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 | 28 | |
29f233cf | 29 | /* |
165c8aed | 30 | * If enabled, frontswap_store will return failure even on success. As |
29f233cf DM |
31 | * a result, the swap subsystem will always write the page to swap, in |
32 | * effect converting frontswap into a writethrough cache. In this mode, | |
33 | * there is no direct reduction in swap writes, but a frontswap backend | |
34 | * can unilaterally "reclaim" any pages in use with no data loss, thus | |
35 | * providing increases control over maximum memory usage due to frontswap. | |
36 | */ | |
37 | static bool frontswap_writethrough_enabled __read_mostly; | |
38 | ||
e3483a5f DM |
39 | /* |
40 | * If enabled, the underlying tmem implementation is capable of doing | |
41 | * exclusive gets, so frontswap_load, on a successful tmem_get must | |
42 | * mark the page as no longer in frontswap AND mark it dirty. | |
43 | */ | |
44 | static bool frontswap_tmem_exclusive_gets_enabled __read_mostly; | |
45 | ||
29f233cf DM |
46 | #ifdef CONFIG_DEBUG_FS |
47 | /* | |
48 | * Counters available via /sys/kernel/debug/frontswap (if debugfs is | |
49 | * properly configured). These are for information only so are not protected | |
50 | * against increment races. | |
51 | */ | |
165c8aed KRW |
52 | static u64 frontswap_loads; |
53 | static u64 frontswap_succ_stores; | |
54 | static u64 frontswap_failed_stores; | |
29f233cf DM |
55 | static u64 frontswap_invalidates; |
56 | ||
165c8aed KRW |
57 | static inline void inc_frontswap_loads(void) { |
58 | frontswap_loads++; | |
29f233cf | 59 | } |
165c8aed KRW |
60 | static inline void inc_frontswap_succ_stores(void) { |
61 | frontswap_succ_stores++; | |
29f233cf | 62 | } |
165c8aed KRW |
63 | static inline void inc_frontswap_failed_stores(void) { |
64 | frontswap_failed_stores++; | |
29f233cf DM |
65 | } |
66 | static inline void inc_frontswap_invalidates(void) { | |
67 | frontswap_invalidates++; | |
68 | } | |
69 | #else | |
165c8aed KRW |
70 | static inline void inc_frontswap_loads(void) { } |
71 | static inline void inc_frontswap_succ_stores(void) { } | |
72 | static inline void inc_frontswap_failed_stores(void) { } | |
29f233cf DM |
73 | static inline void inc_frontswap_invalidates(void) { } |
74 | #endif | |
905cd0e1 DM |
75 | |
76 | /* | |
77 | * Due to the asynchronous nature of the backends loading potentially | |
78 | * _after_ the swap system has been activated, we have chokepoints | |
79 | * on all frontswap functions to not call the backend until the backend | |
80 | * has registered. | |
81 | * | |
82 | * Specifically when no backend is registered (nobody called | |
83 | * frontswap_register_ops) all calls to frontswap_init (which is done via | |
84 | * swapon -> enable_swap_info -> frontswap_init) are registered and remembered | |
85 | * (via the setting of need_init bitmap) but fail to create tmem_pools. When a | |
86 | * backend registers with frontswap at some later point the previous | |
87 | * calls to frontswap_init are executed (by iterating over the need_init | |
88 | * bitmap) to create tmem_pools and set the respective poolids. All of that is | |
89 | * guarded by us using atomic bit operations on the 'need_init' bitmap. | |
90 | * | |
91 | * This would not guards us against the user deciding to call swapoff right as | |
92 | * we are calling the backend to initialize (so swapon is in action). | |
93 | * Fortunatly for us, the swapon_mutex has been taked by the callee so we are | |
