1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie
10 #include <linux/sched/mm.h>
11 #include <linux/sched/task.h>
12 #include <linux/hugetlb.h>
13 #include <linux/mman.h>
14 #include <linux/slab.h>
15 #include <linux/kernel_stat.h>
16 #include <linux/swap.h>
17 #include <linux/vmalloc.h>
18 #include <linux/pagemap.h>
19 #include <linux/namei.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/writeback.h>
24 #include <linux/proc_fs.h>
25 #include <linux/seq_file.h>
26 #include <linux/init.h>
27 #include <linux/ksm.h>
28 #include <linux/rmap.h>
29 #include <linux/security.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mutex.h>
32 #include <linux/capability.h>
33 #include <linux/syscalls.h>
34 #include <linux/memcontrol.h>
35 #include <linux/poll.h>
36 #include <linux/oom.h>
37 #include <linux/frontswap.h>
38 #include <linux/swapfile.h>
39 #include <linux/export.h>
40 #include <linux/swap_slots.h>
41 #include <linux/sort.h>
43 #include <asm/tlbflush.h>
44 #include <linux/swapops.h>
45 #include <linux/swap_cgroup.h>
47 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
49 static void free_swap_count_continuations(struct swap_info_struct
*);
50 static sector_t
map_swap_entry(swp_entry_t
, struct block_device
**);
52 DEFINE_SPINLOCK(swap_lock
);
53 static unsigned int nr_swapfiles
;
54 atomic_long_t nr_swap_pages
;
56 * Some modules use swappable objects and may try to swap them out under
57 * memory pressure (via the shrinker). Before doing so, they may wish to
58 * check to see if any swap space is available.
60 EXPORT_SYMBOL_GPL(nr_swap_pages
);
61 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
62 long total_swap_pages
;
63 static int least_priority
= -1;
65 static const char Bad_file
[] = "Bad swap file entry ";
66 static const char Unused_file
[] = "Unused swap file entry ";
67 static const char Bad_offset
[] = "Bad swap offset entry ";
68 static const char Unused_offset
[] = "Unused swap offset entry ";
71 * all active swap_info_structs
72 * protected with swap_lock, and ordered by priority.
74 PLIST_HEAD(swap_active_head
);
77 * all available (active, not full) swap_info_structs
78 * protected with swap_avail_lock, ordered by priority.
79 * This is used by get_swap_page() instead of swap_active_head
80 * because swap_active_head includes all swap_info_structs,
81 * but get_swap_page() doesn't need to look at full ones.
82 * This uses its own lock instead of swap_lock because when a
83 * swap_info_struct changes between not-full/full, it needs to
84 * add/remove itself to/from this list, but the swap_info_struct->lock
85 * is held and the locking order requires swap_lock to be taken
86 * before any swap_info_struct->lock.
88 static struct plist_head
*swap_avail_heads
;
89 static DEFINE_SPINLOCK(swap_avail_lock
);
91 struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
93 static DEFINE_MUTEX(swapon_mutex
);
95 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
96 /* Activity counter to indicate that a swapon or swapoff has occurred */
97 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
99 atomic_t nr_rotate_swap
= ATOMIC_INIT(0);
101 static struct swap_info_struct
*swap_type_to_swap_info(int type
)
103 if (type
>= READ_ONCE(nr_swapfiles
))
106 smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */
107 return READ_ONCE(swap_info
[type
]);
110 static inline unsigned char swap_count(unsigned char ent
)
112 return ent
& ~SWAP_HAS_CACHE
; /* may include COUNT_CONTINUED flag */
115 /* Reclaim the swap entry anyway if possible */
116 #define TTRS_ANYWAY 0x1
118 * Reclaim the swap entry if there are no more mappings of the
121 #define TTRS_UNMAPPED 0x2
122 /* Reclaim the swap entry if swap is getting full*/
123 #define TTRS_FULL 0x4
125 /* returns 1 if swap entry is freed */
126 static int __try_to_reclaim_swap(struct swap_info_struct
*si
,
127 unsigned long offset
, unsigned long flags
)
129 swp_entry_t entry
= swp_entry(si
->type
, offset
);
133 page
= find_get_page(swap_address_space(entry
), offset
);
137 * When this function is called from scan_swap_map_slots() and it's
138 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
139 * here. We have to use trylock for avoiding deadlock. This is a special
140 * case and you should use try_to_free_swap() with explicit lock_page()
141 * in usual operations.
143 if (trylock_page(page
)) {
144 if ((flags
& TTRS_ANYWAY
) ||
145 ((flags
& TTRS_UNMAPPED
) && !page_mapped(page
)) ||
146 ((flags
& TTRS_FULL
) && mem_cgroup_swap_full(page
)))
147 ret
= try_to_free_swap(page
);
154 static inline struct swap_extent
*first_se(struct swap_info_struct
*sis
)
156 struct rb_node
*rb
= rb_first(&sis
->swap_extent_root
);
157 return rb_entry(rb
, struct swap_extent
, rb_node
);
160 static inline struct swap_extent
*next_se(struct swap_extent
*se
)
162 struct rb_node
*rb
= rb_next(&se
->rb_node
);
163 return rb
? rb_entry(rb
, struct swap_extent
, rb_node
) : NULL
;
167 * swapon tell device that all the old swap contents can be discarded,
168 * to allow the swap device to optimize its wear-levelling.
170 static int discard_swap(struct swap_info_struct
*si
)
172 struct swap_extent
*se
;
173 sector_t start_block
;
177 /* Do not discard the swap header page! */
179 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
180 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
182 err
= blkdev_issue_discard(si
->bdev
, start_block
,
183 nr_blocks
, GFP_KERNEL
, 0);
189 for (se
= next_se(se
); se
; se
= next_se(se
)) {
190 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
191 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
193 err
= blkdev_issue_discard(si
->bdev
, start_block
,
194 nr_blocks
, GFP_KERNEL
, 0);
200 return err
; /* That will often be -EOPNOTSUPP */
203 static struct swap_extent
*
204 offset_to_swap_extent(struct swap_info_struct
*sis
, unsigned long offset
)
206 struct swap_extent
*se
;
209 rb
= sis
->swap_extent_root
.rb_node
;
211 se
= rb_entry(rb
, struct swap_extent
, rb_node
);
212 if (offset
< se
->start_page
)
214 else if (offset
>= se
->start_page
+ se
->nr_pages
)
219 /* It *must* be present */
224 * swap allocation tell device that a cluster of swap can now be discarded,
225 * to allow the swap device to optimize its wear-levelling.
227 static void discard_swap_cluster(struct swap_info_struct
*si
,
228 pgoff_t start_page
, pgoff_t nr_pages
)
230 struct swap_extent
*se
= offset_to_swap_extent(si
, start_page
);
233 pgoff_t offset
= start_page
- se
->start_page
;
234 sector_t start_block
= se
->start_block
+ offset
;
235 sector_t nr_blocks
= se
->nr_pages
- offset
;
237 if (nr_blocks
> nr_pages
)
238 nr_blocks
= nr_pages
;
239 start_page
+= nr_blocks
;
240 nr_pages
-= nr_blocks
;
242 start_block
<<= PAGE_SHIFT
- 9;
243 nr_blocks
<<= PAGE_SHIFT
- 9;
244 if (blkdev_issue_discard(si
->bdev
, start_block
,
245 nr_blocks
, GFP_NOIO
, 0))
252 #ifdef CONFIG_THP_SWAP
253 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
255 #define swap_entry_size(size) (size)
257 #define SWAPFILE_CLUSTER 256
260 * Define swap_entry_size() as constant to let compiler to optimize
261 * out some code if !CONFIG_THP_SWAP
263 #define swap_entry_size(size) 1
265 #define LATENCY_LIMIT 256
267 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
273 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
278 static inline void cluster_set_count(struct swap_cluster_info
*info
,
284 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
285 unsigned int c
, unsigned int f
)
291 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
296 static inline void cluster_set_next(struct swap_cluster_info
*info
,
302 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
303 unsigned int n
, unsigned int f
)
309 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
311 return info
->flags
& CLUSTER_FLAG_FREE
;
314 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
316 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
319 static inline void cluster_set_null(struct swap_cluster_info
*info
)
321 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
325 static inline bool cluster_is_huge(struct swap_cluster_info
*info
)
327 if (IS_ENABLED(CONFIG_THP_SWAP
))
328 return info
->flags
& CLUSTER_FLAG_HUGE
;
332 static inline void cluster_clear_huge(struct swap_cluster_info
*info
)
334 info
->flags
&= ~CLUSTER_FLAG_HUGE
;
337 static inline struct swap_cluster_info
*lock_cluster(struct swap_info_struct
*si
,
338 unsigned long offset
)
340 struct swap_cluster_info
*ci
;
342 ci
= si
->cluster_info
;
344 ci
+= offset
/ SWAPFILE_CLUSTER
;
345 spin_lock(&ci
->lock
);
350 static inline void unlock_cluster(struct swap_cluster_info
*ci
)
353 spin_unlock(&ci
->lock
);
357 * Determine the locking method in use for this device. Return
358 * swap_cluster_info if SSD-style cluster-based locking is in place.
