1 // SPDX-License-Identifier: GPL-2.0-only
5 * This code provides the generic "frontend" layer to call a matching
6 * "backend" driver implementation of frontswap. See
7 * Documentation/vm/frontswap.rst for more information.
9 * Copyright (C) 2009-2012 Oracle Corp. All rights reserved.
10 * Author: Dan Magenheimer
13 #include <linux/mman.h>
14 #include <linux/swap.h>
15 #include <linux/swapops.h>
16 #include <linux/security.h>
17 #include <linux/module.h>
18 #include <linux/debugfs.h>
19 #include <linux/frontswap.h>
20 #include <linux/swapfile.h>
22 DEFINE_STATIC_KEY_FALSE(frontswap_enabled_key);
25 * frontswap_ops are added by frontswap_register_ops, and provide the
26 * frontswap "backend" implementation functions. Multiple implementations
27 * may be registered, but implementations can never deregister. This
28 * is a simple singly-linked list of all registered implementations.
30 static struct frontswap_ops *frontswap_ops __read_mostly;
32 #define for_each_frontswap_ops(ops) \
33 for ((ops) = frontswap_ops; (ops); (ops) = (ops)->next)
35 #ifdef CONFIG_DEBUG_FS
37 * Counters available via /sys/kernel/debug/frontswap (if debugfs is
38 * properly configured). These are for information only so are not protected
39 * against increment races.
41 static u64 frontswap_loads;
42 static u64 frontswap_succ_stores;
43 static u64 frontswap_failed_stores;
44 static u64 frontswap_invalidates;
46 static inline void inc_frontswap_loads(void)
48 data_race(frontswap_loads++);
50 static inline void inc_frontswap_succ_stores(void)
52 data_race(frontswap_succ_stores++);
54 static inline void inc_frontswap_failed_stores(void)
56 data_race(frontswap_failed_stores++);
58 static inline void inc_frontswap_invalidates(void)
60 data_race(frontswap_invalidates++);
63 static inline void inc_frontswap_loads(void) { }
64 static inline void inc_frontswap_succ_stores(void) { }
65 static inline void inc_frontswap_failed_stores(void) { }
66 static inline void inc_frontswap_invalidates(void) { }
70 * Due to the asynchronous nature of the backends loading potentially
71 * _after_ the swap system has been activated, we have chokepoints
72 * on all frontswap functions to not call the backend until the backend
75 * This would not guards us against the user deciding to call swapoff right as
76 * we are calling the backend to initialize (so swapon is in action).
77 * Fortunately for us, the swapon_mutex has been taken by the callee so we are
78 * OK. The other scenario where calls to frontswap_store (called via
79 * swap_writepage) is racing with frontswap_invalidate_area (called via
80 * swapoff) is again guarded by the swap subsystem.
82 * While no backend is registered all calls to frontswap_[store|load|
83 * invalidate_area|invalidate_page] are ignored or fail.
85 * The time between the backend being registered and the swap file system
86 * calling the backend (via the frontswap_* functions) is indeterminate as
87 * frontswap_ops is not atomic_t (or a value guarded by a spinlock).
88 * That is OK as we are comfortable missing some of these calls to the newly
91 * Obviously the opposite (unloading the backend) must be done after all
92 * the frontswap_[store|load|invalidate_area|invalidate_page] start
93 * ignoring or failing the requests. However, there is currently no way
94 * to unload a backend once it is registered.
98 * Register operations for frontswap
100 void frontswap_register_ops(struct frontswap_ops *ops)
102 DECLARE_BITMAP(a, MAX_SWAPFILES);
103 DECLARE_BITMAP(b, MAX_SWAPFILES);
104 struct swap_info_struct *si;
107 bitmap_zero(a, MAX_SWAPFILES);
108 bitmap_zero(b, MAX_SWAPFILES);
110 spin_lock(&swap_lock);
111 plist_for_each_entry(si, &swap_active_head, list) {
112 if (!WARN_ON(!si->frontswap_map))
113 __set_bit(si->type, a);
115 spin_unlock(&swap_lock);
117 /* the new ops needs to know the currently active swap devices */
118 for_each_set_bit(i, a, MAX_SWAPFILES)
122 * Setting frontswap_ops must happen after the ops->init() calls
123 * above; cmpxchg implies smp_mb() which will ensure the init is
124 * complete at this point.
