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
  21 #include <trace/events/bcachefs.h>
  22
  23 #define x(n)    #n,
  24 static const char * const bch2_journal_watermarks[] = {
  25         JOURNAL_WATERMARKS()
  26         NULL
  27 };
  28
  29 static const char * const bch2_journal_errors[] = {
  30         JOURNAL_ERRORS()
  31         NULL
  32 };
  33 #undef x
  34
  35 static inline bool journal_seq_unwritten(struct journal *j, u64 seq)
  36 {
  37         return seq > j->seq_ondisk;
  38 }
  39
  40 static bool __journal_entry_is_open(union journal_res_state state)
  41 {
  42         return state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL;
  43 }
  44
  45 static inline unsigned nr_unwritten_journal_entries(struct journal *j)
  46 {
  47         return atomic64_read(&j->seq) - j->seq_ondisk;
  48 }
  49
  50 static bool journal_entry_is_open(struct journal *j)
  51 {
  52         return __journal_entry_is_open(j->reservations);
  53 }
  54
  55 static inline struct journal_buf *
  56 journal_seq_to_buf(struct journal *j, u64 seq)
  57 {
  58         struct journal_buf *buf = NULL;
  59
  60         EBUG_ON(seq > journal_cur_seq(j));
  61
  62         if (journal_seq_unwritten(j, seq)) {
  63                 buf = j->buf + (seq & JOURNAL_BUF_MASK);
  64                 EBUG_ON(le64_to_cpu(buf->data->seq) != seq);
  65         }
  66         return buf;
  67 }
  68
  69 static void journal_pin_list_init(struct journal_entry_pin_list *p, int count)
  70 {
  71         INIT_LIST_HEAD(&p->list);
  72         INIT_LIST_HEAD(&p->key_cache_list);
  73         INIT_LIST_HEAD(&p->flushed);
  74         atomic_set(&p->count, count);
  75         p->devs.nr = 0;
  76 }
  77
  78 /* journal entry close/open: */
  79
  80 void __bch2_journal_buf_put(struct journal *j)
  81 {
  82         struct bch_fs *c = container_of(j, struct bch_fs, journal);
  83
  84         closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
  85 }
  86
  87 /*
  88  * Returns true if journal entry is now closed:
  89  *
  90  * We don't close a journal_buf until the next journal_buf is finished writing,
  91  * and can be opened again - this also initializes the next journal_buf:
  92  */
  93 static void __journal_entry_close(struct journal *j, unsigned closed_val)
  94 {
  95         struct bch_fs *c = container_of(j, struct bch_fs, journal);
  96         struct journal_buf *buf = journal_cur_buf(j);
  97         union journal_res_state old, new;
  98         u64 v = atomic64_read(&j->reservations.counter);
  99         unsigned sectors;
 100
 101         BUG_ON(closed_val != JOURNAL_ENTRY_CLOSED_VAL &&
 102                closed_val != JOURNAL_ENTRY_ERROR_VAL);
 103
 104         lockdep_assert_held(&j->lock);
 105
 106         do {
 107                 old.v = new.v = v;
 108                 new.cur_entry_offset = closed_val;
 109
 110                 if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL ||
 111                     old.cur_entry_offset == new.cur_entry_offset)
 112                         return;
 113         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 114                                        old.v, new.v)) != old.v);
 115
 116         if (!__journal_entry_is_open(old))
 117                 return;
 118
 119         /* Close out old buffer: */
 120         buf->data->u64s         = cpu_to_le32(old.cur_entry_offset);
 121
 122         sectors = vstruct_blocks_plus(buf->data, c->block_bits,
 123                                       buf->u64s_reserved) << c->block_bits;
 124         BUG_ON(sectors > buf->sectors);
 125         buf->sectors = sectors;
 126
 127         /*
 128          * We have to set last_seq here, _before_ opening a new journal entry:
 129          *
 130          * A threads may replace an old pin with a new pin on their current
 131          * journal reservation - the expectation being that the journal will
 132          * contain either what the old pin protected or what the new pin
 133          * protects.
 134          *
 135          * After the old pin is dropped journal_last_seq() won't include the old
 136          * pin, so we can only write the updated last_seq on the entry that
 137          * contains whatever the new pin protects.
 138          *
 139          * Restated, we can _not_ update last_seq for a given entry if there
 140          * could be a newer entry open with reservations/pins that have been
 141          * taken against it.
