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/pgtable.h>
44 #include <asm/tlbflush.h>
45 #include <linux/swapops.h>
46 #include <linux/swap_cgroup.h>
48 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
50 static void free_swap_count_continuations(struct swap_info_struct
*);
51 static sector_t
map_swap_entry(swp_entry_t
, struct block_device
**);
53 DEFINE_SPINLOCK(swap_lock
);
54 static unsigned int nr_swapfiles
;
55 atomic_long_t nr_swap_pages
;
57 * Some modules use swappable objects and may try to swap them out under
58 * memory pressure (via the shrinker). Before doing so, they may wish to
59 * check to see if any swap space is available.
61 EXPORT_SYMBOL_GPL(nr_swap_pages
);
62 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
63 long total_swap_pages
;
64 static int least_priority
= -1;
66 static const char Bad_file
[] = "Bad swap file entry ";
67 static const char Unused_file
[] = "Unused swap file entry ";
68 static const char Bad_offset
[] = "Bad swap offset entry ";
69 static const char Unused_offset
[] = "Unused swap offset entry ";
72 * all active swap_info_structs
73 * protected with swap_lock, and ordered by priority.
75 PLIST_HEAD(swap_active_head
);
78 * all available (active, not full) swap_info_structs
79 * protected with swap_avail_lock, ordered by priority.
80 * This is used by get_swap_page() instead of swap_active_head
81 * because swap_active_head includes all swap_info_structs,
82 * but get_swap_page() doesn't need to look at full ones.
83 * This uses its own lock instead of swap_lock because when a
84 * swap_info_struct changes between not-full/full, it needs to
85 * add/remove itself to/from this list, but the swap_info_struct->lock
86 * is held and the locking order requires swap_lock to be taken
87 * before any swap_info_struct->lock.
89 static struct plist_head
*swap_avail_heads
;
90 static DEFINE_SPINLOCK(swap_avail_lock
);
92 struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
94 static DEFINE_MUTEX(swapon_mutex
);
96 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
97 /* Activity counter to indicate that a swapon or swapoff has occurred */
98 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
100 atomic_t nr_rotate_swap
= ATOMIC_INIT(0);
102 static struct swap_info_struct
*swap_type_to_swap_info(int type
)
104 if (type
>= READ_ONCE(nr_swapfiles
))
107 smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */
108 return READ_ONCE(swap_info
[type
]);
111 static inline unsigned char swap_count(unsigned char ent
)
113 return ent
& ~SWAP_HAS_CACHE
; /* may include COUNT_CONTINUED flag */
116 /* Reclaim the swap entry anyway if possible */
117 #define TTRS_ANYWAY 0x1
119 * Reclaim the swap entry if there are no more mappings of the
122 #define TTRS_UNMAPPED 0x2
123 /* Reclaim the swap entry if swap is getting full*/
124 #define TTRS_FULL 0x4
126 /* returns 1 if swap entry is freed */
127 static int __try_to_reclaim_swap(struct swap_info_struct
*si
,
128 unsigned long offset
, unsigned long flags
)
130 swp_entry_t entry
= swp_entry(si
->type
, offset
);
134 page
= find_get_page(swap_address_space(entry
), offset
);
138 * When this function is called from scan_swap_map_slots() and it's
139 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
140 * here. We have to use trylock for avoiding deadlock. This is a special
141 * case and you should use try_to_free_swap() with explicit lock_page()
142 * in usual operations.
144 if (trylock_page(page
)) {
145 if ((flags
& TTRS_ANYWAY
) ||
146 ((flags
& TTRS_UNMAPPED
) && !page_mapped(page
)) ||
147 ((flags
& TTRS_FULL
) && mem_cgroup_swap_full(page
)))
148 ret
= try_to_free_swap(page
);
155 static inline struct swap_extent
*first_se(struct swap_info_struct
*sis
)
157 struct rb_node
*rb
= rb_first(&sis
->swap_extent_root
);
158 return rb_entry(rb
, struct swap_extent
, rb_node
);
161 static inline struct swap_extent
*next_se(struct swap_extent
*se
)
163 struct rb_node
*rb
= rb_next(&se
->rb_node
);
164 return rb
? rb_entry(rb
, struct swap_extent
, rb_node
) : NULL
;
168 * swapon tell device that all the old swap contents can be discarded,
169 * to allow the swap device to optimize its wear-levelling.
171 static int discard_swap(struct swap_info_struct
*si
)
173 struct swap_extent
*se
;
174 sector_t start_block
;
178 /* Do not discard the swap header page! */
180 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
181 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
183 err
= blkdev_issue_discard(si
->bdev
, start_block
,
184 nr_blocks
, GFP_KERNEL
, 0);
190 for (se
= next_se(se
); se
; se
= next_se(se
)) {
191 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
192 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
194 err
= blkdev_issue_discard(si
->bdev
, start_block
,
195 nr_blocks
, GFP_KERNEL
, 0);
201 return err
; /* That will often be -EOPNOTSUPP */
204 static struct swap_extent
*
205 offset_to_swap_extent(struct swap_info_struct
*sis
, unsigned long offset
)
207 struct swap_extent
*se
;
210 rb
= sis
->swap_extent_root
.rb_node
;
212 se
= rb_entry(rb
, struct swap_extent
, rb_node
);
213 if (offset
< se
->start_page
)
215 else if (offset
>= se
->start_page
+ se
->nr_pages
)
220 /* It *must* be present */
225 * swap allocation tell device that a cluster of swap can now be discarded,
226 * to allow the swap device to optimize its wear-levelling.
228 static void discard_swap_cluster(struct swap_info_struct
*si
,
229 pgoff_t start_page
, pgoff_t nr_pages
)
231 struct swap_extent
*se
= offset_to_swap_extent(si
, start_page
);
234 pgoff_t offset
= start_page
- se
->start_page
;
235 sector_t start_block
= se
->start_block
+ offset
;
236 sector_t nr_blocks
= se
->nr_pages
- offset
;
238 if (nr_blocks
> nr_pages
)
239 nr_blocks
= nr_pages
;
240 start_page
+= nr_blocks
;
241 nr_pages
-= nr_blocks
;
243 start_block
<<= PAGE_SHIFT
- 9;
244 nr_blocks
<<= PAGE_SHIFT
- 9;
245 if (blkdev_issue_discard(si
->bdev
, start_block
,
246 nr_blocks
, GFP_NOIO
, 0))
253 #ifdef CONFIG_THP_SWAP
254 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
256 #define swap_entry_size(size) (size)
258 #define SWAPFILE_CLUSTER 256
261 * Define swap_entry_size() as constant to let compiler to optimize
262 * out some code if !CONFIG_THP_SWAP
264 #define swap_entry_size(size) 1
266 #define LATENCY_LIMIT 256
268 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
274 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
279 static inline void cluster_set_count(struct swap_cluster_info
*info
,
285 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
286 unsigned int c
, unsigned int f
)
292 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
297 static inline void cluster_set_next(struct swap_cluster_info
*info
,
303 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
304 unsigned int n
, unsigned int f
)
310 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
312 return info
->flags
& CLUSTER_FLAG_FREE
;
315 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
317 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
320 static inline void cluster_set_null(struct swap_cluster_info
*info
)
322 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
326 static inline bool cluster_is_huge(struct swap_cluster_info
*info
)
328 if (IS_ENABLED(CONFIG_THP_SWAP
))
329 return info
->flags
& CLUSTER_FLAG_HUGE
;
333 static inline void cluster_clear_huge(struct swap_cluster_info
*info
)
335 info
->flags
&= ~CLUSTER_FLAG_HUGE
;
338 static inline struct swap_cluster_info
*lock_cluster(struct swap_info_struct
*si
,
339 unsigned long offset
)
341 struct swap_cluster_info
*ci
;
343 ci
= si
->cluster_info
;
345 ci
+= offset
/ SWAPFILE_CLUSTER
;
346 spin_lock(&ci
->lock
);
351 static inline void unlock_cluster(struct swap_cluster_info
*ci
)
354 spin_unlock(&ci
->lock
);
358 * Determine the locking method in use for this device. Return
359 * swap_cluster_info if SSD-style cluster-based locking is in place.
361 static inline struct swap_cluster_info
*lock_cluster_or_swap_info(
362 struct swap_info_struct
*si
, unsigned long offset
)
364 struct swap_cluster_info
*ci
;
366 /* Try to use fine-grained SSD-style locking if available: */
367 ci
= lock_cluster(si
, offset
);
368 /* Otherwise, fall back to traditional, coarse locking: */
370 spin_lock(&si
->lock
);
375 static inline void unlock_cluster_or_swap_info(struct swap_info_struct
*si
,
376 struct swap_cluster_info
*ci
)
381 spin_unlock(&si
->lock
);
384 static inline bool cluster_list_empty(struct swap_cluster_list
*list
)
386 return cluster_is_null(&list
->head
);
389 static inline unsigned int cluster_list_first(struct swap_cluster_list
*list
)
391 return cluster_next(&list
->head
);
394 static void cluster_list_init(struct swap_cluster_list
*list
)
396 cluster_set_null(&list
->head
);
397 cluster_set_null(&list
->tail
);
400 static void cluster_list_add_tail(struct swap_cluster_list
*list
,
401 struct swap_cluster_info
*ci
,
404 if (cluster_list_empty(list
)) {
405 cluster_set_next_flag(&list
->head
, idx
, 0);
406 cluster_set_next_flag(&list
->tail
, idx
, 0);
408 struct swap_cluster_info
*ci_tail
;
409 unsigned int tail
= cluster_next(&list
->tail
);
412 * Nested cluster lock, but both cluster locks are
413 * only acquired when we held swap_info_struct->lock
416 spin_lock_nested(&ci_tail
->lock
, SINGLE_DEPTH_NESTING
);
417 cluster_set_next(ci_tail
, idx
);
418 spin_unlock(&ci_tail
->lock
);
419 cluster_set_next_flag(&list
->tail
, idx
, 0);
423 static unsigned int cluster_list_del_first(struct swap_cluster_list
*list
,
424 struct swap_cluster_info
*ci
)
428 idx
= cluster_next(&list
->head
);
429 if (cluster_next(&list
->tail
) == idx
) {
430 cluster_set_null(&list
->head
);
431 cluster_set_null(&list
->tail
);
433 cluster_set_next_flag(&list
->head
,
434 cluster_next(&ci
[idx
]), 0);
439 /* Add a cluster to discard list and schedule it to do discard */
440 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
444 * If scan_swap_map() can't find a free cluster, it will check
445 * si->swap_map directly. To make sure the discarding cluster isn't
446 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
447 * will be cleared after discard
449 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
450 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
452 cluster_list_add_tail(&si
->discard_clusters
, si
->cluster_info
, idx
);
454 schedule_work(&si
->discard_work
);
457 static void __free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
459 struct swap_cluster_info
*ci
= si
->cluster_info
;
461 cluster_set_flag(ci
+ idx
, CLUSTER_FLAG_FREE
);
462 cluster_list_add_tail(&si
->free_clusters
, ci
, idx
);
466 * Doing discard actually. After a cluster discard is finished, the cluster
467 * will be added to free cluster list. caller should hold si->lock.
469 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
471 struct swap_cluster_info
*info
, *ci
;
474 info
= si
->cluster_info
;
476 while (!cluster_list_empty(&si
->discard_clusters
)) {
477 idx
= cluster_list_del_first(&si
->discard_clusters
, info
);
478 spin_unlock(&si
->lock
);
480 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
483 spin_lock(&si
->lock
);
484 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
485 __free_cluster(si
, idx
);
486 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
487 0, SWAPFILE_CLUSTER
);
492 static void swap_discard_work(struct work_struct
*work
)
494 struct swap_info_struct
*si
;
496 si
= container_of(work
, struct swap_info_struct
, discard_work
);
498 spin_lock(&si
->lock
);
499 swap_do_scheduled_discard(si
);
500 spin_unlock(&si
->lock
);
503 static void alloc_cluster(struct swap_info_struct
*si
, unsigned long idx
)
505 struct swap_cluster_info
*ci
= si
->cluster_info
;
507 VM_BUG_ON(cluster_list_first(&si
->free_clusters
) != idx
);
508 cluster_list_del_first(&si
->free_clusters
, ci
);
509 cluster_set_count_flag(ci
+ idx
, 0, 0);
512 static void free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
514 struct swap_cluster_info
*ci
= si
->cluster_info
+ idx
;
516 VM_BUG_ON(cluster_count(ci
) != 0);
518 * If the swap is discardable, prepare discard the cluster
519 * instead of free it immediately. The cluster will be freed
522 if ((si
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
523 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
524 swap_cluster_schedule_discard(si
, idx
);
528 __free_cluster(si
, idx
);
532 * The cluster corresponding to page_nr will be used. The cluster will be
533 * removed from free cluster list and its usage counter will be increased.
