]>
Commit | Line | Data |
---|---|---|
2b281117 SJ |
1 | /* |
2 | * zswap.c - zswap driver file | |
3 | * | |
4 | * zswap is a backend for frontswap that takes pages that are in the process | |
5 | * of being swapped out and attempts to compress and store them in a | |
6 | * RAM-based memory pool. This can result in a significant I/O reduction on | |
7 | * the swap device and, in the case where decompressing from RAM is faster | |
8 | * than reading from the swap device, can also improve workload performance. | |
9 | * | |
10 | * Copyright (C) 2012 Seth Jennings <sjenning@linux.vnet.ibm.com> | |
11 | * | |
12 | * This program is free software; you can redistribute it and/or | |
13 | * modify it under the terms of the GNU General Public License | |
14 | * as published by the Free Software Foundation; either version 2 | |
15 | * of the License, or (at your option) any later version. | |
16 | * | |
17 | * This program is distributed in the hope that it will be useful, | |
18 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
19 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
20 | * GNU General Public License for more details. | |
21 | */ | |
22 | ||
23 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/cpu.h> | |
27 | #include <linux/highmem.h> | |
28 | #include <linux/slab.h> | |
29 | #include <linux/spinlock.h> | |
30 | #include <linux/types.h> | |
31 | #include <linux/atomic.h> | |
32 | #include <linux/frontswap.h> | |
33 | #include <linux/rbtree.h> | |
34 | #include <linux/swap.h> | |
35 | #include <linux/crypto.h> | |
36 | #include <linux/mempool.h> | |
12d79d64 | 37 | #include <linux/zpool.h> |
2b281117 SJ |
38 | |
39 | #include <linux/mm_types.h> | |
40 | #include <linux/page-flags.h> | |
41 | #include <linux/swapops.h> | |
42 | #include <linux/writeback.h> | |
43 | #include <linux/pagemap.h> | |
44 | ||
45 | /********************************* | |
46 | * statistics | |
47 | **********************************/ | |
12d79d64 DS |
48 | /* Total bytes used by the compressed storage */ |
49 | static u64 zswap_pool_total_size; | |
2b281117 SJ |
50 | /* The number of compressed pages currently stored in zswap */ |
51 | static atomic_t zswap_stored_pages = ATOMIC_INIT(0); | |
52 | ||
53 | /* | |
54 | * The statistics below are not protected from concurrent access for | |
55 | * performance reasons so they may not be a 100% accurate. However, | |
56 | * they do provide useful information on roughly how many times a | |
57 | * certain event is occurring. | |
58 | */ | |
59 | ||
60 | /* Pool limit was hit (see zswap_max_pool_percent) */ | |
61 | static u64 zswap_pool_limit_hit; | |
62 | /* Pages written back when pool limit was reached */ | |
63 | static u64 zswap_written_back_pages; | |
64 | /* Store failed due to a reclaim failure after pool limit was reached */ | |
65 | static u64 zswap_reject_reclaim_fail; | |
66 | /* Compressed page was too big for the allocator to (optimally) store */ | |
67 | static u64 zswap_reject_compress_poor; | |
68 | /* Store failed because underlying allocator could not get memory */ | |
69 | static u64 zswap_reject_alloc_fail; | |
70 | /* Store failed because the entry metadata could not be allocated (rare) */ | |
71 | static u64 zswap_reject_kmemcache_fail; | |
72 | /* Duplicate store was encountered (rare) */ | |
73 | static u64 zswap_duplicate_entry; | |
74 | ||
75 | /********************************* | |
76 | * tunables | |
77 | **********************************/ | |
c00ed16a DS |
78 | |
79 | /* Enable/disable zswap (disabled by default) */ | |
80 | static bool zswap_enabled; | |
81 | module_param_named(enabled, zswap_enabled, bool, 0644); | |
2b281117 | 82 | |
90b0fc26 | 83 | /* Crypto compressor to use */ |
2b281117 | 84 | #define ZSWAP_COMPRESSOR_DEFAULT "lzo" |
c99b42c3 | 85 | static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT; |
90b0fc26 DS |
86 | static int zswap_compressor_param_set(const char *, |
87 | const struct kernel_param *); | |
88 | static struct kernel_param_ops zswap_compressor_param_ops = { | |
89 | .set = zswap_compressor_param_set, | |
c99b42c3 DS |
90 | .get = param_get_charp, |
91 | .free = param_free_charp, | |
90b0fc26 DS |
92 | }; |
93 | module_param_cb(compressor, &zswap_compressor_param_ops, | |
c99b42c3 | 94 | &zswap_compressor, 0644); |
2b281117 | 95 | |
90b0fc26 | 96 | /* Compressed storage zpool to use */ |
12d79d64 | 97 | #define ZSWAP_ZPOOL_DEFAULT "zbud" |
c99b42c3 | 98 | static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT; |
90b0fc26 DS |
99 | static int zswap_zpool_param_set(const char *, const struct kernel_param *); |
100 | static struct kernel_param_ops zswap_zpool_param_ops = { | |
c99b42c3 DS |
101 | .set = zswap_zpool_param_set, |
102 | .get = param_get_charp, | |
103 | .free = param_free_charp, | |
90b0fc26 | 104 | }; |
c99b42c3 | 105 | module_param_cb(zpool, &zswap_zpool_param_ops, &zswap_zpool_type, 0644); |
12d79d64 | 106 | |
90b0fc26 DS |
107 | /* The maximum percentage of memory that the compressed pool can occupy */ |
108 | static unsigned int zswap_max_pool_percent = 20; | |
109 | module_param_named(max_pool_percent, zswap_max_pool_percent, uint, 0644); | |
60105e12 | 110 | |
2b281117 | 111 | /********************************* |
f1c54846 | 112 | * data structures |
2b281117 | 113 | **********************************/ |
2b281117 | 114 | |
f1c54846 DS |
115 | struct zswap_pool { |
116 | struct zpool *zpool; | |
117 | struct crypto_comp * __percpu *tfm; | |
118 | struct kref kref; | |
119 | struct list_head list; | |
200867af | 120 | struct work_struct work; |
f1c54846 DS |
121 | struct notifier_block notifier; |
122 | char tfm_name[CRYPTO_MAX_ALG_NAME]; | |
2b281117 SJ |
123 | }; |
124 | ||
2b281117 SJ |
125 | /* |
126 | * struct zswap_entry | |
127 | * | |
