brd: use XArray instead of radix-tree to index backing pages
[linux-block.git] / mm / zswap.c
CommitLineData
c942fddf 1// SPDX-License-Identifier: GPL-2.0-or-later
2b281117
SJ
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>
2b281117
SJ
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>
1ec3b5fe 27#include <linux/scatterlist.h>
2b281117 28#include <linux/mempool.h>
12d79d64 29#include <linux/zpool.h>
1ec3b5fe 30#include <crypto/acompress.h>
2b281117
SJ
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>
45190f01 37#include <linux/workqueue.h>
2b281117 38
014bb1de
N
39#include "swap.h"
40
2b281117
SJ
41/*********************************
42* statistics
43**********************************/
12d79d64 44/* Total bytes used by the compressed storage */
f6498b77 45u64 zswap_pool_total_size;
2b281117 46/* The number of compressed pages currently stored in zswap */
f6498b77 47atomic_t zswap_stored_pages = ATOMIC_INIT(0);
a85f878b
SD
48/* The number of same-value filled pages currently stored in zswap */
49static atomic_t zswap_same_filled_pages = ATOMIC_INIT(0);
2b281117
SJ
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
45190f01
VW
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
2b281117
SJ
78/*********************************
79* tunables
80**********************************/
c00ed16a 81
bae21db8
DS
82#define ZSWAP_PARAM_UNSET ""
83
141fdeec
LS
84static int zswap_setup(void);
85
bb8b93b5
MS
86/* Enable/disable zswap */
87static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
d7b028f5
DS
88static int zswap_enabled_param_set(const char *,
89 const struct kernel_param *);
83aed6cd 90static const struct kernel_param_ops zswap_enabled_param_ops = {
d7b028f5
DS
91 .set = zswap_enabled_param_set,
92 .get = param_get_bool,
93};
94module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
2b281117 95
90b0fc26 96/* Crypto compressor to use */
bb8b93b5 97static char *zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
90b0fc26
DS
98static int zswap_compressor_param_set(const char *,
99 const struct kernel_param *);
83aed6cd 100static const struct kernel_param_ops zswap_compressor_param_ops = {
90b0fc26 101 .set = zswap_compressor_param_set,
c99b42c3
DS
102 .get = param_get_charp,
103 .free = param_free_charp,
90b0fc26
DS
104};
105module_param_cb(compressor, &zswap_compressor_param_ops,
c99b42c3 106 &zswap_compressor, 0644);
2b281117 107
90b0fc26 108/* Compressed storage zpool to use */
bb8b93b5 109static char *zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
90b0fc26 110static int zswap_zpool_param_set(const char *, const struct kernel_param *);
83aed6cd 111static const struct kernel_param_ops zswap_zpool_param_ops = {
c99b42c3
DS
112 .set = zswap_zpool_param_set,
113 .get = param_get_charp,
114 .free = param_free_charp,
90b0fc26 115};
c99b42c3 116module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
12d79d64 117
90b0fc26
DS
118/* The maximum percentage of memory that the compressed pool can occupy */
119static unsigned int zswap_max_pool_percent = 20;
120module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
60105e12 121
45190f01
VW
122/* The threshold for accepting new pages after the max_pool_percent was hit */
123static unsigned int zswap_accept_thr_percent = 90; /* of max pool size */
124module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
125 uint, 0644);
126
cb325ddd
MS
127/*
128 * Enable/disable handling same-value filled pages (enabled by default).
129 * If disabled every page is considered non-same-value filled.
130 */
a85f878b
SD
131static bool zswap_same_filled_pages_enabled = true;
132module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
133 bool, 0644);
134
cb325ddd
MS
135/* Enable/disable handling non-same-value filled pages (enabled by default) */
136static bool zswap_non_same_filled_pages_enabled = true;
137module_param_named(non_same_filled_pages_enabled, zswap_non_same_filled_pages_enabled,
138 bool, 0644);
139
2b281117 140/*********************************
f1c54846 141* data structures
2b281117 142**********************************/
2b281117 143
1ec3b5fe
BS
144struct crypto_acomp_ctx {
145 struct crypto_acomp *acomp;
146 struct acomp_req *req;
147 struct crypto_wait wait;
148 u8 *dstmem;
149 struct mutex *mutex;
150};
151
f1c54846
DS
152struct zswap_pool {
153 struct zpool *zpool;
1ec3b5fe 154 struct crypto_acomp_ctx __percpu *acomp_ctx;
f1c54846
DS
155 struct kref kref;
156 struct list_head list;
45190f01
VW
157 struct work_struct release_work;
158 struct work_struct shrink_work;
cab7a7e5 159 struct hlist_node node;
f1c54846 160 char tfm_name[CRYPTO_MAX_ALG_NAME];
2b281117
SJ
161};
162
2b281117
SJ
163/*
164 * struct zswap_entry
165 *
166 * This structure contains the metadata for tracking a single compressed
167 * page within zswap.
168 *
169 * rbnode - links the entry into red-black tree for the appropriate swap type
f1c54846 170 * offset - the swap offset for the entry. Index into the red-black tree.
2b281117
SJ
171 * refcount - the number of outstanding reference to the entry. This is needed
172 * to protect against premature freeing of the entry by code
6b452516 173 * concurrent calls to load, invalidate, and writeback. The lock
2b281117
SJ
174 * for the zswap_tree structure that contains the entry must
175 * be held while changing the refcount. Since the lock must
176 * be held, there is no reason to also make refcount atomic.
2b281117 177 * length - the length in bytes of the compressed page data. Needed during
a85f878b 178 * decompression. For a same value filled page length is 0.
