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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcachefs journalling code, for btree insertions
   4  *
   5  * Copyright 2012 Google, Inc.
   6  */
   7
   8 #include "bcachefs.h"
   9 #include "alloc_foreground.h"
  10 #include "bkey_methods.h"
  11 #include "btree_gc.h"
  12 #include "btree_update.h"
  13 #include "buckets.h"
  14 #include "error.h"
  15 #include "journal.h"
  16 #include "journal_io.h"
  17 #include "journal_reclaim.h"
  18 #include "journal_sb.h"
  19 #include "journal_seq_blacklist.h"
  20 #include "trace.h"
  21
  22 #define x(n)    #n,
  23 static const char * const bch2_journal_watermarks[] = {
  24         JOURNAL_WATERMARKS()
  25         NULL
  26 };
  27
  28 static const char * const bch2_journal_errors[] = {
  29         JOURNAL_ERRORS()
  30         NULL
  31 };
  32 #undef x
  33
  34 static inline bool journal_seq_unwritten(struct journal *j, u64 seq)
  35 {
  36         return seq > j->seq_ondisk;
  37 }
  38
  39 static bool __journal_entry_is_open(union journal_res_state state)
  40 {
  41         return state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL;
  42 }
  43
  44 static inline unsigned nr_unwritten_journal_entries(struct journal *j)
  45 {
  46         return atomic64_read(&j->seq) - j->seq_ondisk;
  47 }
  48
  49 static bool journal_entry_is_open(struct journal *j)
  50 {
  51         return __journal_entry_is_open(j->reservations);
  52 }
  53
  54 static inline struct journal_buf *
  55 journal_seq_to_buf(struct journal *j, u64 seq)
  56 {
  57         struct journal_buf *buf = NULL;
  58
  59         EBUG_ON(seq > journal_cur_seq(j));
  60
  61         if (journal_seq_unwritten(j, seq)) {
  62                 buf = j->buf + (seq & JOURNAL_BUF_MASK);
  63                 EBUG_ON(le64_to_cpu(buf->data->seq) != seq);
  64         }
  65         return buf;
  66 }
  67
  68 static void journal_pin_list_init(struct journal_entry_pin_list *p, int count)
  69 {
  70         unsigned i;
  71         for (i = 0; i < ARRAY_SIZE(p->list); i++)
  72                 INIT_LIST_HEAD(&p->list[i]);
  73         INIT_LIST_HEAD(&p->flushed);
  74         atomic_set(&p->count, count);
  75         p->devs.nr = 0;
  76 }
  77
  78 /*
  79  * Detect stuck journal conditions and trigger shutdown. Technically the journal
  80  * can end up stuck for a variety of reasons, such as a blocked I/O, journal
  81  * reservation lockup, etc. Since this is a fatal error with potentially
  82  * unpredictable characteristics, we want to be fairly conservative before we
  83  * decide to shut things down.
  84  *
  85  * Consider the journal stuck when it appears full with no ability to commit
  86  * btree transactions, to discard journal buckets, nor acquire priority
  87  * (reserved watermark) reservation.
  88  */
  89 static inline bool
  90 journal_error_check_stuck(struct journal *j, int error, unsigned flags)
  91 {
  92         struct bch_fs *c = container_of(j, struct bch_fs, journal);
  93         bool stuck = false;
  94         struct printbuf buf = PRINTBUF;
  95
  96         if (!(error == JOURNAL_ERR_journal_full ||
  97               error == JOURNAL_ERR_journal_pin_full) ||
  98             nr_unwritten_journal_entries(j) ||
  99             (flags & JOURNAL_WATERMARK_MASK) != JOURNAL_WATERMARK_reserved)
 100                 return stuck;
 101
 102         spin_lock(&j->lock);
 103
 104         if (j->can_discard) {
 105                 spin_unlock(&j->lock);
 106                 return stuck;
 107         }
 108
 109         stuck = true;
 110
 111         /*
 112          * The journal shutdown path will set ->err_seq, but do it here first to
 113          * serialize against concurrent failures and avoid duplicate error
 114          * reports.
 115          */
 116         if (j->err_seq) {
 117                 spin_unlock(&j->lock);
 118                 return stuck;
 119         }
 120         j->err_seq = journal_cur_seq(j);
 121         spin_unlock(&j->lock);
 122
 123         bch_err(c, "Journal stuck! Hava a pre-reservation but journal full (error %s)",
 124                 bch2_journal_errors[error]);
 125         bch2_journal_debug_to_text(&buf, j);
 126         bch_err(c, "%s", buf.buf);
 127
 128         printbuf_reset(&buf);
 129         bch2_journal_pins_to_text(&buf, j);
 130         bch_err(c, "Journal pins:\n%s", buf.buf);
 131         printbuf_exit(&buf);
 132
 133         bch2_fatal_error(c);
 134         dump_stack();
 135
 136         return stuck;
 137 }
 138
 139 /* journal entry close/open: */
 140
 141 void __bch2_journal_buf_put(struct journal *j)
 142 {
 143         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 144
 145         closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
 146 }
 147
 148 /*
 149  * Returns true if journal entry is now closed:
 150  *
 151  * We don't close a journal_buf until the next journal_buf is finished writing,
 152  * and can be opened again - this also initializes the next journal_buf:
 153  */
 154 static void __journal_entry_close(struct journal *j, unsigned closed_val)
 155 {
 156         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 157         struct journal_buf *buf = journal_cur_buf(j);
 158         union journal_res_state old, new;
 159         u64 v = atomic64_read(&j->reservations.counter);
 160         unsigned sectors;
 161
 162         BUG_ON(closed_val != JOURNAL_ENTRY_CLOSED_VAL &&
 163                closed_val != JOURNAL_ENTRY_ERROR_VAL);
 164
 165         lockdep_assert_held(&j->lock);
 166
 167         do {
 168                 old.v = new.v = v;
 169                 new.cur_entry_offset = closed_val;
 170
 171                 if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL ||
 172                     old.cur_entry_offset == new.cur_entry_offset)
 173                         return;
 174         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 175                                        old.v, new.v)) != old.v);
 176
 177         if (!__journal_entry_is_open(old))
 178                 return;
 179
 180         /* Close out old buffer: */
 181         buf->data->u64s         = cpu_to_le32(old.cur_entry_offset);
 182
 183         sectors = vstruct_blocks_plus(buf->data, c->block_bits,
 184                                       buf->u64s_reserved) << c->block_bits;
 185         BUG_ON(sectors > buf->sectors);
 186         buf->sectors = sectors;
 187
 188         /*
 189          * We have to set last_seq here, _before_ opening a new journal entry:
 190          *
 191          * A threads may replace an old pin with a new pin on their current
 192          * journal reservation - the expectation being that the journal will
 193          * contain either what the old pin protected or what the new pin
 194          * protects.
