Commit 29f233cfffe7fbc6672938117ce7e4154a2f515f
Committed by
Konrad Rzeszutek Wilk
1 parent
38b5faf4b1
Exists in
master
and in
20 other branches
mm: frontswap: core frontswap functionality
This patch, 3of4, provides the core frontswap code that interfaces between the hooks in the swap subsystem and a frontswap backend via frontswap_ops. --- New file added: mm/frontswap.c [v14: add support for writethrough, per suggestion by aarcange@redhat.com] [v11: sjenning@linux.vnet.ibm.com: s/puts/failed_puts/] [v10: sjenning@linux.vnet.ibm.com: fix debugfs calls on 32-bit] [v9: akpm@linux-foundation.org: change "flush" to "invalidate", part 1] [v9: akpm@linux-foundation.org: mark some statics __read_mostly] [v9: akpm@linux-foundation.org: add clarifying comments] [v9: akpm@linux-foundation.org: no need to loop repeating try_to_unuse] [v9: error27@gmail.com: remove superfluous check for NULL] [v8: rebase to 3.0-rc4] [v8: kamezawa.hiroyu@jp.fujitsu.com: add comment to clarify find_next_to_unuse] [v7: rebase to 3.0-rc3] [v7: JBeulich@novell.com: use new static inlines, no-ops if not config'd] [v6: rebase to 3.1-rc1] [v6: lliubbo@gmail.com: use vzalloc] [v6: lliubbo@gmail.com: fix null pointer deref if vzalloc fails] [v6: konrad.wilk@oracl.com: various checks and code clarifications/comments] [v4: rebase to 2.6.39] Signed-off-by: Dan Magenheimer <dan.magenheimer@oracle.com> Acked-by: Jan Beulich <JBeulich@novell.com> Acked-by: Seth Jennings <sjenning@linux.vnet.ibm.com> Cc: Jeremy Fitzhardinge <jeremy@goop.org> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Nitin Gupta <ngupta@vflare.org> Cc: Matthew Wilcox <matthew@wil.cx> Cc: Chris Mason <chris.mason@oracle.com> Cc: Rik Riel <riel@redhat.com> Cc: Andrew Morton <akpm@linux-foundation.org> [v12: Squashed s/flush/invalidate/ in] [v15: A bit of cleanup and seperate DEBUGFS] Signed-off-by: Konrad Wilk <konrad.wilk@oracle.com>
Showing 1 changed file with 314 additions and 0 deletions Side-by-side Diff
mm/frontswap.c
1 | +/* | |
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 | + | |
14 | +#include <linux/mm.h> | |
15 | +#include <linux/mman.h> | |
16 | +#include <linux/swap.h> | |
17 | +#include <linux/swapops.h> | |
18 | +#include <linux/proc_fs.h> | |
19 | +#include <linux/security.h> | |
20 | +#include <linux/capability.h> | |
21 | +#include <linux/module.h> | |
22 | +#include <linux/uaccess.h> | |
23 | +#include <linux/debugfs.h> | |
24 | +#include <linux/frontswap.h> | |
25 | +#include <linux/swapfile.h> | |
26 | + | |
27 | +/* | |
28 | + * frontswap_ops is set by frontswap_register_ops to contain the pointers | |
29 | + * to the frontswap "backend" implementation functions. | |
30 | + */ | |
31 | +static struct frontswap_ops frontswap_ops __read_mostly; | |
32 | + | |
33 | +/* | |
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. | |
37 | + */ | |
38 | +bool frontswap_enabled __read_mostly; | |
39 | +EXPORT_SYMBOL(frontswap_enabled); | |
40 | + | |
41 | +/* | |
42 | + * If enabled, frontswap_put will return failure even on success. As | |
43 | + * a result, the swap subsystem will always write the page to swap, in | |
44 | + * effect converting frontswap into a writethrough cache. In this mode, | |
45 | + * there is no direct reduction in swap writes, but a frontswap backend | |
