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 | ||
14 | #include <linux/mm.h> | |
15 | #include <linux/mman.h> | |
16 | #include <linux/swap.h> | |
17 | #include <linux/swapops.h> | |
18 | #include <linux/proc_fs.h> | |
19 | #include <linux/security.h> | |
20 | #include <linux/capability.h> | |
21 | #include <linux/module.h> | |
22 | #include <linux/uaccess.h> | |
23 | #include <linux/debugfs.h> | |
24 | #include <linux/frontswap.h> | |
25 | #include <linux/swapfile.h> | |
26 | ||
27 | /* | |
28 | * frontswap_ops is set by frontswap_register_ops to contain the pointers | |
29 | * to the frontswap "backend" implementation functions. | |
30 | */ | |
31 | static struct frontswap_ops frontswap_ops __read_mostly; | |
32 | ||
33 | /* | |
34 | * This global enablement flag reduces overhead on systems where frontswap_ops | |
35 | * has not been registered, so is preferred to the slower alternative: a | |
36 | * function call that checks a non-global. | |
37 | */ | |
38 | bool frontswap_enabled __read_mostly; | |
39 | EXPORT_SYMBOL(frontswap_enabled); | |
40 | ||
41 | /* | |
165c8aed | 42 | * If enabled, frontswap_store will return failure even on success. As |
29f233cf DM |
43 | * a result, the swap subsystem will always write the page to swap, in |
44 | * effect converting frontswap into a writethrough cache. In this mode, | |
45 | * there is no direct reduction in swap writes, but a frontswap backend | |
46 | * can unilaterally "reclaim" any pages in use with no data loss, thus | |
47 | * providing increases control over maximum memory usage due to frontswap. | |
48 | */ | |
49 | static bool frontswap_writethrough_enabled __read_mostly; | |
50 | ||
51 | #ifdef CONFIG_DEBUG_FS | |
52 | /* | |
53 | * Counters available via /sys/kernel/debug/frontswap (if debugfs is | |
54 | * properly configured). These are for information only so are not protected | |
55 | * against increment races. | |
56 | */ | |
165c8aed KRW |
57 | static u64 frontswap_loads; |
58 | static u64 frontswap_succ_stores; | |
59 | static u64 frontswap_failed_stores; | |
29f233cf DM |
60 | static u64 frontswap_invalidates; |
61 | ||
165c8aed KRW |
62 | static inline void inc_frontswap_loads(void) { |
63 | frontswap_loads++; | |
29f233cf | 64 | } |
165c8aed KRW |
65 | static inline void inc_frontswap_succ_stores(void) { |
66 | frontswap_succ_stores++; | |
29f233cf | 67 | } |
165c8aed KRW |
68 | static inline void inc_frontswap_failed_stores(void) { |
69 | frontswap_failed_stores++; | |
29f233cf DM |
70 | } |
71 | static inline void inc_frontswap_invalidates(void) { | |
72 | frontswap_invalidates++; | |
73 | } | |
74 | #else | |
165c8aed KRW |
75 | static inline void inc_frontswap_loads(void) { } |
76 | static inline void inc_frontswap_succ_stores(void) { } | |
77 | static inline void inc_frontswap_failed_stores(void) { } | |
29f233cf DM |
78 | static inline void inc_frontswap_invalidates(void) { } |
79 | #endif | |
80 | /* | |
81 | * Register operations for frontswap, returning previous thus allowing | |
82 | * detection of multiple backends and possible nesting. | |
83 | */ | |
84 | struct frontswap_ops frontswap_register_ops(struct frontswap_ops *ops) | |
85 | { | |
86 | struct frontswap_ops old = frontswap_ops; | |
87 | ||
88 | frontswap_ops = *ops; | |
