Commit | Line | Data |
---|---|---|
3c2a0909 S |
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> | |
37 | #include <linux/zpool.h> | |
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 | **********************************/ | |
48 | /* Total bytes used by the compressed storage */ | |
49 | static u64 zswap_pool_total_size; | |
50 | /* Number of memory pages used by the compressed pool */ | |
51 | u64 zswap_pool_pages; | |
52 | /* The number of compressed pages currently stored in zswap */ | |
53 | atomic_t zswap_stored_pages = ATOMIC_INIT(0); | |
54 | ||
55 | /* | |
56 | * The statistics below are not protected from concurrent access for | |
57 | * performance reasons so they may not be a 100% accurate. However, | |
58 | * they do provide useful information on roughly how many times a | |
59 | * certain event is occurring. | |
60 | */ | |
61 | ||
62 | /* Pool limit was hit (see zswap_max_pool_percent) */ | |
63 | static u64 zswap_pool_limit_hit; | |
64 | /* Pages written back when pool limit was reached */ | |
65 | static u64 zswap_written_back_pages; | |
66 | /* Store failed due to a reclaim failure after pool limit was reached */ | |
67 | static u64 zswap_reject_reclaim_fail; | |
68 | /* Compressed page was too big for the allocator to (optimally) store */ | |
69 | static u64 zswap_reject_compress_poor; | |
70 | /* Store failed because underlying allocator could not get memory */ | |
71 | static u64 zswap_reject_alloc_fail; | |
72 | /* Store failed because the entry metadata could not be allocated (rare) */ | |
73 | static u64 zswap_reject_kmemcache_fail; | |
74 | /* Duplicate store was encountered (rare) */ | |
75 | static u64 zswap_duplicate_entry; | |
76 | ||
77 | /* The number of zero pages currently stored in zswap */ | |
78 | static atomic_t zswap_zero_pages = ATOMIC_INIT(0); | |
79 | ||
80 | /********************************* | |
81 | * tunables | |
82 | **********************************/ | |
83 | /* Enable/disable zswap (disabled by default, fixed at boot for now) */ | |
84 | static bool zswap_enabled __read_mostly = 1; | |
85 | module_param_named(enabled, zswap_enabled, bool, 0444); | |
86 | ||
87 | /* Compressor to be used by zswap (fixed at boot for now) */ | |
88 | #define ZSWAP_COMPRESSOR_DEFAULT "lzo" | |
89 | static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT; | |
90 | module_param_named(compressor, zswap_compressor, charp, 0444); | |
91 | ||
92 | /* The maximum percentage of memory that the compressed pool can occupy */ | |
93 | static unsigned int zswap_max_pool_percent = 50; | |
94 | module_param_named(max_pool_percent, | |
95 | zswap_max_pool_percent, uint, 0644); | |
96 | ||
97 | /* Compressed storage to use */ | |
98 | #define ZSWAP_ZPOOL_DEFAULT "zsmalloc" | |
99 | static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT; | |
100 | module_param_named(zpool, zswap_zpool_type, charp, 0444); | |
101 | ||
102 | /* zpool is shared by all of zswap backend */ | |
103 | static struct zpool *zswap_pool; | |
104 | ||
105 | /********************************* | |
106 | * compression functions | |
107 | **********************************/ | |
108 | /* per-cpu compression transforms */ | |
109 | static struct crypto_comp * __percpu *zswap_comp_pcpu_tfms; | |
110 | ||
111 | enum comp_op { | |
112 | ZSWAP_COMPOP_COMPRESS, | |
113 | ZSWAP_COMPOP_DECOMPRESS | |
114 | }; | |
115 | ||
116 | static int zswap_comp_op(enum comp_op op, const u8 *src, unsigned int slen, | |
117 | u8 *dst, unsigned int *dlen) | |
118 | { | |
119 | struct crypto_comp *tfm; | |
120 | int ret; | |
121 | ||
122 | tfm = *per_cpu_ptr(zswap_comp_pcpu_tfms, get_cpu()); | |
123 | switch (op) { | |
124 | case ZSWAP_COMPOP_COMPRESS: | |
125 | ret = crypto_comp_compress(tfm, src, slen, dst, dlen); | |
126 | break; | |
127 | case ZSWAP_COMPOP_DECOMPRESS: | |
128 | ret = crypto_comp_decompress(tfm, src, slen, dst, dlen); | |
129 | break; | |
130 | default: | |
131 | ret = -EINVAL; | |
132 | } | |
133 | ||
134 | put_cpu(); | |
135 | return ret; | |
136 | } | |
137 | ||
138 | static int __init zswap_comp_init(void) | |
139 | { | |
140 | if (!crypto_has_comp(zswap_compressor, 0, 0)) { | |
141 | pr_info("%s compressor not available\n", zswap_compressor); | |
142 | /* fall back to default compressor */ | |
