Merge tag 'nfs-for-4.8-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
[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 /* Enable/disable zswap (disabled by default) */
80 static bool zswap_enabled;
81 module_param_named(enabled, zswap_enabled, bool, 0644);
82
83 /* Crypto compressor to use */
84 #define ZSWAP_COMPRESSOR_DEFAULT "lzo"
85 static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
86 static int zswap_compressor_param_set(const char *,
87                                       const struct kernel_param *);
88 static struct kernel_param_ops zswap_compressor_param_ops = {
89         .set =          zswap_compressor_param_set,
90         .get =          param_get_charp,
91         .free =         param_free_charp,
92 };
93 module_param_cb(compressor, &zswap_compressor_param_ops,
94                 &zswap_compressor, 0644);
95
96 /* Compressed storage zpool to use */
97 #define ZSWAP_ZPOOL_DEFAULT "zbud"
98 static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
99 static int zswap_zpool_param_set(const char *, const struct kernel_param *);
100 static struct kernel_param_ops zswap_zpool_param_ops = {
101         .set =          zswap_zpool_param_set,
102         .get =          param_get_charp,
103         .free =         param_free_charp,
104 };
105 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
106
107 /* The maximum percentage of memory that the compressed pool can occupy */
108 static unsigned int zswap_max_pool_percent = 20;
109 module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
110
111 /*********************************
112 * data structures
113 **********************************/
114
115 struct zswap_pool {
116         struct zpool *zpool;
117         struct crypto_comp * __percpu *tfm;
118         struct kref kref;
119         struct list_head list;
120         struct work_struct work;
121         struct notifier_block notifier;
122         char tfm_name[CRYPTO_MAX_ALG_NAME];
123 };
124
125 /*
126  * struct zswap_entry
127  *
128  * This structure contains the metadata for tracking a single compressed
129  * page within zswap.
130  *
131  * rbnode - links the entry into red-black tree for the appropriate swap type
132  * offset - the swap offset for the entry.  Index into the red-black tree.
133  * refcount - the number of outstanding reference to the entry. This is needed
134  *            to protect against premature freeing of the entry by code
135  *            concurrent calls to load, invalidate, and writeback.  The lock
136  *            for the zswap_tree structure that contains the entry must
137  *            be held while changing the refcount.  Since the lock must
138  *            be held, there is no reason to also make refcount atomic.
139  * length - the length in bytes of the compressed page data.  Needed during
140  *          decompression
141  * pool - the zswap_pool the entry's data is in
142  * handle - zpool allocation handle that stores the compressed page data
143  */
144 struct zswap_entry {
145         struct rb_node rbnode;
146         pgoff_t offset;
147         int refcount;
148         unsigned int length;
149         struct zswap_pool *pool;
150         unsigned long handle;
151 };
152
153 struct zswap_header {
154         swp_entry_t swpentry;
155 };
156
157 /*
158  * The tree lock in the zswap_tree struct protects a few things:
159  * - the rbtree
160  * - the refcount field of each entry in the tree
161  */
162 struct zswap_tree {
163         struct rb_root rbroot;
164         spinlock_t lock;
165 };
166
167 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
168
169 /* RCU-protected iteration */
170 static LIST_HEAD(zswap_pools);
171 /* protects zswap_pools list modification */
172 static DEFINE_SPINLOCK(zswap_pools_lock);
173 /* pool counter to provide unique names to zpool */
174 static atomic_t zswap_pools_count = ATOMIC_INIT(0);
175
176 /* used by param callback function */
177 static bool zswap_init_started;
178
179 /*********************************
180 * helpers and fwd declarations
181 **********************************/
182
183 #define zswap_pool_debug(msg, p)                                \
184         pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name,         \
185                  zpool_get_type((p)->zpool))
186
187 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
188 static int zswap_pool_get(struct zswap_pool *pool);
189 static void zswap_pool_put(struct zswap_pool *pool);
190
191 static const struct zpool_ops zswap_zpool_ops = {
192         .evict = zswap_writeback_entry
193 };
194
195 static bool zswap_is_full(void)
196 {
197         return totalram_pages * zswap_max_pool_percent / 100 <
198                 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
199 }
200
201 static void zswap_update_total_size(void)
202 {
203         struct zswap_pool *pool;
204         u64 total = 0;
205
206         rcu_read_lock();
207
208         list_for_each_entry_rcu(pool, &zswap_pools, list)
209                 total += zpool_get_total_size(pool->zpool);
210
