drivers/md/bcache/journal.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * bcache journalling code, for btree insertions
   4  *
   5  * Copyright 2012 Google, Inc.
   6  */
   7
   8 #include "bcache.h"
   9 #include "btree.h"
  10 #include "debug.h"
  11 #include "extents.h"
  12
  13 #include <trace/events/bcache.h>
  14
  15 /*
  16  * Journal replay/recovery:
  17  *
  18  * This code is all driven from run_cache_set(); we first read the journal
  19  * entries, do some other stuff, then we mark all the keys in the journal
  20  * entries (same as garbage collection would), then we replay them - reinserting
  21  * them into the cache in precisely the same order as they appear in the
  22  * journal.
  23  *
  24  * We only journal keys that go in leaf nodes, which simplifies things quite a
  25  * bit.
  26  */
  27
  28 static void journal_read_endio(struct bio *bio)
  29 {
  30         struct closure *cl = bio->bi_private;
  31
  32         closure_put(cl);
  33 }
  34
  35 static int journal_read_bucket(struct cache *ca, struct list_head *list,
  36                                unsigned int bucket_index)
  37 {
  38         struct journal_device *ja = &ca->journal;
  39         struct bio *bio = &ja->bio;
  40
  41         struct journal_replay *i;
  42         struct jset *j, *data = ca->set->journal.w[0].data;
  43         struct closure cl;
  44         unsigned int len, left, offset = 0;
  45         int ret = 0;
  46         sector_t bucket = bucket_to_sector(ca->set, ca->sb.d[bucket_index]);
  47
  48         closure_init_stack(&cl);
  49
  50         pr_debug("reading %u\n", bucket_index);
  51
  52         while (offset < ca->sb.bucket_size) {
  53 reread:         left = ca->sb.bucket_size - offset;
  54                 len = min_t(unsigned int, left, PAGE_SECTORS << JSET_BITS);
  55
  56                 bio_reset(bio, ca->bdev, REQ_OP_READ);
  57                 bio->bi_iter.bi_sector  = bucket + offset;
  58                 bio->bi_iter.bi_size    = len << 9;
  59
  60                 bio->bi_end_io  = journal_read_endio;
  61                 bio->bi_private = &cl;
  62                 bch_bio_map(bio, data);
  63
  64                 closure_bio_submit(ca->set, bio, &cl);
  65                 closure_sync(&cl);
  66
  67                 /* This function could be simpler now since we no longer write
  68                  * journal entries that overlap bucket boundaries; this means
  69                  * the start of a bucket will always have a valid journal entry
  70                  * if it has any journal entries at all.
  71                  */
  72
  73                 j = data;
  74                 while (len) {
  75                         struct list_head *where;
  76                         size_t blocks, bytes = set_bytes(j);
  77
  78                         if (j->magic != jset_magic(&ca->sb)) {
  79                                 pr_debug("%u: bad magic\n", bucket_index);
  80                                 return ret;
  81                         }
  82
  83                         if (bytes > left << 9 ||
  84                             bytes > PAGE_SIZE << JSET_BITS) {
  85                                 pr_info("%u: too big, %zu bytes, offset %u\n",
  86                                         bucket_index, bytes, offset);
  87                                 return ret;
  88                         }
  89
  90                         if (bytes > len << 9)
  91                                 goto reread;
  92
  93                         if (j->csum != csum_set(j)) {
  94                                 pr_info("%u: bad csum, %zu bytes, offset %u\n",
  95                                         bucket_index, bytes, offset);
  96                                 return ret;
  97                         }
  98
  99                         blocks = set_blocks(j, block_bytes(ca));
 100
 101                         /*
 102                          * Nodes in 'list' are in linear increasing order of
 103                          * i->j.seq, the node on head has the smallest (oldest)
 104                          * journal seq, the node on tail has the biggest
 105                          * (latest) journal seq.
 106                          */
 107
 108                         /*
 109                          * Check from the oldest jset for last_seq. If
 110                          * i->j.seq < j->last_seq, it means the oldest jset
 111                          * in list is expired and useless, remove it from
 112                          * this list. Otherwise, j is a candidate jset for
 113                          * further following checks.
 114                          */
 115                         while (!list_empty(list)) {
 116                                 i = list_first_entry(list,
 117                                         struct journal_replay, list);
 118                                 if (i->j.seq >= j->last_seq)
 119                                         break;
 120                                 list_del(&i->list);
 121                                 kfree(i);
 122                         }
 123
 124                         /* iterate list in reverse order (from latest jset) */
 125                         list_for_each_entry_reverse(i, list, list) {
 126                                 if (j->seq == i->j.seq)
 127                                         goto next_set;
 128
 129                                 /*
 130                                  * if j->seq is less than any i->j.last_seq
 131                                  * in list, j is an expired and useless jset.
