4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
9 #include <linux/hugetlb.h>
10 #include <linux/mman.h>
11 #include <linux/slab.h>
12 #include <linux/kernel_stat.h>
13 #include <linux/swap.h>
14 #include <linux/vmalloc.h>
15 #include <linux/pagemap.h>
16 #include <linux/namei.h>
17 #include <linux/shm.h>
18 #include <linux/blkdev.h>
19 #include <linux/writeback.h>
20 #include <linux/proc_fs.h>
21 #include <linux/seq_file.h>
22 #include <linux/init.h>
23 #include <linux/module.h>
24 #include <linux/rmap.h>
25 #include <linux/security.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mutex.h>
28 #include <linux/capability.h>
29 #include <linux/syscalls.h>
31 #include <asm/pgtable.h>
32 #include <asm/tlbflush.h>
33 #include <linux/swapops.h>
35 DEFINE_SPINLOCK(swap_lock);
36 unsigned int nr_swapfiles;
37 long total_swap_pages;
38 static int swap_overflow;
40 static const char Bad_file[] = "Bad swap file entry ";
41 static const char Unused_file[] = "Unused swap file entry ";
42 static const char Bad_offset[] = "Bad swap offset entry ";
43 static const char Unused_offset[] = "Unused swap offset entry ";
45 struct swap_list_t swap_list = {-1, -1};
47 static struct swap_info_struct swap_info[MAX_SWAPFILES];
49 static DEFINE_MUTEX(swapon_mutex);
52 * We need this because the bdev->unplug_fn can sleep and we cannot
53 * hold swap_lock while calling the unplug_fn. And swap_lock
54 * cannot be turned into a mutex.
56 static DECLARE_RWSEM(swap_unplug_sem);
58 void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
62 down_read(&swap_unplug_sem);
63 entry.val = page_private(page);
64 if (PageSwapCache(page)) {
65 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
66 struct backing_dev_info *bdi;
69 * If the page is removed from swapcache from under us (with a
70 * racy try_to_unuse/swapoff) we need an additional reference
71 * count to avoid reading garbage from page_private(page) above.
72 * If the WARN_ON triggers during a swapoff it maybe the race
73 * condition and it's harmless. However if it triggers without
74 * swapoff it signals a problem.
76 WARN_ON(page_count(page) <= 1);
78 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
79 blk_run_backing_dev(bdi, page);
81 up_read(&swap_unplug_sem);
84 #define SWAPFILE_CLUSTER 256
85 #define LATENCY_LIMIT 256
87 static inline unsigned long scan_swap_map(struct swap_info_struct *si)
89 unsigned long offset, last_in_cluster;
90 int latency_ration = LATENCY_LIMIT;
93 * We try to cluster swap pages by allocating them sequentially
94 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
95 * way, however, we resort to first-free allocation, starting
96 * a new cluster. This prevents us from scattering swap pages
97 * all over the entire swap partition, so that we reduce
98 * overall disk seek times between swap pages. -- sct
99 * But we do now try to find an empty cluster. -Andrea
102 si->flags += SWP_SCANNING;
103 if (unlikely(!si->cluster_nr)) {
104 si->cluster_nr = SWAPFILE_CLUSTER - 1;
105 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER)
107 spin_unlock(&swap_lock);
109 offset = si->lowest_bit;
110 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
112 /* Locate the first empty (unaligned) cluster */
113 for (; last_in_cluster <= si->highest_bit; offset++) {
114 if (si->swap_map[offset])
115 last_in_cluster = offset + SWAPFILE_CLUSTER;
116 else if (offset == last_in_cluster) {
117 spin_lock(&swap_lock);
118 si->cluster_next = offset-SWAPFILE_CLUSTER+1;
121 if (unlikely(--latency_ration < 0)) {
123 latency_ration = LATENCY_LIMIT;
126 spin_lock(&swap_lock);
132 offset = si->cluster_next;
133 if (offset > si->highest_bit)
134 lowest: offset = si->lowest_bit;
135 checks: if (!(si->flags & SWP_WRITEOK))
137 if (!si->highest_bit)
139 if (!si->swap_map[offset]) {
140 if (offset == si->lowest_bit)
142 if (offset == si->highest_bit)
145 if (si->inuse_pages == si->pages) {
146 si->lowest_bit = si->max;
149 si->swap_map[offset] = 1;
150 si->cluster_next = offset + 1;
151 si->flags -= SWP_SCANNING;
155 spin_unlock(&swap_lock);
156 while (++offset <= si->highest_bit) {
157 if (!si->swap_map[offset]) {
158 spin_lock(&swap_lock);
161 if (unlikely(--latency_ration < 0)) {
163 latency_ration = LATENCY_LIMIT;
166 spin_lock(&swap_lock);
170 si->flags -= SWP_SCANNING;
174 swp_entry_t get_swap_page(void)
176 struct swap_info_struct *si;
181 spin_lock(&swap_lock);
182 if (nr_swap_pages <= 0)
186 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
187 si = swap_info + type;
190 (!wrapped && si->prio != swap_info[next].prio)) {
