OSDN Git Service

7ec8bd912d13223a398d5142c33e6e63a1e4f03b
[tomoyo/tomoyo-test1.git] / mm / zswap.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * zswap.c - zswap driver file
4  *
5  * zswap is a backend for frontswap that takes pages that are in the process
6  * of being swapped out and attempts to compress and store them in a
7  * RAM-based memory pool.  This can result in a significant I/O reduction on
8  * the swap device and, in the case where decompressing from RAM is faster
9  * than reading from the swap device, can also improve workload performance.
10  *
11  * Copyright (C) 2012  Seth Jennings <sjenning@linux.vnet.ibm.com>
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 #include <linux/workqueue.h>
36
37 /*********************************
38 * statistics
39 **********************************/
40 /* Total bytes used by the compressed storage */
41 static u64 zswap_pool_total_size;
42 /* The number of compressed pages currently stored in zswap */
43 static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
44 /* The number of same-value filled pages currently stored in zswap */
45 static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
46
47 /*
48  * The statistics below are not protected from concurrent access for
49  * performance reasons so they may not be a 100% accurate.  However,
50  * they do provide useful information on roughly how many times a
51  * certain event is occurring.
52 */
53
54 /* Pool limit was hit (see zswap_max_pool_percent) */
55 static u64 zswap_pool_limit_hit;
56 /* Pages written back when pool limit was reached */
57 static u64 zswap_written_back_pages;
58 /* Store failed due to a reclaim failure after pool limit was reached */
59 static u64 zswap_reject_reclaim_fail;
60 /* Compressed page was too big for the allocator to (optimally) store */
61 static u64 zswap_reject_compress_poor;
62 /* Store failed because underlying allocator could not get memory */
63 static u64 zswap_reject_alloc_fail;
64 /* Store failed because the entry metadata could not be allocated (rare) */
65 static u64 zswap_reject_kmemcache_fail;
66 /* Duplicate store was encountered (rare) */
67 static u64 zswap_duplicate_entry;
68
69 /* Shrinker work queue */
70 static struct workqueue_struct *shrink_wq;
71 /* Pool limit was hit, we need to calm down */
72 static bool zswap_pool_reached_full;
73
74 /*********************************
75 * tunables
76 **********************************/
77
78 #define ZSWAP_PARAM_UNSET ""
79
80 /* Enable/disable zswap (disabled by default) */
81 static bool zswap_enabled;
82 static int zswap_enabled_param_set(const char *,
83                                    const struct kernel_param *);
84 static struct kernel_param_ops zswap_enabled_param_ops = {
85         .set =          zswap_enabled_param_set,
86         .get =          param_get_bool,
87 };
88 module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
89
90 /* Crypto compressor to use */
91 #define ZSWAP_COMPRESSOR_DEFAULT "lzo"
92 static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
93 static int zswap_compressor_param_set(const char *,
94                                       const struct kernel_param *);
95 static struct kernel_param_ops zswap_compressor_param_ops = {
96         .set =          zswap_compressor_param_set,
97         .get =          param_get_charp,
98         .free =         param_free_charp,
99 };
100 module_param_cb(compressor, &zswap_compressor_param_ops,
101                 &zswap_compressor, 0644);
102
103 /* Compressed storage zpool to use */
104 #define ZSWAP_ZPOOL_DEFAULT "zbud"
105 static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
106 static int zswap_zpool_param_set(const char *, const struct kernel_param *);
107 static 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 };
112 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
113
114 /* The maximum percentage of memory that the compressed pool can occupy */
115 static unsigned int zswap_max_pool_percent = 20;
116 module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
117
118 /* The threshold for accepting new pages after the max_pool_percent was hit */
119 static unsigned int zswap_accept_thr_percent = 90; /* of max pool size */
120 module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
121                    uint, 0644);
122
123 /* Enable/disable handling same-value filled pages (enabled by default) */
124 static bool zswap_same_filled_pages_enabled = true;
125 module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
126                    bool, 0644);
127
128 /*********************************
129 * data structures
130 **********************************/
131
132 struct zswap_pool {
133         struct zpool *zpool;
134         struct crypto_comp * __percpu *tfm;
135         struct kref kref;
136         struct list_head list;
137         struct work_struct release_work;
138         struct work_struct shrink_work;
139         struct hlist_node node;
140         char tfm_name[CRYPTO_MAX_ALG_NAME];
141 };
142
143 /*
144  * struct zswap_entry
145  *
146  * This structure contains the metadata for tracking a single compressed
147  * page within zswap.
148  *
149  * rbnode - links the entry into red-black tree for the appropriate swap type
150  * offset - the swap offset for the entry.  Index into the red-black tree.
151  * refcount - the number of outstanding reference to the entry. This is needed
152  *            to protect against premature freeing of the entry by code
153  *            concurrent calls to load, invalidate, and writeback.  The lock
154  *            for the zswap_tree structure that contains the entry must
155  *            be held while changing the refcount.  Since the lock must
156  *            be held, there is no reason to also make refcount atomic.
157  * length - the length in bytes of the compressed page data.  Needed during
158  *          decompression. For a same value filled page length is 0.
