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         sectors = vstruct_blocks_plus(buf->data, c->block_bits,
 188                                       buf->u64s_reserved) << c->block_bits;
 189         BUG_ON(sectors > buf->sectors);
 190         buf->sectors = sectors;
 191
 192         /*
 193          * We have to set last_seq here, _before_ opening a new journal entry:
 194          *
 195          * A threads may replace an old pin with a new pin on their current
 196          * journal reservation - the expectation being that the journal will
 197          * contain either what the old pin protected or what the new pin
 198          * protects.
 199          *
 200          * After the old pin is dropped journal_last_seq() won't include the old
 201          * pin, so we can only write the updated last_seq on the entry that
 202          * contains whatever the new pin protects.
 203          *
 204          * Restated, we can _not_ update last_seq for a given entry if there
 205          * could be a newer entry open with reservations/pins that have been
 206          * taken against it.
 207          *
 208          * Hence, we want update/set last_seq on the current journal entry right
 209          * before we open a new one:
 210          */
 211         buf->last_seq           = journal_last_seq(j);
 212         buf->data->last_seq     = cpu_to_le64(buf->last_seq);
 213         BUG_ON(buf->last_seq > le64_to_cpu(buf->data->seq));
 214
 215         cancel_delayed_work(&j->write_work);
 216
 217         bch2_journal_space_available(j);
 218
 219         __bch2_journal_buf_put(j, old.idx, le64_to_cpu(buf->data->seq));
 220 }
 221
 222 void bch2_journal_halt(struct journal *j)
 223 {
 224         spin_lock(&j->lock);
 225         __journal_entry_close(j, JOURNAL_ENTRY_ERROR_VAL);
 226         if (!j->err_seq)
 227                 j->err_seq = journal_cur_seq(j);
 228         journal_wake(j);
 229         spin_unlock(&j->lock);
 230 }
 231
 232 static bool journal_entry_want_write(struct journal *j)
 233 {
 234         bool ret = !journal_entry_is_open(j) ||
 235                 journal_cur_seq(j) == journal_last_unwritten_seq(j);
 236
 237         /* Don't close it yet if we already have a write in flight: */
 238         if (ret)
 239                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 240         else if (nr_unwritten_journal_entries(j)) {
 241                 struct journal_buf *buf = journal_cur_buf(j);
 242
 243                 if (!buf->flush_time) {
 244                         buf->flush_time = local_clock() ?: 1;
 245                         buf->expires = jiffies;
 246                 }
 247         }
 248
 249         return ret;
 250 }
 251
 252 static bool journal_entry_close(struct journal *j)
 253 {
 254         bool ret;
 255
 256         spin_lock(&j->lock);
 257         ret = journal_entry_want_write(j);
 258         spin_unlock(&j->lock);
 259
 260         return ret;
 261 }
 262
 263 /*
 264  * should _only_ called from journal_res_get() - when we actually want a
 265  * journal reservation - journal entry is open means journal is dirty:
 266  */
 267 static int journal_entry_open(struct journal *j)
 268 {
 269         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 270         struct journal_buf *buf = j->buf +
 271                 ((journal_cur_seq(j) + 1) & JOURNAL_BUF_MASK);
 272         union journal_res_state old, new;
 273         int u64s;
 274         u64 v;
 275
 276         lockdep_assert_held(&j->lock);
 277         BUG_ON(journal_entry_is_open(j));
 278         BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
 279
 280         if (j->blocked)
 281                 return JOURNAL_ERR_blocked;
 282
 283         if (j->cur_entry_error)
 284                 return j->cur_entry_error;
 285
 286         if (bch2_journal_error(j))
 287                 return JOURNAL_ERR_insufficient_devices; /* -EROFS */
 288
 289         if (!fifo_free(&j->pin))
 290                 return JOURNAL_ERR_journal_pin_full;
 291
 292         if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf))
 293                 return JOURNAL_ERR_max_in_flight;
 294
 295         BUG_ON(!j->cur_entry_sectors);
 296
 297         buf->expires            =
 298                 (journal_cur_seq(j) == j->flushed_seq_ondisk
 299                  ? jiffies
 300                  : j->last_flush_write) +
 301                 msecs_to_jiffies(c->opts.journal_flush_delay);
 302
 303         buf->u64s_reserved      = j->entry_u64s_reserved;
 304         buf->disk_sectors       = j->cur_entry_sectors;
 305         buf->sectors            = min(buf->disk_sectors, buf->buf_size >> 9);
 306
 307         u64s = (int) (buf->sectors << 9) / sizeof(u64) -
 308                 journal_entry_overhead(j);
 309         u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 310
 311         if (u64s <= (ssize_t) j->early_journal_entries.nr)
 312                 return JOURNAL_ERR_journal_full;
 313
 314         if (fifo_empty(&j->pin) && j->reclaim_thread)
 315                 wake_up_process(j->reclaim_thread);
 316
 317         /*
 318          * The fifo_push() needs to happen at the same time as j->seq is
 319          * incremented for journal_last_seq() to be calculated correctly
 320          */
 321         atomic64_inc(&j->seq);
 322         journal_pin_list_init(fifo_push_ref(&j->pin), 1);
 323
 324         BUG_ON(j->buf + (journal_cur_seq(j) & JOURNAL_BUF_MASK) != buf);
 325
 326         bkey_extent_init(&buf->key);
 327         buf->noflush    = false;
 328         buf->must_flush = false;
 329         buf->separate_flush = false;
 330         buf->flush_time = 0;
 331
 332         memset(buf->data, 0, sizeof(*buf->data));
 333         buf->data->seq  = cpu_to_le64(journal_cur_seq(j));
 334         buf->data->u64s = 0;
 335
 336         if (j->early_journal_entries.nr) {
 337                 memcpy(buf->data->_data, j->early_journal_entries.data,
 338                        j->early_journal_entries.nr * sizeof(u64));
 339                 le32_add_cpu(&buf->data->u64s, j->early_journal_entries.nr);
 340         }
 341
 342         /*
 343          * Must be set before marking the journal entry as open:
