4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
9 #include <linux/sched/mm.h>
10 #include <linux/hugetlb.h>
11 #include <linux/mman.h>
12 #include <linux/slab.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>
38 #include <linux/swap_slots.h>
40 #include <asm/pgtable.h>
41 #include <asm/tlbflush.h>
42 #include <linux/swapops.h>
43 #include <linux/swap_cgroup.h>
45 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
47 static void free_swap_count_continuations(struct swap_info_struct
*);
48 static sector_t
map_swap_entry(swp_entry_t
, struct block_device
**);
50 DEFINE_SPINLOCK(swap_lock
);
51 static unsigned int nr_swapfiles
;
52 atomic_long_t nr_swap_pages
;
54 * Some modules use swappable objects and may try to swap them out under
55 * memory pressure (via the shrinker). Before doing so, they may wish to
56 * check to see if any swap space is available.
58 EXPORT_SYMBOL_GPL(nr_swap_pages
);
59 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
60 long total_swap_pages
;
61 static int least_priority
;
63 static const char Bad_file
[] = "Bad swap file entry ";
64 static const char Unused_file
[] = "Unused swap file entry ";
65 static const char Bad_offset
[] = "Bad swap offset entry ";
66 static const char Unused_offset
[] = "Unused swap offset entry ";
69 * all active swap_info_structs
70 * protected with swap_lock, and ordered by priority.
72 PLIST_HEAD(swap_active_head
);
75 * all available (active, not full) swap_info_structs
76 * protected with swap_avail_lock, ordered by priority.
77 * This is used by get_swap_page() instead of swap_active_head
78 * because swap_active_head includes all swap_info_structs,
79 * but get_swap_page() doesn't need to look at full ones.
80 * This uses its own lock instead of swap_lock because when a
81 * swap_info_struct changes between not-full/full, it needs to
82 * add/remove itself to/from this list, but the swap_info_struct->lock
83 * is held and the locking order requires swap_lock to be taken
84 * before any swap_info_struct->lock.
86 static PLIST_HEAD(swap_avail_head
);
87 static DEFINE_SPINLOCK(swap_avail_lock
);
89 struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
91 static DEFINE_MUTEX(swapon_mutex
);
93 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
94 /* Activity counter to indicate that a swapon or swapoff has occurred */
95 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
97 static inline unsigned char swap_count(unsigned char ent
)
99 return ent
& ~SWAP_HAS_CACHE
; /* may include SWAP_HAS_CONT flag */
102 /* returns 1 if swap entry is freed */
104 __try_to_reclaim_swap(struct swap_info_struct
*si
, unsigned long offset
)
106 swp_entry_t entry
= swp_entry(si
->type
, offset
);
110 page
= find_get_page(swap_address_space(entry
), swp_offset(entry
));
114 * This function is called from scan_swap_map() and it's called
115 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
116 * We have to use trylock for avoiding deadlock. This is a special
117 * case and you should use try_to_free_swap() with explicit lock_page()
118 * in usual operations.
120 if (trylock_page(page
)) {
121 ret
= try_to_free_swap(page
);
129 * swapon tell device that all the old swap contents can be discarded,
130 * to allow the swap device to optimize its wear-levelling.
132 static int discard_swap(struct swap_info_struct
*si
)
134 struct swap_extent
*se
;
135 sector_t start_block
;
139 /* Do not discard the swap header page! */
140 se
= &si
->first_swap_extent
;
141 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
142 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
144 err
= blkdev_issue_discard(si
->bdev
, start_block
,
145 nr_blocks
, GFP_KERNEL
, 0);
151 list_for_each_entry(se
, &si
->first_swap_extent
.list
, list
) {
152 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
153 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
155 err
= blkdev_issue_discard(si
->bdev
, start_block
,
156 nr_blocks
, GFP_KERNEL
, 0);
162 return err
; /* That will often be -EOPNOTSUPP */
166 * swap allocation tell device that a cluster of swap can now be discarded,
167 * to allow the swap device to optimize its wear-levelling.
169 static void discard_swap_cluster(struct swap_info_struct
*si
,
170 pgoff_t start_page
, pgoff_t nr_pages
)
172 struct swap_extent
*se
= si
->curr_swap_extent
;
173 int found_extent
= 0;
176 if (se
->start_page
<= start_page
&&
177 start_page
< se
->start_page
+ se
->nr_pages
) {
178 pgoff_t offset
= start_page
- se
->start_page
;
179 sector_t start_block
= se
->start_block
+ offset
;
180 sector_t nr_blocks
= se
->nr_pages
- offset
;
182 if (nr_blocks
> nr_pages
)
183 nr_blocks
= nr_pages
;
184 start_page
+= nr_blocks
;
185 nr_pages
-= nr_blocks
;
188 si
->curr_swap_extent
= se
;
190 start_block
<<= PAGE_SHIFT
- 9;
191 nr_blocks
<<= PAGE_SHIFT
- 9;
192 if (blkdev_issue_discard(si
->bdev
, start_block
,
193 nr_blocks
, GFP_NOIO
, 0))
197 se
= list_next_entry(se
, list
);
201 #define SWAPFILE_CLUSTER 256
202 #define LATENCY_LIMIT 256
204 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
210 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
215 static inline void cluster_set_count(struct swap_cluster_info
*info
,
221 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
222 unsigned int c
, unsigned int f
)
228 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
233 static inline void cluster_set_next(struct swap_cluster_info
*info
,
239 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
240 unsigned int n
, unsigned int f
)
246 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
248 return info
->flags
& CLUSTER_FLAG_FREE
;
251 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
253 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
256 static inline void cluster_set_null(struct swap_cluster_info
*info
)
258 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
262 static inline struct swap_cluster_info
*lock_cluster(struct swap_info_struct
*si
,
263 unsigned long offset
)
265 struct swap_cluster_info
*ci
;
267 ci
= si
->cluster_info
;
269 ci
+= offset
/ SWAPFILE_CLUSTER
;
270 spin_lock(&ci
->lock
);
275 static inline void unlock_cluster(struct swap_cluster_info
*ci
)
278 spin_unlock(&ci
->lock
);
281 static inline struct swap_cluster_info
*lock_cluster_or_swap_info(
282 struct swap_info_struct
*si
,
283 unsigned long offset
)
285 struct swap_cluster_info
*ci
;
287 ci
= lock_cluster(si
, offset
);
289 spin_lock(&si
->lock
);
294 static inline void unlock_cluster_or_swap_info(struct swap_info_struct
*si
,
295 struct swap_cluster_info
*ci
)
300 spin_unlock(&si
->lock
);
303 static inline bool cluster_list_empty(struct swap_cluster_list
*list
)
305 return cluster_is_null(&list
->head
);
308 static inline unsigned int cluster_list_first(struct swap_cluster_list
*list
)
310 return cluster_next(&list
->head
);
313 static void cluster_list_init(struct swap_cluster_list
*list
)
315 cluster_set_null(&list
->head
);
316 cluster_set_null(&list
->tail
);
319 static void cluster_list_add_tail(struct swap_cluster_list
*list
,
320 struct swap_cluster_info
*ci
,
323 if (cluster_list_empty(list
)) {
324 cluster_set_next_flag(&list
->head
, idx
, 0);
325 cluster_set_next_flag(&list
->tail
, idx
, 0);
327 struct swap_cluster_info
*ci_tail
;
328 unsigned int tail
= cluster_next(&list
->tail
);
331 * Nested cluster lock, but both cluster locks are
332 * only acquired when we held swap_info_struct->lock
335 spin_lock_nested(&ci_tail
->lock
, SINGLE_DEPTH_NESTING
);
336 cluster_set_next(ci_tail
, idx
);
337 unlock_cluster(ci_tail
);
338 cluster_set_next_flag(&list
->tail
, idx
, 0);
342 static unsigned int cluster_list_del_first(struct swap_cluster_list
*list
,
343 struct swap_cluster_info
*ci
)
347 idx
= cluster_next(&list
->head
);
348 if (cluster_next(&list
->tail
) == idx
) {
349 cluster_set_null(&list
->head
);
350 cluster_set_null(&list
->tail
);
352 cluster_set_next_flag(&list
->head
,
353 cluster_next(&ci
[idx
]), 0);
358 /* Add a cluster to discard list and schedule it to do discard */
359 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
363 * If scan_swap_map() can't find a free cluster, it will check
364 * si->swap_map directly. To make sure the discarding cluster isn't
365 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
366 * will be cleared after discard
368 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
369 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
371 cluster_list_add_tail(&si
->discard_clusters
, si
->cluster_info
, idx
);
373 schedule_work(&si
->discard_work
);
377 * Doing discard actually. After a cluster discard is finished, the cluster
378 * will be added to free cluster list. caller should hold si->lock.
380 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
382 struct swap_cluster_info
*info
, *ci
;
385 info
= si
->cluster_info
;
387 while (!cluster_list_empty(&si
->discard_clusters
)) {
388 idx
= cluster_list_del_first(&si
->discard_clusters
, info
);
389 spin_unlock(&si
->lock
);
391 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
394 spin_lock(&si
->lock
);
395 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
396 cluster_set_flag(ci
, CLUSTER_FLAG_FREE
);
398 cluster_list_add_tail(&si
->free_clusters
, info
, idx
);
399 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
400 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
401 0, SWAPFILE_CLUSTER
);
406 static void swap_discard_work(struct work_struct
*work
)
408 struct swap_info_struct
*si
;
410 si
= container_of(work
, struct swap_info_struct
, discard_work
);
412 spin_lock(&si
->lock
);
413 swap_do_scheduled_discard(si
);
414 spin_unlock(&si
->lock
);
418 * The cluster corresponding to page_nr will be used. The cluster will be
419 * removed from free cluster list and its usage counter will be increased.
