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/mempool.h>
28#include <linux/zpool.h>
29
30#include <linux/mm_types.h>
31#include <linux/page-flags.h>
32#include <linux/swapops.h>
33#include <linux/writeback.h>
34#include <linux/pagemap.h>
35
36
37
38
39
40static u64 zswap_pool_total_size;
41
42static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
43
44static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
45
46
47
48
49
50
51
52
53
54static u64 zswap_pool_limit_hit;
55
56static u64 zswap_written_back_pages;
57
58static u64 zswap_reject_reclaim_fail;
59
60static u64 zswap_reject_compress_poor;
61
62static u64 zswap_reject_alloc_fail;
63
64static u64 zswap_reject_kmemcache_fail;
65
66static u64 zswap_duplicate_entry;
67
68
69
70
71
72#define ZSWAP_PARAM_UNSET ""
73
74
75static bool zswap_enabled;
76static int zswap_enabled_param_set(const char *,
77 const struct kernel_param *);
78static struct kernel_param_ops zswap_enabled_param_ops = {
79 .set = zswap_enabled_param_set,
80 .get = param_get_bool,
81};
82module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
83
84
85#define ZSWAP_COMPRESSOR_DEFAULT "lzo"
86static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
87static int zswap_compressor_param_set(const char *,
88 const struct kernel_param *);
89static struct kernel_param_ops zswap_compressor_param_ops = {
90 .set = zswap_compressor_param_set,
91 .get = param_get_charp,
92 .free = param_free_charp,
93};
94module_param_cb(compressor, &zswap_compressor_param_ops,
95 &zswap_compressor, 0644);
96
97
98#define ZSWAP_ZPOOL_DEFAULT "zbud"
99static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
100static int zswap_zpool_param_set(const char *, const struct kernel_param *);
101static struct kernel_param_ops zswap_zpool_param_ops = {
102 .set = zswap_zpool_param_set,
103 .get = param_get_charp,
104 .free = param_free_charp,
105};
106module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
107
108
109static unsigned int zswap_max_pool_percent = 20;
110module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
111
112
113static bool zswap_same_filled_pages_enabled = true;
114module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
115 bool, 0644);
116
117
118
119
120
121struct zswap_pool {
122 struct zpool *zpool;
123 struct crypto_comp * __percpu *tfm;
124 struct kref kref;
125 struct list_head list;
126 struct work_struct work;
127 struct hlist_node node;
128 char tfm_name[CRYPTO_MAX_ALG_NAME];
129};
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151struct zswap_entry {
152 struct rb_node rbnode;
153 pgoff_t offset;
154 int refcount;
155 unsigned int length;
156 struct zswap_pool *pool;
157 union {
158 unsigned long handle;
159 unsigned long value;
160 };
161};
162
163struct zswap_header {
164 swp_entry_t swpentry;
165};
166
167
168
169
170
171
172struct zswap_tree {
173 struct rb_root rbroot;
174 spinlock_t lock;
175};
176
177static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
178
179
180static LIST_HEAD(zswap_pools);
181
182static DEFINE_SPINLOCK(zswap_pools_lock);
183
184static atomic_t zswap_pools_count = ATOMIC_INIT(0);
185
186
187static bool zswap_init_started;
188
189
190static bool zswap_init_failed;
191
192
193static bool zswap_has_pool;
194
195
196
197
198
199#define zswap_pool_debug(msg, p) \
200 pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name, \
201 zpool_get_type((p)->zpool))
202
203static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
204static int zswap_pool_get(struct zswap_pool *pool);
