4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
9 #include <linux/hugetlb.h>
10 #include <linux/mman.h>
11 #include <linux/slab.h>
12 #include <linux/kernel_stat.h>
13 #include <linux/swap.h>
14 #include <linux/vmalloc.h>
15 #include <linux/pagemap.h>
16 #include <linux/namei.h>
17 #include <linux/shm.h>
18 #include <linux/blkdev.h>
19 #include <linux/writeback.h>
20 #include <linux/proc_fs.h>
21 #include <linux/seq_file.h>
22 #include <linux/init.h>
23 #include <linux/module.h>
24 #include <linux/rmap.h>
25 #include <linux/security.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mutex.h>
28 #include <linux/capability.h>
29 #include <linux/syscalls.h>
30 #include <linux/memcontrol.h>
32 #include <asm/pgtable.h>
33 #include <asm/tlbflush.h>
34 #include <linux/swapops.h>
36 static DEFINE_SPINLOCK(swap_lock
);
37 static unsigned int nr_swapfiles
;
38 long total_swap_pages
;
39 static int swap_overflow
;
40 static int least_priority
;
42 static const char Bad_file
[] = "Bad swap file entry ";
43 static const char Unused_file
[] = "Unused swap file entry ";
44 static const char Bad_offset
[] = "Bad swap offset entry ";
45 static const char Unused_offset
[] = "Unused swap offset entry ";
47 static struct swap_list_t swap_list
= {-1, -1};
49 static struct swap_info_struct swap_info
[MAX_SWAPFILES
];
51 static DEFINE_MUTEX(swapon_mutex
);
54 * We need this because the bdev->unplug_fn can sleep and we cannot
55 * hold swap_lock while calling the unplug_fn. And swap_lock
56 * cannot be turned into a mutex.
58 static DECLARE_RWSEM(swap_unplug_sem
);
60 void swap_unplug_io_fn(struct backing_dev_info
*unused_bdi
, struct page
*page
)
64 down_read(&swap_unplug_sem
);
65 entry
.val
= page_private(page
);
66 if (PageSwapCache(page
)) {
67 struct block_device
*bdev
= swap_info
[swp_type(entry
)].bdev
;
68 struct backing_dev_info
*bdi
;
71 * If the page is removed from swapcache from under us (with a
72 * racy try_to_unuse/swapoff) we need an additional reference
73 * count to avoid reading garbage from page_private(page) above.
74 * If the WARN_ON triggers during a swapoff it maybe the race
75 * condition and it's harmless. However if it triggers without
76 * swapoff it signals a problem.
78 WARN_ON(page_count(page
) <= 1);
80 bdi
= bdev
->bd_inode
->i_mapping
->backing_dev_info
;
81 blk_run_backing_dev(bdi
, page
);
83 up_read(&swap_unplug_sem
);
86 #define SWAPFILE_CLUSTER 256
87 #define LATENCY_LIMIT 256
89 static inline unsigned long scan_swap_map(struct swap_info_struct
*si
)
91 unsigned long offset
, last_in_cluster
;
92 int latency_ration
= LATENCY_LIMIT
;
95 * We try to cluster swap pages by allocating them sequentially
96 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
97 * way, however, we resort to first-free allocation, starting
98 * a new cluster. This prevents us from scattering swap pages
99 * all over the entire swap partition, so that we reduce
100 * overall disk seek times between swap pages. -- sct
101 * But we do now try to find an empty cluster. -Andrea
104 si
->flags
+= SWP_SCANNING
;
105 if (unlikely(!si
->cluster_nr
)) {
106 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
107 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
)
109 spin_unlock(&swap_lock
);
111 offset
= si
->lowest_bit
;
112 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
114 /* Locate the first empty (unaligned) cluster */
115 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
116 if (si
->swap_map
[offset
])
117 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
118 else if (offset
== last_in_cluster
) {
119 spin_lock(&swap_lock
);
120 si
->cluster_next
= offset
-SWAPFILE_CLUSTER
+1;
123 if (unlikely(--latency_ration
< 0)) {
125 latency_ration
= LATENCY_LIMIT
;
128 spin_lock(&swap_lock
);
134 offset
= si
->cluster_next
;
135 if (offset
> si
->highest_bit
)
136 lowest
: offset
= si
->lowest_bit
;
137 checks
: if (!(si
->flags
& SWP_WRITEOK
))
139 if (!si
->highest_bit
)
141 if (!si
->swap_map
[offset
]) {
142 if (offset
== si
->lowest_bit
)
144 if (offset
== si
->highest_bit
)
147 if (si
->inuse_pages
== si
->pages
) {
148 si
->lowest_bit
= si
->max
;
151 si
->swap_map
[offset
] = 1;
152 si
->cluster_next
= offset
+ 1;
153 si
->flags
-= SWP_SCANNING
;
157 spin_unlock(&swap_lock
);
158 while (++offset
<= si
->highest_bit
) {
159 if (!si
->swap_map
[offset
]) {
160 spin_lock(&swap_lock
);
163 if (unlikely(--latency_ration
< 0)) {
165 latency_ration
= LATENCY_LIMIT
;
168 spin_lock(&swap_lock
);
172 si
->flags
-= SWP_SCANNING
;
176 swp_entry_t
get_swap_page(void)
178 struct swap_info_struct
*si
;
183 spin_lock(&swap_lock
);
184 if (nr_swap_pages
<= 0)
188 for (type
= swap_list
.next
; type
>= 0 && wrapped
< 2; type
= next
) {
189 si
= swap_info
+ type
;
192 (!wrapped
&& si
->prio
!= swap_info
[next
].prio
)) {
193 next
= swap_list
.head
;
197 if (!si
->highest_bit
)
199 if (!(si
->flags
& SWP_WRITEOK
))
202 swap_list
.next
= next
;
203 offset
= scan_swap_map(si
);
205 spin_unlock(&swap_lock
);
206 return swp_entry(type
, offset
);
208 next
= swap_list
.next
;
213 spin_unlock(&swap_lock
);
214 return (swp_entry_t
) {0};
217 swp_entry_t
get_swap_page_of_type(int type
)
219 struct swap_info_struct
*si
;
222 spin_lock(&swap_lock
