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
9 #include <linux/blkdev.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/task.h>
13 #include <linux/hugetlb.h>
14 #include <linux/mman.h>
15 #include <linux/slab.h>
16 #include <linux/kernel_stat.h>
17 #include <linux/swap.h>
18 #include <linux/vmalloc.h>
19 #include <linux/pagemap.h>
20 #include <linux/namei.h>
21 #include <linux/shmem_fs.h>
22 #include <linux/blk-cgroup.h>
23 #include <linux/random.h>
24 #include <linux/writeback.h>
25 #include <linux/proc_fs.h>
26 #include <linux/seq_file.h>
27 #include <linux/init.h>
28 #include <linux/ksm.h>
29 #include <linux/rmap.h>
30 #include <linux/security.h>
31 #include <linux/backing-dev.h>
32 #include <linux/mutex.h>
33 #include <linux/capability.h>
34 #include <linux/syscalls.h>
35 #include <linux/memcontrol.h>
36 #include <linux/poll.h>
37 #include <linux/oom.h>
38 #include <linux/swapfile.h>
39 #include <linux/export.h>
40 #include <linux/swap_slots.h>
41 #include <linux/sort.h>
42 #include <linux/completion.h>
43 #include <linux/suspend.h>
44 #include <linux/zswap.h>
46 #include <asm/tlbflush.h>
47 #include <linux/swapops.h>
48 #include <linux/swap_cgroup.h>
52 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
54 static void free_swap_count_continuations(struct swap_info_struct
*);
56 static DEFINE_SPINLOCK(swap_lock
);
57 static unsigned int nr_swapfiles
;
58 atomic_long_t nr_swap_pages
;
60 * Some modules use swappable objects and may try to swap them out under
61 * memory pressure (via the shrinker). Before doing so, they may wish to
62 * check to see if any swap space is available.
64 EXPORT_SYMBOL_GPL(nr_swap_pages
);
65 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
66 long total_swap_pages
;
67 static int least_priority
= -1;
68 unsigned long swapfile_maximum_size
;
69 #ifdef CONFIG_MIGRATION
70 bool swap_migration_ad_supported
;
71 #endif /* CONFIG_MIGRATION */
73 static const char Bad_file
[] = "Bad swap file entry ";
74 static const char Unused_file
[] = "Unused swap file entry ";
75 static const char Bad_offset
[] = "Bad swap offset entry ";
76 static const char Unused_offset
[] = "Unused swap offset entry ";
79 * all active swap_info_structs
80 * protected with swap_lock, and ordered by priority.
82 static PLIST_HEAD(swap_active_head
);
85 * all available (active, not full) swap_info_structs
86 * protected with swap_avail_lock, ordered by priority.
87 * This is used by folio_alloc_swap() instead of swap_active_head
88 * because swap_active_head includes all swap_info_structs,
89 * but folio_alloc_swap() doesn't need to look at full ones.
90 * This uses its own lock instead of swap_lock because when a
91 * swap_info_struct changes between not-full/full, it needs to
92 * add/remove itself to/from this list, but the swap_info_struct->lock
93 * is held and the locking order requires swap_lock to be taken
94 * before any swap_info_struct->lock.
96 static struct plist_head
*swap_avail_heads
;
97 static DEFINE_SPINLOCK(swap_avail_lock
);
99 static struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
101 static DEFINE_MUTEX(swapon_mutex
);
103 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
104 /* Activity counter to indicate that a swapon or swapoff has occurred */
105 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
107 atomic_t nr_rotate_swap
= ATOMIC_INIT(0);
109 static struct swap_info_struct
*swap_type_to_swap_info(int type
)
111 if (type
>= MAX_SWAPFILES
)
114 return READ_ONCE(swap_info
[type
]); /* rcu_dereference() */
117 static inline unsigned char swap_count(unsigned char ent
)
119 return ent
& ~SWAP_HAS_CACHE
; /* may include COUNT_CONTINUED flag */
122 /* Reclaim the swap entry anyway if possible */
123 #define TTRS_ANYWAY 0x1
125 * Reclaim the swap entry if there are no more mappings of the
128 #define TTRS_UNMAPPED 0x2
129 /* Reclaim the swap entry if swap is getting full*/
130 #define TTRS_FULL 0x4
132 /* returns 1 if swap entry is freed */
133 static int __try_to_reclaim_swap(struct swap_info_struct
*si
,
134 unsigned long offset
, unsigned long flags
)
136 swp_entry_t entry
= swp_entry(si
->type
, offset
);
140 folio
= filemap_get_folio(swap_address_space(entry
), offset
);
144 * When this function is called from scan_swap_map_slots() and it's
145 * called by vmscan.c at reclaiming folios. So we hold a folio lock
146 * here. We have to use trylock for avoiding deadlock. This is a special
147 * case and you should use folio_free_swap() with explicit folio_lock()
148 * in usual operations.
150 if (folio_trylock(folio
)) {
151 if ((flags
& TTRS_ANYWAY
) ||
152 ((flags
& TTRS_UNMAPPED
) && !folio_mapped(folio
)) ||
153 ((flags
& TTRS_FULL
) && mem_cgroup_swap_full(folio
)))
154 ret
= folio_free_swap(folio
);
161 static inline struct swap_extent
*first_se(struct swap_info_struct
*sis
)
163 struct rb_node
*rb
= rb_first(&sis
->swap_extent_root
);
164 return rb_entry(rb
, struct swap_extent
, rb_node
);
167 static inline struct swap_extent
*next_se(struct swap_extent
*se
)
169 struct rb_node
*rb
= rb_next(&se
->rb_node
);
170 return rb
? rb_entry(rb
, struct swap_extent
, rb_node
) : NULL
;
174 * swapon tell device that all the old swap contents can be discarded,
175 * to allow the swap device to optimize its wear-levelling.
177 static int discard_swap(struct swap_info_struct
*si
)
179 struct swap_extent
*se
;
180 sector_t start_block
;
184 /* Do not discard the swap header page! */
186 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
187 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
189 err
= blkdev_issue_discard(si
->bdev
, start_block
,
190 nr_blocks
, GFP_KERNEL
);
196 for (se
= next_se(se
); se
; se
= next_se(se
)) {
197 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
198 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
200 err
= blkdev_issue_discard(si
->bdev
, start_block
,
201 nr_blocks
, GFP_KERNEL
);
207 return err
; /* That will often be -EOPNOTSUPP */
210 static struct swap_extent
*
211 offset_to_swap_extent(struct swap_info_struct
*sis
, unsigned long offset
)
213 struct swap_extent
*se
;
216 rb
= sis
->swap_extent_root
.rb_node
;
218 se
= rb_entry(rb
, struct swap_extent
, rb_node
);
219 if (offset
< se
->start_page
)
221 else if (offset
>= se
->start_page
+ se
->nr_pages
)
226 /* It *must* be present */
230 sector_t
swap_page_sector(struct page
*page
)
232 struct swap_info_struct
*sis
= page_swap_info(page
);
233 struct swap_extent
*se
;
237 offset
= __page_file_index(page
);
238 se
= offset_to_swap_extent(sis
, offset
);
239 sector
= se
->start_block
+ (offset
- se
->start_page
);
240 return sector
<< (PAGE_SHIFT
- 9);
244 * swap allocation tell device that a cluster of swap can now be discarded,
245 * to allow the swap device to optimize its wear-levelling.
247 static void discard_swap_cluster(struct swap_info_struct
*si
,
248 pgoff_t start_page
, pgoff_t nr_pages
)
250 struct swap_extent
*se
= offset_to_swap_extent(si
, start_page
);
253 pgoff_t offset
= start_page
- se
->start_page
;
254 sector_t start_block
= se
->start_block
+ offset
;
255 sector_t nr_blocks
= se
->nr_pages
- offset
;
257 if (nr_blocks
> nr_pages
)
258 nr_blocks
= nr_pages
;
259 start_page
+= nr_blocks
;
260 nr_pages
-= nr_blocks
;
262 start_block
<<= PAGE_SHIFT
- 9;
263 nr_blocks
<<= PAGE_SHIFT
- 9;
264 if (blkdev_issue_discard(si
->bdev
, start_block
,
265 nr_blocks
, GFP_NOIO
))
272 #ifdef CONFIG_THP_SWAP
273 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
275 #define swap_entry_size(size) (size)
277 #define SWAPFILE_CLUSTER 256
280 * Define swap_entry_size() as constant to let compiler to optimize
281 * out some code if !CONFIG_THP_SWAP
283 #define swap_entry_size(size) 1
285 #define LATENCY_LIMIT 256
287 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
293 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
298 static inline void cluster_set_count(struct swap_cluster_info
*info
,
304 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
305 unsigned int c
, unsigned int f
)
311 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
316 static inline void cluster_set_next(struct swap_cluster_info
*info
,
322 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
323 unsigned int n
, unsigned int f
)
329 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
331 return info
->flags
& CLUSTER_FLAG_FREE
;
334 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
336 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
339 static inline void cluster_set_null(struct swap_cluster_info
*info
)
341 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
345 static inline bool cluster_is_huge(struct swap_cluster_info
*info
)
347 if (IS_ENABLED(CONFIG_THP_SWAP
))
348 return info
->flags
& CLUSTER_FLAG_HUGE
;
352 static inline void cluster_clear_huge(struct swap_cluster_info
*info
)
354 info
->flags
&= ~CLUSTER_FLAG_HUGE
;
357 static inline struct swap_cluster_info
*lock_cluster(struct swap_info_struct
*si
,
358 unsigned long offset
)
360 struct swap_cluster_info
*ci
;
362 ci
= si
->cluster_info
;
364 ci
+= offset
/ SWAPFILE_CLUSTER
;
365 spin_lock(&ci
->lock
);
370 static inline void unlock_cluster(struct swap_cluster_info
*ci
)
373 spin_unlock(&ci
->lock
);
377 * Determine the locking method in use for this device. Return
378 * swap_cluster_info if SSD-style cluster-based locking is in place.
380 static inline struct swap_cluster_info
*lock_cluster_or_swap_info(
381 struct swap_info_struct
*si
, unsigned long offset
)
383 struct swap_cluster_info
*ci
;
385 /* Try to use fine-grained SSD-style locking if available: */
386 ci
= lock_cluster(si
, offset
);
387 /* Otherwise, fall back to traditional, coarse locking: */
389 spin_lock(&si
->lock
);
394 static inline void unlock_cluster_or_swap_info(struct swap_info_struct
*si
,
395 struct swap_cluster_info
*ci
)
400 spin_unlock(&si
->lock
);
403 static inline bool cluster_list_empty(struct swap_cluster_list
*list
)
405 return cluster_is_null(&list
->head
);
408 static inline unsigned int cluster_list_first(struct swap_cluster_list
*list
)
410 return cluster_next(&list
->head
);
413 static void cluster_list_init(struct swap_cluster_list
*list
)
415 cluster_set_null(&list
->head
);
416 cluster_set_null(&list
->tail
);
419 static void cluster_list_add_tail(struct swap_cluster_list
*list
,
420 struct swap_cluster_info
*ci
,
423 if (cluster_list_empty(list
)) {
424 cluster_set_next_flag(&list
->head
, idx
, 0);
425 cluster_set_next_flag(&list
->tail
, idx
, 0);
427 struct swap_cluster_info
*ci_tail
;
428 unsigned int tail
= cluster_next(&list
->tail
);
431 * Nested cluster lock, but both cluster locks are
432 * only acquired when we held swap_info_struct->lock
435 spin_lock_nested(&ci_tail
->lock
, SINGLE_DEPTH_NESTING
);
436 cluster_set_next(ci_tail
, idx
);
437 spin_unlock(&ci_tail
->lock
);
438 cluster_set_next_flag(&list
->tail
, idx
, 0);
442 static unsigned int cluster_list_del_first(struct swap_cluster_list
*list
,
443 struct swap_cluster_info
*ci
)
447 idx
= cluster_next(&list
->head
);
448 if (cluster_next(&list
->tail
) == idx
) {
449 cluster_set_null(&list
->head
);
450 cluster_set_null(&list
->tail
);
452 cluster_set_next_flag(&list
->head
,
453 cluster_next(&ci
[idx
]), 0);
458 /* Add a cluster to discard list and schedule it to do discard */
459 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
463 * If scan_swap_map_slots() can't find a free cluster, it will check
464 * si->swap_map directly. To make sure the discarding cluster isn't
465 * taken by scan_swap_map_slots(), mark the swap entries bad (occupied).