159  * pool - the zswap_pool the entry's data is in
160  * handle - zpool allocation handle that stores the compressed page data
161  * value - value of the same-value filled pages which have same content
162  */
163 struct zswap_entry {
164         struct rb_node rbnode;
165         pgoff_t offset;
166         int refcount;
167         unsigned int length;
168         struct zswap_pool *pool;
169         union {
170                 unsigned long handle;
171                 unsigned long value;
172         };
173 };
174
175 struct zswap_header {
176         swp_entry_t swpentry;
177 };
178
179 /*
180  * The tree lock in the zswap_tree struct protects a few things:
181  * - the rbtree
182  * - the refcount field of each entry in the tree
183  */
184 struct zswap_tree {
185         struct rb_root rbroot;
186         spinlock_t lock;
187 };
188
189 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
190
191 /* RCU-protected iteration */
192 static LIST_HEAD(zswap_pools);
193 /* protects zswap_pools list modification */
194 static DEFINE_SPINLOCK(zswap_pools_lock);
195 /* pool counter to provide unique names to zpool */
196 static atomic_t zswap_pools_count = ATOMIC_INIT(0);
197
198 /* used by param callback function */
199 static bool zswap_init_started;
200
201 /* fatal error during init */
202 static bool zswap_init_failed;
203
204 /* init completed, but couldn't create the initial pool */
205 static bool zswap_has_pool;
206
207 /*********************************
208 * helpers and fwd declarations
209 **********************************/
210
211 #define zswap_pool_debug(msg, p)                                \
212         pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name,         \
213                  zpool_get_type((p)->zpool))
214
215 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
216 static int zswap_pool_get(struct zswap_pool *pool);
217 static void zswap_pool_put(struct zswap_pool *pool);
218
219 static const struct zpool_ops zswap_zpool_ops = {
220         .evict = zswap_writeback_entry
221 };
222
223 static bool zswap_is_full(void)
224 {
225         return totalram_pages() * zswap_max_pool_percent / 100 <
226                         DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
227 }
228
229 static bool zswap_can_accept(void)
230 {
231         return totalram_pages() * zswap_accept_thr_percent / 100 *
232                                 zswap_max_pool_percent / 100 >
233                         DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
234 }
235
236 static void zswap_update_total_size(void)
237 {
238         struct zswap_pool *pool;
239         u64 total = 0;
240
241         rcu_read_lock();
242
243         list_for_each_entry_rcu(pool, &zswap_pools, list)
244                 total += zpool_get_total_size(pool->zpool);
245
246         rcu_read_unlock();
247
248         zswap_pool_total_size = total;
249 }
250
251 /*********************************
252 * zswap entry functions
253 **********************************/
254 static struct kmem_cache *zswap_entry_cache;
255
256 static int __init zswap_entry_cache_create(void)
257 {
258         zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
259         return zswap_entry_cache == NULL;
260 }
261
262 static void __init zswap_entry_cache_destroy(void)
263 {
264         kmem_cache_destroy(zswap_entry_cache);
265 }
266
267 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
268 {
269         struct zswap_entry *entry;
270         entry = kmem_cache_alloc(zswap_entry_cache, gfp);
271         if (!entry)
272                 return NULL;
273         entry->refcount = 1;
274         RB_CLEAR_NODE(&entry->rbnode);
275         return entry;
276 }
277
278 static void zswap_entry_cache_free(struct zswap_entry *entry)
279 {
280         kmem_cache_free(zswap_entry_cache, entry);
281 }
282
283 /*********************************
284 * rbtree functions
285 **********************************/
286 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
287 {
288         struct rb_node *node = root->rb_node;
289         struct zswap_entry *entry;
290
291         while (node) {
292                 entry = rb_entry(node, struct zswap_entry, rbnode);
293                 if (entry->offset > offset)
294                         node = node->rb_left;
295                 else if (entry->offset < offset)
296                         node = node->rb_right;
297                 else
298                         return entry;
299         }
300         return NULL;
301 }
302
303 /*
304  * In the case that a entry with the same offset is found, a pointer to
305  * the existing entry is stored in dupentry and the function returns -EEXIST
306  */
307 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
308                         struct zswap_entry **dupentry)
309 {
310         struct rb_node **link = &root->rb_node, *parent = NULL;
311         struct zswap_entry *myentry;
312
313         while (*link) {
314                 parent = *link;
315                 myentry = rb_entry(parent, struct zswap_entry, rbnode);
316                 if (myentry->offset > entry->offset)
317                         link = &(*link)->rb_left;
318                 else if (myentry->offset < entry->offset)
319                         link = &(*link)->rb_right;
320                 else {
321                         *dupentry = myentry;
322                         return -EEXIST;
323                 }
324         }
325         rb_link_node(&entry->rbnode, parent, link);
326         rb_insert_color(&entry->rbnode, root);
327         return 0;
328 }
329
330 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
331 {
332         if (!RB_EMPTY_NODE(&entry->rbnode)) {
333                 rb_erase(&entry->rbnode, root);
334                 RB_CLEAR_NODE(&entry->rbnode);
335         }
336 }
337
338 /*
339  * Carries out the common pattern of freeing and entry's zpool allocation,
340  * freeing the entry itself, and decrementing the number of stored pages.
