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/syscalls.h>
30 #include <asm/pgtable.h>
31 #include <asm/tlbflush.h>
32 #include <linux/swapops.h>
34 DEFINE_SPINLOCK(swaplock
);
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 blk_run_backing_dev(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 * How many references to page are currently swapped out?
281 static inline int page_swapcount(struct page
*page
)
284 struct swap_info_struct
*p
;
287 entry
.val
= page
->private;
288 p
= swap_info_get(entry
);
290 /* Subtract the 1 for the swap cache itself */
291 count
= p
->swap_map
[swp_offset(entry
)] - 1;
298 * We can use this swap cache entry directly
299 * if there are no other references to it.
301 int can_share_swap_page(struct page
*page
)
305 BUG_ON(!PageLocked(page
));
306 count
= page_mapcount(page
);
307 if (count
<= 1 && PageSwapCache(page
))
308 count
+= page_swapcount(page
);
313 * Work out if there are any other processes sharing this
314 * swap cache page. Free it if you can. Return success.
316 int remove_exclusive_swap_page(struct page
*page
)
319 struct swap_info_struct
* p
;
322 BUG_ON(PagePrivate(page
));
323 BUG_ON(!PageLocked(page
));
325 if (!PageSwapCache(page
))
327 if (PageWriteback(page
))
329 if (page_count(page
) != 2) /* 2: us + cache */
332 entry
.val
= page
->private;
333 p
= swap_info_get(entry
);
337 /* Is the only swap cache user the cache itself? */
339 if (p
->swap_map
[swp_offset(entry
)] == 1) {
340 /* Recheck the page count with the swapcache lock held.. */
341 write_lock_irq(&swapper_space
.tree_lock
);
342 if ((page_count(page
) == 2) && !PageWriteback(page
)) {
343 __delete_from_swap_cache(page
);
347 write_unlock_irq(&swapper_space
.tree_lock
);
353 page_cache_release(page
);
360 * Free the swap entry like above, but also try to
361 * free the page cache entry if it is the last user.
363 void free_swap_and_cache(swp_entry_t entry
)
365 struct swap_info_struct
* p
;
366 struct page
*page
= NULL
;
368 p
= swap_info_get(entry
);
370 if (swap_entry_free(p
, swp_offset(entry
)) == 1)
371 page
= find_trylock_page(&swapper_space
, entry
.val
);
377 BUG_ON(PagePrivate(page
));
378 page_cache_get(page
);
379 one_user
= (page_count(page
) == 2);
380 /* Only cache user (+us), or swap space full? Free it! */
381 if (!PageWriteback(page
) && (one_user
|| vm_swap_full())) {
382 delete_from_swap_cache(page
);
386 page_cache_release(page
);
391 * Always set the resulting pte to be nowrite (the same as COW pages
392 * after one process has exited). We don't know just how many PTEs will
393 * share this swap entry, so be cautious and let do_wp_page work out
394 * what to do if a write is requested later.
396 * vma->vm_mm->page_table_lock is held.
398 static void unuse_pte(struct vm_area_struct
*vma
, pte_t
*pte
,
399 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
401 inc_mm_counter(vma
->vm_mm
, rss
);
403 set_pte_at(vma
->vm_mm
, addr
, pte
,
404 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
405 page_add_anon_rmap(page
, vma
, addr
);
408 * Move the page to the active list so it is not
409 * immediately swapped out again after swapon.
414 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
415 unsigned long addr
, unsigned long end
,
416 swp_entry_t entry
, struct page
*page
)
419 pte_t swp_pte
= swp_entry_to_pte(entry
);
421 pte
= pte_offset_map(pmd
, addr
);
424 * swapoff spends a _lot_ of time in this loop!
425 * Test inline before going to call unuse_pte.
