blob: e675ae55f87dac005b7b8801f31a27e43d3e6cf6 [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>
28#include <linux/syscalls.h>
29
30#include <asm/pgtable.h>
31#include <asm/tlbflush.h>
32#include <linux/swapops.h>
33
34DEFINE_SPINLOCK(swaplock);
35unsigned int nr_swapfiles;
36long total_swap_pages;
37static int swap_overflow;
38
39EXPORT_SYMBOL(total_swap_pages);
40
41static 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
48struct swap_info_struct swap_info[MAX_SWAPFILES];
49
50static DECLARE_MUTEX(swapon_sem);
51
52/*
53 * We need this because the bdev->unplug_fn can sleep and we cannot
54 * hold swap_list_lock while calling the unplug_fn. And swap_list_lock
55 * cannot be turned into a semaphore.
56 */
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);
64 entry.val = page->private;
65 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
72 * count to avoid reading garbage from page->private above. If
73 * the WARN_ON triggers during a swapoff it maybe the race
74 * 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 Dickins52b7efdb2005-09-03 15:54:39 -0700108 swap_device_unlock(si);
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 Dickins52b7efdb2005-09-03 15:54:39 -0700118 swap_device_lock(si);
Hugh Dickins7dfad412005-09-03 15:54:38 -0700119 si->cluster_next = offset-SWAPFILE_CLUSTER-1;
120 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 Dickins52b7efdb2005-09-03 15:54:39 -0700127 swap_device_lock(si);
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 Dickins52b7efdb2005-09-03 15:54:39 -0700156 swap_device_unlock(si);
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]) {
159 swap_device_lock(si);
160 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 Dickins52b7efdb2005-09-03 15:54:39 -0700167 swap_device_lock(si);
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 swap_list_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;
202 swap_device_lock(si);
203 swap_list_unlock();
204 offset = scan_swap_map(si);
205 swap_device_unlock(si);
206 if (offset)
207 return swp_entry(type, offset);
208 swap_list_lock();
209 next = swap_list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 }
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700211
212 nr_swap_pages++;
213noswap:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 swap_list_unlock();
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -0700215 return (swp_entry_t) {0};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216}
217
218static struct swap_info_struct * swap_info_get(swp_entry_t entry)
219{
220 struct swap_info_struct * p;
221 unsigned long offset, type;
222
223 if (!entry.val)
224 goto out;
225 type = swp_type(entry);
226 if (type >= nr_swapfiles)
227 goto bad_nofile;
228 p = & swap_info[type];
229 if (!(p->flags & SWP_USED))
230 goto bad_device;
231 offset = swp_offset(entry);
232 if (offset >= p->max)
233 goto bad_offset;
234 if (!p->swap_map[offset])
235 goto bad_free;
236 swap_list_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 swap_device_lock(p);
238 return p;
239
240bad_free:
241 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
242 goto out;
243bad_offset:
244 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
245 goto out;
246bad_device:
247 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
248 goto out;
249bad_nofile:
250 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
251out:
252 return NULL;
253}
254
255static void swap_info_put(struct swap_info_struct * p)
256{
257 swap_device_unlock(p);
258 swap_list_unlock();
259}
260
261static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
262{
263 int count = p->swap_map[offset];
264
265 if (count < SWAP_MAP_MAX) {
266 count--;
267 p->swap_map[offset] = count;
268 if (!count) {
269 if (offset < p->lowest_bit)
270 p->lowest_bit = offset;
271 if (offset > p->highest_bit)
272 p->highest_bit = offset;
Hugh Dickins89d09a22005-09-03 15:54:36 -0700273 if (p->prio > swap_info[swap_list.next].prio)
274 swap_list.next = p - swap_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 nr_swap_pages++;
276 p->inuse_pages--;
277 }
278 }
279 return count;
280}
281
282/*
283 * Caller has made sure that the swapdevice corresponding to entry
284 * is still around or has not been recycled.
285 */
286void swap_free(swp_entry_t entry)
287{
288 struct swap_info_struct * p;
289
290 p = swap_info_get(entry);
291 if (p) {
292 swap_entry_free(p, swp_offset(entry));
293 swap_info_put(p);
294 }
295}
296
297/*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700298 * How many references to page are currently swapped out?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700300static inline int page_swapcount(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700302 int count = 0;
303 struct swap_info_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 swp_entry_t entry;
305
306 entry.val = page->private;
307 p = swap_info_get(entry);
308 if (p) {
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700309 /* Subtract the 1 for the swap cache itself */
310 count = p->swap_map[swp_offset(entry)] - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 swap_info_put(p);
312 }
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700313 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314}
315
316/*
317 * We can use this swap cache entry directly
318 * if there are no other references to it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 */
320int can_share_swap_page(struct page *page)
321{
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700322 int count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700324 BUG_ON(!PageLocked(page));
325 count = page_mapcount(page);
326 if (count <= 1 && PageSwapCache(page))
327 count += page_swapcount(page);
328 return count == 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329}
330
331/*
332 * Work out if there are any other processes sharing this
333 * swap cache page. Free it if you can. Return success.
334 */
335int remove_exclusive_swap_page(struct page *page)
336{
337 int retval;
338 struct swap_info_struct * p;
339 swp_entry_t entry;
340
341 BUG_ON(PagePrivate(page));
342 BUG_ON(!PageLocked(page));
343
344 if (!PageSwapCache(page))
345 return 0;
346 if (PageWriteback(page))
347 return 0;
348 if (page_count(page) != 2) /* 2: us + cache */
349 return 0;
350
351 entry.val = page->private;
352 p = swap_info_get(entry);
353 if (!p)
354 return 0;
355
356 /* Is the only swap cache user the cache itself? */
357 retval = 0;
358 if (p->swap_map[swp_offset(entry)] == 1) {
359 /* Recheck the page count with the swapcache lock held.. */
360 write_lock_irq(&swapper_space.tree_lock);
361 if ((page_count(page) == 2) && !PageWriteback(page)) {
362 __delete_from_swap_cache(page);
363 SetPageDirty(page);
364 retval = 1;
365 }
366 write_unlock_irq(&swapper_space.tree_lock);
367 }
368 swap_info_put(p);
369
370 if (retval) {
371 swap_free(entry);
372 page_cache_release(page);
373 }
374
375 return retval;
376}
377
378/*
379 * Free the swap entry like above, but also try to
380 * free the page cache entry if it is the last user.
