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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcachefs journalling code, for btree insertions
   4  *
   5  * Copyright 2012 Google, Inc.
   6  */
   7
   8 #include "bcachefs.h"
   9 #include "alloc_foreground.h"
  10 #include "bkey_methods.h"
  11 #include "btree_gc.h"
  12 #include "btree_update.h"
  13 #include "buckets.h"
  14 #include "error.h"
  15 #include "journal.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_sb.h"
  19 #include "journal_seq_blacklist.h"
  20 #include "trace.h"
  21
  22 static const char * const bch2_journal_errors[] = {
  23 #define x(n)    #n,
  24         JOURNAL_ERRORS()
  25 #undef x
  26         NULL
  27 };
  28
  29 static inline bool journal_seq_unwritten(struct journal *j, u64 seq)
  30 {
  31         return seq > j->seq_ondisk;
  32 }
  33
  34 static bool __journal_entry_is_open(union journal_res_state state)
  35 {
  36         return state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL;
  37 }
  38
  39 static inline unsigned nr_unwritten_journal_entries(struct journal *j)
  40 {
  41         return atomic64_read(&j->seq) - j->seq_ondisk;
  42 }
  43
  44 static bool journal_entry_is_open(struct journal *j)
  45 {
  46         return __journal_entry_is_open(j->reservations);
  47 }
  48
  49 static inline struct journal_buf *
  50 journal_seq_to_buf(struct journal *j, u64 seq)
  51 {
  52         struct journal_buf *buf = NULL;
  53
  54         EBUG_ON(seq > journal_cur_seq(j));
  55
  56         if (journal_seq_unwritten(j, seq)) {
  57                 buf = j->buf + (seq & JOURNAL_BUF_MASK);
  58                 EBUG_ON(le64_to_cpu(buf->data->seq) != seq);
  59         }
  60         return buf;
  61 }
  62
  63 static void journal_pin_list_init(struct journal_entry_pin_list *p, int count)
  64 {
  65         unsigned i;
  66         for (i = 0; i < ARRAY_SIZE(p->list); i++)
  67                 INIT_LIST_HEAD(&p->list[i]);
  68         INIT_LIST_HEAD(&p->flushed);
  69         atomic_set(&p->count, count);
  70         p->devs.nr = 0;
  71 }
  72
  73 /*
  74  * Detect stuck journal conditions and trigger shutdown. Technically the journal
  75  * can end up stuck for a variety of reasons, such as a blocked I/O, journal
  76  * reservation lockup, etc. Since this is a fatal error with potentially
  77  * unpredictable characteristics, we want to be fairly conservative before we
  78  * decide to shut things down.
  79  *
  80  * Consider the journal stuck when it appears full with no ability to commit
  81  * btree transactions, to discard journal buckets, nor acquire priority
  82  * (reserved watermark) reservation.
  83  */
  84 static inline bool
  85 journal_error_check_stuck(struct journal *j, int error, unsigned flags)
  86 {
  87         struct bch_fs *c = container_of(j, struct bch_fs, journal);
  88         bool stuck = false;
  89         struct printbuf buf = PRINTBUF;
  90
  91         if (!(error == JOURNAL_ERR_journal_full ||
  92               error == JOURNAL_ERR_journal_pin_full) ||
  93             nr_unwritten_journal_entries(j) ||
  94             (flags & BCH_WATERMARK_MASK) != BCH_WATERMARK_reclaim)
  95                 return stuck;
  96
  97         spin_lock(&j->lock);
  98
  99         if (j->can_discard) {
 100                 spin_unlock(&j->lock);
 101                 return stuck;
 102         }
 103
 104         stuck = true;
 105
 106         /*
 107          * The journal shutdown path will set ->err_seq, but do it here first to
 108          * serialize against concurrent failures and avoid duplicate error
 109          * reports.
 110          */
 111         if (j->err_seq) {
 112                 spin_unlock(&j->lock);
 113                 return stuck;
 114         }
 115         j->err_seq = journal_cur_seq(j);
 116         spin_unlock(&j->lock);
 117
 118         bch_err(c, "Journal stuck! Hava a pre-reservation but journal full (error %s)",
 119                 bch2_journal_errors[error]);
 120         bch2_journal_debug_to_text(&buf, j);
 121         bch_err(c, "%s", buf.buf);
 122
 123         printbuf_reset(&buf);
 124         bch2_journal_pins_to_text(&buf, j);
 125         bch_err(c, "Journal pins:\n%s", buf.buf);
 126         printbuf_exit(&buf);
 127
 128         bch2_fatal_error(c);
 129         dump_stack();
 130
 131         return stuck;
 132 }
 133
 134 /* journal entry close/open: */
 135
 136 void __bch2_journal_buf_put(struct journal *j)
 137 {
 138         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 139
 140         closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
 141 }
 142
 143 /*
 144  * Returns true if journal entry is now closed:
 145  *
 146  * We don't close a journal_buf until the next journal_buf is finished writing,
 147  * and can be opened again - this also initializes the next journal_buf:
 148  */
 149 static void __journal_entry_close(struct journal *j, unsigned closed_val)
 150 {
 151         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 152         struct journal_buf *buf = journal_cur_buf(j);
 153         union journal_res_state old, new;
 154         u64 v = atomic64_read(&j->reservations.counter);
 155         unsigned sectors;
 156
 157         BUG_ON(closed_val != JOURNAL_ENTRY_CLOSED_VAL &&
 158                closed_val != JOURNAL_ENTRY_ERROR_VAL);
 159
 160         lockdep_assert_held(&j->lock);
 161
 162         do {
 163                 old.v = new.v = v;
 164                 new.cur_entry_offset = closed_val;
 165
 166                 if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL ||
 167                     old.cur_entry_offset == new.cur_entry_offset)
 168                         return;
 169         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 170                                        old.v, new.v)) != old.v);
 171
 172         if (!__journal_entry_is_open(old))
 173                 return;
 174
 175         /* Close out old buffer: */
 176         buf->data->u64s         = cpu_to_le32(old.cur_entry_offset);
 177
 178         sectors = vstruct_blocks_plus(buf->data, c->block_bits,
 179                                       buf->u64s_reserved) << c->block_bits;
 180         BUG_ON(sectors > buf->sectors);
 181         buf->sectors = sectors;
 182
 183         /*
 184          * We have to set last_seq here, _before_ opening a new journal entry:
 185          *
 186          * A threads may replace an old pin with a new pin on their current
 187          * journal reservation - the expectation being that the journal will
 188          * contain either what the old pin protected or what the new pin
 189          * protects.
