4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
8 #include <linux/config.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>
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_SOFTWARE_SUSPEND
431 * Find the swap type that corresponds to given device (if any)
433 * This is needed for software suspend and is done in such a way that inode
434 * aliasing is allowed.
436 int swap_type_of(dev_t device)
440 spin_lock(&swap_lock);
441 for (i = 0; i < nr_swapfiles; i++) {
444 if (!(swap_info[i].flags & SWP_WRITEOK))
447 spin_unlock(&swap_lock);
450 inode = swap_info->swap_file->f_dentry->d_inode;
451 if (S_ISBLK(inode->i_mode) &&
452 device == MKDEV(imajor(inode), iminor(inode))) {
453 spin_unlock(&swap_lock);
457 spin_unlock(&swap_lock);
462 * Return either the total number of swap pages of given type, or the number
463 * of free pages of that type (depending on @free)
465 * This is needed for software suspend
467 unsigned int count_swap_pages(int type, int free)
471 if (type < nr_swapfiles) {
472 spin_lock(&swap_lock);
473 if (swap_info[type].flags & SWP_WRITEOK) {
474 n = swap_info[type].pages;
476 n -= swap_info[type].inuse_pages;
478 spin_unlock(&swap_lock);
485 * No need to decide whether this PTE shares the swap entry with others,
486 * just let do_wp_page work it out if a write is requested later - to
487 * force COW, vm_page_prot omits write permission from any private vma.
489 static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
490 unsigned long addr, swp_entry_t entry, struct page *page)
492 inc_mm_counter(vma->vm_mm, anon_rss);
494 set_pte_at(vma->vm_mm, addr, pte,
495 pte_mkold(mk_pte(page, vma->vm_page_prot)));
496 page_add_anon_rmap(page, vma, addr);
499 * Move the page to the active list so it is not
500 * immediately swapped out again after swapon.
505 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
506 unsigned long addr, unsigned long end,
507 swp_entry_t entry, struct page *page)
509 pte_t swp_pte = swp_entry_to_pte(entry);
514 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
517 * swapoff spends a _lot_ of time in this loop!
518 * Test inline before going to call unuse_pte.
520 if (unlikely(pte_same(*pte, swp_pte))) {
521 unuse_pte(vma, pte++, addr, entry, page);
525 } while (pte++, addr += PAGE_SIZE, addr != end);
526 pte_unmap_unlock(pte - 1, ptl);
530 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
531 unsigned long addr, unsigned long end,
532 swp_entry_t entry, struct page *page)
537 pmd = pmd_offset(pud, addr);
539 next = pmd_addr_end(addr, end);
540 if (pmd_none_or_clear_bad(pmd))
542 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
544 } while (pmd++, addr = next, addr != end);
548 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
549 unsigned long addr, unsigned long end,
550 swp_entry_t entry, struct page *page)
555 pud = pud_offset(pgd, addr);
557 next = pud_addr_end(addr, end);
558 if (pud_none_or_clear_bad(pud))
560 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
562 } while (pud++, addr = next, addr != end);
566 static int unuse_vma(struct vm_area_struct *vma,
567 swp_entry_t entry, struct page *page)
570 unsigned long addr, end, next;
573 addr = page_address_in_vma(page, vma);
577 end = addr + PAGE_SIZE;
579 addr = vma->vm_start;
583 pgd = pgd_offset(vma->vm_mm, addr);
585 next = pgd_addr_end(addr, end);
586 if (pgd_none_or_clear_bad(pgd))
588 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
590 } while (pgd++, addr = next, addr != end);
594 static int unuse_mm(struct mm_struct *mm,
595 swp_entry_t entry, struct page *page)
597 struct vm_area_struct *vma;
599 if (!down_read_trylock(&mm->mmap_sem)) {
601 * Activate page so shrink_cache is unlikely to unmap its
602 * ptes while lock is dropped, so swapoff can make progress.
606 down_read(&mm->mmap_sem);
609 for (vma = mm->mmap; vma; vma = vma->vm_next) {
610 if (vma->anon_vma && unuse_vma(vma, entry, page))
613 up_read(&mm->mmap_sem);
615 * Currently unuse_mm cannot fail, but leave error handling
616 * at call sites for now, since we change it from time to time.
