xref: /linux/fs/bcachefs/btree_cache.c (revision 8886640dade4ae2595fcdce511c8bcc716aa47d3)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "bkey_buf.h"
5 #include "btree_cache.h"
6 #include "btree_io.h"
7 #include "btree_iter.h"
8 #include "btree_locking.h"
9 #include "debug.h"
10 #include "errcode.h"
11 #include "error.h"
12 #include "journal.h"
13 #include "trace.h"
14 
15 #include <linux/prefetch.h>
16 #include <linux/sched/mm.h>
17 
18 const char * const bch2_btree_node_flags[] = {
19 #define x(f)	#f,
20 	BTREE_FLAGS()
21 #undef x
22 	NULL
23 };
24 
25 void bch2_recalc_btree_reserve(struct bch_fs *c)
26 {
27 	unsigned i, reserve = 16;
28 
29 	if (!c->btree_roots_known[0].b)
30 		reserve += 8;
31 
32 	for (i = 0; i < btree_id_nr_alive(c); i++) {
33 		struct btree_root *r = bch2_btree_id_root(c, i);
34 
35 		if (r->b)
36 			reserve += min_t(unsigned, 1, r->b->c.level) * 8;
37 	}
38 
39 	c->btree_cache.reserve = reserve;
40 }
41 
42 static inline unsigned btree_cache_can_free(struct btree_cache *bc)
43 {
44 	return max_t(int, 0, bc->used - bc->reserve);
45 }
46 
47 static void btree_node_to_freedlist(struct btree_cache *bc, struct btree *b)
48 {
49 	if (b->c.lock.readers)
50 		list_move(&b->list, &bc->freed_pcpu);
51 	else
52 		list_move(&b->list, &bc->freed_nonpcpu);
53 }
54 
55 static void btree_node_data_free(struct bch_fs *c, struct btree *b)
56 {
57 	struct btree_cache *bc = &c->btree_cache;
58 
59 	EBUG_ON(btree_node_write_in_flight(b));
60 
61 	clear_btree_node_just_written(b);
62 
63 	kvpfree(b->data, btree_buf_bytes(b));
64 	b->data = NULL;
65 #ifdef __KERNEL__
66 	kvfree(b->aux_data);
67 #else
68 	munmap(b->aux_data, btree_aux_data_bytes(b));
69 #endif
70 	b->aux_data = NULL;
71 
72 	bc->used--;
73 
74 	btree_node_to_freedlist(bc, b);
75 }
76 
77 static int bch2_btree_cache_cmp_fn(struct rhashtable_compare_arg *arg,
78 				   const void *obj)
79 {
80 	const struct btree *b = obj;
81 	const u64 *v = arg->key;
82 
83 	return b->hash_val == *v ? 0 : 1;
84 }
85 
86 static const struct rhashtable_params bch_btree_cache_params = {
87 	.head_offset	= offsetof(struct btree, hash),
88 	.key_offset	= offsetof(struct btree, hash_val),
89 	.key_len	= sizeof(u64),
90 	.obj_cmpfn	= bch2_btree_cache_cmp_fn,
91 };
92 
93 static int btree_node_data_alloc(struct bch_fs *c, struct btree *b, gfp_t gfp)
94 {
95 	BUG_ON(b->data || b->aux_data);
96 
97 	b->data = kvpmalloc(btree_buf_bytes(b), gfp);
98 	if (!b->data)
99 		return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
100 #ifdef __KERNEL__
101 	b->aux_data = kvmalloc(btree_aux_data_bytes(b), gfp);
102 #else
103 	b->aux_data = mmap(NULL, btree_aux_data_bytes(b),
104 			   PROT_READ|PROT_WRITE|PROT_EXEC,
105 			   MAP_PRIVATE|MAP_ANONYMOUS, 0, 0);
106 	if (b->aux_data == MAP_FAILED)
107 		b->aux_data = NULL;
108 #endif
109 	if (!b->aux_data) {
110 		kvpfree(b->data, btree_buf_bytes(b));
111 		b->data = NULL;
112 		return -BCH_ERR_ENOMEM_btree_node_mem_alloc;
113 	}
114 
115 	return 0;
116 }
117 
118 static struct btree *__btree_node_mem_alloc(struct bch_fs *c, gfp_t gfp)
119 {
120 	struct btree *b;
121 
122 	b = kzalloc(sizeof(struct btree), gfp);
123 	if (!b)
124 		return NULL;
125 
126 	bkey_btree_ptr_init(&b->key);
127 	INIT_LIST_HEAD(&b->list);
128 	INIT_LIST_HEAD(&b->write_blocked);
129 	b->byte_order = ilog2(c->opts.btree_node_size);
130 	return b;
131 }
132 
133 struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
134 {
135 	struct btree_cache *bc = &c->btree_cache;
136 	struct btree *b;
137 
138 	b = __btree_node_mem_alloc(c, GFP_KERNEL);
139 	if (!b)
140 		return NULL;
141 
142 	if (btree_node_data_alloc(c, b, GFP_KERNEL)) {
143 		kfree(b);
144 		return NULL;
145 	}
146 
147 	bch2_btree_lock_init(&b->c, 0);
148 
149 	bc->used++;
150 	list_add(&b->list, &bc->freeable);
151 	return b;
152 }
153 
154 /* Btree in memory cache - hash table */
155 
156 void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
157 {
158 	int ret = rhashtable_remove_fast(&bc->table, &b->hash, bch_btree_cache_params);
159 
160 	BUG_ON(ret);
161 
162 	/* Cause future lookups for this node to fail: */
163 	b->hash_val = 0;
164 }
165 
166 int __bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b)
167 {
168 	BUG_ON(b->hash_val);
169 	b->hash_val = btree_ptr_hash_val(&b->key);
170 
171 	return rhashtable_lookup_insert_fast(&bc->table, &b->hash,
172 					     bch_btree_cache_params);
173 }
174 
175 int bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b,
176 				unsigned level, enum btree_id id)
177 {
178 	int ret;
179 
180 	b->c.level	= level;
