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> | |
37 | #include <linux/zbud.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 | /* Number of memory pages used by the compressed pool */ | |
49 | static u64 zswap_pool_pages; | |
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 | **********************************/ | |
78 | /* Enable/disable zswap (disabled by default, fixed at boot for now) */ | |
79 | static bool zswap_enabled __read_mostly; | |
80 | module_param_named(enabled, zswap_enabled, bool, 0); | |
81 | ||
82 | /* Compressor to be used by zswap (fixed at boot for now) */ | |
83 | #define ZSWAP_COMPRESSOR_DEFAULT "lzo" | |
84 | static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT; | |
85 | module_param_named(compressor, zswap_compressor, charp, 0); | |
86 | ||
87 | /* The maximum percentage of memory that the compressed pool can occupy */ | |
88 | static unsigned int zswap_max_pool_percent = 20; | |
89 | module_param_named(max_pool_percent, | |
90 | zswap_max_pool_percent, uint, 0644); | |
91 | ||
92 | /********************************* | |
93 | * compression functions | |
94 | **********************************/ | |
95 | /* per-cpu compression transforms */ | |
96 | static struct crypto_comp * __percpu *zswap_comp_pcpu_tfms; | |
97 | ||
98 | enum comp_op { | |
99 | ZSWAP_COMPOP_COMPRESS, | |
100 | ZSWAP_COMPOP_DECOMPRESS | |
101 | }; | |
102 | ||
103 | static int zswap_comp_op(enum comp_op op, const u8 *src, unsigned int slen, | |
104 | u8 *dst, unsigned int *dlen) | |
105 | { | |
106 | struct crypto_comp *tfm; | |
107 | int ret; | |
108 | ||
109 | tfm = *per_cpu_ptr(zswap_comp_pcpu_tfms, get_cpu()); | |
110 | switch (op) { | |
111 | case ZSWAP_COMPOP_COMPRESS: | |
112 | ret = crypto_comp_compress(tfm, src, slen, dst, dlen); | |
113 | break; | |
114 | case ZSWAP_COMPOP_DECOMPRESS: | |
115 | ret = crypto_comp_decompress(tfm, src, slen, dst, dlen); | |
116 | break; | |
117 | default: | |
118 | ret = -EINVAL; | |
119 | } | |
120 | ||
121 | put_cpu(); | |
122 | return ret; | |
123 | } | |
124 | ||
125 | static int __init zswap_comp_init(void) | |
126 | { | |
127 | if (!crypto_has_comp(zswap_compressor, 0, 0)) { | |
128 | pr_info("%s compressor not available\n", zswap_compressor); | |
129 | /* fall back to default compressor */ | |
130 | zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT; | |
131 | if (!crypto_has_comp(zswap_compressor, 0, 0)) | |
132 | /* can't even load the default compressor */ | |
133 | return -ENODEV; | |
134 | } | |
135 | pr_info("using %s compressor\n", zswap_compressor); | |
136 | ||
137 | /* alloc percpu transforms */ | |
138 | zswap_comp_pcpu_tfms = alloc_percpu(struct crypto_comp *); | |
139 | if (!zswap_comp_pcpu_tfms) | |
140 | return -ENOMEM; | |
141 | return 0; | |
142 | } | |
143 | ||
144 | static void zswap_comp_exit(void) | |
145 | { | |
146 | /* free percpu transforms */ | |
147 | if (zswap_comp_pcpu_tfms) | |
148 | free_percpu(zswap_comp_pcpu_tfms); | |
149 | } | |
150 | ||
151 | /********************************* | |
152 | * data structures | |
153 | **********************************/ | |
154 | /* | |
155 | * struct zswap_entry | |
156 | * | |
157 | * This structure contains the metadata for tracking a single compressed | |
158 | * page within zswap. | |
159 | * | |
160 | * rbnode - links the entry into red-black tree for the appropriate swap type | |
161 | * refcount - the number of outstanding reference to the entry. This is needed | |
162 | * to protect against premature freeing of the entry by code | |
163 | * concurent calls to load, invalidate, and writeback. The lock | |
164 | * for the zswap_tree structure that contains the entry must | |
165 | * be held while changing the refcount. Since the lock must | |
