1 /* 2 * Primary bucket allocation code 3 * 4 * Copyright 2012 Google, Inc. 5 * 6 * Allocation in bcache is done in terms of buckets: 7 * 8 * Each bucket has associated an 8 bit gen; this gen corresponds to the gen in 9 * btree pointers - they must match for the pointer to be considered valid. 10 * 11 * Thus (assuming a bucket has no dirty data or metadata in it) we can reuse a 12 * bucket simply by incrementing its gen. 13 * 14 * The gens (along with the priorities; it's really the gens are important but 15 * the code is named as if it's the priorities) are written in an arbitrary list 16 * of buckets on disk, with a pointer to them in the journal header. 17 * 18 * When we invalidate a bucket, we have to write its new gen to disk and wait 19 * for that write to complete before we use it - otherwise after a crash we 20 * could have pointers that appeared to be good but pointed to data that had 21 * been overwritten. 22 * 23 * Since the gens and priorities are all stored contiguously on disk, we can 24 * batch this up: We fill up the free_inc list with freshly invalidated buckets, 25 * call prio_write(), and when prio_write() finishes we pull buckets off the 26 * free_inc list and optionally discard them. 27 * 28 * free_inc isn't the only freelist - if it was, we'd often to sleep while 29 * priorities and gens were being written before we could allocate. c->free is a 30 * smaller freelist, and buckets on that list are always ready to be used. 31 * 32 * If we've got discards enabled, that happens when a bucket moves from the 33 * free_inc list to the free list. 34 * 35 * There is another freelist, because sometimes we have buckets that we know 36 * have nothing pointing into them - these we can reuse without waiting for 37 * priorities to be rewritten. These come from freed btree nodes and buckets 38 * that garbage collection discovered no longer had valid keys pointing into 39 * them (because they were overwritten). That's the unused list - buckets on the 40 * unused list move to the free list, optionally being discarded in the process. 41 * 42 * It's also important to ensure that gens don't wrap around - with respect to 43 * either the oldest gen in the btree or the gen on disk. This is quite 44 * difficult to do in practice, but we explicitly guard against it anyways - if 45 * a bucket is in danger of wrapping around we simply skip invalidating it that 46 * time around, and we garbage collect or rewrite the priorities sooner than we 47 * would have otherwise. 48 * 49 * bch_bucket_alloc() allocates a single bucket from a specific cache. 50 * 51 * bch_bucket_alloc_set() allocates one or more buckets from different caches 52 * out of a cache set. 53 * 54 * free_some_buckets() drives all the processes described above. It's called 55 * from bch_bucket_alloc() and a few other places that need to make sure free 56 * buckets are ready. 57 * 58 * invalidate_buckets_(lru|fifo)() find buckets that are available to be 59 * invalidated, and then invalidate them and stick them on the free_inc list - 60 * in either lru or fifo order. 