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 | */ | |
1e01c968 | 27 | static struct frontswap_ops *frontswap_ops __read_mostly; |
29f233cf DM |
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 | ||
e3483a5f DM |
47 | /* |
48 | * If enabled, the underlying tmem implementation is capable of doing | |
49 | * exclusive gets, so frontswap_load, on a successful tmem_get must | |
50 | * mark the page as no longer in frontswap AND mark it dirty. | |
51 | */ | |
52 | static bool frontswap_tmem_exclusive_gets_enabled __read_mostly; | |
53 | ||
29f233cf DM |
54 | #ifdef CONFIG_DEBUG_FS |
55 | /* | |
56 | * Counters available via /sys/kernel/debug/frontswap (if debugfs is | |
57 | * properly configured). These are for information only so are not protected | |
58 | * against increment races. | |
59 | */ | |
165c8aed KRW |
60 | static u64 frontswap_loads; |
61 | static u64 frontswap_succ_stores; | |
62 | static u64 frontswap_failed_stores; | |
29f233cf DM |
63 | static u64 frontswap_invalidates; |
64 | ||
165c8aed KRW |
65 | static inline void inc_frontswap_loads(void) { |
66 | frontswap_loads++; | |
29f233cf | 67 | } |
165c8aed KRW |
68 | static inline void inc_frontswap_succ_stores(void) { |
69 | frontswap_succ_stores++; | |
29f233cf | 70 | } |
165c8aed KRW |
71 | static inline void inc_frontswap_failed_stores(void) { |
72 | frontswap_failed_stores++; | |
29f233cf DM |
73 | } |
74 | static inline void inc_frontswap_invalidates(void) { | |
75 | frontswap_invalidates++; | |
76 | } | |
77 | #else | |
165c8aed KRW |
78 | static inline void inc_frontswap_loads(void) { } |
79 | static inline void inc_frontswap_succ_stores(void) { } | |
80 | static inline void inc_frontswap_failed_stores(void) { } | |
29f233cf DM |
81 | static inline void inc_frontswap_invalidates(void) { } |
82 | #endif | |
905cd0e1 DM |
83 | |
84 | /* | |
85 | * Due to the asynchronous nature of the backends loading potentially | |
86 | * _after_ the swap system has been activated, we have chokepoints | |
87 | * on all frontswap functions to not call the backend until the backend | |
88 | * has registered. | |
89 | * | |
90 | * Specifically when no backend is registered (nobody called | |
91 | * frontswap_register_ops) all calls to frontswap_init (which is done via | |
92 | * swapon -> enable_swap_info -> frontswap_init) are registered and remembered | |
93 | * (via the setting of need_init bitmap) but fail to create tmem_pools. When a | |
94 | * backend registers with frontswap at some later point the previous | |
95 | * calls to frontswap_init are executed (by iterating over the need_init | |
96 | * bitmap) to create tmem_pools and set the respective poolids. All of that is | |
97 | * guarded by us using atomic bit operations on the 'need_init' bitmap. | |
98 | * | |
99 | * This would not guards us against the user deciding to call swapoff right as | |
100 | * we are calling the backend to initialize (so swapon is in action). | |
101 | * Fortunatly for us, the swapon_mutex has been taked by the callee so we are | |
102 | * OK. The other scenario where calls to frontswap_store (called via | |
103 | * swap_writepage) is racing with frontswap_invalidate_area (called via | |
104 | * swapoff) is again guarded by the swap subsystem. | |
105 | * | |