621 #ifdef CONFIG_MIGRATION
622 int remove_vma_swap(struct vm_area_struct *vma, struct page *page)
624 swp_entry_t entry = { .val = page_private(page) };
626 return unuse_vma(vma, entry, page);
631 * Scan swap_map from current position to next entry still in use.
632 * Recycle to start on reaching the end, returning 0 when empty.
634 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
637 unsigned int max = si->max;
638 unsigned int i = prev;
642 * No need for swap_lock here: we're just looking
643 * for whether an entry is in use, not modifying it; false
644 * hits are okay, and sys_swapoff() has already prevented new
645 * allocations from this area (while holding swap_lock).
654 * No entries in use at top of swap_map,
655 * loop back to start and recheck there.
661 count = si->swap_map[i];
662 if (count && count != SWAP_MAP_BAD)
669 * We completely avoid races by reading each swap page in advance,
670 * and then search for the process using it. All the necessary
671 * page table adjustments can then be made atomically.
673 static int try_to_unuse(unsigned int type)
675 struct swap_info_struct * si = &swap_info[type];
676 struct mm_struct *start_mm;
677 unsigned short *swap_map;
678 unsigned short swcount;
683 int reset_overflow = 0;
687 * When searching mms for an entry, a good strategy is to
688 * start at the first mm we freed the previous entry from
689 * (though actually we don't notice whether we or coincidence
690 * freed the entry). Initialize this start_mm with a hold.
692 * A simpler strategy would be to start at the last mm we
693 * freed the previous entry from; but that would take less
694 * advantage of mmlist ordering, which clusters forked mms
695 * together, child after parent. If we race with dup_mmap(), we
696 * prefer to resolve parent before child, lest we miss entries
697 * duplicated after we scanned child: using last mm would invert
698 * that. Though it's only a serious concern when an overflowed
699 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
702 atomic_inc(&init_mm.mm_users);
705 * Keep on scanning until all entries have gone. Usually,
706 * one pass through swap_map is enough, but not necessarily:
707 * there are races when an instance of an entry might be missed.
709 while ((i = find_next_to_unuse(si, i)) != 0) {
710 if (signal_pending(current)) {
716 * Get a page for the entry, using the existing swap
717 * cache page if there is one. Otherwise, get a clean
718 * page and read the swap into it.
720 swap_map = &si->swap_map[i];
721 entry = swp_entry(type, i);
722 page = read_swap_cache_async(entry, NULL, 0);
725 * Either swap_duplicate() failed because entry
726 * has been freed independently, and will not be
727 * reused since sys_swapoff() already disabled
728 * allocation from here, or alloc_page() failed.
737 * Don't hold on to start_mm if it looks like exiting.
739 if (atomic_read(&start_mm->mm_users) == 1) {
742 atomic_inc(&init_mm.mm_users);
746 * Wait for and lock page. When do_swap_page races with
747 * try_to_unuse, do_swap_page can handle the fault much
748 * faster than try_to_unuse can locate the entry. This
749 * apparently redundant "wait_on_page_locked" lets try_to_unuse
750 * defer to do_swap_page in such a case - in some tests,
751 * do_swap_page and try_to_unuse repeatedly compete.
753 wait_on_page_locked(page);
754 wait_on_page_writeback(page);
756 wait_on_page_writeback(page);
759 * Remove all references to entry.
760 * Whenever we reach init_mm, there's no address space
761 * to search, but use it as a reminder to search shmem.
766 if (start_mm == &init_mm)
767 shmem = shmem_unuse(entry, page);
769 retval = unuse_mm(start_mm, entry, page);
772 int set_start_mm = (*swap_map >= swcount);
773 struct list_head *p = &start_mm->mmlist;
774 struct mm_struct *new_start_mm = start_mm;
775 struct mm_struct *prev_mm = start_mm;
776 struct mm_struct *mm;
778 atomic_inc(&new_start_mm->mm_users);
779 atomic_inc(&prev_mm->mm_users);
780 spin_lock(&mmlist_lock);
781 while (*swap_map > 1 && !retval &&
782 (p = p->next) != &start_mm->mmlist) {
783 mm = list_entry(p, struct mm_struct, mmlist);
784 if (atomic_inc_return(&mm->mm_users) == 1) {
785 atomic_dec(&mm->mm_users);
788 spin_unlock(&mmlist_lock);
797 else if (mm == &init_mm) {
799 shmem = shmem_unuse(entry, page);
801 retval = unuse_mm(mm, entry, page);
802 if (set_start_mm && *swap_map < swcount) {
804 atomic_inc(&mm->mm_users);
808 spin_lock(&mmlist_lock);
810 spin_unlock(&mmlist_lock);
813 start_mm = new_start_mm;
817 page_cache_release(page);
822 * How could swap count reach 0x7fff when the maximum
823 * pid is 0x7fff, and there's no way to repeat a swap
824 * page within an mm (except in shmem, where it's the
825 * shared object which takes the reference count)?