127 ops->next = frontswap_ops;
128 } while (cmpxchg(&frontswap_ops, ops->next, ops) != ops->next);
130 static_branch_inc(&frontswap_enabled_key);
132 spin_lock(&swap_lock);
133 plist_for_each_entry(si, &swap_active_head, list) {
134 if (si->frontswap_map)
135 __set_bit(si->type, b);
137 spin_unlock(&swap_lock);
140 * On the very unlikely chance that a swap device was added or
141 * removed between setting the "a" list bits and the ops init
142 * calls, we re-check and do init or invalidate for any changed
145 if (unlikely(!bitmap_equal(a, b, MAX_SWAPFILES))) {
146 for (i = 0; i < MAX_SWAPFILES; i++) {
147 if (!test_bit(i, a) && test_bit(i, b))
149 else if (test_bit(i, a) && !test_bit(i, b))
150 ops->invalidate_area(i);
154 EXPORT_SYMBOL(frontswap_register_ops);
157 * Called when a swap device is swapon'd.
159 void __frontswap_init(unsigned type, unsigned long *map)
161 struct swap_info_struct *sis = swap_info[type];
162 struct frontswap_ops *ops;
164 VM_BUG_ON(sis == NULL);
167 * p->frontswap is a bitmap that we MUST have to figure out which page
168 * has gone in frontswap. Without it there is no point of continuing.
173 * Irregardless of whether the frontswap backend has been loaded
174 * before this function or it will be later, we _MUST_ have the
175 * p->frontswap set to something valid to work properly.
177 frontswap_map_set(sis, map);
179 for_each_frontswap_ops(ops)
182 EXPORT_SYMBOL(__frontswap_init);
184 bool __frontswap_test(struct swap_info_struct *sis,
187 if (sis->frontswap_map)
188 return test_bit(offset, sis->frontswap_map);
191 EXPORT_SYMBOL(__frontswap_test);
193 static inline void __frontswap_set(struct swap_info_struct *sis,
196 set_bit(offset, sis->frontswap_map);
197 atomic_inc(&sis->frontswap_pages);
200 static inline void __frontswap_clear(struct swap_info_struct *sis,
203 clear_bit(offset, sis->frontswap_map);
204 atomic_dec(&sis->frontswap_pages);
208 * "Store" data from a page to frontswap and associate it with the page's
209 * swaptype and offset. Page must be locked and in the swap cache.
210 * If frontswap already contains a page with matching swaptype and
211 * offset, the frontswap implementation may either overwrite the data and
212 * return success or invalidate the page from frontswap and return failure.
214 int __frontswap_store(struct page *page)
217 swp_entry_t entry = { .val = page_private(page), };
218 int type = swp_type(entry);
219 struct swap_info_struct *sis = swap_info[type];
220 pgoff_t offset = swp_offset(entry);
221 struct frontswap_ops *ops;
223 VM_BUG_ON(!frontswap_ops);
224 VM_BUG_ON(!PageLocked(page));
225 VM_BUG_ON(sis == NULL);
228 * If a dup, we must remove the old page first; we can't leave the
229 * old page no matter if the store of the new page succeeds or fails,
230 * and we can't rely on the new page replacing the old page as we may
231 * not store to the same implementation that contains the old page.
233 if (__frontswap_test(sis, offset)) {
234 __frontswap_clear(sis, offset);
235 for_each_frontswap_ops(ops)
236 ops->invalidate_page(type, offset);
239 /* Try to store in each implementation, until one succeeds. */
240 for_each_frontswap_ops(ops) {
241 ret = ops->store(type, offset, page);
242 if (!ret) /* successful store */
246 __frontswap_set(sis, offset);
247 inc_frontswap_succ_stores();
249 inc_frontswap_failed_stores();
254 EXPORT_SYMBOL(__frontswap_store);
257 * "Get" data from frontswap associated with swaptype and offset that were
258 * specified when the data was put to frontswap and use it to fill the
259 * specified page with data. Page must be locked and in the swap cache.
261 int __frontswap_load(struct page *page)
264 swp_entry_t entry = { .val = page_private(page), };
265 int type = swp_type(entry);
266 struct swap_info_struct *sis = swap_info[type];
267 pgoff_t offset = swp_offset(entry);
268 struct frontswap_ops *ops;
270 VM_BUG_ON(!frontswap_ops);
271 VM_BUG_ON(!PageLocked(page));
272 VM_BUG_ON(sis == NULL);
274 if (!__frontswap_test(sis, offset))
277 /* Try loading from each implementation, until one succeeds. */
278 for_each_frontswap_ops(ops) {
279 ret = ops->load(type, offset, page);
280 if (!ret) /* successful load */
284 inc_frontswap_loads();
287 EXPORT_SYMBOL(__frontswap_load);
290 * Invalidate any data from frontswap associated with the specified swaptype
291 * and offset so that a subsequent "get" will fail.
