Commit | Line | Data |
---|---|---|
29f233cf DM |
1 | /* |
2 | * Frontswap frontend | |
3 | * | |
4 | * This code provides the generic "frontend" layer to call a matching | |
5 | * "backend" driver implementation of frontswap. See | |
6 | * Documentation/vm/frontswap.txt for more information. | |
7 | * | |
8 | * Copyright (C) 2009-2012 Oracle Corp. All rights reserved. | |
9 | * Author: Dan Magenheimer | |
10 | * | |
11 | * This work is licensed under the terms of the GNU GPL, version 2. | |
12 | */ | |
13 | ||
29f233cf DM |
14 | #include <linux/mman.h> |
15 | #include <linux/swap.h> | |
16 | #include <linux/swapops.h> | |
29f233cf | 17 | #include <linux/security.h> |
29f233cf | 18 | #include <linux/module.h> |
29f233cf DM |
19 | #include <linux/debugfs.h> |
20 | #include <linux/frontswap.h> | |
21 | #include <linux/swapfile.h> | |
22 | ||
23 | /* | |
24 | * frontswap_ops is set by frontswap_register_ops to contain the pointers | |
25 | * to the frontswap "backend" implementation functions. | |
26 | */ | |
27 | static struct frontswap_ops frontswap_ops __read_mostly; | |
28 | ||
29 | /* | |
30 | * This global enablement flag reduces overhead on systems where frontswap_ops | |
31 | * has not been registered, so is preferred to the slower alternative: a | |
32 | * function call that checks a non-global. | |
33 | */ | |
34 | bool frontswap_enabled __read_mostly; | |
35 | EXPORT_SYMBOL(frontswap_enabled); | |
36 | ||
37 | /* | |
165c8aed | 38 | * If enabled, frontswap_store will return failure even on success. As |
29f233cf DM |
39 | * a result, the swap subsystem will always write the page to swap, in |
40 | * effect converting frontswap into a writethrough cache. In this mode, | |
41 | * there is no direct reduction in swap writes, but a frontswap backend | |
42 | * can unilaterally "reclaim" any pages in use with no data loss, thus | |
43 | * providing increases control over maximum memory usage due to frontswap. | |
44 | */ | |
45 | static bool frontswap_writethrough_enabled __read_mostly; | |
46 | ||
47 | #ifdef CONFIG_DEBUG_FS | |
48 | /* | |
49 | * Counters available via /sys/kernel/debug/frontswap (if debugfs is | |
50 | * properly configured). These are for information only so are not protected | |
51 | * against increment races. | |
52 | */ | |
165c8aed KRW |
53 | static u64 frontswap_loads; |
54 | static u64 frontswap_succ_stores; | |
55 | static u64 frontswap_failed_stores; | |
29f233cf DM |
56 | static u64 frontswap_invalidates; |
57 | ||
165c8aed KRW |
58 | static inline void inc_frontswap_loads(void) { |
59 | frontswap_loads++; | |
29f233cf | 60 | } |
165c8aed KRW |
61 | static inline void inc_frontswap_succ_stores(void) { |
62 | frontswap_succ_stores++; | |
29f233cf | 63 | } |
165c8aed KRW |
64 | static inline void inc_frontswap_failed_stores(void) { |
65 | frontswap_failed_stores++; | |
29f233cf DM |
66 | } |
67 | static inline void inc_frontswap_invalidates(void) { | |
68 | frontswap_invalidates++; | |
69 | } | |
70 | #else | |
165c8aed KRW |
71 | static inline void inc_frontswap_loads(void) { } |
72 | static inline void inc_frontswap_succ_stores(void) { } | |
73 | static inline void inc_frontswap_failed_stores(void) { } | |
29f233cf DM |
74 | static inline void inc_frontswap_invalidates(void) { } |
75 | #endif | |
76 | /* | |
77 | * Register operations for frontswap, returning previous thus allowing | |
