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>
32 #include <asm/pgtable.h>
33 #include <asm/tlbflush.h>
34 #include <linux/swapops.h>
36 DEFINE_SPINLOCK(swap_lock
);
37 unsigned int nr_swapfiles
;
38 long total_swap_pages
;
39 static int swap_overflow
;
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 DEFINE_MUTEX(swapon_mutex
);
53 * We need this because the bdev->unplug_fn can sleep and we cannot
54 * hold swap_lock while calling the unplug_fn. And swap_lock
55 * cannot be turned into a mutex.
57 static DECLARE_RWSEM(swap_unplug_sem
);
59 void swap_unplug_io_fn(struct backing_dev_info
*unused_bdi
, struct page
*page
)
63 down_read(&swap_unplug_sem
);
64 entry
.val
= page_private(page
);
65 if (PageSwapCache(page
)) {
66 struct block_device
*bdev
= swap_info
[swp_type(entry
)].bdev
;
67 struct backing_dev_info
*bdi
;
70 * If the page is removed from swapcache from under us (with a
71 * racy try_to_unuse/swapoff) we need an additional reference
72 * count to avoid reading garbage from page_private(page) above.
73 * If the WARN_ON triggers during a swapoff it maybe the race
74 * condition and it's harmless. However if it triggers without
75 * swapoff it signals a problem.
77 WARN_ON(page_count(page
) <= 1);
79 bdi
= bdev
->bd_inode
->i_mapping
->backing_dev_info
;
80 blk_run_backing_dev(bdi
, page
);
82 up_read(&swap_unplug_sem
);
85 #define SWAPFILE_CLUSTER 256
86 #define LATENCY_LIMIT 256
88 static inline unsigned long scan_swap_map(struct swap_info_struct
*si
)
90 unsigned long offset
, last_in_cluster
;
91 int latency_ration
= LATENCY_LIMIT
;
94 * We try to cluster swap pages by allocating them sequentially
95 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
96 * way, however, we resort to first-free allocation, starting
97 * a new cluster. This prevents us from scattering swap pages
98 * all over the entire swap partition, so that we reduce
99 * overall disk seek times between swap pages. -- sct
100 * But we do now try to find an empty cluster. -Andrea
103 si
->flags
+= SWP_SCANNING
;
104 if (unlikely(!si
->cluster_nr
)) {
105 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
106 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
)
108 spin_unlock(&swap_lock
);
110 offset
= si
->lowest_bit
;
111 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
113 /* Locate the first empty (unaligned) cluster */
114 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
115 if (si
->swap_map
[offset
])
116 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
117 else if (offset
== last_in_cluster
) {
118 spin_lock(&swap_lock
);
119 si
->cluster_next
= offset
-SWAPFILE_CLUSTER
-1;
122 if (unlikely(--latency_ration
< 0)) {
124 latency_ration
= LATENCY_LIMIT
;
127 spin_lock(&swap_lock
);
133 offset
= si
->cluster_next
;
134 if (offset
> si
->highest_bit
)
135 lowest
: offset
= si
->lowest_bit
;
136 checks
: if (!(si
->flags
& SWP_WRITEOK
))
138 if (!si
->highest_bit
)
140 if (!si
->swap_map
[offset
]) {
141 if (offset
== si
->lowest_bit
)
143 if (offset
== si
->highest_bit
)
146 if (si
->inuse_pages
== si
->pages
) {
147 si
->lowest_bit
= si
->max
;
150 si
->swap_map
[offset
] = 1;
151 si
->cluster_next
= offset
+ 1;
152 si
->flags
-= SWP_SCANNING
;
156 spin_unlock(&swap_lock
);
157 while (++offset
<= si
->highest_bit
) {
158 if (!si
->swap_map
[offset
]) {
159 spin_lock(&swap_lock
);
162 if (unlikely(--latency_ration
< 0)) {
164 latency_ration
= LATENCY_LIMIT
;
167 spin_lock(&swap_lock
);
171 si
->flags
-= SWP_SCANNING
;
175 swp_entry_t
get_swap_page(void)
177 struct swap_info_struct
*si
;
182 spin_lock(&swap_lock
);
183 if (nr_swap_pages
<= 0)
187 for (type
= swap_list
.next
; type
>= 0 && wrapped
< 2; type
= next
) {
188 si
= swap_info
+ type
;
191 (!wrapped
&& si
->prio
!= swap_info
[next
].prio
)) {
192 next
= swap_list
.head
;
196 if (!si
->highest_bit
)
198 if (!(si
->flags
& SWP_WRITEOK
))
201 swap_list
.next
= next
;
202 offset
= scan_swap_map(si
);
204 spin_unlock(&swap_lock
);
205 return swp_entry(type
, offset
);
207 next
= swap_list
.next
;
212 spin_unlock(&swap_lock
);
213 return (swp_entry_t
) {0};
216 swp_entry_t
get_swap_page_of_type(int type
)
218 struct swap_info_struct
*si
;
221 spin_lock(&swap_lock
);
222 si
= swap_info
+ type
;
223 if (si
->flags
& SWP_WRITEOK
) {
225 offset
= scan_swap_map(si
);
227 spin_unlock(&swap_lock
);
228 return swp_entry(type
, offset
);
232 spin_unlock(&swap_lock
);
233 return (swp_entry_t
) {0};
236 static struct swap_info_struct
* swap_info_get(swp_entry_t entry
)
238 struct swap_info_struct
* p
;
239 unsigned long offset
, type
;
243 type
= swp_type(entry
);
244 if (type
>= nr_swapfiles
)
246 p
= & swap_info
[type
];
247 if (!(p
->flags
& SWP_USED
))
249 offset
= swp_offset(entry
);
250 if (offset
>= p
->max
)
252 if (!p
->swap_map
[offset
])
254 spin_lock(&swap_lock
);
258 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_offset
, entry
.val
);
261 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_offset
, entry
.val
);
264 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_file
, entry
.val
);
267 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_file
, entry
.val
);
272 static int swap_entry_free(struct swap_info_struct
*p
, unsigned long offset
)
274 int count
= p
->swap_map
[offset
];
276 if (count
< SWAP_MAP_MAX
) {
278 p
->swap_map
[offset
] = count
;
280 if (offset
< p
->lowest_bit
)
281 p
->lowest_bit
= offset
;
282 if (offset
> p
->highest_bit
)
283 p
->highest_bit
= offset
;
284 if (p
->prio
> swap_info
[swap_list
.next
].prio
)
285 swap_list
.next
= p
- swap_info
;
294 * Caller has made sure that the swapdevice corresponding to entry
295 * is still around or has not been recycled.