360 static inline struct swap_cluster_info
*lock_cluster_or_swap_info(
361 struct swap_info_struct
*si
, unsigned long offset
)
363 struct swap_cluster_info
*ci
;
365 /* Try to use fine-grained SSD-style locking if available: */
366 ci
= lock_cluster(si
, offset
);
367 /* Otherwise, fall back to traditional, coarse locking: */
369 spin_lock(&si
->lock
);
374 static inline void unlock_cluster_or_swap_info(struct swap_info_struct
*si
,
375 struct swap_cluster_info
*ci
)
380 spin_unlock(&si
->lock
);
383 static inline bool cluster_list_empty(struct swap_cluster_list
*list
)
385 return cluster_is_null(&list
->head
);
388 static inline unsigned int cluster_list_first(struct swap_cluster_list
*list
)
390 return cluster_next(&list
->head
);
393 static void cluster_list_init(struct swap_cluster_list
*list
)
395 cluster_set_null(&list
->head
);
396 cluster_set_null(&list
->tail
);
399 static void cluster_list_add_tail(struct swap_cluster_list
*list
,
400 struct swap_cluster_info
*ci
,
403 if (cluster_list_empty(list
)) {
404 cluster_set_next_flag(&list
->head
, idx
, 0);
405 cluster_set_next_flag(&list
->tail
, idx
, 0);
407 struct swap_cluster_info
*ci_tail
;
408 unsigned int tail
= cluster_next(&list
->tail
);
411 * Nested cluster lock, but both cluster locks are
412 * only acquired when we held swap_info_struct->lock
415 spin_lock_nested(&ci_tail
->lock
, SINGLE_DEPTH_NESTING
);
416 cluster_set_next(ci_tail
, idx
);
417 spin_unlock(&ci_tail
->lock
);
418 cluster_set_next_flag(&list
->tail
, idx
, 0);
422 static unsigned int cluster_list_del_first(struct swap_cluster_list
*list
,
423 struct swap_cluster_info
*ci
)
427 idx
= cluster_next(&list
->head
);
428 if (cluster_next(&list
->tail
) == idx
) {
429 cluster_set_null(&list
->head
);
430 cluster_set_null(&list
->tail
);
432 cluster_set_next_flag(&list
->head
,
433 cluster_next(&ci
[idx
]), 0);
438 /* Add a cluster to discard list and schedule it to do discard */
439 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
443 * If scan_swap_map() can't find a free cluster, it will check
444 * si->swap_map directly. To make sure the discarding cluster isn't
445 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
446 * will be cleared after discard
448 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
449 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
451 cluster_list_add_tail(&si
->discard_clusters
, si
->cluster_info
, idx
);
453 schedule_work(&si
->discard_work
);
456 static void __free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
458 struct swap_cluster_info
*ci
= si
->cluster_info
;
460 cluster_set_flag(ci
+ idx
, CLUSTER_FLAG_FREE
);
461 cluster_list_add_tail(&si
->free_clusters
, ci
, idx
);
465 * Doing discard actually. After a cluster discard is finished, the cluster
466 * will be added to free cluster list. caller should hold si->lock.
468 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
470 struct swap_cluster_info
*info
, *ci
;
473 info
= si
->cluster_info
;
475 while (!cluster_list_empty(&si
->discard_clusters
)) {
476 idx
= cluster_list_del_first(&si
->discard_clusters
, info
);
477 spin_unlock(&si
->lock
);
479 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
482 spin_lock(&si
->lock
);
483 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
484 __free_cluster(si
, idx
);
485 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
486 0, SWAPFILE_CLUSTER
);
491 static void swap_discard_work(struct work_struct
*work
)
493 struct swap_info_struct
*si
;
495 si
= container_of(work
, struct swap_info_struct
, discard_work
);
497 spin_lock(&si
->lock
);
498 swap_do_scheduled_discard(si
);
499 spin_unlock(&si
->lock
);
502 static void alloc_cluster(struct swap_info_struct
*si
, unsigned long idx
)
504 struct swap_cluster_info
*ci
= si
->cluster_info
;
506 VM_BUG_ON(cluster_list_first(&si
->free_clusters
) != idx
);
507 cluster_list_del_first(&si
->free_clusters
, ci
);
508 cluster_set_count_flag(ci
+ idx
, 0, 0);
511 static void free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
513 struct swap_cluster_info
*ci
= si
->cluster_info
+ idx
;
515 VM_BUG_ON(cluster_count(ci
) != 0);
517 * If the swap is discardable, prepare discard the cluster
518 * instead of free it immediately. The cluster will be freed
521 if ((si
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
522 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
523 swap_cluster_schedule_discard(si
, idx
);
527 __free_cluster(si
, idx
);
531 * The cluster corresponding to page_nr will be used. The cluster will be
532 * removed from free cluster list and its usage counter will be increased.
534 static void inc_cluster_info_page(struct swap_info_struct
*p
,
535 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
537 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
541 if (cluster_is_free(&cluster_info
[idx
]))
542 alloc_cluster(p
, idx
);
544 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
545 cluster_set_count(&cluster_info
[idx
],
546 cluster_count(&cluster_info
[idx
]) + 1);
550 * The cluster corresponding to page_nr decreases one usage. If the usage
551 * counter becomes 0, which means no page in the cluster is in using, we can
552 * optionally discard the cluster and add it to free cluster list.
554 static void dec_cluster_info_page(struct swap_info_struct
*p
,
555 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
557 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
562 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
563 cluster_set_count(&cluster_info
[idx
],
564 cluster_count(&cluster_info
[idx
]) - 1);
566 if (cluster_count(&cluster_info
[idx
]) == 0)
567 free_cluster(p
, idx
);
571 * It's possible scan_swap_map() uses a free cluster in the middle of free
572 * cluster list. Avoiding such abuse to avoid list corruption.
575 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
576 unsigned long offset
)
578 struct percpu_cluster
*percpu_cluster
;
581 offset
/= SWAPFILE_CLUSTER
;
582 conflict
= !cluster_list_empty(&si
->free_clusters
) &&
583 offset
!= cluster_list_first(&si
->free_clusters
) &&
584 cluster_is_free(&si
->cluster_info
[offset
]);
589 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
590 cluster_set_null(&percpu_cluster
->index
);
595 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
596 * might involve allocating a new cluster for current CPU too.
598 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
599 unsigned long *offset
, unsigned long *scan_base
)
601 struct percpu_cluster
*cluster
;
602 struct swap_cluster_info
*ci
;
603 unsigned long tmp
, max
;
606 cluster
= this_cpu_ptr(si
->percpu_cluster
);
607 if (cluster_is_null(&cluster
->index
)) {
608 if (!cluster_list_empty(&si
->free_clusters
)) {
609 cluster
->index
= si
->free_clusters
.head
;
610 cluster
->next
= cluster_next(&cluster
->index
) *
612 } else if (!cluster_list_empty(&si
->discard_clusters
)) {
614 * we don't have free cluster but have some clusters in
615 * discarding, do discard now and reclaim them, then
616 * reread cluster_next_cpu since we dropped si->lock
618 swap_do_scheduled_discard(si
);
619 *scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
620 *offset
= *scan_base
;
627 * Other CPUs can use our cluster if they can't find a free cluster,
628 * check if there is still free entry in the cluster
631 max
= min_t(unsigned long, si
->max
,
632 (cluster_next(&cluster
->index
) + 1) * SWAPFILE_CLUSTER
);
634 ci
= lock_cluster(si
, tmp
);
636 if (!si
->swap_map
[tmp
])
643 cluster_set_null(&cluster
->index
);
646 cluster
->next
= tmp
+ 1;
652 static void __del_from_avail_list(struct swap_info_struct
*p
)
657 plist_del(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
660 static void del_from_avail_list(struct swap_info_struct
*p
)
662 spin_lock(&swap_avail_lock
);
663 __del_from_avail_list(p
);
664 spin_unlock(&swap_avail_lock
);
667 static void swap_range_alloc(struct swap_info_struct
*si
, unsigned long offset
,
668 unsigned int nr_entries
)
670 unsigned int end
= offset
+ nr_entries
- 1;
672 if (offset
== si
->lowest_bit
)
673 si
->lowest_bit
+= nr_entries
;
674 if (end
== si
->highest_bit
)
675 si
->highest_bit
-= nr_entries
;
676 si
->inuse_pages
+= nr_entries
;
677 if (si
->inuse_pages
== si
->pages
) {
678 si
->lowest_bit
= si
->max
;
680 del_from_avail_list(si
);
684 static void add_to_avail_list(struct swap_info_struct
*p
)
688 spin_lock(&swap_avail_lock
);
690 WARN_ON(!plist_node_empty(&p
->avail_lists
[nid
]));
691 plist_add(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
693 spin_unlock(&swap_avail_lock
);
696 static void swap_range_free(struct swap_info_struct
*si
, unsigned long offset
,
697 unsigned int nr_entries
)
699 unsigned long end
= offset
+ nr_entries
- 1;
700 void (*swap_slot_free_notify
)(struct block_device
*, unsigned long);
702 if (offset
< si
->lowest_bit
)
703 si
->lowest_bit
= offset
;
704 if (end
> si
->highest_bit
) {
705 bool was_full
= !si
->highest_bit
;
707 si
->highest_bit
= end
;
708 if (was_full
&& (si
->flags
& SWP_WRITEOK
))
709 add_to_avail_list(si
);
711 atomic_long_add(nr_entries
, &nr_swap_pages
);
712 si
->inuse_pages
-= nr_entries
;
713 if (si
->flags
& SWP_BLKDEV
)
714 swap_slot_free_notify
=
715 si
->bdev
->bd_disk
->fops
->swap_slot_free_notify
;
717 swap_slot_free_notify
= NULL
;
718 while (offset
<= end
) {
719 frontswap_invalidate_page(si
->type
, offset
);
720 if (swap_slot_free_notify
)
721 swap_slot_free_notify(si
->bdev
, offset
);
726 static void set_cluster_next(struct swap_info_struct
*si
, unsigned long next
)
730 if (!(si
->flags
& SWP_SOLIDSTATE
)) {
731 si
->cluster_next
= next
;
735 prev
= this_cpu_read(*si
->cluster_next_cpu
);
737 * Cross the swap address space size aligned trunk, choose
738 * another trunk randomly to avoid lock contention on swap
739 * address space if possible.