205static void zswap_pool_put(struct zswap_pool *pool);
206
207static const struct zpool_ops zswap_zpool_ops = {
208 .evict = zswap_writeback_entry
209};
210
211static bool zswap_is_full(void)
212{
213 return totalram_pages() * zswap_max_pool_percent / 100 <
214 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
215}
216
217static void zswap_update_total_size(void)
218{
219 struct zswap_pool *pool;
220 u64 total = 0;
221
222 rcu_read_lock();
223
224 list_for_each_entry_rcu(pool, &zswap_pools, list)
225 total += zpool_get_total_size(pool->zpool);
226
227 rcu_read_unlock();
228
229 zswap_pool_total_size = total;
230}
231
232
233
234
235static struct kmem_cache *zswap_entry_cache;
236
237static int __init zswap_entry_cache_create(void)
238{
239 zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
240 return zswap_entry_cache == NULL;
241}
242
243static void __init zswap_entry_cache_destroy(void)
244{
245 kmem_cache_destroy(zswap_entry_cache);
246}
247
248static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
249{
250 struct zswap_entry *entry;
251 entry = kmem_cache_alloc(zswap_entry_cache, gfp);
252 if (!entry)
253 return NULL;
254 entry->refcount = 1;
255 RB_CLEAR_NODE(&entry->rbnode);
256 return entry;
257}
258
259static void zswap_entry_cache_free(struct zswap_entry *entry)
260{
261 kmem_cache_free(zswap_entry_cache, entry);
262}
263
264
265
266
267static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
268{
269 struct rb_node *node = root->rb_node;
270 struct zswap_entry *entry;
271
272 while (node) {
273 entry = rb_entry(node, struct zswap_entry, rbnode);
274 if (entry->offset > offset)
275 node = node->rb_left;
276 else if (entry->offset < offset)
277 node = node->rb_right;
278 else
279 return entry;
280 }
281 return NULL;
282}
283
284
285
286
287
288static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
289 struct zswap_entry **dupentry)
290{
291 struct rb_node **link = &root->rb_node, *parent = NULL;
292 struct zswap_entry *myentry;
293
294 while (*link) {
295 parent = *link;
296 myentry = rb_entry(parent, struct zswap_entry, rbnode);
297 if (myentry->offset > entry->offset)
298 link = &(*link)->rb_left;
299 else if (myentry->offset < entry->offset)
300 link = &(*link)->rb_right;
301 else {
302 *dupentry = myentry;
303 return -EEXIST;
304 }
305 }
306 rb_link_node(&entry->rbnode, parent, link);
307 rb_insert_color(&entry->rbnode, root);
308 return 0;
309}
310
311static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
312{
313 if (!RB_EMPTY_NODE(&entry->rbnode)) {
314 rb_erase(&entry->rbnode, root);
315 RB_CLEAR_NODE(&entry->rbnode);
316 }
317}
318
319
320
321
322
323static void zswap_free_entry(struct zswap_entry *entry)
324{
325 if (!entry->length)
326 atomic_dec(&zswap_same_filled_pages);
327 else {
328 zpool_free(entry->pool->zpool, entry->handle);
329 zswap_pool_put(entry->pool);
330 }
331 zswap_entry_cache_free(entry);
332 atomic_dec(&zswap_stored_pages);
333 zswap_update_total_size();
334}
335
336
337static void zswap_entry_get(struct zswap_entry *entry)
338{
339 entry->refcount++;
340}
341
342
343
344
345static void zswap_entry_put(struct zswap_tree *tree,
346 struct zswap_entry *entry)
347{
348 int refcount = --entry->refcount;
349
350 BUG_ON(refcount < 0);
351 if (refcount == 0) {
352 zswap_rb_erase(&tree->rbroot, entry);
353 zswap_free_entry(entry);
354 }
355}
356
357
358static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
359 pgoff_t offset)
360{
361 struct zswap_entry *entry;
362
