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>
34 #include <linux/vs_memory.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))
448 spin_unlock(&swap_lock);
451 inode = swap_info[i].swap_file->f_dentry->d_inode;
452 if (S_ISBLK(inode->i_mode) &&
453 device == MKDEV(imajor(inode), iminor(inode))) {
454 spin_unlock(&swap_lock);
458 spin_unlock(&swap_lock);
463 * Return either the total number of swap pages of given type, or the number
464 * of free pages of that type (depending on @free)
466 * This is needed for software suspend
468 unsigned int count_swap_pages(int type, int free)
472 if (type < nr_swapfiles) {
473 spin_lock(&swap_lock);
474 if (swap_info[type].flags & SWP_WRITEOK) {
475 n = swap_info[type].pages;
477 n -= swap_info[type].inuse_pages;
479 spin_unlock(&swap_lock);
486 * No need to decide whether this PTE shares the swap entry with others,
487 * just let do_wp_page work it out if a write is requested later - to
488 * force COW, vm_page_prot omits write permission from any private vma.
490 static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
491 unsigned long addr, swp_entry_t entry, struct page *page)
493 inc_mm_counter(vma->vm_mm, anon_rss);
495 set_pte_at(vma->vm_mm, addr, pte,
496 pte_mkold(mk_pte(page, vma->vm_page_prot)));
497 page_add_anon_rmap(page, vma, addr);
500 * Move the page to the active list so it is not
501 * immediately swapped out again after swapon.
506 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
507 unsigned long addr, unsigned long end,
508 swp_entry_t entry, struct page *page)
510 pte_t swp_pte = swp_entry_to_pte(entry);
515 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
518 * swapoff spends a _lot_ of time in this loop!
519 * Test inline before going to call unuse_pte.
521 if (unlikely(pte_same(*pte, swp_pte))) {
522 unuse_pte(vma, pte++, addr, entry, page);
526 } while (pte++, addr += PAGE_SIZE, addr != end);
527 pte_unmap_unlock(pte - 1, ptl);
531 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
532 unsigned long addr, unsigned long end,
533 swp_entry_t entry, struct page *page)
538 pmd = pmd_offset(pud, addr);
540 next = pmd_addr_end(addr, end);
541 if (pmd_none_or_clear_bad(pmd))
543 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
545 } while (pmd++, addr = next, addr != end);
549 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
550 unsigned long addr, unsigned long end,
551 swp_entry_t entry, struct page *page)
556 pud = pud_offset(pgd, addr);
558 next = pud_addr_end(addr, end);
559 if (pud_none_or_clear_bad(pud))
561 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
563 } while (pud++, addr = next, addr != end);
567 static int unuse_vma(struct vm_area_struct *vma,
568 swp_entry_t entry, struct page *page)
571 unsigned long addr, end, next;
574 addr = page_address_in_vma(page, vma);
578 end = addr + PAGE_SIZE;
580 addr = vma->vm_start;
584 pgd = pgd_offset(vma->vm_mm, addr);
586 next = pgd_addr_end(addr, end);
587 if (pgd_none_or_clear_bad(pgd))
589 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
591 } while (pgd++, addr = next, addr != end);
595 static int unuse_mm(struct mm_struct *mm,
596 swp_entry_t entry, struct page *page)
598 struct vm_area_struct *vma;
600 if (!down_read_trylock(&mm->mmap_sem)) {
602 * Activate page so shrink_cache is unlikely to unmap its
603 * ptes while lock is dropped, so swapoff can make progress.
607 down_read(&mm->mmap_sem);
610 for (vma = mm->mmap; vma; vma = vma->vm_next) {
611 if (vma->anon_vma && unuse_vma(vma, entry, page))
614 up_read(&mm->mmap_sem);
616 * Currently unuse_mm cannot fail, but leave error handling
617 * at call sites for now, since we change it from time to time.
623 * Scan swap_map from current position to next entry still in use.
624 * Recycle to start on reaching the end, returning 0 when empty.
626 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
629 unsigned int max = si->max;
630 unsigned int i = prev;
634 * No need for swap_lock here: we're just looking
635 * for whether an entry is in use, not modifying it; false
636 * hits are okay, and sys_swapoff() has already prevented new
637 * allocations from this area (while holding swap_lock).