211         rcu_read_unlock();
212
213         zswap_pool_total_size = total;
214 }
215
216 /*********************************
217 * zswap entry functions
218 **********************************/
219 static struct kmem_cache *zswap_entry_cache;
220
221 static int __init zswap_entry_cache_create(void)
222 {
223         zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
224         return zswap_entry_cache == NULL;
225 }
226
227 static void __init zswap_entry_cache_destroy(void)
228 {
229         kmem_cache_destroy(zswap_entry_cache);
230 }
231
232 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
233 {
234         struct zswap_entry *entry;
235         entry = kmem_cache_alloc(zswap_entry_cache, gfp);
236         if (!entry)
237                 return NULL;
238         entry->refcount = 1;
239         RB_CLEAR_NODE(&entry->rbnode);
240         return entry;
241 }
242
243 static void zswap_entry_cache_free(struct zswap_entry *entry)
244 {
245         kmem_cache_free(zswap_entry_cache, entry);
246 }
247
248 /*********************************
249 * rbtree functions
250 **********************************/
251 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
252 {
253         struct rb_node *node = root->rb_node;
254         struct zswap_entry *entry;
255
256         while (node) {
257                 entry = rb_entry(node, struct zswap_entry, rbnode);
258                 if (entry->offset > offset)
259                         node = node->rb_left;
260                 else if (entry->offset < offset)
261                         node = node->rb_right;
262                 else
263                         return entry;
264         }
265         return NULL;
266 }
267
268 /*
269  * In the case that a entry with the same offset is found, a pointer to
270  * the existing entry is stored in dupentry and the function returns -EEXIST
271  */
272 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
273                         struct zswap_entry **dupentry)
274 {
275         struct rb_node **link = &root->rb_node, *parent = NULL;
276         struct zswap_entry *myentry;
277
278         while (*link) {
279                 parent = *link;
280                 myentry = rb_entry(parent, struct zswap_entry, rbnode);
281                 if (myentry->offset > entry->offset)
282                         link = &(*link)->rb_left;
283                 else if (myentry->offset < entry->offset)
284                         link = &(*link)->rb_right;
285                 else {
286                         *dupentry = myentry;
287                         return -EEXIST;
288                 }
289         }
290         rb_link_node(&entry->rbnode, parent, link);
291         rb_insert_color(&entry->rbnode, root);
292         return 0;
293 }
294
295 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
296 {
297         if (!RB_EMPTY_NODE(&entry->rbnode)) {
298                 rb_erase(&entry->rbnode, root);
299                 RB_CLEAR_NODE(&entry->rbnode);
300         }
301 }
302
303 /*
304  * Carries out the common pattern of freeing and entry's zpool allocation,
305  * freeing the entry itself, and decrementing the number of stored pages.
306  */
307 static void zswap_free_entry(struct zswap_entry *entry)
308 {
309         zpool_free(entry->pool->zpool, entry->handle);
310         zswap_pool_put(entry->pool);
311         zswap_entry_cache_free(entry);
312         atomic_dec(&zswap_stored_pages);
313         zswap_update_total_size();
314 }
315
316 /* caller must hold the tree lock */
317 static void zswap_entry_get(struct zswap_entry *entry)
318 {
319         entry->refcount++;
320 }
321
322 /* caller must hold the tree lock
323 * remove from the tree and free it, if nobody reference the entry
324 */
325 static void zswap_entry_put(struct zswap_tree *tree,
326                         struct zswap_entry *entry)
327 {
328         int refcount = --entry->refcount;
329
330         BUG_ON(refcount < 0);
331         if (refcount == 0) {
332                 zswap_rb_erase(&tree->rbroot, entry);
333                 zswap_free_entry(entry);
334         }
335 }
336
337 /* caller must hold the tree lock */
338 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
339                                 pgoff_t offset)
340 {
341         struct zswap_entry *entry;
342
343         entry = zswap_rb_search(root, offset);
344         if (entry)
345                 zswap_entry_get(entry);
346
347         return entry;
348 }
349
350 /*********************************
351 * per-cpu code
352 **********************************/
353 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
354
355 static int __zswap_cpu_dstmem_notifier(unsigned long action, unsigned long cpu)
356 {
357         u8 *dst;
358
359         switch (action) {
360         case CPU_UP_PREPARE:
361                 dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
362                 if (!dst) {
363                         pr_err("can't allocate compressor buffer\n");
364                         return NOTIFY_BAD;
365                 }
