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>
45 #include <linux/plist.h>
47 #include <asm/tlbflush.h>
48 #include <linux/swapops.h>
49 #include <linux/swap_cgroup.h>
53 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
55 static void free_swap_count_continuations(struct swap_info_struct
*);
57 static DEFINE_SPINLOCK(swap_lock
);
58 static unsigned int nr_swapfiles
;
59 atomic_long_t nr_swap_pages
;
61 * Some modules use swappable objects and may try to swap them out under
62 * memory pressure (via the shrinker). Before doing so, they may wish to
63 * check to see if any swap space is available.
65 EXPORT_SYMBOL_GPL(nr_swap_pages
);
66 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
67 long total_swap_pages
;
68 static int least_priority
= -1;
69 unsigned long swapfile_maximum_size
;
70 #ifdef CONFIG_MIGRATION
71 bool swap_migration_ad_supported
;
72 #endif /* CONFIG_MIGRATION */
74 static const char Bad_file
[] = "Bad swap file entry ";
75 static const char Unused_file
[] = "Unused swap file entry ";
76 static const char Bad_offset
[] = "Bad swap offset entry ";
77 static const char Unused_offset
[] = "Unused swap offset entry ";
80 * all active swap_info_structs
81 * protected with swap_lock, and ordered by priority.
83 static PLIST_HEAD(swap_active_head
);
86 * all available (active, not full) swap_info_structs
87 * protected with swap_avail_lock, ordered by priority.
88 * This is used by folio_alloc_swap() instead of swap_active_head
89 * because swap_active_head includes all swap_info_structs,
90 * but folio_alloc_swap() doesn't need to look at full ones.
91 * This uses its own lock instead of swap_lock because when a
92 * swap_info_struct changes between not-full/full, it needs to
93 * add/remove itself to/from this list, but the swap_info_struct->lock
94 * is held and the locking order requires swap_lock to be taken
95 * before any swap_info_struct->lock.
97 static struct plist_head
*swap_avail_heads
;
98 static DEFINE_SPINLOCK(swap_avail_lock
);
100 static struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
102 static DEFINE_MUTEX(swapon_mutex
);
104 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
105 /* Activity counter to indicate that a swapon or swapoff has occurred */
106 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
108 atomic_t nr_rotate_swap
= ATOMIC_INIT(0);
110 static struct swap_info_struct
*swap_type_to_swap_info(int type
)
112 if (type
>= MAX_SWAPFILES
)
115 return READ_ONCE(swap_info
[type
]); /* rcu_dereference() */
118 static inline unsigned char swap_count(unsigned char ent
)
120 return ent
& ~SWAP_HAS_CACHE
; /* may include COUNT_CONTINUED flag */
123 /* Reclaim the swap entry anyway if possible */
124 #define TTRS_ANYWAY 0x1
126 * Reclaim the swap entry if there are no more mappings of the
129 #define TTRS_UNMAPPED 0x2
130 /* Reclaim the swap entry if swap is getting full*/
131 #define TTRS_FULL 0x4
133 /* returns 1 if swap entry is freed */
134 static int __try_to_reclaim_swap(struct swap_info_struct
*si
,
135 unsigned long offset
, unsigned long flags
)
137 swp_entry_t entry
= swp_entry(si
->type
, offset
);
141 folio
= filemap_get_folio(swap_address_space(entry
), offset
);
145 * When this function is called from scan_swap_map_slots() and it's
146 * called by vmscan.c at reclaiming folios. So we hold a folio lock
147 * here. We have to use trylock for avoiding deadlock. This is a special
148 * case and you should use folio_free_swap() with explicit folio_lock()
149 * in usual operations.
151 if (folio_trylock(folio
)) {
152 if ((flags
& TTRS_ANYWAY
) ||
153 ((flags
& TTRS_UNMAPPED
) && !folio_mapped(folio
)) ||
154 ((flags
& TTRS_FULL
) && mem_cgroup_swap_full(folio
)))
155 ret
= folio_free_swap(folio
);
162 static inline struct swap_extent
*first_se(struct swap_info_struct
*sis
)
164 struct rb_node
*rb
= rb_first(&sis
->swap_extent_root
);
165 return rb_entry(rb
, struct swap_extent
, rb_node
);
168 static inline struct swap_extent
*next_se(struct swap_extent
*se
)
170 struct rb_node
*rb
= rb_next(&se
->rb_node
);
171 return rb
? rb_entry(rb
, struct swap_extent
, rb_node
) : NULL
;
175 * swapon tell device that all the old swap contents can be discarded,
176 * to allow the swap device to optimize its wear-levelling.
178 static int discard_swap(struct swap_info_struct
*si
)
180 struct swap_extent
*se
;
181 sector_t start_block
;
185 /* Do not discard the swap header page! */
187 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
188 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
190 err
= blkdev_issue_discard(si
->bdev
, start_block
,
191 nr_blocks
, GFP_KERNEL
);
197 for (se
= next_se(se
); se
; se
= next_se(se
)) {
198 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
199 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
201 err
= blkdev_issue_discard(si
->bdev
, start_block
,
202 nr_blocks
, GFP_KERNEL
);
208 return err
; /* That will often be -EOPNOTSUPP */
211 static struct swap_extent
*
212 offset_to_swap_extent(struct swap_info_struct
*sis
, unsigned long offset
)
214 struct swap_extent
*se
;
217 rb
= sis
->swap_extent_root
.rb_node
;
219 se
= rb_entry(rb
, struct swap_extent
, rb_node
);
220 if (offset
< se
->start_page
)
222 else if (offset
>= se
->start_page
+ se
->nr_pages
)
227 /* It *must* be present */
231 sector_t
swap_folio_sector(struct folio
*folio
)
233 struct swap_info_struct
*sis
= swp_swap_info(folio
->swap
);
234 struct swap_extent
*se
;
238 offset
= swp_offset(folio
->swap
);
239 se
= offset_to_swap_extent(sis
, offset
);
240 sector
= se
->start_block
+ (offset
- se
->start_page
);
241 return sector
<< (PAGE_SHIFT
- 9);
245 * swap allocation tell device that a cluster of swap can now be discarded,
246 * to allow the swap device to optimize its wear-levelling.
248 static void discard_swap_cluster(struct swap_info_struct
*si
,
249 pgoff_t start_page
, pgoff_t nr_pages
)
251 struct swap_extent
*se
= offset_to_swap_extent(si
, start_page
);
254 pgoff_t offset
= start_page
- se
->start_page
;
255 sector_t start_block
= se
->start_block
+ offset
;
256 sector_t nr_blocks
= se
->nr_pages
- offset
;
258 if (nr_blocks
> nr_pages
)
259 nr_blocks
= nr_pages
;
260 start_page
+= nr_blocks
;
261 nr_pages
-= nr_blocks
;
263 start_block
<<= PAGE_SHIFT
- 9;
264 nr_blocks
<<= PAGE_SHIFT
- 9;
265 if (blkdev_issue_discard(si
->bdev
, start_block
,
266 nr_blocks
, GFP_NOIO
))
273 #ifdef CONFIG_THP_SWAP
274 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
276 #define swap_entry_size(size) (size)
278 #define SWAPFILE_CLUSTER 256
281 * Define swap_entry_size() as constant to let compiler to optimize
282 * out some code if !CONFIG_THP_SWAP
284 #define swap_entry_size(size) 1
286 #define LATENCY_LIMIT 256
288 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
294 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
299 static inline void cluster_set_count(struct swap_cluster_info
*info
,
305 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
306 unsigned int c
, unsigned int f
)
312 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
317 static inline void cluster_set_next(struct swap_cluster_info
*info
,
323 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
324 unsigned int n
, unsigned int f
)
330 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
332 return info
->flags
& CLUSTER_FLAG_FREE
;
335 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
337 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
340 static inline void cluster_set_null(struct swap_cluster_info
*info
)
342 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
346 static inline bool cluster_is_huge(struct swap_cluster_info
*info
)
348 if (IS_ENABLED(CONFIG_THP_SWAP
))
349 return info
->flags
& CLUSTER_FLAG_HUGE
;
353 static inline void cluster_clear_huge(struct swap_cluster_info
*info
)
355 info
->flags
&= ~CLUSTER_FLAG_HUGE
;
358 static inline struct swap_cluster_info
*lock_cluster(struct swap_info_struct
*si
,
359 unsigned long offset
)
361 struct swap_cluster_info
*ci
;
363 ci
= si
->cluster_info
;
365 ci
+= offset
/ SWAPFILE_CLUSTER
;
366 spin_lock(&ci
->lock
);
371 static inline void unlock_cluster(struct swap_cluster_info
*ci
)
374 spin_unlock(&ci
->lock
);
378 * Determine the locking method in use for this device. Return
379 * swap_cluster_info if SSD-style cluster-based locking is in place.
381 static inline struct swap_cluster_info
*lock_cluster_or_swap_info(
382 struct swap_info_struct
*si
, unsigned long offset
)
384 struct swap_cluster_info
*ci
;
386 /* Try to use fine-grained SSD-style locking if available: */
387 ci
= lock_cluster(si
, offset
);
388 /* Otherwise, fall back to traditional, coarse locking: */
390 spin_lock(&si
->lock
);
395 static inline void unlock_cluster_or_swap_info(struct swap_info_struct
*si
,
396 struct swap_cluster_info
*ci
)
401 spin_unlock(&si
->lock
);
404 static inline bool cluster_list_empty(struct swap_cluster_list
*list
)
406 return cluster_is_null(&list
->head
);
409 static inline unsigned int cluster_list_first(struct swap_cluster_list
*list
)
411 return cluster_next(&list
->head
);
414 static void cluster_list_init(struct swap_cluster_list
*list
)
416 cluster_set_null(&list
->head
);
417 cluster_set_null(&list
->tail
);
420 static void cluster_list_add_tail(struct swap_cluster_list
*list
,
421 struct swap_cluster_info
*ci
,
424 if (cluster_list_empty(list
)) {
425 cluster_set_next_flag(&list
->head
, idx
, 0);
426 cluster_set_next_flag(&list
->tail
, idx
, 0);
428 struct swap_cluster_info
*ci_tail
;
429 unsigned int tail
= cluster_next(&list
->tail
);
432 * Nested cluster lock, but both cluster locks are
433 * only acquired when we held swap_info_struct->lock
436 spin_lock_nested(&ci_tail
->lock
, SINGLE_DEPTH_NESTING
);
437 cluster_set_next(ci_tail
, idx
);
438 spin_unlock(&ci_tail
->lock
);
439 cluster_set_next_flag(&list
->tail
, idx
, 0);
443 static unsigned int cluster_list_del_first(struct swap_cluster_list
*list
,
444 struct swap_cluster_info
*ci
)
448 idx
= cluster_next(&list
->head
);
449 if (cluster_next(&list
->tail
) == idx
) {
450 cluster_set_null(&list
->head
);
451 cluster_set_null(&list
->tail
);
453 cluster_set_next_flag(&list
->head
,
454 cluster_next(&ci
[idx
]), 0);
459 /* Add a cluster to discard list and schedule it to do discard */
460 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
464 * If scan_swap_map_slots() can't find a free cluster, it will check
465 * si->swap_map directly. To make sure the discarding cluster isn't
466 * taken by scan_swap_map_slots(), mark the swap entries bad (occupied).
467 * It will be cleared after discard
469 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
470 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
472 cluster_list_add_tail(&si
->discard_clusters
, si
->cluster_info
, idx
);
474 schedule_work(&si
->discard_work
);
477 static void __free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
479 struct swap_cluster_info
*ci
= si
->cluster_info
;
481 cluster_set_flag(ci
+ idx
, CLUSTER_FLAG_FREE
);
482 cluster_list_add_tail(&si
->free_clusters
, ci
, idx
);
486 * Doing discard actually. After a cluster discard is finished, the cluster
487 * will be added to free cluster list. caller should hold si->lock.
