[PATCH] swsusp: low level interface
[linux-2.6.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         if (!device)
432                 return -EINVAL;
433         spin_lock(&swap_lock);
434         for (i = 0; i < nr_swapfiles; i++) {
435                 struct inode *inode;
436
437                 if (!(swap_info[i].flags & SWP_WRITEOK))
438                         continue;
439                 inode = swap_info->swap_file->f_dentry->d_inode;
440                 if (S_ISBLK(inode->i_mode) &&
441                     device == MKDEV(imajor(inode), iminor(inode))) {
442                         spin_unlock(&swap_lock);
443                         return i;
444                 }
445         }
446         spin_unlock(&swap_lock);
447         return -ENODEV;
448 }
449
450 /*
451  * Return either the total number of swap pages of given type, or the number
452  * of free pages of that type (depending on @free)
453  *
454  * This is needed for software suspend
455  */
456 unsigned int count_swap_pages(int type, int free)
457 {
458         unsigned int n = 0;
459
460         if (type < nr_swapfiles) {
461                 spin_lock(&swap_lock);
462                 if (swap_info[type].flags & SWP_WRITEOK) {
463                         n = swap_info[type].pages;
464                         if (free)
465                                 n -= swap_info[type].inuse_pages;
466                 }
467                 spin_unlock(&swap_lock);
468         }
469         return n;
470 }
471 #endif
472
473 /*
474  * No need to decide whether this PTE shares the swap entry with others,
475  * just let do_wp_page work it out if a write is requested later - to
476  * force COW, vm_page_prot omits write permission from any private vma.
477  */
478 static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
479                 unsigned long addr, swp_entry_t entry, struct page *page)
480 {
481         inc_mm_counter(vma->vm_mm, anon_rss);
482         get_page(page);
483         set_pte_at(vma->vm_mm, addr, pte,
484                    pte_mkold(mk_pte(page, vma->vm_page_prot)));
485         page_add_anon_rmap(page, vma, addr);
486         swap_free(entry);
487         /*
488          * Move the page to the active list so it is not
489          * immediately swapped out again after swapon.
490          */
491         activate_page(page);
492 }
493
494 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
495                                 unsigned long addr, unsigned long end,
496                                 swp_entry_t entry, struct page *page)
497 {
498         pte_t swp_pte = swp_entry_to_pte(entry);
499         pte_t *pte;
500         spinlock_t *ptl;
501         int found = 0;
502
503         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
504         do {
505                 /*
506                  * swapoff spends a _lot_ of time in this loop!
507                  * Test inline before going to call unuse_pte.
508                  */
509                 if (unlikely(pte_same(*pte, swp_pte))) {
510                         unuse_pte(vma, pte++, addr, entry, page);
511                         found = 1;
512                         break;
513                 }
514         } while (pte++, addr += PAGE_SIZE, addr != end);
515         pte_unmap_unlock(pte - 1, ptl);
516         return found;
517 }
518
519 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
520                                 unsigned long addr, unsigned long end,
521                                 swp_entry_t entry, struct page *page)
522 {
523         pmd_t *pmd;
524         unsigned long next;
525
526         pmd = pmd_offset(pud, addr);
527         do {
528                 next = pmd_addr_end(addr, end);
529                 if (pmd_none_or_clear_bad(pmd))
530                         continue;
531                 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
532                         return 1;
533         } while (pmd++, addr = next, addr != end);
534         return 0;
535 }
536
537 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
538                                 unsigned long addr, unsigned long end,
539                                 swp_entry_t entry, struct page *page)
540 {
541         pud_t *pud;
542         unsigned long next;
543
544         pud = pud_offset(pgd, addr);
545         do {
546                 next = pud_addr_end(addr, end);
547                 if (pud_none_or_clear_bad(pud))
548                         continue;
549                 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
550                         return 1;
551         } while (pud++, addr = next, addr != end);
552         return 0;
553 }
554
555 static int unuse_vma(struct vm_area_struct *vma,
556                                 swp_entry_t entry, struct page *page)
557 {
558         pgd_t *pgd;
559         unsigned long addr, end, next;
560
561         if (page->mapping) {
562                 addr = page_address_in_vma(page, vma);
563                 if (addr == -EFAULT)
564                         return 0;
565                 else
566                         end = addr + PAGE_SIZE;
567         } else {
568                 addr = vma->vm_start;
569                 end = vma->vm_end;
570         }
571
572         pgd = pgd_offset(vma->vm_mm, addr);
573         do {
574                 next = pgd_addr_end(addr, end);
575                 if (pgd_none_or_clear_bad(pgd))
576                         continue;
577                 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
578                         return 1;
579         } while (pgd++, addr = next, addr != end);
580         return 0;
581 }
582
583 static int unuse_mm(struct mm_struct *mm,
584                                 swp_entry_t entry, struct page *page)
585 {
586         struct vm_area_struct *vma;
587
588         if (!down_read_trylock(&mm->mmap_sem)) {
589                 /*
590                  * Activate page so shrink_cache is unlikely to unmap its
591                  * ptes while lock is dropped, so swapoff can make progress.
592                  */
593                 activate_page(page);
594                 unlock_page(page);
595                 down_read(&mm->mmap_sem);
596                 lock_page(page);
597         }
598         for (vma = mm->mmap; vma; vma = vma->vm_next) {
599                 if (vma->anon_vma && unuse_vma(vma, entry, page))
600                         break;
601         }
602         up_read(&mm->mmap_sem);
603         /*
604          * Currently unuse_mm cannot fail, but leave error handling
605          * at call sites for now, since we change it from time to time.
