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/export.h>
13 #include <linux/kernel_stat.h>
14 #include <linux/swap.h>
15 #include <linux/vmalloc.h>
16 #include <linux/pagemap.h>
17 #include <linux/namei.h>
18 #include <linux/shmem_fs.h>
19 #include <linux/blkdev.h>
20 #include <linux/random.h>
21 #include <linux/writeback.h>
22 #include <linux/proc_fs.h>
23 #include <linux/seq_file.h>
24 #include <linux/init.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>
33 #include <linux/poll.h>
34 #include <linux/oom.h>
35 #include <linux/frontswap.h>
36 #include <linux/swapfile.h>
37 #include <linux/export.h>
39 #include <asm/pgtable.h>
40 #include <asm/tlbflush.h>
41 #include <linux/swapops.h>
42 #include <linux/page_cgroup.h>
44 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
46 static void free_swap_count_continuations(struct swap_info_struct
*);
48 DEFINE_SPINLOCK(swap_lock
);
49 static unsigned int nr_swapfiles
;
50 atomic_long_t nr_swap_pages
;
51 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
52 long total_swap_pages
;
53 static int least_priority
;
54 static atomic_t highest_priority_index
= ATOMIC_INIT(-1);
56 static const char Bad_file
[] = "Bad swap file entry ";
57 static const char Unused_file
[] = "Unused swap file entry ";
58 static const char Bad_offset
[] = "Bad swap offset entry ";
59 static const char Unused_offset
[] = "Unused swap offset entry ";
61 struct swap_list_t swap_list
= {-1, -1};
63 struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
65 static DEFINE_MUTEX(swapon_mutex
);
67 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
68 /* Activity counter to indicate that a swapon or swapoff has occurred */
69 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
71 static inline unsigned char swap_count(unsigned char ent
)
73 return ent
& ~SWAP_HAS_CACHE
; /* may include SWAP_HAS_CONT flag */
76 /* returns 1 if swap entry is freed */
78 __try_to_reclaim_swap(struct swap_info_struct
*si
, unsigned long offset
)
80 swp_entry_t entry
= swp_entry(si
->type
, offset
);
84 page
= find_get_page(swap_address_space(entry
), entry
.val
);
88 * This function is called from scan_swap_map() and it's called
89 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
90 * We have to use trylock for avoiding deadlock. This is a special
91 * case and you should use try_to_free_swap() with explicit lock_page()
92 * in usual operations.
94 if (trylock_page(page
)) {
95 ret
= try_to_free_swap(page
);
98 page_cache_release(page
);
103 * swapon tell device that all the old swap contents can be discarded,
104 * to allow the swap device to optimize its wear-levelling.
106 static int discard_swap(struct swap_info_struct
*si
)
108 struct swap_extent
*se
;
109 sector_t start_block
;
113 /* Do not discard the swap header page! */
114 se
= &si
->first_swap_extent
;
115 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
116 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
118 err
= blkdev_issue_discard(si
->bdev
, start_block
,
119 nr_blocks
, GFP_KERNEL
, 0);
125 list_for_each_entry(se
, &si
->first_swap_extent
.list
, list
) {
126 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
127 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
129 err
= blkdev_issue_discard(si
->bdev
, start_block
,
130 nr_blocks
, GFP_KERNEL
, 0);
136 return err
; /* That will often be -EOPNOTSUPP */
140 * swap allocation tell device that a cluster of swap can now be discarded,
141 * to allow the swap device to optimize its wear-levelling.
143 static void discard_swap_cluster(struct swap_info_struct
*si
,
144 pgoff_t start_page
, pgoff_t nr_pages
)
146 struct swap_extent
*se
= si
->curr_swap_extent
;
147 int found_extent
= 0;
150 struct list_head
*lh
;
152 if (se
->start_page
<= start_page
&&
153 start_page
< se
->start_page
+ se
->nr_pages
) {
154 pgoff_t offset
= start_page
- se
->start_page
;
155 sector_t start_block
= se
->start_block
+ offset
;
156 sector_t nr_blocks
= se
->nr_pages
- offset
;
158 if (nr_blocks
> nr_pages
)
159 nr_blocks
= nr_pages
;
160 start_page
+= nr_blocks
;
161 nr_pages
-= nr_blocks
;
164 si
->curr_swap_extent
= se
;
166 start_block
<<= PAGE_SHIFT
- 9;
167 nr_blocks
<<= PAGE_SHIFT
- 9;
168 if (blkdev_issue_discard(si
->bdev
, start_block
,
169 nr_blocks
, GFP_NOIO
, 0))
174 se
= list_entry(lh
, struct swap_extent
, list
);
178 static int wait_for_discard(void *word
)
184 #define SWAPFILE_CLUSTER 256
185 #define LATENCY_LIMIT 256
187 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
190 unsigned long offset
;
191 unsigned long scan_base
;
192 unsigned long last_in_cluster
= 0;
193 int latency_ration
= LATENCY_LIMIT
;
194 int found_free_cluster
= 0;
197 * We try to cluster swap pages by allocating them sequentially
198 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
199 * way, however, we resort to first-free allocation, starting
200 * a new cluster. This prevents us from scattering swap pages
201 * all over the entire swap partition, so that we reduce
202 * overall disk seek times between swap pages. -- sct
203 * But we do now try to find an empty cluster. -Andrea
204 * And we let swap pages go all over an SSD partition. Hugh
207 si
->flags
+= SWP_SCANNING
;
208 scan_base
= offset
= si
->cluster_next
;
210 if (unlikely(!si
->cluster_nr
--)) {
211 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
212 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
215 if (si
->flags
& SWP_DISCARDABLE
) {
217 * Start range check on racing allocations, in case
218 * they overlap the cluster we eventually decide on
219 * (we scan without swap_lock to allow preemption).
220 * It's hardly conceivable that cluster_nr could be
221 * wrapped during our scan, but don't depend on it.
223 if (si
->lowest_alloc
)
225 si
->lowest_alloc
= si
->max
;
226 si
->highest_alloc
= 0;
228 spin_unlock(&si
->lock
);
231 * If seek is expensive, start searching for new cluster from
232 * start of partition, to minimize the span of allocated swap.
233 * But if seek is cheap, search from our current position, so
234 * that swap is allocated from all over the partition: if the
235 * Flash Translation Layer only remaps within limited zones,
236 * we don't want to wear out the first zone too quickly.
238 if (!(si
->flags
& SWP_SOLIDSTATE
))
239 scan_base
= offset
= si
->lowest_bit
;
240 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
242 /* Locate the first empty (unaligned) cluster */
243 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
244 if (si
->swap_map
[offset
])
245 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
246 else if (offset
== last_in_cluster
) {
247 spin_lock(&si
->lock
);
248 offset
-= SWAPFILE_CLUSTER
- 1;
249 si
->cluster_next
= offset
;
250 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
251 found_free_cluster
= 1;
254 if (unlikely(--latency_ration
< 0)) {
256 latency_ration
= LATENCY_LIMIT
;
260 offset
= si
->lowest_bit
;
261 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
263 /* Locate the first empty (unaligned) cluster */
264 for (; last_in_cluster
< scan_base
; offset
++) {
265 if (si
->swap_map
[offset
])
266 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
267 else if (offset
== last_in_cluster
) {
268 spin_lock(&si
->lock
);
269 offset
-= SWAPFILE_CLUSTER
- 1;
270 si
->cluster_next
= offset
;
271 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
272 found_free_cluster
= 1;
275 if (unlikely(--latency_ration
< 0)) {
277 latency_ration
= LATENCY_LIMIT
;
282 spin_lock(&si
->lock
);
283 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
284 si
->lowest_alloc
= 0;
288 if (!(si
->flags
& SWP_WRITEOK
))
290 if (!si
->highest_bit
)
292 if (offset
> si
->highest_bit
)
293 scan_base
= offset
= si
->lowest_bit
;
295 /* reuse swap entry of cache-only swap if not busy. */
296 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
298 spin_unlock(&si
->lock
);
299 swap_was_freed
= __try_to_reclaim_swap(si
, offset
);
300 spin_lock(&si
->lock
);
301 /* entry was freed successfully, try to use this again */
304 goto scan
; /* check next one */
307 if (si
->swap_map
[offset
])
310 if (offset
== si
->lowest_bit
)
312 if (offset
== si
->highest_bit
)
315 if (si
->inuse_pages
== si
->pages
) {
316 si
->lowest_bit
= si
->max
;
319 si
->swap_map
[offset
] = usage
;
320 si
->cluster_next
= offset
+ 1;
321 si
->flags
-= SWP_SCANNING
;
323 if (si
->lowest_alloc
) {
325 * Only set when SWP_DISCARDABLE, and there's a scan
326 * for a free cluster in progress or just completed.
