blob: e5fd5385f0cc119005a87decfb10385b67a2474f [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
8#include <linux/config.h>
9#include <linux/mm.h>
10#include <linux/hugetlb.h>
11#include <linux/mman.h>
12#include <linux/slab.h>
13#include <linux/kernel_stat.h>
14#include <linux/swap.h>
15#include <linux/vmalloc.h>
16#include <linux/pagemap.h>
17#include <linux/namei.h>
18#include <linux/shm.h>
19#include <linux/blkdev.h>
20#include <linux/writeback.h>
21#include <linux/proc_fs.h>
22#include <linux/seq_file.h>
23#include <linux/init.h>
24#include <linux/module.h>
25#include <linux/rmap.h>
26#include <linux/security.h>
27#include <linux/backing-dev.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -080028#include <linux/mutex.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080029#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/syscalls.h>
31
32#include <asm/pgtable.h>
33#include <asm/tlbflush.h>
34#include <linux/swapops.h>
35
Hugh Dickins5d337b92005-09-03 15:54:41 -070036DEFINE_SPINLOCK(swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070037unsigned int nr_swapfiles;
38long total_swap_pages;
39static int swap_overflow;
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041static const char Bad_file[] = "Bad swap file entry ";
42static const char Unused_file[] = "Unused swap file entry ";
43static const char Bad_offset[] = "Bad swap offset entry ";
44static const char Unused_offset[] = "Unused swap offset entry ";
45
46struct swap_list_t swap_list = {-1, -1};
47
Rafael J. Wysockif577eb32006-03-23 02:59:59 -080048static struct swap_info_struct swap_info[MAX_SWAPFILES];
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Ingo Molnarfc0abb12006-01-18 17:42:33 -080050static DEFINE_MUTEX(swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52/*
53 * We need this because the bdev->unplug_fn can sleep and we cannot
Hugh Dickins5d337b92005-09-03 15:54:41 -070054 * hold swap_lock while calling the unplug_fn. And swap_lock
Ingo Molnarfc0abb12006-01-18 17:42:33 -080055 * cannot be turned into a mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 */
57static DECLARE_RWSEM(swap_unplug_sem);
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
60{
61 swp_entry_t entry;
62
63 down_read(&swap_unplug_sem);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070064 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 if (PageSwapCache(page)) {
66 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
67 struct backing_dev_info *bdi;
68
69 /*
70 * If the page is removed from swapcache from under us (with a
71 * racy try_to_unuse/swapoff) we need an additional reference
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070072 * count to avoid reading garbage from page_private(page) above.
73 * If the WARN_ON triggers during a swapoff it maybe the race
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 * condition and it's harmless. However if it triggers without
75 * swapoff it signals a problem.
76 */
77 WARN_ON(page_count(page) <= 1);
78
79 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
McMullan, Jasonba323112005-05-16 21:53:40 -070080 blk_run_backing_dev(bdi, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 }
82 up_read(&swap_unplug_sem);
83}
84
Hugh Dickins048c27f2005-09-03 15:54:40 -070085#define SWAPFILE_CLUSTER 256
86#define LATENCY_LIMIT 256
87
Hugh Dickins6eb396d2005-09-03 15:54:35 -070088static inline unsigned long scan_swap_map(struct swap_info_struct *si)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089{
Hugh Dickins7dfad412005-09-03 15:54:38 -070090 unsigned long offset, last_in_cluster;
Hugh Dickins048c27f2005-09-03 15:54:40 -070091 int latency_ration = LATENCY_LIMIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Hugh Dickins7dfad412005-09-03 15:54:38 -070093 /*
94 * We try to cluster swap pages by allocating them sequentially
95 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
96 * way, however, we resort to first-free allocation, starting
97 * a new cluster. This prevents us from scattering swap pages
98 * all over the entire swap partition, so that we reduce
99 * overall disk seek times between swap pages. -- sct
100 * But we do now try to find an empty cluster. -Andrea
101 */
102
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700103 si->flags += SWP_SCANNING;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700104 if (unlikely(!si->cluster_nr)) {
105 si->cluster_nr = SWAPFILE_CLUSTER - 1;
106 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER)
107 goto lowest;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700108 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700109
110 offset = si->lowest_bit;
111 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
112
113 /* Locate the first empty (unaligned) cluster */
114 for (; last_in_cluster <= si->highest_bit; offset++) {
115 if (si->swap_map[offset])
116 last_in_cluster = offset + SWAPFILE_CLUSTER;
117 else if (offset == last_in_cluster) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700118 spin_lock(&swap_lock);
Akinobu Mita9b65ef52006-03-22 00:09:09 -0800119 si->cluster_next = offset-SWAPFILE_CLUSTER+1;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700120 goto cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700122 if (unlikely(--latency_ration < 0)) {
123 cond_resched();
124 latency_ration = LATENCY_LIMIT;
125 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700126 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700127 spin_lock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700128 goto lowest;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700130
131 si->cluster_nr--;
132cluster:
133 offset = si->cluster_next;
134 if (offset > si->highest_bit)
135lowest: offset = si->lowest_bit;
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700136checks: if (!(si->flags & SWP_WRITEOK))
137 goto no_page;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700138 if (!si->highest_bit)
139 goto no_page;
140 if (!si->swap_map[offset]) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700141 if (offset == si->lowest_bit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 si->lowest_bit++;
143 if (offset == si->highest_bit)
144 si->highest_bit--;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700145 si->inuse_pages++;
146 if (si->inuse_pages == si->pages) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 si->lowest_bit = si->max;
148 si->highest_bit = 0;
149 }
150 si->swap_map[offset] = 1;
Hugh Dickins7dfad412005-09-03 15:54:38 -0700151 si->cluster_next = offset + 1;
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700152 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 return offset;
154 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700155
Hugh Dickins5d337b92005-09-03 15:54:41 -0700156 spin_unlock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700157 while (++offset <= si->highest_bit) {
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700158 if (!si->swap_map[offset]) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700159 spin_lock(&swap_lock);
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700160 goto checks;
161 }
Hugh Dickins048c27f2005-09-03 15:54:40 -0700162 if (unlikely(--latency_ration < 0)) {
163 cond_resched();
164 latency_ration = LATENCY_LIMIT;
165 }
Hugh Dickins7dfad412005-09-03 15:54:38 -0700166 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700167 spin_lock(&swap_lock);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700168 goto lowest;
169
170no_page:
Hugh Dickins52b7efdb2005-09-03 15:54:39 -0700171 si->flags -= SWP_SCANNING;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 return 0;
173}
174
175swp_entry_t get_swap_page(void)
176{
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700177 struct swap_info_struct *si;
178 pgoff_t offset;
179 int type, next;
180 int wrapped = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
