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/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-locking.h>
25 #include <linux/security.h>
27 #include <asm/pgtable.h>
28 #include <linux/swapops.h>
30 spinlock_t swaplock
= SPIN_LOCK_UNLOCKED
;
31 unsigned int nr_swapfiles
;
33 static int swap_overflow
;
35 EXPORT_SYMBOL(total_swap_pages
);
37 static const char Bad_file
[] = "Bad swap file entry ";
38 static const char Unused_file
[] = "Unused swap file entry ";
39 static const char Bad_offset
[] = "Bad swap offset entry ";
40 static const char Unused_offset
[] = "Unused swap offset entry ";
42 struct swap_list_t swap_list
= {-1, -1};
44 struct swap_info_struct swap_info
[MAX_SWAPFILES
];
46 #define SWAPFILE_CLUSTER 256
48 static inline int scan_swap_map(struct swap_info_struct
*si
)
52 * We try to cluster swap pages by allocating them
53 * sequentially in swap. Once we've allocated
54 * SWAPFILE_CLUSTER pages this way, however, we resort to
55 * first-free allocation, starting a new cluster. This
56 * prevents us from scattering swap pages all over the entire
57 * swap partition, so that we reduce overall disk seek times
58 * between swap pages. -- sct */
60 while (si
->cluster_next
<= si
->highest_bit
) {
61 offset
= si
->cluster_next
++;
62 if (si
->swap_map
[offset
])
68 si
->cluster_nr
= SWAPFILE_CLUSTER
;
70 /* try to find an empty (even not aligned) cluster. */
71 offset
= si
->lowest_bit
;
73 if (offset
+SWAPFILE_CLUSTER
-1 <= si
->highest_bit
)
76 for (nr
= offset
; nr
< offset
+SWAPFILE_CLUSTER
; nr
++)
80 goto check_next_cluster
;
82 /* We found a completly empty cluster, so start
87 /* No luck, so now go finegrined as usual. -Andrea */
88 for (offset
= si
->lowest_bit
; offset
<= si
->highest_bit
; offset
++) {
89 if (si
->swap_map
[offset
])
91 si
->lowest_bit
= offset
+1;
93 if (offset
== si
->lowest_bit
)
95 if (offset
== si
->highest_bit
)
97 if (si
->lowest_bit
> si
->highest_bit
) {
98 si
->lowest_bit
= si
->max
;
101 si
->swap_map
[offset
] = 1;
104 si
->cluster_next
= offset
+1;
107 si
->lowest_bit
= si
->max
;
112 swp_entry_t
get_swap_page(void)
114 struct swap_info_struct
* p
;
115 unsigned long offset
;
117 int type
, wrapped
= 0;
119 entry
.val
= 0; /* Out of memory */
121 type
= swap_list
.next
;
124 if (nr_swap_pages
<= 0)
128 p
= &swap_info
[type
];
129 if ((p
->flags
& SWP_ACTIVE
) == SWP_ACTIVE
) {
131 offset
= scan_swap_map(p
);
132 swap_device_unlock(p
);
134 entry
= swp_entry(type
,offset
);
135 type
= swap_info
[type
].next
;
137 p
->prio
!= swap_info
[type
].prio
) {
138 swap_list
.next
= swap_list
.head
;
140 swap_list
.next
= type
;
147 if (type
< 0 || p
->prio
!= swap_info
[type
].prio
) {
148 type
= swap_list
.head
;
153 goto out
; /* out of swap space */
160 static struct swap_info_struct
* swap_info_get(swp_entry_t entry
)
162 struct swap_info_struct
* p
;
163 unsigned long offset
, type
;
167 type
= swp_type(entry
);
168 if (type
>= nr_swapfiles
)
170 p
= & swap_info
[type
];
171 if (!(p
->flags
& SWP_USED
))
173 offset
= swp_offset(entry
);
174 if (offset
>= p
->max
)
176 if (!p
->swap_map
[offset
])
179 if (p
->prio
> swap_info
[swap_list
.next
].prio
)
180 swap_list
.next
= type
;
185 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_offset
, entry
.val
);
188 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_offset
, entry
.val
);
191 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_file
, entry
.val
);
194 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_file
, entry
.val
);
199 static void swap_info_put(struct swap_info_struct
* p
)
201 swap_device_unlock(p
);
205 static int swap_entry_free(struct swap_info_struct
*p
, unsigned long offset
)
207 int count
= p
->swap_map
[offset
];
209 if (count
< SWAP_MAP_MAX
) {
211 p
->swap_map
[offset
] = count
;
213 if (offset
< p
->lowest_bit
)
214 p
->lowest_bit
= offset
;
215 if (offset
> p
->highest_bit
)
216 p
->highest_bit
= offset
;
225 * Caller has made sure that the swapdevice corresponding to entry
226 * is still around or has not been recycled.
228 void swap_free(swp_entry_t entry
)
230 struct swap_info_struct
* p
;
232 p
= swap_info_get(entry
);
234 swap_entry_free(p
, swp_offset(entry
));
240 * Check if we're the only user of a swap page,
241 * when the page is locked.
