xref: /linux/fs/bcachefs/btree_key_cache.c (revision 1e525507)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "btree_cache.h"
5 #include "btree_iter.h"
6 #include "btree_key_cache.h"
7 #include "btree_locking.h"
8 #include "btree_update.h"
9 #include "errcode.h"
10 #include "error.h"
11 #include "journal.h"
12 #include "journal_reclaim.h"
13 #include "trace.h"
14 
15 #include <linux/sched/mm.h>
16 
17 static inline bool btree_uses_pcpu_readers(enum btree_id id)
18 {
19 	return id == BTREE_ID_subvolumes;
20 }
21 
22 static struct kmem_cache *bch2_key_cache;
23 
24 static int bch2_btree_key_cache_cmp_fn(struct rhashtable_compare_arg *arg,
25 				       const void *obj)
26 {
27 	const struct bkey_cached *ck = obj;
28 	const struct bkey_cached_key *key = arg->key;
29 
30 	return ck->key.btree_id != key->btree_id ||
31 		!bpos_eq(ck->key.pos, key->pos);
32 }
33 
34 static const struct rhashtable_params bch2_btree_key_cache_params = {
35 	.head_offset	= offsetof(struct bkey_cached, hash),
36 	.key_offset	= offsetof(struct bkey_cached, key),
37 	.key_len	= sizeof(struct bkey_cached_key),
38 	.obj_cmpfn	= bch2_btree_key_cache_cmp_fn,
39 };
40 
41 __flatten
42 inline struct bkey_cached *
43 bch2_btree_key_cache_find(struct bch_fs *c, enum btree_id btree_id, struct bpos pos)
44 {
45 	struct bkey_cached_key key = {
46 		.btree_id	= btree_id,
47 		.pos		= pos,
48 	};
49 
50 	return rhashtable_lookup_fast(&c->btree_key_cache.table, &key,
51 				      bch2_btree_key_cache_params);
52 }
53 
54 static bool bkey_cached_lock_for_evict(struct bkey_cached *ck)
55 {
56 	if (!six_trylock_intent(&ck->c.lock))
57 		return false;
58 
59 	if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
60 		six_unlock_intent(&ck->c.lock);
61 		return false;
62 	}
63 
64 	if (!six_trylock_write(&ck->c.lock)) {
65 		six_unlock_intent(&ck->c.lock);
66 		return false;
67 	}
68 
69 	return true;
70 }
71 
72 static void bkey_cached_evict(struct btree_key_cache *c,
73 			      struct bkey_cached *ck)
74 {
75 	BUG_ON(rhashtable_remove_fast(&c->table, &ck->hash,
76 				      bch2_btree_key_cache_params));
77 	memset(&ck->key, ~0, sizeof(ck->key));
78 
79 	atomic_long_dec(&c->nr_keys);
80 }
81 
82 static void bkey_cached_free(struct btree_key_cache *bc,
83 			     struct bkey_cached *ck)
84 {
85 	struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
86 
87 	BUG_ON(test_bit(BKEY_CACHED_DIRTY, &ck->flags));
88 
89 	ck->btree_trans_barrier_seq =
90 		start_poll_synchronize_srcu(&c->btree_trans_barrier);
91 
92 	if (ck->c.lock.readers) {
93 		list_move_tail(&ck->list, &bc->freed_pcpu);
94 		bc->nr_freed_pcpu++;
95 	} else {
96 		list_move_tail(&ck->list, &bc->freed_nonpcpu);
97 		bc->nr_freed_nonpcpu++;
98 	}
99 	atomic_long_inc(&bc->nr_freed);
100 
101 	kfree(ck->k);
102 	ck->k		= NULL;
103 	ck->u64s	= 0;
104 
105 	six_unlock_write(&ck->c.lock);
106 	six_unlock_intent(&ck->c.lock);
107 }
108 
109 #ifdef __KERNEL__
110 static void __bkey_cached_move_to_freelist_ordered(struct btree_key_cache *bc,
111 						   struct bkey_cached *ck)
112 {
113 	struct bkey_cached *pos;
114 
115 	bc->nr_freed_nonpcpu++;
116 
117 	list_for_each_entry_reverse(pos, &bc->freed_nonpcpu, list) {
118 		if (ULONG_CMP_GE(ck->btree_trans_barrier_seq,
119 				 pos->btree_trans_barrier_seq)) {
120 			list_move(&ck->list, &pos->list);
121 			return;
122 		}
123 	}
124 
125 	list_move(&ck->list, &bc->freed_nonpcpu);
126 }
127 #endif
128 
129 static void bkey_cached_move_to_freelist(struct btree_key_cache *bc,
130 					 struct bkey_cached *ck)
131 {
132 	BUG_ON(test_bit(BKEY_CACHED_DIRTY, &ck->flags));
133 
134 	if (!ck->c.lock.readers) {
135 #ifdef __KERNEL__
136 		struct btree_key_cache_freelist *f;
137 		bool freed = false;
138 
139 		preempt_disable();
140 		f = this_cpu_ptr(bc->pcpu_freed);
141 
142 		if (f->nr < ARRAY_SIZE(f->objs)) {
143 			f->objs[f->nr++] = ck;
144 			freed = true;
