2 * zswap.c - zswap driver file
4 * zswap is a backend for frontswap that takes pages that are in the process
5 * of being swapped out and attempts to compress and store them in a
6 * RAM-based memory pool. This can result in a significant I/O reduction on
7 * the swap device and, in the case where decompressing from RAM is faster
8 * than reading from the swap device, can also improve workload performance.
10 * Copyright (C) 2012 Seth Jennings <sjenning@linux.vnet.ibm.com>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version 2
15 * of the License, or (at your option) any later version.
17 * This program is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU General Public License for more details.
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25 #include <linux/module.h>
26 #include <linux/cpu.h>
27 #include <linux/highmem.h>
28 #include <linux/slab.h>
29 #include <linux/spinlock.h>
30 #include <linux/types.h>
31 #include <linux/atomic.h>
32 #include <linux/frontswap.h>
33 #include <linux/rbtree.h>
34 #include <linux/swap.h>
35 #include <linux/crypto.h>
36 #include <linux/mempool.h>
37 #include <linux/zpool.h>
39 #include <linux/mm_types.h>
40 #include <linux/page-flags.h>
41 #include <linux/swapops.h>
42 #include <linux/writeback.h>
43 #include <linux/pagemap.h>
45 /*********************************
47 **********************************/
48 /* Total bytes used by the compressed storage */
49 static u64 zswap_pool_total_size;
50 /* The number of compressed pages currently stored in zswap */
51 static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
54 * The statistics below are not protected from concurrent access for
55 * performance reasons so they may not be a 100% accurate. However,
56 * they do provide useful information on roughly how many times a
57 * certain event is occurring.
60 /* Pool limit was hit (see zswap_max_pool_percent) */
61 static u64 zswap_pool_limit_hit;
62 /* Pages written back when pool limit was reached */
63 static u64 zswap_written_back_pages;
64 /* Store failed due to a reclaim failure after pool limit was reached */
65 static u64 zswap_reject_reclaim_fail;
66 /* Compressed page was too big for the allocator to (optimally) store */
67 static u64 zswap_reject_compress_poor;
68 /* Store failed because underlying allocator could not get memory */
69 static u64 zswap_reject_alloc_fail;
70 /* Store failed because the entry metadata could not be allocated (rare) */
71 static u64 zswap_reject_kmemcache_fail;
72 /* Duplicate store was encountered (rare) */
73 static u64 zswap_duplicate_entry;
75 /*********************************
77 **********************************/
79 /* Enable/disable zswap (disabled by default) */
80 static bool zswap_enabled;
81 static int zswap_enabled_param_set(const char *,
82 const struct kernel_param *);
83 static struct kernel_param_ops zswap_enabled_param_ops = {
84 .set = zswap_enabled_param_set,
85 .get = param_get_bool,
87 module_param_cb(enabled, &zswap_enabled_param_ops, &zswap_enabled, 0644);
89 /* Crypto compressor to use */
90 #define ZSWAP_COMPRESSOR_DEFAULT "lzo"
91 static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
92 static int zswap_compressor_param_set(const char *,
93 const struct kernel_param *);
94 static struct kernel_param_ops zswap_compressor_param_ops = {
95 .set = zswap_compressor_param_set,
96 .get = param_get_charp,
97 .free = param_free_charp,
99 module_param_cb(compressor, &zswap_compressor_param_ops,
100 &zswap_compressor, 0644);
102 /* Compressed storage zpool to use */
103 #define ZSWAP_ZPOOL_DEFAULT "zbud"
104 static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
105 static int zswap_zpool_param_set(const char *, const struct kernel_param *);
106 static struct kernel_param_ops zswap_zpool_param_ops = {
107 .set = zswap_zpool_param_set,
108 .get = param_get_charp,
109 .free = param_free_charp,
111 module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644);
113 /* The maximum percentage of memory that the compressed pool can occupy */
114 static unsigned int zswap_max_pool_percent = 20;
115 module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644);
117 /*********************************
119 **********************************/
123 struct crypto_comp * __percpu *tfm;
125 struct list_head list;
126 struct rcu_head rcu_head;
127 struct notifier_block notifier;
128 char tfm_name[CRYPTO_MAX_ALG_NAME];
134 * This structure contains the metadata for tracking a single compressed
137 * rbnode - links the entry into red-black tree for the appropriate swap type
138 * offset - the swap offset for the entry. Index into the red-black tree.
