blob: eade24da9310d4cc4e7499e5580d52fbb929701c [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>
30
31#include <asm/pgtable.h>
32#include <asm/tlbflush.h>
33#include <linux/swapops.h>
34
Hugh Dickins5d337b92005-09-03 15:54:41 -070035DEFINE_SPINLOCK(swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070036unsigned int nr_swapfiles;
37long total_swap_pages;
38static int swap_overflow;
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040static const char Bad_file[] = "Bad swap file entry ";
41static const char Unused_file[] = "Unused swap file entry ";
42static const char Bad_offset[] = "Bad swap offset entry ";
43static const char Unused_offset[] = "Unused swap offset entry ";
44
45struct swap_list_t swap_list = {-1, -1};
46
Rafael J. Wysockif577eb32006-03-23 02:59:59 -080047static struct swap_info_struct swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Ingo Molnarfc0abb12006-01-18 17:42:33 -080049static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
51/*
52 * We need this because the bdev->unplug_fn can sleep and we cannot
Hugh Dickins5d337b92005-09-03 15:54:41 -070053 * hold swap_lock while calling the unplug_fn. And swap_lock
Ingo Molnarfc0abb12006-01-18 17:42:33 -080054 * cannot be turned into a mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 */
56static DECLARE_RWSEM(swap_unplug_sem);
57
Linus Torvalds1da177e2005-04-16 15:20:36 -070058void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
59{
60 swp_entry_t entry;
61
62 down_read(&swap_unplug_sem);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070063 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 if (PageSwapCache(page)) {
65 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
66 struct backing_dev_info *bdi;
67
68 /*
69 * If the page is removed from swapcache from under us (with a
70 * racy try_to_unuse/swapoff) we need an additional reference
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070071 * count to avoid reading garbage from page_private(page) above.
72 * If the WARN_ON triggers during a swapoff it maybe the race
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 * condition and it's harmless. However if it triggers without
74 * swapoff it signals a problem.
75 */
76 WARN_ON(page_count(page) <= 1);
77
78 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
McMullan, Jasonba323112005-05-16 21:53:40 -070079 blk_run_backing_dev(bdi, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 }
81 up_read(&swap_unplug_sem);
82}
83
Hugh Dickins048c27f2005-09-03 15:54:40 -070084#define SWAPFILE_CLUSTER 256
85#define LATENCY_LIMIT 256
86
Hugh Dickins6eb396d2005-09-03 15:54:35 -070087static inline unsigned long scan_swap_map(struct swap_info_struct *si)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
Hugh Dickins7dfad412005-09-03 15:54:38 -070089 unsigned long offset, last_in_cluster;
Hugh Dickins048c27f2005-09-03 15:54:40 -070090 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Hugh Dickins7dfad412005-09-03 15:54:38 -070092 /*
93 * We try to cluster swap pages by allocating them sequentially
94 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
95 * way, however, we resort to first-free allocation, starting
96 * a new cluster. This prevents us from scattering swap pages
97 * all over the entire swap partition, so that we reduce
98 * overall disk seek times between swap pages. -- sct
99 * But we do now try to find an empty cluster. -Andrea
100 */
101
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700102 si->flags += SWP_SCANNING;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700103 if (unlikely(!si->cluster_nr)) {
104 si->cluster_nr = SWAPFILE_CLUSTER - 1;
105 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER)
106 goto lowest;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700107 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700108
109 offset = si->lowest_bit;
110 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
111
112 /* Locate the first empty (unaligned) cluster */
113 for (; last_in_cluster <= si->highest_bit; offset++) {
114 if (si->swap_map[offset])
115 last_in_cluster = offset + SWAPFILE_CLUSTER;
116 else if (offset == last_in_cluster) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700117 spin_lock(&swap_lock);
Akinobu Mita9b65ef52006-03-22 00:09:09 -0800118 si->cluster_next = offset-SWAPFILE_CLUSTER+1;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700119 goto cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700121 if (unlikely(--latency_ration < 0)) {
122 cond_resched();
123 latency_ration = LATENCY_LIMIT;
124 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700125 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700126 spin_lock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700127 goto lowest;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700129
130 si->cluster_nr--;
131cluster:
132 offset = si->cluster_next;
133 if (offset > si->highest_bit)
134lowest: offset = si->lowest_bit;
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700135checks: if (!(si->flags & SWP_WRITEOK))
136 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700137 if (!si->highest_bit)
138 goto no_page;
139 if (!si->swap_map[offset]) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700140 if (offset == si->lowest_bit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 si->lowest_bit++;
142 if (offset == si->highest_bit)
143 si->highest_bit--;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700144 si->inuse_pages++;
145 if (si->inuse_pages == si->pages) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 si->lowest_bit = si->max;
147 si->highest_bit = 0;
148 }
149 si->swap_map[offset] = 1;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700150 si->cluster_next = offset + 1;
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700151 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 return offset;
153 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700154
Hugh Dickins5d337b92005-09-03 15:54:41 -0700155 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700156 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700157 if (!si->swap_map[offset]) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700158 spin_lock(&swap_lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700159 goto checks;
160 }
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 Dickins5d337b92005-09-03 15:54:41 -0700166 spin_lock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700167 goto lowest;
168
169no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700170 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 return 0;
172}
173
174swp_entry_t get_swap_page(void)
175{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700176 struct swap_info_struct *si;
177 pgoff_t offset;
178 int type, next;
179 int wrapped = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Hugh Dickins5d337b92005-09-03 15:54:41 -0700181 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 if (nr_swap_pages <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700183 goto noswap;
184 nr_swap_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700186 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
187 si = swap_info + type;
188 next = si->next;
189 if (next < 0 ||
190 (!wrapped && si->prio != swap_info[next].prio)) {
191 next = swap_list.head;
192 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700194
195 if (!si->highest_bit)
196 continue;
197 if (!(si->flags & SWP_WRITEOK))
198 continue;
199
200 swap_list.next = next;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700201 offset = scan_swap_map(si);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700202 if (offset) {
203 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700204 return swp_entry(type, offset);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700205 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700206 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700208
209 nr_swap_pages++;
210noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700211 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700212 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213}
214
Rafael J. Wysocki3a291a22006-01-06 00:16:37 -0800215swp_entry_t get_swap_page_of_type(int type)
216{
217 struct swap_info_struct *si;
218 pgoff_t offset;
219
220 spin_lock(&swap_lock);
221 si = swap_info + type;
222 if (si->flags & SWP_WRITEOK) {
223 nr_swap_pages--;
224 offset = scan_swap_map(si);
225 if (offset) {
226 spin_unlock(&swap_lock);
227 return swp_entry(type, offset);
228 }
229 nr_swap_pages++;
230 }
231 spin_unlock(&swap_lock);
232 return (swp_entry_t) {0};
233}
234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235static struct swap_info_struct * swap_info_get(swp_entry_t entry)
236{
237 struct swap_info_struct * p;
238 unsigned long offset, type;
239
240 if (!entry.val)
241 goto out;
242 type = swp_type(entry);
243 if (type >= nr_swapfiles)
244 goto bad_nofile;
245 p = & swap_info[type];
246 if (!(p->flags & SWP_USED))
247 goto bad_device;
248 offset = swp_offset(entry);
249 if (offset >= p->max)
250 goto bad_offset;
251 if (!p->swap_map[offset])
252 goto bad_free;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700253 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 return p;
255
256bad_free:
257 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
258 goto out;
259bad_offset:
260 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
261 goto out;
262bad_device:
263 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
264 goto out;
265bad_nofile:
266 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
267out:
268 return NULL;
269}
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
272{
273 int count = p->swap_map[offset];
274
275 if (count < SWAP_MAP_MAX) {
276 count--;
277 p->swap_map[offset] = count;
278 if (!count) {
279 if (offset < p->lowest_bit)
280 p->lowest_bit = offset;
281 if (offset > p->highest_bit)
282 p->highest_bit = offset;
Hugh Dickins89d09a22005-09-03 15:54:36 -0700283 if (p->prio > swap_info[swap_list.next].prio)
284 swap_list.next = p - swap_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 nr_swap_pages++;
286 p->inuse_pages--;
287 }
288 }
289 return count;
290}
291
292/*
293 * Caller has made sure that the swapdevice corresponding to entry
294 * is still around or has not been recycled.