243 static int exclusive_swap_page(struct page
*page
)
246 struct swap_info_struct
* p
;
249 entry
.val
= page
->index
;
250 p
= swap_info_get(entry
);
252 /* Is the only swap cache user the cache itself? */
253 if (p
->swap_map
[swp_offset(entry
)] == 1) {
254 /* Recheck the page count with the pagecache lock held.. */
255 spin_lock(&swapper_space
.page_lock
);
256 if (page_count(page
) - !!PagePrivate(page
) == 2)
258 spin_unlock(&swapper_space
.page_lock
);
266 * We can use this swap cache entry directly
267 * if there are no other references to it.
269 * Here "exclusive_swap_page()" does the real
270 * work, but we opportunistically check whether
271 * we need to get all the locks first..
273 int can_share_swap_page(struct page
*page
)
277 if (!PageLocked(page
))
279 switch (page_count(page
)) {
281 if (!PagePrivate(page
))
285 if (!PageSwapCache(page
))
287 retval
= exclusive_swap_page(page
);
290 if (PageReserved(page
))
298 * Work out if there are any other processes sharing this
299 * swap cache page. Free it if you can. Return success.
301 int remove_exclusive_swap_page(struct page
*page
)
304 struct swap_info_struct
* p
;
307 BUG_ON(PagePrivate(page
));
308 BUG_ON(!PageLocked(page
));
310 if (!PageSwapCache(page
))
312 if (PageWriteback(page
))
314 if (page_count(page
) != 2) /* 2: us + cache */
317 entry
.val
= page
->index
;
318 p
= swap_info_get(entry
);
322 /* Is the only swap cache user the cache itself? */
324 if (p
->swap_map
[swp_offset(entry
)] == 1) {
325 /* Recheck the page count with the pagecache lock held.. */
326 spin_lock(&swapper_space
.page_lock
);
327 if ((page_count(page
) == 2) && !PageWriteback(page
)) {
328 __delete_from_swap_cache(page
);
332 spin_unlock(&swapper_space
.page_lock
);
338 page_cache_release(page
);
345 * Free the swap entry like above, but also try to
346 * free the page cache entry if it is the last user.
348 void free_swap_and_cache(swp_entry_t entry
)
350 struct swap_info_struct
* p
;
351 struct page
*page
= NULL
;
353 p
= swap_info_get(entry
);
355 if (swap_entry_free(p
, swp_offset(entry
)) == 1)
356 page
= find_trylock_page(&swapper_space
, entry
.val
);
362 BUG_ON(PagePrivate(page
));
363 page_cache_get(page
);
364 one_user
= (page_count(page
) == 2);
365 /* Only cache user (+us), or swap space full? Free it! */
366 if (!PageWriteback(page
) && (one_user
|| vm_swap_full())) {
367 delete_from_swap_cache(page
);
371 page_cache_release(page
);
376 * The swap entry has been read in advance, and we return 1 to indicate
377 * that the page has been used or is no longer needed.
379 * Always set the resulting pte to be nowrite (the same as COW pages
380 * after one process has exited). We don't know just how many PTEs will
381 * share this swap entry, so be cautious and let do_wp_page work out
382 * what to do if a write is requested later.
384 /* vma->vm_mm->page_table_lock is held */
386 unuse_pte(struct vm_area_struct
*vma
, unsigned long address
, pte_t
*dir
,
387 swp_entry_t entry
, struct page
*page
, struct pte_chain
**pte_chainp
)
391 set_pte(dir
, pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
392 *pte_chainp
= page_add_rmap(page
, dir
, *pte_chainp
);
396 /* vma->vm_mm->page_table_lock is held */
397 static int unuse_pmd(struct vm_area_struct
* vma
, pmd_t
*dir
,
398 unsigned long address
, unsigned long size
, unsigned long offset
,
399 swp_entry_t entry
, struct page
*page
, struct pte_chain
**pte_chainp
)
403 pte_t swp_pte
= swp_entry_to_pte(entry
);
412 pte
= pte_offset_map(dir
, address
);
413 offset
+= address
& PMD_MASK
;
414 address
&= ~PMD_MASK
;
415 end
= address
+ size
;
420 * swapoff spends a _lot_ of time in this loop!
421 * Test inline before going to call unuse_pte.