421 static void inc_cluster_info_page(struct swap_info_struct
*p
,
422 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
424 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
428 if (cluster_is_free(&cluster_info
[idx
])) {
429 VM_BUG_ON(cluster_list_first(&p
->free_clusters
) != idx
);
430 cluster_list_del_first(&p
->free_clusters
, cluster_info
);
431 cluster_set_count_flag(&cluster_info
[idx
], 0, 0);
434 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
435 cluster_set_count(&cluster_info
[idx
],
436 cluster_count(&cluster_info
[idx
]) + 1);
440 * The cluster corresponding to page_nr decreases one usage. If the usage
441 * counter becomes 0, which means no page in the cluster is in using, we can
442 * optionally discard the cluster and add it to free cluster list.
444 static void dec_cluster_info_page(struct swap_info_struct
*p
,
445 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
447 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
452 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
453 cluster_set_count(&cluster_info
[idx
],
454 cluster_count(&cluster_info
[idx
]) - 1);
456 if (cluster_count(&cluster_info
[idx
]) == 0) {
458 * If the swap is discardable, prepare discard the cluster
459 * instead of free it immediately. The cluster will be freed
462 if ((p
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
463 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
464 swap_cluster_schedule_discard(p
, idx
);
468 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
469 cluster_list_add_tail(&p
->free_clusters
, cluster_info
, idx
);
474 * It's possible scan_swap_map() uses a free cluster in the middle of free
475 * cluster list. Avoiding such abuse to avoid list corruption.
478 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
479 unsigned long offset
)
481 struct percpu_cluster
*percpu_cluster
;
484 offset
/= SWAPFILE_CLUSTER
;
485 conflict
= !cluster_list_empty(&si
->free_clusters
) &&
486 offset
!= cluster_list_first(&si
->free_clusters
) &&
487 cluster_is_free(&si
->cluster_info
[offset
]);
492 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
493 cluster_set_null(&percpu_cluster
->index
);
498 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
499 * might involve allocating a new cluster for current CPU too.
501 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
502 unsigned long *offset
, unsigned long *scan_base
)
504 struct percpu_cluster
*cluster
;
505 struct swap_cluster_info
*ci
;
507 unsigned long tmp
, max
;
510 cluster
= this_cpu_ptr(si
->percpu_cluster
);
511 if (cluster_is_null(&cluster
->index
)) {
512 if (!cluster_list_empty(&si
->free_clusters
)) {
513 cluster
->index
= si
->free_clusters
.head
;
514 cluster
->next
= cluster_next(&cluster
->index
) *
516 } else if (!cluster_list_empty(&si
->discard_clusters
)) {
518 * we don't have free cluster but have some clusters in
519 * discarding, do discard now and reclaim them
521 swap_do_scheduled_discard(si
);
522 *scan_base
= *offset
= si
->cluster_next
;
531 * Other CPUs can use our cluster if they can't find a free cluster,
532 * check if there is still free entry in the cluster
535 max
= min_t(unsigned long, si
->max
,
536 (cluster_next(&cluster
->index
) + 1) * SWAPFILE_CLUSTER
);
538 cluster_set_null(&cluster
->index
);
541 ci
= lock_cluster(si
, tmp
);
543 if (!si
->swap_map
[tmp
]) {
551 cluster_set_null(&cluster
->index
);
554 cluster
->next
= tmp
+ 1;
560 static int scan_swap_map_slots(struct swap_info_struct
*si
,
561 unsigned char usage
, int nr
,
564 struct swap_cluster_info
*ci
;
565 unsigned long offset
;
566 unsigned long scan_base
;
567 unsigned long last_in_cluster
= 0;
568 int latency_ration
= LATENCY_LIMIT
;
575 * We try to cluster swap pages by allocating them sequentially
576 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
577 * way, however, we resort to first-free allocation, starting
578 * a new cluster. This prevents us from scattering swap pages
579 * all over the entire swap partition, so that we reduce
580 * overall disk seek times between swap pages. -- sct
581 * But we do now try to find an empty cluster. -Andrea
582 * And we let swap pages go all over an SSD partition. Hugh
585 si
->flags
+= SWP_SCANNING
;
586 scan_base
= offset
= si
->cluster_next
;
589 if (si
->cluster_info
) {
590 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
596 if (unlikely(!si
->cluster_nr
--)) {
597 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
598 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
602 spin_unlock(&si
->lock
);
605 * If seek is expensive, start searching for new cluster from
606 * start of partition, to minimize the span of allocated swap.
607 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
608 * case, just handled by scan_swap_map_try_ssd_cluster() above.
610 scan_base
= offset
= si
->lowest_bit
;
611 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
613 /* Locate the first empty (unaligned) cluster */
614 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
615 if (si
->swap_map
[offset
])
616 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
617 else if (offset
== last_in_cluster
) {
618 spin_lock(&si
->lock
);
619 offset
-= SWAPFILE_CLUSTER
- 1;
620 si
->cluster_next
= offset
;
621 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
624 if (unlikely(--latency_ration
< 0)) {
626 latency_ration
= LATENCY_LIMIT
;
631 spin_lock(&si
->lock
);
632 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
636 if (si
->cluster_info
) {
637 while (scan_swap_map_ssd_cluster_conflict(si
, offset
)) {
638 /* take a break if we already got some slots */
641 if (!scan_swap_map_try_ssd_cluster(si
, &offset
,
646 if (!(si
->flags
& SWP_WRITEOK
))
648 if (!si
->highest_bit
)
650 if (offset
> si
->highest_bit
)
651 scan_base
= offset
= si
->lowest_bit
;
653 ci
= lock_cluster(si
, offset
);
654 /* reuse swap entry of cache-only swap if not busy. */
655 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
658 spin_unlock(&si
->lock
);
659 swap_was_freed
= __try_to_reclaim_swap(si
, offset
);
660 spin_lock(&si
->lock
);
661 /* entry was freed successfully, try to use this again */
664 goto scan
; /* check next one */
667 if (si
->swap_map
[offset
]) {
675 if (offset
== si
->lowest_bit
)
677 if (offset
== si
->highest_bit
)
680 if (si
->inuse_pages
== si
->pages
) {
681 si
->lowest_bit
= si
->max
;
683 spin_lock(&swap_avail_lock
);
684 plist_del(&si
->avail_list
, &swap_avail_head
);
685 spin_unlock(&swap_avail_lock
);
687 si
->swap_map
[offset
] = usage
;
688 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
690 si
->cluster_next
= offset
+ 1;
691 slots
[n_ret
++] = swp_entry(si
->type
, offset
);
693 /* got enough slots or reach max slots? */
694 if ((n_ret
== nr
) || (offset
>= si
->highest_bit
))
697 /* search for next available slot */
699 /* time to take a break? */
700 if (unlikely(--latency_ration
< 0)) {
703 spin_unlock(&si
->lock
);
705 spin_lock(&si
->lock
);
706 latency_ration
= LATENCY_LIMIT
;
709 /* try to get more slots in cluster */
710 if (si
->cluster_info
) {
711 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
719 /* non-ssd case, still more slots in cluster? */
720 if (si
->cluster_nr
&& !si
->swap_map
[offset
]) {
726 si
->flags
-= SWP_SCANNING
;
730 spin_unlock(&si
->lock
);
731 while (++offset
<= si
->highest_bit
) {
732 if (!si
->swap_map
[offset
]) {
733 spin_lock(&si
->lock
);
736 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
737 spin_lock(&si
->lock
);
740 if (unlikely(--latency_ration
< 0)) {
742 latency_ration
= LATENCY_LIMIT
;
745 offset
= si
->lowest_bit
;
746 while (offset
< scan_base
) {
747 if (!si
->swap_map
[offset
]) {
748 spin_lock(&si
->lock
);
751 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
752 spin_lock(&si
->lock
);
755 if (unlikely(--latency_ration
< 0)) {
757 latency_ration
= LATENCY_LIMIT
;
761 spin_lock(&si
->lock
);
764 si
->flags
-= SWP_SCANNING
;
768 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
774 n_ret
= scan_swap_map_slots(si
, usage
, 1, &entry
);
777 return swp_offset(entry
);
783 int get_swap_pages(int n_goal
, swp_entry_t swp_entries
[])
785 struct swap_info_struct
*si
, *next
;
789 avail_pgs
= atomic_long_read(&nr_swap_pages
);
793 if (n_goal
> SWAP_BATCH
)
796 if (n_goal
> avail_pgs
)
799 atomic_long_sub(n_goal
, &nr_swap_pages
);
801 spin_lock(&swap_avail_lock
);
804 plist_for_each_entry_safe(si
, next
, &swap_avail_head
, avail_list
) {
805 /* requeue si to after same-priority siblings */
806 plist_requeue(&si
->avail_list
, &swap_avail_head
);
807 spin_unlock(&swap_avail_lock
);
808 spin_lock(&si
->lock
);
809 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
810 spin_lock(&swap_avail_lock
);
811 if (plist_node_empty(&si
->avail_list
)) {
812 spin_unlock(&si
->lock
);
815 WARN(!si
->highest_bit
,
816 "swap_info %d in list but !highest_bit\n",
818 WARN(!(si
->flags
& SWP_WRITEOK
),
819 "swap_info %d in list but !SWP_WRITEOK\n",
821 plist_del(&si
->avail_list
, &swap_avail_head
);
822 spin_unlock(&si
->lock
);
825 n_ret
= scan_swap_map_slots(si
, SWAP_HAS_CACHE
,
826 n_goal
, swp_entries
);
827 spin_unlock(&si
->lock
);
830 pr_debug("scan_swap_map of si %d failed to find offset\n",
833 spin_lock(&swap_avail_lock
);
836 * if we got here, it's likely that si was almost full before,
837 * and since scan_swap_map() can drop the si->lock, multiple
838 * callers probably all tried to get a page from the same si
839 * and it filled up before we could get one; or, the si filled
840 * up between us dropping swap_avail_lock and taking si->lock.
841 * Since we dropped the swap_avail_lock, the swap_avail_head
842 * list may have been modified; so if next is still in the
843 * swap_avail_head list then try it, otherwise start over
844 * if we have not gotten any slots.