381 */
382void free_swap_and_cache(swp_entry_t entry)
383{
384 struct swap_info_struct * p;
385 struct page *page = NULL;
386
387 p = swap_info_get(entry);
388 if (p) {
389 if (swap_entry_free(p, swp_offset(entry)) == 1)
390 page = find_trylock_page(&swapper_space, entry.val);
391 swap_info_put(p);
392 }
393 if (page) {
394 int one_user;
395
396 BUG_ON(PagePrivate(page));
397 page_cache_get(page);
398 one_user = (page_count(page) == 2);
399 /* Only cache user (+us), or swap space full? Free it! */
400 if (!PageWriteback(page) && (one_user || vm_swap_full())) {
401 delete_from_swap_cache(page);
402 SetPageDirty(page);
403 }
404 unlock_page(page);
405 page_cache_release(page);
406 }
407}
408
409/*
410 * Always set the resulting pte to be nowrite (the same as COW pages
411 * after one process has exited). We don't know just how many PTEs will
412 * share this swap entry, so be cautious and let do_wp_page work out
413 * what to do if a write is requested later.
414 *
415 * vma->vm_mm->page_table_lock is held.
416 */
417static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
418 unsigned long addr, swp_entry_t entry, struct page *page)
419{
420 inc_mm_counter(vma->vm_mm, rss);
421 get_page(page);
422 set_pte_at(vma->vm_mm, addr, pte,
423 pte_mkold(mk_pte(page, vma->vm_page_prot)));
424 page_add_anon_rmap(page, vma, addr);
425 swap_free(entry);
426 /*
427 * Move the page to the active list so it is not
428 * immediately swapped out again after swapon.
429 */
430 activate_page(page);
431}
432
433static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
434 unsigned long addr, unsigned long end,
435 swp_entry_t entry, struct page *page)
436{
437 pte_t *pte;
438 pte_t swp_pte = swp_entry_to_pte(entry);
439
440 pte = pte_offset_map(pmd, addr);
441 do {
442 /*
443 * swapoff spends a _lot_ of time in this loop!
444 * Test inline before going to call unuse_pte.
445 */
446 if (unlikely(pte_same(*pte, swp_pte))) {
447 unuse_pte(vma, pte, addr, entry, page);
448 pte_unmap(pte);
449 return 1;
450 }
451 } while (pte++, addr += PAGE_SIZE, addr != end);
452 pte_unmap(pte - 1);
453 return 0;
454}
455
456static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
457 unsigned long addr, unsigned long end,
458 swp_entry_t entry, struct page *page)
459{
460 pmd_t *pmd;
461 unsigned long next;
462
463 pmd = pmd_offset(pud, addr);
464 do {
465 next = pmd_addr_end(addr, end);
466 if (pmd_none_or_clear_bad(pmd))
467 continue;
468 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
469 return 1;
470 } while (pmd++, addr = next, addr != end);
471 return 0;
472}
473
474static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
475 unsigned long addr, unsigned long end,
476 swp_entry_t entry, struct page *page)
477{
478 pud_t *pud;
479 unsigned long next;
480
481 pud = pud_offset(pgd, addr);
482 do {
483 next = pud_addr_end(addr, end);
484 if (pud_none_or_clear_bad(pud))
485 continue;
486 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
487 return 1;
488 } while (pud++, addr = next, addr != end);
489 return 0;
490}
491
492static int unuse_vma(struct vm_area_struct *vma,
493 swp_entry_t entry, struct page *page)
494{
495 pgd_t *pgd;
496 unsigned long addr, end, next;
497
498 if (page->mapping) {
499 addr = page_address_in_vma(page, vma);
500 if (addr == -EFAULT)
501 return 0;
502 else
503 end = addr + PAGE_SIZE;
504 } else {
505 addr = vma->vm_start;
506 end = vma->vm_end;
507 }
508
509 pgd = pgd_offset(vma->vm_mm, addr);
510 do {
511 next = pgd_addr_end(addr, end);
512 if (pgd_none_or_clear_bad(pgd))
513 continue;
514 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
515 return 1;
516 } while (pgd++, addr = next, addr != end);
517 return 0;
518}
519
520static int unuse_mm(struct mm_struct *mm,
521 swp_entry_t entry, struct page *page)
522{
523 struct vm_area_struct *vma;
524
525 if (!down_read_trylock(&mm->mmap_sem)) {
526 /*
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700527 * Activate page so shrink_cache is unlikely to unmap its
528 * ptes while lock is dropped, so swapoff can make progress.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 */
Hugh Dickinsc475a8a2005-06-21 17:15:12 -0700530 activate_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 unlock_page(page);
532 down_read(&mm->mmap_sem);
533 lock_page(page);
534 }
535 spin_lock(&mm->page_table_lock);
536 for (vma = mm->mmap; vma; vma = vma->vm_next) {
537 if (vma->anon_vma && unuse_vma(vma, entry, page))
538 break;
539 }
540 spin_unlock(&mm->page_table_lock);
541 up_read(&mm->mmap_sem);
542 /*
543 * Currently unuse_mm cannot fail, but leave error handling
544 * at call sites for now, since we change it from time to time.
545 */
546 return 0;
547}
548
549/*
550 * Scan swap_map from current position to next entry still in use.
551 * Recycle to start on reaching the end, returning 0 when empty.
552 */
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700553static unsigned int find_next_to_unuse(struct swap_info_struct *si,
554 unsigned int prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555{
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700556 unsigned int max = si->max;
557 unsigned int i = prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 int count;
559
560 /*
561 * No need for swap_device_lock(si) here: we're just looking
562 * for whether an entry is in use, not modifying it; false
563 * hits are okay, and sys_swapoff() has already prevented new
564 * allocations from this area (while holding swap_list_lock()).
565 */
566 for (;;) {
567 if (++i >= max) {
568 if (!prev) {
569 i = 0;
570 break;
571 }
572 /*
573 * No entries in use at top of swap_map,
574 * loop back to start and recheck there.
