1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * bcachefs journalling code, for btree insertions 4 * 5 * Copyright 2012 Google, Inc. 6 */ 7 8 #include "bcachefs.h" 9 #include "alloc_foreground.h" 10 #include "bkey_methods.h" 11 #include "btree_gc.h" 12 #include "btree_update.h" 13 #include "buckets.h" 14 #include "error.h" 15 #include "journal.h" 16 #include "journal_io.h" 17 #include "journal_reclaim.h" 18 #include "journal_sb.h" 19 #include "journal_seq_blacklist.h" 20 #include "trace.h" 21 22 static const char * const bch2_journal_errors[] = { 23 #define x(n) #n, 24 JOURNAL_ERRORS() 25 #undef x 26 NULL 27 }; 28 29 static inline bool journal_seq_unwritten(struct journal *j, u64 seq) 30 { 31 return seq > j->seq_ondisk; 32 } 33 34 static bool __journal_entry_is_open(union journal_res_state state) 35 { 36 return state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL; 37 } 38 39 static inline unsigned nr_unwritten_journal_entries(struct journal *j) 40 { 41 return atomic64_read(&j->seq) - j->seq_ondisk; 42 } 43 44 static bool journal_entry_is_open(struct journal *j) 45 { 46 return __journal_entry_is_open(j->reservations); 47 } 48 49 static inline struct journal_buf * 50 journal_seq_to_buf(struct journal *j, u64 seq) 51 { 52 struct journal_buf *buf = NULL; 53 54 EBUG_ON(seq > journal_cur_seq(j)); 55 56 if (journal_seq_unwritten(j, seq)) { 57 buf = j->buf + (seq & JOURNAL_BUF_MASK); 58 EBUG_ON(le64_to_cpu(buf->data->seq) != seq); 59 } 60 return buf; 61 } 62 63 static void journal_pin_list_init(struct journal_entry_pin_list *p, int count) 64 { 65 unsigned i; 66 67 for (i = 0; i < ARRAY_SIZE(p->list); i++) 68 INIT_LIST_HEAD(&p->list[i]); 69 INIT_LIST_HEAD(&p->flushed); 70 atomic_set(&p->count, count); 71 p->devs.nr = 0; 72 } 73 74 /* 75 * Detect stuck journal conditions and trigger shutdown. Technically the journal 76 * can end up stuck for a variety of reasons, such as a blocked I/O, journal 77 * reservation lockup, etc. Since this is a fatal error with potentially 78 * unpredictable characteristics, we want to be fairly conservative before we 79 * decide to shut things down. 80 * 81 * Consider the journal stuck when it appears full with no ability to commit 82 * btree transactions, to discard journal buckets, nor acquire priority 83 * (reserved watermark) reservation. 84 */ 85 static inline bool 86 journal_error_check_stuck(struct journal *j, int error, unsigned flags) 87 { 88 struct bch_fs *c = container_of(j, struct bch_fs, journal); 89 bool stuck = false; 90 struct printbuf buf = PRINTBUF; 91 92 if (!(error == JOURNAL_ERR_journal_full || 93 error == JOURNAL_ERR_journal_pin_full) || 94 nr_unwritten_journal_entries(j) || 95 (flags & BCH_WATERMARK_MASK) != BCH_WATERMARK_reclaim) 96 return stuck; 97 98 spin_lock(&j->lock); 99 100 if (j->can_discard) { 101 spin_unlock(&j->lock); 102 return stuck; 103 } 104 105 stuck = true; 106 107 /* 108 * The journal shutdown path will set ->err_seq, but do it here first to 109 * serialize against concurrent failures and avoid duplicate error 110 * reports. 111 */ 112 if (j->err_seq) { 113 spin_unlock(&j->lock); 114 return stuck; 115 } 116 j->err_seq = journal_cur_seq(j); 117 spin_unlock(&j->lock); 118 119 bch_err(c, "Journal stuck! Hava a pre-reservation but journal full (error %s)", 120 bch2_journal_errors[error]); 121 bch2_journal_debug_to_text(&buf, j); 122 bch_err(c, "%s", buf.buf); 123 124 printbuf_reset(&buf); 125 bch2_journal_pins_to_text(&buf, j); 126 bch_err(c, "Journal pins:\n%s", buf.buf); 127 printbuf_exit(&buf); 128 129 bch2_fatal_error(c); 130 dump_stack(); 131 132 return stuck; 133 } 134 135 /* 136 * Final processing when the last reference of a journal buffer has been 137 * dropped. Drop the pin list reference acquired at journal entry open and write 138 * the buffer, if requested. 139 */ 140 void bch2_journal_buf_put_final(struct journal *j, u64 seq, bool write) 141 { 142 struct bch_fs *c = container_of(j, struct bch_fs, journal); 143 144 lockdep_assert_held(&j->lock); 145 146 if (__bch2_journal_pin_put(j, seq)) 147 bch2_journal_reclaim_fast(j); 148 if (write) 149 closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL); 150 } 151 152 /* 153 * Returns true if journal entry is now closed: 154 * 155 * We don't close a journal_buf until the next journal_buf is finished writing, 156 * and can be opened again - this also initializes the next journal_buf: 157 */ 158 static void __journal_entry_close(struct journal *j, unsigned closed_val) 159 { 160 struct bch_fs *c = container_of(j, struct bch_fs, journal); 161 struct journal_buf *buf = journal_cur_buf(j); 162 union journal_res_state old, new; 163 u64 v = atomic64_read(&j->reservations.counter); 164 unsigned sectors; 165 166 BUG_ON(closed_val != JOURNAL_ENTRY_CLOSED_VAL && 167 closed_val != JOURNAL_ENTRY_ERROR_VAL); 168 169 lockdep_assert_held(&j->lock); 170 171 do { 172 old.v = new.v = v; 173 new.cur_entry_offset = closed_val; 174 175 if (old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL || 176 old.cur_entry_offset == new.cur_entry_offset) 177 return; 178 } while ((v = atomic64_cmpxchg(&j->reservations.counter, 179 old.v, new.v)) != old.v); 180 181 if (!