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
  67         for (i = 0; i < ARRAY_SIZE(p->list); i++)
  68                 INIT_LIST_HEAD(&p->list[i]);
  69         INIT_LIST_HEAD(&p->flushed);
  70         atomic_set(&p->count, count);
  71         p->devs.nr = 0;
  72 }
  73
  74 /*
  75  * Detect stuck journal conditions and trigger shutdown. Technically the journal
  76  * can end up stuck for a variety of reasons, such as a blocked I/O, journal
  77  * reservation lockup, etc. Since this is a fatal error with potentially
  78  * unpredictable characteristics, we want to be fairly conservative before we
  79  * decide to shut things down.
  80  *
  81  * Consider the journal stuck when it appears full with no ability to commit
  82  * btree transactions, to discard journal buckets, nor acquire priority
  83  * (reserved watermark) reservation.
  84  */
  85 static inline bool
  86 journal_error_check_stuck(struct journal *j, int error, unsigned flags)
  87 {
  88         struct bch_fs *c = container_of(j, struct bch_fs, journal);
  89         bool stuck = false;
  90         struct printbuf buf = PRINTBUF;
  91
  92         if (!(error == JOURNAL_ERR_journal_full ||
  93               error == JOURNAL_ERR_journal_pin_full) ||
  94             nr_unwritten_journal_entries(j) ||
  95             (flags & BCH_WATERMARK_MASK) != BCH_WATERMARK_reclaim)
  96                 return stuck;
  97
  98         spin_lock(&j->lock);
  99
 100         if (j->can_discard) {
 101                 spin_unlock(&j->lock);
 102                 return stuck;
 103         }
 104
 105         stuck = true;
 106
 107         /*
 108          * The journal shutdown path will set ->err_seq, but do it here first to
 109          * serialize against concurrent failures and avoid duplicate error
 110          * reports.
 111          */
 112         if (j->err_seq) {
 113                 spin_unlock(&j->lock);
 114                 return stuck;
 115         }
 116         j->err_seq = journal_cur_seq(j);
 117         spin_unlock(&j->lock);
 118
 119         bch_err(c, "Journal stuck! Hava a pre-reservation but journal full (error %s)",
 120                 bch2_journal_errors[error]);
 121         bch2_journal_debug_to_text(&buf, j);
 122         bch_err(c, "%s", buf.buf);
 123
 124         printbuf_reset(&buf);
 125         bch2_journal_pins_to_text(&buf, j);
 126         bch_err(c, "Journal pins:\n%s", buf.buf);
 127         printbuf_exit(&buf);
 128
 129         bch2_fatal_error(c);
 130         dump_stack();
 131
 132         return stuck;
 133 }
 134
 135 /*
 136  * Final processing when the last reference of a journal buffer has been
 137  * dropped. Drop the pin list reference acquired at journal entry open and write
 138  * the buffer, if requested.
 139  */
 140 void bch2_journal_buf_put_final(struct journal *j, u64 seq, bool write)
 141 {
 142         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 143
 144         lockdep_assert_held(&j->lock);
 145
 146         if (__bch2_journal_pin_put(j, seq))
 147                 bch2_journal_reclaim_fast(j);
 148         if (write)
 149                 closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
 150 }
 151
 152 /*
 153  * Returns true if journal entry is now closed:
 154  *
 155  * We don't close a journal_buf until the next journal_buf is finished writing,
 156  * and can be opened again - this also initializes the next journal_buf:
 157  */
 158 static void __journal_entry_close(struct journal *j, unsigned closed_val)
 159 {
 160         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 161         struct journal_buf *buf = journal_cur_buf(j);
 162         union journal_res_state old, new;
 163         u64 v = atomic64_read(&j->reservations.counter);
 164         unsigned sectors;
 165
 166         BUG_ON(closed_val != JOURNAL_ENTRY_CLOSED_VAL &&
 167                closed_val != JOURNAL_ENTRY_ERROR_VAL);
 168
 169         lockdep_assert_held(&j->lock);
 170
 171         do {
 172                 old.v = new.v = v;
 173                 new.cur_entry_offset = closed_val;
 174
 175                 if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL ||
 176                     old.cur_entry_offset == new.cur_entry_offset)
 177                         return;
 178         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 179                                        old.v, new.v)) != old.v);
 180
 181         if (!__journal_entry_is_open(old))
 182                 return;
 183
 184         /* Close out old buffer: */
 185         buf->data->u64s         = cpu_to_le32(old.cur_entry_offset);
 186
 187         trace_journal_entry_close(c, vstruct_bytes(buf->data));
 188
 189         sectors = vstruct_blocks_plus(buf->data, c->block_bits,
 190                                       buf->u64s_reserved) << c->block_bits;
 191         BUG_ON(sectors > buf->sectors);
 192         buf->sectors = sectors;
 193
 194         /*
 195          * We have to set last_seq here, _before_ opening a new journal entry:
 196          *
 197          * A threads may replace an old pin with a new pin on their current
 198          * journal reservation - the expectation being that the journal will
 199          * contain either what the old pin protected or what the new pin
 200          * protects.
 201          *
 202          * After the old pin is dropped journal_last_seq() won't include the old
 203          * pin, so we can only write the updated last_seq on the entry that
 204          * contains whatever the new pin protects.
 205          *
 206          * Restated, we can _not_ update last_seq for a given entry if there
 207          * could be a newer entry open with reservations/pins that have been
 208          * taken against it.