328 if (found_free_cluster
) {
330 * To optimize wear-levelling, discard the
331 * old data of the cluster, taking care not to
332 * discard any of its pages that have already
333 * been allocated by racing tasks (offset has
334 * already stepped over any at the beginning).
336 if (offset
< si
->highest_alloc
&&
337 si
->lowest_alloc
<= last_in_cluster
)
338 last_in_cluster
= si
->lowest_alloc
- 1;
339 si
->flags
|= SWP_DISCARDING
;
340 spin_unlock(&si
->lock
);
342 if (offset
< last_in_cluster
)
343 discard_swap_cluster(si
, offset
,
344 last_in_cluster
- offset
+ 1);
346 spin_lock(&si
->lock
);
347 si
->lowest_alloc
= 0;
348 si
->flags
&= ~SWP_DISCARDING
;
350 smp_mb(); /* wake_up_bit advises this */
351 wake_up_bit(&si
->flags
, ilog2(SWP_DISCARDING
));
353 } else if (si
->flags
& SWP_DISCARDING
) {
355 * Delay using pages allocated by racing tasks
356 * until the whole discard has been issued. We
357 * could defer that delay until swap_writepage,
358 * but it's easier to keep this self-contained.
360 spin_unlock(&si
->lock
);
361 wait_on_bit(&si
->flags
, ilog2(SWP_DISCARDING
),
362 wait_for_discard
, TASK_UNINTERRUPTIBLE
);
363 spin_lock(&si
->lock
);
366 * Note pages allocated by racing tasks while
367 * scan for a free cluster is in progress, so
368 * that its final discard can exclude them.
370 if (offset
< si
->lowest_alloc
)
371 si
->lowest_alloc
= offset
;
372 if (offset
> si
->highest_alloc
)
373 si
->highest_alloc
= offset
;
379 spin_unlock(&si
->lock
);
380 while (++offset
<= si
->highest_bit
) {
381 if (!si
->swap_map
[offset
]) {
382 spin_lock(&si
->lock
);
385 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
386 spin_lock(&si
->lock
);
389 if (unlikely(--latency_ration
< 0)) {
391 latency_ration
= LATENCY_LIMIT
;
394 offset
= si
->lowest_bit
;
395 while (++offset
< scan_base
) {
396 if (!si
->swap_map
[offset
]) {
397 spin_lock(&si
->lock
);
400 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
401 spin_lock(&si
->lock
);
404 if (unlikely(--latency_ration
< 0)) {
406 latency_ration
= LATENCY_LIMIT
;
409 spin_lock(&si
->lock
);
412 si
->flags
-= SWP_SCANNING
;
416 swp_entry_t
get_swap_page(void)
418 struct swap_info_struct
*si
;
424 spin_lock(&swap_lock
);
425 if (atomic_long_read(&nr_swap_pages
) <= 0)
427 atomic_long_dec(&nr_swap_pages
);
429 for (type
= swap_list
.next
; type
>= 0 && wrapped
< 2; type
= next
) {
430 hp_index
= atomic_xchg(&highest_priority_index
, -1);
432 * highest_priority_index records current highest priority swap
433 * type which just frees swap entries. If its priority is
434 * higher than that of swap_list.next swap type, we use it. It
435 * isn't protected by swap_lock, so it can be an invalid value
436 * if the corresponding swap type is swapoff. We double check
437 * the flags here. It's even possible the swap type is swapoff
438 * and swapon again and its priority is changed. In such rare
439 * case, low prority swap type might be used, but eventually
440 * high priority swap will be used after several rounds of
443 if (hp_index
!= -1 && hp_index
!= type
&&
444 swap_info
[type
]->prio
< swap_info
[hp_index
]->prio
&&
445 (swap_info
[hp_index
]->flags
& SWP_WRITEOK
)) {
447 swap_list
.next
= type
;
450 si
= swap_info
[type
];
453 (!wrapped
&& si
->prio
!= swap_info
[next
]->prio
)) {
454 next
= swap_list
.head
;
458 spin_lock(&si
->lock
);
459 if (!si
->highest_bit
) {
460 spin_unlock(&si
->lock
);
463 if (!(si
->flags
& SWP_WRITEOK
)) {
464 spin_unlock(&si
->lock
);
468 swap_list
.next
= next
;
470 spin_unlock(&swap_lock
);
471 /* This is called for allocating swap entry for cache */
472 offset
= scan_swap_map(si
, SWAP_HAS_CACHE
);
473 spin_unlock(&si
->lock
);
475 return swp_entry(type
, offset
);
476 spin_lock(&swap_lock
);
477 next
= swap_list
.next
;
480 atomic_long_inc(&nr_swap_pages
);
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 si
= swap_info
[type
];
493 spin_lock(&si
->lock
);
494 if (si
&& (si
->flags
& SWP_WRITEOK
)) {
495 atomic_long_dec(&nr_swap_pages
);
496 /* This is called for allocating swap entry, not cache */
497 offset
= scan_swap_map(si
, 1);
499 spin_unlock(&si
->lock
);
500 return swp_entry(type
, offset
);
502 atomic_long_inc(&nr_swap_pages
);
504 spin_unlock(&si
->lock
);
505 return (swp_entry_t
) {0};
507 EXPORT_SYMBOL_GPL(get_swap_page_of_type
);
510 swp_entry_t
get_swap_page_by_memcg(struct page
*page
)
512 struct swap_info_struct
*si
;
513 int type
, next
, wrapped
;
516 /* Go to original get_swap_page if we have only 1 or no swap area. */
517 if (nr_swapfiles
<= 1)
518 return get_swap_page();
520 spin_lock(&swap_lock
);
521 if (atomic_long_read(&nr_swap_pages
) <= 0)
523 atomic_long_dec(&nr_swap_pages
);
526 if (memcg_is_root(page
)) {
527 /* High to low priority */
529 type
= swap_list
.head
;
531 /* Low to high priority */
533 si
= swap_info
[swap_list
.head
];
537 /* Scan for an empty swap entry */
538 for (; (type
>= 0) || (wrapped
== 0); type
= next
) {
541 next
= swap_list
.head
;
545 si
= swap_info
[type
];
547 spin_lock(&si
->lock
);
548 if (!si
->highest_bit
) {
549 spin_unlock(&si
->lock
);
552 if (!(si
->flags
& SWP_WRITEOK
)) {
553 spin_unlock(&si
->lock
);
556 spin_unlock(&swap_lock
);
557 /* This is called for allocating swap entry for cache */
558 offset
= scan_swap_map(si
, SWAP_HAS_CACHE
);
559 spin_unlock(&si
->lock
);
561 return swp_entry(type
, offset
);
562 spin_lock(&swap_lock
);
565 atomic_long_inc(&nr_swap_pages
);
567 spin_unlock(&swap_lock
);
568 return (swp_entry_t
) {0};
572 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
573 unsigned int prev
, bool frontswap
);
575 void get_swap_range_of_type(int type
, swp_entry_t
*start
, swp_entry_t
*end
,
578 struct swap_info_struct
*si
;
582 *start
= swp_entry(0, 0);
583 *end
= swp_entry(0, 0);
584 si
= swap_info
[type
];
585 spin_lock(&si
->lock
);
586 if (si
&& (si
->flags
& SWP_WRITEOK
)) {
587 /* This is called for allocating swap entry, not cache */
588 start_at
= scan_swap_map(si
, 1);
590 unsigned long stop_at
= find_next_to_unuse(si
, start_at
, 0);
591 if (stop_at
> start_at
)
594 stop_at
= si
->max
- 1;
595 if (stop_at
- start_at
+ 1 > limit
)
596 stop_at
= min_t(unsigned int,
597 start_at
+ limit
- 1,
600 for (i
= start_at
; i
<= stop_at
; i
++)
602 /* first page already done above */
603 si
->inuse_pages
+= stop_at
- start_at
;
605 atomic_long_sub(stop_at
- start_at
+ 1, &nr_swap_pages
);
606 if (start_at
+ 1 == si
->lowest_bit
)
607 si
->lowest_bit
= stop_at
+ 1;
608 if (si
->inuse_pages
== si
->pages
) {
609 si
->lowest_bit
= si
->max
;
612 si
->cluster_next
= stop_at
+ 1;
613 *start
= swp_entry(type
, start_at
);
614 *end
= swp_entry(type
, stop_at
);
617 spin_unlock(&si
->lock
);
619 EXPORT_SYMBOL_GPL(get_swap_range_of_type
);
621 struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
623 struct swap_info_struct
*p
;
624 unsigned long offset
, type
;
628 type
= swp_type(entry
);
629 if (type
>= nr_swapfiles
)
632 if (!(p
->flags
& SWP_USED
))
634 offset
= swp_offset(entry
);
635 if (offset
>= p
->max
)
637 if (!p
->swap_map
[offset
])
643 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_offset
, entry
.val
);
646 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_offset
, entry
.val
);
649 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_file
, entry
.val
);
652 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_file
, entry
.val
);
657 void swap_info_unlock(struct swap_info_struct
*si
)
659 spin_unlock(&si
->lock
);
663 * This swap type frees swap entry, check if it is the highest priority swap
664 * type which just frees swap entry. get_swap_page() uses
665 * highest_priority_index to search highest priority swap type. The
666 * swap_info_struct.lock can't protect us if there are multiple swap types
667 * active, so we use atomic_cmpxchg.