128 | * This structure contains the metadata for tracking a single compressed | |
129 | * page within zswap. | |
130 | * | |
131 | * rbnode - links the entry into red-black tree for the appropriate swap type | |
f1c54846 | 132 | * offset - the swap offset for the entry. Index into the red-black tree. |
2b281117 SJ |
133 | * refcount - the number of outstanding reference to the entry. This is needed |
134 | * to protect against premature freeing of the entry by code | |
6b452516 | 135 | * concurrent calls to load, invalidate, and writeback. The lock |
2b281117 SJ |
136 | * for the zswap_tree structure that contains the entry must |
137 | * be held while changing the refcount. Since the lock must | |
138 | * be held, there is no reason to also make refcount atomic. | |
2b281117 | 139 | * length - the length in bytes of the compressed page data. Needed during |
6b452516 | 140 | * decompression |
f1c54846 DS |
141 | * pool - the zswap_pool the entry's data is in |
142 | * handle - zpool allocation handle that stores the compressed page data | |
2b281117 SJ |
143 | */ |
144 | struct zswap_entry { | |
145 | struct rb_node rbnode; | |
146 | pgoff_t offset; | |
147 | int refcount; | |
148 | unsigned int length; | |
f1c54846 | 149 | struct zswap_pool *pool; |
2b281117 SJ |
150 | unsigned long handle; |
151 | }; | |
152 | ||
153 | struct zswap_header { | |
154 | swp_entry_t swpentry; | |
155 | }; | |
156 | ||
157 | /* | |
158 | * The tree lock in the zswap_tree struct protects a few things: | |
159 | * - the rbtree | |
160 | * - the refcount field of each entry in the tree | |
161 | */ | |
162 | struct zswap_tree { | |
163 | struct rb_root rbroot; | |
164 | spinlock_t lock; | |
2b281117 SJ |
165 | }; |
166 | ||
167 | static struct zswap_tree *zswap_trees[MAX_SWAPFILES]; | |
168 | ||
f1c54846 DS |
169 | /* RCU-protected iteration */ |
170 | static LIST_HEAD(zswap_pools); | |
171 | /* protects zswap_pools list modification */ | |
172 | static DEFINE_SPINLOCK(zswap_pools_lock); | |
32a4e169 DS |
173 | /* pool counter to provide unique names to zpool */ |
174 | static atomic_t zswap_pools_count = ATOMIC_INIT(0); | |
f1c54846 | 175 | |
90b0fc26 DS |
176 | /* used by param callback function */ |
177 | static bool zswap_init_started; | |
178 | ||
f1c54846 DS |
179 | /********************************* |
180 | * helpers and fwd declarations | |
181 | **********************************/ | |
182 | ||
183 | #define zswap_pool_debug(msg, p) \ | |
184 | pr_debug("%s pool %s/%s\n", msg, (p)->tfm_name, \ | |
185 | zpool_get_type((p)->zpool)) | |
186 | ||
187 | static int zswap_writeback_entry(struct zpool *pool, unsigned long handle); | |
188 | static int zswap_pool_get(struct zswap_pool *pool); | |
189 | static void zswap_pool_put(struct zswap_pool *pool); | |
190 | ||
191 | static const struct zpool_ops zswap_zpool_ops = { | |
192 | .evict = zswap_writeback_entry | |
193 | }; | |
194 | ||
195 | static bool zswap_is_full(void) | |
196 | { | |
197 | return totalram_pages * zswap_max_pool_percent / 100 < | |
198 | DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE); | |
199 | } | |
200 | ||
201 | static void zswap_update_total_size(void) | |
202 | { | |
203 | struct zswap_pool *pool; | |
204 | u64 total = 0; | |
205 | ||
206 | rcu_read_lock(); | |
207 | ||
208 | list_for_each_entry_rcu(pool, &zswap_pools, list) | |
209 | total += zpool_get_total_size(pool->zpool); | |
210 | ||
211 | rcu_read_unlock(); | |
212 | ||
213 | zswap_pool_total_size = total; | |
214 | } | |
215 | ||
2b281117 SJ |
216 | /********************************* |
217 | * zswap entry functions | |
218 | **********************************/ | |
219 | static struct kmem_cache *zswap_entry_cache; | |
220 | ||
dd01d7d8 | 221 | static int __init zswap_entry_cache_create(void) |
2b281117 SJ |
222 | { |
223 | zswap_entry_cache = KMEM_CACHE(zswap_entry, 0); | |
5d2d42de | 224 | return zswap_entry_cache == NULL; |
2b281117 SJ |
225 | } |
226 | ||
c119239b | 227 | static void __init zswap_entry_cache_destroy(void) |
2b281117 SJ |
228 | { |
229 | kmem_cache_destroy(zswap_entry_cache); | |
230 | } | |
231 | ||
232 | static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp) | |
233 | { | |
234 | struct zswap_entry *entry; | |
235 | entry = kmem_cache_alloc(zswap_entry_cache, gfp); | |
236 | if (!entry) | |
237 | return NULL; | |
238 | entry->refcount = 1; | |
0ab0abcf | 239 | RB_CLEAR_NODE(&entry->rbnode); |
2b281117 SJ |
240 | return entry; |
241 | } | |
242 | ||
243 | static void zswap_entry_cache_free(struct zswap_entry *entry) | |
244 | { | |
245 | kmem_cache_free(zswap_entry_cache, entry); | |
246 | } | |
247 | ||
2b281117 SJ |
248 | /********************************* |
249 | * rbtree functions | |
250 | **********************************/ | |
251 | static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset) | |
252 | { | |
253 | struct rb_node *node = root->rb_node; | |
254 | struct zswap_entry *entry; | |
255 | ||
256 | while (node) { | |
257 | entry = rb_entry(node, struct zswap_entry, rbnode); | |
258 | if (entry->offset > offset) | |
259 | node = node->rb_left; | |
260 | else if (entry->offset < offset) | |
261 | node = node->rb_right; | |
262 | else | |
263 | return entry; | |
264 | } | |
265 | return NULL; | |
266 | } | |
267 | ||
268 | /* | |
269 | * In the case that a entry with the same offset is found, a pointer to | |
270 | * the existing entry is stored in dupentry and the function returns -EEXIST | |
271 | */ | |
272 | static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry, | |
273 | struct zswap_entry **dupentry) | |
274 | { | |
275 | struct rb_node **link = &root->rb_node, *parent = NULL; | |
276 | struct zswap_entry *myentry; | |
277 | ||
278 | while (*link) { | |
279 | parent = *link; | |