489 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
491 struct swap_cluster_info
*info
, *ci
;
494 info
= si
->cluster_info
;
496 while (!cluster_list_empty(&si
->discard_clusters
)) {
497 idx
= cluster_list_del_first(&si
->discard_clusters
, info
);
498 spin_unlock(&si
->lock
);
500 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
503 spin_lock(&si
->lock
);
504 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
505 __free_cluster(si
, idx
);
506 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
507 0, SWAPFILE_CLUSTER
);
512 static void swap_discard_work(struct work_struct
*work
)
514 struct swap_info_struct
*si
;
516 si
= container_of(work
, struct swap_info_struct
, discard_work
);
518 spin_lock(&si
->lock
);
519 swap_do_scheduled_discard(si
);
520 spin_unlock(&si
->lock
);
523 static void swap_users_ref_free(struct percpu_ref
*ref
)
525 struct swap_info_struct
*si
;
527 si
= container_of(ref
, struct swap_info_struct
, users
);
531 static void alloc_cluster(struct swap_info_struct
*si
, unsigned long idx
)
533 struct swap_cluster_info
*ci
= si
->cluster_info
;
535 VM_BUG_ON(cluster_list_first(&si
->free_clusters
) != idx
);
536 cluster_list_del_first(&si
->free_clusters
, ci
);
537 cluster_set_count_flag(ci
+ idx
, 0, 0);
540 static void free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
542 struct swap_cluster_info
*ci
= si
->cluster_info
+ idx
;
544 VM_BUG_ON(cluster_count(ci
) != 0);
546 * If the swap is discardable, prepare discard the cluster
547 * instead of free it immediately. The cluster will be freed
550 if ((si
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
551 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
552 swap_cluster_schedule_discard(si
, idx
);
556 __free_cluster(si
, idx
);
560 * The cluster corresponding to page_nr will be used. The cluster will be
561 * removed from free cluster list and its usage counter will be increased.
563 static void inc_cluster_info_page(struct swap_info_struct
*p
,
564 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
566 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
570 if (cluster_is_free(&cluster_info
[idx
]))
571 alloc_cluster(p
, idx
);
573 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
574 cluster_set_count(&cluster_info
[idx
],
575 cluster_count(&cluster_info
[idx
]) + 1);
579 * The cluster corresponding to page_nr decreases one usage. If the usage
580 * counter becomes 0, which means no page in the cluster is in using, we can
581 * optionally discard the cluster and add it to free cluster list.
583 static void dec_cluster_info_page(struct swap_info_struct
*p
,
584 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
586 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
591 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
592 cluster_set_count(&cluster_info
[idx
],
593 cluster_count(&cluster_info
[idx
]) - 1);
595 if (cluster_count(&cluster_info
[idx
]) == 0)
596 free_cluster(p
, idx
);
600 * It's possible scan_swap_map_slots() uses a free cluster in the middle of free
601 * cluster list. Avoiding such abuse to avoid list corruption.
604 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
605 unsigned long offset
)
607 struct percpu_cluster
*percpu_cluster
;
610 offset
/= SWAPFILE_CLUSTER
;
611 conflict
= !cluster_list_empty(&si
->free_clusters
) &&
612 offset
!= cluster_list_first(&si
->free_clusters
) &&
613 cluster_is_free(&si
->cluster_info
[offset
]);
618 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
619 cluster_set_null(&percpu_cluster
->index
);
624 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
625 * might involve allocating a new cluster for current CPU too.
627 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
628 unsigned long *offset
, unsigned long *scan_base
)
630 struct percpu_cluster
*cluster
;
631 struct swap_cluster_info
*ci
;
632 unsigned long tmp
, max
;
635 cluster
= this_cpu_ptr(si
->percpu_cluster
);
636 if (cluster_is_null(&cluster
->index
)) {
637 if (!cluster_list_empty(&si
->free_clusters
)) {
638 cluster
->index
= si
->free_clusters
.head
;
639 cluster
->next
= cluster_next(&cluster
->index
) *
641 } else if (!cluster_list_empty(&si
->discard_clusters
)) {
643 * we don't have free cluster but have some clusters in
644 * discarding, do discard now and reclaim them, then
645 * reread cluster_next_cpu since we dropped si->lock
647 swap_do_scheduled_discard(si
);
648 *scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
649 *offset
= *scan_base
;
656 * Other CPUs can use our cluster if they can't find a free cluster,
657 * check if there is still free entry in the cluster
660 max
= min_t(unsigned long, si
->max
,
661 (cluster_next(&cluster
->index
) + 1) * SWAPFILE_CLUSTER
);
663 ci
= lock_cluster(si
, tmp
);
665 if (!si
->swap_map
[tmp
])
672 cluster_set_null(&cluster
->index
);
675 cluster
->next
= tmp
+ 1;
681 static void __del_from_avail_list(struct swap_info_struct
*p
)
685 assert_spin_locked(&p
->lock
);
687 plist_del(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
690 static void del_from_avail_list(struct swap_info_struct
*p
)
692 spin_lock(&swap_avail_lock
);
693 __del_from_avail_list(p
);
694 spin_unlock(&swap_avail_lock
);
697 static void swap_range_alloc(struct swap_info_struct
*si
, unsigned long offset
,
698 unsigned int nr_entries
)
700 unsigned int end
= offset
+ nr_entries
- 1;
702 if (offset
== si
->lowest_bit
)
703 si
->lowest_bit
+= nr_entries
;
704 if (end
== si
->highest_bit
)
705 WRITE_ONCE(si
->highest_bit
, si
->highest_bit
- nr_entries
);
706 WRITE_ONCE(si
->inuse_pages
, si
->inuse_pages
+ nr_entries
);
707 if (si
->inuse_pages
== si
->pages
) {
708 si
->lowest_bit
= si
->max
;
710 del_from_avail_list(si
);
714 static void add_to_avail_list(struct swap_info_struct
*p
)
718 spin_lock(&swap_avail_lock
);
720 plist_add(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
721 spin_unlock(&swap_avail_lock
);
724 static void swap_range_free(struct swap_info_struct
*si
, unsigned long offset
,
725 unsigned int nr_entries
)
727 unsigned long begin
= offset
;
728 unsigned long end
= offset
+ nr_entries
- 1;
729 void (*swap_slot_free_notify
)(struct block_device
*, unsigned long);
731 if (offset
< si
->lowest_bit
)
732 si
->lowest_bit
= offset
;
733 if (end
> si
->highest_bit
) {
734 bool was_full
= !si
->highest_bit
;
736 WRITE_ONCE(si
->highest_bit
, end
);
737 if (was_full
&& (si
->flags
& SWP_WRITEOK
))
738 add_to_avail_list(si
);
740 atomic_long_add(nr_entries
, &nr_swap_pages
);
741 WRITE_ONCE(si
->inuse_pages
, si
->inuse_pages
- nr_entries
);
742 if (si
->flags
& SWP_BLKDEV
)
743 swap_slot_free_notify
=
744 si
->bdev
->bd_disk
->fops
->swap_slot_free_notify
;
746 swap_slot_free_notify
= NULL
;
747 while (offset
<= end
) {
748 arch_swap_invalidate_page(si
->type
, offset
);
749 zswap_invalidate(si
->type
, offset
);
750 if (swap_slot_free_notify
)
751 swap_slot_free_notify(si
->bdev
, offset
);
754 clear_shadow_from_swap_cache(si
->type
, begin
, end
);
757 static void set_cluster_next(struct swap_info_struct
*si
, unsigned long next
)
761 if (!(si
->flags
& SWP_SOLIDSTATE
)) {
762 si
->cluster_next
= next
;
766 prev
= this_cpu_read(*si
->cluster_next_cpu
);
768 * Cross the swap address space size aligned trunk, choose
769 * another trunk randomly to avoid lock contention on swap
770 * address space if possible.
772 if ((prev
>> SWAP_ADDRESS_SPACE_SHIFT
) !=
773 (next
>> SWAP_ADDRESS_SPACE_SHIFT
)) {
774 /* No free swap slots available */
775 if (si
->highest_bit
<= si
->lowest_bit
)
777 next
= get_random_u32_inclusive(si
->lowest_bit
, si
->highest_bit
);
778 next
= ALIGN_DOWN(next
, SWAP_ADDRESS_SPACE_PAGES
);
779 next
= max_t(unsigned int, next
, si
->lowest_bit
);
781 this_cpu_write(*si
->cluster_next_cpu
, next
);
784 static bool swap_offset_available_and_locked(struct swap_info_struct
*si
,
785 unsigned long offset
)
787 if (data_race(!si
->swap_map
[offset
])) {
788 spin_lock(&si
->lock
);
792 if (vm_swap_full() && READ_ONCE(si
->swap_map
[offset
]) == SWAP_HAS_CACHE
) {
793 spin_lock(&si
->lock
);
800 static int scan_swap_map_slots(struct swap_info_struct
*si
,
801 unsigned char usage
, int nr
,
804 struct swap_cluster_info
*ci
;
805 unsigned long offset
;
806 unsigned long scan_base
;
807 unsigned long last_in_cluster
= 0;
808 int latency_ration
= LATENCY_LIMIT
;
810 bool scanned_many
= false;
813 * We try to cluster swap pages by allocating them sequentially
814 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
815 * way, however, we resort to first-free allocation, starting
816 * a new cluster. This prevents us from scattering swap pages
817 * all over the entire swap partition, so that we reduce
818 * overall disk seek times between swap pages. -- sct
819 * But we do now try to find an empty cluster. -Andrea
820 * And we let swap pages go all over an SSD partition. Hugh
823 si
->flags
+= SWP_SCANNING
;
825 * Use percpu scan base for SSD to reduce lock contention on
826 * cluster and swap cache. For HDD, sequential access is more
829 if (si
->flags
& SWP_SOLIDSTATE
)
830 scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
832 scan_base
= si
->cluster_next
;
836 if (si
->cluster_info
) {
837 if (!scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
839 } else if (unlikely(!si
->cluster_nr
--)) {
840 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
841 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
845 spin_unlock(&si
->lock
);
848 * If seek is expensive, start searching for new cluster from
849 * start of partition, to minimize the span of allocated swap.
850 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
851 * case, just handled by scan_swap_map_try_ssd_cluster() above.
853 scan_base
= offset
= si
->lowest_bit
;
854 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
856 /* Locate the first empty (unaligned) cluster */
857 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
858 if (si
->swap_map
[offset
])
859 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
860 else if (offset
== last_in_cluster
) {
861 spin_lock(&si
->lock
);
862 offset
-= SWAPFILE_CLUSTER
- 1;
863 si
->cluster_next
= offset
;
864 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
867 if (unlikely(--latency_ration
< 0)) {
869 latency_ration
= LATENCY_LIMIT
;
874 spin_lock(&si
->lock
);
875 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
879 if (si
->cluster_info
) {
880 while (scan_swap_map_ssd_cluster_conflict(si
, offset
)) {
881 /* take a break if we already got some slots */
884 if (!scan_swap_map_try_ssd_cluster(si
, &offset
,
889 if (!(si
->flags
& SWP_WRITEOK
))
891 if (!si
->highest_bit
)
893 if (offset
> si
->highest_bit
)
894 scan_base
= offset
= si
->lowest_bit
;
896 ci
= lock_cluster(si
, offset
);
897 /* reuse swap entry of cache-only swap if not busy. */
898 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
901 spin_unlock(&si
->lock
);
902 swap_was_freed
= __try_to_reclaim_swap(si
, offset
, TTRS_ANYWAY
);
903 spin_lock(&si
->lock
);
904 /* entry was freed successfully, try to use this again */
907 goto scan
; /* check next one */
910 if (si
->swap_map
[offset
]) {
917 WRITE_ONCE(si
->swap_map
[offset
], usage
);
918 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
921 swap_range_alloc(si
, offset
, 1);
922 slots
[n_ret
++] = swp_entry(si
->type
, offset
);
924 /* got enough slots or reach max slots? */
925 if ((n_ret
== nr
) || (offset
>= si
->highest_bit
))
928 /* search for next available slot */
930 /* time to take a break? */
931 if (unlikely(--latency_ration
< 0)) {
934 spin_unlock(&si
->lock
);
936 spin_lock(&si
->lock
);
937 latency_ration
= LATENCY_LIMIT
;
940 /* try to get more slots in cluster */
941 if (si
->cluster_info
) {
942 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
944 } else if (si
->cluster_nr
&& !si
->swap_map
[++offset
]) {
945 /* non-ssd case, still more slots in cluster? */
951 * Even if there's no free clusters available (fragmented),
952 * try to scan a little more quickly with lock held unless we
953 * have scanned too many slots already.