181 	b->c.btree_id	= id;
182 
183 	mutex_lock(&bc->lock);
184 	ret = __bch2_btree_node_hash_insert(bc, b);
185 	if (!ret)
186 		list_add_tail(&b->list, &bc->live);
187 	mutex_unlock(&bc->lock);
188 
189 	return ret;
190 }
191 
192 __flatten
193 static inline struct btree *btree_cache_find(struct btree_cache *bc,
194 				     const struct bkey_i *k)
195 {
196 	u64 v = btree_ptr_hash_val(k);
197 
198 	return rhashtable_lookup_fast(&bc->table, &v, bch_btree_cache_params);
199 }
200 
201 /*
202  * this version is for btree nodes that have already been freed (we're not
203  * reaping a real btree node)
204  */
205 static int __btree_node_reclaim(struct bch_fs *c, struct btree *b, bool flush)
206 {
207 	struct btree_cache *bc = &c->btree_cache;
208 	int ret = 0;
209 
210 	lockdep_assert_held(&bc->lock);
211 wait_on_io:
212 	if (b->flags & ((1U << BTREE_NODE_dirty)|
213 			(1U << BTREE_NODE_read_in_flight)|
214 			(1U << BTREE_NODE_write_in_flight))) {
215 		if (!flush)
216 			return -BCH_ERR_ENOMEM_btree_node_reclaim;
217 
218 		/* XXX: waiting on IO with btree cache lock held */
219 		bch2_btree_node_wait_on_read(b);
220 		bch2_btree_node_wait_on_write(b);
221 	}
222 
223 	if (!six_trylock_intent(&b->c.lock))
224 		return -BCH_ERR_ENOMEM_btree_node_reclaim;
225 
226 	if (!six_trylock_write(&b->c.lock))
227 		goto out_unlock_intent;
228 
229 	/* recheck under lock */
230 	if (b->flags & ((1U << BTREE_NODE_read_in_flight)|
231 			(1U << BTREE_NODE_write_in_flight))) {
232 		if (!flush)
233 			goto out_unlock;
234 		six_unlock_write(&b->c.lock);
235 		six_unlock_intent(&b->c.lock);
236 		goto wait_on_io;
237 	}
238 
239 	if (btree_node_noevict(b) ||
240 	    btree_node_write_blocked(b) ||
241 	    btree_node_will_make_reachable(b))
242 		goto out_unlock;
243 
244 	if (btree_node_dirty(b)) {
245 		if (!flush)
246 			goto out_unlock;
247 		/*
248 		 * Using the underscore version because we don't want to compact
249 		 * bsets after the write, since this node is about to be evicted
250 		 * - unless btree verify mode is enabled, since it runs out of
251 		 * the post write cleanup:
252 		 */
253 		if (bch2_verify_btree_ondisk)
254 			bch2_btree_node_write(c, b, SIX_LOCK_intent,
255 					      BTREE_WRITE_cache_reclaim);
256 		else
257 			__bch2_btree_node_write(c, b,
258 						BTREE_WRITE_cache_reclaim);
259 
260 		six_unlock_write(&b->c.lock);
261 		six_unlock_intent(&b->c.lock);
262 		goto wait_on_io;
263 	}
264 out:
265 	if (b->hash_val && !ret)
266 		trace_and_count(c, btree_cache_reap, c, b);
267 	return ret;
268 out_unlock:
269 	six_unlock_write(&b->c.lock);
270 out_unlock_intent:
271 	six_unlock_intent(&b->c.lock);
272 	ret = -BCH_ERR_ENOMEM_btree_node_reclaim;
273 	goto out;
274 }
275 
276 static int btree_node_reclaim(struct bch_fs *c, struct btree *b)
277 {
278 	return __btree_node_reclaim(c, b, false);
279 }
280 
281 static int btree_node_write_and_reclaim(struct bch_fs *c, struct btree *b)
282 {
283 	return __btree_node_reclaim(c, b, true);
284 }
285 
286 static unsigned long bch2_btree_cache_scan(struct shrinker *shrink,
287 					   struct shrink_control *sc)
288 {
289 	struct bch_fs *c = shrink->private_data;
290 	struct btree_cache *bc = &c->btree_cache;
291 	struct btree *b, *t;
292 	unsigned long nr = sc->nr_to_scan;
293 	unsigned long can_free = 0;
294 	unsigned long freed = 0;
295 	unsigned long touched = 0;
296 	unsigned i, flags;
297 	unsigned long ret = SHRINK_STOP;
298 	bool trigger_writes = atomic_read(&bc->dirty) + nr >=
299 		bc->used * 3 / 4;
300 
301 	if (bch2_btree_shrinker_disabled)
302 		return SHRINK_STOP;
303 
304 	mutex_lock(&bc->lock);
305 	flags = memalloc_nofs_save();
306 
307 	/*
308 	 * It's _really_ critical that we don't free too many btree nodes - we
309 	 * have to always leave ourselves a reserve. The reserve is how we
310 	 * guarantee that allocating memory for a new btree node can always
311 	 * succeed, so that inserting keys into the btree can always succeed and
312 	 * IO can always make forward progress:
313 	 */
314 	can_free = btree_cache_can_free(bc);
315 	nr = min_t(unsigned long, nr, can_free);
316 
317 	i = 0;
318 	list_for_each_entry_safe(b, t, &bc->freeable, list) {
319 		/*
320 		 * Leave a few nodes on the freeable list, so that a btree split
321 		 * won't have to hit the system allocator:
322 		 */
323 		if (++i <= 3)
324 			continue;
325 
326 		touched++;
327 
328 		if (touched >= nr)
329 			goto out;
330 
331 		if (!btree_node_reclaim(c, b)) {
332 			btree_node_data_free(c, b);
333 			six_unlock_write(&b->c.lock);
334 			six_unlock_intent(&b->c.lock);
