1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * zswap.c - zswap driver file
4 *
5 * zswap is a cache 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/rbtree.h>
24 #include <linux/swap.h>
25 #include <linux/crypto.h>
26 #include <linux/scatterlist.h>
27 #include <linux/mempool.h>
28 #include <linux/zpool.h>
29 #include <crypto/acompress.h>
30 #include <linux/zswap.h>
31 #include <linux/mm_types.h>
32 #include <linux/page-flags.h>
33 #include <linux/swapops.h>
34 #include <linux/writeback.h>
35 #include <linux/pagemap.h>
36 #include <linux/workqueue.h>
37
38 #include "swap.h"
39 #include "internal.h"
40
41 /*********************************
42 * statistics
43 **********************************/
44 /* Total bytes used by the compressed storage */
45 u64 zswap_pool_total_size;
46 /* The number of compressed pages currently stored in zswap */
47 atomic_t zswap_stored_pages = ATOMIC_INIT(0);
48 /* The number of same-value filled pages currently stored in zswap */
49 static 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) */
59 static u64 zswap_pool_limit_hit;
60 /* Pages written back when pool limit was reached */
61 static u64 zswap_written_back_pages;
62 /* Store failed due to a reclaim failure after pool limit was reached */
63 static u64 zswap_reject_reclaim_fail;
64 /* Compressed page was too big for the allocator to (optimally) store */
65 static u64 zswap_reject_compress_poor;
66 /* Store failed because underlying allocator could not get memory */
67 static u64 zswap_reject_alloc_fail;
68 /* Store failed because the entry metadata could not be allocated (rare) */
69 static u64 zswap_reject_kmemcache_fail;
70 /* Duplicate store was encountered (rare) */
71 static u64 zswap_duplicate_entry;
72
73 /* Shrinker work queue */
74 static struct workqueue_struct *shrink_wq;
75 /* Pool limit was hit, we need to calm down */
76 static bool zswap_pool_reached_full;
77
78 /*********************************
79 * tunables
80 **********************************/
81
82 #define ZSWAP_PARAM_UNSET ""
83
84 static int zswap_setup(void);
85
86 /* Enable/disable zswap */
87 static bool zswap_enabled = IS_ENABLED(CONFIG_ZSWAP_DEFAULT_ON);
88 static int zswap_enabled_param_set(const char *,
89 const struct kernel_param *);
90 static const struct kernel_param_ops zswap_enabled_param_ops = {
91 .set = zswap_enabled_param_set,
92 .get = param_get_bool,
93 };
94 module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
95
96 /* Crypto compressor to use */
97 static char *zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
98 static int zswap_compressor_param_set(const char *,
99 const struct kernel_param *);
100 static const struct kernel_param_ops zswap_compressor_param_ops = {
101 .set = zswap_compressor_param_set,
102 .get = param_get_charp,
103 .free = param_free_charp,
104 };
105 module_param_cb(compressor, &zswap_compressor_param_ops,
106 &zswap_compressor, 0644);
107
108 /* Compressed storage zpool to use */
109 static char *zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
110 static int zswap_zpool_param_set(const char *, const struct kernel_param *);
111 static const struct kernel_param_ops zswap_zpool_param_ops = {
112 .set = zswap_zpool_param_set,
113 .get = param_get_charp,
114 .free = param_free_charp,
115 };
116 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
117
118 /* The maximum percentage of memory that the compressed pool can occupy */
119 static unsigned int zswap_max_pool_percent = 20;
120 module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
121
122 /* The threshold for accepting new pages after the max_pool_percent was hit */
123 static unsigned int zswap_accept_thr_percent = 90; /* of max pool size */
124 module_param_named(accept_threshold_percent, zswap_accept_thr_percent,
125 uint, 0644);
126
127 /*
128 * Enable/disable handling same-value filled pages (enabled by default).
129 * If disabled every page is considered non-same-value filled.
130 */
131 static bool zswap_same_filled_pages_enabled = true;
132 module_param_named(same_filled_pages_enabled, zswap_same_filled_pages_enabled,
133 bool, 0644);
134
135 /* Enable/disable handling non-same-value filled pages (enabled by default) */
136 static bool zswap_non_same_filled_pages_enabled = true;
137 module_param_named(non_same_filled_pages_enabled, zswap_non_same_filled_pages_enabled,
138 bool, 0644);
139
140 static bool zswap_exclusive_loads_enabled = IS_ENABLED(
141 CONFIG_ZSWAP_EXCLUSIVE_LOADS_DEFAULT_ON);
142 module_param_named(exclusive_loads, zswap_exclusive_loads_enabled, bool, 0644);
143
144 /* Number of zpools in zswap_pool (empirically determined for scalability) */
145 #define ZSWAP_NR_ZPOOLS 32
146
147 /*********************************
148 * data structures
149 **********************************/
150
151 struct crypto_acomp_ctx {
152 struct crypto_acomp *acomp;
153 struct acomp_req *req;
154 struct crypto_wait wait;
155 u8 *dstmem;
156 struct mutex *mutex;
157 };
158
159 /*
160 * The lock ordering is zswap_tree.lock -> zswap_pool.lru_lock.
161 * The only case where lru_lock is not acquired while holding tree.lock is
162 * when a zswap_entry is taken off the lru for writeback, in that case it
163 * needs to be verified that it's still valid in the tree.
164 */
165 struct zswap_pool {
166 struct zpool *zpools[ZSWAP_NR_ZPOOLS];
167 struct crypto_acomp_ctx __percpu *acomp_ctx;
168 struct kref kref;
169 struct list_head list;
170 struct work_struct release_work;
171 struct work_struct shrink_work;
172 struct hlist_node node;
173 char tfm_name[CRYPTO_MAX_ALG_NAME];
174 struct list_head lru;
175 spinlock_t lru_lock;
176 };
177
178 /*
179 * struct zswap_entry
180 *
181 * This structure contains the metadata for tracking a single compressed
182 * page within zswap.
183 *
184 * rbnode - links the entry into red-black tree for the appropriate swap type
185 * swpentry - associated swap entry, the offset indexes into the red-black tree
186 * refcount - the number of outstanding reference to the entry. This is needed
187 * to protect against premature freeing of the entry by code
188 * concurrent calls to load, invalidate, and writeback. The lock
189 * for the zswap_tree structure that contains the entry must
190 * be held while changing the refcount. Since the lock must
191 * be held, there is no reason to also make refcount atomic.
192 * length - the length in bytes of the compressed page data. Needed during
193 * decompression. For a same value filled page length is 0, and both
194 * pool and lru are invalid and must be ignored.
195 * pool - the zswap_pool the entry's data is in
196 * handle - zpool allocation handle that stores the compressed page data
197 * value - value of the same-value filled pages which have same content
198 * objcg - the obj_cgroup that the compressed memory is charged to
199 * lru - handle to the pool's lru used to evict pages.