646 * No entries in use at top of swap_map,
647 * loop back to start and recheck there.
653 count = si->swap_map[i];
654 if (count && count != SWAP_MAP_BAD)
661 * We completely avoid races by reading each swap page in advance,
662 * and then search for the process using it. All the necessary
663 * page table adjustments can then be made atomically.
665 static int try_to_unuse(unsigned int type)
667 struct swap_info_struct * si = &swap_info[type];
668 struct mm_struct *start_mm;
669 unsigned short *swap_map;
670 unsigned short swcount;
675 int reset_overflow = 0;
679 * When searching mms for an entry, a good strategy is to
680 * start at the first mm we freed the previous entry from
681 * (though actually we don't notice whether we or coincidence
682 * freed the entry). Initialize this start_mm with a hold.
684 * A simpler strategy would be to start at the last mm we
685 * freed the previous entry from; but that would take less
686 * advantage of mmlist ordering, which clusters forked mms
687 * together, child after parent. If we race with dup_mmap(), we
688 * prefer to resolve parent before child, lest we miss entries
689 * duplicated after we scanned child: using last mm would invert
690 * that. Though it's only a serious concern when an overflowed
691 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
694 atomic_inc(&init_mm.mm_users);
697 * Keep on scanning until all entries have gone. Usually,
698 * one pass through swap_map is enough, but not necessarily:
699 * there are races when an instance of an entry might be missed.
701 while ((i = find_next_to_unuse(si, i)) != 0) {
702 if (signal_pending(current)) {
708 * Get a page for the entry, using the existing swap
709 * cache page if there is one. Otherwise, get a clean
710 * page and read the swap into it.
712 swap_map = &si->swap_map[i];
713 entry = swp_entry(type, i);
714 page = read_swap_cache_async(entry, NULL, 0);
717 * Either swap_duplicate() failed because entry
718 * has been freed independently, and will not be
719 * reused since sys_swapoff() already disabled
720 * allocation from here, or alloc_page() failed.
729 * Don't hold on to start_mm if it looks like exiting.
731 if (atomic_read(&start_mm->mm_users) == 1) {
734 atomic_inc(&init_mm.mm_users);
738 * Wait for and lock page. When do_swap_page races with
739 * try_to_unuse, do_swap_page can handle the fault much
740 * faster than try_to_unuse can locate the entry. This
741 * apparently redundant "wait_on_page_locked" lets try_to_unuse
742 * defer to do_swap_page in such a case - in some tests,
743 * do_swap_page and try_to_unuse repeatedly compete.
745 wait_on_page_locked(page);
746 wait_on_page_writeback(page);
748 wait_on_page_writeback(page);
751 * Remove all references to entry.
752 * Whenever we reach init_mm, there's no address space
753 * to search, but use it as a reminder to search shmem.
758 if (start_mm == &init_mm)
759 shmem = shmem_unuse(entry, page);
761 retval = unuse_mm(start_mm, entry, page);
764 int set_start_mm = (*swap_map >= swcount);
765 struct list_head *p = &start_mm->mmlist;
766 struct mm_struct *new_start_mm = start_mm;
767 struct mm_struct *prev_mm = start_mm;
768 struct mm_struct *mm;
770 atomic_inc(&new_start_mm->mm_users);
771 atomic_inc(&prev_mm->mm_users);
772 spin_lock(&mmlist_lock);
773 while (*swap_map > 1 && !retval &&
774 (p = p->next) != &start_mm->mmlist) {
775 mm = list_entry(p, struct mm_struct, mmlist);
776 if (!atomic_inc_not_zero(&mm->mm_users))
778 spin_unlock(&mmlist_lock);
787 else if (mm == &init_mm) {
789 shmem = shmem_unuse(entry, page);
791 retval = unuse_mm(mm, entry, page);
792 if (set_start_mm && *swap_map < swcount) {
794 atomic_inc(&mm->mm_users);
798 spin_lock(&mmlist_lock);
800 spin_unlock(&mmlist_lock);
803 start_mm = new_start_mm;
807 page_cache_release(page);
812 * How could swap count reach 0x7fff when the maximum
813 * pid is 0x7fff, and there's no way to repeat a swap
814 * page within an mm (except in shmem, where it's the
815 * shared object which takes the reference count)?