366                 per_cpu(zswap_dstmem, cpu) = dst;
367                 break;
368         case CPU_DEAD:
369         case CPU_UP_CANCELED:
370                 dst = per_cpu(zswap_dstmem, cpu);
371                 kfree(dst);
372                 per_cpu(zswap_dstmem, cpu) = NULL;
373                 break;
374         default:
375                 break;
376         }
377         return NOTIFY_OK;
378 }
379
380 static int zswap_cpu_dstmem_notifier(struct notifier_block *nb,
381                                      unsigned long action, void *pcpu)
382 {
383         return __zswap_cpu_dstmem_notifier(action, (unsigned long)pcpu);
384 }
385
386 static struct notifier_block zswap_dstmem_notifier = {
387         .notifier_call =        zswap_cpu_dstmem_notifier,
388 };
389
390 static int __init zswap_cpu_dstmem_init(void)
391 {
392         unsigned long cpu;
393
394         cpu_notifier_register_begin();
395         for_each_online_cpu(cpu)
396                 if (__zswap_cpu_dstmem_notifier(CPU_UP_PREPARE, cpu) ==
397                     NOTIFY_BAD)
398                         goto cleanup;
399         __register_cpu_notifier(&zswap_dstmem_notifier);
400         cpu_notifier_register_done();
401         return 0;
402
403 cleanup:
404         for_each_online_cpu(cpu)
405                 __zswap_cpu_dstmem_notifier(CPU_UP_CANCELED, cpu);
406         cpu_notifier_register_done();
407         return -ENOMEM;
408 }
409
410 static void zswap_cpu_dstmem_destroy(void)
411 {
412         unsigned long cpu;
413
414         cpu_notifier_register_begin();
415         for_each_online_cpu(cpu)
416                 __zswap_cpu_dstmem_notifier(CPU_UP_CANCELED, cpu);
417         __unregister_cpu_notifier(&zswap_dstmem_notifier);
418         cpu_notifier_register_done();
419 }
420
421 static int __zswap_cpu_comp_notifier(struct zswap_pool *pool,
422                                      unsigned long action, unsigned long cpu)
423 {
424         struct crypto_comp *tfm;
425
426         switch (action) {
427         case CPU_UP_PREPARE:
428                 if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu)))
429                         break;
430                 tfm = crypto_alloc_comp(pool->tfm_name, 0, 0);
431                 if (IS_ERR_OR_NULL(tfm)) {
432                         pr_err("could not alloc crypto comp %s : %ld\n",
433                                pool->tfm_name, PTR_ERR(tfm));
434                         return NOTIFY_BAD;
435                 }
436                 *per_cpu_ptr(pool->tfm, cpu) = tfm;
437                 break;
438         case CPU_DEAD:
439         case CPU_UP_CANCELED:
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                 break;
445         default:
446                 break;
447         }
448         return NOTIFY_OK;
449 }
450
451 static int zswap_cpu_comp_notifier(struct notifier_block *nb,
452                                    unsigned long action, void *pcpu)
453 {
454         unsigned long cpu = (unsigned long)pcpu;
455         struct zswap_pool *pool = container_of(nb, typeof(*pool), notifier);
456
457         return __zswap_cpu_comp_notifier(pool, action, cpu);
458 }
459
460 static int zswap_cpu_comp_init(struct zswap_pool *pool)
461 {
462         unsigned long cpu;
463
464         memset(&pool->notifier, 0, sizeof(pool->notifier));
465         pool->notifier.notifier_call = zswap_cpu_comp_notifier;
466
467         cpu_notifier_register_begin();
468         for_each_online_cpu(cpu)
469                 if (__zswap_cpu_comp_notifier(pool, CPU_UP_PREPARE, cpu) ==
470                     NOTIFY_BAD)
471                         goto cleanup;
472         __register_cpu_notifier(&pool->notifier);
473         cpu_notifier_register_done();
474         return 0;
475
476 cleanup:
477         for_each_online_cpu(cpu)
478                 __zswap_cpu_comp_notifier(pool, CPU_UP_CANCELED, cpu);
479         cpu_notifier_register_done();
480         return -ENOMEM;
481 }
482
483 static void zswap_cpu_comp_destroy(struct zswap_pool *pool)
484 {
485         unsigned long cpu;
486
487         cpu_notifier_register_begin();
488         for_each_online_cpu(cpu)
489                 __zswap_cpu_comp_notifier(pool, CPU_UP_CANCELED, cpu);
490         __unregister_cpu_notifier(&pool->notifier);
491         cpu_notifier_register_done();
492 }
493
494 /*********************************
495 * pool functions
496 **********************************/
497
498 static struct zswap_pool *__zswap_pool_current(void)
499 {
500         struct zswap_pool *pool;
501
502         pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
503         WARN_ON(!pool);
504
505         return pool;
506 }
507
508 static struct zswap_pool *zswap_pool_current(void)
509 {
510         assert_spin_locked(&zswap_pools_lock);
511
512         return __zswap_pool_current();
513 }
514
515 static struct zswap_pool *zswap_pool_current_get(void)
516 {
517         struct zswap_pool *pool;
518
519         rcu_read_lock();
520
521         pool = __zswap_pool_current();
522         if (!pool || !zswap_pool_get(pool))