 132                                  */
 133                                 if (j->seq < i->j.last_seq)
 134                                         goto next_set;
 135
 136                                 /*
 137                                  * 'where' points to first jset in list which
 138                                  * is elder then j.
 139                                  */
 140                                 if (j->seq > i->j.seq) {
 141                                         where = &i->list;
 142                                         goto add;
 143                                 }
 144                         }
 145
 146                         where = list;
 147 add:
 148                         i = kmalloc(offsetof(struct journal_replay, j) +
 149                                     bytes, GFP_KERNEL);
 150                         if (!i)
 151                                 return -ENOMEM;
 152                         unsafe_memcpy(&i->j, j, bytes,
 153                                 /* "bytes" was calculated by set_bytes() above */);
 154                         /* Add to the location after 'where' points to */
 155                         list_add(&i->list, where);
 156                         ret = 1;
 157
 158                         if (j->seq > ja->seq[bucket_index])
 159                                 ja->seq[bucket_index] = j->seq;
 160 next_set:
 161                         offset  += blocks * ca->sb.block_size;
 162                         len     -= blocks * ca->sb.block_size;
 163                         j = ((void *) j) + blocks * block_bytes(ca);
 164                 }
 165         }
 166
 167         return ret;
 168 }
 169
 170 int bch_journal_read(struct cache_set *c, struct list_head *list)
 171 {
 172 #define read_bucket(b)                                                  \
 173         ({                                                              \
 174                 ret = journal_read_bucket(ca, list, b);                 \
 175                 __set_bit(b, bitmap);                                   \
 176                 if (ret < 0)                                            \
 177                         return ret;                                     \
 178                 ret;                                                    \
 179         })
 180
 181         struct cache *ca = c->cache;
 182         int ret = 0;
 183         struct journal_device *ja = &ca->journal;
 184         DECLARE_BITMAP(bitmap, SB_JOURNAL_BUCKETS);
 185         unsigned int i, l, r, m;
 186         uint64_t seq;
 187
 188         bitmap_zero(bitmap, SB_JOURNAL_BUCKETS);
 189         pr_debug("%u journal buckets\n", ca->sb.njournal_buckets);
 190
 191         /*
 192          * Read journal buckets ordered by golden ratio hash to quickly
 193          * find a sequence of buckets with valid journal entries
 194          */
 195         for (i = 0; i < ca->sb.njournal_buckets; i++) {
 196                 /*
 197                  * We must try the index l with ZERO first for
 198                  * correctness due to the scenario that the journal
 199                  * bucket is circular buffer which might have wrapped
 200                  */
 201                 l = (i * 2654435769U) % ca->sb.njournal_buckets;
 202
 203                 if (test_bit(l, bitmap))
 204                         break;
 205
 206                 if (read_bucket(l))
 207                         goto bsearch;
 208         }
 209
 210         /*
 211          * If that fails, check all the buckets we haven't checked
 212          * already
 213          */
 214         pr_debug("falling back to linear search\n");
 215
 216         for_each_clear_bit(l, bitmap, ca->sb.njournal_buckets)
 217                 if (read_bucket(l))
 218                         goto bsearch;
 219
 220         /* no journal entries on this device? */
 221         if (l == ca->sb.njournal_buckets)
 222                 goto out;
 223 bsearch:
 224         BUG_ON(list_empty(list));
 225
 226         /* Binary search */
 227         m = l;
 228         r = find_next_bit(bitmap, ca->sb.njournal_buckets, l + 1);
 229         pr_debug("starting binary search, l %u r %u\n", l, r);
 230
 231         while (l + 1 < r) {
 232                 seq = list_entry(list->prev, struct journal_replay,
 233                                  list)->j.seq;
 234
 235                 m = (l + r) >> 1;
 236                 read_bucket(m);
 237
 238                 if (seq != list_entry(list->prev, struct journal_replay,
 239                                       list)->j.seq)
 240                         l = m;
 241                 else
 242                         r = m;
 243         }
 244
 245         /*
 246          * Read buckets in reverse order until we stop finding more
 247          * journal entries
 248          */
 249         pr_debug("finishing up: m %u njournal_buckets %u\n",
 250                  m, ca->sb.njournal_buckets);
 251         l = m;
 252
 253         while (1) {
 254                 if (!l--)
 255                         l = ca->sb.njournal_buckets - 1;
 256
 257                 if (l == m)
 258                         break;
 259
 260                 if (test_bit(l, bitmap))