 209          *
 210          * Hence, we want update/set last_seq on the current journal entry right
 211          * before we open a new one:
 212          */
 213         buf->last_seq           = journal_last_seq(j);
 214         buf->data->last_seq     = cpu_to_le64(buf->last_seq);
 215         BUG_ON(buf->last_seq > le64_to_cpu(buf->data->seq));
 216
 217         cancel_delayed_work(&j->write_work);
 218
 219         bch2_journal_space_available(j);
 220
 221         __bch2_journal_buf_put(j, old.idx, le64_to_cpu(buf->data->seq));
 222 }
 223
 224 void bch2_journal_halt(struct journal *j)
 225 {
 226         spin_lock(&j->lock);
 227         __journal_entry_close(j, JOURNAL_ENTRY_ERROR_VAL);
 228         if (!j->err_seq)
 229                 j->err_seq = journal_cur_seq(j);
 230         journal_wake(j);
 231         spin_unlock(&j->lock);
 232 }
 233
 234 static bool journal_entry_want_write(struct journal *j)
 235 {
 236         bool ret = !journal_entry_is_open(j) ||
 237                 journal_cur_seq(j) == journal_last_unwritten_seq(j);
 238
 239         /* Don't close it yet if we already have a write in flight: */
 240         if (ret)
 241                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 242         else if (nr_unwritten_journal_entries(j)) {
 243                 struct journal_buf *buf = journal_cur_buf(j);
 244
 245                 if (!buf->flush_time) {
 246                         buf->flush_time = local_clock() ?: 1;
 247                         buf->expires = jiffies;
 248                 }
 249         }
 250
 251         return ret;
 252 }
 253
 254 static bool journal_entry_close(struct journal *j)
 255 {
 256         bool ret;
 257
 258         spin_lock(&j->lock);
 259         ret = journal_entry_want_write(j);
 260         spin_unlock(&j->lock);
 261
 262         return ret;
 263 }
 264
 265 /*
 266  * should _only_ called from journal_res_get() - when we actually want a
 267  * journal reservation - journal entry is open means journal is dirty:
 268  */
 269 static int journal_entry_open(struct journal *j)
 270 {
 271         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 272         struct journal_buf *buf = j->buf +
 273                 ((journal_cur_seq(j) + 1) & JOURNAL_BUF_MASK);
 274         union journal_res_state old, new;
 275         int u64s;
 276         u64 v;
 277
 278         lockdep_assert_held(&j->lock);
 279         BUG_ON(journal_entry_is_open(j));
 280         BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
 281
 282         if (j->blocked)
 283                 return JOURNAL_ERR_blocked;
 284
 285         if (j->cur_entry_error)
 286                 return j->cur_entry_error;
 287
 288         if (bch2_journal_error(j))
 289                 return JOURNAL_ERR_insufficient_devices; /* -EROFS */
 290
 291         if (!fifo_free(&j->pin))
 292                 return JOURNAL_ERR_journal_pin_full;
 293
 294         if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf))
 295                 return JOURNAL_ERR_max_in_flight;
 296
 297         BUG_ON(!j->cur_entry_sectors);
 298
 299         buf->expires            =
 300                 (journal_cur_seq(j) == j->flushed_seq_ondisk
 301                  ? jiffies
 302                  : j->last_flush_write) +
 303                 msecs_to_jiffies(c->opts.journal_flush_delay);
 304
 305         buf->u64s_reserved      = j->entry_u64s_reserved;
 306         buf->disk_sectors       = j->cur_entry_sectors;
 307         buf->sectors            = min(buf->disk_sectors, buf->buf_size >> 9);
 308
 309         u64s = (int) (buf->sectors << 9) / sizeof(u64) -
 310                 journal_entry_overhead(j);
 311         u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 312
 313         if (u64s <= (ssize_t) j->early_journal_entries.nr)
 314                 return JOURNAL_ERR_journal_full;
 315
 316         if (fifo_empty(&j->pin) && j->reclaim_thread)
 317                 wake_up_process(j->reclaim_thread);
 318
 319         /*
 320          * The fifo_push() needs to happen at the same time as j->seq is
 321          * incremented for journal_last_seq() to be calculated correctly
 322          */
 323         atomic64_inc(&j->seq);
 324         journal_pin_list_init(fifo_push_ref(&j->pin), 1);
 325
 326         BUG_ON(j->pin.back - 1 != atomic64_read(&j->seq));
 327
 328         BUG_ON(j->buf + (journal_cur_seq(j) & JOURNAL_BUF_MASK) != buf);
 329
 330         bkey_extent_init(&buf->key);
 331         buf->noflush    = false;
 332         buf->must_flush = false;
 333         buf->separate_flush = false;
 334         buf->flush_time = 0;
 335
 336         memset(buf->data, 0, sizeof(*buf->data));
 337         buf->data->seq  = cpu_to_le64(journal_cur_seq(j));
 338         buf->data->u64s = 0;
 339
 340         if (j->early_journal_entries.nr) {
 341                 memcpy(buf->data->_data, j->early_journal_entries.data,
 342                        j->early_journal_entries.nr * sizeof(u64));
 343                 le32_add_cpu(&buf->data->u64s, j->early_journal_entries.nr);
 344         }
 345
 346         /*
 347          * Must be set before marking the journal entry as open:
 348          */
 349         j->cur_entry_u64s = u64s;
 350
 351         v = atomic64_read(&j->reservations.counter);
 352         do {
 353                 old.v = new.v = v;
 354
 355                 BUG_ON(old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL);
 356
 357                 new.idx++;
 358                 BUG_ON(journal_state_count(new, new.idx));
 359                 BUG_ON(new.idx != (journal_cur_seq(j) & JOURNAL_BUF_MASK));
 360
 361                 journal_state_inc(&new);
 362
 363                 /* Handle any already added entries */
 364                 new.cur_entry_offset = le32_to_cpu(buf->data->u64s);
 365         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 366                                        old.v, new.v)) != old.v);
 367
 368         mod_delayed_work(c->io_complete_wq,
 369                          &j->write_work,
 370                          msecs_to_jiffies(c->opts.journal_flush_delay));
 371         journal_wake(j);
 372
 373         if (j->early_journal_entries.nr)
 374                 darray_exit(&j->early_journal_entries);
 375         return 0;
 376 }
 377
 378 static bool journal_quiesced(struct journal *j)
 379 {
 380         bool ret = atomic64_read(&j->seq) == j->seq_ondisk;
 381
 382         if (!ret)