61 */ 62 63 #include "bcache.h" 64 #include "btree.h" 65 66 #include <linux/blkdev.h> 67 #include <linux/freezer.h> 68 #include <linux/kthread.h> 69 #include <linux/random.h> 70 #include <trace/events/bcache.h> 71 72 /* Bucket heap / gen */ 73 74 uint8_t bch_inc_gen(struct cache *ca, struct bucket *b) 75 { 76 uint8_t ret = ++b->gen; 77 78 ca->set->need_gc = max(ca->set->need_gc, bucket_gc_gen(b)); 79 WARN_ON_ONCE(ca->set->need_gc > BUCKET_GC_GEN_MAX); 80 81 if (CACHE_SYNC(&ca->set->sb)) { 82 ca->need_save_prio = max(ca->need_save_prio, 83 bucket_disk_gen(b)); 84 WARN_ON_ONCE(ca->need_save_prio > BUCKET_DISK_GEN_MAX); 85 } 86 87 return ret; 88 } 89 90 void bch_rescale_priorities(struct cache_set *c, int sectors) 91 { 92 struct cache *ca; 93 struct bucket *b; 94 unsigned next = c->nbuckets * c->sb.bucket_size / 1024; 95 unsigned i; 96 int r; 97 98 atomic_sub(sectors, &c->rescale); 99 100 do { 101 r = atomic_read(&c->rescale); 102 103 if (r >= 0) 104 return; 105 } while (atomic_cmpxchg(&c->rescale, r, r + next) != r); 106 107 mutex_lock(&c->bucket_lock); 108 109 c->min_prio = USHRT_MAX; 110 111 for_each_cache(ca, c, i) 112 for_each_bucket(b, ca) 113 if (b->prio && 114 b->prio != BTREE_PRIO && 115 !atomic_read(&b->pin)) { 116 b->prio--; 117 c->min_prio = min(c->min_prio, b->prio); 118 } 119 120 mutex_unlock(&c->bucket_lock); 121 } 122 123 /* Allocation */ 124 125 static inline bool can_inc_bucket_gen(struct bucket *b) 126 { 127 return bucket_gc_gen(b) < BUCKET_GC_GEN_MAX && 128 bucket_disk_gen(b) < BUCKET_DISK_GEN_MAX; 129 } 130 131 bool bch_bucket_add_unused(struct cache *ca, struct bucket *b) 132 { 133 BUG_ON(GC_MARK(b) || GC_SECTORS_USED(b)); 134 135 if (fifo_used(&ca->free) > ca->watermark[WATERMARK_MOVINGGC] && 136 CACHE_REPLACEMENT(&ca->sb) == CACHE_REPLACEMENT_FIFO) 137 return false; 138 139 b->prio = 0; 140 141 if (can_inc_bucket_gen(b) && 142 fifo_push(&ca->unused, b - ca->buckets)) { 143 atomic_inc(&b->pin); 144 return true; 145 } 146 147 return false; 148 } 149 150 static bool can_invalidate_bucket(struct cache *ca, struct bucket *b) 151 { 152 return GC_MARK(b) == GC_MARK_RECLAIMABLE && 153 !atomic_read(&b->pin) && 154 can_inc_bucket_gen(b); 155 } 156 157 static void invalidate_one_bucket(struct cache *ca, struct bucket *b) 158 { 159 bch_inc_gen(ca, b); 160 b->prio = INITIAL_PRIO; 161 atomic_inc(&b->pin); 162 fifo_push(&ca->free_inc, b - ca->buckets); 163 } 164 165 #define bucket_prio(b) \ 166 (((unsigned) (b->prio - ca->set->min_prio)) * GC_SECTORS_USED(b)) 167 168 #define bucket_max_cmp(l, r) (bucket_prio(l) < bucket_prio(r)) 169 #define bucket_min_cmp(l, r) (bucket_prio(l) > bucket_prio(r)) 170 171 static void invalidate_buckets_lru(struct cache *ca) 172 { 173 struct bucket *b; 174 ssize_t i; 175 176 ca->heap.used = 0; 177 178 for_each_bucket(b, ca) { 179 /* 180 * If we fill up the unused list, if we then return before 181 * adding anything to the free_inc list we'll skip writing 182 * prios/gens and just go back to allocating from the unused 183 * list: 184 */ 185 if (fifo_full(&ca->unused)) 186 return; 187 188 if (!can_invalidate_bucket(ca, b)) 189 continue; 190 191 if (!GC_SECTORS_USED(b) && 192 bch_bucket_add_unused(ca, b)) 193 continue; 194 195 if (!heap_full(&ca->heap)) 196 heap_add(&ca->heap, b, bucket_max_cmp); 197 else if (bucket_max_cmp(b, heap_peek(&ca->heap))) { 198 ca->heap.data[0] = b; 199 heap_sift(&ca->heap, 0, bucket_max_cmp); 