669 static void set_highest_priority_index(int type
)
671 int old_hp_index
, new_hp_index
;
674 old_hp_index
= atomic_read(&highest_priority_index
);
675 if (old_hp_index
!= -1 &&
676 swap_info
[old_hp_index
]->prio
>= swap_info
[type
]->prio
)
679 } while (atomic_cmpxchg(&highest_priority_index
,
680 old_hp_index
, new_hp_index
) != old_hp_index
);
683 static unsigned char swap_entry_free(struct swap_info_struct
*p
,
684 swp_entry_t entry
, unsigned char usage
)
686 unsigned long offset
= swp_offset(entry
);
688 unsigned char has_cache
;
690 count
= p
->swap_map
[offset
];
691 has_cache
= count
& SWAP_HAS_CACHE
;
692 count
&= ~SWAP_HAS_CACHE
;
694 if (usage
== SWAP_HAS_CACHE
) {
695 VM_BUG_ON(!has_cache
);
697 } else if (count
== SWAP_MAP_SHMEM
) {
699 * Or we could insist on shmem.c using a special
700 * swap_shmem_free() and free_shmem_swap_and_cache()...
703 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
704 if (count
== COUNT_CONTINUED
) {
705 if (swap_count_continued(p
, offset
, count
))
706 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
708 count
= SWAP_MAP_MAX
;
714 mem_cgroup_uncharge_swap(entry
);
716 usage
= count
| has_cache
;
717 p
->swap_map
[offset
] = usage
;
719 /* free if no reference */
721 if (offset
< p
->lowest_bit
)
722 p
->lowest_bit
= offset
;
723 if (offset
> p
->highest_bit
)
724 p
->highest_bit
= offset
;
725 set_highest_priority_index(p
->type
);
726 atomic_long_inc(&nr_swap_pages
);
728 frontswap_invalidate_page(p
->type
, offset
);
729 if (p
->flags
& SWP_BLKDEV
) {
730 struct gendisk
*disk
= p
->bdev
->bd_disk
;
731 if (disk
->fops
->swap_slot_free_notify
)
732 disk
->fops
->swap_slot_free_notify(p
->bdev
,
741 * Caller has made sure that the swapdevice corresponding to entry
742 * is still around or has not been recycled.
744 void swap_free(swp_entry_t entry
)
746 struct swap_info_struct
*p
;
748 p
= swap_info_get(entry
);
750 swap_entry_free(p
, entry
, 1);
751 spin_unlock(&p
->lock
);
756 * Called after dropping swapcache to decrease refcnt to swap entries.
758 void swapcache_free(swp_entry_t entry
, struct page
*page
)
760 struct swap_info_struct
*p
;
763 p
= swap_info_get(entry
);
765 count
= swap_entry_free(p
, entry
, SWAP_HAS_CACHE
);
767 mem_cgroup_uncharge_swapcache(page
, entry
, count
!= 0);
768 spin_unlock(&p
->lock
);
771 EXPORT_SYMBOL_GPL(swap_free
);
774 * How many references to page are currently swapped out?
775 * This does not give an exact answer when swap count is continued,
776 * but does include the high COUNT_CONTINUED flag to allow for that.
778 int page_swapcount(struct page
*page
)
781 struct swap_info_struct
*p
;
784 entry
.val
= page_private(page
);
785 p
= swap_info_get(entry
);
787 count
= swap_count(p
->swap_map
[swp_offset(entry
)]);
788 spin_unlock(&p
->lock
);
794 * We can write to an anon page without COW if there are no other references
795 * to it. And as a side-effect, free up its swap: because the old content
796 * on disk will never be read, and seeking back there to write new content
797 * later would only waste time away from clustering.
799 int reuse_swap_page(struct page
*page
)
803 VM_BUG_ON(!PageLocked(page
));
804 if (unlikely(PageKsm(page
)))
806 count
= page_mapcount(page
);
807 if (count
<= 1 && PageSwapCache(page
)) {
808 count
+= page_swapcount(page
);
809 if (count
== 1 && !PageWriteback(page
)) {
810 delete_from_swap_cache(page
);
818 * If swap is getting full, or if there are no more mappings of this page,
819 * then try_to_free_swap is called to free its swap space.
821 int try_to_free_swap(struct page
*page
)
823 VM_BUG_ON(!PageLocked(page
));
825 if (!PageSwapCache(page
))
827 if (PageWriteback(page
))
829 if (page_swapcount(page
))
833 * Once hibernation has begun to create its image of memory,
834 * there's a danger that one of the calls to try_to_free_swap()
835 * - most probably a call from __try_to_reclaim_swap() while
836 * hibernation is allocating its own swap pages for the image,
837 * but conceivably even a call from memory reclaim - will free
838 * the swap from a page which has already been recorded in the
839 * image as a clean swapcache page, and then reuse its swap for
840 * another page of the image. On waking from hibernation, the
841 * original page might be freed under memory pressure, then
842 * later read back in from swap, now with the wrong data.
844 * Hibration suspends storage while it is writing the image
845 * to disk so check that here.
847 if (pm_suspended_storage())
850 delete_from_swap_cache(page
);
856 * Free the swap entry like above, but also try to
857 * free the page cache entry if it is the last user.
859 int free_swap_and_cache(swp_entry_t entry
)
861 struct swap_info_struct
*p
;
862 struct page
*page
= NULL
;
864 if (non_swap_entry(entry
))
867 p
= swap_info_get(entry
);
869 if (swap_entry_free(p
, entry
, 1) == SWAP_HAS_CACHE
) {
870 page
= find_get_page(swap_address_space(entry
),
872 if (page
&& !trylock_page(page
)) {
873 page_cache_release(page
);
877 spin_unlock(&p
->lock
);
881 * Not mapped elsewhere, or swap space full? Free it!
882 * Also recheck PageSwapCache now page is locked (above).
884 if (PageSwapCache(page
) && !PageWriteback(page
) &&
885 (!page_mapped(page
) || vm_swap_full())) {
886 delete_from_swap_cache(page
);
890 page_cache_release(page
);
895 #ifdef CONFIG_HIBERNATION
897 * Find the swap type that corresponds to given device (if any).
899 * @offset - number of the PAGE_SIZE-sized block of the device, starting
900 * from 0, in which the swap header is expected to be located.
902 * This is needed for the suspend to disk (aka swsusp).
904 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
906 struct block_device
*bdev
= NULL
;
910 bdev
= bdget(device
);
912 spin_lock(&swap_lock
);
913 for (type
= 0; type
< nr_swapfiles
; type
++) {
914 struct swap_info_struct
*sis
= swap_info
[type
];
916 if (!(sis
->flags
& SWP_WRITEOK
))
921 *bdev_p
= bdgrab(sis
->bdev
);
923 spin_unlock(&swap_lock
);
926 if (bdev
== sis
->bdev
) {
927 struct swap_extent
*se
= &sis
->first_swap_extent
;
929 if (se
->start_block
== offset
) {
931 *bdev_p
= bdgrab(sis
->bdev
);
933 spin_unlock(&swap_lock
);
939 spin_unlock(&swap_lock
);
947 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
948 * corresponding to given index in swap_info (swap type).