200 */
201 struct zswap_entry {
202 struct rb_node rbnode;
203 swp_entry_t swpentry;
204 int refcount;
205 unsigned int length;
206 struct zswap_pool *pool;
207 union {
208 unsigned long handle;
209 unsigned long value;
210 };
211 struct obj_cgroup *objcg;
212 struct list_head lru;
213 };
214
215 /*
216 * The tree lock in the zswap_tree struct protects a few things:
217 * - the rbtree
218 * - the refcount field of each entry in the tree
219 */
220 struct zswap_tree {
221 struct rb_root rbroot;
222 spinlock_t lock;
223 };
224
225 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
226
227 /* RCU-protected iteration */
228 static LIST_HEAD(zswap_pools);
229 /* protects zswap_pools list modification */
230 static DEFINE_SPINLOCK(zswap_pools_lock);
231 /* pool counter to provide unique names to zpool */
232 static atomic_t zswap_pools_count = ATOMIC_INIT(0);
233
234 enum zswap_init_type {
235 ZSWAP_UNINIT,
236 ZSWAP_INIT_SUCCEED,
237 ZSWAP_INIT_FAILED
238 };
239
240 static enum zswap_init_type zswap_init_state;
241
242 /* used to ensure the integrity of initialization */
243 static DEFINE_MUTEX(zswap_init_lock);
244
245 /* init completed, but couldn't create the initial pool */
246 static bool zswap_has_pool;
247
248 /*********************************
249 * helpers and fwd declarations
250 **********************************/
251
252 #define zswap_pool_debug(msg, p) \
253 pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name, \
254 zpool_get_type((p)->zpools[0]))
255
256 static int zswap_writeback_entry(struct zswap_entry *entry,
257 struct zswap_tree *tree);
258 static int zswap_pool_get(struct zswap_pool *pool);
259 static void zswap_pool_put(struct zswap_pool *pool);
260
zswap_is_full(void)261 static bool zswap_is_full(void)
262 {
263 return totalram_pages() * zswap_max_pool_percent / 100 <
264 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
265 }
266
zswap_can_accept(void)267 static bool zswap_can_accept(void)
268 {
269 return totalram_pages() * zswap_accept_thr_percent / 100 *
270 zswap_max_pool_percent / 100 >
271 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
272 }
273
zswap_update_total_size(void)274 static void zswap_update_total_size(void)
275 {
276 struct zswap_pool *pool;
277 u64 total = 0;
278 int i;
279
280 rcu_read_lock();
281
282 list_for_each_entry_rcu(pool, &zswap_pools, list)
283 for (i = 0; i < ZSWAP_NR_ZPOOLS; i++)
284 total += zpool_get_total_size(pool->zpools[i]);
285
286 rcu_read_unlock();
287
288 zswap_pool_total_size = total;
289 }
290
291 /*********************************
292 * zswap entry functions
293 **********************************/
294 static struct kmem_cache *zswap_entry_cache;
295
zswap_entry_cache_alloc(gfp_t gfp)296 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
297 {
298 struct zswap_entry *entry;
299 entry = kmem_cache_alloc(zswap_entry_cache, gfp);
300 if (!entry)
301 return NULL;
302 entry->refcount = 1;
303 RB_CLEAR_NODE(&entry->rbnode);
304 return entry;
305 }
306
zswap_entry_cache_free(struct zswap_entry * entry)307 static void zswap_entry_cache_free(struct zswap_entry *entry)
308 {
309 kmem_cache_free(zswap_entry_cache, entry);
310 }
311
312 /*********************************
313 * rbtree functions
314 **********************************/
zswap_rb_search(struct rb_root * root,pgoff_t offset)315 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
316 {
317 struct rb_node *node = root->rb_node;
318 struct zswap_entry *entry;
319 pgoff_t entry_offset;
320
321 while (node) {
322 entry = rb_entry(node, struct zswap_entry, rbnode);
323 entry_offset = swp_offset(entry->swpentry);
324 if (entry_offset > offset)
325 node = node->rb_left;
326 else if (entry_offset < offset)
327 node = node->rb_right;
328 else
329 return entry;
330 }
331 return NULL;
332 }
333
334 /*
335 * In the case that a entry with the same offset is found, a pointer to
336 * the existing entry is stored in dupentry and the function returns -EEXIST
337 */
zswap_rb_insert(struct rb_root * root,struct zswap_entry * entry,struct zswap_entry ** dupentry)338 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
339 struct zswap_entry **dupentry)
340 {
341 struct rb_node **link = &root->rb_node, *parent = NULL;
342 struct zswap_entry *myentry;
343 pgoff_t myentry_offset, entry_offset = swp_offset(entry->swpentry);
344
345 while (*link) {
346 parent = *link;
347 myentry = rb_entry(parent, struct zswap_entry, rbnode);
348 myentry_offset = swp_offset(myentry->swpentry);
349 if (myentry_offset > entry_offset)
350 link = &(*link)->rb_left;
351 else if (myentry_offset < entry_offset)
352 link = &(*link)->rb_right;
353 else {
354 *dupentry = myentry;
355 return -EEXIST;
356 }
357 }
358 rb_link_node(&entry->rbnode, parent, link);
359 rb_insert_color(&entry->rbnode, root);
360 return 0;
361 }
362
zswap_rb_erase(struct rb_root * root,struct zswap_entry * entry)363 static bool zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
364 {
365 if (!RB_EMPTY_NODE(&entry->rbnode)) {
366 rb_erase(&entry->rbnode, root);
367 RB_CLEAR_NODE(&entry->rbnode);
368 return true;
369 }
370 return false;
371 }
372
zswap_find_zpool(struct zswap_entry * entry)373 static struct zpool *zswap_find_zpool(struct zswap_entry *entry)
374 {
375 int i = 0;
376
377 if (ZSWAP_NR_ZPOOLS > 1)
378 i = hash_ptr(entry, ilog2(ZSWAP_NR_ZPOOLS));
379
380 return entry->pool->zpools[i];
381 }
382
383 /*
384 * Carries out the common pattern of freeing and entry's zpool allocation,
385 * freeing the entry itself, and decrementing the number of stored pages.
386 */
zswap_free_entry(struct zswap_entry * entry)387 static void zswap_free_entry(struct zswap_entry *entry)
388 {
389 if (entry->objcg) {
390 obj_cgroup_uncharge_zswap(entry->objcg, entry->length);
391 obj_cgroup_put(entry->objcg);
392 }
393 if (!entry->length)
394 atomic_dec(&zswap_same_filled_pages);
395 else {
396 spin_lock(&entry->pool->lru_lock);
397 list_del(&entry->lru);
398 spin_unlock(&entry->pool->lru_lock);
399 zpool_free(zswap_find_zpool(entry), entry->handle);
400 zswap_pool_put(entry->pool);
401 }
402 zswap_entry_cache_free(entry);
403 atomic_dec(&zswap_stored_pages);
404 zswap_update_total_size();
405 }
406
407 /* caller must hold the tree lock */
zswap_entry_get(struct zswap_entry * entry)408 static void zswap_entry_get(struct zswap_entry *entry)
409 {
410 entry->refcount++;
411 }
412
413 /* caller must hold the tree lock
414 * remove from the tree and free it, if nobody reference the entry
415 */
zswap_entry_put(struct zswap_tree * tree,struct zswap_entry * entry)416 static void zswap_entry_put(struct zswap_tree *tree,
417 struct zswap_entry *entry)
418 {
419 int refcount = --entry->refcount;
420
421 WARN_ON_ONCE(refcount < 0);
422 if (refcount == 0) {
423 WARN_ON_ONCE(!RB_EMPTY_NODE(&entry->rbnode));
424 zswap_free_entry(entry);
425 }
426 }
427
428 /* caller must hold the tree lock */
zswap_entry_find_get(struct rb_root * root,pgoff_t offset)429 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
430 pgoff_t offset)
431 {
432 struct zswap_entry *entry;
433
434 entry = zswap_rb_search(root, offset);
435 if (entry)
436 zswap_entry_get(entry);
437
438 return entry;
439 }
440
441 /*********************************
442 * per-cpu code
443 **********************************/
444 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
445 /*
446 * If users dynamically change the zpool type and compressor at runtime, i.e.
