xref: /linux/mm/zswap.c (revision 5ef12cb4a3a78ffb331c03a795a15eea4ae35155)
1 /*
2  * zswap.c - zswap driver file
3  *
4  * zswap is a backend for frontswap that takes pages that are in the process
5  * of being swapped out and attempts to compress and store them in a
6  * RAM-based memory pool.  This can result in a significant I/O reduction on
7  * the swap device and, in the case where decompressing from RAM is faster
8  * than reading from the swap device, can also improve workload performance.
9  *
10  * Copyright (C) 2012  Seth Jennings <sjenning@linux.vnet.ibm.com>
11  *
12  * This program is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU General Public License
14  * as published by the Free Software Foundation; either version 2
15  * of the License, or (at your option) any later version.
16  *
17  * This program is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20  * GNU General Public License for more details.
21 */
22 
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24 
25 #include <linux/module.h>
26 #include <linux/cpu.h>
27 #include <linux/highmem.h>
28 #include <linux/slab.h>
29 #include <linux/spinlock.h>
30 #include <linux/types.h>
31 #include <linux/atomic.h>
32 #include <linux/frontswap.h>
33 #include <linux/rbtree.h>
34 #include <linux/swap.h>
35 #include <linux/crypto.h>
36 #include <linux/mempool.h>
37 #include <linux/zpool.h>
38 
39 #include <linux/mm_types.h>
40 #include <linux/page-flags.h>
41 #include <linux/swapops.h>
42 #include <linux/writeback.h>
43 #include <linux/pagemap.h>
44 
45 /*********************************
46 * statistics
47 **********************************/
48 /* Total bytes used by the compressed storage */
49 static u64 zswap_pool_total_size;
50 /* The number of compressed pages currently stored in zswap */
51 static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
52 /* The number of same-value filled pages currently stored in zswap */
53 static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
54 
55 /*
56  * The statistics below are not protected from concurrent access for
57  * performance reasons so they may not be a 100% accurate.  However,
58  * they do provide useful information on roughly how many times a
59  * certain event is occurring.
60 */
61 
62 /* Pool limit was hit (see zswap_max_pool_percent) */
63 static u64 zswap_pool_limit_hit;
64 /* Pages written back when pool limit was reached */
65 static u64 zswap_written_back_pages;
66 /* Store failed due to a reclaim failure after pool limit was reached */
67 static u64 zswap_reject_reclaim_fail;
68 /* Compressed page was too big for the allocator to (optimally) store */
69 static u64 zswap_reject_compress_poor;
70 /* Store failed because underlying allocator could not get memory */
71 static u64 zswap_reject_alloc_fail;
72 /* Store failed because the entry metadata could not be allocated (rare) */
73 static u64 zswap_reject_kmemcache_fail;
74 /* Duplicate store was encountered (rare) */
75 static u64 zswap_duplicate_entry;
76 
77 /*********************************
78 * tunables
79 **********************************/
80 
81 #define ZSWAP_PARAM_UNSET ""
82 
83 /* Enable/disable zswap (disabled by default) */
84 static bool zswap_enabled;
85 static int zswap_enabled_param_set(const char *,
86 				   const struct kernel_param *);
87 static struct kernel_param_ops zswap_enabled_param_ops = {
88 	.set =		zswap_enabled_param_set,
89 	.get =		param_get_bool,
90 };
91 module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
92 
93 /* Crypto compressor to use */
94 #define ZSWAP_COMPRESSOR_DEFAULT "lzo"
95 static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
96 static int zswap_compressor_param_set(const char *,
97 				      const struct kernel_param *);
98 static struct kernel_param_ops zswap_compressor_param_ops = {
99 	.set =		zswap_compressor_param_set,
100 	.get =		param_get_charp,
101 	.free =		param_free_charp,
102 };
103 module_param_cb(compressor, &zswap_compressor_param_ops,
104 		&zswap_compressor, 0644);
105 
106 /* Compressed storage zpool to use */
107 #define ZSWAP_ZPOOL_DEFAULT "zbud"
108 static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
109 static int zswap_zpool_param_set(const char *, const struct kernel_param *);
110 static struct kernel_param_ops zswap_zpool_param_ops = {
111 	.set =		zswap_zpool_param_set,
112 	.get =		param_get_charp,
113 	.free =		param_free_charp,
114 };
115 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
116 
117 /* The maximum percentage of memory that the compressed pool can occupy */
118 static unsigned int zswap_max_pool_percent = 20;
119 module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
120 
121 /* Enable/disable handling same-value filled pages (enabled by default) */
122 static bool zswap_same_filled_pages_enabled = true;
123 module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
124 		   bool, 0644);
125 
126 /*********************************
127 * data structures
128 **********************************/
129 
130 struct zswap_pool {
131 	struct zpool *zpool;
132 	struct crypto_comp * __percpu *tfm;
133 	struct kref kref;
134 	struct list_head list;
135 	struct work_struct work;
136 	struct hlist_node node;
137 	char tfm_name[CRYPTO_MAX_ALG_NAME];
138 };
139 
140 /*
141  * struct zswap_entry
142  *
143  * This structure contains the metadata for tracking a single compressed
144  * page within zswap.
145  *
146  * rbnode - links the entry into red-black tree for the appropriate swap type
147  * offset - the swap offset for the entry.  Index into the red-black tree.
148  * refcount - the number of outstanding reference to the entry. This is needed
149  *            to protect against premature freeing of the entry by code
150  *            concurrent calls to load, invalidate, and writeback.  The lock
151  *            for the zswap_tree structure that contains the entry must
152  *            be held while changing the refcount.  Since the lock must
153  *            be held, there is no reason to also make refcount atomic.