846 if (plist_node_empty(&next
->avail_list
))
850 spin_unlock(&swap_avail_lock
);
854 atomic_long_add((long) (n_goal
-n_ret
), &nr_swap_pages
);
859 /* The only caller of this function is now suspend routine */
860 swp_entry_t
get_swap_page_of_type(int type
)
862 struct swap_info_struct
*si
;
865 si
= swap_info
[type
];
866 spin_lock(&si
->lock
);
867 if (si
&& (si
->flags
& SWP_WRITEOK
)) {
868 atomic_long_dec(&nr_swap_pages
);
869 /* This is called for allocating swap entry, not cache */
870 offset
= scan_swap_map(si
, 1);
872 spin_unlock(&si
->lock
);
873 return swp_entry(type
, offset
);
875 atomic_long_inc(&nr_swap_pages
);
877 spin_unlock(&si
->lock
);
878 return (swp_entry_t
) {0};
881 static struct swap_info_struct
*__swap_info_get(swp_entry_t entry
)
883 struct swap_info_struct
*p
;
884 unsigned long offset
, type
;
888 type
= swp_type(entry
);
889 if (type
>= nr_swapfiles
)
892 if (!(p
->flags
& SWP_USED
))
894 offset
= swp_offset(entry
);
895 if (offset
>= p
->max
)
900 pr_err("swap_info_get: %s%08lx\n", Bad_offset
, entry
.val
);
903 pr_err("swap_info_get: %s%08lx\n", Unused_file
, entry
.val
);
906 pr_err("swap_info_get: %s%08lx\n", Bad_file
, entry
.val
);
911 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
913 struct swap_info_struct
*p
;
915 p
= __swap_info_get(entry
);
918 if (!p
->swap_map
[swp_offset(entry
)])
923 pr_err("swap_info_get: %s%08lx\n", Unused_offset
, entry
.val
);
929 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
931 struct swap_info_struct
*p
;
933 p
= _swap_info_get(entry
);
939 static struct swap_info_struct
*swap_info_get_cont(swp_entry_t entry
,
940 struct swap_info_struct
*q
)
942 struct swap_info_struct
*p
;
944 p
= _swap_info_get(entry
);
948 spin_unlock(&q
->lock
);
955 static unsigned char __swap_entry_free(struct swap_info_struct
*p
,
956 swp_entry_t entry
, unsigned char usage
)
958 struct swap_cluster_info
*ci
;
959 unsigned long offset
= swp_offset(entry
);
961 unsigned char has_cache
;
963 ci
= lock_cluster_or_swap_info(p
, offset
);
965 count
= p
->swap_map
[offset
];
967 has_cache
= count
& SWAP_HAS_CACHE
;
968 count
&= ~SWAP_HAS_CACHE
;
970 if (usage
== SWAP_HAS_CACHE
) {
971 VM_BUG_ON(!has_cache
);
973 } else if (count
== SWAP_MAP_SHMEM
) {
975 * Or we could insist on shmem.c using a special
976 * swap_shmem_free() and free_shmem_swap_and_cache()...
979 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
980 if (count
== COUNT_CONTINUED
) {
981 if (swap_count_continued(p
, offset
, count
))
982 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
984 count
= SWAP_MAP_MAX
;
989 usage
= count
| has_cache
;
990 p
->swap_map
[offset
] = usage
? : SWAP_HAS_CACHE
;
992 unlock_cluster_or_swap_info(p
, ci
);
997 static void swap_entry_free(struct swap_info_struct
*p
, swp_entry_t entry
)
999 struct swap_cluster_info
*ci
;
1000 unsigned long offset
= swp_offset(entry
);
1001 unsigned char count
;
1003 ci
= lock_cluster(p
, offset
);
1004 count
= p
->swap_map
[offset
];
1005 VM_BUG_ON(count
!= SWAP_HAS_CACHE
);
1006 p
->swap_map
[offset
] = 0;
1007 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
1010 mem_cgroup_uncharge_swap(entry
);
1011 if (offset
< p
->lowest_bit
)
1012 p
->lowest_bit
= offset
;
1013 if (offset
> p
->highest_bit
) {
1014 bool was_full
= !p
->highest_bit
;
1016 p
->highest_bit
= offset
;
1017 if (was_full
&& (p
->flags
& SWP_WRITEOK
)) {
1018 spin_lock(&swap_avail_lock
);
1019 WARN_ON(!plist_node_empty(&p
->avail_list
));
1020 if (plist_node_empty(&p
->avail_list
))
1021 plist_add(&p
->avail_list
,
1023 spin_unlock(&swap_avail_lock
);
1026 atomic_long_inc(&nr_swap_pages
);
1028 frontswap_invalidate_page(p
->type
, offset
);
1029 if (p
->flags
& SWP_BLKDEV
) {
1030 struct gendisk
*disk
= p
->bdev
->bd_disk
;
1032 if (disk
->fops
->swap_slot_free_notify
)
1033 disk
->fops
->swap_slot_free_notify(p
->bdev
,
1039 * Caller has made sure that the swap device corresponding to entry
1040 * is still around or has not been recycled.
1042 void swap_free(swp_entry_t entry
)
1044 struct swap_info_struct
*p
;
1046 p
= _swap_info_get(entry
);
1048 if (!__swap_entry_free(p
, entry
, 1))
1049 free_swap_slot(entry
);
1054 * Called after dropping swapcache to decrease refcnt to swap entries.
1056 void swapcache_free(swp_entry_t entry
)
1058 struct swap_info_struct
*p
;
1060 p
= _swap_info_get(entry
);
1062 if (!__swap_entry_free(p
, entry
, SWAP_HAS_CACHE
))
1063 free_swap_slot(entry
);
1067 void swapcache_free_entries(swp_entry_t
*entries
, int n
)
1069 struct swap_info_struct
*p
, *prev
;
1077 for (i
= 0; i
< n
; ++i
) {
1078 p
= swap_info_get_cont(entries
[i
], prev
);
1080 swap_entry_free(p
, entries
[i
]);
1086 spin_unlock(&p
->lock
);
1090 * How many references to page are currently swapped out?
1091 * This does not give an exact answer when swap count is continued,
1092 * but does include the high COUNT_CONTINUED flag to allow for that.
1094 int page_swapcount(struct page
*page
)
1097 struct swap_info_struct
*p
;
1098 struct swap_cluster_info
*ci
;
1100 unsigned long offset
;
1102 entry
.val
= page_private(page
);
1103 p
= _swap_info_get(entry
);
1105 offset
= swp_offset(entry
);
1106 ci
= lock_cluster_or_swap_info(p
, offset
);
1107 count
= swap_count(p
->swap_map
[offset
]);
1108 unlock_cluster_or_swap_info(p
, ci
);
1114 * How many references to @entry are currently swapped out?
1115 * This does not give an exact answer when swap count is continued,
1116 * but does include the high COUNT_CONTINUED flag to allow for that.
1118 int __swp_swapcount(swp_entry_t entry
)
1122 struct swap_info_struct
*si
;
1123 struct swap_cluster_info
*ci
;
1125 si
= __swap_info_get(entry
);
1127 offset
= swp_offset(entry
);
1128 ci
= lock_cluster_or_swap_info(si
, offset
);
1129 count
= swap_count(si
->swap_map
[offset
]);
1130 unlock_cluster_or_swap_info(si
, ci
);
1136 * How many references to @entry are currently swapped out?
1137 * This considers COUNT_CONTINUED so it returns exact answer.
1139 int swp_swapcount(swp_entry_t entry
)
1141 int count
, tmp_count
, n
;
1142 struct swap_info_struct
*p
;
1143 struct swap_cluster_info
*ci
;
1148 p
= _swap_info_get(entry
);
1152 offset
= swp_offset(entry
);
1154 ci
= lock_cluster_or_swap_info(p
, offset
);
1156 count
= swap_count(p
->swap_map
[offset
]);
1157 if (!(count
& COUNT_CONTINUED
))
1160 count
&= ~COUNT_CONTINUED
;
1161 n
= SWAP_MAP_MAX
+ 1;
1163 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1164 offset
&= ~PAGE_MASK
;
1165 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1168 page
= list_next_entry(page
, lru
);
1169 map
= kmap_atomic(page
);
1170 tmp_count
= map
[offset
];
1173 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1174 n
*= (SWAP_CONT_MAX
+ 1);
1175 } while (tmp_count
& COUNT_CONTINUED
);
1177 unlock_cluster_or_swap_info(p
, ci
);
1182 * We can write to an anon page without COW if there are no other references
1183 * to it. And as a side-effect, free up its swap: because the old content
1184 * on disk will never be read, and seeking back there to write new content
1185 * later would only waste time away from clustering.
1187 * NOTE: total_mapcount should not be relied upon by the caller if
1188 * reuse_swap_page() returns false, but it may be always overwritten
1189 * (see the other implementation for CONFIG_SWAP=n).
1191 bool reuse_swap_page(struct page
*page
, int *total_mapcount
)
1195 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1196 if (unlikely(PageKsm(page
)))
1198 count
= page_trans_huge_mapcount(page
, total_mapcount
);
1199 if (count
<= 1 && PageSwapCache(page
)) {
1200 count
+= page_swapcount(page
);
1203 if (!PageWriteback(page
)) {
1204 delete_from_swap_cache(page
);
1208 struct swap_info_struct
*p
;
1210 entry
.val
= page_private(page
);
1211 p
= swap_info_get(entry
);
1212 if (p
->flags
& SWP_STABLE_WRITES
) {
1213 spin_unlock(&p
->lock
);
1216 spin_unlock(&p
->lock
);
1224 * If swap is getting full, or if there are no more mappings of this page,
1225 * then try_to_free_swap is called to free its swap space.