335 			freed++;
336 		}
337 	}
338 restart:
339 	list_for_each_entry_safe(b, t, &bc->live, list) {
340 		touched++;
341 
342 		if (btree_node_accessed(b)) {
343 			clear_btree_node_accessed(b);
344 		} else if (!btree_node_reclaim(c, b)) {
345 			freed++;
346 			btree_node_data_free(c, b);
347 
348 			bch2_btree_node_hash_remove(bc, b);
349 			six_unlock_write(&b->c.lock);
350 			six_unlock_intent(&b->c.lock);
351 
352 			if (freed == nr)
353 				goto out_rotate;
354 		} else if (trigger_writes &&
355 			   btree_node_dirty(b) &&
356 			   !btree_node_will_make_reachable(b) &&
357 			   !btree_node_write_blocked(b) &&
358 			   six_trylock_read(&b->c.lock)) {
359 			list_move(&bc->live, &b->list);
360 			mutex_unlock(&bc->lock);
361 			__bch2_btree_node_write(c, b, BTREE_WRITE_cache_reclaim);
362 			six_unlock_read(&b->c.lock);
363 			if (touched >= nr)
364 				goto out_nounlock;
365 			mutex_lock(&bc->lock);
366 			goto restart;
367 		}
368 
369 		if (touched >= nr)
370 			break;
371 	}
372 out_rotate:
373 	if (&t->list != &bc->live)
374 		list_move_tail(&bc->live, &t->list);
375 out:
376 	mutex_unlock(&bc->lock);
377 out_nounlock:
378 	ret = freed;
379 	memalloc_nofs_restore(flags);
380 	trace_and_count(c, btree_cache_scan, sc->nr_to_scan, can_free, ret);
381 	return ret;
382 }
383 
384 static unsigned long bch2_btree_cache_count(struct shrinker *shrink,
385 					    struct shrink_control *sc)
386 {
387 	struct bch_fs *c = shrink->private_data;
388 	struct btree_cache *bc = &c->btree_cache;
389 
390 	if (bch2_btree_shrinker_disabled)
391 		return 0;
392 
393 	return btree_cache_can_free(bc);
394 }
395 
396 void bch2_fs_btree_cache_exit(struct bch_fs *c)
397 {
398 	struct btree_cache *bc = &c->btree_cache;
399 	struct btree *b;
400 	unsigned i, flags;
401 
402 	shrinker_free(bc->shrink);
403 
404 	/* vfree() can allocate memory: */
405 	flags = memalloc_nofs_save();
406 	mutex_lock(&bc->lock);
407 
408 	if (c->verify_data)
409 		list_move(&c->verify_data->list, &bc->live);
410 
411 	kvpfree(c->verify_ondisk, c->opts.btree_node_size);
412 
413 	for (i = 0; i < btree_id_nr_alive(c); i++) {
414 		struct btree_root *r = bch2_btree_id_root(c, i);
415 
416 		if (r->b)
417 			list_add(&r->b->list, &bc->live);
418 	}
419 
420 	list_splice(&bc->freeable, &bc->live);
421 
422 	while (!list_empty(&bc->live)) {
423 		b = list_first_entry(&bc->live, struct btree, list);
424 
425 		BUG_ON(btree_node_read_in_flight(b) ||
426 		       btree_node_write_in_flight(b));
427 
428 		btree_node_data_free(c, b);
429 	}
430 
431 	BUG_ON(!bch2_journal_error(&c->journal) &&
432 	       atomic_read(&c->btree_cache.dirty));
433 
434 	list_splice(&bc->freed_pcpu, &bc->freed_nonpcpu);
435 
436 	while (!list_empty(&bc->freed_nonpcpu)) {
437 		b = list_first_entry(&bc->freed_nonpcpu, struct btree, list);
438 		list_del(&b->list);
439 		six_lock_exit(&b->c.lock);
440 		kfree(b);
441 	}
442 
443 	mutex_unlock(&bc->lock);
444 	memalloc_nofs_restore(flags);
445 
446 	if (bc->table_init_done)
447 		rhashtable_destroy(&bc->table);
448 }
449 
450 int bch2_fs_btree_cache_init(struct bch_fs *c)
451 {
452 	struct btree_cache *bc = &c->btree_cache;
453 	struct shrinker *shrink;
454 	unsigned i;
455 	int ret = 0;
456 
457 	ret = rhashtable_init(&bc->table, &bch_btree_cache_params);
458 	if (ret)
459 		goto err;
460 
461 	bc->table_init_done = true;
462 
463 	bch2_recalc_btree_reserve(c);
464 
465 	for (i = 0; i < bc->reserve; i++)
466 		if (!__bch2_btree_node_mem_alloc(c))
467 			goto err;
468 
469 	list_splice_init(&bc->live, &bc->freeable);
470 
471 	mutex_init(&c->verify_lock);
472 
473 	shrink = shrinker_alloc(0, "%s-btree_cache", c->name);
474 	if (!shrink)
475 		goto err;
476 	bc->shrink = shrink;
477 	shrink->count_objects	= bch2_btree_cache_count;
478 	shrink->scan_objects	= bch2_btree_cache_scan;
479 	shrink->seeks		= 4;
480 	shrink->private_data	= c;
481 	shrinker_register(shrink);
482 
483 	return 0;
484 err:
485 	return -BCH_ERR_ENOMEM_fs_btree_cache_init;
486 }
487 
488 void bch2_fs_btree_cache_init_early(struct btree_cache *bc)
489 {
490 	mutex_init(&bc->lock);
491 	INIT_LIST_HEAD(&bc->live);
492 	INIT_LIST_HEAD(&bc->freeable);
493 	INIT_LIST_HEAD(&bc->freed_pcpu);
494 	INIT_LIST_HEAD(&bc->freed_nonpcpu);
495 }
496 
497 /*
498  * We can only have one thread cannibalizing other cached btree nodes at a time,
499  * or we'll deadlock. We use an open coded mutex to ensure that, which a
500  * cannibalize_bucket() will take. This means every time we unlock the root of
501  * the btree, we need to release this lock if we have it held.
