1
2
3
4
5
6
7
8
9
10
11
12
13
14#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15
16#include <linux/module.h>
17#include <linux/cpu.h>
18#include <linux/highmem.h>
19#include <linux/slab.h>
20#include <linux/spinlock.h>
21#include <linux/types.h>
22#include <linux/atomic.h>
23#include <linux/frontswap.h>
24#include <linux/rbtree.h>
25#include <linux/swap.h>
26#include <linux/crypto.h>
27#include <linux/scatterlist.h>
28#include <linux/mempool.h>
29#include <linux/zpool.h>
30#include <crypto/acompress.h>
31
32#include <linux/mm_types.h>
33#include <linux/page-flags.h>
34#include <linux/swapops.h>
35#include <linux/writeback.h>
36#include <linux/pagemap.h>
37#include <linux/workqueue.h>
38
39
40
41
42
43static u64 zswap_pool_total_size;
44
45static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
46
47static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
48
49
50
51
52
53
54
55
56
57static u64 zswap_pool_limit_hit;
58
59static u64 zswap_written_back_pages;
60
61static u64 zswap_reject_reclaim_fail;
62
63static u64 zswap_reject_compress_poor;
64
65static u64 zswap_reject_alloc_fail;
66
67static u64 zswap_reject_kmemcache_fail;
68
69static u64 zswap_duplicate_entry;
70
71
72static struct workqueue_struct *shrink_wq;
73
74static bool zswap_pool_reached_full;
75
76
77
78
79
80#define ZSWAP_PARAM_UNSET ""
81
82
83static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
84static int zswap_enabled_param_set(const char *,
85 const struct kernel_param *);
86static const struct kernel_param_ops zswap_enabled_param_ops = {
87 .set = zswap_enabled_param_set,
88 .get = param_get_bool,
89};
90module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
91
92
93static char *zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
94static int zswap_compressor_param_set(const char *,
95 const struct kernel_param *);
96static const struct kernel_param_ops zswap_compressor_param_ops = {
97 .set = zswap_compressor_param_set,
98 .get = param_get_charp,
99 .free = param_free_charp,
100};
101module_param_cb(compressor, &zswap_compressor_param_ops,
102 &zswap_compressor, 0644);
103
104
105static char *zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
106static int zswap_zpool_param_set(const char *, const struct kernel_param *);
107static const struct kernel_param_ops zswap_zpool_param_ops = {
108 .set = zswap_zpool_param_set,
109 .get = param_get_charp,
110 .free = param_free_charp,
111};
112module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
113
114
115static unsigned int zswap_max_pool_percent = 20;
116module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
117
118
119static unsigned int zswap_accept_thr_percent = 90;
120module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
121 uint, 0644);
122
123
124static bool zswap_same_filled_pages_enabled = true;
125module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
126 bool, 0644);
127
128
129
130
131
132struct crypto_acomp_ctx {
133 struct crypto_acomp *acomp;
134 struct acomp_req *req;
135 struct crypto_wait wait;
136 u8 *dstmem;
137 struct mutex *mutex;
138};
139
140struct zswap_pool {
141 struct zpool *zpool;
142 struct crypto_acomp_ctx __percpu *acomp_ctx;
143 struct kref kref;
144 struct list_head list;
145 struct work_struct release_work;
146 struct work_struct shrink_work;
147 struct hlist_node node;
148 char tfm_name[CRYPTO_MAX_ALG_NAME];
149};
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171struct zswap_entry {
172 struct rb_node rbnode;
173 pgoff_t offset;
174 int refcount;
175 unsigned int length;
176 struct zswap_pool *pool;
177 union {
178 unsigned long handle;
179 unsigned long value;
180 };
181};
182
183struct zswap_header {
184 swp_entry_t swpentry;
185};
186
187
188
189
190
191
192struct zswap_tree {
193 struct rb_root rbroot;
194 spinlock_t lock;
195};
196
197static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
198
199
200static LIST_HEAD(zswap_pools);
201
202static DEFINE_SPINLOCK(zswap_pools_lock);
203
204static atomic_t zswap_pools_count = ATOMIC_INIT(0);
205
206
207static bool zswap_init_started;
208
209
210static bool zswap_init_failed;
211
212
213static bool zswap_has_pool;
214
215
216
217
218