950 sector_t
swapdev_block(int type
, pgoff_t offset
)
952 struct block_device
*bdev
;
954 if ((unsigned int)type
>= nr_swapfiles
)
956 if (!(swap_info
[type
]->flags
& SWP_WRITEOK
))
958 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
962 * Return either the total number of swap pages of given type, or the number
963 * of free pages of that type (depending on @free)
965 * This is needed for software suspend
967 unsigned int count_swap_pages(int type
, int free
)
971 spin_lock(&swap_lock
);
972 if ((unsigned int)type
< nr_swapfiles
) {
973 struct swap_info_struct
*sis
= swap_info
[type
];
975 spin_lock(&sis
->lock
);
976 if (sis
->flags
& SWP_WRITEOK
) {
979 n
-= sis
->inuse_pages
;
981 spin_unlock(&sis
->lock
);
983 spin_unlock(&swap_lock
);
986 #endif /* CONFIG_HIBERNATION */
989 * No need to decide whether this PTE shares the swap entry with others,
990 * just let do_wp_page work it out if a write is requested later - to
991 * force COW, vm_page_prot omits write permission from any private vma.
993 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
994 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
996 struct page
*swapcache
;
997 struct mem_cgroup
*memcg
;
1003 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1004 if (unlikely(!page
))
1007 if (mem_cgroup_try_charge_swapin(vma
->vm_mm
, page
,
1008 GFP_KERNEL
, &memcg
)) {
1013 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1014 if (unlikely(!pte_same(*pte
, swp_entry_to_pte(entry
)))) {
1015 mem_cgroup_cancel_charge_swapin(memcg
);
1020 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1021 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1023 set_pte_at(vma
->vm_mm
, addr
, pte
,
1024 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1025 if (page
== swapcache
)
1026 page_add_anon_rmap(page
, vma
, addr
);
1027 else /* ksm created a completely new copy */
1028 page_add_new_anon_rmap(page
, vma
, addr
);
1029 mem_cgroup_commit_charge_swapin(page
, memcg
);
1032 * Move the page to the active list so it is not
1033 * immediately swapped out again after swapon.
1035 activate_page(page
);
1037 pte_unmap_unlock(pte
, ptl
);
1039 if (page
!= swapcache
) {
1046 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1047 unsigned long addr
, unsigned long end
,
1048 swp_entry_t entry
, struct page
*page
)
1050 pte_t swp_pte
= swp_entry_to_pte(entry
);
1055 * We don't actually need pte lock while scanning for swp_pte: since
1056 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1057 * page table while we're scanning; though it could get zapped, and on
1058 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1059 * of unmatched parts which look like swp_pte, so unuse_pte must
1060 * recheck under pte lock. Scanning without pte lock lets it be
1061 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1063 pte
= pte_offset_map(pmd
, addr
);
1066 * swapoff spends a _lot_ of time in this loop!
1067 * Test inline before going to call unuse_pte.
1069 if (unlikely(pte_same(*pte
, swp_pte
))) {
1071 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1074 pte
= pte_offset_map(pmd
, addr
);
1076 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1082 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
1083 unsigned long addr
, unsigned long end
,
1084 swp_entry_t entry
, struct page
*page
)
1090 pmd
= pmd_offset(pud
, addr
);
1092 next
= pmd_addr_end(addr
, end
);
1093 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1095 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
);
1098 } while (pmd
++, addr
= next
, addr
!= end
);
1102 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
1103 unsigned long addr
, unsigned long end
,
1104 swp_entry_t entry
, struct page
*page
)
1110 pud
= pud_offset(pgd
, addr
);
1112 next
= pud_addr_end(addr
, end
);
1113 if (pud_none_or_clear_bad(pud
))
1115 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
);
1118 } while (pud
++, addr
= next
, addr
!= end
);
1122 static int unuse_vma(struct vm_area_struct
*vma
,
1123 swp_entry_t entry
, struct page
*page
)
1126 unsigned long addr
, end
, next
;
1129 if (page_anon_vma(page
)) {
1130 addr
= page_address_in_vma(page
, vma
);
1131 if (addr
== -EFAULT
)
1134 end
= addr
+ PAGE_SIZE
;
1136 addr
= vma
->vm_start
;
1140 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1142 next
= pgd_addr_end(addr
, end
);
1143 if (pgd_none_or_clear_bad(pgd
))
1145 ret
= unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
);
1148 } while (pgd
++, addr
= next
, addr
!= end
);
1152 static int unuse_mm(struct mm_struct
*mm
,
1153 swp_entry_t entry
, struct page
*page
)
1155 struct vm_area_struct
*vma
;
1158 if (!down_read_trylock(&mm
->mmap_sem
)) {
1160 * Activate page so shrink_inactive_list is unlikely to unmap
1161 * its ptes while lock is dropped, so swapoff can make progress.
1163 activate_page(page
);
1165 down_read(&mm
->mmap_sem
);
1168 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1169 if (vma
->anon_vma
&& (ret
= unuse_vma(vma
, entry
, page
)))
1172 up_read(&mm
->mmap_sem
);
1173 return (ret
< 0)? ret
: 0;
1177 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1178 * from current position to next entry still in use.
1179 * Recycle to start on reaching the end, returning 0 when empty.
1181 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
1182 unsigned int prev
, bool frontswap
)
1184 unsigned int max
= si
->max
;
1185 unsigned int i
= prev
;
1186 unsigned char count
;
1189 * No need for swap_lock here: we're just looking
1190 * for whether an entry is in use, not modifying it; false
1191 * hits are okay, and sys_swapoff() has already prevented new
1192 * allocations from this area (while holding swap_lock).
1201 * No entries in use at top of swap_map,
1202 * loop back to start and recheck there.
1209 if (frontswap_test(si
, i
))
1214 count
= si
->swap_map
[i
];
1215 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
1222 * We completely avoid races by reading each swap page in advance,
1223 * and then search for the process using it. All the necessary
1224 * page table adjustments can then be made atomically.
1226 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1227 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1229 int try_to_unuse(unsigned int type
, bool frontswap
,
1230 unsigned long pages_to_unuse
)
1232 struct swap_info_struct
*si
= swap_info
[type
];
1233 struct mm_struct
*start_mm
;
1234 unsigned char *swap_map
;
1235 unsigned char swcount
;
1242 * When searching mms for an entry, a good strategy is to
1243 * start at the first mm we freed the previous entry from
1244 * (though actually we don't notice whether we or coincidence
1245 * freed the entry). Initialize this start_mm with a hold.
1247 * A simpler strategy would be to start at the last mm we
1248 * freed the previous entry from; but that would take less
1249 * advantage of mmlist ordering, which clusters forked mms
1250 * together, child after parent. If we race with dup_mmap(), we
1251 * prefer to resolve parent before child, lest we miss entries
1252 * duplicated after we scanned child: using last mm would invert
1255 start_mm
= &init_mm
;
1256 atomic_inc(&init_mm
.mm_users
);
1259 * Keep on scanning until all entries have gone. Usually,
1260 * one pass through swap_map is enough, but not necessarily:
1261 * there are races when an instance of an entry might be missed.
1263 while ((i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
1264 if (signal_pending(current
)) {
1270 * Get a page for the entry, using the existing swap
1271 * cache page if there is one. Otherwise, get a clean
1272 * page and read the swap into it.
1274 swap_map
= &si
->swap_map
[i
];
1275 entry
= swp_entry(type
, i
);
1276 page
= read_swap_cache_async(entry
,
1277 GFP_HIGHUSER_MOVABLE
, NULL
, 0);
1280 * Either swap_duplicate() failed because entry
1281 * has been freed independently, and will not be
1282 * reused since sys_swapoff() already disabled
1283 * allocation from here, or alloc_page() failed.
1292 * Don't hold on to start_mm if it looks like exiting.
1294 if (atomic_read(&start_mm
->mm_users
) == 1) {
1296 start_mm
= &init_mm
;
1297 atomic_inc(&init_mm
.mm_users
);
1301 * Wait for and lock page. When do_swap_page races with
1302 * try_to_unuse, do_swap_page can handle the fault much
1303 * faster than try_to_unuse can locate the entry. This
1304 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1305 * defer to do_swap_page in such a case - in some tests,
1306 * do_swap_page and try_to_unuse repeatedly compete.
1308 wait_on_page_locked(page
);
1309 wait_on_page_writeback(page
);
1311 wait_on_page_writeback(page
);
1314 * Remove all references to entry.
