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 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_trylock_page(&swapper_space, entry.val);
404 spin_unlock(&swap_lock);
409 BUG_ON(PagePrivate(page));
410 page_cache_get(page);
411 one_user = (page_count(page) == 2);
412 /* Only cache user (+us), or swap space full? Free it! */
413 if (!PageWriteback(page) && (one_user || vm_swap_full())) {
414 delete_from_swap_cache(page);
418 page_cache_release(page);
423 * No need to decide whether this PTE shares the swap entry with others,
424 * just let do_wp_page work it out if a write is requested later - to
425 * force COW, vm_page_prot omits write permission from any private vma.
427 static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
428 unsigned long addr, swp_entry_t entry, struct page *page)
430 inc_mm_counter(vma->vm_mm, anon_rss);
432 set_pte_at(vma->vm_mm, addr, pte,
433 pte_mkold(mk_pte(page, vma->vm_page_prot)));
434 page_add_anon_rmap(page, vma, addr);
437 * Move the page to the active list so it is not
438 * immediately swapped out again after swapon.
443 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
444 unsigned long addr, unsigned long end,
445 swp_entry_t entry, struct page *page)
447 pte_t swp_pte = swp_entry_to_pte(entry);
452 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
455 * swapoff spends a _lot_ of time in this loop!
456 * Test inline before going to call unuse_pte.
458 if (unlikely(pte_same(*pte, swp_pte))) {
459 unuse_pte(vma, pte++, addr, entry, page);
463 } while (pte++, addr += PAGE_SIZE, addr != end);
464 pte_unmap_unlock(pte - 1, ptl);
468 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
469 unsigned long addr, unsigned long end,
470 swp_entry_t entry, struct page *page)
475 pmd = pmd_offset(pud, addr);
477 next = pmd_addr_end(addr, end);
478 if (pmd_none_or_clear_bad(pmd))
480 if (unuse_pte_range(vma, pmd, addr, next, entry, page))
482 } while (pmd++, addr = next, addr != end);
486 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
487 unsigned long addr, unsigned long end,
488 swp_entry_t entry, struct page *page)
493 pud = pud_offset(pgd, addr);
495 next = pud_addr_end(addr, end);
496 if (pud_none_or_clear_bad(pud))
498 if (unuse_pmd_range(vma, pud, addr, next, entry, page))
500 } while (pud++, addr = next, addr != end);
504 static int unuse_vma(struct vm_area_struct *vma,
505 swp_entry_t entry, struct page *page)
508 unsigned long addr, end, next;
511 addr = page_address_in_vma(page, vma);
515 end = addr + PAGE_SIZE;
517 addr = vma->vm_start;
521 pgd = pgd_offset(vma->vm_mm, addr);
523 next = pgd_addr_end(addr, end);
524 if (pgd_none_or_clear_bad(pgd))
526 if (unuse_pud_range(vma, pgd, addr, next, entry, page))
528 } while (pgd++, addr = next, addr != end);
532 static int unuse_mm(struct mm_struct *mm,
533 swp_entry_t entry, struct page *page)
535 struct vm_area_struct *vma;
537 if (!down_read_trylock(&mm->mmap_sem)) {
539 * Activate page so shrink_cache is unlikely to unmap its
540 * ptes while lock is dropped, so swapoff can make progress.
544 down_read(&mm->mmap_sem);
547 for (vma = mm->mmap; vma; vma = vma->vm_next) {
548 if (vma->anon_vma && unuse_vma(vma, entry, page))
551 up_read(&mm->mmap_sem);
553 * Currently unuse_mm cannot fail, but leave error handling
554 * at call sites for now, since we change it from time to time.
559 #ifdef CONFIG_MIGRATION
560 int remove_vma_swap(struct vm_area_struct *vma, struct page *page)
562 swp_entry_t entry = { .val = page_private(page) };
564 return unuse_vma(vma, entry, page);
569 * Scan swap_map from current position to next entry still in use.
570 * Recycle to start on reaching the end, returning 0 when empty.