816 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
818 * If that's wrong, then we should worry more about
819 * exit_mmap() and do_munmap() cases described above:
820 * we might be resetting SWAP_MAP_MAX too early here.
821 * We know "Undead"s can happen, they're okay, so don't
822 * report them; but do report if we reset SWAP_MAP_MAX.
824 if (*swap_map == SWAP_MAP_MAX) {
825 spin_lock(&swap_lock);
827 spin_unlock(&swap_lock);
832 * If a reference remains (rare), we would like to leave
833 * the page in the swap cache; but try_to_unmap could
834 * then re-duplicate the entry once we drop page lock,
835 * so we might loop indefinitely; also, that page could
836 * not be swapped out to other storage meanwhile. So:
837 * delete from cache even if there's another reference,
838 * after ensuring that the data has been saved to disk -
839 * since if the reference remains (rarer), it will be
840 * read from disk into another page. Splitting into two
841 * pages would be incorrect if swap supported "shared
842 * private" pages, but they are handled by tmpfs files.
844 * Note shmem_unuse already deleted a swappage from
845 * the swap cache, unless the move to filepage failed:
846 * in which case it left swappage in cache, lowered its
847 * swap count to pass quickly through the loops above,
848 * and now we must reincrement count to try again later.
850 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
851 struct writeback_control wbc = {
852 .sync_mode = WB_SYNC_NONE,
855 swap_writepage(page, &wbc);
857 wait_on_page_writeback(page);
859 if (PageSwapCache(page)) {
861 swap_duplicate(entry);
863 delete_from_swap_cache(page);
867 * So we could skip searching mms once swap count went
868 * to 1, we did not mark any present ptes as dirty: must
869 * mark page dirty so shrink_list will preserve it.
873 page_cache_release(page);
876 * Make sure that we aren't completely killing
877 * interactive performance.
883 if (reset_overflow) {
884 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
891 * After a successful try_to_unuse, if no swap is now in use, we know
892 * we can empty the mmlist. swap_lock must be held on entry and exit.
893 * Note that mmlist_lock nests inside swap_lock, and an mm must be
894 * added to the mmlist just after page_duplicate - before would be racy.
896 static void drain_mmlist(void)
898 struct list_head *p, *next;
901 for (i = 0; i < nr_swapfiles; i++)
902 if (swap_info[i].inuse_pages)
904 spin_lock(&mmlist_lock);
905 list_for_each_safe(p, next, &init_mm.mmlist)
907 spin_unlock(&mmlist_lock);
911 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
912 * corresponds to page offset `offset'.
914 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
916 struct swap_extent *se = sis->curr_swap_extent;
917 struct swap_extent *start_se = se;
920 struct list_head *lh;
922 if (se->start_page <= offset &&
923 offset < (se->start_page + se->nr_pages)) {
924 return se->start_block + (offset - se->start_page);
927 if (lh == &sis->extent_list)
929 se = list_entry(lh, struct swap_extent, list);
930 sis->curr_swap_extent = se;
931 BUG_ON(se == start_se); /* It *must* be present */
936 * Free all of a swapdev's extent information
938 static void destroy_swap_extents(struct swap_info_struct *sis)
940 while (!list_empty(&sis->extent_list)) {
941 struct swap_extent *se;
943 se = list_entry(sis->extent_list.next,
944 struct swap_extent, list);
951 * Add a block range (and the corresponding page range) into this swapdev's
952 * extent list. The extent list is kept sorted in page order.
954 * This function rather assumes that it is called in ascending page order.
957 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
958 unsigned long nr_pages, sector_t start_block)
960 struct swap_extent *se;
961 struct swap_extent *new_se;
962 struct list_head *lh;
964 lh = sis->extent_list.prev; /* The highest page extent */
965 if (lh != &sis->extent_list) {
966 se = list_entry(lh, struct swap_extent, list);
967 BUG_ON(se->start_page + se->nr_pages != start_page);
968 if (se->start_block + se->nr_pages == start_block) {
970 se->nr_pages += nr_pages;
976 * No merge. Insert a new extent, preserving ordering.
