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/random.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/ksm.h>
26 #include <linux/rmap.h>
27 #include <linux/security.h>
28 #include <linux/backing-dev.h>
29 #include <linux/mutex.h>
30 #include <linux/capability.h>
31 #include <linux/syscalls.h>
32 #include <linux/memcontrol.h>
34 #include <asm/pgtable.h>
35 #include <asm/tlbflush.h>
36 #include <linux/swapops.h>
37 #include <linux/page_cgroup.h>
39 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
41 static void free_swap_count_continuations(struct swap_info_struct
*);
42 static sector_t
map_swap_entry(swp_entry_t
, struct block_device
**);
44 static DEFINE_SPINLOCK(swap_lock
);
45 static unsigned int nr_swapfiles
;
47 long total_swap_pages
;
48 static int least_priority
;
50 static const char Bad_file
[] = "Bad swap file entry ";
51 static const char Unused_file
[] = "Unused swap file entry ";
52 static const char Bad_offset
[] = "Bad swap offset entry ";
53 static const char Unused_offset
[] = "Unused swap offset entry ";
55 static struct swap_list_t swap_list
= {-1, -1};
57 static struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
59 static DEFINE_MUTEX(swapon_mutex
);
61 static inline unsigned char swap_count(unsigned char ent
)
63 return ent
& ~SWAP_HAS_CACHE
; /* may include SWAP_HAS_CONT flag */
66 /* returns 1 if swap entry is freed */
68 __try_to_reclaim_swap(struct swap_info_struct
*si
, unsigned long offset
)
70 swp_entry_t entry
= swp_entry(si
->type
, offset
);
74 page
= find_get_page(&swapper_space
, entry
.val
);
78 * This function is called from scan_swap_map() and it's called
79 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
80 * We have to use trylock for avoiding deadlock. This is a special
81 * case and you should use try_to_free_swap() with explicit lock_page()
82 * in usual operations.
84 if (trylock_page(page
)) {
85 ret
= try_to_free_swap(page
);
88 page_cache_release(page
);
93 * We need this because the bdev->unplug_fn can sleep and we cannot
94 * hold swap_lock while calling the unplug_fn. And swap_lock
95 * cannot be turned into a mutex.
97 static DECLARE_RWSEM(swap_unplug_sem
);
99 void swap_unplug_io_fn(struct backing_dev_info
*unused_bdi
, struct page
*page
)
103 down_read(&swap_unplug_sem
);
104 entry
.val
= page_private(page
);
105 if (PageSwapCache(page
)) {
106 struct block_device
*bdev
= swap_info
[swp_type(entry
)]->bdev
;
107 struct backing_dev_info
*bdi
;
110 * If the page is removed from swapcache from under us (with a
111 * racy try_to_unuse/swapoff) we need an additional reference
112 * count to avoid reading garbage from page_private(page) above.
113 * If the WARN_ON triggers during a swapoff it maybe the race
114 * condition and it's harmless. However if it triggers without
115 * swapoff it signals a problem.
117 WARN_ON(page_count(page
) <= 1);
119 bdi
= bdev
->bd_inode
->i_mapping
->backing_dev_info
;
120 blk_run_backing_dev(bdi
, page
);
122 up_read(&swap_unplug_sem
);
126 * swapon tell device that all the old swap contents can be discarded,
127 * to allow the swap device to optimize its wear-levelling.
129 static int discard_swap(struct swap_info_struct
*si
)
131 struct swap_extent
*se
;
132 sector_t start_block
;
136 /* Do not discard the swap header page! */
137 se
= &si
->first_swap_extent
;
138 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
139 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
141 err
= blkdev_issue_discard(si
->bdev
, start_block
,
142 nr_blocks
, GFP_KERNEL
,
143 BLKDEV_IFL_WAIT
| BLKDEV_IFL_BARRIER
);
149 list_for_each_entry(se
, &si
->first_swap_extent
.list
, list
) {
150 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
151 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
153 err
= blkdev_issue_discard(si
->bdev
, start_block
,
154 nr_blocks
, GFP_KERNEL
,
155 BLKDEV_IFL_WAIT
| BLKDEV_IFL_BARRIER
);
161 return err
; /* That will often be -EOPNOTSUPP */
165 * swap allocation tell device that a cluster of swap can now be discarded,
166 * to allow the swap device to optimize its wear-levelling.
168 static void discard_swap_cluster(struct swap_info_struct
*si
,
169 pgoff_t start_page
, pgoff_t nr_pages
)
171 struct swap_extent
*se
= si
->curr_swap_extent
;
172 int found_extent
= 0;
175 struct list_head
*lh
;
177 if (se
->start_page
<= start_page
&&
178 start_page
< se
->start_page
+ se
->nr_pages
) {
179 pgoff_t offset
= start_page
- se
->start_page
;
180 sector_t start_block
= se
->start_block
+ offset
;
181 sector_t nr_blocks
= se
->nr_pages
- offset
;
183 if (nr_blocks
> nr_pages
)
184 nr_blocks
= nr_pages
;
185 start_page
+= nr_blocks
;
186 nr_pages
-= nr_blocks
;
189 si
->curr_swap_extent
= se
;
191 start_block
<<= PAGE_SHIFT
- 9;
192 nr_blocks
<<= PAGE_SHIFT
- 9;
193 if (blkdev_issue_discard(si
->bdev
, start_block
,
194 nr_blocks
, GFP_NOIO
, BLKDEV_IFL_WAIT
|
200 se
= list_entry(lh
, struct swap_extent
, list
);
204 static int wait_for_discard(void *word
)
210 #define SWAPFILE_CLUSTER 256
211 #define LATENCY_LIMIT 256
213 static inline unsigned long scan_swap_map(struct swap_info_struct
*si
,
216 unsigned long offset
;
217 unsigned long scan_base
;
218 unsigned long last_in_cluster
= 0;
219 int latency_ration
= LATENCY_LIMIT
;
220 int found_free_cluster
= 0;
223 * We try to cluster swap pages by allocating them sequentially
224 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
225 * way, however, we resort to first-free allocation, starting
226 * a new cluster. This prevents us from scattering swap pages
227 * all over the entire swap partition, so that we reduce
228 * overall disk seek times between swap pages. -- sct
229 * But we do now try to find an empty cluster. -Andrea
230 * And we let swap pages go all over an SSD partition. Hugh
233 si
->flags
+= SWP_SCANNING
;
234 scan_base
= offset
= si
->cluster_next
;
236 if (unlikely(!si
->cluster_nr
--)) {
237 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
238 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
241 if (si
->flags
& SWP_DISCARDABLE
) {
243 * Start range check on racing allocations, in case
244 * they overlap the cluster we eventually decide on
245 * (we scan without swap_lock to allow preemption).
246 * It's hardly conceivable that cluster_nr could be
247 * wrapped during our scan, but don't depend on it.
249 if (si
->lowest_alloc
)
251 si
->lowest_alloc
= si
->max
;
252 si
->highest_alloc
= 0;
254 spin_unlock(&swap_lock
);
257 * If seek is expensive, start searching for new cluster from
258 * start of partition, to minimize the span of allocated swap.
259 * But if seek is cheap, search from our current position, so
260 * that swap is allocated from all over the partition: if the
261 * Flash Translation Layer only remaps within limited zones,
262 * we don't want to wear out the first zone too quickly.
