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>
37 #include <linux/swap_slots.h>
39 #include <asm/pgtable.h>
40 #include <asm/tlbflush.h>
41 #include <linux/swapops.h>
42 #include <linux/swap_cgroup.h>
44 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
46 static void free_swap_count_continuations(struct swap_info_struct
*);
47 static sector_t
map_swap_entry(swp_entry_t
, struct block_device
**);
49 DEFINE_SPINLOCK(swap_lock
);
50 static unsigned int nr_swapfiles
;
51 atomic_long_t nr_swap_pages
;
53 * Some modules use swappable objects and may try to swap them out under
54 * memory pressure (via the shrinker). Before doing so, they may wish to
55 * check to see if any swap space is available.
57 EXPORT_SYMBOL_GPL(nr_swap_pages
);
58 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
59 long total_swap_pages
;
60 static int least_priority
;
62 static const char Bad_file
[] = "Bad swap file entry ";
63 static const char Unused_file
[] = "Unused swap file entry ";
64 static const char Bad_offset
[] = "Bad swap offset entry ";
65 static const char Unused_offset
[] = "Unused swap offset entry ";
68 * all active swap_info_structs
69 * protected with swap_lock, and ordered by priority.
71 PLIST_HEAD(swap_active_head
);
74 * all available (active, not full) swap_info_structs
75 * protected with swap_avail_lock, ordered by priority.
76 * This is used by get_swap_page() instead of swap_active_head
77 * because swap_active_head includes all swap_info_structs,
78 * but get_swap_page() doesn't need to look at full ones.
79 * This uses its own lock instead of swap_lock because when a
80 * swap_info_struct changes between not-full/full, it needs to
81 * add/remove itself to/from this list, but the swap_info_struct->lock
82 * is held and the locking order requires swap_lock to be taken
83 * before any swap_info_struct->lock.
85 static PLIST_HEAD(swap_avail_head
);
86 static DEFINE_SPINLOCK(swap_avail_lock
);
88 struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
90 static DEFINE_MUTEX(swapon_mutex
);
92 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
93 /* Activity counter to indicate that a swapon or swapoff has occurred */
94 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
96 static inline unsigned char swap_count(unsigned char ent
)
98 return ent
& ~SWAP_HAS_CACHE
; /* may include SWAP_HAS_CONT flag */
101 /* returns 1 if swap entry is freed */
103 __try_to_reclaim_swap(struct swap_info_struct
*si
, unsigned long offset
)
105 swp_entry_t entry
= swp_entry(si
->type
, offset
);
109 page
= find_get_page(swap_address_space(entry
), swp_offset(entry
));
113 * This function is called from scan_swap_map() and it's called
114 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
115 * We have to use trylock for avoiding deadlock. This is a special
116 * case and you should use try_to_free_swap() with explicit lock_page()
117 * in usual operations.
119 if (trylock_page(page
)) {
120 ret
= try_to_free_swap(page
);
128 * swapon tell device that all the old swap contents can be discarded,
129 * to allow the swap device to optimize its wear-levelling.
131 static int discard_swap(struct swap_info_struct
*si
)
133 struct swap_extent
*se
;
134 sector_t start_block
;
138 /* Do not discard the swap header page! */
139 se
= &si
->first_swap_extent
;
140 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
141 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
143 err
= blkdev_issue_discard(si
->bdev
, start_block
,
144 nr_blocks
, GFP_KERNEL
, 0);
150 list_for_each_entry(se
, &si
->first_swap_extent
.list
, list
) {
151 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
152 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
154 err
= blkdev_issue_discard(si
->bdev
, start_block
,
155 nr_blocks
, GFP_KERNEL
, 0);
161 return err
; /* That will often be -EOPNOTSUPP */
165 * swap allocation tell device that a cluster of swap can now be discarded,
166 * to allow the swap device to optimize its wear-levelling.
168 static void discard_swap_cluster(struct swap_info_struct
*si
,
169 pgoff_t start_page
, pgoff_t nr_pages
)
171 struct swap_extent
*se
= si
->curr_swap_extent
;
172 int found_extent
= 0;
175 if (se
->start_page
<= start_page
&&
176 start_page
< se
->start_page
+ se
->nr_pages
) {
177 pgoff_t offset
= start_page
- se
->start_page
;
178 sector_t start_block
= se
->start_block
+ offset
;
179 sector_t nr_blocks
= se
->nr_pages
- offset
;
181 if (nr_blocks
> nr_pages
)
182 nr_blocks
= nr_pages
;
183 start_page
+= nr_blocks
;
184 nr_pages
-= nr_blocks
;
187 si
->curr_swap_extent
= se
;
189 start_block
<<= PAGE_SHIFT
- 9;
190 nr_blocks
<<= PAGE_SHIFT
- 9;
191 if (blkdev_issue_discard(si
->bdev
, start_block
,
192 nr_blocks
, GFP_NOIO
, 0))
196 se
= list_next_entry(se
, list
);
200 #define SWAPFILE_CLUSTER 256
201 #define LATENCY_LIMIT 256
203 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
209 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
214 static inline void cluster_set_count(struct swap_cluster_info
*info
,
220 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
221 unsigned int c
, unsigned int f
)
227 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
232 static inline void cluster_set_next(struct swap_cluster_info
*info
,
238 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
239 unsigned int n
, unsigned int f
)
245 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
247 return info
->flags
& CLUSTER_FLAG_FREE
;
250 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
252 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
255 static inline void cluster_set_null(struct swap_cluster_info
*info
)
257 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
261 static inline struct swap_cluster_info
*lock_cluster(struct swap_info_struct
*si
,
262 unsigned long offset
)
264 struct swap_cluster_info
*ci
;
266 ci
= si
->cluster_info
;
268 ci
+= offset
/ SWAPFILE_CLUSTER
;
269 spin_lock(&ci
->lock
);
274 static inline void unlock_cluster(struct swap_cluster_info
*ci
)
277 spin_unlock(&ci
->lock
);
280 static inline struct swap_cluster_info
*lock_cluster_or_swap_info(
281 struct swap_info_struct
*si
,
282 unsigned long offset
)
284 struct swap_cluster_info
*ci
;
286 ci
= lock_cluster(si
, offset
);
288 spin_lock(&si
->lock
);
293 static inline void unlock_cluster_or_swap_info(struct swap_info_struct
*si
,
294 struct swap_cluster_info
*ci
)
299 spin_unlock(&si
->lock
);
302 static inline bool cluster_list_empty(struct swap_cluster_list
*list
)
304 return cluster_is_null(&list
->head
);
307 static inline unsigned int cluster_list_first(struct swap_cluster_list
*list
)
309 return cluster_next(&list
->head
);
312 static void cluster_list_init(struct swap_cluster_list
*list
)
314 cluster_set_null(&list
->head
);
315 cluster_set_null(&list
->tail
);
318 static void cluster_list_add_tail(struct swap_cluster_list
*list
,
319 struct swap_cluster_info
*ci
,
322 if (cluster_list_empty(list
)) {
323 cluster_set_next_flag(&list
->head
, idx
, 0);
324 cluster_set_next_flag(&list
->tail
, idx
, 0);
326 struct swap_cluster_info
*ci_tail
;
327 unsigned int tail
= cluster_next(&list
->tail
);
330 * Nested cluster lock, but both cluster locks are
331 * only acquired when we held swap_info_struct->lock
334 spin_lock_nested(&ci_tail
->lock
, SINGLE_DEPTH_NESTING
);
335 cluster_set_next(ci_tail
, idx
);
336 unlock_cluster(ci_tail
);
337 cluster_set_next_flag(&list
->tail
, idx
, 0);
341 static unsigned int cluster_list_del_first(struct swap_cluster_list
*list
,
342 struct swap_cluster_info
*ci
)
346 idx
= cluster_next(&list
->head
);
347 if (cluster_next(&list
->tail
) == idx
) {
348 cluster_set_null(&list
->head
);
349 cluster_set_null(&list
->tail
);
351 cluster_set_next_flag(&list
->head
,
352 cluster_next(&ci
[idx
]), 0);
357 /* Add a cluster to discard list and schedule it to do discard */
358 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
362 * If scan_swap_map() can't find a free cluster, it will check
363 * si->swap_map directly. To make sure the discarding cluster isn't
364 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
365 * will be cleared after discard
367 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
368 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
370 cluster_list_add_tail(&si
->discard_clusters
, si
->cluster_info
, idx
);
372 schedule_work(&si
->discard_work
);
376 * Doing discard actually. After a cluster discard is finished, the cluster
377 * will be added to free cluster list. caller should hold si->lock.
379 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
381 struct swap_cluster_info
*info
, *ci
;
384 info
= si
->cluster_info
;
386 while (!cluster_list_empty(&si
->discard_clusters
)) {
387 idx
= cluster_list_del_first(&si
->discard_clusters
, info
);
388 spin_unlock(&si
->lock
);
390 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
393 spin_lock(&si
->lock
);
394 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
395 cluster_set_flag(ci
, CLUSTER_FLAG_FREE
);
397 cluster_list_add_tail(&si
->free_clusters
, info
, idx
);
398 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
399 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
400 0, SWAPFILE_CLUSTER
);
405 static void swap_discard_work(struct work_struct
*work
)
407 struct swap_info_struct
*si
;
409 si
= container_of(work
, struct swap_info_struct
, discard_work
);
411 spin_lock(&si
->lock
);
412 swap_do_scheduled_discard(si
);
413 spin_unlock(&si
->lock
);
417 * The cluster corresponding to page_nr will be used. The cluster will be
418 * removed from free cluster list and its usage counter will be increased.