 344          */
 345         j->cur_entry_u64s = u64s;
 346
 347         v = atomic64_read(&j->reservations.counter);
 348         do {
 349                 old.v = new.v = v;
 350
 351                 BUG_ON(old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL);
 352
 353                 new.idx++;
 354                 BUG_ON(journal_state_count(new, new.idx));
 355                 BUG_ON(new.idx != (journal_cur_seq(j) & JOURNAL_BUF_MASK));
 356
 357                 journal_state_inc(&new);
 358
 359                 /* Handle any already added entries */
 360                 new.cur_entry_offset = le32_to_cpu(buf->data->u64s);
 361         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 362                                        old.v, new.v)) != old.v);
 363
 364         if (j->res_get_blocked_start)
 365                 bch2_time_stats_update(j->blocked_time,
 366                                        j->res_get_blocked_start);
 367         j->res_get_blocked_start = 0;
 368
 369         mod_delayed_work(c->io_complete_wq,
 370                          &j->write_work,
 371                          msecs_to_jiffies(c->opts.journal_flush_delay));
 372         journal_wake(j);
 373
 374         if (j->early_journal_entries.nr)
 375                 darray_exit(&j->early_journal_entries);
 376         return 0;
 377 }
 378
 379 static bool journal_quiesced(struct journal *j)
 380 {
 381         bool ret = atomic64_read(&j->seq) == j->seq_ondisk;
 382
 383         if (!ret)
 384                 journal_entry_close(j);
 385         return ret;
 386 }
 387
 388 static void journal_quiesce(struct journal *j)
 389 {
 390         wait_event(j->wait, journal_quiesced(j));
 391 }
 392
 393 static void journal_write_work(struct work_struct *work)
 394 {
 395         struct journal *j = container_of(work, struct journal, write_work.work);
 396         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 397         long delta;
 398
 399         spin_lock(&j->lock);
 400         if (!__journal_entry_is_open(j->reservations))
 401                 goto unlock;
 402
 403         delta = journal_cur_buf(j)->expires - jiffies;
 404
 405         if (delta > 0)
 406                 mod_delayed_work(c->io_complete_wq, &j->write_work, delta);
 407         else
 408                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 409 unlock:
 410         spin_unlock(&j->lock);
 411 }
 412
 413 static int __journal_res_get(struct journal *j, struct journal_res *res,
 414                              unsigned flags)
 415 {
 416         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 417         struct journal_buf *buf;
 418         bool can_discard;
 419         int ret;
 420 retry:
 421         if (journal_res_get_fast(j, res, flags))
 422                 return 0;
 423
 424         if (bch2_journal_error(j))
 425                 return -BCH_ERR_erofs_journal_err;
 426
 427         spin_lock(&j->lock);
 428
 429         /* check once more in case somebody else shut things down... */
 430         if (bch2_journal_error(j)) {
 431                 spin_unlock(&j->lock);
 432                 return -BCH_ERR_erofs_journal_err;
 433         }
 434
 435         /*
 436          * Recheck after taking the lock, so we don't race with another thread
 437          * that just did journal_entry_open() and call journal_entry_close()
 438          * unnecessarily
 439          */
 440         if (journal_res_get_fast(j, res, flags)) {
 441                 spin_unlock(&j->lock);
 442                 return 0;
 443         }
 444
 445         if ((flags & BCH_WATERMARK_MASK) < j->watermark) {
 446                 /*
 447                  * Don't want to close current journal entry, just need to
 448                  * invoke reclaim:
 449                  */
 450                 ret = JOURNAL_ERR_journal_full;
 451                 goto unlock;
 452         }
 453
 454         /*
 455          * If we couldn't get a reservation because the current buf filled up,
 456          * and we had room for a bigger entry on disk, signal that we want to
 457          * realloc the journal bufs:
 458          */
 459         buf = journal_cur_buf(j);
 460         if (journal_entry_is_open(j) &&
 461             buf->buf_size >> 9 < buf->disk_sectors &&
 462             buf->buf_size < JOURNAL_ENTRY_SIZE_MAX)
 463                 j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1);
 464
 465         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 466         ret = journal_entry_open(j);
 467
 468         if (ret == JOURNAL_ERR_max_in_flight)
 469                 trace_and_count(c, journal_entry_full, c);
 470 unlock:
 471         if ((ret && ret != JOURNAL_ERR_insufficient_devices) &&
 472             !j->res_get_blocked_start) {
 473                 j->res_get_blocked_start = local_clock() ?: 1;
 474                 trace_and_count(c, journal_full, c);
 475         }
 476
 477         can_discard = j->can_discard;
 478         spin_unlock(&j->lock);
 479
 480         if (!ret)
 481                 goto retry;
 482         if (journal_error_check_stuck(j, ret, flags))
 483                 ret = -BCH_ERR_journal_res_get_blocked;
 484
 485         /*
 486          * Journal is full - can't rely on reclaim from work item due to
 487          * freezing:
 488          */
 489         if ((ret == JOURNAL_ERR_journal_full ||
 490              ret == JOURNAL_ERR_journal_pin_full) &&
 491             !(flags & JOURNAL_RES_GET_NONBLOCK)) {
 492                 if (can_discard) {
 493                         bch2_journal_do_discards(j);
 494                         goto retry;
 495                 }
 496
 497                 if (mutex_trylock(&j->reclaim_lock)) {
 498                         bch2_journal_reclaim(j);
 499                         mutex_unlock(&j->reclaim_lock);
 500                 }
 501         }
 502
 503         return ret == JOURNAL_ERR_insufficient_devices
 504                 ? -BCH_ERR_erofs_journal_err
 505                 : -BCH_ERR_journal_res_get_blocked;
 506 }
 507
 508 /*
 509  * Essentially the entry function to the journaling code. When bcachefs is doing
 510  * a btree insert, it calls this function to get the current journal write.