Hugh Dickins5d337b92005-09-03 15:54:41 -0700182 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 if (nr_swap_pages <= 0)
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700184 goto noswap;
185 nr_swap_pages--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700187 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
188 si = swap_info + type;
189 next = si->next;
190 if (next < 0 ||
191 (!wrapped && si->prio != swap_info[next].prio)) {
192 next = swap_list.head;
193 wrapped++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700195
196 if (!si->highest_bit)
197 continue;
198 if (!(si->flags & SWP_WRITEOK))
199 continue;
200
201 swap_list.next = next;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700202 offset = scan_swap_map(si);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700203 if (offset) {
204 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700205 return swp_entry(type, offset);
Hugh Dickins5d337b92005-09-03 15:54:41 -0700206 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700207 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700209
210 nr_swap_pages++;
211noswap:
Hugh Dickins5d337b92005-09-03 15:54:41 -0700212 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700213 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
Rafael J. Wysocki3a291a22006-01-06 00:16:37 -0800216swp_entry_t get_swap_page_of_type(int type)
217{
218 struct swap_info_struct *si;
219 pgoff_t offset;
220
221 spin_lock(&swap_lock);
222 si = swap_info + type;
223 if (si->flags & SWP_WRITEOK) {
224 nr_swap_pages--;
225 offset = scan_swap_map(si);
226 if (offset) {
227 spin_unlock(&swap_lock);
228 return swp_entry(type, offset);
229 }
230 nr_swap_pages++;
231 }
232 spin_unlock(&swap_lock);
233 return (swp_entry_t) {0};
234}
235
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236static struct swap_info_struct * swap_info_get(swp_entry_t entry)
237{
238 struct swap_info_struct * p;
239 unsigned long offset, type;
240
241 if (!entry.val)
242 goto out;
243 type = swp_type(entry);
244 if (type >= nr_swapfiles)
245 goto bad_nofile;
246 p = & swap_info[type];
247 if (!(p->flags & SWP_USED))
248 goto bad_device;
249 offset = swp_offset(entry);
250 if (offset >= p->max)
251 goto bad_offset;
252 if (!p->swap_map[offset])
253 goto bad_free;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700254 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 return p;
256
257bad_free:
258 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
259 goto out;
260bad_offset:
261 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
262 goto out;
263bad_device:
264 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
265 goto out;
266bad_nofile:
267 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
268out:
269 return NULL;
270}
271
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
273{
274 int count = p->swap_map[offset];
275
276 if (count < SWAP_MAP_MAX) {
277 count--;
278 p->swap_map[offset] = count;
279 if (!count) {
280 if (offset < p->lowest_bit)
281 p->lowest_bit = offset;
282 if (offset > p->highest_bit)
283 p->highest_bit = offset;
Hugh Dickins89d09a22005-09-03 15:54:36 -0700284 if (p->prio > swap_info[swap_list.next].prio)
285 swap_list.next = p - swap_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 nr_swap_pages++;
287 p->inuse_pages--;
288 }
289 }
290 return count;
291}
292
293/*
294 * Caller has made sure that the swapdevice corresponding to entry
295 * is still around or has not been recycled.
296 */
297void swap_free(swp_entry_t entry)
298{
299 struct swap_info_struct * p;
300
301 p = swap_info_get(entry);
302 if (p) {
303 swap_entry_free(p, swp_offset(entry));
Hugh Dickins5d337b92005-09-03 15:54:41 -0700304 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 }
306}
307
308/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700309 * How many references to page are currently swapped out?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700311static inline int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700313 int count = 0;
314 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 swp_entry_t entry;
316
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700317 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 p = swap_info_get(entry);
319 if (p) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700320 /* Subtract the 1 for the swap cache itself */
321 count = p->swap_map[swp_offset(entry)] - 1;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700322 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700324 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325}
326
327/*
328 * We can use this swap cache entry directly
329 * if there are no other references to it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 */
331int can_share_swap_page(struct page *page)
332{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700333 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700335 BUG_ON(!PageLocked(page));
336 count = page_mapcount(page);
337 if (count <= 1 && PageSwapCache(page))
338 count += page_swapcount(page);
339 return count == 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
341
342/*
343 * Work out if there are any other processes sharing this
344 * swap cache page. Free it if you can. Return success.
345 */
346int remove_exclusive_swap_page(struct page *page)
347{
348 int retval;
349 struct swap_info_struct * p;
350 swp_entry_t entry;
351
352 BUG_ON(PagePrivate(page));
353 BUG_ON(!PageLocked(page));
354
355 if (!PageSwapCache(page))
356 return 0;
357 if (PageWriteback(page))
358 return 0;
359 if (page_count(page) != 2) /* 2: us + cache */
360 return 0;
361
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700362 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 p = swap_info_get(entry);
364 if (!p)
365 return 0;
366
367 /* Is the only swap cache user the cache itself? */
368 retval = 0;
369 if (p->swap_map[swp_offset(entry)] == 1) {
370 /* Recheck the page count with the swapcache lock held.. */
371 write_lock_irq(&swapper_space.tree_lock);
372 if ((page_count(page) == 2) && !PageWriteback(page)) {
373 __delete_from_swap_cache(page);
374 SetPageDirty(page);
375 retval = 1;
376 }
377 write_unlock_irq(&swapper_space.tree_lock);
378 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700379 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
381 if (retval) {
382 swap_free(entry);
383 page_cache_release(page);
384 }
385
386 return retval;
387}
388
389/*
390 * Free the swap entry like above, but also try to
391 * free the page cache entry if it is the last user.
392 */
393void free_swap_and_cache(swp_entry_t entry)
394{
395 struct swap_info_struct * p;
396 struct page *page = NULL;
397
398 p = swap_info_get(entry);
399 if (p) {
Nick Piggin93fac702006-03-31 02:29:56 -0800400 if (swap_entry_free(p, swp_offset(entry)) == 1) {
401 page = find_get_page(&swapper_space, entry.val);
402 if (page && unlikely(TestSetPageLocked(page))) {
403 page_cache_release(page);
404 page = NULL;
405 }
406 }
Hugh Dickins5d337b92005-09-03 15:54:41 -0700407 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 }
409 if (page) {
410 int one_user;
411
412 BUG_ON(PagePrivate(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 one_user = (page_count(page) == 2);
414 /* Only cache user (+us), or swap space full? Free it! */
Nick Piggin93fac702006-03-31 02:29:56 -0800415 /* Also recheck PageSwapCache after page is locked (above) */
416 if (PageSwapCache(page) && !PageWriteback(page) &&
417 (one_user || vm_swap_full())) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 delete_from_swap_cache(page);
419 SetPageDirty(page);
420 }
421 unlock_page(page);
422 page_cache_release(page);
423 }
424}
425
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800426#ifdef CONFIG_SOFTWARE_SUSPEND
427/*
428 * Find the swap type that corresponds to given device (if any)
429 *
430 * This is needed for software suspend and is done in such a way that inode
431 * aliasing is allowed.