826 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
828 * If that's wrong, then we should worry more about
829 * exit_mmap() and do_munmap() cases described above:
830 * we might be resetting SWAP_MAP_MAX too early here.
831 * We know "Undead"s can happen, they're okay, so don't
832 * report them; but do report if we reset SWAP_MAP_MAX.
834 if (*swap_map == SWAP_MAP_MAX) {
835 spin_lock(&swap_lock);
837 spin_unlock(&swap_lock);
842 * If a reference remains (rare), we would like to leave
843 * the page in the swap cache; but try_to_unmap could
844 * then re-duplicate the entry once we drop page lock,
845 * so we might loop indefinitely; also, that page could
846 * not be swapped out to other storage meanwhile. So:
847 * delete from cache even if there's another reference,
848 * after ensuring that the data has been saved to disk -
849 * since if the reference remains (rarer), it will be
850 * read from disk into another page. Splitting into two
851 * pages would be incorrect if swap supported "shared
852 * private" pages, but they are handled by tmpfs files.
854 * Note shmem_unuse already deleted a swappage from
855 * the swap cache, unless the move to filepage failed:
856 * in which case it left swappage in cache, lowered its
857 * swap count to pass quickly through the loops above,
858 * and now we must reincrement count to try again later.
860 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
861 struct writeback_control wbc = {
862 .sync_mode = WB_SYNC_NONE,
865 swap_writepage(page, &wbc);
867 wait_on_page_writeback(page);
869 if (PageSwapCache(page)) {
871 swap_duplicate(entry);
873 delete_from_swap_cache(page);
877 * So we could skip searching mms once swap count went
878 * to 1, we did not mark any present ptes as dirty: must
879 * mark page dirty so shrink_list will preserve it.
883 page_cache_release(page);
886 * Make sure that we aren't completely killing
887 * interactive performance.
893 if (reset_overflow) {
894 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
901 * After a successful try_to_unuse, if no swap is now in use, we know
902 * we can empty the mmlist. swap_lock must be held on entry and exit.
903 * Note that mmlist_lock nests inside swap_lock, and an mm must be
904 * added to the mmlist just after page_duplicate - before would be racy.
906 static void drain_mmlist(void)
908 struct list_head *p, *next;
911 for (i = 0; i < nr_swapfiles; i++)
912 if (swap_info[i].inuse_pages)
914 spin_lock(&mmlist_lock);
915 list_for_each_safe(p, next, &init_mm.mmlist)
917 spin_unlock(&mmlist_lock);
921 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
922 * corresponds to page offset `offset'.
924 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
926 struct swap_extent *se = sis->curr_swap_extent;
927 struct swap_extent *start_se = se;
930 struct list_head *lh;
932 if (se->start_page <= offset &&
933 offset < (se->start_page + se->nr_pages)) {
934 return se->start_block + (offset - se->start_page);
937 if (lh == &sis->extent_list)
939 se = list_entry(lh, struct swap_extent, list);
940 sis->curr_swap_extent = se;
941 BUG_ON(se == start_se); /* It *must* be present */
946 * Free all of a swapdev's extent information
948 static void destroy_swap_extents(struct swap_info_struct *sis)
950 while (!list_empty(&sis->extent_list)) {
951 struct swap_extent *se;
953 se = list_entry(sis->extent_list.next,
954 struct swap_extent, list);
961 * Add a block range (and the corresponding page range) into this swapdev's
962 * extent list. The extent list is kept sorted in page order.
964 * This function rather assumes that it is called in ascending page order.