145 		}
146 		preempt_enable();
147 
148 		if (!freed) {
149 			mutex_lock(&bc->lock);
150 			preempt_disable();
151 			f = this_cpu_ptr(bc->pcpu_freed);
152 
153 			while (f->nr > ARRAY_SIZE(f->objs) / 2) {
154 				struct bkey_cached *ck2 = f->objs[--f->nr];
155 
156 				__bkey_cached_move_to_freelist_ordered(bc, ck2);
157 			}
158 			preempt_enable();
159 
160 			__bkey_cached_move_to_freelist_ordered(bc, ck);
161 			mutex_unlock(&bc->lock);
162 		}
163 #else
164 		mutex_lock(&bc->lock);
165 		list_move_tail(&ck->list, &bc->freed_nonpcpu);
166 		bc->nr_freed_nonpcpu++;
167 		mutex_unlock(&bc->lock);
168 #endif
169 	} else {
170 		mutex_lock(&bc->lock);
171 		list_move_tail(&ck->list, &bc->freed_pcpu);
172 		bc->nr_freed_pcpu++;
173 		mutex_unlock(&bc->lock);
174 	}
175 }
176 
177 static void bkey_cached_free_fast(struct btree_key_cache *bc,
178 				  struct bkey_cached *ck)
179 {
180 	struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
181 
182 	ck->btree_trans_barrier_seq =
183 		start_poll_synchronize_srcu(&c->btree_trans_barrier);
184 
185 	list_del_init(&ck->list);
186 	atomic_long_inc(&bc->nr_freed);
187 
188 	kfree(ck->k);
189 	ck->k		= NULL;
190 	ck->u64s	= 0;
191 
192 	bkey_cached_move_to_freelist(bc, ck);
193 
194 	six_unlock_write(&ck->c.lock);
195 	six_unlock_intent(&ck->c.lock);
196 }
197 
198 static struct bkey_cached *
199 bkey_cached_alloc(struct btree_trans *trans, struct btree_path *path,
200 		  bool *was_new)
201 {
202 	struct bch_fs *c = trans->c;
203 	struct btree_key_cache *bc = &c->btree_key_cache;
204 	struct bkey_cached *ck = NULL;
205 	bool pcpu_readers = btree_uses_pcpu_readers(path->btree_id);
206 	int ret;
207 
208 	if (!pcpu_readers) {
209 #ifdef __KERNEL__
210 		struct btree_key_cache_freelist *f;
211 
212 		preempt_disable();
213 		f = this_cpu_ptr(bc->pcpu_freed);
214 		if (f->nr)
215 			ck = f->objs[--f->nr];
216 		preempt_enable();
217 
218 		if (!ck) {
219 			mutex_lock(&bc->lock);
220 			preempt_disable();
221 			f = this_cpu_ptr(bc->pcpu_freed);
222 
223 			while (!list_empty(&bc->freed_nonpcpu) &&
224 			       f->nr < ARRAY_SIZE(f->objs) / 2) {
225 				ck = list_last_entry(&bc->freed_nonpcpu, struct bkey_cached, list);
226 				list_del_init(&ck->list);
227 				bc->nr_freed_nonpcpu--;
228 				f->objs[f->nr++] = ck;
229 			}
230 
231 			ck = f->nr ? f->objs[--f->nr] : NULL;
232 			preempt_enable();
233 			mutex_unlock(&bc->lock);
234 		}
235 #else
236 		mutex_lock(&bc->lock);
237 		if (!list_empty(&bc->freed_nonpcpu)) {
238 			ck = list_last_entry(&bc->freed_nonpcpu, struct bkey_cached, list);
239 			list_del_init(&ck->list);
240 			bc->nr_freed_nonpcpu--;
241 		}
242 		mutex_unlock(&bc->lock);
243 #endif
244 	} else {
245 		mutex_lock(&bc->lock);
246 		if (!list_empty(&bc->freed_pcpu)) {
247 			ck = list_last_entry(&bc->freed_pcpu, struct bkey_cached, list);
248 			list_del_init(&ck->list);
249 			bc->nr_freed_pcpu--;
250 		}
251 		mutex_unlock(&bc->lock);
252 	}
253 
254 	if (ck) {
255 		ret = btree_node_lock_nopath(trans, &ck->c, SIX_LOCK_intent, _THIS_IP_);
256 		if (unlikely(ret)) {
257 			bkey_cached_move_to_freelist(bc, ck);
258 			return ERR_PTR(ret);
259 		}
260 
261 		path->l[0].b = (void *) ck;
262 		path->l[0].lock_seq = six_lock_seq(&ck->c.lock);
263 		mark_btree_node_locked(trans, path, 0, BTREE_NODE_INTENT_LOCKED);
264 
265 		ret = bch2_btree_node_lock_write(trans, path, &ck->c);
266 		if (unlikely(ret)) {
267 			btree_node_unlock(trans, path, 0);
268 			bkey_cached_move_to_freelist(bc, ck);
269 			return ERR_PTR(ret);
270 		}
271 
272 		return ck;
273 	}
274 
275 	ck = allocate_dropping_locks(trans, ret,
276 			kmem_cache_zalloc(bch2_key_cache, _gfp));
277 	if (ret) {
278 		kmem_cache_free(bch2_key_cache, ck);
279 		return ERR_PTR(ret);
280 	}
281 
282 	if (!ck)
283 		return NULL;
284 
285 	INIT_LIST_HEAD(&ck->list);
