blob: fbeb4bb8eb50b2d2091db7a5e6a0846d016e7f10 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/swapfile.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/mm.h>
9#include <linux/hugetlb.h>
10#include <linux/mman.h>
11#include <linux/slab.h>
12#include <linux/kernel_stat.h>
13#include <linux/swap.h>
14#include <linux/vmalloc.h>
15#include <linux/pagemap.h>
16#include <linux/namei.h>
17#include <linux/shm.h>
18#include <linux/blkdev.h>
19#include <linux/writeback.h>
20#include <linux/proc_fs.h>
21#include <linux/seq_file.h>
22#include <linux/init.h>
23#include <linux/module.h>
24#include <linux/rmap.h>
25#include <linux/security.h>
26#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080027#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080028#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/syscalls.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080030#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#include <asm/pgtable.h>
33#include <asm/tlbflush.h>
34#include <linux/swapops.h>
35
Adrian Bunk7c363b82008-07-25 19:46:24 -070036static DEFINE_SPINLOCK(swap_lock);
37static unsigned int nr_swapfiles;
Hugh Dickinsb9627162009-01-06 14:39:41 -080038long nr_swap_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039long total_swap_pages;
40static int swap_overflow;
Hugh Dickins78ecba02008-07-23 21:28:23 -070041static int least_priority;
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043static const char Bad_file[] = "Bad swap file entry ";
44static const char Unused_file[] = "Unused swap file entry ";
45static const char Bad_offset[] = "Bad swap offset entry ";
46static const char Unused_offset[] = "Unused swap offset entry ";
47
Adrian Bunk7c363b82008-07-25 19:46:24 -070048static struct swap_list_t swap_list = {-1, -1};
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Rafael J. Wysockif577eb32006-03-23 02:59:59 -080050static struct swap_info_struct swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Ingo Molnarfc0abb12006-01-18 17:42:33 -080052static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54/*
55 * We need this because the bdev->unplug_fn can sleep and we cannot
Hugh Dickins5d337b92005-09-03 15:54:41 -070056 * hold swap_lock while calling the unplug_fn. And swap_lock
Ingo Molnarfc0abb12006-01-18 17:42:33 -080057 * cannot be turned into a mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 */
59static DECLARE_RWSEM(swap_unplug_sem);
60
Linus Torvalds1da177e2005-04-16 15:20:36 -070061void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
62{
63 swp_entry_t entry;
64
65 down_read(&swap_unplug_sem);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070066 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 if (PageSwapCache(page)) {
68 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
69 struct backing_dev_info *bdi;
70
71 /*
72 * If the page is removed from swapcache from under us (with a
73 * racy try_to_unuse/swapoff) we need an additional reference
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070074 * count to avoid reading garbage from page_private(page) above.
75 * If the WARN_ON triggers during a swapoff it maybe the race
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 * condition and it's harmless. However if it triggers without
77 * swapoff it signals a problem.
78 */
79 WARN_ON(page_count(page) <= 1);
80
81 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
McMullan, Jasonba323112005-05-16 21:53:40 -070082 blk_run_backing_dev(bdi, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 }
84 up_read(&swap_unplug_sem);
85}
86
Hugh Dickins6a6ba832009-01-06 14:39:51 -080087/*
88 * swapon tell device that all the old swap contents can be discarded,
89 * to allow the swap device to optimize its wear-levelling.
90 */
91static int discard_swap(struct swap_info_struct *si)
92{
93 struct swap_extent *se;
94 int err = 0;
95
96 list_for_each_entry(se, &si->extent_list, list) {
97 sector_t start_block = se->start_block << (PAGE_SHIFT - 9);
98 pgoff_t nr_blocks = se->nr_pages << (PAGE_SHIFT - 9);
99
100 if (se->start_page == 0) {
101 /* Do not discard the swap header page! */
102 start_block += 1 << (PAGE_SHIFT - 9);
103 nr_blocks -= 1 << (PAGE_SHIFT - 9);
104 if (!nr_blocks)
105 continue;
106 }
107
108 err = blkdev_issue_discard(si->bdev, start_block,
109 nr_blocks, GFP_KERNEL);
110 if (err)
111 break;
112
113 cond_resched();
114 }
115 return err; /* That will often be -EOPNOTSUPP */
116}
117
Hugh Dickins048c27f2005-09-03 15:54:40 -0700118#define SWAPFILE_CLUSTER 256
119#define LATENCY_LIMIT 256
120
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700121static inline unsigned long scan_swap_map(struct swap_info_struct *si)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800123 unsigned long offset;
124 unsigned long last_in_cluster;
Hugh Dickins048c27f2005-09-03 15:54:40 -0700125 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800127 /*
Hugh Dickins7dfad412005-09-03 15:54:38 -0700128 * We try to cluster swap pages by allocating them sequentially
129 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
130 * way, however, we resort to first-free allocation, starting
131 * a new cluster. This prevents us from scattering swap pages
132 * all over the entire swap partition, so that we reduce
133 * overall disk seek times between swap pages. -- sct
134 * But we do now try to find an empty cluster. -Andrea
135 */
136
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700137 si->flags += SWP_SCANNING;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800138 offset = si->cluster_next;
139
140 if (unlikely(!si->cluster_nr--)) {
141 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
142 si->cluster_nr = SWAPFILE_CLUSTER - 1;
143 goto checks;
144 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700145 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700146
147 offset = si->lowest_bit;
148 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
149
150 /* Locate the first empty (unaligned) cluster */
151 for (; last_in_cluster <= si->highest_bit; offset++) {
152 if (si->swap_map[offset])
153 last_in_cluster = offset + SWAPFILE_CLUSTER;
154 else if (offset == last_in_cluster) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700155 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800156 offset -= SWAPFILE_CLUSTER - 1;
157 si->cluster_next = offset;
158 si->cluster_nr = SWAPFILE_CLUSTER - 1;
159 goto checks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700161 if (unlikely(--latency_ration < 0)) {
162 cond_resched();
163 latency_ration = LATENCY_LIMIT;
164 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700165 }
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800166
167 offset = si->lowest_bit;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700168 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800169 si->cluster_nr = SWAPFILE_CLUSTER - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700171
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800172checks:
173 if (!(si->flags & SWP_WRITEOK))
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700174 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700175 if (!si->highest_bit)
176 goto no_page;
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800177 if (offset > si->highest_bit)
178 offset = si->lowest_bit;
179 if (si->swap_map[offset])
180 goto scan;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700181
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800182 if (offset == si->lowest_bit)
183 si->lowest_bit++;
184 if (offset == si->highest_bit)
185 si->highest_bit--;
186 si->inuse_pages++;
187 if (si->inuse_pages == si->pages) {
188 si->lowest_bit = si->max;
189 si->highest_bit = 0;
190 }
191 si->swap_map[offset] = 1;
192 si->cluster_next = offset + 1;
193 si->flags -= SWP_SCANNING;
194 return offset;
195
196scan:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700197 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700198 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700199 if (!si->swap_map[offset]) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700200 spin_lock(&swap_lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700201 goto checks;
202 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700203 if (unlikely(--latency_ration < 0)) {
204 cond_resched();
205 latency_ration = LATENCY_LIMIT;
206 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700207 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700208 spin_lock(&swap_lock);
Hugh Dickinsebebbbe2009-01-06 14:39:50 -0800209 goto checks;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700210
211no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700212 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 return 0;
214}
215
216swp_entry_t get_swap_page(void)
217{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700218 struct swap_info_struct *si;