 383                 journal_entry_close(j);
 384         return ret;
 385 }
 386
 387 static void journal_quiesce(struct journal *j)
 388 {
 389         wait_event(j->wait, journal_quiesced(j));
 390 }
 391
 392 static void journal_write_work(struct work_struct *work)
 393 {
 394         struct journal *j = container_of(work, struct journal, write_work.work);
 395         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 396         long delta;
 397
 398         spin_lock(&j->lock);
 399         if (!__journal_entry_is_open(j->reservations))
 400                 goto unlock;
 401
 402         delta = journal_cur_buf(j)->expires - jiffies;
 403
 404         if (delta > 0)
 405                 mod_delayed_work(c->io_complete_wq, &j->write_work, delta);
 406         else
 407                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 408 unlock:
 409         spin_unlock(&j->lock);
 410 }
 411
 412 static int __journal_res_get(struct journal *j, struct journal_res *res,
 413                              unsigned flags)
 414 {
 415         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 416         struct journal_buf *buf;
 417         bool can_discard;
 418         int ret;
 419 retry:
 420         if (journal_res_get_fast(j, res, flags))
 421                 return 0;
 422
 423         if (bch2_journal_error(j))
 424                 return -BCH_ERR_erofs_journal_err;
 425
 426         spin_lock(&j->lock);
 427
 428         /* check once more in case somebody else shut things down... */
 429         if (bch2_journal_error(j)) {
 430                 spin_unlock(&j->lock);
 431                 return -BCH_ERR_erofs_journal_err;
 432         }
 433
 434         /*
 435          * Recheck after taking the lock, so we don't race with another thread
 436          * that just did journal_entry_open() and call journal_entry_close()
 437          * unnecessarily
 438          */
 439         if (journal_res_get_fast(j, res, flags)) {
 440                 spin_unlock(&j->lock);
 441                 return 0;
 442         }
 443
 444         if ((flags & BCH_WATERMARK_MASK) < j->watermark) {
 445                 /*
 446                  * Don't want to close current journal entry, just need to
 447                  * invoke reclaim:
 448                  */
 449                 ret = JOURNAL_ERR_journal_full;
 450                 goto unlock;
 451         }
 452
 453         /*
 454          * If we couldn't get a reservation because the current buf filled up,
 455          * and we had room for a bigger entry on disk, signal that we want to
 456          * realloc the journal bufs:
 457          */
 458         buf = journal_cur_buf(j);
 459         if (journal_entry_is_open(j) &&
 460             buf->buf_size >> 9 < buf->disk_sectors &&
 461             buf->buf_size < JOURNAL_ENTRY_SIZE_MAX)
 462                 j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1);
 463
 464         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 465         ret = journal_entry_open(j);
 466
 467         if (ret == JOURNAL_ERR_max_in_flight) {
 468                 track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight],
 469                                    &j->max_in_flight_start, true);
 470                 trace_and_count(c, journal_entry_full, c);
 471         }
 472 unlock:
 473         can_discard = j->can_discard;
 474         spin_unlock(&j->lock);
 475
 476         if (!ret)
 477                 goto retry;
 478         if (journal_error_check_stuck(j, ret, flags))
 479                 ret = -BCH_ERR_journal_res_get_blocked;
 480
 481         /*
 482          * Journal is full - can't rely on reclaim from work item due to
 483          * freezing:
 484          */
 485         if ((ret == JOURNAL_ERR_journal_full ||
 486              ret == JOURNAL_ERR_journal_pin_full) &&
 487             !(flags & JOURNAL_RES_GET_NONBLOCK)) {
 488                 if (can_discard) {
 489                         bch2_journal_do_discards(j);
 490                         goto retry;
 491                 }
 492
 493                 if (mutex_trylock(&j->reclaim_lock)) {
 494                         bch2_journal_reclaim(j);
 495                         mutex_unlock(&j->reclaim_lock);
 496                 }
 497         }
 498
 499         return ret == JOURNAL_ERR_insufficient_devices
 500                 ? -BCH_ERR_erofs_journal_err
 501                 : -BCH_ERR_journal_res_get_blocked;
 502 }
 503
 504 /*
 505  * Essentially the entry function to the journaling code. When bcachefs is doing
 506  * a btree insert, it calls this function to get the current journal write.
 507  * Journal write is the structure used set up journal writes. The calling
 508  * function will then add its keys to the structure, queuing them for the next
 509  * write.
 510  *
 511  * To ensure forward progress, the current task must not be holding any
 512  * btree node write locks.
 513  */
 514 int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
 515                                   unsigned flags)
 516 {
 517         int ret;
 518
 519         closure_wait_event(&j->async_wait,
 520                    (ret = __journal_res_get(j, res, flags)) != -BCH_ERR_journal_res_get_blocked ||
 521                    (flags & JOURNAL_RES_GET_NONBLOCK));
 522         return ret;
 523 }
 524
 525 /* journal_entry_res: */
 526
 527 void bch2_journal_entry_res_resize(struct journal *j,
 528                                    struct journal_entry_res *res,
 529                                    unsigned new_u64s)
 530 {
 531         union journal_res_state state;
 532         int d = new_u64s - res->u64s;
 533
 534         spin_lock(&j->lock);
 535
 536         j->entry_u64s_reserved += d;
 537         if (d <= 0)
 538                 goto out;
 539
 540         j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d);
 541         smp_mb();
 542         state = READ_ONCE(j->reservations);
 543
 544         if (state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL &&
 545             state.cur_entry_offset > j->cur_entry_u64s) {
 546                 j->cur_entry_u64s += d;
 547                 /*
 548                  * Not enough room in current journal entry, have to flush it:
 549                  */
 550                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 551         } else {
 552                 journal_cur_buf(j)->u64s_reserved += d;
 553         }
 554 out:
 555         spin_unlock(&j->lock);
 556         res->u64s += d;
 557 }
 558
 559 /* journal flushing: */
 560
 561 /**
 562  * bch2_journal_flush_seq_async - wait for a journal entry to be written
 563  * @j:          journal object
 564  * @seq:        seq to flush
 565  * @parent:     closure object to wait with
 566  * Returns:     1 if @seq has already been flushed, 0 if @seq is being flushed,
 567  *              -EIO if @seq will never be flushed
 568  *
 569  * Like bch2_journal_wait_on_seq, except that it triggers a write immediately if
 570  * necessary
 571  */
 572 int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 573                                  struct closure *parent)
 574 {
 575         struct journal_buf *buf;
 576         int ret = 0;
 577
 578         if (seq <= j->flushed_seq_ondisk)
 579                 return 1;
 580
 581         spin_lock(&j->lock);
 582
 583         if (WARN_ONCE(seq > journal_cur_seq(j),
 584                       "requested to flush journal seq %llu, but currently at %llu",
 585                       seq, journal_cur_seq(j)))
 586                 goto out;
 587
 588         /* Recheck under lock: */
 589         if (j->err_seq && seq >= j->err_seq) {
 590                 ret = -EIO;
 591                 goto out;
 592         }
 593
 594         if (seq <= j->flushed_seq_ondisk) {
 595                 ret = 1;
 596                 goto out;
 597         }
 598
 599         /* if seq was written, but not flushed - flush a newer one instead */
 600         seq = max(seq, journal_last_unwritten_seq(j));
 601
 602 recheck_need_open:
 603         if (seq > journal_cur_seq(j)) {
 604                 struct journal_res res = { 0 };
 605
 606                 if (journal_entry_is_open(j))
 607                         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 608
 609                 spin_unlock(&j->lock);
 610
 611                 ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 612                 if (ret)
 613                         return ret;
 614
 615                 seq = res.seq;
 616                 buf = j->buf + (seq & JOURNAL_BUF_MASK);
 617                 buf->must_flush = true;
 618
 619                 if (!buf->flush_time) {
 620                         buf->flush_time = local_clock() ?: 1;
 621                         buf->expires = jiffies;
 622                 }
 623
 624                 if (parent && !closure_wait(&buf->wait, parent))
 625                         BUG();
 626
 627                 bch2_journal_res_put(j, &res);
 628
 629                 spin_lock(&j->lock);
 630                 goto want_write;
 631         }
 632
 633         /*
 634          * if write was kicked off without a flush, flush the next sequence
 635          * number instead
 636          */
 637         buf = journal_seq_to_buf(j, seq);
 638         if (buf->noflush) {
 639                 seq++;
 640                 goto recheck_need_open;
 641         }
 642
 643         buf->must_flush = true;
 644
 645         if (parent && !closure_wait(&buf->wait, parent))
 646                 BUG();
 647 want_write:
 648         if (seq == journal_cur_seq(j))
 649                 journal_entry_want_write(j);
 650 out:
 651         spin_unlock(&j->lock);
 652         return ret;
 653 }
 654
 655 int bch2_journal_flush_seq(struct journal *j, u64 seq)
 656 {
 657         u64 start_time = local_clock();
 658         int ret, ret2;
 659
 660         /*
 661          * Don't update time_stats when @seq is already flushed:
 662          */
 663         if (seq <= j->flushed_seq_ondisk)
 664                 return 0;
 665
 666         ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)));
 667
 668         if (!ret)
 669                 bch2_time_stats_update(j->flush_seq_time, start_time);
 670
 671         return ret ?: ret2 < 0 ? ret2 : 0;
 672 }
 673
 674 /*
 675  * bch2_journal_flush_async - if there is an open journal entry, or a journal
 676  * still being written, write it and wait for the write to complete
 677  */
 678 void bch2_journal_flush_async(struct journal *j, struct closure *parent)
 679 {
 680         bch2_journal_flush_seq_async(j, atomic64_read(&j->seq), parent);
 681 }
 682
 683 int bch2_journal_flush(struct journal *j)
 684 {
 685         return bch2_journal_flush_seq(j, atomic64_read(&j->seq));
 686 }
 687
 688 /*
 689  * bch2_journal_noflush_seq - tell the journal not to issue any flushes before
 690  * @seq
 691  */
 692 bool bch2_journal_noflush_seq(struct journal *j, u64 seq)
 693 {
 694         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 695         u64 unwritten_seq;
 696         bool ret = false;
 697
 698         if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush)))
 699                 return false;
 700
 701         if (seq <= c->journal.flushed_seq_ondisk)
 702                 return false;
 703
 704         spin_lock(&j->lock);
 705         if (seq <= c->journal.flushed_seq_ondisk)
 706                 goto out;
 707
 708         for (unwritten_seq = journal_last_unwritten_seq(j);
 709              unwritten_seq < seq;
 710              unwritten_seq++) {
 711                 struct journal_buf *buf = journal_seq_to_buf(j, unwritten_seq);
 712
 713                 /* journal write is already in flight, and was a flush write: */
 714                 if (unwritten_seq == journal_last_unwritten_seq(j) && !buf->noflush)
 715                         goto out;
 716
 717                 buf->noflush = true;
 718         }
 719
 720         ret = true;
 721 out:
 722         spin_unlock(&j->lock);
 723         return ret;
 724 }
 725
 726 int bch2_journal_meta(struct journal *j)
 727 {
 728         struct journal_buf *buf;
 729         struct journal_res res;
 730         int ret;
 731
 732         memset(&res, 0, sizeof(res));
 733
 734         ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 735         if (ret)
 736                 return ret;
 737
 738         buf = j->buf + (res.seq & JOURNAL_BUF_MASK);
 739         buf->must_flush = true;
 740
 741         if (!buf->flush_time) {
 742                 buf->flush_time = local_clock() ?: 1;
 743                 buf->expires = jiffies;
 744         }
 745
 746         bch2_journal_res_put(j, &res);
 747
 748         return bch2_journal_flush_seq(j, res.seq);
 749 }
 750
 751 /* block/unlock the journal: */
 752
 753 void bch2_journal_unblock(struct journal *j)
 754 {
 755         spin_lock(&j->lock);
 756         j->blocked--;