200 } 201 } 202 203 for (i = ca->heap.used / 2 - 1; i >= 0; --i) 204 heap_sift(&ca->heap, i, bucket_min_cmp); 205 206 while (!fifo_full(&ca->free_inc)) { 207 if (!heap_pop(&ca->heap, b, bucket_min_cmp)) { 208 /* 209 * We don't want to be calling invalidate_buckets() 210 * multiple times when it can't do anything 211 */ 212 ca->invalidate_needs_gc = 1; 213 wake_up_gc(ca->set); 214 return; 215 } 216 217 invalidate_one_bucket(ca, b); 218 } 219 } 220 221 static void invalidate_buckets_fifo(struct cache *ca) 222 { 223 struct bucket *b; 224 size_t checked = 0; 225 226 while (!fifo_full(&ca->free_inc)) { 227 if (ca->fifo_last_bucket < ca->sb.first_bucket || 228 ca->fifo_last_bucket >= ca->sb.nbuckets) 229 ca->fifo_last_bucket = ca->sb.first_bucket; 230 231 b = ca->buckets + ca->fifo_last_bucket++; 232 233 if (can_invalidate_bucket(ca, b)) 234 invalidate_one_bucket(ca, b); 235 236 if (++checked >= ca->sb.nbuckets) { 237 ca->invalidate_needs_gc = 1; 238 wake_up_gc(ca->set); 239 return; 240 } 241 } 242 } 243 244 static void invalidate_buckets_random(struct cache *ca) 245 { 246 struct bucket *b; 247 size_t checked = 0; 248 249 while (!fifo_full(&ca->free_inc)) { 250 size_t n; 251 get_random_bytes(&n, sizeof(n)); 252 253 n %= (size_t) (ca->sb.nbuckets - ca->sb.first_bucket); 254 n += ca->sb.first_bucket; 255 256 b = ca->buckets + n; 257 258 if (can_invalidate_bucket(ca, b)) 259 invalidate_one_bucket(ca, b); 260 261 if (++checked >= ca->sb.nbuckets / 2) { 262 ca->invalidate_needs_gc = 1; 263 wake_up_gc(ca->set); 264 return; 265 } 266 } 267 } 268 269 static void invalidate_buckets(struct cache *ca) 270 { 271 if (ca->invalidate_needs_gc) 272 return; 273 274 switch (CACHE_REPLACEMENT(&ca->sb)) { 275 case CACHE_REPLACEMENT_LRU: 276 invalidate_buckets_lru(ca); 277 break; 278 case CACHE_REPLACEMENT_FIFO: 279 invalidate_buckets_fifo(ca); 280 break; 281 case CACHE_REPLACEMENT_RANDOM: 282 invalidate_buckets_random(ca); 283 break; 284 } 285 286 trace_bcache_alloc_invalidate(ca); 287 } 288 289 #define allocator_wait(ca, cond) \ 290 do { \ 291 while (1) { \ 292 set_current_state(TASK_INTERRUPTIBLE); \ 293 if (cond) \ 294 break; \ 295 \ 296 mutex_unlock(&(ca)->set->bucket_lock); \ 297 if (kthread_should_stop()) \ 298 return 0; \ 299 \ 300 try_to_freeze(); \ 301 schedule(); \ 302 mutex_lock(&(ca)->set->bucket_lock); \ 303 } \ 304 __set_current_state(TASK_RUNNING); \ 305 } while (0) 306 307 static int bch_allocator_thread(void *arg) 308 { 309 struct cache *ca = arg; 310 311 mutex_lock(&ca->set->bucket_lock); 312 313 while (1) { 314 /* 315 * First, we pull buckets off of the unused and free_inc lists, 316 * possibly issue discards to them, then we add the bucket to 317 * the free list: 318 */ 319 while (1) { 320 long bucket; 321 322 if ((!atomic_read(&ca->set->prio_blocked) || 323 !CACHE_SYNC(&ca->set->sb)) && 324 !fifo_empty(&ca->unused)) 325 fifo_pop(&ca->unused, bucket); 326 else if (!fifo_empty(&ca->free_inc)) 327 fifo_pop(&ca->free_inc, bucket); 328 else 329 break; 330 331 if (ca->discard) { 332 mutex_unlock(&ca->set->bucket_lock); 333 blkdev_issue_discard(ca->bdev, 334 bucket_to_sector(ca->set, bucket), 335 ca->sb.block_size, GFP_KERNEL, 0); 336 mutex_lock(&ca->set->bucket_lock); 337 } 338 339 