420 static void inc_cluster_info_page(struct swap_info_struct
*p
,
421 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
423 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
427 if (cluster_is_free(&cluster_info
[idx
])) {
428 VM_BUG_ON(cluster_list_first(&p
->free_clusters
) != idx
);
429 cluster_list_del_first(&p
->free_clusters
, cluster_info
);
430 cluster_set_count_flag(&cluster_info
[idx
], 0, 0);
433 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
434 cluster_set_count(&cluster_info
[idx
],
435 cluster_count(&cluster_info
[idx
]) + 1);
439 * The cluster corresponding to page_nr decreases one usage. If the usage
440 * counter becomes 0, which means no page in the cluster is in using, we can
441 * optionally discard the cluster and add it to free cluster list.
443 static void dec_cluster_info_page(struct swap_info_struct
*p
,
444 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
446 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
451 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
452 cluster_set_count(&cluster_info
[idx
],
453 cluster_count(&cluster_info
[idx
]) - 1);
455 if (cluster_count(&cluster_info
[idx
]) == 0) {
457 * If the swap is discardable, prepare discard the cluster
458 * instead of free it immediately. The cluster will be freed
461 if ((p
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
462 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
463 swap_cluster_schedule_discard(p
, idx
);
467 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
468 cluster_list_add_tail(&p
->free_clusters
, cluster_info
, idx
);
473 * It's possible scan_swap_map() uses a free cluster in the middle of free
474 * cluster list. Avoiding such abuse to avoid list corruption.
477 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
478 unsigned long offset
)
480 struct percpu_cluster
*percpu_cluster
;
483 offset
/= SWAPFILE_CLUSTER
;
484 conflict
= !cluster_list_empty(&si
->free_clusters
) &&
485 offset
!= cluster_list_first(&si
->free_clusters
) &&
486 cluster_is_free(&si
->cluster_info
[offset
]);
491 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
492 cluster_set_null(&percpu_cluster
->index
);
497 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
498 * might involve allocating a new cluster for current CPU too.
500 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
501 unsigned long *offset
, unsigned long *scan_base
)
503 struct percpu_cluster
*cluster
;
504 struct swap_cluster_info
*ci
;
506 unsigned long tmp
, max
;
509 cluster
= this_cpu_ptr(si
->percpu_cluster
);
510 if (cluster_is_null(&cluster
->index
)) {
511 if (!cluster_list_empty(&si
->free_clusters
)) {
512 cluster
->index
= si
->free_clusters
.head
;
513 cluster
->next
= cluster_next(&cluster
->index
) *
515 } else if (!cluster_list_empty(&si
->discard_clusters
)) {
517 * we don't have free cluster but have some clusters in
518 * discarding, do discard now and reclaim them
520 swap_do_scheduled_discard(si
);
521 *scan_base
= *offset
= si
->cluster_next
;
530 * Other CPUs can use our cluster if they can't find a free cluster,
531 * check if there is still free entry in the cluster
534 max
= min_t(unsigned long, si
->max
,
535 (cluster_next(&cluster
->index
) + 1) * SWAPFILE_CLUSTER
);
537 cluster_set_null(&cluster
->index
);
540 ci
= lock_cluster(si
, tmp
);
542 if (!si
->swap_map
[tmp
]) {
550 cluster_set_null(&cluster
->index
);
553 cluster
->next
= tmp
+ 1;
559 static int scan_swap_map_slots(struct swap_info_struct
*si
,
560 unsigned char usage
, int nr
,
563 struct swap_cluster_info
*ci
;
564 unsigned long offset
;
565 unsigned long scan_base
;
566 unsigned long last_in_cluster
= 0;
567 int latency_ration
= LATENCY_LIMIT
;
574 * We try to cluster swap pages by allocating them sequentially
575 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
576 * way, however, we resort to first-free allocation, starting
577 * a new cluster. This prevents us from scattering swap pages
578 * all over the entire swap partition, so that we reduce
579 * overall disk seek times between swap pages. -- sct
580 * But we do now try to find an empty cluster. -Andrea
581 * And we let swap pages go all over an SSD partition. Hugh
584 si
->flags
+= SWP_SCANNING
;
585 scan_base
= offset
= si
->cluster_next
;
588 if (si
->cluster_info
) {
589 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
595 if (unlikely(!si
->cluster_nr
--)) {
596 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
597 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
601 spin_unlock(&si
->lock
);
604 * If seek is expensive, start searching for new cluster from
605 * start of partition, to minimize the span of allocated swap.
606 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
607 * case, just handled by scan_swap_map_try_ssd_cluster() above.
609 scan_base
= offset
= si
->lowest_bit
;
610 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
612 /* Locate the first empty (unaligned) cluster */
613 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
614 if (si
->swap_map
[offset
])
615 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
616 else if (offset
== last_in_cluster
) {
617 spin_lock(&si
->lock
);
618 offset
-= SWAPFILE_CLUSTER
- 1;
619 si
->cluster_next
= offset
;
620 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
623 if (unlikely(--latency_ration
< 0)) {
625 latency_ration
= LATENCY_LIMIT
;
630 spin_lock(&si
->lock
);
631 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
635 if (si
->cluster_info
) {
636 while (scan_swap_map_ssd_cluster_conflict(si
, offset
)) {
637 /* take a break if we already got some slots */
640 if (!scan_swap_map_try_ssd_cluster(si
, &offset
,
645 if (!(si
->flags
& SWP_WRITEOK
))
647 if (!si
->highest_bit
)
649 if (offset
> si
->highest_bit
)
650 scan_base
= offset
= si
->lowest_bit
;
652 ci
= lock_cluster(si
, offset
);
653 /* reuse swap entry of cache-only swap if not busy. */
654 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
657 spin_unlock(&si
->lock
);
658 swap_was_freed
= __try_to_reclaim_swap(si
, offset
);
659 spin_lock(&si
->lock
);
660 /* entry was freed successfully, try to use this again */
663 goto scan
; /* check next one */
666 if (si
->swap_map
[offset
]) {
674 if (offset
== si
->lowest_bit
)
676 if (offset
== si
->highest_bit
)
679 if (si
->inuse_pages
== si
->pages
) {
680 si
->lowest_bit
= si
->max
;
682 spin_lock(&swap_avail_lock
);
683 plist_del(&si
->avail_list
, &swap_avail_head
);
684 spin_unlock(&swap_avail_lock
);
686 si
->swap_map
[offset
] = usage
;
687 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
689 si
->cluster_next
= offset
+ 1;
690 slots
[n_ret
++] = swp_entry(si
->type
, offset
);
692 /* got enough slots or reach max slots? */
693 if ((n_ret
== nr
) || (offset
>= si
->highest_bit
))
696 /* search for next available slot */
698 /* time to take a break? */
699 if (unlikely(--latency_ration
< 0)) {
702 spin_unlock(&si
->lock
);
704 spin_lock(&si
->lock
);
705 latency_ration
= LATENCY_LIMIT
;
708 /* try to get more slots in cluster */
709 if (si
->cluster_info
) {
710 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
718 /* non-ssd case, still more slots in cluster? */
719 if (si
->cluster_nr
&& !si
->swap_map
[offset
]) {
725 si
->flags
-= SWP_SCANNING
;
729 spin_unlock(&si
->lock
);
730 while (++offset
<= si
->highest_bit
) {
731 if (!si
->swap_map
[offset
]) {
732 spin_lock(&si
->lock
);
735 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
736 spin_lock(&si
->lock
);
739 if (unlikely(--latency_ration
< 0)) {
741 latency_ration
= LATENCY_LIMIT
;
744 offset
= si
->lowest_bit
;
745 while (offset
< scan_base
) {
746 if (!si
->swap_map
[offset
]) {
747 spin_lock(&si
->lock
);
750 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
751 spin_lock(&si
->lock
);
754 if (unlikely(--latency_ration
< 0)) {
756 latency_ration
= LATENCY_LIMIT
;
760 spin_lock(&si
->lock
);
763 si
->flags
-= SWP_SCANNING
;
767 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
773 n_ret
= scan_swap_map_slots(si
, usage
, 1, &entry
);
776 return swp_offset(entry
);
782 int get_swap_pages(int n_goal
, swp_entry_t swp_entries
[])
784 struct swap_info_struct
*si
, *next
;
788 avail_pgs
= atomic_long_read(&nr_swap_pages
);
792 if (n_goal
> SWAP_BATCH
)
795 if (n_goal
> avail_pgs
)
798 atomic_long_sub(n_goal
, &nr_swap_pages
);
800 spin_lock(&swap_avail_lock
);
803 plist_for_each_entry_safe(si
, next
, &swap_avail_head
, avail_list
) {
804 /* requeue si to after same-priority siblings */
805 plist_requeue(&si
->avail_list
, &swap_avail_head
);
806 spin_unlock(&swap_avail_lock
);
807 spin_lock(&si
->lock
);
808 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
809 spin_lock(&swap_avail_lock
);
810 if (plist_node_empty(&si
->avail_list
)) {
811 spin_unlock(&si
->lock
);
814 WARN(!si
->highest_bit
,
815 "swap_info %d in list but !highest_bit\n",
817 WARN(!(si
->flags
& SWP_WRITEOK
),
818 "swap_info %d in list but !SWP_WRITEOK\n",
820 plist_del(&si
->avail_list
, &swap_avail_head
);
821 spin_unlock(&si
->lock
);
824 n_ret
= scan_swap_map_slots(si
, SWAP_HAS_CACHE
,
825 n_goal
, swp_entries
);
826 spin_unlock(&si
->lock
);
829 pr_debug("scan_swap_map of si %d failed to find offset\n",
832 spin_lock(&swap_avail_lock
);
835 * if we got here, it's likely that si was almost full before,
836 * and since scan_swap_map() can drop the si->lock, multiple
837 * callers probably all tried to get a page from the same si
838 * and it filled up before we could get one; or, the si filled
839 * up between us dropping swap_avail_lock and taking si->lock.
840 * Since we dropped the swap_avail_lock, the swap_avail_head
841 * list may have been modified; so if next is still in the
842 * swap_avail_head list then try it, otherwise start over
843 * if we have not gotten any slots.
