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>
30 #include <linux/memcontrol.h>
32 #include <asm/pgtable.h>
33 #include <asm/tlbflush.h>
34 #include <linux/swapops.h>
36 DEFINE_SPINLOCK(swap_lock);
37 unsigned int nr_swapfiles;
38 long total_swap_pages;
39 static int swap_overflow;
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 ";
46 struct swap_list_t swap_list = {-1, -1};
48 static struct swap_info_struct swap_info[MAX_SWAPFILES];
50 static DEFINE_MUTEX(swapon_mutex);
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.
57 static DECLARE_RWSEM(swap_unplug_sem);
59 void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
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;
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.
77 WARN_ON(page_count(page) <= 1);
79 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
80 blk_run_backing_dev(bdi, page);
82 up_read(&swap_unplug_sem);
85 #define SWAPFILE_CLUSTER 256
86 #define LATENCY_LIMIT 256
88 static inline unsigned long scan_swap_map(struct swap_info_struct *si)
90 unsigned long offset, last_in_cluster;
91 int latency_ration = LATENCY_LIMIT;
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
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)
108 spin_unlock(&swap_lock);
110 offset = si->lowest_bit;
111 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
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;
122 if (unlikely(--latency_ration < 0)) {
124 latency_ration = LATENCY_LIMIT;
127 spin_lock(&swap_lock);
133 offset = si->cluster_next;
134 if (offset > si->highest_bit)
135 lowest: offset = si->lowest_bit;
136 checks: if (!(si->flags & SWP_WRITEOK))
138 if (!si->highest_bit)
140 if (!si->swap_map[offset]) {
141 if (offset == si->lowest_bit)
143 if (offset == si->highest_bit)
146 if (si->inuse_pages == si->pages) {
147 si->lowest_bit = si->max;
150 si->swap_map[offset] = 1;
151 si->cluster_next = offset + 1;
152 si->flags -= SWP_SCANNING;
156 spin_unlock(&swap_lock);
157 while (++offset <= si->highest_bit) {
158 if (!si->swap_map[offset]) {
159 spin_lock(&swap_lock);
162 if (unlikely(--latency_ration < 0)) {
164 latency_ration = LATENCY_LIMIT;
167 spin_lock(&swap_lock);
171 si->flags -= SWP_SCANNING;
175 swp_entry_t get_swap_page(void)
177 struct swap_info_struct *si;
182 spin_lock(&swap_lock);
183 if (nr_swap_pages <= 0)
187 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
188 si = swap_info + type;
191 (!wrapped && si->prio != swap_info[next].prio)) {
192 next = swap_list.head;
196 if (!si->highest_bit)
198 if (!(si->flags & SWP_WRITEOK))
201 swap_list.next = next;
202 offset = scan_swap_map(si);
204 spin_unlock(&swap_lock);
205 return swp_entry(type, offset);
207 next = swap_list.next;
212 spin_unlock(&swap_lock);
213 return (swp_entry_t) {0};
216 swp_entry_t get_swap_page_of_type(int type)
218 struct swap_info_struct *si;
221 spin_lock(&swap_lock);
222 si = swap_info + type;
223 if (si->flags & SWP_WRITEOK) {
225 offset = scan_swap_map(si);
227 spin_unlock(&swap_lock);
228 return swp_entry(type, offset);
232 spin_unlock(&swap_lock);
233 return (swp_entry_t) {0};
236 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
238 struct swap_info_struct * p;
239 unsigned long offset, type;
243 type = swp_type(entry);
244 if (type >= nr_swapfiles)
246 p = & swap_info[type];
247 if (!(p->flags & SWP_USED))
249 offset = swp_offset(entry);
250 if (offset >= p->max)
252 if (!p->swap_map[offset])
254 spin_lock(&swap_lock);
258 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
261 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
264 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
267 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
272 static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
274 int count = p->swap_map[offset];
276 if (count < SWAP_MAP_MAX) {
278 p->swap_map[offset] = 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;
294 * Caller has made sure that the swapdevice corresponding to entry
295 * is still around or has not been recycled.
297 void swap_free(swp_entry_t entry)
299 struct swap_info_struct * p;
301 p = swap_info_get(entry);
303 swap_entry_free(p, swp_offset(entry));
304 spin_unlock(&swap_lock);
309 * How many references to page are currently swapped out?
311 static inline int page_swapcount(struct page *page)
314 struct swap_info_struct *p;
317 entry.val = page_private(page);
318 p = swap_info_get(entry);
320 /* Subtract the 1 for the swap cache itself */
321 count = p->swap_map[swp_offset(entry)] - 1;
322 spin_unlock(&swap_lock);
328 * We can use this swap cache entry directly
329 * if there are no other references to it.