219 pgoff_t offset;
220 int type, next;
221 int wrapped = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Hugh Dickins5d337b92005-09-03 15:54:41 -0700223 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 if (nr_swap_pages <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700225 goto noswap;
226 nr_swap_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700228 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
229 si = swap_info + type;
230 next = si->next;
231 if (next < 0 ||
232 (!wrapped && si->prio != swap_info[next].prio)) {
233 next = swap_list.head;
234 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700236
237 if (!si->highest_bit)
238 continue;
239 if (!(si->flags & SWP_WRITEOK))
240 continue;
241
242 swap_list.next = next;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700243 offset = scan_swap_map(si);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700244 if (offset) {
245 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700246 return swp_entry(type, offset);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700247 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700248 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700250
251 nr_swap_pages++;
252noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700253 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700254 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255}
256
Rafael J. Wysocki3a291a22006-01-06 00:16:37 -0800257swp_entry_t get_swap_page_of_type(int type)
258{
259 struct swap_info_struct *si;
260 pgoff_t offset;
261
262 spin_lock(&swap_lock);
263 si = swap_info + type;
264 if (si->flags & SWP_WRITEOK) {
265 nr_swap_pages--;
266 offset = scan_swap_map(si);
267 if (offset) {
268 spin_unlock(&swap_lock);
269 return swp_entry(type, offset);
270 }
271 nr_swap_pages++;
272 }
273 spin_unlock(&swap_lock);
274 return (swp_entry_t) {0};
275}
276
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277static struct swap_info_struct * swap_info_get(swp_entry_t entry)
278{
279 struct swap_info_struct * p;
280 unsigned long offset, type;
281
282 if (!entry.val)
283 goto out;
284 type = swp_type(entry);
285 if (type >= nr_swapfiles)
286 goto bad_nofile;
287 p = & swap_info[type];
288 if (!(p->flags & SWP_USED))
289 goto bad_device;
290 offset = swp_offset(entry);
291 if (offset >= p->max)
292 goto bad_offset;
293 if (!p->swap_map[offset])
294 goto bad_free;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700295 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 return p;
297
298bad_free:
299 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
300 goto out;
301bad_offset:
302 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
303 goto out;
304bad_device:
305 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
306 goto out;
307bad_nofile:
308 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
309out:
310 return NULL;
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800311}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
314{
315 int count = p->swap_map[offset];
316
317 if (count < SWAP_MAP_MAX) {
318 count--;
319 p->swap_map[offset] = count;
320 if (!count) {
321 if (offset < p->lowest_bit)
322 p->lowest_bit = offset;
323 if (offset > p->highest_bit)
324 p->highest_bit = offset;
Hugh Dickins89d09a22005-09-03 15:54:36 -0700325 if (p->prio > swap_info[swap_list.next].prio)
326 swap_list.next = p - swap_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 nr_swap_pages++;
328 p->inuse_pages--;
329 }
330 }
331 return count;
332}
333
334/*
335 * Caller has made sure that the swapdevice corresponding to entry
336 * is still around or has not been recycled.
337 */
338void swap_free(swp_entry_t entry)
339{
340 struct swap_info_struct * p;
341
342 p = swap_info_get(entry);
343 if (p) {
344 swap_entry_free(p, swp_offset(entry));
Hugh Dickins5d337b92005-09-03 15:54:41 -0700345 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 }
347}
348
349/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700350 * How many references to page are currently swapped out?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700352static inline int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700354 int count = 0;
355 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 swp_entry_t entry;
357
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700358 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 p = swap_info_get(entry);
360 if (p) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700361 /* Subtract the 1 for the swap cache itself */
362 count = p->swap_map[swp_offset(entry)] - 1;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700363 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700365 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366}
367
368/*
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800369 * We can write to an anon page without COW if there are no other references
370 * to it. And as a side-effect, free up its swap: because the old content
371 * on disk will never be read, and seeking back there to write new content
372 * later would only waste time away from clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 */
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800374int reuse_swap_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700376 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Hugh Dickins51726b12009-01-06 14:39:25 -0800378 VM_BUG_ON(!PageLocked(page));
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700379 count = page_mapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800380 if (count <= 1 && PageSwapCache(page)) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700381 count += page_swapcount(page);
Hugh Dickins7b1fe592009-01-06 14:39:34 -0800382 if (count == 1 && !PageWriteback(page)) {
383 delete_from_swap_cache(page);
384 SetPageDirty(page);
385 }
386 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700387 return count == 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388}
389
390/*
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800391 * If swap is getting full, or if there are no more mappings of this page,
392 * then try_to_free_swap is called to free its swap space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 */
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800394int try_to_free_swap(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395{
Hugh Dickins51726b12009-01-06 14:39:25 -0800396 VM_BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
398 if (!PageSwapCache(page))
399 return 0;
400 if (PageWriteback(page))
401 return 0;
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800402 if (page_swapcount(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 return 0;
404
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800405 delete_from_swap_cache(page);
406 SetPageDirty(page);
407 return 1;
Rik van Riel68a223942008-10-18 20:26:23 -0700408}
409
410/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 * Free the swap entry like above, but also try to
412 * free the page cache entry if it is the last user.
413 */
414void free_swap_and_cache(swp_entry_t entry)
415{
416 struct swap_info_struct * p;
417 struct page *page = NULL;
418
Christoph Lameter06972122006-06-23 02:03:35 -0700419 if (is_migration_entry(entry))
420 return;
421
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 p = swap_info_get(entry);
423 if (p) {
Nick Piggin93fac702006-03-31 02:29:56 -0800424 if (swap_entry_free(p, swp_offset(entry)) == 1) {
425 page = find_get_page(&swapper_space, entry.val);
Nick Piggin8413ac92008-10-18 20:26:59 -0700426 if (page && !trylock_page(page)) {
Nick Piggin93fac702006-03-31 02:29:56 -0800427 page_cache_release(page);
428 page = NULL;
429 }
430 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700431 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 }
433 if (page) {
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800434 /*
435 * Not mapped elsewhere, or swap space full? Free it!
436 * Also recheck PageSwapCache now page is locked (above).
437 */
Nick Piggin93fac702006-03-31 02:29:56 -0800438 if (PageSwapCache(page) && !PageWriteback(page) &&
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800439 (!page_mapped(page) || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 delete_from_swap_cache(page);
441 SetPageDirty(page);
442 }
443 unlock_page(page);
444 page_cache_release(page);
445 }
446}
447
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200448#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800449/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800450 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800451 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800452 * @offset - number of the PAGE_SIZE-sized block of the device, starting
453 * from 0, in which the swap header is expected to be located.