139 * refcount - the number of outstanding reference to the entry. This is needed
140 * to protect against premature freeing of the entry by code
141 * concurrent calls to load, invalidate, and writeback. The lock
142 * for the zswap_tree structure that contains the entry must
143 * be held while changing the refcount. Since the lock must
144 * be held, there is no reason to also make refcount atomic.
145 * length - the length in bytes of the compressed page data. Needed during
147 * pool - the zswap_pool the entry's data is in
148 * handle - zpool allocation handle that stores the compressed page data
151 struct rb_node rbnode;
155 struct zswap_pool *pool;
156 unsigned long handle;
159 struct zswap_header {
160 swp_entry_t swpentry;
164 * The tree lock in the zswap_tree struct protects a few things:
166 * - the refcount field of each entry in the tree
169 struct rb_root rbroot;
173 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
175 /* RCU-protected iteration */
176 static LIST_HEAD(zswap_pools);
177 /* protects zswap_pools list modification */
178 static DEFINE_SPINLOCK(zswap_pools_lock);
179 /* pool counter to provide unique names to zpool */
180 static atomic_t zswap_pools_count = ATOMIC_INIT(0);
182 /* used by param callback function */
183 static bool zswap_init_started;
185 /* fatal error during init */
186 static bool zswap_init_failed;
188 /*********************************
189 * helpers and fwd declarations
190 **********************************/
192 #define zswap_pool_debug(msg, p) \
193 pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name, \
194 zpool_get_type((p)->zpool))
196 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle);
197 static int zswap_pool_get(struct zswap_pool *pool);
198 static void zswap_pool_put(struct zswap_pool *pool);
200 static const struct zpool_ops zswap_zpool_ops = {
201 .evict = zswap_writeback_entry
204 static bool zswap_is_full(void)
206 return totalram_pages * zswap_max_pool_percent / 100 <
207 DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
210 static void zswap_update_total_size(void)
212 struct zswap_pool *pool;
217 list_for_each_entry_rcu(pool, &zswap_pools, list)
218 total += zpool_get_total_size(pool->zpool);
222 zswap_pool_total_size = total;
225 /*********************************
226 * zswap entry functions
227 **********************************/
228 static struct kmem_cache *zswap_entry_cache;
230 static int __init zswap_entry_cache_create(void)
232 zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
233 return zswap_entry_cache == NULL;
236 static void __init zswap_entry_cache_destroy(void)
238 kmem_cache_destroy(zswap_entry_cache);
241 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
243 struct zswap_entry *entry;
244 entry = kmem_cache_alloc(zswap_entry_cache, gfp);
248 RB_CLEAR_NODE(&entry->rbnode);
252 static void zswap_entry_cache_free(struct zswap_entry *entry)
254 kmem_cache_free(zswap_entry_cache, entry);
257 /*********************************
259 **********************************/
260 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
262 struct rb_node *node = root->rb_node;
263 struct zswap_entry *entry;
266 entry = rb_entry(node, struct zswap_entry, rbnode);
267 if (entry->offset > offset)
268 node = node->rb_left;
269 else if (entry->offset < offset)
270 node = node->rb_right;
278 * In the case that a entry with the same offset is found, a pointer to
279 * the existing entry is stored in dupentry and the function returns -EEXIST
281 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
282 struct zswap_entry **dupentry)
284 struct rb_node **link = &root->rb_node, *parent = NULL;
285 struct zswap_entry *myentry;
289 myentry = rb_entry(parent, struct zswap_entry, rbnode);
290 if (myentry->offset > entry->offset)
291 link = &(*link)->rb_left;
292 else if (myentry->offset < entry->offset)
293 link = &(*link)->rb_right;
299 rb_link_node(&entry->rbnode, parent, link);
300 rb_insert_color(&entry->rbnode, root);
304 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
306 if (!RB_EMPTY_NODE(&entry->rbnode)) {
307 rb_erase(&entry->rbnode, root);
308 RB_CLEAR_NODE(&entry->rbnode);
313 * Carries out the common pattern of freeing and entry's zpool allocation,
314 * freeing the entry itself, and decrementing the number of stored pages.