956 unsigned long scan_limit
;
958 if (offset
< scan_base
)
959 scan_limit
= scan_base
;
961 scan_limit
= si
->highest_bit
;
962 for (; offset
<= scan_limit
&& --latency_ration
> 0;
964 if (!si
->swap_map
[offset
])
970 set_cluster_next(si
, offset
+ 1);
971 si
->flags
-= SWP_SCANNING
;
975 spin_unlock(&si
->lock
);
976 while (++offset
<= READ_ONCE(si
->highest_bit
)) {
977 if (unlikely(--latency_ration
< 0)) {
979 latency_ration
= LATENCY_LIMIT
;
982 if (swap_offset_available_and_locked(si
, offset
))
985 offset
= si
->lowest_bit
;
986 while (offset
< scan_base
) {
987 if (unlikely(--latency_ration
< 0)) {
989 latency_ration
= LATENCY_LIMIT
;
992 if (swap_offset_available_and_locked(si
, offset
))
996 spin_lock(&si
->lock
);
999 si
->flags
-= SWP_SCANNING
;
1003 static int swap_alloc_cluster(struct swap_info_struct
*si
, swp_entry_t
*slot
)
1006 struct swap_cluster_info
*ci
;
1007 unsigned long offset
;
1010 * Should not even be attempting cluster allocations when huge
1011 * page swap is disabled. Warn and fail the allocation.
1013 if (!IS_ENABLED(CONFIG_THP_SWAP
)) {
1018 if (cluster_list_empty(&si
->free_clusters
))
1021 idx
= cluster_list_first(&si
->free_clusters
);
1022 offset
= idx
* SWAPFILE_CLUSTER
;
1023 ci
= lock_cluster(si
, offset
);
1024 alloc_cluster(si
, idx
);
1025 cluster_set_count_flag(ci
, SWAPFILE_CLUSTER
, CLUSTER_FLAG_HUGE
);
1027 memset(si
->swap_map
+ offset
, SWAP_HAS_CACHE
, SWAPFILE_CLUSTER
);
1029 swap_range_alloc(si
, offset
, SWAPFILE_CLUSTER
);
1030 *slot
= swp_entry(si
->type
, offset
);
1035 static void swap_free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
1037 unsigned long offset
= idx
* SWAPFILE_CLUSTER
;
1038 struct swap_cluster_info
*ci
;
1040 ci
= lock_cluster(si
, offset
);
1041 memset(si
->swap_map
+ offset
, 0, SWAPFILE_CLUSTER
);
1042 cluster_set_count_flag(ci
, 0, 0);
1043 free_cluster(si
, idx
);
1045 swap_range_free(si
, offset
, SWAPFILE_CLUSTER
);
1048 int get_swap_pages(int n_goal
, swp_entry_t swp_entries
[], int entry_size
)
1050 unsigned long size
= swap_entry_size(entry_size
);
1051 struct swap_info_struct
*si
, *next
;
1056 /* Only single cluster request supported */
1057 WARN_ON_ONCE(n_goal
> 1 && size
== SWAPFILE_CLUSTER
);
1059 spin_lock(&swap_avail_lock
);
1061 avail_pgs
= atomic_long_read(&nr_swap_pages
) / size
;
1062 if (avail_pgs
<= 0) {
1063 spin_unlock(&swap_avail_lock
);
1067 n_goal
= min3((long)n_goal
, (long)SWAP_BATCH
, avail_pgs
);
1069 atomic_long_sub(n_goal
* size
, &nr_swap_pages
);
1072 node
= numa_node_id();
1073 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[node
], avail_lists
[node
]) {
1074 /* requeue si to after same-priority siblings */
1075 plist_requeue(&si
->avail_lists
[node
], &swap_avail_heads
[node
]);
1076 spin_unlock(&swap_avail_lock
);
1077 spin_lock(&si
->lock
);
1078 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
1079 spin_lock(&swap_avail_lock
);
1080 if (plist_node_empty(&si
->avail_lists
[node
])) {
1081 spin_unlock(&si
->lock
);
1084 WARN(!si
->highest_bit
,
1085 "swap_info %d in list but !highest_bit\n",
1087 WARN(!(si
->flags
& SWP_WRITEOK
),
1088 "swap_info %d in list but !SWP_WRITEOK\n",
1090 __del_from_avail_list(si
);
1091 spin_unlock(&si
->lock
);
1094 if (size
== SWAPFILE_CLUSTER
) {
1095 if (si
->flags
& SWP_BLKDEV
)
1096 n_ret
= swap_alloc_cluster(si
, swp_entries
);
1098 n_ret
= scan_swap_map_slots(si
, SWAP_HAS_CACHE
,
1099 n_goal
, swp_entries
);
1100 spin_unlock(&si
->lock
);
1101 if (n_ret
|| size
== SWAPFILE_CLUSTER
)
1105 spin_lock(&swap_avail_lock
);
1108 * if we got here, it's likely that si was almost full before,
1109 * and since scan_swap_map_slots() can drop the si->lock,
1110 * multiple callers probably all tried to get a page from the
1111 * same si and it filled up before we could get one; or, the si
1112 * filled up between us dropping swap_avail_lock and taking
1113 * si->lock. Since we dropped the swap_avail_lock, the
1114 * swap_avail_head list may have been modified; so if next is
1115 * still in the swap_avail_head list then try it, otherwise
1116 * start over if we have not gotten any slots.
1118 if (plist_node_empty(&next
->avail_lists
[node
]))
1122 spin_unlock(&swap_avail_lock
);
1126 atomic_long_add((long)(n_goal
- n_ret
) * size
,
1132 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
1134 struct swap_info_struct
*p
;
1135 unsigned long offset
;
1139 p
= swp_swap_info(entry
);
1142 if (data_race(!(p
->flags
& SWP_USED
)))
1144 offset
= swp_offset(entry
);
1145 if (offset
>= p
->max
)
1147 if (data_race(!p
->swap_map
[swp_offset(entry
)]))
1152 pr_err("%s: %s%08lx\n", __func__
, Unused_offset
, entry
.val
);
1155 pr_err("%s: %s%08lx\n", __func__
, Bad_offset
, entry
.val
);
1158 pr_err("%s: %s%08lx\n", __func__
, Unused_file
, entry
.val
);
1161 pr_err("%s: %s%08lx\n", __func__
, Bad_file
, entry
.val
);
1166 static struct swap_info_struct
*swap_info_get_cont(swp_entry_t entry
,
1167 struct swap_info_struct
*q
)
1169 struct swap_info_struct
*p
;
1171 p
= _swap_info_get(entry
);
1175 spin_unlock(&q
->lock
);
1177 spin_lock(&p
->lock
);
1182 static unsigned char __swap_entry_free_locked(struct swap_info_struct
*p
,
1183 unsigned long offset
,
1184 unsigned char usage
)
1186 unsigned char count
;
1187 unsigned char has_cache
;
1189 count
= p
->swap_map
[offset
];
1191 has_cache
= count
& SWAP_HAS_CACHE
;
1192 count
&= ~SWAP_HAS_CACHE
;
1194 if (usage
== SWAP_HAS_CACHE
) {
1195 VM_BUG_ON(!has_cache
);
1197 } else if (count
== SWAP_MAP_SHMEM
) {
1199 * Or we could insist on shmem.c using a special
1200 * swap_shmem_free() and free_shmem_swap_and_cache()...
1203 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
1204 if (count
== COUNT_CONTINUED
) {
1205 if (swap_count_continued(p
, offset
, count
))
1206 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
1208 count
= SWAP_MAP_MAX
;
1213 usage
= count
| has_cache
;
1215 WRITE_ONCE(p
->swap_map
[offset
], usage
);
1217 WRITE_ONCE(p
->swap_map
[offset
], SWAP_HAS_CACHE
);
1223 * When we get a swap entry, if there aren't some other ways to
1224 * prevent swapoff, such as the folio in swap cache is locked, page
1225 * table lock is held, etc., the swap entry may become invalid because
1226 * of swapoff. Then, we need to enclose all swap related functions
1227 * with get_swap_device() and put_swap_device(), unless the swap
1228 * functions call get/put_swap_device() by themselves.
1230 * Check whether swap entry is valid in the swap device. If so,
1231 * return pointer to swap_info_struct, and keep the swap entry valid
1232 * via preventing the swap device from being swapoff, until
1233 * put_swap_device() is called. Otherwise return NULL.
1235 * Notice that swapoff or swapoff+swapon can still happen before the
1236 * percpu_ref_tryget_live() in get_swap_device() or after the
1237 * percpu_ref_put() in put_swap_device() if there isn't any other way
1238 * to prevent swapoff. The caller must be prepared for that. For
1239 * example, the following situation is possible.
1243 * ... swapoff+swapon
1244 * __read_swap_cache_async()
1245 * swapcache_prepare()
1246 * __swap_duplicate()
1248 * // verify PTE not changed
1250 * In __swap_duplicate(), the swap_map need to be checked before
1251 * changing partly because the specified swap entry may be for another
1252 * swap device which has been swapoff. And in do_swap_page(), after
1253 * the page is read from the swap device, the PTE is verified not
1254 * changed with the page table locked to check whether the swap device
1255 * has been swapoff or swapoff+swapon.
1257 struct swap_info_struct
*get_swap_device(swp_entry_t entry
)
1259 struct swap_info_struct
*si
;
1260 unsigned long offset
;
1264 si
= swp_swap_info(entry
);
1267 if (!percpu_ref_tryget_live(&si
->users
))
1270 * Guarantee the si->users are checked before accessing other
1271 * fields of swap_info_struct.
1273 * Paired with the spin_unlock() after setup_swap_info() in
1274 * enable_swap_info().
1277 offset
= swp_offset(entry
);
1278 if (offset
>= si
->max
)
1283 pr_err("%s: %s%08lx\n", __func__
, Bad_file
, entry
.val
);
1287 pr_err("%s: %s%08lx\n", __func__
, Bad_offset
, entry
.val
);
1288 percpu_ref_put(&si
->users
);
1292 static unsigned char __swap_entry_free(struct swap_info_struct
*p
,
1295 struct swap_cluster_info
*ci
;
1296 unsigned long offset
= swp_offset(entry
);
1297 unsigned char usage
;
1299 ci
= lock_cluster_or_swap_info(p
, offset
);
1300 usage
= __swap_entry_free_locked(p
, offset
, 1);
1301 unlock_cluster_or_swap_info(p
, ci
);
1303 free_swap_slot(entry
);
1308 static void swap_entry_free(struct swap_info_struct
*p
, swp_entry_t entry
)
1310 struct swap_cluster_info
*ci
;
1311 unsigned long offset
= swp_offset(entry
);
1312 unsigned char count
;
1314 ci
= lock_cluster(p
, offset
);
1315 count
= p
->swap_map
[offset
];
1316 VM_BUG_ON(count
!= SWAP_HAS_CACHE
);
1317 p
->swap_map
[offset
] = 0;
1318 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
1321 mem_cgroup_uncharge_swap(entry
, 1);
1322 swap_range_free(p
, offset
, 1);
1326 * Caller has made sure that the swap device corresponding to entry
1327 * is still around or has not been recycled.