191 next = swap_list.head;
195 if (!si->highest_bit)
197 if (!(si->flags & SWP_WRITEOK))
200 swap_list.next = next;
201 offset = scan_swap_map(si);
203 spin_unlock(&swap_lock);
204 return swp_entry(type, offset);
206 next = swap_list.next;
211 spin_unlock(&swap_lock);
212 return (swp_entry_t) {0};
215 swp_entry_t get_swap_page_of_type(int type)
217 struct swap_info_struct *si;
220 spin_lock(&swap_lock);
221 si = swap_info + type;
222 if (si->flags & SWP_WRITEOK) {
224 offset = scan_swap_map(si);
226 spin_unlock(&swap_lock);
227 return swp_entry(type, offset);
231 spin_unlock(&swap_lock);
232 return (swp_entry_t) {0};
235 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
237 struct swap_info_struct * p;
238 unsigned long offset, type;
242 type = swp_type(entry);
243 if (type >= nr_swapfiles)
245 p = & swap_info[type];
246 if (!(p->flags & SWP_USED))
248 offset = swp_offset(entry);
249 if (offset >= p->max)
251 if (!p->swap_map[offset])
253 spin_lock(&swap_lock);
257 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
260 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
263 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
266 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
271 static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
273 int count = p->swap_map[offset];
275 if (count < SWAP_MAP_MAX) {
277 p->swap_map[offset] = count;
279 if (offset < p->lowest_bit)
280 p->lowest_bit = offset;
281 if (offset > p->highest_bit)
282 p->highest_bit = offset;
283 if (p->prio > swap_info[swap_list.next].prio)
284 swap_list.next = p - swap_info;
293 * Caller has made sure that the swapdevice corresponding to entry
294 * is still around or has not been recycled.
296 void swap_free(swp_entry_t entry)
298 struct swap_info_struct * p;
300 p = swap_info_get(entry);
302 swap_entry_free(p, swp_offset(entry));
303 spin_unlock(&swap_lock);
308 * How many references to page are currently swapped out?
310 static inline int page_swapcount(struct page *page)
313 struct swap_info_struct *p;
316 entry.val = page_private(page);
317 p = swap_info_get(entry);
319 /* Subtract the 1 for the swap cache itself */
320 count = p->swap_map[swp_offset(entry)] - 1;
321 spin_unlock(&swap_lock);
327 * We can use this swap cache entry directly
328 * if there are no other references to it.
330 int can_share_swap_page(struct page *page)
334 BUG_ON(!PageLocked(page));
335 count = page_mapcount(page);
336 if (count <= 1 && PageSwapCache(page))
337 count += page_swapcount(page);
342 * Work out if there are any other processes sharing this
343 * swap cache page. Free it if you can. Return success.
345 int remove_exclusive_swap_page(struct page *page)
348 struct swap_info_struct * p;
351 BUG_ON(PagePrivate(page));
352 BUG_ON(!PageLocked(page));
354 if (!PageSwapCache(page))
356 if (PageWriteback(page))
358 if (page_count(page) != 2) /* 2: us + cache */
361 entry.val = page_private(page);
362 p = swap_info_get(entry);
366 /* Is the only swap cache user the cache itself? */
368 if (p->swap_map[swp_offset(entry)] == 1) {
369 /* Recheck the page count with the swapcache lock held.. */
370 write_lock_irq(&swapper_space.tree_lock);
371 if ((page_count(page) == 2) && !PageWriteback(page)) {
372 __delete_from_swap_cache(page);
376 write_unlock_irq(&swapper_space.tree_lock);
378 spin_unlock(&swap_lock);
382 page_cache_release(page);
389 * Free the swap entry like above, but also try to
390 * free the page cache entry if it is the last user.
392 void free_swap_and_cache(swp_entry_t entry)
394 struct swap_info_struct * p;
395 struct page *page = NULL;
397 if (is_migration_entry(entry))
400 p = swap_info_get(entry);
402 if (swap_entry_free(p, swp_offset(entry)) == 1) {
403 page = find_get_page(&swapper_space, entry.val);
404 if (page && unlikely(TestSetPageLocked(page))) {
405 page_cache_release(page);
409 spin_unlock(&swap_lock);
414 BUG_ON(PagePrivate(page));
415 one_user = (page_count(page) == 2);
416 /* Only cache user (+us), or swap space full? Free it! */
417 /* Also recheck PageSwapCache after page is locked (above) */
418 if (PageSwapCache(page) && !PageWriteback(page) &&
419 (one_user || vm_swap_full())) {
420 delete_from_swap_cache(page);
424 page_cache_release(page);
428 #ifdef CONFIG_SOFTWARE_SUSPEND
430 * Find the swap type that corresponds to given device (if any).
432 * @offset - number of the PAGE_SIZE-sized block of the device, starting
433 * from 0, in which the swap header is expected to be located.
435 * This is needed for the suspend to disk (aka swsusp).