 757         spin_unlock(&j->lock);
 758
 759         journal_wake(j);
 760 }
 761
 762 void bch2_journal_block(struct journal *j)
 763 {
 764         spin_lock(&j->lock);
 765         j->blocked++;
 766         spin_unlock(&j->lock);
 767
 768         journal_quiesce(j);
 769 }
 770
 771 /* allocate journal on a device: */
 772
 773 static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 774                                          bool new_fs, struct closure *cl)
 775 {
 776         struct bch_fs *c = ca->fs;
 777         struct journal_device *ja = &ca->journal;
 778         u64 *new_bucket_seq = NULL, *new_buckets = NULL;
 779         struct open_bucket **ob = NULL;
 780         long *bu = NULL;
 781         unsigned i, pos, nr_got = 0, nr_want = nr - ja->nr;
 782         int ret = 0;
 783
 784         BUG_ON(nr <= ja->nr);
 785
 786         bu              = kcalloc(nr_want, sizeof(*bu), GFP_KERNEL);
 787         ob              = kcalloc(nr_want, sizeof(*ob), GFP_KERNEL);
 788         new_buckets     = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 789         new_bucket_seq  = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 790         if (!bu || !ob || !new_buckets || !new_bucket_seq) {
 791                 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
 792                 goto err_free;
 793         }
 794
 795         for (nr_got = 0; nr_got < nr_want; nr_got++) {
 796                 if (new_fs) {
 797                         bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
 798                         if (bu[nr_got] < 0) {
 799                                 ret = -BCH_ERR_ENOSPC_bucket_alloc;
 800                                 break;
 801                         }
 802                 } else {
 803                         ob[nr_got] = bch2_bucket_alloc(c, ca, BCH_WATERMARK_normal, cl);
 804                         ret = PTR_ERR_OR_ZERO(ob[nr_got]);
 805                         if (ret)
 806                                 break;
 807
 808                         ret = bch2_trans_run(c,
 809                                 bch2_trans_mark_metadata_bucket(trans, ca,
 810                                                 ob[nr_got]->bucket, BCH_DATA_journal,
 811                                                 ca->mi.bucket_size));
 812                         if (ret) {
 813                                 bch2_open_bucket_put(c, ob[nr_got]);
 814                                 bch_err_msg(c, ret, "marking new journal buckets");
 815                                 break;
 816                         }
 817
 818                         bu[nr_got] = ob[nr_got]->bucket;
 819                 }
 820         }
 821
 822         if (!nr_got)
 823                 goto err_free;
 824
 825         /* Don't return an error if we successfully allocated some buckets: */
 826         ret = 0;
 827
 828         if (c) {
 829                 bch2_journal_flush_all_pins(&c->journal);
 830                 bch2_journal_block(&c->journal);
 831                 mutex_lock(&c->sb_lock);
 832         }
 833
 834         memcpy(new_buckets,     ja->buckets,    ja->nr * sizeof(u64));
 835         memcpy(new_bucket_seq,  ja->bucket_seq, ja->nr * sizeof(u64));
 836
 837         BUG_ON(ja->discard_idx > ja->nr);
 838
 839         pos = ja->discard_idx ?: ja->nr;
 840
 841         memmove(new_buckets + pos + nr_got,
 842                 new_buckets + pos,
 843                 sizeof(new_buckets[0]) * (ja->nr - pos));
 844         memmove(new_bucket_seq + pos + nr_got,
 845                 new_bucket_seq + pos,
 846                 sizeof(new_bucket_seq[0]) * (ja->nr - pos));
 847
 848         for (i = 0; i < nr_got; i++) {
 849                 new_buckets[pos + i] = bu[i];
 850                 new_bucket_seq[pos + i] = 0;
 851         }
 852
 853         nr = ja->nr + nr_got;
 854
 855         ret = bch2_journal_buckets_to_sb(c, ca, new_buckets, nr);
 856         if (ret)
 857                 goto err_unblock;
 858
 859         if (!new_fs)
 860                 bch2_write_super(c);
 861
 862         /* Commit: */
 863         if (c)
 864                 spin_lock(&c->journal.lock);
 865
 866         swap(new_buckets,       ja->buckets);
 867         swap(new_bucket_seq,    ja->bucket_seq);
 868         ja->nr = nr;
 869
 870         if (pos <= ja->discard_idx)
 871                 ja->discard_idx = (ja->discard_idx + nr_got) % ja->nr;
 872         if (pos <= ja->dirty_idx_ondisk)
 873                 ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + nr_got) % ja->nr;
 874         if (pos <= ja->dirty_idx)
 875                 ja->dirty_idx = (ja->dirty_idx + nr_got) % ja->nr;
 876         if (pos <= ja->cur_idx)
 877                 ja->cur_idx = (ja->cur_idx + nr_got) % ja->nr;
 878
 879         if (c)
 880                 spin_unlock(&c->journal.lock);
 881 err_unblock:
 882         if (c) {
 883                 bch2_journal_unblock(&c->journal);
 884                 mutex_unlock(&c->sb_lock);
 885         }
 886
 887         if (ret && !new_fs)
 888                 for (i = 0; i < nr_got; i++)
 889                         bch2_trans_run(c,
 890                                 bch2_trans_mark_metadata_bucket(trans, ca,
 891                                                 bu[i], BCH_DATA_free, 0));
 892 err_free:
 893         if (!new_fs)
 894                 for (i = 0; i < nr_got; i++)
 895                         bch2_open_bucket_put(c, ob[i]);
 896
 897         kfree(new_bucket_seq);
 898         kfree(new_buckets);
 899         kfree(ob);
 900         kfree(bu);
 901         return ret;
 902 }
 903
 904 /*
 905  * Allocate more journal space at runtime - not currently making use if it, but
 906  * the code works:
 907  */
 908 int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 909                                 unsigned nr)
 910 {
 911         struct journal_device *ja = &ca->journal;
 912         struct closure cl;
 913         int ret = 0;
 914
 915         closure_init_stack(&cl);
 916
 917         down_write(&c->state_lock);
 918
 919         /* don't handle reducing nr of buckets yet: */
 920         if (nr < ja->nr)
 921                 goto unlock;
 922
 923         while (ja->nr < nr) {
 924                 struct disk_reservation disk_res = { 0, 0, 0 };
 925
 926                 /*
 927                  * note: journal buckets aren't really counted as _sectors_ used yet, so