466 * It will be cleared after discard
468 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
469 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
471 cluster_list_add_tail(&si
->discard_clusters
, si
->cluster_info
, idx
);
473 schedule_work(&si
->discard_work
);
476 static void __free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
478 struct swap_cluster_info
*ci
= si
->cluster_info
;
480 cluster_set_flag(ci
+ idx
, CLUSTER_FLAG_FREE
);
481 cluster_list_add_tail(&si
->free_clusters
, ci
, idx
);
485 * Doing discard actually. After a cluster discard is finished, the cluster
486 * will be added to free cluster list. caller should hold si->lock.
488 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
490 struct swap_cluster_info
*info
, *ci
;
493 info
= si
->cluster_info
;
495 while (!cluster_list_empty(&si
->discard_clusters
)) {
496 idx
= cluster_list_del_first(&si
->discard_clusters
, info
);
497 spin_unlock(&si
->lock
);
499 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
502 spin_lock(&si
->lock
);
503 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
504 __free_cluster(si
, idx
);
505 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
506 0, SWAPFILE_CLUSTER
);
511 static void swap_discard_work(struct work_struct
*work
)
513 struct swap_info_struct
*si
;
515 si
= container_of(work
, struct swap_info_struct
, discard_work
);
517 spin_lock(&si
->lock
);
518 swap_do_scheduled_discard(si
);
519 spin_unlock(&si
->lock
);
522 static void swap_users_ref_free(struct percpu_ref
*ref
)
524 struct swap_info_struct
*si
;
526 si
= container_of(ref
, struct swap_info_struct
, users
);
530 static void alloc_cluster(struct swap_info_struct
*si
, unsigned long idx
)
532 struct swap_cluster_info
*ci
= si
->cluster_info
;
534 VM_BUG_ON(cluster_list_first(&si
->free_clusters
) != idx
);
535 cluster_list_del_first(&si
->free_clusters
, ci
);
536 cluster_set_count_flag(ci
+ idx
, 0, 0);
539 static void free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
541 struct swap_cluster_info
*ci
= si
->cluster_info
+ idx
;
543 VM_BUG_ON(cluster_count(ci
) != 0);
545 * If the swap is discardable, prepare discard the cluster
546 * instead of free it immediately. The cluster will be freed
549 if ((si
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
550 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
551 swap_cluster_schedule_discard(si
, idx
);
555 __free_cluster(si
, idx
);
559 * The cluster corresponding to page_nr will be used. The cluster will be
560 * removed from free cluster list and its usage counter will be increased.
562 static void inc_cluster_info_page(struct swap_info_struct
*p
,
563 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
565 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
569 if (cluster_is_free(&cluster_info
[idx
]))
570 alloc_cluster(p
, idx
);
572 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
573 cluster_set_count(&cluster_info
[idx
],
574 cluster_count(&cluster_info
[idx
]) + 1);
578 * The cluster corresponding to page_nr decreases one usage. If the usage
579 * counter becomes 0, which means no page in the cluster is in using, we can
580 * optionally discard the cluster and add it to free cluster list.
582 static void dec_cluster_info_page(struct swap_info_struct
*p
,
583 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
585 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
590 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
591 cluster_set_count(&cluster_info
[idx
],
592 cluster_count(&cluster_info
[idx
]) - 1);
594 if (cluster_count(&cluster_info
[idx
]) == 0)
595 free_cluster(p
, idx
);
599 * It's possible scan_swap_map_slots() uses a free cluster in the middle of free
600 * cluster list. Avoiding such abuse to avoid list corruption.
603 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
604 unsigned long offset
)
606 struct percpu_cluster
*percpu_cluster
;
609 offset
/= SWAPFILE_CLUSTER
;
610 conflict
= !cluster_list_empty(&si
->free_clusters
) &&
611 offset
!= cluster_list_first(&si
->free_clusters
) &&
612 cluster_is_free(&si
->cluster_info
[offset
]);
617 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
618 cluster_set_null(&percpu_cluster
->index
);
623 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
624 * might involve allocating a new cluster for current CPU too.
626 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
627 unsigned long *offset
, unsigned long *scan_base
)
629 struct percpu_cluster
*cluster
;
630 struct swap_cluster_info
*ci
;
631 unsigned long tmp
, max
;
634 cluster
= this_cpu_ptr(si
->percpu_cluster
);
635 if (cluster_is_null(&cluster
->index
)) {
636 if (!cluster_list_empty(&si
->free_clusters
)) {
637 cluster
->index
= si
->free_clusters
.head
;
638 cluster
->next
= cluster_next(&cluster
->index
) *
640 } else if (!cluster_list_empty(&si
->discard_clusters
)) {
642 * we don't have free cluster but have some clusters in
643 * discarding, do discard now and reclaim them, then
644 * reread cluster_next_cpu since we dropped si->lock
646 swap_do_scheduled_discard(si
);
647 *scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
648 *offset
= *scan_base
;
655 * Other CPUs can use our cluster if they can't find a free cluster,
656 * check if there is still free entry in the cluster
659 max
= min_t(unsigned long, si
->max
,
660 (cluster_next(&cluster
->index
) + 1) * SWAPFILE_CLUSTER
);
662 ci
= lock_cluster(si
, tmp
);
664 if (!si
->swap_map
[tmp
])
671 cluster_set_null(&cluster
->index
);
674 cluster
->next
= tmp
+ 1;
680 static void __del_from_avail_list(struct swap_info_struct
*p
)
684 assert_spin_locked(&p
->lock
);
686 plist_del(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
689 static void del_from_avail_list(struct swap_info_struct
*p
)
691 spin_lock(&swap_avail_lock
);
692 __del_from_avail_list(p
);
693 spin_unlock(&swap_avail_lock
);
696 static void swap_range_alloc(struct swap_info_struct
*si
, unsigned long offset
,
697 unsigned int nr_entries
)
699 unsigned int end
= offset
+ nr_entries
- 1;
701 if (offset
== si
->lowest_bit
)
702 si
->lowest_bit
+= nr_entries
;
703 if (end
== si
->highest_bit
)
704 WRITE_ONCE(si
->highest_bit
, si
->highest_bit
- nr_entries
);
705 WRITE_ONCE(si
->inuse_pages
, si
->inuse_pages
+ nr_entries
);
706 if (si
->inuse_pages
== si
->pages
) {
707 si
->lowest_bit
= si
->max
;
709 del_from_avail_list(si
);
713 static void add_to_avail_list(struct swap_info_struct
*p
)
717 spin_lock(&swap_avail_lock
);
719 plist_add(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
720 spin_unlock(&swap_avail_lock
);
723 static void swap_range_free(struct swap_info_struct
*si
, unsigned long offset
,
724 unsigned int nr_entries
)
726 unsigned long begin
= offset
;
727 unsigned long end
= offset
+ nr_entries
- 1;
728 void (*swap_slot_free_notify
)(struct block_device
*, unsigned long);
730 if (offset
< si
->lowest_bit
)
731 si
->lowest_bit
= offset
;
732 if (end
> si
->highest_bit
) {
733 bool was_full
= !si
->highest_bit
;
735 WRITE_ONCE(si
->highest_bit
, end
);
736 if (was_full
&& (si
->flags
& SWP_WRITEOK
))
737 add_to_avail_list(si
);
739 atomic_long_add(nr_entries
, &nr_swap_pages
);
740 WRITE_ONCE(si
->inuse_pages
, si
->inuse_pages
- nr_entries
);
741 if (si
->flags
& SWP_BLKDEV
)
742 swap_slot_free_notify
=
743 si
->bdev
->bd_disk
->fops
->swap_slot_free_notify
;
745 swap_slot_free_notify
= NULL
;
746 while (offset
<= end
) {
747 arch_swap_invalidate_page(si
->type
, offset
);
748 zswap_invalidate(si
->type
, offset
);
749 if (swap_slot_free_notify
)
750 swap_slot_free_notify(si
->bdev
, offset
);
753 clear_shadow_from_swap_cache(si
->type
, begin
, end
);
756 static void set_cluster_next(struct swap_info_struct
*si
, unsigned long next
)
760 if (!(si
->flags
& SWP_SOLIDSTATE
)) {
761 si
->cluster_next
= next
;
765 prev
= this_cpu_read(*si
->cluster_next_cpu
);
767 * Cross the swap address space size aligned trunk, choose
768 * another trunk randomly to avoid lock contention on swap
769 * address space if possible.
771 if ((prev
>> SWAP_ADDRESS_SPACE_SHIFT
) !=
772 (next
>> SWAP_ADDRESS_SPACE_SHIFT
)) {
773 /* No free swap slots available */
774 if (si
->highest_bit
<= si
->lowest_bit
)
776 next
= get_random_u32_inclusive(si
->lowest_bit
, si
->highest_bit
);
777 next
= ALIGN_DOWN(next
, SWAP_ADDRESS_SPACE_PAGES
);
778 next
= max_t(unsigned int, next
, si
->lowest_bit
);
780 this_cpu_write(*si
->cluster_next_cpu
, next
);
783 static bool swap_offset_available_and_locked(struct swap_info_struct
*si
,
784 unsigned long offset
)
786 if (data_race(!si
->swap_map
[offset
])) {
787 spin_lock(&si
->lock
);
791 if (vm_swap_full() && READ_ONCE(si
->swap_map
[offset
]) == SWAP_HAS_CACHE
) {
792 spin_lock(&si
->lock
);
799 static int scan_swap_map_slots(struct swap_info_struct
*si
,
800 unsigned char usage
, int nr
,
803 struct swap_cluster_info
*ci
;
804 unsigned long offset
;
805 unsigned long scan_base
;
806 unsigned long last_in_cluster
= 0;
807 int latency_ration
= LATENCY_LIMIT
;
809 bool scanned_many
= false;
812 * We try to cluster swap pages by allocating them sequentially
813 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
814 * way, however, we resort to first-free allocation, starting
815 * a new cluster. This prevents us from scattering swap pages
816 * all over the entire swap partition, so that we reduce
817 * overall disk seek times between swap pages. -- sct
818 * But we do now try to find an empty cluster. -Andrea
819 * And we let swap pages go all over an SSD partition. Hugh
822 si
->flags
+= SWP_SCANNING
;
824 * Use percpu scan base for SSD to reduce lock contention on
825 * cluster and swap cache. For HDD, sequential access is more
828 if (si
->flags
& SWP_SOLIDSTATE
)
829 scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
831 scan_base
= si
->cluster_next
;
835 if (si
->cluster_info
) {
836 if (!scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
838 } else if (unlikely(!si
->cluster_nr
--)) {
839 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
840 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
844 spin_unlock(&si
->lock
);
847 * If seek is expensive, start searching for new cluster from
848 * start of partition, to minimize the span of allocated swap.
849 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
850 * case, just handled by scan_swap_map_try_ssd_cluster() above.