437 int swap_type_of(dev_t device, sector_t offset)
439 struct block_device *bdev = NULL;
443 bdev = bdget(device);
445 spin_lock(&swap_lock);
446 for (i = 0; i < nr_swapfiles; i++) {
447 struct swap_info_struct *sis = swap_info + i;
449 if (!(sis->flags & SWP_WRITEOK))
453 spin_unlock(&swap_lock);
456 if (bdev == sis->bdev) {
457 struct swap_extent *se;
459 se = list_entry(sis->extent_list.next,
460 struct swap_extent, list);
461 if (se->start_block == offset) {
462 spin_unlock(&swap_lock);
468 spin_unlock(&swap_lock);
476 * Return either the total number of swap pages of given type, or the number
477 * of free pages of that type (depending on @free)
479 * This is needed for software suspend
481 unsigned int count_swap_pages(int type, int free)
485 if (type < nr_swapfiles) {
486 spin_lock(&swap_lock);
487 if (swap_info[type].flags & SWP_WRITEOK) {
488 n = swap_info[type].pages;
490 n -= swap_info[type].inuse_pages;
492 spin_unlock(&swap_lock);
499 * No need to decide whether this PTE shares the swap entry with others,
500 * just let do_wp_page work it out if a write is requested later - to
501 * force COW, vm_page_prot omits write permission from any private vma.
503 static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
504 unsigned long addr, swp_entry_t entry, struct page *page)
506 inc_mm_counter(vma->vm_mm, anon_rss);
508 set_pte_at(vma->vm_mm, addr, pte,
509 pte_mkold(mk_pte(page, vma->vm_page_prot)));
510 page_add_anon_rmap(page, vma, addr);
513 * Move the page to the active list so it is not
514 * immediately swapped out again after swapon.
519 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
520 unsigned long addr, unsigned long end,
521 swp_entry_t entry, struct page *page)
523 pte_t swp_pte = swp_entry_to_pte(entry);
528 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
531 * swapoff spends a _lot_ of time in this loop!
532 * Test inline before going to call unuse_pte.
534 if (unlikely(pte_same(*pte, swp_pte))) {
535 unuse_pte(vma, pte++, addr, entry, page);
539 } while (pte++, addr += PAGE_SIZE, addr != end);
540 pte_unmap_unlock(pte - 1, ptl);
544 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
545 unsigned long addr, unsigned long end,
546 swp_entry_t entry, struct page *page)
551 pmd = pmd_offset(pud, addr);
553 next = pmd_addr_end(addr, end);
554 if (pmd_none_or_clear_bad(pmd))
556 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
558 } while (pmd++, addr = next, addr != end);
562 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
563 unsigned long addr, unsigned long end,
564 swp_entry_t entry, struct page *page)
569 pud = pud_offset(pgd, addr);
571 next = pud_addr_end(addr, end);
572 if (pud_none_or_clear_bad(pud))
574 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
576 } while (pud++, addr = next, addr != end);
580 static int unuse_vma(struct vm_area_struct *vma,
581 swp_entry_t entry, struct page *page)
584 unsigned long addr, end, next;
587 addr = page_address_in_vma(page, vma);
591 end = addr + PAGE_SIZE;
593 addr = vma->vm_start;
597 pgd = pgd_offset(vma->vm_mm, addr);
599 next = pgd_addr_end(addr, end);
600 if (pgd_none_or_clear_bad(pgd))
602 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
604 } while (pgd++, addr = next, addr != end);
608 static int unuse_mm(struct mm_struct *mm,
609 swp_entry_t entry, struct page *page)
611 struct vm_area_struct *vma;
613 if (!down_read_trylock(&mm->mmap_sem)) {
615 * Activate page so shrink_cache is unlikely to unmap its
616 * ptes while lock is dropped, so swapoff can make progress.
620 down_read(&mm->mmap_sem);
623 for (vma = mm->mmap; vma; vma = vma->vm_next) {
624 if (vma->anon_vma && unuse_vma(vma, entry, page))
627 up_read(&mm->mmap_sem);
629 * Currently unuse_mm cannot fail, but leave error handling
630 * at call sites for now, since we change it from time to time.
636 * Scan swap_map from current position to next entry still in use.
637 * Recycle to start on reaching the end, returning 0 when empty.
639 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
642 unsigned int max = si->max;
643 unsigned int i = prev;
647 * No need for swap_lock here: we're just looking
648 * for whether an entry is in use, not modifying it; false
649 * hits are okay, and sys_swapoff() has already prevented new
650 * allocations from this area (while holding swap_lock).
659 * No entries in use at top of swap_map,
660 * loop back to start and recheck there.
666 count = si->swap_map[i];
667 if (count && count != SWAP_MAP_BAD)
674 * We completely avoid races by reading each swap page in advance,
675 * and then search for the process using it. All the necessary
676 * page table adjustments can then be made atomically.
678 static int try_to_unuse(unsigned int type)
680 struct swap_info_struct * si = &swap_info[type];
681 struct mm_struct *start_mm;
682 unsigned short *swap_map;
683 unsigned short swcount;
688 int reset_overflow = 0;
692 * When searching mms for an entry, a good strategy is to
693 * start at the first mm we freed the previous entry from
694 * (though actually we don't notice whether we or coincidence
695 * freed the entry). Initialize this start_mm with a hold.