143 | zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT; | |
144 | if (!crypto_has_comp(zswap_compressor, 0, 0)) | |
145 | /* can't even load the default compressor */ | |
146 | return -ENODEV; | |
147 | } | |
148 | pr_info("using %s compressor\n", zswap_compressor); | |
149 | ||
150 | /* alloc percpu transforms */ | |
151 | zswap_comp_pcpu_tfms = alloc_percpu(struct crypto_comp *); | |
152 | if (!zswap_comp_pcpu_tfms) | |
153 | return -ENOMEM; | |
154 | return 0; | |
155 | } | |
156 | ||
157 | static void zswap_comp_exit(void) | |
158 | { | |
159 | /* free percpu transforms */ | |
160 | if (zswap_comp_pcpu_tfms) | |
161 | free_percpu(zswap_comp_pcpu_tfms); | |
162 | } | |
163 | ||
164 | /********************************* | |
165 | * data structures | |
166 | **********************************/ | |
167 | /* | |
168 | * struct zswap_entry | |
169 | * | |
170 | * This structure contains the metadata for tracking a single compressed | |
171 | * page within zswap. | |
172 | * | |
173 | * rbnode - links the entry into red-black tree for the appropriate swap type | |
174 | * refcount - the number of outstanding reference to the entry. This is needed | |
175 | * to protect against premature freeing of the entry by code | |
176 | * concurrent calls to load, invalidate, and writeback. The lock | |
177 | * for the zswap_tree structure that contains the entry must | |
178 | * be held while changing the refcount. Since the lock must | |
179 | * be held, there is no reason to also make refcount atomic. | |
180 | * offset - the swap offset for the entry. Index into the red-black tree. | |
181 | * handle - zpool allocation handle that stores the compressed page data | |
182 | * length - the length in bytes of the compressed page data. Needed during | |
183 | * decompression | |
184 | * zero_flag - the flag indicating the page for the zswap_entry is a zero page. | |
185 | * zswap does not store the page during compression. | |
186 | * It memsets the page with 0 during decompression. | |
187 | */ | |
188 | struct zswap_entry { | |
189 | struct rb_node rbnode; | |
190 | pgoff_t offset; | |
191 | int refcount; | |
192 | unsigned int length; | |
193 | unsigned long handle; | |
194 | unsigned char zero_flag; | |
195 | }; | |
196 | ||
197 | struct zswap_header { | |
198 | swp_entry_t swpentry; | |
199 | }; | |
200 | ||
201 | /* | |
202 | * The tree lock in the zswap_tree struct protects a few things: | |
203 | * - the rbtree | |
204 | * - the refcount field of each entry in the tree | |
205 | */ | |
206 | struct zswap_tree { | |
207 | struct rb_root rbroot; | |
208 | spinlock_t lock; | |
209 | }; | |
210 | ||
211 | static struct zswap_tree *zswap_trees[MAX_SWAPFILES]; | |
212 | ||
213 | /********************************* | |
214 | * zswap entry functions | |
215 | **********************************/ | |
216 | static struct kmem_cache *zswap_entry_cache; | |
217 | ||
218 | static int zswap_entry_cache_create(void) | |
219 | { | |
220 | zswap_entry_cache = KMEM_CACHE(zswap_entry, 0); | |
221 | return zswap_entry_cache == NULL; | |
222 | } | |
223 | ||
224 | static void __init zswap_entry_cache_destroy(void) | |
225 | { | |
226 | kmem_cache_destroy(zswap_entry_cache); | |
227 | } | |
228 | ||
229 | static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp) | |
230 | { | |
231 | struct zswap_entry *entry; | |
232 | entry = kmem_cache_alloc(zswap_entry_cache, gfp); | |
233 | if (!entry) | |
234 | return NULL; | |
235 | entry->refcount = 1; | |
236 | entry->zero_flag = 0; | |
237 | RB_CLEAR_NODE(&entry->rbnode); | |
238 | return entry; | |
239 | } | |
240 | ||
241 | static void zswap_entry_cache_free(struct zswap_entry *entry) | |
242 | { | |
243 | kmem_cache_free(zswap_entry_cache, entry); | |
244 | } | |
245 | ||
246 | /********************************* | |
247 | * rbtree functions | |
248 | **********************************/ | |
249 | static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset) | |
250 | { | |
251 | struct rb_node *node = root->rb_node; | |
252 | struct zswap_entry *entry; | |
253 | ||
254 | while (node) { | |
255 | entry = rb_entry(node, struct zswap_entry, rbnode); | |
256 | if (entry->offset > offset) | |
257 | node = node->rb_left; | |
