Merge branch 'akpm'
[linux-2.6/next.git] / mm / frontswap.c
blobb8aebcadb52881c5997d9eed2ccbb56d11390b5d
1 /*
2 * Frontswap frontend
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.
8 * Copyright (C) 2009-2010 Oracle Corp. All rights reserved.
9 * Author: Dan Magenheimer
11 * This work is licensed under the terms of the GNU GPL, version 2.
14 #include <linux/mm.h>
15 #include <linux/mman.h>
16 #include <linux/sysctl.h>
17 #include <linux/swap.h>
18 #include <linux/swapops.h>
19 #include <linux/proc_fs.h>
20 #include <linux/security.h>
21 #include <linux/capability.h>
22 #include <linux/module.h>
23 #include <linux/uaccess.h>
24 #include <linux/frontswap.h>
25 #include <linux/swapfile.h>
28 * frontswap_ops is set by frontswap_register_ops to contain the pointers
29 * to the frontswap "backend" implementation functions.
31 static struct frontswap_ops frontswap_ops;
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.
38 int frontswap_enabled;
39 EXPORT_SYMBOL(frontswap_enabled);
42 * Useful stats available in /sys/kernel/mm/frontswap. These are for
43 * information only so are not protected against increment/decrement races.
45 static unsigned long frontswap_gets;
46 static unsigned long frontswap_succ_puts;
47 static unsigned long frontswap_failed_puts;
48 static unsigned long frontswap_flushes;
51 * Register operations for frontswap, returning previous thus allowing
52 * detection of multiple backends and possible nesting
54 struct frontswap_ops frontswap_register_ops(struct frontswap_ops *ops)
56 struct frontswap_ops old = frontswap_ops;
58 frontswap_ops = *ops;
59 frontswap_enabled = 1;
60 return old;
62 EXPORT_SYMBOL(frontswap_register_ops);
64 /* Called when a swap device is swapon'd */
65 void __frontswap_init(unsigned type)
67 struct swap_info_struct *sis = swap_info[type];
69 BUG_ON(sis == NULL);
70 if (sis->frontswap_map == NULL)
71 return;
72 if (frontswap_enabled)
73 (*frontswap_ops.init)(type);
75 EXPORT_SYMBOL(__frontswap_init);
78 * "Put" data from a page to frontswap and associate it with the page's
79 * swaptype and offset. Page must be locked and in the swap cache.
80 * If frontswap already contains a page with matching swaptype and
81 * offset, the frontswap implmentation may either overwrite the data
82 * and return success or flush the page from frontswap and return failure
84 int __frontswap_put_page(struct page *page)
86 int ret = -1, dup = 0;
87 swp_entry_t entry = { .val = page_private(page), };
88 int type = swp_type(entry);
89 struct swap_info_struct *sis = swap_info[type];
90 pgoff_t offset = swp_offset(entry);
92 BUG_ON(!PageLocked(page));
93 BUG_ON(sis == NULL);
94 if (frontswap_test(sis, offset))
95 dup = 1;
96 ret = (*frontswap_ops.put_page)(type, offset, page);
97 if (ret == 0) {
98 frontswap_set(sis, offset);
99 frontswap_succ_puts++;
100 if (!dup)
101 atomic_inc(&sis->frontswap_pages);
102 } else if (dup) {
104 failed dup always results in automatic flush of
105 the (older) page from frontswap
107 frontswap_clear(sis, offset);
108 atomic_dec(&sis->frontswap_pages);
109 frontswap_failed_puts++;
110 } else
111 frontswap_failed_puts++;
112 return ret;
114 EXPORT_SYMBOL(__frontswap_put_page);
117 * "Get" data from frontswap associated with swaptype and offset that were
118 * specified when the data was put to frontswap and use it to fill the
119 * specified page with data. Page must be locked and in the swap cache
121 int __frontswap_get_page(struct page *page)
123 int ret = -1;
124 swp_entry_t entry = { .val = page_private(page), };
125 int type = swp_type(entry);
126 struct swap_info_struct *sis = swap_info[type];
127 pgoff_t offset = swp_offset(entry);
129 BUG_ON(!PageLocked(page));
130 BUG_ON(sis == NULL);
131 if (frontswap_test(sis, offset))
132 ret = (*frontswap_ops.get_page)(type, offset, page);
133 if (ret == 0)
134 frontswap_gets++;
135 return ret;
137 EXPORT_SYMBOL(__frontswap_get_page);
140 * Flush any data from frontswap associated with the specified swaptype
141 * and offset so that a subsequent "get" will fail.
143 void __frontswap_flush_page(unsigned type, pgoff_t offset)
145 struct swap_info_struct *sis = swap_info[type];
147 BUG_ON(sis == NULL);
148 if (frontswap_test(sis, offset)) {
149 (*frontswap_ops.flush_page)(type, offset);
150 atomic_dec(&sis->frontswap_pages);
151 frontswap_clear(sis, offset);
152 frontswap_flushes++;
155 EXPORT_SYMBOL(__frontswap_flush_page);
158 * Flush all data from frontswap associated with all offsets for the
159 * specified swaptype.