f1c54846
DS
179 * pool - the zswap_pool the entry's data is in
180 * handle - zpool allocation handle that stores the compressed page data
a85f878b 181 * value - value of the same-value filled pages which have same content
2b281117
SJ
182 */
183struct zswap_entry {
184 struct rb_node rbnode;
185 pgoff_t offset;
186 int refcount;
187 unsigned int length;
f1c54846 188 struct zswap_pool *pool;
a85f878b
SD
189 union {
190 unsigned long handle;
191 unsigned long value;
192 };
f4840ccf 193 struct obj_cgroup *objcg;
2b281117
SJ
194};
195
196struct zswap_header {
197 swp_entry_t swpentry;
198};
199
200/*
201 * The tree lock in the zswap_tree struct protects a few things:
202 * - the rbtree
203 * - the refcount field of each entry in the tree
204 */
205struct zswap_tree {
206 struct rb_root rbroot;
207 spinlock_t lock;
2b281117
SJ
208};
209
210static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
211
f1c54846
DS
212/* RCU-protected iteration */
213static LIST_HEAD(zswap_pools);
214/* protects zswap_pools list modification */
215static DEFINE_SPINLOCK(zswap_pools_lock);
32a4e169
DS
216/* pool counter to provide unique names to zpool */
217static atomic_t zswap_pools_count = ATOMIC_INIT(0);
f1c54846 218
9021ccec
LS
219enum zswap_init_type {
220 ZSWAP_UNINIT,
221 ZSWAP_INIT_SUCCEED,
222 ZSWAP_INIT_FAILED
223};
90b0fc26 224
9021ccec 225static enum zswap_init_type zswap_init_state;
90b0fc26 226
141fdeec
LS
227/* used to ensure the integrity of initialization */
228static DEFINE_MUTEX(zswap_init_lock);
d7b028f5 229
ae3d89a7
DS
230/* init completed, but couldn't create the initial pool */
231static bool zswap_has_pool;
232
f1c54846
DS
233/*********************************
234* helpers and fwd declarations
235**********************************/
236
237#define zswap_pool_debug(msg, p) \
238 pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name, \
239 zpool_get_type((p)->zpool))
240
241static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
242static int zswap_pool_get(struct zswap_pool *pool);
243static void zswap_pool_put(struct zswap_pool *pool);
244
245static const struct zpool_ops zswap_zpool_ops = {
246 .evict = zswap_writeback_entry
247};
248
249static bool zswap_is_full(void)
250{
ca79b0c2
AK
251 return totalram_pages() * zswap_max_pool_percent / 100 <
252 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
f1c54846
DS
253}
254
45190f01
VW
255static bool zswap_can_accept(void)
256{
257 return totalram_pages() * zswap_accept_thr_percent / 100 *
258 zswap_max_pool_percent / 100 >
259 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
260}
261
f1c54846
DS
262static void zswap_update_total_size(void)
263{
264 struct zswap_pool *pool;
265 u64 total = 0;
266
267 rcu_read_lock();
268
269 list_for_each_entry_rcu(pool, &zswap_pools, list)
270 total += zpool_get_total_size(pool->zpool);
271
272 rcu_read_unlock();
273
274 zswap_pool_total_size = total;
275}
276
2b281117
SJ
277/*********************************
278* zswap entry functions
279**********************************/
280static struct kmem_cache *zswap_entry_cache;
281
2b281117
SJ
282static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
283{
284 struct zswap_entry *entry;
285 entry = kmem_cache_alloc(zswap_entry_cache, gfp);
286 if (!entry)
287 return NULL;
288 entry->refcount = 1;
0ab0abcf 289 RB_CLEAR_NODE(&entry->rbnode);
2b281117
SJ
290 return entry;
291}
292
293static void zswap_entry_cache_free(struct zswap_entry *entry)
294{
295 kmem_cache_free(zswap_entry_cache, entry);
296}
297
2b281117
SJ
298/*********************************
299* rbtree functions
300**********************************/
301static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
302{
303 struct rb_node *node = root->rb_node;
304 struct zswap_entry *entry;
305
306 while (node) {
307 entry = rb_entry(node, struct zswap_entry, rbnode);
308 if (entry->offset > offset)
309 node = node->rb_left;
310 else if (entry->offset < offset)
311 node = node->rb_right;
312 else
313 return entry;
314 }
315 return NULL;
316}
317
318/*
319 * In the case that a entry with the same offset is found, a pointer to
320 * the existing entry is stored in dupentry and the function returns -EEXIST
321 */
322static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
323 struct zswap_entry **dupentry)
324{
325 struct rb_node **link = &root->rb_node, *parent = NULL;
326 struct zswap_entry *myentry;
327
328 while (*link) {
329 parent = *link;
330 myentry = rb_entry(parent, struct zswap_entry, rbnode);
331 if (myentry->offset > entry->offset)
332 link = &(*link)->rb_left;
333 else if (myentry->offset < entry->offset)
334 link = &(*link)->rb_right;
335 else {
336 *dupentry = myentry;
337 return -EEXIST;
338 }
339 }
340 rb_link_node(&entry->rbnode, parent, link);
341 rb_insert_color(&entry->rbnode, root);
342 return 0;
343}
344
0ab0abcf
WY
345static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
346{
347 if (!RB_EMPTY_NODE(&entry->rbnode)) {
348 rb_erase(&entry->rbnode, root);
349 RB_CLEAR_NODE(&entry->rbnode);
350 }
351}
352
353/*
12d79d64 354 * Carries out the common pattern of freeing and entry's zpool allocation,
0ab0abcf
WY
355 * freeing the entry itself, and decrementing the number of stored pages.
356 */
60105e12 357static void zswap_free_entry(struct zswap_entry *entry)
0ab0abcf 358{
f4840ccf
JW
359 if (entry->objcg) {
360 obj_cgroup_uncharge_zswap(entry->objcg, entry->length);
361 obj_cgroup_put(entry->objcg);
362 }
a85f878b
SD
363 if (!entry->length)
364 atomic_dec(&zswap_same_filled_pages);
365 else {
366 zpool_free(entry->pool->zpool, entry->handle);
367 zswap_pool_put(entry->pool);
368 }
0ab0abcf
WY
369 zswap_entry_cache_free(entry);
370 atomic_dec(&zswap_stored_pages);
f1c54846 371 zswap_update_total_size();
0ab0abcf
WY
372}
373
374/* caller must hold the tree lock */
375static void zswap_entry_get(struct zswap_entry *entry)
376{
377 entry->refcount++;
378}
379
380/* caller must hold the tree lock
381* remove from the tree and free it, if nobody reference the entry
382*/
383static void zswap_entry_put(struct zswap_tree *tree,
384 struct zswap_entry *entry)
385{
386 int refcount = --entry->refcount;
387
388 BUG_ON(refcount < 0);
389 if (refcount == 0) {
390 zswap_rb_erase(&tree->rbroot, entry);
60105e12 391 zswap_free_entry(entry);
0ab0abcf
WY
392 }
393}
394
395/* caller must hold the tree lock */
396static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
397 pgoff_t offset)
398{
b0c9865f 399 struct zswap_entry *entry;
0ab0abcf
WY
400
401 entry = zswap_rb_search(root, offset);
402 if (entry)
403 zswap_entry_get(entry);
404
405 return entry;
406}
407
2b281117
SJ
408/*********************************
409* per-cpu code
410**********************************/
411static DEFINE_PER_CPU(u8 *, zswap_dstmem);
1ec3b5fe
BS
412/*
413 * If users dynamically change the zpool type and compressor at runtime, i.e.
414 * zswap is running, zswap can have more than one zpool on one cpu, but they
415 * are sharing dtsmem. So we need this mutex to be per-cpu.
