mm/sparse.c: fix typo in online_mem_sections
[linux-2.6-microblaze.git] / mm / zswap.c
1 /*
2  * zswap.c - zswap driver file
3  *
4  * zswap is a backend for frontswap that takes pages that are in the process
5  * of being swapped out and attempts to compress and store them in a
6  * RAM-based memory pool.  This can result in a significant I/O reduction on
7  * the swap device and, in the case where decompressing from RAM is faster
8  * than reading from the swap device, can also improve workload performance.
9  *
10  * Copyright (C) 2012  Seth Jennings <sjenning@linux.vnet.ibm.com>
11  *
12  * This program is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU General Public License
14  * as published by the Free Software Foundation; either version 2
15  * of the License, or (at your option) any later version.
16  *
17  * This program is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20  * GNU General Public License for more details.
21 */
22
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24
25 #include <linux/module.h>
26 #include <linux/cpu.h>
27 #include <linux/highmem.h>
28 #include <linux/slab.h>
29 #include <linux/spinlock.h>
30 #include <linux/types.h>
31 #include <linux/atomic.h>
32 #include <linux/frontswap.h>
33 #include <linux/rbtree.h>
34 #include <linux/swap.h>
35 #include <linux/crypto.h>
36 #include <linux/mempool.h>
37 #include <linux/zpool.h>
38
39 #include <linux/mm_types.h>
40 #include <linux/page-flags.h>
41 #include <linux/swapops.h>
42 #include <linux/writeback.h>
43 #include <linux/pagemap.h>
44
45 /*********************************
46 * statistics
47 **********************************/
48 /* Total bytes used by the compressed storage */
49 static u64 zswap_pool_total_size;
50 /* The number of compressed pages currently stored in zswap */
51 static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
52
53 /*
54  * The statistics below are not protected from concurrent access for
55  * performance reasons so they may not be a 100% accurate.  However,
56  * they do provide useful information on roughly how many times a
57  * certain event is occurring.
58 */
59
60 /* Pool limit was hit (see zswap_max_pool_percent) */
61 static u64 zswap_pool_limit_hit;
62 /* Pages written back when pool limit was reached */
63 static u64 zswap_written_back_pages;
64 /* Store failed due to a reclaim failure after pool limit was reached */
65 static u64 zswap_reject_reclaim_fail;
66 /* Compressed page was too big for the allocator to (optimally) store */
67 static u64 zswap_reject_compress_poor;
68 /* Store failed because underlying allocator could not get memory */
69 static u64 zswap_reject_alloc_fail;
70 /* Store failed because the entry metadata could not be allocated (rare) */
71 static u64 zswap_reject_kmemcache_fail;
72 /* Duplicate store was encountered (rare) */
73 static u64 zswap_duplicate_entry;
74
75 /*********************************
76 * tunables
77 **********************************/
78
79 #define ZSWAP_PARAM_UNSET ""
80
81 /* Enable/disable zswap (disabled by default) */
82 static bool zswap_enabled;
83 static int zswap_enabled_param_set(const char *,
84                                    const struct kernel_param *);
85 static struct kernel_param_ops zswap_enabled_param_ops = {
86         .set =          zswap_enabled_param_set,
87         .get =          param_get_bool,
88 };
89 module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
90
91 /* Crypto compressor to use */
92 #define ZSWAP_COMPRESSOR_DEFAULT "lzo"
93 static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
94 static int zswap_compressor_param_set(const char *,
95                                       const struct kernel_param *);
96 static 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 };
101 module_param_cb(compressor, &zswap_compressor_param_ops,
102                 &zswap_compressor, 0644);
103
104 /* Compressed storage zpool to use */
105 #define ZSWAP_ZPOOL_DEFAULT "zbud"
106 static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
107 static int zswap_zpool_param_set(const char *, const struct kernel_param *);
108 static struct kernel_param_ops zswap_zpool_param_ops = {
109         .set =          zswap_zpool_param_set,
110         .get =          param_get_charp,
111         .free =         param_free_charp,
112 };
113 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
114
115 /* The maximum percentage of memory that the compressed pool can occupy */
116 static unsigned int zswap_max_pool_percent = 20;
117 module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
118
119 /*********************************
120 * data structures
121 **********************************/
122
123 struct zswap_pool {
124         struct zpool *zpool;
125         struct crypto_comp * __percpu *tfm;
126         struct kref kref;
127         struct list_head list;
128         struct work_struct work;
129         struct hlist_node node;
130         char tfm_name[CRYPTO_MAX_ALG_NAME];
131 };
132
133 /*
134  * struct zswap_entry
135  *
136  * This structure contains the metadata for tracking a single compressed
137  * page within zswap.
138  *
139  * rbnode - links the entry into red-black tree for the appropriate swap type
140  * offset - the swap offset for the entry.  Index into the red-black tree.
141  * refcount - the number of outstanding reference to the entry. This is needed
142  *            to protect against premature freeing of the entry by code
143  *            concurrent calls to load, invalidate, and writeback.  The lock
144  *            for the zswap_tree structure that contains the entry must
145  *            be held while changing the refcount.  Since the lock must
146  *            be held, there is no reason to also make refcount atomic.