264 if (!(si
->flags
& SWP_SOLIDSTATE
))
265 scan_base
= offset
= si
->lowest_bit
;
266 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
268 /* Locate the first empty (unaligned) cluster */
269 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
270 if (si
->swap_map
[offset
])
271 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
272 else if (offset
== last_in_cluster
) {
273 spin_lock(&swap_lock
);
274 offset
-= SWAPFILE_CLUSTER
- 1;
275 si
->cluster_next
= offset
;
276 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
277 found_free_cluster
= 1;
280 if (unlikely(--latency_ration
< 0)) {
282 latency_ration
= LATENCY_LIMIT
;
286 offset
= si
->lowest_bit
;
287 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
289 /* Locate the first empty (unaligned) cluster */
290 for (; last_in_cluster
< scan_base
; offset
++) {
291 if (si
->swap_map
[offset
])
292 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
293 else if (offset
== last_in_cluster
) {
294 spin_lock(&swap_lock
);
295 offset
-= SWAPFILE_CLUSTER
- 1;
296 si
->cluster_next
= offset
;
297 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
298 found_free_cluster
= 1;
301 if (unlikely(--latency_ration
< 0)) {
303 latency_ration
= LATENCY_LIMIT
;
308 spin_lock(&swap_lock
);
309 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
310 si
->lowest_alloc
= 0;
314 if (!(si
->flags
& SWP_WRITEOK
))
316 if (!si
->highest_bit
)
318 if (offset
> si
->highest_bit
)
319 scan_base
= offset
= si
->lowest_bit
;
321 /* reuse swap entry of cache-only swap if not hibernation. */
323 && usage
== SWAP_HAS_CACHE
324 && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
326 spin_unlock(&swap_lock
);
327 swap_was_freed
= __try_to_reclaim_swap(si
, offset
);
328 spin_lock(&swap_lock
);
329 /* entry was freed successfully, try to use this again */
332 goto scan
; /* check next one */
335 if (si
->swap_map
[offset
])
338 if (offset
== si
->lowest_bit
)
340 if (offset
== si
->highest_bit
)
343 if (si
->inuse_pages
== si
->pages
) {
344 si
->lowest_bit
= si
->max
;
347 si
->swap_map
[offset
] = usage
;
348 si
->cluster_next
= offset
+ 1;
349 si
->flags
-= SWP_SCANNING
;
351 if (si
->lowest_alloc
) {
353 * Only set when SWP_DISCARDABLE, and there's a scan
354 * for a free cluster in progress or just completed.
356 if (found_free_cluster
) {
358 * To optimize wear-levelling, discard the
359 * old data of the cluster, taking care not to
360 * discard any of its pages that have already
361 * been allocated by racing tasks (offset has
362 * already stepped over any at the beginning).
364 if (offset
< si
->highest_alloc
&&
365 si
->lowest_alloc
<= last_in_cluster
)
366 last_in_cluster
= si
->lowest_alloc
- 1;
367 si
->flags
|= SWP_DISCARDING
;
368 spin_unlock(&swap_lock
);
370 if (offset
< last_in_cluster
)
371 discard_swap_cluster(si
, offset
,
372 last_in_cluster
- offset
+ 1);
374 spin_lock(&swap_lock
);
375 si
->lowest_alloc
= 0;
376 si
->flags
&= ~SWP_DISCARDING
;
378 smp_mb(); /* wake_up_bit advises this */
379 wake_up_bit(&si
->flags
, ilog2(SWP_DISCARDING
));
381 } else if (si
->flags
& SWP_DISCARDING
) {
383 * Delay using pages allocated by racing tasks
384 * until the whole discard has been issued. We
385 * could defer that delay until swap_writepage,
386 * but it's easier to keep this self-contained.
388 spin_unlock(&swap_lock
);
389 wait_on_bit(&si
->flags
, ilog2(SWP_DISCARDING
),
390 wait_for_discard
, TASK_UNINTERRUPTIBLE
);
391 spin_lock(&swap_lock
);
394 * Note pages allocated by racing tasks while
395 * scan for a free cluster is in progress, so
396 * that its final discard can exclude them.
398 if (offset
< si
->lowest_alloc
)
399 si
->lowest_alloc
= offset
;
400 if (offset
> si
->highest_alloc
)
401 si
->highest_alloc
= offset
;
407 spin_unlock(&swap_lock
);
408 while (++offset
<= si
->highest_bit
) {
409 if (!si
->swap_map
[offset
]) {
410 spin_lock(&swap_lock
);
413 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
414 spin_lock(&swap_lock
);
417 if (unlikely(--latency_ration
< 0)) {
419 latency_ration
= LATENCY_LIMIT
;
422 offset
= si
->lowest_bit
;
423 while (++offset
< scan_base
) {
424 if (!si
->swap_map
[offset
]) {
425 spin_lock(&swap_lock
);
428 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
429 spin_lock(&swap_lock
);
432 if (unlikely(--latency_ration
< 0)) {
434 latency_ration
= LATENCY_LIMIT
;
437 spin_lock(&swap_lock
);
440 si
->flags
-= SWP_SCANNING
;
444 swp_entry_t
get_swap_page(void)
446 struct swap_info_struct
*si
;
451 spin_lock(&swap_lock
);
452 if (nr_swap_pages
<= 0)
456 for (type
= swap_list
.next
; type
>= 0 && wrapped
< 2; type
= next
) {
457 si
= swap_info
[type
];
460 (!wrapped
&& si
->prio
!= swap_info
[next
]->prio
)) {
461 next
= swap_list
.head
;
465 if (!si
->highest_bit
)
467 if (!(si
->flags
& SWP_WRITEOK
))
470 swap_list
.next
= next
;
471 /* This is called for allocating swap entry for cache */
472 offset
= scan_swap_map(si
, SWAP_HAS_CACHE
);
474 spin_unlock(&swap_lock
);
475 return swp_entry(type
, offset
);
477 next
= swap_list
.next
;
482 spin_unlock(&swap_lock
);
483 return (swp_entry_t
) {0};
486 /* The only caller of this function is now susupend routine */
487 swp_entry_t
get_swap_page_of_type(int type
)
489 struct swap_info_struct
*si
;
492 spin_lock(&swap_lock
);
493 si
= swap_info
[type
];
494 if (si
&& (si
->flags
& SWP_WRITEOK
)) {
496 /* This is called for allocating swap entry, not cache */
497 offset
= scan_swap_map(si
, 1);
499 spin_unlock(&swap_lock
);
500 return swp_entry(type
, offset
);
504 spin_unlock(&swap_lock
);
505 return (swp_entry_t
) {0};
508 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
510 struct swap_info_struct
*p
;
511 unsigned long offset
, type
;
515 type
= swp_type(entry
);
516 if (type
>= nr_swapfiles
)
519 if (!(p
->flags
& SWP_USED
))
521 offset
= swp_offset(entry
);
522 if (offset
>= p
->max
)
524 if (!p
->swap_map
[offset
])
526 spin_lock(&swap_lock
);
530 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_offset
, entry
.val
);
533 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_offset
, entry
.val
);
536 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_file
, entry
.val
);
539 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_file
, entry
.val
);
544 static unsigned char swap_entry_free(struct swap_info_struct
*p
,
545 swp_entry_t entry
, unsigned char usage
)
547 unsigned long offset
= swp_offset(entry
);
549 unsigned char has_cache
;
551 count
= p
->swap_map
[offset
];
552 has_cache
= count
& SWAP_HAS_CACHE
;
553 count
&= ~SWAP_HAS_CACHE
;
555 if (usage
== SWAP_HAS_CACHE
) {
556 VM_BUG_ON(!has_cache
);
558 } else if (count
== SWAP_MAP_SHMEM
) {
560 * Or we could insist on shmem.c using a special
561 * swap_shmem_free() and free_shmem_swap_and_cache()...