89 | frontswap_enabled = true; | |
90 | return old; | |
91 | } | |
92 | EXPORT_SYMBOL(frontswap_register_ops); | |
93 | ||
94 | /* | |
95 | * Enable/disable frontswap writethrough (see above). | |
96 | */ | |
97 | void frontswap_writethrough(bool enable) | |
98 | { | |
99 | frontswap_writethrough_enabled = enable; | |
100 | } | |
101 | EXPORT_SYMBOL(frontswap_writethrough); | |
102 | ||
103 | /* | |
104 | * Called when a swap device is swapon'd. | |
105 | */ | |
106 | void __frontswap_init(unsigned type) | |
107 | { | |
108 | struct swap_info_struct *sis = swap_info[type]; | |
109 | ||
110 | BUG_ON(sis == NULL); | |
111 | if (sis->frontswap_map == NULL) | |
112 | return; | |
113 | if (frontswap_enabled) | |
ef383597 | 114 | frontswap_ops.init(type); |
29f233cf DM |
115 | } |
116 | EXPORT_SYMBOL(__frontswap_init); | |
117 | ||
118 | /* | |
165c8aed | 119 | * "Store" data from a page to frontswap and associate it with the page's |
29f233cf DM |
120 | * swaptype and offset. Page must be locked and in the swap cache. |
121 | * If frontswap already contains a page with matching swaptype and | |
122 | * offset, the frontswap implmentation may either overwrite the data and | |
123 | * return success or invalidate the page from frontswap and return failure. | |
124 | */ | |
165c8aed | 125 | int __frontswap_store(struct page *page) |
29f233cf DM |
126 | { |
127 | int ret = -1, dup = 0; | |
128 | swp_entry_t entry = { .val = page_private(page), }; | |
129 | int type = swp_type(entry); | |
130 | struct swap_info_struct *sis = swap_info[type]; | |
131 | pgoff_t offset = swp_offset(entry); | |
132 | ||
133 | BUG_ON(!PageLocked(page)); | |
134 | BUG_ON(sis == NULL); | |
135 | if (frontswap_test(sis, offset)) | |
136 | dup = 1; | |
ef383597 | 137 | ret = frontswap_ops.store(type, offset, page); |
29f233cf DM |
138 | if (ret == 0) { |
139 | frontswap_set(sis, offset); | |
165c8aed | 140 | inc_frontswap_succ_stores(); |
29f233cf DM |
141 | if (!dup) |
142 | atomic_inc(&sis->frontswap_pages); | |
143 | } else if (dup) { | |
144 | /* | |
145 | failed dup always results in automatic invalidate of | |
146 | the (older) page from frontswap | |
147 | */ | |
148 | frontswap_clear(sis, offset); | |
149 | atomic_dec(&sis->frontswap_pages); | |
165c8aed | 150 | inc_frontswap_failed_stores(); |
4bb3e31e | 151 | } else { |
165c8aed | 152 | inc_frontswap_failed_stores(); |
4bb3e31e | 153 | } |
29f233cf DM |
154 | if (frontswap_writethrough_enabled) |
155 | /* report failure so swap also writes to swap device */ | |
156 | ret = -1; | |
157 | return ret; | |
158 | } | |
165c8aed | 159 | EXPORT_SYMBOL(__frontswap_store); |
29f233cf DM |
160 | |
161 | /* | |
162 | * "Get" data from frontswap associated with swaptype and offset that were | |
163 | * specified when the data was put to frontswap and use it to fill the | |
164 | * specified page with data. Page must be locked and in the swap cache. | |
165 | */ | |
165c8aed | 166 | int __frontswap_load(struct page *page) |
29f233cf DM |
167 | { |
168 | int ret = -1; | |
169 | swp_entry_t entry = { .val = page_private(page), }; | |
170 | int type = swp_type(entry); | |
171 | struct swap_info_struct *sis = swap_info[type]; | |
172 | pgoff_t offset = swp_offset(entry); | |
173 | ||
