Loading...
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/scatterlist.h>
28#include <linux/mempool.h>
29#include <linux/zpool.h>
30#include <crypto/acompress.h>
31
32#include <linux/mm_types.h>
33#include <linux/page-flags.h>
34#include <linux/swapops.h>
35#include <linux/writeback.h>
36#include <linux/pagemap.h>
37#include <linux/workqueue.h>
38
39/*********************************
40* statistics
41**********************************/
42/* Total bytes used by the compressed storage */
43static u64 zswap_pool_total_size;
44/* The number of compressed pages currently stored in zswap */
45static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
46/* The number of same-value filled pages currently stored in zswap */
47static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
48
49/*
50 * The statistics below are not protected from concurrent access for
51 * performance reasons so they may not be a 100% accurate. However,
52 * they do provide useful information on roughly how many times a
53 * certain event is occurring.
54*/
55
56/* Pool limit was hit (see zswap_max_pool_percent) */
57static u64 zswap_pool_limit_hit;
58/* Pages written back when pool limit was reached */
59static u64 zswap_written_back_pages;
60/* Store failed due to a reclaim failure after pool limit was reached */
61static u64 zswap_reject_reclaim_fail;
62/* Compressed page was too big for the allocator to (optimally) store */
63static u64 zswap_reject_compress_poor;
64/* Store failed because underlying allocator could not get memory */
65static u64 zswap_reject_alloc_fail;
66/* Store failed because the entry metadata could not be allocated (rare) */
67static u64 zswap_reject_kmemcache_fail;
68/* Duplicate store was encountered (rare) */
69static u64 zswap_duplicate_entry;
70
71/* Shrinker work queue */
72static struct workqueue_struct *shrink_wq;
73/* Pool limit was hit, we need to calm down */
74static bool zswap_pool_reached_full;
75
76/*********************************
77* tunables
78**********************************/
79
80#define ZSWAP_PARAM_UNSET ""
81
82/* Enable/disable zswap */
83static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
84static int zswap_enabled_param_set(const char *,
85 const struct kernel_param *);
86static const struct kernel_param_ops zswap_enabled_param_ops = {
87 .set = zswap_enabled_param_set,
88 .get = param_get_bool,
89};
90module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
91
92/* Crypto compressor to use */
93static char *zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
94static int zswap_compressor_param_set(const char *,
95 const struct kernel_param *);
96static const struct kernel_param_ops zswap_compressor_param_ops = {
97 .set = zswap_compressor_param_set,
98 .get = param_get_charp,
99 .free = param_free_charp,
100};
101module_param_cb(compressor, &zswap_compressor_param_ops,
102 &zswap_compressor, 0644);
103
104/* Compressed storage zpool to use */
105static char *zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
106static int zswap_zpool_param_set(const char *, const struct kernel_param *);
107static const struct kernel_param_ops zswap_zpool_param_ops = {
108 .set = zswap_zpool_param_set,
109 .get = param_get_charp,
110 .free = param_free_charp,
111};
112module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
113
114/* The maximum percentage of memory that the compressed pool can occupy */
115static unsigned int zswap_max_pool_percent = 20;
116module_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 */
119static unsigned int zswap_accept_thr_percent = 90; /* of max pool size */
120module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
121 uint, 0644);
122
123/* Enable/disable handling same-value filled pages (enabled by default) */
124static bool zswap_same_filled_pages_enabled = true;
125module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
126 bool, 0644);
127
128/*********************************
129* data structures
130**********************************/
131
132struct crypto_acomp_ctx {
133 struct crypto_acomp *acomp;
134 struct acomp_req *req;
135 struct crypto_wait wait;
136 u8 *dstmem;
137 struct mutex *mutex;
138};
139
140struct zswap_pool {
141 struct zpool *zpool;
142 struct crypto_acomp_ctx __percpu *acomp_ctx;
143 struct kref kref;
144 struct list_head list;
145 struct work_struct release_work;
146 struct work_struct shrink_work;
147 struct hlist_node node;
148 char tfm_name[CRYPTO_MAX_ALG_NAME];
149};
150
151/*
152 * struct zswap_entry
153 *
154 * This structure contains the metadata for tracking a single compressed
155 * page within zswap.
156 *
157 * rbnode - links the entry into red-black tree for the appropriate swap type
158 * offset - the swap offset for the entry. Index into the red-black tree.
159 * refcount - the number of outstanding reference to the entry. This is needed
160 * to protect against premature freeing of the entry by code
161 * concurrent calls to load, invalidate, and writeback. The lock
162 * for the zswap_tree structure that contains the entry must
163 * be held while changing the refcount. Since the lock must
164 * be held, there is no reason to also make refcount atomic.
165 * length - the length in bytes of the compressed page data. Needed during
166 * decompression. For a same value filled page length is 0.
167 * pool - the zswap_pool the entry's data is in
168 * handle - zpool allocation handle that stores the compressed page data
169 * value - value of the same-value filled pages which have same content
170 */
171struct zswap_entry {
172 struct rb_node rbnode;
173 pgoff_t offset;
174 int refcount;
175 unsigned int length;
176 struct zswap_pool *pool;
177 union {
178 unsigned long handle;
179 unsigned long value;
180 };
181};
182
183struct zswap_header {
184 swp_entry_t swpentry;
185};
186
187/*
188 * The tree lock in the zswap_tree struct protects a few things:
189 * - the rbtree
190 * - the refcount field of each entry in the tree
191 */
192struct zswap_tree {
193 struct rb_root rbroot;
194 spinlock_t lock;
195};
196
197static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
198
199/* RCU-protected iteration */
200static LIST_HEAD(zswap_pools);
201/* protects zswap_pools list modification */
202static DEFINE_SPINLOCK(zswap_pools_lock);
203/* pool counter to provide unique names to zpool */
204static atomic_t zswap_pools_count = ATOMIC_INIT(0);
205
206/* used by param callback function */
207static bool zswap_init_started;
208
209/* fatal error during init */
210static bool zswap_init_failed;
211
212/* init completed, but couldn't create the initial pool */
213static bool zswap_has_pool;
214
215/*********************************
216* helpers and fwd declarations
217**********************************/
218
219#define zswap_pool_debug(msg, p) \
220 pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name, \
221 zpool_get_type((p)->zpool))
222
223static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
224static int zswap_pool_get(struct zswap_pool *pool);
225static void zswap_pool_put(struct zswap_pool *pool);
226
227static const struct zpool_ops zswap_zpool_ops = {
228 .evict = zswap_writeback_entry
229};
230
231static bool zswap_is_full(void)
232{
233 return totalram_pages() * zswap_max_pool_percent / 100 <
234 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
235}
236
237static bool zswap_can_accept(void)
238{
239 return totalram_pages() * zswap_accept_thr_percent / 100 *
240 zswap_max_pool_percent / 100 >
241 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
242}
243
244static void zswap_update_total_size(void)
245{
246 struct zswap_pool *pool;
247 u64 total = 0;
248
249 rcu_read_lock();
250
251 list_for_each_entry_rcu(pool, &zswap_pools, list)
252 total += zpool_get_total_size(pool->zpool);
253
254 rcu_read_unlock();
255
256 zswap_pool_total_size = total;
257}
258
259/*********************************
260* zswap entry functions
261**********************************/
262static struct kmem_cache *zswap_entry_cache;
263
264static int __init zswap_entry_cache_create(void)
265{
266 zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
267 return zswap_entry_cache == NULL;
268}
269
270static void __init zswap_entry_cache_destroy(void)
271{
272 kmem_cache_destroy(zswap_entry_cache);
273}
274
275static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
276{
277 struct zswap_entry *entry;
278 entry = kmem_cache_alloc(zswap_entry_cache, gfp);
279 if (!entry)
280 return NULL;
281 entry->refcount = 1;
282 RB_CLEAR_NODE(&entry->rbnode);
283 return entry;
284}
285
286static void zswap_entry_cache_free(struct zswap_entry *entry)
287{
288 kmem_cache_free(zswap_entry_cache, entry);
289}
290
291/*********************************
292* rbtree functions
293**********************************/
294static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
295{
296 struct rb_node *node = root->rb_node;
297 struct zswap_entry *entry;
298
299 while (node) {
300 entry = rb_entry(node, struct zswap_entry, rbnode);
301 if (entry->offset > offset)
302 node = node->rb_left;
303 else if (entry->offset < offset)
304 node = node->rb_right;
305 else
306 return entry;
307 }
308 return NULL;
309}
310
311/*
312 * In the case that a entry with the same offset is found, a pointer to
313 * the existing entry is stored in dupentry and the function returns -EEXIST
314 */
315static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
316 struct zswap_entry **dupentry)
317{
318 struct rb_node **link = &root->rb_node, *parent = NULL;
319 struct zswap_entry *myentry;
320
321 while (*link) {
322 parent = *link;
323 myentry = rb_entry(parent, struct zswap_entry, rbnode);
324 if (myentry->offset > entry->offset)
325 link = &(*link)->rb_left;
326 else if (myentry->offset < entry->offset)
327 link = &(*link)->rb_right;
328 else {
329 *dupentry = myentry;
330 return -EEXIST;
331 }
332 }
333 rb_link_node(&entry->rbnode, parent, link);
334 rb_insert_color(&entry->rbnode, root);
335 return 0;
336}
337
338static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
339{
340 if (!RB_EMPTY_NODE(&entry->rbnode)) {
341 rb_erase(&entry->rbnode, root);
342 RB_CLEAR_NODE(&entry->rbnode);
343 }
344}
345
346/*
347 * Carries out the common pattern of freeing and entry's zpool allocation,
348 * freeing the entry itself, and decrementing the number of stored pages.
349 */
350static void zswap_free_entry(struct zswap_entry *entry)
351{
352 if (!entry->length)
353 atomic_dec(&zswap_same_filled_pages);
354 else {
355 zpool_free(entry->pool->zpool, entry->handle);
356 zswap_pool_put(entry->pool);
357 }
358 zswap_entry_cache_free(entry);
359 atomic_dec(&zswap_stored_pages);
360 zswap_update_total_size();
361}
362
363/* caller must hold the tree lock */
364static void zswap_entry_get(struct zswap_entry *entry)
365{
366 entry->refcount++;
367}
368
369/* caller must hold the tree lock
370* remove from the tree and free it, if nobody reference the entry
371*/
372static void zswap_entry_put(struct zswap_tree *tree,
373 struct zswap_entry *entry)
374{
375 int refcount = --entry->refcount;
376
377 BUG_ON(refcount < 0);
378 if (refcount == 0) {
379 zswap_rb_erase(&tree->rbroot, entry);
380 zswap_free_entry(entry);
381 }
382}
383
384/* caller must hold the tree lock */
385static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
386 pgoff_t offset)
387{
388 struct zswap_entry *entry;
389
390 entry = zswap_rb_search(root, offset);
391 if (entry)
392 zswap_entry_get(entry);
393
394 return entry;
395}
396
397/*********************************
398* per-cpu code
399**********************************/
400static DEFINE_PER_CPU(u8 *, zswap_dstmem);
401/*
402 * If users dynamically change the zpool type and compressor at runtime, i.e.
403 * zswap is running, zswap can have more than one zpool on one cpu, but they
404 * are sharing dtsmem. So we need this mutex to be per-cpu.