280 | myentry = rb_entry(parent, struct zswap_entry, rbnode); | |
281 | if (myentry->offset > entry->offset) | |
282 | link = &(*link)->rb_left; | |
283 | else if (myentry->offset < entry->offset) | |
284 | link = &(*link)->rb_right; | |
285 | else { | |
286 | *dupentry = myentry; | |
287 | return -EEXIST; | |
288 | } | |
289 | } | |
290 | rb_link_node(&entry->rbnode, parent, link); | |
291 | rb_insert_color(&entry->rbnode, root); | |
292 | return 0; | |
293 | } | |
294 | ||
0ab0abcf WY |
295 | static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry) |
296 | { | |
297 | if (!RB_EMPTY_NODE(&entry->rbnode)) { | |
298 | rb_erase(&entry->rbnode, root); | |
299 | RB_CLEAR_NODE(&entry->rbnode); | |
300 | } | |
301 | } | |
302 | ||
303 | /* | |
12d79d64 | 304 | * Carries out the common pattern of freeing and entry's zpool allocation, |
0ab0abcf WY |
305 | * freeing the entry itself, and decrementing the number of stored pages. |
306 | */ | |
60105e12 | 307 | static void zswap_free_entry(struct zswap_entry *entry) |
0ab0abcf | 308 | { |
f1c54846 DS |
309 | zpool_free(entry->pool->zpool, entry->handle); |
310 | zswap_pool_put(entry->pool); | |
0ab0abcf WY |
311 | zswap_entry_cache_free(entry); |
312 | atomic_dec(&zswap_stored_pages); | |
f1c54846 | 313 | zswap_update_total_size(); |
0ab0abcf WY |
314 | } |
315 | ||
316 | /* caller must hold the tree lock */ | |
317 | static void zswap_entry_get(struct zswap_entry *entry) | |
318 | { | |
319 | entry->refcount++; | |
320 | } | |
321 | ||
322 | /* caller must hold the tree lock | |
323 | * remove from the tree and free it, if nobody reference the entry | |
324 | */ | |
325 | static void zswap_entry_put(struct zswap_tree *tree, | |
326 | struct zswap_entry *entry) | |
327 | { | |
328 | int refcount = --entry->refcount; | |
329 | ||
330 | BUG_ON(refcount < 0); | |
331 | if (refcount == 0) { | |
332 | zswap_rb_erase(&tree->rbroot, entry); | |
60105e12 | 333 | zswap_free_entry(entry); |
0ab0abcf WY |
334 | } |
335 | } | |
336 | ||
337 | /* caller must hold the tree lock */ | |
338 | static struct zswap_entry *zswap_entry_find_get(struct rb_root *root, | |
339 | pgoff_t offset) | |
340 | { | |
b0c9865f | 341 | struct zswap_entry *entry; |
0ab0abcf WY |
342 | |
343 | entry = zswap_rb_search(root, offset); | |
344 | if (entry) | |
345 | zswap_entry_get(entry); | |
346 | ||
347 | return entry; | |
348 | } | |
349 | ||
2b281117 SJ |
350 | /********************************* |
351 | * per-cpu code | |
352 | **********************************/ | |
353 | static DEFINE_PER_CPU(u8 *, zswap_dstmem); | |
354 | ||
ad7ed770 | 355 | static int zswap_dstmem_prepare(unsigned int cpu) |
2b281117 | 356 | { |
2b281117 SJ |
357 | u8 *dst; |
358 | ||
ad7ed770 SAS |
359 | dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu)); |
360 | if (!dst) { | |
361 | pr_err("can't allocate compressor buffer\n"); | |
362 | return -ENOMEM; | |
2b281117 | 363 | } |
ad7ed770 SAS |
364 | per_cpu(zswap_dstmem, cpu) = dst; |
365 | return 0; | |
2b281117 SJ |
366 | } |
367 | ||
ad7ed770 | 368 | static int zswap_dstmem_dead(unsigned int cpu) |
2b281117 | 369 | { |
ad7ed770 | 370 | u8 *dst; |
2b281117 | 371 | |
ad7ed770 SAS |
372 | dst = per_cpu(zswap_dstmem, cpu); |
373 | kfree(dst); | |
374 | per_cpu(zswap_dstmem, cpu) = NULL; | |
f1c54846 | 375 | |
f1c54846 | 376 | return 0; |
f1c54846 DS |
377 | } |
378 | ||
379 | static int __zswap_cpu_comp_notifier(struct zswap_pool *pool, | |
380 | unsigned long action, unsigned long cpu) | |
381 | { | |
382 | struct crypto_comp *tfm; | |
383 | ||
384 | switch (action) { | |
385 | case CPU_UP_PREPARE: | |
386 | if (WARN_ON(*per_cpu_ptr(pool->tfm, cpu))) | |
387 | break; | |
388 | tfm = crypto_alloc_comp(pool->tfm_name, 0, 0); | |
389 | if (IS_ERR_OR_NULL(tfm)) { | |
390 | pr_err("could not alloc crypto comp %s : %ld\n", | |
391 | pool->tfm_name, PTR_ERR(tfm)); | |
392 | return NOTIFY_BAD; | |
393 | } | |
394 | *per_cpu_ptr(pool->tfm, cpu) = tfm; | |
395 | break; | |
396 | case CPU_DEAD: | |
397 | case CPU_UP_CANCELED: | |
398 | tfm = *per_cpu_ptr(pool->tfm, cpu); | |
399 | if (!IS_ERR_OR_NULL(tfm)) | |
400 | crypto_free_comp(tfm); | |
401 | *per_cpu_ptr(pool->tfm, cpu) = NULL; | |
402 | break; | |
403 | default: | |
404 | break; | |
405 | } | |
406 | return NOTIFY_OK; | |
407 | } | |
408 | ||
409 | static int zswap_cpu_comp_notifier(struct notifier_block *nb, | |
410 | unsigned long action, void *pcpu) | |
411 | { | |
412 | unsigned long cpu = (unsigned long)pcpu; | |
413 | struct zswap_pool *pool = container_of(nb, typeof(*pool), notifier); | |
414 | ||
415 | return __zswap_cpu_comp_notifier(pool, action, cpu); | |
416 | } | |
417 | ||
418 | static int zswap_cpu_comp_init(struct zswap_pool *pool) | |
2b281117 SJ |
419 | { |
420 | unsigned long cpu; | |
421 | ||
f1c54846 DS |
422 | memset(&pool->notifier, 0, sizeof(pool->notifier)); |
423 | pool->notifier.notifier_call = zswap_cpu_comp_notifier; | |
424 | ||
57637824 | 425 | cpu_notifier_register_begin(); |
2b281117 | 426 | for_each_online_cpu(cpu) |
f1c54846 DS |
427 | if (__zswap_cpu_comp_notifier(pool, CPU_UP_PREPARE, cpu) == |
428 | NOTIFY_BAD) | |
2b281117 | 429 | goto cleanup; |
f1c54846 | 430 | __register_cpu_notifier(&pool->notifier); |
57637824 | 431 | cpu_notifier_register_done(); |
2b281117 SJ |
432 | return 0; |
433 | ||
434 | cleanup: | |
435 | for_each_online_cpu(cpu) | |
f1c54846 | 436 | __zswap_cpu_comp_notifier(pool, CPU_UP_CANCELED, cpu); |
57637824 | 437 | cpu_notifier_register_done(); |
2b281117 SJ |
438 | return -ENOMEM; |
439 | } | |
440 | ||
f1c54846 DS |
441 | static void zswap_cpu_comp_destroy(struct zswap_pool *pool) |
442 | { | |
443 | unsigned long cpu; | |
444 | ||
445 | cpu_notifier_register_begin(); | |
446 | for_each_online_cpu(cpu) | |
447 | __zswap_cpu_comp_notifier(pool, CPU_UP_CANCELED, cpu); | |