258 | else if (entry->offset < offset) | |
259 | node = node->rb_right; | |
260 | else | |
261 | return entry; | |
262 | } | |
263 | return NULL; | |
264 | } | |
265 | ||
266 | /* | |
267 | * In the case that a entry with the same offset is found, a pointer to | |
268 | * the existing entry is stored in dupentry and the function returns -EEXIST | |
269 | */ | |
270 | static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry, | |
271 | struct zswap_entry **dupentry) | |
272 | { | |
273 | struct rb_node **link = &root->rb_node, *parent = NULL; | |
274 | struct zswap_entry *myentry; | |
275 | ||
276 | while (*link) { | |
277 | parent = *link; | |
278 | myentry = rb_entry(parent, struct zswap_entry, rbnode); | |
279 | if (myentry->offset > entry->offset) | |
280 | link = &(*link)->rb_left; | |
281 | else if (myentry->offset < entry->offset) | |
282 | link = &(*link)->rb_right; | |
283 | else { | |
284 | *dupentry = myentry; | |
285 | return -EEXIST; | |
286 | } | |
287 | } | |
288 | rb_link_node(&entry->rbnode, parent, link); | |
289 | rb_insert_color(&entry->rbnode, root); | |
290 | return 0; | |
291 | } | |
292 | ||
293 | static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry) | |
294 | { | |
295 | if (!RB_EMPTY_NODE(&entry->rbnode)) { | |
296 | rb_erase(&entry->rbnode, root); | |
297 | RB_CLEAR_NODE(&entry->rbnode); | |
298 | } | |
299 | } | |
300 | ||
301 | /* | |
302 | * Carries out the common pattern of freeing and entry's zpool allocation, | |
303 | * freeing the entry itself, and decrementing the number of stored pages. | |
304 | */ | |
305 | static void zswap_free_entry(struct zswap_entry *entry) | |
306 | { | |
307 | if (entry->zero_flag == 1) { | |
308 | atomic_dec(&zswap_zero_pages); | |
309 | goto zeropage_out; | |
310 | } | |
311 | zpool_free(zswap_pool, entry->handle); | |
312 | zeropage_out: | |
313 | zswap_entry_cache_free(entry); | |
314 | atomic_dec(&zswap_stored_pages); | |
315 | zswap_pool_total_size = zpool_get_total_size(zswap_pool); | |
316 | zswap_pool_pages = zswap_pool_total_size >> PAGE_SHIFT; | |
317 | } | |
318 | ||
319 | /* caller must hold the tree lock */ | |
320 | static void zswap_entry_get(struct zswap_entry *entry) | |
321 | { | |
322 | entry->refcount++; | |
323 | } | |
324 | ||
325 | /* caller must hold the tree lock | |
326 | * remove from the tree and free it, if nobody reference the entry | |
327 | */ | |
328 | static void zswap_entry_put(struct zswap_tree *tree, | |
329 | struct zswap_entry *entry) | |
330 | { | |
331 | int refcount = --entry->refcount; | |
332 | ||
333 | BUG_ON(refcount < 0); | |
334 | if (refcount == 0) { | |
335 | zswap_rb_erase(&tree->rbroot, entry); | |
336 | zswap_free_entry(entry); | |
337 | } | |
338 | } | |
339 | ||
340 | /* caller must hold the tree lock */ | |
341 | static struct zswap_entry *zswap_entry_find_get(struct rb_root *root, | |
342 | pgoff_t offset) | |
343 | { | |
344 | struct zswap_entry *entry = NULL; | |
345 | ||
346 | entry = zswap_rb_search(root, offset); | |
347 | if (entry) | |
348 | zswap_entry_get(entry); | |
349 | ||
350 | return entry; | |
351 | } | |
352 | ||
353 | /********************************* | |
354 | * per-cpu code | |
355 | **********************************/ | |
356 | static DEFINE_PER_CPU(u8 *, zswap_dstmem); | |
357 | ||
358 | static int __zswap_cpu_notifier(unsigned long action, unsigned long cpu) | |
359 | { | |
360 | struct crypto_comp *tfm; | |
361 | u8 *dst; | |
362 | ||
363 | switch (action) { | |
364 | case CPU_UP_PREPARE: | |
365 | tfm = crypto_alloc_comp(zswap_compressor, 0, 0); | |
366 | if (IS_ERR(tfm)) { | |
367 | pr_err("can't allocate compressor transform\n"); | |
368 | return NOTIFY_BAD; | |
369 | } | |
370 | *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = tfm; | |
371 | dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu)); | |
372 | if (!dst) { | |
373 | pr_err("can't allocate compressor buffer\n"); | |
374 | crypto_free_comp(tfm); | |
375 | *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = NULL; | |
376 | return NOTIFY_BAD; | |
377 | } | |
378 | per_cpu(zswap_dstmem, cpu) = dst; | |
379 | break; | |
380 | case CPU_DEAD: | |
381 | case CPU_UP_CANCELED: | |
382 | tfm = *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu); | |
383 | if (tfm) { | |