106 | * While no backend is registered all calls to frontswap_[store|load| | |
107 | * invalidate_area|invalidate_page] are ignored or fail. | |
108 | * | |
109 | * The time between the backend being registered and the swap file system | |
110 | * calling the backend (via the frontswap_* functions) is indeterminate as | |
1e01c968 | 111 | * frontswap_ops is not atomic_t (or a value guarded by a spinlock). |
905cd0e1 DM |
112 | * That is OK as we are comfortable missing some of these calls to the newly |
113 | * registered backend. | |
114 | * | |
115 | * Obviously the opposite (unloading the backend) must be done after all | |
116 | * the frontswap_[store|load|invalidate_area|invalidate_page] start | |
1e01c968 | 117 | * ignorning or failing the requests - at which point frontswap_ops |
905cd0e1 DM |
118 | * would have to be made in some fashion atomic. |
119 | */ | |
120 | static DECLARE_BITMAP(need_init, MAX_SWAPFILES); | |
905cd0e1 | 121 | |
29f233cf DM |
122 | /* |
123 | * Register operations for frontswap, returning previous thus allowing | |
124 | * detection of multiple backends and possible nesting. | |
125 | */ | |
1e01c968 | 126 | struct frontswap_ops *frontswap_register_ops(struct frontswap_ops *ops) |
29f233cf | 127 | { |
1e01c968 | 128 | struct frontswap_ops *old = frontswap_ops; |
905cd0e1 | 129 | int i; |
29f233cf | 130 | |
29f233cf | 131 | frontswap_enabled = true; |
905cd0e1 DM |
132 | |
133 | for (i = 0; i < MAX_SWAPFILES; i++) { | |
134 | if (test_and_clear_bit(i, need_init)) | |
1e01c968 | 135 | ops->init(i); |
905cd0e1 DM |
136 | } |
137 | /* | |
1e01c968 | 138 | * We MUST have frontswap_ops set _after_ the frontswap_init's |
905cd0e1 DM |
139 | * have been called. Otherwise __frontswap_store might fail. Hence |
140 | * the barrier to make sure compiler does not re-order us. | |
141 | */ | |
142 | barrier(); | |
1e01c968 | 143 | frontswap_ops = ops; |
29f233cf DM |
144 | return old; |
145 | } | |
146 | EXPORT_SYMBOL(frontswap_register_ops); | |
147 | ||
148 | /* | |
149 | * Enable/disable frontswap writethrough (see above). | |
150 | */ | |
151 | void frontswap_writethrough(bool enable) | |
152 | { | |
153 | frontswap_writethrough_enabled = enable; | |
154 | } | |
155 | EXPORT_SYMBOL(frontswap_writethrough); | |
156 | ||
e3483a5f DM |
157 | /* |
158 | * Enable/disable frontswap exclusive gets (see above). | |
159 | */ | |
160 | void frontswap_tmem_exclusive_gets(bool enable) | |
161 | { | |
162 | frontswap_tmem_exclusive_gets_enabled = enable; | |
163 | } | |
164 | EXPORT_SYMBOL(frontswap_tmem_exclusive_gets); | |
165 | ||
29f233cf DM |
166 | /* |
167 | * Called when a swap device is swapon'd. | |
168 | */ | |
169 | void __frontswap_init(unsigned type) | |
170 | { | |
171 | struct swap_info_struct *sis = swap_info[type]; | |
172 | ||
1e01c968 | 173 | if (frontswap_ops) { |
905cd0e1 DM |
174 | BUG_ON(sis == NULL); |
175 | if (sis->frontswap_map == NULL) | |
176 | return; | |
1e01c968 | 177 | frontswap_ops->init(type); |
905cd0e1 DM |
178 | } else { |
179 | BUG_ON(type > MAX_SWAPFILES); | |
180 | set_bit(type, need_init); | |
181 | } | |
182 | ||
29f233cf DM |
183 | } |
184 | EXPORT_SYMBOL(__frontswap_init); | |
185 | ||
611edfed SL |
186 | static inline void __frontswap_clear(struct swap_info_struct *sis, pgoff_t offset) |
187 | { | |