__journal_entry_is_open(old)) 182 return; 183 184 /* Close out old buffer: */ 185 buf->data->u64s = cpu_to_le32(old.cur_entry_offset); 186 187 sectors = vstruct_blocks_plus(buf->data, c->block_bits, 188 buf->u64s_reserved) << c->block_bits; 189 BUG_ON(sectors > buf->sectors); 190 buf->sectors = sectors; 191 192 /* 193 * We have to set last_seq here, _before_ opening a new journal entry: 194 * 195 * A threads may replace an old pin with a new pin on their current 196 * journal reservation - the expectation being that the journal will 197 * contain either what the old pin protected or what the new pin 198 * protects. 199 * 200 * After the old pin is dropped journal_last_seq() won't include the old 201 * pin, so we can only write the updated last_seq on the entry that 202 * contains whatever the new pin protects. 203 * 204 * Restated, we can _not_ update last_seq for a given entry if there 205 * could be a newer entry open with reservations/pins that have been 206 * taken against it. 207 * 208 * Hence, we want update/set last_seq on the current journal entry right 209 * before we open a new one: 210 */ 211 buf->last_seq = journal_last_seq(j); 212 buf->data->last_seq = cpu_to_le64(buf->last_seq); 213 BUG_ON(buf->last_seq > le64_to_cpu(buf->data->seq)); 214 215 cancel_delayed_work(&j->write_work); 216 217 bch2_journal_space_available(j); 218 219 __bch2_journal_buf_put(j, old.idx, le64_to_cpu(buf->data->seq)); 220 } 221 222 void bch2_journal_halt(struct journal *j) 223 { 224 spin_lock(&j->lock); 225 __journal_entry_close(j, JOURNAL_ENTRY_ERROR_VAL); 226 if (!j->err_seq) 227 j->err_seq = journal_cur_seq(j); 228 journal_wake(j); 229 spin_unlock(&j->lock); 230 } 231 232 static bool journal_entry_want_write(struct journal *j) 233 { 234 bool ret = !journal_entry_is_open(j) || 235 journal_cur_seq(j) == journal_last_unwritten_seq(j); 236 237 /* Don't close it yet if we already have a write in flight: */ 238 if (ret) 239 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL); 240 else if (nr_unwritten_journal_entries(j)) { 241 struct journal_buf *buf = journal_cur_buf(j); 242 243 if (!buf->flush_time) { 244 buf->flush_time = local_clock() ?: 1; 245 buf->expires = jiffies; 246 } 247 } 248 249 return ret; 250 } 251 252 static bool journal_entry_close(struct journal *j) 253 { 254 bool ret; 255 256 spin_lock(&j->lock); 257 ret = journal_entry_want_write(j); 258 spin_unlock(&j->lock); 259 260 return ret; 261 } 262 263 /* 264 * should _only_ called from journal_res_get() - when we actually want a 265 * journal reservation - journal entry is open means journal is dirty: 266 */ 267 static int journal_entry_open(struct journal *j) 268 { 269 struct bch_fs *c = container_of(j, struct bch_fs, journal); 270 struct journal_buf *buf = j->buf + 271 ((journal_cur_seq(j) + 1) & JOURNAL_BUF_MASK); 272 union journal_res_state old, new; 273 int u64s; 274 u64 v; 275 276 lockdep_assert_held(&j->lock); 277 BUG_ON(journal_entry_is_open(j)); 278 BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb)); 279 280 if (j->blocked) 281 return JOURNAL_ERR_blocked; 282 283 if (j->cur_entry_error) 284 return j->cur_entry_error; 285 286 if (bch2_journal_error(j)) 287 return JOURNAL_ERR_insufficient_devices; /* -EROFS */ 288 289 if (!fifo_free(&j->pin)) 290 return JOURNAL_ERR_journal_pin_full; 291 292 if (nr_unwritten_journal_entries(j) == ARRAY_SIZE(j->buf)) 293 return JOURNAL_ERR_max_in_flight; 294 295 BUG_ON(!j->cur_entry_sectors); 296 297 buf->expires = 298 (journal_cur_seq(j) == j->flushed_seq_ondisk 299 ? jiffies 300 : j->last_flush_write) + 301 msecs_to_jiffies(c->opts.journal_flush_delay); 302 303 buf->u64s_reserved = j->entry_u64s_reserved; 304 buf->disk_sectors = j->cur_entry_sectors; 305 buf->sectors = min(buf->disk_sectors, buf->buf_size >> 9); 306 307 u64s = (int) (buf->sectors << 9) / sizeof(u64) - 308 journal_entry_overhead(j); 309 u64s = clamp_t(int, u64s, 0, JOURNAL_ENTRY_CLOSED_VAL - 1); 310 311 if (u64s <= (ssize_t) j->early_journal_entries.nr) 312 return JOURNAL_ERR_journal_full; 313 314 if (fifo_empty(&j->pin) && j->reclaim_thread) 315 wake_up_process(j->reclaim_thread); 316 317 /* 318 * The fifo_push() needs to happen at the same time as j->seq is 319 * incremented for journal_last_seq() to be calculated correctly 320 */ 321 atomic64_inc(&j->seq); 322 journal_pin_list_init(fifo_push_ref(&j->pin), 1); 323 324 BUG_ON(j->buf + (journal_cur_seq(j) & JOURNAL_BUF_MASK) != buf); 325 326 bkey_extent_init(&buf->key); 327 buf->noflush = false; 328 buf->must_flush = false; 329 buf->separate_flush = false; 330 buf->flush_time = 0; 331 332 memset(buf->data, 0, sizeof(*buf->data)); 333 buf->data->seq = cpu_to_le64(journal_cur_seq(j)); 334 buf->data->u64s = 0; 335 336 if (j->early_journal_entries.nr) { 337 memcpy(buf->data->_data, j->early_journal_entries.data, 338 j->early_journal_entries.nr * sizeof(u64)); 339 le32_add_cpu(&buf->data->u64s, j->early_journal_entries.nr); 340 } 341 342 /* 343 * Must be set before marking the journal entry as open: 344 */ 345 j->cur_entry_u64s = u64s; 346 347 v = atomic64_read(&j->reservations.counter); 348 do { 349 old.v = new.v = v; 350 351 BUG_ON(old.