405 */
406static DEFINE_PER_CPU(struct mutex *, zswap_mutex);
407
408static int zswap_dstmem_prepare(unsigned int cpu)
409{
410 struct mutex *mutex;
411 u8 *dst;
412
413 dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
414 if (!dst)
415 return -ENOMEM;
416
417 mutex = kmalloc_node(sizeof(*mutex), GFP_KERNEL, cpu_to_node(cpu));
418 if (!mutex) {
419 kfree(dst);
420 return -ENOMEM;
421 }
422
423 mutex_init(mutex);
424 per_cpu(zswap_dstmem, cpu) = dst;
425 per_cpu(zswap_mutex, cpu) = mutex;
426 return 0;
427}
428
429static int zswap_dstmem_dead(unsigned int cpu)
430{
431 struct mutex *mutex;
432 u8 *dst;
433
434 mutex = per_cpu(zswap_mutex, cpu);
435 kfree(mutex);
436 per_cpu(zswap_mutex, cpu) = NULL;
437
438 dst = per_cpu(zswap_dstmem, cpu);
439 kfree(dst);
440 per_cpu(zswap_dstmem, cpu) = NULL;
441
442 return 0;
443}
444
445static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
446{
447 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
448 struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
449 struct crypto_acomp *acomp;
450 struct acomp_req *req;
451
452 acomp = crypto_alloc_acomp_node(pool->tfm_name, 0, 0, cpu_to_node(cpu));
453 if (IS_ERR(acomp)) {
454 pr_err("could not alloc crypto acomp %s : %ld\n",
455 pool->tfm_name, PTR_ERR(acomp));
456 return PTR_ERR(acomp);
457 }
458 acomp_ctx->acomp = acomp;
459
460 req = acomp_request_alloc(acomp_ctx->acomp);
461 if (!req) {
462 pr_err("could not alloc crypto acomp_request %s\n",
463 pool->tfm_name);
464 crypto_free_acomp(acomp_ctx->acomp);
465 return -ENOMEM;
466 }
467 acomp_ctx->req = req;
468
469 crypto_init_wait(&acomp_ctx->wait);
470 /*
471 * if the backend of acomp is async zip, crypto_req_done() will wakeup
472 * crypto_wait_req(); if the backend of acomp is scomp, the callback
473 * won't be called, crypto_wait_req() will return without blocking.
474 */
475 acomp_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
476 crypto_req_done, &acomp_ctx->wait);
477
478 acomp_ctx->mutex = per_cpu(zswap_mutex, cpu);
479 acomp_ctx->dstmem = per_cpu(zswap_dstmem, cpu);
480
481 return 0;
482}
483
484static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
485{
486 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
487 struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
488
489 if (!IS_ERR_OR_NULL(acomp_ctx)) {
490 if (!IS_ERR_OR_NULL(acomp_ctx->req))
491 acomp_request_free(acomp_ctx->req);
492 if (!IS_ERR_OR_NULL(acomp_ctx->acomp))
493 crypto_free_acomp(acomp_ctx->acomp);
494 }
495
496 return 0;
497}
498
499/*********************************
500* pool functions
501**********************************/
502
503static struct zswap_pool *__zswap_pool_current(void)
504{
505 struct zswap_pool *pool;
506
507 pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
508 WARN_ONCE(!pool && zswap_has_pool,
509 "%s: no page storage pool!\n", __func__);
510
511 return pool;
512}
513
514static struct zswap_pool *zswap_pool_current(void)
515{
516 assert_spin_locked(&zswap_pools_lock);
517
518 return __zswap_pool_current();
519}
520
521static struct zswap_pool *zswap_pool_current_get(void)
522{
523 struct zswap_pool *pool;
524
525 rcu_read_lock();
526
527 pool = __zswap_pool_current();
528 if (!zswap_pool_get(pool))
529 pool = NULL;
530
531 rcu_read_unlock();
532
533 return pool;
534}
535
536static struct zswap_pool *zswap_pool_last_get(void)
537{
538 struct zswap_pool *pool, *last = NULL;
539
540 rcu_read_lock();
541
542 list_for_each_entry_rcu(pool, &zswap_pools, list)
543 last = pool;
544 WARN_ONCE(!last && zswap_has_pool,
545 "%s: no page storage pool!\n", __func__);
546 if (!zswap_pool_get(last))
547 last = NULL;
548
549 rcu_read_unlock();
550
551 return last;
552}
553
554/* type and compressor must be null-terminated */
555static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
556{
557 struct zswap_pool *pool;
558
559 assert_spin_locked(&zswap_pools_lock);
560
561 list_for_each_entry_rcu(pool, &zswap_pools, list) {
562 if (strcmp(pool->tfm_name, compressor))
563 continue;
564 if (strcmp(zpool_get_type(pool->zpool), type))
565 continue;
566 /* if we can't get it, it's about to be destroyed */
567 if (!zswap_pool_get(pool))
568 continue;
569 return pool;
570 }
571
572 return NULL;
573}
574
575static void shrink_worker(struct work_struct *w)
576{
577 struct zswap_pool *pool = container_of(w, typeof(*pool),
578 shrink_work);
579
580 if (zpool_shrink(pool->zpool, 1, NULL))
581 zswap_reject_reclaim_fail++;
582 zswap_pool_put(pool);
583}
584
585static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
586{
587 struct zswap_pool *pool;
588 char name[38]; /* 'zswap' + 32 char (max) num + \0 */
589 gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
590 int ret;
591
592 if (!zswap_has_pool) {
593 /* if either are unset, pool initialization failed, and we
594 * need both params to be set correctly before trying to
595 * create a pool.
596 */
597 if (!strcmp(type, ZSWAP_PARAM_UNSET))
598 return NULL;
599 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
600 return NULL;
601 }
602
603 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
604 if (!pool)
605 return NULL;
606
607 /* unique name for each pool specifically required by zsmalloc */
608 snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
609
610 pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
611 if (!pool->zpool) {
612 pr_err("%s zpool not available\n", type);
613 goto error;
614 }
615 pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
616
617 strscpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
618
619 pool->acomp_ctx = alloc_percpu(*pool->acomp_ctx);
620 if (!pool->acomp_ctx) {
621 pr_err("percpu alloc failed\n");
622 goto error;
623 }
624
625 ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
626 &pool->node);
627 if (ret)
628 goto error;
629 pr_debug("using %s compressor\n", pool->tfm_name);
630
631 /* being the current pool takes 1 ref; this func expects the
632 * caller to always add the new pool as the current pool
633 */
634 kref_init(&pool->kref);
635 INIT_LIST_HEAD(&pool->list);
636 INIT_WORK(&pool->shrink_work, shrink_worker);
637
638 zswap_pool_debug("created", pool);
639
640 return pool;
641
642error:
643 if (pool->acomp_ctx)
644 free_percpu(pool->acomp_ctx);
645 if (pool->zpool)
646 zpool_destroy_pool(pool->zpool);
647 kfree(pool);
648 return NULL;
649}
650
651static __init struct zswap_pool *__zswap_pool_create_fallback(void)
652{
653 bool has_comp, has_zpool;
654
655 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
656 if (!has_comp && strcmp(zswap_compressor,
657 CONFIG_ZSWAP_COMPRESSOR_DEFAULT)) {
658 pr_err("compressor %s not available, using default %s\n",
659 zswap_compressor, CONFIG_ZSWAP_COMPRESSOR_DEFAULT);
660 param_free_charp(&zswap_compressor);
661 zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
662 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
663 }
664 if (!has_comp) {
665 pr_err("default compressor %s not available\n",
666 zswap_compressor);
667 param_free_charp(&zswap_compressor);
668 zswap_compressor = ZSWAP_PARAM_UNSET;
669 }
670
671 has_zpool = zpool_has_pool(zswap_zpool_type);
672 if (!has_zpool && strcmp(zswap_zpool_type,
673 CONFIG_ZSWAP_ZPOOL_DEFAULT)) {
674 pr_err("zpool %s not available, using default %s\n",
675 zswap_zpool_type, CONFIG_ZSWAP_ZPOOL_DEFAULT);
676 param_free_charp(&zswap_zpool_type);
677 zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
678 has_zpool = zpool_has_pool(zswap_zpool_type);
679 }
680 if (!has_zpool) {
681 pr_err("default zpool %s not available\n",
682 zswap_zpool_type);
683 param_free_charp(&zswap_zpool_type);
684 zswap_zpool_type = ZSWAP_PARAM_UNSET;
685 }
686
687 if (!has_comp || !has_zpool)
688 return NULL;
689
690 return zswap_pool_create(zswap_zpool_type, zswap_compressor);
691}
692
693static void zswap_pool_destroy(struct zswap_pool *pool)
694{
695 zswap_pool_debug("destroying", pool);
696
697 cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
698 free_percpu(pool->acomp_ctx);
699 zpool_destroy_pool(pool->zpool);
700 kfree(pool);
701}
702
703static int __must_check zswap_pool_get(struct zswap_pool *pool)
704{
705 if (!pool)
706 return 0;
707
708 return kref_get_unless_zero(&pool->kref);
709}
710
711static void __zswap_pool_release(struct work_struct *work)
712{
713 struct zswap_pool *pool = container_of(work, typeof(*pool),
714 release_work);
715
716 synchronize_rcu();
717
718 /* nobody should have been able to get a kref... */
719 WARN_ON(kref_get_unless_zero(&pool->kref));
720
721 /* pool is now off zswap_pools list and has no references. */
722 zswap_pool_destroy(pool);
723}
724
725static void __zswap_pool_empty(struct kref *kref)
726{
727 struct zswap_pool *pool;
728
729 pool = container_of(kref, typeof(*pool), kref);
730
731 spin_lock(&zswap_pools_lock);
732
733 WARN_ON(pool == zswap_pool_current());
734
735 list_del_rcu(&pool->list);
736
737 INIT_WORK(&pool->release_work, __zswap_pool_release);
738 schedule_work(&pool->release_work);
739
740 spin_unlock(&zswap_pools_lock);
741}
742
743static void zswap_pool_put(struct zswap_pool *pool)
744{
745 kref_put(&pool->kref, __zswap_pool_empty);
746}
747
748/*********************************
749* param callbacks
750**********************************/
751
752/* val must be a null-terminated string */
753static int __zswap_param_set(const char *val, const struct kernel_param *kp,
754 char *type, char *compressor)
755{
756 struct zswap_pool *pool, *put_pool = NULL;
757 char *s = strstrip((char *)val);
758 int ret;
759
760 if (zswap_init_failed) {
761 pr_err("can't set param, initialization failed\n");
762 return -ENODEV;
763 }
764
765 /* no change required */
766 if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
767 return 0;
768
769 /* if this is load-time (pre-init) param setting,
770 * don't create a pool; that's done during init.
771 */
772 if (!zswap_init_started)
773 return param_set_charp(s, kp);
774
775 if (!type) {
776 if (!zpool_has_pool(s)) {
777 pr_err("zpool %s not available\n", s);
778 return -ENOENT;
779 }
780 type = s;
781 } else if (!compressor) {
782 if (!crypto_has_acomp(s, 0, 0)) {
783 pr_err("compressor %s not available\n", s);
784 return -ENOENT;
785 }
786 compressor = s;
787 } else {
788 WARN_ON(1);
789 return -EINVAL;
790 }
791
792 spin_lock(&zswap_pools_lock);
793
794 pool = zswap_pool_find_get(type, compressor);
795 if (pool) {
796 zswap_pool_debug("using existing", pool);
797 WARN_ON(pool == zswap_pool_current());
798 list_del_rcu(&pool->list);
799 }
800
801 spin_unlock(&zswap_pools_lock);
802
803 if (!pool)
804 pool = zswap_pool_create(type, compressor);
805
806 if (pool)
807 ret = param_set_charp(s, kp);
808 else
809 ret = -EINVAL;
810
811 spin_lock(&zswap_pools_lock);
812
813 if (!ret) {
814 put_pool = zswap_pool_current();
815 list_add_rcu(&pool->list, &zswap_pools);
816 zswap_has_pool = true;
817 } else if (pool) {
818 /* add the possibly pre-existing pool to the end of the pools
819 * list; if it's new (and empty) then it'll be removed and
820 * destroyed by the put after we drop the lock
821 */
822 list_add_tail_rcu(&pool->list, &zswap_pools);
823 put_pool = pool;
824 }
825
826 spin_unlock(&zswap_pools_lock);
827
828 if (!zswap_has_pool && !pool) {
829 /* if initial pool creation failed, and this pool creation also
830 * failed, maybe both compressor and zpool params were bad.
