[PATCH] swsusp: userland interface
[linux-3.10.git] / mm / swapfile.c
1 /*
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/mutex.h>
29 #include <linux/capability.h>
30 #include <linux/syscalls.h>
31
32 #include <asm/pgtable.h>
33 #include <asm/tlbflush.h>
34 #include <linux/swapops.h>
35
36 DEFINE_SPINLOCK(swap_lock);
37 unsigned int nr_swapfiles;
38 long total_swap_pages;
39 static int swap_overflow;
40
41 static const char Bad_file[] = "Bad swap file entry ";
42 static const char Unused_file[] = "Unused swap file entry ";
43 static const char Bad_offset[] = "Bad swap offset entry ";
44 static const char Unused_offset[] = "Unused swap offset entry ";
45
46 struct swap_list_t swap_list = {-1, -1};
47
48 static struct swap_info_struct swap_info[MAX_SWAPFILES];
49
50 static DEFINE_MUTEX(swapon_mutex);
51
52 /*
53  * We need this because the bdev->unplug_fn can sleep and we cannot
54  * hold swap_lock while calling the unplug_fn. And swap_lock
55  * cannot be turned into a mutex.
56  */
57 static DECLARE_RWSEM(swap_unplug_sem);
58
59 void 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(page);
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(page) above.
73                  * If 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;
80                 blk_run_backing_dev(bdi, page);
81         }
82         up_read(&swap_unplug_sem);
83 }
84
85 #define SWAPFILE_CLUSTER        256
86 #define LATENCY_LIMIT           256
87
88 static inline unsigned long scan_swap_map(struct swap_info_struct *si)
89 {
90         unsigned long offset, last_in_cluster;
91         int latency_ration = LATENCY_LIMIT;
92
93         /* 
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
103         si->flags += SWP_SCANNING;
104         if (unlikely(!si->cluster_nr)) {
105                 si->cluster_nr = SWAPFILE_CLUSTER - 1;
106                 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER)
107                         goto lowest;
108                 spin_unlock(&swap_lock);
109
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) {
118                                 spin_lock(&swap_lock);
119                                 si->cluster_next = offset-SWAPFILE_CLUSTER+1;
120                                 goto cluster;
121                         }
122                         if (unlikely(--latency_ration < 0)) {
123                                 cond_resched();
124                                 latency_ration = LATENCY_LIMIT;
125                         }
126                 }
127                 spin_lock(&swap_lock);
128                 goto lowest;
129         }
130
131         si->cluster_nr--;
132 cluster:
133         offset = si->cluster_next;
134         if (offset > si->highest_bit)
135 lowest:         offset = si->lowest_bit;
136 checks: if (!(si->flags & SWP_WRITEOK))
137                 goto no_page;
138         if (!si->highest_bit)
139                 goto no_page;
140         if (!si->swap_map[offset]) {
141                 if (offset == si->lowest_bit)
142                         si->lowest_bit++;
143                 if (offset == si->highest_bit)
144                         si->highest_bit--;
145                 si->inuse_pages++;
146                 if (si->inuse_pages == si->pages) {
147                         si->lowest_bit = si->max;
148                         si->highest_bit = 0;
149                 }
150                 si->swap_map[offset] = 1;
151                 si->cluster_next = offset + 1;
152                 si->flags -= SWP_SCANNING;
153                 return offset;
154         }
155
156         spin_unlock(&swap_lock);
157         while (++offset <= si->highest_bit) {
158                 if (!si->swap_map[offset]) {
159                         spin_lock(&swap_lock);
160                         goto checks;
161                 }
162                 if (unlikely(--latency_ration < 0)) {
163                         cond_resched();
164                         latency_ration = LATENCY_LIMIT;
165                 }
166         }
167         spin_lock(&swap_lock);
168         goto lowest;
169
170 no_page:
171         si->flags -= SWP_SCANNING;
172         return 0;
173 }
174
175 swp_entry_t get_swap_page(void)
176 {
177         struct swap_info_struct *si;
178         pgoff_t offset;
179         int type, next;
180         int wrapped = 0;
181
182         spin_lock(&swap_lock);
183         if (nr_swap_pages <= 0)
184                 goto noswap;
185         nr_swap_pages--;
186
187         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++;
194                 }
195
196                 if (!si->highest_bit)
197                         continue;
198                 if (!(si->flags & SWP_WRITEOK))
199                         continue;
200
201                 swap_list.next = next;
202                 offset = scan_swap_map(si);
203                 if (offset) {
204                         spin_unlock(&swap_lock);
205                         return swp_entry(type, offset);
206                 }
207                 next = swap_list.next;
208         }
209
210         nr_swap_pages++;
211 noswap:
212         spin_unlock(&swap_lock);
213         return (swp_entry_t) {0};
214 }
215
216 swp_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
236 static 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;
254         spin_lock(&swap_lock);
255         return p;
256
257 bad_free:
258         printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
259         goto out;
260 bad_offset:
261         printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
262         goto out;
263 bad_device:
264         printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
265         goto out;
266 bad_nofile:
267         printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
268 out:
269         return NULL;
270 }       
271
272 static 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;
284                         if (p->prio > swap_info[swap_list.next].prio)
285                                 swap_list.next = p - swap_info;
286                         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  */
297 void 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));
304                 spin_unlock(&swap_lock);
305         }
306 }
307
308 /*
309  * How many references to page are currently swapped out?
310  */
311 static inline int page_swapcount(struct page *page)
312 {
313         int count = 0;
314         struct swap_info_struct *p;
315         swp_entry_t entry;
316
317         entry.val = page_private(page);
318         p = swap_info_get(entry);
319         if (p) {
320                 /* Subtract the 1 for the swap cache itself */
321                 count = p->swap_map[swp_offset(entry)] - 1;
322                 spin_unlock(&swap_lock);
323         }
324         return count;
325 }
326
327 /*
328  * We can use this swap cache entry directly
329  * if there are no other references to it.