316 static void zswap_free_entry(struct zswap_entry *entry)
318 zpool_free(entry->pool->zpool, entry->handle);
319 zswap_pool_put(entry->pool);
320 zswap_entry_cache_free(entry);
321 atomic_dec(&zswap_stored_pages);
322 zswap_update_total_size();
325 /* caller must hold the tree lock */
326 static void zswap_entry_get(struct zswap_entry *entry)
331 /* caller must hold the tree lock
332 * remove from the tree and free it, if nobody reference the entry
334 static void zswap_entry_put(struct zswap_tree *tree,
335 struct zswap_entry *entry)
337 int refcount = --entry->refcount;
339 BUG_ON(refcount < 0);
341 zswap_rb_erase(&tree->rbroot, entry);
342 zswap_free_entry(entry);
346 /* caller must hold the tree lock */
347 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
350 struct zswap_entry *entry;
352 entry = zswap_rb_search(root, offset);
354 zswap_entry_get(entry);
359 /*********************************
361 **********************************/
362 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
364 static int __zswap_cpu_dstmem_notifier(unsigned long action, unsigned long cpu)
370 dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
372 pr_err("can't allocate compressor buffer\n");
375 per_cpu(zswap_dstmem, cpu) = dst;
378 case CPU_UP_CANCELED:
379 dst = per_cpu(zswap_dstmem, cpu);
381 per_cpu(zswap_dstmem, cpu) = NULL;
389 static int zswap_cpu_dstmem_notifier(struct notifier_block *nb,
390 unsigned long action, void *pcpu)
392 return __zswap_cpu_dstmem_notifier(action, (unsigned long)pcpu);
395 static struct notifier_block zswap_dstmem_notifier = {
396 .notifier_call = zswap_cpu_dstmem_notifier,
399 static int __init zswap_cpu_dstmem_init(void)
403 cpu_notifier_register_begin();
404 for_each_online_cpu(cpu)
405 if (__zswap_cpu_dstmem_notifier(CPU_UP_PREPARE, cpu) ==
408 __register_cpu_notifier(&zswap_dstmem_notifier);
409 cpu_notifier_register_done();
413 for_each_online_cpu(cpu)
414 __zswap_cpu_dstmem_notifier(CPU_UP_CANCELED, cpu);
415 cpu_notifier_register_done();
419 static void zswap_cpu_dstmem_destroy(void)
423 cpu_notifier_register_begin();
424 for_each_online_cpu(cpu)
425 __zswap_cpu_dstmem_notifier(CPU_UP_CANCELED, cpu);
426 __unregister_cpu_notifier(&zswap_dstmem_notifier);
427 cpu_notifier_register_done();
430 static int __zswap_cpu_comp_notifier(struct zswap_pool *pool,
431 unsigned long action, unsigned long cpu)
433 struct crypto_comp *tfm;
437 if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu)))
439 tfm = crypto_alloc_comp(pool->tfm_name, 0, 0);
440 if (IS_ERR_OR_NULL(tfm)) {
441 pr_err("could not alloc crypto comp %s : %ld\n",
442 pool->tfm_name, PTR_ERR(tfm));
445 *per_cpu_ptr(pool->tfm, cpu) = tfm;
448 case CPU_UP_CANCELED:
449 tfm = *per_cpu_ptr(pool->tfm, cpu);
450 if (!IS_ERR_OR_NULL(tfm))
451 crypto_free_comp(tfm);
452 *per_cpu_ptr(pool->tfm, cpu) = NULL;
460 static int zswap_cpu_comp_notifier(struct notifier_block *nb,
461 unsigned long action, void *pcpu)
463 unsigned long cpu = (unsigned long)pcpu;
464 struct zswap_pool *pool = container_of(nb, typeof(*pool), notifier);
466 return __zswap_cpu_comp_notifier(pool, action, cpu);
469 static int zswap_cpu_comp_init(struct zswap_pool *pool)
473 memset(&pool->notifier, 0, sizeof(pool->notifier));
474 pool->notifier.notifier_call = zswap_cpu_comp_notifier;
476 cpu_notifier_register_begin();
477 for_each_online_cpu(cpu)
478 if (__zswap_cpu_comp_notifier(pool, CPU_UP_PREPARE, cpu) ==
481 __register_cpu_notifier(&pool->notifier);
482 cpu_notifier_register_done();
486 for_each_online_cpu(cpu)
487 __zswap_cpu_comp_notifier(pool, CPU_UP_CANCELED, cpu);
488 cpu_notifier_register_done();