423 if (unlikely(pte_same(*pte
, swp_pte
))) {
424 unuse_pte(vma
, offset
+ address
, pte
,
425 entry
, page
, pte_chainp
);
429 address
+= PAGE_SIZE
;
431 } while (address
&& (address
< end
));
436 /* vma->vm_mm->page_table_lock is held */
437 static int unuse_pgd(struct vm_area_struct
* vma
, pgd_t
*dir
,
438 unsigned long address
, unsigned long size
,
439 swp_entry_t entry
, struct page
*page
, struct pte_chain
**pte_chainp
)
442 unsigned long offset
, end
;
451 pmd
= pmd_offset(dir
, address
);
452 offset
= address
& PGDIR_MASK
;
453 address
&= ~PGDIR_MASK
;
454 end
= address
+ size
;
455 if (end
> PGDIR_SIZE
)
460 if (unuse_pmd(vma
, pmd
, address
, end
- address
,
461 offset
, entry
, page
, pte_chainp
))
463 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
465 } while (address
&& (address
< end
));
469 /* vma->vm_mm->page_table_lock is held */
470 static int unuse_vma(struct vm_area_struct
* vma
, pgd_t
*pgdir
,
471 swp_entry_t entry
, struct page
*page
, struct pte_chain
**pte_chainp
)
473 unsigned long start
= vma
->vm_start
, end
= vma
->vm_end
;
478 if (unuse_pgd(vma
, pgdir
, start
, end
- start
,
479 entry
, page
, pte_chainp
))
481 start
= (start
+ PGDIR_SIZE
) & PGDIR_MASK
;
483 } while (start
&& (start
< end
));
487 static int unuse_process(struct mm_struct
* mm
,
488 swp_entry_t entry
, struct page
* page
)
490 struct vm_area_struct
* vma
;
491 struct pte_chain
*pte_chain
;
493 pte_chain
= pte_chain_alloc(GFP_KERNEL
);
498 * Go through process' page directory.
500 spin_lock(&mm
->page_table_lock
);
501 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
502 pgd_t
* pgd
= pgd_offset(mm
, vma
->vm_start
);
503 if (unuse_vma(vma
, pgd
, entry
, page
, &pte_chain
))
506 spin_unlock(&mm
->page_table_lock
);
507 pte_chain_free(pte_chain
);
512 * Scan swap_map from current position to next entry still in use.
513 * Recycle to start on reaching the end, returning 0 when empty.
515 static int find_next_to_unuse(struct swap_info_struct
*si
, int prev
)
522 * No need for swap_device_lock(si) here: we're just looking
523 * for whether an entry is in use, not modifying it; false
524 * hits are okay, and sys_swapoff() has already prevented new
525 * allocations from this area (while holding swap_list_lock()).
534 * No entries in use at top of swap_map,
535 * loop back to start and recheck there.
541 count
= si
->swap_map
[i
];
542 if (count
&& count
!= SWAP_MAP_BAD
)
549 * We completely avoid races by reading each swap page in advance,
550 * and then search for the process using it. All the necessary
551 * page table adjustments can then be made atomically.
553 static int try_to_unuse(unsigned int type
)
555 struct swap_info_struct
* si
= &swap_info
[type
];
556 struct mm_struct
*start_mm
;
557 unsigned short *swap_map
;
558 unsigned short swcount
;
563 int reset_overflow
= 0;
567 * When searching mms for an entry, a good strategy is to
568 * start at the first mm we freed the previous entry from
569 * (though actually we don't notice whether we or coincidence
570 * freed the entry). Initialize this start_mm with a hold.
572 * A simpler strategy would be to start at the last mm we
573 * freed the previous entry from; but that would take less
574 * advantage of mmlist ordering (now preserved by swap_out()),
575 * which clusters forked address spaces together, most recent
576 * child immediately after parent. If we race with dup_mmap(),
577 * we very much want to resolve parent before child, otherwise
578 * we may miss some entries: using last mm would invert that.
581 atomic_inc(&init_mm
.mm_users
);
584 * Keep on scanning until all entries have gone. Usually,
585 * one pass through swap_map is enough, but not necessarily:
586 * mmput() removes mm from mmlist before exit_mmap() and its
587 * zap_page_range(). That's not too bad, those entries are
588 * on their way out, and handled faster there than here.
589 * do_munmap() behaves similarly, taking the range out of mm's
590 * vma list before zap_page_range(). But unfortunately, when
591 * unmapping a part of a vma, it takes the whole out first,
592 * then reinserts what's left after (might even reschedule if
593 * open() method called) - so swap entries may be invisible
594 * to swapoff for a while, then reappear - but that is rare.
596 while ((i
= find_next_to_unuse(si
, i
))) {
597 if (signal_pending(current
)) {
603 * Get a page for the entry, using the existing swap
604 * cache page if there is one. Otherwise, get a clean
605 * page and read the swap into it.
607 swap_map
= &si
->swap_map
[i
];
608 entry
= swp_entry(type
, i
);
609 page
= read_swap_cache_async(entry
);
612 * Either swap_duplicate() failed because entry
613 * has been freed independently, and will not be
614 * reused since sys_swapoff() already disabled
615 * allocation from here, or alloc_page() failed.
624 * Don't hold on to start_mm if it looks like exiting.
626 if (atomic_read(&start_mm
->mm_users
) == 1) {
629 atomic_inc(&init_mm
.mm_users
);
633 * Wait for and lock page. When do_swap_page races with
634 * try_to_unuse, do_swap_page can handle the fault much
635 * faster than try_to_unuse can locate the entry. This
636 * apparently redundant "wait_on_page_locked" lets try_to_unuse
637 * defer to do_swap_page in such a case - in some tests,
638 * do_swap_page and try_to_unuse repeatedly compete.