154  * length - the length in bytes of the compressed page data.  Needed during
155  *          decompression. For a same value filled page length is 0.
156  * pool - the zswap_pool the entry's data is in
157  * handle - zpool allocation handle that stores the compressed page data
158  * value - value of the same-value filled pages which have same content
159  */
160 struct zswap_entry {
161 	struct rb_node rbnode;
162 	pgoff_t offset;
163 	int refcount;
164 	unsigned int length;
165 	struct zswap_pool *pool;
166 	union {
167 		unsigned long handle;
168 		unsigned long value;
169 	};
170 };
171 
172 struct zswap_header {
173 	swp_entry_t swpentry;
174 };
175 
176 /*
177  * The tree lock in the zswap_tree struct protects a few things:
178  * - the rbtree
179  * - the refcount field of each entry in the tree
180  */
181 struct zswap_tree {
182 	struct rb_root rbroot;
183 	spinlock_t lock;
184 };
185 
186 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
187 
188 /* RCU-protected iteration */
189 static LIST_HEAD(zswap_pools);
190 /* protects zswap_pools list modification */
191 static DEFINE_SPINLOCK(zswap_pools_lock);
192 /* pool counter to provide unique names to zpool */
193 static atomic_t zswap_pools_count = ATOMIC_INIT(0);
194 
195 /* used by param callback function */
196 static bool zswap_init_started;
197 
198 /* fatal error during init */
199 static bool zswap_init_failed;
200 
201 /* init completed, but couldn't create the initial pool */
202 static bool zswap_has_pool;
203 
204 /*********************************
205 * helpers and fwd declarations
206 **********************************/
207 
208 #define zswap_pool_debug(msg, p)				\
209 	pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name,		\
210 		 zpool_get_type((p)->zpool))
211 
212 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
213 static int zswap_pool_get(struct zswap_pool *pool);
214 static void zswap_pool_put(struct zswap_pool *pool);
215 
216 static const struct zpool_ops zswap_zpool_ops = {
217 	.evict = zswap_writeback_entry
218 };
219 
220 static bool zswap_is_full(void)
221 {
222 	return totalram_pages * zswap_max_pool_percent / 100 <
223 		DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
224 }
225 
226 static void zswap_update_total_size(void)
227 {
228 	struct zswap_pool *pool;
229 	u64 total = 0;
230 
231 	rcu_read_lock();
232 
233 	list_for_each_entry_rcu(pool, &zswap_pools, list)
234 		total += zpool_get_total_size(pool->zpool);
235 
236 	rcu_read_unlock();
237 
238 	zswap_pool_total_size = total;
239 }
240 
241 /*********************************
242 * zswap entry functions
243 **********************************/
244 static struct kmem_cache *zswap_entry_cache;
245 
246 static int __init zswap_entry_cache_create(void)
247 {
248 	zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
249 	return zswap_entry_cache == NULL;
250 }
251 
252 static void __init zswap_entry_cache_destroy(void)
253 {
254 	kmem_cache_destroy(zswap_entry_cache);
255 }
256 
257 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
258 {
259 	struct zswap_entry *entry;
260 	entry = kmem_cache_alloc(zswap_entry_cache, gfp);
261 	if (!entry)
262 		return NULL;
263 	entry->refcount = 1;
264 	RB_CLEAR_NODE(&entry->rbnode);
265 	return entry;
266 }
267 
268 static void zswap_entry_cache_free(struct zswap_entry *entry)
269 {
270 	kmem_cache_free(zswap_entry_cache, entry);
271 }
272 
273 /*********************************
274 * rbtree functions
275 **********************************/
276 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
277 {
278 	struct rb_node *node = root->rb_node;
279 	struct zswap_entry *entry;
280 
281 	while (node) {
282 		entry = rb_entry(node, struct zswap_entry, rbnode);
283 		if (entry->offset > offset)
284 			node = node->rb_left;
285 		else if (entry->offset < offset)
286 			node = node->rb_right;
287 		else
288 			return entry;
289 	}
290 	return NULL;
291 }
292 
293 /*
294  * In the case that a entry with the same offset is found, a pointer to
295  * the existing entry is stored in dupentry and the function returns -EEXIST
296  */
297 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
298 			struct zswap_entry **dupentry)
299 {
300 	struct rb_node **link = &root->rb_node, *parent = NULL;
301 	struct zswap_entry *myentry;
302 
303 	while (*link) {
304 		parent = *link;
305 		myentry = rb_entry(parent, struct zswap_entry, rbnode);
306 		if (myentry->offset > entry->offset)
307 			link = &(*link)->rb_left;
308 		else if (myentry->offset < entry->offset)
309 			link = &(*link)->rb_right;
310 		else {
311 			*dupentry = myentry;
312 			return -EEXIST;
313 		}
314 	}
315 	rb_link_node(&entry->rbnode, parent, link);
316 	rb_insert_color(&entry->rbnode, root);
317 	return 0;
318 }
319 
320 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
321 {
322 	if (!RB_EMPTY_NODE(&entry->rbnode)) {
323 		rb_erase(&entry->rbnode, root);
324 		RB_CLEAR_NODE(&entry->rbnode);
325 	}
326 }
327 
328 /*
329  * Carries out the common pattern of freeing and entry's zpool allocation,
330  * freeing the entry itself, and decrementing the number of stored pages.