978 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
981 new_se->start_page = start_page;
982 new_se->nr_pages = nr_pages;
983 new_se->start_block = start_block;
985 list_add_tail(&new_se->list, &sis->extent_list);
990 * A `swap extent' is a simple thing which maps a contiguous range of pages
991 * onto a contiguous range of disk blocks. An ordered list of swap extents
992 * is built at swapon time and is then used at swap_writepage/swap_readpage
993 * time for locating where on disk a page belongs.
995 * If the swapfile is an S_ISBLK block device, a single extent is installed.
996 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
997 * swap files identically.
999 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1000 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1001 * swapfiles are handled *identically* after swapon time.
1003 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1004 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1005 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1006 * requirements, they are simply tossed out - we will never use those blocks
1009 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1010 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1011 * which will scribble on the fs.
1013 * The amount of disk space which a single swap extent represents varies.
1014 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1015 * extents in the list. To avoid much list walking, we cache the previous
1016 * search location in `curr_swap_extent', and start new searches from there.
1017 * This is extremely effective. The average number of iterations in
1018 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1020 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1022 struct inode *inode;
1023 unsigned blocks_per_page;
1024 unsigned long page_no;
1026 sector_t probe_block;
1027 sector_t last_block;
1028 sector_t lowest_block = -1;
1029 sector_t highest_block = 0;
1033 inode = sis->swap_file->f_mapping->host;
1034 if (S_ISBLK(inode->i_mode)) {
1035 ret = add_swap_extent(sis, 0, sis->max, 0);
1040 blkbits = inode->i_blkbits;
1041 blocks_per_page = PAGE_SIZE >> blkbits;
1044 * Map all the blocks into the extent list. This code doesn't try
1049 last_block = i_size_read(inode) >> blkbits;
1050 while ((probe_block + blocks_per_page) <= last_block &&
1051 page_no < sis->max) {
1052 unsigned block_in_page;
1053 sector_t first_block;
1055 first_block = bmap(inode, probe_block);
1056 if (first_block == 0)
1060 * It must be PAGE_SIZE aligned on-disk
1062 if (first_block & (blocks_per_page - 1)) {
1067 for (block_in_page = 1; block_in_page < blocks_per_page;
1071 block = bmap(inode, probe_block + block_in_page);
1074 if (block != first_block + block_in_page) {
1081 first_block >>= (PAGE_SHIFT - blkbits);
1082 if (page_no) { /* exclude the header page */
1083 if (first_block < lowest_block)
1084 lowest_block = first_block;
1085 if (first_block > highest_block)
1086 highest_block = first_block;
1090 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1092 ret = add_swap_extent(sis, page_no, 1, first_block);
1097 probe_block += blocks_per_page;
1102 *span = 1 + highest_block - lowest_block;
1104 page_no = 1; /* force Empty message */
1106 sis->pages = page_no - 1;
1107 sis->highest_bit = page_no - 1;
1109 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1110 struct swap_extent, list);
1113 printk(KERN_ERR "swapon: swapfile has holes\n");
1119 #if 0 /* We don't need this yet */
1120 #include <linux/backing-dev.h>
1121 int page_queue_congested(struct page *page)
1123 struct backing_dev_info *bdi;
1125 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1127 if (PageSwapCache(page)) {
1128 swp_entry_t entry = { .val = page_private(page) };
1129 struct swap_info_struct *sis;
1131 sis = get_swap_info_struct(swp_type(entry));
1132 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1134 bdi = page->mapping->backing_dev_info;
1135 return bdi_write_congested(bdi);
1139 asmlinkage long sys_swapoff(const char __user * specialfile)