492 static void zswap_cpu_comp_destroy(struct zswap_pool *pool)
496 cpu_notifier_register_begin();
497 for_each_online_cpu(cpu)
498 __zswap_cpu_comp_notifier(pool, CPU_UP_CANCELED, cpu);
499 __unregister_cpu_notifier(&pool->notifier);
500 cpu_notifier_register_done();
503 /*********************************
505 **********************************/
507 static struct zswap_pool *__zswap_pool_current(void)
509 struct zswap_pool *pool;
511 pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list);
517 static struct zswap_pool *zswap_pool_current(void)
519 assert_spin_locked(&zswap_pools_lock);
521 return __zswap_pool_current();
524 static struct zswap_pool *zswap_pool_current_get(void)
526 struct zswap_pool *pool;
530 pool = __zswap_pool_current();
531 if (!pool || !zswap_pool_get(pool))
539 static struct zswap_pool *zswap_pool_last_get(void)
541 struct zswap_pool *pool, *last = NULL;
545 list_for_each_entry_rcu(pool, &zswap_pools, list)
547 if (!WARN_ON(!last) && !zswap_pool_get(last))
555 /* type and compressor must be null-terminated */
556 static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor)
558 struct zswap_pool *pool;
560 assert_spin_locked(&zswap_pools_lock);
562 list_for_each_entry_rcu(pool, &zswap_pools, list) {
563 if (strcmp(pool->tfm_name, compressor))
565 if (strcmp(zpool_get_type(pool->zpool), type))
567 /* if we can't get it, it's about to be destroyed */
568 if (!zswap_pool_get(pool))
576 static struct zswap_pool *zswap_pool_create(char *type, char *compressor)
578 struct zswap_pool *pool;
579 char name[38]; /* 'zswap' + 32 char (max) num + \0 */
580 gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM;
582 pool = kzalloc(sizeof(*pool), GFP_KERNEL);
584 pr_err("pool alloc failed\n");
588 /* unique name for each pool specifically required by zsmalloc */
589 snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count));
591 pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops);
593 pr_err("%s zpool not available\n", type);
596 pr_debug("using %s zpool\n", zpool_get_type(pool->zpool));
598 strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name));
599 pool->tfm = alloc_percpu(struct crypto_comp *);
601 pr_err("percpu alloc failed\n");
605 if (zswap_cpu_comp_init(pool))
607 pr_debug("using %s compressor\n", pool->tfm_name);
609 /* being the current pool takes 1 ref; this func expects the
610 * caller to always add the new pool as the current pool
612 kref_init(&pool->kref);
613 INIT_LIST_HEAD(&pool->list);
615 zswap_pool_debug("created", pool);
620 free_percpu(pool->tfm);
622 zpool_destroy_pool(pool->zpool);
627 static __init struct zswap_pool *__zswap_pool_create_fallback(void)
629 if (!crypto_has_comp(zswap_compressor, 0, 0)) {
630 if (!strcmp(zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT)) {
631 pr_err("default compressor %s not available\n",
635 pr_err("compressor %s not available, using default %s\n",
636 zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT);
637 param_free_charp(&zswap_compressor);
638 zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
640 if (!zpool_has_pool(zswap_zpool_type)) {
641 if (!strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) {
642 pr_err("default zpool %s not available\n",
646 pr_err("zpool %s not available, using default %s\n",
647 zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT);
648 param_free_charp(&zswap_zpool_type);
649 zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
652 return zswap_pool_create(zswap_zpool_type, zswap_compressor);
655 static void zswap_pool_destroy(struct zswap_pool *pool)
657 zswap_pool_debug("destroying", pool);
659 zswap_cpu_comp_destroy(pool);