330  */
331 int can_share_swap_page(struct page *page)
332 {
333         int count;
334
335         BUG_ON(!PageLocked(page));
336         count = page_mapcount(page);
337         if (count <= 1 && PageSwapCache(page))
338                 count += page_swapcount(page);
339         return count == 1;
340 }
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  */
346 int 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
362         entry.val = page_private(page);
363         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         }
379         spin_unlock(&swap_lock);
380
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  */
393 void 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) {
400                 if (swap_entry_free(p, swp_offset(entry)) == 1)
401                         page = find_trylock_page(&swapper_space, entry.val);
402                 spin_unlock(&swap_lock);
403         }
404         if (page) {
405                 int one_user;
406
407                 BUG_ON(PagePrivate(page));
408                 page_cache_get(page);
409                 one_user = (page_count(page) == 2);
410                 /* Only cache user (+us), or swap space full? Free it! */
411                 if (!PageWriteback(page) && (one_user || vm_swap_full())) {
412                         delete_from_swap_cache(page);
413                         SetPageDirty(page);
414                 }
415                 unlock_page(page);
416                 page_cache_release(page);
417         }
418 }
419
420 #ifdef CONFIG_SOFTWARE_SUSPEND
421 /*
422  * Find the swap type that corresponds to given device (if any)
423  *
424  * This is needed for software suspend and is done in such a way that inode
425  * aliasing is allowed.
426  */
427 int swap_type_of(dev_t device)
428 {
429         int i;
430
431         spin_lock(&swap_lock);
432         for (i = 0; i < nr_swapfiles; i++) {
433                 struct inode *inode;
434
435                 if (!(swap_info[i].flags & SWP_WRITEOK))
436                         continue;
437                 if (!device) {
438                         spin_unlock(&swap_lock);
439                         return i;
440                 }
441                 inode = swap_info->swap_file->f_dentry->d_inode;
442                 if (S_ISBLK(inode->i_mode) &&
443                     device == MKDEV(imajor(inode), iminor(inode))) {
444                         spin_unlock(&swap_lock);
445                         return i;
446                 }
447         }
448         spin_unlock(&swap_lock);
449         return -ENODEV;
450 }
451
452 /*
453  * Return either the total number of swap pages of given type, or the number
454  * of free pages of that type (depending on @free)
455  *
456  * This is needed for software suspend
457  */
458 unsigned int count_swap_pages(int type, int free)
459 {
460         unsigned int n = 0;
461
462         if (type < nr_swapfiles) {
463                 spin_lock(&swap_lock);
464                 if (swap_info[type].flags & SWP_WRITEOK) {
465                         n = swap_info[type].pages;
466                         if (free)
467                                 n -= swap_info[type].inuse_pages;
468                 }
469                 spin_unlock(&swap_lock);
470         }
471         return n;
472 }
473 #endif
474
475 /*
476  * No need to decide whether this PTE shares the swap entry with others,
477  * just let do_wp_page work it out if a write is requested later - to
478  * force COW, vm_page_prot omits write permission from any private vma.
479  */
480 static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
481                 unsigned long addr, swp_entry_t entry, struct page *page)
482 {
483         inc_mm_counter(vma->vm_mm, anon_rss);
484         get_page(page);
485         set_pte_at(vma->vm_mm, addr, pte,
486                    pte_mkold(mk_pte(page, vma->vm_page_prot)));
487         page_add_anon_rmap(page, vma, addr);
488         swap_free(entry);
489         /*
490          * Move the page to the active list so it is not
491          * immediately swapped out again after swapon.
492          */
493         activate_page(page);
494 }
495
496 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
497                                 unsigned long addr, unsigned long end,
498                                 swp_entry_t entry, struct page *page)
499 {
500         pte_t swp_pte = swp_entry_to_pte(entry);
501         pte_t *pte;
502         spinlock_t *ptl;
503         int found = 0;
504
505         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
506         do {
507                 /*
508                  * swapoff spends a _lot_ of time in this loop!
509                  * Test inline before going to call unuse_pte.
510                  */
511                 if (unlikely(pte_same(*pte, swp_pte))) {
512                         unuse_pte(vma, pte++, addr, entry, page);
513                         found = 1;
514                         break;
515                 }
516         } while (pte++, addr += PAGE_SIZE, addr != end);
517         pte_unmap_unlock(pte - 1, ptl);
518         return found;
519 }
520
521 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
522                                 unsigned long addr, unsigned long end,
523                                 swp_entry_t entry, struct page *page)
524 {
525         pmd_t *pmd;
526         unsigned long next;
527
528         pmd = pmd_offset(pud, addr);
529         do {
530                 next = pmd_addr_end(addr, end);
531                 if (pmd_none_or_clear_bad(pmd))
532                         continue;
533                 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
534                         return 1;
535         } while (pmd++, addr = next, addr != end);
536         return 0;
537 }
538
539 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
540                                 unsigned long addr, unsigned long end,
541                                 swp_entry_t entry, struct page *page)
542 {
543         pud_t *pud;
544         unsigned long next;
545
546         pud = pud_offset(pgd, addr);
547         do {
548                 next = pud_addr_end(addr, end);
549                 if (pud_none_or_clear_bad(pud))
550                         continue;
551                 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
552                         return 1;
553         } while (pud++, addr = next, addr != end);
554         return 0;
555 }
556
557 static int unuse_vma(struct vm_area_struct *vma,
558                                 swp_entry_t entry, struct page *page)
559 {
560         pgd_t *pgd;
561         unsigned long addr, end, next;
562
563         if (page->mapping) {
564                 addr = page_address_in_vma(page, vma);
565                 if (addr == -EFAULT)
566                         return 0;
567                 else
568                         end = addr + PAGE_SIZE;
569         } else {
570                 addr = vma->vm_start;
571                 end = vma->vm_end;
572         }
573
574         pgd = pgd_offset(vma->vm_mm, addr);
575         do {
576                 next = pgd_addr_end(addr, end);
577                 if (pgd_none_or_clear_bad(pgd))
578                         continue;
579                 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
580                         return 1;
581         } while (pgd++, addr = next, addr != end);
582         return 0;
583 }
584
585 static int unuse_mm(struct mm_struct *mm,
586                                 swp_entry_t entry, struct page *page)
587 {
588         struct vm_area_struct *vma;
589
590         if (!down_read_trylock(&mm->mmap_sem)) {
591                 /*
592                  * Activate page so shrink_cache is unlikely to unmap its
593                  * ptes while lock is dropped, so swapoff can make progress.
594                  */
595                 activate_page(page);
596                 unlock_page(page);
597                 down_read(&mm->mmap_sem);
598                 lock_page(page);
599         }
600         for (vma = mm->mmap; vma; vma = vma->vm_next) {
601                 if (vma->anon_vma && unuse_vma(vma, entry, page))
602                         break;
603         }
604         up_read(&mm->mmap_sem);
605         /*
606          * Currently unuse_mm cannot fail, but leave error handling
607          * at call sites for now, since we change it from time to time.