286 	bch2_btree_lock_init(&ck->c, pcpu_readers ? SIX_LOCK_INIT_PCPU : 0);
287 
288 	ck->c.cached = true;
289 	BUG_ON(!six_trylock_intent(&ck->c.lock));
290 	BUG_ON(!six_trylock_write(&ck->c.lock));
291 	*was_new = true;
292 	return ck;
293 }
294 
295 static struct bkey_cached *
296 bkey_cached_reuse(struct btree_key_cache *c)
297 {
298 	struct bucket_table *tbl;
299 	struct rhash_head *pos;
300 	struct bkey_cached *ck;
301 	unsigned i;
302 
303 	mutex_lock(&c->lock);
304 	rcu_read_lock();
305 	tbl = rht_dereference_rcu(c->table.tbl, &c->table);
306 	for (i = 0; i < tbl->size; i++)
307 		rht_for_each_entry_rcu(ck, pos, tbl, i, hash) {
308 			if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
309 			    bkey_cached_lock_for_evict(ck)) {
310 				bkey_cached_evict(c, ck);
311 				goto out;
312 			}
313 		}
314 	ck = NULL;
315 out:
316 	rcu_read_unlock();
317 	mutex_unlock(&c->lock);
318 	return ck;
319 }
320 
321 static struct bkey_cached *
322 btree_key_cache_create(struct btree_trans *trans, struct btree_path *path)
323 {
324 	struct bch_fs *c = trans->c;
325 	struct btree_key_cache *bc = &c->btree_key_cache;
326 	struct bkey_cached *ck;
327 	bool was_new = false;
328 
329 	ck = bkey_cached_alloc(trans, path, &was_new);
330 	if (IS_ERR(ck))
331 		return ck;
332 
333 	if (unlikely(!ck)) {
334 		ck = bkey_cached_reuse(bc);
335 		if (unlikely(!ck)) {
336 			bch_err(c, "error allocating memory for key cache item, btree %s",
337 				bch2_btree_id_str(path->btree_id));
338 			return ERR_PTR(-BCH_ERR_ENOMEM_btree_key_cache_create);
339 		}
340 
341 		mark_btree_node_locked(trans, path, 0, BTREE_NODE_INTENT_LOCKED);
342 	}
343 
344 	ck->c.level		= 0;
345 	ck->c.btree_id		= path->btree_id;
346 	ck->key.btree_id	= path->btree_id;
347 	ck->key.pos		= path->pos;
348 	ck->valid		= false;
349 	ck->flags		= 1U << BKEY_CACHED_ACCESSED;
350 
351 	if (unlikely(rhashtable_lookup_insert_fast(&bc->table,
352 					  &ck->hash,
353 					  bch2_btree_key_cache_params))) {
354 		/* We raced with another fill: */
355 
356 		if (likely(was_new)) {
357 			six_unlock_write(&ck->c.lock);
358 			six_unlock_intent(&ck->c.lock);
359 			kfree(ck);
360 		} else {
361 			bkey_cached_free_fast(bc, ck);
362 		}
363 
364 		mark_btree_node_locked(trans, path, 0, BTREE_NODE_UNLOCKED);
365 		return NULL;
366 	}
367 
368 	atomic_long_inc(&bc->nr_keys);
369 
370 	six_unlock_write(&ck->c.lock);
371 
372 	return ck;
373 }
374 
375 static int btree_key_cache_fill(struct btree_trans *trans,
376 				struct btree_path *ck_path,
377 				struct bkey_cached *ck)
378 {
379 	struct btree_iter iter;
380 	struct bkey_s_c k;
381 	unsigned new_u64s = 0;
382 	struct bkey_i *new_k = NULL;
383 	int ret;
384 
385 	bch2_trans_iter_init(trans, &iter, ck->key.btree_id, ck->key.pos,
386 			     BTREE_ITER_KEY_CACHE_FILL|
387 			     BTREE_ITER_CACHED_NOFILL);
388 	iter.flags &= ~BTREE_ITER_WITH_JOURNAL;
389 	k = bch2_btree_iter_peek_slot(&iter);
390 	ret = bkey_err(k);
391 	if (ret)
392 		goto err;
393 
394 	if (!bch2_btree_node_relock(trans, ck_path, 0)) {
395 		trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
396 		ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_fill);
397 		goto err;
398 	}
399 
400 	/*
401 	 * bch2_varint_decode can read past the end of the buffer by at
402 	 * most 7 bytes (it won't be used):
403 	 */
404 	new_u64s = k.k->u64s + 1;
405 
406 	/*
407 	 * Allocate some extra space so that the transaction commit path is less
408 	 * likely to have to reallocate, since that requires a transaction
409 	 * restart:
410 	 */
411 	new_u64s = min(256U, (new_u64s * 3) / 2);
412 
413 	if (new_u64s > ck->u64s) {
414 		new_u64s = roundup_pow_of_two(new_u64s);
415 		new_k = kmalloc(new_u64s * sizeof(u64), GFP_NOWAIT|__GFP_NOWARN);
416 		if (!new_k) {
417 			bch2_trans_unlock(trans);
418 
419 			new_k = kmalloc(new_u64s * sizeof(u64), GFP_KERNEL);
420 			if (!new_k) {