297 void swap_free(swp_entry_t entry
)
299 struct swap_info_struct
* p
;
301 p
= swap_info_get(entry
);
303 swap_entry_free(p
, swp_offset(entry
));
304 spin_unlock(&swap_lock
);
309 * How many references to page are currently swapped out?
311 static inline int page_swapcount(struct page
*page
)
314 struct swap_info_struct
*p
;
317 entry
.val
= page_private(page
);
318 p
= swap_info_get(entry
);
320 /* Subtract the 1 for the swap cache itself */
321 count
= p
->swap_map
[swp_offset(entry
)] - 1;
322 spin_unlock(&swap_lock
);
328 * We can use this swap cache entry directly
329 * if there are no other references to it.
331 int can_share_swap_page(struct page
*page
)
335 BUG_ON(!PageLocked(page
));
336 count
= page_mapcount(page
);
337 if (count
<= 1 && PageSwapCache(page
))
338 count
+= page_swapcount(page
);
343 * Work out if there are any other processes sharing this
344 * swap cache page. Free it if you can. Return success.
346 int remove_exclusive_swap_page(struct page
*page
)
349 struct swap_info_struct
* p
;
352 BUG_ON(PagePrivate(page
));
353 BUG_ON(!PageLocked(page
));
355 if (!PageSwapCache(page
))
357 if (PageWriteback(page
))
359 if (page_count(page
) != 2) /* 2: us + cache */
362 entry
.val
= page_private(page
);
363 p
= swap_info_get(entry
);
367 /* Is the only swap cache user the cache itself? */
369 if (p
->swap_map
[swp_offset(entry
)] == 1) {
370 /* Recheck the page count with the swapcache lock held.. */
371 write_lock_irq(&swapper_space
.tree_lock
);
372 if ((page_count(page
) == 2) && !PageWriteback(page
)) {
373 __delete_from_swap_cache(page
);
377 write_unlock_irq(&swapper_space
.tree_lock
);
379 spin_unlock(&swap_lock
);
383 page_cache_release(page
);
390 * Free the swap entry like above, but also try to
391 * free the page cache entry if it is the last user.
393 void free_swap_and_cache(swp_entry_t entry
)
395 struct swap_info_struct
* p
;
396 struct page
*page
= NULL
;
398 p
= swap_info_get(entry
);
400 if (swap_entry_free(p
, swp_offset(entry
)) == 1)
401 page
= find_trylock_page(&swapper_space
, entry
.val
);
402 spin_unlock(&swap_lock
);
407 BUG_ON(PagePrivate(page
));
408 page_cache_get(page
);
409 one_user
= (page_count(page
) == 2);
410 /* Only cache user (+us), or swap space full? Free it! */
411 if (!PageWriteback(page
) && (one_user
|| vm_swap_full())) {
412 delete_from_swap_cache(page
);
416 page_cache_release(page
);
421 * No need to decide whether this PTE shares the swap entry with others,
422 * just let do_wp_page work it out if a write is requested later - to
423 * force COW, vm_page_prot omits write permission from any private vma.
425 static void unuse_pte(struct vm_area_struct
*vma
, pte_t
*pte
,
426 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
428 inc_mm_counter(vma
->vm_mm
, anon_rss
);
430 set_pte_at(vma
->vm_mm
, addr
, pte
,
431 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
432 page_add_anon_rmap(page
, vma
, addr
);
435 * Move the page to the active list so it is not
436 * immediately swapped out again after swapon.
441 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
442 unsigned long addr
, unsigned long end
,
443 swp_entry_t entry
, struct page
*page
)
445 pte_t swp_pte
= swp_entry_to_pte(entry
);
450 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
453 * swapoff spends a _lot_ of time in this loop!
454 * Test inline before going to call unuse_pte.