502  */
503 void bch2_btree_cache_cannibalize_unlock(struct btree_trans *trans)
504 {
505 	struct bch_fs *c = trans->c;
506 	struct btree_cache *bc = &c->btree_cache;
507 
508 	if (bc->alloc_lock == current) {
509 		trace_and_count(c, btree_cache_cannibalize_unlock, trans);
510 		bc->alloc_lock = NULL;
511 		closure_wake_up(&bc->alloc_wait);
512 	}
513 }
514 
515 int bch2_btree_cache_cannibalize_lock(struct btree_trans *trans, struct closure *cl)
516 {
517 	struct bch_fs *c = trans->c;
518 	struct btree_cache *bc = &c->btree_cache;
519 	struct task_struct *old;
520 
521 	old = cmpxchg(&bc->alloc_lock, NULL, current);
522 	if (old == NULL || old == current)
523 		goto success;
524 
525 	if (!cl) {
526 		trace_and_count(c, btree_cache_cannibalize_lock_fail, trans);
527 		return -BCH_ERR_ENOMEM_btree_cache_cannibalize_lock;
528 	}
529 
530 	closure_wait(&bc->alloc_wait, cl);
531 
532 	/* Try again, after adding ourselves to waitlist */
533 	old = cmpxchg(&bc->alloc_lock, NULL, current);
534 	if (old == NULL || old == current) {
535 		/* We raced */
536 		closure_wake_up(&bc->alloc_wait);
537 		goto success;
538 	}
539 
540 	trace_and_count(c, btree_cache_cannibalize_lock_fail, trans);
541 	return -BCH_ERR_btree_cache_cannibalize_lock_blocked;
542 
543 success:
544 	trace_and_count(c, btree_cache_cannibalize_lock, trans);
545 	return 0;
546 }
547 
548 static struct btree *btree_node_cannibalize(struct bch_fs *c)
549 {
550 	struct btree_cache *bc = &c->btree_cache;
551 	struct btree *b;
552 
553 	list_for_each_entry_reverse(b, &bc->live, list)
554 		if (!btree_node_reclaim(c, b))
555 			return b;
556 
557 	while (1) {
558 		list_for_each_entry_reverse(b, &bc->live, list)
559 			if (!btree_node_write_and_reclaim(c, b))
560 				return b;
561 
562 		/*
563 		 * Rare case: all nodes were intent-locked.
564 		 * Just busy-wait.
565 		 */
566 		WARN_ONCE(1, "btree cache cannibalize failed\n");
567 		cond_resched();
568 	}
569 }
570 
571 struct btree *bch2_btree_node_mem_alloc(struct btree_trans *trans, bool pcpu_read_locks)
572 {
573 	struct bch_fs *c = trans->c;
574 	struct btree_cache *bc = &c->btree_cache;
575 	struct list_head *freed = pcpu_read_locks
576 		? &bc->freed_pcpu
577 		: &bc->freed_nonpcpu;
578 	struct btree *b, *b2;
579 	u64 start_time = local_clock();
580 	unsigned flags;
581 
582 	flags = memalloc_nofs_save();
583 	mutex_lock(&bc->lock);
584 
585 	/*
586 	 * We never free struct btree itself, just the memory that holds the on
587 	 * disk node. Check the freed list before allocating a new one:
588 	 */
589 	list_for_each_entry(b, freed, list)
590 		if (!btree_node_reclaim(c, b)) {
591 			list_del_init(&b->list);
592 			goto got_node;
593 		}
594 
595 	b = __btree_node_mem_alloc(c, GFP_NOWAIT|__GFP_NOWARN);
596 	if (!b) {
597 		mutex_unlock(&bc->lock);
598 		bch2_trans_unlock(trans);
599 		b = __btree_node_mem_alloc(c, GFP_KERNEL);
600 		if (!b)
601 			goto err;
602 		mutex_lock(&bc->lock);
603 	}
604 
605 	bch2_btree_lock_init(&b->c, pcpu_read_locks ? SIX_LOCK_INIT_PCPU : 0);
606 
607 	BUG_ON(!six_trylock_intent(&b->c.lock));
608 	BUG_ON(!six_trylock_write(&b->c.lock));
609 got_node:
610 
611 	/*
612 	 * btree_free() doesn't free memory; it sticks the node on the end of
613 	 * the list. Check if there's any freed nodes there:
614 	 */
615 	list_for_each_entry(b2, &bc->freeable, list)
616 		if (!btree_node_reclaim(c, b2)) {
617 			swap(b->data, b2->data);
618 			swap(b->aux_data, b2->aux_data);
619 			btree_node_to_freedlist(bc, b2);
620 			six_unlock_write(&b2->c.lock);
621 			six_unlock_intent(&b2->c.lock);
622 			goto got_mem;
623 		}
624 
625 	mutex_unlock(&bc->lock);
626 
627 	if (btree_node_data_alloc(c, b, GFP_NOWAIT|__GFP_NOWARN)) {
628 		bch2_trans_unlock(trans);
629 		if (btree_node_data_alloc(c, b, GFP_KERNEL|__GFP_NOWARN))
630 			goto err;
631 	}
632 
633 	mutex_lock(&bc->lock);
634 	bc->used++;
635 got_mem:
636 	mutex_unlock(&bc->lock);
637 
638 	BUG_ON(btree_node_hashed(b));
639 	BUG_ON(btree_node_dirty(b));
640 	BUG_ON(btree_node_write_in_flight(b));
641 out:
642 	b->flags		= 0;
643 	b->written		= 0;
644 	b->nsets		= 0;
645 	b->sib_u64s[0]		= 0;
646 	b->sib_u64s[1]		= 0;
647 	b->whiteout_u64s	= 0;
648 	bch2_btree_keys_init(b);