697 * A simpler strategy would be to start at the last mm we
698 * freed the previous entry from; but that would take less
699 * advantage of mmlist ordering, which clusters forked mms
700 * together, child after parent. If we race with dup_mmap(), we
701 * prefer to resolve parent before child, lest we miss entries
702 * duplicated after we scanned child: using last mm would invert
703 * that. Though it's only a serious concern when an overflowed
704 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
707 atomic_inc(&init_mm.mm_users);
710 * Keep on scanning until all entries have gone. Usually,
711 * one pass through swap_map is enough, but not necessarily:
712 * there are races when an instance of an entry might be missed.
714 while ((i = find_next_to_unuse(si, i)) != 0) {
715 if (signal_pending(current)) {
721 * Get a page for the entry, using the existing swap
722 * cache page if there is one. Otherwise, get a clean
723 * page and read the swap into it.
725 swap_map = &si->swap_map[i];
726 entry = swp_entry(type, i);
727 page = read_swap_cache_async(entry, NULL, 0);
730 * Either swap_duplicate() failed because entry
731 * has been freed independently, and will not be
732 * reused since sys_swapoff() already disabled
733 * allocation from here, or alloc_page() failed.
742 * Don't hold on to start_mm if it looks like exiting.
744 if (atomic_read(&start_mm->mm_users) == 1) {
747 atomic_inc(&init_mm.mm_users);
751 * Wait for and lock page. When do_swap_page races with
752 * try_to_unuse, do_swap_page can handle the fault much
753 * faster than try_to_unuse can locate the entry. This
754 * apparently redundant "wait_on_page_locked" lets try_to_unuse
755 * defer to do_swap_page in such a case - in some tests,
756 * do_swap_page and try_to_unuse repeatedly compete.
758 wait_on_page_locked(page);
759 wait_on_page_writeback(page);
761 wait_on_page_writeback(page);
764 * Remove all references to entry.
765 * Whenever we reach init_mm, there's no address space
766 * to search, but use it as a reminder to search shmem.
771 if (start_mm == &init_mm)
772 shmem = shmem_unuse(entry, page);
774 retval = unuse_mm(start_mm, entry, page);
777 int set_start_mm = (*swap_map >= swcount);
778 struct list_head *p = &start_mm->mmlist;
779 struct mm_struct *new_start_mm = start_mm;
780 struct mm_struct *prev_mm = start_mm;
781 struct mm_struct *mm;
783 atomic_inc(&new_start_mm->mm_users);
784 atomic_inc(&prev_mm->mm_users);
785 spin_lock(&mmlist_lock);
786 while (*swap_map > 1 && !retval &&
787 (p = p->next) != &start_mm->mmlist) {
788 mm = list_entry(p, struct mm_struct, mmlist);
789 if (!atomic_inc_not_zero(&mm->mm_users))
791 spin_unlock(&mmlist_lock);
800 else if (mm == &init_mm) {
802 shmem = shmem_unuse(entry, page);
804 retval = unuse_mm(mm, entry, page);
805 if (set_start_mm && *swap_map < swcount) {
807 atomic_inc(&mm->mm_users);
811 spin_lock(&mmlist_lock);
813 spin_unlock(&mmlist_lock);
816 start_mm = new_start_mm;
820 page_cache_release(page);
825 * How could swap count reach 0x7fff when the maximum
826 * pid is 0x7fff, and there's no way to repeat a swap
827 * page within an mm (except in shmem, where it's the
828 * shared object which takes the reference count)?
829 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
831 * If that's wrong, then we should worry more about
832 * exit_mmap() and do_munmap() cases described above:
833 * we might be resetting SWAP_MAP_MAX too early here.
834 * We know "Undead"s can happen, they're okay, so don't
835 * report them; but do report if we reset SWAP_MAP_MAX.
837 if (*swap_map == SWAP_MAP_MAX) {
838 spin_lock(&swap_lock);
840 spin_unlock(&swap_lock);
845 * If a reference remains (rare), we would like to leave
846 * the page in the swap cache; but try_to_unmap could
847 * then re-duplicate the entry once we drop page lock,
848 * so we might loop indefinitely; also, that page could
849 * not be swapped out to other storage meanwhile. So:
850 * delete from cache even if there's another reference,
851 * after ensuring that the data has been saved to disk -
852 * since if the reference remains (rarer), it will be
853 * read from disk into another page. Splitting into two
854 * pages would be incorrect if swap supported "shared
855 * private" pages, but they are handled by tmpfs files.
857 * Note shmem_unuse already deleted a swappage from
858 * the swap cache, unless the move to filepage failed:
859 * in which case it left swappage in cache, lowered its
860 * swap count to pass quickly through the loops above,
861 * and now we must reincrement count to try again later.