845 if (plist_node_empty(&next
->avail_list
))
849 spin_unlock(&swap_avail_lock
);
853 atomic_long_add((long) (n_goal
-n_ret
), &nr_swap_pages
);
858 /* The only caller of this function is now suspend routine */
859 swp_entry_t
get_swap_page_of_type(int type
)
861 struct swap_info_struct
*si
;
864 si
= swap_info
[type
];
865 spin_lock(&si
->lock
);
866 if (si
&& (si
->flags
& SWP_WRITEOK
)) {
867 atomic_long_dec(&nr_swap_pages
);
868 /* This is called for allocating swap entry, not cache */
869 offset
= scan_swap_map(si
, 1);
871 spin_unlock(&si
->lock
);
872 return swp_entry(type
, offset
);
874 atomic_long_inc(&nr_swap_pages
);
876 spin_unlock(&si
->lock
);
877 return (swp_entry_t
) {0};
880 static struct swap_info_struct
*__swap_info_get(swp_entry_t entry
)
882 struct swap_info_struct
*p
;
883 unsigned long offset
, type
;
887 type
= swp_type(entry
);
888 if (type
>= nr_swapfiles
)
891 if (!(p
->flags
& SWP_USED
))
893 offset
= swp_offset(entry
);
894 if (offset
>= p
->max
)
899 pr_err("swap_info_get: %s%08lx\n", Bad_offset
, entry
.val
);
902 pr_err("swap_info_get: %s%08lx\n", Unused_file
, entry
.val
);
905 pr_err("swap_info_get: %s%08lx\n", Bad_file
, entry
.val
);
910 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
912 struct swap_info_struct
*p
;
914 p
= __swap_info_get(entry
);
917 if (!p
->swap_map
[swp_offset(entry
)])
922 pr_err("swap_info_get: %s%08lx\n", Unused_offset
, entry
.val
);
928 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
930 struct swap_info_struct
*p
;
932 p
= _swap_info_get(entry
);
938 static struct swap_info_struct
*swap_info_get_cont(swp_entry_t entry
,
939 struct swap_info_struct
*q
)
941 struct swap_info_struct
*p
;
943 p
= _swap_info_get(entry
);
947 spin_unlock(&q
->lock
);
954 static unsigned char __swap_entry_free(struct swap_info_struct
*p
,
955 swp_entry_t entry
, unsigned char usage
)
957 struct swap_cluster_info
*ci
;
958 unsigned long offset
= swp_offset(entry
);
960 unsigned char has_cache
;
962 ci
= lock_cluster_or_swap_info(p
, offset
);
964 count
= p
->swap_map
[offset
];
966 has_cache
= count
& SWAP_HAS_CACHE
;
967 count
&= ~SWAP_HAS_CACHE
;
969 if (usage
== SWAP_HAS_CACHE
) {
970 VM_BUG_ON(!has_cache
);
972 } else if (count
== SWAP_MAP_SHMEM
) {
974 * Or we could insist on shmem.c using a special
975 * swap_shmem_free() and free_shmem_swap_and_cache()...
978 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
979 if (count
== COUNT_CONTINUED
) {
980 if (swap_count_continued(p
, offset
, count
))
981 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
983 count
= SWAP_MAP_MAX
;
988 usage
= count
| has_cache
;
989 p
->swap_map
[offset
] = usage
? : SWAP_HAS_CACHE
;
991 unlock_cluster_or_swap_info(p
, ci
);
996 static void swap_entry_free(struct swap_info_struct
*p
, swp_entry_t entry
)
998 struct swap_cluster_info
*ci
;
999 unsigned long offset
= swp_offset(entry
);
1000 unsigned char count
;
1002 ci
= lock_cluster(p
, offset
);
1003 count
= p
->swap_map
[offset
];
1004 VM_BUG_ON(count
!= SWAP_HAS_CACHE
);
1005 p
->swap_map
[offset
] = 0;
1006 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
1009 mem_cgroup_uncharge_swap(entry
);
1010 if (offset
< p
->lowest_bit
)
1011 p
->lowest_bit
= offset
;
1012 if (offset
> p
->highest_bit
) {
1013 bool was_full
= !p
->highest_bit
;
1015 p
->highest_bit
= offset
;
1016 if (was_full
&& (p
->flags
& SWP_WRITEOK
)) {
1017 spin_lock(&swap_avail_lock
);
1018 WARN_ON(!plist_node_empty(&p
->avail_list
));
1019 if (plist_node_empty(&p
->avail_list
))
1020 plist_add(&p
->avail_list
,
1022 spin_unlock(&swap_avail_lock
);
1025 atomic_long_inc(&nr_swap_pages
);
1027 frontswap_invalidate_page(p
->type
, offset
);
1028 if (p
->flags
& SWP_BLKDEV
) {
1029 struct gendisk
*disk
= p
->bdev
->bd_disk
;
1031 if (disk
->fops
->swap_slot_free_notify
)
1032 disk
->fops
->swap_slot_free_notify(p
->bdev
,
1038 * Caller has made sure that the swap device corresponding to entry
1039 * is still around or has not been recycled.
1041 void swap_free(swp_entry_t entry
)
1043 struct swap_info_struct
*p
;
1045 p
= _swap_info_get(entry
);
1047 if (!__swap_entry_free(p
, entry
, 1))
1048 free_swap_slot(entry
);
1053 * Called after dropping swapcache to decrease refcnt to swap entries.
1055 void swapcache_free(swp_entry_t entry
)
1057 struct swap_info_struct
*p
;
1059 p
= _swap_info_get(entry
);
1061 if (!__swap_entry_free(p
, entry
, SWAP_HAS_CACHE
))
1062 free_swap_slot(entry
);
1066 void swapcache_free_entries(swp_entry_t
*entries
, int n
)
1068 struct swap_info_struct
*p
, *prev
;
1076 for (i
= 0; i
< n
; ++i
) {
1077 p
= swap_info_get_cont(entries
[i
], prev
);
1079 swap_entry_free(p
, entries
[i
]);
1085 spin_unlock(&p
->lock
);
1089 * How many references to page are currently swapped out?
1090 * This does not give an exact answer when swap count is continued,
1091 * but does include the high COUNT_CONTINUED flag to allow for that.
1093 int page_swapcount(struct page
*page
)
1096 struct swap_info_struct
*p
;
1097 struct swap_cluster_info
*ci
;
1099 unsigned long offset
;
1101 entry
.val
= page_private(page
);
1102 p
= _swap_info_get(entry
);
1104 offset
= swp_offset(entry
);
1105 ci
= lock_cluster_or_swap_info(p
, offset
);
1106 count
= swap_count(p
->swap_map
[offset
]);
1107 unlock_cluster_or_swap_info(p
, ci
);
1113 * How many references to @entry are currently swapped out?
1114 * This does not give an exact answer when swap count is continued,
1115 * but does include the high COUNT_CONTINUED flag to allow for that.
1117 int __swp_swapcount(swp_entry_t entry
)
1121 struct swap_info_struct
*si
;
1122 struct swap_cluster_info
*ci
;
1124 si
= __swap_info_get(entry
);
1126 offset
= swp_offset(entry
);
1127 ci
= lock_cluster_or_swap_info(si
, offset
);
1128 count
= swap_count(si
->swap_map
[offset
]);
1129 unlock_cluster_or_swap_info(si
, ci
);
1135 * How many references to @entry are currently swapped out?
1136 * This considers COUNT_CONTINUED so it returns exact answer.
1138 int swp_swapcount(swp_entry_t entry
)
1140 int count
, tmp_count
, n
;
1141 struct swap_info_struct
*p
;
1142 struct swap_cluster_info
*ci
;
1147 p
= _swap_info_get(entry
);
1151 offset
= swp_offset(entry
);
1153 ci
= lock_cluster_or_swap_info(p
, offset
);
1155 count
= swap_count(p
->swap_map
[offset
]);
1156 if (!(count
& COUNT_CONTINUED
))
1159 count
&= ~COUNT_CONTINUED
;
1160 n
= SWAP_MAP_MAX
+ 1;
1162 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1163 offset
&= ~PAGE_MASK
;
1164 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1167 page
= list_next_entry(page
, lru
);
1168 map
= kmap_atomic(page
);
1169 tmp_count
= map
[offset
];
1172 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1173 n
*= (SWAP_CONT_MAX
+ 1);
1174 } while (tmp_count
& COUNT_CONTINUED
);
1176 unlock_cluster_or_swap_info(p
, ci
);
1181 * We can write to an anon page without COW if there are no other references
1182 * to it. And as a side-effect, free up its swap: because the old content
1183 * on disk will never be read, and seeking back there to write new content
1184 * later would only waste time away from clustering.
1186 * NOTE: total_mapcount should not be relied upon by the caller if
1187 * reuse_swap_page() returns false, but it may be always overwritten
1188 * (see the other implementation for CONFIG_SWAP=n).
1190 bool reuse_swap_page(struct page
*page
, int *total_mapcount
)
1194 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1195 if (unlikely(PageKsm(page
)))
1197 count
= page_trans_huge_mapcount(page
, total_mapcount
);
1198 if (count
<= 1 && PageSwapCache(page
)) {
1199 count
+= page_swapcount(page
);
1202 if (!PageWriteback(page
)) {
1203 delete_from_swap_cache(page
);
1207 struct swap_info_struct
*p
;
1209 entry
.val
= page_private(page
);
1210 p
= swap_info_get(entry
);
1211 if (p
->flags
& SWP_STABLE_WRITES
) {
1212 spin_unlock(&p
->lock
);
1215 spin_unlock(&p
->lock
);
1223 * If swap is getting full, or if there are no more mappings of this page,
1224 * then try_to_free_swap is called to free its swap space.