649 	set_btree_node_accessed(b);
650 
651 	bch2_time_stats_update(&c->times[BCH_TIME_btree_node_mem_alloc],
652 			       start_time);
653 
654 	memalloc_nofs_restore(flags);
655 	return b;
656 err:
657 	mutex_lock(&bc->lock);
658 
659 	/* Try to cannibalize another cached btree node: */
660 	if (bc->alloc_lock == current) {
661 		b2 = btree_node_cannibalize(c);
662 		clear_btree_node_just_written(b2);
663 		bch2_btree_node_hash_remove(bc, b2);
664 
665 		if (b) {
666 			swap(b->data, b2->data);
667 			swap(b->aux_data, b2->aux_data);
668 			btree_node_to_freedlist(bc, b2);
669 			six_unlock_write(&b2->c.lock);
670 			six_unlock_intent(&b2->c.lock);
671 		} else {
672 			b = b2;
673 			list_del_init(&b->list);
674 		}
675 
676 		mutex_unlock(&bc->lock);
677 
678 		trace_and_count(c, btree_cache_cannibalize, trans);
679 		goto out;
680 	}
681 
682 	mutex_unlock(&bc->lock);
683 	memalloc_nofs_restore(flags);
684 	return ERR_PTR(-BCH_ERR_ENOMEM_btree_node_mem_alloc);
685 }
686 
687 /* Slowpath, don't want it inlined into btree_iter_traverse() */
688 static noinline struct btree *bch2_btree_node_fill(struct btree_trans *trans,
689 				struct btree_path *path,
690 				const struct bkey_i *k,
691 				enum btree_id btree_id,
692 				unsigned level,
693 				enum six_lock_type lock_type,
694 				bool sync)
695 {
696 	struct bch_fs *c = trans->c;
697 	struct btree_cache *bc = &c->btree_cache;
698 	struct btree *b;
699 	u32 seq;
700 
701 	BUG_ON(level + 1 >= BTREE_MAX_DEPTH);
702 	/*
703 	 * Parent node must be locked, else we could read in a btree node that's
704 	 * been freed:
705 	 */
706 	if (path && !bch2_btree_node_relock(trans, path, level + 1)) {
707 		trace_and_count(c, trans_restart_relock_parent_for_fill, trans, _THIS_IP_, path);
708 		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_fill_relock));
709 	}
710 
711 	b = bch2_btree_node_mem_alloc(trans, level != 0);
712 
713 	if (bch2_err_matches(PTR_ERR_OR_ZERO(b), ENOMEM)) {
714 		trans->memory_allocation_failure = true;
715 		trace_and_count(c, trans_restart_memory_allocation_failure, trans, _THIS_IP_, path);
716 		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_fill_mem_alloc_fail));
717 	}
718 
719 	if (IS_ERR(b))
720 		return b;
721 
722 	bkey_copy(&b->key, k);
723 	if (bch2_btree_node_hash_insert(bc, b, level, btree_id)) {
724 		/* raced with another fill: */
725 
726 		/* mark as unhashed... */
727 		b->hash_val = 0;
728 
729 		mutex_lock(&bc->lock);
730 		list_add(&b->list, &bc->freeable);
731 		mutex_unlock(&bc->lock);
732 
733 		six_unlock_write(&b->c.lock);
734 		six_unlock_intent(&b->c.lock);
735 		return NULL;
736 	}
737 
738 	set_btree_node_read_in_flight(b);
739 
740 	six_unlock_write(&b->c.lock);
741 	seq = six_lock_seq(&b->c.lock);
742 	six_unlock_intent(&b->c.lock);
743 
744 	/* Unlock before doing IO: */
745 	if (path && sync)
746 		bch2_trans_unlock_noassert(trans);
747 
748 	bch2_btree_node_read(trans, b, sync);
749 
750 	if (!sync)
751 		return NULL;
752 
753 	if (path) {
754 		int ret = bch2_trans_relock(trans) ?:
755 			bch2_btree_path_relock_intent(trans, path);
756 		if (ret) {
757 			BUG_ON(!trans->restarted);
758 			return ERR_PTR(ret);
759 		}
760 	}
761 
762 	if (!six_relock_type(&b->c.lock, lock_type, seq)) {
763 		if (path)
764 			trace_and_count(c, trans_restart_relock_after_fill, trans, _THIS_IP_, path);
765 		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_relock_after_fill));
766 	}
767 
768 	return b;
769 }
770 
771 static noinline void btree_bad_header(struct bch_fs *c, struct btree *b)
772 {
773 	struct printbuf buf = PRINTBUF;
774 
775 	if (c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_allocations)
776 		return;
777 
778 	prt_printf(&buf,
779 	       "btree node header doesn't match ptr\n"
780 	       "btree %s level %u\n"
781 	       "ptr: ",
782 	       bch2_btree_id_str(b->c.btree_id), b->c.level);
783 	bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
784 
785 	prt_printf(&buf, "\nheader: btree %s level %llu\n"
786 	       "min ",
787 	       bch2_btree_id_str(BTREE_NODE_ID(b->data)),
788 	       BTREE_NODE_LEVEL(b->data));
789 	bch2_bpos_to_text(&buf, b->data->min_key);
790 
791 	prt_printf(&buf, "\nmax ");
792 	bch2_bpos_to_text(&buf, b->data->max_key);