 190          *
 191          * After the old pin is dropped journal_last_seq() won't include the old
 192          * pin, so we can only write the updated last_seq on the entry that
 193          * contains whatever the new pin protects.
 194          *
 195          * Restated, we can _not_ update last_seq for a given entry if there
 196          * could be a newer entry open with reservations/pins that have been
 197          * taken against it.
 198          *
 199          * Hence, we want update/set last_seq on the current journal entry right
 200          * before we open a new one:
 201          */
 202         buf->last_seq           = journal_last_seq(j);
 203         buf->data->last_seq     = cpu_to_le64(buf->last_seq);
 204         BUG_ON(buf->last_seq > le64_to_cpu(buf->data->seq));
 205
 206         __bch2_journal_pin_put(j, le64_to_cpu(buf->data->seq));
 207
 208         cancel_delayed_work(&j->write_work);
 209
 210         bch2_journal_space_available(j);
 211
 212         bch2_journal_buf_put(j, old.idx);
 213 }
 214
 215 void bch2_journal_halt(struct journal *j)
 216 {
 217         spin_lock(&j->lock);
 218         __journal_entry_close(j, JOURNAL_ENTRY_ERROR_VAL);
 219         if (!j->err_seq)
 220                 j->err_seq = journal_cur_seq(j);
 221         journal_wake(j);
 222         spin_unlock(&j->lock);
 223 }
 224
 225 static bool journal_entry_want_write(struct journal *j)
 226 {
 227         bool ret = !journal_entry_is_open(j) ||
 228                 journal_cur_seq(j) == journal_last_unwritten_seq(j);
 229
 230         /* Don't close it yet if we already have a write in flight: */
 231         if (ret)
 232                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 233         else if (nr_unwritten_journal_entries(j)) {
 234                 struct journal_buf *buf = journal_cur_buf(j);
 235
 236                 if (!buf->flush_time) {
 237                         buf->flush_time = local_clock() ?: 1;
 238                         buf->expires = jiffies;
 239                 }
 240         }
 241
 242         return ret;
 243 }
 244
 245 static bool journal_entry_close(struct journal *j)
 246 {
 247         bool ret;
 248
 249         spin_lock(&j->lock);
 250         ret = journal_entry_want_write(j);
 251         spin_unlock(&j->lock);
 252
 253         return ret;
 254 }
 255
 256 /*
 257  * should _only_ called from journal_res_get() - when we actually want a
 258  * journal reservation - journal entry is open means journal is dirty:
 259  */
 260 static int journal_entry_open(struct journal *j)
 261 {
 262         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 263         struct journal_buf *buf = j->buf +
 264                 ((journal_cur_seq(j) + 1) & JOURNAL_BUF_MASK);
 265         union journal_res_state old, new;
 266         int u64s;
 267         u64 v;
 268
 269         lockdep_assert_held(&j->lock);
 270         BUG_ON(journal_entry_is_open(j));
 271         BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
 272
 273         if (j->blocked)
 274                 return JOURNAL_ERR_blocked;
 275
 276         if (j->cur_entry_error)
 277                 return j->cur_entry_error;
 278
 279         if (bch2_journal_error(j))
 280                 return JOURNAL_ERR_insufficient_devices; /* -EROFS */
 281
 282         if (!fifo_free(&j->pin))
 283                 return JOURNAL_ERR_journal_pin_full;
 284
 285         if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf))
 286                 return JOURNAL_ERR_max_in_flight;
 287
 288         BUG_ON(!j->cur_entry_sectors);
 289
 290         buf->expires            =
 291                 (journal_cur_seq(j) == j->flushed_seq_ondisk
 292                  ? jiffies
 293                  : j->last_flush_write) +
 294                 msecs_to_jiffies(c->opts.journal_flush_delay);
 295
 296         buf->u64s_reserved      = j->entry_u64s_reserved;
 297         buf->disk_sectors       = j->cur_entry_sectors;
 298         buf->sectors            = min(buf->disk_sectors, buf->buf_size >> 9);
 299
 300         u64s = (int) (buf->sectors << 9) / sizeof(u64) -
 301                 journal_entry_overhead(j);
 302         u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 303
 304         if (u64s <= (ssize_t) j->early_journal_entries.nr)
 305                 return JOURNAL_ERR_journal_full;
 306
 307         if (fifo_empty(&j->pin) && j->reclaim_thread)
 308                 wake_up_process(j->reclaim_thread);
 309
 310         /*
 311          * The fifo_push() needs to happen at the same time as j->seq is
 312          * incremented for journal_last_seq() to be calculated correctly
 313          */
 314         atomic64_inc(&j->seq);
 315         journal_pin_list_init(fifo_push_ref(&j->pin), 1);
 316
 317         BUG_ON(j->buf + (journal_cur_seq(j) & JOURNAL_BUF_MASK) != buf);
 318
 319         bkey_extent_init(&buf->key);
 320         buf->noflush    = false;
 321         buf->must_flush = false;
 322         buf->separate_flush = false;
 323         buf->flush_time = 0;
 324
 325         memset(buf->data, 0, sizeof(*buf->data));
 326         buf->data->seq  = cpu_to_le64(journal_cur_seq(j));
 327         buf->data->u64s = 0;
 328
 329         if (j->early_journal_entries.nr) {
 330                 memcpy(buf->data->_data, j->early_journal_entries.data,
 331                        j->early_journal_entries.nr * sizeof(u64));
 332                 le32_add_cpu(&buf->data->u64s, j->early_journal_entries.nr);
 333         }
 334
 335         /*
 336          * Must be set before marking the journal entry as open:
 337          */
 338         j->cur_entry_u64s = u64s;
 339
 340         v = atomic64_read(&j->reservations.counter);
 341         do {
 342                 old.v = new.v = v;
 343
 344                 BUG_ON(old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL);
 345
 346                 new.idx++;
 347                 BUG_ON(journal_state_count(new, new.idx));
 348                 BUG_ON(new.idx != (journal_cur_seq(j) & JOURNAL_BUF_MASK));
 349
