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>
34 spinlock_t swaplock = SPIN_LOCK_UNLOCKED;
35 unsigned int nr_swapfiles;
36 long total_swap_pages;
37 static int swap_overflow;
39 EXPORT_SYMBOL(total_swap_pages);
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 struct swap_info_struct swap_info[MAX_SWAPFILES];
50 static DECLARE_MUTEX(swapon_sem);
53 * We need this because the bdev->unplug_fn can sleep and we cannot
54 * hold swap_list_lock while calling the unplug_fn. And swap_list_lock
55 * cannot be turned into a semaphore.
57 static DECLARE_RWSEM(swap_unplug_sem);
59 #define SWAPFILE_CLUSTER 256
61 void swap_unplug_io_fn(struct backing_dev_info *unused_bdi, struct page *page)
65 down_read(&swap_unplug_sem);
66 entry.val = page->private;
67 if (PageSwapCache(page)) {
68 struct block_device *bdev = swap_info[swp_type(entry)].bdev;
69 struct backing_dev_info *bdi;
72 * If the page is removed from swapcache from under us (with a
73 * racy try_to_unuse/swapoff) we need an additional reference
74 * count to avoid reading garbage from page->private above. If
75 * the WARN_ON triggers during a swapoff it maybe the race
76 * condition and it's harmless. However if it triggers without
77 * swapoff it signals a problem.
79 WARN_ON(page_count(page) <= 1);
81 bdi = bdev->bd_inode->i_mapping->backing_dev_info;
82 bdi->unplug_io_fn(bdi, page);
84 up_read(&swap_unplug_sem);
87 static inline int scan_swap_map(struct swap_info_struct *si)
91 * We try to cluster swap pages by allocating them
92 * sequentially in swap. Once we've allocated
93 * SWAPFILE_CLUSTER pages this way, however, we resort to
94 * first-free allocation, starting a new cluster. This
95 * prevents us from scattering swap pages all over the entire
96 * swap partition, so that we reduce overall disk seek times
97 * between swap pages. -- sct */
99 while (si->cluster_next <= si->highest_bit) {
100 offset = si->cluster_next++;
101 if (si->swap_map[offset])
107 si->cluster_nr = SWAPFILE_CLUSTER;
109 /* try to find an empty (even not aligned) cluster. */
110 offset = si->lowest_bit;
112 if (offset+SWAPFILE_CLUSTER-1 <= si->highest_bit)
115 for (nr = offset; nr < offset+SWAPFILE_CLUSTER; nr++)
116 if (si->swap_map[nr])
119 goto check_next_cluster;
121 /* We found a completly empty cluster, so start
126 /* No luck, so now go finegrined as usual. -Andrea */
127 for (offset = si->lowest_bit; offset <= si->highest_bit ; offset++) {
128 if (si->swap_map[offset])
130 si->lowest_bit = offset+1;
132 if (offset == si->lowest_bit)
134 if (offset == si->highest_bit)
136 if (si->lowest_bit > si->highest_bit) {
137 si->lowest_bit = si->max;
140 si->swap_map[offset] = 1;
143 si->cluster_next = offset+1;
146 si->lowest_bit = si->max;
151 swp_entry_t get_swap_page(void)
153 struct swap_info_struct * p;
154 unsigned long offset;
156 int type, wrapped = 0;
158 entry.val = 0; /* Out of memory */
160 type = swap_list.next;
163 if (nr_swap_pages <= 0)
167 p = &swap_info[type];
168 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
170 offset = scan_swap_map(p);
171 swap_device_unlock(p);
173 entry = swp_entry(type,offset);
174 type = swap_info[type].next;
176 p->prio != swap_info[type].prio) {
177 swap_list.next = swap_list.head;
179 swap_list.next = type;
186 if (type < 0 || p->prio != swap_info[type].prio) {
187 type = swap_list.head;
192 goto out; /* out of swap space */
199 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
201 struct swap_info_struct * p;
202 unsigned long offset, type;
206 type = swp_type(entry);
207 if (type >= nr_swapfiles)
209 p = & swap_info[type];
210 if (!(p->flags & SWP_USED))
212 offset = swp_offset(entry);
213 if (offset >= p->max)
215 if (!p->swap_map[offset])
218 if (p->prio > swap_info[swap_list.next].prio)
219 swap_list.next = type;
224 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
227 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
230 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
233 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
238 static void swap_info_put(struct swap_info_struct * p)
240 swap_device_unlock(p);
244 static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
246 int count = p->swap_map[offset];
248 if (count < SWAP_MAP_MAX) {
250 p->swap_map[offset] = count;
252 if (offset < p->lowest_bit)
253 p->lowest_bit = offset;
254 if (offset > p->highest_bit)
255 p->highest_bit = offset;
264 * Caller has made sure that the swapdevice corresponding to entry
265 * is still around or has not been recycled.