535 static void inc_cluster_info_page(struct swap_info_struct
*p
,
536 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
538 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
542 if (cluster_is_free(&cluster_info
[idx
]))
543 alloc_cluster(p
, idx
);
545 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
546 cluster_set_count(&cluster_info
[idx
],
547 cluster_count(&cluster_info
[idx
]) + 1);
551 * The cluster corresponding to page_nr decreases one usage. If the usage
552 * counter becomes 0, which means no page in the cluster is in using, we can
553 * optionally discard the cluster and add it to free cluster list.
555 static void dec_cluster_info_page(struct swap_info_struct
*p
,
556 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
558 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
563 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
564 cluster_set_count(&cluster_info
[idx
],
565 cluster_count(&cluster_info
[idx
]) - 1);
567 if (cluster_count(&cluster_info
[idx
]) == 0)
568 free_cluster(p
, idx
);
572 * It's possible scan_swap_map() uses a free cluster in the middle of free
573 * cluster list. Avoiding such abuse to avoid list corruption.
576 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
577 unsigned long offset
)
579 struct percpu_cluster
*percpu_cluster
;
582 offset
/= SWAPFILE_CLUSTER
;
583 conflict
= !cluster_list_empty(&si
->free_clusters
) &&
584 offset
!= cluster_list_first(&si
->free_clusters
) &&
585 cluster_is_free(&si
->cluster_info
[offset
]);
590 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
591 cluster_set_null(&percpu_cluster
->index
);
596 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
597 * might involve allocating a new cluster for current CPU too.
599 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
600 unsigned long *offset
, unsigned long *scan_base
)
602 struct percpu_cluster
*cluster
;
603 struct swap_cluster_info
*ci
;
604 unsigned long tmp
, max
;
607 cluster
= this_cpu_ptr(si
->percpu_cluster
);
608 if (cluster_is_null(&cluster
->index
)) {
609 if (!cluster_list_empty(&si
->free_clusters
)) {
610 cluster
->index
= si
->free_clusters
.head
;
611 cluster
->next
= cluster_next(&cluster
->index
) *
613 } else if (!cluster_list_empty(&si
->discard_clusters
)) {
615 * we don't have free cluster but have some clusters in
616 * discarding, do discard now and reclaim them, then
617 * reread cluster_next_cpu since we dropped si->lock
619 swap_do_scheduled_discard(si
);
620 *scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
621 *offset
= *scan_base
;
628 * Other CPUs can use our cluster if they can't find a free cluster,
629 * check if there is still free entry in the cluster
632 max
= min_t(unsigned long, si
->max
,
633 (cluster_next(&cluster
->index
) + 1) * SWAPFILE_CLUSTER
);
635 ci
= lock_cluster(si
, tmp
);
637 if (!si
->swap_map
[tmp
])
644 cluster_set_null(&cluster
->index
);
647 cluster
->next
= tmp
+ 1;
653 static void __del_from_avail_list(struct swap_info_struct
*p
)
658 plist_del(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
661 static void del_from_avail_list(struct swap_info_struct
*p
)
663 spin_lock(&swap_avail_lock
);
664 __del_from_avail_list(p
);
665 spin_unlock(&swap_avail_lock
);
668 static void swap_range_alloc(struct swap_info_struct
*si
, unsigned long offset
,
669 unsigned int nr_entries
)
671 unsigned int end
= offset
+ nr_entries
- 1;
673 if (offset
== si
->lowest_bit
)
674 si
->lowest_bit
+= nr_entries
;
675 if (end
== si
->highest_bit
)
676 si
->highest_bit
-= nr_entries
;
677 si
->inuse_pages
+= nr_entries
;
678 if (si
->inuse_pages
== si
->pages
) {
679 si
->lowest_bit
= si
->max
;
681 del_from_avail_list(si
);
685 static void add_to_avail_list(struct swap_info_struct
*p
)
689 spin_lock(&swap_avail_lock
);
691 WARN_ON(!plist_node_empty(&p
->avail_lists
[nid
]));
692 plist_add(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
694 spin_unlock(&swap_avail_lock
);
697 static void swap_range_free(struct swap_info_struct
*si
, unsigned long offset
,
698 unsigned int nr_entries
)
700 unsigned long end
= offset
+ nr_entries
- 1;
701 void (*swap_slot_free_notify
)(struct block_device
*, unsigned long);
703 if (offset
< si
->lowest_bit
)
704 si
->lowest_bit
= offset
;
705 if (end
> si
->highest_bit
) {
706 bool was_full
= !si
->highest_bit
;
708 si
->highest_bit
= end
;
709 if (was_full
&& (si
->flags
& SWP_WRITEOK
))
710 add_to_avail_list(si
);
712 atomic_long_add(nr_entries
, &nr_swap_pages
);
713 si
->inuse_pages
-= nr_entries
;
714 if (si
->flags
& SWP_BLKDEV
)
715 swap_slot_free_notify
=
716 si
->bdev
->bd_disk
->fops
->swap_slot_free_notify
;
718 swap_slot_free_notify
= NULL
;
719 while (offset
<= end
) {
720 frontswap_invalidate_page(si
->type
, offset
);
721 if (swap_slot_free_notify
)
722 swap_slot_free_notify(si
->bdev
, offset
);
727 static void set_cluster_next(struct swap_info_struct
*si
, unsigned long next
)
731 if (!(si
->flags
& SWP_SOLIDSTATE
)) {
732 si
->cluster_next
= next
;
736 prev
= this_cpu_read(*si
->cluster_next_cpu
);
738 * Cross the swap address space size aligned trunk, choose
739 * another trunk randomly to avoid lock contention on swap
740 * address space if possible.
742 if ((prev
>> SWAP_ADDRESS_SPACE_SHIFT
) !=
743 (next
>> SWAP_ADDRESS_SPACE_SHIFT
)) {
744 /* No free swap slots available */
745 if (si
->highest_bit
<= si
->lowest_bit
)
747 next
= si
->lowest_bit
+
748 prandom_u32_max(si
->highest_bit
- si
->lowest_bit
+ 1);
749 next
= ALIGN_DOWN(next
, SWAP_ADDRESS_SPACE_PAGES
);
750 next
= max_t(unsigned int, next
, si
->lowest_bit
);
752 this_cpu_write(*si
->cluster_next_cpu
, next
);
755 static int scan_swap_map_slots(struct swap_info_struct
*si
,
756 unsigned char usage
, int nr
,
759 struct swap_cluster_info
*ci
;
760 unsigned long offset
;
761 unsigned long scan_base
;
762 unsigned long last_in_cluster
= 0;
763 int latency_ration
= LATENCY_LIMIT
;
765 bool scanned_many
= false;
768 * We try to cluster swap pages by allocating them sequentially
769 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
770 * way, however, we resort to first-free allocation, starting
771 * a new cluster. This prevents us from scattering swap pages
772 * all over the entire swap partition, so that we reduce
773 * overall disk seek times between swap pages. -- sct
774 * But we do now try to find an empty cluster. -Andrea
775 * And we let swap pages go all over an SSD partition. Hugh
778 si
->flags
+= SWP_SCANNING
;
780 * Use percpu scan base for SSD to reduce lock contention on
781 * cluster and swap cache. For HDD, sequential access is more
784 if (si
->flags
& SWP_SOLIDSTATE
)
785 scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
787 scan_base
= si
->cluster_next
;
791 if (si
->cluster_info
) {
792 if (!scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
794 } else if (unlikely(!si
->cluster_nr
--)) {
795 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
796 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
800 spin_unlock(&si
->lock
);
803 * If seek is expensive, start searching for new cluster from
804 * start of partition, to minimize the span of allocated swap.
805 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
806 * case, just handled by scan_swap_map_try_ssd_cluster() above.
808 scan_base
= offset
= si
->lowest_bit
;
809 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
811 /* Locate the first empty (unaligned) cluster */
812 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
813 if (si
->swap_map
[offset
])
814 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
815 else if (offset
== last_in_cluster
) {
816 spin_lock(&si
->lock
);
817 offset
-= SWAPFILE_CLUSTER
- 1;
818 si
->cluster_next
= offset
;
819 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
822 if (unlikely(--latency_ration
< 0)) {
824 latency_ration
= LATENCY_LIMIT
;
829 spin_lock(&si
->lock
);
830 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
834 if (si
->cluster_info
) {
835 while (scan_swap_map_ssd_cluster_conflict(si
, offset
)) {
836 /* take a break if we already got some slots */
839 if (!scan_swap_map_try_ssd_cluster(si
, &offset
,
844 if (!(si
->flags
& SWP_WRITEOK
))
846 if (!si
->highest_bit
)
848 if (offset
> si
->highest_bit
)
849 scan_base
= offset
= si
->lowest_bit
;
851 ci
= lock_cluster(si
, offset
);
852 /* reuse swap entry of cache-only swap if not busy. */
853 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
856 spin_unlock(&si
->lock
);
857 swap_was_freed
= __try_to_reclaim_swap(si
, offset
, TTRS_ANYWAY
);
858 spin_lock(&si
->lock
);
859 /* entry was freed successfully, try to use this again */
862 goto scan
; /* check next one */
865 if (si
->swap_map
[offset
]) {
872 si
->swap_map
[offset
] = usage
;
873 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
876 swap_range_alloc(si
, offset
, 1);
877 slots
[n_ret
++] = swp_entry(si
->type
, offset
);
879 /* got enough slots or reach max slots? */
880 if ((n_ret
== nr
) || (offset
>= si
->highest_bit
))
883 /* search for next available slot */
885 /* time to take a break? */
886 if (unlikely(--latency_ration
< 0)) {
889 spin_unlock(&si
->lock
);
891 spin_lock(&si
->lock
);
892 latency_ration
= LATENCY_LIMIT
;
895 /* try to get more slots in cluster */
896 if (si
->cluster_info
) {
897 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
899 } else if (si
->cluster_nr
&& !si
->swap_map
[++offset
]) {
900 /* non-ssd case, still more slots in cluster? */
906 * Even if there's no free clusters available (fragmented),
907 * try to scan a little more quickly with lock held unless we
908 * have scanned too many slots already.
911 unsigned long scan_limit
;
913 if (offset
< scan_base
)
914 scan_limit
= scan_base
;
916 scan_limit
= si
->highest_bit
;
917 for (; offset
<= scan_limit
&& --latency_ration
> 0;
919 if (!si
->swap_map
[offset
])
925 set_cluster_next(si
, offset
+ 1);
926 si
->flags
-= SWP_SCANNING
;
930 spin_unlock(&si
->lock
);
931 while (++offset
<= si
->highest_bit
) {
932 if (!si
->swap_map
[offset
]) {
933 spin_lock(&si
->lock
);
936 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
937 spin_lock(&si
->lock
);
940 if (unlikely(--latency_ration
< 0)) {
942 latency_ration
= LATENCY_LIMIT
;
946 offset
= si
->lowest_bit
;
947 while (offset
< scan_base
) {
948 if (!si
->swap_map
[offset
]) {
949 spin_lock(&si
->lock
);
952 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
953 spin_lock(&si
->lock
);
956 if (unlikely(--latency_ration
< 0)) {
958 latency_ration
= LATENCY_LIMIT
;
963 spin_lock(&si
->lock
);
966 si
->flags
-= SWP_SCANNING
;
970 static int swap_alloc_cluster(struct swap_info_struct
*si
, swp_entry_t
*slot
)
973 struct swap_cluster_info
*ci
;
974 unsigned long offset
, i
;
978 * Should not even be attempting cluster allocations when huge
979 * page swap is disabled. Warn and fail the allocation.