331  */
332 static void zswap_free_entry(struct zswap_entry *entry)
333 {
334 	if (!entry->length)
335 		atomic_dec(&zswap_same_filled_pages);
336 	else {
337 		zpool_free(entry->pool->zpool, entry->handle);
338 		zswap_pool_put(entry->pool);
339 	}
340 	zswap_entry_cache_free(entry);
341 	atomic_dec(&zswap_stored_pages);
342 	zswap_update_total_size();
343 }
344 
345 /* caller must hold the tree lock */
346 static void zswap_entry_get(struct zswap_entry *entry)
347 {
348 	entry->refcount++;
349 }
350 
351 /* caller must hold the tree lock
352 * remove from the tree and free it, if nobody reference the entry
353 */
354 static void zswap_entry_put(struct zswap_tree *tree,
355 			struct zswap_entry *entry)
356 {
357 	int refcount = --entry->refcount;
358 
359 	BUG_ON(refcount < 0);
360 	if (refcount == 0) {
361 		zswap_rb_erase(&tree->rbroot, entry);
362 		zswap_free_entry(entry);
363 	}
364 }
365 
366 /* caller must hold the tree lock */
367 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
368 				pgoff_t offset)
369 {
370 	struct zswap_entry *entry;
371 
372 	entry = zswap_rb_search(root, offset);
373 	if (entry)
374 		zswap_entry_get(entry);
375 
376 	return entry;
377 }
378 
379 /*********************************
380 * per-cpu code
381 **********************************/
382 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
383 
384 static int zswap_dstmem_prepare(unsigned int cpu)
385 {
386 	u8 *dst;
387 
388 	dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
389 	if (!dst)
390 		return -ENOMEM;
391 
392 	per_cpu(zswap_dstmem, cpu) = dst;
393 	return 0;
394 }
395 
396 static int zswap_dstmem_dead(unsigned int cpu)
397 {
398 	u8 *dst;
399 
400 	dst = per_cpu(zswap_dstmem, cpu);
401 	kfree(dst);
402 	per_cpu(zswap_dstmem, cpu) = NULL;
403 
404 	return 0;
405 }
406 
407 static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
408 {
409 	struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
410 	struct crypto_comp *tfm;
411 
412 	if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu)))
413 		return 0;
414 
415 	tfm = crypto_alloc_comp(pool->tfm_name, 0, 0);
416 	if (IS_ERR_OR_NULL(tfm)) {
417 		pr_err("could not alloc crypto comp %s : %ld\n",
418 		       pool->tfm_name, PTR_ERR(tfm));
419 		return -ENOMEM;
420 	}
421 	*per_cpu_ptr(pool->tfm, cpu) = tfm;
422 	return 0;
423 }
424 
425 static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
426 {
427 	struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
428 	struct crypto_comp *tfm;
429 
430 	tfm = *per_cpu_ptr(pool->tfm, cpu);
431 	if (!IS_ERR_OR_NULL(tfm))
432 		crypto_free_comp(tfm);
433 	*per_cpu_ptr(pool->tfm, cpu) = NULL;
434 	return 0;
435 }
436 
437 /*********************************
438 * pool functions
439 **********************************/
440 
441 static struct zswap_pool *__zswap_pool_current(void)
442 {
443 	struct zswap_pool *pool;
444 
445 	pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
446 	WARN_ONCE(!pool && zswap_has_pool,
447 		  "%s: no page storage pool!\n", __func__);
448 
449 	return pool;
450 }
451 
452 static struct zswap_pool *zswap_pool_current(void)
453 {
454 	assert_spin_locked(&zswap_pools_lock);
455 
456 	return __zswap_pool_current();
457 }
458 
459 static struct zswap_pool *zswap_pool_current_get(void)
460 {
461 	struct zswap_pool *pool;
462 
463 	rcu_read_lock();
464 
465 	pool = __zswap_pool_current();
466 	if (!zswap_pool_get(pool))
467 		pool = NULL;
468 
469 	rcu_read_unlock();
470 
471 	return pool;
472 }
473 
474 static struct zswap_pool *zswap_pool_last_get(void)
475 {
476 	struct zswap_pool *pool, *last = NULL;
477 
478 	rcu_read_lock();
479 
480 	list_for_each_entry_rcu(pool, &zswap_pools, list)
481 		last = pool;
482 	WARN_ONCE(!last && zswap_has_pool,
483 		  "%s: no page storage pool!\n", __func__);
484 	if (!zswap_pool_get(last))
485 		last = NULL;
486 
487 	rcu_read_unlock();
488 
489 	return last;
490 }
491 
492 /* type and compressor must be null-terminated */
493 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
494 {
495 	struct zswap_pool *pool;
496 
497 	assert_spin_locked(&zswap_pools_lock);
498 
499 	list_for_each_entry_rcu(pool, &zswap_pools, list) {
500 		if (strcmp(pool->tfm_name, compressor))
501 			continue;
502 		if (strcmp(zpool_get_type(pool->zpool), type))
503 			continue;
504 		/* if we can't get it, it's about to be destroyed */
505 		if (!zswap_pool_get(pool))
506 			continue;
507 		return pool;
508 	}
509 
510 	return NULL;
511 }
512 
513 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
514 {
515 	struct zswap_pool *pool;
516 	char name[38]; /* 'zswap' + 32 char (max) num + \0 */
517 	gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
518 	int ret;
519 
520 	if (!zswap_has_pool) {
521 		/* if either are unset, pool initialization failed, and we
522 		 * need both params to be set correctly before trying to
523 		 * create a pool.
