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>
29 #include <asm/pgtable.h>
30 #include <asm/tlbflush.h>
31 #include <linux/swapops.h>
32 #include <linux/vs_base.h>
33 #include <linux/vs_memory.h>
35 spinlock_t swaplock = SPIN_LOCK_UNLOCKED;
36 unsigned int nr_swapfiles;
37 long total_swap_pages;
38 static int swap_overflow;
40 EXPORT_SYMBOL(total_swap_pages);
42 static const char Bad_file[] = "Bad swap file entry ";
43 static const char Unused_file[] = "Unused swap file entry ";
44 static const char Bad_offset[] = "Bad swap offset entry ";
45 static const char Unused_offset[] = "Unused swap offset entry ";
47 struct swap_list_t swap_list = {-1, -1};
49 struct swap_info_struct swap_info[MAX_SWAPFILES];
51 static DECLARE_MUTEX(swapon_sem);
54 * We need this because the bdev->unplug_fn can sleep and we cannot
55 * hold swap_list_lock while calling the unplug_fn. And swap_list_lock
56 * cannot be turned into a semaphore.
58 static DECLARE_RWSEM(swap_unplug_sem);
60 #define SWAPFILE_CLUSTER 256
62 void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
66 down_read(&swap_unplug_sem);
67 entry.val = page->private;
68 if (PageSwapCache(page)) {
69 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
70 struct backing_dev_info *bdi;
73 * If the page is removed from swapcache from under us (with a
74 * racy try_to_unuse/swapoff) we need an additional reference
75 * count to avoid reading garbage from page->private above. If
76 * the WARN_ON triggers during a swapoff it maybe the race
77 * condition and it's harmless. However if it triggers without
78 * swapoff it signals a problem.
80 WARN_ON(page_count(page) <= 1);
82 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
83 bdi->unplug_io_fn(bdi, page);
85 up_read(&swap_unplug_sem);
88 static inline int scan_swap_map(struct swap_info_struct *si)
92 * We try to cluster swap pages by allocating them
93 * sequentially in swap. Once we've allocated
94 * SWAPFILE_CLUSTER pages this way, however, we resort to
95 * first-free allocation, starting a new cluster. This
96 * prevents us from scattering swap pages all over the entire
97 * swap partition, so that we reduce overall disk seek times
98 * between swap pages. -- sct */
100 while (si->cluster_next <= si->highest_bit) {
101 offset = si->cluster_next++;
102 if (si->swap_map[offset])
108 si->cluster_nr = SWAPFILE_CLUSTER;
110 /* try to find an empty (even not aligned) cluster. */
111 offset = si->lowest_bit;
113 if (offset+SWAPFILE_CLUSTER-1 <= si->highest_bit)
116 for (nr = offset; nr < offset+SWAPFILE_CLUSTER; nr++)
117 if (si->swap_map[nr])
120 goto check_next_cluster;
122 /* We found a completly empty cluster, so start
127 /* No luck, so now go finegrined as usual. -Andrea */
128 for (offset = si->lowest_bit; offset <= si->highest_bit ; offset++) {
129 if (si->swap_map[offset])
131 si->lowest_bit = offset+1;
133 if (offset == si->lowest_bit)
135 if (offset == si->highest_bit)
137 if (si->lowest_bit > si->highest_bit) {
138 si->lowest_bit = si->max;
141 si->swap_map[offset] = 1;
144 si->cluster_next = offset+1;
147 si->lowest_bit = si->max;
152 swp_entry_t get_swap_page(void)
154 struct swap_info_struct * p;
155 unsigned long offset;
157 int type, wrapped = 0;
159 entry.val = 0; /* Out of memory */
161 type = swap_list.next;
164 if (nr_swap_pages <= 0)
168 p = &swap_info[type];
169 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
171 offset = scan_swap_map(p);
172 swap_device_unlock(p);
174 entry = swp_entry(type,offset);
175 type = swap_info[type].next;
177 p->prio != swap_info[type].prio) {
178 swap_list.next = swap_list.head;
180 swap_list.next = type;
187 if (type < 0 || p->prio != swap_info[type].prio) {
188 type = swap_list.head;
193 goto out; /* out of swap space */
200 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
202 struct swap_info_struct * p;
203 unsigned long offset, type;
207 type = swp_type(entry);
208 if (type >= nr_swapfiles)
210 p = & swap_info[type];
211 if (!(p->flags & SWP_USED))
213 offset = swp_offset(entry);
214 if (offset >= p->max)
216 if (!p->swap_map[offset])
219 if (p->prio > swap_info[swap_list.next].prio)
220 swap_list.next = type;
225 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
228 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
231 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
234 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
239 static void swap_info_put(struct swap_info_struct * p)
241 swap_device_unlock(p);
245 static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
247 int count = p->swap_map[offset];
249 if (count < SWAP_MAP_MAX) {
251 p->swap_map[offset] = count;
253 if (offset < p->lowest_bit)
254 p->lowest_bit = offset;
255 if (offset > p->highest_bit)
256 p->highest_bit = offset;
265 * Caller has made sure that the swapdevice corresponding to entry
266 * is still around or has not been recycled.