267 void swap_free(swp_entry_t entry)
269 struct swap_info_struct * p;
271 p = swap_info_get(entry);
273 swap_entry_free(p, swp_offset(entry));
279 * Check if we're the only user of a swap page,
280 * when the page is locked.
282 static int exclusive_swap_page(struct page *page)
285 struct swap_info_struct * p;
288 entry.val = page->private;
289 p = swap_info_get(entry);
291 /* Is the only swap cache user the cache itself? */
292 if (p->swap_map[swp_offset(entry)] == 1) {
293 /* Recheck the page count with the swapcache lock held.. */
294 spin_lock_irq(&swapper_space.tree_lock);
295 if (page_count(page) == 2)
297 spin_unlock_irq(&swapper_space.tree_lock);
305 * We can use this swap cache entry directly
306 * if there are no other references to it.
308 * Here "exclusive_swap_page()" does the real
309 * work, but we opportunistically check whether
310 * we need to get all the locks first..
312 int can_share_swap_page(struct page *page)
316 if (!PageLocked(page))
318 switch (page_count(page)) {
320 if (!PagePrivate(page))
324 if (!PageSwapCache(page))
326 retval = exclusive_swap_page(page);
329 if (PageReserved(page))
337 * Work out if there are any other processes sharing this
338 * swap cache page. Free it if you can. Return success.
340 int remove_exclusive_swap_page(struct page *page)
343 struct swap_info_struct * p;
346 BUG_ON(PagePrivate(page));
347 BUG_ON(!PageLocked(page));
349 if (!PageSwapCache(page))
351 if (PageWriteback(page))
353 if (page_count(page) != 2) /* 2: us + cache */
356 entry.val = page->private;
357 p = swap_info_get(entry);
361 /* Is the only swap cache user the cache itself? */
363 if (p->swap_map[swp_offset(entry)] == 1) {
364 /* Recheck the page count with the swapcache lock held.. */
365 spin_lock_irq(&swapper_space.tree_lock);
366 if ((page_count(page) == 2) && !PageWriteback(page)) {
367 __delete_from_swap_cache(page);
371 spin_unlock_irq(&swapper_space.tree_lock);
377 page_cache_release(page);
384 * Free the swap entry like above, but also try to
385 * free the page cache entry if it is the last user.
387 void free_swap_and_cache(swp_entry_t entry)
389 struct swap_info_struct * p;
390 struct page *page = NULL;
392 p = swap_info_get(entry);
394 if (swap_entry_free(p, swp_offset(entry)) == 1) {
395 spin_lock_irq(&swapper_space.tree_lock);
396 page = radix_tree_lookup(&swapper_space.page_tree,
398 if (page && TestSetPageLocked(page))
400 spin_unlock_irq(&swapper_space.tree_lock);
407 BUG_ON(PagePrivate(page));
408 page_cache_get(page);
409 one_user = (page_count(page) == 2);
410 /* Only cache user (+us), or swap space full? Free it! */
411 if (!PageWriteback(page) && (one_user || vm_swap_full())) {
412 delete_from_swap_cache(page);
416 page_cache_release(page);
421 * The swap entry has been read in advance, and we return 1 to indicate
422 * that the page has been used or is no longer needed.
424 * Always set the resulting pte to be nowrite (the same as COW pages
425 * after one process has exited). We don't know just how many PTEs will
426 * share this swap entry, so be cautious and let do_wp_page work out
427 * what to do if a write is requested later.
429 /* vma->vm_mm->page_table_lock is held */
431 unuse_pte(struct vm_area_struct *vma, unsigned long address, pte_t *dir,
432 swp_entry_t entry, struct page *page)
434 // vma->vm_mm->rss++;
435 vx_rsspages_inc(vma->vm_mm);
437 set_pte(dir, pte_mkold(mk_pte(page, vma->vm_page_prot)));
438 page_add_anon_rmap(page, vma, address);
442 /* vma->vm_mm->page_table_lock is held */
443 static unsigned long unuse_pmd(struct vm_area_struct * vma, pmd_t *dir,
444 unsigned long address, unsigned long size, unsigned long offset,
445 swp_entry_t entry, struct page *page)
449 pte_t swp_pte = swp_entry_to_pte(entry);
458 pte = pte_offset_map(dir, address);
459 offset += address & PMD_MASK;
460 address &= ~PMD_MASK;
461 end = address + size;
466 * swapoff spends a _lot_ of time in this loop!