608          */
609         return 0;
610 }
611
612 #ifdef CONFIG_MIGRATION
613 int remove_vma_swap(struct vm_area_struct *vma, struct page *page)
614 {
615         swp_entry_t entry = { .val = page_private(page) };
616
617         return unuse_vma(vma, entry, page);
618 }
619 #endif
620
621 /*
622  * Scan swap_map from current position to next entry still in use.
623  * Recycle to start on reaching the end, returning 0 when empty.
624  */
625 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
626                                         unsigned int prev)
627 {
628         unsigned int max = si->max;
629         unsigned int i = prev;
630         int count;
631
632         /*
633          * No need for swap_lock here: we're just looking
634          * for whether an entry is in use, not modifying it; false
635          * hits are okay, and sys_swapoff() has already prevented new
636          * allocations from this area (while holding swap_lock).
637          */
638         for (;;) {
639                 if (++i >= max) {
640                         if (!prev) {
641                                 i = 0;
642                                 break;
643                         }
644                         /*
645                          * No entries in use at top of swap_map,
646                          * loop back to start and recheck there.
647                          */
648                         max = prev + 1;
649                         prev = 0;
650                         i = 1;
651                 }
652                 count = si->swap_map[i];
653                 if (count && count != SWAP_MAP_BAD)
654                         break;
655         }
656         return i;
657 }
658
659 /*
660  * We completely avoid races by reading each swap page in advance,
661  * and then search for the process using it.  All the necessary
662  * page table adjustments can then be made atomically.
663  */
664 static int try_to_unuse(unsigned int type)
665 {
666         struct swap_info_struct * si = &swap_info[type];
667         struct mm_struct *start_mm;
668         unsigned short *swap_map;
669         unsigned short swcount;
670         struct page *page;
671         swp_entry_t entry;
672         unsigned int i = 0;
673         int retval = 0;
674         int reset_overflow = 0;
675         int shmem;
676
677         /*
678          * When searching mms for an entry, a good strategy is to
679          * start at the first mm we freed the previous entry from
680          * (though actually we don't notice whether we or coincidence
681          * freed the entry).  Initialize this start_mm with a hold.
682          *
683          * A simpler strategy would be to start at the last mm we
684          * freed the previous entry from; but that would take less
685          * advantage of mmlist ordering, which clusters forked mms
686          * together, child after parent.  If we race with dup_mmap(), we
687          * prefer to resolve parent before child, lest we miss entries
688          * duplicated after we scanned child: using last mm would invert
689          * that.  Though it's only a serious concern when an overflowed
690          * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
691          */
692         start_mm = &init_mm;
693         atomic_inc(&init_mm.mm_users);
694
695         /*
696          * Keep on scanning until all entries have gone.  Usually,
697          * one pass through swap_map is enough, but not necessarily:
698          * there are races when an instance of an entry might be missed.
699          */
700         while ((i = find_next_to_unuse(si, i)) != 0) {
701                 if (signal_pending(current)) {
702                         retval = -EINTR;
703                         break;
704                 }
705
706                 /* 
707                  * Get a page for the entry, using the existing swap
708                  * cache page if there is one.  Otherwise, get a clean
709                  * page and read the swap into it. 
710                  */
711                 swap_map = &si->swap_map[i];
712                 entry = swp_entry(type, i);
713 again:
714                 page = read_swap_cache_async(entry, NULL, 0);
715                 if (!page) {
716                         /*
717                          * Either swap_duplicate() failed because entry
718                          * has been freed independently, and will not be
719                          * reused since sys_swapoff() already disabled
720                          * allocation from here, or alloc_page() failed.
721                          */
722                         if (!*swap_map)
723                                 continue;
724                         retval = -ENOMEM;
725                         break;
726                 }
727
728                 /*
729                  * Don't hold on to start_mm if it looks like exiting.
730                  */
731                 if (atomic_read(&start_mm->mm_users) == 1) {
732                         mmput(start_mm);
733                         start_mm = &init_mm;
734                         atomic_inc(&init_mm.mm_users);
735                 }
736
737                 /*
738                  * Wait for and lock page.  When do_swap_page races with
739                  * try_to_unuse, do_swap_page can handle the fault much
740                  * faster than try_to_unuse can locate the entry.  This
741                  * apparently redundant "wait_on_page_locked" lets try_to_unuse
742                  * defer to do_swap_page in such a case - in some tests,
743                  * do_swap_page and try_to_unuse repeatedly compete.
744                  */
745                 wait_on_page_locked(page);
746                 wait_on_page_writeback(page);
747                 lock_page(page);
748                 if (!PageSwapCache(page)) {
749                         /* Page migration has occured */
750                         unlock_page(page);
751                         page_cache_release(page);
752                         goto again;
753                 }
754                 wait_on_page_writeback(page);
755
756                 /*
757                  * Remove all references to entry.
758                  * Whenever we reach init_mm, there's no address space
759                  * to search, but use it as a reminder to search shmem.
760                  */
761                 shmem = 0;
762                 swcount = *swap_map;
763                 if (swcount > 1) {
764                         if (start_mm == &init_mm)
765                                 shmem = shmem_unuse(entry, page);
766                         else
767                                 retval = unuse_mm(start_mm, entry, page);
768                 }
769                 if (*swap_map > 1) {
770                         int set_start_mm = (*swap_map >= swcount);
771                         struct list_head *p = &start_mm->mmlist;
772                         struct mm_struct *new_start_mm = start_mm;
773                         struct mm_struct *prev_mm = start_mm;
774                         struct mm_struct *mm;
775
776                         atomic_inc(&new_start_mm->mm_users);
777                         atomic_inc(&prev_mm->mm_users);
778                         spin_lock(&mmlist_lock);
779                         while (*swap_map > 1 && !retval &&
780                                         (p = p->next) != &start_mm->mmlist) {
781                                 mm = list_entry(p, struct mm_struct, mmlist);
782                                 if (atomic_inc_return(&mm->mm_users) == 1) {
783                                         atomic_dec(&mm->mm_users);
784                                         continue;
785                                 }
786                                 spin_unlock(&mmlist_lock);
787                                 mmput(prev_mm);
788                                 prev_mm = mm;
789
790                                 cond_resched();
791
792                                 swcount = *swap_map;
793                                 if (swcount <= 1)
794                                         ;
795                                 else if (mm == &init_mm) {
796                                         set_start_mm = 1;
797                                         shmem = shmem_unuse(entry, page);
798                                 } else
799                                         retval = unuse_mm(mm, entry, page);
800                                 if (set_start_mm && *swap_map < swcount) {
801                                         mmput(new_start_mm);
802                                         atomic_inc(&mm->mm_users);
803                                         new_start_mm = mm;
804                                         set_start_mm = 0;
805                                 }
806                                 spin_lock(&mmlist_lock);
807                         }
808                         spin_unlock(&mmlist_lock);
809                         mmput(prev_mm);
810                         mmput(start_mm);
811                         start_mm = new_start_mm;
812                 }
813                 if (retval) {
814                         unlock_page(page);
815                         page_cache_release(page);
816                         break;
817                 }
818
819                 /*
820                  * How could swap count reach 0x7fff when the maximum
821                  * pid is 0x7fff, and there's no way to repeat a swap
822                  * page within an mm (except in shmem, where it's the
823                  * shared object which takes the reference count)?