1227 int try_to_free_swap(struct page
*page
)
1229 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1231 if (!PageSwapCache(page
))
1233 if (PageWriteback(page
))
1235 if (page_swapcount(page
))
1239 * Once hibernation has begun to create its image of memory,
1240 * there's a danger that one of the calls to try_to_free_swap()
1241 * - most probably a call from __try_to_reclaim_swap() while
1242 * hibernation is allocating its own swap pages for the image,
1243 * but conceivably even a call from memory reclaim - will free
1244 * the swap from a page which has already been recorded in the
1245 * image as a clean swapcache page, and then reuse its swap for
1246 * another page of the image. On waking from hibernation, the
1247 * original page might be freed under memory pressure, then
1248 * later read back in from swap, now with the wrong data.
1250 * Hibernation suspends storage while it is writing the image
1251 * to disk so check that here.
1253 if (pm_suspended_storage())
1256 delete_from_swap_cache(page
);
1262 * Free the swap entry like above, but also try to
1263 * free the page cache entry if it is the last user.
1265 int free_swap_and_cache(swp_entry_t entry
)
1267 struct swap_info_struct
*p
;
1268 struct page
*page
= NULL
;
1269 unsigned char count
;
1271 if (non_swap_entry(entry
))
1274 p
= _swap_info_get(entry
);
1276 count
= __swap_entry_free(p
, entry
, 1);
1277 if (count
== SWAP_HAS_CACHE
) {
1278 page
= find_get_page(swap_address_space(entry
),
1280 if (page
&& !trylock_page(page
)) {
1285 free_swap_slot(entry
);
1289 * Not mapped elsewhere, or swap space full? Free it!
1290 * Also recheck PageSwapCache now page is locked (above).
1292 if (PageSwapCache(page
) && !PageWriteback(page
) &&
1293 (!page_mapped(page
) || mem_cgroup_swap_full(page
))) {
1294 delete_from_swap_cache(page
);
1303 #ifdef CONFIG_HIBERNATION
1305 * Find the swap type that corresponds to given device (if any).
1307 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1308 * from 0, in which the swap header is expected to be located.
1310 * This is needed for the suspend to disk (aka swsusp).
1312 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
1314 struct block_device
*bdev
= NULL
;
1318 bdev
= bdget(device
);
1320 spin_lock(&swap_lock
);
1321 for (type
= 0; type
< nr_swapfiles
; type
++) {
1322 struct swap_info_struct
*sis
= swap_info
[type
];
1324 if (!(sis
->flags
& SWP_WRITEOK
))
1329 *bdev_p
= bdgrab(sis
->bdev
);
1331 spin_unlock(&swap_lock
);
1334 if (bdev
== sis
->bdev
) {
1335 struct swap_extent
*se
= &sis
->first_swap_extent
;
1337 if (se
->start_block
== offset
) {
1339 *bdev_p
= bdgrab(sis
->bdev
);
1341 spin_unlock(&swap_lock
);
1347 spin_unlock(&swap_lock
);
1355 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1356 * corresponding to given index in swap_info (swap type).
1358 sector_t
swapdev_block(int type
, pgoff_t offset
)
1360 struct block_device
*bdev
;
1362 if ((unsigned int)type
>= nr_swapfiles
)
1364 if (!(swap_info
[type
]->flags
& SWP_WRITEOK
))
1366 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
1370 * Return either the total number of swap pages of given type, or the number
1371 * of free pages of that type (depending on @free)
1373 * This is needed for software suspend
1375 unsigned int count_swap_pages(int type
, int free
)
1379 spin_lock(&swap_lock
);
1380 if ((unsigned int)type
< nr_swapfiles
) {
1381 struct swap_info_struct
*sis
= swap_info
[type
];
1383 spin_lock(&sis
->lock
);
1384 if (sis
->flags
& SWP_WRITEOK
) {
1387 n
-= sis
->inuse_pages
;
1389 spin_unlock(&sis
->lock
);
1391 spin_unlock(&swap_lock
);
1394 #endif /* CONFIG_HIBERNATION */
1396 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1398 return pte_same(pte_swp_clear_soft_dirty(pte
), swp_pte
);
1402 * No need to decide whether this PTE shares the swap entry with others,
1403 * just let do_wp_page work it out if a write is requested later - to
1404 * force COW, vm_page_prot omits write permission from any private vma.
1406 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1407 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1409 struct page
*swapcache
;
1410 struct mem_cgroup
*memcg
;
1416 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1417 if (unlikely(!page
))
1420 if (mem_cgroup_try_charge(page
, vma
->vm_mm
, GFP_KERNEL
,
1426 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1427 if (unlikely(!pte_same_as_swp(*pte
, swp_entry_to_pte(entry
)))) {
1428 mem_cgroup_cancel_charge(page
, memcg
, false);
1433 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1434 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1436 set_pte_at(vma
->vm_mm
, addr
, pte
,
1437 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1438 if (page
== swapcache
) {
1439 page_add_anon_rmap(page
, vma
, addr
, false);
1440 mem_cgroup_commit_charge(page
, memcg
, true, false);
1441 } else { /* ksm created a completely new copy */
1442 page_add_new_anon_rmap(page
, vma
, addr
, false);
1443 mem_cgroup_commit_charge(page
, memcg
, false, false);
1444 lru_cache_add_active_or_unevictable(page
, vma
);
1448 * Move the page to the active list so it is not
1449 * immediately swapped out again after swapon.
1451 activate_page(page
);
1453 pte_unmap_unlock(pte
, ptl
);
1455 if (page
!= swapcache
) {
1462 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1463 unsigned long addr
, unsigned long end
,
1464 swp_entry_t entry
, struct page
*page
)
1466 pte_t swp_pte
= swp_entry_to_pte(entry
);
1471 * We don't actually need pte lock while scanning for swp_pte: since
1472 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1473 * page table while we're scanning; though it could get zapped, and on
1474 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1475 * of unmatched parts which look like swp_pte, so unuse_pte must
1476 * recheck under pte lock. Scanning without pte lock lets it be
1477 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1479 pte
= pte_offset_map(pmd
, addr
);
1482 * swapoff spends a _lot_ of time in this loop!
1483 * Test inline before going to call unuse_pte.
1485 if (unlikely(pte_same_as_swp(*pte
, swp_pte
))) {
1487 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1490 pte
= pte_offset_map(pmd
, addr
);
1492 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1498 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
1499 unsigned long addr
, unsigned long end
,
1500 swp_entry_t entry
, struct page
*page
)
1506 pmd
= pmd_offset(pud
, addr
);
1509 next
= pmd_addr_end(addr
, end
);
1510 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1512 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
);
1515 } while (pmd
++, addr
= next
, addr
!= end
);
1519 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
1520 unsigned long addr
, unsigned long end
,
1521 swp_entry_t entry
, struct page
*page
)
1527 pud
= pud_offset(pgd
, addr
);
1529 next
= pud_addr_end(addr
, end
);
1530 if (pud_none_or_clear_bad(pud
))
1532 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
);
1535 } while (pud
++, addr
= next
, addr
!= end
);
1539 static int unuse_vma(struct vm_area_struct
*vma
,
1540 swp_entry_t entry
, struct page
*page
)
1543 unsigned long addr
, end
, next
;
1546 if (page_anon_vma(page
)) {
1547 addr
= page_address_in_vma(page
, vma
);
1548 if (addr
== -EFAULT
)
1551 end
= addr
+ PAGE_SIZE
;
1553 addr
= vma
->vm_start
;
1557 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1559 next
= pgd_addr_end(addr
, end
);
1560 if (pgd_none_or_clear_bad(pgd
))
1562 ret
= unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
);
1565 } while (pgd
++, addr
= next
, addr
!= end
);
1569 static int unuse_mm(struct mm_struct
*mm
,
1570 swp_entry_t entry
, struct page
*page
)
1572 struct vm_area_struct
*vma
;
1575 if (!down_read_trylock(&mm
->mmap_sem
)) {
1577 * Activate page so shrink_inactive_list is unlikely to unmap
1578 * its ptes while lock is dropped, so swapoff can make progress.
1580 activate_page(page
);
1582 down_read(&mm
->mmap_sem
);
1585 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1586 if (vma
->anon_vma
&& (ret
= unuse_vma(vma
, entry
, page
)))
1590 up_read(&mm
->mmap_sem
);
1591 return (ret
< 0)? ret
: 0;
1595 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1596 * from current position to next entry still in use.
1597 * Recycle to start on reaching the end, returning 0 when empty.
1599 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
1600 unsigned int prev
, bool frontswap
)
1602 unsigned int max
= si
->max
;
1603 unsigned int i
= prev
;
1604 unsigned char count
;
1607 * No need for swap_lock here: we're just looking
1608 * for whether an entry is in use, not modifying it; false
1609 * hits are okay, and sys_swapoff() has already prevented new
1610 * allocations from this area (while holding swap_lock).
1619 * No entries in use at top of swap_map,
1620 * loop back to start and recheck there.
1626 count
= READ_ONCE(si
->swap_map
[i
]);
1627 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
1628 if (!frontswap
|| frontswap_test(si
, i
))
1630 if ((i
% LATENCY_LIMIT
) == 0)
1637 * We completely avoid races by reading each swap page in advance,
1638 * and then search for the process using it. All the necessary
1639 * page table adjustments can then be made atomically.
1641 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1642 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1644 int try_to_unuse(unsigned int type
, bool frontswap
,
1645 unsigned long pages_to_unuse
)
1647 struct swap_info_struct
*si
= swap_info
[type
];
1648 struct mm_struct
*start_mm
;
1649 volatile unsigned char *swap_map
; /* swap_map is accessed without
1650 * locking. Mark it as volatile
1651 * to prevent compiler doing
1654 unsigned char swcount
;
1661 * When searching mms for an entry, a good strategy is to
1662 * start at the first mm we freed the previous entry from
1663 * (though actually we don't notice whether we or coincidence
1664 * freed the entry). Initialize this start_mm with a hold.