467 * Test inline before going to call unuse_pte.
469 if (unlikely(pte_same(*pte, swp_pte))) {
470 unuse_pte(vma, offset + address, pte, entry, page);
474 * Move the page to the active list so it is not
475 * immediately swapped out again after swapon.
479 /* add 1 since address may be 0 */
480 return 1 + offset + address;
482 address += PAGE_SIZE;
484 } while (address && (address < end));
489 /* vma->vm_mm->page_table_lock is held */
490 static unsigned long unuse_pgd(struct vm_area_struct * vma, pgd_t *dir,
491 unsigned long address, unsigned long size,
492 swp_entry_t entry, struct page *page)
495 unsigned long offset, end;
496 unsigned long foundaddr;
505 pmd = pmd_offset(dir, address);
506 offset = address & PGDIR_MASK;
507 address &= ~PGDIR_MASK;
508 end = address + size;
509 if (end > PGDIR_SIZE)
514 foundaddr = unuse_pmd(vma, pmd, address, end - address,
515 offset, entry, page);
518 address = (address + PMD_SIZE) & PMD_MASK;
520 } while (address && (address < end));
524 /* vma->vm_mm->page_table_lock is held */
525 static unsigned long unuse_vma(struct vm_area_struct * vma, pgd_t *pgdir,
526 swp_entry_t entry, struct page *page)
528 unsigned long start = vma->vm_start, end = vma->vm_end;
529 unsigned long foundaddr;
534 foundaddr = unuse_pgd(vma, pgdir, start, end - start,
538 start = (start + PGDIR_SIZE) & PGDIR_MASK;
540 } while (start && (start < end));
544 static int unuse_process(struct mm_struct * mm,
545 swp_entry_t entry, struct page* page)
547 struct vm_area_struct* vma;
548 unsigned long foundaddr = 0;
551 * Go through process' page directory.
553 if (!down_read_trylock(&mm->mmap_sem)) {
555 * Our reference to the page stops try_to_unmap_one from
556 * unmapping its ptes, so swapoff can make progress.
559 down_read(&mm->mmap_sem);
562 spin_lock(&mm->page_table_lock);
563 for (vma = mm->mmap; vma; vma = vma->vm_next) {
564 if (!is_vm_hugetlb_page(vma)) {
565 pgd_t * pgd = pgd_offset(mm, vma->vm_start);
566 foundaddr = unuse_vma(vma, pgd, entry, page);
571 spin_unlock(&mm->page_table_lock);
572 up_read(&mm->mmap_sem);
574 * Currently unuse_process cannot fail, but leave error handling
575 * at call sites for now, since we change it from time to time.
581 * Scan swap_map from current position to next entry still in use.
582 * Recycle to start on reaching the end, returning 0 when empty.
584 static int find_next_to_unuse(struct swap_info_struct *si, int prev)
591 * No need for swap_device_lock(si) here: we're just looking
592 * for whether an entry is in use, not modifying it; false
593 * hits are okay, and sys_swapoff() has already prevented new
594 * allocations from this area (while holding swap_list_lock()).
603 * No entries in use at top of swap_map,
604 * loop back to start and recheck there.
610 count = si->swap_map[i];
611 if (count && count != SWAP_MAP_BAD)
618 * We completely avoid races by reading each swap page in advance,
619 * and then search for the process using it. All the necessary
620 * page table adjustments can then be made atomically.
622 static int try_to_unuse(unsigned int type)
624 struct swap_info_struct * si = &swap_info[type];
625 struct mm_struct *start_mm;
626 unsigned short *swap_map;
627 unsigned short swcount;
632 int reset_overflow = 0;
636 * When searching mms for an entry, a good strategy is to
637 * start at the first mm we freed the previous entry from
638 * (though actually we don't notice whether we or coincidence
639 * freed the entry). Initialize this start_mm with a hold.
641 * A simpler strategy would be to start at the last mm we
642 * freed the previous entry from; but that would take less
643 * advantage of mmlist ordering (now preserved by swap_out()),
644 * which clusters forked address spaces together, most recent
645 * child immediately after parent. If we race with dup_mmap(),
646 * we very much want to resolve parent before child, otherwise
647 * we may miss some entries: using last mm would invert that.
650 atomic_inc(&init_mm.mm_users);
653 * Keep on scanning until all entries have gone. Usually,
654 * one pass through swap_map is enough, but not necessarily:
655 * mmput() removes mm from mmlist before exit_mmap() and its
656 * zap_page_range(). That's not too bad, those entries are
657 * on their way out, and handled faster there than here.