640 wait_on_page_locked(page
);
641 wait_on_page_writeback(page
);
643 wait_on_page_writeback(page
);
646 * Remove all references to entry, without blocking.
647 * Whenever we reach init_mm, there's no address space
648 * to search, but use it as a reminder to search shmem.
653 if (start_mm
== &init_mm
)
654 shmem
= shmem_unuse(entry
, page
);
656 retval
= unuse_process(start_mm
, entry
, page
);
659 int set_start_mm
= (*swap_map
>= swcount
);
660 struct list_head
*p
= &start_mm
->mmlist
;
661 struct mm_struct
*new_start_mm
= start_mm
;
662 struct mm_struct
*prev_mm
= start_mm
;
663 struct mm_struct
*mm
;
665 atomic_inc(&new_start_mm
->mm_users
);
666 atomic_inc(&prev_mm
->mm_users
);
667 spin_lock(&mmlist_lock
);
668 while (*swap_map
> 1 && !retval
&&
669 (p
= p
->next
) != &start_mm
->mmlist
) {
670 mm
= list_entry(p
, struct mm_struct
, mmlist
);
671 atomic_inc(&mm
->mm_users
);
672 spin_unlock(&mmlist_lock
);
681 else if (mm
== &init_mm
) {
683 shmem
= shmem_unuse(entry
, page
);
685 retval
= unuse_process(mm
, entry
, page
);
686 if (set_start_mm
&& *swap_map
< swcount
) {
688 atomic_inc(&mm
->mm_users
);
692 spin_lock(&mmlist_lock
);
694 spin_unlock(&mmlist_lock
);
697 start_mm
= new_start_mm
;
701 page_cache_release(page
);
706 * How could swap count reach 0x7fff when the maximum
707 * pid is 0x7fff, and there's no way to repeat a swap
708 * page within an mm (except in shmem, where it's the
709 * shared object which takes the reference count)?
710 * We believe SWAP_MAP_MAX cannot occur in Linux 2.4.
712 * If that's wrong, then we should worry more about
713 * exit_mmap() and do_munmap() cases described above:
714 * we might be resetting SWAP_MAP_MAX too early here.
715 * We know "Undead"s can happen, they're okay, so don't
716 * report them; but do report if we reset SWAP_MAP_MAX.
718 if (*swap_map
== SWAP_MAP_MAX
) {
719 swap_device_lock(si
);
721 swap_device_unlock(si
);
726 * If a reference remains (rare), we would like to leave
727 * the page in the swap cache; but try_to_unmap could
728 * then re-duplicate the entry once we drop page lock,
729 * so we might loop indefinitely; also, that page could
730 * not be swapped out to other storage meanwhile. So:
731 * delete from cache even if there's another reference,
732 * after ensuring that the data has been saved to disk -
733 * since if the reference remains (rarer), it will be
734 * read from disk into another page. Splitting into two
735 * pages would be incorrect if swap supported "shared
736 * private" pages, but they are handled by tmpfs files.
738 * Note shmem_unuse already deleted a swappage from
739 * the swap cache, unless the move to filepage failed:
740 * in which case it left swappage in cache, lowered its
741 * swap count to pass quickly through the loops above,
742 * and now we must reincrement count to try again later.
744 if ((*swap_map
> 1) && PageDirty(page
) && PageSwapCache(page
)) {
745 struct writeback_control wbc
= {
746 .sync_mode
= WB_SYNC_NONE
,
749 swap_writepage(page
, &wbc
);
751 wait_on_page_writeback(page
);
753 if (PageSwapCache(page
)) {
755 swap_duplicate(entry
);
757 delete_from_swap_cache(page
);
761 * So we could skip searching mms once swap count went
762 * to 1, we did not mark any present ptes as dirty: must
763 * mark page dirty so shrink_list will preserve it.
767 page_cache_release(page
);
770 * Make sure that we aren't completely killing
771 * interactive performance.
777 if (reset_overflow
) {
778 printk(KERN_WARNING
"swapoff: cleared swap entry overflow\n");
785 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
786 * corresponds to page offset `offset'.
788 sector_t
map_swap_page(struct swap_info_struct
*sis
, pgoff_t offset
)
790 struct swap_extent
*se
= sis
->curr_swap_extent
;
791 struct swap_extent
*start_se
= se
;
794 struct list_head
*lh
;
796 if (se
->start_page
<= offset
&&
797 offset
< (se
->start_page
+ se
->nr_pages
)) {
798 return se
->start_block
+ (offset
- se
->start_page
);
801 if (lh
== &sis
->extent_list
)
803 se
= list_entry(lh
, struct swap_extent
, list
);
804 sis
->curr_swap_extent
= se
;
805 BUG_ON(se
== start_se
); /* It *must* be present */
810 * Free all of a swapdev's extent information
812 static void destroy_swap_extents(struct swap_info_struct
*sis
)
814 while (!list_empty(&sis
->extent_list
)) {
815 struct swap_extent
*se
;
817 se
= list_entry(sis
->extent_list
.next
,
818 struct swap_extent
, list
);
826 * Add a block range (and the corresponding page range) into this swapdev's
827 * extent list. The extent list is kept sorted in block order.