);
223 si
= swap_info
+ type
;
224 if (si
->flags
& SWP_WRITEOK
) {
226 offset
= scan_swap_map(si
);
228 spin_unlock(&swap_lock
);
229 return swp_entry(type
, offset
);
233 spin_unlock(&swap_lock
);
234 return (swp_entry_t
) {0};
237 static struct swap_info_struct
* swap_info_get(swp_entry_t entry
)
239 struct swap_info_struct
* p
;
240 unsigned long offset
, type
;
244 type
= swp_type(entry
);
245 if (type
>= nr_swapfiles
)
247 p
= & swap_info
[type
];
248 if (!(p
->flags
& SWP_USED
))
250 offset
= swp_offset(entry
);
251 if (offset
>= p
->max
)
253 if (!p
->swap_map
[offset
])
255 spin_lock(&swap_lock
);
259 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_offset
, entry
.val
);
262 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_offset
, entry
.val
);
265 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_file
, entry
.val
);
268 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_file
, entry
.val
);
273 static int swap_entry_free(struct swap_info_struct
*p
, unsigned long offset
)
275 int count
= p
->swap_map
[offset
];
277 if (count
< SWAP_MAP_MAX
) {
279 p
->swap_map
[offset
] = count
;
281 if (offset
< p
->lowest_bit
)
282 p
->lowest_bit
= offset
;
283 if (offset
> p
->highest_bit
)
284 p
->highest_bit
= offset
;
285 if (p
->prio
> swap_info
[swap_list
.next
].prio
)
286 swap_list
.next
= p
- swap_info
;
295 * Caller has made sure that the swapdevice corresponding to entry
296 * is still around or has not been recycled.
298 void swap_free(swp_entry_t entry
)
300 struct swap_info_struct
* p
;
302 p
= swap_info_get(entry
);
304 swap_entry_free(p
, swp_offset(entry
));
305 spin_unlock(&swap_lock
);
310 * How many references to page are currently swapped out?
312 static inline int page_swapcount(struct page
*page
)
315 struct swap_info_struct
*p
;
318 entry
.val
= page_private(page
);
319 p
= swap_info_get(entry
);
321 /* Subtract the 1 for the swap cache itself */
322 count
= p
->swap_map
[swp_offset(entry
)] - 1;
323 spin_unlock(&swap_lock
);
329 * We can write to an anon page without COW if there are no other references
330 * to it. And as a side-effect, free up its swap: because the old content
331 * on disk will never be read, and seeking back there to write new content
332 * later would only waste time away from clustering.
334 int reuse_swap_page(struct page
*page
)
338 VM_BUG_ON(!PageLocked(page
));
339 count
= page_mapcount(page
);
340 if (count
<= 1 && PageSwapCache(page
)) {
341 count
+= page_swapcount(page
);
342 if (count
== 1 && !PageWriteback(page
)) {
343 delete_from_swap_cache(page
);
351 * If swap is getting full, or if there are no more mappings of this page,
352 * then try_to_free_swap is called to free its swap space.
354 int try_to_free_swap(struct page
*page
)
356 VM_BUG_ON(!PageLocked(page
));
358 if (!PageSwapCache(page
))
360 if (PageWriteback(page
))
362 if (page_swapcount(page
))
365 delete_from_swap_cache(page
);
371 * Free the swap entry like above, but also try to
372 * free the page cache entry if it is the last user.
374 void free_swap_and_cache(swp_entry_t entry
)
376 struct swap_info_struct
* p
;
377 struct page
*page
= NULL
;
379 if (is_migration_entry(entry
))
382 p
= swap_info_get(entry
);
384 if (swap_entry_free(p
, swp_offset(entry
)) == 1) {
385 page
= find_get_page(&swapper_space
, entry
.val
);
386 if (page
&& !trylock_page(page
)) {
387 page_cache_release(page
);
391 spin_unlock(&swap_lock
);
395 * Not mapped elsewhere, or swap space full? Free it!
396 * Also recheck PageSwapCache now page is locked (above).
398 if (PageSwapCache(page
) && !PageWriteback(page
) &&
399 (!page_mapped(page
) || vm_swap_full())) {
400 delete_from_swap_cache(page
);
404 page_cache_release(page
);
408 #ifdef CONFIG_HIBERNATION
410 * Find the swap type that corresponds to given device (if any).
412 * @offset - number of the PAGE_SIZE-sized block of the device, starting
413 * from 0, in which the swap header is expected to be located.
415 * This is needed for the suspend to disk (aka swsusp).
417 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
419 struct block_device
*bdev
= NULL
;
423 bdev
= bdget(device
);
425 spin_lock(&swap_lock
);
426 for (i
= 0; i
< nr_swapfiles
; i
++) {
427 struct swap_info_struct
*sis
= swap_info
+ i
;
429 if (!(sis
->flags
& SWP_WRITEOK
))
436 spin_unlock(&swap_lock
);
439 if (bdev
== sis
->bdev
) {
440 struct swap_extent
*se
;
442 se
= list_entry(sis
->extent_list
.next
,
443 struct swap_extent
, list
);
444 if (se
->start_block
== offset
) {
448 spin_unlock(&swap_lock
);
454 spin_unlock(&swap_lock
);
462 * Return either the total number of swap pages of given type, or the number
463 * of free pages of that type (depending on @free)
465 * This is needed for software suspend
467 unsigned int count_swap_pages(int type
, int free
)
471 if (type
< nr_swapfiles
) {
472 spin_lock(&swap_lock
);
473 if (swap_info
[type
].flags
& SWP_WRITEOK
) {
474 n
= swap_info
[type
].pages
;
476 n
-= swap_info
[type
].inuse_pages
;
478 spin_unlock(&swap_lock
);
485 * No need to decide whether this PTE shares the swap entry with others,
486 * just let do_wp_page work it out if a write is requested later - to
487 * force COW, vm_page_prot omits write permission from any private vma.
