blacklist_bad_size,
 449                 "invalid journal seq blacklist entry: bad size")) {
 450                 journal_entry_null_range(entry, vstruct_next(entry));
 451         }
 452 fsck_err:
 453         return ret;
 454 }
 455
 456 static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs *c,
 457                                             struct jset_entry *entry)
 458 {
 459         struct jset_entry_blacklist *bl =
 460                 container_of(entry, struct jset_entry_blacklist, entry);
 461
 462         prt_printf(out, "seq=%llu", le64_to_cpu(bl->seq));
 463 }
 464
 465 static int journal_entry_blacklist_v2_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         struct jset_entry_blacklist_v2 *bl_entry;
 472         int ret = 0;
 473
 474         if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2,
 475                                  c, version, jset, entry,
 476                                  journal_entry_blacklist_v2_bad_size,
 477                 "invalid journal seq blacklist entry: bad size")) {
 478                 journal_entry_null_range(entry, vstruct_next(entry));
 479                 goto out;
 480         }
 481
 482         bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry);
 483
 484         if (journal_entry_err_on(le64_to_cpu(bl_entry->start) >
 485                                  le64_to_cpu(bl_entry->end),
 486                                  c, version, jset, entry,
 487                                  journal_entry_blacklist_v2_start_past_end,
 488                 "invalid journal seq blacklist entry: start > end")) {
 489                 journal_entry_null_range(entry, vstruct_next(entry));
 490         }
 491 out:
 492 fsck_err:
 493         return ret;
 494 }
 495
 496 static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_fs *c,
 497                                                struct jset_entry *entry)
 498 {
 499         struct jset_entry_blacklist_v2 *bl =
 500                 container_of(entry, struct jset_entry_blacklist_v2, entry);
 501
 502         prt_printf(out, "start=%llu end=%llu",
 503                le64_to_cpu(bl->start),
 504                le64_to_cpu(bl->end));
 505 }
 506
 507 static int journal_entry_usage_validate(struct bch_fs *c,
 508                                 struct jset *jset,
 509                                 struct jset_entry *entry,
 510                                 unsigned version, int big_endian,
 511                                 enum bkey_invalid_flags flags)
 512 {
 513         struct jset_entry_usage *u =
 514                 container_of(entry, struct jset_entry_usage, entry);
 515         unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 516         int ret = 0;
 517
 518         if (journal_entry_err_on(bytes < sizeof(*u),
 519                                  c, version, jset, entry,
 520                                  journal_entry_usage_bad_size,
 521                                  "invalid journal entry usage: bad size")) {
 522                 journal_entry_null_range(entry, vstruct_next(entry));
 523                 return ret;
 524         }
 525
 526 fsck_err:
 527         return ret;
 528 }
 529
 530 static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c,
 531                                         struct jset_entry *entry)
 532 {
 533         struct jset_entry_usage *u =
 534                 container_of(entry, struct jset_entry_usage, entry);
 535
 536         prt_printf(out, "type=%s v=%llu",
 537                bch2_fs_usage_types[u->entry.btree_id],
 538                le64_to_cpu(u->v));
 539 }
 540
 541 static int journal_entry_data_usage_validate(struct bch_fs *c,
 542                                 struct jset *jset,
 543                                 struct jset_entry *entry,
 544                                 unsigned version, int big_endian,
 545                                 enum bkey_invalid_flags flags)
 546 {
 547         struct jset_entry_data_usage *u =
 548                 container_of(entry, struct jset_entry_data_usage, entry);
 549         unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 550         int ret = 0;
 551
 552         if (journal_entry_err_on(bytes < sizeof(*u) ||
 553                                  bytes < sizeof(*u) + u->r.nr_devs,
 554                                  c, version, jset, entry,
 555                                  journal_entry_data_usage_bad_size,
 556                                  "invalid journal entry usage: bad size")) {
 557                 journal_entry_null_range(entry, vstruct_next(entry));
 558                 return ret;
 559         }
 560
 561 fsck_err:
 562         return ret;
 563 }
 564
 565 static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs *c,
 566                                              struct jset_entry *entry)
 567 {
 568         struct jset_entry_data_usage *u =
 569                 container_of(entry, struct jset_entry_data_usage, entry);
 570
 571         bch2_replicas_entry_to_text(out, &u->r);
 572         prt_printf(out, "=%llu", le64_to_cpu(u->v));
 573 }
 574
 575 static int journal_entry_clock_validate(struct bch_fs *c,
 576                                 struct jset *jset,
 577                                 struct jset_entry *entry,
 578                                 unsigned version, int big_endian,
 579                                 enum bkey_invalid_flags flags)
 580 {
 581         struct jset_entry_clock *clock =
 582                 container_of(entry, struct jset_entry_clock, entry);
 583         unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 584         int ret = 0;
 585
 586         if (journal_entry_err_on(bytes != sizeof(*clock),
 587                                  c, version, jset, entry,
 588                                  journal_entry_clock_bad_size,
 589                                  "bad size")) {
 590                 journal_entry_null_range(entry, vstruct_next(entry));
 591                 return ret;
 592         }
 593
 594         if (journal_entry_err_on(clock->rw > 1,
 595                                  c, version, jset, entry,
 596                                  journal_entry_clock_bad_rw,
 597                                  "bad rw")) {
 598                 journal_entry_null_range(entry, vstruct_next(entry));
 599                 return ret;
 600         }
 601
 602 fsck_err:
 603         return ret;
 604 }
 605
 606 static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c,
 607                                         struct jset_entry *entry)
 608 {
 609         struct jset_entry_clock *clock =
 610                 container_of(entry, struct jset_entry_clock, entry);
 611
 612         prt_printf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time));
 613 }
 614
 615 static int journal_entry_dev_usage_validate(struct bch_fs *c,
 616                                 struct jset *jset,
 617                                 struct jset_entry *entry,
 618                                 unsigned version, int big_endian,
 619                                 enum bkey_invalid_flags flags)
 620 {
 621         struct jset_entry_dev_usage *u =
 622                 container_of(entry, struct jset_entry_dev_usage, entry);
 623         unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
 624         unsigned expected = sizeof(*u);
 625         unsigned dev;
 626         int ret = 0;
 627
 628         if (journal_entry_err_on(bytes < expected,
 629                                  c, version, jset, entry,
 630                                  journal_entry_dev_usage_bad_size,
 631                                  "bad size (%u < %u)",
 632                                  bytes, expected)) {
 633                 journal_entry_null_range(entry, vstruct_next(entry));
 634                 return ret;
 635         }
 636
 637         dev = le32_to_cpu(u->dev);
 638