447 * zswap is running, zswap can have more than one zpool on one cpu, but they
448 * are sharing dtsmem. So we need this mutex to be per-cpu.
449 */
450 static DEFINE_PER_CPU(struct mutex *, zswap_mutex);
451
zswap_dstmem_prepare(unsigned int cpu)452 static int zswap_dstmem_prepare(unsigned int cpu)
453 {
454 struct mutex *mutex;
455 u8 *dst;
456
457 dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
458 if (!dst)
459 return -ENOMEM;
460
461 mutex = kmalloc_node(sizeof(*mutex), GFP_KERNEL, cpu_to_node(cpu));
462 if (!mutex) {
463 kfree(dst);
464 return -ENOMEM;
465 }
466
467 mutex_init(mutex);
468 per_cpu(zswap_dstmem, cpu) = dst;
469 per_cpu(zswap_mutex, cpu) = mutex;
470 return 0;
471 }
472
zswap_dstmem_dead(unsigned int cpu)473 static int zswap_dstmem_dead(unsigned int cpu)
474 {
475 struct mutex *mutex;
476 u8 *dst;
477
478 mutex = per_cpu(zswap_mutex, cpu);
479 kfree(mutex);
480 per_cpu(zswap_mutex, cpu) = NULL;
481
482 dst = per_cpu(zswap_dstmem, cpu);
483 kfree(dst);
484 per_cpu(zswap_dstmem, cpu) = NULL;
485
486 return 0;
487 }
488
zswap_cpu_comp_prepare(unsigned int cpu,struct hlist_node * node)489 static int zswap_cpu_comp_prepare(unsigned int cpu, struct hlist_node *node)
490 {
491 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
492 struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
493 struct crypto_acomp *acomp;
494 struct acomp_req *req;
495
496 acomp = crypto_alloc_acomp_node(pool->tfm_name, 0, 0, cpu_to_node(cpu));
497 if (IS_ERR(acomp)) {
498 pr_err("could not alloc crypto acomp %s : %ld\n",
499 pool->tfm_name, PTR_ERR(acomp));
500 return PTR_ERR(acomp);
501 }
502 acomp_ctx->acomp = acomp;
503
504 req = acomp_request_alloc(acomp_ctx->acomp);
505 if (!req) {
506 pr_err("could not alloc crypto acomp_request %s\n",
507 pool->tfm_name);
508 crypto_free_acomp(acomp_ctx->acomp);
509 return -ENOMEM;
510 }
511 acomp_ctx->req = req;
512
513 crypto_init_wait(&acomp_ctx->wait);
514 /*
515 * if the backend of acomp is async zip, crypto_req_done() will wakeup
516 * crypto_wait_req(); if the backend of acomp is scomp, the callback
517 * won't be called, crypto_wait_req() will return without blocking.
518 */
519 acomp_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
520 crypto_req_done, &acomp_ctx->wait);
521
522 acomp_ctx->mutex = per_cpu(zswap_mutex, cpu);
523 acomp_ctx->dstmem = per_cpu(zswap_dstmem, cpu);
524
525 return 0;
526 }
527
zswap_cpu_comp_dead(unsigned int cpu,struct hlist_node * node)528 static int zswap_cpu_comp_dead(unsigned int cpu, struct hlist_node *node)
529 {
530 struct zswap_pool *pool = hlist_entry(node, struct zswap_pool, node);
531 struct crypto_acomp_ctx *acomp_ctx = per_cpu_ptr(pool->acomp_ctx, cpu);
532
533 if (!IS_ERR_OR_NULL(acomp_ctx)) {
534 if (!IS_ERR_OR_NULL(acomp_ctx->req))
535 acomp_request_free(acomp_ctx->req);
536 if (!IS_ERR_OR_NULL(acomp_ctx->acomp))
537 crypto_free_acomp(acomp_ctx->acomp);
538 }
539
540 return 0;
541 }
542
543 /*********************************
544 * pool functions
545 **********************************/
546
__zswap_pool_current(void)547 static struct zswap_pool *__zswap_pool_current(void)
548 {
549 struct zswap_pool *pool;
550
551 pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
552 WARN_ONCE(!pool && zswap_has_pool,
553 "%s: no page storage pool!\n", __func__);
554
555 return pool;
556 }
557
zswap_pool_current(void)558 static struct zswap_pool *zswap_pool_current(void)
559 {
560 assert_spin_locked(&zswap_pools_lock);
561
562 return __zswap_pool_current();
563 }
564
zswap_pool_current_get(void)565 static struct zswap_pool *zswap_pool_current_get(void)
566 {
567 struct zswap_pool *pool;
568
569 rcu_read_lock();
570
571 pool = __zswap_pool_current();
572 if (!zswap_pool_get(pool))
573 pool = NULL;
574
575 rcu_read_unlock();
576
577 return pool;
578 }
579
zswap_pool_last_get(void)580 static struct zswap_pool *zswap_pool_last_get(void)
581 {
582 struct zswap_pool *pool, *last = NULL;
583
584 rcu_read_lock();
585
586 list_for_each_entry_rcu(pool, &zswap_pools, list)
587 last = pool;
588 WARN_ONCE(!last && zswap_has_pool,
589 "%s: no page storage pool!\n", __func__);
590 if (!zswap_pool_get(last))
591 last = NULL;
592
593 rcu_read_unlock();
594
595 return last;
596 }
597
598 /* type and compressor must be null-terminated */
zswap_pool_find_get(char * type,char * compressor)599 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
600 {
601 struct zswap_pool *pool;
602
603 assert_spin_locked(&zswap_pools_lock);
604
605 list_for_each_entry_rcu(pool, &zswap_pools, list) {
606 if (strcmp(pool->tfm_name, compressor))
607 continue;
608 /* all zpools share the same type */
609 if (strcmp(zpool_get_type(pool->zpools[0]), type))
610 continue;
611 /* if we can't get it, it's about to be destroyed */
612 if (!zswap_pool_get(pool))
613 continue;
614 return pool;
615 }
616
617 return NULL;
618 }
619
620 /*
621 * If the entry is still valid in the tree, drop the initial ref and remove it
622 * from the tree. This function must be called with an additional ref held,
623 * otherwise it may race with another invalidation freeing the entry.
624 */
zswap_invalidate_entry(struct zswap_tree * tree,struct zswap_entry * entry)625 static void zswap_invalidate_entry(struct zswap_tree *tree,
626 struct zswap_entry *entry)
627 {
628 if (zswap_rb_erase(&tree->rbroot, entry))
629 zswap_entry_put(tree, entry);
630 }
631
zswap_reclaim_entry(struct zswap_pool * pool)632 static int zswap_reclaim_entry(struct zswap_pool *pool)
633 {
634 struct zswap_entry *entry;
635 struct zswap_tree *tree;
636 pgoff_t swpoffset;
637 int ret;
638
639 /* Get an entry off the LRU */
640 spin_lock(&pool->lru_lock);
641 if (list_empty(&pool->lru)) {
642 spin_unlock(&pool->lru_lock);
643 return -EINVAL;
644 }
645 entry = list_last_entry(&pool->lru, struct zswap_entry, lru);
646 list_del_init(&entry->lru);
647 /*
648 * Once the lru lock is dropped, the entry might get freed. The
649 * swpoffset is copied to the stack, and entry isn't deref'd again
650 * until the entry is verified to still be alive in the tree.