219#define zswap_pool_debug(msg, p) \
220 pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name, \
221 zpool_get_type((p)->zpool))
222
223static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
224static int zswap_pool_get(struct zswap_pool *pool);
225static void zswap_pool_put(struct zswap_pool *pool);
226
227static const struct zpool_ops zswap_zpool_ops = {
228 .evict = zswap_writeback_entry
229};
230
231static bool zswap_is_full(void)
232{
233 return totalram_pages() * zswap_max_pool_percent / 100 <
234 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
235}
236
237static bool zswap_can_accept(void)
238{
239 return totalram_pages() * zswap_accept_thr_percent / 100 *
240 zswap_max_pool_percent / 100 >
241 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
242}
243
244static void zswap_update_total_size(void)
245{
246 struct zswap_pool *pool;
247 u64 total = 0;
248
249 rcu_read_lock();
250
251 list_for_each_entry_rcu(pool, &zswap_pools, list)
252 total += zpool_get_total_size(pool->zpool);
253
254 rcu_read_unlock();
255
256 zswap_pool_total_size = total;
257}
258
259
260
261
262static struct kmem_cache *zswap_entry_cache;
263
264static int __init zswap_entry_cache_create(void)
265{
266 zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
267 return zswap_entry_cache == NULL;
268}
269
270static void __init zswap_entry_cache_destroy(void)
271{
272 kmem_cache_destroy(zswap_entry_cache);
273}
274
275static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
276{
277 struct zswap_entry *entry;
278 entry = kmem_cache_alloc(zswap_entry_cache, gfp);
279 if (!entry)
280 return NULL;
281 entry->refcount = 1;
282 RB_CLEAR_NODE(&entry->rbnode);
283 return entry;
284}
285
286static void zswap_entry_cache_free(struct zswap_entry *entry)
287{
288 kmem_cache_free(zswap_entry_cache, entry);
289}
290
291
292
293
294static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
295{
296 struct rb_node *node = root->rb_node;
297 struct zswap_entry *entry;
298
299 while (node) {
300 entry = rb_entry(node, struct zswap_entry, rbnode);
301 if (entry->offset > offset)
302 node = node->rb_left;
303 else if (entry->offset < offset)
304 node = node->rb_right;
305 else
306 return entry;
307 }
308 return NULL;
309}
310
311
312
313
314
315static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
316 struct zswap_entry **dupentry)
317{
318 struct rb_node **link = &root->rb_node, *parent = NULL;
319 struct zswap_entry *myentry;
320
321 while (*link) {
322 parent = *link;
323 myentry = rb_entry(parent, struct zswap_entry, rbnode);
324 if (myentry->offset > entry->offset)
325 link = &(*link)->rb_left;
326 else if (myentry->offset < entry->offset)
327 link = &(*link)->rb_right;
328 else {
329 *dupentry = myentry;
330 return -EEXIST;
331 }
332 }
333 rb_link_node(&entry->rbnode, parent, link);
334 rb_insert_color(&entry->rbnode, root);
335 return 0;
336}
337
338static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
339{
340 if (!RB_EMPTY_NODE(&entry->rbnode)) {
341 rb_erase(&entry->rbnode, root);
342 RB_CLEAR_NODE(&entry->rbnode);
343 }
344}
345
346
347
348
349
350static void zswap_free_entry(struct zswap_entry *entry)
351{
352 if (!entry->length)
353 atomic_dec(&zswap_same_filled_pages);
354 else {
355 zpool_free(entry->pool->zpool, entry->handle);
356 zswap_pool_put(entry->pool);
357 }
358 zswap_entry_cache_free(entry);
359 atomic_dec(&zswap_stored_pages);
360 zswap_update_total_size();
361}
362
363
364static void zswap_entry_get(struct zswap_entry *entry)
365{
366 entry->refcount++;
367}
368
369
370
371
372static void zswap_entry_put(struct zswap_tree *tree,
373 struct zswap_entry *entry)
374{
375 int refcount = --entry->refcount;
376
377 BUG_ON(refcount < 0);
378 if (refcount == 0) {
379 zswap_rb_erase(&tree->rbroot, entry);
380 zswap_free_entry(entry);
381 }
382}
383
384
385static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
386 pgoff_t offset)
387{
388 struct zswap_entry *entry;
389
390 entry = zswap_rb_search(root, offset);
391 if (entry)
392 zswap_entry_get(entry);
393
394 return entry;
395}
396
397
398
399
400static DEFINE_PER_CPU(u8 *, zswap_dstmem);
401
402
403
404
405