 350                 journal_state_inc(&new);
 351
 352                 /* Handle any already added entries */
 353                 new.cur_entry_offset = le32_to_cpu(buf->data->u64s);
 354         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 355                                        old.v, new.v)) != old.v);
 356
 357         if (j->res_get_blocked_start)
 358                 bch2_time_stats_update(j->blocked_time,
 359                                        j->res_get_blocked_start);
 360         j->res_get_blocked_start = 0;
 361
 362         mod_delayed_work(c->io_complete_wq,
 363                          &j->write_work,
 364                          msecs_to_jiffies(c->opts.journal_flush_delay));
 365         journal_wake(j);
 366
 367         if (j->early_journal_entries.nr)
 368                 darray_exit(&j->early_journal_entries);
 369         return 0;
 370 }
 371
 372 static bool journal_quiesced(struct journal *j)
 373 {
 374         bool ret = atomic64_read(&j->seq) == j->seq_ondisk;
 375
 376         if (!ret)
 377                 journal_entry_close(j);
 378         return ret;
 379 }
 380
 381 static void journal_quiesce(struct journal *j)
 382 {
 383         wait_event(j->wait, journal_quiesced(j));
 384 }
 385
 386 static void journal_write_work(struct work_struct *work)
 387 {
 388         struct journal *j = container_of(work, struct journal, write_work.work);
 389         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 390         long delta;
 391
 392         spin_lock(&j->lock);
 393         if (!__journal_entry_is_open(j->reservations))
 394                 goto unlock;
 395
 396         delta = journal_cur_buf(j)->expires - jiffies;
 397
 398         if (delta > 0)
 399                 mod_delayed_work(c->io_complete_wq, &j->write_work, delta);
 400         else
 401                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 402 unlock:
 403         spin_unlock(&j->lock);
 404 }
 405
 406 static int __journal_res_get(struct journal *j, struct journal_res *res,
 407                              unsigned flags)
 408 {
 409         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 410         struct journal_buf *buf;
 411         bool can_discard;
 412         int ret;
 413 retry:
 414         if (journal_res_get_fast(j, res, flags))
 415                 return 0;
 416
 417         if (bch2_journal_error(j))
 418                 return -BCH_ERR_erofs_journal_err;
 419
 420         spin_lock(&j->lock);
 421
 422         /* check once more in case somebody else shut things down... */
 423         if (bch2_journal_error(j)) {
 424                 spin_unlock(&j->lock);
 425                 return -BCH_ERR_erofs_journal_err;
 426         }
 427
 428         /*
 429          * Recheck after taking the lock, so we don't race with another thread
 430          * that just did journal_entry_open() and call journal_entry_close()
 431          * unnecessarily
 432          */
 433         if (journal_res_get_fast(j, res, flags)) {
 434                 spin_unlock(&j->lock);
 435                 return 0;
 436         }
 437
 438         if ((flags & BCH_WATERMARK_MASK) < j->watermark) {
 439                 /*
 440                  * Don't want to close current journal entry, just need to
 441                  * invoke reclaim:
 442                  */
 443                 ret = JOURNAL_ERR_journal_full;
 444                 goto unlock;
 445         }
 446
 447         /*
 448          * If we couldn't get a reservation because the current buf filled up,
 449          * and we had room for a bigger entry on disk, signal that we want to
 450          * realloc the journal bufs:
 451          */
 452         buf = journal_cur_buf(j);
 453         if (journal_entry_is_open(j) &&
 454             buf->buf_size >> 9 < buf->disk_sectors &&
 455             buf->buf_size < JOURNAL_ENTRY_SIZE_MAX)
 456                 j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1);
 457
 458         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 459         ret = journal_entry_open(j);
 460
 461         if (ret == JOURNAL_ERR_max_in_flight)
 462                 trace_and_count(c, journal_entry_full, c);
 463 unlock:
 464         if ((ret && ret != JOURNAL_ERR_insufficient_devices) &&
 465             !j->res_get_blocked_start) {
 466                 j->res_get_blocked_start = local_clock() ?: 1;
 467                 trace_and_count(c, journal_full, c);
 468         }
 469
 470         can_discard = j->can_discard;
 471         spin_unlock(&j->lock);
 472
 473         if (!ret)
 474                 goto retry;
 475         if (journal_error_check_stuck(j, ret, flags))
 476                 ret = -BCH_ERR_journal_res_get_blocked;
 477
 478         /*
 479          * Journal is full - can't rely on reclaim from work item due to
 480          * freezing:
 481          */
 482         if ((ret == JOURNAL_ERR_journal_full ||
 483              ret == JOURNAL_ERR_journal_pin_full) &&
 484             !(flags & JOURNAL_RES_GET_NONBLOCK)) {
 485                 if (can_discard) {
 486                         bch2_journal_do_discards(j);
 487                         goto retry;
 488                 }
 489
 490                 if (mutex_trylock(&j->reclaim_lock)) {
 491                         bch2_journal_reclaim(j);
 492                         mutex_unlock(&j->reclaim_lock);
 493                 }
 494         }
 495
 496         return ret == JOURNAL_ERR_insufficient_devices
 497                 ? -BCH_ERR_erofs_journal_err
 498                 : -BCH_ERR_journal_res_get_blocked;
 499 }
 500
 501 /*
 502  * Essentially the entry function to the journaling code. When bcachefs is doing
 503  * a btree insert, it calls this function to get the current journal write.
 504  * Journal write is the structure used set up journal writes. The calling
 505  * function will then add its keys to the structure, queuing them for the next
 506  * write.
 507  *
 508  * To ensure forward progress, the current task must not be holding any
 509  * btree node write locks.