454 *
455 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800456 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800457int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800458{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800459 struct block_device *bdev = NULL;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800460 int i;
461
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800462 if (device)
463 bdev = bdget(device);
464
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800465 spin_lock(&swap_lock);
466 for (i = 0; i < nr_swapfiles; i++) {
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800467 struct swap_info_struct *sis = swap_info + i;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800468
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800469 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800470 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700471
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800472 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800473 if (bdev_p)
474 *bdev_p = sis->bdev;
475
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800476 spin_unlock(&swap_lock);
477 return i;
478 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800479 if (bdev == sis->bdev) {
480 struct swap_extent *se;
481
482 se = list_entry(sis->extent_list.next,
483 struct swap_extent, list);
484 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800485 if (bdev_p)
486 *bdev_p = sis->bdev;
487
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800488 spin_unlock(&swap_lock);
489 bdput(bdev);
490 return i;
491 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800492 }
493 }
494 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800495 if (bdev)
496 bdput(bdev);
497
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800498 return -ENODEV;
499}
500
501/*
502 * Return either the total number of swap pages of given type, or the number
503 * of free pages of that type (depending on @free)
504 *
505 * This is needed for software suspend
506 */
507unsigned int count_swap_pages(int type, int free)
508{
509 unsigned int n = 0;
510
511 if (type < nr_swapfiles) {
512 spin_lock(&swap_lock);
513 if (swap_info[type].flags & SWP_WRITEOK) {
514 n = swap_info[type].pages;
515 if (free)
516 n -= swap_info[type].inuse_pages;
517 }
518 spin_unlock(&swap_lock);
519 }
520 return n;
521}
522#endif
523
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700525 * No need to decide whether this PTE shares the swap entry with others,
526 * just let do_wp_page work it out if a write is requested later - to
527 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 */
Hugh Dickins044d66c2008-02-07 00:14:04 -0800529static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 unsigned long addr, swp_entry_t entry, struct page *page)
531{
Hugh Dickins044d66c2008-02-07 00:14:04 -0800532 spinlock_t *ptl;
533 pte_t *pte;
534 int ret = 1;
535
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800536 if (mem_cgroup_charge(page, vma->vm_mm, GFP_KERNEL))
Hugh Dickins044d66c2008-02-07 00:14:04 -0800537 ret = -ENOMEM;
538
539 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
540 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
541 if (ret > 0)
542 mem_cgroup_uncharge_page(page);
543 ret = 0;
544 goto out;
545 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800546
Hugh Dickins42946212005-10-29 18:16:05 -0700547 inc_mm_counter(vma->vm_mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 get_page(page);
549 set_pte_at(vma->vm_mm, addr, pte,
550 pte_mkold(mk_pte(page, vma->vm_page_prot)));
551 page_add_anon_rmap(page, vma, addr);
552 swap_free(entry);
553 /*
554 * Move the page to the active list so it is not
555 * immediately swapped out again after swapon.
556 */
557 activate_page(page);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800558out:
559 pte_unmap_unlock(pte, ptl);
560 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561}
562
563static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
564 unsigned long addr, unsigned long end,
565 swp_entry_t entry, struct page *page)
566{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700568 pte_t *pte;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800569 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Hugh Dickins044d66c2008-02-07 00:14:04 -0800571 /*
572 * We don't actually need pte lock while scanning for swp_pte: since
573 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
574 * page table while we're scanning; though it could get zapped, and on
575 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
576 * of unmatched parts which look like swp_pte, so unuse_pte must
577 * recheck under pte lock. Scanning without pte lock lets it be
578 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
579 */
580 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 do {
582 /*
583 * swapoff spends a _lot_ of time in this loop!
584 * Test inline before going to call unuse_pte.
585 */
586 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins044d66c2008-02-07 00:14:04 -0800587 pte_unmap(pte);
588 ret = unuse_pte(vma, pmd, addr, entry, page);
589 if (ret)
590 goto out;
591 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 }
593 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins044d66c2008-02-07 00:14:04 -0800594 pte_unmap(pte - 1);
595out:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800596 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597}
598
599static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
600 unsigned long addr, unsigned long end,
601 swp_entry_t entry, struct page *page)
602{
603 pmd_t *pmd;
604 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800605 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
607 pmd = pmd_offset(pud, addr);
608 do {
609 next = pmd_addr_end(addr, end);
610 if (pmd_none_or_clear_bad(pmd))
611 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800612 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
613 if (ret)
614 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 } while (pmd++, addr = next, addr != end);
616 return 0;
617}
618
619static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
620 unsigned long addr, unsigned long end,
621 swp_entry_t entry, struct page *page)
622{
623 pud_t *pud;
624 unsigned long next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800625 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
627 pud = pud_offset(pgd, addr);
628 do {
629 next = pud_addr_end(addr, end);
630 if (pud_none_or_clear_bad(pud))
631 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800632 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
633 if (ret)
634 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 } while (pud++, addr = next, addr != end);
636 return 0;
637}
638
639static int unuse_vma(struct vm_area_struct *vma,
640 swp_entry_t entry, struct page *page)
641{
642 pgd_t *pgd;
643 unsigned long addr, end, next;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800644 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
646 if (page->mapping) {
647 addr = page_address_in_vma(page, vma);
648 if (addr == -EFAULT)
649 return 0;
650 else
651 end = addr + PAGE_SIZE;
652 } else {
653 addr = vma->vm_start;
654 end = vma->vm_end;
655 }
656
657 pgd = pgd_offset(vma->vm_mm, addr);
658 do {
659 next = pgd_addr_end(addr, end);
660 if (pgd_none_or_clear_bad(pgd))
661 continue;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800662 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
663 if (ret)
664 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 } while (pgd++, addr = next, addr != end);
666 return 0;
667}
668
669static int unuse_mm(struct mm_struct *mm,
670 swp_entry_t entry, struct page *page)
671{
672 struct vm_area_struct *vma;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800673 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
675 if (!down_read_trylock(&mm->mmap_sem)) {
676 /*
Fernando Luis Vazquez Cao7d034312008-07-29 22:33:41 -0700677 * Activate page so shrink_inactive_list is unlikely to unmap
678 * its ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700680 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 unlock_page(page);
682 down_read(&mm->mmap_sem);
683 lock_page(page);
684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 for (vma = mm->mmap; vma; vma = vma->vm_next) {
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800686 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 break;
688 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 up_read(&mm->mmap_sem);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800690 return (ret < 0)? ret: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691}
692
693/*
694 * Scan swap_map from current position to next entry still in use.
695 * Recycle to start on reaching the end, returning 0 when empty.
696 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700697static unsigned int find_next_to_unuse(struct swap_info_struct *si,
698 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699{
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700700 unsigned int max = si->max;
701 unsigned int i = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 int count;
703
704 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -0700705 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 * for whether an entry is in use, not modifying it; false
707 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -0700708 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 */
710 for (;;) {
711 if (++i >= max) {
712 if (!prev) {
713 i = 0;
714 break;
715 }
716 /*
717 * No entries in use at top of swap_map,
718 * loop back to start and recheck there.
719 */
720 max = prev + 1;
721 prev = 0;
722 i = 1;
723 }
724 count = si->swap_map[i];
725 if (count && count != SWAP_MAP_BAD)
726 break;
727 }
728 return i;
729}
730
731/*
732 * We completely avoid races by reading each swap page in advance,
733 * and then search for the process using it. All the necessary
734 * page table adjustments can then be made atomically.
735 */
736static int try_to_unuse(unsigned int type)
737{
738 struct swap_info_struct * si = &swap_info[type];
739 struct mm_struct *start_mm;
740 unsigned short *swap_map;
741 unsigned short swcount;
742 struct page *page;
743 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700744 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 int retval = 0;
746 int reset_overflow = 0;
747 int shmem;
748
749 /*
750 * When searching mms for an entry, a good strategy is to
751 * start at the first mm we freed the previous entry from
752 * (though actually we don't notice whether we or coincidence
753 * freed the entry). Initialize this start_mm with a hold.