94 | * OK. The other scenario where calls to frontswap_store (called via | |
95 | * swap_writepage) is racing with frontswap_invalidate_area (called via | |
96 | * swapoff) is again guarded by the swap subsystem. | |
97 | * | |
98 | * While no backend is registered all calls to frontswap_[store|load| | |
99 | * invalidate_area|invalidate_page] are ignored or fail. | |
100 | * | |
101 | * The time between the backend being registered and the swap file system | |
102 | * calling the backend (via the frontswap_* functions) is indeterminate as | |
1e01c968 | 103 | * frontswap_ops is not atomic_t (or a value guarded by a spinlock). |
905cd0e1 DM |
104 | * That is OK as we are comfortable missing some of these calls to the newly |
105 | * registered backend. | |
106 | * | |
107 | * Obviously the opposite (unloading the backend) must be done after all | |
108 | * the frontswap_[store|load|invalidate_area|invalidate_page] start | |
1e01c968 | 109 | * ignorning or failing the requests - at which point frontswap_ops |
905cd0e1 DM |
110 | * would have to be made in some fashion atomic. |
111 | */ | |
112 | static DECLARE_BITMAP(need_init, MAX_SWAPFILES); | |
905cd0e1 | 113 | |
29f233cf DM |
114 | /* |
115 | * Register operations for frontswap, returning previous thus allowing | |
116 | * detection of multiple backends and possible nesting. | |
117 | */ | |
1e01c968 | 118 | struct frontswap_ops *frontswap_register_ops(struct frontswap_ops *ops) |
29f233cf | 119 | { |
1e01c968 | 120 | struct frontswap_ops *old = frontswap_ops; |
905cd0e1 | 121 | int i; |
29f233cf | 122 | |
905cd0e1 DM |
123 | for (i = 0; i < MAX_SWAPFILES; i++) { |
124 | if (test_and_clear_bit(i, need_init)) | |
1e01c968 | 125 | ops->init(i); |
905cd0e1 DM |
126 | } |
127 | /* | |
1e01c968 | 128 | * We MUST have frontswap_ops set _after_ the frontswap_init's |
905cd0e1 DM |
129 | * have been called. Otherwise __frontswap_store might fail. Hence |
130 | * the barrier to make sure compiler does not re-order us. | |
131 | */ | |
132 | barrier(); | |
1e01c968 | 133 | frontswap_ops = ops; |
29f233cf DM |
134 | return old; |
135 | } | |
136 | EXPORT_SYMBOL(frontswap_register_ops); | |
137 | ||
138 | /* | |
139 | * Enable/disable frontswap writethrough (see above). | |
140 | */ | |
141 | void frontswap_writethrough(bool enable) | |
142 | { | |
143 | frontswap_writethrough_enabled = enable; | |
144 | } | |
145 | EXPORT_SYMBOL(frontswap_writethrough); | |
146 | ||
e3483a5f DM |
147 | /* |
148 | * Enable/disable frontswap exclusive gets (see above). | |
149 | */ | |
150 | void frontswap_tmem_exclusive_gets(bool enable) | |
151 | { | |
152 | frontswap_tmem_exclusive_gets_enabled = enable; | |
153 | } | |
154 | EXPORT_SYMBOL(frontswap_tmem_exclusive_gets); | |
155 | ||
29f233cf DM |
156 | /* |
157 | * Called when a swap device is swapon'd. | |
158 | */ | |
159 | void __frontswap_init(unsigned type) | |
160 | { | |
161 | struct swap_info_struct *sis = swap_info[type]; | |
162 | ||
1e01c968 | 163 | if (frontswap_ops) { |
905cd0e1 DM |
164 | BUG_ON(sis == NULL); |
165 | if (sis->frontswap_map == NULL) | |
166 | return; | |
1e01c968 | 167 | frontswap_ops->init(type); |
905cd0e1 DM |
168 | } else { |
169 | BUG_ON(type > MAX_SWAPFILES); | |
170 | set_bit(type, need_init); | |
171 | } | |
172 | ||
29f233cf DM |
173 | } |
174 | EXPORT_SYMBOL(__frontswap_init); | |
175 | ||
f066ea23 BL |
176 | bool __frontswap_test(struct swap_info_struct *sis, |
177 | pgoff_t offset) | |