147  * length - the length in bytes of the compressed page data.  Needed during
148  *          decompression
149  * pool - the zswap_pool the entry's data is in
150  * handle - zpool allocation handle that stores the compressed page data
151  */
152 struct zswap_entry {
153         struct rb_node rbnode;
154         pgoff_t offset;
155         int refcount;
156         unsigned int length;
157         struct zswap_pool *pool;
158         unsigned long handle;
159 };
160
161 struct zswap_header {
162         swp_entry_t swpentry;
163 };
164
165 /*
166  * The tree lock in the zswap_tree struct protects a few things:
167  * - the rbtree
168  * - the refcount field of each entry in the tree
169  */
170 struct zswap_tree {
171         struct rb_root rbroot;
172         spinlock_t lock;
173 };
174
175 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
176
177 /* RCU-protected iteration */
178 static LIST_HEAD(zswap_pools);
179 /* protects zswap_pools list modification */
180 static DEFINE_SPINLOCK(zswap_pools_lock);
181 /* pool counter to provide unique names to zpool */
182 static atomic_t zswap_pools_count = ATOMIC_INIT(0);
183
184 /* used by param callback function */
185 static bool zswap_init_started;
186
187 /* fatal error during init */
188 static bool zswap_init_failed;
189
190 /* init completed, but couldn't create the initial pool */
191 static bool zswap_has_pool;
192
193 /*********************************
194 * helpers and fwd declarations
195 **********************************/
196
197 #define zswap_pool_debug(msg, p)                                \
198         pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name,         \
199                  zpool_get_type((p)->zpool))
200
201 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
202 static int zswap_pool_get(struct zswap_pool *pool);
203 static void zswap_pool_put(struct zswap_pool *pool);
204
205 static const struct zpool_ops zswap_zpool_ops = {
206         .evict = zswap_writeback_entry
207 };
208
209 static bool zswap_is_full(void)
210 {
211         return totalram_pages * zswap_max_pool_percent / 100 <
212                 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
213 }
214
215 static void zswap_update_total_size(void)
216 {
217         struct zswap_pool *pool;
218         u64 total = 0;
219
220         rcu_read_lock();
221
222         list_for_each_entry_rcu(pool, &zswap_pools, list)
223                 total += zpool_get_total_size(pool->zpool);
224
225         rcu_read_unlock();
226
227         zswap_pool_total_size = total;
228 }
229
230 /*********************************
231 * zswap entry functions
232 **********************************/
233 static struct kmem_cache *zswap_entry_cache;
234
235 static int __init zswap_entry_cache_create(void)
236 {
237         zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
238         return zswap_entry_cache == NULL;
239 }
240
241 static void __init zswap_entry_cache_destroy(void)
242 {
243         kmem_cache_destroy(zswap_entry_cache);
244 }
245
246 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
247 {
248         struct zswap_entry *entry;
249         entry = kmem_cache_alloc(zswap_entry_cache, gfp);
250         if (!entry)
251                 return NULL;
252         entry->refcount = 1;
253         RB_CLEAR_NODE(&entry->rbnode);
254         return entry;
255 }
256
257 static void zswap_entry_cache_free(struct zswap_entry *entry)
258 {
259         kmem_cache_free(zswap_entry_cache, entry);
260 }
261
262 /*********************************
263 * rbtree functions
264 **********************************/
265 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
266 {
267         struct rb_node *node = root->rb_node;
268         struct zswap_entry *entry;
269
270         while (node) {
271                 entry = rb_entry(node, struct zswap_entry, rbnode);
272                 if (entry->offset > offset)
273                         node = node->rb_left;
274                 else if (entry->offset < offset)
275                         node = node->rb_right;
276                 else
277                         return entry;
278         }
279         return NULL;
280 }
281
282 /*
283  * In the case that a entry with the same offset is found, a pointer to
284  * the existing entry is stored in dupentry and the function returns -EEXIST
285  */
286 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
287                         struct zswap_entry **dupentry)
288 {
289         struct rb_node **link = &root->rb_node, *parent = NULL;
290         struct zswap_entry *myentry;
291
292         while (*link) {
293                 parent = *link;
294                 myentry = rb_entry(parent, struct zswap_entry, rbnode);
295                 if (myentry->offset > entry->offset)
296                         link = &(*link)->rb_left;
297                 else if (myentry->offset < entry->offset)
298                         link = &(*link)->rb_right;
299                 else {
300                         *dupentry = myentry;
301                         return -EEXIST;
302                 }
303         }
304         rb_link_node(&entry->rbnode, parent, link);
305         rb_insert_color(&entry->rbnode, root);
306         return 0;
307 }
308
309 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
310 {
311         if (!RB_EMPTY_NODE(&entry->rbnode)) {
312                 rb_erase(&entry->rbnode, root);
313                 RB_CLEAR_NODE(&entry->rbnode);
314         }
315 }
316
317 /*
318  * Carries out the common pattern of freeing and entry's zpool allocation,
319  * freeing the entry itself, and decrementing the number of stored pages.