564 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
565 if (count
== COUNT_CONTINUED
) {
566 if (swap_count_continued(p
, offset
, count
))
567 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
569 count
= SWAP_MAP_MAX
;
575 mem_cgroup_uncharge_swap(entry
);
577 usage
= count
| has_cache
;
578 p
->swap_map
[offset
] = usage
;
580 /* free if no reference */
582 struct gendisk
*disk
= p
->bdev
->bd_disk
;
583 if (offset
< p
->lowest_bit
)
584 p
->lowest_bit
= offset
;
585 if (offset
> p
->highest_bit
)
586 p
->highest_bit
= offset
;
587 if (swap_list
.next
>= 0 &&
588 p
->prio
> swap_info
[swap_list
.next
]->prio
)
589 swap_list
.next
= p
->type
;
592 if ((p
->flags
& SWP_BLKDEV
) &&
593 disk
->fops
->swap_slot_free_notify
)
594 disk
->fops
->swap_slot_free_notify(p
->bdev
, offset
);
601 * Caller has made sure that the swapdevice corresponding to entry
602 * is still around or has not been recycled.
604 void swap_free(swp_entry_t entry
)
606 struct swap_info_struct
*p
;
608 p
= swap_info_get(entry
);
610 swap_entry_free(p
, entry
, 1);
611 spin_unlock(&swap_lock
);
616 * Called after dropping swapcache to decrease refcnt to swap entries.
618 void swapcache_free(swp_entry_t entry
, struct page
*page
)
620 struct swap_info_struct
*p
;
623 p
= swap_info_get(entry
);
625 count
= swap_entry_free(p
, entry
, SWAP_HAS_CACHE
);
627 mem_cgroup_uncharge_swapcache(page
, entry
, count
!= 0);
628 spin_unlock(&swap_lock
);
633 * How many references to page are currently swapped out?
634 * This does not give an exact answer when swap count is continued,
635 * but does include the high COUNT_CONTINUED flag to allow for that.
637 static inline int page_swapcount(struct page
*page
)
640 struct swap_info_struct
*p
;
643 entry
.val
= page_private(page
);
644 p
= swap_info_get(entry
);
646 count
= swap_count(p
->swap_map
[swp_offset(entry
)]);
647 spin_unlock(&swap_lock
);
653 * We can write to an anon page without COW if there are no other references
654 * to it. And as a side-effect, free up its swap: because the old content
655 * on disk will never be read, and seeking back there to write new content
656 * later would only waste time away from clustering.
658 int reuse_swap_page(struct page
*page
)
662 VM_BUG_ON(!PageLocked(page
));
663 if (unlikely(PageKsm(page
)))
665 count
= page_mapcount(page
);
666 if (count
<= 1 && PageSwapCache(page
)) {
667 count
+= page_swapcount(page
);
668 if (count
== 1 && !PageWriteback(page
)) {
669 delete_from_swap_cache(page
);
677 * If swap is getting full, or if there are no more mappings of this page,
678 * then try_to_free_swap is called to free its swap space.
680 int try_to_free_swap(struct page
*page
)
682 VM_BUG_ON(!PageLocked(page
));
684 if (!PageSwapCache(page
))
686 if (PageWriteback(page
))
688 if (page_swapcount(page
))
691 delete_from_swap_cache(page
);
697 * Free the swap entry like above, but also try to
698 * free the page cache entry if it is the last user.
700 int free_swap_and_cache(swp_entry_t entry
)
702 struct swap_info_struct
*p
;
703 struct page
*page
= NULL
;
705 if (non_swap_entry(entry
))
708 p
= swap_info_get(entry
);
710 if (swap_entry_free(p
, entry
, 1) == SWAP_HAS_CACHE
) {
711 page
= find_get_page(&swapper_space
, entry
.val
);
712 if (page
&& !trylock_page(page
)) {
713 page_cache_release(page
);
717 spin_unlock(&swap_lock
);
721 * Not mapped elsewhere, or swap space full? Free it!
722 * Also recheck PageSwapCache now page is locked (above).
724 if (PageSwapCache(page
) && !PageWriteback(page
) &&
725 (!page_mapped(page
) || vm_swap_full())) {
726 delete_from_swap_cache(page
);
730 page_cache_release(page
);
735 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
737 * mem_cgroup_count_swap_user - count the user of a swap entry
738 * @ent: the swap entry to be checked
739 * @pagep: the pointer for the swap cache page of the entry to be stored
741 * Returns the number of the user of the swap entry. The number is valid only
742 * for swaps of anonymous pages.
743 * If the entry is found on swap cache, the page is stored to pagep with
744 * refcount of it being incremented.
746 int mem_cgroup_count_swap_user(swp_entry_t ent
, struct page
**pagep
)
749 struct swap_info_struct
*p
;
752 page
= find_get_page(&swapper_space
, ent
.val
);
754 count
+= page_mapcount(page
);
755 p
= swap_info_get(ent
);
757 count
+= swap_count(p
->swap_map
[swp_offset(ent
)]);
758 spin_unlock(&swap_lock
);
766 #ifdef CONFIG_HIBERNATION
768 * Find the swap type that corresponds to given device (if any).
770 * @offset - number of the PAGE_SIZE-sized block of the device, starting
771 * from 0, in which the swap header is expected to be located.
773 * This is needed for the suspend to disk (aka swsusp).
775 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
777 struct block_device
*bdev
= NULL
;
781 bdev
= bdget(device
);
783 spin_lock(&swap_lock
);
784 for (type
= 0; type
< nr_swapfiles
; type
++) {
785 struct swap_info_struct
*sis
= swap_info
[type
];
787 if (!(sis
->flags
& SWP_WRITEOK
))
792 *bdev_p
= bdgrab(sis
->bdev
);
794 spin_unlock(&swap_lock
);
797 if (bdev
== sis
->bdev
) {
798 struct swap_extent
*se
= &sis
->first_swap_extent
;
800 if (se
->start_block
== offset
) {
802 *bdev_p
= bdgrab(sis
->bdev
);
804 spin_unlock(&swap_lock
);
810 spin_unlock(&swap_lock
);
818 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
819 * corresponding to given index in swap_info (swap type).
821 sector_t
swapdev_block(int type
, pgoff_t offset
)
823 struct block_device
*bdev
;
825 if ((unsigned int)type
>= nr_swapfiles
)
827 if (!(swap_info
[type
]->flags
& SWP_WRITEOK
))
829 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
833 * Return either the total number of swap pages of given type, or the number
834 * of free pages of that type (depending on @free)
836 * This is needed for software suspend
838 unsigned int count_swap_pages(int type
, int free
)
842 spin_lock(&swap_lock
);
843 if ((unsigned int)type
< nr_swapfiles
) {
844 struct swap_info_struct
*sis
= swap_info
[type
];
846 if (sis
->flags
& SWP_WRITEOK
) {
849 n
-= sis
->inuse_pages
;
852 spin_unlock(&swap_lock
);
855 #endif /* CONFIG_HIBERNATION */
858 * No need to decide whether this PTE shares the swap entry with others,
859 * just let do_wp_page work it out if a write is requested later - to
860 * force COW, vm_page_prot omits write permission from any private vma.
862 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
863 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
865 struct mem_cgroup
*ptr
= NULL
;
870 if (mem_cgroup_try_charge_swapin(vma
->vm_mm
, page
, GFP_KERNEL
, &ptr
)) {
875 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
876 if (unlikely(!pte_same(*pte
, swp_entry_to_pte(entry
)))) {
878 mem_cgroup_cancel_charge_swapin(ptr
);
883 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
884 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
886 set_pte_at(vma
->vm_mm
, addr
, pte
,
887 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
888 page_add_anon_rmap(page
, vma
, addr
);
889 mem_cgroup_commit_charge_swapin(page
, ptr
);
892 * Move the page to the active list so it is not
893 * immediately swapped out again after swapon.