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL); 352 353 new.idx++; 354 BUG_ON(journal_state_count(new, new.idx)); 355 BUG_ON(new.idx != (journal_cur_seq(j) & JOURNAL_BUF_MASK)); 356 357 journal_state_inc(&new); 358 359 /* Handle any already added entries */ 360 new.cur_entry_offset = le32_to_cpu(buf->data->u64s); 361 } while ((v = atomic64_cmpxchg(&j->reservations.counter, 362 old.v, new.v)) != old.v); 363 364 if (j->res_get_blocked_start) 365 bch2_time_stats_update(j->blocked_time, 366 j->res_get_blocked_start); 367 j->res_get_blocked_start = 0; 368 369 mod_delayed_work(c->io_complete_wq, 370 &j->write_work, 371 msecs_to_jiffies(c->opts.journal_flush_delay)); 372 journal_wake(j); 373 374 if (j->early_journal_entries.nr) 375 darray_exit(&j->early_journal_entries); 376 return 0; 377 } 378 379 static bool journal_quiesced(struct journal *j) 380 { 381 bool ret = atomic64_read(&j->seq) == j->seq_ondisk; 382 383 if (!ret) 384 journal_entry_close(j); 385 return ret; 386 } 387 388 static void journal_quiesce(struct journal *j) 389 { 390 wait_event(j->wait, journal_quiesced(j)); 391 } 392 393 static void journal_write_work(struct work_struct *work) 394 { 395 struct journal *j = container_of(work, struct journal, write_work.work); 396 struct bch_fs *c = container_of(j, struct bch_fs, journal); 397 long delta; 398 399 spin_lock(&j->lock); 400 if (!__journal_entry_is_open(j->reservations)) 401 goto unlock; 402 403 delta = journal_cur_buf(j)->expires - jiffies; 404 405 if (delta > 0) 406 mod_delayed_work(c->io_complete_wq, &j->write_work, delta); 407 else 408 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL); 409 unlock: 410 spin_unlock(&j->lock); 411 } 412 413 static int __journal_res_get(struct journal *j, struct journal_res *res, 414 unsigned flags) 415 { 416 struct bch_fs *c = container_of(j, struct bch_fs, journal); 417 struct journal_buf *buf; 418 bool can_discard; 419 int ret; 420 retry: 421 if (journal_res_get_fast(j, res, flags)) 422 return 0; 423 424 if (bch2_journal_error(j)) 425 return -BCH_ERR_erofs_journal_err; 426 427 spin_lock(&j->lock); 428 429 /* check once more in case somebody else shut things down... */ 430 if (bch2_journal_error(j)) { 431 spin_unlock(&j->lock); 432 return -BCH_ERR_erofs_journal_err; 433 } 434 435 /* 436 * Recheck after taking the lock, so we don't race with another thread 437 * that just did journal_entry_open() and call journal_entry_close() 438 * unnecessarily 439 */ 440 if (journal_res_get_fast(j, res, flags)) { 441 spin_unlock(&j->lock); 442 return 0; 443 } 444 445 if ((flags & BCH_WATERMARK_MASK) < j->watermark) { 446 /* 447 * Don't want to close current journal entry, just need to 448 * invoke reclaim: 449 */ 450 ret = JOURNAL_ERR_journal_full; 451 goto unlock; 452 } 453 454 /* 455 * If we couldn't get a reservation because the current buf filled up, 456 * and we had room for a bigger entry on disk, signal that we want to 457 * realloc the journal bufs: 458 */ 459 buf = journal_cur_buf(j); 460 if (journal_entry_is_open(j) && 461 buf->buf_size >> 9 < buf->disk_sectors && 462 buf->buf_size < JOURNAL_ENTRY_SIZE_MAX) 463 j->buf_size_want = max(j->buf_size_want, buf->buf_size << 1); 464 465 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL); 466 ret = journal_entry_open(j); 467 468 if (ret == JOURNAL_ERR_max_in_flight) 469 trace_and_count(c, journal_entry_full, c); 470 unlock: 471 if ((ret && ret != JOURNAL_ERR_insufficient_devices) && 472 !j->res_get_blocked_start) { 473 j->res_get_blocked_start = local_clock() ?: 1; 474 trace_and_count(c, journal_full, c); 475 } 476 477 can_discard = j->can_discard; 478 spin_unlock(&j->lock); 479 480 if (!ret) 481 goto retry; 482 if (journal_error_check_stuck(j, ret, flags)) 483 ret = -BCH_ERR_journal_res_get_blocked; 484 485 /* 486 * Journal is full - can't rely on reclaim from work item due to 487 * freezing: 488 */ 489 if ((ret == JOURNAL_ERR_journal_full || 490 ret == JOURNAL_ERR_journal_pin_full) && 491 !(flags & JOURNAL_RES_GET_NONBLOCK)) { 492 if (can_discard) { 493 bch2_journal_do_discards(j); 494 goto retry; 495 } 496 497 if (mutex_trylock(&j->reclaim_lock)) { 498 bch2_journal_reclaim(j); 499 mutex_unlock(&j->reclaim_lock); 500 } 501 } 502 503 return ret == JOURNAL_ERR_insufficient_devices 504 ? -BCH_ERR_erofs_journal_err 505 : -BCH_ERR_journal_res_get_blocked; 506 } 507 508 /* 509 * Essentially the entry function to the journaling code. When bcachefs is doing 510 * a btree insert, it calls this function to get the current journal write. 511 * Journal write is the structure used set up journal writes. The calling 512 * function will then add its keys to the structure, queuing them for the next 513 * write. 514 * 515 * To ensure forward progress, the current task must not be holding any 516 * btree node write locks. 