824                  * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
825                  *
826                  * If that's wrong, then we should worry more about
827                  * exit_mmap() and do_munmap() cases described above:
828                  * we might be resetting SWAP_MAP_MAX too early here.
829                  * We know "Undead"s can happen, they're okay, so don't
830                  * report them; but do report if we reset SWAP_MAP_MAX.
831                  */
832                 if (*swap_map == SWAP_MAP_MAX) {
833                         spin_lock(&swap_lock);
834                         *swap_map = 1;
835                         spin_unlock(&swap_lock);
836                         reset_overflow = 1;
837                 }
838
839                 /*
840                  * If a reference remains (rare), we would like to leave
841                  * the page in the swap cache; but try_to_unmap could
842                  * then re-duplicate the entry once we drop page lock,
843                  * so we might loop indefinitely; also, that page could
844                  * not be swapped out to other storage meanwhile.  So:
845                  * delete from cache even if there's another reference,
846                  * after ensuring that the data has been saved to disk -
847                  * since if the reference remains (rarer), it will be
848                  * read from disk into another page.  Splitting into two
849                  * pages would be incorrect if swap supported "shared
850                  * private" pages, but they are handled by tmpfs files.
851                  *
852                  * Note shmem_unuse already deleted a swappage from
853                  * the swap cache, unless the move to filepage failed:
854                  * in which case it left swappage in cache, lowered its
855                  * swap count to pass quickly through the loops above,
856                  * and now we must reincrement count to try again later.
857                  */
858                 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
859                         struct writeback_control wbc = {
860                                 .sync_mode = WB_SYNC_NONE,
861                         };
862
863                         swap_writepage(page, &wbc);
864                         lock_page(page);
865                         wait_on_page_writeback(page);
866                 }
867                 if (PageSwapCache(page)) {
868                         if (shmem)
869                                 swap_duplicate(entry);
870                         else
871                                 delete_from_swap_cache(page);
872                 }
873
874                 /*
875                  * So we could skip searching mms once swap count went
876                  * to 1, we did not mark any present ptes as dirty: must
877                  * mark page dirty so shrink_list will preserve it.
878                  */
879                 SetPageDirty(page);
880                 unlock_page(page);
881                 page_cache_release(page);
882
883                 /*
884                  * Make sure that we aren't completely killing
885                  * interactive performance.
886                  */
887                 cond_resched();
888         }
889
890         mmput(start_mm);
891         if (reset_overflow) {
892                 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
893                 swap_overflow = 0;
894         }
895         return retval;
896 }
897
898 /*
899  * After a successful try_to_unuse, if no swap is now in use, we know
900  * we can empty the mmlist.  swap_lock must be held on entry and exit.
901  * Note that mmlist_lock nests inside swap_lock, and an mm must be
902  * added to the mmlist just after page_duplicate - before would be racy.
903  */
904 static void drain_mmlist(void)
905 {
906         struct list_head *p, *next;
907         unsigned int i;
908
909         for (i = 0; i < nr_swapfiles; i++)
910                 if (swap_info[i].inuse_pages)
911                         return;
912         spin_lock(&mmlist_lock);
913         list_for_each_safe(p, next, &init_mm.mmlist)
914                 list_del_init(p);
915         spin_unlock(&mmlist_lock);
916 }
917
918 /*
919  * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
920  * corresponds to page offset `offset'.
921  */
922 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
923 {
924         struct swap_extent *se = sis->curr_swap_extent;
925         struct swap_extent *start_se = se;
926
927         for ( ; ; ) {
928                 struct list_head *lh;
929
930                 if (se->start_page <= offset &&
931                                 offset < (se->start_page + se->nr_pages)) {
932                         return se->start_block + (offset - se->start_page);
933                 }
934                 lh = se->list.next;
935                 if (lh == &sis->extent_list)
936                         lh = lh->next;
937                 se = list_entry(lh, struct swap_extent, list);
938                 sis->curr_swap_extent = se;
939                 BUG_ON(se == start_se);         /* It *must* be present */
940         }
941 }
942
943 /*
944  * Free all of a swapdev's extent information
945  */
946 static void destroy_swap_extents(struct swap_info_struct *sis)
947 {
948         while (!list_empty(&sis->extent_list)) {
949                 struct swap_extent *se;
950
951                 se = list_entry(sis->extent_list.next,
952                                 struct swap_extent, list);
953                 list_del(&se->list);
954                 kfree(se);
955         }
956 }
957
958 /*
959  * Add a block range (and the corresponding page range) into this swapdev's
960  * extent list.  The extent list is kept sorted in page order.
961  *
962  * This function rather assumes that it is called in ascending page order.
963  */
964 static int
965 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
966                 unsigned long nr_pages, sector_t start_block)
967 {
968         struct swap_extent *se;
969         struct swap_extent *new_se;
970         struct list_head *lh;
971
972         lh = sis->extent_list.prev;     /* The highest page extent */
973         if (lh != &sis->extent_list) {
974                 se = list_entry(lh, struct swap_extent, list);
975                 BUG_ON(se->start_page + se->nr_pages != start_page);
976                 if (se->start_block + se->nr_pages == start_block) {
977                         /* Merge it */
978                         se->nr_pages += nr_pages;
979                         return 0;
980                 }
981         }
982
983         /*
984          * No merge.  Insert a new extent, preserving ordering.
985          */
986         new_se = kmalloc(sizeof(*se), GFP_KERNEL);
987         if (new_se == NULL)
988                 return -ENOMEM;
989         new_se->start_page = start_page;
990         new_se->nr_pages = nr_pages;
991         new_se->start_block = start_block;
992
993         list_add_tail(&new_se->list, &sis->extent_list);
994         return 1;
995 }
996
997 /*
998  * A `swap extent' is a simple thing which maps a contiguous range of pages
999  * onto a contiguous range of disk blocks.  An ordered list of swap extents
1000  * is built at swapon time and is then used at swap_writepage/swap_readpage
1001  * time for locating where on disk a page belongs.