416 */
417static DEFINE_PER_CPU(struct mutex *, zswap_mutex);
2b281117 418
ad7ed770 419static int zswap_dstmem_prepare(unsigned int cpu)
2b281117 420{
1ec3b5fe 421 struct mutex *mutex;
2b281117
SJ
422 u8 *dst;
423
ad7ed770 424 dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
2b2695f5 425 if (!dst)
ad7ed770 426 return -ENOMEM;
2b2695f5 427
1ec3b5fe
BS
428 mutex = kmalloc_node(sizeof(*mutex), GFP_KERNEL, cpu_to_node(cpu));
429 if (!mutex) {
430 kfree(dst);
431 return -ENOMEM;
432 }
433
434 mutex_init(mutex);
ad7ed770 435 per_cpu(zswap_dstmem, cpu) = dst;
1ec3b5fe 436 per_cpu(zswap_mutex, cpu) = mutex;
ad7ed770 437 return 0;
2b281117
SJ
438}
439
ad7ed770 440static int zswap_dstmem_dead(unsigned int cpu)
2b281117 441{
1ec3b5fe 442 struct mutex *mutex;
ad7ed770 443 u8 *dst;
2b281117 444
1ec3b5fe
BS
445 mutex = per_cpu(zswap_mutex, cpu);
446 kfree(mutex);
447 per_cpu(zswap_mutex, cpu) = NULL;
448
ad7ed770
SAS
449 dst = per_cpu(zswap_dstmem, cpu);
450 kfree(dst);
451 per_cpu(zswap_dstmem, cpu) = NULL;
f1c54846 452
f1c54846 453 return 0;
f1c54846
DS
454}
455
cab7a7e5 456static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
f1c54846 457{
cab7a7e5 458 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
1ec3b5fe
BS
459 struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
460 struct crypto_acomp *acomp;
461 struct acomp_req *req;
462
463 acomp = crypto_alloc_acomp_node(pool->tfm_name, 0, 0, cpu_to_node(cpu));
464 if (IS_ERR(acomp)) {
465 pr_err("could not alloc crypto acomp %s : %ld\n",
466 pool->tfm_name, PTR_ERR(acomp));
467 return PTR_ERR(acomp);
468 }
469 acomp_ctx->acomp = acomp;
f1c54846 470
1ec3b5fe
BS
471 req = acomp_request_alloc(acomp_ctx->acomp);
472 if (!req) {
473 pr_err("could not alloc crypto acomp_request %s\n",
474 pool->tfm_name);
475 crypto_free_acomp(acomp_ctx->acomp);
cab7a7e5
SAS
476 return -ENOMEM;
477 }
1ec3b5fe
BS
478 acomp_ctx->req = req;
479
480 crypto_init_wait(&acomp_ctx->wait);
481 /*
482 * if the backend of acomp is async zip, crypto_req_done() will wakeup
483 * crypto_wait_req(); if the backend of acomp is scomp, the callback
484 * won't be called, crypto_wait_req() will return without blocking.
485 */
486 acomp_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
487 crypto_req_done, &acomp_ctx->wait);
488
489 acomp_ctx->mutex = per_cpu(zswap_mutex, cpu);
490 acomp_ctx->dstmem = per_cpu(zswap_dstmem, cpu);
491
2b281117 492 return 0;
2b281117
SJ
493}
494
cab7a7e5 495static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
f1c54846 496{
cab7a7e5 497 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
1ec3b5fe
BS
498 struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
499
500 if (!IS_ERR_OR_NULL(acomp_ctx)) {
501 if (!IS_ERR_OR_NULL(acomp_ctx->req))
502 acomp_request_free(acomp_ctx->req);
503 if (!IS_ERR_OR_NULL(acomp_ctx->acomp))
504 crypto_free_acomp(acomp_ctx->acomp);
505 }
f1c54846 506
cab7a7e5 507 return 0;
f1c54846
DS
508}
509
2b281117 510/*********************************
f1c54846 511* pool functions
2b281117 512**********************************/
f1c54846
DS
513
514static struct zswap_pool *__zswap_pool_current(void)
2b281117 515{
f1c54846
DS
516 struct zswap_pool *pool;
517
518 pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
ae3d89a7
DS
519 WARN_ONCE(!pool && zswap_has_pool,
520 "%s: no page storage pool!\n", __func__);
f1c54846
DS
521
522 return pool;
523}
524
525static struct zswap_pool *zswap_pool_current(void)
526{
527 assert_spin_locked(&zswap_pools_lock);
528
529 return __zswap_pool_current();
530}
531
532static struct zswap_pool *zswap_pool_current_get(void)
533{
534 struct zswap_pool *pool;
535
536 rcu_read_lock();
537
538 pool = __zswap_pool_current();
ae3d89a7 539 if (!zswap_pool_get(pool))
f1c54846
DS
540 pool = NULL;
541
542 rcu_read_unlock();
543
544 return pool;
545}
546
547static struct zswap_pool *zswap_pool_last_get(void)
548{
549 struct zswap_pool *pool, *last = NULL;
550
551 rcu_read_lock();
552
553 list_for_each_entry_rcu(pool, &zswap_pools, list)
554 last = pool;
ae3d89a7
DS
555 WARN_ONCE(!last && zswap_has_pool,
556 "%s: no page storage pool!\n", __func__);
557 if (!zswap_pool_get(last))
f1c54846
DS
558 last = NULL;
559
560 rcu_read_unlock();
561
562 return last;
563}
564
8bc8b228 565/* type and compressor must be null-terminated */
f1c54846
DS
566static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
567{
568 struct zswap_pool *pool;
569
570 assert_spin_locked(&zswap_pools_lock);
571
572 list_for_each_entry_rcu(pool, &zswap_pools, list) {
8bc8b228 573 if (strcmp(pool->tfm_name, compressor))
f1c54846 574 continue;
8bc8b228 575 if (strcmp(zpool_get_type(pool->zpool), type))
f1c54846
DS
576 continue;
577 /* if we can't get it, it's about to be destroyed */
578 if (!zswap_pool_get(pool))
579 continue;
580 return pool;
581 }
582
583 return NULL;
584}
585
45190f01
VW
586static void shrink_worker(struct work_struct *w)
587{
588 struct zswap_pool *pool = container_of(w, typeof(*pool),
589 shrink_work);
590
591 if (zpool_shrink(pool->zpool, 1, NULL))
592 zswap_reject_reclaim_fail++;
593 zswap_pool_put(pool);
594}
595
f1c54846
DS
596static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
597{
598 struct zswap_pool *pool;
32a4e169 599 char name[38]; /* 'zswap' + 32 char (max) num + \0 */
d0164adc 600 gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
cab7a7e5 601 int ret;
f1c54846 602
bae21db8
DS
603 if (!zswap_has_pool) {
604 /* if either are unset, pool initialization failed, and we
605 * need both params to be set correctly before trying to
606 * create a pool.