489 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
490 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
496 if (mem_cgroup_charge(page
, vma
->vm_mm
, GFP_KERNEL
))
499 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
500 if (unlikely(!pte_same(*pte
, swp_entry_to_pte(entry
)))) {
502 mem_cgroup_uncharge_page(page
);
507 inc_mm_counter(vma
->vm_mm
, anon_rss
);
509 set_pte_at(vma
->vm_mm
, addr
, pte
,
510 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
511 page_add_anon_rmap(page
, vma
, addr
);
514 * Move the page to the active list so it is not
515 * immediately swapped out again after swapon.
519 pte_unmap_unlock(pte
, ptl
);
523 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
524 unsigned long addr
, unsigned long end
,
525 swp_entry_t entry
, struct page
*page
)
527 pte_t swp_pte
= swp_entry_to_pte(entry
);
532 * We don't actually need pte lock while scanning for swp_pte: since
533 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
534 * page table while we're scanning; though it could get zapped, and on
535 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
536 * of unmatched parts which look like swp_pte, so unuse_pte must
537 * recheck under pte lock. Scanning without pte lock lets it be
538 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
540 pte
= pte_offset_map(pmd
, addr
);
543 * swapoff spends a _lot_ of time in this loop!
544 * Test inline before going to call unuse_pte.
546 if (unlikely(pte_same(*pte
, swp_pte
))) {
548 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
551 pte
= pte_offset_map(pmd
, addr
);
553 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
559 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
560 unsigned long addr
, unsigned long end
,
561 swp_entry_t entry
, struct page
*page
)
567 pmd
= pmd_offset(pud
, addr
);
569 next
= pmd_addr_end(addr
, end
);
570 if (pmd_none_or_clear_bad(pmd
))
572 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
);
575 } while (pmd
++, addr
= next
, addr
!= end
);
579 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
580 unsigned long addr
, unsigned long end
,
581 swp_entry_t entry
, struct page
*page
)
587 pud
= pud_offset(pgd
, addr
);
589 next
= pud_addr_end(addr
, end
);
590 if (pud_none_or_clear_bad(pud
))
592 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
);
595 } while (pud
++, addr
= next
, addr
!= end
);
599 static int unuse_vma(struct vm_area_struct
*vma
,
600 swp_entry_t entry
, struct page
*page
)
603 unsigned long addr
, end
, next
;
607 addr
= page_address_in_vma(page
, vma
);
611 end
= addr
+ PAGE_SIZE
;
613 addr
= vma
->vm_start
;
617 pgd
= pgd_offset(vma
->vm_mm
, addr
);
619 next
= pgd_addr_end(addr
, end
);
620 if (pgd_none_or_clear_bad(pgd
))
622 ret
= unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
);
625 } while (pgd
++, addr
= next
, addr
!= end
);
629 static int unuse_mm(struct mm_struct
*mm
,
630 swp_entry_t entry
, struct page
*page
)
632 struct vm_area_struct
*vma
;
635 if (!down_read_trylock(&mm
->mmap_sem
)) {
637 * Activate page so shrink_inactive_list is unlikely to unmap
638 * its ptes while lock is dropped, so swapoff can make progress.
642 down_read(&mm
->mmap_sem
);
645 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
646 if (vma
->anon_vma
&& (ret
= unuse_vma(vma
, entry
, page
)))
649 up_read(&mm
->mmap_sem
);
650 return (ret
< 0)? ret
: 0;
654 * Scan swap_map from current position to next entry still in use.
655 * Recycle to start on reaching the end, returning 0 when empty.
657 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
660 unsigned int max
= si
->max
;
661 unsigned int i
= prev
;
665 * No need for swap_lock here: we're just looking
666 * for whether an entry is in use, not modifying it; false
667 * hits are okay, and sys_swapoff() has already prevented new
668 * allocations from this area (while holding swap_lock).
677 * No entries in use at top of swap_map,
678 * loop back to start and recheck there.
684 count
= si
->swap_map
[i
];
685 if (count
&& count
!= SWAP_MAP_BAD
)
692 * We completely avoid races by reading each swap page in advance,
693 * and then search for the process using it. All the necessary
694 * page table adjustments can then be made atomically.
696 static int try_to_unuse(unsigned int type
)
698 struct swap_info_struct
* si
= &swap_info
[type
];
699 struct mm_struct
*start_mm
;
700 unsigned short *swap_map
;
701 unsigned short swcount
;
706 int reset_overflow
= 0;
710 * When searching mms for an entry, a good strategy is to
711 * start at the first mm we freed the previous entry from
712 * (though actually we don't notice whether we or coincidence
713 * freed the entry). Initialize this start_mm with a hold.
715 * A simpler strategy would be to start at the last mm we
716 * freed the previous entry from; but that would take less
717 * advantage of mmlist ordering, which clusters forked mms
718 * together, child after parent. If we race with dup_mmap(), we
719 * prefer to resolve parent before child, lest we miss entries
720 * duplicated after we scanned child: using last mm would invert
721 * that. Though it's only a serious concern when an overflowed
722 * swap count is reset from SWAP_MAP_MAX, preventing a rescan.
725 atomic_inc(&init_mm
.mm_users
);
728 * Keep on scanning until all entries have gone. Usually,
729 * one pass through swap_map is enough, but not necessarily:
730 * there are races when an instance of an entry might be missed.
732 while ((i
= find_next_to_unuse(si
, i
)) != 0) {
733 if (signal_pending(current
)) {
739 * Get a page for the entry, using the existing swap
740 * cache page if there is one. Otherwise, get a clean
741 * page and read the swap into it.
