4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
8 #include <linux/config.h>
10 #include <linux/hugetlb.h>
11 #include <linux/mman.h>
12 #include <linux/slab.h>
13 #include <linux/kernel_stat.h>
14 #include <linux/swap.h>
15 #include <linux/vmalloc.h>
16 #include <linux/pagemap.h>
17 #include <linux/namei.h>
18 #include <linux/shm.h>
19 #include <linux/blkdev.h>
20 #include <linux/writeback.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/init.h>
24 #include <linux/module.h>
25 #include <linux/rmap.h>
26 #include <linux/security.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mutex.h>
29 #include <linux/capability.h>
30 #include <linux/syscalls.h>
31 #include <linux/vs_base.h>
32 #include <linux/vs_memory.h>
34 #include <asm/pgtable.h>
35 #include <asm/tlbflush.h>
36 #include <linux/swapops.h>
38 DEFINE_SPINLOCK(swap_lock);
39 unsigned int nr_swapfiles;
40 long total_swap_pages;
41 static int swap_overflow;
43 static const char Bad_file[] = "Bad swap file entry ";
44 static const char Unused_file[] = "Unused swap file entry ";
45 static const char Bad_offset[] = "Bad swap offset entry ";
46 static const char Unused_offset[] = "Unused swap offset entry ";
48 struct swap_list_t swap_list = {-1, -1};
50 static struct swap_info_struct swap_info[MAX_SWAPFILES];
52 static DEFINE_MUTEX(swapon_mutex);
55 * We need this because the bdev->unplug_fn can sleep and we cannot
56 * hold swap_lock while calling the unplug_fn. And swap_lock
57 * cannot be turned into a mutex.
59 static DECLARE_RWSEM(swap_unplug_sem);
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(page);
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(page) above.
75 * If 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 blk_run_backing_dev(bdi, page);
84 up_read(&swap_unplug_sem);
87 #define SWAPFILE_CLUSTER 256
88 #define LATENCY_LIMIT 256
90 static inline unsigned long scan_swap_map(struct swap_info_struct *si)
92 unsigned long offset, last_in_cluster;
93 int latency_ration = LATENCY_LIMIT;
96 * We try to cluster swap pages by allocating them sequentially
97 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
98 * way, however, we resort to first-free allocation, starting
99 * a new cluster. This prevents us from scattering swap pages
100 * all over the entire swap partition, so that we reduce
101 * overall disk seek times between swap pages. -- sct
102 * But we do now try to find an empty cluster. -Andrea
105 si->flags += SWP_SCANNING;
106 if (unlikely(!si->cluster_nr)) {
107 si->cluster_nr = SWAPFILE_CLUSTER - 1;
108 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER)
110 spin_unlock(&swap_lock);
112 offset = si->lowest_bit;
113 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
115 /* Locate the first empty (unaligned) cluster */
116 for (; last_in_cluster <= si->highest_bit; offset++) {
117 if (si->swap_map[offset])
118 last_in_cluster = offset + SWAPFILE_CLUSTER;
119 else if (offset == last_in_cluster) {
120 spin_lock(&swap_lock);
121 si->cluster_next = offset-SWAPFILE_CLUSTER+1;
124 if (unlikely(--latency_ration < 0)) {
126 latency_ration = LATENCY_LIMIT;
129 spin_lock(&swap_lock);
135 offset = si->cluster_next;
136 if (offset > si->highest_bit)
137 lowest: offset = si->lowest_bit;
138 checks: if (!(si->flags & SWP_WRITEOK))
140 if (!si->highest_bit)
142 if (!si->swap_map[offset]) {
143 if (offset == si->lowest_bit)
145 if (offset == si->highest_bit)
148 if (si->inuse_pages == si->pages) {
149 si->lowest_bit = si->max;
152 si->swap_map[offset] = 1;
153 si->cluster_next = offset + 1;
154 si->flags -= SWP_SCANNING;
158 spin_unlock(&swap_lock);
159 while (++offset <= si->highest_bit) {
160 if (!si->swap_map[offset]) {
161 spin_lock(&swap_lock);
164 if (unlikely(--latency_ration < 0)) {
166 latency_ration = LATENCY_LIMIT;
169 spin_lock(&swap_lock);
173 si->flags -= SWP_SCANNING;
177 swp_entry_t get_swap_page(void)
179 struct swap_info_struct *si;
184 spin_lock(&swap_lock);
185 if (nr_swap_pages <= 0)
189 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
190 si = swap_info + type;
193 (!wrapped && si->prio != swap_info[next].prio)) {
194 next = swap_list.head;
198 if (!si->highest_bit)
200 if (!(si->flags & SWP_WRITEOK))
203 swap_list.next = next;
204 offset = scan_swap_map(si);
206 spin_unlock(&swap_lock);