863 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
864 struct writeback_control wbc = {
865 .sync_mode = WB_SYNC_NONE,
868 swap_writepage(page, &wbc);
870 wait_on_page_writeback(page);
872 if (PageSwapCache(page)) {
874 swap_duplicate(entry);
876 delete_from_swap_cache(page);
880 * So we could skip searching mms once swap count went
881 * to 1, we did not mark any present ptes as dirty: must
882 * mark page dirty so shrink_list will preserve it.
886 page_cache_release(page);
889 * Make sure that we aren't completely killing
890 * interactive performance.
896 if (reset_overflow) {
897 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
904 * After a successful try_to_unuse, if no swap is now in use, we know
905 * we can empty the mmlist. swap_lock must be held on entry and exit.
906 * Note that mmlist_lock nests inside swap_lock, and an mm must be
907 * added to the mmlist just after page_duplicate - before would be racy.
909 static void drain_mmlist(void)
911 struct list_head *p, *next;
914 for (i = 0; i < nr_swapfiles; i++)
915 if (swap_info[i].inuse_pages)
917 spin_lock(&mmlist_lock);
918 list_for_each_safe(p, next, &init_mm.mmlist)
920 spin_unlock(&mmlist_lock);
924 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
925 * corresponds to page offset `offset'.
927 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
929 struct swap_extent *se = sis->curr_swap_extent;
930 struct swap_extent *start_se = se;
933 struct list_head *lh;
935 if (se->start_page <= offset &&
936 offset < (se->start_page + se->nr_pages)) {
937 return se->start_block + (offset - se->start_page);
940 if (lh == &sis->extent_list)
942 se = list_entry(lh, struct swap_extent, list);
943 sis->curr_swap_extent = se;
944 BUG_ON(se == start_se); /* It *must* be present */
948 #ifdef CONFIG_SOFTWARE_SUSPEND
950 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
951 * corresponding to given index in swap_info (swap type).
953 sector_t swapdev_block(int swap_type, pgoff_t offset)
955 struct swap_info_struct *sis;
957 if (swap_type >= nr_swapfiles)
960 sis = swap_info + swap_type;
961 return (sis->flags & SWP_WRITEOK) ? map_swap_page(sis, offset) : 0;
963 #endif /* CONFIG_SOFTWARE_SUSPEND */
966 * Free all of a swapdev's extent information
968 static void destroy_swap_extents(struct swap_info_struct *sis)
970 while (!list_empty(&sis->extent_list)) {
971 struct swap_extent *se;
973 se = list_entry(sis->extent_list.next,
974 struct swap_extent, list);
981 * Add a block range (and the corresponding page range) into this swapdev's
982 * extent list. The extent list is kept sorted in page order.
984 * This function rather assumes that it is called in ascending page order.
987 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
988 unsigned long nr_pages, sector_t start_block)
990 struct swap_extent *se;
991 struct swap_extent *new_se;
992 struct list_head *lh;
994 lh = sis->extent_list.prev; /* The highest page extent */
995 if (lh != &sis->extent_list) {
996 se = list_entry(lh, struct swap_extent, list);
997 BUG_ON(se->start_page + se->nr_pages != start_page);
998 if (se->start_block + se->nr_pages == start_block) {
1000 se->nr_pages += nr_pages;
1006 * No merge. Insert a new extent, preserving ordering.
1008 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1011 new_se->start_page = start_page;
1012 new_se->nr_pages = nr_pages;
1013 new_se->start_block = start_block;
1015 list_add_tail(&new_se->list, &sis->extent_list);
1020 * A `swap extent' is a simple thing which maps a contiguous range of pages
1021 * onto a contiguous range of disk blocks. An ordered list of swap extents
1022 * is built at swapon time and is then used at swap_writepage/swap_readpage
1023 * time for locating where on disk a page belongs.
1025 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1026 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1027 * swap files identically.
1029 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1030 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1031 * swapfiles are handled *identically* after swapon time.
1033 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1034 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1035 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1036 * requirements, they are simply tossed out - we will never use those blocks
1039 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1040 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1041 * which will scribble on the fs.
1043 * The amount of disk space which a single swap extent represents varies.
1044 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1045 * extents in the list. To avoid much list walking, we cache the previous
1046 * search location in `curr_swap_extent', and start new searches from there.