606          */
607         return 0;
608 }
609
610 #ifdef CONFIG_MIGRATION
611 int remove_vma_swap(struct vm_area_struct *vma, struct page *page)
612 {
613         swp_entry_t entry = { .val = page_private(page) };
614
615         return unuse_vma(vma, entry, page);
616 }
617 #endif
618
619 /*
620  * Scan swap_map from current position to next entry still in use.
621  * Recycle to start on reaching the end, returning 0 when empty.
622  */
623 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
624                                         unsigned int prev)
625 {
626         unsigned int max = si->max;
627         unsigned int i = prev;
628         int count;
629
630         /*
631          * No need for swap_lock here: we're just looking
632          * for whether an entry is in use, not modifying it; false
633          * hits are okay, and sys_swapoff() has already prevented new
634          * allocations from this area (while holding swap_lock).
635          */
636         for (;;) {
637                 if (++i >= max) {
638                         if (!prev) {
639                                 i = 0;
640                                 break;
641                         }
642                         /*
643                          * No entries in use at top of swap_map,
644                          * loop back to start and recheck there.
645                          */
646                         max = prev + 1;
647                         prev = 0;
648                         i = 1;
649                 }
650                 count = si->swap_map[i];
651                 if (count && count != SWAP_MAP_BAD)
652                         break;
653         }
654         return i;
655 }
656
657 /*
658  * We completely avoid races by reading each swap page in advance,
659  * and then search for the process using it.  All the necessary
660  * page table adjustments can then be made atomically.
661  */
662 static int try_to_unuse(unsigned int type)
663 {
664         struct swap_info_struct * si = &swap_info[type];
665         struct mm_struct *start_mm;
666         unsigned short *swap_map;
667         unsigned short swcount;
668         struct page *page;
669         swp_entry_t entry;
670         unsigned int i = 0;
671         int retval = 0;
672         int reset_overflow = 0;
673         int shmem;
674
675         /*
676          * When searching mms for an entry, a good strategy is to
677          * start at the first mm we freed the previous entry from
678          * (though actually we don't notice whether we or coincidence
679          * freed the entry).  Initialize this start_mm with a hold.
680          *
681          * A simpler strategy would be to start at the last mm we
682          * freed the previous entry from; but that would take less
683          * advantage of mmlist ordering, which clusters forked mms
684          * together, child after parent.  If we race with dup_mmap(), we
685          * prefer to resolve parent before child, lest we miss entries
686          * duplicated after we scanned child: using last mm would invert
687          * that.  Though it's only a serious concern when an overflowed
688          * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
689          */
690         start_mm = &init_mm;
691         atomic_inc(&init_mm.mm_users);
692
693         /*
694          * Keep on scanning until all entries have gone.  Usually,
695          * one pass through swap_map is enough, but not necessarily:
696          * there are races when an instance of an entry might be missed.
697          */
698         while ((i = find_next_to_unuse(si, i)) != 0) {
699                 if (signal_pending(current)) {
700                         retval = -EINTR;
701                         break;
702                 }
703
704                 /* 
705                  * Get a page for the entry, using the existing swap
706                  * cache page if there is one.  Otherwise, get a clean
707                  * page and read the swap into it. 
708                  */
709                 swap_map = &si->swap_map[i];
710                 entry = swp_entry(type, i);
711 again:
712                 page = read_swap_cache_async(entry, NULL, 0);
713                 if (!page) {
714                         /*
715                          * Either swap_duplicate() failed because entry
716                          * has been freed independently, and will not be
717                          * reused since sys_swapoff() already disabled
718                          * allocation from here, or alloc_page() failed.
719                          */
720                         if (!*swap_map)
721                                 continue;
722                         retval = -ENOMEM;
723                         break;
724                 }
725
726                 /*
727                  * Don't hold on to start_mm if it looks like exiting.
728                  */
729                 if (atomic_read(&start_mm->mm_users) == 1) {
730                         mmput(start_mm);
731                         start_mm = &init_mm;
732                         atomic_inc(&init_mm.mm_users);
733                 }
734
735                 /*
736                  * Wait for and lock page.  When do_swap_page races with
737                  * try_to_unuse, do_swap_page can handle the fault much
738                  * faster than try_to_unuse can locate the entry.  This
739                  * apparently redundant "wait_on_page_locked" lets try_to_unuse
740                  * defer to do_swap_page in such a case - in some tests,
741                  * do_swap_page and try_to_unuse repeatedly compete.
742                  */
743                 wait_on_page_locked(page);
744                 wait_on_page_writeback(page);
745                 lock_page(page);
746                 if (!PageSwapCache(page)) {
747                         /* Page migration has occured */
748                         unlock_page(page);
749                         page_cache_release(page);
750                         goto again;
751                 }
752                 wait_on_page_writeback(page);
753
754                 /*
755                  * Remove all references to entry.
756                  * Whenever we reach init_mm, there's no address space
757                  * to search, but use it as a reminder to search shmem.