1226 int try_to_free_swap(struct page
*page
)
1228 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1230 if (!PageSwapCache(page
))
1232 if (PageWriteback(page
))
1234 if (page_swapcount(page
))
1238 * Once hibernation has begun to create its image of memory,
1239 * there's a danger that one of the calls to try_to_free_swap()
1240 * - most probably a call from __try_to_reclaim_swap() while
1241 * hibernation is allocating its own swap pages for the image,
1242 * but conceivably even a call from memory reclaim - will free
1243 * the swap from a page which has already been recorded in the
1244 * image as a clean swapcache page, and then reuse its swap for
1245 * another page of the image. On waking from hibernation, the
1246 * original page might be freed under memory pressure, then
1247 * later read back in from swap, now with the wrong data.
1249 * Hibernation suspends storage while it is writing the image
1250 * to disk so check that here.
1252 if (pm_suspended_storage())
1255 delete_from_swap_cache(page
);
1261 * Free the swap entry like above, but also try to
1262 * free the page cache entry if it is the last user.
1264 int free_swap_and_cache(swp_entry_t entry
)
1266 struct swap_info_struct
*p
;
1267 struct page
*page
= NULL
;
1268 unsigned char count
;
1270 if (non_swap_entry(entry
))
1273 p
= _swap_info_get(entry
);
1275 count
= __swap_entry_free(p
, entry
, 1);
1276 if (count
== SWAP_HAS_CACHE
) {
1277 page
= find_get_page(swap_address_space(entry
),
1279 if (page
&& !trylock_page(page
)) {
1284 free_swap_slot(entry
);
1288 * Not mapped elsewhere, or swap space full? Free it!
1289 * Also recheck PageSwapCache now page is locked (above).
1291 if (PageSwapCache(page
) && !PageWriteback(page
) &&
1292 (!page_mapped(page
) || mem_cgroup_swap_full(page
))) {
1293 delete_from_swap_cache(page
);
1302 #ifdef CONFIG_HIBERNATION
1304 * Find the swap type that corresponds to given device (if any).
1306 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1307 * from 0, in which the swap header is expected to be located.
1309 * This is needed for the suspend to disk (aka swsusp).
1311 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
1313 struct block_device
*bdev
= NULL
;
1317 bdev
= bdget(device
);
1319 spin_lock(&swap_lock
);
1320 for (type
= 0; type
< nr_swapfiles
; type
++) {
1321 struct swap_info_struct
*sis
= swap_info
[type
];
1323 if (!(sis
->flags
& SWP_WRITEOK
))
1328 *bdev_p
= bdgrab(sis
->bdev
);
1330 spin_unlock(&swap_lock
);
1333 if (bdev
== sis
->bdev
) {
1334 struct swap_extent
*se
= &sis
->first_swap_extent
;
1336 if (se
->start_block
== offset
) {
1338 *bdev_p
= bdgrab(sis
->bdev
);
1340 spin_unlock(&swap_lock
);
1346 spin_unlock(&swap_lock
);
1354 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1355 * corresponding to given index in swap_info (swap type).
1357 sector_t
swapdev_block(int type
, pgoff_t offset
)
1359 struct block_device
*bdev
;
1361 if ((unsigned int)type
>= nr_swapfiles
)
1363 if (!(swap_info
[type
]->flags
& SWP_WRITEOK
))
1365 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
1369 * Return either the total number of swap pages of given type, or the number
1370 * of free pages of that type (depending on @free)
1372 * This is needed for software suspend
1374 unsigned int count_swap_pages(int type
, int free
)
1378 spin_lock(&swap_lock
);
1379 if ((unsigned int)type
< nr_swapfiles
) {
1380 struct swap_info_struct
*sis
= swap_info
[type
];
1382 spin_lock(&sis
->lock
);
1383 if (sis
->flags
& SWP_WRITEOK
) {
1386 n
-= sis
->inuse_pages
;
1388 spin_unlock(&sis
->lock
);
1390 spin_unlock(&swap_lock
);
1393 #endif /* CONFIG_HIBERNATION */
1395 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1397 return pte_same(pte_swp_clear_soft_dirty(pte
), swp_pte
);
1401 * No need to decide whether this PTE shares the swap entry with others,
1402 * just let do_wp_page work it out if a write is requested later - to
1403 * force COW, vm_page_prot omits write permission from any private vma.
1405 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1406 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1408 struct page
*swapcache
;
1409 struct mem_cgroup
*memcg
;
1415 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1416 if (unlikely(!page
))
1419 if (mem_cgroup_try_charge(page
, vma
->vm_mm
, GFP_KERNEL
,
1425 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1426 if (unlikely(!pte_same_as_swp(*pte
, swp_entry_to_pte(entry
)))) {
1427 mem_cgroup_cancel_charge(page
, memcg
, false);
1432 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1433 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1435 set_pte_at(vma
->vm_mm
, addr
, pte
,
1436 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1437 if (page
== swapcache
) {
1438 page_add_anon_rmap(page
, vma
, addr
, false);
1439 mem_cgroup_commit_charge(page
, memcg
, true, false);
1440 } else { /* ksm created a completely new copy */
1441 page_add_new_anon_rmap(page
, vma
, addr
, false);
1442 mem_cgroup_commit_charge(page
, memcg
, false, false);
1443 lru_cache_add_active_or_unevictable(page
, vma
);
1447 * Move the page to the active list so it is not
1448 * immediately swapped out again after swapon.
1450 activate_page(page
);
1452 pte_unmap_unlock(pte
, ptl
);
1454 if (page
!= swapcache
) {
1461 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1462 unsigned long addr
, unsigned long end
,
1463 swp_entry_t entry
, struct page
*page
)
1465 pte_t swp_pte
= swp_entry_to_pte(entry
);
1470 * We don't actually need pte lock while scanning for swp_pte: since
1471 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1472 * page table while we're scanning; though it could get zapped, and on
1473 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1474 * of unmatched parts which look like swp_pte, so unuse_pte must
1475 * recheck under pte lock. Scanning without pte lock lets it be
1476 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1478 pte
= pte_offset_map(pmd
, addr
);
1481 * swapoff spends a _lot_ of time in this loop!
1482 * Test inline before going to call unuse_pte.
1484 if (unlikely(pte_same_as_swp(*pte
, swp_pte
))) {
1486 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1489 pte
= pte_offset_map(pmd
, addr
);
1491 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1497 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
1498 unsigned long addr
, unsigned long end
,
1499 swp_entry_t entry
, struct page
*page
)
1505 pmd
= pmd_offset(pud
, addr
);
1508 next
= pmd_addr_end(addr
, end
);
1509 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1511 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
);
1514 } while (pmd
++, addr
= next
, addr
!= end
);
1518 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
1519 unsigned long addr
, unsigned long end
,
1520 swp_entry_t entry
, struct page
*page
)
1526 pud
= pud_offset(pgd
, addr
);
1528 next
= pud_addr_end(addr
, end
);
1529 if (pud_none_or_clear_bad(pud
))
1531 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
);
1534 } while (pud
++, addr
= next
, addr
!= end
);
1538 static int unuse_vma(struct vm_area_struct
*vma
,
1539 swp_entry_t entry
, struct page
*page
)
1542 unsigned long addr
, end
, next
;
1545 if (page_anon_vma(page
)) {
1546 addr
= page_address_in_vma(page
, vma
);
1547 if (addr
== -EFAULT
)
1550 end
= addr
+ PAGE_SIZE
;
1552 addr
= vma
->vm_start
;
1556 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1558 next
= pgd_addr_end(addr
, end
);
1559 if (pgd_none_or_clear_bad(pgd
))
1561 ret
= unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
);
1564 } while (pgd
++, addr
= next
, addr
!= end
);
1568 static int unuse_mm(struct mm_struct
*mm
,
1569 swp_entry_t entry
, struct page
*page
)
1571 struct vm_area_struct
*vma
;
1574 if (!down_read_trylock(&mm
->mmap_sem
)) {
1576 * Activate page so shrink_inactive_list is unlikely to unmap
1577 * its ptes while lock is dropped, so swapoff can make progress.
1579 activate_page(page
);
1581 down_read(&mm
->mmap_sem
);
1584 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1585 if (vma
->anon_vma
&& (ret
= unuse_vma(vma
, entry
, page
)))
1589 up_read(&mm
->mmap_sem
);
1590 return (ret
< 0)? ret
: 0;
1594 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1595 * from current position to next entry still in use.
1596 * Recycle to start on reaching the end, returning 0 when empty.
1598 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
1599 unsigned int prev
, bool frontswap
)
1601 unsigned int max
= si
->max
;
1602 unsigned int i
= prev
;
1603 unsigned char count
;
1606 * No need for swap_lock here: we're just looking
1607 * for whether an entry is in use, not modifying it; false
1608 * hits are okay, and sys_swapoff() has already prevented new
1609 * allocations from this area (while holding swap_lock).
1618 * No entries in use at top of swap_map,
1619 * loop back to start and recheck there.
1625 count
= READ_ONCE(si
->swap_map
[i
]);
1626 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
1627 if (!frontswap
|| frontswap_test(si
, i
))
1629 if ((i
% LATENCY_LIMIT
) == 0)
1636 * We completely avoid races by reading each swap page in advance,
1637 * and then search for the process using it. All the necessary
1638 * page table adjustments can then be made atomically.
1640 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1641 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1643 int try_to_unuse(unsigned int type
, bool frontswap
,
1644 unsigned long pages_to_unuse
)
1646 struct swap_info_struct
*si
= swap_info
[type
];
1647 struct mm_struct
*start_mm
;
1648 volatile unsigned char *swap_map
; /* swap_map is accessed without
1649 * locking. Mark it as volatile
1650 * to prevent compiler doing
1653 unsigned char swcount
;
1660 * When searching mms for an entry, a good strategy is to
1661 * start at the first mm we freed the previous entry from
1662 * (though actually we don't notice whether we or coincidence
1663 * freed the entry). Initialize this start_mm with a hold.