981 if (!IS_ENABLED(CONFIG_THP_SWAP
)) {
986 if (cluster_list_empty(&si
->free_clusters
))
989 idx
= cluster_list_first(&si
->free_clusters
);
990 offset
= idx
* SWAPFILE_CLUSTER
;
991 ci
= lock_cluster(si
, offset
);
992 alloc_cluster(si
, idx
);
993 cluster_set_count_flag(ci
, SWAPFILE_CLUSTER
, CLUSTER_FLAG_HUGE
);
995 map
= si
->swap_map
+ offset
;
996 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++)
997 map
[i
] = SWAP_HAS_CACHE
;
999 swap_range_alloc(si
, offset
, SWAPFILE_CLUSTER
);
1000 *slot
= swp_entry(si
->type
, offset
);
1005 static void swap_free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
1007 unsigned long offset
= idx
* SWAPFILE_CLUSTER
;
1008 struct swap_cluster_info
*ci
;
1010 ci
= lock_cluster(si
, offset
);
1011 memset(si
->swap_map
+ offset
, 0, SWAPFILE_CLUSTER
);
1012 cluster_set_count_flag(ci
, 0, 0);
1013 free_cluster(si
, idx
);
1015 swap_range_free(si
, offset
, SWAPFILE_CLUSTER
);
1018 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
1019 unsigned char usage
)
1024 n_ret
= scan_swap_map_slots(si
, usage
, 1, &entry
);
1027 return swp_offset(entry
);
1033 int get_swap_pages(int n_goal
, swp_entry_t swp_entries
[], int entry_size
)
1035 unsigned long size
= swap_entry_size(entry_size
);
1036 struct swap_info_struct
*si
, *next
;
1041 /* Only single cluster request supported */
1042 WARN_ON_ONCE(n_goal
> 1 && size
== SWAPFILE_CLUSTER
);
1044 avail_pgs
= atomic_long_read(&nr_swap_pages
) / size
;
1048 n_goal
= min3((long)n_goal
, (long)SWAP_BATCH
, avail_pgs
);
1050 atomic_long_sub(n_goal
* size
, &nr_swap_pages
);
1052 spin_lock(&swap_avail_lock
);
1055 node
= numa_node_id();
1056 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[node
], avail_lists
[node
]) {
1057 /* requeue si to after same-priority siblings */
1058 plist_requeue(&si
->avail_lists
[node
], &swap_avail_heads
[node
]);
1059 spin_unlock(&swap_avail_lock
);
1060 spin_lock(&si
->lock
);
1061 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
1062 spin_lock(&swap_avail_lock
);
1063 if (plist_node_empty(&si
->avail_lists
[node
])) {
1064 spin_unlock(&si
->lock
);
1067 WARN(!si
->highest_bit
,
1068 "swap_info %d in list but !highest_bit\n",
1070 WARN(!(si
->flags
& SWP_WRITEOK
),
1071 "swap_info %d in list but !SWP_WRITEOK\n",
1073 __del_from_avail_list(si
);
1074 spin_unlock(&si
->lock
);
1077 if (size
== SWAPFILE_CLUSTER
) {
1078 if (!(si
->flags
& SWP_FS
))
1079 n_ret
= swap_alloc_cluster(si
, swp_entries
);
1081 n_ret
= scan_swap_map_slots(si
, SWAP_HAS_CACHE
,
1082 n_goal
, swp_entries
);
1083 spin_unlock(&si
->lock
);
1084 if (n_ret
|| size
== SWAPFILE_CLUSTER
)
1086 pr_debug("scan_swap_map of si %d failed to find offset\n",
1089 spin_lock(&swap_avail_lock
);
1092 * if we got here, it's likely that si was almost full before,
1093 * and since scan_swap_map() can drop the si->lock, multiple
1094 * callers probably all tried to get a page from the same si
1095 * and it filled up before we could get one; or, the si filled
1096 * up between us dropping swap_avail_lock and taking si->lock.
1097 * Since we dropped the swap_avail_lock, the swap_avail_head
1098 * list may have been modified; so if next is still in the
1099 * swap_avail_head list then try it, otherwise start over
1100 * if we have not gotten any slots.
1102 if (plist_node_empty(&next
->avail_lists
[node
]))
1106 spin_unlock(&swap_avail_lock
);
1110 atomic_long_add((long)(n_goal
- n_ret
) * size
,
1116 /* The only caller of this function is now suspend routine */
1117 swp_entry_t
get_swap_page_of_type(int type
)
1119 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1125 spin_lock(&si
->lock
);
1126 if (si
->flags
& SWP_WRITEOK
) {
1127 atomic_long_dec(&nr_swap_pages
);
1128 /* This is called for allocating swap entry, not cache */
1129 offset
= scan_swap_map(si
, 1);
1131 spin_unlock(&si
->lock
);
1132 return swp_entry(type
, offset
);
1134 atomic_long_inc(&nr_swap_pages
);
1136 spin_unlock(&si
->lock
);
1138 return (swp_entry_t
) {0};
1141 static struct swap_info_struct
*__swap_info_get(swp_entry_t entry
)
1143 struct swap_info_struct
*p
;
1144 unsigned long offset
;
1148 p
= swp_swap_info(entry
);
1151 if (!(p
->flags
& SWP_USED
))
1153 offset
= swp_offset(entry
);
1154 if (offset
>= p
->max
)
1159 pr_err("swap_info_get: %s%08lx\n", Bad_offset
, entry
.val
);
1162 pr_err("swap_info_get: %s%08lx\n", Unused_file
, entry
.val
);
1165 pr_err("swap_info_get: %s%08lx\n", Bad_file
, entry
.val
);
1170 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
1172 struct swap_info_struct
*p
;
1174 p
= __swap_info_get(entry
);
1177 if (!p
->swap_map
[swp_offset(entry
)])
1182 pr_err("swap_info_get: %s%08lx\n", Unused_offset
, entry
.val
);
1188 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
1190 struct swap_info_struct
*p
;
1192 p
= _swap_info_get(entry
);
1194 spin_lock(&p
->lock
);
1198 static struct swap_info_struct
*swap_info_get_cont(swp_entry_t entry
,
1199 struct swap_info_struct
*q
)
1201 struct swap_info_struct
*p
;
1203 p
= _swap_info_get(entry
);
1207 spin_unlock(&q
->lock
);
1209 spin_lock(&p
->lock
);
1214 static unsigned char __swap_entry_free_locked(struct swap_info_struct
*p
,
1215 unsigned long offset
,
1216 unsigned char usage
)
1218 unsigned char count
;
1219 unsigned char has_cache
;
1221 count
= p
->swap_map
[offset
];
1223 has_cache
= count
& SWAP_HAS_CACHE
;
1224 count
&= ~SWAP_HAS_CACHE
;
1226 if (usage
== SWAP_HAS_CACHE
) {
1227 VM_BUG_ON(!has_cache
);
1229 } else if (count
== SWAP_MAP_SHMEM
) {
1231 * Or we could insist on shmem.c using a special
1232 * swap_shmem_free() and free_shmem_swap_and_cache()...
1235 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
1236 if (count
== COUNT_CONTINUED
) {
1237 if (swap_count_continued(p
, offset
, count
))
1238 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
1240 count
= SWAP_MAP_MAX
;
1245 usage
= count
| has_cache
;
1246 p
->swap_map
[offset
] = usage
? : SWAP_HAS_CACHE
;
1252 * Check whether swap entry is valid in the swap device. If so,
1253 * return pointer to swap_info_struct, and keep the swap entry valid
1254 * via preventing the swap device from being swapoff, until
1255 * put_swap_device() is called. Otherwise return NULL.
1257 * The entirety of the RCU read critical section must come before the
1258 * return from or after the call to synchronize_rcu() in
1259 * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is
1260 * true, the si->map, si->cluster_info, etc. must be valid in the
1263 * Notice that swapoff or swapoff+swapon can still happen before the
1264 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1265 * in put_swap_device() if there isn't any other way to prevent
1266 * swapoff, such as page lock, page table lock, etc. The caller must
1267 * be prepared for that. For example, the following situation is
1272 * ... swapoff+swapon
1273 * __read_swap_cache_async()
1274 * swapcache_prepare()
1275 * __swap_duplicate()
1277 * // verify PTE not changed
1279 * In __swap_duplicate(), the swap_map need to be checked before
1280 * changing partly because the specified swap entry may be for another
1281 * swap device which has been swapoff. And in do_swap_page(), after
1282 * the page is read from the swap device, the PTE is verified not
1283 * changed with the page table locked to check whether the swap device
1284 * has been swapoff or swapoff+swapon.
1286 struct swap_info_struct
*get_swap_device(swp_entry_t entry
)
1288 struct swap_info_struct
*si
;
1289 unsigned long offset
;
1293 si
= swp_swap_info(entry
);
1298 if (!(si
->flags
& SWP_VALID
))
1300 offset
= swp_offset(entry
);
1301 if (offset
>= si
->max
)
1306 pr_err("%s: %s%08lx\n", __func__
, Bad_file
, entry
.val
);
1314 static unsigned char __swap_entry_free(struct swap_info_struct
*p
,
1317 struct swap_cluster_info
*ci
;
1318 unsigned long offset
= swp_offset(entry
);
1319 unsigned char usage
;
1321 ci
= lock_cluster_or_swap_info(p
, offset
);
1322 usage
= __swap_entry_free_locked(p
, offset
, 1);
1323 unlock_cluster_or_swap_info(p
, ci
);
1325 free_swap_slot(entry
);
1330 static void swap_entry_free(struct swap_info_struct
*p
, swp_entry_t entry
)
1332 struct swap_cluster_info
*ci
;
1333 unsigned long offset
= swp_offset(entry
);
1334 unsigned char count
;
1336 ci
= lock_cluster(p
, offset
);
1337 count
= p
->swap_map
[offset
];
1338 VM_BUG_ON(count
!= SWAP_HAS_CACHE
);
1339 p
->swap_map
[offset
] = 0;
1340 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
1343 mem_cgroup_uncharge_swap(entry
, 1);
1344 swap_range_free(p
, offset
, 1);
1348 * Caller has made sure that the swap device corresponding to entry
1349 * is still around or has not been recycled.
1351 void swap_free(swp_entry_t entry
)
1353 struct swap_info_struct
*p
;
1355 p
= _swap_info_get(entry
);
1357 __swap_entry_free(p
, entry
);
1361 * Called after dropping swapcache to decrease refcnt to swap entries.
1363 void put_swap_page(struct page
*page
, swp_entry_t entry
)
1365 unsigned long offset
= swp_offset(entry
);
1366 unsigned long idx
= offset
/ SWAPFILE_CLUSTER
;
1367 struct swap_cluster_info
*ci
;
1368 struct swap_info_struct
*si
;
1370 unsigned int i
, free_entries
= 0;
1372 int size
= swap_entry_size(hpage_nr_pages(page
));
1374 si
= _swap_info_get(entry
);
1378 ci
= lock_cluster_or_swap_info(si
, offset
);
1379 if (size
== SWAPFILE_CLUSTER
) {
1380 VM_BUG_ON(!cluster_is_huge(ci
));
1381 map
= si
->swap_map
+ offset
;
1382 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1384 VM_BUG_ON(!(val
& SWAP_HAS_CACHE
));
1385 if (val
== SWAP_HAS_CACHE
)
1388 cluster_clear_huge(ci
);
1389 if (free_entries
== SWAPFILE_CLUSTER
) {
1390 unlock_cluster_or_swap_info(si
, ci
);
1391 spin_lock(&si
->lock
);
1392 mem_cgroup_uncharge_swap(entry
, SWAPFILE_CLUSTER
);
1393 swap_free_cluster(si
, idx
);
1394 spin_unlock(&si
->lock
);
1398 for (i
= 0; i
< size
; i
++, entry
.val
++) {
1399 if (!__swap_entry_free_locked(si
, offset
+ i
, SWAP_HAS_CACHE
)) {
1400 unlock_cluster_or_swap_info(si
, ci
);
1401 free_swap_slot(entry
);
1404 lock_cluster_or_swap_info(si
, offset
);
1407 unlock_cluster_or_swap_info(si
, ci
);
1410 #ifdef CONFIG_THP_SWAP
1411 int split_swap_cluster(swp_entry_t entry
)
1413 struct swap_info_struct
*si
;
1414 struct swap_cluster_info
*ci
;
1415 unsigned long offset
= swp_offset(entry
);
1417 si
= _swap_info_get(entry
);
1420 ci
= lock_cluster(si
, offset
);
1421 cluster_clear_huge(ci
);
1427 static int swp_entry_cmp(const void *ent1
, const void *ent2
)
1429 const swp_entry_t
*e1
= ent1
, *e2
= ent2
;
1431 return (int)swp_type(*e1
) - (int)swp_type(*e2
);
1434 void swapcache_free_entries(swp_entry_t
*entries
, int n
)
1436 struct swap_info_struct
*p
, *prev
;
1446 * Sort swap entries by swap device, so each lock is only taken once.