572 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
575 unsigned int max = si->max;
576 unsigned int i = prev;
580 * No need for swap_lock here: we're just looking
581 * for whether an entry is in use, not modifying it; false
582 * hits are okay, and sys_swapoff() has already prevented new
583 * allocations from this area (while holding swap_lock).
592 * No entries in use at top of swap_map,
593 * loop back to start and recheck there.
599 count = si->swap_map[i];
600 if (count && count != SWAP_MAP_BAD)
607 * We completely avoid races by reading each swap page in advance,
608 * and then search for the process using it. All the necessary
609 * page table adjustments can then be made atomically.
611 static int try_to_unuse(unsigned int type)
613 struct swap_info_struct * si = &swap_info[type];
614 struct mm_struct *start_mm;
615 unsigned short *swap_map;
616 unsigned short swcount;
621 int reset_overflow = 0;
625 * When searching mms for an entry, a good strategy is to
626 * start at the first mm we freed the previous entry from
627 * (though actually we don't notice whether we or coincidence
628 * freed the entry). Initialize this start_mm with a hold.
630 * A simpler strategy would be to start at the last mm we
631 * freed the previous entry from; but that would take less
632 * advantage of mmlist ordering, which clusters forked mms
633 * together, child after parent. If we race with dup_mmap(), we
634 * prefer to resolve parent before child, lest we miss entries
635 * duplicated after we scanned child: using last mm would invert
636 * that. Though it's only a serious concern when an overflowed
637 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
640 atomic_inc(&init_mm.mm_users);
643 * Keep on scanning until all entries have gone. Usually,
644 * one pass through swap_map is enough, but not necessarily:
645 * there are races when an instance of an entry might be missed.
647 while ((i = find_next_to_unuse(si, i)) != 0) {
648 if (signal_pending(current)) {
654 * Get a page for the entry, using the existing swap
655 * cache page if there is one. Otherwise, get a clean
656 * page and read the swap into it.
658 swap_map = &si->swap_map[i];
659 entry = swp_entry(type, i);
661 page = read_swap_cache_async(entry, NULL, 0);
664 * Either swap_duplicate() failed because entry
665 * has been freed independently, and will not be
666 * reused since sys_swapoff() already disabled
667 * allocation from here, or alloc_page() failed.
676 * Don't hold on to start_mm if it looks like exiting.
678 if (atomic_read(&start_mm->mm_users) == 1) {
681 atomic_inc(&init_mm.mm_users);
685 * Wait for and lock page. When do_swap_page races with
686 * try_to_unuse, do_swap_page can handle the fault much
687 * faster than try_to_unuse can locate the entry. This
688 * apparently redundant "wait_on_page_locked" lets try_to_unuse
689 * defer to do_swap_page in such a case - in some tests,
690 * do_swap_page and try_to_unuse repeatedly compete.
692 wait_on_page_locked(page);
693 wait_on_page_writeback(page);
695 if (!PageSwapCache(page)) {
696 /* Page migration has occured */
698 page_cache_release(page);
701 wait_on_page_writeback(page);
704 * Remove all references to entry.
705 * Whenever we reach init_mm, there's no address space
706 * to search, but use it as a reminder to search shmem.
711 if (start_mm == &init_mm)
712 shmem = shmem_unuse(entry, page);
714 retval = unuse_mm(start_mm, entry, page);
717 int set_start_mm = (*swap_map >= swcount);
718 struct list_head *p = &start_mm->mmlist;
719 struct mm_struct *new_start_mm = start_mm;
720 struct mm_struct *prev_mm = start_mm;
721 struct mm_struct *mm;
723 atomic_inc(&new_start_mm->mm_users);
724 atomic_inc(&prev_mm->mm_users);
725 spin_lock(&mmlist_lock);
726 while (*swap_map > 1 && !retval &&
727 (p = p->next) != &start_mm->mmlist) {
728 mm = list_entry(p, struct mm_struct, mmlist);
729 if (atomic_inc_return(&mm->mm_users) == 1) {
730 atomic_dec(&mm->mm_users);
733 spin_unlock(&mmlist_lock);
742 else if (mm == &init_mm) {
744 shmem = shmem_unuse(entry, page);
746 retval = unuse_mm(mm, entry, page);
747 if (set_start_mm && *swap_map < swcount) {
749 atomic_inc(&mm->mm_users);
753 spin_lock(&mmlist_lock);
755 spin_unlock(&mmlist_lock);
758 start_mm = new_start_mm;
762 page_cache_release(page);
767 * How could swap count reach 0x7fff when the maximum
768 * pid is 0x7fff, and there's no way to repeat a swap
769 * page within an mm (except in shmem, where it's the
770 * shared object which takes the reference count)?
