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  * returns:
 204  * 0:           success
 205  * -ENOSPC:     journal currently full, must invoke reclaim
 206  * -EAGAIN:     journal blocked, must wait
 207  * -EROFS:      insufficient rw devices or journal error
 208  */
 209 static int journal_entry_open(struct journal *j)
 210 {
 211         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 212         struct journal_buf *buf = j->buf +
 213                 ((journal_cur_seq(j) + 1) & JOURNAL_BUF_MASK);
 214         union journal_res_state old, new;
 215         int u64s;
 216         u64 v;
 217
 218         lockdep_assert_held(&j->lock);
 219         BUG_ON(journal_entry_is_open(j));
 220         BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
 221
 222         if (j->blocked)
 223                 return JOURNAL_ERR_blocked;
 224
 225         if (j->cur_entry_error)
 226                 return j->cur_entry_error;
 227
 228         if (bch2_journal_error(j))
 229                 return JOURNAL_ERR_insufficient_devices; /* -EROFS */
 230
 231         if (!fifo_free(&j->pin))
 232                 return JOURNAL_ERR_journal_pin_full;
 233
 234         if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf) - 1)
 235                 return JOURNAL_ERR_max_in_flight;
 236
 237         BUG_ON(!j->cur_entry_sectors);
 238
 239         buf->expires            =
 240                 (journal_cur_seq(j) == j->flushed_seq_ondisk
 241                  ? jiffies
 242                  : j->last_flush_write) +
 243                 msecs_to_jiffies(c->opts.journal_flush_delay);
 244
 245         buf->u64s_reserved      = j->entry_u64s_reserved;
 246         buf->disk_sectors       = j->cur_entry_sectors;
 247         buf->sectors            = min(buf->disk_sectors, buf->buf_size >> 9);
 248
 249         u64s = (int) (buf->sectors << 9) / sizeof(u64) -
 250                 journal_entry_overhead(j);
 251         u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1);
 252
 253         if (u64s <= 0)
 254                 return JOURNAL_ERR_journal_full;
 255
 256         if (fifo_empty(&j->pin) && j->reclaim_thread)
 257                 wake_up_process(j->reclaim_thread);
 258
 259         /*
 260          * The fifo_push() needs to happen at the same time as j->seq is
 261          * incremented for journal_last_seq() to be calculated correctly
 262          */
 263         atomic64_inc(&j->seq);
 264         journal_pin_list_init(fifo_push_ref(&j->pin), 1);
 265
 266         BUG_ON(j->buf + (journal_cur_seq(j) & JOURNAL_BUF_MASK) != buf);
 267
 268         bkey_extent_init(&buf->key);
 269         buf->noflush    = false;
 270         buf->must_flush = false;
 271         buf->separate_flush = false;
 272         buf->flush_time = 0;
 273
 274         memset(buf->data, 0, sizeof(*buf->data));
 275         buf->data->seq  = cpu_to_le64(journal_cur_seq(j));
 276         buf->data->u64s = 0;
 277
 278         /*
 279          * Must be set before marking the journal entry as open:
 280          */
 281         j->cur_entry_u64s = u64s;
 282
 283         v = atomic64_read(&j->reservations.counter);
 284         do {
 285                 old.v = new.v = v;
 286
 287                 BUG_ON(old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL);
 288
 289                 new.idx++;
 290                 BUG_ON(journal_state_count(new, new.idx));
 291                 BUG_ON(new.idx != (journal_cur_seq(j) & JOURNAL_BUF_MASK));
 292
 293                 journal_state_inc(&new);
 294                 new.cur_entry_offset = 0;
 295         } while ((v = atomic64_cmpxchg(&j->reservations.counter,
 296                                        old.v, new.v)) != old.v);
 297
 298         if (j->res_get_blocked_start)
 299                 bch2_time_stats_update(j->blocked_time,
 300                                        j->res_get_blocked_start);
 301         j->res_get_blocked_start = 0;
 302
 303         mod_delayed_work(c->io_complete_wq,
 304                          &j->write_work,
 305                          msecs_to_jiffies(c->opts.journal_flush_delay));
 306         journal_wake(j);
 307         return 0;
 308 }
 309
 310 static bool journal_quiesced(struct journal *j)
 311 {
 312         bool ret = atomic64_read(&j->seq) == j->seq_ondisk;
 313
 314         if (!ret)
 315                 journal_entry_close(j);
 316         return ret;
 317 }
 318
 319 static void journal_quiesce(struct journal *j)
 320 {
 321         wait_event(j->wait, journal_quiesced(j));
 322 }
 323
 324 static void journal_write_work(struct work_struct *work)
 325 {
 326         struct journal *j = container_of(work, struct journal, write_work.work);
 327         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 328         long delta;
 329
 330         spin_lock(&j->lock);
 331         if (!__journal_entry_is_open(j->reservations))
 332                 goto unlock;