 261                         continue;
 262
 263                 if (!read_bucket(l))
 264                         break;
 265         }
 266
 267         seq = 0;
 268
 269         for (i = 0; i < ca->sb.njournal_buckets; i++)
 270                 if (ja->seq[i] > seq) {
 271                         seq = ja->seq[i];
 272                         /*
 273                          * When journal_reclaim() goes to allocate for
 274                          * the first time, it'll use the bucket after
 275                          * ja->cur_idx
 276                          */
 277                         ja->cur_idx = i;
 278                         ja->last_idx = ja->discard_idx = (i + 1) %
 279                                 ca->sb.njournal_buckets;
 280
 281                 }
 282
 283 out:
 284         if (!list_empty(list))
 285                 c->journal.seq = list_entry(list->prev,
 286                                             struct journal_replay,
 287                                             list)->j.seq;
 288
 289         return 0;
 290 #undef read_bucket
 291 }
 292
 293 void bch_journal_mark(struct cache_set *c, struct list_head *list)
 294 {
 295         atomic_t p = { 0 };
 296         struct bkey *k;
 297         struct journal_replay *i;
 298         struct journal *j = &c->journal;
 299         uint64_t last = j->seq;
 300
 301         /*
 302          * journal.pin should never fill up - we never write a journal
 303          * entry when it would fill up. But if for some reason it does, we
 304          * iterate over the list in reverse order so that we can just skip that
 305          * refcount instead of bugging.
 306          */
 307
 308         list_for_each_entry_reverse(i, list, list) {
 309                 BUG_ON(last < i->j.seq);
 310                 i->pin = NULL;
 311
 312                 while (last-- != i->j.seq)
 313                         if (fifo_free(&j->pin) > 1) {
 314                                 fifo_push_front(&j->pin, p);
 315                                 atomic_set(&fifo_front(&j->pin), 0);
 316                         }
 317
 318                 if (fifo_free(&j->pin) > 1) {
 319                         fifo_push_front(&j->pin, p);
 320                         i->pin = &fifo_front(&j->pin);
 321                         atomic_set(i->pin, 1);
 322                 }
 323
 324                 for (k = i->j.start;
 325                      k < bset_bkey_last(&i->j);
 326                      k = bkey_next(k))
 327                         if (!__bch_extent_invalid(c, k)) {
 328                                 unsigned int j;
 329
 330                                 for (j = 0; j < KEY_PTRS(k); j++)
 331                                         if (ptr_available(c, k, j))
 332                                                 atomic_inc(&PTR_BUCKET(c, k, j)->pin);
 333
 334                                 bch_initial_mark_key(c, 0, k);
 335                         }
 336         }
 337 }
 338
 339 static bool is_discard_enabled(struct cache_set *s)
 340 {
 341         struct cache *ca = s->cache;
 342
 343         if (ca->discard)
 344                 return true;
 345
 346         return false;
 347 }
 348
 349 int bch_journal_replay(struct cache_set *s, struct list_head *list)
 350 {
 351         int ret = 0, keys = 0, entries = 0;
 352         struct bkey *k;
 353         struct journal_replay *i =
 354                 list_entry(list->prev, struct journal_replay, list);
 355
 356         uint64_t start = i->j.last_seq, end = i->j.seq, n = start;
 357         struct keylist keylist;
 358
 359         list_for_each_entry(i, list, list) {
 360                 BUG_ON(i->pin && atomic_read(i->pin) != 1);
 361
 362                 if (n != i->j.seq) {
 363                         if (n == start && is_discard_enabled(s))
 364                                 pr_info("journal entries %llu-%llu may be discarded! (replaying %llu-%llu)\n",
 365                                         n, i->j.seq - 1, start, end);
 366                         else {
 367                                 pr_err("journal entries %llu-%llu missing! (replaying %llu-%llu)\n",
 368                                         n, i->j.seq - 1, start, end);
 369                                 ret = -EIO;
 370                                 goto err;
 371                         }
 372                 }
 373
 374                 for (k = i->j.start;
 375                      k < bset_bkey_last(&i->j);
 376                      k = bkey_next(k)) {
 377                         trace_bcache_journal_replay_key(k);
 378
 379                         bch_keylist_init_single(&keylist, k);
 380
 381                         ret = bch_btree_insert(s, &keylist, i->pin, NULL);
 382                         if (ret)
 383                                 goto err;
 384
 385                         BUG_ON(!bch_keylist_empty(&keylist));
 386                         keys++;
 387
 388                         cond_resched();
 389                 }
 390
 391                 if (i->pin)
 392                         atomic_dec(i->pin);
 393                 n = i->j.seq + 1;
 394                 entries++;
 395         }
 396
 397         pr_info("journal replay done, %i keys in %i entries, seq %llu\n",