 195          *
 196          * After the old pin is dropped journal_last_seq() won't include the old
 197          * pin, so we can only write the updated last_seq on the entry that
 198          * contains whatever the new pin protects.
 199          *
 200          * Restated, we can _not_ update last_seq for a given entry if there
 201          * could be a newer entry open with reservations/pins that have been
 202          * taken against it.
 203          *
 204          * Hence, we want update/set last_seq on the current journal entry right
 205          * before we open a new one:
 206          */
 207         buf->last_seq           = journal_last_seq(j);
 208         buf->data->last_seq     = cpu_to_le64(buf->last_seq);
 209         BUG_ON(buf->last_seq > le64_to_cpu(buf->data->seq));
 210
 211         __bch2_journal_pin_put(j, le64_to_cpu(buf->data->seq));
 212
 213         cancel_delayed_work(&j->write_work);
 214
 215         bch2_journal_space_available(j);
 216
 217         bch2_journal_buf_put(j, old.idx);
 218 }
 219
 220 void bch2_journal_halt(struct journal *j)
 221 {
 222         spin_lock(&j->lock);
 223         __journal_entry_close(j, JOURNAL_ENTRY_ERROR_VAL);
 224         if (!j->err_seq)
 225                 j->err_seq = journal_cur_seq(j);
 226         journal_wake(j);
 227         spin_unlock(&j->lock);
 228 }
 229
 230 static bool journal_entry_want_write(struct journal *j)
 231 {
 232         bool ret = !journal_entry_is_open(j) ||
 233                 journal_cur_seq(j) == journal_last_unwritten_seq(j);
 234
 235         /* Don't close it yet if we already have a write in flight: */
 236         if (ret)
 237                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 238         else if (nr_unwritten_journal_entries(j)) {
 239                 struct journal_buf *buf = journal_cur_buf(j);
 240
 241                 if (!buf->flush_time) {
 242                         buf->flush_time = local_clock() ?: 1;
 243                         buf->expires = jiffies;
 244                 }
 245         }
 246
 247         return ret;
 248 }
 249
 250 static bool journal_entry_close(struct journal *j)
 251 {
 252         bool ret;
 253
 254         spin_lock(&j->lock);
 255         ret = journal_entry_want_write(j);
 256         spin_unlock(&j->lock);
 257
 258         return ret;
 259 }
 260
 261 /*
 262  * should _only_ called from journal_res_get() - when we actually want a
 263  * journal reservation - journal entry is open means journal is dirty:
 264  */
 265 static int journal_entry_open(struct journal *j)
 266 {
 267         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 268         struct journal_buf *buf = j->buf +
 269                 ((journal_cur_seq(j) + 1) & JOURNAL_BUF_MASK);
 270         union journal_res_state old, new;
 271         int u64s;
 272         u64 v;
 273
 274         lockdep_assert_held(&j->lock);
 275         BUG_ON(journal_entry_is_open(j));
 276         BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
 277
 278         if (j->blocked)
 279                 return JOURNAL_ERR_blocked;
 280
 281         if (j->cur_entry_error)
 282                 return j->cur_entry_error;
 283
 284         if (bch2_journal_error(j))
 285                 return JOURNAL_ERR_insufficient_devices; /* -EROFS */
 286
 287         if (!fifo_free(&j->pin))
 288                 return JOURNAL_ERR_journal_pin_full;
 289
 290         if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf))
 291                 return JOURNAL_ERR_max_in_flight;
 292
 293         BUG_ON(!j->cur_entry_sectors);
 294
 295         buf->expires            =
 296                 (journal_cur_seq(j) == j->flushed_seq_ondisk
 297                  ? jiffies
 298                  : j->last_flush_write) +
 299                 msecs_to_jiffies(c->opts.journal_flush_delay);
 300
 301         buf->u64s_reserved      = j->entry_u64s_reserved;
 302         buf->disk_sectors       = j->cur_entry_sectors;
 303         buf->sectors            = min(buf->disk_sectors, buf->buf_size >> 9);
 304
 305         u64s = (int) (buf->sectors << 9) / sizeof(u64) -
 306                 journal_entry_overhead(j);
 307         u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 308
 309         if (u64s <= (ssize_t) j->early_journal_entries.nr)
 310                 return JOURNAL_ERR_journal_full;
 311
 312         if (fifo_empty(&j->pin) && j->reclaim_thread)
 313                 wake_up_process(j->reclaim_thread);
 314
 315         /*
 316          * The fifo_push() needs to happen at the same time as j->seq is
 317          * incremented for journal_last_seq() to be calculated correctly
 318          */
 319         atomic64_inc(&j->seq);
 320         journal_pin_list_init(fifo_push_ref(&j->pin), 1);
 321
 322         BUG_ON(j->buf + (journal_cur_seq(j) & JOURNAL_BUF_MASK) != buf);
 323
 324         bkey_extent_init(&buf->key);
 325         buf->noflush    = false;
 326         buf->must_flush = false;
 327         buf->separate_flush = false;
 328         buf->flush_time = 0;
 329
 330         memset(buf->data, 0, sizeof(*buf->data));
 331         buf->data->seq  = cpu_to_le64(journal_cur_seq(j));
 332         buf->data->u64s = 0;
 333
 334         if (j->early_journal_entries.nr) {
 335                 memcpy(buf->data->_data, j->early_journal_entries.data,
 336                        j->early_journal_entries.nr * sizeof(u64));
 337                 le32_add_cpu(&buf->data->u64s, j->early_journal_entries.nr);
 338         }
 339
 340         /*
 341          * Must be set before marking the journal entry as open:
 342          */
 343         j->cur_entry_u64s = u64s;
 344
 345         v = atomic64_read(&j->reservations.counter);
 346         do {
 347                 old.v = new.v = v;
 348
 349                 BUG_ON(old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL);
 350
 351                 new.idx++;
 352                 BUG_ON(journal_state_count(new, new.idx));
 353                 BUG_ON(new.idx != (journal_cur_seq(j) & JOURNAL_BUF_MASK));
 354