166 | * be held, there is no reason to also make refcount atomic. | |
167 | * offset - the swap offset for the entry. Index into the red-black tree. | |
168 | * handle - zsmalloc allocation handle that stores the compressed page data | |
169 | * length - the length in bytes of the compressed page data. Needed during | |
170 | * decompression | |
171 | */ | |
172 | struct zswap_entry { | |
173 | struct rb_node rbnode; | |
174 | pgoff_t offset; | |
175 | int refcount; | |
176 | unsigned int length; | |
177 | unsigned long handle; | |
178 | }; | |
179 | ||
180 | struct zswap_header { | |
181 | swp_entry_t swpentry; | |
182 | }; | |
183 | ||
184 | /* | |
185 | * The tree lock in the zswap_tree struct protects a few things: | |
186 | * - the rbtree | |
187 | * - the refcount field of each entry in the tree | |
188 | */ | |
189 | struct zswap_tree { | |
190 | struct rb_root rbroot; | |
191 | spinlock_t lock; | |
192 | struct zbud_pool *pool; | |
193 | }; | |
194 | ||
195 | static struct zswap_tree *zswap_trees[MAX_SWAPFILES]; | |
196 | ||
197 | /********************************* | |
198 | * zswap entry functions | |
199 | **********************************/ | |
200 | static struct kmem_cache *zswap_entry_cache; | |
201 | ||
202 | static int zswap_entry_cache_create(void) | |
203 | { | |
204 | zswap_entry_cache = KMEM_CACHE(zswap_entry, 0); | |
205 | return (zswap_entry_cache == NULL); | |
206 | } | |
207 | ||
208 | static void zswap_entry_cache_destory(void) | |
209 | { | |
210 | kmem_cache_destroy(zswap_entry_cache); | |
211 | } | |
212 | ||
213 | static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp) | |
214 | { | |
215 | struct zswap_entry *entry; | |
216 | entry = kmem_cache_alloc(zswap_entry_cache, gfp); | |
217 | if (!entry) | |
218 | return NULL; | |
219 | entry->refcount = 1; | |
220 | return entry; | |
221 | } | |
222 | ||
223 | static void zswap_entry_cache_free(struct zswap_entry *entry) | |
224 | { | |
225 | kmem_cache_free(zswap_entry_cache, entry); | |
226 | } | |
227 | ||
228 | /* caller must hold the tree lock */ | |
229 | static void zswap_entry_get(struct zswap_entry *entry) | |
230 | { | |
231 | entry->refcount++; | |
232 | } | |
233 | ||
234 | /* caller must hold the tree lock */ | |
235 | static int zswap_entry_put(struct zswap_entry *entry) | |
236 | { | |
237 | entry->refcount--; | |
238 | return entry->refcount; | |
239 | } | |
240 | ||
241 | /********************************* | |
242 | * rbtree functions | |
243 | **********************************/ | |
244 | static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset) | |
245 | { | |
246 | struct rb_node *node = root->rb_node; | |
247 | struct zswap_entry *entry; | |
248 | ||
249 | while (node) { | |
250 | entry = rb_entry(node, struct zswap_entry, rbnode); | |
251 | if (entry->offset > offset) | |
252 | node = node->rb_left; | |
253 | else if (entry->offset < offset) | |
254 | node = node->rb_right; | |
255 | else | |
256 | return entry; | |
257 | } | |
258 | return NULL; | |
259 | } | |
260 | ||
261 | /* | |
262 | * In the case that a entry with the same offset is found, a pointer to | |
263 | * the existing entry is stored in dupentry and the function returns -EEXIST | |
264 | */ | |
265 | static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry, | |
266 | struct zswap_entry **dupentry) | |
267 | { | |
268 | struct rb_node **link = &root->rb_node, *parent = NULL; | |
269 | struct zswap_entry *myentry; | |
270 | ||
271 | while (*link) { | |
272 | parent = *link; | |
273 | myentry = rb_entry(parent, struct zswap_entry, rbnode); | |
274 | if (myentry->offset > entry->offset) | |
275 | link = &(*link)->rb_left; | |
276 | else if (myentry->offset < entry->offset) | |
277 | link = &(*link)->rb_right; | |