456 if (unlikely(pte_same(*pte
, swp_pte
))) {
457 unuse_pte(vma
, pte
++, addr
, entry
, page
);
461 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
462 pte_unmap_unlock(pte
- 1, ptl
);
466 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
467 unsigned long addr
, unsigned long end
,
468 swp_entry_t entry
, struct page
*page
)
473 pmd
= pmd_offset(pud
, addr
);
475 next
= pmd_addr_end(addr
, end
);
476 if (pmd_none_or_clear_bad(pmd
))
478 if (unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
))
480 } while (pmd
++, addr
= next
, addr
!= end
);
484 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
485 unsigned long addr
, unsigned long end
,
486 swp_entry_t entry
, struct page
*page
)
491 pud
= pud_offset(pgd
, addr
);
493 next
= pud_addr_end(addr
, end
);
494 if (pud_none_or_clear_bad(pud
))
496 if (unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
))
498 } while (pud
++, addr
= next
, addr
!= end
);
502 static int unuse_vma(struct vm_area_struct
*vma
,
503 swp_entry_t entry
, struct page
*page
)
506 unsigned long addr
, end
, next
;
509 addr
= page_address_in_vma(page
, vma
);
513 end
= addr
+ PAGE_SIZE
;
515 addr
= vma
->vm_start
;
519 pgd
= pgd_offset(vma
->vm_mm
, addr
);
521 next
= pgd_addr_end(addr
, end
);
522 if (pgd_none_or_clear_bad(pgd
))
524 if (unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
))
526 } while (pgd
++, addr
= next
, addr
!= end
);
530 static int unuse_mm(struct mm_struct
*mm
,
531 swp_entry_t entry
, struct page
*page
)
533 struct vm_area_struct
*vma
;
535 if (!down_read_trylock(&mm
->mmap_sem
)) {
537 * Activate page so shrink_cache is unlikely to unmap its
538 * ptes while lock is dropped, so swapoff can make progress.
542 down_read(&mm
->mmap_sem
);
545 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
546 if (vma
->anon_vma
&& unuse_vma(vma
, entry
, page
))
549 up_read(&mm
->mmap_sem
);
551 * Currently unuse_mm cannot fail, but leave error handling
552 * at call sites for now, since we change it from time to time.
558 * Scan swap_map from current position to next entry still in use.
559 * Recycle to start on reaching the end, returning 0 when empty.
561 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
564 unsigned int max
= si
->max
;
565 unsigned int i
= prev
;
569 * No need for swap_lock here: we're just looking
570 * for whether an entry is in use, not modifying it; false
571 * hits are okay, and sys_swapoff() has already prevented new
572 * allocations from this area (while holding swap_lock).
581 * No entries in use at top of swap_map,
582 * loop back to start and recheck there.
588 count
= si
->swap_map
[i
];
589 if (count
&& count
!= SWAP_MAP_BAD
)
596 * We completely avoid races by reading each swap page in advance,
597 * and then search for the process using it. All the necessary
598 * page table adjustments can then be made atomically.
600 static int try_to_unuse(unsigned int type
)
602 struct swap_info_struct
* si
= &swap_info
[type
];
603 struct mm_struct
*start_mm
;
604 unsigned short *swap_map
;
605 unsigned short swcount
;
610 int reset_overflow
= 0;
614 * When searching mms for an entry, a good strategy is to
615 * start at the first mm we freed the previous entry from
616 * (though actually we don't notice whether we or coincidence
617 * freed the entry). Initialize this start_mm with a hold.
619 * A simpler strategy would be to start at the last mm we
620 * freed the previous entry from; but that would take less
621 * advantage of mmlist ordering, which clusters forked mms
622 * together, child after parent. If we race with dup_mmap(), we
623 * prefer to resolve parent before child, lest we miss entries
624 * duplicated after we scanned child: using last mm would invert
625 * that. Though it's only a serious concern when an overflowed
626 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
629 atomic_inc(&init_mm
.mm_users
);
632 * Keep on scanning until all entries have gone. Usually,
633 * one pass through swap_map is enough, but not necessarily:
634 * there are races when an instance of an entry might be missed.
636 while ((i
= find_next_to_unuse(si
, i
)) != 0) {
637 if (signal_pending(current
)) {
643 * Get a page for the entry, using the existing swap
644 * cache page if there is one. Otherwise, get a clean
645 * page and read the swap into it.
647 swap_map
= &si
->swap_map
[i
];
648 entry
= swp_entry(type
, i
);
650 page
= read_swap_cache_async(entry
, NULL
, 0);
653 * Either swap_duplicate() failed because entry
654 * has been freed independently, and will not be
655 * reused since sys_swapoff() already disabled
656 * allocation from here, or alloc_page() failed.
665 * Don't hold on to start_mm if it looks like exiting.
667 if (atomic_read(&start_mm
->mm_users
) == 1) {
670 atomic_inc(&init_mm
.mm_users
);
674 * Wait for and lock page. When do_swap_page races with
675 * try_to_unuse, do_swap_page can handle the fault much
676 * faster than try_to_unuse can locate the entry. This
677 * apparently redundant "wait_on_page_locked" lets try_to_unuse
678 * defer to do_swap_page in such a case - in some tests,
679 * do_swap_page and try_to_unuse repeatedly compete.
681 wait_on_page_locked(page
);
682 wait_on_page_writeback(page
);
684 if (!PageSwapCache(page
)) {
685 /* Page migration has occured */
687 page_cache_release(page
);
690 wait_on_page_writeback(page
);
693 * Remove all references to entry.
694 * Whenever we reach init_mm, there's no address space
695 * to search, but use it as a reminder to search shmem.
