1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie
10 #include <linux/sched/mm.h>
11 #include <linux/sched/task.h>
12 #include <linux/hugetlb.h>
13 #include <linux/mman.h>
14 #include <linux/slab.h>
15 #include <linux/kernel_stat.h>
16 #include <linux/swap.h>
17 #include <linux/vmalloc.h>
18 #include <linux/pagemap.h>
19 #include <linux/namei.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/writeback.h>
24 #include <linux/proc_fs.h>
25 #include <linux/seq_file.h>
26 #include <linux/init.h>
27 #include <linux/ksm.h>
28 #include <linux/rmap.h>
29 #include <linux/security.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mutex.h>
32 #include <linux/capability.h>
33 #include <linux/syscalls.h>
34 #include <linux/memcontrol.h>
35 #include <linux/poll.h>
36 #include <linux/oom.h>
37 #include <linux/frontswap.h>
38 #include <linux/swapfile.h>
39 #include <linux/export.h>
40 #include <linux/swap_slots.h>
41 #include <linux/sort.h>
43 #include <asm/tlbflush.h>
44 #include <linux/swapops.h>
45 #include <linux/swap_cgroup.h>
47 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
49 static void free_swap_count_continuations(struct swap_info_struct
*);
50 static sector_t
map_swap_entry(swp_entry_t
, struct block_device
**);
52 DEFINE_SPINLOCK(swap_lock
);
53 static unsigned int nr_swapfiles
;
54 atomic_long_t nr_swap_pages
;
56 * Some modules use swappable objects and may try to swap them out under
57 * memory pressure (via the shrinker). Before doing so, they may wish to
58 * check to see if any swap space is available.
60 EXPORT_SYMBOL_GPL(nr_swap_pages
);
61 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
62 long total_swap_pages
;
63 static int least_priority
= -1;
65 static const char Bad_file
[] = "Bad swap file entry ";
66 static const char Unused_file
[] = "Unused swap file entry ";
67 static const char Bad_offset
[] = "Bad swap offset entry ";
68 static const char Unused_offset
[] = "Unused swap offset entry ";
71 * all active swap_info_structs
72 * protected with swap_lock, and ordered by priority.
74 PLIST_HEAD(swap_active_head
);
77 * all available (active, not full) swap_info_structs
78 * protected with swap_avail_lock, ordered by priority.
79 * This is used by get_swap_page() instead of swap_active_head
80 * because swap_active_head includes all swap_info_structs,
81 * but get_swap_page() doesn't need to look at full ones.
82 * This uses its own lock instead of swap_lock because when a
83 * swap_info_struct changes between not-full/full, it needs to
84 * add/remove itself to/from this list, but the swap_info_struct->lock
85 * is held and the locking order requires swap_lock to be taken
86 * before any swap_info_struct->lock.
88 static struct plist_head
*swap_avail_heads
;
89 static DEFINE_SPINLOCK(swap_avail_lock
);
91 struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
93 static DEFINE_MUTEX(swapon_mutex
);
95 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
96 /* Activity counter to indicate that a swapon or swapoff has occurred */
97 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
99 atomic_t nr_rotate_swap
= ATOMIC_INIT(0);
101 static struct swap_info_struct
*swap_type_to_swap_info(int type
)
103 if (type
>= READ_ONCE(nr_swapfiles
))
106 smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */
107 return READ_ONCE(swap_info
[type
]);
110 static inline unsigned char swap_count(unsigned char ent
)
112 return ent
& ~SWAP_HAS_CACHE
; /* may include COUNT_CONTINUED flag */
115 /* Reclaim the swap entry anyway if possible */
116 #define TTRS_ANYWAY 0x1
118 * Reclaim the swap entry if there are no more mappings of the
121 #define TTRS_UNMAPPED 0x2
122 /* Reclaim the swap entry if swap is getting full*/
123 #define TTRS_FULL 0x4
125 /* returns 1 if swap entry is freed */
126 static int __try_to_reclaim_swap(struct swap_info_struct
*si
,
127 unsigned long offset
, unsigned long flags
)
129 swp_entry_t entry
= swp_entry(si
->type
, offset
);
133 page
= find_get_page(swap_address_space(entry
), offset
);
137 * When this function is called from scan_swap_map_slots() and it's
138 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
139 * here. We have to use trylock for avoiding deadlock. This is a special
140 * case and you should use try_to_free_swap() with explicit lock_page()
141 * in usual operations.
143 if (trylock_page(page
)) {
144 if ((flags
& TTRS_ANYWAY
) ||
145 ((flags
& TTRS_UNMAPPED
) && !page_mapped(page
)) ||
146 ((flags
& TTRS_FULL
) && mem_cgroup_swap_full(page
)))
147 ret
= try_to_free_swap(page
);
154 static inline struct swap_extent
*first_se(struct swap_info_struct
*sis
)
156 struct rb_node
*rb
= rb_first(&sis
->swap_extent_root
);
157 return rb_entry(rb
, struct swap_extent
, rb_node
);
160 static inline struct swap_extent
*next_se(struct swap_extent
*se
)
162 struct rb_node
*rb
= rb_next(&se
->rb_node
);
163 return rb
? rb_entry(rb
, struct swap_extent
, rb_node
) : NULL
;
167 * swapon tell device that all the old swap contents can be discarded,
168 * to allow the swap device to optimize its wear-levelling.
170 static int discard_swap(struct swap_info_struct
*si
)
172 struct swap_extent
*se
;
173 sector_t start_block
;
177 /* Do not discard the swap header page! */
179 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
180 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
182 err
= blkdev_issue_discard(si
->bdev
, start_block
,
183 nr_blocks
, GFP_KERNEL
, 0);
189 for (se
= next_se(se
); se
; se
= next_se(se
)) {
190 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
191 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
193 err
= blkdev_issue_discard(si
->bdev
, start_block
,
194 nr_blocks
, GFP_KERNEL
, 0);
200 return err
; /* That will often be -EOPNOTSUPP */
203 static struct swap_extent
*
204 offset_to_swap_extent(struct swap_info_struct
*sis
, unsigned long offset
)
206 struct swap_extent
*se
;
209 rb
= sis
->swap_extent_root
.rb_node
;
211 se
= rb_entry(rb
, struct swap_extent
, rb_node
);
212 if (offset
< se
->start_page
)
214 else if (offset
>= se
->start_page
+ se
->nr_pages
)
219 /* It *must* be present */
224 * swap allocation tell device that a cluster of swap can now be discarded,
225 * to allow the swap device to optimize its wear-levelling.
227 static void discard_swap_cluster(struct swap_info_struct
*si
,
228 pgoff_t start_page
, pgoff_t nr_pages
)
230 struct swap_extent
*se
= offset_to_swap_extent(si
, start_page
);
233 pgoff_t offset
= start_page
- se
->start_page
;
234 sector_t start_block
= se
->start_block
+ offset
;
235 sector_t nr_blocks
= se
->nr_pages
- offset
;
237 if (nr_blocks
> nr_pages
)
238 nr_blocks
= nr_pages
;
239 start_page
+= nr_blocks
;
240 nr_pages
-= nr_blocks
;
242 start_block
<<= PAGE_SHIFT
- 9;
243 nr_blocks
<<= PAGE_SHIFT
- 9;
244 if (blkdev_issue_discard(si
->bdev
, start_block
,
245 nr_blocks
, GFP_NOIO
, 0))
252 #ifdef CONFIG_THP_SWAP
253 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
255 #define swap_entry_size(size) (size)
257 #define SWAPFILE_CLUSTER 256
260 * Define swap_entry_size() as constant to let compiler to optimize
261 * out some code if !CONFIG_THP_SWAP
263 #define swap_entry_size(size) 1
265 #define LATENCY_LIMIT 256
267 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
273 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
278 static inline void cluster_set_count(struct swap_cluster_info
*info
,
284 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
285 unsigned int c
, unsigned int f
)
291 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
296 static inline void cluster_set_next(struct swap_cluster_info
*info
,
302 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
303 unsigned int n
, unsigned int f
)
309 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
311 return info
->flags
& CLUSTER_FLAG_FREE
;
314 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
316 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
319 static inline void cluster_set_null(struct swap_cluster_info
*info
)
321 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
325 static inline bool cluster_is_huge(struct swap_cluster_info
*info
)
327 if (IS_ENABLED(CONFIG_THP_SWAP
))
328 return info
->flags
& CLUSTER_FLAG_HUGE
;
332 static inline void cluster_clear_huge(struct swap_cluster_info
*info
)
334 info
->flags
&= ~CLUSTER_FLAG_HUGE
;
337 static inline struct swap_cluster_info
*lock_cluster(struct swap_info_struct
*si
,
338 unsigned long offset
)
340 struct swap_cluster_info
*ci
;
342 ci
= si
->cluster_info
;
344 ci
+= offset
/ SWAPFILE_CLUSTER
;
345 spin_lock(&ci
->lock
);
350 static inline void unlock_cluster(struct swap_cluster_info
*ci
)
353 spin_unlock(&ci
->lock
);
357 * Determine the locking method in use for this device. Return
358 * swap_cluster_info if SSD-style cluster-based locking is in place.
360 static inline struct swap_cluster_info
*lock_cluster_or_swap_info(
361 struct swap_info_struct
*si
, unsigned long offset
)
363 struct swap_cluster_info
*ci
;
365 /* Try to use fine-grained SSD-style locking if available: */
366 ci
= lock_cluster(si
, offset
);
367 /* Otherwise, fall back to traditional, coarse locking: */
369 spin_lock(&si
->lock
);
374 static inline void unlock_cluster_or_swap_info(struct swap_info_struct
*si
,
375 struct swap_cluster_info
*ci
)
380 spin_unlock(&si
->lock
);
383 static inline bool cluster_list_empty(struct swap_cluster_list
*list
)
385 return cluster_is_null(&list
->head
);
388 static inline unsigned int cluster_list_first(struct swap_cluster_list
*list
)
390 return cluster_next(&list
->head
);
393 static void cluster_list_init(struct swap_cluster_list
*list
)
395 cluster_set_null(&list
->head
);
396 cluster_set_null(&list
->tail
);
399 static void cluster_list_add_tail(struct swap_cluster_list
*list
,
400 struct swap_cluster_info
*ci
,
403 if (cluster_list_empty(list
)) {
404 cluster_set_next_flag(&list
->head
, idx
, 0);
405 cluster_set_next_flag(&list
->tail
, idx
, 0);
407 struct swap_cluster_info
*ci_tail
;
408 unsigned int tail
= cluster_next(&list
->tail
);
411 * Nested cluster lock, but both cluster locks are
412 * only acquired when we held swap_info_struct->lock
415 spin_lock_nested(&ci_tail
->lock
, SINGLE_DEPTH_NESTING
);
416 cluster_set_next(ci_tail
, idx
);
417 spin_unlock(&ci_tail
->lock
);
418 cluster_set_next_flag(&list
->tail
, idx
, 0);
422 static unsigned int cluster_list_del_first(struct swap_cluster_list
*list
,
423 struct swap_cluster_info
*ci
)
427 idx
= cluster_next(&list
->head
);
428 if (cluster_next(&list
->tail
) == idx
) {
429 cluster_set_null(&list
->head
);
430 cluster_set_null(&list
->tail
);
432 cluster_set_next_flag(&list
->head
,
433 cluster_next(&ci
[idx
]), 0);
438 /* Add a cluster to discard list and schedule it to do discard */
439 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
443 * If scan_swap_map() can't find a free cluster, it will check
444 * si->swap_map directly. To make sure the discarding cluster isn't
445 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
446 * will be cleared after discard
448 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
449 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
451 cluster_list_add_tail(&si
->discard_clusters
, si
->cluster_info
, idx
);
453 schedule_work(&si
->discard_work
);
456 static void __free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
458 struct swap_cluster_info
*ci
= si
->cluster_info
;
460 cluster_set_flag(ci
+ idx
, CLUSTER_FLAG_FREE
);
461 cluster_list_add_tail(&si
->free_clusters
, ci
, idx
);
465 * Doing discard actually. After a cluster discard is finished, the cluster
466 * will be added to free cluster list. caller should hold si->lock.