278 | else { | |
279 | *dupentry = myentry; | |
280 | return -EEXIST; | |
281 | } | |
282 | } | |
283 | rb_link_node(&entry->rbnode, parent, link); | |
284 | rb_insert_color(&entry->rbnode, root); | |
285 | return 0; | |
286 | } | |
287 | ||
288 | /********************************* | |
289 | * per-cpu code | |
290 | **********************************/ | |
291 | static DEFINE_PER_CPU(u8 *, zswap_dstmem); | |
292 | ||
293 | static int __zswap_cpu_notifier(unsigned long action, unsigned long cpu) | |
294 | { | |
295 | struct crypto_comp *tfm; | |
296 | u8 *dst; | |
297 | ||
298 | switch (action) { | |
299 | case CPU_UP_PREPARE: | |
300 | tfm = crypto_alloc_comp(zswap_compressor, 0, 0); | |
301 | if (IS_ERR(tfm)) { | |
302 | pr_err("can't allocate compressor transform\n"); | |
303 | return NOTIFY_BAD; | |
304 | } | |
305 | *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = tfm; | |
306 | dst = kmalloc(PAGE_SIZE * 2, GFP_KERNEL); | |
307 | if (!dst) { | |
308 | pr_err("can't allocate compressor buffer\n"); | |
309 | crypto_free_comp(tfm); | |
310 | *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = NULL; | |
311 | return NOTIFY_BAD; | |
312 | } | |
313 | per_cpu(zswap_dstmem, cpu) = dst; | |
314 | break; | |
315 | case CPU_DEAD: | |
316 | case CPU_UP_CANCELED: | |
317 | tfm = *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu); | |
318 | if (tfm) { | |
319 | crypto_free_comp(tfm); | |
320 | *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = NULL; | |
321 | } | |
322 | dst = per_cpu(zswap_dstmem, cpu); | |
323 | kfree(dst); | |
324 | per_cpu(zswap_dstmem, cpu) = NULL; | |
325 | break; | |
326 | default: | |
327 | break; | |
328 | } | |
329 | return NOTIFY_OK; | |
330 | } | |
331 | ||
332 | static int zswap_cpu_notifier(struct notifier_block *nb, | |
333 | unsigned long action, void *pcpu) | |
334 | { | |
335 | unsigned long cpu = (unsigned long)pcpu; | |
336 | return __zswap_cpu_notifier(action, cpu); | |
337 | } | |
338 | ||
339 | static struct notifier_block zswap_cpu_notifier_block = { | |
340 | .notifier_call = zswap_cpu_notifier | |
341 | }; | |
342 | ||
343 | static int zswap_cpu_init(void) | |
344 | { | |
345 | unsigned long cpu; | |
346 | ||
347 | get_online_cpus(); | |
348 | for_each_online_cpu(cpu) | |
349 | if (__zswap_cpu_notifier(CPU_UP_PREPARE, cpu) != NOTIFY_OK) | |
350 | goto cleanup; | |
351 | register_cpu_notifier(&zswap_cpu_notifier_block); | |
352 | put_online_cpus(); | |
353 | return 0; | |
354 | ||
355 | cleanup: | |
356 | for_each_online_cpu(cpu) | |
357 | __zswap_cpu_notifier(CPU_UP_CANCELED, cpu); | |
358 | put_online_cpus(); | |
359 | return -ENOMEM; | |
360 | } | |
361 | ||
362 | /********************************* | |
363 | * helpers | |
364 | **********************************/ | |
365 | static bool zswap_is_full(void) | |
366 | { | |
367 | return (totalram_pages * zswap_max_pool_percent / 100 < | |
368 | zswap_pool_pages); | |
369 | } | |
370 | ||
371 | /* | |
372 | * Carries out the common pattern of freeing and entry's zsmalloc allocation, | |
373 | * freeing the entry itself, and decrementing the number of stored pages. | |
374 | */ | |
375 | static void zswap_free_entry(struct zswap_tree *tree, struct zswap_entry *entry) | |
376 | { | |
377 | zbud_free(tree->pool, entry->handle); | |
378 | zswap_entry_cache_free(entry); | |
379 | atomic_dec(&zswap_stored_pages); | |
380 | zswap_pool_pages = zbud_get_pool_size(tree->pool); | |
381 | } | |
382 | ||
383 | /********************************* | |
384 | * writeback code | |
385 | **********************************/ | |
386 | /* return enum for zswap_get_swap_cache_page */ | |
387 | enum zswap_get_swap_ret { | |
388 | ZSWAP_SWAPCACHE_NEW, | |
389 | ZSWAP_SWAPCACHE_EXIST, | |