207 return swp_entry(type, offset);
209 next = swap_list.next;
214 spin_unlock(&swap_lock);
215 return (swp_entry_t) {0};
218 swp_entry_t get_swap_page_of_type(int type)
220 struct swap_info_struct *si;
223 spin_lock(&swap_lock);
224 si = swap_info + type;
225 if (si->flags & SWP_WRITEOK) {
227 offset = scan_swap_map(si);
229 spin_unlock(&swap_lock);
230 return swp_entry(type, offset);
234 spin_unlock(&swap_lock);
235 return (swp_entry_t) {0};
238 static struct swap_info_struct * swap_info_get(swp_entry_t entry)
240 struct swap_info_struct * p;
241 unsigned long offset, type;
245 type = swp_type(entry);
246 if (type >= nr_swapfiles)
248 p = & swap_info[type];
249 if (!(p->flags & SWP_USED))
251 offset = swp_offset(entry);
252 if (offset >= p->max)
254 if (!p->swap_map[offset])
256 spin_lock(&swap_lock);
260 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_offset, entry.val);
263 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_offset, entry.val);
266 printk(KERN_ERR "swap_free: %s%08lx\n", Unused_file, entry.val);
269 printk(KERN_ERR "swap_free: %s%08lx\n", Bad_file, entry.val);
274 static int swap_entry_free(struct swap_info_struct *p, unsigned long offset)
276 int count = p->swap_map[offset];
278 if (count < SWAP_MAP_MAX) {
280 p->swap_map[offset] = count;
282 if (offset < p->lowest_bit)
283 p->lowest_bit = offset;
284 if (offset > p->highest_bit)
285 p->highest_bit = offset;
286 if (p->prio > swap_info[swap_list.next].prio)
287 swap_list.next = p - swap_info;
296 * Caller has made sure that the swapdevice corresponding to entry
297 * is still around or has not been recycled.
299 void swap_free(swp_entry_t entry)
301 struct swap_info_struct * p;
303 p = swap_info_get(entry);
305 swap_entry_free(p, swp_offset(entry));
306 spin_unlock(&swap_lock);
311 * How many references to page are currently swapped out?
313 static inline int page_swapcount(struct page *page)
316 struct swap_info_struct *p;
319 entry.val = page_private(page);
320 p = swap_info_get(entry);
322 /* Subtract the 1 for the swap cache itself */
323 count = p->swap_map[swp_offset(entry)] - 1;
324 spin_unlock(&swap_lock);
330 * We can use this swap cache entry directly
331 * if there are no other references to it.
333 int can_share_swap_page(struct page *page)
337 BUG_ON(!PageLocked(page));
338 count = page_mapcount(page);
339 if (count <= 1 && PageSwapCache(page))
340 count += page_swapcount(page);
345 * Work out if there are any other processes sharing this
346 * swap cache page. Free it if you can. Return success.
348 int remove_exclusive_swap_page(struct page *page)
351 struct swap_info_struct * p;
354 BUG_ON(PagePrivate(page));
355 BUG_ON(!PageLocked(page));
357 if (!PageSwapCache(page))
359 if (PageWriteback(page))
361 if (page_count(page) != 2) /* 2: us + cache */
364 entry.val = page_private(page);
365 p = swap_info_get(entry);
369 /* Is the only swap cache user the cache itself? */
371 if (p->swap_map[swp_offset(entry)] == 1) {
372 /* Recheck the page count with the swapcache lock held.. */
373 write_lock_irq(&swapper_space.tree_lock);
374 if ((page_count(page) == 2) && !PageWriteback(page)) {
375 __delete_from_swap_cache(page);
379 write_unlock_irq(&swapper_space.tree_lock);
381 spin_unlock(&swap_lock);
385 page_cache_release(page);
392 * Free the swap entry like above, but also try to
393 * free the page cache entry if it is the last user.
395 void free_swap_and_cache(swp_entry_t entry)
397 struct swap_info_struct * p;
398 struct page *page = NULL;
400 p = swap_info_get(entry);
402 if (swap_entry_free(p, swp_offset(entry)) == 1) {
403 page = find_get_page(&swapper_space, entry.val);
404 if (page && unlikely(TestSetPageLocked(page))) {
405 page_cache_release(page);
409 spin_unlock(&swap_lock);
414 BUG_ON(PagePrivate(page));
415 one_user = (page_count(page) == 2);
416 /* Only cache user (+us), or swap space full? Free it! */
417 /* Also recheck PageSwapCache after page is locked (above) */
418 if (PageSwapCache(page) && !PageWriteback(page) &&
419 (one_user || vm_swap_full())) {
420 delete_from_swap_cache(page);
424 page_cache_release(page);
428 #ifdef CONFIG_SOFTWARE_SUSPEND
430 * Find the swap type that corresponds to given device (if any)
432 * This is needed for software suspend and is done in such a way that inode
433 * aliasing is allowed.