1447 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1448 * so low that it isn't necessary to optimize further.
1450 if (nr_swapfiles
> 1)
1451 sort(entries
, n
, sizeof(entries
[0]), swp_entry_cmp
, NULL
);
1452 for (i
= 0; i
< n
; ++i
) {
1453 p
= swap_info_get_cont(entries
[i
], prev
);
1455 swap_entry_free(p
, entries
[i
]);
1459 spin_unlock(&p
->lock
);
1463 * How many references to page are currently swapped out?
1464 * This does not give an exact answer when swap count is continued,
1465 * but does include the high COUNT_CONTINUED flag to allow for that.
1467 int page_swapcount(struct page
*page
)
1470 struct swap_info_struct
*p
;
1471 struct swap_cluster_info
*ci
;
1473 unsigned long offset
;
1475 entry
.val
= page_private(page
);
1476 p
= _swap_info_get(entry
);
1478 offset
= swp_offset(entry
);
1479 ci
= lock_cluster_or_swap_info(p
, offset
);
1480 count
= swap_count(p
->swap_map
[offset
]);
1481 unlock_cluster_or_swap_info(p
, ci
);
1486 int __swap_count(swp_entry_t entry
)
1488 struct swap_info_struct
*si
;
1489 pgoff_t offset
= swp_offset(entry
);
1492 si
= get_swap_device(entry
);
1494 count
= swap_count(si
->swap_map
[offset
]);
1495 put_swap_device(si
);
1500 static int swap_swapcount(struct swap_info_struct
*si
, swp_entry_t entry
)
1503 pgoff_t offset
= swp_offset(entry
);
1504 struct swap_cluster_info
*ci
;
1506 ci
= lock_cluster_or_swap_info(si
, offset
);
1507 count
= swap_count(si
->swap_map
[offset
]);
1508 unlock_cluster_or_swap_info(si
, ci
);
1513 * How many references to @entry are currently swapped out?
1514 * This does not give an exact answer when swap count is continued,
1515 * but does include the high COUNT_CONTINUED flag to allow for that.
1517 int __swp_swapcount(swp_entry_t entry
)
1520 struct swap_info_struct
*si
;
1522 si
= get_swap_device(entry
);
1524 count
= swap_swapcount(si
, entry
);
1525 put_swap_device(si
);
1531 * How many references to @entry are currently swapped out?
1532 * This considers COUNT_CONTINUED so it returns exact answer.
1534 int swp_swapcount(swp_entry_t entry
)
1536 int count
, tmp_count
, n
;
1537 struct swap_info_struct
*p
;
1538 struct swap_cluster_info
*ci
;
1543 p
= _swap_info_get(entry
);
1547 offset
= swp_offset(entry
);
1549 ci
= lock_cluster_or_swap_info(p
, offset
);
1551 count
= swap_count(p
->swap_map
[offset
]);
1552 if (!(count
& COUNT_CONTINUED
))
1555 count
&= ~COUNT_CONTINUED
;
1556 n
= SWAP_MAP_MAX
+ 1;
1558 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1559 offset
&= ~PAGE_MASK
;
1560 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1563 page
= list_next_entry(page
, lru
);
1564 map
= kmap_atomic(page
);
1565 tmp_count
= map
[offset
];
1568 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1569 n
*= (SWAP_CONT_MAX
+ 1);
1570 } while (tmp_count
& COUNT_CONTINUED
);
1572 unlock_cluster_or_swap_info(p
, ci
);
1576 static bool swap_page_trans_huge_swapped(struct swap_info_struct
*si
,
1579 struct swap_cluster_info
*ci
;
1580 unsigned char *map
= si
->swap_map
;
1581 unsigned long roffset
= swp_offset(entry
);
1582 unsigned long offset
= round_down(roffset
, SWAPFILE_CLUSTER
);
1586 ci
= lock_cluster_or_swap_info(si
, offset
);
1587 if (!ci
|| !cluster_is_huge(ci
)) {
1588 if (swap_count(map
[roffset
]))
1592 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1593 if (swap_count(map
[offset
+ i
])) {
1599 unlock_cluster_or_swap_info(si
, ci
);
1603 static bool page_swapped(struct page
*page
)
1606 struct swap_info_struct
*si
;
1608 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
)))
1609 return page_swapcount(page
) != 0;
1611 page
= compound_head(page
);
1612 entry
.val
= page_private(page
);
1613 si
= _swap_info_get(entry
);
1615 return swap_page_trans_huge_swapped(si
, entry
);
1619 static int page_trans_huge_map_swapcount(struct page
*page
, int *total_mapcount
,
1620 int *total_swapcount
)
1622 int i
, map_swapcount
, _total_mapcount
, _total_swapcount
;
1623 unsigned long offset
= 0;
1624 struct swap_info_struct
*si
;
1625 struct swap_cluster_info
*ci
= NULL
;
1626 unsigned char *map
= NULL
;
1627 int mapcount
, swapcount
= 0;
1629 /* hugetlbfs shouldn't call it */
1630 VM_BUG_ON_PAGE(PageHuge(page
), page
);
1632 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
))) {
1633 mapcount
= page_trans_huge_mapcount(page
, total_mapcount
);
1634 if (PageSwapCache(page
))
1635 swapcount
= page_swapcount(page
);
1636 if (total_swapcount
)
1637 *total_swapcount
= swapcount
;
1638 return mapcount
+ swapcount
;
1641 page
= compound_head(page
);
1643 _total_mapcount
= _total_swapcount
= map_swapcount
= 0;
1644 if (PageSwapCache(page
)) {
1647 entry
.val
= page_private(page
);
1648 si
= _swap_info_get(entry
);
1651 offset
= swp_offset(entry
);
1655 ci
= lock_cluster(si
, offset
);
1656 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
1657 mapcount
= atomic_read(&page
[i
]._mapcount
) + 1;
1658 _total_mapcount
+= mapcount
;
1660 swapcount
= swap_count(map
[offset
+ i
]);
1661 _total_swapcount
+= swapcount
;
1663 map_swapcount
= max(map_swapcount
, mapcount
+ swapcount
);
1666 if (PageDoubleMap(page
)) {
1668 _total_mapcount
-= HPAGE_PMD_NR
;
1670 mapcount
= compound_mapcount(page
);
1671 map_swapcount
+= mapcount
;
1672 _total_mapcount
+= mapcount
;
1674 *total_mapcount
= _total_mapcount
;
1675 if (total_swapcount
)
1676 *total_swapcount
= _total_swapcount
;
1678 return map_swapcount
;
1682 * We can write to an anon page without COW if there are no other references
1683 * to it. And as a side-effect, free up its swap: because the old content
1684 * on disk will never be read, and seeking back there to write new content
1685 * later would only waste time away from clustering.
1687 * NOTE: total_map_swapcount should not be relied upon by the caller if
1688 * reuse_swap_page() returns false, but it may be always overwritten
1689 * (see the other implementation for CONFIG_SWAP=n).
1691 bool reuse_swap_page(struct page
*page
, int *total_map_swapcount
)
1693 int count
, total_mapcount
, total_swapcount
;
1695 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1696 if (unlikely(PageKsm(page
)))
1698 count
= page_trans_huge_map_swapcount(page
, &total_mapcount
,
1700 if (total_map_swapcount
)
1701 *total_map_swapcount
= total_mapcount
+ total_swapcount
;
1702 if (count
== 1 && PageSwapCache(page
) &&
1703 (likely(!PageTransCompound(page
)) ||
1704 /* The remaining swap count will be freed soon */
1705 total_swapcount
== page_swapcount(page
))) {
1706 if (!PageWriteback(page
)) {
1707 page
= compound_head(page
);
1708 delete_from_swap_cache(page
);
1712 struct swap_info_struct
*p
;
1714 entry
.val
= page_private(page
);
1715 p
= swap_info_get(entry
);
1716 if (p
->flags
& SWP_STABLE_WRITES
) {
1717 spin_unlock(&p
->lock
);
1720 spin_unlock(&p
->lock
);
1728 * If swap is getting full, or if there are no more mappings of this page,
1729 * then try_to_free_swap is called to free its swap space.
1731 int try_to_free_swap(struct page
*page
)
1733 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1735 if (!PageSwapCache(page
))
1737 if (PageWriteback(page
))
1739 if (page_swapped(page
))
1743 * Once hibernation has begun to create its image of memory,
1744 * there's a danger that one of the calls to try_to_free_swap()
1745 * - most probably a call from __try_to_reclaim_swap() while
1746 * hibernation is allocating its own swap pages for the image,
1747 * but conceivably even a call from memory reclaim - will free
1748 * the swap from a page which has already been recorded in the
1749 * image as a clean swapcache page, and then reuse its swap for
1750 * another page of the image. On waking from hibernation, the
1751 * original page might be freed under memory pressure, then
1752 * later read back in from swap, now with the wrong data.
1754 * Hibernation suspends storage while it is writing the image
1755 * to disk so check that here.
1757 if (pm_suspended_storage())
1760 page
= compound_head(page
);
1761 delete_from_swap_cache(page
);
1767 * Free the swap entry like above, but also try to
1768 * free the page cache entry if it is the last user.
1770 int free_swap_and_cache(swp_entry_t entry
)
1772 struct swap_info_struct
*p
;
1773 unsigned char count
;
1775 if (non_swap_entry(entry
))
1778 p
= _swap_info_get(entry
);
1780 count
= __swap_entry_free(p
, entry
);
1781 if (count
== SWAP_HAS_CACHE
&&
1782 !swap_page_trans_huge_swapped(p
, entry
))
1783 __try_to_reclaim_swap(p
, swp_offset(entry
),
1784 TTRS_UNMAPPED
| TTRS_FULL
);
1789 #ifdef CONFIG_HIBERNATION
1791 * Find the swap type that corresponds to given device (if any).
1793 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1794 * from 0, in which the swap header is expected to be located.
1796 * This is needed for the suspend to disk (aka swsusp).
1798 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
1800 struct block_device
*bdev
= NULL
;
1804 bdev
= bdget(device
);
1806 spin_lock(&swap_lock
);
1807 for (type
= 0; type
< nr_swapfiles
; type
++) {
1808 struct swap_info_struct
*sis
= swap_info
[type
];
1810 if (!(sis
->flags
& SWP_WRITEOK
))
1815 *bdev_p
= bdgrab(sis
->bdev
);
1817 spin_unlock(&swap_lock
);
1820 if (bdev
== sis
->bdev
) {
1821 struct swap_extent
*se
= first_se(sis
);
1823 if (se
->start_block
== offset
) {
1825 *bdev_p
= bdgrab(sis
->bdev
);
1827 spin_unlock(&swap_lock
);
1833 spin_unlock(&swap_lock
);
1841 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1842 * corresponding to given index in swap_info (swap type).