771 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
773 * If that's wrong, then we should worry more about
774 * exit_mmap() and do_munmap() cases described above:
775 * we might be resetting SWAP_MAP_MAX too early here.
776 * We know "Undead"s can happen, they're okay, so don't
777 * report them; but do report if we reset SWAP_MAP_MAX.
779 if (*swap_map == SWAP_MAP_MAX) {
780 spin_lock(&swap_lock);
782 spin_unlock(&swap_lock);
787 * If a reference remains (rare), we would like to leave
788 * the page in the swap cache; but try_to_unmap could
789 * then re-duplicate the entry once we drop page lock,
790 * so we might loop indefinitely; also, that page could
791 * not be swapped out to other storage meanwhile. So:
792 * delete from cache even if there's another reference,
793 * after ensuring that the data has been saved to disk -
794 * since if the reference remains (rarer), it will be
795 * read from disk into another page. Splitting into two
796 * pages would be incorrect if swap supported "shared
797 * private" pages, but they are handled by tmpfs files.
799 * Note shmem_unuse already deleted a swappage from
800 * the swap cache, unless the move to filepage failed:
801 * in which case it left swappage in cache, lowered its
802 * swap count to pass quickly through the loops above,
803 * and now we must reincrement count to try again later.
805 if ((*swap_map > 1) && PageDirty(page) && PageSwapCache(page)) {
806 struct writeback_control wbc = {
807 .sync_mode = WB_SYNC_NONE,
810 swap_writepage(page, &wbc);
812 wait_on_page_writeback(page);
814 if (PageSwapCache(page)) {
816 swap_duplicate(entry);
818 delete_from_swap_cache(page);
822 * So we could skip searching mms once swap count went
823 * to 1, we did not mark any present ptes as dirty: must
824 * mark page dirty so shrink_list will preserve it.
828 page_cache_release(page);
831 * Make sure that we aren't completely killing
832 * interactive performance.
838 if (reset_overflow) {
839 printk(KERN_WARNING "swapoff: cleared swap entry overflow\n");
846 * After a successful try_to_unuse, if no swap is now in use, we know
847 * we can empty the mmlist. swap_lock must be held on entry and exit.
848 * Note that mmlist_lock nests inside swap_lock, and an mm must be
849 * added to the mmlist just after page_duplicate - before would be racy.
851 static void drain_mmlist(void)
853 struct list_head *p, *next;
856 for (i = 0; i < nr_swapfiles; i++)
857 if (swap_info[i].inuse_pages)
859 spin_lock(&mmlist_lock);
860 list_for_each_safe(p, next, &init_mm.mmlist)
862 spin_unlock(&mmlist_lock);
866 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
867 * corresponds to page offset `offset'.
869 sector_t map_swap_page(struct swap_info_struct *sis, pgoff_t offset)
871 struct swap_extent *se = sis->curr_swap_extent;
872 struct swap_extent *start_se = se;
875 struct list_head *lh;
877 if (se->start_page <= offset &&
878 offset < (se->start_page + se->nr_pages)) {
879 return se->start_block + (offset - se->start_page);
882 if (lh == &sis->extent_list)
884 se = list_entry(lh, struct swap_extent, list);
885 sis->curr_swap_extent = se;
886 BUG_ON(se == start_se); /* It *must* be present */
891 * Free all of a swapdev's extent information
893 static void destroy_swap_extents(struct swap_info_struct *sis)
895 while (!list_empty(&sis->extent_list)) {
896 struct swap_extent *se;
898 se = list_entry(sis->extent_list.next,
899 struct swap_extent, list);
906 * Add a block range (and the corresponding page range) into this swapdev's
907 * extent list. The extent list is kept sorted in page order.