435 int swap_type_of(dev_t device)
439 spin_lock(&swap_lock);
440 for (i = 0; i < nr_swapfiles; i++) {
443 if (!(swap_info[i].flags & SWP_WRITEOK))
447 spin_unlock(&swap_lock);
450 inode = swap_info[i].swap_file->f_dentry->d_inode;
451 if (S_ISBLK(inode->i_mode) &&
452 device == MKDEV(imajor(inode), iminor(inode))) {
453 spin_unlock(&swap_lock);
457 spin_unlock(&swap_lock);
462 * Return either the total number of swap pages of given type, or the number
463 * of free pages of that type (depending on @free)
465 * This is needed for software suspend
467 unsigned int count_swap_pages(int type, int free)
471 if (type < nr_swapfiles) {
472 spin_lock(&swap_lock);
473 if (swap_info[type].flags & SWP_WRITEOK) {
474 n = swap_info[type].pages;
476 n -= swap_info[type].inuse_pages;
478 spin_unlock(&swap_lock);
485 * No need to decide whether this PTE shares the swap entry with others,
486 * just let do_wp_page work it out if a write is requested later - to
487 * force COW, vm_page_prot omits write permission from any private vma.
489 static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
490 unsigned long addr, swp_entry_t entry, struct page *page)
492 inc_mm_counter(vma->vm_mm, anon_rss);
494 set_pte_at(vma->vm_mm, addr, pte,
495 pte_mkold(mk_pte(page, vma->vm_page_prot)));
496 page_add_anon_rmap(page, vma, addr);
499 * Move the page to the active list so it is not
500 * immediately swapped out again after swapon.
505 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
506 unsigned long addr, unsigned long end,
507 swp_entry_t entry, struct page *page)
509 pte_t swp_pte = swp_entry_to_pte(entry);
514 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
517 * swapoff spends a _lot_ of time in this loop!
518 * Test inline before going to call unuse_pte.
520 if (unlikely(pte_same(*pte, swp_pte))) {
521 unuse_pte(vma, pte++, addr, entry, page);
525 } while (pte++, addr += PAGE_SIZE, addr != end);
526 pte_unmap_unlock(pte - 1, ptl);
530 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
531 unsigned long addr, unsigned long end,
532 swp_entry_t entry, struct page *page)
537 pmd = pmd_offset(pud, addr);
539 next = pmd_addr_end(addr, end);
540 if (pmd_none_or_clear_bad(pmd))
542 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
544 } while (pmd++, addr = next, addr != end);
548 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
549 unsigned long addr, unsigned long end,
550 swp_entry_t entry, struct page *page)
555 pud = pud_offset(pgd, addr);
557 next = pud_addr_end(addr, end);
558 if (pud_none_or_clear_bad(pud))
560 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
562 } while (pud++, addr = next, addr != end);
566 static int unuse_vma(struct vm_area_struct *vma,
567 swp_entry_t entry, struct page *page)
570 unsigned long addr, end, next;
573 addr = page_address_in_vma(page, vma);
577 end = addr + PAGE_SIZE;
579 addr = vma->vm_start;
583 pgd = pgd_offset(vma->vm_mm, addr);
585 next = pgd_addr_end(addr, end);
586 if (pgd_none_or_clear_bad(pgd))
588 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
590 } while (pgd++, addr = next, addr != end);
594 static int unuse_mm(struct mm_struct *mm,
595 swp_entry_t entry, struct page *page)
597 struct vm_area_struct *vma;
599 if (!down_read_trylock(&mm->mmap_sem)) {
601 * Activate page so shrink_cache is unlikely to unmap its
602 * ptes while lock is dropped, so swapoff can make progress.
606 down_read(&mm->mmap_sem);
609 for (vma = mm->mmap; vma; vma = vma->vm_next) {
610 if (vma->anon_vma && unuse_vma(vma, entry, page))
613 up_read(&mm->mmap_sem);
615 * Currently unuse_mm cannot fail, but leave error handling
616 * at call sites for now, since we change it from time to time.
621 #ifdef CONFIG_MIGRATION
622 int remove_vma_swap(struct vm_area_struct *vma, struct page *page)
624 swp_entry_t entry = { .val = page_private(page) };
626 return unuse_vma(vma, entry, page);
631 * Scan swap_map from current position to next entry still in use.
632 * Recycle to start on reaching the end, returning 0 when empty.
634 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
637 unsigned int max = si->max;
638 unsigned int i = prev;
642 * No need for swap_lock here: we're just looking
643 * for whether an entry is in use, not modifying it; false
644 * hits are okay, and sys_swapoff() has already prevented new
645 * allocations from this area (while holding swap_lock).
654 * No entries in use at top of swap_map,
655 * loop back to start and recheck there.
661 count = si->swap_map[i];
662 if (count && count != SWAP_MAP_BAD)
669 * We completely avoid races by reading each swap page in advance,
670 * and then search for the process using it. All the necessary
671 * page table adjustments can then be made atomically.
673 static int try_to_unuse(unsigned int type)
675 struct swap_info_struct * si = &swap_info[type];
676 struct mm_struct *start_mm;
677 unsigned short *swap_map;
678 unsigned short swcount;
683 int reset_overflow = 0;
687 * When searching mms for an entry, a good strategy is to
688 * start at the first mm we freed the previous entry from
689 * (though actually we don't notice whether we or coincidence
690 * freed the entry). Initialize this start_mm with a hold.