67d13fe8 | 390 | ZSWAP_SWAPCACHE_FAIL, |
2b281117 SJ |
391 | }; |
392 | ||
393 | /* | |
394 | * zswap_get_swap_cache_page | |
395 | * | |
396 | * This is an adaption of read_swap_cache_async() | |
397 | * | |
398 | * This function tries to find a page with the given swap entry | |
399 | * in the swapper_space address space (the swap cache). If the page | |
400 | * is found, it is returned in retpage. Otherwise, a page is allocated, | |
401 | * added to the swap cache, and returned in retpage. | |
402 | * | |
403 | * If success, the swap cache page is returned in retpage | |
67d13fe8 WY |
404 | * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache |
405 | * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated, | |
406 | * the new page is added to swapcache and locked | |
407 | * Returns ZSWAP_SWAPCACHE_FAIL on error | |
2b281117 SJ |
408 | */ |
409 | static int zswap_get_swap_cache_page(swp_entry_t entry, | |
410 | struct page **retpage) | |
411 | { | |
412 | struct page *found_page, *new_page = NULL; | |
822518dc | 413 | struct address_space *swapper_space = swap_address_space(entry); |
2b281117 SJ |
414 | int err; |
415 | ||
416 | *retpage = NULL; | |
417 | do { | |
418 | /* | |
419 | * First check the swap cache. Since this is normally | |
420 | * called after lookup_swap_cache() failed, re-calling | |
421 | * that would confuse statistics. | |
422 | */ | |
423 | found_page = find_get_page(swapper_space, entry.val); | |
424 | if (found_page) | |
425 | break; | |
426 | ||
427 | /* | |
428 | * Get a new page to read into from swap. | |
429 | */ | |
430 | if (!new_page) { | |
431 | new_page = alloc_page(GFP_KERNEL); | |
432 | if (!new_page) | |
433 | break; /* Out of memory */ | |
434 | } | |
435 | ||
436 | /* | |
437 | * call radix_tree_preload() while we can wait. | |
438 | */ | |
439 | err = radix_tree_preload(GFP_KERNEL); | |
440 | if (err) | |
441 | break; | |
442 | ||
443 | /* | |
444 | * Swap entry may have been freed since our caller observed it. | |
445 | */ | |
446 | err = swapcache_prepare(entry); | |
447 | if (err == -EEXIST) { /* seems racy */ | |
448 | radix_tree_preload_end(); | |
449 | continue; | |
450 | } | |
451 | if (err) { /* swp entry is obsolete ? */ | |
452 | radix_tree_preload_end(); | |
453 | break; | |
454 | } | |
455 | ||
456 | /* May fail (-ENOMEM) if radix-tree node allocation failed. */ | |
457 | __set_page_locked(new_page); | |
458 | SetPageSwapBacked(new_page); | |
459 | err = __add_to_swap_cache(new_page, entry); | |
460 | if (likely(!err)) { | |
461 | radix_tree_preload_end(); | |
462 | lru_cache_add_anon(new_page); | |
463 | *retpage = new_page; | |
464 | return ZSWAP_SWAPCACHE_NEW; | |
465 | } | |
466 | radix_tree_preload_end(); | |
467 | ClearPageSwapBacked(new_page); | |
468 | __clear_page_locked(new_page); | |
469 | /* | |
470 | * add_to_swap_cache() doesn't return -EEXIST, so we can safely | |
471 | * clear SWAP_HAS_CACHE flag. | |
472 | */ | |
473 | swapcache_free(entry, NULL); | |
474 | } while (err != -ENOMEM); | |
475 | ||
476 | if (new_page) | |
477 | page_cache_release(new_page); | |
478 | if (!found_page) | |
67d13fe8 | 479 | return ZSWAP_SWAPCACHE_FAIL; |
2b281117 SJ |
480 | *retpage = found_page; |
481 | return ZSWAP_SWAPCACHE_EXIST; | |
482 | } | |
483 | ||
484 | /* | |
485 | * Attempts to free an entry by adding a page to the swap cache, | |
486 | * decompressing the entry data into the page, and issuing a | |
487 | * bio write to write the page back to the swap device. | |
488 | * | |
489 | * This can be thought of as a "resumed writeback" of the page | |
490 | * to the swap device. We are basically resuming the same swap | |
491 | * writeback path that was intercepted with the frontswap_store() | |