427 if (unlikely(pte_same(*pte
, swp_pte
))) {
428 unuse_pte(vma
, pte
, addr
, entry
, page
);
432 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
437 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
438 unsigned long addr
, unsigned long end
,
439 swp_entry_t entry
, struct page
*page
)
444 pmd
= pmd_offset(pud
, addr
);
446 next
= pmd_addr_end(addr
, end
);
447 if (pmd_none_or_clear_bad(pmd
))
449 if (unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
))
451 } while (pmd
++, addr
= next
, addr
!= end
);
455 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
456 unsigned long addr
, unsigned long end
,
457 swp_entry_t entry
, struct page
*page
)
462 pud
= pud_offset(pgd
, addr
);
464 next
= pud_addr_end(addr
, end
);
465 if (pud_none_or_clear_bad(pud
))
467 if (unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
))
469 } while (pud
++, addr
= next
, addr
!= end
);
473 static int unuse_vma(struct vm_area_struct
*vma
,
474 swp_entry_t entry
, struct page
*page
)
477 unsigned long addr
, end
, next
;
480 addr
= page_address_in_vma(page
, vma
);
484 end
= addr
+ PAGE_SIZE
;
486 addr
= vma
->vm_start
;
490 pgd
= pgd_offset(vma
->vm_mm
, addr
);
492 next
= pgd_addr_end(addr
, end
);
493 if (pgd_none_or_clear_bad(pgd
))
495 if (unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
))
497 } while (pgd
++, addr
= next
, addr
!= end
);
501 static int unuse_mm(struct mm_struct
*mm
,
502 swp_entry_t entry
, struct page
*page
)
504 struct vm_area_struct
*vma
;
506 if (!down_read_trylock(&mm
->mmap_sem
)) {
508 * Activate page so shrink_cache is unlikely to unmap its
509 * ptes while lock is dropped, so swapoff can make progress.
513 down_read(&mm
->mmap_sem
);
516 spin_lock(&mm
->page_table_lock
);
517 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
518 if (vma
->anon_vma
&& unuse_vma(vma
, entry
, page
))
521 spin_unlock(&mm
->page_table_lock
);
522 up_read(&mm
->mmap_sem
);
524 * Currently unuse_mm cannot fail, but leave error handling
525 * at call sites for now, since we change it from time to time.
531 * Scan swap_map from current position to next entry still in use.
532 * Recycle to start on reaching the end, returning 0 when empty.
534 static int find_next_to_unuse(struct swap_info_struct
*si
, int prev
)
541 * No need for swap_device_lock(si) here: we're just looking
542 * for whether an entry is in use, not modifying it; false
543 * hits are okay, and sys_swapoff() has already prevented new
544 * allocations from this area (while holding swap_list_lock()).
553 * No entries in use at top of swap_map,
554 * loop back to start and recheck there.
560 count
= si
->swap_map
[i
];
561 if (count
&& count
!= SWAP_MAP_BAD
)
568 * We completely avoid races by reading each swap page in advance,
569 * and then search for the process using it. All the necessary
570 * page table adjustments can then be made atomically.
572 static int try_to_unuse(unsigned int type
)
574 struct swap_info_struct
* si
= &swap_info
[type
];
575 struct mm_struct
*start_mm
;
576 unsigned short *swap_map
;
577 unsigned short swcount
;
582 int reset_overflow
= 0;
586 * When searching mms for an entry, a good strategy is to
587 * start at the first mm we freed the previous entry from
588 * (though actually we don't notice whether we or coincidence
589 * freed the entry). Initialize this start_mm with a hold.
591 * A simpler strategy would be to start at the last mm we
592 * freed the previous entry from; but that would take less
593 * advantage of mmlist ordering, which clusters forked mms
594 * together, child after parent. If we race with dup_mmap(), we
595 * prefer to resolve parent before child, lest we miss entries
596 * duplicated after we scanned child: using last mm would invert
597 * that. Though it's only a serious concern when an overflowed
598 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
601 atomic_inc(&init_mm
.mm_users
);
604 * Keep on scanning until all entries have gone. Usually,
605 * one pass through swap_map is enough, but not necessarily:
606 * there are races when an instance of an entry might be missed.
608 while ((i
= find_next_to_unuse(si
, i
)) != 0) {
609 if (signal_pending(current
)) {
615 * Get a page for the entry, using the existing swap
616 * cache page if there is one. Otherwise, get a clean
617 * page and read the swap into it.
619 swap_map
= &si
->swap_map
[i
];
620 entry
= swp_entry(type
, i
);
621 page
= read_swap_cache_async(entry
, NULL
, 0);
624 * Either swap_duplicate() failed because entry
625 * has been freed independently, and will not be
626 * reused since sys_swapoff() already disabled
627 * allocation from here, or alloc_page() failed.
636 * Don't hold on to start_mm if it looks like exiting.
638 if (atomic_read(&start_mm
->mm_users
) == 1) {
641 atomic_inc(&init_mm
.mm_users
);
645 * Wait for and lock page. When do_swap_page races with
646 * try_to_unuse, do_swap_page can handle the fault much
647 * faster than try_to_unuse can locate the entry. This
648 * apparently redundant "wait_on_page_locked" lets try_to_unuse
649 * defer to do_swap_page in such a case - in some tests,
650 * do_swap_page and try_to_unuse repeatedly compete.