909 * This function rather assumes that it is called in ascending page order.
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.prev; /* The highest page extent */
920 if (lh != &sis->extent_list) {
921 se = list_entry(lh, struct swap_extent, list);
922 BUG_ON(se->start_page + se->nr_pages != start_page);
923 if (se->start_block + se->nr_pages == start_block) {
925 se->nr_pages += nr_pages;
931 * No merge. Insert a new extent, preserving ordering.
933 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
936 new_se->start_page = start_page;
937 new_se->nr_pages = nr_pages;
938 new_se->start_block = start_block;
940 list_add_tail(&new_se->list, &sis->extent_list);
945 * A `swap extent' is a simple thing which maps a contiguous range of pages
946 * onto a contiguous range of disk blocks. An ordered list of swap extents
947 * is built at swapon time and is then used at swap_writepage/swap_readpage
948 * time for locating where on disk a page belongs.
950 * If the swapfile is an S_ISBLK block device, a single extent is installed.
951 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
952 * swap files identically.
954 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
955 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
956 * swapfiles are handled *identically* after swapon time.
958 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
959 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
960 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
961 * requirements, they are simply tossed out - we will never use those blocks
964 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
965 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
966 * which will scribble on the fs.
968 * The amount of disk space which a single swap extent represents varies.
969 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
970 * extents in the list. To avoid much list walking, we cache the previous
971 * search location in `curr_swap_extent', and start new searches from there.
972 * This is extremely effective. The average number of iterations in
973 * map_swap_page() has been measured at about 0.3 per page. - akpm.
975 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
978 unsigned blocks_per_page;
979 unsigned long page_no;
981 sector_t probe_block;
983 sector_t lowest_block = -1;
984 sector_t highest_block = 0;
988 inode = sis->swap_file->f_mapping->host;
989 if (S_ISBLK(inode->i_mode)) {
990 ret = add_swap_extent(sis, 0, sis->max, 0);
995 blkbits = inode->i_blkbits;
996 blocks_per_page = PAGE_SIZE >> blkbits;
999 * Map all the blocks into the extent list. This code doesn't try
1004 last_block = i_size_read(inode) >> blkbits;
1005 while ((probe_block + blocks_per_page) <= last_block &&
1006 page_no < sis->max) {
1007 unsigned block_in_page;
1008 sector_t first_block;
1010 first_block = bmap(inode, probe_block);
1011 if (first_block == 0)
1015 * It must be PAGE_SIZE aligned on-disk
1017 if (first_block & (blocks_per_page - 1)) {
1022 for (block_in_page = 1; block_in_page < blocks_per_page;
1026 block = bmap(inode, probe_block + block_in_page);
1029 if (block != first_block + block_in_page) {
1036 first_block >>= (PAGE_SHIFT - blkbits);
1037 if (page_no) { /* exclude the header page */
1038 if (first_block < lowest_block)
1039 lowest_block = first_block;
1040 if (first_block > highest_block)
1041 highest_block = first_block;
1045 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1047 ret = add_swap_extent(sis, page_no, 1, first_block);
1052 probe_block += blocks_per_page;
1057 *span = 1 + highest_block - lowest_block;
1059 page_no = 1; /* force Empty message */
1061 sis->pages = page_no - 1;
1062 sis->highest_bit = page_no - 1;
1064 sis->curr_swap_extent = list_entry(sis->extent_list.prev,
1065 struct swap_extent, list);
1068 printk(KERN_ERR "swapon: swapfile has holes\n");
1074 #if 0 /* We don't need this yet */
1075 #include <linux/backing-dev.h>