651 */
652 swpoffset = swp_offset(entry->swpentry);
653 tree = zswap_trees[swp_type(entry->swpentry)];
654 spin_unlock(&pool->lru_lock);
655
656 /* Check for invalidate() race */
657 spin_lock(&tree->lock);
658 if (entry != zswap_rb_search(&tree->rbroot, swpoffset)) {
659 ret = -EAGAIN;
660 goto unlock;
661 }
662 /* Hold a reference to prevent a free during writeback */
663 zswap_entry_get(entry);
664 spin_unlock(&tree->lock);
665
666 ret = zswap_writeback_entry(entry, tree);
667
668 spin_lock(&tree->lock);
669 if (ret) {
670 /* Writeback failed, put entry back on LRU */
671 spin_lock(&pool->lru_lock);
672 list_move(&entry->lru, &pool->lru);
673 spin_unlock(&pool->lru_lock);
674 goto put_unlock;
675 }
676
677 /*
678 * Writeback started successfully, the page now belongs to the
679 * swapcache. Drop the entry from zswap - unless invalidate already
680 * took it out while we had the tree->lock released for IO.
681 */
682 zswap_invalidate_entry(tree, entry);
683
684 put_unlock:
685 /* Drop local reference */
686 zswap_entry_put(tree, entry);
687 unlock:
688 spin_unlock(&tree->lock);
689 return ret ? -EAGAIN : 0;
690 }
691
shrink_worker(struct work_struct * w)692 static void shrink_worker(struct work_struct *w)
693 {
694 struct zswap_pool *pool = container_of(w, typeof(*pool),
695 shrink_work);
696 int ret, failures = 0;
697
698 do {
699 ret = zswap_reclaim_entry(pool);
700 if (ret) {
701 zswap_reject_reclaim_fail++;
702 if (ret != -EAGAIN)
703 break;
704 if (++failures == MAX_RECLAIM_RETRIES)
705 break;
706 }
707 cond_resched();
708 } while (!zswap_can_accept());
709 zswap_pool_put(pool);
710 }
711
zswap_pool_create(char * type,char * compressor)712 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
713 {
714 int i;
715 struct zswap_pool *pool;
716 char name[38]; /* 'zswap' + 32 char (max) num + \0 */
717 gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
718 int ret;
719
720 if (!zswap_has_pool) {
721 /* if either are unset, pool initialization failed, and we
722 * need both params to be set correctly before trying to
723 * create a pool.
724 */
725 if (!strcmp(type, ZSWAP_PARAM_UNSET))
726 return NULL;
727 if (!strcmp(compressor, ZSWAP_PARAM_UNSET))
728 return NULL;
729 }
730
731 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
732 if (!pool)
733 return NULL;
734
735 for (i = 0; i < ZSWAP_NR_ZPOOLS; i++) {
736 /* unique name for each pool specifically required by zsmalloc */
737 snprintf(name, 38, "zswap%x",
738 atomic_inc_return(&zswap_pools_count));
739
740 pool->zpools[i] = zpool_create_pool(type, name, gfp);
741 if (!pool->zpools[i]) {
742 pr_err("%s zpool not available\n", type);
743 goto error;
744 }
745 }
746 pr_debug("using %s zpool\n", zpool_get_type(pool->zpools[0]));
747
748 strscpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
749
750 pool->acomp_ctx = alloc_percpu(*pool->acomp_ctx);
751 if (!pool->acomp_ctx) {
752 pr_err("percpu alloc failed\n");
753 goto error;
754 }
755
756 ret = cpuhp_state_add_instance(CPUHP_MM_ZSWP_POOL_PREPARE,
757 &pool->node);
758 if (ret)
759 goto error;
760 pr_debug("using %s compressor\n", pool->tfm_name);
761
762 /* being the current pool takes 1 ref; this func expects the
763 * caller to always add the new pool as the current pool
764 */
765 kref_init(&pool->kref);
766 INIT_LIST_HEAD(&pool->list);
767 INIT_LIST_HEAD(&pool->lru);
768 spin_lock_init(&pool->lru_lock);
769 INIT_WORK(&pool->shrink_work, shrink_worker);
770
771 zswap_pool_debug("created", pool);
772
773 return pool;
774
775 error:
776 if (pool->acomp_ctx)
777 free_percpu(pool->acomp_ctx);
778 while (i--)
779 zpool_destroy_pool(pool->zpools[i]);
780 kfree(pool);
781 return NULL;
782 }
783
__zswap_pool_create_fallback(void)784 static struct zswap_pool *__zswap_pool_create_fallback(void)
785 {
786 bool has_comp, has_zpool;
787
788 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
789 if (!has_comp && strcmp(zswap_compressor,
790 CONFIG_ZSWAP_COMPRESSOR_DEFAULT)) {
791 pr_err("compressor %s not available, using default %s\n",
792 zswap_compressor, CONFIG_ZSWAP_COMPRESSOR_DEFAULT);
793 param_free_charp(&zswap_compressor);
794 zswap_compressor = CONFIG_ZSWAP_COMPRESSOR_DEFAULT;
795 has_comp = crypto_has_acomp(zswap_compressor, 0, 0);
796 }
797 if (!has_comp) {
798 pr_err("default compressor %s not available\n",
799 zswap_compressor);
800 param_free_charp(&zswap_compressor);
801 zswap_compressor = ZSWAP_PARAM_UNSET;
802 }
803
804 has_zpool = zpool_has_pool(zswap_zpool_type);
805 if (!has_zpool && strcmp(zswap_zpool_type,
806 CONFIG_ZSWAP_ZPOOL_DEFAULT)) {
807 pr_err("zpool %s not available, using default %s\n",
808 zswap_zpool_type, CONFIG_ZSWAP_ZPOOL_DEFAULT);
809 param_free_charp(&zswap_zpool_type);
810 zswap_zpool_type = CONFIG_ZSWAP_ZPOOL_DEFAULT;
811 has_zpool = zpool_has_pool(zswap_zpool_type);
812 }
813 if (!has_zpool) {
814 pr_err("default zpool %s not available\n",
815 zswap_zpool_type);
816 param_free_charp(&zswap_zpool_type);
817 zswap_zpool_type = ZSWAP_PARAM_UNSET;
818 }
819
820 if (!has_comp || !has_zpool)
821 return NULL;
822
823 return zswap_pool_create(zswap_zpool_type, zswap_compressor);
824 }
825
zswap_pool_destroy(struct zswap_pool * pool)826 static void zswap_pool_destroy(struct zswap_pool *pool)
827 {
828 int i;
829
830 zswap_pool_debug("destroying", pool);
831
832 cpuhp_state_remove_instance(CPUHP_MM_ZSWP_POOL_PREPARE, &pool->node);
833 free_percpu(pool->acomp_ctx);
834 for (i = 0; i < ZSWAP_NR_ZPOOLS; i++)
835 zpool_destroy_pool(pool->zpools[i]);
836 kfree(pool);
837 }
838
zswap_pool_get(struct zswap_pool * pool)839 static int __must_check zswap_pool_get(struct zswap_pool *pool)
840 {
841 if (!pool)
842 return 0;
843
844 return kref_get_unless_zero(&pool->kref);
845 }
846
__zswap_pool_release(struct work_struct * work)847 static void __zswap_pool_release(struct work_struct *work)
848 {
849 struct zswap_pool *pool = container_of(work, typeof(*pool),
850 release_work);
851
852 synchronize_rcu();
853
854 /* nobody should have been able to get a kref... */