517 */ 518 int bch2_journal_res_get_slowpath(struct journal *j, struct journal_res *res, 519 unsigned flags) 520 { 521 int ret; 522 523 closure_wait_event(&j->async_wait, 524 (ret = __journal_res_get(j, res, flags)) != -BCH_ERR_journal_res_get_blocked || 525 (flags & JOURNAL_RES_GET_NONBLOCK)); 526 return ret; 527 } 528 529 /* journal_entry_res: */ 530 531 void bch2_journal_entry_res_resize(struct journal *j, 532 struct journal_entry_res *res, 533 unsigned new_u64s) 534 { 535 union journal_res_state state; 536 int d = new_u64s - res->u64s; 537 538 spin_lock(&j->lock); 539 540 j->entry_u64s_reserved += d; 541 if (d <= 0) 542 goto out; 543 544 j->cur_entry_u64s = max_t(int, 0, j->cur_entry_u64s - d); 545 smp_mb(); 546 state = READ_ONCE(j->reservations); 547 548 if (state.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL && 549 state.cur_entry_offset > j->cur_entry_u64s) { 550 j->cur_entry_u64s += d; 551 /* 552 * Not enough room in current journal entry, have to flush it: 553 */ 554 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL); 555 } else { 556 journal_cur_buf(j)->u64s_reserved += d; 557 } 558 out: 559 spin_unlock(&j->lock); 560 res->u64s += d; 561 } 562 563 /* journal flushing: */ 564 565 /** 566 * bch2_journal_flush_seq_async - wait for a journal entry to be written 567 * @j: journal object 568 * @seq: seq to flush 569 * @parent: closure object to wait with 570 * Returns: 1 if @seq has already been flushed, 0 if @seq is being flushed, 571 * -EIO if @seq will never be flushed 572 * 573 * Like bch2_journal_wait_on_seq, except that it triggers a write immediately if 574 * necessary 575 */ 576 int bch2_journal_flush_seq_async(struct journal *j, u64 seq, 577 struct closure *parent) 578 { 579 struct journal_buf *buf; 580 int ret = 0; 581 582 if (seq <= j->flushed_seq_ondisk) 583 return 1; 584 585 spin_lock(&j->lock); 586 587 if (WARN_ONCE(seq > journal_cur_seq(j), 588 "requested to flush journal seq %llu, but currently at %llu", 589 seq, journal_cur_seq(j))) 590 goto out; 591 592 /* Recheck under lock: */ 593 if (j->err_seq && seq >= j->err_seq) { 594 ret = -EIO; 595 goto out; 596 } 597 598 if (seq <= j->flushed_seq_ondisk) { 599 ret = 1; 600 goto out; 601 } 602 603 /* if seq was written, but not flushed - flush a newer one instead */ 604 seq = max(seq, journal_last_unwritten_seq(j)); 605 606 recheck_need_open: 607 if (seq > journal_cur_seq(j)) { 608 struct journal_res res = { 0 }; 609 610 if (journal_entry_is_open(j)) 611 __journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL); 612 613 spin_unlock(&j->lock); 614 615 ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0); 616 if (ret) 617 return ret; 618 619 seq = res.seq; 620 buf = j->buf + (seq & JOURNAL_BUF_MASK); 621 buf->must_flush = true; 622 623 if (!buf->flush_time) { 624 buf->flush_time = local_clock() ?: 1; 625 buf->expires = jiffies; 626 } 627 628 if (parent && !closure_wait(&buf->wait, parent)) 629 BUG(); 630 631 bch2_journal_res_put(j, &res); 632 633 spin_lock(&j->lock); 634 goto want_write; 635 } 636 637 /* 638 * if write was kicked off without a flush, flush the next sequence 639 * number instead 640 */ 641 buf = journal_seq_to_buf(j, seq); 642 if (buf->noflush) { 643 seq++; 644 goto recheck_need_open; 645 } 646 647 buf->must_flush = true; 648 649 if (parent && !closure_wait(&buf->wait, parent)) 650 BUG(); 651 want_write: 652 if (seq == journal_cur_seq(j)) 653 journal_entry_want_write(j); 654 out: 655 spin_unlock(&j->lock); 656 return ret; 657 } 658 659 int bch2_journal_flush_seq(struct journal *j, u64 seq) 660 { 661 u64 start_time = local_clock(); 662 int ret, ret2; 663 664 /* 665 * Don't update time_stats when @seq is already flushed: 666 */ 667 if (seq <= j->flushed_seq_ondisk) 668 return 0; 669 670 ret = wait_event_interruptible(j->wait, (ret2 = bch2_journal_flush_seq_async(j, seq, NULL))); 671 672 if (!ret) 673 bch2_time_stats_update(j->flush_seq_time, start_time); 674 675 return ret ?: ret2 < 0 ? ret2 : 0; 676 } 677 678 /* 679 * bch2_journal_flush_async - if there is an open journal entry, or a journal 680 * still being written, write it and wait for the write to complete 681 */ 682 void bch2_journal_flush_async(struct journal *j, struct closure *parent) 683 { 684 bch2_journal_flush_seq_async(j, atomic64_read(&j->seq), parent); 685 } 686 687 int bch2_journal_flush(struct journal *j) 688 { 689 return bch2_journal_flush_seq(j, atomic64_read(&j->seq)); 690 } 691 692 /* 693 * bch2_journal_noflush_seq - tell the journal not to issue any flushes before 694 * @seq 695 */ 696 bool bch2_journal_noflush_seq(struct journal *j, u64 seq) 697 { 698 struct bch_fs *c = container_of(j, struct bch_fs, journal); 699 u64 unwritten_seq; 700 bool ret = false; 701 702 if (!