 928                  * we don't need the disk reservation to avoid the BUG_ON() in buckets.c
 929                  * when space used goes up without a reservation - but we do need the
 930                  * reservation to ensure we'll actually be able to allocate:
 931                  *
 932                  * XXX: that's not right, disk reservations only ensure a
 933                  * filesystem-wide allocation will succeed, this is a device
 934                  * specific allocation - we can hang here:
 935                  */
 936
 937                 ret = bch2_disk_reservation_get(c, &disk_res,
 938                                                 bucket_to_sector(ca, nr - ja->nr), 1, 0);
 939                 if (ret)
 940                         break;
 941
 942                 ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl);
 943
 944                 bch2_disk_reservation_put(c, &disk_res);
 945
 946                 closure_sync(&cl);
 947
 948                 if (ret && ret != -BCH_ERR_bucket_alloc_blocked)
 949                         break;
 950         }
 951
 952         if (ret)
 953                 bch_err_fn(c, ret);
 954 unlock:
 955         up_write(&c->state_lock);
 956         return ret;
 957 }
 958
 959 int bch2_dev_journal_alloc(struct bch_dev *ca)
 960 {
 961         unsigned nr;
 962         int ret;
 963
 964         if (dynamic_fault("bcachefs:add:journal_alloc")) {
 965                 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
 966                 goto err;
 967         }
 968
 969         /* 1/128th of the device by default: */
 970         nr = ca->mi.nbuckets >> 7;
 971
 972         /*
 973          * clamp journal size to 8192 buckets or 8GB (in sectors), whichever
 974          * is smaller:
 975          */
 976         nr = clamp_t(unsigned, nr,
 977                      BCH_JOURNAL_BUCKETS_MIN,
 978                      min(1 << 13,
 979                          (1 << 24) / ca->mi.bucket_size));
 980
 981         ret = __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
 982 err:
 983         if (ret)
 984                 bch_err_fn(ca, ret);
 985         return ret;
 986 }
 987
 988 int bch2_fs_journal_alloc(struct bch_fs *c)
 989 {
 990         struct bch_dev *ca;
 991         unsigned i;
 992
 993         for_each_online_member(ca, c, i) {
 994                 if (ca->journal.nr)
 995                         continue;
 996
 997                 int ret = bch2_dev_journal_alloc(ca);
 998                 if (ret) {
 999                         percpu_ref_put(&ca->io_ref);
1000                         return ret;
1001                 }
1002         }
1003
1004         return 0;
1005 }
1006
1007 /* startup/shutdown: */
1008
1009 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
1010 {
1011         bool ret = false;
1012         u64 seq;
1013
1014         spin_lock(&j->lock);
1015         for (seq = journal_last_unwritten_seq(j);
1016              seq <= journal_cur_seq(j) && !ret;
1017              seq++) {
1018                 struct journal_buf *buf = journal_seq_to_buf(j, seq);
1019
1020                 if (bch2_bkey_has_device_c(bkey_i_to_s_c(&buf->key), dev_idx))
1021                         ret = true;
1022         }
1023         spin_unlock(&j->lock);
1024
1025         return ret;
1026 }
1027
1028 void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca)
1029 {
1030         wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx));
1031 }
1032
1033 void bch2_fs_journal_stop(struct journal *j)
1034 {
1035         bch2_journal_reclaim_stop(j);
1036         bch2_journal_flush_all_pins(j);
1037
1038         wait_event(j->wait, journal_entry_close(j));
1039
1040         /*
1041          * Always write a new journal entry, to make sure the clock hands are up
1042          * to date (and match the superblock)
1043          */
1044         bch2_journal_meta(j);
1045
1046         journal_quiesce(j);
1047
1048         BUG_ON(!bch2_journal_error(j) &&
1049                test_bit(JOURNAL_REPLAY_DONE, &j->flags) &&
1050                j->last_empty_seq != journal_cur_seq(j));
1051
1052         cancel_delayed_work_sync(&j->write_work);
1053 }
1054
1055 int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
1056 {
1057         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1058         struct journal_entry_pin_list *p;
1059         struct journal_replay *i, **_i;
1060         struct genradix_iter iter;
1061         bool had_entries = false;
1062         unsigned ptr;
1063         u64 last_seq = cur_seq, nr, seq;
1064
1065         genradix_for_each_reverse(&c->journal_entries, iter, _i) {
1066                 i = *_i;
1067
1068                 if (!i || i->ignore)
1069                         continue;
1070
1071                 last_seq = le64_to_cpu(i->j.last_seq);
1072                 break;
1073         }
1074
1075         nr = cur_seq - last_seq;
1076
1077         if (nr + 1 > j->pin.size) {
1078                 free_fifo(&j->pin);
1079                 init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL);
1080                 if (!j->pin.data) {
1081                         bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
1082                         return -BCH_ERR_ENOMEM_journal_pin_fifo;
1083                 }
1084         }
1085
1086         j->replay_journal_seq   = last_seq;
1087         j->replay_journal_seq_end = cur_seq;
1088         j->last_seq_ondisk      = last_seq;
1089         j->flushed_seq_ondisk   = cur_seq - 1;
1090         j->seq_ondisk           = cur_seq - 1;
1091         j->pin.front            = last_seq;
1092         j->pin.back             = cur_seq;
1093         atomic64_set(&j->seq, cur_seq - 1);
1094
1095         fifo_for_each_entry_ptr(p, &j->pin, seq)
1096                 journal_pin_list_init(p, 1);
1097
1098         genradix_for_each(&c->journal_entries, iter, _i) {
1099                 i = *_i;
1100
1101                 if (!i || i->ignore)
1102                         continue;
1103
1104                 seq = le64_to_cpu(i->j.seq);
1105                 BUG_ON(seq >= cur_seq);
1106
1107                 if (seq < last_seq)
1108                         continue;
1109
1110                 if (journal_entry_empty(&i->j))
1111                         j->last_empty_seq = le64_to_cpu(i->j.seq);
1112
1113                 p = journal_seq_pin(j, seq);
1114
1115                 p->devs.nr = 0;
1116                 for (ptr = 0; ptr < i->nr_ptrs; ptr++)
1117                         bch2_dev_list_add_dev(&p->devs, i->ptrs[ptr].dev);
