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/shmem_fs.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/ksm.h>
25 #include <linux/rmap.h>
26 #include <linux/security.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mutex.h>
29 #include <linux/capability.h>
30 #include <linux/syscalls.h>
31 #include <linux/memcontrol.h>
32 #include <linux/poll.h>
33 #include <linux/oom.h>
34 #include <linux/frontswap.h>
35 #include <linux/swapfile.h>
36 #include <linux/export.h>
38 #include <asm/pgtable.h>
39 #include <asm/tlbflush.h>
40 #include <linux/swapops.h>
41 #include <linux/page_cgroup.h>
43 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
45 static void free_swap_count_continuations(struct swap_info_struct
*);
46 static sector_t
map_swap_entry(swp_entry_t
, struct block_device
**);
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
;
55 static const char Bad_file
[] = "Bad swap file entry ";
56 static const char Unused_file
[] = "Unused swap file entry ";
57 static const char Bad_offset
[] = "Bad swap offset entry ";
58 static const char Unused_offset
[] = "Unused swap offset entry ";
61 * all active swap_info_structs
62 * protected with swap_lock, and ordered by priority.
64 LIST_HEAD(swap_list_head
);
66 struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
68 static DEFINE_MUTEX(swapon_mutex
);
70 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
71 /* Activity counter to indicate that a swapon or swapoff has occurred */
72 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
74 static inline unsigned char swap_count(unsigned char ent
)
76 return ent
& ~SWAP_HAS_CACHE
; /* may include SWAP_HAS_CONT flag */
79 /* returns 1 if swap entry is freed */
81 __try_to_reclaim_swap(struct swap_info_struct
*si
, unsigned long offset
)
83 swp_entry_t entry
= swp_entry(si
->type
, offset
);
87 page
= find_get_page(swap_address_space(entry
), entry
.val
);
91 * This function is called from scan_swap_map() and it's called
92 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
93 * We have to use trylock for avoiding deadlock. This is a special
94 * case and you should use try_to_free_swap() with explicit lock_page()
95 * in usual operations.
97 if (trylock_page(page
)) {
98 ret
= try_to_free_swap(page
);
101 page_cache_release(page
);
106 * swapon tell device that all the old swap contents can be discarded,
107 * to allow the swap device to optimize its wear-levelling.
109 static int discard_swap(struct swap_info_struct
*si
)
111 struct swap_extent
*se
;
112 sector_t start_block
;
116 /* Do not discard the swap header page! */
117 se
= &si
->first_swap_extent
;
118 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
119 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
121 err
= blkdev_issue_discard(si
->bdev
, start_block
,
122 nr_blocks
, GFP_KERNEL
, 0);
128 list_for_each_entry(se
, &si
->first_swap_extent
.list
, list
) {
129 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
130 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
132 err
= blkdev_issue_discard(si
->bdev
, start_block
,
133 nr_blocks
, GFP_KERNEL
, 0);
139 return err
; /* That will often be -EOPNOTSUPP */
143 * swap allocation tell device that a cluster of swap can now be discarded,
144 * to allow the swap device to optimize its wear-levelling.
146 static void discard_swap_cluster(struct swap_info_struct
*si
,
147 pgoff_t start_page
, pgoff_t nr_pages
)
149 struct swap_extent
*se
= si
->curr_swap_extent
;
150 int found_extent
= 0;
153 struct list_head
*lh
;
155 if (se
->start_page
<= start_page
&&
156 start_page
< se
->start_page
+ se
->nr_pages
) {
157 pgoff_t offset
= start_page
- se
->start_page
;
158 sector_t start_block
= se
->start_block
+ offset
;
159 sector_t nr_blocks
= se
->nr_pages
- offset
;
161 if (nr_blocks
> nr_pages
)
162 nr_blocks
= nr_pages
;
163 start_page
+= nr_blocks
;
164 nr_pages
-= nr_blocks
;
167 si
->curr_swap_extent
= se
;
169 start_block
<<= PAGE_SHIFT
- 9;
170 nr_blocks
<<= PAGE_SHIFT
- 9;
171 if (blkdev_issue_discard(si
->bdev
, start_block
,
172 nr_blocks
, GFP_NOIO
, 0))
177 se
= list_entry(lh
, struct swap_extent
, list
);
181 #define SWAPFILE_CLUSTER 256
182 #define LATENCY_LIMIT 256
184 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
190 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
195 static inline void cluster_set_count(struct swap_cluster_info
*info
,
201 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
202 unsigned int c
, unsigned int f
)
208 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
213 static inline void cluster_set_next(struct swap_cluster_info
*info
,
219 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
220 unsigned int n
, unsigned int f
)
226 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
228 return info
->flags
& CLUSTER_FLAG_FREE
;
231 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
233 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
236 static inline void cluster_set_null(struct swap_cluster_info
*info
)
238 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
242 /* Add a cluster to discard list and schedule it to do discard */
243 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
247 * If scan_swap_map() can't find a free cluster, it will check
248 * si->swap_map directly. To make sure the discarding cluster isn't
249 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
250 * will be cleared after discard
252 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
253 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
255 if (cluster_is_null(&si
->discard_cluster_head
)) {
256 cluster_set_next_flag(&si
->discard_cluster_head
,
258 cluster_set_next_flag(&si
->discard_cluster_tail
,
261 unsigned int tail
= cluster_next(&si
->discard_cluster_tail
);
262 cluster_set_next(&si
->cluster_info
[tail
], idx
);
263 cluster_set_next_flag(&si
->discard_cluster_tail
,
267 schedule_work(&si
->discard_work
);
271 * Doing discard actually. After a cluster discard is finished, the cluster
272 * will be added to free cluster list. caller should hold si->lock.
274 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
276 struct swap_cluster_info
*info
;
279 info
= si
->cluster_info
;
281 while (!cluster_is_null(&si
->discard_cluster_head
)) {
282 idx
= cluster_next(&si
->discard_cluster_head
);
284 cluster_set_next_flag(&si
->discard_cluster_head
,
285 cluster_next(&info
[idx
]), 0);
286 if (cluster_next(&si
->discard_cluster_tail
) == idx
) {
287 cluster_set_null(&si
->discard_cluster_head
);
288 cluster_set_null(&si
->discard_cluster_tail
);
290 spin_unlock(&si
->lock
);
292 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
295 spin_lock(&si
->lock
);
296 cluster_set_flag(&info
[idx
], CLUSTER_FLAG_FREE
);
297 if (cluster_is_null(&si
->free_cluster_head
)) {
298 cluster_set_next_flag(&si
->free_cluster_head
,
300 cluster_set_next_flag(&si
->free_cluster_tail
,
305 tail
= cluster_next(&si
->free_cluster_tail
);
306 cluster_set_next(&info
[tail
], idx
);
307 cluster_set_next_flag(&si
->free_cluster_tail
,
310 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
311 0, SWAPFILE_CLUSTER
);
315 static void swap_discard_work(struct work_struct
*work
)
317 struct swap_info_struct
*si
;
319 si
= container_of(work
, struct swap_info_struct
, discard_work
);
321 spin_lock(&si
->lock
);
322 swap_do_scheduled_discard(si
);
323 spin_unlock(&si
->lock
);
327 * The cluster corresponding to page_nr will be used. The cluster will be
328 * removed from free cluster list and its usage counter will be increased.
330 static void inc_cluster_info_page(struct swap_info_struct
*p
,
331 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
333 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
337 if (cluster_is_free(&cluster_info
[idx
])) {
338 VM_BUG_ON(cluster_next(&p
->free_cluster_head
) != idx
);
339 cluster_set_next_flag(&p
->free_cluster_head
,
340 cluster_next(&cluster_info
[idx
]), 0);
341 if (cluster_next(&p
->free_cluster_tail
) == idx
) {
342 cluster_set_null(&p
->free_cluster_tail
);
343 cluster_set_null(&p
->free_cluster_head
);
345 cluster_set_count_flag(&cluster_info
[idx
], 0, 0);
348 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
349 cluster_set_count(&cluster_info
[idx
],
350 cluster_count(&cluster_info
[idx
]) + 1);
354 * The cluster corresponding to page_nr decreases one usage. If the usage
355 * counter becomes 0, which means no page in the cluster is in using, we can
356 * optionally discard the cluster and add it to free cluster list.
358 static void dec_cluster_info_page(struct swap_info_struct
*p
,
359 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
361 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
366 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
367 cluster_set_count(&cluster_info
[idx
],
368 cluster_count(&cluster_info
[idx
]) - 1);
370 if (cluster_count(&cluster_info
[idx
]) == 0) {
372 * If the swap is discardable, prepare discard the cluster
373 * instead of free it immediately. The cluster will be freed
376 if ((p
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
377 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
378 swap_cluster_schedule_discard(p
, idx
);
382 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
383 if (cluster_is_null(&p
->free_cluster_head
)) {
384 cluster_set_next_flag(&p
->free_cluster_head
, idx
, 0);
385 cluster_set_next_flag(&p
->free_cluster_tail
, idx
, 0);
387 unsigned int tail
= cluster_next(&p
->free_cluster_tail
);
388 cluster_set_next(&cluster_info
[tail
], idx
);
389 cluster_set_next_flag(&p
->free_cluster_tail
, idx
, 0);
395 * It's possible scan_swap_map() uses a free cluster in the middle of free
396 * cluster list. Avoiding such abuse to avoid list corruption.