575 */
576 max = prev + 1;
577 prev = 0;
578 i = 1;
579 }
580 count = si->swap_map[i];
581 if (count && count != SWAP_MAP_BAD)
582 break;
583 }
584 return i;
585}
586
587/*
588 * We completely avoid races by reading each swap page in advance,
589 * and then search for the process using it. All the necessary
590 * page table adjustments can then be made atomically.
591 */
592static int try_to_unuse(unsigned int type)
593{
594 struct swap_info_struct * si = &swap_info[type];
595 struct mm_struct *start_mm;
596 unsigned short *swap_map;
597 unsigned short swcount;
598 struct page *page;
599 swp_entry_t entry;
Hugh Dickins6eb396d2005-09-03 15:54:35 -0700600 unsigned int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 int retval = 0;
602 int reset_overflow = 0;
603 int shmem;
604
605 /*
606 * When searching mms for an entry, a good strategy is to
607 * start at the first mm we freed the previous entry from
608 * (though actually we don't notice whether we or coincidence
609 * freed the entry). Initialize this start_mm with a hold.
610 *
611 * A simpler strategy would be to start at the last mm we
612 * freed the previous entry from; but that would take less
613 * advantage of mmlist ordering, which clusters forked mms
614 * together, child after parent. If we race with dup_mmap(), we
615 * prefer to resolve parent before child, lest we miss entries
616 * duplicated after we scanned child: using last mm would invert
617 * that. Though it's only a serious concern when an overflowed
618 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
619 */
620 start_mm = &init_mm;
621 atomic_inc(&init_mm.mm_users);
622
623 /*
624 * Keep on scanning until all entries have gone. Usually,
625 * one pass through swap_map is enough, but not necessarily:
626 * there are races when an instance of an entry might be missed.
627 */
628 while ((i = find_next_to_unuse(si, i)) != 0) {
629 if (signal_pending(current)) {
630 retval = -EINTR;
631 break;
632 }
633
634 /*
635 * Get a page for the entry, using the existing swap
636 * cache page if there is one. Otherwise, get a clean
637 * page and read the swap into it.
638 */
639 swap_map = &si->swap_map[i];
640 entry = swp_entry(type, i);
641 page = read_swap_cache_async(entry, NULL, 0);
642 if (!page) {
643 /*
644 * Either swap_duplicate() failed because entry
645 * has been freed independently, and will not be
646 * reused since sys_swapoff() already disabled
647 * allocation from here, or alloc_page() failed.
648 */
649 if (!*swap_map)
650 continue;
651 retval = -ENOMEM;
652 break;
653 }
654
655 /*
656 * Don't hold on to start_mm if it looks like exiting.
657 */
658 if (atomic_read(&start_mm->mm_users) == 1) {
659 mmput(start_mm);
660 start_mm = &init_mm;
661 atomic_inc(&init_mm.mm_users);
662 }
663
664 /*
665 * Wait for and lock page. When do_swap_page races with
666 * try_to_unuse, do_swap_page can handle the fault much
667 * faster than try_to_unuse can locate the entry. This
668 * apparently redundant "wait_on_page_locked" lets try_to_unuse
669 * defer to do_swap_page in such a case - in some tests,
670 * do_swap_page and try_to_unuse repeatedly compete.
671 */
672 wait_on_page_locked(page);
673 wait_on_page_writeback(page);
674 lock_page(page);
675 wait_on_page_writeback(page);
676
677 /*
678 * Remove all references to entry.
679 * Whenever we reach init_mm, there's no address space
680 * to search, but use it as a reminder to search shmem.
681 */
682 shmem = 0;
683 swcount = *swap_map;
684 if (swcount > 1) {
685 if (start_mm == &init_mm)
686 shmem = shmem_unuse(entry, page);
687 else
688 retval = unuse_mm(start_mm, entry, page);
689 }
690 if (*swap_map > 1) {
691 int set_start_mm = (*swap_map >= swcount);
692 struct list_head *p = &start_mm->mmlist;
693 struct mm_struct *new_start_mm = start_mm;
694 struct mm_struct *prev_mm = start_mm;
695 struct mm_struct *mm;
696
697 atomic_inc(&new_start_mm->mm_users);
698 atomic_inc(&prev_mm->mm_users);
699 spin_lock(&mmlist_lock);
700 while (*swap_map > 1 && !retval &&
701 (p = p->next) != &start_mm->mmlist) {
702 mm = list_entry(p, struct mm_struct, mmlist);
703 if (atomic_inc_return(&mm->mm_users) == 1) {
704 atomic_dec(&mm->mm_users);
705 continue;
706 }
707 spin_unlock(&mmlist_lock);
708 mmput(prev_mm);
709 prev_mm = mm;
710
711 cond_resched();
712
713 swcount = *swap_map;
714 if (swcount <= 1)
715 ;
716 else if (mm == &init_mm) {
717 set_start_mm = 1;
718 shmem = shmem_unuse(entry, page);
719 } else
720 retval = unuse_mm(mm, entry, page);
721 if (set_start_mm && *swap_map < swcount) {
722 mmput(new_start_mm);
723 atomic_inc(&mm->mm_users);
724 new_start_mm = mm;
725 set_start_mm = 0;
726 }
727 spin_lock(&mmlist_lock);
728 }
729 spin_unlock(&mmlist_lock);
730 mmput(prev_mm);
731 mmput(start_mm);
732 start_mm = new_start_mm;
733 }
734 if (retval) {
735 unlock_page(page);
736 page_cache_release(page);
737 break;
738 }
739
740 /*
741 * How could swap count reach 0x7fff when the maximum
742 * pid is 0x7fff, and there's no way to repeat a swap
743 * page within an mm (except in shmem, where it's the
744 * shared object which takes the reference count)?
745 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
746 *
747 * If that's wrong, then we should worry more about
748 * exit_mmap() and do_munmap() cases described above:
749 * we might be resetting SWAP_MAP_MAX too early here.
750 * We know "Undead"s can happen, they're okay, so don't
751 * report them; but do report if we reset SWAP_MAP_MAX.