 511  * Journal write is the structure used set up journal writes. The calling
 512  * function will then add its keys to the structure, queuing them for the next
 513  * write.
 514  *
 515  * To ensure forward progress, the current task must not be holding any
 516  * btree node write locks.
 517  */
 518 int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
 519                                   unsigned flags)
 520 {
 521         int ret;
 522
 523         closure_wait_event(&j->async_wait,
 524                    (ret = __journal_res_get(j, res, flags)) != -BCH_ERR_journal_res_get_blocked ||
 525                    (flags & JOURNAL_RES_GET_NONBLOCK));
 526         return ret;
 527 }
 528
 529 /* journal_preres: */
 530
 531 static bool journal_preres_available(struct journal *j,
 532                                      struct journal_preres *res,
 533                                      unsigned new_u64s,
 534                                      unsigned flags)
 535 {
 536         bool ret = bch2_journal_preres_get_fast(j, res, new_u64s, flags, true);
 537
 538         if (!ret && mutex_trylock(&j->reclaim_lock)) {
 539                 bch2_journal_reclaim(j);
 540                 mutex_unlock(&j->reclaim_lock);
 541         }
 542
 543         return ret;
 544 }
 545
 546 int __bch2_journal_preres_get(struct journal *j,
 547                               struct journal_preres *res,
 548                               unsigned new_u64s,
 549                               unsigned flags)
 550 {
 551         int ret;
 552
 553         closure_wait_event(&j->preres_wait,
 554                    (ret = bch2_journal_error(j)) ||
 555                    journal_preres_available(j, res, new_u64s, flags));
 556         return ret;
 557 }
 558
 559 /* journal_entry_res: */
 560
 561 void bch2_journal_entry_res_resize(struct journal *j,
 562                                    struct journal_entry_res *res,
 563                                    unsigned new_u64s)
 564 {
 565         union journal_res_state state;
 566         int d = new_u64s - res->u64s;
 567
 568         spin_lock(&j->lock);
 569
 570         j->entry_u64s_reserved += d;
 571         if (d <= 0)
 572                 goto out;
 573
 574         j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d);
 575         smp_mb();
 576         state = READ_ONCE(j->reservations);
 577
 578         if (state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL &&
 579             state.cur_entry_offset > j->cur_entry_u64s) {
 580                 j->cur_entry_u64s += d;
 581                 /*
 582                  * Not enough room in current journal entry, have to flush it:
 583                  */
 584                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 585         } else {
 586                 journal_cur_buf(j)->u64s_reserved += d;
 587         }
 588 out:
 589         spin_unlock(&j->lock);
 590         res->u64s += d;
 591 }
 592
 593 /* journal flushing: */
 594
 595 /**
 596  * bch2_journal_flush_seq_async - wait for a journal entry to be written
 597  * @j:          journal object
 598  * @seq:        seq to flush
 599  * @parent:     closure object to wait with
 600  * Returns:     1 if @seq has already been flushed, 0 if @seq is being flushed,
 601  *              -EIO if @seq will never be flushed
 602  *
 603  * Like bch2_journal_wait_on_seq, except that it triggers a write immediately if
 604  * necessary
 605  */
 606 int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 607                                  struct closure *parent)
 608 {
 609         struct journal_buf *buf;
 610         int ret = 0;
 611
 612         if (seq <= j->flushed_seq_ondisk)
 613                 return 1;
 614
 615         spin_lock(&j->lock);
 616
 617         if (WARN_ONCE(seq > journal_cur_seq(j),
 618                       "requested to flush journal seq %llu, but currently at %llu",
 619                       seq, journal_cur_seq(j)))
 620                 goto out;
 621
 622         /* Recheck under lock: */
 623         if (j->err_seq && seq >= j->err_seq) {
 624                 ret = -EIO;
 625                 goto out;
 626         }
 627
 628         if (seq <= j->flushed_seq_ondisk) {
 629                 ret = 1;
 630                 goto out;
 631         }
 632
 633         /* if seq was written, but not flushed - flush a newer one instead */
 634         seq = max(seq, journal_last_unwritten_seq(j));
 635
 636 recheck_need_open:
 637         if (seq > journal_cur_seq(j)) {
 638                 struct journal_res res = { 0 };
 639
 640                 if (journal_entry_is_open(j))
 641                         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 642
 643                 spin_unlock(&j->lock);
 644
 645                 ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 646                 if (ret)
 647                         return ret;
 648
 649                 seq = res.seq;
 650                 buf = j->buf + (seq & JOURNAL_BUF_MASK);
 651                 buf->must_flush = true;
 652
 653                 if (!buf->flush_time) {
 654                         buf->flush_time = local_clock() ?: 1;
 655                         buf->expires = jiffies;
 656                 }
 657
 658                 if (parent && !closure_wait(&buf->wait, parent))
 659                         BUG();
 660
 661                 bch2_journal_res_put(j, &res);
 662
 663                 spin_lock(&j->lock);
 664                 goto want_write;
 665         }
 666
 667         /*
 668          * if write was kicked off without a flush, flush the next sequence
 669          * number instead
 670          */
 671         buf = journal_seq_to_buf(j, seq);
 672         if (buf->noflush) {
 673                 seq++;
 674                 goto recheck_need_open;