399 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
400 unsigned long offset
)
402 struct percpu_cluster
*percpu_cluster
;
405 offset
/= SWAPFILE_CLUSTER
;
406 conflict
= !cluster_is_null(&si
->free_cluster_head
) &&
407 offset
!= cluster_next(&si
->free_cluster_head
) &&
408 cluster_is_free(&si
->cluster_info
[offset
]);
413 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
414 cluster_set_null(&percpu_cluster
->index
);
419 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
420 * might involve allocating a new cluster for current CPU too.
422 static void scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
423 unsigned long *offset
, unsigned long *scan_base
)
425 struct percpu_cluster
*cluster
;
430 cluster
= this_cpu_ptr(si
->percpu_cluster
);
431 if (cluster_is_null(&cluster
->index
)) {
432 if (!cluster_is_null(&si
->free_cluster_head
)) {
433 cluster
->index
= si
->free_cluster_head
;
434 cluster
->next
= cluster_next(&cluster
->index
) *
436 } else if (!cluster_is_null(&si
->discard_cluster_head
)) {
438 * we don't have free cluster but have some clusters in
439 * discarding, do discard now and reclaim them
441 swap_do_scheduled_discard(si
);
442 *scan_base
= *offset
= si
->cluster_next
;
451 * Other CPUs can use our cluster if they can't find a free cluster,
452 * check if there is still free entry in the cluster
455 while (tmp
< si
->max
&& tmp
< (cluster_next(&cluster
->index
) + 1) *
457 if (!si
->swap_map
[tmp
]) {
464 cluster_set_null(&cluster
->index
);
467 cluster
->next
= tmp
+ 1;
472 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
475 unsigned long offset
;
476 unsigned long scan_base
;
477 unsigned long last_in_cluster
= 0;
478 int latency_ration
= LATENCY_LIMIT
;
481 * We try to cluster swap pages by allocating them sequentially
482 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
483 * way, however, we resort to first-free allocation, starting
484 * a new cluster. This prevents us from scattering swap pages
485 * all over the entire swap partition, so that we reduce
486 * overall disk seek times between swap pages. -- sct
487 * But we do now try to find an empty cluster. -Andrea
488 * And we let swap pages go all over an SSD partition. Hugh
491 si
->flags
+= SWP_SCANNING
;
492 scan_base
= offset
= si
->cluster_next
;
495 if (si
->cluster_info
) {
496 scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
);
500 if (unlikely(!si
->cluster_nr
--)) {
501 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
502 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
506 spin_unlock(&si
->lock
);
509 * If seek is expensive, start searching for new cluster from
510 * start of partition, to minimize the span of allocated swap.
511 * But if seek is cheap, search from our current position, so
512 * that swap is allocated from all over the partition: if the
513 * Flash Translation Layer only remaps within limited zones,
514 * we don't want to wear out the first zone too quickly.
516 if (!(si
->flags
& SWP_SOLIDSTATE
))
517 scan_base
= offset
= si
->lowest_bit
;
518 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
520 /* Locate the first empty (unaligned) cluster */
521 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
522 if (si
->swap_map
[offset
])
523 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
524 else if (offset
== last_in_cluster
) {
525 spin_lock(&si
->lock
);
526 offset
-= SWAPFILE_CLUSTER
- 1;
527 si
->cluster_next
= offset
;
528 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
531 if (unlikely(--latency_ration
< 0)) {
533 latency_ration
= LATENCY_LIMIT
;
537 offset
= si
->lowest_bit
;
538 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
540 /* Locate the first empty (unaligned) cluster */
541 for (; last_in_cluster
< scan_base
; offset
++) {
542 if (si
->swap_map
[offset
])
543 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
544 else if (offset
== last_in_cluster
) {
545 spin_lock(&si
->lock
);
546 offset
-= SWAPFILE_CLUSTER
- 1;
547 si
->cluster_next
= offset
;
548 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
551 if (unlikely(--latency_ration
< 0)) {
553 latency_ration
= LATENCY_LIMIT
;
558 spin_lock(&si
->lock
);
559 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
563 if (si
->cluster_info
) {
564 while (scan_swap_map_ssd_cluster_conflict(si
, offset
))
565 scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
);
567 if (!(si
->flags
& SWP_WRITEOK
))
569 if (!si
->highest_bit
)
571 if (offset
> si
->highest_bit
)
572 scan_base
= offset
= si
->lowest_bit
;
574 /* reuse swap entry of cache-only swap if not busy. */
575 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
577 spin_unlock(&si
->lock
);
578 swap_was_freed
= __try_to_reclaim_swap(si
, offset
);
579 spin_lock(&si
->lock
);
580 /* entry was freed successfully, try to use this again */
583 goto scan
; /* check next one */
586 if (si
->swap_map
[offset
])
589 if (offset
== si
->lowest_bit
)
591 if (offset
== si
->highest_bit
)
594 if (si
->inuse_pages
== si
->pages
) {
595 si
->lowest_bit
= si
->max
;
598 si
->swap_map
[offset
] = usage
;
599 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
600 si
->cluster_next
= offset
+ 1;
601 si
->flags
-= SWP_SCANNING
;
606 spin_unlock(&si
->lock
);
607 while (++offset
<= si
->highest_bit
) {
608 if (!si
->swap_map
[offset
]) {
609 spin_lock(&si
->lock
);
612 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
613 spin_lock(&si
->lock
);
616 if (unlikely(--latency_ration
< 0)) {
618 latency_ration
= LATENCY_LIMIT
;
621 offset
= si
->lowest_bit
;
622 while (offset
< scan_base
) {
623 if (!si
->swap_map
[offset
]) {
624 spin_lock(&si
->lock
);
627 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
628 spin_lock(&si
->lock
);
631 if (unlikely(--latency_ration
< 0)) {
633 latency_ration
= LATENCY_LIMIT
;
637 spin_lock(&si
->lock
);
640 si
->flags
-= SWP_SCANNING
;
644 swp_entry_t
get_swap_page(void)
646 struct swap_info_struct
*si
, *next
;
648 struct list_head
*tmp
;
650 spin_lock(&swap_lock
);
651 if (atomic_long_read(&nr_swap_pages
) <= 0)
653 atomic_long_dec(&nr_swap_pages
);
655 list_for_each(tmp
, &swap_list_head
) {
656 si
= list_entry(tmp
, typeof(*si
), list
);
657 spin_lock(&si
->lock
);
658 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
659 spin_unlock(&si
->lock
);
664 * rotate the current swap_info that we're going to use
665 * to after any other swap_info that have the same prio,
666 * so that all equal-priority swap_info get used equally
669 list_for_each_entry_continue(next
, &swap_list_head
, list
) {
670 if (si
->prio
!= next
->prio
)
672 list_rotate_left(&si
->list
);
676 spin_unlock(&swap_lock
);
677 /* This is called for allocating swap entry for cache */
678 offset
= scan_swap_map(si
, SWAP_HAS_CACHE
);
679 spin_unlock(&si
->lock
);
681 return swp_entry(si
->type
, offset
);
682 spin_lock(&swap_lock
);
684 * if we got here, it's likely that si was almost full before,
685 * and since scan_swap_map() can drop the si->lock, multiple
686 * callers probably all tried to get a page from the same si
687 * and it filled up before we could get one. So we need to
688 * try again. Since we dropped the swap_lock, there may now
689 * be non-full higher priority swap_infos, and this si may have
690 * even been removed from the list (although very unlikely).
693 tmp
= &swap_list_head
;
696 atomic_long_inc(&nr_swap_pages
);
698 spin_unlock(&swap_lock
);
699 return (swp_entry_t
) {0};
702 /* The only caller of this function is now suspend routine */
703 swp_entry_t
get_swap_page_of_type(int type
)
705 struct swap_info_struct
*si
;
708 si
= swap_info
[type
];
709 spin_lock(&si
->lock
);
710 if (si
&& (si
->flags
& SWP_WRITEOK
)) {
711 atomic_long_dec(&nr_swap_pages
);
712 /* This is called for allocating swap entry, not cache */
713 offset
= scan_swap_map(si
, 1);
715 spin_unlock(&si
->lock
);
716 return swp_entry(type
, offset
);
718 atomic_long_inc(&nr_swap_pages
);
720 spin_unlock(&si
->lock
);
721 return (swp_entry_t
) {0};
724 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
726 struct swap_info_struct
*p
;
727 unsigned long offset
, type
;
731 type
= swp_type(entry
);
732 if (type
>= nr_swapfiles
)
735 if (!(p
->flags
& SWP_USED
))
737 offset
= swp_offset(entry
);
738 if (offset
>= p
->max
)
740 if (!p
->swap_map
[offset
])
746 pr_err("swap_free: %s%08lx\n", Unused_offset
, entry
.val
);
749 pr_err("swap_free: %s%08lx\n", Bad_offset
, entry
.val
);
752 pr_err("swap_free: %s%08lx\n", Unused_file
, entry
.val
);
755 pr_err("swap_free: %s%08lx\n", Bad_file
, entry
.val
);
760 static unsigned char swap_entry_free(struct swap_info_struct
*p
,
761 swp_entry_t entry
, unsigned char usage
)
763 unsigned long offset
= swp_offset(entry
);
765 unsigned char has_cache
;
767 count
= p
->swap_map
[offset
];
768 has_cache
= count
& SWAP_HAS_CACHE
;
769 count
&= ~SWAP_HAS_CACHE
;
771 if (usage
== SWAP_HAS_CACHE
) {
772 VM_BUG_ON(!has_cache
);
774 } else if (count
== SWAP_MAP_SHMEM
) {
776 * Or we could insist on shmem.c using a special
777 * swap_shmem_free() and free_shmem_swap_and_cache()...