341  */
342 static void zswap_free_entry(struct zswap_entry *entry)
343 {
344         if (!entry->length)
345                 atomic_dec(&zswap_same_filled_pages);
346         else {
347                 zpool_free(entry->pool->zpool, entry->handle);
348                 zswap_pool_put(entry->pool);
349         }
350         zswap_entry_cache_free(entry);
351         atomic_dec(&zswap_stored_pages);
352         zswap_update_total_size();
353 }
354
355 /* caller must hold the tree lock */
356 static void zswap_entry_get(struct zswap_entry *entry)
357 {
358         entry->refcount++;
359 }
360
361 /* caller must hold the tree lock
362 * remove from the tree and free it, if nobody reference the entry
363 */
364 static void zswap_entry_put(struct zswap_tree *tree,
365                         struct zswap_entry *entry)
366 {
367         int refcount = --entry->refcount;
368
369         BUG_ON(refcount < 0);
370         if (refcount == 0) {
371                 zswap_rb_erase(&tree->rbroot, entry);
372                 zswap_free_entry(entry);
373         }
374 }
375
376 /* caller must hold the tree lock */
377 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
378                                 pgoff_t offset)
379 {
380         struct zswap_entry *entry;
381
382         entry = zswap_rb_search(root, offset);
383         if (entry)
384                 zswap_entry_get(entry);
385
386         return entry;
387 }
388
389 /*********************************
390 * per-cpu code
391 **********************************/
392 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
393
394 static int zswap_dstmem_prepare(unsigned int cpu)
395 {
396         u8 *dst;
397
398         dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
399         if (!dst)
400                 return -ENOMEM;
401
402         per_cpu(zswap_dstmem, cpu) = dst;
403         return 0;
404 }
405
406 static int zswap_dstmem_dead(unsigned int cpu)
407 {
408         u8 *dst;
409
410         dst = per_cpu(zswap_dstmem, cpu);
411         kfree(dst);
412         per_cpu(zswap_dstmem, cpu) = NULL;
413
414         return 0;
415 }
416
417 static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
418 {
419         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
420         struct crypto_comp *tfm;
421
422         if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu)))
423                 return 0;
424
425         tfm = crypto_alloc_comp(pool->tfm_name, 0, 0);
426         if (IS_ERR_OR_NULL(tfm)) {
427                 pr_err("could not alloc crypto comp %s : %ld\n",
428                        pool->tfm_name, PTR_ERR(tfm));
429                 return -ENOMEM;
430         }
431         *per_cpu_ptr(pool->tfm, cpu) = tfm;
432         return 0;
433 }
434
435 static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
436 {
437         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
438         struct crypto_comp *tfm;
439
440         tfm = *per_cpu_ptr(pool->tfm, cpu);
441         if (!IS_ERR_OR_NULL(tfm))
442                 crypto_free_comp(tfm);
443         *per_cpu_ptr(pool->tfm, cpu) = NULL;
444         return 0;
445 }
446
447 /*********************************
448 * pool functions
449 **********************************/
450
451 static struct zswap_pool *__zswap_pool_current(void)
452 {
453         struct zswap_pool *pool;
454
455         pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
456         WARN_ONCE(!pool && zswap_has_pool,
457                   "%s: no page storage pool!\n", __func__);
458
459         return pool;
460 }
461
462 static struct zswap_pool *zswap_pool_current(void)
463 {
464         assert_spin_locked(&zswap_pools_lock);
465
466         return __zswap_pool_current();
467 }
468
469 static struct zswap_pool *zswap_pool_current_get(void)
470 {
471         struct zswap_pool *pool;
472
473         rcu_read_lock();
474
475         pool = __zswap_pool_current();
476         if (!zswap_pool_get(pool))
477                 pool = NULL;
478
479         rcu_read_unlock();
480
481         return pool;
482 }
483
484 static struct zswap_pool *zswap_pool_last_get(void)
485 {
486         struct zswap_pool *pool, *last = NULL;
487
488         rcu_read_lock();
489
490         list_for_each_entry_rcu(pool, &zswap_pools, list)
491                 last = pool;
492         WARN_ONCE(!last && zswap_has_pool,
493                   "%s: no page storage pool!\n", __func__);
494         if (!zswap_pool_get(last))
495                 last = NULL;
496
497         rcu_read_unlock();
498
499         return last;
500 }
501
502 /* type and compressor must be null-terminated */
503 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
504 {
505         struct zswap_pool *pool;
506
507         assert_spin_locked(&zswap_pools_lock);
508
509         list_for_each_entry_rcu(pool, &zswap_pools, list) {
510                 if (strcmp(pool->tfm_name, compressor))
511                         continue;
512                 if (strcmp(zpool_get_type(pool->zpool), type))
513                         continue;
514                 /* if we can't get it, it's about to be destroyed */
515                 if (!zswap_pool_get(pool))
516                         continue;
517                 return pool;
518         }
519
520         return NULL;
521 }
522
523 static void shrink_worker(struct work_struct *w)
524 {
525         struct zswap_pool *pool = container_of(w, typeof(*pool),
526                                                 shrink_work);
527
528         if (zpool_shrink(pool->zpool, 1, NULL))
529                 zswap_reject_reclaim_fail++;
530         zswap_pool_put(pool);
531 }
532
533 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
534 {
535         struct zswap_pool *pool;
536         char name[38]; /* 'zswap' + 32 char (max) num + \0 */
537         gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
538         int ret;
539
540         if (!zswap_has_pool) {
541                 /* if either are unset, pool initialization failed, and we
542                  * need both params to be set correctly before trying to
543                  * create a pool.