 398                 keys, entries, end);
 399 err:
 400         while (!list_empty(list)) {
 401                 i = list_first_entry(list, struct journal_replay, list);
 402                 list_del(&i->list);
 403                 kfree(i);
 404         }
 405
 406         return ret;
 407 }
 408
 409 void bch_journal_space_reserve(struct journal *j)
 410 {
 411         j->do_reserve = true;
 412 }
 413
 414 /* Journalling */
 415
 416 static void btree_flush_write(struct cache_set *c)
 417 {
 418         struct btree *b, *t, *btree_nodes[BTREE_FLUSH_NR];
 419         unsigned int i, nr;
 420         int ref_nr;
 421         atomic_t *fifo_front_p, *now_fifo_front_p;
 422         size_t mask;
 423
 424         if (c->journal.btree_flushing)
 425                 return;
 426
 427         spin_lock(&c->journal.flush_write_lock);
 428         if (c->journal.btree_flushing) {
 429                 spin_unlock(&c->journal.flush_write_lock);
 430                 return;
 431         }
 432         c->journal.btree_flushing = true;
 433         spin_unlock(&c->journal.flush_write_lock);
 434
 435         /* get the oldest journal entry and check its refcount */
 436         spin_lock(&c->journal.lock);
 437         fifo_front_p = &fifo_front(&c->journal.pin);
 438         ref_nr = atomic_read(fifo_front_p);
 439         if (ref_nr <= 0) {
 440                 /*
 441                  * do nothing if no btree node references
 442                  * the oldest journal entry
 443                  */
 444                 spin_unlock(&c->journal.lock);
 445                 goto out;
 446         }
 447         spin_unlock(&c->journal.lock);
 448
 449         mask = c->journal.pin.mask;
 450         nr = 0;
 451         atomic_long_inc(&c->flush_write);
 452         memset(btree_nodes, 0, sizeof(btree_nodes));
 453
 454         mutex_lock(&c->bucket_lock);
 455         list_for_each_entry_safe_reverse(b, t, &c->btree_cache, list) {
 456                 /*
 457                  * It is safe to get now_fifo_front_p without holding
 458                  * c->journal.lock here, because we don't need to know
 459                  * the exactly accurate value, just check whether the
 460                  * front pointer of c->journal.pin is changed.
 461                  */
 462                 now_fifo_front_p = &fifo_front(&c->journal.pin);
 463                 /*
 464                  * If the oldest journal entry is reclaimed and front
 465                  * pointer of c->journal.pin changes, it is unnecessary
 466                  * to scan c->btree_cache anymore, just quit the loop and
 467                  * flush out what we have already.
 468                  */
 469                 if (now_fifo_front_p != fifo_front_p)
 470                         break;
 471                 /*
 472                  * quit this loop if all matching btree nodes are
 473                  * scanned and record in btree_nodes[] already.
 474                  */
 475                 ref_nr = atomic_read(fifo_front_p);
 476                 if (nr >= ref_nr)
 477                         break;
 478
 479                 if (btree_node_journal_flush(b))
 480                         pr_err("BUG: flush_write bit should not be set here!\n");
 481
 482                 mutex_lock(&b->write_lock);
 483
 484                 if (!btree_node_dirty(b)) {
 485                         mutex_unlock(&b->write_lock);
 486                         continue;
 487                 }
 488
 489                 if (!btree_current_write(b)->journal) {
 490                         mutex_unlock(&b->write_lock);
 491                         continue;
 492                 }
 493
 494                 /*
 495                  * Only select the btree node which exactly references
 496                  * the oldest journal entry.
 497                  *
 498                  * If the journal entry pointed by fifo_front_p is
 499                  * reclaimed in parallel, don't worry:
 500                  * - the list_for_each_xxx loop will quit when checking
 501                  *   next now_fifo_front_p.
 502                  * - If there are matched nodes recorded in btree_nodes[],
 503                  *   they are clean now (this is why and how the oldest
 504                  *   journal entry can be reclaimed). These selected nodes
 505                  *   will be ignored and skipped in the following for-loop.
 506                  */
 507                 if (((btree_current_write(b)->journal - fifo_front_p) &
 508                      mask) != 0) {
 509                         mutex_unlock(&b->write_lock);
 510                         continue;
 511                 }
 512
 513                 set_btree_node_journal_flush(b);
 514
 515                 mutex_unlock(&b->write_lock);
 516
 517                 btree_nodes[nr++] = b;
 518                 /*
 519                  * To avoid holding c->bucket_lock too long time,
 520                  * only scan for BTREE_FLUSH_NR matched btree nodes
 521                  * at most. If there are more btree nodes reference
 522                  * the oldest journal entry, try to flush them next
 523                  * time when btree_flush_write() is called.