780 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
781 if (count
== COUNT_CONTINUED
) {
782 if (swap_count_continued(p
, offset
, count
))
783 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
785 count
= SWAP_MAP_MAX
;
791 mem_cgroup_uncharge_swap(entry
);
793 usage
= count
| has_cache
;
794 p
->swap_map
[offset
] = usage
;
796 /* free if no reference */
798 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
799 if (offset
< p
->lowest_bit
)
800 p
->lowest_bit
= offset
;
801 if (offset
> p
->highest_bit
)
802 p
->highest_bit
= offset
;
803 atomic_long_inc(&nr_swap_pages
);
805 frontswap_invalidate_page(p
->type
, offset
);
806 if (p
->flags
& SWP_BLKDEV
) {
807 struct gendisk
*disk
= p
->bdev
->bd_disk
;
808 if (disk
->fops
->swap_slot_free_notify
)
809 disk
->fops
->swap_slot_free_notify(p
->bdev
,
818 * Caller has made sure that the swap device corresponding to entry
819 * is still around or has not been recycled.
821 void swap_free(swp_entry_t entry
)
823 struct swap_info_struct
*p
;
825 p
= swap_info_get(entry
);
827 swap_entry_free(p
, entry
, 1);
828 spin_unlock(&p
->lock
);
833 * Called after dropping swapcache to decrease refcnt to swap entries.
835 void swapcache_free(swp_entry_t entry
, struct page
*page
)
837 struct swap_info_struct
*p
;
840 p
= swap_info_get(entry
);
842 count
= swap_entry_free(p
, entry
, SWAP_HAS_CACHE
);
844 mem_cgroup_uncharge_swapcache(page
, entry
, count
!= 0);
845 spin_unlock(&p
->lock
);
850 * How many references to page are currently swapped out?
851 * This does not give an exact answer when swap count is continued,
852 * but does include the high COUNT_CONTINUED flag to allow for that.
854 int page_swapcount(struct page
*page
)
857 struct swap_info_struct
*p
;
860 entry
.val
= page_private(page
);
861 p
= swap_info_get(entry
);
863 count
= swap_count(p
->swap_map
[swp_offset(entry
)]);
864 spin_unlock(&p
->lock
);
870 * We can write to an anon page without COW if there are no other references
871 * to it. And as a side-effect, free up its swap: because the old content
872 * on disk will never be read, and seeking back there to write new content
873 * later would only waste time away from clustering.
875 int reuse_swap_page(struct page
*page
)
879 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
880 if (unlikely(PageKsm(page
)))
882 count
= page_mapcount(page
);
883 if (count
<= 1 && PageSwapCache(page
)) {
884 count
+= page_swapcount(page
);
885 if (count
== 1 && !PageWriteback(page
)) {
886 delete_from_swap_cache(page
);
894 * If swap is getting full, or if there are no more mappings of this page,
895 * then try_to_free_swap is called to free its swap space.
897 int try_to_free_swap(struct page
*page
)
899 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
901 if (!PageSwapCache(page
))
903 if (PageWriteback(page
))
905 if (page_swapcount(page
))
909 * Once hibernation has begun to create its image of memory,
910 * there's a danger that one of the calls to try_to_free_swap()
911 * - most probably a call from __try_to_reclaim_swap() while
912 * hibernation is allocating its own swap pages for the image,
913 * but conceivably even a call from memory reclaim - will free
914 * the swap from a page which has already been recorded in the
915 * image as a clean swapcache page, and then reuse its swap for
916 * another page of the image. On waking from hibernation, the
917 * original page might be freed under memory pressure, then
918 * later read back in from swap, now with the wrong data.
920 * Hibernation suspends storage while it is writing the image
921 * to disk so check that here.
923 if (pm_suspended_storage())
926 delete_from_swap_cache(page
);
932 * Free the swap entry like above, but also try to
933 * free the page cache entry if it is the last user.
935 int free_swap_and_cache(swp_entry_t entry
)
937 struct swap_info_struct
*p
;
938 struct page
*page
= NULL
;
940 if (non_swap_entry(entry
))
943 p
= swap_info_get(entry
);
945 if (swap_entry_free(p
, entry
, 1) == SWAP_HAS_CACHE
) {
946 page
= find_get_page(swap_address_space(entry
),
948 if (page
&& !trylock_page(page
)) {
949 page_cache_release(page
);
953 spin_unlock(&p
->lock
);
957 * Not mapped elsewhere, or swap space full? Free it!
958 * Also recheck PageSwapCache now page is locked (above).
960 if (PageSwapCache(page
) && !PageWriteback(page
) &&
961 (!page_mapped(page
) || vm_swap_full())) {
962 delete_from_swap_cache(page
);
966 page_cache_release(page
);
971 #ifdef CONFIG_HIBERNATION
973 * Find the swap type that corresponds to given device (if any).
975 * @offset - number of the PAGE_SIZE-sized block of the device, starting
976 * from 0, in which the swap header is expected to be located.
978 * This is needed for the suspend to disk (aka swsusp).
980 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
982 struct block_device
*bdev
= NULL
;
986 bdev
= bdget(device
);
988 spin_lock(&swap_lock
);
989 for (type
= 0; type
< nr_swapfiles
; type
++) {
990 struct swap_info_struct
*sis
= swap_info
[type
];
992 if (!(sis
->flags
& SWP_WRITEOK
))
997 *bdev_p
= bdgrab(sis
->bdev
);
999 spin_unlock(&swap_lock
);
1002 if (bdev
== sis
->bdev
) {
1003 struct swap_extent
*se
= &sis
->first_swap_extent
;
1005 if (se
->start_block
== offset
) {
1007 *bdev_p
= bdgrab(sis
->bdev
);
1009 spin_unlock(&swap_lock
);
1015 spin_unlock(&swap_lock
);
1023 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1024 * corresponding to given index in swap_info (swap type).
1026 sector_t
swapdev_block(int type
, pgoff_t offset
)
1028 struct block_device
*bdev
;
1030 if ((unsigned int)type
>= nr_swapfiles
)
1032 if (!(swap_info
[type
]->flags
& SWP_WRITEOK
))
1034 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
1038 * Return either the total number of swap pages of given type, or the number
1039 * of free pages of that type (depending on @free)
1041 * This is needed for software suspend
1043 unsigned int count_swap_pages(int type
, int free
)
1047 spin_lock(&swap_lock
);
1048 if ((unsigned int)type
< nr_swapfiles
) {
1049 struct swap_info_struct
*sis
= swap_info
[type
];
1051 spin_lock(&sis
->lock
);
1052 if (sis
->flags
& SWP_WRITEOK
) {
1055 n
-= sis
->inuse_pages
;
1057 spin_unlock(&sis
->lock
);
1059 spin_unlock(&swap_lock
);
1062 #endif /* CONFIG_HIBERNATION */
1064 static inline int maybe_same_pte(pte_t pte
, pte_t swp_pte
)
1066 #ifdef CONFIG_MEM_SOFT_DIRTY
1068 * When pte keeps soft dirty bit the pte generated
1069 * from swap entry does not has it, still it's same
1070 * pte from logical point of view.
1072 pte_t swp_pte_dirty
= pte_swp_mksoft_dirty(swp_pte
);
1073 return pte_same(pte
, swp_pte
) || pte_same(pte
, swp_pte_dirty
);
1075 return pte_same(pte
, swp_pte
);
1080 * No need to decide whether this PTE shares the swap entry with others,
1081 * just let do_wp_page work it out if a write is requested later - to
1082 * force COW, vm_page_prot omits write permission from any private vma.