967 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
968 unsigned long nr_pages, sector_t start_block)
970 struct swap_extent *se;
971 struct swap_extent *new_se;
972 struct list_head *lh;
974 lh = sis->extent_list.prev; /* The highest page extent */
975 if (lh != &sis->extent_list) {
976 se = list_entry(lh, struct swap_extent, list);
977 BUG_ON(se->start_page + se->nr_pages != start_page);
978 if (se->start_block + se->nr_pages == start_block) {
980 se->nr_pages += nr_pages;
986 * No merge. Insert a new extent, preserving ordering.
988 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
991 new_se->start_page = start_page;
992 new_se->nr_pages = nr_pages;
993 new_se->start_block = start_block;
995 list_add_tail(&new_se->list, &sis->extent_list);
1000 * A `swap extent' is a simple thing which maps a contiguous range of pages
1001 * onto a contiguous range of disk blocks. An ordered list of swap extents
1002 * is built at swapon time and is then used at swap_writepage/swap_readpage
1003 * time for locating where on disk a page belongs.
1005 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1006 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1007 * swap files identically.
1009 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1010 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1011 * swapfiles are handled *identically* after swapon time.
1013 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1014 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1015 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1016 * requirements, they are simply tossed out - we will never use those blocks
1019 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1020 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1021 * which will scribble on the fs.
1023 * The amount of disk space which a single swap extent represents varies.
1024 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1025 * extents in the list. To avoid much list walking, we cache the previous
1026 * search location in `curr_swap_extent', and start new searches from there.
1027 * This is extremely effective. The average number of iterations in
1028 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1030 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1032 struct inode *inode;
1033 unsigned blocks_per_page;
1034 unsigned long page_no;
1036 sector_t probe_block;
1037 sector_t last_block;
1038 sector_t lowest_block = -1;
1039 sector_t highest_block = 0;
1043 inode = sis->swap_file->f_mapping->host;
1044 if (S_ISBLK(inode->i_mode)) {
1045 ret = add_swap_extent(sis, 0, sis->max, 0);
1050 blkbits = inode->i_blkbits;
1051 blocks_per_page = PAGE_SIZE >> blkbits;
1054 * Map all the blocks into the extent list. This code doesn't try
1059 last_block = i_size_read(inode) >> blkbits;
1060 while ((probe_block + blocks_per_page) <= last_block &&
1061 page_no < sis->max) {
1062 unsigned block_in_page;
1063 sector_t first_block;
1065 first_block = bmap(inode, probe_block);
1066 if (first_block == 0)
1070 * It must be PAGE_SIZE aligned on-disk
1072 if (first_block & (blocks_per_page - 1)) {
1077 for (block_in_page = 1; block_in_page < blocks_per_page;
1081 block = bmap(inode, probe_block + block_in_page);
1084 if (block != first_block + block_in_page) {
1091 first_block >>= (PAGE_SHIFT - blkbits);
1092 if (page_no) { /* exclude the header page */
1093 if (first_block < lowest_block)
1094 lowest_block = first_block;
1095 if (first_block > highest_block)
1096 highest_block = first_block;
1100 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1102 ret = add_swap_extent(sis, page_no, 1, first_block);
1107 probe_block += blocks_per_page;
1112 *span = 1 + highest_block - lowest_block;
1114 page_no = 1; /* force Empty message */
1116 sis->pages = page_no - 1;
1117 sis->highest_bit = page_no - 1;
1119 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1120 struct swap_extent, list);
1123 printk(KERN_ERR "swapon: swapfile has holes\n");
1129 #if 0 /* We don't need this yet */
1130 #include <linux/backing-dev.h>
1131 int page_queue_congested(struct page *page)
1133 struct backing_dev_info *bdi;
1135 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1137 if (PageSwapCache(page)) {