758                  */
759                 shmem = 0;
760                 swcount = *swap_map;
761                 if (swcount > 1) {
762                         if (start_mm == &init_mm)
763                                 shmem = shmem_unuse(entry, page);
764                         else
765                                 retval = unuse_mm(start_mm, entry, page);
766                 }
767                 if (*swap_map > 1) {
768                         int set_start_mm = (*swap_map >= swcount);
769                         struct list_head *p = &start_mm->mmlist;
770                         struct mm_struct *new_start_mm = start_mm;
771                         struct mm_struct *prev_mm = start_mm;
772                         struct mm_struct *mm;
773
774                         atomic_inc(&new_start_mm->mm_users);
775                         atomic_inc(&prev_mm->mm_users);
776                         spin_lock(&mmlist_lock);
777                         while (*swap_map > 1 && !retval &&
778                                         (p = p->next) != &start_mm->mmlist) {
779                                 mm = list_entry(p, struct mm_struct, mmlist);
780                                 if (atomic_inc_return(&mm->mm_users) == 1) {
781                                         atomic_dec(&mm->mm_users);
782                                         continue;
783                                 }
784                                 spin_unlock(&mmlist_lock);
785                                 mmput(prev_mm);
786                                 prev_mm = mm;
787
788                                 cond_resched();
789
790                                 swcount = *swap_map;
791                                 if (swcount <= 1)
792                                         ;
793                                 else if (mm == &init_mm) {
794                                         set_start_mm = 1;
795                                         shmem = shmem_unuse(entry, page);
796                                 } else
797                                         retval = unuse_mm(mm, entry, page);
798                                 if (set_start_mm && *swap_map < swcount) {
799                                         mmput(new_start_mm);
800                                         atomic_inc(&mm->mm_users);
801                                         new_start_mm = mm;
802                                         set_start_mm = 0;
803                                 }
804                                 spin_lock(&mmlist_lock);
805                         }
806                         spin_unlock(&mmlist_lock);
807                         mmput(prev_mm);
808                         mmput(start_mm);
809                         start_mm = new_start_mm;
810                 }
811                 if (retval) {
812                         unlock_page(page);
813                         page_cache_release(page);
814                         break;
815                 }
816
817                 /*
818                  * How could swap count reach 0x7fff when the maximum
819                  * pid is 0x7fff, and there's no way to repeat a swap
820                  * page within an mm (except in shmem, where it's the
821                  * shared object which takes the reference count)?
822                  * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
823                  *
824                  * If that's wrong, then we should worry more about
825                  * exit_mmap() and do_munmap() cases described above:
826                  * we might be resetting SWAP_MAP_MAX too early here.
827                  * We know "Undead"s can happen, they're okay, so don't
828                  * report them; but do report if we reset SWAP_MAP_MAX.
829                  */
830                 if (*swap_map == SWAP_MAP_MAX) {
831                         spin_lock(&swap_lock);
832                         *swap_map = 1;
833                         spin_unlock(&swap_lock);
834                         reset_overflow = 1;
835                 }
836
837                 /*
838                  * If a reference remains (rare), we would like to leave
839                  * the page in the swap cache; but try_to_unmap could
840                  * then re-duplicate the entry once we drop page lock,
841                  * so we might loop indefinitely; also, that page could
842                  * not be swapped out to other storage meanwhile.  So:
843                  * delete from cache even if there's another reference,
844                  * after ensuring that the data has been saved to disk -
845                  * since if the reference remains (rarer), it will be
846                  * read from disk into another page.  Splitting into two
847                  * pages would be incorrect if swap supported "shared
848                  * private" pages, but they are handled by tmpfs files.
849                  *
850                  * Note shmem_unuse already deleted a swappage from
851                  * the swap cache, unless the move to filepage failed:
852                  * in which case it left swappage in cache, lowered its
853                  * swap count to pass quickly through the loops above,
854                  * and now we must reincrement count to try again later.
855                  */
856                 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
857                         struct writeback_control wbc = {
858                                 .sync_mode = WB_SYNC_NONE,
859                         };
860
861                         swap_writepage(page, &wbc);
862                         lock_page(page);
863                         wait_on_page_writeback(page);
864                 }
865                 if (PageSwapCache(page)) {
866                         if (shmem)
867                                 swap_duplicate(entry);
868                         else
869                                 delete_from_swap_cache(page);
870                 }
871
872                 /*
873                  * So we could skip searching mms once swap count went
874                  * to 1, we did not mark any present ptes as dirty: must
875                  * mark page dirty so shrink_list will preserve it.
876                  */
877                 SetPageDirty(page);
878                 unlock_page(page);
879                 page_cache_release(page);
880
881                 /*
882                  * Make sure that we aren't completely killing
883                  * interactive performance.
884                  */
885                 cond_resched();
886         }
887
888         mmput(start_mm);
889         if (reset_overflow) {
890                 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
891                 swap_overflow = 0;
892         }
893         return retval;
894 }
895
896 /*
897  * After a successful try_to_unuse, if no swap is now in use, we know
898  * we can empty the mmlist.  swap_lock must be held on entry and exit.
899  * Note that mmlist_lock nests inside swap_lock, and an mm must be
900  * added to the mmlist just after page_duplicate - before would be racy.
901  */
902 static void drain_mmlist(void)
903 {
904         struct list_head *p, *next;
905         unsigned int i;
906
907         for (i = 0; i < nr_swapfiles; i++)
908                 if (swap_info[i].inuse_pages)
909                         return;
910         spin_lock(&mmlist_lock);
911         list_for_each_safe(p, next, &init_mm.mmlist)
912                 list_del_init(p);
913         spin_unlock(&mmlist_lock);
914 }
915
916 /*
917  * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
918  * corresponds to page offset `offset'.