320  */
321 static void zswap_free_entry(struct zswap_entry *entry)
322 {
323         zpool_free(entry->pool->zpool, entry->handle);
324         zswap_pool_put(entry->pool);
325         zswap_entry_cache_free(entry);
326         atomic_dec(&zswap_stored_pages);
327         zswap_update_total_size();
328 }
329
330 /* caller must hold the tree lock */
331 static void zswap_entry_get(struct zswap_entry *entry)
332 {
333         entry->refcount++;
334 }
335
336 /* caller must hold the tree lock
337 * remove from the tree and free it, if nobody reference the entry
338 */
339 static void zswap_entry_put(struct zswap_tree *tree,
340                         struct zswap_entry *entry)
341 {
342         int refcount = --entry->refcount;
343
344         BUG_ON(refcount < 0);
345         if (refcount == 0) {
346                 zswap_rb_erase(&tree->rbroot, entry);
347                 zswap_free_entry(entry);
348         }
349 }
350
351 /* caller must hold the tree lock */
352 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
353                                 pgoff_t offset)
354 {
355         struct zswap_entry *entry;
356
357         entry = zswap_rb_search(root, offset);
358         if (entry)
359                 zswap_entry_get(entry);
360
361         return entry;
362 }
363
364 /*********************************
365 * per-cpu code
366 **********************************/
367 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
368
369 static int zswap_dstmem_prepare(unsigned int cpu)
370 {
371         u8 *dst;
372
373         dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
374         if (!dst)
375                 return -ENOMEM;
376
377         per_cpu(zswap_dstmem, cpu) = dst;
378         return 0;
379 }
380
381 static int zswap_dstmem_dead(unsigned int cpu)
382 {
383         u8 *dst;
384
385         dst = per_cpu(zswap_dstmem, cpu);
386         kfree(dst);
387         per_cpu(zswap_dstmem, cpu) = NULL;
388
389         return 0;
390 }
391
392 static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
393 {
394         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
395         struct crypto_comp *tfm;
396
397         if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu)))
398                 return 0;
399
400         tfm = crypto_alloc_comp(pool->tfm_name, 0, 0);
401         if (IS_ERR_OR_NULL(tfm)) {
402                 pr_err("could not alloc crypto comp %s : %ld\n",
403                        pool->tfm_name, PTR_ERR(tfm));
404                 return -ENOMEM;
405         }
406         *per_cpu_ptr(pool->tfm, cpu) = tfm;
407         return 0;
408 }
409
410 static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
411 {
412         struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
413         struct crypto_comp *tfm;
414
415         tfm = *per_cpu_ptr(pool->tfm, cpu);
416         if (!IS_ERR_OR_NULL(tfm))
417                 crypto_free_comp(tfm);
418         *per_cpu_ptr(pool->tfm, cpu) = NULL;
419         return 0;
420 }
421
422 /*********************************
423 * pool functions
424 **********************************/
425
426 static struct zswap_pool *__zswap_pool_current(void)
427 {
428         struct zswap_pool *pool;
429
430         pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
431         WARN_ONCE(!pool && zswap_has_pool,
432                   "%s: no page storage pool!\n", __func__);
433
434         return pool;
435 }
436
437 static struct zswap_pool *zswap_pool_current(void)
438 {
439         assert_spin_locked(&zswap_pools_lock);
440
441         return __zswap_pool_current();
442 }
443
444 static struct zswap_pool *zswap_pool_current_get(void)
445 {
446         struct zswap_pool *pool;
447
448         rcu_read_lock();
449
450         pool = __zswap_pool_current();
451         if (!zswap_pool_get(pool))
452                 pool = NULL;
453
454         rcu_read_unlock();
455
456         return pool;
457 }
458
459 static struct zswap_pool *zswap_pool_last_get(void)
460 {
461         struct zswap_pool *pool, *last = NULL;
462
463         rcu_read_lock();
464
465         list_for_each_entry_rcu(pool, &zswap_pools, list)
466                 last = pool;
467         WARN_ONCE(!last && zswap_has_pool,
468                   "%s: no page storage pool!\n", __func__);
469         if (!zswap_pool_get(last))
470                 last = NULL;
471
472         rcu_read_unlock();
473
474         return last;
475 }
476
477 /* type and compressor must be null-terminated */
478 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
479 {
480         struct zswap_pool *pool;
481
482         assert_spin_locked(&zswap_pools_lock);
483
484         list_for_each_entry_rcu(pool, &zswap_pools, list) {
485                 if (strcmp(pool->tfm_name, compressor))
486                         continue;
487                 if (strcmp(zpool_get_type(pool->zpool), type))
488                         continue;
489                 /* if we can't get it, it's about to be destroyed */
490                 if (!zswap_pool_get(pool))
491                         continue;
492                 return pool;
493         }
494
495         return NULL;
496 }
497
498 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
499 {
500         struct zswap_pool *pool;
501         char name[38]; /* 'zswap' + 32 char (max) num + \0 */
502         gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
503         int ret;
504
505         if (!zswap_has_pool) {
506                 /* if either are unset, pool initialization failed, and we
507                  * need both params to be set correctly before trying to
508                  * create a pool.