829 * This function rather assumes that it is called in ascending sector_t order.
830 * It doesn't look for extent coalescing opportunities.
833 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
834 unsigned long nr_pages
, sector_t start_block
)
836 struct swap_extent
*se
;
837 struct swap_extent
*new_se
;
838 struct list_head
*lh
;
840 lh
= sis
->extent_list
.next
; /* The highest-addressed block */
841 while (lh
!= &sis
->extent_list
) {
842 se
= list_entry(lh
, struct swap_extent
, list
);
843 if (se
->start_block
+ se
->nr_pages
== start_block
) {
845 se
->nr_pages
+= nr_pages
;
852 * No merge. Insert a new extent, preserving ordering.
854 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
857 new_se
->start_page
= start_page
;
858 new_se
->nr_pages
= nr_pages
;
859 new_se
->start_block
= start_block
;
861 lh
= sis
->extent_list
.prev
; /* The lowest block */
862 while (lh
!= &sis
->extent_list
) {
863 se
= list_entry(lh
, struct swap_extent
, list
);
864 if (se
->start_block
> start_block
)
868 list_add_tail(&new_se
->list
, lh
);
874 * A `swap extent' is a simple thing which maps a contiguous range of pages
875 * onto a contiguous range of disk blocks. An ordered list of swap extents
876 * is built at swapon time and is then used at swap_writepage/swap_readpage
877 * time for locating where on disk a page belongs.
879 * If the swapfile is an S_ISBLK block device, a single extent is installed.
880 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
881 * swap files identically.
883 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
884 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
885 * swapfiles are handled *identically* after swapon time.
887 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
888 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
889 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
890 * requirements, they are simply tossed out - we will never use those blocks
893 * For S_ISREG swapfiles we hold i_sem across the life of the swapon. This
894 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
895 * which will scribble on the fs.
897 * The amount of disk space which a single swap extent represents varies.
898 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
899 * extents in the list. To avoid much list walking, we cache the previous
900 * search location in `curr_swap_extent', and start new searches from there.
901 * This is extremely effective. The average number of iterations in
902 * map_swap_page() has been measured at about 0.3 per page. - akpm.
904 static int setup_swap_extents(struct swap_info_struct
*sis
)
907 unsigned blocks_per_page
;
908 unsigned long page_no
;
910 sector_t probe_block
;
914 inode
= sis
->swap_file
->f_dentry
->d_inode
;
915 if (S_ISBLK(inode
->i_mode
)) {
916 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
920 blkbits
= inode
->i_blkbits
;
921 blocks_per_page
= PAGE_SIZE
>> blkbits
;
924 * Map all the blocks into the extent list. This code doesn't try
929 last_block
= i_size_read(inode
) >> blkbits
;
930 while ((probe_block
+ blocks_per_page
) <= last_block
&&
931 page_no
< sis
->max
) {
932 unsigned block_in_page
;
933 sector_t first_block
;
935 first_block
= bmap(inode
, probe_block
);
936 if (first_block
== 0)
940 * It must be PAGE_SIZE aligned on-disk
942 if (first_block
& (blocks_per_page
- 1)) {
947 for (block_in_page
= 1; block_in_page
< blocks_per_page
;
951 block
= bmap(inode
, probe_block
+ block_in_page
);
954 if (block
!= first_block
+ block_in_page
) {
962 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
964 ret
= add_swap_extent(sis
, page_no
, 1,
965 first_block
>> (PAGE_SHIFT
- blkbits
));
969 probe_block
+= blocks_per_page
;
977 sis
->highest_bit
= page_no
- 1;
979 sis
->curr_swap_extent
= list_entry(sis
