1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie
10 #include <linux/sched/mm.h>
11 #include <linux/sched/task.h>
12 #include <linux/hugetlb.h>
13 #include <linux/mman.h>
14 #include <linux/slab.h>
15 #include <linux/kernel_stat.h>
16 #include <linux/swap.h>
17 #include <linux/vmalloc.h>
18 #include <linux/pagemap.h>
19 #include <linux/namei.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/writeback.h>
24 #include <linux/proc_fs.h>
25 #include <linux/seq_file.h>
26 #include <linux/init.h>
27 #include <linux/ksm.h>
28 #include <linux/rmap.h>
29 #include <linux/security.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mutex.h>
32 #include <linux/capability.h>
33 #include <linux/syscalls.h>
34 #include <linux/memcontrol.h>
35 #include <linux/poll.h>
36 #include <linux/oom.h>
37 #include <linux/frontswap.h>
38 #include <linux/swapfile.h>
39 #include <linux/export.h>
40 #include <linux/swap_slots.h>
41 #include <linux/sort.h>
43 #include <asm/tlbflush.h>
44 #include <linux/swapops.h>
45 #include <linux/swap_cgroup.h>
47 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
49 static void free_swap_count_continuations(struct swap_info_struct
*);
51 DEFINE_SPINLOCK(swap_lock
);
52 static unsigned int nr_swapfiles
;
53 atomic_long_t nr_swap_pages
;
55 * Some modules use swappable objects and may try to swap them out under
56 * memory pressure (via the shrinker). Before doing so, they may wish to
57 * check to see if any swap space is available.
59 EXPORT_SYMBOL_GPL(nr_swap_pages
);
60 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
61 long total_swap_pages
;
62 static int least_priority
= -1;
64 static const char Bad_file
[] = "Bad swap file entry ";
65 static const char Unused_file
[] = "Unused swap file entry ";
66 static const char Bad_offset
[] = "Bad swap offset entry ";
67 static const char Unused_offset
[] = "Unused swap offset entry ";
70 * all active swap_info_structs
71 * protected with swap_lock, and ordered by priority.
73 PLIST_HEAD(swap_active_head
);
76 * all available (active, not full) swap_info_structs
77 * protected with swap_avail_lock, ordered by priority.
78 * This is used by get_swap_page() instead of swap_active_head
79 * because swap_active_head includes all swap_info_structs,
80 * but get_swap_page() doesn't need to look at full ones.
81 * This uses its own lock instead of swap_lock because when a
82 * swap_info_struct changes between not-full/full, it needs to
83 * add/remove itself to/from this list, but the swap_info_struct->lock
84 * is held and the locking order requires swap_lock to be taken
85 * before any swap_info_struct->lock.
87 static struct plist_head
*swap_avail_heads
;
88 static DEFINE_SPINLOCK(swap_avail_lock
);
90 struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
92 static DEFINE_MUTEX(swapon_mutex
);
94 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
95 /* Activity counter to indicate that a swapon or swapoff has occurred */
96 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
98 atomic_t nr_rotate_swap
= ATOMIC_INIT(0);
100 static struct swap_info_struct
*swap_type_to_swap_info(int type
)
102 if (type
>= READ_ONCE(nr_swapfiles
))
105 smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */
106 return READ_ONCE(swap_info
[type
]);
109 static inline unsigned char swap_count(unsigned char ent
)
111 return ent
& ~SWAP_HAS_CACHE
; /* may include COUNT_CONTINUED flag */
114 /* Reclaim the swap entry anyway if possible */
115 #define TTRS_ANYWAY 0x1
117 * Reclaim the swap entry if there are no more mappings of the
120 #define TTRS_UNMAPPED 0x2
121 /* Reclaim the swap entry if swap is getting full*/
122 #define TTRS_FULL 0x4
124 /* returns 1 if swap entry is freed */
125 static int __try_to_reclaim_swap(struct swap_info_struct
*si
,
126 unsigned long offset
, unsigned long flags
)
128 swp_entry_t entry
= swp_entry(si
->type
, offset
);
132 page
= find_get_page(swap_address_space(entry
), offset
);
136 * When this function is called from scan_swap_map_slots() and it's
137 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
138 * here. We have to use trylock for avoiding deadlock. This is a special
139 * case and you should use try_to_free_swap() with explicit lock_page()
140 * in usual operations.
142 if (trylock_page(page
)) {
143 if ((flags
& TTRS_ANYWAY
) ||
144 ((flags
& TTRS_UNMAPPED
) && !page_mapped(page
)) ||
145 ((flags
& TTRS_FULL
) && mem_cgroup_swap_full(page
)))
146 ret
= try_to_free_swap(page
);
153 static inline struct swap_extent
*first_se(struct swap_info_struct
*sis
)
155 struct rb_node
*rb
= rb_first(&sis
->swap_extent_root
);
156 return rb_entry(rb
, struct swap_extent
, rb_node
);
159 static inline struct swap_extent
*next_se(struct swap_extent
*se
)
161 struct rb_node
*rb
= rb_next(&se
->rb_node
);
162 return rb
? rb_entry(rb
, struct swap_extent
, rb_node
) : NULL
;
166 * swapon tell device that all the old swap contents can be discarded,
167 * to allow the swap device to optimize its wear-levelling.
169 static int discard_swap(struct swap_info_struct
*si
)
171 struct swap_extent
*se
;
172 sector_t start_block
;
176 /* Do not discard the swap header page! */
178 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
179 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
181 err
= blkdev_issue_discard(si
->bdev
, start_block
,
182 nr_blocks
, GFP_KERNEL
, 0);
188 for (se
= next_se(se
); se
; se
= next_se(se
)) {
189 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
190 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
192 err
= blkdev_issue_discard(si
->bdev
, start_block
,
193 nr_blocks
, GFP_KERNEL
, 0);
199 return err
; /* That will often be -EOPNOTSUPP */
202 static struct swap_extent
*
203 offset_to_swap_extent(struct swap_info_struct
*sis
, unsigned long offset
)
205 struct swap_extent
*se
;
208 rb
= sis
->swap_extent_root
.rb_node
;
210 se
= rb_entry(rb
, struct swap_extent
, rb_node
);
211 if (offset
< se
->start_page
)
213 else if (offset
>= se
->start_page
+ se
->nr_pages
)
218 /* It *must* be present */
222 sector_t
swap_page_sector(struct page
*page
)
224 struct swap_info_struct
*sis
= page_swap_info(page
);
225 struct swap_extent
*se
;
229 offset
= __page_file_index(page
);
230 se
= offset_to_swap_extent(sis
, offset
);
231 sector
= se
->start_block
+ (offset
- se
->start_page
);
232 return sector
<< (PAGE_SHIFT
- 9);
236 * swap allocation tell device that a cluster of swap can now be discarded,
237 * to allow the swap device to optimize its wear-levelling.
239 static void discard_swap_cluster(struct swap_info_struct
*si
,
240 pgoff_t start_page
, pgoff_t nr_pages
)
242 struct swap_extent
*se
= offset_to_swap_extent(si
, start_page
);
245 pgoff_t offset
= start_page
- se
->start_page
;
246 sector_t start_block
= se
->start_block
+ offset
;
247 sector_t nr_blocks
= se
->nr_pages
- offset
;
249 if (nr_blocks
> nr_pages
)
250 nr_blocks
= nr_pages
;
251 start_page
+= nr_blocks
;
252 nr_pages
-= nr_blocks
;
254 start_block
<<= PAGE_SHIFT
- 9;
255 nr_blocks
<<= PAGE_SHIFT
- 9;
256 if (blkdev_issue_discard(si
->bdev
, start_block
,
257 nr_blocks
, GFP_NOIO
, 0))
264 #ifdef CONFIG_THP_SWAP
265 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
267 #define swap_entry_size(size) (size)
269 #define SWAPFILE_CLUSTER 256
272 * Define swap_entry_size() as constant to let compiler to optimize
273 * out some code if !CONFIG_THP_SWAP
275 #define swap_entry_size(size) 1
277 #define LATENCY_LIMIT 256
279 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
285 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
290 static inline void cluster_set_count(struct swap_cluster_info
*info
,
296 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
297 unsigned int c
, unsigned int f
)
303 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
308 static inline void cluster_set_next(struct swap_cluster_info
*info
,
314 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
315 unsigned int n
, unsigned int f
)
321 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
323 return info
->flags
& CLUSTER_FLAG_FREE
;
326 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
328 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
331 static inline void cluster_set_null(struct swap_cluster_info
*info
)
333 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
337 static inline bool cluster_is_huge(struct swap_cluster_info
*info
)
339 if (IS_ENABLED(CONFIG_THP_SWAP
))
340 return info
->flags
& CLUSTER_FLAG_HUGE
;
344 static inline void cluster_clear_huge(struct swap_cluster_info
*info
)
346 info
->flags
&= ~CLUSTER_FLAG_HUGE
;
349 static inline struct swap_cluster_info
*lock_cluster(struct swap_info_struct
*si
,
350 unsigned long offset
)
352 struct swap_cluster_info
*ci
;
354 ci
= si
->cluster_info
;
356 ci
+= offset
/ SWAPFILE_CLUSTER
;
357 spin_lock(&ci
->lock
);
362 static inline void unlock_cluster(struct swap_cluster_info
*ci
)
365 spin_unlock(&ci
->lock
);
369 * Determine the locking method in use for this device. Return
370 * swap_cluster_info if SSD-style cluster-based locking is in place.
372 static inline struct swap_cluster_info
*lock_cluster_or_swap_info(
373 struct swap_info_struct
*si
, unsigned long offset
)
375 struct swap_cluster_info
*ci
;
377 /* Try to use fine-grained SSD-style locking if available: */
378 ci
= lock_cluster(si
, offset
);
379 /* Otherwise, fall back to traditional, coarse locking: */
381 spin_lock(&si
->lock
);
386 static inline void unlock_cluster_or_swap_info(struct swap_info_struct
*si
,
387 struct swap_cluster_info
*ci
)
392 spin_unlock(&si
->lock
);
395 static inline bool cluster_list_empty(struct swap_cluster_list
*list
)
397 return cluster_is_null(&list
->head
);
400 static inline unsigned int cluster_list_first(struct swap_cluster_list
*list
)
402 return cluster_next(&list
->head
);
405 static void cluster_list_init(struct swap_cluster_list
*list
)
407 cluster_set_null(&list
->head
);
408 cluster_set_null(&list
->tail
);
411 static void cluster_list_add_tail(struct swap_cluster_list
*list
,
412 struct swap_cluster_info
*ci
,
415 if (cluster_list_empty(list
)) {
416 cluster_set_next_flag(&list
->head
, idx
, 0);
417 cluster_set_next_flag(&list
->tail
, idx
, 0);
419 struct swap_cluster_info
*ci_tail
;
420 unsigned int tail
= cluster_next(&list
->tail
);
423 * Nested cluster lock, but both cluster locks are
424 * only acquired when we held swap_info_struct->lock
427 spin_lock_nested(&ci_tail
->lock
, SINGLE_DEPTH_NESTING
);
428 cluster_set_next(ci_tail
, idx
);
429 spin_unlock(&ci_tail
->lock
);
430 cluster_set_next_flag(&list
->tail
, idx
, 0);
434 static unsigned int cluster_list_del_first(struct swap_cluster_list
*list
,
435 struct swap_cluster_info
*ci
)
439 idx
= cluster_next(&list
->head
);
440 if (cluster_next(&list
->tail
) == idx
) {
441 cluster_set_null(&list
->head
);
442 cluster_set_null(&list
->tail
);
444 cluster_set_next_flag(&list
->head
,
445 cluster_next(&ci
[idx
]), 0);
450 /* Add a cluster to discard list and schedule it to do discard */
451 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
455 * If scan_swap_map() can't find a free cluster, it will check
456 * si->swap_map directly. To make sure the discarding cluster isn't
457 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
458 * will be cleared after discard
460 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
461 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
463 cluster_list_add_tail(&si
->discard_clusters
, si
->cluster_info
, idx
);
465 schedule_work(&si
->discard_work
);
468 static void __free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
470 struct swap_cluster_info
*ci
= si
->cluster_info
;
472 cluster_set_flag(ci
+ idx
, CLUSTER_FLAG_FREE
);
473 cluster_list_add_tail(&si
->free_clusters
, ci
, idx
);
477 * Doing discard actually. After a cluster discard is finished, the cluster
478 * will be added to free cluster list. caller should hold si->lock.