897 pte_unmap_unlock(pte
, ptl
);
902 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
903 unsigned long addr
, unsigned long end
,
904 swp_entry_t entry
, struct page
*page
)
906 pte_t swp_pte
= swp_entry_to_pte(entry
);
911 * We don't actually need pte lock while scanning for swp_pte: since
912 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
913 * page table while we're scanning; though it could get zapped, and on
914 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
915 * of unmatched parts which look like swp_pte, so unuse_pte must
916 * recheck under pte lock. Scanning without pte lock lets it be
917 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
919 pte
= pte_offset_map(pmd
, addr
);
922 * swapoff spends a _lot_ of time in this loop!
923 * Test inline before going to call unuse_pte.
925 if (unlikely(pte_same(*pte
, swp_pte
))) {
927 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
930 pte
= pte_offset_map(pmd
, addr
);
932 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
938 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
939 unsigned long addr
, unsigned long end
,
940 swp_entry_t entry
, struct page
*page
)
946 pmd
= pmd_offset(pud
, addr
);
948 next
= pmd_addr_end(addr
, end
);
949 if (pmd_none_or_clear_bad(pmd
))
951 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
);
954 } while (pmd
++, addr
= next
, addr
!= end
);
958 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
959 unsigned long addr
, unsigned long end
,
960 swp_entry_t entry
, struct page
*page
)
966 pud
= pud_offset(pgd
, addr
);
968 next
= pud_addr_end(addr
, end
);
969 if (pud_none_or_clear_bad(pud
))
971 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
);
974 } while (pud
++, addr
= next
, addr
!= end
);
978 static int unuse_vma(struct vm_area_struct
*vma
,
979 swp_entry_t entry
, struct page
*page
)
982 unsigned long addr
, end
, next
;
985 if (page_anon_vma(page
)) {
986 addr
= page_address_in_vma(page
, vma
);
990 end
= addr
+ PAGE_SIZE
;
992 addr
= vma
->vm_start
;
996 pgd
= pgd_offset(vma
->vm_mm
, addr
);
998 next
= pgd_addr_end(addr
, end
);
999 if (pgd_none_or_clear_bad(pgd
))
1001 ret
= unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
);
1004 } while (pgd
++, addr
= next
, addr
!= end
);
1008 static int unuse_mm(struct mm_struct
*mm
,
1009 swp_entry_t entry
, struct page
*page
)
1011 struct vm_area_struct
*vma
;
1014 if (!down_read_trylock(&mm
->mmap_sem
)) {
1016 * Activate page so shrink_inactive_list is unlikely to unmap
1017 * its ptes while lock is dropped, so swapoff can make progress.
1019 activate_page(page
);
1021 down_read(&mm
->mmap_sem
);
1024 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1025 if (vma
->anon_vma
&& (ret
= unuse_vma(vma
, entry
, page
)))
1028 up_read(&mm
->mmap_sem
);
1029 return (ret
< 0)? ret
: 0;
1033 * Scan swap_map from current position to next entry still in use.
1034 * Recycle to start on reaching the end, returning 0 when empty.
1036 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
1039 unsigned int max
= si
->max
;
1040 unsigned int i
= prev
;
1041 unsigned char count
;
1044 * No need for swap_lock here: we're just looking
1045 * for whether an entry is in use, not modifying it; false
1046 * hits are okay, and sys_swapoff() has already prevented new
1047 * allocations from this area (while holding swap_lock).
1056 * No entries in use at top of swap_map,
1057 * loop back to start and recheck there.
1063 count
= si
->swap_map
[i
];
1064 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
1071 * We completely avoid races by reading each swap page in advance,
1072 * and then search for the process using it. All the necessary
1073 * page table adjustments can then be made atomically.
1075 static int try_to_unuse(unsigned int type
)
1077 struct swap_info_struct
*si
= swap_info
[type
];
1078 struct mm_struct
*start_mm
;
1079 unsigned char *swap_map
;
1080 unsigned char swcount
;
1087 * When searching mms for an entry, a good strategy is to
1088 * start at the first mm we freed the previous entry from
1089 * (though actually we don't notice whether we or coincidence
1090 * freed the entry). Initialize this start_mm with a hold.
1092 * A simpler strategy would be to start at the last mm we
1093 * freed the previous entry from; but that would take less
1094 * advantage of mmlist ordering, which clusters forked mms
1095 * together, child after parent. If we race with dup_mmap(), we
1096 * prefer to resolve parent before child, lest we miss entries
1097 * duplicated after we scanned child: using last mm would invert
1100 start_mm
= &init_mm
;
1101 atomic_inc(&init_mm
.mm_users
);
1104 * Keep on scanning until all entries have gone. Usually,
1105 * one pass through swap_map is enough, but not necessarily:
1106 * there are races when an instance of an entry might be missed.
1108 while ((i
= find_next_to_unuse(si
, i
)) != 0) {
1109 if (signal_pending(current
)) {
1115 * Get a page for the entry, using the existing swap
1116 * cache page if there is one. Otherwise, get a clean
1117 * page and read the swap into it.
1119 swap_map
= &si
->swap_map
[i
];
1120 entry
= swp_entry(type
, i
);
1121 page
= read_swap_cache_async(entry
,
1122 GFP_HIGHUSER_MOVABLE
, NULL
, 0);
1125 * Either swap_duplicate() failed because entry
1126 * has been freed independently, and will not be
1127 * reused since sys_swapoff() already disabled
1128 * allocation from here, or alloc_page() failed.
1137 * Don't hold on to start_mm if it looks like exiting.
1139 if (atomic_read(&start_mm
->mm_users
) == 1) {
1141 start_mm
= &init_mm
;
1142 atomic_inc(&init_mm
.mm_users
);
1146 * Wait for and lock page. When do_swap_page races with
1147 * try_to_unuse, do_swap_page can handle the fault much
1148 * faster than try_to_unuse can locate the entry. This
1149 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1150 * defer to do_swap_page in such a case - in some tests,
1151 * do_swap_page and try_to_unuse repeatedly compete.
1153 wait_on_page_locked(page
);
1154 wait_on_page_writeback(page
);
1156 wait_on_page_writeback(page
);
1159 * Remove all references to entry.
1161 swcount
= *swap_map
;
1162 if (swap_count(swcount
) == SWAP_MAP_SHMEM
) {
1163 retval
= shmem_unuse(entry
, page
);
1164 /* page has already been unlocked and released */
1169 if (swap_count(swcount
) && start_mm
!= &init_mm
)
1170 retval
= unuse_mm(start_mm
, entry
, page
);
1172 if (swap_count(*swap_map
)) {
1173 int set_start_mm
= (*swap_map
>= swcount
);
1174 struct list_head
*p
= &start_mm
->mmlist
;
1175 struct mm_struct
*new_start_mm
= start_mm
;
1176 struct mm_struct
*prev_mm
= start_mm
;
1177 struct mm_struct
*mm
;
1179 atomic_inc(&new_start_mm
->mm_users
);
1180 atomic_inc(&prev_mm
->mm_users
);
1181 spin_lock(&mmlist_lock
);
1182 while (swap_count(*swap_map
) && !retval
&&
1183 (p
= p
->next
) != &start_mm
->mmlist
) {
1184 mm
= list_entry(p
, struct mm_struct
, mmlist
);
1185 if (!atomic_inc_not_zero(&mm
->mm_users
))
1187 spin_unlock(&mmlist_lock
);
1193 swcount
= *swap_map
;
1194 if (!swap_count(swcount
)) /* any usage ? */
1196 else if (mm
== &init_mm
)
1199 retval
= unuse_mm(mm
, entry
, page
);
1201 if (set_start_mm
&& *swap_map
< swcount
) {
1202 mmput(new_start_mm
);
1203 atomic_inc(&mm
->mm_users
);
1207 spin_lock(&mmlist_lock
);
1209 spin_unlock(&mmlist_lock
);
1212 start_mm
= new_start_mm
;
1216 page_cache_release(page
);
1221 * If a reference remains (rare), we would like to leave
1222 * the page in the swap cache; but try_to_unmap could
1223 * then re-duplicate the entry once we drop page lock,
1224 * so we might loop indefinitely; also, that page could
1225 * not be swapped out to other storage meanwhile. So:
1226 * delete from cache even if there's another reference,
1227 * after ensuring that the data has been saved to disk -
1228 * since if the reference remains (rarer), it will be
1229 * read from disk into another page. Splitting into two
1230 * pages would be incorrect if swap supported "shared
1231 * private" pages, but they are handled by tmpfs files.