allocator_wait(ca, !fifo_full(&ca->free)); 340 341 fifo_push(&ca->free, bucket); 342 wake_up(&ca->set->bucket_wait); 343 } 344 345 /* 346 * We've run out of free buckets, we need to find some buckets 347 * we can invalidate. First, invalidate them in memory and add 348 * them to the free_inc list: 349 */ 350 351 allocator_wait(ca, ca->set->gc_mark_valid && 352 (ca->need_save_prio > 64 || 353 !ca->invalidate_needs_gc)); 354 invalidate_buckets(ca); 355 356 /* 357 * Now, we write their new gens to disk so we can start writing 358 * new stuff to them: 359 */ 360 allocator_wait(ca, !atomic_read(&ca->set->prio_blocked)); 361 if (CACHE_SYNC(&ca->set->sb) && 362 (!fifo_empty(&ca->free_inc) || 363 ca->need_save_prio > 64)) 364 bch_prio_write(ca); 365 } 366 } 367 368 long bch_bucket_alloc(struct cache *ca, unsigned watermark, bool wait) 369 { 370 DEFINE_WAIT(w); 371 struct bucket *b; 372 long r; 373 374 /* fastpath */ 375 if (fifo_used(&ca->free) > ca->watermark[watermark]) { 376 fifo_pop(&ca->free, r); 377 goto out; 378 } 379 380 if (!wait) 381 return -1; 382 383 while (1) { 384 if (fifo_used(&ca->free) > ca->watermark[watermark]) { 385 fifo_pop(&ca->free, r); 386 break; 387 } 388 389 prepare_to_wait(&ca->set->bucket_wait, &w, 390 TASK_UNINTERRUPTIBLE); 391 392 mutex_unlock(&ca->set->bucket_lock); 393 schedule(); 394 mutex_lock(&ca->set->bucket_lock); 395 } 396 397 finish_wait(&ca->set->bucket_wait, &w); 398 out: 399 wake_up_process(ca->alloc_thread); 400 401 if (expensive_debug_checks(ca->set)) { 402 size_t iter; 403 long i; 404 405 for (iter = 0; iter < prio_buckets(ca) * 2; iter++) 406 BUG_ON(ca->prio_buckets[iter] == (uint64_t) r); 407 408 fifo_for_each(i, &ca->free, iter) 409 BUG_ON(i == r); 410 fifo_for_each(i, &ca->free_inc, iter) 411 BUG_ON(i == r); 412 fifo_for_each(i, &ca->unused, iter) 413 BUG_ON(i == r); 414 } 415 416 b = ca->buckets + r; 417 418 BUG_ON(atomic_read(&b->pin) != 1); 419 420 SET_GC_SECTORS_USED(b, ca->sb.bucket_size); 421 422 if (watermark <= WATERMARK_METADATA) { 423 SET_GC_MARK(b, GC_MARK_METADATA); 424 SET_GC_MOVE(b, 0); 425 b->prio = BTREE_PRIO; 426 } else { 427 SET_GC_MARK(b, GC_MARK_RECLAIMABLE); 428 SET_GC_MOVE(b, 0); 429 b->prio = INITIAL_PRIO; 430 } 431 432 return r; 433 } 434 435 void bch_bucket_free(struct cache_set *c, struct bkey *k) 436 { 437 unsigned i; 438 439 for (i = 0; i < KEY_PTRS(k); i++) { 440 struct bucket *b = PTR_BUCKET(c, k, i); 441 442 SET_GC_MARK(b, GC_MARK_RECLAIMABLE); 443 SET_GC_SECTORS_USED(b, 0); 444 bch_bucket_add_unused(PTR_CACHE(c, k, i), b); 445 } 446 } 447 448 int __bch_bucket_alloc_set(struct cache_set *c, unsigned watermark, 449 struct bkey *k, int n, bool wait) 450 { 451 int i; 452 453 lockdep_assert_held(&c->bucket_lock); 454 BUG_ON(!n || n > c->caches_loaded || n > 8); 455 456 bkey_init(k); 457 458 /* sort by free space/prio of oldest data in caches */ 459 460 for (i = 0; i < n; i++) { 461 struct cache *ca = c->cache_by_alloc[i]; 462 long b = bch_bucket_alloc(ca, watermark, wait); 463 464 if (b == -1) 465 goto err; 466 467 k->ptr[i] = PTR(ca->buckets[b].gen, 468 bucket_to_sector(c, b), 469 ca->sb.nr_this_dev); 470 471 SET_KEY_PTRS(k, i + 1); 472 } 473 474 return 0; 475 err: 476 bch_bucket_free(c, k); 477 bkey_put(c, k); 478 return -1; 479 } 