1047 * This is extremely effective. The average number of iterations in
1048 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1050 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1052 struct inode *inode;
1053 unsigned blocks_per_page;
1054 unsigned long page_no;
1056 sector_t probe_block;
1057 sector_t last_block;
1058 sector_t lowest_block = -1;
1059 sector_t highest_block = 0;
1063 inode = sis->swap_file->f_mapping->host;
1064 if (S_ISBLK(inode->i_mode)) {
1065 ret = add_swap_extent(sis, 0, sis->max, 0);
1070 blkbits = inode->i_blkbits;
1071 blocks_per_page = PAGE_SIZE >> blkbits;
1074 * Map all the blocks into the extent list. This code doesn't try
1079 last_block = i_size_read(inode) >> blkbits;
1080 while ((probe_block + blocks_per_page) <= last_block &&
1081 page_no < sis->max) {
1082 unsigned block_in_page;
1083 sector_t first_block;
1085 first_block = bmap(inode, probe_block);
1086 if (first_block == 0)
1090 * It must be PAGE_SIZE aligned on-disk
1092 if (first_block & (blocks_per_page - 1)) {
1097 for (block_in_page = 1; block_in_page < blocks_per_page;
1101 block = bmap(inode, probe_block + block_in_page);
1104 if (block != first_block + block_in_page) {
1111 first_block >>= (PAGE_SHIFT - blkbits);
1112 if (page_no) { /* exclude the header page */
1113 if (first_block < lowest_block)
1114 lowest_block = first_block;
1115 if (first_block > highest_block)
1116 highest_block = first_block;
1120 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1122 ret = add_swap_extent(sis, page_no, 1, first_block);
1127 probe_block += blocks_per_page;
1132 *span = 1 + highest_block - lowest_block;
1134 page_no = 1; /* force Empty message */
1136 sis->pages = page_no - 1;
1137 sis->highest_bit = page_no - 1;
1139 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1140 struct swap_extent, list);
1143 printk(KERN_ERR "swapon: swapfile has holes\n");
1149 #if 0 /* We don't need this yet */
1150 #include <linux/backing-dev.h>
1151 int page_queue_congested(struct page *page)
1153 struct backing_dev_info *bdi;
1155 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1157 if (PageSwapCache(page)) {
1158 swp_entry_t entry = { .val = page_private(page) };
1159 struct swap_info_struct *sis;
1161 sis = get_swap_info_struct(swp_type(entry));
1162 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1164 bdi = page->mapping->backing_dev_info;
1165 return bdi_write_congested(bdi);
1169 asmlinkage long sys_swapoff(const char __user * specialfile)
1171 struct swap_info_struct * p = NULL;
1172 unsigned short *swap_map;
1173 struct file *swap_file, *victim;
1174 struct address_space *mapping;
1175 struct inode *inode;
1180 if (!capable(CAP_SYS_ADMIN))
1183 pathname = getname(specialfile);
1184 err = PTR_ERR(pathname);
1185 if (IS_ERR(pathname))
1188 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1190 err = PTR_ERR(victim);
1194 mapping = victim->f_mapping;
1196 spin_lock(&swap_lock);
1197 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1198 p = swap_info + type;
1199 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1200 if (p->swap_file->f_mapping == mapping)
1207 spin_unlock(&swap_lock);
1210 if (!security_vm_enough_memory(p->pages))
1211 vm_unacct_memory(p->pages);
1214 spin_unlock(&swap_lock);
1218 swap_list.head = p->next;
1220 swap_info[prev].next = p->next;
1222 if (type == swap_list.next) {
1223 /* just pick something that's safe... */
1224 swap_list.next = swap_list.head;
1226 nr_swap_pages -= p->pages;
1227 total_swap_pages -= p->pages;
1228 p->flags &= ~SWP_WRITEOK;
1229 spin_unlock(&swap_lock);
1231 current->flags |= PF_SWAPOFF;
1232 err = try_to_unuse(type);
1233 current->flags &= ~PF_SWAPOFF;
1236 /* re-insert swap space back into swap_list */
1237 spin_lock(&swap_lock);
1238 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1239 if (p->prio >= swap_info[i].prio)
1243 swap_list.head = swap_list.next = p - swap_info;
1245 swap_info[prev].next = p - swap_info;
1246 nr_swap_pages += p->pages;
1247 total_swap_pages += p->pages;
1248 p->flags |= SWP_WRITEOK;
1249 spin_unlock(&swap_lock);
1253 /* wait for any unplug function to finish */
1254 down_write(&swap_unplug_sem);
1255 up_write(&swap_unplug_sem);
1257 destroy_swap_extents(p);
1258 mutex_lock(&swapon_mutex);
1259 spin_lock(&swap_lock);
1262 /* wait for anyone still in scan_swap_map */
1263 p->highest_bit = 0; /* cuts scans short */