480 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
482 struct swap_cluster_info
*info
, *ci
;
485 info
= si
->cluster_info
;
487 while (!cluster_list_empty(&si
->discard_clusters
)) {
488 idx
= cluster_list_del_first(&si
->discard_clusters
, info
);
489 spin_unlock(&si
->lock
);
491 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
494 spin_lock(&si
->lock
);
495 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
496 __free_cluster(si
, idx
);
497 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
498 0, SWAPFILE_CLUSTER
);
503 static void swap_discard_work(struct work_struct
*work
)
505 struct swap_info_struct
*si
;
507 si
= container_of(work
, struct swap_info_struct
, discard_work
);
509 spin_lock(&si
->lock
);
510 swap_do_scheduled_discard(si
);
511 spin_unlock(&si
->lock
);
514 static void alloc_cluster(struct swap_info_struct
*si
, unsigned long idx
)
516 struct swap_cluster_info
*ci
= si
->cluster_info
;
518 VM_BUG_ON(cluster_list_first(&si
->free_clusters
) != idx
);
519 cluster_list_del_first(&si
->free_clusters
, ci
);
520 cluster_set_count_flag(ci
+ idx
, 0, 0);
523 static void free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
525 struct swap_cluster_info
*ci
= si
->cluster_info
+ idx
;
527 VM_BUG_ON(cluster_count(ci
) != 0);
529 * If the swap is discardable, prepare discard the cluster
530 * instead of free it immediately. The cluster will be freed
533 if ((si
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
534 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
535 swap_cluster_schedule_discard(si
, idx
);
539 __free_cluster(si
, idx
);
543 * The cluster corresponding to page_nr will be used. The cluster will be
544 * removed from free cluster list and its usage counter will be increased.
546 static void inc_cluster_info_page(struct swap_info_struct
*p
,
547 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
549 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
553 if (cluster_is_free(&cluster_info
[idx
]))
554 alloc_cluster(p
, idx
);
556 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
557 cluster_set_count(&cluster_info
[idx
],
558 cluster_count(&cluster_info
[idx
]) + 1);
562 * The cluster corresponding to page_nr decreases one usage. If the usage
563 * counter becomes 0, which means no page in the cluster is in using, we can
564 * optionally discard the cluster and add it to free cluster list.
566 static void dec_cluster_info_page(struct swap_info_struct
*p
,
567 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
569 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
574 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
575 cluster_set_count(&cluster_info
[idx
],
576 cluster_count(&cluster_info
[idx
]) - 1);
578 if (cluster_count(&cluster_info
[idx
]) == 0)
579 free_cluster(p
, idx
);
583 * It's possible scan_swap_map() uses a free cluster in the middle of free
584 * cluster list. Avoiding such abuse to avoid list corruption.
587 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
588 unsigned long offset
)
590 struct percpu_cluster
*percpu_cluster
;
593 offset
/= SWAPFILE_CLUSTER
;
594 conflict
= !cluster_list_empty(&si
->free_clusters
) &&
595 offset
!= cluster_list_first(&si
->free_clusters
) &&
596 cluster_is_free(&si
->cluster_info
[offset
]);
601 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
602 cluster_set_null(&percpu_cluster
->index
);
607 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
608 * might involve allocating a new cluster for current CPU too.
610 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
611 unsigned long *offset
, unsigned long *scan_base
)
613 struct percpu_cluster
*cluster
;
614 struct swap_cluster_info
*ci
;
615 unsigned long tmp
, max
;
618 cluster
= this_cpu_ptr(si
->percpu_cluster
);
619 if (cluster_is_null(&cluster
->index
)) {
620 if (!cluster_list_empty(&si
->free_clusters
)) {
621 cluster
->index
= si
->free_clusters
.head
;
622 cluster
->next
= cluster_next(&cluster
->index
) *
624 } else if (!cluster_list_empty(&si
->discard_clusters
)) {
626 * we don't have free cluster but have some clusters in
627 * discarding, do discard now and reclaim them, then
628 * reread cluster_next_cpu since we dropped si->lock
630 swap_do_scheduled_discard(si
);
631 *scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
632 *offset
= *scan_base
;
639 * Other CPUs can use our cluster if they can't find a free cluster,
640 * check if there is still free entry in the cluster
643 max
= min_t(unsigned long, si
->max
,
644 (cluster_next(&cluster
->index
) + 1) * SWAPFILE_CLUSTER
);
646 ci
= lock_cluster(si
, tmp
);
648 if (!si
->swap_map
[tmp
])
655 cluster_set_null(&cluster
->index
);
658 cluster
->next
= tmp
+ 1;
664 static void __del_from_avail_list(struct swap_info_struct
*p
)
669 plist_del(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
672 static void del_from_avail_list(struct swap_info_struct
*p
)
674 spin_lock(&swap_avail_lock
);
675 __del_from_avail_list(p
);
676 spin_unlock(&swap_avail_lock
);
679 static void swap_range_alloc(struct swap_info_struct
*si
, unsigned long offset
,
680 unsigned int nr_entries
)
682 unsigned int end
= offset
+ nr_entries
- 1;
684 if (offset
== si
->lowest_bit
)
685 si
->lowest_bit
+= nr_entries
;
686 if (end
== si
->highest_bit
)
687 WRITE_ONCE(si
->highest_bit
, si
->highest_bit
- nr_entries
);
688 si
->inuse_pages
+= nr_entries
;
689 if (si
->inuse_pages
== si
->pages
) {
690 si
->lowest_bit
= si
->max
;
692 del_from_avail_list(si
);
696 static void add_to_avail_list(struct swap_info_struct
*p
)
700 spin_lock(&swap_avail_lock
);
702 WARN_ON(!plist_node_empty(&p
->avail_lists
[nid
]));
703 plist_add(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
705 spin_unlock(&swap_avail_lock
);
708 static void swap_range_free(struct swap_info_struct
*si
, unsigned long offset
,
709 unsigned int nr_entries
)
711 unsigned long begin
= offset
;
712 unsigned long end
= offset
+ nr_entries
- 1;
713 void (*swap_slot_free_notify
)(struct block_device
*, unsigned long);
715 if (offset
< si
->lowest_bit
)
716 si
->lowest_bit
= offset
;
717 if (end
> si
->highest_bit
) {
718 bool was_full
= !si
->highest_bit
;
720 WRITE_ONCE(si
->highest_bit
, end
);
721 if (was_full
&& (si
->flags
& SWP_WRITEOK
))
722 add_to_avail_list(si
);
724 atomic_long_add(nr_entries
, &nr_swap_pages
);
725 si
->inuse_pages
-= nr_entries
;
726 if (si
->flags
& SWP_BLKDEV
)
727 swap_slot_free_notify
=
728 si
->bdev
->bd_disk
->fops
->swap_slot_free_notify
;
730 swap_slot_free_notify
= NULL
;
731 while (offset
<= end
) {
732 arch_swap_invalidate_page(si
->type
, offset
);
733 frontswap_invalidate_page(si
->type
, offset
);
734 if (swap_slot_free_notify
)
735 swap_slot_free_notify(si
->bdev
, offset
);
738 clear_shadow_from_swap_cache(si
->type
, begin
, end
);
741 static void set_cluster_next(struct swap_info_struct
*si
, unsigned long next
)
745 if (!(si
->flags
& SWP_SOLIDSTATE
)) {
746 si
->cluster_next
= next
;
750 prev
= this_cpu_read(*si
->cluster_next_cpu
);
752 * Cross the swap address space size aligned trunk, choose
753 * another trunk randomly to avoid lock contention on swap
754 * address space if possible.
756 if ((prev
>> SWAP_ADDRESS_SPACE_SHIFT
) !=
757 (next
>> SWAP_ADDRESS_SPACE_SHIFT
)) {
758 /* No free swap slots available */
759 if (si
->highest_bit
<= si
->lowest_bit
)
761 next
= si
->lowest_bit
+
762 prandom_u32_max(si
->highest_bit
- si
->lowest_bit
+ 1);
763 next
= ALIGN_DOWN(next
, SWAP_ADDRESS_SPACE_PAGES
);
764 next
= max_t(unsigned int, next
, si
->lowest_bit
);
766 this_cpu_write(*si
->cluster_next_cpu
, next
);
769 static int scan_swap_map_slots(struct swap_info_struct
*si
,
770 unsigned char usage
, int nr
,
773 struct swap_cluster_info
*ci
;
774 unsigned long offset
;
775 unsigned long scan_base
;
776 unsigned long last_in_cluster
= 0;
777 int latency_ration
= LATENCY_LIMIT
;
779 bool scanned_many
= false;
782 * We try to cluster swap pages by allocating them sequentially
783 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
784 * way, however, we resort to first-free allocation, starting
785 * a new cluster. This prevents us from scattering swap pages
786 * all over the entire swap partition, so that we reduce
787 * overall disk seek times between swap pages. -- sct
788 * But we do now try to find an empty cluster. -Andrea
789 * And we let swap pages go all over an SSD partition. Hugh
792 si
->flags
+= SWP_SCANNING
;
794 * Use percpu scan base for SSD to reduce lock contention on
795 * cluster and swap cache. For HDD, sequential access is more
798 if (si
->flags
& SWP_SOLIDSTATE
)
799 scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
801 scan_base
= si
->cluster_next
;
805 if (si
->cluster_info
) {
806 if (!scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
808 } else if (unlikely(!si
->cluster_nr
--)) {
809 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
810 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
814 spin_unlock(&si
->lock
);
817 * If seek is expensive, start searching for new cluster from
818 * start of partition, to minimize the span of allocated swap.
819 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
820 * case, just handled by scan_swap_map_try_ssd_cluster() above.
822 scan_base
= offset
= si
->lowest_bit
;
823 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
825 /* Locate the first empty (unaligned) cluster */
826 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
827 if (si
->swap_map
[offset
])
828 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
829 else if (offset
== last_in_cluster
) {
830 spin_lock(&si
->lock
);
831 offset
-= SWAPFILE_CLUSTER
- 1;
832 si
->cluster_next
= offset
;
833 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
836 if (unlikely(--latency_ration
< 0)) {
838 latency_ration
= LATENCY_LIMIT
;
843 spin_lock(&si
->lock
);
844 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
848 if (si
->cluster_info
) {
849 while (scan_swap_map_ssd_cluster_conflict(si
, offset
)) {
850 /* take a break if we already got some slots */
853 if (!scan_swap_map_try_ssd_cluster(si
, &offset
,
858 if (!(si
->flags
& SWP_WRITEOK
))
860 if (!si
->highest_bit
)
862 if (offset
> si
->highest_bit
)
863 scan_base
= offset
= si
->lowest_bit
;
865 ci
= lock_cluster(si
, offset
);
866 /* reuse swap entry of cache-only swap if not busy. */
867 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
870 spin_unlock(&si
->lock
);
871 swap_was_freed
= __try_to_reclaim_swap(si
, offset
, TTRS_ANYWAY
);
872 spin_lock(&si
->lock
);
873 /* entry was freed successfully, try to use this again */
876 goto scan
; /* check next one */
879 if (si
->swap_map
[offset
]) {
886 WRITE_ONCE(si
->swap_map
[offset
], usage
);
887 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
890 swap_range_alloc(si
, offset
, 1);
891 slots
[n_ret
++] = swp_entry(si
->type
, offset
);
893 /* got enough slots or reach max slots? */
894 if ((n_ret
== nr
) || (offset
>= si
->highest_bit
))
897 /* search for next available slot */
899 /* time to take a break? */
900 if (unlikely(--latency_ration
< 0)) {
903 spin_unlock(&si
->lock
);
905 spin_lock(&si
->lock
);
906 latency_ration
= LATENCY_LIMIT
;
909 /* try to get more slots in cluster */
910 if (si
->cluster_info
) {
911 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
913 } else if (si
->cluster_nr
&& !si
->swap_map
[++offset
]) {
914 /* non-ssd case, still more slots in cluster? */
920 * Even if there's no free clusters available (fragmented),
921 * try to scan a little more quickly with lock held unless we
922 * have scanned too many slots already.