752 */
753 if (*swap_map == SWAP_MAP_MAX) {
754 swap_device_lock(si);
755 *swap_map = 1;
756 swap_device_unlock(si);
757 reset_overflow = 1;
758 }
759
760 /*
761 * If a reference remains (rare), we would like to leave
762 * the page in the swap cache; but try_to_unmap could
763 * then re-duplicate the entry once we drop page lock,
764 * so we might loop indefinitely; also, that page could
765 * not be swapped out to other storage meanwhile. So:
766 * delete from cache even if there's another reference,
767 * after ensuring that the data has been saved to disk -
768 * since if the reference remains (rarer), it will be
769 * read from disk into another page. Splitting into two
770 * pages would be incorrect if swap supported "shared
771 * private" pages, but they are handled by tmpfs files.
772 *
773 * Note shmem_unuse already deleted a swappage from
774 * the swap cache, unless the move to filepage failed:
775 * in which case it left swappage in cache, lowered its
776 * swap count to pass quickly through the loops above,
777 * and now we must reincrement count to try again later.
778 */
779 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
780 struct writeback_control wbc = {
781 .sync_mode = WB_SYNC_NONE,
782 };
783
784 swap_writepage(page, &wbc);
785 lock_page(page);
786 wait_on_page_writeback(page);
787 }
788 if (PageSwapCache(page)) {
789 if (shmem)
790 swap_duplicate(entry);
791 else
792 delete_from_swap_cache(page);
793 }
794
795 /*
796 * So we could skip searching mms once swap count went
797 * to 1, we did not mark any present ptes as dirty: must
798 * mark page dirty so shrink_list will preserve it.
799 */
800 SetPageDirty(page);
801 unlock_page(page);
802 page_cache_release(page);
803
804 /*
805 * Make sure that we aren't completely killing
806 * interactive performance.
807 */
808 cond_resched();
809 }
810
811 mmput(start_mm);
812 if (reset_overflow) {
813 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
814 swap_overflow = 0;
815 }
816 return retval;
817}
818
819/*
820 * After a successful try_to_unuse, if no swap is now in use, we know we
821 * can empty the mmlist. swap_list_lock must be held on entry and exit.
822 * Note that mmlist_lock nests inside swap_list_lock, and an mm must be
823 * added to the mmlist just after page_duplicate - before would be racy.
824 */
825static void drain_mmlist(void)
826{
827 struct list_head *p, *next;
828 unsigned int i;
829
830 for (i = 0; i < nr_swapfiles; i++)
831 if (swap_info[i].inuse_pages)
832 return;
833 spin_lock(&mmlist_lock);
834 list_for_each_safe(p, next, &init_mm.mmlist)
835 list_del_init(p);
836 spin_unlock(&mmlist_lock);
837}
838
839/*
840 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
841 * corresponds to page offset `offset'.
842 */
843sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
844{
845 struct swap_extent *se = sis->curr_swap_extent;
846 struct swap_extent *start_se = se;
847
848 for ( ; ; ) {
849 struct list_head *lh;
850
851 if (se->start_page <= offset &&
852 offset < (se->start_page + se->nr_pages)) {
853 return se->start_block + (offset - se->start_page);
854 }
Hugh Dickins11d31882005-09-03 15:54:34 -0700855 lh = se->list.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 if (lh == &sis->extent_list)
Hugh Dickins11d31882005-09-03 15:54:34 -0700857 lh = lh->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 se = list_entry(lh, struct swap_extent, list);
859 sis->curr_swap_extent = se;
860 BUG_ON(se == start_se); /* It *must* be present */
861 }
862}
863
864/*
865 * Free all of a swapdev's extent information
866 */
867static void destroy_swap_extents(struct swap_info_struct *sis)
868{
869 while (!list_empty(&sis->extent_list)) {
870 struct swap_extent *se;
871
872 se = list_entry(sis->extent_list.next,
873 struct swap_extent, list);
874 list_del(&se->list);
875 kfree(se);
876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877}
878
879/*
880 * Add a block range (and the corresponding page range) into this swapdev's
Hugh Dickins11d31882005-09-03 15:54:34 -0700881 * extent list. The extent list is kept sorted in page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 *
Hugh Dickins11d31882005-09-03 15:54:34 -0700883 * This function rather assumes that it is called in ascending page order.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 */
885static int
886add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
887 unsigned long nr_pages, sector_t start_block)
888{
889 struct swap_extent *se;
890 struct swap_extent *new_se;
891 struct list_head *lh;
892
Hugh Dickins11d31882005-09-03 15:54:34 -0700893 lh = sis->extent_list.prev; /* The highest page extent */
894 if (lh != &sis->extent_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 se = list_entry(lh, struct swap_extent, list);
Hugh Dickins11d31882005-09-03 15:54:34 -0700896 BUG_ON(se->start_page + se->nr_pages != start_page);
897 if (se->start_block + se->nr_pages == start_block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 /* Merge it */
899 se->nr_pages += nr_pages;
900 return 0;
901 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 }
903
904 /*
905 * No merge. Insert a new extent, preserving ordering.
906 */
907 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
908 if (new_se == NULL)
909 return -ENOMEM;
910 new_se->start_page = start_page;
911 new_se->nr_pages = nr_pages;
912 new_se->start_block = start_block;
913
Hugh Dickins11d31882005-09-03 15:54:34 -0700914 list_add_tail(&new_se->list, &sis->extent_list);
Hugh Dickins53092a72005-09-03 15:54:34 -0700915 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916}
917
918/*
919 * A `swap extent' is a simple thing which maps a contiguous range of pages
920 * onto a contiguous range of disk blocks. An ordered list of swap extents
921 * is built at swapon time and is then used at swap_writepage/swap_readpage
922 * time for locating where on disk a page belongs.
923 *
924 * If the swapfile is an S_ISBLK block device, a single extent is installed.
925 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
926 * swap files identically.
927 *
928 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
929 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
930 * swapfiles are handled *identically* after swapon time.
931 *
932 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
933 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
934 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
935 * requirements, they are simply tossed out - we will never use those blocks
936 * for swapping.
937 *
Hugh Dickinsb0d9bcd2005-09-03 15:54:31 -0700938 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
940 * which will scribble on the fs.