1316 swcount
= *swap_map
;
1317 if (swap_count(swcount
) == SWAP_MAP_SHMEM
) {
1318 retval
= shmem_unuse(entry
, page
);
1319 /* page has already been unlocked and released */
1324 if (swap_count(swcount
) && start_mm
!= &init_mm
)
1325 retval
= unuse_mm(start_mm
, entry
, page
);
1327 if (swap_count(*swap_map
)) {
1328 int set_start_mm
= (*swap_map
>= swcount
);
1329 struct list_head
*p
= &start_mm
->mmlist
;
1330 struct mm_struct
*new_start_mm
= start_mm
;
1331 struct mm_struct
*prev_mm
= start_mm
;
1332 struct mm_struct
*mm
;
1334 atomic_inc(&new_start_mm
->mm_users
);
1335 atomic_inc(&prev_mm
->mm_users
);
1336 spin_lock(&mmlist_lock
);
1337 while (swap_count(*swap_map
) && !retval
&&
1338 (p
= p
->next
) != &start_mm
->mmlist
) {
1339 mm
= list_entry(p
, struct mm_struct
, mmlist
);
1340 if (!atomic_inc_not_zero(&mm
->mm_users
))
1342 spin_unlock(&mmlist_lock
);
1348 swcount
= *swap_map
;
1349 if (!swap_count(swcount
)) /* any usage ? */
1351 else if (mm
== &init_mm
)
1354 retval
= unuse_mm(mm
, entry
, page
);
1356 if (set_start_mm
&& *swap_map
< swcount
) {
1357 mmput(new_start_mm
);
1358 atomic_inc(&mm
->mm_users
);
1362 spin_lock(&mmlist_lock
);
1364 spin_unlock(&mmlist_lock
);
1367 start_mm
= new_start_mm
;
1371 page_cache_release(page
);
1376 * If a reference remains (rare), we would like to leave
1377 * the page in the swap cache; but try_to_unmap could
1378 * then re-duplicate the entry once we drop page lock,
1379 * so we might loop indefinitely; also, that page could
1380 * not be swapped out to other storage meanwhile. So:
1381 * delete from cache even if there's another reference,
1382 * after ensuring that the data has been saved to disk -
1383 * since if the reference remains (rarer), it will be
1384 * read from disk into another page. Splitting into two
1385 * pages would be incorrect if swap supported "shared
1386 * private" pages, but they are handled by tmpfs files.
1388 * Given how unuse_vma() targets one particular offset
1389 * in an anon_vma, once the anon_vma has been determined,
1390 * this splitting happens to be just what is needed to
1391 * handle where KSM pages have been swapped out: re-reading
1392 * is unnecessarily slow, but we can fix that later on.
1394 if (swap_count(*swap_map
) &&
1395 PageDirty(page
) && PageSwapCache(page
)) {
1396 struct writeback_control wbc
= {
1397 .sync_mode
= WB_SYNC_NONE
,
1400 swap_writepage(page
, &wbc
);
1402 wait_on_page_writeback(page
);
1406 * It is conceivable that a racing task removed this page from
1407 * swap cache just before we acquired the page lock at the top,
1408 * or while we dropped it in unuse_mm(). The page might even
1409 * be back in swap cache on another swap area: that we must not
1410 * delete, since it may not have been written out to swap yet.
1412 if (PageSwapCache(page
) &&
1413 likely(page_private(page
) == entry
.val
))
1414 delete_from_swap_cache(page
);
1417 * So we could skip searching mms once swap count went
1418 * to 1, we did not mark any present ptes as dirty: must
1419 * mark page dirty so shrink_page_list will preserve it.
1423 page_cache_release(page
);
1426 * Make sure that we aren't completely killing
1427 * interactive performance.
1430 if (frontswap
&& pages_to_unuse
> 0) {
1431 if (!--pages_to_unuse
)
1441 * After a successful try_to_unuse, if no swap is now in use, we know
1442 * we can empty the mmlist. swap_lock must be held on entry and exit.
1443 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1444 * added to the mmlist just after page_duplicate - before would be racy.
1446 static void drain_mmlist(void)
1448 struct list_head
*p
, *next
;
1451 for (type
= 0; type
< nr_swapfiles
; type
++)
1452 if (swap_info
[type
]->inuse_pages
)
1454 spin_lock(&mmlist_lock
);
1455 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
1457 spin_unlock(&mmlist_lock
);
1461 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1462 * corresponds to page offset for the specified swap entry.
1463 * Note that the type of this function is sector_t, but it returns page offset
1464 * into the bdev, not sector offset.
1466 sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
1468 struct swap_info_struct
*sis
;
1469 struct swap_extent
*start_se
;
1470 struct swap_extent
*se
;
1473 sis
= swap_info
[swp_type(entry
)];
1476 offset
= swp_offset(entry
);
1477 start_se
= sis
->curr_swap_extent
;
1481 struct list_head
*lh
;
1483 if (se
->start_page
<= offset
&&
1484 offset
< (se
->start_page
+ se
->nr_pages
)) {
1485 return se
->start_block
+ (offset
- se
->start_page
);
1488 se
= list_entry(lh
, struct swap_extent
, list
);
1489 sis
->curr_swap_extent
= se
;
1490 BUG_ON(se
== start_se
); /* It *must* be present */
1493 EXPORT_SYMBOL_GPL(map_swap_entry
);
1496 * Returns the page offset into bdev for the specified page's swap entry.
1498 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
1501 entry
.val
= page_private(page
);
1502 return map_swap_entry(entry
, bdev
);
1506 * Free all of a swapdev's extent information
1508 static void destroy_swap_extents(struct swap_info_struct
*sis
)
1510 while (!list_empty(&sis
->first_swap_extent
.list
)) {
1511 struct swap_extent
*se
;
1513 se
= list_entry(sis
->first_swap_extent
.list
.next
,
1514 struct swap_extent
, list
);
1515 list_del(&se
->list
);
1519 if (sis
->flags
& SWP_FILE
) {
1520 struct file
*swap_file
= sis
->swap_file
;
1521 struct address_space
*mapping
= swap_file
->f_mapping
;
1523 sis
->flags
&= ~SWP_FILE
;
1524 mapping
->a_ops
->swap_deactivate(swap_file
);
1529 * Add a block range (and the corresponding page range) into this swapdev's
1530 * extent list. The extent list is kept sorted in page order.
1532 * This function rather assumes that it is called in ascending page order.
1535 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
1536 unsigned long nr_pages
, sector_t start_block
)
1538 struct swap_extent
*se
;
1539 struct swap_extent
*new_se
;
1540 struct list_head
*lh
;
1542 if (start_page
== 0) {
1543 se
= &sis
->first_swap_extent
;
1544 sis
->curr_swap_extent
= se
;
1546 se
->nr_pages
= nr_pages
;
1547 se
->start_block
= start_block
;
1550 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
1551 se
= list_entry(lh
, struct swap_extent
, list
);
1552 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
1553 if (se
->start_block
+ se
->nr_pages
== start_block
) {
1555 se
->nr_pages
+= nr_pages
;
1561 * No merge. Insert a new extent, preserving ordering.
1563 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
1566 new_se
->start_page
= start_page
;
1567 new_se
->nr_pages
= nr_pages
;
1568 new_se
->start_block
= start_block
;
1570 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
1575 * A `swap extent' is a simple thing which maps a contiguous range of pages
1576 * onto a contiguous range of disk blocks. An ordered list of swap extents
1577 * is built at swapon time and is then used at swap_writepage/swap_readpage
1578 * time for locating where on disk a page belongs.
1580 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1581 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1582 * swap files identically.
1584 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1585 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1586 * swapfiles are handled *identically* after swapon time.
1588 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1589 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1590 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1591 * requirements, they are simply tossed out - we will never use those blocks
1594 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1595 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1596 * which will scribble on the fs.
1598 * The amount of disk space which a single swap extent represents varies.
1599 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1600 * extents in the list. To avoid much list walking, we cache the previous
1601 * search location in `curr_swap_extent', and start new searches from there.