448 | __unregister_cpu_notifier(&pool->notifier); | |
449 | cpu_notifier_register_done(); | |
450 | } | |
451 | ||
2b281117 | 452 | /********************************* |
f1c54846 | 453 | * pool functions |
2b281117 | 454 | **********************************/ |
f1c54846 DS |
455 | |
456 | static struct zswap_pool *__zswap_pool_current(void) | |
2b281117 | 457 | { |
f1c54846 DS |
458 | struct zswap_pool *pool; |
459 | ||
460 | pool = list_first_or_null_rcu(&zswap_pools, typeof(*pool), list); | |
461 | WARN_ON(!pool); | |
462 | ||
463 | return pool; | |
464 | } | |
465 | ||
466 | static struct zswap_pool *zswap_pool_current(void) | |
467 | { | |
468 | assert_spin_locked(&zswap_pools_lock); | |
469 | ||
470 | return __zswap_pool_current(); | |
471 | } | |
472 | ||
473 | static struct zswap_pool *zswap_pool_current_get(void) | |
474 | { | |
475 | struct zswap_pool *pool; | |
476 | ||
477 | rcu_read_lock(); | |
478 | ||
479 | pool = __zswap_pool_current(); | |
480 | if (!pool || !zswap_pool_get(pool)) | |
481 | pool = NULL; | |
482 | ||
483 | rcu_read_unlock(); | |
484 | ||
485 | return pool; | |
486 | } | |
487 | ||
488 | static struct zswap_pool *zswap_pool_last_get(void) | |
489 | { | |
490 | struct zswap_pool *pool, *last = NULL; | |
491 | ||
492 | rcu_read_lock(); | |
493 | ||
494 | list_for_each_entry_rcu(pool, &zswap_pools, list) | |
495 | last = pool; | |
496 | if (!WARN_ON(!last) && !zswap_pool_get(last)) | |
497 | last = NULL; | |
498 | ||
499 | rcu_read_unlock(); | |
500 | ||
501 | return last; | |
502 | } | |
503 | ||
8bc8b228 | 504 | /* type and compressor must be null-terminated */ |
f1c54846 DS |
505 | static struct zswap_pool *zswap_pool_find_get(char *type, char *compressor) |
506 | { | |
507 | struct zswap_pool *pool; | |
508 | ||
509 | assert_spin_locked(&zswap_pools_lock); | |
510 | ||
511 | list_for_each_entry_rcu(pool, &zswap_pools, list) { | |
8bc8b228 | 512 | if (strcmp(pool->tfm_name, compressor)) |
f1c54846 | 513 | continue; |
8bc8b228 | 514 | if (strcmp(zpool_get_type(pool->zpool), type)) |
f1c54846 DS |
515 | continue; |
516 | /* if we can't get it, it's about to be destroyed */ | |
517 | if (!zswap_pool_get(pool)) | |
518 | continue; | |
519 | return pool; | |
520 | } | |
521 | ||
522 | return NULL; | |
523 | } | |
524 | ||
525 | static struct zswap_pool *zswap_pool_create(char *type, char *compressor) | |
526 | { | |
527 | struct zswap_pool *pool; | |
32a4e169 | 528 | char name[38]; /* 'zswap' + 32 char (max) num + \0 */ |
d0164adc | 529 | gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM; |
f1c54846 DS |
530 | |
531 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); | |
532 | if (!pool) { | |
533 | pr_err("pool alloc failed\n"); | |
534 | return NULL; | |
535 | } | |
536 | ||
32a4e169 DS |
537 | /* unique name for each pool specifically required by zsmalloc */ |
538 | snprintf(name, 38, "zswap%x", atomic_inc_return(&zswap_pools_count)); | |
539 | ||
540 | pool->zpool = zpool_create_pool(type, name, gfp, &zswap_zpool_ops); | |
f1c54846 DS |
541 | if (!pool->zpool) { |
542 | pr_err("%s zpool not available\n", type); | |
543 | goto error; | |
544 | } | |
545 | pr_debug("using %s zpool\n", zpool_get_type(pool->zpool)); | |
546 | ||
547 | strlcpy(pool->tfm_name, compressor, sizeof(pool->tfm_name)); | |
548 | pool->tfm = alloc_percpu(struct crypto_comp *); | |
549 | if (!pool->tfm) { | |
550 | pr_err("percpu alloc failed\n"); | |
551 | goto error; | |
552 | } | |
553 | ||
554 | if (zswap_cpu_comp_init(pool)) | |
555 | goto error; | |
556 | pr_debug("using %s compressor\n", pool->tfm_name); | |
557 | ||
558 | /* being the current pool takes 1 ref; this func expects the | |
559 | * caller to always add the new pool as the current pool | |
560 | */ | |
561 | kref_init(&pool->kref); | |
562 | INIT_LIST_HEAD(&pool->list); | |
563 | ||
564 | zswap_pool_debug("created", pool); | |
565 | ||
566 | return pool; | |
567 | ||
568 | error: | |
569 | free_percpu(pool->tfm); | |
570 | if (pool->zpool) | |
571 | zpool_destroy_pool(pool->zpool); | |
572 | kfree(pool); | |
573 | return NULL; | |
574 | } | |
575 | ||
c99b42c3 | 576 | static __init struct zswap_pool *__zswap_pool_create_fallback(void) |
f1c54846 DS |
577 | { |
578 | if (!crypto_has_comp(zswap_compressor, 0, 0)) { | |
c99b42c3 DS |
579 | if (!strcmp(zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT)) { |
580 | pr_err("default compressor %s not available\n", | |
581 | zswap_compressor); | |
582 | return NULL; | |
583 | } | |
f1c54846 DS |
584 | pr_err("compressor %s not available, using default %s\n", |
585 | zswap_compressor, ZSWAP_COMPRESSOR_DEFAULT); | |
c99b42c3 DS |
586 | param_free_charp(&zswap_compressor); |
587 | zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT; | |
f1c54846 DS |
588 | } |
589 | if (!zpool_has_pool(zswap_zpool_type)) { | |
c99b42c3 DS |
590 | if (!strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) { |
591 | pr_err("default zpool %s not available\n", | |
592 | zswap_zpool_type); | |
593 | return NULL; | |
594 | } | |
f1c54846 DS |
595 | pr_err("zpool %s not available, using default %s\n", |
596 | zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT); | |
c99b42c3 DS |
597 | param_free_charp(&zswap_zpool_type); |
598 | zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT; | |
f1c54846 DS |
599 | } |
600 | ||
601 | return zswap_pool_create(zswap_zpool_type, zswap_compressor); | |
602 | } | |
603 | ||
604 | static void zswap_pool_destroy(struct zswap_pool *pool) | |
605 | { | |
606 | zswap_pool_debug("destroying", pool); | |
607 | ||
608 | zswap_cpu_comp_destroy(pool); | |
609 | free_percpu(pool->tfm); | |
610 | zpool_destroy_pool(pool->zpool); | |
611 | kfree(pool); | |
612 | } | |
613 | ||
614 | static int __must_check zswap_pool_get(struct zswap_pool *pool) | |