46 | + * can unilaterally "reclaim" any pages in use with no data loss, thus | |
47 | + * providing increases control over maximum memory usage due to frontswap. | |
48 | + */ | |
49 | +static bool frontswap_writethrough_enabled __read_mostly; | |
50 | + | |
51 | +#ifdef CONFIG_DEBUG_FS | |
52 | +/* | |
53 | + * Counters available via /sys/kernel/debug/frontswap (if debugfs is | |
54 | + * properly configured). These are for information only so are not protected | |
55 | + * against increment races. | |
56 | + */ | |
57 | +static u64 frontswap_gets; | |
58 | +static u64 frontswap_succ_puts; | |
59 | +static u64 frontswap_failed_puts; | |
60 | +static u64 frontswap_invalidates; | |
61 | + | |
62 | +static inline void inc_frontswap_gets(void) { | |
63 | + frontswap_gets++; | |
64 | +} | |
65 | +static inline void inc_frontswap_succ_puts(void) { | |
66 | + frontswap_succ_puts++; | |
67 | +} | |
68 | +static inline void inc_frontswap_failed_puts(void) { | |
69 | + frontswap_failed_puts++; | |
70 | +} | |
71 | +static inline void inc_frontswap_invalidates(void) { | |
72 | + frontswap_invalidates++; | |
73 | +} | |
74 | +#else | |
75 | +static inline void inc_frontswap_gets(void) { } | |
76 | +static inline void inc_frontswap_succ_puts(void) { } | |
77 | +static inline void inc_frontswap_failed_puts(void) { } | |
78 | +static inline void inc_frontswap_invalidates(void) { } | |
79 | +#endif | |
80 | +/* | |
81 | + * Register operations for frontswap, returning previous thus allowing | |
82 | + * detection of multiple backends and possible nesting. | |
83 | + */ | |
84 | +struct frontswap_ops frontswap_register_ops(struct frontswap_ops *ops) | |
85 | +{ | |
86 | + struct frontswap_ops old = frontswap_ops; | |
87 | + | |
88 | + frontswap_ops = *ops; | |
89 | + frontswap_enabled = true; | |
90 | + return old; | |
91 | +} | |
92 | +EXPORT_SYMBOL(frontswap_register_ops); | |
93 | + | |
94 | +/* | |
95 | + * Enable/disable frontswap writethrough (see above). | |
96 | + */ | |
97 | +void frontswap_writethrough(bool enable) | |
98 | +{ | |
99 | + frontswap_writethrough_enabled = enable; | |
100 | +} | |
101 | +EXPORT_SYMBOL(frontswap_writethrough); | |
102 | + | |
103 | +/* | |
104 | + * Called when a swap device is swapon'd. | |
105 | + */ | |
106 | +void __frontswap_init(unsigned type) | |
107 | +{ | |
108 | + struct swap_info_struct *sis = swap_info[type]; | |
109 | + | |
110 | + BUG_ON(sis == NULL); | |
111 | + if (sis->frontswap_map == NULL) | |
112 | + return; | |
113 | + if (frontswap_enabled) | |
114 | + (*frontswap_ops.init)(type); | |
115 | +} | |
116 | +EXPORT_SYMBOL(__frontswap_init); | |
117 | + | |
118 | +/* | |
119 | + * "Put" data from a page to frontswap and associate it with the page's | |
120 | + * swaptype and offset. Page must be locked and in the swap cache. | |
121 | + * If frontswap already contains a page with matching swaptype and | |
122 | + * offset, the frontswap implmentation may either overwrite the data and | |
123 | + * return success or invalidate the page from frontswap and return failure. | |
124 | + */ | |
125 | +int __frontswap_put_page(struct page *page) | |
126 | +{ | |
127 | + int ret = -1, dup = 0; | |
128 | + swp_entry_t entry = { .val = page_private(page), }; | |
129 | + int type = swp_type(entry); | |