855 WARN_ON(kref_get_unless_zero(&pool->kref));
856
857 /* pool is now off zswap_pools list and has no references. */
858 zswap_pool_destroy(pool);
859 }
860
__zswap_pool_empty(struct kref * kref)861 static void __zswap_pool_empty(struct kref *kref)
862 {
863 struct zswap_pool *pool;
864
865 pool = container_of(kref, typeof(*pool), kref);
866
867 spin_lock(&zswap_pools_lock);
868
869 WARN_ON(pool == zswap_pool_current());
870
871 list_del_rcu(&pool->list);
872
873 INIT_WORK(&pool->release_work, __zswap_pool_release);
874 schedule_work(&pool->release_work);
875
876 spin_unlock(&zswap_pools_lock);
877 }
878
zswap_pool_put(struct zswap_pool * pool)879 static void zswap_pool_put(struct zswap_pool *pool)
880 {
881 kref_put(&pool->kref, __zswap_pool_empty);
882 }
883
884 /*********************************
885 * param callbacks
886 **********************************/
887
zswap_pool_changed(const char * s,const struct kernel_param * kp)888 static bool zswap_pool_changed(const char *s, const struct kernel_param *kp)
889 {
890 /* no change required */
891 if (!strcmp(s, *(char **)kp->arg) && zswap_has_pool)
892 return false;
893 return true;
894 }
895
896 /* val must be a null-terminated string */
__zswap_param_set(const char * val,const struct kernel_param * kp,char * type,char * compressor)897 static int __zswap_param_set(const char *val, const struct kernel_param *kp,
898 char *type, char *compressor)
899 {
900 struct zswap_pool *pool, *put_pool = NULL;
901 char *s = strstrip((char *)val);
902 int ret = 0;
903 bool new_pool = false;
904
905 mutex_lock(&zswap_init_lock);
906 switch (zswap_init_state) {
907 case ZSWAP_UNINIT:
908 /* if this is load-time (pre-init) param setting,
909 * don't create a pool; that's done during init.
910 */
911 ret = param_set_charp(s, kp);
912 break;
913 case ZSWAP_INIT_SUCCEED:
914 new_pool = zswap_pool_changed(s, kp);
915 break;
916 case ZSWAP_INIT_FAILED:
917 pr_err("can't set param, initialization failed\n");
918 ret = -ENODEV;
919 }
920 mutex_unlock(&zswap_init_lock);
921
922 /* no need to create a new pool, return directly */
923 if (!new_pool)
924 return ret;
925
926 if (!type) {
927 if (!zpool_has_pool(s)) {
928 pr_err("zpool %s not available\n", s);
929 return -ENOENT;
930 }
931 type = s;
932 } else if (!compressor) {
933 if (!crypto_has_acomp(s, 0, 0)) {
934 pr_err("compressor %s not available\n", s);
935 return -ENOENT;
936 }
937 compressor = s;
938 } else {
939 WARN_ON(1);
940 return -EINVAL;
941 }
942
943 spin_lock(&zswap_pools_lock);
944
945 pool = zswap_pool_find_get(type, compressor);
946 if (pool) {
947 zswap_pool_debug("using existing", pool);
948 WARN_ON(pool == zswap_pool_current());
949 list_del_rcu(&pool->list);
950 }
951
952 spin_unlock(&zswap_pools_lock);
953
954 if (!pool)
955 pool = zswap_pool_create(type, compressor);
956
957 if (pool)
958 ret = param_set_charp(s, kp);
959 else
960 ret = -EINVAL;
961
962 spin_lock(&zswap_pools_lock);
963
964 if (!ret) {
965 put_pool = zswap_pool_current();
966 list_add_rcu(&pool->list, &zswap_pools);
967 zswap_has_pool = true;
968 } else if (pool) {
969 /* add the possibly pre-existing pool to the end of the pools
970 * list; if it's new (and empty) then it'll be removed and
971 * destroyed by the put after we drop the lock
972 */
973 list_add_tail_rcu(&pool->list, &zswap_pools);
974 put_pool = pool;
975 }
976
977 spin_unlock(&zswap_pools_lock);
978
979 if (!zswap_has_pool && !pool) {
980 /* if initial pool creation failed, and this pool creation also
981 * failed, maybe both compressor and zpool params were bad.
982 * Allow changing this param, so pool creation will succeed
983 * when the other param is changed. We already verified this
984 * param is ok in the zpool_has_pool() or crypto_has_acomp()
985 * checks above.
986 */
987 ret = param_set_charp(s, kp);
988 }
989
990 /* drop the ref from either the old current pool,
991 * or the new pool we failed to add
992 */
993 if (put_pool)
994 zswap_pool_put(put_pool);
995
996 return ret;
997 }
998
zswap_compressor_param_set(const char * val,const struct kernel_param * kp)999 static int zswap_compressor_param_set(const char *val,
1000 const struct kernel_param *kp)
1001 {
1002 return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
1003 }
1004
zswap_zpool_param_set(const char * val,const struct kernel_param * kp)1005 static int zswap_zpool_param_set(const char *val,
1006 const struct kernel_param *kp)
1007 {
1008 return __zswap_param_set(val, kp, NULL, zswap_compressor);
1009 }
1010
zswap_enabled_param_set(const char * val,const struct kernel_param * kp)1011 static int zswap_enabled_param_set(const char *val,
1012 const struct kernel_param *kp)
1013 {
1014 int ret = -ENODEV;
1015
1016 /* if this is load-time (pre-init) param setting, only set param. */
1017 if (system_state != SYSTEM_RUNNING)
1018 return param_set_bool(val, kp);
1019
1020 mutex_lock(&zswap_init_lock);
1021 switch (zswap_init_state) {
1022 case ZSWAP_UNINIT:
1023 if (zswap_setup())
1024 break;
1025 fallthrough;
1026 case ZSWAP_INIT_SUCCEED:
1027 if (!zswap_has_pool)
1028 pr_err("can't enable, no pool configured\n");
1029 else
1030 ret = param_set_bool(val, kp);
1031 break;
1032 case ZSWAP_INIT_FAILED:
1033 pr_err("can't enable, initialization failed\n");
1034 }
1035 mutex_unlock(&zswap_init_lock);
1036
1037 return ret;
1038 }
1039
1040 /*********************************
1041 * writeback code
1042 **********************************/
1043 /*
1044 * Attempts to free an entry by adding a page to the swap cache,
1045 * decompressing the entry data into the page, and issuing a
1046 * bio write to write the page back to the swap device.
1047 *
1048 * This can be thought of as a "resumed writeback" of the page
1049 * to the swap device. We are basically resuming the same swap
1050 * writeback path that was intercepted with the zswap_store()
1051 * in the first place. After the page has been decompressed into
1052 * the swap cache, the compressed version stored by zswap can be
1053 * freed.