658 * do_munmap() behaves similarly, taking the range out of mm's
659 * vma list before zap_page_range(). But unfortunately, when
660 * unmapping a part of a vma, it takes the whole out first,
661 * then reinserts what's left after (might even reschedule if
662 * open() method called) - so swap entries may be invisible
663 * to swapoff for a while, then reappear - but that is rare.
665 while ((i = find_next_to_unuse(si, i)) != 0) {
666 if (signal_pending(current)) {
672 * Get a page for the entry, using the existing swap
673 * cache page if there is one. Otherwise, get a clean
674 * page and read the swap into it.
676 swap_map = &si->swap_map[i];
677 entry = swp_entry(type, i);
678 page = read_swap_cache_async(entry, NULL, 0);
681 * Either swap_duplicate() failed because entry
682 * has been freed independently, and will not be
683 * reused since sys_swapoff() already disabled
684 * allocation from here, or alloc_page() failed.
693 * Don't hold on to start_mm if it looks like exiting.
695 if (atomic_read(&start_mm->mm_users) == 1) {
698 atomic_inc(&init_mm.mm_users);
702 * Wait for and lock page. When do_swap_page races with
703 * try_to_unuse, do_swap_page can handle the fault much
704 * faster than try_to_unuse can locate the entry. This
705 * apparently redundant "wait_on_page_locked" lets try_to_unuse
706 * defer to do_swap_page in such a case - in some tests,
707 * do_swap_page and try_to_unuse repeatedly compete.
709 wait_on_page_locked(page);
710 wait_on_page_writeback(page);
712 wait_on_page_writeback(page);
715 * Remove all references to entry, without blocking.
716 * Whenever we reach init_mm, there's no address space
717 * to search, but use it as a reminder to search shmem.
722 if (start_mm == &init_mm)
723 shmem = shmem_unuse(entry, page);
725 retval = unuse_process(start_mm, entry, page);
728 int set_start_mm = (*swap_map >= swcount);
729 struct list_head *p = &start_mm->mmlist;
730 struct mm_struct *new_start_mm = start_mm;
731 struct mm_struct *prev_mm = start_mm;
732 struct mm_struct *mm;
734 atomic_inc(&new_start_mm->mm_users);
735 atomic_inc(&prev_mm->mm_users);
736 spin_lock(&mmlist_lock);
737 while (*swap_map > 1 && !retval &&
738 (p = p->next) != &start_mm->mmlist) {
739 mm = list_entry(p, struct mm_struct, mmlist);
740 atomic_inc(&mm->mm_users);
741 spin_unlock(&mmlist_lock);
750 else if (mm == &init_mm) {
752 shmem = shmem_unuse(entry, page);
754 retval = unuse_process(mm, entry, page);
755 if (set_start_mm && *swap_map < swcount) {
757 atomic_inc(&mm->mm_users);
761 spin_lock(&mmlist_lock);
763 spin_unlock(&mmlist_lock);
766 start_mm = new_start_mm;
770 page_cache_release(page);
775 * How could swap count reach 0x7fff when the maximum
776 * pid is 0x7fff, and there's no way to repeat a swap
777 * page within an mm (except in shmem, where it's the
778 * shared object which takes the reference count)?
779 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
781 * If that's wrong, then we should worry more about
782 * exit_mmap() and do_munmap() cases described above:
783 * we might be resetting SWAP_MAP_MAX too early here.
784 * We know "Undead"s can happen, they're okay, so don't
785 * report them; but do report if we reset SWAP_MAP_MAX.
787 if (*swap_map == SWAP_MAP_MAX) {
788 swap_device_lock(si);
790 swap_device_unlock(si);
795 * If a reference remains (rare), we would like to leave
796 * the page in the swap cache; but try_to_unmap could
797 * then re-duplicate the entry once we drop page lock,
798 * so we might loop indefinitely; also, that page could
799 * not be swapped out to other storage meanwhile. So:
800 * delete from cache even if there's another reference,
801 * after ensuring that the data has been saved to disk -
802 * since if the reference remains (rarer), it will be
803 * read from disk into another page. Splitting into two
804 * pages would be incorrect if swap supported "shared
805 * private" pages, but they are handled by tmpfs files.
807 * Note shmem_unuse already deleted a swappage from
808 * the swap cache, unless the move to filepage failed:
809 * in which case it left swappage in cache, lowered its
810 * swap count to pass quickly through the loops above,
811 * and now we must reincrement count to try again later.