331 int can_share_swap_page(struct page *page)
335 BUG_ON(!PageLocked(page));
336 count = page_mapcount(page);
337 if (count <= 1 && PageSwapCache(page))
338 count += page_swapcount(page);
343 * Work out if there are any other processes sharing this
344 * swap cache page. Free it if you can. Return success.
346 int remove_exclusive_swap_page(struct page *page)
349 struct swap_info_struct * p;
352 BUG_ON(PagePrivate(page));
353 BUG_ON(!PageLocked(page));
355 if (!PageSwapCache(page))
357 if (PageWriteback(page))
359 if (page_count(page) != 2) /* 2: us + cache */
362 entry.val = page_private(page);
363 p = swap_info_get(entry);
367 /* Is the only swap cache user the cache itself? */
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);
377 write_unlock_irq(&swapper_space.tree_lock);
379 spin_unlock(&swap_lock);
383 page_cache_release(page);
390 * Free the swap entry like above, but also try to
391 * free the page cache entry if it is the last user.
393 void free_swap_and_cache(swp_entry_t entry)
395 struct swap_info_struct * p;
396 struct page *page = NULL;
398 if (is_migration_entry(entry))
401 p = swap_info_get(entry);
403 if (swap_entry_free(p, swp_offset(entry)) == 1) {
404 page = find_get_page(&swapper_space, entry.val);
405 if (page && unlikely(TestSetPageLocked(page))) {
406 page_cache_release(page);
410 spin_unlock(&swap_lock);
415 BUG_ON(PagePrivate(page));
416 one_user = (page_count(page) == 2);
417 /* Only cache user (+us), or swap space full? Free it! */
418 /* Also recheck PageSwapCache after page is locked (above) */
419 if (PageSwapCache(page) && !PageWriteback(page) &&
420 (one_user || vm_swap_full())) {
421 delete_from_swap_cache(page);
425 page_cache_release(page);
429 #ifdef CONFIG_HIBERNATION
431 * Find the swap type that corresponds to given device (if any).
433 * @offset - number of the PAGE_SIZE-sized block of the device, starting
434 * from 0, in which the swap header is expected to be located.
436 * This is needed for the suspend to disk (aka swsusp).
438 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
440 struct block_device *bdev = NULL;
444 bdev = bdget(device);
446 spin_lock(&swap_lock);
447 for (i = 0; i < nr_swapfiles; i++) {
448 struct swap_info_struct *sis = swap_info + i;
450 if (!(sis->flags & SWP_WRITEOK))
457 spin_unlock(&swap_lock);
460 if (bdev == sis->bdev) {
461 struct swap_extent *se;
463 se = list_entry(sis->extent_list.next,
464 struct swap_extent, list);
465 if (se->start_block == offset) {
469 spin_unlock(&swap_lock);
475 spin_unlock(&swap_lock);
483 * Return either the total number of swap pages of given type, or the number
484 * of free pages of that type (depending on @free)
486 * This is needed for software suspend
488 unsigned int count_swap_pages(int type, int free)
492 if (type < nr_swapfiles) {
493 spin_lock(&swap_lock);
494 if (swap_info[type].flags & SWP_WRITEOK) {
495 n = swap_info[type].pages;
497 n -= swap_info[type].inuse_pages;
499 spin_unlock(&swap_lock);
506 * No need to decide whether this PTE shares the swap entry with others,
507 * just let do_wp_page work it out if a write is requested later - to
508 * force COW, vm_page_prot omits write permission from any private vma.
510 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
511 unsigned long addr, swp_entry_t entry, struct page *page)
517 if (mem_cgroup_charge(page, vma->vm_mm, GFP_KERNEL))
520 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
521 if (unlikely(!pte_same(*pte, swp_entry_to_pte(entry)))) {
523 mem_cgroup_uncharge_page(page);
528 inc_mm_counter(vma->vm_mm, anon_rss);
530 set_pte_at(vma->vm_mm, addr, pte,
531 pte_mkold(mk_pte(page, vma->vm_page_prot)));
532 page_add_anon_rmap(page, vma, addr);
535 * Move the page to the active list so it is not
536 * immediately swapped out again after swapon.
540 pte_unmap_unlock(pte, ptl);
544 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
545 unsigned long addr, unsigned long end,
546 swp_entry_t entry, struct page *page)
548 pte_t swp_pte = swp_entry_to_pte(entry);
553 * We don't actually need pte lock while scanning for swp_pte: since
554 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
555 * page table while we're scanning; though it could get zapped, and on
556 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
557 * of unmatched parts which look like swp_pte, so unuse_pte must
558 * recheck under pte lock. Scanning without pte lock lets it be
559 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
561 pte = pte_offset_map(pmd, addr);
564 * swapoff spends a _lot_ of time in this loop!
565 * Test inline before going to call unuse_pte.