1054 */
zswap_writeback_entry(struct zswap_entry * entry,struct zswap_tree * tree)1055 static int zswap_writeback_entry(struct zswap_entry *entry,
1056 struct zswap_tree *tree)
1057 {
1058 swp_entry_t swpentry = entry->swpentry;
1059 struct page *page;
1060 struct scatterlist input, output;
1061 struct crypto_acomp_ctx *acomp_ctx;
1062 struct zpool *pool = zswap_find_zpool(entry);
1063 bool page_was_allocated;
1064 u8 *src, *tmp = NULL;
1065 unsigned int dlen;
1066 int ret;
1067 struct writeback_control wbc = {
1068 .sync_mode = WB_SYNC_NONE,
1069 };
1070
1071 if (!zpool_can_sleep_mapped(pool)) {
1072 tmp = kmalloc(PAGE_SIZE, GFP_KERNEL);
1073 if (!tmp)
1074 return -ENOMEM;
1075 }
1076
1077 /* try to allocate swap cache page */
1078 page = __read_swap_cache_async(swpentry, GFP_KERNEL, NULL, 0,
1079 &page_was_allocated);
1080 if (!page) {
1081 ret = -ENOMEM;
1082 goto fail;
1083 }
1084
1085 /* Found an existing page, we raced with load/swapin */
1086 if (!page_was_allocated) {
1087 put_page(page);
1088 ret = -EEXIST;
1089 goto fail;
1090 }
1091
1092 /*
1093 * Page is locked, and the swapcache is now secured against
1094 * concurrent swapping to and from the slot. Verify that the
1095 * swap entry hasn't been invalidated and recycled behind our
1096 * backs (our zswap_entry reference doesn't prevent that), to
1097 * avoid overwriting a new swap page with old compressed data.
1098 */
1099 spin_lock(&tree->lock);
1100 if (zswap_rb_search(&tree->rbroot, swp_offset(entry->swpentry)) != entry) {
1101 spin_unlock(&tree->lock);
1102 delete_from_swap_cache(page_folio(page));
1103 unlock_page(page);
1104 put_page(page);
1105 ret = -ENOMEM;
1106 goto fail;
1107 }
1108 spin_unlock(&tree->lock);
1109
1110 /* decompress */
1111 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1112 dlen = PAGE_SIZE;
1113
1114 src = zpool_map_handle(pool, entry->handle, ZPOOL_MM_RO);
1115 if (!zpool_can_sleep_mapped(pool)) {
1116 memcpy(tmp, src, entry->length);
1117 src = tmp;
1118 zpool_unmap_handle(pool, entry->handle);
1119 }
1120
1121 mutex_lock(acomp_ctx->mutex);
1122 sg_init_one(&input, src, entry->length);
1123 sg_init_table(&output, 1);
1124 sg_set_page(&output, page, PAGE_SIZE, 0);
1125 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1126 ret = crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait);
1127 dlen = acomp_ctx->req->dlen;
1128 mutex_unlock(acomp_ctx->mutex);
1129
1130 if (!zpool_can_sleep_mapped(pool))
1131 kfree(tmp);
1132 else
1133 zpool_unmap_handle(pool, entry->handle);
1134
1135 BUG_ON(ret);
1136 BUG_ON(dlen != PAGE_SIZE);
1137
1138 /* page is up to date */
1139 SetPageUptodate(page);
1140
1141 /* move it to the tail of the inactive list after end_writeback */
1142 SetPageReclaim(page);
1143
1144 /* start writeback */
1145 __swap_writepage(page, &wbc);
1146 put_page(page);
1147 zswap_written_back_pages++;
1148
1149 return ret;
1150
1151 fail:
1152 if (!zpool_can_sleep_mapped(pool))
1153 kfree(tmp);
1154
1155 /*
1156 * If we get here because the page is already in swapcache, a
1157 * load may be happening concurrently. It is safe and okay to
1158 * not free the entry. It is also okay to return !0.
1159 */
1160 return ret;
1161 }
1162
zswap_is_page_same_filled(void * ptr,unsigned long * value)1163 static int zswap_is_page_same_filled(void *ptr, unsigned long *value)
1164 {
1165 unsigned long *page;
1166 unsigned long val;
1167 unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
1168
1169 page = (unsigned long *)ptr;
1170 val = page[0];
1171
1172 if (val != page[last_pos])
1173 return 0;
1174
1175 for (pos = 1; pos < last_pos; pos++) {
1176 if (val != page[pos])
1177 return 0;
1178 }
1179
1180 *value = val;
1181
1182 return 1;
1183 }
1184
zswap_fill_page(void * ptr,unsigned long value)1185 static void zswap_fill_page(void *ptr, unsigned long value)
1186 {
1187 unsigned long *page;
1188
1189 page = (unsigned long *)ptr;
1190 memset_l(page, value, PAGE_SIZE / sizeof(unsigned long));
1191 }
1192
zswap_store(struct folio * folio)1193 bool zswap_store(struct folio *folio)
1194 {
1195 swp_entry_t swp = folio->swap;
1196 int type = swp_type(swp);
1197 pgoff_t offset = swp_offset(swp);
1198 struct page *page = &folio->page;
1199 struct zswap_tree *tree = zswap_trees[type];
1200 struct zswap_entry *entry, *dupentry;
1201 struct scatterlist input, output;
1202 struct crypto_acomp_ctx *acomp_ctx;
1203 struct obj_cgroup *objcg = NULL;
1204 struct zswap_pool *pool;
1205 struct zpool *zpool;
1206 unsigned int dlen = PAGE_SIZE;
1207 unsigned long handle, value;
1208 char *buf;
1209 u8 *src, *dst;
1210 gfp_t gfp;
1211 int ret;
1212
1213 VM_WARN_ON_ONCE(!folio_test_locked(folio));
1214 VM_WARN_ON_ONCE(!folio_test_swapcache(folio));
1215
1216 /* Large folios aren't supported */
1217 if (folio_test_large(folio))
1218 return false;
1219
1220 if (!tree)
1221 return false;
1222
1223 /*
1224 * If this is a duplicate, it must be removed before attempting to store
1225 * it, otherwise, if the store fails the old page won't be removed from
1226 * the tree, and it might be written back overriding the new data.
1227 */
1228 spin_lock(&tree->lock);
1229 dupentry = zswap_rb_search(&tree->rbroot, offset);
1230 if (dupentry) {
1231 zswap_duplicate_entry++;
1232 zswap_invalidate_entry(tree, dupentry);
1233 }
1234 spin_unlock(&tree->lock);
1235
1236 if (!zswap_enabled)
1237 return false;
1238
1239 /*
1240 * XXX: zswap reclaim does not work with cgroups yet. Without a
1241 * cgroup-aware entry LRU, we will push out entries system-wide based on
1242 * local cgroup limits.