1602 * This is extremely effective. The average number of iterations in
1603 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1605 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
1607 struct file
*swap_file
= sis
->swap_file
;
1608 struct address_space
*mapping
= swap_file
->f_mapping
;
1609 struct inode
*inode
= mapping
->host
;
1612 if (S_ISBLK(inode
->i_mode
)) {
1613 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1618 if (mapping
->a_ops
->swap_activate
) {
1619 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
1621 sis
->flags
|= SWP_FILE
;
1622 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1628 return generic_swapfile_activate(sis
, swap_file
, span
);
1631 static void _enable_swap_info(struct swap_info_struct
*p
, int prio
,
1632 unsigned char *swap_map
)
1639 p
->prio
= --least_priority
;
1640 p
->swap_map
= swap_map
;
1641 p
->flags
|= SWP_WRITEOK
;
1642 atomic_long_add(p
->pages
, &nr_swap_pages
);
1643 total_swap_pages
+= p
->pages
;
1645 /* insert swap space into swap_list: */
1647 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
]->next
) {
1648 if (p
->prio
>= swap_info
[i
]->prio
)
1654 swap_list
.head
= swap_list
.next
= p
->type
;
1656 swap_info
[prev
]->next
= p
->type
;
1659 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
1660 unsigned char *swap_map
,
1661 unsigned long *frontswap_map
)
1663 frontswap_init(p
->type
, frontswap_map
);
1664 spin_lock(&swap_lock
);
1665 spin_lock(&p
->lock
);
1666 _enable_swap_info(p
, prio
, swap_map
);
1667 spin_unlock(&p
->lock
);
1668 spin_unlock(&swap_lock
);
1671 static void reinsert_swap_info(struct swap_info_struct
*p
)
1673 spin_lock(&swap_lock
);
1674 spin_lock(&p
->lock
);
1675 _enable_swap_info(p
, p
->prio
, p
->swap_map
);
1676 spin_unlock(&p
->lock
);
1677 spin_unlock(&swap_lock
);
1680 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
1682 struct swap_info_struct
*p
= NULL
;
1683 unsigned char *swap_map
;
1684 unsigned long *frontswap_map
;
1685 struct file
*swap_file
, *victim
;
1686 struct address_space
*mapping
;
1687 struct inode
*inode
;
1688 struct filename
*pathname
;
1692 if (!capable(CAP_SYS_ADMIN
))
1696 // FIXME: Turn it off due to current->mm may be NULL in kernel space
1697 // by calling sys_swapoff(swapfilename) in disable_swapfile() @ tuxonice_swap.c
1698 pr_warn("[HIB/SWAP] [%s] file(%s) current(%p/%d/%s) current->mm(%p)\n", __func__
, specialfile
, current
, current
->pid
, current
->comm
, current
->mm
);
1699 WARN_ON(!current
->mm
);
1701 BUG_ON(!current
->mm
);
1704 pathname
= getname(specialfile
);
1705 if (IS_ERR(pathname
))
1706 return PTR_ERR(pathname
);
1708 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
1709 err
= PTR_ERR(victim
);
1713 mapping
= victim
->f_mapping
;
1715 spin_lock(&swap_lock
);
1716 for (type
= swap_list
.head
; type
>= 0; type
= swap_info
[type
]->next
) {
1717 p
= swap_info
[type
];
1718 if (p
->flags
& SWP_WRITEOK
) {
1719 if (p
->swap_file
->f_mapping
== mapping
)
1726 spin_unlock(&swap_lock
);
1729 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
1730 vm_unacct_memory(p
->pages
);
1733 spin_unlock(&swap_lock
);
1737 swap_list
.head
= p
->next
;
1739 swap_info
[prev
]->next
= p
->next
;
1740 if (type
== swap_list
.next
) {
1741 /* just pick something that's safe... */
1742 swap_list
.next
= swap_list
.head
;
1744 spin_lock(&p
->lock
);
1746 for (i
= p
->next
; i
>= 0; i
= swap_info
[i
]->next
)
1747 swap_info
[i
]->prio
= p
->prio
--;
1750 atomic_long_sub(p
->pages
, &nr_swap_pages
);
1751 total_swap_pages
-= p
->pages
;
1752 p
->flags
&= ~SWP_WRITEOK
;
1753 spin_unlock(&p
->lock
);
1754 spin_unlock(&swap_lock
);
1756 set_current_oom_origin();
1757 err
= try_to_unuse(type
, false, 0); /* force all pages to be unused */
1758 clear_current_oom_origin();
1761 /* re-insert swap space back into swap_list */
1762 reinsert_swap_info(p
);
1766 destroy_swap_extents(p
);
1767 if (p
->flags
& SWP_CONTINUED
)
1768 free_swap_count_continuations(p
);
1770 mutex_lock(&swapon_mutex
);
1771 spin_lock(&swap_lock
);
1772 spin_lock(&p
->lock
);
1775 /* wait for anyone still in scan_swap_map */
1776 p
->highest_bit
= 0; /* cuts scans short */
1777 while (p
->flags
>= SWP_SCANNING
) {
1778 spin_unlock(&p
->lock
);
1779 spin_unlock(&swap_lock
);
1780 schedule_timeout_uninterruptible(1);
1781 spin_lock(&swap_lock
);
1782 spin_lock(&p
->lock
);
1785 swap_file
= p
->swap_file
;
1786 p
->swap_file
= NULL
;
1788 swap_map
= p
->swap_map
;
1791 frontswap_map
= frontswap_map_get(p
);
1792 frontswap_map_set(p
, NULL
);
1793 spin_unlock(&p
->lock
);
1794 spin_unlock(&swap_lock
);
1795 frontswap_invalidate_area(type
);
1796 mutex_unlock(&swapon_mutex
);
1798 vfree(frontswap_map
);
1799 /* Destroy swap account informatin */
1800 swap_cgroup_swapoff(type
);
1802 inode
= mapping
->host
;
1803 if (S_ISBLK(inode
->i_mode
)) {
1804 struct block_device
*bdev
= I_BDEV(inode
);
1805 set_blocksize(bdev
, p
->old_block_size
);
1806 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1808 mutex_lock(&inode
->i_mutex
);
1809 inode
->i_flags
&= ~S_SWAPFILE
;
1810 mutex_unlock(&inode
->i_mutex
);
1812 filp_close(swap_file
, NULL
);
1814 atomic_inc(&proc_poll_event
);
1815 wake_up_interruptible(&proc_poll_wait
);
1818 filp_close(victim
, NULL
);
1823 EXPORT_SYMBOL_GPL(sys_swapoff
);
1825 #ifdef CONFIG_PROC_FS
1826 static unsigned swaps_poll(struct file
*file
, poll_table
*wait
)
1828 struct seq_file
*seq
= file
->private_data
;
1830 poll_wait(file
, &proc_poll_wait
, wait
);
1832 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
1833 seq
->poll_event
= atomic_read(&proc_poll_event
);
1834 return POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
1837 return POLLIN
| POLLRDNORM
;
1841 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
1843 struct swap_info_struct
*si
;
1847 mutex_lock(&swapon_mutex
);
1850 return SEQ_START_TOKEN
;
1852 for (type
= 0; type
< nr_swapfiles
; type
++) {
1853 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1854 si
= swap_info
[type
];
1855 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
1864 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
1866 struct swap_info_struct
*si
= v
;
1869 if (v
== SEQ_START_TOKEN
)
1872 type
= si
->type
+ 1;
1874 for (; type
< nr_swapfiles
; type
++) {
1875 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1876 si
= swap_info
[type
];
1877 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
1886 static void swap_stop(struct seq_file
*swap
, void *v
)
1888 mutex_unlock(&swapon_mutex
);
1891 static int swap_show(struct seq_file
*swap
, void *v
)
1893 struct swap_info_struct
*si
= v
;
1897 if (si
== SEQ_START_TOKEN
) {
1898 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1902 file
= si
->swap_file
;
1903 len
= seq_path(swap
, &file
->f_path
, " \t\n\\");
1904 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
1905 len
< 40 ? 40 - len
: 1, " ",
1906 S_ISBLK(file_inode(file
)->i_mode
) ?
1907 "partition" : "file\t",
1908 si
->pages
<< (PAGE_SHIFT
- 10),
1909 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
1914 static const struct seq_operations swaps_op
= {
1915 .start
= swap_start
,
1921 static int swaps_open(struct inode
*inode
, struct file
*file
)
1923 struct seq_file
*seq
;
1926 ret
= seq_open(file
, &swaps_op
);
1930 seq
= file
->private_data
;
1931 seq
->poll_event
= atomic_read(&proc_poll_event
);
1935 static const struct file_operations proc_swaps_operations
= {
1938 .llseek
= seq_lseek
,
1939 .release
= seq_release
,
1943 static int __init
procswaps_init(void)
1945 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
1948 __initcall(procswaps_init
);
1949 #endif /* CONFIG_PROC_FS */
1951 #ifdef MAX_SWAPFILES_CHECK
1952 static int __init
max_swapfiles_check(void)
1954 MAX_SWAPFILES_CHECK();
1957 late_initcall(max_swapfiles_check
);
1960 static struct swap_info_struct
*alloc_swap_info(void)
1962 struct swap_info_struct
*p
;
1965 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
1967 return ERR_PTR(-ENOMEM
);
1969 spin_lock(&swap_lock
);
1970 for (type
= 0; type
< nr_swapfiles
; type
++) {
1971 if (!(swap_info
[type
]->flags
& SWP_USED
))
1974 if (type
>= MAX_SWAPFILES
) {
1975 spin_unlock(&swap_lock
);
1977 return ERR_PTR(-EPERM
);
1979 if (type
>= nr_swapfiles
) {
1981 swap_info
[type
] = p
;
1983 * Write swap_info[type] before nr_swapfiles, in case a
1984 * racing procfs swap_start() or swap_next() is reading them.