384 | crypto_free_comp(tfm); | |
385 | *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = NULL; | |
386 | } | |
387 | dst = per_cpu(zswap_dstmem, cpu); | |
388 | kfree(dst); | |
389 | per_cpu(zswap_dstmem, cpu) = NULL; | |
390 | break; | |
391 | default: | |
392 | break; | |
393 | } | |
394 | return NOTIFY_OK; | |
395 | } | |
396 | ||
397 | static int zswap_cpu_notifier(struct notifier_block *nb, | |
398 | unsigned long action, void *pcpu) | |
399 | { | |
400 | unsigned long cpu = (unsigned long)pcpu; | |
401 | return __zswap_cpu_notifier(action, cpu); | |
402 | } | |
403 | ||
404 | static struct notifier_block zswap_cpu_notifier_block = { | |
405 | .notifier_call = zswap_cpu_notifier | |
406 | }; | |
407 | ||
408 | static int zswap_cpu_init(void) | |
409 | { | |
410 | unsigned long cpu; | |
411 | ||
412 | cpu_notifier_register_begin(); | |
413 | for_each_online_cpu(cpu) | |
414 | if (__zswap_cpu_notifier(CPU_UP_PREPARE, cpu) != NOTIFY_OK) | |
415 | goto cleanup; | |
416 | __register_cpu_notifier(&zswap_cpu_notifier_block); | |
417 | cpu_notifier_register_done(); | |
418 | return 0; | |
419 | ||
420 | cleanup: | |
421 | for_each_online_cpu(cpu) | |
422 | __zswap_cpu_notifier(CPU_UP_CANCELED, cpu); | |
423 | cpu_notifier_register_done(); | |
424 | return -ENOMEM; | |
425 | } | |
426 | ||
427 | /********************************* | |
428 | * helpers | |
429 | **********************************/ | |
430 | static bool zswap_is_full(void) | |
431 | { | |
432 | return totalram_pages * zswap_max_pool_percent / 100 < | |
433 | DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE); | |
434 | } | |
435 | ||
436 | /********************************* | |
437 | * writeback code | |
438 | **********************************/ | |
439 | /* return enum for zswap_get_swap_cache_page */ | |
440 | enum zswap_get_swap_ret { | |
441 | ZSWAP_SWAPCACHE_NEW, | |
442 | ZSWAP_SWAPCACHE_EXIST, | |
443 | ZSWAP_SWAPCACHE_FAIL, | |
444 | }; | |
445 | ||
446 | /* | |
447 | * zswap_get_swap_cache_page | |
448 | * | |
449 | * This is an adaption of read_swap_cache_async() | |
450 | * | |
451 | * This function tries to find a page with the given swap entry | |
452 | * in the swapper_space address space (the swap cache). If the page | |
453 | * is found, it is returned in retpage. Otherwise, a page is allocated, | |
454 | * added to the swap cache, and returned in retpage. | |
455 | * | |
456 | * If success, the swap cache page is returned in retpage | |
457 | * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache | |
458 | * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated, | |
459 | * the new page is added to swapcache and locked | |
460 | * Returns ZSWAP_SWAPCACHE_FAIL on error | |
461 | */ | |
462 | static int zswap_get_swap_cache_page(swp_entry_t entry, | |
463 | struct page **retpage) | |
464 | { | |
465 | struct page *found_page, *new_page = NULL; | |
466 | struct address_space *swapper_space = swap_address_space(entry); | |
467 | int err; | |
468 | ||
469 | *retpage = NULL; | |
470 | do { | |
471 | /* | |
472 | * First check the swap cache. Since this is normally | |
473 | * called after lookup_swap_cache() failed, re-calling | |
474 | * that would confuse statistics. | |
475 | */ | |
476 | found_page = find_get_page(swapper_space, entry.val); | |
477 | if (found_page) | |
478 | break; | |
479 | ||
480 | /* | |
481 | * Get a new page to read into from swap. | |
482 | */ | |
483 | if (!new_page) { | |
484 | new_page = alloc_page(GFP_KERNEL); | |
485 | if (!new_page) | |
486 | break; /* Out of memory */ | |
487 | } | |
488 | ||
489 | /* | |
490 | * call radix_tree_preload() while we can wait. | |
491 | */ | |
492 | err = radix_tree_preload(GFP_KERNEL); | |
493 | if (err) | |
494 | break; | |
495 | ||
496 | /* | |
497 | * Swap entry may have been freed since our caller observed it. | |
498 | */ | |
499 | err = swapcache_prepare(entry); | |
500 | if (err == -EEXIST) { /* seems racy */ | |
501 | radix_tree_preload_end(); | |
502 | continue; | |
503 | } | |
504 | if (err) { /* swp entry is obsolete ? */ | |
505 | radix_tree_preload_end(); | |
506 | break; | |
507 | } | |
508 | ||
509 | /* May fail (-ENOMEM) if radix-tree node allocation failed. */ | |
510 | __set_page_locked(new_page); | |