652 wait_on_page_locked(page
);
653 wait_on_page_writeback(page
);
655 wait_on_page_writeback(page
);
658 * Remove all references to entry.
659 * Whenever we reach init_mm, there's no address space
660 * to search, but use it as a reminder to search shmem.
665 if (start_mm
== &init_mm
)
666 shmem
= shmem_unuse(entry
, page
);
668 retval
= unuse_mm(start_mm
, entry
, page
);
671 int set_start_mm
= (*swap_map
>= swcount
);
672 struct list_head
*p
= &start_mm
->mmlist
;
673 struct mm_struct
*new_start_mm
= start_mm
;
674 struct mm_struct
*prev_mm
= start_mm
;
675 struct mm_struct
*mm
;
677 atomic_inc(&new_start_mm
->mm_users
);
678 atomic_inc(&prev_mm
->mm_users
);
679 spin_lock(&mmlist_lock
);
680 while (*swap_map
> 1 && !retval
&&
681 (p
= p
->next
) != &start_mm
->mmlist
) {
682 mm
= list_entry(p
, struct mm_struct
, mmlist
);
683 if (atomic_inc_return(&mm
->mm_users
) == 1) {
684 atomic_dec(&mm
->mm_users
);
687 spin_unlock(&mmlist_lock
);
696 else if (mm
== &init_mm
) {
698 shmem
= shmem_unuse(entry
, page
);
700 retval
= unuse_mm(mm
, entry
, page
);
701 if (set_start_mm
&& *swap_map
< swcount
) {
703 atomic_inc(&mm
->mm_users
);
707 spin_lock(&mmlist_lock
);
709 spin_unlock(&mmlist_lock
);
712 start_mm
= new_start_mm
;
716 page_cache_release(page
);
721 * How could swap count reach 0x7fff when the maximum
722 * pid is 0x7fff, and there's no way to repeat a swap
723 * page within an mm (except in shmem, where it's the
724 * shared object which takes the reference count)?
725 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
727 * If that's wrong, then we should worry more about
728 * exit_mmap() and do_munmap() cases described above:
729 * we might be resetting SWAP_MAP_MAX too early here.
730 * We know "Undead"s can happen, they're okay, so don't
731 * report them; but do report if we reset SWAP_MAP_MAX.
733 if (*swap_map
== SWAP_MAP_MAX
) {
734 swap_device_lock(si
);
736 swap_device_unlock(si
);
741 * If a reference remains (rare), we would like to leave
742 * the page in the swap cache; but try_to_unmap could
743 * then re-duplicate the entry once we drop page lock,
744 * so we might loop indefinitely; also, that page could
745 * not be swapped out to other storage meanwhile. So:
746 * delete from cache even if there's another reference,
747 * after ensuring that the data has been saved to disk -
748 * since if the reference remains (rarer), it will be
749 * read from disk into another page. Splitting into two
750 * pages would be incorrect if swap supported "shared
751 * private" pages, but they are handled by tmpfs files.
753 * Note shmem_unuse already deleted a swappage from
754 * the swap cache, unless the move to filepage failed:
755 * in which case it left swappage in cache, lowered its
756 * swap count to pass quickly through the loops above,
757 * and now we must reincrement count to try again later.
759 if ((*swap_map
> 1) && PageDirty(page
) && PageSwapCache(page
)) {
760 struct writeback_control wbc
= {
761 .sync_mode
= WB_SYNC_NONE
,
764 swap_writepage(page
, &wbc
);
766 wait_on_page_writeback(page
);
768 if (PageSwapCache(page
)) {
770 swap_duplicate(entry
);
772 delete_from_swap_cache(page
);
776 * So we could skip searching mms once swap count went
777 * to 1, we did not mark any present ptes as dirty: must
778 * mark page dirty so shrink_list will preserve it.
782 page_cache_release(page
);
785 * Make sure that we aren't completely killing
786 * interactive performance.
792 if (reset_overflow
) {
793 printk(KERN_WARNING
"swapoff: cleared swap entry overflow\n");
800 * After a successful try_to_unuse, if no swap is now in use, we know we
801 * can empty the mmlist. swap_list_lock must be held on entry and exit.
802 * Note that mmlist_lock nests inside swap_list_lock, and an mm must be
803 * added to the mmlist just after page_duplicate - before would be racy.