->extent_list
.prev
,
980 struct swap_extent
, list
);
983 printk(KERN_ERR
"swapon: swapfile has holes\n");
989 #if 0 /* We don't need this yet */
990 #include <linux/backing-dev.h>
991 int page_queue_congested(struct page
*page
)
993 struct backing_dev_info
*bdi
;
995 BUG_ON(!PageLocked(page
)); /* It pins the swap_info_struct */
997 bdi
= page
->mapping
->backing_dev_info
;
998 if (PageSwapCache(page
)) {
999 swp_entry_t entry
= { .val
= page
->index
};
1000 struct swap_info_struct
*sis
;
1002 sis
= get_swap_info_struct(swp_type(entry
));
1003 bdi
= sis
->bdev
->bd_inode
->i_mapping
->backing_dev_info
;
1005 return bdi_write_congested(bdi
);
1009 asmlinkage
long sys_swapoff(const char __user
* specialfile
)
1011 struct swap_info_struct
* p
= NULL
;
1012 unsigned short *swap_map
;
1013 struct file
*swap_file
, *victim
;
1014 struct address_space
*mapping
;
1019 if (!capable(CAP_SYS_ADMIN
))
1022 pathname
= getname(specialfile
);
1023 err
= PTR_ERR(pathname
);
1024 if (IS_ERR(pathname
))
1027 victim
= filp_open(pathname
, O_RDWR
, 0);
1029 err
= PTR_ERR(victim
);
1033 mapping
= victim
->f_dentry
->d_inode
->i_mapping
;
1036 for (type
= swap_list
.head
; type
>= 0; type
= swap_info
[type
].next
) {
1037 p
= swap_info
+ type
;
1038 if ((p
->flags
& SWP_ACTIVE
) == SWP_ACTIVE
) {
1039 if (p
->swap_file
->f_dentry
->d_inode
->i_mapping
==mapping
)
1049 if (!security_vm_enough_memory(p
->pages
))
1050 vm_unacct_memory(p
->pages
);
1057 swap_list
.head
= p
->next
;
1059 swap_info
[prev
].next
= p
->next
;
1061 if (type
== swap_list
.next
) {
1062 /* just pick something that's safe... */
1063 swap_list
.next
= swap_list
.head
;
1065 nr_swap_pages
-= p
->pages
;
1066 total_swap_pages
-= p
->pages
;
1067 p
->flags
&= ~SWP_WRITEOK
;
1069 current
->flags
|= PF_SWAPOFF
;
1070 err
= try_to_unuse(type
);
1071 current
->flags
&= ~PF_SWAPOFF
;
1073 /* re-insert swap space back into swap_list */
1075 for (prev
= -1, i
= swap_list
.head
; i
>= 0; prev
= i
, i
= swap_info
[i
].next
)
1076 if (p
->prio
>= swap_info
[i
].prio
)
1080 swap_list
.head
= swap_list
.next
= p
- swap_info
;
1082 swap_info
[prev
].next
= p
- swap_info
;
1083 nr_swap_pages
+= p
->pages
;
1084 total_swap_pages
+= p
->pages
;
1085 p
->flags
|= SWP_WRITEOK
;
1090 swap_device_lock(p
);
1091 swap_file
= p
->swap_file
;
1092 p
->swap_file
= NULL
;
1094 swap_map
= p
->swap_map
;
1097 destroy_swap_extents(p
);
1098 swap_device_unlock(p
);
1101 if (S_ISBLK(swap_file
->f_dentry
->d_inode
->i_mode
)) {
1102 struct block_device
*bdev
;
1103 bdev
= swap_file
->f_dentry
->d_inode
->i_bdev
;
1104 set_blocksize(bdev
, p
->old_block_size
);
1107 up(&swap_file
->f_dentry
->d_inode
->i_mapping
->host
->i_sem
);
1109 filp_close(swap_file
, NULL
);
1113 filp_close(victim
, NULL
);
1118 #ifdef CONFIG_PROC_FS
1120 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
1122 struct swap_info_struct
*ptr
= swap_info
;
1128 for (i
= 0; i
< nr_swapfiles
; i
++, ptr
++) {
1129 if (!(ptr
->flags
& SWP_USED
) || !ptr
->swap_map
)
1138 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
1140 struct swap_info_struct
*ptr
= v
;
1141 void *endptr
= (void *) swap_info
+ nr_swapfiles
* sizeof(struct swap_info_struct
);
1143 for (++ptr
; ptr
< (struct swap_info_struct
*) endptr
; ptr
++) {
1144 if (!(ptr
->flags
& SWP_USED
) || !ptr
->swap_map
)
1153 static void swap_stop(struct seq_file
*swap
, void *v
)
1158 static int swap_show(struct seq_file
*swap
, void *v
)
1160 struct swap_info_struct
*ptr
= v
;
1165 seq_puts(swap
, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1167 file
= ptr
->swap_file
;
1168 len
= seq_path(swap
, file
->f_vfsmnt
, file
->f_dentry
, " \t\n\\");
1169 seq_printf(swap
, "%*s %s\t%d\t%ld\t%d\n",
1170 len
< 40 ? 40 - len
: 1, " ",
1171 S_ISBLK(file
->f_dentry
->d_inode
->i_mode
) ?