432 */
433int swap_type_of(dev_t device)
434{
435 int i;
436
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800437 spin_lock(&swap_lock);
438 for (i = 0; i < nr_swapfiles; i++) {
439 struct inode *inode;
440
441 if (!(swap_info[i].flags & SWP_WRITEOK))
442 continue;
Rafael J. Wysocki6e1819d2006-03-23 03:00:03 -0800443 if (!device) {
444 spin_unlock(&swap_lock);
445 return i;
446 }
Rafael J. Wysockif577eb32006-03-23 02:59:59 -0800447 inode = swap_info->swap_file->f_dentry->d_inode;
448 if (S_ISBLK(inode->i_mode) &&
449 device == MKDEV(imajor(inode), iminor(inode))) {
450 spin_unlock(&swap_lock);
451 return i;
452 }
453 }
454 spin_unlock(&swap_lock);
455 return -ENODEV;
456}
457
458/*
459 * Return either the total number of swap pages of given type, or the number
460 * of free pages of that type (depending on @free)
461 *
462 * This is needed for software suspend
463 */
464unsigned int count_swap_pages(int type, int free)
465{
466 unsigned int n = 0;
467
468 if (type < nr_swapfiles) {
469 spin_lock(&swap_lock);
470 if (swap_info[type].flags & SWP_WRITEOK) {
471 n = swap_info[type].pages;
472 if (free)
473 n -= swap_info[type].inuse_pages;
474 }
475 spin_unlock(&swap_lock);
476 }
477 return n;
478}
479#endif
480
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481/*
Hugh Dickins72866f62005-10-29 18:15:55 -0700482 * No need to decide whether this PTE shares the swap entry with others,
483 * just let do_wp_page work it out if a write is requested later - to
484 * force COW, vm_page_prot omits write permission from any private vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 */
486static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
487 unsigned long addr, swp_entry_t entry, struct page *page)
488{
Hugh Dickins42946212005-10-29 18:16:05 -0700489 inc_mm_counter(vma->vm_mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 get_page(page);
491 set_pte_at(vma->vm_mm, addr, pte,
492 pte_mkold(mk_pte(page, vma->vm_page_prot)));
493 page_add_anon_rmap(page, vma, addr);
494 swap_free(entry);
495 /*
496 * Move the page to the active list so it is not
497 * immediately swapped out again after swapon.
498 */
499 activate_page(page);
500}
501
502static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
503 unsigned long addr, unsigned long end,
504 swp_entry_t entry, struct page *page)
505{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 pte_t swp_pte = swp_entry_to_pte(entry);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700507 pte_t *pte;
508 spinlock_t *ptl;
509 int found = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Hugh Dickins705e87c2005-10-29 18:16:27 -0700511 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 do {
513 /*
514 * swapoff spends a _lot_ of time in this loop!
515 * Test inline before going to call unuse_pte.
516 */
517 if (unlikely(pte_same(*pte, swp_pte))) {
Hugh Dickins705e87c2005-10-29 18:16:27 -0700518 unuse_pte(vma, pte++, addr, entry, page);
519 found = 1;
520 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 }
522 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700523 pte_unmap_unlock(pte - 1, ptl);
524 return found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525}
526
527static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
528 unsigned long addr, unsigned long end,
529 swp_entry_t entry, struct page *page)
530{
531 pmd_t *pmd;
532 unsigned long next;
533
534 pmd = pmd_offset(pud, addr);
535 do {
536 next = pmd_addr_end(addr, end);
537 if (pmd_none_or_clear_bad(pmd))
538 continue;
539 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
540 return 1;
541 } while (pmd++, addr = next, addr != end);
542 return 0;
543}
544
545static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
546 unsigned long addr, unsigned long end,
547 swp_entry_t entry, struct page *page)
548{
549 pud_t *pud;
550 unsigned long next;
551
552 pud = pud_offset(pgd, addr);
553 do {
554 next = pud_addr_end(addr, end);
555 if (pud_none_or_clear_bad(pud))
556 continue;
557 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
558 return 1;
559 } while (pud++, addr = next, addr != end);
560 return 0;
561}
562
563static int unuse_vma(struct vm_area_struct *vma,
564 swp_entry_t entry, struct page *page)
565{
566 pgd_t *pgd;
567 unsigned long addr, end, next;
568
569 if (page->mapping) {
570 addr = page_address_in_vma(page, vma);
571 if (addr == -EFAULT)
572 return 0;
573 else
574 end = addr + PAGE_SIZE;
575 } else {
576 addr = vma->vm_start;
577 end = vma->vm_end;
578 }
579
580 pgd = pgd_offset(vma->vm_mm, addr);
581 do {
582 next = pgd_addr_end(addr, end);
583 if (pgd_none_or_clear_bad(pgd))
584 continue;
585 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
586 return 1;
587 } while (pgd++, addr = next, addr != end);
588 return 0;
589}
590
591static int unuse_mm(struct mm_struct *mm,
592 swp_entry_t entry, struct page *page)
593{
594 struct vm_area_struct *vma;
595
596 if (!down_read_trylock(&mm->mmap_sem)) {
597 /*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700598 * Activate page so shrink_cache is unlikely to unmap its
599 * ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700601 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 unlock_page(page);
603 down_read(&mm->mmap_sem);
604 lock_page(page);
605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 for (vma = mm->mmap; vma; vma = vma->vm_next) {
607 if (vma->anon_vma && unuse_vma(vma, entry, page))
608 break;
609 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 up_read(&mm->mmap_sem);
611 /*
612 * Currently unuse_mm cannot fail, but leave error handling
613 * at call sites for now, since we change it from time to time.
614 */
615 return 0;
616}
617
Christoph Lametera3351e52006-02-01 03:05:39 -0800618#ifdef CONFIG_MIGRATION
619int remove_vma_swap(struct vm_area_struct *vma, struct page *page)
620{
621 swp_entry_t entry = { .val = page_private(page) };
622
623 return unuse_vma(vma, entry, page);
624}
625#endif
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627/*
628 * Scan swap_map from current position to next entry still in use.
629 * Recycle to start on reaching the end, returning 0 when empty.
630 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700631static unsigned int find_next_to_unuse(struct swap_info_struct *si,
632 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633{
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700634 unsigned int max = si->max;
635 unsigned int i = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 int count;
637
638 /*
Hugh Dickins5d337b92005-09-03 15:54:41 -0700639 * No need for swap_lock here: we're just looking
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 * for whether an entry is in use, not modifying it; false
641 * hits are okay, and sys_swapoff() has already prevented new
Hugh Dickins5d337b92005-09-03 15:54:41 -0700642 * allocations from this area (while holding swap_lock).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 */
644 for (;;) {
645 if (++i >= max) {
646 if (!prev) {
647 i = 0;
648 break;
649 }
650 /*
651 * No entries in use at top of swap_map,
652 * loop back to start and recheck there.
653 */
654 max = prev + 1;
655 prev = 0;
656 i = 1;
657 }
658 count = si->swap_map[i];
659 if (count && count != SWAP_MAP_BAD)
660 break;
661 }
662 return i;
663}
664
665/*
666 * We completely avoid races by reading each swap page in advance,
667 * and then search for the process using it. All the necessary
668 * page table adjustments can then be made atomically.