363 entry = zswap_rb_search(root, offset);
364 if (entry)
365 zswap_entry_get(entry);
366
367 return entry;
368}
369
370
371
372
373static DEFINE_PER_CPU(u8 *, zswap_dstmem);
374
375static int zswap_dstmem_prepare(unsigned int cpu)
376{
377 u8 *dst;
378
379 dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
380 if (!dst)
381 return -ENOMEM;
382
383 per_cpu(zswap_dstmem, cpu) = dst;
384 return 0;
385}
386
387static int zswap_dstmem_dead(unsigned int cpu)
388{
389 u8 *dst;
390
391 dst = per_cpu(zswap_dstmem, cpu);
392 kfree(dst);
393 per_cpu(zswap_dstmem, cpu) = NULL;
394
395 return 0;
396}
397
398static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
399{
400 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
401 struct crypto_comp *tfm;
402
403 if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu)))
404 return 0;
405
406 tfm = crypto_alloc_comp(pool->tfm_name, 0, 0);
407 if (IS_ERR_OR_NULL(tfm)) {
408 pr_err("could not alloc crypto comp %s : %ld\n",
409 pool->tfm_name, PTR_ERR(tfm));
410 return -ENOMEM;
411 }
412 *per_cpu_ptr(pool->tfm, cpu) = tfm;
413 return 0;
414}
415
416static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
417{
418 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
419 struct crypto_comp *tfm;
420
421 tfm = *per_cpu_ptr(pool->tfm, cpu);
422 if (!IS_ERR_OR_NULL(tfm))
423 crypto_free_comp(tfm);
424 *per_cpu_ptr(pool->tfm, cpu) = NULL;
425 return 0;
426}
427
428
429
430
431
432static struct zswap_pool *__zswap_pool_current(void)
433{
434 struct zswap_pool *pool;
435
436 pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
437 WARN_ONCE(!pool && zswap_has_pool,
438 "%s: no page storage pool!\n", __func__);
439
440 return pool;
441}
442
443static struct zswap_pool *zswap_pool_current(void)
444{
445 assert_spin_locked(&zswap_pools_lock);
446
447 return __zswap_pool_current();
448}
449
450static struct zswap_pool *zswap_pool_current_get(void)
451{
452 struct zswap_pool *pool;
453
454 rcu_read_lock();
455
456 pool = __zswap_pool_current();
457 if (!zswap_pool_get(pool))
458 pool = NULL;
459
460 rcu_read_unlock();
461
462 return pool;
463}
464
465static struct zswap_pool *zswap_pool_last_get(void)
466{
467 struct zswap_pool *pool, *last = NULL;
468
469 rcu_read_lock();
470
471 list_for_each_entry_rcu(pool, &zswap_pools, list)
472 last = pool;
473 WARN_ONCE(!last && zswap_has_pool,
474 "%s: no page storage pool!\n", __func__);
475 if (!zswap_pool_get(last))
476 last = NULL;
477
478 rcu_read_unlock();
479
480 return last;
481}
482
483
484static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
485{
486 struct zswap_pool *pool;
487
488 assert_spin_locked(&zswap_pools_lock);
489
490 list_for_each_entry_rcu(pool, &zswap_pools, list) {
491 if (strcmp(pool->tfm_name, compressor))
492 continue;
493 if (strcmp(zpool_get_type(pool->zpool), type))
494 continue;
495
496 if (!zswap_pool_get(pool))
497 continue;
498 return pool;
499 }
500
501 return NULL;
502}
503
504static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
505{
506 struct zswap_pool *pool;
507 char name[38];
508 gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
509 int ret;
510
511 if (!zswap_has_pool) {
512
513
514
515
516 if (!strcmp(type, ZSWAP_PARAM_UNSET))
517 return NULL;
518 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
519 return NULL;
520 }
521
522 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
523 if (!pool)
524 return NULL;
525
526
527 snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
528
529 pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