 142          *
 143          * Hence, we want update/set last_seq on the current journal entry right
 144          * before we open a new one:
 145          */
 146         buf->last_seq           = journal_last_seq(j);
 147         buf->data->last_seq     = cpu_to_le64(buf->last_seq);
 148         BUG_ON(buf->last_seq > le64_to_cpu(buf->data->seq));
 149
 150         __bch2_journal_pin_put(j, le64_to_cpu(buf->data->seq));
 151
 152         cancel_delayed_work(&j->write_work);
 153
 154         bch2_journal_space_available(j);
 155
 156         bch2_journal_buf_put(j, old.idx);
 157 }
 158
 159 void bch2_journal_halt(struct journal *j)
 160 {
 161         spin_lock(&j->lock);
 162         __journal_entry_close(j, JOURNAL_ENTRY_ERROR_VAL);
 163         if (!j->err_seq)
 164                 j->err_seq = journal_cur_seq(j);
 165         spin_unlock(&j->lock);
 166 }
 167
 168 static bool journal_entry_want_write(struct journal *j)
 169 {
 170         bool ret = !journal_entry_is_open(j) ||
 171                 journal_cur_seq(j) == journal_last_unwritten_seq(j);
 172
 173         /* Don't close it yet if we already have a write in flight: */
 174         if (ret)
 175                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 176         else if (nr_unwritten_journal_entries(j)) {
 177                 struct journal_buf *buf = journal_cur_buf(j);
 178
 179                 if (!buf->flush_time) {
 180                         buf->flush_time = local_clock() ?: 1;
 181                         buf->expires = jiffies;
 182                 }
 183         }
 184
 185         return ret;
 186 }
 187
 188 static bool journal_entry_close(struct journal *j)
 189 {
 190         bool ret;
 191
 192         spin_lock(&j->lock);
 193         ret = journal_entry_want_write(j);
 194         spin_unlock(&j->lock);
 195
 196         return ret;
 197 }
 198
 199 /*
 200  * should _only_ called from journal_res_get() - when we actually want a
 201  * journal reservation - journal entry is open means journal is dirty:
 202  */
 203 static int journal_entry_open(struct journal *j)
 204 {
 205         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 206         struct journal_buf *buf = j->buf +
 207                 ((journal_cur_seq(j) + 1) & JOURNAL_BUF_MASK);
 208         union journal_res_state old, new;
 209         int u64s;
 210         u64 v;
 211
 212         lockdep_assert_held(&j->lock);
 213         BUG_ON(journal_entry_is_open(j));
 214         BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
 215
 216         if (j->blocked)
 217                 return JOURNAL_ERR_blocked;
 218
 219         if (j->cur_entry_error)
 220                 return j->cur_entry_error;
 221
 222         if (bch2_journal_error(j))
 223                 return JOURNAL_ERR_insufficient_devices; /* -EROFS */
 224
 225         if (!fifo_free(&j->pin))
 226                 return JOURNAL_ERR_journal_pin_full;
 227
 228         if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf))
 229                 return JOURNAL_ERR_max_in_flight;
 230
 231         BUG_ON(!j->cur_entry_sectors);
 232
 233         buf->expires            =
 234                 (journal_cur_seq(j) == j->flushed_seq_ondisk
 235                  ? jiffies
 236                  : j->last_flush_write) +
 237                 msecs_to_jiffies(c->opts.journal_flush_delay);
 238
 239         buf->u64s_reserved      = j->entry_u64s_reserved;
 240         buf->disk_sectors       = j->cur_entry_sectors;
 241         buf->sectors            = min(buf->disk_sectors, buf->buf_size >> 9);
 242
 243         u64s = (int) (buf->sectors << 9) / sizeof(u64) -
 244                 journal_entry_overhead(j);
 245         u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 246
 247         if (u64s <= (ssize_t) j->early_journal_entries.nr)
 248                 return JOURNAL_ERR_journal_full;
 249
 250         if (fifo_empty(&j->pin) && j->reclaim_thread)
 251                 wake_up_process(j->reclaim_thread);
 252
 253         /*
 254          * The fifo_push() needs to happen at the same time as j->seq is
 255          * incremented for journal_last_seq() to be calculated correctly
 256          */
 257         atomic64_inc(&j->seq);
 258         journal_pin_list_init(fifo_push_ref(&j->pin), 1);
 259
 260         BUG_ON(j->buf + (journal_cur_seq(j) & JOURNAL_BUF_MASK) != buf);
 261
 262         bkey_extent_init(&buf->key);
 263         buf->noflush    = false;
 264         buf->must_flush = false;
 265         buf->separate_flush = false;
 266         buf->flush_time = 0;
 267
 268         memset(buf->data, 0, sizeof(*buf->data));
 269         buf->data->seq  = cpu_to_le64(journal_cur_seq(j));
 270         buf->data->u64s = 0;
 271
 272         if (j->early_journal_entries.nr) {
 273                 memcpy(buf->data->_data, j->early_journal_entries.data,
 274                        j->early_journal_entries.nr * sizeof(u64));
 275                 le32_add_cpu(&buf->data->u64s, j->early_journal_entries.nr);
 276         }
 277
 278         /*
 279          * Must be set before marking the journal entry as open:
 280          */
 281         j->cur_entry_u64s = u64s;
 282
 283         v = atomic64_read(&j->reservations.counter);
 284         do {
 285                 old.v = new.v = v;
 286
 287                 BUG_ON(old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL);
 288
 289                 new.idx++;
 290                 BUG_ON(journal_state_count(new, new.idx));
 291                 BUG_ON(new.idx != (journal_cur_seq(j) & JOURNAL_BUF_MASK));
 292
 293                 journal_state_inc(&new);
 294
 295                 /* Handle any already added entries */
 296                 new.cur_entry_offset = le32_to_cpu(buf->data->u64s);
 297         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 298                                        old.v, new.v)) != old.v);
 299
 300         if (j->res_get_blocked_start)
 301                 bch2_time_stats_update(j->blocked_time,
 302                                        j->res_get_blocked_start);
 303         j->res_get_blocked_start = 0;
 304
 305         mod_delayed_work(c->io_complete_wq,
 306                          &j->write_work,
 307                          msecs_to_jiffies(c->opts.journal_flush_delay));
 308         journal_wake(j);
 309
 310         if (j->early_journal_entries.nr)
 311                 darray_exit(&j->early_journal_entries);
 312         return 0;
 313 }
 314
 315 static bool journal_quiesced(struct journal *j)
 316 {
 317         bool ret = atomic64_read(&j->seq) == j->seq_ondisk;
 318
 319         if (!ret)
 320                 journal_entry_close(j);