421 				bch_err(trans->c, "error allocating memory for key cache key, btree %s u64s %u",
422 					bch2_btree_id_str(ck->key.btree_id), new_u64s);
423 				ret = -BCH_ERR_ENOMEM_btree_key_cache_fill;
424 				goto err;
425 			}
426 
427 			if (!bch2_btree_node_relock(trans, ck_path, 0)) {
428 				kfree(new_k);
429 				trace_and_count(trans->c, trans_restart_relock_key_cache_fill, trans, _THIS_IP_, ck_path);
430 				ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_fill);
431 				goto err;
432 			}
433 
434 			ret = bch2_trans_relock(trans);
435 			if (ret) {
436 				kfree(new_k);
437 				goto err;
438 			}
439 		}
440 	}
441 
442 	ret = bch2_btree_node_lock_write(trans, ck_path, &ck_path->l[0].b->c);
443 	if (ret) {
444 		kfree(new_k);
445 		goto err;
446 	}
447 
448 	if (new_k) {
449 		kfree(ck->k);
450 		ck->u64s = new_u64s;
451 		ck->k = new_k;
452 	}
453 
454 	bkey_reassemble(ck->k, k);
455 	ck->valid = true;
456 	bch2_btree_node_unlock_write(trans, ck_path, ck_path->l[0].b);
457 
458 	/* We're not likely to need this iterator again: */
459 	set_btree_iter_dontneed(&iter);
460 err:
461 	bch2_trans_iter_exit(trans, &iter);
462 	return ret;
463 }
464 
465 static noinline int
466 bch2_btree_path_traverse_cached_slowpath(struct btree_trans *trans, struct btree_path *path,
467 					 unsigned flags)
468 {
469 	struct bch_fs *c = trans->c;
470 	struct bkey_cached *ck;
471 	int ret = 0;
472 
473 	BUG_ON(path->level);
474 
475 	path->l[1].b = NULL;
476 
477 	if (bch2_btree_node_relock_notrace(trans, path, 0)) {
478 		ck = (void *) path->l[0].b;
479 		goto fill;
480 	}
481 retry:
482 	ck = bch2_btree_key_cache_find(c, path->btree_id, path->pos);
483 	if (!ck) {
484 		ck = btree_key_cache_create(trans, path);
485 		ret = PTR_ERR_OR_ZERO(ck);
486 		if (ret)
487 			goto err;
488 		if (!ck)
489 			goto retry;
490 
491 		mark_btree_node_locked(trans, path, 0, BTREE_NODE_INTENT_LOCKED);
492 		path->locks_want = 1;
493 	} else {
494 		enum six_lock_type lock_want = __btree_lock_want(path, 0);
495 
496 		ret = btree_node_lock(trans, path, (void *) ck, 0,
497 				      lock_want, _THIS_IP_);
498 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
499 			goto err;
500 
501 		BUG_ON(ret);
502 
503 		if (ck->key.btree_id != path->btree_id ||
504 		    !bpos_eq(ck->key.pos, path->pos)) {
505 			six_unlock_type(&ck->c.lock, lock_want);
506 			goto retry;
507 		}
508 
509 		mark_btree_node_locked(trans, path, 0,
510 				       (enum btree_node_locked_type) lock_want);
511 	}
512 
513 	path->l[0].lock_seq	= six_lock_seq(&ck->c.lock);
514 	path->l[0].b		= (void *) ck;
515 fill:
516 	path->uptodate = BTREE_ITER_UPTODATE;
517 
518 	if (!ck->valid && !(flags & BTREE_ITER_CACHED_NOFILL)) {
519 		/*
520 		 * Using the underscore version because we haven't set
521 		 * path->uptodate yet:
522 		 */
523 		if (!path->locks_want &&
524 		    !__bch2_btree_path_upgrade(trans, path, 1, NULL)) {
525 			trace_and_count(trans->c, trans_restart_key_cache_upgrade, trans, _THIS_IP_);
526 			ret = btree_trans_restart(trans, BCH_ERR_transaction_restart_key_cache_upgrade);
527 			goto err;
528 		}
529 
530 		ret = btree_key_cache_fill(trans, path, ck);
531 		if (ret)
532 			goto err;
533 
534 		ret = bch2_btree_path_relock(trans, path, _THIS_IP_);
535 		if (ret)
536 			goto err;
537 
538 		path->uptodate = BTREE_ITER_UPTODATE;
539 	}
540 
541 	if (!test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
542 		set_bit(BKEY_CACHED_ACCESSED, &ck->flags);
543 
544 	BUG_ON(btree_node_locked_type(path, 0) != btree_lock_want(path, 0));
545 	BUG_ON(path->uptodate);
546 
547 	return ret;
548 err:
549 	path->uptodate = BTREE_ITER_NEED_TRAVERSE;
550 	if (!bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
551 		btree_node_unlock(trans, path, 0);
552 		path->l[0].b = ERR_PTR(ret);
553 	}
554 	return ret;
555 }
556 
557 int bch2_btree_path_traverse_cached(struct btree_trans *trans, struct btree_path *path,