925 unsigned long scan_limit
;
927 if (offset
< scan_base
)
928 scan_limit
= scan_base
;
930 scan_limit
= si
->highest_bit
;
931 for (; offset
<= scan_limit
&& --latency_ration
> 0;
933 if (!si
->swap_map
[offset
])
939 set_cluster_next(si
, offset
+ 1);
940 si
->flags
-= SWP_SCANNING
;
944 spin_unlock(&si
->lock
);
945 while (++offset
<= READ_ONCE(si
->highest_bit
)) {
946 if (data_race(!si
->swap_map
[offset
])) {
947 spin_lock(&si
->lock
);
950 if (vm_swap_full() &&
951 READ_ONCE(si
->swap_map
[offset
]) == SWAP_HAS_CACHE
) {
952 spin_lock(&si
->lock
);
955 if (unlikely(--latency_ration
< 0)) {
957 latency_ration
= LATENCY_LIMIT
;
961 offset
= si
->lowest_bit
;
962 while (offset
< scan_base
) {
963 if (data_race(!si
->swap_map
[offset
])) {
964 spin_lock(&si
->lock
);
967 if (vm_swap_full() &&
968 READ_ONCE(si
->swap_map
[offset
]) == SWAP_HAS_CACHE
) {
969 spin_lock(&si
->lock
);
972 if (unlikely(--latency_ration
< 0)) {
974 latency_ration
= LATENCY_LIMIT
;
979 spin_lock(&si
->lock
);
982 si
->flags
-= SWP_SCANNING
;
986 static int swap_alloc_cluster(struct swap_info_struct
*si
, swp_entry_t
*slot
)
989 struct swap_cluster_info
*ci
;
990 unsigned long offset
;
993 * Should not even be attempting cluster allocations when huge
994 * page swap is disabled. Warn and fail the allocation.
996 if (!IS_ENABLED(CONFIG_THP_SWAP
)) {
1001 if (cluster_list_empty(&si
->free_clusters
))
1004 idx
= cluster_list_first(&si
->free_clusters
);
1005 offset
= idx
* SWAPFILE_CLUSTER
;
1006 ci
= lock_cluster(si
, offset
);
1007 alloc_cluster(si
, idx
);
1008 cluster_set_count_flag(ci
, SWAPFILE_CLUSTER
, CLUSTER_FLAG_HUGE
);
1010 memset(si
->swap_map
+ offset
, SWAP_HAS_CACHE
, SWAPFILE_CLUSTER
);
1012 swap_range_alloc(si
, offset
, SWAPFILE_CLUSTER
);
1013 *slot
= swp_entry(si
->type
, offset
);
1018 static void swap_free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
1020 unsigned long offset
= idx
* SWAPFILE_CLUSTER
;
1021 struct swap_cluster_info
*ci
;
1023 ci
= lock_cluster(si
, offset
);
1024 memset(si
->swap_map
+ offset
, 0, SWAPFILE_CLUSTER
);
1025 cluster_set_count_flag(ci
, 0, 0);
1026 free_cluster(si
, idx
);
1028 swap_range_free(si
, offset
, SWAPFILE_CLUSTER
);
1031 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
1032 unsigned char usage
)
1037 n_ret
= scan_swap_map_slots(si
, usage
, 1, &entry
);
1040 return swp_offset(entry
);
1046 int get_swap_pages(int n_goal
, swp_entry_t swp_entries
[], int entry_size
)
1048 unsigned long size
= swap_entry_size(entry_size
);
1049 struct swap_info_struct
*si
, *next
;
1054 /* Only single cluster request supported */
1055 WARN_ON_ONCE(n_goal
> 1 && size
== SWAPFILE_CLUSTER
);
1057 spin_lock(&swap_avail_lock
);
1059 avail_pgs
= atomic_long_read(&nr_swap_pages
) / size
;
1060 if (avail_pgs
<= 0) {
1061 spin_unlock(&swap_avail_lock
);
1065 n_goal
= min3((long)n_goal
, (long)SWAP_BATCH
, avail_pgs
);
1067 atomic_long_sub(n_goal
* size
, &nr_swap_pages
);
1070 node
= numa_node_id();
1071 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[node
], avail_lists
[node
]) {
1072 /* requeue si to after same-priority siblings */
1073 plist_requeue(&si
->avail_lists
[node
], &swap_avail_heads
[node
]);
1074 spin_unlock(&swap_avail_lock
);
1075 spin_lock(&si
->lock
);
1076 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
1077 spin_lock(&swap_avail_lock
);
1078 if (plist_node_empty(&si
->avail_lists
[node
])) {
1079 spin_unlock(&si
->lock
);
1082 WARN(!si
->highest_bit
,
1083 "swap_info %d in list but !highest_bit\n",
1085 WARN(!(si
->flags
& SWP_WRITEOK
),
1086 "swap_info %d in list but !SWP_WRITEOK\n",
1088 __del_from_avail_list(si
);
1089 spin_unlock(&si
->lock
);
1092 if (size
== SWAPFILE_CLUSTER
) {
1093 if (si
->flags
& SWP_BLKDEV
)
1094 n_ret
= swap_alloc_cluster(si
, swp_entries
);
1096 n_ret
= scan_swap_map_slots(si
, SWAP_HAS_CACHE
,
1097 n_goal
, swp_entries
);
1098 spin_unlock(&si
->lock
);
1099 if (n_ret
|| size
== SWAPFILE_CLUSTER
)
1101 pr_debug("scan_swap_map of si %d failed to find offset\n",
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() can drop the si->lock, multiple
1109 * callers probably all tried to get a page from the same si
1110 * and it filled up before we could get one; or, the si filled
1111 * up between us dropping swap_avail_lock and taking si->lock.
1112 * Since we dropped the swap_avail_lock, the swap_avail_head
1113 * list may have been modified; so if next is still in the
1114 * swap_avail_head list then try it, otherwise start over
1115 * 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 /* The only caller of this function is now suspend routine */
1132 swp_entry_t
get_swap_page_of_type(int type
)
1134 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1140 spin_lock(&si
->lock
);
1141 if (si
->flags
& SWP_WRITEOK
) {
1142 /* This is called for allocating swap entry, not cache */
1143 offset
= scan_swap_map(si
, 1);
1145 atomic_long_dec(&nr_swap_pages
);
1146 spin_unlock(&si
->lock
);
1147 return swp_entry(type
, offset
);
1150 spin_unlock(&si
->lock
);
1152 return (swp_entry_t
) {0};
1155 static struct swap_info_struct
*__swap_info_get(swp_entry_t entry
)
1157 struct swap_info_struct
*p
;
1158 unsigned long offset
;
1162 p
= swp_swap_info(entry
);
1165 if (data_race(!(p
->flags
& SWP_USED
)))
1167 offset
= swp_offset(entry
);
1168 if (offset
>= p
->max
)
1173 pr_err("%s: %s%08lx\n", __func__
, Bad_offset
, entry
.val
);
1176 pr_err("%s: %s%08lx\n", __func__
, Unused_file
, entry
.val
);
1179 pr_err("%s: %s%08lx\n", __func__
, Bad_file
, entry
.val
);
1184 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
1186 struct swap_info_struct
*p
;
1188 p
= __swap_info_get(entry
);
1191 if (data_race(!p
->swap_map
[swp_offset(entry
)]))
1196 pr_err("%s: %s%08lx\n", __func__
, Unused_offset
, entry
.val
);
1201 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
1203 struct swap_info_struct
*p
;
1205 p
= _swap_info_get(entry
);
1207 spin_lock(&p
->lock
);
1211 static struct swap_info_struct
*swap_info_get_cont(swp_entry_t entry
,
1212 struct swap_info_struct
*q
)
1214 struct swap_info_struct
*p
;
1216 p
= _swap_info_get(entry
);
1220 spin_unlock(&q
->lock
);
1222 spin_lock(&p
->lock
);
1227 static unsigned char __swap_entry_free_locked(struct swap_info_struct
*p
,
1228 unsigned long offset
,
1229 unsigned char usage
)
1231 unsigned char count
;
1232 unsigned char has_cache
;
1234 count
= p
->swap_map
[offset
];
1236 has_cache
= count
& SWAP_HAS_CACHE
;
1237 count
&= ~SWAP_HAS_CACHE
;
1239 if (usage
== SWAP_HAS_CACHE
) {
1240 VM_BUG_ON(!has_cache
);
1242 } else if (count
== SWAP_MAP_SHMEM
) {
1244 * Or we could insist on shmem.c using a special
1245 * swap_shmem_free() and free_shmem_swap_and_cache()...
1248 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
1249 if (count
== COUNT_CONTINUED
) {
1250 if (swap_count_continued(p
, offset
, count
))
1251 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
1253 count
= SWAP_MAP_MAX
;
1258 usage
= count
| has_cache
;
1260 WRITE_ONCE(p
->swap_map
[offset
], usage
);
1262 WRITE_ONCE(p
->swap_map
[offset
], SWAP_HAS_CACHE
);
1268 * Check whether swap entry is valid in the swap device. If so,
1269 * return pointer to swap_info_struct, and keep the swap entry valid
1270 * via preventing the swap device from being swapoff, until
1271 * put_swap_device() is called. Otherwise return NULL.
1273 * The entirety of the RCU read critical section must come before the
1274 * return from or after the call to synchronize_rcu() in
1275 * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is
1276 * true, the si->map, si->cluster_info, etc. must be valid in the
1279 * Notice that swapoff or swapoff+swapon can still happen before the
1280 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1281 * in put_swap_device() if there isn't any other way to prevent
1282 * swapoff, such as page lock, page table lock, etc. The caller must
1283 * be prepared for that. For example, the following situation is
1288 * ... swapoff+swapon
1289 * __read_swap_cache_async()
1290 * swapcache_prepare()
1291 * __swap_duplicate()
1293 * // verify PTE not changed
1295 * In __swap_duplicate(), the swap_map need to be checked before
1296 * changing partly because the specified swap entry may be for another
1297 * swap device which has been swapoff. And in do_swap_page(), after
1298 * the page is read from the swap device, the PTE is verified not
1299 * changed with the page table locked to check whether the swap device
1300 * has been swapoff or swapoff+swapon.
1302 struct swap_info_struct
*get_swap_device(swp_entry_t entry
)
1304 struct swap_info_struct
*si
;
1305 unsigned long offset
;
1309 si
= swp_swap_info(entry
);
1314 if (data_race(!(si
->flags
& SWP_VALID
)))
1316 offset
= swp_offset(entry
);
1317 if (offset
>= si
->max
)
1322 pr_err("%s: %s%08lx\n", __func__
, Bad_file
, entry
.val
);
1330 static unsigned char __swap_entry_free(struct swap_info_struct
*p
,
1333 struct swap_cluster_info
*ci
;
1334 unsigned long offset
= swp_offset(entry
);
1335 unsigned char usage
;
1337 ci
= lock_cluster_or_swap_info(p
, offset
);
1338 usage
= __swap_entry_free_locked(p
, offset
, 1);
1339 unlock_cluster_or_swap_info(p
, ci
);
1341 free_swap_slot(entry
);
1346 static void swap_entry_free(struct swap_info_struct
*p
, swp_entry_t entry
)
1348 struct swap_cluster_info
*ci
;
1349 unsigned long offset
= swp_offset(entry
);
1350 unsigned char count
;
1352 ci
= lock_cluster(p
, offset
);
1353 count
= p
->swap_map
[offset
];
1354 VM_BUG_ON(count
!= SWAP_HAS_CACHE
);
1355 p
->swap_map
[offset
] = 0;
1356 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
1359 mem_cgroup_uncharge_swap(entry
, 1);
1360 swap_range_free(p
, offset
, 1);
1364 * Caller has made sure that the swap device corresponding to entry
1365 * is still around or has not been recycled.
1367 void swap_free(swp_entry_t entry
)
1369 struct swap_info_struct
*p
;
1371 p
= _swap_info_get(entry
);
1373 __swap_entry_free(p
, entry
);
1377 * Called after dropping swapcache to decrease refcnt to swap entries.