919  */
920 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
921 {
922         struct swap_extent *se = sis->curr_swap_extent;
923         struct swap_extent *start_se = se;
924
925         for ( ; ; ) {
926                 struct list_head *lh;
927
928                 if (se->start_page <= offset &&
929                                 offset < (se->start_page + se->nr_pages)) {
930                         return se->start_block + (offset - se->start_page);
931                 }
932                 lh = se->list.next;
933                 if (lh == &sis->extent_list)
934                         lh = lh->next;
935                 se = list_entry(lh, struct swap_extent, list);
936                 sis->curr_swap_extent = se;
937                 BUG_ON(se == start_se);         /* It *must* be present */
938         }
939 }
940
941 /*
942  * Free all of a swapdev's extent information
943  */
944 static void destroy_swap_extents(struct swap_info_struct *sis)
945 {
946         while (!list_empty(&sis->extent_list)) {
947                 struct swap_extent *se;
948
949                 se = list_entry(sis->extent_list.next,
950                                 struct swap_extent, list);
951                 list_del(&se->list);
952                 kfree(se);
953         }
954 }
955
956 /*
957  * Add a block range (and the corresponding page range) into this swapdev's
958  * extent list.  The extent list is kept sorted in page order.
959  *
960  * This function rather assumes that it is called in ascending page order.
961  */
962 static int
963 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
964                 unsigned long nr_pages, sector_t start_block)
965 {
966         struct swap_extent *se;
967         struct swap_extent *new_se;
968         struct list_head *lh;
969
970         lh = sis->extent_list.prev;     /* The highest page extent */
971         if (lh != &sis->extent_list) {
972                 se = list_entry(lh, struct swap_extent, list);
973                 BUG_ON(se->start_page + se->nr_pages != start_page);
974                 if (se->start_block + se->nr_pages == start_block) {
975                         /* Merge it */
976                         se->nr_pages += nr_pages;
977                         return 0;
978                 }
979         }
980
981         /*
982          * No merge.  Insert a new extent, preserving ordering.
983          */
984         new_se = kmalloc(sizeof(*se), GFP_KERNEL);
985         if (new_se == NULL)
986                 return -ENOMEM;
987         new_se->start_page = start_page;
988         new_se->nr_pages = nr_pages;
989         new_se->start_block = start_block;
990
991         list_add_tail(&new_se->list, &sis->extent_list);
992         return 1;
993 }
994
995 /*
996  * A `swap extent' is a simple thing which maps a contiguous range of pages
997  * onto a contiguous range of disk blocks.  An ordered list of swap extents
998  * is built at swapon time and is then used at swap_writepage/swap_readpage
999  * time for locating where on disk a page belongs.
1000  *
1001  * If the swapfile is an S_ISBLK block device, a single extent is installed.
1002  * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1003  * swap files identically.
1004  *
1005  * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1006  * extent list operates in PAGE_SIZE disk blocks.  Both S_ISREG and S_ISBLK
1007  * swapfiles are handled *identically* after swapon time.
1008  *
1009  * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1010  * and will parse them into an ordered extent list, in PAGE_SIZE chunks.  If
1011  * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1012  * requirements, they are simply tossed out - we will never use those blocks
1013  * for swapping.
1014  *
1015  * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon.  This
1016  * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1017  * which will scribble on the fs.
1018  *
1019  * The amount of disk space which a single swap extent represents varies.
1020  * Typically it is in the 1-4 megabyte range.  So we can have hundreds of
1021  * extents in the list.  To avoid much list walking, we cache the previous
1022  * search location in `curr_swap_extent', and start new searches from there.
1023  * This is extremely effective.  The average number of iterations in
1024  * map_swap_page() has been measured at about 0.3 per page.  - akpm.
1025  */
1026 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1027 {
1028         struct inode *inode;
1029         unsigned blocks_per_page;
1030         unsigned long page_no;
1031         unsigned blkbits;
1032         sector_t probe_block;
1033         sector_t last_block;
1034         sector_t lowest_block = -1;
1035         sector_t highest_block = 0;
1036         int nr_extents = 0;
1037         int ret;
1038
1039         inode = sis->swap_file->f_mapping->host;
1040         if (S_ISBLK(inode->i_mode)) {
1041                 ret = add_swap_extent(sis, 0, sis->max, 0);
1042                 *span = sis->pages;
1043                 goto done;
1044         }
1045
1046         blkbits = inode->i_blkbits;
1047         blocks_per_page = PAGE_SIZE >> blkbits;
1048
1049         /*
1050          * Map all the blocks into the extent list.  This code doesn't try
1051          * to be very smart.