 321         return ret;
 322 }
 323
 324 static void journal_quiesce(struct journal *j)
 325 {
 326         wait_event(j->wait, journal_quiesced(j));
 327 }
 328
 329 static void journal_write_work(struct work_struct *work)
 330 {
 331         struct journal *j = container_of(work, struct journal, write_work.work);
 332         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 333         long delta;
 334
 335         spin_lock(&j->lock);
 336         if (!__journal_entry_is_open(j->reservations))
 337                 goto unlock;
 338
 339         delta = journal_cur_buf(j)->expires - jiffies;
 340
 341         if (delta > 0)
 342                 mod_delayed_work(c->io_complete_wq, &j->write_work, delta);
 343         else
 344                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 345 unlock:
 346         spin_unlock(&j->lock);
 347 }
 348
 349 static int __journal_res_get(struct journal *j, struct journal_res *res,
 350                              unsigned flags)
 351 {
 352         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 353         struct journal_buf *buf;
 354         bool can_discard;
 355         int ret;
 356 retry:
 357         if (journal_res_get_fast(j, res, flags))
 358                 return 0;
 359
 360         if (bch2_journal_error(j))
 361                 return -BCH_ERR_erofs_journal_err;
 362
 363         spin_lock(&j->lock);
 364
 365         /*
 366          * Recheck after taking the lock, so we don't race with another thread
 367          * that just did journal_entry_open() and call journal_entry_close()
 368          * unnecessarily
 369          */
 370         if (journal_res_get_fast(j, res, flags)) {
 371                 spin_unlock(&j->lock);
 372                 return 0;
 373         }
 374
 375         if ((flags & JOURNAL_WATERMARK_MASK) < j->watermark) {
 376                 /*
 377                  * Don't want to close current journal entry, just need to
 378                  * invoke reclaim:
 379                  */
 380                 ret = JOURNAL_ERR_journal_full;
 381                 goto unlock;
 382         }
 383
 384         /*
 385          * If we couldn't get a reservation because the current buf filled up,
 386          * and we had room for a bigger entry on disk, signal that we want to
 387          * realloc the journal bufs:
 388          */
 389         buf = journal_cur_buf(j);
 390         if (journal_entry_is_open(j) &&
 391             buf->buf_size >> 9 < buf->disk_sectors &&
 392             buf->buf_size < JOURNAL_ENTRY_SIZE_MAX)
 393                 j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1);
 394
 395         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 396         ret = journal_entry_open(j);
 397
 398         if (ret == JOURNAL_ERR_max_in_flight)
 399                 trace_and_count(c, journal_entry_full, c);
 400 unlock:
 401         if ((ret && ret != JOURNAL_ERR_insufficient_devices) &&
 402             !j->res_get_blocked_start) {
 403                 j->res_get_blocked_start = local_clock() ?: 1;
 404                 trace_and_count(c, journal_full, c);
 405         }
 406
 407         can_discard = j->can_discard;
 408         spin_unlock(&j->lock);
 409
 410         if (!ret)
 411                 goto retry;
 412
 413         if ((ret == JOURNAL_ERR_journal_full ||
 414              ret == JOURNAL_ERR_journal_pin_full) &&
 415             !can_discard &&
 416             !nr_unwritten_journal_entries(j) &&
 417             (flags & JOURNAL_WATERMARK_MASK) == JOURNAL_WATERMARK_reserved) {
 418                 struct printbuf buf = PRINTBUF;
 419
 420                 bch_err(c, "Journal stuck! Hava a pre-reservation but journal full (ret %s)",
 421                         bch2_journal_errors[ret]);
 422
 423                 bch2_journal_debug_to_text(&buf, j);
 424                 bch_err(c, "%s", buf.buf);
 425
 426                 printbuf_reset(&buf);
 427                 bch2_journal_pins_to_text(&buf, j);
 428                 bch_err(c, "Journal pins:\n%s", buf.buf);
 429
 430                 printbuf_exit(&buf);
 431                 bch2_fatal_error(c);
 432                 dump_stack();
 433         }
 434
 435         /*
 436          * Journal is full - can't rely on reclaim from work item due to
 437          * freezing:
 438          */
 439         if ((ret == JOURNAL_ERR_journal_full ||
 440              ret == JOURNAL_ERR_journal_pin_full) &&
 441             !(flags & JOURNAL_RES_GET_NONBLOCK)) {
 442                 if (can_discard) {
 443                         bch2_journal_do_discards(j);
 444                         goto retry;
 445                 }
 446
 447                 if (mutex_trylock(&j->reclaim_lock)) {
 448                         bch2_journal_reclaim(j);
 449                         mutex_unlock(&j->reclaim_lock);
 450                 }
 451         }
 452
 453         return ret == JOURNAL_ERR_insufficient_devices
 454                 ? -EROFS
 455                 : -BCH_ERR_journal_res_get_blocked;
 456 }
 457
 458 /*
 459  * Essentially the entry function to the journaling code. When bcachefs is doing
 460  * a btree insert, it calls this function to get the current journal write.
 461  * Journal write is the structure used set up journal writes. The calling
 462  * function will then add its keys to the structure, queuing them for the next
 463  * write.
 464  *
 465  * To ensure forward progress, the current task must not be holding any
 466  * btree node write locks.
 467  */
 468 int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
 469                                   unsigned flags)
 470 {
 471         int ret;
 472
 473         closure_wait_event(&j->async_wait,
 474                    (ret = __journal_res_get(j, res, flags)) !=
 475                    -BCH_ERR_journal_res_get_blocked||
 476                    (flags & JOURNAL_RES_GET_NONBLOCK));
 477         return ret;
 478 }
 479
 480 /* journal_preres: */
 481
 482 static bool journal_preres_available(struct journal *j,
 483                                      struct journal_preres *res,
 484                                      unsigned new_u64s,
 485                                      unsigned flags)
 486 {
 487         bool ret = bch2_journal_preres_get_fast(j, res, new_u64s, flags, true);
 488
 489         if (!ret && mutex_trylock(&j->reclaim_lock)) {
 490                 bch2_journal_reclaim(j);
 491                 mutex_unlock(&j->reclaim_lock);
 492         }
 493
 494         return ret;
 495 }
 496
 497 int __bch2_journal_preres_get(struct journal *j,
 498                               struct journal_preres *res,