 333
 334         delta = journal_cur_buf(j)->expires - jiffies;
 335
 336         if (delta > 0)
 337                 mod_delayed_work(c->io_complete_wq, &j->write_work, delta);
 338         else
 339                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 340 unlock:
 341         spin_unlock(&j->lock);
 342 }
 343
 344 static int __journal_res_get(struct journal *j, struct journal_res *res,
 345                              unsigned flags)
 346 {
 347         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 348         struct journal_buf *buf;
 349         bool can_discard;
 350         int ret;
 351 retry:
 352         if (journal_res_get_fast(j, res, flags))
 353                 return 0;
 354
 355         if (bch2_journal_error(j))
 356                 return -EROFS;
 357
 358         spin_lock(&j->lock);
 359
 360         /*
 361          * Recheck after taking the lock, so we don't race with another thread
 362          * that just did journal_entry_open() and call journal_entry_close()
 363          * unnecessarily
 364          */
 365         if (journal_res_get_fast(j, res, flags)) {
 366                 spin_unlock(&j->lock);
 367                 return 0;
 368         }
 369
 370         if ((flags & JOURNAL_WATERMARK_MASK) < j->watermark) {
 371                 /*
 372                  * Don't want to close current journal entry, just need to
 373                  * invoke reclaim:
 374                  */
 375                 ret = JOURNAL_ERR_journal_full;
 376                 goto unlock;
 377         }
 378
 379         /*
 380          * If we couldn't get a reservation because the current buf filled up,
 381          * and we had room for a bigger entry on disk, signal that we want to
 382          * realloc the journal bufs:
 383          */
 384         buf = journal_cur_buf(j);
 385         if (journal_entry_is_open(j) &&
 386             buf->buf_size >> 9 < buf->disk_sectors &&
 387             buf->buf_size < JOURNAL_ENTRY_SIZE_MAX)
 388                 j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1);
 389
 390         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 391         ret = journal_entry_open(j);
 392
 393         if (ret == JOURNAL_ERR_max_in_flight)
 394                 trace_and_count(c, journal_entry_full, c);
 395 unlock:
 396         if ((ret && ret != JOURNAL_ERR_insufficient_devices) &&
 397             !j->res_get_blocked_start) {
 398                 j->res_get_blocked_start = local_clock() ?: 1;
 399                 trace_and_count(c, journal_full, c);
 400         }
 401
 402         can_discard = j->can_discard;
 403         spin_unlock(&j->lock);
 404
 405         if (!ret)
 406                 goto retry;
 407
 408         if ((ret == JOURNAL_ERR_journal_full ||
 409              ret == JOURNAL_ERR_journal_pin_full) &&
 410             !can_discard &&
 411             !nr_unwritten_journal_entries(j) &&
 412             (flags & JOURNAL_WATERMARK_MASK) == JOURNAL_WATERMARK_reserved) {
 413                 struct printbuf buf = PRINTBUF;
 414
 415                 bch_err(c, "Journal stuck! Hava a pre-reservation but journal full (ret %s)",
 416                         bch2_journal_errors[ret]);
 417
 418                 bch2_journal_debug_to_text(&buf, j);
 419                 bch_err(c, "%s", buf.buf);
 420
 421                 printbuf_reset(&buf);
 422                 bch2_journal_pins_to_text(&buf, j);
 423                 bch_err(c, "Journal pins:\n%s", buf.buf);
 424
 425                 printbuf_exit(&buf);
 426                 bch2_fatal_error(c);
 427                 dump_stack();
 428         }
 429
 430         /*
 431          * Journal is full - can't rely on reclaim from work item due to
 432          * freezing:
 433          */
 434         if ((ret == JOURNAL_ERR_journal_full ||
 435              ret == JOURNAL_ERR_journal_pin_full) &&
 436             !(flags & JOURNAL_RES_GET_NONBLOCK)) {
 437                 if (can_discard) {
 438                         bch2_journal_do_discards(j);
 439                         goto retry;
 440                 }
 441
 442                 if (mutex_trylock(&j->reclaim_lock)) {
 443                         bch2_journal_reclaim(j);
 444                         mutex_unlock(&j->reclaim_lock);
 445                 }
 446         }
 447
 448         return ret == JOURNAL_ERR_insufficient_devices ? -EROFS : -EAGAIN;
 449 }
 450
 451 /*
 452  * Essentially the entry function to the journaling code. When bcachefs is doing
 453  * a btree insert, it calls this function to get the current journal write.
 454  * Journal write is the structure used set up journal writes. The calling
 455  * function will then add its keys to the structure, queuing them for the next
 456  * write.
 457  *
 458  * To ensure forward progress, the current task must not be holding any
 459  * btree node write locks.