 524                  */
 525                 if (nr == BTREE_FLUSH_NR)
 526                         break;
 527         }
 528         mutex_unlock(&c->bucket_lock);
 529
 530         for (i = 0; i < nr; i++) {
 531                 b = btree_nodes[i];
 532                 if (!b) {
 533                         pr_err("BUG: btree_nodes[%d] is NULL\n", i);
 534                         continue;
 535                 }
 536
 537                 /* safe to check without holding b->write_lock */
 538                 if (!btree_node_journal_flush(b)) {
 539                         pr_err("BUG: bnode %p: journal_flush bit cleaned\n", b);
 540                         continue;
 541                 }
 542
 543                 mutex_lock(&b->write_lock);
 544                 if (!btree_current_write(b)->journal) {
 545                         clear_bit(BTREE_NODE_journal_flush, &b->flags);
 546                         mutex_unlock(&b->write_lock);
 547                         pr_debug("bnode %p: written by others\n", b);
 548                         continue;
 549                 }
 550
 551                 if (!btree_node_dirty(b)) {
 552                         clear_bit(BTREE_NODE_journal_flush, &b->flags);
 553                         mutex_unlock(&b->write_lock);
 554                         pr_debug("bnode %p: dirty bit cleaned by others\n", b);
 555                         continue;
 556                 }
 557
 558                 __bch_btree_node_write(b, NULL);
 559                 clear_bit(BTREE_NODE_journal_flush, &b->flags);
 560                 mutex_unlock(&b->write_lock);
 561         }
 562
 563 out:
 564         spin_lock(&c->journal.flush_write_lock);
 565         c->journal.btree_flushing = false;
 566         spin_unlock(&c->journal.flush_write_lock);
 567 }
 568
 569 #define last_seq(j)     ((j)->seq - fifo_used(&(j)->pin) + 1)
 570
 571 static void journal_discard_endio(struct bio *bio)
 572 {
 573         struct journal_device *ja =
 574                 container_of(bio, struct journal_device, discard_bio);
 575         struct cache *ca = container_of(ja, struct cache, journal);
 576
 577         atomic_set(&ja->discard_in_flight, DISCARD_DONE);
 578
 579         closure_wake_up(&ca->set->journal.wait);
 580         closure_put(&ca->set->cl);
 581 }
 582
 583 static void journal_discard_work(struct work_struct *work)
 584 {
 585         struct journal_device *ja =
 586                 container_of(work, struct journal_device, discard_work);
 587
 588         submit_bio(&ja->discard_bio);
 589 }
 590
 591 static void do_journal_discard(struct cache *ca)
 592 {
 593         struct journal_device *ja = &ca->journal;
 594         struct bio *bio = &ja->discard_bio;
 595
 596         if (!ca->discard) {
 597                 ja->discard_idx = ja->last_idx;
 598                 return;
 599         }
 600
 601         switch (atomic_read(&ja->discard_in_flight)) {
 602         case DISCARD_IN_FLIGHT:
 603                 return;
 604
 605         case DISCARD_DONE:
 606                 ja->discard_idx = (ja->discard_idx + 1) %
 607                         ca->sb.njournal_buckets;
 608
 609                 atomic_set(&ja->discard_in_flight, DISCARD_READY);
 610                 fallthrough;
 611
 612         case DISCARD_READY:
 613                 if (ja->discard_idx == ja->last_idx)
 614                         return;
 615
 616                 atomic_set(&ja->discard_in_flight, DISCARD_IN_FLIGHT);
 617
 618                 bio_init(bio, ca->bdev, bio->bi_inline_vecs, 1, REQ_OP_DISCARD);
 619                 bio->bi_iter.bi_sector  = bucket_to_sector(ca->set,
 620                                                 ca->sb.d[ja->discard_idx]);
 621                 bio->bi_iter.bi_size    = bucket_bytes(ca);
 622                 bio->bi_end_io          = journal_discard_endio;
 623
 624                 closure_get(&ca->set->cl);
 625                 INIT_WORK(&ja->discard_work, journal_discard_work);
 626                 queue_work(bch_journal_wq, &ja->discard_work);
 627         }
 628 }
 629
 630 static unsigned int free_journal_buckets(struct cache_set *c)
 631 {
 632         struct journal *j = &c->journal;
 633         struct cache *ca = c->cache;
 634         struct journal_device *ja = &c->cache->journal;
 635         unsigned int n;
 636
 637         /* In case njournal_buckets is not power of 2 */
 638         if (ja->cur_idx >= ja->discard_idx)