509                  */
510                 if (!strcmp(type, ZSWAP_PARAM_UNSET))
511                         return NULL;
512                 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
513                         return NULL;
514         }
515
516         pool = kzalloc(sizeof(*pool), GFP_KERNEL);
517         if (!pool)
518                 return NULL;
519
520         /* unique name for each pool specifically required by zsmalloc */
521         snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
522
523         pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
524         if (!pool->zpool) {
525                 pr_err("%s zpool not available\n", type);
526                 goto error;
527         }
528         pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
529
530         strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
531         pool->tfm = alloc_percpu(struct crypto_comp *);
532         if (!pool->tfm) {
533                 pr_err("percpu alloc failed\n");
534                 goto error;
535         }
536
537         ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
538                                        &pool->node);
539         if (ret)
540                 goto error;
541         pr_debug("using %s compressor\n", pool->tfm_name);
542
543         /* being the current pool takes 1 ref; this func expects the
544          * caller to always add the new pool as the current pool
545          */
546         kref_init(&pool->kref);
547         INIT_LIST_HEAD(&pool->list);
548
549         zswap_pool_debug("created", pool);
550
551         return pool;
552
553 error:
554         free_percpu(pool->tfm);
555         if (pool->zpool)
556                 zpool_destroy_pool(pool->zpool);
557         kfree(pool);
558         return NULL;
559 }
560
561 static __init struct zswap_pool *__zswap_pool_create_fallback(void)
562 {
563         bool has_comp, has_zpool;
564
565         has_comp = crypto_has_comp(zswap_compressor, 0, 0);
566         if (!has_comp && strcmp(zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT)) {
567                 pr_err("compressor %s not available, using default %s\n",
568                        zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT);
569                 param_free_charp(&zswap_compressor);
570                 zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
571                 has_comp = crypto_has_comp(zswap_compressor, 0, 0);
572         }
573         if (!has_comp) {
574                 pr_err("default compressor %s not available\n",
575                        zswap_compressor);
576                 param_free_charp(&zswap_compressor);
577                 zswap_compressor = ZSWAP_PARAM_UNSET;
578         }
579
580         has_zpool = zpool_has_pool(zswap_zpool_type);
581         if (!has_zpool && strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) {
582                 pr_err("zpool %s not available, using default %s\n",
583                        zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT);
584                 param_free_charp(&zswap_zpool_type);
585                 zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
586                 has_zpool = zpool_has_pool(zswap_zpool_type);
587         }
588         if (!has_zpool) {
589                 pr_err("default zpool %s not available\n",
590                        zswap_zpool_type);
591                 param_free_charp(&zswap_zpool_type);
592                 zswap_zpool_type = ZSWAP_PARAM_UNSET;
593         }
594
595         if (!has_comp || !has_zpool)
596                 return NULL;
597
598         return zswap_pool_create(zswap_zpool_type, zswap_compressor);
599 }
600
601 static void zswap_pool_destroy(struct zswap_pool *pool)
602 {
603         zswap_pool_debug("destroying", pool);
604
605         cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
606         free_percpu(pool->tfm);
607         zpool_destroy_pool(pool->zpool);
608         kfree(pool);
609 }
610
611 static int __must_check zswap_pool_get(struct zswap_pool *pool)
612 {
613         if (!pool)
614                 return 0;
615
616         return kref_get_unless_zero(&pool->kref);
617 }
618
619 static void __zswap_pool_release(struct work_struct *work)
620 {
621         struct zswap_pool *pool = container_of(work, typeof(*pool), work);
622
623         synchronize_rcu();
624
625         /* nobody should have been able to get a kref... */
626         WARN_ON(kref_get_unless_zero(&pool->kref));
627
628         /* pool is now off zswap_pools list and has no references. */
629         zswap_pool_destroy(pool);
630 }
631
632 static void __zswap_pool_empty(struct kref *kref)
633 {
634         struct zswap_pool *pool;
635
636         pool = container_of(kref, typeof(*pool), kref);
637
638         spin_lock(&zswap_pools_lock);
639
640         WARN_ON(pool == zswap_pool_current());
641
642         list_del_rcu(&pool->list);
643
644         INIT_WORK(&pool->work, __zswap_pool_release);
645         schedule_work(&pool->work);
646
647         spin_unlock(&zswap_pools_lock);
648 }
649
650 static void zswap_pool_put(struct zswap_pool *pool)
651 {
652         kref_put(&pool->kref, __zswap_pool_empty);
653 }
654
655 /*********************************
656 * param callbacks
657 **********************************/
658
659 /* val must be a null-terminated string */
660 static int __zswap_param_set(const char *val, const struct kernel_param *kp,
661                              char *type, char *compressor)
662 {
663         struct zswap_pool *pool, *put_pool = NULL;
664         char *s = strstrip((char *)val);
665         int ret;
666
667         if (zswap_init_failed) {
668                 pr_err("can't set param, initialization failed\n");
669                 return -ENODEV;
670         }
671
672         /* no change required */
673         if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
674                 return 0;
675
676         /* if this is load-time (pre-init) param setting,
677          * don't create a pool; that's done during init.