1084 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1085 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1087 struct page
*swapcache
;
1088 struct mem_cgroup
*memcg
;
1094 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1095 if (unlikely(!page
))
1098 if (mem_cgroup_try_charge_swapin(vma
->vm_mm
, page
,
1099 GFP_KERNEL
, &memcg
)) {
1104 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1105 if (unlikely(!maybe_same_pte(*pte
, swp_entry_to_pte(entry
)))) {
1106 mem_cgroup_cancel_charge_swapin(memcg
);
1111 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1112 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1114 set_pte_at(vma
->vm_mm
, addr
, pte
,
1115 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1116 if (page
== swapcache
)
1117 page_add_anon_rmap(page
, vma
, addr
);
1118 else /* ksm created a completely new copy */
1119 page_add_new_anon_rmap(page
, vma
, addr
);
1120 mem_cgroup_commit_charge_swapin(page
, memcg
);
1123 * Move the page to the active list so it is not
1124 * immediately swapped out again after swapon.
1126 activate_page(page
);
1128 pte_unmap_unlock(pte
, ptl
);
1130 if (page
!= swapcache
) {
1137 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1138 unsigned long addr
, unsigned long end
,
1139 swp_entry_t entry
, struct page
*page
)
1141 pte_t swp_pte
= swp_entry_to_pte(entry
);
1146 * We don't actually need pte lock while scanning for swp_pte: since
1147 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1148 * page table while we're scanning; though it could get zapped, and on
1149 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1150 * of unmatched parts which look like swp_pte, so unuse_pte must
1151 * recheck under pte lock. Scanning without pte lock lets it be
1152 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1154 pte
= pte_offset_map(pmd
, addr
);
1157 * swapoff spends a _lot_ of time in this loop!
1158 * Test inline before going to call unuse_pte.
1160 if (unlikely(maybe_same_pte(*pte
, swp_pte
))) {
1162 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1165 pte
= pte_offset_map(pmd
, addr
);
1167 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1173 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
1174 unsigned long addr
, unsigned long end
,
1175 swp_entry_t entry
, struct page
*page
)
1181 pmd
= pmd_offset(pud
, addr
);
1183 next
= pmd_addr_end(addr
, end
);
1184 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1186 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
);
1189 } while (pmd
++, addr
= next
, addr
!= end
);
1193 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
1194 unsigned long addr
, unsigned long end
,
1195 swp_entry_t entry
, struct page
*page
)
1201 pud
= pud_offset(pgd
, addr
);
1203 next
= pud_addr_end(addr
, end
);
1204 if (pud_none_or_clear_bad(pud
))
1206 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
);
1209 } while (pud
++, addr
= next
, addr
!= end
);
1213 static int unuse_vma(struct vm_area_struct
*vma
,
1214 swp_entry_t entry
, struct page
*page
)
1217 unsigned long addr
, end
, next
;
1220 if (page_anon_vma(page
)) {
1221 addr
= page_address_in_vma(page
, vma
);
1222 if (addr
== -EFAULT
)
1225 end
= addr
+ PAGE_SIZE
;
1227 addr
= vma
->vm_start
;
1231 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1233 next
= pgd_addr_end(addr
, end
);
1234 if (pgd_none_or_clear_bad(pgd
))
1236 ret
= unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
);
1239 } while (pgd
++, addr
= next
, addr
!= end
);
1243 static int unuse_mm(struct mm_struct
*mm
,
1244 swp_entry_t entry
, struct page
*page
)
1246 struct vm_area_struct
*vma
;
1249 if (!down_read_trylock(&mm
->mmap_sem
)) {
1251 * Activate page so shrink_inactive_list is unlikely to unmap
1252 * its ptes while lock is dropped, so swapoff can make progress.
1254 activate_page(page
);
1256 down_read(&mm
->mmap_sem
);
1259 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1260 if (vma
->anon_vma
&& (ret
= unuse_vma(vma
, entry
, page
)))
1263 up_read(&mm
->mmap_sem
);
1264 return (ret
< 0)? ret
: 0;
1268 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1269 * from current position to next entry still in use.
1270 * Recycle to start on reaching the end, returning 0 when empty.
1272 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
1273 unsigned int prev
, bool frontswap
)
1275 unsigned int max
= si
->max
;
1276 unsigned int i
= prev
;
1277 unsigned char count
;
1280 * No need for swap_lock here: we're just looking
1281 * for whether an entry is in use, not modifying it; false
1282 * hits are okay, and sys_swapoff() has already prevented new
1283 * allocations from this area (while holding swap_lock).
1292 * No entries in use at top of swap_map,
1293 * loop back to start and recheck there.
1300 if (frontswap_test(si
, i
))
1305 count
= ACCESS_ONCE(si
->swap_map
[i
]);
1306 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
1313 * We completely avoid races by reading each swap page in advance,
1314 * and then search for the process using it. All the necessary
1315 * page table adjustments can then be made atomically.
1317 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1318 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1320 int try_to_unuse(unsigned int type
, bool frontswap
,
1321 unsigned long pages_to_unuse
)
1323 struct swap_info_struct
*si
= swap_info
[type
];
1324 struct mm_struct
*start_mm
;
1325 volatile unsigned char *swap_map
; /* swap_map is accessed without
1326 * locking. Mark it as volatile
1327 * to prevent compiler doing
1330 unsigned char swcount
;
1337 * When searching mms for an entry, a good strategy is to
1338 * start at the first mm we freed the previous entry from
1339 * (though actually we don't notice whether we or coincidence
1340 * freed the entry). Initialize this start_mm with a hold.
1342 * A simpler strategy would be to start at the last mm we
1343 * freed the previous entry from; but that would take less
1344 * advantage of mmlist ordering, which clusters forked mms
1345 * together, child after parent. If we race with dup_mmap(), we
1346 * prefer to resolve parent before child, lest we miss entries
1347 * duplicated after we scanned child: using last mm would invert
1350 start_mm
= &init_mm
;
1351 atomic_inc(&init_mm
.mm_users
);
1354 * Keep on scanning until all entries have gone. Usually,
1355 * one pass through swap_map is enough, but not necessarily:
1356 * there are races when an instance of an entry might be missed.
1358 while ((i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
1359 if (signal_pending(current
)) {
1365 * Get a page for the entry, using the existing swap
1366 * cache page if there is one. Otherwise, get a clean
1367 * page and read the swap into it.
1369 swap_map
= &si
->swap_map
[i
];
1370 entry
= swp_entry(type
, i
);
1371 page
= read_swap_cache_async(entry
,
1372 GFP_HIGHUSER_MOVABLE
, NULL
, 0);
1375 * Either swap_duplicate() failed because entry
1376 * has been freed independently, and will not be
1377 * reused since sys_swapoff() already disabled
1378 * allocation from here, or alloc_page() failed.
1380 swcount
= *swap_map
;
1382 * We don't hold lock here, so the swap entry could be
1383 * SWAP_MAP_BAD (when the cluster is discarding).
1384 * Instead of fail out, We can just skip the swap
1385 * entry because swapoff will wait for discarding
1388 if (!swcount
|| swcount
== SWAP_MAP_BAD
)
1395 * Don't hold on to start_mm if it looks like exiting.
1397 if (atomic_read(&start_mm
->mm_users
) == 1) {
1399 start_mm
= &init_mm
;
1400 atomic_inc(&init_mm
.mm_users
);
1404 * Wait for and lock page. When do_swap_page races with
1405 * try_to_unuse, do_swap_page can handle the fault much
1406 * faster than try_to_unuse can locate the entry. This
1407 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1408 * defer to do_swap_page in such a case - in some tests,
1409 * do_swap_page and try_to_unuse repeatedly compete.
1411 wait_on_page_locked(page
);
1412 wait_on_page_writeback(page
);
1414 wait_on_page_writeback(page
);
1417 * Remove all references to entry.
1419 swcount
= *swap_map
;
1420 if (swap_count(swcount
) == SWAP_MAP_SHMEM
) {
1421 retval
= shmem_unuse(entry
, page
);
1422 /* page has already been unlocked and released */
1427 if (swap_count(swcount
) && start_mm
!= &init_mm
)
1428 retval
= unuse_mm(start_mm
, entry
, page
);
1430 if (swap_count(*swap_map
)) {
1431 int set_start_mm
= (*swap_map
>= swcount
);
1432 struct list_head
*p
= &start_mm
->mmlist
;
1433 struct mm_struct
*new_start_mm
= start_mm
;
1434 struct mm_struct
*prev_mm
= start_mm
;
1435 struct mm_struct
*mm
;
1437 atomic_inc(&new_start_mm
->mm_users
);
1438 atomic_inc(&prev_mm
->mm_users
);
1439 spin_lock(&mmlist_lock
);
1440 while (swap_count(*swap_map
) && !retval
&&
1441 (p
= p
->next
) != &start_mm
->mmlist
) {
1442 mm
= list_entry(p
, struct mm_struct
, mmlist
);
1443 if (!atomic_inc_not_zero(&mm
->mm_users
))
1445 spin_unlock(&mmlist_lock
);
1451 swcount
= *swap_map
;
1452 if (!swap_count(swcount
)) /* any usage ? */
1454 else if (mm
== &init_mm
)
1457 retval
= unuse_mm(mm
, entry
, page
);
1459 if (set_start_mm
&& *swap_map
< swcount
) {
1460 mmput(new_start_mm
);
1461 atomic_inc(&mm
->mm_users
);
1465 spin_lock(&mmlist_lock
);
1467 spin_unlock(&mmlist_lock
);
1470 start_mm
= new_start_mm
;
1474 page_cache_release(page
);
1479 * If a reference remains (rare), we would like to leave
1480 * the page in the swap cache; but try_to_unmap could
1481 * then re-duplicate the entry once we drop page lock,
1482 * so we might loop indefinitely; also, that page could
1483 * not be swapped out to other storage meanwhile. So:
1484 * delete from cache even if there's another reference,
1485 * after ensuring that the data has been saved to disk -
1486 * since if the reference remains (rarer), it will be
1487 * read from disk into another page. Splitting into two
1488 * pages would be incorrect if swap supported "shared
1489 * private" pages, but they are handled by tmpfs files.