468 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
470 struct swap_cluster_info
*info
, *ci
;
473 info
= si
->cluster_info
;
475 while (!cluster_list_empty(&si
->discard_clusters
)) {
476 idx
= cluster_list_del_first(&si
->discard_clusters
, info
);
477 spin_unlock(&si
->lock
);
479 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
482 spin_lock(&si
->lock
);
483 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
484 __free_cluster(si
, idx
);
485 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
486 0, SWAPFILE_CLUSTER
);
491 static void swap_discard_work(struct work_struct
*work
)
493 struct swap_info_struct
*si
;
495 si
= container_of(work
, struct swap_info_struct
, discard_work
);
497 spin_lock(&si
->lock
);
498 swap_do_scheduled_discard(si
);
499 spin_unlock(&si
->lock
);
502 static void alloc_cluster(struct swap_info_struct
*si
, unsigned long idx
)
504 struct swap_cluster_info
*ci
= si
->cluster_info
;
506 VM_BUG_ON(cluster_list_first(&si
->free_clusters
) != idx
);
507 cluster_list_del_first(&si
->free_clusters
, ci
);
508 cluster_set_count_flag(ci
+ idx
, 0, 0);
511 static void free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
513 struct swap_cluster_info
*ci
= si
->cluster_info
+ idx
;
515 VM_BUG_ON(cluster_count(ci
) != 0);
517 * If the swap is discardable, prepare discard the cluster
518 * instead of free it immediately. The cluster will be freed
521 if ((si
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
522 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
523 swap_cluster_schedule_discard(si
, idx
);
527 __free_cluster(si
, idx
);
531 * The cluster corresponding to page_nr will be used. The cluster will be
532 * removed from free cluster list and its usage counter will be increased.
534 static void inc_cluster_info_page(struct swap_info_struct
*p
,
535 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
537 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
541 if (cluster_is_free(&cluster_info
[idx
]))
542 alloc_cluster(p
, idx
);
544 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
545 cluster_set_count(&cluster_info
[idx
],
546 cluster_count(&cluster_info
[idx
]) + 1);
550 * The cluster corresponding to page_nr decreases one usage. If the usage
551 * counter becomes 0, which means no page in the cluster is in using, we can
552 * optionally discard the cluster and add it to free cluster list.
554 static void dec_cluster_info_page(struct swap_info_struct
*p
,
555 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
557 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
562 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
563 cluster_set_count(&cluster_info
[idx
],
564 cluster_count(&cluster_info
[idx
]) - 1);
566 if (cluster_count(&cluster_info
[idx
]) == 0)
567 free_cluster(p
, idx
);
571 * It's possible scan_swap_map() uses a free cluster in the middle of free
572 * cluster list. Avoiding such abuse to avoid list corruption.
575 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
576 unsigned long offset
)
578 struct percpu_cluster
*percpu_cluster
;
581 offset
/= SWAPFILE_CLUSTER
;
582 conflict
= !cluster_list_empty(&si
->free_clusters
) &&
583 offset
!= cluster_list_first(&si
->free_clusters
) &&
584 cluster_is_free(&si
->cluster_info
[offset
]);
589 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
590 cluster_set_null(&percpu_cluster
->index
);
595 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
596 * might involve allocating a new cluster for current CPU too.
598 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
599 unsigned long *offset
, unsigned long *scan_base
)
601 struct percpu_cluster
*cluster
;
602 struct swap_cluster_info
*ci
;
603 unsigned long tmp
, max
;
606 cluster
= this_cpu_ptr(si
->percpu_cluster
);
607 if (cluster_is_null(&cluster
->index
)) {
608 if (!cluster_list_empty(&si
->free_clusters
)) {
609 cluster
->index
= si
->free_clusters
.head
;
610 cluster
->next
= cluster_next(&cluster
->index
) *
612 } else if (!cluster_list_empty(&si
->discard_clusters
)) {
614 * we don't have free cluster but have some clusters in
615 * discarding, do discard now and reclaim them, then
616 * reread cluster_next_cpu since we dropped si->lock
618 swap_do_scheduled_discard(si
);
619 *scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
620 *offset
= *scan_base
;
627 * Other CPUs can use our cluster if they can't find a free cluster,
628 * check if there is still free entry in the cluster
631 max
= min_t(unsigned long, si
->max
,
632 (cluster_next(&cluster
->index
) + 1) * SWAPFILE_CLUSTER
);
634 ci
= lock_cluster(si
, tmp
);
636 if (!si
->swap_map
[tmp
])
643 cluster_set_null(&cluster
->index
);
646 cluster
->next
= tmp
+ 1;
652 static void __del_from_avail_list(struct swap_info_struct
*p
)
657 plist_del(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
660 static void del_from_avail_list(struct swap_info_struct
*p
)
662 spin_lock(&swap_avail_lock
);
663 __del_from_avail_list(p
);
664 spin_unlock(&swap_avail_lock
);
667 static void swap_range_alloc(struct swap_info_struct
*si
, unsigned long offset
,
668 unsigned int nr_entries
)
670 unsigned int end
= offset
+ nr_entries
- 1;
672 if (offset
== si
->lowest_bit
)
673 si
->lowest_bit
+= nr_entries
;
674 if (end
== si
->highest_bit
)
675 WRITE_ONCE(si
->highest_bit
, si
->highest_bit
- nr_entries
);
676 si
->inuse_pages
+= nr_entries
;
677 if (si
->inuse_pages
== si
->pages
) {
678 si
->lowest_bit
= si
->max
;
680 del_from_avail_list(si
);
684 static void add_to_avail_list(struct swap_info_struct
*p
)
688 spin_lock(&swap_avail_lock
);
690 WARN_ON(!plist_node_empty(&p
->avail_lists
[nid
]));
691 plist_add(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
693 spin_unlock(&swap_avail_lock
);
696 static void swap_range_free(struct swap_info_struct
*si
, unsigned long offset
,
697 unsigned int nr_entries
)
699 unsigned long begin
= offset
;
700 unsigned long end
= offset
+ nr_entries
- 1;
701 void (*swap_slot_free_notify
)(struct block_device
*, unsigned long);
703 if (offset
< si
->lowest_bit
)
704 si
->lowest_bit
= offset
;
705 if (end
> si
->highest_bit
) {
706 bool was_full
= !si
->highest_bit
;
708 WRITE_ONCE(si
->highest_bit
, end
);
709 if (was_full
&& (si
->flags
& SWP_WRITEOK
))
710 add_to_avail_list(si
);
712 atomic_long_add(nr_entries
, &nr_swap_pages
);
713 si
->inuse_pages
-= nr_entries
;
714 if (si
->flags
& SWP_BLKDEV
)
715 swap_slot_free_notify
=
716 si
->bdev
->bd_disk
->fops
->swap_slot_free_notify
;
718 swap_slot_free_notify
= NULL
;
719 while (offset
<= end
) {
720 frontswap_invalidate_page(si
->type
, offset
);
721 if (swap_slot_free_notify
)
722 swap_slot_free_notify(si
->bdev
, offset
);
725 clear_shadow_from_swap_cache(si
->type
, begin
, end
);
728 static void set_cluster_next(struct swap_info_struct
*si
, unsigned long next
)
732 if (!(si
->flags
& SWP_SOLIDSTATE
)) {
733 si
->cluster_next
= next
;
737 prev
= this_cpu_read(*si
->cluster_next_cpu
);
739 * Cross the swap address space size aligned trunk, choose
740 * another trunk randomly to avoid lock contention on swap
741 * address space if possible.
743 if ((prev
>> SWAP_ADDRESS_SPACE_SHIFT
) !=
744 (next
>> SWAP_ADDRESS_SPACE_SHIFT
)) {
745 /* No free swap slots available */
746 if (si
->highest_bit
<= si
->lowest_bit
)
748 next
= si
->lowest_bit
+
749 prandom_u32_max(si
->highest_bit
- si
->lowest_bit
+ 1);
750 next
= ALIGN_DOWN(next
, SWAP_ADDRESS_SPACE_PAGES
);
751 next
= max_t(unsigned int, next
, si
->lowest_bit
);
753 this_cpu_write(*si
->cluster_next_cpu
, next
);
756 static int scan_swap_map_slots(struct swap_info_struct
*si
,
757 unsigned char usage
, int nr
,
760 struct swap_cluster_info
*ci
;
761 unsigned long offset
;
762 unsigned long scan_base
;
763 unsigned long last_in_cluster
= 0;
764 int latency_ration
= LATENCY_LIMIT
;
766 bool scanned_many
= false;
769 * We try to cluster swap pages by allocating them sequentially
770 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
771 * way, however, we resort to first-free allocation, starting
772 * a new cluster. This prevents us from scattering swap pages
773 * all over the entire swap partition, so that we reduce
774 * overall disk seek times between swap pages. -- sct
775 * But we do now try to find an empty cluster. -Andrea
776 * And we let swap pages go all over an SSD partition. Hugh
779 si
->flags
+= SWP_SCANNING
;
781 * Use percpu scan base for SSD to reduce lock contention on
782 * cluster and swap cache. For HDD, sequential access is more
785 if (si
->flags
& SWP_SOLIDSTATE
)
786 scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
788 scan_base
= si
->cluster_next
;
792 if (si
->cluster_info
) {
793 if (!scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
795 } else if (unlikely(!si
->cluster_nr
--)) {
796 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
797 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
801 spin_unlock(&si
->lock
);
804 * If seek is expensive, start searching for new cluster from
805 * start of partition, to minimize the span of allocated swap.
806 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
807 * case, just handled by scan_swap_map_try_ssd_cluster() above.
809 scan_base
= offset
= si
->lowest_bit
;
810 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
812 /* Locate the first empty (unaligned) cluster */
813 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
814 if (si
->swap_map
[offset
])
815 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
816 else if (offset
== last_in_cluster
) {
817 spin_lock(&si
->lock
);
818 offset
-= SWAPFILE_CLUSTER
- 1;
819 si
->cluster_next
= offset
;
820 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
823 if (unlikely(--latency_ration
< 0)) {
825 latency_ration
= LATENCY_LIMIT
;
830 spin_lock(&si
->lock
);
831 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
835 if (si
->cluster_info
) {
836 while (scan_swap_map_ssd_cluster_conflict(si
, offset
)) {
837 /* take a break if we already got some slots */
840 if (!scan_swap_map_try_ssd_cluster(si
, &offset
,
845 if (!(si
->flags
& SWP_WRITEOK
))
847 if (!si
->highest_bit
)
849 if (offset
> si
->highest_bit
)
850 scan_base
= offset
= si
->lowest_bit
;
852 ci
= lock_cluster(si
, offset
);
853 /* reuse swap entry of cache-only swap if not busy. */
854 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
857 spin_unlock(&si
->lock
);
858 swap_was_freed
= __try_to_reclaim_swap(si
, offset
, TTRS_ANYWAY
);
859 spin_lock(&si
->lock
);
860 /* entry was freed successfully, try to use this again */
863 goto scan
; /* check next one */
866 if (si
->swap_map
[offset
]) {
873 WRITE_ONCE(si
->swap_map
[offset
], usage
);
874 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
877 swap_range_alloc(si
, offset
, 1);
878 slots
[n_ret
++] = swp_entry(si
->type
, offset
);
880 /* got enough slots or reach max slots? */
881 if ((n_ret
== nr
) || (offset
>= si
->highest_bit
))
884 /* search for next available slot */
886 /* time to take a break? */
887 if (unlikely(--latency_ration
< 0)) {
890 spin_unlock(&si
->lock
);
892 spin_lock(&si
->lock
);
893 latency_ration
= LATENCY_LIMIT
;
896 /* try to get more slots in cluster */
897 if (si
->cluster_info
) {
898 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
900 } else if (si
->cluster_nr
&& !si
->swap_map
[++offset
]) {
901 /* non-ssd case, still more slots in cluster? */
907 * Even if there's no free clusters available (fragmented),
908 * try to scan a little more quickly with lock held unless we
909 * have scanned too many slots already.