567 if (unlikely(pte_same(*pte, swp_pte))) {
569 ret = unuse_pte(vma, pmd, addr, entry, page);
572 pte = pte_offset_map(pmd, addr);
574 } while (pte++, addr += PAGE_SIZE, addr != end);
580 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
581 unsigned long addr, unsigned long end,
582 swp_entry_t entry, struct page *page)
588 pmd = pmd_offset(pud, addr);
590 next = pmd_addr_end(addr, end);
591 if (pmd_none_or_clear_bad(pmd))
593 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
596 } while (pmd++, addr = next, addr != end);
600 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
601 unsigned long addr, unsigned long end,
602 swp_entry_t entry, struct page *page)
608 pud = pud_offset(pgd, addr);
610 next = pud_addr_end(addr, end);
611 if (pud_none_or_clear_bad(pud))
613 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
616 } while (pud++, addr = next, addr != end);
620 static int unuse_vma(struct vm_area_struct *vma,
621 swp_entry_t entry, struct page *page)
624 unsigned long addr, end, next;
628 addr = page_address_in_vma(page, vma);
632 end = addr + PAGE_SIZE;
634 addr = vma->vm_start;
638 pgd = pgd_offset(vma->vm_mm, addr);
640 next = pgd_addr_end(addr, end);
641 if (pgd_none_or_clear_bad(pgd))
643 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
646 } while (pgd++, addr = next, addr != end);
650 static int unuse_mm(struct mm_struct *mm,
651 swp_entry_t entry, struct page *page)
653 struct vm_area_struct *vma;
656 if (!down_read_trylock(&mm->mmap_sem)) {
658 * Activate page so shrink_cache is unlikely to unmap its
659 * ptes while lock is dropped, so swapoff can make progress.
663 down_read(&mm->mmap_sem);
666 for (vma = mm->mmap; vma; vma = vma->vm_next) {
667 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
670 up_read(&mm->mmap_sem);
671 return (ret < 0)? ret: 0;
675 * Scan swap_map from current position to next entry still in use.
676 * Recycle to start on reaching the end, returning 0 when empty.
678 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
681 unsigned int max = si->max;
682 unsigned int i = prev;
686 * No need for swap_lock here: we're just looking
687 * for whether an entry is in use, not modifying it; false
688 * hits are okay, and sys_swapoff() has already prevented new
689 * allocations from this area (while holding swap_lock).
698 * No entries in use at top of swap_map,
699 * loop back to start and recheck there.
705 count = si->swap_map[i];
706 if (count && count != SWAP_MAP_BAD)
713 * We completely avoid races by reading each swap page in advance,
714 * and then search for the process using it. All the necessary
715 * page table adjustments can then be made atomically.
717 static int try_to_unuse(unsigned int type)
719 struct swap_info_struct * si = &swap_info[type];
720 struct mm_struct *start_mm;
721 unsigned short *swap_map;
722 unsigned short swcount;
727 int reset_overflow = 0;
731 * When searching mms for an entry, a good strategy is to
732 * start at the first mm we freed the previous entry from
733 * (though actually we don't notice whether we or coincidence
734 * freed the entry). Initialize this start_mm with a hold.
736 * A simpler strategy would be to start at the last mm we
737 * freed the previous entry from; but that would take less
738 * advantage of mmlist ordering, which clusters forked mms
739 * together, child after parent. If we race with dup_mmap(), we
740 * prefer to resolve parent before child, lest we miss entries
741 * duplicated after we scanned child: using last mm would invert
742 * that. Though it's only a serious concern when an overflowed
743 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
746 atomic_inc(&init_mm.mm_users);
749 * Keep on scanning until all entries have gone. Usually,
750 * one pass through swap_map is enough, but not necessarily:
751 * there are races when an instance of an entry might be missed.
753 while ((i = find_next_to_unuse(si, i)) != 0) {
754 if (signal_pending(current)) {
760 * Get a page for the entry, using the existing swap
761 * cache page if there is one. Otherwise, get a clean
762 * page and read the swap into it.
764 swap_map = &si->swap_map[i];
765 entry = swp_entry(type, i);
766 page = read_swap_cache_async(entry,
767 GFP_HIGHUSER_MOVABLE, NULL, 0);
770 * Either swap_duplicate() failed because entry
771 * has been freed independently, and will not be
772 * reused since sys_swapoff() already disabled
773 * allocation from here, or alloc_page() failed.
782 * Don't hold on to start_mm if it looks like exiting.
784 if (atomic_read(&start_mm->mm_users) == 1) {
787 atomic_inc(&init_mm.mm_users);
791 * Wait for and lock page. When do_swap_page races with
792 * try_to_unuse, do_swap_page can handle the fault much
793 * faster than try_to_unuse can locate the entry. This
794 * apparently redundant "wait_on_page_locked" lets try_to_unuse
795 * defer to do_swap_page in such a case - in some tests,
796 * do_swap_page and try_to_unuse repeatedly compete.