1141 struct swap_info_struct * p = NULL;
1142 unsigned short *swap_map;
1143 struct file *swap_file, *victim;
1144 struct address_space *mapping;
1145 struct inode *inode;
1150 if (!capable(CAP_SYS_ADMIN))
1153 pathname = getname(specialfile);
1154 err = PTR_ERR(pathname);
1155 if (IS_ERR(pathname))
1158 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1160 err = PTR_ERR(victim);
1164 mapping = victim->f_mapping;
1166 spin_lock(&swap_lock);
1167 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1168 p = swap_info + type;
1169 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1170 if (p->swap_file->f_mapping == mapping)
1177 spin_unlock(&swap_lock);
1180 if (!security_vm_enough_memory(p->pages))
1181 vm_unacct_memory(p->pages);
1184 spin_unlock(&swap_lock);
1188 swap_list.head = p->next;
1190 swap_info[prev].next = p->next;
1192 if (type == swap_list.next) {
1193 /* just pick something that's safe... */
1194 swap_list.next = swap_list.head;
1196 nr_swap_pages -= p->pages;
1197 total_swap_pages -= p->pages;
1198 p->flags &= ~SWP_WRITEOK;
1199 spin_unlock(&swap_lock);
1201 current->flags |= PF_SWAPOFF;
1202 err = try_to_unuse(type);
1203 current->flags &= ~PF_SWAPOFF;
1206 /* re-insert swap space back into swap_list */
1207 spin_lock(&swap_lock);
1208 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1209 if (p->prio >= swap_info[i].prio)
1213 swap_list.head = swap_list.next = p - swap_info;
1215 swap_info[prev].next = p - swap_info;
1216 nr_swap_pages += p->pages;
1217 total_swap_pages += p->pages;
1218 p->flags |= SWP_WRITEOK;
1219 spin_unlock(&swap_lock);
1223 /* wait for any unplug function to finish */
1224 down_write(&swap_unplug_sem);
1225 up_write(&swap_unplug_sem);
1227 destroy_swap_extents(p);
1228 mutex_lock(&swapon_mutex);
1229 spin_lock(&swap_lock);
1232 /* wait for anyone still in scan_swap_map */
1233 p->highest_bit = 0; /* cuts scans short */
1234 while (p->flags >= SWP_SCANNING) {
1235 spin_unlock(&swap_lock);
1236 schedule_timeout_uninterruptible(1);
1237 spin_lock(&swap_lock);
1240 swap_file = p->swap_file;
1241 p->swap_file = NULL;
1243 swap_map = p->swap_map;
1246 spin_unlock(&swap_lock);
1247 mutex_unlock(&swapon_mutex);
1249 inode = mapping->host;
1250 if (S_ISBLK(inode->i_mode)) {
1251 struct block_device *bdev = I_BDEV(inode);
1252 set_blocksize(bdev, p->old_block_size);
1255 mutex_lock(&inode->i_mutex);
1256 inode->i_flags &= ~S_SWAPFILE;
1257 mutex_unlock(&inode->i_mutex);
1259 filp_close(swap_file, NULL);
1263 filp_close(victim, NULL);
1268 #ifdef CONFIG_PROC_FS
1270 static void *swap_start(struct seq_file *swap, loff_t *pos)
1272 struct swap_info_struct *ptr = swap_info;
1276 mutex_lock(&swapon_mutex);
1278 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1279 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1288 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1290 struct swap_info_struct *ptr = v;
1291 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1293 for (++ptr; ptr < endptr; ptr++) {
1294 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1303 static void swap_stop(struct seq_file *swap, void *v)
1305 mutex_unlock(&swapon_mutex);
1308 static int swap_show(struct seq_file *swap, void *v)
1310 struct swap_info_struct *ptr = v;
1315 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1317 file = ptr->swap_file;
1318 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1319 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1320 len < 40 ? 40 - len : 1, " ",
1321 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1322 "partition" : "file\t",
1323 ptr->pages << (PAGE_SHIFT - 10),
1324 ptr->inuse_pages << (PAGE_SHIFT - 10),
1329 static struct seq_operations swaps_op = {
1330 .start = swap_start,
1336 static int swaps_open(struct inode *inode, struct file *file)
1338 return seq_open(file, &swaps_op);
1341 static struct file_operations proc_swaps_operations = {
1344 .llseek = seq_lseek,
1345 .release = seq_release,
1348 static int __init procswaps_init(void)
1350 struct proc_dir_entry *entry;
1352 entry = create_proc_entry("swaps", 0, NULL);