 639         if (journal_entry_err_on(!bch2_dev_exists2(c, dev),
 640                                  c, version, jset, entry,
 641                                  journal_entry_dev_usage_bad_dev,
 642                                  "bad dev")) {
 643                 journal_entry_null_range(entry, vstruct_next(entry));
 644                 return ret;
 645         }
 646
 647         if (journal_entry_err_on(u->pad,
 648                                  c, version, jset, entry,
 649                                  journal_entry_dev_usage_bad_pad,
 650                                  "bad pad")) {
 651                 journal_entry_null_range(entry, vstruct_next(entry));
 652                 return ret;
 653         }
 654
 655 fsck_err:
 656         return ret;
 657 }
 658
 659 static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs *c,
 660                                             struct jset_entry *entry)
 661 {
 662         struct jset_entry_dev_usage *u =
 663                 container_of(entry, struct jset_entry_dev_usage, entry);
 664         unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 665
 666         prt_printf(out, "dev=%u", le32_to_cpu(u->dev));
 667
 668         for (i = 0; i < nr_types; i++) {
 669                 if (i < BCH_DATA_NR)
 670                         prt_printf(out, " %s", bch2_data_types[i]);
 671                 else
 672                         prt_printf(out, " (unknown data type %u)", i);
 673                 prt_printf(out, ": buckets=%llu sectors=%llu fragmented=%llu",
 674                        le64_to_cpu(u->d[i].buckets),
 675                        le64_to_cpu(u->d[i].sectors),
 676                        le64_to_cpu(u->d[i].fragmented));
 677         }
 678
 679         prt_printf(out, " buckets_ec: %llu", le64_to_cpu(u->buckets_ec));
 680 }
 681
 682 static int journal_entry_log_validate(struct bch_fs *c,
 683                                 struct jset *jset,
 684                                 struct jset_entry *entry,
 685                                 unsigned version, int big_endian,
 686                                 enum bkey_invalid_flags flags)
 687 {
 688         return 0;
 689 }
 690
 691 static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c,
 692                                       struct jset_entry *entry)
 693 {
 694         struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry);
 695         unsigned bytes = vstruct_bytes(entry) - offsetof(struct jset_entry_log, d);
 696
 697         prt_printf(out, "%.*s", bytes, l->d);
 698 }
 699
 700 static int journal_entry_overwrite_validate(struct bch_fs *c,
 701                                 struct jset *jset,
 702                                 struct jset_entry *entry,
 703                                 unsigned version, int big_endian,
 704                                 enum bkey_invalid_flags flags)
 705 {
 706         return journal_entry_btree_keys_validate(c, jset, entry,
 707                                 version, big_endian, READ);
 708 }
 709
 710 static void journal_entry_overwrite_to_text(struct printbuf *out, struct bch_fs *c,
 711                                             struct jset_entry *entry)
 712 {
 713         journal_entry_btree_keys_to_text(out, c, entry);
 714 }
 715
 716 struct jset_entry_ops {
 717         int (*validate)(struct bch_fs *, struct jset *,
 718                         struct jset_entry *, unsigned, int,
 719                         enum bkey_invalid_flags);
 720         void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *);
 721 };
 722
 723 static const struct jset_entry_ops bch2_jset_entry_ops[] = {
 724 #define x(f, nr)                                                \
 725         [BCH_JSET_ENTRY_##f]    = (struct jset_entry_ops) {     \
 726                 .validate       = journal_entry_##f##_validate, \
 727                 .to_text        = journal_entry_##f##_to_text,  \
 728         },
 729         BCH_JSET_ENTRY_TYPES()
 730 #undef x
 731 };
 732
 733 int bch2_journal_entry_validate(struct bch_fs *c,
 734                                 struct jset *jset,
 735                                 struct jset_entry *entry,
 736                                 unsigned version, int big_endian,
 737                                 enum bkey_invalid_flags flags)
 738 {
 739         return entry->type < BCH_JSET_ENTRY_NR
 740                 ? bch2_jset_entry_ops[entry->type].validate(c, jset, entry,
 741                                 version, big_endian, flags)
 742                 : 0;
 743 }
 744
 745 void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c,
 746                                 struct jset_entry *entry)
 747 {
 748         if (entry->type < BCH_JSET_ENTRY_NR) {
 749                 prt_printf(out, "%s: ", bch2_jset_entry_types[entry->type]);
 750                 bch2_jset_entry_ops[entry->type].to_text(out, c, entry);
 751         } else {
 752                 prt_printf(out, "(unknown type %u)", entry->type);
 753         }
 754 }
 755
 756 static int jset_validate_entries(struct bch_fs *c, struct jset *jset,
 757                                  enum bkey_invalid_flags flags)
 758 {
 759         struct jset_entry *entry;
 760         unsigned version = le32_to_cpu(jset->version);
 761         int ret = 0;
 762
 763         vstruct_for_each(jset, entry) {
 764                 if (journal_entry_err_on(vstruct_next(entry) > vstruct_last(jset),
 765                                 c, version, jset, entry,
 766                                 journal_entry_past_jset_end,
 767                                 "journal entry extends past end of jset")) {
 768                         jset->u64s = cpu_to_le32((u64 *) entry - jset->_data);
 769                         break;
 770                 }
 771
 772                 ret = bch2_journal_entry_validate(c, jset, entry,
 773                                         version, JSET_BIG_ENDIAN(jset), flags);
 774                 if (ret)
 775                         break;
 776         }
 777 fsck_err:
 778         return ret;
 779 }
 780
 781 static int jset_validate(struct bch_fs *c,
 782                          struct bch_dev *ca,
 783                          struct jset *jset, u64 sector,
 784                          enum bkey_invalid_flags flags)
 785 {
 786         unsigned version;
 787         int ret = 0;
 788
 789         if (le64_to_cpu(jset->magic) != jset_magic(c))
 790                 return JOURNAL_ENTRY_NONE;
 791
 792         version = le32_to_cpu(jset->version);
 793         if (journal_entry_err_on(!bch2_version_compatible(version),
 794                         c, version, jset, NULL,
 795                         jset_unsupported_version,
 796                         "%s sector %llu seq %llu: incompatible journal entry version %u.%u",
 797                         ca ? ca->name : c->name,
 798                         sector, le64_to_cpu(jset->seq),
 799                         BCH_VERSION_MAJOR(version),
 800                         BCH_VERSION_MINOR(version))) {
 801                 /* don't try to continue: */
 802                 return -EINVAL;
 803         }
 804
 805         if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)),
 806                         c, version, jset, NULL,
 807                         jset_unknown_csum,
 808                         "%s sector %llu seq %llu: journal entry with unknown csum type %llu",
 809                         ca ? ca->name : c->name,
 810                         sector, le64_to_cpu(jset->seq),
 811                         JSET_CSUM_TYPE(jset)))
 812                 ret = JOURNAL_ENTRY_BAD;
 813
 814         /* last_seq is ignored when JSET_NO_FLUSH is true */
 815         if (journal_entry_err_on(!JSET_NO_FLUSH(jset) &&
 816                                  le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq),
 817                                  c, version, jset, NULL,