805 static void drain_mmlist(void)
807 struct list_head
*p
, *next
;
810 for (i
= 0; i
< nr_swapfiles
; i
++)
811 if (swap_info
[i
].inuse_pages
)
813 spin_lock(&mmlist_lock
);
814 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
816 spin_unlock(&mmlist_lock
);
820 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
821 * corresponds to page offset `offset'.
823 sector_t
map_swap_page(struct swap_info_struct
*sis
, pgoff_t offset
)
825 struct swap_extent
*se
= sis
->curr_swap_extent
;
826 struct swap_extent
*start_se
= se
;
829 struct list_head
*lh
;
831 if (se
->start_page
<= offset
&&
832 offset
< (se
->start_page
+ se
->nr_pages
)) {
833 return se
->start_block
+ (offset
- se
->start_page
);
836 if (lh
== &sis
->extent_list
)
838 se
= list_entry(lh
, struct swap_extent
, list
);
839 sis
->curr_swap_extent
= se
;
840 BUG_ON(se
== start_se
); /* It *must* be present */
845 * Free all of a swapdev's extent information
847 static void destroy_swap_extents(struct swap_info_struct
*sis
)
849 while (!list_empty(&sis
->extent_list
)) {
850 struct swap_extent
*se
;
852 se
= list_entry(sis
->extent_list
.next
,
853 struct swap_extent
, list
);
860 * Add a block range (and the corresponding page range) into this swapdev's
861 * extent list. The extent list is kept sorted in page order.
863 * This function rather assumes that it is called in ascending page order.
866 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
867 unsigned long nr_pages
, sector_t start_block
)
869 struct swap_extent
*se
;
870 struct swap_extent
*new_se
;
871 struct list_head
*lh
;
873 lh
= sis
->extent_list
.prev
; /* The highest page extent */
874 if (lh
!= &sis
->extent_list
) {
875 se
= list_entry(lh
, struct swap_extent
, list
);
876 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
877 if (se
->start_block
+ se
->nr_pages
== start_block
) {
879 se
->nr_pages
+= nr_pages
;
885 * No merge. Insert a new extent, preserving ordering.
887 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
890 new_se
->start_page
= start_page
;
891 new_se
->nr_pages
= nr_pages
;
892 new_se
->start_block
= start_block
;
894 list_add_tail(&new_se
->list
, &sis
->extent_list
);
899 * A `swap extent' is a simple thing which maps a contiguous range of pages
900 * onto a contiguous range of disk blocks. An ordered list of swap extents
901 * is built at swapon time and is then used at swap_writepage/swap_readpage
902 * time for locating where on disk a page belongs.
904 * If the swapfile is an S_ISBLK block device, a single extent is installed.
905 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
906 * swap files identically.
908 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
909 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
910 * swapfiles are handled *identically* after swapon time.
912 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
913 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
914 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
915 * requirements, they are simply tossed out - we will never use those blocks
918 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
919 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
920 * which will scribble on the fs.
922 * The amount of disk space which a single swap extent represents varies.
923 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
924 * extents in the list. To avoid much list walking, we cache the previous
925 * search location in `curr_swap_extent', and start new searches from there.
926 * This is extremely effective. The average number of iterations in
927 * map_swap_page() has been measured at about 0.3 per page. - akpm.