678          */
679         if (!zswap_init_started)
680                 return param_set_charp(s, kp);
681
682         if (!type) {
683                 if (!zpool_has_pool(s)) {
684                         pr_err("zpool %s not available\n", s);
685                         return -ENOENT;
686                 }
687                 type = s;
688         } else if (!compressor) {
689                 if (!crypto_has_comp(s, 0, 0)) {
690                         pr_err("compressor %s not available\n", s);
691                         return -ENOENT;
692                 }
693                 compressor = s;
694         } else {
695                 WARN_ON(1);
696                 return -EINVAL;
697         }
698
699         spin_lock(&zswap_pools_lock);
700
701         pool = zswap_pool_find_get(type, compressor);
702         if (pool) {
703                 zswap_pool_debug("using existing", pool);
704                 WARN_ON(pool == zswap_pool_current());
705                 list_del_rcu(&pool->list);
706         }
707
708         spin_unlock(&zswap_pools_lock);
709
710         if (!pool)
711                 pool = zswap_pool_create(type, compressor);
712
713         if (pool)
714                 ret = param_set_charp(s, kp);
715         else
716                 ret = -EINVAL;
717
718         spin_lock(&zswap_pools_lock);
719
720         if (!ret) {
721                 put_pool = zswap_pool_current();
722                 list_add_rcu(&pool->list, &zswap_pools);
723                 zswap_has_pool = true;
724         } else if (pool) {
725                 /* add the possibly pre-existing pool to the end of the pools
726                  * list; if it's new (and empty) then it'll be removed and
727                  * destroyed by the put after we drop the lock
728                  */
729                 list_add_tail_rcu(&pool->list, &zswap_pools);
730                 put_pool = pool;
731         }
732
733         spin_unlock(&zswap_pools_lock);
734
735         if (!zswap_has_pool && !pool) {
736                 /* if initial pool creation failed, and this pool creation also
737                  * failed, maybe both compressor and zpool params were bad.
738                  * Allow changing this param, so pool creation will succeed
739                  * when the other param is changed. We already verified this
740                  * param is ok in the zpool_has_pool() or crypto_has_comp()
741                  * checks above.
742                  */
743                 ret = param_set_charp(s, kp);
744         }
745
746         /* drop the ref from either the old current pool,
747          * or the new pool we failed to add
748          */
749         if (put_pool)
750                 zswap_pool_put(put_pool);
751
752         return ret;
753 }
754
755 static int zswap_compressor_param_set(const char *val,
756                                       const struct kernel_param *kp)
757 {
758         return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
759 }
760
761 static int zswap_zpool_param_set(const char *val,
762                                  const struct kernel_param *kp)
763 {
764         return __zswap_param_set(val, kp, NULL, zswap_compressor);
765 }
766
767 static int zswap_enabled_param_set(const char *val,
768                                    const struct kernel_param *kp)
769 {
770         if (zswap_init_failed) {
771                 pr_err("can't enable, initialization failed\n");
772                 return -ENODEV;
773         }
774         if (!zswap_has_pool && zswap_init_started) {
775                 pr_err("can't enable, no pool configured\n");
776                 return -ENODEV;
777         }
778
779         return param_set_bool(val, kp);
780 }
781
782 /*********************************
783 * writeback code
784 **********************************/
785 /* return enum for zswap_get_swap_cache_page */
786 enum zswap_get_swap_ret {
787         ZSWAP_SWAPCACHE_NEW,
788         ZSWAP_SWAPCACHE_EXIST,
789         ZSWAP_SWAPCACHE_FAIL,
790 };
791
792 /*
793  * zswap_get_swap_cache_page
794  *
795  * This is an adaption of read_swap_cache_async()
796  *
797  * This function tries to find a page with the given swap entry
798  * in the swapper_space address space (the swap cache).  If the page
799  * is found, it is returned in retpage.  Otherwise, a page is allocated,
800  * added to the swap cache, and returned in retpage.
801  *
802  * If success, the swap cache page is returned in retpage
803  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
804  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
805  *     the new page is added to swapcache and locked
806  * Returns ZSWAP_SWAPCACHE_FAIL on error
807  */
808 static int zswap_get_swap_cache_page(swp_entry_t entry,
809                                 struct page **retpage)
810 {
811         bool page_was_allocated;
812
813         *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
814                         NULL, 0, &page_was_allocated);
815         if (page_was_allocated)
816                 return ZSWAP_SWAPCACHE_NEW;
817         if (!*retpage)
818                 return ZSWAP_SWAPCACHE_FAIL;
819         return ZSWAP_SWAPCACHE_EXIST;
820 }
821
822 /*
823  * Attempts to free an entry by adding a page to the swap cache,
824  * decompressing the entry data into the page, and issuing a
825  * bio write to write the page back to the swap device.
826  *
827  * This can be thought of as a "resumed writeback" of the page
828  * to the swap device.  We are basically resuming the same swap
829  * writeback path that was intercepted with the frontswap_store()
830  * in the first place.  After the page has been decompressed into
831  * the swap cache, the compressed version stored by zswap can be
832  * freed.