295 */
296void swap_free(swp_entry_t entry)
297{
298 struct swap_info_struct * p;
299
300 p = swap_info_get(entry);
301 if (p) {
302 swap_entry_free(p, swp_offset(entry));
Hugh Dickins5d337b92005-09-03 15:54:41 -0700303 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 }
305}
306
307/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700308 * How many references to page are currently swapped out?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700310static inline int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700312 int count = 0;
313 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 swp_entry_t entry;
315
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700316 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 p = swap_info_get(entry);
318 if (p) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700319 /* Subtract the 1 for the swap cache itself */
320 count = p->swap_map[swp_offset(entry)] - 1;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700321 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700323 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324}
325
326/*
327 * We can use this swap cache entry directly
328 * if there are no other references to it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 */
330int can_share_swap_page(struct page *page)
331{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700332 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700334 BUG_ON(!PageLocked(page));
335 count = page_mapcount(page);
336 if (count <= 1 && PageSwapCache(page))
337 count += page_swapcount(page);
338 return count == 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339}
340
341/*
342 * Work out if there are any other processes sharing this
343 * swap cache page. Free it if you can. Return success.
344 */
345int remove_exclusive_swap_page(struct page *page)
346{
347 int retval;
348 struct swap_info_struct * p;
349 swp_entry_t entry;
350
351 BUG_ON(PagePrivate(page));
352 BUG_ON(!PageLocked(page));
353
354 if (!PageSwapCache(page))
355 return 0;
356 if (PageWriteback(page))
357 return 0;
358 if (page_count(page) != 2) /* 2: us + cache */
359 return 0;
360
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700361 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 p = swap_info_get(entry);
363 if (!p)
364 return 0;
365
366 /* Is the only swap cache user the cache itself? */
367 retval = 0;
368 if (p->swap_map[swp_offset(entry)] == 1) {
369 /* Recheck the page count with the swapcache lock held.. */
370 write_lock_irq(&swapper_space.tree_lock);
371 if ((page_count(page) == 2) && !PageWriteback(page)) {
372 __delete_from_swap_cache(page);
373 SetPageDirty(page);
374 retval = 1;
375 }
376 write_unlock_irq(&swapper_space.tree_lock);
377 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700378 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
380 if (retval) {
381 swap_free(entry);
382 page_cache_release(page);
383 }
384
385 return retval;
386}
387
388/*
389 * Free the swap entry like above, but also try to
390 * free the page cache entry if it is the last user.
391 */
392void free_swap_and_cache(swp_entry_t entry)
393{
394 struct swap_info_struct * p;
395 struct page *page = NULL;
396
Christoph Lameter06972122006-06-23 02:03:35 -0700397 if (is_migration_entry(entry))
398 return;
399
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 p = swap_info_get(entry);
401 if (p) {
Nick Piggin93fac702006-03-31 02:29:56 -0800402 if (swap_entry_free(p, swp_offset(entry)) == 1) {
403 page = find_get_page(&swapper_space, entry.val);
404 if (page && unlikely(TestSetPageLocked(page))) {
405 page_cache_release(page);
406 page = NULL;
407 }
408 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700409 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 }
411 if (page) {
412 int one_user;
413
414 BUG_ON(PagePrivate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 one_user = (page_count(page) == 2);
416 /* Only cache user (+us), or swap space full? Free it! */
Nick Piggin93fac702006-03-31 02:29:56 -0800417 /* Also recheck PageSwapCache after page is locked (above) */
418 if (PageSwapCache(page) && !PageWriteback(page) &&
419 (one_user || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 delete_from_swap_cache(page);
421 SetPageDirty(page);
422 }
423 unlock_page(page);
424 page_cache_release(page);
425 }
426}
427
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200428#ifdef CONFIG_HIBERNATION
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800429/*
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800430 * Find the swap type that corresponds to given device (if any).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800431 *
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800432 * @offset - number of the PAGE_SIZE-sized block of the device, starting
433 * from 0, in which the swap header is expected to be located.
434 *
435 * This is needed for the suspend to disk (aka swsusp).
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800436 */
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800437int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800438{
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800439 struct block_device *bdev = NULL;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800440 int i;
441
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800442 if (device)
443 bdev = bdget(device);
444
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800445 spin_lock(&swap_lock);
446 for (i = 0; i < nr_swapfiles; i++) {
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800447 struct swap_info_struct *sis = swap_info + i;
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800448
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800449 if (!(sis->flags & SWP_WRITEOK))
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800450 continue;
Rafael J. Wysockib6b5bce2006-08-27 01:23:25 -0700451
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800452 if (!bdev) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800453 if (bdev_p)
454 *bdev_p = sis->bdev;
455
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800456 spin_unlock(&swap_lock);
457 return i;
458 }
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800459 if (bdev == sis->bdev) {
460 struct swap_extent *se;
461
462 se = list_entry(sis->extent_list.next,
463 struct swap_extent, list);
464 if (se->start_block == offset) {
Rafael J. Wysocki7bf23682007-01-05 16:36:28 -0800465 if (bdev_p)
466 *bdev_p = sis->bdev;
467
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800468 spin_unlock(&swap_lock);
469 bdput(bdev);
470 return i;
471 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800472 }
473 }
474 spin_unlock(&swap_lock);
Rafael J. Wysocki915bae92006-12-06 20:34:07 -0800475 if (bdev)
476 bdput(bdev);
477
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800478 return -ENODEV;
479}
480
481/*
482 * Return either the total number of swap pages of given type, or the number
483 * of free pages of that type (depending on @free)
484 *
485 * This is needed for software suspend
486 */
487unsigned int count_swap_pages(int type, int free)
488{
489 unsigned int n = 0;
490
491 if (type < nr_swapfiles) {
492 spin_lock(&swap_lock);
493 if (swap_info[type].flags & SWP_WRITEOK) {
494 n = swap_info[type].pages;
495 if (free)
496 n -= swap_info[type].inuse_pages;
497 }
498 spin_unlock(&swap_lock);
499 }
500 return n;
501}
502#endif
503
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700505 * No need to decide whether this PTE shares the swap entry with others,
506 * just let do_wp_page work it out if a write is requested later - to
507 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 */
Hugh Dickins2e441882008-02-04 22:28:46 -0800509static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 unsigned long addr, swp_entry_t entry, struct page *page)
511{
Hugh Dickins2e441882008-02-04 22:28:46 -0800512 spinlock_t *ptl;
513 pte_t *pte;
514 int found = 1;
515
516 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
517 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
518 found = 0;
519 goto out;
520 }
521
Hugh Dickins42946212005-10-29 18:16:05 -0700522 inc_mm_counter(vma->vm_mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 get_page(page);
524 set_pte_at(vma->vm_mm, addr, pte,
525 pte_mkold(mk_pte(page, vma->vm_page_prot)));
526 page_add_anon_rmap(page, vma, addr);
527 swap_free(entry);
528 /*
529 * Move the page to the active list so it is not
530 * immediately swapped out again after swapon.