 499                               unsigned new_u64s,
 500                               unsigned flags)
 501 {
 502         int ret;
 503
 504         closure_wait_event(&j->preres_wait,
 505                    (ret = bch2_journal_error(j)) ||
 506                    journal_preres_available(j, res, new_u64s, flags));
 507         return ret;
 508 }
 509
 510 /* journal_entry_res: */
 511
 512 void bch2_journal_entry_res_resize(struct journal *j,
 513                                    struct journal_entry_res *res,
 514                                    unsigned new_u64s)
 515 {
 516         union journal_res_state state;
 517         int d = new_u64s - res->u64s;
 518
 519         spin_lock(&j->lock);
 520
 521         j->entry_u64s_reserved += d;
 522         if (d <= 0)
 523                 goto out;
 524
 525         j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d);
 526         smp_mb();
 527         state = READ_ONCE(j->reservations);
 528
 529         if (state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL &&
 530             state.cur_entry_offset > j->cur_entry_u64s) {
 531                 j->cur_entry_u64s += d;
 532                 /*
 533                  * Not enough room in current journal entry, have to flush it:
 534                  */
 535                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 536         } else {
 537                 journal_cur_buf(j)->u64s_reserved += d;
 538         }
 539 out:
 540         spin_unlock(&j->lock);
 541         res->u64s += d;
 542 }
 543
 544 /* journal flushing: */
 545
 546 /**
 547  * bch2_journal_flush_seq_async - wait for a journal entry to be written
 548  *
 549  * like bch2_journal_wait_on_seq, except that it triggers a write immediately if
 550  * necessary
 551  */
 552 int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 553                                  struct closure *parent)
 554 {
 555         struct journal_buf *buf;
 556         int ret = 0;
 557
 558         if (seq <= j->flushed_seq_ondisk)
 559                 return 1;
 560
 561         spin_lock(&j->lock);
 562
 563         if (WARN_ONCE(seq > journal_cur_seq(j),
 564                       "requested to flush journal seq %llu, but currently at %llu",
 565                       seq, journal_cur_seq(j)))
 566                 goto out;
 567
 568         /* Recheck under lock: */
 569         if (j->err_seq && seq >= j->err_seq) {
 570                 ret = -EIO;
 571                 goto out;
 572         }
 573
 574         if (seq <= j->flushed_seq_ondisk) {
 575                 ret = 1;
 576                 goto out;
 577         }
 578
 579         /* if seq was written, but not flushed - flush a newer one instead */
 580         seq = max(seq, journal_last_unwritten_seq(j));
 581
 582 recheck_need_open:
 583         if (seq > journal_cur_seq(j)) {
 584                 struct journal_res res = { 0 };
 585
 586                 if (journal_entry_is_open(j))
 587                         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 588
 589                 spin_unlock(&j->lock);
 590
 591                 ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 592                 if (ret)
 593                         return ret;
 594
 595                 seq = res.seq;
 596                 buf = j->buf + (seq & JOURNAL_BUF_MASK);
 597                 buf->must_flush = true;
 598
 599                 if (!buf->flush_time) {
 600                         buf->flush_time = local_clock() ?: 1;
 601                         buf->expires = jiffies;
 602                 }
 603
 604                 if (parent && !closure_wait(&buf->wait, parent))
 605                         BUG();
 606
 607                 bch2_journal_res_put(j, &res);
 608
 609                 spin_lock(&j->lock);
 610                 goto want_write;
 611         }
 612
 613         /*
 614          * if write was kicked off without a flush, flush the next sequence
 615          * number instead
 616          */
 617         buf = journal_seq_to_buf(j, seq);
 618         if (buf->noflush) {
 619                 seq++;
 620                 goto recheck_need_open;
 621         }
 622
 623         buf->must_flush = true;
 624
 625         if (parent && !closure_wait(&buf->wait, parent))
 626                 BUG();
 627 want_write:
 628         if (seq == journal_cur_seq(j))
 629                 journal_entry_want_write(j);
 630 out:
 631         spin_unlock(&j->lock);
 632         return ret;
 633 }
 634
 635 int bch2_journal_flush_seq(struct journal *j, u64 seq)
 636 {
 637         u64 start_time = local_clock();
 638         int ret, ret2;
 639
 640         /*
 641          * Don't update time_stats when @seq is already flushed:
 642          */
 643         if (seq <= j->flushed_seq_ondisk)
 644                 return 0;
 645
 646         ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)));
 647
 648         if (!ret)
 649                 bch2_time_stats_update(j->flush_seq_time, start_time);
 650
 651         return ret ?: ret2 < 0 ? ret2 : 0;
 652 }
 653
 654 /*
 655  * bch2_journal_flush_async - if there is an open journal entry, or a journal
 656  * still being written, write it and wait for the write to complete
 657  */
 658 void bch2_journal_flush_async(struct journal *j, struct closure *parent)
 659 {
 660         bch2_journal_flush_seq_async(j, atomic64_read(&j->seq), parent);
 661 }
 662
 663 int bch2_journal_flush(struct journal *j)
 664 {
 665         return bch2_journal_flush_seq(j, atomic64_read(&j->seq));
 666 }
 667
 668 /*
 669  * bch2_journal_noflush_seq - tell the journal not to issue any flushes before
 670  * @seq
 671  */
 672 bool bch2_journal_noflush_seq(struct journal *j, u64 seq)
 673 {
 674         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 675         u64 unwritten_seq;
 676         bool ret = false;
 677
 678         if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush)))
 679                 return false;
 680
 681         if (seq <= c->journal.flushed_seq_ondisk)
 682                 return false;
 683
 684         spin_lock(&j->lock);
 685         if (seq <= c->journal.flushed_seq_ondisk)
 686                 goto out;
 687
 688         for (unwritten_seq = journal_last_unwritten_seq(j);
 689              unwritten_seq < seq;
 690              unwritten_seq++) {
 691                 struct journal_buf *buf = journal_seq_to_buf(j, unwritten_seq);
 692
 693                 /* journal write is already in flight, and was a flush write: */
 694                 if (unwritten_seq == journal_last_unwritten_seq(j) && !buf->noflush)
 695                         goto out;
 696
 697                 buf->noflush = true;