941 *
942 * The amount of disk space which a single swap extent represents varies.
943 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
944 * extents in the list. To avoid much list walking, we cache the previous
945 * search location in `curr_swap_extent', and start new searches from there.
946 * This is extremely effective. The average number of iterations in
947 * map_swap_page() has been measured at about 0.3 per page. - akpm.
948 */
Hugh Dickins53092a72005-09-03 15:54:34 -0700949static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950{
951 struct inode *inode;
952 unsigned blocks_per_page;
953 unsigned long page_no;
954 unsigned blkbits;
955 sector_t probe_block;
956 sector_t last_block;
Hugh Dickins53092a72005-09-03 15:54:34 -0700957 sector_t lowest_block = -1;
958 sector_t highest_block = 0;
959 int nr_extents = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 int ret;
961
962 inode = sis->swap_file->f_mapping->host;
963 if (S_ISBLK(inode->i_mode)) {
964 ret = add_swap_extent(sis, 0, sis->max, 0);
Hugh Dickins53092a72005-09-03 15:54:34 -0700965 *span = sis->pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 goto done;
967 }
968
969 blkbits = inode->i_blkbits;
970 blocks_per_page = PAGE_SIZE >> blkbits;
971
972 /*
973 * Map all the blocks into the extent list. This code doesn't try
974 * to be very smart.
975 */
976 probe_block = 0;
977 page_no = 0;
978 last_block = i_size_read(inode) >> blkbits;
979 while ((probe_block + blocks_per_page) <= last_block &&
980 page_no < sis->max) {
981 unsigned block_in_page;
982 sector_t first_block;
983
984 first_block = bmap(inode, probe_block);
985 if (first_block == 0)
986 goto bad_bmap;
987
988 /*
989 * It must be PAGE_SIZE aligned on-disk
990 */
991 if (first_block & (blocks_per_page - 1)) {
992 probe_block++;
993 goto reprobe;
994 }
995
996 for (block_in_page = 1; block_in_page < blocks_per_page;
997 block_in_page++) {
998 sector_t block;
999
1000 block = bmap(inode, probe_block + block_in_page);
1001 if (block == 0)
1002 goto bad_bmap;
1003 if (block != first_block + block_in_page) {
1004 /* Discontiguity */
1005 probe_block++;
1006 goto reprobe;
1007 }
1008 }
1009
Hugh Dickins53092a72005-09-03 15:54:34 -07001010 first_block >>= (PAGE_SHIFT - blkbits);
1011 if (page_no) { /* exclude the header page */
1012 if (first_block < lowest_block)
1013 lowest_block = first_block;
1014 if (first_block > highest_block)
1015 highest_block = first_block;
1016 }
1017
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 /*
1019 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1020 */
Hugh Dickins53092a72005-09-03 15:54:34 -07001021 ret = add_swap_extent(sis, page_no, 1, first_block);
1022 if (ret < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 goto out;
Hugh Dickins53092a72005-09-03 15:54:34 -07001024 nr_extents += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 page_no++;
1026 probe_block += blocks_per_page;
1027reprobe:
1028 continue;
1029 }
Hugh Dickins53092a72005-09-03 15:54:34 -07001030 ret = nr_extents;
1031 *span = 1 + highest_block - lowest_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 if (page_no == 0)
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001033 page_no = 1; /* force Empty message */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 sis->max = page_no;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001035 sis->pages = page_no - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 sis->highest_bit = page_no - 1;
1037done:
1038 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1039 struct swap_extent, list);
1040 goto out;
1041bad_bmap:
1042 printk(KERN_ERR "swapon: swapfile has holes\n");
1043 ret = -EINVAL;
1044out:
1045 return ret;
1046}
1047
1048#if 0 /* We don't need this yet */
1049#include <linux/backing-dev.h>
1050int page_queue_congested(struct page *page)
1051{
1052 struct backing_dev_info *bdi;
1053
1054 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1055
1056 if (PageSwapCache(page)) {
1057 swp_entry_t entry = { .val = page->private };
1058 struct swap_info_struct *sis;
1059
1060 sis = get_swap_info_struct(swp_type(entry));
1061 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1062 } else
1063 bdi = page->mapping->backing_dev_info;
1064 return bdi_write_congested(bdi);
1065}
1066#endif
1067
1068asmlinkage long sys_swapoff(const char __user * specialfile)
1069{
1070 struct swap_info_struct * p = NULL;
1071 unsigned short *swap_map;
1072 struct file *swap_file, *victim;
1073 struct address_space *mapping;
1074 struct inode *inode;
1075 char * pathname;
1076 int i, type, prev;
1077 int err;
1078
1079 if (!capable(CAP_SYS_ADMIN))
1080 return -EPERM;
1081
1082 pathname = getname(specialfile);
1083 err = PTR_ERR(pathname);
1084 if (IS_ERR(pathname))
1085 goto out;
1086
1087 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1088 putname(pathname);
1089 err = PTR_ERR(victim);
1090 if (IS_ERR(victim))
1091 goto out;
1092
1093 mapping = victim->f_mapping;
1094 prev = -1;
1095 swap_list_lock();
1096 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1097 p = swap_info + type;
1098 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1099 if (p->swap_file->f_mapping == mapping)
1100 break;
1101 }
1102 prev = type;
1103 }
1104 if (type < 0) {
1105 err = -EINVAL;
1106 swap_list_unlock();
1107 goto out_dput;
1108 }
1109 if (!security_vm_enough_memory(p->pages))
1110 vm_unacct_memory(p->pages);
1111 else {
1112 err = -ENOMEM;
1113 swap_list_unlock();
1114 goto out_dput;
1115 }
1116 if (prev < 0) {
1117 swap_list.head = p->next;
1118 } else {
1119 swap_info[prev].next = p->next;
1120 }
1121 if (type == swap_list.next) {
1122 /* just pick something that's safe... */
1123 swap_list.next = swap_list.head;
1124 }
1125 nr_swap_pages -= p->pages;
1126 total_swap_pages -= p->pages;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001127 swap_device_lock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 p->flags &= ~SWP_WRITEOK;
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001129 swap_device_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 swap_list_unlock();
Hugh Dickinsfb4f88d2005-09-03 15:54:37 -07001131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 current->flags |= PF_SWAPOFF;
1133 err = try_to_unuse(type);
1134 current->flags &= ~PF_SWAPOFF;