 355                 journal_state_inc(&new);
 356
 357                 /* Handle any already added entries */
 358                 new.cur_entry_offset = le32_to_cpu(buf->data->u64s);
 359         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 360                                        old.v, new.v)) != old.v);
 361
 362         if (j->res_get_blocked_start)
 363                 bch2_time_stats_update(j->blocked_time,
 364                                        j->res_get_blocked_start);
 365         j->res_get_blocked_start = 0;
 366
 367         mod_delayed_work(c->io_complete_wq,
 368                          &j->write_work,
 369                          msecs_to_jiffies(c->opts.journal_flush_delay));
 370         journal_wake(j);
 371
 372         if (j->early_journal_entries.nr)
 373                 darray_exit(&j->early_journal_entries);
 374         return 0;
 375 }
 376
 377 static bool journal_quiesced(struct journal *j)
 378 {
 379         bool ret = atomic64_read(&j->seq) == j->seq_ondisk;
 380
 381         if (!ret)
 382                 journal_entry_close(j);
 383         return ret;
 384 }
 385
 386 static void journal_quiesce(struct journal *j)
 387 {
 388         wait_event(j->wait, journal_quiesced(j));
 389 }
 390
 391 static void journal_write_work(struct work_struct *work)
 392 {
 393         struct journal *j = container_of(work, struct journal, write_work.work);
 394         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 395         long delta;
 396
 397         spin_lock(&j->lock);
 398         if (!__journal_entry_is_open(j->reservations))
 399                 goto unlock;
 400
 401         delta = journal_cur_buf(j)->expires - jiffies;
 402
 403         if (delta > 0)
 404                 mod_delayed_work(c->io_complete_wq, &j->write_work, delta);
 405         else
 406                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 407 unlock:
 408         spin_unlock(&j->lock);
 409 }
 410
 411 static int __journal_res_get(struct journal *j, struct journal_res *res,
 412                              unsigned flags)
 413 {
 414         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 415         struct journal_buf *buf;
 416         bool can_discard;
 417         int ret;
 418 retry:
 419         if (journal_res_get_fast(j, res, flags))
 420                 return 0;
 421
 422         if (bch2_journal_error(j))
 423                 return -BCH_ERR_erofs_journal_err;
 424
 425         spin_lock(&j->lock);
 426
 427         /* check once more in case somebody else shut things down... */
 428         if (bch2_journal_error(j)) {
 429                 spin_unlock(&j->lock);
 430                 return -BCH_ERR_erofs_journal_err;
 431         }
 432
 433         /*
 434          * Recheck after taking the lock, so we don't race with another thread
 435          * that just did journal_entry_open() and call journal_entry_close()
 436          * unnecessarily
 437          */
 438         if (journal_res_get_fast(j, res, flags)) {
 439                 spin_unlock(&j->lock);
 440                 return 0;
 441         }
 442
 443         if ((flags & JOURNAL_WATERMARK_MASK) < j->watermark) {
 444                 /*
 445                  * Don't want to close current journal entry, just need to
 446                  * invoke reclaim:
 447                  */
 448                 ret = JOURNAL_ERR_journal_full;
 449                 goto unlock;
 450         }
 451
 452         /*
 453          * If we couldn't get a reservation because the current buf filled up,
 454          * and we had room for a bigger entry on disk, signal that we want to
 455          * realloc the journal bufs:
 456          */
 457         buf = journal_cur_buf(j);
 458         if (journal_entry_is_open(j) &&
 459             buf->buf_size >> 9 < buf->disk_sectors &&
 460             buf->buf_size < JOURNAL_ENTRY_SIZE_MAX)
 461                 j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1);
 462
 463         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 464         ret = journal_entry_open(j);
 465
 466         if (ret == JOURNAL_ERR_max_in_flight)
 467                 trace_and_count(c, journal_entry_full, c);
 468 unlock:
 469         if ((ret && ret != JOURNAL_ERR_insufficient_devices) &&
 470             !j->res_get_blocked_start) {
 471                 j->res_get_blocked_start = local_clock() ?: 1;
 472                 trace_and_count(c, journal_full, c);
 473         }
 474
 475         can_discard = j->can_discard;
 476         spin_unlock(&j->lock);
 477
 478         if (!ret)
 479                 goto retry;
 480         if (journal_error_check_stuck(j, ret, flags))
 481                 ret = -BCH_ERR_journal_res_get_blocked;
 482
 483         /*
 484          * Journal is full - can't rely on reclaim from work item due to
 485          * freezing:
 486          */
 487         if ((ret == JOURNAL_ERR_journal_full ||
 488              ret == JOURNAL_ERR_journal_pin_full) &&
 489             !(flags & JOURNAL_RES_GET_NONBLOCK)) {
 490                 if (can_discard) {
 491                         bch2_journal_do_discards(j);
 492                         goto retry;
 493                 }
 494
 495                 if (mutex_trylock(&j->reclaim_lock)) {
 496                         bch2_journal_reclaim(j);
 497                         mutex_unlock(&j->reclaim_lock);
 498                 }
 499         }
 500
 501         return ret == JOURNAL_ERR_insufficient_devices
 502                 ? -EROFS
 503                 : -BCH_ERR_journal_res_get_blocked;
 504 }
 505
 506 /*
 507  * Essentially the entry function to the journaling code. When bcachefs is doing
 508  * a btree insert, it calls this function to get the current journal write.
 509  * Journal write is the structure used set up journal writes. The calling
 510  * function will then add its keys to the structure, queuing them for the next
 511  * write.
 512  *
 513  * To ensure forward progress, the current task must not be holding any
 514  * btree node write locks.