 818                                  jset_last_seq_newer_than_seq,
 819                                  "invalid journal entry: last_seq > seq (%llu > %llu)",
 820                                  le64_to_cpu(jset->last_seq),
 821                                  le64_to_cpu(jset->seq))) {
 822                 jset->last_seq = jset->seq;
 823                 return JOURNAL_ENTRY_BAD;
 824         }
 825
 826         ret = jset_validate_entries(c, jset, flags);
 827 fsck_err:
 828         return ret;
 829 }
 830
 831 static int jset_validate_early(struct bch_fs *c,
 832                          struct bch_dev *ca,
 833                          struct jset *jset, u64 sector,
 834                          unsigned bucket_sectors_left,
 835                          unsigned sectors_read)
 836 {
 837         size_t bytes = vstruct_bytes(jset);
 838         unsigned version;
 839         enum bkey_invalid_flags flags = BKEY_INVALID_JOURNAL;
 840         int ret = 0;
 841
 842         if (le64_to_cpu(jset->magic) != jset_magic(c))
 843                 return JOURNAL_ENTRY_NONE;
 844
 845         version = le32_to_cpu(jset->version);
 846         if (journal_entry_err_on(!bch2_version_compatible(version),
 847                         c, version, jset, NULL,
 848                         jset_unsupported_version,
 849                         "%s sector %llu seq %llu: unknown journal entry version %u.%u",
 850                         ca ? ca->name : c->name,
 851                         sector, le64_to_cpu(jset->seq),
 852                         BCH_VERSION_MAJOR(version),
 853                         BCH_VERSION_MINOR(version))) {
 854                 /* don't try to continue: */
 855                 return -EINVAL;
 856         }
 857
 858         if (bytes > (sectors_read << 9) &&
 859             sectors_read < bucket_sectors_left)
 860                 return JOURNAL_ENTRY_REREAD;
 861
 862         if (journal_entry_err_on(bytes > bucket_sectors_left << 9,
 863                         c, version, jset, NULL,
 864                         jset_past_bucket_end,
 865                         "%s sector %llu seq %llu: journal entry too big (%zu bytes)",
 866                         ca ? ca->name : c->name,
 867                         sector, le64_to_cpu(jset->seq), bytes))
 868                 le32_add_cpu(&jset->u64s,
 869                              -((bytes - (bucket_sectors_left << 9)) / 8));
 870 fsck_err:
 871         return ret;
 872 }
 873
 874 struct journal_read_buf {
 875         void            *data;
 876         size_t          size;
 877 };
 878
 879 static int journal_read_buf_realloc(struct journal_read_buf *b,
 880                                     size_t new_size)
 881 {
 882         void *n;
 883
 884         /* the bios are sized for this many pages, max: */
 885         if (new_size > JOURNAL_ENTRY_SIZE_MAX)
 886                 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
 887
 888         new_size = roundup_pow_of_two(new_size);
 889         n = kvpmalloc(new_size, GFP_KERNEL);
 890         if (!n)
 891                 return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
 892
 893         kvpfree(b->data, b->size);
 894         b->data = n;
 895         b->size = new_size;
 896         return 0;
 897 }
 898
 899 static int journal_read_bucket(struct bch_dev *ca,
 900                                struct journal_read_buf *buf,
 901                                struct journal_list *jlist,
 902                                unsigned bucket)
 903 {
 904         struct bch_fs *c = ca->fs;
 905         struct journal_device *ja = &ca->journal;
 906         struct jset *j = NULL;
 907         unsigned sectors, sectors_read = 0;
 908         u64 offset = bucket_to_sector(ca, ja->buckets[bucket]),
 909             end = offset + ca->mi.bucket_size;
 910         bool saw_bad = false, csum_good;
 911         int ret = 0;
 912
 913         pr_debug("reading %u", bucket);
 914
 915         while (offset < end) {
 916                 if (!sectors_read) {
 917                         struct bio *bio;
 918                         unsigned nr_bvecs;
 919 reread:
 920                         sectors_read = min_t(unsigned,
 921                                 end - offset, buf->size >> 9);
 922                         nr_bvecs = buf_pages(buf->data, sectors_read << 9);
 923
 924                         bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
 925                         bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ);
 926
 927                         bio->bi_iter.bi_sector = offset;
 928                         bch2_bio_map(bio, buf->data, sectors_read << 9);
 929
 930                         ret = submit_bio_wait(bio);
 931                         kfree(bio);
 932
 933                         if (bch2_dev_io_err_on(ret, ca, BCH_MEMBER_ERROR_read,
 934                                                "journal read error: sector %llu",
 935                                                offset) ||
 936                             bch2_meta_read_fault("journal")) {
 937                                 /*
 938                                  * We don't error out of the recovery process
 939                                  * here, since the relevant journal entry may be
 940                                  * found on a different device, and missing or
 941                                  * no journal entries will be handled later
 942                                  */
 943                                 return 0;
 944                         }
 945
 946                         j = buf->data;
 947                 }
 948
 949                 ret = jset_validate_early(c, ca, j, offset,
 950                                     end - offset, sectors_read);
 951                 switch (ret) {
 952                 case 0:
 953                         sectors = vstruct_sectors(j, c->block_bits);
 954                         break;
 955                 case JOURNAL_ENTRY_REREAD:
 956                         if (vstruct_bytes(j) > buf->size) {
 957                                 ret = journal_read_buf_realloc(buf,
 958                                                         vstruct_bytes(j));
 959                                 if (ret)
 960                                         return ret;
 961                         }
 962                         goto reread;
 963                 case JOURNAL_ENTRY_NONE:
 964                         if (!saw_bad)
 965                                 return 0;
 966                         /*
 967                          * On checksum error we don't really trust the size
 968                          * field of the journal entry we read, so try reading
 969                          * again at next block boundary:
 970                          */
 971                         sectors = block_sectors(c);
 972                         goto next_block;
 973                 default:
 974                         return ret;
 975                 }
 976
 977                 /*
 978                  * This happens sometimes if we don't have discards on -
 979                  * when we've partially overwritten a bucket with new
 980                  * journal entries. We don't need the rest of the
 981                  * bucket:
 982                  */
 983                 if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket])
 984                         return 0;
 985
 986                 ja->bucket_seq[bucket] = le64_to_cpu(j->seq);
 987
 988                 csum_good = jset_csum_good(c, j);
 989                 if (bch2_dev_io_err_on(!csum_good, ca, BCH_MEMBER_ERROR_checksum,
 990                                        "journal checksum error"))
 991                         saw_bad = true;
 992
 993                 ret = bch2_encrypt(c, JSET_CSUM_TYPE(j), journal_nonce(j),
 994                              j->encrypted_start,