524 		 */
525 		if (!strcmp(type, ZSWAP_PARAM_UNSET))
526 			return NULL;
527 		if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
528 			return NULL;
529 	}
530 
531 	pool = kzalloc(sizeof(*pool), GFP_KERNEL);
532 	if (!pool)
533 		return NULL;
534 
535 	/* unique name for each pool specifically required by zsmalloc */
536 	snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
537 
538 	pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
539 	if (!pool->zpool) {
540 		pr_err("%s zpool not available\n", type);
541 		goto error;
542 	}
543 	pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
544 
545 	strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
546 	pool->tfm = alloc_percpu(struct crypto_comp *);
547 	if (!pool->tfm) {
548 		pr_err("percpu alloc failed\n");
549 		goto error;
550 	}
551 
552 	ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
553 				       &pool->node);
554 	if (ret)
555 		goto error;
556 	pr_debug("using %s compressor\n", pool->tfm_name);
557 
558 	/* being the current pool takes 1 ref; this func expects the
559 	 * caller to always add the new pool as the current pool
560 	 */
561 	kref_init(&pool->kref);
562 	INIT_LIST_HEAD(&pool->list);
563 
564 	zswap_pool_debug("created", pool);
565 
566 	return pool;
567 
568 error:
569 	free_percpu(pool->tfm);
570 	if (pool->zpool)
571 		zpool_destroy_pool(pool->zpool);
572 	kfree(pool);
573 	return NULL;
574 }
575 
576 static __init struct zswap_pool *__zswap_pool_create_fallback(void)
577 {
578 	bool has_comp, has_zpool;
579 
580 	has_comp = crypto_has_comp(zswap_compressor, 0, 0);
581 	if (!has_comp && strcmp(zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT)) {
582 		pr_err("compressor %s not available, using default %s\n",
583 		       zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT);
584 		param_free_charp(&zswap_compressor);
585 		zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
586 		has_comp = crypto_has_comp(zswap_compressor, 0, 0);
587 	}
588 	if (!has_comp) {
589 		pr_err("default compressor %s not available\n",
590 		       zswap_compressor);
591 		param_free_charp(&zswap_compressor);
592 		zswap_compressor = ZSWAP_PARAM_UNSET;
593 	}
594 
595 	has_zpool = zpool_has_pool(zswap_zpool_type);
596 	if (!has_zpool && strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) {
597 		pr_err("zpool %s not available, using default %s\n",
598 		       zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT);
599 		param_free_charp(&zswap_zpool_type);
600 		zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
601 		has_zpool = zpool_has_pool(zswap_zpool_type);
602 	}
603 	if (!has_zpool) {
604 		pr_err("default zpool %s not available\n",
605 		       zswap_zpool_type);
606 		param_free_charp(&zswap_zpool_type);
607 		zswap_zpool_type = ZSWAP_PARAM_UNSET;
608 	}
609 
610 	if (!has_comp || !has_zpool)
611 		return NULL;
612 
613 	return zswap_pool_create(zswap_zpool_type, zswap_compressor);
614 }
615 
616 static void zswap_pool_destroy(struct zswap_pool *pool)
617 {
618 	zswap_pool_debug("destroying", pool);
619 
620 	cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
621 	free_percpu(pool->tfm);
622 	zpool_destroy_pool(pool->zpool);
623 	kfree(pool);
624 }
625 
626 static int __must_check zswap_pool_get(struct zswap_pool *pool)
627 {
628 	if (!pool)
629 		return 0;
630 
631 	return kref_get_unless_zero(&pool->kref);
632 }
633 
634 static void __zswap_pool_release(struct work_struct *work)
635 {
636 	struct zswap_pool *pool = container_of(work, typeof(*pool), work);
637 
638 	synchronize_rcu();
639 
640 	/* nobody should have been able to get a kref... */
641 	WARN_ON(kref_get_unless_zero(&pool->kref));
642 
643 	/* pool is now off zswap_pools list and has no references. */
644 	zswap_pool_destroy(pool);
645 }
646 
647 static void __zswap_pool_empty(struct kref *kref)
648 {
649 	struct zswap_pool *pool;
650 
651 	pool = container_of(kref, typeof(*pool), kref);
652 
653 	spin_lock(&zswap_pools_lock);
654 
655 	WARN_ON(pool == zswap_pool_current());
656 
657 	list_del_rcu(&pool->list);
658 
659 	INIT_WORK(&pool->work, __zswap_pool_release);
660 	schedule_work(&pool->work);
661 
662 	spin_unlock(&zswap_pools_lock);
663 }
664 
665 static void zswap_pool_put(struct zswap_pool *pool)
666 {
667 	kref_put(&pool->kref, __zswap_pool_empty);
668 }
669 
670 /*********************************
671 * param callbacks
672 **********************************/
673 
674 /* val must be a null-terminated string */
675 static int __zswap_param_set(const char *val, const struct kernel_param *kp,
676 			     char *type, char *compressor)
677 {
678 	struct zswap_pool *pool, *put_pool = NULL;
679 	char *s = strstrip((char *)val);
680 	int ret;
681 
682 	if (zswap_init_failed) {
683 		pr_err("can't set param, initialization failed\n");
684 		return -ENODEV;
685 	}
686 
687 	/* no change required */
688 	if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
689 		return 0;
690 
691 	/* if this is load-time (pre-init) param setting,
692 	 * don't create a pool; that's done during init.