530 if (!pool->zpool) {
531 pr_err("%s zpool not available\n", type);
532 goto error;
533 }
534 pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
535
536 strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
537 pool->tfm = alloc_percpu(struct crypto_comp *);
538 if (!pool->tfm) {
539 pr_err("percpu alloc failed\n");
540 goto error;
541 }
542
543 ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
544 &pool->node);
545 if (ret)
546 goto error;
547 pr_debug("using %s compressor\n", pool->tfm_name);
548
549
550
551
552 kref_init(&pool->kref);
553 INIT_LIST_HEAD(&pool->list);
554
555 zswap_pool_debug("created", pool);
556
557 return pool;
558
559error:
560 free_percpu(pool->tfm);
561 if (pool->zpool)
562 zpool_destroy_pool(pool->zpool);
563 kfree(pool);
564 return NULL;
565}
566
567static __init struct zswap_pool *__zswap_pool_create_fallback(void)
568{
569 bool has_comp, has_zpool;
570
571 has_comp = crypto_has_comp(zswap_compressor, 0, 0);
572 if (!has_comp && strcmp(zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT)) {
573 pr_err("compressor %s not available, using default %s\n",
574 zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT);
575 param_free_charp(&zswap_compressor);
576 zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
577 has_comp = crypto_has_comp(zswap_compressor, 0, 0);
578 }
579 if (!has_comp) {
580 pr_err("default compressor %s not available\n",
581 zswap_compressor);
582 param_free_charp(&zswap_compressor);
583 zswap_compressor = ZSWAP_PARAM_UNSET;
584 }
585
586 has_zpool = zpool_has_pool(zswap_zpool_type);
587 if (!has_zpool && strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) {
588 pr_err("zpool %s not available, using default %s\n",
589 zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT);
590 param_free_charp(&zswap_zpool_type);
591 zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
592 has_zpool = zpool_has_pool(zswap_zpool_type);
593 }
594 if (!has_zpool) {
595 pr_err("default zpool %s not available\n",
596 zswap_zpool_type);
597 param_free_charp(&zswap_zpool_type);
598 zswap_zpool_type = ZSWAP_PARAM_UNSET;
599 }
600
601 if (!has_comp || !has_zpool)
602 return NULL;
603
604 return zswap_pool_create(zswap_zpool_type, zswap_compressor);
605}
606
607static void zswap_pool_destroy(struct zswap_pool *pool)
608{
609 zswap_pool_debug("destroying", pool);
610
611 cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
612 free_percpu(pool->tfm);
613 zpool_destroy_pool(pool->zpool);
614 kfree(pool);
615}
616
617static int __must_check zswap_pool_get(struct zswap_pool *pool)
618{
619 if (!pool)
620 return 0;
621
622 return kref_get_unless_zero(&pool->kref);
623}
624
625static void __zswap_pool_release(struct work_struct *work)
626{
627 struct zswap_pool *pool = container_of(work, typeof(*pool), work);
628
629 synchronize_rcu();
630
631
632 WARN_ON(kref_get_unless_zero(&pool->kref));
633
634
635 zswap_pool_destroy(pool);
636}
637
638static void __zswap_pool_empty(struct kref *kref)
639{
640 struct zswap_pool *pool;
641
642 pool = container_of(kref, typeof(*pool), kref);
643
644 spin_lock(&zswap_pools_lock);
645
646 WARN_ON(pool == zswap_pool_current());
647
648 list_del_rcu(&pool->list);
649
650 INIT_WORK(&pool->work, __zswap_pool_release);
651 schedule_work(&pool->work);
652
653 spin_unlock(&zswap_pools_lock);
654}
655
656static void zswap_pool_put(struct zswap_pool *pool)
657{
658 kref_put(&pool->kref, __zswap_pool_empty);
659}
660
661
662
663
664
665