(c->sb.features & (1ULL << BCH_FEATURE_journal_no_flush))) 703 return false; 704 705 if (seq <= c->journal.flushed_seq_ondisk) 706 return false; 707 708 spin_lock(&j->lock); 709 if (seq <= c->journal.flushed_seq_ondisk) 710 goto out; 711 712 for (unwritten_seq = journal_last_unwritten_seq(j); 713 unwritten_seq < seq; 714 unwritten_seq++) { 715 struct journal_buf *buf = journal_seq_to_buf(j, unwritten_seq); 716 717 /* journal write is already in flight, and was a flush write: */ 718 if (unwritten_seq == journal_last_unwritten_seq(j) && !buf->noflush) 719 goto out; 720 721 buf->noflush = true; 722 } 723 724 ret = true; 725 out: 726 spin_unlock(&j->lock); 727 return ret; 728 } 729 730 int bch2_journal_meta(struct journal *j) 731 { 732 struct journal_buf *buf; 733 struct journal_res res; 734 int ret; 735 736 memset(&res, 0, sizeof(res)); 737 738 ret = bch2_journal_res_get(j, &res, jset_u64s(0), 0); 739 if (ret) 740 return ret; 741 742 buf = j->buf + (res.seq & JOURNAL_BUF_MASK); 743 buf->must_flush = true; 744 745 if (!buf->flush_time) { 746 buf->flush_time = local_clock() ?: 1; 747 buf->expires = jiffies; 748 } 749 750 bch2_journal_res_put(j, &res); 751 752 return bch2_journal_flush_seq(j, res.seq); 753 } 754 755 /* block/unlock the journal: */ 756 757 void bch2_journal_unblock(struct journal *j) 758 { 759 spin_lock(&j->lock); 760 j->blocked--; 761 spin_unlock(&j->lock); 762 763 journal_wake(j); 764 } 765 766 void bch2_journal_block(struct journal *j) 767 { 768 spin_lock(&j->lock); 769 j->blocked++; 770 spin_unlock(&j->lock); 771 772 journal_quiesce(j); 773 } 774 775 /* allocate journal on a device: */ 776 777 static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr, 778 bool new_fs, struct closure *cl) 779 { 780 struct bch_fs *c = ca->fs; 781 struct journal_device *ja = &ca->journal; 782 u64 *new_bucket_seq = NULL, *new_buckets = NULL; 783 struct open_bucket **ob = NULL; 784 long *bu = NULL; 785 unsigned i, pos, nr_got = 0, nr_want = nr - ja->nr; 786 int ret = 0; 787 788 BUG_ON(nr <= ja->nr); 789 790 bu = kcalloc(nr_want, sizeof(*bu), GFP_KERNEL); 791 ob = kcalloc(nr_want, sizeof(*ob), GFP_KERNEL); 792 new_buckets = kcalloc(nr, sizeof(u64), GFP_KERNEL); 793 new_bucket_seq = kcalloc(nr, sizeof(u64), GFP_KERNEL); 794 if (!bu || !ob || !new_buckets || !new_bucket_seq) { 795 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets; 796 goto err_free; 797 } 798 799 for (nr_got = 0; nr_got < nr_want; nr_got++) { 800 if (new_fs) { 801 bu[nr_got] = bch2_bucket_alloc_new_fs(ca); 802 if (bu[nr_got] < 0) { 803 ret = -BCH_ERR_ENOSPC_bucket_alloc; 804 break; 805 } 806 } else { 807 ob[nr_got] = bch2_bucket_alloc(c, ca, BCH_WATERMARK_normal, cl); 808 ret = PTR_ERR_OR_ZERO(ob[nr_got]); 809 if (ret) 810 break; 811 812 ret = bch2_trans_run(c, 813 bch2_trans_mark_metadata_bucket(trans, ca, 814 ob[nr_got]->bucket, BCH_DATA_journal, 815 ca->mi.bucket_size)); 816 if (ret) { 817 bch2_open_bucket_put(c, ob[nr_got]); 818 bch_err_msg(c, ret, "marking new journal buckets"); 819 break; 820 } 821 822 bu[nr_got] = ob[nr_got]->bucket; 823 } 824 } 825 826 if (!nr_got) 827 goto err_free; 828 829 /* Don't return an error if we successfully allocated some buckets: */ 830 ret = 0; 831 832 if (c) { 833 bch2_journal_flush_all_pins(&c->journal); 834 bch2_journal_block(&c->journal); 835 mutex_lock(&c->sb_lock); 836 } 837 838 memcpy(new_buckets, ja->buckets, ja->nr * sizeof(u64)); 839 memcpy(new_bucket_seq, ja->bucket_seq, ja->nr * sizeof(u64)); 840 841 BUG_ON(ja->discard_idx > ja->nr); 842 843 pos = ja->discard_idx ?: ja->nr; 844 845 memmove(new_buckets + pos + nr_got, 846 new_buckets + pos, 847 sizeof(new_buckets[0]) * (ja->nr - pos)); 848 memmove(new_bucket_seq + pos + nr_got, 849 new_bucket_seq + pos, 850 sizeof(new_bucket_seq[0]) * (ja->nr - pos)); 851 852 for (i = 0; i < nr_got; i++) { 853 new_buckets[pos + i] = bu[i]; 854 new_bucket_seq[pos + i] = 0; 855 } 856 857 nr = ja->nr + nr_got; 858 859 ret = bch2_journal_buckets_to_sb(c, ca, new_buckets, nr); 860 if (ret) 861 goto err_unblock; 862 863 if (!new_fs) 864 bch2_write_super(c); 865 866 /* Commit: */ 867 if (c) 868 spin_lock(&c->journal.lock); 869 870 swap(new_buckets, ja->buckets); 871 swap(new_bucket_seq, ja->bucket_seq); 872 ja->nr = nr; 873 874 if (pos <= ja->discard_idx) 875 ja->discard_idx = (ja->discard_idx + nr_got) % ja->nr; 876 if (pos <= ja->dirty_idx_ondisk) 877 ja->dirty_idx_ondisk = (ja->dirty_idx_ondisk + nr_got) % ja->nr; 878 if (pos <= ja->dirty_idx) 879 ja->dirty_idx = (ja->dirty_idx + nr_got) % ja->nr; 880 if (pos <= ja->cur_idx) 881 ja->cur_idx = (ja->cur_idx + nr_got) % ja->nr; 882 883 if (c) 884 spin_unlock(&c->journal.lock); 885 err_unblock: 886 if (c) { 887 bch2_journal_unblock(&c->journal); 888 mutex_unlock(&c->sb_lock); 889 } 890 891 if (ret && !new_fs) 892 for (i = 0; i < nr_got; i++) 893 bch2_trans_run(c, 894 bch2_trans_mark_metadata_bucket(trans, ca, 895 bu[i], BCH_DATA_free, 0)); 896 err_free: 897 if (!new_fs) 898 for (i = 0; i < nr_got; i++) 899 bch2_open_bucket_put(c, ob[i]); 900 901 kfree(new_bucket_seq); 902 kfree(new_buckets); 903 kfree(ob); 904 kfree(bu); 905 return ret; 906 } 907 908 /* 909 * Allocate more journal space at runtime - not currently making use if it, but 910 * the code works: 911 */ 912 int bch2_set_nr_journal_buckets(struct bch_fs *c, struct bch_dev *ca, 913 unsigned nr) 914 { 915 struct journal_device *ja = &ca->journal; 916 struct closure cl; 917 int ret = 0; 918 919 closure_init_stack(&cl); 920 921 down_write(&c->state_lock); 922 923 /* don't handle reducing nr of buckets yet: */ 924 if (nr < ja->nr) 925 goto unlock; 926 927 while (ja->nr < nr) { 928 struct disk_reservation disk_res = { 0, 0, 0 }; 929 930 /* 931 * note: journal buckets aren't really counted as _sectors_ used yet, so 932 * we don't need the disk reservation to avoid the BUG_ON() in buckets.c 933 * when space used goes up without a reservation - but we do need the 934 * reservation to ensure we'll actually be able to allocate: 935 * 936 * XXX: that's not right, disk reservations only ensure a 937 * filesystem-wide allocation will succeed, this is a device 938 * specific allocation - we can hang here: 939 */ 940 941 ret = bch2_disk_reservation_get(c, &disk_res, 942 bucket_to_sector(ca, nr - ja->nr), 1, 0); 943 if (ret) 944 break; 945 946 ret = __bch2_set_nr_journal_buckets(ca, nr, false, &cl); 947 948 bch2_disk_reservation_put(c, &disk_res); 949 950 closure_sync(&cl); 951 952 if (ret && ret != -BCH_ERR_bucket_alloc_blocked) 953 break; 954 } 955 956 if (ret) 957 bch_err_fn(c, ret); 958 unlock: 959 up_write(&c->state_lock); 960 return ret; 961 } 962 963 int bch2_dev_journal_alloc(struct bch_dev *ca) 964 { 965 unsigned nr; 966 int ret; 967 968 if (dynamic_fault("bcachefs:add:journal_alloc")) { 969 ret = -BCH_ERR_ENOMEM_set_nr_journal_buckets; 970 goto err; 971 } 972 973 /* 1/128th of the device by default: */ 974 nr = ca->mi.nbuckets >> 7; 975 976 /* 977 * clamp journal size to 8192 buckets or 8GB (in sectors), whichever 978 * is smaller: 979 */ 980 nr = clamp_t(unsigned, nr, 981 BCH_JOURNAL_BUCKETS_MIN, 982 min(1 << 13, 983 (1 << 24) / ca->mi.bucket_size)); 984 985 ret = __bch2_set_nr_journal_buckets(ca, nr, true, NULL); 986 err: 987 if (ret) 988 bch_err_fn(ca, ret); 989 return ret; 990 } 991 992 int bch2_fs_journal_alloc(struct bch_fs *c) 993 { 994 struct bch_dev *ca; 995 unsigned i; 996 997 for_each_online_member(ca, c, i) { 998 if (ca->journal.nr) 999 continue; 1000 1001 int ret = bch2_dev_journal_alloc(ca); 1002 if (ret) { 1003 percpu_ref_put(&ca->io_ref); 1004 return ret; 1005 } 1006 } 1007 1008 return 0; 1009 } 1010 1011 /* startup/shutdown: */ 1012 1013 static bool bch2_journal_writing_to_device(struct journal *j, unsigned dev_idx) 1014 { 1015 bool ret = false; 1016 u64 seq; 1017 1018 spin_lock(&j->lock); 1019 for (seq = journal_last_unwritten_seq(j); 1020 seq <= journal_cur_seq(j) && !ret; 1021 seq++) { 1022 struct journal_buf *buf = journal_seq_to_buf(j, seq); 1023 1024 if (bch2_bkey_has_device_c(bkey_i_to_s_c(&buf->key), dev_idx)) 1025 ret = true; 1026 } 1027 spin_unlock(&j->lock); 1028 1029 return ret; 1030 } 1031 1032 void bch2_dev_journal_stop(struct journal *j, struct bch_dev *ca) 1033 { 1034 wait_event(j->wait, !bch2_journal_writing_to_device(j, ca->dev_idx)); 1035 } 1036 1037 void bch2_fs_journal_stop(struct journal *j) 1038 { 1039 bch2_journal_reclaim_stop(j); 1040 bch2_journal_flush_all_pins(j); 1041 1042 wait_event(j->wait, journal_entry_close(j)); 1043 1044 /* 1045 * Always write a new journal entry, to make sure the clock hands are up 1046 * to date (and match the superblock) 1047 */ 1048 bch2_journal_meta(j); 1049 1050 journal_quiesce(j); 1051 1052 BUG_ON(!bch2_journal_error(j) && 1053 test_bit(JOURNAL_REPLAY_DONE, &j->flags) && 1054 j->last_empty_seq != journal_cur_seq(j)); 1055 1056 cancel_delayed_work_sync(&j->write_work); 1057 } 1058 1059 int bch2_fs_journal_start(struct journal *j, u64 cur_seq) 1060 { 1061 struct bch_fs *c = container_of(j, struct bch_fs, journal); 1062 struct journal_entry_pin_list *p; 1063 struct journal_replay *i, **_i; 1064 struct genradix_iter iter; 1065 bool had_entries = false; 1066 unsigned ptr; 1067 u64 last_seq = cur_seq, nr, seq; 1068 1069 genradix_for_each_reverse(&c->journal_entries, iter, _i) { 1070 i = *_i; 1071 1072 if (!i || i->ignore) 1073 continue; 1074 1075 last_seq = le64_to_cpu(i->j.last_seq); 1076 break; 1077 } 1078 1079 nr = cur_seq - last_seq; 1080 1081 if (nr + 1 > j->pin.size) { 1082 free_fifo(&j->pin); 1083 init_fifo(&j->pin, roundup_pow_of_two(nr + 1), GFP_KERNEL); 1084 if (!j->pin.data) { 1085 bch_err(c, "error reallocating journal fifo (%llu open entries)", nr); 1086 return -BCH_ERR_ENOMEM_journal_pin_fifo; 1087 } 1088 } 1089 1090 j->replay_journal_seq = last_seq; 1091 j->replay_journal_seq_end = cur_seq; 1092 j->last_seq_ondisk = last_seq; 1093 j->flushed_seq_ondisk = cur_seq - 1; 1094 j->seq_ondisk = cur_seq - 1; 1095 j->pin.front = last_seq; 1096 j->pin.back = cur_seq; 1097 atomic64_set(&j->seq, cur_seq - 1); 1098 1099 fifo_for_each_entry_ptr(p, &j->pin, seq) 1100 journal_pin_list_init(p, 1); 1101 1102 genradix_for_each(&c->journal_entries, iter, _i) { 1103 i = *_i; 1104 1105 if (!i || i->ignore) 1106 continue; 1107 1108 seq = le64_to_cpu(i->j.seq); 1109 BUG_ON(seq >= cur_seq); 1110 1111 if (seq < last_seq) 1112 continue; 1113 1114 if (journal_entry_empty(&i->j)) 1115 j->last_empty_seq = le64_to_cpu(i->j.seq); 