833  */
834 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
835 {
836         struct zswap_header *zhdr;
837         swp_entry_t swpentry;
838         struct zswap_tree *tree;
839         pgoff_t offset;
840         struct zswap_entry *entry;
841         struct page *page;
842         struct crypto_comp *tfm;
843         u8 *src, *dst;
844         unsigned int dlen;
845         int ret;
846         struct writeback_control wbc = {
847                 .sync_mode = WB_SYNC_NONE,
848         };
849
850         /* extract swpentry from data */
851         zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
852         swpentry = zhdr->swpentry; /* here */
853         zpool_unmap_handle(pool, handle);
854         tree = zswap_trees[swp_type(swpentry)];
855         offset = swp_offset(swpentry);
856
857         /* find and ref zswap entry */
858         spin_lock(&tree->lock);
859         entry = zswap_entry_find_get(&tree->rbroot, offset);
860         if (!entry) {
861                 /* entry was invalidated */
862                 spin_unlock(&tree->lock);
863                 return 0;
864         }
865         spin_unlock(&tree->lock);
866         BUG_ON(offset != entry->offset);
867
868         /* try to allocate swap cache page */
869         switch (zswap_get_swap_cache_page(swpentry, &page)) {
870         case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
871                 ret = -ENOMEM;
872                 goto fail;
873
874         case ZSWAP_SWAPCACHE_EXIST:
875                 /* page is already in the swap cache, ignore for now */
876                 put_page(page);
877                 ret = -EEXIST;
878                 goto fail;
879
880         case ZSWAP_SWAPCACHE_NEW: /* page is locked */
881                 /* decompress */
882                 dlen = PAGE_SIZE;
883                 src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle,
884                                 ZPOOL_MM_RO) + sizeof(struct zswap_header);
885                 dst = kmap_atomic(page);
886                 tfm = *get_cpu_ptr(entry->pool->tfm);
887                 ret = crypto_comp_decompress(tfm, src, entry->length,
888                                              dst, &dlen);
889                 put_cpu_ptr(entry->pool->tfm);
890                 kunmap_atomic(dst);
891                 zpool_unmap_handle(entry->pool->zpool, entry->handle);
892                 BUG_ON(ret);
893                 BUG_ON(dlen != PAGE_SIZE);
894
895                 /* page is up to date */
896                 SetPageUptodate(page);
897         }
898
899         /* move it to the tail of the inactive list after end_writeback */
900         SetPageReclaim(page);
901
902         /* start writeback */
903         __swap_writepage(page, &wbc, end_swap_bio_write);
904         put_page(page);
905         zswap_written_back_pages++;
906
907         spin_lock(&tree->lock);
908         /* drop local reference */
909         zswap_entry_put(tree, entry);
910
911         /*
912         * There are two possible situations for entry here:
913         * (1) refcount is 1(normal case),  entry is valid and on the tree
914         * (2) refcount is 0, entry is freed and not on the tree
915         *     because invalidate happened during writeback
916         *  search the tree and free the entry if find entry
917         */
918         if (entry == zswap_rb_search(&tree->rbroot, offset))
919                 zswap_entry_put(tree, entry);
920         spin_unlock(&tree->lock);
921
922         goto end;
923
924         /*
925         * if we get here due to ZSWAP_SWAPCACHE_EXIST
926         * a load may happening concurrently
927         * it is safe and okay to not free the entry
928         * if we free the entry in the following put
929         * it it either okay to return !0
930         */
931 fail:
932         spin_lock(&tree->lock);
933         zswap_entry_put(tree, entry);
934         spin_unlock(&tree->lock);
935
936 end:
937         return ret;
938 }
939
940 static int zswap_shrink(void)
941 {
942         struct zswap_pool *pool;
943         int ret;
944
945         pool = zswap_pool_last_get();
946         if (!pool)
947                 return -ENOENT;
948
949         ret = zpool_shrink(pool->zpool, 1, NULL);
950
951         zswap_pool_put(pool);
952
953         return ret;
954 }
955
956 /*********************************
957 * frontswap hooks
958 **********************************/
959 /* attempts to compress and store an single page */
960 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
961                                 struct page *page)
962 {
963         struct zswap_tree *tree = zswap_trees[type];
964         struct zswap_entry *entry, *dupentry;
965         struct crypto_comp *tfm;
966         int ret;
967         unsigned int dlen = PAGE_SIZE, len;
968         unsigned long handle;
969         char *buf;
970         u8 *src, *dst;
971         struct zswap_header *zhdr;
972
973         if (!zswap_enabled || !tree) {
974                 ret = -ENODEV;
975                 goto reject;
976         }
977
978         /* reclaim space if needed */
979         if (zswap_is_full()) {
980                 zswap_pool_limit_hit++;
981                 if (zswap_shrink()) {
982                         zswap_reject_reclaim_fail++;
983                         ret = -ENOMEM;
984                         goto reject;
985                 }
986         }
987