1491 * Given how unuse_vma() targets one particular offset
1492 * in an anon_vma, once the anon_vma has been determined,
1493 * this splitting happens to be just what is needed to
1494 * handle where KSM pages have been swapped out: re-reading
1495 * is unnecessarily slow, but we can fix that later on.
1497 if (swap_count(*swap_map
) &&
1498 PageDirty(page
) && PageSwapCache(page
)) {
1499 struct writeback_control wbc
= {
1500 .sync_mode
= WB_SYNC_NONE
,
1503 swap_writepage(page
, &wbc
);
1505 wait_on_page_writeback(page
);
1509 * It is conceivable that a racing task removed this page from
1510 * swap cache just before we acquired the page lock at the top,
1511 * or while we dropped it in unuse_mm(). The page might even
1512 * be back in swap cache on another swap area: that we must not
1513 * delete, since it may not have been written out to swap yet.
1515 if (PageSwapCache(page
) &&
1516 likely(page_private(page
) == entry
.val
))
1517 delete_from_swap_cache(page
);
1520 * So we could skip searching mms once swap count went
1521 * to 1, we did not mark any present ptes as dirty: must
1522 * mark page dirty so shrink_page_list will preserve it.
1526 page_cache_release(page
);
1529 * Make sure that we aren't completely killing
1530 * interactive performance.
1533 if (frontswap
&& pages_to_unuse
> 0) {
1534 if (!--pages_to_unuse
)
1544 * After a successful try_to_unuse, if no swap is now in use, we know
1545 * we can empty the mmlist. swap_lock must be held on entry and exit.
1546 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1547 * added to the mmlist just after page_duplicate - before would be racy.
1549 static void drain_mmlist(void)
1551 struct list_head
*p
, *next
;
1554 for (type
= 0; type
< nr_swapfiles
; type
++)
1555 if (swap_info
[type
]->inuse_pages
)
1557 spin_lock(&mmlist_lock
);
1558 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
1560 spin_unlock(&mmlist_lock
);
1564 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1565 * corresponds to page offset for the specified swap entry.
1566 * Note that the type of this function is sector_t, but it returns page offset
1567 * into the bdev, not sector offset.
1569 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
1571 struct swap_info_struct
*sis
;
1572 struct swap_extent
*start_se
;
1573 struct swap_extent
*se
;
1576 sis
= swap_info
[swp_type(entry
)];
1579 offset
= swp_offset(entry
);
1580 start_se
= sis
->curr_swap_extent
;
1584 struct list_head
*lh
;
1586 if (se
->start_page
<= offset
&&
1587 offset
< (se
->start_page
+ se
->nr_pages
)) {
1588 return se
->start_block
+ (offset
- se
->start_page
);
1591 se
= list_entry(lh
, struct swap_extent
, list
);
1592 sis
->curr_swap_extent
= se
;
1593 BUG_ON(se
== start_se
); /* It *must* be present */
1598 * Returns the page offset into bdev for the specified page's swap entry.
1600 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
1603 entry
.val
= page_private(page
);
1604 return map_swap_entry(entry
, bdev
);
1608 * Free all of a swapdev's extent information
1610 static void destroy_swap_extents(struct swap_info_struct
*sis
)
1612 while (!list_empty(&sis
->first_swap_extent
.list
)) {
1613 struct swap_extent
*se
;
1615 se
= list_entry(sis
->first_swap_extent
.list
.next
,
1616 struct swap_extent
, list
);
1617 list_del(&se
->list
);
1621 if (sis
->flags
& SWP_FILE
) {
1622 struct file
*swap_file
= sis
->swap_file
;
1623 struct address_space
*mapping
= swap_file
->f_mapping
;
1625 sis
->flags
&= ~SWP_FILE
;
1626 mapping
->a_ops
->swap_deactivate(swap_file
);
1631 * Add a block range (and the corresponding page range) into this swapdev's
1632 * extent list. The extent list is kept sorted in page order.
1634 * This function rather assumes that it is called in ascending page order.
1637 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
1638 unsigned long nr_pages
, sector_t start_block
)
1640 struct swap_extent
*se
;
1641 struct swap_extent
*new_se
;
1642 struct list_head
*lh
;
1644 if (start_page
== 0) {
1645 se
= &sis
->first_swap_extent
;
1646 sis
->curr_swap_extent
= se
;
1648 se
->nr_pages
= nr_pages
;
1649 se
->start_block
= start_block
;
1652 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
1653 se
= list_entry(lh
, struct swap_extent
, list
);
1654 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
1655 if (se
->start_block
+ se
->nr_pages
== start_block
) {
1657 se
->nr_pages
+= nr_pages
;
1663 * No merge. Insert a new extent, preserving ordering.
1665 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
1668 new_se
->start_page
= start_page
;
1669 new_se
->nr_pages
= nr_pages
;
1670 new_se
->start_block
= start_block
;
1672 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
1677 * A `swap extent' is a simple thing which maps a contiguous range of pages
1678 * onto a contiguous range of disk blocks. An ordered list of swap extents
1679 * is built at swapon time and is then used at swap_writepage/swap_readpage
1680 * time for locating where on disk a page belongs.
1682 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1683 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1684 * swap files identically.
1686 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1687 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1688 * swapfiles are handled *identically* after swapon time.
1690 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1691 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1692 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1693 * requirements, they are simply tossed out - we will never use those blocks
1696 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1697 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1698 * which will scribble on the fs.
1700 * The amount of disk space which a single swap extent represents varies.
1701 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1702 * extents in the list. To avoid much list walking, we cache the previous
1703 * search location in `curr_swap_extent', and start new searches from there.
1704 * This is extremely effective. The average number of iterations in
1705 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1707 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
1709 struct file
*swap_file
= sis
->swap_file
;
1710 struct address_space
*mapping
= swap_file
->f_mapping
;
1711 struct inode
*inode
= mapping
->host
;
1714 if (S_ISBLK(inode
->i_mode
)) {
1715 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1720 if (mapping
->a_ops
->swap_activate
) {
1721 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
1723 sis
->flags
|= SWP_FILE
;
1724 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1730 return generic_swapfile_activate(sis
, swap_file
, span
);
1733 static void _enable_swap_info(struct swap_info_struct
*p
, int prio
,
1734 unsigned char *swap_map
,
1735 struct swap_cluster_info
*cluster_info
)
1737 struct swap_info_struct
*si
;
1742 p
->prio
= --least_priority
;
1743 p
->swap_map
= swap_map
;
1744 p
->cluster_info
= cluster_info
;
1745 p
->flags
|= SWP_WRITEOK
;
1746 atomic_long_add(p
->pages
, &nr_swap_pages
);
1747 total_swap_pages
+= p
->pages
;
1749 assert_spin_locked(&swap_lock
);
1750 BUG_ON(!list_empty(&p
->list
));
1752 * insert into swap list; the list is in priority order,
1753 * so that get_swap_page() can get a page from the highest
1754 * priority swap_info_struct with available page(s), and
1755 * swapoff can adjust the auto-assigned (i.e. negative) prio
1756 * values for any lower-priority swap_info_structs when
1757 * removing a negative-prio swap_info_struct
1759 list_for_each_entry(si
, &swap_list_head
, list
) {
1760 if (p