 995                              vstruct_end(j) - (void *) j->encrypted_start);
 996                 bch2_fs_fatal_err_on(ret, c,
 997                                 "error decrypting journal entry: %i", ret);
 998
 999                 mutex_lock(&jlist->lock);
1000                 ret = journal_entry_add(c, ca, (struct journal_ptr) {
1001                                         .csum_good      = csum_good,
1002                                         .dev            = ca->dev_idx,
1003                                         .bucket         = bucket,
1004                                         .bucket_offset  = offset -
1005                                                 bucket_to_sector(ca, ja->buckets[bucket]),
1006                                         .sector         = offset,
1007                                         }, jlist, j);
1008                 mutex_unlock(&jlist->lock);
1009
1010                 switch (ret) {
1011                 case JOURNAL_ENTRY_ADD_OK:
1012                         break;
1013                 case JOURNAL_ENTRY_ADD_OUT_OF_RANGE:
1014                         break;
1015                 default:
1016                         return ret;
1017                 }
1018 next_block:
1019                 pr_debug("next");
1020                 offset          += sectors;
1021                 sectors_read    -= sectors;
1022                 j = ((void *) j) + (sectors << 9);
1023         }
1024
1025         return 0;
1026 }
1027
1028 static void bch2_journal_read_device(struct closure *cl)
1029 {
1030         struct journal_device *ja =
1031                 container_of(cl, struct journal_device, read);
1032         struct bch_dev *ca = container_of(ja, struct bch_dev, journal);
1033         struct bch_fs *c = ca->fs;
1034         struct journal_list *jlist =
1035                 container_of(cl->parent, struct journal_list, cl);
1036         struct journal_replay *r, **_r;
1037         struct genradix_iter iter;
1038         struct journal_read_buf buf = { NULL, 0 };
1039         unsigned i;
1040         int ret = 0;
1041
1042         if (!ja->nr)
1043                 goto out;
1044
1045         ret = journal_read_buf_realloc(&buf, PAGE_SIZE);
1046         if (ret)
1047                 goto err;
1048
1049         pr_debug("%u journal buckets", ja->nr);
1050
1051         for (i = 0; i < ja->nr; i++) {
1052                 ret = journal_read_bucket(ca, &buf, jlist, i);
1053                 if (ret)
1054                         goto err;
1055         }
1056
1057         ja->sectors_free = ca->mi.bucket_size;
1058
1059         mutex_lock(&jlist->lock);
1060         genradix_for_each_reverse(&c->journal_entries, iter, _r) {
1061                 r = *_r;
1062
1063                 if (!r)
1064                         continue;
1065
1066                 for (i = 0; i < r->nr_ptrs; i++) {
1067                         if (r->ptrs[i].dev == ca->dev_idx) {
1068                                 unsigned wrote = bucket_remainder(ca, r->ptrs[i].sector) +
1069                                         vstruct_sectors(&r->j, c->block_bits);
1070
1071                                 ja->cur_idx = r->ptrs[i].bucket;
1072                                 ja->sectors_free = ca->mi.bucket_size - wrote;
1073                                 goto found;
1074                         }
1075                 }
1076         }
1077 found:
1078         mutex_unlock(&jlist->lock);
1079
1080         if (ja->bucket_seq[ja->cur_idx] &&
1081             ja->sectors_free == ca->mi.bucket_size) {
1082 #if 0
1083                 /*
1084                  * Debug code for ZNS support, where we (probably) want to be
1085                  * correlated where we stopped in the journal to the zone write
1086                  * points:
1087                  */
1088                 bch_err(c, "ja->sectors_free == ca->mi.bucket_size");
1089                 bch_err(c, "cur_idx %u/%u", ja->cur_idx, ja->nr);
1090                 for (i = 0; i < 3; i++) {
1091                         unsigned idx = (ja->cur_idx + ja->nr - 1 + i) % ja->nr;
1092
1093                         bch_err(c, "bucket_seq[%u] = %llu", idx, ja->bucket_seq[idx]);
1094                 }
1095 #endif
1096                 ja->sectors_free = 0;
1097         }
1098
1099         /*
1100          * Set dirty_idx to indicate the entire journal is full and needs to be
1101          * reclaimed - journal reclaim will immediately reclaim whatever isn't
1102          * pinned when it first runs:
1103          */
1104         ja->discard_idx = ja->dirty_idx_ondisk =
1105                 ja->dirty_idx = (ja->cur_idx + 1) % ja->nr;
1106 out:
1107         bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret);
1108         kvpfree(buf.data, buf.size);
1109         percpu_ref_put(&ca->io_ref);
1110         closure_return(cl);
1111         return;
1112 err:
1113         mutex_lock(&jlist->lock);
1114         jlist->ret = ret;
1115         mutex_unlock(&jlist->lock);
1116         goto out;
1117 }
1118
1119 void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
1120                                struct journal_replay *j)
1121 {
1122         unsigned i;
1123
1124         for (i = 0; i < j->nr_ptrs; i++) {
1125                 struct bch_dev *ca = bch_dev_bkey_exists(c, j->ptrs[i].dev);
1126                 u64 offset;
1127
1128                 div64_u64_rem(j->ptrs[i].sector, ca->mi.bucket_size, &offset);
1129
1130                 if (i)
1131                         prt_printf(out, " ");
1132                 prt_printf(out, "%u:%u:%u (sector %llu)",
1133                        j->ptrs[i].dev,
1134                        j->ptrs[i].bucket,
1135                        j->ptrs[i].bucket_offset,
1136                        j->ptrs[i].sector);
1137         }
1138 }
1139
1140 int bch2_journal_read(struct bch_fs *c,
1141                       u64 *last_seq,
1142                       u64 *blacklist_seq,
1143                       u64 *start_seq)
1144 {
1145         struct journal_list jlist;
1146         struct journal_replay *i, **_i, *prev = NULL;
1147         struct genradix_iter radix_iter;
1148         struct bch_dev *ca;
1149         unsigned iter;
1150         struct printbuf buf = PRINTBUF;
1151         bool degraded = false, last_write_torn = false;
1152         u64 seq;
1153         int ret = 0;
1154
1155         closure_init_stack(&jlist.cl);
1156         mutex_init(&jlist.lock);
1157         jlist.last_seq = 0;
1158         jlist.ret = 0;
1159
1160         for_each_member_device(ca, c, iter) {
1161                 if (!c->opts.fsck &&
1162                     !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal)))
1163                         continue;
1164
1165                 if ((ca->mi.state == BCH_MEMBER_STATE_rw ||
1166                      ca->mi.state == BCH_MEMBER_STATE_ro) &&
1167                     percpu_ref_tryget(&ca->io_ref))
1168                         closure_call(&ca->journal.read,
1169                                      bch2_journal_read_device,
1170                                      system_unbound_wq,
1171                                      &jlist.cl);
1172                 else
1173                         degraded = true;
1174         }
1175
1176         closure_sync(&jlist.cl);
1177
1178         if (jlist.ret)
1179                 return jlist.ret;
1180
1181         *last_seq       = 0;
1182         *start_seq      = 0;
1183         *blacklist_seq  = 0;
1184
1185         /*
1186          * Find most recent flush entry, and ignore newer non flush entries -
1187          * those entries will be blacklisted:
1188          */
1189         genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) {
1190                 enum bkey_invalid_flags flags = BKEY_INVALID_JOURNAL;
1191
1192                 i = *_i;
1193
1194                 if (!i || i->ignore)
1195                         continue;
1196
1197                 if (!*start_seq)