1665 * A simpler strategy would be to start at the last mm we
1666 * freed the previous entry from; but that would take less
1667 * advantage of mmlist ordering, which clusters forked mms
1668 * together, child after parent. If we race with dup_mmap(), we
1669 * prefer to resolve parent before child, lest we miss entries
1670 * duplicated after we scanned child: using last mm would invert
1673 start_mm
= &init_mm
;
1674 atomic_inc(&init_mm
.mm_users
);
1677 * Keep on scanning until all entries have gone. Usually,
1678 * one pass through swap_map is enough, but not necessarily:
1679 * there are races when an instance of an entry might be missed.
1681 while ((i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
1682 if (signal_pending(current
)) {
1688 * Get a page for the entry, using the existing swap
1689 * cache page if there is one. Otherwise, get a clean
1690 * page and read the swap into it.
1692 swap_map
= &si
->swap_map
[i
];
1693 entry
= swp_entry(type
, i
);
1694 page
= read_swap_cache_async(entry
,
1695 GFP_HIGHUSER_MOVABLE
, NULL
, 0);
1698 * Either swap_duplicate() failed because entry
1699 * has been freed independently, and will not be
1700 * reused since sys_swapoff() already disabled
1701 * allocation from here, or alloc_page() failed.
1703 swcount
= *swap_map
;
1705 * We don't hold lock here, so the swap entry could be
1706 * SWAP_MAP_BAD (when the cluster is discarding).
1707 * Instead of fail out, We can just skip the swap
1708 * entry because swapoff will wait for discarding
1711 if (!swcount
|| swcount
== SWAP_MAP_BAD
)
1718 * Don't hold on to start_mm if it looks like exiting.
1720 if (atomic_read(&start_mm
->mm_users
) == 1) {
1722 start_mm
= &init_mm
;
1723 atomic_inc(&init_mm
.mm_users
);
1727 * Wait for and lock page. When do_swap_page races with
1728 * try_to_unuse, do_swap_page can handle the fault much
1729 * faster than try_to_unuse can locate the entry. This
1730 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1731 * defer to do_swap_page in such a case - in some tests,
1732 * do_swap_page and try_to_unuse repeatedly compete.
1734 wait_on_page_locked(page
);
1735 wait_on_page_writeback(page
);
1737 wait_on_page_writeback(page
);
1740 * Remove all references to entry.
1742 swcount
= *swap_map
;
1743 if (swap_count(swcount
) == SWAP_MAP_SHMEM
) {
1744 retval
= shmem_unuse(entry
, page
);
1745 /* page has already been unlocked and released */
1750 if (swap_count(swcount
) && start_mm
!= &init_mm
)
1751 retval
= unuse_mm(start_mm
, entry
, page
);
1753 if (swap_count(*swap_map
)) {
1754 int set_start_mm
= (*swap_map
>= swcount
);
1755 struct list_head
*p
= &start_mm
->mmlist
;
1756 struct mm_struct
*new_start_mm
= start_mm
;
1757 struct mm_struct
*prev_mm
= start_mm
;
1758 struct mm_struct
*mm
;
1760 atomic_inc(&new_start_mm
->mm_users
);
1761 atomic_inc(&prev_mm
->mm_users
);
1762 spin_lock(&mmlist_lock
);
1763 while (swap_count(*swap_map
) && !retval
&&
1764 (p
= p
->next
) != &start_mm
->mmlist
) {
1765 mm
= list_entry(p
, struct mm_struct
, mmlist
);
1766 if (!atomic_inc_not_zero(&mm
->mm_users
))
1768 spin_unlock(&mmlist_lock
);
1774 swcount
= *swap_map
;
1775 if (!swap_count(swcount
)) /* any usage ? */
1777 else if (mm
== &init_mm
)
1780 retval
= unuse_mm(mm
, entry
, page
);
1782 if (set_start_mm
&& *swap_map
< swcount
) {
1783 mmput(new_start_mm
);
1784 atomic_inc(&mm
->mm_users
);
1788 spin_lock(&mmlist_lock
);
1790 spin_unlock(&mmlist_lock
);
1793 start_mm
= new_start_mm
;
1802 * If a reference remains (rare), we would like to leave
1803 * the page in the swap cache; but try_to_unmap could
1804 * then re-duplicate the entry once we drop page lock,
1805 * so we might loop indefinitely; also, that page could
1806 * not be swapped out to other storage meanwhile. So:
1807 * delete from cache even if there's another reference,
1808 * after ensuring that the data has been saved to disk -
1809 * since if the reference remains (rarer), it will be
1810 * read from disk into another page. Splitting into two
1811 * pages would be incorrect if swap supported "shared
1812 * private" pages, but they are handled by tmpfs files.
1814 * Given how unuse_vma() targets one particular offset
1815 * in an anon_vma, once the anon_vma has been determined,
1816 * this splitting happens to be just what is needed to
1817 * handle where KSM pages have been swapped out: re-reading
1818 * is unnecessarily slow, but we can fix that later on.
1820 if (swap_count(*swap_map
) &&
1821 PageDirty(page
) && PageSwapCache(page
)) {
1822 struct writeback_control wbc
= {
1823 .sync_mode
= WB_SYNC_NONE
,
1826 swap_writepage(page
, &wbc
);
1828 wait_on_page_writeback(page
);
1832 * It is conceivable that a racing task removed this page from
1833 * swap cache just before we acquired the page lock at the top,
1834 * or while we dropped it in unuse_mm(). The page might even
1835 * be back in swap cache on another swap area: that we must not
1836 * delete, since it may not have been written out to swap yet.
1838 if (PageSwapCache(page
) &&
1839 likely(page_private(page
) == entry
.val
))
1840 delete_from_swap_cache(page
);
1843 * So we could skip searching mms once swap count went
1844 * to 1, we did not mark any present ptes as dirty: must
1845 * mark page dirty so shrink_page_list will preserve it.
1852 * Make sure that we aren't completely killing
1853 * interactive performance.
1856 if (frontswap
&& pages_to_unuse
> 0) {
1857 if (!--pages_to_unuse
)
1867 * After a successful try_to_unuse, if no swap is now in use, we know
1868 * we can empty the mmlist. swap_lock must be held on entry and exit.
1869 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1870 * added to the mmlist just after page_duplicate - before would be racy.
1872 static void drain_mmlist(void)
1874 struct list_head
*p
, *next
;
1877 for (type
= 0; type
< nr_swapfiles
; type
++)
1878 if (swap_info
[type
]->inuse_pages
)
1880 spin_lock(&mmlist_lock
);
1881 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
1883 spin_unlock(&mmlist_lock
);
1887 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1888 * corresponds to page offset for the specified swap entry.
1889 * Note that the type of this function is sector_t, but it returns page offset
1890 * into the bdev, not sector offset.
1892 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
1894 struct swap_info_struct
*sis
;
1895 struct swap_extent
*start_se
;
1896 struct swap_extent
*se
;
1899 sis
= swap_info
[swp_type(entry
)];
1902 offset
= swp_offset(entry
);
1903 start_se
= sis
->curr_swap_extent
;
1907 if (se
->start_page
<= offset
&&
1908 offset
< (se
->start_page
+ se
->nr_pages
)) {
1909 return se
->start_block
+ (offset
- se
->start_page
);
1911 se
= list_next_entry(se
, list
);
1912 sis
->curr_swap_extent
= se
;
1913 BUG_ON(se
== start_se
); /* It *must* be present */
1918 * Returns the page offset into bdev for the specified page's swap entry.
1920 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
1923 entry
.val
= page_private(page
);
1924 return map_swap_entry(entry
, bdev
);
1928 * Free all of a swapdev's extent information
1930 static void destroy_swap_extents(struct swap_info_struct
*sis
)
1932 while (!list_empty(&sis
->first_swap_extent
.list
)) {
1933 struct swap_extent
*se
;
1935 se
= list_first_entry(&sis
->first_swap_extent
.list
,
1936 struct swap_extent
, list
);
1937 list_del(&se
->list
);
1941 if (sis
->flags
& SWP_FILE
) {
1942 struct file
*swap_file
= sis
->swap_file
;
1943 struct address_space
*mapping
= swap_file
->f_mapping
;
1945 sis
->flags
&= ~SWP_FILE
;
1946 mapping
->a_ops
->swap_deactivate(swap_file
);
1951 * Add a block range (and the corresponding page range) into this swapdev's
1952 * extent list. The extent list is kept sorted in page order.
1954 * This function rather assumes that it is called in ascending page order.
1957 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
1958 unsigned long nr_pages
, sector_t start_block
)
1960 struct swap_extent
*se
;
1961 struct swap_extent
*new_se
;
1962 struct list_head
*lh
;
1964 if (start_page
== 0) {
1965 se
= &sis
->first_swap_extent
;
1966 sis
->curr_swap_extent
= se
;
1968 se
->nr_pages
= nr_pages
;
1969 se
->start_block
= start_block
;
1972 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
1973 se
= list_entry(lh
, struct swap_extent
, list
);
1974 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
1975 if (se
->start_block
+ se
->nr_pages
== start_block
) {
1977 se
->nr_pages
+= nr_pages
;
1983 * No merge. Insert a new extent, preserving ordering.
1985 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
1988 new_se
->start_page
= start_page
;
1989 new_se
->nr_pages
= nr_pages
;
1990 new_se
->start_block
= start_block
;
1992 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
1997 * A `swap extent' is a simple thing which maps a contiguous range of pages
1998 * onto a contiguous range of disk blocks. An ordered list of swap extents
1999 * is built at swapon time and is then used at swap_writepage/swap_readpage
2000 * time for locating where on disk a page belongs.
2002 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2003 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2004 * swap files identically.
2006 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2007 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2008 * swapfiles are handled *identically* after swapon time.
2010 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2011 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2012 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2013 * requirements, they are simply tossed out - we will never use those blocks
2016 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
2017 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
2018 * which will scribble on the fs.
2020 * The amount of disk space which a single swap extent represents varies.
2021 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2022 * extents in the list. To avoid much list walking, we cache the previous
2023 * search location in `curr_swap_extent', and start new searches from there.