798 wait_on_page_locked(page);
799 wait_on_page_writeback(page);
801 wait_on_page_writeback(page);
804 * Remove all references to entry.
805 * Whenever we reach init_mm, there's no address space
806 * to search, but use it as a reminder to search shmem.
811 if (start_mm == &init_mm)
812 shmem = shmem_unuse(entry, page);
814 retval = unuse_mm(start_mm, entry, page);
817 int set_start_mm = (*swap_map >= swcount);
818 struct list_head *p = &start_mm->mmlist;
819 struct mm_struct *new_start_mm = start_mm;
820 struct mm_struct *prev_mm = start_mm;
821 struct mm_struct *mm;
823 atomic_inc(&new_start_mm->mm_users);
824 atomic_inc(&prev_mm->mm_users);
825 spin_lock(&mmlist_lock);
826 while (*swap_map > 1 && !retval && !shmem &&
827 (p = p->next) != &start_mm->mmlist) {
828 mm = list_entry(p, struct mm_struct, mmlist);
829 if (!atomic_inc_not_zero(&mm->mm_users))
831 spin_unlock(&mmlist_lock);
840 else if (mm == &init_mm) {
842 shmem = shmem_unuse(entry, page);
844 retval = unuse_mm(mm, entry, page);
845 if (set_start_mm && *swap_map < swcount) {
847 atomic_inc(&mm->mm_users);
851 spin_lock(&mmlist_lock);
853 spin_unlock(&mmlist_lock);
856 start_mm = new_start_mm;
859 /* page has already been unlocked and released */
867 page_cache_release(page);
872 * How could swap count reach 0x7fff when the maximum
873 * pid is 0x7fff, and there's no way to repeat a swap
874 * page within an mm (except in shmem, where it's the
875 * shared object which takes the reference count)?
876 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
878 * If that's wrong, then we should worry more about
879 * exit_mmap() and do_munmap() cases described above:
880 * we might be resetting SWAP_MAP_MAX too early here.
881 * We know "Undead"s can happen, they're okay, so don't
882 * report them; but do report if we reset SWAP_MAP_MAX.
884 if (*swap_map == SWAP_MAP_MAX) {
885 spin_lock(&swap_lock);
887 spin_unlock(&swap_lock);
892 * If a reference remains (rare), we would like to leave
893 * the page in the swap cache; but try_to_unmap could
894 * then re-duplicate the entry once we drop page lock,
895 * so we might loop indefinitely; also, that page could
896 * not be swapped out to other storage meanwhile. So:
897 * delete from cache even if there's another reference,
898 * after ensuring that the data has been saved to disk -
899 * since if the reference remains (rarer), it will be
900 * read from disk into another page. Splitting into two
901 * pages would be incorrect if swap supported "shared
902 * private" pages, but they are handled by tmpfs files.
904 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
905 struct writeback_control wbc = {
906 .sync_mode = WB_SYNC_NONE,
909 swap_writepage(page, &wbc);
911 wait_on_page_writeback(page);
913 if (PageSwapCache(page))
914 delete_from_swap_cache(page);
917 * So we could skip searching mms once swap count went
918 * to 1, we did not mark any present ptes as dirty: must
919 * mark page dirty so shrink_page_list will preserve it.
923 page_cache_release(page);
926 * Make sure that we aren't completely killing
927 * interactive performance.
933 if (reset_overflow) {
934 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
941 * After a successful try_to_unuse, if no swap is now in use, we know
942 * we can empty the mmlist. swap_lock must be held on entry and exit.
943 * Note that mmlist_lock nests inside swap_lock, and an mm must be
944 * added to the mmlist just after page_duplicate - before would be racy.
946 static void drain_mmlist(void)
948 struct list_head *p, *next;
951 for (i = 0; i < nr_swapfiles; i++)
952 if (swap_info[i].inuse_pages)
954 spin_lock(&mmlist_lock);
955 list_for_each_safe(p, next, &init_mm.mmlist)
957 spin_unlock(&mmlist_lock);
961 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
962 * corresponds to page offset `offset'.
964 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
966 struct swap_extent *se = sis->curr_swap_extent;
967 struct swap_extent *start_se = se;
970 struct list_head *lh;
972 if (se->start_page <= offset &&
973 offset < (se->start_page + se->nr_pages)) {
974 return se->start_block + (offset - se->start_page);
977 if (lh == &sis->extent_list)
979 se = list_entry(lh, struct swap_extent, list);
980 sis->curr_swap_extent = se;
981 BUG_ON(se == start_se); /* It *must* be present */
985 #ifdef CONFIG_HIBERNATION
987 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
988 * corresponding to given index in swap_info (swap type).