666static int __zswap_param_set(const char *val, const struct kernel_param *kp,
667 char *type, char *compressor)
668{
669 struct zswap_pool *pool, *put_pool = NULL;
670 char *s = strstrip((char *)val);
671 int ret;
672
673 if (zswap_init_failed) {
674 pr_err("can't set param, initialization failed\n");
675 return -ENODEV;
676 }
677
678
679 if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
680 return 0;
681
682
683
684
685 if (!zswap_init_started)
686 return param_set_charp(s, kp);
687
688 if (!type) {
689 if (!zpool_has_pool(s)) {
690 pr_err("zpool %s not available\n", s);
691 return -ENOENT;
692 }
693 type = s;
694 } else if (!compressor) {
695 if (!crypto_has_comp(s, 0, 0)) {
696 pr_err("compressor %s not available\n", s);
697 return -ENOENT;
698 }
699 compressor = s;
700 } else {
701 WARN_ON(1);
702 return -EINVAL;
703 }
704
705 spin_lock(&zswap_pools_lock);
706
707 pool = zswap_pool_find_get(type, compressor);
708 if (pool) {
709 zswap_pool_debug("using existing", pool);
710 WARN_ON(pool == zswap_pool_current());
711 list_del_rcu(&pool->list);
712 }
713
714 spin_unlock(&zswap_pools_lock);
715
716 if (!pool)
717 pool = zswap_pool_create(type, compressor);
718
719 if (pool)
720 ret = param_set_charp(s, kp);
721 else
722 ret = -EINVAL;
723
724 spin_lock(&zswap_pools_lock);
725
726 if (!ret) {
727 put_pool = zswap_pool_current();
728 list_add_rcu(&pool->list, &zswap_pools);
729 zswap_has_pool = true;
730 } else if (pool) {
731
732
733
734
735 list_add_tail_rcu(&pool->list, &zswap_pools);
736 put_pool = pool;
737 }
738
739 spin_unlock(&zswap_pools_lock);
740
741 if (!zswap_has_pool && !pool) {
742
743
744
745
746
747
748
749 ret = param_set_charp(s, kp);
750 }
751
752
753
754
755 if (put_pool)
756 zswap_pool_put(put_pool);
757
758 return ret;
759}
760
761static int zswap_compressor_param_set(const char *val,
762 const struct kernel_param *kp)
763{
764 return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
765}
766
767static int zswap_zpool_param_set(const char *val,
768 const struct kernel_param *kp)
769{
770 return __zswap_param_set(val, kp, NULL, zswap_compressor);
771}
772
773static int zswap_enabled_param_set(const char *val,
774 const struct kernel_param *kp)
775{
776 if (zswap_init_failed) {
777 pr_err("can't enable, initialization failed\n");
778 return -ENODEV;
779 }
780 if (!zswap_has_pool && zswap_init_started) {
781 pr_err("can't enable, no pool configured\n");
782 return -ENODEV;
783 }
784
785 return param_set_bool(val, kp);
786}
787
788
789
790
791
792enum zswap_get_swap_ret {
793 ZSWAP_SWAPCACHE_NEW,
794 ZSWAP_SWAPCACHE_EXIST,
795 ZSWAP_SWAPCACHE_FAIL,
796};
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814static int zswap_get_swap_cache_page(swp_entry_t entry,
815 struct page **retpage)
816{
817 bool page_was_allocated;
818
819 *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
820 NULL, 0, &page_was_allocated);
821 if (page_was_allocated)
822 return ZSWAP_SWAPCACHE_NEW;
823 if (!*retpage)
824 return ZSWAP_SWAPCACHE_FAIL;
825 return ZSWAP_SWAPCACHE_EXIST;
826}
827
828
829
830
831
832
833
834
835
836
837
838
839
840static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
841{
842 struct zswap_header *zhdr;
843 swp_entry_t swpentry;
844 struct zswap_tree *tree;
845 pgoff_t offset;
846 struct zswap_entry *entry;
847 struct page *page;
848 struct crypto_comp *tfm;
849 u8 *src, *dst;
850 unsigned int dlen;
851 int ret;
852 struct writeback_control wbc = {
853 .sync_mode = WB_SYNC_NONE,
854 };
855
856
857 zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