544                  */
545                 if (!strcmp(type, ZSWAP_PARAM_UNSET))
546                         return NULL;
547                 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
548                         return NULL;
549         }
550
551         pool = kzalloc(sizeof(*pool), GFP_KERNEL);
552         if (!pool)
553                 return NULL;
554
555         /* unique name for each pool specifically required by zsmalloc */
556         snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
557
558         pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
559         if (!pool->zpool) {
560                 pr_err("%s zpool not available\n", type);
561                 goto error;
562         }
563         pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
564
565         strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
566         pool->tfm = alloc_percpu(struct crypto_comp *);
567         if (!pool->tfm) {
568                 pr_err("percpu alloc failed\n");
569                 goto error;
570         }
571
572         ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
573                                        &pool->node);
574         if (ret)
575                 goto error;
576         pr_debug("using %s compressor\n", pool->tfm_name);
577
578         /* being the current pool takes 1 ref; this func expects the
579          * caller to always add the new pool as the current pool
580          */
581         kref_init(&pool->kref);
582         INIT_LIST_HEAD(&pool->list);
583         INIT_WORK(&pool->shrink_work, shrink_worker);
584
585         zswap_pool_debug("created", pool);
586
587         return pool;
588
589 error:
590         free_percpu(pool->tfm);
591         if (pool->zpool)
592                 zpool_destroy_pool(pool->zpool);
593         kfree(pool);
594         return NULL;
595 }
596
597 static __init struct zswap_pool *__zswap_pool_create_fallback(void)
598 {
599         bool has_comp, has_zpool;
600
601         has_comp = crypto_has_comp(zswap_compressor, 0, 0);
602         if (!has_comp && strcmp(zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT)) {
603                 pr_err("compressor %s not available, using default %s\n",
604                        zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT);
605                 param_free_charp(&zswap_compressor);
606                 zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
607                 has_comp = crypto_has_comp(zswap_compressor, 0, 0);
608         }
609         if (!has_comp) {
610                 pr_err("default compressor %s not available\n",
611                        zswap_compressor);
612                 param_free_charp(&zswap_compressor);
613                 zswap_compressor = ZSWAP_PARAM_UNSET;
614         }
615
616         has_zpool = zpool_has_pool(zswap_zpool_type);
617         if (!has_zpool && strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) {
618                 pr_err("zpool %s not available, using default %s\n",
619                        zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT);
620                 param_free_charp(&zswap_zpool_type);
621                 zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
622                 has_zpool = zpool_has_pool(zswap_zpool_type);
623         }
624         if (!has_zpool) {
625                 pr_err("default zpool %s not available\n",
626                        zswap_zpool_type);
627                 param_free_charp(&zswap_zpool_type);
628                 zswap_zpool_type = ZSWAP_PARAM_UNSET;
629         }
630
631         if (!has_comp || !has_zpool)
632                 return NULL;
633
634         return zswap_pool_create(zswap_zpool_type, zswap_compressor);
635 }
636
637 static void zswap_pool_destroy(struct zswap_pool *pool)
638 {
639         zswap_pool_debug("destroying", pool);
640
641         cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
642         free_percpu(pool->tfm);
643         zpool_destroy_pool(pool->zpool);
644         kfree(pool);
645 }
646
647 static int __must_check zswap_pool_get(struct zswap_pool *pool)
648 {
649         if (!pool)
650                 return 0;
651
652         return kref_get_unless_zero(&pool->kref);
653 }
654
655 static void __zswap_pool_release(struct work_struct *work)
656 {
657         struct zswap_pool *pool = container_of(work, typeof(*pool),
658                                                 release_work);
659
660         synchronize_rcu();
661
662         /* nobody should have been able to get a kref... */
663         WARN_ON(kref_get_unless_zero(&pool->kref));
664
665         /* pool is now off zswap_pools list and has no references. */
666         zswap_pool_destroy(pool);
667 }
668
669 static void __zswap_pool_empty(struct kref *kref)
670 {
671         struct zswap_pool *pool;
672
673         pool = container_of(kref, typeof(*pool), kref);
674
675         spin_lock(&zswap_pools_lock);
676
677         WARN_ON(pool == zswap_pool_current());
678
679         list_del_rcu(&pool->list);
680
681         INIT_WORK(&pool->release_work, __zswap_pool_release);
682         schedule_work(&pool->release_work);
683
684         spin_unlock(&zswap_pools_lock);
685 }
686
687 static void zswap_pool_put(struct zswap_pool *pool)
688 {
689         kref_put(&pool->kref, __zswap_pool_empty);
690 }
691
692 /*********************************
693 * param callbacks
694 **********************************/
695
696 /* val must be a null-terminated string */
697 static int __zswap_param_set(const char *val, const struct kernel_param *kp,
698                              char *type, char *compressor)
699 {
700         struct zswap_pool *pool, *put_pool = NULL;
701         char *s = strstrip((char *)val);
702         int ret;
703
704         if (zswap_init_failed) {
705                 pr_err("can't set param, initialization failed\n");
706                 return -ENODEV;
707         }
708
709         /* no change required */
710         if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
711                 return 0;
712
713         /* if this is load-time (pre-init) param setting,
714          * don't create a pool; that's done during init.