692 * A simpler strategy would be to start at the last mm we
693 * freed the previous entry from; but that would take less
694 * advantage of mmlist ordering, which clusters forked mms
695 * together, child after parent. If we race with dup_mmap(), we
696 * prefer to resolve parent before child, lest we miss entries
697 * duplicated after we scanned child: using last mm would invert
698 * that. Though it's only a serious concern when an overflowed
699 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
702 atomic_inc(&init_mm.mm_users);
705 * Keep on scanning until all entries have gone. Usually,
706 * one pass through swap_map is enough, but not necessarily:
707 * there are races when an instance of an entry might be missed.
709 while ((i = find_next_to_unuse(si, i)) != 0) {
710 if (signal_pending(current)) {
716 * Get a page for the entry, using the existing swap
717 * cache page if there is one. Otherwise, get a clean
718 * page and read the swap into it.
720 swap_map = &si->swap_map[i];
721 entry = swp_entry(type, i);
723 page = read_swap_cache_async(entry, NULL, 0);
726 * Either swap_duplicate() failed because entry
727 * has been freed independently, and will not be
728 * reused since sys_swapoff() already disabled
729 * allocation from here, or alloc_page() failed.
738 * Don't hold on to start_mm if it looks like exiting.
740 if (atomic_read(&start_mm->mm_users) == 1) {
743 atomic_inc(&init_mm.mm_users);
747 * Wait for and lock page. When do_swap_page races with
748 * try_to_unuse, do_swap_page can handle the fault much
749 * faster than try_to_unuse can locate the entry. This
750 * apparently redundant "wait_on_page_locked" lets try_to_unuse
751 * defer to do_swap_page in such a case - in some tests,
752 * do_swap_page and try_to_unuse repeatedly compete.
754 wait_on_page_locked(page);
755 wait_on_page_writeback(page);
757 if (!PageSwapCache(page)) {
758 /* Page migration has occured */
760 page_cache_release(page);
763 wait_on_page_writeback(page);
766 * Remove all references to entry.
767 * Whenever we reach init_mm, there's no address space
768 * to search, but use it as a reminder to search shmem.
773 if (start_mm == &init_mm)
774 shmem = shmem_unuse(entry, page);
776 retval = unuse_mm(start_mm, entry, page);
779 int set_start_mm = (*swap_map >= swcount);
780 struct list_head *p = &start_mm->mmlist;
781 struct mm_struct *new_start_mm = start_mm;
782 struct mm_struct *prev_mm = start_mm;
783 struct mm_struct *mm;
785 atomic_inc(&new_start_mm->mm_users);
786 atomic_inc(&prev_mm->mm_users);
787 spin_lock(&mmlist_lock);
788 while (*swap_map > 1 && !retval &&
789 (p = p->next) != &start_mm->mmlist) {
790 mm = list_entry(p, struct mm_struct, mmlist);
791 if (atomic_inc_return(&mm->mm_users) == 1) {
792 atomic_dec(&mm->mm_users);
795 spin_unlock(&mmlist_lock);
804 else if (mm == &init_mm) {
806 shmem = shmem_unuse(entry, page);
808 retval = unuse_mm(mm, entry, page);
809 if (set_start_mm && *swap_map < swcount) {
811 atomic_inc(&mm->mm_users);
815 spin_lock(&mmlist_lock);
817 spin_unlock(&mmlist_lock);
820 start_mm = new_start_mm;
824 page_cache_release(page);
829 * How could swap count reach 0x7fff when the maximum
830 * pid is 0x7fff, and there's no way to repeat a swap
831 * page within an mm (except in shmem, where it's the
832 * shared object which takes the reference count)?
833 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
835 * If that's wrong, then we should worry more about
836 * exit_mmap() and do_munmap() cases described above:
837 * we might be resetting SWAP_MAP_MAX too early here.
838 * We know "Undead"s can happen, they're okay, so don't
839 * report them; but do report if we reset SWAP_MAP_MAX.
841 if (*swap_map == SWAP_MAP_MAX) {
842 spin_lock(&swap_lock);
844 spin_unlock(&swap_lock);
849 * If a reference remains (rare), we would like to leave
850 * the page in the swap cache; but try_to_unmap could
851 * then re-duplicate the entry once we drop page lock,
852 * so we might loop indefinitely; also, that page could
853 * not be swapped out to other storage meanwhile. So:
854 * delete from cache even if there's another reference,
855 * after ensuring that the data has been saved to disk -
856 * since if the reference remains (rarer), it will be
857 * read from disk into another page. Splitting into two
858 * pages would be incorrect if swap supported "shared
859 * private" pages, but they are handled by tmpfs files.
861 * Note shmem_unuse already deleted a swappage from
862 * the swap cache, unless the move to filepage failed:
863 * in which case it left swappage in cache, lowered its
864 * swap count to pass quickly through the loops above,
865 * and now we must reincrement count to try again later.