268 void swap_free(swp_entry_t entry)
270 struct swap_info_struct * p;
272 p = swap_info_get(entry);
274 swap_entry_free(p, swp_offset(entry));
280 * Check if we're the only user of a swap page,
281 * when the page is locked.
283 static int exclusive_swap_page(struct page *page)
286 struct swap_info_struct * p;
289 entry.val = page->private;
290 p = swap_info_get(entry);
292 /* Is the only swap cache user the cache itself? */
293 if (p->swap_map[swp_offset(entry)] == 1) {
294 /* Recheck the page count with the swapcache lock held.. */
295 spin_lock_irq(&swapper_space.tree_lock);
296 if (page_count(page) == 2)
298 spin_unlock_irq(&swapper_space.tree_lock);
306 * We can use this swap cache entry directly
307 * if there are no other references to it.
309 * Here "exclusive_swap_page()" does the real
310 * work, but we opportunistically check whether
311 * we need to get all the locks first..
313 int can_share_swap_page(struct page *page)
317 if (!PageLocked(page))
319 switch (page_count(page)) {
321 if (!PagePrivate(page))
325 if (!PageSwapCache(page))
327 retval = exclusive_swap_page(page);
330 if (PageReserved(page))
338 * Work out if there are any other processes sharing this
339 * swap cache page. Free it if you can. Return success.
341 int remove_exclusive_swap_page(struct page *page)
344 struct swap_info_struct * p;
347 BUG_ON(PagePrivate(page));
348 BUG_ON(!PageLocked(page));
350 if (!PageSwapCache(page))
352 if (PageWriteback(page))
354 if (page_count(page) != 2) /* 2: us + cache */
357 entry.val = page->private;
358 p = swap_info_get(entry);
362 /* Is the only swap cache user the cache itself? */
364 if (p->swap_map[swp_offset(entry)] == 1) {
365 /* Recheck the page count with the swapcache lock held.. */
366 spin_lock_irq(&swapper_space.tree_lock);
367 if ((page_count(page) == 2) && !PageWriteback(page)) {
368 __delete_from_swap_cache(page);
372 spin_unlock_irq(&swapper_space.tree_lock);
378 page_cache_release(page);
385 * Free the swap entry like above, but also try to
386 * free the page cache entry if it is the last user.
388 void free_swap_and_cache(swp_entry_t entry)
390 struct swap_info_struct * p;
391 struct page *page = NULL;
393 p = swap_info_get(entry);
395 if (swap_entry_free(p, swp_offset(entry)) == 1) {
396 spin_lock_irq(&swapper_space.tree_lock);
397 page = radix_tree_lookup(&swapper_space.page_tree,
399 if (page && TestSetPageLocked(page))
401 spin_unlock_irq(&swapper_space.tree_lock);
408 BUG_ON(PagePrivate(page));
409 page_cache_get(page);
410 one_user = (page_count(page) == 2);
411 /* Only cache user (+us), or swap space full? Free it! */
412 if (!PageWriteback(page) && (one_user || vm_swap_full())) {
413 delete_from_swap_cache(page);
417 page_cache_release(page);
422 * The swap entry has been read in advance, and we return 1 to indicate
423 * that the page has been used or is no longer needed.
425 * Always set the resulting pte to be nowrite (the same as COW pages
426 * after one process has exited). We don't know just how many PTEs will
427 * share this swap entry, so be cautious and let do_wp_page work out
428 * what to do if a write is requested later.
430 /* vma->vm_mm->page_table_lock is held */
432 unuse_pte(struct vm_area_struct *vma, unsigned long address, pte_t *dir,
433 swp_entry_t entry, struct page *page)
435 // vma->vm_mm->rss++;
436 vx_rsspages_inc(vma->vm_mm);
438 set_pte(dir, pte_mkold(mk_pte(page, vma->vm_page_prot)));
439 page_add_anon_rmap(page, vma, address);
443 /* vma->vm_mm->page_table_lock is held */
444 static unsigned long unuse_pmd(struct vm_area_struct * vma, pmd_t *dir,
445 unsigned long address, unsigned long size, unsigned long offset,
446 swp_entry_t entry, struct page *page)
450 pte_t swp_pte = swp_entry_to_pte(entry);
459 pte = pte_offset_map(dir, address);
460 offset += address & PMD_MASK;
461 address &= ~PMD_MASK;
462 end = address + size;
467 * swapoff spends a _lot_ of time in this loop!