831 * Allow changing this param, so pool creation will succeed
832 * when the other param is changed. We already verified this
833 * param is ok in the zpool_has_pool() or crypto_has_acomp()
834 * checks above.
835 */
836 ret = param_set_charp(s, kp);
837 }
838
839 /* drop the ref from either the old current pool,
840 * or the new pool we failed to add
841 */
842 if (put_pool)
843 zswap_pool_put(put_pool);
844
845 return ret;
846}
847
848static int zswap_compressor_param_set(const char *val,
849 const struct kernel_param *kp)
850{
851 return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
852}
853
854static int zswap_zpool_param_set(const char *val,
855 const struct kernel_param *kp)
856{
857 return __zswap_param_set(val, kp, NULL, zswap_compressor);
858}
859
860static int zswap_enabled_param_set(const char *val,
861 const struct kernel_param *kp)
862{
863 if (zswap_init_failed) {
864 pr_err("can't enable, initialization failed\n");
865 return -ENODEV;
866 }
867 if (!zswap_has_pool && zswap_init_started) {
868 pr_err("can't enable, no pool configured\n");
869 return -ENODEV;
870 }
871
872 return param_set_bool(val, kp);
873}
874
875/*********************************
876* writeback code
877**********************************/
878/* return enum for zswap_get_swap_cache_page */
879enum zswap_get_swap_ret {
880 ZSWAP_SWAPCACHE_NEW,
881 ZSWAP_SWAPCACHE_EXIST,
882 ZSWAP_SWAPCACHE_FAIL,
883};
884
885/*
886 * zswap_get_swap_cache_page
887 *
888 * This is an adaption of read_swap_cache_async()
889 *
890 * This function tries to find a page with the given swap entry
891 * in the swapper_space address space (the swap cache). If the page
892 * is found, it is returned in retpage. Otherwise, a page is allocated,
893 * added to the swap cache, and returned in retpage.
894 *
895 * If success, the swap cache page is returned in retpage
896 * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
897 * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
898 * the new page is added to swapcache and locked
899 * Returns ZSWAP_SWAPCACHE_FAIL on error
900 */
901static int zswap_get_swap_cache_page(swp_entry_t entry,
902 struct page **retpage)
903{
904 bool page_was_allocated;
905
906 *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
907 NULL, 0, &page_was_allocated);
908 if (page_was_allocated)
909 return ZSWAP_SWAPCACHE_NEW;
910 if (!*retpage)
911 return ZSWAP_SWAPCACHE_FAIL;
912 return ZSWAP_SWAPCACHE_EXIST;
913}
914
915/*
916 * Attempts to free an entry by adding a page to the swap cache,
917 * decompressing the entry data into the page, and issuing a
918 * bio write to write the page back to the swap device.
919 *
920 * This can be thought of as a "resumed writeback" of the page
921 * to the swap device. We are basically resuming the same swap
922 * writeback path that was intercepted with the frontswap_store()
923 * in the first place. After the page has been decompressed into
924 * the swap cache, the compressed version stored by zswap can be
925 * freed.
926 */
927static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
928{
929 struct zswap_header *zhdr;
930 swp_entry_t swpentry;
931 struct zswap_tree *tree;
932 pgoff_t offset;
933 struct zswap_entry *entry;
934 struct page *page;
935 struct scatterlist input, output;
936 struct crypto_acomp_ctx *acomp_ctx;
937
938 u8 *src, *tmp = NULL;
939 unsigned int dlen;
940 int ret;
941 struct writeback_control wbc = {
942 .sync_mode = WB_SYNC_NONE,
943 };
944
945 if (!zpool_can_sleep_mapped(pool)) {
946 tmp = kmalloc(PAGE_SIZE, GFP_ATOMIC);
947 if (!tmp)
948 return -ENOMEM;
949 }
950
951 /* extract swpentry from data */
952 zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
953 swpentry = zhdr->swpentry; /* here */
954 tree = zswap_trees[swp_type(swpentry)];
955 offset = swp_offset(swpentry);
956
957 /* find and ref zswap entry */
958 spin_lock(&tree->lock);
959 entry = zswap_entry_find_get(&tree->rbroot, offset);
960 if (!entry) {
961 /* entry was invalidated */
962 spin_unlock(&tree->lock);
963 zpool_unmap_handle(pool, handle);
964 kfree(tmp);
965 return 0;
966 }
967 spin_unlock(&tree->lock);
968 BUG_ON(offset != entry->offset);
969
970 src = (u8 *)zhdr + sizeof(struct zswap_header);
971 if (!zpool_can_sleep_mapped(pool)) {
972 memcpy(tmp, src, entry->length);
973 src = tmp;
974 zpool_unmap_handle(pool, handle);
975 }
976
977 /* try to allocate swap cache page */
978 switch (zswap_get_swap_cache_page(swpentry, &page)) {
979 case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
980 ret = -ENOMEM;
981 goto fail;
982
983 case ZSWAP_SWAPCACHE_EXIST:
984 /* page is already in the swap cache, ignore for now */
985 put_page(page);
986 ret = -EEXIST;
987 goto fail;
988
989 case ZSWAP_SWAPCACHE_NEW: /* page is locked */
990 /* decompress */
991 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
992 dlen = PAGE_SIZE;
993
994 mutex_lock(acomp_ctx->mutex);
995 sg_init_one(&input, src, entry->length);
996 sg_init_table(&output, 1);
997 sg_set_page(&output, page, PAGE_SIZE, 0);
998 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
999 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1000 dlen = acomp_ctx->req->dlen;
1001 mutex_unlock(acomp_ctx->mutex);
1002
1003 BUG_ON(ret);
1004 BUG_ON(dlen != PAGE_SIZE);
1005
1006 /* page is up to date */
1007 SetPageUptodate(page);
1008 }
1009
1010 /* move it to the tail of the inactive list after end_writeback */
1011 SetPageReclaim(page);
1012
1013 /* start writeback */
1014 __swap_writepage(page, &wbc, end_swap_bio_write);
1015 put_page(page);
1016 zswap_written_back_pages++;
1017
1018 spin_lock(&tree->lock);
1019 /* drop local reference */
1020 zswap_entry_put(tree, entry);
1021
1022 /*
1023 * There are two possible situations for entry here:
1024 * (1) refcount is 1(normal case), entry is valid and on the tree
1025 * (2) refcount is 0, entry is freed and not on the tree
1026 * because invalidate happened during writeback
1027 * search the tree and free the entry if find entry
1028 */
1029 if (entry == zswap_rb_search(&tree->rbroot, offset))
1030 zswap_entry_put(tree, entry);
1031 spin_unlock(&tree->lock);
1032
1033 goto end;
1034
1035 /*
1036 * if we get here due to ZSWAP_SWAPCACHE_EXIST
1037 * a load may be happening concurrently.
1038 * it is safe and okay to not free the entry.
1039 * if we free the entry in the following put
1040 * it is also okay to return !0
1041 */
1042fail:
1043 spin_lock(&tree->lock);
1044 zswap_entry_put(tree, entry);
1045 spin_unlock(&tree->lock);
1046
1047end:
1048 if (zpool_can_sleep_mapped(pool))
1049 zpool_unmap_handle(pool, handle);
1050 else
1051 kfree(tmp);
1052
1053 return ret;
1054}
1055
1056static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
1057{
1058 unsigned int pos;
1059 unsigned long *page;
1060
1061 page = (unsigned long *)ptr;
1062 for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
1063 if (page[pos] != page[0])
1064 return 0;
1065 }
1066 *value = page[0];
1067 return 1;
1068}
1069
1070static void zswap_fill_page(void *ptr, unsigned long value)
1071{
1072 unsigned long *page;
1073
1074 page = (unsigned long *)ptr;
1075 memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
1076}
1077
1078/*********************************
1079* frontswap hooks
1080**********************************/
1081/* attempts to compress and store an single page */
1082static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1083 struct page *page)
1084{
1085 struct zswap_tree *tree = zswap_trees[type];
1086 struct zswap_entry *entry, *dupentry;
1087 struct scatterlist input, output;
1088 struct crypto_acomp_ctx *acomp_ctx;
1089 int ret;
1090 unsigned int hlen, dlen = PAGE_SIZE;
1091 unsigned long handle, value;
1092 char *buf;
1093 u8 *src, *dst;
1094 struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1095 gfp_t gfp;
1096
1097 /* THP isn't supported */
1098 if (PageTransHuge(page)) {
1099 ret = -EINVAL;
1100 goto reject;
1101 }
1102
1103 if (!zswap_enabled || !tree) {
1104 ret = -ENODEV;
1105 goto reject;
1106 }
1107
1108 /* reclaim space if needed */
1109 if (zswap_is_full()) {
1110 struct zswap_pool *pool;
1111
1112 zswap_pool_limit_hit++;
1113 zswap_pool_reached_full = true;
1114 pool = zswap_pool_last_get();
1115 if (pool)
1116 queue_work(shrink_wq, &pool->shrink_work);
1117 ret = -ENOMEM;
1118 goto reject;
1119 }
1120
1121 if (zswap_pool_reached_full) {
1122 if (!zswap_can_accept()) {
1123 ret = -ENOMEM;
1124 goto reject;
1125 } else
1126 zswap_pool_reached_full = false;
1127 }
1128
1129 /* allocate entry */
1130 entry = zswap_entry_cache_alloc(GFP_KERNEL);
1131 if (!entry) {
1132 zswap_reject_kmemcache_fail++;
1133 ret = -ENOMEM;
1134 goto reject;
1135 }
1136
1137 if (zswap_same_filled_pages_enabled) {
1138 src = kmap_atomic(page);
1139 if (zswap_is_page_same_filled(src, &value)) {
1140 kunmap_atomic(src);
1141 entry->offset = offset;
1142 entry->length = 0;
1143 entry->value = value;
1144 atomic_inc(&zswap_same_filled_pages);
1145 goto insert_entry;
1146 }
1147 kunmap_atomic(src);
1148 }
1149
1150 /* if entry is successfully added, it keeps the reference */
1151 entry->pool = zswap_pool_current_get();
1152 if (!entry->pool) {
1153 ret = -EINVAL;
1154 goto freepage;
1155 }
1156
1157 /* compress */
1158 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1159
1160 mutex_lock(acomp_ctx->mutex);
1161
1162 dst = acomp_ctx->dstmem;
1163 sg_init_table(&input, 1);
1164 sg_set_page(&input, page, PAGE_SIZE, 0);
1165
1166 /* zswap_dstmem is of size (PAGE_SIZE * 2). Reflect same in sg_list */
1167 sg_init_one(&output, dst, PAGE_SIZE * 2);
1168 acomp_request_set_params(acomp_ctx->req, &input, &output, PAGE_SIZE, dlen);
1169 /*
1170 * it maybe looks a little bit silly that we send an asynchronous request,
1171 * then wait for its completion synchronously. This makes the process look
1172 * synchronous in fact.