867 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
868 struct writeback_control wbc = {
869 .sync_mode = WB_SYNC_NONE,
872 swap_writepage(page, &wbc);
874 wait_on_page_writeback(page);
876 if (PageSwapCache(page)) {
878 swap_duplicate(entry);
880 delete_from_swap_cache(page);
884 * So we could skip searching mms once swap count went
885 * to 1, we did not mark any present ptes as dirty: must
886 * mark page dirty so shrink_list will preserve it.
890 page_cache_release(page);
893 * Make sure that we aren't completely killing
894 * interactive performance.
900 if (reset_overflow) {
901 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
908 * After a successful try_to_unuse, if no swap is now in use, we know
909 * we can empty the mmlist. swap_lock must be held on entry and exit.
910 * Note that mmlist_lock nests inside swap_lock, and an mm must be
911 * added to the mmlist just after page_duplicate - before would be racy.
913 static void drain_mmlist(void)
915 struct list_head *p, *next;
918 for (i = 0; i < nr_swapfiles; i++)
919 if (swap_info[i].inuse_pages)
921 spin_lock(&mmlist_lock);
922 list_for_each_safe(p, next, &init_mm.mmlist)
924 spin_unlock(&mmlist_lock);
928 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
929 * corresponds to page offset `offset'.
931 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
933 struct swap_extent *se = sis->curr_swap_extent;
934 struct swap_extent *start_se = se;
937 struct list_head *lh;
939 if (se->start_page <= offset &&
940 offset < (se->start_page + se->nr_pages)) {
941 return se->start_block + (offset - se->start_page);
944 if (lh == &sis->extent_list)
946 se = list_entry(lh, struct swap_extent, list);
947 sis->curr_swap_extent = se;
948 BUG_ON(se == start_se); /* It *must* be present */
953 * Free all of a swapdev's extent information
955 static void destroy_swap_extents(struct swap_info_struct *sis)
957 while (!list_empty(&sis->extent_list)) {
958 struct swap_extent *se;
960 se = list_entry(sis->extent_list.next,
961 struct swap_extent, list);
968 * Add a block range (and the corresponding page range) into this swapdev's
969 * extent list. The extent list is kept sorted in page order.
971 * This function rather assumes that it is called in ascending page order.
974 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
975 unsigned long nr_pages, sector_t start_block)
977 struct swap_extent *se;
978 struct swap_extent *new_se;
979 struct list_head *lh;
981 lh = sis->extent_list.prev; /* The highest page extent */
982 if (lh != &sis->extent_list) {
983 se = list_entry(lh, struct swap_extent, list);
984 BUG_ON(se->start_page + se->nr_pages != start_page);
985 if (se->start_block + se->nr_pages == start_block) {
987 se->nr_pages += nr_pages;
993 * No merge. Insert a new extent, preserving ordering.
995 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
998 new_se->start_page = start_page;
999 new_se->nr_pages = nr_pages;
1000 new_se->start_block = start_block;
1002 list_add_tail(&new_se->list, &sis->extent_list);
1007 * A `swap extent' is a simple thing which maps a contiguous range of pages
1008 * onto a contiguous range of disk blocks. An ordered list of swap extents
1009 * is built at swapon time and is then used at swap_writepage/swap_readpage
1010 * time for locating where on disk a page belongs.
1012 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1013 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1014 * swap files identically.
1016 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1017 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1018 * swapfiles are handled *identically* after swapon time.
1020 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1021 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1022 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1023 * requirements, they are simply tossed out - we will never use those blocks
1026 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1027 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1028 * which will scribble on the fs.
1030 * The amount of disk space which a single swap extent represents varies.
1031 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1032 * extents in the list. To avoid much list walking, we cache the previous
1033 * search location in `curr_swap_extent', and start new searches from there.