468 * Test inline before going to call unuse_pte.
470 if (unlikely(pte_same(*pte, swp_pte))) {
471 unuse_pte(vma, offset + address, pte, entry, page);
475 * Move the page to the active list so it is not
476 * immediately swapped out again after swapon.
480 /* add 1 since address may be 0 */
481 return 1 + offset + address;
483 address += PAGE_SIZE;
485 } while (address && (address < end));
490 /* vma->vm_mm->page_table_lock is held */
491 static unsigned long unuse_pgd(struct vm_area_struct * vma, pgd_t *dir,
492 unsigned long address, unsigned long size,
493 swp_entry_t entry, struct page *page)
496 unsigned long offset, end;
497 unsigned long foundaddr;
506 pmd = pmd_offset(dir, address);
507 offset = address & PGDIR_MASK;
508 address &= ~PGDIR_MASK;
509 end = address + size;
510 if (end > PGDIR_SIZE)
515 foundaddr = unuse_pmd(vma, pmd, address, end - address,
516 offset, entry, page);
519 address = (address + PMD_SIZE) & PMD_MASK;
521 } while (address && (address < end));
525 /* vma->vm_mm->page_table_lock is held */
526 static unsigned long unuse_vma(struct vm_area_struct * vma,
527 swp_entry_t entry, struct page *page)
530 unsigned long start, end;
531 unsigned long foundaddr;
534 start = page_address_in_vma(page, vma);
535 if (start == -EFAULT)
538 end = start + PAGE_SIZE;
540 start = vma->vm_start;
543 pgdir = pgd_offset(vma->vm_mm, start);
545 foundaddr = unuse_pgd(vma, pgdir, start, end - start,
549 start = (start + PGDIR_SIZE) & PGDIR_MASK;
551 } while (start && (start < end));
555 static int unuse_process(struct mm_struct * mm,
556 swp_entry_t entry, struct page* page)
558 struct vm_area_struct* vma;
559 unsigned long foundaddr = 0;
562 * Go through process' page directory.
564 if (!down_read_trylock(&mm->mmap_sem)) {
566 * Our reference to the page stops try_to_unmap_one from
567 * unmapping its ptes, so swapoff can make progress.
570 down_read(&mm->mmap_sem);
573 spin_lock(&mm->page_table_lock);
574 for (vma = mm->mmap; vma; vma = vma->vm_next) {
576 foundaddr = unuse_vma(vma, entry, page);
581 spin_unlock(&mm->page_table_lock);
582 up_read(&mm->mmap_sem);
584 * Currently unuse_process cannot fail, but leave error handling
585 * at call sites for now, since we change it from time to time.
591 * Scan swap_map from current position to next entry still in use.
592 * Recycle to start on reaching the end, returning 0 when empty.
594 static int find_next_to_unuse(struct swap_info_struct *si, int prev)
601 * No need for swap_device_lock(si) here: we're just looking
602 * for whether an entry is in use, not modifying it; false
603 * hits are okay, and sys_swapoff() has already prevented new
604 * allocations from this area (while holding swap_list_lock()).
613 * No entries in use at top of swap_map,
614 * loop back to start and recheck there.
620 count = si->swap_map[i];
621 if (count && count != SWAP_MAP_BAD)
628 * We completely avoid races by reading each swap page in advance,
629 * and then search for the process using it. All the necessary
630 * page table adjustments can then be made atomically.
632 static int try_to_unuse(unsigned int type)
634 struct swap_info_struct * si = &swap_info[type];
635 struct mm_struct *start_mm;
636 unsigned short *swap_map;
637 unsigned short swcount;
642 int reset_overflow = 0;
646 * When searching mms for an entry, a good strategy is to
647 * start at the first mm we freed the previous entry from
648 * (though actually we don't notice whether we or coincidence
649 * freed the entry). Initialize this start_mm with a hold.
651 * A simpler strategy would be to start at the last mm we
652 * freed the previous entry from; but that would take less
653 * advantage of mmlist ordering (now preserved by swap_out()),
654 * which clusters forked address spaces together, most recent
655 * child immediately after parent. If we race with dup_mmap(),
656 * we very much want to resolve parent before child, otherwise
657 * we may miss some entries: using last mm would invert that.