406static DEFINE_PER_CPU(struct mutex *, zswap_mutex);
407
408static int zswap_dstmem_prepare(unsigned int cpu)
409{
410 struct mutex *mutex;
411 u8 *dst;
412
413 dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
414 if (!dst)
415 return -ENOMEM;
416
417 mutex = kmalloc_node(sizeof(*mutex), GFP_KERNEL, cpu_to_node(cpu));
418 if (!mutex) {
419 kfree(dst);
420 return -ENOMEM;
421 }
422
423 mutex_init(mutex);
424 per_cpu(zswap_dstmem, cpu) = dst;
425 per_cpu(zswap_mutex, cpu) = mutex;
426 return 0;
427}
428
429static int zswap_dstmem_dead(unsigned int cpu)
430{
431 struct mutex *mutex;
432 u8 *dst;
433
434 mutex = per_cpu(zswap_mutex, cpu);
435 kfree(mutex);
436 per_cpu(zswap_mutex, cpu) = NULL;
437
438 dst = per_cpu(zswap_dstmem, cpu);
439 kfree(dst);
440 per_cpu(zswap_dstmem, cpu) = NULL;
441
442 return 0;
443}
444
445static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
446{
447 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
448 struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
449 struct crypto_acomp *acomp;
450 struct acomp_req *req;
451
452 acomp = crypto_alloc_acomp_node(pool->tfm_name, 0, 0, cpu_to_node(cpu));
453 if (IS_ERR(acomp)) {
454 pr_err("could not alloc crypto acomp %s : %ld\n",
455 pool->tfm_name, PTR_ERR(acomp));
456 return PTR_ERR(acomp);
457 }
458 acomp_ctx->acomp = acomp;
459
460 req = acomp_request_alloc(acomp_ctx->acomp);
461 if (!req) {
462 pr_err("could not alloc crypto acomp_request %s\n",
463 pool->tfm_name);
464 crypto_free_acomp(acomp_ctx->acomp);
465 return -ENOMEM;
466 }
467 acomp_ctx->req = req;
468
469 crypto_init_wait(&acomp_ctx->wait);
470
471
472
473
474
475 acomp_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
476 crypto_req_done, &acomp_ctx->wait);
477
478 acomp_ctx->mutex = per_cpu(zswap_mutex, cpu);
479 acomp_ctx->dstmem = per_cpu(zswap_dstmem, cpu);
480
481 return 0;
482}
483
484static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
485{
486 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
487 struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
488
489 if (!IS_ERR_OR_NULL(acomp_ctx)) {
490 if (!IS_ERR_OR_NULL(acomp_ctx->req))
491 acomp_request_free(acomp_ctx->req);
492 if (!IS_ERR_OR_NULL(acomp_ctx->acomp))
493 crypto_free_acomp(acomp_ctx->acomp);
494 }
495
496 return 0;
497}
498
499
500
501
502
503static struct zswap_pool *__zswap_pool_current(void)
504{
505 struct zswap_pool *pool;
506
507 pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
508 WARN_ONCE(!pool && zswap_has_pool,
509 "%s: no page storage pool!\n", __func__);
510
511 return pool;
512}
513
514static struct zswap_pool *zswap_pool_current(void)
515{
516 assert_spin_locked(&zswap_pools_lock);
517
518 return __zswap_pool_current();
519}
520
521static struct zswap_pool *zswap_pool_current_get(void)
522{
523 struct zswap_pool *pool;
524
525 rcu_read_lock();
526
527 pool = __zswap_pool_current();
528 if (!zswap_pool_get(pool))
529 pool = NULL;
530
531 rcu_read_unlock();
532
533 return pool;
534}
535
536static struct zswap_pool *zswap_pool_last_get(void)
537{
538 struct zswap_pool *pool, *last = NULL;
539
540 rcu_read_lock();
541
542 list_for_each_entry_rcu(pool, &zswap_pools, list)
543 last = pool;
544 WARN_ONCE(!last && zswap_has_pool,
545 "%s: no page storage pool!\n", __func__);
546 if (!zswap_pool_get(last))
547 last = NULL;
548
549 rcu_read_unlock();
550
551 return last;
552}
553
554
555static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
556{
557 struct zswap_pool *pool;
558
559 assert_spin_locked(&zswap_pools_lock);
560
561 list_for_each_entry_rcu(pool, &zswap_pools, list) {
562 if (strcmp(pool->tfm_name, compressor))
563 continue;
564 if (strcmp(zpool_get_type(pool->zpool), type))
565 continue;
566
567 if (!zswap_pool_get(pool))
568 continue;
569 return pool;
570 }
571
572 return NULL;
573}
574
575static void shrink_worker(struct work_struct *w)
576{
577 struct zswap_pool *pool = container_of(w, typeof(*pool),
578 shrink_work);
579
580 if (zpool_shrink(pool->zpool, 1, NULL))
581 zswap_reject_reclaim_fail++;
582 zswap_pool_put(pool);
583}
584
585static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
586{
587 struct zswap_pool *pool;
588 char name[38];