1198                         *blacklist_seq = *start_seq = le64_to_cpu(i->j.seq) + 1;
1199
1200                 if (JSET_NO_FLUSH(&i->j)) {
1201                         i->ignore = true;
1202                         continue;
1203                 }
1204
1205                 if (!last_write_torn && !i->csum_good) {
1206                         last_write_torn = true;
1207                         i->ignore = true;
1208                         continue;
1209                 }
1210
1211                 if (journal_entry_err_on(le64_to_cpu(i->j.last_seq) > le64_to_cpu(i->j.seq),
1212                                          c, le32_to_cpu(i->j.version), &i->j, NULL,
1213                                          jset_last_seq_newer_than_seq,
1214                                          "invalid journal entry: last_seq > seq (%llu > %llu)",
1215                                          le64_to_cpu(i->j.last_seq),
1216                                          le64_to_cpu(i->j.seq)))
1217                         i->j.last_seq = i->j.seq;
1218
1219                 *last_seq       = le64_to_cpu(i->j.last_seq);
1220                 *blacklist_seq  = le64_to_cpu(i->j.seq) + 1;
1221                 break;
1222         }
1223
1224         if (!*start_seq) {
1225                 bch_info(c, "journal read done, but no entries found");
1226                 return 0;
1227         }
1228
1229         if (!*last_seq) {
1230                 fsck_err(c, dirty_but_no_journal_entries_post_drop_nonflushes,
1231                          "journal read done, but no entries found after dropping non-flushes");
1232                 return 0;
1233         }
1234
1235         bch_info(c, "journal read done, replaying entries %llu-%llu",
1236                  *last_seq, *blacklist_seq - 1);
1237
1238         if (*start_seq != *blacklist_seq)
1239                 bch_info(c, "dropped unflushed entries %llu-%llu",
1240                          *blacklist_seq, *start_seq - 1);
1241
1242         /* Drop blacklisted entries and entries older than last_seq: */
1243         genradix_for_each(&c->journal_entries, radix_iter, _i) {
1244                 i = *_i;
1245
1246                 if (!i || i->ignore)
1247                         continue;
1248
1249                 seq = le64_to_cpu(i->j.seq);
1250                 if (seq < *last_seq) {
1251                         journal_replay_free(c, i);
1252                         continue;
1253                 }
1254
1255                 if (bch2_journal_seq_is_blacklisted(c, seq, true)) {
1256                         fsck_err_on(!JSET_NO_FLUSH(&i->j), c,
1257                                     jset_seq_blacklisted,
1258                                     "found blacklisted journal entry %llu", seq);
1259                         i->ignore = true;
1260                 }
1261         }
1262
1263         /* Check for missing entries: */
1264         seq = *last_seq;
1265         genradix_for_each(&c->journal_entries, radix_iter, _i) {
1266                 i = *_i;
1267
1268                 if (!i || i->ignore)
1269                         continue;
1270
1271                 BUG_ON(seq > le64_to_cpu(i->j.seq));
1272
1273                 while (seq < le64_to_cpu(i->j.seq)) {
1274                         u64 missing_start, missing_end;
1275                         struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
1276
1277                         while (seq < le64_to_cpu(i->j.seq) &&
1278                                bch2_journal_seq_is_blacklisted(c, seq, false))
1279                                 seq++;
1280
1281                         if (seq == le64_to_cpu(i->j.seq))
1282                                 break;
1283
1284                         missing_start = seq;
1285
1286                         while (seq < le64_to_cpu(i->j.seq) &&
1287                                !bch2_journal_seq_is_blacklisted(c, seq, false))
1288                                 seq++;
1289
1290                         if (prev) {
1291                                 bch2_journal_ptrs_to_text(&buf1, c, prev);
1292                                 prt_printf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits));
1293                         } else
1294                                 prt_printf(&buf1, "(none)");
1295                         bch2_journal_ptrs_to_text(&buf2, c, i);
1296
1297                         missing_end = seq - 1;
1298                         fsck_err(c, journal_entries_missing,
1299                                  "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
1300                                  "  prev at %s\n"
1301                                  "  next at %s",
1302                                  missing_start, missing_end,
1303                                  *last_seq, *blacklist_seq - 1,
1304                                  buf1.buf, buf2.buf);
1305
1306                         printbuf_exit(&buf1);
1307                         printbuf_exit(&buf2);
1308                 }
1309
1310                 prev = i;
1311                 seq++;
1312         }
1313
1314         genradix_for_each(&c->journal_entries, radix_iter, _i) {
1315                 struct bch_replicas_padded replicas = {
1316                         .e.data_type = BCH_DATA_journal,
1317                         .e.nr_required = 1,
1318                 };
1319                 unsigned ptr;
1320
1321                 i = *_i;
1322                 if (!i || i->ignore)
1323                         continue;
1324
1325                 for (ptr = 0; ptr < i->nr_ptrs; ptr++) {
1326                         ca = bch_dev_bkey_exists(c, i->ptrs[ptr].dev);
1327
1328                         if (!i->ptrs[ptr].csum_good)
1329                                 bch_err_dev_offset(ca, i->ptrs[ptr].sector,
1330                                                    "invalid journal checksum, seq %llu%s",
1331                                                    le64_to_cpu(i->j.seq),
1332                                                    i->csum_good ? " (had good copy on another device)" : "");
1333                 }
1334
1335                 ret = jset_validate(c,
1336                                     bch_dev_bkey_exists(c, i->ptrs[0].dev),
1337                                     &i->j,
1338                                     i->ptrs[0].sector,
1339                                     READ);
1340                 if (ret)
1341                         goto err;
1342
1343                 for (ptr = 0; ptr < i->nr_ptrs; ptr++)
1344                         replicas.e.devs[replicas.e.nr_devs++] = i->ptrs[ptr].dev;
1345
1346                 bch2_replicas_entry_sort(&replicas.e);
1347
1348                 printbuf_reset(&buf);
1349                 bch2_replicas_entry_to_text(&buf, &replicas.e);
1350
1351                 if (!degraded &&
1352                     !bch2_replicas_marked(c, &replicas.e) &&
1353                     (le64_to_cpu(i->j.seq) == *last_seq ||
1354                      fsck_err(c, journal_entry_replicas_not_marked,
1355                               "superblock not marked as containing replicas for journal entry %llu\n  %s",
1356                               le64_to_cpu(i->j.seq), buf.buf))) {
1357                         ret = bch2_mark_replicas(c, &replicas.e);
1358                         if (ret)
1359                                 goto err;
1360                 }
1361         }
1362 err:
1363 fsck_err:
1364         printbuf_exit(&buf);
1365         return ret;
1366 }
1367
1368 /* journal write: */
1369
1370 static void __journal_write_alloc(struct journal *j,
1371                                   struct journal_buf *w,
1372                                   struct dev_alloc_list *devs_sorted,
1373                                   unsigned sectors,
1374                                   unsigned *replicas,
1375                                   unsigned replicas_want)
1376 {