700 if (start_mm
== &init_mm
)
701 shmem
= shmem_unuse(entry
, page
);
703 retval
= unuse_mm(start_mm
, entry
, page
);
706 int set_start_mm
= (*swap_map
>= swcount
);
707 struct list_head
*p
= &start_mm
->mmlist
;
708 struct mm_struct
*new_start_mm
= start_mm
;
709 struct mm_struct
*prev_mm
= start_mm
;
710 struct mm_struct
*mm
;
712 atomic_inc(&new_start_mm
->mm_users
);
713 atomic_inc(&prev_mm
->mm_users
);
714 spin_lock(&mmlist_lock
);
715 while (*swap_map
> 1 && !retval
&&
716 (p
= p
->next
) != &start_mm
->mmlist
) {
717 mm
= list_entry(p
, struct mm_struct
, mmlist
);
718 if (atomic_inc_return(&mm
->mm_users
) == 1) {
719 atomic_dec(&mm
->mm_users
);
722 spin_unlock(&mmlist_lock
);
731 else if (mm
== &init_mm
) {
733 shmem
= shmem_unuse(entry
, page
);
735 retval
= unuse_mm(mm
, entry
, page
);
736 if (set_start_mm
&& *swap_map
< swcount
) {
738 atomic_inc(&mm
->mm_users
);
742 spin_lock(&mmlist_lock
);
744 spin_unlock(&mmlist_lock
);
747 start_mm
= new_start_mm
;
751 page_cache_release(page
);
756 * How could swap count reach 0x7fff when the maximum
757 * pid is 0x7fff, and there's no way to repeat a swap
758 * page within an mm (except in shmem, where it's the
759 * shared object which takes the reference count)?
760 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
762 * If that's wrong, then we should worry more about
763 * exit_mmap() and do_munmap() cases described above:
764 * we might be resetting SWAP_MAP_MAX too early here.
765 * We know "Undead"s can happen, they're okay, so don't
766 * report them; but do report if we reset SWAP_MAP_MAX.
768 if (*swap_map
== SWAP_MAP_MAX
) {
769 spin_lock(&swap_lock
);
771 spin_unlock(&swap_lock
);
776 * If a reference remains (rare), we would like to leave
777 * the page in the swap cache; but try_to_unmap could
778 * then re-duplicate the entry once we drop page lock,
779 * so we might loop indefinitely; also, that page could
780 * not be swapped out to other storage meanwhile. So:
781 * delete from cache even if there's another reference,
782 * after ensuring that the data has been saved to disk -
783 * since if the reference remains (rarer), it will be
784 * read from disk into another page. Splitting into two
785 * pages would be incorrect if swap supported "shared
786 * private" pages, but they are handled by tmpfs files.
788 * Note shmem_unuse already deleted a swappage from
789 * the swap cache, unless the move to filepage failed:
790 * in which case it left swappage in cache, lowered its
791 * swap count to pass quickly through the loops above,
792 * and now we must reincrement count to try again later.
794 if ((*swap_map
> 1) && PageDirty(page
) && PageSwapCache(page
)) {
795 struct writeback_control wbc
= {
796 .sync_mode
= WB_SYNC_NONE
,
799 swap_writepage(page
, &wbc
);
801 wait_on_page_writeback(page
);
803 if (PageSwapCache(page
)) {
805 swap_duplicate(entry
);
807 delete_from_swap_cache(page
);
811 * So we could skip searching mms once swap count went
812 * to 1, we did not mark any present ptes as dirty: must
813 * mark page dirty so shrink_list will preserve it.
817 page_cache_release(page
);
820 * Make sure that we aren't completely killing
821 * interactive performance.
827 if (reset_overflow
) {
828 printk(KERN_WARNING
"swapoff: cleared swap entry overflow\n");
835 * After a successful try_to_unuse, if no swap is now in use, we know
836 * we can empty the mmlist. swap_lock must be held on entry and exit.
837 * Note that mmlist_lock nests inside swap_lock, and an mm must be
838 * added to the mmlist just after page_duplicate - before would be racy.
840 static void drain_mmlist(void)
842 struct list_head
*p
, *next
;
845 for (i
= 0; i
< nr_swapfiles
; i
++)
846 if (swap_info
[i
].inuse_pages
)
848 spin_lock(&mmlist_lock
);
849 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
851 spin_unlock(&mmlist_lock
);
855 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
856 * corresponds to page offset `offset'.
858 sector_t
map_swap_page(struct swap_info_struct
*sis
, pgoff_t offset
)
860 struct swap_extent
*se
= sis
->curr_swap_extent
;
861 struct swap_extent
*start_se
= se
;
864 struct list_head
*lh
;
866 if (se
->start_page
<= offset
&&
867 offset
< (se
->start_page
+ se
->nr_pages
)) {
868 return se
->start_block
+ (offset
- se
->start_page
);
871 if (lh
== &sis
->extent_list
)
873 se
= list_entry(lh
, struct swap_extent
, list
);
874 sis
->curr_swap_extent
= se
;
875 BUG_ON(se
== start_se
); /* It *must* be present */
880 * Free all of a swapdev's extent information
882 static void destroy_swap_extents(struct swap_info_struct
*sis
)
884 while (!list_empty(&sis
->extent_list
)) {
885 struct swap_extent
*se
;
887 se
= list_entry(sis
->extent_list
.next
,
888 struct swap_extent
, list
);
895 * Add a block range (and the corresponding page range) into this swapdev's
896 * extent list. The extent list is kept sorted in page order.
898 * This function rather assumes that it is called in ascending page order.