523                 pool = NULL;
524
525         rcu_read_unlock();
526
527         return pool;
528 }
529
530 static struct zswap_pool *zswap_pool_last_get(void)
531 {
532         struct zswap_pool *pool, *last = NULL;
533
534         rcu_read_lock();
535
536         list_for_each_entry_rcu(pool, &zswap_pools, list)
537                 last = pool;
538         if (!WARN_ON(!last) && !zswap_pool_get(last))
539                 last = NULL;
540
541         rcu_read_unlock();
542
543         return last;
544 }
545
546 /* type and compressor must be null-terminated */
547 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
548 {
549         struct zswap_pool *pool;
550
551         assert_spin_locked(&zswap_pools_lock);
552
553         list_for_each_entry_rcu(pool, &zswap_pools, list) {
554                 if (strcmp(pool->tfm_name, compressor))
555                         continue;
556                 if (strcmp(zpool_get_type(pool->zpool), type))
557                         continue;
558                 /* if we can't get it, it's about to be destroyed */
559                 if (!zswap_pool_get(pool))
560                         continue;
561                 return pool;
562         }
563
564         return NULL;
565 }
566
567 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
568 {
569         struct zswap_pool *pool;
570         char name[38]; /* 'zswap' + 32 char (max) num + \0 */
571         gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
572
573         pool = kzalloc(sizeof(*pool), GFP_KERNEL);
574         if (!pool) {
575                 pr_err("pool alloc failed\n");
576                 return NULL;
577         }
578
579         /* unique name for each pool specifically required by zsmalloc */
580         snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
581
582         pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
583         if (!pool->zpool) {
584                 pr_err("%s zpool not available\n", type);
585                 goto error;
586         }
587         pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
588
589         strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
590         pool->tfm = alloc_percpu(struct crypto_comp *);
591         if (!pool->tfm) {
592                 pr_err("percpu alloc failed\n");
593                 goto error;
594         }
595
596         if (zswap_cpu_comp_init(pool))
597                 goto error;
598         pr_debug("using %s compressor\n", pool->tfm_name);
599
600         /* being the current pool takes 1 ref; this func expects the
601          * caller to always add the new pool as the current pool
602          */
603         kref_init(&pool->kref);
604         INIT_LIST_HEAD(&pool->list);
605
606         zswap_pool_debug("created", pool);
607
608         return pool;
609
610 error:
611         free_percpu(pool->tfm);
612         if (pool->zpool)
613                 zpool_destroy_pool(pool->zpool);
614         kfree(pool);
615         return NULL;
616 }
617
618 static __init struct zswap_pool *__zswap_pool_create_fallback(void)
619 {
620         if (!crypto_has_comp(zswap_compressor, 0, 0)) {
621                 if (!strcmp(zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT)) {
622                         pr_err("default compressor %s not available\n",
623                                zswap_compressor);
624                         return NULL;
625                 }
626                 pr_err("compressor %s not available, using default %s\n",
627                        zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT);
628                 param_free_charp(&zswap_compressor);
629                 zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
630         }
631         if (!zpool_has_pool(zswap_zpool_type)) {
632                 if (!strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) {
633                         pr_err("default zpool %s not available\n",
634                                zswap_zpool_type);
635                         return NULL;
636                 }
637                 pr_err("zpool %s not available, using default %s\n",
638                        zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT);
639                 param_free_charp(&zswap_zpool_type);
640                 zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
641         }
642
643         return zswap_pool_create(zswap_zpool_type, zswap_compressor);
644 }
645
646 static void zswap_pool_destroy(struct zswap_pool *pool)
647 {
648         zswap_pool_debug("destroying", pool);
649
650         zswap_cpu_comp_destroy(pool);
651         free_percpu(pool->tfm);
652         zpool_destroy_pool(pool->zpool);
653         kfree(pool);
654 }
655
656 static int __must_check zswap_pool_get(struct zswap_pool *pool)
657 {
658         return kref_get_unless_zero(&pool->kref);
659 }
660
661 static void __zswap_pool_release(struct work_struct *work)
662 {
663         struct zswap_pool *pool = container_of(work, typeof(*pool), work);
664
665         synchronize_rcu();
666