1377         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1378         struct journal_device *ja;
1379         struct bch_dev *ca;
1380         unsigned i;
1381
1382         if (*replicas >= replicas_want)
1383                 return;
1384
1385         for (i = 0; i < devs_sorted->nr; i++) {
1386                 ca = rcu_dereference(c->devs[devs_sorted->devs[i]]);
1387                 if (!ca)
1388                         continue;
1389
1390                 ja = &ca->journal;
1391
1392                 /*
1393                  * Check that we can use this device, and aren't already using
1394                  * it:
1395                  */
1396                 if (!ca->mi.durability ||
1397                     ca->mi.state != BCH_MEMBER_STATE_rw ||
1398                     !ja->nr ||
1399                     bch2_bkey_has_device_c(bkey_i_to_s_c(&w->key), ca->dev_idx) ||
1400                     sectors > ja->sectors_free)
1401                         continue;
1402
1403                 bch2_dev_stripe_increment(ca, &j->wp.stripe);
1404
1405                 bch2_bkey_append_ptr(&w->key,
1406                         (struct bch_extent_ptr) {
1407                                   .offset = bucket_to_sector(ca,
1408                                         ja->buckets[ja->cur_idx]) +
1409                                         ca->mi.bucket_size -
1410                                         ja->sectors_free,
1411                                   .dev = ca->dev_idx,
1412                 });
1413
1414                 ja->sectors_free -= sectors;
1415                 ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1416
1417                 *replicas += ca->mi.durability;
1418
1419                 if (*replicas >= replicas_want)
1420                         break;
1421         }
1422 }
1423
1424 /**
1425  * journal_write_alloc - decide where to write next journal entry
1426  *
1427  * @j:          journal object
1428  * @w:          journal buf (entry to be written)
1429  *
1430  * Returns: 0 on success, or -EROFS on failure
1431  */
1432 static int journal_write_alloc(struct journal *j, struct journal_buf *w)
1433 {
1434         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1435         struct bch_devs_mask devs;
1436         struct journal_device *ja;
1437         struct bch_dev *ca;
1438         struct dev_alloc_list devs_sorted;
1439         unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1440         unsigned target = c->opts.metadata_target ?:
1441                 c->opts.foreground_target;
1442         unsigned i, replicas = 0, replicas_want =
1443                 READ_ONCE(c->opts.metadata_replicas);
1444
1445         rcu_read_lock();
1446 retry:
1447         devs = target_rw_devs(c, BCH_DATA_journal, target);
1448
1449         devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs);
1450
1451         __journal_write_alloc(j, w, &devs_sorted,
1452                               sectors, &replicas, replicas_want);
1453
1454         if (replicas >= replicas_want)
1455                 goto done;
1456
1457         for (i = 0; i < devs_sorted.nr; i++) {
1458                 ca = rcu_dereference(c->devs[devs_sorted.devs[i]]);
1459                 if (!ca)
1460                         continue;
1461
1462                 ja = &ca->journal;
1463
1464                 if (sectors > ja->sectors_free &&
1465                     sectors <= ca->mi.bucket_size &&
1466                     bch2_journal_dev_buckets_available(j, ja,
1467                                         journal_space_discarded)) {
1468                         ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
1469                         ja->sectors_free = ca->mi.bucket_size;
1470
1471                         /*
1472                          * ja->bucket_seq[ja->cur_idx] must always have
1473                          * something sensible:
1474                          */
1475                         ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1476                 }
1477         }
1478
1479         __journal_write_alloc(j, w, &devs_sorted,
1480                               sectors, &replicas, replicas_want);
1481
1482         if (replicas < replicas_want && target) {
1483                 /* Retry from all devices: */
1484                 target = 0;
1485                 goto retry;
1486         }
1487 done:
1488         rcu_read_unlock();
1489
1490         BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX);
1491
1492         return replicas >= c->opts.metadata_replicas_required ? 0 : -EROFS;
1493 }
1494
1495 static void journal_buf_realloc(struct journal *j, struct journal_buf *buf)
1496 {
1497         /* we aren't holding j->lock: */
1498         unsigned new_size = READ_ONCE(j->buf_size_want);
1499         void *new_buf;
1500
1501         if (buf->buf_size >= new_size)
1502                 return;
1503
1504         new_buf = kvpmalloc(new_size, GFP_NOFS|__GFP_NOWARN);
1505         if (!new_buf)
1506                 return;
1507
1508         memcpy(new_buf, buf->data, buf->buf_size);
1509
1510         spin_lock(&j->lock);
1511         swap(buf->data,         new_buf);
1512         swap(buf->buf_size,     new_size);
1513         spin_unlock(&j->lock);
1514
1515         kvpfree(new_buf, new_size);
1516 }
1517
1518 static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j)
1519 {
1520         return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK);
1521 }
1522
1523 static void journal_write_done(struct closure *cl)
1524 {
1525         struct journal *j = container_of(cl, struct journal, io);
1526         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1527         struct journal_buf *w = journal_last_unwritten_buf(j);
1528         struct bch_replicas_padded replicas;
1529         union journal_res_state old, new;
1530         u64 v, seq;
1531         int err = 0;
1532
1533         bch2_time_stats_update(!JSET_NO_FLUSH(w->data)
1534                                ? j->flush_write_time
1535                                : j->noflush_write_time, j->write_start_time);
1536
1537         if (!w->devs_written.nr) {
1538                 bch_err(c, "unable to write journal to sufficient devices");
1539                 err = -EIO;
1540         } else {
1541                 bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
1542                                          w->devs_written);
1543                 if (bch2_mark_replicas(c, &replicas.e))
1544                         err = -EIO;
1545         }
1546
1547         if (err)
1548                 bch2_fatal_error(c);
1549
1550         spin_lock(&j->lock);
1551         seq = le64_to_cpu(w->data->seq);
1552
1553         if (seq >= j->pin.front)
1554                 journal_seq_pin(j, seq)->devs = w->devs_written;
1555
1556         if (!err) {
1557                 if (!JSET_NO_FLUSH(w->data)) {
1558                         j->flushed_seq_ondisk = seq;
1559                         j->last_seq_ondisk = w->last_seq;
1560
1561                         bch2_do_discards(c);
1562                         closure_wake_up(&c->freelist_wait);
1563
1564                         bch2_reset_alloc_cursors(c);
1565                 }
1566         } else if (!j->err_seq || seq < j->err_seq)
1567                 j->err_seq      = seq;
1568
1569         j->seq_ondisk           = seq;
1570
1571         /*
1572          * Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard
1573          * more buckets:
1574          *
1575          * Must come before signaling write completion, for
1576          * bch2_fs_journal_stop():
1577          */
1578         if (j->watermark != BCH_WATERMARK_stripe)
1579                 journal_reclaim_kick(&c->journal);
1580
1581         /* also must come before signalling write completion: */
1582         closure_debug_destroy(cl);
1583
1584         v = atomic64_read(&j->reservations.counter);
1585         do {
1586                 old.v = new.v = v;