901 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
902 unsigned long nr_pages
, sector_t start_block
)
904 struct swap_extent
*se
;
905 struct swap_extent
*new_se
;
906 struct list_head
*lh
;
908 lh
= sis
->extent_list
.prev
; /* The highest page extent */
909 if (lh
!= &sis
->extent_list
) {
910 se
= list_entry(lh
, struct swap_extent
, list
);
911 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
912 if (se
->start_block
+ se
->nr_pages
== start_block
) {
914 se
->nr_pages
+= nr_pages
;
920 * No merge. Insert a new extent, preserving ordering.
922 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
925 new_se
->start_page
= start_page
;
926 new_se
->nr_pages
= nr_pages
;
927 new_se
->start_block
= start_block
;
929 list_add_tail(&new_se
->list
, &sis
->extent_list
);
934 * A `swap extent' is a simple thing which maps a contiguous range of pages
935 * onto a contiguous range of disk blocks. An ordered list of swap extents
936 * is built at swapon time and is then used at swap_writepage/swap_readpage
937 * time for locating where on disk a page belongs.
939 * If the swapfile is an S_ISBLK block device, a single extent is installed.
940 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
941 * swap files identically.
943 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
944 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
945 * swapfiles are handled *identically* after swapon time.
947 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
948 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
949 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
950 * requirements, they are simply tossed out - we will never use those blocks
953 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
954 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
955 * which will scribble on the fs.
957 * The amount of disk space which a single swap extent represents varies.
958 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
959 * extents in the list. To avoid much list walking, we cache the previous
960 * search location in `curr_swap_extent', and start new searches from there.
961 * This is extremely effective. The average number of iterations in
962 * map_swap_page() has been measured at about 0.3 per page. - akpm.
964 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
967 unsigned blocks_per_page
;
968 unsigned long page_no
;
970 sector_t probe_block
;
972 sector_t lowest_block
= -1;
973 sector_t highest_block
= 0;
977 inode
= sis
->swap_file
->f_mapping
->host
;
978 if (S_ISBLK(inode
->i_mode
)) {
979 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
984 blkbits
= inode
->i_blkbits
;
985 blocks_per_page
= PAGE_SIZE
>> blkbits
;
988 * Map all the blocks into the extent list. This code doesn't try
993 last_block
= i_size_read(inode
) >> blkbits
;
994 while ((probe_block
+ blocks_per_page
) <= last_block
&&
995 page_no
< sis
->max
) {
996 unsigned block_in_page
;
997 sector_t first_block
;
999 first_block
= bmap(inode
, probe_block
);
1000 if (first_block
== 0)
1004 * It must be PAGE_SIZE aligned on-disk
1006 if (first_block
& (blocks_per_page
- 1)) {
1011 for (block_in_page
= 1; block_in_page
< blocks_per_page
;
1015 block
= bmap(inode
, probe_block
+ block_in_page
);
1018 if (block
!= first_block
+ block_in_page
) {
1025 first_block
>>= (PAGE_SHIFT
- blkbits
);
1026 if (page_no
) { /* exclude the header page */
1027 if (first_block
< lowest_block
)
1028 lowest_block
= first_block
;
1029 if (first_block
> highest_block
)
1030 highest_block
= first_block
;
1034 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1036 ret
= add_swap_extent(sis
, page_no
, 1, first_block
);
1041 probe_block
+= blocks_per_page
;
1046 *span
= 1 + highest_block
- lowest_block
;
1048 page_no
= 1; /* force Empty message */
1050 sis
->pages
= page_no
- 1;
1051 sis
->highest_bit
= page_no
- 1;
1053 sis
->curr_swap_extent
= list_entry(sis
->extent_list
.prev
,
1054 struct swap_extent
, list
);
1057 printk(KERN_ERR
"swapon: swapfile has holes\n");
1063 #if 0 /* We don't need this yet */
1064 #include <linux/backing-dev.h>
1065 int page_queue_congested(struct page
*page
)
1067 struct backing_dev_info
*bdi
;
1069 BUG_ON(!PageLocked(page
)); /* It pins the swap_info_struct */
1071 if (PageSwapCache(page
)) {
1072 swp_entry_t entry
= { .val
= page_private(page
) };
1073 struct swap_info_struct
*sis
;
1075 sis
= get_swap_info_struct(swp_type(entry
));
1076 bdi
= sis
->bdev
->bd_inode
->i_mapping
->backing_dev_info
;
1078 bdi
= page
->mapping
->backing_dev_info
;
1079 return bdi_write_congested(bdi
);
1083 asmlinkage
long sys_swapoff(const char __user
* specialfile
)
1085 struct swap_info_struct
* p
= NULL
;
1086 unsigned short *swap_map
;
1087 struct file
*swap_file
, *victim
;
1088 struct address_space
*mapping
;
1089 struct inode
*inode
;
1094 if (!capable(CAP_SYS_ADMIN
))
1097 pathname
= getname(specialfile
);
1098 err
= PTR_ERR(pathname
);
1099 if (IS_ERR(pathname
))
1102 victim
= filp_open(pathname
, O_RDWR
|O_LARGEFILE
, 0);
1104 err
= PTR_ERR(victim
);
1108 mapping
= victim
->f_mapping
;
1110 spin_lock(&swap_lock
);
1111 for (type
= swap_list
.head
; type
>= 0; type
= swap_info
[type
].next
) {
1112 p
= swap_info
+ type
;