743 swap_map
= &si
->swap_map
[i
];
744 entry
= swp_entry(type
, i
);
745 page
= read_swap_cache_async(entry
,
746 GFP_HIGHUSER_MOVABLE
, NULL
, 0);
749 * Either swap_duplicate() failed because entry
750 * has been freed independently, and will not be
751 * reused since sys_swapoff() already disabled
752 * allocation from here, or alloc_page() failed.
761 * Don't hold on to start_mm if it looks like exiting.
763 if (atomic_read(&start_mm
->mm_users
) == 1) {
766 atomic_inc(&init_mm
.mm_users
);
770 * Wait for and lock page. When do_swap_page races with
771 * try_to_unuse, do_swap_page can handle the fault much
772 * faster than try_to_unuse can locate the entry. This
773 * apparently redundant "wait_on_page_locked" lets try_to_unuse
774 * defer to do_swap_page in such a case - in some tests,
775 * do_swap_page and try_to_unuse repeatedly compete.
777 wait_on_page_locked(page
);
778 wait_on_page_writeback(page
);
780 wait_on_page_writeback(page
);
783 * Remove all references to entry.
784 * Whenever we reach init_mm, there's no address space
785 * to search, but use it as a reminder to search shmem.
790 if (start_mm
== &init_mm
)
791 shmem
= shmem_unuse(entry
, page
);
793 retval
= unuse_mm(start_mm
, entry
, page
);
796 int set_start_mm
= (*swap_map
>= swcount
);
797 struct list_head
*p
= &start_mm
->mmlist
;
798 struct mm_struct
*new_start_mm
= start_mm
;
799 struct mm_struct
*prev_mm
= start_mm
;
800 struct mm_struct
*mm
;
802 atomic_inc(&new_start_mm
->mm_users
);
803 atomic_inc(&prev_mm
->mm_users
);
804 spin_lock(&mmlist_lock
);
805 while (*swap_map
> 1 && !retval
&& !shmem
&&
806 (p
= p
->next
) != &start_mm
->mmlist
) {
807 mm
= list_entry(p
, struct mm_struct
, mmlist
);
808 if (!atomic_inc_not_zero(&mm
->mm_users
))
810 spin_unlock(&mmlist_lock
);
819 else if (mm
== &init_mm
) {
821 shmem
= shmem_unuse(entry
, page
);
823 retval
= unuse_mm(mm
, entry
, page
);
824 if (set_start_mm
&& *swap_map
< swcount
) {
826 atomic_inc(&mm
->mm_users
);
830 spin_lock(&mmlist_lock
);
832 spin_unlock(&mmlist_lock
);
835 start_mm
= new_start_mm
;
838 /* page has already been unlocked and released */
846 page_cache_release(page
);
851 * How could swap count reach 0x7fff when the maximum
852 * pid is 0x7fff, and there's no way to repeat a swap
853 * page within an mm (except in shmem, where it's the
854 * shared object which takes the reference count)?
855 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
857 * If that's wrong, then we should worry more about
858 * exit_mmap() and do_munmap() cases described above:
859 * we might be resetting SWAP_MAP_MAX too early here.
860 * We know "Undead"s can happen, they're okay, so don't
861 * report them; but do report if we reset SWAP_MAP_MAX.
863 if (*swap_map
== SWAP_MAP_MAX
) {
864 spin_lock(&swap_lock
);
866 spin_unlock(&swap_lock
);
871 * If a reference remains (rare), we would like to leave
872 * the page in the swap cache; but try_to_unmap could
873 * then re-duplicate the entry once we drop page lock,
874 * so we might loop indefinitely; also, that page could
875 * not be swapped out to other storage meanwhile. So:
876 * delete from cache even if there's another reference,
877 * after ensuring that the data has been saved to disk -
878 * since if the reference remains (rarer), it will be
879 * read from disk into another page. Splitting into two
880 * pages would be incorrect if swap supported "shared
881 * private" pages, but they are handled by tmpfs files.
883 if ((*swap_map
> 1) && PageDirty(page
) && PageSwapCache(page
)) {
884 struct writeback_control wbc
= {
885 .sync_mode
= WB_SYNC_NONE
,
888 swap_writepage(page
, &wbc
);
890 wait_on_page_writeback(page
);
894 * It is conceivable that a racing task removed this page from
895 * swap cache just before we acquired the page lock at the top,
896 * or while we dropped it in unuse_mm(). The page might even
897 * be back in swap cache on another swap area: that we must not
898 * delete, since it may not have been written out to swap yet.
900 if (PageSwapCache(page
) &&
901 likely(page_private(page
) == entry
.val
))
902 delete_from_swap_cache(page
);
905 * So we could skip searching mms once swap count went
906 * to 1, we did not mark any present ptes as dirty: must
907 * mark page dirty so shrink_page_list will preserve it.
911 page_cache_release(page
);
914 * Make sure that we aren't completely killing
915 * interactive performance.
921 if (reset_overflow
) {
922 printk(KERN_WARNING
"swapoff: cleared swap entry overflow\n");
929 * After a successful try_to_unuse, if no swap is now in use, we know
930 * we can empty the mmlist. swap_lock must be held on entry and exit.
931 * Note that mmlist_lock nests inside swap_lock, and an mm must be
932 * added to the mmlist just after page_duplicate - before would be racy.
934 static void drain_mmlist(void)
936 struct list_head
*p
, *next
;
939 for (i
= 0; i
< nr_swapfiles
; i
++)
940 if (swap_info
[i
].inuse_pages
)
942 spin_lock(&mmlist_lock
);
943 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
945 spin_unlock(&mmlist_lock
);
949 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
950 * corresponds to page offset `offset'.