589 gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
590 int ret;
591
592 if (!zswap_has_pool) {
593
594
595
596
597 if (!strcmp(type, ZSWAP_PARAM_UNSET))
598 return NULL;
599 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
600 return NULL;
601 }
602
603 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
604 if (!pool)
605 return NULL;
606
607
608 snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
609
610 pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
611 if (!pool->zpool) {
612 pr_err("%s zpool not available\n", type);
613 goto error;
614 }
615 pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
616
617 strscpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
618
619 pool->acomp_ctx = alloc_percpu(*pool->acomp_ctx);
620 if (!pool->acomp_ctx) {
621 pr_err("percpu alloc failed\n");
622 goto error;
623 }
624
625 ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
626 &pool->node);
627 if (ret)
628 goto error;
629 pr_debug("using %s compressor\n", pool->tfm_name);
630
631
632
633
634 kref_init(&pool->kref);
635 INIT_LIST_HEAD(&pool->list);
636 INIT_WORK(&pool->shrink_work, shrink_worker);
637
638 zswap_pool_debug("created", pool);
639
640 return pool;
641
642error:
643 if (pool->acomp_ctx)
644 free_percpu(pool->acomp_ctx);
645 if (pool->zpool)
646 zpool_destroy_pool(pool->zpool);
647 kfree(pool);
648 return NULL;
649}
650
651static __init struct zswap_pool *__zswap_pool_create_fallback(void)
652{
653 bool has_comp, has_zpool;
654
655 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
656 if (!has_comp && strcmp(zswap_compressor,
657 CONFIG_ZSWAP_COMPRESSOR_DEFAULT)) {
658 pr_err("compressor %s not available, using default %s\n",
659 zswap_compressor, CONFIG_ZSWAP_COMPRESSOR_DEFAULT);
660 param_free_charp(&zswap_compressor);
661 zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
662 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
663 }
664 if (!has_comp) {
665 pr_err("default compressor %s not available\n",
666 zswap_compressor);
667 param_free_charp(&zswap_compressor);
668 zswap_compressor = ZSWAP_PARAM_UNSET;
669 }
670
671 has_zpool = zpool_has_pool(zswap_zpool_type);
672 if (!has_zpool && strcmp(zswap_zpool_type,
673 CONFIG_ZSWAP_ZPOOL_DEFAULT)) {
674 pr_err("zpool %s not available, using default %s\n",
675 zswap_zpool_type, CONFIG_ZSWAP_ZPOOL_DEFAULT);
676 param_free_charp(&zswap_zpool_type);
677 zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
678 has_zpool = zpool_has_pool(zswap_zpool_type);
679 }
680 if (!has_zpool) {
681 pr_err("default zpool %s not available\n",
682 zswap_zpool_type);
683 param_free_charp(&zswap_zpool_type);
684 zswap_zpool_type = ZSWAP_PARAM_UNSET;
685 }
686
687 if (!has_comp || !has_zpool)
688 return NULL;
689
690 return zswap_pool_create(zswap_zpool_type, zswap_compressor);
691}
692
693static void zswap_pool_destroy(struct zswap_pool *pool)
694{
695 zswap_pool_debug("destroying", pool);
696
697 cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
698 free_percpu(pool->acomp_ctx);
699 zpool_destroy_pool(pool->zpool);
700 kfree(pool);
701}
702
703static int __must_check zswap_pool_get(struct zswap_pool *pool)
704{
705 if (!pool)
706 return 0;
707
708 return kref_get_unless_zero(&pool->kref);
709}
710
711static void __zswap_pool_release(struct work_struct *work)
712{
713 struct zswap_pool *pool = container_of(work, typeof(*pool),
714 release_work);
715
716 synchronize_rcu();
717
718
719 WARN_ON(kref_get_unless_zero(&pool->kref));
720
721
722 zswap_pool_destroy(pool);
723}
724
725static void __zswap_pool_empty(struct kref *kref)
726{
727 struct zswap_pool *pool;
728
729 pool = container_of(kref, typeof(*pool), kref);
730
731 spin_lock(&zswap_pools_lock);
732
733 WARN_ON(pool == zswap_pool_current());
734
735 list_del_rcu(&pool->list);
736
737 INIT_WORK(&pool->release_work, __zswap_pool_release);
738 schedule_work(&pool->release_work);
739
740 spin_unlock(&zswap_pools_lock);
741}
742
743static void zswap_pool_put(struct zswap_pool *pool)
744{
745 kref_put(&pool->kref, __zswap_pool_empty);
746}
747
748
749
750
751
752
753static int __zswap_param_set(const char *val, const struct kernel_param *kp,
754 char *type, char *compressor)