813 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
814 struct writeback_control wbc = {
815 .sync_mode = WB_SYNC_NONE,
818 swap_writepage(page, &wbc);
820 wait_on_page_writeback(page);
822 if (PageSwapCache(page)) {
824 swap_duplicate(entry);
826 delete_from_swap_cache(page);
830 * So we could skip searching mms once swap count went
831 * to 1, we did not mark any present ptes as dirty: must
832 * mark page dirty so shrink_list will preserve it.
836 page_cache_release(page);
839 * Make sure that we aren't completely killing
840 * interactive performance.
846 if (reset_overflow) {
847 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
854 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
855 * corresponds to page offset `offset'.
857 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
859 struct swap_extent *se = sis->curr_swap_extent;
860 struct swap_extent *start_se = se;
863 struct list_head *lh;
865 if (se->start_page <= offset &&
866 offset < (se->start_page + se->nr_pages)) {
867 return se->start_block + (offset - se->start_page);
870 if (lh == &sis->extent_list)
872 se = list_entry(lh, struct swap_extent, list);
873 sis->curr_swap_extent = se;
874 BUG_ON(se == start_se); /* It *must* be present */
879 * Free all of a swapdev's extent information
881 static void destroy_swap_extents(struct swap_info_struct *sis)
883 while (!list_empty(&sis->extent_list)) {
884 struct swap_extent *se;
886 se = list_entry(sis->extent_list.next,
887 struct swap_extent, list);
895 * Add a block range (and the corresponding page range) into this swapdev's
896 * extent list. The extent list is kept sorted in block order.
898 * This function rather assumes that it is called in ascending sector_t order.
899 * It doesn't look for extent coalescing opportunities.
902 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
903 unsigned long nr_pages, sector_t start_block)
905 struct swap_extent *se;
906 struct swap_extent *new_se;
907 struct list_head *lh;
909 lh = sis->extent_list.next; /* The highest-addressed block */
910 while (lh != &sis->extent_list) {
911 se = list_entry(lh, struct swap_extent, list);
912 if (se->start_block + se->nr_pages == start_block &&
913 se->start_page + se->nr_pages == start_page) {
915 se->nr_pages += nr_pages;
922 * No merge. Insert a new extent, preserving ordering.
924 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
927 new_se->start_page = start_page;
928 new_se->nr_pages = nr_pages;
929 new_se->start_block = start_block;
931 lh = sis->extent_list.prev; /* The lowest block */
932 while (lh != &sis->extent_list) {
933 se = list_entry(lh, struct swap_extent, list);
934 if (se->start_block > start_block)
938 list_add_tail(&new_se->list, lh);
944 * A `swap extent' is a simple thing which maps a contiguous range of pages
945 * onto a contiguous range of disk blocks. An ordered list of swap extents
946 * is built at swapon time and is then used at swap_writepage/swap_readpage
947 * time for locating where on disk a page belongs.
949 * If the swapfile is an S_ISBLK block device, a single extent is installed.
950 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
951 * swap files identically.
953 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
954 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
955 * swapfiles are handled *identically* after swapon time.
957 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
958 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
959 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
960 * requirements, they are simply tossed out - we will never use those blocks
963 * For S_ISREG swapfiles we hold i_sem across the life of the swapon. This
964 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
965 * which will scribble on the fs.
967 * The amount of disk space which a single swap extent represents varies.
968 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
969 * extents in the list. To avoid much list walking, we cache the previous
970 * search location in `curr_swap_extent', and start new searches from there.
971 * This is extremely effective. The average number of iterations in
972 * map_swap_page() has been measured at about 0.3 per page. - akpm.