1844 sector_t
swapdev_block(int type
, pgoff_t offset
)
1846 struct block_device
*bdev
;
1847 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1849 if (!si
|| !(si
->flags
& SWP_WRITEOK
))
1851 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
1855 * Return either the total number of swap pages of given type, or the number
1856 * of free pages of that type (depending on @free)
1858 * This is needed for software suspend
1860 unsigned int count_swap_pages(int type
, int free
)
1864 spin_lock(&swap_lock
);
1865 if ((unsigned int)type
< nr_swapfiles
) {
1866 struct swap_info_struct
*sis
= swap_info
[type
];
1868 spin_lock(&sis
->lock
);
1869 if (sis
->flags
& SWP_WRITEOK
) {
1872 n
-= sis
->inuse_pages
;
1874 spin_unlock(&sis
->lock
);
1876 spin_unlock(&swap_lock
);
1879 #endif /* CONFIG_HIBERNATION */
1881 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1883 return pte_same(pte_swp_clear_soft_dirty(pte
), swp_pte
);
1887 * No need to decide whether this PTE shares the swap entry with others,
1888 * just let do_wp_page work it out if a write is requested later - to
1889 * force COW, vm_page_prot omits write permission from any private vma.
1891 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1892 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1894 struct page
*swapcache
;
1900 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1901 if (unlikely(!page
))
1904 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1905 if (unlikely(!pte_same_as_swp(*pte
, swp_entry_to_pte(entry
)))) {
1910 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1911 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1913 set_pte_at(vma
->vm_mm
, addr
, pte
,
1914 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1915 if (page
== swapcache
) {
1916 page_add_anon_rmap(page
, vma
, addr
, false);
1917 } else { /* ksm created a completely new copy */
1918 page_add_new_anon_rmap(page
, vma
, addr
, false);
1919 lru_cache_add_active_or_unevictable(page
, vma
);
1923 * Move the page to the active list so it is not
1924 * immediately swapped out again after swapon.
1926 activate_page(page
);
1928 pte_unmap_unlock(pte
, ptl
);
1929 if (page
!= swapcache
) {
1936 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1937 unsigned long addr
, unsigned long end
,
1938 unsigned int type
, bool frontswap
,
1939 unsigned long *fs_pages_to_unuse
)
1944 struct swap_info_struct
*si
;
1945 unsigned long offset
;
1947 volatile unsigned char *swap_map
;
1949 si
= swap_info
[type
];
1950 pte
= pte_offset_map(pmd
, addr
);
1952 struct vm_fault vmf
;
1954 if (!is_swap_pte(*pte
))
1957 entry
= pte_to_swp_entry(*pte
);
1958 if (swp_type(entry
) != type
)
1961 offset
= swp_offset(entry
);
1962 if (frontswap
&& !frontswap_test(si
, offset
))
1966 swap_map
= &si
->swap_map
[offset
];
1967 page
= lookup_swap_cache(entry
, vma
, addr
);
1972 page
= swapin_readahead(entry
, GFP_HIGHUSER_MOVABLE
,
1976 if (*swap_map
== 0 || *swap_map
== SWAP_MAP_BAD
)
1982 wait_on_page_writeback(page
);
1983 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1990 try_to_free_swap(page
);
1994 if (*fs_pages_to_unuse
&& !--(*fs_pages_to_unuse
)) {
1995 ret
= FRONTSWAP_PAGES_UNUSED
;
1999 pte
= pte_offset_map(pmd
, addr
);
2000 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
2008 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
2009 unsigned long addr
, unsigned long end
,
2010 unsigned int type
, bool frontswap
,
2011 unsigned long *fs_pages_to_unuse
)
2017 pmd
= pmd_offset(pud
, addr
);
2020 next
= pmd_addr_end(addr
, end
);
2021 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
2023 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, type
,
2024 frontswap
, fs_pages_to_unuse
);
2027 } while (pmd
++, addr
= next
, addr
!= end
);
2031 static inline int unuse_pud_range(struct vm_area_struct
*vma
, p4d_t
*p4d
,
2032 unsigned long addr
, unsigned long end
,
2033 unsigned int type
, bool frontswap
,
2034 unsigned long *fs_pages_to_unuse
)
2040 pud
= pud_offset(p4d
, addr
);
2042 next
= pud_addr_end(addr
, end
);
2043 if (pud_none_or_clear_bad(pud
))
2045 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, type
,
2046 frontswap
, fs_pages_to_unuse
);
2049 } while (pud
++, addr
= next
, addr
!= end
);
2053 static inline int unuse_p4d_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
2054 unsigned long addr
, unsigned long end
,
2055 unsigned int type
, bool frontswap
,
2056 unsigned long *fs_pages_to_unuse
)
2062 p4d
= p4d_offset(pgd
, addr
);
2064 next
= p4d_addr_end(addr
, end
);
2065 if (p4d_none_or_clear_bad(p4d
))
2067 ret
= unuse_pud_range(vma
, p4d
, addr
, next
, type
,
2068 frontswap
, fs_pages_to_unuse
);
2071 } while (p4d
++, addr
= next
, addr
!= end
);
2075 static int unuse_vma(struct vm_area_struct
*vma
, unsigned int type
,
2076 bool frontswap
, unsigned long *fs_pages_to_unuse
)
2079 unsigned long addr
, end
, next
;
2082 addr
= vma
->vm_start
;
2085 pgd
= pgd_offset(vma
->vm_mm
, addr
);
2087 next
= pgd_addr_end(addr
, end
);
2088 if (pgd_none_or_clear_bad(pgd
))
2090 ret
= unuse_p4d_range(vma
, pgd
, addr
, next
, type
,
2091 frontswap
, fs_pages_to_unuse
);
2094 } while (pgd
++, addr
= next
, addr
!= end
);
2098 static int unuse_mm(struct mm_struct
*mm
, unsigned int type
,
2099 bool frontswap
, unsigned long *fs_pages_to_unuse
)
2101 struct vm_area_struct
*vma
;
2104 down_read(&mm
->mmap_sem
);
2105 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2106 if (vma
->anon_vma
) {
2107 ret
= unuse_vma(vma
, type
, frontswap
,
2114 up_read(&mm
->mmap_sem
);
2119 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2120 * from current position to next entry still in use. Return 0
2121 * if there are no inuse entries after prev till end of the map.
2123 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
2124 unsigned int prev
, bool frontswap
)
2127 unsigned char count
;
2130 * No need for swap_lock here: we're just looking
2131 * for whether an entry is in use, not modifying it; false
2132 * hits are okay, and sys_swapoff() has already prevented new
2133 * allocations from this area (while holding swap_lock).
2135 for (i
= prev
+ 1; i
< si
->max
; i
++) {
2136 count
= READ_ONCE(si
->swap_map
[i
]);
2137 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
2138 if (!frontswap
|| frontswap_test(si
, i
))
2140 if ((i
% LATENCY_LIMIT
) == 0)
2151 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2152 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2154 int try_to_unuse(unsigned int type
, bool frontswap
,
2155 unsigned long pages_to_unuse
)
2157 struct mm_struct
*prev_mm
;
2158 struct mm_struct
*mm
;
2159 struct list_head
*p
;
2161 struct swap_info_struct
*si
= swap_info
[type
];
2166 if (!READ_ONCE(si
->inuse_pages
))
2173 retval
= shmem_unuse(type
, frontswap
, &pages_to_unuse
);
2180 spin_lock(&mmlist_lock
);
2181 p
= &init_mm
.mmlist
;
2182 while (READ_ONCE(si
->inuse_pages
) &&
2183 !signal_pending(current
) &&
2184 (p
= p
->next
) != &init_mm
.mmlist
) {
2186 mm
= list_entry(p
, struct mm_struct
, mmlist
);
2187 if (!mmget_not_zero(mm
))
2189 spin_unlock(&mmlist_lock
);
2192 retval
= unuse_mm(mm
, type
, frontswap
, &pages_to_unuse
);
2200 * Make sure that we aren't completely killing
2201 * interactive performance.
2204 spin_lock(&mmlist_lock
);
2206 spin_unlock(&mmlist_lock
);
2211 while (READ_ONCE(si
->inuse_pages
) &&
2212 !signal_pending(current
) &&
2213 (i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
2215 entry
= swp_entry(type
, i
);
2216 page
= find_get_page(swap_address_space(entry
), i
);
2221 * It is conceivable that a racing task removed this page from
2222 * swap cache just before we acquired the page lock. The page
2223 * might even be back in swap cache on another swap area. But
2224 * that is okay, try_to_free_swap() only removes stale pages.
2227 wait_on_page_writeback(page
);
2228 try_to_free_swap(page
);
2233 * For frontswap, we just need to unuse pages_to_unuse, if
2234 * it was specified. Need not check frontswap again here as
2235 * we already zeroed out pages_to_unuse if not frontswap.
2237 if (pages_to_unuse
&& --pages_to_unuse
== 0)
2242 * Lets check again to see if there are still swap entries in the map.
2243 * If yes, we would need to do retry the unuse logic again.
2244 * Under global memory pressure, swap entries can be reinserted back
2245 * into process space after the mmlist loop above passes over them.
2247 * Limit the number of retries? No: when mmget_not_zero() above fails,
2248 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2249 * at its own independent pace; and even shmem_writepage() could have
2250 * been preempted after get_swap_page(), temporarily hiding that swap.
2251 * It's easy and robust (though cpu-intensive) just to keep retrying.
2253 if (READ_ONCE(si
->inuse_pages
)) {
2254 if (!signal_pending(current
))
2259 return (retval
== FRONTSWAP_PAGES_UNUSED
) ? 0 : retval
;
2263 * After a successful try_to_unuse, if no swap is now in use, we know
2264 * we can empty the mmlist. swap_lock must be held on entry and exit.
2265 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2266 * added to the mmlist just after page_duplicate - before would be racy.
2268 static void drain_mmlist(void)
2270 struct list_head
*p
, *next
;
2273 for (type
= 0; type
< nr_swapfiles
; type
++)
2274 if (swap_info
[type
]->inuse_pages
)
2276 spin_lock(&mmlist_lock
);
2277 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
2279 spin_unlock(&mmlist_lock
);
2283 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2284 * corresponds to page offset for the specified swap entry.
2285 * Note that the type of this function is sector_t, but it returns page offset
2286 * into the bdev, not sector offset.
2288 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
2290 struct swap_info_struct
*sis
;
2291 struct swap_extent
*se
;
2294 sis
= swp_swap_info(entry
);
2297 offset
= swp_offset(entry
);
2298 se
= offset_to_swap_extent(sis
, offset
);
2299 return se
->start_block
+ (offset
- se
->start_page
);
2303 * Returns the page offset into bdev for the specified page's swap entry.
2305 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
2308 entry
.val
= page_private(page
);
2309 return map_swap_entry(entry
, bdev
);
2313 * Free all of a swapdev's extent information
2315 static void destroy_swap_extents(struct swap_info_struct
*sis
)
2317 while (!RB_EMPTY_ROOT(&sis
->swap_extent_root
)) {
2318 struct rb_node
*rb
= sis
->swap_extent_root
.rb_node
;
2319 struct swap_extent
*se
= rb_entry(rb
, struct swap_extent
, rb_node
);
2321 rb_erase(rb
, &sis
->swap_extent_root
);
2325 if (sis
->flags
& SWP_ACTIVATED
) {
2326 struct file
*swap_file
= sis
->swap_file
;
2327 struct address_space
*mapping
= swap_file
->f_mapping
;
2329 sis
->flags
&= ~SWP_ACTIVATED
;
2330 if (mapping
->a_ops
->swap_deactivate
)
2331 mapping
->a_ops
->swap_deactivate(swap_file
);
2336 * Add a block range (and the corresponding page range) into this swapdev's
2339 * This function rather assumes that it is called in ascending page order.
2342 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
2343 unsigned long nr_pages
, sector_t start_block
)
2345 struct rb_node
**link
= &sis
->swap_extent_root
.rb_node
, *parent
= NULL
;
2346 struct swap_extent
*se
;
2347 struct swap_extent
*new_se
;
2350 * place the new node at the right most since the
2351 * function is called in ascending page order.