531 */
532 activate_page(page);
Hugh Dickins2e441882008-02-04 22:28:46 -0800533out:
534 pte_unmap_unlock(pte, ptl);
535 return found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537
538static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
539 unsigned long addr, unsigned long end,
540 swp_entry_t entry, struct page *page)
541{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700543 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700544 int found = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Hugh Dickins2e441882008-02-04 22:28:46 -0800546 /*
547 * We don't actually need pte lock while scanning for swp_pte: since
548 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
549 * page table while we're scanning; though it could get zapped, and on
550 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
551 * of unmatched parts which look like swp_pte, so unuse_pte must
552 * recheck under pte lock. Scanning without pte lock lets it be
553 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
554 */
555 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 do {
557 /*
558 * swapoff spends a _lot_ of time in this loop!
559 * Test inline before going to call unuse_pte.
560 */
561 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins2e441882008-02-04 22:28:46 -0800562 pte_unmap(pte);
563 found = unuse_pte(vma, pmd, addr, entry, page);
564 if (found)
565 goto out;
566 pte = pte_offset_map(pmd, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 }
568 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins2e441882008-02-04 22:28:46 -0800569 pte_unmap(pte - 1);
570out:
Hugh Dickins705e87c2005-10-29 18:16:27 -0700571 return found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572}
573
574static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
575 unsigned long addr, unsigned long end,
576 swp_entry_t entry, struct page *page)
577{
578 pmd_t *pmd;
579 unsigned long next;
580
581 pmd = pmd_offset(pud, addr);
582 do {
583 next = pmd_addr_end(addr, end);
584 if (pmd_none_or_clear_bad(pmd))
585 continue;
586 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
587 return 1;
588 } while (pmd++, addr = next, addr != end);
589 return 0;
590}
591
592static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
593 unsigned long addr, unsigned long end,
594 swp_entry_t entry, struct page *page)
595{
596 pud_t *pud;
597 unsigned long next;
598
599 pud = pud_offset(pgd, addr);
600 do {
601 next = pud_addr_end(addr, end);
602 if (pud_none_or_clear_bad(pud))
603 continue;
604 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
605 return 1;
606 } while (pud++, addr = next, addr != end);
607 return 0;
608}
609
610static int unuse_vma(struct vm_area_struct *vma,
611 swp_entry_t entry, struct page *page)
612{
613 pgd_t *pgd;
614 unsigned long addr, end, next;
615
616 if (page->mapping) {
617 addr = page_address_in_vma(page, vma);
618 if (addr == -EFAULT)
619 return 0;
620 else
621 end = addr + PAGE_SIZE;
622 } else {
623 addr = vma->vm_start;
624 end = vma->vm_end;
625 }
626
627 pgd = pgd_offset(vma->vm_mm, addr);
628 do {
629 next = pgd_addr_end(addr, end);
630 if (pgd_none_or_clear_bad(pgd))
631 continue;
632 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
633 return 1;
634 } while (pgd++, addr = next, addr != end);
635 return 0;
636}
637
638static int unuse_mm(struct mm_struct *mm,
639 swp_entry_t entry, struct page *page)
640{
641 struct vm_area_struct *vma;
642
643 if (!down_read_trylock(&mm->mmap_sem)) {
644 /*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700645 * Activate page so shrink_cache is unlikely to unmap its
646 * ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700648 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 unlock_page(page);
650 down_read(&mm->mmap_sem);
651 lock_page(page);
652 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 for (vma = mm->mmap; vma; vma = vma->vm_next) {
654 if (vma->anon_vma && unuse_vma(vma, entry, page))
655 break;
656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 up_read(&mm->mmap_sem);
658 /*
659 * Currently unuse_mm cannot fail, but leave error handling
660 * at call sites for now, since we change it from time to time.
661 */
662 return 0;
663}
664
665/*
666 * Scan swap_map from current position to next entry still in use.
667 * Recycle to start on reaching the end, returning 0 when empty.
668 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700669static unsigned int find_next_to_unuse(struct swap_info_struct *si,
670 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671{
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700672 unsigned int max = si->max;
673 unsigned int i = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 int count;
675
676 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -0700677 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 * for whether an entry is in use, not modifying it; false
679 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -0700680 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 */
682 for (;;) {
683 if (++i >= max) {
684 if (!prev) {
685 i = 0;
686 break;
687 }
688 /*
689 * No entries in use at top of swap_map,
690 * loop back to start and recheck there.
691 */
692 max = prev + 1;
693 prev = 0;
694 i = 1;
695 }
696 count = si->swap_map[i];
697 if (count && count != SWAP_MAP_BAD)
698 break;
699 }
700 return i;
701}
702
703/*
704 * We completely avoid races by reading each swap page in advance,
705 * and then search for the process using it. All the necessary
706 * page table adjustments can then be made atomically.
707 */
708static int try_to_unuse(unsigned int type)
709{
710 struct swap_info_struct * si = &swap_info[type];
711 struct mm_struct *start_mm;
712 unsigned short *swap_map;
713 unsigned short swcount;
714 struct page *page;
715 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700716 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 int retval = 0;
718 int reset_overflow = 0;
719 int shmem;
720
721 /*
722 * When searching mms for an entry, a good strategy is to
723 * start at the first mm we freed the previous entry from
724 * (though actually we don't notice whether we or coincidence
725 * freed the entry). Initialize this start_mm with a hold.