1135
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 if (err) {
1137 /* re-insert swap space back into swap_list */
1138 swap_list_lock();
1139 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1140 if (p->prio >= swap_info[i].prio)
1141 break;
1142 p->next = i;
1143 if (prev < 0)
1144 swap_list.head = swap_list.next = p - swap_info;
1145 else
1146 swap_info[prev].next = p - swap_info;
1147 nr_swap_pages += p->pages;
1148 total_swap_pages += p->pages;
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001149 swap_device_lock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 p->flags |= SWP_WRITEOK;
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001151 swap_device_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 swap_list_unlock();
1153 goto out_dput;
1154 }
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001155
1156 /* wait for any unplug function to finish */
1157 down_write(&swap_unplug_sem);
1158 up_write(&swap_unplug_sem);
1159
1160 /* wait for anyone still in scan_swap_map */
1161 swap_device_lock(p);
1162 p->highest_bit = 0; /* cuts scans short */
1163 while (p->flags >= SWP_SCANNING) {
1164 swap_device_unlock(p);
1165 set_current_state(TASK_UNINTERRUPTIBLE);
1166 schedule_timeout(1);
1167 swap_device_lock(p);
1168 }
1169 swap_device_unlock(p);
1170
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001171 destroy_swap_extents(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 down(&swapon_sem);
1173 swap_list_lock();
1174 drain_mmlist();
1175 swap_device_lock(p);
1176 swap_file = p->swap_file;
1177 p->swap_file = NULL;
1178 p->max = 0;
1179 swap_map = p->swap_map;
1180 p->swap_map = NULL;
1181 p->flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 swap_device_unlock(p);
1183 swap_list_unlock();
1184 up(&swapon_sem);
1185 vfree(swap_map);
1186 inode = mapping->host;
1187 if (S_ISBLK(inode->i_mode)) {
1188 struct block_device *bdev = I_BDEV(inode);
1189 set_blocksize(bdev, p->old_block_size);
1190 bd_release(bdev);
1191 } else {
1192 down(&inode->i_sem);
1193 inode->i_flags &= ~S_SWAPFILE;
1194 up(&inode->i_sem);
1195 }
1196 filp_close(swap_file, NULL);
1197 err = 0;
1198
1199out_dput:
1200 filp_close(victim, NULL);
1201out:
1202 return err;
1203}
1204
1205#ifdef CONFIG_PROC_FS
1206/* iterator */
1207static void *swap_start(struct seq_file *swap, loff_t *pos)
1208{
1209 struct swap_info_struct *ptr = swap_info;
1210 int i;
1211 loff_t l = *pos;
1212
1213 down(&swapon_sem);
1214
1215 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1216 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1217 continue;
1218 if (!l--)
1219 return ptr;
1220 }
1221
1222 return NULL;
1223}
1224
1225static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1226{
1227 struct swap_info_struct *ptr = v;
1228 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1229
1230 for (++ptr; ptr < endptr; ptr++) {
1231 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1232 continue;
1233 ++*pos;
1234 return ptr;
1235 }
1236
1237 return NULL;
1238}
1239
1240static void swap_stop(struct seq_file *swap, void *v)
1241{
1242 up(&swapon_sem);
1243}
1244
1245static int swap_show(struct seq_file *swap, void *v)
1246{
1247 struct swap_info_struct *ptr = v;
1248 struct file *file;
1249 int len;
1250
1251 if (v == swap_info)
1252 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1253
1254 file = ptr->swap_file;
1255 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001256 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 len < 40 ? 40 - len : 1, " ",
1258 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1259 "partition" : "file\t",
1260 ptr->pages << (PAGE_SHIFT - 10),
1261 ptr->inuse_pages << (PAGE_SHIFT - 10),
1262 ptr->prio);
1263 return 0;
1264}
1265
1266static struct seq_operations swaps_op = {
1267 .start = swap_start,
1268 .next = swap_next,
1269 .stop = swap_stop,
1270 .show = swap_show
1271};
1272
1273static int swaps_open(struct inode *inode, struct file *file)
1274{
1275 return seq_open(file, &swaps_op);
1276}
1277
1278static struct file_operations proc_swaps_operations = {
1279 .open = swaps_open,
1280 .read = seq_read,
1281 .llseek = seq_lseek,
1282 .release = seq_release,
1283};
1284
1285static int __init procswaps_init(void)
1286{
1287 struct proc_dir_entry *entry;
1288
1289 entry = create_proc_entry("swaps", 0, NULL);
1290 if (entry)
1291 entry->proc_fops = &proc_swaps_operations;
1292 return 0;
1293}
1294__initcall(procswaps_init);
1295#endif /* CONFIG_PROC_FS */
1296
1297/*
1298 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1299 *
1300 * The swapon system call
1301 */
1302asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1303{
1304 struct swap_info_struct * p;
1305 char *name = NULL;
1306 struct block_device *bdev = NULL;
1307 struct file *swap_file = NULL;
1308 struct address_space *mapping;
1309 unsigned int type;
1310 int i, prev;
1311 int error;
1312 static int least_priority;
1313 union swap_header *swap_header = NULL;
1314 int swap_header_version;
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001315 unsigned int nr_good_pages = 0;
1316 int nr_extents = 0;
Hugh Dickins53092a72005-09-03 15:54:34 -07001317 sector_t span;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 unsigned long maxpages = 1;
1319 int swapfilesize;
1320 unsigned short *swap_map;
1321 struct page *page = NULL;
1322 struct inode *inode = NULL;
1323 int did_down = 0;
1324
1325 if (!capable(CAP_SYS_ADMIN))
1326 return -EPERM;
1327 swap_list_lock();
1328 p = swap_info;
1329 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1330 if (!(p->flags & SWP_USED))
1331 break;
1332 error = -EPERM;
1333 /*
1334 * Test if adding another swap device is possible. There are
1335 * two limiting factors: 1) the number of bits for the swap
1336 * type swp_entry_t definition and 2) the number of bits for
1337 * the swap type in the swap ptes as defined by the different
1338 * architectures. To honor both limitations a swap entry
1339 * with swap offset 0 and swap type ~0UL is created, encoded
1340 * to a swap pte, decoded to a swp_entry_t again and finally
1341 * the swap type part is extracted. This will mask all bits
1342 * from the initial ~0UL that can't be encoded in either the
1343 * swp_entry_t or the architecture definition of a swap pte.