1329 void swap_free(swp_entry_t entry
)
1331 struct swap_info_struct
*p
;
1333 p
= _swap_info_get(entry
);
1335 __swap_entry_free(p
, entry
);
1339 * Called after dropping swapcache to decrease refcnt to swap entries.
1341 void put_swap_folio(struct folio
*folio
, swp_entry_t entry
)
1343 unsigned long offset
= swp_offset(entry
);
1344 unsigned long idx
= offset
/ SWAPFILE_CLUSTER
;
1345 struct swap_cluster_info
*ci
;
1346 struct swap_info_struct
*si
;
1348 unsigned int i
, free_entries
= 0;
1350 int size
= swap_entry_size(folio_nr_pages(folio
));
1352 si
= _swap_info_get(entry
);
1356 ci
= lock_cluster_or_swap_info(si
, offset
);
1357 if (size
== SWAPFILE_CLUSTER
) {
1358 VM_BUG_ON(!cluster_is_huge(ci
));
1359 map
= si
->swap_map
+ offset
;
1360 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1362 VM_BUG_ON(!(val
& SWAP_HAS_CACHE
));
1363 if (val
== SWAP_HAS_CACHE
)
1366 cluster_clear_huge(ci
);
1367 if (free_entries
== SWAPFILE_CLUSTER
) {
1368 unlock_cluster_or_swap_info(si
, ci
);
1369 spin_lock(&si
->lock
);
1370 mem_cgroup_uncharge_swap(entry
, SWAPFILE_CLUSTER
);
1371 swap_free_cluster(si
, idx
);
1372 spin_unlock(&si
->lock
);
1376 for (i
= 0; i
< size
; i
++, entry
.val
++) {
1377 if (!__swap_entry_free_locked(si
, offset
+ i
, SWAP_HAS_CACHE
)) {
1378 unlock_cluster_or_swap_info(si
, ci
);
1379 free_swap_slot(entry
);
1382 lock_cluster_or_swap_info(si
, offset
);
1385 unlock_cluster_or_swap_info(si
, ci
);
1388 #ifdef CONFIG_THP_SWAP
1389 int split_swap_cluster(swp_entry_t entry
)
1391 struct swap_info_struct
*si
;
1392 struct swap_cluster_info
*ci
;
1393 unsigned long offset
= swp_offset(entry
);
1395 si
= _swap_info_get(entry
);
1398 ci
= lock_cluster(si
, offset
);
1399 cluster_clear_huge(ci
);
1405 static int swp_entry_cmp(const void *ent1
, const void *ent2
)
1407 const swp_entry_t
*e1
= ent1
, *e2
= ent2
;
1409 return (int)swp_type(*e1
) - (int)swp_type(*e2
);
1412 void swapcache_free_entries(swp_entry_t
*entries
, int n
)
1414 struct swap_info_struct
*p
, *prev
;
1424 * Sort swap entries by swap device, so each lock is only taken once.
1425 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1426 * so low that it isn't necessary to optimize further.
1428 if (nr_swapfiles
> 1)
1429 sort(entries
, n
, sizeof(entries
[0]), swp_entry_cmp
, NULL
);
1430 for (i
= 0; i
< n
; ++i
) {
1431 p
= swap_info_get_cont(entries
[i
], prev
);
1433 swap_entry_free(p
, entries
[i
]);
1437 spin_unlock(&p
->lock
);
1440 int __swap_count(swp_entry_t entry
)
1442 struct swap_info_struct
*si
= swp_swap_info(entry
);
1443 pgoff_t offset
= swp_offset(entry
);
1445 return swap_count(si
->swap_map
[offset
]);
1449 * How many references to @entry are currently swapped out?
1450 * This does not give an exact answer when swap count is continued,
1451 * but does include the high COUNT_CONTINUED flag to allow for that.
1453 int swap_swapcount(struct swap_info_struct
*si
, swp_entry_t entry
)
1455 pgoff_t offset
= swp_offset(entry
);
1456 struct swap_cluster_info
*ci
;
1459 ci
= lock_cluster_or_swap_info(si
, offset
);
1460 count
= swap_count(si
->swap_map
[offset
]);
1461 unlock_cluster_or_swap_info(si
, ci
);
1466 * How many references to @entry are currently swapped out?
1467 * This considers COUNT_CONTINUED so it returns exact answer.
1469 int swp_swapcount(swp_entry_t entry
)
1471 int count
, tmp_count
, n
;
1472 struct swap_info_struct
*p
;
1473 struct swap_cluster_info
*ci
;
1478 p
= _swap_info_get(entry
);
1482 offset
= swp_offset(entry
);
1484 ci
= lock_cluster_or_swap_info(p
, offset
);
1486 count
= swap_count(p
->swap_map
[offset
]);
1487 if (!(count
& COUNT_CONTINUED
))
1490 count
&= ~COUNT_CONTINUED
;
1491 n
= SWAP_MAP_MAX
+ 1;
1493 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1494 offset
&= ~PAGE_MASK
;
1495 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1498 page
= list_next_entry(page
, lru
);
1499 map
= kmap_local_page(page
);
1500 tmp_count
= map
[offset
];
1503 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1504 n
*= (SWAP_CONT_MAX
+ 1);
1505 } while (tmp_count
& COUNT_CONTINUED
);
1507 unlock_cluster_or_swap_info(p
, ci
);
1511 static bool swap_page_trans_huge_swapped(struct swap_info_struct
*si
,
1514 struct swap_cluster_info
*ci
;
1515 unsigned char *map
= si
->swap_map
;
1516 unsigned long roffset
= swp_offset(entry
);
1517 unsigned long offset
= round_down(roffset
, SWAPFILE_CLUSTER
);
1521 ci
= lock_cluster_or_swap_info(si
, offset
);
1522 if (!ci
|| !cluster_is_huge(ci
)) {
1523 if (swap_count(map
[roffset
]))
1527 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1528 if (swap_count(map
[offset
+ i
])) {
1534 unlock_cluster_or_swap_info(si
, ci
);
1538 static bool folio_swapped(struct folio
*folio
)
1540 swp_entry_t entry
= folio
->swap
;
1541 struct swap_info_struct
*si
= _swap_info_get(entry
);
1546 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!folio_test_large(folio
)))
1547 return swap_swapcount(si
, entry
) != 0;
1549 return swap_page_trans_huge_swapped(si
, entry
);
1553 * folio_free_swap() - Free the swap space used for this folio.
1554 * @folio: The folio to remove.
1556 * If swap is getting full, or if there are no more mappings of this folio,
1557 * then call folio_free_swap to free its swap space.
1559 * Return: true if we were able to release the swap space.
1561 bool folio_free_swap(struct folio
*folio
)
1563 VM_BUG_ON_FOLIO(!folio_test_locked(folio
), folio
);
1565 if (!folio_test_swapcache(folio
))
1567 if (folio_test_writeback(folio
))
1569 if (folio_swapped(folio
))
1573 * Once hibernation has begun to create its image of memory,
1574 * there's a danger that one of the calls to folio_free_swap()
1575 * - most probably a call from __try_to_reclaim_swap() while
1576 * hibernation is allocating its own swap pages for the image,
1577 * but conceivably even a call from memory reclaim - will free
1578 * the swap from a folio which has already been recorded in the
1579 * image as a clean swapcache folio, and then reuse its swap for
1580 * another page of the image. On waking from hibernation, the
1581 * original folio might be freed under memory pressure, then
1582 * later read back in from swap, now with the wrong data.
1584 * Hibernation suspends storage while it is writing the image
1585 * to disk so check that here.
1587 if (pm_suspended_storage())
1590 delete_from_swap_cache(folio
);
1591 folio_set_dirty(folio
);
1596 * Free the swap entry like above, but also try to
1597 * free the page cache entry if it is the last user.
1599 int free_swap_and_cache(swp_entry_t entry
)
1601 struct swap_info_struct
*p
;
1602 unsigned char count
;
1604 if (non_swap_entry(entry
))
1607 p
= _swap_info_get(entry
);
1609 count
= __swap_entry_free(p
, entry
);
1610 if (count
== SWAP_HAS_CACHE
&&
1611 !swap_page_trans_huge_swapped(p
, entry
))
1612 __try_to_reclaim_swap(p
, swp_offset(entry
),
1613 TTRS_UNMAPPED
| TTRS_FULL
);
1618 #ifdef CONFIG_HIBERNATION
1620 swp_entry_t
get_swap_page_of_type(int type
)
1622 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1623 swp_entry_t entry
= {0};
1628 /* This is called for allocating swap entry, not cache */
1629 spin_lock(&si
->lock
);
1630 if ((si
->flags
& SWP_WRITEOK
) && scan_swap_map_slots(si
, 1, 1, &entry
))
1631 atomic_long_dec(&nr_swap_pages
);
1632 spin_unlock(&si
->lock
);
1638 * Find the swap type that corresponds to given device (if any).
1640 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1641 * from 0, in which the swap header is expected to be located.
1643 * This is needed for the suspend to disk (aka swsusp).
1645 int swap_type_of(dev_t device
, sector_t offset
)
1652 spin_lock(&swap_lock
);
1653 for (type
= 0; type
< nr_swapfiles
; type
++) {
1654 struct swap_info_struct
*sis
= swap_info
[type
];
1656 if (!(sis
->flags
& SWP_WRITEOK
))
1659 if (device
== sis
->bdev
->bd_dev
) {
1660 struct swap_extent
*se
= first_se(sis
);
1662 if (se
->start_block
== offset
) {
1663 spin_unlock(&swap_lock
);
1668 spin_unlock(&swap_lock
);
1672 int find_first_swap(dev_t
*device
)
1676 spin_lock(&swap_lock
);
1677 for (type
= 0; type
< nr_swapfiles
; type
++) {
1678 struct swap_info_struct
*sis
= swap_info
[type
];
1680 if (!(sis
->flags
& SWP_WRITEOK
))
1682 *device
= sis
->bdev
->bd_dev
;
1683 spin_unlock(&swap_lock
);
1686 spin_unlock(&swap_lock
);
1691 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1692 * corresponding to given index in swap_info (swap type).
1694 sector_t
swapdev_block(int type
, pgoff_t offset
)
1696 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1697 struct swap_extent
*se
;
1699 if (!si
|| !(si
->flags
& SWP_WRITEOK
))
1701 se
= offset_to_swap_extent(si
, offset
);
1702 return se
->start_block
+ (offset
- se
->start_page
);
1706 * Return either the total number of swap pages of given type, or the number
1707 * of free pages of that type (depending on @free)
1709 * This is needed for software suspend
1711 unsigned int count_swap_pages(int type
, int free
)
1715 spin_lock(&swap_lock
);
1716 if ((unsigned int)type
< nr_swapfiles
) {
1717 struct swap_info_struct
*sis
= swap_info
[type
];
1719 spin_lock(&sis
->lock
);
1720 if (sis
->flags
& SWP_WRITEOK
) {
1723 n
-= sis
->inuse_pages
;
1725 spin_unlock(&sis
->lock
);
1727 spin_unlock(&swap_lock
);
1730 #endif /* CONFIG_HIBERNATION */
1732 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1734 return pte_same(pte_swp_clear_flags(pte
), swp_pte
);
1738 * No need to decide whether this PTE shares the swap entry with others,
1739 * just let do_wp_page work it out if a write is requested later - to
1740 * force COW, vm_page_prot omits write permission from any private vma.