929 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
932 unsigned blocks_per_page
;
933 unsigned long page_no
;
935 sector_t probe_block
;
937 sector_t lowest_block
= -1;
938 sector_t highest_block
= 0;
942 inode
= sis
->swap_file
->f_mapping
->host
;
943 if (S_ISBLK(inode
->i_mode
)) {
944 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
949 blkbits
= inode
->i_blkbits
;
950 blocks_per_page
= PAGE_SIZE
>> blkbits
;
953 * Map all the blocks into the extent list. This code doesn't try
958 last_block
= i_size_read(inode
) >> blkbits
;
959 while ((probe_block
+ blocks_per_page
) <= last_block
&&
960 page_no
< sis
->max
) {
961 unsigned block_in_page
;
962 sector_t first_block
;
964 first_block
= bmap(inode
, probe_block
);
965 if (first_block
== 0)
969 * It must be PAGE_SIZE aligned on-disk
971 if (first_block
& (blocks_per_page
- 1)) {
976 for (block_in_page
= 1; block_in_page
< blocks_per_page
;
980 block
= bmap(inode
, probe_block
+ block_in_page
);
983 if (block
!= first_block
+ block_in_page
) {
990 first_block
>>= (PAGE_SHIFT
- blkbits
);
991 if (page_no
) { /* exclude the header page */
992 if (first_block
< lowest_block
)
993 lowest_block
= first_block
;
994 if (first_block
> highest_block
)
995 highest_block
= first_block
;
999 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1001 ret
= add_swap_extent(sis
, page_no
, 1, first_block
);
1006 probe_block
+= blocks_per_page
;
1011 *span
= 1 + highest_block
- lowest_block
;
1013 page_no
= 1; /* force Empty message */
1015 sis
->pages
= page_no
- 1;
1016 sis
->highest_bit
= page_no
- 1;
1018 sis
->curr_swap_extent
= list_entry(sis
->extent_list
.prev
,
1019 struct swap_extent
, list
);
1022 printk(KERN_ERR
"swapon: swapfile has holes\n");
1028 #if 0 /* We don't need this yet */
1029 #include <linux/backing-dev.h>
1030 int page_queue_congested(struct page
*page
)
1032 struct backing_dev_info
*bdi
;
1034 BUG_ON(!PageLocked(page
)); /* It pins the swap_info_struct */
1036 if (PageSwapCache(page
)) {
1037 swp_entry_t entry
= { .val
= page
->private };
1038 struct swap_info_struct
*sis
;
1040 sis
= get_swap_info_struct(swp_type(entry
));
1041 bdi
= sis
->bdev
->bd_inode
->i_mapping
->backing_dev_info
;
1043 bdi
= page
->mapping
->backing_dev_info
;
1044 return bdi_write_congested(bdi
);
1048 asmlinkage
long sys_swapoff(const char __user
* specialfile
)
1050 struct swap_info_struct
* p
= NULL
;
1051 unsigned short *swap_map
;
1052 struct file
*swap_file
, *victim
;
1053 struct address_space
*mapping
;
1054 struct inode
*inode
;
1059 if (!capable(CAP_SYS_ADMIN
))
1062 pathname
= getname(specialfile
);
1063 err
= PTR_ERR(pathname
);
1064 if (IS_ERR(pathname
))
1067 victim
= filp_open(pathname
, O_RDWR
|O_LARGEFILE
, 0);
1069 err
= PTR_ERR(victim
);
1073 mapping
= victim
->f_mapping
;
1076 for (type
= swap_list
.head
; type
>= 0; type
= swap_info
[type
].next
) {
1077 p
= swap_info
+ type
;
1078 if ((p
->flags
& SWP_ACTIVE
) == SWP_ACTIVE
) {
1079 if (p
->swap_file
->f_mapping
== mapping
)
1089 if (!security_vm_enough_memory(p
->pages
))
1090 vm_unacct_memory(p
->pages
);
1097 swap_list
.head
= p
->next
;
1099 swap_info
[prev
].next
= p
->next
;
1101 if (type
== swap_list
.next
) {
1102 /* just pick something that's safe... */
1103 swap_list
.next
= swap_list
.head
;
1105 nr_swap_pages
-= p
->pages
;
1106 total_swap_pages
-= p
->pages
;
1107 p
->flags
&= ~SWP_WRITEOK
;
1109 current
->flags
|= PF_SWAPOFF
;
1110 err
= try_to_unuse(type
);
1111 current
->flags
&= ~PF_SWAPOFF
;
1113 /* wait for any unplug function to finish */
1114 down_write(&swap_unplug_sem
);
1115 up_write(&swap_unplug_sem
);
1118 /* re-insert swap space back into swap_list */
1120 for (prev
= -1, i
= swap_list
.head
; i
>= 0; prev
= i
, i
= swap_info