1138 swp_entry_t entry = { .val = page_private(page) };
1139 struct swap_info_struct *sis;
1141 sis = get_swap_info_struct(swp_type(entry));
1142 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1144 bdi = page->mapping->backing_dev_info;
1145 return bdi_write_congested(bdi);
1149 asmlinkage long sys_swapoff(const char __user * specialfile)
1151 struct swap_info_struct * p = NULL;
1152 unsigned short *swap_map;
1153 struct file *swap_file, *victim;
1154 struct address_space *mapping;
1155 struct inode *inode;
1160 if (!capable(CAP_SYS_ADMIN))
1163 pathname = getname(specialfile);
1164 err = PTR_ERR(pathname);
1165 if (IS_ERR(pathname))
1168 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1170 err = PTR_ERR(victim);
1174 mapping = victim->f_mapping;
1176 spin_lock(&swap_lock);
1177 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1178 p = swap_info + type;
1179 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1180 if (p->swap_file->f_mapping == mapping)
1187 spin_unlock(&swap_lock);
1190 if (!security_vm_enough_memory(p->pages))
1191 vm_unacct_memory(p->pages);
1194 spin_unlock(&swap_lock);
1198 swap_list.head = p->next;
1200 swap_info[prev].next = p->next;
1202 if (type == swap_list.next) {
1203 /* just pick something that's safe... */
1204 swap_list.next = swap_list.head;
1206 nr_swap_pages -= p->pages;
1207 total_swap_pages -= p->pages;
1208 p->flags &= ~SWP_WRITEOK;
1209 spin_unlock(&swap_lock);
1211 current->flags |= PF_SWAPOFF;
1212 err = try_to_unuse(type);
1213 current->flags &= ~PF_SWAPOFF;
1216 /* re-insert swap space back into swap_list */
1217 spin_lock(&swap_lock);
1218 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1219 if (p->prio >= swap_info[i].prio)
1223 swap_list.head = swap_list.next = p - swap_info;
1225 swap_info[prev].next = p - swap_info;
1226 nr_swap_pages += p->pages;
1227 total_swap_pages += p->pages;
1228 p->flags |= SWP_WRITEOK;
1229 spin_unlock(&swap_lock);
1233 /* wait for any unplug function to finish */
1234 down_write(&swap_unplug_sem);
1235 up_write(&swap_unplug_sem);
1237 destroy_swap_extents(p);
1238 mutex_lock(&swapon_mutex);
1239 spin_lock(&swap_lock);
1242 /* wait for anyone still in scan_swap_map */
1243 p->highest_bit = 0; /* cuts scans short */
1244 while (p->flags >= SWP_SCANNING) {
1245 spin_unlock(&swap_lock);
1246 schedule_timeout_uninterruptible(1);
1247 spin_lock(&swap_lock);
1250 swap_file = p->swap_file;
1251 p->swap_file = NULL;
1253 swap_map = p->swap_map;
1256 spin_unlock(&swap_lock);
1257 mutex_unlock(&swapon_mutex);
1259 inode = mapping->host;
1260 if (S_ISBLK(inode->i_mode)) {
1261 struct block_device *bdev = I_BDEV(inode);
1262 set_blocksize(bdev, p->old_block_size);
1265 mutex_lock(&inode->i_mutex);
1266 inode->i_flags &= ~S_SWAPFILE;
1267 mutex_unlock(&inode->i_mutex);
1269 filp_close(swap_file, NULL);
1273 filp_close(victim, NULL);
1278 #ifdef CONFIG_PROC_FS
1280 static void *swap_start(struct seq_file *swap, loff_t *pos)
1282 struct swap_info_struct *ptr = swap_info;
1286 mutex_lock(&swapon_mutex);
1288 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1289 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1298 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1300 struct swap_info_struct *ptr = v;
1301 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1303 for (++ptr; ptr < endptr; ptr++) {
1304 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1313 static void swap_stop(struct seq_file *swap, void *v)
1315 mutex_unlock(&swapon_mutex);
1318 static int swap_show(struct seq_file *swap, void *v)
1320 struct swap_info_struct *ptr = v;
1325 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1327 file = ptr->swap_file;
1328 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1329 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1330 len < 40 ? 40 - len : 1, " ",
1331 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1332 "partition" : "file\t",
1333 ptr->pages << (PAGE_SHIFT - 10),
1334 ptr->inuse_pages << (PAGE_SHIFT - 10),
1339 static struct seq_operations swaps_op = {
1340 .start = swap_start,
1346 static int swaps_open(struct inode *inode, struct file *file)