755{
756 struct zswap_pool *pool, *put_pool = NULL;
757 char *s = strstrip((char *)val);
758 int ret;
759
760 if (zswap_init_failed) {
761 pr_err("can't set param, initialization failed\n");
762 return -ENODEV;
763 }
764
765
766 if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
767 return 0;
768
769
770
771
772 if (!zswap_init_started)
773 return param_set_charp(s, kp);
774
775 if (!type) {
776 if (!zpool_has_pool(s)) {
777 pr_err("zpool %s not available\n", s);
778 return -ENOENT;
779 }
780 type = s;
781 } else if (!compressor) {
782 if (!crypto_has_acomp(s, 0, 0)) {
783 pr_err("compressor %s not available\n", s);
784 return -ENOENT;
785 }
786 compressor = s;
787 } else {
788 WARN_ON(1);
789 return -EINVAL;
790 }
791
792 spin_lock(&zswap_pools_lock);
793
794 pool = zswap_pool_find_get(type, compressor);
795 if (pool) {
796 zswap_pool_debug("using existing", pool);
797 WARN_ON(pool == zswap_pool_current());
798 list_del_rcu(&pool->list);
799 }
800
801 spin_unlock(&zswap_pools_lock);
802
803 if (!pool)
804 pool = zswap_pool_create(type, compressor);
805
806 if (pool)
807 ret = param_set_charp(s, kp);
808 else
809 ret = -EINVAL;
810
811 spin_lock(&zswap_pools_lock);
812
813 if (!ret) {
814 put_pool = zswap_pool_current();
815 list_add_rcu(&pool->list, &zswap_pools);
816 zswap_has_pool = true;
817 } else if (pool) {
818
819
820
821
822 list_add_tail_rcu(&pool->list, &zswap_pools);
823 put_pool = pool;
824 }
825
826 spin_unlock(&zswap_pools_lock);
827
828 if (!zswap_has_pool && !pool) {
829
830
831
832
833
834
835
836 ret = param_set_charp(s, kp);
837 }
838
839
840
841
842 if (put_pool)
843 zswap_pool_put(put_pool);
844
845 return ret;
846}
847
848static int zswap_compressor_param_set(const char *val,
849 const struct kernel_param *kp)
850{
851 return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
852}
853
854static int zswap_zpool_param_set(const char *val,
855 const struct kernel_param *kp)
856{
857 return __zswap_param_set(val, kp, NULL, zswap_compressor);
858}
859
860static int zswap_enabled_param_set(const char *val,
861 const struct kernel_param *kp)
862{
863 if (zswap_init_failed) {
864 pr_err("can't enable, initialization failed\n");
865 return -ENODEV;
866 }
867 if (!zswap_has_pool && zswap_init_started) {
868 pr_err("can't enable, no pool configured\n");
869 return -ENODEV;
870 }
871
872 return param_set_bool(val, kp);
873}
874
875
876
877
878
879enum zswap_get_swap_ret {
880 ZSWAP_SWAPCACHE_NEW,
881 ZSWAP_SWAPCACHE_EXIST,
882 ZSWAP_SWAPCACHE_FAIL,
883};
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901static int zswap_get_swap_cache_page(swp_entry_t entry,
902 struct page **retpage)
903{
904 bool page_was_allocated;
905
906 *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
907 NULL, 0, &page_was_allocated);
908 if (page_was_allocated)
909 return ZSWAP_SWAPCACHE_NEW;
910 if (!*retpage)
911 return ZSWAP_SWAPCACHE_FAIL;
912 return ZSWAP_SWAPCACHE_EXIST;
913}
914
915
916
917
918
919
920
921
922
923
924
925
926
927static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
928{
929 struct zswap_header *zhdr;
930 swp_entry_t swpentry;
931 struct zswap_tree *tree;
932 pgoff_t offset;
933 struct zswap_entry *entry;
934 struct page *page;
935 struct scatterlist input, output;
936 struct crypto_acomp_ctx *acomp_ctx;
937
938 u8 *src, *tmp = NULL;
939 unsigned int dlen;
940 int ret;
941 struct writeback_control wbc = {
942 .sync_mode = WB_SYNC_NONE,
943 };
944
945 if (!zpool_can_sleep_mapped(pool)) {
946 tmp = kmalloc(PAGE_SIZE, GFP_ATOMIC);
947 if (!tmp)
948 return -ENOMEM;
949 }
950
951
952 zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
953 swpentry = zhdr->swpentry;
954 tree = zswap_trees[swp_type(swpentry)];
955 offset = swp_offset(swpentry);
956
957
958 spin_lock(&tree->lock);
959 entry = zswap_entry_find_get(&tree->rbroot, offset);
960 if (!entry) {
961
962 spin_unlock(&tree->lock);
963 zpool_unmap_handle(pool, handle);
964 kfree(tmp);
965 return 0;
966 }
967 spin_unlock(&tree->lock);
968 BUG_ON(offset != entry->offset);
969
970 src = (u8 *)zhdr + sizeof(struct zswap_header);
971 if (!zpool_can_sleep_mapped(pool)) {
972 memcpy(tmp, src, entry->length);
973 src = tmp;