607 */
608 if (!strcmp(type, ZSWAP_PARAM_UNSET))
609 return NULL;
610 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
611 return NULL;
612 }
613
f1c54846 614 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
f4ae0ce0 615 if (!pool)
f1c54846 616 return NULL;
f1c54846 617
32a4e169
DS
618 /* unique name for each pool specifically required by zsmalloc */
619 snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
620
621 pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
f1c54846
DS
622 if (!pool->zpool) {
623 pr_err("%s zpool not available\n", type);
624 goto error;
625 }
626 pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
627
79cd4202 628 strscpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
1ec3b5fe
BS
629
630 pool->acomp_ctx = alloc_percpu(*pool->acomp_ctx);
631 if (!pool->acomp_ctx) {
f1c54846
DS
632 pr_err("percpu alloc failed\n");
633 goto error;
634 }
635
cab7a7e5
SAS
636 ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
637 &pool->node);
638 if (ret)
f1c54846
DS
639 goto error;
640 pr_debug("using %s compressor\n", pool->tfm_name);
641
642 /* being the current pool takes 1 ref; this func expects the
643 * caller to always add the new pool as the current pool
644 */
645 kref_init(&pool->kref);
646 INIT_LIST_HEAD(&pool->list);
45190f01 647 INIT_WORK(&pool->shrink_work, shrink_worker);
f1c54846
DS
648
649 zswap_pool_debug("created", pool);
650
651 return pool;
652
653error:
1ec3b5fe
BS
654 if (pool->acomp_ctx)
655 free_percpu(pool->acomp_ctx);
f1c54846
DS
656 if (pool->zpool)
657 zpool_destroy_pool(pool->zpool);
658 kfree(pool);
659 return NULL;
660}
661
141fdeec 662static struct zswap_pool *__zswap_pool_create_fallback(void)
f1c54846 663{
bae21db8
DS
664 bool has_comp, has_zpool;
665
1ec3b5fe 666 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
bb8b93b5
MS
667 if (!has_comp && strcmp(zswap_compressor,
668 CONFIG_ZSWAP_COMPRESSOR_DEFAULT)) {
f1c54846 669 pr_err("compressor %s not available, using default %s\n",
bb8b93b5 670 zswap_compressor, CONFIG_ZSWAP_COMPRESSOR_DEFAULT);
c99b42c3 671 param_free_charp(&zswap_compressor);
bb8b93b5 672 zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
1ec3b5fe 673 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
f1c54846 674 }
bae21db8
DS
675 if (!has_comp) {
676 pr_err("default compressor %s not available\n",
677 zswap_compressor);
678 param_free_charp(&zswap_compressor);
679 zswap_compressor = ZSWAP_PARAM_UNSET;
680 }
681
682 has_zpool = zpool_has_pool(zswap_zpool_type);
bb8b93b5
MS
683 if (!has_zpool && strcmp(zswap_zpool_type,
684 CONFIG_ZSWAP_ZPOOL_DEFAULT)) {
f1c54846 685 pr_err("zpool %s not available, using default %s\n",
bb8b93b5 686 zswap_zpool_type, CONFIG_ZSWAP_ZPOOL_DEFAULT);
c99b42c3 687 param_free_charp(&zswap_zpool_type);
bb8b93b5 688 zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
bae21db8 689 has_zpool = zpool_has_pool(zswap_zpool_type);
f1c54846 690 }
bae21db8
DS
691 if (!has_zpool) {
692 pr_err("default zpool %s not available\n",
693 zswap_zpool_type);
694 param_free_charp(&zswap_zpool_type);
695 zswap_zpool_type = ZSWAP_PARAM_UNSET;
696 }
697
698 if (!has_comp || !has_zpool)
699 return NULL;
f1c54846
DS
700
701 return zswap_pool_create(zswap_zpool_type, zswap_compressor);
702}
703
704static void zswap_pool_destroy(struct zswap_pool *pool)
705{
706 zswap_pool_debug("destroying", pool);
707
cab7a7e5 708 cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
1ec3b5fe 709 free_percpu(pool->acomp_ctx);
f1c54846
DS
710 zpool_destroy_pool(pool->zpool);
711 kfree(pool);
712}
713
714static int __must_check zswap_pool_get(struct zswap_pool *pool)
715{
ae3d89a7
DS
716 if (!pool)
717 return 0;
718
f1c54846
DS
719 return kref_get_unless_zero(&pool->kref);
720}
721
200867af 722static void __zswap_pool_release(struct work_struct *work)
f1c54846 723{
45190f01
VW
724 struct zswap_pool *pool = container_of(work, typeof(*pool),
725 release_work);
200867af
DS
726
727 synchronize_rcu();
f1c54846
DS
728
729 /* nobody should have been able to get a kref... */
730 WARN_ON(kref_get_unless_zero(&pool->kref));
731
732 /* pool is now off zswap_pools list and has no references. */
733 zswap_pool_destroy(pool);
734}
735
736static void __zswap_pool_empty(struct kref *kref)
737{
738 struct zswap_pool *pool;
739
740 pool = container_of(kref, typeof(*pool), kref);
741
742 spin_lock(&zswap_pools_lock);
743
744 WARN_ON(pool == zswap_pool_current());
745
746 list_del_rcu(&pool->list);
200867af 747
45190f01
VW
748 INIT_WORK(&pool->release_work, __zswap_pool_release);
749 schedule_work(&pool->release_work);
f1c54846
DS
750
751 spin_unlock(&zswap_pools_lock);
752}
753
754static void zswap_pool_put(struct zswap_pool *pool)
755{
756 kref_put(&pool->kref, __zswap_pool_empty);
2b281117
SJ
757}
758
90b0fc26
DS
759/*********************************
760* param callbacks
761**********************************/
762
141fdeec
LS
763static bool zswap_pool_changed(const char *s, const struct kernel_param *kp)
764{
765 /* no change required */
766 if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
767 return false;
768 return true;
769}
770
c99b42c3 771/* val must be a null-terminated string */
90b0fc26
DS
772static int __zswap_param_set(const char *val, const struct kernel_param *kp,
773 char *type, char *compressor)
774{
775 struct zswap_pool *pool, *put_pool = NULL;
c99b42c3 776 char *s = strstrip((char *)val);
141fdeec
LS
777 int ret = 0;
778 bool new_pool = false;
90b0fc26 779
141fdeec 780 mutex_lock(&zswap_init_lock);
9021ccec
LS
781 switch (zswap_init_state) {
782 case ZSWAP_UNINIT:
783 /* if this is load-time (pre-init) param setting,
784 * don't create a pool; that's done during init.
785 */
141fdeec
LS
786 ret = param_set_charp(s, kp);
787 break;
9021ccec 788 case ZSWAP_INIT_SUCCEED:
141fdeec 789 new_pool = zswap_pool_changed(s, kp);
9021ccec
LS
790 break;
791 case ZSWAP_INIT_FAILED:
d7b028f5 792 pr_err("can't set param, initialization failed\n");
141fdeec 793 ret = -ENODEV;
d7b028f5 794 }
141fdeec 795 mutex_unlock(&zswap_init_lock);
d7b028f5 796
141fdeec
LS
797 /* no need to create a new pool, return directly */
798 if (!new_pool)
799 return ret;
90b0fc26
DS
800
801 if (!type) {
c99b42c3
DS
802 if (!zpool_has_pool(s)) {
803 pr_err("zpool %s not available\n", s);
90b0fc26
DS
804 return -ENOENT;
805 }
c99b42c3 806 type = s;
90b0fc26 807 } else if (!compressor) {
1ec3b5fe 808 if (!crypto_has_acomp(s, 0, 0)) {
c99b42c3 809 pr_err("compressor %s not available\n", s);
90b0fc26
DS
810 return -ENOENT;
811 }
c99b42c3
DS
812 compressor = s;
813 } else {
814 WARN_ON(1);
815 return -EINVAL;
90b0fc26
DS
816 }
817
818 spin_lock(&zswap_pools_lock);
819
820 pool = zswap_pool_find_get(type, compressor);
821 if (pool) {
822 zswap_pool_debug("using existing", pool);
fd5bb66c 823 WARN_ON(pool == zswap_pool_current());
90b0fc26 824 list_del_rcu(&pool->list);
90b0fc26
DS
825 }
826
fd5bb66c
DS
827 spin_unlock(&zswap_pools_lock);
828
829 if (!pool)
830 pool = zswap_pool_create(type, compressor);
831
90b0fc26 832 if (pool)
c99b42c3 833 ret = param_set_charp(s, kp);
90b0fc26
DS
834 else
835 ret = -EINVAL;
836
fd5bb66c
DS
837 spin_lock(&zswap_pools_lock);
838
90b0fc26
DS
839 if (!ret) {
840 put_pool = zswap_pool_current();
841 list_add_rcu(&pool->list, &zswap_pools);
ae3d89a7 842 zswap_has_pool = true;
90b0fc26
DS
843 } else if (pool) {
844 /* add the possibly pre-existing pool to the end of the pools
845 * list; if it's new (and empty) then it'll be removed and
846 * destroyed by the put after we drop the lock
847 */
848 list_add_tail_rcu(&pool->list, &zswap_pools);
849 put_pool = pool;
fd5bb66c
DS
850 }
851
852 spin_unlock(&zswap_pools_lock);
853
854 if (!zswap_has_pool && !pool) {
ae3d89a7
DS
855 /* if initial pool creation failed, and this pool creation also
856 * failed, maybe both compressor and zpool params were bad.