667         /* nobody should have been able to get a kref... */
668         WARN_ON(kref_get_unless_zero(&pool->kref));
669
670         /* pool is now off zswap_pools list and has no references. */
671         zswap_pool_destroy(pool);
672 }
673
674 static void __zswap_pool_empty(struct kref *kref)
675 {
676         struct zswap_pool *pool;
677
678         pool = container_of(kref, typeof(*pool), kref);
679
680         spin_lock(&zswap_pools_lock);
681
682         WARN_ON(pool == zswap_pool_current());
683
684         list_del_rcu(&pool->list);
685
686         INIT_WORK(&pool->work, __zswap_pool_release);
687         schedule_work(&pool->work);
688
689         spin_unlock(&zswap_pools_lock);
690 }
691
692 static void zswap_pool_put(struct zswap_pool *pool)
693 {
694         kref_put(&pool->kref, __zswap_pool_empty);
695 }
696
697 /*********************************
698 * param callbacks
699 **********************************/
700
701 /* val must be a null-terminated string */
702 static int __zswap_param_set(const char *val, const struct kernel_param *kp,
703                              char *type, char *compressor)
704 {
705         struct zswap_pool *pool, *put_pool = NULL;
706         char *s = strstrip((char *)val);
707         int ret;
708
709         /* no change required */
710         if (!strcmp(s, *(char **)kp->arg))
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                 list_del_rcu(&pool->list);
742         } else {
743                 spin_unlock(&zswap_pools_lock);
744                 pool = zswap_pool_create(type, compressor);
745                 spin_lock(&zswap_pools_lock);
746         }
747
748         if (pool)
749                 ret = param_set_charp(s, kp);
750         else
751                 ret = -EINVAL;
752
753         if (!ret) {
754                 put_pool = zswap_pool_current();
755                 list_add_rcu(&pool->list, &zswap_pools);
756         } else if (pool) {
757                 /* add the possibly pre-existing pool to the end of the pools
758                  * list; if it's new (and empty) then it'll be removed and
759                  * destroyed by the put after we drop the lock
760                  */
761                 list_add_tail_rcu(&pool->list, &zswap_pools);
762                 put_pool = pool;
763         }
764
765         spin_unlock(&zswap_pools_lock);
766
767         /* drop the ref from either the old current pool,
768          * or the new pool we failed to add
769          */
770         if (put_pool)
771                 zswap_pool_put(put_pool);
772
773         return ret;
774 }
775
776 static int zswap_compressor_param_set(const char *val,
777                                       const struct kernel_param *kp)
778 {
779         return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
780 }
781
782 static int zswap_zpool_param_set(const char *val,
783                                  const struct kernel_param *kp)
784 {
785         return __zswap_param_set(val, kp, NULL, zswap_compressor);
786 }
787
788 /*********************************
789 * writeback code
790 **********************************/
791 /* return enum for zswap_get_swap_cache_page */
792 enum zswap_get_swap_ret {
793         ZSWAP_SWAPCACHE_NEW,
794         ZSWAP_SWAPCACHE_EXIST,
795         ZSWAP_SWAPCACHE_FAIL,
796 };
797
798 /*
799  * zswap_get_swap_cache_page
800  *
801  * This is an adaption of read_swap_cache_async()
802  *
803  * This function tries to find a page with the given swap entry
804  * in the swapper_space address space (the swap cache).  If the page
805  * is found, it is returned in retpage.  Otherwise, a page is allocated,
806  * added to the swap cache, and returned in retpage.
807  *
808  * If success, the swap cache page is returned in retpage
809  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
810  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
811  *     the new page is added to swapcache and locked
812  * Returns ZSWAP_SWAPCACHE_FAIL on error
813  */
814 static int zswap_get_swap_cache_page(swp_entry_t entry,
815                                 struct page **retpage)
816 {
817         bool page_was_allocated;
818
819         *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
820                         NULL, 0, &page_was_allocated);
821         if (page_was_allocated)
822                 return ZSWAP_SWAPCACHE_NEW;
823         if (!*retpage)
824                 return ZSWAP_SWAPCACHE_FAIL;
825         return ZSWAP_SWAPCACHE_EXIST;
826 }
827
828 /*
829  * Attempts to free an entry by adding a page to the swap cache,
830  * decompressing the entry data into the page, and issuing a
831  * bio write to write the page back to the swap device.
832  *
833  * This can be thought of as a "resumed writeback" of the page
834  * to the swap device.  We are basically resuming the same swap
835  * writeback path that was intercepted with the frontswap_store()
836  * in the first place.  After the page has been decompressed into
837  * the swap cache, the compressed version stored by zswap can be
838  * freed.