1233 * Given how unuse_vma() targets one particular offset
1234 * in an anon_vma, once the anon_vma has been determined,
1235 * this splitting happens to be just what is needed to
1236 * handle where KSM pages have been swapped out: re-reading
1237 * is unnecessarily slow, but we can fix that later on.
1239 if (swap_count(*swap_map
) &&
1240 PageDirty(page
) && PageSwapCache(page
)) {
1241 struct writeback_control wbc
= {
1242 .sync_mode
= WB_SYNC_NONE
,
1245 swap_writepage(page
, &wbc
);
1247 wait_on_page_writeback(page
);
1251 * It is conceivable that a racing task removed this page from
1252 * swap cache just before we acquired the page lock at the top,
1253 * or while we dropped it in unuse_mm(). The page might even
1254 * be back in swap cache on another swap area: that we must not
1255 * delete, since it may not have been written out to swap yet.
1257 if (PageSwapCache(page
) &&
1258 likely(page_private(page
) == entry
.val
))
1259 delete_from_swap_cache(page
);
1262 * So we could skip searching mms once swap count went
1263 * to 1, we did not mark any present ptes as dirty: must
1264 * mark page dirty so shrink_page_list will preserve it.
1268 page_cache_release(page
);
1271 * Make sure that we aren't completely killing
1272 * interactive performance.
1282 * After a successful try_to_unuse, if no swap is now in use, we know
1283 * we can empty the mmlist. swap_lock must be held on entry and exit.
1284 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1285 * added to the mmlist just after page_duplicate - before would be racy.
1287 static void drain_mmlist(void)
1289 struct list_head
*p
, *next
;
1292 for (type
= 0; type
< nr_swapfiles
; type
++)
1293 if (swap_info
[type
]->inuse_pages
)
1295 spin_lock(&mmlist_lock
);
1296 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
1298 spin_unlock(&mmlist_lock
);
1302 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1303 * corresponds to page offset for the specified swap entry.
1304 * Note that the type of this function is sector_t, but it returns page offset
1305 * into the bdev, not sector offset.
1307 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
1309 struct swap_info_struct
*sis
;
1310 struct swap_extent
*start_se
;
1311 struct swap_extent
*se
;
1314 sis
= swap_info
[swp_type(entry
)];
1317 offset
= swp_offset(entry
);
1318 start_se
= sis
->curr_swap_extent
;
1322 struct list_head
*lh
;
1324 if (se
->start_page
<= offset
&&
1325 offset
< (se
->start_page
+ se
->nr_pages
)) {
1326 return se
->start_block
+ (offset
- se
->start_page
);
1329 se
= list_entry(lh
, struct swap_extent
, list
);
1330 sis
->curr_swap_extent
= se
;
1331 BUG_ON(se
== start_se
); /* It *must* be present */
1336 * Returns the page offset into bdev for the specified page's swap entry.
1338 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
1341 entry
.val
= page_private(page
);
1342 return map_swap_entry(entry
, bdev
);
1346 * Free all of a swapdev's extent information
1348 static void destroy_swap_extents(struct swap_info_struct
*sis
)
1350 while (!list_empty(&sis
->first_swap_extent
.list
)) {
1351 struct swap_extent
*se
;
1353 se
= list_entry(sis
->first_swap_extent
.list
.next
,
1354 struct swap_extent
, list
);
1355 list_del(&se
->list
);
1361 * Add a block range (and the corresponding page range) into this swapdev's
1362 * extent list. The extent list is kept sorted in page order.
1364 * This function rather assumes that it is called in ascending page order.
1367 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
1368 unsigned long nr_pages
, sector_t start_block
)
1370 struct swap_extent
*se
;
1371 struct swap_extent
*new_se
;
1372 struct list_head
*lh
;
1374 if (start_page
== 0) {
1375 se
= &sis
->first_swap_extent
;
1376 sis
->curr_swap_extent
= se
;
1378 se
->nr_pages
= nr_pages
;
1379 se
->start_block
= start_block
;
1382 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
1383 se
= list_entry(lh
, struct swap_extent
, list
);
1384 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
1385 if (se
->start_block
+ se
->nr_pages
== start_block
) {
1387 se
->nr_pages
+= nr_pages
;
1393 * No merge. Insert a new extent, preserving ordering.
1395 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
1398 new_se
->start_page
= start_page
;
1399 new_se
->nr_pages
= nr_pages
;
1400 new_se
->start_block
= start_block
;
1402 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
1407 * A `swap extent' is a simple thing which maps a contiguous range of pages
1408 * onto a contiguous range of disk blocks. An ordered list of swap extents
1409 * is built at swapon time and is then used at swap_writepage/swap_readpage
1410 * time for locating where on disk a page belongs.
1412 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1413 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1414 * swap files identically.
1416 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1417 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1418 * swapfiles are handled *identically* after swapon time.
1420 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1421 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1422 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1423 * requirements, they are simply tossed out - we will never use those blocks
1426 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1427 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1428 * which will scribble on the fs.
1430 * The amount of disk space which a single swap extent represents varies.
1431 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1432 * extents in the list. To avoid much list walking, we cache the previous
1433 * search location in `curr_swap_extent', and start new searches from there.