715          */
716         if (!zswap_init_started)
717                 return param_set_charp(s, kp);
718
719         if (!type) {
720                 if (!zpool_has_pool(s)) {
721                         pr_err("zpool %s not available\n", s);
722                         return -ENOENT;
723                 }
724                 type = s;
725         } else if (!compressor) {
726                 if (!crypto_has_comp(s, 0, 0)) {
727                         pr_err("compressor %s not available\n", s);
728                         return -ENOENT;
729                 }
730                 compressor = s;
731         } else {
732                 WARN_ON(1);
733                 return -EINVAL;
734         }
735
736         spin_lock(&zswap_pools_lock);
737
738         pool = zswap_pool_find_get(type, compressor);
739         if (pool) {
740                 zswap_pool_debug("using existing", pool);
741                 WARN_ON(pool == zswap_pool_current());
742                 list_del_rcu(&pool->list);
743         }
744
745         spin_unlock(&zswap_pools_lock);
746
747         if (!pool)
748                 pool = zswap_pool_create(type, compressor);
749
750         if (pool)
751                 ret = param_set_charp(s, kp);
752         else
753                 ret = -EINVAL;
754
755         spin_lock(&zswap_pools_lock);
756
757         if (!ret) {
758                 put_pool = zswap_pool_current();
759                 list_add_rcu(&pool->list, &zswap_pools);
760                 zswap_has_pool = true;
761         } else if (pool) {
762                 /* add the possibly pre-existing pool to the end of the pools
763                  * list; if it's new (and empty) then it'll be removed and
764                  * destroyed by the put after we drop the lock
765                  */
766                 list_add_tail_rcu(&pool->list, &zswap_pools);
767                 put_pool = pool;
768         }
769
770         spin_unlock(&zswap_pools_lock);
771
772         if (!zswap_has_pool && !pool) {
773                 /* if initial pool creation failed, and this pool creation also
774                  * failed, maybe both compressor and zpool params were bad.
775                  * Allow changing this param, so pool creation will succeed
776                  * when the other param is changed. We already verified this
777                  * param is ok in the zpool_has_pool() or crypto_has_comp()
778                  * checks above.
779                  */
780                 ret = param_set_charp(s, kp);
781         }
782
783         /* drop the ref from either the old current pool,
784          * or the new pool we failed to add
785          */
786         if (put_pool)
787                 zswap_pool_put(put_pool);
788
789         return ret;
790 }
791
792 static int zswap_compressor_param_set(const char *val,
793                                       const struct kernel_param *kp)
794 {
795         return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
796 }
797
798 static int zswap_zpool_param_set(const char *val,
799                                  const struct kernel_param *kp)
800 {
801         return __zswap_param_set(val, kp, NULL, zswap_compressor);
802 }
803
804 static int zswap_enabled_param_set(const char *val,
805                                    const struct kernel_param *kp)
806 {
807         if (zswap_init_failed) {
808                 pr_err("can't enable, initialization failed\n");
809                 return -ENODEV;
810         }
811         if (!zswap_has_pool && zswap_init_started) {
812                 pr_err("can't enable, no pool configured\n");
813                 return -ENODEV;
814         }
815
816         return param_set_bool(val, kp);
817 }
818
819 /*********************************
820 * writeback code
821 **********************************/
822 /* return enum for zswap_get_swap_cache_page */
823 enum zswap_get_swap_ret {
824         ZSWAP_SWAPCACHE_NEW,
825         ZSWAP_SWAPCACHE_EXIST,
826         ZSWAP_SWAPCACHE_FAIL,
827 };
828
829 /*
830  * zswap_get_swap_cache_page
831  *
832  * This is an adaption of read_swap_cache_async()
833  *
834  * This function tries to find a page with the given swap entry
835  * in the swapper_space address space (the swap cache).  If the page
836  * is found, it is returned in retpage.  Otherwise, a page is allocated,
837  * added to the swap cache, and returned in retpage.
838  *
839  * If success, the swap cache page is returned in retpage
840  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
841  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
842  *     the new page is added to swapcache and locked
843  * Returns ZSWAP_SWAPCACHE_FAIL on error
844  */
845 static int zswap_get_swap_cache_page(swp_entry_t entry,
846                                 struct page **retpage)
847 {
848         bool page_was_allocated;
849
850         *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
851                         NULL, 0, &page_was_allocated);
852         if (page_was_allocated)
853                 return ZSWAP_SWAPCACHE_NEW;
854         if (!*retpage)
855                 return ZSWAP_SWAPCACHE_FAIL;
856         return ZSWAP_SWAPCACHE_EXIST;
857 }
858
859 /*
860  * Attempts to free an entry by adding a page to the swap cache,
861  * decompressing the entry data into the page, and issuing a
862  * bio write to write the page back to the swap device.
863  *
864  * This can be thought of as a "resumed writeback" of the page
865  * to the swap device.  We are basically resuming the same swap
866  * writeback path that was intercepted with the frontswap_store()
867  * in the first place.  After the page has been decompressed into
868  * the swap cache, the compressed version stored by zswap can be
869  * freed.