839  */
840 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
841 {
842         struct zswap_header *zhdr;
843         swp_entry_t swpentry;
844         struct zswap_tree *tree;
845         pgoff_t offset;
846         struct zswap_entry *entry;
847         struct page *page;
848         struct crypto_comp *tfm;
849         u8 *src, *dst;
850         unsigned int dlen;
851         int ret;
852         struct writeback_control wbc = {
853                 .sync_mode = WB_SYNC_NONE,
854         };
855
856         /* extract swpentry from data */
857         zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
858         swpentry = zhdr->swpentry; /* here */
859         zpool_unmap_handle(pool, handle);
860         tree = zswap_trees[swp_type(swpentry)];
861         offset = swp_offset(swpentry);
862
863         /* find and ref zswap entry */
864         spin_lock(&tree->lock);
865         entry = zswap_entry_find_get(&tree->rbroot, offset);
866         if (!entry) {
867                 /* entry was invalidated */
868                 spin_unlock(&tree->lock);
869                 return 0;
870         }
871         spin_unlock(&tree->lock);
872         BUG_ON(offset != entry->offset);
873
874         /* try to allocate swap cache page */
875         switch (zswap_get_swap_cache_page(swpentry, &page)) {
876         case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
877                 ret = -ENOMEM;
878                 goto fail;
879
880         case ZSWAP_SWAPCACHE_EXIST:
881                 /* page is already in the swap cache, ignore for now */
882                 put_page(page);
883                 ret = -EEXIST;
884                 goto fail;
885
886         case ZSWAP_SWAPCACHE_NEW: /* page is locked */
887                 /* decompress */
888                 dlen = PAGE_SIZE;
889                 src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle,
890                                 ZPOOL_MM_RO) + sizeof(struct zswap_header);
891                 dst = kmap_atomic(page);
892                 tfm = *get_cpu_ptr(entry->pool->tfm);
893                 ret = crypto_comp_decompress(tfm, src, entry->length,
894                                              dst, &dlen);
895                 put_cpu_ptr(entry->pool->tfm);
896                 kunmap_atomic(dst);
897                 zpool_unmap_handle(entry->pool->zpool, entry->handle);
898                 BUG_ON(ret);
899                 BUG_ON(dlen != PAGE_SIZE);
900
901                 /* page is up to date */
902                 SetPageUptodate(page);
903         }
904
905         /* move it to the tail of the inactive list after end_writeback */
906         SetPageReclaim(page);
907
908         /* start writeback */
909         __swap_writepage(page, &wbc, end_swap_bio_write);
910         put_page(page);
911         zswap_written_back_pages++;
912
913         spin_lock(&tree->lock);
914         /* drop local reference */
915         zswap_entry_put(tree, entry);
916
917         /*
918         * There are two possible situations for entry here:
919         * (1) refcount is 1(normal case),  entry is valid and on the tree
920         * (2) refcount is 0, entry is freed and not on the tree
921         *     because invalidate happened during writeback
922         *  search the tree and free the entry if find entry
923         */
924         if (entry == zswap_rb_search(&tree->rbroot, offset))
925                 zswap_entry_put(tree, entry);
926         spin_unlock(&tree->lock);
927
928         goto end;
929
930         /*
931         * if we get here due to ZSWAP_SWAPCACHE_EXIST
932         * a load may happening concurrently
933         * it is safe and okay to not free the entry
934         * if we free the entry in the following put
935         * it it either okay to return !0
936         */
937 fail:
938         spin_lock(&tree->lock);
939         zswap_entry_put(tree, entry);
940         spin_unlock(&tree->lock);
941
942 end:
943         return ret;
944 }
945
946 static int zswap_shrink(void)
947 {
948         struct zswap_pool *pool;
949         int ret;
950
951         pool = zswap_pool_last_get();
952         if (!pool)
953                 return -ENOENT;
954
955         ret = zpool_shrink(pool->zpool, 1, NULL);
956
957         zswap_pool_put(pool);
958
959         return ret;
960 }
961
962 /*********************************
963 * frontswap hooks
964 **********************************/
965 /* attempts to compress and store an single page */
966 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
967                                 struct page *page)
968 {
969         struct zswap_tree *tree = zswap_trees[type];
970         struct zswap_entry *entry, *dupentry;
971         struct crypto_comp *tfm;
972         int ret;
973         unsigned int dlen = PAGE_SIZE, len;
974         unsigned long handle;
975         char *buf;
976         u8 *src, *dst;
977         struct zswap_header *zhdr;
978
979         if (!zswap_enabled || !tree) {
980                 ret = -ENODEV;
981                 goto reject;
982         }
983
984         /* reclaim space if needed */
985         if (zswap_is_full()) {
986                 zswap_pool_limit_hit++;