1118
1119                 had_entries = true;
1120         }
1121
1122         if (!had_entries)
1123                 j->last_empty_seq = cur_seq;
1124
1125         spin_lock(&j->lock);
1126
1127         set_bit(JOURNAL_STARTED, &j->flags);
1128         j->last_flush_write = jiffies;
1129
1130         j->reservations.idx = j->reservations.unwritten_idx = journal_cur_seq(j);
1131         j->reservations.unwritten_idx++;
1132
1133         c->last_bucket_seq_cleanup = journal_cur_seq(j);
1134
1135         bch2_journal_space_available(j);
1136         spin_unlock(&j->lock);
1137
1138         return bch2_journal_reclaim_start(j);
1139 }
1140
1141 /* init/exit: */
1142
1143 void bch2_dev_journal_exit(struct bch_dev *ca)
1144 {
1145         kfree(ca->journal.bio);
1146         kfree(ca->journal.buckets);
1147         kfree(ca->journal.bucket_seq);
1148
1149         ca->journal.bio         = NULL;
1150         ca->journal.buckets     = NULL;
1151         ca->journal.bucket_seq  = NULL;
1152 }
1153
1154 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
1155 {
1156         struct journal_device *ja = &ca->journal;
1157         struct bch_sb_field_journal *journal_buckets =
1158                 bch2_sb_field_get(sb, journal);
1159         struct bch_sb_field_journal_v2 *journal_buckets_v2 =
1160                 bch2_sb_field_get(sb, journal_v2);
1161         unsigned i, nr_bvecs;
1162
1163         ja->nr = 0;
1164
1165         if (journal_buckets_v2) {
1166                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1167
1168                 for (i = 0; i < nr; i++)
1169                         ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr);
1170         } else if (journal_buckets) {
1171                 ja->nr = bch2_nr_journal_buckets(journal_buckets);
1172         }
1173
1174         ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1175         if (!ja->bucket_seq)
1176                 return -BCH_ERR_ENOMEM_dev_journal_init;
1177
1178         nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
1179
1180         ca->journal.bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1181         if (!ca->journal.bio)
1182                 return -BCH_ERR_ENOMEM_dev_journal_init;
1183
1184         bio_init(ca->journal.bio, NULL, ca->journal.bio->bi_inline_vecs, nr_bvecs, 0);
1185
1186         ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1187         if (!ja->buckets)
1188                 return -BCH_ERR_ENOMEM_dev_journal_init;
1189
1190         if (journal_buckets_v2) {
1191                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1192                 unsigned j, dst = 0;
1193
1194                 for (i = 0; i < nr; i++)
1195                         for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
1196                                 ja->buckets[dst++] =
1197                                         le64_to_cpu(journal_buckets_v2->d[i].start) + j;
1198         } else if (journal_buckets) {
1199                 for (i = 0; i < ja->nr; i++)
1200                         ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
1201         }
1202
1203         return 0;
1204 }
1205
1206 void bch2_fs_journal_exit(struct journal *j)
1207 {
1208         unsigned i;
1209
1210         darray_exit(&j->early_journal_entries);
1211
1212         for (i = 0; i < ARRAY_SIZE(j->buf); i++)
1213                 kvpfree(j->buf[i].data, j->buf[i].buf_size);
1214         free_fifo(&j->pin);
1215 }
1216
1217 int bch2_fs_journal_init(struct journal *j)
1218 {
1219         static struct lock_class_key res_key;
1220         unsigned i;
1221
1222         spin_lock_init(&j->lock);
1223         spin_lock_init(&j->err_lock);
1224         init_waitqueue_head(&j->wait);
1225         INIT_DELAYED_WORK(&j->write_work, journal_write_work);
1226         init_waitqueue_head(&j->reclaim_wait);
1227         init_waitqueue_head(&j->pin_flush_wait);
1228         mutex_init(&j->reclaim_lock);
1229         mutex_init(&j->discard_lock);
1230
1231         lockdep_init_map(&j->res_map, "journal res", &res_key, 0);
1232
1233         atomic64_set(&j->reservations.counter,
1234                 ((union journal_res_state)
1235                  { .cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL }).v);
1236
1237         if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)))
1238                 return -BCH_ERR_ENOMEM_journal_pin_fifo;
1239
1240         for (i = 0; i < ARRAY_SIZE(j->buf); i++) {
1241                 j->buf[i].buf_size = JOURNAL_ENTRY_SIZE_MIN;
1242                 j->buf[i].data = kvpmalloc(j->buf[i].buf_size, GFP_KERNEL);
1243                 if (!j->buf[i].data)
1244                         return -BCH_ERR_ENOMEM_journal_buf;
1245         }
1246
1247         j->pin.front = j->pin.back = 1;
1248         return 0;
1249 }
1250
1251 /* debug: */
1252
1253 void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1254 {
1255         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1256         union journal_res_state s;
1257         struct bch_dev *ca;
1258         unsigned long now = jiffies;
1259         u64 nr_writes = j->nr_flush_writes + j->nr_noflush_writes;
1260         u64 seq;
1261         unsigned i;
1262
1263         if (!out->nr_tabstops)
1264                 printbuf_tabstop_push(out, 24);
1265         out->atomic++;
1266
1267         rcu_read_lock();
1268         s = READ_ONCE(j->reservations);
1269
1270         prt_printf(out, "dirty journal entries:\t%llu/%llu\n",  fifo_used(&j->pin), j->pin.size);
1271         prt_printf(out, "seq:\t\t\t%llu\n",                     journal_cur_seq(j));
1272         prt_printf(out, "seq_ondisk:\t\t%llu\n",                j->seq_ondisk);
1273         prt_printf(out, "last_seq:\t\t%llu\n",                  journal_last_seq(j));
1274         prt_printf(out, "last_seq_ondisk:\t%llu\n",             j->last_seq_ondisk);
1275         prt_printf(out, "flushed_seq_ondisk:\t%llu\n",          j->flushed_seq_ondisk);
1276         prt_printf(out, "watermark:\t\t%s\n",                   bch2_watermarks[j->watermark]);
1277         prt_printf(out, "each entry reserved:\t%u\n",           j->entry_u64s_reserved);
1278         prt_printf(out, "nr flush writes:\t%llu\n",             j->nr_flush_writes);
1279         prt_printf(out, "nr noflush writes:\t%llu\n",           j->nr_noflush_writes);
1280         prt_printf(out, "average write size:\t");