990 sector_t swapdev_block(int swap_type, pgoff_t offset)
992 struct swap_info_struct *sis;
994 if (swap_type >= nr_swapfiles)
997 sis = swap_info + swap_type;
998 return (sis->flags & SWP_WRITEOK) ? map_swap_page(sis, offset) : 0;
1000 #endif /* CONFIG_HIBERNATION */
1003 * Free all of a swapdev's extent information
1005 static void destroy_swap_extents(struct swap_info_struct *sis)
1007 while (!list_empty(&sis->extent_list)) {
1008 struct swap_extent *se;
1010 se = list_entry(sis->extent_list.next,
1011 struct swap_extent, list);
1012 list_del(&se->list);
1018 * Add a block range (and the corresponding page range) into this swapdev's
1019 * extent list. The extent list is kept sorted in page order.
1021 * This function rather assumes that it is called in ascending page order.
1024 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1025 unsigned long nr_pages, sector_t start_block)
1027 struct swap_extent *se;
1028 struct swap_extent *new_se;
1029 struct list_head *lh;
1031 lh = sis->extent_list.prev; /* The highest page extent */
1032 if (lh != &sis->extent_list) {
1033 se = list_entry(lh, struct swap_extent, list);
1034 BUG_ON(se->start_page + se->nr_pages != start_page);
1035 if (se->start_block + se->nr_pages == start_block) {
1037 se->nr_pages += nr_pages;
1043 * No merge. Insert a new extent, preserving ordering.
1045 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1048 new_se->start_page = start_page;
1049 new_se->nr_pages = nr_pages;
1050 new_se->start_block = start_block;
1052 list_add_tail(&new_se->list, &sis->extent_list);
1057 * A `swap extent' is a simple thing which maps a contiguous range of pages
1058 * onto a contiguous range of disk blocks. An ordered list of swap extents
1059 * is built at swapon time and is then used at swap_writepage/swap_readpage
1060 * time for locating where on disk a page belongs.
1062 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1063 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1064 * swap files identically.
1066 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1067 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1068 * swapfiles are handled *identically* after swapon time.
1070 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1071 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1072 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1073 * requirements, they are simply tossed out - we will never use those blocks
1076 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1077 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1078 * which will scribble on the fs.
1080 * The amount of disk space which a single swap extent represents varies.
1081 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1082 * extents in the list. To avoid much list walking, we cache the previous
1083 * search location in `curr_swap_extent', and start new searches from there.
1084 * This is extremely effective. The average number of iterations in
1085 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1087 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1089 struct inode *inode;
1090 unsigned blocks_per_page;
1091 unsigned long page_no;
1093 sector_t probe_block;
1094 sector_t last_block;
1095 sector_t lowest_block = -1;
1096 sector_t highest_block = 0;
1100 inode = sis->swap_file->f_mapping->host;
1101 if (S_ISBLK(inode->i_mode)) {
1102 ret = add_swap_extent(sis, 0, sis->max, 0);
1107 blkbits = inode->i_blkbits;
1108 blocks_per_page = PAGE_SIZE >> blkbits;
1111 * Map all the blocks into the extent list. This code doesn't try
1116 last_block = i_size_read(inode) >> blkbits;
1117 while ((probe_block + blocks_per_page) <= last_block &&
1118 page_no < sis->max) {
1119 unsigned block_in_page;
1120 sector_t first_block;
1122 first_block = bmap(inode, probe_block);
1123 if (first_block == 0)
1127 * It must be PAGE_SIZE aligned on-disk
1129 if (first_block & (blocks_per_page - 1)) {
1134 for (block_in_page = 1; block_in_page < blocks_per_page;
1138 block = bmap(inode, probe_block + block_in_page);
1141 if (block != first_block + block_in_page) {
1148 first_block >>= (PAGE_SHIFT - blkbits);
1149 if (page_no) { /* exclude the header page */
1150 if (first_block < lowest_block)
1151 lowest_block = first_block;
1152 if (first_block > highest_block)
1153 highest_block = first_block;
1157 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1159 ret = add_swap_extent(sis, page_no, 1, first_block);
1164 probe_block += blocks_per_page;