1587                 BUG_ON(journal_state_count(new, new.unwritten_idx));
1588
1589                 new.unwritten_idx++;
1590         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
1591                                        old.v, new.v)) != old.v);
1592
1593         bch2_journal_space_available(j);
1594
1595         track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight],
1596                            &j->max_in_flight_start, false);
1597
1598         closure_wake_up(&w->wait);
1599         journal_wake(j);
1600
1601         if (!journal_state_count(new, new.unwritten_idx) &&
1602             journal_last_unwritten_seq(j) <= journal_cur_seq(j)) {
1603                 spin_unlock(&j->lock);
1604                 closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
1605         } else if (journal_last_unwritten_seq(j) == journal_cur_seq(j) &&
1606                    new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) {
1607                 struct journal_buf *buf = journal_cur_buf(j);
1608                 long delta = buf->expires - jiffies;
1609
1610                 /*
1611                  * We don't close a journal entry to write it while there's
1612                  * previous entries still in flight - the current journal entry
1613                  * might want to be written now:
1614                  */
1615
1616                 spin_unlock(&j->lock);
1617                 mod_delayed_work(c->io_complete_wq, &j->write_work, max(0L, delta));
1618         } else {
1619                 spin_unlock(&j->lock);
1620         }
1621 }
1622
1623 static void journal_write_endio(struct bio *bio)
1624 {
1625         struct bch_dev *ca = bio->bi_private;
1626         struct journal *j = &ca->fs->journal;
1627         struct journal_buf *w = journal_last_unwritten_buf(j);
1628         unsigned long flags;
1629
1630         if (bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_write,
1631                                "error writing journal entry %llu: %s",
1632                                le64_to_cpu(w->data->seq),
1633                                bch2_blk_status_to_str(bio->bi_status)) ||
1634             bch2_meta_write_fault("journal")) {
1635                 spin_lock_irqsave(&j->err_lock, flags);
1636                 bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx);
1637                 spin_unlock_irqrestore(&j->err_lock, flags);
1638         }
1639
1640         closure_put(&j->io);
1641         percpu_ref_put(&ca->io_ref);
1642 }
1643
1644 static void do_journal_write(struct closure *cl)
1645 {
1646         struct journal *j = container_of(cl, struct journal, io);
1647         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1648         struct bch_dev *ca;
1649         struct journal_buf *w = journal_last_unwritten_buf(j);
1650         struct bch_extent_ptr *ptr;
1651         struct bio *bio;
1652         unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1653
1654         extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) {
1655                 ca = bch_dev_bkey_exists(c, ptr->dev);
1656                 if (!percpu_ref_tryget(&ca->io_ref)) {
1657                         /* XXX: fix this */
1658                         bch_err(c, "missing device for journal write\n");
1659                         continue;
1660                 }
1661
1662                 this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal],
1663                              sectors);
1664
1665                 bio = ca->journal.bio;
1666                 bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
1667                 bio->bi_iter.bi_sector  = ptr->offset;
1668                 bio->bi_end_io          = journal_write_endio;
1669                 bio->bi_private         = ca;
1670
1671                 BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector);
1672                 ca->prev_journal_sector = bio->bi_iter.bi_sector;
1673
1674                 if (!JSET_NO_FLUSH(w->data))
1675                         bio->bi_opf    |= REQ_FUA;
1676                 if (!JSET_NO_FLUSH(w->data) && !w->separate_flush)
1677                         bio->bi_opf    |= REQ_PREFLUSH;
1678
1679                 bch2_bio_map(bio, w->data, sectors << 9);
1680
1681                 trace_and_count(c, journal_write, bio);
1682                 closure_bio_submit(bio, cl);
1683
1684                 ca->journal.bucket_seq[ca->journal.cur_idx] =
1685                         le64_to_cpu(w->data->seq);
1686         }
1687
1688         continue_at(cl, journal_write_done, c->io_complete_wq);
1689 }
1690
1691 static int bch2_journal_write_prep(struct journal *j, struct journal_buf *w)
1692 {
1693         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1694         struct jset_entry *start, *end, *i, *next, *prev = NULL;
1695         struct jset *jset = w->data;
1696         unsigned sectors, bytes, u64s;
1697         bool validate_before_checksum = false;
1698         unsigned long btree_roots_have = 0;
1699         int ret;
1700
1701         /*
1702          * Simple compaction, dropping empty jset_entries (from journal
1703          * reservations that weren't fully used) and merging jset_entries that
1704          * can be.
1705          *
1706          * If we wanted to be really fancy here, we could sort all the keys in
1707          * the jset and drop keys that were overwritten - probably not worth it:
1708          */
1709         vstruct_for_each_safe(jset, i, next) {
1710                 unsigned u64s = le16_to_cpu(i->u64s);
1711
1712                 /* Empty entry: */
1713                 if (!u64s)
1714                         continue;
1715
1716                 /*
1717                  * New btree roots are set by journalling them; when the journal
1718                  * entry gets written we have to propagate them to
1719                  * c->btree_roots
1720                  *
1721                  * But, every journal entry we write has to contain all the
1722                  * btree roots (at least for now); so after we copy btree roots
1723                  * to c->btree_roots we have to get any missing btree roots and
1724                  * add them to this journal entry:
1725                  */
1726                 if (i->type == BCH_JSET_ENTRY_btree_root) {
1727                         bch2_journal_entry_to_btree_root(c, i);
1728                         __set_bit(i->btree_id, &btree_roots_have);
1729                 }
1730
1731                 /* Can we merge with previous entry? */
1732                 if (prev &&
1733                     i->btree_id == prev->btree_id &&
1734                     i->level    == prev->level &&
1735                     i->type     == prev->type &&
1736                     i->type     == BCH_JSET_ENTRY_btree_keys &&
1737                     le16_to_cpu(prev->u64s) + u64s <= U16_MAX) {
1738                         memmove_u64s_down(vstruct_next(prev),
1739                                           i->_data,
1740                                           u64s);
1741                         le16_add_cpu(&prev->u64s, u64s);
1742                         continue;
1743                 }
1744
1745                 /* Couldn't merge, move i into new position (after prev): */
1746                 prev = prev ? vstruct_next(prev) : jset->start;
1747                 if (i != prev)
1748                         memmove_u64s_down(prev, i, jset_u64s(u64s));
1749         }
1750
1751         prev = prev ? vstruct_next(prev) : jset->start;
1752         jset->u64s = cpu_to_le32((u64 *) prev - jset->_data);
1753
1754         start = end = vstruct_last(jset);
1755
1756         end     = bch2_btree_roots_to_journal_entries(c, end, btree_roots_have);
1757
1758         bch2_journal_super_entries_add_common(c, &end,
1759                                 le64_to_cpu(jset->seq));
1760         u64s    = (u64 *) end - (u64 *) start;