660 atomic_inc(&init_mm.mm_users);
663 * Keep on scanning until all entries have gone. Usually,
664 * one pass through swap_map is enough, but not necessarily:
665 * mmput() removes mm from mmlist before exit_mmap() and its
666 * zap_page_range(). That's not too bad, those entries are
667 * on their way out, and handled faster there than here.
668 * do_munmap() behaves similarly, taking the range out of mm's
669 * vma list before zap_page_range(). But unfortunately, when
670 * unmapping a part of a vma, it takes the whole out first,
671 * then reinserts what's left after (might even reschedule if
672 * open() method called) - so swap entries may be invisible
673 * to swapoff for a while, then reappear - but that is rare.
675 while ((i = find_next_to_unuse(si, i)) != 0) {
676 if (signal_pending(current)) {
682 * Get a page for the entry, using the existing swap
683 * cache page if there is one. Otherwise, get a clean
684 * page and read the swap into it.
686 swap_map = &si->swap_map[i];
687 entry = swp_entry(type, i);
688 page = read_swap_cache_async(entry, NULL, 0);
691 * Either swap_duplicate() failed because entry
692 * has been freed independently, and will not be
693 * reused since sys_swapoff() already disabled
694 * allocation from here, or alloc_page() failed.
703 * Don't hold on to start_mm if it looks like exiting.
705 if (atomic_read(&start_mm->mm_users) == 1) {
708 atomic_inc(&init_mm.mm_users);
712 * Wait for and lock page. When do_swap_page races with
713 * try_to_unuse, do_swap_page can handle the fault much
714 * faster than try_to_unuse can locate the entry. This
715 * apparently redundant "wait_on_page_locked" lets try_to_unuse
716 * defer to do_swap_page in such a case - in some tests,
717 * do_swap_page and try_to_unuse repeatedly compete.
719 wait_on_page_locked(page);
720 wait_on_page_writeback(page);
722 wait_on_page_writeback(page);
725 * Remove all references to entry, without blocking.
726 * Whenever we reach init_mm, there's no address space
727 * to search, but use it as a reminder to search shmem.
732 if (start_mm == &init_mm)
733 shmem = shmem_unuse(entry, page);
735 retval = unuse_process(start_mm, entry, page);
738 int set_start_mm = (*swap_map >= swcount);
739 struct list_head *p = &start_mm->mmlist;
740 struct mm_struct *new_start_mm = start_mm;
741 struct mm_struct *prev_mm = start_mm;
742 struct mm_struct *mm;
744 atomic_inc(&new_start_mm->mm_users);
745 atomic_inc(&prev_mm->mm_users);
746 spin_lock(&mmlist_lock);
747 while (*swap_map > 1 && !retval &&
748 (p = p->next) != &start_mm->mmlist) {
749 mm = list_entry(p, struct mm_struct, mmlist);
750 atomic_inc(&mm->mm_users);
751 spin_unlock(&mmlist_lock);
760 else if (mm == &init_mm) {
762 shmem = shmem_unuse(entry, page);
764 retval = unuse_process(mm, entry, page);
765 if (set_start_mm && *swap_map < swcount) {
767 atomic_inc(&mm->mm_users);
771 spin_lock(&mmlist_lock);
773 spin_unlock(&mmlist_lock);
776 start_mm = new_start_mm;
780 page_cache_release(page);
785 * How could swap count reach 0x7fff when the maximum
786 * pid is 0x7fff, and there's no way to repeat a swap
787 * page within an mm (except in shmem, where it's the
788 * shared object which takes the reference count)?
789 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
791 * If that's wrong, then we should worry more about
792 * exit_mmap() and do_munmap() cases described above:
793 * we might be resetting SWAP_MAP_MAX too early here.
794 * We know "Undead"s can happen, they're okay, so don't
795 * report them; but do report if we reset SWAP_MAP_MAX.
797 if (*swap_map == SWAP_MAP_MAX) {
798 swap_device_lock(si);
800 swap_device_unlock(si);
805 * If a reference remains (rare), we would like to leave
806 * the page in the swap cache; but try_to_unmap could
807 * then re-duplicate the entry once we drop page lock,
808 * so we might loop indefinitely; also, that page could
809 * not be swapped out to other storage meanwhile. So:
810 * delete from cache even if there's another reference,
811 * after ensuring that the data has been saved to disk -
812 * since if the reference remains (rarer), it will be
813 * read from disk into another page. Splitting into two
814 * pages would be incorrect if swap supported "shared
815 * private" pages, but they are handled by tmpfs files.