1172 "partition" : "file\t",
1173 ptr
->pages
<< (PAGE_SHIFT
- 10),
1174 ptr
->inuse_pages
<< (PAGE_SHIFT
- 10),
1179 static struct seq_operations swaps_op
= {
1180 .start
= swap_start
,
1186 static int swaps_open(struct inode
*inode
, struct file
*file
)
1188 return seq_open(file
, &swaps_op
);
1191 static struct file_operations proc_swaps_operations
= {
1194 .llseek
= seq_lseek
,
1195 .release
= seq_release
,
1198 static int __init
procswaps_init(void)
1200 struct proc_dir_entry
*entry
;
1202 entry
= create_proc_entry("swaps", 0, NULL
);
1204 entry
->proc_fops
= &proc_swaps_operations
;
1207 __initcall(procswaps_init
);
1208 #endif /* CONFIG_PROC_FS */
1211 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1213 * The swapon system call
1215 asmlinkage
long sys_swapon(const char __user
* specialfile
, int swap_flags
)
1217 struct swap_info_struct
* p
;
1219 struct block_device
*bdev
= NULL
;
1220 struct file
*swap_file
= NULL
;
1221 struct address_space
*mapping
;
1225 static int least_priority
= 0;
1226 union swap_header
*swap_header
= 0;
1227 int swap_header_version
;
1228 int nr_good_pages
= 0;
1229 unsigned long maxpages
= 1;
1231 unsigned short *swap_map
;
1232 struct page
*page
= NULL
;
1233 struct inode
*inode
;
1234 struct inode
*downed_inode
= NULL
;
1236 if (!capable(CAP_SYS_ADMIN
))
1240 for (type
= 0 ; type
< nr_swapfiles
; type
++,p
++)
1241 if (!(p
->flags
& SWP_USED
))
1244 if (type
>= MAX_SWAPFILES
) {
1248 if (type
>= nr_swapfiles
)
1249 nr_swapfiles
= type
+1;
1250 INIT_LIST_HEAD(&p
->extent_list
);
1251 p
->flags
= SWP_USED
;
1253 p
->swap_file
= NULL
;
1254 p
->old_block_size
= 0;
1260 p
->sdev_lock
= SPIN_LOCK_UNLOCKED
;
1262 if (swap_flags
& SWAP_FLAG_PREFER
) {
1264 (swap_flags
& SWAP_FLAG_PRIO_MASK
)>>SWAP_FLAG_PRIO_SHIFT
;
1266 p
->prio
= --least_priority
;
1269 name
= getname(specialfile
);
1270 error
= PTR_ERR(name
);
1273 swap_file
= filp_open(name
, O_RDWR
, 0);
1274 error
= PTR_ERR(swap_file
);
1275 if (IS_ERR(swap_file
)) {
1280 p
->swap_file
= swap_file
;
1281 inode
= swap_file
->f_dentry
->d_inode
;
1282 mapping
= swap_file
->f_dentry
->d_inode
->i_mapping
;
1285 for (i
= 0; i
< nr_swapfiles
; i
++) {
1286 struct swap_info_struct
*q
= &swap_info
[i
];
1288 if (i
== type
|| !q
->swap_file
)
1290 if (mapping
== q
->swap_file
->f_dentry
->d_inode
->i_mapping
)
1295 if (S_ISBLK(inode
->i_mode
)) {
1296 bdev
= inode
->i_bdev
;
1297 error
= bd_claim(bdev
, sys_swapon
);
1302 p
->old_block_size
= block_size(bdev
);
1303 error
= set_blocksize(inode
->i_bdev
, PAGE_SIZE
);
1307 } else if (S_ISREG(inode
->i_mode
)) {
1308 p
->bdev
= inode
->i_sb
->s_bdev
;
1309 downed_inode
= mapping
->host
;
1310 down(&downed_inode
->i_sem
);
1315 swapfilesize
= i_size_read(mapping
->host
) >> PAGE_SHIFT
;
1318 * Read the swap header.
1320 page
= read_cache_page(mapping
, 0,
1321 (filler_t
*)mapping
->a_ops
->readpage
, swap_file
);
1323 error
= PTR_ERR(page
);
1326 wait_on_page_locked(page
);
1327 if (!PageUptodate(page
))
1330 swap_header
= page_address(page
);
1332 if (!memcmp("SWAP-SPACE",swap_header
->magic
.magic
,10))
1333 swap_header_version
= 1;
1334 else if (!memcmp("SWAPSPACE2",swap_header
->magic
.magic
,10))
1335 swap_header_version
= 2;
1337 printk("Unable to find swap-space signature\n");
1342 switch (swap_header_version
) {
1344 printk(KERN_ERR
"version 0 swap is no longer supported. "
1345 "Use mkswap -v1 %s\n", name
);
1349 /* Check the swap header's sub-version and the size of
1350 the swap file and bad block lists */
1351 if (swap_header
->info
.version
!= 1) {
1353 "Unable to handle swap header version %d\n",
1354 swap_header
->info
.version
);
1360 maxpages
= swp_offset(swp_entry(0,~0UL)) - 1;
1361 if (maxpages
> swap_header
->info
.last_page
)
1362 maxpages
= swap_header
->info
.last_page
;
1363 p
->highest_bit
= maxpages
- 1;
1366 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
1369 /* OK, set up the swap map and apply the bad block list */