912 unsigned long scan_limit
;
914 if (offset
< scan_base
)
915 scan_limit
= scan_base
;
917 scan_limit
= si
->highest_bit
;
918 for (; offset
<= scan_limit
&& --latency_ration
> 0;
920 if (!si
->swap_map
[offset
])
926 set_cluster_next(si
, offset
+ 1);
927 si
->flags
-= SWP_SCANNING
;
931 spin_unlock(&si
->lock
);
932 while (++offset
<= READ_ONCE(si
->highest_bit
)) {
933 if (data_race(!si
->swap_map
[offset
])) {
934 spin_lock(&si
->lock
);
937 if (vm_swap_full() &&
938 READ_ONCE(si
->swap_map
[offset
]) == SWAP_HAS_CACHE
) {
939 spin_lock(&si
->lock
);
942 if (unlikely(--latency_ration
< 0)) {
944 latency_ration
= LATENCY_LIMIT
;
948 offset
= si
->lowest_bit
;
949 while (offset
< scan_base
) {
950 if (data_race(!si
->swap_map
[offset
])) {
951 spin_lock(&si
->lock
);
954 if (vm_swap_full() &&
955 READ_ONCE(si
->swap_map
[offset
]) == SWAP_HAS_CACHE
) {
956 spin_lock(&si
->lock
);
959 if (unlikely(--latency_ration
< 0)) {
961 latency_ration
= LATENCY_LIMIT
;
966 spin_lock(&si
->lock
);
969 si
->flags
-= SWP_SCANNING
;
973 static int swap_alloc_cluster(struct swap_info_struct
*si
, swp_entry_t
*slot
)
976 struct swap_cluster_info
*ci
;
977 unsigned long offset
, i
;
981 * Should not even be attempting cluster allocations when huge
982 * page swap is disabled. Warn and fail the allocation.
984 if (!IS_ENABLED(CONFIG_THP_SWAP
)) {
989 if (cluster_list_empty(&si
->free_clusters
))
992 idx
= cluster_list_first(&si
->free_clusters
);
993 offset
= idx
* SWAPFILE_CLUSTER
;
994 ci
= lock_cluster(si
, offset
);
995 alloc_cluster(si
, idx
);
996 cluster_set_count_flag(ci
, SWAPFILE_CLUSTER
, CLUSTER_FLAG_HUGE
);
998 map
= si
->swap_map
+ offset
;
999 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++)
1000 map
[i
] = SWAP_HAS_CACHE
;
1002 swap_range_alloc(si
, offset
, SWAPFILE_CLUSTER
);
1003 *slot
= swp_entry(si
->type
, offset
);
1008 static void swap_free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
1010 unsigned long offset
= idx
* SWAPFILE_CLUSTER
;
1011 struct swap_cluster_info
*ci
;
1013 ci
= lock_cluster(si
, offset
);
1014 memset(si
->swap_map
+ offset
, 0, SWAPFILE_CLUSTER
);
1015 cluster_set_count_flag(ci
, 0, 0);
1016 free_cluster(si
, idx
);
1018 swap_range_free(si
, offset
, SWAPFILE_CLUSTER
);
1021 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
1022 unsigned char usage
)
1027 n_ret
= scan_swap_map_slots(si
, usage
, 1, &entry
);
1030 return swp_offset(entry
);
1036 int get_swap_pages(int n_goal
, swp_entry_t swp_entries
[], int entry_size
)
1038 unsigned long size
= swap_entry_size(entry_size
);
1039 struct swap_info_struct
*si
, *next
;
1044 /* Only single cluster request supported */
1045 WARN_ON_ONCE(n_goal
> 1 && size
== SWAPFILE_CLUSTER
);
1047 avail_pgs
= atomic_long_read(&nr_swap_pages
) / size
;
1051 n_goal
= min3((long)n_goal
, (long)SWAP_BATCH
, avail_pgs
);
1053 atomic_long_sub(n_goal
* size
, &nr_swap_pages
);
1055 spin_lock(&swap_avail_lock
);
1058 node
= numa_node_id();
1059 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[node
], avail_lists
[node
]) {
1060 /* requeue si to after same-priority siblings */
1061 plist_requeue(&si
->avail_lists
[node
], &swap_avail_heads
[node
]);
1062 spin_unlock(&swap_avail_lock
);
1063 spin_lock(&si
->lock
);
1064 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
1065 spin_lock(&swap_avail_lock
);
1066 if (plist_node_empty(&si
->avail_lists
[node
])) {
1067 spin_unlock(&si
->lock
);
1070 WARN(!si
->highest_bit
,
1071 "swap_info %d in list but !highest_bit\n",
1073 WARN(!(si
->flags
& SWP_WRITEOK
),
1074 "swap_info %d in list but !SWP_WRITEOK\n",
1076 __del_from_avail_list(si
);
1077 spin_unlock(&si
->lock
);
1080 if (size
== SWAPFILE_CLUSTER
) {
1081 if (!(si
->flags
& SWP_FS
))
1082 n_ret
= swap_alloc_cluster(si
, swp_entries
);
1084 n_ret
= scan_swap_map_slots(si
, SWAP_HAS_CACHE
,
1085 n_goal
, swp_entries
);
1086 spin_unlock(&si
->lock
);
1087 if (n_ret
|| size
== SWAPFILE_CLUSTER
)
1089 pr_debug("scan_swap_map of si %d failed to find offset\n",
1092 spin_lock(&swap_avail_lock
);
1095 * if we got here, it's likely that si was almost full before,
1096 * and since scan_swap_map() can drop the si->lock, multiple
1097 * callers probably all tried to get a page from the same si
1098 * and it filled up before we could get one; or, the si filled
1099 * up between us dropping swap_avail_lock and taking si->lock.
1100 * Since we dropped the swap_avail_lock, the swap_avail_head
1101 * list may have been modified; so if next is still in the
1102 * swap_avail_head list then try it, otherwise start over
1103 * if we have not gotten any slots.
1105 if (plist_node_empty(&next
->avail_lists
[node
]))
1109 spin_unlock(&swap_avail_lock
);
1113 atomic_long_add((long)(n_goal
- n_ret
) * size
,
1119 /* The only caller of this function is now suspend routine */
1120 swp_entry_t
get_swap_page_of_type(int type
)
1122 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1128 spin_lock(&si
->lock
);
1129 if (si
->flags
& SWP_WRITEOK
) {
1130 atomic_long_dec(&nr_swap_pages
);
1131 /* This is called for allocating swap entry, not cache */
1132 offset
= scan_swap_map(si
, 1);
1134 spin_unlock(&si
->lock
);
1135 return swp_entry(type
, offset
);
1137 atomic_long_inc(&nr_swap_pages
);
1139 spin_unlock(&si
->lock
);
1141 return (swp_entry_t
) {0};
1144 static struct swap_info_struct
*__swap_info_get(swp_entry_t entry
)
1146 struct swap_info_struct
*p
;
1147 unsigned long offset
;
1151 p
= swp_swap_info(entry
);
1154 if (data_race(!(p
->flags
& SWP_USED
)))
1156 offset
= swp_offset(entry
);
1157 if (offset
>= p
->max
)
1162 pr_err("swap_info_get: %s%08lx\n", Bad_offset
, entry
.val
);
1165 pr_err("swap_info_get: %s%08lx\n", Unused_file
, entry
.val
);
1168 pr_err("swap_info_get: %s%08lx\n", Bad_file
, entry
.val
);
1173 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
1175 struct swap_info_struct
*p
;
1177 p
= __swap_info_get(entry
);
1180 if (data_race(!p
->swap_map
[swp_offset(entry
)]))
1185 pr_err("swap_info_get: %s%08lx\n", Unused_offset
, entry
.val
);
1191 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
1193 struct swap_info_struct
*p
;
1195 p
= _swap_info_get(entry
);
1197 spin_lock(&p
->lock
);
1201 static struct swap_info_struct
*swap_info_get_cont(swp_entry_t entry
,
1202 struct swap_info_struct
*q
)
1204 struct swap_info_struct
*p
;
1206 p
= _swap_info_get(entry
);
1210 spin_unlock(&q
->lock
);
1212 spin_lock(&p
->lock
);
1217 static unsigned char __swap_entry_free_locked(struct swap_info_struct
*p
,
1218 unsigned long offset
,
1219 unsigned char usage
)
1221 unsigned char count
;
1222 unsigned char has_cache
;
1224 count
= p
->swap_map
[offset
];
1226 has_cache
= count
& SWAP_HAS_CACHE
;
1227 count
&= ~SWAP_HAS_CACHE
;
1229 if (usage
== SWAP_HAS_CACHE
) {
1230 VM_BUG_ON(!has_cache
);
1232 } else if (count
== SWAP_MAP_SHMEM
) {
1234 * Or we could insist on shmem.c using a special
1235 * swap_shmem_free() and free_shmem_swap_and_cache()...
1238 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
1239 if (count
== COUNT_CONTINUED
) {
1240 if (swap_count_continued(p
, offset
, count
))
1241 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
1243 count
= SWAP_MAP_MAX
;
1248 usage
= count
| has_cache
;
1250 WRITE_ONCE(p
->swap_map
[offset
], usage
);
1252 WRITE_ONCE(p
->swap_map
[offset
], SWAP_HAS_CACHE
);
1258 * Check whether swap entry is valid in the swap device. If so,
1259 * return pointer to swap_info_struct, and keep the swap entry valid
1260 * via preventing the swap device from being swapoff, until
1261 * put_swap_device() is called. Otherwise return NULL.
1263 * The entirety of the RCU read critical section must come before the
1264 * return from or after the call to synchronize_rcu() in
1265 * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is
1266 * true, the si->map, si->cluster_info, etc. must be valid in the
1269 * Notice that swapoff or swapoff+swapon can still happen before the
1270 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1271 * in put_swap_device() if there isn't any other way to prevent
1272 * swapoff, such as page lock, page table lock, etc. The caller must
1273 * be prepared for that. For example, the following situation is
1278 * ... swapoff+swapon
1279 * __read_swap_cache_async()
1280 * swapcache_prepare()
1281 * __swap_duplicate()
1283 * // verify PTE not changed
1285 * In __swap_duplicate(), the swap_map need to be checked before
1286 * changing partly because the specified swap entry may be for another
1287 * swap device which has been swapoff. And in do_swap_page(), after
1288 * the page is read from the swap device, the PTE is verified not
1289 * changed with the page table locked to check whether the swap device
1290 * has been swapoff or swapoff+swapon.
1292 struct swap_info_struct
*get_swap_device(swp_entry_t entry
)
1294 struct swap_info_struct
*si
;
1295 unsigned long offset
;
1299 si
= swp_swap_info(entry
);
1304 if (data_race(!(si
->flags
& SWP_VALID
)))
1306 offset
= swp_offset(entry
);
1307 if (offset
>= si
->max
)
1312 pr_err("%s: %s%08lx\n", __func__
, Bad_file
, entry
.val
);
1320 static unsigned char __swap_entry_free(struct swap_info_struct
*p
,
1323 struct swap_cluster_info
*ci
;
1324 unsigned long offset
= swp_offset(entry
);
1325 unsigned char usage
;
1327 ci
= lock_cluster_or_swap_info(p
, offset
);
1328 usage
= __swap_entry_free_locked(p
, offset
, 1);
1329 unlock_cluster_or_swap_info(p
, ci
);
1331 free_swap_slot(entry
);
1336 static void swap_entry_free(struct swap_info_struct
*p
, swp_entry_t entry
)
1338 struct swap_cluster_info
*ci
;
1339 unsigned long offset
= swp_offset(entry
);
1340 unsigned char count
;
1342 ci
= lock_cluster(p
, offset
);
1343 count
= p
->swap_map
[offset
];
1344 VM_BUG_ON(count
!= SWAP_HAS_CACHE
);
1345 p
->swap_map
[offset
] = 0;
1346 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
1349 mem_cgroup_uncharge_swap(entry
, 1);
1350 swap_range_free(p
, offset
, 1);
1354 * Caller has made sure that the swap device corresponding to entry
1355 * is still around or has not been recycled.
1357 void swap_free(swp_entry_t entry
)
1359 struct swap_info_struct
*p
;
1361 p
= _swap_info_get(entry
);
1363 __swap_entry_free(p
, entry
);
1367 * Called after dropping swapcache to decrease refcnt to swap entries.