726 *
727 * A simpler strategy would be to start at the last mm we
728 * freed the previous entry from; but that would take less
729 * advantage of mmlist ordering, which clusters forked mms
730 * together, child after parent. If we race with dup_mmap(), we
731 * prefer to resolve parent before child, lest we miss entries
732 * duplicated after we scanned child: using last mm would invert
733 * that. Though it's only a serious concern when an overflowed
734 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
735 */
736 start_mm = &init_mm;
737 atomic_inc(&init_mm.mm_users);
738
739 /*
740 * Keep on scanning until all entries have gone. Usually,
741 * one pass through swap_map is enough, but not necessarily:
742 * there are races when an instance of an entry might be missed.
743 */
744 while ((i = find_next_to_unuse(si, i)) != 0) {
745 if (signal_pending(current)) {
746 retval = -EINTR;
747 break;
748 }
749
750 /*
751 * Get a page for the entry, using the existing swap
752 * cache page if there is one. Otherwise, get a clean
753 * page and read the swap into it.
754 */
755 swap_map = &si->swap_map[i];
756 entry = swp_entry(type, i);
Hugh Dickins02098fe2008-02-04 22:28:42 -0800757 page = read_swap_cache_async(entry,
758 GFP_HIGHUSER_MOVABLE, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 if (!page) {
760 /*
761 * Either swap_duplicate() failed because entry
762 * has been freed independently, and will not be
763 * reused since sys_swapoff() already disabled
764 * allocation from here, or alloc_page() failed.
765 */
766 if (!*swap_map)
767 continue;
768 retval = -ENOMEM;
769 break;
770 }
771
772 /*
773 * Don't hold on to start_mm if it looks like exiting.
774 */
775 if (atomic_read(&start_mm->mm_users) == 1) {
776 mmput(start_mm);
777 start_mm = &init_mm;
778 atomic_inc(&init_mm.mm_users);
779 }
780
781 /*
782 * Wait for and lock page. When do_swap_page races with
783 * try_to_unuse, do_swap_page can handle the fault much
784 * faster than try_to_unuse can locate the entry. This
785 * apparently redundant "wait_on_page_locked" lets try_to_unuse
786 * defer to do_swap_page in such a case - in some tests,
787 * do_swap_page and try_to_unuse repeatedly compete.
788 */
789 wait_on_page_locked(page);
790 wait_on_page_writeback(page);
791 lock_page(page);
792 wait_on_page_writeback(page);
793
794 /*
795 * Remove all references to entry.
796 * Whenever we reach init_mm, there's no address space
797 * to search, but use it as a reminder to search shmem.
798 */
799 shmem = 0;
800 swcount = *swap_map;
801 if (swcount > 1) {
802 if (start_mm == &init_mm)
803 shmem = shmem_unuse(entry, page);
804 else
805 retval = unuse_mm(start_mm, entry, page);
806 }
807 if (*swap_map > 1) {
808 int set_start_mm = (*swap_map >= swcount);
809 struct list_head *p = &start_mm->mmlist;
810 struct mm_struct *new_start_mm = start_mm;
811 struct mm_struct *prev_mm = start_mm;
812 struct mm_struct *mm;
813
814 atomic_inc(&new_start_mm->mm_users);
815 atomic_inc(&prev_mm->mm_users);
816 spin_lock(&mmlist_lock);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800817 while (*swap_map > 1 && !retval && !shmem &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 (p = p->next) != &start_mm->mmlist) {
819 mm = list_entry(p, struct mm_struct, mmlist);
Hugh Dickins70af7c52006-06-23 02:03:44 -0700820 if (!atomic_inc_not_zero(&mm->mm_users))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 spin_unlock(&mmlist_lock);
823 mmput(prev_mm);
824 prev_mm = mm;
825
826 cond_resched();
827
828 swcount = *swap_map;
829 if (swcount <= 1)
830 ;
831 else if (mm == &init_mm) {
832 set_start_mm = 1;
833 shmem = shmem_unuse(entry, page);
834 } else
835 retval = unuse_mm(mm, entry, page);
836 if (set_start_mm && *swap_map < swcount) {
837 mmput(new_start_mm);
838 atomic_inc(&mm->mm_users);
839 new_start_mm = mm;
840 set_start_mm = 0;
841 }
842 spin_lock(&mmlist_lock);
843 }
844 spin_unlock(&mmlist_lock);
845 mmput(prev_mm);
846 mmput(start_mm);
847 start_mm = new_start_mm;
848 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800849 if (shmem) {
850 /* page has already been unlocked and released */
851 if (shmem > 0)
852 continue;
853 retval = shmem;
854 break;
855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 if (retval) {
857 unlock_page(page);
858 page_cache_release(page);
859 break;
860 }
861
862 /*
863 * How could swap count reach 0x7fff when the maximum
864 * pid is 0x7fff, and there's no way to repeat a swap
865 * page within an mm (except in shmem, where it's the
866 * shared object which takes the reference count)?
867 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
868 *
869 * If that's wrong, then we should worry more about
870 * exit_mmap() and do_munmap() cases described above:
871 * we might be resetting SWAP_MAP_MAX too early here.
872 * We know "Undead"s can happen, they're okay, so don't
873 * report them; but do report if we reset SWAP_MAP_MAX.
874 */
875 if (*swap_map == SWAP_MAP_MAX) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700876 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 *swap_map = 1;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700878 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 reset_overflow = 1;
880 }
881
882 /*
883 * If a reference remains (rare), we would like to leave
884 * the page in the swap cache; but try_to_unmap could
885 * then re-duplicate the entry once we drop page lock,
886 * so we might loop indefinitely; also, that page could
887 * not be swapped out to other storage meanwhile. So:
888 * delete from cache even if there's another reference,
889 * after ensuring that the data has been saved to disk -
890 * since if the reference remains (rarer), it will be
891 * read from disk into another page. Splitting into two
892 * pages would be incorrect if swap supported "shared
893 * private" pages, but they are handled by tmpfs files.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 */
895 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
896 struct writeback_control wbc = {
897 .sync_mode = WB_SYNC_NONE,
898 };
899
900 swap_writepage(page, &wbc);
901 lock_page(page);
902 wait_on_page_writeback(page);
903 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800904 if (PageSwapCache(page))
905 delete_from_swap_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
907 /*
908 * So we could skip searching mms once swap count went
909 * to 1, we did not mark any present ptes as dirty: must
Anderson Briglia2706a1b2007-07-15 23:38:09 -0700910 * mark page dirty so shrink_page_list will preserve it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 */
912 SetPageDirty(page);
913 unlock_page(page);
914 page_cache_release(page);
915
916 /*
917 * Make sure that we aren't completely killing
918 * interactive performance.
919 */
920 cond_resched();
921 }
922
923 mmput(start_mm);
924 if (reset_overflow) {
925 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
926 swap_overflow = 0;
927 }
928 return retval;
929}
930
931/*
Hugh Dickins5d337b92005-09-03 15:54:41 -0700932 * After a successful try_to_unuse, if no swap is now in use, we know
933 * we can empty the mmlist. swap_lock must be held on entry and exit.
934 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 * added to the mmlist just after page_duplicate - before would be racy.