 460  */
 461 int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res,
 462                                   unsigned flags)
 463 {
 464         int ret;
 465
 466         closure_wait_event(&j->async_wait,
 467                    (ret = __journal_res_get(j, res, flags)) != -EAGAIN ||
 468                    (flags & JOURNAL_RES_GET_NONBLOCK));
 469         return ret;
 470 }
 471
 472 /* journal_preres: */
 473
 474 static bool journal_preres_available(struct journal *j,
 475                                      struct journal_preres *res,
 476                                      unsigned new_u64s,
 477                                      unsigned flags)
 478 {
 479         bool ret = bch2_journal_preres_get_fast(j, res, new_u64s, flags, true);
 480
 481         if (!ret && mutex_trylock(&j->reclaim_lock)) {
 482                 bch2_journal_reclaim(j);
 483                 mutex_unlock(&j->reclaim_lock);
 484         }
 485
 486         return ret;
 487 }
 488
 489 int __bch2_journal_preres_get(struct journal *j,
 490                               struct journal_preres *res,
 491                               unsigned new_u64s,
 492                               unsigned flags)
 493 {
 494         int ret;
 495
 496         closure_wait_event(&j->preres_wait,
 497                    (ret = bch2_journal_error(j)) ||
 498                    journal_preres_available(j, res, new_u64s, flags));
 499         return ret;
 500 }
 501
 502 /* journal_entry_res: */
 503
 504 void bch2_journal_entry_res_resize(struct journal *j,
 505                                    struct journal_entry_res *res,
 506                                    unsigned new_u64s)
 507 {
 508         union journal_res_state state;
 509         int d = new_u64s - res->u64s;
 510
 511         spin_lock(&j->lock);
 512
 513         j->entry_u64s_reserved += d;
 514         if (d <= 0)
 515                 goto out;
 516
 517         j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d);
 518         smp_mb();
 519         state = READ_ONCE(j->reservations);
 520
 521         if (state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL &&
 522             state.cur_entry_offset > j->cur_entry_u64s) {
 523                 j->cur_entry_u64s += d;
 524                 /*
 525                  * Not enough room in current journal entry, have to flush it:
 526                  */
 527                 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 528         } else {
 529                 journal_cur_buf(j)->u64s_reserved += d;
 530         }
 531 out:
 532         spin_unlock(&j->lock);
 533         res->u64s += d;
 534 }
 535
 536 /* journal flushing: */
 537
 538 /**
 539  * bch2_journal_flush_seq_async - wait for a journal entry to be written
 540  *
 541  * like bch2_journal_wait_on_seq, except that it triggers a write immediately if
 542  * necessary
 543  */
 544 int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 545                                  struct closure *parent)
 546 {
 547         struct journal_buf *buf;
 548         int ret = 0;
 549
 550         if (seq <= j->flushed_seq_ondisk)
 551                 return 1;
 552
 553         spin_lock(&j->lock);
 554
 555         if (WARN_ONCE(seq > journal_cur_seq(j),
 556                       "requested to flush journal seq %llu, but currently at %llu",
 557                       seq, journal_cur_seq(j)))
 558                 goto out;
 559
 560         /* Recheck under lock: */
 561         if (j->err_seq && seq >= j->err_seq) {
 562                 ret = -EIO;
 563                 goto out;
 564         }
 565
 566         if (seq <= j->flushed_seq_ondisk) {
 567                 ret = 1;
 568                 goto out;
 569         }
 570
 571         /* if seq was written, but not flushed - flush a newer one instead */
 572         seq = max(seq, journal_last_unwritten_seq(j));
 573
 574 recheck_need_open:
 575         if (seq > journal_cur_seq(j)) {
 576                 struct journal_res res = { 0 };
 577
 578                 if (journal_entry_is_open(j))
 579                         __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL);
 580
 581                 spin_unlock(&j->lock);
 582
 583                 ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 584                 if (ret)
 585                         return ret;
 586
 587                 seq = res.seq;
 588                 buf = j->buf + (seq & JOURNAL_BUF_MASK);
 589                 buf->must_flush = true;
 590
 591                 if (!buf->flush_time) {
 592                         buf->flush_time = local_clock() ?: 1;
 593                         buf->expires = jiffies;
 594                 }
 595
 596                 if (parent && !closure_wait(&buf->wait, parent))
 597                         BUG();
 598
 599                 bch2_journal_res_put(j, &res);
 600
 601                 spin_lock(&j->lock);
 602                 goto want_write;
 603         }
 604
 605         /*
 606          * if write was kicked off without a flush, flush the next sequence
 607          * number instead
 608          */
 609         buf = journal_seq_to_buf(j, seq);
 610         if (buf->noflush) {
 611                 seq++;
 612                 goto recheck_need_open;
 613         }
 614
 615         buf->must_flush = true;
 616
 617         if (parent && !closure_wait(&buf->wait, parent))
 618                 BUG();
 619 want_write:
 620         if (seq == journal_cur_seq(j))
 621                 journal_entry_want_write(j);
 622 out:
 623         spin_unlock(&j->lock);
 624         return ret;
 625 }
 626
 627 int bch2_journal_flush_seq(struct journal *j, u64 seq)
 628 {
 629         u64 start_time = local_clock();
 630         int ret, ret2;
 631
 632         /*
 633          * Don't update time_stats when @seq is already flushed:
 634          */
 635         if (seq <= j->flushed_seq_ondisk)
 636                 return 0;
 637
 638         ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL)));
 639
 640         if (!ret)
 641                 bch2_time_stats_update(j->flush_seq_time, start_time);
 642
 643         return ret ?: ret2 < 0 ? ret2 : 0;
 644 }
 645
 646 /*
 647  * bch2_journal_flush_async - if there is an open journal entry, or a journal
 648  * still being written, write it and wait for the write to complete
 649  */
 650 void bch2_journal_flush_async(struct journal *j, struct closure *parent)
 651 {
 652         bch2_journal_flush_seq_async(j, atomic64_read(&j->seq), parent);
 653 }
 654
 655 int bch2_journal_flush(struct journal *j)
 656 {
 657         return bch2_journal_flush_seq(j, atomic64_read(&j->seq));
 658 }
 659
 660 /*
 661  * bch2_journal_noflush_seq - tell the journal not to issue any flushes before
 662  * @seq
 663  */
 664 bool bch2_journal_noflush_seq(struct journal *j, u64 seq)
 665 {
 666         struct bch_fs *c = container_of(j, struct bch_fs, journal);
 667         u64 unwritten_seq;
 668         bool ret = false;
 669
 670         if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush)))
 671                 return false;
 672
 673         if (seq <= c->journal.flushed_seq_ondisk)
 674                 return false;
 675
 676         spin_lock(&j->lock);
 677         if (seq <= c->journal.flushed_seq_ondisk)
 678                 goto out;
 679
 680         for (unwritten_seq = journal_last_unwritten_seq(j);
 681              unwritten_seq < seq;
 682              unwritten_seq++) {
 683                 struct journal_buf *buf = journal_seq_to_buf(j, unwritten_seq);
 684
 685                 /* journal write is already in flight, and was a flush write: */
 686                 if (unwritten_seq == journal_last_unwritten_seq(j) && !buf->noflush)
 687                         goto out;
 688
 689                 buf->noflush = true;
 690         }
 691
 692         ret = true;
 693 out:
 694         spin_unlock(&j->lock);
 695         return ret;
 696 }
 697
 698 int bch2_journal_meta(struct journal *j)
 699 {
 700         struct journal_buf *buf;
 701         struct journal_res res;
 702         int ret;
 703
 704         memset(&res, 0, sizeof(res));
 705
 706         ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0);
 707         if (ret)
 708                 return ret;
 709
 710         buf = j->buf + (res.seq & JOURNAL_BUF_MASK);
 711         buf->must_flush = true;
 712
 713         if (!buf->flush_time) {
 714                 buf->flush_time = local_clock() ?: 1;
 715                 buf->expires = jiffies;
 716         }
 717
 718         bch2_journal_res_put(j, &res);
 719
 720         return bch2_journal_flush_seq(j, res.seq);
 721 }
 722
 723 int bch2_journal_log_msg(struct journal *j, const char *fmt, ...)