660 free_percpu(pool->tfm);
661 zpool_destroy_pool(pool->zpool);
665 static int __must_check zswap_pool_get(struct zswap_pool *pool)
667 return kref_get_unless_zero(&pool->kref);
670 static void __zswap_pool_release(struct rcu_head *head)
672 struct zswap_pool *pool = container_of(head, typeof(*pool), rcu_head);
674 /* nobody should have been able to get a kref... */
675 WARN_ON(kref_get_unless_zero(&pool->kref));
677 /* pool is now off zswap_pools list and has no references. */
678 zswap_pool_destroy(pool);
681 static void __zswap_pool_empty(struct kref *kref)
683 struct zswap_pool *pool;
685 pool = container_of(kref, typeof(*pool), kref);
687 spin_lock(&zswap_pools_lock);
689 WARN_ON(pool == zswap_pool_current());
691 list_del_rcu(&pool->list);
692 call_rcu(&pool->rcu_head, __zswap_pool_release);
694 spin_unlock(&zswap_pools_lock);
697 static void zswap_pool_put(struct zswap_pool *pool)
699 kref_put(&pool->kref, __zswap_pool_empty);
702 /*********************************
704 **********************************/
706 /* val must be a null-terminated string */
707 static int __zswap_param_set(const char *val, const struct kernel_param *kp,
708 char *type, char *compressor)
710 struct zswap_pool *pool, *put_pool = NULL;
711 char *s = strstrip((char *)val);
714 if (zswap_init_failed) {
715 pr_err("can't set param, initialization failed\n");
719 /* no change required */
720 if (!strcmp(s, *(char **)kp->arg))
723 /* if this is load-time (pre-init) param setting,
724 * don't create a pool; that's done during init.
726 if (!zswap_init_started)
727 return param_set_charp(s, kp);
730 if (!zpool_has_pool(s)) {
731 pr_err("zpool %s not available\n", s);
735 } else if (!compressor) {
736 if (!crypto_has_comp(s, 0, 0)) {
737 pr_err("compressor %s not available\n", s);
746 spin_lock(&zswap_pools_lock);
748 pool = zswap_pool_find_get(type, compressor);
750 zswap_pool_debug("using existing", pool);
751 list_del_rcu(&pool->list);
753 spin_unlock(&zswap_pools_lock);
754 pool = zswap_pool_create(type, compressor);
755 spin_lock(&zswap_pools_lock);
759 ret = param_set_charp(s, kp);
764 put_pool = zswap_pool_current();
765 list_add_rcu(&pool->list, &zswap_pools);
767 /* add the possibly pre-existing pool to the end of the pools
768 * list; if it's new (and empty) then it'll be removed and
769 * destroyed by the put after we drop the lock
771 list_add_tail_rcu(&pool->list, &zswap_pools);
775 spin_unlock(&zswap_pools_lock);
777 /* drop the ref from either the old current pool,
778 * or the new pool we failed to add
781 zswap_pool_put(put_pool);
786 static int zswap_compressor_param_set(const char *val,
787 const struct kernel_param *kp)
789 return __zswap_param_set(val, kp, zswap_zpool_type, NULL);
792 static int zswap_zpool_param_set(const char *val,
793 const struct kernel_param *kp)
795 return __zswap_param_set(val, kp, NULL, zswap_compressor);
798 static int zswap_enabled_param_set(const char *val,
799 const struct kernel_param *kp)
801 if (zswap_init_failed) {
802 pr_err("can't enable, initialization failed\n");
806 return param_set_bool(val, kp);
809 /*********************************
811 **********************************/
812 /* return enum for zswap_get_swap_cache_page */
813 enum zswap_get_swap_ret {
815 ZSWAP_SWAPCACHE_EXIST,
816 ZSWAP_SWAPCACHE_FAIL,
820 * zswap_get_swap_cache_page
822 * This is an adaption of read_swap_cache_async()
824 * This function tries to find a page with the given swap entry
825 * in the swapper_space address space (the swap cache). If the page
826 * is found, it is returned in retpage. Otherwise, a page is allocated,
827 * added to the swap cache, and returned in retpage.