 675         }
 676
 677         buf->must_flush = true;
 678
 679         if (parent && !closure_wait(&buf->wait, parent))
 680                 BUG();
 681 want_write:
 682         if (seq == journal_cur_seq(j))
 683                 journal_entry_want_write(j);
 684 out:
 685         spin_unlock(&j->lock);
 686         return ret;
 687 }
 688
 689 int bch2_journal_flush_seq(struct journal *j, u64 seq)
 690 {
 691         u64 start_time = local_clock();
 692         int ret, ret2;
 693
 694         /*
 695          * Don't update time_stats when @seq is already flushed:
 696          */
 697         if (seq <= j->flushed_seq_ondisk)
 698                 return 0;
 699
 700         ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)));
 701
 702         if (!ret)
 703                 bch2_time_stats_update(j->flush_seq_time, start_time);
 704
 705         return ret ?: ret2 < 0 ? ret2 : 0;
 706 }
 707
 708 /*
 709  * bch2_journal_flush_async - if there is an open journal entry, or a journal
 710  * still being written, write it and wait for the write to complete
 711  */
 712 void bch2_journal_flush_async(struct journal *j, struct closure *parent)
 713 {
 714         bch2_journal_flush_seq_async(j, atomic64_read(&j->seq), parent);
 715 }
 716
 717 int bch2_journal_flush(struct journal *j)
 718 {
 719         return bch2_journal_flush_seq(j, atomic64_read(&j->seq));
 720 }
 721
 722 /*
 723  * bch2_journal_noflush_seq - tell the journal not to issue any flushes before
 724  * @seq
 725  */
 726 bool bch2_journal_noflush_seq(struct journal *j, u64 seq)
 727 {
 728         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 729         u64 unwritten_seq;
 730         bool ret = false;
 731
 732         if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush)))
 733                 return false;
 734
 735         if (seq <= c->journal.flushed_seq_ondisk)
 736                 return false;
 737
 738         spin_lock(&j->lock);
 739         if (seq <= c->journal.flushed_seq_ondisk)
 740                 goto out;
 741
 742         for (unwritten_seq = journal_last_unwritten_seq(j);
 743              unwritten_seq < seq;
 744              unwritten_seq++) {
 745                 struct journal_buf *buf = journal_seq_to_buf(j, unwritten_seq);
 746
 747                 /* journal write is already in flight, and was a flush write: */
 748                 if (unwritten_seq == journal_last_unwritten_seq(j) && !buf->noflush)
 749                         goto out;
 750
 751                 buf->noflush = true;
 752         }
 753
 754         ret = true;
 755 out:
 756         spin_unlock(&j->lock);
 757         return ret;
 758 }
 759
 760 int bch2_journal_meta(struct journal *j)
 761 {
 762         struct journal_buf *buf;
 763         struct journal_res res;
 764         int ret;
 765
 766         memset(&res, 0, sizeof(res));
 767
 768         ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 769         if (ret)
 770                 return ret;
 771
 772         buf = j->buf + (res.seq & JOURNAL_BUF_MASK);
 773         buf->must_flush = true;
 774
 775         if (!buf->flush_time) {
 776                 buf->flush_time = local_clock() ?: 1;
 777                 buf->expires = jiffies;
 778         }
 779
 780         bch2_journal_res_put(j, &res);
 781
 782         return bch2_journal_flush_seq(j, res.seq);
 783 }
 784
 785 /* block/unlock the journal: */
 786
 787 void bch2_journal_unblock(struct journal *j)
 788 {
 789         spin_lock(&j->lock);
 790         j->blocked--;
 791         spin_unlock(&j->lock);
 792
 793         journal_wake(j);
 794 }
 795
 796 void bch2_journal_block(struct journal *j)
 797 {
 798         spin_lock(&j->lock);
 799         j->blocked++;
 800         spin_unlock(&j->lock);
 801
 802         journal_quiesce(j);
 803 }
 804
 805 /* allocate journal on a device: */
 806
 807 static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 808                                          bool new_fs, struct closure *cl)
 809 {
 810         struct bch_fs *c = ca->fs;
 811         struct journal_device *ja = &ca->journal;
 812         u64 *new_bucket_seq = NULL, *new_buckets = NULL;
 813         struct open_bucket **ob = NULL;
 814         long *bu = NULL;
 815         unsigned i, pos, nr_got = 0, nr_want = nr - ja->nr;
 816         int ret = 0;
 817
 818         BUG_ON(nr <= ja->nr);
 819
 820         bu              = kcalloc(nr_want, sizeof(*bu), GFP_KERNEL);
 821         ob              = kcalloc(nr_want, sizeof(*ob), GFP_KERNEL);
 822         new_buckets     = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 823         new_bucket_seq  = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 824         if (!bu || !ob || !new_buckets || !new_bucket_seq) {
 825                 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
 826                 goto err_free;
 827         }
 828
 829         for (nr_got = 0; nr_got < nr_want; nr_got++) {
 830                 if (new_fs) {
 831                         bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
 832                         if (bu[nr_got] < 0) {
 833                                 ret = -BCH_ERR_ENOSPC_bucket_alloc;
 834                                 break;
 835                         }
 836                 } else {
 837                         ob[nr_got] = bch2_bucket_alloc(c, ca, BCH_WATERMARK_normal, cl);
 838                         ret = PTR_ERR_OR_ZERO(ob[nr_got]);
 839                         if (ret)
 840                                 break;
 841
 842                         ret = bch2_trans_run(c,
 843                                 bch2_trans_mark_metadata_bucket(trans, ca,
 844                                                 ob[nr_got]->bucket, BCH_DATA_journal,
 845                                                 ca->mi.bucket_size));