1985 * (We never shrink nr_swapfiles, we never free this entry.)
1991 p
= swap_info
[type
];
1993 * Do not memset this entry: a racing procfs swap_next()
1994 * would be relying on p->type to remain valid.
1997 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
1998 p
->flags
= SWP_USED
;
2000 spin_unlock(&swap_lock
);
2001 spin_lock_init(&p
->lock
);
2006 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2010 if (S_ISBLK(inode
->i_mode
)) {
2011 p
->bdev
= bdgrab(I_BDEV(inode
));
2012 error
= blkdev_get(p
->bdev
,
2013 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
,
2019 p
->old_block_size
= block_size(p
->bdev
);
2020 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2023 p
->flags
|= SWP_BLKDEV
;
2024 } else if (S_ISREG(inode
->i_mode
)) {
2025 p
->bdev
= inode
->i_sb
->s_bdev
;
2026 mutex_lock(&inode
->i_mutex
);
2027 if (IS_SWAPFILE(inode
))
2035 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2036 union swap_header
*swap_header
,
2037 struct inode
*inode
)
2040 unsigned long maxpages
;
2041 unsigned long swapfilepages
;
2043 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2044 printk(KERN_ERR
"Unable to find swap-space signature\n");
2048 /* swap partition endianess hack... */
2049 if (swab32(swap_header
->info
.version
) == 1) {
2050 swab32s(&swap_header
->info
.version
);
2051 swab32s(&swap_header
->info
.last_page
);
2052 swab32s(&swap_header
->info
.nr_badpages
);
2053 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2055 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2056 swab32s(&swap_header
->info
.badpages
[i
]);
2058 /* Check the swap header's sub-version */
2059 if (swap_header
->info
.version
!= 1) {
2061 "Unable to handle swap header version %d\n",
2062 swap_header
->info
.version
);
2067 p
->cluster_next
= 1;
2071 * Find out how many pages are allowed for a single swap
2072 * device. There are two limiting factors: 1) the number
2073 * of bits for the swap offset in the swp_entry_t type, and
2074 * 2) the number of bits in the swap pte as defined by the
2075 * different architectures. In order to find the
2076 * largest possible bit mask, a swap entry with swap type 0
2077 * and swap offset ~0UL is created, encoded to a swap pte,
2078 * decoded to a swp_entry_t again, and finally the swap
2079 * offset is extracted. This will mask all the bits from
2080 * the initial ~0UL mask that can't be encoded in either
2081 * the swp_entry_t or the architecture definition of a
2084 maxpages
= swp_offset(pte_to_swp_entry(
2085 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2086 if (maxpages
> swap_header
->info
.last_page
) {
2087 maxpages
= swap_header
->info
.last_page
+ 1;
2088 /* p->max is an unsigned int: don't overflow it */
2089 if ((unsigned int)maxpages
== 0)
2090 maxpages
= UINT_MAX
;
2092 p
->highest_bit
= maxpages
- 1;
2096 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
2097 if (swapfilepages
&& maxpages
> swapfilepages
) {
2099 "Swap area shorter than signature indicates\n");
2102 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
2104 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2110 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
2111 union swap_header
*swap_header
,
2112 unsigned char *swap_map
,
2113 unsigned long maxpages
,
2117 unsigned int nr_good_pages
;
2120 nr_good_pages
= maxpages
- 1; /* omit header page */
2122 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
2123 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
2124 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
2126 if (page_nr
< maxpages
) {
2127 swap_map
[page_nr
] = SWAP_MAP_BAD
;
2132 if (nr_good_pages
) {
2133 swap_map
[0] = SWAP_MAP_BAD
;
2135 p
->pages
= nr_good_pages
;
2136 nr_extents
= setup_swap_extents(p
, span
);
2139 nr_good_pages
= p
->pages
;
2141 if (!nr_good_pages
) {
2142 printk(KERN_WARNING
"Empty swap-file\n");
2149 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
2151 struct swap_info_struct
*p
;
2152 struct filename
*name
;
2153 struct file
*swap_file
= NULL
;
2154 struct address_space
*mapping
;
2158 union swap_header
*swap_header
;
2161 unsigned long maxpages
;
2162 unsigned char *swap_map
= NULL
;
2163 unsigned long *frontswap_map
= NULL
;
2164 struct page
*page
= NULL
;
2165 struct inode
*inode
= NULL
;
2167 if (swap_flags
& ~SWAP_FLAGS_VALID
)
2170 if (!capable(CAP_SYS_ADMIN
))
2173 p
= alloc_swap_info();
2177 name
= getname(specialfile
);
2179 error
= PTR_ERR(name
);
2183 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
2184 if (IS_ERR(swap_file
)) {
2185 error
= PTR_ERR(swap_file
);
2190 p
->swap_file
= swap_file
;
2191 mapping
= swap_file
->f_mapping
;
2193 for (i
= 0; i
< nr_swapfiles
; i
++) {
2194 struct swap_info_struct
*q
= swap_info
[i
];
2196 if (q
== p
|| !q
->swap_file
)
2198 if (mapping
== q
->swap_file
->f_mapping
) {
2204 inode
= mapping
->host
;
2205 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
2206 error
= claim_swapfile(p
, inode
);
2207 if (unlikely(error
))
2211 * Read the swap header.
2213 if (!mapping
->a_ops
->readpage
) {
2217 page
= read_mapping_page(mapping
, 0, swap_file
);
2219 error
= PTR_ERR(page
);
2222 swap_header
= kmap(page
);
2224 maxpages
= read_swap_header(p
, swap_header
, inode
);
2225 if (unlikely(!maxpages
)) {
2230 /* OK, set up the swap map and apply the bad block list */
2231 swap_map
= vzalloc(maxpages
);
2237 error
= swap_cgroup_swapon(p
->type
, maxpages
);
2241 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
2243 if (unlikely(nr_extents
< 0)) {
2247 /* frontswap enabled? set up bit-per-page map for frontswap */
2248 if (frontswap_enabled
)
2249 frontswap_map
= vzalloc(BITS_TO_LONGS(maxpages
) * sizeof(long));
2252 if (blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
2253 p
->flags
|= SWP_SOLIDSTATE
;
2254 p
->cluster_next
= 1 + (prandom_u32() % p
->highest_bit
);
2256 if ((swap_flags
& SWAP_FLAG_DISCARD
) && discard_swap(p
) == 0)
2257 p
->flags
|= SWP_DISCARDABLE
;
2260 mutex_lock(&swapon_mutex
);
2262 if (swap_flags
& SWAP_FLAG_PREFER
)
2264 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
2265 enable_swap_info(p
, prio
, swap_map
, frontswap_map
);
2267 printk(KERN_INFO
"Adding %uk swap on %s. "
2268 "Priority:%d extents:%d across:%lluk %s%s%s\n",
2269 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
2270 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
2271 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
2272 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
2273 (frontswap_map
) ? "FS" : "");
2275 mutex_unlock(&swapon_mutex
);
2276 atomic_inc(&proc_poll_event
);
2277 wake_up_interruptible(&proc_poll_wait
);
2279 if (S_ISREG(inode
->i_mode
))
2280 inode
->i_flags
|= S_SWAPFILE
;
2284 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
2285 set_blocksize(p
->bdev
, p
->old_block_size
);
2286 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2288 destroy_swap_extents(p
);
2289 swap_cgroup_swapoff(p
->type
);
2290 spin_lock(&swap_lock
);
2291 p
->swap_file
= NULL
;
2293 spin_unlock(&swap_lock
);
2296 if (inode
&& S_ISREG(inode
->i_mode
)) {
2297 mutex_unlock(&inode
->i_mutex
);
2300 filp_close(swap_file
, NULL
);
2303 if (page
&& !IS_ERR(page
)) {
2305 page_cache_release(page
);
2309 if (inode
&& S_ISREG(inode
->i_mode
))
2310 mutex_unlock(&inode
->i_mutex
);
2313 EXPORT_SYMBOL_GPL(sys_swapon
);
2315 void si_swapinfo(struct sysinfo
*val
)
2318 unsigned long nr_to_be_unused
= 0;
2320 spin_lock(&swap_lock
);
2321 for (type
= 0; type
< nr_swapfiles
; type
++) {
2322 struct swap_info_struct
*si
= swap_info
[type
];
2324 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
2325 nr_to_be_unused
+= si
->inuse_pages
;
2327 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
2328 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
2329 spin_unlock(&swap_lock
);
2331 EXPORT_SYMBOL_GPL(si_swapinfo
);
2334 * Verify that a swap entry is valid and increment its swap map count.