 510  */
 511 int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
 512                                   unsigned flags)
 513 {
 514         int ret;
 515
 516         closure_wait_event(&j->async_wait,
 517                    (ret = __journal_res_get(j, res, flags)) !=
 518                    -BCH_ERR_journal_res_get_blocked||
 519                    (flags & JOURNAL_RES_GET_NONBLOCK));
 520         return ret;
 521 }
 522
 523 /* journal_preres: */
 524
 525 static bool journal_preres_available(struct journal *j,
 526                                      struct journal_preres *res,
 527                                      unsigned new_u64s,
 528                                      unsigned flags)
 529 {
 530         bool ret = bch2_journal_preres_get_fast(j, res, new_u64s, flags, true);
 531
 532         if (!ret && mutex_trylock(&j->reclaim_lock)) {
 533                 bch2_journal_reclaim(j);
 534                 mutex_unlock(&j->reclaim_lock);
 535         }
 536
 537         return ret;
 538 }
 539
 540 int __bch2_journal_preres_get(struct journal *j,
 541                               struct journal_preres *res,
 542                               unsigned new_u64s,
 543                               unsigned flags)
 544 {
 545         int ret;
 546
 547         closure_wait_event(&j->preres_wait,
 548                    (ret = bch2_journal_error(j)) ||
 549                    journal_preres_available(j, res, new_u64s, flags));
 550         return ret;
 551 }
 552
 553 /* journal_entry_res: */
 554
 555 void bch2_journal_entry_res_resize(struct journal *j,
 556                                    struct journal_entry_res *res,
 557                                    unsigned new_u64s)
 558 {
 559         union journal_res_state state;
 560         int d = new_u64s - res->u64s;
 561
 562         spin_lock(&j->lock);
 563
 564         j->entry_u64s_reserved += d;
 565         if (d <= 0)
 566                 goto out;
 567
 568         j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d);
 569         smp_mb();
 570         state = READ_ONCE(j->reservations);
 571
 572         if (state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL &&
 573             state.cur_entry_offset > j->cur_entry_u64s) {
 574                 j->cur_entry_u64s += d;
 575                 /*
 576                  * Not enough room in current journal entry, have to flush it:
 577                  */
 578                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 579         } else {
 580                 journal_cur_buf(j)->u64s_reserved += d;
 581         }
 582 out:
 583         spin_unlock(&j->lock);
 584         res->u64s += d;
 585 }
 586
 587 /* journal flushing: */
 588
 589 /**
 590  * bch2_journal_flush_seq_async - wait for a journal entry to be written
 591  *
 592  * like bch2_journal_wait_on_seq, except that it triggers a write immediately if
 593  * necessary
 594  */
 595 int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 596                                  struct closure *parent)
 597 {
 598         struct journal_buf *buf;
 599         int ret = 0;
 600
 601         if (seq <= j->flushed_seq_ondisk)
 602                 return 1;
 603
 604         spin_lock(&j->lock);
 605
 606         if (WARN_ONCE(seq > journal_cur_seq(j),
 607                       "requested to flush journal seq %llu, but currently at %llu",
 608                       seq, journal_cur_seq(j)))
 609                 goto out;
 610
 611         /* Recheck under lock: */
 612         if (j->err_seq && seq >= j->err_seq) {
 613                 ret = -EIO;
 614                 goto out;
 615         }
 616
 617         if (seq <= j->flushed_seq_ondisk) {
 618                 ret = 1;
 619                 goto out;
 620         }
 621
 622         /* if seq was written, but not flushed - flush a newer one instead */
 623         seq = max(seq, journal_last_unwritten_seq(j));
 624
 625 recheck_need_open:
 626         if (seq > journal_cur_seq(j)) {
 627                 struct journal_res res = { 0 };
 628
 629                 if (journal_entry_is_open(j))
 630                         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 631
 632                 spin_unlock(&j->lock);
 633
 634                 ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 635                 if (ret)
 636                         return ret;
 637
 638                 seq = res.seq;
 639                 buf = j->buf + (seq & JOURNAL_BUF_MASK);
 640                 buf->must_flush = true;
 641
 642                 if (!buf->flush_time) {
 643                         buf->flush_time = local_clock() ?: 1;
 644                         buf->expires = jiffies;
 645                 }
 646
 647                 if (parent && !closure_wait(&buf->wait, parent))
 648                         BUG();
 649
 650                 bch2_journal_res_put(j, &res);
 651
 652                 spin_lock(&j->lock);
 653                 goto want_write;
 654         }
 655
 656         /*
 657          * if write was kicked off without a flush, flush the next sequence
 658          * number instead
 659          */
 660         buf = journal_seq_to_buf(j, seq);
 661         if (buf->noflush) {
 662                 seq++;
 663                 goto recheck_need_open;
 664         }
 665
 666         buf->must_flush = true;
 667
 668         if (parent && !closure_wait(&buf->wait, parent))
 669                 BUG();
 670 want_write:
 671         if (seq == journal_cur_seq(j))
 672                 journal_entry_want_write(j);
 673 out:
 674         spin_unlock(&j->lock);
 675         return ret;
 676 }
 677
 678 int bch2_journal_flush_seq(struct journal *j, u64 seq)
 679 {
 680         u64 start_time = local_clock();
 681         int ret, ret2;
 682
 683         /*
 684          * Don't update time_stats when @seq is already flushed:
 685          */
 686         if (seq <= j->flushed_seq_ondisk)
 687                 return 0;
 688
 689         ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)));
 690
 691         if (!ret)
 692                 bch2_time_stats_update(j->flush_seq_time, start_time);
 693
 694         return ret ?: ret2 < 0 ? ret2 : 0;
 695 }
 696
 697 /*
 698  * bch2_journal_flush_async - if there is an open journal entry, or a journal
 699  * still being written, write it and wait for the write to complete
 700  */
 701 void bch2_journal_flush_async(struct journal *j, struct closure *parent)
 702 {
 703         bch2_journal_flush_seq_async(j, atomic64_read(&j->seq), parent);
 704 }
 705
 706 int bch2_journal_flush(struct journal *j)
 707 {
 708         return bch2_journal_flush_seq(j, atomic64_read(&j->seq));
 709 }
 710
 711 /*
 712  * bch2_journal_noflush_seq - tell the journal not to issue any flushes before