78 | * detection of multiple backends and possible nesting. | |
79 | */ | |
80 | struct frontswap_ops frontswap_register_ops(struct frontswap_ops *ops) | |
81 | { | |
82 | struct frontswap_ops old = frontswap_ops; | |
83 | ||
84 | frontswap_ops = *ops; | |
85 | frontswap_enabled = true; | |
86 | return old; | |
87 | } | |
88 | EXPORT_SYMBOL(frontswap_register_ops); | |
89 | ||
90 | /* | |
91 | * Enable/disable frontswap writethrough (see above). | |
92 | */ | |
93 | void frontswap_writethrough(bool enable) | |
94 | { | |
95 | frontswap_writethrough_enabled = enable; | |
96 | } | |
97 | EXPORT_SYMBOL(frontswap_writethrough); | |
98 | ||
99 | /* | |
100 | * Called when a swap device is swapon'd. | |
101 | */ | |
102 | void __frontswap_init(unsigned type) | |
103 | { | |
104 | struct swap_info_struct *sis = swap_info[type]; | |
105 | ||
106 | BUG_ON(sis == NULL); | |
107 | if (sis->frontswap_map == NULL) | |
108 | return; | |
f9f08103 | 109 | frontswap_ops.init(type); |
29f233cf DM |
110 | } |
111 | EXPORT_SYMBOL(__frontswap_init); | |
112 | ||
611edfed SL |
113 | static inline void __frontswap_clear(struct swap_info_struct *sis, pgoff_t offset) |
114 | { | |
115 | frontswap_clear(sis, offset); | |
116 | atomic_dec(&sis->frontswap_pages); | |
117 | } | |
118 | ||
29f233cf | 119 | /* |
165c8aed | 120 | * "Store" data from a page to frontswap and associate it with the page's |
29f233cf DM |
121 | * swaptype and offset. Page must be locked and in the swap cache. |
122 | * If frontswap already contains a page with matching swaptype and | |
1d00015e | 123 | * offset, the frontswap implementation may either overwrite the data and |
29f233cf DM |
124 | * return success or invalidate the page from frontswap and return failure. |
125 | */ | |
165c8aed | 126 | int __frontswap_store(struct page *page) |
29f233cf DM |
127 | { |
128 | int ret = -1, dup = 0; | |
129 | swp_entry_t entry = { .val = page_private(page), }; | |
130 | int type = swp_type(entry); | |
131 | struct swap_info_struct *sis = swap_info[type]; | |
132 | pgoff_t offset = swp_offset(entry); | |
133 | ||
134 | BUG_ON(!PageLocked(page)); | |
135 | BUG_ON(sis == NULL); | |
136 | if (frontswap_test(sis, offset)) | |
137 | dup = 1; | |
ef383597 | 138 | ret = frontswap_ops.store(type, offset, page); |
29f233cf DM |
139 | if (ret == 0) { |
140 | frontswap_set(sis, offset); | |
165c8aed | 141 | inc_frontswap_succ_stores(); |
29f233cf DM |
142 | if (!dup) |
143 | atomic_inc(&sis->frontswap_pages); | |
d9674dda | 144 | } else { |
29f233cf DM |
145 | /* |
146 | failed dup always results in automatic invalidate of | |
147 | the (older) page from frontswap | |
148 | */ | |
165c8aed | 149 | inc_frontswap_failed_stores(); |
611edfed SL |
150 | if (dup) |
151 | __frontswap_clear(sis, offset); | |
4bb3e31e | 152 | } |
29f233cf DM |
153 | if (frontswap_writethrough_enabled) |
154 | /* report failure so swap also writes to swap device */ | |
155 | ret = -1; | |
156 | return ret; | |
157 | } | |
165c8aed | 158 | EXPORT_SYMBOL(__frontswap_store); |
29f233cf DM |
159 | |
160 | /* | |
161 | * "Get" data from frontswap associated with swaptype and offset that were | |
162 | * specified when the data was put to frontswap and use it to fill the | |
163 | * specified page with data. Page must be locked and in the swap cache. | |
164 | */ | |
165c8aed | 165 | int __frontswap_load(struct page *page) |