130 | + struct swap_info_struct *sis = swap_info[type]; | |
131 | + pgoff_t offset = swp_offset(entry); | |
132 | + | |
133 | + BUG_ON(!PageLocked(page)); | |
134 | + BUG_ON(sis == NULL); | |
135 | + if (frontswap_test(sis, offset)) | |
136 | + dup = 1; | |
137 | + ret = (*frontswap_ops.put_page)(type, offset, page); | |
138 | + if (ret == 0) { | |
139 | + frontswap_set(sis, offset); | |
140 | + inc_frontswap_succ_puts(); | |
141 | + if (!dup) | |
142 | + atomic_inc(&sis->frontswap_pages); | |
143 | + } else if (dup) { | |
144 | + /* | |
145 | + failed dup always results in automatic invalidate of | |
146 | + the (older) page from frontswap | |
147 | + */ | |
148 | + frontswap_clear(sis, offset); | |
149 | + atomic_dec(&sis->frontswap_pages); | |
150 | + inc_frontswap_failed_puts(); | |
151 | + } else | |
152 | + inc_frontswap_failed_puts(); | |
153 | + if (frontswap_writethrough_enabled) | |
154 | + /* report failure so swap also writes to swap device */ | |
155 | + ret = -1; | |
156 | + return ret; | |
157 | +} | |
158 | +EXPORT_SYMBOL(__frontswap_put_page); | |
159 | + | |
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 | + */ | |
165 | +int __frontswap_get_page(struct page *page) | |
166 | +{ | |
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)) | |
176 | + ret = (*frontswap_ops.get_page)(type, offset, page); | |
177 | + if (ret == 0) | |
178 | + inc_frontswap_gets(); | |
179 | + return ret; | |
180 | +} | |
181 | +EXPORT_SYMBOL(__frontswap_get_page); | |
182 | + | |
183 | +/* | |
184 | + * Invalidate any data from frontswap associated with the specified swaptype | |
185 | + * and offset so that a subsequent "get" will fail. | |
186 | + */ | |
187 | +void __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)) { | |
193 | + (*frontswap_ops.invalidate_page)(type, offset); | |
194 | + atomic_dec(&sis->frontswap_pages); | |
195 | + frontswap_clear(sis, offset); | |
196 | + inc_frontswap_invalidates(); | |
197 | + } | |
198 | +} | |
199 | +EXPORT_SYMBOL(__frontswap_invalidate_page); | |
200 | + | |
201 | +/* | |
202 | + * Invalidate all data from frontswap associated with all offsets for the | |
203 | + * specified swaptype. | |
204 | + */ | |
205 | +void __frontswap_invalidate_area(unsigned type) | |
206 | +{ | |
207 | + struct swap_info_struct *sis = swap_info[type]; | |
208 | + | |
209 | + BUG_ON(sis == NULL); | |
210 | + if (sis->frontswap_map == NULL) | |
211 | + return; | |
212 | + (*frontswap_ops.invalidate_area)(type); | |
213 | + atomic_set(&sis->frontswap_pages, 0); | |
214 | + memset(sis->frontswap_map, 0, sis->max / sizeof(long)); | |
215 | +} | |
216 | +EXPORT_SYMBOL(__frontswap_invalidate_area); | |
217 | + | |
218 | +/* | |
219 | + * Frontswap, like a true swap device, may unnecessarily retain pages | |
220 | + * under certain circumstances; "shrink" frontswap is essentially a | |
221 | + * "partial swapoff" and works by calling try_to_unuse to attempt to | |
222 | + * unuse enough frontswap pages to attempt to -- subject to memory | |
223 | + * constraints -- reduce the number of pages in frontswap to the | |
224 | + * number given in the parameter target_pages. | |
225 | + */ | |
226 | +void frontswap_shrink(unsigned long target_pages) | |
227 | +{ | |
228 | + struct swap_info_struct *si = NULL; | |