558 				    unsigned flags)
559 {
560 	struct bch_fs *c = trans->c;
561 	struct bkey_cached *ck;
562 	int ret = 0;
563 
564 	EBUG_ON(path->level);
565 
566 	path->l[1].b = NULL;
567 
568 	if (bch2_btree_node_relock_notrace(trans, path, 0)) {
569 		ck = (void *) path->l[0].b;
570 		goto fill;
571 	}
572 retry:
573 	ck = bch2_btree_key_cache_find(c, path->btree_id, path->pos);
574 	if (!ck) {
575 		return bch2_btree_path_traverse_cached_slowpath(trans, path, flags);
576 	} else {
577 		enum six_lock_type lock_want = __btree_lock_want(path, 0);
578 
579 		ret = btree_node_lock(trans, path, (void *) ck, 0,
580 				      lock_want, _THIS_IP_);
581 		EBUG_ON(ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart));
582 
583 		if (ret)
584 			return ret;
585 
586 		if (ck->key.btree_id != path->btree_id ||
587 		    !bpos_eq(ck->key.pos, path->pos)) {
588 			six_unlock_type(&ck->c.lock, lock_want);
589 			goto retry;
590 		}
591 
592 		mark_btree_node_locked(trans, path, 0,
593 				       (enum btree_node_locked_type) lock_want);
594 	}
595 
596 	path->l[0].lock_seq	= six_lock_seq(&ck->c.lock);
597 	path->l[0].b		= (void *) ck;
598 fill:
599 	if (!ck->valid)
600 		return bch2_btree_path_traverse_cached_slowpath(trans, path, flags);
601 
602 	if (!test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
603 		set_bit(BKEY_CACHED_ACCESSED, &ck->flags);
604 
605 	path->uptodate = BTREE_ITER_UPTODATE;
606 	EBUG_ON(!ck->valid);
607 	EBUG_ON(btree_node_locked_type(path, 0) != btree_lock_want(path, 0));
608 
609 	return ret;
610 }
611 
612 static int btree_key_cache_flush_pos(struct btree_trans *trans,
613 				     struct bkey_cached_key key,
614 				     u64 journal_seq,
615 				     unsigned commit_flags,
616 				     bool evict)
617 {
618 	struct bch_fs *c = trans->c;
619 	struct journal *j = &c->journal;
620 	struct btree_iter c_iter, b_iter;
621 	struct bkey_cached *ck = NULL;
622 	int ret;
623 
624 	bch2_trans_iter_init(trans, &b_iter, key.btree_id, key.pos,
625 			     BTREE_ITER_SLOTS|
626 			     BTREE_ITER_INTENT|
627 			     BTREE_ITER_ALL_SNAPSHOTS);
628 	bch2_trans_iter_init(trans, &c_iter, key.btree_id, key.pos,
629 			     BTREE_ITER_CACHED|
630 			     BTREE_ITER_INTENT);
631 	b_iter.flags &= ~BTREE_ITER_WITH_KEY_CACHE;
632 
633 	ret = bch2_btree_iter_traverse(&c_iter);
634 	if (ret)
635 		goto out;
636 
637 	ck = (void *) btree_iter_path(trans, &c_iter)->l[0].b;
638 	if (!ck)
639 		goto out;
640 
641 	if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
642 		if (evict)
643 			goto evict;
644 		goto out;
645 	}
646 
647 	BUG_ON(!ck->valid);
648 
649 	if (journal_seq && ck->journal.seq != journal_seq)
650 		goto out;
651 
652 	trans->journal_res.seq = ck->journal.seq;
653 
654 	/*
655 	 * If we're at the end of the journal, we really want to free up space
656 	 * in the journal right away - we don't want to pin that old journal
657 	 * sequence number with a new btree node write, we want to re-journal
658 	 * the update
659 	 */
660 	if (ck->journal.seq == journal_last_seq(j))
661 		commit_flags |= BCH_WATERMARK_reclaim;
662 
663 	if (ck->journal.seq != journal_last_seq(j) ||
664 	    !test_bit(JOURNAL_SPACE_LOW, &c->journal.flags))
665 		commit_flags |= BCH_TRANS_COMMIT_no_journal_res;
666 
667 	ret   = bch2_btree_iter_traverse(&b_iter) ?:
668 		bch2_trans_update(trans, &b_iter, ck->k,
669 				  BTREE_UPDATE_KEY_CACHE_RECLAIM|
670 				  BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE|
671 				  BTREE_TRIGGER_NORUN) ?:
672 		bch2_trans_commit(trans, NULL, NULL,
673 				  BCH_TRANS_COMMIT_no_check_rw|
674 				  BCH_TRANS_COMMIT_no_enospc|
675 				  commit_flags);
676 
677 	bch2_fs_fatal_err_on(ret &&
678 			     !bch2_err_matches(ret, BCH_ERR_transaction_restart) &&
679 			     !bch2_err_matches(ret, BCH_ERR_journal_reclaim_would_deadlock) &&