1034 * This is extremely effective. The average number of iterations in
1035 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1037 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1039 struct inode *inode;
1040 unsigned blocks_per_page;
1041 unsigned long page_no;
1043 sector_t probe_block;
1044 sector_t last_block;
1045 sector_t lowest_block = -1;
1046 sector_t highest_block = 0;
1050 inode = sis->swap_file->f_mapping->host;
1051 if (S_ISBLK(inode->i_mode)) {
1052 ret = add_swap_extent(sis, 0, sis->max, 0);
1057 blkbits = inode->i_blkbits;
1058 blocks_per_page = PAGE_SIZE >> blkbits;
1061 * Map all the blocks into the extent list. This code doesn't try
1066 last_block = i_size_read(inode) >> blkbits;
1067 while ((probe_block + blocks_per_page) <= last_block &&
1068 page_no < sis->max) {
1069 unsigned block_in_page;
1070 sector_t first_block;
1072 first_block = bmap(inode, probe_block);
1073 if (first_block == 0)
1077 * It must be PAGE_SIZE aligned on-disk
1079 if (first_block & (blocks_per_page - 1)) {
1084 for (block_in_page = 1; block_in_page < blocks_per_page;
1088 block = bmap(inode, probe_block + block_in_page);
1091 if (block != first_block + block_in_page) {
1098 first_block >>= (PAGE_SHIFT - blkbits);
1099 if (page_no) { /* exclude the header page */
1100 if (first_block < lowest_block)
1101 lowest_block = first_block;
1102 if (first_block > highest_block)
1103 highest_block = first_block;
1107 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1109 ret = add_swap_extent(sis, page_no, 1, first_block);
1114 probe_block += blocks_per_page;
1119 *span = 1 + highest_block - lowest_block;
1121 page_no = 1; /* force Empty message */
1123 sis->pages = page_no - 1;
1124 sis->highest_bit = page_no - 1;
1126 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1127 struct swap_extent, list);
1130 printk(KERN_ERR "swapon: swapfile has holes\n");
1136 #if 0 /* We don't need this yet */
1137 #include <linux/backing-dev.h>
1138 int page_queue_congested(struct page *page)
1140 struct backing_dev_info *bdi;
1142 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1144 if (PageSwapCache(page)) {
1145 swp_entry_t entry = { .val = page_private(page) };
1146 struct swap_info_struct *sis;
1148 sis = get_swap_info_struct(swp_type(entry));
1149 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1151 bdi = page->mapping->backing_dev_info;
1152 return bdi_write_congested(bdi);
1156 asmlinkage long sys_swapoff(const char __user * specialfile)
1158 struct swap_info_struct * p = NULL;
1159 unsigned short *swap_map;
1160 struct file *swap_file, *victim;
1161 struct address_space *mapping;
1162 struct inode *inode;
1167 if (!capable(CAP_SYS_ADMIN))
1170 pathname = getname(specialfile);
1171 err = PTR_ERR(pathname);
1172 if (IS_ERR(pathname))
1175 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1177 err = PTR_ERR(victim);
1181 mapping = victim->f_mapping;
1183 spin_lock(&swap_lock);
1184 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1185 p = swap_info + type;
1186 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1187 if (p->swap_file->f_mapping == mapping)
1194 spin_unlock(&swap_lock);
1197 if (!security_vm_enough_memory(p->pages))
1198 vm_unacct_memory(p->pages);
1201 spin_unlock(&swap_lock);
1205 swap_list.head = p->next;
1207 swap_info[prev].next = p->next;
1209 if (type == swap_list.next) {
1210 /* just pick something that's safe... */
1211 swap_list.next = swap_list.head;
1213 nr_swap_pages -= p->pages;
1214 total_swap_pages -= p->pages;
1215 p->flags &= ~SWP_WRITEOK;
1216 spin_unlock(&swap_lock);
1218 current->flags |= PF_SWAPOFF;
1219 err = try_to_unuse(type);
1220 current->flags &= ~PF_SWAPOFF;
1223 /* re-insert swap space back into swap_list */
1224 spin_lock(&swap_lock);
1225 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1226 if (p->prio >= swap_info[i].prio)
1230 swap_list.head = swap_list.next = p - swap_info;
1232 swap_info[prev].next = p - swap_info;
1233 nr_swap_pages += p->pages;
1234 total_swap_pages += p->pages;
1235 p->flags |= SWP_WRITEOK;
1236 spin_unlock(&swap_lock);
1240 /* wait for any unplug function to finish */
1241 down_write(&swap_unplug_sem);
1242 up_write(&swap_unplug_sem);
1244 destroy_swap_extents(p);
1245 mutex_lock(&swapon_mutex);
1246 spin_lock(&swap_lock);
1249 /* wait for anyone still in scan_swap_map */
1250 p->highest_bit = 0; /* cuts scans short */
1251 while (p->flags >= SWP_SCANNING) {
1252 spin_unlock(&swap_lock);
1253 schedule_timeout_uninterruptible(1);
1254 spin_lock(&swap_lock);
1257 swap_file = p->swap_file;
1258 p->swap_file = NULL;
1260 swap_map = p->swap_map;
1263 spin_unlock(&swap_lock);
1264 mutex_unlock(&swapon_mutex);