1076 int page_queue_congested(struct page *page)
1078 struct backing_dev_info *bdi;
1080 BUG_ON(!PageLocked(page)); /* It pins the swap_info_struct */
1082 if (PageSwapCache(page)) {
1083 swp_entry_t entry = { .val = page_private(page) };
1084 struct swap_info_struct *sis;
1086 sis = get_swap_info_struct(swp_type(entry));
1087 bdi = sis->bdev->bd_inode->i_mapping->backing_dev_info;
1089 bdi = page->mapping->backing_dev_info;
1090 return bdi_write_congested(bdi);
1094 asmlinkage long sys_swapoff(const char __user * specialfile)
1096 struct swap_info_struct * p = NULL;
1097 unsigned short *swap_map;
1098 struct file *swap_file, *victim;
1099 struct address_space *mapping;
1100 struct inode *inode;
1105 if (!capable(CAP_SYS_ADMIN))
1108 pathname = getname(specialfile);
1109 err = PTR_ERR(pathname);
1110 if (IS_ERR(pathname))
1113 victim = filp_open(pathname, O_RDWR|O_LARGEFILE, 0);
1115 err = PTR_ERR(victim);
1119 mapping = victim->f_mapping;
1121 spin_lock(&swap_lock);
1122 for (type = swap_list.head; type >= 0; type = swap_info[type].next) {
1123 p = swap_info + type;
1124 if ((p->flags & SWP_ACTIVE) == SWP_ACTIVE) {
1125 if (p->swap_file->f_mapping == mapping)
1132 spin_unlock(&swap_lock);
1135 if (!security_vm_enough_memory(p->pages))
1136 vm_unacct_memory(p->pages);
1139 spin_unlock(&swap_lock);
1143 swap_list.head = p->next;
1145 swap_info[prev].next = p->next;
1147 if (type == swap_list.next) {
1148 /* just pick something that's safe... */
1149 swap_list.next = swap_list.head;
1151 nr_swap_pages -= p->pages;
1152 total_swap_pages -= p->pages;
1153 p->flags &= ~SWP_WRITEOK;
1154 spin_unlock(&swap_lock);
1156 current->flags |= PF_SWAPOFF;
1157 err = try_to_unuse(type);
1158 current->flags &= ~PF_SWAPOFF;
1161 /* re-insert swap space back into swap_list */
1162 spin_lock(&swap_lock);
1163 for (prev = -1, i = swap_list.head; i >= 0; prev = i, i = swap_info[i].next)
1164 if (p->prio >= swap_info[i].prio)
1168 swap_list.head = swap_list.next = p - swap_info;
1170 swap_info[prev].next = p - swap_info;
1171 nr_swap_pages += p->pages;
1172 total_swap_pages += p->pages;
1173 p->flags |= SWP_WRITEOK;
1174 spin_unlock(&swap_lock);
1178 /* wait for any unplug function to finish */
1179 down_write(&swap_unplug_sem);
1180 up_write(&swap_unplug_sem);
1182 destroy_swap_extents(p);
1183 mutex_lock(&swapon_mutex);
1184 spin_lock(&swap_lock);
1187 /* wait for anyone still in scan_swap_map */
1188 p->highest_bit = 0; /* cuts scans short */
1189 while (p->flags >= SWP_SCANNING) {
1190 spin_unlock(&swap_lock);
1191 schedule_timeout_uninterruptible(1);
1192 spin_lock(&swap_lock);
1195 swap_file = p->swap_file;
1196 p->swap_file = NULL;
1198 swap_map = p->swap_map;
1201 spin_unlock(&swap_lock);
1202 mutex_unlock(&swapon_mutex);
1204 inode = mapping->host;
1205 if (S_ISBLK(inode->i_mode)) {
1206 struct block_device *bdev = I_BDEV(inode);
1207 set_blocksize(bdev, p->old_block_size);
1210 mutex_lock(&inode->i_mutex);
1211 inode->i_flags &= ~S_SWAPFILE;
1212 mutex_unlock(&inode->i_mutex);
1214 filp_close(swap_file, NULL);
1218 filp_close(victim, NULL);
1223 #ifdef CONFIG_PROC_FS
1225 static void *swap_start(struct seq_file *swap, loff_t *pos)
1227 struct swap_info_struct *ptr = swap_info;
1231 mutex_lock(&swapon_mutex);
1233 for (i = 0; i < nr_swapfiles; i++, ptr++) {
1234 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1243 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
1245 struct swap_info_struct *ptr = v;
1246 struct swap_info_struct *endptr = swap_info + nr_swapfiles;
1248 for (++ptr; ptr < endptr; ptr++) {
1249 if (!(ptr->flags & SWP_USED) || !ptr->swap_map)
1258 static void swap_stop(struct seq_file *swap, void *v)
1260 mutex_unlock(&swapon_mutex);
1263 static int swap_show(struct seq_file *swap, void *v)
1265 struct swap_info_struct *ptr = v;
1270 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1272 file = ptr->swap_file;
1273 len = seq_path(swap, file->f_vfsmnt, file->f_dentry, " \t\n\\");