754 *
755 * A simpler strategy would be to start at the last mm we
756 * freed the previous entry from; but that would take less
757 * advantage of mmlist ordering, which clusters forked mms
758 * together, child after parent. If we race with dup_mmap(), we
759 * prefer to resolve parent before child, lest we miss entries
760 * duplicated after we scanned child: using last mm would invert
761 * that. Though it's only a serious concern when an overflowed
762 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
763 */
764 start_mm = &init_mm;
765 atomic_inc(&init_mm.mm_users);
766
767 /*
768 * Keep on scanning until all entries have gone. Usually,
769 * one pass through swap_map is enough, but not necessarily:
770 * there are races when an instance of an entry might be missed.
771 */
772 while ((i = find_next_to_unuse(si, i)) != 0) {
773 if (signal_pending(current)) {
774 retval = -EINTR;
775 break;
776 }
777
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800778 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 * Get a page for the entry, using the existing swap
780 * cache page if there is one. Otherwise, get a clean
Hugh Dickins886bb7e2009-01-06 14:39:48 -0800781 * page and read the swap into it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 */
783 swap_map = &si->swap_map[i];
784 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -0800785 page = read_swap_cache_async(entry,
786 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 if (!page) {
788 /*
789 * Either swap_duplicate() failed because entry
790 * has been freed independently, and will not be
791 * reused since sys_swapoff() already disabled
792 * allocation from here, or alloc_page() failed.
793 */
794 if (!*swap_map)
795 continue;
796 retval = -ENOMEM;
797 break;
798 }
799
800 /*
801 * Don't hold on to start_mm if it looks like exiting.
802 */
803 if (atomic_read(&start_mm->mm_users) == 1) {
804 mmput(start_mm);
805 start_mm = &init_mm;
806 atomic_inc(&init_mm.mm_users);
807 }
808
809 /*
810 * Wait for and lock page. When do_swap_page races with
811 * try_to_unuse, do_swap_page can handle the fault much
812 * faster than try_to_unuse can locate the entry. This
813 * apparently redundant "wait_on_page_locked" lets try_to_unuse
814 * defer to do_swap_page in such a case - in some tests,
815 * do_swap_page and try_to_unuse repeatedly compete.
816 */
817 wait_on_page_locked(page);
818 wait_on_page_writeback(page);
819 lock_page(page);
820 wait_on_page_writeback(page);
821
822 /*
823 * Remove all references to entry.
824 * Whenever we reach init_mm, there's no address space
825 * to search, but use it as a reminder to search shmem.
826 */
827 shmem = 0;
828 swcount = *swap_map;
829 if (swcount > 1) {
830 if (start_mm == &init_mm)
831 shmem = shmem_unuse(entry, page);
832 else
833 retval = unuse_mm(start_mm, entry, page);
834 }
835 if (*swap_map > 1) {
836 int set_start_mm = (*swap_map >= swcount);
837 struct list_head *p = &start_mm->mmlist;
838 struct mm_struct *new_start_mm = start_mm;
839 struct mm_struct *prev_mm = start_mm;
840 struct mm_struct *mm;
841
842 atomic_inc(&new_start_mm->mm_users);
843 atomic_inc(&prev_mm->mm_users);
844 spin_lock(&mmlist_lock);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800845 while (*swap_map > 1 && !retval && !shmem &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 (p = p->next) != &start_mm->mmlist) {
847 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -0700848 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 spin_unlock(&mmlist_lock);
851 mmput(prev_mm);
852 prev_mm = mm;
853
854 cond_resched();
855
856 swcount = *swap_map;
857 if (swcount <= 1)
858 ;
859 else if (mm == &init_mm) {
860 set_start_mm = 1;
861 shmem = shmem_unuse(entry, page);
862 } else
863 retval = unuse_mm(mm, entry, page);
864 if (set_start_mm && *swap_map < swcount) {
865 mmput(new_start_mm);
866 atomic_inc(&mm->mm_users);
867 new_start_mm = mm;
868 set_start_mm = 0;
869 }
870 spin_lock(&mmlist_lock);
871 }
872 spin_unlock(&mmlist_lock);
873 mmput(prev_mm);
874 mmput(start_mm);
875 start_mm = new_start_mm;
876 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800877 if (shmem) {
878 /* page has already been unlocked and released */
879 if (shmem > 0)
880 continue;
881 retval = shmem;
882 break;
883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 if (retval) {
885 unlock_page(page);
886 page_cache_release(page);
887 break;
888 }
889
890 /*
891 * How could swap count reach 0x7fff when the maximum
892 * pid is 0x7fff, and there's no way to repeat a swap
893 * page within an mm (except in shmem, where it's the
894 * shared object which takes the reference count)?
895 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
896 *
897 * If that's wrong, then we should worry more about
898 * exit_mmap() and do_munmap() cases described above:
899 * we might be resetting SWAP_MAP_MAX too early here.
900 * We know "Undead"s can happen, they're okay, so don't
901 * report them; but do report if we reset SWAP_MAP_MAX.
902 */
903 if (*swap_map == SWAP_MAP_MAX) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700904 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 *swap_map = 1;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700906 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 reset_overflow = 1;
908 }
909
910 /*
911 * If a reference remains (rare), we would like to leave
912 * the page in the swap cache; but try_to_unmap could
913 * then re-duplicate the entry once we drop page lock,
914 * so we might loop indefinitely; also, that page could
915 * not be swapped out to other storage meanwhile. So:
916 * delete from cache even if there's another reference,
917 * after ensuring that the data has been saved to disk -
918 * since if the reference remains (rarer), it will be
919 * read from disk into another page. Splitting into two
920 * pages would be incorrect if swap supported "shared
921 * private" pages, but they are handled by tmpfs files.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 */
923 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
924 struct writeback_control wbc = {
925 .sync_mode = WB_SYNC_NONE,
926 };
927
928 swap_writepage(page, &wbc);
929 lock_page(page);
930 wait_on_page_writeback(page);
931 }
Hugh Dickins68bdc8d2009-01-06 14:39:37 -0800932
933 /*
934 * It is conceivable that a racing task removed this page from
935 * swap cache just before we acquired the page lock at the top,
936 * or while we dropped it in unuse_mm(). The page might even
937 * be back in swap cache on another swap area: that we must not
938 * delete, since it may not have been written out to swap yet.
939 */
940 if (PageSwapCache(page) &&
941 likely(page_private(page) == entry.val))
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800942 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 /*
945 * So we could skip searching mms once swap count went
946 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700947 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 */
949 SetPageDirty(page);
950 unlock_page(page);
951 page_cache_release(page);
952
953 /*
954 * Make sure that we aren't completely killing
955 * interactive performance.
956 */
957 cond_resched();
958 }
959
960 mmput(start_mm);
961 if (reset_overflow) {
962 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
963 swap_overflow = 0;
964 }
965 return retval;
966}
967
968/*
Hugh Dickins5d337b92005-09-03 15:54:41 -0700969 * After a successful try_to_unuse, if no swap is now in use, we know
970 * we can empty the mmlist. swap_lock must be held on entry and exit.