1002  *
1003  * If the swapfile is an S_ISBLK block device, a single extent is installed.
1004  * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1005  * swap files identically.
1006  *
1007  * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1008  * extent list operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
1009  * swapfiles are handled *identically* after swapon time.
1010  *
1011  * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1012  * and will parse them into an ordered extent list, in PAGE_SIZE chunks.  If
1013  * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1014  * requirements, they are simply tossed out - we will never use those blocks
1015  * for swapping.
1016  *
1017  * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon.  This
1018  * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1019  * which will scribble on the fs.
1020  *
1021  * The amount of disk space which a single swap extent represents varies.
1022  * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
1023  * extents in the list.  To avoid much list walking, we cache the previous
1024  * search location in `curr_swap_extent', and start new searches from there.
1025  * This is extremely effective.  The average number of iterations in
1026  * map_swap_page() has been measured at about 0.3 per page.  - akpm.
1027  */
1028 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1029 {
1030         struct inode *inode;
1031         unsigned blocks_per_page;
1032         unsigned long page_no;
1033         unsigned blkbits;
1034         sector_t probe_block;
1035         sector_t last_block;
1036         sector_t lowest_block = -1;
1037         sector_t highest_block = 0;
1038         int nr_extents = 0;
1039         int ret;
1040
1041         inode = sis->swap_file->f_mapping->host;
1042         if (S_ISBLK(inode->i_mode)) {
1043                 ret = add_swap_extent(sis, 0, sis->max, 0);
1044                 *span = sis->pages;
1045                 goto done;
1046         }
1047
1048         blkbits = inode->i_blkbits;
1049         blocks_per_page = PAGE_SIZE >> blkbits;
1050
1051         /*
1052          * Map all the blocks into the extent list.  This code doesn't try
1053          * to be very smart.
1054          */
1055         probe_block = 0;
1056         page_no = 0;
1057         last_block = i_size_read(inode) >> blkbits;
1058         while ((probe_block + blocks_per_page) <= last_block &&
1059                         page_no < sis->max) {
1060                 unsigned block_in_page;
1061                 sector_t first_block;
1062
1063                 first_block = bmap(inode, probe_block);
1064                 if (first_block == 0)
1065                         goto bad_bmap;
1066
1067                 /*
1068                  * It must be PAGE_SIZE aligned on-disk
1069                  */
1070                 if (first_block & (blocks_per_page - 1)) {
1071                         probe_block++;
1072                         goto reprobe;
1073                 }
1074
1075                 for (block_in_page = 1; block_in_page < blocks_per_page;
1076                                         block_in_page++) {
1077                         sector_t block;
1078
1079                         block = bmap(inode, probe_block + block_in_page);
1080                         if (block == 0)
1081                                 goto bad_bmap;
1082                         if (block != first_block + block_in_page) {
1083                                 /* Discontiguity */
1084                                 probe_block++;
1085                                 goto reprobe;
1086                         }
1087                 }
1088
1089                 first_block >>= (PAGE_SHIFT - blkbits);
1090                 if (page_no) {  /* exclude the header page */
1091                         if (first_block < lowest_block)
1092                                 lowest_block = first_block;
1093                         if (first_block > highest_block)
1094                                 highest_block = first_block;
1095                 }
1096
1097                 /*
1098                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1099                  */
1100                 ret = add_swap_extent(sis, page_no, 1, first_block);
1101                 if (ret < 0)
1102                         goto out;
1103                 nr_extents += ret;
1104                 page_no++;
1105                 probe_block += blocks_per_page;
1106 reprobe:
1107                 continue;
1108         }
1109         ret = nr_extents;
1110         *span = 1 + highest_block - lowest_block;
1111         if (page_no == 0)
1112                 page_no = 1;    /* force Empty message */
1113         sis->max = page_no;
1114         sis->pages = page_no - 1;
1115         sis->highest_bit = page_no - 1;
1116 done:
1117         sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1118                                         struct swap_extent, list);
1119         goto out;
1120 bad_bmap:
1121         printk(KERN_ERR "swapon: swapfile has holes\n");
1122         ret = -EINVAL;
1123 out:
1124         return ret;
1125 }
1126
1127 #if 0   /* We don't need this yet */
1128 #include <linux/backing-dev.h>
1129 int page_queue_congested(struct page *page)
1130 {
1131         struct backing_dev_info *bdi;
1132
1133         BUG_ON(!PageLocked(page));      /* It pins the swap_info_struct */
1134
1135         if (PageSwapCache(page)) {
1136                 swp_entry_t entry = { .val = page_private(page) };
1137                 struct swap_info_struct *sis;
1138
1139                 sis = get_swap_info_struct(swp_type(entry));
1140                 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1141         } else
1142                 bdi = page->mapping->backing_dev_info;
1143         return bdi_write_congested(bdi);
1144 }
1145 #endif
1146
1147 asmlinkage long sys_swapoff(const char __user * specialfile)
1148 {
1149         struct swap_info_struct * p = NULL;
1150         unsigned short *swap_map;
1151         struct file *swap_file, *victim;
1152         struct address_space *mapping;
1153         struct inode *inode;
1154         char * pathname;
1155         int i, type, prev;
1156         int err;
1157         
1158         if (!capable(CAP_SYS_ADMIN))
1159                 return -EPERM;
1160
1161         pathname = getname(specialfile);
1162         err = PTR_ERR(pathname);
1163         if (IS_ERR(pathname))
1164                 goto out;
1165
1166         victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1167         putname(pathname);
1168         err = PTR_ERR(victim);
1169         if (IS_ERR(victim))
1170                 goto out;
1171
1172         mapping = victim->f_mapping;
1173         prev = -1;
1174         spin_lock(&swap_lock);
1175         for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1176                 p = swap_info + type;
1177                 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1178                         if (p->swap_file->f_mapping == mapping)
1179                                 break;
1180                 }
1181                 prev = type;
1182         }
1183         if (type < 0) {
1184                 err = -EINVAL;
1185                 spin_unlock(&swap_lock);
1186                 goto out_dput;
1187         }
1188         if (!security_vm_enough_memory(p->pages))
1189                 vm_unacct_memory(p->pages);
1190         else {
1191                 err = -ENOMEM;
1192                 spin_unlock(&swap_lock);
1193                 goto out_dput;
1194         }
1195         if (prev < 0) {
1196                 swap_list.head = p->next;
1197         } else {
1198                 swap_info[prev].next = p->next;
1199         }