857 * Allow changing this param, so pool creation will succeed
858 * when the other param is changed. We already verified this
1ec3b5fe 859 * param is ok in the zpool_has_pool() or crypto_has_acomp()
ae3d89a7
DS
860 * checks above.
861 */
862 ret = param_set_charp(s, kp);
90b0fc26
DS
863 }
864
90b0fc26
DS
865 /* drop the ref from either the old current pool,
866 * or the new pool we failed to add
867 */
868 if (put_pool)
869 zswap_pool_put(put_pool);
870
871 return ret;
872}
873
874static int zswap_compressor_param_set(const char *val,
875 const struct kernel_param *kp)
876{
877 return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
878}
879
880static int zswap_zpool_param_set(const char *val,
881 const struct kernel_param *kp)
882{
883 return __zswap_param_set(val, kp, NULL, zswap_compressor);
884}
885
d7b028f5
DS
886static int zswap_enabled_param_set(const char *val,
887 const struct kernel_param *kp)
888{
141fdeec
LS
889 int ret = -ENODEV;
890
891 /* if this is load-time (pre-init) param setting, only set param. */
892 if (system_state != SYSTEM_RUNNING)
893 return param_set_bool(val, kp);
894
895 mutex_lock(&zswap_init_lock);
9021ccec
LS
896 switch (zswap_init_state) {
897 case ZSWAP_UNINIT:
141fdeec
LS
898 if (zswap_setup())
899 break;
900 fallthrough;
9021ccec 901 case ZSWAP_INIT_SUCCEED:
141fdeec 902 if (!zswap_has_pool)
9021ccec 903 pr_err("can't enable, no pool configured\n");
141fdeec
LS
904 else
905 ret = param_set_bool(val, kp);
906 break;
9021ccec 907 case ZSWAP_INIT_FAILED:
d7b028f5 908 pr_err("can't enable, initialization failed\n");
ae3d89a7 909 }
141fdeec 910 mutex_unlock(&zswap_init_lock);
d7b028f5 911
141fdeec 912 return ret;
d7b028f5
DS
913}
914
2b281117
SJ
915/*********************************
916* writeback code
917**********************************/
918/* return enum for zswap_get_swap_cache_page */
919enum zswap_get_swap_ret {
920 ZSWAP_SWAPCACHE_NEW,
921 ZSWAP_SWAPCACHE_EXIST,
67d13fe8 922 ZSWAP_SWAPCACHE_FAIL,
2b281117
SJ
923};
924
925/*
926 * zswap_get_swap_cache_page
927 *
928 * This is an adaption of read_swap_cache_async()
929 *
930 * This function tries to find a page with the given swap entry
931 * in the swapper_space address space (the swap cache). If the page
932 * is found, it is returned in retpage. Otherwise, a page is allocated,
933 * added to the swap cache, and returned in retpage.
934 *
935 * If success, the swap cache page is returned in retpage
67d13fe8
WY
936 * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
937 * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
938 * the new page is added to swapcache and locked
939 * Returns ZSWAP_SWAPCACHE_FAIL on error
2b281117
SJ
940 */
941static int zswap_get_swap_cache_page(swp_entry_t entry,
942 struct page **retpage)
943{
5b999aad 944 bool page_was_allocated;
2b281117 945
5b999aad
DS
946 *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
947 NULL, 0, &page_was_allocated);
948 if (page_was_allocated)
949 return ZSWAP_SWAPCACHE_NEW;
950 if (!*retpage)
67d13fe8 951 return ZSWAP_SWAPCACHE_FAIL;
2b281117
SJ
952 return ZSWAP_SWAPCACHE_EXIST;
953}
954
955/*
956 * Attempts to free an entry by adding a page to the swap cache,
957 * decompressing the entry data into the page, and issuing a
958 * bio write to write the page back to the swap device.
959 *
960 * This can be thought of as a "resumed writeback" of the page
961 * to the swap device. We are basically resuming the same swap
962 * writeback path that was intercepted with the frontswap_store()
963 * in the first place. After the page has been decompressed into
964 * the swap cache, the compressed version stored by zswap can be
965 * freed.
966 */
12d79d64 967static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
2b281117
SJ
968{
969 struct zswap_header *zhdr;
970 swp_entry_t swpentry;
971 struct zswap_tree *tree;
972 pgoff_t offset;
973 struct zswap_entry *entry;
974 struct page *page;
1ec3b5fe
BS
975 struct scatterlist input, output;
976 struct crypto_acomp_ctx *acomp_ctx;
977
fc6697a8 978 u8 *src, *tmp = NULL;
2b281117 979 unsigned int dlen;
0ab0abcf 980 int ret;
2b281117
SJ
981 struct writeback_control wbc = {
982 .sync_mode = WB_SYNC_NONE,
983 };
984
fc6697a8 985 if (!zpool_can_sleep_mapped(pool)) {
8d9b6370 986 tmp = kmalloc(PAGE_SIZE, GFP_KERNEL);
fc6697a8
TT
987 if (!tmp)
988 return -ENOMEM;
989 }
990
2b281117 991 /* extract swpentry from data */
12d79d64 992 zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
2b281117 993 swpentry = zhdr->swpentry; /* here */
2b281117
SJ
994 tree = zswap_trees[swp_type(swpentry)];
995 offset = swp_offset(swpentry);
6b3379e8 996 zpool_unmap_handle(pool, handle);
2b281117
SJ
997
998 /* find and ref zswap entry */
999 spin_lock(&tree->lock);
0ab0abcf 1000 entry = zswap_entry_find_get(&tree->rbroot, offset);
2b281117
SJ
1001 if (!entry) {
1002 /* entry was invalidated */
1003 spin_unlock(&tree->lock);
fc6697a8 1004 kfree(tmp);
2b281117
SJ
1005 return 0;
1006 }
2b281117
SJ
1007 spin_unlock(&tree->lock);
1008 BUG_ON(offset != entry->offset);
1009
1010 /* try to allocate swap cache page */
1011 switch (zswap_get_swap_cache_page(swpentry, &page)) {
67d13fe8 1012 case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
2b281117
SJ
1013 ret = -ENOMEM;
1014 goto fail;
1015
67d13fe8 1016 case ZSWAP_SWAPCACHE_EXIST:
2b281117 1017 /* page is already in the swap cache, ignore for now */
09cbfeaf 1018 put_page(page);
2b281117
SJ
1019 ret = -EEXIST;
1020 goto fail;
1021
1022 case ZSWAP_SWAPCACHE_NEW: /* page is locked */
1023 /* decompress */
1ec3b5fe 1024 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
2b281117 1025 dlen = PAGE_SIZE;
fc6697a8 1026
6b3379e8
JW
1027 zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
1028 src = (u8 *)zhdr + sizeof(struct zswap_header);
1029 if (!zpool_can_sleep_mapped(pool)) {
1030 memcpy(tmp, src, entry->length);
1031 src = tmp;
1032 zpool_unmap_handle(pool, handle);
1033 }
1034
1ec3b5fe
BS
1035 mutex_lock(acomp_ctx->mutex);
1036 sg_init_one(&input, src, entry->length);
1037 sg_init_table(&output, 1);
1038 sg_set_page(&output, page, PAGE_SIZE, 0);
1039 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1040 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1041 dlen = acomp_ctx->req->dlen;
1042 mutex_unlock(acomp_ctx->mutex);
1043
6b3379e8
JW
1044 if (!zpool_can_sleep_mapped(pool))
1045 kfree(tmp);
1046 else
1047 zpool_unmap_handle(pool, handle);
1048
2b281117
SJ
1049 BUG_ON(ret);
1050 BUG_ON(dlen != PAGE_SIZE);
1051
1052 /* page is up to date */
1053 SetPageUptodate(page);
1054 }
1055
b349acc7
WY
1056 /* move it to the tail of the inactive list after end_writeback */
1057 SetPageReclaim(page);
1058
2b281117 1059 /* start writeback */
cf1e3fe4 1060 __swap_writepage(page, &wbc);
09cbfeaf 1061 put_page(page);
2b281117
SJ
1062 zswap_written_back_pages++;
1063
1064 spin_lock(&tree->lock);
2b281117 1065 /* drop local reference */
0ab0abcf 1066 zswap_entry_put(tree, entry);
2b281117
SJ
1067
1068 /*
0ab0abcf
WY
1069 * There are two possible situations for entry here:
1070 * (1) refcount is 1(normal case), entry is valid and on the tree
1071 * (2) refcount is 0, entry is freed and not on the tree
1072 * because invalidate happened during writeback
1073 * search the tree and free the entry if find entry
1074 */
1075 if (entry == zswap_rb_search(&tree->rbroot, offset))
1076 zswap_entry_put(tree, entry);
2b281117 1077 spin_unlock(&tree->lock);
2b281117 1078
6b3379e8
JW
1079 return ret;
1080
1081fail:
1082 if (!zpool_can_sleep_mapped(pool))
1083 kfree(tmp);
0ab0abcf
WY
1084
1085 /*
1086 * if we get here due to ZSWAP_SWAPCACHE_EXIST
c0c641d7
RD
1087 * a load may be happening concurrently.