 846                         if (ret) {
 847                                 bch2_open_bucket_put(c, ob[nr_got]);
 848                                 bch_err_msg(c, ret, "marking new journal buckets");
 849                                 break;
 850                         }
 851
 852                         bu[nr_got] = ob[nr_got]->bucket;
 853                 }
 854         }
 855
 856         if (!nr_got)
 857                 goto err_free;
 858
 859         /* Don't return an error if we successfully allocated some buckets: */
 860         ret = 0;
 861
 862         if (c) {
 863                 bch2_journal_flush_all_pins(&c->journal);
 864                 bch2_journal_block(&c->journal);
 865                 mutex_lock(&c->sb_lock);
 866         }
 867
 868         memcpy(new_buckets,     ja->buckets,    ja->nr * sizeof(u64));
 869         memcpy(new_bucket_seq,  ja->bucket_seq, ja->nr * sizeof(u64));
 870
 871         BUG_ON(ja->discard_idx > ja->nr);
 872
 873         pos = ja->discard_idx ?: ja->nr;
 874
 875         memmove(new_buckets + pos + nr_got,
 876                 new_buckets + pos,
 877                 sizeof(new_buckets[0]) * (ja->nr - pos));
 878         memmove(new_bucket_seq + pos + nr_got,
 879                 new_bucket_seq + pos,
 880                 sizeof(new_bucket_seq[0]) * (ja->nr - pos));
 881
 882         for (i = 0; i < nr_got; i++) {
 883                 new_buckets[pos + i] = bu[i];
 884                 new_bucket_seq[pos + i] = 0;
 885         }
 886
 887         nr = ja->nr + nr_got;
 888
 889         ret = bch2_journal_buckets_to_sb(c, ca, new_buckets, nr);
 890         if (ret)
 891                 goto err_unblock;
 892
 893         if (!new_fs)
 894                 bch2_write_super(c);
 895
 896         /* Commit: */
 897         if (c)
 898                 spin_lock(&c->journal.lock);
 899
 900         swap(new_buckets,       ja->buckets);
 901         swap(new_bucket_seq,    ja->bucket_seq);
 902         ja->nr = nr;
 903
 904         if (pos <= ja->discard_idx)
 905                 ja->discard_idx = (ja->discard_idx + nr_got) % ja->nr;
 906         if (pos <= ja->dirty_idx_ondisk)
 907                 ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + nr_got) % ja->nr;
 908         if (pos <= ja->dirty_idx)
 909                 ja->dirty_idx = (ja->dirty_idx + nr_got) % ja->nr;
 910         if (pos <= ja->cur_idx)
 911                 ja->cur_idx = (ja->cur_idx + nr_got) % ja->nr;
 912
 913         if (c)
 914                 spin_unlock(&c->journal.lock);
 915 err_unblock:
 916         if (c) {
 917                 bch2_journal_unblock(&c->journal);
 918                 mutex_unlock(&c->sb_lock);
 919         }
 920
 921         if (ret && !new_fs)
 922                 for (i = 0; i < nr_got; i++)
 923                         bch2_trans_run(c,
 924                                 bch2_trans_mark_metadata_bucket(trans, ca,
 925                                                 bu[i], BCH_DATA_free, 0));
 926 err_free:
 927         if (!new_fs)
 928                 for (i = 0; i < nr_got; i++)
 929                         bch2_open_bucket_put(c, ob[i]);
 930
 931         kfree(new_bucket_seq);
 932         kfree(new_buckets);
 933         kfree(ob);
 934         kfree(bu);
 935         return ret;
 936 }
 937
 938 /*
 939  * Allocate more journal space at runtime - not currently making use if it, but
 940  * the code works:
 941  */
 942 int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 943                                 unsigned nr)
 944 {
 945         struct journal_device *ja = &ca->journal;
 946         struct closure cl;
 947         int ret = 0;
 948
 949         closure_init_stack(&cl);
 950
 951         down_write(&c->state_lock);
 952
 953         /* don't handle reducing nr of buckets yet: */
 954         if (nr < ja->nr)
 955                 goto unlock;
 956
 957         while (ja->nr < nr) {
 958                 struct disk_reservation disk_res = { 0, 0, 0 };
 959
 960                 /*
 961                  * note: journal buckets aren't really counted as _sectors_ used yet, so
 962                  * we don't need the disk reservation to avoid the BUG_ON() in buckets.c
 963                  * when space used goes up without a reservation - but we do need the
 964                  * reservation to ensure we'll actually be able to allocate:
 965                  *
 966                  * XXX: that's not right, disk reservations only ensure a
 967                  * filesystem-wide allocation will succeed, this is a device
 968                  * specific allocation - we can hang here:
 969                  */
 970
 971                 ret = bch2_disk_reservation_get(c, &disk_res,
 972                                                 bucket_to_sector(ca, nr - ja->nr), 1, 0);
 973                 if (ret)
 974                         break;
 975
 976                 ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl);
 977
 978                 bch2_disk_reservation_put(c, &disk_res);
 979
 980                 closure_sync(&cl);
 981
 982                 if (ret && ret != -BCH_ERR_bucket_alloc_blocked)
 983                         break;
 984         }
 985
 986         if (ret)
 987                 bch_err_fn(c, ret);
 988 unlock:
 989         up_write(&c->state_lock);
 990         return ret;
 991 }
 992
 993 int bch2_dev_journal_alloc(struct bch_dev *ca)
 994 {
 995         unsigned nr;
 996         int ret;
 997
 998         if (dynamic_fault("bcachefs:add:journal_alloc")) {
 999                 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
1000                 goto err;
1001         }
1002
1003         /* 1/128th of the device by default: */
1004         nr = ca->mi.nbuckets >> 7;
1005
1006         /*
1007          * clamp journal size to 8192 buckets or 8GB (in sectors), whichever
