blob: 6b3e71a2cd483a70c845d33136719c43739a6c68 [file] [log] [blame]
Dan Magenheimer29f233c2012-04-09 17:09:27 -06001/*
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
Dan Magenheimer29f233c2012-04-09 17:09:27 -060014#include <linux/mman.h>
15#include <linux/swap.h>
16#include <linux/swapops.h>
Dan Magenheimer29f233c2012-04-09 17:09:27 -060017#include <linux/security.h>
Dan Magenheimer29f233c2012-04-09 17:09:27 -060018#include <linux/module.h>
Dan Magenheimer29f233c2012-04-09 17:09:27 -060019#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 */
27static 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 */
34bool frontswap_enabled __read_mostly;
35EXPORT_SYMBOL(frontswap_enabled);
36
37/*
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -040038 * If enabled, frontswap_store will return failure even on success. As
Dan Magenheimer29f233c2012-04-09 17:09:27 -060039 * 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 */
45static 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 */
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -040053static u64 frontswap_loads;
54static u64 frontswap_succ_stores;
55static u64 frontswap_failed_stores;
Dan Magenheimer29f233c2012-04-09 17:09:27 -060056static u64 frontswap_invalidates;
57
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -040058static inline void inc_frontswap_loads(void) {
59 frontswap_loads++;
Dan Magenheimer29f233c2012-04-09 17:09:27 -060060}
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -040061static inline void inc_frontswap_succ_stores(void) {
62 frontswap_succ_stores++;
Dan Magenheimer29f233c2012-04-09 17:09:27 -060063}
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -040064static inline void inc_frontswap_failed_stores(void) {
65 frontswap_failed_stores++;
Dan Magenheimer29f233c2012-04-09 17:09:27 -060066}
67static inline void inc_frontswap_invalidates(void) {
68 frontswap_invalidates++;
69}
70#else
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -040071static inline void inc_frontswap_loads(void) { }
72static inline void inc_frontswap_succ_stores(void) { }
73static inline void inc_frontswap_failed_stores(void) { }
Dan Magenheimer29f233c2012-04-09 17:09:27 -060074static 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 */
80struct 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}
88EXPORT_SYMBOL(frontswap_register_ops);
89
90/*
91 * Enable/disable frontswap writethrough (see above).
92 */
93void frontswap_writethrough(bool enable)
94{
95 frontswap_writethrough_enabled = enable;
96}
97EXPORT_SYMBOL(frontswap_writethrough);
98
99/*
100 * Called when a swap device is swapon'd.
101 */
102void __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;
Sasha Levinf9f08102012-06-10 12:51:05 +0200109 frontswap_ops.init(type);
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600110}
111EXPORT_SYMBOL(__frontswap_init);
112
Sasha Levin611edfe2012-06-10 12:51:07 +0200113static 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
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600119/*
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400120 * "Store" data from a page to frontswap and associate it with the page's
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600121 * swaptype and offset. Page must be locked and in the swap cache.
122 * If frontswap already contains a page with matching swaptype and
Wanpeng Li1d000152012-06-16 20:37:48 +0800123 * offset, the frontswap implementation may either overwrite the data and
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600124 * return success or invalidate the page from frontswap and return failure.
125 */
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400126int __frontswap_store(struct page *page)
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600127{
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;
Sasha Levinef383592012-06-10 12:50:59 +0200138 ret = frontswap_ops.store(type, offset, page);
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600139 if (ret == 0) {
140 frontswap_set(sis, offset);
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400141 inc_frontswap_succ_stores();
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600142 if (!dup)
143 atomic_inc(&sis->frontswap_pages);
Sasha Levind9674dd2012-06-10 12:51:04 +0200144 } else {
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600145 /*
146 failed dup always results in automatic invalidate of
147 the (older) page from frontswap
148 */
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400149 inc_frontswap_failed_stores();
Sasha Levin611edfe2012-06-10 12:51:07 +0200150 if (dup)
151 __frontswap_clear(sis, offset);
Sasha Levin4bb3e312012-06-10 12:51:00 +0200152 }
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600153 if (frontswap_writethrough_enabled)
154 /* report failure so swap also writes to swap device */
155 ret = -1;
156 return ret;
157}
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400158EXPORT_SYMBOL(__frontswap_store);
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600159
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 */
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400165int __frontswap_load(struct page *page)
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600166{
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))
Sasha Levinef383592012-06-10 12:50:59 +0200176 ret = frontswap_ops.load(type, offset, page);
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600177 if (ret == 0)
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400178 inc_frontswap_loads();
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600179 return ret;
180}
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400181EXPORT_SYMBOL(__frontswap_load);
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600182
183/*
184 * Invalidate any data from frontswap associated with the specified swaptype
185 * and offset so that a subsequent "get" will fail.
186 */
187void __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)) {
Sasha Levinef383592012-06-10 12:50:59 +0200193 frontswap_ops.invalidate_page(type, offset);
Sasha Levin611edfe2012-06-10 12:51:07 +0200194 __frontswap_clear(sis, offset);
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600195 inc_frontswap_invalidates();
196 }
197}
198EXPORT_SYMBOL(__frontswap_invalidate_page);
199
200/*
201 * Invalidate all data from frontswap associated with all offsets for the
202 * specified swaptype.
203 */
204void __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;
Sasha Levinef383592012-06-10 12:50:59 +0200211 frontswap_ops.invalidate_area(type);
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600212 atomic_set(&sis->frontswap_pages, 0);
213 memset(sis->frontswap_map, 0, sis->max / sizeof(long));
214}
215EXPORT_SYMBOL(__frontswap_invalidate_area);
216
Sasha Levin96253442012-06-10 12:51:01 +0200217static 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
Sasha Levinf1166952012-06-10 12:51:02 +0200231static 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
Sasha Levin69217b42012-06-10 12:51:03 +0200266static 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
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600284/*
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 */
292void frontswap_shrink(unsigned long target_pages)
293{
Sasha Levinf1166952012-06-10 12:51:02 +0200294 unsigned long pages_to_unuse = 0;
295 int type, ret;
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600296
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);
Sasha Levin69217b42012-06-10 12:51:03 +0200303 ret = __frontswap_shrink(target_pages, &pages_to_unuse, &type);
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600304 spin_unlock(&swap_lock);
Sasha Levin69217b42012-06-10 12:51:03 +0200305 if (ret == 0 && pages_to_unuse)
306 try_to_unuse(type, true, pages_to_unuse);
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600307 return;
308}
309EXPORT_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 */
316unsigned long frontswap_curr_pages(void)
317{
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600318 unsigned long totalpages = 0;
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600319
320 spin_lock(&swap_lock);
Sasha Levin96253442012-06-10 12:51:01 +0200321 totalpages = __frontswap_curr_pages();
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600322 spin_unlock(&swap_lock);
Sasha Levin96253442012-06-10 12:51:01 +0200323
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600324 return totalpages;
325}
326EXPORT_SYMBOL(frontswap_curr_pages);
327
328static 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;
Konrad Rzeszutek Wilk165c8ae2012-05-15 11:32:15 -0400334 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);
Dan Magenheimer29f233c2012-04-09 17:09:27 -0600338 debugfs_create_u64("invalidates", S_IRUGO,
339 root, &frontswap_invalidates);
340#endif
341 return 0;
342}
343
344module_init(init_frontswap);