952 sector_t
map_swap_page(struct swap_info_struct
*sis
, pgoff_t offset
)
954 struct swap_extent
*se
= sis
->curr_swap_extent
;
955 struct swap_extent
*start_se
= se
;
958 struct list_head
*lh
;
960 if (se
->start_page
<= offset
&&
961 offset
< (se
->start_page
+ se
->nr_pages
)) {
962 return se
->start_block
+ (offset
- se
->start_page
);
965 if (lh
== &sis
->extent_list
)
967 se
= list_entry(lh
, struct swap_extent
, list
);
968 sis
->curr_swap_extent
= se
;
969 BUG_ON(se
== start_se
); /* It *must* be present */
973 #ifdef CONFIG_HIBERNATION
975 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
976 * corresponding to given index in swap_info (swap type).
978 sector_t
swapdev_block(int swap_type
, pgoff_t offset
)
980 struct swap_info_struct
*sis
;
982 if (swap_type
>= nr_swapfiles
)
985 sis
= swap_info
+ swap_type
;
986 return (sis
->flags
& SWP_WRITEOK
) ? map_swap_page(sis
, offset
) : 0;
988 #endif /* CONFIG_HIBERNATION */
991 * Free all of a swapdev's extent information
993 static void destroy_swap_extents(struct swap_info_struct
*sis
)
995 while (!list_empty(&sis
->extent_list
)) {
996 struct swap_extent
*se
;
998 se
= list_entry(sis
->extent_list
.next
,
999 struct swap_extent
, list
);
1000 list_del(&se
->list
);
1006 * Add a block range (and the corresponding page range) into this swapdev's
1007 * extent list. The extent list is kept sorted in page order.
1009 * This function rather assumes that it is called in ascending page order.
1012 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
1013 unsigned long nr_pages
, sector_t start_block
)
1015 struct swap_extent
*se
;
1016 struct swap_extent
*new_se
;
1017 struct list_head
*lh
;
1019 lh
= sis
->extent_list
.prev
; /* The highest page extent */
1020 if (lh
!= &sis
->extent_list
) {
1021 se
= list_entry(lh
, struct swap_extent
, list
);
1022 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
1023 if (se
->start_block
+ se
->nr_pages
== start_block
) {
1025 se
->nr_pages
+= nr_pages
;
1031 * No merge. Insert a new extent, preserving ordering.
1033 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
1036 new_se
->start_page
= start_page
;
1037 new_se
->nr_pages
= nr_pages
;
1038 new_se
->start_block
= start_block
;
1040 list_add_tail(&new_se
->list
, &sis
->extent_list
);
1045 * A `swap extent' is a simple thing which maps a contiguous range of pages
1046 * onto a contiguous range of disk blocks. An ordered list of swap extents
1047 * is built at swapon time and is then used at swap_writepage/swap_readpage
1048 * time for locating where on disk a page belongs.
1050 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1051 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1052 * swap files identically.
1054 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1055 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1056 * swapfiles are handled *identically* after swapon time.
1058 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1059 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1060 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1061 * requirements, they are simply tossed out - we will never use those blocks
1064 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1065 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1066 * which will scribble on the fs.
1068 * The amount of disk space which a single swap extent represents varies.
1069 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1070 * extents in the list. To avoid much list walking, we cache the previous
1071 * search location in `curr_swap_extent', and start new searches from there.
1072 * This is extremely effective. The average number of iterations in
1073 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1075 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
1077 struct inode
*inode
;
1078 unsigned blocks_per_page
;
1079 unsigned long page_no
;
1081 sector_t probe_block
;
1082 sector_t last_block
;
1083 sector_t lowest_block
= -1;
1084 sector_t highest_block
= 0;
1088 inode
= sis
->swap_file
->f_mapping
->host
;
1089 if (S_ISBLK(inode
->i_mode
)) {
1090 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1095 blkbits
= inode
->i_blkbits
;
1096 blocks_per_page
= PAGE_SIZE
>> blkbits
;
1099 * Map all the blocks into the extent list. This code doesn't try
1104 last_block
= i_size_read(inode
) >> blkbits
;
1105 while ((probe_block
+ blocks_per_page
) <= last_block
&&
1106 page_no
< sis
->max
) {
1107 unsigned block_in_page
;
1108 sector_t first_block
;
1110 first_block
= bmap(inode
, probe_block
);
1111 if (first_block
== 0)
1115 * It must be PAGE_SIZE aligned on-disk
1117 if (first_block
& (blocks_per_page
- 1)) {
1122 for (block_in_page
= 1; block_in_page
< blocks_per_page
;
1126 block
= bmap(inode
, probe_block
+ block_in_page
);
1129 if (block
!= first_block
+ block_in_page
) {
1136 first_block
>>= (PAGE_SHIFT
- blkbits
);
1137 if (page_no
) { /* exclude the header page */
1138 if (first_block
< lowest_block
)
1139 lowest_block
= first_block
;
1140 if (first_block
> highest_block
)