1379 void put_swap_page(struct page
*page
, swp_entry_t entry
)
1381 unsigned long offset
= swp_offset(entry
);
1382 unsigned long idx
= offset
/ SWAPFILE_CLUSTER
;
1383 struct swap_cluster_info
*ci
;
1384 struct swap_info_struct
*si
;
1386 unsigned int i
, free_entries
= 0;
1388 int size
= swap_entry_size(thp_nr_pages(page
));
1390 si
= _swap_info_get(entry
);
1394 ci
= lock_cluster_or_swap_info(si
, offset
);
1395 if (size
== SWAPFILE_CLUSTER
) {
1396 VM_BUG_ON(!cluster_is_huge(ci
));
1397 map
= si
->swap_map
+ offset
;
1398 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1400 VM_BUG_ON(!(val
& SWAP_HAS_CACHE
));
1401 if (val
== SWAP_HAS_CACHE
)
1404 cluster_clear_huge(ci
);
1405 if (free_entries
== SWAPFILE_CLUSTER
) {
1406 unlock_cluster_or_swap_info(si
, ci
);
1407 spin_lock(&si
->lock
);
1408 mem_cgroup_uncharge_swap(entry
, SWAPFILE_CLUSTER
);
1409 swap_free_cluster(si
, idx
);
1410 spin_unlock(&si
->lock
);
1414 for (i
= 0; i
< size
; i
++, entry
.val
++) {
1415 if (!__swap_entry_free_locked(si
, offset
+ i
, SWAP_HAS_CACHE
)) {
1416 unlock_cluster_or_swap_info(si
, ci
);
1417 free_swap_slot(entry
);
1420 lock_cluster_or_swap_info(si
, offset
);
1423 unlock_cluster_or_swap_info(si
, ci
);
1426 #ifdef CONFIG_THP_SWAP
1427 int split_swap_cluster(swp_entry_t entry
)
1429 struct swap_info_struct
*si
;
1430 struct swap_cluster_info
*ci
;
1431 unsigned long offset
= swp_offset(entry
);
1433 si
= _swap_info_get(entry
);
1436 ci
= lock_cluster(si
, offset
);
1437 cluster_clear_huge(ci
);
1443 static int swp_entry_cmp(const void *ent1
, const void *ent2
)
1445 const swp_entry_t
*e1
= ent1
, *e2
= ent2
;
1447 return (int)swp_type(*e1
) - (int)swp_type(*e2
);
1450 void swapcache_free_entries(swp_entry_t
*entries
, int n
)
1452 struct swap_info_struct
*p
, *prev
;
1462 * Sort swap entries by swap device, so each lock is only taken once.
1463 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1464 * so low that it isn't necessary to optimize further.
1466 if (nr_swapfiles
> 1)
1467 sort(entries
, n
, sizeof(entries
[0]), swp_entry_cmp
, NULL
);
1468 for (i
= 0; i
< n
; ++i
) {
1469 p
= swap_info_get_cont(entries
[i
], prev
);
1471 swap_entry_free(p
, entries
[i
]);
1475 spin_unlock(&p
->lock
);
1479 * How many references to page are currently swapped out?
1480 * This does not give an exact answer when swap count is continued,
1481 * but does include the high COUNT_CONTINUED flag to allow for that.
1483 int page_swapcount(struct page
*page
)
1486 struct swap_info_struct
*p
;
1487 struct swap_cluster_info
*ci
;
1489 unsigned long offset
;
1491 entry
.val
= page_private(page
);
1492 p
= _swap_info_get(entry
);
1494 offset
= swp_offset(entry
);
1495 ci
= lock_cluster_or_swap_info(p
, offset
);
1496 count
= swap_count(p
->swap_map
[offset
]);
1497 unlock_cluster_or_swap_info(p
, ci
);
1502 int __swap_count(swp_entry_t entry
)
1504 struct swap_info_struct
*si
;
1505 pgoff_t offset
= swp_offset(entry
);
1508 si
= get_swap_device(entry
);
1510 count
= swap_count(si
->swap_map
[offset
]);
1511 put_swap_device(si
);
1516 static int swap_swapcount(struct swap_info_struct
*si
, swp_entry_t entry
)
1519 pgoff_t offset
= swp_offset(entry
);
1520 struct swap_cluster_info
*ci
;
1522 ci
= lock_cluster_or_swap_info(si
, offset
);
1523 count
= swap_count(si
->swap_map
[offset
]);
1524 unlock_cluster_or_swap_info(si
, ci
);
1529 * How many references to @entry are currently swapped out?
1530 * This does not give an exact answer when swap count is continued,
1531 * but does include the high COUNT_CONTINUED flag to allow for that.
1533 int __swp_swapcount(swp_entry_t entry
)
1536 struct swap_info_struct
*si
;
1538 si
= get_swap_device(entry
);
1540 count
= swap_swapcount(si
, entry
);
1541 put_swap_device(si
);
1547 * How many references to @entry are currently swapped out?
1548 * This considers COUNT_CONTINUED so it returns exact answer.
1550 int swp_swapcount(swp_entry_t entry
)
1552 int count
, tmp_count
, n
;
1553 struct swap_info_struct
*p
;
1554 struct swap_cluster_info
*ci
;
1559 p
= _swap_info_get(entry
);
1563 offset
= swp_offset(entry
);
1565 ci
= lock_cluster_or_swap_info(p
, offset
);
1567 count
= swap_count(p
->swap_map
[offset
]);
1568 if (!(count
& COUNT_CONTINUED
))
1571 count
&= ~COUNT_CONTINUED
;
1572 n
= SWAP_MAP_MAX
+ 1;
1574 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1575 offset
&= ~PAGE_MASK
;
1576 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1579 page
= list_next_entry(page
, lru
);
1580 map
= kmap_atomic(page
);
1581 tmp_count
= map
[offset
];
1584 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1585 n
*= (SWAP_CONT_MAX
+ 1);
1586 } while (tmp_count
& COUNT_CONTINUED
);
1588 unlock_cluster_or_swap_info(p
, ci
);
1592 static bool swap_page_trans_huge_swapped(struct swap_info_struct
*si
,
1595 struct swap_cluster_info
*ci
;
1596 unsigned char *map
= si
->swap_map
;
1597 unsigned long roffset
= swp_offset(entry
);
1598 unsigned long offset
= round_down(roffset
, SWAPFILE_CLUSTER
);
1602 ci
= lock_cluster_or_swap_info(si
, offset
);
1603 if (!ci
|| !cluster_is_huge(ci
)) {
1604 if (swap_count(map
[roffset
]))
1608 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1609 if (swap_count(map
[offset
+ i
])) {
1615 unlock_cluster_or_swap_info(si
, ci
);
1619 static bool page_swapped(struct page
*page
)
1622 struct swap_info_struct
*si
;
1624 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
)))
1625 return page_swapcount(page
) != 0;
1627 page
= compound_head(page
);
1628 entry
.val
= page_private(page
);
1629 si
= _swap_info_get(entry
);
1631 return swap_page_trans_huge_swapped(si
, entry
);
1635 static int page_trans_huge_map_swapcount(struct page
*page
, int *total_mapcount
,
1636 int *total_swapcount
)
1638 int i
, map_swapcount
, _total_mapcount
, _total_swapcount
;
1639 unsigned long offset
= 0;
1640 struct swap_info_struct
*si
;
1641 struct swap_cluster_info
*ci
= NULL
;
1642 unsigned char *map
= NULL
;
1643 int mapcount
, swapcount
= 0;
1645 /* hugetlbfs shouldn't call it */
1646 VM_BUG_ON_PAGE(PageHuge(page
), page
);
1648 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
))) {
1649 mapcount
= page_trans_huge_mapcount(page
, total_mapcount
);
1650 if (PageSwapCache(page
))
1651 swapcount
= page_swapcount(page
);
1652 if (total_swapcount
)
1653 *total_swapcount
= swapcount
;
1654 return mapcount
+ swapcount
;
1657 page
= compound_head(page
);
1659 _total_mapcount
= _total_swapcount
= map_swapcount
= 0;
1660 if (PageSwapCache(page
)) {
1663 entry
.val
= page_private(page
);
1664 si
= _swap_info_get(entry
);
1667 offset
= swp_offset(entry
);
1671 ci
= lock_cluster(si
, offset
);
1672 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
1673 mapcount
= atomic_read(&page
[i
]._mapcount
) + 1;
1674 _total_mapcount
+= mapcount
;
1676 swapcount
= swap_count(map
[offset
+ i
]);
1677 _total_swapcount
+= swapcount
;
1679 map_swapcount
= max(map_swapcount
, mapcount
+ swapcount
);
1682 if (PageDoubleMap(page
)) {
1684 _total_mapcount
-= HPAGE_PMD_NR
;
1686 mapcount
= compound_mapcount(page
);
1687 map_swapcount
+= mapcount
;
1688 _total_mapcount
+= mapcount
;
1690 *total_mapcount
= _total_mapcount
;
1691 if (total_swapcount
)
1692 *total_swapcount
= _total_swapcount
;
1694 return map_swapcount
;
1698 * We can write to an anon page without COW if there are no other references
1699 * to it. And as a side-effect, free up its swap: because the old content
1700 * on disk will never be read, and seeking back there to write new content
1701 * later would only waste time away from clustering.
1703 * NOTE: total_map_swapcount should not be relied upon by the caller if
1704 * reuse_swap_page() returns false, but it may be always overwritten
1705 * (see the other implementation for CONFIG_SWAP=n).
1707 bool reuse_swap_page(struct page
*page
, int *total_map_swapcount
)
1709 int count
, total_mapcount
, total_swapcount
;
1711 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1712 if (unlikely(PageKsm(page
)))
1714 count
= page_trans_huge_map_swapcount(page
, &total_mapcount
,
1716 if (total_map_swapcount
)
1717 *total_map_swapcount
= total_mapcount
+ total_swapcount
;
1718 if (count
== 1 && PageSwapCache(page
) &&
1719 (likely(!PageTransCompound(page
)) ||
1720 /* The remaining swap count will be freed soon */
1721 total_swapcount
== page_swapcount(page
))) {
1722 if (!PageWriteback(page
)) {
1723 page
= compound_head(page
);
1724 delete_from_swap_cache(page
);
1728 struct swap_info_struct
*p
;
1730 entry
.val
= page_private(page
);
1731 p
= swap_info_get(entry
);
1732 if (p
->flags
& SWP_STABLE_WRITES
) {
1733 spin_unlock(&p
->lock
);
1736 spin_unlock(&p
->lock
);
1744 * If swap is getting full, or if there are no more mappings of this page,
1745 * then try_to_free_swap is called to free its swap space.
1747 int try_to_free_swap(struct page
*page
)
1749 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1751 if (!PageSwapCache(page
))
1753 if (PageWriteback(page
))
1755 if (page_swapped(page
))
1759 * Once hibernation has begun to create its image of memory,
1760 * there's a danger that one of the calls to try_to_free_swap()
1761 * - most probably a call from __try_to_reclaim_swap() while
1762 * hibernation is allocating its own swap pages for the image,
1763 * but conceivably even a call from memory reclaim - will free
1764 * the swap from a page which has already been recorded in the
1765 * image as a clean swapcache page, and then reuse its swap for
1766 * another page of the image. On waking from hibernation, the
1767 * original page might be freed under memory pressure, then
1768 * later read back in from swap, now with the wrong data.
1770 * Hibernation suspends storage while it is writing the image
1771 * to disk so check that here.
1773 if (pm_suspended_storage())
1776 page
= compound_head(page
);
1777 delete_from_swap_cache(page
);
1783 * Free the swap entry like above, but also try to
1784 * free the page cache entry if it is the last user.
1786 int free_swap_and_cache(swp_entry_t entry
)
1788 struct swap_info_struct
*p
;
1789 unsigned char count
;
1791 if (non_swap_entry(entry
))
1794 p
= _swap_info_get(entry
);
1796 count
= __swap_entry_free(p
, entry
);
1797 if (count
== SWAP_HAS_CACHE
&&
1798 !swap_page_trans_huge_swapped(p
, entry
))
1799 __try_to_reclaim_swap(p
, swp_offset(entry
),
1800 TTRS_UNMAPPED
| TTRS_FULL
);
1805 #ifdef CONFIG_HIBERNATION
1807 * Find the swap type that corresponds to given device (if any).
1809 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1810 * from 0, in which the swap header is expected to be located.
1812 * This is needed for the suspend to disk (aka swsusp).