2024 * This is extremely effective. The average number of iterations in
2025 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2027 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
2029 struct file
*swap_file
= sis
->swap_file
;
2030 struct address_space
*mapping
= swap_file
->f_mapping
;
2031 struct inode
*inode
= mapping
->host
;
2034 if (S_ISBLK(inode
->i_mode
)) {
2035 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2040 if (mapping
->a_ops
->swap_activate
) {
2041 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
2043 sis
->flags
|= SWP_FILE
;
2044 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2050 return generic_swapfile_activate(sis
, swap_file
, span
);
2053 static void _enable_swap_info(struct swap_info_struct
*p
, int prio
,
2054 unsigned char *swap_map
,
2055 struct swap_cluster_info
*cluster_info
)
2060 p
->prio
= --least_priority
;
2062 * the plist prio is negated because plist ordering is
2063 * low-to-high, while swap ordering is high-to-low
2065 p
->list
.prio
= -p
->prio
;
2066 p
->avail_list
.prio
= -p
->prio
;
2067 p
->swap_map
= swap_map
;
2068 p
->cluster_info
= cluster_info
;
2069 p
->flags
|= SWP_WRITEOK
;
2070 atomic_long_add(p
->pages
, &nr_swap_pages
);
2071 total_swap_pages
+= p
->pages
;
2073 assert_spin_locked(&swap_lock
);
2075 * both lists are plists, and thus priority ordered.
2076 * swap_active_head needs to be priority ordered for swapoff(),
2077 * which on removal of any swap_info_struct with an auto-assigned
2078 * (i.e. negative) priority increments the auto-assigned priority
2079 * of any lower-priority swap_info_structs.
2080 * swap_avail_head needs to be priority ordered for get_swap_page(),
2081 * which allocates swap pages from the highest available priority
2084 plist_add(&p
->list
, &swap_active_head
);
2085 spin_lock(&swap_avail_lock
);
2086 plist_add(&p
->avail_list
, &swap_avail_head
);
2087 spin_unlock(&swap_avail_lock
);
2090 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
2091 unsigned char *swap_map
,
2092 struct swap_cluster_info
*cluster_info
,
2093 unsigned long *frontswap_map
)
2095 frontswap_init(p
->type
, frontswap_map
);
2096 spin_lock(&swap_lock
);
2097 spin_lock(&p
->lock
);
2098 _enable_swap_info(p
, prio
, swap_map
, cluster_info
);
2099 spin_unlock(&p
->lock
);
2100 spin_unlock(&swap_lock
);
2103 static void reinsert_swap_info(struct swap_info_struct
*p
)
2105 spin_lock(&swap_lock
);
2106 spin_lock(&p
->lock
);
2107 _enable_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
2108 spin_unlock(&p
->lock
);
2109 spin_unlock(&swap_lock
);
2112 bool has_usable_swap(void)
2116 spin_lock(&swap_lock
);
2117 if (plist_head_empty(&swap_active_head
))
2119 spin_unlock(&swap_lock
);
2123 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
2125 struct swap_info_struct
*p
= NULL
;
2126 unsigned char *swap_map
;
2127 struct swap_cluster_info
*cluster_info
;
2128 unsigned long *frontswap_map
;
2129 struct file
*swap_file
, *victim
;
2130 struct address_space
*mapping
;
2131 struct inode
*inode
;
2132 struct filename
*pathname
;
2134 unsigned int old_block_size
;
2136 if (!capable(CAP_SYS_ADMIN
))
2139 BUG_ON(!current
->mm
);
2141 pathname
= getname(specialfile
);
2142 if (IS_ERR(pathname
))
2143 return PTR_ERR(pathname
);
2145 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
2146 err
= PTR_ERR(victim
);
2150 mapping
= victim
->f_mapping
;
2151 spin_lock(&swap_lock
);
2152 plist_for_each_entry(p
, &swap_active_head
, list
) {
2153 if (p
->flags
& SWP_WRITEOK
) {
2154 if (p
->swap_file
->f_mapping
== mapping
) {
2162 spin_unlock(&swap_lock
);
2165 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2166 vm_unacct_memory(p
->pages
);
2169 spin_unlock(&swap_lock
);
2172 spin_lock(&swap_avail_lock
);
2173 plist_del(&p
->avail_list
, &swap_avail_head
);
2174 spin_unlock(&swap_avail_lock
);
2175 spin_lock(&p
->lock
);
2177 struct swap_info_struct
*si
= p
;
2179 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2182 si
->avail_list
.prio
--;
2186 plist_del(&p
->list
, &swap_active_head
);
2187 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2188 total_swap_pages
-= p
->pages
;
2189 p
->flags
&= ~SWP_WRITEOK
;
2190 spin_unlock(&p
->lock
);
2191 spin_unlock(&swap_lock
);
2193 disable_swap_slots_cache_lock();
2195 set_current_oom_origin();
2196 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
2197 clear_current_oom_origin();
2200 /* re-insert swap space back into swap_list */
2201 reinsert_swap_info(p
);
2202 reenable_swap_slots_cache_unlock();
2206 reenable_swap_slots_cache_unlock();
2208 flush_work(&p
->discard_work
);
2210 destroy_swap_extents(p
);
2211 if (p
->flags
& SWP_CONTINUED
)
2212 free_swap_count_continuations(p
);
2214 mutex_lock(&swapon_mutex
);
2215 spin_lock(&swap_lock
);
2216 spin_lock(&p
->lock
);
2219 /* wait for anyone still in scan_swap_map */
2220 p
->highest_bit
= 0; /* cuts scans short */
2221 while (p
->flags
>= SWP_SCANNING
) {
2222 spin_unlock(&p
->lock
);
2223 spin_unlock(&swap_lock
);
2224 schedule_timeout_uninterruptible(1);
2225 spin_lock(&swap_lock
);
2226 spin_lock(&p
->lock
);
2229 swap_file
= p
->swap_file
;
2230 old_block_size
= p
->old_block_size
;
2231 p
->swap_file
= NULL
;
2233 swap_map
= p
->swap_map
;
2235 cluster_info
= p
->cluster_info
;
2236 p
->cluster_info
= NULL
;
2237 frontswap_map
= frontswap_map_get(p
);
2238 spin_unlock(&p
->lock
);
2239 spin_unlock(&swap_lock
);
2240 frontswap_invalidate_area(p
->type
);
2241 frontswap_map_set(p
, NULL
);
2242 mutex_unlock(&swapon_mutex
);
2243 free_percpu(p
->percpu_cluster
);
2244 p
->percpu_cluster
= NULL
;
2246 vfree(cluster_info
);
2247 vfree(frontswap_map
);
2248 /* Destroy swap account information */
2249 swap_cgroup_swapoff(p
->type
);
2250 exit_swap_address_space(p
->type
);
2252 inode
= mapping
->host
;
2253 if (S_ISBLK(inode
->i_mode
)) {
2254 struct block_device
*bdev
= I_BDEV(inode
);
2255 set_blocksize(bdev
, old_block_size
);
2256 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2259 inode
->i_flags
&= ~S_SWAPFILE
;
2260 inode_unlock(inode
);
2262 filp_close(swap_file
, NULL
);
2265 * Clear the SWP_USED flag after all resources are freed so that swapon
2266 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2267 * not hold p->lock after we cleared its SWP_WRITEOK.
2269 spin_lock(&swap_lock
);
2271 spin_unlock(&swap_lock
);
2274 atomic_inc(&proc_poll_event
);
2275 wake_up_interruptible(&proc_poll_wait
);
2278 filp_close(victim
, NULL
);
2284 #ifdef CONFIG_PROC_FS
2285 static unsigned swaps_poll(struct file
*file
, poll_table
*wait
)
2287 struct seq_file
*seq
= file
->private_data
;
2289 poll_wait(file
, &proc_poll_wait
, wait
);
2291 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2292 seq
->poll_event
= atomic_read(&proc_poll_event
);
2293 return POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
2296 return POLLIN
| POLLRDNORM
;
2300 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2302 struct swap_info_struct
*si
;
2306 mutex_lock(&swapon_mutex
);
2309 return SEQ_START_TOKEN
;
2311 for (type
= 0; type
< nr_swapfiles
; type
++) {
2312 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2313 si
= swap_info
[type
];
2314 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2323 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2325 struct swap_info_struct
*si
= v
;
2328 if (v
== SEQ_START_TOKEN
)
2331 type
= si
->type
+ 1;
2333 for (; type
< nr_swapfiles
; type
++) {
2334 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2335 si
= swap_info
[type
];
2336 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2345 static void swap_stop(struct seq_file
*swap
, void *v
)
2347 mutex_unlock(&swapon_mutex
);
2350 static int swap_show(struct seq_file
*swap
, void *v
)
2352 struct swap_info_struct
*si
= v
;
2356 if (si
== SEQ_START_TOKEN
) {
2357 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2361 file
= si
->swap_file
;
2362 len
= seq_file_path(swap
, file
, " \t\n\\");
2363 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
2364 len
< 40 ? 40 - len
: 1, " ",
2365 S_ISBLK(file_inode(file
)->i_mode
) ?
2366 "partition" : "file\t",
2367 si
->pages
<< (PAGE_SHIFT
- 10),
2368 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
2373 static const struct seq_operations swaps_op
= {
2374 .start
= swap_start
,
2380 static int swaps_open(struct inode
*inode
, struct file
*file
)
2382 struct seq_file
*seq
;
2385 ret
= seq_open(file
, &swaps_op
);
2389 seq
= file
->private_data
;
2390 seq
->poll_event
= atomic_read(&proc_poll_event
);
2394 static const struct file_operations proc_swaps_operations
= {
2397 .llseek
= seq_lseek
,
2398 .release
= seq_release
,
2402 static int __init
procswaps_init(void)
2404 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
2407 __initcall(procswaps_init
);
2408 #endif /* CONFIG_PROC_FS */
2410 #ifdef MAX_SWAPFILES_CHECK
2411 static int __init
max_swapfiles_check(void)
2413 MAX_SWAPFILES_CHECK();
2416 late_initcall(max_swapfiles_check
);
2419 static struct swap_info_struct
*alloc_swap_info(void)
2421 struct swap_info_struct
*p
;
2424 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
2426 return ERR_PTR(-ENOMEM
);
2428 spin_lock(&swap_lock
);
2429 for (type
= 0; type
< nr_swapfiles
; type
++) {
2430 if (!(swap_info
[type
]->flags
& SWP_USED
))
2433 if (type
>= MAX_SWAPFILES
) {
2434 spin_unlock(&swap_lock
);
2436 return ERR_PTR(-EPERM
);
2438 if (type
>= nr_swapfiles
) {
2440 swap_info
[type
] = p
;
2442 * Write swap_info[type] before nr_swapfiles, in case a
2443 * racing procfs swap_start() or swap_next() is reading them.