1370 if (!(p
->swap_map
= vmalloc(maxpages
* sizeof(short)))) {
1376 memset(p
->swap_map
, 0, maxpages
* sizeof(short));
1377 for (i
=0; i
<swap_header
->info
.nr_badpages
; i
++) {
1378 int page
= swap_header
->info
.badpages
[i
];
1379 if (page
<= 0 || page
>= swap_header
->info
.last_page
)
1382 p
->swap_map
[page
] = SWAP_MAP_BAD
;
1384 nr_good_pages
= swap_header
->info
.last_page
-
1385 swap_header
->info
.nr_badpages
-
1386 1 /* header page */;
1391 if (swapfilesize
&& maxpages
> swapfilesize
) {
1393 "Swap area shorter than signature indicates\n");
1397 if (!nr_good_pages
) {
1398 printk(KERN_WARNING
"Empty swap-file\n");
1402 p
->swap_map
[0] = SWAP_MAP_BAD
;
1404 p
->pages
= nr_good_pages
;
1406 if (setup_swap_extents(p
))
1410 swap_device_lock(p
);
1411 p
->flags
= SWP_ACTIVE
;
1412 nr_swap_pages
+= nr_good_pages
;
1413 total_swap_pages
+= nr_good_pages
;
1414 printk(KERN_INFO
"Adding %dk swap on %s. Priority:%d extents:%d\n",
1415 nr_good_pages
<<(PAGE_SHIFT
-10), name
,
1416 p
->prio
, p
->nr_extents
);
1418 /* insert swap space into swap_list: */
1420 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
].next
) {
1421 if (p
->prio
>= swap_info
[i
].prio
) {
1428 swap_list
.head
= swap_list
.next
= p
- swap_info
;
1430 swap_info
[prev
].next
= p
- swap_info
;
1432 swap_device_unlock(p
);
1438 set_blocksize(bdev
, p
->old_block_size
);
1443 swap_map
= p
->swap_map
;
1444 p
->swap_file
= NULL
;
1447 if (!(swap_flags
& SWAP_FLAG_PREFER
))
1450 destroy_swap_extents(p
);
1453 if (swap_file
&& !IS_ERR(swap_file
))
1454 filp_close(swap_file
, NULL
);
1456 if (page
&& !IS_ERR(page
)) {
1458 page_cache_release(page
);
1462 if (error
&& downed_inode
)
1463 up(&downed_inode
->i_sem
);
1467 void si_swapinfo(struct sysinfo
*val
)
1470 unsigned long nr_to_be_unused
= 0;
1473 for (i
= 0; i
< nr_swapfiles
; i
++) {
1474 if (!(swap_info
[i
].flags
& SWP_USED
) ||
1475 (swap_info
[i
].flags
& SWP_WRITEOK
))
1477 nr_to_be_unused
+= swap_info
[i
].inuse_pages
;
1479 val
->freeswap
= nr_swap_pages
+ nr_to_be_unused
;
1480 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
1485 * Verify that a swap entry is valid and increment its swap map count.
1487 * Note: if swap_map[] reaches SWAP_MAP_MAX the entries are treated as
1488 * "permanent", but will be reclaimed by the next swapoff.
1490 int swap_duplicate(swp_entry_t entry
)
1492 struct swap_info_struct
* p
;
1493 unsigned long offset
, type
;
1496 type
= swp_type(entry
);
1497 if (type
>= nr_swapfiles
)
1499 p
= type
+ swap_info
;
1500 offset
= swp_offset(entry
);
1502 swap_device_lock(p
);
1503 if (offset
< p
->max
&& p
->swap_map
[offset
]) {
1504 if (p
->swap_map
[offset
] < SWAP_MAP_MAX
- 1) {
1505 p
->swap_map
[offset
]++;
1507 } else if (p
->swap_map
[offset
] <= SWAP_MAP_MAX
) {
1508 if (swap_overflow
++ < 5)
1509 printk(KERN_WARNING
"swap_dup: swap entry overflow\n");
1510 p
->swap_map
[offset
] = SWAP_MAP_MAX
;
1514 swap_device_unlock(p
);
1519 printk(KERN_ERR
"swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
1523 struct swap_info_struct
*
1524 get_swap_info_struct(unsigned type
)
1526 return &swap_info
[type
];
1530 * swap_device_lock prevents swap_map being freed. Don't grab an extra
1531 * reference on the swaphandle, it doesn't matter if it becomes unused.
1533 int valid_swaphandles(swp_entry_t entry
, unsigned long *offset
)
1535 int ret
= 0, i
= 1 << page_cluster
;
1537 struct swap_info_struct
*swapdev
= swp_type(entry
) + swap_info
;
1539 if (!page_cluster
) /* no readahead */
1541 toff
= (swp_offset(entry
) >> page_cluster
) << page_cluster
;
1542 if (!toff
) /* first page is swap header */
1546 swap_device_lock(swapdev
);
1548 /* Don't read-ahead past the end of the swap area */
1549 if (toff
>= swapdev
->max
)
1551 /* Don't read in free or bad pages */
1552 if (!swapdev
->swap_map
[toff
])
1554 if (swapdev
->swap_map
[toff
] == SWAP_MAP_BAD
)
1559 swap_device_unlock(swapdev
);