1200         if (type == swap_list.next) {
1201                 /* just pick something that's safe... */
1202                 swap_list.next = swap_list.head;
1203         }
1204         nr_swap_pages -= p->pages;
1205         total_swap_pages -= p->pages;
1206         p->flags &= ~SWP_WRITEOK;
1207         spin_unlock(&swap_lock);
1208
1209         current->flags |= PF_SWAPOFF;
1210         err = try_to_unuse(type);
1211         current->flags &= ~PF_SWAPOFF;
1212
1213         if (err) {
1214                 /* re-insert swap space back into swap_list */
1215                 spin_lock(&swap_lock);
1216                 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1217                         if (p->prio >= swap_info[i].prio)
1218                                 break;
1219                 p->next = i;
1220                 if (prev < 0)
1221                         swap_list.head = swap_list.next = p - swap_info;
1222                 else
1223                         swap_info[prev].next = p - swap_info;
1224                 nr_swap_pages += p->pages;
1225                 total_swap_pages += p->pages;
1226                 p->flags |= SWP_WRITEOK;
1227                 spin_unlock(&swap_lock);
1228                 goto out_dput;
1229         }
1230
1231         /* wait for any unplug function to finish */
1232         down_write(&swap_unplug_sem);
1233         up_write(&swap_unplug_sem);
1234
1235         destroy_swap_extents(p);
1236         mutex_lock(&swapon_mutex);
1237         spin_lock(&swap_lock);
1238         drain_mmlist();
1239
1240         /* wait for anyone still in scan_swap_map */
1241         p->highest_bit = 0;             /* cuts scans short */
1242         while (p->flags >= SWP_SCANNING) {
1243                 spin_unlock(&swap_lock);
1244                 schedule_timeout_uninterruptible(1);
1245                 spin_lock(&swap_lock);
1246         }
1247
1248         swap_file = p->swap_file;
1249         p->swap_file = NULL;
1250         p->max = 0;
1251         swap_map = p->swap_map;
1252         p->swap_map = NULL;
1253         p->flags = 0;
1254         spin_unlock(&swap_lock);
1255         mutex_unlock(&swapon_mutex);
1256         vfree(swap_map);
1257         inode = mapping->host;
1258         if (S_ISBLK(inode->i_mode)) {
1259                 struct block_device *bdev = I_BDEV(inode);
1260                 set_blocksize(bdev, p->old_block_size);
1261                 bd_release(bdev);
1262         } else {
1263                 mutex_lock(&inode->i_mutex);
1264                 inode->i_flags &= ~S_SWAPFILE;
1265                 mutex_unlock(&inode->i_mutex);
1266         }
1267         filp_close(swap_file, NULL);
1268         err = 0;
1269
1270 out_dput:
1271         filp_close(victim, NULL);
1272 out:
1273         return err;
1274 }
1275
1276 #ifdef CONFIG_PROC_FS
1277 /* iterator */
1278 static void *swap_start(struct seq_file *swap, loff_t *pos)
1279 {
1280         struct swap_info_struct *ptr = swap_info;
1281         int i;
1282         loff_t l = *pos;
1283
1284         mutex_lock(&swapon_mutex);
1285
1286         for (i = 0; i < nr_swapfiles; i++, ptr++) {
1287                 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1288                         continue;
1289                 if (!l--)
1290                         return ptr;
1291         }
1292
1293         return NULL;
1294 }
1295
1296 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1297 {
1298         struct swap_info_struct *ptr = v;
1299         struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1300
1301         for (++ptr; ptr < endptr; ptr++) {
1302                 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1303                         continue;
1304                 ++*pos;
1305                 return ptr;
1306         }
1307
1308         return NULL;
1309 }
1310
1311 static void swap_stop(struct seq_file *swap, void *v)
1312 {
1313         mutex_unlock(&swapon_mutex);
1314 }
1315
1316 static int swap_show(struct seq_file *swap, void *v)
1317 {
1318         struct swap_info_struct *ptr = v;
1319         struct file *file;
1320         int len;
1321
1322         if (v == swap_info)
1323                 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1324
1325         file = ptr->swap_file;
1326         len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1327         seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1328                        len < 40 ? 40 - len : 1, " ",
1329                        S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1330                                 "partition" : "file\t",
1331                        ptr->pages << (PAGE_SHIFT - 10),
1332                        ptr->inuse_pages << (PAGE_SHIFT - 10),
1333                        ptr->prio);
1334         return 0;
1335 }
1336
1337 static struct seq_operations swaps_op = {
1338         .start =        swap_start,
1339         .next =         swap_next,
1340         .stop =         swap_stop,
1341         .show =         swap_show
1342 };
1343
1344 static int swaps_open(struct inode *inode, struct file *file)
1345 {
1346         return seq_open(file, &swaps_op);
1347 }
1348
1349 static struct file_operations proc_swaps_operations = {
1350         .open           = swaps_open,
1351         .read           = seq_read,
1352         .llseek         = seq_lseek,
1353         .release        = seq_release,
1354 };
1355
1356 static int __init procswaps_init(void)
1357 {
1358         struct proc_dir_entry *entry;
1359
1360         entry = create_proc_entry("swaps", 0, NULL);
1361         if (entry)
1362                 entry->proc_fops = &proc_swaps_operations;
1363         return 0;
1364 }
1365 __initcall(procswaps_init);
1366 #endif /* CONFIG_PROC_FS */
1367
1368 /*
1369  * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1370  *
1371  * The swapon system call
1372  */
1373 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1374 {
1375         struct swap_info_struct * p;
1376         char *name = NULL;
1377         struct block_device *bdev = NULL;
1378         struct file *swap_file = NULL;
1379         struct address_space *mapping;
1380         unsigned int type;
1381         int i, prev;
1382         int error;
1383         static int least_priority;
1384         union swap_header *swap_header = NULL;
1385         int swap_header_version;
1386         unsigned int nr_good_pages = 0;
1387         int nr_extents = 0;
1388         sector_t span;
1389         unsigned long maxpages = 1;
1390         int swapfilesize;
1391         unsigned short *swap_map;
1392         struct page *page = NULL;
1393         struct inode *inode = NULL;
1394         int did_down = 0;
1395
1396         if (!capable(CAP_SYS_ADMIN))
1397                 return -EPERM;
1398         spin_lock(&swap_lock);
1399         p = swap_info;
1400         for (type = 0 ; type < nr_swapfiles ; type++,p++)
1401                 if (!(p->flags & SWP_USED))
1402                         break;
1403         error = -EPERM;
1404         /*
1405          * Test if adding another swap device is possible. There are
1406          * two limiting factors: 1) the number of bits for the swap
1407          * type swp_entry_t definition and 2) the number of bits for
1408          * the swap type in the swap ptes as defined by the different
1409          * architectures. To honor both limitations a swap entry
1410          * with swap offset 0 and swap type ~0UL is created, encoded
1411          * to a swap pte, decoded to a swp_entry_t again and finally
1412          * the swap type part is extracted. This will mask all bits
1413          * from the initial ~0UL that can't be encoded in either the
1414          * swp_entry_t or the architecture definition of a swap pte.