2355 link
= &parent
->rb_right
;
2359 se
= rb_entry(parent
, struct swap_extent
, rb_node
);
2360 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
2361 if (se
->start_block
+ se
->nr_pages
== start_block
) {
2363 se
->nr_pages
+= nr_pages
;
2368 /* No merge, insert a new extent. */
2369 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
2372 new_se
->start_page
= start_page
;
2373 new_se
->nr_pages
= nr_pages
;
2374 new_se
->start_block
= start_block
;
2376 rb_link_node(&new_se
->rb_node
, parent
, link
);
2377 rb_insert_color(&new_se
->rb_node
, &sis
->swap_extent_root
);
2380 EXPORT_SYMBOL_GPL(add_swap_extent
);
2383 * A `swap extent' is a simple thing which maps a contiguous range of pages
2384 * onto a contiguous range of disk blocks. An ordered list of swap extents
2385 * is built at swapon time and is then used at swap_writepage/swap_readpage
2386 * time for locating where on disk a page belongs.
2388 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2389 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2390 * swap files identically.
2392 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2393 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2394 * swapfiles are handled *identically* after swapon time.
2396 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2397 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2398 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2399 * requirements, they are simply tossed out - we will never use those blocks
2402 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2403 * prevents users from writing to the swap device, which will corrupt memory.
2405 * The amount of disk space which a single swap extent represents varies.
2406 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2407 * extents in the list. To avoid much list walking, we cache the previous
2408 * search location in `curr_swap_extent', and start new searches from there.
2409 * This is extremely effective. The average number of iterations in
2410 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2412 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
2414 struct file
*swap_file
= sis
->swap_file
;
2415 struct address_space
*mapping
= swap_file
->f_mapping
;
2416 struct inode
*inode
= mapping
->host
;
2419 if (S_ISBLK(inode
->i_mode
)) {
2420 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2425 if (mapping
->a_ops
->swap_activate
) {
2426 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
2428 sis
->flags
|= SWP_ACTIVATED
;
2430 sis
->flags
|= SWP_FS
;
2431 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2437 return generic_swapfile_activate(sis
, swap_file
, span
);
2440 static int swap_node(struct swap_info_struct
*p
)
2442 struct block_device
*bdev
;
2447 bdev
= p
->swap_file
->f_inode
->i_sb
->s_bdev
;
2449 return bdev
? bdev
->bd_disk
->node_id
: NUMA_NO_NODE
;
2452 static void setup_swap_info(struct swap_info_struct
*p
, int prio
,
2453 unsigned char *swap_map
,
2454 struct swap_cluster_info
*cluster_info
)
2461 p
->prio
= --least_priority
;
2463 * the plist prio is negated because plist ordering is
2464 * low-to-high, while swap ordering is high-to-low
2466 p
->list
.prio
= -p
->prio
;
2469 p
->avail_lists
[i
].prio
= -p
->prio
;
2471 if (swap_node(p
) == i
)
2472 p
->avail_lists
[i
].prio
= 1;
2474 p
->avail_lists
[i
].prio
= -p
->prio
;
2477 p
->swap_map
= swap_map
;
2478 p
->cluster_info
= cluster_info
;
2481 static void _enable_swap_info(struct swap_info_struct
*p
)
2483 p
->flags
|= SWP_WRITEOK
| SWP_VALID
;
2484 atomic_long_add(p
->pages
, &nr_swap_pages
);
2485 total_swap_pages
+= p
->pages
;
2487 assert_spin_locked(&swap_lock
);
2489 * both lists are plists, and thus priority ordered.
2490 * swap_active_head needs to be priority ordered for swapoff(),
2491 * which on removal of any swap_info_struct with an auto-assigned
2492 * (i.e. negative) priority increments the auto-assigned priority
2493 * of any lower-priority swap_info_structs.
2494 * swap_avail_head needs to be priority ordered for get_swap_page(),
2495 * which allocates swap pages from the highest available priority
2498 plist_add(&p
->list
, &swap_active_head
);
2499 add_to_avail_list(p
);
2502 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
2503 unsigned char *swap_map
,
2504 struct swap_cluster_info
*cluster_info
,
2505 unsigned long *frontswap_map
)
2507 frontswap_init(p
->type
, frontswap_map
);
2508 spin_lock(&swap_lock
);
2509 spin_lock(&p
->lock
);
2510 setup_swap_info(p
, prio
, swap_map
, cluster_info
);
2511 spin_unlock(&p
->lock
);
2512 spin_unlock(&swap_lock
);
2514 * Guarantee swap_map, cluster_info, etc. fields are valid
2515 * between get/put_swap_device() if SWP_VALID bit is set
2518 spin_lock(&swap_lock
);
2519 spin_lock(&p
->lock
);
2520 _enable_swap_info(p
);
2521 spin_unlock(&p
->lock
);
2522 spin_unlock(&swap_lock
);
2525 static void reinsert_swap_info(struct swap_info_struct
*p
)
2527 spin_lock(&swap_lock
);
2528 spin_lock(&p
->lock
);
2529 setup_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
2530 _enable_swap_info(p
);
2531 spin_unlock(&p
->lock
);
2532 spin_unlock(&swap_lock
);
2535 bool has_usable_swap(void)
2539 spin_lock(&swap_lock
);
2540 if (plist_head_empty(&swap_active_head
))
2542 spin_unlock(&swap_lock
);
2546 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
2548 struct swap_info_struct
*p
= NULL
;
2549 unsigned char *swap_map
;
2550 struct swap_cluster_info
*cluster_info
;
2551 unsigned long *frontswap_map
;
2552 struct file
*swap_file
, *victim
;
2553 struct address_space
*mapping
;
2554 struct inode
*inode
;
2555 struct filename
*pathname
;
2557 unsigned int old_block_size
;
2559 if (!capable(CAP_SYS_ADMIN
))
2562 BUG_ON(!current
->mm
);
2564 pathname
= getname(specialfile
);
2565 if (IS_ERR(pathname
))
2566 return PTR_ERR(pathname
);
2568 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
2569 err
= PTR_ERR(victim
);
2573 mapping
= victim
->f_mapping
;
2574 spin_lock(&swap_lock
);
2575 plist_for_each_entry(p
, &swap_active_head
, list
) {
2576 if (p
->flags
& SWP_WRITEOK
) {
2577 if (p
->swap_file
->f_mapping
== mapping
) {
2585 spin_unlock(&swap_lock
);
2588 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2589 vm_unacct_memory(p
->pages
);
2592 spin_unlock(&swap_lock
);
2595 del_from_avail_list(p
);
2596 spin_lock(&p
->lock
);
2598 struct swap_info_struct
*si
= p
;
2601 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2604 for_each_node(nid
) {
2605 if (si
->avail_lists
[nid
].prio
!= 1)
2606 si
->avail_lists
[nid
].prio
--;
2611 plist_del(&p
->list
, &swap_active_head
);
2612 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2613 total_swap_pages
-= p
->pages
;
2614 p
->flags
&= ~SWP_WRITEOK
;
2615 spin_unlock(&p
->lock
);
2616 spin_unlock(&swap_lock
);
2618 disable_swap_slots_cache_lock();
2620 set_current_oom_origin();
2621 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
2622 clear_current_oom_origin();
2625 /* re-insert swap space back into swap_list */
2626 reinsert_swap_info(p
);
2627 reenable_swap_slots_cache_unlock();
2631 reenable_swap_slots_cache_unlock();
2633 spin_lock(&swap_lock
);
2634 spin_lock(&p
->lock
);
2635 p
->flags
&= ~SWP_VALID
; /* mark swap device as invalid */
2636 spin_unlock(&p
->lock
);
2637 spin_unlock(&swap_lock
);
2639 * wait for swap operations protected by get/put_swap_device()
2644 flush_work(&p
->discard_work
);
2646 destroy_swap_extents(p
);
2647 if (p
->flags
& SWP_CONTINUED
)
2648 free_swap_count_continuations(p
);
2650 if (!p
->bdev
|| !blk_queue_nonrot(bdev_get_queue(p
->bdev
)))
2651 atomic_dec(&nr_rotate_swap
);
2653 mutex_lock(&swapon_mutex
);
2654 spin_lock(&swap_lock
);
2655 spin_lock(&p
->lock
);
2658 /* wait for anyone still in scan_swap_map */
2659 p
->highest_bit
= 0; /* cuts scans short */
2660 while (p
->flags
>= SWP_SCANNING
) {
2661 spin_unlock(&p
->lock
);
2662 spin_unlock(&swap_lock
);
2663 schedule_timeout_uninterruptible(1);
2664 spin_lock(&swap_lock
);
2665 spin_lock(&p
->lock
);
2668 swap_file
= p
->swap_file
;
2669 old_block_size
= p
->old_block_size
;
2670 p
->swap_file
= NULL
;
2672 swap_map
= p
->swap_map
;
2674 cluster_info
= p
->cluster_info
;
2675 p
->cluster_info
= NULL
;
2676 frontswap_map
= frontswap_map_get(p
);
2677 spin_unlock(&p
->lock
);
2678 spin_unlock(&swap_lock
);
2679 frontswap_invalidate_area(p
->type
);
2680 frontswap_map_set(p
, NULL
);
2681 mutex_unlock(&swapon_mutex
);
2682 free_percpu(p
->percpu_cluster
);
2683 p
->percpu_cluster
= NULL
;
2684 free_percpu(p
->cluster_next_cpu
);
2685 p
->cluster_next_cpu
= NULL
;
2687 kvfree(cluster_info
);
2688 kvfree(frontswap_map
);
2689 /* Destroy swap account information */
2690 swap_cgroup_swapoff(p
->type
);
2691 exit_swap_address_space(p
->type
);
2693 inode
= mapping
->host
;
2694 if (S_ISBLK(inode
->i_mode
)) {
2695 struct block_device
*bdev
= I_BDEV(inode
);
2697 set_blocksize(bdev
, old_block_size
);
2698 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2702 inode
->i_flags
&= ~S_SWAPFILE
;
2703 inode_unlock(inode
);
2704 filp_close(swap_file
, NULL
);
2707 * Clear the SWP_USED flag after all resources are freed so that swapon
2708 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2709 * not hold p->lock after we cleared its SWP_WRITEOK.
2711 spin_lock(&swap_lock
);
2713 spin_unlock(&swap_lock
);
2716 atomic_inc(&proc_poll_event
);
2717 wake_up_interruptible(&proc_poll_wait
);
2720 filp_close(victim
, NULL
);
2726 #ifdef CONFIG_PROC_FS
2727 static __poll_t
swaps_poll(struct file
*file
, poll_table
*wait
)
2729 struct seq_file
*seq
= file
->private_data
;
2731 poll_wait(file
, &proc_poll_wait
, wait
);
2733 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2734 seq
->poll_event
= atomic_read(&proc_poll_event
);
2735 return EPOLLIN
| EPOLLRDNORM
| EPOLLERR
| EPOLLPRI
;
2738 return EPOLLIN
| EPOLLRDNORM
;
2742 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2744 struct swap_info_struct
*si
;
2748 mutex_lock(&swapon_mutex
);
2751 return SEQ_START_TOKEN
;
2753 for (type
= 0; (si
= swap_type_to_swap_info(type
)); type
++) {
2754 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2763 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2765 struct swap_info_struct
*si
= v
;
2768 if (v
== SEQ_START_TOKEN
)
2771 type
= si
->type
+ 1;
2774 for (; (si
= swap_type_to_swap_info(type
)); type
++) {
2775 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2783 static void swap_stop(struct seq_file
*swap
, void *v
)
2785 mutex_unlock(&swapon_mutex
);
2788 static int swap_show(struct seq_file
*swap
, void *v
)
2790 struct swap_info_struct
*si
= v
;
2793 unsigned int bytes
, inuse
;
2795 if (si
== SEQ_START_TOKEN
) {
2796 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2800 bytes
= si
->pages
<< (PAGE_SHIFT
- 10);
2801 inuse
= si
->inuse_pages
<< (PAGE_SHIFT
- 10);
2803 file
= si
->swap_file
;
2804 len
= seq_file_path(swap
, file
, " \t\n\\");
2805 seq_printf(swap
, "%*s%s\t%u\t%s%u\t%s%d\n",
2806 len
< 40 ? 40 - len
: 1, " ",
2807 S_ISBLK(file_inode(file
)->i_mode
) ?