->prio
>= si
->prio
) {
1761 list_add_tail(&p
->list
, &si
->list
);
1766 * this covers two cases:
1767 * 1) p->prio is less than all existing prio
1768 * 2) the swap list is empty
1770 list_add_tail(&p
->list
, &swap_list_head
);
1773 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
1774 unsigned char *swap_map
,
1775 struct swap_cluster_info
*cluster_info
,
1776 unsigned long *frontswap_map
)
1778 frontswap_init(p
->type
, frontswap_map
);
1779 spin_lock(&swap_lock
);
1780 spin_lock(&p
->lock
);
1781 _enable_swap_info(p
, prio
, swap_map
, cluster_info
);
1782 spin_unlock(&p
->lock
);
1783 spin_unlock(&swap_lock
);
1786 static void reinsert_swap_info(struct swap_info_struct
*p
)
1788 spin_lock(&swap_lock
);
1789 spin_lock(&p
->lock
);
1790 _enable_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
1791 spin_unlock(&p
->lock
);
1792 spin_unlock(&swap_lock
);
1795 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
1797 struct swap_info_struct
*p
= NULL
;
1798 unsigned char *swap_map
;
1799 struct swap_cluster_info
*cluster_info
;
1800 unsigned long *frontswap_map
;
1801 struct file
*swap_file
, *victim
;
1802 struct address_space
*mapping
;
1803 struct inode
*inode
;
1804 struct filename
*pathname
;
1806 unsigned int old_block_size
;
1808 if (!capable(CAP_SYS_ADMIN
))
1811 BUG_ON(!current
->mm
);
1813 pathname
= getname(specialfile
);
1814 if (IS_ERR(pathname
))
1815 return PTR_ERR(pathname
);
1817 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
1818 err
= PTR_ERR(victim
);
1822 mapping
= victim
->f_mapping
;
1823 spin_lock(&swap_lock
);
1824 list_for_each_entry(p
, &swap_list_head
, list
) {
1825 if (p
->flags
& SWP_WRITEOK
) {
1826 if (p
->swap_file
->f_mapping
== mapping
) {
1834 spin_unlock(&swap_lock
);
1837 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
1838 vm_unacct_memory(p
->pages
);
1841 spin_unlock(&swap_lock
);
1844 spin_lock(&p
->lock
);
1846 struct swap_info_struct
*si
= p
;
1848 list_for_each_entry_continue(si
, &swap_list_head
, list
) {
1853 list_del_init(&p
->list
);
1854 atomic_long_sub(p
->pages
, &nr_swap_pages
);
1855 total_swap_pages
-= p
->pages
;
1856 p
->flags
&= ~SWP_WRITEOK
;
1857 spin_unlock(&p
->lock
);
1858 spin_unlock(&swap_lock
);
1860 set_current_oom_origin();
1861 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
1862 clear_current_oom_origin();
1865 /* re-insert swap space back into swap_list */
1866 reinsert_swap_info(p
);
1870 flush_work(&p
->discard_work
);
1872 destroy_swap_extents(p
);
1873 if (p
->flags
& SWP_CONTINUED
)
1874 free_swap_count_continuations(p
);
1876 mutex_lock(&swapon_mutex
);
1877 spin_lock(&swap_lock
);
1878 spin_lock(&p
->lock
);
1881 /* wait for anyone still in scan_swap_map */
1882 p
->highest_bit
= 0; /* cuts scans short */
1883 while (p
->flags
>= SWP_SCANNING
) {
1884 spin_unlock(&p
->lock
);
1885 spin_unlock(&swap_lock
);
1886 schedule_timeout_uninterruptible(1);
1887 spin_lock(&swap_lock
);
1888 spin_lock(&p
->lock
);
1891 swap_file
= p
->swap_file
;
1892 old_block_size
= p
->old_block_size
;
1893 p
->swap_file
= NULL
;
1895 swap_map
= p
->swap_map
;
1897 cluster_info
= p
->cluster_info
;
1898 p
->cluster_info
= NULL
;
1899 frontswap_map
= frontswap_map_get(p
);
1900 spin_unlock(&p
->lock
);
1901 spin_unlock(&swap_lock
);
1902 frontswap_invalidate_area(p
->type
);
1903 frontswap_map_set(p
, NULL
);
1904 mutex_unlock(&swapon_mutex
);
1905 free_percpu(p
->percpu_cluster
);
1906 p
->percpu_cluster
= NULL
;
1908 vfree(cluster_info
);
1909 vfree(frontswap_map
);
1910 /* Destroy swap account information */
1911 swap_cgroup_swapoff(p
->type
);
1913 inode
= mapping
->host
;
1914 if (S_ISBLK(inode
->i_mode
)) {
1915 struct block_device
*bdev
= I_BDEV(inode
);
1916 set_blocksize(bdev
, old_block_size
);
1917 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1919 mutex_lock(&inode
->i_mutex
);
1920 inode
->i_flags
&= ~S_SWAPFILE
;
1921 mutex_unlock(&inode
->i_mutex
);
1923 filp_close(swap_file
, NULL
);
1926 * Clear the SWP_USED flag after all resources are freed so that swapon
1927 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1928 * not hold p->lock after we cleared its SWP_WRITEOK.
1930 spin_lock(&swap_lock
);
1932 spin_unlock(&swap_lock
);
1935 atomic_inc(&proc_poll_event
);
1936 wake_up_interruptible(&proc_poll_wait
);
1939 filp_close(victim
, NULL
);
1945 #ifdef CONFIG_PROC_FS
1946 static unsigned swaps_poll(struct file
*file
, poll_table
*wait
)
1948 struct seq_file
*seq
= file
->private_data
;
1950 poll_wait(file
, &proc_poll_wait
, wait
);
1952 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
1953 seq
->poll_event
= atomic_read(&proc_poll_event
);
1954 return POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
1957 return POLLIN
| POLLRDNORM
;
1961 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
1963 struct swap_info_struct
*si
;
1967 mutex_lock(&swapon_mutex
);
1970 return SEQ_START_TOKEN
;
1972 for (type
= 0; type
< nr_swapfiles
; type
++) {
1973 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1974 si
= swap_info
[type
];
1975 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
1984 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
1986 struct swap_info_struct
*si
= v
;
1989 if (v
== SEQ_START_TOKEN
)
1992 type
= si
->type
+ 1;
1994 for (; type
< nr_swapfiles
; type
++) {
1995 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1996 si
= swap_info
[type
];
1997 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2006 static void swap_stop(struct seq_file
*swap
, void *v
)
2008 mutex_unlock(&swapon_mutex
);
2011 static int swap_show(struct seq_file
*swap
, void *v
)
2013 struct swap_info_struct
*si
= v
;
2017 if (si
== SEQ_START_TOKEN
) {
2018 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2022 file
= si
->swap_file
;
2023 len
= seq_path(swap
, &file
->f_path
, " \t\n\\");
2024 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
2025 len
< 40 ? 40 - len
: 1, " ",
2026 S_ISBLK(file_inode(file
)->i_mode
) ?
2027 "partition" : "file\t",
2028 si
->pages
<< (PAGE_SHIFT
- 10),
2029 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
2034 static const struct seq_operations swaps_op
= {
2035 .start
= swap_start
,
2041 static int swaps_open(struct inode
*inode
, struct file
*file
)
2043 struct seq_file
*seq
;
2046 ret
= seq_open(file
, &swaps_op
);
2050 seq
= file
->private_data
;
2051 seq
->poll_event
= atomic_read(&proc_poll_event
);
2055 static const struct file_operations proc_swaps_operations
= {
2058 .llseek
= seq_lseek
,
2059 .release
= seq_release
,
2063 static int __init
procswaps_init(void)
2065 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
2068 __initcall(procswaps_init
);
2069 #endif /* CONFIG_PROC_FS */
2071 #ifdef MAX_SWAPFILES_CHECK
2072 static int __init
max_swapfiles_check(void)
2074 MAX_SWAPFILES_CHECK();
2077 late_initcall(max_swapfiles_check
);
2080 static struct swap_info_struct
*alloc_swap_info(void)
2082 struct swap_info_struct
*p
;
2085 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
2087 return ERR_PTR(-ENOMEM
);
2089 spin_lock(&swap_lock
);
2090 for (type
= 0; type
< nr_swapfiles
; type
++) {
2091 if (!(swap_info
[type
]->flags
& SWP_USED
))
2094 if (type
>= MAX_SWAPFILES
) {
2095 spin_unlock(&swap_lock
);
2097 return ERR_PTR(-EPERM
);
2099 if (type
>= nr_swapfiles
) {
2101 swap_info
[type
] = p
;
2103 * Write swap_info[type] before nr_swapfiles, in case a
2104 * racing procfs swap_start() or swap_next() is reading them.
2105 * (We never shrink nr_swapfiles, we never free this entry.)
2111 p
= swap_info
[type
];
2113 * Do not memset this entry: a racing procfs swap_next()
2114 * would be relying on p->type to remain valid.