 698         }
 699
 700         ret = true;
 701 out:
 702         spin_unlock(&j->lock);
 703         return ret;
 704 }
 705
 706 int bch2_journal_meta(struct journal *j)
 707 {
 708         struct journal_buf *buf;
 709         struct journal_res res;
 710         int ret;
 711
 712         memset(&res, 0, sizeof(res));
 713
 714         ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 715         if (ret)
 716                 return ret;
 717
 718         buf = j->buf + (res.seq & JOURNAL_BUF_MASK);
 719         buf->must_flush = true;
 720
 721         if (!buf->flush_time) {
 722                 buf->flush_time = local_clock() ?: 1;
 723                 buf->expires = jiffies;
 724         }
 725
 726         bch2_journal_res_put(j, &res);
 727
 728         return bch2_journal_flush_seq(j, res.seq);
 729 }
 730
 731 /* block/unlock the journal: */
 732
 733 void bch2_journal_unblock(struct journal *j)
 734 {
 735         spin_lock(&j->lock);
 736         j->blocked--;
 737         spin_unlock(&j->lock);
 738
 739         journal_wake(j);
 740 }
 741
 742 void bch2_journal_block(struct journal *j)
 743 {
 744         spin_lock(&j->lock);
 745         j->blocked++;
 746         spin_unlock(&j->lock);
 747
 748         journal_quiesce(j);
 749 }
 750
 751 /* allocate journal on a device: */
 752
 753 static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 754                                          bool new_fs, struct closure *cl)
 755 {
 756         struct bch_fs *c = ca->fs;
 757         struct journal_device *ja = &ca->journal;
 758         u64 *new_bucket_seq = NULL, *new_buckets = NULL;
 759         struct open_bucket **ob = NULL;
 760         long *bu = NULL;
 761         unsigned i, nr_got = 0, nr_want = nr - ja->nr;
 762         unsigned old_nr                 = ja->nr;
 763         unsigned old_discard_idx        = ja->discard_idx;
 764         unsigned old_dirty_idx_ondisk   = ja->dirty_idx_ondisk;
 765         unsigned old_dirty_idx          = ja->dirty_idx;
 766         unsigned old_cur_idx            = ja->cur_idx;
 767         int ret = 0;
 768
 769         if (c) {
 770                 bch2_journal_flush_all_pins(&c->journal);
 771                 bch2_journal_block(&c->journal);
 772                 mutex_lock(&c->sb_lock);
 773         }
 774
 775         bu              = kcalloc(nr_want, sizeof(*bu), GFP_KERNEL);
 776         ob              = kcalloc(nr_want, sizeof(*ob), GFP_KERNEL);
 777         new_buckets     = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 778         new_bucket_seq  = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 779         if (!bu || !ob || !new_buckets || !new_bucket_seq) {
 780                 ret = -ENOMEM;
 781                 goto err_unblock;
 782         }
 783
 784         for (nr_got = 0; nr_got < nr_want; nr_got++) {
 785                 if (new_fs) {
 786                         bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
 787                         if (bu[nr_got] < 0) {
 788                                 ret = -BCH_ERR_ENOSPC_bucket_alloc;
 789                                 break;
 790                         }
 791                 } else {
 792                         ob[nr_got] = bch2_bucket_alloc(c, ca, RESERVE_none,
 793                                                false, cl);
 794                         ret = PTR_ERR_OR_ZERO(ob[nr_got]);
 795                         if (ret)
 796                                 break;
 797
 798                         bu[nr_got] = ob[nr_got]->bucket;
 799                 }
 800         }
 801
 802         if (!nr_got)
 803                 goto err_unblock;
 804
 805         /*
 806          * We may be called from the device add path, before the new device has
 807          * actually been added to the running filesystem:
 808          */
 809         if (!new_fs)
 810                 spin_lock(&c->journal.lock);
 811
 812         memcpy(new_buckets,     ja->buckets,    ja->nr * sizeof(u64));
 813         memcpy(new_bucket_seq,  ja->bucket_seq, ja->nr * sizeof(u64));
 814         swap(new_buckets,       ja->buckets);
 815         swap(new_bucket_seq,    ja->bucket_seq);
 816
 817         for (i = 0; i < nr_got; i++) {
 818                 unsigned pos = ja->discard_idx ?: ja->nr;
 819                 long b = bu[i];
 820
 821                 __array_insert_item(ja->buckets,                ja->nr, pos);
 822                 __array_insert_item(ja->bucket_seq,             ja->nr, pos);
 823                 ja->nr++;
 824
 825                 ja->buckets[pos] = b;
 826                 ja->bucket_seq[pos] = 0;
 827
 828                 if (pos <= ja->discard_idx)
 829                         ja->discard_idx = (ja->discard_idx + 1) % ja->nr;
 830                 if (pos <= ja->dirty_idx_ondisk)
 831                         ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + 1) % ja->nr;
 832                 if (pos <= ja->dirty_idx)
 833                         ja->dirty_idx = (ja->dirty_idx + 1) % ja->nr;
 834                 if (pos <= ja->cur_idx)
 835                         ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
 836         }
 837
 838         ret = bch2_journal_buckets_to_sb(c, ca);
 839         if (ret) {
 840                 /* Revert: */
 841                 swap(new_buckets,       ja->buckets);
 842                 swap(new_bucket_seq,    ja->bucket_seq);
 843                 ja->nr                  = old_nr;
 844                 ja->discard_idx         = old_discard_idx;
 845                 ja->dirty_idx_ondisk    = old_dirty_idx_ondisk;
 846                 ja->dirty_idx           = old_dirty_idx;
 847                 ja->cur_idx             = old_cur_idx;
 848         }
 849
 850         if (!new_fs)
 851                 spin_unlock(&c->journal.lock);
 852
 853         if (ja->nr != old_nr && !new_fs)
 854                 bch2_write_super(c);
 855
 856         if (c)
 857                 bch2_journal_unblock(&c->journal);
 858
 859         if (ret)
 860                 goto err;
 861
 862         if (!new_fs) {
 863                 for (i = 0; i < nr_got; i++) {
 864                         ret = bch2_trans_run(c,
 865                                 bch2_trans_mark_metadata_bucket(&trans, ca,
 866                                                 bu[i], BCH_DATA_journal,
 867                                                 ca->mi.bucket_size));
 868                         if (ret) {
 869                                 bch2_fs_inconsistent(c, "error marking new journal buckets: %i", ret);
 870                                 goto err;
 871                         }
 872                 }