2336 * Returns error code in following case.
2338 * - swp_entry is invalid -> EINVAL
2339 * - swp_entry is migration entry -> EINVAL
2340 * - swap-cache reference is requested but there is already one. -> EEXIST
2341 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2342 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2344 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
2346 struct swap_info_struct
*p
;
2347 unsigned long offset
, type
;
2348 unsigned char count
;
2349 unsigned char has_cache
;
2352 if (non_swap_entry(entry
))
2355 type
= swp_type(entry
);
2356 if (type
>= nr_swapfiles
)
2358 p
= swap_info
[type
];
2359 offset
= swp_offset(entry
);
2361 spin_lock(&p
->lock
);
2362 if (unlikely(offset
>= p
->max
))
2365 count
= p
->swap_map
[offset
];
2366 has_cache
= count
& SWAP_HAS_CACHE
;
2367 count
&= ~SWAP_HAS_CACHE
;
2370 if (usage
== SWAP_HAS_CACHE
) {
2372 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2373 if (!has_cache
&& count
)
2374 has_cache
= SWAP_HAS_CACHE
;
2375 else if (has_cache
) /* someone else added cache */
2377 else /* no users remaining */
2380 } else if (count
|| has_cache
) {
2382 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
2384 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
2386 else if (swap_count_continued(p
, offset
, count
))
2387 count
= COUNT_CONTINUED
;
2391 err
= -ENOENT
; /* unused swap entry */
2393 p
->swap_map
[offset
] = count
| has_cache
;
2396 spin_unlock(&p
->lock
);
2401 printk(KERN_ERR
"swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
2406 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2407 * (in which case its reference count is never incremented).
2409 void swap_shmem_alloc(swp_entry_t entry
)
2411 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
2415 * Increase reference count of swap entry by 1.
2416 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2417 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2418 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2419 * might occur if a page table entry has got corrupted.
2421 int swap_duplicate(swp_entry_t entry
)
2425 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
2426 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
2431 * @entry: swap entry for which we allocate swap cache.
2433 * Called when allocating swap cache for existing swap entry,
2434 * This can return error codes. Returns 0 at success.
2435 * -EBUSY means there is a swap cache.
2436 * Note: return code is different from swap_duplicate().
2438 int swapcache_prepare(swp_entry_t entry
)
2440 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
2443 struct swap_info_struct
*page_swap_info(struct page
*page
)
2445 swp_entry_t swap
= { .val
= page_private(page
) };
2446 BUG_ON(!PageSwapCache(page
));
2447 return swap_info
[swp_type(swap
)];
2451 * out-of-line __page_file_ methods to avoid include hell.
2453 struct address_space
*__page_file_mapping(struct page
*page
)
2455 VM_BUG_ON(!PageSwapCache(page
));
2456 return page_swap_info(page
)->swap_file
->f_mapping
;
2458 EXPORT_SYMBOL_GPL(__page_file_mapping
);
2460 pgoff_t
__page_file_index(struct page
*page
)
2462 swp_entry_t swap
= { .val
= page_private(page
) };
2463 VM_BUG_ON(!PageSwapCache(page
));
2464 return swp_offset(swap
);
2467 EXPORT_SYMBOL_GPL(__page_file_index
);
2469 struct swap_info_struct
*get_swap_info_struct(unsigned type
)
2471 return swap_info
[type
];
2473 EXPORT_SYMBOL_GPL(get_swap_info_struct
);
2476 * add_swap_count_continuation - called when a swap count is duplicated
2477 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2478 * page of the original vmalloc'ed swap_map, to hold the continuation count
2479 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2480 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2482 * These continuation pages are seldom referenced: the common paths all work
2483 * on the original swap_map, only referring to a continuation page when the
2484 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2486 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2487 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2488 * can be called after dropping locks.
2490 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
2492 struct swap_info_struct
*si
;
2495 struct page
*list_page
;
2497 unsigned char count
;
2500 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2501 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2503 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
2505 si
= swap_info_get(entry
);
2508 * An acceptable race has occurred since the failing
2509 * __swap_duplicate(): the swap entry has been freed,
2510 * perhaps even the whole swap_map cleared for swapoff.
2515 offset
= swp_offset(entry
);
2516 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
2518 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
2520 * The higher the swap count, the more likely it is that tasks
2521 * will race to add swap count continuation: we need to avoid
2522 * over-provisioning.
2528 spin_unlock(&si
->lock
);
2533 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2534 * no architecture is using highmem pages for kernel pagetables: so it
2535 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2537 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2538 offset
&= ~PAGE_MASK
;
2541 * Page allocation does not initialize the page's lru field,
2542 * but it does always reset its private field.
2544 if (!page_private(head
)) {
2545 BUG_ON(count
& COUNT_CONTINUED
);
2546 INIT_LIST_HEAD(&head
->lru
);
2547 set_page_private(head
, SWP_CONTINUED
);
2548 si
->flags
|= SWP_CONTINUED
;
2551 list_for_each_entry(list_page
, &head
->lru
, lru
) {
2555 * If the previous map said no continuation, but we've found
2556 * a continuation page, free our allocation and use this one.
2558 if (!(count
& COUNT_CONTINUED
))
2561 map
= kmap_atomic(list_page
) + offset
;
2566 * If this continuation count now has some space in it,
2567 * free our allocation and use this one.
2569 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
2573 list_add_tail(&page
->lru
, &head
->lru
);
2574 page
= NULL
; /* now it's attached, don't free it */
2576 spin_unlock(&si
->lock
);
2584 * swap_count_continued - when the original swap_map count is incremented
2585 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2586 * into, carry if so, or else fail until a new continuation page is allocated;
2587 * when the original swap_map count is decremented from 0 with continuation,
2588 * borrow from the continuation and report whether it still holds more.
2589 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2591 static bool swap_count_continued(struct swap_info_struct
*si
,
2592 pgoff_t offset
, unsigned char count
)
2598 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2599 if (page_private(head
) != SWP_CONTINUED
) {
2600 BUG_ON(count
& COUNT_CONTINUED
);
2601 return false; /* need to add count continuation */
2604 offset
&= ~PAGE_MASK
;
2605 page
= list_entry(head
->lru
.next
, struct page
, lru
);
2606 map
= kmap_atomic(page
) + offset
;
2608 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
2609 goto init_map
; /* jump over SWAP_CONT_MAX checks */
2611 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
2613 * Think of how you add 1 to 999
2615 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
2617 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2618 BUG_ON(page
== head
);
2619 map
= kmap_atomic(page
) + offset
;
2621 if (*map
== SWAP_CONT_MAX
) {
2623 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2625 return false; /* add count continuation */
2626 map
= kmap_atomic(page
) + offset
;
2627 init_map
: *map
= 0; /* we didn't zero the page */
2631 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2632 while (page
!= head
) {
2633 map
= kmap_atomic(page
) + offset
;
2634 *map
= COUNT_CONTINUED
;
2636 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2638 return true; /* incremented */
2640 } else { /* decrementing */
2642 * Think of how you subtract 1 from 1000
2644 BUG_ON(count
!= COUNT_CONTINUED
);
2645 while (*map
== COUNT_CONTINUED
) {
2647 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2648 BUG_ON(page
== head
);
2649 map
= kmap_atomic(page
) + offset
;
2656 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2657 while (page
!= head
) {
2658 map
= kmap_atomic(page
) + offset
;
2659 *map
= SWAP_CONT_MAX
| count
;
2660 count
= COUNT_CONTINUED
;
2662 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2664 return count
== COUNT_CONTINUED
;
2669 * free_swap_count_continuations - swapoff free all the continuation pages
2670 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2672 static void free_swap_count_continuations(struct swap_info_struct
*si
)
2676 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
2678 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2679 if (page_private(head
)) {
2680 struct list_head
*this, *next
;
2681 list_for_each_safe(this, next
, &head
->lru
) {
2683 page
= list_entry(this, struct page
, lru
);