1354 entry->proc_fops = &proc_swaps_operations;
1357 __initcall(procswaps_init);
1358 #endif /* CONFIG_PROC_FS */
1361 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1363 * The swapon system call
1365 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1367 struct swap_info_struct * p;
1369 struct block_device *bdev = NULL;
1370 struct file *swap_file = NULL;
1371 struct address_space *mapping;
1375 static int least_priority;
1376 union swap_header *swap_header = NULL;
1377 int swap_header_version;
1378 unsigned int nr_good_pages = 0;
1381 unsigned long maxpages = 1;
1383 unsigned short *swap_map;
1384 struct page *page = NULL;
1385 struct inode *inode = NULL;
1388 if (!capable(CAP_SYS_ADMIN))
1390 spin_lock(&swap_lock);
1392 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1393 if (!(p->flags & SWP_USED))
1396 if (type >= MAX_SWAPFILES) {
1397 spin_unlock(&swap_lock);
1400 if (type >= nr_swapfiles)
1401 nr_swapfiles = type+1;
1402 INIT_LIST_HEAD(&p->extent_list);
1403 p->flags = SWP_USED;
1404 p->swap_file = NULL;
1405 p->old_block_size = 0;
1412 if (swap_flags & SWAP_FLAG_PREFER) {
1414 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1416 p->prio = --least_priority;
1418 spin_unlock(&swap_lock);
1419 name = getname(specialfile);
1420 error = PTR_ERR(name);
1425 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1426 error = PTR_ERR(swap_file);
1427 if (IS_ERR(swap_file)) {
1432 p->swap_file = swap_file;
1433 mapping = swap_file->f_mapping;
1434 inode = mapping->host;
1437 for (i = 0; i < nr_swapfiles; i++) {
1438 struct swap_info_struct *q = &swap_info[i];
1440 if (i == type || !q->swap_file)
1442 if (mapping == q->swap_file->f_mapping)
1447 if (S_ISBLK(inode->i_mode)) {
1448 bdev = I_BDEV(inode);
1449 error = bd_claim(bdev, sys_swapon);
1455 p->old_block_size = block_size(bdev);
1456 error = set_blocksize(bdev, PAGE_SIZE);
1460 } else if (S_ISREG(inode->i_mode)) {
1461 p->bdev = inode->i_sb->s_bdev;
1462 mutex_lock(&inode->i_mutex);
1464 if (IS_SWAPFILE(inode)) {
1472 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1475 * Read the swap header.
1477 if (!mapping->a_ops->readpage) {
1481 page = read_mapping_page(mapping, 0, swap_file);
1483 error = PTR_ERR(page);
1486 wait_on_page_locked(page);
1487 if (!PageUptodate(page))
1490 swap_header = page_address(page);
1492 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1493 swap_header_version = 1;
1494 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1495 swap_header_version = 2;
1497 printk(KERN_ERR "Unable to find swap-space signature\n");
1502 switch (swap_header_version) {
1504 printk(KERN_ERR "version 0 swap is no longer supported. "
1505 "Use mkswap -v1 %s\n", name);
1509 /* Check the swap header's sub-version and the size of
1510 the swap file and bad block lists */
1511 if (swap_header->info.version != 1) {
1513 "Unable to handle swap header version %d\n",
1514 swap_header->info.version);
1520 p->cluster_next = 1;
1523 * Find out how many pages are allowed for a single swap
1524 * device. There are two limiting factors: 1) the number of
1525 * bits for the swap offset in the swp_entry_t type and
1526 * 2) the number of bits in the a swap pte as defined by
1527 * the different architectures. In order to find the
1528 * largest possible bit mask a swap entry with swap type 0
1529 * and swap offset ~0UL is created, encoded to a swap pte,
1530 * decoded to a swp_entry_t again and finally the swap
1531 * offset is extracted. This will mask all the bits from
1532 * the initial ~0UL mask that can't be encoded in either
1533 * the swp_entry_t or the architecture definition of a
1536 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1537 if (maxpages > swap_header->info.last_page)
1538 maxpages = swap_header->info.last_page;
1539 p->highest_bit = maxpages - 1;
1544 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1546 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1549 /* OK, set up the swap map and apply the bad block list */
1550 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1556 memset(p->swap_map, 0, maxpages * sizeof(short));