1169 *span = 1 + highest_block - lowest_block;
1171 page_no = 1; /* force Empty message */
1173 sis->pages = page_no - 1;
1174 sis->highest_bit = page_no - 1;
1176 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1177 struct swap_extent, list);
1180 printk(KERN_ERR "swapon: swapfile has holes\n");
1186 #if 0 /* We don't need this yet */
1187 #include <linux/backing-dev.h>
1188 int page_queue_congested(struct page *page)
1190 struct backing_dev_info *bdi;
1192 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1194 if (PageSwapCache(page)) {
1195 swp_entry_t entry = { .val = page_private(page) };
1196 struct swap_info_struct *sis;
1198 sis = get_swap_info_struct(swp_type(entry));
1199 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1201 bdi = page->mapping->backing_dev_info;
1202 return bdi_write_congested(bdi);
1206 asmlinkage long sys_swapoff(const char __user * specialfile)
1208 struct swap_info_struct * p = NULL;
1209 unsigned short *swap_map;
1210 struct file *swap_file, *victim;
1211 struct address_space *mapping;
1212 struct inode *inode;
1217 if (!capable(CAP_SYS_ADMIN))
1220 pathname = getname(specialfile);
1221 err = PTR_ERR(pathname);
1222 if (IS_ERR(pathname))
1225 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1227 err = PTR_ERR(victim);
1231 mapping = victim->f_mapping;
1233 spin_lock(&swap_lock);
1234 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1235 p = swap_info + type;
1236 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1237 if (p->swap_file->f_mapping == mapping)
1244 spin_unlock(&swap_lock);
1247 if (!security_vm_enough_memory(p->pages))
1248 vm_unacct_memory(p->pages);
1251 spin_unlock(&swap_lock);
1255 swap_list.head = p->next;
1257 swap_info[prev].next = p->next;
1259 if (type == swap_list.next) {
1260 /* just pick something that's safe... */
1261 swap_list.next = swap_list.head;
1263 nr_swap_pages -= p->pages;
1264 total_swap_pages -= p->pages;
1265 p->flags &= ~SWP_WRITEOK;
1266 spin_unlock(&swap_lock);
1268 current->flags |= PF_SWAPOFF;
1269 err = try_to_unuse(type);
1270 current->flags &= ~PF_SWAPOFF;
1273 /* re-insert swap space back into swap_list */
1274 spin_lock(&swap_lock);
1275 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1276 if (p->prio >= swap_info[i].prio)
1280 swap_list.head = swap_list.next = p - swap_info;
1282 swap_info[prev].next = p - swap_info;
1283 nr_swap_pages += p->pages;
1284 total_swap_pages += p->pages;
1285 p->flags |= SWP_WRITEOK;
1286 spin_unlock(&swap_lock);
1290 /* wait for any unplug function to finish */
1291 down_write(&swap_unplug_sem);
1292 up_write(&swap_unplug_sem);
1294 destroy_swap_extents(p);
1295 mutex_lock(&swapon_mutex);
1296 spin_lock(&swap_lock);
1299 /* wait for anyone still in scan_swap_map */
1300 p->highest_bit = 0; /* cuts scans short */
1301 while (p->flags >= SWP_SCANNING) {
1302 spin_unlock(&swap_lock);
1303 schedule_timeout_uninterruptible(1);
1304 spin_lock(&swap_lock);
1307 swap_file = p->swap_file;
1308 p->swap_file = NULL;
1310 swap_map = p->swap_map;
1313 spin_unlock(&swap_lock);
1314 mutex_unlock(&swapon_mutex);
1316 inode = mapping->host;
1317 if (S_ISBLK(inode->i_mode)) {
1318 struct block_device *bdev = I_BDEV(inode);
1319 set_blocksize(bdev, p->old_block_size);
1322 mutex_lock(&inode->i_mutex);
1323 inode->i_flags &= ~S_SWAPFILE;
1324 mutex_unlock(&inode->i_mutex);
1326 filp_close(swap_file, NULL);
1330 filp_close(victim, NULL);
1335 #ifdef CONFIG_PROC_FS
1337 static void *swap_start(struct seq_file *swap, loff_t *pos)
1339 struct swap_info_struct *ptr = swap_info;
1343 mutex_lock(&swapon_mutex);
1346 return SEQ_START_TOKEN;
1348 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1349 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1358 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1360 struct swap_info_struct *ptr;
1361 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1363 if (v == SEQ_START_TOKEN)
1370 for (; ptr < endptr; ptr++) {
1371 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1380 static void swap_stop(struct seq_file *swap, void *v)
1382 mutex_unlock(&swapon_mutex);
1385 static int swap_show(struct seq_file *swap, void *v)
1387 struct swap_info_struct *ptr = v;
1391 if (ptr == SEQ_START_TOKEN) {
1392 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1396 file = ptr->swap_file;
1397 len = seq_path(swap, &file->f_path, " \t\n\\");
1398 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1399 len < 40 ? 40 - len : 1, " ",