[i
].next
)
1121 if (p
->prio
>= swap_info
[i
].prio
)
1125 swap_list
.head
= swap_list
.next
= p
- swap_info
;
1127 swap_info
[prev
].next
= p
- swap_info
;
1128 nr_swap_pages
+= p
->pages
;
1129 total_swap_pages
+= p
->pages
;
1130 p
->flags
|= SWP_WRITEOK
;
1134 destroy_swap_extents(p
);
1138 swap_device_lock(p
);
1139 swap_file
= p
->swap_file
;
1140 p
->swap_file
= NULL
;
1142 swap_map
= p
->swap_map
;
1145 swap_device_unlock(p
);
1149 inode
= mapping
->host
;
1150 if (S_ISBLK(inode
->i_mode
)) {
1151 struct block_device
*bdev
= I_BDEV(inode
);
1152 set_blocksize(bdev
, p
->old_block_size
);
1155 down(&inode
->i_sem
);
1156 inode
->i_flags
&= ~S_SWAPFILE
;
1159 filp_close(swap_file
, NULL
);
1163 filp_close(victim
, NULL
);
1168 #ifdef CONFIG_PROC_FS
1170 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
1172 struct swap_info_struct
*ptr
= swap_info
;
1178 for (i
= 0; i
< nr_swapfiles
; i
++, ptr
++) {
1179 if (!(ptr
->flags
& SWP_USED
) || !ptr
->swap_map
)
1188 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
1190 struct swap_info_struct
*ptr
= v
;
1191 struct swap_info_struct
*endptr
= swap_info
+ nr_swapfiles
;
1193 for (++ptr
; ptr
< endptr
; ptr
++) {
1194 if (!(ptr
->flags
& SWP_USED
) || !ptr
->swap_map
)
1203 static void swap_stop(struct seq_file
*swap
, void *v
)
1208 static int swap_show(struct seq_file
*swap
, void *v
)
1210 struct swap_info_struct
*ptr
= v
;
1215 seq_puts(swap
, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1217 file
= ptr
->swap_file
;
1218 len
= seq_path(swap
, file
->f_vfsmnt
, file
->f_dentry
, " \t\n\\");
1219 seq_printf(swap
, "%*s%s\t%d\t%ld\t%d\n",
1220 len
< 40 ? 40 - len
: 1, " ",
1221 S_ISBLK(file
->f_dentry
->d_inode
->i_mode
) ?
1222 "partition" : "file\t",
1223 ptr
->pages
<< (PAGE_SHIFT
- 10),
1224 ptr
->inuse_pages
<< (PAGE_SHIFT
- 10),
1229 static struct seq_operations swaps_op
= {
1230 .start
= swap_start
,
1236 static int swaps_open(struct inode
*inode
, struct file
*file
)
1238 return seq_open(file
, &swaps_op
);
1241 static struct file_operations proc_swaps_operations
= {
1244 .llseek
= seq_lseek
,
1245 .release
= seq_release
,
1248 static int __init
procswaps_init(void)
1250 struct proc_dir_entry
*entry
;
1252 entry
= create_proc_entry("swaps", 0, NULL
);
1254 entry
->proc_fops
= &proc_swaps_operations
;
1257 __initcall(procswaps_init
);
1258 #endif /* CONFIG_PROC_FS */
1261 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1263 * The swapon system call
1265 asmlinkage
long sys_swapon(const char __user
* specialfile
, int swap_flags
)
1267 struct swap_info_struct
* p
;
1269 struct block_device
*bdev
= NULL
;
1270 struct file
*swap_file
= NULL
;
1271 struct address_space
*mapping
;
1275 static int least_priority
;
1276 union swap_header
*swap_header
= NULL
;
1277 int swap_header_version
;
1278 int nr_good_pages
= 0;
1281 unsigned long maxpages
= 1;
1283 unsigned short *swap_map
;
1284 struct page
*page
= NULL
;
1285 struct inode
*inode
= NULL
;
1288 if (!capable(CAP_SYS_ADMIN
))
1292 for (type
= 0 ; type
< nr_swapfiles
; type
++,p
++)
1293 if (!(p
->flags
& SWP_USED
))
1297 * Test if adding another swap device is possible. There are
1298 * two limiting factors: 1) the number of bits for the swap
1299 * type swp_entry_t definition and 2) the number of bits for
1300 * the swap type in the swap ptes as defined by the different
1301 * architectures. To honor both limitations a swap entry
1302 * with swap offset 0 and swap type ~0UL is created, encoded
1303 * to a swap pte, decoded to a swp_entry_t again and finally
1304 * the swap type part is extracted. This will mask all bits
1305 * from the initial ~0UL that can't be encoded in either the
1306 * swp_entry_t or the architecture definition of a swap pte.
