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) - 1)
 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 <= 0)
 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         /*
 273          * Must be set before marking the journal entry as open:
 274          */
 275         j->cur_entry_u64s = u64s;
 276
 277         v = atomic64_read(&j->reservations.counter);
 278         do {
 279                 old.v = new.v = v;
 280
 281                 BUG_ON(old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL);
 282
 283                 new.idx++;
 284                 BUG_ON(journal_state_count(new, new.idx));
 285                 BUG_ON(new.idx != (journal_cur_seq(j) & JOURNAL_BUF_MASK));
 286
 287                 journal_state_inc(&new);
 288                 new.cur_entry_offset = 0;
 289         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 290                                        old.v, new.v)) != old.v);
 291
 292         if (j->res_get_blocked_start)
 293                 bch2_time_stats_update(j->blocked_time,
 294                                        j->res_get_blocked_start);
 295         j->res_get_blocked_start = 0;
 296
 297         mod_delayed_work(c->io_complete_wq,
 298                          &j->write_work,
 299                          msecs_to_jiffies(c->opts.journal_flush_delay));
 300         journal_wake(j);
 301         return 0;
 302 }
 303
 304 static bool journal_quiesced(struct journal *j)
 305 {
 306         bool ret = atomic64_read(&j->seq) == j->seq_ondisk;
 307
 308         if (!ret)
 309                 journal_entry_close(j);
 310         return ret;
 311 }
 312
 313 static void journal_quiesce(struct journal *j)
 314 {
 315         wait_event(j->wait, journal_quiesced(j));
 316 }
 317
 318 static void journal_write_work(struct work_struct *work)
 319 {
 320         struct journal *j = container_of(work, struct journal, write_work.work);
 321         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 322         long delta;
 323
 324         spin_lock(&j->lock);
 325         if (!__journal_entry_is_open(j->reservations))
 326                 goto unlock;
 327
 328         delta = journal_cur_buf(j)->expires - jiffies;
 329
 330         if (delta > 0)
 331                 mod_delayed_work(c->io_complete_wq, &j->write_work, delta);
 332         else
 333                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 334 unlock:
 335         spin_unlock(&j->lock);
 336 }
 337
 338 static int __journal_res_get(struct journal *j, struct journal_res *res,
 339                              unsigned flags)
 340 {
 341         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 342         struct journal_buf *buf;
 343         bool can_discard;
 344         int ret;
 345 retry:
 346         if (journal_res_get_fast(j, res, flags))
 347                 return 0;
 348
 349         if (bch2_journal_error(j))
 350                 return -BCH_ERR_erofs_journal_err;
 351
 352         spin_lock(&j->lock);
 353
 354         /*
 355          * Recheck after taking the lock, so we don't race with another thread
 356          * that just did journal_entry_open() and call journal_entry_close()
 357          * unnecessarily
 358          */
 359         if (journal_res_get_fast(j, res, flags)) {
 360                 spin_unlock(&j->lock);
 361                 return 0;
 362         }
 363
 364         if ((flags & JOURNAL_WATERMARK_MASK) < j->watermark) {
 365                 /*
 366                  * Don't want to close current journal entry, just need to
 367                  * invoke reclaim:
 368                  */
 369                 ret = JOURNAL_ERR_journal_full;
 370                 goto unlock;
 371         }
 372
 373         /*
 374          * If we couldn't get a reservation because the current buf filled up,
 375          * and we had room for a bigger entry on disk, signal that we want to
 376          * realloc the journal bufs:
 377          */
 378         buf = journal_cur_buf(j);
 379         if (journal_entry_is_open(j) &&
 380             buf->buf_size >> 9 < buf->disk_sectors &&
 381             buf->buf_size < JOURNAL_ENTRY_SIZE_MAX)
 382                 j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1);
 383
 384         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 385         ret = journal_entry_open(j);
 386
 387         if (ret == JOURNAL_ERR_max_in_flight)
 388                 trace_and_count(c, journal_entry_full, c);
 389 unlock:
 390         if ((ret && ret != JOURNAL_ERR_insufficient_devices) &&
 391             !j->res_get_blocked_start) {
 392                 j->res_get_blocked_start = local_clock() ?: 1;
 393                 trace_and_count(c, journal_full, c);
 394         }
 395
 396         can_discard = j->can_discard;
 397         spin_unlock(&j->lock);
 398
 399         if (!ret)
 400                 goto retry;
 401
 402         if ((ret == JOURNAL_ERR_journal_full ||
 403              ret == JOURNAL_ERR_journal_pin_full) &&
 404             !can_discard &&
 405             !nr_unwritten_journal_entries(j) &&
 406             (flags & JOURNAL_WATERMARK_MASK) == JOURNAL_WATERMARK_reserved) {
 407                 struct printbuf buf = PRINTBUF;
 408
 409                 bch_err(c, "Journal stuck! Hava a pre-reservation but journal full (ret %s)",
 410                         bch2_journal_errors[ret]);
 411
 412                 bch2_journal_debug_to_text(&buf, j);
 413                 bch_err(c, "%s", buf.buf);
 414
 415                 printbuf_reset(&buf);
 416                 bch2_journal_pins_to_text(&buf, j);
 417                 bch_err(c, "Journal pins:\n%s", buf.buf);
 418
 419                 printbuf_exit(&buf);
 420                 bch2_fatal_error(c);
 421                 dump_stack();
 422         }
 423
 424         /*
 425          * Journal is full - can't rely on reclaim from work item due to
 426          * freezing:
 427          */
 428         if ((ret == JOURNAL_ERR_journal_full ||
 429              ret == JOURNAL_ERR_journal_pin_full) &&
 430             !(flags & JOURNAL_RES_GET_NONBLOCK)) {
 431                 if (can_discard) {
 432                         bch2_journal_do_discards(j);
 433                         goto retry;
 434                 }
 435
 436                 if (mutex_trylock(&j->reclaim_lock)) {
 437                         bch2_journal_reclaim(j);
 438                         mutex_unlock(&j->reclaim_lock);
 439                 }
 440         }
 441
 442         return ret == JOURNAL_ERR_insufficient_devices
 443                 ? -EROFS
 444                 : -BCH_ERR_journal_res_get_blocked;
 445 }
 446
 447 /*
 448  * Essentially the entry function to the journaling code. When bcachefs is doing
 449  * a btree insert, it calls this function to get the current journal write.