829 * If success, the swap cache page is returned in retpage
830 * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
831 * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
832 * the new page is added to swapcache and locked
833 * Returns ZSWAP_SWAPCACHE_FAIL on error
835 static int zswap_get_swap_cache_page(swp_entry_t entry,
836 struct page **retpage)
838 bool page_was_allocated;
840 *retpage = __read_swap_cache_async(entry, GFP_KERNEL,
841 NULL, 0, &page_was_allocated);
842 if (page_was_allocated)
843 return ZSWAP_SWAPCACHE_NEW;
845 return ZSWAP_SWAPCACHE_FAIL;
846 return ZSWAP_SWAPCACHE_EXIST;
850 * Attempts to free an entry by adding a page to the swap cache,
851 * decompressing the entry data into the page, and issuing a
852 * bio write to write the page back to the swap device.
854 * This can be thought of as a "resumed writeback" of the page
855 * to the swap device. We are basically resuming the same swap
856 * writeback path that was intercepted with the frontswap_store()
857 * in the first place. After the page has been decompressed into
858 * the swap cache, the compressed version stored by zswap can be
861 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
863 struct zswap_header *zhdr;
864 swp_entry_t swpentry;
865 struct zswap_tree *tree;
867 struct zswap_entry *entry;
869 struct crypto_comp *tfm;
873 struct writeback_control wbc = {
874 .sync_mode = WB_SYNC_NONE,
877 /* extract swpentry from data */
878 zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
879 swpentry = zhdr->swpentry; /* here */
880 zpool_unmap_handle(pool, handle);
881 tree = zswap_trees[swp_type(swpentry)];
882 offset = swp_offset(swpentry);
884 /* find and ref zswap entry */
885 spin_lock(&tree->lock);
886 entry = zswap_entry_find_get(&tree->rbroot, offset);
888 /* entry was invalidated */
889 spin_unlock(&tree->lock);
892 spin_unlock(&tree->lock);
893 BUG_ON(offset != entry->offset);
895 /* try to allocate swap cache page */
896 switch (zswap_get_swap_cache_page(swpentry, &page)) {
897 case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
901 case ZSWAP_SWAPCACHE_EXIST:
902 /* page is already in the swap cache, ignore for now */
903 page_cache_release(page);
907 case ZSWAP_SWAPCACHE_NEW: /* page is locked */
910 src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle,
911 ZPOOL_MM_RO) + sizeof(struct zswap_header);
912 dst = kmap_atomic(page);
913 tfm = *get_cpu_ptr(entry->pool->tfm);
914 ret = crypto_comp_decompress(tfm, src, entry->length,
916 put_cpu_ptr(entry->pool->tfm);
918 zpool_unmap_handle(entry->pool->zpool, entry->handle);
920 BUG_ON(dlen != PAGE_SIZE);
922 /* page is up to date */
923 SetPageUptodate(page);
926 /* move it to the tail of the inactive list after end_writeback */
927 SetPageReclaim(page);
929 /* start writeback */
930 __swap_writepage(page, &wbc, end_swap_bio_write);
931 page_cache_release(page);
932 zswap_written_back_pages++;
934 spin_lock(&tree->lock);
935 /* drop local reference */
936 zswap_entry_put(tree, entry);
939 * There are two possible situations for entry here:
940 * (1) refcount is 1(normal case), entry is valid and on the tree
941 * (2) refcount is 0, entry is freed and not on the tree
942 * because invalidate happened during writeback
943 * search the tree and free the entry if find entry
945 if (entry == zswap_rb_search(&tree->rbroot, offset))
946 zswap_entry_put(tree, entry);
947 spin_unlock(&tree->lock);
952 * if we get here due to ZSWAP_SWAPCACHE_EXIST
953 * a load may happening concurrently
954 * it is safe and okay to not free the entry
955 * if we free the entry in the following put
956 * it it either okay to return !0
959 spin_lock(&tree->lock);
960 zswap_entry_put(tree, entry);
961 spin_unlock(&tree->lock);