1666 * A simpler strategy would be to start at the last mm we
1667 * freed the previous entry from; but that would take less
1668 * advantage of mmlist ordering, which clusters forked mms
1669 * together, child after parent. If we race with dup_mmap(), we
1670 * prefer to resolve parent before child, lest we miss entries
1671 * duplicated after we scanned child: using last mm would invert
1674 start_mm
= &init_mm
;
1678 * Keep on scanning until all entries have gone. Usually,
1679 * one pass through swap_map is enough, but not necessarily:
1680 * there are races when an instance of an entry might be missed.
1682 while ((i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
1683 if (signal_pending(current
)) {
1689 * Get a page for the entry, using the existing swap
1690 * cache page if there is one. Otherwise, get a clean
1691 * page and read the swap into it.
1693 swap_map
= &si
->swap_map
[i
];
1694 entry
= swp_entry(type
, i
);
1695 page
= read_swap_cache_async(entry
,
1696 GFP_HIGHUSER_MOVABLE
, NULL
, 0);
1699 * Either swap_duplicate() failed because entry
1700 * has been freed independently, and will not be
1701 * reused since sys_swapoff() already disabled
1702 * allocation from here, or alloc_page() failed.
1704 swcount
= *swap_map
;
1706 * We don't hold lock here, so the swap entry could be
1707 * SWAP_MAP_BAD (when the cluster is discarding).
1708 * Instead of fail out, We can just skip the swap
1709 * entry because swapoff will wait for discarding
1712 if (!swcount
|| swcount
== SWAP_MAP_BAD
)
1719 * Don't hold on to start_mm if it looks like exiting.
1721 if (atomic_read(&start_mm
->mm_users
) == 1) {
1723 start_mm
= &init_mm
;
1728 * Wait for and lock page. When do_swap_page races with
1729 * try_to_unuse, do_swap_page can handle the fault much
1730 * faster than try_to_unuse can locate the entry. This
1731 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1732 * defer to do_swap_page in such a case - in some tests,
1733 * do_swap_page and try_to_unuse repeatedly compete.
1735 wait_on_page_locked(page
);
1736 wait_on_page_writeback(page
);
1738 wait_on_page_writeback(page
);
1741 * Remove all references to entry.
1743 swcount
= *swap_map
;
1744 if (swap_count(swcount
) == SWAP_MAP_SHMEM
) {
1745 retval
= shmem_unuse(entry
, page
);
1746 /* page has already been unlocked and released */
1751 if (swap_count(swcount
) && start_mm
!= &init_mm
)
1752 retval
= unuse_mm(start_mm
, entry
, page
);
1754 if (swap_count(*swap_map
)) {
1755 int set_start_mm
= (*swap_map
>= swcount
);
1756 struct list_head
*p
= &start_mm
->mmlist
;
1757 struct mm_struct
*new_start_mm
= start_mm
;
1758 struct mm_struct
*prev_mm
= start_mm
;
1759 struct mm_struct
*mm
;
1761 mmget(new_start_mm
);
1763 spin_lock(&mmlist_lock
);
1764 while (swap_count(*swap_map
) && !retval
&&
1765 (p
= p
->next
) != &start_mm
->mmlist
) {
1766 mm
= list_entry(p
, struct mm_struct
, mmlist
);
1767 if (!mmget_not_zero(mm
))
1769 spin_unlock(&mmlist_lock
);
1775 swcount
= *swap_map
;
1776 if (!swap_count(swcount
)) /* any usage ? */
1778 else if (mm
== &init_mm
)
1781 retval
= unuse_mm(mm
, entry
, page
);
1783 if (set_start_mm
&& *swap_map
< swcount
) {
1784 mmput(new_start_mm
);
1789 spin_lock(&mmlist_lock
);
1791 spin_unlock(&mmlist_lock
);
1794 start_mm
= new_start_mm
;
1803 * If a reference remains (rare), we would like to leave
1804 * the page in the swap cache; but try_to_unmap could
1805 * then re-duplicate the entry once we drop page lock,
1806 * so we might loop indefinitely; also, that page could
1807 * not be swapped out to other storage meanwhile. So:
1808 * delete from cache even if there's another reference,
1809 * after ensuring that the data has been saved to disk -
1810 * since if the reference remains (rarer), it will be
1811 * read from disk into another page. Splitting into two
1812 * pages would be incorrect if swap supported "shared
1813 * private" pages, but they are handled by tmpfs files.
1815 * Given how unuse_vma() targets one particular offset
1816 * in an anon_vma, once the anon_vma has been determined,
1817 * this splitting happens to be just what is needed to
1818 * handle where KSM pages have been swapped out: re-reading
1819 * is unnecessarily slow, but we can fix that later on.
1821 if (swap_count(*swap_map
) &&
1822 PageDirty(page
) && PageSwapCache(page
)) {
1823 struct writeback_control wbc
= {
1824 .sync_mode
= WB_SYNC_NONE
,
1827 swap_writepage(page
, &wbc
);
1829 wait_on_page_writeback(page
);
1833 * It is conceivable that a racing task removed this page from
1834 * swap cache just before we acquired the page lock at the top,
1835 * or while we dropped it in unuse_mm(). The page might even
1836 * be back in swap cache on another swap area: that we must not
1837 * delete, since it may not have been written out to swap yet.
1839 if (PageSwapCache(page
) &&
1840 likely(page_private(page
) == entry
.val
))
1841 delete_from_swap_cache(page
);
1844 * So we could skip searching mms once swap count went
1845 * to 1, we did not mark any present ptes as dirty: must
1846 * mark page dirty so shrink_page_list will preserve it.
1853 * Make sure that we aren't completely killing
1854 * interactive performance.
1857 if (frontswap
&& pages_to_unuse
> 0) {
1858 if (!--pages_to_unuse
)
1868 * After a successful try_to_unuse, if no swap is now in use, we know
1869 * we can empty the mmlist. swap_lock must be held on entry and exit.
1870 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1871 * added to the mmlist just after page_duplicate - before would be racy.
1873 static void drain_mmlist(void)
1875 struct list_head
*p
, *next
;
1878 for (type
= 0; type
< nr_swapfiles
; type
++)
1879 if (swap_info
[type
]->inuse_pages
)
1881 spin_lock(&mmlist_lock
);
1882 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
1884 spin_unlock(&mmlist_lock
);
1888 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1889 * corresponds to page offset for the specified swap entry.
1890 * Note that the type of this function is sector_t, but it returns page offset
1891 * into the bdev, not sector offset.
1893 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
1895 struct swap_info_struct
*sis
;
1896 struct swap_extent
*start_se
;
1897 struct swap_extent
*se
;
1900 sis
= swap_info
[swp_type(entry
)];
1903 offset
= swp_offset(entry
);
1904 start_se
= sis
->curr_swap_extent
;
1908 if (se
->start_page
<= offset
&&
1909 offset
< (se
->start_page
+ se
->nr_pages
)) {
1910 return se
->start_block
+ (offset
- se
->start_page
);
1912 se
= list_next_entry(se
, list
);
1913 sis
->curr_swap_extent
= se
;
1914 BUG_ON(se
== start_se
); /* It *must* be present */
1919 * Returns the page offset into bdev for the specified page's swap entry.
1921 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
1924 entry
.val
= page_private(page
);
1925 return map_swap_entry(entry
, bdev
);
1929 * Free all of a swapdev's extent information
1931 static void destroy_swap_extents(struct swap_info_struct
*sis
)
1933 while (!list_empty(&sis
->first_swap_extent
.list
)) {
1934 struct swap_extent
*se
;
1936 se
= list_first_entry(&sis
->first_swap_extent
.list
,
1937 struct swap_extent
, list
);
1938 list_del(&se
->list
);
1942 if (sis
->flags
& SWP_FILE
) {
1943 struct file
*swap_file
= sis
->swap_file
;
1944 struct address_space
*mapping
= swap_file
->f_mapping
;
1946 sis
->flags
&= ~SWP_FILE
;
1947 mapping
->a_ops
->swap_deactivate(swap_file
);
1952 * Add a block range (and the corresponding page range) into this swapdev's
1953 * extent list. The extent list is kept sorted in page order.
1955 * This function rather assumes that it is called in ascending page order.
1958 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
1959 unsigned long nr_pages
, sector_t start_block
)
1961 struct swap_extent
*se
;
1962 struct swap_extent
*new_se
;
1963 struct list_head
*lh
;
1965 if (start_page
== 0) {
1966 se
= &sis
->first_swap_extent
;
1967 sis
->curr_swap_extent
= se
;
1969 se
->nr_pages
= nr_pages
;
1970 se
->start_block
= start_block
;
1973 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
1974 se
= list_entry(lh
, struct swap_extent
, list
);
1975 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
1976 if (se
->start_block
+ se
->nr_pages
== start_block
) {
1978 se
->nr_pages
+= nr_pages
;
1984 * No merge. Insert a new extent, preserving ordering.
1986 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
1989 new_se
->start_page
= start_page
;
1990 new_se
->nr_pages
= nr_pages
;
1991 new_se
->start_block
= start_block
;
1993 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
1998 * A `swap extent' is a simple thing which maps a contiguous range of pages
1999 * onto a contiguous range of disk blocks. An ordered list of swap extents
2000 * is built at swapon time and is then used at swap_writepage/swap_readpage
2001 * time for locating where on disk a page belongs.
2003 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2004 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2005 * swap files identically.
2007 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2008 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2009 * swapfiles are handled *identically* after swapon time.
2011 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2012 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2013 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2014 * requirements, they are simply tossed out - we will never use those blocks
2017 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
2018 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
2019 * which will scribble on the fs.
2021 * The amount of disk space which a single swap extent represents varies.
2022 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2023 * extents in the list. To avoid much list walking, we cache the previous
2024 * search location in `curr_swap_extent', and start new searches from there.