 450  * Journal write is the structure used set up journal writes. The calling
 451  * function will then add its keys to the structure, queuing them for the next
 452  * write.
 453  *
 454  * To ensure forward progress, the current task must not be holding any
 455  * btree node write locks.
 456  */
 457 int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
 458                                   unsigned flags)
 459 {
 460         int ret;
 461
 462         closure_wait_event(&j->async_wait,
 463                    (ret = __journal_res_get(j, res, flags)) !=
 464                    -BCH_ERR_journal_res_get_blocked||
 465                    (flags & JOURNAL_RES_GET_NONBLOCK));
 466         return ret;
 467 }
 468
 469 /* journal_preres: */
 470
 471 static bool journal_preres_available(struct journal *j,
 472                                      struct journal_preres *res,
 473                                      unsigned new_u64s,
 474                                      unsigned flags)
 475 {
 476         bool ret = bch2_journal_preres_get_fast(j, res, new_u64s, flags, true);
 477
 478         if (!ret && mutex_trylock(&j->reclaim_lock)) {
 479                 bch2_journal_reclaim(j);
 480                 mutex_unlock(&j->reclaim_lock);
 481         }
 482
 483         return ret;
 484 }
 485
 486 int __bch2_journal_preres_get(struct journal *j,
 487                               struct journal_preres *res,
 488                               unsigned new_u64s,
 489                               unsigned flags)
 490 {
 491         int ret;
 492
 493         closure_wait_event(&j->preres_wait,
 494                    (ret = bch2_journal_error(j)) ||
 495                    journal_preres_available(j, res, new_u64s, flags));
 496         return ret;
 497 }
 498
 499 /* journal_entry_res: */
 500
 501 void bch2_journal_entry_res_resize(struct journal *j,
 502                                    struct journal_entry_res *res,
 503                                    unsigned new_u64s)
 504 {
 505         union journal_res_state state;
 506         int d = new_u64s - res->u64s;
 507
 508         spin_lock(&j->lock);
 509
 510         j->entry_u64s_reserved += d;
 511         if (d <= 0)
 512                 goto out;
 513
 514         j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d);
 515         smp_mb();
 516         state = READ_ONCE(j->reservations);
 517
 518         if (state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL &&
 519             state.cur_entry_offset > j->cur_entry_u64s) {
 520                 j->cur_entry_u64s += d;
 521                 /*
 522                  * Not enough room in current journal entry, have to flush it:
 523                  */
 524                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 525         } else {
 526                 journal_cur_buf(j)->u64s_reserved += d;
 527         }
 528 out:
 529         spin_unlock(&j->lock);
 530         res->u64s += d;
 531 }
 532
 533 /* journal flushing: */
 534
 535 /**
 536  * bch2_journal_flush_seq_async - wait for a journal entry to be written
 537  *
 538  * like bch2_journal_wait_on_seq, except that it triggers a write immediately if
 539  * necessary
 540  */
 541 int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 542                                  struct closure *parent)
 543 {
 544         struct journal_buf *buf;
 545         int ret = 0;
 546
 547         if (seq <= j->flushed_seq_ondisk)
 548                 return 1;
 549
 550         spin_lock(&j->lock);
 551
 552         if (WARN_ONCE(seq > journal_cur_seq(j),
 553                       "requested to flush journal seq %llu, but currently at %llu",
 554                       seq, journal_cur_seq(j)))
 555                 goto out;
 556
 557         /* Recheck under lock: */
 558         if (j->err_seq && seq >= j->err_seq) {
 559                 ret = -EIO;
 560                 goto out;
 561         }
 562
 563         if (seq <= j->flushed_seq_ondisk) {
 564                 ret = 1;
 565                 goto out;
 566         }
 567
 568         /* if seq was written, but not flushed - flush a newer one instead */
 569         seq = max(seq, journal_last_unwritten_seq(j));
 570
 571 recheck_need_open:
 572         if (seq > journal_cur_seq(j)) {
 573                 struct journal_res res = { 0 };
 574
 575                 if (journal_entry_is_open(j))
 576                         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 577
 578                 spin_unlock(&j->lock);