669 */
670static int try_to_unuse(unsigned int type)
671{
672 struct swap_info_struct * si = &swap_info[type];
673 struct mm_struct *start_mm;
674 unsigned short *swap_map;
675 unsigned short swcount;
676 struct page *page;
677 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700678 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 int retval = 0;
680 int reset_overflow = 0;
681 int shmem;
682
683 /*
684 * When searching mms for an entry, a good strategy is to
685 * start at the first mm we freed the previous entry from
686 * (though actually we don't notice whether we or coincidence
687 * freed the entry). Initialize this start_mm with a hold.
688 *
689 * A simpler strategy would be to start at the last mm we
690 * freed the previous entry from; but that would take less
691 * advantage of mmlist ordering, which clusters forked mms
692 * together, child after parent. If we race with dup_mmap(), we
693 * prefer to resolve parent before child, lest we miss entries
694 * duplicated after we scanned child: using last mm would invert
695 * that. Though it's only a serious concern when an overflowed
696 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
697 */
698 start_mm = &init_mm;
699 atomic_inc(&init_mm.mm_users);
700
701 /*
702 * Keep on scanning until all entries have gone. Usually,
703 * one pass through swap_map is enough, but not necessarily:
704 * there are races when an instance of an entry might be missed.
705 */
706 while ((i = find_next_to_unuse(si, i)) != 0) {
707 if (signal_pending(current)) {
708 retval = -EINTR;
709 break;
710 }
711
712 /*
713 * Get a page for the entry, using the existing swap
714 * cache page if there is one. Otherwise, get a clean
715 * page and read the swap into it.
716 */
717 swap_map = &si->swap_map[i];
718 entry = swp_entry(type, i);
Christoph Lameterb16664e2006-02-01 03:05:36 -0800719again:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 page = read_swap_cache_async(entry, NULL, 0);
721 if (!page) {
722 /*
723 * Either swap_duplicate() failed because entry
724 * has been freed independently, and will not be
725 * reused since sys_swapoff() already disabled
726 * allocation from here, or alloc_page() failed.
727 */
728 if (!*swap_map)
729 continue;
730 retval = -ENOMEM;
731 break;
732 }
733
734 /*
735 * Don't hold on to start_mm if it looks like exiting.
736 */
737 if (atomic_read(&start_mm->mm_users) == 1) {
738 mmput(start_mm);
739 start_mm = &init_mm;
740 atomic_inc(&init_mm.mm_users);
741 }
742
743 /*
744 * Wait for and lock page. When do_swap_page races with
745 * try_to_unuse, do_swap_page can handle the fault much
746 * faster than try_to_unuse can locate the entry. This
747 * apparently redundant "wait_on_page_locked" lets try_to_unuse
748 * defer to do_swap_page in such a case - in some tests,
749 * do_swap_page and try_to_unuse repeatedly compete.
750 */
751 wait_on_page_locked(page);
752 wait_on_page_writeback(page);
753 lock_page(page);
Christoph Lameterb16664e2006-02-01 03:05:36 -0800754 if (!PageSwapCache(page)) {
755 /* Page migration has occured */
756 unlock_page(page);
757 page_cache_release(page);
758 goto again;
759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 wait_on_page_writeback(page);
761
762 /*
763 * Remove all references to entry.
764 * Whenever we reach init_mm, there's no address space
765 * to search, but use it as a reminder to search shmem.
766 */
767 shmem = 0;
768 swcount = *swap_map;
769 if (swcount > 1) {
770 if (start_mm == &init_mm)
771 shmem = shmem_unuse(entry, page);
772 else
773 retval = unuse_mm(start_mm, entry, page);
774 }
775 if (*swap_map > 1) {
776 int set_start_mm = (*swap_map >= swcount);
777 struct list_head *p = &start_mm->mmlist;
778 struct mm_struct *new_start_mm = start_mm;
779 struct mm_struct *prev_mm = start_mm;
780 struct mm_struct *mm;
781
782 atomic_inc(&new_start_mm->mm_users);
783 atomic_inc(&prev_mm->mm_users);
784 spin_lock(&mmlist_lock);
785 while (*swap_map > 1 && !retval &&
786 (p = p->next) != &start_mm->mmlist) {
787 mm = list_entry(p, struct mm_struct, mmlist);
788 if (atomic_inc_return(&mm->mm_users) == 1) {
789 atomic_dec(&mm->mm_users);
790 continue;
791 }
792 spin_unlock(&mmlist_lock);
793 mmput(prev_mm);
794 prev_mm = mm;
795
796 cond_resched();
797
798 swcount = *swap_map;
799 if (swcount <= 1)
800 ;
801 else if (mm == &init_mm) {
802 set_start_mm = 1;
803 shmem = shmem_unuse(entry, page);
804 } else
805 retval = unuse_mm(mm, entry, page);
806 if (set_start_mm && *swap_map < swcount) {
807 mmput(new_start_mm);
808 atomic_inc(&mm->mm_users);
809 new_start_mm = mm;
810 set_start_mm = 0;
811 }
812 spin_lock(&mmlist_lock);
813 }
814 spin_unlock(&mmlist_lock);
815 mmput(prev_mm);
816 mmput(start_mm);
817 start_mm = new_start_mm;
818 }
819 if (retval) {
820 unlock_page(page);
821 page_cache_release(page);
822 break;
823 }
824
825 /*
826 * How could swap count reach 0x7fff when the maximum
827 * pid is 0x7fff, and there's no way to repeat a swap
828 * page within an mm (except in shmem, where it's the
829 * shared object which takes the reference count)?
830 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
831 *
832 * If that's wrong, then we should worry more about
833 * exit_mmap() and do_munmap() cases described above:
834 * we might be resetting SWAP_MAP_MAX too early here.
835 * We know "Undead"s can happen, they're okay, so don't
836 * report them; but do report if we reset SWAP_MAP_MAX.
837 */
838 if (*swap_map == SWAP_MAP_MAX) {
Hugh Dickins5d337b92005-09-03 15:54:41 -0700839 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 *swap_map = 1;
Hugh Dickins5d337b92005-09-03 15:54:41 -0700841 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 reset_overflow = 1;
843 }
844
845 /*
846 * If a reference remains (rare), we would like to leave
847 * the page in the swap cache; but try_to_unmap could
848 * then re-duplicate the entry once we drop page lock,
849 * so we might loop indefinitely; also, that page could
850 * not be swapped out to other storage meanwhile. So:
851 * delete from cache even if there's another reference,
852 * after ensuring that the data has been saved to disk -
853 * since if the reference remains (rarer), it will be
854 * read from disk into another page. Splitting into two
855 * pages would be incorrect if swap supported "shared
856 * private" pages, but they are handled by tmpfs files.
857 *
858 * Note shmem_unuse already deleted a swappage from
859 * the swap cache, unless the move to filepage failed:
860 * in which case it left swappage in cache, lowered its
861 * swap count to pass quickly through the loops above,
862 * and now we must reincrement count to try again later.