793 
794 	bch2_fs_inconsistent(c, "%s", buf.buf);
795 	printbuf_exit(&buf);
796 }
797 
798 static inline void btree_check_header(struct bch_fs *c, struct btree *b)
799 {
800 	if (b->c.btree_id != BTREE_NODE_ID(b->data) ||
801 	    b->c.level != BTREE_NODE_LEVEL(b->data) ||
802 	    !bpos_eq(b->data->max_key, b->key.k.p) ||
803 	    (b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
804 	     !bpos_eq(b->data->min_key,
805 		      bkey_i_to_btree_ptr_v2(&b->key)->v.min_key)))
806 		btree_bad_header(c, b);
807 }
808 
809 static struct btree *__bch2_btree_node_get(struct btree_trans *trans, struct btree_path *path,
810 					   const struct bkey_i *k, unsigned level,
811 					   enum six_lock_type lock_type,
812 					   unsigned long trace_ip)
813 {
814 	struct bch_fs *c = trans->c;
815 	struct btree_cache *bc = &c->btree_cache;
816 	struct btree *b;
817 	struct bset_tree *t;
818 	bool need_relock = false;
819 	int ret;
820 
821 	EBUG_ON(level >= BTREE_MAX_DEPTH);
822 retry:
823 	b = btree_cache_find(bc, k);
824 	if (unlikely(!b)) {
825 		/*
826 		 * We must have the parent locked to call bch2_btree_node_fill(),
827 		 * else we could read in a btree node from disk that's been
828 		 * freed:
829 		 */
830 		b = bch2_btree_node_fill(trans, path, k, path->btree_id,
831 					 level, lock_type, true);
832 		need_relock = true;
833 
834 		/* We raced and found the btree node in the cache */
835 		if (!b)
836 			goto retry;
837 
838 		if (IS_ERR(b))
839 			return b;
840 	} else {
841 		if (btree_node_read_locked(path, level + 1))
842 			btree_node_unlock(trans, path, level + 1);
843 
844 		ret = btree_node_lock(trans, path, &b->c, level, lock_type, trace_ip);
845 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
846 			return ERR_PTR(ret);
847 
848 		BUG_ON(ret);
849 
850 		if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
851 			     b->c.level != level ||
852 			     race_fault())) {
853 			six_unlock_type(&b->c.lock, lock_type);
854 			if (bch2_btree_node_relock(trans, path, level + 1))
855 				goto retry;
856 
857 			trace_and_count(c, trans_restart_btree_node_reused, trans, trace_ip, path);
858 			return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_lock_node_reused));
859 		}
860 
861 		/* avoid atomic set bit if it's not needed: */
862 		if (!btree_node_accessed(b))
863 			set_btree_node_accessed(b);
864 	}
865 
866 	if (unlikely(btree_node_read_in_flight(b))) {
867 		u32 seq = six_lock_seq(&b->c.lock);
868 
869 		six_unlock_type(&b->c.lock, lock_type);
870 		bch2_trans_unlock(trans);
871 		need_relock = true;
872 
873 		bch2_btree_node_wait_on_read(b);
874 
875 		/*
876 		 * should_be_locked is not set on this path yet, so we need to
877 		 * relock it specifically:
878 		 */
879 		if (!six_relock_type(&b->c.lock, lock_type, seq))
880 			goto retry;
881 	}
882 
883 	if (unlikely(need_relock)) {
884 		ret = bch2_trans_relock(trans) ?:
885 			bch2_btree_path_relock_intent(trans, path);
886 		if (ret) {
887 			six_unlock_type(&b->c.lock, lock_type);
888 			return ERR_PTR(ret);
889 		}
890 	}
891 
892 	prefetch(b->aux_data);
893 
894 	for_each_bset(b, t) {
895 		void *p = (u64 *) b->aux_data + t->aux_data_offset;
896 
897 		prefetch(p + L1_CACHE_BYTES * 0);
898 		prefetch(p + L1_CACHE_BYTES * 1);
899 		prefetch(p + L1_CACHE_BYTES * 2);
900 	}
901 
902 	if (unlikely(btree_node_read_error(b))) {
903 		six_unlock_type(&b->c.lock, lock_type);
904 		return ERR_PTR(-EIO);
905 	}
906 
907 	EBUG_ON(b->c.btree_id != path->btree_id);
908 	EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
909 	btree_check_header(c, b);
910 
911 	return b;
912 }
913 
914 /**
915  * bch2_btree_node_get - find a btree node in the cache and lock it, reading it
916  * in from disk if necessary.
917  *
918  * @trans:	btree transaction object
919  * @path:	btree_path being traversed
920  * @k:		pointer to btree node (generally KEY_TYPE_btree_ptr_v2)
921  * @level:	level of btree node being looked up (0 == leaf node)
922  * @lock_type:	SIX_LOCK_read or SIX_LOCK_intent
923  * @trace_ip:	ip of caller of btree iterator code (i.e. caller of bch2_btree_iter_peek())
924  *
925  * The btree node will have either a read or a write lock held, depending on
926  * the @write parameter.