1434 * This is extremely effective. The average number of iterations in
1435 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1437 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
1439 struct inode
*inode
;
1440 unsigned blocks_per_page
;
1441 unsigned long page_no
;
1443 sector_t probe_block
;
1444 sector_t last_block
;
1445 sector_t lowest_block
= -1;
1446 sector_t highest_block
= 0;
1450 inode
= sis
->swap_file
->f_mapping
->host
;
1451 if (S_ISBLK(inode
->i_mode
)) {
1452 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1457 blkbits
= inode
->i_blkbits
;
1458 blocks_per_page
= PAGE_SIZE
>> blkbits
;
1461 * Map all the blocks into the extent list. This code doesn't try
1466 last_block
= i_size_read(inode
) >> blkbits
;
1467 while ((probe_block
+ blocks_per_page
) <= last_block
&&
1468 page_no
< sis
->max
) {
1469 unsigned block_in_page
;
1470 sector_t first_block
;
1472 first_block
= bmap(inode
, probe_block
);
1473 if (first_block
== 0)
1477 * It must be PAGE_SIZE aligned on-disk
1479 if (first_block
& (blocks_per_page
- 1)) {
1484 for (block_in_page
= 1; block_in_page
< blocks_per_page
;
1488 block
= bmap(inode
, probe_block
+ block_in_page
);
1491 if (block
!= first_block
+ block_in_page
) {
1498 first_block
>>= (PAGE_SHIFT
- blkbits
);
1499 if (page_no
) { /* exclude the header page */
1500 if (first_block
< lowest_block
)
1501 lowest_block
= first_block
;
1502 if (first_block
> highest_block
)
1503 highest_block
= first_block
;
1507 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1509 ret
= add_swap_extent(sis
, page_no
, 1, first_block
);
1514 probe_block
+= blocks_per_page
;
1519 *span
= 1 + highest_block
- lowest_block
;
1521 page_no
= 1; /* force Empty message */
1523 sis
->pages
= page_no
- 1;
1524 sis
->highest_bit
= page_no
- 1;
1528 printk(KERN_ERR
"swapon: swapfile has holes\n");
1533 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
1535 struct swap_info_struct
*p
= NULL
;
1536 unsigned char *swap_map
;
1537 struct file
*swap_file
, *victim
;
1538 struct address_space
*mapping
;
1539 struct inode
*inode
;
1544 if (!capable(CAP_SYS_ADMIN
))
1547 pathname
= getname(specialfile
);
1548 err
= PTR_ERR(pathname
);
1549 if (IS_ERR(pathname
))
1552 victim
= filp_open(pathname
, O_RDWR
|O_LARGEFILE
, 0);
1554 err
= PTR_ERR(victim
);
1558 mapping
= victim
->f_mapping
;
1560 spin_lock(&swap_lock
);
1561 for (type
= swap_list
.head
; type
>= 0; type
= swap_info
[type
]->next
) {
1562 p
= swap_info
[type
];
1563 if (p
->flags
& SWP_WRITEOK
) {
1564 if (p
->swap_file
->f_mapping
== mapping
)
1571 spin_unlock(&swap_lock
);
1574 if (!security_vm_enough_memory(p
->pages
))
1575 vm_unacct_memory(p
->pages
);
1578 spin_unlock(&swap_lock
);
1582 swap_list
.head
= p
->next
;
1584 swap_info
[prev
]->next
= p
->next
;
1585 if (type
== swap_list
.next
) {
1586 /* just pick something that's safe... */
1587 swap_list
.next
= swap_list
.head
;
1590 for (i
= p
->next
; i
>= 0; i
= swap_info
[i
]->next
)
1591 swap_info
[i
]->prio
= p
->prio
--;
1594 nr_swap_pages
-= p
->pages
;
1595 total_swap_pages
-= p
->pages
;
1596 p
->flags
&= ~SWP_WRITEOK
;
1597 spin_unlock(&swap_lock
);
1599 current
->flags
|= PF_OOM_ORIGIN
;
1600 err
= try_to_unuse(type
);
1601 current
->flags
&= ~PF_OOM_ORIGIN
;
1604 /* re-insert swap space back into swap_list */
1605 spin_lock(&swap_lock
);
1607 p
->prio
= --least_priority
;
1609 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
]->next
) {
1610 if (p
->prio
>= swap_info
[i
]->prio
)
1616 swap_list
.head
= swap_list
.next
= type
;
1618 swap_info
[prev
]->next
= type
;
1619 nr_swap_pages
+= p
->pages
;
1620 total_swap_pages
+= p
->pages
;
1621 p
->flags
|= SWP_WRITEOK
;
1622 spin_unlock(&swap_lock
);
1626 /* wait for any unplug function to finish */
1627 down_write(&swap_unplug_sem
);
1628 up_write(&swap_unplug_sem
);
1630 destroy_swap_extents(p
);
1631 if (p
->flags
& SWP_CONTINUED
)
1632 free_swap_count_continuations(p
);
1634 mutex_lock(&swapon_mutex
);
1635 spin_lock(&swap_lock
);
1638 /* wait for anyone still in scan_swap_map */
1639 p
->highest_bit
= 0; /* cuts scans short */
1640 while (p
->flags
>= SWP_SCANNING
) {
1641 spin_unlock(&swap_lock
);
1642 schedule_timeout_uninterruptible(1);
1643 spin_lock(&swap_lock
);
1646 swap_file
= p
->swap_file
;
1647 p
->swap_file
= NULL
;
1649 swap_map
= p
->swap_map
;
1652 spin_unlock(&swap_lock
);
1653 mutex_unlock(&swapon_mutex
);
1655 /* Destroy swap account informatin */
1656 swap_cgroup_swapoff(type
);
1658 inode
= mapping
->host
;
1659 if (S_ISBLK(inode
->i_mode
)) {
1660 struct block_device
*bdev
= I_BDEV(inode
);
1661 set_blocksize(bdev
, p
->old_block_size
);
1664 mutex_lock(&inode
->i_mutex
);
1665 inode
->i_flags
&= ~S_SWAPFILE
;
1666 mutex_unlock(&inode
->i_mutex
);
1668 filp_close(swap_file
, NULL
);
1672 filp_close(victim
, NULL
);
1677 #ifdef CONFIG_PROC_FS
1679 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
1681 struct swap_info_struct
*si
;
1685 mutex_lock(&swapon_mutex
);
1688 return SEQ_START_TOKEN
;
1690 for (type
= 0; type
< nr_swapfiles
; type
++) {
1691 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1692 si
= swap_info
[type
];
1693 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
1702 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
1704 struct swap_info_struct
*si
= v
;
1707 if (v
== SEQ_START_TOKEN
)
1710 type
= si
->type
+ 1;
1712 for (; type
< nr_swapfiles
; type
++) {
1713 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1714 si
= swap_info
[type
];
1715 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
1724 static void swap_stop(struct seq_file
*swap
, void *v
)
1726 mutex_unlock(&swapon_mutex
);
1729 static int swap_show(struct seq_file
*swap
, void *v
)
1731 struct swap_info_struct
*si
= v
;
1735 if (si
== SEQ_START_TOKEN
) {
1736 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1740 file
= si
->swap_file
;
1741 len
= seq_path(swap
, &file
->f_path
, " \t\n\\");
1742 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
1743 len
< 40 ? 40 - len
: 1, " ",
1744 S_ISBLK(file
->f_path
.dentry
->d_inode
->i_mode
) ?
1745 "partition" : "file\t",
1746 si
->pages
<< (PAGE_SHIFT
- 10),
1747 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
1752 static const struct seq_operations swaps_op
= {
1753 .start
= swap_start
,
1759 static int swaps_open(struct inode
*inode
, struct file
*file
)
1761 return seq_open(file
, &swaps_op
);
1764 static const struct file_operations proc_swaps_operations
= {
1767 .llseek
= seq_lseek
,
1768 .release
= seq_release
,
1771 static int __init
procswaps_init(void)
1773 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
1776 __initcall(procswaps_init
);
1777 #endif /* CONFIG_PROC_FS */
1779 #ifdef MAX_SWAPFILES_CHECK
1780 static int __init
max_swapfiles_check(void)
1782 MAX_SWAPFILES_CHECK();
1785 late_initcall(max_swapfiles_check
);
1789 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1791 * The swapon system call
1793 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
1795 struct swap_info_struct
*p
;
1797 struct block_device
*bdev
= NULL
;
1798 struct file
*swap_file
= NULL
;
1799 struct address_space
*mapping
;
1803 union swap_header
*swap_header
;
1804 unsigned int nr_good_pages
;
1807 unsigned long maxpages
;
1808 unsigned long swapfilepages
;
1809 unsigned char *swap_map
= NULL
;
1810 struct page
*page
= NULL
;
1811 struct inode
*inode
= NULL
;
1814 if (!capable(CAP_SYS_ADMIN
))
1817 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
1821 spin_lock(&swap_lock
);
1822 for (type
= 0; type
< nr_swapfiles
; type
++) {
1823 if (!(swap_info
[type
]->flags
& SWP_USED
))
1827 if (type
>= MAX_SWAPFILES
) {
1828 spin_unlock(&swap_lock
);
1832 if (type
>= nr_swapfiles
) {
1834 swap_info
[type
] = p
;
1836 * Write swap_info[type] before nr_swapfiles, in case a
1837 * racing procfs swap_start() or swap_next() is reading them.
1838 * (We never shrink nr_swapfiles, we never free this entry.)
1844 p
= swap_info
[type
];
1846 * Do not memset this entry: a racing procfs swap_next()
1847 * would be relying on p->type to remain valid.