293 void __frontswap_invalidate_page(unsigned type, pgoff_t offset)
295 struct swap_info_struct *sis = swap_info[type];
296 struct frontswap_ops *ops;
298 VM_BUG_ON(!frontswap_ops);
299 VM_BUG_ON(sis == NULL);
301 if (!__frontswap_test(sis, offset))
304 for_each_frontswap_ops(ops)
305 ops->invalidate_page(type, offset);
306 __frontswap_clear(sis, offset);
307 inc_frontswap_invalidates();
309 EXPORT_SYMBOL(__frontswap_invalidate_page);
312 * Invalidate all data from frontswap associated with all offsets for the
313 * specified swaptype.
315 void __frontswap_invalidate_area(unsigned type)
317 struct swap_info_struct *sis = swap_info[type];
318 struct frontswap_ops *ops;
320 VM_BUG_ON(!frontswap_ops);
321 VM_BUG_ON(sis == NULL);
323 if (sis->frontswap_map == NULL)
326 for_each_frontswap_ops(ops)
327 ops->invalidate_area(type);
328 atomic_set(&sis->frontswap_pages, 0);
329 bitmap_zero(sis->frontswap_map, sis->max);
331 EXPORT_SYMBOL(__frontswap_invalidate_area);
333 static unsigned long __frontswap_curr_pages(void)
335 unsigned long totalpages = 0;
336 struct swap_info_struct *si = NULL;
338 assert_spin_locked(&swap_lock);
339 plist_for_each_entry(si, &swap_active_head, list)
340 totalpages += atomic_read(&si->frontswap_pages);
344 static int __frontswap_unuse_pages(unsigned long total, unsigned long *unused,
348 struct swap_info_struct *si = NULL;
349 int si_frontswap_pages;
350 unsigned long total_pages_to_unuse = total;
351 unsigned long pages = 0, pages_to_unuse = 0;
353 assert_spin_locked(&swap_lock);
354 plist_for_each_entry(si, &swap_active_head, list) {
355 si_frontswap_pages = atomic_read(&si->frontswap_pages);
356 if (total_pages_to_unuse < si_frontswap_pages) {
357 pages = pages_to_unuse = total_pages_to_unuse;
359 pages = si_frontswap_pages;
360 pages_to_unuse = 0; /* unuse all */
362 /* ensure there is enough RAM to fetch pages from frontswap */
363 if (security_vm_enough_memory_mm(current->mm, pages)) {
367 vm_unacct_memory(pages);
368 *unused = pages_to_unuse;
378 * Used to check if it's necessary and feasible to unuse pages.
379 * Return 1 when nothing to do, 0 when need to shrink pages,
380 * error code when there is an error.
382 static int __frontswap_shrink(unsigned long target_pages,
383 unsigned long *pages_to_unuse,
386 unsigned long total_pages = 0, total_pages_to_unuse;
388 assert_spin_locked(&swap_lock);
390 total_pages = __frontswap_curr_pages();
391 if (total_pages <= target_pages) {
396 total_pages_to_unuse = total_pages - target_pages;
397 return __frontswap_unuse_pages(total_pages_to_unuse, pages_to_unuse, type);
401 * Frontswap, like a true swap device, may unnecessarily retain pages
402 * under certain circumstances; "shrink" frontswap is essentially a
403 * "partial swapoff" and works by calling try_to_unuse to attempt to
404 * unuse enough frontswap pages to attempt to -- subject to memory
405 * constraints -- reduce the number of pages in frontswap to the
406 * number given in the parameter target_pages.
408 void frontswap_shrink(unsigned long target_pages)
410 unsigned long pages_to_unuse = 0;
414 * we don't want to hold swap_lock while doing a very
415 * lengthy try_to_unuse, but swap_list may change
416 * so restart scan from swap_active_head each time
418 spin_lock(&swap_lock);
419 ret = __frontswap_shrink(target_pages, &pages_to_unuse, &type);
420 spin_unlock(&swap_lock);
422 try_to_unuse(type, true, pages_to_unuse);
425 EXPORT_SYMBOL(frontswap_shrink);
428 * Count and return the number of frontswap pages across all
429 * swap devices. This is exported so that backend drivers can
430 * determine current usage without reading debugfs.
432 unsigned long frontswap_curr_pages(void)
434 unsigned long totalpages = 0;
436 spin_lock(&swap_lock);
437 totalpages = __frontswap_curr_pages();
438 spin_unlock(&swap_lock);
442 EXPORT_SYMBOL(frontswap_curr_pages);
444 static int __init init_frontswap(void)
446 #ifdef CONFIG_DEBUG_FS
447 struct dentry *root = debugfs_create_dir("frontswap", NULL);
450 debugfs_create_u64("loads", 0444, root, &frontswap_loads);
451 debugfs_create_u64("succ_stores", 0444, root, &frontswap_succ_stores);
452 debugfs_create_u64("failed_stores", 0444, root,
453 &frontswap_failed_stores);
454 debugfs_create_u64("invalidates", 0444, root, &frontswap_invalidates);
459 module_init(init_frontswap);