1052          */
1053         probe_block = 0;
1054         page_no = 0;
1055         last_block = i_size_read(inode) >> blkbits;
1056         while ((probe_block + blocks_per_page) <= last_block &&
1057                         page_no < sis->max) {
1058                 unsigned block_in_page;
1059                 sector_t first_block;
1060
1061                 first_block = bmap(inode, probe_block);
1062                 if (first_block == 0)
1063                         goto bad_bmap;
1064
1065                 /*
1066                  * It must be PAGE_SIZE aligned on-disk
1067                  */
1068                 if (first_block & (blocks_per_page - 1)) {
1069                         probe_block++;
1070                         goto reprobe;
1071                 }
1072
1073                 for (block_in_page = 1; block_in_page < blocks_per_page;
1074                                         block_in_page++) {
1075                         sector_t block;
1076
1077                         block = bmap(inode, probe_block + block_in_page);
1078                         if (block == 0)
1079                                 goto bad_bmap;
1080                         if (block != first_block + block_in_page) {
1081                                 /* Discontiguity */
1082                                 probe_block++;
1083                                 goto reprobe;
1084                         }
1085                 }
1086
1087                 first_block >>= (PAGE_SHIFT - blkbits);
1088                 if (page_no) {  /* exclude the header page */
1089                         if (first_block < lowest_block)
1090                                 lowest_block = first_block;
1091                         if (first_block > highest_block)
1092                                 highest_block = first_block;
1093                 }
1094
1095                 /*
1096                  * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1097                  */
1098                 ret = add_swap_extent(sis, page_no, 1, first_block);
1099                 if (ret < 0)
1100                         goto out;
1101                 nr_extents += ret;
1102                 page_no++;
1103                 probe_block += blocks_per_page;
1104 reprobe:
1105                 continue;
1106         }
1107         ret = nr_extents;
1108         *span = 1 + highest_block - lowest_block;
1109         if (page_no == 0)
1110                 page_no = 1;    /* force Empty message */
1111         sis->max = page_no;
1112         sis->pages = page_no - 1;
1113         sis->highest_bit = page_no - 1;
1114 done:
1115         sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1116                                         struct swap_extent, list);
1117         goto out;
1118 bad_bmap:
1119         printk(KERN_ERR "swapon: swapfile has holes\n");
1120         ret = -EINVAL;
1121 out:
1122         return ret;
1123 }
1124
1125 #if 0   /* We don't need this yet */
1126 #include <linux/backing-dev.h>
1127 int page_queue_congested(struct page *page)
1128 {
1129         struct backing_dev_info *bdi;
1130
1131         BUG_ON(!PageLocked(page));      /* It pins the swap_info_struct */
1132
1133         if (PageSwapCache(page)) {
1134                 swp_entry_t entry = { .val = page_private(page) };
1135                 struct swap_info_struct *sis;
1136
1137                 sis = get_swap_info_struct(swp_type(entry));
1138                 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1139         } else
1140                 bdi = page->mapping->backing_dev_info;
1141         return bdi_write_congested(bdi);
1142 }
1143 #endif
1144
1145 asmlinkage long sys_swapoff(const char __user * specialfile)
1146 {
1147         struct swap_info_struct * p = NULL;
1148         unsigned short *swap_map;
1149         struct file *swap_file, *victim;
1150         struct address_space *mapping;
1151         struct inode *inode;
1152         char * pathname;
1153         int i, type, prev;
1154         int err;
1155         
1156         if (!capable(CAP_SYS_ADMIN))
1157                 return -EPERM;
1158
1159         pathname = getname(specialfile);
1160         err = PTR_ERR(pathname);
1161         if (IS_ERR(pathname))
1162                 goto out;
1163
1164         victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1165         putname(pathname);
1166         err = PTR_ERR(victim);
1167         if (IS_ERR(victim))
1168                 goto out;
1169
1170         mapping = victim->f_mapping;
1171         prev = -1;
1172         spin_lock(&swap_lock);
1173         for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1174                 p = swap_info + type;
1175                 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1176                         if (p->swap_file->f_mapping == mapping)
1177                                 break;
1178                 }
1179                 prev = type;
1180         }
1181         if (type < 0) {
1182                 err = -EINVAL;
1183                 spin_unlock(&swap_lock);
1184                 goto out_dput;
1185         }
1186         if (!security_vm_enough_memory(p->pages))
1187                 vm_unacct_memory(p->pages);
1188         else {
1189                 err = -ENOMEM;
1190                 spin_unlock(&swap_lock);
1191                 goto out_dput;
1192         }
1193         if (prev < 0) {
1194                 swap_list.head = p->next;
1195         } else {
1196                 swap_info[prev].next = p->next;
1197         }
1198         if (type == swap_list.next) {
1199                 /* just pick something that's safe... */
1200                 swap_list.next = swap_list.head;
1201         }
1202         nr_swap_pages -= p->pages;
1203         total_swap_pages -= p->pages;
1204         p->flags &= ~SWP_WRITEOK;
1205         spin_unlock(&swap_lock);
1206
1207         current->flags |= PF_SWAPOFF;
1208         err = try_to_unuse(type);
1209         current->flags &= ~PF_SWAPOFF;
1210
1211         if (err) {
1212                 /* re-insert swap space back into swap_list */
1213                 spin_lock(&swap_lock);
1214                 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1215                         if (p->prio >= swap_info[i].prio)
1216                                 break;
1217                 p->next = i;
1218                 if (prev < 0)
1219                         swap_list.head = swap_list.next = p - swap_info;
1220                 else
1221                         swap_info[prev].next = p - swap_info;
1222                 nr_swap_pages += p->pages;
1223                 total_swap_pages += p->pages;