178 | { | |
179 | bool ret = false; | |
180 | ||
181 | if (frontswap_ops && sis->frontswap_map) | |
182 | ret = test_bit(offset, sis->frontswap_map); | |
183 | return ret; | |
184 | } | |
185 | EXPORT_SYMBOL(__frontswap_test); | |
186 | ||
187 | static inline void __frontswap_clear(struct swap_info_struct *sis, | |
188 | pgoff_t offset) | |
611edfed | 189 | { |
f066ea23 | 190 | clear_bit(offset, sis->frontswap_map); |
611edfed SL |
191 | atomic_dec(&sis->frontswap_pages); |
192 | } | |
193 | ||
29f233cf | 194 | /* |
165c8aed | 195 | * "Store" data from a page to frontswap and associate it with the page's |
29f233cf DM |
196 | * swaptype and offset. Page must be locked and in the swap cache. |
197 | * If frontswap already contains a page with matching swaptype and | |
1d00015e | 198 | * offset, the frontswap implementation may either overwrite the data and |
29f233cf DM |
199 | * return success or invalidate the page from frontswap and return failure. |
200 | */ | |
165c8aed | 201 | int __frontswap_store(struct page *page) |
29f233cf DM |
202 | { |
203 | int ret = -1, dup = 0; | |
204 | swp_entry_t entry = { .val = page_private(page), }; | |
205 | int type = swp_type(entry); | |
206 | struct swap_info_struct *sis = swap_info[type]; | |
207 | pgoff_t offset = swp_offset(entry); | |
208 | ||
f066ea23 BL |
209 | /* |
210 | * Return if no backend registed. | |
211 | * Don't need to inc frontswap_failed_stores here. | |
212 | */ | |
213 | if (!frontswap_ops) | |
905cd0e1 | 214 | return ret; |
905cd0e1 | 215 | |
29f233cf DM |
216 | BUG_ON(!PageLocked(page)); |
217 | BUG_ON(sis == NULL); | |
f066ea23 | 218 | if (__frontswap_test(sis, offset)) |
29f233cf | 219 | dup = 1; |
1e01c968 | 220 | ret = frontswap_ops->store(type, offset, page); |
29f233cf | 221 | if (ret == 0) { |
f066ea23 | 222 | set_bit(offset, sis->frontswap_map); |
165c8aed | 223 | inc_frontswap_succ_stores(); |
29f233cf DM |
224 | if (!dup) |
225 | atomic_inc(&sis->frontswap_pages); | |
d9674dda | 226 | } else { |
29f233cf DM |
227 | /* |
228 | failed dup always results in automatic invalidate of | |
229 | the (older) page from frontswap | |
230 | */ | |
165c8aed | 231 | inc_frontswap_failed_stores(); |
611edfed SL |
232 | if (dup) |
233 | __frontswap_clear(sis, offset); | |
4bb3e31e | 234 | } |
29f233cf DM |
235 | if (frontswap_writethrough_enabled) |
236 | /* report failure so swap also writes to swap device */ | |
237 | ret = -1; | |
238 | return ret; | |
239 | } | |
165c8aed | 240 | EXPORT_SYMBOL(__frontswap_store); |
29f233cf DM |
241 | |
242 | /* | |
243 | * "Get" data from frontswap associated with swaptype and offset that were | |
244 | * specified when the data was put to frontswap and use it to fill the | |
245 | * specified page with data. Page must be locked and in the swap cache. | |
246 | */ | |
165c8aed | 247 | int __frontswap_load(struct page *page) |
29f233cf DM |
248 | { |
249 | int ret = -1; | |
250 | swp_entry_t entry = { .val = page_private(page), }; | |
251 | int type = swp_type(entry); | |
252 | struct swap_info_struct *sis = swap_info[type]; | |
253 | pgoff_t offset = swp_offset(entry); | |
254 | ||
255 | BUG_ON(!PageLocked(page)); | |
256 | BUG_ON(sis == NULL); | |
f066ea23 BL |
257 | /* |
258 | * __frontswap_test() will check whether there is backend registered | |
259 | */ | |
260 | if (__frontswap_test(sis, offset)) | |
1e01c968 | 261 | ret = frontswap_ops->load(type, offset, page); |
e3483a5f | 262 | if (ret == 0) { |