1369 void put_swap_page(struct page
*page
, swp_entry_t entry
)
1371 unsigned long offset
= swp_offset(entry
);
1372 unsigned long idx
= offset
/ SWAPFILE_CLUSTER
;
1373 struct swap_cluster_info
*ci
;
1374 struct swap_info_struct
*si
;
1376 unsigned int i
, free_entries
= 0;
1378 int size
= swap_entry_size(thp_nr_pages(page
));
1380 si
= _swap_info_get(entry
);
1384 ci
= lock_cluster_or_swap_info(si
, offset
);
1385 if (size
== SWAPFILE_CLUSTER
) {
1386 VM_BUG_ON(!cluster_is_huge(ci
));
1387 map
= si
->swap_map
+ offset
;
1388 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1390 VM_BUG_ON(!(val
& SWAP_HAS_CACHE
));
1391 if (val
== SWAP_HAS_CACHE
)
1394 cluster_clear_huge(ci
);
1395 if (free_entries
== SWAPFILE_CLUSTER
) {
1396 unlock_cluster_or_swap_info(si
, ci
);
1397 spin_lock(&si
->lock
);
1398 mem_cgroup_uncharge_swap(entry
, SWAPFILE_CLUSTER
);
1399 swap_free_cluster(si
, idx
);
1400 spin_unlock(&si
->lock
);
1404 for (i
= 0; i
< size
; i
++, entry
.val
++) {
1405 if (!__swap_entry_free_locked(si
, offset
+ i
, SWAP_HAS_CACHE
)) {
1406 unlock_cluster_or_swap_info(si
, ci
);
1407 free_swap_slot(entry
);
1410 lock_cluster_or_swap_info(si
, offset
);
1413 unlock_cluster_or_swap_info(si
, ci
);
1416 #ifdef CONFIG_THP_SWAP
1417 int split_swap_cluster(swp_entry_t entry
)
1419 struct swap_info_struct
*si
;
1420 struct swap_cluster_info
*ci
;
1421 unsigned long offset
= swp_offset(entry
);
1423 si
= _swap_info_get(entry
);
1426 ci
= lock_cluster(si
, offset
);
1427 cluster_clear_huge(ci
);
1433 static int swp_entry_cmp(const void *ent1
, const void *ent2
)
1435 const swp_entry_t
*e1
= ent1
, *e2
= ent2
;
1437 return (int)swp_type(*e1
) - (int)swp_type(*e2
);
1440 void swapcache_free_entries(swp_entry_t
*entries
, int n
)
1442 struct swap_info_struct
*p
, *prev
;
1452 * Sort swap entries by swap device, so each lock is only taken once.
1453 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1454 * so low that it isn't necessary to optimize further.
1456 if (nr_swapfiles
> 1)
1457 sort(entries
, n
, sizeof(entries
[0]), swp_entry_cmp
, NULL
);
1458 for (i
= 0; i
< n
; ++i
) {
1459 p
= swap_info_get_cont(entries
[i
], prev
);
1461 swap_entry_free(p
, entries
[i
]);
1465 spin_unlock(&p
->lock
);
1469 * How many references to page are currently swapped out?
1470 * This does not give an exact answer when swap count is continued,
1471 * but does include the high COUNT_CONTINUED flag to allow for that.
1473 int page_swapcount(struct page
*page
)
1476 struct swap_info_struct
*p
;
1477 struct swap_cluster_info
*ci
;
1479 unsigned long offset
;
1481 entry
.val
= page_private(page
);
1482 p
= _swap_info_get(entry
);
1484 offset
= swp_offset(entry
);
1485 ci
= lock_cluster_or_swap_info(p
, offset
);
1486 count
= swap_count(p
->swap_map
[offset
]);
1487 unlock_cluster_or_swap_info(p
, ci
);
1492 int __swap_count(swp_entry_t entry
)
1494 struct swap_info_struct
*si
;
1495 pgoff_t offset
= swp_offset(entry
);
1498 si
= get_swap_device(entry
);
1500 count
= swap_count(si
->swap_map
[offset
]);
1501 put_swap_device(si
);
1506 static int swap_swapcount(struct swap_info_struct
*si
, swp_entry_t entry
)
1509 pgoff_t offset
= swp_offset(entry
);
1510 struct swap_cluster_info
*ci
;
1512 ci
= lock_cluster_or_swap_info(si
, offset
);
1513 count
= swap_count(si
->swap_map
[offset
]);
1514 unlock_cluster_or_swap_info(si
, ci
);
1519 * How many references to @entry are currently swapped out?
1520 * This does not give an exact answer when swap count is continued,
1521 * but does include the high COUNT_CONTINUED flag to allow for that.
1523 int __swp_swapcount(swp_entry_t entry
)
1526 struct swap_info_struct
*si
;
1528 si
= get_swap_device(entry
);
1530 count
= swap_swapcount(si
, entry
);
1531 put_swap_device(si
);
1537 * How many references to @entry are currently swapped out?
1538 * This considers COUNT_CONTINUED so it returns exact answer.
1540 int swp_swapcount(swp_entry_t entry
)
1542 int count
, tmp_count
, n
;
1543 struct swap_info_struct
*p
;
1544 struct swap_cluster_info
*ci
;
1549 p
= _swap_info_get(entry
);
1553 offset
= swp_offset(entry
);
1555 ci
= lock_cluster_or_swap_info(p
, offset
);
1557 count
= swap_count(p
->swap_map
[offset
]);
1558 if (!(count
& COUNT_CONTINUED
))
1561 count
&= ~COUNT_CONTINUED
;
1562 n
= SWAP_MAP_MAX
+ 1;
1564 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1565 offset
&= ~PAGE_MASK
;
1566 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1569 page
= list_next_entry(page
, lru
);
1570 map
= kmap_atomic(page
);
1571 tmp_count
= map
[offset
];
1574 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1575 n
*= (SWAP_CONT_MAX
+ 1);
1576 } while (tmp_count
& COUNT_CONTINUED
);
1578 unlock_cluster_or_swap_info(p
, ci
);
1582 static bool swap_page_trans_huge_swapped(struct swap_info_struct
*si
,
1585 struct swap_cluster_info
*ci
;
1586 unsigned char *map
= si
->swap_map
;
1587 unsigned long roffset
= swp_offset(entry
);
1588 unsigned long offset
= round_down(roffset
, SWAPFILE_CLUSTER
);
1592 ci
= lock_cluster_or_swap_info(si
, offset
);
1593 if (!ci
|| !cluster_is_huge(ci
)) {
1594 if (swap_count(map
[roffset
]))
1598 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1599 if (swap_count(map
[offset
+ i
])) {
1605 unlock_cluster_or_swap_info(si
, ci
);
1609 static bool page_swapped(struct page
*page
)
1612 struct swap_info_struct
*si
;
1614 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
)))
1615 return page_swapcount(page
) != 0;
1617 page
= compound_head(page
);
1618 entry
.val
= page_private(page
);
1619 si
= _swap_info_get(entry
);
1621 return swap_page_trans_huge_swapped(si
, entry
);
1625 static int page_trans_huge_map_swapcount(struct page
*page
, int *total_mapcount
,
1626 int *total_swapcount
)
1628 int i
, map_swapcount
, _total_mapcount
, _total_swapcount
;
1629 unsigned long offset
= 0;
1630 struct swap_info_struct
*si
;
1631 struct swap_cluster_info
*ci
= NULL
;
1632 unsigned char *map
= NULL
;
1633 int mapcount
, swapcount
= 0;
1635 /* hugetlbfs shouldn't call it */
1636 VM_BUG_ON_PAGE(PageHuge(page
), page
);
1638 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
))) {
1639 mapcount
= page_trans_huge_mapcount(page
, total_mapcount
);
1640 if (PageSwapCache(page
))
1641 swapcount
= page_swapcount(page
);
1642 if (total_swapcount
)
1643 *total_swapcount
= swapcount
;
1644 return mapcount
+ swapcount
;
1647 page
= compound_head(page
);
1649 _total_mapcount
= _total_swapcount
= map_swapcount
= 0;
1650 if (PageSwapCache(page
)) {
1653 entry
.val
= page_private(page
);
1654 si
= _swap_info_get(entry
);
1657 offset
= swp_offset(entry
);
1661 ci
= lock_cluster(si
, offset
);
1662 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
1663 mapcount
= atomic_read(&page
[i
]._mapcount
) + 1;
1664 _total_mapcount
+= mapcount
;
1666 swapcount
= swap_count(map
[offset
+ i
]);
1667 _total_swapcount
+= swapcount
;
1669 map_swapcount
= max(map_swapcount
, mapcount
+ swapcount
);
1672 if (PageDoubleMap(page
)) {
1674 _total_mapcount
-= HPAGE_PMD_NR
;
1676 mapcount
= compound_mapcount(page
);
1677 map_swapcount
+= mapcount
;
1678 _total_mapcount
+= mapcount
;
1680 *total_mapcount
= _total_mapcount
;
1681 if (total_swapcount
)
1682 *total_swapcount
= _total_swapcount
;
1684 return map_swapcount
;
1688 * We can write to an anon page without COW if there are no other references
1689 * to it. And as a side-effect, free up its swap: because the old content
1690 * on disk will never be read, and seeking back there to write new content
1691 * later would only waste time away from clustering.
1693 * NOTE: total_map_swapcount should not be relied upon by the caller if
1694 * reuse_swap_page() returns false, but it may be always overwritten
1695 * (see the other implementation for CONFIG_SWAP=n).
1697 bool reuse_swap_page(struct page
*page
, int *total_map_swapcount
)
1699 int count
, total_mapcount
, total_swapcount
;
1701 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1702 if (unlikely(PageKsm(page
)))
1704 count
= page_trans_huge_map_swapcount(page
, &total_mapcount
,
1706 if (total_map_swapcount
)
1707 *total_map_swapcount
= total_mapcount
+ total_swapcount
;
1708 if (count
== 1 && PageSwapCache(page
) &&
1709 (likely(!PageTransCompound(page
)) ||
1710 /* The remaining swap count will be freed soon */
1711 total_swapcount
== page_swapcount(page
))) {
1712 if (!PageWriteback(page
)) {
1713 page
= compound_head(page
);
1714 delete_from_swap_cache(page
);
1718 struct swap_info_struct
*p
;
1720 entry
.val
= page_private(page
);
1721 p
= swap_info_get(entry
);
1722 if (p
->flags
& SWP_STABLE_WRITES
) {
1723 spin_unlock(&p
->lock
);
1726 spin_unlock(&p
->lock
);
1734 * If swap is getting full, or if there are no more mappings of this page,
1735 * then try_to_free_swap is called to free its swap space.
1737 int try_to_free_swap(struct page
*page
)
1739 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1741 if (!PageSwapCache(page
))
1743 if (PageWriteback(page
))
1745 if (page_swapped(page
))
1749 * Once hibernation has begun to create its image of memory,
1750 * there's a danger that one of the calls to try_to_free_swap()
1751 * - most probably a call from __try_to_reclaim_swap() while
1752 * hibernation is allocating its own swap pages for the image,
1753 * but conceivably even a call from memory reclaim - will free
1754 * the swap from a page which has already been recorded in the
1755 * image as a clean swapcache page, and then reuse its swap for
1756 * another page of the image. On waking from hibernation, the
1757 * original page might be freed under memory pressure, then
1758 * later read back in from swap, now with the wrong data.
1760 * Hibernation suspends storage while it is writing the image
1761 * to disk so check that here.
1763 if (pm_suspended_storage())
1766 page
= compound_head(page
);
1767 delete_from_swap_cache(page
);
1773 * Free the swap entry like above, but also try to
1774 * free the page cache entry if it is the last user.
1776 int free_swap_and_cache(swp_entry_t entry
)
1778 struct swap_info_struct
*p
;
1779 unsigned char count
;
1781 if (non_swap_entry(entry
))
1784 p
= _swap_info_get(entry
);
1786 count
= __swap_entry_free(p
, entry
);
1787 if (count
== SWAP_HAS_CACHE
&&
1788 !swap_page_trans_huge_swapped(p
, entry
))
1789 __try_to_reclaim_swap(p
, swp_offset(entry
),
1790 TTRS_UNMAPPED
| TTRS_FULL
);
1795 #ifdef CONFIG_HIBERNATION
1797 * Find the swap type that corresponds to given device (if any).
1799 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1800 * from 0, in which the swap header is expected to be located.
1802 * This is needed for the suspend to disk (aka swsusp).