 713  * @seq
 714  */
 715 bool bch2_journal_noflush_seq(struct journal *j, u64 seq)
 716 {
 717         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 718         u64 unwritten_seq;
 719         bool ret = false;
 720
 721         if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush)))
 722                 return false;
 723
 724         if (seq <= c->journal.flushed_seq_ondisk)
 725                 return false;
 726
 727         spin_lock(&j->lock);
 728         if (seq <= c->journal.flushed_seq_ondisk)
 729                 goto out;
 730
 731         for (unwritten_seq = journal_last_unwritten_seq(j);
 732              unwritten_seq < seq;
 733              unwritten_seq++) {
 734                 struct journal_buf *buf = journal_seq_to_buf(j, unwritten_seq);
 735
 736                 /* journal write is already in flight, and was a flush write: */
 737                 if (unwritten_seq == journal_last_unwritten_seq(j) && !buf->noflush)
 738                         goto out;
 739
 740                 buf->noflush = true;
 741         }
 742
 743         ret = true;
 744 out:
 745         spin_unlock(&j->lock);
 746         return ret;
 747 }
 748
 749 int bch2_journal_meta(struct journal *j)
 750 {
 751         struct journal_buf *buf;
 752         struct journal_res res;
 753         int ret;
 754
 755         memset(&res, 0, sizeof(res));
 756
 757         ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 758         if (ret)
 759                 return ret;
 760
 761         buf = j->buf + (res.seq & JOURNAL_BUF_MASK);
 762         buf->must_flush = true;
 763
 764         if (!buf->flush_time) {
 765                 buf->flush_time = local_clock() ?: 1;
 766                 buf->expires = jiffies;
 767         }
 768
 769         bch2_journal_res_put(j, &res);
 770
 771         return bch2_journal_flush_seq(j, res.seq);
 772 }
 773
 774 /* block/unlock the journal: */
 775
 776 void bch2_journal_unblock(struct journal *j)
 777 {
 778         spin_lock(&j->lock);
 779         j->blocked--;
 780         spin_unlock(&j->lock);
 781
 782         journal_wake(j);
 783 }
 784
 785 void bch2_journal_block(struct journal *j)
 786 {
 787         spin_lock(&j->lock);
 788         j->blocked++;
 789         spin_unlock(&j->lock);
 790
 791         journal_quiesce(j);
 792 }
 793
 794 /* allocate journal on a device: */
 795
 796 static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 797                                          bool new_fs, struct closure *cl)
 798 {
 799         struct bch_fs *c = ca->fs;
 800         struct journal_device *ja = &ca->journal;
 801         u64 *new_bucket_seq = NULL, *new_buckets = NULL;
 802         struct open_bucket **ob = NULL;
 803         long *bu = NULL;
 804         unsigned i, pos, nr_got = 0, nr_want = nr - ja->nr;
 805         int ret = 0;
 806
 807         BUG_ON(nr <= ja->nr);
 808
 809         bu              = kcalloc(nr_want, sizeof(*bu), GFP_KERNEL);
 810         ob              = kcalloc(nr_want, sizeof(*ob), GFP_KERNEL);
 811         new_buckets     = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 812         new_bucket_seq  = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 813         if (!bu || !ob || !new_buckets || !new_bucket_seq) {
 814                 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
 815                 goto err_free;
 816         }
 817
 818         for (nr_got = 0; nr_got < nr_want; nr_got++) {
 819                 if (new_fs) {
 820                         bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
 821                         if (bu[nr_got] < 0) {
 822                                 ret = -BCH_ERR_ENOSPC_bucket_alloc;
 823                                 break;
 824                         }
 825                 } else {
 826                         ob[nr_got] = bch2_bucket_alloc(c, ca, BCH_WATERMARK_normal, cl);
 827                         ret = PTR_ERR_OR_ZERO(ob[nr_got]);
 828                         if (ret)
 829                                 break;
 830
 831                         ret = bch2_trans_run(c,
 832                                 bch2_trans_mark_metadata_bucket(&trans, ca,
 833                                                 ob[nr_got]->bucket, BCH_DATA_journal,
 834                                                 ca->mi.bucket_size));
 835                         if (ret) {
 836                                 bch2_open_bucket_put(c, ob[nr_got]);
 837                                 bch_err(c, "error marking new journal buckets: %s", bch2_err_str(ret));
 838                                 break;
 839                         }
 840
 841                         bu[nr_got] = ob[nr_got]->bucket;
 842                 }
 843         }
 844
 845         if (!nr_got)
 846                 goto err_free;
 847
 848         /* Don't return an error if we successfully allocated some buckets: */
 849         ret = 0;
 850
 851         if (c) {
 852                 bch2_journal_flush_all_pins(&c->journal);
 853                 bch2_journal_block(&c->journal);
 854                 mutex_lock(&c->sb_lock);
 855         }
 856
 857         memcpy(new_buckets,     ja->buckets,    ja->nr * sizeof(u64));
 858         memcpy(new_bucket_seq,  ja->bucket_seq, ja->nr * sizeof(u64));
 859
 860         BUG_ON(ja->discard_idx > ja->nr);
 861
 862         pos = ja->discard_idx ?: ja->nr;
 863
 864         memmove(new_buckets + pos + nr_got,
 865                 new_buckets + pos,
 866                 sizeof(new_buckets[0]) * (ja->nr - pos));
 867         memmove(new_bucket_seq + pos + nr_got,
 868                 new_bucket_seq + pos,
 869                 sizeof(new_bucket_seq[0]) * (ja->nr - pos));
 870
 871         for (i = 0; i < nr_got; i++) {
 872                 new_buckets[pos + i] = bu[i];
 873                 new_bucket_seq[pos + i] = 0;
 874         }
 875
 876         nr = ja->nr + nr_got;
 877
 878         ret = bch2_journal_buckets_to_sb(c, ca, new_buckets, nr);
 879         if (ret)
 880                 goto err_unblock;
 881
 882         if (!new_fs)
 883                 bch2_write_super(c);
 884
 885         /* Commit: */
 886         if (c)
 887                 spin_lock(&c->journal.lock);
 888
 889         swap(new_buckets,       ja->buckets);
 890         swap(new_bucket_seq,    ja->bucket_seq);
 891         ja->nr = nr;
 892
 893         if (pos <= ja->discard_idx)
 894                 ja->discard_idx = (ja->discard_idx + nr_got) % ja->nr;
 895         if (pos <= ja->dirty_idx_ondisk)
 896                 ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + nr_got) % ja->nr;
 897         if (pos <= ja->dirty_idx)
 898                 ja->dirty_idx = (ja->dirty_idx + nr_got) % ja->nr;
 899         if (pos <= ja->cur_idx)