858 swpentry = zhdr->swpentry;
859 zpool_unmap_handle(pool, handle);
860 tree = zswap_trees[swp_type(swpentry)];
861 offset = swp_offset(swpentry);
862
863
864 spin_lock(&tree->lock);
865 entry = zswap_entry_find_get(&tree->rbroot, offset);
866 if (!entry) {
867
868 spin_unlock(&tree->lock);
869 return 0;
870 }
871 spin_unlock(&tree->lock);
872 BUG_ON(offset != entry->offset);
873
874
875 switch (zswap_get_swap_cache_page(swpentry, &page)) {
876 case ZSWAP_SWAPCACHE_FAIL:
877 ret = -ENOMEM;
878 goto fail;
879
880 case ZSWAP_SWAPCACHE_EXIST:
881
882 put_page(page);
883 ret = -EEXIST;
884 goto fail;
885
886 case ZSWAP_SWAPCACHE_NEW:
887
888 dlen = PAGE_SIZE;
889 src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle,
890 ZPOOL_MM_RO) + sizeof(struct zswap_header);
891 dst = kmap_atomic(page);
892 tfm = *get_cpu_ptr(entry->pool->tfm);
893 ret = crypto_comp_decompress(tfm, src, entry->length,
894 dst, &dlen);
895 put_cpu_ptr(entry->pool->tfm);
896 kunmap_atomic(dst);
897 zpool_unmap_handle(entry->pool->zpool, entry->handle);
898 BUG_ON(ret);
899 BUG_ON(dlen != PAGE_SIZE);
900
901
902 SetPageUptodate(page);
903 }
904
905
906 SetPageReclaim(page);
907
908
909 __swap_writepage(page, &wbc, end_swap_bio_write);
910 put_page(page);
911 zswap_written_back_pages++;
912
913 spin_lock(&tree->lock);
914
915 zswap_entry_put(tree, entry);
916
917
918
919
920
921
922
923
924 if (entry == zswap_rb_search(&tree->rbroot, offset))
925 zswap_entry_put(tree, entry);
926 spin_unlock(&tree->lock);
927
928 goto end;
929
930
931
932
933
934
935
936
937fail:
938 spin_lock(&tree->lock);
939 zswap_entry_put(tree, entry);
940 spin_unlock(&tree->lock);
941
942end:
943 return ret;
944}
945
946static int zswap_shrink(void)
947{
948 struct zswap_pool *pool;
949 int ret;
950
951 pool = zswap_pool_last_get();
952 if (!pool)
953 return -ENOENT;
954
955 ret = zpool_shrink(pool->zpool, 1, NULL);
956
957 zswap_pool_put(pool);
958
959 return ret;
960}
961
962static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
963{
964 unsigned int pos;
965 unsigned long *page;
966
967 page = (unsigned long *)ptr;
968 for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
969 if (page[pos] != page[0])
970 return 0;
971 }
972 *value = page[0];
973 return 1;
974}
975
976static void zswap_fill_page(void *ptr, unsigned long value)
977{
978 unsigned long *page;
979
980 page = (unsigned long *)ptr;
981 memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
982}
983
984
985
986
987
988static int zswap_frontswap_store(unsigned type, pgoff_t offset,
989 struct page *page)
990{
991 struct zswap_tree *tree = zswap_trees[type];
992 struct zswap_entry *entry, *dupentry;
993 struct crypto_comp *tfm;
994 int ret;
995 unsigned int hlen, dlen = PAGE_SIZE;
996 unsigned long handle, value;
997 char *buf;
998 u8 *src, *dst;
999 struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1000
1001
1002 if (PageTransHuge(page)) {
1003 ret = -EINVAL;
1004 goto reject;
1005 }
1006
1007 if (!zswap_enabled || !tree) {
1008 ret = -ENODEV;
1009 goto reject;
1010 }
1011
1012
1013 if (zswap_is_full()) {
1014 zswap_pool_limit_hit++;
1015 if (zswap_shrink()) {
1016 zswap_reject_reclaim_fail++;
1017 ret = -ENOMEM;
1018 goto reject;
1019 }
1020
1021
1022
1023
1024
1025 if (zswap_is_full()) {
1026 ret = -ENOMEM;
1027 goto reject;
1028 }
1029 }
1030
1031
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
1053 entry->pool = zswap_pool_current_get();
1054 if (!entry->pool) {
1055 ret = -EINVAL;
1056 goto freepage;
1057 }
1058
1059