1804 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
1806 struct block_device
*bdev
= NULL
;
1810 bdev
= bdget(device
);
1812 spin_lock(&swap_lock
);
1813 for (type
= 0; type
< nr_swapfiles
; type
++) {
1814 struct swap_info_struct
*sis
= swap_info
[type
];
1816 if (!(sis
->flags
& SWP_WRITEOK
))
1821 *bdev_p
= bdgrab(sis
->bdev
);
1823 spin_unlock(&swap_lock
);
1826 if (bdev
== sis
->bdev
) {
1827 struct swap_extent
*se
= first_se(sis
);
1829 if (se
->start_block
== offset
) {
1831 *bdev_p
= bdgrab(sis
->bdev
);
1833 spin_unlock(&swap_lock
);
1839 spin_unlock(&swap_lock
);
1847 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1848 * corresponding to given index in swap_info (swap type).
1850 sector_t
swapdev_block(int type
, pgoff_t offset
)
1852 struct block_device
*bdev
;
1853 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1855 if (!si
|| !(si
->flags
& SWP_WRITEOK
))
1857 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
1861 * Return either the total number of swap pages of given type, or the number
1862 * of free pages of that type (depending on @free)
1864 * This is needed for software suspend
1866 unsigned int count_swap_pages(int type
, int free
)
1870 spin_lock(&swap_lock
);
1871 if ((unsigned int)type
< nr_swapfiles
) {
1872 struct swap_info_struct
*sis
= swap_info
[type
];
1874 spin_lock(&sis
->lock
);
1875 if (sis
->flags
& SWP_WRITEOK
) {
1878 n
-= sis
->inuse_pages
;
1880 spin_unlock(&sis
->lock
);
1882 spin_unlock(&swap_lock
);
1885 #endif /* CONFIG_HIBERNATION */
1887 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1889 return pte_same(pte_swp_clear_soft_dirty(pte
), swp_pte
);
1893 * No need to decide whether this PTE shares the swap entry with others,
1894 * just let do_wp_page work it out if a write is requested later - to
1895 * force COW, vm_page_prot omits write permission from any private vma.
1897 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1898 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1900 struct page
*swapcache
;
1906 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1907 if (unlikely(!page
))
1910 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1911 if (unlikely(!pte_same_as_swp(*pte
, swp_entry_to_pte(entry
)))) {
1916 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1917 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1919 set_pte_at(vma
->vm_mm
, addr
, pte
,
1920 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1921 if (page
== swapcache
) {
1922 page_add_anon_rmap(page
, vma
, addr
, false);
1923 } else { /* ksm created a completely new copy */
1924 page_add_new_anon_rmap(page
, vma
, addr
, false);
1925 lru_cache_add_inactive_or_unevictable(page
, vma
);
1929 * Move the page to the active list so it is not
1930 * immediately swapped out again after swapon.
1932 activate_page(page
);
1934 pte_unmap_unlock(pte
, ptl
);
1935 if (page
!= swapcache
) {
1942 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1943 unsigned long addr
, unsigned long end
,
1944 unsigned int type
, bool frontswap
,
1945 unsigned long *fs_pages_to_unuse
)
1950 struct swap_info_struct
*si
;
1951 unsigned long offset
;
1953 volatile unsigned char *swap_map
;
1955 si
= swap_info
[type
];
1956 pte
= pte_offset_map(pmd
, addr
);
1958 struct vm_fault vmf
;
1960 if (!is_swap_pte(*pte
))
1963 entry
= pte_to_swp_entry(*pte
);
1964 if (swp_type(entry
) != type
)
1967 offset
= swp_offset(entry
);
1968 if (frontswap
&& !frontswap_test(si
, offset
))
1972 swap_map
= &si
->swap_map
[offset
];
1973 page
= lookup_swap_cache(entry
, vma
, addr
);
1978 page
= swapin_readahead(entry
, GFP_HIGHUSER_MOVABLE
,
1982 if (*swap_map
== 0 || *swap_map
== SWAP_MAP_BAD
)
1988 wait_on_page_writeback(page
);
1989 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1996 try_to_free_swap(page
);
2000 if (*fs_pages_to_unuse
&& !--(*fs_pages_to_unuse
)) {
2001 ret
= FRONTSWAP_PAGES_UNUSED
;
2005 pte
= pte_offset_map(pmd
, addr
);
2006 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
2014 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
2015 unsigned long addr
, unsigned long end
,
2016 unsigned int type
, bool frontswap
,
2017 unsigned long *fs_pages_to_unuse
)
2023 pmd
= pmd_offset(pud
, addr
);
2026 next
= pmd_addr_end(addr
, end
);
2027 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
2029 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, type
,
2030 frontswap
, fs_pages_to_unuse
);
2033 } while (pmd
++, addr
= next
, addr
!= end
);
2037 static inline int unuse_pud_range(struct vm_area_struct
*vma
, p4d_t
*p4d
,
2038 unsigned long addr
, unsigned long end
,
2039 unsigned int type
, bool frontswap
,
2040 unsigned long *fs_pages_to_unuse
)
2046 pud
= pud_offset(p4d
, addr
);
2048 next
= pud_addr_end(addr
, end
);
2049 if (pud_none_or_clear_bad(pud
))
2051 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, type
,
2052 frontswap
, fs_pages_to_unuse
);
2055 } while (pud
++, addr
= next
, addr
!= end
);
2059 static inline int unuse_p4d_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
2060 unsigned long addr
, unsigned long end
,
2061 unsigned int type
, bool frontswap
,
2062 unsigned long *fs_pages_to_unuse
)
2068 p4d
= p4d_offset(pgd
, addr
);
2070 next
= p4d_addr_end(addr
, end
);
2071 if (p4d_none_or_clear_bad(p4d
))
2073 ret
= unuse_pud_range(vma
, p4d
, addr
, next
, type
,
2074 frontswap
, fs_pages_to_unuse
);
2077 } while (p4d
++, addr
= next
, addr
!= end
);
2081 static int unuse_vma(struct vm_area_struct
*vma
, unsigned int type
,
2082 bool frontswap
, unsigned long *fs_pages_to_unuse
)
2085 unsigned long addr
, end
, next
;
2088 addr
= vma
->vm_start
;
2091 pgd
= pgd_offset(vma
->vm_mm
, addr
);
2093 next
= pgd_addr_end(addr
, end
);
2094 if (pgd_none_or_clear_bad(pgd
))
2096 ret
= unuse_p4d_range(vma
, pgd
, addr
, next
, type
,
2097 frontswap
, fs_pages_to_unuse
);
2100 } while (pgd
++, addr
= next
, addr
!= end
);
2104 static int unuse_mm(struct mm_struct
*mm
, unsigned int type
,
2105 bool frontswap
, unsigned long *fs_pages_to_unuse
)
2107 struct vm_area_struct
*vma
;
2111 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2112 if (vma
->anon_vma
) {
2113 ret
= unuse_vma(vma
, type
, frontswap
,
2120 mmap_read_unlock(mm
);
2125 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2126 * from current position to next entry still in use. Return 0
2127 * if there are no inuse entries after prev till end of the map.
2129 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
2130 unsigned int prev
, bool frontswap
)
2133 unsigned char count
;
2136 * No need for swap_lock here: we're just looking
2137 * for whether an entry is in use, not modifying it; false
2138 * hits are okay, and sys_swapoff() has already prevented new
2139 * allocations from this area (while holding swap_lock).
2141 for (i
= prev
+ 1; i
< si
->max
; i
++) {
2142 count
= READ_ONCE(si
->swap_map
[i
]);
2143 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
2144 if (!frontswap
|| frontswap_test(si
, i
))
2146 if ((i
% LATENCY_LIMIT
) == 0)
2157 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2158 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2160 int try_to_unuse(unsigned int type
, bool frontswap
,
2161 unsigned long pages_to_unuse
)
2163 struct mm_struct
*prev_mm
;
2164 struct mm_struct
*mm
;
2165 struct list_head
*p
;
2167 struct swap_info_struct
*si
= swap_info
[type
];
2172 if (!READ_ONCE(si
->inuse_pages
))
2179 retval
= shmem_unuse(type
, frontswap
, &pages_to_unuse
);
2186 spin_lock(&mmlist_lock
);
2187 p
= &init_mm
.mmlist
;
2188 while (READ_ONCE(si
->inuse_pages
) &&
2189 !signal_pending(current
) &&
2190 (p
= p
->next
) != &init_mm
.mmlist
) {
2192 mm
= list_entry(p
, struct mm_struct
, mmlist
);
2193 if (!mmget_not_zero(mm
))
2195 spin_unlock(&mmlist_lock
);
2198 retval
= unuse_mm(mm
, type
, frontswap
, &pages_to_unuse
);
2206 * Make sure that we aren't completely killing
2207 * interactive performance.
2210 spin_lock(&mmlist_lock
);
2212 spin_unlock(&mmlist_lock
);
2217 while (READ_ONCE(si
->inuse_pages
) &&
2218 !signal_pending(current
) &&
2219 (i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
2221 entry
= swp_entry(type
, i
);
2222 page
= find_get_page(swap_address_space(entry
), i
);
2227 * It is conceivable that a racing task removed this page from
2228 * swap cache just before we acquired the page lock. The page
2229 * might even be back in swap cache on another swap area. But
2230 * that is okay, try_to_free_swap() only removes stale pages.
2233 wait_on_page_writeback(page
);
2234 try_to_free_swap(page
);
2239 * For frontswap, we just need to unuse pages_to_unuse, if
2240 * it was specified. Need not check frontswap again here as
2241 * we already zeroed out pages_to_unuse if not frontswap.
2243 if (pages_to_unuse
&& --pages_to_unuse
== 0)
2248 * Lets check again to see if there are still swap entries in the map.
2249 * If yes, we would need to do retry the unuse logic again.
2250 * Under global memory pressure, swap entries can be reinserted back
2251 * into process space after the mmlist loop above passes over them.
2253 * Limit the number of retries? No: when mmget_not_zero() above fails,
2254 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2255 * at its own independent pace; and even shmem_writepage() could have
2256 * been preempted after get_swap_page(), temporarily hiding that swap.
2257 * It's easy and robust (though cpu-intensive) just to keep retrying.
2259 if (READ_ONCE(si
->inuse_pages
)) {
2260 if (!signal_pending(current
))
2265 return (retval
== FRONTSWAP_PAGES_UNUSED
) ? 0 : retval
;
2269 * After a successful try_to_unuse, if no swap is now in use, we know
2270 * we can empty the mmlist. swap_lock must be held on entry and exit.
2271 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2272 * added to the mmlist just after page_duplicate - before would be racy.
2274 static void drain_mmlist(void)
2276 struct list_head
*p
, *next
;
2279 for (type
= 0; type
< nr_swapfiles
; type
++)
2280 if (swap_info
[type
]->inuse_pages
)
2282 spin_lock(&mmlist_lock
);
2283 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
2285 spin_unlock(&mmlist_lock
);
2289 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2290 * corresponds to page offset for the specified swap entry.
2291 * Note that the type of this function is sector_t, but it returns page offset
2292 * into the bdev, not sector offset.
2294 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
2296 struct swap_info_struct
*sis
;
2297 struct swap_extent
*se
;
2300 sis
= swp_swap_info(entry
);
2303 offset
= swp_offset(entry
);
2304 se
= offset_to_swap_extent(sis
, offset
);
2305 return se
->start_block
+ (offset
- se
->start_page
);
2309 * Returns the page offset into bdev for the specified page's swap entry.
2311 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
2314 entry
.val
= page_private(page
);
2315 return map_swap_entry(entry
, bdev
);
2319 * Free all of a swapdev's extent information
2321 static void destroy_swap_extents(struct swap_info_struct
*sis
)
2323 while (!RB_EMPTY_ROOT(&sis
->swap_extent_root
)) {
2324 struct rb_node
*rb
= sis
->swap_extent_root
.rb_node
;
2325 struct swap_extent
*se
= rb_entry(rb
, struct swap_extent
, rb_node
);
2327 rb_erase(rb
, &sis
->swap_extent_root
);
2331 if (sis
->flags
& SWP_ACTIVATED
) {
2332 struct file
*swap_file
= sis
->swap_file
;
2333 struct address_space
*mapping
= swap_file
->f_mapping
;
2335 sis
->flags
&= ~SWP_ACTIVATED
;
2336 if (mapping
->a_ops
->swap_deactivate
)
2337 mapping
->a_ops
->swap_deactivate(swap_file
);
2342 * Add a block range (and the corresponding page range) into this swapdev's
2345 * This function rather assumes that it is called in ascending page order.