1116 1117 p = journal_seq_pin(j, seq); 1118 1119 p->devs.nr = 0; 1120 for (ptr = 0; ptr < i->nr_ptrs; ptr++) 1121 bch2_dev_list_add_dev(&p->devs, i->ptrs[ptr].dev); 1122 1123 had_entries = true; 1124 } 1125 1126 if (!had_entries) 1127 j->last_empty_seq = cur_seq; 1128 1129 spin_lock(&j->lock); 1130 1131 set_bit(JOURNAL_STARTED, &j->flags); 1132 j->last_flush_write = jiffies; 1133 1134 j->reservations.idx = j->reservations.unwritten_idx = journal_cur_seq(j); 1135 j->reservations.unwritten_idx++; 1136 1137 c->last_bucket_seq_cleanup = journal_cur_seq(j); 1138 1139 bch2_journal_space_available(j); 1140 spin_unlock(&j->lock); 1141 1142 return bch2_journal_reclaim_start(j); 1143 } 1144 1145 /* init/exit: */ 1146 1147 void bch2_dev_journal_exit(struct bch_dev *ca) 1148 { 1149 kfree(ca->journal.bio); 1150 kfree(ca->journal.buckets); 1151 kfree(ca->journal.bucket_seq); 1152 1153 ca->journal.bio = NULL; 1154 ca->journal.buckets = NULL; 1155 ca->journal.bucket_seq = NULL; 1156 } 1157 1158 int bch2_dev_journal_init(struct bch_dev *ca, struct bch_sb *sb) 1159 { 1160 struct journal_device *ja = &ca->journal; 1161 struct bch_sb_field_journal *journal_buckets = 1162 bch2_sb_field_get(sb, journal); 1163 struct bch_sb_field_journal_v2 *journal_buckets_v2 = 1164 bch2_sb_field_get(sb, journal_v2); 1165 unsigned i, nr_bvecs; 1166 1167 ja->nr = 0; 1168 1169 if (journal_buckets_v2) { 1170 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2); 1171 1172 for (i = 0; i < nr; i++) 1173 ja->nr += le64_to_cpu(journal_buckets_v2->d[i].nr); 1174 } else if (journal_buckets) { 1175 ja->nr = bch2_nr_journal_buckets(journal_buckets); 1176 } 1177 1178 ja->bucket_seq = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL); 1179 if (!ja->bucket_seq) 1180 return -BCH_ERR_ENOMEM_dev_journal_init; 1181 1182 nr_bvecs = DIV_ROUND_UP(JOURNAL_ENTRY_SIZE_MAX, PAGE_SIZE); 1183 1184 ca->journal.bio = bio_kmalloc(nr_bvecs, GFP_KERNEL); 1185 if (!ca->journal.bio) 1186 return -BCH_ERR_ENOMEM_dev_journal_init; 1187 1188 bio_init(ca->journal.bio, NULL, ca->journal.bio->bi_inline_vecs, nr_bvecs, 0); 1189 1190 ja->buckets = kcalloc(ja->nr, sizeof(u64), GFP_KERNEL); 1191 if (!ja->buckets) 1192 return -BCH_ERR_ENOMEM_dev_journal_init; 1193 1194 if (journal_buckets_v2) { 1195 unsigned nr = bch2_sb_field_journal_v2_nr_entries(journal_buckets_v2); 1196 unsigned j, dst = 0; 1197 1198 for (i = 0; i < nr; i++) 1199 for (j = 0; j < le64_to_cpu(journal_buckets_v2->d[i].nr); j++) 1200 ja->buckets[dst++] = 1201 le64_to_cpu(journal_buckets_v2->d[i].start) + j; 1202 } else if (journal_buckets) { 1203 for (i = 0; i < ja->nr; i++) 1204 ja->buckets[i] = le64_to_cpu(journal_buckets->buckets[i]); 1205 } 1206 1207 return 0; 1208 } 1209 1210 void bch2_fs_journal_exit(struct journal *j) 1211 { 1212 unsigned i; 1213 1214 darray_exit(&j->early_journal_entries); 1215 1216 for (i = 0; i < ARRAY_SIZE(j->buf); i++) 1217 kvpfree(j->buf[i].data, j->buf[i].buf_size); 1218 free_fifo(&j->pin); 1219 } 1220 1221 int bch2_fs_journal_init(struct journal *j) 1222 { 1223 static struct lock_class_key res_key; 1224 unsigned i; 1225 1226 spin_lock_init(&j->lock); 1227 spin_lock_init(&j->err_lock); 1228 init_waitqueue_head(&j->wait); 1229 INIT_DELAYED_WORK(&j->write_work, journal_write_work); 1230 init_waitqueue_head(&j->reclaim_wait); 1231 init_waitqueue_head(&j->pin_flush_wait); 1232 mutex_init(&j->reclaim_lock); 1233 mutex_init(&j->discard_lock); 1234 1235 lockdep_init_map(&j->res_map, "journal res", &res_key, 0); 1236 1237 atomic64_set(&j->reservations.counter, 1238 ((union journal_res_state) 1239 { .cur_entry_offset = JOURNAL_ENTRY_CLOSED_VAL }).v); 1240 1241 if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL))) 1242 return -BCH_ERR_ENOMEM_journal_pin_fifo; 1243 1244 for (i = 0; i < ARRAY_SIZE(j->buf); i++) { 1245 j->buf[i].buf_size = JOURNAL_ENTRY_SIZE_MIN; 1246 j->buf[i].data = kvpmalloc(j->buf[i].buf_size, GFP_KERNEL); 1247 if (!j->buf[i].data) 1248 return -BCH_ERR_ENOMEM_journal_buf; 1249 } 1250 1251 j->pin.front = j->pin.back = 1; 1252 return 0; 1253 } 1254 1255 /* debug: */ 1256 1257 void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j) 1258 { 1259 struct bch_fs *c = container_of(j, struct bch_fs, journal); 1260 union journal_res_state s; 1261 struct bch_dev *ca; 1262 unsigned long now = jiffies; 1263 u64 seq; 1264 unsigned i; 1265 1266 if (!out->nr_tabstops) 1267 printbuf_tabstop_push(out, 24); 1268 out->atomic++; 1269 1270 rcu_read_lock(); 1271 s = READ_ONCE(j->reservations); 1272 1273 prt_printf(out, "dirty journal entries:\t%llu/%llu\n", fifo_used(&j->pin), j->pin.size); 1274 prt_printf(out, "seq:\t\t\t%llu\n", journal_cur_seq(j)); 1275 prt_printf(out, "seq_ondisk:\t\t%llu\n", j->seq_ondisk); 1276 prt_printf(out, "last_seq:\t\t%llu\n", journal_last_seq(j)); 1277 prt_printf(out, "last_seq_ondisk:\t%llu\n", j->last_seq_ondisk); 1278 prt_printf(out, "flushed_seq_ondisk:\t%llu\n", j->flushed_seq_ondisk); 1279 prt_printf(out, "watermark:\t\t%s\n", bch2_watermarks[j->watermark]); 1280 prt_printf(out, "each entry