 579
 580                 ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 581                 if (ret)
 582                         return ret;
 583
 584                 seq = res.seq;
 585                 buf = j->buf + (seq & JOURNAL_BUF_MASK);
 586                 buf->must_flush = true;
 587
 588                 if (!buf->flush_time) {
 589                         buf->flush_time = local_clock() ?: 1;
 590                         buf->expires = jiffies;
 591                 }
 592
 593                 if (parent && !closure_wait(&buf->wait, parent))
 594                         BUG();
 595
 596                 bch2_journal_res_put(j, &res);
 597
 598                 spin_lock(&j->lock);
 599                 goto want_write;
 600         }
 601
 602         /*
 603          * if write was kicked off without a flush, flush the next sequence
 604          * number instead
 605          */
 606         buf = journal_seq_to_buf(j, seq);
 607         if (buf->noflush) {
 608                 seq++;
 609                 goto recheck_need_open;
 610         }
 611
 612         buf->must_flush = true;
 613
 614         if (parent && !closure_wait(&buf->wait, parent))
 615                 BUG();
 616 want_write:
 617         if (seq == journal_cur_seq(j))
 618                 journal_entry_want_write(j);
 619 out:
 620         spin_unlock(&j->lock);
 621         return ret;
 622 }
 623
 624 int bch2_journal_flush_seq(struct journal *j, u64 seq)
 625 {
 626         u64 start_time = local_clock();
 627         int ret, ret2;
 628
 629         /*
 630          * Don't update time_stats when @seq is already flushed:
 631          */
 632         if (seq <= j->flushed_seq_ondisk)
 633                 return 0;
 634
 635         ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)));
 636
 637         if (!ret)
 638                 bch2_time_stats_update(j->flush_seq_time, start_time);
 639
 640         return ret ?: ret2 < 0 ? ret2 : 0;
 641 }
 642
 643 /*
 644  * bch2_journal_flush_async - if there is an open journal entry, or a journal
 645  * still being written, write it and wait for the write to complete
 646  */
 647 void bch2_journal_flush_async(struct journal *j, struct closure *parent)
 648 {
 649         bch2_journal_flush_seq_async(j, atomic64_read(&j->seq), parent);
 650 }
 651
 652 int bch2_journal_flush(struct journal *j)
 653 {
 654         return bch2_journal_flush_seq(j, atomic64_read(&j->seq));
 655 }
 656
 657 /*
 658  * bch2_journal_noflush_seq - tell the journal not to issue any flushes before
 659  * @seq
 660  */
 661 bool bch2_journal_noflush_seq(struct journal *j, u64 seq)
 662 {
 663         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 664         u64 unwritten_seq;
 665         bool ret = false;
 666
 667         if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush)))
 668                 return false;
 669
 670         if (seq <= c->journal.flushed_seq_ondisk)
 671                 return false;
 672
 673         spin_lock(&j->lock);
 674         if (seq <= c->journal.flushed_seq_ondisk)
 675                 goto out;
 676
 677         for (unwritten_seq = journal_last_unwritten_seq(j);
 678              unwritten_seq < seq;
 679              unwritten_seq++) {
 680                 struct journal_buf *buf = journal_seq_to_buf(j, unwritten_seq);
 681
 682                 /* journal write is already in flight, and was a flush write: */
 683                 if (unwritten_seq == journal_last_unwritten_seq(j) && !buf->noflush)
 684                         goto out;
 685
 686                 buf->noflush = true;
 687         }
 688
 689         ret = true;
 690 out:
 691         spin_unlock(&j->lock);
 692         return ret;
 693 }
 694
 695 int bch2_journal_meta(struct journal *j)
 696 {
 697         struct journal_buf *buf;
 698         struct journal_res res;
 699         int ret;
 700
 701         memset(&res, 0, sizeof(res));
 702
 703         ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 704         if (ret)
 705                 return ret;
 706
 707         buf = j->buf + (res.seq & JOURNAL_BUF_MASK);
 708         buf->must_flush = true;
 709
 710         if (!buf->flush_time) {
 711                 buf->flush_time = local_clock() ?: 1;
 712                 buf->expires = jiffies;
 713         }
 714
 715         bch2_journal_res_put(j, &res);
 716
 717         return bch2_journal_flush_seq(j, res.seq);
 718 }
 719
 720 int bch2_journal_log_msg(struct journal *j, const char *fmt, ...)