615 | { | |
616 | return kref_get_unless_zero(&pool->kref); | |
617 | } | |
618 | ||
200867af | 619 | static void __zswap_pool_release(struct work_struct *work) |
f1c54846 | 620 | { |
200867af DS |
621 | struct zswap_pool *pool = container_of(work, typeof(*pool), work); |
622 | ||
623 | synchronize_rcu(); | |
f1c54846 DS |
624 | |
625 | /* nobody should have been able to get a kref... */ | |
626 | WARN_ON(kref_get_unless_zero(&pool->kref)); | |
627 | ||
628 | /* pool is now off zswap_pools list and has no references. */ | |
629 | zswap_pool_destroy(pool); | |
630 | } | |
631 | ||
632 | static void __zswap_pool_empty(struct kref *kref) | |
633 | { | |
634 | struct zswap_pool *pool; | |
635 | ||
636 | pool = container_of(kref, typeof(*pool), kref); | |
637 | ||
638 | spin_lock(&zswap_pools_lock); | |
639 | ||
640 | WARN_ON(pool == zswap_pool_current()); | |
641 | ||
642 | list_del_rcu(&pool->list); | |
200867af DS |
643 | |
644 | INIT_WORK(&pool->work, __zswap_pool_release); | |
645 | schedule_work(&pool->work); | |
f1c54846 DS |
646 | |
647 | spin_unlock(&zswap_pools_lock); | |
648 | } | |
649 | ||
650 | static void zswap_pool_put(struct zswap_pool *pool) | |
651 | { | |
652 | kref_put(&pool->kref, __zswap_pool_empty); | |
2b281117 SJ |
653 | } |
654 | ||
90b0fc26 DS |
655 | /********************************* |
656 | * param callbacks | |
657 | **********************************/ | |
658 | ||
c99b42c3 | 659 | /* val must be a null-terminated string */ |
90b0fc26 DS |
660 | static int __zswap_param_set(const char *val, const struct kernel_param *kp, |
661 | char *type, char *compressor) | |
662 | { | |
663 | struct zswap_pool *pool, *put_pool = NULL; | |
c99b42c3 | 664 | char *s = strstrip((char *)val); |
90b0fc26 DS |
665 | int ret; |
666 | ||
c99b42c3 DS |
667 | /* no change required */ |
668 | if (!strcmp(s, *(char **)kp->arg)) | |
669 | return 0; | |
90b0fc26 DS |
670 | |
671 | /* if this is load-time (pre-init) param setting, | |
672 | * don't create a pool; that's done during init. | |
673 | */ | |
674 | if (!zswap_init_started) | |
c99b42c3 | 675 | return param_set_charp(s, kp); |
90b0fc26 DS |
676 | |
677 | if (!type) { | |
c99b42c3 DS |
678 | if (!zpool_has_pool(s)) { |
679 | pr_err("zpool %s not available\n", s); | |
90b0fc26 DS |
680 | return -ENOENT; |
681 | } | |
c99b42c3 | 682 | type = s; |
90b0fc26 | 683 | } else if (!compressor) { |
c99b42c3 DS |
684 | if (!crypto_has_comp(s, 0, 0)) { |
685 | pr_err("compressor %s not available\n", s); | |
90b0fc26 DS |
686 | return -ENOENT; |
687 | } | |
c99b42c3 DS |
688 | compressor = s; |
689 | } else { | |
690 | WARN_ON(1); | |
691 | return -EINVAL; | |
90b0fc26 DS |
692 | } |
693 | ||
694 | spin_lock(&zswap_pools_lock); | |
695 | ||
696 | pool = zswap_pool_find_get(type, compressor); | |
697 | if (pool) { | |
698 | zswap_pool_debug("using existing", pool); | |
699 | list_del_rcu(&pool->list); | |
700 | } else { | |
701 | spin_unlock(&zswap_pools_lock); | |
702 | pool = zswap_pool_create(type, compressor); | |
703 | spin_lock(&zswap_pools_lock); | |
704 | } | |
705 | ||
706 | if (pool) | |
c99b42c3 | 707 | ret = param_set_charp(s, kp); |
90b0fc26 DS |
708 | else |
709 | ret = -EINVAL; | |
710 | ||
711 | if (!ret) { | |
712 | put_pool = zswap_pool_current(); | |
713 | list_add_rcu(&pool->list, &zswap_pools); | |
714 | } else if (pool) { | |
715 | /* add the possibly pre-existing pool to the end of the pools | |
716 | * list; if it's new (and empty) then it'll be removed and | |
717 | * destroyed by the put after we drop the lock | |
718 | */ | |
719 | list_add_tail_rcu(&pool->list, &zswap_pools); | |
720 | put_pool = pool; | |
721 | } | |
722 | ||
723 | spin_unlock(&zswap_pools_lock); | |
724 | ||
725 | /* drop the ref from either the old current pool, | |
726 | * or the new pool we failed to add | |
727 | */ | |
728 | if (put_pool) | |
729 | zswap_pool_put(put_pool); | |
730 | ||
731 | return ret; | |
732 | } | |
733 | ||
734 | static int zswap_compressor_param_set(const char *val, | |
735 | const struct kernel_param *kp) | |
736 | { | |
737 | return __zswap_param_set(val, kp, zswap_zpool_type, NULL); | |
738 | } | |
739 | ||
740 | static int zswap_zpool_param_set(const char *val, | |
741 | const struct kernel_param *kp) | |
742 | { | |
743 | return __zswap_param_set(val, kp, NULL, zswap_compressor); | |
744 | } | |
745 | ||
2b281117 SJ |
746 | /********************************* |
747 | * writeback code | |
748 | **********************************/ | |
749 | /* return enum for zswap_get_swap_cache_page */ | |
750 | enum zswap_get_swap_ret { | |
751 | ZSWAP_SWAPCACHE_NEW, | |
752 | ZSWAP_SWAPCACHE_EXIST, | |
67d13fe8 | 753 | ZSWAP_SWAPCACHE_FAIL, |
2b281117 SJ |
754 | }; |
755 | ||
756 | /* | |
757 | * zswap_get_swap_cache_page | |
758 | * | |
759 | * This is an adaption of read_swap_cache_async() | |
760 | * | |
761 | * This function tries to find a page with the given swap entry | |
762 | * in the swapper_space address space (the swap cache). If the page | |
763 | * is found, it is returned in retpage. Otherwise, a page is allocated, | |
764 | * added to the swap cache, and returned in retpage. | |
765 | * | |
766 | * If success, the swap cache page is returned in retpage | |
67d13fe8 WY |
767 | * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache |
768 | * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated, | |
769 | * the new page is added to swapcache and locked | |
770 | * Returns ZSWAP_SWAPCACHE_FAIL on error | |
2b281117 SJ |
771 | */ |
772 | static int zswap_get_swap_cache_page(swp_entry_t entry, | |
773 | struct page **retpage) | |
774 | { | |
5b999aad | 775 | bool page_was_allocated; |
2b281117 | 776 | |
5b999aad DS |
777 | *retpage = __read_swap_cache_async(entry, GFP_KERNEL, |