174 | BUG_ON(!PageLocked(page)); | |
175 | BUG_ON(sis == NULL); | |
176 | if (frontswap_test(sis, offset)) | |
ef383597 | 177 | ret = frontswap_ops.load(type, offset, page); |
29f233cf | 178 | if (ret == 0) |
165c8aed | 179 | inc_frontswap_loads(); |
29f233cf DM |
180 | return ret; |
181 | } | |
165c8aed | 182 | EXPORT_SYMBOL(__frontswap_load); |
29f233cf DM |
183 | |
184 | /* | |
185 | * Invalidate any data from frontswap associated with the specified swaptype | |
186 | * and offset so that a subsequent "get" will fail. | |
187 | */ | |
188 | void __frontswap_invalidate_page(unsigned type, pgoff_t offset) | |
189 | { | |
190 | struct swap_info_struct *sis = swap_info[type]; | |
191 | ||
192 | BUG_ON(sis == NULL); | |
193 | if (frontswap_test(sis, offset)) { | |
ef383597 | 194 | frontswap_ops.invalidate_page(type, offset); |
29f233cf DM |
195 | atomic_dec(&sis->frontswap_pages); |
196 | frontswap_clear(sis, offset); | |
197 | inc_frontswap_invalidates(); | |
198 | } | |
199 | } | |
200 | EXPORT_SYMBOL(__frontswap_invalidate_page); | |
201 | ||
202 | /* | |
203 | * Invalidate all data from frontswap associated with all offsets for the | |
204 | * specified swaptype. | |
205 | */ | |
206 | void __frontswap_invalidate_area(unsigned type) | |
207 | { | |
208 | struct swap_info_struct *sis = swap_info[type]; | |
209 | ||
210 | BUG_ON(sis == NULL); | |
211 | if (sis->frontswap_map == NULL) | |
212 | return; | |
ef383597 | 213 | frontswap_ops.invalidate_area(type); |
29f233cf DM |
214 | atomic_set(&sis->frontswap_pages, 0); |
215 | memset(sis->frontswap_map, 0, sis->max / sizeof(long)); | |
216 | } | |
217 | EXPORT_SYMBOL(__frontswap_invalidate_area); | |
218 | ||
96253444 SL |
219 | static unsigned long __frontswap_curr_pages(void) |
220 | { | |
221 | int type; | |
222 | unsigned long totalpages = 0; | |
223 | struct swap_info_struct *si = NULL; | |
224 | ||
225 | assert_spin_locked(&swap_lock); | |
226 | for (type = swap_list.head; type >= 0; type = si->next) { | |
227 | si = swap_info[type]; | |
228 | totalpages += atomic_read(&si->frontswap_pages); | |
229 | } | |
230 | return totalpages; | |
231 | } | |
232 | ||
f116695a SL |
233 | static int __frontswap_unuse_pages(unsigned long total, unsigned long *unused, |
234 | int *swapid) | |
235 | { | |
236 | int ret = -EINVAL; | |
237 | struct swap_info_struct *si = NULL; | |
238 | int si_frontswap_pages; | |
239 | unsigned long total_pages_to_unuse = total; | |
240 | unsigned long pages = 0, pages_to_unuse = 0; | |
241 | int type; | |
242 | ||
243 | assert_spin_locked(&swap_lock); | |
244 | for (type = swap_list.head; type >= 0; type = si->next) { | |
245 | si = swap_info[type]; | |
246 | si_frontswap_pages = atomic_read(&si->frontswap_pages); | |
247 | if (total_pages_to_unuse < si_frontswap_pages) { | |
248 | pages = pages_to_unuse = total_pages_to_unuse; | |
249 | } else { | |
250 | pages = si_frontswap_pages; | |
251 | pages_to_unuse = 0; /* unuse all */ | |
252 | } | |
253 | /* ensure there is enough RAM to fetch pages from frontswap */ | |
254 | if (security_vm_enough_memory_mm(current->mm, pages)) { | |
255 | ret = -ENOMEM; | |
256 | continue; | |
257 | } | |
258 | vm_unacct_memory(pages); | |
259 | *unused = pages_to_unuse; | |
260 | *swapid = type; | |