863 */
864 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
865 struct writeback_control wbc = {
866 .sync_mode = WB_SYNC_NONE,
867 };
868
869 swap_writepage(page, &wbc);
870 lock_page(page);
871 wait_on_page_writeback(page);
872 }
873 if (PageSwapCache(page)) {
874 if (shmem)
875 swap_duplicate(entry);
876 else
877 delete_from_swap_cache(page);
878 }
879
880 /*
881 * So we could skip searching mms once swap count went
882 * to 1, we did not mark any present ptes as dirty: must
883 * mark page dirty so shrink_list will preserve it.
884 */
885 SetPageDirty(page);
886 unlock_page(page);
887 page_cache_release(page);
888
889 /*
890 * Make sure that we aren't completely killing
891 * interactive performance.
892 */
893 cond_resched();
894 }
895
896 mmput(start_mm);
897 if (reset_overflow) {
898 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
899 swap_overflow = 0;
900 }
901 return retval;
902}
903
904/*
Hugh Dickins5d337b92005-09-03 15:54:41 -0700905 * After a successful try_to_unuse, if no swap is now in use, we know
906 * we can empty the mmlist. swap_lock must be held on entry and exit.
907 * Note that mmlist_lock nests inside swap_lock, and an mm must be
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 * added to the mmlist just after page_duplicate - before would be racy.
909 */
910static void drain_mmlist(void)
911{
912 struct list_head *p, *next;
913 unsigned int i;
914
915 for (i = 0; i < nr_swapfiles; i++)
916 if (swap_info[i].inuse_pages)
917 return;
918 spin_lock(&mmlist_lock);
919 list_for_each_safe(p, next, &init_mm.mmlist)
920 list_del_init(p);
921 spin_unlock(&mmlist_lock);
922}
923
924/*
925 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
926 * corresponds to page offset `offset'.
927 */
928sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
929{
930 struct swap_extent *se = sis->curr_swap_extent;
931 struct swap_extent *start_se = se;
932
933 for ( ; ; ) {
934 struct list_head *lh;
935
936 if (se->start_page <= offset &&
937 offset < (se->start_page + se->nr_pages)) {
938 return se->start_block + (offset - se->start_page);
939 }
Hugh Dickins11d31882005-09-03 15:54:34 -0700940 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 if (lh == &sis->extent_list)
Hugh Dickins11d31882005-09-03 15:54:34 -0700942 lh = lh->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 se = list_entry(lh, struct swap_extent, list);
944 sis->curr_swap_extent = se;
945 BUG_ON(se == start_se); /* It *must* be present */
946 }
947}
948
949/*
950 * Free all of a swapdev's extent information
951 */
952static void destroy_swap_extents(struct swap_info_struct *sis)
953{
954 while (!list_empty(&sis->extent_list)) {
955 struct swap_extent *se;
956
957 se = list_entry(sis->extent_list.next,
958 struct swap_extent, list);
959 list_del(&se->list);
960 kfree(se);
961 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962}
963
964/*
965 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -0700966 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 *
Hugh Dickins11d31882005-09-03 15:54:34 -0700968 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 */
970static int
971add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
972 unsigned long nr_pages, sector_t start_block)
973{
974 struct swap_extent *se;
975 struct swap_extent *new_se;
976 struct list_head *lh;
977
Hugh Dickins11d31882005-09-03 15:54:34 -0700978 lh = sis->extent_list.prev; /* The highest page extent */
979 if (lh != &sis->extent_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -0700981 BUG_ON(se->start_page + se->nr_pages != start_page);
982 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 /* Merge it */
984 se->nr_pages += nr_pages;
985 return 0;
986 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 }
988
989 /*
990 * No merge. Insert a new extent, preserving ordering.
991 */
992 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
993 if (new_se == NULL)
994 return -ENOMEM;
995 new_se->start_page = start_page;
996 new_se->nr_pages = nr_pages;
997 new_se->start_block = start_block;
998
Hugh Dickins11d31882005-09-03 15:54:34 -0700999 list_add_tail(&new_se->list, &sis->extent_list);
Hugh Dickins53092a72005-09-03 15:54:34 -07001000 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001}
1002
1003/*
1004 * A `swap extent' is a simple thing which maps a contiguous range of pages
1005 * onto a contiguous range of disk blocks. An ordered list of swap extents
1006 * is built at swapon time and is then used at swap_writepage/swap_readpage
1007 * time for locating where on disk a page belongs.
1008 *
1009 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1010 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1011 * swap files identically.
1012 *
1013 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1014 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1015 * swapfiles are handled *identically* after swapon time.
1016 *
1017 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1018 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1019 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1020 * requirements, they are simply tossed out - we will never use those blocks
1021 * for swapping.
1022 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -07001023 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1025 * which will scribble on the fs.
1026 *
1027 * The amount of disk space which a single swap extent represents varies.
1028 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1029 * extents in the list. To avoid much list walking, we cache the previous
1030 * search location in `curr_swap_extent', and start new searches from there.
1031 * This is extremely effective. The average number of iterations in
1032 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1033 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001034static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035{
1036 struct inode *inode;
1037 unsigned blocks_per_page;
1038 unsigned long page_no;
1039 unsigned blkbits;
1040 sector_t probe_block;
1041 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -07001042 sector_t lowest_block = -1;
1043 sector_t highest_block = 0;
1044 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 int ret;
1046
1047 inode = sis->swap_file->f_mapping->host;
1048 if (S_ISBLK(inode->i_mode)) {
1049 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -07001050 *span = sis->pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 goto done;
1052 }
1053
1054 blkbits = inode->i_blkbits;
1055 blocks_per_page = PAGE_SIZE >> blkbits;
1056
1057 /*
1058 * Map all the blocks into the extent list. This code doesn't try
1059 * to be very smart.
