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) 2009-2012 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/mman.h>
15 #include <linux/swap.h>
16 #include <linux/swapops.h>
17 #include <linux/security.h>
18 #include <linux/module.h>
19 #include <linux/debugfs.h>
20 #include <linux/frontswap.h>
21 #include <linux/swapfile.h>
24 * frontswap_ops is set by frontswap_register_ops to contain the pointers
25 * to the frontswap "backend" implementation functions.
27 static struct frontswap_ops frontswap_ops __read_mostly;
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.
34 bool frontswap_enabled __read_mostly;
35 EXPORT_SYMBOL(frontswap_enabled);
38 * If enabled, frontswap_store will return failure even on success. As
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.
45 static bool frontswap_writethrough_enabled __read_mostly;
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.
52 static bool frontswap_tmem_exclusive_gets_enabled __read_mostly;
54 #ifdef CONFIG_DEBUG_FS
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.
60 static u64 frontswap_loads;
61 static u64 frontswap_succ_stores;
62 static u64 frontswap_failed_stores;
63 static u64 frontswap_invalidates;
65 static inline void inc_frontswap_loads(void) {
68 static inline void inc_frontswap_succ_stores(void) {
69 frontswap_succ_stores++;
71 static inline void inc_frontswap_failed_stores(void) {
72 frontswap_failed_stores++;
74 static inline void inc_frontswap_invalidates(void) {
75 frontswap_invalidates++;
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) { }
81 static inline void inc_frontswap_invalidates(void) { }
84 * Register operations for frontswap, returning previous thus allowing
85 * detection of multiple backends and possible nesting.
87 struct frontswap_ops frontswap_register_ops(struct frontswap_ops *ops)
89 struct frontswap_ops old = frontswap_ops;
92 frontswap_enabled = true;
95 EXPORT_SYMBOL(frontswap_register_ops);
98 * Enable/disable frontswap writethrough (see above).
100 void frontswap_writethrough(bool enable)
102 frontswap_writethrough_enabled = enable;
104 EXPORT_SYMBOL(frontswap_writethrough);
107 * Enable/disable frontswap exclusive gets (see above).
109 void frontswap_tmem_exclusive_gets(bool enable)
111 frontswap_tmem_exclusive_gets_enabled = enable;
113 EXPORT_SYMBOL(frontswap_tmem_exclusive_gets);
116 * Called when a swap device is swapon'd.
118 void __frontswap_init(unsigned type)
120 struct swap_info_struct *sis = swap_info[type];
123 if (sis->frontswap_map == NULL)
125 frontswap_ops.init(type);
127 EXPORT_SYMBOL(__frontswap_init);
129 static inline void __frontswap_clear(struct swap_info_struct *sis, pgoff_t offset)
131 frontswap_clear(sis, offset);
132 atomic_dec(&sis->frontswap_pages);
136 * "Store" data from a page to frontswap and associate it with the page's
137 * swaptype and offset. Page must be locked and in the swap cache.
138 * If frontswap already contains a page with matching swaptype and
139 * offset, the frontswap implementation may either overwrite the data and
140 * return success or invalidate the page from frontswap and return failure.
142 int __frontswap_store(struct page *page)
144 int ret = -1, dup = 0;
145 swp_entry_t entry = { .val = page_private(page), };
146 int type = swp_type(entry);
147 struct swap_info_struct *sis = swap_info[type];
148 pgoff_t offset = swp_offset(entry);
150 BUG_ON(!PageLocked(page));
152 if (frontswap_test(sis, offset))
154 ret = frontswap_ops.store(type, offset, page);
156 frontswap_set(sis, offset);
157 inc_frontswap_succ_stores();
159 atomic_inc(&sis->frontswap_pages);
162 failed dup always results in automatic invalidate of
163 the (older) page from frontswap
165 inc_frontswap_failed_stores();
167 __frontswap_clear(sis, offset);
169 if (frontswap_writethrough_enabled)
170 /* report failure so swap also writes to swap device */
174 EXPORT_SYMBOL(__frontswap_store);
177 * "Get" data from frontswap associated with swaptype and offset that were
178 * specified when the data was put to frontswap and use it to fill the
179 * specified page with data. Page must be locked and in the swap cache.
181 int __frontswap_load(struct page *page)
184 swp_entry_t entry = { .val = page_private(page), };
185 int type = swp_type(entry);
186 struct swap_info_struct *sis = swap_info[type];
187 pgoff_t offset = swp_offset(entry);
189 BUG_ON(!PageLocked(page));
191 if (frontswap_test(sis, offset))
192 ret = frontswap_ops.load(type, offset, page);
194 inc_frontswap_loads();
195 if (frontswap_tmem_exclusive_gets_enabled) {
197 frontswap_clear(sis, offset);
202 EXPORT_SYMBOL(__frontswap_load);
205 * Invalidate any data from frontswap associated with the specified swaptype
206 * and offset so that a subsequent "get" will fail.