2117 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
2118 INIT_LIST_HEAD(&p
->list
);
2119 p
->flags
= SWP_USED
;
2120 spin_unlock(&swap_lock
);
2121 spin_lock_init(&p
->lock
);
2126 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2130 if (S_ISBLK(inode
->i_mode
)) {
2131 p
->bdev
= bdgrab(I_BDEV(inode
));
2132 error
= blkdev_get(p
->bdev
,
2133 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
,
2139 p
->old_block_size
= block_size(p
->bdev
);
2140 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2143 p
->flags
|= SWP_BLKDEV
;
2144 } else if (S_ISREG(inode
->i_mode
)) {
2145 p
->bdev
= inode
->i_sb
->s_bdev
;
2146 mutex_lock(&inode
->i_mutex
);
2147 if (IS_SWAPFILE(inode
))
2155 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2156 union swap_header
*swap_header
,
2157 struct inode
*inode
)
2160 unsigned long maxpages
;
2161 unsigned long swapfilepages
;
2162 unsigned long last_page
;
2164 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2165 pr_err("Unable to find swap-space signature\n");
2169 /* swap partition endianess hack... */
2170 if (swab32(swap_header
->info
.version
) == 1) {
2171 swab32s(&swap_header
->info
.version
);
2172 swab32s(&swap_header
->info
.last_page
);
2173 swab32s(&swap_header
->info
.nr_badpages
);
2174 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2175 swab32s(&swap_header
->info
.badpages
[i
]);
2177 /* Check the swap header's sub-version */
2178 if (swap_header
->info
.version
!= 1) {
2179 pr_warn("Unable to handle swap header version %d\n",
2180 swap_header
->info
.version
);
2185 p
->cluster_next
= 1;
2189 * Find out how many pages are allowed for a single swap
2190 * device. There are two limiting factors: 1) the number
2191 * of bits for the swap offset in the swp_entry_t type, and
2192 * 2) the number of bits in the swap pte as defined by the
2193 * different architectures. In order to find the
2194 * largest possible bit mask, a swap entry with swap type 0
2195 * and swap offset ~0UL is created, encoded to a swap pte,
2196 * decoded to a swp_entry_t again, and finally the swap
2197 * offset is extracted. This will mask all the bits from
2198 * the initial ~0UL mask that can't be encoded in either
2199 * the swp_entry_t or the architecture definition of a
2202 maxpages
= swp_offset(pte_to_swp_entry(
2203 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2204 last_page
= swap_header
->info
.last_page
;
2205 if (last_page
> maxpages
) {
2206 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2207 maxpages
<< (PAGE_SHIFT
- 10),
2208 last_page
<< (PAGE_SHIFT
- 10));
2210 if (maxpages
> last_page
) {
2211 maxpages
= last_page
+ 1;
2212 /* p->max is an unsigned int: don't overflow it */
2213 if ((unsigned int)maxpages
== 0)
2214 maxpages
= UINT_MAX
;
2216 p
->highest_bit
= maxpages
- 1;
2220 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
2221 if (swapfilepages
&& maxpages
> swapfilepages
) {
2222 pr_warn("Swap area shorter than signature indicates\n");
2225 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
2227 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2233 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
2234 union swap_header
*swap_header
,
2235 unsigned char *swap_map
,
2236 struct swap_cluster_info
*cluster_info
,
2237 unsigned long maxpages
,
2241 unsigned int nr_good_pages
;
2243 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2244 unsigned long idx
= p
->cluster_next
/ SWAPFILE_CLUSTER
;
2246 nr_good_pages
= maxpages
- 1; /* omit header page */
2248 cluster_set_null(&p
->free_cluster_head
);
2249 cluster_set_null(&p
->free_cluster_tail
);
2250 cluster_set_null(&p
->discard_cluster_head
);
2251 cluster_set_null(&p
->discard_cluster_tail
);
2253 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
2254 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
2255 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
2257 if (page_nr
< maxpages
) {
2258 swap_map
[page_nr
] = SWAP_MAP_BAD
;
2261 * Haven't marked the cluster free yet, no list
2262 * operation involved
2264 inc_cluster_info_page(p
, cluster_info
, page_nr
);
2268 /* Haven't marked the cluster free yet, no list operation involved */
2269 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
2270 inc_cluster_info_page(p
, cluster_info
, i
);
2272 if (nr_good_pages
) {
2273 swap_map
[0] = SWAP_MAP_BAD
;
2275 * Not mark the cluster free yet, no list
2276 * operation involved
2278 inc_cluster_info_page(p
, cluster_info
, 0);
2280 p
->pages
= nr_good_pages
;
2281 nr_extents
= setup_swap_extents(p
, span
);
2284 nr_good_pages
= p
->pages
;
2286 if (!nr_good_pages
) {
2287 pr_warn("Empty swap-file\n");
2294 for (i
= 0; i
< nr_clusters
; i
++) {
2295 if (!cluster_count(&cluster_info
[idx
])) {
2296 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
2297 if (cluster_is_null(&p
->free_cluster_head
)) {
2298 cluster_set_next_flag(&p
->free_cluster_head
,
2300 cluster_set_next_flag(&p
->free_cluster_tail
,
2305 tail
= cluster_next(&p
->free_cluster_tail
);
2306 cluster_set_next(&cluster_info
[tail
], idx
);
2307 cluster_set_next_flag(&p
->free_cluster_tail
,
2312 if (idx
== nr_clusters
)
2319 * Helper to sys_swapon determining if a given swap
2320 * backing device queue supports DISCARD operations.
2322 static bool swap_discardable(struct swap_info_struct
*si
)
2324 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
2326 if (!q
|| !blk_queue_discard(q
))
2332 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
2334 struct swap_info_struct
*p
;
2335 struct filename
*name
;
2336 struct file
*swap_file
= NULL
;
2337 struct address_space
*mapping
;
2341 union swap_header
*swap_header
;
2344 unsigned long maxpages
;
2345 unsigned char *swap_map
= NULL
;
2346 struct swap_cluster_info
*cluster_info
= NULL
;
2347 unsigned long *frontswap_map
= NULL
;
2348 struct page
*page
= NULL
;
2349 struct inode
*inode
= NULL
;
2351 if (swap_flags
& ~SWAP_FLAGS_VALID
)
2354 if (!capable(CAP_SYS_ADMIN
))
2357 p
= alloc_swap_info();
2361 INIT_WORK(&p
->discard_work
, swap_discard_work
);
2363 name
= getname(specialfile
);
2365 error
= PTR_ERR(name
);
2369 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
2370 if (IS_ERR(swap_file
)) {
2371 error
= PTR_ERR(swap_file
);
2376 p
->swap_file
= swap_file
;
2377 mapping
= swap_file
->f_mapping
;
2379 for (i
= 0; i
< nr_swapfiles
; i
++) {
2380 struct swap_info_struct
*q
= swap_info
[i
];
2382 if (q
== p
|| !q
->swap_file
)
2384 if (mapping
== q
->swap_file
->f_mapping
) {
2390 inode
= mapping
->host
;
2391 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
2392 error
= claim_swapfile(p
, inode
);
2393 if (unlikely(error
))
2397 * Read the swap header.
2399 if (!mapping
->a_ops
->readpage
) {
2403 page
= read_mapping_page(mapping
, 0, swap_file
);
2405 error
= PTR_ERR(page
);
2408 swap_header
= kmap(page
);
2410 maxpages
= read_swap_header(p
, swap_header
, inode
);
2411 if (unlikely(!maxpages
)) {
2416 /* OK, set up the swap map and apply the bad block list */
2417 swap_map
= vzalloc(maxpages
);
2422 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
2423 p
->flags
|= SWP_SOLIDSTATE
;
2425 * select a random position to start with to help wear leveling
2428 p
->cluster_next
= 1 + (prandom_u32() % p
->highest_bit
);
2430 cluster_info
= vzalloc(DIV_ROUND_UP(maxpages
,
2431 SWAPFILE_CLUSTER
) * sizeof(*cluster_info
));
2432 if (!cluster_info
) {
2436 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
2437 if (!p
->percpu_cluster
) {
2441 for_each_possible_cpu(i
) {
2442 struct percpu_cluster
*cluster
;
2443 cluster
= per_cpu_ptr(p
->percpu_cluster
, i
);
2444 cluster_set_null(&cluster
->index
);
2448 error
= swap_cgroup_swapon(p
->type
, maxpages
);
2452 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
2453 cluster_info
, maxpages
, &span
);
2454 if (unlikely(nr_extents
< 0)) {
2458 /* frontswap enabled? set up bit-per-page map for frontswap */
2459 if (frontswap_enabled
)
2460 frontswap_map
= vzalloc(BITS_TO_LONGS(maxpages
) * sizeof(long));
2462 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
2464 * When discard is enabled for swap with no particular
2465 * policy flagged, we set all swap discard flags here in
2466 * order to sustain backward compatibility with older
2467 * swapon(8) releases.
2469 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
2473 * By flagging sys_swapon, a sysadmin can tell us to
2474 * either do single-time area discards only, or to just
2475 * perform discards for released swap page-clusters.
2476 * Now it's time to adjust the p->flags accordingly.