492 | * in the first place. After the page has been decompressed into | |
493 | * the swap cache, the compressed version stored by zswap can be | |
494 | * freed. | |
495 | */ | |
496 | static int zswap_writeback_entry(struct zbud_pool *pool, unsigned long handle) | |
497 | { | |
498 | struct zswap_header *zhdr; | |
499 | swp_entry_t swpentry; | |
500 | struct zswap_tree *tree; | |
501 | pgoff_t offset; | |
502 | struct zswap_entry *entry; | |
503 | struct page *page; | |
504 | u8 *src, *dst; | |
505 | unsigned int dlen; | |
506 | int ret, refcount; | |
507 | struct writeback_control wbc = { | |
508 | .sync_mode = WB_SYNC_NONE, | |
509 | }; | |
510 | ||
511 | /* extract swpentry from data */ | |
512 | zhdr = zbud_map(pool, handle); | |
513 | swpentry = zhdr->swpentry; /* here */ | |
514 | zbud_unmap(pool, handle); | |
515 | tree = zswap_trees[swp_type(swpentry)]; | |
516 | offset = swp_offset(swpentry); | |
517 | BUG_ON(pool != tree->pool); | |
518 | ||
519 | /* find and ref zswap entry */ | |
520 | spin_lock(&tree->lock); | |
521 | entry = zswap_rb_search(&tree->rbroot, offset); | |
522 | if (!entry) { | |
523 | /* entry was invalidated */ | |
524 | spin_unlock(&tree->lock); | |
525 | return 0; | |
526 | } | |
527 | zswap_entry_get(entry); | |
528 | spin_unlock(&tree->lock); | |
529 | BUG_ON(offset != entry->offset); | |
530 | ||
531 | /* try to allocate swap cache page */ | |
532 | switch (zswap_get_swap_cache_page(swpentry, &page)) { | |
67d13fe8 | 533 | case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */ |
2b281117 SJ |
534 | ret = -ENOMEM; |
535 | goto fail; | |
536 | ||
67d13fe8 | 537 | case ZSWAP_SWAPCACHE_EXIST: |
2b281117 SJ |
538 | /* page is already in the swap cache, ignore for now */ |
539 | page_cache_release(page); | |
540 | ret = -EEXIST; | |
541 | goto fail; | |
542 | ||
543 | case ZSWAP_SWAPCACHE_NEW: /* page is locked */ | |
544 | /* decompress */ | |
545 | dlen = PAGE_SIZE; | |
546 | src = (u8 *)zbud_map(tree->pool, entry->handle) + | |
547 | sizeof(struct zswap_header); | |
548 | dst = kmap_atomic(page); | |
549 | ret = zswap_comp_op(ZSWAP_COMPOP_DECOMPRESS, src, | |
550 | entry->length, dst, &dlen); | |
551 | kunmap_atomic(dst); | |
552 | zbud_unmap(tree->pool, entry->handle); | |
553 | BUG_ON(ret); | |
554 | BUG_ON(dlen != PAGE_SIZE); | |
555 | ||
556 | /* page is up to date */ | |
557 | SetPageUptodate(page); | |
558 | } | |
559 | ||
b349acc7 WY |
560 | /* move it to the tail of the inactive list after end_writeback */ |
561 | SetPageReclaim(page); | |
562 | ||
2b281117 SJ |
563 | /* start writeback */ |
564 | __swap_writepage(page, &wbc, end_swap_bio_write); | |
565 | page_cache_release(page); | |
566 | zswap_written_back_pages++; | |
567 | ||
568 | spin_lock(&tree->lock); | |
569 | ||
570 | /* drop local reference */ | |
571 | zswap_entry_put(entry); | |
572 | /* drop the initial reference from entry creation */ | |
573 | refcount = zswap_entry_put(entry); | |
574 | ||
575 | /* | |
576 | * There are three possible values for refcount here: | |
577 | * (1) refcount is 1, load is in progress, unlink from rbtree, | |
578 | * load will free | |
579 | * (2) refcount is 0, (normal case) entry is valid, | |
580 | * remove from rbtree and free entry | |
581 | * (3) refcount is -1, invalidate happened during writeback, | |
582 | * free entry | |
583 | */ | |
584 | if (refcount >= 0) { | |
585 | /* no invalidate yet, remove from rbtree */ | |
586 | rb_erase(&entry->rbnode, &tree->rbroot); | |
587 | } | |
588 | spin_unlock(&tree->lock); | |
589 | if (refcount <= 0) { | |
590 | /* free the entry */ | |
591 | zswap_free_entry(tree, entry); | |
592 | return 0; | |
593 | } | |
594 | return -EAGAIN; | |
595 | ||
596 | fail: | |
597 | spin_lock(&tree->lock); | |
67d13fe8 WY |