1008          * is smaller:
1009          */
1010         nr = clamp_t(unsigned, nr,
1011                      BCH_JOURNAL_BUCKETS_MIN,
1012                      min(1 << 13,
1013                          (1 << 24) / ca->mi.bucket_size));
1014
1015         ret = __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
1016 err:
1017         if (ret)
1018                 bch_err_fn(ca, ret);
1019         return ret;
1020 }
1021
1022 int bch2_fs_journal_alloc(struct bch_fs *c)
1023 {
1024         struct bch_dev *ca;
1025         unsigned i;
1026
1027         for_each_online_member(ca, c, i) {
1028                 if (ca->journal.nr)
1029                         continue;
1030
1031                 int ret = bch2_dev_journal_alloc(ca);
1032                 if (ret) {
1033                         percpu_ref_put(&ca->io_ref);
1034                         return ret;
1035                 }
1036         }
1037
1038         return 0;
1039 }
1040
1041 /* startup/shutdown: */
1042
1043 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
1044 {
1045         bool ret = false;
1046         u64 seq;
1047
1048         spin_lock(&j->lock);
1049         for (seq = journal_last_unwritten_seq(j);
1050              seq <= journal_cur_seq(j) && !ret;
1051              seq++) {
1052                 struct journal_buf *buf = journal_seq_to_buf(j, seq);
1053
1054                 if (bch2_bkey_has_device_c(bkey_i_to_s_c(&buf->key), dev_idx))
1055                         ret = true;
1056         }
1057         spin_unlock(&j->lock);
1058
1059         return ret;
1060 }
1061
1062 void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca)
1063 {
1064         wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx));
1065 }
1066
1067 void bch2_fs_journal_stop(struct journal *j)
1068 {
1069         bch2_journal_reclaim_stop(j);
1070         bch2_journal_flush_all_pins(j);
1071
1072         wait_event(j->wait, journal_entry_close(j));
1073
1074         /*
1075          * Always write a new journal entry, to make sure the clock hands are up
1076          * to date (and match the superblock)
1077          */
1078         bch2_journal_meta(j);
1079
1080         journal_quiesce(j);
1081
1082         BUG_ON(!bch2_journal_error(j) &&
1083                test_bit(JOURNAL_REPLAY_DONE, &j->flags) &&
1084                j->last_empty_seq != journal_cur_seq(j));
1085
1086         cancel_delayed_work_sync(&j->write_work);
1087 }
1088
1089 int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
1090 {
1091         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1092         struct journal_entry_pin_list *p;
1093         struct journal_replay *i, **_i;
1094         struct genradix_iter iter;
1095         bool had_entries = false;
1096         unsigned ptr;
1097         u64 last_seq = cur_seq, nr, seq;
1098
1099         genradix_for_each_reverse(&c->journal_entries, iter, _i) {
1100                 i = *_i;
1101
1102                 if (!i || i->ignore)
1103                         continue;
1104
1105                 last_seq = le64_to_cpu(i->j.last_seq);
1106                 break;
1107         }
1108
1109         nr = cur_seq - last_seq;
1110
1111         if (nr + 1 > j->pin.size) {
1112                 free_fifo(&j->pin);
1113                 init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL);
1114                 if (!j->pin.data) {
1115                         bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
1116                         return -BCH_ERR_ENOMEM_journal_pin_fifo;
1117                 }
1118         }
1119
1120         j->replay_journal_seq   = last_seq;
1121         j->replay_journal_seq_end = cur_seq;
1122         j->last_seq_ondisk      = last_seq;
1123         j->flushed_seq_ondisk   = cur_seq - 1;
1124         j->seq_ondisk           = cur_seq - 1;
1125         j->pin.front            = last_seq;
1126         j->pin.back             = cur_seq;
1127         atomic64_set(&j->seq, cur_seq - 1);
1128
1129         fifo_for_each_entry_ptr(p, &j->pin, seq)
1130                 journal_pin_list_init(p, 1);
1131
1132         genradix_for_each(&c->journal_entries, iter, _i) {
1133                 i = *_i;
1134
1135                 if (!i || i->ignore)
1136                         continue;
1137
1138                 seq = le64_to_cpu(i->j.seq);
1139                 BUG_ON(seq >= cur_seq);
1140
1141                 if (seq < last_seq)
1142                         continue;
1143
1144                 if (journal_entry_empty(&i->j))
1145                         j->last_empty_seq = le64_to_cpu(i->j.seq);
1146
1147                 p = journal_seq_pin(j, seq);
1148
1149                 p->devs.nr = 0;
1150                 for (ptr = 0; ptr < i->nr_ptrs; ptr++)
1151                         bch2_dev_list_add_dev(&p->devs, i->ptrs[ptr].dev);
1152
1153                 had_entries = true;
1154         }
1155
1156         if (!had_entries)
1157                 j->last_empty_seq = cur_seq;
1158
1159         spin_lock(&j->lock);
1160
1161         set_bit(JOURNAL_STARTED, &j->flags);
1162         j->last_flush_write = jiffies;
1163
1164         j->reservations.idx = j->reservations.unwritten_idx = journal_cur_seq(j);
1165         j->reservations.unwritten_idx++;
1166
1167         c->last_bucket_seq_cleanup = journal_cur_seq(j);
1168
1169         bch2_journal_space_available(j);
1170         spin_unlock(&j->lock);
1171
1172         return bch2_journal_reclaim_start(j);
1173 }
1174
1175 /* init/exit: */
1176
1177 void bch2_dev_journal_exit(struct bch_dev *ca)
1178 {
1179         kfree(ca->journal.bio);
1180         kfree(ca->journal.buckets);
1181         kfree(ca->journal.bucket_seq);
1182
1183         ca->journal.bio         = NULL;
1184         ca->journal.buckets     = NULL;
1185         ca->journal.bucket_seq  = NULL;
1186 }
1187