778 | NULL, 0, &page_was_allocated); | |
779 | if (page_was_allocated) | |
780 | return ZSWAP_SWAPCACHE_NEW; | |
781 | if (!*retpage) | |
67d13fe8 | 782 | return ZSWAP_SWAPCACHE_FAIL; |
2b281117 SJ |
783 | return ZSWAP_SWAPCACHE_EXIST; |
784 | } | |
785 | ||
786 | /* | |
787 | * Attempts to free an entry by adding a page to the swap cache, | |
788 | * decompressing the entry data into the page, and issuing a | |
789 | * bio write to write the page back to the swap device. | |
790 | * | |
791 | * This can be thought of as a "resumed writeback" of the page | |
792 | * to the swap device. We are basically resuming the same swap | |
793 | * writeback path that was intercepted with the frontswap_store() | |
794 | * in the first place. After the page has been decompressed into | |
795 | * the swap cache, the compressed version stored by zswap can be | |
796 | * freed. | |
797 | */ | |
12d79d64 | 798 | static int zswap_writeback_entry(struct zpool *pool, unsigned long handle) |
2b281117 SJ |
799 | { |
800 | struct zswap_header *zhdr; | |
801 | swp_entry_t swpentry; | |
802 | struct zswap_tree *tree; | |
803 | pgoff_t offset; | |
804 | struct zswap_entry *entry; | |
805 | struct page *page; | |
f1c54846 | 806 | struct crypto_comp *tfm; |
2b281117 SJ |
807 | u8 *src, *dst; |
808 | unsigned int dlen; | |
0ab0abcf | 809 | int ret; |
2b281117 SJ |
810 | struct writeback_control wbc = { |
811 | .sync_mode = WB_SYNC_NONE, | |
812 | }; | |
813 | ||
814 | /* extract swpentry from data */ | |
12d79d64 | 815 | zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO); |
2b281117 | 816 | swpentry = zhdr->swpentry; /* here */ |
12d79d64 | 817 | zpool_unmap_handle(pool, handle); |
2b281117 SJ |
818 | tree = zswap_trees[swp_type(swpentry)]; |
819 | offset = swp_offset(swpentry); | |
2b281117 SJ |
820 | |
821 | /* find and ref zswap entry */ | |
822 | spin_lock(&tree->lock); | |
0ab0abcf | 823 | entry = zswap_entry_find_get(&tree->rbroot, offset); |
2b281117 SJ |
824 | if (!entry) { |
825 | /* entry was invalidated */ | |
826 | spin_unlock(&tree->lock); | |
827 | return 0; | |
828 | } | |
2b281117 SJ |
829 | spin_unlock(&tree->lock); |
830 | BUG_ON(offset != entry->offset); | |
831 | ||
832 | /* try to allocate swap cache page */ | |
833 | switch (zswap_get_swap_cache_page(swpentry, &page)) { | |
67d13fe8 | 834 | case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */ |
2b281117 SJ |
835 | ret = -ENOMEM; |
836 | goto fail; | |
837 | ||
67d13fe8 | 838 | case ZSWAP_SWAPCACHE_EXIST: |
2b281117 | 839 | /* page is already in the swap cache, ignore for now */ |
09cbfeaf | 840 | put_page(page); |
2b281117 SJ |
841 | ret = -EEXIST; |
842 | goto fail; | |
843 | ||
844 | case ZSWAP_SWAPCACHE_NEW: /* page is locked */ | |
845 | /* decompress */ | |
846 | dlen = PAGE_SIZE; | |
f1c54846 | 847 | src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle, |
12d79d64 | 848 | ZPOOL_MM_RO) + sizeof(struct zswap_header); |
2b281117 | 849 | dst = kmap_atomic(page); |
f1c54846 DS |
850 | tfm = *get_cpu_ptr(entry->pool->tfm); |
851 | ret = crypto_comp_decompress(tfm, src, entry->length, | |
852 | dst, &dlen); | |
853 | put_cpu_ptr(entry->pool->tfm); | |
2b281117 | 854 | kunmap_atomic(dst); |
f1c54846 | 855 | zpool_unmap_handle(entry->pool->zpool, entry->handle); |
2b281117 SJ |
856 | BUG_ON(ret); |
857 | BUG_ON(dlen != PAGE_SIZE); | |
858 | ||
859 | /* page is up to date */ | |
860 | SetPageUptodate(page); | |
861 | } | |
862 | ||
b349acc7 WY |
863 | /* move it to the tail of the inactive list after end_writeback */ |
864 | SetPageReclaim(page); | |
865 | ||
2b281117 SJ |
866 | /* start writeback */ |
867 | __swap_writepage(page, &wbc, end_swap_bio_write); | |
09cbfeaf | 868 | put_page(page); |
2b281117 SJ |
869 | zswap_written_back_pages++; |
870 | ||
871 | spin_lock(&tree->lock); | |
2b281117 | 872 | /* drop local reference */ |
0ab0abcf | 873 | zswap_entry_put(tree, entry); |
2b281117 SJ |
874 | |
875 | /* | |
0ab0abcf WY |
876 | * There are two possible situations for entry here: |
877 | * (1) refcount is 1(normal case), entry is valid and on the tree | |
878 | * (2) refcount is 0, entry is freed and not on the tree | |
879 | * because invalidate happened during writeback | |
880 | * search the tree and free the entry if find entry | |
881 | */ | |
882 | if (entry == zswap_rb_search(&tree->rbroot, offset)) | |
883 | zswap_entry_put(tree, entry); | |
2b281117 | 884 | spin_unlock(&tree->lock); |
2b281117 | 885 | |
0ab0abcf WY |
886 | goto end; |
887 | ||
888 | /* | |
889 | * if we get here due to ZSWAP_SWAPCACHE_EXIST | |
890 | * a load may happening concurrently | |
891 | * it is safe and okay to not free the entry | |
892 | * if we free the entry in the following put | |
893 | * it it either okay to return !0 | |
894 | */ | |
2b281117 SJ |
895 | fail: |
896 | spin_lock(&tree->lock); | |
0ab0abcf | 897 | zswap_entry_put(tree, entry); |
2b281117 | 898 | spin_unlock(&tree->lock); |
0ab0abcf WY |
899 | |
900 | end: | |
2b281117 SJ |
901 | return ret; |
902 | } | |
903 | ||
f1c54846 DS |
904 | static int zswap_shrink(void) |
905 | { | |
906 | struct zswap_pool *pool; | |
907 | int ret; | |
908 | ||
909 | pool = zswap_pool_last_get(); | |
910 | if (!pool) | |
911 | return -ENOENT; | |
912 | ||
913 | ret = zpool_shrink(pool->zpool, 1, NULL); | |
914 | ||
915 | zswap_pool_put(pool); | |
916 | ||
917 | return ret; | |
918 | } | |
919 | ||
2b281117 SJ |
920 | /********************************* |
921 | * frontswap hooks | |
922 | **********************************/ | |
923 | /* attempts to compress and store an single page */ | |
924 | static int zswap_frontswap_store(unsigned type, pgoff_t offset, | |
925 | struct page *page) | |
926 | { | |
927 | struct zswap_tree *tree = zswap_trees[type]; | |