 721 {
 722         struct jset_entry_log *entry;
 723         struct journal_res res = { 0 };
 724         unsigned msglen, u64s;
 725         va_list args;
 726         int ret;
 727
 728         va_start(args, fmt);
 729         msglen = vsnprintf(NULL, 0, fmt, args) + 1;
 730         va_end(args);
 731
 732         u64s = jset_u64s(DIV_ROUND_UP(msglen, sizeof(u64)));
 733
 734         ret = bch2_journal_res_get(j, &res, u64s, 0);
 735         if (ret)
 736                 return ret;
 737
 738         entry = container_of(journal_res_entry(j, &res),
 739                              struct jset_entry_log, entry);
 740         memset(entry, 0, u64s * sizeof(u64));
 741         entry->entry.type = BCH_JSET_ENTRY_log;
 742         entry->entry.u64s = u64s - 1;
 743
 744         va_start(args, fmt);
 745         vsnprintf(entry->d, INT_MAX, fmt, args);
 746         va_end(args);
 747
 748         bch2_journal_res_put(j, &res);
 749
 750         return bch2_journal_flush_seq(j, res.seq);
 751 }
 752
 753 /* block/unlock the journal: */
 754
 755 void bch2_journal_unblock(struct journal *j)
 756 {
 757         spin_lock(&j->lock);
 758         j->blocked--;
 759         spin_unlock(&j->lock);
 760
 761         journal_wake(j);
 762 }
 763
 764 void bch2_journal_block(struct journal *j)
 765 {
 766         spin_lock(&j->lock);
 767         j->blocked++;
 768         spin_unlock(&j->lock);
 769
 770         journal_quiesce(j);
 771 }
 772
 773 /* allocate journal on a device: */
 774
 775 static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 776                                          bool new_fs, struct closure *cl)
 777 {
 778         struct bch_fs *c = ca->fs;
 779         struct journal_device *ja = &ca->journal;
 780         u64 *new_bucket_seq = NULL, *new_buckets = NULL;
 781         struct open_bucket **ob = NULL;
 782         long *bu = NULL;
 783         unsigned i, nr_got = 0, nr_want = nr - ja->nr;
 784         unsigned old_nr                 = ja->nr;
 785         unsigned old_discard_idx        = ja->discard_idx;
 786         unsigned old_dirty_idx_ondisk   = ja->dirty_idx_ondisk;
 787         unsigned old_dirty_idx          = ja->dirty_idx;
 788         unsigned old_cur_idx            = ja->cur_idx;
 789         int ret = 0;
 790
 791         if (c) {
 792                 bch2_journal_flush_all_pins(&c->journal);
 793                 bch2_journal_block(&c->journal);
 794         }
 795
 796         bu              = kcalloc(nr_want, sizeof(*bu), GFP_KERNEL);
 797         ob              = kcalloc(nr_want, sizeof(*ob), GFP_KERNEL);
 798         new_buckets     = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 799         new_bucket_seq  = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 800         if (!bu || !ob || !new_buckets || !new_bucket_seq) {
 801                 ret = -ENOMEM;
 802                 goto err_unblock;
 803         }
 804
 805         for (nr_got = 0; nr_got < nr_want; nr_got++) {
 806                 if (new_fs) {
 807                         bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
 808                         if (bu[nr_got] < 0) {
 809                                 ret = -BCH_ERR_ENOSPC_bucket_alloc;
 810                                 break;
 811                         }
 812                 } else {
 813                         ob[nr_got] = bch2_bucket_alloc(c, ca, RESERVE_none,
 814                                                false, cl);
 815                         ret = PTR_ERR_OR_ZERO(ob[nr_got]);
 816                         if (ret)
 817                                 break;
 818
 819                         bu[nr_got] = ob[nr_got]->bucket;
 820                 }
 821         }
 822
 823         if (!nr_got)
 824                 goto err_unblock;
 825
 826         /*
 827          * We may be called from the device add path, before the new device has
 828          * actually been added to the running filesystem:
 829          */
 830         if (!new_fs)
 831                 spin_lock(&c->journal.lock);
 832
 833         memcpy(new_buckets,     ja->buckets,    ja->nr * sizeof(u64));
 834         memcpy(new_bucket_seq,  ja->bucket_seq, ja->nr * sizeof(u64));
 835         swap(new_buckets,       ja->buckets);
 836         swap(new_bucket_seq,    ja->bucket_seq);
 837
 838         for (i = 0; i < nr_got; i++) {
 839                 unsigned pos = ja->discard_idx ?: ja->nr;
 840                 long b = bu[i];
 841
 842                 __array_insert_item(ja->buckets,                ja->nr, pos);
 843                 __array_insert_item(ja->bucket_seq,             ja->nr, pos);
 844                 ja->nr++;
 845
 846                 ja->buckets[pos] = b;
 847                 ja->bucket_seq[pos] = 0;
 848
 849                 if (pos <= ja->discard_idx)
 850                         ja->discard_idx = (ja->discard_idx + 1) % ja->nr;
 851                 if (pos <= ja->dirty_idx_ondisk)
 852                         ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + 1) % ja->nr;
 853                 if (pos <= ja->dirty_idx)
 854                         ja->dirty_idx = (ja->dirty_idx + 1) % ja->nr;
 855                 if (pos <= ja->cur_idx)
 856                         ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
 857         }
 858
 859         ret = bch2_journal_buckets_to_sb(c, ca);