1141 highest_block
= first_block
;
1145 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1147 ret
= add_swap_extent(sis
, page_no
, 1, first_block
);
1152 probe_block
+= blocks_per_page
;
1157 *span
= 1 + highest_block
- lowest_block
;
1159 page_no
= 1; /* force Empty message */
1161 sis
->pages
= page_no
- 1;
1162 sis
->highest_bit
= page_no
- 1;
1164 sis
->curr_swap_extent
= list_entry(sis
->extent_list
.prev
,
1165 struct swap_extent
, list
);
1168 printk(KERN_ERR
"swapon: swapfile has holes\n");
1174 #if 0 /* We don't need this yet */
1175 #include <linux/backing-dev.h>
1176 int page_queue_congested(struct page
*page
)
1178 struct backing_dev_info
*bdi
;
1180 VM_BUG_ON(!PageLocked(page
)); /* It pins the swap_info_struct */
1182 if (PageSwapCache(page
)) {
1183 swp_entry_t entry
= { .val
= page_private(page
) };
1184 struct swap_info_struct
*sis
;
1186 sis
= get_swap_info_struct(swp_type(entry
));
1187 bdi
= sis
->bdev
->bd_inode
->i_mapping
->backing_dev_info
;
1189 bdi
= page
->mapping
->backing_dev_info
;
1190 return bdi_write_congested(bdi
);
1194 asmlinkage
long sys_swapoff(const char __user
* specialfile
)
1196 struct swap_info_struct
* p
= NULL
;
1197 unsigned short *swap_map
;
1198 struct file
*swap_file
, *victim
;
1199 struct address_space
*mapping
;
1200 struct inode
*inode
;
1205 if (!capable(CAP_SYS_ADMIN
))
1208 pathname
= getname(specialfile
);
1209 err
= PTR_ERR(pathname
);
1210 if (IS_ERR(pathname
))
1213 victim
= filp_open(pathname
, O_RDWR
|O_LARGEFILE
, 0);
1215 err
= PTR_ERR(victim
);
1219 mapping
= victim
->f_mapping
;
1221 spin_lock(&swap_lock
);
1222 for (type
= swap_list
.head
; type
>= 0; type
= swap_info
[type
].next
) {
1223 p
= swap_info
+ type
;
1224 if ((p
->flags
& SWP_ACTIVE
) == SWP_ACTIVE
) {
1225 if (p
->swap_file
->f_mapping
== mapping
)
1232 spin_unlock(&swap_lock
);
1235 if (!security_vm_enough_memory(p
->pages
))
1236 vm_unacct_memory(p
->pages
);
1239 spin_unlock(&swap_lock
);
1243 swap_list
.head
= p
->next
;
1245 swap_info
[prev
].next
= p
->next
;
1247 if (type
== swap_list
.next
) {
1248 /* just pick something that's safe... */
1249 swap_list
.next
= swap_list
.head
;
1252 for (i
= p
->next
; i
>= 0; i
= swap_info
[i
].next
)
1253 swap_info
[i
].prio
= p
->prio
--;
1256 nr_swap_pages
-= p
->pages
;
1257 total_swap_pages
-= p
->pages
;
1258 p
->flags
&= ~SWP_WRITEOK
;
1259 spin_unlock(&swap_lock
);
1261 current
->flags
|= PF_SWAPOFF
;
1262 err
= try_to_unuse(type
);
1263 current
->flags
&= ~PF_SWAPOFF
;
1266 /* re-insert swap space back into swap_list */
1267 spin_lock(&swap_lock
);
1269 p
->prio
= --least_priority
;
1271 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
].next
) {
1272 if (p
->prio
>= swap_info
[i
].prio
)
1278 swap_list
.head
= swap_list
.next
= p
- swap_info
;
1280 swap_info
[prev
].next
= p
- swap_info
;
1281 nr_swap_pages
+= p
->pages
;
1282 total_swap_pages
+= p
->pages
;
1283 p
->flags
|= SWP_WRITEOK
;
1284 spin_unlock(&swap_lock
);
1288 /* wait for any unplug function to finish */
1289 down_write(&swap_unplug_sem
);
1290 up_write(&swap_unplug_sem
);
1292 destroy_swap_extents(p
);
1293 mutex_lock(&swapon_mutex
);
1294 spin_lock(&swap_lock
);
1297 /* wait for anyone still in scan_swap_map */
1298 p
->highest_bit
= 0; /* cuts scans short */
1299 while (p
->flags
>= SWP_SCANNING
) {
1300 spin_unlock(&swap_lock
);
1301 schedule_timeout_uninterruptible(1);
1302 spin_lock(&swap_lock
);
1305 swap_file
= p
->swap_file
;
1306 p
->swap_file
= NULL
;
1308 swap_map
= p
->swap_map
;
1311 spin_unlock(&swap_lock
);
1312 mutex_unlock(&swapon_mutex
);
1314 inode
= mapping
->host
;
1315 if (S_ISBLK(inode
->i_mode
)) {
1316 struct block_device
*bdev
= I_BDEV(inode
);
1317 set_blocksize(bdev
, p
->old_block_size
);
1320 mutex_lock(&inode
->i_mutex
);
1321 inode
->i_flags
&= ~S_SWAPFILE
;
1322 mutex_unlock(&inode
->i_mutex
);
1324 filp_close(swap_file
, NULL
);
1328 filp_close(victim
, NULL
);
1333 #ifdef CONFIG_PROC_FS
1335 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
1337 struct swap_info_struct
*ptr
= swap_info
;
1341 mutex_lock(&swapon_mutex
);
1344 return SEQ_START_TOKEN
;
1346 for (i
= 0; i
< nr_swapfiles
; i
++, ptr
++) {
1347 if (!(ptr
->flags
& SWP_USED
) || !ptr
->swap_map
)
1356 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
1358 struct swap_info_struct
*ptr
;
1359 struct swap_info_struct
*endptr
= swap_info
+ nr_swapfiles
;
1361 if (v
== SEQ_START_TOKEN
)
1368 for (; ptr
< endptr
; ptr
++) {
1369 if (!(ptr
->flags
& SWP_USED
) || !ptr
->swap_map
)
1378 static void swap_stop(struct seq_file
*swap
, void *v
)
1380 mutex_unlock(&swapon_mutex
);
1383 static int swap_show(struct seq_file
*swap
, void *v
)
1385 struct swap_info_struct
*ptr
= v
;
1389 if (ptr
== SEQ_START_TOKEN
) {
1390 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1394 file
= ptr
->swap_file
;
1395 len
= seq_path(swap
, &file
->f_path
, " \t\n\\");
1396 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
1397 len
< 40 ? 40 - len
: 1, " ",
1398 S_ISBLK(file
->f_path
.dentry
->d_inode
->i_mode
) ?