928 | struct zswap_entry *entry, *dupentry; | |
f1c54846 | 929 | struct crypto_comp *tfm; |
2b281117 SJ |
930 | int ret; |
931 | unsigned int dlen = PAGE_SIZE, len; | |
932 | unsigned long handle; | |
933 | char *buf; | |
934 | u8 *src, *dst; | |
935 | struct zswap_header *zhdr; | |
936 | ||
c00ed16a | 937 | if (!zswap_enabled || !tree) { |
2b281117 SJ |
938 | ret = -ENODEV; |
939 | goto reject; | |
940 | } | |
941 | ||
942 | /* reclaim space if needed */ | |
943 | if (zswap_is_full()) { | |
944 | zswap_pool_limit_hit++; | |
f1c54846 | 945 | if (zswap_shrink()) { |
2b281117 SJ |
946 | zswap_reject_reclaim_fail++; |
947 | ret = -ENOMEM; | |
948 | goto reject; | |
949 | } | |
950 | } | |
951 | ||
952 | /* allocate entry */ | |
953 | entry = zswap_entry_cache_alloc(GFP_KERNEL); | |
954 | if (!entry) { | |
955 | zswap_reject_kmemcache_fail++; | |
956 | ret = -ENOMEM; | |
957 | goto reject; | |
958 | } | |
959 | ||
f1c54846 DS |
960 | /* if entry is successfully added, it keeps the reference */ |
961 | entry->pool = zswap_pool_current_get(); | |
962 | if (!entry->pool) { | |
963 | ret = -EINVAL; | |
964 | goto freepage; | |
965 | } | |
966 | ||
2b281117 SJ |
967 | /* compress */ |
968 | dst = get_cpu_var(zswap_dstmem); | |
f1c54846 | 969 | tfm = *get_cpu_ptr(entry->pool->tfm); |
2b281117 | 970 | src = kmap_atomic(page); |
f1c54846 | 971 | ret = crypto_comp_compress(tfm, src, PAGE_SIZE, dst, &dlen); |
2b281117 | 972 | kunmap_atomic(src); |
f1c54846 | 973 | put_cpu_ptr(entry->pool->tfm); |
2b281117 SJ |
974 | if (ret) { |
975 | ret = -EINVAL; | |
f1c54846 | 976 | goto put_dstmem; |
2b281117 SJ |
977 | } |
978 | ||
979 | /* store */ | |
980 | len = dlen + sizeof(struct zswap_header); | |
f1c54846 | 981 | ret = zpool_malloc(entry->pool->zpool, len, |
d0164adc MG |
982 | __GFP_NORETRY | __GFP_NOWARN | __GFP_KSWAPD_RECLAIM, |
983 | &handle); | |
2b281117 SJ |
984 | if (ret == -ENOSPC) { |
985 | zswap_reject_compress_poor++; | |
f1c54846 | 986 | goto put_dstmem; |
2b281117 SJ |
987 | } |
988 | if (ret) { | |
989 | zswap_reject_alloc_fail++; | |
f1c54846 | 990 | goto put_dstmem; |
2b281117 | 991 | } |
f1c54846 | 992 | zhdr = zpool_map_handle(entry->pool->zpool, handle, ZPOOL_MM_RW); |
2b281117 SJ |
993 | zhdr->swpentry = swp_entry(type, offset); |
994 | buf = (u8 *)(zhdr + 1); | |
995 | memcpy(buf, dst, dlen); | |
f1c54846 | 996 | zpool_unmap_handle(entry->pool->zpool, handle); |
2b281117 SJ |
997 | put_cpu_var(zswap_dstmem); |
998 | ||
999 | /* populate entry */ | |
1000 | entry->offset = offset; | |
1001 | entry->handle = handle; | |
1002 | entry->length = dlen; | |
1003 | ||
1004 | /* map */ | |
1005 | spin_lock(&tree->lock); | |
1006 | do { | |
1007 | ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry); | |
1008 | if (ret == -EEXIST) { | |
1009 | zswap_duplicate_entry++; | |
1010 | /* remove from rbtree */ | |
0ab0abcf WY |
1011 | zswap_rb_erase(&tree->rbroot, dupentry); |
1012 | zswap_entry_put(tree, dupentry); | |
2b281117 SJ |
1013 | } |
1014 | } while (ret == -EEXIST); | |
1015 | spin_unlock(&tree->lock); | |
1016 | ||
1017 | /* update stats */ | |
1018 | atomic_inc(&zswap_stored_pages); | |
f1c54846 | 1019 | zswap_update_total_size(); |
2b281117 SJ |
1020 | |
1021 | return 0; | |
1022 | ||
f1c54846 | 1023 | put_dstmem: |
2b281117 | 1024 | put_cpu_var(zswap_dstmem); |
f1c54846 DS |
1025 | zswap_pool_put(entry->pool); |
1026 | freepage: | |
2b281117 SJ |
1027 | zswap_entry_cache_free(entry); |
1028 | reject: | |
1029 | return ret; | |
1030 | } | |
1031 | ||
1032 | /* | |
1033 | * returns 0 if the page was successfully decompressed | |
1034 | * return -1 on entry not found or error | |
1035 | */ | |
1036 | static int zswap_frontswap_load(unsigned type, pgoff_t offset, | |
1037 | struct page *page) | |
1038 | { | |
1039 | struct zswap_tree *tree = zswap_trees[type]; | |
1040 | struct zswap_entry *entry; | |
f1c54846 | 1041 | struct crypto_comp *tfm; |
2b281117 SJ |
1042 | u8 *src, *dst; |
1043 | unsigned int dlen; | |
0ab0abcf | 1044 | int ret; |
2b281117 SJ |
1045 | |
1046 | /* find */ | |
1047 | spin_lock(&tree->lock); | |
0ab0abcf | 1048 | entry = zswap_entry_find_get(&tree->rbroot, offset); |
2b281117 SJ |
1049 | if (!entry) { |
1050 | /* entry was written back */ | |
1051 | spin_unlock(&tree->lock); | |
1052 | return -1; | |
1053 | } | |
2b281117 SJ |
1054 | spin_unlock(&tree->lock); |
1055 | ||
1056 | /* decompress */ | |
1057 | dlen = PAGE_SIZE; | |
f1c54846 | 1058 | src = (u8 *)zpool_map_handle(entry->pool->zpool, entry->handle, |
12d79d64 | 1059 | ZPOOL_MM_RO) + sizeof(struct zswap_header); |
2b281117 | 1060 | dst = kmap_atomic(page); |
f1c54846 DS |
1061 | tfm = *get_cpu_ptr(entry->pool->tfm); |
1062 | ret = crypto_comp_decompress(tfm, src, entry->length, dst, &dlen); | |
1063 | put_cpu_ptr(entry->pool->tfm); | |
2b281117 | 1064 | kunmap_atomic(dst); |
f1c54846 | 1065 | zpool_unmap_handle(entry->pool->zpool, entry->handle); |
2b281117 SJ |
1066 | BUG_ON(ret); |
1067 | ||
1068 | spin_lock(&tree->lock); | |
0ab0abcf | 1069 | zswap_entry_put(tree, entry); |
2b281117 SJ |
1070 | spin_unlock(&tree->lock); |
1071 | ||
2b281117 SJ |
1072 | return 0; |
1073 | } | |
1074 | ||
1075 | /* frees an entry in zswap */ | |
1076 | static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset) | |
1077 | { | |
1078 | struct zswap_tree *tree = zswap_trees[type]; | |
1079 | struct zswap_entry *entry; | |
2b281117 SJ |
1080 | |
1081 | /* find */ | |
1082 | spin_lock(&tree->lock); | |
1083 | entry = zswap_rb_search(&tree->rbroot, offset); | |
1084 | if (!entry) { | |
1085 | /* entry was written back */ | |
1086 | spin_unlock(&tree->lock); | |
1087 | return; | |
1088 | } | |
1089 | ||
1090 | /* remove from rbtree */ | |
0ab0abcf | 1091 | zswap_rb_erase(&tree->rbroot, entry); |
2b281117 SJ |
1092 | |