 724 {
 725         struct jset_entry_log *entry;
 726         struct journal_res res = { 0 };
 727         unsigned msglen, u64s;
 728         va_list args;
 729         int ret;
 730
 731         va_start(args, fmt);
 732         msglen = vsnprintf(NULL, 0, fmt, args) + 1;
 733         va_end(args);
 734
 735         u64s = jset_u64s(DIV_ROUND_UP(msglen, sizeof(u64)));
 736
 737         ret = bch2_journal_res_get(j, &res, u64s, 0);
 738         if (ret)
 739                 return ret;
 740
 741         entry = container_of(journal_res_entry(j, &res),
 742                              struct jset_entry_log, entry);
 743         memset(entry, 0, u64s * sizeof(u64));
 744         entry->entry.type = BCH_JSET_ENTRY_log;
 745         entry->entry.u64s = u64s - 1;
 746
 747         va_start(args, fmt);
 748         vsnprintf(entry->d, INT_MAX, fmt, args);
 749         va_end(args);
 750
 751         bch2_journal_res_put(j, &res);
 752
 753         return bch2_journal_flush_seq(j, res.seq);
 754 }
 755
 756 /* block/unlock the journal: */
 757
 758 void bch2_journal_unblock(struct journal *j)
 759 {
 760         spin_lock(&j->lock);
 761         j->blocked--;
 762         spin_unlock(&j->lock);
 763
 764         journal_wake(j);
 765 }
 766
 767 void bch2_journal_block(struct journal *j)
 768 {
 769         spin_lock(&j->lock);
 770         j->blocked++;
 771         spin_unlock(&j->lock);
 772
 773         journal_quiesce(j);
 774 }
 775
 776 /* allocate journal on a device: */
 777
 778 static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 779                                          bool new_fs, struct closure *cl)
 780 {
 781         struct bch_fs *c = ca->fs;
 782         struct journal_device *ja = &ca->journal;
 783         u64 *new_bucket_seq = NULL, *new_buckets = NULL;
 784         struct open_bucket **ob = NULL;
 785         long *bu = NULL;
 786         unsigned i, nr_got = 0, nr_want = nr - ja->nr;
 787         unsigned old_nr                 = ja->nr;
 788         unsigned old_discard_idx        = ja->discard_idx;
 789         unsigned old_dirty_idx_ondisk   = ja->dirty_idx_ondisk;
 790         unsigned old_dirty_idx          = ja->dirty_idx;
 791         unsigned old_cur_idx            = ja->cur_idx;
 792         int ret = 0;
 793
 794         if (c) {
 795                 bch2_journal_flush_all_pins(&c->journal);
 796                 bch2_journal_block(&c->journal);
 797         }
 798
 799         bu              = kcalloc(nr_want, sizeof(*bu), GFP_KERNEL);
 800         ob              = kcalloc(nr_want, sizeof(*ob), GFP_KERNEL);
 801         new_buckets     = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 802         new_bucket_seq  = kcalloc(nr, sizeof(u64), GFP_KERNEL);
 803         if (!bu || !ob || !new_buckets || !new_bucket_seq) {
 804                 ret = -ENOMEM;
 805                 goto err_unblock;
 806         }
 807
 808         for (nr_got = 0; nr_got < nr_want; nr_got++) {
 809                 if (new_fs) {
 810                         bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
 811                         if (bu[nr_got] < 0) {
 812                                 ret = -BCH_ERR_ENOSPC_bucket_alloc;
 813                                 break;
 814                         }
 815                 } else {
 816                         ob[nr_got] = bch2_bucket_alloc(c, ca, RESERVE_none,
 817                                                false, cl);
 818                         if (IS_ERR(ob[nr_got])) {
 819                                 ret = cl
 820                                         ? -EAGAIN
 821                                         : -BCH_ERR_ENOSPC_bucket_alloc;
 822                                 break;
 823                         }
 824
 825                         bu[nr_got] = ob[nr_got]->bucket;
 826                 }
 827         }
 828
 829         if (!nr_got)
 830                 goto err_unblock;
 831
 832         /*
 833          * We may be called from the device add path, before the new device has
 834          * actually been added to the running filesystem:
 835          */
 836         if (!new_fs)
 837                 spin_lock(&c->journal.lock);
 838
 839         memcpy(new_buckets,     ja->buckets,    ja->nr * sizeof(u64));
 840         memcpy(new_bucket_seq,  ja->bucket_seq, ja->nr * sizeof(u64));
 841         swap(new_buckets,       ja->buckets);
 842         swap(new_bucket_seq,    ja->bucket_seq);
 843
 844         for (i = 0; i < nr_got; i++) {
 845                 unsigned pos = ja->discard_idx ?: ja->nr;
 846                 long b = bu[i];
 847
 848                 __array_insert_item(ja->buckets,                ja->nr, pos);
 849                 __array_insert_item(ja->bucket_seq,             ja->nr, pos);
 850                 ja->nr++;
 851
 852                 ja->buckets[pos] = b;
 853                 ja->bucket_seq[pos] = 0;
 854
 855                 if (pos <= ja->discard_idx)
 856                         ja->discard_idx = (ja->discard_idx + 1) % ja->nr;
 857                 if (pos <= ja->dirty_idx_ondisk)
 858                         ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + 1) % ja->nr;
 859                 if (pos <= ja->dirty_idx)
 860                         ja->dirty_idx = (ja->dirty_idx + 1) % ja->nr;
 861                 if (pos <= ja->cur_idx)