2478 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
2479 p
->flags
&= ~SWP_PAGE_DISCARD
;
2480 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
2481 p
->flags
&= ~SWP_AREA_DISCARD
;
2483 /* issue a swapon-time discard if it's still required */
2484 if (p
->flags
& SWP_AREA_DISCARD
) {
2485 int err
= discard_swap(p
);
2487 pr_err("swapon: discard_swap(%p): %d\n",
2492 mutex_lock(&swapon_mutex
);
2494 if (swap_flags
& SWAP_FLAG_PREFER
)
2496 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
2497 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
2499 pr_info("Adding %uk swap on %s. "
2500 "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2501 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
2502 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
2503 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
2504 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
2505 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
2506 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
2507 (frontswap_map
) ? "FS" : "");
2509 mutex_unlock(&swapon_mutex
);
2510 atomic_inc(&proc_poll_event
);
2511 wake_up_interruptible(&proc_poll_wait
);
2513 if (S_ISREG(inode
->i_mode
))
2514 inode
->i_flags
|= S_SWAPFILE
;
2518 free_percpu(p
->percpu_cluster
);
2519 p
->percpu_cluster
= NULL
;
2520 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
2521 set_blocksize(p
->bdev
, p
->old_block_size
);
2522 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2524 destroy_swap_extents(p
);
2525 swap_cgroup_swapoff(p
->type
);
2526 spin_lock(&swap_lock
);
2527 p
->swap_file
= NULL
;
2529 spin_unlock(&swap_lock
);
2531 vfree(cluster_info
);
2533 if (inode
&& S_ISREG(inode
->i_mode
)) {
2534 mutex_unlock(&inode
->i_mutex
);
2537 filp_close(swap_file
, NULL
);
2540 if (page
&& !IS_ERR(page
)) {
2542 page_cache_release(page
);
2546 if (inode
&& S_ISREG(inode
->i_mode
))
2547 mutex_unlock(&inode
->i_mutex
);
2551 void si_swapinfo(struct sysinfo
*val
)
2554 unsigned long nr_to_be_unused
= 0;
2556 spin_lock(&swap_lock
);
2557 for (type
= 0; type
< nr_swapfiles
; type
++) {
2558 struct swap_info_struct
*si
= swap_info
[type
];
2560 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
2561 nr_to_be_unused
+= si
->inuse_pages
;
2563 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
2564 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
2565 spin_unlock(&swap_lock
);
2569 * Verify that a swap entry is valid and increment its swap map count.
2571 * Returns error code in following case.
2573 * - swp_entry is invalid -> EINVAL
2574 * - swp_entry is migration entry -> EINVAL
2575 * - swap-cache reference is requested but there is already one. -> EEXIST
2576 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2577 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2579 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
2581 struct swap_info_struct
*p
;
2582 unsigned long offset
, type
;
2583 unsigned char count
;
2584 unsigned char has_cache
;
2587 if (non_swap_entry(entry
))
2590 type
= swp_type(entry
);
2591 if (type
>= nr_swapfiles
)
2593 p
= swap_info
[type
];
2594 offset
= swp_offset(entry
);
2596 spin_lock(&p
->lock
);
2597 if (unlikely(offset
>= p
->max
))
2600 count
= p
->swap_map
[offset
];
2603 * swapin_readahead() doesn't check if a swap entry is valid, so the
2604 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2606 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
2611 has_cache
= count
& SWAP_HAS_CACHE
;
2612 count
&= ~SWAP_HAS_CACHE
;
2615 if (usage
== SWAP_HAS_CACHE
) {
2617 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2618 if (!has_cache
&& count
)
2619 has_cache
= SWAP_HAS_CACHE
;
2620 else if (has_cache
) /* someone else added cache */
2622 else /* no users remaining */
2625 } else if (count
|| has_cache
) {
2627 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
2629 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
2631 else if (swap_count_continued(p
, offset
, count
))
2632 count
= COUNT_CONTINUED
;
2636 err
= -ENOENT
; /* unused swap entry */
2638 p
->swap_map
[offset
] = count
| has_cache
;
2641 spin_unlock(&p
->lock
);
2646 pr_err("swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
2651 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2652 * (in which case its reference count is never incremented).
2654 void swap_shmem_alloc(swp_entry_t entry
)
2656 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
2660 * Increase reference count of swap entry by 1.
2661 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2662 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2663 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2664 * might occur if a page table entry has got corrupted.
2666 int swap_duplicate(swp_entry_t entry
)
2670 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
2671 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
2676 * @entry: swap entry for which we allocate swap cache.
2678 * Called when allocating swap cache for existing swap entry,
2679 * This can return error codes. Returns 0 at success.
2680 * -EBUSY means there is a swap cache.
2681 * Note: return code is different from swap_duplicate().
2683 int swapcache_prepare(swp_entry_t entry
)
2685 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
2688 struct swap_info_struct
*page_swap_info(struct page
*page
)
2690 swp_entry_t swap
= { .val
= page_private(page
) };
2691 BUG_ON(!PageSwapCache(page
));
2692 return swap_info
[swp_type(swap
)];
2696 * out-of-line __page_file_ methods to avoid include hell.
2698 struct address_space
*__page_file_mapping(struct page
*page
)
2700 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
2701 return page_swap_info(page
)->swap_file
->f_mapping
;
2703 EXPORT_SYMBOL_GPL(__page_file_mapping
);
2705 pgoff_t
__page_file_index(struct page
*page
)
2707 swp_entry_t swap
= { .val
= page_private(page
) };
2708 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
2709 return swp_offset(swap
);
2711 EXPORT_SYMBOL_GPL(__page_file_index
);
2714 * add_swap_count_continuation - called when a swap count is duplicated
2715 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2716 * page of the original vmalloc'ed swap_map, to hold the continuation count
2717 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2718 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2720 * These continuation pages are seldom referenced: the common paths all work
2721 * on the original swap_map, only referring to a continuation page when the
2722 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2724 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2725 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2726 * can be called after dropping locks.
2728 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
2730 struct swap_info_struct
*si
;
2733 struct page
*list_page
;
2735 unsigned char count
;
2738 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2739 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2741 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
2743 si
= swap_info_get(entry
);
2746 * An acceptable race has occurred since the failing
2747 * __swap_duplicate(): the swap entry has been freed,
2748 * perhaps even the whole swap_map cleared for swapoff.
2753 offset
= swp_offset(entry
);
2754 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
2756 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
2758 * The higher the swap count, the more likely it is that tasks
2759 * will race to add swap count continuation: we need to avoid
2760 * over-provisioning.
2766 spin_unlock(&si
->lock
);
2771 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2772 * no architecture is using highmem pages for kernel page tables: so it
2773 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
2775 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2776 offset
&= ~PAGE_MASK
;
2779 * Page allocation does not initialize the page's lru field,
2780 * but it does always reset its private field.
2782 if (!page_private(head
)) {
2783 BUG_ON(count
& COUNT_CONTINUED
);
2784 INIT_LIST_HEAD(&head
->lru
);
2785 set_page_private(head
, SWP_CONTINUED
);
2786 si
->flags
|= SWP_CONTINUED
;
2789 list_for_each_entry(list_page
, &head
->lru
, lru
) {
2793 * If the previous map said no continuation, but we've found
2794 * a continuation page, free our allocation and use this one.
2796 if (!(count
& COUNT_CONTINUED
))
2799 map
= kmap_atomic(list_page
) + offset
;
2804 * If this continuation count now has some space in it,
2805 * free our allocation and use this one.
2807 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
2811 list_add_tail(&page
->lru
, &head
->lru
);
2812 page
= NULL
; /* now it's attached, don't free it */
2814 spin_unlock(&si
->lock
);
2822 * swap_count_continued - when the original swap_map count is incremented
2823 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2824 * into, carry if so, or else fail until a new continuation page is allocated;
2825 * when the original swap_map count is decremented from 0 with continuation,
2826 * borrow from the continuation and report whether it still holds more.
2827 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2829 static bool swap_count_continued(struct swap_info_struct
*si
,
2830 pgoff_t offset
, unsigned char count
)
2836 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2837 if (page_private(head
) != SWP_CONTINUED
) {
2838 BUG_ON(count
& COUNT_CONTINUED
);
2839 return false; /* need to add count continuation */
2842 offset
&= ~PAGE_MASK
;
2843 page
= list_entry(head
->lru
.next
, struct page
, lru
);
2844 map
= kmap_atomic(page
) + offset
;
2846 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
2847 goto init_map
; /* jump over SWAP_CONT_MAX checks */
2849 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
2851 * Think of how you add 1 to 999
2853 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
2855 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2856 BUG_ON(page
== head
);
2857 map
= kmap_atomic(page
) + offset
;
2859 if (*map
== SWAP_CONT_MAX
) {
2861 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2863 return false; /* add count continuation */
2864 map
= kmap_atomic(page
) + offset
;
2865 init_map
: *map
= 0; /* we didn't zero the page */
2869 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2870 while (page
!= head
) {
2871 map
= kmap_atomic(page
) + offset
;
2872 *map
= COUNT_CONTINUED
;
2874 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2876 return true; /* incremented */
2878 } else { /* decrementing */
2880 * Think of how you subtract 1 from 1000
2882 BUG_ON(count
!= COUNT_CONTINUED
);
2883 while (*map
== COUNT_CONTINUED
) {
2885 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2886 BUG_ON(page
== head
);
2887 map
= kmap_atomic(page
) + offset
;
2894 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2895 while (page
!= head
) {
2896 map
= kmap_atomic(page
) + offset
;
2897 *map
= SWAP_CONT_MAX
| count
;
2898 count
= COUNT_CONTINUED
;
2900 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2902 return count
== COUNT_CONTINUED
;
2907 * free_swap_count_continuations - swapoff free all the continuation pages
2908 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2910 static void free_swap_count_continuations(struct swap_info_struct
*si
)
2914 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
2916 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2917 if (page_private(head
)) {
2918 struct list_head
*this, *next
;
2919 list_for_each_safe(this, next
, &head
->lru
) {
2921 page
= list_entry(this, struct page
, lru
);