1224                 p->flags |= SWP_WRITEOK;
1225                 spin_unlock(&swap_lock);
1226                 goto out_dput;
1227         }
1228
1229         /* wait for any unplug function to finish */
1230         down_write(&swap_unplug_sem);
1231         up_write(&swap_unplug_sem);
1232
1233         destroy_swap_extents(p);
1234         mutex_lock(&swapon_mutex);
1235         spin_lock(&swap_lock);
1236         drain_mmlist();
1237
1238         /* wait for anyone still in scan_swap_map */
1239         p->highest_bit = 0;             /* cuts scans short */
1240         while (p->flags >= SWP_SCANNING) {
1241                 spin_unlock(&swap_lock);
1242                 schedule_timeout_uninterruptible(1);
1243                 spin_lock(&swap_lock);
1244         }
1245
1246         swap_file = p->swap_file;
1247         p->swap_file = NULL;
1248         p->max = 0;
1249         swap_map = p->swap_map;
1250         p->swap_map = NULL;
1251         p->flags = 0;
1252         spin_unlock(&swap_lock);
1253         mutex_unlock(&swapon_mutex);
1254         vfree(swap_map);
1255         inode = mapping->host;
1256         if (S_ISBLK(inode->i_mode)) {
1257                 struct block_device *bdev = I_BDEV(inode);
1258                 set_blocksize(bdev, p->old_block_size);
1259                 bd_release(bdev);
1260         } else {
1261                 mutex_lock(&inode->i_mutex);
1262                 inode->i_flags &= ~S_SWAPFILE;
1263                 mutex_unlock(&inode->i_mutex);
1264         }
1265         filp_close(swap_file, NULL);
1266         err = 0;
1267
1268 out_dput:
1269         filp_close(victim, NULL);
1270 out:
1271         return err;
1272 }
1273
1274 #ifdef CONFIG_PROC_FS
1275 /* iterator */
1276 static void *swap_start(struct seq_file *swap, loff_t *pos)
1277 {
1278         struct swap_info_struct *ptr = swap_info;
1279         int i;
1280         loff_t l = *pos;
1281
1282         mutex_lock(&swapon_mutex);
1283
1284         for (i = 0; i < nr_swapfiles; i++, ptr++) {
1285                 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1286                         continue;
1287                 if (!l--)
1288                         return ptr;
1289         }
1290
1291         return NULL;
1292 }
1293
1294 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1295 {
1296         struct swap_info_struct *ptr = v;
1297         struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1298
1299         for (++ptr; ptr < endptr; ptr++) {
1300                 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1301                         continue;
1302                 ++*pos;
1303                 return ptr;
1304         }
1305
1306         return NULL;
1307 }
1308
1309 static void swap_stop(struct seq_file *swap, void *v)
1310 {
1311         mutex_unlock(&swapon_mutex);
1312 }
1313
1314 static int swap_show(struct seq_file *swap, void *v)
1315 {
1316         struct swap_info_struct *ptr = v;
1317         struct file *file;
1318         int len;
1319
1320         if (v == swap_info)
1321                 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1322
1323         file = ptr->swap_file;
1324         len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1325         seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1326                        len < 40 ? 40 - len : 1, " ",
1327                        S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1328                                 "partition" : "file\t",
1329                        ptr->pages << (PAGE_SHIFT - 10),
1330                        ptr->inuse_pages << (PAGE_SHIFT - 10),
1331                        ptr->prio);
1332         return 0;
1333 }
1334
1335 static struct seq_operations swaps_op = {
1336         .start =        swap_start,
1337         .next =         swap_next,
1338         .stop =         swap_stop,
1339         .show =         swap_show
1340 };
1341
1342 static int swaps_open(struct inode *inode, struct file *file)
1343 {
1344         return seq_open(file, &swaps_op);
1345 }
1346
1347 static struct file_operations proc_swaps_operations = {
1348         .open           = swaps_open,
1349         .read           = seq_read,
1350         .llseek         = seq_lseek,
1351         .release        = seq_release,
1352 };
1353
1354 static int __init procswaps_init(void)
1355 {
1356         struct proc_dir_entry *entry;
1357
1358         entry = create_proc_entry("swaps", 0, NULL);
1359         if (entry)
1360                 entry->proc_fops = &proc_swaps_operations;
1361         return 0;
1362 }
1363 __initcall(procswaps_init);
1364 #endif /* CONFIG_PROC_FS */
1365
1366 /*
1367  * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1368  *
1369  * The swapon system call
1370  */
1371 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1372 {
1373         struct swap_info_struct * p;
1374         char *name = NULL;
1375         struct block_device *bdev = NULL;
1376         struct file *swap_file = NULL;
1377         struct address_space *mapping;
1378         unsigned int type;
1379         int i, prev;
1380         int error;
1381         static int least_priority;
1382         union swap_header *swap_header = NULL;
1383         int swap_header_version;
1384         unsigned int nr_good_pages = 0;
1385         int nr_extents = 0;
1386         sector_t span;
1387         unsigned long maxpages = 1;
1388         int swapfilesize;
1389         unsigned short *swap_map;
1390         struct page *page = NULL;
1391         struct inode *inode = NULL;
1392         int did_down = 0;
1393
1394         if (!capable(CAP_SYS_ADMIN))
1395                 return -EPERM;
1396         spin_lock(&swap_lock);
1397         p = swap_info;
1398         for (type = 0 ; type < nr_swapfiles ; type++,p++)
1399                 if (!(p->flags & SWP_USED))
1400                         break;
1401         error = -EPERM;
1402         /*
1403          * Test if adding another swap device is possible. There are
1404          * two limiting factors: 1) the number of bits for the swap
1405          * type swp_entry_t definition and 2) the number of bits for
1406          * the swap type in the swap ptes as defined by the different
1407          * architectures. To honor both limitations a swap entry
1408          * with swap offset 0 and swap type ~0UL is created, encoded
1409          * to a swap pte, decoded to a swp_entry_t again and finally
1410          * the swap type part is extracted. This will mask all bits
1411          * from the initial ~0UL that can't be encoded in either the
1412          * swp_entry_t or the architecture definition of a swap pte.