2444 * (We never shrink nr_swapfiles, we never free this entry.)
2450 p
= swap_info
[type
];
2452 * Do not memset this entry: a racing procfs swap_next()
2453 * would be relying on p->type to remain valid.
2456 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
2457 plist_node_init(&p
->list
, 0);
2458 plist_node_init(&p
->avail_list
, 0);
2459 p
->flags
= SWP_USED
;
2460 spin_unlock(&swap_lock
);
2461 spin_lock_init(&p
->lock
);
2466 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2470 if (S_ISBLK(inode
->i_mode
)) {
2471 p
->bdev
= bdgrab(I_BDEV(inode
));
2472 error
= blkdev_get(p
->bdev
,
2473 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, p
);
2478 p
->old_block_size
= block_size(p
->bdev
);
2479 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2482 p
->flags
|= SWP_BLKDEV
;
2483 } else if (S_ISREG(inode
->i_mode
)) {
2484 p
->bdev
= inode
->i_sb
->s_bdev
;
2486 if (IS_SWAPFILE(inode
))
2494 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2495 union swap_header
*swap_header
,
2496 struct inode
*inode
)
2499 unsigned long maxpages
;
2500 unsigned long swapfilepages
;
2501 unsigned long last_page
;
2503 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2504 pr_err("Unable to find swap-space signature\n");
2508 /* swap partition endianess hack... */
2509 if (swab32(swap_header
->info
.version
) == 1) {
2510 swab32s(&swap_header
->info
.version
);
2511 swab32s(&swap_header
->info
.last_page
);
2512 swab32s(&swap_header
->info
.nr_badpages
);
2513 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2515 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2516 swab32s(&swap_header
->info
.badpages
[i
]);
2518 /* Check the swap header's sub-version */
2519 if (swap_header
->info
.version
!= 1) {
2520 pr_warn("Unable to handle swap header version %d\n",
2521 swap_header
->info
.version
);
2526 p
->cluster_next
= 1;
2530 * Find out how many pages are allowed for a single swap
2531 * device. There are two limiting factors: 1) the number
2532 * of bits for the swap offset in the swp_entry_t type, and
2533 * 2) the number of bits in the swap pte as defined by the
2534 * different architectures. In order to find the
2535 * largest possible bit mask, a swap entry with swap type 0
2536 * and swap offset ~0UL is created, encoded to a swap pte,
2537 * decoded to a swp_entry_t again, and finally the swap
2538 * offset is extracted. This will mask all the bits from
2539 * the initial ~0UL mask that can't be encoded in either
2540 * the swp_entry_t or the architecture definition of a
2543 maxpages
= swp_offset(pte_to_swp_entry(
2544 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2545 last_page
= swap_header
->info
.last_page
;
2546 if (last_page
> maxpages
) {
2547 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2548 maxpages
<< (PAGE_SHIFT
- 10),
2549 last_page
<< (PAGE_SHIFT
- 10));
2551 if (maxpages
> last_page
) {
2552 maxpages
= last_page
+ 1;
2553 /* p->max is an unsigned int: don't overflow it */
2554 if ((unsigned int)maxpages
== 0)
2555 maxpages
= UINT_MAX
;
2557 p
->highest_bit
= maxpages
- 1;
2561 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
2562 if (swapfilepages
&& maxpages
> swapfilepages
) {
2563 pr_warn("Swap area shorter than signature indicates\n");
2566 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
2568 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2574 #define SWAP_CLUSTER_INFO_COLS \
2575 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2576 #define SWAP_CLUSTER_SPACE_COLS \
2577 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2578 #define SWAP_CLUSTER_COLS \
2579 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2581 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
2582 union swap_header
*swap_header
,
2583 unsigned char *swap_map
,
2584 struct swap_cluster_info
*cluster_info
,
2585 unsigned long maxpages
,
2589 unsigned int nr_good_pages
;
2591 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2592 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
2593 unsigned long i
, idx
;
2595 nr_good_pages
= maxpages
- 1; /* omit header page */
2597 cluster_list_init(&p
->free_clusters
);
2598 cluster_list_init(&p
->discard_clusters
);
2600 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
2601 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
2602 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
2604 if (page_nr
< maxpages
) {
2605 swap_map
[page_nr
] = SWAP_MAP_BAD
;
2608 * Haven't marked the cluster free yet, no list
2609 * operation involved
2611 inc_cluster_info_page(p
, cluster_info
, page_nr
);
2615 /* Haven't marked the cluster free yet, no list operation involved */
2616 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
2617 inc_cluster_info_page(p
, cluster_info
, i
);
2619 if (nr_good_pages
) {
2620 swap_map
[0] = SWAP_MAP_BAD
;
2622 * Not mark the cluster free yet, no list
2623 * operation involved
2625 inc_cluster_info_page(p
, cluster_info
, 0);
2627 p
->pages
= nr_good_pages
;
2628 nr_extents
= setup_swap_extents(p
, span
);
2631 nr_good_pages
= p
->pages
;
2633 if (!nr_good_pages
) {
2634 pr_warn("Empty swap-file\n");
2643 * Reduce false cache line sharing between cluster_info and
2644 * sharing same address space.
2646 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
2647 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
2648 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
2649 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
2650 if (idx
>= nr_clusters
)
2652 if (cluster_count(&cluster_info
[idx
]))
2654 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
2655 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
2663 * Helper to sys_swapon determining if a given swap
2664 * backing device queue supports DISCARD operations.
2666 static bool swap_discardable(struct swap_info_struct
*si
)
2668 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
2670 if (!q
|| !blk_queue_discard(q
))
2676 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
2678 struct swap_info_struct
*p
;
2679 struct filename
*name
;
2680 struct file
*swap_file
= NULL
;
2681 struct address_space
*mapping
;
2684 union swap_header
*swap_header
;
2687 unsigned long maxpages
;
2688 unsigned char *swap_map
= NULL
;
2689 struct swap_cluster_info
*cluster_info
= NULL
;
2690 unsigned long *frontswap_map
= NULL
;
2691 struct page
*page
= NULL
;
2692 struct inode
*inode
= NULL
;
2694 if (swap_flags
& ~SWAP_FLAGS_VALID
)
2697 if (!capable(CAP_SYS_ADMIN
))
2700 p
= alloc_swap_info();
2704 INIT_WORK(&p
->discard_work
, swap_discard_work
);
2706 name
= getname(specialfile
);
2708 error
= PTR_ERR(name
);
2712 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
2713 if (IS_ERR(swap_file
)) {
2714 error
= PTR_ERR(swap_file
);
2719 p
->swap_file
= swap_file
;
2720 mapping
= swap_file
->f_mapping
;
2721 inode
= mapping
->host
;
2723 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
2724 error
= claim_swapfile(p
, inode
);
2725 if (unlikely(error
))
2729 * Read the swap header.
2731 if (!mapping
->a_ops
->readpage
) {
2735 page
= read_mapping_page(mapping
, 0, swap_file
);
2737 error
= PTR_ERR(page
);
2740 swap_header
= kmap(page
);
2742 maxpages
= read_swap_header(p
, swap_header
, inode
);
2743 if (unlikely(!maxpages
)) {
2748 /* OK, set up the swap map and apply the bad block list */
2749 swap_map
= vzalloc(maxpages
);
2755 if (bdi_cap_stable_pages_required(inode_to_bdi(inode
)))
2756 p
->flags
|= SWP_STABLE_WRITES
;
2758 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
2760 unsigned long ci
, nr_cluster
;
2762 p
->flags
|= SWP_SOLIDSTATE
;
2764 * select a random position to start with to help wear leveling
2767 p
->cluster_next
= 1 + (prandom_u32() % p
->highest_bit
);
2768 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2770 cluster_info
= vzalloc(nr_cluster
* sizeof(*cluster_info
));
2771 if (!cluster_info
) {
2776 for (ci
= 0; ci
< nr_cluster
; ci
++)
2777 spin_lock_init(&((cluster_info
+ ci
)->lock
));
2779 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
2780 if (!p
->percpu_cluster
) {
2784 for_each_possible_cpu(cpu
) {
2785 struct percpu_cluster
*cluster
;
2786 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
2787 cluster_set_null(&cluster
->index
);
2791 error
= swap_cgroup_swapon(p
->type
, maxpages
);
2795 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
2796 cluster_info
, maxpages
, &span
);
2797 if (unlikely(nr_extents
< 0)) {
2801 /* frontswap enabled? set up bit-per-page map for frontswap */
2802 if (IS_ENABLED(CONFIG_FRONTSWAP
))
2803 frontswap_map
= vzalloc(BITS_TO_LONGS(maxpages
) * sizeof(long));
2805 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
2807 * When discard is enabled for swap with no particular
2808 * policy flagged, we set all swap discard flags here in
2809 * order to sustain backward compatibility with older
2810 * swapon(8) releases.
2812 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
2816 * By flagging sys_swapon, a sysadmin can tell us to
2817 * either do single-time area discards only, or to just
2818 * perform discards for released swap page-clusters.
2819 * Now it's time to adjust the p->flags accordingly.