1266 inode = mapping->host;
1267 if (S_ISBLK(inode->i_mode)) {
1268 struct block_device *bdev = I_BDEV(inode);
1269 set_blocksize(bdev, p->old_block_size);
1272 mutex_lock(&inode->i_mutex);
1273 inode->i_flags &= ~S_SWAPFILE;
1274 mutex_unlock(&inode->i_mutex);
1276 filp_close(swap_file, NULL);
1280 filp_close(victim, NULL);
1285 #ifdef CONFIG_PROC_FS
1287 static void *swap_start(struct seq_file *swap, loff_t *pos)
1289 struct swap_info_struct *ptr = swap_info;
1293 mutex_lock(&swapon_mutex);
1295 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1296 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1305 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1307 struct swap_info_struct *ptr = v;
1308 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1310 for (++ptr; ptr < endptr; ptr++) {
1311 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1320 static void swap_stop(struct seq_file *swap, void *v)
1322 mutex_unlock(&swapon_mutex);
1325 static int swap_show(struct seq_file *swap, void *v)
1327 struct swap_info_struct *ptr = v;
1332 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1334 file = ptr->swap_file;
1335 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1336 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1337 len < 40 ? 40 - len : 1, " ",
1338 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1339 "partition" : "file\t",
1340 ptr->pages << (PAGE_SHIFT - 10),
1341 ptr->inuse_pages << (PAGE_SHIFT - 10),
1346 static struct seq_operations swaps_op = {
1347 .start = swap_start,
1353 static int swaps_open(struct inode *inode, struct file *file)
1355 return seq_open(file, &swaps_op);
1358 static struct file_operations proc_swaps_operations = {
1361 .llseek = seq_lseek,
1362 .release = seq_release,
1365 static int __init procswaps_init(void)
1367 struct proc_dir_entry *entry;
1369 entry = create_proc_entry("swaps", 0, NULL);
1371 entry->proc_fops = &proc_swaps_operations;
1374 __initcall(procswaps_init);
1375 #endif /* CONFIG_PROC_FS */
1378 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1380 * The swapon system call
1382 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1384 struct swap_info_struct * p;
1386 struct block_device *bdev = NULL;
1387 struct file *swap_file = NULL;
1388 struct address_space *mapping;
1392 static int least_priority;
1393 union swap_header *swap_header = NULL;
1394 int swap_header_version;
1395 unsigned int nr_good_pages = 0;
1398 unsigned long maxpages = 1;
1400 unsigned short *swap_map;
1401 struct page *page = NULL;
1402 struct inode *inode = NULL;
1405 if (!capable(CAP_SYS_ADMIN))
1407 spin_lock(&swap_lock);
1409 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1410 if (!(p->flags & SWP_USED))
1414 * Test if adding another swap device is possible. There are
1415 * two limiting factors: 1) the number of bits for the swap
1416 * type swp_entry_t definition and 2) the number of bits for
1417 * the swap type in the swap ptes as defined by the different
1418 * architectures. To honor both limitations a swap entry
1419 * with swap offset 0 and swap type ~0UL is created, encoded
1420 * to a swap pte, decoded to a swp_entry_t again and finally
1421 * the swap type part is extracted. This will mask all bits
1422 * from the initial ~0UL that can't be encoded in either the
1423 * swp_entry_t or the architecture definition of a swap pte.
1425 if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1426 spin_unlock(&swap_lock);
1429 if (type >= nr_swapfiles)
1430 nr_swapfiles = type+1;
1431 INIT_LIST_HEAD(&p->extent_list);
1432 p->flags = SWP_USED;
1433 p->swap_file = NULL;
1434 p->old_block_size = 0;
1441 if (swap_flags & SWAP_FLAG_PREFER) {
1443 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1445 p->prio = --least_priority;
1447 spin_unlock(&swap_lock);
1448 name = getname(specialfile);
1449 error = PTR_ERR(name);
1454 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1455 error = PTR_ERR(swap_file);
1456 if (IS_ERR(swap_file)) {
1461 p->swap_file = swap_file;
1462 mapping = swap_file->f_mapping;
1463 inode = mapping->host;
1466 for (i = 0; i < nr_swapfiles; i++) {
1467 struct swap_info_struct *q = &swap_info[i];
1469 if (i == type || !q->swap_file)
1471 if (mapping == q->swap_file->f_mapping)
1476 if (S_ISBLK(inode->i_mode)) {
1477 bdev = I_BDEV(inode);
1478 error = bd_claim(bdev, sys_swapon);
1484 p->old_block_size = block_size(bdev);
1485 error = set_blocksize(bdev, PAGE_SIZE);
1489 } else if (S_ISREG(inode->i_mode)) {
1490 p->bdev = inode->i_sb->s_bdev;
1491 mutex_lock(&inode->i_mutex);
1493 if (IS_SWAPFILE(inode)) {
1501 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1504 * Read the swap header.