1344 */
1345 if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1346 swap_list_unlock();
1347 goto out;
1348 }
1349 if (type >= nr_swapfiles)
1350 nr_swapfiles = type+1;
1351 INIT_LIST_HEAD(&p->extent_list);
1352 p->flags = SWP_USED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 p->swap_file = NULL;
1354 p->old_block_size = 0;
1355 p->swap_map = NULL;
1356 p->lowest_bit = 0;
1357 p->highest_bit = 0;
1358 p->cluster_nr = 0;
1359 p->inuse_pages = 0;
1360 spin_lock_init(&p->sdev_lock);
1361 p->next = -1;
1362 if (swap_flags & SWAP_FLAG_PREFER) {
1363 p->prio =
1364 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1365 } else {
1366 p->prio = --least_priority;
1367 }
1368 swap_list_unlock();
1369 name = getname(specialfile);
1370 error = PTR_ERR(name);
1371 if (IS_ERR(name)) {
1372 name = NULL;
1373 goto bad_swap_2;
1374 }
1375 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1376 error = PTR_ERR(swap_file);
1377 if (IS_ERR(swap_file)) {
1378 swap_file = NULL;
1379 goto bad_swap_2;
1380 }
1381
1382 p->swap_file = swap_file;
1383 mapping = swap_file->f_mapping;
1384 inode = mapping->host;
1385
1386 error = -EBUSY;
1387 for (i = 0; i < nr_swapfiles; i++) {
1388 struct swap_info_struct *q = &swap_info[i];
1389
1390 if (i == type || !q->swap_file)
1391 continue;
1392 if (mapping == q->swap_file->f_mapping)
1393 goto bad_swap;
1394 }
1395
1396 error = -EINVAL;
1397 if (S_ISBLK(inode->i_mode)) {
1398 bdev = I_BDEV(inode);
1399 error = bd_claim(bdev, sys_swapon);
1400 if (error < 0) {
1401 bdev = NULL;
1402 goto bad_swap;
1403 }
1404 p->old_block_size = block_size(bdev);
1405 error = set_blocksize(bdev, PAGE_SIZE);
1406 if (error < 0)
1407 goto bad_swap;
1408 p->bdev = bdev;
1409 } else if (S_ISREG(inode->i_mode)) {
1410 p->bdev = inode->i_sb->s_bdev;
1411 down(&inode->i_sem);
1412 did_down = 1;
1413 if (IS_SWAPFILE(inode)) {
1414 error = -EBUSY;
1415 goto bad_swap;
1416 }
1417 } else {
1418 goto bad_swap;
1419 }
1420
1421 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1422
1423 /*
1424 * Read the swap header.
1425 */
1426 if (!mapping->a_ops->readpage) {
1427 error = -EINVAL;
1428 goto bad_swap;
1429 }
1430 page = read_cache_page(mapping, 0,
1431 (filler_t *)mapping->a_ops->readpage, swap_file);
1432 if (IS_ERR(page)) {
1433 error = PTR_ERR(page);
1434 goto bad_swap;
1435 }
1436 wait_on_page_locked(page);
1437 if (!PageUptodate(page))
1438 goto bad_swap;
1439 kmap(page);
1440 swap_header = page_address(page);
1441
1442 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1443 swap_header_version = 1;
1444 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1445 swap_header_version = 2;
1446 else {
1447 printk("Unable to find swap-space signature\n");
1448 error = -EINVAL;
1449 goto bad_swap;
1450 }
1451
1452 switch (swap_header_version) {
1453 case 1:
1454 printk(KERN_ERR "version 0 swap is no longer supported. "
1455 "Use mkswap -v1 %s\n", name);
1456 error = -EINVAL;
1457 goto bad_swap;
1458 case 2:
1459 /* Check the swap header's sub-version and the size of
1460 the swap file and bad block lists */
1461 if (swap_header->info.version != 1) {
1462 printk(KERN_WARNING
1463 "Unable to handle swap header version %d\n",
1464 swap_header->info.version);
1465 error = -EINVAL;
1466 goto bad_swap;
1467 }
1468
1469 p->lowest_bit = 1;
Hugh Dickins52b7efdb2005-09-03 15:54:39 -07001470 p->cluster_next = 1;
1471
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 /*
1473 * Find out how many pages are allowed for a single swap
1474 * device. There are two limiting factors: 1) the number of
1475 * bits for the swap offset in the swp_entry_t type and
1476 * 2) the number of bits in the a swap pte as defined by
1477 * the different architectures. In order to find the
1478 * largest possible bit mask a swap entry with swap type 0
1479 * and swap offset ~0UL is created, encoded to a swap pte,
1480 * decoded to a swp_entry_t again and finally the swap
1481 * offset is extracted. This will mask all the bits from
1482 * the initial ~0UL mask that can't be encoded in either
1483 * the swp_entry_t or the architecture definition of a
1484 * swap pte.