741 if ((prev
>> SWAP_ADDRESS_SPACE_SHIFT
) !=
742 (next
>> SWAP_ADDRESS_SPACE_SHIFT
)) {
743 /* No free swap slots available */
744 if (si
->highest_bit
<= si
->lowest_bit
)
746 next
= si
->lowest_bit
+
747 prandom_u32_max(si
->highest_bit
- si
->lowest_bit
+ 1);
748 next
= ALIGN_DOWN(next
, SWAP_ADDRESS_SPACE_PAGES
);
749 next
= max_t(unsigned int, next
, si
->lowest_bit
);
751 this_cpu_write(*si
->cluster_next_cpu
, next
);
754 static int scan_swap_map_slots(struct swap_info_struct
*si
,
755 unsigned char usage
, int nr
,
758 struct swap_cluster_info
*ci
;
759 unsigned long offset
;
760 unsigned long scan_base
;
761 unsigned long last_in_cluster
= 0;
762 int latency_ration
= LATENCY_LIMIT
;
764 bool scanned_many
= false;
767 * We try to cluster swap pages by allocating them sequentially
768 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
769 * way, however, we resort to first-free allocation, starting
770 * a new cluster. This prevents us from scattering swap pages
771 * all over the entire swap partition, so that we reduce
772 * overall disk seek times between swap pages. -- sct
773 * But we do now try to find an empty cluster. -Andrea
774 * And we let swap pages go all over an SSD partition. Hugh
777 si
->flags
+= SWP_SCANNING
;
779 * Use percpu scan base for SSD to reduce lock contention on
780 * cluster and swap cache. For HDD, sequential access is more
783 if (si
->flags
& SWP_SOLIDSTATE
)
784 scan_base
= this_cpu_read(*si
->cluster_next_cpu
);
786 scan_base
= si
->cluster_next
;
790 if (si
->cluster_info
) {
791 if (!scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
793 } else if (unlikely(!si
->cluster_nr
--)) {
794 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
795 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
799 spin_unlock(&si
->lock
);
802 * If seek is expensive, start searching for new cluster from
803 * start of partition, to minimize the span of allocated swap.
804 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
805 * case, just handled by scan_swap_map_try_ssd_cluster() above.
807 scan_base
= offset
= si
->lowest_bit
;
808 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
810 /* Locate the first empty (unaligned) cluster */
811 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
812 if (si
->swap_map
[offset
])
813 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
814 else if (offset
== last_in_cluster
) {
815 spin_lock(&si
->lock
);
816 offset
-= SWAPFILE_CLUSTER
- 1;
817 si
->cluster_next
= offset
;
818 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
821 if (unlikely(--latency_ration
< 0)) {
823 latency_ration
= LATENCY_LIMIT
;
828 spin_lock(&si
->lock
);
829 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
833 if (si
->cluster_info
) {
834 while (scan_swap_map_ssd_cluster_conflict(si
, offset
)) {
835 /* take a break if we already got some slots */
838 if (!scan_swap_map_try_ssd_cluster(si
, &offset
,
843 if (!(si
->flags
& SWP_WRITEOK
))
845 if (!si
->highest_bit
)
847 if (offset
> si
->highest_bit
)
848 scan_base
= offset
= si
->lowest_bit
;
850 ci
= lock_cluster(si
, offset
);
851 /* reuse swap entry of cache-only swap if not busy. */
852 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
855 spin_unlock(&si
->lock
);
856 swap_was_freed
= __try_to_reclaim_swap(si
, offset
, TTRS_ANYWAY
);
857 spin_lock(&si
->lock
);
858 /* entry was freed successfully, try to use this again */
861 goto scan
; /* check next one */
864 if (si
->swap_map
[offset
]) {
871 si
->swap_map
[offset
] = usage
;
872 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
875 swap_range_alloc(si
, offset
, 1);
876 slots
[n_ret
++] = swp_entry(si
->type
, offset
);
878 /* got enough slots or reach max slots? */
879 if ((n_ret
== nr
) || (offset
>= si
->highest_bit
))
882 /* search for next available slot */
884 /* time to take a break? */
885 if (unlikely(--latency_ration
< 0)) {
888 spin_unlock(&si
->lock
);
890 spin_lock(&si
->lock
);
891 latency_ration
= LATENCY_LIMIT
;
894 /* try to get more slots in cluster */
895 if (si
->cluster_info
) {
896 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
898 } else if (si
->cluster_nr
&& !si
->swap_map
[++offset
]) {
899 /* non-ssd case, still more slots in cluster? */
905 * Even if there's no free clusters available (fragmented),
906 * try to scan a little more quickly with lock held unless we
907 * have scanned too many slots already.
910 unsigned long scan_limit
;
912 if (offset
< scan_base
)
913 scan_limit
= scan_base
;
915 scan_limit
= si
->highest_bit
;
916 for (; offset
<= scan_limit
&& --latency_ration
> 0;
918 if (!si
->swap_map
[offset
])
924 set_cluster_next(si
, offset
+ 1);
925 si
->flags
-= SWP_SCANNING
;
929 spin_unlock(&si
->lock
);
930 while (++offset
<= si
->highest_bit
) {
931 if (!si
->swap_map
[offset
]) {
932 spin_lock(&si
->lock
);
935 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
936 spin_lock(&si
->lock
);
939 if (unlikely(--latency_ration
< 0)) {
941 latency_ration
= LATENCY_LIMIT
;
945 offset
= si
->lowest_bit
;
946 while (offset
< scan_base
) {
947 if (!si
->swap_map
[offset
]) {
948 spin_lock(&si
->lock
);
951 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
952 spin_lock(&si
->lock
);
955 if (unlikely(--latency_ration
< 0)) {
957 latency_ration
= LATENCY_LIMIT
;
962 spin_lock(&si
->lock
);
965 si
->flags
-= SWP_SCANNING
;
969 static int swap_alloc_cluster(struct swap_info_struct
*si
, swp_entry_t
*slot
)
972 struct swap_cluster_info
*ci
;
973 unsigned long offset
, i
;
977 * Should not even be attempting cluster allocations when huge
978 * page swap is disabled. Warn and fail the allocation.