480 481 int bch_bucket_alloc_set(struct cache_set *c, unsigned watermark, 482 struct bkey *k, int n, bool wait) 483 { 484 int ret; 485 mutex_lock(&c->bucket_lock); 486 ret = __bch_bucket_alloc_set(c, watermark, k, n, wait); 487 mutex_unlock(&c->bucket_lock); 488 return ret; 489 } 490 491 /* Sector allocator */ 492 493 struct open_bucket { 494 struct list_head list; 495 unsigned last_write_point; 496 unsigned sectors_free; 497 BKEY_PADDED(key); 498 }; 499 500 /* 501 * We keep multiple buckets open for writes, and try to segregate different 502 * write streams for better cache utilization: first we look for a bucket where 503 * the last write to it was sequential with the current write, and failing that 504 * we look for a bucket that was last used by the same task. 505 * 506 * The ideas is if you've got multiple tasks pulling data into the cache at the 507 * same time, you'll get better cache utilization if you try to segregate their 508 * data and preserve locality. 509 * 510 * For example, say you've starting Firefox at the same time you're copying a 511 * bunch of files. Firefox will likely end up being fairly hot and stay in the 512 * cache awhile, but the data you copied might not be; if you wrote all that 513 * data to the same buckets it'd get invalidated at the same time. 514 * 515 * Both of those tasks will be doing fairly random IO so we can't rely on 516 * detecting sequential IO to segregate their data, but going off of the task 517 * should be a sane heuristic. 518 */ 519 static struct open_bucket *pick_data_bucket(struct cache_set *c, 520 const struct bkey *search, 521 unsigned write_point, 522 struct bkey *alloc) 523 { 524 struct open_bucket *ret, *ret_task = NULL; 525 526 list_for_each_entry_reverse(ret, &c->data_buckets, list) 527 if (!bkey_cmp(&ret->key, search)) 528 goto found; 529 else if (ret->last_write_point == write_point) 530 ret_task = ret; 531 532 ret = ret_task ?: list_first_entry(&c->data_buckets, 533 struct open_bucket, list); 534 found: 535 if (!ret->sectors_free && KEY_PTRS(alloc)) { 536 ret->sectors_free = c->sb.bucket_size; 537 bkey_copy(&ret->key, alloc); 538 bkey_init(alloc); 539 } 540 541 if (!ret->sectors_free) 542 ret = NULL; 543 544 return ret; 545 } 546 547 /* 548 * Allocates some space in the cache to write to, and k to point to the newly 549 * allocated space, and updates KEY_SIZE(k) and KEY_OFFSET(k) (to point to the 550 * end of the newly allocated space). 551 * 552 * May allocate fewer sectors than @sectors, KEY_SIZE(k) indicates how many 553 * sectors were actually allocated. 554 * 555 * If s->writeback is true, will not fail. 556 */ 557 bool bch_alloc_sectors(struct cache_set *c, struct bkey *k, unsigned sectors, 558 unsigned write_point, unsigned write_prio, bool wait) 559 { 560 struct open_bucket *b; 561 BKEY_PADDED(key) alloc; 562 unsigned i; 563 564 /* 565 * We might have to allocate a new bucket, which we can't do with a 566 * spinlock held. So if we have to allocate, we drop the lock, allocate 567 * and then retry. KEY_PTRS() indicates whether alloc points to 568 * allocated bucket(s). 569 */ 570 571 bkey_init(&alloc.key); 572 spin_lock(&c->data_bucket_lock); 573 574 while (!