161 void __frontswap_flush_area(unsigned type)
163 struct swap_info_struct *sis = swap_info[type];
165 BUG_ON(sis == NULL);
166 if (sis->frontswap_map == NULL)
167 return;
168 (*frontswap_ops.flush_area)(type);
169 atomic_set(&sis->frontswap_pages, 0);
170 memset(sis->frontswap_map, 0, sis->max / sizeof(long));
172 EXPORT_SYMBOL(__frontswap_flush_area);
175 * Frontswap, like a true swap device, may unnecessarily retain pages
176 * under certain circumstances; "shrink" frontswap is essentially a
177 * "partial swapoff" and works by calling try_to_unuse to attempt to
178 * unuse enough frontswap pages to attempt to -- subject to memory
179 * constraints -- reduce the number of pages in frontswap
181 void frontswap_shrink(unsigned long target_pages)
183 int wrapped = 0;
184 bool locked = false;
186 /* try a few times to maximize chance of try_to_unuse success */
187 for (wrapped = 0; wrapped < 3; wrapped++) {
189 struct swap_info_struct *si = NULL;
190 int si_frontswap_pages;
191 unsigned long total_pages = 0, total_pages_to_unuse;
192 unsigned long pages = 0, pages_to_unuse = 0;
193 int type;
196 * we don't want to hold swap_lock while doing a very
197 * lengthy try_to_unuse, but swap_list may change
198 * so restart scan from swap_list.head each time
200 spin_lock(&swap_lock);
201 locked = true;
202 total_pages = 0;
203 for (type = swap_list.head; type >= 0; type = si->next) {
204 si = swap_info[type];
205 total_pages += atomic_read(&si->frontswap_pages);
207 if (total_pages <= target_pages)
208 goto out;
209 total_pages_to_unuse = total_pages - target_pages;
210 for (type = swap_list.head; type >= 0; type = si->next) {
211 si = swap_info[type];
212 si_frontswap_pages = atomic_read(&si->frontswap_pages);
213 if (total_pages_to_unuse < si_frontswap_pages)
214 pages = pages_to_unuse = total_pages_to_unuse;
215 else {
216 pages = si_frontswap_pages;
217 pages_to_unuse = 0; /* unuse all */
219 if (security_vm_enough_memory_kern(pages))
220 continue;
221 vm_unacct_memory(pages);
222 break;
224 if (type < 0)
225 goto out;
226 locked = false;
227 spin_unlock(&swap_lock);
228 try_to_unuse(type, true, pages_to_unuse);
231 out:
232 if (locked)
233 spin_unlock(&swap_lock);
234 return;
236 EXPORT_SYMBOL(frontswap_shrink);
239 * Count and return the number of pages frontswap pages across all
240 * swap devices. This is exported so that a kernel module can
241 * determine current usage without reading sysfs.
243 unsigned long frontswap_curr_pages(void)
245 int type;
246 unsigned long totalpages = 0;
247 struct swap_info_struct *si = NULL;
249 spin_lock(&swap_lock);
250 for (type = swap_list.head; type >= 0; type = si->next) {
251 si = swap_info[type];
252 if (si != NULL)
253 totalpages += atomic_read(&si->frontswap_pages);
255 spin_unlock(&swap_lock);
256 return totalpages;
258 EXPORT_SYMBOL(frontswap_curr_pages);
260 #ifdef CONFIG_SYSFS
262 /* see Documentation/ABI/xxx/sysfs-kernel-mm-frontswap */
264 #define FRONTSWAP_ATTR_RO(_name) \
265 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)
266 #define FRONTSWAP_ATTR(_name) \
267 static struct kobj_attribute _name##_attr = \
268 __ATTR(_name, 0644, _name##_show, _name##_store)
270 static ssize_t curr_pages_show(struct kobject *kobj,
271 struct kobj_attribute *attr, char *buf)
273 return sprintf(buf, "%lu\n", frontswap_curr_pages());
276 static ssize_t curr_pages_store(struct kobject *kobj,
277 struct kobj_attribute *attr,
278 const char *buf, size_t count)
280 unsigned long target_pages;
282 if (strict_strtoul(buf, 10, &target_pages))
283 return -EINVAL;
285 frontswap_shrink(target_pages);
287 return count;
289 FRONTSWAP_ATTR(curr_pages);
291 static ssize_t succ_puts_show(struct kobject *kobj,
292 struct kobj_attribute *attr, char *buf)
294 return sprintf(buf, "%lu\n", frontswap_succ_puts);
296 FRONTSWAP_ATTR_RO(succ_puts);
298 static ssize_t failed_puts_show(struct kobject *kobj,
299 struct kobj_attribute *attr, char *buf)
301 return sprintf(buf, "%lu\n", frontswap_failed_puts);
303 FRONTSWAP_ATTR_RO(failed_puts);
305 static ssize_t gets_show(struct kobject *kobj,
306 struct kobj_attribute *attr, char *buf)
308 return sprintf(buf, "%lu\n", frontswap_gets);
310 FRONTSWAP_ATTR_RO(gets);
312 static ssize_t flushes_show(struct kobject *kobj,
313 struct kobj_attribute *attr, char *buf)
315 return sprintf(buf, "%lu\n", frontswap_flushes);
317 FRONTSWAP_ATTR_RO(flushes);
319 static struct attribute *frontswap_attrs[] = {
320 &curr_pages_attr.attr,
321 &succ_puts_attr.attr,
322 &failed_puts_attr.attr,
323 &gets_attr.attr,
324 &flushes_attr.attr,
325 NULL,
328 static struct attribute_group frontswap_attr_group = {
329 .attrs = frontswap_attrs,
330 .name = "frontswap",
333 #endif /* CONFIG_SYSFS */
335 static int __init init_frontswap(void)
337 int err = 0;
339 #ifdef CONFIG_SYSFS
340 err = sysfs_create_group(mm_kobj, &frontswap_attr_group);
341 #endif /* CONFIG_SYSFS */
342 return err;
345 static void __exit exit_frontswap(void)
347 frontswap_shrink(0UL);
350 module_init(init_frontswap);
351 module_exit(exit_frontswap);