2025 * This is extremely effective. The average number of iterations in
2026 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2028 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
2030 struct file
*swap_file
= sis
->swap_file
;
2031 struct address_space
*mapping
= swap_file
->f_mapping
;
2032 struct inode
*inode
= mapping
->host
;
2035 if (S_ISBLK(inode
->i_mode
)) {
2036 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2041 if (mapping
->a_ops
->swap_activate
) {
2042 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
2044 sis
->flags
|= SWP_FILE
;
2045 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2051 return generic_swapfile_activate(sis
, swap_file
, span
);
2054 static void _enable_swap_info(struct swap_info_struct
*p
, int prio
,
2055 unsigned char *swap_map
,
2056 struct swap_cluster_info
*cluster_info
)
2061 p
->prio
= --least_priority
;
2063 * the plist prio is negated because plist ordering is
2064 * low-to-high, while swap ordering is high-to-low
2066 p
->list
.prio
= -p
->prio
;
2067 p
->avail_list
.prio
= -p
->prio
;
2068 p
->swap_map
= swap_map
;
2069 p
->cluster_info
= cluster_info
;
2070 p
->flags
|= SWP_WRITEOK
;
2071 atomic_long_add(p
->pages
, &nr_swap_pages
);
2072 total_swap_pages
+= p
->pages
;
2074 assert_spin_locked(&swap_lock
);
2076 * both lists are plists, and thus priority ordered.
2077 * swap_active_head needs to be priority ordered for swapoff(),
2078 * which on removal of any swap_info_struct with an auto-assigned
2079 * (i.e. negative) priority increments the auto-assigned priority
2080 * of any lower-priority swap_info_structs.
2081 * swap_avail_head needs to be priority ordered for get_swap_page(),
2082 * which allocates swap pages from the highest available priority
2085 plist_add(&p
->list
, &swap_active_head
);
2086 spin_lock(&swap_avail_lock
);
2087 plist_add(&p
->avail_list
, &swap_avail_head
);
2088 spin_unlock(&swap_avail_lock
);
2091 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
2092 unsigned char *swap_map
,
2093 struct swap_cluster_info
*cluster_info
,
2094 unsigned long *frontswap_map
)
2096 frontswap_init(p
->type
, frontswap_map
);
2097 spin_lock(&swap_lock
);
2098 spin_lock(&p
->lock
);
2099 _enable_swap_info(p
, prio
, swap_map
, cluster_info
);
2100 spin_unlock(&p
->lock
);
2101 spin_unlock(&swap_lock
);
2104 static void reinsert_swap_info(struct swap_info_struct
*p
)
2106 spin_lock(&swap_lock
);
2107 spin_lock(&p
->lock
);
2108 _enable_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
2109 spin_unlock(&p
->lock
);
2110 spin_unlock(&swap_lock
);
2113 bool has_usable_swap(void)
2117 spin_lock(&swap_lock
);
2118 if (plist_head_empty(&swap_active_head
))
2120 spin_unlock(&swap_lock
);
2124 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
2126 struct swap_info_struct
*p
= NULL
;
2127 unsigned char *swap_map
;
2128 struct swap_cluster_info
*cluster_info
;
2129 unsigned long *frontswap_map
;
2130 struct file
*swap_file
, *victim
;
2131 struct address_space
*mapping
;
2132 struct inode
*inode
;
2133 struct filename
*pathname
;
2135 unsigned int old_block_size
;
2137 if (!capable(CAP_SYS_ADMIN
))
2140 BUG_ON(!current
->mm
);
2142 pathname
= getname(specialfile
);
2143 if (IS_ERR(pathname
))
2144 return PTR_ERR(pathname
);
2146 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
2147 err
= PTR_ERR(victim
);
2151 mapping
= victim
->f_mapping
;
2152 spin_lock(&swap_lock
);
2153 plist_for_each_entry(p
, &swap_active_head
, list
) {
2154 if (p
->flags
& SWP_WRITEOK
) {
2155 if (p
->swap_file
->f_mapping
== mapping
) {
2163 spin_unlock(&swap_lock
);
2166 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2167 vm_unacct_memory(p
->pages
);
2170 spin_unlock(&swap_lock
);
2173 spin_lock(&swap_avail_lock
);
2174 plist_del(&p
->avail_list
, &swap_avail_head
);
2175 spin_unlock(&swap_avail_lock
);
2176 spin_lock(&p
->lock
);
2178 struct swap_info_struct
*si
= p
;
2180 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2183 si
->avail_list
.prio
--;
2187 plist_del(&p
->list
, &swap_active_head
);
2188 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2189 total_swap_pages
-= p
->pages
;
2190 p
->flags
&= ~SWP_WRITEOK
;
2191 spin_unlock(&p
->lock
);
2192 spin_unlock(&swap_lock
);
2194 disable_swap_slots_cache_lock();
2196 set_current_oom_origin();
2197 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
2198 clear_current_oom_origin();
2201 /* re-insert swap space back into swap_list */
2202 reinsert_swap_info(p
);
2203 reenable_swap_slots_cache_unlock();
2207 reenable_swap_slots_cache_unlock();
2209 flush_work(&p
->discard_work
);
2211 destroy_swap_extents(p
);
2212 if (p
->flags
& SWP_CONTINUED
)
2213 free_swap_count_continuations(p
);
2215 mutex_lock(&swapon_mutex
);
2216 spin_lock(&swap_lock
);
2217 spin_lock(&p
->lock
);
2220 /* wait for anyone still in scan_swap_map */
2221 p
->highest_bit
= 0; /* cuts scans short */
2222 while (p
->flags
>= SWP_SCANNING
) {
2223 spin_unlock(&p
->lock
);
2224 spin_unlock(&swap_lock
);
2225 schedule_timeout_uninterruptible(1);
2226 spin_lock(&swap_lock
);
2227 spin_lock(&p
->lock
);
2230 swap_file
= p
->swap_file
;
2231 old_block_size
= p
->old_block_size
;
2232 p
->swap_file
= NULL
;
2234 swap_map
= p
->swap_map
;
2236 cluster_info
= p
->cluster_info
;
2237 p
->cluster_info
= NULL
;
2238 frontswap_map
= frontswap_map_get(p
);
2239 spin_unlock(&p
->lock
);
2240 spin_unlock(&swap_lock
);
2241 frontswap_invalidate_area(p
->type
);
2242 frontswap_map_set(p
, NULL
);
2243 mutex_unlock(&swapon_mutex
);
2244 free_percpu(p
->percpu_cluster
);
2245 p
->percpu_cluster
= NULL
;
2247 vfree(cluster_info
);
2248 vfree(frontswap_map
);
2249 /* Destroy swap account information */
2250 swap_cgroup_swapoff(p
->type
);
2251 exit_swap_address_space(p
->type
);
2253 inode
= mapping
->host
;
2254 if (S_ISBLK(inode
->i_mode
)) {
2255 struct block_device
*bdev
= I_BDEV(inode
);
2256 set_blocksize(bdev
, old_block_size
);
2257 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2260 inode
->i_flags
&= ~S_SWAPFILE
;
2261 inode_unlock(inode
);
2263 filp_close(swap_file
, NULL
);
2266 * Clear the SWP_USED flag after all resources are freed so that swapon
2267 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2268 * not hold p->lock after we cleared its SWP_WRITEOK.
2270 spin_lock(&swap_lock
);
2272 spin_unlock(&swap_lock
);
2275 atomic_inc(&proc_poll_event
);
2276 wake_up_interruptible(&proc_poll_wait
);
2279 filp_close(victim
, NULL
);
2285 #ifdef CONFIG_PROC_FS
2286 static unsigned swaps_poll(struct file
*file
, poll_table
*wait
)
2288 struct seq_file
*seq
= file
->private_data
;
2290 poll_wait(file
, &proc_poll_wait
, wait
);
2292 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2293 seq
->poll_event
= atomic_read(&proc_poll_event
);
2294 return POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
2297 return POLLIN
| POLLRDNORM
;
2301 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2303 struct swap_info_struct
*si
;
2307 mutex_lock(&swapon_mutex
);
2310 return SEQ_START_TOKEN
;
2312 for (type
= 0; type
< nr_swapfiles
; type
++) {
2313 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2314 si
= swap_info
[type
];
2315 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2324 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2326 struct swap_info_struct
*si
= v
;
2329 if (v
== SEQ_START_TOKEN
)
2332 type
= si
->type
+ 1;
2334 for (; type
< nr_swapfiles
; type
++) {
2335 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2336 si
= swap_info
[type
];
2337 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2346 static void swap_stop(struct seq_file
*swap
, void *v
)
2348 mutex_unlock(&swapon_mutex
);
2351 static int swap_show(struct seq_file
*swap
, void *v
)
2353 struct swap_info_struct
*si
= v
;
2357 if (si
== SEQ_START_TOKEN
) {
2358 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2362 file
= si
->swap_file
;
2363 len
= seq_file_path(swap
, file
, " \t\n\\");
2364 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
2365 len
< 40 ? 40 - len
: 1, " ",
2366 S_ISBLK(file_inode(file
)->i_mode
) ?
2367 "partition" : "file\t",
2368 si
->pages
<< (PAGE_SHIFT
- 10),
2369 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
2374 static const struct seq_operations swaps_op
= {
2375 .start
= swap_start
,
2381 static int swaps_open(struct inode
*inode
, struct file
*file
)
2383 struct seq_file
*seq
;
2386 ret
= seq_open(file
, &swaps_op
);
2390 seq
= file
->private_data
;
2391 seq
->poll_event
= atomic_read(&proc_poll_event
);
2395 static const struct file_operations proc_swaps_operations
= {
2398 .llseek
= seq_lseek
,
2399 .release
= seq_release
,
2403 static int __init
procswaps_init(void)
2405 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
2408 __initcall(procswaps_init
);
2409 #endif /* CONFIG_PROC_FS */
2411 #ifdef MAX_SWAPFILES_CHECK
2412 static int __init
max_swapfiles_check(void)
2414 MAX_SWAPFILES_CHECK();
2417 late_initcall(max_swapfiles_check
);
2420 static struct swap_info_struct
*alloc_swap_info(void)
2422 struct swap_info_struct
*p
;
2425 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
2427 return ERR_PTR(-ENOMEM
);
2429 spin_lock(&swap_lock
);
2430 for (type
= 0; type
< nr_swapfiles
; type
++) {
2431 if (!(swap_info
[type
]->flags
& SWP_USED
))
2434 if (type
>= MAX_SWAPFILES
) {
2435 spin_unlock(&swap_lock
);
2437 return ERR_PTR(-EPERM
);
2439 if (type
>= nr_swapfiles
) {
2441 swap_info
[type
] = p
;
2443 * Write swap_info[type] before nr_swapfiles, in case a
2444 * racing procfs swap_start() or swap_next() is reading them.