936 */
937static void drain_mmlist(void)
938{
939 struct list_head *p, *next;
940 unsigned int i;
941
942 for (i = 0; i < nr_swapfiles; i++)
943 if (swap_info[i].inuse_pages)
944 return;
945 spin_lock(&mmlist_lock);
946 list_for_each_safe(p, next, &init_mm.mmlist)
947 list_del_init(p);
948 spin_unlock(&mmlist_lock);
949}
950
951/*
952 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
953 * corresponds to page offset `offset'.
954 */
955sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
956{
957 struct swap_extent *se = sis->curr_swap_extent;
958 struct swap_extent *start_se = se;
959
960 for ( ; ; ) {
961 struct list_head *lh;
962
963 if (se->start_page <= offset &&
964 offset < (se->start_page + se->nr_pages)) {
965 return se->start_block + (offset - se->start_page);
966 }
Hugh Dickins11d31882005-09-03 15:54:34 -0700967 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 if (lh == &sis->extent_list)
Hugh Dickins11d31882005-09-03 15:54:34 -0700969 lh = lh->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 se = list_entry(lh, struct swap_extent, list);
971 sis->curr_swap_extent = se;
972 BUG_ON(se == start_se); /* It *must* be present */
973 }
974}
975
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200976#ifdef CONFIG_HIBERNATION
Rafael J. Wysocki3aef83e2006-12-06 20:34:10 -0800977/*
978 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
979 * corresponding to given index in swap_info (swap type).
980 */
981sector_t swapdev_block(int swap_type, pgoff_t offset)
982{
983 struct swap_info_struct *sis;
984
985 if (swap_type >= nr_swapfiles)
986 return 0;
987
988 sis = swap_info + swap_type;
989 return (sis->flags & SWP_WRITEOK) ? map_swap_page(sis, offset) : 0;
990}
Rafael J. Wysockib0cb1a12007-07-29 23:24:36 +0200991#endif /* CONFIG_HIBERNATION */
Rafael J. Wysocki3aef83e2006-12-06 20:34:10 -0800992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993/*
994 * Free all of a swapdev's extent information
995 */
996static void destroy_swap_extents(struct swap_info_struct *sis)
997{
998 while (!list_empty(&sis->extent_list)) {
999 struct swap_extent *se;
1000
1001 se = list_entry(sis->extent_list.next,
1002 struct swap_extent, list);
1003 list_del(&se->list);
1004 kfree(se);
1005 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006}
1007
1008/*
1009 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -07001010 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 *
Hugh Dickins11d31882005-09-03 15:54:34 -07001012 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 */
1014static int
1015add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1016 unsigned long nr_pages, sector_t start_block)
1017{
1018 struct swap_extent *se;
1019 struct swap_extent *new_se;
1020 struct list_head *lh;
1021
Hugh Dickins11d31882005-09-03 15:54:34 -07001022 lh = sis->extent_list.prev; /* The highest page extent */
1023 if (lh != &sis->extent_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -07001025 BUG_ON(se->start_page + se->nr_pages != start_page);
1026 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 /* Merge it */
1028 se->nr_pages += nr_pages;
1029 return 0;
1030 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 }
1032
1033 /*
1034 * No merge. Insert a new extent, preserving ordering.
1035 */
1036 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1037 if (new_se == NULL)
1038 return -ENOMEM;
1039 new_se->start_page = start_page;
1040 new_se->nr_pages = nr_pages;
1041 new_se->start_block = start_block;
1042
Hugh Dickins11d31882005-09-03 15:54:34 -07001043 list_add_tail(&new_se->list, &sis->extent_list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001044 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045}
1046
1047/*
1048 * A `swap extent' is a simple thing which maps a contiguous range of pages
1049 * onto a contiguous range of disk blocks. An ordered list of swap extents
1050 * is built at swapon time and is then used at swap_writepage/swap_readpage
1051 * time for locating where on disk a page belongs.
1052 *
1053 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1054 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1055 * swap files identically.
1056 *
1057 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1058 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1059 * swapfiles are handled *identically* after swapon time.
1060 *
1061 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1062 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1063 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1064 * requirements, they are simply tossed out - we will never use those blocks
1065 * for swapping.
1066 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001067 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1069 * which will scribble on the fs.
1070 *
1071 * The amount of disk space which a single swap extent represents varies.
1072 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1073 * extents in the list. To avoid much list walking, we cache the previous
1074 * search location in `curr_swap_extent', and start new searches from there.
1075 * This is extremely effective. The average number of iterations in
1076 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1077 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001078static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079{
1080 struct inode *inode;
1081 unsigned blocks_per_page;
1082 unsigned long page_no;
1083 unsigned blkbits;
1084 sector_t probe_block;
1085 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001086 sector_t lowest_block = -1;
1087 sector_t highest_block = 0;
1088 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 int ret;
1090
1091 inode = sis->swap_file->f_mapping->host;
1092 if (S_ISBLK(inode->i_mode)) {
1093 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001094 *span = sis->pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 goto done;
1096 }
1097
1098 blkbits = inode->i_blkbits;
1099 blocks_per_page = PAGE_SIZE >> blkbits;
1100
1101 /*
1102 * Map all the blocks into the extent list. This code doesn't try
1103 * to be very smart.