852 scan_base
= offset
= si
->lowest_bit
;
853 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
855 /* Locate the first empty (unaligned) cluster */
856 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
857 if (si
->swap_map
[offset
])
858 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
859 else if (offset
== last_in_cluster
) {
860 spin_lock(&si
->lock
);
861 offset
-= SWAPFILE_CLUSTER
- 1;
862 si
->cluster_next
= offset
;
863 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
866 if (unlikely(--latency_ration
< 0)) {
868 latency_ration
= LATENCY_LIMIT
;
873 spin_lock(&si
->lock
);
874 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
878 if (si
->cluster_info
) {
879 while (scan_swap_map_ssd_cluster_conflict(si
, offset
)) {
880 /* take a break if we already got some slots */
883 if (!scan_swap_map_try_ssd_cluster(si
, &offset
,
888 if (!(si
->flags
& SWP_WRITEOK
))
890 if (!si
->highest_bit
)
892 if (offset
> si
->highest_bit
)
893 scan_base
= offset
= si
->lowest_bit
;
895 ci
= lock_cluster(si
, offset
);
896 /* reuse swap entry of cache-only swap if not busy. */
897 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
900 spin_unlock(&si
->lock
);
901 swap_was_freed
= __try_to_reclaim_swap(si
, offset
, TTRS_ANYWAY
);
902 spin_lock(&si
->lock
);
903 /* entry was freed successfully, try to use this again */
906 goto scan
; /* check next one */
909 if (si
->swap_map
[offset
]) {
916 WRITE_ONCE(si
->swap_map
[offset
], usage
);
917 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
920 swap_range_alloc(si
, offset
, 1);
921 slots
[n_ret
++] = swp_entry(si
->type
, offset
);
923 /* got enough slots or reach max slots? */
924 if ((n_ret
== nr
) || (offset
>= si
->highest_bit
))
927 /* search for next available slot */
929 /* time to take a break? */
930 if (unlikely(--latency_ration
< 0)) {
933 spin_unlock(&si
->lock
);
935 spin_lock(&si
->lock
);
936 latency_ration
= LATENCY_LIMIT
;
939 /* try to get more slots in cluster */
940 if (si
->cluster_info
) {
941 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
943 } else if (si
->cluster_nr
&& !si
->swap_map
[++offset
]) {
944 /* non-ssd case, still more slots in cluster? */
950 * Even if there's no free clusters available (fragmented),
951 * try to scan a little more quickly with lock held unless we
952 * have scanned too many slots already.
955 unsigned long scan_limit
;
957 if (offset
< scan_base
)
958 scan_limit
= scan_base
;
960 scan_limit
= si
->highest_bit
;
961 for (; offset
<= scan_limit
&& --latency_ration
> 0;
963 if (!si
->swap_map
[offset
])
969 set_cluster_next(si
, offset
+ 1);
970 si
->flags
-= SWP_SCANNING
;
974 spin_unlock(&si
->lock
);
975 while (++offset
<= READ_ONCE(si
->highest_bit
)) {
976 if (unlikely(--latency_ration
< 0)) {
978 latency_ration
= LATENCY_LIMIT
;
981 if (swap_offset_available_and_locked(si
, offset
))
984 offset
= si
->lowest_bit
;
985 while (offset
< scan_base
) {
986 if (unlikely(--latency_ration
< 0)) {
988 latency_ration
= LATENCY_LIMIT
;
991 if (swap_offset_available_and_locked(si
, offset
))
995 spin_lock(&si
->lock
);
998 si
->flags
-= SWP_SCANNING
;
1002 static int swap_alloc_cluster(struct swap_info_struct
*si
, swp_entry_t
*slot
)
1005 struct swap_cluster_info
*ci
;
1006 unsigned long offset
;
1009 * Should not even be attempting cluster allocations when huge
1010 * page swap is disabled. Warn and fail the allocation.
1012 if (!IS_ENABLED(CONFIG_THP_SWAP
)) {
1017 if (cluster_list_empty(&si
->free_clusters
))
1020 idx
= cluster_list_first(&si
->free_clusters
);
1021 offset
= idx
* SWAPFILE_CLUSTER
;
1022 ci
= lock_cluster(si
, offset
);
1023 alloc_cluster(si
, idx
);
1024 cluster_set_count_flag(ci
, SWAPFILE_CLUSTER
, CLUSTER_FLAG_HUGE
);
1026 memset(si
->swap_map
+ offset
, SWAP_HAS_CACHE
, SWAPFILE_CLUSTER
);
1028 swap_range_alloc(si
, offset
, SWAPFILE_CLUSTER
);
1029 *slot
= swp_entry(si
->type
, offset
);
1034 static void swap_free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
1036 unsigned long offset
= idx
* SWAPFILE_CLUSTER
;
1037 struct swap_cluster_info
*ci
;
1039 ci
= lock_cluster(si
, offset
);
1040 memset(si
->swap_map
+ offset
, 0, SWAPFILE_CLUSTER
);
1041 cluster_set_count_flag(ci
, 0, 0);
1042 free_cluster(si
, idx
);
1044 swap_range_free(si
, offset
, SWAPFILE_CLUSTER
);
1047 int get_swap_pages(int n_goal
, swp_entry_t swp_entries
[], int entry_size
)
1049 unsigned long size
= swap_entry_size(entry_size
);
1050 struct swap_info_struct
*si
, *next
;
1055 /* Only single cluster request supported */
1056 WARN_ON_ONCE(n_goal
> 1 && size
== SWAPFILE_CLUSTER
);
1058 spin_lock(&swap_avail_lock
);
1060 avail_pgs
= atomic_long_read(&nr_swap_pages
) / size
;
1061 if (avail_pgs
<= 0) {
1062 spin_unlock(&swap_avail_lock
);
1066 n_goal
= min3((long)n_goal
, (long)SWAP_BATCH
, avail_pgs
);
1068 atomic_long_sub(n_goal
* size
, &nr_swap_pages
);
1071 node
= numa_node_id();
1072 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[node
], avail_lists
[node
]) {
1073 /* requeue si to after same-priority siblings */
1074 plist_requeue(&si
->avail_lists
[node
], &swap_avail_heads
[node
]);
1075 spin_unlock(&swap_avail_lock
);
1076 spin_lock(&si
->lock
);
1077 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
1078 spin_lock(&swap_avail_lock
);
1079 if (plist_node_empty(&si
->avail_lists
[node
])) {
1080 spin_unlock(&si
->lock
);
1083 WARN(!si
->highest_bit
,
1084 "swap_info %d in list but !highest_bit\n",
1086 WARN(!(si
->flags
& SWP_WRITEOK
),
1087 "swap_info %d in list but !SWP_WRITEOK\n",
1089 __del_from_avail_list(si
);
1090 spin_unlock(&si
->lock
);
1093 if (size
== SWAPFILE_CLUSTER
) {
1094 if (si
->flags
& SWP_BLKDEV
)
1095 n_ret
= swap_alloc_cluster(si
, swp_entries
);
1097 n_ret
= scan_swap_map_slots(si
, SWAP_HAS_CACHE
,
1098 n_goal
, swp_entries
);
1099 spin_unlock(&si
->lock
);
1100 if (n_ret
|| size
== SWAPFILE_CLUSTER
)
1104 spin_lock(&swap_avail_lock
);
1107 * if we got here, it's likely that si was almost full before,
1108 * and since scan_swap_map_slots() can drop the si->lock,
1109 * multiple callers probably all tried to get a page from the
1110 * same si and it filled up before we could get one; or, the si
1111 * filled up between us dropping swap_avail_lock and taking
1112 * si->lock. Since we dropped the swap_avail_lock, the
1113 * swap_avail_head list may have been modified; so if next is
1114 * still in the swap_avail_head list then try it, otherwise
1115 * start over if we have not gotten any slots.
1117 if (plist_node_empty(&next
->avail_lists
[node
]))
1121 spin_unlock(&swap_avail_lock
);
1125 atomic_long_add((long)(n_goal
- n_ret
) * size
,
1131 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
1133 struct swap_info_struct
*p
;
1134 unsigned long offset
;
1138 p
= swp_swap_info(entry
);
1141 if (data_race(!(p
->flags
& SWP_USED
)))
1143 offset
= swp_offset(entry
);
1144 if (offset
>= p
->max
)
1146 if (data_race(!p
->swap_map
[swp_offset(entry
)]))
1151 pr_err("%s: %s%08lx\n", __func__
, Unused_offset
, entry
.val
);
1154 pr_err("%s: %s%08lx\n", __func__
, Bad_offset
, entry
.val
);
1157 pr_err("%s: %s%08lx\n", __func__
, Unused_file
, entry
.val
);
1160 pr_err("%s: %s%08lx\n", __func__
, Bad_file
, entry
.val
);
1165 static struct swap_info_struct
*swap_info_get_cont(swp_entry_t entry
,
1166 struct swap_info_struct
*q
)
1168 struct swap_info_struct
*p
;
1170 p
= _swap_info_get(entry
);
1174 spin_unlock(&q
->lock
);
1176 spin_lock(&p
->lock
);
1181 static unsigned char __swap_entry_free_locked(struct swap_info_struct
*p
,
1182 unsigned long offset
,
1183 unsigned char usage
)
1185 unsigned char count
;
1186 unsigned char has_cache
;
1188 count
= p
->swap_map
[offset
];
1190 has_cache
= count
& SWAP_HAS_CACHE
;
1191 count
&= ~SWAP_HAS_CACHE
;
1193 if (usage
== SWAP_HAS_CACHE
) {
1194 VM_BUG_ON(!has_cache
);
1196 } else if (count
== SWAP_MAP_SHMEM
) {
1198 * Or we could insist on shmem.c using a special
1199 * swap_shmem_free() and free_shmem_swap_and_cache()...
1202 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
1203 if (count
== COUNT_CONTINUED
) {
1204 if (swap_count_continued(p
, offset
, count
))
1205 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
1207 count
= SWAP_MAP_MAX
;
1212 usage
= count
| has_cache
;
1214 WRITE_ONCE(p
->swap_map
[offset
], usage
);
1216 WRITE_ONCE(p
->swap_map
[offset
], SWAP_HAS_CACHE
);
1222 * When we get a swap entry, if there aren't some other ways to
1223 * prevent swapoff, such as the folio in swap cache is locked, page
1224 * table lock is held, etc., the swap entry may become invalid because
1225 * of swapoff. Then, we need to enclose all swap related functions
1226 * with get_swap_device() and put_swap_device(), unless the swap
1227 * functions call get/put_swap_device() by themselves.
1229 * Check whether swap entry is valid in the swap device. If so,
1230 * return pointer to swap_info_struct, and keep the swap entry valid
1231 * via preventing the swap device from being swapoff, until
1232 * put_swap_device() is called. Otherwise return NULL.
1234 * Notice that swapoff or swapoff+swapon can still happen before the
1235 * percpu_ref_tryget_live() in get_swap_device() or after the
1236 * percpu_ref_put() in put_swap_device() if there isn't any other way
1237 * to prevent swapoff. The caller must be prepared for that. For
1238 * example, the following situation is possible.
1242 * ... swapoff+swapon
1243 * __read_swap_cache_async()
1244 * swapcache_prepare()
1245 * __swap_duplicate()
1247 * // verify PTE not changed
1249 * In __swap_duplicate(), the swap_map need to be checked before
1250 * changing partly because the specified swap entry may be for another
1251 * swap device which has been swapoff. And in do_swap_page(), after
1252 * the page is read from the swap device, the PTE is verified not
1253 * changed with the page table locked to check whether the swap device
1254 * has been swapoff or swapoff+swapon.
1256 struct swap_info_struct
*get_swap_device(swp_entry_t entry
)
1258 struct swap_info_struct
*si
;
1259 unsigned long offset
;
1263 si
= swp_swap_info(entry
);
1266 if (!percpu_ref_tryget_live(&si
->users
))
1269 * Guarantee the si->users are checked before accessing other
1270 * fields of swap_info_struct.
1272 * Paired with the spin_unlock() after setup_swap_info() in
1273 * enable_swap_info().
1276 offset
= swp_offset(entry
);
1277 if (offset
>= si
->max
)
1282 pr_err("%s: %s%08lx\n", __func__
, Bad_file
, entry
.val
);
1286 pr_err("%s: %s%08lx\n", __func__
, Bad_offset
, entry
.val
);
1287 percpu_ref_put(&si
->users
);
1291 static unsigned char __swap_entry_free(struct swap_info_struct
*p
,
1294 struct swap_cluster_info
*ci
;
1295 unsigned long offset
= swp_offset(entry
);
1296 unsigned char usage
;
1298 ci
= lock_cluster_or_swap_info(p
, offset
);
1299 usage
= __swap_entry_free_locked(p
, offset
, 1);
1300 unlock_cluster_or_swap_info(p
, ci
);
1302 free_swap_slot(entry
);
1307 static void swap_entry_free(struct swap_info_struct
*p
, swp_entry_t entry
)
1309 struct swap_cluster_info
*ci
;
1310 unsigned long offset
= swp_offset(entry
);
1311 unsigned char count
;
1313 ci
= lock_cluster(p
, offset
);
1314 count
= p
->swap_map
[offset
];
1315 VM_BUG_ON(count
!= SWAP_HAS_CACHE
);
1316 p
->swap_map
[offset
] = 0;
1317 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
1320 mem_cgroup_uncharge_swap(entry
, 1);
1321 swap_range_free(p
, offset
, 1);
1325 * Caller has made sure that the swap device corresponding to entry
1326 * is still around or has not been recycled.