1850 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
1851 p
->flags
= SWP_USED
;
1853 spin_unlock(&swap_lock
);
1855 name
= getname(specialfile
);
1856 error
= PTR_ERR(name
);
1861 swap_file
= filp_open(name
, O_RDWR
|O_LARGEFILE
, 0);
1862 error
= PTR_ERR(swap_file
);
1863 if (IS_ERR(swap_file
)) {
1868 p
->swap_file
= swap_file
;
1869 mapping
= swap_file
->f_mapping
;
1870 inode
= mapping
->host
;
1873 for (i
= 0; i
< nr_swapfiles
; i
++) {
1874 struct swap_info_struct
*q
= swap_info
[i
];
1876 if (i
== type
|| !q
->swap_file
)
1878 if (mapping
== q
->swap_file
->f_mapping
)
1883 if (S_ISBLK(inode
->i_mode
)) {
1884 bdev
= I_BDEV(inode
);
1885 error
= bd_claim(bdev
, sys_swapon
);
1891 p
->old_block_size
= block_size(bdev
);
1892 error
= set_blocksize(bdev
, PAGE_SIZE
);
1896 p
->flags
|= SWP_BLKDEV
;
1897 } else if (S_ISREG(inode
->i_mode
)) {
1898 p
->bdev
= inode
->i_sb
->s_bdev
;
1899 mutex_lock(&inode
->i_mutex
);
1901 if (IS_SWAPFILE(inode
)) {
1909 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
1912 * Read the swap header.
1914 if (!mapping
->a_ops
->readpage
) {
1918 page
= read_mapping_page(mapping
, 0, swap_file
);
1920 error
= PTR_ERR(page
);
1923 swap_header
= kmap(page
);
1925 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
1926 printk(KERN_ERR
"Unable to find swap-space signature\n");
1931 /* swap partition endianess hack... */
1932 if (swab32(swap_header
->info
.version
) == 1) {
1933 swab32s(&swap_header
->info
.version
);
1934 swab32s(&swap_header
->info
.last_page
);
1935 swab32s(&swap_header
->info
.nr_badpages
);
1936 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
1937 swab32s(&swap_header
->info
.badpages
[i
]);
1939 /* Check the swap header's sub-version */
1940 if (swap_header
->info
.version
!= 1) {
1942 "Unable to handle swap header version %d\n",
1943 swap_header
->info
.version
);
1949 p
->cluster_next
= 1;
1953 * Find out how many pages are allowed for a single swap
1954 * device. There are two limiting factors: 1) the number of
1955 * bits for the swap offset in the swp_entry_t type and
1956 * 2) the number of bits in the a swap pte as defined by
1957 * the different architectures. In order to find the
1958 * largest possible bit mask a swap entry with swap type 0
1959 * and swap offset ~0UL is created, encoded to a swap pte,
1960 * decoded to a swp_entry_t again and finally the swap
1961 * offset is extracted. This will mask all the bits from
1962 * the initial ~0UL mask that can't be encoded in either
1963 * the swp_entry_t or the architecture definition of a
1966 maxpages
= swp_offset(pte_to_swp_entry(
1967 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
1968 if (maxpages
> swap_header
->info
.last_page
) {
1969 maxpages
= swap_header
->info
.last_page
+ 1;
1970 /* p->max is an unsigned int: don't overflow it */
1971 if ((unsigned int)maxpages
== 0)
1972 maxpages
= UINT_MAX
;
1974 p
->highest_bit
= maxpages
- 1;
1979 if (swapfilepages
&& maxpages
> swapfilepages
) {
1981 "Swap area shorter than signature indicates\n");
1984 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
1986 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
1989 /* OK, set up the swap map and apply the bad block list */
1990 swap_map
= vmalloc(maxpages
);
1996 memset(swap_map
, 0, maxpages
);
1997 nr_good_pages
= maxpages
- 1; /* omit header page */
1999 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
2000 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
2001 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
) {
2005 if (page_nr
< maxpages
) {
2006 swap_map
[page_nr
] = SWAP_MAP_BAD
;
2011 error
= swap_cgroup_swapon(type
, maxpages
);
2015 if (nr_good_pages
) {
2016 swap_map
[0] = SWAP_MAP_BAD
;
2018 p
->pages
= nr_good_pages
;
2019 nr_extents
= setup_swap_extents(p
, &span
);
2020 if (nr_extents
< 0) {
2024 nr_good_pages
= p
->pages
;
2026 if (!nr_good_pages
) {
2027 printk(KERN_WARNING
"Empty swap-file\n");
2033 if (blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
2034 p
->flags
|= SWP_SOLIDSTATE
;
2035 p
->cluster_next
= 1 + (random32() % p
->highest_bit
);
2037 if (discard_swap(p
) == 0)
2038 p
->flags
|= SWP_DISCARDABLE
;
2041 mutex_lock(&swapon_mutex
);
2042 spin_lock(&swap_lock
);
2043 if (swap_flags
& SWAP_FLAG_PREFER
)
2045 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
2047 p
->prio
= --least_priority
;
2048 p
->swap_map
= swap_map
;
2049 p
->flags
|= SWP_WRITEOK
;
2050 nr_swap_pages
+= nr_good_pages
;
2051 total_swap_pages
+= nr_good_pages
;
2053 printk(KERN_INFO
"Adding %uk swap on %s. "
2054 "Priority:%d extents:%d across:%lluk %s%s\n",
2055 nr_good_pages
<<(PAGE_SHIFT
-10), name
, p
->prio
,
2056 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
2057 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
2058 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "");
2060 /* insert swap space into swap_list: */
2062 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
]->next
) {
2063 if (p
->prio
>= swap_info
[i
]->prio
)
2069 swap_list
.head
= swap_list
.next
= type
;
2071 swap_info
[prev
]->next
= type
;
2072 spin_unlock(&swap_lock
);
2073 mutex_unlock(&swapon_mutex
);
2078 set_blocksize(bdev
, p
->old_block_size
);
2081 destroy_swap_extents(p
);
2082 swap_cgroup_swapoff(type
);
2084 spin_lock(&swap_lock
);
2085 p
->swap_file
= NULL
;
2087 spin_unlock(&swap_lock
);
2090 filp_close(swap_file
, NULL
);
2092 if (page
&& !IS_ERR(page
)) {
2094 page_cache_release(page
);
2100 inode
->i_flags
|= S_SWAPFILE
;
2101 mutex_unlock(&inode
->i_mutex
);
2106 void si_swapinfo(struct sysinfo
*val
)
2109 unsigned long nr_to_be_unused
= 0;
2111 spin_lock(&swap_lock
);
2112 for (type
= 0; type
< nr_swapfiles
; type
++) {
2113 struct swap_info_struct
*si
= swap_info
[type
];
2115 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
2116 nr_to_be_unused
+= si
->inuse_pages
;
2118 val
->freeswap
= nr_swap_pages
+ nr_to_be_unused
;
2119 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
2120 spin_unlock(&swap_lock
);
2124 * Verify that a swap entry is valid and increment its swap map count.
2126 * Returns error code in following case.
2128 * - swp_entry is invalid -> EINVAL
2129 * - swp_entry is migration entry -> EINVAL
2130 * - swap-cache reference is requested but there is already one. -> EEXIST
2131 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2132 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2134 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
2136 struct swap_info_struct
*p
;
2137 unsigned long offset
, type
;
2138 unsigned char count
;
2139 unsigned char has_cache
;
2142 if (non_swap_entry(entry
))
2145 type
= swp_type(entry
);
2146 if (type
>= nr_swapfiles
)
2148 p
= swap_info
[type
];
2149 offset
= swp_offset(entry
);
2151 spin_lock(&swap_lock
);
2152 if (unlikely(offset
>= p
->max
))
2155 count
= p
->swap_map
[offset
];
2156 has_cache
= count
& SWAP_HAS_CACHE
;
2157 count
&= ~SWAP_HAS_CACHE
;
2160 if (usage
== SWAP_HAS_CACHE
) {
2162 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2163 if (!has_cache
&& count
)
2164 has_cache
= SWAP_HAS_CACHE
;
2165 else if (has_cache
) /* someone else added cache */
2167 else /* no users remaining */
2170 } else if (count
|| has_cache
) {
2172 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
2174 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
2176 else if (swap_count_continued(p
, offset
, count
))
2177 count
= COUNT_CONTINUED
;
2181 err
= -ENOENT
; /* unused swap entry */
2183 p
->swap_map
[offset
] = count
| has_cache
;
2186 spin_unlock(&swap_lock
);
2191 printk(KERN_ERR
"swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
2196 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2197 * (in which case its reference count is never incremented).