1814 int swap_type_of(dev_t device
, sector_t offset
)
1821 spin_lock(&swap_lock
);
1822 for (type
= 0; type
< nr_swapfiles
; type
++) {
1823 struct swap_info_struct
*sis
= swap_info
[type
];
1825 if (!(sis
->flags
& SWP_WRITEOK
))
1828 if (device
== sis
->bdev
->bd_dev
) {
1829 struct swap_extent
*se
= first_se(sis
);
1831 if (se
->start_block
== offset
) {
1832 spin_unlock(&swap_lock
);
1837 spin_unlock(&swap_lock
);
1841 int find_first_swap(dev_t
*device
)
1845 spin_lock(&swap_lock
);
1846 for (type
= 0; type
< nr_swapfiles
; type
++) {
1847 struct swap_info_struct
*sis
= swap_info
[type
];
1849 if (!(sis
->flags
& SWP_WRITEOK
))
1851 *device
= sis
->bdev
->bd_dev
;
1852 spin_unlock(&swap_lock
);
1855 spin_unlock(&swap_lock
);
1860 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1861 * corresponding to given index in swap_info (swap type).
1863 sector_t
swapdev_block(int type
, pgoff_t offset
)
1865 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1866 struct swap_extent
*se
;
1868 if (!si
|| !(si
->flags
& SWP_WRITEOK
))
1870 se
= offset_to_swap_extent(si
, offset
);
1871 return se
->start_block
+ (offset
- se
->start_page
);
1875 * Return either the total number of swap pages of given type, or the number
1876 * of free pages of that type (depending on @free)
1878 * This is needed for software suspend
1880 unsigned int count_swap_pages(int type
, int free
)
1884 spin_lock(&swap_lock
);
1885 if ((unsigned int)type
< nr_swapfiles
) {
1886 struct swap_info_struct
*sis
= swap_info
[type
];
1888 spin_lock(&sis
->lock
);
1889 if (sis
->flags
& SWP_WRITEOK
) {
1892 n
-= sis
->inuse_pages
;
1894 spin_unlock(&sis
->lock
);
1896 spin_unlock(&swap_lock
);
1899 #endif /* CONFIG_HIBERNATION */
1901 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1903 return pte_same(pte_swp_clear_soft_dirty(pte
), swp_pte
);
1907 * No need to decide whether this PTE shares the swap entry with others,
1908 * just let do_wp_page work it out if a write is requested later - to
1909 * force COW, vm_page_prot omits write permission from any private vma.
1911 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1912 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1914 struct page
*swapcache
;
1920 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1921 if (unlikely(!page
))
1924 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1925 if (unlikely(!pte_same_as_swp(*pte
, swp_entry_to_pte(entry
)))) {
1930 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1931 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1933 set_pte_at(vma
->vm_mm
, addr
, pte
,
1934 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1935 if (page
== swapcache
) {
1936 page_add_anon_rmap(page
, vma
, addr
, false);
1937 } else { /* ksm created a completely new copy */
1938 page_add_new_anon_rmap(page
, vma
, addr
, false);
1939 lru_cache_add_inactive_or_unevictable(page
, vma
);
1943 pte_unmap_unlock(pte
, ptl
);
1944 if (page
!= swapcache
) {
1951 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1952 unsigned long addr
, unsigned long end
,
1953 unsigned int type
, bool frontswap
,
1954 unsigned long *fs_pages_to_unuse
)
1959 struct swap_info_struct
*si
;
1960 unsigned long offset
;
1962 volatile unsigned char *swap_map
;
1964 si
= swap_info
[type
];
1965 pte
= pte_offset_map(pmd
, addr
);
1967 if (!is_swap_pte(*pte
))
1970 entry
= pte_to_swp_entry(*pte
);
1971 if (swp_type(entry
) != type
)
1974 offset
= swp_offset(entry
);
1975 if (frontswap
&& !frontswap_test(si
, offset
))
1979 swap_map
= &si
->swap_map
[offset
];
1980 page
= lookup_swap_cache(entry
, vma
, addr
);
1982 struct vm_fault vmf
= {
1988 page
= swapin_readahead(entry
, GFP_HIGHUSER_MOVABLE
,
1992 if (*swap_map
== 0 || *swap_map
== SWAP_MAP_BAD
)
1998 wait_on_page_writeback(page
);
1999 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
2006 try_to_free_swap(page
);
2010 if (*fs_pages_to_unuse
&& !--(*fs_pages_to_unuse
)) {
2011 ret
= FRONTSWAP_PAGES_UNUSED
;
2015 pte
= pte_offset_map(pmd
, addr
);
2016 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
2024 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
2025 unsigned long addr
, unsigned long end
,
2026 unsigned int type
, bool frontswap
,
2027 unsigned long *fs_pages_to_unuse
)
2033 pmd
= pmd_offset(pud
, addr
);
2036 next
= pmd_addr_end(addr
, end
);
2037 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
2039 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, type
,
2040 frontswap
, fs_pages_to_unuse
);
2043 } while (pmd
++, addr
= next
, addr
!= end
);
2047 static inline int unuse_pud_range(struct vm_area_struct
*vma
, p4d_t
*p4d
,
2048 unsigned long addr
, unsigned long end
,
2049 unsigned int type
, bool frontswap
,
2050 unsigned long *fs_pages_to_unuse
)
2056 pud
= pud_offset(p4d
, addr
);
2058 next
= pud_addr_end(addr
, end
);
2059 if (pud_none_or_clear_bad(pud
))
2061 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, type
,
2062 frontswap
, fs_pages_to_unuse
);
2065 } while (pud
++, addr
= next
, addr
!= end
);
2069 static inline int unuse_p4d_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
2070 unsigned long addr
, unsigned long end
,
2071 unsigned int type
, bool frontswap
,
2072 unsigned long *fs_pages_to_unuse
)
2078 p4d
= p4d_offset(pgd
, addr
);
2080 next
= p4d_addr_end(addr
, end
);
2081 if (p4d_none_or_clear_bad(p4d
))
2083 ret
= unuse_pud_range(vma
, p4d
, addr
, next
, type
,
2084 frontswap
, fs_pages_to_unuse
);
2087 } while (p4d
++, addr
= next
, addr
!= end
);
2091 static int unuse_vma(struct vm_area_struct
*vma
, unsigned int type
,
2092 bool frontswap
, unsigned long *fs_pages_to_unuse
)
2095 unsigned long addr
, end
, next
;
2098 addr
= vma
->vm_start
;
2101 pgd
= pgd_offset(vma
->vm_mm
, addr
);
2103 next
= pgd_addr_end(addr
, end
);
2104 if (pgd_none_or_clear_bad(pgd
))
2106 ret
= unuse_p4d_range(vma
, pgd
, addr
, next
, type
,
2107 frontswap
, fs_pages_to_unuse
);
2110 } while (pgd
++, addr
= next
, addr
!= end
);
2114 static int unuse_mm(struct mm_struct
*mm
, unsigned int type
,
2115 bool frontswap
, unsigned long *fs_pages_to_unuse
)
2117 struct vm_area_struct
*vma
;
2121 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2122 if (vma
->anon_vma
) {
2123 ret
= unuse_vma(vma
, type
, frontswap
,
2130 mmap_read_unlock(mm
);
2135 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2136 * from current position to next entry still in use. Return 0
2137 * if there are no inuse entries after prev till end of the map.
2139 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
2140 unsigned int prev
, bool frontswap
)
2143 unsigned char count
;
2146 * No need for swap_lock here: we're just looking
2147 * for whether an entry is in use, not modifying it; false
2148 * hits are okay, and sys_swapoff() has already prevented new
2149 * allocations from this area (while holding swap_lock).
2151 for (i
= prev
+ 1; i
< si
->max
; i
++) {
2152 count
= READ_ONCE(si
->swap_map
[i
]);
2153 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
2154 if (!frontswap
|| frontswap_test(si
, i
))
2156 if ((i
% LATENCY_LIMIT
) == 0)
2167 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2168 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2170 int try_to_unuse(unsigned int type
, bool frontswap
,
2171 unsigned long pages_to_unuse
)
2173 struct mm_struct
*prev_mm
;
2174 struct mm_struct
*mm
;
2175 struct list_head
*p
;
2177 struct swap_info_struct
*si
= swap_info
[type
];
2182 if (!READ_ONCE(si
->inuse_pages
))
2189 retval
= shmem_unuse(type
, frontswap
, &pages_to_unuse
);
2196 spin_lock(&mmlist_lock
);
2197 p
= &init_mm
.mmlist
;
2198 while (READ_ONCE(si
->inuse_pages
) &&
2199 !signal_pending(current
) &&
2200 (p
= p
->next
) != &init_mm
.mmlist
) {
2202 mm
= list_entry(p
, struct mm_struct
, mmlist
);
2203 if (!mmget_not_zero(mm
))
2205 spin_unlock(&mmlist_lock
);
2208 retval
= unuse_mm(mm
, type
, frontswap
, &pages_to_unuse
);
2216 * Make sure that we aren't completely killing
2217 * interactive performance.
2220 spin_lock(&mmlist_lock
);
2222 spin_unlock(&mmlist_lock
);
2227 while (READ_ONCE(si
->inuse_pages
) &&
2228 !signal_pending(current
) &&
2229 (i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
2231 entry
= swp_entry(type
, i
);
2232 page
= find_get_page(swap_address_space(entry
), i
);
2237 * It is conceivable that a racing task removed this page from
2238 * swap cache just before we acquired the page lock. The page
2239 * might even be back in swap cache on another swap area. But
2240 * that is okay, try_to_free_swap() only removes stale pages.
2243 wait_on_page_writeback(page
);
2244 try_to_free_swap(page
);
2249 * For frontswap, we just need to unuse pages_to_unuse, if
2250 * it was specified. Need not check frontswap again here as
2251 * we already zeroed out pages_to_unuse if not frontswap.
2253 if (pages_to_unuse
&& --pages_to_unuse
== 0)
2258 * Lets check again to see if there are still swap entries in the map.
2259 * If yes, we would need to do retry the unuse logic again.
2260 * Under global memory pressure, swap entries can be reinserted back
2261 * into process space after the mmlist loop above passes over them.
2263 * Limit the number of retries? No: when mmget_not_zero() above fails,
2264 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2265 * at its own independent pace; and even shmem_writepage() could have
2266 * been preempted after get_swap_page(), temporarily hiding that swap.
2267 * It's easy and robust (though cpu-intensive) just to keep retrying.
2269 if (READ_ONCE(si
->inuse_pages
)) {
2270 if (!signal_pending(current
))
2275 return (retval
== FRONTSWAP_PAGES_UNUSED
) ? 0 : retval
;
2279 * After a successful try_to_unuse, if no swap is now in use, we know
2280 * we can empty the mmlist. swap_lock must be held on entry and exit.
2281 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2282 * added to the mmlist just after page_duplicate - before would be racy.
2284 static void drain_mmlist(void)
2286 struct list_head
*p
, *next
;
2289 for (type
= 0; type
< nr_swapfiles
; type
++)
2290 if (swap_info
[type
]->inuse_pages
)
2292 spin_lock(&mmlist_lock
);
2293 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
2295 spin_unlock(&mmlist_lock
);
2299 * Free all of a swapdev's extent information
2301 static void destroy_swap_extents(struct swap_info_struct
*sis
)
2303 while (!RB_EMPTY_ROOT(&sis
->swap_extent_root
)) {
2304 struct rb_node
*rb
= sis
->swap_extent_root
.rb_node
;
2305 struct swap_extent
*se
= rb_entry(rb
, struct swap_extent
, rb_node
);
2307 rb_erase(rb
, &sis
->swap_extent_root
);
2311 if (sis
->flags
& SWP_ACTIVATED
) {
2312 struct file
*swap_file
= sis
->swap_file
;
2313 struct address_space
*mapping
= swap_file
->f_mapping
;
2315 sis
->flags
&= ~SWP_ACTIVATED
;
2316 if (mapping
->a_ops
->swap_deactivate
)
2317 mapping
->a_ops
->swap_deactivate(swap_file
);
2322 * Add a block range (and the corresponding page range) into this swapdev's
2325 * This function rather assumes that it is called in ascending page order.
2328 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
2329 unsigned long nr_pages
, sector_t start_block
)
2331 struct rb_node
**link
= &sis
->swap_extent_root
.rb_node
, *parent
= NULL
;
2332 struct swap_extent
*se
;
2333 struct swap_extent
*new_se
;
2336 * place the new node at the right most since the
2337 * function is called in ascending page order.