1274 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
1275 len < 40 ? 40 - len : 1, " ",
1276 S_ISBLK(file->f_dentry->d_inode->i_mode) ?
1277 "partition" : "file\t",
1278 ptr->pages << (PAGE_SHIFT - 10),
1279 ptr->inuse_pages << (PAGE_SHIFT - 10),
1284 static struct seq_operations swaps_op = {
1285 .start = swap_start,
1291 static int swaps_open(struct inode *inode, struct file *file)
1293 return seq_open(file, &swaps_op);
1296 static struct file_operations proc_swaps_operations = {
1299 .llseek = seq_lseek,
1300 .release = seq_release,
1303 static int __init procswaps_init(void)
1305 struct proc_dir_entry *entry;
1307 entry = create_proc_entry("swaps", 0, NULL);
1309 entry->proc_fops = &proc_swaps_operations;
1312 __initcall(procswaps_init);
1313 #endif /* CONFIG_PROC_FS */
1316 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1318 * The swapon system call
1320 asmlinkage long sys_swapon(const char __user * specialfile, int swap_flags)
1322 struct swap_info_struct * p;
1324 struct block_device *bdev = NULL;
1325 struct file *swap_file = NULL;
1326 struct address_space *mapping;
1330 static int least_priority;
1331 union swap_header *swap_header = NULL;
1332 int swap_header_version;
1333 unsigned int nr_good_pages = 0;
1336 unsigned long maxpages = 1;
1338 unsigned short *swap_map;
1339 struct page *page = NULL;
1340 struct inode *inode = NULL;
1343 if (!capable(CAP_SYS_ADMIN))
1345 spin_lock(&swap_lock);
1347 for (type = 0 ; type < nr_swapfiles ; type++,p++)
1348 if (!(p->flags & SWP_USED))
1352 * Test if adding another swap device is possible. There are
1353 * two limiting factors: 1) the number of bits for the swap
1354 * type swp_entry_t definition and 2) the number of bits for
1355 * the swap type in the swap ptes as defined by the different
1356 * architectures. To honor both limitations a swap entry
1357 * with swap offset 0 and swap type ~0UL is created, encoded
1358 * to a swap pte, decoded to a swp_entry_t again and finally
1359 * the swap type part is extracted. This will mask all bits
1360 * from the initial ~0UL that can't be encoded in either the
1361 * swp_entry_t or the architecture definition of a swap pte.
1363 if (type > swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1364 spin_unlock(&swap_lock);
1367 if (type >= nr_swapfiles)
1368 nr_swapfiles = type+1;
1369 INIT_LIST_HEAD(&p->extent_list);
1370 p->flags = SWP_USED;
1371 p->swap_file = NULL;
1372 p->old_block_size = 0;
1379 if (swap_flags & SWAP_FLAG_PREFER) {
1381 (swap_flags & SWAP_FLAG_PRIO_MASK)>>SWAP_FLAG_PRIO_SHIFT;
1383 p->prio = --least_priority;
1385 spin_unlock(&swap_lock);
1386 name = getname(specialfile);
1387 error = PTR_ERR(name);
1392 swap_file = filp_open(name, O_RDWR|O_LARGEFILE, 0);
1393 error = PTR_ERR(swap_file);
1394 if (IS_ERR(swap_file)) {
1399 p->swap_file = swap_file;
1400 mapping = swap_file->f_mapping;
1401 inode = mapping->host;
1404 for (i = 0; i < nr_swapfiles; i++) {
1405 struct swap_info_struct *q = &swap_info[i];
1407 if (i == type || !q->swap_file)
1409 if (mapping == q->swap_file->f_mapping)
1414 if (S_ISBLK(inode->i_mode)) {
1415 bdev = I_BDEV(inode);
1416 error = bd_claim(bdev, sys_swapon);
1422 p->old_block_size = block_size(bdev);
1423 error = set_blocksize(bdev, PAGE_SIZE);
1427 } else if (S_ISREG(inode->i_mode)) {
1428 p->bdev = inode->i_sb->s_bdev;
1429 mutex_lock(&inode->i_mutex);
1431 if (IS_SWAPFILE(inode)) {
1439 swapfilesize = i_size_read(inode) >> PAGE_SHIFT;
1442 * Read the swap header.