1399 "partition" : "file\t",
1400 ptr
->pages
<< (PAGE_SHIFT
- 10),
1401 ptr
->inuse_pages
<< (PAGE_SHIFT
- 10),
1406 static const struct seq_operations swaps_op
= {
1407 .start
= swap_start
,
1413 static int swaps_open(struct inode
*inode
, struct file
*file
)
1415 return seq_open(file
, &swaps_op
);
1418 static const struct file_operations proc_swaps_operations
= {
1421 .llseek
= seq_lseek
,
1422 .release
= seq_release
,
1425 static int __init
procswaps_init(void)
1427 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
1430 __initcall(procswaps_init
);
1431 #endif /* CONFIG_PROC_FS */
1433 #ifdef MAX_SWAPFILES_CHECK
1434 static int __init
max_swapfiles_check(void)
1436 MAX_SWAPFILES_CHECK();
1439 late_initcall(max_swapfiles_check
);
1443 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1445 * The swapon system call
1447 asmlinkage
long sys_swapon(const char __user
* specialfile
, int swap_flags
)
1449 struct swap_info_struct
* p
;
1451 struct block_device
*bdev
= NULL
;
1452 struct file
*swap_file
= NULL
;
1453 struct address_space
*mapping
;
1457 union swap_header
*swap_header
= NULL
;
1458 int swap_header_version
;
1459 unsigned int nr_good_pages
= 0;
1462 unsigned long maxpages
= 1;
1464 unsigned short *swap_map
= NULL
;
1465 struct page
*page
= NULL
;
1466 struct inode
*inode
= NULL
;
1469 if (!capable(CAP_SYS_ADMIN
))
1471 spin_lock(&swap_lock
);
1473 for (type
= 0 ; type
< nr_swapfiles
; type
++,p
++)
1474 if (!(p
->flags
& SWP_USED
))
1477 if (type
>= MAX_SWAPFILES
) {
1478 spin_unlock(&swap_lock
);
1481 if (type
>= nr_swapfiles
)
1482 nr_swapfiles
= type
+1;
1483 memset(p
, 0, sizeof(*p
));
1484 INIT_LIST_HEAD(&p
->extent_list
);
1485 p
->flags
= SWP_USED
;
1487 spin_unlock(&swap_lock
);
1488 name
= getname(specialfile
);
1489 error
= PTR_ERR(name
);
1494 swap_file
= filp_open(name
, O_RDWR
|O_LARGEFILE
, 0);
1495 error
= PTR_ERR(swap_file
);
1496 if (IS_ERR(swap_file
)) {
1501 p
->swap_file
= swap_file
;
1502 mapping
= swap_file
->f_mapping
;
1503 inode
= mapping
->host
;
1506 for (i
= 0; i
< nr_swapfiles
; i
++) {
1507 struct swap_info_struct
*q
= &swap_info
[i
];
1509 if (i
== type
|| !q
->swap_file
)
1511 if (mapping
== q
->swap_file
->f_mapping
)
1516 if (S_ISBLK(inode
->i_mode
)) {
1517 bdev
= I_BDEV(inode
);
1518 error
= bd_claim(bdev
, sys_swapon
);
1524 p
->old_block_size
= block_size(bdev
);
1525 error
= set_blocksize(bdev
, PAGE_SIZE
);
1529 } else if (S_ISREG(inode
->i_mode
)) {
1530 p
->bdev
= inode
->i_sb
->s_bdev
;
1531 mutex_lock(&inode
->i_mutex
);
1533 if (IS_SWAPFILE(inode
)) {
1541 swapfilesize
= i_size_read(inode
) >> PAGE_SHIFT
;
1544 * Read the swap header.
1546 if (!mapping
->a_ops
->readpage
) {
1550 page
= read_mapping_page(mapping
, 0, swap_file
);
1552 error
= PTR_ERR(page
);
1556 swap_header
= page_address(page
);
1558 if (!memcmp("SWAP-SPACE",swap_header
->magic
.magic
,10))
1559 swap_header_version
= 1;
1560 else if (!memcmp("SWAPSPACE2",swap_header
->magic
.magic
,10))
1561 swap_header_version
= 2;
1563 printk(KERN_ERR
"Unable to find swap-space signature\n");
1568 switch (swap_header_version
) {
1570 printk(KERN_ERR
"version 0 swap is no longer supported. "
1571 "Use mkswap -v1 %s\n", name
);
1575 /* swap partition endianess hack... */
1576 if (swab32(swap_header
->info
.version
) == 1) {
1577 swab32s(&swap_header
->info
.version
);
1578 swab32s(&swap_header
->info
.last_page
);
1579 swab32s(&swap_header
->info
.nr_badpages
);
1580 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
1581 swab32s(&swap_header
->info
.badpages
[i
]);
1583 /* Check the swap header's sub-version and the size of
1584 the swap file and bad block lists */
1585 if (swap_header
->info
.version
!= 1) {
1587 "Unable to handle swap header version %d\n",
1588 swap_header
->info
.version
);
1594 p
->cluster_next
= 1;
1597 * Find out how many pages are allowed for a single swap
1598 * device. There are two limiting factors: 1) the number of
1599 * bits for the swap offset in the swp_entry_t type and
1600 * 2) the number of bits in the a swap pte as defined by
1601 * the different architectures. In order to find the
1602 * largest possible bit mask a swap entry with swap type 0
1603 * and swap offset ~0UL is created, encoded to a swap pte,
1604 * decoded to a swp_entry_t again and finally the swap
1605 * offset is extracted. This will mask all the bits from
1606 * the initial ~0UL mask that can't be encoded in either
1607 * the swp_entry_t or the architecture definition of a
1610 maxpages
= swp_offset(pte_to_swp_entry(swp_entry_to_pte(swp_entry(0,~0UL)))) - 1;
1611 if (maxpages
> swap_header
->info
.last_page
)
1612 maxpages
= swap_header
->info
.last_page
;
1613 p
->highest_bit
= maxpages
- 1;
1618 if (swapfilesize
&& maxpages
> swapfilesize
) {
1620 "Swap area shorter than signature indicates\n");
1623 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
1625 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