2341 link
= &parent
->rb_right
;
2345 se
= rb_entry(parent
, struct swap_extent
, rb_node
);
2346 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
2347 if (se
->start_block
+ se
->nr_pages
== start_block
) {
2349 se
->nr_pages
+= nr_pages
;
2354 /* No merge, insert a new extent. */
2355 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
2358 new_se
->start_page
= start_page
;
2359 new_se
->nr_pages
= nr_pages
;
2360 new_se
->start_block
= start_block
;
2362 rb_link_node(&new_se
->rb_node
, parent
, link
);
2363 rb_insert_color(&new_se
->rb_node
, &sis
->swap_extent_root
);
2366 EXPORT_SYMBOL_GPL(add_swap_extent
);
2369 * A `swap extent' is a simple thing which maps a contiguous range of pages
2370 * onto a contiguous range of disk blocks. An ordered list of swap extents
2371 * is built at swapon time and is then used at swap_writepage/swap_readpage
2372 * time for locating where on disk a page belongs.
2374 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2375 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2376 * swap files identically.
2378 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2379 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2380 * swapfiles are handled *identically* after swapon time.
2382 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2383 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2384 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2385 * requirements, they are simply tossed out - we will never use those blocks
2388 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2389 * prevents users from writing to the swap device, which will corrupt memory.
2391 * The amount of disk space which a single swap extent represents varies.
2392 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2393 * extents in the list. To avoid much list walking, we cache the previous
2394 * search location in `curr_swap_extent', and start new searches from there.
2395 * This is extremely effective. The average number of iterations in
2396 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2398 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
2400 struct file
*swap_file
= sis
->swap_file
;
2401 struct address_space
*mapping
= swap_file
->f_mapping
;
2402 struct inode
*inode
= mapping
->host
;
2405 if (S_ISBLK(inode
->i_mode
)) {
2406 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2411 if (mapping
->a_ops
->swap_activate
) {
2412 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
2414 sis
->flags
|= SWP_ACTIVATED
;
2416 sis
->flags
|= SWP_FS_OPS
;
2417 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2423 return generic_swapfile_activate(sis
, swap_file
, span
);
2426 static int swap_node(struct swap_info_struct
*p
)
2428 struct block_device
*bdev
;
2433 bdev
= p
->swap_file
->f_inode
->i_sb
->s_bdev
;
2435 return bdev
? bdev
->bd_disk
->node_id
: NUMA_NO_NODE
;
2438 static void setup_swap_info(struct swap_info_struct
*p
, int prio
,
2439 unsigned char *swap_map
,
2440 struct swap_cluster_info
*cluster_info
)
2447 p
->prio
= --least_priority
;
2449 * the plist prio is negated because plist ordering is
2450 * low-to-high, while swap ordering is high-to-low
2452 p
->list
.prio
= -p
->prio
;
2455 p
->avail_lists
[i
].prio
= -p
->prio
;
2457 if (swap_node(p
) == i
)
2458 p
->avail_lists
[i
].prio
= 1;
2460 p
->avail_lists
[i
].prio
= -p
->prio
;
2463 p
->swap_map
= swap_map
;
2464 p
->cluster_info
= cluster_info
;
2467 static void _enable_swap_info(struct swap_info_struct
*p
)
2469 p
->flags
|= SWP_WRITEOK
| SWP_VALID
;
2470 atomic_long_add(p
->pages
, &nr_swap_pages
);
2471 total_swap_pages
+= p
->pages
;
2473 assert_spin_locked(&swap_lock
);
2475 * both lists are plists, and thus priority ordered.
2476 * swap_active_head needs to be priority ordered for swapoff(),
2477 * which on removal of any swap_info_struct with an auto-assigned
2478 * (i.e. negative) priority increments the auto-assigned priority
2479 * of any lower-priority swap_info_structs.
2480 * swap_avail_head needs to be priority ordered for get_swap_page(),
2481 * which allocates swap pages from the highest available priority
2484 plist_add(&p
->list
, &swap_active_head
);
2485 add_to_avail_list(p
);
2488 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
2489 unsigned char *swap_map
,
2490 struct swap_cluster_info
*cluster_info
,
2491 unsigned long *frontswap_map
)
2493 frontswap_init(p
->type
, frontswap_map
);
2494 spin_lock(&swap_lock
);
2495 spin_lock(&p
->lock
);
2496 setup_swap_info(p
, prio
, swap_map
, cluster_info
);
2497 spin_unlock(&p
->lock
);
2498 spin_unlock(&swap_lock
);
2500 * Guarantee swap_map, cluster_info, etc. fields are valid
2501 * between get/put_swap_device() if SWP_VALID bit is set
2504 spin_lock(&swap_lock
);
2505 spin_lock(&p
->lock
);
2506 _enable_swap_info(p
);
2507 spin_unlock(&p
->lock
);
2508 spin_unlock(&swap_lock
);
2511 static void reinsert_swap_info(struct swap_info_struct
*p
)
2513 spin_lock(&swap_lock
);
2514 spin_lock(&p
->lock
);
2515 setup_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
2516 _enable_swap_info(p
);
2517 spin_unlock(&p
->lock
);
2518 spin_unlock(&swap_lock
);
2521 bool has_usable_swap(void)
2525 spin_lock(&swap_lock
);
2526 if (plist_head_empty(&swap_active_head
))
2528 spin_unlock(&swap_lock
);
2532 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
2534 struct swap_info_struct
*p
= NULL
;
2535 unsigned char *swap_map
;
2536 struct swap_cluster_info
*cluster_info
;
2537 unsigned long *frontswap_map
;
2538 struct file
*swap_file
, *victim
;
2539 struct address_space
*mapping
;
2540 struct inode
*inode
;
2541 struct filename
*pathname
;
2543 unsigned int old_block_size
;
2545 if (!capable(CAP_SYS_ADMIN
))
2548 BUG_ON(!current
->mm
);
2550 pathname
= getname(specialfile
);
2551 if (IS_ERR(pathname
))
2552 return PTR_ERR(pathname
);
2554 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
2555 err
= PTR_ERR(victim
);
2559 mapping
= victim
->f_mapping
;
2560 spin_lock(&swap_lock
);
2561 plist_for_each_entry(p
, &swap_active_head
, list
) {
2562 if (p
->flags
& SWP_WRITEOK
) {
2563 if (p
->swap_file
->f_mapping
== mapping
) {
2571 spin_unlock(&swap_lock
);
2574 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2575 vm_unacct_memory(p
->pages
);
2578 spin_unlock(&swap_lock
);
2581 del_from_avail_list(p
);
2582 spin_lock(&p
->lock
);
2584 struct swap_info_struct
*si
= p
;
2587 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2590 for_each_node(nid
) {
2591 if (si
->avail_lists
[nid
].prio
!= 1)
2592 si
->avail_lists
[nid
].prio
--;
2597 plist_del(&p
->list
, &swap_active_head
);
2598 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2599 total_swap_pages
-= p
->pages
;
2600 p
->flags
&= ~SWP_WRITEOK
;
2601 spin_unlock(&p
->lock
);
2602 spin_unlock(&swap_lock
);
2604 disable_swap_slots_cache_lock();
2606 set_current_oom_origin();
2607 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
2608 clear_current_oom_origin();
2611 /* re-insert swap space back into swap_list */
2612 reinsert_swap_info(p
);
2613 reenable_swap_slots_cache_unlock();
2617 reenable_swap_slots_cache_unlock();
2619 spin_lock(&swap_lock
);
2620 spin_lock(&p
->lock
);
2621 p
->flags
&= ~SWP_VALID
; /* mark swap device as invalid */
2622 spin_unlock(&p
->lock
);
2623 spin_unlock(&swap_lock
);
2625 * wait for swap operations protected by get/put_swap_device()
2630 flush_work(&p
->discard_work
);
2632 destroy_swap_extents(p
);
2633 if (p
->flags
& SWP_CONTINUED
)
2634 free_swap_count_continuations(p
);
2636 if (!p
->bdev
|| !blk_queue_nonrot(bdev_get_queue(p
->bdev
)))
2637 atomic_dec(&nr_rotate_swap
);
2639 mutex_lock(&swapon_mutex
);
2640 spin_lock(&swap_lock
);
2641 spin_lock(&p
->lock
);
2644 /* wait for anyone still in scan_swap_map */
2645 p
->highest_bit
= 0; /* cuts scans short */
2646 while (p
->flags
>= SWP_SCANNING
) {
2647 spin_unlock(&p
->lock
);
2648 spin_unlock(&swap_lock
);
2649 schedule_timeout_uninterruptible(1);
2650 spin_lock(&swap_lock
);
2651 spin_lock(&p
->lock
);
2654 swap_file
= p
->swap_file
;
2655 old_block_size
= p
->old_block_size
;
2656 p
->swap_file
= NULL
;
2658 swap_map
= p
->swap_map
;
2660 cluster_info
= p
->cluster_info
;
2661 p
->cluster_info
= NULL
;
2662 frontswap_map
= frontswap_map_get(p
);
2663 spin_unlock(&p
->lock
);
2664 spin_unlock(&swap_lock
);
2665 arch_swap_invalidate_area(p
->type
);
2666 frontswap_invalidate_area(p
->type
);
2667 frontswap_map_set(p
, NULL
);
2668 mutex_unlock(&swapon_mutex
);
2669 free_percpu(p
->percpu_cluster
);
2670 p
->percpu_cluster
= NULL
;
2671 free_percpu(p
->cluster_next_cpu
);
2672 p
->cluster_next_cpu
= NULL
;
2674 kvfree(cluster_info
);
2675 kvfree(frontswap_map
);
2676 /* Destroy swap account information */
2677 swap_cgroup_swapoff(p
->type
);
2678 exit_swap_address_space(p
->type
);
2680 inode
= mapping
->host
;
2681 if (S_ISBLK(inode
->i_mode
)) {
2682 struct block_device
*bdev
= I_BDEV(inode
);
2684 set_blocksize(bdev
, old_block_size
);
2685 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2689 inode
->i_flags
&= ~S_SWAPFILE
;
2690 inode_unlock(inode
);
2691 filp_close(swap_file
, NULL
);
2694 * Clear the SWP_USED flag after all resources are freed so that swapon
2695 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2696 * not hold p->lock after we cleared its SWP_WRITEOK.
2698 spin_lock(&swap_lock
);
2700 spin_unlock(&swap_lock
);
2703 atomic_inc(&proc_poll_event
);
2704 wake_up_interruptible(&proc_poll_wait
);
2707 filp_close(victim
, NULL
);
2713 #ifdef CONFIG_PROC_FS
2714 static __poll_t
swaps_poll(struct file
*file
, poll_table
*wait
)
2716 struct seq_file
*seq
= file
->private_data
;
2718 poll_wait(file
, &proc_poll_wait
, wait
);
2720 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2721 seq
->poll_event
= atomic_read(&proc_poll_event
);
2722 return EPOLLIN
| EPOLLRDNORM
| EPOLLERR
| EPOLLPRI
;
2725 return EPOLLIN
| EPOLLRDNORM
;
2729 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2731 struct swap_info_struct
*si
;
2735 mutex_lock(&swapon_mutex
);
2738 return SEQ_START_TOKEN
;
2740 for (type
= 0; (si
= swap_type_to_swap_info(type
)); type
++) {
2741 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2750 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2752 struct swap_info_struct
*si
= v
;
2755 if (v
== SEQ_START_TOKEN
)
2758 type
= si
->type
+ 1;
2761 for (; (si
= swap_type_to_swap_info(type
)); type
++) {
2762 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2770 static void swap_stop(struct seq_file
*swap
, void *v
)
2772 mutex_unlock(&swapon_mutex
);
2775 static int swap_show(struct seq_file
*swap
, void *v
)
2777 struct swap_info_struct
*si
= v
;
2780 unsigned int bytes
, inuse
;
2782 if (si
== SEQ_START_TOKEN
) {
2783 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2787 bytes
= si
->pages
<< (PAGE_SHIFT
- 10);
2788 inuse
= si
->inuse_pages
<< (PAGE_SHIFT
- 10);
2790 file
= si
->swap_file
;
2791 len
= seq_file_path(swap
, file
, " \t\n\\");
2792 seq_printf(swap
, "%*s%s\t%u\t%s%u\t%s%d\n",
2793 len
< 40 ? 40 - len
: 1, " ",
2794 S_ISBLK(file_inode(file
)->i_mode
) ?