2821 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
2822 p
->flags
&= ~SWP_PAGE_DISCARD
;
2823 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
2824 p
->flags
&= ~SWP_AREA_DISCARD
;
2826 /* issue a swapon-time discard if it's still required */
2827 if (p
->flags
& SWP_AREA_DISCARD
) {
2828 int err
= discard_swap(p
);
2830 pr_err("swapon: discard_swap(%p): %d\n",
2835 error
= init_swap_address_space(p
->type
, maxpages
);
2839 mutex_lock(&swapon_mutex
);
2841 if (swap_flags
& SWAP_FLAG_PREFER
)
2843 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
2844 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
2846 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2847 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
2848 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
2849 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
2850 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
2851 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
2852 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
2853 (frontswap_map
) ? "FS" : "");
2855 mutex_unlock(&swapon_mutex
);
2856 atomic_inc(&proc_poll_event
);
2857 wake_up_interruptible(&proc_poll_wait
);
2859 if (S_ISREG(inode
->i_mode
))
2860 inode
->i_flags
|= S_SWAPFILE
;
2864 free_percpu(p
->percpu_cluster
);
2865 p
->percpu_cluster
= NULL
;
2866 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
2867 set_blocksize(p
->bdev
, p
->old_block_size
);
2868 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2870 destroy_swap_extents(p
);
2871 swap_cgroup_swapoff(p
->type
);
2872 spin_lock(&swap_lock
);
2873 p
->swap_file
= NULL
;
2875 spin_unlock(&swap_lock
);
2877 vfree(cluster_info
);
2879 if (inode
&& S_ISREG(inode
->i_mode
)) {
2880 inode_unlock(inode
);
2883 filp_close(swap_file
, NULL
);
2886 if (page
&& !IS_ERR(page
)) {
2892 if (inode
&& S_ISREG(inode
->i_mode
))
2893 inode_unlock(inode
);
2895 enable_swap_slots_cache();
2899 void si_swapinfo(struct sysinfo
*val
)
2902 unsigned long nr_to_be_unused
= 0;
2904 spin_lock(&swap_lock
);
2905 for (type
= 0; type
< nr_swapfiles
; type
++) {
2906 struct swap_info_struct
*si
= swap_info
[type
];
2908 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
2909 nr_to_be_unused
+= si
->inuse_pages
;
2911 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
2912 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
2913 spin_unlock(&swap_lock
);
2917 * Verify that a swap entry is valid and increment its swap map count.
2919 * Returns error code in following case.
2921 * - swp_entry is invalid -> EINVAL
2922 * - swp_entry is migration entry -> EINVAL
2923 * - swap-cache reference is requested but there is already one. -> EEXIST
2924 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2925 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2927 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
2929 struct swap_info_struct
*p
;
2930 struct swap_cluster_info
*ci
;
2931 unsigned long offset
, type
;
2932 unsigned char count
;
2933 unsigned char has_cache
;
2936 if (non_swap_entry(entry
))
2939 type
= swp_type(entry
);
2940 if (type
>= nr_swapfiles
)
2942 p
= swap_info
[type
];
2943 offset
= swp_offset(entry
);
2944 if (unlikely(offset
>= p
->max
))
2947 ci
= lock_cluster_or_swap_info(p
, offset
);
2949 count
= p
->swap_map
[offset
];
2952 * swapin_readahead() doesn't check if a swap entry is valid, so the
2953 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2955 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
2960 has_cache
= count
& SWAP_HAS_CACHE
;
2961 count
&= ~SWAP_HAS_CACHE
;
2964 if (usage
== SWAP_HAS_CACHE
) {
2966 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2967 if (!has_cache
&& count
)
2968 has_cache
= SWAP_HAS_CACHE
;
2969 else if (has_cache
) /* someone else added cache */
2971 else /* no users remaining */
2974 } else if (count
|| has_cache
) {
2976 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
2978 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
2980 else if (swap_count_continued(p
, offset
, count
))
2981 count
= COUNT_CONTINUED
;
2985 err
= -ENOENT
; /* unused swap entry */
2987 p
->swap_map
[offset
] = count
| has_cache
;
2990 unlock_cluster_or_swap_info(p
, ci
);
2995 pr_err("swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
3000 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3001 * (in which case its reference count is never incremented).
3003 void swap_shmem_alloc(swp_entry_t entry
)
3005 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
3009 * Increase reference count of swap entry by 1.
3010 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3011 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3012 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3013 * might occur if a page table entry has got corrupted.
3015 int swap_duplicate(swp_entry_t entry
)
3019 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
3020 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
3025 * @entry: swap entry for which we allocate swap cache.
3027 * Called when allocating swap cache for existing swap entry,
3028 * This can return error codes. Returns 0 at success.
3029 * -EBUSY means there is a swap cache.
3030 * Note: return code is different from swap_duplicate().
3032 int swapcache_prepare(swp_entry_t entry
)
3034 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
3037 struct swap_info_struct
*page_swap_info(struct page
*page
)
3039 swp_entry_t swap
= { .val
= page_private(page
) };
3040 return swap_info
[swp_type(swap
)];
3044 * out-of-line __page_file_ methods to avoid include hell.
3046 struct address_space
*__page_file_mapping(struct page
*page
)
3048 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
3049 return page_swap_info(page
)->swap_file
->f_mapping
;
3051 EXPORT_SYMBOL_GPL(__page_file_mapping
);
3053 pgoff_t
__page_file_index(struct page
*page
)
3055 swp_entry_t swap
= { .val
= page_private(page
) };
3056 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
3057 return swp_offset(swap
);
3059 EXPORT_SYMBOL_GPL(__page_file_index
);
3062 * add_swap_count_continuation - called when a swap count is duplicated
3063 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3064 * page of the original vmalloc'ed swap_map, to hold the continuation count
3065 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3066 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3068 * These continuation pages are seldom referenced: the common paths all work
3069 * on the original swap_map, only referring to a continuation page when the
3070 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3072 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3073 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3074 * can be called after dropping locks.
3076 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
3078 struct swap_info_struct
*si
;
3079 struct swap_cluster_info
*ci
;
3082 struct page
*list_page
;
3084 unsigned char count
;
3087 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3088 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3090 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
3092 si
= swap_info_get(entry
);
3095 * An acceptable race has occurred since the failing
3096 * __swap_duplicate(): the swap entry has been freed,
3097 * perhaps even the whole swap_map cleared for swapoff.
3102 offset
= swp_offset(entry
);
3104 ci
= lock_cluster(si
, offset
);
3106 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
3108 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
3110 * The higher the swap count, the more likely it is that tasks
3111 * will race to add swap count continuation: we need to avoid
3112 * over-provisioning.
3119 spin_unlock(&si
->lock
);
3124 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3125 * no architecture is using highmem pages for kernel page tables: so it
3126 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3128 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3129 offset
&= ~PAGE_MASK
;
3132 * Page allocation does not initialize the page's lru field,
3133 * but it does always reset its private field.
3135 if (!page_private(head
)) {
3136 BUG_ON(count
& COUNT_CONTINUED
);
3137 INIT_LIST_HEAD(&head
->lru
);
3138 set_page_private(head
, SWP_CONTINUED
);
3139 si
->flags
|= SWP_CONTINUED
;
3142 list_for_each_entry(list_page
, &head
->lru
, lru
) {
3146 * If the previous map said no continuation, but we've found
3147 * a continuation page, free our allocation and use this one.
3149 if (!(count
& COUNT_CONTINUED
))
3152 map
= kmap_atomic(list_page
) + offset
;
3157 * If this continuation count now has some space in it,
3158 * free our allocation and use this one.
3160 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
3164 list_add_tail(&page
->lru
, &head
->lru
);
3165 page
= NULL
; /* now it's attached, don't free it */
3168 spin_unlock(&si
->lock
);
3176 * swap_count_continued - when the original swap_map count is incremented
3177 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3178 * into, carry if so, or else fail until a new continuation page is allocated;
3179 * when the original swap_map count is decremented from 0 with continuation,
3180 * borrow from the continuation and report whether it still holds more.
3181 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3184 static bool swap_count_continued(struct swap_info_struct
*si
,
3185 pgoff_t offset
, unsigned char count
)
3191 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3192 if (page_private(head
) != SWP_CONTINUED
) {
3193 BUG_ON(count
& COUNT_CONTINUED
);
3194 return false; /* need to add count continuation */
3197 offset
&= ~PAGE_MASK
;
3198 page
= list_entry(head
->lru
.next
, struct page
, lru
);
3199 map
= kmap_atomic(page
) + offset
;
3201 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3202 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3204 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3206 * Think of how you add 1 to 999
3208 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3210 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3211 BUG_ON(page
== head
);
3212 map
= kmap_atomic(page
) + offset
;
3214 if (*map
== SWAP_CONT_MAX
) {
3216 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3218 return false; /* add count continuation */
3219 map
= kmap_atomic(page
) + offset
;
3220 init_map
: *map
= 0; /* we didn't zero the page */
3224 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3225 while (page
!= head
) {
3226 map
= kmap_atomic(page
) + offset
;
3227 *map
= COUNT_CONTINUED
;
3229 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3231 return true; /* incremented */
3233 } else { /* decrementing */
3235 * Think of how you subtract 1 from 1000
3237 BUG_ON(count
!= COUNT_CONTINUED
);
3238 while (*map
== COUNT_CONTINUED
) {
3240 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3241 BUG_ON(page
== head
);
3242 map
= kmap_atomic(page
) + offset
;
3249 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3250 while (page
!= head
) {
3251 map
= kmap_atomic(page
) + offset
;
3252 *map
= SWAP_CONT_MAX
| count
;
3253 count
= COUNT_CONTINUED
;
3255 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3257 return count
== COUNT_CONTINUED
;
3262 * free_swap_count_continuations - swapoff free all the continuation pages
3263 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3265 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3269 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3271 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3272 if (page_private(head
)) {
3273 struct page
*page
, *next
;
3275 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3276 list_del(&page
->lru
);