229 | + int si_frontswap_pages; | |
230 | + unsigned long total_pages = 0, total_pages_to_unuse; | |
231 | + unsigned long pages = 0, pages_to_unuse = 0; | |
232 | + int type; | |
233 | + bool locked = false; | |
234 | + | |
235 | + /* | |
236 | + * we don't want to hold swap_lock while doing a very | |
237 | + * lengthy try_to_unuse, but swap_list may change | |
238 | + * so restart scan from swap_list.head each time | |
239 | + */ | |
240 | + spin_lock(&swap_lock); | |
241 | + locked = true; | |
242 | + total_pages = 0; | |
243 | + for (type = swap_list.head; type >= 0; type = si->next) { | |
244 | + si = swap_info[type]; | |
245 | + total_pages += atomic_read(&si->frontswap_pages); | |
246 | + } | |
247 | + if (total_pages <= target_pages) | |
248 | + goto out; | |
249 | + total_pages_to_unuse = total_pages - target_pages; | |
250 | + for (type = swap_list.head; type >= 0; type = si->next) { | |
251 | + si = swap_info[type]; | |
252 | + si_frontswap_pages = atomic_read(&si->frontswap_pages); | |
253 | + if (total_pages_to_unuse < si_frontswap_pages) | |
254 | + pages = pages_to_unuse = total_pages_to_unuse; | |
255 | + else { | |
256 | + pages = si_frontswap_pages; | |
257 | + pages_to_unuse = 0; /* unuse all */ | |
258 | + } | |
259 | + /* ensure there is enough RAM to fetch pages from frontswap */ | |
260 | + if (security_vm_enough_memory_mm(current->mm, pages)) | |
261 | + continue; | |
262 | + vm_unacct_memory(pages); | |
263 | + break; | |
264 | + } | |
265 | + if (type < 0) | |
266 | + goto out; | |
267 | + locked = false; | |
268 | + spin_unlock(&swap_lock); | |
269 | + try_to_unuse(type, true, pages_to_unuse); | |
270 | +out: | |
271 | + if (locked) | |
272 | + spin_unlock(&swap_lock); | |
273 | + return; | |
274 | +} | |
275 | +EXPORT_SYMBOL(frontswap_shrink); | |
276 | + | |
277 | +/* | |
278 | + * Count and return the number of frontswap pages across all | |
279 | + * swap devices. This is exported so that backend drivers can | |
280 | + * determine current usage without reading debugfs. | |
281 | + */ | |
282 | +unsigned long frontswap_curr_pages(void) | |
283 | +{ | |
284 | + int type; | |
285 | + unsigned long totalpages = 0; | |
286 | + struct swap_info_struct *si = NULL; | |
287 | + | |
288 | + spin_lock(&swap_lock); | |
289 | + for (type = swap_list.head; type >= 0; type = si->next) { | |
290 | + si = swap_info[type]; | |
291 | + totalpages += atomic_read(&si->frontswap_pages); | |
292 | + } | |
293 | + spin_unlock(&swap_lock); | |
294 | + return totalpages; | |
295 | +} | |
296 | +EXPORT_SYMBOL(frontswap_curr_pages); | |
297 | + | |
298 | +static int __init init_frontswap(void) | |
299 | +{ | |
300 | +#ifdef CONFIG_DEBUG_FS | |
301 | + struct dentry *root = debugfs_create_dir("frontswap", NULL); | |
302 | + if (root == NULL) | |
303 | + return -ENXIO; | |
304 | + debugfs_create_u64("gets", S_IRUGO, root, &frontswap_gets); | |
305 | + debugfs_create_u64("succ_puts", S_IRUGO, root, &frontswap_succ_puts); | |
306 | + debugfs_create_u64("failed_puts", S_IRUGO, root, | |
307 | + &frontswap_failed_puts); | |
308 | + debugfs_create_u64("invalidates", S_IRUGO, | |
309 | + root, &frontswap_invalidates); | |
310 | +#endif | |
311 | + return 0; | |
312 | +} | |
313 | + | |
314 | +module_init(init_frontswap); |