 873         }
 874 err:
 875         if (c)
 876                 mutex_unlock(&c->sb_lock);
 877
 878         if (ob && !new_fs)
 879                 for (i = 0; i < nr_got; i++)
 880                         bch2_open_bucket_put(c, ob[i]);
 881
 882         kfree(new_bucket_seq);
 883         kfree(new_buckets);
 884         kfree(ob);
 885         kfree(bu);
 886
 887         return ret;
 888 err_unblock:
 889         if (c)
 890                 bch2_journal_unblock(&c->journal);
 891         goto err;
 892 }
 893
 894 /*
 895  * Allocate more journal space at runtime - not currently making use if it, but
 896  * the code works:
 897  */
 898 int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 899                                 unsigned nr)
 900 {
 901         struct journal_device *ja = &ca->journal;
 902         struct closure cl;
 903         int ret = 0;
 904
 905         /* don't handle reducing nr of buckets yet: */
 906         if (nr < ja->nr)
 907                 return 0;
 908
 909         closure_init_stack(&cl);
 910
 911         while (ja->nr != nr) {
 912                 struct disk_reservation disk_res = { 0, 0 };
 913
 914                 /*
 915                  * note: journal buckets aren't really counted as _sectors_ used yet, so
 916                  * we don't need the disk reservation to avoid the BUG_ON() in buckets.c
 917                  * when space used goes up without a reservation - but we do need the
 918                  * reservation to ensure we'll actually be able to allocate:
 919                  *
 920                  * XXX: that's not right, disk reservations only ensure a
 921                  * filesystem-wide allocation will succeed, this is a device
 922                  * specific allocation - we can hang here:
 923                  */
 924
 925                 ret = bch2_disk_reservation_get(c, &disk_res,
 926                                                 bucket_to_sector(ca, nr - ja->nr), 1, 0);
 927                 if (ret)
 928                         break;
 929
 930                 ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl);
 931
 932                 bch2_disk_reservation_put(c, &disk_res);
 933
 934                 closure_sync(&cl);
 935
 936                 if (ret && ret != -BCH_ERR_bucket_alloc_blocked)
 937                         break;
 938         }
 939
 940         if (ret)
 941                 bch_err(c, "%s: err %s", __func__, bch2_err_str(ret));
 942
 943         return ret;
 944 }
 945
 946 int bch2_dev_journal_alloc(struct bch_dev *ca)
 947 {
 948         unsigned nr;
 949
 950         if (dynamic_fault("bcachefs:add:journal_alloc"))
 951                 return -ENOMEM;
 952
 953         /* 1/128th of the device by default: */
 954         nr = ca->mi.nbuckets >> 7;
 955
 956         /*
 957          * clamp journal size to 8192 buckets or 8GB (in sectors), whichever
 958          * is smaller:
 959          */
 960         nr = clamp_t(unsigned, nr,
 961                      BCH_JOURNAL_BUCKETS_MIN,
 962                      min(1 << 13,
 963                          (1 << 24) / ca->mi.bucket_size));
 964
 965         return __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
 966 }
 967
 968 /* startup/shutdown: */
 969
 970 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
 971 {
 972         bool ret = false;
 973         u64 seq;
 974
 975         spin_lock(&j->lock);
 976         for (seq = journal_last_unwritten_seq(j);
 977              seq <= journal_cur_seq(j) && !ret;
 978              seq++) {
 979                 struct journal_buf *buf = journal_seq_to_buf(j, seq);
 980
 981                 if (bch2_bkey_has_device(bkey_i_to_s_c(&buf->key), dev_idx))
 982                         ret = true;
 983         }
 984         spin_unlock(&j->lock);
 985
 986         return ret;
 987 }
 988
 989 void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca)
 990 {
 991         wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx));
 992 }
 993
 994 void bch2_fs_journal_stop(struct journal *j)
 995 {
 996         bch2_journal_reclaim_stop(j);
 997         bch2_journal_flush_all_pins(j);
 998
 999         wait_event(j->wait, journal_entry_close(j));
1000
1001         /*
1002          * Always write a new journal entry, to make sure the clock hands are up
1003          * to date (and match the superblock)
1004          */
1005         bch2_journal_meta(j);
1006
1007         journal_quiesce(j);
1008
1009         BUG_ON(!bch2_journal_error(j) &&
1010                test_bit(JOURNAL_REPLAY_DONE, &j->flags) &&
1011                j->last_empty_seq != journal_cur_seq(j));
1012
1013         cancel_delayed_work_sync(&j->write_work);
1014 }
1015
1016 int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
1017 {
1018         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1019         struct journal_entry_pin_list *p;
1020         struct journal_replay *i, **_i;
1021         struct genradix_iter iter;
1022         bool had_entries = false;
1023         unsigned ptr;
1024         u64 last_seq = cur_seq, nr, seq;
1025
1026         genradix_for_each_reverse(&c->journal_entries, iter, _i) {
1027                 i = *_i;
1028
1029                 if (!i || i->ignore)
1030                         continue;
1031
1032                 last_seq = le64_to_cpu(i->j.last_seq);
1033                 break;
1034         }
1035
1036         nr = cur_seq - last_seq;
1037
1038         if (nr + 1 > j->pin.size) {
1039                 free_fifo(&j->pin);
1040                 init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL);
1041                 if (!j->pin.data) {
1042                         bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
1043                         return -ENOMEM;
1044                 }
1045         }
1046
1047         j->replay_journal_seq   = last_seq;
1048         j->replay_journal_seq_end = cur_seq;
1049         j->last_seq_ondisk      = last_seq;
1050         j->flushed_seq_ondisk   = cur_seq - 1;
1051         j->seq_ondisk           = cur_seq - 1;
1052         j->pin.front            = last_seq;
1053         j->pin.back             = cur_seq;
1054         atomic64_set(&j->seq, cur_seq - 1);
1055
1056         fifo_for_each_entry_ptr(p, &j->pin, seq)
1057                 journal_pin_list_init(p, 1);
1058
1059         genradix_for_each(&c->journal_entries, iter, _i) {
1060                 i = *_i;
1061
1062                 if (!i || i->ignore)
1063                         continue;
1064
1065                 seq = le64_to_cpu(i->j.seq);
1066                 BUG_ON(seq >= cur_seq);
1067
1068                 if (seq < last_seq)