2348 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
2349 unsigned long nr_pages
, sector_t start_block
)
2351 struct rb_node
**link
= &sis
->swap_extent_root
.rb_node
, *parent
= NULL
;
2352 struct swap_extent
*se
;
2353 struct swap_extent
*new_se
;
2356 * place the new node at the right most since the
2357 * function is called in ascending page order.
2361 link
= &parent
->rb_right
;
2365 se
= rb_entry(parent
, struct swap_extent
, rb_node
);
2366 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
2367 if (se
->start_block
+ se
->nr_pages
== start_block
) {
2369 se
->nr_pages
+= nr_pages
;
2374 /* No merge, insert a new extent. */
2375 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
2378 new_se
->start_page
= start_page
;
2379 new_se
->nr_pages
= nr_pages
;
2380 new_se
->start_block
= start_block
;
2382 rb_link_node(&new_se
->rb_node
, parent
, link
);
2383 rb_insert_color(&new_se
->rb_node
, &sis
->swap_extent_root
);
2386 EXPORT_SYMBOL_GPL(add_swap_extent
);
2389 * A `swap extent' is a simple thing which maps a contiguous range of pages
2390 * onto a contiguous range of disk blocks. An ordered list of swap extents
2391 * is built at swapon time and is then used at swap_writepage/swap_readpage
2392 * time for locating where on disk a page belongs.
2394 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2395 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2396 * swap files identically.
2398 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2399 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2400 * swapfiles are handled *identically* after swapon time.
2402 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2403 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2404 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2405 * requirements, they are simply tossed out - we will never use those blocks
2408 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2409 * prevents users from writing to the swap device, which will corrupt memory.
2411 * The amount of disk space which a single swap extent represents varies.
2412 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2413 * extents in the list. To avoid much list walking, we cache the previous
2414 * search location in `curr_swap_extent', and start new searches from there.
2415 * This is extremely effective. The average number of iterations in
2416 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2418 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
2420 struct file
*swap_file
= sis
->swap_file
;
2421 struct address_space
*mapping
= swap_file
->f_mapping
;
2422 struct inode
*inode
= mapping
->host
;
2425 if (S_ISBLK(inode
->i_mode
)) {
2426 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2431 if (mapping
->a_ops
->swap_activate
) {
2432 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
2434 sis
->flags
|= SWP_ACTIVATED
;
2436 sis
->flags
|= SWP_FS
;
2437 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2443 return generic_swapfile_activate(sis
, swap_file
, span
);
2446 static int swap_node(struct swap_info_struct
*p
)
2448 struct block_device
*bdev
;
2453 bdev
= p
->swap_file
->f_inode
->i_sb
->s_bdev
;
2455 return bdev
? bdev
->bd_disk
->node_id
: NUMA_NO_NODE
;
2458 static void setup_swap_info(struct swap_info_struct
*p
, int prio
,
2459 unsigned char *swap_map
,
2460 struct swap_cluster_info
*cluster_info
)
2467 p
->prio
= --least_priority
;
2469 * the plist prio is negated because plist ordering is
2470 * low-to-high, while swap ordering is high-to-low
2472 p
->list
.prio
= -p
->prio
;
2475 p
->avail_lists
[i
].prio
= -p
->prio
;
2477 if (swap_node(p
) == i
)
2478 p
->avail_lists
[i
].prio
= 1;
2480 p
->avail_lists
[i
].prio
= -p
->prio
;
2483 p
->swap_map
= swap_map
;
2484 p
->cluster_info
= cluster_info
;
2487 static void _enable_swap_info(struct swap_info_struct
*p
)
2489 p
->flags
|= SWP_WRITEOK
| SWP_VALID
;
2490 atomic_long_add(p
->pages
, &nr_swap_pages
);
2491 total_swap_pages
+= p
->pages
;
2493 assert_spin_locked(&swap_lock
);
2495 * both lists are plists, and thus priority ordered.
2496 * swap_active_head needs to be priority ordered for swapoff(),
2497 * which on removal of any swap_info_struct with an auto-assigned
2498 * (i.e. negative) priority increments the auto-assigned priority
2499 * of any lower-priority swap_info_structs.
2500 * swap_avail_head needs to be priority ordered for get_swap_page(),
2501 * which allocates swap pages from the highest available priority
2504 plist_add(&p
->list
, &swap_active_head
);
2505 add_to_avail_list(p
);
2508 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
2509 unsigned char *swap_map
,
2510 struct swap_cluster_info
*cluster_info
,
2511 unsigned long *frontswap_map
)
2513 frontswap_init(p
->type
, frontswap_map
);
2514 spin_lock(&swap_lock
);
2515 spin_lock(&p
->lock
);
2516 setup_swap_info(p
, prio
, swap_map
, cluster_info
);
2517 spin_unlock(&p
->lock
);
2518 spin_unlock(&swap_lock
);
2520 * Guarantee swap_map, cluster_info, etc. fields are valid
2521 * between get/put_swap_device() if SWP_VALID bit is set
2524 spin_lock(&swap_lock
);
2525 spin_lock(&p
->lock
);
2526 _enable_swap_info(p
);
2527 spin_unlock(&p
->lock
);
2528 spin_unlock(&swap_lock
);
2531 static void reinsert_swap_info(struct swap_info_struct
*p
)
2533 spin_lock(&swap_lock
);
2534 spin_lock(&p
->lock
);
2535 setup_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
2536 _enable_swap_info(p
);
2537 spin_unlock(&p
->lock
);
2538 spin_unlock(&swap_lock
);
2541 bool has_usable_swap(void)
2545 spin_lock(&swap_lock
);
2546 if (plist_head_empty(&swap_active_head
))
2548 spin_unlock(&swap_lock
);
2552 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
2554 struct swap_info_struct
*p
= NULL
;
2555 unsigned char *swap_map
;
2556 struct swap_cluster_info
*cluster_info
;
2557 unsigned long *frontswap_map
;
2558 struct file
*swap_file
, *victim
;
2559 struct address_space
*mapping
;
2560 struct inode
*inode
;
2561 struct filename
*pathname
;
2563 unsigned int old_block_size
;
2565 if (!capable(CAP_SYS_ADMIN
))
2568 BUG_ON(!current
->mm
);
2570 pathname
= getname(specialfile
);
2571 if (IS_ERR(pathname
))
2572 return PTR_ERR(pathname
);
2574 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
2575 err
= PTR_ERR(victim
);
2579 mapping
= victim
->f_mapping
;
2580 spin_lock(&swap_lock
);
2581 plist_for_each_entry(p
, &swap_active_head
, list
) {
2582 if (p
->flags
& SWP_WRITEOK
) {
2583 if (p
->swap_file
->f_mapping
== mapping
) {
2591 spin_unlock(&swap_lock
);
2594 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2595 vm_unacct_memory(p
->pages
);
2598 spin_unlock(&swap_lock
);
2601 del_from_avail_list(p
);
2602 spin_lock(&p
->lock
);
2604 struct swap_info_struct
*si
= p
;
2607 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2610 for_each_node(nid
) {
2611 if (si
->avail_lists
[nid
].prio
!= 1)
2612 si
->avail_lists
[nid
].prio
--;
2617 plist_del(&p
->list
, &swap_active_head
);
2618 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2619 total_swap_pages
-= p
->pages
;
2620 p
->flags
&= ~SWP_WRITEOK
;
2621 spin_unlock(&p
->lock
);
2622 spin_unlock(&swap_lock
);
2624 disable_swap_slots_cache_lock();
2626 set_current_oom_origin();
2627 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
2628 clear_current_oom_origin();
2631 /* re-insert swap space back into swap_list */
2632 reinsert_swap_info(p
);
2633 reenable_swap_slots_cache_unlock();
2637 reenable_swap_slots_cache_unlock();
2639 spin_lock(&swap_lock
);
2640 spin_lock(&p
->lock
);
2641 p
->flags
&= ~SWP_VALID
; /* mark swap device as invalid */
2642 spin_unlock(&p
->lock
);
2643 spin_unlock(&swap_lock
);
2645 * wait for swap operations protected by get/put_swap_device()
2650 flush_work(&p
->discard_work
);
2652 destroy_swap_extents(p
);
2653 if (p
->flags
& SWP_CONTINUED
)
2654 free_swap_count_continuations(p
);
2656 if (!p
->bdev
|| !blk_queue_nonrot(bdev_get_queue(p
->bdev
)))
2657 atomic_dec(&nr_rotate_swap
);
2659 mutex_lock(&swapon_mutex
);
2660 spin_lock(&swap_lock
);
2661 spin_lock(&p
->lock
);
2664 /* wait for anyone still in scan_swap_map */
2665 p
->highest_bit
= 0; /* cuts scans short */
2666 while (p
->flags
>= SWP_SCANNING
) {
2667 spin_unlock(&p
->lock
);
2668 spin_unlock(&swap_lock
);
2669 schedule_timeout_uninterruptible(1);
2670 spin_lock(&swap_lock
);
2671 spin_lock(&p
->lock
);
2674 swap_file
= p
->swap_file
;
2675 old_block_size
= p
->old_block_size
;
2676 p
->swap_file
= NULL
;
2678 swap_map
= p
->swap_map
;
2680 cluster_info
= p
->cluster_info
;
2681 p
->cluster_info
= NULL
;
2682 frontswap_map
= frontswap_map_get(p
);
2683 spin_unlock(&p
->lock
);
2684 spin_unlock(&swap_lock
);
2685 frontswap_invalidate_area(p
->type
);
2686 frontswap_map_set(p
, NULL
);
2687 mutex_unlock(&swapon_mutex
);
2688 free_percpu(p
->percpu_cluster
);
2689 p
->percpu_cluster
= NULL
;
2690 free_percpu(p
->cluster_next_cpu
);
2691 p
->cluster_next_cpu
= NULL
;
2693 kvfree(cluster_info
);
2694 kvfree(frontswap_map
);
2695 /* Destroy swap account information */
2696 swap_cgroup_swapoff(p
->type
);
2697 exit_swap_address_space(p
->type
);
2699 inode
= mapping
->host
;
2700 if (S_ISBLK(inode
->i_mode
)) {
2701 struct block_device
*bdev
= I_BDEV(inode
);
2703 set_blocksize(bdev
, old_block_size
);
2704 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2708 inode
->i_flags
&= ~S_SWAPFILE
;
2709 inode_unlock(inode
);
2710 filp_close(swap_file
, NULL
);
2713 * Clear the SWP_USED flag after all resources are freed so that swapon
2714 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2715 * not hold p->lock after we cleared its SWP_WRITEOK.
2717 spin_lock(&swap_lock
);
2719 spin_unlock(&swap_lock
);
2722 atomic_inc(&proc_poll_event
);
2723 wake_up_interruptible(&proc_poll_wait
);
2726 filp_close(victim
, NULL
);
2732 #ifdef CONFIG_PROC_FS
2733 static __poll_t
swaps_poll(struct file
*file
, poll_table
*wait
)
2735 struct seq_file
*seq
= file
->private_data
;
2737 poll_wait(file
, &proc_poll_wait
, wait
);
2739 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2740 seq
->poll_event
= atomic_read(&proc_poll_event
);
2741 return EPOLLIN
| EPOLLRDNORM
| EPOLLERR
| EPOLLPRI
;
2744 return EPOLLIN
| EPOLLRDNORM
;
2748 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2750 struct swap_info_struct
*si
;
2754 mutex_lock(&swapon_mutex
);
2757 return SEQ_START_TOKEN
;
2759 for (type
= 0; (si
= swap_type_to_swap_info(type
)); type
++) {
2760 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2769 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2771 struct swap_info_struct
*si
= v
;
2774 if (v
== SEQ_START_TOKEN
)
2777 type
= si
->type
+ 1;
2780 for (; (si
= swap_type_to_swap_info(type
)); type
++) {
2781 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2789 static void swap_stop(struct seq_file
*swap
, void *v
)
2791 mutex_unlock(&swapon_mutex
);
2794 static int swap_show(struct seq_file
*swap
, void *v
)
2796 struct swap_info_struct
*si
= v
;
2799 unsigned int bytes
, inuse
;
2801 if (si
== SEQ_START_TOKEN
) {
2802 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2806 bytes
= si
->pages
<< (PAGE_SHIFT
- 10);
2807 inuse
= si
->inuse_pages
<< (PAGE_SHIFT
- 10);
2809 file
= si
->swap_file
;
2810 len
= seq_file_path(swap
, file
, " \t\n\\");
2811 seq_printf(swap
, "%*s%s\t%u\t%s%u\t%s%d\n",
2812 len
< 40 ? 40 - len
: 1, " ",
2813 S_ISBLK(file_inode(file
)->i_mode
) ?