987                 if (zswap_shrink()) {
988                         zswap_reject_reclaim_fail++;
989                         ret = -ENOMEM;
990                         goto reject;
991                 }
992         }
993
994         /* allocate entry */
995         entry = zswap_entry_cache_alloc(GFP_KERNEL);
996         if (!entry) {
997                 zswap_reject_kmemcache_fail++;
998                 ret = -ENOMEM;
999                 goto reject;
1000         }
1001
1002         /* if entry is successfully added, it keeps the reference */
1003         entry->pool = zswap_pool_current_get();
1004         if (!entry->pool) {
1005                 ret = -EINVAL;
1006                 goto freepage;
1007         }
1008
1009         /* compress */
1010         dst = get_cpu_var(zswap_dstmem);
1011         tfm = *get_cpu_ptr(entry->pool->tfm);
1012         src = kmap_atomic(page);
1013         ret = crypto_comp_compress(tfm, src, PAGE_SIZE, dst, &dlen);
1014         kunmap_atomic(src);
1015         put_cpu_ptr(entry->pool->tfm);
1016         if (ret) {
1017                 ret = -EINVAL;
1018                 goto put_dstmem;
1019         }
1020
1021         /* store */
1022         len = dlen + sizeof(struct zswap_header);
1023         ret = zpool_malloc(entry->pool->zpool, len,
1024                            __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM,
1025                            &handle);
1026         if (ret == -ENOSPC) {
1027                 zswap_reject_compress_poor++;
1028                 goto put_dstmem;
1029         }
1030         if (ret) {
1031                 zswap_reject_alloc_fail++;
1032                 goto put_dstmem;
1033         }
1034         zhdr = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW);
1035         zhdr->swpentry = swp_entry(type, offset);
1036         buf = (u8 *)(zhdr + 1);
1037         memcpy(buf, dst, dlen);
1038         zpool_unmap_handle(entry->pool->zpool, handle);
1039         put_cpu_var(zswap_dstmem);
1040
1041         /* populate entry */
1042         entry->offset = offset;
1043         entry->handle = handle;
1044         entry->length = dlen;
1045
1046         /* map */
1047         spin_lock(&tree->lock);
1048         do {
1049                 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1050                 if (ret == -EEXIST) {
1051                         zswap_duplicate_entry++;
1052                         /* remove from rbtree */
1053                         zswap_rb_erase(&tree->rbroot, dupentry);
1054                         zswap_entry_put(tree, dupentry);
1055                 }
1056         } while (ret == -EEXIST);
1057         spin_unlock(&tree->lock);
1058
1059         /* update stats */
1060         atomic_inc(&zswap_stored_pages);
1061         zswap_update_total_size();
1062
1063         return 0;
1064
1065 put_dstmem:
1066         put_cpu_var(zswap_dstmem);
1067         zswap_pool_put(entry->pool);
1068 freepage:
1069         zswap_entry_cache_free(entry);
1070 reject:
1071         return ret;
1072 }
1073
1074 /*
1075  * returns 0 if the page was successfully decompressed
1076  * return -1 on entry not found or error
1077 */
1078 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1079                                 struct page *page)
1080 {
1081         struct zswap_tree *tree = zswap_trees[type];
1082         struct zswap_entry *entry;
1083         struct crypto_comp *tfm;
1084         u8 *src, *dst;
1085         unsigned int dlen;
1086         int ret;
1087
1088         /* find */
1089         spin_lock(&tree->lock);
1090         entry = zswap_entry_find_get(&tree->rbroot, offset);
1091         if (!entry) {
1092                 /* entry was written back */
1093                 spin_unlock(&tree->lock);
1094                 return -1;
1095         }
1096         spin_unlock(&tree->lock);
1097
1098         /* decompress */
1099         dlen = PAGE_SIZE;
1100         src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle,
1101                         ZPOOL_MM_RO) + sizeof(struct zswap_header);
1102         dst = kmap_atomic(page);
1103         tfm = *get_cpu_ptr(entry->pool->tfm);
1104         ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen);
1105         put_cpu_ptr(entry->pool->tfm);
1106         kunmap_atomic(dst);
1107         zpool_unmap_handle(entry->pool->zpool, entry->handle);
1108         BUG_ON(ret);
1109
1110         spin_lock(&tree->lock);
1111         zswap_entry_put(tree, entry);
1112         spin_unlock(&tree->lock);
1113
1114         return 0;
1115 }
1116
1117 /* frees an entry in zswap */
1118 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1119 {
1120         struct zswap_tree *tree = zswap_trees[type];
1121         struct zswap_entry *entry;
1122
1123         /* find */
1124         spin_lock(&tree->lock);
1125         entry = zswap_rb_search(&tree->rbroot, offset);
1126         if (!entry) {
1127                 /* entry was written back */
1128                 spin_unlock(&tree->lock);
1129                 return;
1130         }
1131
1132         /* remove from rbtree */
1133         zswap_rb_erase(&tree->rbroot, entry);
1134