reserved:\t%u\n", j->entry_u64s_reserved); 1281 prt_printf(out, "nr flush writes:\t%llu\n", j->nr_flush_writes); 1282 prt_printf(out, "nr noflush writes:\t%llu\n", j->nr_noflush_writes); 1283 prt_printf(out, "nr direct reclaim:\t%llu\n", j->nr_direct_reclaim); 1284 prt_printf(out, "nr background reclaim:\t%llu\n", j->nr_background_reclaim); 1285 prt_printf(out, "reclaim kicked:\t\t%u\n", j->reclaim_kicked); 1286 prt_printf(out, "reclaim runs in:\t%u ms\n", time_after(j->next_reclaim, now) 1287 ? jiffies_to_msecs(j->next_reclaim - jiffies) : 0); 1288 prt_printf(out, "current entry sectors:\t%u\n", j->cur_entry_sectors); 1289 prt_printf(out, "current entry error:\t%s\n", bch2_journal_errors[j->cur_entry_error]); 1290 prt_printf(out, "current entry:\t\t"); 1291 1292 switch (s.cur_entry_offset) { 1293 case JOURNAL_ENTRY_ERROR_VAL: 1294 prt_printf(out, "error"); 1295 break; 1296 case JOURNAL_ENTRY_CLOSED_VAL: 1297 prt_printf(out, "closed"); 1298 break; 1299 default: 1300 prt_printf(out, "%u/%u", s.cur_entry_offset, j->cur_entry_u64s); 1301 break; 1302 } 1303 1304 prt_newline(out); 1305 1306 for (seq = journal_cur_seq(j); 1307 seq >= journal_last_unwritten_seq(j); 1308 --seq) { 1309 i = seq & JOURNAL_BUF_MASK; 1310 1311 prt_printf(out, "unwritten entry:"); 1312 prt_tab(out); 1313 prt_printf(out, "%llu", seq); 1314 prt_newline(out); 1315 printbuf_indent_add(out, 2); 1316 1317 prt_printf(out, "refcount:"); 1318 prt_tab(out); 1319 prt_printf(out, "%u", journal_state_count(s, i)); 1320 prt_newline(out); 1321 1322 prt_printf(out, "sectors:"); 1323 prt_tab(out); 1324 prt_printf(out, "%u", j->buf[i].sectors); 1325 prt_newline(out); 1326 1327 prt_printf(out, "expires"); 1328 prt_tab(out); 1329 prt_printf(out, "%li jiffies", j->buf[i].expires - jiffies); 1330 prt_newline(out); 1331 1332 printbuf_indent_sub(out, 2); 1333 } 1334 1335 prt_printf(out, 1336 "replay done:\t\t%i\n", 1337 test_bit(JOURNAL_REPLAY_DONE, &j->flags)); 1338 1339 prt_printf(out, "space:\n"); 1340 prt_printf(out, "\tdiscarded\t%u:%u\n", 1341 j->space[journal_space_discarded].next_entry, 1342 j->space[journal_space_discarded].total); 1343 prt_printf(out, "\tclean ondisk\t%u:%u\n", 1344 j->space[journal_space_clean_ondisk].next_entry, 1345 j->space[journal_space_clean_ondisk].total); 1346 prt_printf(out, "\tclean\t\t%u:%u\n", 1347 j->space[journal_space_clean].next_entry, 1348 j->space[journal_space_clean].total); 1349 prt_printf(out, "\ttotal\t\t%u:%u\n", 1350 j->space[journal_space_total].next_entry, 1351 j->space[journal_space_total].total); 1352 1353 for_each_member_device_rcu(ca, c, i, 1354 &c->rw_devs[BCH_DATA_journal]) { 1355 struct journal_device *ja = &ca->journal; 1356 1357 if (!test_bit(ca->dev_idx, c->rw_devs[BCH_DATA_journal].d)) 1358 continue; 1359 1360 if (!ja->nr) 1361 continue; 1362 1363 prt_printf(out, "dev %u:\n", i); 1364 prt_printf(out, "\tnr\t\t%u\n", ja->nr); 1365 prt_printf(out, "\tbucket size\t%u\n", ca->mi.bucket_size); 1366 prt_printf(out, "\tavailable\t%u:%u\n", bch2_journal_dev_buckets_available(j, ja, journal_space_discarded), ja->sectors_free); 1367 prt_printf(out, "\tdiscard_idx\t%u\n", ja->discard_idx); 1368 prt_printf(out, "\tdirty_ondisk\t%u (seq %llu)\n", ja->dirty_idx_ondisk, ja->bucket_seq[ja->dirty_idx_ondisk]); 1369 prt_printf(out, "\tdirty_idx\t%u (seq %llu)\n", ja->dirty_idx, ja->bucket_seq[ja->dirty_idx]); 1370 prt_printf(out, "\tcur_idx\t\t%u (seq %llu)\n", ja->cur_idx, ja->bucket_seq[ja->cur_idx]); 1371 } 1372 1373 rcu_read_unlock(); 1374 1375 --out->atomic; 1376 } 1377 1378 void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j) 1379 { 1380 spin_lock(&j->lock); 1381 __bch2_journal_debug_to_text(out, j); 1382 spin_unlock(&j->lock); 1383 } 1384 1385 bool bch2_journal_seq_pins_to_text(struct printbuf *out, struct journal *j, u64 *seq) 1386 { 1387 struct journal_entry_pin_list *pin_list; 1388 struct journal_entry_pin *pin; 1389 unsigned i; 1390 1391 spin_lock(&j->lock); 1392 *seq = max(*seq, j->pin.front); 1393 1394 if (*seq >= j->pin.back) { 1395 spin_unlock(&j->lock); 1396 return true; 1397 } 1398 1399 out->atomic++; 1400 1401 pin_list = journal_seq_pin(j, *seq); 1402 1403 prt_printf(out, "%llu: count %u", *seq, atomic_read(&pin_list->count)); 1404 prt_newline(out); 1405 printbuf_indent_add(out, 2); 1406 1407 for (i = 0; i < ARRAY_SIZE(pin_list->list); i++) 1408 list_for_each_entry(pin, &pin_list->list[i], list) { 1409 prt_printf(out, "\t%px %ps", pin, pin->flush); 1410 prt_newline(out); 1411 } 1412 1413 if (!list_empty(&pin_list->flushed)) { 1414 prt_printf(out, "flushed:"); 1415 prt_newline(out); 1416 } 1417 1418 list_for_each_entry(pin, &pin_list->flushed, list) { 1419 prt_printf(out, "\t%px %ps", pin, pin->flush); 1420 prt_newline(out); 1421 } 1422 1423 printbuf_indent_sub(out, 2); 1424 1425 --out->atomic; 1426 spin_unlock(&j->lock); 1427 1428 return false; 1429 } 1430 1431 void bch2_journal_pins_to_text(struct printbuf *out, struct journal *j) 1432 { 1433 u64 seq = 0; 1434 1435 while (!bch2_journal_seq_pins_to_text(out, j, &seq)) 1436 seq++; 1437 } 1438