980 if (!IS_ENABLED(CONFIG_THP_SWAP
)) {
985 if (cluster_list_empty(&si
->free_clusters
))
988 idx
= cluster_list_first(&si
->free_clusters
);
989 offset
= idx
* SWAPFILE_CLUSTER
;
990 ci
= lock_cluster(si
, offset
);
991 alloc_cluster(si
, idx
);
992 cluster_set_count_flag(ci
, SWAPFILE_CLUSTER
, CLUSTER_FLAG_HUGE
);
994 map
= si
->swap_map
+ offset
;
995 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++)
996 map
[i
] = SWAP_HAS_CACHE
;
998 swap_range_alloc(si
, offset
, SWAPFILE_CLUSTER
);
999 *slot
= swp_entry(si
->type
, offset
);
1004 static void swap_free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
1006 unsigned long offset
= idx
* SWAPFILE_CLUSTER
;
1007 struct swap_cluster_info
*ci
;
1009 ci
= lock_cluster(si
, offset
);
1010 memset(si
->swap_map
+ offset
, 0, SWAPFILE_CLUSTER
);
1011 cluster_set_count_flag(ci
, 0, 0);
1012 free_cluster(si
, idx
);
1014 swap_range_free(si
, offset
, SWAPFILE_CLUSTER
);
1017 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
1018 unsigned char usage
)
1023 n_ret
= scan_swap_map_slots(si
, usage
, 1, &entry
);
1026 return swp_offset(entry
);
1032 int get_swap_pages(int n_goal
, swp_entry_t swp_entries
[], int entry_size
)
1034 unsigned long size
= swap_entry_size(entry_size
);
1035 struct swap_info_struct
*si
, *next
;
1040 /* Only single cluster request supported */
1041 WARN_ON_ONCE(n_goal
> 1 && size
== SWAPFILE_CLUSTER
);
1043 avail_pgs
= atomic_long_read(&nr_swap_pages
) / size
;
1047 n_goal
= min3((long)n_goal
, (long)SWAP_BATCH
, avail_pgs
);
1049 atomic_long_sub(n_goal
* size
, &nr_swap_pages
);
1051 spin_lock(&swap_avail_lock
);
1054 node
= numa_node_id();
1055 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[node
], avail_lists
[node
]) {
1056 /* requeue si to after same-priority siblings */
1057 plist_requeue(&si
->avail_lists
[node
], &swap_avail_heads
[node
]);
1058 spin_unlock(&swap_avail_lock
);
1059 spin_lock(&si
->lock
);
1060 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
1061 spin_lock(&swap_avail_lock
);
1062 if (plist_node_empty(&si
->avail_lists
[node
])) {
1063 spin_unlock(&si
->lock
);
1066 WARN(!si
->highest_bit
,
1067 "swap_info %d in list but !highest_bit\n",
1069 WARN(!(si
->flags
& SWP_WRITEOK
),
1070 "swap_info %d in list but !SWP_WRITEOK\n",
1072 __del_from_avail_list(si
);
1073 spin_unlock(&si
->lock
);
1076 if (size
== SWAPFILE_CLUSTER
) {
1077 if (!(si
->flags
& SWP_FS
))
1078 n_ret
= swap_alloc_cluster(si
, swp_entries
);
1080 n_ret
= scan_swap_map_slots(si
, SWAP_HAS_CACHE
,
1081 n_goal
, swp_entries
);
1082 spin_unlock(&si
->lock
);
1083 if (n_ret
|| size
== SWAPFILE_CLUSTER
)
1085 pr_debug("scan_swap_map of si %d failed to find offset\n",
1088 spin_lock(&swap_avail_lock
);
1091 * if we got here, it's likely that si was almost full before,
1092 * and since scan_swap_map() can drop the si->lock, multiple
1093 * callers probably all tried to get a page from the same si
1094 * and it filled up before we could get one; or, the si filled
1095 * up between us dropping swap_avail_lock and taking si->lock.
1096 * Since we dropped the swap_avail_lock, the swap_avail_head
1097 * list may have been modified; so if next is still in the
1098 * swap_avail_head list then try it, otherwise start over
1099 * if we have not gotten any slots.
1101 if (plist_node_empty(&next
->avail_lists
[node
]))
1105 spin_unlock(&swap_avail_lock
);
1109 atomic_long_add((long)(n_goal
- n_ret
) * size
,
1115 /* The only caller of this function is now suspend routine */
1116 swp_entry_t
get_swap_page_of_type(int type
)
1118 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1124 spin_lock(&si
->lock
);
1125 if (si
->flags
& SWP_WRITEOK
) {
1126 atomic_long_dec(&nr_swap_pages
);
1127 /* This is called for allocating swap entry, not cache */
1128 offset
= scan_swap_map(si
, 1);
1130 spin_unlock(&si
->lock
);
1131 return swp_entry(type
, offset
);
1133 atomic_long_inc(&nr_swap_pages
);
1135 spin_unlock(&si
->lock
);
1137 return (swp_entry_t
) {0};
1140 static struct swap_info_struct
*__swap_info_get(swp_entry_t entry
)
1142 struct swap_info_struct
*p
;
1143 unsigned long offset
;
1147 p
= swp_swap_info(entry
);
1150 if (!(p
->flags
& SWP_USED
))
1152 offset
= swp_offset(entry
);
1153 if (offset
>= p
->max
)
1158 pr_err("swap_info_get: %s%08lx\n", Bad_offset
, entry
.val
);
1161 pr_err("swap_info_get: %s%08lx\n", Unused_file
, entry
.val
);
1164 pr_err("swap_info_get: %s%08lx\n", Bad_file
, entry
.val
);
1169 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
1171 struct swap_info_struct
*p
;
1173 p
= __swap_info_get(entry
);
1176 if (!p
->swap_map
[swp_offset(entry
)])
1181 pr_err("swap_info_get: %s%08lx\n", Unused_offset
, entry
.val
);
1187 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
1189 struct swap_info_struct
*p
;
1191 p
= _swap_info_get(entry
);
1193 spin_lock(&p
->lock
);
1197 static struct swap_info_struct
*swap_info_get_cont(swp_entry_t entry
,
1198 struct swap_info_struct
*q
)
1200 struct swap_info_struct
*p
;
1202 p
= _swap_info_get(entry
);
1206 spin_unlock(&q
->lock
);
1208 spin_lock(&p
->lock
);
1213 static unsigned char __swap_entry_free_locked(struct swap_info_struct
*p
,
1214 unsigned long offset
,
1215 unsigned char usage
)
1217 unsigned char count
;
1218 unsigned char has_cache
;
1220 count
= p
->swap_map
[offset
];
1222 has_cache
= count
& SWAP_HAS_CACHE
;
1223 count
&= ~SWAP_HAS_CACHE
;
1225 if (usage
== SWAP_HAS_CACHE
) {
1226 VM_BUG_ON(!has_cache
);
1228 } else if (count
== SWAP_MAP_SHMEM
) {
1230 * Or we could insist on shmem.c using a special
1231 * swap_shmem_free() and free_shmem_swap_and_cache()...
1234 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
1235 if (count
== COUNT_CONTINUED
) {
1236 if (swap_count_continued(p
, offset
, count
))
1237 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
1239 count
= SWAP_MAP_MAX
;
1244 usage
= count
| has_cache
;
1245 p
->swap_map
[offset
] = usage
? : SWAP_HAS_CACHE
;
1251 * Check whether swap entry is valid in the swap device. If so,
1252 * return pointer to swap_info_struct, and keep the swap entry valid
1253 * via preventing the swap device from being swapoff, until
1254 * put_swap_device() is called. Otherwise return NULL.
1256 * The entirety of the RCU read critical section must come before the
1257 * return from or after the call to synchronize_rcu() in
1258 * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is
1259 * true, the si->map, si->cluster_info, etc. must be valid in the
1262 * Notice that swapoff or swapoff+swapon can still happen before the
1263 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1264 * in put_swap_device() if there isn't any other way to prevent
1265 * swapoff, such as page lock, page table lock, etc. The caller must
1266 * be prepared for that. For example, the following situation is
1271 * ... swapoff+swapon
1272 * __read_swap_cache_async()
1273 * swapcache_prepare()
1274 * __swap_duplicate()
1276 * // verify PTE not changed
1278 * In __swap_duplicate(), the swap_map need to be checked before
1279 * changing partly because the specified swap entry may be for another
1280 * swap device which has been swapoff. And in do_swap_page(), after
1281 * the page is read from the swap device, the PTE is verified not
1282 * changed with the page table locked to check whether the swap device
1283 * has been swapoff or swapoff+swapon.
1285 struct swap_info_struct
*get_swap_device(swp_entry_t entry
)
1287 struct swap_info_struct
*si
;
1288 unsigned long offset
;
1292 si
= swp_swap_info(entry
);
1297 if (!(si
->flags
& SWP_VALID
))
1299 offset
= swp_offset(entry
);
1300 if (offset
>= si
->max
)
1305 pr_err("%s: %s%08lx\n", __func__
, Bad_file
, entry
.val
);
1313 static unsigned char __swap_entry_free(struct swap_info_struct
*p
,
1316 struct swap_cluster_info
*ci
;
1317 unsigned long offset
= swp_offset(entry
);
1318 unsigned char usage
;
1320 ci
= lock_cluster_or_swap_info(p
, offset
);
1321 usage
= __swap_entry_free_locked(p
, offset
, 1);
1322 unlock_cluster_or_swap_info(p
, ci
);
1324 free_swap_slot(entry
);
1329 static void swap_entry_free(struct swap_info_struct
*p
, swp_entry_t entry
)
1331 struct swap_cluster_info
*ci
;
1332 unsigned long offset
= swp_offset(entry
);
1333 unsigned char count
;
1335 ci
= lock_cluster(p
, offset
);
1336 count
= p
->swap_map
[offset
];
1337 VM_BUG_ON(count
!= SWAP_HAS_CACHE
);
1338 p
->swap_map
[offset
] = 0;
1339 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
1342 mem_cgroup_uncharge_swap(entry
, 1);
1343 swap_range_free(p
, offset
, 1);
1347 * Caller has made sure that the swap device corresponding to entry
1348 * is still around or has not been recycled.