511 | SetPageSwapBacked(new_page); | |
512 | err = __add_to_swap_cache(new_page, entry); | |
513 | if (likely(!err)) { | |
514 | radix_tree_preload_end(); | |
515 | lru_cache_add_anon(new_page); | |
516 | *retpage = new_page; | |
517 | return ZSWAP_SWAPCACHE_NEW; | |
518 | } | |
519 | radix_tree_preload_end(); | |
520 | ClearPageSwapBacked(new_page); | |
521 | __clear_page_locked(new_page); | |
522 | /* | |
523 | * add_to_swap_cache() doesn't return -EEXIST, so we can safely | |
524 | * clear SWAP_HAS_CACHE flag. | |
525 | */ | |
526 | swapcache_free(entry, NULL); | |
527 | } while (err != -ENOMEM); | |
528 | ||
529 | if (new_page) | |
530 | page_cache_release(new_page); | |
531 | if (!found_page) | |
532 | return ZSWAP_SWAPCACHE_FAIL; | |
533 | *retpage = found_page; | |
534 | return ZSWAP_SWAPCACHE_EXIST; | |
535 | } | |
536 | ||
537 | /* | |
538 | * Attempts to free an entry by adding a page to the swap cache, | |
539 | * decompressing the entry data into the page, and issuing a | |
540 | * bio write to write the page back to the swap device. | |
541 | * | |
542 | * This can be thought of as a "resumed writeback" of the page | |
543 | * to the swap device. We are basically resuming the same swap | |
544 | * writeback path that was intercepted with the frontswap_store() | |
545 | * in the first place. After the page has been decompressed into | |
546 | * the swap cache, the compressed version stored by zswap can be | |
547 | * freed. | |
548 | */ | |
549 | static int zswap_writeback_entry(struct zpool *pool, unsigned long handle) | |
550 | { | |
551 | struct zswap_header *zhdr; | |
552 | swp_entry_t swpentry; | |
553 | struct zswap_tree *tree; | |
554 | pgoff_t offset; | |
555 | struct zswap_entry *entry; | |
556 | struct page *page; | |
557 | u8 *src, *dst; | |
558 | unsigned int dlen; | |
559 | int ret; | |
560 | struct writeback_control wbc = { | |
561 | .sync_mode = WB_SYNC_NONE, | |
562 | }; | |
563 | ||
564 | /* extract swpentry from data */ | |
565 | zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO); | |
566 | swpentry = zhdr->swpentry; /* here */ | |
567 | zpool_unmap_handle(pool, handle); | |
568 | tree = zswap_trees[swp_type(swpentry)]; | |
569 | offset = swp_offset(swpentry); | |
570 | ||
571 | /* find and ref zswap entry */ | |
572 | spin_lock(&tree->lock); | |
573 | entry = zswap_entry_find_get(&tree->rbroot, offset); | |
574 | if (!entry) { | |
575 | /* entry was invalidated */ | |
576 | spin_unlock(&tree->lock); | |
577 | return 0; | |
578 | } | |
579 | spin_unlock(&tree->lock); | |
580 | BUG_ON(offset != entry->offset); | |
581 | ||
582 | /* try to allocate swap cache page */ | |
583 | switch (zswap_get_swap_cache_page(swpentry, &page)) { | |
584 | case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */ | |
585 | ret = -ENOMEM; | |
586 | goto fail; | |
587 | ||
588 | case ZSWAP_SWAPCACHE_EXIST: | |
589 | /* page is already in the swap cache, ignore for now */ | |
590 | page_cache_release(page); | |
591 | ret = -EEXIST; | |
592 | goto fail; | |
593 | ||
594 | case ZSWAP_SWAPCACHE_NEW: /* page is locked */ | |
595 | /* decompress */ | |
596 | dlen = PAGE_SIZE; | |
597 | src = (u8 *)zpool_map_handle(zswap_pool, entry->handle, | |
598 | ZPOOL_MM_RO) + sizeof(struct zswap_header); | |
599 | dst = kmap_atomic(page); | |
600 | ret = zswap_comp_op(ZSWAP_COMPOP_DECOMPRESS, src, | |
601 | entry->length, dst, &dlen); | |
602 | kunmap_atomic(dst); | |
603 | zpool_unmap_handle(zswap_pool, entry->handle); | |
604 | BUG_ON(ret); | |
605 | BUG_ON(dlen != PAGE_SIZE); | |
606 | ||
607 | /* page is up to date */ | |
608 | SetPageUptodate(page); | |
609 | } | |
610 | ||
611 | /* move it to the tail of the inactive list after end_writeback */ | |
612 | SetPageReclaim(page); | |
613 | ||
614 | /* start writeback */ | |
615 | __swap_writepage(page, &wbc, end_swap_bio_write); | |
616 | page_cache_release(page); | |
617 | zswap_written_back_pages++; | |
618 | ||
619 | spin_lock(&tree->lock); | |
620 | /* drop local reference */ | |
621 | zswap_entry_put(tree, entry); | |
622 | ||
623 | /* | |
624 | * There are two possible situations for entry here: | |
625 | * (1) refcount is 1(normal case), entry is valid and on the tree | |
626 | * (2) refcount is 0, entry is freed and not on the tree | |
