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) { }
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
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.
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.
106 * While no backend is registered all calls to frontswap_[store|load|
107 * invalidate_area|invalidate_page] are ignored or fail.
109 * The time between the backend being registered and the swap file system
110 * calling the backend (via the frontswap_* functions) is indeterminate as
111 * frontswap_ops is not atomic_t (or a value guarded by a spinlock).
112 * That is OK as we are comfortable missing some of these calls to the newly
113 * registered backend.
115 * Obviously the opposite (unloading the backend) must be done after all
116 * the frontswap_[store|load|invalidate_area|invalidate_page] start
117 * ignorning or failing the requests - at which point frontswap_ops
118 * would have to be made in some fashion atomic.
120 static DECLARE_BITMAP(need_init, MAX_SWAPFILES);
123 * Register operations for frontswap, returning previous thus allowing
124 * detection of multiple backends and possible nesting.
126 struct frontswap_ops *frontswap_register_ops(struct frontswap_ops *ops)
128 struct frontswap_ops *old = frontswap_ops;
131 frontswap_enabled = true;
133 for (i = 0; i < MAX_SWAPFILES; i++) {
134 if (test_and_clear_bit(i, need_init))
138 * We MUST have frontswap_ops set _after_ the frontswap_init's
139 * have been called. Otherwise __frontswap_store might fail. Hence
140 * the barrier to make sure compiler does not re-order us.
146 EXPORT_SYMBOL(frontswap_register_ops);
149 * Enable/disable frontswap writethrough (see above).
151 void frontswap_writethrough(bool enable)
153 frontswap_writethrough_enabled = enable;
155 EXPORT_SYMBOL(frontswap_writethrough);
158 * Enable/disable frontswap exclusive gets (see above).
160 void frontswap_tmem_exclusive_gets(bool enable)
162 frontswap_tmem_exclusive_gets_enabled = enable;
164 EXPORT_SYMBOL(frontswap_tmem_exclusive_gets);
167 * Called when a swap device is swapon'd.
169 void __frontswap_init(unsigned type)
171 struct swap_info_struct *sis = swap_info[type];
175 if (sis->frontswap_map == NULL)
177 frontswap_ops->init(type);
179 BUG_ON(type > MAX_SWAPFILES);
180 set_bit(type, need_init);
184 EXPORT_SYMBOL(__frontswap_init);
186 static inline void __frontswap_clear(struct swap_info_struct *sis, pgoff_t offset)
188 frontswap_clear(sis, offset);
189 atomic_dec(&sis->frontswap_pages);
193 * "Store" data from a page to frontswap and associate it with the page's
194 * swaptype and offset. Page must be locked and in the swap cache.
195 * If frontswap already contains a page with matching swaptype and
196 * offset, the frontswap implementation may either overwrite the data and
197 * return success or invalidate the page from frontswap and return failure.
199 int __frontswap_store(struct page *page)
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);
207 if (!frontswap_ops) {
208 inc_frontswap_failed_stores();
212 BUG_ON(!PageLocked(page));
214 if (frontswap_test(sis, offset))
216 ret = frontswap_ops->store(type, offset, page);
218 frontswap_set(sis, offset);
219 inc_frontswap_succ_stores();
221 atomic_inc(&sis->frontswap_pages);
224 failed dup always results in automatic invalidate of
225 the (older) page from frontswap
227 inc_frontswap_failed_stores();
229 __frontswap_clear(sis, offset);
231 if (frontswap_writethrough_enabled)
232 /* report failure so swap also writes to swap device */
236 EXPORT_SYMBOL(__frontswap_store);
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.
243 int __frontswap_load(struct page *page)
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);
254 BUG_ON(!PageLocked(page));
256 if (frontswap_test(sis, offset))
257 ret = frontswap_ops->load(type, offset, page);
259 inc_frontswap_loads();
260 if (frontswap_tmem_exclusive_gets_enabled) {
262 frontswap_clear(sis, offset);
267 EXPORT_SYMBOL(__frontswap_load);
270 * Invalidate any data from frontswap associated with the specified swaptype
271 * and offset so that a subsequent "get" will fail.
273 void __frontswap_invalidate_page(unsigned type, pgoff_t offset)
275 struct swap_info_struct *sis = swap_info[type];
281 if (frontswap_test(sis, offset)) {
282 frontswap_ops->invalidate_page(type, offset);
283 __frontswap_clear(sis, offset);
284 inc_frontswap_invalidates();
287 EXPORT_SYMBOL(__frontswap_invalidate_page);
290 * Invalidate all data from frontswap associated with all offsets for the
291 * specified swaptype.
293 void __frontswap_invalidate_area(unsigned type)
295 struct swap_info_struct *sis = swap_info[type];
299 if (sis->frontswap_map == NULL)
301 frontswap_ops->invalidate_area(type);
302 atomic_set(&sis->frontswap_pages, 0);
303 memset(sis->frontswap_map, 0, sis->max / sizeof(long));
305 clear_bit(type, need_init);
307 EXPORT_SYMBOL(__frontswap_invalidate_area);
309 static unsigned long __frontswap_curr_pages(void)
312 unsigned long totalpages = 0;
313 struct swap_info_struct *si = NULL;
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);
323 static int __frontswap_unuse_pages(unsigned long total, unsigned long *unused,
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;
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;
340 pages = si_frontswap_pages;
341 pages_to_unuse = 0; /* unuse all */
343 /* ensure there is enough RAM to fetch pages from frontswap */
344 if (security_vm_enough_memory_mm(current->mm, pages)) {
348 vm_unacct_memory(pages);
349 *unused = pages_to_unuse;
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.
363 static int __frontswap_shrink(unsigned long target_pages,
364 unsigned long *pages_to_unuse,
367 unsigned long total_pages = 0, total_pages_to_unuse;
369 assert_spin_locked(&swap_lock);
371 total_pages = __frontswap_curr_pages();
372 if (total_pages <= target_pages) {
377 total_pages_to_unuse = total_pages - target_pages;
378 return __frontswap_unuse_pages(total_pages_to_unuse, pages_to_unuse, type);
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.
389 void frontswap_shrink(unsigned long target_pages)
391 unsigned long pages_to_unuse = 0;
392 int uninitialized_var(type), ret;
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
399 spin_lock(&swap_lock);
400 ret = __frontswap_shrink(target_pages, &pages_to_unuse, &type);
401 spin_unlock(&swap_lock);
403 try_to_unuse(type, true, pages_to_unuse);
406 EXPORT_SYMBOL(frontswap_shrink);
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.
413 unsigned long frontswap_curr_pages(void)
415 unsigned long totalpages = 0;
417 spin_lock(&swap_lock);
418 totalpages = __frontswap_curr_pages();
419 spin_unlock(&swap_lock);
423 EXPORT_SYMBOL(frontswap_curr_pages);
425 static int __init init_frontswap(void)
427 #ifdef CONFIG_DEBUG_FS
428 struct dentry *root = debugfs_create_dir("frontswap", NULL);
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);
435 debugfs_create_u64("invalidates", S_IRUGO,
436 root, &frontswap_invalidates);
438 frontswap_enabled = 1;
442 module_init(init_frontswap);