2795 "partition" : "file\t",
2796 bytes
, bytes
< 10000000 ? "\t" : "",
2797 inuse
, inuse
< 10000000 ? "\t" : "",
2802 static const struct seq_operations swaps_op
= {
2803 .start
= swap_start
,
2809 static int swaps_open(struct inode
*inode
, struct file
*file
)
2811 struct seq_file
*seq
;
2814 ret
= seq_open(file
, &swaps_op
);
2818 seq
= file
->private_data
;
2819 seq
->poll_event
= atomic_read(&proc_poll_event
);
2823 static const struct proc_ops swaps_proc_ops
= {
2824 .proc_flags
= PROC_ENTRY_PERMANENT
,
2825 .proc_open
= swaps_open
,
2826 .proc_read
= seq_read
,
2827 .proc_lseek
= seq_lseek
,
2828 .proc_release
= seq_release
,
2829 .proc_poll
= swaps_poll
,
2832 static int __init
procswaps_init(void)
2834 proc_create("swaps", 0, NULL
, &swaps_proc_ops
);
2837 __initcall(procswaps_init
);
2838 #endif /* CONFIG_PROC_FS */
2840 #ifdef MAX_SWAPFILES_CHECK
2841 static int __init
max_swapfiles_check(void)
2843 MAX_SWAPFILES_CHECK();
2846 late_initcall(max_swapfiles_check
);
2849 static struct swap_info_struct
*alloc_swap_info(void)
2851 struct swap_info_struct
*p
;
2852 struct swap_info_struct
*defer
= NULL
;
2856 p
= kvzalloc(struct_size(p
, avail_lists
, nr_node_ids
), GFP_KERNEL
);
2858 return ERR_PTR(-ENOMEM
);
2860 spin_lock(&swap_lock
);
2861 for (type
= 0; type
< nr_swapfiles
; type
++) {
2862 if (!(swap_info
[type
]->flags
& SWP_USED
))
2865 if (type
>= MAX_SWAPFILES
) {
2866 spin_unlock(&swap_lock
);
2868 return ERR_PTR(-EPERM
);
2870 if (type
>= nr_swapfiles
) {
2872 WRITE_ONCE(swap_info
[type
], p
);
2874 * Write swap_info[type] before nr_swapfiles, in case a
2875 * racing procfs swap_start() or swap_next() is reading them.
2876 * (We never shrink nr_swapfiles, we never free this entry.)
2879 WRITE_ONCE(nr_swapfiles
, nr_swapfiles
+ 1);
2882 p
= swap_info
[type
];
2884 * Do not memset this entry: a racing procfs swap_next()
2885 * would be relying on p->type to remain valid.
2888 p
->swap_extent_root
= RB_ROOT
;
2889 plist_node_init(&p
->list
, 0);
2891 plist_node_init(&p
->avail_lists
[i
], 0);
2892 p
->flags
= SWP_USED
;
2893 spin_unlock(&swap_lock
);
2895 spin_lock_init(&p
->lock
);
2896 spin_lock_init(&p
->cont_lock
);
2901 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2905 if (S_ISBLK(inode
->i_mode
)) {
2906 p
->bdev
= blkdev_get_by_dev(inode
->i_rdev
,
2907 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, p
);
2908 if (IS_ERR(p
->bdev
)) {
2909 error
= PTR_ERR(p
->bdev
);
2913 p
->old_block_size
= block_size(p
->bdev
);
2914 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2918 * Zoned block devices contain zones that have a sequential
2919 * write only restriction. Hence zoned block devices are not
2920 * suitable for swapping. Disallow them here.
2922 if (blk_queue_is_zoned(p
->bdev
->bd_disk
->queue
))
2924 p
->flags
|= SWP_BLKDEV
;
2925 } else if (S_ISREG(inode
->i_mode
)) {
2926 p
->bdev
= inode
->i_sb
->s_bdev
;
2934 * Find out how many pages are allowed for a single swap device. There
2935 * are two limiting factors:
2936 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2937 * 2) the number of bits in the swap pte, as defined by the different
2940 * In order to find the largest possible bit mask, a swap entry with
2941 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2942 * decoded to a swp_entry_t again, and finally the swap offset is
2945 * This will mask all the bits from the initial ~0UL mask that can't
2946 * be encoded in either the swp_entry_t or the architecture definition
2949 unsigned long generic_max_swapfile_size(void)
2951 return swp_offset(pte_to_swp_entry(
2952 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2955 /* Can be overridden by an architecture for additional checks. */
2956 __weak
unsigned long max_swapfile_size(void)
2958 return generic_max_swapfile_size();
2961 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2962 union swap_header
*swap_header
,
2963 struct inode
*inode
)
2966 unsigned long maxpages
;
2967 unsigned long swapfilepages
;
2968 unsigned long last_page
;
2970 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2971 pr_err("Unable to find swap-space signature\n");
2975 /* swap partition endianess hack... */
2976 if (swab32(swap_header
->info
.version
) == 1) {
2977 swab32s(&swap_header
->info
.version
);
2978 swab32s(&swap_header
->info
.last_page
);
2979 swab32s(&swap_header
->info
.nr_badpages
);
2980 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2982 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2983 swab32s(&swap_header
->info
.badpages
[i
]);
2985 /* Check the swap header's sub-version */
2986 if (swap_header
->info
.version
!= 1) {
2987 pr_warn("Unable to handle swap header version %d\n",
2988 swap_header
->info
.version
);
2993 p
->cluster_next
= 1;
2996 maxpages
= max_swapfile_size();
2997 last_page
= swap_header
->info
.last_page
;
2999 pr_warn("Empty swap-file\n");
3002 if (last_page
> maxpages
) {
3003 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
3004 maxpages
<< (PAGE_SHIFT
- 10),
3005 last_page
<< (PAGE_SHIFT
- 10));
3007 if (maxpages
> last_page
) {
3008 maxpages
= last_page
+ 1;
3009 /* p->max is an unsigned int: don't overflow it */
3010 if ((unsigned int)maxpages
== 0)
3011 maxpages
= UINT_MAX
;
3013 p
->highest_bit
= maxpages
- 1;
3017 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
3018 if (swapfilepages
&& maxpages
> swapfilepages
) {
3019 pr_warn("Swap area shorter than signature indicates\n");
3022 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
3024 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
3030 #define SWAP_CLUSTER_INFO_COLS \
3031 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3032 #define SWAP_CLUSTER_SPACE_COLS \
3033 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3034 #define SWAP_CLUSTER_COLS \
3035 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3037 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
3038 union swap_header
*swap_header
,
3039 unsigned char *swap_map
,
3040 struct swap_cluster_info
*cluster_info
,
3041 unsigned long maxpages
,
3045 unsigned int nr_good_pages
;
3047 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3048 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
3049 unsigned long i
, idx
;
3051 nr_good_pages
= maxpages
- 1; /* omit header page */
3053 cluster_list_init(&p
->free_clusters
);
3054 cluster_list_init(&p
->discard_clusters
);
3056 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
3057 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
3058 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
3060 if (page_nr
< maxpages
) {
3061 swap_map
[page_nr
] = SWAP_MAP_BAD
;
3064 * Haven't marked the cluster free yet, no list
3065 * operation involved
3067 inc_cluster_info_page(p
, cluster_info
, page_nr
);
3071 /* Haven't marked the cluster free yet, no list operation involved */
3072 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
3073 inc_cluster_info_page(p
, cluster_info
, i
);
3075 if (nr_good_pages
) {
3076 swap_map
[0] = SWAP_MAP_BAD
;
3078 * Not mark the cluster free yet, no list
3079 * operation involved
3081 inc_cluster_info_page(p
, cluster_info
, 0);
3083 p
->pages
= nr_good_pages
;
3084 nr_extents
= setup_swap_extents(p
, span
);
3087 nr_good_pages
= p
->pages
;
3089 if (!nr_good_pages
) {
3090 pr_warn("Empty swap-file\n");
3099 * Reduce false cache line sharing between cluster_info and
3100 * sharing same address space.
3102 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
3103 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
3104 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
3105 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
3106 if (idx
>= nr_clusters
)
3108 if (cluster_count(&cluster_info
[idx
]))
3110 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
3111 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
3119 * Helper to sys_swapon determining if a given swap
3120 * backing device queue supports DISCARD operations.
3122 static bool swap_discardable(struct swap_info_struct
*si
)
3124 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
3126 if (!q
|| !blk_queue_discard(q
))
3132 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
3134 struct swap_info_struct
*p
;
3135 struct filename
*name
;
3136 struct file
*swap_file
= NULL
;
3137 struct address_space
*mapping
;
3140 union swap_header
*swap_header
;
3143 unsigned long maxpages
;
3144 unsigned char *swap_map
= NULL
;
3145 struct swap_cluster_info
*cluster_info
= NULL
;
3146 unsigned long *frontswap_map
= NULL
;
3147 struct page
*page
= NULL
;
3148 struct inode
*inode
= NULL
;
3149 bool inced_nr_rotate_swap
= false;
3151 if (swap_flags
& ~SWAP_FLAGS_VALID
)
3154 if (!capable(CAP_SYS_ADMIN
))
3157 if (!swap_avail_heads
)
3160 p
= alloc_swap_info();
3164 INIT_WORK(&p
->discard_work
, swap_discard_work
);
3166 name
= getname(specialfile
);
3168 error
= PTR_ERR(name
);
3172 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
3173 if (IS_ERR(swap_file
)) {
3174 error
= PTR_ERR(swap_file
);
3179 p
->swap_file
= swap_file
;
3180 mapping
= swap_file
->f_mapping
;
3181 inode
= mapping
->host
;
3183 error
= claim_swapfile(p
, inode
);
3184 if (unlikely(error
))
3188 if (IS_SWAPFILE(inode
)) {
3190 goto bad_swap_unlock_inode
;
3194 * Read the swap header.
3196 if (!mapping
->a_ops
->readpage
) {
3198 goto bad_swap_unlock_inode
;
3200 page
= read_mapping_page(mapping
, 0, swap_file
);
3202 error
= PTR_ERR(page
);
3203 goto bad_swap_unlock_inode
;
3205 swap_header
= kmap(page
);
3207 maxpages
= read_swap_header(p
, swap_header
, inode
);
3208 if (unlikely(!maxpages
)) {
3210 goto bad_swap_unlock_inode
;
3213 /* OK, set up the swap map and apply the bad block list */
3214 swap_map
= vzalloc(maxpages
);
3217 goto bad_swap_unlock_inode
;
3220 if (p
->bdev
&& blk_queue_stable_writes(p
->bdev
->bd_disk
->queue
))
3221 p
->flags
|= SWP_STABLE_WRITES
;
3223 if (p
->bdev
&& p
->bdev
->bd_disk
->fops
->rw_page
)
3224 p
->flags
|= SWP_SYNCHRONOUS_IO
;
3226 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
3228 unsigned long ci
, nr_cluster
;
3230 p
->flags
|= SWP_SOLIDSTATE
;
3231 p
->cluster_next_cpu
= alloc_percpu(unsigned int);
3232 if (!p
->cluster_next_cpu
) {
3234 goto bad_swap_unlock_inode
;
3237 * select a random position to start with to help wear leveling
3240 for_each_possible_cpu(cpu
) {
3241 per_cpu(*p
->cluster_next_cpu
, cpu
) =
3242 1 + prandom_u32_max(p
->highest_bit
);
3244 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3246 cluster_info
= kvcalloc(nr_cluster
, sizeof(*cluster_info
),
3248 if (!cluster_info
) {
3250 goto bad_swap_unlock_inode
;
3253 for (ci
= 0; ci
< nr_cluster
; ci
++)
3254 spin_lock_init(&((cluster_info
+ ci
)->lock
));
3256 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
3257 if (!p
->percpu_cluster
) {
3259 goto bad_swap_unlock_inode
;
3261 for_each_possible_cpu(cpu
) {
3262 struct percpu_cluster
*cluster
;
3263 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
3264 cluster_set_null(&cluster
->index
);
3267 atomic_inc(&nr_rotate_swap
);
3268 inced_nr_rotate_swap
= true;
3271 error
= swap_cgroup_swapon(p
->type
, maxpages
);
3273 goto bad_swap_unlock_inode
;
3275 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
3276 cluster_info
, maxpages
, &span
);
3277 if (unlikely(nr_extents
< 0)) {
3279 goto bad_swap_unlock_inode
;
3281 /* frontswap enabled? set up bit-per-page map for frontswap */
3282 if (IS_ENABLED(CONFIG_FRONTSWAP
))
3283 frontswap_map
= kvcalloc(BITS_TO_LONGS(maxpages
),
3287 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
3289 * When discard is enabled for swap with no particular
3290 * policy flagged, we set all swap discard flags here in
3291 * order to sustain backward compatibility with older
3292 * swapon(8) releases.