1348 return seq_open(file, &swaps_op);
1351 static struct file_operations proc_swaps_operations = {
1354 .llseek = seq_lseek,
1355 .release = seq_release,
1358 static int __init procswaps_init(void)
1360 struct proc_dir_entry *entry;
1362 entry = create_proc_entry("swaps", 0, NULL);
1364 entry->proc_fops = &proc_swaps_operations;
1367 __initcall(procswaps_init);
1368 #endif /* CONFIG_PROC_FS */
1371 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1373 * The swapon system call
1375 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1377 struct swap_info_struct * p;
1379 struct block_device *bdev = NULL;
1380 struct file *swap_file = NULL;
1381 struct address_space *mapping;
1385 static int least_priority;
1386 union swap_header *swap_header = NULL;
1387 int swap_header_version;
1388 unsigned int nr_good_pages = 0;
1391 unsigned long maxpages = 1;
1393 unsigned short *swap_map;
1394 struct page *page = NULL;
1395 struct inode *inode = NULL;
1398 if (!capable(CAP_SYS_ADMIN))
1400 spin_lock(&swap_lock);
1402 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1403 if (!(p->flags & SWP_USED))
1406 if (type >= MAX_SWAPFILES) {
1407 spin_unlock(&swap_lock);
1410 if (type >= nr_swapfiles)
1411 nr_swapfiles = type+1;
1412 INIT_LIST_HEAD(&p->extent_list);
1413 p->flags = SWP_USED;
1414 p->swap_file = NULL;
1415 p->old_block_size = 0;
1422 if (swap_flags & SWAP_FLAG_PREFER) {
1424 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1426 p->prio = --least_priority;
1428 spin_unlock(&swap_lock);
1429 name = getname(specialfile);
1430 error = PTR_ERR(name);
1435 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1436 error = PTR_ERR(swap_file);
1437 if (IS_ERR(swap_file)) {
1442 p->swap_file = swap_file;
1443 mapping = swap_file->f_mapping;
1444 inode = mapping->host;
1447 for (i = 0; i < nr_swapfiles; i++) {
1448 struct swap_info_struct *q = &swap_info[i];
1450 if (i == type || !q->swap_file)
1452 if (mapping == q->swap_file->f_mapping)
1457 if (S_ISBLK(inode->i_mode)) {
1458 bdev = I_BDEV(inode);
1459 error = bd_claim(bdev, sys_swapon);
1465 p->old_block_size = block_size(bdev);
1466 error = set_blocksize(bdev, PAGE_SIZE);
1470 } else if (S_ISREG(inode->i_mode)) {
1471 p->bdev = inode->i_sb->s_bdev;
1472 mutex_lock(&inode->i_mutex);
1474 if (IS_SWAPFILE(inode)) {
1482 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1485 * Read the swap header.
1487 if (!mapping->a_ops->readpage) {
1491 page = read_cache_page(mapping, 0,
1492 (filler_t *)mapping->a_ops->readpage, swap_file);
1494 error = PTR_ERR(page);
1497 wait_on_page_locked(page);
1498 if (!PageUptodate(page))
1501 swap_header = page_address(page);
1503 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1504 swap_header_version = 1;
1505 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1506 swap_header_version = 2;
1508 printk(KERN_ERR "Unable to find swap-space signature\n");
1513 switch (swap_header_version) {
1515 printk(KERN_ERR "version 0 swap is no longer supported. "
1516 "Use mkswap -v1 %s\n", name);
1520 /* Check the swap header's sub-version and the size of
1521 the swap file and bad block lists */
1522 if (swap_header->info.version != 1) {
1524 "Unable to handle swap header version %d\n",
1525 swap_header->info.version);
1531 p->cluster_next = 1;
1534 * Find out how many pages are allowed for a single swap
1535 * device. There are two limiting factors: 1) the number of
1536 * bits for the swap offset in the swp_entry_t type and
1537 * 2) the number of bits in the a swap pte as defined by
1538 * the different architectures. In order to find the
1539 * largest possible bit mask a swap entry with swap type 0
1540 * and swap offset ~0UL is created, encoded to a swap pte,
1541 * decoded to a swp_entry_t again and finally the swap
1542 * offset is extracted. This will mask all the bits from
1543 * the initial ~0UL mask that can't be encoded in either
1544 * the swp_entry_t or the architecture definition of a
1547 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1548 if (maxpages > swap_header->info.last_page)
1549 maxpages = swap_header->info.last_page;
1550 p->highest_bit = maxpages - 1;