1104 */
1105 probe_block = 0;
1106 page_no = 0;
1107 last_block = i_size_read(inode) >> blkbits;
1108 while ((probe_block + blocks_per_page) <= last_block &&
1109 page_no < sis->max) {
1110 unsigned block_in_page;
1111 sector_t first_block;
1112
1113 first_block = bmap(inode, probe_block);
1114 if (first_block == 0)
1115 goto bad_bmap;
1116
1117 /*
1118 * It must be PAGE_SIZE aligned on-disk
1119 */
1120 if (first_block & (blocks_per_page - 1)) {
1121 probe_block++;
1122 goto reprobe;
1123 }
1124
1125 for (block_in_page = 1; block_in_page < blocks_per_page;
1126 block_in_page++) {
1127 sector_t block;
1128
1129 block = bmap(inode, probe_block + block_in_page);
1130 if (block == 0)
1131 goto bad_bmap;
1132 if (block != first_block + block_in_page) {
1133 /* Discontiguity */
1134 probe_block++;
1135 goto reprobe;
1136 }
1137 }
1138
Hugh Dickins53092a72005-09-03 15:54:34 -07001139 first_block >>= (PAGE_SHIFT - blkbits);
1140 if (page_no) { /* exclude the header page */
1141 if (first_block < lowest_block)
1142 lowest_block = first_block;
1143 if (first_block > highest_block)
1144 highest_block = first_block;
1145 }
1146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 /*
1148 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1149 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001150 ret = add_swap_extent(sis, page_no, 1, first_block);
1151 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001153 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 page_no++;
1155 probe_block += blocks_per_page;
1156reprobe:
1157 continue;
1158 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001159 ret = nr_extents;
1160 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001162 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001164 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 sis->highest_bit = page_no - 1;
1166done:
1167 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1168 struct swap_extent, list);
1169 goto out;
1170bad_bmap:
1171 printk(KERN_ERR "swapon: swapfile has holes\n");
1172 ret = -EINVAL;
1173out:
1174 return ret;
1175}
1176
1177#if 0 /* We don't need this yet */
1178#include <linux/backing-dev.h>
1179int page_queue_congested(struct page *page)
1180{
1181 struct backing_dev_info *bdi;
1182
1183 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1184
1185 if (PageSwapCache(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001186 swp_entry_t entry = { .val = page_private(page) };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 struct swap_info_struct *sis;
1188
1189 sis = get_swap_info_struct(swp_type(entry));
1190 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1191 } else
1192 bdi = page->mapping->backing_dev_info;
1193 return bdi_write_congested(bdi);
1194}
1195#endif
1196
1197asmlinkage long sys_swapoff(const char __user * specialfile)
1198{
1199 struct swap_info_struct * p = NULL;
1200 unsigned short *swap_map;
1201 struct file *swap_file, *victim;
1202 struct address_space *mapping;
1203 struct inode *inode;
1204 char * pathname;
1205 int i, type, prev;
1206 int err;
1207
1208 if (!capable(CAP_SYS_ADMIN))
1209 return -EPERM;
1210
1211 pathname = getname(specialfile);
1212 err = PTR_ERR(pathname);
1213 if (IS_ERR(pathname))
1214 goto out;
1215
1216 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1217 putname(pathname);
1218 err = PTR_ERR(victim);
1219 if (IS_ERR(victim))
1220 goto out;
1221
1222 mapping = victim->f_mapping;
1223 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001224 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1226 p = swap_info + type;
1227 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1228 if (p->swap_file->f_mapping == mapping)
1229 break;
1230 }
1231 prev = type;
1232 }
1233 if (type < 0) {
1234 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001235 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 goto out_dput;
1237 }
1238 if (!security_vm_enough_memory(p->pages))
1239 vm_unacct_memory(p->pages);
1240 else {
1241 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001242 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 goto out_dput;
1244 }
1245 if (prev < 0) {
1246 swap_list.head = p->next;
1247 } else {
1248 swap_info[prev].next = p->next;
1249 }
1250 if (type == swap_list.next) {
1251 /* just pick something that's safe... */
1252 swap_list.next = swap_list.head;
1253 }
1254 nr_swap_pages -= p->pages;
1255 total_swap_pages -= p->pages;
1256 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001257 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 current->flags |= PF_SWAPOFF;
1260 err = try_to_unuse(type);
1261 current->flags &= ~PF_SWAPOFF;
1262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 if (err) {
1264 /* re-insert swap space back into swap_list */
Hugh Dickins5d337b92005-09-03 15:54:41 -07001265 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1267 if (p->prio >= swap_info[i].prio)
1268 break;
1269 p->next = i;
1270 if (prev < 0)
1271 swap_list.head = swap_list.next = p - swap_info;
1272 else
1273 swap_info[prev].next = p - swap_info;
1274 nr_swap_pages += p->pages;
1275 total_swap_pages += p->pages;
1276 p->flags |= SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001277 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 goto out_dput;
1279 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001280
1281 /* wait for any unplug function to finish */
1282 down_write(&swap_unplug_sem);
1283 up_write(&swap_unplug_sem);
1284
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001285 destroy_swap_extents(p);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001286 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001287 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001289
1290 /* wait for anyone still in scan_swap_map */
1291 p->highest_bit = 0; /* cuts scans short */
1292 while (p->flags >= SWP_SCANNING) {
1293 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001294 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001295 spin_lock(&swap_lock);
1296 }
1297
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 swap_file = p->swap_file;
1299 p->swap_file = NULL;
1300 p->max = 0;
1301 swap_map = p->swap_map;
1302 p->swap_map = NULL;
1303 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001304 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001305 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 vfree(swap_map);
1307 inode = mapping->host;
1308 if (S_ISBLK(inode->i_mode)) {
1309 struct block_device *bdev = I_BDEV(inode);
1310 set_blocksize(bdev, p->old_block_size);
1311 bd_release(bdev);
1312 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001313 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001315 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 }
1317 filp_close(swap_file, NULL);
1318 err = 0;
1319
1320out_dput:
1321 filp_close(victim, NULL);
1322out:
1323 return err;
1324}
1325
1326#ifdef CONFIG_PROC_FS
1327/* iterator */
1328static void *swap_start(struct seq_file *swap, loff_t *pos)
1329{
1330 struct swap_info_struct *ptr = swap_info;
1331 int i;
1332 loff_t l = *pos;
1333
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001334 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001336 if (!l)
1337 return SEQ_START_TOKEN;
1338
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1340 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1341 continue;
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001342 if (!--l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 return ptr;
1344 }
1345
1346 return NULL;
1347}
1348
1349static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1350{
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001351 struct swap_info_struct *ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1353
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001354 if (v == SEQ_START_TOKEN)
1355 ptr = swap_info;
1356 else {
1357 ptr = v;
1358 ptr++;
1359 }
1360