(b = pick_data_bucket(c, k, write_point, &alloc.key))) { 575 unsigned watermark = write_prio 576 ? WATERMARK_MOVINGGC 577 : WATERMARK_NONE; 578 579 spin_unlock(&c->data_bucket_lock); 580 581 if (bch_bucket_alloc_set(c, watermark, &alloc.key, 1, wait)) 582 return false; 583 584 spin_lock(&c->data_bucket_lock); 585 } 586 587 /* 588 * If we had to allocate, we might race and not need to allocate the 589 * second time we call find_data_bucket(). If we allocated a bucket but 590 * didn't use it, drop the refcount bch_bucket_alloc_set() took: 591 */ 592 if (KEY_PTRS(&alloc.key)) 593 bkey_put(c, &alloc.key); 594 595 for (i = 0; i < KEY_PTRS(&b->key); i++) 596 EBUG_ON(ptr_stale(c, &b->key, i)); 597 598 /* Set up the pointer to the space we're allocating: */ 599 600 for (i = 0; i < KEY_PTRS(&b->key); i++) 601 k->ptr[i] = b->key.ptr[i]; 602 603 sectors = min(sectors, b->sectors_free); 604 605 SET_KEY_OFFSET(k, KEY_OFFSET(k) + sectors); 606 SET_KEY_SIZE(k, sectors); 607 SET_KEY_PTRS(k, KEY_PTRS(&b->key)); 608 609 /* 610 * Move b to the end of the lru, and keep track of what this bucket was 611 * last used for: 612 */ 613 list_move_tail(&b->list, &c->data_buckets); 614 bkey_copy_key(&b->key, k); 615 b->last_write_point = write_point; 616 617 b->sectors_free -= sectors; 618 619 for (i = 0; i < KEY_PTRS(&b->key); i++) { 620 SET_PTR_OFFSET(&b->key, i, PTR_OFFSET(&b->key, i) + sectors); 621 622 atomic_long_add(sectors, 623 &PTR_CACHE(c, &b->key, i)->sectors_written); 624 } 625 626 if (b->sectors_free < c->sb.block_size) 627 b->sectors_free = 0; 628 629 /* 630 * k takes refcounts on the buckets it points to until it's inserted 631 * into the btree, but if we're done with this bucket we just transfer 632 * get_data_bucket()'s refcount. 633 */ 634 if (b->sectors_free) 635 for (i = 0; i < KEY_PTRS(&b->key); i++) 636 atomic_inc(&PTR_BUCKET(c, &b->key, i)->pin); 637 638 spin_unlock(&c->data_bucket_lock); 639 return true; 640 } 641 642 /* Init */ 643 644 void bch_open_buckets_free(struct cache_set *c) 645 { 646 struct open_bucket *b; 647 648 while (!list_empty(&c->data_buckets)) { 649 b = list_first_entry(&c->data_buckets, 650 struct open_bucket, list); 651 list_del(&b->list); 652 kfree(b); 653 } 654 } 655 656 int bch_open_buckets_alloc(struct cache_set *c) 657 { 658 int i; 659 660 spin_lock_init(&c->data_bucket_lock); 661 662 for (i = 0; i < 6; i++) { 663 struct open_bucket *b = kzalloc(sizeof(*b), GFP_KERNEL); 664 if (!b) 665 return -ENOMEM; 666 667 list_add(&b->list, &c->data_buckets); 668 } 669 670 return 0; 671 } 672 673 int bch_cache_allocator_start(struct cache *ca) 674 { 675 struct task_struct *k = kthread_run(bch_allocator_thread, 676 ca, "bcache_allocator"); 677 if (IS_ERR(k)) 678 return PTR_ERR(k); 679 680 ca->alloc_thread = k; 681 return 0; 682 } 683 684 int bch_cache_allocator_init(struct cache *ca) 685 { 686 /* 687 * Reserve: 688 * Prio/gen writes first 689 * Then 8 for btree allocations 690 * Then half for the moving garbage collector 691 */ 692 693 ca->watermark[WATERMARK_PRIO] = 0; 694 695 ca->watermark[WATERMARK_METADATA] = prio_buckets(ca); 696 697 ca->watermark[WATERMARK_MOVINGGC] = 8 + 698 ca->watermark[WATERMARK_METADATA]; 699 700 ca->watermark[WATERMARK_NONE] = ca->free.size / 2 + 701 ca->watermark[WATERMARK_MOVINGGC]; 702 703 return 0; 704 } 705