1264 while (p->flags >= SWP_SCANNING) {
1265 spin_unlock(&swap_lock);
1266 schedule_timeout_uninterruptible(1);
1267 spin_lock(&swap_lock);
1270 swap_file = p->swap_file;
1271 p->swap_file = NULL;
1273 swap_map = p->swap_map;
1276 spin_unlock(&swap_lock);
1277 mutex_unlock(&swapon_mutex);
1279 inode = mapping->host;
1280 if (S_ISBLK(inode->i_mode)) {
1281 struct block_device *bdev = I_BDEV(inode);
1282 set_blocksize(bdev, p->old_block_size);
1285 mutex_lock(&inode->i_mutex);
1286 inode->i_flags &= ~S_SWAPFILE;
1287 mutex_unlock(&inode->i_mutex);
1289 filp_close(swap_file, NULL);
1293 filp_close(victim, NULL);
1298 #ifdef CONFIG_PROC_FS
1300 static void *swap_start(struct seq_file *swap, loff_t *pos)
1302 struct swap_info_struct *ptr = swap_info;
1306 mutex_lock(&swapon_mutex);
1309 return SEQ_START_TOKEN;
1311 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1312 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1321 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1323 struct swap_info_struct *ptr;
1324 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1326 if (v == SEQ_START_TOKEN)
1333 for (; ptr < endptr; ptr++) {
1334 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1343 static void swap_stop(struct seq_file *swap, void *v)
1345 mutex_unlock(&swapon_mutex);
1348 static int swap_show(struct seq_file *swap, void *v)
1350 struct swap_info_struct *ptr = v;
1354 if (ptr == SEQ_START_TOKEN) {
1355 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1359 file = ptr->swap_file;
1360 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1361 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1362 len < 40 ? 40 - len : 1, " ",
1363 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1364 "partition" : "file\t",
1365 ptr->pages << (PAGE_SHIFT - 10),
1366 ptr->inuse_pages << (PAGE_SHIFT - 10),
1371 static const struct seq_operations swaps_op = {
1372 .start = swap_start,
1378 static int swaps_open(struct inode *inode, struct file *file)
1380 return seq_open(file, &swaps_op);
1383 static const struct file_operations proc_swaps_operations = {
1386 .llseek = seq_lseek,
1387 .release = seq_release,
1390 static int __init procswaps_init(void)
1392 struct proc_dir_entry *entry;
1394 entry = create_proc_entry("swaps", 0, NULL);
1396 entry->proc_fops = &proc_swaps_operations;
1399 __initcall(procswaps_init);
1400 #endif /* CONFIG_PROC_FS */
1403 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1405 * The swapon system call
1407 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1409 struct swap_info_struct * p;
1411 struct block_device *bdev = NULL;
1412 struct file *swap_file = NULL;
1413 struct address_space *mapping;
1417 static int least_priority;
1418 union swap_header *swap_header = NULL;
1419 int swap_header_version;
1420 unsigned int nr_good_pages = 0;
1423 unsigned long maxpages = 1;
1425 unsigned short *swap_map;
1426 struct page *page = NULL;
1427 struct inode *inode = NULL;
1430 if (!capable(CAP_SYS_ADMIN))
1432 spin_lock(&swap_lock);
1434 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1435 if (!(p->flags & SWP_USED))
1438 if (type >= MAX_SWAPFILES) {
1439 spin_unlock(&swap_lock);
1442 if (type >= nr_swapfiles)
1443 nr_swapfiles = type+1;
1444 INIT_LIST_HEAD(&p->extent_list);
1445 p->flags = SWP_USED;
1446 p->swap_file = NULL;
1447 p->old_block_size = 0;
1454 if (swap_flags & SWAP_FLAG_PREFER) {
1456 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1458 p->prio = --least_priority;
1460 spin_unlock(&swap_lock);
1461 name = getname(specialfile);
1462 error = PTR_ERR(name);
1467 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1468 error = PTR_ERR(swap_file);
1469 if (IS_ERR(swap_file)) {
1474 p->swap_file = swap_file;
1475 mapping = swap_file->f_mapping;
1476 inode = mapping->host;
1479 for (i = 0; i < nr_swapfiles; i++) {
1480 struct swap_info_struct *q = &swap_info[i];
1482 if (i == type || !q->swap_file)
1484 if (mapping == q->swap_file->f_mapping)
1489 if (S_ISBLK(inode->i_mode)) {
1490 bdev = I_BDEV(inode);
1491 error = bd_claim(bdev, sys_swapon);
1497 p->old_block_size = block_size(bdev);
1498 error = set_blocksize(bdev, PAGE_SIZE);
1502 } else if (S_ISREG(inode->i_mode)) {
1503 p->bdev = inode->i_sb->s_bdev;
1504 mutex_lock(&inode->i_mutex);
1506 if (IS_SWAPFILE(inode)) {
1514 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1517 * Read the swap header.