693 	 */
694 	if (!zswap_init_started)
695 		return param_set_charp(s, kp);
696 
697 	if (!type) {
698 		if (!zpool_has_pool(s)) {
699 			pr_err("zpool %s not available\n", s);
700 			return -ENOENT;
701 		}
702 		type = s;
703 	} else if (!compressor) {
704 		if (!crypto_has_comp(s, 0, 0)) {
705 			pr_err("compressor %s not available\n", s);
706 			return -ENOENT;
707 		}
708 		compressor = s;
709 	} else {
710 		WARN_ON(1);
711 		return -EINVAL;
712 	}
713 
714 	spin_lock(&zswap_pools_lock);
715 
716 	pool = zswap_pool_find_get(type, compressor);
717 	if (pool) {
718 		zswap_pool_debug("using existing", pool);
719 		WARN_ON(pool == zswap_pool_current());
720 		list_del_rcu(&pool->list);
721 	}
722 
723 	spin_unlock(&zswap_pools_lock);
724 
725 	if (!pool)
726 		pool = zswap_pool_create(type, compressor);
727 
728 	if (pool)
729 		ret = param_set_charp(s, kp);
730 	else
731 		ret = -EINVAL;
732 
733 	spin_lock(&zswap_pools_lock);
734 
735 	if (!ret) {
736 		put_pool = zswap_pool_current();
737 		list_add_rcu(&pool->list, &zswap_pools);
738 		zswap_has_pool = true;
739 	} else if (pool) {
740 		/* add the possibly pre-existing pool to the end of the pools
741 		 * list; if it's new (and empty) then it'll be removed and
742 		 * destroyed by the put after we drop the lock
743 		 */
744 		list_add_tail_rcu(&pool->list, &zswap_pools);
745 		put_pool = pool;
746 	}
747 
748 	spin_unlock(&zswap_pools_lock);
749 
750 	if (!zswap_has_pool && !pool) {
751 		/* if initial pool creation failed, and this pool creation also
752 		 * failed, maybe both compressor and zpool params were bad.
753 		 * Allow changing this param, so pool creation will succeed
754 		 * when the other param is changed. We already verified this
755 		 * param is ok in the zpool_has_pool() or crypto_has_comp()
756 		 * checks above.
757 		 */
758 		ret = param_set_charp(s, kp);
759 	}
760 
761 	/* drop the ref from either the old current pool,
762 	 * or the new pool we failed to add
763 	 */
764 	if (put_pool)
765 		zswap_pool_put(put_pool);
766 
767 	return ret;
768 }
769 
770 static int zswap_compressor_param_set(const char *val,
771 				      const struct kernel_param *kp)
772 {
773 	return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
774 }
775 
776 static int zswap_zpool_param_set(const char *val,
777 				 const struct kernel_param *kp)
778 {
779 	return __zswap_param_set(val, kp, NULL, zswap_compressor);
780 }
781 
782 static int zswap_enabled_param_set(const char *val,
783 				   const struct kernel_param *kp)
784 {
785 	if (zswap_init_failed) {
786 		pr_err("can't enable, initialization failed\n");
787 		return -ENODEV;
788 	}
789 	if (!zswap_has_pool && zswap_init_started) {
790 		pr_err("can't enable, no pool configured\n");
791 		return -ENODEV;
792 	}
793 
794 	return param_set_bool(val, kp);
795 }
796 
797 /*********************************
798 * writeback code
799 **********************************/
800 /* return enum for zswap_get_swap_cache_page */
801 enum zswap_get_swap_ret {
802 	ZSWAP_SWAPCACHE_NEW,
803 	ZSWAP_SWAPCACHE_EXIST,
804 	ZSWAP_SWAPCACHE_FAIL,
805 };
806 
807 /*
808  * zswap_get_swap_cache_page
809  *
810  * This is an adaption of read_swap_cache_async()
811  *
812  * This function tries to find a page with the given swap entry
813  * in the swapper_space address space (the swap cache).  If the page
814  * is found, it is returned in retpage.  Otherwise, a page is allocated,
815  * added to the swap cache, and returned in retpage.
816  *
817  * If success, the swap cache page is returned in retpage
818  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
819  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
820  *     the new page is added to swapcache and locked
821  * Returns ZSWAP_SWAPCACHE_FAIL on error
822  */
823 static int zswap_get_swap_cache_page(swp_entry_t entry,
824 				struct page **retpage)
825 {
826 	bool page_was_allocated;
827 
828 	*retpage = __read_swap_cache_async(entry, GFP_KERNEL,
829 			NULL, 0, &page_was_allocated);
830 	if (page_was_allocated)
831 		return ZSWAP_SWAPCACHE_NEW;
832 	if (!*retpage)
833 		return ZSWAP_SWAPCACHE_FAIL;
834 	return ZSWAP_SWAPCACHE_EXIST;
835 }
836 
837 /*
838  * Attempts to free an entry by adding a page to the swap cache,
839  * decompressing the entry data into the page, and issuing a
840  * bio write to write the page back to the swap device.
841  *
842  * This can be thought of as a "resumed writeback" of the page
843  * to the swap device.  We are basically resuming the same swap
844  * writeback path that was intercepted with the frontswap_store()
845  * in the first place.  After the page has been decompressed into
846  * the swap cache, the compressed version stored by zswap can be
847  * freed.