974 zpool_unmap_handle(pool, handle);
975 }
976
977
978 switch (zswap_get_swap_cache_page(swpentry, &page)) {
979 case ZSWAP_SWAPCACHE_FAIL:
980 ret = -ENOMEM;
981 goto fail;
982
983 case ZSWAP_SWAPCACHE_EXIST:
984
985 put_page(page);
986 ret = -EEXIST;
987 goto fail;
988
989 case ZSWAP_SWAPCACHE_NEW:
990
991 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
992 dlen = PAGE_SIZE;
993
994 mutex_lock(acomp_ctx->mutex);
995 sg_init_one(&input, src, entry->length);
996 sg_init_table(&output, 1);
997 sg_set_page(&output, page, PAGE_SIZE, 0);
998 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
999 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1000 dlen = acomp_ctx->req->dlen;
1001 mutex_unlock(acomp_ctx->mutex);
1002
1003 BUG_ON(ret);
1004 BUG_ON(dlen != PAGE_SIZE);
1005
1006
1007 SetPageUptodate(page);
1008 }
1009
1010
1011 SetPageReclaim(page);
1012
1013
1014 __swap_writepage(page, &wbc, end_swap_bio_write);
1015 put_page(page);
1016 zswap_written_back_pages++;
1017
1018 spin_lock(&tree->lock);
1019
1020 zswap_entry_put(tree, entry);
1021
1022
1023
1024
1025
1026
1027
1028
1029 if (entry == zswap_rb_search(&tree->rbroot, offset))
1030 zswap_entry_put(tree, entry);
1031 spin_unlock(&tree->lock);
1032
1033 goto end;
1034
1035
1036
1037
1038
1039
1040
1041
1042fail:
1043 spin_lock(&tree->lock);
1044 zswap_entry_put(tree, entry);
1045 spin_unlock(&tree->lock);
1046
1047end:
1048 if (zpool_can_sleep_mapped(pool))
1049 zpool_unmap_handle(pool, handle);
1050 else
1051 kfree(tmp);
1052
1053 return ret;
1054}
1055
1056static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
1057{
1058 unsigned int pos;
1059 unsigned long *page;
1060
1061 page = (unsigned long *)ptr;
1062 for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
1063 if (page[pos] != page[0])
1064 return 0;
1065 }
1066 *value = page[0];
1067 return 1;
1068}
1069
1070static void zswap_fill_page(void *ptr, unsigned long value)
1071{
1072 unsigned long *page;
1073
1074 page = (unsigned long *)ptr;
1075 memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
1076}
1077
1078
1079
1080
1081
1082static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1083 struct page *page)
1084{
1085 struct zswap_tree *tree = zswap_trees[type];
1086 struct zswap_entry *entry, *dupentry;
1087 struct scatterlist input, output;
1088 struct crypto_acomp_ctx *acomp_ctx;
1089 int ret;
1090 unsigned int hlen, dlen = PAGE_SIZE;
1091 unsigned long handle, value;
1092 char *buf;
1093 u8 *src, *dst;
1094 struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1095 gfp_t gfp;
1096
1097
1098 if (PageTransHuge(page)) {
1099 ret = -EINVAL;
1100 goto reject;
1101 }
1102
1103 if (!zswap_enabled || !tree) {
1104 ret = -ENODEV;
1105 goto reject;
1106 }
1107
1108
1109 if (zswap_is_full()) {
1110 struct zswap_pool *pool;
1111
1112 zswap_pool_limit_hit++;
1113 zswap_pool_reached_full = true;
1114 pool = zswap_pool_last_get();
1115 if (pool)
1116 queue_work(shrink_wq, &pool->shrink_work);
1117 ret = -ENOMEM;
1118 goto reject;
1119 }
1120
1121 if (zswap_pool_reached_full) {
1122 if (!zswap_can_accept()) {
1123 ret = -ENOMEM;
1124 goto reject;
1125 } else
1126 zswap_pool_reached_full = false;
1127 }
1128
1129
1130 entry = zswap_entry_cache_alloc(GFP_KERNEL);
1131 if (!entry) {
1132 zswap_reject_kmemcache_fail++;
1133 ret = -ENOMEM;
1134 goto reject;
1135 }
1136
1137 if (zswap_same_filled_pages_enabled) {
1138 src = kmap_atomic(page);
1139 if (zswap_is_page_same_filled(src, &value)) {
1140 kunmap_atomic(src);
1141 entry->offset = offset;
1142 entry->length = 0;
1143 entry->value = value;
1144 atomic_inc(&zswap_same_filled_pages);
1145 goto insert_entry;
1146 }
1147 kunmap_atomic(src);
1148 }
1149
1150
1151 entry->pool = zswap_pool_current_get();
1152 if (!entry->pool) {
1153 ret = -EINVAL;
1154 goto freepage;
1155 }
1156
1157
1158 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1159
1160 mutex_lock(acomp_ctx->mutex);
1161
1162 dst = acomp_ctx->dstmem;
1163 sg_init_table(&input, 1);
1164 sg_set_page(&input, page, PAGE_SIZE, 0);
1165
1166
1167 sg_init_one(&output, dst, PAGE_SIZE * 2);
1168 acomp_request_set_params(acomp_ctx->req, &input, &output, PAGE_SIZE, dlen);