1519 if (!mapping->a_ops->readpage) {
1523 page = read_mapping_page(mapping, 0, swap_file);
1525 error = PTR_ERR(page);
1528 wait_on_page_locked(page);
1529 if (!PageUptodate(page))
1532 swap_header = page_address(page);
1534 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1535 swap_header_version = 1;
1536 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1537 swap_header_version = 2;
1539 printk(KERN_ERR "Unable to find swap-space signature\n");
1544 switch (swap_header_version) {
1546 printk(KERN_ERR "version 0 swap is no longer supported. "
1547 "Use mkswap -v1 %s\n", name);
1551 /* Check the swap header's sub-version and the size of
1552 the swap file and bad block lists */
1553 if (swap_header->info.version != 1) {
1555 "Unable to handle swap header version %d\n",
1556 swap_header->info.version);
1562 p->cluster_next = 1;
1565 * Find out how many pages are allowed for a single swap
1566 * device. There are two limiting factors: 1) the number of
1567 * bits for the swap offset in the swp_entry_t type and
1568 * 2) the number of bits in the a swap pte as defined by
1569 * the different architectures. In order to find the
1570 * largest possible bit mask a swap entry with swap type 0
1571 * and swap offset ~0UL is created, encoded to a swap pte,
1572 * decoded to a swp_entry_t again and finally the swap
1573 * offset is extracted. This will mask all the bits from
1574 * the initial ~0UL mask that can't be encoded in either
1575 * the swp_entry_t or the architecture definition of a
1578 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1579 if (maxpages > swap_header->info.last_page)
1580 maxpages = swap_header->info.last_page;
1581 p->highest_bit = maxpages - 1;
1586 if (swapfilesize && maxpages > swapfilesize) {
1588 "Swap area shorter than signature indicates\n");
1591 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1593 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1596 /* OK, set up the swap map and apply the bad block list */
1597 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1603 memset(p->swap_map, 0, maxpages * sizeof(short));
1604 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1605 int page_nr = swap_header->info.badpages[i];
1606 if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
1609 p->swap_map[page_nr] = SWAP_MAP_BAD;
1611 nr_good_pages = swap_header->info.last_page -
1612 swap_header->info.nr_badpages -
1613 1 /* header page */;
1618 if (nr_good_pages) {
1619 p->swap_map[0] = SWAP_MAP_BAD;
1621 p->pages = nr_good_pages;
1622 nr_extents = setup_swap_extents(p, &span);
1623 if (nr_extents < 0) {
1627 nr_good_pages = p->pages;
1629 if (!nr_good_pages) {
1630 printk(KERN_WARNING "Empty swap-file\n");
1635 mutex_lock(&swapon_mutex);
1636 spin_lock(&swap_lock);
1637 p->flags = SWP_ACTIVE;
1638 nr_swap_pages += nr_good_pages;
1639 total_swap_pages += nr_good_pages;
1641 printk(KERN_INFO "Adding %uk swap on %s. "
1642 "Priority:%d extents:%d across:%lluk\n",
1643 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1644 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
1646 /* insert swap space into swap_list: */
1648 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1649 if (p->prio >= swap_info[i].prio) {
1656 swap_list.head = swap_list.next = p - swap_info;
1658 swap_info[prev].next = p - swap_info;
1660 spin_unlock(&swap_lock);
1661 mutex_unlock(&swapon_mutex);
1666 set_blocksize(bdev, p->old_block_size);
1669 destroy_swap_extents(p);
1671 spin_lock(&swap_lock);
1672 swap_map = p->swap_map;
1673 p->swap_file = NULL;
1676 if (!(swap_flags & SWAP_FLAG_PREFER))
1678 spin_unlock(&swap_lock);
1681 filp_close(swap_file, NULL);
1683 if (page && !IS_ERR(page)) {
1685 page_cache_release(page);
1691 inode->i_flags |= S_SWAPFILE;
1692 mutex_unlock(&inode->i_mutex);
1697 void si_swapinfo(struct sysinfo *val)
1700 unsigned long nr_to_be_unused = 0;
1702 spin_lock(&swap_lock);
1703 for (i = 0; i < nr_swapfiles; i++) {
1704 if (!(swap_info[i].flags & SWP_USED) ||
1705 (swap_info[i].flags & SWP_WRITEOK))
1707 nr_to_be_unused += swap_info[i].inuse_pages;
1709 val->freeswap = nr_swap_pages + nr_to_be_unused;
1710 val->totalswap = total_swap_pages + nr_to_be_unused;
1711 spin_unlock(&swap_lock);
1715 * Verify that a swap entry is valid and increment its swap map count.
1717 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1718 * "permanent", but will be reclaimed by the next swapoff.
1720 int swap_duplicate(swp_entry_t entry)
1722 struct swap_info_struct * p;
1723 unsigned long offset, type;
1726 if (is_migration_entry(entry))
1729 type = swp_type(entry);
1730 if (type >= nr_swapfiles)
1732 p = type + swap_info;
1733 offset = swp_offset(entry);
1735 spin_lock(&swap_lock);
1736 if (offset < p->max && p->swap_map[offset]) {
1737 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1738 p->swap_map[offset]++;
1740 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1741 if (swap_overflow++ < 5)
1742 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1743 p->swap_map[offset] = SWAP_MAP_MAX;
1747 spin_unlock(&swap_lock);
1752 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1756 struct swap_info_struct *
1757 get_swap_info_struct(unsigned type)
1759 return &swap_info[type];
1763 * swap_lock prevents swap_map being freed. Don't grab an extra
1764 * reference on the swaphandle, it doesn't matter if it becomes unused.
1766 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1768 int our_page_cluster = page_cluster;
1769 int ret = 0, i = 1 << our_page_cluster;
1771 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1773 if (!our_page_cluster) /* no readahead */
1775 toff = (swp_offset(entry) >> our_page_cluster) << our_page_cluster;
1776 if (!toff) /* first page is swap header */
1780 spin_lock(&swap_lock);
1782 /* Don't read-ahead past the end of the swap area */
1783 if (toff >= swapdev->max)
1785 /* Don't read in free or bad pages */
1786 if (!swapdev->swap_map[toff])
1788 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1793 spin_unlock(&swap_lock);