1113 if ((p
->flags
& SWP_ACTIVE
) == SWP_ACTIVE
) {
1114 if (p
->swap_file
->f_mapping
== mapping
)
1121 spin_unlock(&swap_lock
);
1124 if (!security_vm_enough_memory(p
->pages
))
1125 vm_unacct_memory(p
->pages
);
1128 spin_unlock(&swap_lock
);
1132 swap_list
.head
= p
->next
;
1134 swap_info
[prev
].next
= p
->next
;
1136 if (type
== swap_list
.next
) {
1137 /* just pick something that's safe... */
1138 swap_list
.next
= swap_list
.head
;
1140 nr_swap_pages
-= p
->pages
;
1141 total_swap_pages
-= p
->pages
;
1142 p
->flags
&= ~SWP_WRITEOK
;
1143 spin_unlock(&swap_lock
);
1145 current
->flags
|= PF_SWAPOFF
;
1146 err
= try_to_unuse(type
);
1147 current
->flags
&= ~PF_SWAPOFF
;
1150 /* re-insert swap space back into swap_list */
1151 spin_lock(&swap_lock
);
1152 for (prev
= -1, i
= swap_list
.head
; i
>= 0; prev
= i
, i
= swap_info
[i
].next
)
1153 if (p
->prio
>= swap_info
[i
].prio
)
1157 swap_list
.head
= swap_list
.next
= p
- swap_info
;
1159 swap_info
[prev
].next
= p
- swap_info
;
1160 nr_swap_pages
+= p
->pages
;
1161 total_swap_pages
+= p
->pages
;
1162 p
->flags
|= SWP_WRITEOK
;
1163 spin_unlock(&swap_lock
);
1167 /* wait for any unplug function to finish */
1168 down_write(&swap_unplug_sem
);
1169 up_write(&swap_unplug_sem
);
1171 destroy_swap_extents(p
);
1172 mutex_lock(&swapon_mutex
);
1173 spin_lock(&swap_lock
);
1176 /* wait for anyone still in scan_swap_map */
1177 p
->highest_bit
= 0; /* cuts scans short */
1178 while (p
->flags
>= SWP_SCANNING
) {
1179 spin_unlock(&swap_lock
);
1180 schedule_timeout_uninterruptible(1);
1181 spin_lock(&swap_lock
);
1184 swap_file
= p
->swap_file
;
1185 p
->swap_file
= NULL
;
1187 swap_map
= p
->swap_map
;
1190 spin_unlock(&swap_lock
);
1191 mutex_unlock(&swapon_mutex
);
1193 inode
= mapping
->host
;
1194 if (S_ISBLK(inode
->i_mode
)) {
1195 struct block_device
*bdev
= I_BDEV(inode
);
1196 set_blocksize(bdev
, p
->old_block_size
);
1199 mutex_lock(&inode
->i_mutex
);
1200 inode
->i_flags
&= ~S_SWAPFILE
;
1201 mutex_unlock(&inode
->i_mutex
);
1203 filp_close(swap_file
, NULL
);
1207 filp_close(victim
, NULL
);
1212 #ifdef CONFIG_PROC_FS
1214 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
1216 struct swap_info_struct
*ptr
= swap_info
;
1220 mutex_lock(&swapon_mutex
);
1222 for (i
= 0; i
< nr_swapfiles
; i
++, ptr
++) {
1223 if (!(ptr
->flags
& SWP_USED
) || !ptr
->swap_map
)
1232 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
1234 struct swap_info_struct
*ptr
= v
;
1235 struct swap_info_struct
*endptr
= swap_info
+ nr_swapfiles
;
1237 for (++ptr
; ptr
< endptr
; ptr
++) {
1238 if (!(ptr
->flags
& SWP_USED
) || !ptr
->swap_map
)
1247 static void swap_stop(struct seq_file
*swap
, void *v
)
1249 mutex_unlock(&swapon_mutex
);
1252 static int swap_show(struct seq_file
*swap
, void *v
)
1254 struct swap_info_struct
*ptr
= v
;
1259 seq_puts(swap
, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1261 file
= ptr
->swap_file
;
1262 len
= seq_path(swap
, file
->f_vfsmnt
, file
->f_dentry
, " \t\n\\");
1263 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
1264 len
< 40 ? 40 - len
: 1, " ",
1265 S_ISBLK(file
->f_dentry
->d_inode
->i_mode
) ?
1266 "partition" : "file\t",
1267 ptr
->pages
<< (PAGE_SHIFT
- 10),
1268 ptr
->inuse_pages
<< (PAGE_SHIFT
- 10),
1273 static struct seq_operations swaps_op
= {
1274 .start
= swap_start
,
1280 static int swaps_open(struct inode
*inode
, struct file
*file
)
1282 return seq_open(file
, &swaps_op
);
1285 static struct file_operations proc_swaps_operations
= {
1288 .llseek
= seq_lseek
,
1289 .release
= seq_release
,
1292 static int __init
procswaps_init(void)
1294 struct proc_dir_entry
*entry
;
1296 entry
= create_proc_entry("swaps", 0, NULL
);
1298 entry
->proc_fops
= &proc_swaps_operations
;
1301 __initcall(procswaps_init
);
1302 #endif /* CONFIG_PROC_FS */
1305 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1307 * The swapon system call
1309 asmlinkage
long sys_swapon(const char __user
* specialfile
, int swap_flags
)
1311 struct swap_info_struct
* p
;
1313 struct block_device
*bdev
= NULL
;
1314 struct file
*swap_file
= NULL
;
1315 struct address_space
*mapping
;
1319 static int least_priority
;
1320 union swap_header
*swap_header
= NULL
;
1321 int swap_header_version
;
1322 unsigned int nr_good_pages
= 0;
1325 unsigned long maxpages
= 1;
1327 unsigned short *swap_map
;
1328 struct page
*page
= NULL
;
1329 struct inode
*inode
= NULL
;
1332 if (!capable(CAP_SYS_ADMIN
))
1334 spin_lock(&swap_lock
);
1336 for (type
= 0 ; type
< nr_swapfiles
; type
++,p
++)
1337 if (!(p
->flags
& SWP_USED
))
1341 * Test if adding another swap device is possible. There are
1342 * two limiting factors: 1) the number of bits for the swap
1343 * type swp_entry_t definition and 2) the number of bits for
1344 * the swap type in the swap ptes as defined by the different
1345 * architectures. To honor both limitations a swap entry
1346 * with swap offset 0 and swap type ~0UL is created, encoded
1347 * to a swap pte, decoded to a swp_entry_t again and finally
1348 * the swap type part is extracted. This will mask all bits
1349 * from the initial ~0UL that can't be encoded in either the
1350 * swp_entry_t or the architecture definition of a swap pte.