627 | * because invalidate happened during writeback | |
628 | * search the tree and free the entry if find entry | |
629 | */ | |
630 | if (entry == zswap_rb_search(&tree->rbroot, offset)) | |
631 | zswap_entry_put(tree, entry); | |
632 | spin_unlock(&tree->lock); | |
633 | ||
634 | goto end; | |
635 | ||
636 | /* | |
637 | * if we get here due to ZSWAP_SWAPCACHE_EXIST | |
638 | * a load may happening concurrently | |
639 | * it is safe and okay to not free the entry | |
640 | * if we free the entry in the following put | |
641 | * it it either okay to return !0 | |
642 | */ | |
643 | fail: | |
644 | spin_lock(&tree->lock); | |
645 | zswap_entry_put(tree, entry); | |
646 | spin_unlock(&tree->lock); | |
647 | ||
648 | end: | |
649 | return ret; | |
650 | } | |
651 | ||
652 | static int page_zero_filled(void *ptr) | |
653 | { | |
654 | unsigned int pos; | |
655 | unsigned long *page; | |
656 | ||
657 | page = (unsigned long *)ptr; | |
658 | ||
659 | for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) { | |
660 | if (page[pos]) | |
661 | return 0; | |
662 | } | |
663 | ||
664 | return 1; | |
665 | } | |
666 | ||
667 | /********************************* | |
668 | * frontswap hooks | |
669 | **********************************/ | |
670 | /* attempts to compress and store an single page */ | |
671 | static int zswap_frontswap_store(unsigned type, pgoff_t offset, | |
672 | struct page *page) | |
673 | { | |
674 | struct zswap_tree *tree = zswap_trees[type]; | |
675 | struct zswap_entry *entry, *dupentry; | |
676 | int ret; | |
677 | unsigned int dlen = PAGE_SIZE, len; | |
678 | unsigned long handle; | |
679 | char *buf; | |
680 | u8 *src, *dst; | |
681 | struct zswap_header *zhdr; | |
682 | ||
683 | if (!tree) { | |
684 | ret = -ENODEV; | |
685 | goto reject; | |
686 | } | |
687 | ||
688 | /* if this page got EIO on pageout before, give up immediately */ | |
689 | if (PageError(page)) { | |
690 | ret = -ENOMEM; | |
691 | goto reject; | |
692 | } | |
693 | ||
694 | /* reclaim space if needed */ | |
695 | if (zswap_is_full()) { | |
696 | zswap_pool_limit_hit++; | |
697 | if (zpool_shrink(zswap_pool, 1, NULL)) { | |
698 | zswap_reject_reclaim_fail++; | |
699 | ret = -ENOMEM; | |
700 | goto reject; | |
701 | } | |
702 | } | |
703 | ||
704 | /* allocate entry */ | |
705 | entry = zswap_entry_cache_alloc(GFP_KERNEL); | |
706 | if (!entry) { | |
707 | zswap_reject_kmemcache_fail++; | |
708 | ret = -ENOMEM; | |
709 | goto reject; | |
710 | } | |
711 | ||
712 | /* compress */ | |
713 | src = kmap_atomic(page); | |
714 | if (page_zero_filled(src)) { | |
715 | atomic_inc(&zswap_zero_pages); | |
716 | entry->zero_flag = 1; | |
717 | kunmap_atomic(src); | |
718 | ||
719 | handle = 0; | |
720 | dlen = PAGE_SIZE; | |
721 | goto zeropage_out; | |
722 | } | |
723 | dst = get_cpu_var(zswap_dstmem); | |
724 | ||
725 | ret = zswap_comp_op(ZSWAP_COMPOP_COMPRESS, src, PAGE_SIZE, dst, &dlen); | |
726 | kunmap_atomic(src); | |
727 | if (ret) { | |
728 | ret = -EINVAL; | |
729 | goto freepage; | |
730 | } | |
731 | ||
732 | /* store */ | |
733 | len = dlen + sizeof(struct zswap_header); | |
734 | ret = zpool_malloc(zswap_pool, len, __GFP_NORETRY | __GFP_NOWARN, | |
735 | &handle); | |
736 | if (ret == -ENOSPC) { | |
737 | zswap_reject_compress_poor++; | |
738 | goto freepage; | |
739 | } | |
740 | if (ret) { | |
741 | zswap_reject_alloc_fail++; | |
742 | goto freepage; | |
743 | } | |
744 | zhdr = zpool_map_handle(zswap_pool, handle, ZPOOL_MM_RW); | |
745 | zhdr->swpentry = swp_entry(type, offset); | |
746 | buf = (u8 *)(zhdr + 1); | |
747 | memcpy(buf, dst, dlen); | |
748 | zpool_unmap_handle(zswap_pool, handle); | |
749 | put_cpu_var(zswap_dstmem); | |
750 | ||
751 | zeropage_out: | |
752 | /* populate entry */ | |
753 | entry->offset = offset; | |
754 | entry->handle = handle; | |
755 | entry->length = dlen; | |
756 | ||
757 | /* map */ | |
758 | spin_lock(&tree->lock); | |
759 | do { | |
760 | ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry); | |
761 | if (ret == -EEXIST) { | |
762 | zswap_duplicate_entry++; | |
763 | /* remove from rbtree */ | |
764 | zswap_rb_erase(&tree->rbroot, dupentry); | |
765 | zswap_entry_put(tree, dupentry); | |
766 | } | |
767 | } while (ret == -EEXIST); | |