188 | frontswap_clear(sis, offset); | |
189 | atomic_dec(&sis->frontswap_pages); | |
190 | } | |
191 | ||
29f233cf | 192 | /* |
165c8aed | 193 | * "Store" data from a page to frontswap and associate it with the page's |
29f233cf DM |
194 | * swaptype and offset. Page must be locked and in the swap cache. |
195 | * If frontswap already contains a page with matching swaptype and | |
1d00015e | 196 | * offset, the frontswap implementation may either overwrite the data and |
29f233cf DM |
197 | * return success or invalidate the page from frontswap and return failure. |
198 | */ | |
165c8aed | 199 | int __frontswap_store(struct page *page) |
29f233cf DM |
200 | { |
201 | int ret = -1, dup = 0; | |
202 | swp_entry_t entry = { .val = page_private(page), }; | |
203 | int type = swp_type(entry); | |
204 | struct swap_info_struct *sis = swap_info[type]; | |
205 | pgoff_t offset = swp_offset(entry); | |
206 | ||
1e01c968 | 207 | if (!frontswap_ops) { |
905cd0e1 DM |
208 | inc_frontswap_failed_stores(); |
209 | return ret; | |
210 | } | |
211 | ||
29f233cf DM |
212 | BUG_ON(!PageLocked(page)); |
213 | BUG_ON(sis == NULL); | |
214 | if (frontswap_test(sis, offset)) | |
215 | dup = 1; | |
1e01c968 | 216 | ret = frontswap_ops->store(type, offset, page); |
29f233cf DM |
217 | if (ret == 0) { |
218 | frontswap_set(sis, offset); | |
165c8aed | 219 | inc_frontswap_succ_stores(); |
29f233cf DM |
220 | if (!dup) |
221 | atomic_inc(&sis->frontswap_pages); | |
d9674dda | 222 | } else { |
29f233cf DM |
223 | /* |
224 | failed dup always results in automatic invalidate of | |
225 | the (older) page from frontswap | |
226 | */ | |
165c8aed | 227 | inc_frontswap_failed_stores(); |
611edfed SL |
228 | if (dup) |
229 | __frontswap_clear(sis, offset); | |
4bb3e31e | 230 | } |
29f233cf DM |
231 | if (frontswap_writethrough_enabled) |
232 | /* report failure so swap also writes to swap device */ | |
233 | ret = -1; | |
234 | return ret; | |
235 | } | |
165c8aed | 236 | EXPORT_SYMBOL(__frontswap_store); |
29f233cf DM |
237 | |
238 | /* | |
239 | * "Get" data from frontswap associated with swaptype and offset that were | |
240 | * specified when the data was put to frontswap and use it to fill the | |
241 | * specified page with data. Page must be locked and in the swap cache. | |
242 | */ | |
165c8aed | 243 | int __frontswap_load(struct page *page) |
29f233cf DM |
244 | { |
245 | int ret = -1; | |
246 | swp_entry_t entry = { .val = page_private(page), }; | |
247 | int type = swp_type(entry); | |
248 | struct swap_info_struct *sis = swap_info[type]; | |
249 | pgoff_t offset = swp_offset(entry); | |
250 | ||
1e01c968 | 251 | if (!frontswap_ops) |
905cd0e1 DM |
252 | return ret; |
253 | ||
29f233cf DM |
254 | BUG_ON(!PageLocked(page)); |
255 | BUG_ON(sis == NULL); | |
256 | if (frontswap_test(sis, offset)) | |
1e01c968 | 257 | ret = frontswap_ops->load(type, offset, page); |
e3483a5f | 258 | if (ret == 0) { |
165c8aed | 259 | inc_frontswap_loads(); |
e3483a5f DM |
260 | if (frontswap_tmem_exclusive_gets_enabled) { |
261 | SetPageDirty(page); | |
262 | frontswap_clear(sis, offset); | |
263 | } | |
264 | } | |
29f233cf DM |
265 | return ret; |
266 | } | |
165c8aed | 267 | EXPORT_SYMBOL(__frontswap_load); |
29f233cf DM |
268 | |
269 | /* | |