1328 void swap_free(swp_entry_t entry
)
1330 struct swap_info_struct
*p
;
1332 p
= _swap_info_get(entry
);
1334 __swap_entry_free(p
, entry
);
1338 * Called after dropping swapcache to decrease refcnt to swap entries.
1340 void put_swap_folio(struct folio
*folio
, swp_entry_t entry
)
1342 unsigned long offset
= swp_offset(entry
);
1343 unsigned long idx
= offset
/ SWAPFILE_CLUSTER
;
1344 struct swap_cluster_info
*ci
;
1345 struct swap_info_struct
*si
;
1347 unsigned int i
, free_entries
= 0;
1349 int size
= swap_entry_size(folio_nr_pages(folio
));
1351 si
= _swap_info_get(entry
);
1355 ci
= lock_cluster_or_swap_info(si
, offset
);
1356 if (size
== SWAPFILE_CLUSTER
) {
1357 VM_BUG_ON(!cluster_is_huge(ci
));
1358 map
= si
->swap_map
+ offset
;
1359 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1361 VM_BUG_ON(!(val
& SWAP_HAS_CACHE
));
1362 if (val
== SWAP_HAS_CACHE
)
1365 cluster_clear_huge(ci
);
1366 if (free_entries
== SWAPFILE_CLUSTER
) {
1367 unlock_cluster_or_swap_info(si
, ci
);
1368 spin_lock(&si
->lock
);
1369 mem_cgroup_uncharge_swap(entry
, SWAPFILE_CLUSTER
);
1370 swap_free_cluster(si
, idx
);
1371 spin_unlock(&si
->lock
);
1375 for (i
= 0; i
< size
; i
++, entry
.val
++) {
1376 if (!__swap_entry_free_locked(si
, offset
+ i
, SWAP_HAS_CACHE
)) {
1377 unlock_cluster_or_swap_info(si
, ci
);
1378 free_swap_slot(entry
);
1381 lock_cluster_or_swap_info(si
, offset
);
1384 unlock_cluster_or_swap_info(si
, ci
);
1387 #ifdef CONFIG_THP_SWAP
1388 int split_swap_cluster(swp_entry_t entry
)
1390 struct swap_info_struct
*si
;
1391 struct swap_cluster_info
*ci
;
1392 unsigned long offset
= swp_offset(entry
);
1394 si
= _swap_info_get(entry
);
1397 ci
= lock_cluster(si
, offset
);
1398 cluster_clear_huge(ci
);
1404 static int swp_entry_cmp(const void *ent1
, const void *ent2
)
1406 const swp_entry_t
*e1
= ent1
, *e2
= ent2
;
1408 return (int)swp_type(*e1
) - (int)swp_type(*e2
);
1411 void swapcache_free_entries(swp_entry_t
*entries
, int n
)
1413 struct swap_info_struct
*p
, *prev
;
1423 * Sort swap entries by swap device, so each lock is only taken once.
1424 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1425 * so low that it isn't necessary to optimize further.
1427 if (nr_swapfiles
> 1)
1428 sort(entries
, n
, sizeof(entries
[0]), swp_entry_cmp
, NULL
);
1429 for (i
= 0; i
< n
; ++i
) {
1430 p
= swap_info_get_cont(entries
[i
], prev
);
1432 swap_entry_free(p
, entries
[i
]);
1436 spin_unlock(&p
->lock
);
1439 int __swap_count(swp_entry_t entry
)
1441 struct swap_info_struct
*si
= swp_swap_info(entry
);
1442 pgoff_t offset
= swp_offset(entry
);
1444 return swap_count(si
->swap_map
[offset
]);
1448 * How many references to @entry are currently swapped out?
1449 * This does not give an exact answer when swap count is continued,
1450 * but does include the high COUNT_CONTINUED flag to allow for that.
1452 int swap_swapcount(struct swap_info_struct
*si
, swp_entry_t entry
)
1454 pgoff_t offset
= swp_offset(entry
);
1455 struct swap_cluster_info
*ci
;
1458 ci
= lock_cluster_or_swap_info(si
, offset
);
1459 count
= swap_count(si
->swap_map
[offset
]);
1460 unlock_cluster_or_swap_info(si
, ci
);
1465 * How many references to @entry are currently swapped out?
1466 * This considers COUNT_CONTINUED so it returns exact answer.
1468 int swp_swapcount(swp_entry_t entry
)
1470 int count
, tmp_count
, n
;
1471 struct swap_info_struct
*p
;
1472 struct swap_cluster_info
*ci
;
1477 p
= _swap_info_get(entry
);
1481 offset
= swp_offset(entry
);
1483 ci
= lock_cluster_or_swap_info(p
, offset
);
1485 count
= swap_count(p
->swap_map
[offset
]);
1486 if (!(count
& COUNT_CONTINUED
))
1489 count
&= ~COUNT_CONTINUED
;
1490 n
= SWAP_MAP_MAX
+ 1;
1492 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1493 offset
&= ~PAGE_MASK
;
1494 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1497 page
= list_next_entry(page
, lru
);
1498 map
= kmap_atomic(page
);
1499 tmp_count
= map
[offset
];
1502 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1503 n
*= (SWAP_CONT_MAX
+ 1);
1504 } while (tmp_count
& COUNT_CONTINUED
);
1506 unlock_cluster_or_swap_info(p
, ci
);
1510 static bool swap_page_trans_huge_swapped(struct swap_info_struct
*si
,
1513 struct swap_cluster_info
*ci
;
1514 unsigned char *map
= si
->swap_map
;
1515 unsigned long roffset
= swp_offset(entry
);
1516 unsigned long offset
= round_down(roffset
, SWAPFILE_CLUSTER
);
1520 ci
= lock_cluster_or_swap_info(si
, offset
);
1521 if (!ci
|| !cluster_is_huge(ci
)) {
1522 if (swap_count(map
[roffset
]))
1526 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1527 if (swap_count(map
[offset
+ i
])) {
1533 unlock_cluster_or_swap_info(si
, ci
);
1537 static bool folio_swapped(struct folio
*folio
)
1539 swp_entry_t entry
= folio
->swap
;
1540 struct swap_info_struct
*si
= _swap_info_get(entry
);
1545 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!folio_test_large(folio
)))
1546 return swap_swapcount(si
, entry
) != 0;
1548 return swap_page_trans_huge_swapped(si
, entry
);
1552 * folio_free_swap() - Free the swap space used for this folio.
1553 * @folio: The folio to remove.
1555 * If swap is getting full, or if there are no more mappings of this folio,
1556 * then call folio_free_swap to free its swap space.
1558 * Return: true if we were able to release the swap space.
1560 bool folio_free_swap(struct folio
*folio
)
1562 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
1564 if (!folio_test_swapcache(folio
))
1566 if (folio_test_writeback(folio
))
1568 if (folio_swapped(folio
))
1572 * Once hibernation has begun to create its image of memory,
1573 * there's a danger that one of the calls to folio_free_swap()
1574 * - most probably a call from __try_to_reclaim_swap() while
1575 * hibernation is allocating its own swap pages for the image,
1576 * but conceivably even a call from memory reclaim - will free
1577 * the swap from a folio which has already been recorded in the
1578 * image as a clean swapcache folio, and then reuse its swap for
1579 * another page of the image. On waking from hibernation, the
1580 * original folio might be freed under memory pressure, then
1581 * later read back in from swap, now with the wrong data.
1583 * Hibernation suspends storage while it is writing the image
1584 * to disk so check that here.
1586 if (pm_suspended_storage())
1589 delete_from_swap_cache(folio
);
1590 folio_set_dirty(folio
);
1595 * Free the swap entry like above, but also try to
1596 * free the page cache entry if it is the last user.
1598 int free_swap_and_cache(swp_entry_t entry
)
1600 struct swap_info_struct
*p
;
1601 unsigned char count
;
1603 if (non_swap_entry(entry
))
1606 p
= _swap_info_get(entry
);
1608 count
= __swap_entry_free(p
, entry
);
1609 if (count
== SWAP_HAS_CACHE
&&
1610 !swap_page_trans_huge_swapped(p
, entry
))
1611 __try_to_reclaim_swap(p
, swp_offset(entry
),
1612 TTRS_UNMAPPED
| TTRS_FULL
);
1617 #ifdef CONFIG_HIBERNATION
1619 swp_entry_t
get_swap_page_of_type(int type
)
1621 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1622 swp_entry_t entry
= {0};
1627 /* This is called for allocating swap entry, not cache */
1628 spin_lock(&si
->lock
);
1629 if ((si
->flags
& SWP_WRITEOK
) && scan_swap_map_slots(si
, 1, 1, &entry
))
1630 atomic_long_dec(&nr_swap_pages
);
1631 spin_unlock(&si
->lock
);
1637 * Find the swap type that corresponds to given device (if any).
1639 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1640 * from 0, in which the swap header is expected to be located.
1642 * This is needed for the suspend to disk (aka swsusp).
1644 int swap_type_of(dev_t device
, sector_t offset
)
1651 spin_lock(&swap_lock
);
1652 for (type
= 0; type
< nr_swapfiles
; type
++) {
1653 struct swap_info_struct
*sis
= swap_info
[type
];
1655 if (!(sis
->flags
& SWP_WRITEOK
))
1658 if (device
== sis
->bdev
->bd_dev
) {
1659 struct swap_extent
*se
= first_se(sis
);
1661 if (se
->start_block
== offset
) {
1662 spin_unlock(&swap_lock
);
1667 spin_unlock(&swap_lock
);
1671 int find_first_swap(dev_t
*device
)
1675 spin_lock(&swap_lock
);
1676 for (type
= 0; type
< nr_swapfiles
; type
++) {
1677 struct swap_info_struct
*sis
= swap_info
[type
];
1679 if (!(sis
->flags
& SWP_WRITEOK
))
1681 *device
= sis
->bdev
->bd_dev
;
1682 spin_unlock(&swap_lock
);
1685 spin_unlock(&swap_lock
);
1690 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1691 * corresponding to given index in swap_info (swap type).
1693 sector_t
swapdev_block(int type
, pgoff_t offset
)
1695 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1696 struct swap_extent
*se
;
1698 if (!si
|| !(si
->flags
& SWP_WRITEOK
))
1700 se
= offset_to_swap_extent(si
, offset
);
1701 return se
->start_block
+ (offset
- se
->start_page
);
1705 * Return either the total number of swap pages of given type, or the number
1706 * of free pages of that type (depending on @free)
1708 * This is needed for software suspend
1710 unsigned int count_swap_pages(int type
, int free
)
1714 spin_lock(&swap_lock
);
1715 if ((unsigned int)type
< nr_swapfiles
) {
1716 struct swap_info_struct
*sis
= swap_info
[type
];
1718 spin_lock(&sis
->lock
);
1719 if (sis
->flags
& SWP_WRITEOK
) {
1722 n
-= sis
->inuse_pages
;
1724 spin_unlock(&sis
->lock
);
1726 spin_unlock(&swap_lock
);
1729 #endif /* CONFIG_HIBERNATION */
1731 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1733 return pte_same(pte_swp_clear_flags(pte
), swp_pte
);
1737 * No need to decide whether this PTE shares the swap entry with others,
1738 * just let do_wp_page work it out if a write is requested later - to
1739 * force COW, vm_page_prot omits write permission from any private vma.