974 static int setup_swap_extents(struct swap_info_struct *sis)
977 unsigned blocks_per_page;
978 unsigned long page_no;
980 sector_t probe_block;
984 inode = sis->swap_file->f_mapping->host;
985 if (S_ISBLK(inode->i_mode)) {
986 ret = add_swap_extent(sis, 0, sis->max, 0);
990 blkbits = inode->i_blkbits;
991 blocks_per_page = PAGE_SIZE >> blkbits;
994 * Map all the blocks into the extent list. This code doesn't try
999 last_block = i_size_read(inode) >> blkbits;
1000 while ((probe_block + blocks_per_page) <= last_block &&
1001 page_no < sis->max) {
1002 unsigned block_in_page;
1003 sector_t first_block;
1005 first_block = bmap(inode, probe_block);
1006 if (first_block == 0)
1010 * It must be PAGE_SIZE aligned on-disk
1012 if (first_block & (blocks_per_page - 1)) {
1017 for (block_in_page = 1; block_in_page < blocks_per_page;
1021 block = bmap(inode, probe_block + block_in_page);
1024 if (block != first_block + block_in_page) {
1032 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1034 ret = add_swap_extent(sis, page_no, 1,
1035 first_block >> (PAGE_SHIFT - blkbits));
1039 probe_block += blocks_per_page;
1047 sis->highest_bit = page_no - 1;
1049 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1050 struct swap_extent, list);
1053 printk(KERN_ERR "swapon: swapfile has holes\n");
1059 #if 0 /* We don't need this yet */
1060 #include <linux/backing-dev.h>
1061 int page_queue_congested(struct page *page)
1063 struct backing_dev_info *bdi;
1065 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1067 if (PageSwapCache(page)) {
1068 swp_entry_t entry = { .val = page->private };
1069 struct swap_info_struct *sis;
1071 sis = get_swap_info_struct(swp_type(entry));
1072 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1074 bdi = page->mapping->backing_dev_info;
1075 return bdi_write_congested(bdi);
1079 asmlinkage long sys_swapoff(const char __user * specialfile)
1081 struct swap_info_struct * p = NULL;
1082 unsigned short *swap_map;
1083 struct file *swap_file, *victim;
1084 struct address_space *mapping;
1085 struct inode *inode;
1090 if (!capable(CAP_SYS_ADMIN))
1093 pathname = getname(specialfile);
1094 err = PTR_ERR(pathname);
1095 if (IS_ERR(pathname))
1098 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1100 err = PTR_ERR(victim);
1104 mapping = victim->f_mapping;
1107 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1108 p = swap_info + type;
1109 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1110 if (p->swap_file->f_mapping == mapping)
1120 if (!security_vm_enough_memory(p->pages))
1121 vm_unacct_memory(p->pages);
1128 swap_list.head = p->next;
1130 swap_info[prev].next = p->next;
1132 if (type == swap_list.next) {
1133 /* just pick something that's safe... */
1134 swap_list.next = swap_list.head;
1136 nr_swap_pages -= p->pages;
1137 total_swap_pages -= p->pages;
1138 p->flags &= ~SWP_WRITEOK;
1140 current->flags |= PF_SWAPOFF;
1141 err = try_to_unuse(type);
1142 current->flags &= ~PF_SWAPOFF;
1144 /* wait for any unplug function to finish */
1145 down_write(&swap_unplug_sem);
1146 up_write(&swap_unplug_sem);
1149 /* re-insert swap space back into swap_list */
1151 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1152 if (p->prio >= swap_info[i].prio)
1156 swap_list.head = swap_list.next = p - swap_info;
1158 swap_info[prev].next = p - swap_info;
1159 nr_swap_pages += p->pages;
1160 total_swap_pages += p->pages;
1161 p->flags |= SWP_WRITEOK;
1167 swap_device_lock(p);
1168 swap_file = p->swap_file;
1169 p->swap_file = NULL;
1171 swap_map = p->swap_map;
1174 destroy_swap_extents(p);
1175 swap_device_unlock(p);
1179 inode = mapping->host;
1180 if (S_ISBLK(inode->i_mode)) {
1181 struct block_device *bdev = I_BDEV(inode);
1182 set_blocksize(bdev, p->old_block_size);
1185 down(&inode->i_sem);
1186 inode->i_flags &= ~S_SWAPFILE;
1189 filp_close(swap_file, NULL);
1193 filp_close(victim, NULL);
1198 #ifdef CONFIG_PROC_FS
1200 static void *swap_start(struct seq_file *swap, loff_t *pos)
1202 struct swap_info_struct *ptr = swap_info;