598 | refcount = zswap_entry_put(entry); |
599 | if (refcount <= 0) { | |
600 | /* invalidate happened, consider writeback as success */ | |
601 | zswap_free_entry(tree, entry); | |
602 | ret = 0; | |
603 | } | |
2b281117 SJ |
604 | spin_unlock(&tree->lock); |
605 | return ret; | |
606 | } | |
607 | ||
608 | /********************************* | |
609 | * frontswap hooks | |
610 | **********************************/ | |
611 | /* attempts to compress and store an single page */ | |
612 | static int zswap_frontswap_store(unsigned type, pgoff_t offset, | |
613 | struct page *page) | |
614 | { | |
615 | struct zswap_tree *tree = zswap_trees[type]; | |
616 | struct zswap_entry *entry, *dupentry; | |
617 | int ret; | |
618 | unsigned int dlen = PAGE_SIZE, len; | |
619 | unsigned long handle; | |
620 | char *buf; | |
621 | u8 *src, *dst; | |
622 | struct zswap_header *zhdr; | |
623 | ||
624 | if (!tree) { | |
625 | ret = -ENODEV; | |
626 | goto reject; | |
627 | } | |
628 | ||
629 | /* reclaim space if needed */ | |
630 | if (zswap_is_full()) { | |
631 | zswap_pool_limit_hit++; | |
632 | if (zbud_reclaim_page(tree->pool, 8)) { | |
633 | zswap_reject_reclaim_fail++; | |
634 | ret = -ENOMEM; | |
635 | goto reject; | |
636 | } | |
637 | } | |
638 | ||
639 | /* allocate entry */ | |
640 | entry = zswap_entry_cache_alloc(GFP_KERNEL); | |
641 | if (!entry) { | |
642 | zswap_reject_kmemcache_fail++; | |
643 | ret = -ENOMEM; | |
644 | goto reject; | |
645 | } | |
646 | ||
647 | /* compress */ | |
648 | dst = get_cpu_var(zswap_dstmem); | |
649 | src = kmap_atomic(page); | |
650 | ret = zswap_comp_op(ZSWAP_COMPOP_COMPRESS, src, PAGE_SIZE, dst, &dlen); | |
651 | kunmap_atomic(src); | |
652 | if (ret) { | |
653 | ret = -EINVAL; | |
654 | goto freepage; | |
655 | } | |
656 | ||
657 | /* store */ | |
658 | len = dlen + sizeof(struct zswap_header); | |
659 | ret = zbud_alloc(tree->pool, len, __GFP_NORETRY | __GFP_NOWARN, | |
660 | &handle); | |
661 | if (ret == -ENOSPC) { | |
662 | zswap_reject_compress_poor++; | |
663 | goto freepage; | |
664 | } | |
665 | if (ret) { | |
666 | zswap_reject_alloc_fail++; | |
667 | goto freepage; | |
668 | } | |
669 | zhdr = zbud_map(tree->pool, handle); | |
670 | zhdr->swpentry = swp_entry(type, offset); | |
671 | buf = (u8 *)(zhdr + 1); | |
672 | memcpy(buf, dst, dlen); | |
673 | zbud_unmap(tree->pool, handle); | |
674 | put_cpu_var(zswap_dstmem); | |
675 | ||
676 | /* populate entry */ | |
677 | entry->offset = offset; | |
678 | entry->handle = handle; | |
679 | entry->length = dlen; | |
680 | ||
681 | /* map */ | |
682 | spin_lock(&tree->lock); | |
683 | do { | |
684 | ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry); | |
685 | if (ret == -EEXIST) { | |
686 | zswap_duplicate_entry++; | |
687 | /* remove from rbtree */ | |
688 | rb_erase(&dupentry->rbnode, &tree->rbroot); | |
689 | if (!zswap_entry_put(dupentry)) { | |
690 | /* free */ | |
691 | zswap_free_entry(tree, dupentry); | |
692 | } | |
693 | } | |
694 | } while (ret == -EEXIST); | |
695 | spin_unlock(&tree->lock); | |
696 | ||
697 | /* update stats */ | |
698 | atomic_inc(&zswap_stored_pages); | |
699 | zswap_pool_pages = zbud_get_pool_size(tree->pool); | |
700 | ||
701 | return 0; | |
702 | ||
703 | freepage: | |
704 | put_cpu_var(zswap_dstmem); | |
705 | zswap_entry_cache_free(entry); | |
706 | reject: | |
707 | return ret; | |
708 | } | |
709 | ||
710 | /* | |
711 | * returns 0 if the page was successfully decompressed | |
712 | * return -1 on entry not found or error | |
713 | */ | |
714 | static int zswap_frontswap_load(unsigned type, pgoff_t offset, | |
715 | struct page *page) | |
716 | { | |
717 | struct zswap_tree *tree = zswap_trees[type]; | |
718 | struct zswap_entry *entry; | |
719 | u8 *src, *dst; | |