988         /* allocate entry */
989         entry = zswap_entry_cache_alloc(GFP_KERNEL);
990         if (!entry) {
991                 zswap_reject_kmemcache_fail++;
992                 ret = -ENOMEM;
993                 goto reject;
994         }
995
996         /* if entry is successfully added, it keeps the reference */
997         entry->pool = zswap_pool_current_get();
998         if (!entry->pool) {
999                 ret = -EINVAL;
1000                 goto freepage;
1001         }
1002
1003         /* compress */
1004         dst = get_cpu_var(zswap_dstmem);
1005         tfm = *get_cpu_ptr(entry->pool->tfm);
1006         src = kmap_atomic(page);
1007         ret = crypto_comp_compress(tfm, src, PAGE_SIZE, dst, &dlen);
1008         kunmap_atomic(src);
1009         put_cpu_ptr(entry->pool->tfm);
1010         if (ret) {
1011                 ret = -EINVAL;
1012                 goto put_dstmem;
1013         }
1014
1015         /* store */
1016         len = dlen + sizeof(struct zswap_header);
1017         ret = zpool_malloc(entry->pool->zpool, len,
1018                            __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM,
1019                            &handle);
1020         if (ret == -ENOSPC) {
1021                 zswap_reject_compress_poor++;
1022                 goto put_dstmem;
1023         }
1024         if (ret) {
1025                 zswap_reject_alloc_fail++;
1026                 goto put_dstmem;
1027         }
1028         zhdr = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW);
1029         zhdr->swpentry = swp_entry(type, offset);
1030         buf = (u8 *)(zhdr + 1);
1031         memcpy(buf, dst, dlen);
1032         zpool_unmap_handle(entry->pool->zpool, handle);
1033         put_cpu_var(zswap_dstmem);
1034
1035         /* populate entry */
1036         entry->offset = offset;
1037         entry->handle = handle;
1038         entry->length = dlen;
1039
1040         /* map */
1041         spin_lock(&tree->lock);
1042         do {
1043                 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1044                 if (ret == -EEXIST) {
1045                         zswap_duplicate_entry++;
1046                         /* remove from rbtree */
1047                         zswap_rb_erase(&tree->rbroot, dupentry);
1048                         zswap_entry_put(tree, dupentry);
1049                 }
1050         } while (ret == -EEXIST);
1051         spin_unlock(&tree->lock);
1052
1053         /* update stats */
1054         atomic_inc(&zswap_stored_pages);
1055         zswap_update_total_size();
1056
1057         return 0;
1058
1059 put_dstmem:
1060         put_cpu_var(zswap_dstmem);
1061         zswap_pool_put(entry->pool);
1062 freepage:
1063         zswap_entry_cache_free(entry);
1064 reject:
1065         return ret;
1066 }
1067
1068 /*
1069  * returns 0 if the page was successfully decompressed
1070  * return -1 on entry not found or error
1071 */
1072 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1073                                 struct page *page)
1074 {
1075         struct zswap_tree *tree = zswap_trees[type];
1076         struct zswap_entry *entry;
1077         struct crypto_comp *tfm;
1078         u8 *src, *dst;
1079         unsigned int dlen;
1080         int ret;
1081
1082         /* find */
1083         spin_lock(&tree->lock);
1084         entry = zswap_entry_find_get(&tree->rbroot, offset);
1085         if (!entry) {
1086                 /* entry was written back */
1087                 spin_unlock(&tree->lock);
1088                 return -1;
1089         }
1090         spin_unlock(&tree->lock);
1091
1092         /* decompress */
1093         dlen = PAGE_SIZE;
1094         src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle,
1095                         ZPOOL_MM_RO) + sizeof(struct zswap_header);
1096         dst = kmap_atomic(page);
1097         tfm = *get_cpu_ptr(entry->pool->tfm);
1098         ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen);
1099         put_cpu_ptr(entry->pool->tfm);
1100         kunmap_atomic(dst);
1101         zpool_unmap_handle(entry->pool->zpool, entry->handle);
1102         BUG_ON(ret);
1103
1104         spin_lock(&tree->lock);
1105         zswap_entry_put(tree, entry);
1106         spin_unlock(&tree->lock);
1107
1108         return 0;
1109 }
1110
1111 /* frees an entry in zswap */
1112 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1113 {
1114         struct zswap_tree *tree = zswap_trees[type];
1115         struct zswap_entry *entry;
1116
1117         /* find */
1118         spin_lock(&tree->lock);
1119         entry = zswap_rb_search(&tree->rbroot, offset);
1120         if (!entry) {
1121                 /* entry was written back */
1122                 spin_unlock(&tree->lock);
1123                 return;
1124         }
1125
1126         /* remove from rbtree */
1127         zswap_rb_erase(&tree->rbroot, entry);
1128
1129         /* drop the initial reference from entry creation */
1130         zswap_entry_put(tree, entry);
1131
1132         spin_unlock(&tree->lock);
1133 }
1134
1135 /* frees all zswap entries for the given swap type */
1136 static void zswap_frontswap_invalidate_area(unsigned type)
1137 {
1138         struct zswap_tree *tree = zswap_trees[type];
1139         struct zswap_entry *entry, *n;
1140
1141         if (!tree)
1142                 return;
1143
1144         /* walk the tree and free everything */