 862                         ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
 863         }
 864
 865         ret = bch2_journal_buckets_to_sb(c, ca);
 866         if (ret) {
 867                 /* Revert: */
 868                 swap(new_buckets,       ja->buckets);
 869                 swap(new_bucket_seq,    ja->bucket_seq);
 870                 ja->nr                  = old_nr;
 871                 ja->discard_idx         = old_discard_idx;
 872                 ja->dirty_idx_ondisk    = old_dirty_idx_ondisk;
 873                 ja->dirty_idx           = old_dirty_idx;
 874                 ja->cur_idx             = old_cur_idx;
 875         }
 876
 877         if (!new_fs)
 878                 spin_unlock(&c->journal.lock);
 879
 880         if (c)
 881                 bch2_journal_unblock(&c->journal);
 882
 883         if (ret)
 884                 goto err;
 885
 886         if (!new_fs) {
 887                 for (i = 0; i < nr_got; i++) {
 888                         ret = bch2_trans_run(c,
 889                                 bch2_trans_mark_metadata_bucket(&trans, ca,
 890                                                 bu[i], BCH_DATA_journal,
 891                                                 ca->mi.bucket_size));
 892                         if (ret) {
 893                                 bch2_fs_inconsistent(c, "error marking new journal buckets: %i", ret);
 894                                 goto err;
 895                         }
 896                 }
 897         }
 898 err:
 899         if (ob && !new_fs)
 900                 for (i = 0; i < nr_got; i++)
 901                         bch2_open_bucket_put(c, ob[i]);
 902
 903         kfree(new_bucket_seq);
 904         kfree(new_buckets);
 905         kfree(ob);
 906         kfree(bu);
 907
 908         return ret;
 909 err_unblock:
 910         if (c)
 911                 bch2_journal_unblock(&c->journal);
 912         goto err;
 913 }
 914
 915 /*
 916  * Allocate more journal space at runtime - not currently making use if it, but
 917  * the code works:
 918  */
 919 int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca,
 920                                 unsigned nr)
 921 {
 922         struct journal_device *ja = &ca->journal;
 923         struct closure cl;
 924         unsigned current_nr;
 925         int ret = 0;
 926
 927         /* don't handle reducing nr of buckets yet: */
 928         if (nr < ja->nr)
 929                 return 0;
 930
 931         closure_init_stack(&cl);
 932
 933         while (ja->nr != nr && (ret == 0 || ret == -EAGAIN)) {
 934                 struct disk_reservation disk_res = { 0, 0 };
 935
 936                 closure_sync(&cl);
 937
 938                 mutex_lock(&c->sb_lock);
 939                 current_nr = ja->nr;
 940
 941                 /*
 942                  * note: journal buckets aren't really counted as _sectors_ used yet, so
 943                  * we don't need the disk reservation to avoid the BUG_ON() in buckets.c
 944                  * when space used goes up without a reservation - but we do need the
 945                  * reservation to ensure we'll actually be able to allocate:
 946                  */
 947
 948                 ret = bch2_disk_reservation_get(c, &disk_res,
 949                                                 bucket_to_sector(ca, nr - ja->nr), 1, 0);
 950                 if (ret) {
 951                         mutex_unlock(&c->sb_lock);
 952                         return ret;
 953                 }
 954
 955                 ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl);
 956
 957                 bch2_disk_reservation_put(c, &disk_res);
 958
 959                 if (ja->nr != current_nr)
 960                         bch2_write_super(c);
 961                 mutex_unlock(&c->sb_lock);
 962         }
 963
 964         return ret;
 965 }
 966
 967 int bch2_dev_journal_alloc(struct bch_dev *ca)
 968 {
 969         unsigned nr;
 970         int ret;
 971
 972         if (dynamic_fault("bcachefs:add:journal_alloc"))
 973                 return -ENOMEM;
 974
 975         /* 1/128th of the device by default: */
 976         nr = ca->mi.nbuckets >> 7;
 977
 978         /*
 979          * clamp journal size to 8192 buckets or 8GB (in sectors), whichever
 980          * is smaller:
 981          */
 982         nr = clamp_t(unsigned, nr,
 983                      BCH_JOURNAL_BUCKETS_MIN,
 984                      min(1 << 13,
 985                          (1 << 24) / ca->mi.bucket_size));
 986
 987         if (ca->fs)
 988                 mutex_lock(&ca->fs->sb_lock);
 989
 990         ret = __bch2_set_nr_journal_buckets(ca, nr, true, NULL);
 991
 992         if (ca->fs)
 993                 mutex_unlock(&ca->fs->sb_lock);
 994
 995         return ret;
 996 }
 997
 998 /* startup/shutdown: */
 999
1000 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx)
1001 {
1002         bool ret = false;
1003         u64 seq;
1004
1005         spin_lock(&j->lock);
1006         for (seq = journal_last_unwritten_seq(j);
1007              seq <= journal_cur_seq(j) && !ret;
1008              seq++) {
1009                 struct journal_buf *buf = journal_seq_to_buf(j, seq);
1010
1011                 if (bch2_bkey_has_device(bkey_i_to_s_c(&buf->key), dev_idx))
1012                         ret = true;
1013         }
1014         spin_unlock(&j->lock);
1015
1016         return ret;