2814 "partition" : "file\t",
2815 bytes
, bytes
< 10000000 ? "\t" : "",
2816 inuse
, inuse
< 10000000 ? "\t" : "",
2821 static const struct seq_operations swaps_op
= {
2822 .start
= swap_start
,
2828 static int swaps_open(struct inode
*inode
, struct file
*file
)
2830 struct seq_file
*seq
;
2833 ret
= seq_open(file
, &swaps_op
);
2837 seq
= file
->private_data
;
2838 seq
->poll_event
= atomic_read(&proc_poll_event
);
2842 static const struct proc_ops swaps_proc_ops
= {
2843 .proc_flags
= PROC_ENTRY_PERMANENT
,
2844 .proc_open
= swaps_open
,
2845 .proc_read
= seq_read
,
2846 .proc_lseek
= seq_lseek
,
2847 .proc_release
= seq_release
,
2848 .proc_poll
= swaps_poll
,
2851 static int __init
procswaps_init(void)
2853 proc_create("swaps", 0, NULL
, &swaps_proc_ops
);
2856 __initcall(procswaps_init
);
2857 #endif /* CONFIG_PROC_FS */
2859 #ifdef MAX_SWAPFILES_CHECK
2860 static int __init
max_swapfiles_check(void)
2862 MAX_SWAPFILES_CHECK();
2865 late_initcall(max_swapfiles_check
);
2868 static struct swap_info_struct
*alloc_swap_info(void)
2870 struct swap_info_struct
*p
;
2874 p
= kvzalloc(struct_size(p
, avail_lists
, nr_node_ids
), GFP_KERNEL
);
2876 return ERR_PTR(-ENOMEM
);
2878 spin_lock(&swap_lock
);
2879 for (type
= 0; type
< nr_swapfiles
; type
++) {
2880 if (!(swap_info
[type
]->flags
& SWP_USED
))
2883 if (type
>= MAX_SWAPFILES
) {
2884 spin_unlock(&swap_lock
);
2886 return ERR_PTR(-EPERM
);
2888 if (type
>= nr_swapfiles
) {
2890 WRITE_ONCE(swap_info
[type
], p
);
2892 * Write swap_info[type] before nr_swapfiles, in case a
2893 * racing procfs swap_start() or swap_next() is reading them.
2894 * (We never shrink nr_swapfiles, we never free this entry.)
2897 WRITE_ONCE(nr_swapfiles
, nr_swapfiles
+ 1);
2900 p
= swap_info
[type
];
2902 * Do not memset this entry: a racing procfs swap_next()
2903 * would be relying on p->type to remain valid.
2906 p
->swap_extent_root
= RB_ROOT
;
2907 plist_node_init(&p
->list
, 0);
2909 plist_node_init(&p
->avail_lists
[i
], 0);
2910 p
->flags
= SWP_USED
;
2911 spin_unlock(&swap_lock
);
2912 spin_lock_init(&p
->lock
);
2913 spin_lock_init(&p
->cont_lock
);
2918 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2922 if (S_ISBLK(inode
->i_mode
)) {
2923 p
->bdev
= bdgrab(I_BDEV(inode
));
2924 error
= blkdev_get(p
->bdev
,
2925 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, p
);
2930 p
->old_block_size
= block_size(p
->bdev
);
2931 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2935 * Zoned block devices contain zones that have a sequential
2936 * write only restriction. Hence zoned block devices are not
2937 * suitable for swapping. Disallow them here.
2939 if (blk_queue_is_zoned(p
->bdev
->bd_disk
->queue
))
2941 p
->flags
|= SWP_BLKDEV
;
2942 } else if (S_ISREG(inode
->i_mode
)) {
2943 p
->bdev
= inode
->i_sb
->s_bdev
;
2951 * Find out how many pages are allowed for a single swap device. There
2952 * are two limiting factors:
2953 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2954 * 2) the number of bits in the swap pte, as defined by the different
2957 * In order to find the largest possible bit mask, a swap entry with
2958 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2959 * decoded to a swp_entry_t again, and finally the swap offset is
2962 * This will mask all the bits from the initial ~0UL mask that can't
2963 * be encoded in either the swp_entry_t or the architecture definition
2966 unsigned long generic_max_swapfile_size(void)
2968 return swp_offset(pte_to_swp_entry(
2969 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2972 /* Can be overridden by an architecture for additional checks. */
2973 __weak
unsigned long max_swapfile_size(void)
2975 return generic_max_swapfile_size();
2978 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2979 union swap_header
*swap_header
,
2980 struct inode
*inode
)
2983 unsigned long maxpages
;
2984 unsigned long swapfilepages
;
2985 unsigned long last_page
;
2987 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2988 pr_err("Unable to find swap-space signature\n");
2992 /* swap partition endianess hack... */
2993 if (swab32(swap_header
->info
.version
) == 1) {
2994 swab32s(&swap_header
->info
.version
);
2995 swab32s(&swap_header
->info
.last_page
);
2996 swab32s(&swap_header
->info
.nr_badpages
);
2997 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2999 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
3000 swab32s(&swap_header
->info
.badpages
[i
]);
3002 /* Check the swap header's sub-version */
3003 if (swap_header
->info
.version
!= 1) {
3004 pr_warn("Unable to handle swap header version %d\n",
3005 swap_header
->info
.version
);
3010 p
->cluster_next
= 1;
3013 maxpages
= max_swapfile_size();
3014 last_page
= swap_header
->info
.last_page
;
3016 pr_warn("Empty swap-file\n");
3019 if (last_page
> maxpages
) {
3020 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
3021 maxpages
<< (PAGE_SHIFT
- 10),
3022 last_page
<< (PAGE_SHIFT
- 10));
3024 if (maxpages
> last_page
) {
3025 maxpages
= last_page
+ 1;
3026 /* p->max is an unsigned int: don't overflow it */
3027 if ((unsigned int)maxpages
== 0)
3028 maxpages
= UINT_MAX
;
3030 p
->highest_bit
= maxpages
- 1;
3034 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
3035 if (swapfilepages
&& maxpages
> swapfilepages
) {
3036 pr_warn("Swap area shorter than signature indicates\n");
3039 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
3041 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
3047 #define SWAP_CLUSTER_INFO_COLS \
3048 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3049 #define SWAP_CLUSTER_SPACE_COLS \
3050 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3051 #define SWAP_CLUSTER_COLS \
3052 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3054 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
3055 union swap_header
*swap_header
,
3056 unsigned char *swap_map
,
3057 struct swap_cluster_info
*cluster_info
,
3058 unsigned long maxpages
,
3062 unsigned int nr_good_pages
;
3064 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3065 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
3066 unsigned long i
, idx
;
3068 nr_good_pages
= maxpages
- 1; /* omit header page */
3070 cluster_list_init(&p
->free_clusters
);
3071 cluster_list_init(&p
->discard_clusters
);
3073 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
3074 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
3075 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
3077 if (page_nr
< maxpages
) {
3078 swap_map
[page_nr
] = SWAP_MAP_BAD
;
3081 * Haven't marked the cluster free yet, no list
3082 * operation involved
3084 inc_cluster_info_page(p
, cluster_info
, page_nr
);
3088 /* Haven't marked the cluster free yet, no list operation involved */
3089 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
3090 inc_cluster_info_page(p
, cluster_info
, i
);
3092 if (nr_good_pages
) {
3093 swap_map
[0] = SWAP_MAP_BAD
;
3095 * Not mark the cluster free yet, no list
3096 * operation involved
3098 inc_cluster_info_page(p
, cluster_info
, 0);
3100 p
->pages
= nr_good_pages
;
3101 nr_extents
= setup_swap_extents(p
, span
);
3104 nr_good_pages
= p
->pages
;
3106 if (!nr_good_pages
) {
3107 pr_warn("Empty swap-file\n");
3116 * Reduce false cache line sharing between cluster_info and
3117 * sharing same address space.
3119 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
3120 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
3121 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
3122 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
3123 if (idx
>= nr_clusters
)
3125 if (cluster_count(&cluster_info
[idx
]))
3127 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
3128 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
3136 * Helper to sys_swapon determining if a given swap
3137 * backing device queue supports DISCARD operations.
3139 static bool swap_discardable(struct swap_info_struct
*si
)
3141 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
3143 if (!q
|| !blk_queue_discard(q
))
3149 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
3151 struct swap_info_struct
*p
;
3152 struct filename
*name
;
3153 struct file
*swap_file
= NULL
;
3154 struct address_space
*mapping
;
3157 union swap_header
*swap_header
;
3160 unsigned long maxpages
;
3161 unsigned char *swap_map
= NULL
;
3162 struct swap_cluster_info
*cluster_info
= NULL
;
3163 unsigned long *frontswap_map
= NULL
;
3164 struct page
*page
= NULL
;
3165 struct inode
*inode
= NULL
;
3166 bool inced_nr_rotate_swap
= false;
3168 if (swap_flags
& ~SWAP_FLAGS_VALID
)
3171 if (!capable(CAP_SYS_ADMIN
))
3174 if (!swap_avail_heads
)
3177 p
= alloc_swap_info();
3181 INIT_WORK(&p
->discard_work
, swap_discard_work
);
3183 name
= getname(specialfile
);
3185 error
= PTR_ERR(name
);
3189 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
3190 if (IS_ERR(swap_file
)) {
3191 error
= PTR_ERR(swap_file
);
3196 p
->swap_file
= swap_file
;
3197 mapping
= swap_file
->f_mapping
;
3198 inode
= mapping
->host
;
3200 error
= claim_swapfile(p
, inode
);
3201 if (unlikely(error
))
3205 if (IS_SWAPFILE(inode
)) {
3207 goto bad_swap_unlock_inode
;
3211 * Read the swap header.
3213 if (!mapping
->a_ops
->readpage
) {
3215 goto bad_swap_unlock_inode
;
3217 page
= read_mapping_page(mapping
, 0, swap_file
);
3219 error
= PTR_ERR(page
);
3220 goto bad_swap_unlock_inode
;
3222 swap_header
= kmap(page
);
3224 maxpages
= read_swap_header(p
, swap_header
, inode
);
3225 if (unlikely(!maxpages
)) {
3227 goto bad_swap_unlock_inode
;
3230 /* OK, set up the swap map and apply the bad block list */
3231 swap_map
= vzalloc(maxpages
);
3234 goto bad_swap_unlock_inode
;
3237 if (bdi_cap_stable_pages_required(inode_to_bdi(inode
)))
3238 p
->flags
|= SWP_STABLE_WRITES
;
3240 if (bdi_cap_synchronous_io(inode_to_bdi(inode
)))
3241 p
->flags
|= SWP_SYNCHRONOUS_IO
;
3243 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
3245 unsigned long ci
, nr_cluster
;
3247 p
->flags
|= SWP_SOLIDSTATE
;
3248 p
->cluster_next_cpu
= alloc_percpu(unsigned int);
3249 if (!p
->cluster_next_cpu
) {
3251 goto bad_swap_unlock_inode
;
3254 * select a random position to start with to help wear leveling
3257 for_each_possible_cpu(cpu
) {
3258 per_cpu(*p
->cluster_next_cpu
, cpu
) =
3259 1 + prandom_u32_max(p
->highest_bit
);
3261 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3263 cluster_info
= kvcalloc(nr_cluster
, sizeof(*cluster_info
),
3265 if (!cluster_info
) {
3267 goto bad_swap_unlock_inode
;
3270 for (ci
= 0; ci
< nr_cluster
; ci
++)
3271 spin_lock_init(&((cluster_info
+ ci
)->lock
));
3273 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
3274 if (!p
->percpu_cluster
) {
3276 goto bad_swap_unlock_inode
;
3278 for_each_possible_cpu(cpu
) {
3279 struct percpu_cluster
*cluster
;
3280 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
3281 cluster_set_null(&cluster
->index
);
3284 atomic_inc(&nr_rotate_swap
);
3285 inced_nr_rotate_swap
= true;
3288 error
= swap_cgroup_swapon(p
->type
, maxpages
);
3290 goto bad_swap_unlock_inode
;
3292 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
3293 cluster_info
, maxpages
, &span
);
3294 if (unlikely(nr_extents
< 0)) {
3296 goto bad_swap_unlock_inode
;
3298 /* frontswap enabled? set up bit-per-page map for frontswap */
3299 if (IS_ENABLED(CONFIG_FRONTSWAP
))
3300 frontswap_map
= kvcalloc(BITS_TO_LONGS(maxpages
),
3304 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
3306 * When discard is enabled for swap with no particular
3307 * policy flagged, we set all swap discard flags here in
3308 * order to sustain backward compatibility with older
3309 * swapon(8) releases.