971 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 * added to the mmlist just after page_duplicate - before would be racy.
973 */
974static void drain_mmlist(void)
975{
976 struct list_head *p, *next;
977 unsigned int i;
978
979 for (i = 0; i < nr_swapfiles; i++)
980 if (swap_info[i].inuse_pages)
981 return;
982 spin_lock(&mmlist_lock);
983 list_for_each_safe(p, next, &init_mm.mmlist)
984 list_del_init(p);
985 spin_unlock(&mmlist_lock);
986}
987
988/*
989 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
990 * corresponds to page offset `offset'.
991 */
992sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
993{
994 struct swap_extent *se = sis->curr_swap_extent;
995 struct swap_extent *start_se = se;
996
997 for ( ; ; ) {
998 struct list_head *lh;
999
1000 if (se->start_page <= offset &&
1001 offset < (se->start_page + se->nr_pages)) {
1002 return se->start_block + (offset - se->start_page);
1003 }
Hugh Dickins11d31882005-09-03 15:54:34 -07001004 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 if (lh == &sis->extent_list)
Hugh Dickins11d31882005-09-03 15:54:34 -07001006 lh = lh->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 se = list_entry(lh, struct swap_extent, list);
1008 sis->curr_swap_extent = se;
1009 BUG_ON(se == start_se); /* It *must* be present */
1010 }
1011}
1012
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001013#ifdef CONFIG_HIBERNATION
Rafael J. Wysocki3aef83e2006-12-06 20:34:10 -08001014/*
1015 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1016 * corresponding to given index in swap_info (swap type).
1017 */
1018sector_t swapdev_block(int swap_type, pgoff_t offset)
1019{
1020 struct swap_info_struct *sis;
1021
1022 if (swap_type >= nr_swapfiles)
1023 return 0;
1024
1025 sis = swap_info + swap_type;
1026 return (sis->flags & SWP_WRITEOK) ? map_swap_page(sis, offset) : 0;
1027}
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +02001028#endif /* CONFIG_HIBERNATION */
Rafael J. Wysocki3aef83e2006-12-06 20:34:10 -08001029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030/*
1031 * Free all of a swapdev's extent information
1032 */
1033static void destroy_swap_extents(struct swap_info_struct *sis)
1034{
1035 while (!list_empty(&sis->extent_list)) {
1036 struct swap_extent *se;
1037
1038 se = list_entry(sis->extent_list.next,
1039 struct swap_extent, list);
1040 list_del(&se->list);
1041 kfree(se);
1042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043}
1044
1045/*
1046 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001047 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001049 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 */
1051static int
1052add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1053 unsigned long nr_pages, sector_t start_block)
1054{
1055 struct swap_extent *se;
1056 struct swap_extent *new_se;
1057 struct list_head *lh;
1058
Hugh Dickins11d31882005-09-03 15:54:34 -07001059 lh = sis->extent_list.prev; /* The highest page extent */
1060 if (lh != &sis->extent_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001062 BUG_ON(se->start_page + se->nr_pages != start_page);
1063 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 /* Merge it */
1065 se->nr_pages += nr_pages;
1066 return 0;
1067 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 }
1069
1070 /*
1071 * No merge. Insert a new extent, preserving ordering.
1072 */
1073 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1074 if (new_se == NULL)
1075 return -ENOMEM;
1076 new_se->start_page = start_page;
1077 new_se->nr_pages = nr_pages;
1078 new_se->start_block = start_block;
1079
Hugh Dickins11d31882005-09-03 15:54:34 -07001080 list_add_tail(&new_se->list, &sis->extent_list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001081 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082}
1083
1084/*
1085 * A `swap extent' is a simple thing which maps a contiguous range of pages
1086 * onto a contiguous range of disk blocks. An ordered list of swap extents
1087 * is built at swapon time and is then used at swap_writepage/swap_readpage
1088 * time for locating where on disk a page belongs.
1089 *
1090 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1091 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1092 * swap files identically.
1093 *
1094 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1095 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1096 * swapfiles are handled *identically* after swapon time.
1097 *
1098 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1099 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1100 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1101 * requirements, they are simply tossed out - we will never use those blocks
1102 * for swapping.
1103 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001104 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1106 * which will scribble on the fs.
1107 *
1108 * The amount of disk space which a single swap extent represents varies.
1109 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1110 * extents in the list. To avoid much list walking, we cache the previous
1111 * search location in `curr_swap_extent', and start new searches from there.
1112 * This is extremely effective. The average number of iterations in
1113 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1114 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001115static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116{
1117 struct inode *inode;
1118 unsigned blocks_per_page;
1119 unsigned long page_no;
1120 unsigned blkbits;
1121 sector_t probe_block;
1122 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001123 sector_t lowest_block = -1;
1124 sector_t highest_block = 0;
1125 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 int ret;
1127
1128 inode = sis->swap_file->f_mapping->host;
1129 if (S_ISBLK(inode->i_mode)) {
1130 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001131 *span = sis->pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 goto done;
1133 }
1134
1135 blkbits = inode->i_blkbits;
1136 blocks_per_page = PAGE_SIZE >> blkbits;
1137
1138 /*
1139 * Map all the blocks into the extent list. This code doesn't try
1140 * to be very smart.