848  */
849 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
850 {
851 	struct zswap_header *zhdr;
852 	swp_entry_t swpentry;
853 	struct zswap_tree *tree;
854 	pgoff_t offset;
855 	struct zswap_entry *entry;
856 	struct page *page;
857 	struct crypto_comp *tfm;
858 	u8 *src, *dst;
859 	unsigned int dlen;
860 	int ret;
861 	struct writeback_control wbc = {
862 		.sync_mode = WB_SYNC_NONE,
863 	};
864 
865 	/* extract swpentry from data */
866 	zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
867 	swpentry = zhdr->swpentry; /* here */
868 	zpool_unmap_handle(pool, handle);
869 	tree = zswap_trees[swp_type(swpentry)];
870 	offset = swp_offset(swpentry);
871 
872 	/* find and ref zswap entry */
873 	spin_lock(&tree->lock);
874 	entry = zswap_entry_find_get(&tree->rbroot, offset);
875 	if (!entry) {
876 		/* entry was invalidated */
877 		spin_unlock(&tree->lock);
878 		return 0;
879 	}
880 	spin_unlock(&tree->lock);
881 	BUG_ON(offset != entry->offset);
882 
883 	/* try to allocate swap cache page */
884 	switch (zswap_get_swap_cache_page(swpentry, &page)) {
885 	case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
886 		ret = -ENOMEM;
887 		goto fail;
888 
889 	case ZSWAP_SWAPCACHE_EXIST:
890 		/* page is already in the swap cache, ignore for now */
891 		put_page(page);
892 		ret = -EEXIST;
893 		goto fail;
894 
895 	case ZSWAP_SWAPCACHE_NEW: /* page is locked */
896 		/* decompress */
897 		dlen = PAGE_SIZE;
898 		src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle,
899 				ZPOOL_MM_RO) + sizeof(struct zswap_header);
900 		dst = kmap_atomic(page);
901 		tfm = *get_cpu_ptr(entry->pool->tfm);
902 		ret = crypto_comp_decompress(tfm, src, entry->length,
903 					     dst, &dlen);
904 		put_cpu_ptr(entry->pool->tfm);
905 		kunmap_atomic(dst);
906 		zpool_unmap_handle(entry->pool->zpool, entry->handle);
907 		BUG_ON(ret);
908 		BUG_ON(dlen != PAGE_SIZE);
909 
910 		/* page is up to date */
911 		SetPageUptodate(page);
912 	}
913 
914 	/* move it to the tail of the inactive list after end_writeback */
915 	SetPageReclaim(page);
916 
917 	/* start writeback */
918 	__swap_writepage(page, &wbc, end_swap_bio_write);
919 	put_page(page);
920 	zswap_written_back_pages++;
921 
922 	spin_lock(&tree->lock);
923 	/* drop local reference */
924 	zswap_entry_put(tree, entry);
925 
926 	/*
927 	* There are two possible situations for entry here:
928 	* (1) refcount is 1(normal case),  entry is valid and on the tree
929 	* (2) refcount is 0, entry is freed and not on the tree
930 	*     because invalidate happened during writeback
931 	*  search the tree and free the entry if find entry
932 	*/
933 	if (entry == zswap_rb_search(&tree->rbroot, offset))
934 		zswap_entry_put(tree, entry);
935 	spin_unlock(&tree->lock);
936 
937 	goto end;
938 
939 	/*
940 	* if we get here due to ZSWAP_SWAPCACHE_EXIST
941 	* a load may happening concurrently
942 	* it is safe and okay to not free the entry
943 	* if we free the entry in the following put
944 	* it it either okay to return !0
945 	*/
946 fail:
947 	spin_lock(&tree->lock);
948 	zswap_entry_put(tree, entry);
949 	spin_unlock(&tree->lock);
950 
951 end:
952 	return ret;
953 }
954 
955 static int zswap_shrink(void)
956 {
957 	struct zswap_pool *pool;
958 	int ret;
959 
960 	pool = zswap_pool_last_get();
961 	if (!pool)
962 		return -ENOENT;
963 
964 	ret = zpool_shrink(pool->zpool, 1, NULL);
965 
966 	zswap_pool_put(pool);
967 
968 	return ret;
969 }
970 
971 static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
972 {
973 	unsigned int pos;
974 	unsigned long *page;
975 
976 	page = (unsigned long *)ptr;
977 	for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
978 		if (page[pos] != page[0])
979 			return 0;
980 	}
981 	*value = page[0];
982 	return 1;
983 }
984 
985 static void zswap_fill_page(void *ptr, unsigned long value)
986 {
987 	unsigned long *page;
988 
989 	page = (unsigned long *)ptr;
990 	memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
991 }
992 
993 /*********************************
994 * frontswap hooks
995 **********************************/
996 /* attempts to compress and store an single page */
997 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
998 				struct page *page)
999 {
1000 	struct zswap_tree *tree = zswap_trees[type];
1001 	struct zswap_entry *entry, *dupentry;
1002 	struct crypto_comp *tfm;
1003 	int ret;
1004 	unsigned int hlen, dlen = PAGE_SIZE;
1005 	unsigned long handle, value;
1006 	char *buf;
1007 	u8 *src, *dst;
1008 	struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1009 
1010 	/* THP isn't supported */
1011 	if (PageTransHuge(page)) {
1012 		ret = -EINVAL;
1013 		goto reject;
1014 	}
1015 
1016 	if (!zswap_enabled || !tree) {
1017 		ret = -ENODEV;
1018 		goto reject;
1019 	}
1020 
1021 	/* reclaim space if needed */
1022 	if (zswap_is_full()) {
1023 		zswap_pool_limit_hit++;