1350 void swap_free(swp_entry_t entry
)
1352 struct swap_info_struct
*p
;
1354 p
= _swap_info_get(entry
);
1356 __swap_entry_free(p
, entry
);
1360 * Called after dropping swapcache to decrease refcnt to swap entries.
1362 void put_swap_page(struct page
*page
, swp_entry_t entry
)
1364 unsigned long offset
= swp_offset(entry
);
1365 unsigned long idx
= offset
/ SWAPFILE_CLUSTER
;
1366 struct swap_cluster_info
*ci
;
1367 struct swap_info_struct
*si
;
1369 unsigned int i
, free_entries
= 0;
1371 int size
= swap_entry_size(hpage_nr_pages(page
));
1373 si
= _swap_info_get(entry
);
1377 ci
= lock_cluster_or_swap_info(si
, offset
);
1378 if (size
== SWAPFILE_CLUSTER
) {
1379 VM_BUG_ON(!cluster_is_huge(ci
));
1380 map
= si
->swap_map
+ offset
;
1381 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1383 VM_BUG_ON(!(val
& SWAP_HAS_CACHE
));
1384 if (val
== SWAP_HAS_CACHE
)
1387 cluster_clear_huge(ci
);
1388 if (free_entries
== SWAPFILE_CLUSTER
) {
1389 unlock_cluster_or_swap_info(si
, ci
);
1390 spin_lock(&si
->lock
);
1391 mem_cgroup_uncharge_swap(entry
, SWAPFILE_CLUSTER
);
1392 swap_free_cluster(si
, idx
);
1393 spin_unlock(&si
->lock
);
1397 for (i
= 0; i
< size
; i
++, entry
.val
++) {
1398 if (!__swap_entry_free_locked(si
, offset
+ i
, SWAP_HAS_CACHE
)) {
1399 unlock_cluster_or_swap_info(si
, ci
);
1400 free_swap_slot(entry
);
1403 lock_cluster_or_swap_info(si
, offset
);
1406 unlock_cluster_or_swap_info(si
, ci
);
1409 #ifdef CONFIG_THP_SWAP
1410 int split_swap_cluster(swp_entry_t entry
)
1412 struct swap_info_struct
*si
;
1413 struct swap_cluster_info
*ci
;
1414 unsigned long offset
= swp_offset(entry
);
1416 si
= _swap_info_get(entry
);
1419 ci
= lock_cluster(si
, offset
);
1420 cluster_clear_huge(ci
);
1426 static int swp_entry_cmp(const void *ent1
, const void *ent2
)
1428 const swp_entry_t
*e1
= ent1
, *e2
= ent2
;
1430 return (int)swp_type(*e1
) - (int)swp_type(*e2
);
1433 void swapcache_free_entries(swp_entry_t
*entries
, int n
)
1435 struct swap_info_struct
*p
, *prev
;
1445 * Sort swap entries by swap device, so each lock is only taken once.
1446 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1447 * so low that it isn't necessary to optimize further.
1449 if (nr_swapfiles
> 1)
1450 sort(entries
, n
, sizeof(entries
[0]), swp_entry_cmp
, NULL
);
1451 for (i
= 0; i
< n
; ++i
) {
1452 p
= swap_info_get_cont(entries
[i
], prev
);
1454 swap_entry_free(p
, entries
[i
]);
1458 spin_unlock(&p
->lock
);
1462 * How many references to page are currently swapped out?
1463 * This does not give an exact answer when swap count is continued,
1464 * but does include the high COUNT_CONTINUED flag to allow for that.
1466 int page_swapcount(struct page
*page
)
1469 struct swap_info_struct
*p
;
1470 struct swap_cluster_info
*ci
;
1472 unsigned long offset
;
1474 entry
.val
= page_private(page
);
1475 p
= _swap_info_get(entry
);
1477 offset
= swp_offset(entry
);
1478 ci
= lock_cluster_or_swap_info(p
, offset
);
1479 count
= swap_count(p
->swap_map
[offset
]);
1480 unlock_cluster_or_swap_info(p
, ci
);
1485 int __swap_count(swp_entry_t entry
)
1487 struct swap_info_struct
*si
;
1488 pgoff_t offset
= swp_offset(entry
);
1491 si
= get_swap_device(entry
);
1493 count
= swap_count(si
->swap_map
[offset
]);
1494 put_swap_device(si
);
1499 static int swap_swapcount(struct swap_info_struct
*si
, swp_entry_t entry
)
1502 pgoff_t offset
= swp_offset(entry
);
1503 struct swap_cluster_info
*ci
;
1505 ci
= lock_cluster_or_swap_info(si
, offset
);
1506 count
= swap_count(si
->swap_map
[offset
]);
1507 unlock_cluster_or_swap_info(si
, ci
);
1512 * How many references to @entry are currently swapped out?
1513 * This does not give an exact answer when swap count is continued,
1514 * but does include the high COUNT_CONTINUED flag to allow for that.
1516 int __swp_swapcount(swp_entry_t entry
)
1519 struct swap_info_struct
*si
;
1521 si
= get_swap_device(entry
);
1523 count
= swap_swapcount(si
, entry
);
1524 put_swap_device(si
);
1530 * How many references to @entry are currently swapped out?
1531 * This considers COUNT_CONTINUED so it returns exact answer.
1533 int swp_swapcount(swp_entry_t entry
)
1535 int count
, tmp_count
, n
;
1536 struct swap_info_struct
*p
;
1537 struct swap_cluster_info
*ci
;
1542 p
= _swap_info_get(entry
);
1546 offset
= swp_offset(entry
);
1548 ci
= lock_cluster_or_swap_info(p
, offset
);
1550 count
= swap_count(p
->swap_map
[offset
]);
1551 if (!(count
& COUNT_CONTINUED
))
1554 count
&= ~COUNT_CONTINUED
;
1555 n
= SWAP_MAP_MAX
+ 1;
1557 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1558 offset
&= ~PAGE_MASK
;
1559 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1562 page
= list_next_entry(page
, lru
);
1563 map
= kmap_atomic(page
);
1564 tmp_count
= map
[offset
];
1567 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1568 n
*= (SWAP_CONT_MAX
+ 1);
1569 } while (tmp_count
& COUNT_CONTINUED
);
1571 unlock_cluster_or_swap_info(p
, ci
);
1575 static bool swap_page_trans_huge_swapped(struct swap_info_struct
*si
,
1578 struct swap_cluster_info
*ci
;
1579 unsigned char *map
= si
->swap_map
;
1580 unsigned long roffset
= swp_offset(entry
);
1581 unsigned long offset
= round_down(roffset
, SWAPFILE_CLUSTER
);
1585 ci
= lock_cluster_or_swap_info(si
, offset
);
1586 if (!ci
|| !cluster_is_huge(ci
)) {
1587 if (swap_count(map
[roffset
]))
1591 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1592 if (swap_count(map
[offset
+ i
])) {
1598 unlock_cluster_or_swap_info(si
, ci
);
1602 static bool page_swapped(struct page
*page
)
1605 struct swap_info_struct
*si
;
1607 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
)))
1608 return page_swapcount(page
) != 0;
1610 page
= compound_head(page
);
1611 entry
.val
= page_private(page
);
1612 si
= _swap_info_get(entry
);
1614 return swap_page_trans_huge_swapped(si
, entry
);
1618 static int page_trans_huge_map_swapcount(struct page
*page
, int *total_mapcount
,
1619 int *total_swapcount
)
1621 int i
, map_swapcount
, _total_mapcount
, _total_swapcount
;
1622 unsigned long offset
= 0;
1623 struct swap_info_struct
*si
;
1624 struct swap_cluster_info
*ci
= NULL
;
1625 unsigned char *map
= NULL
;
1626 int mapcount
, swapcount
= 0;
1628 /* hugetlbfs shouldn't call it */
1629 VM_BUG_ON_PAGE(PageHuge(page
), page
);
1631 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
))) {
1632 mapcount
= page_trans_huge_mapcount(page
, total_mapcount
);
1633 if (PageSwapCache(page
))
1634 swapcount
= page_swapcount(page
);
1635 if (total_swapcount
)
1636 *total_swapcount
= swapcount
;
1637 return mapcount
+ swapcount
;
1640 page
= compound_head(page
);
1642 _total_mapcount
= _total_swapcount
= map_swapcount
= 0;
1643 if (PageSwapCache(page
)) {
1646 entry
.val
= page_private(page
);
1647 si
= _swap_info_get(entry
);
1650 offset
= swp_offset(entry
);
1654 ci
= lock_cluster(si
, offset
);
1655 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
1656 mapcount
= atomic_read(&page
[i
]._mapcount
) + 1;
1657 _total_mapcount
+= mapcount
;
1659 swapcount
= swap_count(map
[offset
+ i
]);
1660 _total_swapcount
+= swapcount
;
1662 map_swapcount
= max(map_swapcount
, mapcount
+ swapcount
);
1665 if (PageDoubleMap(page
)) {
1667 _total_mapcount
-= HPAGE_PMD_NR
;
1669 mapcount
= compound_mapcount(page
);
1670 map_swapcount
+= mapcount
;
1671 _total_mapcount
+= mapcount
;
1673 *total_mapcount
= _total_mapcount
;
1674 if (total_swapcount
)
1675 *total_swapcount
= _total_swapcount
;
1677 return map_swapcount
;
1681 * We can write to an anon page without COW if there are no other references
1682 * to it. And as a side-effect, free up its swap: because the old content
1683 * on disk will never be read, and seeking back there to write new content
1684 * later would only waste time away from clustering.