1742 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1743 unsigned long addr
, swp_entry_t entry
, struct folio
*folio
)
1746 struct folio
*swapcache
;
1748 pte_t
*pte
, new_pte
, old_pte
;
1749 bool hwpoisoned
= false;
1753 folio
= ksm_might_need_to_copy(folio
, vma
, addr
);
1754 if (unlikely(!folio
))
1756 else if (unlikely(folio
== ERR_PTR(-EHWPOISON
))) {
1761 page
= folio_file_page(folio
, swp_offset(entry
));
1762 if (PageHWPoison(page
))
1765 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1766 if (unlikely(!pte
|| !pte_same_as_swp(ptep_get(pte
),
1767 swp_entry_to_pte(entry
)))) {
1772 old_pte
= ptep_get(pte
);
1774 if (unlikely(hwpoisoned
|| !folio_test_uptodate(folio
))) {
1775 swp_entry_t swp_entry
;
1777 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1779 swp_entry
= make_hwpoison_entry(page
);
1781 swp_entry
= make_poisoned_swp_entry();
1783 new_pte
= swp_entry_to_pte(swp_entry
);
1789 * Some architectures may have to restore extra metadata to the page
1790 * when reading from swap. This metadata may be indexed by swap entry
1791 * so this must be called before swap_free().
1793 arch_swap_restore(entry
, folio
);
1795 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1796 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1798 if (folio
== swapcache
) {
1799 rmap_t rmap_flags
= RMAP_NONE
;
1802 * See do_swap_page(): writeback would be problematic.
1803 * However, we do a folio_wait_writeback() just before this
1804 * call and have the folio locked.
1806 VM_BUG_ON_FOLIO(folio_test_writeback(folio
), folio
);
1807 if (pte_swp_exclusive(old_pte
))
1808 rmap_flags
|= RMAP_EXCLUSIVE
;
1810 folio_add_anon_rmap_pte(folio
, page
, vma
, addr
, rmap_flags
);
1811 } else { /* ksm created a completely new copy */
1812 folio_add_new_anon_rmap(folio
, vma
, addr
);
1813 folio_add_lru_vma(folio
, vma
);
1815 new_pte
= pte_mkold(mk_pte(page
, vma
->vm_page_prot
));
1816 if (pte_swp_soft_dirty(old_pte
))
1817 new_pte
= pte_mksoft_dirty(new_pte
);
1818 if (pte_swp_uffd_wp(old_pte
))
1819 new_pte
= pte_mkuffd_wp(new_pte
);
1821 set_pte_at(vma
->vm_mm
, addr
, pte
, new_pte
);
1825 pte_unmap_unlock(pte
, ptl
);
1826 if (folio
!= swapcache
) {
1827 folio_unlock(folio
);
1833 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1834 unsigned long addr
, unsigned long end
,
1838 struct swap_info_struct
*si
;
1840 si
= swap_info
[type
];
1842 struct folio
*folio
;
1843 unsigned long offset
;
1844 unsigned char swp_count
;
1850 pte
= pte_offset_map(pmd
, addr
);
1855 ptent
= ptep_get_lockless(pte
);
1857 if (!is_swap_pte(ptent
))
1860 entry
= pte_to_swp_entry(ptent
);
1861 if (swp_type(entry
) != type
)
1864 offset
= swp_offset(entry
);
1868 folio
= swap_cache_get_folio(entry
, vma
, addr
);
1871 struct vm_fault vmf
= {
1874 .real_address
= addr
,
1878 page
= swapin_readahead(entry
, GFP_HIGHUSER_MOVABLE
,
1881 folio
= page_folio(page
);
1884 swp_count
= READ_ONCE(si
->swap_map
[offset
]);
1885 if (swp_count
== 0 || swp_count
== SWAP_MAP_BAD
)
1891 folio_wait_writeback(folio
);
1892 ret
= unuse_pte(vma
, pmd
, addr
, entry
, folio
);
1894 folio_unlock(folio
);
1899 folio_free_swap(folio
);
1900 folio_unlock(folio
);
1902 } while (addr
+= PAGE_SIZE
, addr
!= end
);
1909 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
1910 unsigned long addr
, unsigned long end
,
1917 pmd
= pmd_offset(pud
, addr
);
1920 next
= pmd_addr_end(addr
, end
);
1921 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, type
);
1924 } while (pmd
++, addr
= next
, addr
!= end
);
1928 static inline int unuse_pud_range(struct vm_area_struct
*vma
, p4d_t
*p4d
,
1929 unsigned long addr
, unsigned long end
,
1936 pud
= pud_offset(p4d
, addr
);
1938 next
= pud_addr_end(addr
, end
);
1939 if (pud_none_or_clear_bad(pud
))
1941 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, type
);
1944 } while (pud
++, addr
= next
, addr
!= end
);
1948 static inline int unuse_p4d_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
1949 unsigned long addr
, unsigned long end
,
1956 p4d
= p4d_offset(pgd
, addr
);
1958 next
= p4d_addr_end(addr
, end
);
1959 if (p4d_none_or_clear_bad(p4d
))
1961 ret
= unuse_pud_range(vma
, p4d
, addr
, next
, type
);
1964 } while (p4d
++, addr
= next
, addr
!= end
);
1968 static int unuse_vma(struct vm_area_struct
*vma
, unsigned int type
)
1971 unsigned long addr
, end
, next
;
1974 addr
= vma
->vm_start
;
1977 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1979 next
= pgd_addr_end(addr
, end
);
1980 if (pgd_none_or_clear_bad(pgd
))
1982 ret
= unuse_p4d_range(vma
, pgd
, addr
, next
, type
);
1985 } while (pgd
++, addr
= next
, addr
!= end
);
1989 static int unuse_mm(struct mm_struct
*mm
, unsigned int type
)
1991 struct vm_area_struct
*vma
;
1993 VMA_ITERATOR(vmi
, mm
, 0);
1996 for_each_vma(vmi
, vma
) {
1997 if (vma
->anon_vma
) {
1998 ret
= unuse_vma(vma
, type
);
2005 mmap_read_unlock(mm
);
2010 * Scan swap_map from current position to next entry still in use.
2011 * Return 0 if there are no inuse entries after prev till end of
2014 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
2018 unsigned char count
;
2021 * No need for swap_lock here: we're just looking
2022 * for whether an entry is in use, not modifying it; false
2023 * hits are okay, and sys_swapoff() has already prevented new
2024 * allocations from this area (while holding swap_lock).
2026 for (i
= prev
+ 1; i
< si
->max
; i
++) {
2027 count
= READ_ONCE(si
->swap_map
[i
]);
2028 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
2030 if ((i
% LATENCY_LIMIT
) == 0)
2040 static int try_to_unuse(unsigned int type
)
2042 struct mm_struct
*prev_mm
;
2043 struct mm_struct
*mm
;
2044 struct list_head
*p
;
2046 struct swap_info_struct
*si
= swap_info
[type
];
2047 struct folio
*folio
;
2051 if (!READ_ONCE(si
->inuse_pages
))
2055 retval
= shmem_unuse(type
);
2062 spin_lock(&mmlist_lock
);
2063 p
= &init_mm
.mmlist
;
2064 while (READ_ONCE(si
->inuse_pages
) &&
2065 !signal_pending(current
) &&
2066 (p
= p
->next
) != &init_mm
.mmlist
) {
2068 mm
= list_entry(p
, struct mm_struct
, mmlist
);
2069 if (!mmget_not_zero(mm
))
2071 spin_unlock(&mmlist_lock
);
2074 retval
= unuse_mm(mm
, type
);
2081 * Make sure that we aren't completely killing
2082 * interactive performance.
2085 spin_lock(&mmlist_lock
);
2087 spin_unlock(&mmlist_lock
);
2092 while (READ_ONCE(si
->inuse_pages
) &&
2093 !signal_pending(current
) &&
2094 (i
= find_next_to_unuse(si
, i
)) != 0) {
2096 entry
= swp_entry(type
, i
);
2097 folio
= filemap_get_folio(swap_address_space(entry
), i
);
2102 * It is conceivable that a racing task removed this folio from
2103 * swap cache just before we acquired the page lock. The folio
2104 * might even be back in swap cache on another swap area. But
2105 * that is okay, folio_free_swap() only removes stale folios.
2108 folio_wait_writeback(folio
);
2109 folio_free_swap(folio
);
2110 folio_unlock(folio
);
2115 * Lets check again to see if there are still swap entries in the map.
2116 * If yes, we would need to do retry the unuse logic again.
2117 * Under global memory pressure, swap entries can be reinserted back
2118 * into process space after the mmlist loop above passes over them.
2120 * Limit the number of retries? No: when mmget_not_zero()
2121 * above fails, that mm is likely to be freeing swap from
2122 * exit_mmap(), which proceeds at its own independent pace;
2123 * and even shmem_writepage() could have been preempted after
2124 * folio_alloc_swap(), temporarily hiding that swap. It's easy
2125 * and robust (though cpu-intensive) just to keep retrying.
2127 if (READ_ONCE(si
->inuse_pages
)) {
2128 if (!signal_pending(current
))
2137 * After a successful try_to_unuse, if no swap is now in use, we know
2138 * we can empty the mmlist. swap_lock must be held on entry and exit.
2139 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2140 * added to the mmlist just after page_duplicate - before would be racy.
2142 static void drain_mmlist(void)
2144 struct list_head
*p
, *next
;
2147 for (type
= 0; type
< nr_swapfiles
; type
++)
2148 if (swap_info
[type
]->inuse_pages
)
2150 spin_lock(&mmlist_lock
);
2151 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
2153 spin_unlock(&mmlist_lock
);
2157 * Free all of a swapdev's extent information
2159 static void destroy_swap_extents(struct swap_info_struct
*sis
)
2161 while (!RB_EMPTY_ROOT(&sis
->swap_extent_root
)) {
2162 struct rb_node
*rb
= sis
->swap_extent_root
.rb_node
;
2163 struct swap_extent
*se
= rb_entry(rb
, struct swap_extent
, rb_node
);
2165 rb_erase(rb
, &sis
->swap_extent_root
);
2169 if (sis
->flags
& SWP_ACTIVATED
) {
2170 struct file
*swap_file
= sis
->swap_file
;
2171 struct address_space
*mapping
= swap_file
->f_mapping
;
2173 sis
->flags
&= ~SWP_ACTIVATED
;
2174 if (mapping
->a_ops
->swap_deactivate
)
2175 mapping
->a_ops
->swap_deactivate(swap_file
);
2180 * Add a block range (and the corresponding page range) into this swapdev's
2183 * This function rather assumes that it is called in ascending page order.
2186 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
2187 unsigned long nr_pages
, sector_t start_block
)
2189 struct rb_node
**link
= &sis
->swap_extent_root
.rb_node
, *parent
= NULL
;
2190 struct swap_extent
*se
;
2191 struct swap_extent
*new_se
;
2194 * place the new node at the right most since the
2195 * function is called in ascending page order.
2199 link
= &parent
->rb_right
;
2203 se
= rb_entry(parent
, struct swap_extent
, rb_node
);
2204 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
2205 if (se
->start_block
+ se
->nr_pages
== start_block
) {
2207 se
->nr_pages
+= nr_pages
;
2212 /* No merge, insert a new extent. */
2213 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
2216 new_se
->start_page
= start_page
;
2217 new_se
->nr_pages
= nr_pages
;
2218 new_se
->start_block
= start_block
;
2220 rb_link_node(&new_se
->rb_node
, parent
, link
);
2221 rb_insert_color(&new_se
->rb_node
, &sis
->swap_extent_root
);
2224 EXPORT_SYMBOL_GPL(add_swap_extent
);
2227 * A `swap extent' is a simple thing which maps a contiguous range of pages
2228 * onto a contiguous range of disk blocks. A rbtree of swap extents is
2229 * built at swapon time and is then used at swap_writepage/swap_read_folio
2230 * time for locating where on disk a page belongs.
2232 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2233 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2234 * swap files identically.
2236 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2237 * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2238 * swapfiles are handled *identically* after swapon time.
2240 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2241 * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray
2242 * blocks are found which do not fall within the PAGE_SIZE alignment
2243 * requirements, they are simply tossed out - we will never use those blocks
2246 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2247 * prevents users from writing to the swap device, which will corrupt memory.
2249 * The amount of disk space which a single swap extent represents varies.