720 | unsigned int dlen; | |
721 | int refcount, ret; | |
722 | ||
723 | /* find */ | |
724 | spin_lock(&tree->lock); | |
725 | entry = zswap_rb_search(&tree->rbroot, offset); | |
726 | if (!entry) { | |
727 | /* entry was written back */ | |
728 | spin_unlock(&tree->lock); | |
729 | return -1; | |
730 | } | |
731 | zswap_entry_get(entry); | |
732 | spin_unlock(&tree->lock); | |
733 | ||
734 | /* decompress */ | |
735 | dlen = PAGE_SIZE; | |
736 | src = (u8 *)zbud_map(tree->pool, entry->handle) + | |
737 | sizeof(struct zswap_header); | |
738 | dst = kmap_atomic(page); | |
739 | ret = zswap_comp_op(ZSWAP_COMPOP_DECOMPRESS, src, entry->length, | |
740 | dst, &dlen); | |
741 | kunmap_atomic(dst); | |
742 | zbud_unmap(tree->pool, entry->handle); | |
743 | BUG_ON(ret); | |
744 | ||
745 | spin_lock(&tree->lock); | |
746 | refcount = zswap_entry_put(entry); | |
747 | if (likely(refcount)) { | |
748 | spin_unlock(&tree->lock); | |
749 | return 0; | |
750 | } | |
751 | spin_unlock(&tree->lock); | |
752 | ||
753 | /* | |
754 | * We don't have to unlink from the rbtree because | |
755 | * zswap_writeback_entry() or zswap_frontswap_invalidate page() | |
756 | * has already done this for us if we are the last reference. | |
757 | */ | |
758 | /* free */ | |
759 | ||
760 | zswap_free_entry(tree, entry); | |
761 | ||
762 | return 0; | |
763 | } | |
764 | ||
765 | /* frees an entry in zswap */ | |
766 | static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset) | |
767 | { | |
768 | struct zswap_tree *tree = zswap_trees[type]; | |
769 | struct zswap_entry *entry; | |
770 | int refcount; | |
771 | ||
772 | /* find */ | |
773 | spin_lock(&tree->lock); | |
774 | entry = zswap_rb_search(&tree->rbroot, offset); | |
775 | if (!entry) { | |
776 | /* entry was written back */ | |
777 | spin_unlock(&tree->lock); | |
778 | return; | |
779 | } | |
780 | ||
781 | /* remove from rbtree */ | |
782 | rb_erase(&entry->rbnode, &tree->rbroot); | |
783 | ||
784 | /* drop the initial reference from entry creation */ | |
785 | refcount = zswap_entry_put(entry); | |
786 | ||
787 | spin_unlock(&tree->lock); | |
788 | ||
789 | if (refcount) { | |
790 | /* writeback in progress, writeback will free */ | |
791 | return; | |
792 | } | |
793 | ||
794 | /* free */ | |
795 | zswap_free_entry(tree, entry); | |
796 | } | |
797 | ||
798 | /* frees all zswap entries for the given swap type */ | |
799 | static void zswap_frontswap_invalidate_area(unsigned type) | |
800 | { | |
801 | struct zswap_tree *tree = zswap_trees[type]; | |
0bd42136 | 802 | struct zswap_entry *entry, *n; |
2b281117 SJ |
803 | |
804 | if (!tree) | |
805 | return; | |
806 | ||
807 | /* walk the tree and free everything */ | |
808 | spin_lock(&tree->lock); | |
0bd42136 | 809 | rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode) { |
2b281117 SJ |
810 | zbud_free(tree->pool, entry->handle); |
811 | zswap_entry_cache_free(entry); | |
812 | atomic_dec(&zswap_stored_pages); | |
813 | } | |
814 | tree->rbroot = RB_ROOT; | |
815 | spin_unlock(&tree->lock); | |
aa9bca05 WY |
816 | |
817 | zbud_destroy_pool(tree->pool); | |
818 | kfree(tree); | |
819 | zswap_trees[type] = NULL; | |
2b281117 SJ |
820 | } |
821 | ||
822 | static struct zbud_ops zswap_zbud_ops = { | |
823 | .evict = zswap_writeback_entry | |
824 | }; | |
825 | ||
826 | static void zswap_frontswap_init(unsigned type) | |
827 | { | |
828 | struct zswap_tree *tree; | |
829 | ||
830 | tree = kzalloc(sizeof(struct zswap_tree), GFP_KERNEL); | |
831 | if (!tree) | |
832 | goto err; | |
833 | tree->pool = zbud_create_pool(GFP_KERNEL, &zswap_zbud_ops); | |
834 | if (!tree->pool) | |
835 | goto freetree; | |
836 | tree->rbroot = RB_ROOT; | |
837 | spin_lock_init(&tree->lock); | |