1145         spin_lock(&tree->lock);
1146         rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1147                 zswap_free_entry(entry);
1148         tree->rbroot = RB_ROOT;
1149         spin_unlock(&tree->lock);
1150         kfree(tree);
1151         zswap_trees[type] = NULL;
1152 }
1153
1154 static void zswap_frontswap_init(unsigned type)
1155 {
1156         struct zswap_tree *tree;
1157
1158         tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1159         if (!tree) {
1160                 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1161                 return;
1162         }
1163
1164         tree->rbroot = RB_ROOT;
1165         spin_lock_init(&tree->lock);
1166         zswap_trees[type] = tree;
1167 }
1168
1169 static struct frontswap_ops zswap_frontswap_ops = {
1170         .store = zswap_frontswap_store,
1171         .load = zswap_frontswap_load,
1172         .invalidate_page = zswap_frontswap_invalidate_page,
1173         .invalidate_area = zswap_frontswap_invalidate_area,
1174         .init = zswap_frontswap_init
1175 };
1176
1177 /*********************************
1178 * debugfs functions
1179 **********************************/
1180 #ifdef CONFIG_DEBUG_FS
1181 #include <linux/debugfs.h>
1182
1183 static struct dentry *zswap_debugfs_root;
1184
1185 static int __init zswap_debugfs_init(void)
1186 {
1187         if (!debugfs_initialized())
1188                 return -ENODEV;
1189
1190         zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1191         if (!zswap_debugfs_root)
1192                 return -ENOMEM;
1193
1194         debugfs_create_u64("pool_limit_hit", S_IRUGO,
1195                         zswap_debugfs_root, &zswap_pool_limit_hit);
1196         debugfs_create_u64("reject_reclaim_fail", S_IRUGO,
1197                         zswap_debugfs_root, &zswap_reject_reclaim_fail);
1198         debugfs_create_u64("reject_alloc_fail", S_IRUGO,
1199                         zswap_debugfs_root, &zswap_reject_alloc_fail);
1200         debugfs_create_u64("reject_kmemcache_fail", S_IRUGO,
1201                         zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1202         debugfs_create_u64("reject_compress_poor", S_IRUGO,
1203                         zswap_debugfs_root, &zswap_reject_compress_poor);
1204         debugfs_create_u64("written_back_pages", S_IRUGO,
1205                         zswap_debugfs_root, &zswap_written_back_pages);
1206         debugfs_create_u64("duplicate_entry", S_IRUGO,
1207                         zswap_debugfs_root, &zswap_duplicate_entry);
1208         debugfs_create_u64("pool_total_size", S_IRUGO,
1209                         zswap_debugfs_root, &zswap_pool_total_size);
1210         debugfs_create_atomic_t("stored_pages", S_IRUGO,
1211                         zswap_debugfs_root, &zswap_stored_pages);
1212
1213         return 0;
1214 }
1215
1216 static void __exit zswap_debugfs_exit(void)
1217 {
1218         debugfs_remove_recursive(zswap_debugfs_root);
1219 }
1220 #else
1221 static int __init zswap_debugfs_init(void)
1222 {
1223         return 0;
1224 }
1225
1226 static void __exit zswap_debugfs_exit(void) { }
1227 #endif
1228
1229 /*********************************
1230 * module init and exit
1231 **********************************/
1232 static int __init init_zswap(void)
1233 {
1234         struct zswap_pool *pool;
1235         int ret;
1236
1237         zswap_init_started = true;
1238
1239         if (zswap_entry_cache_create()) {
1240                 pr_err("entry cache creation failed\n");
1241                 goto cache_fail;
1242         }
1243
1244         ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1245                                 zswap_dstmem_prepare, zswap_dstmem_dead);
1246         if (ret) {
1247                 pr_err("dstmem alloc failed\n");
1248                 goto dstmem_fail;
1249         }
1250
1251         ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1252                                       "mm/zswap_pool:prepare",
1253                                       zswap_cpu_comp_prepare,
1254                                       zswap_cpu_comp_dead);
1255         if (ret)
1256                 goto hp_fail;
1257
1258         pool = __zswap_pool_create_fallback();
1259         if (pool) {
1260                 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1261                         zpool_get_type(pool->zpool));
1262                 list_add(&pool->list, &zswap_pools);
1263                 zswap_has_pool = true;
1264         } else {
1265                 pr_err("pool creation failed\n");
1266                 zswap_enabled = false;
1267         }
1268
1269         frontswap_register_ops(&zswap_frontswap_ops);
1270         if (zswap_debugfs_init())
1271                 pr_warn("debugfs initialization failed\n");
1272         return 0;
1273
1274 hp_fail:
1275         cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1276 dstmem_fail:
1277         zswap_entry_cache_destroy();
1278 cache_fail:
1279         /* if built-in, we aren't unloaded on failure; don't allow use */
1280         zswap_init_failed = true;
1281         zswap_enabled = false;
1282         return -ENOMEM;
1283 }
1284 /* must be late so crypto has time to come up */
1285 late_initcall(init_zswap);
1286
1287 MODULE_LICENSE("GPL");
1288 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1289 MODULE_DESCRIPTION("Compressed cache for swap pages");