 515  */
 516 int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
 517                                   unsigned flags)
 518 {
 519         int ret;
 520
 521         closure_wait_event(&j->async_wait,
 522                    (ret = __journal_res_get(j, res, flags)) !=
 523                    -BCH_ERR_journal_res_get_blocked||
 524                    (flags & JOURNAL_RES_GET_NONBLOCK));
 525         return ret;
 526 }
 527
 528 /* journal_preres: */
 529
 530 static bool journal_preres_available(struct journal *j,
 531                                      struct journal_preres *res,
 532                                      unsigned new_u64s,
 533                                      unsigned flags)
 534 {
 535         bool ret = bch2_journal_preres_get_fast(j, res, new_u64s, flags, true);
 536
 537         if (!ret && mutex_trylock(&j->reclaim_lock)) {
 538                 bch2_journal_reclaim(j);
 539                 mutex_unlock(&j->reclaim_lock);
 540         }
 541
 542         return ret;
 543 }
 544
 545 int __bch2_journal_preres_get(struct journal *j,
 546                               struct journal_preres *res,
 547                               unsigned new_u64s,
 548                               unsigned flags)
 549 {
 550         int ret;
 551
 552         closure_wait_event(&j->preres_wait,
 553                    (ret = bch2_journal_error(j)) ||
 554                    journal_preres_available(j, res, new_u64s, flags));
 555         return ret;
 556 }
 557
 558 /* journal_entry_res: */
 559
 560 void bch2_journal_entry_res_resize(struct journal *j,
 561                                    struct journal_entry_res *res,
 562                                    unsigned new_u64s)
 563 {
 564         union journal_res_state state;
 565         int d = new_u64s - res->u64s;
 566
 567         spin_lock(&j->lock);
 568
 569         j->entry_u64s_reserved += d;
 570         if (d <= 0)
 571                 goto out;
 572
 573         j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d);
 574         smp_mb();
 575         state = READ_ONCE(j->reservations);
 576
 577         if (state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL &&
 578             state.cur_entry_offset > j->cur_entry_u64s) {
 579                 j->cur_entry_u64s += d;
 580                 /*
 581                  * Not enough room in current journal entry, have to flush it:
 582                  */
 583                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 584         } else {
 585                 journal_cur_buf(j)->u64s_reserved += d;
 586         }
 587 out:
 588         spin_unlock(&j->lock);
 589         res->u64s += d;
 590 }
 591
 592 /* journal flushing: */
 593
 594 /**
 595  * bch2_journal_flush_seq_async - wait for a journal entry to be written
 596  *
 597  * like bch2_journal_wait_on_seq, except that it triggers a write immediately if
 598  * necessary
 599  */
 600 int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 601                                  struct closure *parent)
 602 {
 603         struct journal_buf *buf;
 604         int ret = 0;
 605
 606         if (seq <= j->flushed_seq_ondisk)
 607                 return 1;
 608
 609         spin_lock(&j->lock);
 610
 611         if (WARN_ONCE(seq > journal_cur_seq(j),
 612                       "requested to flush journal seq %llu, but currently at %llu",
 613                       seq, journal_cur_seq(j)))
 614                 goto out;
 615
 616         /* Recheck under lock: */
 617         if (j->err_seq && seq >= j->err_seq) {
 618                 ret = -EIO;
 619                 goto out;
 620         }
 621
 622         if (seq <= j->flushed_seq_ondisk) {
 623                 ret = 1;
 624                 goto out;
 625         }
 626
 627         /* if seq was written, but not flushed - flush a newer one instead */
 628         seq = max(seq, journal_last_unwritten_seq(j));
 629
 630 recheck_need_open:
 631         if (seq > journal_cur_seq(j)) {
 632                 struct journal_res res = { 0 };
 633
 634                 if (journal_entry_is_open(j))
 635                         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 636
 637                 spin_unlock(&j->lock);
 638
 639                 ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 640                 if (ret)
 641                         return ret;
 642
 643                 seq = res.seq;
 644                 buf = j->buf + (seq & JOURNAL_BUF_MASK);
 645                 buf->must_flush = true;
 646
 647                 if (!buf->flush_time) {
 648                         buf->flush_time = local_clock() ?: 1;
 649                         buf->expires = jiffies;
 650                 }
 651
 652                 if (parent && !closure_wait(&buf->wait, parent))
 653                         BUG();
 654
 655                 bch2_journal_res_put(j, &res);
 656
 657                 spin_lock(&j->lock);
 658                 goto want_write;
 659         }
 660
 661         /*
 662          * if write was kicked off without a flush, flush the next sequence
 663          * number instead
 664          */
 665         buf = journal_seq_to_buf(j, seq);
 666         if (buf->noflush) {
 667                 seq++;
 668                 goto recheck_need_open;
 669         }
 670
 671         buf->must_flush = true;
 672
 673         if (parent && !closure_wait(&buf->wait, parent))
 674                 BUG();
 675 want_write:
 676         if (seq == journal_cur_seq(j))
 677                 journal_entry_want_write(j);
 678 out:
 679         spin_unlock(&j->lock);
 680         return ret;
 681 }
 682
 683 int bch2_journal_flush_seq(struct journal *j, u64 seq)
 684 {
 685         u64 start_time = local_clock();
 686         int ret, ret2;
 687
 688         /*
 689          * Don't update time_stats when @seq is already flushed:
 690          */
 691         if (seq <= j->flushed_seq_ondisk)
 692                 return 0;
 693
 694         ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)));
 695
 696         if (!ret)
 697                 bch2_time_stats_update(j->flush_seq_time, start_time);
 698
 699         return ret ?: ret2 < 0 ? ret2 : 0;
 700 }
 701
 702 /*
 703  * bch2_journal_flush_async - if there is an open journal entry, or a journal
 704  * still being written, write it and wait for the write to complete
 705  */
 706 void bch2_journal_flush_async(struct journal *j, struct closure *parent)
 707 {
 708         bch2_journal_flush_seq_async(j, atomic64_read(&j->seq), parent);
 709 }
 710
 711 int bch2_journal_flush(struct journal *j)
 712 {
 713         return bch2_journal_flush_seq(j, atomic64_read(&j->seq));
 714 }
 715
 716 /*
 717  * bch2_journal_noflush_seq - tell the journal not to issue any flushes before
 718  * @seq
 719  */