270 | * Invalidate any data from frontswap associated with the specified swaptype | |
271 | * and offset so that a subsequent "get" will fail. | |
272 | */ | |
273 | void __frontswap_invalidate_page(unsigned type, pgoff_t offset) | |
274 | { | |
275 | struct swap_info_struct *sis = swap_info[type]; | |
276 | ||
1e01c968 | 277 | if (!frontswap_ops) |
905cd0e1 DM |
278 | return; |
279 | ||
29f233cf DM |
280 | BUG_ON(sis == NULL); |
281 | if (frontswap_test(sis, offset)) { | |
1e01c968 | 282 | frontswap_ops->invalidate_page(type, offset); |
611edfed | 283 | __frontswap_clear(sis, offset); |
29f233cf DM |
284 | inc_frontswap_invalidates(); |
285 | } | |
286 | } | |
287 | EXPORT_SYMBOL(__frontswap_invalidate_page); | |
288 | ||
289 | /* | |
290 | * Invalidate all data from frontswap associated with all offsets for the | |
291 | * specified swaptype. | |
292 | */ | |
293 | void __frontswap_invalidate_area(unsigned type) | |
294 | { | |
295 | struct swap_info_struct *sis = swap_info[type]; | |
296 | ||
1e01c968 | 297 | if (frontswap_ops) { |
905cd0e1 DM |
298 | BUG_ON(sis == NULL); |
299 | if (sis->frontswap_map == NULL) | |
300 | return; | |
1e01c968 | 301 | frontswap_ops->invalidate_area(type); |
905cd0e1 DM |
302 | atomic_set(&sis->frontswap_pages, 0); |
303 | memset(sis->frontswap_map, 0, sis->max / sizeof(long)); | |
304 | } | |
305 | clear_bit(type, need_init); | |
29f233cf DM |
306 | } |
307 | EXPORT_SYMBOL(__frontswap_invalidate_area); | |
308 | ||
96253444 SL |
309 | static unsigned long __frontswap_curr_pages(void) |
310 | { | |
311 | int type; | |
312 | unsigned long totalpages = 0; | |
313 | struct swap_info_struct *si = NULL; | |
314 | ||
315 | assert_spin_locked(&swap_lock); | |
316 | for (type = swap_list.head; type >= 0; type = si->next) { | |
317 | si = swap_info[type]; | |
318 | totalpages += atomic_read(&si->frontswap_pages); | |
319 | } | |
320 | return totalpages; | |
321 | } | |
322 | ||
f116695a SL |
323 | static int __frontswap_unuse_pages(unsigned long total, unsigned long *unused, |
324 | int *swapid) | |
325 | { | |
326 | int ret = -EINVAL; | |
327 | struct swap_info_struct *si = NULL; | |
328 | int si_frontswap_pages; | |
329 | unsigned long total_pages_to_unuse = total; | |
330 | unsigned long pages = 0, pages_to_unuse = 0; | |
331 | int type; | |
332 | ||
333 | assert_spin_locked(&swap_lock); | |
334 | for (type = swap_list.head; type >= 0; type = si->next) { | |
335 | si = swap_info[type]; | |
336 | si_frontswap_pages = atomic_read(&si->frontswap_pages); | |
337 | if (total_pages_to_unuse < si_frontswap_pages) { | |
338 | pages = pages_to_unuse = total_pages_to_unuse; | |
339 | } else { | |
340 | pages = si_frontswap_pages; | |
341 | pages_to_unuse = 0; /* unuse all */ | |
342 | } | |
343 | /* ensure there is enough RAM to fetch pages from frontswap */ | |
344 | if (security_vm_enough_memory_mm(current->mm, pages)) { | |
345 | ret = -ENOMEM; | |
346 | continue; | |
347 | } | |
348 | vm_unacct_memory(pages); | |
349 | *unused = pages_to_unuse; | |
350 | *swapid = type; | |
351 | ret = 0; | |
352 | break; | |
353 | } | |
354 | ||
355 | return ret; | |
356 | } | |
357 | ||
a00bb1e9 ZD |
358 | /* |
359 | * Used to check if it's necessory and feasible to unuse pages. | |
360 | * Return 1 when nothing to do, 0 when need to shink pages, | |