1352 if (type
> swp_type(pte_to_swp_entry(swp_entry_to_pte(swp_entry(~0UL,0))))) {
1353 spin_unlock(&swap_lock
);
1356 if (type
>= nr_swapfiles
)
1357 nr_swapfiles
= type
+1;
1358 INIT_LIST_HEAD(&p
->extent_list
);
1359 p
->flags
= SWP_USED
;
1360 p
->swap_file
= NULL
;
1361 p
->old_block_size
= 0;
1368 if (swap_flags
& SWAP_FLAG_PREFER
) {
1370 (swap_flags
& SWAP_FLAG_PRIO_MASK
)>>SWAP_FLAG_PRIO_SHIFT
;
1372 p
->prio
= --least_priority
;
1374 spin_unlock(&swap_lock
);
1375 name
= getname(specialfile
);
1376 error
= PTR_ERR(name
);
1381 swap_file
= filp_open(name
, O_RDWR
|O_LARGEFILE
, 0);
1382 error
= PTR_ERR(swap_file
);
1383 if (IS_ERR(swap_file
)) {
1388 p
->swap_file
= swap_file
;
1389 mapping
= swap_file
->f_mapping
;
1390 inode
= mapping
->host
;
1393 for (i
= 0; i
< nr_swapfiles
; i
++) {
1394 struct swap_info_struct
*q
= &swap_info
[i
];
1396 if (i
== type
|| !q
->swap_file
)
1398 if (mapping
== q
->swap_file
->f_mapping
)
1403 if (S_ISBLK(inode
->i_mode
)) {
1404 bdev
= I_BDEV(inode
);
1405 error
= bd_claim(bdev
, sys_swapon
);
1411 p
->old_block_size
= block_size(bdev
);
1412 error
= set_blocksize(bdev
, PAGE_SIZE
);
1416 } else if (S_ISREG(inode
->i_mode
)) {
1417 p
->bdev
= inode
->i_sb
->s_bdev
;
1418 mutex_lock(&inode
->i_mutex
);
1420 if (IS_SWAPFILE(inode
)) {
1428 swapfilesize
= i_size_read(inode
) >> PAGE_SHIFT
;
1431 * Read the swap header.
1433 if (!mapping
->a_ops
->readpage
) {
1437 page
= read_cache_page(mapping
, 0,
1438 (filler_t
*)mapping
->a_ops
->readpage
, swap_file
);
1440 error
= PTR_ERR(page
);
1443 wait_on_page_locked(page
);
1444 if (!PageUptodate(page
))
1447 swap_header
= page_address(page
);
1449 if (!memcmp("SWAP-SPACE",swap_header
->magic
.magic
,10))
1450 swap_header_version
= 1;
1451 else if (!memcmp("SWAPSPACE2",swap_header
->magic
.magic
,10))
1452 swap_header_version
= 2;
1454 printk(KERN_ERR
"Unable to find swap-space signature\n");
1459 switch (swap_header_version
) {
1461 printk(KERN_ERR
"version 0 swap is no longer supported. "
1462 "Use mkswap -v1 %s\n", name
);
1466 /* Check the swap header's sub-version and the size of
1467 the swap file and bad block lists */
1468 if (swap_header
->info
.version
!= 1) {
1470 "Unable to handle swap header version %d\n",
1471 swap_header
->info
.version
);
1477 p
->cluster_next
= 1;
1480 * Find out how many pages are allowed for a single swap
1481 * device. There are two limiting factors: 1) the number of
1482 * bits for the swap offset in the swp_entry_t type and
1483 * 2) the number of bits in the a swap pte as defined by
1484 * the different architectures. In order to find the
1485 * largest possible bit mask a swap entry with swap type 0
1486 * and swap offset ~0UL is created, encoded to a swap pte,
1487 * decoded to a swp_entry_t again and finally the swap
1488 * offset is extracted. This will mask all the bits from
1489 * the initial ~0UL mask that can't be encoded in either
1490 * the swp_entry_t or the architecture definition of a
1493 maxpages
= swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1494 if (maxpages
> swap_header
->info
.last_page
)
1495 maxpages
= swap_header
->info
.last_page
;
1496 p
->highest_bit
= maxpages
- 1;
1501 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
1503 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
1506 /* OK, set up the swap map and apply the bad block list */
1507 if (!(p
->swap_map
= vmalloc(maxpages
* sizeof(short)))) {
1513 memset(p
->swap_map
, 0, maxpages
* sizeof(short));
1514 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
1515 int page_nr
= swap_header
->info
.badpages
[i
];
1516 if (page_nr
<= 0 || page_nr
>= swap_header
->info
.last_page
)
1519 p
->swap_map
[page_nr
] = SWAP_MAP_BAD
;
1521 nr_good_pages
= swap_header
->info
.last_page
-
1522 swap_header
->info