1173 * Theoretically, acomp supports users send multiple acomp requests in one
1174 * acomp instance, then get those requests done simultaneously. but in this
1175 * case, frontswap actually does store and load page by page, there is no
1176 * existing method to send the second page before the first page is done
1177 * in one thread doing frontswap.
1178 * but in different threads running on different cpu, we have different
1179 * acomp instance, so multiple threads can do (de)compression in parallel.
1180 */
1181 ret = crypto_wait_req(crypto_acomp_compress(acomp_ctx->req), &acomp_ctx->wait);
1182 dlen = acomp_ctx->req->dlen;
1183
1184 if (ret) {
1185 ret = -EINVAL;
1186 goto put_dstmem;
1187 }
1188
1189 /* store */
1190 hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1191 gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1192 if (zpool_malloc_support_movable(entry->pool->zpool))
1193 gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1194 ret = zpool_malloc(entry->pool->zpool, hlen + dlen, gfp, &handle);
1195 if (ret == -ENOSPC) {
1196 zswap_reject_compress_poor++;
1197 goto put_dstmem;
1198 }
1199 if (ret) {
1200 zswap_reject_alloc_fail++;
1201 goto put_dstmem;
1202 }
1203 buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_WO);
1204 memcpy(buf, &zhdr, hlen);
1205 memcpy(buf + hlen, dst, dlen);
1206 zpool_unmap_handle(entry->pool->zpool, handle);
1207 mutex_unlock(acomp_ctx->mutex);
1208
1209 /* populate entry */
1210 entry->offset = offset;
1211 entry->handle = handle;
1212 entry->length = dlen;
1213
1214insert_entry:
1215 /* map */
1216 spin_lock(&tree->lock);
1217 do {
1218 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1219 if (ret == -EEXIST) {
1220 zswap_duplicate_entry++;
1221 /* remove from rbtree */
1222 zswap_rb_erase(&tree->rbroot, dupentry);
1223 zswap_entry_put(tree, dupentry);
1224 }
1225 } while (ret == -EEXIST);
1226 spin_unlock(&tree->lock);
1227
1228 /* update stats */
1229 atomic_inc(&zswap_stored_pages);
1230 zswap_update_total_size();
1231
1232 return 0;
1233
1234put_dstmem:
1235 mutex_unlock(acomp_ctx->mutex);
1236 zswap_pool_put(entry->pool);
1237freepage:
1238 zswap_entry_cache_free(entry);
1239reject:
1240 return ret;
1241}
1242
1243/*
1244 * returns 0 if the page was successfully decompressed
1245 * return -1 on entry not found or error
1246*/
1247static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1248 struct page *page)
1249{
1250 struct zswap_tree *tree = zswap_trees[type];
1251 struct zswap_entry *entry;
1252 struct scatterlist input, output;
1253 struct crypto_acomp_ctx *acomp_ctx;
1254 u8 *src, *dst, *tmp;
1255 unsigned int dlen;
1256 int ret;
1257
1258 /* find */
1259 spin_lock(&tree->lock);
1260 entry = zswap_entry_find_get(&tree->rbroot, offset);
1261 if (!entry) {
1262 /* entry was written back */
1263 spin_unlock(&tree->lock);
1264 return -1;
1265 }
1266 spin_unlock(&tree->lock);
1267
1268 if (!entry->length) {
1269 dst = kmap_atomic(page);
1270 zswap_fill_page(dst, entry->value);
1271 kunmap_atomic(dst);
1272 ret = 0;
1273 goto freeentry;
1274 }
1275
1276 if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1277
1278 tmp = kmalloc(entry->length, GFP_ATOMIC);
1279 if (!tmp) {
1280 ret = -ENOMEM;
1281 goto freeentry;
1282 }
1283 }
1284
1285 /* decompress */
1286 dlen = PAGE_SIZE;
1287 src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1288 if (zpool_evictable(entry->pool->zpool))
1289 src += sizeof(struct zswap_header);
1290
1291 if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1292
1293 memcpy(tmp, src, entry->length);
1294 src = tmp;
1295
1296 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1297 }
1298
1299 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1300 mutex_lock(acomp_ctx->mutex);
1301 sg_init_one(&input, src, entry->length);
1302 sg_init_table(&output, 1);
1303 sg_set_page(&output, page, PAGE_SIZE, 0);
1304 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1305 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1306 mutex_unlock(acomp_ctx->mutex);
1307
1308 if (zpool_can_sleep_mapped(entry->pool->zpool))
1309 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1310 else
1311 kfree(tmp);
1312
1313 BUG_ON(ret);
1314
1315freeentry:
1316 spin_lock(&tree->lock);
1317 zswap_entry_put(tree, entry);
1318 spin_unlock(&tree->lock);
1319
1320 return ret;
1321}
1322
1323/* frees an entry in zswap */
1324static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1325{
1326 struct zswap_tree *tree = zswap_trees[type];
1327 struct zswap_entry *entry;
1328
1329 /* find */
1330 spin_lock(&tree->lock);
1331 entry = zswap_rb_search(&tree->rbroot, offset);
1332 if (!entry) {
1333 /* entry was written back */
1334 spin_unlock(&tree->lock);
1335 return;
1336 }
1337
1338 /* remove from rbtree */
1339 zswap_rb_erase(&tree->rbroot, entry);
1340
1341 /* drop the initial reference from entry creation */
1342 zswap_entry_put(tree, entry);
1343
1344 spin_unlock(&tree->lock);
1345}
1346
1347/* frees all zswap entries for the given swap type */
1348static void zswap_frontswap_invalidate_area(unsigned type)
1349{
1350 struct zswap_tree *tree = zswap_trees[type];
1351 struct zswap_entry *entry, *n;
1352
1353 if (!tree)
1354 return;
1355
1356 /* walk the tree and free everything */
1357 spin_lock(&tree->lock);
1358 rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1359 zswap_free_entry(entry);
1360 tree->rbroot = RB_ROOT;
1361 spin_unlock(&tree->lock);
1362 kfree(tree);
1363 zswap_trees[type] = NULL;
1364}
1365
1366static void zswap_frontswap_init(unsigned type)
1367{
1368 struct zswap_tree *tree;
1369
1370 tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1371 if (!tree) {
1372 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1373 return;
1374 }
1375
1376 tree->rbroot = RB_ROOT;
1377 spin_lock_init(&tree->lock);
1378 zswap_trees[type] = tree;
1379}
1380
1381static struct frontswap_ops zswap_frontswap_ops = {
1382 .store = zswap_frontswap_store,
1383 .load = zswap_frontswap_load,
1384 .invalidate_page = zswap_frontswap_invalidate_page,
1385 .invalidate_area = zswap_frontswap_invalidate_area,
1386 .init = zswap_frontswap_init
1387};
1388
1389/*********************************
1390* debugfs functions
1391**********************************/
1392#ifdef CONFIG_DEBUG_FS
1393#include <linux/debugfs.h>
1394
1395static struct dentry *zswap_debugfs_root;
1396
1397static int __init zswap_debugfs_init(void)
1398{
1399 if (!debugfs_initialized())
1400 return -ENODEV;
1401
1402 zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1403
1404 debugfs_create_u64("pool_limit_hit", 0444,
1405 zswap_debugfs_root, &zswap_pool_limit_hit);
1406 debugfs_create_u64("reject_reclaim_fail", 0444,
1407 zswap_debugfs_root, &zswap_reject_reclaim_fail);
1408 debugfs_create_u64("reject_alloc_fail", 0444,
1409 zswap_debugfs_root, &zswap_reject_alloc_fail);
1410 debugfs_create_u64("reject_kmemcache_fail", 0444,
1411 zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1412 debugfs_create_u64("reject_compress_poor", 0444,
1413 zswap_debugfs_root, &zswap_reject_compress_poor);
1414 debugfs_create_u64("written_back_pages", 0444,
1415 zswap_debugfs_root, &zswap_written_back_pages);
1416 debugfs_create_u64("duplicate_entry", 0444,
1417 zswap_debugfs_root, &zswap_duplicate_entry);
1418 debugfs_create_u64("pool_total_size", 0444,
1419 zswap_debugfs_root, &zswap_pool_total_size);
1420 debugfs_create_atomic_t("stored_pages", 0444,
1421 zswap_debugfs_root, &zswap_stored_pages);
1422 debugfs_create_atomic_t("same_filled_pages", 0444,
1423 zswap_debugfs_root, &zswap_same_filled_pages);
1424
1425 return 0;
1426}
1427#else
1428static int __init zswap_debugfs_init(void)
1429{
1430 return 0;
1431}
1432#endif
1433
1434/*********************************
1435* module init and exit
1436**********************************/
1437static int __init init_zswap(void)
1438{
1439 struct zswap_pool *pool;
1440 int ret;
1441
1442 zswap_init_started = true;
1443
1444 if (zswap_entry_cache_create()) {
1445 pr_err("entry cache creation failed\n");
1446 goto cache_fail;
1447 }
1448
1449 ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1450 zswap_dstmem_prepare, zswap_dstmem_dead);
1451 if (ret) {
1452 pr_err("dstmem alloc failed\n");
1453 goto dstmem_fail;
1454 }
1455
1456 ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1457 "mm/zswap_pool:prepare",
1458 zswap_cpu_comp_prepare,
1459 zswap_cpu_comp_dead);
1460 if (ret)
1461 goto hp_fail;
1462
1463 pool = __zswap_pool_create_fallback();
1464 if (pool) {
1465 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1466 zpool_get_type(pool->zpool));
1467 list_add(&pool->list, &zswap_pools);
1468 zswap_has_pool = true;
1469 } else {
1470 pr_err("pool creation failed\n");
1471 zswap_enabled = false;
1472 }
1473
1474 shrink_wq = create_workqueue("zswap-shrink");
1475 if (!shrink_wq)
1476 goto fallback_fail;
1477
1478 frontswap_register_ops(&zswap_frontswap_ops);
1479 if (zswap_debugfs_init())
1480 pr_warn("debugfs initialization failed\n");
1481 return 0;
1482
1483fallback_fail:
1484 if (pool)
1485 zswap_pool_destroy(pool);
1486hp_fail:
1487 cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1488dstmem_fail:
1489 zswap_entry_cache_destroy();
1490cache_fail:
1491 /* if built-in, we aren't unloaded on failure; don't allow use */
1492 zswap_init_failed = true;
1493 zswap_enabled = false;
1494 return -ENOMEM;
1495}
1496/* must be late so crypto has time to come up */
1497late_initcall(init_zswap);
1498
1499MODULE_LICENSE("GPL");
1500MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1501MODULE_DESCRIPTION("Compressed cache for swap pages");
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/scatterlist.h>
28#include <linux/mempool.h>
29#include <linux/zpool.h>
30#include <crypto/acompress.h>
31
32#include <linux/mm_types.h>
33#include <linux/page-flags.h>
34#include <linux/swapops.h>
35#include <linux/writeback.h>
36#include <linux/pagemap.h>
37#include <linux/workqueue.h>
38
39#include "swap.h"
40
41/*********************************
42* statistics
43**********************************/
44/* Total bytes used by the compressed storage */
45u64 zswap_pool_total_size;
46/* The number of compressed pages currently stored in zswap */
47atomic_t zswap_stored_pages = ATOMIC_INIT(0);
48/* The number of same-value filled pages currently stored in zswap */
49static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
50
51/*
52 * The statistics below are not protected from concurrent access for
53 * performance reasons so they may not be a 100% accurate. However,
54 * they do provide useful information on roughly how many times a
55 * certain event is occurring.