1741 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1742 unsigned long addr
, swp_entry_t entry
, struct folio
*folio
)
1744 struct page
*page
= folio_file_page(folio
, swp_offset(entry
));
1745 struct page
*swapcache
;
1747 pte_t
*pte
, new_pte
, old_pte
;
1748 bool hwpoisoned
= PageHWPoison(page
);
1752 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1753 if (unlikely(!page
))
1755 else if (unlikely(PTR_ERR(page
) == -EHWPOISON
))
1758 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1759 if (unlikely(!pte
|| !pte_same_as_swp(ptep_get(pte
),
1760 swp_entry_to_pte(entry
)))) {
1765 old_pte
= ptep_get(pte
);
1767 if (unlikely(hwpoisoned
|| !PageUptodate(page
))) {
1768 swp_entry_t swp_entry
;
1770 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1772 swp_entry
= make_hwpoison_entry(swapcache
);
1775 swp_entry
= make_poisoned_swp_entry();
1777 new_pte
= swp_entry_to_pte(swp_entry
);
1783 * Some architectures may have to restore extra metadata to the page
1784 * when reading from swap. This metadata may be indexed by swap entry
1785 * so this must be called before swap_free().
1787 arch_swap_restore(entry
, page_folio(page
));
1789 /* See do_swap_page() */
1790 BUG_ON(!PageAnon(page
) && PageMappedToDisk(page
));
1791 BUG_ON(PageAnon(page
) && PageAnonExclusive(page
));
1793 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1794 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1796 if (page
== swapcache
) {
1797 rmap_t rmap_flags
= RMAP_NONE
;
1800 * See do_swap_page(): PageWriteback() would be problematic.
1801 * However, we do a wait_on_page_writeback() just before this
1802 * call and have the page locked.
1804 VM_BUG_ON_PAGE(PageWriteback(page
), page
);
1805 if (pte_swp_exclusive(old_pte
))
1806 rmap_flags
|= RMAP_EXCLUSIVE
;
1808 page_add_anon_rmap(page
, vma
, addr
, rmap_flags
);
1809 } else { /* ksm created a completely new copy */
1810 page_add_new_anon_rmap(page
, vma
, addr
);
1811 lru_cache_add_inactive_or_unevictable(page
, vma
);
1813 new_pte
= pte_mkold(mk_pte(page
, vma
->vm_page_prot
));
1814 if (pte_swp_soft_dirty(old_pte
))
1815 new_pte
= pte_mksoft_dirty(new_pte
);
1816 if (pte_swp_uffd_wp(old_pte
))
1817 new_pte
= pte_mkuffd_wp(new_pte
);
1819 set_pte_at(vma
->vm_mm
, addr
, pte
, new_pte
);
1823 pte_unmap_unlock(pte
, ptl
);
1824 if (page
!= swapcache
) {
1831 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1832 unsigned long addr
, unsigned long end
,
1836 struct swap_info_struct
*si
;
1838 si
= swap_info
[type
];
1840 struct folio
*folio
;
1841 unsigned long offset
;
1842 unsigned char swp_count
;
1848 pte
= pte_offset_map(pmd
, addr
);
1853 ptent
= ptep_get_lockless(pte
);
1855 if (!is_swap_pte(ptent
))
1858 entry
= pte_to_swp_entry(ptent
);
1859 if (swp_type(entry
) != type
)
1862 offset
= swp_offset(entry
);
1866 folio
= swap_cache_get_folio(entry
, vma
, addr
);
1869 struct vm_fault vmf
= {
1872 .real_address
= addr
,
1876 page
= swapin_readahead(entry
, GFP_HIGHUSER_MOVABLE
,
1879 folio
= page_folio(page
);
1882 swp_count
= READ_ONCE(si
->swap_map
[offset
]);
1883 if (swp_count
== 0 || swp_count
== SWAP_MAP_BAD
)
1889 folio_wait_writeback(folio
);
1890 ret
= unuse_pte(vma
, pmd
, addr
, entry
, folio
);
1892 folio_unlock(folio
);
1897 folio_free_swap(folio
);
1898 folio_unlock(folio
);
1900 } while (addr
+= PAGE_SIZE
, addr
!= end
);
1907 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
1908 unsigned long addr
, unsigned long end
,
1915 pmd
= pmd_offset(pud
, addr
);
1918 next
= pmd_addr_end(addr
, end
);
1919 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, type
);
1922 } while (pmd
++, addr
= next
, addr
!= end
);
1926 static inline int unuse_pud_range(struct vm_area_struct
*vma
, p4d_t
*p4d
,
1927 unsigned long addr
, unsigned long end
,
1934 pud
= pud_offset(p4d
, addr
);
1936 next
= pud_addr_end(addr
, end
);
1937 if (pud_none_or_clear_bad(pud
))
1939 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, type
);
1942 } while (pud
++, addr
= next
, addr
!= end
);
1946 static inline int unuse_p4d_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
1947 unsigned long addr
, unsigned long end
,
1954 p4d
= p4d_offset(pgd
, addr
);
1956 next
= p4d_addr_end(addr
, end
);
1957 if (p4d_none_or_clear_bad(p4d
))
1959 ret
= unuse_pud_range(vma
, p4d
, addr
, next
, type
);
1962 } while (p4d
++, addr
= next
, addr
!= end
);
1966 static int unuse_vma(struct vm_area_struct
*vma
, unsigned int type
)
1969 unsigned long addr
, end
, next
;
1972 addr
= vma
->vm_start
;
1975 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1977 next
= pgd_addr_end(addr
, end
);
1978 if (pgd_none_or_clear_bad(pgd
))
1980 ret
= unuse_p4d_range(vma
, pgd
, addr
, next
, type
);
1983 } while (pgd
++, addr
= next
, addr
!= end
);
1987 static int unuse_mm(struct mm_struct
*mm
, unsigned int type
)
1989 struct vm_area_struct
*vma
;
1991 VMA_ITERATOR(vmi
, mm
, 0);
1994 for_each_vma(vmi
, vma
) {
1995 if (vma
->anon_vma
) {
1996 ret
= unuse_vma(vma
, type
);
2003 mmap_read_unlock(mm
);
2008 * Scan swap_map from current position to next entry still in use.
2009 * Return 0 if there are no inuse entries after prev till end of
2012 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
2016 unsigned char count
;
2019 * No need for swap_lock here: we're just looking
2020 * for whether an entry is in use, not modifying it; false
2021 * hits are okay, and sys_swapoff() has already prevented new
2022 * allocations from this area (while holding swap_lock).
2024 for (i
= prev
+ 1; i
< si
->max
; i
++) {
2025 count
= READ_ONCE(si
->swap_map
[i
]);
2026 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
2028 if ((i
% LATENCY_LIMIT
) == 0)
2038 static int try_to_unuse(unsigned int type
)
2040 struct mm_struct
*prev_mm
;
2041 struct mm_struct
*mm
;
2042 struct list_head
*p
;
2044 struct swap_info_struct
*si
= swap_info
[type
];
2045 struct folio
*folio
;
2049 if (!READ_ONCE(si
->inuse_pages
))
2053 retval
= shmem_unuse(type
);
2060 spin_lock(&mmlist_lock
);
2061 p
= &init_mm
.mmlist
;
2062 while (READ_ONCE(si
->inuse_pages
) &&
2063 !signal_pending(current
) &&
2064 (p
= p
->next
) != &init_mm
.mmlist
) {
2066 mm
= list_entry(p
, struct mm_struct
, mmlist
);
2067 if (!mmget_not_zero(mm
))
2069 spin_unlock(&mmlist_lock
);
2072 retval
= unuse_mm(mm
, type
);
2079 * Make sure that we aren't completely killing
2080 * interactive performance.
2083 spin_lock(&mmlist_lock
);
2085 spin_unlock(&mmlist_lock
);
2090 while (READ_ONCE(si
->inuse_pages
) &&
2091 !signal_pending(current
) &&
2092 (i
= find_next_to_unuse(si
, i
)) != 0) {
2094 entry
= swp_entry(type
, i
);
2095 folio
= filemap_get_folio(swap_address_space(entry
), i
);
2100 * It is conceivable that a racing task removed this folio from
2101 * swap cache just before we acquired the page lock. The folio
2102 * might even be back in swap cache on another swap area. But
2103 * that is okay, folio_free_swap() only removes stale folios.
2106 folio_wait_writeback(folio
);
2107 folio_free_swap(folio
);
2108 folio_unlock(folio
);
2113 * Lets check again to see if there are still swap entries in the map.
2114 * If yes, we would need to do retry the unuse logic again.
2115 * Under global memory pressure, swap entries can be reinserted back
2116 * into process space after the mmlist loop above passes over them.
2118 * Limit the number of retries? No: when mmget_not_zero()
2119 * above fails, that mm is likely to be freeing swap from
2120 * exit_mmap(), which proceeds at its own independent pace;
2121 * and even shmem_writepage() could have been preempted after
2122 * folio_alloc_swap(), temporarily hiding that swap. It's easy
2123 * and robust (though cpu-intensive) just to keep retrying.
2125 if (READ_ONCE(si
->inuse_pages
)) {
2126 if (!signal_pending(current
))
2135 * After a successful try_to_unuse, if no swap is now in use, we know
2136 * we can empty the mmlist. swap_lock must be held on entry and exit.
2137 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2138 * added to the mmlist just after page_duplicate - before would be racy.
2140 static void drain_mmlist(void)
2142 struct list_head
*p
, *next
;
2145 for (type
= 0; type
< nr_swapfiles
; type
++)
2146 if (swap_info
[type
]->inuse_pages
)
2148 spin_lock(&mmlist_lock
);
2149 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
2151 spin_unlock(&mmlist_lock
);
2155 * Free all of a swapdev's extent information
2157 static void destroy_swap_extents(struct swap_info_struct
*sis
)
2159 while (!RB_EMPTY_ROOT(&sis
->swap_extent_root
)) {
2160 struct rb_node
*rb
= sis
->swap_extent_root
.rb_node
;
2161 struct swap_extent
*se
= rb_entry(rb
, struct swap_extent
, rb_node
);
2163 rb_erase(rb
, &sis
->swap_extent_root
);
2167 if (sis
->flags
& SWP_ACTIVATED
) {
2168 struct file
*swap_file
= sis
->swap_file
;
2169 struct address_space
*mapping
= swap_file
->f_mapping
;
2171 sis
->flags
&= ~SWP_ACTIVATED
;
2172 if (mapping
->a_ops
->swap_deactivate
)
2173 mapping
->a_ops
->swap_deactivate(swap_file
);
2178 * Add a block range (and the corresponding page range) into this swapdev's
2181 * This function rather assumes that it is called in ascending page order.
2184 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
2185 unsigned long nr_pages
, sector_t start_block
)
2187 struct rb_node
**link
= &sis
->swap_extent_root
.rb_node
, *parent
= NULL
;
2188 struct swap_extent
*se
;
2189 struct swap_extent
*new_se
;
2192 * place the new node at the right most since the
2193 * function is called in ascending page order.
2197 link
= &parent
->rb_right
;
2201 se
= rb_entry(parent
, struct swap_extent
, rb_node
);
2202 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
2203 if (se
->start_block
+ se
->nr_pages
== start_block
) {
2205 se
->nr_pages
+= nr_pages
;
2210 /* No merge, insert a new extent. */
2211 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
2214 new_se
->start_page
= start_page
;
2215 new_se
->nr_pages
= nr_pages
;
2216 new_se
->start_block
= start_block
;
2218 rb_link_node(&new_se
->rb_node
, parent
, link
);
2219 rb_insert_color(&new_se
->rb_node
, &sis
->swap_extent_root
);
2222 EXPORT_SYMBOL_GPL(add_swap_extent
);
2225 * A `swap extent' is a simple thing which maps a contiguous range of pages
2226 * onto a contiguous range of disk blocks. A rbtree of swap extents is
2227 * built at swapon time and is then used at swap_writepage/swap_readpage
2228 * time for locating where on disk a page belongs.
2230 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2231 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2232 * swap files identically.
2234 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2235 * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2236 * swapfiles are handled *identically* after swapon time.