 720 bool bch2_journal_noflush_seq(struct journal *j, u64 seq)
 721 {
 722         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 723         u64 unwritten_seq;
 724         bool ret = false;
 725
 726         if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush)))
 727                 return false;
 728
 729         if (seq <= c->journal.flushed_seq_ondisk)
 730                 return false;
 731
 732         spin_lock(&j->lock);
 733         if (seq <= c->journal.flushed_seq_ondisk)
 734                 goto out;
 735
 736         for (unwritten_seq = journal_last_unwritten_seq(j);
 737              unwritten_seq < seq;
 738              unwritten_seq++) {
 739                 struct journal_buf *buf = journal_seq_to_buf(j, unwritten_seq);
 740
 741                 /* journal write is already in flight, and was a flush write: */
 742                 if (unwritten_seq == journal_last_unwritten_seq(j) && !buf->noflush)
 743                         goto out;
 744
 745                 buf->noflush = true;
 746         }
 747
 748         ret = true;
 749 out:
 750         spin_unlock(&j->lock);
 751         return ret;
 752 }
 753
 754 int bch2_journal_meta(struct journal *j)
 755 {
 756         struct journal_buf *buf;
 757         struct journal_res res;
 758         int ret;
 759
 760         memset(&res, 0, sizeof(res));
 761
 762         ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 763         if (ret)
 764                 return ret;
 765
 766         buf = j->buf + (res.seq & JOURNAL_BUF_MASK);
 767         buf->must_flush = true;
 768
 769         if (!buf->flush_time) {
 770                 buf->flush_time = local_clock() ?: 1;
 771                 buf->expires = jiffies;
 772         }
 773
 774         bch2_journal_res_put(j, &res);
 775
 776         return bch2_journal_flush_seq(j, res.seq);
 777 }
 778
 779 /* block/unlock the journal: */
 780
 781 void bch2_journal_unblock(struct journal *j)
 782 {
 783         spin_lock(&j->lock);
 784         j->blocked--;
 785         spin_unlock(&j->lock);
 786
 787         journal_wake(j);
 788 }
 789
 790 void bch2_journal_block(struct journal *j)
 791 {
 792         spin_lock(&j->lock);
 793         j->blocked++;
 794         spin_unlock(&j->lock);
 795
 796         journal_quiesce(j);
 797 }
 798
 799 /* allocate journal on a device: */
 800
 801 static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 802                                          bool new_fs, struct closure *cl)
 803 {
 804         struct bch_fs *c = ca->fs;
 805         struct journal_device *ja = &ca->journal;
 806         u64 *new_bucket_seq = NULL, *new_buckets = NULL;
 807         struct open_bucket **ob = NULL;
 808         long *bu = NULL;
 809         unsigned i, pos, nr_got = 0, nr_want = nr - ja->nr;
 810         int ret = 0;
 811
 812         BUG_ON(nr <= ja->nr);
 813
 814         bu              = kcalloc(nr_want, sizeof(*bu), GFP_KERNEL);
 815         ob              = kcalloc(nr_want, sizeof(*ob), GFP_KERNEL);
 816         new_buckets     = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 817         new_bucket_seq  = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 818         if (!bu || !ob || !new_buckets || !new_bucket_seq) {
 819                 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets;
 820                 goto err_free;
 821         }
 822
 823         for (nr_got = 0; nr_got < nr_want; nr_got++) {
 824                 if (new_fs) {
 825                         bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
 826                         if (bu[nr_got] < 0) {
 827                                 ret = -BCH_ERR_ENOSPC_bucket_alloc;
 828                                 break;
 829                         }
 830                 } else {
 831                         ob[nr_got] = bch2_bucket_alloc(c, ca, RESERVE_none, cl);
 832                         ret = PTR_ERR_OR_ZERO(ob[nr_got]);
 833                         if (ret)
 834                                 break;
 835
 836                         ret = bch2_trans_run(c,
 837                                 bch2_trans_mark_metadata_bucket(&trans, ca,
 838                                                 ob[nr_got]->bucket, BCH_DATA_journal,
 839                                                 ca->mi.bucket_size));
 840                         if (ret) {
 841                                 bch2_open_bucket_put(c, ob[nr_got]);
 842                                 bch_err(c, "error marking new journal buckets: %s", bch2_err_str(ret));
 843                                 break;
 844                         }
 845
 846                         bu[nr_got] = ob[nr_got]->bucket;
 847                 }
 848         }
 849
 850         if (!nr_got)
 851                 goto err_free;
 852
 853         /* Don't return an error if we successfully allocated some buckets: */
 854         ret = 0;
 855
 856         if (c) {
 857                 bch2_journal_flush_all_pins(&c->journal);
 858                 bch2_journal_block(&c->journal);
 859                 mutex_lock(&c->sb_lock);
 860         }
 861
 862         memcpy(new_buckets,     ja->buckets,    ja->nr * sizeof(u64));
 863         memcpy(new_bucket_seq,  ja->bucket_seq, ja->nr * sizeof(u64));
 864
 865         BUG_ON(ja->discard_idx > ja->nr);
 866
 867         pos = ja->discard_idx ?: ja->nr;
 868
 869         memmove(new_buckets + pos + nr_got,
 870                 new_buckets + pos,
 871                 sizeof(new_buckets[0]) * (ja->nr - pos));
 872         memmove(new_bucket_seq + pos + nr_got,
 873                 new_bucket_seq + pos,
 874                 sizeof(new_bucket_seq[0]) * (ja->nr - pos));
 875
 876         for (i = 0; i < nr_got; i++) {
 877                 new_buckets[pos + i] = bu[i];
 878                 new_bucket_seq[pos + i] = 0;
 879         }
 880
 881         nr = ja->nr + nr_got;
 882
 883         ret = bch2_journal_buckets_to_sb(c, ca, new_buckets, nr);
 884         if (ret)
 885                 goto err_unblock;
 886
 887         if (!new_fs)
 888                 bch2_write_super(c);
 889
 890         /* Commit: */
 891         if (c)
 892                 spin_lock(&c->journal.lock);
 893
 894         swap(new_buckets,       ja->buckets);
 895         swap(new_bucket_seq,    ja->bucket_seq);
 896         ja->nr = nr;
 897
 898         if (pos <= ja->discard_idx)
 899                 ja->discard_idx = (ja->discard_idx + nr_got) % ja->nr;
 900         if (pos <= ja->dirty_idx_ondisk)
 901                 ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + nr_got) % ja->nr;
 902         if (pos <= ja->dirty_idx)
 903                 ja->dirty_idx = (ja->dirty_idx + nr_got) % ja->nr;
 904         if (pos <= ja->cur_idx)
 905                 ja->cur_idx = (ja->cur_idx + nr_got) % ja->nr;
 906
 907         if (c)