1141 */
1142 probe_block = 0;
1143 page_no = 0;
1144 last_block = i_size_read(inode) >> blkbits;
1145 while ((probe_block + blocks_per_page) <= last_block &&
1146 page_no < sis->max) {
1147 unsigned block_in_page;
1148 sector_t first_block;
1149
1150 first_block = bmap(inode, probe_block);
1151 if (first_block == 0)
1152 goto bad_bmap;
1153
1154 /*
1155 * It must be PAGE_SIZE aligned on-disk
1156 */
1157 if (first_block & (blocks_per_page - 1)) {
1158 probe_block++;
1159 goto reprobe;
1160 }
1161
1162 for (block_in_page = 1; block_in_page < blocks_per_page;
1163 block_in_page++) {
1164 sector_t block;
1165
1166 block = bmap(inode, probe_block + block_in_page);
1167 if (block == 0)
1168 goto bad_bmap;
1169 if (block != first_block + block_in_page) {
1170 /* Discontiguity */
1171 probe_block++;
1172 goto reprobe;
1173 }
1174 }
1175
Hugh Dickins53092a72005-09-03 15:54:34 -07001176 first_block >>= (PAGE_SHIFT - blkbits);
1177 if (page_no) { /* exclude the header page */
1178 if (first_block < lowest_block)
1179 lowest_block = first_block;
1180 if (first_block > highest_block)
1181 highest_block = first_block;
1182 }
1183
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 /*
1185 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1186 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001187 ret = add_swap_extent(sis, page_no, 1, first_block);
1188 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001190 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 page_no++;
1192 probe_block += blocks_per_page;
1193reprobe:
1194 continue;
1195 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001196 ret = nr_extents;
1197 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001199 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001201 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 sis->highest_bit = page_no - 1;
1203done:
1204 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1205 struct swap_extent, list);
1206 goto out;
1207bad_bmap:
1208 printk(KERN_ERR "swapon: swapfile has holes\n");
1209 ret = -EINVAL;
1210out:
1211 return ret;
1212}
1213
1214#if 0 /* We don't need this yet */
1215#include <linux/backing-dev.h>
1216int page_queue_congested(struct page *page)
1217{
1218 struct backing_dev_info *bdi;
1219
Hugh Dickins51726b12009-01-06 14:39:25 -08001220 VM_BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
1222 if (PageSwapCache(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001223 swp_entry_t entry = { .val = page_private(page) };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 struct swap_info_struct *sis;
1225
1226 sis = get_swap_info_struct(swp_type(entry));
1227 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1228 } else
1229 bdi = page->mapping->backing_dev_info;
1230 return bdi_write_congested(bdi);
1231}
1232#endif
1233
1234asmlinkage long sys_swapoff(const char __user * specialfile)
1235{
1236 struct swap_info_struct * p = NULL;
1237 unsigned short *swap_map;
1238 struct file *swap_file, *victim;
1239 struct address_space *mapping;
1240 struct inode *inode;
1241 char * pathname;
1242 int i, type, prev;
1243 int err;
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001244
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 if (!capable(CAP_SYS_ADMIN))
1246 return -EPERM;
1247
1248 pathname = getname(specialfile);
1249 err = PTR_ERR(pathname);
1250 if (IS_ERR(pathname))
1251 goto out;
1252
1253 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1254 putname(pathname);
1255 err = PTR_ERR(victim);
1256 if (IS_ERR(victim))
1257 goto out;
1258
1259 mapping = victim->f_mapping;
1260 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001261 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1263 p = swap_info + type;
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001264 if (p->flags & SWP_WRITEOK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 if (p->swap_file->f_mapping == mapping)
1266 break;
1267 }
1268 prev = type;
1269 }
1270 if (type < 0) {
1271 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001272 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 goto out_dput;
1274 }
1275 if (!security_vm_enough_memory(p->pages))
1276 vm_unacct_memory(p->pages);
1277 else {
1278 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001279 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 goto out_dput;
1281 }
1282 if (prev < 0) {
1283 swap_list.head = p->next;
1284 } else {
1285 swap_info[prev].next = p->next;
1286 }
1287 if (type == swap_list.next) {
1288 /* just pick something that's safe... */
1289 swap_list.next = swap_list.head;
1290 }
Hugh Dickins78ecba02008-07-23 21:28:23 -07001291 if (p->prio < 0) {
1292 for (i = p->next; i >= 0; i = swap_info[i].next)
1293 swap_info[i].prio = p->prio--;
1294 least_priority++;
1295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 nr_swap_pages -= p->pages;
1297 total_swap_pages -= p->pages;
1298 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001299 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001300
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 current->flags |= PF_SWAPOFF;
1302 err = try_to_unuse(type);
1303 current->flags &= ~PF_SWAPOFF;
1304
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 if (err) {
1306 /* re-insert swap space back into swap_list */
Hugh Dickins5d337b92005-09-03 15:54:41 -07001307 spin_lock(&swap_lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001308 if (p->prio < 0)
1309 p->prio = --least_priority;
1310 prev = -1;
1311 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 if (p->prio >= swap_info[i].prio)
1313 break;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001314 prev = i;
1315 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 p->next = i;
1317 if (prev < 0)
1318 swap_list.head = swap_list.next = p - swap_info;
1319 else
1320 swap_info[prev].next = p - swap_info;
1321 nr_swap_pages += p->pages;
1322 total_swap_pages += p->pages;
1323 p->flags |= SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001324 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 goto out_dput;
1326 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001327
1328 /* wait for any unplug function to finish */
1329 down_write(&swap_unplug_sem);
1330 up_write(&swap_unplug_sem);
1331
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001332 destroy_swap_extents(p);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001333 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001334 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001336
1337 /* wait for anyone still in scan_swap_map */
1338 p->highest_bit = 0; /* cuts scans short */
1339 while (p->flags >= SWP_SCANNING) {
1340 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001341 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001342 spin_lock(&swap_lock);
1343 }
1344
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 swap_file = p->swap_file;
1346 p->swap_file = NULL;
1347 p->max = 0;
1348 swap_map = p->swap_map;
1349 p->swap_map = NULL;
1350 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001351 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001352 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 vfree(swap_map);
1354 inode = mapping->host;
1355 if (S_ISBLK(inode->i_mode)) {
1356 struct block_device *bdev = I_BDEV(inode);
1357 set_blocksize(bdev, p->old_block_size);
1358 bd_release(bdev);
1359 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001360 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001362 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 }
1364 filp_close(swap_file, NULL);
1365 err = 0;
1366
1367out_dput:
1368 filp_close(victim, NULL);
1369out:
1370 return err;
1371}
1372
1373#ifdef CONFIG_PROC_FS
1374/* iterator */
1375static void *swap_start(struct seq_file *swap, loff_t *pos)
1376{
1377 struct swap_info_struct *ptr = swap_info;
1378 int i;
1379 loff_t l = *pos;
1380
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001381 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001383 if (!l)
1384 return SEQ_START_TOKEN;
1385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1387 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1388 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001389 if (!--l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 return ptr;
1391 }
1392
1393 return NULL;
1394}
1395
1396static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1397{