3311 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
3315 * By flagging sys_swapon, a sysadmin can tell us to
3316 * either do single-time area discards only, or to just
3317 * perform discards for released swap page-clusters.
3318 * Now it's time to adjust the p->flags accordingly.
3320 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
3321 p
->flags
&= ~SWP_PAGE_DISCARD
;
3322 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
3323 p
->flags
&= ~SWP_AREA_DISCARD
;
3325 /* issue a swapon-time discard if it's still required */
3326 if (p
->flags
& SWP_AREA_DISCARD
) {
3327 int err
= discard_swap(p
);
3329 pr_err("swapon: discard_swap(%p): %d\n",
3334 error
= init_swap_address_space(p
->type
, maxpages
);
3336 goto bad_swap_unlock_inode
;
3339 * Flush any pending IO and dirty mappings before we start using this
3342 inode
->i_flags
|= S_SWAPFILE
;
3343 error
= inode_drain_writes(inode
);
3345 inode
->i_flags
&= ~S_SWAPFILE
;
3346 goto bad_swap_unlock_inode
;
3349 mutex_lock(&swapon_mutex
);
3351 if (swap_flags
& SWAP_FLAG_PREFER
)
3353 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
3354 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
3356 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3357 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
3358 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
3359 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
3360 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
3361 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
3362 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
3363 (frontswap_map
) ? "FS" : "");
3365 mutex_unlock(&swapon_mutex
);
3366 atomic_inc(&proc_poll_event
);
3367 wake_up_interruptible(&proc_poll_wait
);
3371 bad_swap_unlock_inode
:
3372 inode_unlock(inode
);
3374 free_percpu(p
->percpu_cluster
);
3375 p
->percpu_cluster
= NULL
;
3376 free_percpu(p
->cluster_next_cpu
);
3377 p
->cluster_next_cpu
= NULL
;
3378 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
3379 set_blocksize(p
->bdev
, p
->old_block_size
);
3380 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
3383 destroy_swap_extents(p
);
3384 swap_cgroup_swapoff(p
->type
);
3385 spin_lock(&swap_lock
);
3386 p
->swap_file
= NULL
;
3388 spin_unlock(&swap_lock
);
3390 kvfree(cluster_info
);
3391 kvfree(frontswap_map
);
3392 if (inced_nr_rotate_swap
)
3393 atomic_dec(&nr_rotate_swap
);
3395 filp_close(swap_file
, NULL
);
3397 if (page
&& !IS_ERR(page
)) {
3404 inode_unlock(inode
);
3406 enable_swap_slots_cache();
3410 void si_swapinfo(struct sysinfo
*val
)
3413 unsigned long nr_to_be_unused
= 0;
3415 spin_lock(&swap_lock
);
3416 for (type
= 0; type
< nr_swapfiles
; type
++) {
3417 struct swap_info_struct
*si
= swap_info
[type
];
3419 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
3420 nr_to_be_unused
+= si
->inuse_pages
;
3422 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
3423 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
3424 spin_unlock(&swap_lock
);
3428 * Verify that a swap entry is valid and increment its swap map count.
3430 * Returns error code in following case.
3432 * - swp_entry is invalid -> EINVAL
3433 * - swp_entry is migration entry -> EINVAL
3434 * - swap-cache reference is requested but there is already one. -> EEXIST
3435 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3436 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3438 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
3440 struct swap_info_struct
*p
;
3441 struct swap_cluster_info
*ci
;
3442 unsigned long offset
;
3443 unsigned char count
;
3444 unsigned char has_cache
;
3447 p
= get_swap_device(entry
);
3451 offset
= swp_offset(entry
);
3452 ci
= lock_cluster_or_swap_info(p
, offset
);
3454 count
= p
->swap_map
[offset
];
3457 * swapin_readahead() doesn't check if a swap entry is valid, so the
3458 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3460 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
3465 has_cache
= count
& SWAP_HAS_CACHE
;
3466 count
&= ~SWAP_HAS_CACHE
;
3469 if (usage
== SWAP_HAS_CACHE
) {
3471 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3472 if (!has_cache
&& count
)
3473 has_cache
= SWAP_HAS_CACHE
;
3474 else if (has_cache
) /* someone else added cache */
3476 else /* no users remaining */
3479 } else if (count
|| has_cache
) {
3481 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
3483 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
3485 else if (swap_count_continued(p
, offset
, count
))
3486 count
= COUNT_CONTINUED
;
3490 err
= -ENOENT
; /* unused swap entry */
3492 WRITE_ONCE(p
->swap_map
[offset
], count
| has_cache
);
3495 unlock_cluster_or_swap_info(p
, ci
);
3503 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3504 * (in which case its reference count is never incremented).
3506 void swap_shmem_alloc(swp_entry_t entry
)
3508 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
3512 * Increase reference count of swap entry by 1.
3513 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3514 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3515 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3516 * might occur if a page table entry has got corrupted.
3518 int swap_duplicate(swp_entry_t entry
)
3522 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
3523 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
3528 * @entry: swap entry for which we allocate swap cache.
3530 * Called when allocating swap cache for existing swap entry,
3531 * This can return error codes. Returns 0 at success.
3532 * -EEXIST means there is a swap cache.
3533 * Note: return code is different from swap_duplicate().
3535 int swapcache_prepare(swp_entry_t entry
)
3537 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
3540 struct swap_info_struct
*swp_swap_info(swp_entry_t entry
)
3542 return swap_type_to_swap_info(swp_type(entry
));
3545 struct swap_info_struct
*page_swap_info(struct page
*page
)
3547 swp_entry_t entry
= { .val
= page_private(page
) };
3548 return swp_swap_info(entry
);
3552 * out-of-line __page_file_ methods to avoid include hell.
3554 struct address_space
*__page_file_mapping(struct page
*page
)
3556 return page_swap_info(page
)->swap_file
->f_mapping
;
3558 EXPORT_SYMBOL_GPL(__page_file_mapping
);
3560 pgoff_t
__page_file_index(struct page
*page
)
3562 swp_entry_t swap
= { .val
= page_private(page
) };
3563 return swp_offset(swap
);
3565 EXPORT_SYMBOL_GPL(__page_file_index
);
3568 * add_swap_count_continuation - called when a swap count is duplicated
3569 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3570 * page of the original vmalloc'ed swap_map, to hold the continuation count
3571 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3572 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3574 * These continuation pages are seldom referenced: the common paths all work
3575 * on the original swap_map, only referring to a continuation page when the
3576 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3578 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3579 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3580 * can be called after dropping locks.
3582 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
3584 struct swap_info_struct
*si
;
3585 struct swap_cluster_info
*ci
;
3588 struct page
*list_page
;
3590 unsigned char count
;
3594 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3595 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3597 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
3599 si
= get_swap_device(entry
);
3602 * An acceptable race has occurred since the failing
3603 * __swap_duplicate(): the swap device may be swapoff
3607 spin_lock(&si
->lock
);
3609 offset
= swp_offset(entry
);
3611 ci
= lock_cluster(si
, offset
);
3613 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
3615 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
3617 * The higher the swap count, the more likely it is that tasks
3618 * will race to add swap count continuation: we need to avoid
3619 * over-provisioning.
3630 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3631 * no architecture is using highmem pages for kernel page tables: so it
3632 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3634 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3635 offset
&= ~PAGE_MASK
;
3637 spin_lock(&si
->cont_lock
);
3639 * Page allocation does not initialize the page's lru field,
3640 * but it does always reset its private field.
3642 if (!page_private(head
)) {
3643 BUG_ON(count
& COUNT_CONTINUED
);
3644 INIT_LIST_HEAD(&head
->lru
);
3645 set_page_private(head
, SWP_CONTINUED
);
3646 si
->flags
|= SWP_CONTINUED
;
3649 list_for_each_entry(list_page
, &head
->lru
, lru
) {
3653 * If the previous map said no continuation, but we've found
3654 * a continuation page, free our allocation and use this one.
3656 if (!(count
& COUNT_CONTINUED
))
3657 goto out_unlock_cont
;
3659 map
= kmap_atomic(list_page
) + offset
;
3664 * If this continuation count now has some space in it,
3665 * free our allocation and use this one.
3667 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
3668 goto out_unlock_cont
;
3671 list_add_tail(&page
->lru
, &head
->lru
);
3672 page
= NULL
; /* now it's attached, don't free it */
3674 spin_unlock(&si
->cont_lock
);
3677 spin_unlock(&si
->lock
);
3678 put_swap_device(si
);
3686 * swap_count_continued - when the original swap_map count is incremented
3687 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3688 * into, carry if so, or else fail until a new continuation page is allocated;
3689 * when the original swap_map count is decremented from 0 with continuation,
3690 * borrow from the continuation and report whether it still holds more.
3691 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3694 static bool swap_count_continued(struct swap_info_struct
*si
,
3695 pgoff_t offset
, unsigned char count
)
3702 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3703 if (page_private(head
) != SWP_CONTINUED
) {
3704 BUG_ON(count
& COUNT_CONTINUED
);
3705 return false; /* need to add count continuation */
3708 spin_lock(&si
->cont_lock
);
3709 offset
&= ~PAGE_MASK
;
3710 page
= list_next_entry(head
, lru
);
3711 map
= kmap_atomic(page
) + offset
;
3713 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3714 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3716 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3718 * Think of how you add 1 to 999
3720 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3722 page
= list_next_entry(page
, lru
);
3723 BUG_ON(page
== head
);
3724 map
= kmap_atomic(page
) + offset
;
3726 if (*map
== SWAP_CONT_MAX
) {
3728 page
= list_next_entry(page
, lru
);
3730 ret
= false; /* add count continuation */
3733 map
= kmap_atomic(page
) + offset
;
3734 init_map
: *map
= 0; /* we didn't zero the page */
3738 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3739 map
= kmap_atomic(page
) + offset
;
3740 *map
= COUNT_CONTINUED
;
3743 ret
= true; /* incremented */
3745 } else { /* decrementing */
3747 * Think of how you subtract 1 from 1000
3749 BUG_ON(count
!= COUNT_CONTINUED
);
3750 while (*map
== COUNT_CONTINUED
) {
3752 page
= list_next_entry(page
, lru
);
3753 BUG_ON(page
== head
);
3754 map
= kmap_atomic(page
) + offset
;
3761 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3762 map
= kmap_atomic(page
) + offset
;
3763 *map
= SWAP_CONT_MAX
| count
;
3764 count
= COUNT_CONTINUED
;
3767 ret
= count
== COUNT_CONTINUED
;
3770 spin_unlock(&si
->cont_lock
);
3775 * free_swap_count_continuations - swapoff free all the continuation pages
3776 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3778 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3782 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3784 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3785 if (page_private(head
)) {
3786 struct page
*page
, *next
;
3788 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3789 list_del(&page
->lru
);
3796 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3797 void cgroup_throttle_swaprate(struct page
*page
, gfp_t gfp_mask
)
3799 struct swap_info_struct
*si
, *next
;
3800 int nid
= page_to_nid(page
);
3802 if (!(gfp_mask
& __GFP_IO
))
3805 if (!blk_cgroup_congested())
3809 * We've already scheduled a throttle, avoid taking the global swap
3812 if (current
->throttle_queue
)
3815 spin_lock(&swap_avail_lock
);
3816 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[nid
],
3819 blkcg_schedule_throttle(bdev_get_queue(si
->bdev
), true);
3823 spin_unlock(&swap_avail_lock
);
3827 static int __init
swapfile_init(void)
3831 swap_avail_heads
= kmalloc_array(nr_node_ids
, sizeof(struct plist_head
),
3833 if (!swap_avail_heads
) {
3834 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3839 plist_head_init(&swap_avail_heads
[nid
]);
3843 subsys_initcall(swapfile_init
);