361 | * error code when there is an error. | |
362 | */ | |
69217b4c SL |
363 | static int __frontswap_shrink(unsigned long target_pages, |
364 | unsigned long *pages_to_unuse, | |
365 | int *type) | |
366 | { | |
367 | unsigned long total_pages = 0, total_pages_to_unuse; | |
368 | ||
369 | assert_spin_locked(&swap_lock); | |
370 | ||
371 | total_pages = __frontswap_curr_pages(); | |
372 | if (total_pages <= target_pages) { | |
373 | /* Nothing to do */ | |
374 | *pages_to_unuse = 0; | |
a00bb1e9 | 375 | return 1; |
69217b4c SL |
376 | } |
377 | total_pages_to_unuse = total_pages - target_pages; | |
378 | return __frontswap_unuse_pages(total_pages_to_unuse, pages_to_unuse, type); | |
379 | } | |
380 | ||
29f233cf DM |
381 | /* |
382 | * Frontswap, like a true swap device, may unnecessarily retain pages | |
383 | * under certain circumstances; "shrink" frontswap is essentially a | |
384 | * "partial swapoff" and works by calling try_to_unuse to attempt to | |
385 | * unuse enough frontswap pages to attempt to -- subject to memory | |
386 | * constraints -- reduce the number of pages in frontswap to the | |
387 | * number given in the parameter target_pages. | |
388 | */ | |
389 | void frontswap_shrink(unsigned long target_pages) | |
390 | { | |
f116695a | 391 | unsigned long pages_to_unuse = 0; |
6b982fcf | 392 | int uninitialized_var(type), ret; |
29f233cf DM |
393 | |
394 | /* | |
395 | * we don't want to hold swap_lock while doing a very | |
396 | * lengthy try_to_unuse, but swap_list may change | |
397 | * so restart scan from swap_list.head each time | |
398 | */ | |
399 | spin_lock(&swap_lock); | |
69217b4c | 400 | ret = __frontswap_shrink(target_pages, &pages_to_unuse, &type); |
29f233cf | 401 | spin_unlock(&swap_lock); |
a00bb1e9 | 402 | if (ret == 0) |
69217b4c | 403 | try_to_unuse(type, true, pages_to_unuse); |
29f233cf DM |
404 | return; |
405 | } | |
406 | EXPORT_SYMBOL(frontswap_shrink); | |
407 | ||
408 | /* | |
409 | * Count and return the number of frontswap pages across all | |
410 | * swap devices. This is exported so that backend drivers can | |
411 | * determine current usage without reading debugfs. | |
412 | */ | |
413 | unsigned long frontswap_curr_pages(void) | |
414 | { | |
29f233cf | 415 | unsigned long totalpages = 0; |
29f233cf DM |
416 | |
417 | spin_lock(&swap_lock); | |
96253444 | 418 | totalpages = __frontswap_curr_pages(); |
29f233cf | 419 | spin_unlock(&swap_lock); |
96253444 | 420 | |
29f233cf DM |
421 | return totalpages; |
422 | } | |
423 | EXPORT_SYMBOL(frontswap_curr_pages); | |
424 | ||
425 | static int __init init_frontswap(void) | |
426 | { | |
427 | #ifdef CONFIG_DEBUG_FS | |
428 | struct dentry *root = debugfs_create_dir("frontswap", NULL); | |
429 | if (root == NULL) | |
430 | return -ENXIO; | |
165c8aed KRW |
431 | debugfs_create_u64("loads", S_IRUGO, root, &frontswap_loads); |
432 | debugfs_create_u64("succ_stores", S_IRUGO, root, &frontswap_succ_stores); | |
433 | debugfs_create_u64("failed_stores", S_IRUGO, root, | |
434 | &frontswap_failed_stores); | |
29f233cf DM |
435 | debugfs_create_u64("invalidates", S_IRUGO, |
436 | root, &frontswap_invalidates); | |
437 | #endif | |
905cd0e1 | 438 | frontswap_enabled = 1; |
29f233cf DM |
439 | return 0; |
440 | } | |
441 | ||
442 | module_init(init_frontswap); |