56*/
57
58/* Pool limit was hit (see zswap_max_pool_percent) */
59static u64 zswap_pool_limit_hit;
60/* Pages written back when pool limit was reached */
61static u64 zswap_written_back_pages;
62/* Store failed due to a reclaim failure after pool limit was reached */
63static u64 zswap_reject_reclaim_fail;
64/* Compressed page was too big for the allocator to (optimally) store */
65static u64 zswap_reject_compress_poor;
66/* Store failed because underlying allocator could not get memory */
67static u64 zswap_reject_alloc_fail;
68/* Store failed because the entry metadata could not be allocated (rare) */
69static u64 zswap_reject_kmemcache_fail;
70/* Duplicate store was encountered (rare) */
71static u64 zswap_duplicate_entry;
72
73/* Shrinker work queue */
74static struct workqueue_struct *shrink_wq;
75/* Pool limit was hit, we need to calm down */
76static bool zswap_pool_reached_full;
77
78/*********************************
79* tunables
80**********************************/
81
82#define ZSWAP_PARAM_UNSET ""
83
84/* Enable/disable zswap */
85static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
86static int zswap_enabled_param_set(const char *,
87 const struct kernel_param *);
88static const struct kernel_param_ops zswap_enabled_param_ops = {
89 .set = zswap_enabled_param_set,
90 .get = param_get_bool,
91};
92module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
93
94/* Crypto compressor to use */
95static char *zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
96static int zswap_compressor_param_set(const char *,
97 const struct kernel_param *);
98static const struct kernel_param_ops zswap_compressor_param_ops = {
99 .set = zswap_compressor_param_set,
100 .get = param_get_charp,
101 .free = param_free_charp,
102};
103module_param_cb(compressor, &zswap_compressor_param_ops,
104 &zswap_compressor, 0644);
105
106/* Compressed storage zpool to use */
107static char *zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
108static int zswap_zpool_param_set(const char *, const struct kernel_param *);
109static const struct kernel_param_ops zswap_zpool_param_ops = {
110 .set = zswap_zpool_param_set,
111 .get = param_get_charp,
112 .free = param_free_charp,
113};
114module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
115
116/* The maximum percentage of memory that the compressed pool can occupy */
117static unsigned int zswap_max_pool_percent = 20;
118module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
119
120/* The threshold for accepting new pages after the max_pool_percent was hit */
121static unsigned int zswap_accept_thr_percent = 90; /* of max pool size */
122module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
123 uint, 0644);
124
125/*
126 * Enable/disable handling same-value filled pages (enabled by default).
127 * If disabled every page is considered non-same-value filled.
128 */
129static bool zswap_same_filled_pages_enabled = true;
130module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
131 bool, 0644);
132
133/* Enable/disable handling non-same-value filled pages (enabled by default) */
134static bool zswap_non_same_filled_pages_enabled = true;
135module_param_named(non_same_filled_pages_enabled, zswap_non_same_filled_pages_enabled,
136 bool, 0644);
137
138/*********************************
139* data structures
140**********************************/
141
142struct crypto_acomp_ctx {
143 struct crypto_acomp *acomp;
144 struct acomp_req *req;
145 struct crypto_wait wait;
146 u8 *dstmem;
147 struct mutex *mutex;
148};
149
150struct zswap_pool {
151 struct zpool *zpool;
152 struct crypto_acomp_ctx __percpu *acomp_ctx;
153 struct kref kref;
154 struct list_head list;
155 struct work_struct release_work;
156 struct work_struct shrink_work;
157 struct hlist_node node;
158 char tfm_name[CRYPTO_MAX_ALG_NAME];
159};
160
161/*
162 * struct zswap_entry
163 *
164 * This structure contains the metadata for tracking a single compressed
165 * page within zswap.
166 *
167 * rbnode - links the entry into red-black tree for the appropriate swap type
168 * offset - the swap offset for the entry. Index into the red-black tree.
169 * refcount - the number of outstanding reference to the entry. This is needed
170 * to protect against premature freeing of the entry by code
171 * concurrent calls to load, invalidate, and writeback. The lock
172 * for the zswap_tree structure that contains the entry must
173 * be held while changing the refcount. Since the lock must
174 * be held, there is no reason to also make refcount atomic.
175 * length - the length in bytes of the compressed page data. Needed during
176 * decompression. For a same value filled page length is 0.
177 * pool - the zswap_pool the entry's data is in
178 * handle - zpool allocation handle that stores the compressed page data
179 * value - value of the same-value filled pages which have same content
180 */
181struct zswap_entry {
182 struct rb_node rbnode;
183 pgoff_t offset;
184 int refcount;
185 unsigned int length;
186 struct zswap_pool *pool;
187 union {
188 unsigned long handle;
189 unsigned long value;
190 };
191 struct obj_cgroup *objcg;
192};
193
194struct zswap_header {
195 swp_entry_t swpentry;
196};
197
198/*
199 * The tree lock in the zswap_tree struct protects a few things:
200 * - the rbtree
201 * - the refcount field of each entry in the tree
202 */
203struct zswap_tree {
204 struct rb_root rbroot;
205 spinlock_t lock;
206};
207
208static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
209
210/* RCU-protected iteration */
211static LIST_HEAD(zswap_pools);
212/* protects zswap_pools list modification */
213static DEFINE_SPINLOCK(zswap_pools_lock);
214/* pool counter to provide unique names to zpool */
215static atomic_t zswap_pools_count = ATOMIC_INIT(0);
216
217/* used by param callback function */
218static bool zswap_init_started;
219
220/* fatal error during init */
221static bool zswap_init_failed;
222
223/* init completed, but couldn't create the initial pool */
224static bool zswap_has_pool;
225
226/*********************************
227* helpers and fwd declarations
228**********************************/
229
230#define zswap_pool_debug(msg, p) \
231 pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name, \
232 zpool_get_type((p)->zpool))
233
234static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
235static int zswap_pool_get(struct zswap_pool *pool);
236static void zswap_pool_put(struct zswap_pool *pool);
237
238static const struct zpool_ops zswap_zpool_ops = {
239 .evict = zswap_writeback_entry
240};
241
242static bool zswap_is_full(void)
243{
244 return totalram_pages() * zswap_max_pool_percent / 100 <
245 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
246}
247
248static bool zswap_can_accept(void)
249{
250 return totalram_pages() * zswap_accept_thr_percent / 100 *
251 zswap_max_pool_percent / 100 >
252 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
253}
254
255static void zswap_update_total_size(void)
256{
257 struct zswap_pool *pool;
258 u64 total = 0;
259
260 rcu_read_lock();
261
262 list_for_each_entry_rcu(pool, &zswap_pools, list)
263 total += zpool_get_total_size(pool->zpool);
264
265 rcu_read_unlock();
266
267 zswap_pool_total_size = total;
268}
269
270/*********************************
271* zswap entry functions
272**********************************/
273static struct kmem_cache *zswap_entry_cache;
274
275static int __init zswap_entry_cache_create(void)
276{
277 zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
278 return zswap_entry_cache == NULL;
279}
280
281static void __init zswap_entry_cache_destroy(void)
282{
283 kmem_cache_destroy(zswap_entry_cache);
284}
285
286static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
287{
288 struct zswap_entry *entry;
289 entry = kmem_cache_alloc(zswap_entry_cache, gfp);
290 if (!entry)
291 return NULL;
292 entry->refcount = 1;
293 RB_CLEAR_NODE(&entry->rbnode);
294 return entry;
295}
296
297static void zswap_entry_cache_free(struct zswap_entry *entry)
298{
299 kmem_cache_free(zswap_entry_cache, entry);
300}
301
302/*********************************
303* rbtree functions
304**********************************/
305static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
306{
307 struct rb_node *node = root->rb_node;
308 struct zswap_entry *entry;
309
310 while (node) {
311 entry = rb_entry(node, struct zswap_entry, rbnode);
312 if (entry->offset > offset)
313 node = node->rb_left;
314 else if (entry->offset < offset)
315 node = node->rb_right;
316 else
317 return entry;
318 }
319 return NULL;
320}
321
322/*
323 * In the case that a entry with the same offset is found, a pointer to
324 * the existing entry is stored in dupentry and the function returns -EEXIST
325 */
326static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
327 struct zswap_entry **dupentry)
328{
329 struct rb_node **link = &root->rb_node, *parent = NULL;
330 struct zswap_entry *myentry;
331
332 while (*link) {
333 parent = *link;
334 myentry = rb_entry(parent, struct zswap_entry, rbnode);
335 if (myentry->offset > entry->offset)
336 link = &(*link)->rb_left;
337 else if (myentry->offset < entry->offset)
338 link = &(*link)->rb_right;
339 else {
340 *dupentry = myentry;
341 return -EEXIST;
342 }
343 }
344 rb_link_node(&entry->rbnode, parent, link);
345 rb_insert_color(&entry->rbnode, root);
346 return 0;
347}
348
349static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
350{
351 if (!RB_EMPTY_NODE(&entry->rbnode)) {
352 rb_erase(&entry->rbnode, root);
353 RB_CLEAR_NODE(&entry->rbnode);
354 }
355}
356
357/*
358 * Carries out the common pattern of freeing and entry's zpool allocation,
359 * freeing the entry itself, and decrementing the number of stored pages.
360 */
361static void zswap_free_entry(struct zswap_entry *entry)
362{
363 if (entry->objcg) {
364 obj_cgroup_uncharge_zswap(entry->objcg, entry->length);
365 obj_cgroup_put(entry->objcg);
366 }
367 if (!entry->length)
368 atomic_dec(&zswap_same_filled_pages);
369 else {
370 zpool_free(entry->pool->zpool, entry->handle);
371 zswap_pool_put(entry->pool);
372 }
373 zswap_entry_cache_free(entry);
374 atomic_dec(&zswap_stored_pages);
375 zswap_update_total_size();
376}
377
378/* caller must hold the tree lock */
379static void zswap_entry_get(struct zswap_entry *entry)
380{
381 entry->refcount++;
382}
383
384/* caller must hold the tree lock
385* remove from the tree and free it, if nobody reference the entry
386*/
387static void zswap_entry_put(struct zswap_tree *tree,
388 struct zswap_entry *entry)
389{
390 int refcount = --entry->refcount;
391
392 BUG_ON(refcount < 0);
393 if (refcount == 0) {
394 zswap_rb_erase(&tree->rbroot, entry);
395 zswap_free_entry(entry);
396 }
397}
398
399/* caller must hold the tree lock */
400static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
401 pgoff_t offset)
402{
403 struct zswap_entry *entry;
404
405 entry = zswap_rb_search(root, offset);
406 if (entry)
407 zswap_entry_get(entry);
408
409 return entry;
410}
411
412/*********************************
413* per-cpu code
414**********************************/
415static DEFINE_PER_CPU(u8 *, zswap_dstmem);
416/*
417 * If users dynamically change the zpool type and compressor at runtime, i.e.
418 * zswap is running, zswap can have more than one zpool on one cpu, but they
419 * are sharing dtsmem. So we need this mutex to be per-cpu.