1243 */
1244 objcg = get_obj_cgroup_from_folio(folio);
1245 if (objcg && !obj_cgroup_may_zswap(objcg))
1246 goto reject;
1247
1248 /* reclaim space if needed */
1249 if (zswap_is_full()) {
1250 zswap_pool_limit_hit++;
1251 zswap_pool_reached_full = true;
1252 goto shrink;
1253 }
1254
1255 if (zswap_pool_reached_full) {
1256 if (!zswap_can_accept())
1257 goto shrink;
1258 else
1259 zswap_pool_reached_full = false;
1260 }
1261
1262 /* allocate entry */
1263 entry = zswap_entry_cache_alloc(GFP_KERNEL);
1264 if (!entry) {
1265 zswap_reject_kmemcache_fail++;
1266 goto reject;
1267 }
1268
1269 if (zswap_same_filled_pages_enabled) {
1270 src = kmap_atomic(page);
1271 if (zswap_is_page_same_filled(src, &value)) {
1272 kunmap_atomic(src);
1273 entry->swpentry = swp_entry(type, offset);
1274 entry->length = 0;
1275 entry->value = value;
1276 atomic_inc(&zswap_same_filled_pages);
1277 goto insert_entry;
1278 }
1279 kunmap_atomic(src);
1280 }
1281
1282 if (!zswap_non_same_filled_pages_enabled)
1283 goto freepage;
1284
1285 /* if entry is successfully added, it keeps the reference */
1286 entry->pool = zswap_pool_current_get();
1287 if (!entry->pool)
1288 goto freepage;
1289
1290 /* compress */
1291 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1292
1293 mutex_lock(acomp_ctx->mutex);
1294
1295 dst = acomp_ctx->dstmem;
1296 sg_init_table(&input, 1);
1297 sg_set_page(&input, page, PAGE_SIZE, 0);
1298
1299 /* zswap_dstmem is of size (PAGE_SIZE * 2). Reflect same in sg_list */
1300 sg_init_one(&output, dst, PAGE_SIZE * 2);
1301 acomp_request_set_params(acomp_ctx->req, &input, &output, PAGE_SIZE, dlen);
1302 /*
1303 * it maybe looks a little bit silly that we send an asynchronous request,
1304 * then wait for its completion synchronously. This makes the process look
1305 * synchronous in fact.
1306 * Theoretically, acomp supports users send multiple acomp requests in one
1307 * acomp instance, then get those requests done simultaneously. but in this
1308 * case, zswap actually does store and load page by page, there is no
1309 * existing method to send the second page before the first page is done
1310 * in one thread doing zwap.
1311 * but in different threads running on different cpu, we have different
1312 * acomp instance, so multiple threads can do (de)compression in parallel.
1313 */
1314 ret = crypto_wait_req(crypto_acomp_compress(acomp_ctx->req), &acomp_ctx->wait);
1315 dlen = acomp_ctx->req->dlen;
1316
1317 if (ret)
1318 goto put_dstmem;
1319
1320 /* store */
1321 zpool = zswap_find_zpool(entry);
1322 gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
1323 if (zpool_malloc_support_movable(zpool))
1324 gfp |= __GFP_HIGHMEM | __GFP_MOVABLE;
1325 ret = zpool_malloc(zpool, dlen, gfp, &handle);
1326 if (ret == -ENOSPC) {
1327 zswap_reject_compress_poor++;
1328 goto put_dstmem;
1329 }
1330 if (ret) {
1331 zswap_reject_alloc_fail++;
1332 goto put_dstmem;
1333 }
1334 buf = zpool_map_handle(zpool, handle, ZPOOL_MM_WO);
1335 memcpy(buf, dst, dlen);
1336 zpool_unmap_handle(zpool, handle);
1337 mutex_unlock(acomp_ctx->mutex);
1338
1339 /* populate entry */
1340 entry->swpentry = swp_entry(type, offset);
1341 entry->handle = handle;
1342 entry->length = dlen;
1343
1344 insert_entry:
1345 entry->objcg = objcg;
1346 if (objcg) {
1347 obj_cgroup_charge_zswap(objcg, entry->length);
1348 /* Account before objcg ref is moved to tree */
1349 count_objcg_event(objcg, ZSWPOUT);
1350 }
1351
1352 /* map */
1353 spin_lock(&tree->lock);
1354 /*
1355 * A duplicate entry should have been removed at the beginning of this
1356 * function. Since the swap entry should be pinned, if a duplicate is
1357 * found again here it means that something went wrong in the swap
1358 * cache.
1359 */
1360 while (zswap_rb_insert(&tree->rbroot, entry, &dupentry) == -EEXIST) {
1361 WARN_ON(1);
1362 zswap_duplicate_entry++;
1363 zswap_invalidate_entry(tree, dupentry);
1364 }
1365 if (entry->length) {
1366 spin_lock(&entry->pool->lru_lock);
1367 list_add(&entry->lru, &entry->pool->lru);
1368 spin_unlock(&entry->pool->lru_lock);
1369 }
1370 spin_unlock(&tree->lock);
1371
1372 /* update stats */
1373 atomic_inc(&zswap_stored_pages);
1374 zswap_update_total_size();
1375 count_vm_event(ZSWPOUT);
1376
1377 return true;
1378
1379 put_dstmem:
1380 mutex_unlock(acomp_ctx->mutex);
1381 zswap_pool_put(entry->pool);
1382 freepage:
1383 zswap_entry_cache_free(entry);
1384 reject:
1385 if (objcg)
1386 obj_cgroup_put(objcg);
1387 return false;
1388
1389 shrink:
1390 pool = zswap_pool_last_get();
1391 if (pool && !queue_work(shrink_wq, &pool->shrink_work))
1392 zswap_pool_put(pool);
1393 goto reject;
1394 }
1395
zswap_load(struct folio * folio)1396 bool zswap_load(struct folio *folio)
1397 {
1398 swp_entry_t swp = folio->swap;
1399 int type = swp_type(swp);
1400 pgoff_t offset = swp_offset(swp);
1401 struct page *page = &folio->page;
1402 struct zswap_tree *tree = zswap_trees[type];
1403 struct zswap_entry *entry;
1404 struct scatterlist input, output;
1405 struct crypto_acomp_ctx *acomp_ctx;
1406 u8 *src, *dst, *tmp;
1407 struct zpool *zpool;
1408 unsigned int dlen;
1409 bool ret;
1410
1411 VM_WARN_ON_ONCE(!folio_test_locked(folio));
1412
1413 /* find */
1414 spin_lock(&tree->lock);
1415 entry = zswap_entry_find_get(&tree->rbroot, offset);
1416 if (!entry) {
1417 spin_unlock(&tree->lock);
1418 return false;
1419 }
1420 spin_unlock(&tree->lock);
1421
1422 if (!entry->length) {
1423 dst = kmap_atomic(page);
1424 zswap_fill_page(dst, entry->value);
1425 kunmap_atomic(dst);
1426 ret = true;
1427 goto stats;
1428 }
1429
1430 zpool = zswap_find_zpool(entry);
1431 if (!zpool_can_sleep_mapped(zpool)) {
1432 tmp = kmalloc(entry->length, GFP_KERNEL);
1433 if (!tmp) {
1434 ret = false;
1435 goto freeentry;
1436 }
1437 }
1438
1439 /* decompress */
1440 dlen = PAGE_SIZE;
1441 src = zpool_map_handle(zpool, entry->handle, ZPOOL_MM_RO);
1442
1443 if (!zpool_can_sleep_mapped(zpool)) {
1444 memcpy(tmp, src, entry->length);
1445 src = tmp;
1446 zpool_unmap_handle(zpool, entry->handle);
1447 }
1448
1449 acomp_ctx = raw_cpu_ptr(entry->pool->acomp_ctx);