165c8aed | 263 | inc_frontswap_loads(); |
e3483a5f DM |
264 | if (frontswap_tmem_exclusive_gets_enabled) { |
265 | SetPageDirty(page); | |
f066ea23 | 266 | __frontswap_clear(sis, offset); |
e3483a5f DM |
267 | } |
268 | } | |
29f233cf DM |
269 | return ret; |
270 | } | |
165c8aed | 271 | EXPORT_SYMBOL(__frontswap_load); |
29f233cf DM |
272 | |
273 | /* | |
274 | * Invalidate any data from frontswap associated with the specified swaptype | |
275 | * and offset so that a subsequent "get" will fail. | |
276 | */ | |
277 | void __frontswap_invalidate_page(unsigned type, pgoff_t offset) | |
278 | { | |
279 | struct swap_info_struct *sis = swap_info[type]; | |
280 | ||
281 | BUG_ON(sis == NULL); | |
f066ea23 BL |
282 | /* |
283 | * __frontswap_test() will check whether there is backend registered | |
284 | */ | |
285 | if (__frontswap_test(sis, offset)) { | |
1e01c968 | 286 | frontswap_ops->invalidate_page(type, offset); |
611edfed | 287 | __frontswap_clear(sis, offset); |
29f233cf DM |
288 | inc_frontswap_invalidates(); |
289 | } | |
290 | } | |
291 | EXPORT_SYMBOL(__frontswap_invalidate_page); | |
292 | ||
293 | /* | |
294 | * Invalidate all data from frontswap associated with all offsets for the | |
295 | * specified swaptype. | |
296 | */ | |
297 | void __frontswap_invalidate_area(unsigned type) | |
298 | { | |
299 | struct swap_info_struct *sis = swap_info[type]; | |
300 | ||
1e01c968 | 301 | if (frontswap_ops) { |
905cd0e1 DM |
302 | BUG_ON(sis == NULL); |
303 | if (sis->frontswap_map == NULL) | |
304 | return; | |
1e01c968 | 305 | frontswap_ops->invalidate_area(type); |
905cd0e1 DM |
306 | atomic_set(&sis->frontswap_pages, 0); |
307 | memset(sis->frontswap_map, 0, sis->max / sizeof(long)); | |
308 | } | |
309 | clear_bit(type, need_init); | |
29f233cf DM |
310 | } |
311 | EXPORT_SYMBOL(__frontswap_invalidate_area); | |
312 | ||
96253444 SL |
313 | static unsigned long __frontswap_curr_pages(void) |
314 | { | |
315 | int type; | |
316 | unsigned long totalpages = 0; | |
317 | struct swap_info_struct *si = NULL; | |
318 | ||
319 | assert_spin_locked(&swap_lock); | |
320 | for (type = swap_list.head; type >= 0; type = si->next) { | |
321 | si = swap_info[type]; | |
322 | totalpages += atomic_read(&si->frontswap_pages); | |
323 | } | |
324 | return totalpages; | |
325 | } | |
326 | ||
f116695a SL |
327 | static int __frontswap_unuse_pages(unsigned long total, unsigned long *unused, |
328 | int *swapid) | |
329 | { | |
330 | int ret = -EINVAL; | |
331 | struct swap_info_struct *si = NULL; | |
332 | int si_frontswap_pages; | |
333 | unsigned long total_pages_to_unuse = total; | |
334 | unsigned long pages = 0, pages_to_unuse = 0; | |
335 | int type; | |
336 | ||
337 | assert_spin_locked(&swap_lock); | |
338 | for (type = swap_list.head; type >= 0; type = si->next) { | |
339 | si = swap_info[type]; | |
340 | si_frontswap_pages = atomic_read(&si->frontswap_pages); | |
341 | if (total_pages_to_unuse < si_frontswap_pages) { | |
342 | pages = pages_to_unuse = total_pages_to_unuse; | |
343 | } else { | |
344 | pages = si_frontswap_pages; | |
345 | pages_to_unuse = 0; /* unuse all */ | |
346 | } | |
347 | /* ensure there is enough RAM to fetch pages from frontswap */ | |
348 | if (security_vm_enough_memory_mm(current->mm, pages)) { | |
349 | ret = -ENOMEM; | |
350 | continue; | |
351 | } | |
352 | vm_unacct_memory(pages); | |
353 | *unused = pages_to_unuse; | |
354 | *swapid = type; | |
355 | ret = 0; | |
356 | break; | |
357 | } | |