1308 if (type
> swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1312 if (type
>= nr_swapfiles
)
1313 nr_swapfiles
= type
+1;
1314 INIT_LIST_HEAD(&p
->extent_list
);
1315 p
->flags
= SWP_USED
;
1316 p
->swap_file
= NULL
;
1317 p
->old_block_size
= 0;
1323 spin_lock_init(&p
->sdev_lock
);
1325 if (swap_flags
& SWAP_FLAG_PREFER
) {
1327 (swap_flags
& SWAP_FLAG_PRIO_MASK
)>>SWAP_FLAG_PRIO_SHIFT
;
1329 p
->prio
= --least_priority
;
1332 name
= getname(specialfile
);
1333 error
= PTR_ERR(name
);
1338 swap_file
= filp_open(name
, O_RDWR
|O_LARGEFILE
, 0);
1339 error
= PTR_ERR(swap_file
);
1340 if (IS_ERR(swap_file
)) {
1345 p
->swap_file
= swap_file
;
1346 mapping
= swap_file
->f_mapping
;
1347 inode
= mapping
->host
;
1350 for (i
= 0; i
< nr_swapfiles
; i
++) {
1351 struct swap_info_struct
*q
= &swap_info
[i
];
1353 if (i
== type
|| !q
->swap_file
)
1355 if (mapping
== q
->swap_file
->f_mapping
)
1360 if (S_ISBLK(inode
->i_mode
)) {
1361 bdev
= I_BDEV(inode
);
1362 error
= bd_claim(bdev
, sys_swapon
);
1367 p
->old_block_size
= block_size(bdev
);
1368 error
= set_blocksize(bdev
, PAGE_SIZE
);
1372 } else if (S_ISREG(inode
->i_mode
)) {
1373 p
->bdev
= inode
->i_sb
->s_bdev
;
1374 down(&inode
->i_sem
);
1376 if (IS_SWAPFILE(inode
)) {
1384 swapfilesize
= i_size_read(inode
) >> PAGE_SHIFT
;
1387 * Read the swap header.
1389 if (!mapping
->a_ops
->readpage
) {
1393 page
= read_cache_page(mapping
, 0,
1394 (filler_t
*)mapping
->a_ops
->readpage
, swap_file
);
1396 error
= PTR_ERR(page
);
1399 wait_on_page_locked(page
);
1400 if (!PageUptodate(page
))
1403 swap_header
= page_address(page
);
1405 if (!memcmp("SWAP-SPACE",swap_header
->magic
.magic
,10))
1406 swap_header_version
= 1;
1407 else if (!memcmp("SWAPSPACE2",swap_header
->magic
.magic
,10))
1408 swap_header_version
= 2;
1410 printk("Unable to find swap-space signature\n");
1415 switch (swap_header_version
) {
1417 printk(KERN_ERR
"version 0 swap is no longer supported. "
1418 "Use mkswap -v1 %s\n", name
);
1422 /* Check the swap header's sub-version and the size of
1423 the swap file and bad block lists */
1424 if (swap_header
->info
.version
!= 1) {
1426 "Unable to handle swap header version %d\n",
1427 swap_header
->info
.version
);
1434 * Find out how many pages are allowed for a single swap
1435 * device. There are two limiting factors: 1) the number of
1436 * bits for the swap offset in the swp_entry_t type and
1437 * 2) the number of bits in the a swap pte as defined by
1438 * the different architectures. In order to find the
1439 * largest possible bit mask a swap entry with swap type 0
1440 * and swap offset ~0UL is created, encoded to a swap pte,
1441 * decoded to a swp_entry_t again and finally the swap
1442 * offset is extracted. This will mask all the bits from
1443 * the initial ~0UL mask that can't be encoded in either
1444 * the swp_entry_t or the architecture definition of a
1447 maxpages
= swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1448 if (maxpages
> swap_header
->info
.last_page
)
1449 maxpages
= swap_header
->info
.last_page
;
1450 p
->highest_bit
= maxpages
- 1;
1455 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
1457 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
1460 /* OK, set up the swap map and apply the bad block list */
1461 if (!(p
->swap_map
= vmalloc(maxpages
* sizeof(short)))) {
1467 memset(p
->swap_map
, 0, maxpages
* sizeof(short));
1468 for (i
=0; i
<swap_header
->info
.nr_badpages
; i
++) {
1469 int page
= swap_header
->info
.badpages
[i
];
1470 if (page
<= 0 || page
>= swap_header
->info
.last_page
)
1473 p
->swap_map
[page
] = SWAP_MAP_BAD
;
1475 nr_good_pages