.nr_badpages
-
1523 1 /* header page */;
1528 if (swapfilesize
&& maxpages
> swapfilesize
) {
1530 "Swap area shorter than signature indicates\n");
1534 if (nr_good_pages
) {
1535 p
->swap_map
[0] = SWAP_MAP_BAD
;
1537 p
->pages
= nr_good_pages
;
1538 nr_extents
= setup_swap_extents(p
, &span
);
1539 if (nr_extents
< 0) {
1543 nr_good_pages
= p
->pages
;
1545 if (!nr_good_pages
) {
1546 printk(KERN_WARNING
"Empty swap-file\n");
1551 mutex_lock(&swapon_mutex
);
1552 spin_lock(&swap_lock
);
1553 p
->flags
= SWP_ACTIVE
;
1554 nr_swap_pages
+= nr_good_pages
;
1555 total_swap_pages
+= nr_good_pages
;
1557 printk(KERN_INFO
"Adding %uk swap on %s. "
1558 "Priority:%d extents:%d across:%lluk\n",
1559 nr_good_pages
<<(PAGE_SHIFT
-10), name
, p
->prio
,
1560 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10));
1562 /* insert swap space into swap_list: */
1564 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
].next
) {
1565 if (p
->prio
>= swap_info
[i
].prio
) {
1572 swap_list
.head
= swap_list
.next
= p
- swap_info
;
1574 swap_info
[prev
].next
= p
- swap_info
;
1576 spin_unlock(&swap_lock
);
1577 mutex_unlock(&swapon_mutex
);
1582 set_blocksize(bdev
, p
->old_block_size
);
1585 destroy_swap_extents(p
);
1587 spin_lock(&swap_lock
);
1588 swap_map
= p
->swap_map
;
1589 p
->swap_file
= NULL
;
1592 if (!(swap_flags
& SWAP_FLAG_PREFER
))
1594 spin_unlock(&swap_lock
);
1597 filp_close(swap_file
, NULL
);
1599 if (page
&& !IS_ERR(page
)) {
1601 page_cache_release(page
);
1607 inode
->i_flags
|= S_SWAPFILE
;
1608 mutex_unlock(&inode
->i_mutex
);
1613 void si_swapinfo(struct sysinfo
*val
)
1616 unsigned long nr_to_be_unused
= 0;
1618 spin_lock(&swap_lock
);
1619 for (i
= 0; i
< nr_swapfiles
; i
++) {
1620 if (!(swap_info
[i
].flags
& SWP_USED
) ||
1621 (swap_info
[i
].flags
& SWP_WRITEOK
))
1623 nr_to_be_unused
+= swap_info
[i
].inuse_pages
;
1625 val
->freeswap
= nr_swap_pages
+ nr_to_be_unused
;
1626 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
1627 spin_unlock(&swap_lock
);
1631 * Verify that a swap entry is valid and increment its swap map count.
1633 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1634 * "permanent", but will be reclaimed by the next swapoff.
1636 int swap_duplicate(swp_entry_t entry
)
1638 struct swap_info_struct
* p
;
1639 unsigned long offset
, type
;
1642 type
= swp_type(entry
);
1643 if (type
>= nr_swapfiles
)
1645 p
= type
+ swap_info
;
1646 offset
= swp_offset(entry
);
1648 spin_lock(&swap_lock
);
1649 if (offset
< p
->max
&& p
->swap_map
[offset
]) {
1650 if (p
->swap_map
[offset
] < SWAP_MAP_MAX
- 1) {
1651 p
->swap_map
[offset
]++;
1653 } else if (p
->swap_map
[offset
] <= SWAP_MAP_MAX
) {
1654 if (swap_overflow
++ < 5)
1655 printk(KERN_WARNING
"swap_dup: swap entry overflow\n");
1656 p
->swap_map
[offset
] = SWAP_MAP_MAX
;
1660 spin_unlock(&swap_lock
);
1665 printk(KERN_ERR
"swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
1669 struct swap_info_struct
*
1670 get_swap_info_struct(unsigned type
)
1672 return &swap_info
[type
];
1676 * swap_lock prevents swap_map being freed. Don't grab an extra
1677 * reference on the swaphandle, it doesn't matter if it becomes unused.
1679 int valid_swaphandles(swp_entry_t entry
, unsigned long *offset
)
1681 int ret
= 0, i
= 1 << page_cluster
;
1683 struct swap_info_struct
*swapdev
= swp_type(entry
) + swap_info
;
1685 if (!page_cluster
) /* no readahead */
1687 toff
= (swp_offset(entry
) >> page_cluster
) << page_cluster
;
1688 if (!toff
) /* first page is swap header */
1692 spin_lock(&swap_lock
);
1694 /* Don't read-ahead past the end of the swap area */
1695 if (toff
>= swapdev
->max
)
1697 /* Don't read in free or bad pages */
1698 if (!swapdev
->swap_map
[toff
])
1700 if (swapdev
->swap_map
[toff
] == SWAP_MAP_BAD
)
1705 spin_unlock(&swap_lock
);