1450 mutex_lock(acomp_ctx->mutex);
1451 sg_init_one(&input, src, entry->length);
1452 sg_init_table(&output, 1);
1453 sg_set_page(&output, page, PAGE_SIZE, 0);
1454 acomp_request_set_params(acomp_ctx->req, &input, &output, entry->length, dlen);
1455 if (crypto_wait_req(crypto_acomp_decompress(acomp_ctx->req), &acomp_ctx->wait))
1456 WARN_ON(1);
1457 mutex_unlock(acomp_ctx->mutex);
1458
1459 if (zpool_can_sleep_mapped(zpool))
1460 zpool_unmap_handle(zpool, entry->handle);
1461 else
1462 kfree(tmp);
1463
1464 ret = true;
1465 stats:
1466 count_vm_event(ZSWPIN);
1467 if (entry->objcg)
1468 count_objcg_event(entry->objcg, ZSWPIN);
1469 freeentry:
1470 spin_lock(&tree->lock);
1471 if (ret && zswap_exclusive_loads_enabled) {
1472 zswap_invalidate_entry(tree, entry);
1473 folio_mark_dirty(folio);
1474 } else if (entry->length) {
1475 spin_lock(&entry->pool->lru_lock);
1476 list_move(&entry->lru, &entry->pool->lru);
1477 spin_unlock(&entry->pool->lru_lock);
1478 }
1479 zswap_entry_put(tree, entry);
1480 spin_unlock(&tree->lock);
1481
1482 return ret;
1483 }
1484
zswap_invalidate(int type,pgoff_t offset)1485 void zswap_invalidate(int type, pgoff_t offset)
1486 {
1487 struct zswap_tree *tree = zswap_trees[type];
1488 struct zswap_entry *entry;
1489
1490 /* find */
1491 spin_lock(&tree->lock);
1492 entry = zswap_rb_search(&tree->rbroot, offset);
1493 if (!entry) {
1494 /* entry was written back */
1495 spin_unlock(&tree->lock);
1496 return;
1497 }
1498 zswap_invalidate_entry(tree, entry);
1499 spin_unlock(&tree->lock);
1500 }
1501
zswap_swapon(int type)1502 void zswap_swapon(int type)
1503 {
1504 struct zswap_tree *tree;
1505
1506 tree = kzalloc(sizeof(*tree), GFP_KERNEL);
1507 if (!tree) {
1508 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1509 return;
1510 }
1511
1512 tree->rbroot = RB_ROOT;
1513 spin_lock_init(&tree->lock);
1514 zswap_trees[type] = tree;
1515 }
1516
zswap_swapoff(int type)1517 void zswap_swapoff(int type)
1518 {
1519 struct zswap_tree *tree = zswap_trees[type];
1520 struct zswap_entry *entry, *n;
1521
1522 if (!tree)
1523 return;
1524
1525 /* walk the tree and free everything */
1526 spin_lock(&tree->lock);
1527 rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1528 zswap_free_entry(entry);
1529 tree->rbroot = RB_ROOT;
1530 spin_unlock(&tree->lock);
1531 kfree(tree);
1532 zswap_trees[type] = NULL;
1533 }
1534
1535 /*********************************
1536 * debugfs functions
1537 **********************************/
1538 #ifdef CONFIG_DEBUG_FS
1539 #include <linux/debugfs.h>
1540
1541 static struct dentry *zswap_debugfs_root;
1542
zswap_debugfs_init(void)1543 static int zswap_debugfs_init(void)
1544 {
1545 if (!debugfs_initialized())
1546 return -ENODEV;
1547
1548 zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1549
1550 debugfs_create_u64("pool_limit_hit", 0444,
1551 zswap_debugfs_root, &zswap_pool_limit_hit);
1552 debugfs_create_u64("reject_reclaim_fail", 0444,
1553 zswap_debugfs_root, &zswap_reject_reclaim_fail);
1554 debugfs_create_u64("reject_alloc_fail", 0444,
1555 zswap_debugfs_root, &zswap_reject_alloc_fail);
1556 debugfs_create_u64("reject_kmemcache_fail", 0444,
1557 zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1558 debugfs_create_u64("reject_compress_poor", 0444,
1559 zswap_debugfs_root, &zswap_reject_compress_poor);
1560 debugfs_create_u64("written_back_pages", 0444,
1561 zswap_debugfs_root, &zswap_written_back_pages);
1562 debugfs_create_u64("duplicate_entry", 0444,
1563 zswap_debugfs_root, &zswap_duplicate_entry);
1564 debugfs_create_u64("pool_total_size", 0444,
1565 zswap_debugfs_root, &zswap_pool_total_size);
1566 debugfs_create_atomic_t("stored_pages", 0444,
1567 zswap_debugfs_root, &zswap_stored_pages);
1568 debugfs_create_atomic_t("same_filled_pages", 0444,
1569 zswap_debugfs_root, &zswap_same_filled_pages);
1570
1571 return 0;
1572 }
1573 #else
zswap_debugfs_init(void)1574 static int zswap_debugfs_init(void)
1575 {
1576 return 0;
1577 }
1578 #endif
1579
1580 /*********************************
1581 * module init and exit
1582 **********************************/
zswap_setup(void)1583 static int zswap_setup(void)
1584 {
1585 struct zswap_pool *pool;
1586 int ret;
1587
1588 zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
1589 if (!zswap_entry_cache) {
1590 pr_err("entry cache creation failed\n");
1591 goto cache_fail;
1592 }
1593
1594 ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare",
1595 zswap_dstmem_prepare, zswap_dstmem_dead);
1596 if (ret) {
1597 pr_err("dstmem alloc failed\n");
1598 goto dstmem_fail;
1599 }
1600
1601 ret = cpuhp_setup_state_multi(CPUHP_MM_ZSWP_POOL_PREPARE,
1602 "mm/zswap_pool:prepare",
1603 zswap_cpu_comp_prepare,
1604 zswap_cpu_comp_dead);
1605 if (ret)
1606 goto hp_fail;
1607
1608 pool = __zswap_pool_create_fallback();
1609 if (pool) {
1610 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1611 zpool_get_type(pool->zpools[0]));
1612 list_add(&pool->list, &zswap_pools);
1613 zswap_has_pool = true;
1614 } else {
1615 pr_err("pool creation failed\n");
1616 zswap_enabled = false;
1617 }
1618
1619 shrink_wq = create_workqueue("zswap-shrink");
1620 if (!shrink_wq)
1621 goto fallback_fail;
1622
1623 if (zswap_debugfs_init())
1624 pr_warn("debugfs initialization failed\n");
1625 zswap_init_state = ZSWAP_INIT_SUCCEED;
1626 return 0;
1627
1628 fallback_fail:
1629 if (pool)
1630 zswap_pool_destroy(pool);
1631 hp_fail:
1632 cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE);
1633 dstmem_fail:
1634 kmem_cache_destroy(zswap_entry_cache);
1635 cache_fail:
1636 /* if built-in, we aren't unloaded on failure; don't allow use */
1637 zswap_init_state = ZSWAP_INIT_FAILED;
1638 zswap_enabled = false;
1639 return -ENOMEM;
1640 }
1641
zswap_init(void)1642 static int __init zswap_init(void)
1643 {
1644 if (!zswap_enabled)
1645 return 0;
1646 return zswap_setup();
1647 }
1648 /* must be late so crypto has time to come up */
1649 late_initcall(zswap_init);
1650
1651 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1652 MODULE_DESCRIPTION("Compressed cache for swap pages");
1653