 900                 ja->cur_idx = (ja->cur_idx + nr_got) % ja->nr;
 901
 902         if (c)
 903                 spin_unlock(&c->journal.lock);
 904 err_unblock:
 905         if (c) {
 906                 bch2_journal_unblock(&c->journal);
 907                 mutex_unlock(&c->sb_lock);
 908         }
 909
 910         if (ret && !new_fs)
 911                 for (i = 0; i < nr_got; i++)
 912                         bch2_trans_run(c,
 913                                 bch2_trans_mark_metadata_bucket(&trans, ca,
 914                                                 bu[i], BCH_DATA_free, 0));
 915 err_free:
 916         if (!new_fs)
 917                 for (i = 0; i < nr_got; i++)
 918                         bch2_open_bucket_put(c, ob[i]);
 919
 920         kfree(new_bucket_seq);
 921         kfree(new_buckets);
 922         kfree(ob);
 923         kfree(bu);
 924         return ret;
 925 }
 926
 927 /*
 928  * Allocate more journal space at runtime - not currently making use if it, but
 929  * the code works:
 930  */
 931 int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 932                                 unsigned nr)
 933 {
 934         struct journal_device *ja = &ca->journal;
 935         struct closure cl;
 936         int ret = 0;
 937
 938         closure_init_stack(&cl);
 939
 940         down_write(&c->state_lock);
 941
 942         /* don't handle reducing nr of buckets yet: */
 943         if (nr < ja->nr)
 944                 goto unlock;
 945
 946         while (ja->nr < nr) {
 947                 struct disk_reservation disk_res = { 0, 0 };
 948
 949                 /*
 950                  * note: journal buckets aren't really counted as _sectors_ used yet, so
 951                  * we don't need the disk reservation to avoid the BUG_ON() in buckets.c
 952                  * when space used goes up without a reservation - but we do need the
 953                  * reservation to ensure we'll actually be able to allocate:
 954                  *
 955                  * XXX: that's not right, disk reservations only ensure a
 956                  * filesystem-wide allocation will succeed, this is a device
 957                  * specific allocation - we can hang here:
 958                  */
 959
 960                 ret = bch2_disk_reservation_get(c, &disk_res,
 961                                                 bucket_to_sector(ca, nr - ja->nr), 1, 0);
 962                 if (ret)
 963                         break;
 964
 965                 ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl);
 966
 967                 bch2_disk_reservation_put(c, &disk_res);
 968
 969                 closure_sync(&cl);
 970
 971                 if (ret && ret != -BCH_ERR_bucket_alloc_blocked)
 972                         break;
 973         }
 974
 975         if (ret)
 976                 bch_err_fn(c, ret);
 977 unlock:
 978         up_write(&c->state_lock);
 979         return ret;
 980 }
 981
 982 int bch2_dev_journal_alloc(struct bch_dev *ca)
 983 {
 984         unsigned nr;
 985         int ret;
 986
 987         if (dynamic_fault("bcachefs:add:journal_alloc")) {
 988                 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
 989                 goto err;
 990         }
 991
 992         /* 1/128th of the device by default: */
 993         nr = ca->mi.nbuckets >> 7;
 994
 995         /*
 996          * clamp journal size to 8192 buckets or 8GB (in sectors), whichever
 997          * is smaller:
 998          */
 999         nr = clamp_t(unsigned, nr,
1000                      BCH_JOURNAL_BUCKETS_MIN,
1001                      min(1 << 13,
1002                          (1 << 24) / ca->mi.bucket_size));
1003
1004         ret = __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
1005 err:
1006         if (ret)
1007                 bch_err_fn(ca, ret);
1008         return ret;
1009 }
1010
1011 /* startup/shutdown: */
1012
1013 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
1014 {
1015         bool ret = false;
1016         u64 seq;
1017
1018         spin_lock(&j->lock);
1019         for (seq = journal_last_unwritten_seq(j);
1020              seq <= journal_cur_seq(j) && !ret;
1021              seq++) {
1022                 struct journal_buf *buf = journal_seq_to_buf(j, seq);
1023
1024                 if (bch2_bkey_has_device_c(bkey_i_to_s_c(&buf->key), dev_idx))
1025                         ret = true;
1026         }
1027         spin_unlock(&j->lock);
1028
1029         return ret;
1030 }
1031
1032 void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca)
1033 {
1034         wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx));
1035 }
1036
1037 void bch2_fs_journal_stop(struct journal *j)
1038 {
1039         bch2_journal_reclaim_stop(j);
1040         bch2_journal_flush_all_pins(j);
1041
1042         wait_event(j->wait, journal_entry_close(j));
1043
1044         /*
1045          * Always write a new journal entry, to make sure the clock hands are up
1046          * to date (and match the superblock)
1047          */
1048         bch2_journal_meta(j);
1049
1050         journal_quiesce(j);
1051
1052         BUG_ON(!bch2_journal_error(j) &&
1053                test_bit(JOURNAL_REPLAY_DONE, &j->flags) &&
1054                j->last_empty_seq != journal_cur_seq(j));
1055
1056         cancel_delayed_work_sync(&j->write_work);
1057 }
1058
1059 int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
1060 {
1061         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1062         struct journal_entry_pin_list *p;
1063         struct journal_replay *i, **_i;
1064         struct genradix_iter iter;
1065         bool had_entries = false;
1066         unsigned ptr;
1067         u64 last_seq = cur_seq, nr, seq;
1068
1069         genradix_for_each_reverse(&c->journal_entries, iter, _i) {
1070                 i = *_i;
1071
1072                 if (!i || i->ignore)
1073                         continue;
1074
1075                 last_seq = le64_to_cpu(i->j.last_seq);
1076                 break;
1077         }
1078
1079         nr = cur_seq - last_seq;
1080
1081         if (nr + 1 > j->pin.size) {
1082                 free_fifo(&j->pin);
1083                 init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL);
1084                 if (!j->pin.data) {
1085                         bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
1086                         return -BCH_ERR_ENOMEM_journal_pin_fifo;
1087                 }
1088         }
1089
1090         j->replay_journal_seq   = last_seq;
1091         j->replay_journal_seq_end = cur_seq;
1092         j->last_seq_ondisk      = last_seq;
1093         j->flushed_seq_ondisk   = cur_seq - 1;
1094         j->seq_ondisk           = cur_seq - 1;
1095         j->pin.front            = last_seq;