1088 * it is safe and okay to not free the entry.
0ab0abcf 1089 * if we free the entry in the following put
c0c641d7 1090 * it is also okay to return !0
0ab0abcf 1091 */
2b281117 1092 spin_lock(&tree->lock);
0ab0abcf 1093 zswap_entry_put(tree, entry);
2b281117 1094 spin_unlock(&tree->lock);
0ab0abcf 1095
2b281117
SJ
1096 return ret;
1097}
1098
a85f878b
SD
1099static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
1100{
a85f878b 1101 unsigned long *page;
62bf1258
TS
1102 unsigned long val;
1103 unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
a85f878b
SD
1104
1105 page = (unsigned long *)ptr;
62bf1258
TS
1106 val = page[0];
1107
1108 if (val != page[last_pos])
1109 return 0;
1110
1111 for (pos = 1; pos < last_pos; pos++) {
1112 if (val != page[pos])
a85f878b
SD
1113 return 0;
1114 }
62bf1258
TS
1115
1116 *value = val;
1117
a85f878b
SD
1118 return 1;
1119}
1120
1121static void zswap_fill_page(void *ptr, unsigned long value)
1122{
1123 unsigned long *page;
1124
1125 page = (unsigned long *)ptr;
1126 memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
1127}
1128
2b281117
SJ
1129/*********************************
1130* frontswap hooks
1131**********************************/
1132/* attempts to compress and store an single page */
1133static int zswap_frontswap_store(unsigned type, pgoff_t offset,
1134 struct page *page)
1135{
1136 struct zswap_tree *tree = zswap_trees[type];
1137 struct zswap_entry *entry, *dupentry;
1ec3b5fe
BS
1138 struct scatterlist input, output;
1139 struct crypto_acomp_ctx *acomp_ctx;
f4840ccf
JW
1140 struct obj_cgroup *objcg = NULL;
1141 struct zswap_pool *pool;
2b281117 1142 int ret;
9c3760eb 1143 unsigned int hlen, dlen = PAGE_SIZE;
a85f878b 1144 unsigned long handle, value;
2b281117
SJ
1145 char *buf;
1146 u8 *src, *dst;
9c3760eb 1147 struct zswap_header zhdr = { .swpentry = swp_entry(type, offset) };
d2fcd82b 1148 gfp_t gfp;
2b281117 1149
7ba71669
HY
1150 /* THP isn't supported */
1151 if (PageTransHuge(page)) {
1152 ret = -EINVAL;
1153 goto reject;
1154 }
1155
c00ed16a 1156 if (!zswap_enabled || !tree) {
2b281117
SJ
1157 ret = -ENODEV;
1158 goto reject;
1159 }
1160
f4840ccf
JW
1161 objcg = get_obj_cgroup_from_page(page);
1162 if (objcg && !obj_cgroup_may_zswap(objcg))
1163 goto shrink;
1164
2b281117
SJ
1165 /* reclaim space if needed */
1166 if (zswap_is_full()) {
1167 zswap_pool_limit_hit++;
45190f01 1168 zswap_pool_reached_full = true;
f4840ccf 1169 goto shrink;
45190f01 1170 }
16e536ef 1171
45190f01
VW
1172 if (zswap_pool_reached_full) {
1173 if (!zswap_can_accept()) {
16e536ef
LW
1174 ret = -ENOMEM;
1175 goto reject;
45190f01
VW
1176 } else
1177 zswap_pool_reached_full = false;
2b281117
SJ
1178 }
1179
1180 /* allocate entry */
1181 entry = zswap_entry_cache_alloc(GFP_KERNEL);
1182 if (!entry) {
1183 zswap_reject_kmemcache_fail++;
1184 ret = -ENOMEM;
1185 goto reject;
1186 }
1187
a85f878b
SD
1188 if (zswap_same_filled_pages_enabled) {
1189 src = kmap_atomic(page);
1190 if (zswap_is_page_same_filled(src, &value)) {
1191 kunmap_atomic(src);
1192 entry->offset = offset;
1193 entry->length = 0;
1194 entry->value = value;
1195 atomic_inc(&zswap_same_filled_pages);
1196 goto insert_entry;
1197 }
1198 kunmap_atomic(src);
1199 }
1200
cb325ddd
MS
1201 if (!zswap_non_same_filled_pages_enabled) {
1202 ret = -EINVAL;
1203 goto freepage;
1204 }
1205
f1c54846
DS
1206 /* if entry is successfully added, it keeps the reference */
1207 entry->pool = zswap_pool_current_get();
1208 if (!entry->pool) {
1209 ret = -EINVAL;
1210 goto freepage;
1211 }
1212
2b281117 1213 /* compress */
1ec3b5fe
BS
1214 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1215
1216 mutex_lock(acomp_ctx->mutex);
1217
1218 dst = acomp_ctx->dstmem;
1219 sg_init_table(&input, 1);
1220 sg_set_page(&input, page, PAGE_SIZE, 0);
1221
1222 /* zswap_dstmem is of size (PAGE_SIZE * 2). Reflect same in sg_list */
1223 sg_init_one(&output, dst, PAGE_SIZE * 2);
1224 acomp_request_set_params(acomp_ctx->req, &input, &output, PAGE_SIZE, dlen);
1225 /*
1226 * it maybe looks a little bit silly that we send an asynchronous request,
1227 * then wait for its completion synchronously. This makes the process look
1228 * synchronous in fact.