 908                 spin_unlock(&c->journal.lock);
 909 err_unblock:
 910         if (c) {
 911                 bch2_journal_unblock(&c->journal);
 912                 mutex_unlock(&c->sb_lock);
 913         }
 914
 915         if (ret && !new_fs)
 916                 for (i = 0; i < nr_got; i++)
 917                         bch2_trans_run(c,
 918                                 bch2_trans_mark_metadata_bucket(&trans, ca,
 919                                                 bu[i], BCH_DATA_free, 0));
 920 err_free:
 921         if (!new_fs)
 922                 for (i = 0; i < nr_got; i++)
 923                         bch2_open_bucket_put(c, ob[i]);
 924
 925         kfree(new_bucket_seq);
 926         kfree(new_buckets);
 927         kfree(ob);
 928         kfree(bu);
 929         return ret;
 930 }
 931
 932 /*
 933  * Allocate more journal space at runtime - not currently making use if it, but
 934  * the code works:
 935  */
 936 int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 937                                 unsigned nr)
 938 {
 939         struct journal_device *ja = &ca->journal;
 940         struct closure cl;
 941         int ret = 0;
 942
 943         closure_init_stack(&cl);
 944
 945         down_write(&c->state_lock);
 946
 947         /* don't handle reducing nr of buckets yet: */
 948         if (nr < ja->nr)
 949                 goto unlock;
 950
 951         while (ja->nr < nr) {
 952                 struct disk_reservation disk_res = { 0, 0 };
 953
 954                 /*
 955                  * note: journal buckets aren't really counted as _sectors_ used yet, so
 956                  * we don't need the disk reservation to avoid the BUG_ON() in buckets.c
 957                  * when space used goes up without a reservation - but we do need the
 958                  * reservation to ensure we'll actually be able to allocate:
 959                  *
 960                  * XXX: that's not right, disk reservations only ensure a
 961                  * filesystem-wide allocation will succeed, this is a device
 962                  * specific allocation - we can hang here:
 963                  */
 964
 965                 ret = bch2_disk_reservation_get(c, &disk_res,
 966                                                 bucket_to_sector(ca, nr - ja->nr), 1, 0);
 967                 if (ret)
 968                         break;
 969
 970                 ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl);
 971
 972                 bch2_disk_reservation_put(c, &disk_res);
 973
 974                 closure_sync(&cl);
 975
 976                 if (ret && ret != -BCH_ERR_bucket_alloc_blocked)
 977                         break;
 978         }
 979
 980         if (ret)
 981                 bch_err(c, "%s: err %s", __func__, bch2_err_str(ret));
 982 unlock:
 983         up_write(&c->state_lock);
 984         return ret;
 985 }
 986
 987 int bch2_dev_journal_alloc(struct bch_dev *ca)
 988 {
 989         unsigned nr;
 990
 991         if (dynamic_fault("bcachefs:add:journal_alloc"))
 992                 return -BCH_ERR_ENOMEM_set_nr_journal_buckets;
 993
 994         /* 1/128th of the device by default: */
 995         nr = ca->mi.nbuckets >> 7;
 996
 997         /*
 998          * clamp journal size to 8192 buckets or 8GB (in sectors), whichever
 999          * is smaller:
1000          */
1001         nr = clamp_t(unsigned, nr,
1002                      BCH_JOURNAL_BUCKETS_MIN,
1003                      min(1 << 13,
1004                          (1 << 24) / ca->mi.bucket_size));
1005
1006         return __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
1007 }
1008
1009 /* startup/shutdown: */
1010
1011 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
1012 {
1013         bool ret = false;
1014         u64 seq;
1015
1016         spin_lock(&j->lock);
1017         for (seq = journal_last_unwritten_seq(j);
1018              seq <= journal_cur_seq(j) && !ret;
1019              seq++) {
1020                 struct journal_buf *buf = journal_seq_to_buf(j, seq);
1021
1022                 if (bch2_bkey_has_device_c(bkey_i_to_s_c(&buf->key), dev_idx))
1023                         ret = true;
1024         }
1025         spin_unlock(&j->lock);
1026
1027         return ret;
1028 }
1029
1030 void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca)
1031 {
1032         wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx));
1033 }
1034
1035 void bch2_fs_journal_stop(struct journal *j)
1036 {
1037         bch2_journal_reclaim_stop(j);
1038         bch2_journal_flush_all_pins(j);
1039
1040         wait_event(j->wait, journal_entry_close(j));
1041
1042         /*
1043          * Always write a new journal entry, to make sure the clock hands are up
1044          * to date (and match the superblock)
1045          */
1046         bch2_journal_meta(j);
1047
1048         journal_quiesce(j);
1049
1050         BUG_ON(!bch2_journal_error(j) &&
1051                test_bit(JOURNAL_REPLAY_DONE, &j->flags) &&
1052                j->last_empty_seq != journal_cur_seq(j));
1053
1054         cancel_delayed_work_sync(&j->write_work);
1055 }
1056
1057 int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
1058 {
1059         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1060         struct journal_entry_pin_list *p;
1061         struct journal_replay *i, **_i;
1062         struct genradix_iter iter;
1063         bool had_entries = false;
1064         unsigned ptr;
1065         u64 last_seq = cur_seq, nr, seq;
1066
1067         genradix_for_each_reverse(&c->journal_entries, iter, _i) {
1068                 i = *_i;
1069
1070                 if (!i || i->ignore)
1071                         continue;
1072
1073                 last_seq = le64_to_cpu(i->j.last_seq);
1074                 break;
1075         }
1076
1077         nr = cur_seq - last_seq;
1078
1079         if (nr + 1 > j->pin.size) {
1080                 free_fifo(&j->pin);
1081                 init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL);
1082                 if (!j->pin.data) {
1083                         bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
1084                         return -BCH_ERR_ENOMEM_journal_pin_fifo;
1085                 }
1086         }
1087
1088         j->replay_journal_seq   = last_seq;
1089         j->replay_journal_seq_end = cur_seq;
1090         j->last_seq_ondisk      = last_seq;
1091         j->flushed_seq_ondisk   = cur_seq - 1;
1092         j->seq_ondisk           = cur_seq - 1;
1093         j->pin.front            = last_seq;
1094         j->pin.back             = cur_seq;
1095         atomic64_set(&j->seq, cur_seq - 1);
1096
1097         fifo_for_each_entry_ptr(p, &j->pin, seq)
1098                 journal_pin_list_init(p, 1);
1099
1100         genradix_for_each(&c->journal_entries, iter, _i) {