927  *
928  * Returns: btree node or ERR_PTR()
929  */
930 struct btree *bch2_btree_node_get(struct btree_trans *trans, struct btree_path *path,
931 				  const struct bkey_i *k, unsigned level,
932 				  enum six_lock_type lock_type,
933 				  unsigned long trace_ip)
934 {
935 	struct bch_fs *c = trans->c;
936 	struct btree *b;
937 	struct bset_tree *t;
938 	int ret;
939 
940 	EBUG_ON(level >= BTREE_MAX_DEPTH);
941 
942 	b = btree_node_mem_ptr(k);
943 
944 	/*
945 	 * Check b->hash_val _before_ calling btree_node_lock() - this might not
946 	 * be the node we want anymore, and trying to lock the wrong node could
947 	 * cause an unneccessary transaction restart:
948 	 */
949 	if (unlikely(!c->opts.btree_node_mem_ptr_optimization ||
950 		     !b ||
951 		     b->hash_val != btree_ptr_hash_val(k)))
952 		return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
953 
954 	if (btree_node_read_locked(path, level + 1))
955 		btree_node_unlock(trans, path, level + 1);
956 
957 	ret = btree_node_lock(trans, path, &b->c, level, lock_type, trace_ip);
958 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
959 		return ERR_PTR(ret);
960 
961 	BUG_ON(ret);
962 
963 	if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
964 		     b->c.level != level ||
965 		     race_fault())) {
966 		six_unlock_type(&b->c.lock, lock_type);
967 		if (bch2_btree_node_relock(trans, path, level + 1))
968 			return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
969 
970 		trace_and_count(c, trans_restart_btree_node_reused, trans, trace_ip, path);
971 		return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_lock_node_reused));
972 	}
973 
974 	if (unlikely(btree_node_read_in_flight(b))) {
975 		six_unlock_type(&b->c.lock, lock_type);
976 		return __bch2_btree_node_get(trans, path, k, level, lock_type, trace_ip);
977 	}
978 
979 	prefetch(b->aux_data);
980 
981 	for_each_bset(b, t) {
982 		void *p = (u64 *) b->aux_data + t->aux_data_offset;
983 
984 		prefetch(p + L1_CACHE_BYTES * 0);
985 		prefetch(p + L1_CACHE_BYTES * 1);
986 		prefetch(p + L1_CACHE_BYTES * 2);
987 	}
988 
989 	/* avoid atomic set bit if it's not needed: */
990 	if (!btree_node_accessed(b))
991 		set_btree_node_accessed(b);
992 
993 	if (unlikely(btree_node_read_error(b))) {
994 		six_unlock_type(&b->c.lock, lock_type);
995 		return ERR_PTR(-EIO);
996 	}
997 
998 	EBUG_ON(b->c.btree_id != path->btree_id);
999 	EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
1000 	btree_check_header(c, b);
1001 
1002 	return b;
1003 }
1004 
1005 struct btree *bch2_btree_node_get_noiter(struct btree_trans *trans,
1006 					 const struct bkey_i *k,
1007 					 enum btree_id btree_id,
1008 					 unsigned level,
1009 					 bool nofill)
1010 {
1011 	struct bch_fs *c = trans->c;
1012 	struct btree_cache *bc = &c->btree_cache;
1013 	struct btree *b;
1014 	struct bset_tree *t;
1015 	int ret;
1016 
1017 	EBUG_ON(level >= BTREE_MAX_DEPTH);
1018 
1019 	if (c->opts.btree_node_mem_ptr_optimization) {
1020 		b = btree_node_mem_ptr(k);
1021 		if (b)
1022 			goto lock_node;
1023 	}
1024 retry:
1025 	b = btree_cache_find(bc, k);
1026 	if (unlikely(!b)) {
1027 		if (nofill)
1028 			goto out;
1029 
1030 		b = bch2_btree_node_fill(trans, NULL, k, btree_id,
1031 					 level, SIX_LOCK_read, true);
1032 
1033 		/* We raced and found the btree node in the cache */
1034 		if (!b)
1035 			goto retry;
1036 
1037 		if (IS_ERR(b) &&
1038 		    !bch2_btree_cache_cannibalize_lock(trans, NULL))
1039 			goto retry;
1040 
1041 		if (IS_ERR(b))
1042 			goto out;
1043 	} else {
1044 lock_node:
1045 		ret = btree_node_lock_nopath(trans, &b->c, SIX_LOCK_read, _THIS_IP_);
1046 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1047 			return ERR_PTR(ret);
1048 
1049 		BUG_ON(ret);
1050 
1051 		if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
1052 			     b->c.btree_id != btree_id ||
1053 			     b->c.level != level)) {
1054 			six_unlock_read(&b->c.lock);
1055 			goto retry;
1056 		}
1057 	}
1058 
1059 	/* XXX: waiting on IO with btree locks held: */
1060 	__bch2_btree_node_wait_on_read(b);
1061 
1062 	prefetch(b->aux_data);
1063 
1064 	for_each_bset(b, t) {
1065 		void *p = (u64 *) b->aux_data + t->aux_data_offset;
1066 
1067 		prefetch(p + L1_CACHE_BYTES * 0);
1068 		prefetch(p + L1_CACHE_BYTES * 1);
1069 		prefetch(p + L1_CACHE_BYTES * 2);
1070 	}
1071 
1072 	/* avoid atomic set bit if it's not needed: */
1073 	if (!btree_node_accessed(b))