1229 * Theoretically, acomp supports users send multiple acomp requests in one
1230 * acomp instance, then get those requests done simultaneously. but in this
1231 * case, frontswap actually does store and load page by page, there is no
1232 * existing method to send the second page before the first page is done
1233 * in one thread doing frontswap.
1234 * but in different threads running on different cpu, we have different
1235 * acomp instance, so multiple threads can do (de)compression in parallel.
1236 */
1237 ret = crypto_wait_req(crypto_acomp_compress(acomp_ctx->req), &acomp_ctx->wait);
1238 dlen = acomp_ctx->req->dlen;
1239
2b281117
SJ
1240 if (ret) {
1241 ret = -EINVAL;
f1c54846 1242 goto put_dstmem;
2b281117
SJ
1243 }
1244
1245 /* store */
9c3760eb 1246 hlen = zpool_evictable(entry->pool->zpool) ? sizeof(zhdr) : 0;
d2fcd82b
HZ
1247 gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1248 if (zpool_malloc_support_movable(entry->pool->zpool))
1249 gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1250 ret = zpool_malloc(entry->pool->zpool, hlen + dlen, gfp, &handle);
2b281117
SJ
1251 if (ret == -ENOSPC) {
1252 zswap_reject_compress_poor++;
f1c54846 1253 goto put_dstmem;
2b281117
SJ
1254 }
1255 if (ret) {
1256 zswap_reject_alloc_fail++;
f1c54846 1257 goto put_dstmem;
2b281117 1258 }
ae34af1f 1259 buf = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_WO);
9c3760eb
YZ
1260 memcpy(buf, &zhdr, hlen);
1261 memcpy(buf + hlen, dst, dlen);
f1c54846 1262 zpool_unmap_handle(entry->pool->zpool, handle);
1ec3b5fe 1263 mutex_unlock(acomp_ctx->mutex);
2b281117
SJ
1264
1265 /* populate entry */
1266 entry->offset = offset;
1267 entry->handle = handle;
1268 entry->length = dlen;
1269
a85f878b 1270insert_entry:
f4840ccf
JW
1271 entry->objcg = objcg;
1272 if (objcg) {
1273 obj_cgroup_charge_zswap(objcg, entry->length);
1274 /* Account before objcg ref is moved to tree */
1275 count_objcg_event(objcg, ZSWPOUT);
1276 }
1277
2b281117
SJ
1278 /* map */
1279 spin_lock(&tree->lock);
1280 do {
1281 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1282 if (ret == -EEXIST) {
1283 zswap_duplicate_entry++;
1284 /* remove from rbtree */
0ab0abcf
WY
1285 zswap_rb_erase(&tree->rbroot, dupentry);
1286 zswap_entry_put(tree, dupentry);
2b281117
SJ
1287 }
1288 } while (ret == -EEXIST);
1289 spin_unlock(&tree->lock);
1290
1291 /* update stats */
1292 atomic_inc(&zswap_stored_pages);
f1c54846 1293 zswap_update_total_size();
f6498b77 1294 count_vm_event(ZSWPOUT);
2b281117
SJ
1295
1296 return 0;
1297
f1c54846 1298put_dstmem:
1ec3b5fe 1299 mutex_unlock(acomp_ctx->mutex);
f1c54846
DS
1300 zswap_pool_put(entry->pool);
1301freepage:
2b281117
SJ
1302 zswap_entry_cache_free(entry);
1303reject:
f4840ccf
JW
1304 if (objcg)
1305 obj_cgroup_put(objcg);
2b281117 1306 return ret;
f4840ccf
JW
1307
1308shrink:
1309 pool = zswap_pool_last_get();
1310 if (pool)
1311 queue_work(shrink_wq, &pool->shrink_work);
1312 ret = -ENOMEM;
1313 goto reject;
2b281117
SJ
1314}
1315
1316/*
1317 * returns 0 if the page was successfully decompressed
1318 * return -1 on entry not found or error
1319*/
1320static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1321 struct page *page)
1322{
1323 struct zswap_tree *tree = zswap_trees[type];
1324 struct zswap_entry *entry;
1ec3b5fe
BS
1325 struct scatterlist input, output;
1326 struct crypto_acomp_ctx *acomp_ctx;
fc6697a8 1327 u8 *src, *dst, *tmp;
2b281117 1328 unsigned int dlen;
0ab0abcf 1329 int ret;
2b281117
SJ
1330
1331 /* find */
1332 spin_lock(&tree->lock);
0ab0abcf 1333 entry = zswap_entry_find_get(&tree->rbroot, offset);
2b281117
SJ
1334 if (!entry) {
1335 /* entry was written back */
1336 spin_unlock(&tree->lock);
1337 return -1;
1338 }
2b281117
SJ
1339 spin_unlock(&tree->lock);
1340
a85f878b
SD
1341 if (!entry->length) {
1342 dst = kmap_atomic(page);
1343 zswap_fill_page(dst, entry->value);
1344 kunmap_atomic(dst);
fc6697a8 1345 ret = 0;
f6498b77 1346 goto stats;
a85f878b
SD
1347 }
1348
fc6697a8 1349 if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
8d9b6370 1350 tmp = kmalloc(entry->length, GFP_KERNEL);
fc6697a8
TT
1351 if (!tmp) {
1352 ret = -ENOMEM;
1353 goto freeentry;
1354 }
1355 }
1356
2b281117
SJ
1357 /* decompress */
1358 dlen = PAGE_SIZE;
9c3760eb
YZ
1359 src = zpool_map_handle(entry->pool->zpool, entry->handle, ZPOOL_MM_RO);
1360 if (zpool_evictable(entry->pool->zpool))
1361 src += sizeof(struct zswap_header);
1ec3b5fe 1362
fc6697a8 1363 if (!zpool_can_sleep_mapped(entry->pool->zpool)) {
fc6697a8
TT
1364 memcpy(tmp, src, entry->length);
1365 src = tmp;
fc6697a8
TT
1366 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1367 }
1368
1ec3b5fe
BS
1369 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1370 mutex_lock(acomp_ctx->mutex);
1371 sg_init_one(&input, src, entry->length);
1372 sg_init_table(&output, 1);
1373 sg_set_page(&output, page, PAGE_SIZE, 0);
1374 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1375 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1376 mutex_unlock(acomp_ctx->mutex);
1377
fc6697a8
TT
1378 if (zpool_can_sleep_mapped(entry->pool->zpool))
1379 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1380 else
1381 kfree(tmp);
1382
2b281117 1383 BUG_ON(ret);
f6498b77
JW
1384stats:
1385 count_vm_event(ZSWPIN);
f4840ccf
JW
1386 if (entry->objcg)
1387 count_objcg_event(entry->objcg, ZSWPIN);
a85f878b 1388freeentry:
2b281117 1389 spin_lock(&tree->lock);
0ab0abcf 1390 zswap_entry_put(tree, entry);
2b281117
SJ
1391 spin_unlock(&tree->lock);
1392
fc6697a8 1393 return ret;
2b281117
SJ
1394}
1395
1396/* frees an entry in zswap */
1397static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1398{
1399 struct zswap_tree *tree = zswap_trees[type];
1400 struct zswap_entry *entry;
2b281117
SJ
1401
1402 /* find */
1403 spin_lock(&tree->lock);
1404 entry = zswap_rb_search(&tree->rbroot, offset);
1405 if (!entry) {
1406 /* entry was written back */