817 * Note shmem_unuse already deleted a swappage from
818 * the swap cache, unless the move to filepage failed:
819 * in which case it left swappage in cache, lowered its
820 * swap count to pass quickly through the loops above,
821 * and now we must reincrement count to try again later.
823 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
824 struct writeback_control wbc = {
825 .sync_mode = WB_SYNC_NONE,
828 swap_writepage(page, &wbc);
830 wait_on_page_writeback(page);
832 if (PageSwapCache(page)) {
834 swap_duplicate(entry);
836 delete_from_swap_cache(page);
840 * So we could skip searching mms once swap count went
841 * to 1, we did not mark any present ptes as dirty: must
842 * mark page dirty so shrink_list will preserve it.
846 page_cache_release(page);
849 * Make sure that we aren't completely killing
850 * interactive performance.
856 if (reset_overflow) {
857 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
864 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
865 * corresponds to page offset `offset'.
867 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
869 struct swap_extent *se = sis->curr_swap_extent;
870 struct swap_extent *start_se = se;
873 struct list_head *lh;
875 if (se->start_page <= offset &&
876 offset < (se->start_page + se->nr_pages)) {
877 return se->start_block + (offset - se->start_page);
880 if (lh == &sis->extent_list)
882 se = list_entry(lh, struct swap_extent, list);
883 sis->curr_swap_extent = se;
884 BUG_ON(se == start_se); /* It *must* be present */
889 * Free all of a swapdev's extent information
891 static void destroy_swap_extents(struct swap_info_struct *sis)
893 while (!list_empty(&sis->extent_list)) {
894 struct swap_extent *se;
896 se = list_entry(sis->extent_list.next,
897 struct swap_extent, list);
905 * Add a block range (and the corresponding page range) into this swapdev's
906 * extent list. The extent list is kept sorted in block order.
908 * This function rather assumes that it is called in ascending sector_t order.
909 * It doesn't look for extent coalescing opportunities.
912 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
913 unsigned long nr_pages, sector_t start_block)
915 struct swap_extent *se;
916 struct swap_extent *new_se;
917 struct list_head *lh;
919 lh = sis->extent_list.next; /* The highest-addressed block */
920 while (lh != &sis->extent_list) {
921 se = list_entry(lh, struct swap_extent, list);
922 if (se->start_block + se->nr_pages == start_block &&
923 se->start_page + se->nr_pages == start_page) {
925 se->nr_pages += nr_pages;
932 * No merge. Insert a new extent, preserving ordering.
934 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
937 new_se->start_page = start_page;
938 new_se->nr_pages = nr_pages;
939 new_se->start_block = start_block;
941 lh = sis->extent_list.prev; /* The lowest block */
942 while (lh != &sis->extent_list) {
943 se = list_entry(lh, struct swap_extent, list);
944 if (se->start_block > start_block)
948 list_add_tail(&new_se->list, lh);
954 * A `swap extent' is a simple thing which maps a contiguous range of pages
955 * onto a contiguous range of disk blocks. An ordered list of swap extents
956 * is built at swapon time and is then used at swap_writepage/swap_readpage
957 * time for locating where on disk a page belongs.
959 * If the swapfile is an S_ISBLK block device, a single extent is installed.
960 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
961 * swap files identically.
963 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
964 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
965 * swapfiles are handled *identically* after swapon time.
967 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
968 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
969 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
970 * requirements, they are simply tossed out - we will never use those blocks
973 * For S_ISREG swapfiles we hold i_sem across the life of the swapon. This
974 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
975 * which will scribble on the fs.
977 * The amount of disk space which a single swap extent represents varies.
978 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
979 * extents in the list. To avoid much list walking, we cache the previous
980 * search location in `curr_swap_extent', and start new searches from there.
981 * This is extremely effective. The average number of iterations in
982 * map_swap_page() has been measured at about 0.3 per page. - akpm.