1017 }
1018
1019 void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca)
1020 {
1021         wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx));
1022 }
1023
1024 void bch2_fs_journal_stop(struct journal *j)
1025 {
1026         bch2_journal_reclaim_stop(j);
1027         bch2_journal_flush_all_pins(j);
1028
1029         wait_event(j->wait, journal_entry_close(j));
1030
1031         /*
1032          * Always write a new journal entry, to make sure the clock hands are up
1033          * to date (and match the superblock)
1034          */
1035         bch2_journal_meta(j);
1036
1037         journal_quiesce(j);
1038
1039         BUG_ON(!bch2_journal_error(j) &&
1040                test_bit(JOURNAL_REPLAY_DONE, &j->flags) &&
1041                j->last_empty_seq != journal_cur_seq(j));
1042
1043         cancel_delayed_work_sync(&j->write_work);
1044 }
1045
1046 int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
1047 {
1048         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1049         struct journal_entry_pin_list *p;
1050         struct journal_replay *i, **_i;
1051         struct genradix_iter iter;
1052         bool had_entries = false;
1053         unsigned ptr;
1054         u64 last_seq = cur_seq, nr, seq;
1055
1056         genradix_for_each_reverse(&c->journal_entries, iter, _i) {
1057                 i = *_i;
1058
1059                 if (!i || i->ignore)
1060                         continue;
1061
1062                 last_seq = le64_to_cpu(i->j.last_seq);
1063                 break;
1064         }
1065
1066         nr = cur_seq - last_seq;
1067
1068         if (nr + 1 > j->pin.size) {
1069                 free_fifo(&j->pin);
1070                 init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL);
1071                 if (!j->pin.data) {
1072                         bch_err(c, "error reallocating journal fifo (%llu open entries)", nr);
1073                         return -ENOMEM;
1074                 }
1075         }
1076
1077         j->replay_journal_seq   = last_seq;
1078         j->replay_journal_seq_end = cur_seq;
1079         j->last_seq_ondisk      = last_seq;
1080         j->flushed_seq_ondisk   = cur_seq - 1;
1081         j->seq_ondisk           = cur_seq - 1;
1082         j->pin.front            = last_seq;
1083         j->pin.back             = cur_seq;
1084         atomic64_set(&j->seq, cur_seq - 1);
1085
1086         fifo_for_each_entry_ptr(p, &j->pin, seq)
1087                 journal_pin_list_init(p, 1);
1088
1089         genradix_for_each(&c->journal_entries, iter, _i) {
1090                 i = *_i;
1091
1092                 if (!i || i->ignore)
1093                         continue;
1094
1095                 seq = le64_to_cpu(i->j.seq);
1096                 BUG_ON(seq >= cur_seq);
1097
1098                 if (seq < last_seq)
1099                         continue;
1100
1101                 if (journal_entry_empty(&i->j))
1102                         j->last_empty_seq = le64_to_cpu(i->j.seq);
1103
1104                 p = journal_seq_pin(j, seq);
1105
1106                 p->devs.nr = 0;
1107                 for (ptr = 0; ptr < i->nr_ptrs; ptr++)
1108                         bch2_dev_list_add_dev(&p->devs, i->ptrs[ptr].dev);
1109
1110                 had_entries = true;
1111         }
1112
1113         if (!had_entries)
1114                 j->last_empty_seq = cur_seq;
1115
1116         spin_lock(&j->lock);
1117
1118         set_bit(JOURNAL_STARTED, &j->flags);
1119         j->last_flush_write = jiffies;
1120
1121         j->reservations.idx = j->reservations.unwritten_idx = journal_cur_seq(j);
1122         j->reservations.unwritten_idx++;
1123
1124         c->last_bucket_seq_cleanup = journal_cur_seq(j);
1125
1126         bch2_journal_space_available(j);
1127         spin_unlock(&j->lock);
1128
1129         return bch2_journal_reclaim_start(j);
1130 }
1131
1132 /* init/exit: */
1133
1134 void bch2_dev_journal_exit(struct bch_dev *ca)
1135 {
1136         kfree(ca->journal.bio);
1137         kfree(ca->journal.buckets);
1138         kfree(ca->journal.bucket_seq);
1139
1140         ca->journal.bio         = NULL;
1141         ca->journal.buckets     = NULL;
1142         ca->journal.bucket_seq  = NULL;
1143 }
1144
1145 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb)
1146 {
1147         struct journal_device *ja = &ca->journal;
1148         struct bch_sb_field_journal *journal_buckets =
1149                 bch2_sb_get_journal(sb);
1150         struct bch_sb_field_journal_v2 *journal_buckets_v2 =
1151                 bch2_sb_get_journal_v2(sb);
1152         unsigned i, nr_bvecs;
1153
1154         ja->nr = 0;
1155
1156         if (journal_buckets_v2) {
1157                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1158
1159                 for (i = 0; i < nr; i++)
1160                         ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr);
1161         } else if (journal_buckets) {
1162                 ja->nr = bch2_nr_journal_buckets(journal_buckets);
1163         }
1164
1165         ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1166         if (!ja->bucket_seq)
1167                 return -ENOMEM;
1168
1169         nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE);
1170
1171         ca->journal.bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1172         if (!ca->journal.bio)
1173                 return -ENOMEM;