1407 spin_unlock(&tree->lock);
1408 return;
1409 }
1410
1411 /* remove from rbtree */
0ab0abcf 1412 zswap_rb_erase(&tree->rbroot, entry);
2b281117
SJ
1413
1414 /* drop the initial reference from entry creation */
0ab0abcf 1415 zswap_entry_put(tree, entry);
2b281117
SJ
1416
1417 spin_unlock(&tree->lock);
2b281117
SJ
1418}
1419
1420/* frees all zswap entries for the given swap type */
1421static void zswap_frontswap_invalidate_area(unsigned type)
1422{
1423 struct zswap_tree *tree = zswap_trees[type];
0bd42136 1424 struct zswap_entry *entry, *n;
2b281117
SJ
1425
1426 if (!tree)
1427 return;
1428
1429 /* walk the tree and free everything */
1430 spin_lock(&tree->lock);
0ab0abcf 1431 rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
60105e12 1432 zswap_free_entry(entry);
2b281117
SJ
1433 tree->rbroot = RB_ROOT;
1434 spin_unlock(&tree->lock);
aa9bca05
WY
1435 kfree(tree);
1436 zswap_trees[type] = NULL;
2b281117
SJ
1437}
1438
2b281117
SJ
1439static void zswap_frontswap_init(unsigned type)
1440{
1441 struct zswap_tree *tree;
1442
9cd1f701 1443 tree = kzalloc(sizeof(*tree), GFP_KERNEL);
60105e12
MK
1444 if (!tree) {
1445 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1446 return;
1447 }
1448
2b281117
SJ
1449 tree->rbroot = RB_ROOT;
1450 spin_lock_init(&tree->lock);
1451 zswap_trees[type] = tree;
2b281117
SJ
1452}
1453
1da0d94a 1454static const struct frontswap_ops zswap_frontswap_ops = {
2b281117
SJ
1455 .store = zswap_frontswap_store,
1456 .load = zswap_frontswap_load,
1457 .invalidate_page = zswap_frontswap_invalidate_page,
1458 .invalidate_area = zswap_frontswap_invalidate_area,
1459 .init = zswap_frontswap_init
1460};
1461
1462/*********************************
1463* debugfs functions
1464**********************************/
1465#ifdef CONFIG_DEBUG_FS
1466#include <linux/debugfs.h>
1467
1468static struct dentry *zswap_debugfs_root;
1469
141fdeec 1470static int zswap_debugfs_init(void)
2b281117
SJ
1471{
1472 if (!debugfs_initialized())
1473 return -ENODEV;
1474
1475 zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
2b281117 1476
0825a6f9
JP
1477 debugfs_create_u64("pool_limit_hit", 0444,
1478 zswap_debugfs_root, &zswap_pool_limit_hit);
1479 debugfs_create_u64("reject_reclaim_fail", 0444,
1480 zswap_debugfs_root, &zswap_reject_reclaim_fail);
1481 debugfs_create_u64("reject_alloc_fail", 0444,
1482 zswap_debugfs_root, &zswap_reject_alloc_fail);
1483 debugfs_create_u64("reject_kmemcache_fail", 0444,
1484 zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1485 debugfs_create_u64("reject_compress_poor", 0444,
1486 zswap_debugfs_root, &zswap_reject_compress_poor);
1487 debugfs_create_u64("written_back_pages", 0444,
1488 zswap_debugfs_root, &zswap_written_back_pages);
1489 debugfs_create_u64("duplicate_entry", 0444,
1490 zswap_debugfs_root, &zswap_duplicate_entry);
1491 debugfs_create_u64("pool_total_size", 0444,
1492 zswap_debugfs_root, &zswap_pool_total_size);
1493 debugfs_create_atomic_t("stored_pages", 0444,
1494 zswap_debugfs_root, &zswap_stored_pages);
a85f878b 1495 debugfs_create_atomic_t("same_filled_pages", 0444,
0825a6f9 1496 zswap_debugfs_root, &zswap_same_filled_pages);
2b281117
SJ
1497
1498 return 0;
1499}
2b281117 1500#else
141fdeec 1501static int zswap_debugfs_init(void)
2b281117
SJ
1502{
1503 return 0;
1504}
2b281117
SJ
1505#endif
1506
1507/*********************************
1508* module init and exit
1509**********************************/
141fdeec 1510static int zswap_setup(void)
2b281117 1511{
f1c54846 1512 struct zswap_pool *pool;
ad7ed770 1513 int ret;
60105e12 1514
b7919122
LS
1515 zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
1516 if (!zswap_entry_cache) {
2b281117 1517 pr_err("entry cache creation failed\n");
f1c54846 1518 goto cache_fail;
2b281117 1519 }
f1c54846 1520
ad7ed770
SAS
1521 ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1522 zswap_dstmem_prepare, zswap_dstmem_dead);
1523 if (ret) {
f1c54846
DS
1524 pr_err("dstmem alloc failed\n");
1525 goto dstmem_fail;
2b281117 1526 }
f1c54846 1527
cab7a7e5
SAS
1528 ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1529 "mm/zswap_pool:prepare",
1530 zswap_cpu_comp_prepare,
1531 zswap_cpu_comp_dead);
1532 if (ret)
1533 goto hp_fail;
1534
f1c54846 1535 pool = __zswap_pool_create_fallback();
ae3d89a7
DS
1536 if (pool) {
1537 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1538 zpool_get_type(pool->zpool));
1539 list_add(&pool->list, &zswap_pools);
1540 zswap_has_pool = true;
1541 } else {
f1c54846 1542 pr_err("pool creation failed\n");
ae3d89a7 1543 zswap_enabled = false;
2b281117 1544 }
60105e12 1545
45190f01
VW
1546 shrink_wq = create_workqueue("zswap-shrink");
1547 if (!shrink_wq)
1548 goto fallback_fail;
1549
1da0d94a
CH
1550 ret = frontswap_register_ops(&zswap_frontswap_ops);
1551 if (ret)
1552 goto destroy_wq;
2b281117
SJ
1553 if (zswap_debugfs_init())
1554 pr_warn("debugfs initialization failed\n");
9021ccec 1555 zswap_init_state = ZSWAP_INIT_SUCCEED;
2b281117 1556 return 0;
f1c54846 1557
1da0d94a
CH
1558destroy_wq:
1559 destroy_workqueue(shrink_wq);
45190f01 1560fallback_fail:
38aeb071
DC
1561 if (pool)
1562 zswap_pool_destroy(pool);
cab7a7e5 1563hp_fail:
ad7ed770 1564 cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
f1c54846 1565dstmem_fail:
b7919122 1566 kmem_cache_destroy(zswap_entry_cache);
f1c54846 1567cache_fail:
d7b028f5 1568 /* if built-in, we aren't unloaded on failure; don't allow use */
9021ccec 1569 zswap_init_state = ZSWAP_INIT_FAILED;
d7b028f5 1570 zswap_enabled = false;
2b281117
SJ
1571 return -ENOMEM;
1572}
141fdeec
LS
1573
1574static int __init zswap_init(void)
1575{
1576 if (!zswap_enabled)
1577 return 0;
1578 return zswap_setup();
1579}
2b281117 1580/* must be late so crypto has time to come up */
141fdeec 1581late_initcall(zswap_init);
2b281117 1582
68386da8 1583MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
2b281117 1584MODULE_DESCRIPTION("Compressed cache for swap pages");