1555 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1557 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1560 /* OK, set up the swap map and apply the bad block list */
1561 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1567 memset(p->swap_map, 0, maxpages * sizeof(short));
1568 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1569 int page_nr = swap_header->info.badpages[i];
1570 if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
1573 p->swap_map[page_nr] = SWAP_MAP_BAD;
1575 nr_good_pages = swap_header->info.last_page -
1576 swap_header->info.nr_badpages -
1577 1 /* header page */;
1582 if (swapfilesize && maxpages > swapfilesize) {
1584 "Swap area shorter than signature indicates\n");
1588 if (nr_good_pages) {
1589 p->swap_map[0] = SWAP_MAP_BAD;
1591 p->pages = nr_good_pages;
1592 nr_extents = setup_swap_extents(p, &span);
1593 if (nr_extents < 0) {
1597 nr_good_pages = p->pages;
1599 if (!nr_good_pages) {
1600 printk(KERN_WARNING "Empty swap-file\n");
1605 mutex_lock(&swapon_mutex);
1606 spin_lock(&swap_lock);
1607 p->flags = SWP_ACTIVE;
1608 nr_swap_pages += nr_good_pages;
1609 total_swap_pages += nr_good_pages;
1611 printk(KERN_INFO "Adding %uk swap on %s. "
1612 "Priority:%d extents:%d across:%lluk\n",
1613 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1614 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
1616 /* insert swap space into swap_list: */
1618 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1619 if (p->prio >= swap_info[i].prio) {
1626 swap_list.head = swap_list.next = p - swap_info;
1628 swap_info[prev].next = p - swap_info;
1630 spin_unlock(&swap_lock);
1631 mutex_unlock(&swapon_mutex);
1636 set_blocksize(bdev, p->old_block_size);
1639 destroy_swap_extents(p);
1641 spin_lock(&swap_lock);
1642 swap_map = p->swap_map;
1643 p->swap_file = NULL;
1646 if (!(swap_flags & SWAP_FLAG_PREFER))
1648 spin_unlock(&swap_lock);
1651 filp_close(swap_file, NULL);
1653 if (page && !IS_ERR(page)) {
1655 page_cache_release(page);
1661 inode->i_flags |= S_SWAPFILE;
1662 mutex_unlock(&inode->i_mutex);
1667 void si_swapinfo(struct sysinfo *val)
1670 unsigned long nr_to_be_unused = 0;
1672 spin_lock(&swap_lock);
1673 for (i = 0; i < nr_swapfiles; i++) {
1674 if (!(swap_info[i].flags & SWP_USED) ||
1675 (swap_info[i].flags & SWP_WRITEOK))
1677 nr_to_be_unused += swap_info[i].inuse_pages;
1679 val->freeswap = nr_swap_pages + nr_to_be_unused;
1680 val->totalswap = total_swap_pages + nr_to_be_unused;
1681 spin_unlock(&swap_lock);
1685 * Verify that a swap entry is valid and increment its swap map count.
1687 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1688 * "permanent", but will be reclaimed by the next swapoff.
1690 int swap_duplicate(swp_entry_t entry)
1692 struct swap_info_struct * p;
1693 unsigned long offset, type;
1696 if (is_migration_entry(entry))
1699 type = swp_type(entry);
1700 if (type >= nr_swapfiles)
1702 p = type + swap_info;
1703 offset = swp_offset(entry);
1705 spin_lock(&swap_lock);
1706 if (offset < p->max && p->swap_map[offset]) {
1707 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1708 p->swap_map[offset]++;
1710 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1711 if (swap_overflow++ < 5)
1712 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1713 p->swap_map[offset] = SWAP_MAP_MAX;
1717 spin_unlock(&swap_lock);
1722 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1726 struct swap_info_struct *
1727 get_swap_info_struct(unsigned type)
1729 return &swap_info[type];
1733 * swap_lock prevents swap_map being freed. Don't grab an extra
1734 * reference on the swaphandle, it doesn't matter if it becomes unused.
1736 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1738 int ret = 0, i = 1 << page_cluster;
1740 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1742 if (!page_cluster) /* no readahead */
1744 toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1745 if (!toff) /* first page is swap header */
1749 spin_lock(&swap_lock);
1751 /* Don't read-ahead past the end of the swap area */
1752 if (toff >= swapdev->max)
1754 /* Don't read in free or bad pages */
1755 if (!swapdev->swap_map[toff])
1757 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1762 spin_unlock(&swap_lock);