1485 */
1486 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1487 if (maxpages > swap_header->info.last_page)
1488 maxpages = swap_header->info.last_page;
1489 p->highest_bit = maxpages - 1;
1490
1491 error = -EINVAL;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001492 if (!maxpages)
1493 goto bad_swap;
1494 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1495 goto bad_swap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1497 goto bad_swap;
1498
1499 /* OK, set up the swap map and apply the bad block list */
1500 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1501 error = -ENOMEM;
1502 goto bad_swap;
1503 }
1504
1505 error = 0;
1506 memset(p->swap_map, 0, maxpages * sizeof(short));
1507 for (i=0; i<swap_header->info.nr_badpages; i++) {
1508 int page = swap_header->info.badpages[i];
1509 if (page <= 0 || page >= swap_header->info.last_page)
1510 error = -EINVAL;
1511 else
1512 p->swap_map[page] = SWAP_MAP_BAD;
1513 }
1514 nr_good_pages = swap_header->info.last_page -
1515 swap_header->info.nr_badpages -
1516 1 /* header page */;
1517 if (error)
1518 goto bad_swap;
1519 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 if (swapfilesize && maxpages > swapfilesize) {
1522 printk(KERN_WARNING
1523 "Swap area shorter than signature indicates\n");
1524 error = -EINVAL;
1525 goto bad_swap;
1526 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001527 if (nr_good_pages) {
1528 p->swap_map[0] = SWAP_MAP_BAD;
1529 p->max = maxpages;
1530 p->pages = nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07001531 nr_extents = setup_swap_extents(p, &span);
1532 if (nr_extents < 0) {
1533 error = nr_extents;
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001534 goto bad_swap;
Hugh Dickins53092a72005-09-03 15:54:34 -07001535 }
Hugh Dickinse2244ec2005-09-03 15:54:32 -07001536 nr_good_pages = p->pages;
1537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 if (!nr_good_pages) {
1539 printk(KERN_WARNING "Empty swap-file\n");
1540 error = -EINVAL;
1541 goto bad_swap;
1542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
1544 down(&swapon_sem);
1545 swap_list_lock();
1546 swap_device_lock(p);
1547 p->flags = SWP_ACTIVE;
1548 nr_swap_pages += nr_good_pages;
1549 total_swap_pages += nr_good_pages;
Hugh Dickins53092a72005-09-03 15:54:34 -07001550
Hugh Dickins6eb396d2005-09-03 15:54:35 -07001551 printk(KERN_INFO "Adding %uk swap on %s. "
Hugh Dickins53092a72005-09-03 15:54:34 -07001552 "Priority:%d extents:%d across:%lluk\n",
1553 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1554 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
1556 /* insert swap space into swap_list: */
1557 prev = -1;
1558 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1559 if (p->prio >= swap_info[i].prio) {
1560 break;
1561 }
1562 prev = i;
1563 }
1564 p->next = i;
1565 if (prev < 0) {
1566 swap_list.head = swap_list.next = p - swap_info;
1567 } else {
1568 swap_info[prev].next = p - swap_info;
1569 }
1570 swap_device_unlock(p);
1571 swap_list_unlock();
1572 up(&swapon_sem);
1573 error = 0;
1574 goto out;
1575bad_swap:
1576 if (bdev) {
1577 set_blocksize(bdev, p->old_block_size);
1578 bd_release(bdev);
1579 }
Hugh Dickins4cd3bb12005-09-03 15:54:33 -07001580 destroy_swap_extents(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581bad_swap_2:
1582 swap_list_lock();
1583 swap_map = p->swap_map;
1584 p->swap_file = NULL;
1585 p->swap_map = NULL;
1586 p->flags = 0;
1587 if (!(swap_flags & SWAP_FLAG_PREFER))
1588 ++least_priority;
1589 swap_list_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 vfree(swap_map);
1591 if (swap_file)
1592 filp_close(swap_file, NULL);
1593out:
1594 if (page && !IS_ERR(page)) {
1595 kunmap(page);
1596 page_cache_release(page);
1597 }
1598 if (name)
1599 putname(name);
1600 if (did_down) {
1601 if (!error)
1602 inode->i_flags |= S_SWAPFILE;
1603 up(&inode->i_sem);
1604 }
1605 return error;
1606}
1607
1608void si_swapinfo(struct sysinfo *val)
1609{
1610 unsigned int i;
1611 unsigned long nr_to_be_unused = 0;
1612
1613 swap_list_lock();
1614 for (i = 0; i < nr_swapfiles; i++) {
1615 if (!(swap_info[i].flags & SWP_USED) ||
1616 (swap_info[i].flags & SWP_WRITEOK))
1617 continue;
1618 nr_to_be_unused += swap_info[i].inuse_pages;
1619 }
1620 val->freeswap = nr_swap_pages + nr_to_be_unused;
1621 val->totalswap = total_swap_pages + nr_to_be_unused;
1622 swap_list_unlock();
1623}
1624
1625/*
1626 * Verify that a swap entry is valid and increment its swap map count.
1627 *
1628 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1629 * "permanent", but will be reclaimed by the next swapoff.
1630 */
1631int swap_duplicate(swp_entry_t entry)
1632{
1633 struct swap_info_struct * p;
1634 unsigned long offset, type;
1635 int result = 0;
1636
1637 type = swp_type(entry);
1638 if (type >= nr_swapfiles)
1639 goto bad_file;
1640 p = type + swap_info;
1641 offset = swp_offset(entry);
1642
1643 swap_device_lock(p);
1644 if (offset < p->max && p->swap_map[offset]) {
1645 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1646 p->swap_map[offset]++;
1647 result = 1;
1648 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1649 if (swap_overflow++ < 5)
1650 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1651 p->swap_map[offset] = SWAP_MAP_MAX;
1652 result = 1;
1653 }
1654 }
1655 swap_device_unlock(p);
1656out:
1657 return result;
1658
1659bad_file:
1660 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1661 goto out;
1662}
1663
1664struct swap_info_struct *
1665get_swap_info_struct(unsigned type)
1666{
1667 return &swap_info[type];
1668}
1669
1670/*
1671 * swap_device_lock prevents swap_map being freed. Don't grab an extra
1672 * reference on the swaphandle, it doesn't matter if it becomes unused.
1673 */
1674int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1675{
1676 int ret = 0, i = 1 << page_cluster;
1677 unsigned long toff;
1678 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1679
1680 if (!page_cluster) /* no readahead */
1681 return 0;
1682 toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1683 if (!toff) /* first page is swap header */
1684 toff++, i--;
1685 *offset = toff;
1686
1687 swap_device_lock(swapdev);
1688 do {
1689 /* Don't read-ahead past the end of the swap area */
1690 if (toff >= swapdev->max)
1691 break;
1692 /* Don't read in free or bad pages */
1693 if (!swapdev->swap_map[toff])
1694 break;
1695 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1696 break;
1697 toff++;
1698 ret++;
1699 } while (--i);
1700 swap_device_unlock(swapdev);
1701 return ret;
1702}