984 static int setup_swap_extents(struct swap_info_struct *sis)
987 unsigned blocks_per_page;
988 unsigned long page_no;
990 sector_t probe_block;
994 inode = sis->swap_file->f_mapping->host;
995 if (S_ISBLK(inode->i_mode)) {
996 ret = add_swap_extent(sis, 0, sis->max, 0);
1000 blkbits = inode->i_blkbits;
1001 blocks_per_page = PAGE_SIZE >> blkbits;
1004 * Map all the blocks into the extent list. This code doesn't try
1009 last_block = i_size_read(inode) >> blkbits;
1010 while ((probe_block + blocks_per_page) <= last_block &&
1011 page_no < sis->max) {
1012 unsigned block_in_page;
1013 sector_t first_block;
1015 first_block = bmap(inode, probe_block);
1016 if (first_block == 0)
1020 * It must be PAGE_SIZE aligned on-disk
1022 if (first_block & (blocks_per_page - 1)) {
1027 for (block_in_page = 1; block_in_page < blocks_per_page;
1031 block = bmap(inode, probe_block + block_in_page);
1034 if (block != first_block + block_in_page) {
1042 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1044 ret = add_swap_extent(sis, page_no, 1,
1045 first_block >> (PAGE_SHIFT - blkbits));
1049 probe_block += blocks_per_page;
1057 sis->highest_bit = page_no - 1;
1059 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1060 struct swap_extent, list);
1063 printk(KERN_ERR "swapon: swapfile has holes\n");
1069 #if 0 /* We don't need this yet */
1070 #include <linux/backing-dev.h>
1071 int page_queue_congested(struct page *page)
1073 struct backing_dev_info *bdi;
1075 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1077 if (PageSwapCache(page)) {
1078 swp_entry_t entry = { .val = page->private };
1079 struct swap_info_struct *sis;
1081 sis = get_swap_info_struct(swp_type(entry));
1082 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1084 bdi = page->mapping->backing_dev_info;
1085 return bdi_write_congested(bdi);
1089 asmlinkage long sys_swapoff(const char __user * specialfile)
1091 struct swap_info_struct * p = NULL;
1092 unsigned short *swap_map;
1093 struct file *swap_file, *victim;
1094 struct address_space *mapping;
1095 struct inode *inode;
1100 if (!capable(CAP_SYS_ADMIN))
1103 pathname = getname(specialfile);
1104 err = PTR_ERR(pathname);
1105 if (IS_ERR(pathname))
1108 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1110 err = PTR_ERR(victim);
1114 mapping = victim->f_mapping;
1117 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1118 p = swap_info + type;
1119 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1120 if (p->swap_file->f_mapping == mapping)
1130 if (!security_vm_enough_memory(p->pages))
1131 vm_unacct_memory(p->pages);
1138 swap_list.head = p->next;
1140 swap_info[prev].next = p->next;
1142 if (type == swap_list.next) {
1143 /* just pick something that's safe... */
1144 swap_list.next = swap_list.head;
1146 nr_swap_pages -= p->pages;
1147 total_swap_pages -= p->pages;
1148 p->flags &= ~SWP_WRITEOK;
1150 current->flags |= PF_SWAPOFF;
1151 err = try_to_unuse(type);
1152 current->flags &= ~PF_SWAPOFF;
1154 /* wait for any unplug function to finish */
1155 down_write(&swap_unplug_sem);
1156 up_write(&swap_unplug_sem);
1159 /* re-insert swap space back into swap_list */
1161 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1162 if (p->prio >= swap_info[i].prio)
1166 swap_list.head = swap_list.next = p - swap_info;
1168 swap_info[prev].next = p - swap_info;
1169 nr_swap_pages += p->pages;
1170 total_swap_pages += p->pages;
1171 p->flags |= SWP_WRITEOK;
1177 swap_device_lock(p);
1178 swap_file = p->swap_file;
1179 p->swap_file = NULL;
1181 swap_map = p->swap_map;
1184 destroy_swap_extents(p);
1185 swap_device_unlock(p);
1189 inode = mapping->host;
1190 if (S_ISBLK(inode->i_mode)) {
1191 struct block_device *bdev = I_BDEV(inode);
1192 set_blocksize(bdev, p->old_block_size);
1195 down(&inode->i_sem);
1196 inode->i_flags &= ~S_SWAPFILE;
1199 filp_close(swap_file, NULL);
1203 filp_close(victim, NULL);
1208 #ifdef CONFIG_PROC_FS
1210 static void *swap_start(struct seq_file *swap, loff_t *pos)
1212 struct swap_info_struct *ptr = swap_info;