680 			     !bch2_journal_error(j), c,
681 			     "flushing key cache: %s", bch2_err_str(ret));
682 	if (ret)
683 		goto out;
684 
685 	bch2_journal_pin_drop(j, &ck->journal);
686 
687 	struct btree_path *path = btree_iter_path(trans, &c_iter);
688 	BUG_ON(!btree_node_locked(path, 0));
689 
690 	if (!evict) {
691 		if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
692 			clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
693 			atomic_long_dec(&c->btree_key_cache.nr_dirty);
694 		}
695 	} else {
696 		struct btree_path *path2;
697 		unsigned i;
698 evict:
699 		trans_for_each_path(trans, path2, i)
700 			if (path2 != path)
701 				__bch2_btree_path_unlock(trans, path2);
702 
703 		bch2_btree_node_lock_write_nofail(trans, path, &ck->c);
704 
705 		if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
706 			clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
707 			atomic_long_dec(&c->btree_key_cache.nr_dirty);
708 		}
709 
710 		mark_btree_node_locked_noreset(path, 0, BTREE_NODE_UNLOCKED);
711 		bkey_cached_evict(&c->btree_key_cache, ck);
712 		bkey_cached_free_fast(&c->btree_key_cache, ck);
713 	}
714 out:
715 	bch2_trans_iter_exit(trans, &b_iter);
716 	bch2_trans_iter_exit(trans, &c_iter);
717 	return ret;
718 }
719 
720 int bch2_btree_key_cache_journal_flush(struct journal *j,
721 				struct journal_entry_pin *pin, u64 seq)
722 {
723 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
724 	struct bkey_cached *ck =
725 		container_of(pin, struct bkey_cached, journal);
726 	struct bkey_cached_key key;
727 	struct btree_trans *trans = bch2_trans_get(c);
728 	int srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
729 	int ret = 0;
730 
731 	btree_node_lock_nopath_nofail(trans, &ck->c, SIX_LOCK_read);
732 	key = ck->key;
733 
734 	if (ck->journal.seq != seq ||
735 	    !test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
736 		six_unlock_read(&ck->c.lock);
737 		goto unlock;
738 	}
739 
740 	if (ck->seq != seq) {
741 		bch2_journal_pin_update(&c->journal, ck->seq, &ck->journal,
742 					bch2_btree_key_cache_journal_flush);
743 		six_unlock_read(&ck->c.lock);
744 		goto unlock;
745 	}
746 	six_unlock_read(&ck->c.lock);
747 
748 	ret = lockrestart_do(trans,
749 		btree_key_cache_flush_pos(trans, key, seq,
750 				BCH_TRANS_COMMIT_journal_reclaim, false));
751 unlock:
752 	srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);
753 
754 	bch2_trans_put(trans);
755 	return ret;
756 }
757 
758 bool bch2_btree_insert_key_cached(struct btree_trans *trans,
759 				  unsigned flags,
760 				  struct btree_insert_entry *insert_entry)
761 {
762 	struct bch_fs *c = trans->c;
763 	struct bkey_cached *ck = (void *) (trans->paths + insert_entry->path)->l[0].b;
764 	struct bkey_i *insert = insert_entry->k;
765 	bool kick_reclaim = false;
766 
767 	BUG_ON(insert->k.u64s > ck->u64s);
768 
769 	bkey_copy(ck->k, insert);
770 	ck->valid = true;
771 
772 	if (!test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
773 		EBUG_ON(test_bit(BCH_FS_clean_shutdown, &c->flags));
774 		set_bit(BKEY_CACHED_DIRTY, &ck->flags);
775 		atomic_long_inc(&c->btree_key_cache.nr_dirty);
776 
777 		if (bch2_nr_btree_keys_need_flush(c))
778 			kick_reclaim = true;
779 	}
780 
781 	/*
782 	 * To minimize lock contention, we only add the journal pin here and
783 	 * defer pin updates to the flush callback via ->seq. Be careful not to
784 	 * update ->seq on nojournal commits because we don't want to update the
785 	 * pin to a seq that doesn't include journal updates on disk. Otherwise
786 	 * we risk losing the update after a crash.
787 	 *
788 	 * The only exception is if the pin is not active in the first place. We
789 	 * have to add the pin because journal reclaim drives key cache
790 	 * flushing. The flush callback will not proceed unless ->seq matches
791 	 * the latest pin, so make sure it starts with a consistent value.