1361 for (; ptr < endptr; ptr++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1363 continue;
1364 ++*pos;
1365 return ptr;
1366 }
1367
1368 return NULL;
1369}
1370
1371static void swap_stop(struct seq_file *swap, void *v)
1372{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001373 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374}
1375
1376static int swap_show(struct seq_file *swap, void *v)
1377{
1378 struct swap_info_struct *ptr = v;
1379 struct file *file;
1380 int len;
1381
Suleiman Souhlal881e4aa2006-12-06 20:32:28 -08001382 if (ptr == SEQ_START_TOKEN) {
1383 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1384 return 0;
1385 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
1387 file = ptr->swap_file;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001388 len = seq_path(swap, file->f_path.mnt, file->f_path.dentry, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001389 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 len < 40 ? 40 - len : 1, " ",
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001391 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 "partition" : "file\t",
1393 ptr->pages << (PAGE_SHIFT - 10),
1394 ptr->inuse_pages << (PAGE_SHIFT - 10),
1395 ptr->prio);
1396 return 0;
1397}
1398
Helge Deller15ad7cd2006-12-06 20:40:36 -08001399static const struct seq_operations swaps_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 .start = swap_start,
1401 .next = swap_next,
1402 .stop = swap_stop,
1403 .show = swap_show
1404};
1405
1406static int swaps_open(struct inode *inode, struct file *file)
1407{
1408 return seq_open(file, &swaps_op);
1409}
1410
Helge Deller15ad7cd2006-12-06 20:40:36 -08001411static const struct file_operations proc_swaps_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 .open = swaps_open,
1413 .read = seq_read,
1414 .llseek = seq_lseek,
1415 .release = seq_release,
1416};
1417
1418static int __init procswaps_init(void)
1419{
1420 struct proc_dir_entry *entry;
1421
1422 entry = create_proc_entry("swaps", 0, NULL);
1423 if (entry)
1424 entry->proc_fops = &proc_swaps_operations;
1425 return 0;
1426}
1427__initcall(procswaps_init);
1428#endif /* CONFIG_PROC_FS */
1429
1430/*
1431 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1432 *
1433 * The swapon system call
1434 */
1435asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1436{
1437 struct swap_info_struct * p;
1438 char *name = NULL;
1439 struct block_device *bdev = NULL;
1440 struct file *swap_file = NULL;
1441 struct address_space *mapping;
1442 unsigned int type;
1443 int i, prev;
1444 int error;
1445 static int least_priority;
1446 union swap_header *swap_header = NULL;
1447 int swap_header_version;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001448 unsigned int nr_good_pages = 0;
1449 int nr_extents = 0;
Hugh Dickins53092a72005-09-03 15:54:34 -07001450 sector_t span;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 unsigned long maxpages = 1;
1452 int swapfilesize;
1453 unsigned short *swap_map;
1454 struct page *page = NULL;
1455 struct inode *inode = NULL;
1456 int did_down = 0;
1457
1458 if (!capable(CAP_SYS_ADMIN))
1459 return -EPERM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001460 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 p = swap_info;
1462 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1463 if (!(p->flags & SWP_USED))
1464 break;
1465 error = -EPERM;
Christoph Lameter06972122006-06-23 02:03:35 -07001466 if (type >= MAX_SWAPFILES) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001467 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 goto out;
1469 }
1470 if (type >= nr_swapfiles)
1471 nr_swapfiles = type+1;
1472 INIT_LIST_HEAD(&p->extent_list);
1473 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 p->swap_file = NULL;
1475 p->old_block_size = 0;
1476 p->swap_map = NULL;
1477 p->lowest_bit = 0;
1478 p->highest_bit = 0;
1479 p->cluster_nr = 0;
1480 p->inuse_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 p->next = -1;
1482 if (swap_flags & SWAP_FLAG_PREFER) {
1483 p->prio =
1484 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1485 } else {
1486 p->prio = --least_priority;
1487 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07001488 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 name = getname(specialfile);
1490 error = PTR_ERR(name);
1491 if (IS_ERR(name)) {
1492 name = NULL;
1493 goto bad_swap_2;
1494 }
1495 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1496 error = PTR_ERR(swap_file);
1497 if (IS_ERR(swap_file)) {
1498 swap_file = NULL;
1499 goto bad_swap_2;
1500 }
1501
1502 p->swap_file = swap_file;
1503 mapping = swap_file->f_mapping;
1504 inode = mapping->host;
1505
1506 error = -EBUSY;
1507 for (i = 0; i < nr_swapfiles; i++) {
1508 struct swap_info_struct *q = &swap_info[i];
1509
1510 if (i == type || !q->swap_file)
1511 continue;
1512 if (mapping == q->swap_file->f_mapping)
1513 goto bad_swap;
1514 }
1515
1516 error = -EINVAL;
1517 if (S_ISBLK(inode->i_mode)) {
1518 bdev = I_BDEV(inode);
1519 error = bd_claim(bdev, sys_swapon);
1520 if (error < 0) {
1521 bdev = NULL;
Rob Landleyf7b3a432005-09-22 21:44:27 -07001522 error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 goto bad_swap;
1524 }
1525 p->old_block_size = block_size(bdev);
1526 error = set_blocksize(bdev, PAGE_SIZE);
1527 if (error < 0)
1528 goto bad_swap;
1529 p->bdev = bdev;
1530 } else if (S_ISREG(inode->i_mode)) {
1531 p->bdev = inode->i_sb->s_bdev;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001532 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 did_down = 1;
1534 if (IS_SWAPFILE(inode)) {
1535 error = -EBUSY;
1536 goto bad_swap;
1537 }
1538 } else {
1539 goto bad_swap;
1540 }
1541
1542 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1543
1544 /*
1545 * Read the swap header.
1546 */
1547 if (!mapping->a_ops->readpage) {
1548 error = -EINVAL;
1549 goto bad_swap;
1550 }
Pekka Enberg090d2b12006-06-23 02:05:08 -07001551 page = read_mapping_page(mapping, 0, swap_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (IS_ERR(page)) {
1553 error = PTR_ERR(page);
1554 goto bad_swap;
1555 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 kmap(page);
1557 swap_header = page_address(page);
1558
1559 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1560 swap_header_version = 1;
1561 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1562 swap_header_version = 2;
1563 else {
Jesper Juhle97a3112006-01-11 01:50:28 +01001564 printk(KERN_ERR "Unable to find swap-space signature\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 error = -EINVAL;
1566 goto bad_swap;
1567 }
1568
1569 switch (swap_header_version) {
1570 case 1:
1571 printk(KERN_ERR "version 0 swap is no longer supported. "
1572 "Use mkswap -v1 %s\n", name);
1573 error = -EINVAL;
1574 goto bad_swap;
1575 case 2:
1576 /* Check the swap header's sub-version and the size of
1577 the swap file and bad block lists */
1578 if (swap_header->info.version != 1) {
1579 printk(KERN_WARNING
1580 "Unable to handle swap header version %d\n",
1581 swap_header->info.version);
1582 error = -EINVAL;
1583 goto bad_swap;
1584 }
1585
1586 p->lowest_bit = 1;
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001587 p->cluster_next = 1;
1588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 /*
1590 * Find out how many pages are allowed for a single swap
1591 * device. There are two limiting factors: 1) the number of
1592 * bits for the swap offset in the swp_entry_t type and
1593 * 2) the number of bits in the a swap pte as defined by
1594 * the different architectures. In order to find the
1595 * largest possible bit mask a swap entry with swap type 0
1596 * and swap offset ~0UL is created, encoded to a swap pte,
1597 * decoded to a swp_entry_t again and finally the swap
1598 * offset is extracted. This will mask all the bits from
1599 * the initial ~0UL mask that can't be encoded in either
1600 * the swp_entry_t or the architecture definition of a
1601 * swap pte.