2250 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2251 * extents in the rbtree. - akpm.
2253 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
2255 struct file
*swap_file
= sis
->swap_file
;
2256 struct address_space
*mapping
= swap_file
->f_mapping
;
2257 struct inode
*inode
= mapping
->host
;
2260 if (S_ISBLK(inode
->i_mode
)) {
2261 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2266 if (mapping
->a_ops
->swap_activate
) {
2267 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
2270 sis
->flags
|= SWP_ACTIVATED
;
2271 if ((sis
->flags
& SWP_FS_OPS
) &&
2272 sio_pool_init() != 0) {
2273 destroy_swap_extents(sis
);
2279 return generic_swapfile_activate(sis
, swap_file
, span
);
2282 static int swap_node(struct swap_info_struct
*p
)
2284 struct block_device
*bdev
;
2289 bdev
= p
->swap_file
->f_inode
->i_sb
->s_bdev
;
2291 return bdev
? bdev
->bd_disk
->node_id
: NUMA_NO_NODE
;
2294 static void setup_swap_info(struct swap_info_struct
*p
, int prio
,
2295 unsigned char *swap_map
,
2296 struct swap_cluster_info
*cluster_info
)
2303 p
->prio
= --least_priority
;
2305 * the plist prio is negated because plist ordering is
2306 * low-to-high, while swap ordering is high-to-low
2308 p
->list
.prio
= -p
->prio
;
2311 p
->avail_lists
[i
].prio
= -p
->prio
;
2313 if (swap_node(p
) == i
)
2314 p
->avail_lists
[i
].prio
= 1;
2316 p
->avail_lists
[i
].prio
= -p
->prio
;
2319 p
->swap_map
= swap_map
;
2320 p
->cluster_info
= cluster_info
;
2323 static void _enable_swap_info(struct swap_info_struct
*p
)
2325 p
->flags
|= SWP_WRITEOK
;
2326 atomic_long_add(p
->pages
, &nr_swap_pages
);
2327 total_swap_pages
+= p
->pages
;
2329 assert_spin_locked(&swap_lock
);
2331 * both lists are plists, and thus priority ordered.
2332 * swap_active_head needs to be priority ordered for swapoff(),
2333 * which on removal of any swap_info_struct with an auto-assigned
2334 * (i.e. negative) priority increments the auto-assigned priority
2335 * of any lower-priority swap_info_structs.
2336 * swap_avail_head needs to be priority ordered for folio_alloc_swap(),
2337 * which allocates swap pages from the highest available priority
2340 plist_add(&p
->list
, &swap_active_head
);
2342 /* add to available list iff swap device is not full */
2344 add_to_avail_list(p
);
2347 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
2348 unsigned char *swap_map
,
2349 struct swap_cluster_info
*cluster_info
)
2351 zswap_swapon(p
->type
);
2353 spin_lock(&swap_lock
);
2354 spin_lock(&p
->lock
);
2355 setup_swap_info(p
, prio
, swap_map
, cluster_info
);
2356 spin_unlock(&p
->lock
);
2357 spin_unlock(&swap_lock
);
2359 * Finished initializing swap device, now it's safe to reference it.
2361 percpu_ref_resurrect(&p
->users
);
2362 spin_lock(&swap_lock
);
2363 spin_lock(&p
->lock
);
2364 _enable_swap_info(p
);
2365 spin_unlock(&p
->lock
);
2366 spin_unlock(&swap_lock
);
2369 static void reinsert_swap_info(struct swap_info_struct
*p
)
2371 spin_lock(&swap_lock
);
2372 spin_lock(&p
->lock
);
2373 setup_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
2374 _enable_swap_info(p
);
2375 spin_unlock(&p
->lock
);
2376 spin_unlock(&swap_lock
);
2379 bool has_usable_swap(void)
2383 spin_lock(&swap_lock
);
2384 if (plist_head_empty(&swap_active_head
))
2386 spin_unlock(&swap_lock
);
2390 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
2392 struct swap_info_struct
*p
= NULL
;
2393 unsigned char *swap_map
;
2394 struct swap_cluster_info
*cluster_info
;
2395 struct file
*swap_file
, *victim
;
2396 struct address_space
*mapping
;
2397 struct inode
*inode
;
2398 struct filename
*pathname
;
2400 unsigned int old_block_size
;
2402 if (!capable(CAP_SYS_ADMIN
))
2405 BUG_ON(!current
->mm
);
2407 pathname
= getname(specialfile
);
2408 if (IS_ERR(pathname
))
2409 return PTR_ERR(pathname
);
2411 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
2412 err
= PTR_ERR(victim
);
2416 mapping
= victim
->f_mapping
;
2417 spin_lock(&swap_lock
);
2418 plist_for_each_entry(p
, &swap_active_head
, list
) {
2419 if (p
->flags
& SWP_WRITEOK
) {
2420 if (p
->swap_file
->f_mapping
== mapping
) {
2428 spin_unlock(&swap_lock
);
2431 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2432 vm_unacct_memory(p
->pages
);
2435 spin_unlock(&swap_lock
);
2438 spin_lock(&p
->lock
);
2439 del_from_avail_list(p
);
2441 struct swap_info_struct
*si
= p
;
2444 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2447 for_each_node(nid
) {
2448 if (si
->avail_lists
[nid
].prio
!= 1)
2449 si
->avail_lists
[nid
].prio
--;
2454 plist_del(&p
->list
, &swap_active_head
);
2455 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2456 total_swap_pages
-= p
->pages
;
2457 p
->flags
&= ~SWP_WRITEOK
;
2458 spin_unlock(&p
->lock
);
2459 spin_unlock(&swap_lock
);
2461 disable_swap_slots_cache_lock();
2463 set_current_oom_origin();
2464 err
= try_to_unuse(p
->type
);
2465 clear_current_oom_origin();
2468 /* re-insert swap space back into swap_list */
2469 reinsert_swap_info(p
);
2470 reenable_swap_slots_cache_unlock();
2474 reenable_swap_slots_cache_unlock();
2477 * Wait for swap operations protected by get/put_swap_device()
2480 * We need synchronize_rcu() here to protect the accessing to
2481 * the swap cache data structure.
2483 percpu_ref_kill(&p
->users
);
2485 wait_for_completion(&p
->comp
);
2487 flush_work(&p
->discard_work
);
2489 destroy_swap_extents(p
);
2490 if (p
->flags
& SWP_CONTINUED
)
2491 free_swap_count_continuations(p
);
2493 if (!p
->bdev
|| !bdev_nonrot(p
->bdev
))
2494 atomic_dec(&nr_rotate_swap
);
2496 mutex_lock(&swapon_mutex
);
2497 spin_lock(&swap_lock
);
2498 spin_lock(&p
->lock
);
2501 /* wait for anyone still in scan_swap_map_slots */
2502 p
->highest_bit
= 0; /* cuts scans short */
2503 while (p
->flags
>= SWP_SCANNING
) {
2504 spin_unlock(&p
->lock
);
2505 spin_unlock(&swap_lock
);
2506 schedule_timeout_uninterruptible(1);
2507 spin_lock(&swap_lock
);
2508 spin_lock(&p
->lock
);
2511 swap_file
= p
->swap_file
;
2512 old_block_size
= p
->old_block_size
;
2513 p
->swap_file
= NULL
;
2515 swap_map
= p
->swap_map
;
2517 cluster_info
= p
->cluster_info
;
2518 p
->cluster_info
= NULL
;
2519 spin_unlock(&p
->lock
);
2520 spin_unlock(&swap_lock
);
2521 arch_swap_invalidate_area(p
->type
);
2522 zswap_swapoff(p
->type
);
2523 mutex_unlock(&swapon_mutex
);
2524 free_percpu(p
->percpu_cluster
);
2525 p
->percpu_cluster
= NULL
;
2526 free_percpu(p
->cluster_next_cpu
);
2527 p
->cluster_next_cpu
= NULL
;
2529 kvfree(cluster_info
);
2530 /* Destroy swap account information */
2531 swap_cgroup_swapoff(p
->type
);
2532 exit_swap_address_space(p
->type
);
2534 inode
= mapping
->host
;
2536 set_blocksize(p
->bdev
, old_block_size
);
2538 p
->bdev_file
= NULL
;
2542 inode
->i_flags
&= ~S_SWAPFILE
;
2543 inode_unlock(inode
);
2544 filp_close(swap_file
, NULL
);
2547 * Clear the SWP_USED flag after all resources are freed so that swapon
2548 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2549 * not hold p->lock after we cleared its SWP_WRITEOK.
2551 spin_lock(&swap_lock
);
2553 spin_unlock(&swap_lock
);
2556 atomic_inc(&proc_poll_event
);
2557 wake_up_interruptible(&proc_poll_wait
);
2560 filp_close(victim
, NULL
);
2566 #ifdef CONFIG_PROC_FS
2567 static __poll_t
swaps_poll(struct file
*file
, poll_table
*wait
)
2569 struct seq_file
*seq
= file
->private_data
;
2571 poll_wait(file
, &proc_poll_wait
, wait
);
2573 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2574 seq
->poll_event
= atomic_read(&proc_poll_event
);
2575 return EPOLLIN
| EPOLLRDNORM
| EPOLLERR
| EPOLLPRI
;
2578 return EPOLLIN
| EPOLLRDNORM
;
2582 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2584 struct swap_info_struct
*si
;
2588 mutex_lock(&swapon_mutex
);
2591 return SEQ_START_TOKEN
;
2593 for (type
= 0; (si
= swap_type_to_swap_info(type
)); type
++) {
2594 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2603 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2605 struct swap_info_struct
*si
= v
;
2608 if (v
== SEQ_START_TOKEN
)
2611 type
= si
->type
+ 1;
2614 for (; (si
= swap_type_to_swap_info(type
)); type
++) {
2615 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2623 static void swap_stop(struct seq_file
*swap
, void *v
)
2625 mutex_unlock(&swapon_mutex
);
2628 static int swap_show(struct seq_file
*swap
, void *v
)
2630 struct swap_info_struct
*si
= v
;
2633 unsigned long bytes
, inuse
;
2635 if (si
== SEQ_START_TOKEN
) {
2636 seq_puts(swap
, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2640 bytes
= K(si
->pages
);
2641 inuse
= K(READ_ONCE(si
->inuse_pages
));
2643 file
= si
->swap_file
;
2644 len
= seq_file_path(swap
, file
, " \t\n\\");
2645 seq_printf(swap
, "%*s%s\t%lu\t%s%lu\t%s%d\n",
2646 len
< 40 ? 40 - len
: 1, " ",
2647 S_ISBLK(file_inode(file
)->i_mode
) ?
2648 "partition" : "file\t",
2649 bytes
, bytes
< 10000000 ? "\t" : "",
2650 inuse
, inuse
< 10000000 ? "\t" : "",
2655 static const struct seq_operations swaps_op
= {
2656 .start
= swap_start
,
2662 static int swaps_open(struct inode
*inode
, struct file
*file
)
2664 struct seq_file
*seq
;
2667 ret
= seq_open(file
, &swaps_op
);
2671 seq
= file
->private_data
;
2672 seq
->poll_event
= atomic_read(&proc_poll_event
);
2676 static const struct proc_ops swaps_proc_ops
= {
2677 .proc_flags
= PROC_ENTRY_PERMANENT
,
2678 .proc_open
= swaps_open
,
2679 .proc_read
= seq_read
,
2680 .proc_lseek
= seq_lseek
,
2681 .proc_release
= seq_release
,
2682 .proc_poll
= swaps_poll
,
2685 static int __init
procswaps_init(void)
2687 proc_create("swaps", 0, NULL
, &swaps_proc_ops
);
2690 __initcall(procswaps_init
);
2691 #endif /* CONFIG_PROC_FS */
2693 #ifdef MAX_SWAPFILES_CHECK
2694 static int __init
max_swapfiles_check(void)
2696 MAX_SWAPFILES_CHECK();
2699 late_initcall(max_swapfiles_check
);
2702 static struct swap_info_struct
*alloc_swap_info(void)
2704 struct swap_info_struct
*p
;
2705 struct swap_info_struct
*defer
= NULL
;
2709 p
= kvzalloc(struct_size(p
, avail_lists
, nr_node_ids
), GFP_KERNEL
);
2711 return ERR_PTR(-ENOMEM
);
2713 if (percpu_ref_init(&p
->users
, swap_users_ref_free
,
2714 PERCPU_REF_INIT_DEAD
, GFP_KERNEL
)) {
2716 return ERR_PTR(-ENOMEM
);
2719 spin_lock(&swap_lock
);
2720 for (type
= 0; type
< nr_swapfiles
; type
++) {
2721 if (!(swap_info
[type
]->flags
& SWP_USED
))
2724 if (type
>= MAX_SWAPFILES
) {
2725 spin_unlock(&swap_lock
);
2726 percpu_ref_exit(&p
->users
);
2728 return ERR_PTR(-EPERM
);
2730 if (type
>= nr_swapfiles
) {
2733 * Publish the swap_info_struct after initializing it.