 860         if (ret) {
 861                 /* Revert: */
 862                 swap(new_buckets,       ja->buckets);
 863                 swap(new_bucket_seq,    ja->bucket_seq);
 864                 ja->nr                  = old_nr;
 865                 ja->discard_idx         = old_discard_idx;
 866                 ja->dirty_idx_ondisk    = old_dirty_idx_ondisk;
 867                 ja->dirty_idx           = old_dirty_idx;
 868                 ja->cur_idx             = old_cur_idx;
 869         }
 870
 871         if (!new_fs)
 872                 spin_unlock(&c->journal.lock);
 873
 874         if (c)
 875                 bch2_journal_unblock(&c->journal);
 876
 877         if (ret)
 878                 goto err;
 879
 880         if (!new_fs) {
 881                 for (i = 0; i < nr_got; i++) {
 882                         ret = bch2_trans_run(c,
 883                                 bch2_trans_mark_metadata_bucket(&trans, ca,
 884                                                 bu[i], BCH_DATA_journal,
 885                                                 ca->mi.bucket_size));
 886                         if (ret) {
 887                                 bch2_fs_inconsistent(c, "error marking new journal buckets: %i", ret);
 888                                 goto err;
 889                         }
 890                 }
 891         }
 892 err:
 893         if (ob && !new_fs)
 894                 for (i = 0; i < nr_got; i++)
 895                         bch2_open_bucket_put(c, ob[i]);
 896
 897         kfree(new_bucket_seq);
 898         kfree(new_buckets);
 899         kfree(ob);
 900         kfree(bu);
 901
 902         return ret;
 903 err_unblock:
 904         if (c)
 905                 bch2_journal_unblock(&c->journal);
 906         goto err;
 907 }
 908
 909 /*
 910  * Allocate more journal space at runtime - not currently making use if it, but
 911  * the code works:
 912  */
 913 int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 914                                 unsigned nr)
 915 {
 916         struct journal_device *ja = &ca->journal;
 917         struct closure cl;
 918         unsigned current_nr;
 919         int ret = 0;
 920
 921         /* don't handle reducing nr of buckets yet: */
 922         if (nr < ja->nr)
 923                 return 0;
 924
 925         closure_init_stack(&cl);
 926
 927         while (ja->nr != nr && (ret == 0 || ret == -BCH_ERR_bucket_alloc_blocked)) {
 928                 struct disk_reservation disk_res = { 0, 0 };
 929
 930                 closure_sync(&cl);
 931
 932                 mutex_lock(&c->sb_lock);
 933                 current_nr = ja->nr;
 934
 935                 /*
 936                  * note: journal buckets aren't really counted as _sectors_ used yet, so
 937                  * we don't need the disk reservation to avoid the BUG_ON() in buckets.c
 938                  * when space used goes up without a reservation - but we do need the
 939                  * reservation to ensure we'll actually be able to allocate:
 940                  */
 941
 942                 ret = bch2_disk_reservation_get(c, &disk_res,
 943                                                 bucket_to_sector(ca, nr - ja->nr), 1, 0);
 944                 if (ret) {
 945                         mutex_unlock(&c->sb_lock);
 946                         return ret;
 947                 }
 948
 949                 ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl);
 950
 951                 bch2_disk_reservation_put(c, &disk_res);
 952
 953                 if (ja->nr != current_nr)
 954                         bch2_write_super(c);
 955                 mutex_unlock(&c->sb_lock);
 956         }
 957
 958         return ret;
 959 }
 960
 961 int bch2_dev_journal_alloc(struct bch_dev *ca)
 962 {
 963         unsigned nr;
 964         int ret;
 965
 966         if (dynamic_fault("bcachefs:add:journal_alloc"))
 967                 return -ENOMEM;
 968
 969         /* 1/128th of the device by default: */
 970         nr = ca->mi.nbuckets >> 7;
 971
 972         /*
 973          * clamp journal size to 8192 buckets or 8GB (in sectors), whichever
 974          * is smaller:
 975          */
 976         nr = clamp_t(unsigned, nr,
 977                      BCH_JOURNAL_BUCKETS_MIN,
 978                      min(1 << 13,
 979                          (1 << 24) / ca->mi.bucket_size));
 980
 981         if (ca->fs)
 982                 mutex_lock(&ca->fs->sb_lock);
 983
 984         ret = __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
 985
 986         if (ca->fs)
 987                 mutex_unlock(&ca->fs->sb_lock);
 988
 989         return ret;
 990 }
 991
 992 /* startup/shutdown: */
 993
 994 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
 995 {
 996         bool ret = false;
 997         u64 seq;
 998
 999         spin_lock(&j->lock);
1000         for (seq = journal_last_unwritten_seq(j);
1001              seq <= journal_cur_seq(j) && !ret;
1002              seq++) {
1003                 struct journal_buf *buf = journal_seq_to_buf(j, seq);
1004
1005                 if (bch2_bkey_has_device(bkey_i_to_s_c(&buf->key), dev_idx))
1006                         ret = true;
1007         }
1008         spin_unlock(&j->lock);
1009
1010         return ret;
1011 }
1012
1013 void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca)