2445 * (We never shrink nr_swapfiles, we never free this entry.)
2451 p
= swap_info
[type
];
2453 * Do not memset this entry: a racing procfs swap_next()
2454 * would be relying on p->type to remain valid.
2457 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
2458 plist_node_init(&p
->list
, 0);
2459 plist_node_init(&p
->avail_list
, 0);
2460 p
->flags
= SWP_USED
;
2461 spin_unlock(&swap_lock
);
2462 spin_lock_init(&p
->lock
);
2467 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2471 if (S_ISBLK(inode
->i_mode
)) {
2472 p
->bdev
= bdgrab(I_BDEV(inode
));
2473 error
= blkdev_get(p
->bdev
,
2474 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, p
);
2479 p
->old_block_size
= block_size(p
->bdev
);
2480 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2483 p
->flags
|= SWP_BLKDEV
;
2484 } else if (S_ISREG(inode
->i_mode
)) {
2485 p
->bdev
= inode
->i_sb
->s_bdev
;
2487 if (IS_SWAPFILE(inode
))
2495 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2496 union swap_header
*swap_header
,
2497 struct inode
*inode
)
2500 unsigned long maxpages
;
2501 unsigned long swapfilepages
;
2502 unsigned long last_page
;
2504 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2505 pr_err("Unable to find swap-space signature\n");
2509 /* swap partition endianess hack... */
2510 if (swab32(swap_header
->info
.version
) == 1) {
2511 swab32s(&swap_header
->info
.version
);
2512 swab32s(&swap_header
->info
.last_page
);
2513 swab32s(&swap_header
->info
.nr_badpages
);
2514 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2516 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2517 swab32s(&swap_header
->info
.badpages
[i
]);
2519 /* Check the swap header's sub-version */
2520 if (swap_header
->info
.version
!= 1) {
2521 pr_warn("Unable to handle swap header version %d\n",
2522 swap_header
->info
.version
);
2527 p
->cluster_next
= 1;
2531 * Find out how many pages are allowed for a single swap
2532 * device. There are two limiting factors: 1) the number
2533 * of bits for the swap offset in the swp_entry_t type, and
2534 * 2) the number of bits in the swap pte as defined by the
2535 * different architectures. In order to find the
2536 * largest possible bit mask, a swap entry with swap type 0
2537 * and swap offset ~0UL is created, encoded to a swap pte,
2538 * decoded to a swp_entry_t again, and finally the swap
2539 * offset is extracted. This will mask all the bits from
2540 * the initial ~0UL mask that can't be encoded in either
2541 * the swp_entry_t or the architecture definition of a
2544 maxpages
= swp_offset(pte_to_swp_entry(
2545 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2546 last_page
= swap_header
->info
.last_page
;
2547 if (last_page
> maxpages
) {
2548 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2549 maxpages
<< (PAGE_SHIFT
- 10),
2550 last_page
<< (PAGE_SHIFT
- 10));
2552 if (maxpages
> last_page
) {
2553 maxpages
= last_page
+ 1;
2554 /* p->max is an unsigned int: don't overflow it */
2555 if ((unsigned int)maxpages
== 0)
2556 maxpages
= UINT_MAX
;
2558 p
->highest_bit
= maxpages
- 1;
2562 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
2563 if (swapfilepages
&& maxpages
> swapfilepages
) {
2564 pr_warn("Swap area shorter than signature indicates\n");
2567 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
2569 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2575 #define SWAP_CLUSTER_INFO_COLS \
2576 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2577 #define SWAP_CLUSTER_SPACE_COLS \
2578 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2579 #define SWAP_CLUSTER_COLS \
2580 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2582 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
2583 union swap_header
*swap_header
,
2584 unsigned char *swap_map
,
2585 struct swap_cluster_info
*cluster_info
,
2586 unsigned long maxpages
,
2590 unsigned int nr_good_pages
;
2592 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2593 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
2594 unsigned long i
, idx
;
2596 nr_good_pages
= maxpages
- 1; /* omit header page */
2598 cluster_list_init(&p
->free_clusters
);
2599 cluster_list_init(&p
->discard_clusters
);
2601 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
2602 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
2603 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
2605 if (page_nr
< maxpages
) {
2606 swap_map
[page_nr
] = SWAP_MAP_BAD
;
2609 * Haven't marked the cluster free yet, no list
2610 * operation involved
2612 inc_cluster_info_page(p
, cluster_info
, page_nr
);
2616 /* Haven't marked the cluster free yet, no list operation involved */
2617 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
2618 inc_cluster_info_page(p
, cluster_info
, i
);
2620 if (nr_good_pages
) {
2621 swap_map
[0] = SWAP_MAP_BAD
;
2623 * Not mark the cluster free yet, no list
2624 * operation involved
2626 inc_cluster_info_page(p
, cluster_info
, 0);
2628 p
->pages
= nr_good_pages
;
2629 nr_extents
= setup_swap_extents(p
, span
);
2632 nr_good_pages
= p
->pages
;
2634 if (!nr_good_pages
) {
2635 pr_warn("Empty swap-file\n");
2644 * Reduce false cache line sharing between cluster_info and
2645 * sharing same address space.
2647 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
2648 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
2649 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
2650 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
2651 if (idx
>= nr_clusters
)
2653 if (cluster_count(&cluster_info
[idx
]))
2655 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
2656 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
2664 * Helper to sys_swapon determining if a given swap
2665 * backing device queue supports DISCARD operations.
2667 static bool swap_discardable(struct swap_info_struct
*si
)
2669 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
2671 if (!q
|| !blk_queue_discard(q
))
2677 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
2679 struct swap_info_struct
*p
;
2680 struct filename
*name
;
2681 struct file
*swap_file
= NULL
;
2682 struct address_space
*mapping
;
2685 union swap_header
*swap_header
;
2688 unsigned long maxpages
;
2689 unsigned char *swap_map
= NULL
;
2690 struct swap_cluster_info
*cluster_info
= NULL
;
2691 unsigned long *frontswap_map
= NULL
;
2692 struct page
*page
= NULL
;
2693 struct inode
*inode
= NULL
;
2695 if (swap_flags
& ~SWAP_FLAGS_VALID
)
2698 if (!capable(CAP_SYS_ADMIN
))
2701 p
= alloc_swap_info();
2705 INIT_WORK(&p
->discard_work
, swap_discard_work
);
2707 name
= getname(specialfile
);
2709 error
= PTR_ERR(name
);
2713 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
2714 if (IS_ERR(swap_file
)) {
2715 error
= PTR_ERR(swap_file
);
2720 p
->swap_file
= swap_file
;
2721 mapping
= swap_file
->f_mapping
;
2722 inode
= mapping
->host
;
2724 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
2725 error
= claim_swapfile(p
, inode
);
2726 if (unlikely(error
))
2730 * Read the swap header.
2732 if (!mapping
->a_ops
->readpage
) {
2736 page
= read_mapping_page(mapping
, 0, swap_file
);
2738 error
= PTR_ERR(page
);
2741 swap_header
= kmap(page
);
2743 maxpages
= read_swap_header(p
, swap_header
, inode
);
2744 if (unlikely(!maxpages
)) {
2749 /* OK, set up the swap map and apply the bad block list */
2750 swap_map
= vzalloc(maxpages
);
2756 if (bdi_cap_stable_pages_required(inode_to_bdi(inode
)))
2757 p
->flags
|= SWP_STABLE_WRITES
;
2759 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
2761 unsigned long ci
, nr_cluster
;
2763 p
->flags
|= SWP_SOLIDSTATE
;
2765 * select a random position to start with to help wear leveling
2768 p
->cluster_next
= 1 + (prandom_u32() % p
->highest_bit
);
2769 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2771 cluster_info
= vzalloc(nr_cluster
* sizeof(*cluster_info
));
2772 if (!cluster_info
) {
2777 for (ci
= 0; ci
< nr_cluster
; ci
++)
2778 spin_lock_init(&((cluster_info
+ ci
)->lock
));
2780 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
2781 if (!p
->percpu_cluster
) {
2785 for_each_possible_cpu(cpu
) {
2786 struct percpu_cluster
*cluster
;
2787 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
2788 cluster_set_null(&cluster
->index
);
2792 error
= swap_cgroup_swapon(p
->type
, maxpages
);
2796 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
2797 cluster_info
, maxpages
, &span
);
2798 if (unlikely(nr_extents
< 0)) {
2802 /* frontswap enabled? set up bit-per-page map for frontswap */
2803 if (IS_ENABLED(CONFIG_FRONTSWAP
))
2804 frontswap_map
= vzalloc(BITS_TO_LONGS(maxpages
) * sizeof(long));
2806 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
2808 * When discard is enabled for swap with no particular
2809 * policy flagged, we set all swap discard flags here in
2810 * order to sustain backward compatibility with older
2811 * swapon(8) releases.
2813 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
2817 * By flagging sys_swapon, a sysadmin can tell us to
2818 * either do single-time area discards only, or to just
2819 * perform discards for released swap page-clusters.
2820 * Now it's time to adjust the p->flags accordingly.