1188 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
1189 {
1190         struct journal_device *ja = &ca->journal;
1191         struct bch_sb_field_journal *journal_buckets =
1192                 bch2_sb_field_get(sb, journal);
1193         struct bch_sb_field_journal_v2 *journal_buckets_v2 =
1194                 bch2_sb_field_get(sb, journal_v2);
1195         unsigned i, nr_bvecs;
1196
1197         ja->nr = 0;
1198
1199         if (journal_buckets_v2) {
1200                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1201
1202                 for (i = 0; i < nr; i++)
1203                         ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr);
1204         } else if (journal_buckets) {
1205                 ja->nr = bch2_nr_journal_buckets(journal_buckets);
1206         }
1207
1208         ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1209         if (!ja->bucket_seq)
1210                 return -BCH_ERR_ENOMEM_dev_journal_init;
1211
1212         nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
1213
1214         ca->journal.bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1215         if (!ca->journal.bio)
1216                 return -BCH_ERR_ENOMEM_dev_journal_init;
1217
1218         bio_init(ca->journal.bio, NULL, ca->journal.bio->bi_inline_vecs, nr_bvecs, 0);
1219
1220         ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1221         if (!ja->buckets)
1222                 return -BCH_ERR_ENOMEM_dev_journal_init;
1223
1224         if (journal_buckets_v2) {
1225                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1226                 unsigned j, dst = 0;
1227
1228                 for (i = 0; i < nr; i++)
1229                         for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
1230                                 ja->buckets[dst++] =
1231                                         le64_to_cpu(journal_buckets_v2->d[i].start) + j;
1232         } else if (journal_buckets) {
1233                 for (i = 0; i < ja->nr; i++)
1234                         ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
1235         }
1236
1237         return 0;
1238 }
1239
1240 void bch2_fs_journal_exit(struct journal *j)
1241 {
1242         unsigned i;
1243
1244         darray_exit(&j->early_journal_entries);
1245
1246         for (i = 0; i < ARRAY_SIZE(j->buf); i++)
1247                 kvpfree(j->buf[i].data, j->buf[i].buf_size);
1248         free_fifo(&j->pin);
1249 }
1250
1251 int bch2_fs_journal_init(struct journal *j)
1252 {
1253         static struct lock_class_key res_key;
1254         unsigned i;
1255
1256         spin_lock_init(&j->lock);
1257         spin_lock_init(&j->err_lock);
1258         init_waitqueue_head(&j->wait);
1259         INIT_DELAYED_WORK(&j->write_work, journal_write_work);
1260         init_waitqueue_head(&j->reclaim_wait);
1261         init_waitqueue_head(&j->pin_flush_wait);
1262         mutex_init(&j->reclaim_lock);
1263         mutex_init(&j->discard_lock);
1264
1265         lockdep_init_map(&j->res_map, "journal res", &res_key, 0);
1266
1267         atomic64_set(&j->reservations.counter,
1268                 ((union journal_res_state)
1269                  { .cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL }).v);
1270
1271         if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)))
1272                 return -BCH_ERR_ENOMEM_journal_pin_fifo;
1273
1274         for (i = 0; i < ARRAY_SIZE(j->buf); i++) {
1275                 j->buf[i].buf_size = JOURNAL_ENTRY_SIZE_MIN;
1276                 j->buf[i].data = kvpmalloc(j->buf[i].buf_size, GFP_KERNEL);
1277                 if (!j->buf[i].data)
1278                         return -BCH_ERR_ENOMEM_journal_buf;
1279         }
1280
1281         j->pin.front = j->pin.back = 1;
1282         return 0;
1283 }
1284
1285 /* debug: */
1286
1287 void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1288 {
1289         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1290         union journal_res_state s;
1291         struct bch_dev *ca;
1292         unsigned long now = jiffies;
1293         u64 seq;
1294         unsigned i;
1295
1296         if (!out->nr_tabstops)
1297                 printbuf_tabstop_push(out, 24);
1298         out->atomic++;
1299
1300         rcu_read_lock();
1301         s = READ_ONCE(j->reservations);
1302
1303         prt_printf(out, "dirty journal entries:\t%llu/%llu\n",  fifo_used(&j->pin), j->pin.size);
1304         prt_printf(out, "seq:\t\t\t%llu\n",                     journal_cur_seq(j));
1305         prt_printf(out, "seq_ondisk:\t\t%llu\n",                j->seq_ondisk);
1306         prt_printf(out, "last_seq:\t\t%llu\n",          journal_last_seq(j));
1307         prt_printf(out, "last_seq_ondisk:\t%llu\n",             j->last_seq_ondisk);
1308         prt_printf(out, "flushed_seq_ondisk:\t%llu\n",  j->flushed_seq_ondisk);
1309         prt_printf(out, "prereserved:\t\t%u/%u\n",              j->prereserved.reserved, j->prereserved.remaining);
1310         prt_printf(out, "watermark:\t\t%s\n",           bch2_watermarks[j->watermark]);
1311         prt_printf(out, "each entry reserved:\t%u\n",   j->entry_u64s_reserved);
1312         prt_printf(out, "nr flush writes:\t%llu\n",             j->nr_flush_writes);
1313         prt_printf(out, "nr noflush writes:\t%llu\n",   j->nr_noflush_writes);
1314         prt_printf(out, "nr direct reclaim:\t%llu\n",   j->nr_direct_reclaim);
1315         prt_printf(out, "nr background reclaim:\t%llu\n",       j->nr_background_reclaim);
1316         prt_printf(out, "reclaim kicked:\t\t%u\n",              j->reclaim_kicked);
1317         prt_printf(out, "reclaim runs in:\t%u ms\n",    time_after(j->next_reclaim, now)
1318                ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0);
1319         prt_printf(out, "current entry sectors:\t%u\n", j->cur_entry_sectors);