768 | spin_unlock(&tree->lock); | |
769 | ||
770 | /* update stats */ | |
771 | atomic_inc(&zswap_stored_pages); | |
772 | zswap_pool_total_size = zpool_get_total_size(zswap_pool); | |
773 | zswap_pool_pages = zswap_pool_total_size >> PAGE_SHIFT; | |
774 | ||
775 | return 0; | |
776 | ||
777 | freepage: | |
778 | put_cpu_var(zswap_dstmem); | |
779 | zswap_entry_cache_free(entry); | |
780 | reject: | |
781 | return ret; | |
782 | } | |
783 | ||
784 | /* | |
785 | * returns 0 if the page was successfully decompressed | |
786 | * return -1 on entry not found or error | |
787 | */ | |
788 | static int zswap_frontswap_load(unsigned type, pgoff_t offset, | |
789 | struct page *page) | |
790 | { | |
791 | struct zswap_tree *tree = zswap_trees[type]; | |
792 | struct zswap_entry *entry; | |
793 | u8 *src, *dst; | |
794 | unsigned int dlen; | |
795 | int ret; | |
796 | ||
797 | /* find */ | |
798 | spin_lock(&tree->lock); | |
799 | entry = zswap_entry_find_get(&tree->rbroot, offset); | |
800 | if (!entry) { | |
801 | /* entry was written back */ | |
802 | spin_unlock(&tree->lock); | |
803 | return -1; | |
804 | } | |
805 | spin_unlock(&tree->lock); | |
806 | ||
807 | if (entry->zero_flag == 1) { | |
808 | dst = kmap_atomic(page); | |
809 | memset(dst, 0, PAGE_SIZE); | |
810 | kunmap_atomic(dst); | |
811 | goto zeropage_out; | |
812 | } | |
813 | ||
814 | /* decompress */ | |
815 | dlen = PAGE_SIZE; | |
816 | src = (u8 *)zpool_map_handle(zswap_pool, entry->handle, | |
817 | ZPOOL_MM_RO) + sizeof(struct zswap_header); | |
818 | dst = kmap_atomic(page); | |
819 | ret = zswap_comp_op(ZSWAP_COMPOP_DECOMPRESS, src, entry->length, | |
820 | dst, &dlen); | |
821 | kunmap_atomic(dst); | |
822 | zpool_unmap_handle(zswap_pool, entry->handle); | |
823 | BUG_ON(ret); | |
824 | ||
825 | zeropage_out: | |
826 | spin_lock(&tree->lock); | |
827 | zswap_entry_put(tree, entry); | |
828 | spin_unlock(&tree->lock); | |
829 | ||
830 | return 0; | |
831 | } | |
832 | ||
833 | /* frees an entry in zswap */ | |
834 | static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset) | |
835 | { | |
836 | struct zswap_tree *tree = zswap_trees[type]; | |
837 | struct zswap_entry *entry; | |
838 | ||
839 | /* find */ | |
840 | spin_lock(&tree->lock); | |
841 | entry = zswap_rb_search(&tree->rbroot, offset); | |
842 | if (!entry) { | |
843 | /* entry was written back */ | |
844 | spin_unlock(&tree->lock); | |
845 | return; | |
846 | } | |
847 | ||
848 | /* remove from rbtree */ | |
849 | zswap_rb_erase(&tree->rbroot, entry); | |
850 | ||
851 | /* drop the initial reference from entry creation */ | |
852 | zswap_entry_put(tree, entry); | |
853 | ||
854 | spin_unlock(&tree->lock); | |
855 | } | |
856 | ||
857 | /* frees all zswap entries for the given swap type */ | |
858 | static void zswap_frontswap_invalidate_area(unsigned type) | |
859 | { | |
860 | struct zswap_tree *tree = zswap_trees[type]; | |
861 | struct zswap_entry *entry, *n; | |
862 | ||
863 | if (!tree) | |
864 | return; | |
865 | ||
866 | /* walk the tree and free everything */ | |
867 | spin_lock(&tree->lock); | |
868 | rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode) | |
869 | zswap_free_entry(entry); | |
870 | tree->rbroot = RB_ROOT; | |
871 | spin_unlock(&tree->lock); | |
872 | kfree(tree); | |
873 | zswap_trees[type] = NULL; | |
874 | } | |
875 | ||
876 | static struct zpool_ops zswap_zpool_ops = { | |
877 | .evict = zswap_writeback_entry | |
878 | }; | |
879 | ||
880 | static void zswap_frontswap_init(unsigned type) | |
881 | { | |
882 | struct zswap_tree *tree; | |
883 | ||
884 | tree = kzalloc(sizeof(struct zswap_tree), GFP_KERNEL); | |
885 | if (!tree) { | |
886 | pr_err("alloc failed, zswap disabled for swap type %d\n", type); | |
887 | return; | |
888 | } | |
889 | ||
890 | tree->rbroot = RB_ROOT; | |
891 | spin_lock_init(&tree->lock); | |
892 | zswap_trees[type] = tree; | |
893 | } | |
894 | ||
895 | static struct frontswap_ops zswap_frontswap_ops = { | |
896 | .store = zswap_frontswap_store, | |
897 | .load = zswap_frontswap_load, | |
898 | .invalidate_page = zswap_frontswap_invalidate_page, | |
899 | .invalidate_area = zswap_frontswap_invalidate_area, | |
900 | .init = zswap_frontswap_init | |
901 | }; | |
902 | ||