1096         j->pin.back             = cur_seq;
1097         atomic64_set(&j->seq, cur_seq - 1);
1098
1099         fifo_for_each_entry_ptr(p, &j->pin, seq)
1100                 journal_pin_list_init(p, 1);
1101
1102         genradix_for_each(&c->journal_entries, iter, _i) {
1103                 i = *_i;
1104
1105                 if (!i || i->ignore)
1106                         continue;
1107
1108                 seq = le64_to_cpu(i->j.seq);
1109                 BUG_ON(seq >= cur_seq);
1110
1111                 if (seq < last_seq)
1112                         continue;
1113
1114                 if (journal_entry_empty(&i->j))
1115                         j->last_empty_seq = le64_to_cpu(i->j.seq);
1116
1117                 p = journal_seq_pin(j, seq);
1118
1119                 p->devs.nr = 0;
1120                 for (ptr = 0; ptr < i->nr_ptrs; ptr++)
1121                         bch2_dev_list_add_dev(&p->devs, i->ptrs[ptr].dev);
1122
1123                 had_entries = true;
1124         }
1125
1126         if (!had_entries)
1127                 j->last_empty_seq = cur_seq;
1128
1129         spin_lock(&j->lock);
1130
1131         set_bit(JOURNAL_STARTED, &j->flags);
1132         j->last_flush_write = jiffies;
1133
1134         j->reservations.idx = j->reservations.unwritten_idx = journal_cur_seq(j);
1135         j->reservations.unwritten_idx++;
1136
1137         c->last_bucket_seq_cleanup = journal_cur_seq(j);
1138
1139         bch2_journal_space_available(j);
1140         spin_unlock(&j->lock);
1141
1142         return bch2_journal_reclaim_start(j);
1143 }
1144
1145 /* init/exit: */
1146
1147 void bch2_dev_journal_exit(struct bch_dev *ca)
1148 {
1149         kfree(ca->journal.bio);
1150         kfree(ca->journal.buckets);
1151         kfree(ca->journal.bucket_seq);
1152
1153         ca->journal.bio         = NULL;
1154         ca->journal.buckets     = NULL;
1155         ca->journal.bucket_seq  = NULL;
1156 }
1157
1158 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
1159 {
1160         struct journal_device *ja = &ca->journal;
1161         struct bch_sb_field_journal *journal_buckets =
1162                 bch2_sb_get_journal(sb);
1163         struct bch_sb_field_journal_v2 *journal_buckets_v2 =
1164                 bch2_sb_get_journal_v2(sb);
1165         unsigned i, nr_bvecs;
1166
1167         ja->nr = 0;
1168
1169         if (journal_buckets_v2) {
1170                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1171
1172                 for (i = 0; i < nr; i++)
1173                         ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr);
1174         } else if (journal_buckets) {
1175                 ja->nr = bch2_nr_journal_buckets(journal_buckets);
1176         }
1177
1178         ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1179         if (!ja->bucket_seq)
1180                 return -BCH_ERR_ENOMEM_dev_journal_init;
1181
1182         nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
1183
1184         ca->journal.bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1185         if (!ca->journal.bio)
1186                 return -BCH_ERR_ENOMEM_dev_journal_init;
1187
1188         bio_init(ca->journal.bio, NULL, ca->journal.bio->bi_inline_vecs, nr_bvecs, 0);
1189
1190         ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1191         if (!ja->buckets)
1192                 return -BCH_ERR_ENOMEM_dev_journal_init;
1193
1194         if (journal_buckets_v2) {
1195                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1196                 unsigned j, dst = 0;
1197
1198                 for (i = 0; i < nr; i++)
1199                         for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
1200                                 ja->buckets[dst++] =
1201                                         le64_to_cpu(journal_buckets_v2->d[i].start) + j;
1202         } else if (journal_buckets) {
1203                 for (i = 0; i < ja->nr; i++)
1204                         ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
1205         }
1206
1207         return 0;
1208 }
1209
1210 void bch2_fs_journal_exit(struct journal *j)
1211 {
1212         unsigned i;
1213
1214         darray_exit(&j->early_journal_entries);
1215
1216         for (i = 0; i < ARRAY_SIZE(j->buf); i++)
1217                 kvpfree(j->buf[i].data, j->buf[i].buf_size);
1218         free_fifo(&j->pin);
1219 }
1220
1221 int bch2_fs_journal_init(struct journal *j)
1222 {
1223         static struct lock_class_key res_key;
1224         unsigned i;
1225
1226         spin_lock_init(&j->lock);
1227         spin_lock_init(&j->err_lock);
1228         init_waitqueue_head(&j->wait);
1229         INIT_DELAYED_WORK(&j->write_work, journal_write_work);
1230         init_waitqueue_head(&j->reclaim_wait);
1231         init_waitqueue_head(&j->pin_flush_wait);
1232         mutex_init(&j->reclaim_lock);
1233         mutex_init(&j->discard_lock);
1234
1235         lockdep_init_map(&j->res_map, "journal res", &res_key, 0);
1236
1237         atomic64_set(&j->reservations.counter,
1238                 ((union journal_res_state)
1239                  { .cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL }).v);
1240
1241         if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)))
1242                 return -BCH_ERR_ENOMEM_journal_pin_fifo;
1243
1244         for (i = 0; i < ARRAY_SIZE(j->buf); i++) {
1245                 j->buf[i].buf_size = JOURNAL_ENTRY_SIZE_MIN;
1246                 j->buf[i].data = kvpmalloc(j->buf[i].buf_size, GFP_KERNEL);
1247                 if (!j->buf[i].data)
1248                         return -BCH_ERR_ENOMEM_journal_buf;
1249         }
1250
1251         j->pin.front = j->pin.back = 1;
1252         return 0;
1253 }
1254
1255 /* debug: */
1256
1257 void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1258 {
1259         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1260         union journal_res_state s;
1261         struct bch_dev *ca;
1262         unsigned long now = jiffies;
1263         u64 seq;
1264         unsigned i;
1265
1266         if (!out->nr_tabstops)
1267                 printbuf_tabstop_push(out, 24);
1268         out->atomic++;
1269
1270         rcu_read_lock();
1271         s = READ_ONCE(j->reservations);
1272
1273         prt_printf(out, "dirty journal entries:\t%llu/%llu\n",  fifo_used(&j->pin), j->pin.size);
1274         prt_printf(out, "seq:\t\t\t%llu\n",                     journal_cur_seq(j));
1275         prt_printf(out, "seq_ondisk:\t\t%llu\n",                j->seq_ondisk);
1276         prt_printf(out, "last_seq:\t\t%llu\n",          journal_last_seq(j));
1277         prt_printf(out, "last_seq_ondisk:\t%llu\n",             j->last_seq_ondisk);
1278         prt_printf(out, "flushed_seq_ondisk:\t%llu\n",  j->flushed_seq_ondisk);