1686 * NOTE: total_map_swapcount should not be relied upon by the caller if
1687 * reuse_swap_page() returns false, but it may be always overwritten
1688 * (see the other implementation for CONFIG_SWAP=n).
1690 bool reuse_swap_page(struct page
*page
, int *total_map_swapcount
)
1692 int count
, total_mapcount
, total_swapcount
;
1694 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1695 if (unlikely(PageKsm(page
)))
1697 count
= page_trans_huge_map_swapcount(page
, &total_mapcount
,
1699 if (total_map_swapcount
)
1700 *total_map_swapcount
= total_mapcount
+ total_swapcount
;
1701 if (count
== 1 && PageSwapCache(page
) &&
1702 (likely(!PageTransCompound(page
)) ||
1703 /* The remaining swap count will be freed soon */
1704 total_swapcount
== page_swapcount(page
))) {
1705 if (!PageWriteback(page
)) {
1706 page
= compound_head(page
);
1707 delete_from_swap_cache(page
);
1711 struct swap_info_struct
*p
;
1713 entry
.val
= page_private(page
);
1714 p
= swap_info_get(entry
);
1715 if (p
->flags
& SWP_STABLE_WRITES
) {
1716 spin_unlock(&p
->lock
);
1719 spin_unlock(&p
->lock
);
1727 * If swap is getting full, or if there are no more mappings of this page,
1728 * then try_to_free_swap is called to free its swap space.
1730 int try_to_free_swap(struct page
*page
)
1732 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1734 if (!PageSwapCache(page
))
1736 if (PageWriteback(page
))
1738 if (page_swapped(page
))
1742 * Once hibernation has begun to create its image of memory,
1743 * there's a danger that one of the calls to try_to_free_swap()
1744 * - most probably a call from __try_to_reclaim_swap() while
1745 * hibernation is allocating its own swap pages for the image,
1746 * but conceivably even a call from memory reclaim - will free
1747 * the swap from a page which has already been recorded in the
1748 * image as a clean swapcache page, and then reuse its swap for
1749 * another page of the image. On waking from hibernation, the
1750 * original page might be freed under memory pressure, then
1751 * later read back in from swap, now with the wrong data.
1753 * Hibernation suspends storage while it is writing the image
1754 * to disk so check that here.
1756 if (pm_suspended_storage())
1759 page
= compound_head(page
);
1760 delete_from_swap_cache(page
);
1766 * Free the swap entry like above, but also try to
1767 * free the page cache entry if it is the last user.
1769 int free_swap_and_cache(swp_entry_t entry
)
1771 struct swap_info_struct
*p
;
1772 unsigned char count
;
1774 if (non_swap_entry(entry
))
1777 p
= _swap_info_get(entry
);
1779 count
= __swap_entry_free(p
, entry
);
1780 if (count
== SWAP_HAS_CACHE
&&
1781 !swap_page_trans_huge_swapped(p
, entry
))
1782 __try_to_reclaim_swap(p
, swp_offset(entry
),
1783 TTRS_UNMAPPED
| TTRS_FULL
);
1788 #ifdef CONFIG_HIBERNATION
1790 * Find the swap type that corresponds to given device (if any).
1792 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1793 * from 0, in which the swap header is expected to be located.
1795 * This is needed for the suspend to disk (aka swsusp).
1797 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
1799 struct block_device
*bdev
= NULL
;
1803 bdev
= bdget(device
);
1805 spin_lock(&swap_lock
);
1806 for (type
= 0; type
< nr_swapfiles
; type
++) {
1807 struct swap_info_struct
*sis
= swap_info
[type
];
1809 if (!(sis
->flags
& SWP_WRITEOK
))
1814 *bdev_p
= bdgrab(sis
->bdev
);
1816 spin_unlock(&swap_lock
);
1819 if (bdev
== sis
->bdev
) {
1820 struct swap_extent
*se
= first_se(sis
);
1822 if (se
->start_block
== offset
) {
1824 *bdev_p
= bdgrab(sis
->bdev
);
1826 spin_unlock(&swap_lock
);
1832 spin_unlock(&swap_lock
);
1840 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1841 * corresponding to given index in swap_info (swap type).
1843 sector_t
swapdev_block(int type
, pgoff_t offset
)
1845 struct block_device
*bdev
;
1846 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1848 if (!si
|| !(si
->flags
& SWP_WRITEOK
))
1850 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
1854 * Return either the total number of swap pages of given type, or the number
1855 * of free pages of that type (depending on @free)
1857 * This is needed for software suspend
1859 unsigned int count_swap_pages(int type
, int free
)
1863 spin_lock(&swap_lock
);
1864 if ((unsigned int)type
< nr_swapfiles
) {
1865 struct swap_info_struct
*sis
= swap_info
[type
];
1867 spin_lock(&sis
->lock
);
1868 if (sis
->flags
& SWP_WRITEOK
) {
1871 n
-= sis
->inuse_pages
;
1873 spin_unlock(&sis
->lock
);
1875 spin_unlock(&swap_lock
);
1878 #endif /* CONFIG_HIBERNATION */
1880 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1882 return pte_same(pte_swp_clear_soft_dirty(pte
), swp_pte
);
1886 * No need to decide whether this PTE shares the swap entry with others,
1887 * just let do_wp_page work it out if a write is requested later - to
1888 * force COW, vm_page_prot omits write permission from any private vma.
1890 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1891 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1893 struct page
*swapcache
;
1899 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1900 if (unlikely(!page
))
1903 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1904 if (unlikely(!pte_same_as_swp(*pte
, swp_entry_to_pte(entry
)))) {
1909 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1910 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1912 set_pte_at(vma
->vm_mm
, addr
, pte
,
1913 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1914 if (page
== swapcache
) {
1915 page_add_anon_rmap(page
, vma
, addr
, false);
1916 } else { /* ksm created a completely new copy */
1917 page_add_new_anon_rmap(page
, vma
, addr
, false);
1918 lru_cache_add_active_or_unevictable(page
, vma
);
1922 * Move the page to the active list so it is not
1923 * immediately swapped out again after swapon.