420 */
421static DEFINE_PER_CPU(struct mutex *, zswap_mutex);
422
423static int zswap_dstmem_prepare(unsigned int cpu)
424{
425 struct mutex *mutex;
426 u8 *dst;
427
428 dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
429 if (!dst)
430 return -ENOMEM;
431
432 mutex = kmalloc_node(sizeof(*mutex), GFP_KERNEL, cpu_to_node(cpu));
433 if (!mutex) {
434 kfree(dst);
435 return -ENOMEM;
436 }
437
438 mutex_init(mutex);
439 per_cpu(zswap_dstmem, cpu) = dst;
440 per_cpu(zswap_mutex, cpu) = mutex;
441 return 0;
442}
443
444static int zswap_dstmem_dead(unsigned int cpu)
445{
446 struct mutex *mutex;
447 u8 *dst;
448
449 mutex = per_cpu(zswap_mutex, cpu);
450 kfree(mutex);
451 per_cpu(zswap_mutex, cpu) = NULL;
452
453 dst = per_cpu(zswap_dstmem, cpu);
454 kfree(dst);
455 per_cpu(zswap_dstmem, cpu) = NULL;
456
457 return 0;
458}
459
460static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
461{
462 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
463 struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
464 struct crypto_acomp *acomp;
465 struct acomp_req *req;
466
467 acomp = crypto_alloc_acomp_node(pool->tfm_name, 0, 0, cpu_to_node(cpu));
468 if (IS_ERR(acomp)) {
469 pr_err("could not alloc crypto acomp %s : %ld\n",
470 pool->tfm_name, PTR_ERR(acomp));
471 return PTR_ERR(acomp);
472 }
473 acomp_ctx->acomp = acomp;
474
475 req = acomp_request_alloc(acomp_ctx->acomp);
476 if (!req) {
477 pr_err("could not alloc crypto acomp_request %s\n",
478 pool->tfm_name);
479 crypto_free_acomp(acomp_ctx->acomp);
480 return -ENOMEM;
481 }
482 acomp_ctx->req = req;
483
484 crypto_init_wait(&acomp_ctx->wait);
485 /*
486 * if the backend of acomp is async zip, crypto_req_done() will wakeup
487 * crypto_wait_req(); if the backend of acomp is scomp, the callback
488 * won't be called, crypto_wait_req() will return without blocking.
489 */
490 acomp_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
491 crypto_req_done, &acomp_ctx->wait);
492
493 acomp_ctx->mutex = per_cpu(zswap_mutex, cpu);
494 acomp_ctx->dstmem = per_cpu(zswap_dstmem, cpu);
495
496 return 0;
497}
498
499static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
500{
501 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
502 struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
503
504 if (!IS_ERR_OR_NULL(acomp_ctx)) {
505 if (!IS_ERR_OR_NULL(acomp_ctx->req))
506 acomp_request_free(acomp_ctx->req);
507 if (!IS_ERR_OR_NULL(acomp_ctx->acomp))
508 crypto_free_acomp(acomp_ctx->acomp);
509 }
510
511 return 0;
512}
513
514/*********************************
515* pool functions
516**********************************/
517
518static struct zswap_pool *__zswap_pool_current(void)
519{
520 struct zswap_pool *pool;
521
522 pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
523 WARN_ONCE(!pool && zswap_has_pool,
524 "%s: no page storage pool!\n", __func__);
525
526 return pool;
527}
528
529static struct zswap_pool *zswap_pool_current(void)
530{
531 assert_spin_locked(&zswap_pools_lock);
532
533 return __zswap_pool_current();
534}
535
536static struct zswap_pool *zswap_pool_current_get(void)
537{
538 struct zswap_pool *pool;
539
540 rcu_read_lock();
541
542 pool = __zswap_pool_current();
543 if (!zswap_pool_get(pool))
544 pool = NULL;
545
546 rcu_read_unlock();
547
548 return pool;
549}
550
551static struct zswap_pool *zswap_pool_last_get(void)
552{
553 struct zswap_pool *pool, *last = NULL;
554
555 rcu_read_lock();
556
557 list_for_each_entry_rcu(pool, &zswap_pools, list)
558 last = pool;
559 WARN_ONCE(!last && zswap_has_pool,
560 "%s: no page storage pool!\n", __func__);
561 if (!zswap_pool_get(last))
562 last = NULL;
563
564 rcu_read_unlock();
565
566 return last;
567}
568
569/* type and compressor must be null-terminated */
570static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
571{
572 struct zswap_pool *pool;
573
574 assert_spin_locked(&zswap_pools_lock);
575
576 list_for_each_entry_rcu(pool, &zswap_pools, list) {
577 if (strcmp(pool->tfm_name, compressor))
578 continue;
579 if (strcmp(zpool_get_type(pool->zpool), type))
580 continue;
581 /* if we can't get it, it's about to be destroyed */
582 if (!zswap_pool_get(pool))
583 continue;
584 return pool;
585 }
586
587 return NULL;
588}
589
590static void shrink_worker(struct work_struct *w)
591{
592 struct zswap_pool *pool = container_of(w, typeof(*pool),
593 shrink_work);
594
595 if (zpool_shrink(pool->zpool, 1, NULL))
596 zswap_reject_reclaim_fail++;
597 zswap_pool_put(pool);
598}
599
600static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
601{
602 struct zswap_pool *pool;
603 char name[38]; /* 'zswap' + 32 char (max) num + \0 */
604 gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
605 int ret;
606
607 if (!zswap_has_pool) {
608 /* if either are unset, pool initialization failed, and we
609 * need both params to be set correctly before trying to
610 * create a pool.
611 */
612 if (!strcmp(type, ZSWAP_PARAM_UNSET))
613 return NULL;
614 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
615 return NULL;
616 }
617
618 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
619 if (!pool)
620 return NULL;
621
622 /* unique name for each pool specifically required by zsmalloc */
623 snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
624
625 pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
626 if (!pool->zpool) {
627 pr_err("%s zpool not available\n", type);
628 goto error;
629 }
630 pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
631
632 strscpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
633
634 pool->acomp_ctx = alloc_percpu(*pool->acomp_ctx);
635 if (!pool->acomp_ctx) {
636 pr_err("percpu alloc failed\n");
637 goto error;
638 }
639
640 ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
641 &pool->node);
642 if (ret)
643 goto error;
644 pr_debug("using %s compressor\n", pool->tfm_name);
645
646 /* being the current pool takes 1 ref; this func expects the
647 * caller to always add the new pool as the current pool
648 */
649 kref_init(&pool->kref);
650 INIT_LIST_HEAD(&pool->list);
651 INIT_WORK(&pool->shrink_work, shrink_worker);
652
653 zswap_pool_debug("created", pool);
654
655 return pool;
656
657error:
658 if (pool->acomp_ctx)
659 free_percpu(pool->acomp_ctx);
660 if (pool->zpool)
661 zpool_destroy_pool(pool->zpool);
662 kfree(pool);
663 return NULL;
664}
665
666static __init struct zswap_pool *__zswap_pool_create_fallback(void)
667{
668 bool has_comp, has_zpool;
669
670 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
671 if (!has_comp && strcmp(zswap_compressor,
672 CONFIG_ZSWAP_COMPRESSOR_DEFAULT)) {
673 pr_err("compressor %s not available, using default %s\n",
674 zswap_compressor, CONFIG_ZSWAP_COMPRESSOR_DEFAULT);
675 param_free_charp(&zswap_compressor);
676 zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
677 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
678 }
679 if (!has_comp) {
680 pr_err("default compressor %s not available\n",
681 zswap_compressor);
682 param_free_charp(&zswap_compressor);
683 zswap_compressor = ZSWAP_PARAM_UNSET;
684 }
685
686 has_zpool = zpool_has_pool(zswap_zpool_type);
687 if (!has_zpool && strcmp(zswap_zpool_type,
688 CONFIG_ZSWAP_ZPOOL_DEFAULT)) {
689 pr_err("zpool %s not available, using default %s\n",
690 zswap_zpool_type, CONFIG_ZSWAP_ZPOOL_DEFAULT);
691 param_free_charp(&zswap_zpool_type);
692 zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
693 has_zpool = zpool_has_pool(zswap_zpool_type);
694 }
695 if (!has_zpool) {
696 pr_err("default zpool %s not available\n",
697 zswap_zpool_type);
698 param_free_charp(&zswap_zpool_type);
699 zswap_zpool_type = ZSWAP_PARAM_UNSET;
700 }
701
702 if (!has_comp || !has_zpool)
703 return NULL;
704
705 return zswap_pool_create(zswap_zpool_type, zswap_compressor);
706}
707
708static void zswap_pool_destroy(struct zswap_pool *pool)
709{
710 zswap_pool_debug("destroying", pool);
711
712 cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
713 free_percpu(pool->acomp_ctx);
714 zpool_destroy_pool(pool->zpool);
715 kfree(pool);
716}
717
718static int __must_check zswap_pool_get(struct zswap_pool *pool)
719{
720 if (!pool)
721 return 0;
722
723 return kref_get_unless_zero(&pool->kref);
724}
725
726static void __zswap_pool_release(struct work_struct *work)
727{
728 struct zswap_pool *pool = container_of(work, typeof(*pool),
729 release_work);
730
731 synchronize_rcu();
732
733 /* nobody should have been able to get a kref... */
734 WARN_ON(kref_get_unless_zero(&pool->kref));
735
736 /* pool is now off zswap_pools list and has no references. */
737 zswap_pool_destroy(pool);
738}
739
740static void __zswap_pool_empty(struct kref *kref)
741{
742 struct zswap_pool *pool;
743
744 pool = container_of(kref, typeof(*pool), kref);
745
746 spin_lock(&zswap_pools_lock);
747
748 WARN_ON(pool == zswap_pool_current());
749
750 list_del_rcu(&pool->list);
751
752 INIT_WORK(&pool->release_work, __zswap_pool_release);
753 schedule_work(&pool->release_work);
754
755 spin_unlock(&zswap_pools_lock);
756}
757
758static void zswap_pool_put(struct zswap_pool *pool)
759{
760 kref_put(&pool->kref, __zswap_pool_empty);
761}
762
763/*********************************
764* param callbacks
765**********************************/
766
767/* val must be a null-terminated string */
768static int __zswap_param_set(const char *val, const struct kernel_param *kp,
769 char *type, char *compressor)
770{
771 struct zswap_pool *pool, *put_pool = NULL;
772 char *s = strstrip((char *)val);
773 int ret;
774
775 if (zswap_init_failed) {
776 pr_err("can't set param, initialization failed\n");
777 return -ENODEV;
778 }
779
780 /* no change required */
781 if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
782 return 0;
783
784 /* if this is load-time (pre-init) param setting,
785 * don't create a pool; that's done during init.
786 */
787 if (!zswap_init_started)
788 return param_set_charp(s, kp);
789
790 if (!type) {
791 if (!zpool_has_pool(s)) {
792 pr_err("zpool %s not available\n", s);
793 return -ENOENT;
794 }
795 type = s;
796 } else if (!compressor) {
797 if (!crypto_has_acomp(s, 0, 0)) {
798 pr_err("compressor %s not available\n", s);
799 return -ENOENT;
800 }
801 compressor = s;
802 } else {
803 WARN_ON(1);
804 return -EINVAL;
805 }
806
807 spin_lock(&zswap_pools_lock);
808
809 pool = zswap_pool_find_get(type, compressor);
810 if (pool) {
811 zswap_pool_debug("using existing", pool);
812 WARN_ON(pool == zswap_pool_current());
813 list_del_rcu(&pool->list);
814 }
815
816 spin_unlock(&zswap_pools_lock);
817
818 if (!pool)
819 pool = zswap_pool_create(type, compressor);
820
821 if (pool)
822 ret = param_set_charp(s, kp);
823 else
824 ret = -EINVAL;
825
826 spin_lock(&zswap_pools_lock);
827
828 if (!ret) {
829 put_pool = zswap_pool_current();
830 list_add_rcu(&pool->list, &zswap_pools);
831 zswap_has_pool = true;
832 } else if (pool) {
833 /* add the possibly pre-existing pool to the end of the pools
834 * list; if it's new (and empty) then it'll be removed and
835 * destroyed by the put after we drop the lock
836 */
837 list_add_tail_rcu(&pool->list, &zswap_pools);
838 put_pool = pool;
839 }
840
841 spin_unlock(&zswap_pools_lock);
842
843 if (!zswap_has_pool && !pool) {
844 /* if initial pool creation failed, and this pool creation also
845 * failed, maybe both compressor and zpool params were bad.