1557 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1558 int page_nr = swap_header->info.badpages[i];
1559 if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
1562 p->swap_map[page_nr] = SWAP_MAP_BAD;
1564 nr_good_pages = swap_header->info.last_page -
1565 swap_header->info.nr_badpages -
1566 1 /* header page */;
1571 if (swapfilesize && maxpages > swapfilesize) {
1573 "Swap area shorter than signature indicates\n");
1577 if (nr_good_pages) {
1578 p->swap_map[0] = SWAP_MAP_BAD;
1580 p->pages = nr_good_pages;
1581 nr_extents = setup_swap_extents(p, &span);
1582 if (nr_extents < 0) {
1586 nr_good_pages = p->pages;
1588 if (!nr_good_pages) {
1589 printk(KERN_WARNING "Empty swap-file\n");
1594 mutex_lock(&swapon_mutex);
1595 spin_lock(&swap_lock);
1596 p->flags = SWP_ACTIVE;
1597 nr_swap_pages += nr_good_pages;
1598 total_swap_pages += nr_good_pages;
1600 printk(KERN_INFO "Adding %uk swap on %s. "
1601 "Priority:%d extents:%d across:%lluk\n",
1602 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1603 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
1605 /* insert swap space into swap_list: */
1607 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1608 if (p->prio >= swap_info[i].prio) {
1615 swap_list.head = swap_list.next = p - swap_info;
1617 swap_info[prev].next = p - swap_info;
1619 spin_unlock(&swap_lock);
1620 mutex_unlock(&swapon_mutex);
1625 set_blocksize(bdev, p->old_block_size);
1628 destroy_swap_extents(p);
1630 spin_lock(&swap_lock);
1631 swap_map = p->swap_map;
1632 p->swap_file = NULL;
1635 if (!(swap_flags & SWAP_FLAG_PREFER))
1637 spin_unlock(&swap_lock);
1640 filp_close(swap_file, NULL);
1642 if (page && !IS_ERR(page)) {
1644 page_cache_release(page);
1650 inode->i_flags |= S_SWAPFILE;
1651 mutex_unlock(&inode->i_mutex);
1656 void si_swapinfo(struct sysinfo *val)
1659 unsigned long nr_to_be_unused = 0;
1661 spin_lock(&swap_lock);
1662 for (i = 0; i < nr_swapfiles; i++) {
1663 if (!(swap_info[i].flags & SWP_USED) ||
1664 (swap_info[i].flags & SWP_WRITEOK))
1666 nr_to_be_unused += swap_info[i].inuse_pages;
1668 val->freeswap = nr_swap_pages + nr_to_be_unused;
1669 val->totalswap = total_swap_pages + nr_to_be_unused;
1670 spin_unlock(&swap_lock);
1671 if (vx_flags(VXF_VIRT_MEM, 0))
1672 vx_vsi_swapinfo(val);
1676 * Verify that a swap entry is valid and increment its swap map count.
1678 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1679 * "permanent", but will be reclaimed by the next swapoff.
1681 int swap_duplicate(swp_entry_t entry)
1683 struct swap_info_struct * p;
1684 unsigned long offset, type;
1687 if (is_migration_entry(entry))
1690 type = swp_type(entry);
1691 if (type >= nr_swapfiles)
1693 p = type + swap_info;
1694 offset = swp_offset(entry);
1696 spin_lock(&swap_lock);
1697 if (offset < p->max && p->swap_map[offset]) {
1698 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1699 p->swap_map[offset]++;
1701 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1702 if (swap_overflow++ < 5)
1703 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1704 p->swap_map[offset] = SWAP_MAP_MAX;
1708 spin_unlock(&swap_lock);
1713 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1717 struct swap_info_struct *
1718 get_swap_info_struct(unsigned type)
1720 return &swap_info[type];
1724 * swap_lock prevents swap_map being freed. Don't grab an extra
1725 * reference on the swaphandle, it doesn't matter if it becomes unused.
1727 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1729 int ret = 0, i = 1 << page_cluster;
1731 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1733 if (!page_cluster) /* no readahead */
1735 toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1736 if (!toff) /* first page is swap header */
1740 spin_lock(&swap_lock);
1742 /* Don't read-ahead past the end of the swap area */
1743 if (toff >= swapdev->max)
1745 /* Don't read in free or bad pages */
1746 if (!swapdev->swap_map[toff])
1748 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1753 spin_unlock(&swap_lock);