1024 		if (zswap_shrink()) {
1025 			zswap_reject_reclaim_fail++;
1026 			ret = -ENOMEM;
1027 			goto reject;
1028 		}
1029 	}
1030 
1031 	/* allocate entry */
1032 	entry = zswap_entry_cache_alloc(GFP_KERNEL);
1033 	if (!entry) {
1034 		zswap_reject_kmemcache_fail++;
1035 		ret = -ENOMEM;
1036 		goto reject;
1037 	}
1038 
1039 	if (zswap_same_filled_pages_enabled) {
1040 		src = kmap_atomic(page);
1041 		if (zswap_is_page_same_filled(src, &value)) {
1042 			kunmap_atomic(src);
1043 			entry->offset = offset;
1044 			entry->length = 0;
1045 			entry->value = value;
1046 			atomic_inc(&zswap_same_filled_pages);
1047 			goto insert_entry;
1048 		}
1049 		kunmap_atomic(src);
1050 	}
1051 
1052 	/* if entry is successfully added, it keeps the reference */
1053 	entry->pool = zswap_pool_current_get();
1054 	if (!entry->pool) {
1055 		ret = -EINVAL;
1056 		goto freepage;
1057 	}
1058 
1059 	/* compress */
1060 	dst = get_cpu_var(zswap_dstmem);
1061 	tfm = *get_cpu_ptr(entry->pool->tfm);
1062 	src = kmap_atomic(page);
1063 	ret = crypto_comp_compress(tfm, src, PAGE_SIZE, dst, &dlen);
1064 	kunmap_atomic(src);
1065 	put_cpu_ptr(entry->pool->tfm);
1066 	if (ret) {
1067 		ret = -EINVAL;
1068 		goto put_dstmem;
1069 	}
1070 
1071 	/* store */
1072 	hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1073 	ret = zpool_malloc(entry->pool->zpool, hlen + dlen,
1074 			   __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM,
1075 			   &handle);
1076 	if (ret == -ENOSPC) {
1077 		zswap_reject_compress_poor++;
1078 		goto put_dstmem;
1079 	}
1080 	if (ret) {
1081 		zswap_reject_alloc_fail++;
1082 		goto put_dstmem;
1083 	}
1084 	buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW);
1085 	memcpy(buf, &zhdr, hlen);
1086 	memcpy(buf + hlen, dst, dlen);
1087 	zpool_unmap_handle(entry->pool->zpool, handle);
1088 	put_cpu_var(zswap_dstmem);
1089 
1090 	/* populate entry */
1091 	entry->offset = offset;
1092 	entry->handle = handle;
1093 	entry->length = dlen;
1094 
1095 insert_entry:
1096 	/* map */
1097 	spin_lock(&tree->lock);
1098 	do {
1099 		ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1100 		if (ret == -EEXIST) {
1101 			zswap_duplicate_entry++;
1102 			/* remove from rbtree */
1103 			zswap_rb_erase(&tree->rbroot, dupentry);
1104 			zswap_entry_put(tree, dupentry);
1105 		}
1106 	} while (ret == -EEXIST);
1107 	spin_unlock(&tree->lock);
1108 
1109 	/* update stats */
1110 	atomic_inc(&zswap_stored_pages);
1111 	zswap_update_total_size();
1112 
1113 	return 0;
1114 
1115 put_dstmem:
1116 	put_cpu_var(zswap_dstmem);
1117 	zswap_pool_put(entry->pool);
1118 freepage:
1119 	zswap_entry_cache_free(entry);
1120 reject:
1121 	return ret;
1122 }
1123 
1124 /*
1125  * returns 0 if the page was successfully decompressed
1126  * return -1 on entry not found or error
1127 */
1128 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1129 				struct page *page)
1130 {
1131 	struct zswap_tree *tree = zswap_trees[type];
1132 	struct zswap_entry *entry;
1133 	struct crypto_comp *tfm;
1134 	u8 *src, *dst;
1135 	unsigned int dlen;
1136 	int ret;
1137 
1138 	/* find */
1139 	spin_lock(&tree->lock);
1140 	entry = zswap_entry_find_get(&tree->rbroot, offset);
1141 	if (!entry) {
1142 		/* entry was written back */
1143 		spin_unlock(&tree->lock);
1144 		return -1;
1145 	}
1146 	spin_unlock(&tree->lock);
1147 
1148 	if (!entry->length) {
1149 		dst = kmap_atomic(page);
1150 		zswap_fill_page(dst, entry->value);
1151 		kunmap_atomic(dst);
1152 		goto freeentry;
1153 	}
1154 
1155 	/* decompress */
1156 	dlen = PAGE_SIZE;
1157 	src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1158 	if (zpool_evictable(entry->pool->zpool))
1159 		src += sizeof(struct zswap_header);
1160 	dst = kmap_atomic(page);
1161 	tfm = *get_cpu_ptr(entry->pool->tfm);
1162 	ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen);
1163 	put_cpu_ptr(entry->pool->tfm);
1164 	kunmap_atomic(dst);
1165 	zpool_unmap_handle(entry->pool->zpool, entry->handle);
1166 	BUG_ON(ret);
1167 
1168 freeentry:
1169 	spin_lock(&tree->lock);
1170 	zswap_entry_put(tree, entry);
1171 	spin_unlock(&tree->lock);
1172 
1173 	return 0;
1174 }
1175 
1176 /* frees an entry in zswap */
1177 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1178 {
1179 	struct zswap_tree *tree = zswap_trees[type];
1180 	struct zswap_entry *entry;
1181 
1182 	/* find */
1183 	spin_lock(&tree->lock);
1184 	entry = zswap_rb_search(&tree->rbroot, offset);
1185 	if (!entry) {
1186 		/* entry was written back */
1187 		spin_unlock(&tree->lock);
1188 		return;
1189 	}
1190 
1191 	/* remove from rbtree */
1192 	zswap_rb_erase(&tree->rbroot, entry);
1193 
1194 	/* drop the initial reference from entry creation */
1195 	zswap_entry_put(tree, entry);
1196 
1197 	spin_unlock(&tree->lock);
1198 }
1199 
1200 /* frees all zswap entries for the given swap type */