1014 {
1015         wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx));
1016 }
1017
1018 void bch2_fs_journal_stop(struct journal *j)
1019 {
1020         bch2_journal_reclaim_stop(j);
1021         bch2_journal_flush_all_pins(j);
1022
1023         wait_event(j->wait, journal_entry_close(j));
1024
1025         /*
1026          * Always write a new journal entry, to make sure the clock hands are up
1027          * to date (and match the superblock)
1028          */
1029         bch2_journal_meta(j);
1030
1031         journal_quiesce(j);
1032
1033         BUG_ON(!bch2_journal_error(j) &&
1034                test_bit(JOURNAL_REPLAY_DONE, &j->flags) &&
1035                j->last_empty_seq != journal_cur_seq(j));
1036
1037         cancel_delayed_work_sync(&j->write_work);
1038 }
1039
1040 int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
1041 {
1042         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1043         struct journal_entry_pin_list *p;
1044         struct journal_replay *i, **_i;
1045         struct genradix_iter iter;
1046         bool had_entries = false;
1047         unsigned ptr;
1048         u64 last_seq = cur_seq, nr, seq;
1049
1050         genradix_for_each_reverse(&c->journal_entries, iter, _i) {
1051                 i = *_i;
1052
1053                 if (!i || i->ignore)
1054                         continue;
1055
1056                 last_seq = le64_to_cpu(i->j.last_seq);
1057                 break;
1058         }
1059
1060         nr = cur_seq - last_seq;
1061
1062         if (nr + 1 > j->pin.size) {
1063                 free_fifo(&j->pin);
1064                 init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL);
1065                 if (!j->pin.data) {
1066                         bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
1067                         return -ENOMEM;
1068                 }
1069         }
1070
1071         j->replay_journal_seq   = last_seq;
1072         j->replay_journal_seq_end = cur_seq;
1073         j->last_seq_ondisk      = last_seq;
1074         j->flushed_seq_ondisk   = cur_seq - 1;
1075         j->seq_ondisk           = cur_seq - 1;
1076         j->pin.front            = last_seq;
1077         j->pin.back             = cur_seq;
1078         atomic64_set(&j->seq, cur_seq - 1);
1079
1080         fifo_for_each_entry_ptr(p, &j->pin, seq)
1081                 journal_pin_list_init(p, 1);
1082
1083         genradix_for_each(&c->journal_entries, iter, _i) {
1084                 i = *_i;
1085
1086                 if (!i || i->ignore)
1087                         continue;
1088
1089                 seq = le64_to_cpu(i->j.seq);
1090                 BUG_ON(seq >= cur_seq);
1091
1092                 if (seq < last_seq)
1093                         continue;
1094
1095                 if (journal_entry_empty(&i->j))
1096                         j->last_empty_seq = le64_to_cpu(i->j.seq);
1097
1098                 p = journal_seq_pin(j, seq);
1099
1100                 p->devs.nr = 0;
1101                 for (ptr = 0; ptr < i->nr_ptrs; ptr++)
1102                         bch2_dev_list_add_dev(&p->devs, i->ptrs[ptr].dev);
1103
1104                 had_entries = true;
1105         }
1106
1107         if (!had_entries)
1108                 j->last_empty_seq = cur_seq;
1109
1110         spin_lock(&j->lock);
1111
1112         set_bit(JOURNAL_STARTED, &j->flags);
1113         j->last_flush_write = jiffies;
1114
1115         j->reservations.idx = j->reservations.unwritten_idx = journal_cur_seq(j);
1116         j->reservations.unwritten_idx++;
1117
1118         c->last_bucket_seq_cleanup = journal_cur_seq(j);
1119
1120         bch2_journal_space_available(j);
1121         spin_unlock(&j->lock);
1122
1123         return bch2_journal_reclaim_start(j);
1124 }
1125
1126 /* init/exit: */
1127
1128 void bch2_dev_journal_exit(struct bch_dev *ca)
1129 {
1130         kfree(ca->journal.bio);
1131         kfree(ca->journal.buckets);
1132         kfree(ca->journal.bucket_seq);
1133
1134         ca->journal.bio         = NULL;
1135         ca->journal.buckets     = NULL;
1136         ca->journal.bucket_seq  = NULL;
1137 }
1138
1139 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
1140 {
1141         struct journal_device *ja = &ca->journal;
1142         struct bch_sb_field_journal *journal_buckets =
1143                 bch2_sb_get_journal(sb);
1144         struct bch_sb_field_journal_v2 *journal_buckets_v2 =
1145                 bch2_sb_get_journal_v2(sb);
1146         unsigned i, nr_bvecs;
1147
1148         ja->nr = 0;
1149
1150         if (journal_buckets_v2) {
1151                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1152
1153                 for (i = 0; i < nr; i++)
1154                         ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr);
1155         } else if (journal_buckets) {
1156                 ja->nr = bch2_nr_journal_buckets(journal_buckets);
1157         }
1158
1159         ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1160         if (!ja->bucket_seq)
1161                 return -ENOMEM;
1162
1163         nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
1164
1165         ca->journal.bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1166         if (!ca->journal.bio)
1167                 return -ENOMEM;
1168