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001398 struct swap_info_struct *ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1400
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001401 if (v == SEQ_START_TOKEN)
1402 ptr = swap_info;
1403 else {
1404 ptr = v;
1405 ptr++;
1406 }
1407
1408 for (; ptr < endptr; ptr++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1410 continue;
1411 ++*pos;
1412 return ptr;
1413 }
1414
1415 return NULL;
1416}
1417
1418static void swap_stop(struct seq_file *swap, void *v)
1419{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001420 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421}
1422
1423static int swap_show(struct seq_file *swap, void *v)
1424{
1425 struct swap_info_struct *ptr = v;
1426 struct file *file;
1427 int len;
1428
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001429 if (ptr == SEQ_START_TOKEN) {
1430 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1431 return 0;
1432 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
1434 file = ptr->swap_file;
Jan Blunckc32c2f62008-02-14 19:38:43 -08001435 len = seq_path(swap, &file->f_path, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001436 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001437 len < 40 ? 40 - len : 1, " ",
1438 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 "partition" : "file\t",
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001440 ptr->pages << (PAGE_SHIFT - 10),
1441 ptr->inuse_pages << (PAGE_SHIFT - 10),
1442 ptr->prio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 return 0;
1444}
1445
Helge Deller15ad7cd2006-12-06 20:40:36 -08001446static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 .start = swap_start,
1448 .next = swap_next,
1449 .stop = swap_stop,
1450 .show = swap_show
1451};
1452
1453static int swaps_open(struct inode *inode, struct file *file)
1454{
1455 return seq_open(file, &swaps_op);
1456}
1457
Helge Deller15ad7cd2006-12-06 20:40:36 -08001458static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 .open = swaps_open,
1460 .read = seq_read,
1461 .llseek = seq_lseek,
1462 .release = seq_release,
1463};
1464
1465static int __init procswaps_init(void)
1466{
Denis V. Lunev3d71f862008-04-29 01:02:13 -07001467 proc_create("swaps", 0, NULL, &proc_swaps_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 return 0;
1469}
1470__initcall(procswaps_init);
1471#endif /* CONFIG_PROC_FS */
1472
Jan Beulich17963162008-12-16 11:35:24 +00001473#ifdef MAX_SWAPFILES_CHECK
1474static int __init max_swapfiles_check(void)
1475{
1476 MAX_SWAPFILES_CHECK();
1477 return 0;
1478}
1479late_initcall(max_swapfiles_check);
1480#endif
1481
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482/*
1483 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1484 *
1485 * The swapon system call
1486 */
1487asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1488{
1489 struct swap_info_struct * p;
1490 char *name = NULL;
1491 struct block_device *bdev = NULL;
1492 struct file *swap_file = NULL;
1493 struct address_space *mapping;
1494 unsigned int type;
1495 int i, prev;
1496 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 union swap_header *swap_header = NULL;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001498 unsigned int nr_good_pages = 0;
1499 int nr_extents = 0;
Hugh Dickins53092a72005-09-03 15:54:34 -07001500 sector_t span;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 unsigned long maxpages = 1;
Hugh Dickins73fd8742009-01-06 14:39:47 -08001502 unsigned long swapfilepages;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001503 unsigned short *swap_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 struct page *page = NULL;
1505 struct inode *inode = NULL;
1506 int did_down = 0;
1507
1508 if (!capable(CAP_SYS_ADMIN))
1509 return -EPERM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001510 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 p = swap_info;
1512 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1513 if (!(p->flags & SWP_USED))
1514 break;
1515 error = -EPERM;
Christoph Lameter06972122006-06-23 02:03:35 -07001516 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001517 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 goto out;
1519 }
1520 if (type >= nr_swapfiles)
1521 nr_swapfiles = type+1;
Hugh Dickins78ecba02008-07-23 21:28:23 -07001522 memset(p, 0, sizeof(*p));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 INIT_LIST_HEAD(&p->extent_list);
1524 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 p->next = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001526 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 name = getname(specialfile);
1528 error = PTR_ERR(name);
1529 if (IS_ERR(name)) {
1530 name = NULL;
1531 goto bad_swap_2;
1532 }
1533 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1534 error = PTR_ERR(swap_file);
1535 if (IS_ERR(swap_file)) {
1536 swap_file = NULL;
1537 goto bad_swap_2;
1538 }
1539
1540 p->swap_file = swap_file;
1541 mapping = swap_file->f_mapping;
1542 inode = mapping->host;
1543
1544 error = -EBUSY;
1545 for (i = 0; i < nr_swapfiles; i++) {
1546 struct swap_info_struct *q = &swap_info[i];
1547
1548 if (i == type || !q->swap_file)
1549 continue;
1550 if (mapping == q->swap_file->f_mapping)
1551 goto bad_swap;
1552 }
1553
1554 error = -EINVAL;
1555 if (S_ISBLK(inode->i_mode)) {
1556 bdev = I_BDEV(inode);
1557 error = bd_claim(bdev, sys_swapon);
1558 if (error < 0) {
1559 bdev = NULL;
Rob Landleyf7b3a432005-09-22 21:44:27 -07001560 error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 goto bad_swap;
1562 }
1563 p->old_block_size = block_size(bdev);
1564 error = set_blocksize(bdev, PAGE_SIZE);
1565 if (error < 0)
1566 goto bad_swap;
1567 p->bdev = bdev;
1568 } else if (S_ISREG(inode->i_mode)) {
1569 p->bdev = inode->i_sb->s_bdev;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001570 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 did_down = 1;
1572 if (IS_SWAPFILE(inode)) {
1573 error = -EBUSY;
1574 goto bad_swap;
1575 }
1576 } else {
1577 goto bad_swap;
1578 }
1579
Hugh Dickins73fd8742009-01-06 14:39:47 -08001580 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
1582 /*
1583 * Read the swap header.
1584 */
1585 if (!mapping->a_ops->readpage) {
1586 error = -EINVAL;
1587 goto bad_swap;
1588 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07001589 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 if (IS_ERR(page)) {
1591 error = PTR_ERR(page);
1592 goto bad_swap;
1593 }
Hugh Dickins81e33972009-01-06 14:39:49 -08001594 swap_header = kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595
Hugh Dickins81e33972009-01-06 14:39:49 -08001596 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
Jesper Juhle97a3112006-01-11 01:50:28 +01001597 printk(KERN_ERR "Unable to find swap-space signature\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 error = -EINVAL;
1599 goto bad_swap;
1600 }
Hugh Dickins886bb7e2009-01-06 14:39:48 -08001601
Hugh Dickins81e33972009-01-06 14:39:49 -08001602 /* swap partition endianess hack... */
1603 if (swab32(swap_header->info.version) == 1) {
1604 swab32s(&swap_header->info.version);
1605 swab32s(&swap_header->info.last_page);
1606 swab32s(&swap_header->info.nr_badpages);
1607 for (i = 0; i < swap_header->info.nr_badpages; i++)
1608 swab32s(&swap_header->info.badpages[i]);
1609 }
1610 /* Check the swap header's sub-version */
1611 if (swap_header->info.version != 1) {
1612 printk(KERN_WARNING
1613 "Unable to handle swap header version %d\n",
1614 swap_header->info.version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 error = -EINVAL;
1616 goto bad_swap;
Hugh Dickins81e33972009-01-06 14:39:49 -08001617 }
1618
1619 p->lowest_bit = 1;
1620 p->cluster_next = 1;
1621
1622 /*
1623 * Find out how many pages are allowed for a single swap
1624 * device. There are two limiting factors: 1) the number of
1625 * bits for the swap offset in the swp_entry_t type and
1626 * 2) the number of bits in the a swap pte as defined by
1627 * the different architectures. In order to find the
1628 * largest possible bit mask a swap entry with swap type 0
1629 * and swap offset ~0UL is created, encoded to a swap pte,
1630 * decoded to a swp_entry_t again and finally the swap
1631 * offset is extracted. This will mask all the bits from
1632 * the initial ~0UL mask that can't be encoded in either
1633 * the swp_entry_t or the architecture definition of a
1634 * swap pte.