2822 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
2823 p
->flags
&= ~SWP_PAGE_DISCARD
;
2824 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
2825 p
->flags
&= ~SWP_AREA_DISCARD
;
2827 /* issue a swapon-time discard if it's still required */
2828 if (p
->flags
& SWP_AREA_DISCARD
) {
2829 int err
= discard_swap(p
);
2831 pr_err("swapon: discard_swap(%p): %d\n",
2836 error
= init_swap_address_space(p
->type
, maxpages
);
2840 mutex_lock(&swapon_mutex
);
2842 if (swap_flags
& SWAP_FLAG_PREFER
)
2844 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
2845 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
2847 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2848 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
2849 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
2850 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
2851 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
2852 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
2853 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
2854 (frontswap_map
) ? "FS" : "");
2856 mutex_unlock(&swapon_mutex
);
2857 atomic_inc(&proc_poll_event
);
2858 wake_up_interruptible(&proc_poll_wait
);
2860 if (S_ISREG(inode
->i_mode
))
2861 inode
->i_flags
|= S_SWAPFILE
;
2865 free_percpu(p
->percpu_cluster
);
2866 p
->percpu_cluster
= NULL
;
2867 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
2868 set_blocksize(p
->bdev
, p
->old_block_size
);
2869 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2871 destroy_swap_extents(p
);
2872 swap_cgroup_swapoff(p
->type
);
2873 spin_lock(&swap_lock
);
2874 p
->swap_file
= NULL
;
2876 spin_unlock(&swap_lock
);
2878 vfree(cluster_info
);
2880 if (inode
&& S_ISREG(inode
->i_mode
)) {
2881 inode_unlock(inode
);
2884 filp_close(swap_file
, NULL
);
2887 if (page
&& !IS_ERR(page
)) {
2893 if (inode
&& S_ISREG(inode
->i_mode
))
2894 inode_unlock(inode
);
2896 enable_swap_slots_cache();
2900 void si_swapinfo(struct sysinfo
*val
)
2903 unsigned long nr_to_be_unused
= 0;
2905 spin_lock(&swap_lock
);
2906 for (type
= 0; type
< nr_swapfiles
; type
++) {
2907 struct swap_info_struct
*si
= swap_info
[type
];
2909 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
2910 nr_to_be_unused
+= si
->inuse_pages
;
2912 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
2913 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
2914 spin_unlock(&swap_lock
);
2918 * Verify that a swap entry is valid and increment its swap map count.
2920 * Returns error code in following case.
2922 * - swp_entry is invalid -> EINVAL
2923 * - swp_entry is migration entry -> EINVAL
2924 * - swap-cache reference is requested but there is already one. -> EEXIST
2925 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2926 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2928 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
2930 struct swap_info_struct
*p
;
2931 struct swap_cluster_info
*ci
;
2932 unsigned long offset
, type
;
2933 unsigned char count
;
2934 unsigned char has_cache
;
2937 if (non_swap_entry(entry
))
2940 type
= swp_type(entry
);
2941 if (type
>= nr_swapfiles
)
2943 p
= swap_info
[type
];
2944 offset
= swp_offset(entry
);
2945 if (unlikely(offset
>= p
->max
))
2948 ci
= lock_cluster_or_swap_info(p
, offset
);
2950 count
= p
->swap_map
[offset
];
2953 * swapin_readahead() doesn't check if a swap entry is valid, so the
2954 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2956 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
2961 has_cache
= count
& SWAP_HAS_CACHE
;
2962 count
&= ~SWAP_HAS_CACHE
;
2965 if (usage
== SWAP_HAS_CACHE
) {
2967 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2968 if (!has_cache
&& count
)
2969 has_cache
= SWAP_HAS_CACHE
;
2970 else if (has_cache
) /* someone else added cache */
2972 else /* no users remaining */
2975 } else if (count
|| has_cache
) {
2977 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
2979 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
2981 else if (swap_count_continued(p
, offset
, count
))
2982 count
= COUNT_CONTINUED
;
2986 err
= -ENOENT
; /* unused swap entry */
2988 p
->swap_map
[offset
] = count
| has_cache
;
2991 unlock_cluster_or_swap_info(p
, ci
);
2996 pr_err("swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
3001 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3002 * (in which case its reference count is never incremented).
3004 void swap_shmem_alloc(swp_entry_t entry
)
3006 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
3010 * Increase reference count of swap entry by 1.
3011 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3012 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3013 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3014 * might occur if a page table entry has got corrupted.
3016 int swap_duplicate(swp_entry_t entry
)
3020 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
3021 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
3026 * @entry: swap entry for which we allocate swap cache.
3028 * Called when allocating swap cache for existing swap entry,
3029 * This can return error codes. Returns 0 at success.
3030 * -EBUSY means there is a swap cache.
3031 * Note: return code is different from swap_duplicate().
3033 int swapcache_prepare(swp_entry_t entry
)
3035 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
3038 struct swap_info_struct
*page_swap_info(struct page
*page
)
3040 swp_entry_t swap
= { .val
= page_private(page
) };
3041 return swap_info
[swp_type(swap
)];
3045 * out-of-line __page_file_ methods to avoid include hell.
3047 struct address_space
*__page_file_mapping(struct page
*page
)
3049 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
3050 return page_swap_info(page
)->swap_file
->f_mapping
;
3052 EXPORT_SYMBOL_GPL(__page_file_mapping
);
3054 pgoff_t
__page_file_index(struct page
*page
)
3056 swp_entry_t swap
= { .val
= page_private(page
) };
3057 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
3058 return swp_offset(swap
);
3060 EXPORT_SYMBOL_GPL(__page_file_index
);
3063 * add_swap_count_continuation - called when a swap count is duplicated
3064 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3065 * page of the original vmalloc'ed swap_map, to hold the continuation count
3066 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3067 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3069 * These continuation pages are seldom referenced: the common paths all work
3070 * on the original swap_map, only referring to a continuation page when the
3071 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3073 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3074 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3075 * can be called after dropping locks.
3077 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
3079 struct swap_info_struct
*si
;
3080 struct swap_cluster_info
*ci
;
3083 struct page
*list_page
;
3085 unsigned char count
;
3088 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3089 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3091 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
3093 si
= swap_info_get(entry
);
3096 * An acceptable race has occurred since the failing
3097 * __swap_duplicate(): the swap entry has been freed,
3098 * perhaps even the whole swap_map cleared for swapoff.
3103 offset
= swp_offset(entry
);
3105 ci
= lock_cluster(si
, offset
);
3107 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
3109 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
3111 * The higher the swap count, the more likely it is that tasks
3112 * will race to add swap count continuation: we need to avoid
3113 * over-provisioning.
3120 spin_unlock(&si
->lock
);
3125 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3126 * no architecture is using highmem pages for kernel page tables: so it
3127 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3129 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3130 offset
&= ~PAGE_MASK
;
3133 * Page allocation does not initialize the page's lru field,
3134 * but it does always reset its private field.
3136 if (!page_private(head
)) {
3137 BUG_ON(count
& COUNT_CONTINUED
);
3138 INIT_LIST_HEAD(&head
->lru
);
3139 set_page_private(head
, SWP_CONTINUED
);
3140 si
->flags
|= SWP_CONTINUED
;
3143 list_for_each_entry(list_page
, &head
->lru
, lru
) {
3147 * If the previous map said no continuation, but we've found
3148 * a continuation page, free our allocation and use this one.
3150 if (!(count
& COUNT_CONTINUED
))
3153 map
= kmap_atomic(list_page
) + offset
;
3158 * If this continuation count now has some space in it,
3159 * free our allocation and use this one.
3161 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
3165 list_add_tail(&page
->lru
, &head
->lru
);
3166 page
= NULL
; /* now it's attached, don't free it */
3169 spin_unlock(&si
->lock
);
3177 * swap_count_continued - when the original swap_map count is incremented
3178 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3179 * into, carry if so, or else fail until a new continuation page is allocated;
3180 * when the original swap_map count is decremented from 0 with continuation,
3181 * borrow from the continuation and report whether it still holds more.
3182 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3185 static bool swap_count_continued(struct swap_info_struct
*si
,
3186 pgoff_t offset
, unsigned char count
)
3192 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3193 if (page_private(head
) != SWP_CONTINUED
) {
3194 BUG_ON(count
& COUNT_CONTINUED
);
3195 return false; /* need to add count continuation */
3198 offset
&= ~PAGE_MASK
;
3199 page
= list_entry(head
->lru
.next
, struct page
, lru
);
3200 map
= kmap_atomic(page
) + offset
;
3202 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3203 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3205 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3207 * Think of how you add 1 to 999
3209 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3211 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3212 BUG_ON(page
== head
);
3213 map
= kmap_atomic(page
) + offset
;
3215 if (*map
== SWAP_CONT_MAX
) {
3217 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3219 return false; /* add count continuation */
3220 map
= kmap_atomic(page
) + offset
;
3221 init_map
: *map
= 0; /* we didn't zero the page */
3225 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3226 while (page
!= head
) {
3227 map
= kmap_atomic(page
) + offset
;
3228 *map
= COUNT_CONTINUED
;
3230 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3232 return true; /* incremented */
3234 } else { /* decrementing */
3236 * Think of how you subtract 1 from 1000
3238 BUG_ON(count
!= COUNT_CONTINUED
);
3239 while (*map
== COUNT_CONTINUED
) {
3241 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3242 BUG_ON(page
== head
);
3243 map
= kmap_atomic(page
) + offset
;
3250 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3251 while (page
!= head
) {
3252 map
= kmap_atomic(page
) + offset
;
3253 *map
= SWAP_CONT_MAX
| count
;
3254 count
= COUNT_CONTINUED
;
3256 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3258 return count
== COUNT_CONTINUED
;
3263 * free_swap_count_continuations - swapoff free all the continuation pages
3264 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3266 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3270 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3272 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3273 if (page_private(head
)) {
3274 struct page
*page
, *next
;
3276 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3277 list_del(&page
->lru
);