 639                 n = ca->sb.njournal_buckets +  ja->discard_idx - ja->cur_idx;
 640         else
 641                 n = ja->discard_idx - ja->cur_idx;
 642
 643         if (n > (1 + j->do_reserve))
 644                 return n - (1 + j->do_reserve);
 645
 646         return 0;
 647 }
 648
 649 static void journal_reclaim(struct cache_set *c)
 650 {
 651         struct bkey *k = &c->journal.key;
 652         struct cache *ca = c->cache;
 653         uint64_t last_seq;
 654         struct journal_device *ja = &ca->journal;
 655         atomic_t p __maybe_unused;
 656
 657         atomic_long_inc(&c->reclaim);
 658
 659         while (!atomic_read(&fifo_front(&c->journal.pin)))
 660                 fifo_pop(&c->journal.pin, p);
 661
 662         last_seq = last_seq(&c->journal);
 663
 664         /* Update last_idx */
 665
 666         while (ja->last_idx != ja->cur_idx &&
 667                ja->seq[ja->last_idx] < last_seq)
 668                 ja->last_idx = (ja->last_idx + 1) %
 669                         ca->sb.njournal_buckets;
 670
 671         do_journal_discard(ca);
 672
 673         if (c->journal.blocks_free)
 674                 goto out;
 675
 676         if (!free_journal_buckets(c))
 677                 goto out;
 678
 679         ja->cur_idx = (ja->cur_idx + 1) % ca->sb.njournal_buckets;
 680         k->ptr[0] = MAKE_PTR(0,
 681                              bucket_to_sector(c, ca->sb.d[ja->cur_idx]),
 682                              ca->sb.nr_this_dev);
 683         atomic_long_inc(&c->reclaimed_journal_buckets);
 684
 685         bkey_init(k);
 686         SET_KEY_PTRS(k, 1);
 687         c->journal.blocks_free = ca->sb.bucket_size >> c->block_bits;
 688
 689 out:
 690         if (!journal_full(&c->journal))
 691                 __closure_wake_up(&c->journal.wait);
 692 }
 693
 694 void bch_journal_next(struct journal *j)
 695 {
 696         atomic_t p = { 1 };
 697
 698         j->cur = (j->cur == j->w)
 699                 ? &j->w[1]
 700                 : &j->w[0];
 701
 702         /*
 703          * The fifo_push() needs to happen at the same time as j->seq is
 704          * incremented for last_seq() to be calculated correctly
 705          */
 706         BUG_ON(!fifo_push(&j->pin, p));
 707         atomic_set(&fifo_back(&j->pin), 1);
 708
 709         j->cur->data->seq       = ++j->seq;
 710         j->cur->dirty           = false;
 711         j->cur->need_write      = false;
 712         j->cur->data->keys      = 0;
 713
 714         if (fifo_full(&j->pin))
 715                 pr_debug("journal_pin full (%zu)\n", fifo_used(&j->pin));
 716 }
 717
 718 static void journal_write_endio(struct bio *bio)
 719 {
 720         struct journal_write *w = bio->bi_private;
 721
 722         cache_set_err_on(bio->bi_status, w->c, "journal io error");
 723         closure_put(&w->c->journal.io);
 724 }
 725
 726 static CLOSURE_CALLBACK(journal_write);
 727
 728 static CLOSURE_CALLBACK(journal_write_done)
 729 {
 730         closure_type(j, struct journal, io);
 731         struct journal_write *w = (j->cur == j->w)
 732                 ? &j->w[1]
 733                 : &j->w[0];
 734
 735         __closure_wake_up(&w->wait);
 736         continue_at_nobarrier(cl, journal_write, bch_journal_wq);
 737 }
 738
 739 static CLOSURE_CALLBACK(journal_write_unlock)
 740         __releases(&c->journal.lock)
 741 {
 742         closure_type(c, struct cache_set, journal.io);
 743
 744         c->journal.io_in_flight = 0;
 745         spin_unlock(&c->journal.lock);
 746 }
 747
 748 static CLOSURE_CALLBACK(journal_write_unlocked)
 749         __releases(c->journal.lock)
 750 {
 751         closure_type(c, struct cache_set, journal.io);
 752         struct cache *ca = c->cache;
 753         struct journal_write *w = c->journal.cur;
 754         struct bkey *k = &c->journal.key;
 755         unsigned int i, sectors = set_blocks(w->data, block_bytes(ca)) *
 756                 ca->sb.block_size;
 757
 758         struct bio *bio;
 759         struct bio_list list;
 760
 761         bio_list_init(&list);
 762
 763         if (!w->need_write) {
 764                 closure_return_with_destructor(cl, journal_write_unlock);
 765                 return;
 766         } else if (journal_full(&c->journal)) {
 767                 journal_reclaim(c);
 768                 spin_unlock(&c->journal.lock);
 769
 770                 btree_flush_write(c);