1415          */
1416         if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1417                 spin_unlock(&swap_lock);
1418                 goto out;
1419         }
1420         if (type >= nr_swapfiles)
1421                 nr_swapfiles = type+1;
1422         INIT_LIST_HEAD(&p->extent_list);
1423         p->flags = SWP_USED;
1424         p->swap_file = NULL;
1425         p->old_block_size = 0;
1426         p->swap_map = NULL;
1427         p->lowest_bit = 0;
1428         p->highest_bit = 0;
1429         p->cluster_nr = 0;
1430         p->inuse_pages = 0;
1431         p->next = -1;
1432         if (swap_flags & SWAP_FLAG_PREFER) {
1433                 p->prio =
1434                   (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1435         } else {
1436                 p->prio = --least_priority;
1437         }
1438         spin_unlock(&swap_lock);
1439         name = getname(specialfile);
1440         error = PTR_ERR(name);
1441         if (IS_ERR(name)) {
1442                 name = NULL;
1443                 goto bad_swap_2;
1444         }
1445         swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1446         error = PTR_ERR(swap_file);
1447         if (IS_ERR(swap_file)) {
1448                 swap_file = NULL;
1449                 goto bad_swap_2;
1450         }
1451
1452         p->swap_file = swap_file;
1453         mapping = swap_file->f_mapping;
1454         inode = mapping->host;
1455
1456         error = -EBUSY;
1457         for (i = 0; i < nr_swapfiles; i++) {
1458                 struct swap_info_struct *q = &swap_info[i];
1459
1460                 if (i == type || !q->swap_file)
1461                         continue;
1462                 if (mapping == q->swap_file->f_mapping)
1463                         goto bad_swap;
1464         }
1465
1466         error = -EINVAL;
1467         if (S_ISBLK(inode->i_mode)) {
1468                 bdev = I_BDEV(inode);
1469                 error = bd_claim(bdev, sys_swapon);
1470                 if (error < 0) {
1471                         bdev = NULL;
1472                         error = -EINVAL;
1473                         goto bad_swap;
1474                 }
1475                 p->old_block_size = block_size(bdev);
1476                 error = set_blocksize(bdev, PAGE_SIZE);
1477                 if (error < 0)
1478                         goto bad_swap;
1479                 p->bdev = bdev;
1480         } else if (S_ISREG(inode->i_mode)) {
1481                 p->bdev = inode->i_sb->s_bdev;
1482                 mutex_lock(&inode->i_mutex);
1483                 did_down = 1;
1484                 if (IS_SWAPFILE(inode)) {
1485                         error = -EBUSY;
1486                         goto bad_swap;
1487                 }
1488         } else {
1489                 goto bad_swap;
1490         }
1491
1492         swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1493
1494         /*
1495          * Read the swap header.
1496          */
1497         if (!mapping->a_ops->readpage) {
1498                 error = -EINVAL;
1499                 goto bad_swap;
1500         }
1501         page = read_cache_page(mapping, 0,
1502                         (filler_t *)mapping->a_ops->readpage, swap_file);
1503         if (IS_ERR(page)) {
1504                 error = PTR_ERR(page);
1505                 goto bad_swap;
1506         }
1507         wait_on_page_locked(page);
1508         if (!PageUptodate(page))
1509                 goto bad_swap;
1510         kmap(page);
1511         swap_header = page_address(page);
1512
1513         if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1514                 swap_header_version = 1;
1515         else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1516                 swap_header_version = 2;
1517         else {
1518                 printk(KERN_ERR "Unable to find swap-space signature\n");
1519                 error = -EINVAL;
1520                 goto bad_swap;
1521         }
1522         
1523         switch (swap_header_version) {
1524         case 1:
1525                 printk(KERN_ERR "version 0 swap is no longer supported. "
1526                         "Use mkswap -v1 %s\n", name);
1527                 error = -EINVAL;
1528                 goto bad_swap;
1529         case 2:
1530                 /* Check the swap header's sub-version and the size of
1531                    the swap file and bad block lists */
1532                 if (swap_header->info.version != 1) {
1533                         printk(KERN_WARNING
1534                                "Unable to handle swap header version %d\n",
1535                                swap_header->info.version);
1536                         error = -EINVAL;
1537                         goto bad_swap;
1538                 }
1539
1540                 p->lowest_bit  = 1;
1541                 p->cluster_next = 1;
1542
1543                 /*
1544                  * Find out how many pages are allowed for a single swap
1545                  * device. There are two limiting factors: 1) the number of
1546                  * bits for the swap offset in the swp_entry_t type and
1547                  * 2) the number of bits in the a swap pte as defined by
1548                  * the different architectures. In order to find the
1549                  * largest possible bit mask a swap entry with swap type 0
1550                  * and swap offset ~0UL is created, encoded to a swap pte,
1551                  * decoded to a swp_entry_t again and finally the swap
1552                  * offset is extracted. This will mask all the bits from
1553                  * the initial ~0UL mask that can't be encoded in either
1554                  * the swp_entry_t or the architecture definition of a
1555                  * swap pte.