2734 * Note that kvzalloc() above zeroes all its fields.
2736 smp_store_release(&swap_info
[type
], p
); /* rcu_assign_pointer() */
2740 p
= swap_info
[type
];
2742 * Do not memset this entry: a racing procfs swap_next()
2743 * would be relying on p->type to remain valid.
2746 p
->swap_extent_root
= RB_ROOT
;
2747 plist_node_init(&p
->list
, 0);
2749 plist_node_init(&p
->avail_lists
[i
], 0);
2750 p
->flags
= SWP_USED
;
2751 spin_unlock(&swap_lock
);
2753 percpu_ref_exit(&defer
->users
);
2756 spin_lock_init(&p
->lock
);
2757 spin_lock_init(&p
->cont_lock
);
2758 init_completion(&p
->comp
);
2763 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2767 if (S_ISBLK(inode
->i_mode
)) {
2768 p
->bdev_file
= bdev_file_open_by_dev(inode
->i_rdev
,
2769 BLK_OPEN_READ
| BLK_OPEN_WRITE
, p
, NULL
);
2770 if (IS_ERR(p
->bdev_file
)) {
2771 error
= PTR_ERR(p
->bdev_file
);
2772 p
->bdev_file
= NULL
;
2775 p
->bdev
= file_bdev(p
->bdev_file
);
2776 p
->old_block_size
= block_size(p
->bdev
);
2777 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2781 * Zoned block devices contain zones that have a sequential
2782 * write only restriction. Hence zoned block devices are not
2783 * suitable for swapping. Disallow them here.
2785 if (bdev_is_zoned(p
->bdev
))
2787 p
->flags
|= SWP_BLKDEV
;
2788 } else if (S_ISREG(inode
->i_mode
)) {
2789 p
->bdev
= inode
->i_sb
->s_bdev
;
2797 * Find out how many pages are allowed for a single swap device. There
2798 * are two limiting factors:
2799 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2800 * 2) the number of bits in the swap pte, as defined by the different
2803 * In order to find the largest possible bit mask, a swap entry with
2804 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2805 * decoded to a swp_entry_t again, and finally the swap offset is
2808 * This will mask all the bits from the initial ~0UL mask that can't
2809 * be encoded in either the swp_entry_t or the architecture definition
2812 unsigned long generic_max_swapfile_size(void)
2814 return swp_offset(pte_to_swp_entry(
2815 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2818 /* Can be overridden by an architecture for additional checks. */
2819 __weak
unsigned long arch_max_swapfile_size(void)
2821 return generic_max_swapfile_size();
2824 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2825 union swap_header
*swap_header
,
2826 struct inode
*inode
)
2829 unsigned long maxpages
;
2830 unsigned long swapfilepages
;
2831 unsigned long last_page
;
2833 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2834 pr_err("Unable to find swap-space signature\n");
2838 /* swap partition endianness hack... */
2839 if (swab32(swap_header
->info
.version
) == 1) {
2840 swab32s(&swap_header
->info
.version
);
2841 swab32s(&swap_header
->info
.last_page
);
2842 swab32s(&swap_header
->info
.nr_badpages
);
2843 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2845 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2846 swab32s(&swap_header
->info
.badpages
[i
]);
2848 /* Check the swap header's sub-version */
2849 if (swap_header
->info
.version
!= 1) {
2850 pr_warn("Unable to handle swap header version %d\n",
2851 swap_header
->info
.version
);
2856 p
->cluster_next
= 1;
2859 maxpages
= swapfile_maximum_size
;
2860 last_page
= swap_header
->info
.last_page
;
2862 pr_warn("Empty swap-file\n");
2865 if (last_page
> maxpages
) {
2866 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2867 K(maxpages
), K(last_page
));
2869 if (maxpages
> last_page
) {
2870 maxpages
= last_page
+ 1;
2871 /* p->max is an unsigned int: don't overflow it */
2872 if ((unsigned int)maxpages
== 0)
2873 maxpages
= UINT_MAX
;
2875 p
->highest_bit
= maxpages
- 1;
2879 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
2880 if (swapfilepages
&& maxpages
> swapfilepages
) {
2881 pr_warn("Swap area shorter than signature indicates\n");
2884 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
2886 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2892 #define SWAP_CLUSTER_INFO_COLS \
2893 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2894 #define SWAP_CLUSTER_SPACE_COLS \
2895 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2896 #define SWAP_CLUSTER_COLS \
2897 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2899 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
2900 union swap_header
*swap_header
,
2901 unsigned char *swap_map
,
2902 struct swap_cluster_info
*cluster_info
,
2903 unsigned long maxpages
,
2907 unsigned int nr_good_pages
;
2909 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2910 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
2911 unsigned long i
, idx
;
2913 nr_good_pages
= maxpages
- 1; /* omit header page */
2915 cluster_list_init(&p
->free_clusters
);
2916 cluster_list_init(&p
->discard_clusters
);
2918 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
2919 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
2920 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
2922 if (page_nr
< maxpages
) {
2923 swap_map
[page_nr
] = SWAP_MAP_BAD
;
2926 * Haven't marked the cluster free yet, no list
2927 * operation involved
2929 inc_cluster_info_page(p
, cluster_info
, page_nr
);
2933 /* Haven't marked the cluster free yet, no list operation involved */
2934 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
2935 inc_cluster_info_page(p
, cluster_info
, i
);
2937 if (nr_good_pages
) {
2938 swap_map
[0] = SWAP_MAP_BAD
;
2940 * Not mark the cluster free yet, no list
2941 * operation involved
2943 inc_cluster_info_page(p
, cluster_info
, 0);
2945 p
->pages
= nr_good_pages
;
2946 nr_extents
= setup_swap_extents(p
, span
);
2949 nr_good_pages
= p
->pages
;
2951 if (!nr_good_pages
) {
2952 pr_warn("Empty swap-file\n");
2961 * Reduce false cache line sharing between cluster_info and
2962 * sharing same address space.
2964 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
2965 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
2966 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
2967 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
2968 if (idx
>= nr_clusters
)
2970 if (cluster_count(&cluster_info
[idx
]))
2972 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
2973 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
2980 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
2982 struct swap_info_struct
*p
;
2983 struct filename
*name
;
2984 struct file
*swap_file
= NULL
;
2985 struct address_space
*mapping
;
2986 struct dentry
*dentry
;
2989 union swap_header
*swap_header
;
2992 unsigned long maxpages
;
2993 unsigned char *swap_map
= NULL
;
2994 struct swap_cluster_info
*cluster_info
= NULL
;
2995 struct page
*page
= NULL
;
2996 struct inode
*inode
= NULL
;
2997 bool inced_nr_rotate_swap
= false;
2999 if (swap_flags
& ~SWAP_FLAGS_VALID
)
3002 if (!capable(CAP_SYS_ADMIN
))
3005 if (!swap_avail_heads
)
3008 p
= alloc_swap_info();
3012 INIT_WORK(&p
->discard_work
, swap_discard_work
);
3014 name
= getname(specialfile
);
3016 error
= PTR_ERR(name
);
3020 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
3021 if (IS_ERR(swap_file
)) {
3022 error
= PTR_ERR(swap_file
);
3027 p
->swap_file
= swap_file
;
3028 mapping
= swap_file
->f_mapping
;
3029 dentry
= swap_file
->f_path
.dentry
;
3030 inode
= mapping
->host
;
3032 error
= claim_swapfile(p
, inode
);
3033 if (unlikely(error
))
3037 if (d_unlinked(dentry
) || cant_mount(dentry
)) {
3039 goto bad_swap_unlock_inode
;
3041 if (IS_SWAPFILE(inode
)) {
3043 goto bad_swap_unlock_inode
;
3047 * Read the swap header.
3049 if (!mapping
->a_ops
->read_folio
) {
3051 goto bad_swap_unlock_inode
;
3053 page
= read_mapping_page(mapping
, 0, swap_file
);
3055 error
= PTR_ERR(page
);
3056 goto bad_swap_unlock_inode
;
3058 swap_header
= kmap(page
);
3060 maxpages
= read_swap_header(p
, swap_header
, inode
);
3061 if (unlikely(!maxpages
)) {
3063 goto bad_swap_unlock_inode
;
3066 /* OK, set up the swap map and apply the bad block list */
3067 swap_map
= vzalloc(maxpages
);
3070 goto bad_swap_unlock_inode
;
3073 if (p
->bdev
&& bdev_stable_writes(p
->bdev
))
3074 p
->flags
|= SWP_STABLE_WRITES
;
3076 if (p
->bdev
&& bdev_synchronous(p
->bdev
))
3077 p
->flags
|= SWP_SYNCHRONOUS_IO
;
3079 if (p
->bdev
&& bdev_nonrot(p
->bdev
)) {
3081 unsigned long ci
, nr_cluster
;
3083 p
->flags
|= SWP_SOLIDSTATE
;
3084 p
->cluster_next_cpu
= alloc_percpu(unsigned int);
3085 if (!p
->cluster_next_cpu
) {
3087 goto bad_swap_unlock_inode
;
3090 * select a random position to start with to help wear leveling
3093 for_each_possible_cpu(cpu
) {
3094 per_cpu(*p
->cluster_next_cpu
, cpu
) =
3095 get_random_u32_inclusive(1, p
->highest_bit
);
3097 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3099 cluster_info
= kvcalloc(nr_cluster
, sizeof(*cluster_info
),
3101 if (!cluster_info
) {
3103 goto bad_swap_unlock_inode
;
3106 for (ci
= 0; ci
< nr_cluster
; ci
++)
3107 spin_lock_init(&((cluster_info
+ ci
)->lock
));
3109 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
3110 if (!p
->percpu_cluster
) {
3112 goto bad_swap_unlock_inode
;
3114 for_each_possible_cpu(cpu
) {
3115 struct percpu_cluster
*cluster
;
3116 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
3117 cluster_set_null(&cluster
->index
);
3120 atomic_inc(&nr_rotate_swap
);
3121 inced_nr_rotate_swap
= true;
3124 error
= swap_cgroup_swapon(p
->type
, maxpages
);
3126 goto bad_swap_unlock_inode
;
3128 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
3129 cluster_info
, maxpages
, &span
);
3130 if (unlikely(nr_extents
< 0)) {
3132 goto bad_swap_unlock_inode
;
3135 if ((swap_flags
& SWAP_FLAG_DISCARD
) &&
3136 p
->bdev
&& bdev_max_discard_sectors(p
->bdev
)) {
3138 * When discard is enabled for swap with no particular
3139 * policy flagged, we set all swap discard flags here in
3140 * order to sustain backward compatibility with older
3141 * swapon(8) releases.
3143 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
3147 * By flagging sys_swapon, a sysadmin can tell us to
3148 * either do single-time area discards only, or to just
3149 * perform discards for released swap page-clusters.