 771                 continue_at(cl, journal_write, bch_journal_wq);
 772                 return;
 773         }
 774
 775         c->journal.blocks_free -= set_blocks(w->data, block_bytes(ca));
 776
 777         w->data->btree_level = c->root->level;
 778
 779         bkey_copy(&w->data->btree_root, &c->root->key);
 780         bkey_copy(&w->data->uuid_bucket, &c->uuid_bucket);
 781
 782         w->data->prio_bucket[ca->sb.nr_this_dev] = ca->prio_buckets[0];
 783         w->data->magic          = jset_magic(&ca->sb);
 784         w->data->version        = BCACHE_JSET_VERSION;
 785         w->data->last_seq       = last_seq(&c->journal);
 786         w->data->csum           = csum_set(w->data);
 787
 788         for (i = 0; i < KEY_PTRS(k); i++) {
 789                 ca = c->cache;
 790                 bio = &ca->journal.bio;
 791
 792                 atomic_long_add(sectors, &ca->meta_sectors_written);
 793
 794                 bio_reset(bio, ca->bdev, REQ_OP_WRITE |
 795                           REQ_SYNC | REQ_META | REQ_PREFLUSH | REQ_FUA);
 796                 bio->bi_iter.bi_sector  = PTR_OFFSET(k, i);
 797                 bio->bi_iter.bi_size = sectors << 9;
 798
 799                 bio->bi_end_io  = journal_write_endio;
 800                 bio->bi_private = w;
 801                 bch_bio_map(bio, w->data);
 802
 803                 trace_bcache_journal_write(bio, w->data->keys);
 804                 bio_list_add(&list, bio);
 805
 806                 SET_PTR_OFFSET(k, i, PTR_OFFSET(k, i) + sectors);
 807
 808                 ca->journal.seq[ca->journal.cur_idx] = w->data->seq;
 809         }
 810
 811         /* If KEY_PTRS(k) == 0, this jset gets lost in air */
 812         BUG_ON(i == 0);
 813
 814         atomic_dec_bug(&fifo_back(&c->journal.pin));
 815         bch_journal_next(&c->journal);
 816         journal_reclaim(c);
 817
 818         spin_unlock(&c->journal.lock);
 819
 820         while ((bio = bio_list_pop(&list)))
 821                 closure_bio_submit(c, bio, cl);
 822
 823         continue_at(cl, journal_write_done, NULL);
 824 }
 825
 826 static CLOSURE_CALLBACK(journal_write)
 827 {
 828         closure_type(c, struct cache_set, journal.io);
 829
 830         spin_lock(&c->journal.lock);
 831         journal_write_unlocked(&cl->work);
 832 }
 833
 834 static void journal_try_write(struct cache_set *c)
 835         __releases(c->journal.lock)
 836 {
 837         struct closure *cl = &c->journal.io;
 838         struct journal_write *w = c->journal.cur;
 839
 840         w->need_write = true;
 841
 842         if (!c->journal.io_in_flight) {
 843                 c->journal.io_in_flight = 1;
 844                 closure_call(cl, journal_write_unlocked, NULL, &c->cl);
 845         } else {
 846                 spin_unlock(&c->journal.lock);
 847         }
 848 }
 849
 850 static struct journal_write *journal_wait_for_write(struct cache_set *c,
 851                                                     unsigned int nkeys)
 852         __acquires(&c->journal.lock)
 853 {
 854         size_t sectors;
 855         struct closure cl;
 856         bool wait = false;
 857         struct cache *ca = c->cache;
 858
 859         closure_init_stack(&cl);
 860
 861         spin_lock(&c->journal.lock);
 862
 863         while (1) {
 864                 struct journal_write *w = c->journal.cur;
 865
 866                 sectors = __set_blocks(w->data, w->data->keys + nkeys,
 867                                        block_bytes(ca)) * ca->sb.block_size;
 868
 869                 if (sectors <= min_t(size_t,
 870                                      c->journal.blocks_free * ca->sb.block_size,
 871                                      PAGE_SECTORS << JSET_BITS))
 872                         return w;
 873
 874                 if (wait)
 875                         closure_wait(&c->journal.wait, &cl);
 876
 877                 if (!journal_full(&c->journal)) {
 878                         if (wait)
 879                                 trace_bcache_journal_entry_full(c);
 880
 881                         /*
 882                          * XXX: If we were inserting so many keys that they
 883                          * won't fit in an _empty_ journal write, we'll
 884                          * deadlock. For now, handle this in
 885                          * bch_keylist_realloc() - but something to think about.