838 | zswap_trees[type] = tree; | |
839 | return; | |
840 | ||
841 | freetree: | |
842 | kfree(tree); | |
843 | err: | |
844 | pr_err("alloc failed, zswap disabled for swap type %d\n", type); | |
845 | } | |
846 | ||
847 | static struct frontswap_ops zswap_frontswap_ops = { | |
848 | .store = zswap_frontswap_store, | |
849 | .load = zswap_frontswap_load, | |
850 | .invalidate_page = zswap_frontswap_invalidate_page, | |
851 | .invalidate_area = zswap_frontswap_invalidate_area, | |
852 | .init = zswap_frontswap_init | |
853 | }; | |
854 | ||
855 | /********************************* | |
856 | * debugfs functions | |
857 | **********************************/ | |
858 | #ifdef CONFIG_DEBUG_FS | |
859 | #include <linux/debugfs.h> | |
860 | ||
861 | static struct dentry *zswap_debugfs_root; | |
862 | ||
863 | static int __init zswap_debugfs_init(void) | |
864 | { | |
865 | if (!debugfs_initialized()) | |
866 | return -ENODEV; | |
867 | ||
868 | zswap_debugfs_root = debugfs_create_dir("zswap", NULL); | |
869 | if (!zswap_debugfs_root) | |
870 | return -ENOMEM; | |
871 | ||
872 | debugfs_create_u64("pool_limit_hit", S_IRUGO, | |
873 | zswap_debugfs_root, &zswap_pool_limit_hit); | |
874 | debugfs_create_u64("reject_reclaim_fail", S_IRUGO, | |
875 | zswap_debugfs_root, &zswap_reject_reclaim_fail); | |
876 | debugfs_create_u64("reject_alloc_fail", S_IRUGO, | |
877 | zswap_debugfs_root, &zswap_reject_alloc_fail); | |
878 | debugfs_create_u64("reject_kmemcache_fail", S_IRUGO, | |
879 | zswap_debugfs_root, &zswap_reject_kmemcache_fail); | |
880 | debugfs_create_u64("reject_compress_poor", S_IRUGO, | |
881 | zswap_debugfs_root, &zswap_reject_compress_poor); | |
882 | debugfs_create_u64("written_back_pages", S_IRUGO, | |
883 | zswap_debugfs_root, &zswap_written_back_pages); | |
884 | debugfs_create_u64("duplicate_entry", S_IRUGO, | |
885 | zswap_debugfs_root, &zswap_duplicate_entry); | |
886 | debugfs_create_u64("pool_pages", S_IRUGO, | |
887 | zswap_debugfs_root, &zswap_pool_pages); | |
888 | debugfs_create_atomic_t("stored_pages", S_IRUGO, | |
889 | zswap_debugfs_root, &zswap_stored_pages); | |
890 | ||
891 | return 0; | |
892 | } | |
893 | ||
894 | static void __exit zswap_debugfs_exit(void) | |
895 | { | |
896 | debugfs_remove_recursive(zswap_debugfs_root); | |
897 | } | |
898 | #else | |
899 | static int __init zswap_debugfs_init(void) | |
900 | { | |
901 | return 0; | |
902 | } | |
903 | ||
904 | static void __exit zswap_debugfs_exit(void) { } | |
905 | #endif | |
906 | ||
907 | /********************************* | |
908 | * module init and exit | |
909 | **********************************/ | |
910 | static int __init init_zswap(void) | |
911 | { | |
912 | if (!zswap_enabled) | |
913 | return 0; | |
914 | ||
915 | pr_info("loading zswap\n"); | |
916 | if (zswap_entry_cache_create()) { | |
917 | pr_err("entry cache creation failed\n"); | |
918 | goto error; | |
919 | } | |
920 | if (zswap_comp_init()) { | |
921 | pr_err("compressor initialization failed\n"); | |
922 | goto compfail; | |
923 | } | |
924 | if (zswap_cpu_init()) { | |
925 | pr_err("per-cpu initialization failed\n"); | |
926 | goto pcpufail; | |
927 | } | |
928 | frontswap_register_ops(&zswap_frontswap_ops); | |
929 | if (zswap_debugfs_init()) | |
930 | pr_warn("debugfs initialization failed\n"); | |
931 | return 0; | |
932 | pcpufail: | |
933 | zswap_comp_exit(); | |
934 | compfail: | |
935 | zswap_entry_cache_destory(); | |
936 | error: | |
937 | return -ENOMEM; | |
938 | } | |
939 | /* must be late so crypto has time to come up */ | |
940 | late_initcall(init_zswap); | |
941 | ||
942 | MODULE_LICENSE("GPL"); | |
943 | MODULE_AUTHOR("Seth Jennings <sjenning@linux.vnet.ibm.com>"); | |
944 | MODULE_DESCRIPTION("Compressed cache for swap pages"); |