1208 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1209 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1218 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1220 struct swap_info_struct *ptr = v;
1221 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1223 for (++ptr; ptr < endptr; ptr++) {
1224 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1233 static void swap_stop(struct seq_file *swap, void *v)
1238 static int swap_show(struct seq_file *swap, void *v)
1240 struct swap_info_struct *ptr = v;
1245 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1247 file = ptr->swap_file;
1248 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1249 seq_printf(swap, "%*s%s\t%d\t%ld\t%d\n",
1250 len < 40 ? 40 - len : 1, " ",
1251 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1252 "partition" : "file\t",
1253 ptr->pages << (PAGE_SHIFT - 10),
1254 ptr->inuse_pages << (PAGE_SHIFT - 10),
1259 static struct seq_operations swaps_op = {
1260 .start = swap_start,
1266 static int swaps_open(struct inode *inode, struct file *file)
1268 return seq_open(file, &swaps_op);
1271 static struct file_operations proc_swaps_operations = {
1274 .llseek = seq_lseek,
1275 .release = seq_release,
1278 static int __init procswaps_init(void)
1280 struct proc_dir_entry *entry;
1282 entry = create_proc_entry("swaps", 0, NULL);
1284 entry->proc_fops = &proc_swaps_operations;
1287 __initcall(procswaps_init);
1288 #endif /* CONFIG_PROC_FS */
1291 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1293 * The swapon system call
1295 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1297 struct swap_info_struct * p;
1299 struct block_device *bdev = NULL;
1300 struct file *swap_file = NULL;
1301 struct address_space *mapping;
1305 static int least_priority;
1306 union swap_header *swap_header = NULL;
1307 int swap_header_version;
1308 int nr_good_pages = 0;
1309 unsigned long maxpages = 1;
1311 unsigned short *swap_map;
1312 struct page *page = NULL;
1313 struct inode *inode = NULL;
1316 if (!capable(CAP_SYS_ADMIN))
1320 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1321 if (!(p->flags & SWP_USED))
1325 * Test if adding another swap device is possible. There are
1326 * two limiting factors: 1) the number of bits for the swap
1327 * type swp_entry_t definition and 2) the number of bits for
1328 * the swap type in the swap ptes as defined by the different
1329 * architectures. To honor both limitations a swap entry
1330 * with swap offset 0 and swap type ~0UL is created, encoded
1331 * to a swap pte, decoded to a swp_entry_t again and finally
1332 * the swap type part is extracted. This will mask all bits
1333 * from the initial ~0UL that can't be encoded in either the
1334 * swp_entry_t or the architecture definition of a swap pte.
1336 if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1340 if (type >= nr_swapfiles)
1341 nr_swapfiles = type+1;
1342 INIT_LIST_HEAD(&p->extent_list);
1343 p->flags = SWP_USED;
1345 p->swap_file = NULL;
1346 p->old_block_size = 0;
1352 p->sdev_lock = SPIN_LOCK_UNLOCKED;
1354 if (swap_flags & SWAP_FLAG_PREFER) {
1356 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1358 p->prio = --least_priority;
1361 name = getname(specialfile);
1362 error = PTR_ERR(name);
1367 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1368 error = PTR_ERR(swap_file);
1369 if (IS_ERR(swap_file)) {
1374 p->swap_file = swap_file;
1375 mapping = swap_file->f_mapping;
1376 inode = mapping->host;
1379 for (i = 0; i < nr_swapfiles; i++) {
1380 struct swap_info_struct *q = &swap_info[i];
1382 if (i == type || !q->swap_file)
1384 if (mapping == q->swap_file->f_mapping)
1389 if (S_ISBLK(inode->i_mode)) {
1390 bdev = I_BDEV(inode);
1391 error = bd_claim(bdev, sys_swapon);
1396 p->old_block_size = block_size(bdev);
1397 error = set_blocksize(bdev, PAGE_SIZE);
1401 } else if (S_ISREG(inode->i_mode)) {
1402 p->bdev = inode->i_sb->s_bdev;
1403 down(&inode->i_sem);
1405 if (IS_SWAPFILE(inode)) {
1413 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1416 * Read the swap header.