1628 /* OK, set up the swap map and apply the bad block list */
1629 swap_map
= vmalloc(maxpages
* sizeof(short));
1636 memset(swap_map
, 0, maxpages
* sizeof(short));
1637 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
1638 int page_nr
= swap_header
->info
.badpages
[i
];
1639 if (page_nr
<= 0 || page_nr
>= swap_header
->info
.last_page
)
1642 swap_map
[page_nr
] = SWAP_MAP_BAD
;
1644 nr_good_pages
= swap_header
->info
.last_page
-
1645 swap_header
->info
.nr_badpages
-
1646 1 /* header page */;
1651 if (nr_good_pages
) {
1652 swap_map
[0] = SWAP_MAP_BAD
;
1654 p
->pages
= nr_good_pages
;
1655 nr_extents
= setup_swap_extents(p
, &span
);
1656 if (nr_extents
< 0) {
1660 nr_good_pages
= p
->pages
;
1662 if (!nr_good_pages
) {
1663 printk(KERN_WARNING
"Empty swap-file\n");
1668 mutex_lock(&swapon_mutex
);
1669 spin_lock(&swap_lock
);
1670 if (swap_flags
& SWAP_FLAG_PREFER
)
1672 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
1674 p
->prio
= --least_priority
;
1675 p
->swap_map
= swap_map
;
1676 p
->flags
= SWP_ACTIVE
;
1677 nr_swap_pages
+= nr_good_pages
;
1678 total_swap_pages
+= nr_good_pages
;
1680 printk(KERN_INFO
"Adding %uk swap on %s. "
1681 "Priority:%d extents:%d across:%lluk\n",
1682 nr_good_pages
<<(PAGE_SHIFT
-10), name
, p
->prio
,
1683 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10));
1685 /* insert swap space into swap_list: */
1687 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
].next
) {
1688 if (p
->prio
>= swap_info
[i
].prio
) {
1695 swap_list
.head
= swap_list
.next
= p
- swap_info
;
1697 swap_info
[prev
].next
= p
- swap_info
;
1699 spin_unlock(&swap_lock
);
1700 mutex_unlock(&swapon_mutex
);
1705 set_blocksize(bdev
, p
->old_block_size
);
1708 destroy_swap_extents(p
);
1710 spin_lock(&swap_lock
);
1711 p
->swap_file
= NULL
;
1713 spin_unlock(&swap_lock
);
1716 filp_close(swap_file
, NULL
);
1718 if (page
&& !IS_ERR(page
)) {
1720 page_cache_release(page
);
1726 inode
->i_flags
|= S_SWAPFILE
;
1727 mutex_unlock(&inode
->i_mutex
);
1732 void si_swapinfo(struct sysinfo
*val
)
1735 unsigned long nr_to_be_unused
= 0;
1737 spin_lock(&swap_lock
);
1738 for (i
= 0; i
< nr_swapfiles
; i
++) {
1739 if (!(swap_info
[i
].flags
& SWP_USED
) ||
1740 (swap_info
[i
].flags
& SWP_WRITEOK
))
1742 nr_to_be_unused
+= swap_info
[i
].inuse_pages
;
1744 val
->freeswap
= nr_swap_pages
+ nr_to_be_unused
;
1745 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
1746 spin_unlock(&swap_lock
);
1750 * Verify that a swap entry is valid and increment its swap map count.
1752 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1753 * "permanent", but will be reclaimed by the next swapoff.
1755 int swap_duplicate(swp_entry_t entry
)
1757 struct swap_info_struct
* p
;
1758 unsigned long offset
, type
;
1761 if (is_migration_entry(entry
))
1764 type
= swp_type(entry
);
1765 if (type
>= nr_swapfiles
)
1767 p
= type
+ swap_info
;
1768 offset
= swp_offset(entry
);
1770 spin_lock(&swap_lock
);
1771 if (offset
< p
->max
&& p
->swap_map
[offset
]) {
1772 if (p
->swap_map
[offset
] < SWAP_MAP_MAX
- 1) {
1773 p
->swap_map
[offset
]++;
1775 } else if (p
->swap_map
[offset
] <= SWAP_MAP_MAX
) {
1776 if (swap_overflow
++ < 5)
1777 printk(KERN_WARNING
"swap_dup: swap entry overflow\n");
1778 p
->swap_map
[offset
] = SWAP_MAP_MAX
;
1782 spin_unlock(&swap_lock
);
1787 printk(KERN_ERR
"swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
1791 struct swap_info_struct
*
1792 get_swap_info_struct(unsigned type
)
1794 return &swap_info
[type
];
1798 * swap_lock prevents swap_map being freed. Don't grab an extra
1799 * reference on the swaphandle, it doesn't matter if it becomes unused.
1801 int valid_swaphandles(swp_entry_t entry
, unsigned long *offset
)
1803 struct swap_info_struct
*si
;
1804 int our_page_cluster
= page_cluster
;
1805 pgoff_t target
, toff
;
1809 if (!our_page_cluster
) /* no readahead */
1812 si
= &swap_info
[swp_type(entry
)];
1813 target
= swp_offset(entry
);
1814 base
= (target
>> our_page_cluster
) << our_page_cluster
;
1815 end
= base
+ (1 << our_page_cluster
);
1816 if (!base
) /* first page is swap header */
1819 spin_lock(&swap_lock
);
1820 if (end
> si
->max
) /* don't go beyond end of map */
1823 /* Count contiguous allocated slots above our target */
1824 for (toff
= target
; ++toff
< end
; nr_pages
++) {
1825 /* Don't read in free or bad pages */
1826 if (!si
->swap_map
[toff
])
1828 if (si
->swap_map
[toff
] == SWAP_MAP_BAD
)
1831 /* Count contiguous allocated slots below our target */
1832 for (toff
= target
; --toff
>= base
; nr_pages
++) {
1833 /* Don't read in free or bad pages */
1834 if (!si
->swap_map
[toff
])
1836 if (si
->swap_map
[toff
] == SWAP_MAP_BAD
)
1839 spin_unlock(&swap_lock
);
1842 * Indicate starting offset, and return number of pages to get:
1843 * if only 1, say 0, since there's then no readahead to be done.
1846 return nr_pages
? ++nr_pages
: 0;