1174
1175         bio_init(ca->journal.bio, NULL, ca->journal.bio->bi_inline_vecs, nr_bvecs, 0);
1176
1177         ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL);
1178         if (!ja->buckets)
1179                 return -ENOMEM;
1180
1181         if (journal_buckets_v2) {
1182                 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2);
1183                 unsigned j, dst = 0;
1184
1185                 for (i = 0; i < nr; i++)
1186                         for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++)
1187                                 ja->buckets[dst++] =
1188                                         le64_to_cpu(journal_buckets_v2->d[i].start) + j;
1189         } else if (journal_buckets) {
1190                 for (i = 0; i < ja->nr; i++)
1191                         ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]);
1192         }
1193
1194         return 0;
1195 }
1196
1197 void bch2_fs_journal_exit(struct journal *j)
1198 {
1199         unsigned i;
1200
1201         for (i = 0; i < ARRAY_SIZE(j->buf); i++)
1202                 kvpfree(j->buf[i].data, j->buf[i].buf_size);
1203         free_fifo(&j->pin);
1204 }
1205
1206 int bch2_fs_journal_init(struct journal *j)
1207 {
1208         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1209         static struct lock_class_key res_key;
1210         unsigned i;
1211         int ret = 0;
1212
1213         pr_verbose_init(c->opts, "");
1214
1215         spin_lock_init(&j->lock);
1216         spin_lock_init(&j->err_lock);
1217         init_waitqueue_head(&j->wait);
1218         INIT_DELAYED_WORK(&j->write_work, journal_write_work);
1219         init_waitqueue_head(&j->reclaim_wait);
1220         init_waitqueue_head(&j->pin_flush_wait);
1221         mutex_init(&j->reclaim_lock);
1222         mutex_init(&j->discard_lock);
1223
1224         lockdep_init_map(&j->res_map, "journal res", &res_key, 0);
1225
1226         atomic64_set(&j->reservations.counter,
1227                 ((union journal_res_state)
1228                  { .cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL }).v);
1229
1230         if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL))) {
1231                 ret = -ENOMEM;
1232                 goto out;
1233         }
1234
1235         for (i = 0; i < ARRAY_SIZE(j->buf); i++) {
1236                 j->buf[i].buf_size = JOURNAL_ENTRY_SIZE_MIN;
1237                 j->buf[i].data = kvpmalloc(j->buf[i].buf_size, GFP_KERNEL);
1238                 if (!j->buf[i].data) {
1239                         ret = -ENOMEM;
1240                         goto out;
1241                 }
1242         }
1243
1244         j->pin.front = j->pin.back = 1;
1245 out:
1246         pr_verbose_init(c->opts, "ret %i", ret);
1247         return ret;
1248 }
1249
1250 /* debug: */
1251
1252 void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1253 {
1254         struct bch_fs *c = container_of(j, struct bch_fs, journal);
1255         union journal_res_state s;
1256         struct bch_dev *ca;
1257         unsigned long now = jiffies;
1258         u64 seq;
1259         unsigned i;
1260
1261         if (!out->nr_tabstops)
1262                 printbuf_tabstop_push(out, 24);
1263         out->atomic++;
1264
1265         rcu_read_lock();
1266         s = READ_ONCE(j->reservations);
1267
1268         prt_printf(out, "dirty journal entries:\t%llu/%llu\n",  fifo_used(&j->pin), j->pin.size);
1269         prt_printf(out, "seq:\t\t\t%llu\n",                     journal_cur_seq(j));
1270         prt_printf(out, "seq_ondisk:\t\t%llu\n",                j->seq_ondisk);
1271         prt_printf(out, "last_seq:\t\t%llu\n",          journal_last_seq(j));
1272         prt_printf(out, "last_seq_ondisk:\t%llu\n",             j->last_seq_ondisk);
1273         prt_printf(out, "flushed_seq_ondisk:\t%llu\n",  j->flushed_seq_ondisk);
1274         prt_printf(out, "prereserved:\t\t%u/%u\n",              j->prereserved.reserved, j->prereserved.remaining);
1275         prt_printf(out, "watermark:\t\t%s\n",           bch2_journal_watermarks[j->watermark]);
1276         prt_printf(out, "each entry reserved:\t%u\n",   j->entry_u64s_reserved);
1277         prt_printf(out, "nr flush writes:\t%llu\n",             j->nr_flush_writes);
1278         prt_printf(out, "nr noflush writes:\t%llu\n",   j->nr_noflush_writes);
1279         prt_printf(out, "nr direct reclaim:\t%llu\n",   j->nr_direct_reclaim);
1280         prt_printf(out, "nr background reclaim:\t%llu\n",       j->nr_background_reclaim);
1281         prt_printf(out, "reclaim kicked:\t\t%u\n",              j->reclaim_kicked);
1282         prt_printf(out, "reclaim runs in:\t%u ms\n",    time_after(j->next_reclaim, now)
1283                ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0);
1284         prt_printf(out, "current entry sectors:\t%u\n", j->cur_entry_sectors);
1285         prt_printf(out, "current entry error:\t%s\n",   bch2_journal_errors[j->cur_entry_error]);
1286         prt_printf(out, "current entry:\t\t");
1287
1288         switch (s.cur_entry_offset) {
1289         case JOURNAL_ENTRY_ERROR_VAL:
1290                 prt_printf(out, "error");
1291                 break;
1292         case JOURNAL_ENTRY_CLOSED_VAL:
1293                 prt_printf(out, "closed");
1294                 break;
1295         default:
1296                 prt_printf(out, "%u/%u", s.cur_entry_offset, j->cur_entry_u64s);