1060 */
1061 probe_block = 0;
1062 page_no = 0;
1063 last_block = i_size_read(inode) >> blkbits;
1064 while ((probe_block + blocks_per_page) <= last_block &&
1065 page_no < sis->max) {
1066 unsigned block_in_page;
1067 sector_t first_block;
1068
1069 first_block = bmap(inode, probe_block);
1070 if (first_block == 0)
1071 goto bad_bmap;
1072
1073 /*
1074 * It must be PAGE_SIZE aligned on-disk
1075 */
1076 if (first_block & (blocks_per_page - 1)) {
1077 probe_block++;
1078 goto reprobe;
1079 }
1080
1081 for (block_in_page = 1; block_in_page < blocks_per_page;
1082 block_in_page++) {
1083 sector_t block;
1084
1085 block = bmap(inode, probe_block + block_in_page);
1086 if (block == 0)
1087 goto bad_bmap;
1088 if (block != first_block + block_in_page) {
1089 /* Discontiguity */
1090 probe_block++;
1091 goto reprobe;
1092 }
1093 }
1094
Hugh Dickins53092a72005-09-03 15:54:34 -07001095 first_block >>= (PAGE_SHIFT - blkbits);
1096 if (page_no) { /* exclude the header page */
1097 if (first_block < lowest_block)
1098 lowest_block = first_block;
1099 if (first_block > highest_block)
1100 highest_block = first_block;
1101 }
1102
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 /*
1104 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1105 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001106 ret = add_swap_extent(sis, page_no, 1, first_block);
1107 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001109 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 page_no++;
1111 probe_block += blocks_per_page;
1112reprobe:
1113 continue;
1114 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001115 ret = nr_extents;
1116 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001118 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001120 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 sis->highest_bit = page_no - 1;
1122done:
1123 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1124 struct swap_extent, list);
1125 goto out;
1126bad_bmap:
1127 printk(KERN_ERR "swapon: swapfile has holes\n");
1128 ret = -EINVAL;
1129out:
1130 return ret;
1131}
1132
1133#if 0 /* We don't need this yet */
1134#include <linux/backing-dev.h>
1135int page_queue_congested(struct page *page)
1136{
1137 struct backing_dev_info *bdi;
1138
1139 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1140
1141 if (PageSwapCache(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -07001142 swp_entry_t entry = { .val = page_private(page) };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 struct swap_info_struct *sis;
1144
1145 sis = get_swap_info_struct(swp_type(entry));
1146 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1147 } else
1148 bdi = page->mapping->backing_dev_info;
1149 return bdi_write_congested(bdi);
1150}
1151#endif
1152
1153asmlinkage long sys_swapoff(const char __user * specialfile)
1154{
1155 struct swap_info_struct * p = NULL;
1156 unsigned short *swap_map;
1157 struct file *swap_file, *victim;
1158 struct address_space *mapping;
1159 struct inode *inode;
1160 char * pathname;
1161 int i, type, prev;
1162 int err;
1163
1164 if (!capable(CAP_SYS_ADMIN))
1165 return -EPERM;
1166
1167 pathname = getname(specialfile);
1168 err = PTR_ERR(pathname);
1169 if (IS_ERR(pathname))
1170 goto out;
1171
1172 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1173 putname(pathname);
1174 err = PTR_ERR(victim);
1175 if (IS_ERR(victim))
1176 goto out;
1177
1178 mapping = victim->f_mapping;
1179 prev = -1;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001180 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1182 p = swap_info + type;
1183 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1184 if (p->swap_file->f_mapping == mapping)
1185 break;
1186 }
1187 prev = type;
1188 }
1189 if (type < 0) {
1190 err = -EINVAL;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001191 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 goto out_dput;
1193 }
1194 if (!security_vm_enough_memory(p->pages))
1195 vm_unacct_memory(p->pages);
1196 else {
1197 err = -ENOMEM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001198 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 goto out_dput;
1200 }
1201 if (prev < 0) {
1202 swap_list.head = p->next;
1203 } else {
1204 swap_info[prev].next = p->next;
1205 }
1206 if (type == swap_list.next) {
1207 /* just pick something that's safe... */
1208 swap_list.next = swap_list.head;
1209 }
1210 nr_swap_pages -= p->pages;
1211 total_swap_pages -= p->pages;
1212 p->flags &= ~SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001213 spin_unlock(&swap_lock);
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 current->flags |= PF_SWAPOFF;
1216 err = try_to_unuse(type);
1217 current->flags &= ~PF_SWAPOFF;
1218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 if (err) {
1220 /* re-insert swap space back into swap_list */
Hugh Dickins5d337b92005-09-03 15:54:41 -07001221 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1223 if (p->prio >= swap_info[i].prio)
1224 break;
1225 p->next = i;
1226 if (prev < 0)
1227 swap_list.head = swap_list.next = p - swap_info;
1228 else
1229 swap_info[prev].next = p - swap_info;
1230 nr_swap_pages += p->pages;
1231 total_swap_pages += p->pages;
1232 p->flags |= SWP_WRITEOK;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001233 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 goto out_dput;
1235 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001236
1237 /* wait for any unplug function to finish */
1238 down_write(&swap_unplug_sem);
1239 up_write(&swap_unplug_sem);
1240
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001241 destroy_swap_extents(p);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001242 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001243 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 drain_mmlist();
Hugh Dickins5d337b92005-09-03 15:54:41 -07001245
1246 /* wait for anyone still in scan_swap_map */
1247 p->highest_bit = 0; /* cuts scans short */
1248 while (p->flags >= SWP_SCANNING) {
1249 spin_unlock(&swap_lock);
Nishanth Aravamudan13e4b572005-09-10 00:27:25 -07001250 schedule_timeout_uninterruptible(1);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001251 spin_lock(&swap_lock);
1252 }
1253
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 swap_file = p->swap_file;
1255 p->swap_file = NULL;
1256 p->max = 0;
1257 swap_map = p->swap_map;
1258 p->swap_map = NULL;
1259 p->flags = 0;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001260 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001261 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 vfree(swap_map);
1263 inode = mapping->host;
1264 if (S_ISBLK(inode->i_mode)) {
1265 struct block_device *bdev = I_BDEV(inode);
1266 set_blocksize(bdev, p->old_block_size);
1267 bd_release(bdev);
1268 } else {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001269 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 inode->i_flags &= ~S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001271 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 }
1273 filp_close(swap_file, NULL);
1274 err = 0;
1275
1276out_dput:
1277 filp_close(victim, NULL);
1278out:
1279 return err;
1280}
1281
1282#ifdef CONFIG_PROC_FS
1283/* iterator */
1284static void *swap_start(struct seq_file *swap, loff_t *pos)
1285{
1286 struct swap_info_struct *ptr = swap_info;
1287 int i;
1288 loff_t l = *pos;
1289
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001290 mutex_lock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
1292 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1293 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1294 continue;
1295 if (!l--)
1296 return ptr;
1297 }
1298
1299 return NULL;
1300}
1301
1302static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1303{
1304 struct swap_info_struct *ptr = v;
1305 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1306
1307 for (++ptr; ptr < endptr; ptr++) {
1308 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1309 continue;
1310 ++*pos;
1311 return ptr;
1312 }
1313
1314 return NULL;
1315}
1316
1317static void swap_stop(struct seq_file *swap, void *v)
1318{
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001319 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320}
1321
1322static int swap_show(struct seq_file *swap, void *v)
1323{
1324 struct swap_info_struct *ptr = v;
1325 struct file *file;
1326 int len;
1327
1328 if (v == swap_info)
1329 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1330
1331 file = ptr->swap_file;
1332 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001333 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 len < 40 ? 40 - len : 1, " ",
1335 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1336 "partition" : "file\t",
1337 ptr->pages << (PAGE_SHIFT - 10),
1338 ptr->inuse_pages << (PAGE_SHIFT - 10),
1339 ptr->prio);
1340 return 0;
1341}
1342
1343static struct seq_operations swaps_op = {
1344 .start = swap_start,
1345 .next = swap_next,
1346 .stop = swap_stop,
1347 .show = swap_show
1348};
1349
1350static int swaps_open(struct inode *inode, struct file *file)
1351{
1352 return seq_open(file, &swaps_op);
1353}
1354
1355static struct file_operations proc_swaps_operations = {
1356 .open = swaps_open,
1357 .read = seq_read,
1358 .llseek = seq_lseek,
1359 .release = seq_release,
1360};
1361
1362static int __init procswaps_init(void)
1363{
1364 struct proc_dir_entry *entry;
1365
1366 entry = create_proc_entry("swaps", 0, NULL);
1367 if (entry)
1368 entry->proc_fops = &proc_swaps_operations;
1369 return 0;
1370}
1371__initcall(procswaps_init);
1372#endif /* CONFIG_PROC_FS */
1373
1374/*
1375 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1376 *
1377 * The swapon system call
1378 */
1379asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1380{
1381 struct swap_info_struct * p;
1382 char *name = NULL;
1383 struct block_device *bdev = NULL;
1384 struct file *swap_file = NULL;
1385 struct address_space *mapping;
1386 unsigned int type;
1387 int i, prev;
1388 int error;
1389 static int least_priority;
1390 union swap_header *swap_header = NULL;
1391 int swap_header_version;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001392 unsigned int nr_good_pages = 0;
1393 int nr_extents = 0;
Hugh Dickins53092a72005-09-03 15:54:34 -07001394 sector_t span;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 unsigned long maxpages = 1;
1396 int swapfilesize;
1397 unsigned short *swap_map;
1398 struct page *page = NULL;
1399 struct inode *inode = NULL;
1400 int did_down = 0;
1401
1402 if (!capable(CAP_SYS_ADMIN))
1403 return -EPERM;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001404 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 p = swap_info;
1406 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1407 if (!(p->flags & SWP_USED))
1408 break;
1409 error = -EPERM;
1410 /*
1411 * Test if adding another swap device is possible. There are
1412 * two limiting factors: 1) the number of bits for the swap
1413 * type swp_entry_t definition and 2) the number of bits for
1414 * the swap type in the swap ptes as defined by the different
1415 * architectures. To honor both limitations a swap entry
1416 * with swap offset 0 and swap type ~0UL is created, encoded
1417 * to a swap pte, decoded to a swp_entry_t again and finally
1418 * the swap type part is extracted. This will mask all bits
1419 * from the initial ~0UL that can't be encoded in either the
1420 * swp_entry_t or the architecture definition of a swap pte.