1635 */
1636 maxpages = swp_offset(pte_to_swp_entry(
1637 swp_entry_to_pte(swp_entry(0, ~0UL)))) - 1;
1638 if (maxpages > swap_header->info.last_page)
1639 maxpages = swap_header->info.last_page;
1640 p->highest_bit = maxpages - 1;
1641
1642 error = -EINVAL;
1643 if (!maxpages)
1644 goto bad_swap;
1645 if (swapfilepages && maxpages > swapfilepages) {
1646 printk(KERN_WARNING
1647 "Swap area shorter than signature indicates\n");
1648 goto bad_swap;
1649 }
1650 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1651 goto bad_swap;
1652 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1653 goto bad_swap;
1654
1655 /* OK, set up the swap map and apply the bad block list */
1656 swap_map = vmalloc(maxpages * sizeof(short));
1657 if (!swap_map) {
1658 error = -ENOMEM;
1659 goto bad_swap;
1660 }
1661
1662 memset(swap_map, 0, maxpages * sizeof(short));
1663 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1664 int page_nr = swap_header->info.badpages[i];
1665 if (page_nr <= 0 || page_nr >= swap_header->info.last_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 error = -EINVAL;
1667 goto bad_swap;
1668 }
Hugh Dickins81e33972009-01-06 14:39:49 -08001669 swap_map[page_nr] = SWAP_MAP_BAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 }
Hugh Dickins81e33972009-01-06 14:39:49 -08001671 nr_good_pages = swap_header->info.last_page -
1672 swap_header->info.nr_badpages -
1673 1 /* header page */;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001674
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001675 if (nr_good_pages) {
Hugh Dickins78ecba02008-07-23 21:28:23 -07001676 swap_map[0] = SWAP_MAP_BAD;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001677 p->max = maxpages;
1678 p->pages = nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07001679 nr_extents = setup_swap_extents(p, &span);
1680 if (nr_extents < 0) {
1681 error = nr_extents;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001682 goto bad_swap;
Hugh Dickins53092a72005-09-03 15:54:34 -07001683 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001684 nr_good_pages = p->pages;
1685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 if (!nr_good_pages) {
1687 printk(KERN_WARNING "Empty swap-file\n");
1688 error = -EINVAL;
1689 goto bad_swap;
1690 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691
Hugh Dickins6a6ba832009-01-06 14:39:51 -08001692 if (discard_swap(p) == 0)
1693 p->flags |= SWP_DISCARDABLE;
1694
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001695 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001696 spin_lock(&swap_lock);
Hugh Dickins78ecba02008-07-23 21:28:23 -07001697 if (swap_flags & SWAP_FLAG_PREFER)
1698 p->prio =
1699 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
1700 else
1701 p->prio = --least_priority;
1702 p->swap_map = swap_map;
Hugh Dickins22c6f8f2009-01-06 14:39:48 -08001703 p->flags |= SWP_WRITEOK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 nr_swap_pages += nr_good_pages;
1705 total_swap_pages += nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07001706
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001707 printk(KERN_INFO "Adding %uk swap on %s. "
Hugh Dickins6a6ba832009-01-06 14:39:51 -08001708 "Priority:%d extents:%d across:%lluk%s\n",
Hugh Dickins53092a72005-09-03 15:54:34 -07001709 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
Hugh Dickins6a6ba832009-01-06 14:39:51 -08001710 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
1711 (p->flags & SWP_DISCARDABLE) ? " D" : "");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712
1713 /* insert swap space into swap_list: */
1714 prev = -1;
1715 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1716 if (p->prio >= swap_info[i].prio) {
1717 break;
1718 }
1719 prev = i;
1720 }
1721 p->next = i;
1722 if (prev < 0) {
1723 swap_list.head = swap_list.next = p - swap_info;
1724 } else {
1725 swap_info[prev].next = p - swap_info;
1726 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07001727 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001728 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 error = 0;
1730 goto out;
1731bad_swap:
1732 if (bdev) {
1733 set_blocksize(bdev, p->old_block_size);
1734 bd_release(bdev);
1735 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001736 destroy_swap_extents(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737bad_swap_2:
Hugh Dickins5d337b92005-09-03 15:54:41 -07001738 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 p->swap_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001741 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 vfree(swap_map);
1743 if (swap_file)
1744 filp_close(swap_file, NULL);
1745out:
1746 if (page && !IS_ERR(page)) {
1747 kunmap(page);
1748 page_cache_release(page);
1749 }
1750 if (name)
1751 putname(name);
1752 if (did_down) {
1753 if (!error)
1754 inode->i_flags |= S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001755 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 }
1757 return error;
1758}
1759
1760void si_swapinfo(struct sysinfo *val)
1761{
1762 unsigned int i;
1763 unsigned long nr_to_be_unused = 0;
1764
Hugh Dickins5d337b92005-09-03 15:54:41 -07001765 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 for (i = 0; i < nr_swapfiles; i++) {
1767 if (!(swap_info[i].flags & SWP_USED) ||
1768 (swap_info[i].flags & SWP_WRITEOK))
1769 continue;
1770 nr_to_be_unused += swap_info[i].inuse_pages;
1771 }
1772 val->freeswap = nr_swap_pages + nr_to_be_unused;
1773 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001774 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775}
1776
1777/*
1778 * Verify that a swap entry is valid and increment its swap map count.
1779 *
1780 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1781 * "permanent", but will be reclaimed by the next swapoff.
1782 */
1783int swap_duplicate(swp_entry_t entry)
1784{
1785 struct swap_info_struct * p;
1786 unsigned long offset, type;
1787 int result = 0;
1788
Christoph Lameter06972122006-06-23 02:03:35 -07001789 if (is_migration_entry(entry))
1790 return 1;
1791
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 type = swp_type(entry);
1793 if (type >= nr_swapfiles)
1794 goto bad_file;
1795 p = type + swap_info;
1796 offset = swp_offset(entry);
1797
Hugh Dickins5d337b92005-09-03 15:54:41 -07001798 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 if (offset < p->max && p->swap_map[offset]) {
1800 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1801 p->swap_map[offset]++;
1802 result = 1;
1803 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1804 if (swap_overflow++ < 5)
1805 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1806 p->swap_map[offset] = SWAP_MAP_MAX;
1807 result = 1;
1808 }
1809 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07001810 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811out:
1812 return result;
1813
1814bad_file:
1815 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1816 goto out;
1817}
1818
1819struct swap_info_struct *
1820get_swap_info_struct(unsigned type)
1821{
1822 return &swap_info[type];
1823}
1824
1825/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001826 * swap_lock prevents swap_map being freed. Don't grab an extra
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 * reference on the swaphandle, it doesn't matter if it becomes unused.
1828 */
1829int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1830{
Hugh Dickins89528982008-02-04 22:28:45 -08001831 struct swap_info_struct *si;
Hugh Dickins3f9e7942006-09-29 02:01:26 -07001832 int our_page_cluster = page_cluster;
Hugh Dickins89528982008-02-04 22:28:45 -08001833 pgoff_t target, toff;
1834 pgoff_t base, end;
1835 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
Hugh Dickins3f9e7942006-09-29 02:01:26 -07001837 if (!our_page_cluster) /* no readahead */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 return 0;
Hugh Dickins89528982008-02-04 22:28:45 -08001839
1840 si = &swap_info[swp_type(entry)];
1841 target = swp_offset(entry);
1842 base = (target >> our_page_cluster) << our_page_cluster;
1843 end = base + (1 << our_page_cluster);
1844 if (!base) /* first page is swap header */
1845 base++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846
Hugh Dickins5d337b92005-09-03 15:54:41 -07001847 spin_lock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08001848 if (end > si->max) /* don't go beyond end of map */
1849 end = si->max;
1850
1851 /* Count contiguous allocated slots above our target */
1852 for (toff = target; ++toff < end; nr_pages++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 /* Don't read in free or bad pages */
Hugh Dickins89528982008-02-04 22:28:45 -08001854 if (!si->swap_map[toff])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 break;
Hugh Dickins89528982008-02-04 22:28:45 -08001856 if (si->swap_map[toff] == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 break;
Hugh Dickins89528982008-02-04 22:28:45 -08001858 }
1859 /* Count contiguous allocated slots below our target */
1860 for (toff = target; --toff >= base; nr_pages++) {
1861 /* Don't read in free or bad pages */
1862 if (!si->swap_map[toff])
1863 break;
1864 if (si->swap_map[toff] == SWAP_MAP_BAD)
1865 break;
1866 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07001867 spin_unlock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08001868
1869 /*
1870 * Indicate starting offset, and return number of pages to get:
1871 * if only 1, say 0, since there's then no readahead to be done.
1872 */
1873 *offset = ++toff;
1874 return nr_pages? ++nr_pages: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875}