1101                 i = *_i;
1102
1103                 if (!i || i->ignore)
1104                         continue;
1105
1106                 seq = le64_to_cpu(i->j.seq);
1107                 BUG_ON(seq >= cur_seq);
1108
1109                 if (seq < last_seq)
1110                         continue;
1111
1112                 if (journal_entry_empty(&i->j))
1113                         j->last_empty_seq = le64_to_cpu(i->j.seq);
1114
1115                 p = journal_seq_pin(j, seq);
1116
1117                 p->devs.nr = 0;
1118                 for (ptr = 0; ptr < i->nr_ptrs; ptr++)
1119                         bch2_dev_list_add_dev(&p->devs, i->ptrs[ptr].dev);
1120
1121                 had_entries = true;
1122         }
1123
1124         if (!had_entries)
1125                 j->last_empty_seq = cur_seq;
1126
1127         spin_lock(&j->lock);
1128
1129         set_bit(JOURNAL_STARTED, &j->flags);
1130         j->last_flush_write = jiffies;
1131
1132         j->reservations.idx = j->reservations.unwritten_idx = journal_cur_seq(j);
1133         j->reservations.unwritten_idx++;
1134
1135         c->last_bucket_seq_cleanup = journal_cur_seq(j);
1136
1137         bch2_journal_space_available(j);
1138         spin_unlock(&j->lock);
1139
1140         return bch2_journal_reclaim_start(j);
1141 }
1142
1143 /* init/exit: */
1144
1145 void bch2_dev_journal_exit(struct bch_dev *ca)
1146 {
1147         kfree(ca->journal.bio);
1148         kfree(ca->journal.buckets);
1149         kfree(ca->journal.bucket_seq);
1150
1151         ca->journal.bio         = NULL;
1152         ca->journal.buckets     = NULL;
1153         ca->journal.bucket_seq  = NULL;
1154 }
1155
1156 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
1157 {
1158         struct journal_device *ja = &ca->journal;
1159         struct bch_sb_field_journal *journal_buckets =
1160                 bch2_sb_get_journal(sb);
1161         struct bch_sb_field_journal_v2 *journal_buckets_v2 =
1162                 bch2_sb_get_journal_v2(sb);
1163         unsigned i, nr_bvecs;
1164
1165         ja->nr = 0;
1166
1167         if (journal_buckets_v2) {
1168                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1169
1170                 for (i = 0; i < nr; i++)
1171                         ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr);
1172         } else if (journal_buckets) {
1173                 ja->nr = bch2_nr_journal_buckets(journal_buckets);
1174         }
1175
1176         ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1177         if (!ja->bucket_seq)
1178                 return -BCH_ERR_ENOMEM_dev_journal_init;
1179
1180         nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
1181
1182         ca->journal.bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1183         if (!ca->journal.bio)
1184                 return -BCH_ERR_ENOMEM_dev_journal_init;
1185
1186         bio_init(ca->journal.bio, NULL, ca->journal.bio->bi_inline_vecs, nr_bvecs, 0);
1187
1188         ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1189         if (!ja->buckets)
1190                 return -BCH_ERR_ENOMEM_dev_journal_init;
1191
1192         if (journal_buckets_v2) {
1193                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1194                 unsigned j, dst = 0;
1195
1196                 for (i = 0; i < nr; i++)
1197                         for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
1198                                 ja->buckets[dst++] =
1199                                         le64_to_cpu(journal_buckets_v2->d[i].start) + j;
1200         } else if (journal_buckets) {
1201                 for (i = 0; i < ja->nr; i++)
1202                         ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
1203         }
1204
1205         return 0;
1206 }
1207
1208 void bch2_fs_journal_exit(struct journal *j)
1209 {
1210         unsigned i;
1211
1212         darray_exit(&j->early_journal_entries);
1213
1214         for (i = 0; i < ARRAY_SIZE(j->buf); i++)
1215                 kvpfree(j->buf[i].data, j->buf[i].buf_size);
1216         free_fifo(&j->pin);
1217 }
1218
1219 int bch2_fs_journal_init(struct journal *j)
1220 {
1221         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1222         static struct lock_class_key res_key;
1223         unsigned i;
1224         int ret = 0;
1225
1226         pr_verbose_init(c->opts, "");
1227
1228         spin_lock_init(&j->lock);
1229         spin_lock_init(&j->err_lock);
1230         init_waitqueue_head(&j->wait);
1231         INIT_DELAYED_WORK(&j->write_work, journal_write_work);
1232         init_waitqueue_head(&j->reclaim_wait);
1233         init_waitqueue_head(&j->pin_flush_wait);
1234         mutex_init(&j->reclaim_lock);
1235         mutex_init(&j->discard_lock);
1236
1237         lockdep_init_map(&j->res_map, "journal res", &res_key, 0);
1238
1239         atomic64_set(&j->reservations.counter,
1240                 ((union journal_res_state)
1241                  { .cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL }).v);
1242
1243         if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL))) {
1244                 ret = -BCH_ERR_ENOMEM_journal_pin_fifo;
1245                 goto out;
1246         }
1247
1248         for (i = 0; i < ARRAY_SIZE(j->buf); i++) {
1249                 j->buf[i].buf_size = JOURNAL_ENTRY_SIZE_MIN;
1250                 j->buf[i].data = kvpmalloc(j->buf[i].buf_size, GFP_KERNEL);
1251                 if (!j->buf[i].data) {
1252                         ret = -BCH_ERR_ENOMEM_journal_buf;
1253                         goto out;
1254                 }
1255         }
1256
1257         j->pin.front = j->pin.back = 1;
1258 out:
1259         pr_verbose_init(c->opts, "ret %i", ret);
1260         return ret;
1261 }
1262
1263 /* debug: */
1264
1265 void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1266 {
1267         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1268         union journal_res_state s;
1269         struct bch_dev *ca;
1270         unsigned long now = jiffies;
1271         u64 seq;
1272         unsigned i;
1273
1274         if (!out->nr_tabstops)
1275                 printbuf_tabstop_push(out, 24);
1276         out->atomic++;
1277
1278         rcu_read_lock();
1279         s = READ_ONCE(j->reservations);
1280
1281         prt_printf(out, "dirty journal entries:\t%llu/%llu\n",  fifo_used(&j->pin), j->pin.size);
1282         prt_printf(out, "seq:\t\t\t%llu\n",                     journal_cur_seq(j));
1283         prt_printf(out, "seq_ondisk:\t\t%llu\n",                j->seq_ondisk);
1284         prt_printf(out, "last_seq:\t\t%llu\n",          journal_last_seq(j));
1285         prt_printf(out, "last_seq_ondisk:\t%llu\n",             j->last_seq_ondisk);
1286         prt_printf(out, "flushed_seq_ondisk:\t%llu\n",  j->flushed_seq_ondisk);