1421 */
1422 if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
Hugh Dickins5d337b92005-09-03 15:54:41 -07001423 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 goto out;
1425 }
1426 if (type >= nr_swapfiles)
1427 nr_swapfiles = type+1;
1428 INIT_LIST_HEAD(&p->extent_list);
1429 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 p->swap_file = NULL;
1431 p->old_block_size = 0;
1432 p->swap_map = NULL;
1433 p->lowest_bit = 0;
1434 p->highest_bit = 0;
1435 p->cluster_nr = 0;
1436 p->inuse_pages = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 p->next = -1;
1438 if (swap_flags & SWAP_FLAG_PREFER) {
1439 p->prio =
1440 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1441 } else {
1442 p->prio = --least_priority;
1443 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07001444 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 name = getname(specialfile);
1446 error = PTR_ERR(name);
1447 if (IS_ERR(name)) {
1448 name = NULL;
1449 goto bad_swap_2;
1450 }
1451 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1452 error = PTR_ERR(swap_file);
1453 if (IS_ERR(swap_file)) {
1454 swap_file = NULL;
1455 goto bad_swap_2;
1456 }
1457
1458 p->swap_file = swap_file;
1459 mapping = swap_file->f_mapping;
1460 inode = mapping->host;
1461
1462 error = -EBUSY;
1463 for (i = 0; i < nr_swapfiles; i++) {
1464 struct swap_info_struct *q = &swap_info[i];
1465
1466 if (i == type || !q->swap_file)
1467 continue;
1468 if (mapping == q->swap_file->f_mapping)
1469 goto bad_swap;
1470 }
1471
1472 error = -EINVAL;
1473 if (S_ISBLK(inode->i_mode)) {
1474 bdev = I_BDEV(inode);
1475 error = bd_claim(bdev, sys_swapon);
1476 if (error < 0) {
1477 bdev = NULL;
Rob Landleyf7b3a432005-09-22 21:44:27 -07001478 error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 goto bad_swap;
1480 }
1481 p->old_block_size = block_size(bdev);
1482 error = set_blocksize(bdev, PAGE_SIZE);
1483 if (error < 0)
1484 goto bad_swap;
1485 p->bdev = bdev;
1486 } else if (S_ISREG(inode->i_mode)) {
1487 p->bdev = inode->i_sb->s_bdev;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001488 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 did_down = 1;
1490 if (IS_SWAPFILE(inode)) {
1491 error = -EBUSY;
1492 goto bad_swap;
1493 }
1494 } else {
1495 goto bad_swap;
1496 }
1497
1498 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1499
1500 /*
1501 * Read the swap header.
1502 */
1503 if (!mapping->a_ops->readpage) {
1504 error = -EINVAL;
1505 goto bad_swap;
1506 }
1507 page = read_cache_page(mapping, 0,
1508 (filler_t *)mapping->a_ops->readpage, swap_file);
1509 if (IS_ERR(page)) {
1510 error = PTR_ERR(page);
1511 goto bad_swap;
1512 }
1513 wait_on_page_locked(page);
1514 if (!PageUptodate(page))
1515 goto bad_swap;
1516 kmap(page);
1517 swap_header = page_address(page);
1518
1519 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1520 swap_header_version = 1;
1521 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1522 swap_header_version = 2;
1523 else {
Jesper Juhle97a3112006-01-11 01:50:28 +01001524 printk(KERN_ERR "Unable to find swap-space signature\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 error = -EINVAL;
1526 goto bad_swap;
1527 }
1528
1529 switch (swap_header_version) {
1530 case 1:
1531 printk(KERN_ERR "version 0 swap is no longer supported. "
1532 "Use mkswap -v1 %s\n", name);
1533 error = -EINVAL;
1534 goto bad_swap;
1535 case 2:
1536 /* Check the swap header's sub-version and the size of
1537 the swap file and bad block lists */
1538 if (swap_header->info.version != 1) {
1539 printk(KERN_WARNING
1540 "Unable to handle swap header version %d\n",
1541 swap_header->info.version);
1542 error = -EINVAL;
1543 goto bad_swap;
1544 }
1545
1546 p->lowest_bit = 1;
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001547 p->cluster_next = 1;
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 /*
1550 * Find out how many pages are allowed for a single swap
1551 * device. There are two limiting factors: 1) the number of
1552 * bits for the swap offset in the swp_entry_t type and
1553 * 2) the number of bits in the a swap pte as defined by
1554 * the different architectures. In order to find the
1555 * largest possible bit mask a swap entry with swap type 0
1556 * and swap offset ~0UL is created, encoded to a swap pte,
1557 * decoded to a swp_entry_t again and finally the swap
1558 * offset is extracted. This will mask all the bits from
1559 * the initial ~0UL mask that can't be encoded in either
1560 * the swp_entry_t or the architecture definition of a
1561 * swap pte.