1297                 break;
1298         }
1299
1300         prt_newline(out);
1301
1302         for (seq = journal_cur_seq(j);
1303              seq >= journal_last_unwritten_seq(j);
1304              --seq) {
1305                 i = seq & JOURNAL_BUF_MASK;
1306
1307                 prt_printf(out, "unwritten entry:");
1308                 prt_tab(out);
1309                 prt_printf(out, "%llu", seq);
1310                 prt_newline(out);
1311                 printbuf_indent_add(out, 2);
1312
1313                 prt_printf(out, "refcount:");
1314                 prt_tab(out);
1315                 prt_printf(out, "%u", journal_state_count(s, i));
1316                 prt_newline(out);
1317
1318                 prt_printf(out, "sectors:");
1319                 prt_tab(out);
1320                 prt_printf(out, "%u", j->buf[i].sectors);
1321                 prt_newline(out);
1322
1323                 prt_printf(out, "expires");
1324                 prt_tab(out);
1325                 prt_printf(out, "%li jiffies", j->buf[i].expires - jiffies);
1326                 prt_newline(out);
1327
1328                 printbuf_indent_sub(out, 2);
1329         }
1330
1331         prt_printf(out,
1332                "replay done:\t\t%i\n",
1333                test_bit(JOURNAL_REPLAY_DONE,    &j->flags));
1334
1335         prt_printf(out, "space:\n");
1336         prt_printf(out, "\tdiscarded\t%u:%u\n",
1337                j->space[journal_space_discarded].next_entry,
1338                j->space[journal_space_discarded].total);
1339         prt_printf(out, "\tclean ondisk\t%u:%u\n",
1340                j->space[journal_space_clean_ondisk].next_entry,
1341                j->space[journal_space_clean_ondisk].total);
1342         prt_printf(out, "\tclean\t\t%u:%u\n",
1343                j->space[journal_space_clean].next_entry,
1344                j->space[journal_space_clean].total);
1345         prt_printf(out, "\ttotal\t\t%u:%u\n",
1346                j->space[journal_space_total].next_entry,
1347                j->space[journal_space_total].total);
1348
1349         for_each_member_device_rcu(ca, c, i,
1350                                    &c->rw_devs[BCH_DATA_journal]) {
1351                 struct journal_device *ja = &ca->journal;
1352
1353                 if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d))
1354                         continue;
1355
1356                 if (!ja->nr)
1357                         continue;
1358
1359                 prt_printf(out, "dev %u:\n",            i);
1360                 prt_printf(out, "\tnr\t\t%u\n",         ja->nr);
1361                 prt_printf(out, "\tbucket size\t%u\n",  ca->mi.bucket_size);
1362                 prt_printf(out, "\tavailable\t%u:%u\n", bch2_journal_dev_buckets_available(j, ja, journal_space_discarded), ja->sectors_free);
1363                 prt_printf(out, "\tdiscard_idx\t%u\n",  ja->discard_idx);
1364                 prt_printf(out, "\tdirty_ondisk\t%u (seq %llu)\n", ja->dirty_idx_ondisk,        ja->bucket_seq[ja->dirty_idx_ondisk]);
1365                 prt_printf(out, "\tdirty_idx\t%u (seq %llu)\n", ja->dirty_idx,          ja->bucket_seq[ja->dirty_idx]);
1366                 prt_printf(out, "\tcur_idx\t\t%u (seq %llu)\n", ja->cur_idx,            ja->bucket_seq[ja->cur_idx]);
1367         }
1368
1369         rcu_read_unlock();
1370
1371         --out->atomic;
1372 }
1373
1374 void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
1375 {
1376         spin_lock(&j->lock);
1377         __bch2_journal_debug_to_text(out, j);
1378         spin_unlock(&j->lock);
1379 }
1380
1381 bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq)
1382 {
1383         struct journal_entry_pin_list *pin_list;
1384         struct journal_entry_pin *pin;
1385
1386         spin_lock(&j->lock);
1387         *seq = max(*seq, j->pin.front);
1388
1389         if (*seq >= j->pin.back) {
1390                 spin_unlock(&j->lock);
1391                 return true;
1392         }
1393
1394         out->atomic++;
1395
1396         pin_list = journal_seq_pin(j, *seq);
1397
1398         prt_printf(out, "%llu: count %u", *seq, atomic_read(&pin_list->count));
1399         prt_newline(out);
1400         printbuf_indent_add(out, 2);
1401
1402         list_for_each_entry(pin, &pin_list->list, list) {
1403                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1404                 prt_newline(out);
1405         }
1406
1407         list_for_each_entry(pin, &pin_list->key_cache_list, list) {
1408                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1409                 prt_newline(out);
1410         }
1411
1412         if (!list_empty(&pin_list->flushed)) {
1413                 prt_printf(out, "flushed:");
1414                 prt_newline(out);
1415         }
1416
1417         list_for_each_entry(pin, &pin_list->flushed, list) {
1418                 prt_printf(out, "\t%px %ps", pin, pin->flush);
1419                 prt_newline(out);
1420         }
1421
1422         printbuf_indent_sub(out, 2);
1423
1424         --out->atomic;
1425         spin_unlock(&j->lock);
1426
1427         return false;
1428 }
1429
1430 void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j)
1431 {
1432         u64 seq = 0;
1433
1434         while (!bch2_journal_seq_pins_to_text(out, j, &seq))
1435                 seq++;
1436 }