3294 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
3298 * By flagging sys_swapon, a sysadmin can tell us to
3299 * either do single-time area discards only, or to just
3300 * perform discards for released swap page-clusters.
3301 * Now it's time to adjust the p->flags accordingly.
3303 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
3304 p
->flags
&= ~SWP_PAGE_DISCARD
;
3305 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
3306 p
->flags
&= ~SWP_AREA_DISCARD
;
3308 /* issue a swapon-time discard if it's still required */
3309 if (p
->flags
& SWP_AREA_DISCARD
) {
3310 int err
= discard_swap(p
);
3312 pr_err("swapon: discard_swap(%p): %d\n",
3317 error
= init_swap_address_space(p
->type
, maxpages
);
3319 goto bad_swap_unlock_inode
;
3322 * Flush any pending IO and dirty mappings before we start using this
3325 inode
->i_flags
|= S_SWAPFILE
;
3326 error
= inode_drain_writes(inode
);
3328 inode
->i_flags
&= ~S_SWAPFILE
;
3329 goto free_swap_address_space
;
3332 mutex_lock(&swapon_mutex
);
3334 if (swap_flags
& SWAP_FLAG_PREFER
)
3336 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
3337 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
3339 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3340 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
3341 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
3342 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
3343 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
3344 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
3345 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
3346 (frontswap_map
) ? "FS" : "");
3348 mutex_unlock(&swapon_mutex
);
3349 atomic_inc(&proc_poll_event
);
3350 wake_up_interruptible(&proc_poll_wait
);
3354 free_swap_address_space
:
3355 exit_swap_address_space(p
->type
);
3356 bad_swap_unlock_inode
:
3357 inode_unlock(inode
);
3359 free_percpu(p
->percpu_cluster
);
3360 p
->percpu_cluster
= NULL
;
3361 free_percpu(p
->cluster_next_cpu
);
3362 p
->cluster_next_cpu
= NULL
;
3363 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
3364 set_blocksize(p
->bdev
, p
->old_block_size
);
3365 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
3368 destroy_swap_extents(p
);
3369 swap_cgroup_swapoff(p
->type
);
3370 spin_lock(&swap_lock
);
3371 p
->swap_file
= NULL
;
3373 spin_unlock(&swap_lock
);
3375 kvfree(cluster_info
);
3376 kvfree(frontswap_map
);
3377 if (inced_nr_rotate_swap
)
3378 atomic_dec(&nr_rotate_swap
);
3380 filp_close(swap_file
, NULL
);
3382 if (page
&& !IS_ERR(page
)) {
3389 inode_unlock(inode
);
3391 enable_swap_slots_cache();
3395 void si_swapinfo(struct sysinfo
*val
)
3398 unsigned long nr_to_be_unused
= 0;
3400 spin_lock(&swap_lock
);
3401 for (type
= 0; type
< nr_swapfiles
; type
++) {
3402 struct swap_info_struct
*si
= swap_info
[type
];
3404 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
3405 nr_to_be_unused
+= si
->inuse_pages
;
3407 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
3408 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
3409 spin_unlock(&swap_lock
);
3413 * Verify that a swap entry is valid and increment its swap map count.
3415 * Returns error code in following case.
3417 * - swp_entry is invalid -> EINVAL
3418 * - swp_entry is migration entry -> EINVAL
3419 * - swap-cache reference is requested but there is already one. -> EEXIST
3420 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3421 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3423 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
3425 struct swap_info_struct
*p
;
3426 struct swap_cluster_info
*ci
;
3427 unsigned long offset
;
3428 unsigned char count
;
3429 unsigned char has_cache
;
3432 p
= get_swap_device(entry
);
3436 offset
= swp_offset(entry
);
3437 ci
= lock_cluster_or_swap_info(p
, offset
);
3439 count
= p
->swap_map
[offset
];
3442 * swapin_readahead() doesn't check if a swap entry is valid, so the
3443 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3445 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
3450 has_cache
= count
& SWAP_HAS_CACHE
;
3451 count
&= ~SWAP_HAS_CACHE
;
3454 if (usage
== SWAP_HAS_CACHE
) {
3456 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3457 if (!has_cache
&& count
)
3458 has_cache
= SWAP_HAS_CACHE
;
3459 else if (has_cache
) /* someone else added cache */
3461 else /* no users remaining */
3464 } else if (count
|| has_cache
) {
3466 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
3468 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
3470 else if (swap_count_continued(p
, offset
, count
))
3471 count
= COUNT_CONTINUED
;
3475 err
= -ENOENT
; /* unused swap entry */
3477 WRITE_ONCE(p
->swap_map
[offset
], count
| has_cache
);
3480 unlock_cluster_or_swap_info(p
, ci
);
3487 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3488 * (in which case its reference count is never incremented).
3490 void swap_shmem_alloc(swp_entry_t entry
)
3492 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
3496 * Increase reference count of swap entry by 1.
3497 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3498 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3499 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3500 * might occur if a page table entry has got corrupted.
3502 int swap_duplicate(swp_entry_t entry
)
3506 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
3507 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
3512 * @entry: swap entry for which we allocate swap cache.
3514 * Called when allocating swap cache for existing swap entry,
3515 * This can return error codes. Returns 0 at success.
3516 * -EEXIST means there is a swap cache.
3517 * Note: return code is different from swap_duplicate().
3519 int swapcache_prepare(swp_entry_t entry
)
3521 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
3524 struct swap_info_struct
*swp_swap_info(swp_entry_t entry
)
3526 return swap_type_to_swap_info(swp_type(entry
));
3529 struct swap_info_struct
*page_swap_info(struct page
*page
)
3531 swp_entry_t entry
= { .val
= page_private(page
) };
3532 return swp_swap_info(entry
);
3536 * out-of-line __page_file_ methods to avoid include hell.
3538 struct address_space
*__page_file_mapping(struct page
*page
)
3540 return page_swap_info(page
)->swap_file
->f_mapping
;
3542 EXPORT_SYMBOL_GPL(__page_file_mapping
);
3544 pgoff_t
__page_file_index(struct page
*page
)
3546 swp_entry_t swap
= { .val
= page_private(page
) };
3547 return swp_offset(swap
);
3549 EXPORT_SYMBOL_GPL(__page_file_index
);
3552 * add_swap_count_continuation - called when a swap count is duplicated
3553 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3554 * page of the original vmalloc'ed swap_map, to hold the continuation count
3555 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3556 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3558 * These continuation pages are seldom referenced: the common paths all work
3559 * on the original swap_map, only referring to a continuation page when the
3560 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3562 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3563 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3564 * can be called after dropping locks.
3566 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
3568 struct swap_info_struct
*si
;
3569 struct swap_cluster_info
*ci
;
3572 struct page
*list_page
;
3574 unsigned char count
;
3578 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3579 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3581 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
3583 si
= get_swap_device(entry
);
3586 * An acceptable race has occurred since the failing
3587 * __swap_duplicate(): the swap device may be swapoff
3591 spin_lock(&si
->lock
);
3593 offset
= swp_offset(entry
);
3595 ci
= lock_cluster(si
, offset
);
3597 count
= swap_count(si
->swap_map
[offset
]);
3599 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
3601 * The higher the swap count, the more likely it is that tasks
3602 * will race to add swap count continuation: we need to avoid
3603 * over-provisioning.
3614 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3615 * no architecture is using highmem pages for kernel page tables: so it
3616 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3618 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3619 offset
&= ~PAGE_MASK
;
3621 spin_lock(&si
->cont_lock
);
3623 * Page allocation does not initialize the page's lru field,
3624 * but it does always reset its private field.
3626 if (!page_private(head
)) {
3627 BUG_ON(count
& COUNT_CONTINUED
);
3628 INIT_LIST_HEAD(&head
->lru
);
3629 set_page_private(head
, SWP_CONTINUED
);
3630 si
->flags
|= SWP_CONTINUED
;
3633 list_for_each_entry(list_page
, &head
->lru
, lru
) {
3637 * If the previous map said no continuation, but we've found
3638 * a continuation page, free our allocation and use this one.
3640 if (!(count
& COUNT_CONTINUED
))
3641 goto out_unlock_cont
;
3643 map
= kmap_atomic(list_page
) + offset
;
3648 * If this continuation count now has some space in it,
3649 * free our allocation and use this one.
3651 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
3652 goto out_unlock_cont
;
3655 list_add_tail(&page
->lru
, &head
->lru
);
3656 page
= NULL
; /* now it's attached, don't free it */
3658 spin_unlock(&si
->cont_lock
);
3661 spin_unlock(&si
->lock
);
3662 put_swap_device(si
);
3670 * swap_count_continued - when the original swap_map count is incremented
3671 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3672 * into, carry if so, or else fail until a new continuation page is allocated;
3673 * when the original swap_map count is decremented from 0 with continuation,
3674 * borrow from the continuation and report whether it still holds more.
3675 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3678 static bool swap_count_continued(struct swap_info_struct
*si
,
3679 pgoff_t offset
, unsigned char count
)
3686 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3687 if (page_private(head
) != SWP_CONTINUED
) {
3688 BUG_ON(count
& COUNT_CONTINUED
);
3689 return false; /* need to add count continuation */
3692 spin_lock(&si
->cont_lock
);
3693 offset
&= ~PAGE_MASK
;
3694 page
= list_next_entry(head
, lru
);
3695 map
= kmap_atomic(page
) + offset
;
3697 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3698 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3700 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3702 * Think of how you add 1 to 999
3704 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3706 page
= list_next_entry(page
, lru
);
3707 BUG_ON(page
== head
);
3708 map
= kmap_atomic(page
) + offset
;
3710 if (*map
== SWAP_CONT_MAX
) {
3712 page
= list_next_entry(page
, lru
);
3714 ret
= false; /* add count continuation */
3717 map
= kmap_atomic(page
) + offset
;
3718 init_map
: *map
= 0; /* we didn't zero the page */
3722 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3723 map
= kmap_atomic(page
) + offset
;
3724 *map
= COUNT_CONTINUED
;
3727 ret
= true; /* incremented */
3729 } else { /* decrementing */
3731 * Think of how you subtract 1 from 1000
3733 BUG_ON(count
!= COUNT_CONTINUED
);
3734 while (*map
== COUNT_CONTINUED
) {
3736 page
= list_next_entry(page
, lru
);
3737 BUG_ON(page
== head
);
3738 map
= kmap_atomic(page
) + offset
;
3745 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3746 map
= kmap_atomic(page
) + offset
;
3747 *map
= SWAP_CONT_MAX
| count
;
3748 count
= COUNT_CONTINUED
;
3751 ret
= count
== COUNT_CONTINUED
;
3754 spin_unlock(&si
->cont_lock
);
3759 * free_swap_count_continuations - swapoff free all the continuation pages
3760 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3762 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3766 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3768 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3769 if (page_private(head
)) {
3770 struct page
*page
, *next
;
3772 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3773 list_del(&page
->lru
);
3780 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3781 void cgroup_throttle_swaprate(struct page
*page
, gfp_t gfp_mask
)
3783 struct swap_info_struct
*si
, *next
;
3784 int nid
= page_to_nid(page
);
3786 if (!(gfp_mask
& __GFP_IO
))
3789 if (!blk_cgroup_congested())
3793 * We've already scheduled a throttle, avoid taking the global swap
3796 if (current
->throttle_queue
)
3799 spin_lock(&swap_avail_lock
);
3800 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[nid
],
3803 blkcg_schedule_throttle(bdev_get_queue(si
->bdev
), true);
3807 spin_unlock(&swap_avail_lock
);
3811 static int __init
swapfile_init(void)
3815 swap_avail_heads
= kmalloc_array(nr_node_ids
, sizeof(struct plist_head
),
3817 if (!swap_avail_heads
) {
3818 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3823 plist_head_init(&swap_avail_heads
[nid
]);
3827 subsys_initcall(swapfile_init
);