1093 | /* drop the initial reference from entry creation */ | |
0ab0abcf | 1094 | zswap_entry_put(tree, entry); |
2b281117 SJ |
1095 | |
1096 | spin_unlock(&tree->lock); | |
2b281117 SJ |
1097 | } |
1098 | ||
1099 | /* frees all zswap entries for the given swap type */ | |
1100 | static void zswap_frontswap_invalidate_area(unsigned type) | |
1101 | { | |
1102 | struct zswap_tree *tree = zswap_trees[type]; | |
0bd42136 | 1103 | struct zswap_entry *entry, *n; |
2b281117 SJ |
1104 | |
1105 | if (!tree) | |
1106 | return; | |
1107 | ||
1108 | /* walk the tree and free everything */ | |
1109 | spin_lock(&tree->lock); | |
0ab0abcf | 1110 | rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode) |
60105e12 | 1111 | zswap_free_entry(entry); |
2b281117 SJ |
1112 | tree->rbroot = RB_ROOT; |
1113 | spin_unlock(&tree->lock); | |
aa9bca05 WY |
1114 | kfree(tree); |
1115 | zswap_trees[type] = NULL; | |
2b281117 SJ |
1116 | } |
1117 | ||
2b281117 SJ |
1118 | static void zswap_frontswap_init(unsigned type) |
1119 | { | |
1120 | struct zswap_tree *tree; | |
1121 | ||
1122 | tree = kzalloc(sizeof(struct zswap_tree), GFP_KERNEL); | |
60105e12 MK |
1123 | if (!tree) { |
1124 | pr_err("alloc failed, zswap disabled for swap type %d\n", type); | |
1125 | return; | |
1126 | } | |
1127 | ||
2b281117 SJ |
1128 | tree->rbroot = RB_ROOT; |
1129 | spin_lock_init(&tree->lock); | |
1130 | zswap_trees[type] = tree; | |
2b281117 SJ |
1131 | } |
1132 | ||
1133 | static struct frontswap_ops zswap_frontswap_ops = { | |
1134 | .store = zswap_frontswap_store, | |
1135 | .load = zswap_frontswap_load, | |
1136 | .invalidate_page = zswap_frontswap_invalidate_page, | |
1137 | .invalidate_area = zswap_frontswap_invalidate_area, | |
1138 | .init = zswap_frontswap_init | |
1139 | }; | |
1140 | ||
1141 | /********************************* | |
1142 | * debugfs functions | |
1143 | **********************************/ | |
1144 | #ifdef CONFIG_DEBUG_FS | |
1145 | #include <linux/debugfs.h> | |
1146 | ||
1147 | static struct dentry *zswap_debugfs_root; | |
1148 | ||
1149 | static int __init zswap_debugfs_init(void) | |
1150 | { | |
1151 | if (!debugfs_initialized()) | |
1152 | return -ENODEV; | |
1153 | ||
1154 | zswap_debugfs_root = debugfs_create_dir("zswap", NULL); | |
1155 | if (!zswap_debugfs_root) | |
1156 | return -ENOMEM; | |
1157 | ||
1158 | debugfs_create_u64("pool_limit_hit", S_IRUGO, | |
1159 | zswap_debugfs_root, &zswap_pool_limit_hit); | |
1160 | debugfs_create_u64("reject_reclaim_fail", S_IRUGO, | |
1161 | zswap_debugfs_root, &zswap_reject_reclaim_fail); | |
1162 | debugfs_create_u64("reject_alloc_fail", S_IRUGO, | |
1163 | zswap_debugfs_root, &zswap_reject_alloc_fail); | |
1164 | debugfs_create_u64("reject_kmemcache_fail", S_IRUGO, | |
1165 | zswap_debugfs_root, &zswap_reject_kmemcache_fail); | |
1166 | debugfs_create_u64("reject_compress_poor", S_IRUGO, | |
1167 | zswap_debugfs_root, &zswap_reject_compress_poor); | |
1168 | debugfs_create_u64("written_back_pages", S_IRUGO, | |
1169 | zswap_debugfs_root, &zswap_written_back_pages); | |
1170 | debugfs_create_u64("duplicate_entry", S_IRUGO, | |
1171 | zswap_debugfs_root, &zswap_duplicate_entry); | |
12d79d64 DS |
1172 | debugfs_create_u64("pool_total_size", S_IRUGO, |
1173 | zswap_debugfs_root, &zswap_pool_total_size); | |
2b281117 SJ |
1174 | debugfs_create_atomic_t("stored_pages", S_IRUGO, |
1175 | zswap_debugfs_root, &zswap_stored_pages); | |
1176 | ||
1177 | return 0; | |
1178 | } | |
1179 | ||
1180 | static void __exit zswap_debugfs_exit(void) | |
1181 | { | |
1182 | debugfs_remove_recursive(zswap_debugfs_root); | |
1183 | } | |
1184 | #else | |
1185 | static int __init zswap_debugfs_init(void) | |
1186 | { | |
1187 | return 0; | |
1188 | } | |
1189 | ||
1190 | static void __exit zswap_debugfs_exit(void) { } | |
1191 | #endif | |
1192 | ||
1193 | /********************************* | |
1194 | * module init and exit | |
1195 | **********************************/ | |
1196 | static int __init init_zswap(void) | |
1197 | { | |
f1c54846 | 1198 | struct zswap_pool *pool; |
ad7ed770 | 1199 | int ret; |
60105e12 | 1200 | |
90b0fc26 DS |
1201 | zswap_init_started = true; |
1202 | ||
2b281117 SJ |
1203 | if (zswap_entry_cache_create()) { |
1204 | pr_err("entry cache creation failed\n"); | |
f1c54846 | 1205 | goto cache_fail; |
2b281117 | 1206 | } |
f1c54846 | 1207 | |
ad7ed770 SAS |
1208 | ret = cpuhp_setup_state(CPUHP_MM_ZSWP_MEM_PREPARE, "mm/zswap:prepare", |
1209 | zswap_dstmem_prepare, zswap_dstmem_dead); | |
1210 | if (ret) { | |
f1c54846 DS |
1211 | pr_err("dstmem alloc failed\n"); |
1212 | goto dstmem_fail; | |
2b281117 | 1213 | } |
f1c54846 DS |
1214 | |
1215 | pool = __zswap_pool_create_fallback(); | |
1216 | if (!pool) { | |
1217 | pr_err("pool creation failed\n"); | |
1218 | goto pool_fail; | |
2b281117 | 1219 | } |
f1c54846 DS |
1220 | pr_info("loaded using pool %s/%s\n", pool->tfm_name, |
1221 | zpool_get_type(pool->zpool)); | |
1222 | ||
1223 | list_add(&pool->list, &zswap_pools); | |
60105e12 | 1224 | |
2b281117 SJ |
1225 | frontswap_register_ops(&zswap_frontswap_ops); |
1226 | if (zswap_debugfs_init()) | |
1227 | pr_warn("debugfs initialization failed\n"); | |
1228 | return 0; | |
f1c54846 DS |
1229 | |
1230 | pool_fail: | |
ad7ed770 | 1231 | cpuhp_remove_state(CPUHP_MM_ZSWP_MEM_PREPARE); |
f1c54846 | 1232 | dstmem_fail: |
c119239b | 1233 | zswap_entry_cache_destroy(); |
f1c54846 | 1234 | cache_fail: |
2b281117 SJ |
1235 | return -ENOMEM; |
1236 | } | |
1237 | /* must be late so crypto has time to come up */ | |
1238 | late_initcall(init_zswap); | |
1239 | ||
1240 | MODULE_LICENSE("GPL"); | |
68386da8 | 1241 | MODULE_AUTHOR("Seth Jennings <sjennings@variantweb.net>"); |
2b281117 | 1242 | MODULE_DESCRIPTION("Compressed cache for swap pages"); |