2238 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2239 * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray
2240 * blocks are found which do not fall within the PAGE_SIZE alignment
2241 * requirements, they are simply tossed out - we will never use those blocks
2244 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2245 * prevents users from writing to the swap device, which will corrupt memory.
2247 * The amount of disk space which a single swap extent represents varies.
2248 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2249 * extents in the rbtree. - akpm.
2251 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
2253 struct file
*swap_file
= sis
->swap_file
;
2254 struct address_space
*mapping
= swap_file
->f_mapping
;
2255 struct inode
*inode
= mapping
->host
;
2258 if (S_ISBLK(inode
->i_mode
)) {
2259 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2264 if (mapping
->a_ops
->swap_activate
) {
2265 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
2268 sis
->flags
|= SWP_ACTIVATED
;
2269 if ((sis
->flags
& SWP_FS_OPS
) &&
2270 sio_pool_init() != 0) {
2271 destroy_swap_extents(sis
);
2277 return generic_swapfile_activate(sis
, swap_file
, span
);
2280 static int swap_node(struct swap_info_struct
*p
)
2282 struct block_device
*bdev
;
2287 bdev
= p
->swap_file
->f_inode
->i_sb
->s_bdev
;
2289 return bdev
? bdev
->bd_disk
->node_id
: NUMA_NO_NODE
;
2292 static void setup_swap_info(struct swap_info_struct
*p
, int prio
,
2293 unsigned char *swap_map
,
2294 struct swap_cluster_info
*cluster_info
)
2301 p
->prio
= --least_priority
;
2303 * the plist prio is negated because plist ordering is
2304 * low-to-high, while swap ordering is high-to-low
2306 p
->list
.prio
= -p
->prio
;
2309 p
->avail_lists
[i
].prio
= -p
->prio
;
2311 if (swap_node(p
) == i
)
2312 p
->avail_lists
[i
].prio
= 1;
2314 p
->avail_lists
[i
].prio
= -p
->prio
;
2317 p
->swap_map
= swap_map
;
2318 p
->cluster_info
= cluster_info
;
2321 static void _enable_swap_info(struct swap_info_struct
*p
)
2323 p
->flags
|= SWP_WRITEOK
;
2324 atomic_long_add(p
->pages
, &nr_swap_pages
);
2325 total_swap_pages
+= p
->pages
;
2327 assert_spin_locked(&swap_lock
);
2329 * both lists are plists, and thus priority ordered.
2330 * swap_active_head needs to be priority ordered for swapoff(),
2331 * which on removal of any swap_info_struct with an auto-assigned
2332 * (i.e. negative) priority increments the auto-assigned priority
2333 * of any lower-priority swap_info_structs.
2334 * swap_avail_head needs to be priority ordered for folio_alloc_swap(),
2335 * which allocates swap pages from the highest available priority
2338 plist_add(&p
->list
, &swap_active_head
);
2340 /* add to available list iff swap device is not full */
2342 add_to_avail_list(p
);
2345 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
2346 unsigned char *swap_map
,
2347 struct swap_cluster_info
*cluster_info
)
2349 zswap_swapon(p
->type
);
2351 spin_lock(&swap_lock
);
2352 spin_lock(&p
->lock
);
2353 setup_swap_info(p
, prio
, swap_map
, cluster_info
);
2354 spin_unlock(&p
->lock
);
2355 spin_unlock(&swap_lock
);
2357 * Finished initializing swap device, now it's safe to reference it.
2359 percpu_ref_resurrect(&p
->users
);
2360 spin_lock(&swap_lock
);
2361 spin_lock(&p
->lock
);
2362 _enable_swap_info(p
);
2363 spin_unlock(&p
->lock
);
2364 spin_unlock(&swap_lock
);
2367 static void reinsert_swap_info(struct swap_info_struct
*p
)
2369 spin_lock(&swap_lock
);
2370 spin_lock(&p
->lock
);
2371 setup_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
2372 _enable_swap_info(p
);
2373 spin_unlock(&p
->lock
);
2374 spin_unlock(&swap_lock
);
2377 bool has_usable_swap(void)
2381 spin_lock(&swap_lock
);
2382 if (plist_head_empty(&swap_active_head
))
2384 spin_unlock(&swap_lock
);
2388 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
2390 struct swap_info_struct
*p
= NULL
;
2391 unsigned char *swap_map
;
2392 struct swap_cluster_info
*cluster_info
;
2393 struct file
*swap_file
, *victim
;
2394 struct address_space
*mapping
;
2395 struct inode
*inode
;
2396 struct filename
*pathname
;
2398 unsigned int old_block_size
;
2400 if (!capable(CAP_SYS_ADMIN
))
2403 BUG_ON(!current
->mm
);
2405 pathname
= getname(specialfile
);
2406 if (IS_ERR(pathname
))
2407 return PTR_ERR(pathname
);
2409 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
2410 err
= PTR_ERR(victim
);
2414 mapping
= victim
->f_mapping
;
2415 spin_lock(&swap_lock
);
2416 plist_for_each_entry(p
, &swap_active_head
, list
) {
2417 if (p
->flags
& SWP_WRITEOK
) {
2418 if (p
->swap_file
->f_mapping
== mapping
) {
2426 spin_unlock(&swap_lock
);
2429 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2430 vm_unacct_memory(p
->pages
);
2433 spin_unlock(&swap_lock
);
2436 spin_lock(&p
->lock
);
2437 del_from_avail_list(p
);
2439 struct swap_info_struct
*si
= p
;
2442 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2445 for_each_node(nid
) {
2446 if (si
->avail_lists
[nid
].prio
!= 1)
2447 si
->avail_lists
[nid
].prio
--;
2452 plist_del(&p
->list
, &swap_active_head
);
2453 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2454 total_swap_pages
-= p
->pages
;
2455 p
->flags
&= ~SWP_WRITEOK
;
2456 spin_unlock(&p
->lock
);
2457 spin_unlock(&swap_lock
);
2459 disable_swap_slots_cache_lock();
2461 set_current_oom_origin();
2462 err
= try_to_unuse(p
->type
);
2463 clear_current_oom_origin();
2466 /* re-insert swap space back into swap_list */
2467 reinsert_swap_info(p
);
2468 reenable_swap_slots_cache_unlock();
2472 reenable_swap_slots_cache_unlock();
2475 * Wait for swap operations protected by get/put_swap_device()
2478 * We need synchronize_rcu() here to protect the accessing to
2479 * the swap cache data structure.
2481 percpu_ref_kill(&p
->users
);
2483 wait_for_completion(&p
->comp
);
2485 flush_work(&p
->discard_work
);
2487 destroy_swap_extents(p
);
2488 if (p
->flags
& SWP_CONTINUED
)
2489 free_swap_count_continuations(p
);
2491 if (!p
->bdev
|| !bdev_nonrot(p
->bdev
))
2492 atomic_dec(&nr_rotate_swap
);
2494 mutex_lock(&swapon_mutex
);
2495 spin_lock(&swap_lock
);
2496 spin_lock(&p
->lock
);
2499 /* wait for anyone still in scan_swap_map_slots */
2500 p
->highest_bit
= 0; /* cuts scans short */
2501 while (p
->flags
>= SWP_SCANNING
) {
2502 spin_unlock(&p
->lock
);
2503 spin_unlock(&swap_lock
);
2504 schedule_timeout_uninterruptible(1);
2505 spin_lock(&swap_lock
);
2506 spin_lock(&p
->lock
);
2509 swap_file
= p
->swap_file
;
2510 old_block_size
= p
->old_block_size
;
2511 p
->swap_file
= NULL
;
2513 swap_map
= p
->swap_map
;
2515 cluster_info
= p
->cluster_info
;
2516 p
->cluster_info
= NULL
;
2517 spin_unlock(&p
->lock
);
2518 spin_unlock(&swap_lock
);
2519 arch_swap_invalidate_area(p
->type
);
2520 zswap_swapoff(p
->type
);
2521 mutex_unlock(&swapon_mutex
);
2522 free_percpu(p
->percpu_cluster
);
2523 p
->percpu_cluster
= NULL
;
2524 free_percpu(p
->cluster_next_cpu
);
2525 p
->cluster_next_cpu
= NULL
;
2527 kvfree(cluster_info
);
2528 /* Destroy swap account information */
2529 swap_cgroup_swapoff(p
->type
);
2530 exit_swap_address_space(p
->type
);
2532 inode
= mapping
->host
;
2533 if (S_ISBLK(inode
->i_mode
)) {
2534 struct block_device
*bdev
= I_BDEV(inode
);
2536 set_blocksize(bdev
, old_block_size
);
2537 blkdev_put(bdev
, p
);
2541 inode
->i_flags
&= ~S_SWAPFILE
;
2542 inode_unlock(inode
);
2543 filp_close(swap_file
, NULL
);
2546 * Clear the SWP_USED flag after all resources are freed so that swapon
2547 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2548 * not hold p->lock after we cleared its SWP_WRITEOK.
2550 spin_lock(&swap_lock
);
2552 spin_unlock(&swap_lock
);
2555 atomic_inc(&proc_poll_event
);
2556 wake_up_interruptible(&proc_poll_wait
);
2559 filp_close(victim
, NULL
);
2565 #ifdef CONFIG_PROC_FS
2566 static __poll_t
swaps_poll(struct file
*file
, poll_table
*wait
)
2568 struct seq_file
*seq
= file
->private_data
;
2570 poll_wait(file
, &proc_poll_wait
, wait
);
2572 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2573 seq
->poll_event
= atomic_read(&proc_poll_event
);
2574 return EPOLLIN
| EPOLLRDNORM
| EPOLLERR
| EPOLLPRI
;
2577 return EPOLLIN
| EPOLLRDNORM
;
2581 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2583 struct swap_info_struct
*si
;
2587 mutex_lock(&swapon_mutex
);
2590 return SEQ_START_TOKEN
;
2592 for (type
= 0; (si
= swap_type_to_swap_info(type
)); type
++) {
2593 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2602 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2604 struct swap_info_struct
*si
= v
;
2607 if (v
== SEQ_START_TOKEN
)
2610 type
= si
->type
+ 1;
2613 for (; (si
= swap_type_to_swap_info(type
)); type
++) {
2614 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2622 static void swap_stop(struct seq_file
*swap
, void *v
)
2624 mutex_unlock(&swapon_mutex
);
2627 static int swap_show(struct seq_file
*swap
, void *v
)
2629 struct swap_info_struct
*si
= v
;
2632 unsigned long bytes
, inuse
;
2634 if (si
== SEQ_START_TOKEN
) {
2635 seq_puts(swap
, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2639 bytes
= K(si
->pages
);
2640 inuse
= K(READ_ONCE(si
->inuse_pages
));
2642 file
= si
->swap_file
;
2643 len
= seq_file_path(swap
, file
, " \t\n\\");
2644 seq_printf(swap
, "%*s%s\t%lu\t%s%lu\t%s%d\n",
2645 len
< 40 ? 40 - len
: 1, " ",
2646 S_ISBLK(file_inode(file
)->i_mode
) ?