1413          */
1414         if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1415                 spin_unlock(&swap_lock);
1416                 goto out;
1417         }
1418         if (type >= nr_swapfiles)
1419                 nr_swapfiles = type+1;
1420         INIT_LIST_HEAD(&p->extent_list);
1421         p->flags = SWP_USED;
1422         p->swap_file = NULL;
1423         p->old_block_size = 0;
1424         p->swap_map = NULL;
1425         p->lowest_bit = 0;
1426         p->highest_bit = 0;
1427         p->cluster_nr = 0;
1428         p->inuse_pages = 0;
1429         p->next = -1;
1430         if (swap_flags & SWAP_FLAG_PREFER) {
1431                 p->prio =
1432                   (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1433         } else {
1434                 p->prio = --least_priority;
1435         }
1436         spin_unlock(&swap_lock);
1437         name = getname(specialfile);
1438         error = PTR_ERR(name);
1439         if (IS_ERR(name)) {
1440                 name = NULL;
1441                 goto bad_swap_2;
1442         }
1443         swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1444         error = PTR_ERR(swap_file);
1445         if (IS_ERR(swap_file)) {
1446                 swap_file = NULL;
1447                 goto bad_swap_2;
1448         }
1449
1450         p->swap_file = swap_file;
1451         mapping = swap_file->f_mapping;
1452         inode = mapping->host;
1453
1454         error = -EBUSY;
1455         for (i = 0; i < nr_swapfiles; i++) {
1456                 struct swap_info_struct *q = &swap_info[i];
1457
1458                 if (i == type || !q->swap_file)
1459                         continue;
1460                 if (mapping == q->swap_file->f_mapping)
1461                         goto bad_swap;
1462         }
1463
1464         error = -EINVAL;
1465         if (S_ISBLK(inode->i_mode)) {
1466                 bdev = I_BDEV(inode);
1467                 error = bd_claim(bdev, sys_swapon);
1468                 if (error < 0) {
1469                         bdev = NULL;
1470                         error = -EINVAL;
1471                         goto bad_swap;
1472                 }
1473                 p->old_block_size = block_size(bdev);
1474                 error = set_blocksize(bdev, PAGE_SIZE);
1475                 if (error < 0)
1476                         goto bad_swap;
1477                 p->bdev = bdev;
1478         } else if (S_ISREG(inode->i_mode)) {
1479                 p->bdev = inode->i_sb->s_bdev;
1480                 mutex_lock(&inode->i_mutex);
1481                 did_down = 1;
1482                 if (IS_SWAPFILE(inode)) {
1483                         error = -EBUSY;
1484                         goto bad_swap;
1485                 }
1486         } else {
1487                 goto bad_swap;
1488         }
1489
1490         swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1491
1492         /*
1493          * Read the swap header.
1494          */
1495         if (!mapping->a_ops->readpage) {
1496                 error = -EINVAL;
1497                 goto bad_swap;
1498         }
1499         page = read_cache_page(mapping, 0,
1500                         (filler_t *)mapping->a_ops->readpage, swap_file);
1501         if (IS_ERR(page)) {
1502                 error = PTR_ERR(page);
1503                 goto bad_swap;
1504         }
1505         wait_on_page_locked(page);
1506         if (!PageUptodate(page))
1507                 goto bad_swap;
1508         kmap(page);
1509         swap_header = page_address(page);
1510
1511         if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1512                 swap_header_version = 1;
1513         else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1514                 swap_header_version = 2;
1515         else {
1516                 printk(KERN_ERR "Unable to find swap-space signature\n");
1517                 error = -EINVAL;
1518                 goto bad_swap;
1519         }
1520         
1521         switch (swap_header_version) {
1522         case 1:
1523                 printk(KERN_ERR "version 0 swap is no longer supported. "
1524                         "Use mkswap -v1 %s\n", name);
1525                 error = -EINVAL;
1526                 goto bad_swap;
1527         case 2:
1528                 /* Check the swap header's sub-version and the size of
1529                    the swap file and bad block lists */
1530                 if (swap_header->info.version != 1) {
1531                         printk(KERN_WARNING
1532                                "Unable to handle swap header version %d\n",
1533                                swap_header->info.version);
1534                         error = -EINVAL;
1535                         goto bad_swap;
1536                 }
1537
1538                 p->lowest_bit  = 1;
1539                 p->cluster_next = 1;
1540
1541                 /*
1542                  * Find out how many pages are allowed for a single swap
1543                  * device. There are two limiting factors: 1) the number of
1544                  * bits for the swap offset in the swp_entry_t type and
1545                  * 2) the number of bits in the a swap pte as defined by
1546                  * the different architectures. In order to find the
1547                  * largest possible bit mask a swap entry with swap type 0
1548                  * and swap offset ~0UL is created, encoded to a swap pte,
1549                  * decoded to a swp_entry_t again and finally the swap
1550                  * offset is extracted. This will mask all the bits from
1551                  * the initial ~0UL mask that can't be encoded in either
1552                  * the swp_entry_t or the architecture definition of a
1553                  * swap pte.