1418 if (!mapping->a_ops->readpage) {
1422 page = read_cache_page(mapping, 0,
1423 (filler_t *)mapping->a_ops->readpage, swap_file);
1425 error = PTR_ERR(page);
1428 wait_on_page_locked(page);
1429 if (!PageUptodate(page))
1432 swap_header = page_address(page);
1434 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1435 swap_header_version = 1;
1436 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1437 swap_header_version = 2;
1439 printk("Unable to find swap-space signature\n");
1444 switch (swap_header_version) {
1446 printk(KERN_ERR "version 0 swap is no longer supported. "
1447 "Use mkswap -v1 %s\n", name);
1451 /* Check the swap header's sub-version and the size of
1452 the swap file and bad block lists */
1453 if (swap_header->info.version != 1) {
1455 "Unable to handle swap header version %d\n",
1456 swap_header->info.version);
1463 * Find out how many pages are allowed for a single swap
1464 * device. There are two limiting factors: 1) the number of
1465 * bits for the swap offset in the swp_entry_t type and
1466 * 2) the number of bits in the a swap pte as defined by
1467 * the different architectures. In order to find the
1468 * largest possible bit mask a swap entry with swap type 0
1469 * and swap offset ~0UL is created, encoded to a swap pte,
1470 * decoded to a swp_entry_t again and finally the swap
1471 * offset is extracted. This will mask all the bits from
1472 * the initial ~0UL mask that can't be encoded in either
1473 * the swp_entry_t or the architecture definition of a
1476 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1477 if (maxpages > swap_header->info.last_page)
1478 maxpages = swap_header->info.last_page;
1479 p->highest_bit = maxpages - 1;
1482 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1485 /* OK, set up the swap map and apply the bad block list */
1486 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1492 memset(p->swap_map, 0, maxpages * sizeof(short));
1493 for (i=0; i<swap_header->info.nr_badpages; i++) {
1494 int page = swap_header->info.badpages[i];
1495 if (page <= 0 || page >= swap_header->info.last_page)
1498 p->swap_map[page] = SWAP_MAP_BAD;
1500 nr_good_pages = swap_header->info.last_page -
1501 swap_header->info.nr_badpages -
1502 1 /* header page */;
1507 if (swapfilesize && maxpages > swapfilesize) {
1509 "Swap area shorter than signature indicates\n");
1513 if (!nr_good_pages) {
1514 printk(KERN_WARNING "Empty swap-file\n");
1518 p->swap_map[0] = SWAP_MAP_BAD;
1520 p->pages = nr_good_pages;
1522 error = setup_swap_extents(p);
1528 swap_device_lock(p);
1529 p->flags = SWP_ACTIVE;
1530 nr_swap_pages += nr_good_pages;
1531 total_swap_pages += nr_good_pages;
1532 printk(KERN_INFO "Adding %dk swap on %s. Priority:%d extents:%d\n",
1533 nr_good_pages<<(PAGE_SHIFT-10), name,
1534 p->prio, p->nr_extents);
1536 /* insert swap space into swap_list: */
1538 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1539 if (p->prio >= swap_info[i].prio) {
1546 swap_list.head = swap_list.next = p - swap_info;
1548 swap_info[prev].next = p - swap_info;
1550 swap_device_unlock(p);
1557 set_blocksize(bdev, p->old_block_size);
1562 swap_map = p->swap_map;
1563 p->swap_file = NULL;
1566 if (!(swap_flags & SWAP_FLAG_PREFER))
1569 destroy_swap_extents(p);
1573 filp_close(swap_file, NULL);
1575 if (page && !IS_ERR(page)) {
1577 page_cache_release(page);
1583 inode->i_flags |= S_SWAPFILE;
1589 void si_swapinfo(struct sysinfo *val)
1592 unsigned long nr_to_be_unused = 0;
1595 for (i = 0; i < nr_swapfiles; i++) {
1596 if (!(swap_info[i].flags & SWP_USED) ||
1597 (swap_info[i].flags & SWP_WRITEOK))
1599 nr_to_be_unused += swap_info[i].inuse_pages;
1601 val->freeswap = nr_swap_pages + nr_to_be_unused;
1602 val->totalswap = total_swap_pages + nr_to_be_unused;
1604 if (vx_flags(VXF_VIRT_MEM, 0))
1605 vx_vsi_swapinfo(val);
1609 * Verify that a swap entry is valid and increment its swap map count.
1611 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1612 * "permanent", but will be reclaimed by the next swapoff.
1614 int swap_duplicate(swp_entry_t entry)
1616 struct swap_info_struct * p;
1617 unsigned long offset, type;
1620 type = swp_type(entry);
1621 if (type >= nr_swapfiles)
1623 p = type + swap_info;
1624 offset = swp_offset(entry);
1626 swap_device_lock(p);
1627 if (offset < p->max && p->swap_map[offset]) {
1628 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1629 p->swap_map[offset]++;
1631 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1632 if (swap_overflow++ < 5)
1633 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1634 p->swap_map[offset] = SWAP_MAP_MAX;
1638 swap_device_unlock(p);
1643 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1647 struct swap_info_struct *
1648 get_swap_info_struct(unsigned type)
1650 return &swap_info[type];
1654 * swap_device_lock prevents swap_map being freed. Don't grab an extra
1655 * reference on the swaphandle, it doesn't matter if it becomes unused.
1657 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1659 int ret = 0, i = 1 << page_cluster;
1661 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1663 if (!page_cluster) /* no readahead */
1665 toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1666 if (!toff) /* first page is swap header */
1670 swap_device_lock(swapdev);
1672 /* Don't read-ahead past the end of the swap area */
1673 if (toff >= swapdev->max)
1675 /* Don't read in free or bad pages */
1676 if (!swapdev->swap_map[toff])
1678 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1683 swap_device_unlock(swapdev);