846 * Allow changing this param, so pool creation will succeed
847 * when the other param is changed. We already verified this
848 * param is ok in the zpool_has_pool() or crypto_has_acomp()
849 * checks above.
850 */
851 ret = param_set_charp(s, kp);
852 }
853
854 /* drop the ref from either the old current pool,
855 * or the new pool we failed to add
856 */
857 if (put_pool)
858 zswap_pool_put(put_pool);
859
860 return ret;
861}
862
863static int zswap_compressor_param_set(const char *val,
864 const struct kernel_param *kp)
865{
866 return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
867}
868
869static int zswap_zpool_param_set(const char *val,
870 const struct kernel_param *kp)
871{
872 return __zswap_param_set(val, kp, NULL, zswap_compressor);
873}
874
875static int zswap_enabled_param_set(const char *val,
876 const struct kernel_param *kp)
877{
878 if (zswap_init_failed) {
879 pr_err("can't enable, initialization failed\n");
880 return -ENODEV;
881 }
882 if (!zswap_has_pool && zswap_init_started) {
883 pr_err("can't enable, no pool configured\n");
884 return -ENODEV;
885 }
886
887 return param_set_bool(val, kp);
888}
889
890/*********************************
891* writeback code
892**********************************/
893/* return enum for zswap_get_swap_cache_page */
894enum zswap_get_swap_ret {
895 ZSWAP_SWAPCACHE_NEW,
896 ZSWAP_SWAPCACHE_EXIST,
897 ZSWAP_SWAPCACHE_FAIL,
898};
899
900/*
901 * zswap_get_swap_cache_page
902 *
903 * This is an adaption of read_swap_cache_async()
904 *
905 * This function tries to find a page with the given swap entry
906 * in the swapper_space address space (the swap cache). If the page
907 * is found, it is returned in retpage. Otherwise, a page is allocated,
908 * added to the swap cache, and returned in retpage.
909 *
910 * If success, the swap cache page is returned in retpage
911 * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
912 * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
913 * the new page is added to swapcache and locked
914 * Returns ZSWAP_SWAPCACHE_FAIL on error
915 */
916static int zswap_get_swap_cache_page(swp_entry_t entry,
917 struct page **retpage)
918{
919 bool page_was_allocated;
920
921 *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
922 NULL, 0, &page_was_allocated);
923 if (page_was_allocated)
924 return ZSWAP_SWAPCACHE_NEW;
925 if (!*retpage)
926 return ZSWAP_SWAPCACHE_FAIL;
927 return ZSWAP_SWAPCACHE_EXIST;
928}
929
930/*
931 * Attempts to free an entry by adding a page to the swap cache,
932 * decompressing the entry data into the page, and issuing a
933 * bio write to write the page back to the swap device.
934 *
935 * This can be thought of as a "resumed writeback" of the page
936 * to the swap device. We are basically resuming the same swap
937 * writeback path that was intercepted with the frontswap_store()
938 * in the first place. After the page has been decompressed into
939 * the swap cache, the compressed version stored by zswap can be
940 * freed.
941 */
942static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
943{
944 struct zswap_header *zhdr;
945 swp_entry_t swpentry;
946 struct zswap_tree *tree;
947 pgoff_t offset;
948 struct zswap_entry *entry;
949 struct page *page;
950 struct scatterlist input, output;
951 struct crypto_acomp_ctx *acomp_ctx;
952
953 u8 *src, *tmp = NULL;
954 unsigned int dlen;
955 int ret;
956 struct writeback_control wbc = {
957 .sync_mode = WB_SYNC_NONE,
958 };
959
960 if (!zpool_can_sleep_mapped(pool)) {
961 tmp = kmalloc(PAGE_SIZE, GFP_KERNEL);
962 if (!tmp)
963 return -ENOMEM;
964 }
965
966 /* extract swpentry from data */
967 zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
968 swpentry = zhdr->swpentry; /* here */
969 tree = zswap_trees[swp_type(swpentry)];
970 offset = swp_offset(swpentry);
971 zpool_unmap_handle(pool, handle);
972
973 /* find and ref zswap entry */
974 spin_lock(&tree->lock);
975 entry = zswap_entry_find_get(&tree->rbroot, offset);
976 if (!entry) {
977 /* entry was invalidated */
978 spin_unlock(&tree->lock);
979 kfree(tmp);
980 return 0;
981 }
982 spin_unlock(&tree->lock);
983 BUG_ON(offset != entry->offset);
984
985 /* try to allocate swap cache page */
986 switch (zswap_get_swap_cache_page(swpentry, &page)) {
987 case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
988 ret = -ENOMEM;
989 goto fail;
990
991 case ZSWAP_SWAPCACHE_EXIST:
992 /* page is already in the swap cache, ignore for now */
993 put_page(page);
994 ret = -EEXIST;
995 goto fail;
996
997 case ZSWAP_SWAPCACHE_NEW: /* page is locked */
998 /* decompress */
999 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1000 dlen = PAGE_SIZE;
1001
1002 zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
1003 src = (u8 *)zhdr + sizeof(struct zswap_header);
1004 if (!zpool_can_sleep_mapped(pool)) {
1005 memcpy(tmp, src, entry->length);
1006 src = tmp;
1007 zpool_unmap_handle(pool, handle);
1008 }
1009
1010 mutex_lock(acomp_ctx->mutex);
1011 sg_init_one(&input, src, entry->length);
1012 sg_init_table(&output, 1);
1013 sg_set_page(&output, page, PAGE_SIZE, 0);
1014 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1015 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1016 dlen = acomp_ctx->req->dlen;
1017 mutex_unlock(acomp_ctx->mutex);
1018
1019 if (!zpool_can_sleep_mapped(pool))
1020 kfree(tmp);
1021 else
1022 zpool_unmap_handle(pool, handle);
1023
1024 BUG_ON(ret);
1025 BUG_ON(dlen != PAGE_SIZE);
1026
1027 /* page is up to date */
1028 SetPageUptodate(page);
1029 }
1030
1031 /* move it to the tail of the inactive list after end_writeback */
1032 SetPageReclaim(page);
1033
1034 /* start writeback */
1035 __swap_writepage(page, &wbc);
1036 put_page(page);
1037 zswap_written_back_pages++;
1038
1039 spin_lock(&tree->lock);
1040 /* drop local reference */
1041 zswap_entry_put(tree, entry);
1042
1043 /*
1044 * There are two possible situations for entry here:
1045 * (1) refcount is 1(normal case), entry is valid and on the tree
1046 * (2) refcount is 0, entry is freed and not on the tree
1047 * because invalidate happened during writeback
1048 * search the tree and free the entry if find entry
1049 */
1050 if (entry == zswap_rb_search(&tree->rbroot, offset))
1051 zswap_entry_put(tree, entry);
1052 spin_unlock(&tree->lock);
1053
1054 return ret;
1055
1056fail:
1057 if (!zpool_can_sleep_mapped(pool))
1058 kfree(tmp);
1059
1060 /*
1061 * if we get here due to ZSWAP_SWAPCACHE_EXIST
1062 * a load may be happening concurrently.
1063 * it is safe and okay to not free the entry.
1064 * if we free the entry in the following put
1065 * it is also okay to return !0
1066 */
1067 spin_lock(&tree->lock);
1068 zswap_entry_put(tree, entry);
1069 spin_unlock(&tree->lock);
1070
1071 return ret;
1072}
1073
1074static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
1075{
1076 unsigned int pos;
1077 unsigned long *page;
1078
1079 page = (unsigned long *)ptr;
1080 for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
1081 if (page[pos] != page[0])
1082 return 0;
1083 }
1084 *value = page[0];
1085 return 1;
1086}
1087
1088static void zswap_fill_page(void *ptr, unsigned long value)
1089{
1090 unsigned long *page;
1091
1092 page = (unsigned long *)ptr;
1093 memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
1094}
1095
1096/*********************************
1097* frontswap hooks
1098**********************************/
1099/* attempts to compress and store an single page */
1100static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1101 struct page *page)
1102{
1103 struct zswap_tree *tree = zswap_trees[type];
1104 struct zswap_entry *entry, *dupentry;
1105 struct scatterlist input, output;
1106 struct crypto_acomp_ctx *acomp_ctx;
1107 struct obj_cgroup *objcg = NULL;
1108 struct zswap_pool *pool;
1109 int ret;
1110 unsigned int hlen, dlen = PAGE_SIZE;
1111 unsigned long handle, value;
1112 char *buf;
1113 u8 *src, *dst;
1114 struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
1115 gfp_t gfp;
1116
1117 /* THP isn't supported */
1118 if (PageTransHuge(page)) {
1119 ret = -EINVAL;
1120 goto reject;
1121 }
1122
1123 if (!zswap_enabled || !tree) {
1124 ret = -ENODEV;
1125 goto reject;
1126 }
1127
1128 objcg = get_obj_cgroup_from_page(page);
1129 if (objcg && !obj_cgroup_may_zswap(objcg))
1130 goto shrink;
1131
1132 /* reclaim space if needed */
1133 if (zswap_is_full()) {
1134 zswap_pool_limit_hit++;
1135 zswap_pool_reached_full = true;
1136 goto shrink;
1137 }
1138
1139 if (zswap_pool_reached_full) {
1140 if (!zswap_can_accept()) {
1141 ret = -ENOMEM;
1142 goto reject;
1143 } else
1144 zswap_pool_reached_full = false;
1145 }
1146
1147 /* allocate entry */
1148 entry = zswap_entry_cache_alloc(GFP_KERNEL);
1149 if (!entry) {
1150 zswap_reject_kmemcache_fail++;
1151 ret = -ENOMEM;
1152 goto reject;
1153 }
1154
1155 if (zswap_same_filled_pages_enabled) {
1156 src = kmap_atomic(page);
1157 if (zswap_is_page_same_filled(src, &value)) {
1158 kunmap_atomic(src);
1159 entry->offset = offset;
1160 entry->length = 0;
1161 entry->value = value;
1162 atomic_inc(&zswap_same_filled_pages);
1163 goto insert_entry;
1164 }
1165 kunmap_atomic(src);
1166 }
1167
1168 if (!zswap_non_same_filled_pages_enabled) {
1169 ret = -EINVAL;
1170 goto freepage;
1171 }
1172
1173 /* if entry is successfully added, it keeps the reference */
1174 entry->pool = zswap_pool_current_get();
1175 if (!entry->pool) {
1176 ret = -EINVAL;
1177 goto freepage;
1178 }
1179
1180 /* compress */
1181 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1182
1183 mutex_lock(acomp_ctx->mutex);
1184
1185 dst = acomp_ctx->dstmem;
1186 sg_init_table(&input, 1);
1187 sg_set_page(&input, page, PAGE_SIZE, 0);
1188
1189 /* zswap_dstmem is of size (PAGE_SIZE * 2). Reflect same in sg_list */
1190 sg_init_one(&output, dst, PAGE_SIZE * 2);
1191 acomp_request_set_params(acomp_ctx->req, &input, &output, PAGE_SIZE, dlen);
1192 /*
1193 * it maybe looks a little bit silly that we send an asynchronous request,
1194 * then wait for its completion synchronously. This makes the process look
1195 * synchronous in fact.