= swap_header
->info
.last_page
-
1476 swap_header
->info
.nr_badpages
-
1477 1 /* header page */;
1482 if (swapfilesize
&& maxpages
> swapfilesize
) {
1484 "Swap area shorter than signature indicates\n");
1488 if (nr_good_pages
) {
1489 p
->swap_map
[0] = SWAP_MAP_BAD
;
1491 p
->pages
= nr_good_pages
;
1492 nr_extents
= setup_swap_extents(p
, &span
);
1493 if (nr_extents
< 0) {
1497 nr_good_pages
= p
->pages
;
1499 if (!nr_good_pages
) {
1500 printk(KERN_WARNING
"Empty swap-file\n");
1507 swap_device_lock(p
);
1508 p
->flags
= SWP_ACTIVE
;
1509 nr_swap_pages
+= nr_good_pages
;
1510 total_swap_pages
+= nr_good_pages
;
1512 printk(KERN_INFO
"Adding %dk swap on %s. "
1513 "Priority:%d extents:%d across:%lluk\n",
1514 nr_good_pages
<<(PAGE_SHIFT
-10), name
, p
->prio
,
1515 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10));
1517 /* insert swap space into swap_list: */
1519 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
].next
) {
1520 if (p
->prio
>= swap_info
[i
].prio
) {
1527 swap_list
.head
= swap_list
.next
= p
- swap_info
;
1529 swap_info
[prev
].next
= p
- swap_info
;
1531 swap_device_unlock(p
);
1538 set_blocksize(bdev
, p
->old_block_size
);
1541 destroy_swap_extents(p
);
1544 swap_map
= p
->swap_map
;
1545 p
->swap_file
= NULL
;
1548 if (!(swap_flags
& SWAP_FLAG_PREFER
))
1553 filp_close(swap_file
, NULL
);
1555 if (page
&& !IS_ERR(page
)) {
1557 page_cache_release(page
);
1563 inode
->i_flags
|= S_SWAPFILE
;
1569 void si_swapinfo(struct sysinfo
*val
)
1572 unsigned long nr_to_be_unused
= 0;
1575 for (i
= 0; i
< nr_swapfiles
; i
++) {
1576 if (!(swap_info
[i
].flags
& SWP_USED
) ||
1577 (swap_info
[i
].flags
& SWP_WRITEOK
))
1579 nr_to_be_unused
+= swap_info
[i
].inuse_pages
;
1581 val
->freeswap
= nr_swap_pages
+ nr_to_be_unused
;
1582 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
1587 * Verify that a swap entry is valid and increment its swap map count.
1589 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1590 * "permanent", but will be reclaimed by the next swapoff.
1592 int swap_duplicate(swp_entry_t entry
)
1594 struct swap_info_struct
* p
;
1595 unsigned long offset
, type
;
1598 type
= swp_type(entry
);
1599 if (type
>= nr_swapfiles
)
1601 p
= type
+ swap_info
;
1602 offset
= swp_offset(entry
);
1604 swap_device_lock(p
);
1605 if (offset
< p
->max
&& p
->swap_map
[offset
]) {
1606 if (p
->swap_map
[offset
] < SWAP_MAP_MAX
- 1) {
1607 p
->swap_map
[offset
]++;
1609 } else if (p
->swap_map
[offset
] <= SWAP_MAP_MAX
) {
1610 if (swap_overflow
++ < 5)
1611 printk(KERN_WARNING
"swap_dup: swap entry overflow\n");
1612 p
->swap_map
[offset
] = SWAP_MAP_MAX
;
1616 swap_device_unlock(p
);
1621 printk(KERN_ERR
"swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
1625 struct swap_info_struct
*
1626 get_swap_info_struct(unsigned type
)
1628 return &swap_info
[type
];
1632 * swap_device_lock prevents swap_map being freed. Don't grab an extra
1633 * reference on the swaphandle, it doesn't matter if it becomes unused.
1635 int valid_swaphandles(swp_entry_t entry
, unsigned long *offset
)
1637 int ret
= 0, i
= 1 << page_cluster
;
1639 struct swap_info_struct
*swapdev
= swp_type(entry
) + swap_info
;
1641 if (!page_cluster
) /* no readahead */
1643 toff
= (swp_offset(entry
) >> page_cluster
) << page_cluster
;
1644 if (!toff
) /* first page is swap header */
1648 swap_device_lock(swapdev
);
1650 /* Don't read-ahead past the end of the swap area */
1651 if (toff
>= swapdev
->max
)
1653 /* Don't read in free or bad pages */
1654 if (!swapdev
->swap_map
[toff
])
1656 if (swapdev
->swap_map
[toff
] == SWAP_MAP_BAD
)
1661 swap_device_unlock(swapdev
);