1400 S_ISBLK(file->f_path.dentry->d_inode->i_mode) ?
1401 "partition" : "file\t",
1402 ptr->pages << (PAGE_SHIFT - 10),
1403 ptr->inuse_pages << (PAGE_SHIFT - 10),
1408 static const struct seq_operations swaps_op = {
1409 .start = swap_start,
1415 static int swaps_open(struct inode *inode, struct file *file)
1417 return seq_open(file, &swaps_op);
1420 static const struct file_operations proc_swaps_operations = {
1423 .llseek = seq_lseek,
1424 .release = seq_release,
1427 static int __init procswaps_init(void)
1429 proc_create("swaps", 0, NULL, &proc_swaps_operations);
1432 __initcall(procswaps_init);
1433 #endif /* CONFIG_PROC_FS */
1436 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1438 * The swapon system call
1440 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1442 struct swap_info_struct * p;
1444 struct block_device *bdev = NULL;
1445 struct file *swap_file = NULL;
1446 struct address_space *mapping;
1450 static int least_priority;
1451 union swap_header *swap_header = NULL;
1452 int swap_header_version;
1453 unsigned int nr_good_pages = 0;
1456 unsigned long maxpages = 1;
1458 unsigned short *swap_map;
1459 struct page *page = NULL;
1460 struct inode *inode = NULL;
1463 if (!capable(CAP_SYS_ADMIN))
1465 spin_lock(&swap_lock);
1467 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1468 if (!(p->flags & SWP_USED))
1471 if (type >= MAX_SWAPFILES) {
1472 spin_unlock(&swap_lock);
1475 if (type >= nr_swapfiles)
1476 nr_swapfiles = type+1;
1477 INIT_LIST_HEAD(&p->extent_list);
1478 p->flags = SWP_USED;
1479 p->swap_file = NULL;
1480 p->old_block_size = 0;
1487 if (swap_flags & SWAP_FLAG_PREFER) {
1489 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1491 p->prio = --least_priority;
1493 spin_unlock(&swap_lock);
1494 name = getname(specialfile);
1495 error = PTR_ERR(name);
1500 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1501 error = PTR_ERR(swap_file);
1502 if (IS_ERR(swap_file)) {
1507 p->swap_file = swap_file;
1508 mapping = swap_file->f_mapping;
1509 inode = mapping->host;
1512 for (i = 0; i < nr_swapfiles; i++) {
1513 struct swap_info_struct *q = &swap_info[i];
1515 if (i == type || !q->swap_file)
1517 if (mapping == q->swap_file->f_mapping)
1522 if (S_ISBLK(inode->i_mode)) {
1523 bdev = I_BDEV(inode);
1524 error = bd_claim(bdev, sys_swapon);
1530 p->old_block_size = block_size(bdev);
1531 error = set_blocksize(bdev, PAGE_SIZE);
1535 } else if (S_ISREG(inode->i_mode)) {
1536 p->bdev = inode->i_sb->s_bdev;
1537 mutex_lock(&inode->i_mutex);
1539 if (IS_SWAPFILE(inode)) {
1547 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1550 * Read the swap header.
1552 if (!mapping->a_ops->readpage) {
1556 page = read_mapping_page(mapping, 0, swap_file);
1558 error = PTR_ERR(page);
1562 swap_header = page_address(page);
1564 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1565 swap_header_version = 1;
1566 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1567 swap_header_version = 2;
1569 printk(KERN_ERR "Unable to find swap-space signature\n");
1574 switch (swap_header_version) {
1576 printk(KERN_ERR "version 0 swap is no longer supported. "
1577 "Use mkswap -v1 %s\n", name);
1581 /* swap partition endianess hack... */
1582 if (swab32(swap_header->info.version) == 1) {
1583 swab32s(&swap_header->info.version);
1584 swab32s(&swap_header->info.last_page);
1585 swab32s(&swap_header->info.nr_badpages);
1586 for (i = 0; i < swap_header->info.nr_badpages; i++)
1587 swab32s(&swap_header->info.badpages[i]);
1589 /* Check the swap header's sub-version and the size of
1590 the swap file and bad block lists */
1591 if (swap_header->info.version != 1) {
1593 "Unable to handle swap header version %d\n",
1594 swap_header->info.version);
1600 p->cluster_next = 1;
1603 * Find out how many pages are allowed for a single swap
1604 * device. There are two limiting factors: 1) the number of
1605 * bits for the swap offset in the swp_entry_t type and
1606 * 2) the number of bits in the a swap pte as defined by
1607 * the different architectures. In order to find the
1608 * largest possible bit mask a swap entry with swap type 0
1609 * and swap offset ~0UL is created, encoded to a swap pte,
1610 * decoded to a swp_entry_t again and finally the swap
1611 * offset is extracted. This will mask all the bits from
1612 * the initial ~0UL mask that can't be encoded in either
1613 * the swp_entry_t or the architecture definition of a
1616 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1617 if (maxpages > swap_header->info.last_page)
1618 maxpages = swap_header->info.last_page;
1619 p->highest_bit = maxpages - 1;
1624 if (swapfilesize && maxpages > swapfilesize) {
1626 "Swap area shorter than signature indicates\n");