3150 * Now it's time to adjust the p->flags accordingly.
3152 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
3153 p
->flags
&= ~SWP_PAGE_DISCARD
;
3154 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
3155 p
->flags
&= ~SWP_AREA_DISCARD
;
3157 /* issue a swapon-time discard if it's still required */
3158 if (p
->flags
& SWP_AREA_DISCARD
) {
3159 int err
= discard_swap(p
);
3161 pr_err("swapon: discard_swap(%p): %d\n",
3166 error
= init_swap_address_space(p
->type
, maxpages
);
3168 goto bad_swap_unlock_inode
;
3171 * Flush any pending IO and dirty mappings before we start using this
3174 inode
->i_flags
|= S_SWAPFILE
;
3175 error
= inode_drain_writes(inode
);
3177 inode
->i_flags
&= ~S_SWAPFILE
;
3178 goto free_swap_address_space
;
3181 mutex_lock(&swapon_mutex
);
3183 if (swap_flags
& SWAP_FLAG_PREFER
)
3185 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
3186 enable_swap_info(p
, prio
, swap_map
, cluster_info
);
3188 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s\n",
3189 K(p
->pages
), name
->name
, p
->prio
, nr_extents
,
3190 K((unsigned long long)span
),
3191 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
3192 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
3193 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
3194 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "");
3196 mutex_unlock(&swapon_mutex
);
3197 atomic_inc(&proc_poll_event
);
3198 wake_up_interruptible(&proc_poll_wait
);
3202 free_swap_address_space
:
3203 exit_swap_address_space(p
->type
);
3204 bad_swap_unlock_inode
:
3205 inode_unlock(inode
);
3207 free_percpu(p
->percpu_cluster
);
3208 p
->percpu_cluster
= NULL
;
3209 free_percpu(p
->cluster_next_cpu
);
3210 p
->cluster_next_cpu
= NULL
;
3212 set_blocksize(p
->bdev
, p
->old_block_size
);
3214 p
->bdev_file
= NULL
;
3217 destroy_swap_extents(p
);
3218 swap_cgroup_swapoff(p
->type
);
3219 spin_lock(&swap_lock
);
3220 p
->swap_file
= NULL
;
3222 spin_unlock(&swap_lock
);
3224 kvfree(cluster_info
);
3225 if (inced_nr_rotate_swap
)
3226 atomic_dec(&nr_rotate_swap
);
3228 filp_close(swap_file
, NULL
);
3230 if (page
&& !IS_ERR(page
)) {
3237 inode_unlock(inode
);
3239 enable_swap_slots_cache();
3243 void si_swapinfo(struct sysinfo
*val
)
3246 unsigned long nr_to_be_unused
= 0;
3248 spin_lock(&swap_lock
);
3249 for (type
= 0; type
< nr_swapfiles
; type
++) {
3250 struct swap_info_struct
*si
= swap_info
[type
];
3252 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
3253 nr_to_be_unused
+= READ_ONCE(si
->inuse_pages
);
3255 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
3256 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
3257 spin_unlock(&swap_lock
);
3261 * Verify that a swap entry is valid and increment its swap map count.
3263 * Returns error code in following case.
3265 * - swp_entry is invalid -> EINVAL
3266 * - swp_entry is migration entry -> EINVAL
3267 * - swap-cache reference is requested but there is already one. -> EEXIST
3268 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3269 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3271 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
3273 struct swap_info_struct
*p
;
3274 struct swap_cluster_info
*ci
;
3275 unsigned long offset
;
3276 unsigned char count
;
3277 unsigned char has_cache
;
3280 p
= swp_swap_info(entry
);
3282 offset
= swp_offset(entry
);
3283 ci
= lock_cluster_or_swap_info(p
, offset
);
3285 count
= p
->swap_map
[offset
];
3288 * swapin_readahead() doesn't check if a swap entry is valid, so the
3289 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3291 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
3296 has_cache
= count
& SWAP_HAS_CACHE
;
3297 count
&= ~SWAP_HAS_CACHE
;
3300 if (usage
== SWAP_HAS_CACHE
) {
3302 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3303 if (!has_cache
&& count
)
3304 has_cache
= SWAP_HAS_CACHE
;
3305 else if (has_cache
) /* someone else added cache */
3307 else /* no users remaining */
3310 } else if (count
|| has_cache
) {
3312 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
3314 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
3316 else if (swap_count_continued(p
, offset
, count
))
3317 count
= COUNT_CONTINUED
;
3321 err
= -ENOENT
; /* unused swap entry */
3323 WRITE_ONCE(p
->swap_map
[offset
], count
| has_cache
);
3326 unlock_cluster_or_swap_info(p
, ci
);
3331 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3332 * (in which case its reference count is never incremented).
3334 void swap_shmem_alloc(swp_entry_t entry
)
3336 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
3340 * Increase reference count of swap entry by 1.
3341 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3342 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3343 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3344 * might occur if a page table entry has got corrupted.
3346 int swap_duplicate(swp_entry_t entry
)
3350 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
3351 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
3356 * @entry: swap entry for which we allocate swap cache.
3358 * Called when allocating swap cache for existing swap entry,
3359 * This can return error codes. Returns 0 at success.
3360 * -EEXIST means there is a swap cache.
3361 * Note: return code is different from swap_duplicate().
3363 int swapcache_prepare(swp_entry_t entry
)
3365 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
3368 struct swap_info_struct
*swp_swap_info(swp_entry_t entry
)
3370 return swap_type_to_swap_info(swp_type(entry
));
3374 * out-of-line methods to avoid include hell.
3376 struct address_space
*swapcache_mapping(struct folio
*folio
)
3378 return swp_swap_info(folio
->swap
)->swap_file
->f_mapping
;
3380 EXPORT_SYMBOL_GPL(swapcache_mapping
);
3382 pgoff_t
__page_file_index(struct page
*page
)
3384 swp_entry_t swap
= page_swap_entry(page
);
3385 return swp_offset(swap
);
3387 EXPORT_SYMBOL_GPL(__page_file_index
);
3390 * add_swap_count_continuation - called when a swap count is duplicated
3391 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3392 * page of the original vmalloc'ed swap_map, to hold the continuation count
3393 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3394 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3396 * These continuation pages are seldom referenced: the common paths all work
3397 * on the original swap_map, only referring to a continuation page when the
3398 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3400 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3401 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3402 * can be called after dropping locks.
3404 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
3406 struct swap_info_struct
*si
;
3407 struct swap_cluster_info
*ci
;
3410 struct page
*list_page
;
3412 unsigned char count
;
3416 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3417 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3419 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
3421 si
= get_swap_device(entry
);
3424 * An acceptable race has occurred since the failing
3425 * __swap_duplicate(): the swap device may be swapoff
3429 spin_lock(&si
->lock
);
3431 offset
= swp_offset(entry
);
3433 ci
= lock_cluster(si
, offset
);
3435 count
= swap_count(si
->swap_map
[offset
]);
3437 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
3439 * The higher the swap count, the more likely it is that tasks
3440 * will race to add swap count continuation: we need to avoid
3441 * over-provisioning.
3451 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3452 offset
&= ~PAGE_MASK
;
3454 spin_lock(&si
->cont_lock
);
3456 * Page allocation does not initialize the page's lru field,
3457 * but it does always reset its private field.
3459 if (!page_private(head
)) {
3460 BUG_ON(count
& COUNT_CONTINUED
);
3461 INIT_LIST_HEAD(&head
->lru
);
3462 set_page_private(head
, SWP_CONTINUED
);
3463 si
->flags
|= SWP_CONTINUED
;
3466 list_for_each_entry(list_page
, &head
->lru
, lru
) {
3470 * If the previous map said no continuation, but we've found
3471 * a continuation page, free our allocation and use this one.
3473 if (!(count
& COUNT_CONTINUED
))
3474 goto out_unlock_cont
;
3476 map
= kmap_local_page(list_page
) + offset
;
3481 * If this continuation count now has some space in it,
3482 * free our allocation and use this one.
3484 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
3485 goto out_unlock_cont
;
3488 list_add_tail(&page
->lru
, &head
->lru
);
3489 page
= NULL
; /* now it's attached, don't free it */
3491 spin_unlock(&si
->cont_lock
);
3494 spin_unlock(&si
->lock
);
3495 put_swap_device(si
);
3503 * swap_count_continued - when the original swap_map count is incremented
3504 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3505 * into, carry if so, or else fail until a new continuation page is allocated;
3506 * when the original swap_map count is decremented from 0 with continuation,
3507 * borrow from the continuation and report whether it still holds more.
3508 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3511 static bool swap_count_continued(struct swap_info_struct
*si
,
3512 pgoff_t offset
, unsigned char count
)
3519 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3520 if (page_private(head
) != SWP_CONTINUED
) {
3521 BUG_ON(count
& COUNT_CONTINUED
);
3522 return false; /* need to add count continuation */
3525 spin_lock(&si
->cont_lock
);
3526 offset
&= ~PAGE_MASK
;
3527 page
= list_next_entry(head
, lru
);
3528 map
= kmap_local_page(page
) + offset
;
3530 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3531 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3533 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3535 * Think of how you add 1 to 999
3537 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3539 page
= list_next_entry(page
, lru
);
3540 BUG_ON(page
== head
);
3541 map
= kmap_local_page(page
) + offset
;
3543 if (*map
== SWAP_CONT_MAX
) {
3545 page
= list_next_entry(page
, lru
);
3547 ret
= false; /* add count continuation */
3550 map
= kmap_local_page(page
) + offset
;
3551 init_map
: *map
= 0; /* we didn't zero the page */
3555 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3556 map
= kmap_local_page(page
) + offset
;
3557 *map
= COUNT_CONTINUED
;
3560 ret
= true; /* incremented */
3562 } else { /* decrementing */
3564 * Think of how you subtract 1 from 1000
3566 BUG_ON(count
!= COUNT_CONTINUED
);
3567 while (*map
== COUNT_CONTINUED
) {
3569 page
= list_next_entry(page
, lru
);
3570 BUG_ON(page
== head
);
3571 map
= kmap_local_page(page
) + offset
;
3578 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3579 map
= kmap_local_page(page
) + offset
;
3580 *map
= SWAP_CONT_MAX
| count
;
3581 count
= COUNT_CONTINUED
;
3584 ret
= count
== COUNT_CONTINUED
;
3587 spin_unlock(&si
->cont_lock
);
3592 * free_swap_count_continuations - swapoff free all the continuation pages
3593 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3595 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3599 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3601 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3602 if (page_private(head
)) {
3603 struct page
*page
, *next
;
3605 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3606 list_del(&page
->lru
);
3613 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3614 void __folio_throttle_swaprate(struct folio
*folio
, gfp_t gfp
)
3616 struct swap_info_struct
*si
, *next
;
3617 int nid
= folio_nid(folio
);
3619 if (!(gfp
& __GFP_IO
))
3622 if (!blk_cgroup_congested())
3626 * We've already scheduled a throttle, avoid taking the global swap
3629 if (current
->throttle_disk
)
3632 spin_lock(&swap_avail_lock
);
3633 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[nid
],
3636 blkcg_schedule_throttle(si
->bdev
->bd_disk
, true);
3640 spin_unlock(&swap_avail_lock
);
3644 static int __init
swapfile_init(void)
3648 swap_avail_heads
= kmalloc_array(nr_node_ids
, sizeof(struct plist_head
),
3650 if (!swap_avail_heads
) {
3651 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3656 plist_head_init(&swap_avail_heads
[nid
]);
3658 swapfile_maximum_size
= arch_max_swapfile_size();
3660 #ifdef CONFIG_MIGRATION
3661 if (swapfile_maximum_size
>= (1UL << SWP_MIG_TOTAL_BITS
))
3662 swap_migration_ad_supported
= true;
3663 #endif /* CONFIG_MIGRATION */
3667 subsys_initcall(swapfile_init
);