2808 "partition" : "file\t",
2809 bytes
, bytes
< 10000000 ? "\t" : "",
2810 inuse
, inuse
< 10000000 ? "\t" : "",
2815 static const struct seq_operations swaps_op
= {
2816 .start
= swap_start
,
2822 static int swaps_open(struct inode
*inode
, struct file
*file
)
2824 struct seq_file
*seq
;
2827 ret
= seq_open(file
, &swaps_op
);
2831 seq
= file
->private_data
;
2832 seq
->poll_event
= atomic_read(&proc_poll_event
);
2836 static const struct proc_ops swaps_proc_ops
= {
2837 .proc_flags
= PROC_ENTRY_PERMANENT
,
2838 .proc_open
= swaps_open
,
2839 .proc_read
= seq_read
,
2840 .proc_lseek
= seq_lseek
,
2841 .proc_release
= seq_release
,
2842 .proc_poll
= swaps_poll
,
2845 static int __init
procswaps_init(void)
2847 proc_create("swaps", 0, NULL
, &swaps_proc_ops
);
2850 __initcall(procswaps_init
);
2851 #endif /* CONFIG_PROC_FS */
2853 #ifdef MAX_SWAPFILES_CHECK
2854 static int __init
max_swapfiles_check(void)
2856 MAX_SWAPFILES_CHECK();
2859 late_initcall(max_swapfiles_check
);
2862 static struct swap_info_struct
*alloc_swap_info(void)
2864 struct swap_info_struct
*p
;
2868 p
= kvzalloc(struct_size(p
, avail_lists
, nr_node_ids
), GFP_KERNEL
);
2870 return ERR_PTR(-ENOMEM
);
2872 spin_lock(&swap_lock
);
2873 for (type
= 0; type
< nr_swapfiles
; type
++) {
2874 if (!(swap_info
[type
]->flags
& SWP_USED
))
2877 if (type
>= MAX_SWAPFILES
) {
2878 spin_unlock(&swap_lock
);
2880 return ERR_PTR(-EPERM
);
2882 if (type
>= nr_swapfiles
) {
2884 WRITE_ONCE(swap_info
[type
], p
);
2886 * Write swap_info[type] before nr_swapfiles, in case a
2887 * racing procfs swap_start() or swap_next() is reading them.
2888 * (We never shrink nr_swapfiles, we never free this entry.)
2891 WRITE_ONCE(nr_swapfiles
, nr_swapfiles
+ 1);
2894 p
= swap_info
[type
];
2896 * Do not memset this entry: a racing procfs swap_next()
2897 * would be relying on p->type to remain valid.
2900 p
->swap_extent_root
= RB_ROOT
;
2901 plist_node_init(&p
->list
, 0);
2903 plist_node_init(&p
->avail_lists
[i
], 0);
2904 p
->flags
= SWP_USED
;
2905 spin_unlock(&swap_lock
);
2906 spin_lock_init(&p
->lock
);
2907 spin_lock_init(&p
->cont_lock
);
2912 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2916 if (S_ISBLK(inode
->i_mode
)) {
2917 p
->bdev
= bdgrab(I_BDEV(inode
));
2918 error
= blkdev_get(p
->bdev
,
2919 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, p
);
2924 p
->old_block_size
= block_size(p
->bdev
);
2925 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2929 * Zoned block devices contain zones that have a sequential
2930 * write only restriction. Hence zoned block devices are not
2931 * suitable for swapping. Disallow them here.
2933 if (blk_queue_is_zoned(p
->bdev
->bd_queue
))
2935 p
->flags
|= SWP_BLKDEV
;
2936 } else if (S_ISREG(inode
->i_mode
)) {
2937 p
->bdev
= inode
->i_sb
->s_bdev
;
2945 * Find out how many pages are allowed for a single swap device. There
2946 * are two limiting factors:
2947 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2948 * 2) the number of bits in the swap pte, as defined by the different
2951 * In order to find the largest possible bit mask, a swap entry with
2952 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2953 * decoded to a swp_entry_t again, and finally the swap offset is
2956 * This will mask all the bits from the initial ~0UL mask that can't
2957 * be encoded in either the swp_entry_t or the architecture definition
2960 unsigned long generic_max_swapfile_size(void)
2962 return swp_offset(pte_to_swp_entry(
2963 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2966 /* Can be overridden by an architecture for additional checks. */
2967 __weak
unsigned long max_swapfile_size(void)
2969 return generic_max_swapfile_size();
2972 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2973 union swap_header
*swap_header
,
2974 struct inode
*inode
)
2977 unsigned long maxpages
;
2978 unsigned long swapfilepages
;
2979 unsigned long last_page
;
2981 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2982 pr_err("Unable to find swap-space signature\n");
2986 /* swap partition endianess hack... */
2987 if (swab32(swap_header
->info
.version
) == 1) {
2988 swab32s(&swap_header
->info
.version
);
2989 swab32s(&swap_header
->info
.last_page
);
2990 swab32s(&swap_header
->info
.nr_badpages
);
2991 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2993 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2994 swab32s(&swap_header
->info
.badpages
[i
]);
2996 /* Check the swap header's sub-version */
2997 if (swap_header
->info
.version
!= 1) {
2998 pr_warn("Unable to handle swap header version %d\n",
2999 swap_header
->info
.version
);
3004 p
->cluster_next
= 1;
3007 maxpages
= max_swapfile_size();
3008 last_page
= swap_header
->info
.last_page
;
3010 pr_warn("Empty swap-file\n");
3013 if (last_page
> maxpages
) {
3014 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
3015 maxpages
<< (PAGE_SHIFT
- 10),
3016 last_page
<< (PAGE_SHIFT
- 10));
3018 if (maxpages
> last_page
) {
3019 maxpages
= last_page
+ 1;
3020 /* p->max is an unsigned int: don't overflow it */
3021 if ((unsigned int)maxpages
== 0)
3022 maxpages
= UINT_MAX
;
3024 p
->highest_bit
= maxpages
- 1;
3028 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
3029 if (swapfilepages
&& maxpages
> swapfilepages
) {
3030 pr_warn("Swap area shorter than signature indicates\n");
3033 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
3035 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
3041 #define SWAP_CLUSTER_INFO_COLS \
3042 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3043 #define SWAP_CLUSTER_SPACE_COLS \
3044 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3045 #define SWAP_CLUSTER_COLS \
3046 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3048 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
3049 union swap_header
*swap_header
,
3050 unsigned char *swap_map
,
3051 struct swap_cluster_info
*cluster_info
,
3052 unsigned long maxpages
,
3056 unsigned int nr_good_pages
;
3058 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3059 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
3060 unsigned long i
, idx
;
3062 nr_good_pages
= maxpages
- 1; /* omit header page */
3064 cluster_list_init(&p
->free_clusters
);
3065 cluster_list_init(&p
->discard_clusters
);
3067 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
3068 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
3069 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
3071 if (page_nr
< maxpages
) {
3072 swap_map
[page_nr
] = SWAP_MAP_BAD
;
3075 * Haven't marked the cluster free yet, no list
3076 * operation involved
3078 inc_cluster_info_page(p
, cluster_info
, page_nr
);
3082 /* Haven't marked the cluster free yet, no list operation involved */
3083 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
3084 inc_cluster_info_page(p
, cluster_info
, i
);
3086 if (nr_good_pages
) {
3087 swap_map
[0] = SWAP_MAP_BAD
;
3089 * Not mark the cluster free yet, no list
3090 * operation involved
3092 inc_cluster_info_page(p
, cluster_info
, 0);
3094 p
->pages
= nr_good_pages
;
3095 nr_extents
= setup_swap_extents(p
, span
);
3098 nr_good_pages
= p
->pages
;
3100 if (!nr_good_pages
) {
3101 pr_warn("Empty swap-file\n");
3110 * Reduce false cache line sharing between cluster_info and
3111 * sharing same address space.
3113 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
3114 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
3115 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
3116 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
3117 if (idx
>= nr_clusters
)
3119 if (cluster_count(&cluster_info
[idx
]))
3121 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
3122 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
3130 * Helper to sys_swapon determining if a given swap
3131 * backing device queue supports DISCARD operations.
3133 static bool swap_discardable(struct swap_info_struct
*si
)
3135 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
3137 if (!q
|| !blk_queue_discard(q
))
3143 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
3145 struct swap_info_struct
*p
;
3146 struct filename
*name
;
3147 struct file
*swap_file
= NULL
;
3148 struct address_space
*mapping
;
3151 union swap_header
*swap_header
;
3154 unsigned long maxpages
;
3155 unsigned char *swap_map
= NULL
;
3156 struct swap_cluster_info
*cluster_info
= NULL
;
3157 unsigned long *frontswap_map
= NULL
;
3158 struct page
*page
= NULL
;
3159 struct inode
*inode
= NULL
;
3160 bool inced_nr_rotate_swap
= false;
3162 if (swap_flags
& ~SWAP_FLAGS_VALID
)
3165 if (!capable(CAP_SYS_ADMIN
))
3168 if (!swap_avail_heads
)
3171 p
= alloc_swap_info();
3175 INIT_WORK(&p
->discard_work
, swap_discard_work
);
3177 name
= getname(specialfile
);
3179 error
= PTR_ERR(name
);
3183 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
3184 if (IS_ERR(swap_file
)) {
3185 error
= PTR_ERR(swap_file
);
3190 p
->swap_file
= swap_file
;
3191 mapping
= swap_file
->f_mapping
;
3192 inode
= mapping
->host
;
3194 error
= claim_swapfile(p
, inode
);
3195 if (unlikely(error
))
3199 if (IS_SWAPFILE(inode
)) {
3201 goto bad_swap_unlock_inode
;
3205 * Read the swap header.
3207 if (!mapping
->a_ops
->readpage
) {
3209 goto bad_swap_unlock_inode
;
3211 page
= read_mapping_page(mapping
, 0, swap_file
);
3213 error
= PTR_ERR(page
);
3214 goto bad_swap_unlock_inode
;
3216 swap_header
= kmap(page
);
3218 maxpages
= read_swap_header(p
, swap_header
, inode
);
3219 if (unlikely(!maxpages
)) {
3221 goto bad_swap_unlock_inode
;
3224 /* OK, set up the swap map and apply the bad block list */
3225 swap_map
= vzalloc(maxpages
);
3228 goto bad_swap_unlock_inode
;
3231 if (bdi_cap_stable_pages_required(inode_to_bdi(inode
)))
3232 p
->flags
|= SWP_STABLE_WRITES
;
3234 if (bdi_cap_synchronous_io(inode_to_bdi(inode
)))
3235 p
->flags
|= SWP_SYNCHRONOUS_IO
;
3237 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
3239 unsigned long ci
, nr_cluster
;
3241 p
->flags
|= SWP_SOLIDSTATE
;
3242 p
->cluster_next_cpu
= alloc_percpu(unsigned int);
3243 if (!p
->cluster_next_cpu
) {
3245 goto bad_swap_unlock_inode
;
3248 * select a random position to start with to help wear leveling
3251 for_each_possible_cpu(cpu
) {
3252 per_cpu(*p
->cluster_next_cpu
, cpu
) =
3253 1 + prandom_u32_max(p
->highest_bit
);
3255 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3257 cluster_info
= kvcalloc(nr_cluster
, sizeof(*cluster_info
),
3259 if (!cluster_info
) {
3261 goto bad_swap_unlock_inode
;
3264 for (ci
= 0; ci
< nr_cluster
; ci
++)
3265 spin_lock_init(&((cluster_info
+ ci
)->lock
));
3267 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
3268 if (!p
->percpu_cluster
) {
3270 goto bad_swap_unlock_inode
;
3272 for_each_possible_cpu(cpu
) {
3273 struct percpu_cluster
*cluster
;
3274 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
3275 cluster_set_null(&cluster
->index
);
3278 atomic_inc(&nr_rotate_swap
);
3279 inced_nr_rotate_swap
= true;
3282 error
= swap_cgroup_swapon(p
->type
, maxpages
);
3284 goto bad_swap_unlock_inode
;
3286 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
3287 cluster_info
, maxpages
, &span
);
3288 if (unlikely(nr_extents
< 0)) {
3290 goto bad_swap_unlock_inode
;
3292 /* frontswap enabled? set up bit-per-page map for frontswap */
3293 if (IS_ENABLED(CONFIG_FRONTSWAP
))
3294 frontswap_map
= kvcalloc(BITS_TO_LONGS(maxpages
),
3298 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
3300 * When discard is enabled for swap with no particular
3301 * policy flagged, we set all swap discard flags here in
3302 * order to sustain backward compatibility with older
3303 * swapon(8) releases.