870  */
871 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
872 {
873         struct zswap_header *zhdr;
874         swp_entry_t swpentry;
875         struct zswap_tree *tree;
876         pgoff_t offset;
877         struct zswap_entry *entry;
878         struct page *page;
879         struct crypto_comp *tfm;
880         u8 *src, *dst;
881         unsigned int dlen;
882         int ret;
883         struct writeback_control wbc = {
884                 .sync_mode = WB_SYNC_NONE,
885         };
886
887         /* extract swpentry from data */
888         zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
889         swpentry = zhdr->swpentry; /* here */
890         tree = zswap_trees[swp_type(swpentry)];
891         offset = swp_offset(swpentry);
892
893         /* find and ref zswap entry */
894         spin_lock(&tree->lock);
895         entry = zswap_entry_find_get(&tree->rbroot, offset);
896         if (!entry) {
897                 /* entry was invalidated */
898                 spin_unlock(&tree->lock);
899                 zpool_unmap_handle(pool, handle);
900                 return 0;
901         }
902         spin_unlock(&tree->lock);
903         BUG_ON(offset != entry->offset);
904
905         /* try to allocate swap cache page */
906         switch (zswap_get_swap_cache_page(swpentry, &page)) {
907         case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
908                 ret = -ENOMEM;
909                 goto fail;
910
911         case ZSWAP_SWAPCACHE_EXIST:
912                 /* page is already in the swap cache, ignore for now */
913                 put_page(page);
914                 ret = -EEXIST;
915                 goto fail;
916
917         case ZSWAP_SWAPCACHE_NEW: /* page is locked */
918                 /* decompress */
919                 dlen = PAGE_SIZE;
920                 src = (u8 *)zhdr + sizeof(struct zswap_header);
921                 dst = kmap_atomic(page);
922                 tfm = *get_cpu_ptr(entry->pool->tfm);
923                 ret = crypto_comp_decompress(tfm, src, entry->length,
924                                              dst, &dlen);
925                 put_cpu_ptr(entry->pool->tfm);
926                 kunmap_atomic(dst);
927                 BUG_ON(ret);
928                 BUG_ON(dlen != PAGE_SIZE);
929
930                 /* page is up to date */
931                 SetPageUptodate(page);
932         }
933
934         /* move it to the tail of the inactive list after end_writeback */
935         SetPageReclaim(page);
936
937         /* start writeback */
938         __swap_writepage(page, &wbc, end_swap_bio_write);
939         put_page(page);
940         zswap_written_back_pages++;
941
942         spin_lock(&tree->lock);
943         /* drop local reference */
944         zswap_entry_put(tree, entry);
945
946         /*
947         * There are two possible situations for entry here:
948         * (1) refcount is 1(normal case),  entry is valid and on the tree
949         * (2) refcount is 0, entry is freed and not on the tree
950         *     because invalidate happened during writeback
951         *  search the tree and free the entry if find entry
952         */
953         if (entry == zswap_rb_search(&tree->rbroot, offset))
954                 zswap_entry_put(tree, entry);
955         spin_unlock(&tree->lock);
956
957         goto end;
958
959         /*
960         * if we get here due to ZSWAP_SWAPCACHE_EXIST
961         * a load may happening concurrently
962         * it is safe and okay to not free the entry
963         * if we free the entry in the following put
964         * it it either okay to return !0
965         */
966 fail:
967         spin_lock(&tree->lock);
968         zswap_entry_put(tree, entry);
969         spin_unlock(&tree->lock);
970
971 end:
972         zpool_unmap_handle(pool, handle);
973         return ret;
974 }
975
976 static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
977 {
978         unsigned int pos;
979         unsigned long *page;
980
981         page = (unsigned long *)ptr;
982         for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
983                 if (page[pos] != page[0])
984                         return 0;
985         }
986         *value = page[0];
987         return 1;
988 }
989
990 static void zswap_fill_page(void *ptr, unsigned long value)
991 {
992         unsigned long *page;
993
994         page = (unsigned long *)ptr;
995         memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
996 }
997
998 /*********************************
999 * frontswap hooks
1000 **********************************/
1001 /* attempts to compress and store an single page */
1002 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1003                                 struct page *page)
1004 {
1005         struct zswap_tree *tree = zswap_trees[type];
1006         struct zswap_entry *entry, *dupentry;
1007         struct crypto_comp *tfm;
1008         int ret;
1009         unsigned int hlen, dlen = PAGE_SIZE;
1010         unsigned long handle, value;
1011         char *buf;
1012         u8 *src, *dst;
1013         struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1014         gfp_t gfp;
1015
1016         /* THP isn't supported */
1017         if (PageTransHuge(page)) {
1018                 ret = -EINVAL;
1019                 goto reject;
1020         }
1021
1022         if (!zswap_enabled || !tree) {
1023                 ret = -ENODEV;
1024                 goto reject;
1025         }
1026
1027         /* reclaim space if needed */
1028         if (zswap_is_full()) {
1029                 struct zswap_pool *pool;
1030
1031                 zswap_pool_limit_hit++;
1032                 zswap_pool_reached_full = true;
1033                 pool = zswap_pool_last_get();
1034                 if (pool)
1035                         queue_work(shrink_wq, &pool->shrink_work);
1036                 ret = -ENOMEM;
1037                 goto reject;
1038         }
1039
1040         if (zswap_pool_reached_full) {
1041                if (!zswap_can_accept()) {
1042                         ret = -ENOMEM;