1169         bio_init(ca->journal.bio, NULL, ca->journal.bio->bi_inline_vecs, nr_bvecs, 0);
1170
1171         ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1172         if (!ja->buckets)
1173                 return -ENOMEM;
1174
1175         if (journal_buckets_v2) {
1176                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1177                 unsigned j, dst = 0;
1178
1179                 for (i = 0; i < nr; i++)
1180                         for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
1181                                 ja->buckets[dst++] =
1182                                         le64_to_cpu(journal_buckets_v2->d[i].start) + j;
1183         } else if (journal_buckets) {
1184                 for (i = 0; i < ja->nr; i++)
1185                         ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
1186         }
1187
1188         return 0;
1189 }
1190
1191 void bch2_fs_journal_exit(struct journal *j)
1192 {
1193         unsigned i;
1194
1195         for (i = 0; i < ARRAY_SIZE(j->buf); i++)
1196                 kvpfree(j->buf[i].data, j->buf[i].buf_size);
1197         free_fifo(&j->pin);
1198 }
1199
1200 int bch2_fs_journal_init(struct journal *j)
1201 {
1202         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1203         static struct lock_class_key res_key;
1204         unsigned i;
1205         int ret = 0;
1206
1207         pr_verbose_init(c->opts, "");
1208
1209         spin_lock_init(&j->lock);
1210         spin_lock_init(&j->err_lock);
1211         init_waitqueue_head(&j->wait);
1212         INIT_DELAYED_WORK(&j->write_work, journal_write_work);
1213         init_waitqueue_head(&j->reclaim_wait);
1214         init_waitqueue_head(&j->pin_flush_wait);
1215         mutex_init(&j->reclaim_lock);
1216         mutex_init(&j->discard_lock);
1217
1218         lockdep_init_map(&j->res_map, "journal res", &res_key, 0);
1219
1220         atomic64_set(&j->reservations.counter,
1221                 ((union journal_res_state)
1222                  { .cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL }).v);
1223
1224         if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL))) {
1225                 ret = -ENOMEM;
1226                 goto out;
1227         }
1228
1229         for (i = 0; i < ARRAY_SIZE(j->buf); i++) {
1230                 j->buf[i].buf_size = JOURNAL_ENTRY_SIZE_MIN;
1231                 j->buf[i].data = kvpmalloc(j->buf[i].buf_size, GFP_KERNEL);
1232                 if (!j->buf[i].data) {
1233                         ret = -ENOMEM;
1234                         goto out;
1235                 }
1236         }
1237
1238         j->pin.front = j->pin.back = 1;
1239 out:
1240         pr_verbose_init(c->opts, "ret %i", ret);
1241         return ret;
1242 }
1243
1244 /* debug: */
1245
1246 void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1247 {
1248         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1249         union journal_res_state s;
1250         struct bch_dev *ca;
1251         unsigned long now = jiffies;
1252         u64 seq;
1253         unsigned i;
1254
1255         if (!out->nr_tabstops)
1256                 printbuf_tabstop_push(out, 24);
1257         out->atomic++;
1258
1259         rcu_read_lock();
1260         s = READ_ONCE(j->reservations);
1261
1262         prt_printf(out, "dirty journal entries:\t%llu/%llu\n",  fifo_used(&j->pin), j->pin.size);
1263         prt_printf(out, "seq:\t\t\t%llu\n",                     journal_cur_seq(j));
1264         prt_printf(out, "seq_ondisk:\t\t%llu\n",                j->seq_ondisk);
1265         prt_printf(out, "last_seq:\t\t%llu\n",          journal_last_seq(j));
1266         prt_printf(out, "last_seq_ondisk:\t%llu\n",             j->last_seq_ondisk);
1267         prt_printf(out, "flushed_seq_ondisk:\t%llu\n",  j->flushed_seq_ondisk);
1268         prt_printf(out, "prereserved:\t\t%u/%u\n",              j->prereserved.reserved, j->prereserved.remaining);
1269         prt_printf(out, "watermark:\t\t%s\n",           bch2_journal_watermarks[j->watermark]);
1270         prt_printf(out, "each entry reserved:\t%u\n",   j->entry_u64s_reserved);
1271         prt_printf(out, "nr flush writes:\t%llu\n",             j->nr_flush_writes);
1272         prt_printf(out, "nr noflush writes:\t%llu\n",   j->nr_noflush_writes);
1273         prt_printf(out, "nr direct reclaim:\t%llu\n",   j->nr_direct_reclaim);
1274         prt_printf(out, "nr background reclaim:\t%llu\n",       j->nr_background_reclaim);
1275         prt_printf(out, "reclaim kicked:\t\t%u\n",              j->reclaim_kicked);
1276         prt_printf(out, "reclaim runs in:\t%u ms\n",    time_after(j->next_reclaim, now)
1277                ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0);
1278         prt_printf(out, "current entry sectors:\t%u\n", j->cur_entry_sectors);
1279         prt_printf(out, "current entry error:\t%s\n",   bch2_journal_errors[j->cur_entry_error]);
1280         prt_printf(out, "current entry:\t\t");
1281
1282         switch (s.cur_entry_offset) {
1283         case JOURNAL_ENTRY_ERROR_VAL:
1284                 prt_printf(out, "error");
1285                 break;
1286         case JOURNAL_ENTRY_CLOSED_VAL:
1287                 prt_printf(out, "closed");
1288                 break;
1289         default:
1290                 prt_printf(out, "%u/%u", s.cur_entry_offset, j->cur_entry_u64s);