3305 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
3309 * By flagging sys_swapon, a sysadmin can tell us to
3310 * either do single-time area discards only, or to just
3311 * perform discards for released swap page-clusters.
3312 * Now it's time to adjust the p->flags accordingly.
3314 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
3315 p
->flags
&= ~SWP_PAGE_DISCARD
;
3316 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
3317 p
->flags
&= ~SWP_AREA_DISCARD
;
3319 /* issue a swapon-time discard if it's still required */
3320 if (p
->flags
& SWP_AREA_DISCARD
) {
3321 int err
= discard_swap(p
);
3323 pr_err("swapon: discard_swap(%p): %d\n",
3328 error
= init_swap_address_space(p
->type
, maxpages
);
3330 goto bad_swap_unlock_inode
;
3333 * Flush any pending IO and dirty mappings before we start using this
3336 inode
->i_flags
|= S_SWAPFILE
;
3337 error
= inode_drain_writes(inode
);
3339 inode
->i_flags
&= ~S_SWAPFILE
;
3340 goto bad_swap_unlock_inode
;
3343 mutex_lock(&swapon_mutex
);
3345 if (swap_flags
& SWAP_FLAG_PREFER
)
3347 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
3348 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
3350 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3351 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
3352 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
3353 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
3354 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
3355 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
3356 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
3357 (frontswap_map
) ? "FS" : "");
3359 mutex_unlock(&swapon_mutex
);
3360 atomic_inc(&proc_poll_event
);
3361 wake_up_interruptible(&proc_poll_wait
);
3365 bad_swap_unlock_inode
:
3366 inode_unlock(inode
);
3368 free_percpu(p
->percpu_cluster
);
3369 p
->percpu_cluster
= NULL
;
3370 free_percpu(p
->cluster_next_cpu
);
3371 p
->cluster_next_cpu
= NULL
;
3372 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
3373 set_blocksize(p
->bdev
, p
->old_block_size
);
3374 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
3377 destroy_swap_extents(p
);
3378 swap_cgroup_swapoff(p
->type
);
3379 spin_lock(&swap_lock
);
3380 p
->swap_file
= NULL
;
3382 spin_unlock(&swap_lock
);
3384 kvfree(cluster_info
);
3385 kvfree(frontswap_map
);
3386 if (inced_nr_rotate_swap
)
3387 atomic_dec(&nr_rotate_swap
);
3389 filp_close(swap_file
, NULL
);
3391 if (page
&& !IS_ERR(page
)) {
3398 inode_unlock(inode
);
3400 enable_swap_slots_cache();
3404 void si_swapinfo(struct sysinfo
*val
)
3407 unsigned long nr_to_be_unused
= 0;
3409 spin_lock(&swap_lock
);
3410 for (type
= 0; type
< nr_swapfiles
; type
++) {
3411 struct swap_info_struct
*si
= swap_info
[type
];
3413 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
3414 nr_to_be_unused
+= si
->inuse_pages
;
3416 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
3417 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
3418 spin_unlock(&swap_lock
);
3422 * Verify that a swap entry is valid and increment its swap map count.
3424 * Returns error code in following case.
3426 * - swp_entry is invalid -> EINVAL
3427 * - swp_entry is migration entry -> EINVAL
3428 * - swap-cache reference is requested but there is already one. -> EEXIST
3429 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3430 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3432 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
3434 struct swap_info_struct
*p
;
3435 struct swap_cluster_info
*ci
;
3436 unsigned long offset
;
3437 unsigned char count
;
3438 unsigned char has_cache
;
3441 p
= get_swap_device(entry
);
3445 offset
= swp_offset(entry
);
3446 ci
= lock_cluster_or_swap_info(p
, offset
);
3448 count
= p
->swap_map
[offset
];
3451 * swapin_readahead() doesn't check if a swap entry is valid, so the
3452 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3454 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
3459 has_cache
= count
& SWAP_HAS_CACHE
;
3460 count
&= ~SWAP_HAS_CACHE
;
3463 if (usage
== SWAP_HAS_CACHE
) {
3465 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3466 if (!has_cache
&& count
)
3467 has_cache
= SWAP_HAS_CACHE
;
3468 else if (has_cache
) /* someone else added cache */
3470 else /* no users remaining */
3473 } else if (count
|| has_cache
) {
3475 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
3477 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
3479 else if (swap_count_continued(p
, offset
, count
))
3480 count
= COUNT_CONTINUED
;
3484 err
= -ENOENT
; /* unused swap entry */
3486 p
->swap_map
[offset
] = count
| has_cache
;
3489 unlock_cluster_or_swap_info(p
, ci
);
3497 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3498 * (in which case its reference count is never incremented).
3500 void swap_shmem_alloc(swp_entry_t entry
)
3502 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
3506 * Increase reference count of swap entry by 1.
3507 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3508 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3509 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3510 * might occur if a page table entry has got corrupted.
3512 int swap_duplicate(swp_entry_t entry
)
3516 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
3517 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
3522 * @entry: swap entry for which we allocate swap cache.
3524 * Called when allocating swap cache for existing swap entry,
3525 * This can return error codes. Returns 0 at success.
3526 * -EEXIST means there is a swap cache.
3527 * Note: return code is different from swap_duplicate().
3529 int swapcache_prepare(swp_entry_t entry
)
3531 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
3534 struct swap_info_struct
*swp_swap_info(swp_entry_t entry
)
3536 return swap_type_to_swap_info(swp_type(entry
));
3539 struct swap_info_struct
*page_swap_info(struct page
*page
)
3541 swp_entry_t entry
= { .val
= page_private(page
) };
3542 return swp_swap_info(entry
);
3546 * out-of-line __page_file_ methods to avoid include hell.
3548 struct address_space
*__page_file_mapping(struct page
*page
)
3550 return page_swap_info(page
)->swap_file
->f_mapping
;
3552 EXPORT_SYMBOL_GPL(__page_file_mapping
);
3554 pgoff_t
__page_file_index(struct page
*page
)
3556 swp_entry_t swap
= { .val
= page_private(page
) };
3557 return swp_offset(swap
);
3559 EXPORT_SYMBOL_GPL(__page_file_index
);
3562 * add_swap_count_continuation - called when a swap count is duplicated
3563 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3564 * page of the original vmalloc'ed swap_map, to hold the continuation count
3565 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3566 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3568 * These continuation pages are seldom referenced: the common paths all work
3569 * on the original swap_map, only referring to a continuation page when the
3570 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3572 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3573 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3574 * can be called after dropping locks.
3576 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
3578 struct swap_info_struct
*si
;
3579 struct swap_cluster_info
*ci
;
3582 struct page
*list_page
;
3584 unsigned char count
;
3588 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3589 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3591 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
3593 si
= get_swap_device(entry
);
3596 * An acceptable race has occurred since the failing
3597 * __swap_duplicate(): the swap device may be swapoff
3601 spin_lock(&si
->lock
);
3603 offset
= swp_offset(entry
);
3605 ci
= lock_cluster(si
, offset
);
3607 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
3609 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
3611 * The higher the swap count, the more likely it is that tasks
3612 * will race to add swap count continuation: we need to avoid
3613 * over-provisioning.
3624 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3625 * no architecture is using highmem pages for kernel page tables: so it
3626 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3628 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3629 offset
&= ~PAGE_MASK
;
3631 spin_lock(&si
->cont_lock
);
3633 * Page allocation does not initialize the page's lru field,
3634 * but it does always reset its private field.
3636 if (!page_private(head
)) {
3637 BUG_ON(count
& COUNT_CONTINUED
);
3638 INIT_LIST_HEAD(&head
->lru
);
3639 set_page_private(head
, SWP_CONTINUED
);
3640 si
->flags
|= SWP_CONTINUED
;
3643 list_for_each_entry(list_page
, &head
->lru
, lru
) {
3647 * If the previous map said no continuation, but we've found
3648 * a continuation page, free our allocation and use this one.
3650 if (!(count
& COUNT_CONTINUED
))
3651 goto out_unlock_cont
;
3653 map
= kmap_atomic(list_page
) + offset
;
3658 * If this continuation count now has some space in it,
3659 * free our allocation and use this one.
3661 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
3662 goto out_unlock_cont
;
3665 list_add_tail(&page
->lru
, &head
->lru
);
3666 page
= NULL
; /* now it's attached, don't free it */
3668 spin_unlock(&si
->cont_lock
);
3671 spin_unlock(&si
->lock
);
3672 put_swap_device(si
);
3680 * swap_count_continued - when the original swap_map count is incremented
3681 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3682 * into, carry if so, or else fail until a new continuation page is allocated;
3683 * when the original swap_map count is decremented from 0 with continuation,
3684 * borrow from the continuation and report whether it still holds more.
3685 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3688 static bool swap_count_continued(struct swap_info_struct
*si
,
3689 pgoff_t offset
, unsigned char count
)
3696 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3697 if (page_private(head
) != SWP_CONTINUED
) {
3698 BUG_ON(count
& COUNT_CONTINUED
);
3699 return false; /* need to add count continuation */
3702 spin_lock(&si
->cont_lock
);
3703 offset
&= ~PAGE_MASK
;
3704 page
= list_next_entry(head
, lru
);
3705 map
= kmap_atomic(page
) + offset
;
3707 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3708 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3710 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3712 * Think of how you add 1 to 999
3714 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3716 page
= list_next_entry(page
, lru
);
3717 BUG_ON(page
== head
);
3718 map
= kmap_atomic(page
) + offset
;
3720 if (*map
== SWAP_CONT_MAX
) {
3722 page
= list_next_entry(page
, lru
);
3724 ret
= false; /* add count continuation */
3727 map
= kmap_atomic(page
) + offset
;
3728 init_map
: *map
= 0; /* we didn't zero the page */
3732 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3733 map
= kmap_atomic(page
) + offset
;
3734 *map
= COUNT_CONTINUED
;
3737 ret
= true; /* incremented */
3739 } else { /* decrementing */
3741 * Think of how you subtract 1 from 1000
3743 BUG_ON(count
!= COUNT_CONTINUED
);
3744 while (*map
== COUNT_CONTINUED
) {
3746 page
= list_next_entry(page
, lru
);
3747 BUG_ON(page
== head
);
3748 map
= kmap_atomic(page
) + offset
;
3755 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3756 map
= kmap_atomic(page
) + offset
;
3757 *map
= SWAP_CONT_MAX
| count
;
3758 count
= COUNT_CONTINUED
;
3761 ret
= count
== COUNT_CONTINUED
;
3764 spin_unlock(&si
->cont_lock
);
3769 * free_swap_count_continuations - swapoff free all the continuation pages
3770 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3772 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3776 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3778 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3779 if (page_private(head
)) {
3780 struct page
*page
, *next
;
3782 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3783 list_del(&page
->lru
);
3790 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3791 void cgroup_throttle_swaprate(struct page
*page
, gfp_t gfp_mask
)
3793 struct swap_info_struct
*si
, *next
;
3794 int nid
= page_to_nid(page
);
3796 if (!(gfp_mask
& __GFP_IO
))
3799 if (!blk_cgroup_congested())
3803 * We've already scheduled a throttle, avoid taking the global swap
3806 if (current
->throttle_queue
)
3809 spin_lock(&swap_avail_lock
);
3810 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[nid
],
3813 blkcg_schedule_throttle(bdev_get_queue(si
->bdev
), true);
3817 spin_unlock(&swap_avail_lock
);
3821 static int __init
swapfile_init(void)
3825 swap_avail_heads
= kmalloc_array(nr_node_ids
, sizeof(struct plist_head
),
3827 if (!swap_avail_heads
) {
3828 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3833 plist_head_init(&swap_avail_heads
[nid
]);
3837 subsys_initcall(swapfile_init
);