1444 if (!mapping->a_ops->readpage) {
1448 page = read_cache_page(mapping, 0,
1449 (filler_t *)mapping->a_ops->readpage, swap_file);
1451 error = PTR_ERR(page);
1454 wait_on_page_locked(page);
1455 if (!PageUptodate(page))
1458 swap_header = page_address(page);
1460 if (!memcmp("SWAP-SPACE",swap_header->magic.magic,10))
1461 swap_header_version = 1;
1462 else if (!memcmp("SWAPSPACE2",swap_header->magic.magic,10))
1463 swap_header_version = 2;
1465 printk(KERN_ERR "Unable to find swap-space signature\n");
1470 switch (swap_header_version) {
1472 printk(KERN_ERR "version 0 swap is no longer supported. "
1473 "Use mkswap -v1 %s\n", name);
1477 /* Check the swap header's sub-version and the size of
1478 the swap file and bad block lists */
1479 if (swap_header->info.version != 1) {
1481 "Unable to handle swap header version %d\n",
1482 swap_header->info.version);
1488 p->cluster_next = 1;
1491 * Find out how many pages are allowed for a single swap
1492 * device. There are two limiting factors: 1) the number of
1493 * bits for the swap offset in the swp_entry_t type and
1494 * 2) the number of bits in the a swap pte as defined by
1495 * the different architectures. In order to find the
1496 * largest possible bit mask a swap entry with swap type 0
1497 * and swap offset ~0UL is created, encoded to a swap pte,
1498 * decoded to a swp_entry_t again and finally the swap
1499 * offset is extracted. This will mask all the bits from
1500 * the initial ~0UL mask that can't be encoded in either
1501 * the swp_entry_t or the architecture definition of a
1504 maxpages = swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1505 if (maxpages > swap_header->info.last_page)
1506 maxpages = swap_header->info.last_page;
1507 p->highest_bit = maxpages - 1;
1512 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
1514 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
1517 /* OK, set up the swap map and apply the bad block list */
1518 if (!(p->swap_map = vmalloc(maxpages * sizeof(short)))) {
1524 memset(p->swap_map, 0, maxpages * sizeof(short));
1525 for (i = 0; i < swap_header->info.nr_badpages; i++) {
1526 int page_nr = swap_header->info.badpages[i];
1527 if (page_nr <= 0 || page_nr >= swap_header->info.last_page)
1530 p->swap_map[page_nr] = SWAP_MAP_BAD;
1532 nr_good_pages = swap_header->info.last_page -
1533 swap_header->info.nr_badpages -
1534 1 /* header page */;
1539 if (swapfilesize && maxpages > swapfilesize) {
1541 "Swap area shorter than signature indicates\n");
1545 if (nr_good_pages) {
1546 p->swap_map[0] = SWAP_MAP_BAD;
1548 p->pages = nr_good_pages;
1549 nr_extents = setup_swap_extents(p, &span);
1550 if (nr_extents < 0) {
1554 nr_good_pages = p->pages;
1556 if (!nr_good_pages) {
1557 printk(KERN_WARNING "Empty swap-file\n");
1562 mutex_lock(&swapon_mutex);
1563 spin_lock(&swap_lock);
1564 p->flags = SWP_ACTIVE;
1565 nr_swap_pages += nr_good_pages;
1566 total_swap_pages += nr_good_pages;
1568 printk(KERN_INFO "Adding %uk swap on %s. "
1569 "Priority:%d extents:%d across:%lluk\n",