967 static int zswap_shrink(void)
969 struct zswap_pool *pool;
972 pool = zswap_pool_last_get();
976 ret = zpool_shrink(pool->zpool, 1, NULL);
978 zswap_pool_put(pool);
983 /*********************************
985 **********************************/
986 /* attempts to compress and store an single page */
987 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
990 struct zswap_tree *tree = zswap_trees[type];
991 struct zswap_entry *entry, *dupentry;
992 struct crypto_comp *tfm;
994 unsigned int dlen = PAGE_SIZE, len;
995 unsigned long handle;
998 struct zswap_header *zhdr;
1000 if (!zswap_enabled || !tree) {
1005 /* reclaim space if needed */
1006 if (zswap_is_full()) {
1007 zswap_pool_limit_hit++;
1008 if (zswap_shrink()) {
1009 zswap_reject_reclaim_fail++;
1015 /* allocate entry */
1016 entry = zswap_entry_cache_alloc(GFP_KERNEL);
1018 zswap_reject_kmemcache_fail++;
1023 /* if entry is successfully added, it keeps the reference */
1024 entry->pool = zswap_pool_current_get();
1031 dst = get_cpu_var(zswap_dstmem);
1032 tfm = *get_cpu_ptr(entry->pool->tfm);
1033 src = kmap_atomic(page);
1034 ret = crypto_comp_compress(tfm, src, PAGE_SIZE, dst, &dlen);
1036 put_cpu_ptr(entry->pool->tfm);
1043 len = dlen + sizeof(struct zswap_header);
1044 ret = zpool_malloc(entry->pool->zpool, len,
1045 __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM,
1047 if (ret == -ENOSPC) {
1048 zswap_reject_compress_poor++;
1052 zswap_reject_alloc_fail++;
1055 zhdr = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW);
1056 zhdr->swpentry = swp_entry(type, offset);
1057 buf = (u8 *)(zhdr + 1);
1058 memcpy(buf, dst, dlen);
1059 zpool_unmap_handle(entry->pool->zpool, handle);
1060 put_cpu_var(zswap_dstmem);
1062 /* populate entry */
1063 entry->offset = offset;
1064 entry->handle = handle;
1065 entry->length = dlen;
1068 spin_lock(&tree->lock);
1070 ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
1071 if (ret == -EEXIST) {
1072 zswap_duplicate_entry++;
1073 /* remove from rbtree */
1074 zswap_rb_erase(&tree->rbroot, dupentry);
1075 zswap_entry_put(tree, dupentry);
1077 } while (ret == -EEXIST);
1078 spin_unlock(&tree->lock);
1081 atomic_inc(&zswap_stored_pages);
1082 zswap_update_total_size();
1087 put_cpu_var(zswap_dstmem);
1088 zswap_pool_put(entry->pool);
1090 zswap_entry_cache_free(entry);
1096 * returns 0 if the page was successfully decompressed
1097 * return -1 on entry not found or error
1099 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
1102 struct zswap_tree *tree = zswap_trees[type];
1103 struct zswap_entry *entry;
1104 struct crypto_comp *tfm;
1110 spin_lock(&tree->lock);
1111 entry = zswap_entry_find_get(&tree->rbroot, offset);
1113 /* entry was written back */
1114 spin_unlock(&tree->lock);
1117 spin_unlock(&tree->lock);
1121 src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle,
1122 ZPOOL_MM_RO) + sizeof(struct zswap_header);
1123 dst = kmap_atomic(page);
1124 tfm = *get_cpu_ptr(entry->pool->tfm);
1125 ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen);
1126 put_cpu_ptr(entry->pool->tfm);
1128 zpool_unmap_handle(entry->pool->zpool, entry->handle);
1131 spin_lock(&tree->lock);
1132 zswap_entry_put(tree, entry);
1133 spin_unlock(&tree->lock);
1138 /* frees an entry in zswap */
1139 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
1141 struct zswap_tree *tree = zswap_trees[type];
1142 struct zswap_entry *entry;
1145 spin_lock(&tree->lock);
1146 entry = zswap_rb_search(&tree->rbroot, offset);
1148 /* entry was written back */
1149 spin_unlock(&tree->lock);
1153 /* remove from rbtree */
1154 zswap_rb_erase(&tree->rbroot, entry);
1156 /* drop the initial reference from entry creation */
1157 zswap_entry_put(tree, entry);
1159 spin_unlock(&tree->lock);