29f233cf DM |
166 | { |
167 | int ret = -1; | |
168 | swp_entry_t entry = { .val = page_private(page), }; | |
169 | int type = swp_type(entry); | |
170 | struct swap_info_struct *sis = swap_info[type]; | |
171 | pgoff_t offset = swp_offset(entry); | |
172 | ||
173 | BUG_ON(!PageLocked(page)); | |
174 | BUG_ON(sis == NULL); | |
175 | if (frontswap_test(sis, offset)) | |
ef383597 | 176 | ret = frontswap_ops.load(type, offset, page); |
29f233cf | 177 | if (ret == 0) |
165c8aed | 178 | inc_frontswap_loads(); |
29f233cf DM |
179 | return ret; |
180 | } | |
165c8aed | 181 | EXPORT_SYMBOL(__frontswap_load); |
29f233cf DM |
182 | |
183 | /* | |
184 | * Invalidate any data from frontswap associated with the specified swaptype | |
185 | * and offset so that a subsequent "get" will fail. | |
186 | */ | |
187 | void __frontswap_invalidate_page(unsigned type, pgoff_t offset) | |
188 | { | |
189 | struct swap_info_struct *sis = swap_info[type]; | |
190 | ||
191 | BUG_ON(sis == NULL); | |
192 | if (frontswap_test(sis, offset)) { | |
ef383597 | 193 | frontswap_ops.invalidate_page(type, offset); |
611edfed | 194 | __frontswap_clear(sis, offset); |
29f233cf DM |
195 | inc_frontswap_invalidates(); |
196 | } | |
197 | } | |
198 | EXPORT_SYMBOL(__frontswap_invalidate_page); | |
199 | ||
200 | /* | |
201 | * Invalidate all data from frontswap associated with all offsets for the | |
202 | * specified swaptype. | |
203 | */ | |
204 | void __frontswap_invalidate_area(unsigned type) | |
205 | { | |
206 | struct swap_info_struct *sis = swap_info[type]; | |
207 | ||
208 | BUG_ON(sis == NULL); | |
209 | if (sis->frontswap_map == NULL) | |
210 | return; | |
ef383597 | 211 | frontswap_ops.invalidate_area(type); |
29f233cf DM |
212 | atomic_set(&sis->frontswap_pages, 0); |
213 | memset(sis->frontswap_map, 0, sis->max / sizeof(long)); | |
214 | } | |
215 | EXPORT_SYMBOL(__frontswap_invalidate_area); | |
216 | ||
96253444 SL |
217 | static unsigned long __frontswap_curr_pages(void) |
218 | { | |
219 | int type; | |
220 | unsigned long totalpages = 0; | |
221 | struct swap_info_struct *si = NULL; | |
222 | ||
223 | assert_spin_locked(&swap_lock); | |
224 | for (type = swap_list.head; type >= 0; type = si->next) { | |
225 | si = swap_info[type]; | |
226 | totalpages += atomic_read(&si->frontswap_pages); | |
227 | } | |
228 | return totalpages; | |
229 | } | |
230 | ||
f116695a SL |
231 | static int __frontswap_unuse_pages(unsigned long total, unsigned long *unused, |
232 | int *swapid) | |
233 | { | |
234 | int ret = -EINVAL; | |
235 | struct swap_info_struct *si = NULL; | |
236 | int si_frontswap_pages; | |
237 | unsigned long total_pages_to_unuse = total; | |
238 | unsigned long pages = 0, pages_to_unuse = 0; | |
239 | int type; | |
240 | ||
241 | assert_spin_locked(&swap_lock); | |
242 | for (type = swap_list.head; type >= 0; type = si->next) { | |
243 | si = swap_info[type]; | |
244 | si_frontswap_pages = atomic_read(&si->frontswap_pages); | |
245 | if (total_pages_to_unuse < si_frontswap_pages) { | |
246 | pages = pages_to_unuse = total_pages_to_unuse; | |
247 | } else { | |
248 | pages = si_frontswap_pages; | |
249 | pages_to_unuse = 0; /* unuse all */ | |
250 | } | |
251 | /* ensure there is enough RAM to fetch pages from frontswap */ | |
252 | if (security_vm_enough_memory_mm(current->mm, pages)) { | |
253 | ret = -ENOMEM; | |
254 | continue; | |