1135         /* drop the initial reference from entry creation */
1136         zswap_entry_put(tree, entry);
1137
1138         spin_unlock(&tree->lock);
1139 }
1140
1141 /* frees all zswap entries for the given swap type */
1142 static void zswap_frontswap_invalidate_area(unsigned type)
1143 {
1144         struct zswap_tree *tree = zswap_trees[type];
1145         struct zswap_entry *entry, *n;
1146
1147         if (!tree)
1148                 return;
1149
1150         /* walk the tree and free everything */
1151         spin_lock(&tree->lock);
1152         rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1153                 zswap_free_entry(entry);
1154         tree->rbroot = RB_ROOT;
1155         spin_unlock(&tree->lock);
1156         kfree(tree);
1157         zswap_trees[type] = NULL;
1158 }
1159
1160 static void zswap_frontswap_init(unsigned type)
1161 {
1162         struct zswap_tree *tree;
1163
1164         tree = kzalloc(sizeof(struct zswap_tree), GFP_KERNEL);
1165         if (!tree) {
1166                 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1167                 return;
1168         }
1169
1170         tree->rbroot = RB_ROOT;
1171         spin_lock_init(&tree->lock);
1172         zswap_trees[type] = tree;
1173 }
1174
1175 static struct frontswap_ops zswap_frontswap_ops = {
1176         .store = zswap_frontswap_store,
1177         .load = zswap_frontswap_load,
1178         .invalidate_page = zswap_frontswap_invalidate_page,
1179         .invalidate_area = zswap_frontswap_invalidate_area,
1180         .init = zswap_frontswap_init
1181 };
1182
1183 /*********************************
1184 * debugfs functions
1185 **********************************/
1186 #ifdef CONFIG_DEBUG_FS
1187 #include <linux/debugfs.h>
1188
1189 static struct dentry *zswap_debugfs_root;
1190
1191 static int __init zswap_debugfs_init(void)
1192 {
1193         if (!debugfs_initialized())
1194                 return -ENODEV;
1195
1196         zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1197         if (!zswap_debugfs_root)
1198                 return -ENOMEM;
1199
1200         debugfs_create_u64("pool_limit_hit", S_IRUGO,
1201                         zswap_debugfs_root, &zswap_pool_limit_hit);
1202         debugfs_create_u64("reject_reclaim_fail", S_IRUGO,
1203                         zswap_debugfs_root, &zswap_reject_reclaim_fail);
1204         debugfs_create_u64("reject_alloc_fail", S_IRUGO,
1205                         zswap_debugfs_root, &zswap_reject_alloc_fail);
1206         debugfs_create_u64("reject_kmemcache_fail", S_IRUGO,
1207                         zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1208         debugfs_create_u64("reject_compress_poor", S_IRUGO,
1209                         zswap_debugfs_root, &zswap_reject_compress_poor);
1210         debugfs_create_u64("written_back_pages", S_IRUGO,
1211                         zswap_debugfs_root, &zswap_written_back_pages);
1212         debugfs_create_u64("duplicate_entry", S_IRUGO,
1213                         zswap_debugfs_root, &zswap_duplicate_entry);
1214         debugfs_create_u64("pool_total_size", S_IRUGO,
1215                         zswap_debugfs_root, &zswap_pool_total_size);
1216         debugfs_create_atomic_t("stored_pages", S_IRUGO,
1217                         zswap_debugfs_root, &zswap_stored_pages);
1218
1219         return 0;
1220 }
1221
1222 static void __exit zswap_debugfs_exit(void)
1223 {
1224         debugfs_remove_recursive(zswap_debugfs_root);
1225 }
1226 #else
1227 static int __init zswap_debugfs_init(void)
1228 {
1229         return 0;
1230 }
1231
1232 static void __exit zswap_debugfs_exit(void) { }
1233 #endif
1234
1235 /*********************************
1236 * module init and exit
1237 **********************************/
1238 static int __init init_zswap(void)
1239 {
1240         struct zswap_pool *pool;
1241
1242         zswap_init_started = true;
1243
1244         if (zswap_entry_cache_create()) {
1245                 pr_err("entry cache creation failed\n");
1246                 goto cache_fail;
1247         }
1248
1249         if (zswap_cpu_dstmem_init()) {
1250                 pr_err("dstmem alloc failed\n");
1251                 goto dstmem_fail;
1252         }
1253
1254         pool = __zswap_pool_create_fallback();
1255         if (!pool) {
1256                 pr_err("pool creation failed\n");
1257                 goto pool_fail;
1258         }
1259         pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1260                 zpool_get_type(pool->zpool));
1261
1262         list_add(&pool->list, &zswap_pools);
1263
1264         frontswap_register_ops(&zswap_frontswap_ops);
1265         if (zswap_debugfs_init())
1266                 pr_warn("debugfs initialization failed\n");
1267         return 0;
1268
1269 pool_fail:
1270         zswap_cpu_dstmem_destroy();
1271 dstmem_fail:
1272         zswap_entry_cache_destroy();
1273 cache_fail:
1274         return -ENOMEM;
1275 }
1276 /* must be late so crypto has time to come up */
1277 late_initcall(init_zswap);
1278
1279 MODULE_LICENSE("GPL");
1280 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1281 MODULE_DESCRIPTION("Compressed cache for swap pages");