1556                  */
1557                 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1558                 if (maxpages > swap_header->info.last_page)
1559                         maxpages = swap_header->info.last_page;
1560                 p->highest_bit = maxpages - 1;
1561
1562                 error = -EINVAL;
1563                 if (!maxpages)
1564                         goto bad_swap;
1565                 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1566                         goto bad_swap;
1567                 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1568                         goto bad_swap;
1569
1570                 /* OK, set up the swap map and apply the bad block list */
1571                 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1572                         error = -ENOMEM;
1573                         goto bad_swap;
1574                 }
1575
1576                 error = 0;
1577                 memset(p->swap_map, 0, maxpages * sizeof(short));
1578                 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1579                         int page_nr = swap_header->info.badpages[i];
1580                         if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
1581                                 error = -EINVAL;
1582                         else
1583                                 p->swap_map[page_nr] = SWAP_MAP_BAD;
1584                 }
1585                 nr_good_pages = swap_header->info.last_page -
1586                                 swap_header->info.nr_badpages -
1587                                 1 /* header page */;
1588                 if (error)
1589                         goto bad_swap;
1590         }
1591
1592         if (swapfilesize && maxpages > swapfilesize) {
1593                 printk(KERN_WARNING
1594                        "Swap area shorter than signature indicates\n");
1595                 error = -EINVAL;
1596                 goto bad_swap;
1597         }
1598         if (nr_good_pages) {
1599                 p->swap_map[0] = SWAP_MAP_BAD;
1600                 p->max = maxpages;
1601                 p->pages = nr_good_pages;
1602                 nr_extents = setup_swap_extents(p, &span);
1603                 if (nr_extents < 0) {
1604                         error = nr_extents;
1605                         goto bad_swap;
1606                 }
1607                 nr_good_pages = p->pages;
1608         }
1609         if (!nr_good_pages) {
1610                 printk(KERN_WARNING "Empty swap-file\n");
1611                 error = -EINVAL;
1612                 goto bad_swap;
1613         }
1614
1615         mutex_lock(&swapon_mutex);
1616         spin_lock(&swap_lock);
1617         p->flags = SWP_ACTIVE;
1618         nr_swap_pages += nr_good_pages;
1619         total_swap_pages += nr_good_pages;
1620
1621         printk(KERN_INFO "Adding %uk swap on %s.  "
1622                         "Priority:%d extents:%d across:%lluk\n",
1623                 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1624                 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
1625
1626         /* insert swap space into swap_list: */
1627         prev = -1;
1628         for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1629                 if (p->prio >= swap_info[i].prio) {
1630                         break;
1631                 }
1632                 prev = i;
1633         }
1634         p->next = i;
1635         if (prev < 0) {
1636                 swap_list.head = swap_list.next = p - swap_info;
1637         } else {
1638                 swap_info[prev].next = p - swap_info;
1639         }
1640         spin_unlock(&swap_lock);
1641         mutex_unlock(&swapon_mutex);
1642         error = 0;
1643         goto out;
1644 bad_swap:
1645         if (bdev) {
1646                 set_blocksize(bdev, p->old_block_size);
1647                 bd_release(bdev);
1648         }
1649         destroy_swap_extents(p);
1650 bad_swap_2:
1651         spin_lock(&swap_lock);
1652         swap_map = p->swap_map;
1653         p->swap_file = NULL;
1654         p->swap_map = NULL;
1655         p->flags = 0;
1656         if (!(swap_flags & SWAP_FLAG_PREFER))
1657                 ++least_priority;
1658         spin_unlock(&swap_lock);
1659         vfree(swap_map);
1660         if (swap_file)
1661                 filp_close(swap_file, NULL);
1662 out:
1663         if (page && !IS_ERR(page)) {
1664                 kunmap(page);
1665                 page_cache_release(page);
1666         }
1667         if (name)
1668                 putname(name);
1669         if (did_down) {
1670                 if (!error)
1671                         inode->i_flags |= S_SWAPFILE;
1672                 mutex_unlock(&inode->i_mutex);
1673         }
1674         return error;
1675 }
1676
1677 void si_swapinfo(struct sysinfo *val)
1678 {
1679         unsigned int i;
1680         unsigned long nr_to_be_unused = 0;
1681
1682         spin_lock(&swap_lock);
1683         for (i = 0; i < nr_swapfiles; i++) {
1684                 if (!(swap_info[i].flags & SWP_USED) ||
1685                      (swap_info[i].flags & SWP_WRITEOK))
1686                         continue;
1687                 nr_to_be_unused += swap_info[i].inuse_pages;
1688         }
1689         val->freeswap = nr_swap_pages + nr_to_be_unused;
1690         val->totalswap = total_swap_pages + nr_to_be_unused;
1691         spin_unlock(&swap_lock);
1692 }
1693
1694 /*
1695  * Verify that a swap entry is valid and increment its swap map count.
1696  *
1697  * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1698  * "permanent", but will be reclaimed by the next swapoff.
1699  */
1700 int swap_duplicate(swp_entry_t entry)
1701 {
1702         struct swap_info_struct * p;
1703         unsigned long offset, type;
1704         int result = 0;
1705
1706         type = swp_type(entry);
1707         if (type >= nr_swapfiles)
1708                 goto bad_file;
1709         p = type + swap_info;
1710         offset = swp_offset(entry);
1711
1712         spin_lock(&swap_lock);
1713         if (offset < p->max && p->swap_map[offset]) {
1714                 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1715                         p->swap_map[offset]++;
1716                         result = 1;
1717                 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1718                         if (swap_overflow++ < 5)
1719                                 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1720                         p->swap_map[offset] = SWAP_MAP_MAX;
1721                         result = 1;
1722                 }
1723         }
1724         spin_unlock(&swap_lock);
1725 out:
1726         return result;
1727
1728 bad_file:
1729         printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1730         goto out;
1731 }
1732
1733 struct swap_info_struct *
1734 get_swap_info_struct(unsigned type)
1735 {
1736         return &swap_info[type];
1737 }
1738
1739 /*
1740  * swap_lock prevents swap_map being freed. Don't grab an extra
1741  * reference on the swaphandle, it doesn't matter if it becomes unused.
1742  */
1743 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1744 {
1745         int ret = 0, i = 1 << page_cluster;
1746         unsigned long toff;
1747         struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1748
1749         if (!page_cluster)      /* no readahead */
1750                 return 0;
1751         toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1752         if (!toff)              /* first page is swap header */
1753                 toff++, i--;
1754         *offset = toff;
1755
1756         spin_lock(&swap_lock);
1757         do {
1758                 /* Don't read-ahead past the end of the swap area */
1759                 if (toff >= swapdev->max)
1760                         break;
1761                 /* Don't read in free or bad pages */
1762                 if (!swapdev->swap_map[toff])
1763                         break;
1764                 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1765                         break;
1766                 toff++;
1767                 ret++;
1768         } while (--i);
1769         spin_unlock(&swap_lock);
1770         return ret;
1771 }