1925 activate_page(page
);
1927 pte_unmap_unlock(pte
, ptl
);
1928 if (page
!= swapcache
) {
1935 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1936 unsigned long addr
, unsigned long end
,
1937 unsigned int type
, bool frontswap
,
1938 unsigned long *fs_pages_to_unuse
)
1943 struct swap_info_struct
*si
;
1944 unsigned long offset
;
1946 volatile unsigned char *swap_map
;
1948 si
= swap_info
[type
];
1949 pte
= pte_offset_map(pmd
, addr
);
1951 struct vm_fault vmf
;
1953 if (!is_swap_pte(*pte
))
1956 entry
= pte_to_swp_entry(*pte
);
1957 if (swp_type(entry
) != type
)
1960 offset
= swp_offset(entry
);
1961 if (frontswap
&& !frontswap_test(si
, offset
))
1965 swap_map
= &si
->swap_map
[offset
];
1966 page
= lookup_swap_cache(entry
, vma
, addr
);
1971 page
= swapin_readahead(entry
, GFP_HIGHUSER_MOVABLE
,
1975 if (*swap_map
== 0 || *swap_map
== SWAP_MAP_BAD
)
1981 wait_on_page_writeback(page
);
1982 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1989 try_to_free_swap(page
);
1993 if (*fs_pages_to_unuse
&& !--(*fs_pages_to_unuse
)) {
1994 ret
= FRONTSWAP_PAGES_UNUSED
;
1998 pte
= pte_offset_map(pmd
, addr
);
1999 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
2007 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
2008 unsigned long addr
, unsigned long end
,
2009 unsigned int type
, bool frontswap
,
2010 unsigned long *fs_pages_to_unuse
)
2016 pmd
= pmd_offset(pud
, addr
);
2019 next
= pmd_addr_end(addr
, end
);
2020 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
2022 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, type
,
2023 frontswap
, fs_pages_to_unuse
);
2026 } while (pmd
++, addr
= next
, addr
!= end
);
2030 static inline int unuse_pud_range(struct vm_area_struct
*vma
, p4d_t
*p4d
,
2031 unsigned long addr
, unsigned long end
,
2032 unsigned int type
, bool frontswap
,
2033 unsigned long *fs_pages_to_unuse
)
2039 pud
= pud_offset(p4d
, addr
);
2041 next
= pud_addr_end(addr
, end
);
2042 if (pud_none_or_clear_bad(pud
))
2044 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, type
,
2045 frontswap
, fs_pages_to_unuse
);
2048 } while (pud
++, addr
= next
, addr
!= end
);
2052 static inline int unuse_p4d_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
2053 unsigned long addr
, unsigned long end
,
2054 unsigned int type
, bool frontswap
,
2055 unsigned long *fs_pages_to_unuse
)
2061 p4d
= p4d_offset(pgd
, addr
);
2063 next
= p4d_addr_end(addr
, end
);
2064 if (p4d_none_or_clear_bad(p4d
))
2066 ret
= unuse_pud_range(vma
, p4d
, addr
, next
, type
,
2067 frontswap
, fs_pages_to_unuse
);
2070 } while (p4d
++, addr
= next
, addr
!= end
);
2074 static int unuse_vma(struct vm_area_struct
*vma
, unsigned int type
,
2075 bool frontswap
, unsigned long *fs_pages_to_unuse
)
2078 unsigned long addr
, end
, next
;
2081 addr
= vma
->vm_start
;
2084 pgd
= pgd_offset(vma
->vm_mm
, addr
);
2086 next
= pgd_addr_end(addr
, end
);
2087 if (pgd_none_or_clear_bad(pgd
))
2089 ret
= unuse_p4d_range(vma
, pgd
, addr
, next
, type
,
2090 frontswap
, fs_pages_to_unuse
);
2093 } while (pgd
++, addr
= next
, addr
!= end
);
2097 static int unuse_mm(struct mm_struct
*mm
, unsigned int type
,
2098 bool frontswap
, unsigned long *fs_pages_to_unuse
)
2100 struct vm_area_struct
*vma
;
2104 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2105 if (vma
->anon_vma
) {
2106 ret
= unuse_vma(vma
, type
, frontswap
,
2113 mmap_read_unlock(mm
);
2118 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2119 * from current position to next entry still in use. Return 0
2120 * if there are no inuse entries after prev till end of the map.
2122 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
2123 unsigned int prev
, bool frontswap
)
2126 unsigned char count
;
2129 * No need for swap_lock here: we're just looking
2130 * for whether an entry is in use, not modifying it; false
2131 * hits are okay, and sys_swapoff() has already prevented new
2132 * allocations from this area (while holding swap_lock).
2134 for (i
= prev
+ 1; i
< si
->max
; i
++) {
2135 count
= READ_ONCE(si
->swap_map
[i
]);
2136 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
2137 if (!frontswap
|| frontswap_test(si
, i
))
2139 if ((i
% LATENCY_LIMIT
) == 0)
2150 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2151 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2153 int try_to_unuse(unsigned int type
, bool frontswap
,
2154 unsigned long pages_to_unuse
)
2156 struct mm_struct
*prev_mm
;
2157 struct mm_struct
*mm
;
2158 struct list_head
*p
;
2160 struct swap_info_struct
*si
= swap_info
[type
];
2165 if (!READ_ONCE(si
->inuse_pages
))
2172 retval
= shmem_unuse(type
, frontswap
, &pages_to_unuse
);
2179 spin_lock(&mmlist_lock
);
2180 p
= &init_mm
.mmlist
;
2181 while (READ_ONCE(si
->inuse_pages
) &&
2182 !signal_pending(current
) &&
2183 (p
= p
->next
) != &init_mm
.mmlist
) {
2185 mm
= list_entry(p
, struct mm_struct
, mmlist
);
2186 if (!mmget_not_zero(mm
))
2188 spin_unlock(&mmlist_lock
);
2191 retval
= unuse_mm(mm
, type
, frontswap
, &pages_to_unuse
);
2199 * Make sure that we aren't completely killing
2200 * interactive performance.
2203 spin_lock(&mmlist_lock
);
2205 spin_unlock(&mmlist_lock
);
2210 while (READ_ONCE(si
->inuse_pages
) &&
2211 !signal_pending(current
) &&
2212 (i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
2214 entry
= swp_entry(type
, i
);
2215 page
= find_get_page(swap_address_space(entry
), i
);
2220 * It is conceivable that a racing task removed this page from
2221 * swap cache just before we acquired the page lock. The page
2222 * might even be back in swap cache on another swap area. But
2223 * that is okay, try_to_free_swap() only removes stale pages.
2226 wait_on_page_writeback(page
);
2227 try_to_free_swap(page
);
2232 * For frontswap, we just need to unuse pages_to_unuse, if
2233 * it was specified. Need not check frontswap again here as
2234 * we already zeroed out pages_to_unuse if not frontswap.
2236 if (pages_to_unuse
&& --pages_to_unuse
== 0)
2241 * Lets check again to see if there are still swap entries in the map.
2242 * If yes, we would need to do retry the unuse logic again.
2243 * Under global memory pressure, swap entries can be reinserted back
2244 * into process space after the mmlist loop above passes over them.
2246 * Limit the number of retries? No: when mmget_not_zero() above fails,
2247 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2248 * at its own independent pace; and even shmem_writepage() could have
2249 * been preempted after get_swap_page(), temporarily hiding that swap.
2250 * It's easy and robust (though cpu-intensive) just to keep retrying.
2252 if (READ_ONCE(si
->inuse_pages
)) {
2253 if (!signal_pending(current
))
2258 return (retval
== FRONTSWAP_PAGES_UNUSED
) ? 0 : retval
;
2262 * After a successful try_to_unuse, if no swap is now in use, we know
2263 * we can empty the mmlist. swap_lock must be held on entry and exit.
2264 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2265 * added to the mmlist just after page_duplicate - before would be racy.
2267 static void drain_mmlist(void)
2269 struct list_head
*p
, *next
;
2272 for (type
= 0; type
< nr_swapfiles
; type
++)
2273 if (swap_info
[type
]->inuse_pages
)
2275 spin_lock(&mmlist_lock
);
2276 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
2278 spin_unlock(&mmlist_lock
);
2282 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2283 * corresponds to page offset for the specified swap entry.
2284 * Note that the type of this function is sector_t, but it returns page offset
2285 * into the bdev, not sector offset.
2287 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
2289 struct swap_info_struct
*sis
;
2290 struct swap_extent
*se
;
2293 sis
= swp_swap_info(entry
);
2296 offset
= swp_offset(entry
);
2297 se
= offset_to_swap_extent(sis
, offset
);
2298 return se
->start_block
+ (offset
- se
->start_page
);
2302 * Returns the page offset into bdev for the specified page's swap entry.
2304 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
2307 entry
.val
= page_private(page
);
2308 return map_swap_entry(entry
, bdev
);
2312 * Free all of a swapdev's extent information
2314 static void destroy_swap_extents(struct swap_info_struct
*sis
)
2316 while (!RB_EMPTY_ROOT(&sis
->swap_extent_root
)) {
2317 struct rb_node
*rb
= sis
->swap_extent_root
.rb_node
;
2318 struct swap_extent
*se
= rb_entry(rb
, struct swap_extent
, rb_node
);
2320 rb_erase(rb
, &sis
->swap_extent_root
);
2324 if (sis
->flags
& SWP_ACTIVATED
) {
2325 struct file
*swap_file
= sis
->swap_file
;
2326 struct address_space
*mapping
= swap_file
->f_mapping
;
2328 sis
->flags
&= ~SWP_ACTIVATED
;
2329 if (mapping
->a_ops
->swap_deactivate
)
2330 mapping
->a_ops
->swap_deactivate(swap_file
);
2335 * Add a block range (and the corresponding page range) into this swapdev's
2338 * This function rather assumes that it is called in ascending page order.
2341 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
2342 unsigned long nr_pages
, sector_t start_block
)
2344 struct rb_node
**link
= &sis
->swap_extent_root
.rb_node
, *parent
= NULL
;
2345 struct swap_extent
*se
;
2346 struct swap_extent
*new_se
;
2349 * place the new node at the right most since the
2350 * function is called in ascending page order.
2354 link
= &parent
->rb_right
;
2358 se
= rb_entry(parent
, struct swap_extent
, rb_node
);
2359 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
2360 if (se
->start_block
+ se
->nr_pages
== start_block
) {
2362 se
->nr_pages
+= nr_pages
;
2367 /* No merge, insert a new extent. */
2368 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
2371 new_se
->start_page
= start_page
;
2372 new_se
->nr_pages
= nr_pages
;
2373 new_se
->start_block
= start_block
;
2375 rb_link_node(&new_se
->rb_node
, parent
, link
);
2376 rb_insert_color(&new_se
->rb_node
, &sis
->swap_extent_root
);
2379 EXPORT_SYMBOL_GPL(add_swap_extent
);
2382 * A `swap extent' is a simple thing which maps a contiguous range of pages
2383 * onto a contiguous range of disk blocks. An ordered list of swap extents
2384 * is built at swapon time and is then used at swap_writepage/swap_readpage
2385 * time for locating where on disk a page belongs.