903 | /********************************* | |
904 | * debugfs functions | |
905 | **********************************/ | |
906 | #ifdef CONFIG_DEBUG_FS | |
907 | #include <linux/debugfs.h> | |
908 | ||
909 | static struct dentry *zswap_debugfs_root; | |
910 | ||
911 | static int __init zswap_debugfs_init(void) | |
912 | { | |
913 | if (!debugfs_initialized()) | |
914 | return -ENODEV; | |
915 | ||
916 | zswap_debugfs_root = debugfs_create_dir("zswap", NULL); | |
917 | if (!zswap_debugfs_root) | |
918 | return -ENOMEM; | |
919 | ||
920 | debugfs_create_u64("pool_limit_hit", S_IRUGO, | |
921 | zswap_debugfs_root, &zswap_pool_limit_hit); | |
922 | debugfs_create_u64("reject_reclaim_fail", S_IRUGO, | |
923 | zswap_debugfs_root, &zswap_reject_reclaim_fail); | |
924 | debugfs_create_u64("reject_alloc_fail", S_IRUGO, | |
925 | zswap_debugfs_root, &zswap_reject_alloc_fail); | |
926 | debugfs_create_u64("reject_kmemcache_fail", S_IRUGO, | |
927 | zswap_debugfs_root, &zswap_reject_kmemcache_fail); | |
928 | debugfs_create_u64("reject_compress_poor", S_IRUGO, | |
929 | zswap_debugfs_root, &zswap_reject_compress_poor); | |
930 | debugfs_create_u64("written_back_pages", S_IRUGO, | |
931 | zswap_debugfs_root, &zswap_written_back_pages); | |
932 | debugfs_create_u64("duplicate_entry", S_IRUGO, | |
933 | zswap_debugfs_root, &zswap_duplicate_entry); | |
934 | debugfs_create_u64("pool_total_size", S_IRUGO, | |
935 | zswap_debugfs_root, &zswap_pool_total_size); | |
936 | debugfs_create_u64("pool_pages", S_IRUGO, | |
937 | zswap_debugfs_root, &zswap_pool_pages); | |
938 | debugfs_create_atomic_t("stored_pages", S_IRUGO, | |
939 | zswap_debugfs_root, &zswap_stored_pages); | |
940 | debugfs_create_atomic_t("zero_pages", S_IRUGO, | |
941 | zswap_debugfs_root, &zswap_zero_pages); | |
942 | ||
943 | return 0; | |
944 | } | |
945 | ||
946 | static void __exit zswap_debugfs_exit(void) | |
947 | { | |
948 | debugfs_remove_recursive(zswap_debugfs_root); | |
949 | } | |
950 | #else | |
951 | static int __init zswap_debugfs_init(void) | |
952 | { | |
953 | return 0; | |
954 | } | |
955 | ||
956 | static void __exit zswap_debugfs_exit(void) { } | |
957 | #endif | |
958 | ||
959 | /********************************* | |
960 | * module init and exit | |
961 | **********************************/ | |
962 | static int __init init_zswap(void) | |
963 | { | |
964 | gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN | __GFP_HIGHMEM; | |
965 | ||
966 | if (!zswap_enabled) | |
967 | return 0; | |
968 | ||
969 | pr_info("loading zswap\n"); | |
970 | ||
971 | zswap_pool = zpool_create_pool(zswap_zpool_type, gfp, &zswap_zpool_ops); | |
972 | if (!zswap_pool && strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) { | |
973 | pr_info("%s zpool not available\n", zswap_zpool_type); | |
974 | zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT; | |
975 | zswap_pool = zpool_create_pool(zswap_zpool_type, gfp, | |
976 | &zswap_zpool_ops); | |
977 | } | |
978 | if (!zswap_pool) { | |
979 | pr_err("%s zpool not available\n", zswap_zpool_type); | |
980 | pr_err("zpool creation failed\n"); | |
981 | goto error; | |
982 | } | |
983 | pr_info("using %s pool\n", zswap_zpool_type); | |
984 | ||
985 | if (zswap_entry_cache_create()) { | |
986 | pr_err("entry cache creation failed\n"); | |
987 | goto cachefail; | |
988 | } | |
989 | if (zswap_comp_init()) { | |
990 | pr_err("compressor initialization failed\n"); | |
991 | goto compfail; | |
992 | } | |
993 | if (zswap_cpu_init()) { | |
994 | pr_err("per-cpu initialization failed\n"); | |
995 | goto pcpufail; | |
996 | } | |
997 | ||
998 | frontswap_register_ops(&zswap_frontswap_ops); | |
999 | if (zswap_debugfs_init()) | |
1000 | pr_warn("debugfs initialization failed\n"); | |
1001 | return 0; | |
1002 | pcpufail: | |
1003 | zswap_comp_exit(); | |
1004 | compfail: | |
1005 | zswap_entry_cache_destroy(); | |
1006 | cachefail: | |
1007 | zpool_destroy_pool(zswap_pool); | |
1008 | error: | |
1009 | return -ENOMEM; | |
1010 | } | |
1011 | /* must be late so crypto has time to come up */ | |
1012 | late_initcall(init_zswap); | |
1013 | ||
1014 | MODULE_LICENSE("GPL"); | |
1015 | MODULE_AUTHOR("Seth Jennings <sjenning@linux.vnet.ibm.com>"); | |
1016 | MODULE_DESCRIPTION("Compressed cache for swap pages"); |