1218 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1219 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1228 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1230 struct swap_info_struct *ptr = v;
1231 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1233 for (++ptr; ptr < endptr; ptr++) {
1234 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1243 static void swap_stop(struct seq_file *swap, void *v)
1248 static int swap_show(struct seq_file *swap, void *v)
1250 struct swap_info_struct *ptr = v;
1255 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1257 file = ptr->swap_file;
1258 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1259 seq_printf(swap, "%*s%s\t%d\t%ld\t%d\n",
1260 len < 40 ? 40 - len : 1, " ",
1261 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1262 "partition" : "file\t",
1263 ptr->pages << (PAGE_SHIFT - 10),
1264 ptr->inuse_pages << (PAGE_SHIFT - 10),
1269 static struct seq_operations swaps_op = {
1270 .start = swap_start,
1276 static int swaps_open(struct inode *inode, struct file *file)
1278 return seq_open(file, &swaps_op);
1281 static struct file_operations proc_swaps_operations = {
1284 .llseek = seq_lseek,
1285 .release = seq_release,
1288 static int __init procswaps_init(void)
1290 struct proc_dir_entry *entry;
1292 entry = create_proc_entry("swaps", 0, NULL);
1294 entry->proc_fops = &proc_swaps_operations;
1297 __initcall(procswaps_init);
1298 #endif /* CONFIG_PROC_FS */
1301 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1303 * The swapon system call
1305 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1307 struct swap_info_struct * p;
1309 struct block_device *bdev = NULL;
1310 struct file *swap_file = NULL;
1311 struct address_space *mapping;
1315 static int least_priority;
1316 union swap_header *swap_header = NULL;
1317 int swap_header_version;
1318 int nr_good_pages = 0;
1319 unsigned long maxpages = 1;
1321 unsigned short *swap_map;
1322 struct page *page = NULL;
1323 struct inode *inode = NULL;
1326 if (!capable(CAP_SYS_ADMIN))
1330 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1331 if (!(p->flags & SWP_USED))
1335 * Test if adding another swap device is possible. There are
1336 * two limiting factors: 1) the number of bits for the swap
1337 * type swp_entry_t definition and 2) the number of bits for
1338 * the swap type in the swap ptes as defined by the different
1339 * architectures. To honor both limitations a swap entry
1340 * with swap offset 0 and swap type ~0UL is created, encoded
1341 * to a swap pte, decoded to a swp_entry_t again and finally
1342 * the swap type part is extracted. This will mask all bits
1343 * from the initial ~0UL that can't be encoded in either the
1344 * swp_entry_t or the architecture definition of a swap pte.
1346 if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1350 if (type >= nr_swapfiles)
1351 nr_swapfiles = type+1;
1352 INIT_LIST_HEAD(&p->extent_list);
1353 p->flags = SWP_USED;
1355 p->swap_file = NULL;
1356 p->old_block_size = 0;
1362 p->sdev_lock = SPIN_LOCK_UNLOCKED;
1364 if (swap_flags & SWAP_FLAG_PREFER) {
1366 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1368 p->prio = --least_priority;
1371 name = getname(specialfile);
1372 error = PTR_ERR(name);
1377 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1378 error = PTR_ERR(swap_file);
1379 if (IS_ERR(swap_file)) {
1384 p->swap_file = swap_file;
1385 mapping = swap_file->f_mapping;
1386 inode = mapping->host;
1389 for (i = 0; i < nr_swapfiles; i++) {
1390 struct swap_info_struct *q = &swap_info[i];
1392 if (i == type || !q->swap_file)
1394 if (mapping == q->swap_file->f_mapping)
1399 if (S_ISBLK(inode->i_mode)) {
1400 bdev = I_BDEV(inode);
1401 error = bd_claim(bdev, sys_swapon);
1406 p->old_block_size = block_size(bdev);
1407 error = set_blocksize(bdev, PAGE_SIZE);
1411 } else if (S_ISREG(inode->i_mode)) {
1412 p->bdev = inode->i_sb->s_bdev;
1413 down(&inode->i_sem);
1415 if (IS_SWAPFILE(inode)) {
1423 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1426 * Read the swap header.