1562 */
1563 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1564 if (maxpages > swap_header->info.last_page)
1565 maxpages = swap_header->info.last_page;
1566 p->highest_bit = maxpages - 1;
1567
1568 error = -EINVAL;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001569 if (!maxpages)
1570 goto bad_swap;
1571 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1572 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1574 goto bad_swap;
Tobias Klausercd105df2006-01-08 01:00:59 -08001575
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 /* OK, set up the swap map and apply the bad block list */
1577 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1578 error = -ENOMEM;
1579 goto bad_swap;
1580 }
1581
1582 error = 0;
1583 memset(p->swap_map, 0, maxpages * sizeof(short));
Tobias Klausercd105df2006-01-08 01:00:59 -08001584 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1585 int page_nr = swap_header->info.badpages[i];
1586 if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 error = -EINVAL;
1588 else
Tobias Klausercd105df2006-01-08 01:00:59 -08001589 p->swap_map[page_nr] = SWAP_MAP_BAD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 }
1591 nr_good_pages = swap_header->info.last_page -
1592 swap_header->info.nr_badpages -
1593 1 /* header page */;
Tobias Klausercd105df2006-01-08 01:00:59 -08001594 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 goto bad_swap;
1596 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001597
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 if (swapfilesize && maxpages > swapfilesize) {
1599 printk(KERN_WARNING
1600 "Swap area shorter than signature indicates\n");
1601 error = -EINVAL;
1602 goto bad_swap;
1603 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001604 if (nr_good_pages) {
1605 p->swap_map[0] = SWAP_MAP_BAD;
1606 p->max = maxpages;
1607 p->pages = nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07001608 nr_extents = setup_swap_extents(p, &span);
1609 if (nr_extents < 0) {
1610 error = nr_extents;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001611 goto bad_swap;
Hugh Dickins53092a72005-09-03 15:54:34 -07001612 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001613 nr_good_pages = p->pages;
1614 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 if (!nr_good_pages) {
1616 printk(KERN_WARNING "Empty swap-file\n");
1617 error = -EINVAL;
1618 goto bad_swap;
1619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001621 mutex_lock(&swapon_mutex);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001622 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 p->flags = SWP_ACTIVE;
1624 nr_swap_pages += nr_good_pages;
1625 total_swap_pages += nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07001626
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001627 printk(KERN_INFO "Adding %uk swap on %s. "
Hugh Dickins53092a72005-09-03 15:54:34 -07001628 "Priority:%d extents:%d across:%lluk\n",
1629 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1630 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
1632 /* insert swap space into swap_list: */
1633 prev = -1;
1634 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1635 if (p->prio >= swap_info[i].prio) {
1636 break;
1637 }
1638 prev = i;
1639 }
1640 p->next = i;
1641 if (prev < 0) {
1642 swap_list.head = swap_list.next = p - swap_info;
1643 } else {
1644 swap_info[prev].next = p - swap_info;
1645 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07001646 spin_unlock(&swap_lock);
Ingo Molnarfc0abb12006-01-18 17:42:33 -08001647 mutex_unlock(&swapon_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 error = 0;
1649 goto out;
1650bad_swap:
1651 if (bdev) {
1652 set_blocksize(bdev, p->old_block_size);
1653 bd_release(bdev);
1654 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001655 destroy_swap_extents(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656bad_swap_2:
Hugh Dickins5d337b92005-09-03 15:54:41 -07001657 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 swap_map = p->swap_map;
1659 p->swap_file = NULL;
1660 p->swap_map = NULL;
1661 p->flags = 0;
1662 if (!(swap_flags & SWAP_FLAG_PREFER))
1663 ++least_priority;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001664 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 vfree(swap_map);
1666 if (swap_file)
1667 filp_close(swap_file, NULL);
1668out:
1669 if (page && !IS_ERR(page)) {
1670 kunmap(page);
1671 page_cache_release(page);
1672 }
1673 if (name)
1674 putname(name);
1675 if (did_down) {
1676 if (!error)
1677 inode->i_flags |= S_SWAPFILE;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001678 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 }
1680 return error;
1681}
1682
1683void si_swapinfo(struct sysinfo *val)
1684{
1685 unsigned int i;
1686 unsigned long nr_to_be_unused = 0;
1687
Hugh Dickins5d337b92005-09-03 15:54:41 -07001688 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 for (i = 0; i < nr_swapfiles; i++) {
1690 if (!(swap_info[i].flags & SWP_USED) ||
1691 (swap_info[i].flags & SWP_WRITEOK))
1692 continue;
1693 nr_to_be_unused += swap_info[i].inuse_pages;
1694 }
1695 val->freeswap = nr_swap_pages + nr_to_be_unused;
1696 val->totalswap = total_swap_pages + nr_to_be_unused;
Hugh Dickins5d337b92005-09-03 15:54:41 -07001697 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698}
1699
1700/*
1701 * Verify that a swap entry is valid and increment its swap map count.
1702 *
1703 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1704 * "permanent", but will be reclaimed by the next swapoff.
1705 */
1706int swap_duplicate(swp_entry_t entry)
1707{
1708 struct swap_info_struct * p;
1709 unsigned long offset, type;
1710 int result = 0;
1711
1712 type = swp_type(entry);
1713 if (type >= nr_swapfiles)
1714 goto bad_file;
1715 p = type + swap_info;
1716 offset = swp_offset(entry);
1717
Hugh Dickins5d337b92005-09-03 15:54:41 -07001718 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 if (offset < p->max && p->swap_map[offset]) {
1720 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1721 p->swap_map[offset]++;
1722 result = 1;
1723 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1724 if (swap_overflow++ < 5)
1725 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1726 p->swap_map[offset] = SWAP_MAP_MAX;
1727 result = 1;
1728 }
1729 }
Hugh Dickins5d337b92005-09-03 15:54:41 -07001730 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731out:
1732 return result;
1733
1734bad_file:
1735 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1736 goto out;
1737}
1738
1739struct swap_info_struct *
1740get_swap_info_struct(unsigned type)
1741{
1742 return &swap_info[type];
1743}
1744
1745/*
Hugh Dickins5d337b92005-09-03 15:54:41 -07001746 * swap_lock prevents swap_map being freed. Don't grab an extra
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 * reference on the swaphandle, it doesn't matter if it becomes unused.
1748 */
1749int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1750{
1751 int ret = 0, i = 1 << page_cluster;
1752 unsigned long toff;
1753 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1754
1755 if (!page_cluster) /* no readahead */
1756 return 0;
1757 toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1758 if (!toff) /* first page is swap header */
1759 toff++, i--;
1760 *offset = toff;
1761
Hugh Dickins5d337b92005-09-03 15:54:41 -07001762 spin_lock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 do {
1764 /* Don't read-ahead past the end of the swap area */
1765 if (toff >= swapdev->max)
1766 break;
1767 /* Don't read in free or bad pages */
1768 if (!swapdev->swap_map[toff])
1769 break;
1770 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1771 break;
1772 toff++;
1773 ret++;
1774 } while (--i);
Hugh Dickins5d337b92005-09-03 15:54:41 -07001775 spin_unlock(&swap_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 return ret;
1777}