1043                         goto reject;
1044                 } else
1045                         zswap_pool_reached_full = false;
1046         }
1047
1048         /* allocate entry */
1049         entry = zswap_entry_cache_alloc(GFP_KERNEL);
1050         if (!entry) {
1051                 zswap_reject_kmemcache_fail++;
1052                 ret = -ENOMEM;
1053                 goto reject;
1054         }
1055
1056         if (zswap_same_filled_pages_enabled) {
1057                 src = kmap_atomic(page);
1058                 if (zswap_is_page_same_filled(src, &value)) {
1059                         kunmap_atomic(src);
1060                         entry->offset = offset;
1061                         entry->length = 0;
1062                         entry->value = value;
1063                         atomic_inc(&zswap_same_filled_pages);
1064                         goto insert_entry;
1065                 }
1066                 kunmap_atomic(src);
1067         }
1068
1069         /* if entry is successfully added, it keeps the reference */
1070         entry->pool = zswap_pool_current_get();
1071         if (!entry->pool) {
1072                 ret = -EINVAL;
1073                 goto freepage;
1074         }
1075
1076         /* compress */
1077         dst = get_cpu_var(zswap_dstmem);
1078         tfm = *get_cpu_ptr(entry->pool->tfm);
1079         src = kmap_atomic(page);
1080         ret = crypto_comp_compress(tfm, src, PAGE_SIZE, dst, &dlen);
1081         kunmap_atomic(src);
1082         put_cpu_ptr(entry->pool->tfm);
1083         if (ret) {
1084                 ret = -EINVAL;
1085                 goto put_dstmem;
1086         }
1087
1088         /* store */
1089         hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1090         gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1091         if (zpool_malloc_support_movable(entry->pool->zpool))
1092                 gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1093         ret = zpool_malloc(entry->pool->zpool, hlen + dlen, gfp, &handle);
1094         if (ret == -ENOSPC) {
1095                 zswap_reject_compress_poor++;
1096                 goto put_dstmem;
1097         }
1098         if (ret) {
1099                 zswap_reject_alloc_fail++;
1100                 goto put_dstmem;
1101         }
1102         buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW);
1103         memcpy(buf, &zhdr, hlen);
1104         memcpy(buf + hlen, dst, dlen);
1105         zpool_unmap_handle(entry->pool->zpool, handle);
1106         put_cpu_var(zswap_dstmem);
1107
1108         /* populate entry */
1109         entry->offset = offset;
1110         entry->handle = handle;
1111         entry->length = dlen;
1112
1113 insert_entry:
1114         /* map */
1115         spin_lock(&tree->lock);
1116         do {
1117                 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1118                 if (ret == -EEXIST) {
1119                         zswap_duplicate_entry++;
1120                         /* remove from rbtree */
1121                         zswap_rb_erase(&tree->rbroot, dupentry);
1122                         zswap_entry_put(tree, dupentry);
1123                 }
1124         } while (ret == -EEXIST);
1125         spin_unlock(&tree->lock);
1126
1127         /* update stats */
1128         atomic_inc(&zswap_stored_pages);
1129         zswap_update_total_size();
1130
1131         return 0;
1132
1133 put_dstmem:
1134         put_cpu_var(zswap_dstmem);
1135         zswap_pool_put(entry->pool);
1136 freepage:
1137         zswap_entry_cache_free(entry);
1138 reject:
1139         return ret;
1140 }
1141
1142 /*
1143  * returns 0 if the page was successfully decompressed
1144  * return -1 on entry not found or error
1145 */
1146 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1147                                 struct page *page)
1148 {
1149         struct zswap_tree *tree = zswap_trees[type];
1150         struct zswap_entry *entry;
1151         struct crypto_comp *tfm;
1152         u8 *src, *dst;
1153         unsigned int dlen;
1154         int ret;
1155
1156         /* find */
1157         spin_lock(&tree->lock);
1158         entry = zswap_entry_find_get(&tree->rbroot, offset);
1159         if (!entry) {
1160                 /* entry was written back */
1161                 spin_unlock(&tree->lock);
1162                 return -1;
1163         }
1164         spin_unlock(&tree->lock);
1165
1166         if (!entry->length) {
1167                 dst = kmap_atomic(page);
1168                 zswap_fill_page(dst, entry->value);
1169                 kunmap_atomic(dst);
1170                 goto freeentry;
1171         }
1172
1173         /* decompress */
1174         dlen = PAGE_SIZE;
1175         src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1176         if (zpool_evictable(entry->pool->zpool))
1177                 src += sizeof(struct zswap_header);
1178         dst = kmap_atomic(page);
1179         tfm = *get_cpu_ptr(entry->pool->tfm);
1180         ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen);
1181         put_cpu_ptr(entry->pool->tfm);
1182         kunmap_atomic(dst);
1183         zpool_unmap_handle(entry->pool->zpool, entry->handle);
1184         BUG_ON(ret);
1185
1186 freeentry:
1187         spin_lock(&tree->lock);
1188         zswap_entry_put(tree, entry);
1189         spin_unlock(&tree->lock);
1190
1191         return 0;
1192 }
1193
1194 /* frees an entry in zswap */
1195 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1196 {
1197         struct zswap_tree *tree = zswap_trees[type];
1198         struct zswap_entry *entry;
1199
1200         /* find */
1201         spin_lock(&tree->lock);
1202         entry = zswap_rb_search(&tree->rbroot, offset);
1203         if (!entry) {
1204                 /* entry was written back */
1205                 spin_unlock(&tree->lock);
1206                 return;
1207         }
1208
1209         /* remove from rbtree */
1210         zswap_rb_erase(&tree->rbroot, entry);
1211
1212         /* drop the initial reference from entry creation */
1213         zswap_entry_put(tree, entry);