 886                          */
 887                         BUG_ON(!w->data->keys);
 888
 889                         journal_try_write(c); /* unlocks */
 890                 } else {
 891                         if (wait)
 892                                 trace_bcache_journal_full(c);
 893
 894                         journal_reclaim(c);
 895                         spin_unlock(&c->journal.lock);
 896
 897                         btree_flush_write(c);
 898                 }
 899
 900                 closure_sync(&cl);
 901                 spin_lock(&c->journal.lock);
 902                 wait = true;
 903         }
 904 }
 905
 906 static void journal_write_work(struct work_struct *work)
 907 {
 908         struct cache_set *c = container_of(to_delayed_work(work),
 909                                            struct cache_set,
 910                                            journal.work);
 911         spin_lock(&c->journal.lock);
 912         if (c->journal.cur->dirty)
 913                 journal_try_write(c);
 914         else
 915                 spin_unlock(&c->journal.lock);
 916 }
 917
 918 /*
 919  * Entry point to the journalling code - bio_insert() and btree_invalidate()
 920  * pass bch_journal() a list of keys to be journalled, and then
 921  * bch_journal() hands those same keys off to btree_insert_async()
 922  */
 923
 924 atomic_t *bch_journal(struct cache_set *c,
 925                       struct keylist *keys,
 926                       struct closure *parent)
 927 {
 928         struct journal_write *w;
 929         atomic_t *ret;
 930
 931         /* No journaling if CACHE_SET_IO_DISABLE set already */
 932         if (unlikely(test_bit(CACHE_SET_IO_DISABLE, &c->flags)))
 933                 return NULL;
 934
 935         if (!CACHE_SYNC(&c->cache->sb))
 936                 return NULL;
 937
 938         w = journal_wait_for_write(c, bch_keylist_nkeys(keys));
 939
 940         memcpy(bset_bkey_last(w->data), keys->keys, bch_keylist_bytes(keys));
 941         w->data->keys += bch_keylist_nkeys(keys);
 942
 943         ret = &fifo_back(&c->journal.pin);
 944         atomic_inc(ret);
 945
 946         if (parent) {
 947                 closure_wait(&w->wait, parent);
 948                 journal_try_write(c);
 949         } else if (!w->dirty) {
 950                 w->dirty = true;
 951                 queue_delayed_work(bch_flush_wq, &c->journal.work,
 952                                    msecs_to_jiffies(c->journal_delay_ms));
 953                 spin_unlock(&c->journal.lock);
 954         } else {
 955                 spin_unlock(&c->journal.lock);
 956         }
 957
 958
 959         return ret;
 960 }
 961
 962 void bch_journal_meta(struct cache_set *c, struct closure *cl)
 963 {
 964         struct keylist keys;
 965         atomic_t *ref;
 966
 967         bch_keylist_init(&keys);
 968
 969         ref = bch_journal(c, &keys, cl);
 970         if (ref)
 971                 atomic_dec_bug(ref);
 972 }
 973
 974 void bch_journal_free(struct cache_set *c)
 975 {
 976         free_pages((unsigned long) c->journal.w[1].data, JSET_BITS);
 977         free_pages((unsigned long) c->journal.w[0].data, JSET_BITS);
 978         free_fifo(&c->journal.pin);
 979 }
 980
 981 int bch_journal_alloc(struct cache_set *c)
 982 {
 983         struct journal *j = &c->journal;
 984
 985         spin_lock_init(&j->lock);
 986         spin_lock_init(&j->flush_write_lock);
 987         INIT_DELAYED_WORK(&j->work, journal_write_work);
 988
 989         c->journal_delay_ms = 100;
 990
 991         j->w[0].c = c;
 992         j->w[1].c = c;
 993
 994         if (!(init_fifo(&j->pin, JOURNAL_PIN, GFP_KERNEL)) ||
 995             !(j->w[0].data = (void *) __get_free_pages(GFP_KERNEL|__GFP_COMP, JSET_BITS)) ||
 996             !(j->w[1].data = (void *) __get_free_pages(GFP_KERNEL|__GFP_COMP, JSET_BITS)))
 997                 return -ENOMEM;
 998
 999         return 0;
1000 }