255 | } | |
256 | vm_unacct_memory(pages); | |
257 | *unused = pages_to_unuse; | |
258 | *swapid = type; | |
259 | ret = 0; | |
260 | break; | |
261 | } | |
262 | ||
263 | return ret; | |
264 | } | |
265 | ||
69217b4c SL |
266 | static int __frontswap_shrink(unsigned long target_pages, |
267 | unsigned long *pages_to_unuse, | |
268 | int *type) | |
269 | { | |
270 | unsigned long total_pages = 0, total_pages_to_unuse; | |
271 | ||
272 | assert_spin_locked(&swap_lock); | |
273 | ||
274 | total_pages = __frontswap_curr_pages(); | |
275 | if (total_pages <= target_pages) { | |
276 | /* Nothing to do */ | |
277 | *pages_to_unuse = 0; | |
278 | return 0; | |
279 | } | |
280 | total_pages_to_unuse = total_pages - target_pages; | |
281 | return __frontswap_unuse_pages(total_pages_to_unuse, pages_to_unuse, type); | |
282 | } | |
283 | ||
29f233cf DM |
284 | /* |
285 | * Frontswap, like a true swap device, may unnecessarily retain pages | |
286 | * under certain circumstances; "shrink" frontswap is essentially a | |
287 | * "partial swapoff" and works by calling try_to_unuse to attempt to | |
288 | * unuse enough frontswap pages to attempt to -- subject to memory | |
289 | * constraints -- reduce the number of pages in frontswap to the | |
290 | * number given in the parameter target_pages. | |
291 | */ | |
292 | void frontswap_shrink(unsigned long target_pages) | |
293 | { | |
f116695a SL |
294 | unsigned long pages_to_unuse = 0; |
295 | int type, ret; | |
29f233cf DM |
296 | |
297 | /* | |
298 | * we don't want to hold swap_lock while doing a very | |
299 | * lengthy try_to_unuse, but swap_list may change | |
300 | * so restart scan from swap_list.head each time | |
301 | */ | |
302 | spin_lock(&swap_lock); | |
69217b4c | 303 | ret = __frontswap_shrink(target_pages, &pages_to_unuse, &type); |
29f233cf | 304 | spin_unlock(&swap_lock); |
69217b4c SL |
305 | if (ret == 0 && pages_to_unuse) |
306 | try_to_unuse(type, true, pages_to_unuse); | |
29f233cf DM |
307 | return; |
308 | } | |
309 | EXPORT_SYMBOL(frontswap_shrink); | |
310 | ||
311 | /* | |
312 | * Count and return the number of frontswap pages across all | |
313 | * swap devices. This is exported so that backend drivers can | |
314 | * determine current usage without reading debugfs. | |
315 | */ | |
316 | unsigned long frontswap_curr_pages(void) | |
317 | { | |
29f233cf | 318 | unsigned long totalpages = 0; |
29f233cf DM |
319 | |
320 | spin_lock(&swap_lock); | |
96253444 | 321 | totalpages = __frontswap_curr_pages(); |
29f233cf | 322 | spin_unlock(&swap_lock); |
96253444 | 323 | |
29f233cf DM |
324 | return totalpages; |
325 | } | |
326 | EXPORT_SYMBOL(frontswap_curr_pages); | |
327 | ||
328 | static int __init init_frontswap(void) | |
329 | { | |
330 | #ifdef CONFIG_DEBUG_FS | |
331 | struct dentry *root = debugfs_create_dir("frontswap", NULL); | |
332 | if (root == NULL) | |
333 | return -ENXIO; | |
165c8aed KRW |
334 | debugfs_create_u64("loads", S_IRUGO, root, &frontswap_loads); |
335 | debugfs_create_u64("succ_stores", S_IRUGO, root, &frontswap_succ_stores); | |
336 | debugfs_create_u64("failed_stores", S_IRUGO, root, | |
337 | &frontswap_failed_stores); | |
29f233cf DM |
338 | debugfs_create_u64("invalidates", S_IRUGO, |
339 | root, &frontswap_invalidates); | |
340 | #endif | |
341 | return 0; | |
342 | } | |
343 | ||
344 | module_init(init_frontswap); |