1279         prt_printf(out, "prereserved:\t\t%u/%u\n",              j->prereserved.reserved, j->prereserved.remaining);
1280         prt_printf(out, "watermark:\t\t%s\n",           bch2_watermarks[j->watermark]);
1281         prt_printf(out, "each entry reserved:\t%u\n",   j->entry_u64s_reserved);
1282         prt_printf(out, "nr flush writes:\t%llu\n",             j->nr_flush_writes);
1283         prt_printf(out, "nr noflush writes:\t%llu\n",   j->nr_noflush_writes);
1284         prt_printf(out, "nr direct reclaim:\t%llu\n",   j->nr_direct_reclaim);
1285         prt_printf(out, "nr background reclaim:\t%llu\n",       j->nr_background_reclaim);
1286         prt_printf(out, "reclaim kicked:\t\t%u\n",              j->reclaim_kicked);
1287         prt_printf(out, "reclaim runs in:\t%u ms\n",    time_after(j->next_reclaim, now)
1288                ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0);
1289         prt_printf(out, "current entry sectors:\t%u\n", j->cur_entry_sectors);
1290         prt_printf(out, "current entry error:\t%s\n",   bch2_journal_errors[j->cur_entry_error]);
1291         prt_printf(out, "current entry:\t\t");
1292
1293         switch (s.cur_entry_offset) {
1294         case JOURNAL_ENTRY_ERROR_VAL:
1295                 prt_printf(out, "error");
1296                 break;
1297         case JOURNAL_ENTRY_CLOSED_VAL:
1298                 prt_printf(out, "closed");
1299                 break;
1300         default:
1301                 prt_printf(out, "%u/%u", s.cur_entry_offset, j->cur_entry_u64s);
1302                 break;
1303         }
1304
1305         prt_newline(out);
1306
1307         for (seq = journal_cur_seq(j);
1308              seq >= journal_last_unwritten_seq(j);
1309              --seq) {
1310                 i = seq & JOURNAL_BUF_MASK;
1311
1312                 prt_printf(out, "unwritten entry:");
1313                 prt_tab(out);
1314                 prt_printf(out, "%llu", seq);
1315                 prt_newline(out);
1316                 printbuf_indent_add(out, 2);
1317
1318                 prt_printf(out, "refcount:");
1319                 prt_tab(out);
1320                 prt_printf(out, "%u", journal_state_count(s, i));
1321                 prt_newline(out);
1322
1323                 prt_printf(out, "sectors:");
1324                 prt_tab(out);
1325                 prt_printf(out, "%u", j->buf[i].sectors);
1326                 prt_newline(out);
1327
1328                 prt_printf(out, "expires");
1329                 prt_tab(out);
1330                 prt_printf(out, "%li jiffies", j->buf[i].expires - jiffies);
1331                 prt_newline(out);
1332
1333                 printbuf_indent_sub(out, 2);
1334         }
1335
1336         prt_printf(out,
1337                "replay done:\t\t%i\n",
1338                test_bit(JOURNAL_REPLAY_DONE,    &j->flags));
1339
1340         prt_printf(out, "space:\n");
1341         prt_printf(out, "\tdiscarded\t%u:%u\n",
1342                j->space[journal_space_discarded].next_entry,
1343                j->space[journal_space_discarded].total);
1344         prt_printf(out, "\tclean ondisk\t%u:%u\n",
1345                j->space[journal_space_clean_ondisk].next_entry,
1346                j->space[journal_space_clean_ondisk].total);
1347         prt_printf(out, "\tclean\t\t%u:%u\n",
1348                j->space[journal_space_clean].next_entry,
1349                j->space[journal_space_clean].total);
1350         prt_printf(out, "\ttotal\t\t%u:%u\n",
1351                j->space[journal_space_total].next_entry,
1352                j->space[journal_space_total].total);
1353
1354         for_each_member_device_rcu(ca, c, i,
1355                                    &c->rw_devs[BCH_DATA_journal]) {
1356                 struct journal_device *ja = &ca->journal;
1357
1358                 if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d))
1359                         continue;
1360
1361                 if (!ja->nr)
1362                         continue;
1363
1364                 prt_printf(out, "dev %u:\n",            i);
1365                 prt_printf(out, "\tnr\t\t%u\n",         ja->nr);
1366                 prt_printf(out, "\tbucket size\t%u\n",  ca->mi.bucket_size);
1367                 prt_printf(out, "\tavailable\t%u:%u\n", bch2_journal_dev_buckets_available(j, ja, journal_space_discarded), ja->sectors_free);
1368                 prt_printf(out, "\tdiscard_idx\t%u\n",  ja->discard_idx);
1369                 prt_printf(out, "\tdirty_ondisk\t%u (seq %llu)\n", ja->dirty_idx_ondisk,        ja->bucket_seq[ja->dirty_idx_ondisk]);
1370                 prt_printf(out, "\tdirty_idx\t%u (seq %llu)\n", ja->dirty_idx,          ja->bucket_seq[ja->dirty_idx]);
1371                 prt_printf(out, "\tcur_idx\t\t%u (seq %llu)\n", ja->cur_idx,            ja->bucket_seq[ja->cur_idx]);
1372         }
1373
1374         rcu_read_unlock();
1375
1376         --out->atomic;
1377 }
1378
1379 void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1380 {
1381         spin_lock(&j->lock);
1382         __bch2_journal_debug_to_text(out, j);
1383         spin_unlock(&j->lock);
1384 }
1385
1386 bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq)
1387 {
1388         struct journal_entry_pin_list *pin_list;
1389         struct journal_entry_pin *pin;
1390         unsigned i;
1391
1392         spin_lock(&j->lock);
1393         *seq = max(*seq, j->pin.front);
1394
1395         if (*seq >= j->pin.back) {
1396                 spin_unlock(&j->lock);
1397                 return true;
1398         }
1399
1400         out->atomic++;
1401
1402         pin_list = journal_seq_pin(j, *seq);
1403
1404         prt_printf(out, "%llu: count %u", *seq, atomic_read(&pin_list->count));
1405         prt_newline(out);
1406         printbuf_indent_add(out, 2);
1407
1408         for (i = 0; i < ARRAY_SIZE(pin_list->list); i++)
1409                 list_for_each_entry(pin, &pin_list->list[i], list) {
1410                         prt_printf(out, "\t%px %ps", pin, pin->flush);
1411                         prt_newline(out);
1412                 }
1413
1414         if (!list_empty(&pin_list->flushed)) {
1415                 prt_printf(out, "flushed:");
1416                 prt_newline(out);
1417         }
1418
1419         list_for_each_entry(pin, &pin_list->flushed, list) {
1420                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1421                 prt_newline(out);
1422         }
1423
1424         printbuf_indent_sub(out, 2);
1425
1426         --out->atomic;
1427         spin_unlock(&j->lock);
1428
1429         return false;
1430 }
1431
1432 void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
1433 {
1434         u64 seq = 0;
1435
1436         while (!bch2_journal_seq_pins_to_text(out, j, &seq))
1437                 seq++;
1438 }