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181 ret = crypto_wait_req(crypto_acomp_compress(acomp_ctx->req), &acomp_ctx->wait);
1182 dlen = acomp_ctx->req->dlen;
1183
1184 if (ret) {
1185 ret = -EINVAL;
1186 goto put_dstmem;
1187 }
1188
1189
1190 hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1191 gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1192 if (zpool_malloc_support_movable(entry->pool->zpool))
1193 gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1194 ret = zpool_malloc(entry->pool->zpool, hlen + dlen, gfp, &handle);
1195 if (ret == -ENOSPC) {
1196 zswap_reject_compress_poor++;
1197 goto put_dstmem;
1198 }
1199 if (ret) {
1200 zswap_reject_alloc_fail++;
1201 goto put_dstmem;
1202 }
1203 buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_WO);
1204 memcpy(buf, &zhdr, hlen);
1205 memcpy(buf + hlen, dst, dlen);
1206 zpool_unmap_handle(entry->pool->zpool, handle);
1207 mutex_unlock(acomp_ctx->mutex);
1208
1209
1210 entry->offset = offset;
1211 entry->handle = handle;
1212 entry->length = dlen;
1213
1214insert_entry:
1215
1216 spin_lock(&tree->lock);
1217 do {
1218 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1219 if (ret == -EEXIST) {
1220 zswap_duplicate_entry++;
1221
1222 zswap_rb_erase(&tree->rbroot, dupentry);
1223 zswap_entry_put(tree, dupentry);
1224 }
1225 } while (ret == -EEXIST);
1226 spin_unlock(&tree->lock);
1227
1228
1229 atomic_inc(&zswap_stored_pages);
1230 zswap_update_total_size();
1231
1232 return 0;
1233
1234put_dstmem:
1235 mutex_unlock(acomp_ctx->mutex);
1236 zswap_pool_put(entry->pool);
1237freepage:
1238 zswap_entry_cache_free(entry);
1239reject:
1240 return ret;
1241}
1242
1243
1244
1245
1246
1247static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1248 struct page *page)
1249{
1250 struct zswap_tree *tree = zswap_trees[type];
1251 struct zswap_entry *entry;
1252 struct scatterlist input, output;
1253 struct crypto_acomp_ctx *acomp_ctx;
1254 u8 *src, *dst, *tmp;
1255 unsigned int dlen;
1256 int ret;
1257
1258
1259 spin_lock(&tree->lock);
1260 entry = zswap_entry_find_get(&tree->rbroot, offset);
1261 if (!entry) {
1262
1263 spin_unlock(&tree->lock);
1264 return -1;
1265 }
1266 spin_unlock(&tree->lock);
1267
1268 if (!entry->length) {
1269 dst = kmap_atomic(page);
1270 zswap_fill_page(dst, entry->value);
1271 kunmap_atomic(dst);
1272 ret = 0;
1273 goto freeentry;
1274 }
1275
1276 if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1277
1278 tmp = kmalloc(entry->length, GFP_ATOMIC);
1279 if (!tmp) {
1280 ret = -ENOMEM;
1281 goto freeentry;
1282 }
1283 }
1284
1285
1286 dlen = PAGE_SIZE;
1287 src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1288 if (zpool_evictable(entry->pool->zpool))
1289 src += sizeof(struct zswap_header);
1290
1291 if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1292
1293 memcpy(tmp, src, entry->length);
1294 src = tmp;
1295
1296 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1297 }
1298
1299 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1300 mutex_lock(acomp_ctx->mutex);
1301 sg_init_one(&input, src, entry->length);
1302 sg_init_table(&output, 1);
1303 sg_set_page(&output, page, PAGE_SIZE, 0);
1304 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1305 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1306 mutex_unlock(acomp_ctx->mutex);
1307
1308 if (zpool_can_sleep_mapped(entry->pool->zpool))
1309 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1310 else
1311 kfree(tmp);
1312
1313 BUG_ON(ret);
1314
1315freeentry:
1316 spin_lock(&tree->lock);
1317 zswap_entry_put(tree, entry);
1318 spin_unlock(&tree->lock);
1319
1320 return ret;
1321}
1322
1323
1324static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1325{
1326 struct zswap_tree *tree = zswap_trees[type];
1327 struct zswap_entry *entry;
1328
1329
1330 spin_lock(&tree->lock);
1331 entry = zswap_rb_search(&tree->rbroot, offset);
1332 if (!entry) {
1333
1334 spin_unlock(&tree->lock);
1335 return;
1336 }
1337
1338
1339 zswap_rb_erase(&tree->rbroot, entry);
1340
1341
1342 zswap_entry_put(tree, entry);
1343
1344 spin_unlock(&tree->lock);
1345}
1346
1347
1348static void zswap_frontswap_invalidate_area(unsigned type)
1349{