2647 "partition" : "file\t",
2648 bytes
, bytes
< 10000000 ? "\t" : "",
2649 inuse
, inuse
< 10000000 ? "\t" : "",
2654 static const struct seq_operations swaps_op
= {
2655 .start
= swap_start
,
2661 static int swaps_open(struct inode
*inode
, struct file
*file
)
2663 struct seq_file
*seq
;
2666 ret
= seq_open(file
, &swaps_op
);
2670 seq
= file
->private_data
;
2671 seq
->poll_event
= atomic_read(&proc_poll_event
);
2675 static const struct proc_ops swaps_proc_ops
= {
2676 .proc_flags
= PROC_ENTRY_PERMANENT
,
2677 .proc_open
= swaps_open
,
2678 .proc_read
= seq_read
,
2679 .proc_lseek
= seq_lseek
,
2680 .proc_release
= seq_release
,
2681 .proc_poll
= swaps_poll
,
2684 static int __init
procswaps_init(void)
2686 proc_create("swaps", 0, NULL
, &swaps_proc_ops
);
2689 __initcall(procswaps_init
);
2690 #endif /* CONFIG_PROC_FS */
2692 #ifdef MAX_SWAPFILES_CHECK
2693 static int __init
max_swapfiles_check(void)
2695 MAX_SWAPFILES_CHECK();
2698 late_initcall(max_swapfiles_check
);
2701 static struct swap_info_struct
*alloc_swap_info(void)
2703 struct swap_info_struct
*p
;
2704 struct swap_info_struct
*defer
= NULL
;
2708 p
= kvzalloc(struct_size(p
, avail_lists
, nr_node_ids
), GFP_KERNEL
);
2710 return ERR_PTR(-ENOMEM
);
2712 if (percpu_ref_init(&p
->users
, swap_users_ref_free
,
2713 PERCPU_REF_INIT_DEAD
, GFP_KERNEL
)) {
2715 return ERR_PTR(-ENOMEM
);
2718 spin_lock(&swap_lock
);
2719 for (type
= 0; type
< nr_swapfiles
; type
++) {
2720 if (!(swap_info
[type
]->flags
& SWP_USED
))
2723 if (type
>= MAX_SWAPFILES
) {
2724 spin_unlock(&swap_lock
);
2725 percpu_ref_exit(&p
->users
);
2727 return ERR_PTR(-EPERM
);
2729 if (type
>= nr_swapfiles
) {
2732 * Publish the swap_info_struct after initializing it.
2733 * Note that kvzalloc() above zeroes all its fields.
2735 smp_store_release(&swap_info
[type
], p
); /* rcu_assign_pointer() */
2739 p
= swap_info
[type
];
2741 * Do not memset this entry: a racing procfs swap_next()
2742 * would be relying on p->type to remain valid.
2745 p
->swap_extent_root
= RB_ROOT
;
2746 plist_node_init(&p
->list
, 0);
2748 plist_node_init(&p
->avail_lists
[i
], 0);
2749 p
->flags
= SWP_USED
;
2750 spin_unlock(&swap_lock
);
2752 percpu_ref_exit(&defer
->users
);
2755 spin_lock_init(&p
->lock
);
2756 spin_lock_init(&p
->cont_lock
);
2757 init_completion(&p
->comp
);
2762 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2766 if (S_ISBLK(inode
->i_mode
)) {
2767 p
->bdev
= blkdev_get_by_dev(inode
->i_rdev
,
2768 BLK_OPEN_READ
| BLK_OPEN_WRITE
, p
, NULL
);
2769 if (IS_ERR(p
->bdev
)) {
2770 error
= PTR_ERR(p
->bdev
);
2774 p
->old_block_size
= block_size(p
->bdev
);
2775 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2779 * Zoned block devices contain zones that have a sequential
2780 * write only restriction. Hence zoned block devices are not
2781 * suitable for swapping. Disallow them here.
2783 if (bdev_is_zoned(p
->bdev
))
2785 p
->flags
|= SWP_BLKDEV
;
2786 } else if (S_ISREG(inode
->i_mode
)) {
2787 p
->bdev
= inode
->i_sb
->s_bdev
;
2795 * Find out how many pages are allowed for a single swap device. There
2796 * are two limiting factors:
2797 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2798 * 2) the number of bits in the swap pte, as defined by the different
2801 * In order to find the largest possible bit mask, a swap entry with
2802 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2803 * decoded to a swp_entry_t again, and finally the swap offset is
2806 * This will mask all the bits from the initial ~0UL mask that can't
2807 * be encoded in either the swp_entry_t or the architecture definition
2810 unsigned long generic_max_swapfile_size(void)
2812 return swp_offset(pte_to_swp_entry(
2813 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2816 /* Can be overridden by an architecture for additional checks. */
2817 __weak
unsigned long arch_max_swapfile_size(void)
2819 return generic_max_swapfile_size();
2822 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2823 union swap_header
*swap_header
,
2824 struct inode
*inode
)
2827 unsigned long maxpages
;
2828 unsigned long swapfilepages
;
2829 unsigned long last_page
;
2831 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2832 pr_err("Unable to find swap-space signature\n");
2836 /* swap partition endianness hack... */
2837 if (swab32(swap_header
->info
.version
) == 1) {
2838 swab32s(&swap_header
->info
.version
);
2839 swab32s(&swap_header
->info
.last_page
);
2840 swab32s(&swap_header
->info
.nr_badpages
);
2841 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2843 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2844 swab32s(&swap_header
->info
.badpages
[i
]);
2846 /* Check the swap header's sub-version */
2847 if (swap_header
->info
.version
!= 1) {
2848 pr_warn("Unable to handle swap header version %d\n",
2849 swap_header
->info
.version
);
2854 p
->cluster_next
= 1;
2857 maxpages
= swapfile_maximum_size
;
2858 last_page
= swap_header
->info
.last_page
;
2860 pr_warn("Empty swap-file\n");
2863 if (last_page
> maxpages
) {
2864 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2865 K(maxpages
), K(last_page
));
2867 if (maxpages
> last_page
) {
2868 maxpages
= last_page
+ 1;
2869 /* p->max is an unsigned int: don't overflow it */
2870 if ((unsigned int)maxpages
== 0)
2871 maxpages
= UINT_MAX
;
2873 p
->highest_bit
= maxpages
- 1;
2877 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
2878 if (swapfilepages
&& maxpages
> swapfilepages
) {
2879 pr_warn("Swap area shorter than signature indicates\n");
2882 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
2884 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2890 #define SWAP_CLUSTER_INFO_COLS \
2891 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2892 #define SWAP_CLUSTER_SPACE_COLS \
2893 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2894 #define SWAP_CLUSTER_COLS \
2895 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2897 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
2898 union swap_header
*swap_header
,
2899 unsigned char *swap_map
,
2900 struct swap_cluster_info
*cluster_info
,
2901 unsigned long maxpages
,
2905 unsigned int nr_good_pages
;
2907 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2908 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
2909 unsigned long i
, idx
;
2911 nr_good_pages
= maxpages
- 1; /* omit header page */
2913 cluster_list_init(&p
->free_clusters
);
2914 cluster_list_init(&p
->discard_clusters
);
2916 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
2917 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
2918 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
2920 if (page_nr
< maxpages
) {
2921 swap_map
[page_nr
] = SWAP_MAP_BAD
;
2924 * Haven't marked the cluster free yet, no list
2925 * operation involved
2927 inc_cluster_info_page(p
, cluster_info
, page_nr
);
2931 /* Haven't marked the cluster free yet, no list operation involved */
2932 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
2933 inc_cluster_info_page(p
, cluster_info
, i
);
2935 if (nr_good_pages
) {
2936 swap_map
[0] = SWAP_MAP_BAD
;
2938 * Not mark the cluster free yet, no list
2939 * operation involved
2941 inc_cluster_info_page(p
, cluster_info
, 0);
2943 p
->pages
= nr_good_pages
;
2944 nr_extents
= setup_swap_extents(p
, span
);
2947 nr_good_pages
= p
->pages
;
2949 if (!nr_good_pages
) {
2950 pr_warn("Empty swap-file\n");
2959 * Reduce false cache line sharing between cluster_info and
2960 * sharing same address space.
2962 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
2963 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
2964 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
2965 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
2966 if (idx
>= nr_clusters
)
2968 if (cluster_count(&cluster_info
[idx
]))
2970 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
2971 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
2978 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
2980 struct swap_info_struct
*p
;
2981 struct filename
*name
;
2982 struct file
*swap_file
= NULL
;
2983 struct address_space
*mapping
;
2984 struct dentry
*dentry
;
2987 union swap_header
*swap_header
;
2990 unsigned long maxpages
;
2991 unsigned char *swap_map
= NULL
;
2992 struct swap_cluster_info
*cluster_info
= NULL
;
2993 struct page
*page
= NULL
;
2994 struct inode
*inode
= NULL
;
2995 bool inced_nr_rotate_swap
= false;
2997 if (swap_flags
& ~SWAP_FLAGS_VALID
)
3000 if (!capable(CAP_SYS_ADMIN
))
3003 if (!swap_avail_heads
)
3006 p
= alloc_swap_info();
3010 INIT_WORK(&p
->discard_work
, swap_discard_work
);
3012 name
= getname(specialfile
);
3014 error
= PTR_ERR(name
);
3018 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
3019 if (IS_ERR(swap_file
)) {
3020 error
= PTR_ERR(swap_file
);
3025 p
->swap_file
= swap_file
;
3026 mapping
= swap_file
->f_mapping
;
3027 dentry
= swap_file
->f_path
.dentry
;
3028 inode
= mapping
->host
;
3030 error
= claim_swapfile(p
, inode
);
3031 if (unlikely(error
))
3035 if (d_unlinked(dentry
) || cant_mount(dentry
)) {
3037 goto bad_swap_unlock_inode
;
3039 if (IS_SWAPFILE(inode
)) {
3041 goto bad_swap_unlock_inode
;
3045 * Read the swap header.
3047 if (!mapping
->a_ops
->read_folio
) {
3049 goto bad_swap_unlock_inode
;
3051 page
= read_mapping_page(mapping
, 0, swap_file
);
3053 error
= PTR_ERR(page
);
3054 goto bad_swap_unlock_inode
;
3056 swap_header
= kmap(page
);
3058 maxpages
= read_swap_header(p
, swap_header
, inode
);
3059 if (unlikely(!maxpages
)) {
3061 goto bad_swap_unlock_inode
;
3064 /* OK, set up the swap map and apply the bad block list */
3065 swap_map
= vzalloc(maxpages
);
3068 goto bad_swap_unlock_inode
;
3071 if (p
->bdev
&& bdev_stable_writes(p
->bdev
))
3072 p
->flags
|= SWP_STABLE_WRITES
;
3074 if (p
->bdev
&& bdev_synchronous(p
->bdev
))
3075 p
->flags
|= SWP_SYNCHRONOUS_IO
;
3077 if (p
->bdev
&& bdev_nonrot(p
->bdev
)) {
3079 unsigned long ci
, nr_cluster
;
3081 p
->flags
|= SWP_SOLIDSTATE
;
3082 p
->cluster_next_cpu
= alloc_percpu(unsigned int);
3083 if (!p
->cluster_next_cpu
) {
3085 goto bad_swap_unlock_inode
;
3088 * select a random position to start with to help wear leveling
3091 for_each_possible_cpu(cpu
) {
3092 per_cpu(*p
->cluster_next_cpu
, cpu
) =
3093 get_random_u32_inclusive(1, p
->highest_bit
);
3095 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3097 cluster_info
= kvcalloc(nr_cluster
, sizeof(*cluster_info
),
3099 if (!cluster_info
) {
3101 goto bad_swap_unlock_inode
;
3104 for (ci
= 0; ci
< nr_cluster
; ci
++)
3105 spin_lock_init(&((cluster_info
+ ci
)->lock
));
3107 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
3108 if (!p
->percpu_cluster
) {
3110 goto bad_swap_unlock_inode
;
3112 for_each_possible_cpu(cpu
) {
3113 struct percpu_cluster
*cluster
;
3114 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
3115 cluster_set_null(&cluster
->index
);
3118 atomic_inc(&nr_rotate_swap
);
3119 inced_nr_rotate_swap
= true;
3122 error
= swap_cgroup_swapon(p
->type
, maxpages
);
3124 goto bad_swap_unlock_inode
;
3126 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
3127 cluster_info
, maxpages
, &span
);
3128 if (unlikely(nr_extents
< 0)) {
3130 goto bad_swap_unlock_inode
;
3133 if ((swap_flags
& SWAP_FLAG_DISCARD
) &&
3134 p
->bdev
&& bdev_max_discard_sectors(p
->bdev
)) {
3136 * When discard is enabled for swap with no particular
3137 * policy flagged, we set all swap discard flags here in
3138 * order to sustain backward compatibility with older
3139 * swapon(8) releases.
3141 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
3145 * By flagging sys_swapon, a sysadmin can tell us to
3146 * either do single-time area discards only, or to just
3147 * perform discards for released swap page-clusters.