1069                         continue;
1070
1071                 if (journal_entry_empty(&i->j))
1072                         j->last_empty_seq = le64_to_cpu(i->j.seq);
1073
1074                 p = journal_seq_pin(j, seq);
1075
1076                 p->devs.nr = 0;
1077                 for (ptr = 0; ptr < i->nr_ptrs; ptr++)
1078                         bch2_dev_list_add_dev(&p->devs, i->ptrs[ptr].dev);
1079
1080                 had_entries = true;
1081         }
1082
1083         if (!had_entries)
1084                 j->last_empty_seq = cur_seq;
1085
1086         spin_lock(&j->lock);
1087
1088         set_bit(JOURNAL_STARTED, &j->flags);
1089         j->last_flush_write = jiffies;
1090
1091         j->reservations.idx = j->reservations.unwritten_idx = journal_cur_seq(j);
1092         j->reservations.unwritten_idx++;
1093
1094         c->last_bucket_seq_cleanup = journal_cur_seq(j);
1095
1096         bch2_journal_space_available(j);
1097         spin_unlock(&j->lock);
1098
1099         return bch2_journal_reclaim_start(j);
1100 }
1101
1102 /* init/exit: */
1103
1104 void bch2_dev_journal_exit(struct bch_dev *ca)
1105 {
1106         kfree(ca->journal.bio);
1107         kfree(ca->journal.buckets);
1108         kfree(ca->journal.bucket_seq);
1109
1110         ca->journal.bio         = NULL;
1111         ca->journal.buckets     = NULL;
1112         ca->journal.bucket_seq  = NULL;
1113 }
1114
1115 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
1116 {
1117         struct journal_device *ja = &ca->journal;
1118         struct bch_sb_field_journal *journal_buckets =
1119                 bch2_sb_get_journal(sb);
1120         struct bch_sb_field_journal_v2 *journal_buckets_v2 =
1121                 bch2_sb_get_journal_v2(sb);
1122         unsigned i, nr_bvecs;
1123
1124         ja->nr = 0;
1125
1126         if (journal_buckets_v2) {
1127                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1128
1129                 for (i = 0; i < nr; i++)
1130                         ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr);
1131         } else if (journal_buckets) {
1132                 ja->nr = bch2_nr_journal_buckets(journal_buckets);
1133         }
1134
1135         ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1136         if (!ja->bucket_seq)
1137                 return -ENOMEM;
1138
1139         nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
1140
1141         ca->journal.bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1142         if (!ca->journal.bio)
1143                 return -ENOMEM;
1144
1145         bio_init(ca->journal.bio, NULL, ca->journal.bio->bi_inline_vecs, nr_bvecs, 0);
1146
1147         ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1148         if (!ja->buckets)
1149                 return -ENOMEM;
1150
1151         if (journal_buckets_v2) {
1152                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1153                 unsigned j, dst = 0;
1154
1155                 for (i = 0; i < nr; i++)
1156                         for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
1157                                 ja->buckets[dst++] =
1158                                         le64_to_cpu(journal_buckets_v2->d[i].start) + j;
1159         } else if (journal_buckets) {
1160                 for (i = 0; i < ja->nr; i++)
1161                         ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
1162         }
1163
1164         return 0;
1165 }
1166
1167 void bch2_fs_journal_exit(struct journal *j)
1168 {
1169         unsigned i;
1170
1171         darray_exit(&j->early_journal_entries);
1172
1173         for (i = 0; i < ARRAY_SIZE(j->buf); i++)
1174                 kvpfree(j->buf[i].data, j->buf[i].buf_size);
1175         free_fifo(&j->pin);
1176 }
1177
1178 int bch2_fs_journal_init(struct journal *j)
1179 {
1180         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1181         static struct lock_class_key res_key;
1182         unsigned i;
1183         int ret = 0;
1184
1185         pr_verbose_init(c->opts, "");
1186
1187         spin_lock_init(&j->lock);
1188         spin_lock_init(&j->err_lock);
1189         init_waitqueue_head(&j->wait);
1190         INIT_DELAYED_WORK(&j->write_work, journal_write_work);
1191         init_waitqueue_head(&j->reclaim_wait);
1192         init_waitqueue_head(&j->pin_flush_wait);
1193         mutex_init(&j->reclaim_lock);
1194         mutex_init(&j->discard_lock);
1195
1196         lockdep_init_map(&j->res_map, "journal res", &res_key, 0);
1197
1198         atomic64_set(&j->reservations.counter,
1199                 ((union journal_res_state)
1200                  { .cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL }).v);
1201
1202         if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL))) {
1203                 ret = -ENOMEM;
1204                 goto out;
1205         }
1206
1207         for (i = 0; i < ARRAY_SIZE(j->buf); i++) {
1208                 j->buf[i].buf_size = JOURNAL_ENTRY_SIZE_MIN;
1209                 j->buf[i].data = kvpmalloc(j->buf[i].buf_size, GFP_KERNEL);
1210                 if (!j->buf[i].data) {
1211                         ret = -ENOMEM;
1212                         goto out;
1213                 }
1214         }
1215
1216         j->pin.front = j->pin.back = 1;
1217 out:
1218         pr_verbose_init(c->opts, "ret %i", ret);
1219         return ret;
1220 }
1221
1222 /* debug: */
1223
1224 void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1225 {
1226         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1227         union journal_res_state s;
1228         struct bch_dev *ca;
1229         unsigned long now = jiffies;
1230         u64 seq;
1231         unsigned i;
1232
1233         if (!out->nr_tabstops)
1234                 printbuf_tabstop_push(out, 24);
1235         out->atomic++;
1236
1237         rcu_read_lock();
1238         s = READ_ONCE(j->reservations);
1239
1240         prt_printf(out, "dirty journal entries:\t%llu/%llu\n",  fifo_used(&j->pin), j->pin.size);
1241         prt_printf(out, "seq:\t\t\t%llu\n",                     journal_cur_seq(j));
1242         prt_printf(out, "seq_ondisk:\t\t%llu\n",                j->seq_ondisk);
1243         prt_printf(out, "last_seq:\t\t%llu\n",          journal_last_seq(j));
1244         prt_printf(out, "last_seq_ondisk:\t%llu\n",             j->last_seq_ondisk);
1245         prt_printf(out, "flushed_seq_ondisk:\t%llu\n",  j->flushed_seq_ondisk);
1246         prt_printf(out, "prereserved:\t\t%u/%u\n",              j->prereserved.reserved, j->prereserved.remaining);
1247         prt_printf(out, "watermark:\t\t%s\n",           bch2_journal_watermarks[j->watermark]);