1350 struct zswap_tree *tree = zswap_trees[type];
1351 struct zswap_entry *entry, *n;
1352
1353 if (!tree)
1354 return;
1355
1356
1357 spin_lock(&tree->lock);
1358 rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1359 zswap_free_entry(entry);
1360 tree->rbroot = RB_ROOT;
1361 spin_unlock(&tree->lock);
1362 kfree(tree);
1363 zswap_trees[type] = NULL;
1364}
1365
1366static void zswap_frontswap_init(unsigned type)
1367{
1368 struct zswap_tree *tree;
1369
1370 tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1371 if (!tree) {
1372 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1373 return;
1374 }
1375
1376 tree->rbroot = RB_ROOT;
1377 spin_lock_init(&tree->lock);
1378 zswap_trees[type] = tree;
1379}
1380
1381static struct frontswap_ops zswap_frontswap_ops = {
1382 .store = zswap_frontswap_store,
1383 .load = zswap_frontswap_load,
1384 .invalidate_page = zswap_frontswap_invalidate_page,
1385 .invalidate_area = zswap_frontswap_invalidate_area,
1386 .init = zswap_frontswap_init
1387};
1388
1389
1390
1391
1392#ifdef CONFIG_DEBUG_FS
1393#include <linux/debugfs.h>
1394
1395static struct dentry *zswap_debugfs_root;
1396
1397static int __init zswap_debugfs_init(void)
1398{
1399 if (!debugfs_initialized())
1400 return -ENODEV;
1401
1402 zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1403
1404 debugfs_create_u64("pool_limit_hit", 0444,
1405 zswap_debugfs_root, &zswap_pool_limit_hit);
1406 debugfs_create_u64("reject_reclaim_fail", 0444,
1407 zswap_debugfs_root, &zswap_reject_reclaim_fail);
1408 debugfs_create_u64("reject_alloc_fail", 0444,
1409 zswap_debugfs_root, &zswap_reject_alloc_fail);
1410 debugfs_create_u64("reject_kmemcache_fail", 0444,
1411 zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1412 debugfs_create_u64("reject_compress_poor", 0444,
1413 zswap_debugfs_root, &zswap_reject_compress_poor);
1414 debugfs_create_u64("written_back_pages", 0444,
1415 zswap_debugfs_root, &zswap_written_back_pages);
1416 debugfs_create_u64("duplicate_entry", 0444,
1417 zswap_debugfs_root, &zswap_duplicate_entry);
1418 debugfs_create_u64("pool_total_size", 0444,
1419 zswap_debugfs_root, &zswap_pool_total_size);
1420 debugfs_create_atomic_t("stored_pages", 0444,
1421 zswap_debugfs_root, &zswap_stored_pages);
1422 debugfs_create_atomic_t("same_filled_pages", 0444,
1423 zswap_debugfs_root, &zswap_same_filled_pages);
1424
1425 return 0;
1426}
1427#else
1428static int __init zswap_debugfs_init(void)
1429{
1430 return 0;
1431}
1432#endif
1433
1434
1435
1436
1437static int __init init_zswap(void)
1438{
1439 struct zswap_pool *pool;
1440 int ret;
1441
1442 zswap_init_started = true;
1443
1444 if (zswap_entry_cache_create()) {
1445 pr_err("entry cache creation failed\n");
1446 goto cache_fail;
1447 }
1448
1449 ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1450 zswap_dstmem_prepare, zswap_dstmem_dead);
1451 if (ret) {
1452 pr_err("dstmem alloc failed\n");
1453 goto dstmem_fail;
1454 }
1455
1456 ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1457 "mm/zswap_pool:prepare",
1458 zswap_cpu_comp_prepare,
1459 zswap_cpu_comp_dead);
1460 if (ret)
1461 goto hp_fail;
1462
1463 pool = __zswap_pool_create_fallback();
1464 if (pool) {
1465 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1466 zpool_get_type(pool->zpool));
1467 list_add(&pool->list, &zswap_pools);
1468 zswap_has_pool = true;
1469 } else {
1470 pr_err("pool creation failed\n");
1471 zswap_enabled = false;
1472 }
1473
1474 shrink_wq = create_workqueue("zswap-shrink");
1475 if (!shrink_wq)
1476 goto fallback_fail;
1477
1478 frontswap_register_ops(&zswap_frontswap_ops);
1479 if (zswap_debugfs_init())
1480 pr_warn("debugfs initialization failed\n");
1481 return 0;
1482
1483fallback_fail:
1484 if (pool)
1485 zswap_pool_destroy(pool);
1486hp_fail:
1487 cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1488dstmem_fail:
1489 zswap_entry_cache_destroy();
1490cache_fail:
1491
1492 zswap_init_failed = true;
1493 zswap_enabled = false;
1494 return -ENOMEM;
1495}
1496
1497late_initcall(init_zswap);
1498
1499MODULE_LICENSE("GPL");
1500MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1501MODULE_DESCRIPTION("Compressed cache for swap pages");
1502