1196 * Theoretically, acomp supports users send multiple acomp requests in one
1197 * acomp instance, then get those requests done simultaneously. but in this
1198 * case, frontswap actually does store and load page by page, there is no
1199 * existing method to send the second page before the first page is done
1200 * in one thread doing frontswap.
1201 * but in different threads running on different cpu, we have different
1202 * acomp instance, so multiple threads can do (de)compression in parallel.
1203 */
1204 ret = crypto_wait_req(crypto_acomp_compress(acomp_ctx->req), &acomp_ctx->wait);
1205 dlen = acomp_ctx->req->dlen;
1206
1207 if (ret) {
1208 ret = -EINVAL;
1209 goto put_dstmem;
1210 }
1211
1212 /* store */
1213 hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
1214 gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1215 if (zpool_malloc_support_movable(entry->pool->zpool))
1216 gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1217 ret = zpool_malloc(entry->pool->zpool, hlen + dlen, gfp, &handle);
1218 if (ret == -ENOSPC) {
1219 zswap_reject_compress_poor++;
1220 goto put_dstmem;
1221 }
1222 if (ret) {
1223 zswap_reject_alloc_fail++;
1224 goto put_dstmem;
1225 }
1226 buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_WO);
1227 memcpy(buf, &zhdr, hlen);
1228 memcpy(buf + hlen, dst, dlen);
1229 zpool_unmap_handle(entry->pool->zpool, handle);
1230 mutex_unlock(acomp_ctx->mutex);
1231
1232 /* populate entry */
1233 entry->offset = offset;
1234 entry->handle = handle;
1235 entry->length = dlen;
1236
1237insert_entry:
1238 entry->objcg = objcg;
1239 if (objcg) {
1240 obj_cgroup_charge_zswap(objcg, entry->length);
1241 /* Account before objcg ref is moved to tree */
1242 count_objcg_event(objcg, ZSWPOUT);
1243 }
1244
1245 /* map */
1246 spin_lock(&tree->lock);
1247 do {
1248 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1249 if (ret == -EEXIST) {
1250 zswap_duplicate_entry++;
1251 /* remove from rbtree */
1252 zswap_rb_erase(&tree->rbroot, dupentry);
1253 zswap_entry_put(tree, dupentry);
1254 }
1255 } while (ret == -EEXIST);
1256 spin_unlock(&tree->lock);
1257
1258 /* update stats */
1259 atomic_inc(&zswap_stored_pages);
1260 zswap_update_total_size();
1261 count_vm_event(ZSWPOUT);
1262
1263 return 0;
1264
1265put_dstmem:
1266 mutex_unlock(acomp_ctx->mutex);
1267 zswap_pool_put(entry->pool);
1268freepage:
1269 zswap_entry_cache_free(entry);
1270reject:
1271 if (objcg)
1272 obj_cgroup_put(objcg);
1273 return ret;
1274
1275shrink:
1276 pool = zswap_pool_last_get();
1277 if (pool)
1278 queue_work(shrink_wq, &pool->shrink_work);
1279 ret = -ENOMEM;
1280 goto reject;
1281}
1282
1283/*
1284 * returns 0 if the page was successfully decompressed
1285 * return -1 on entry not found or error
1286*/
1287static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1288 struct page *page)
1289{
1290 struct zswap_tree *tree = zswap_trees[type];
1291 struct zswap_entry *entry;
1292 struct scatterlist input, output;
1293 struct crypto_acomp_ctx *acomp_ctx;
1294 u8 *src, *dst, *tmp;
1295 unsigned int dlen;
1296 int ret;
1297
1298 /* find */
1299 spin_lock(&tree->lock);
1300 entry = zswap_entry_find_get(&tree->rbroot, offset);
1301 if (!entry) {
1302 /* entry was written back */
1303 spin_unlock(&tree->lock);
1304 return -1;
1305 }
1306 spin_unlock(&tree->lock);
1307
1308 if (!entry->length) {
1309 dst = kmap_atomic(page);
1310 zswap_fill_page(dst, entry->value);
1311 kunmap_atomic(dst);
1312 ret = 0;
1313 goto stats;
1314 }
1315
1316 if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1317 tmp = kmalloc(entry->length, GFP_KERNEL);
1318 if (!tmp) {
1319 ret = -ENOMEM;
1320 goto freeentry;
1321 }
1322 }
1323
1324 /* decompress */
1325 dlen = PAGE_SIZE;
1326 src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1327 if (zpool_evictable(entry->pool->zpool))
1328 src += sizeof(struct zswap_header);
1329
1330 if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
1331 memcpy(tmp, src, entry->length);
1332 src = tmp;
1333 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1334 }
1335
1336 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1337 mutex_lock(acomp_ctx->mutex);
1338 sg_init_one(&input, src, entry->length);
1339 sg_init_table(&output, 1);
1340 sg_set_page(&output, page, PAGE_SIZE, 0);
1341 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1342 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1343 mutex_unlock(acomp_ctx->mutex);
1344
1345 if (zpool_can_sleep_mapped(entry->pool->zpool))
1346 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1347 else
1348 kfree(tmp);
1349
1350 BUG_ON(ret);
1351stats:
1352 count_vm_event(ZSWPIN);
1353 if (entry->objcg)
1354 count_objcg_event(entry->objcg, ZSWPIN);
1355freeentry:
1356 spin_lock(&tree->lock);
1357 zswap_entry_put(tree, entry);
1358 spin_unlock(&tree->lock);
1359
1360 return ret;
1361}
1362
1363/* frees an entry in zswap */
1364static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1365{
1366 struct zswap_tree *tree = zswap_trees[type];
1367 struct zswap_entry *entry;
1368
1369 /* find */
1370 spin_lock(&tree->lock);
1371 entry = zswap_rb_search(&tree->rbroot, offset);
1372 if (!entry) {
1373 /* entry was written back */
1374 spin_unlock(&tree->lock);
1375 return;
1376 }
1377
1378 /* remove from rbtree */
1379 zswap_rb_erase(&tree->rbroot, entry);
1380
1381 /* drop the initial reference from entry creation */
1382 zswap_entry_put(tree, entry);
1383
1384 spin_unlock(&tree->lock);
1385}
1386
1387/* frees all zswap entries for the given swap type */
1388static void zswap_frontswap_invalidate_area(unsigned type)
1389{
1390 struct zswap_tree *tree = zswap_trees[type];
1391 struct zswap_entry *entry, *n;
1392
1393 if (!tree)
1394 return;
1395
1396 /* walk the tree and free everything */
1397 spin_lock(&tree->lock);
1398 rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1399 zswap_free_entry(entry);
1400 tree->rbroot = RB_ROOT;
1401 spin_unlock(&tree->lock);
1402 kfree(tree);
1403 zswap_trees[type] = NULL;
1404}
1405
1406static void zswap_frontswap_init(unsigned type)
1407{
1408 struct zswap_tree *tree;
1409
1410 tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1411 if (!tree) {
1412 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1413 return;
1414 }
1415
1416 tree->rbroot = RB_ROOT;
1417 spin_lock_init(&tree->lock);
1418 zswap_trees[type] = tree;
1419}
1420
1421static const struct frontswap_ops zswap_frontswap_ops = {
1422 .store = zswap_frontswap_store,
1423 .load = zswap_frontswap_load,
1424 .invalidate_page = zswap_frontswap_invalidate_page,
1425 .invalidate_area = zswap_frontswap_invalidate_area,
1426 .init = zswap_frontswap_init
1427};
1428
1429/*********************************
1430* debugfs functions
1431**********************************/
1432#ifdef CONFIG_DEBUG_FS
1433#include <linux/debugfs.h>
1434
1435static struct dentry *zswap_debugfs_root;
1436
1437static int __init zswap_debugfs_init(void)
1438{
1439 if (!debugfs_initialized())
1440 return -ENODEV;
1441
1442 zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1443
1444 debugfs_create_u64("pool_limit_hit", 0444,
1445 zswap_debugfs_root, &zswap_pool_limit_hit);
1446 debugfs_create_u64("reject_reclaim_fail", 0444,
1447 zswap_debugfs_root, &zswap_reject_reclaim_fail);
1448 debugfs_create_u64("reject_alloc_fail", 0444,
1449 zswap_debugfs_root, &zswap_reject_alloc_fail);
1450 debugfs_create_u64("reject_kmemcache_fail", 0444,
1451 zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1452 debugfs_create_u64("reject_compress_poor", 0444,
1453 zswap_debugfs_root, &zswap_reject_compress_poor);
1454 debugfs_create_u64("written_back_pages", 0444,
1455 zswap_debugfs_root, &zswap_written_back_pages);
1456 debugfs_create_u64("duplicate_entry", 0444,
1457 zswap_debugfs_root, &zswap_duplicate_entry);
1458 debugfs_create_u64("pool_total_size", 0444,
1459 zswap_debugfs_root, &zswap_pool_total_size);
1460 debugfs_create_atomic_t("stored_pages", 0444,
1461 zswap_debugfs_root, &zswap_stored_pages);
1462 debugfs_create_atomic_t("same_filled_pages", 0444,
1463 zswap_debugfs_root, &zswap_same_filled_pages);
1464
1465 return 0;
1466}
1467#else
1468static int __init zswap_debugfs_init(void)
1469{
1470 return 0;
1471}
1472#endif
1473
1474/*********************************
1475* module init and exit
1476**********************************/
1477static int __init init_zswap(void)
1478{
1479 struct zswap_pool *pool;
1480 int ret;
1481
1482 zswap_init_started = true;
1483
1484 if (zswap_entry_cache_create()) {
1485 pr_err("entry cache creation failed\n");
1486 goto cache_fail;
1487 }
1488
1489 ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1490 zswap_dstmem_prepare, zswap_dstmem_dead);
1491 if (ret) {
1492 pr_err("dstmem alloc failed\n");
1493 goto dstmem_fail;
1494 }
1495
1496 ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1497 "mm/zswap_pool:prepare",
1498 zswap_cpu_comp_prepare,
1499 zswap_cpu_comp_dead);
1500 if (ret)
1501 goto hp_fail;
1502
1503 pool = __zswap_pool_create_fallback();
1504 if (pool) {
1505 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1506 zpool_get_type(pool->zpool));
1507 list_add(&pool->list, &zswap_pools);
1508 zswap_has_pool = true;
1509 } else {
1510 pr_err("pool creation failed\n");
1511 zswap_enabled = false;
1512 }
1513
1514 shrink_wq = create_workqueue("zswap-shrink");
1515 if (!shrink_wq)
1516 goto fallback_fail;
1517
1518 ret = frontswap_register_ops(&zswap_frontswap_ops);
1519 if (ret)
1520 goto destroy_wq;
1521 if (zswap_debugfs_init())
1522 pr_warn("debugfs initialization failed\n");
1523 return 0;
1524
1525destroy_wq:
1526 destroy_workqueue(shrink_wq);
1527fallback_fail:
1528 if (pool)
1529 zswap_pool_destroy(pool);
1530hp_fail:
1531 cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1532dstmem_fail:
1533 zswap_entry_cache_destroy();
1534cache_fail:
1535 /* if built-in, we aren't unloaded on failure; don't allow use */
1536 zswap_init_failed = true;
1537 zswap_enabled = false;
1538 return -ENOMEM;
1539}
1540/* must be late so crypto has time to come up */
1541late_initcall(init_zswap);
1542
1543MODULE_LICENSE("GPL");
1544MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1545MODULE_DESCRIPTION("Compressed cache for swap pages");