1281         prt_human_readable_u64(out, nr_writes ? div64_u64(j->entry_bytes_written, nr_writes) : 0);
1282         prt_newline(out);
1283         prt_printf(out, "nr direct reclaim:\t%llu\n",           j->nr_direct_reclaim);
1284         prt_printf(out, "nr background reclaim:\t%llu\n",       j->nr_background_reclaim);
1285         prt_printf(out, "reclaim kicked:\t\t%u\n",              j->reclaim_kicked);
1286         prt_printf(out, "reclaim runs in:\t%u ms\n",            time_after(j->next_reclaim, now)
1287                ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0);
1288         prt_printf(out, "current entry sectors:\t%u\n",         j->cur_entry_sectors);
1289         prt_printf(out, "current entry error:\t%s\n",           bch2_journal_errors[j->cur_entry_error]);
1290         prt_printf(out, "current entry:\t\t");
1291
1292         switch (s.cur_entry_offset) {
1293         case JOURNAL_ENTRY_ERROR_VAL:
1294                 prt_printf(out, "error");
1295                 break;
1296         case JOURNAL_ENTRY_CLOSED_VAL:
1297                 prt_printf(out, "closed");
1298                 break;
1299         default:
1300                 prt_printf(out, "%u/%u", s.cur_entry_offset, j->cur_entry_u64s);
1301                 break;
1302         }
1303
1304         prt_newline(out);
1305
1306         for (seq = journal_cur_seq(j);
1307              seq >= journal_last_unwritten_seq(j);
1308              --seq) {
1309                 i = seq & JOURNAL_BUF_MASK;
1310
1311                 prt_printf(out, "unwritten entry:");
1312                 prt_tab(out);
1313                 prt_printf(out, "%llu", seq);
1314                 prt_newline(out);
1315                 printbuf_indent_add(out, 2);
1316
1317                 prt_printf(out, "refcount:");
1318                 prt_tab(out);
1319                 prt_printf(out, "%u", journal_state_count(s, i));
1320                 prt_newline(out);
1321
1322                 prt_printf(out, "sectors:");
1323                 prt_tab(out);
1324                 prt_printf(out, "%u", j->buf[i].sectors);
1325                 prt_newline(out);
1326
1327                 prt_printf(out, "expires");
1328                 prt_tab(out);
1329                 prt_printf(out, "%li jiffies", j->buf[i].expires - jiffies);
1330                 prt_newline(out);
1331
1332                 printbuf_indent_sub(out, 2);
1333         }
1334
1335         prt_printf(out,
1336                "replay done:\t\t%i\n",
1337                test_bit(JOURNAL_REPLAY_DONE,    &j->flags));
1338
1339         prt_printf(out, "space:\n");
1340         prt_printf(out, "\tdiscarded\t%u:%u\n",
1341                j->space[journal_space_discarded].next_entry,
1342                j->space[journal_space_discarded].total);
1343         prt_printf(out, "\tclean ondisk\t%u:%u\n",
1344                j->space[journal_space_clean_ondisk].next_entry,
1345                j->space[journal_space_clean_ondisk].total);
1346         prt_printf(out, "\tclean\t\t%u:%u\n",
1347                j->space[journal_space_clean].next_entry,
1348                j->space[journal_space_clean].total);
1349         prt_printf(out, "\ttotal\t\t%u:%u\n",
1350                j->space[journal_space_total].next_entry,
1351                j->space[journal_space_total].total);
1352
1353         for_each_member_device_rcu(ca, c, i,
1354                                    &c->rw_devs[BCH_DATA_journal]) {
1355                 struct journal_device *ja = &ca->journal;
1356
1357                 if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d))
1358                         continue;
1359
1360                 if (!ja->nr)
1361                         continue;
1362
1363                 prt_printf(out, "dev %u:\n",            i);
1364                 prt_printf(out, "\tnr\t\t%u\n",         ja->nr);
1365                 prt_printf(out, "\tbucket size\t%u\n",  ca->mi.bucket_size);
1366                 prt_printf(out, "\tavailable\t%u:%u\n", bch2_journal_dev_buckets_available(j, ja, journal_space_discarded), ja->sectors_free);
1367                 prt_printf(out, "\tdiscard_idx\t%u\n",  ja->discard_idx);
1368                 prt_printf(out, "\tdirty_ondisk\t%u (seq %llu)\n", ja->dirty_idx_ondisk,        ja->bucket_seq[ja->dirty_idx_ondisk]);
1369                 prt_printf(out, "\tdirty_idx\t%u (seq %llu)\n", ja->dirty_idx,          ja->bucket_seq[ja->dirty_idx]);
1370                 prt_printf(out, "\tcur_idx\t\t%u (seq %llu)\n", ja->cur_idx,            ja->bucket_seq[ja->cur_idx]);
1371         }
1372
1373         rcu_read_unlock();
1374
1375         --out->atomic;
1376 }
1377
1378 void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1379 {
1380         spin_lock(&j->lock);
1381         __bch2_journal_debug_to_text(out, j);
1382         spin_unlock(&j->lock);
1383 }
1384
1385 bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq)
1386 {
1387         struct journal_entry_pin_list *pin_list;
1388         struct journal_entry_pin *pin;
1389         unsigned i;
1390
1391         spin_lock(&j->lock);
1392         *seq = max(*seq, j->pin.front);
1393
1394         if (*seq >= j->pin.back) {
1395                 spin_unlock(&j->lock);
1396                 return true;
1397         }
1398
1399         out->atomic++;
1400
1401         pin_list = journal_seq_pin(j, *seq);
1402
1403         prt_printf(out, "%llu: count %u", *seq, atomic_read(&pin_list->count));
1404         prt_newline(out);
1405         printbuf_indent_add(out, 2);
1406
1407         for (i = 0; i < ARRAY_SIZE(pin_list->list); i++)
1408                 list_for_each_entry(pin, &pin_list->list[i], list) {
1409                         prt_printf(out, "\t%px %ps", pin, pin->flush);
1410                         prt_newline(out);
1411                 }
1412
1413         if (!list_empty(&pin_list->flushed)) {
1414                 prt_printf(out, "flushed:");
1415                 prt_newline(out);
1416         }
1417
1418         list_for_each_entry(pin, &pin_list->flushed, list) {
1419                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1420                 prt_newline(out);
1421         }
1422
1423         printbuf_indent_sub(out, 2);
1424
1425         --out->atomic;
1426         spin_unlock(&j->lock);
1427
1428         return false;
1429 }
1430
1431 void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
1432 {
1433         u64 seq = 0;
1434
1435         while (!bch2_journal_seq_pins_to_text(out, j, &seq))
1436                 seq++;
1437 }