2199 void swap_shmem_alloc(swp_entry_t entry
)
2201 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
2205 * Increase reference count of swap entry by 1.
2206 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2207 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2208 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2209 * might occur if a page table entry has got corrupted.
2211 int swap_duplicate(swp_entry_t entry
)
2215 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
2216 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
2221 * @entry: swap entry for which we allocate swap cache.
2223 * Called when allocating swap cache for existing swap entry,
2224 * This can return error codes. Returns 0 at success.
2225 * -EBUSY means there is a swap cache.
2226 * Note: return code is different from swap_duplicate().
2228 int swapcache_prepare(swp_entry_t entry
)
2230 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
2234 * swap_lock prevents swap_map being freed. Don't grab an extra
2235 * reference on the swaphandle, it doesn't matter if it becomes unused.
2237 int valid_swaphandles(swp_entry_t entry
, unsigned long *offset
)
2239 struct swap_info_struct
*si
;
2240 int our_page_cluster
= page_cluster
;
2241 pgoff_t target
, toff
;
2245 if (!our_page_cluster
) /* no readahead */
2248 si
= swap_info
[swp_type(entry
)];
2249 target
= swp_offset(entry
);
2250 base
= (target
>> our_page_cluster
) << our_page_cluster
;
2251 end
= base
+ (1 << our_page_cluster
);
2252 if (!base
) /* first page is swap header */
2255 spin_lock(&swap_lock
);
2256 if (end
> si
->max
) /* don't go beyond end of map */
2259 /* Count contiguous allocated slots above our target */
2260 for (toff
= target
; ++toff
< end
; nr_pages
++) {
2261 /* Don't read in free or bad pages */
2262 if (!si
->swap_map
[toff
])
2264 if (swap_count(si
->swap_map
[toff
]) == SWAP_MAP_BAD
)
2267 /* Count contiguous allocated slots below our target */
2268 for (toff
= target
; --toff
>= base
; nr_pages
++) {
2269 /* Don't read in free or bad pages */
2270 if (!si
->swap_map
[toff
])
2272 if (swap_count(si
->swap_map
[toff
]) == SWAP_MAP_BAD
)
2275 spin_unlock(&swap_lock
);
2278 * Indicate starting offset, and return number of pages to get:
2279 * if only 1, say 0, since there's then no readahead to be done.
2282 return nr_pages
? ++nr_pages
: 0;
2286 * add_swap_count_continuation - called when a swap count is duplicated
2287 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2288 * page of the original vmalloc'ed swap_map, to hold the continuation count
2289 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2290 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2292 * These continuation pages are seldom referenced: the common paths all work
2293 * on the original swap_map, only referring to a continuation page when the
2294 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2296 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2297 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2298 * can be called after dropping locks.
2300 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
2302 struct swap_info_struct
*si
;
2305 struct page
*list_page
;
2307 unsigned char count
;
2310 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2311 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2313 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
2315 si
= swap_info_get(entry
);
2318 * An acceptable race has occurred since the failing
2319 * __swap_duplicate(): the swap entry has been freed,
2320 * perhaps even the whole swap_map cleared for swapoff.
2325 offset
= swp_offset(entry
);
2326 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
2328 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
2330 * The higher the swap count, the more likely it is that tasks
2331 * will race to add swap count continuation: we need to avoid
2332 * over-provisioning.
2338 spin_unlock(&swap_lock
);
2343 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2344 * no architecture is using highmem pages for kernel pagetables: so it
2345 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2347 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2348 offset
&= ~PAGE_MASK
;
2351 * Page allocation does not initialize the page's lru field,
2352 * but it does always reset its private field.
2354 if (!page_private(head
)) {
2355 BUG_ON(count
& COUNT_CONTINUED
);
2356 INIT_LIST_HEAD(&head
->lru
);
2357 set_page_private(head
, SWP_CONTINUED
);
2358 si
->flags
|= SWP_CONTINUED
;
2361 list_for_each_entry(list_page
, &head
->lru
, lru
) {
2365 * If the previous map said no continuation, but we've found
2366 * a continuation page, free our allocation and use this one.
2368 if (!(count
& COUNT_CONTINUED
))
2371 map
= kmap_atomic(list_page
, KM_USER0
) + offset
;
2373 kunmap_atomic(map
, KM_USER0
);
2376 * If this continuation count now has some space in it,
2377 * free our allocation and use this one.
2379 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
2383 list_add_tail(&page
->lru
, &head
->lru
);
2384 page
= NULL
; /* now it's attached, don't free it */
2386 spin_unlock(&swap_lock
);
2394 * swap_count_continued - when the original swap_map count is incremented
2395 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2396 * into, carry if so, or else fail until a new continuation page is allocated;
2397 * when the original swap_map count is decremented from 0 with continuation,
2398 * borrow from the continuation and report whether it still holds more.
2399 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2401 static bool swap_count_continued(struct swap_info_struct
*si
,
2402 pgoff_t offset
, unsigned char count
)
2408 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2409 if (page_private(head
) != SWP_CONTINUED
) {
2410 BUG_ON(count
& COUNT_CONTINUED
);
2411 return false; /* need to add count continuation */
2414 offset
&= ~PAGE_MASK
;
2415 page
= list_entry(head
->lru
.next
, struct page
, lru
);
2416 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2418 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
2419 goto init_map
; /* jump over SWAP_CONT_MAX checks */
2421 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
2423 * Think of how you add 1 to 999
2425 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
2426 kunmap_atomic(map
, KM_USER0
);
2427 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2428 BUG_ON(page
== head
);
2429 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2431 if (*map
== SWAP_CONT_MAX
) {
2432 kunmap_atomic(map
, KM_USER0
);
2433 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2435 return false; /* add count continuation */
2436 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2437 init_map
: *map
= 0; /* we didn't zero the page */
2440 kunmap_atomic(map
, KM_USER0
);
2441 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2442 while (page
!= head
) {
2443 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2444 *map
= COUNT_CONTINUED
;
2445 kunmap_atomic(map
, KM_USER0
);
2446 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2448 return true; /* incremented */
2450 } else { /* decrementing */
2452 * Think of how you subtract 1 from 1000
2454 BUG_ON(count
!= COUNT_CONTINUED
);
2455 while (*map
== COUNT_CONTINUED
) {
2456 kunmap_atomic(map
, KM_USER0
);
2457 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2458 BUG_ON(page
== head
);
2459 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2465 kunmap_atomic(map
, KM_USER0
);
2466 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2467 while (page
!= head
) {
2468 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2469 *map
= SWAP_CONT_MAX
| count
;
2470 count
= COUNT_CONTINUED
;
2471 kunmap_atomic(map
, KM_USER0
);
2472 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2474 return count
== COUNT_CONTINUED
;
2479 * free_swap_count_continuations - swapoff free all the continuation pages
2480 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2482 static void free_swap_count_continuations(struct swap_info_struct
*si
)
2486 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
2488 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2489 if (page_private(head
)) {
2490 struct list_head
*this, *next
;
2491 list_for_each_safe(this, next
, &head
->lru
) {
2493 page
= list_entry(this, struct page
, lru
);