1570 nr_good_pages<<(PAGE_SHIFT-10), name, p->prio,
1571 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10));
1573 /* insert swap space into swap_list: */
1575 for (i = swap_list.head; i >= 0; i = swap_info[i].next) {
1576 if (p->prio >= swap_info[i].prio) {
1583 swap_list.head = swap_list.next = p - swap_info;
1585 swap_info[prev].next = p - swap_info;
1587 spin_unlock(&swap_lock);
1588 mutex_unlock(&swapon_mutex);
1593 set_blocksize(bdev, p->old_block_size);
1596 destroy_swap_extents(p);
1598 spin_lock(&swap_lock);
1599 swap_map = p->swap_map;
1600 p->swap_file = NULL;
1603 if (!(swap_flags & SWAP_FLAG_PREFER))
1605 spin_unlock(&swap_lock);
1608 filp_close(swap_file, NULL);
1610 if (page && !IS_ERR(page)) {
1612 page_cache_release(page);
1618 inode->i_flags |= S_SWAPFILE;
1619 mutex_unlock(&inode->i_mutex);
1624 void si_swapinfo(struct sysinfo *val)
1627 unsigned long nr_to_be_unused = 0;
1629 spin_lock(&swap_lock);
1630 for (i = 0; i < nr_swapfiles; i++) {
1631 if (!(swap_info[i].flags & SWP_USED) ||
1632 (swap_info[i].flags & SWP_WRITEOK))
1634 nr_to_be_unused += swap_info[i].inuse_pages;
1636 val->freeswap = nr_swap_pages + nr_to_be_unused;
1637 val->totalswap = total_swap_pages + nr_to_be_unused;
1638 spin_unlock(&swap_lock);
1639 if (vx_flags(VXF_VIRT_MEM, 0))
1640 vx_vsi_swapinfo(val);
1644 * Verify that a swap entry is valid and increment its swap map count.
1646 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1647 * "permanent", but will be reclaimed by the next swapoff.
1649 int swap_duplicate(swp_entry_t entry)
1651 struct swap_info_struct * p;
1652 unsigned long offset, type;
1655 type = swp_type(entry);
1656 if (type >= nr_swapfiles)
1658 p = type + swap_info;
1659 offset = swp_offset(entry);
1661 spin_lock(&swap_lock);
1662 if (offset < p->max && p->swap_map[offset]) {
1663 if (p->swap_map[offset] < SWAP_MAP_MAX - 1) {
1664 p->swap_map[offset]++;
1666 } else if (p->swap_map[offset] <= SWAP_MAP_MAX) {
1667 if (swap_overflow++ < 5)
1668 printk(KERN_WARNING "swap_dup: swap entry overflow\n");
1669 p->swap_map[offset] = SWAP_MAP_MAX;
1673 spin_unlock(&swap_lock);
1678 printk(KERN_ERR "swap_dup: %s%08lx\n", Bad_file, entry.val);
1682 struct swap_info_struct *
1683 get_swap_info_struct(unsigned type)
1685 return &swap_info[type];
1689 * swap_lock prevents swap_map being freed. Don't grab an extra
1690 * reference on the swaphandle, it doesn't matter if it becomes unused.
1692 int valid_swaphandles(swp_entry_t entry, unsigned long *offset)
1694 int ret = 0, i = 1 << page_cluster;
1696 struct swap_info_struct *swapdev = swp_type(entry) + swap_info;
1698 if (!page_cluster) /* no readahead */
1700 toff = (swp_offset(entry) >> page_cluster) << page_cluster;
1701 if (!toff) /* first page is swap header */
1705 spin_lock(&swap_lock);
1707 /* Don't read-ahead past the end of the swap area */
1708 if (toff >= swapdev->max)
1710 /* Don't read in free or bad pages */
1711 if (!swapdev->swap_map[toff])
1713 if (swapdev->swap_map[toff] == SWAP_MAP_BAD)
1718 spin_unlock(&swap_lock);