1320         prt_printf(out, "current entry error:\t%s\n",   bch2_journal_errors[j->cur_entry_error]);
1321         prt_printf(out, "current entry:\t\t");
1322
1323         switch (s.cur_entry_offset) {
1324         case JOURNAL_ENTRY_ERROR_VAL:
1325                 prt_printf(out, "error");
1326                 break;
1327         case JOURNAL_ENTRY_CLOSED_VAL:
1328                 prt_printf(out, "closed");
1329                 break;
1330         default:
1331                 prt_printf(out, "%u/%u", s.cur_entry_offset, j->cur_entry_u64s);
1332                 break;
1333         }
1334
1335         prt_newline(out);
1336
1337         for (seq = journal_cur_seq(j);
1338              seq >= journal_last_unwritten_seq(j);
1339              --seq) {
1340                 i = seq & JOURNAL_BUF_MASK;
1341
1342                 prt_printf(out, "unwritten entry:");
1343                 prt_tab(out);
1344                 prt_printf(out, "%llu", seq);
1345                 prt_newline(out);
1346                 printbuf_indent_add(out, 2);
1347
1348                 prt_printf(out, "refcount:");
1349                 prt_tab(out);
1350                 prt_printf(out, "%u", journal_state_count(s, i));
1351                 prt_newline(out);
1352
1353                 prt_printf(out, "sectors:");
1354                 prt_tab(out);
1355                 prt_printf(out, "%u", j->buf[i].sectors);
1356                 prt_newline(out);
1357
1358                 prt_printf(out, "expires");
1359                 prt_tab(out);
1360                 prt_printf(out, "%li jiffies", j->buf[i].expires - jiffies);
1361                 prt_newline(out);
1362
1363                 printbuf_indent_sub(out, 2);
1364         }
1365
1366         prt_printf(out,
1367                "replay done:\t\t%i\n",
1368                test_bit(JOURNAL_REPLAY_DONE,    &j->flags));
1369
1370         prt_printf(out, "space:\n");
1371         prt_printf(out, "\tdiscarded\t%u:%u\n",
1372                j->space[journal_space_discarded].next_entry,
1373                j->space[journal_space_discarded].total);
1374         prt_printf(out, "\tclean ondisk\t%u:%u\n",
1375                j->space[journal_space_clean_ondisk].next_entry,
1376                j->space[journal_space_clean_ondisk].total);
1377         prt_printf(out, "\tclean\t\t%u:%u\n",
1378                j->space[journal_space_clean].next_entry,
1379                j->space[journal_space_clean].total);
1380         prt_printf(out, "\ttotal\t\t%u:%u\n",
1381                j->space[journal_space_total].next_entry,
1382                j->space[journal_space_total].total);
1383
1384         for_each_member_device_rcu(ca, c, i,
1385                                    &c->rw_devs[BCH_DATA_journal]) {
1386                 struct journal_device *ja = &ca->journal;
1387
1388                 if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d))
1389                         continue;
1390
1391                 if (!ja->nr)
1392                         continue;
1393
1394                 prt_printf(out, "dev %u:\n",            i);
1395                 prt_printf(out, "\tnr\t\t%u\n",         ja->nr);
1396                 prt_printf(out, "\tbucket size\t%u\n",  ca->mi.bucket_size);
1397                 prt_printf(out, "\tavailable\t%u:%u\n", bch2_journal_dev_buckets_available(j, ja, journal_space_discarded), ja->sectors_free);
1398                 prt_printf(out, "\tdiscard_idx\t%u\n",  ja->discard_idx);
1399                 prt_printf(out, "\tdirty_ondisk\t%u (seq %llu)\n", ja->dirty_idx_ondisk,        ja->bucket_seq[ja->dirty_idx_ondisk]);
1400                 prt_printf(out, "\tdirty_idx\t%u (seq %llu)\n", ja->dirty_idx,          ja->bucket_seq[ja->dirty_idx]);
1401                 prt_printf(out, "\tcur_idx\t\t%u (seq %llu)\n", ja->cur_idx,            ja->bucket_seq[ja->cur_idx]);
1402         }
1403
1404         rcu_read_unlock();
1405
1406         --out->atomic;
1407 }
1408
1409 void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1410 {
1411         spin_lock(&j->lock);
1412         __bch2_journal_debug_to_text(out, j);
1413         spin_unlock(&j->lock);
1414 }
1415
1416 bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq)
1417 {
1418         struct journal_entry_pin_list *pin_list;
1419         struct journal_entry_pin *pin;
1420         unsigned i;
1421
1422         spin_lock(&j->lock);
1423         *seq = max(*seq, j->pin.front);
1424
1425         if (*seq >= j->pin.back) {
1426                 spin_unlock(&j->lock);
1427                 return true;
1428         }
1429
1430         out->atomic++;
1431
1432         pin_list = journal_seq_pin(j, *seq);
1433
1434         prt_printf(out, "%llu: count %u", *seq, atomic_read(&pin_list->count));
1435         prt_newline(out);
1436         printbuf_indent_add(out, 2);
1437
1438         for (i = 0; i < ARRAY_SIZE(pin_list->list); i++)
1439                 list_for_each_entry(pin, &pin_list->list[i], list) {
1440                         prt_printf(out, "\t%px %ps", pin, pin->flush);
1441                         prt_newline(out);
1442                 }
1443
1444         if (!list_empty(&pin_list->flushed)) {
1445                 prt_printf(out, "flushed:");
1446                 prt_newline(out);
1447         }
1448
1449         list_for_each_entry(pin, &pin_list->flushed, list) {
1450                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1451                 prt_newline(out);
1452         }
1453
1454         printbuf_indent_sub(out, 2);
1455
1456         --out->atomic;
1457         spin_unlock(&j->lock);
1458
1459         return false;
1460 }
1461
1462 void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
1463 {
1464         u64 seq = 0;
1465
1466         while (!bch2_journal_seq_pins_to_text(out, j, &seq))
1467                 seq++;
1468 }