358 | ||
359 | return ret; | |
360 | } | |
361 | ||
a00bb1e9 ZD |
362 | /* |
363 | * Used to check if it's necessory and feasible to unuse pages. | |
364 | * Return 1 when nothing to do, 0 when need to shink pages, | |
365 | * error code when there is an error. | |
366 | */ | |
69217b4c SL |
367 | static int __frontswap_shrink(unsigned long target_pages, |
368 | unsigned long *pages_to_unuse, | |
369 | int *type) | |
370 | { | |
371 | unsigned long total_pages = 0, total_pages_to_unuse; | |
372 | ||
373 | assert_spin_locked(&swap_lock); | |
374 | ||
375 | total_pages = __frontswap_curr_pages(); | |
376 | if (total_pages <= target_pages) { | |
377 | /* Nothing to do */ | |
378 | *pages_to_unuse = 0; | |
a00bb1e9 | 379 | return 1; |
69217b4c SL |
380 | } |
381 | total_pages_to_unuse = total_pages - target_pages; | |
382 | return __frontswap_unuse_pages(total_pages_to_unuse, pages_to_unuse, type); | |
383 | } | |
384 | ||
29f233cf DM |
385 | /* |
386 | * Frontswap, like a true swap device, may unnecessarily retain pages | |
387 | * under certain circumstances; "shrink" frontswap is essentially a | |
388 | * "partial swapoff" and works by calling try_to_unuse to attempt to | |
389 | * unuse enough frontswap pages to attempt to -- subject to memory | |
390 | * constraints -- reduce the number of pages in frontswap to the | |
391 | * number given in the parameter target_pages. | |
392 | */ | |
393 | void frontswap_shrink(unsigned long target_pages) | |
394 | { | |
f116695a | 395 | unsigned long pages_to_unuse = 0; |
6b982fcf | 396 | int uninitialized_var(type), ret; |
29f233cf DM |
397 | |
398 | /* | |
399 | * we don't want to hold swap_lock while doing a very | |
400 | * lengthy try_to_unuse, but swap_list may change | |
401 | * so restart scan from swap_list.head each time | |
402 | */ | |
403 | spin_lock(&swap_lock); | |
69217b4c | 404 | ret = __frontswap_shrink(target_pages, &pages_to_unuse, &type); |
29f233cf | 405 | spin_unlock(&swap_lock); |
a00bb1e9 | 406 | if (ret == 0) |
69217b4c | 407 | try_to_unuse(type, true, pages_to_unuse); |
29f233cf DM |
408 | return; |
409 | } | |
410 | EXPORT_SYMBOL(frontswap_shrink); | |
411 | ||
412 | /* | |
413 | * Count and return the number of frontswap pages across all | |
414 | * swap devices. This is exported so that backend drivers can | |
415 | * determine current usage without reading debugfs. | |
416 | */ | |
417 | unsigned long frontswap_curr_pages(void) | |
418 | { | |
29f233cf | 419 | unsigned long totalpages = 0; |
29f233cf DM |
420 | |
421 | spin_lock(&swap_lock); | |
96253444 | 422 | totalpages = __frontswap_curr_pages(); |
29f233cf | 423 | spin_unlock(&swap_lock); |
96253444 | 424 | |
29f233cf DM |
425 | return totalpages; |
426 | } | |
427 | EXPORT_SYMBOL(frontswap_curr_pages); | |
428 | ||
429 | static int __init init_frontswap(void) | |
430 | { | |
431 | #ifdef CONFIG_DEBUG_FS | |
432 | struct dentry *root = debugfs_create_dir("frontswap", NULL); | |
433 | if (root == NULL) | |
434 | return -ENXIO; | |
165c8aed KRW |
435 | debugfs_create_u64("loads", S_IRUGO, root, &frontswap_loads); |
436 | debugfs_create_u64("succ_stores", S_IRUGO, root, &frontswap_succ_stores); | |
437 | debugfs_create_u64("failed_stores", S_IRUGO, root, | |
438 | &frontswap_failed_stores); | |
29f233cf DM |
439 | debugfs_create_u64("invalidates", S_IRUGO, |
440 | root, &frontswap_invalidates); | |
441 | #endif | |
442 | return 0; | |
443 | } | |
444 | ||
445 | module_init(init_frontswap); |