1629 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1631 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1634 /* OK, set up the swap map and apply the bad block list */
1635 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1641 memset(p->swap_map, 0, maxpages * sizeof(short));
1642 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1643 int page_nr = swap_header->info.badpages[i];
1644 if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
1647 p->swap_map[page_nr] = SWAP_MAP_BAD;
1649 nr_good_pages = swap_header->info.last_page -
1650 swap_header->info.nr_badpages -
1651 1 /* header page */;
1656 if (nr_good_pages) {
1657 p->swap_map[0] = SWAP_MAP_BAD;
1659 p->pages = nr_good_pages;
1660 nr_extents = setup_swap_extents(p, &span);
1661 if (nr_extents < 0) {
1665 nr_good_pages = p->pages;
1667 if (!nr_good_pages) {
1668 printk(KERN_WARNING "Empty swap-file\n");
1673 mutex_lock(&swapon_mutex);
1674 spin_lock(&swap_lock);
1675 p->flags = SWP_ACTIVE;
1676 nr_swap_pages += nr_good_pages;
1677 total_swap_pages += nr_good_pages;
1679 printk(KERN_INFO "Adding %uk swap on %s. "
1680 "Priority:%d extents:%d across:%lluk\n",
1681 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1682 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
1684 /* insert swap space into swap_list: */
1686 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1687 if (p->prio >= swap_info[i].prio) {
1694 swap_list.head = swap_list.next = p - swap_info;
1696 swap_info[prev].next = p - swap_info;
1698 spin_unlock(&swap_lock);
1699 mutex_unlock(&swapon_mutex);
1704 set_blocksize(bdev, p->old_block_size);
1707 destroy_swap_extents(p);
1709 spin_lock(&swap_lock);
1710 swap_map = p->swap_map;
1711 p->swap_file = NULL;
1714 if (!(swap_flags & SWAP_FLAG_PREFER))
1716 spin_unlock(&swap_lock);
1719 filp_close(swap_file, NULL);
1721 if (page && !IS_ERR(page)) {
1723 page_cache_release(page);
1729 inode->i_flags |= S_SWAPFILE;
1730 mutex_unlock(&inode->i_mutex);
1735 void si_swapinfo(struct sysinfo *val)
1738 unsigned long nr_to_be_unused = 0;
1740 spin_lock(&swap_lock);
1741 for (i = 0; i < nr_swapfiles; i++) {
1742 if (!(swap_info[i].flags & SWP_USED) ||
1743 (swap_info[i].flags & SWP_WRITEOK))
1745 nr_to_be_unused += swap_info[i].inuse_pages;
1747 val->freeswap = nr_swap_pages + nr_to_be_unused;
1748 val->totalswap = total_swap_pages + nr_to_be_unused;
1749 spin_unlock(&swap_lock);
1753 * Verify that a swap entry is valid and increment its swap map count.
1755 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1756 * "permanent", but will be reclaimed by the next swapoff.
1758 int swap_duplicate(swp_entry_t entry)
1760 struct swap_info_struct * p;
1761 unsigned long offset, type;
1764 if (is_migration_entry(entry))
1767 type = swp_type(entry);
1768 if (type >= nr_swapfiles)
1770 p = type + swap_info;
1771 offset = swp_offset(entry);
1773 spin_lock(&swap_lock);
1774 if (offset < p->max && p->swap_map[offset]) {
1775 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1776 p->swap_map[offset]++;
1778 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1779 if (swap_overflow++ < 5)
1780 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1781 p->swap_map[offset] = SWAP_MAP_MAX;
1785 spin_unlock(&swap_lock);
1790 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1794 struct swap_info_struct *
1795 get_swap_info_struct(unsigned type)
1797 return &swap_info[type];
1801 * swap_lock prevents swap_map being freed. Don't grab an extra
1802 * reference on the swaphandle, it doesn't matter if it becomes unused.
1804 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1806 struct swap_info_struct *si;
1807 int our_page_cluster = page_cluster;
1808 pgoff_t target, toff;
1812 if (!our_page_cluster) /* no readahead */
1815 si = &swap_info[swp_type(entry)];
1816 target = swp_offset(entry);
1817 base = (target >> our_page_cluster) << our_page_cluster;
1818 end = base + (1 << our_page_cluster);
1819 if (!base) /* first page is swap header */
1822 spin_lock(&swap_lock);
1823 if (end > si->max) /* don't go beyond end of map */
1826 /* Count contiguous allocated slots above our target */
1827 for (toff = target; ++toff < end; nr_pages++) {
1828 /* Don't read in free or bad pages */
1829 if (!si->swap_map[toff])
1831 if (si->swap_map[toff] == SWAP_MAP_BAD)
1834 /* Count contiguous allocated slots below our target */
1835 for (toff = target; --toff >= base; nr_pages++) {
1836 /* Don't read in free or bad pages */
1837 if (!si->swap_map[toff])
1839 if (si->swap_map[toff] == SWAP_MAP_BAD)
1842 spin_unlock(&swap_lock);
1845 * Indicate starting offset, and return number of pages to get:
1846 * if only 1, say 0, since there's then no readahead to be done.
1849 return nr_pages? ++nr_pages: 0;