1248         prt_printf(out, "each entry reserved:\t%u\n",   j->entry_u64s_reserved);
1249         prt_printf(out, "nr flush writes:\t%llu\n",             j->nr_flush_writes);
1250         prt_printf(out, "nr noflush writes:\t%llu\n",   j->nr_noflush_writes);
1251         prt_printf(out, "nr direct reclaim:\t%llu\n",   j->nr_direct_reclaim);
1252         prt_printf(out, "nr background reclaim:\t%llu\n",       j->nr_background_reclaim);
1253         prt_printf(out, "reclaim kicked:\t\t%u\n",              j->reclaim_kicked);
1254         prt_printf(out, "reclaim runs in:\t%u ms\n",    time_after(j->next_reclaim, now)
1255                ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0);
1256         prt_printf(out, "current entry sectors:\t%u\n", j->cur_entry_sectors);
1257         prt_printf(out, "current entry error:\t%s\n",   bch2_journal_errors[j->cur_entry_error]);
1258         prt_printf(out, "current entry:\t\t");
1259
1260         switch (s.cur_entry_offset) {
1261         case JOURNAL_ENTRY_ERROR_VAL:
1262                 prt_printf(out, "error");
1263                 break;
1264         case JOURNAL_ENTRY_CLOSED_VAL:
1265                 prt_printf(out, "closed");
1266                 break;
1267         default:
1268                 prt_printf(out, "%u/%u", s.cur_entry_offset, j->cur_entry_u64s);
1269                 break;
1270         }
1271
1272         prt_newline(out);
1273
1274         for (seq = journal_cur_seq(j);
1275              seq >= journal_last_unwritten_seq(j);
1276              --seq) {
1277                 i = seq & JOURNAL_BUF_MASK;
1278
1279                 prt_printf(out, "unwritten entry:");
1280                 prt_tab(out);
1281                 prt_printf(out, "%llu", seq);
1282                 prt_newline(out);
1283                 printbuf_indent_add(out, 2);
1284
1285                 prt_printf(out, "refcount:");
1286                 prt_tab(out);
1287                 prt_printf(out, "%u", journal_state_count(s, i));
1288                 prt_newline(out);
1289
1290                 prt_printf(out, "sectors:");
1291                 prt_tab(out);
1292                 prt_printf(out, "%u", j->buf[i].sectors);
1293                 prt_newline(out);
1294
1295                 prt_printf(out, "expires");
1296                 prt_tab(out);
1297                 prt_printf(out, "%li jiffies", j->buf[i].expires - jiffies);
1298                 prt_newline(out);
1299
1300                 printbuf_indent_sub(out, 2);
1301         }
1302
1303         prt_printf(out,
1304                "replay done:\t\t%i\n",
1305                test_bit(JOURNAL_REPLAY_DONE,    &j->flags));
1306
1307         prt_printf(out, "space:\n");
1308         prt_printf(out, "\tdiscarded\t%u:%u\n",
1309                j->space[journal_space_discarded].next_entry,
1310                j->space[journal_space_discarded].total);
1311         prt_printf(out, "\tclean ondisk\t%u:%u\n",
1312                j->space[journal_space_clean_ondisk].next_entry,
1313                j->space[journal_space_clean_ondisk].total);
1314         prt_printf(out, "\tclean\t\t%u:%u\n",
1315                j->space[journal_space_clean].next_entry,
1316                j->space[journal_space_clean].total);
1317         prt_printf(out, "\ttotal\t\t%u:%u\n",
1318                j->space[journal_space_total].next_entry,
1319                j->space[journal_space_total].total);
1320
1321         for_each_member_device_rcu(ca, c, i,
1322                                    &c->rw_devs[BCH_DATA_journal]) {
1323                 struct journal_device *ja = &ca->journal;
1324
1325                 if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d))
1326                         continue;
1327
1328                 if (!ja->nr)
1329                         continue;
1330
1331                 prt_printf(out, "dev %u:\n",            i);
1332                 prt_printf(out, "\tnr\t\t%u\n",         ja->nr);
1333                 prt_printf(out, "\tbucket size\t%u\n",  ca->mi.bucket_size);
1334                 prt_printf(out, "\tavailable\t%u:%u\n", bch2_journal_dev_buckets_available(j, ja, journal_space_discarded), ja->sectors_free);
1335                 prt_printf(out, "\tdiscard_idx\t%u\n",  ja->discard_idx);
1336                 prt_printf(out, "\tdirty_ondisk\t%u (seq %llu)\n", ja->dirty_idx_ondisk,        ja->bucket_seq[ja->dirty_idx_ondisk]);
1337                 prt_printf(out, "\tdirty_idx\t%u (seq %llu)\n", ja->dirty_idx,          ja->bucket_seq[ja->dirty_idx]);
1338                 prt_printf(out, "\tcur_idx\t\t%u (seq %llu)\n", ja->cur_idx,            ja->bucket_seq[ja->cur_idx]);
1339         }
1340
1341         rcu_read_unlock();
1342
1343         --out->atomic;
1344 }
1345
1346 void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1347 {
1348         spin_lock(&j->lock);
1349         __bch2_journal_debug_to_text(out, j);
1350         spin_unlock(&j->lock);
1351 }
1352
1353 bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq)
1354 {
1355         struct journal_entry_pin_list *pin_list;
1356         struct journal_entry_pin *pin;
1357
1358         spin_lock(&j->lock);
1359         *seq = max(*seq, j->pin.front);
1360
1361         if (*seq >= j->pin.back) {
1362                 spin_unlock(&j->lock);
1363                 return true;
1364         }
1365
1366         out->atomic++;
1367
1368         pin_list = journal_seq_pin(j, *seq);
1369
1370         prt_printf(out, "%llu: count %u", *seq, atomic_read(&pin_list->count));
1371         prt_newline(out);
1372         printbuf_indent_add(out, 2);
1373
1374         list_for_each_entry(pin, &pin_list->list, list) {
1375                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1376                 prt_newline(out);
1377         }
1378
1379         list_for_each_entry(pin, &pin_list->key_cache_list, list) {
1380                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1381                 prt_newline(out);
1382         }
1383
1384         if (!list_empty(&pin_list->flushed)) {
1385                 prt_printf(out, "flushed:");
1386                 prt_newline(out);
1387         }
1388
1389         list_for_each_entry(pin, &pin_list->flushed, list) {
1390                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1391                 prt_newline(out);
1392         }
1393
1394         printbuf_indent_sub(out, 2);
1395
1396         --out->atomic;
1397         spin_unlock(&j->lock);
1398
1399         return false;
1400 }
1401
1402 void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
1403 {
1404         u64 seq = 0;
1405
1406         while (!bch2_journal_seq_pins_to_text(out, j, &seq))
1407                 seq++;
1408 }