1201 static void zswap_frontswap_invalidate_area(unsigned type)
1202 {
1203 	struct zswap_tree *tree = zswap_trees[type];
1204 	struct zswap_entry *entry, *n;
1205 
1206 	if (!tree)
1207 		return;
1208 
1209 	/* walk the tree and free everything */
1210 	spin_lock(&tree->lock);
1211 	rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1212 		zswap_free_entry(entry);
1213 	tree->rbroot = RB_ROOT;
1214 	spin_unlock(&tree->lock);
1215 	kfree(tree);
1216 	zswap_trees[type] = NULL;
1217 }
1218 
1219 static void zswap_frontswap_init(unsigned type)
1220 {
1221 	struct zswap_tree *tree;
1222 
1223 	tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1224 	if (!tree) {
1225 		pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1226 		return;
1227 	}
1228 
1229 	tree->rbroot = RB_ROOT;
1230 	spin_lock_init(&tree->lock);
1231 	zswap_trees[type] = tree;
1232 }
1233 
1234 static struct frontswap_ops zswap_frontswap_ops = {
1235 	.store = zswap_frontswap_store,
1236 	.load = zswap_frontswap_load,
1237 	.invalidate_page = zswap_frontswap_invalidate_page,
1238 	.invalidate_area = zswap_frontswap_invalidate_area,
1239 	.init = zswap_frontswap_init
1240 };
1241 
1242 /*********************************
1243 * debugfs functions
1244 **********************************/
1245 #ifdef CONFIG_DEBUG_FS
1246 #include <linux/debugfs.h>
1247 
1248 static struct dentry *zswap_debugfs_root;
1249 
1250 static int __init zswap_debugfs_init(void)
1251 {
1252 	if (!debugfs_initialized())
1253 		return -ENODEV;
1254 
1255 	zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1256 	if (!zswap_debugfs_root)
1257 		return -ENOMEM;
1258 
1259 	debugfs_create_u64("pool_limit_hit", S_IRUGO,
1260 			zswap_debugfs_root, &zswap_pool_limit_hit);
1261 	debugfs_create_u64("reject_reclaim_fail", S_IRUGO,
1262 			zswap_debugfs_root, &zswap_reject_reclaim_fail);
1263 	debugfs_create_u64("reject_alloc_fail", S_IRUGO,
1264 			zswap_debugfs_root, &zswap_reject_alloc_fail);
1265 	debugfs_create_u64("reject_kmemcache_fail", S_IRUGO,
1266 			zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1267 	debugfs_create_u64("reject_compress_poor", S_IRUGO,
1268 			zswap_debugfs_root, &zswap_reject_compress_poor);
1269 	debugfs_create_u64("written_back_pages", S_IRUGO,
1270 			zswap_debugfs_root, &zswap_written_back_pages);
1271 	debugfs_create_u64("duplicate_entry", S_IRUGO,
1272 			zswap_debugfs_root, &zswap_duplicate_entry);
1273 	debugfs_create_u64("pool_total_size", S_IRUGO,
1274 			zswap_debugfs_root, &zswap_pool_total_size);
1275 	debugfs_create_atomic_t("stored_pages", S_IRUGO,
1276 			zswap_debugfs_root, &zswap_stored_pages);
1277 	debugfs_create_atomic_t("same_filled_pages", 0444,
1278 			zswap_debugfs_root, &zswap_same_filled_pages);
1279 
1280 	return 0;
1281 }
1282 
1283 static void __exit zswap_debugfs_exit(void)
1284 {
1285 	debugfs_remove_recursive(zswap_debugfs_root);
1286 }
1287 #else
1288 static int __init zswap_debugfs_init(void)
1289 {
1290 	return 0;
1291 }
1292 
1293 static void __exit zswap_debugfs_exit(void) { }
1294 #endif
1295 
1296 /*********************************
1297 * module init and exit
1298 **********************************/
1299 static int __init init_zswap(void)
1300 {
1301 	struct zswap_pool *pool;
1302 	int ret;
1303 
1304 	zswap_init_started = true;
1305 
1306 	if (zswap_entry_cache_create()) {
1307 		pr_err("entry cache creation failed\n");
1308 		goto cache_fail;
1309 	}
1310 
1311 	ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1312 				zswap_dstmem_prepare, zswap_dstmem_dead);
1313 	if (ret) {
1314 		pr_err("dstmem alloc failed\n");
1315 		goto dstmem_fail;
1316 	}
1317 
1318 	ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1319 				      "mm/zswap_pool:prepare",
1320 				      zswap_cpu_comp_prepare,
1321 				      zswap_cpu_comp_dead);
1322 	if (ret)
1323 		goto hp_fail;
1324 
1325 	pool = __zswap_pool_create_fallback();
1326 	if (pool) {
1327 		pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1328 			zpool_get_type(pool->zpool));
1329 		list_add(&pool->list, &zswap_pools);
1330 		zswap_has_pool = true;
1331 	} else {
1332 		pr_err("pool creation failed\n");
1333 		zswap_enabled = false;
1334 	}
1335 
1336 	frontswap_register_ops(&zswap_frontswap_ops);
1337 	if (zswap_debugfs_init())
1338 		pr_warn("debugfs initialization failed\n");
1339 	return 0;
1340 
1341 hp_fail:
1342 	cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1343 dstmem_fail:
1344 	zswap_entry_cache_destroy();
1345 cache_fail:
1346 	/* if built-in, we aren't unloaded on failure; don't allow use */
1347 	zswap_init_failed = true;
1348 	zswap_enabled = false;
1349 	return -ENOMEM;
1350 }
1351 /* must be late so crypto has time to come up */
1352 late_initcall(init_zswap);
1353 
1354 MODULE_LICENSE("GPL");
1355 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1356 MODULE_DESCRIPTION("Compressed cache for swap pages");
1357