261 | ret = 0; | |
262 | break; | |
263 | } | |
264 | ||
265 | return ret; | |
266 | } | |
267 | ||
69217b4c SL |
268 | static int __frontswap_shrink(unsigned long target_pages, |
269 | unsigned long *pages_to_unuse, | |
270 | int *type) | |
271 | { | |
272 | unsigned long total_pages = 0, total_pages_to_unuse; | |
273 | ||
274 | assert_spin_locked(&swap_lock); | |
275 | ||
276 | total_pages = __frontswap_curr_pages(); | |
277 | if (total_pages <= target_pages) { | |
278 | /* Nothing to do */ | |
279 | *pages_to_unuse = 0; | |
280 | return 0; | |
281 | } | |
282 | total_pages_to_unuse = total_pages - target_pages; | |
283 | return __frontswap_unuse_pages(total_pages_to_unuse, pages_to_unuse, type); | |
284 | } | |
285 | ||
29f233cf DM |
286 | /* |
287 | * Frontswap, like a true swap device, may unnecessarily retain pages | |
288 | * under certain circumstances; "shrink" frontswap is essentially a | |
289 | * "partial swapoff" and works by calling try_to_unuse to attempt to | |
290 | * unuse enough frontswap pages to attempt to -- subject to memory | |
291 | * constraints -- reduce the number of pages in frontswap to the | |
292 | * number given in the parameter target_pages. | |
293 | */ | |
294 | void frontswap_shrink(unsigned long target_pages) | |
295 | { | |
f116695a SL |
296 | unsigned long pages_to_unuse = 0; |
297 | int type, ret; | |
29f233cf DM |
298 | |
299 | /* | |
300 | * we don't want to hold swap_lock while doing a very | |
301 | * lengthy try_to_unuse, but swap_list may change | |
302 | * so restart scan from swap_list.head each time | |
303 | */ | |
304 | spin_lock(&swap_lock); | |
69217b4c | 305 | ret = __frontswap_shrink(target_pages, &pages_to_unuse, &type); |
29f233cf | 306 | spin_unlock(&swap_lock); |
69217b4c SL |
307 | if (ret == 0 && pages_to_unuse) |
308 | try_to_unuse(type, true, pages_to_unuse); | |
29f233cf DM |
309 | return; |
310 | } | |
311 | EXPORT_SYMBOL(frontswap_shrink); | |
312 | ||
313 | /* | |
314 | * Count and return the number of frontswap pages across all | |
315 | * swap devices. This is exported so that backend drivers can | |
316 | * determine current usage without reading debugfs. | |
317 | */ | |
318 | unsigned long frontswap_curr_pages(void) | |
319 | { | |
29f233cf | 320 | unsigned long totalpages = 0; |
29f233cf DM |
321 | |
322 | spin_lock(&swap_lock); | |
96253444 | 323 | totalpages = __frontswap_curr_pages(); |
29f233cf | 324 | spin_unlock(&swap_lock); |
96253444 | 325 | |
29f233cf DM |
326 | return totalpages; |
327 | } | |
328 | EXPORT_SYMBOL(frontswap_curr_pages); | |
329 | ||
330 | static int __init init_frontswap(void) | |
331 | { | |
332 | #ifdef CONFIG_DEBUG_FS | |
333 | struct dentry *root = debugfs_create_dir("frontswap", NULL); | |
334 | if (root == NULL) | |
335 | return -ENXIO; | |
165c8aed KRW |
336 | debugfs_create_u64("loads", S_IRUGO, root, &frontswap_loads); |
337 | debugfs_create_u64("succ_stores", S_IRUGO, root, &frontswap_succ_stores); | |
338 | debugfs_create_u64("failed_stores", S_IRUGO, root, | |
339 | &frontswap_failed_stores); | |
29f233cf DM |
340 | debugfs_create_u64("invalidates", S_IRUGO, |
341 | root, &frontswap_invalidates); | |
342 | #endif | |
343 | return 0; | |
344 | } | |
345 | ||
346 | module_init(init_frontswap); |