1428 if (!mapping->a_ops->readpage) {
1432 page = read_cache_page(mapping, 0,
1433 (filler_t *)mapping->a_ops->readpage, swap_file);
1435 error = PTR_ERR(page);
1438 wait_on_page_locked(page);
1439 if (!PageUptodate(page))
1442 swap_header = page_address(page);
1444 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1445 swap_header_version = 1;
1446 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1447 swap_header_version = 2;
1449 printk("Unable to find swap-space signature\n");
1454 switch (swap_header_version) {
1456 printk(KERN_ERR "version 0 swap is no longer supported. "
1457 "Use mkswap -v1 %s\n", name);
1461 /* Check the swap header's sub-version and the size of
1462 the swap file and bad block lists */
1463 if (swap_header->info.version != 1) {
1465 "Unable to handle swap header version %d\n",
1466 swap_header->info.version);
1473 * Find out how many pages are allowed for a single swap
1474 * device. There are two limiting factors: 1) the number of
1475 * bits for the swap offset in the swp_entry_t type and
1476 * 2) the number of bits in the a swap pte as defined by
1477 * the different architectures. In order to find the
1478 * largest possible bit mask a swap entry with swap type 0
1479 * and swap offset ~0UL is created, encoded to a swap pte,
1480 * decoded to a swp_entry_t again and finally the swap
1481 * offset is extracted. This will mask all the bits from
1482 * the initial ~0UL mask that can't be encoded in either
1483 * the swp_entry_t or the architecture definition of a
1486 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1487 if (maxpages > swap_header->info.last_page)
1488 maxpages = swap_header->info.last_page;
1489 p->highest_bit = maxpages - 1;
1492 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1495 /* OK, set up the swap map and apply the bad block list */
1496 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1502 memset(p->swap_map, 0, maxpages * sizeof(short));
1503 for (i=0; i<swap_header->info.nr_badpages; i++) {
1504 int page = swap_header->info.badpages[i];
1505 if (page <= 0 || page >= swap_header->info.last_page)
1508 p->swap_map[page] = SWAP_MAP_BAD;
1510 nr_good_pages = swap_header->info.last_page -
1511 swap_header->info.nr_badpages -
1512 1 /* header page */;
1517 if (swapfilesize && maxpages > swapfilesize) {
1519 "Swap area shorter than signature indicates\n");
1523 if (!nr_good_pages) {
1524 printk(KERN_WARNING "Empty swap-file\n");
1528 p->swap_map[0] = SWAP_MAP_BAD;
1530 p->pages = nr_good_pages;
1532 error = setup_swap_extents(p);
1538 swap_device_lock(p);
1539 p->flags = SWP_ACTIVE;
1540 nr_swap_pages += nr_good_pages;
1541 total_swap_pages += nr_good_pages;
1542 printk(KERN_INFO "Adding %dk swap on %s. Priority:%d extents:%d\n",
1543 nr_good_pages<<(PAGE_SHIFT-10), name,
1544 p->prio, p->nr_extents);
1546 /* insert swap space into swap_list: */
1548 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1549 if (p->prio >= swap_info[i].prio) {
1556 swap_list.head = swap_list.next = p - swap_info;
1558 swap_info[prev].next = p - swap_info;
1560 swap_device_unlock(p);
1567 set_blocksize(bdev, p->old_block_size);
1572 swap_map = p->swap_map;
1573 p->swap_file = NULL;
1576 if (!(swap_flags & SWAP_FLAG_PREFER))
1579 destroy_swap_extents(p);
1583 filp_close(swap_file, NULL);
1585 if (page && !IS_ERR(page)) {
1587 page_cache_release(page);
1593 inode->i_flags |= S_SWAPFILE;
1599 void si_swapinfo(struct sysinfo *val)
1602 unsigned long nr_to_be_unused = 0;
1605 for (i = 0; i < nr_swapfiles; i++) {
1606 if (!(swap_info[i].flags & SWP_USED) ||
1607 (swap_info[i].flags & SWP_WRITEOK))
1609 nr_to_be_unused += swap_info[i].inuse_pages;
1611 val->freeswap = nr_swap_pages + nr_to_be_unused;
1612 val->totalswap = total_swap_pages + nr_to_be_unused;
1614 if (vx_flags(VXF_VIRT_MEM, 0))
1615 vx_vsi_swapinfo(val);
1619 * Verify that a swap entry is valid and increment its swap map count.
1621 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1622 * "permanent", but will be reclaimed by the next swapoff.
1624 int swap_duplicate(swp_entry_t entry)
1626 struct swap_info_struct * p;
1627 unsigned long offset, type;
1630 type = swp_type(entry);
1631 if (type >= nr_swapfiles)
1633 p = type + swap_info;
1634 offset = swp_offset(entry);
1636 swap_device_lock(p);
1637 if (offset < p->max && p->swap_map[offset]) {
1638 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1639 p->swap_map[offset]++;
1641 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1642 if (swap_overflow++ < 5)
1643 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1644 p->swap_map[offset] = SWAP_MAP_MAX;
1648 swap_device_unlock(p);
1653 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1657 struct swap_info_struct *
1658 get_swap_info_struct(unsigned type)
1660 return &swap_info[type];
1664 * swap_device_lock prevents swap_map being freed. Don't grab an extra
1665 * reference on the swaphandle, it doesn't matter if it becomes unused.
1667 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1669 int ret = 0, i = 1 << page_cluster;
1671 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1673 if (!page_cluster) /* no readahead */
1675 toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1676 if (!toff) /* first page is swap header */
1680 swap_device_lock(swapdev);
1682 /* Don't read-ahead past the end of the swap area */
1683 if (toff >= swapdev->max)
1685 /* Don't read in free or bad pages */
1686 if (!swapdev->swap_map[toff])
1688 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1693 swap_device_unlock(swapdev);