1602 */
1603 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1604 if (maxpages > swap_header->info.last_page)
1605 maxpages = swap_header->info.last_page;
1606 p->highest_bit = maxpages - 1;
1607
1608 error = -EINVAL;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001609 if (!maxpages)
1610 goto bad_swap;
Eric Sandeen5d1854e2006-12-06 20:33:06 -08001611 if (swapfilesize && maxpages > swapfilesize) {
1612 printk(KERN_WARNING
1613 "Swap area shorter than signature indicates\n");
1614 goto bad_swap;
1615 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001616 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1617 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1619 goto bad_swap;
Tobias Klausercd105df2006-01-08 01:00:59 -08001620
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 /* OK, set up the swap map and apply the bad block list */
1622 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1623 error = -ENOMEM;
1624 goto bad_swap;
1625 }
1626
1627 error = 0;
1628 memset(p->swap_map, 0, maxpages * sizeof(short));
Tobias Klausercd105df2006-01-08 01:00:59 -08001629 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1630 int page_nr = swap_header->info.badpages[i];
1631 if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 error = -EINVAL;
1633 else
Tobias Klausercd105df2006-01-08 01:00:59 -08001634 p->swap_map[page_nr] = SWAP_MAP_BAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 }
1636 nr_good_pages = swap_header->info.last_page -
1637 swap_header->info.nr_badpages -
1638 1 /* header page */;
Tobias Klausercd105df2006-01-08 01:00:59 -08001639 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 goto bad_swap;
1641 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001642
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001643 if (nr_good_pages) {
1644 p->swap_map[0] = SWAP_MAP_BAD;
1645 p->max = maxpages;
1646 p->pages = nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07001647 nr_extents = setup_swap_extents(p, &span);
1648 if (nr_extents < 0) {
1649 error = nr_extents;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001650 goto bad_swap;
Hugh Dickins53092a72005-09-03 15:54:34 -07001651 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001652 nr_good_pages = p->pages;
1653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 if (!nr_good_pages) {
1655 printk(KERN_WARNING "Empty swap-file\n");
1656 error = -EINVAL;
1657 goto bad_swap;
1658 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001660 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001661 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 p->flags = SWP_ACTIVE;
1663 nr_swap_pages += nr_good_pages;
1664 total_swap_pages += nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07001665
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001666 printk(KERN_INFO "Adding %uk swap on %s. "
Hugh Dickins53092a72005-09-03 15:54:34 -07001667 "Priority:%d extents:%d across:%lluk\n",
1668 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1669 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670
1671 /* insert swap space into swap_list: */
1672 prev = -1;
1673 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1674 if (p->prio >= swap_info[i].prio) {
1675 break;
1676 }
1677 prev = i;
1678 }
1679 p->next = i;
1680 if (prev < 0) {
1681 swap_list.head = swap_list.next = p - swap_info;
1682 } else {
1683 swap_info[prev].next = p - swap_info;
1684 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07001685 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001686 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 error = 0;
1688 goto out;
1689bad_swap:
1690 if (bdev) {
1691 set_blocksize(bdev, p->old_block_size);
1692 bd_release(bdev);
1693 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001694 destroy_swap_extents(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695bad_swap_2:
Hugh Dickins5d337b92005-09-03 15:54:41 -07001696 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 swap_map = p->swap_map;
1698 p->swap_file = NULL;
1699 p->swap_map = NULL;
1700 p->flags = 0;
1701 if (!(swap_flags & SWAP_FLAG_PREFER))
1702 ++least_priority;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001703 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 vfree(swap_map);
1705 if (swap_file)
1706 filp_close(swap_file, NULL);
1707out:
1708 if (page && !IS_ERR(page)) {
1709 kunmap(page);
1710 page_cache_release(page);
1711 }
1712 if (name)
1713 putname(name);
1714 if (did_down) {
1715 if (!error)
1716 inode->i_flags |= S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001717 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 }
1719 return error;
1720}
1721
1722void si_swapinfo(struct sysinfo *val)
1723{
1724 unsigned int i;
1725 unsigned long nr_to_be_unused = 0;
1726
Hugh Dickins5d337b92005-09-03 15:54:41 -07001727 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 for (i = 0; i < nr_swapfiles; i++) {
1729 if (!(swap_info[i].flags & SWP_USED) ||
1730 (swap_info[i].flags & SWP_WRITEOK))
1731 continue;
1732 nr_to_be_unused += swap_info[i].inuse_pages;
1733 }
1734 val->freeswap = nr_swap_pages + nr_to_be_unused;
1735 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001736 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737}
1738
1739/*
1740 * Verify that a swap entry is valid and increment its swap map count.
1741 *
1742 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1743 * "permanent", but will be reclaimed by the next swapoff.
1744 */
1745int swap_duplicate(swp_entry_t entry)
1746{
1747 struct swap_info_struct * p;
1748 unsigned long offset, type;
1749 int result = 0;
1750
Christoph Lameter06972122006-06-23 02:03:35 -07001751 if (is_migration_entry(entry))
1752 return 1;
1753
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 type = swp_type(entry);
1755 if (type >= nr_swapfiles)
1756 goto bad_file;
1757 p = type + swap_info;
1758 offset = swp_offset(entry);
1759
Hugh Dickins5d337b92005-09-03 15:54:41 -07001760 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 if (offset < p->max && p->swap_map[offset]) {
1762 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1763 p->swap_map[offset]++;
1764 result = 1;
1765 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1766 if (swap_overflow++ < 5)
1767 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1768 p->swap_map[offset] = SWAP_MAP_MAX;
1769 result = 1;
1770 }
1771 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07001772 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773out:
1774 return result;
1775
1776bad_file:
1777 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1778 goto out;
1779}
1780
1781struct swap_info_struct *
1782get_swap_info_struct(unsigned type)
1783{
1784 return &swap_info[type];
1785}
1786
1787/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001788 * swap_lock prevents swap_map being freed. Don't grab an extra
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 * reference on the swaphandle, it doesn't matter if it becomes unused.
1790 */
1791int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1792{
Hugh Dickins89528982008-02-04 22:28:45 -08001793 struct swap_info_struct *si;
Hugh Dickins3f9e7942006-09-29 02:01:26 -07001794 int our_page_cluster = page_cluster;
Hugh Dickins89528982008-02-04 22:28:45 -08001795 pgoff_t target, toff;
1796 pgoff_t base, end;
1797 int nr_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
Hugh Dickins3f9e7942006-09-29 02:01:26 -07001799 if (!our_page_cluster) /* no readahead */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 return 0;
Hugh Dickins89528982008-02-04 22:28:45 -08001801
1802 si = &swap_info[swp_type(entry)];
1803 target = swp_offset(entry);
1804 base = (target >> our_page_cluster) << our_page_cluster;
1805 end = base + (1 << our_page_cluster);
1806 if (!base) /* first page is swap header */
1807 base++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
Hugh Dickins5d337b92005-09-03 15:54:41 -07001809 spin_lock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08001810 if (end > si->max) /* don't go beyond end of map */
1811 end = si->max;
1812
1813 /* Count contiguous allocated slots above our target */
1814 for (toff = target; ++toff < end; nr_pages++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 /* Don't read in free or bad pages */
Hugh Dickins89528982008-02-04 22:28:45 -08001816 if (!si->swap_map[toff])
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 break;
Hugh Dickins89528982008-02-04 22:28:45 -08001818 if (si->swap_map[toff] == SWAP_MAP_BAD)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 break;
Hugh Dickins89528982008-02-04 22:28:45 -08001820 }
1821 /* Count contiguous allocated slots below our target */
1822 for (toff = target; --toff >= base; nr_pages++) {
1823 /* Don't read in free or bad pages */
1824 if (!si->swap_map[toff])
1825 break;
1826 if (si->swap_map[toff] == SWAP_MAP_BAD)
1827 break;
1828 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07001829 spin_unlock(&swap_lock);
Hugh Dickins89528982008-02-04 22:28:45 -08001830
1831 /*
1832 * Indicate starting offset, and return number of pages to get:
1833 * if only 1, say 0, since there's then no readahead to be done.
1834 */
1835 *offset = ++toff;
1836 return nr_pages? ++nr_pages: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837}