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
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
1091 entry->offset = offset;
1092 entry->handle = handle;
1093 entry->length = dlen;
1094
1095insert_entry:
1096
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
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
1110 atomic_inc(&zswap_stored_pages);
1111 zswap_update_total_size();
1112
1113 return 0;
1114
1115put_dstmem:
1116 put_cpu_var(zswap_dstmem);
1117 zswap_pool_put(entry->pool);
1118freepage:
1119 zswap_entry_cache_free(entry);
1120reject:
1121 return ret;
1122}
1123
1124
1125
1126
1127
1128static 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
1139 spin_lock(&tree->lock);
1140 entry = zswap_entry_find_get(&tree->rbroot, offset);
1141 if (!entry) {
1142
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
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
1168freeentry:
1169 spin_lock(&tree->lock);
1170 zswap_entry_put(tree, entry);
1171 spin_unlock(&tree->lock);
1172
1173 return 0;
1174}
1175
1176
1177static 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
1183 spin_lock(&tree->lock);
1184 entry = zswap_rb_search(&tree->rbroot, offset);
1185 if (!entry) {
1186
1187 spin_unlock(&tree->lock);
1188 return;
1189 }
1190
1191
1192 zswap_rb_erase(&tree->rbroot, entry);
1193
1194
1195 zswap_entry_put(tree, entry);
1196
1197 spin_unlock(&tree->lock);
1198}
1199
1200
1201static 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
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
1219static 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
1234static 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
1244
1245#ifdef CONFIG_DEBUG_FS
1246#include <linux/debugfs.h>
1247
1248static struct dentry *zswap_debugfs_root;
1249
1250static 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", 0444,
1260 zswap_debugfs_root, &zswap_pool_limit_hit);
1261 debugfs_create_u64("reject_reclaim_fail", 0444,
1262 zswap_debugfs_root, &zswap_reject_reclaim_fail);
1263 debugfs_create_u64("reject_alloc_fail", 0444,
1264 zswap_debugfs_root, &zswap_reject_alloc_fail);
1265 debugfs_create_u64("reject_kmemcache_fail", 0444,
1266 zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1267 debugfs_create_u64("reject_compress_poor", 0444,
1268 zswap_debugfs_root, &zswap_reject_compress_poor);
1269 debugfs_create_u64("written_back_pages", 0444,
1270 zswap_debugfs_root, &zswap_written_back_pages);
1271 debugfs_create_u64("duplicate_entry", 0444,
1272 zswap_debugfs_root, &zswap_duplicate_entry);
1273 debugfs_create_u64("pool_total_size", 0444,
1274 zswap_debugfs_root, &zswap_pool_total_size);
1275 debugfs_create_atomic_t("stored_pages", 0444,
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
1283static void __exit zswap_debugfs_exit(void)
1284{
1285 debugfs_remove_recursive(zswap_debugfs_root);
1286}
1287#else
1288static int __init zswap_debugfs_init(void)
1289{
1290 return 0;
1291}
1292
1293static void __exit zswap_debugfs_exit(void) { }
1294#endif
1295
1296
1297
1298
1299static 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
1341hp_fail:
1342 cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1343dstmem_fail:
1344 zswap_entry_cache_destroy();
1345cache_fail:
1346
1347 zswap_init_failed = true;
1348 zswap_enabled = false;
1349 return -ENOMEM;
1350}
1351
1352late_initcall(init_zswap);
1353
1354MODULE_LICENSE("GPL");
1355MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1356MODULE_DESCRIPTION("Compressed cache for swap pages");
1357