1162 /* frees all zswap entries for the given swap type */
1163 static void zswap_frontswap_invalidate_area(unsigned type)
1165 struct zswap_tree *tree = zswap_trees[type];
1166 struct zswap_entry *entry, *n;
1171 /* walk the tree and free everything */
1172 spin_lock(&tree->lock);
1173 rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
1174 zswap_free_entry(entry);
1175 tree->rbroot = RB_ROOT;
1176 spin_unlock(&tree->lock);
1178 zswap_trees[type] = NULL;
1181 static void zswap_frontswap_init(unsigned type)
1183 struct zswap_tree *tree;
1185 tree = kzalloc(sizeof(struct zswap_tree), GFP_KERNEL);
1187 pr_err("alloc failed, zswap disabled for swap type %d\n", type);
1191 tree->rbroot = RB_ROOT;
1192 spin_lock_init(&tree->lock);
1193 zswap_trees[type] = tree;
1196 static struct frontswap_ops zswap_frontswap_ops = {
1197 .store = zswap_frontswap_store,
1198 .load = zswap_frontswap_load,
1199 .invalidate_page = zswap_frontswap_invalidate_page,
1200 .invalidate_area = zswap_frontswap_invalidate_area,
1201 .init = zswap_frontswap_init
1204 /*********************************
1206 **********************************/
1207 #ifdef CONFIG_DEBUG_FS
1208 #include <linux/debugfs.h>
1210 static struct dentry *zswap_debugfs_root;
1212 static int __init zswap_debugfs_init(void)
1214 if (!debugfs_initialized())
1217 zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
1218 if (!zswap_debugfs_root)
1221 debugfs_create_u64("pool_limit_hit", S_IRUGO,
1222 zswap_debugfs_root, &zswap_pool_limit_hit);
1223 debugfs_create_u64("reject_reclaim_fail", S_IRUGO,
1224 zswap_debugfs_root, &zswap_reject_reclaim_fail);
1225 debugfs_create_u64("reject_alloc_fail", S_IRUGO,
1226 zswap_debugfs_root, &zswap_reject_alloc_fail);
1227 debugfs_create_u64("reject_kmemcache_fail", S_IRUGO,
1228 zswap_debugfs_root, &zswap_reject_kmemcache_fail);
1229 debugfs_create_u64("reject_compress_poor", S_IRUGO,
1230 zswap_debugfs_root, &zswap_reject_compress_poor);
1231 debugfs_create_u64("written_back_pages", S_IRUGO,
1232 zswap_debugfs_root, &zswap_written_back_pages);
1233 debugfs_create_u64("duplicate_entry", S_IRUGO,
1234 zswap_debugfs_root, &zswap_duplicate_entry);
1235 debugfs_create_u64("pool_total_size", S_IRUGO,
1236 zswap_debugfs_root, &zswap_pool_total_size);
1237 debugfs_create_atomic_t("stored_pages", S_IRUGO,
1238 zswap_debugfs_root, &zswap_stored_pages);
1243 static void __exit zswap_debugfs_exit(void)
1245 debugfs_remove_recursive(zswap_debugfs_root);
1248 static int __init zswap_debugfs_init(void)
1253 static void __exit zswap_debugfs_exit(void) { }
1256 /*********************************
1257 * module init and exit
1258 **********************************/
1259 static int __init init_zswap(void)
1261 struct zswap_pool *pool;
1263 zswap_init_started = true;
1265 if (zswap_entry_cache_create()) {
1266 pr_err("entry cache creation failed\n");
1270 if (zswap_cpu_dstmem_init()) {
1271 pr_err("dstmem alloc failed\n");
1275 pool = __zswap_pool_create_fallback();
1277 pr_err("pool creation failed\n");
1280 pr_info("loaded using pool %s/%s\n", pool->tfm_name,
1281 zpool_get_type(pool->zpool));
1283 list_add(&pool->list, &zswap_pools);
1285 frontswap_register_ops(&zswap_frontswap_ops);
1286 if (zswap_debugfs_init())
1287 pr_warn("debugfs initialization failed\n");
1291 zswap_cpu_dstmem_destroy();
1293 zswap_entry_cache_destroy();
1295 /* if built-in, we aren't unloaded on failure; don't allow use */
1296 zswap_init_failed = true;
1297 zswap_enabled = false;
1300 /* must be late so crypto has time to come up */
1301 late_initcall(init_zswap);
1303 MODULE_LICENSE("GPL");
1304 MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>");
1305 MODULE_DESCRIPTION("Compressed cache for swap pages");