1291                 break;
1292         }
1293
1294         prt_newline(out);
1295
1296         for (seq = journal_cur_seq(j);
1297              seq >= journal_last_unwritten_seq(j);
1298              --seq) {
1299                 i = seq & JOURNAL_BUF_MASK;
1300
1301                 prt_printf(out, "unwritten entry:");
1302                 prt_tab(out);
1303                 prt_printf(out, "%llu", seq);
1304                 prt_newline(out);
1305                 printbuf_indent_add(out, 2);
1306
1307                 prt_printf(out, "refcount:");
1308                 prt_tab(out);
1309                 prt_printf(out, "%u", journal_state_count(s, i));
1310                 prt_newline(out);
1311
1312                 prt_printf(out, "sectors:");
1313                 prt_tab(out);
1314                 prt_printf(out, "%u", j->buf[i].sectors);
1315                 prt_newline(out);
1316
1317                 prt_printf(out, "expires");
1318                 prt_tab(out);
1319                 prt_printf(out, "%li jiffies", j->buf[i].expires - jiffies);
1320                 prt_newline(out);
1321
1322                 printbuf_indent_sub(out, 2);
1323         }
1324
1325         prt_printf(out,
1326                "replay done:\t\t%i\n",
1327                test_bit(JOURNAL_REPLAY_DONE,    &j->flags));
1328
1329         prt_printf(out, "space:\n");
1330         prt_printf(out, "\tdiscarded\t%u:%u\n",
1331                j->space[journal_space_discarded].next_entry,
1332                j->space[journal_space_discarded].total);
1333         prt_printf(out, "\tclean ondisk\t%u:%u\n",
1334                j->space[journal_space_clean_ondisk].next_entry,
1335                j->space[journal_space_clean_ondisk].total);
1336         prt_printf(out, "\tclean\t\t%u:%u\n",
1337                j->space[journal_space_clean].next_entry,
1338                j->space[journal_space_clean].total);
1339         prt_printf(out, "\ttotal\t\t%u:%u\n",
1340                j->space[journal_space_total].next_entry,
1341                j->space[journal_space_total].total);
1342
1343         for_each_member_device_rcu(ca, c, i,
1344                                    &c->rw_devs[BCH_DATA_journal]) {
1345                 struct journal_device *ja = &ca->journal;
1346
1347                 if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d))
1348                         continue;
1349
1350                 if (!ja->nr)
1351                         continue;
1352
1353                 prt_printf(out, "dev %u:\n",            i);
1354                 prt_printf(out, "\tnr\t\t%u\n",         ja->nr);
1355                 prt_printf(out, "\tbucket size\t%u\n",  ca->mi.bucket_size);
1356                 prt_printf(out, "\tavailable\t%u:%u\n", bch2_journal_dev_buckets_available(j, ja, journal_space_discarded), ja->sectors_free);
1357                 prt_printf(out, "\tdiscard_idx\t%u\n",  ja->discard_idx);
1358                 prt_printf(out, "\tdirty_ondisk\t%u (seq %llu)\n", ja->dirty_idx_ondisk,        ja->bucket_seq[ja->dirty_idx_ondisk]);
1359                 prt_printf(out, "\tdirty_idx\t%u (seq %llu)\n", ja->dirty_idx,          ja->bucket_seq[ja->dirty_idx]);
1360                 prt_printf(out, "\tcur_idx\t\t%u (seq %llu)\n", ja->cur_idx,            ja->bucket_seq[ja->cur_idx]);
1361         }
1362
1363         rcu_read_unlock();
1364
1365         --out->atomic;
1366 }
1367
1368 void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1369 {
1370         spin_lock(&j->lock);
1371         __bch2_journal_debug_to_text(out, j);
1372         spin_unlock(&j->lock);
1373 }
1374
1375 bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq)
1376 {
1377         struct journal_entry_pin_list *pin_list;
1378         struct journal_entry_pin *pin;
1379
1380         spin_lock(&j->lock);
1381         *seq = max(*seq, j->pin.front);
1382
1383         if (*seq >= j->pin.back) {
1384                 spin_unlock(&j->lock);
1385                 return true;
1386         }
1387
1388         out->atomic++;
1389
1390         pin_list = journal_seq_pin(j, *seq);
1391
1392         prt_printf(out, "%llu: count %u", *seq, atomic_read(&pin_list->count));
1393         prt_newline(out);
1394         printbuf_indent_add(out, 2);
1395
1396         list_for_each_entry(pin, &pin_list->list, list) {
1397                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1398                 prt_newline(out);
1399         }
1400
1401         list_for_each_entry(pin, &pin_list->key_cache_list, list) {
1402                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1403                 prt_newline(out);
1404         }
1405
1406         if (!list_empty(&pin_list->flushed)) {
1407                 prt_printf(out, "flushed:");
1408                 prt_newline(out);
1409         }
1410
1411         list_for_each_entry(pin, &pin_list->flushed, list) {
1412                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1413                 prt_newline(out);
1414         }
1415
1416         printbuf_indent_sub(out, 2);
1417
1418         --out->atomic;
1419         spin_unlock(&j->lock);
1420
1421         return false;
1422 }
1423
1424 void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
1425 {
1426         u64 seq = 0;
1427
1428         while (!bch2_journal_seq_pins_to_text(out, j, &seq))
1429                 seq++;
1430 }