792 	 */
793 	if (!(insert_entry->flags & BTREE_UPDATE_NOJOURNAL) ||
794 	    !journal_pin_active(&ck->journal)) {
795 		ck->seq = trans->journal_res.seq;
796 	}
797 	bch2_journal_pin_add(&c->journal, trans->journal_res.seq,
798 			     &ck->journal, bch2_btree_key_cache_journal_flush);
799 
800 	if (kick_reclaim)
801 		journal_reclaim_kick(&c->journal);
802 	return true;
803 }
804 
805 void bch2_btree_key_cache_drop(struct btree_trans *trans,
806 			       struct btree_path *path)
807 {
808 	struct bch_fs *c = trans->c;
809 	struct bkey_cached *ck = (void *) path->l[0].b;
810 
811 	BUG_ON(!ck->valid);
812 
813 	/*
814 	 * We just did an update to the btree, bypassing the key cache: the key
815 	 * cache key is now stale and must be dropped, even if dirty:
816 	 */
817 	if (test_bit(BKEY_CACHED_DIRTY, &ck->flags)) {
818 		clear_bit(BKEY_CACHED_DIRTY, &ck->flags);
819 		atomic_long_dec(&c->btree_key_cache.nr_dirty);
820 		bch2_journal_pin_drop(&c->journal, &ck->journal);
821 	}
822 
823 	ck->valid = false;
824 }
825 
826 static unsigned long bch2_btree_key_cache_scan(struct shrinker *shrink,
827 					   struct shrink_control *sc)
828 {
829 	struct bch_fs *c = shrink->private_data;
830 	struct btree_key_cache *bc = &c->btree_key_cache;
831 	struct bucket_table *tbl;
832 	struct bkey_cached *ck, *t;
833 	size_t scanned = 0, freed = 0, nr = sc->nr_to_scan;
834 	unsigned start, flags;
835 	int srcu_idx;
836 
837 	mutex_lock(&bc->lock);
838 	srcu_idx = srcu_read_lock(&c->btree_trans_barrier);
839 	flags = memalloc_nofs_save();
840 
841 	/*
842 	 * Newest freed entries are at the end of the list - once we hit one
843 	 * that's too new to be freed, we can bail out:
844 	 */
845 	scanned += bc->nr_freed_nonpcpu;
846 
847 	list_for_each_entry_safe(ck, t, &bc->freed_nonpcpu, list) {
848 		if (!poll_state_synchronize_srcu(&c->btree_trans_barrier,
849 						 ck->btree_trans_barrier_seq))
850 			break;
851 
852 		list_del(&ck->list);
853 		six_lock_exit(&ck->c.lock);
854 		kmem_cache_free(bch2_key_cache, ck);
855 		atomic_long_dec(&bc->nr_freed);
856 		freed++;
857 		bc->nr_freed_nonpcpu--;
858 	}
859 
860 	if (scanned >= nr)
861 		goto out;
862 
863 	scanned += bc->nr_freed_pcpu;
864 
865 	list_for_each_entry_safe(ck, t, &bc->freed_pcpu, list) {
866 		if (!poll_state_synchronize_srcu(&c->btree_trans_barrier,
867 						 ck->btree_trans_barrier_seq))
868 			break;
869 
870 		list_del(&ck->list);
871 		six_lock_exit(&ck->c.lock);
872 		kmem_cache_free(bch2_key_cache, ck);
873 		atomic_long_dec(&bc->nr_freed);
874 		freed++;
875 		bc->nr_freed_pcpu--;
876 	}
877 
878 	if (scanned >= nr)
879 		goto out;
880 
881 	rcu_read_lock();
882 	tbl = rht_dereference_rcu(bc->table.tbl, &bc->table);
883 	if (bc->shrink_iter >= tbl->size)
884 		bc->shrink_iter = 0;
885 	start = bc->shrink_iter;
886 
887 	do {
888 		struct rhash_head *pos, *next;
889 
890 		pos = rht_ptr_rcu(rht_bucket(tbl, bc->shrink_iter));
891 
892 		while (!rht_is_a_nulls(pos)) {
893 			next = rht_dereference_bucket_rcu(pos->next, tbl, bc->shrink_iter);
894 			ck = container_of(pos, struct bkey_cached, hash);
895 
896 			if (test_bit(BKEY_CACHED_DIRTY, &ck->flags))
897 				goto next;
898 
899 			if (test_bit(BKEY_CACHED_ACCESSED, &ck->flags))
900 				clear_bit(BKEY_CACHED_ACCESSED, &ck->flags);
901 			else if (bkey_cached_lock_for_evict(ck)) {
902 				bkey_cached_evict(bc, ck);
903 				bkey_cached_free(bc, ck);
904 			}
905 
906 			scanned++;
907 			if (scanned >= nr)
908 				break;
909 next:
910 			pos = next;
911 		}
912 
913 		bc->shrink_iter++;
914 		if (bc->shrink_iter >= tbl->size)
915 			bc->shrink_iter = 0;
916 	} while (scanned < nr && bc->shrink_iter != start);
917 
918 	rcu_read_unlock();
919 out:
920 	memalloc_nofs_restore(flags);
921 	srcu_read_unlock(&c->btree_trans_barrier, srcu_idx);
922 	mutex_unlock(&bc->lock);
923 
924 	return freed;
925 }
926 
927 static unsigned long bch2_btree_key_cache_count(struct shrinker *shrink,
928 					    struct shrink_control *sc)
929 {
930 	struct bch_fs *c = shrink->private_data;
931 	struct btree_key_cache *bc = &c->btree_key_cache;