1554                  */
1555                 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1556                 if (maxpages > swap_header->info.last_page)
1557                         maxpages = swap_header->info.last_page;
1558                 p->highest_bit = maxpages - 1;
1559
1560                 error = -EINVAL;
1561                 if (!maxpages)
1562                         goto bad_swap;
1563                 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1564                         goto bad_swap;
1565                 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1566                         goto bad_swap;
1567
1568                 /* OK, set up the swap map and apply the bad block list */
1569                 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1570                         error = -ENOMEM;
1571                         goto bad_swap;
1572                 }
1573
1574                 error = 0;
1575                 memset(p->swap_map, 0, maxpages * sizeof(short));
1576                 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1577                         int page_nr = swap_header->info.badpages[i];
1578                         if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
1579                                 error = -EINVAL;
1580                         else
1581                                 p->swap_map[page_nr] = SWAP_MAP_BAD;
1582                 }
1583                 nr_good_pages = swap_header->info.last_page -
1584                                 swap_header->info.nr_badpages -
1585                                 1 /* header page */;
1586                 if (error)
1587                         goto bad_swap;
1588         }
1589
1590         if (swapfilesize && maxpages > swapfilesize) {
1591                 printk(KERN_WARNING
1592                        "Swap area shorter than signature indicates\n");
1593                 error = -EINVAL;
1594                 goto bad_swap;
1595         }
1596         if (nr_good_pages) {
1597                 p->swap_map[0] = SWAP_MAP_BAD;
1598                 p->max = maxpages;
1599                 p->pages = nr_good_pages;
1600                 nr_extents = setup_swap_extents(p, &span);
1601                 if (nr_extents < 0) {
1602                         error = nr_extents;
1603                         goto bad_swap;
1604                 }
1605                 nr_good_pages = p->pages;
1606         }
1607         if (!nr_good_pages) {
1608                 printk(KERN_WARNING "Empty swap-file\n");
1609                 error = -EINVAL;
1610                 goto bad_swap;
1611         }
1612
1613         mutex_lock(&swapon_mutex);
1614         spin_lock(&swap_lock);
1615         p->flags = SWP_ACTIVE;
1616         nr_swap_pages += nr_good_pages;
1617         total_swap_pages += nr_good_pages;
1618
1619         printk(KERN_INFO "Adding %uk swap on %s.  "
1620                         "Priority:%d extents:%d across:%lluk\n",
1621                 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1622                 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
1623
1624         /* insert swap space into swap_list: */
1625         prev = -1;
1626         for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1627                 if (p->prio >= swap_info[i].prio) {
1628                         break;
1629                 }
1630                 prev = i;
1631         }
1632         p->next = i;
1633         if (prev < 0) {
1634                 swap_list.head = swap_list.next = p - swap_info;
1635         } else {
1636                 swap_info[prev].next = p - swap_info;
1637         }
1638         spin_unlock(&swap_lock);
1639         mutex_unlock(&swapon_mutex);
1640         error = 0;
1641         goto out;
1642 bad_swap:
1643         if (bdev) {
1644                 set_blocksize(bdev, p->old_block_size);
1645                 bd_release(bdev);
1646         }
1647         destroy_swap_extents(p);
1648 bad_swap_2:
1649         spin_lock(&swap_lock);
1650         swap_map = p->swap_map;
1651         p->swap_file = NULL;
1652         p->swap_map = NULL;
1653         p->flags = 0;
1654         if (!(swap_flags & SWAP_FLAG_PREFER))
1655                 ++least_priority;
1656         spin_unlock(&swap_lock);
1657         vfree(swap_map);
1658         if (swap_file)
1659                 filp_close(swap_file, NULL);
1660 out:
1661         if (page && !IS_ERR(page)) {
1662                 kunmap(page);
1663                 page_cache_release(page);
1664         }
1665         if (name)
1666                 putname(name);
1667         if (did_down) {
1668                 if (!error)
1669                         inode->i_flags |= S_SWAPFILE;
1670                 mutex_unlock(&inode->i_mutex);
1671         }
1672         return error;
1673 }
1674
1675 void si_swapinfo(struct sysinfo *val)
1676 {
1677         unsigned int i;
1678         unsigned long nr_to_be_unused = 0;
1679
1680         spin_lock(&swap_lock);
1681         for (i = 0; i < nr_swapfiles; i++) {
1682                 if (!(swap_info[i].flags & SWP_USED) ||
1683                      (swap_info[i].flags & SWP_WRITEOK))
1684                         continue;
1685                 nr_to_be_unused += swap_info[i].inuse_pages;
1686         }
1687         val->freeswap = nr_swap_pages + nr_to_be_unused;
1688         val->totalswap = total_swap_pages + nr_to_be_unused;
1689         spin_unlock(&swap_lock);
1690 }
1691
1692 /*
1693  * Verify that a swap entry is valid and increment its swap map count.
1694  *
1695  * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1696  * "permanent", but will be reclaimed by the next swapoff.
1697  */
1698 int swap_duplicate(swp_entry_t entry)
1699 {
1700         struct swap_info_struct * p;
1701         unsigned long offset, type;
1702         int result = 0;
1703
1704         type = swp_type(entry);
1705         if (type >= nr_swapfiles)
1706                 goto bad_file;
1707         p = type + swap_info;
1708         offset = swp_offset(entry);
1709
1710         spin_lock(&swap_lock);
1711         if (offset < p->max && p->swap_map[offset]) {
1712                 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1713                         p->swap_map[offset]++;
1714                         result = 1;
1715                 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1716                         if (swap_overflow++ < 5)
1717                                 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1718                         p->swap_map[offset] = SWAP_MAP_MAX;
1719                         result = 1;
1720                 }
1721         }
1722         spin_unlock(&swap_lock);
1723 out:
1724         return result;
1725
1726 bad_file:
1727         printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1728         goto out;
1729 }
1730
1731 struct swap_info_struct *
1732 get_swap_info_struct(unsigned type)
1733 {
1734         return &swap_info[type];
1735 }
1736
1737 /*
1738  * swap_lock prevents swap_map being freed. Don't grab an extra
1739  * reference on the swaphandle, it doesn't matter if it becomes unused.
1740  */
1741 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1742 {
1743         int ret = 0, i = 1 << page_cluster;
1744         unsigned long toff;
1745         struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1746
1747         if (!page_cluster)      /* no readahead */
1748                 return 0;
1749         toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1750         if (!toff)              /* first page is swap header */
1751                 toff++, i--;
1752         *offset = toff;
1753
1754         spin_lock(&swap_lock);
1755         do {
1756                 /* Don't read-ahead past the end of the swap area */
1757                 if (toff >= swapdev->max)
1758                         break;
1759                 /* Don't read in free or bad pages */
1760                 if (!swapdev->swap_map[toff])
1761                         break;
1762                 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1763                         break;
1764                 toff++;
1765                 ret++;
1766         } while (--i);
1767         spin_unlock(&swap_lock);
1768         return ret;
1769 }