1506 if (!mapping->a_ops->readpage) {
1510 page = read_cache_page(mapping, 0,
1511 (filler_t *)mapping->a_ops->readpage, swap_file);
1513 error = PTR_ERR(page);
1516 wait_on_page_locked(page);
1517 if (!PageUptodate(page))
1520 swap_header = page_address(page);
1522 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1523 swap_header_version = 1;
1524 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1525 swap_header_version = 2;
1527 printk(KERN_ERR "Unable to find swap-space signature\n");
1532 switch (swap_header_version) {
1534 printk(KERN_ERR "version 0 swap is no longer supported. "
1535 "Use mkswap -v1 %s\n", name);
1539 /* Check the swap header's sub-version and the size of
1540 the swap file and bad block lists */
1541 if (swap_header->info.version != 1) {
1543 "Unable to handle swap header version %d\n",
1544 swap_header->info.version);
1550 p->cluster_next = 1;
1553 * Find out how many pages are allowed for a single swap
1554 * device. There are two limiting factors: 1) the number of
1555 * bits for the swap offset in the swp_entry_t type and
1556 * 2) the number of bits in the a swap pte as defined by
1557 * the different architectures. In order to find the
1558 * largest possible bit mask a swap entry with swap type 0
1559 * and swap offset ~0UL is created, encoded to a swap pte,
1560 * decoded to a swp_entry_t again and finally the swap
1561 * offset is extracted. This will mask all the bits from
1562 * the initial ~0UL mask that can't be encoded in either
1563 * the swp_entry_t or the architecture definition of a
1566 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1567 if (maxpages > swap_header->info.last_page)
1568 maxpages = swap_header->info.last_page;
1569 p->highest_bit = maxpages - 1;
1574 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1576 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1579 /* OK, set up the swap map and apply the bad block list */
1580 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1586 memset(p->swap_map, 0, maxpages * sizeof(short));
1587 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1588 int page_nr = swap_header->info.badpages[i];
1589 if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
1592 p->swap_map[page_nr] = SWAP_MAP_BAD;
1594 nr_good_pages = swap_header->info.last_page -
1595 swap_header->info.nr_badpages -
1596 1 /* header page */;
1601 if (swapfilesize && maxpages > swapfilesize) {
1603 "Swap area shorter than signature indicates\n");
1607 if (nr_good_pages) {
1608 p->swap_map[0] = SWAP_MAP_BAD;
1610 p->pages = nr_good_pages;
1611 nr_extents = setup_swap_extents(p, &span);
1612 if (nr_extents < 0) {
1616 nr_good_pages = p->pages;
1618 if (!nr_good_pages) {
1619 printk(KERN_WARNING "Empty swap-file\n");
1624 mutex_lock(&swapon_mutex);
1625 spin_lock(&swap_lock);
1626 p->flags = SWP_ACTIVE;
1627 nr_swap_pages += nr_good_pages;
1628 total_swap_pages += nr_good_pages;
1630 printk(KERN_INFO "Adding %uk swap on %s. "
1631 "Priority:%d extents:%d across:%lluk\n",
1632 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1633 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
1635 /* insert swap space into swap_list: */
1637 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1638 if (p->prio >= swap_info[i].prio) {
1645 swap_list.head = swap_list.next = p - swap_info;
1647 swap_info[prev].next = p - swap_info;
1649 spin_unlock(&swap_lock);
1650 mutex_unlock(&swapon_mutex);
1655 set_blocksize(bdev, p->old_block_size);
1658 destroy_swap_extents(p);
1660 spin_lock(&swap_lock);
1661 swap_map = p->swap_map;
1662 p->swap_file = NULL;
1665 if (!(swap_flags & SWAP_FLAG_PREFER))
1667 spin_unlock(&swap_lock);
1670 filp_close(swap_file, NULL);
1672 if (page && !IS_ERR(page)) {
1674 page_cache_release(page);
1680 inode->i_flags |= S_SWAPFILE;
1681 mutex_unlock(&inode->i_mutex);
1686 void si_swapinfo(struct sysinfo *val)
1689 unsigned long nr_to_be_unused = 0;
1691 spin_lock(&swap_lock);
1692 for (i = 0; i < nr_swapfiles; i++) {
1693 if (!(swap_info[i].flags & SWP_USED) ||
1694 (swap_info[i].flags & SWP_WRITEOK))
1696 nr_to_be_unused += swap_info[i].inuse_pages;
1698 val->freeswap = nr_swap_pages + nr_to_be_unused;
1699 val->totalswap = total_swap_pages + nr_to_be_unused;
1700 spin_unlock(&swap_lock);
1701 if (vx_flags(VXF_VIRT_MEM, 0))
1702 vx_vsi_swapinfo(val);
1706 * Verify that a swap entry is valid and increment its swap map count.
1708 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1709 * "permanent", but will be reclaimed by the next swapoff.
1711 int swap_duplicate(swp_entry_t entry)
1713 struct swap_info_struct * p;
1714 unsigned long offset, type;
1717 type = swp_type(entry);
1718 if (type >= nr_swapfiles)
1720 p = type + swap_info;
1721 offset = swp_offset(entry);
1723 spin_lock(&swap_lock);
1724 if (offset < p->max && p->swap_map[offset]) {
1725 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1726 p->swap_map[offset]++;
1728 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1729 if (swap_overflow++ < 5)
1730 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1731 p->swap_map[offset] = SWAP_MAP_MAX;
1735 spin_unlock(&swap_lock);
1740 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1744 struct swap_info_struct *
1745 get_swap_info_struct(unsigned type)
1747 return &swap_info[type];
1751 * swap_lock prevents swap_map being freed. Don't grab an extra
1752 * reference on the swaphandle, it doesn't matter if it becomes unused.
1754 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1756 int ret = 0, i = 1 << page_cluster;
1758 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1760 if (!page_cluster) /* no readahead */
1762 toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1763 if (!toff) /* first page is swap header */
1767 spin_lock(&swap_lock);
1769 /* Don't read-ahead past the end of the swap area */
1770 if (toff >= swapdev->max)
1772 /* Don't read in free or bad pages */
1773 if (!swapdev->swap_map[toff])
1775 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1780 spin_unlock(&swap_lock);