3148 * Now it's time to adjust the p->flags accordingly.
3150 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
3151 p
->flags
&= ~SWP_PAGE_DISCARD
;
3152 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
3153 p
->flags
&= ~SWP_AREA_DISCARD
;
3155 /* issue a swapon-time discard if it's still required */
3156 if (p
->flags
& SWP_AREA_DISCARD
) {
3157 int err
= discard_swap(p
);
3159 pr_err("swapon: discard_swap(%p): %d\n",
3164 error
= init_swap_address_space(p
->type
, maxpages
);
3166 goto bad_swap_unlock_inode
;
3169 * Flush any pending IO and dirty mappings before we start using this
3172 inode
->i_flags
|= S_SWAPFILE
;
3173 error
= inode_drain_writes(inode
);
3175 inode
->i_flags
&= ~S_SWAPFILE
;
3176 goto free_swap_address_space
;
3179 mutex_lock(&swapon_mutex
);
3181 if (swap_flags
& SWAP_FLAG_PREFER
)
3183 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
3184 enable_swap_info(p
, prio
, swap_map
, cluster_info
);
3186 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s\n",
3187 K(p
->pages
), name
->name
, p
->prio
, nr_extents
,
3188 K((unsigned long long)span
),
3189 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
3190 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
3191 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
3192 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "");
3194 mutex_unlock(&swapon_mutex
);
3195 atomic_inc(&proc_poll_event
);
3196 wake_up_interruptible(&proc_poll_wait
);
3200 free_swap_address_space
:
3201 exit_swap_address_space(p
->type
);
3202 bad_swap_unlock_inode
:
3203 inode_unlock(inode
);
3205 free_percpu(p
->percpu_cluster
);
3206 p
->percpu_cluster
= NULL
;
3207 free_percpu(p
->cluster_next_cpu
);
3208 p
->cluster_next_cpu
= NULL
;
3209 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
3210 set_blocksize(p
->bdev
, p
->old_block_size
);
3211 blkdev_put(p
->bdev
, p
);
3214 destroy_swap_extents(p
);
3215 swap_cgroup_swapoff(p
->type
);
3216 spin_lock(&swap_lock
);
3217 p
->swap_file
= NULL
;
3219 spin_unlock(&swap_lock
);
3221 kvfree(cluster_info
);
3222 if (inced_nr_rotate_swap
)
3223 atomic_dec(&nr_rotate_swap
);
3225 filp_close(swap_file
, NULL
);
3227 if (page
&& !IS_ERR(page
)) {
3234 inode_unlock(inode
);
3236 enable_swap_slots_cache();
3240 void si_swapinfo(struct sysinfo
*val
)
3243 unsigned long nr_to_be_unused
= 0;
3245 spin_lock(&swap_lock
);
3246 for (type
= 0; type
< nr_swapfiles
; type
++) {
3247 struct swap_info_struct
*si
= swap_info
[type
];
3249 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
3250 nr_to_be_unused
+= READ_ONCE(si
->inuse_pages
);
3252 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
3253 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
3254 spin_unlock(&swap_lock
);
3258 * Verify that a swap entry is valid and increment its swap map count.
3260 * Returns error code in following case.
3262 * - swp_entry is invalid -> EINVAL
3263 * - swp_entry is migration entry -> EINVAL
3264 * - swap-cache reference is requested but there is already one. -> EEXIST
3265 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3266 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3268 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
3270 struct swap_info_struct
*p
;
3271 struct swap_cluster_info
*ci
;
3272 unsigned long offset
;
3273 unsigned char count
;
3274 unsigned char has_cache
;
3277 p
= swp_swap_info(entry
);
3279 offset
= swp_offset(entry
);
3280 ci
= lock_cluster_or_swap_info(p
, offset
);
3282 count
= p
->swap_map
[offset
];
3285 * swapin_readahead() doesn't check if a swap entry is valid, so the
3286 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3288 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
3293 has_cache
= count
& SWAP_HAS_CACHE
;
3294 count
&= ~SWAP_HAS_CACHE
;
3297 if (usage
== SWAP_HAS_CACHE
) {
3299 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3300 if (!has_cache
&& count
)
3301 has_cache
= SWAP_HAS_CACHE
;
3302 else if (has_cache
) /* someone else added cache */
3304 else /* no users remaining */
3307 } else if (count
|| has_cache
) {
3309 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
3311 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
3313 else if (swap_count_continued(p
, offset
, count
))
3314 count
= COUNT_CONTINUED
;
3318 err
= -ENOENT
; /* unused swap entry */
3320 WRITE_ONCE(p
->swap_map
[offset
], count
| has_cache
);
3323 unlock_cluster_or_swap_info(p
, ci
);
3328 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3329 * (in which case its reference count is never incremented).
3331 void swap_shmem_alloc(swp_entry_t entry
)
3333 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
3337 * Increase reference count of swap entry by 1.
3338 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3339 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3340 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3341 * might occur if a page table entry has got corrupted.
3343 int swap_duplicate(swp_entry_t entry
)
3347 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
3348 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
3353 * @entry: swap entry for which we allocate swap cache.
3355 * Called when allocating swap cache for existing swap entry,
3356 * This can return error codes. Returns 0 at success.
3357 * -EEXIST means there is a swap cache.
3358 * Note: return code is different from swap_duplicate().
3360 int swapcache_prepare(swp_entry_t entry
)
3362 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
3365 struct swap_info_struct
*swp_swap_info(swp_entry_t entry
)
3367 return swap_type_to_swap_info(swp_type(entry
));
3370 struct swap_info_struct
*page_swap_info(struct page
*page
)
3372 swp_entry_t entry
= page_swap_entry(page
);
3373 return swp_swap_info(entry
);
3377 * out-of-line methods to avoid include hell.
3379 struct address_space
*swapcache_mapping(struct folio
*folio
)
3381 return page_swap_info(&folio
->page
)->swap_file
->f_mapping
;
3383 EXPORT_SYMBOL_GPL(swapcache_mapping
);
3385 pgoff_t
__page_file_index(struct page
*page
)
3387 swp_entry_t swap
= page_swap_entry(page
);
3388 return swp_offset(swap
);
3390 EXPORT_SYMBOL_GPL(__page_file_index
);
3393 * add_swap_count_continuation - called when a swap count is duplicated
3394 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3395 * page of the original vmalloc'ed swap_map, to hold the continuation count
3396 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3397 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3399 * These continuation pages are seldom referenced: the common paths all work
3400 * on the original swap_map, only referring to a continuation page when the
3401 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3403 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3404 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3405 * can be called after dropping locks.
3407 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
3409 struct swap_info_struct
*si
;
3410 struct swap_cluster_info
*ci
;
3413 struct page
*list_page
;
3415 unsigned char count
;
3419 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3420 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3422 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
3424 si
= get_swap_device(entry
);
3427 * An acceptable race has occurred since the failing
3428 * __swap_duplicate(): the swap device may be swapoff
3432 spin_lock(&si
->lock
);
3434 offset
= swp_offset(entry
);
3436 ci
= lock_cluster(si
, offset
);
3438 count
= swap_count(si
->swap_map
[offset
]);
3440 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
3442 * The higher the swap count, the more likely it is that tasks
3443 * will race to add swap count continuation: we need to avoid
3444 * over-provisioning.
3454 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3455 offset
&= ~PAGE_MASK
;
3457 spin_lock(&si
->cont_lock
);
3459 * Page allocation does not initialize the page's lru field,
3460 * but it does always reset its private field.
3462 if (!page_private(head
)) {
3463 BUG_ON(count
& COUNT_CONTINUED
);
3464 INIT_LIST_HEAD(&head
->lru
);
3465 set_page_private(head
, SWP_CONTINUED
);
3466 si
->flags
|= SWP_CONTINUED
;
3469 list_for_each_entry(list_page
, &head
->lru
, lru
) {
3473 * If the previous map said no continuation, but we've found
3474 * a continuation page, free our allocation and use this one.
3476 if (!(count
& COUNT_CONTINUED
))
3477 goto out_unlock_cont
;
3479 map
= kmap_atomic(list_page
) + offset
;
3484 * If this continuation count now has some space in it,
3485 * free our allocation and use this one.
3487 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
3488 goto out_unlock_cont
;
3491 list_add_tail(&page
->lru
, &head
->lru
);
3492 page
= NULL
; /* now it's attached, don't free it */
3494 spin_unlock(&si
->cont_lock
);
3497 spin_unlock(&si
->lock
);
3498 put_swap_device(si
);
3506 * swap_count_continued - when the original swap_map count is incremented
3507 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3508 * into, carry if so, or else fail until a new continuation page is allocated;
3509 * when the original swap_map count is decremented from 0 with continuation,
3510 * borrow from the continuation and report whether it still holds more.
3511 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3514 static bool swap_count_continued(struct swap_info_struct
*si
,
3515 pgoff_t offset
, unsigned char count
)
3522 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3523 if (page_private(head
) != SWP_CONTINUED
) {
3524 BUG_ON(count
& COUNT_CONTINUED
);
3525 return false; /* need to add count continuation */
3528 spin_lock(&si
->cont_lock
);
3529 offset
&= ~PAGE_MASK
;
3530 page
= list_next_entry(head
, lru
);
3531 map
= kmap_atomic(page
) + offset
;
3533 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3534 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3536 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3538 * Think of how you add 1 to 999
3540 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3542 page
= list_next_entry(page
, lru
);
3543 BUG_ON(page
== head
);
3544 map
= kmap_atomic(page
) + offset
;
3546 if (*map
== SWAP_CONT_MAX
) {
3548 page
= list_next_entry(page
, lru
);
3550 ret
= false; /* add count continuation */
3553 map
= kmap_atomic(page
) + offset
;
3554 init_map
: *map
= 0; /* we didn't zero the page */
3558 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3559 map
= kmap_atomic(page
) + offset
;
3560 *map
= COUNT_CONTINUED
;
3563 ret
= true; /* incremented */
3565 } else { /* decrementing */
3567 * Think of how you subtract 1 from 1000
3569 BUG_ON(count
!= COUNT_CONTINUED
);
3570 while (*map
== COUNT_CONTINUED
) {
3572 page
= list_next_entry(page
, lru
);
3573 BUG_ON(page
== head
);
3574 map
= kmap_atomic(page
) + offset
;
3581 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3582 map
= kmap_atomic(page
) + offset
;
3583 *map
= SWAP_CONT_MAX
| count
;
3584 count
= COUNT_CONTINUED
;
3587 ret
= count
== COUNT_CONTINUED
;
3590 spin_unlock(&si
->cont_lock
);
3595 * free_swap_count_continuations - swapoff free all the continuation pages
3596 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3598 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3602 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3604 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3605 if (page_private(head
)) {
3606 struct page
*page
, *next
;
3608 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3609 list_del(&page
->lru
);
3616 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3617 void __folio_throttle_swaprate(struct folio
*folio
, gfp_t gfp
)
3619 struct swap_info_struct
*si
, *next
;
3620 int nid
= folio_nid(folio
);
3622 if (!(gfp
& __GFP_IO
))
3625 if (!blk_cgroup_congested())
3629 * We've already scheduled a throttle, avoid taking the global swap
3632 if (current
->throttle_disk
)
3635 spin_lock(&swap_avail_lock
);
3636 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[nid
],
3639 blkcg_schedule_throttle(si
->bdev
->bd_disk
, true);
3643 spin_unlock(&swap_avail_lock
);
3647 static int __init
swapfile_init(void)
3651 swap_avail_heads
= kmalloc_array(nr_node_ids
, sizeof(struct plist_head
),
3653 if (!swap_avail_heads
) {
3654 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3659 plist_head_init(&swap_avail_heads
[nid
]);
3661 swapfile_maximum_size
= arch_max_swapfile_size();
3663 #ifdef CONFIG_MIGRATION
3664 if (swapfile_maximum_size
>= (1UL << SWP_MIG_TOTAL_BITS
))
3665 swap_migration_ad_supported
= true;
3666 #endif /* CONFIG_MIGRATION */
3670 subsys_initcall(swapfile_init
);