1074 		set_btree_node_accessed(b);
1075 
1076 	if (unlikely(btree_node_read_error(b))) {
1077 		six_unlock_read(&b->c.lock);
1078 		b = ERR_PTR(-EIO);
1079 		goto out;
1080 	}
1081 
1082 	EBUG_ON(b->c.btree_id != btree_id);
1083 	EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
1084 	btree_check_header(c, b);
1085 out:
1086 	bch2_btree_cache_cannibalize_unlock(trans);
1087 	return b;
1088 }
1089 
1090 int bch2_btree_node_prefetch(struct btree_trans *trans,
1091 			     struct btree_path *path,
1092 			     const struct bkey_i *k,
1093 			     enum btree_id btree_id, unsigned level)
1094 {
1095 	struct bch_fs *c = trans->c;
1096 	struct btree_cache *bc = &c->btree_cache;
1097 	struct btree *b;
1098 
1099 	BUG_ON(trans && !btree_node_locked(path, level + 1));
1100 	BUG_ON(level >= BTREE_MAX_DEPTH);
1101 
1102 	b = btree_cache_find(bc, k);
1103 	if (b)
1104 		return 0;
1105 
1106 	b = bch2_btree_node_fill(trans, path, k, btree_id,
1107 				 level, SIX_LOCK_read, false);
1108 	return PTR_ERR_OR_ZERO(b);
1109 }
1110 
1111 void bch2_btree_node_evict(struct btree_trans *trans, const struct bkey_i *k)
1112 {
1113 	struct bch_fs *c = trans->c;
1114 	struct btree_cache *bc = &c->btree_cache;
1115 	struct btree *b;
1116 
1117 	b = btree_cache_find(bc, k);
1118 	if (!b)
1119 		return;
1120 wait_on_io:
1121 	/* not allowed to wait on io with btree locks held: */
1122 
1123 	/* XXX we're called from btree_gc which will be holding other btree
1124 	 * nodes locked
1125 	 */
1126 	__bch2_btree_node_wait_on_read(b);
1127 	__bch2_btree_node_wait_on_write(b);
1128 
1129 	btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
1130 	btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
1131 
1132 	if (btree_node_dirty(b)) {
1133 		__bch2_btree_node_write(c, b, BTREE_WRITE_cache_reclaim);
1134 		six_unlock_write(&b->c.lock);
1135 		six_unlock_intent(&b->c.lock);
1136 		goto wait_on_io;
1137 	}
1138 
1139 	BUG_ON(btree_node_dirty(b));
1140 
1141 	mutex_lock(&bc->lock);
1142 	btree_node_data_free(c, b);
1143 	bch2_btree_node_hash_remove(bc, b);
1144 	mutex_unlock(&bc->lock);
1145 
1146 	six_unlock_write(&b->c.lock);
1147 	six_unlock_intent(&b->c.lock);
1148 }
1149 
1150 const char *bch2_btree_id_str(enum btree_id btree)
1151 {
1152 	return btree < BTREE_ID_NR ? __bch2_btree_ids[btree] : "(unknown)";
1153 }
1154 
1155 void bch2_btree_pos_to_text(struct printbuf *out, struct bch_fs *c, const struct btree *b)
1156 {
1157 	prt_printf(out, "%s level %u/%u\n  ",
1158 	       bch2_btree_id_str(b->c.btree_id),
1159 	       b->c.level,
1160 	       bch2_btree_id_root(c, b->c.btree_id)->level);
1161 	bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
1162 }
1163 
1164 void bch2_btree_node_to_text(struct printbuf *out, struct bch_fs *c, const struct btree *b)
1165 {
1166 	struct bset_stats stats;
1167 
1168 	memset(&stats, 0, sizeof(stats));
1169 
1170 	bch2_btree_keys_stats(b, &stats);
1171 
1172 	prt_printf(out, "l %u ", b->c.level);
1173 	bch2_bpos_to_text(out, b->data->min_key);
1174 	prt_printf(out, " - ");
1175 	bch2_bpos_to_text(out, b->data->max_key);
1176 	prt_printf(out, ":\n"
1177 	       "    ptrs: ");
1178 	bch2_val_to_text(out, c, bkey_i_to_s_c(&b->key));
1179 	prt_newline(out);
1180 
1181 	prt_printf(out,
1182 	       "    format: ");
1183 	bch2_bkey_format_to_text(out, &b->format);
1184 
1185 	prt_printf(out,
1186 	       "    unpack fn len: %u\n"
1187 	       "    bytes used %zu/%zu (%zu%% full)\n"
1188 	       "    sib u64s: %u, %u (merge threshold %u)\n"
1189 	       "    nr packed keys %u\n"
1190 	       "    nr unpacked keys %u\n"
1191 	       "    floats %zu\n"
1192 	       "    failed unpacked %zu\n",
1193 	       b->unpack_fn_len,
1194 	       b->nr.live_u64s * sizeof(u64),
1195 	       btree_buf_bytes(b) - sizeof(struct btree_node),
1196 	       b->nr.live_u64s * 100 / btree_max_u64s(c),
1197 	       b->sib_u64s[0],
1198 	       b->sib_u64s[1],
1199 	       c->btree_foreground_merge_threshold,
1200 	       b->nr.packed_keys,
1201 	       b->nr.unpacked_keys,
1202 	       stats.floats,
1203 	       stats.failed);
1204 }
1205 
1206 void bch2_btree_cache_to_text(struct printbuf *out, const struct bch_fs *c)
1207 {
1208 	prt_printf(out, "nr nodes:\t\t%u\n", c->btree_cache.used);
1209 	prt_printf(out, "nr dirty:\t\t%u\n", atomic_read(&c->btree_cache.dirty));
1210 	prt_printf(out, "cannibalize lock:\t%p\n", c->btree_cache.alloc_lock);
1211 }
1212