208 void __frontswap_invalidate_page(unsigned type, pgoff_t offset)
210 struct swap_info_struct *sis = swap_info[type];
213 if (frontswap_test(sis, offset)) {
214 frontswap_ops.invalidate_page(type, offset);
215 __frontswap_clear(sis, offset);
216 inc_frontswap_invalidates();
219 EXPORT_SYMBOL(__frontswap_invalidate_page);
222 * Invalidate all data from frontswap associated with all offsets for the
223 * specified swaptype.
225 void __frontswap_invalidate_area(unsigned type)
227 struct swap_info_struct *sis = swap_info[type];
230 if (sis->frontswap_map == NULL)
232 frontswap_ops.invalidate_area(type);
233 atomic_set(&sis->frontswap_pages, 0);
234 memset(sis->frontswap_map, 0, sis->max / sizeof(long));
236 EXPORT_SYMBOL(__frontswap_invalidate_area);
238 static unsigned long __frontswap_curr_pages(void)
241 unsigned long totalpages = 0;
242 struct swap_info_struct *si = NULL;
244 assert_spin_locked(&swap_lock);
245 for (type = swap_list.head; type >= 0; type = si->next) {
246 si = swap_info[type];
247 totalpages += atomic_read(&si->frontswap_pages);
252 static int __frontswap_unuse_pages(unsigned long total, unsigned long *unused,
256 struct swap_info_struct *si = NULL;
257 int si_frontswap_pages;
258 unsigned long total_pages_to_unuse = total;
259 unsigned long pages = 0, pages_to_unuse = 0;
262 assert_spin_locked(&swap_lock);
263 for (type = swap_list.head; type >= 0; type = si->next) {
264 si = swap_info[type];
265 si_frontswap_pages = atomic_read(&si->frontswap_pages);
266 if (total_pages_to_unuse < si_frontswap_pages) {
267 pages = pages_to_unuse = total_pages_to_unuse;
269 pages = si_frontswap_pages;
270 pages_to_unuse = 0; /* unuse all */
272 /* ensure there is enough RAM to fetch pages from frontswap */
273 if (security_vm_enough_memory_mm(current->mm, pages)) {
277 vm_unacct_memory(pages);
278 *unused = pages_to_unuse;
288 * Used to check if it's necessory and feasible to unuse pages.
289 * Return 1 when nothing to do, 0 when need to shink pages,
290 * error code when there is an error.
292 static int __frontswap_shrink(unsigned long target_pages,
293 unsigned long *pages_to_unuse,
296 unsigned long total_pages = 0, total_pages_to_unuse;
298 assert_spin_locked(&swap_lock);
300 total_pages = __frontswap_curr_pages();
301 if (total_pages <= target_pages) {
306 total_pages_to_unuse = total_pages - target_pages;
307 return __frontswap_unuse_pages(total_pages_to_unuse, pages_to_unuse, type);
311 * Frontswap, like a true swap device, may unnecessarily retain pages
312 * under certain circumstances; "shrink" frontswap is essentially a
313 * "partial swapoff" and works by calling try_to_unuse to attempt to
314 * unuse enough frontswap pages to attempt to -- subject to memory
315 * constraints -- reduce the number of pages in frontswap to the
316 * number given in the parameter target_pages.
318 void frontswap_shrink(unsigned long target_pages)
320 unsigned long pages_to_unuse = 0;
321 int uninitialized_var(type), ret;
324 * we don't want to hold swap_lock while doing a very
325 * lengthy try_to_unuse, but swap_list may change
326 * so restart scan from swap_list.head each time
328 spin_lock(&swap_lock);
329 ret = __frontswap_shrink(target_pages, &pages_to_unuse, &type);
330 spin_unlock(&swap_lock);
332 try_to_unuse(type, true, pages_to_unuse);
335 EXPORT_SYMBOL(frontswap_shrink);
338 * Count and return the number of frontswap pages across all
339 * swap devices. This is exported so that backend drivers can
340 * determine current usage without reading debugfs.
342 unsigned long frontswap_curr_pages(void)
344 unsigned long totalpages = 0;
346 spin_lock(&swap_lock);
347 totalpages = __frontswap_curr_pages();
348 spin_unlock(&swap_lock);
352 EXPORT_SYMBOL(frontswap_curr_pages);
354 static int __init init_frontswap(void)
356 #ifdef CONFIG_DEBUG_FS
357 struct dentry *root = debugfs_create_dir("frontswap", NULL);
360 debugfs_create_u64("loads", S_IRUGO, root, &frontswap_loads);
361 debugfs_create_u64("succ_stores", S_IRUGO, root, &frontswap_succ_stores);
362 debugfs_create_u64("failed_stores", S_IRUGO, root,
363 &frontswap_failed_stores);
364 debugfs_create_u64("invalidates", S_IRUGO,
365 root, &frontswap_invalidates);
370 module_init(init_frontswap);