1761         BUG_ON(u64s > j->entry_u64s_reserved);
1762
1763         le32_add_cpu(&jset->u64s, u64s);
1764
1765         sectors = vstruct_sectors(jset, c->block_bits);
1766         bytes   = vstruct_bytes(jset);
1767
1768         if (sectors > w->sectors) {
1769                 bch2_fs_fatal_error(c, "aieeee! journal write overran available space, %zu > %u (extra %u reserved %u/%u)",
1770                                     vstruct_bytes(jset), w->sectors << 9,
1771                                     u64s, w->u64s_reserved, j->entry_u64s_reserved);
1772                 return -EINVAL;
1773         }
1774
1775         jset->magic             = cpu_to_le64(jset_magic(c));
1776         jset->version           = cpu_to_le32(c->sb.version);
1777
1778         SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN);
1779         SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c));
1780
1781         if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset))
1782                 j->last_empty_seq = le64_to_cpu(jset->seq);
1783
1784         if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset)))
1785                 validate_before_checksum = true;
1786
1787         if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current)
1788                 validate_before_checksum = true;
1789
1790         if (validate_before_checksum &&
1791             (ret = jset_validate(c, NULL, jset, 0, WRITE)))
1792                 return ret;
1793
1794         ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
1795                     jset->encrypted_start,
1796                     vstruct_end(jset) - (void *) jset->encrypted_start);
1797         if (bch2_fs_fatal_err_on(ret, c,
1798                         "error decrypting journal entry: %i", ret))
1799                 return ret;
1800
1801         jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset),
1802                                   journal_nonce(jset), jset);
1803
1804         if (!validate_before_checksum &&
1805             (ret = jset_validate(c, NULL, jset, 0, WRITE)))
1806                 return ret;
1807
1808         memset((void *) jset + bytes, 0, (sectors << 9) - bytes);
1809         return 0;
1810 }
1811
1812 static int bch2_journal_write_pick_flush(struct journal *j, struct journal_buf *w)
1813 {
1814         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1815         int error = bch2_journal_error(j);
1816
1817         /*
1818          * If the journal is in an error state - we did an emergency shutdown -
1819          * we prefer to continue doing journal writes. We just mark them as
1820          * noflush so they'll never be used, but they'll still be visible by the
1821          * list_journal tool - this helps in debugging.
1822          *
1823          * There's a caveat: the first journal write after marking the
1824          * superblock dirty must always be a flush write, because on startup
1825          * from a clean shutdown we didn't necessarily read the journal and the
1826          * new journal write might overwrite whatever was in the journal
1827          * previously - we can't leave the journal without any flush writes in
1828          * it.
1829          *
1830          * So if we're in an error state, and we're still starting up, we don't
1831          * write anything at all.
1832          */
1833         if (error && test_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags))
1834                 return -EIO;
1835
1836         if (error ||
1837             w->noflush ||
1838             (!w->must_flush &&
1839              (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) &&
1840              test_bit(JOURNAL_MAY_SKIP_FLUSH, &j->flags))) {
1841                      w->noflush = true;
1842                 SET_JSET_NO_FLUSH(w->data, true);
1843                 w->data->last_seq       = 0;
1844                 w->last_seq             = 0;
1845
1846                 j->nr_noflush_writes++;
1847         } else {
1848                 j->last_flush_write = jiffies;
1849                 j->nr_flush_writes++;
1850                 clear_bit(JOURNAL_NEED_FLUSH_WRITE, &j->flags);
1851         }
1852
1853         return 0;
1854 }
1855
1856 void bch2_journal_write(struct closure *cl)
1857 {
1858         struct journal *j = container_of(cl, struct journal, io);
1859         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1860         struct bch_dev *ca;
1861         struct journal_buf *w = journal_last_unwritten_buf(j);
1862         struct bch_replicas_padded replicas;
1863         struct bio *bio;
1864         struct printbuf journal_debug_buf = PRINTBUF;
1865         unsigned i, nr_rw_members = 0;
1866         int ret;
1867
1868         BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
1869
1870         j->write_start_time = local_clock();
1871
1872         spin_lock(&j->lock);
1873         ret = bch2_journal_write_pick_flush(j, w);
1874         spin_unlock(&j->lock);
1875         if (ret)
1876                 goto err;
1877
1878         journal_buf_realloc(j, w);
1879
1880         ret = bch2_journal_write_prep(j, w);
1881         if (ret)
1882                 goto err;
1883
1884         j->entry_bytes_written += vstruct_bytes(w->data);
1885
1886         while (1) {
1887                 spin_lock(&j->lock);
1888                 ret = journal_write_alloc(j, w);
1889                 if (!ret || !j->can_discard)
1890                         break;
1891
1892                 spin_unlock(&j->lock);
1893                 bch2_journal_do_discards(j);
1894         }
1895
1896         if (ret) {
1897                 __bch2_journal_debug_to_text(&journal_debug_buf, j);
1898                 spin_unlock(&j->lock);
1899                 bch_err(c, "Unable to allocate journal write:\n%s",
1900                         journal_debug_buf.buf);
1901                 printbuf_exit(&journal_debug_buf);
1902                 goto err;
1903         }
1904
1905         /*
1906          * write is allocated, no longer need to account for it in
1907          * bch2_journal_space_available():
1908          */
1909         w->sectors = 0;
1910
1911         /*
1912          * journal entry has been compacted and allocated, recalculate space
1913          * available:
1914          */
1915         bch2_journal_space_available(j);
1916         spin_unlock(&j->lock);
1917
1918         w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key));
1919
1920         if (c->opts.nochanges)
1921                 goto no_io;
1922
1923         for_each_rw_member(ca, c, i)
1924                 nr_rw_members++;
1925
1926         if (nr_rw_members > 1)
1927                 w->separate_flush = true;
1928
1929         /*
1930          * Mark journal replicas before we submit the write to guarantee
1931          * recovery will find the journal entries after a crash.
1932          */
1933         bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
1934                                  w->devs_written);
1935         ret = bch2_mark_replicas(c, &replicas.e);
1936         if (ret)
1937                 goto err;
1938
1939         if (!JSET_NO_FLUSH(w->data) && w->separate_flush) {
1940                 for_each_rw_member(ca, c, i) {
1941                         percpu_ref_get(&ca->io_ref);
1942
1943                         bio = ca->journal.bio;
1944                         bio_reset(bio, ca->disk_sb.bdev, REQ_OP_FLUSH);
1945                         bio->bi_end_io          = journal_write_endio;
1946                         bio->bi_private         = ca;
1947                         closure_bio_submit(bio, cl);
1948                 }
1949         }
1950
1951         continue_at(cl, do_journal_write, c->io_complete_wq);
1952         return;
1953 no_io:
1954         continue_at(cl, journal_write_done, c->io_complete_wq);
1955         return;
1956 err:
1957         bch2_fatal_error(c);
1958         continue_at(cl, journal_write_done, c->io_complete_wq);
1959 }