1287         prt_printf(out, "prereserved:\t\t%u/%u\n",              j->prereserved.reserved, j->prereserved.remaining);
1288         prt_printf(out, "watermark:\t\t%s\n",           bch2_journal_watermarks[j->watermark]);
1289         prt_printf(out, "each entry reserved:\t%u\n",   j->entry_u64s_reserved);
1290         prt_printf(out, "nr flush writes:\t%llu\n",             j->nr_flush_writes);
1291         prt_printf(out, "nr noflush writes:\t%llu\n",   j->nr_noflush_writes);
1292         prt_printf(out, "nr direct reclaim:\t%llu\n",   j->nr_direct_reclaim);
1293         prt_printf(out, "nr background reclaim:\t%llu\n",       j->nr_background_reclaim);
1294         prt_printf(out, "reclaim kicked:\t\t%u\n",              j->reclaim_kicked);
1295         prt_printf(out, "reclaim runs in:\t%u ms\n",    time_after(j->next_reclaim, now)
1296                ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0);
1297         prt_printf(out, "current entry sectors:\t%u\n", j->cur_entry_sectors);
1298         prt_printf(out, "current entry error:\t%s\n",   bch2_journal_errors[j->cur_entry_error]);
1299         prt_printf(out, "current entry:\t\t");
1300
1301         switch (s.cur_entry_offset) {
1302         case JOURNAL_ENTRY_ERROR_VAL:
1303                 prt_printf(out, "error");
1304                 break;
1305         case JOURNAL_ENTRY_CLOSED_VAL:
1306                 prt_printf(out, "closed");
1307                 break;
1308         default:
1309                 prt_printf(out, "%u/%u", s.cur_entry_offset, j->cur_entry_u64s);
1310                 break;
1311         }
1312
1313         prt_newline(out);
1314
1315         for (seq = journal_cur_seq(j);
1316              seq >= journal_last_unwritten_seq(j);
1317              --seq) {
1318                 i = seq & JOURNAL_BUF_MASK;
1319
1320                 prt_printf(out, "unwritten entry:");
1321                 prt_tab(out);
1322                 prt_printf(out, "%llu", seq);
1323                 prt_newline(out);
1324                 printbuf_indent_add(out, 2);
1325
1326                 prt_printf(out, "refcount:");
1327                 prt_tab(out);
1328                 prt_printf(out, "%u", journal_state_count(s, i));
1329                 prt_newline(out);
1330
1331                 prt_printf(out, "sectors:");
1332                 prt_tab(out);
1333                 prt_printf(out, "%u", j->buf[i].sectors);
1334                 prt_newline(out);
1335
1336                 prt_printf(out, "expires");
1337                 prt_tab(out);
1338                 prt_printf(out, "%li jiffies", j->buf[i].expires - jiffies);
1339                 prt_newline(out);
1340
1341                 printbuf_indent_sub(out, 2);
1342         }
1343
1344         prt_printf(out,
1345                "replay done:\t\t%i\n",
1346                test_bit(JOURNAL_REPLAY_DONE,    &j->flags));
1347
1348         prt_printf(out, "space:\n");
1349         prt_printf(out, "\tdiscarded\t%u:%u\n",
1350                j->space[journal_space_discarded].next_entry,
1351                j->space[journal_space_discarded].total);
1352         prt_printf(out, "\tclean ondisk\t%u:%u\n",
1353                j->space[journal_space_clean_ondisk].next_entry,
1354                j->space[journal_space_clean_ondisk].total);
1355         prt_printf(out, "\tclean\t\t%u:%u\n",
1356                j->space[journal_space_clean].next_entry,
1357                j->space[journal_space_clean].total);
1358         prt_printf(out, "\ttotal\t\t%u:%u\n",
1359                j->space[journal_space_total].next_entry,
1360                j->space[journal_space_total].total);
1361
1362         for_each_member_device_rcu(ca, c, i,
1363                                    &c->rw_devs[BCH_DATA_journal]) {
1364                 struct journal_device *ja = &ca->journal;
1365
1366                 if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d))
1367                         continue;
1368
1369                 if (!ja->nr)
1370                         continue;
1371
1372                 prt_printf(out, "dev %u:\n",            i);
1373                 prt_printf(out, "\tnr\t\t%u\n",         ja->nr);
1374                 prt_printf(out, "\tbucket size\t%u\n",  ca->mi.bucket_size);
1375                 prt_printf(out, "\tavailable\t%u:%u\n", bch2_journal_dev_buckets_available(j, ja, journal_space_discarded), ja->sectors_free);
1376                 prt_printf(out, "\tdiscard_idx\t%u\n",  ja->discard_idx);
1377                 prt_printf(out, "\tdirty_ondisk\t%u (seq %llu)\n", ja->dirty_idx_ondisk,        ja->bucket_seq[ja->dirty_idx_ondisk]);
1378                 prt_printf(out, "\tdirty_idx\t%u (seq %llu)\n", ja->dirty_idx,          ja->bucket_seq[ja->dirty_idx]);
1379                 prt_printf(out, "\tcur_idx\t\t%u (seq %llu)\n", ja->cur_idx,            ja->bucket_seq[ja->cur_idx]);
1380         }
1381
1382         rcu_read_unlock();
1383
1384         --out->atomic;
1385 }
1386
1387 void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1388 {
1389         spin_lock(&j->lock);
1390         __bch2_journal_debug_to_text(out, j);
1391         spin_unlock(&j->lock);
1392 }
1393
1394 bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq)
1395 {
1396         struct journal_entry_pin_list *pin_list;
1397         struct journal_entry_pin *pin;
1398         unsigned i;
1399
1400         spin_lock(&j->lock);
1401         *seq = max(*seq, j->pin.front);
1402
1403         if (*seq >= j->pin.back) {
1404                 spin_unlock(&j->lock);
1405                 return true;
1406         }
1407
1408         out->atomic++;
1409
1410         pin_list = journal_seq_pin(j, *seq);
1411
1412         prt_printf(out, "%llu: count %u", *seq, atomic_read(&pin_list->count));
1413         prt_newline(out);
1414         printbuf_indent_add(out, 2);
1415
1416         for (i = 0; i < ARRAY_SIZE(pin_list->list); i++)
1417                 list_for_each_entry(pin, &pin_list->list[i], list) {
1418                         prt_printf(out, "\t%px %ps", pin, pin->flush);
1419                         prt_newline(out);
1420                 }
1421
1422         if (!list_empty(&pin_list->flushed)) {
1423                 prt_printf(out, "flushed:");
1424                 prt_newline(out);
1425         }
1426
1427         list_for_each_entry(pin, &pin_list->flushed, list) {
1428                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1429                 prt_newline(out);
1430         }
1431
1432         printbuf_indent_sub(out, 2);
1433
1434         --out->atomic;
1435         spin_unlock(&j->lock);
1436
1437         return false;
1438 }
1439
1440 void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
1441 {
1442         u64 seq = 0;
1443
1444         while (!bch2_journal_seq_pins_to_text(out, j, &seq))
1445                 seq++;
1446 }