932 	long nr = atomic_long_read(&bc->nr_keys) -
933 		atomic_long_read(&bc->nr_dirty);
934 
935 	return max(0L, nr);
936 }
937 
938 void bch2_fs_btree_key_cache_exit(struct btree_key_cache *bc)
939 {
940 	struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
941 	struct bucket_table *tbl;
942 	struct bkey_cached *ck, *n;
943 	struct rhash_head *pos;
944 	LIST_HEAD(items);
945 	unsigned i;
946 #ifdef __KERNEL__
947 	int cpu;
948 #endif
949 
950 	shrinker_free(bc->shrink);
951 
952 	mutex_lock(&bc->lock);
953 
954 	/*
955 	 * The loop is needed to guard against racing with rehash:
956 	 */
957 	while (atomic_long_read(&bc->nr_keys)) {
958 		rcu_read_lock();
959 		tbl = rht_dereference_rcu(bc->table.tbl, &bc->table);
960 		if (tbl)
961 			for (i = 0; i < tbl->size; i++)
962 				rht_for_each_entry_rcu(ck, pos, tbl, i, hash) {
963 					bkey_cached_evict(bc, ck);
964 					list_add(&ck->list, &items);
965 				}
966 		rcu_read_unlock();
967 	}
968 
969 #ifdef __KERNEL__
970 	for_each_possible_cpu(cpu) {
971 		struct btree_key_cache_freelist *f =
972 			per_cpu_ptr(bc->pcpu_freed, cpu);
973 
974 		for (i = 0; i < f->nr; i++) {
975 			ck = f->objs[i];
976 			list_add(&ck->list, &items);
977 		}
978 	}
979 #endif
980 
981 	BUG_ON(list_count_nodes(&bc->freed_pcpu) != bc->nr_freed_pcpu);
982 	BUG_ON(list_count_nodes(&bc->freed_nonpcpu) != bc->nr_freed_nonpcpu);
983 
984 	list_splice(&bc->freed_pcpu,	&items);
985 	list_splice(&bc->freed_nonpcpu,	&items);
986 
987 	mutex_unlock(&bc->lock);
988 
989 	list_for_each_entry_safe(ck, n, &items, list) {
990 		cond_resched();
991 
992 		list_del(&ck->list);
993 		kfree(ck->k);
994 		six_lock_exit(&ck->c.lock);
995 		kmem_cache_free(bch2_key_cache, ck);
996 	}
997 
998 	if (atomic_long_read(&bc->nr_dirty) &&
999 	    !bch2_journal_error(&c->journal) &&
1000 	    test_bit(BCH_FS_was_rw, &c->flags))
1001 		panic("btree key cache shutdown error: nr_dirty nonzero (%li)\n",
1002 		      atomic_long_read(&bc->nr_dirty));
1003 
1004 	if (atomic_long_read(&bc->nr_keys))
1005 		panic("btree key cache shutdown error: nr_keys nonzero (%li)\n",
1006 		      atomic_long_read(&bc->nr_keys));
1007 
1008 	if (bc->table_init_done)
1009 		rhashtable_destroy(&bc->table);
1010 
1011 	free_percpu(bc->pcpu_freed);
1012 }
1013 
1014 void bch2_fs_btree_key_cache_init_early(struct btree_key_cache *c)
1015 {
1016 	mutex_init(&c->lock);
1017 	INIT_LIST_HEAD(&c->freed_pcpu);
1018 	INIT_LIST_HEAD(&c->freed_nonpcpu);
1019 }
1020 
1021 int bch2_fs_btree_key_cache_init(struct btree_key_cache *bc)
1022 {
1023 	struct bch_fs *c = container_of(bc, struct bch_fs, btree_key_cache);
1024 	struct shrinker *shrink;
1025 
1026 #ifdef __KERNEL__
1027 	bc->pcpu_freed = alloc_percpu(struct btree_key_cache_freelist);
1028 	if (!bc->pcpu_freed)
1029 		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
1030 #endif
1031 
1032 	if (rhashtable_init(&bc->table, &bch2_btree_key_cache_params))
1033 		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
1034 
1035 	bc->table_init_done = true;
1036 
1037 	shrink = shrinker_alloc(0, "%s-btree_key_cache", c->name);
1038 	if (!shrink)
1039 		return -BCH_ERR_ENOMEM_fs_btree_cache_init;
1040 	bc->shrink = shrink;
1041 	shrink->seeks		= 0;
1042 	shrink->count_objects	= bch2_btree_key_cache_count;
1043 	shrink->scan_objects	= bch2_btree_key_cache_scan;
1044 	shrink->private_data	= c;
1045 	shrinker_register(shrink);
1046 	return 0;
1047 }
1048 
1049 void bch2_btree_key_cache_to_text(struct printbuf *out, struct btree_key_cache *c)
1050 {
1051 	prt_printf(out, "nr_freed:\t%lu",	atomic_long_read(&c->nr_freed));
1052 	prt_newline(out);
1053 	prt_printf(out, "nr_keys:\t%lu",	atomic_long_read(&c->nr_keys));
1054 	prt_newline(out);
1055 	prt_printf(out, "nr_dirty:\t%lu",	atomic_long_read(&c->nr_dirty));
1056 	prt_newline(out);
1057 }
1058 
1059 void bch2_btree_key_cache_exit(void)
1060 {
1061 	kmem_cache_destroy(bch2_key_cache);
1062 }
1063 
1064 int __init bch2_btree_key_cache_init(void)
1065 {
1066 	bch2_key_cache = KMEM_CACHE(bkey_cached, SLAB_RECLAIM_ACCOUNT);
1067 	if (!bch2_key_cache)
1068 		return -ENOMEM;
1069 
1070 	return 0;
1071 }
1072