1214
1215         spin_unlock(&tree->lock);
1216 }
1217
1218 /* frees all zswap entries for the given swap type */
1219 static void zswap_frontswap_invalidate_area(unsigned type)
1220 {
1221         struct zswap_tree *tree = zswap_trees[type];
1222         struct zswap_entry *entry, *n;
1223
1224         if (!tree)
1225                 return;
1226
1227         /* walk the tree and free everything */
1228         spin_lock(&tree->lock);
1229         rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1230                 zswap_free_entry(entry);
1231         tree->rbroot = RB_ROOT;
1232         spin_unlock(&tree->lock);
1233         kfree(tree);
1234         zswap_trees[type] = NULL;
1235 }
1236
1237 static void zswap_frontswap_init(unsigned type)
1238 {
1239         struct zswap_tree *tree;
1240
1241         tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1242         if (!tree) {
1243                 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1244                 return;
1245         }
1246
1247         tree->rbroot = RB_ROOT;
1248         spin_lock_init(&tree->lock);
1249         zswap_trees[type] = tree;
1250 }
1251
1252 static struct frontswap_ops zswap_frontswap_ops = {
1253         .store = zswap_frontswap_store,
1254         .load = zswap_frontswap_load,
1255         .invalidate_page = zswap_frontswap_invalidate_page,
1256         .invalidate_area = zswap_frontswap_invalidate_area,
1257         .init = zswap_frontswap_init
1258 };
1259
1260 /*********************************
1261 * debugfs functions
1262 **********************************/
1263 #ifdef CONFIG_DEBUG_FS
1264 #include <linux/debugfs.h>
1265
1266 static struct dentry *zswap_debugfs_root;
1267
1268 static int __init zswap_debugfs_init(void)
1269 {
1270         if (!debugfs_initialized())
1271                 return -ENODEV;
1272
1273         zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1274
1275         debugfs_create_u64("pool_limit_hit", 0444,
1276                            zswap_debugfs_root, &zswap_pool_limit_hit);
1277         debugfs_create_u64("reject_reclaim_fail", 0444,
1278                            zswap_debugfs_root, &zswap_reject_reclaim_fail);
1279         debugfs_create_u64("reject_alloc_fail", 0444,
1280                            zswap_debugfs_root, &zswap_reject_alloc_fail);
1281         debugfs_create_u64("reject_kmemcache_fail", 0444,
1282                            zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1283         debugfs_create_u64("reject_compress_poor", 0444,
1284                            zswap_debugfs_root, &zswap_reject_compress_poor);
1285         debugfs_create_u64("written_back_pages", 0444,
1286                            zswap_debugfs_root, &zswap_written_back_pages);
1287         debugfs_create_u64("duplicate_entry", 0444,
1288                            zswap_debugfs_root, &zswap_duplicate_entry);
1289         debugfs_create_u64("pool_total_size", 0444,
1290                            zswap_debugfs_root, &zswap_pool_total_size);
1291         debugfs_create_atomic_t("stored_pages", 0444,
1292                                 zswap_debugfs_root, &zswap_stored_pages);
1293         debugfs_create_atomic_t("same_filled_pages", 0444,
1294                                 zswap_debugfs_root, &zswap_same_filled_pages);
1295
1296         return 0;
1297 }
1298
1299 static void __exit zswap_debugfs_exit(void)
1300 {
1301         debugfs_remove_recursive(zswap_debugfs_root);
1302 }
1303 #else
1304 static int __init zswap_debugfs_init(void)
1305 {
1306         return 0;
1307 }
1308
1309 static void __exit zswap_debugfs_exit(void) { }
1310 #endif
1311
1312 /*********************************
1313 * module init and exit
1314 **********************************/
1315 static int __init init_zswap(void)
1316 {
1317         struct zswap_pool *pool;
1318         int ret;
1319
1320         zswap_init_started = true;
1321
1322         if (zswap_entry_cache_create()) {
1323                 pr_err("entry cache creation failed\n");
1324                 goto cache_fail;
1325         }
1326
1327         ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1328                                 zswap_dstmem_prepare, zswap_dstmem_dead);
1329         if (ret) {
1330                 pr_err("dstmem alloc failed\n");
1331                 goto dstmem_fail;
1332         }
1333
1334         ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1335                                       "mm/zswap_pool:prepare",
1336                                       zswap_cpu_comp_prepare,
1337                                       zswap_cpu_comp_dead);
1338         if (ret)
1339                 goto hp_fail;
1340
1341         pool = __zswap_pool_create_fallback();
1342         if (pool) {
1343                 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1344                         zpool_get_type(pool->zpool));
1345                 list_add(&pool->list, &zswap_pools);
1346                 zswap_has_pool = true;
1347         } else {
1348                 pr_err("pool creation failed\n");
1349                 zswap_enabled = false;
1350         }
1351
1352         shrink_wq = create_workqueue("zswap-shrink");
1353         if (!shrink_wq)
1354                 goto fallback_fail;
1355
1356         frontswap_register_ops(&zswap_frontswap_ops);
1357         if (zswap_debugfs_init())
1358                 pr_warn("debugfs initialization failed\n");
1359         return 0;
1360
1361 fallback_fail:
1362         zswap_pool_destroy(pool);
1363 hp_fail:
1364         cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1365 dstmem_fail:
1366         zswap_entry_cache_destroy();
1367 cache_fail:
1368         /* if built-in, we aren't unloaded on failure; don't allow use */
1369         zswap_init_failed = true;
1370         zswap_enabled = false;
1371         return -ENOMEM;
1372 }
1373 /* must be late so crypto has time to come up */
1374 late_initcall(init_zswap);
1375
1376 MODULE_LICENSE("GPL");
1377 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1378 MODULE_DESCRIPTION("Compressed cache for swap pages");