2387 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2388 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2389 * swap files identically.
2391 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2392 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2393 * swapfiles are handled *identically* after swapon time.
2395 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2396 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2397 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2398 * requirements, they are simply tossed out - we will never use those blocks
2401 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2402 * prevents users from writing to the swap device, which will corrupt memory.
2404 * The amount of disk space which a single swap extent represents varies.
2405 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2406 * extents in the list. To avoid much list walking, we cache the previous
2407 * search location in `curr_swap_extent', and start new searches from there.
2408 * This is extremely effective. The average number of iterations in
2409 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2411 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
2413 struct file
*swap_file
= sis
->swap_file
;
2414 struct address_space
*mapping
= swap_file
->f_mapping
;
2415 struct inode
*inode
= mapping
->host
;
2418 if (S_ISBLK(inode
->i_mode
)) {
2419 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2424 if (mapping
->a_ops
->swap_activate
) {
2425 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
2427 sis
->flags
|= SWP_ACTIVATED
;
2429 sis
->flags
|= SWP_FS
;
2430 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2436 return generic_swapfile_activate(sis
, swap_file
, span
);
2439 static int swap_node(struct swap_info_struct
*p
)
2441 struct block_device
*bdev
;
2446 bdev
= p
->swap_file
->f_inode
->i_sb
->s_bdev
;
2448 return bdev
? bdev
->bd_disk
->node_id
: NUMA_NO_NODE
;
2451 static void setup_swap_info(struct swap_info_struct
*p
, int prio
,
2452 unsigned char *swap_map
,
2453 struct swap_cluster_info
*cluster_info
)
2460 p
->prio
= --least_priority
;
2462 * the plist prio is negated because plist ordering is
2463 * low-to-high, while swap ordering is high-to-low
2465 p
->list
.prio
= -p
->prio
;
2468 p
->avail_lists
[i
].prio
= -p
->prio
;
2470 if (swap_node(p
) == i
)
2471 p
->avail_lists
[i
].prio
= 1;
2473 p
->avail_lists
[i
].prio
= -p
->prio
;
2476 p
->swap_map
= swap_map
;
2477 p
->cluster_info
= cluster_info
;
2480 static void _enable_swap_info(struct swap_info_struct
*p
)
2482 p
->flags
|= SWP_WRITEOK
| SWP_VALID
;
2483 atomic_long_add(p
->pages
, &nr_swap_pages
);
2484 total_swap_pages
+= p
->pages
;
2486 assert_spin_locked(&swap_lock
);
2488 * both lists are plists, and thus priority ordered.
2489 * swap_active_head needs to be priority ordered for swapoff(),
2490 * which on removal of any swap_info_struct with an auto-assigned
2491 * (i.e. negative) priority increments the auto-assigned priority
2492 * of any lower-priority swap_info_structs.
2493 * swap_avail_head needs to be priority ordered for get_swap_page(),
2494 * which allocates swap pages from the highest available priority
2497 plist_add(&p
->list
, &swap_active_head
);
2498 add_to_avail_list(p
);
2501 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
2502 unsigned char *swap_map
,
2503 struct swap_cluster_info
*cluster_info
,
2504 unsigned long *frontswap_map
)
2506 frontswap_init(p
->type
, frontswap_map
);
2507 spin_lock(&swap_lock
);
2508 spin_lock(&p
->lock
);
2509 setup_swap_info(p
, prio
, swap_map
, cluster_info
);
2510 spin_unlock(&p
->lock
);
2511 spin_unlock(&swap_lock
);
2513 * Guarantee swap_map, cluster_info, etc. fields are valid
2514 * between get/put_swap_device() if SWP_VALID bit is set
2517 spin_lock(&swap_lock
);
2518 spin_lock(&p
->lock
);
2519 _enable_swap_info(p
);
2520 spin_unlock(&p
->lock
);
2521 spin_unlock(&swap_lock
);
2524 static void reinsert_swap_info(struct swap_info_struct
*p
)
2526 spin_lock(&swap_lock
);
2527 spin_lock(&p
->lock
);
2528 setup_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
2529 _enable_swap_info(p
);
2530 spin_unlock(&p
->lock
);
2531 spin_unlock(&swap_lock
);
2534 bool has_usable_swap(void)
2538 spin_lock(&swap_lock
);
2539 if (plist_head_empty(&swap_active_head
))
2541 spin_unlock(&swap_lock
);
2545 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
2547 struct swap_info_struct
*p
= NULL
;
2548 unsigned char *swap_map
;
2549 struct swap_cluster_info
*cluster_info
;
2550 unsigned long *frontswap_map
;
2551 struct file
*swap_file
, *victim
;
2552 struct address_space
*mapping
;
2553 struct inode
*inode
;
2554 struct filename
*pathname
;
2556 unsigned int old_block_size
;
2558 if (!capable(CAP_SYS_ADMIN
))
2561 BUG_ON(!current
->mm
);
2563 pathname
= getname(specialfile
);
2564 if (IS_ERR(pathname
))
2565 return PTR_ERR(pathname
);
2567 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
2568 err
= PTR_ERR(victim
);
2572 mapping
= victim
->f_mapping
;
2573 spin_lock(&swap_lock
);
2574 plist_for_each_entry(p
, &swap_active_head
, list
) {
2575 if (p
->flags
& SWP_WRITEOK
) {
2576 if (p
->swap_file
->f_mapping
== mapping
) {
2584 spin_unlock(&swap_lock
);
2587 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2588 vm_unacct_memory(p
->pages
);
2591 spin_unlock(&swap_lock
);
2594 del_from_avail_list(p
);
2595 spin_lock(&p
->lock
);
2597 struct swap_info_struct
*si
= p
;
2600 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2603 for_each_node(nid
) {
2604 if (si
->avail_lists
[nid
].prio
!= 1)
2605 si
->avail_lists
[nid
].prio
--;
2610 plist_del(&p
->list
, &swap_active_head
);
2611 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2612 total_swap_pages
-= p
->pages
;
2613 p
->flags
&= ~SWP_WRITEOK
;
2614 spin_unlock(&p
->lock
);
2615 spin_unlock(&swap_lock
);
2617 disable_swap_slots_cache_lock();
2619 set_current_oom_origin();
2620 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
2621 clear_current_oom_origin();
2624 /* re-insert swap space back into swap_list */
2625 reinsert_swap_info(p
);
2626 reenable_swap_slots_cache_unlock();
2630 reenable_swap_slots_cache_unlock();
2632 spin_lock(&swap_lock
);
2633 spin_lock(&p
->lock
);
2634 p
->flags
&= ~SWP_VALID
; /* mark swap device as invalid */
2635 spin_unlock(&p
->lock
);
2636 spin_unlock(&swap_lock
);
2638 * wait for swap operations protected by get/put_swap_device()
2643 flush_work(&p
->discard_work
);
2645 destroy_swap_extents(p
);
2646 if (p
->flags
& SWP_CONTINUED
)
2647 free_swap_count_continuations(p
);
2649 if (!p
->bdev
|| !blk_queue_nonrot(bdev_get_queue(p
->bdev
)))
2650 atomic_dec(&nr_rotate_swap
);
2652 mutex_lock(&swapon_mutex
);
2653 spin_lock(&swap_lock
);
2654 spin_lock(&p
->lock
);
2657 /* wait for anyone still in scan_swap_map */
2658 p
->highest_bit
= 0; /* cuts scans short */
2659 while (p
->flags
>= SWP_SCANNING
) {
2660 spin_unlock(&p
->lock
);
2661 spin_unlock(&swap_lock
);
2662 schedule_timeout_uninterruptible(1);
2663 spin_lock(&swap_lock
);
2664 spin_lock(&p
->lock
);
2667 swap_file
= p
->swap_file
;
2668 old_block_size
= p
->old_block_size
;
2669 p
->swap_file
= NULL
;
2671 swap_map
= p
->swap_map
;
2673 cluster_info
= p
->cluster_info
;
2674 p
->cluster_info
= NULL
;
2675 frontswap_map
= frontswap_map_get(p
);
2676 spin_unlock(&p
->lock
);
2677 spin_unlock(&swap_lock
);
2678 frontswap_invalidate_area(p
->type
);
2679 frontswap_map_set(p
, NULL
);
2680 mutex_unlock(&swapon_mutex
);
2681 free_percpu(p
->percpu_cluster
);
2682 p
->percpu_cluster
= NULL
;
2683 free_percpu(p
->cluster_next_cpu
);
2684 p
->cluster_next_cpu
= NULL
;
2686 kvfree(cluster_info
);
2687 kvfree(frontswap_map
);
2688 /* Destroy swap account information */
2689 swap_cgroup_swapoff(p
->type
);
2690 exit_swap_address_space(p
->type
);
2692 inode
= mapping
->host
;
2693 if (S_ISBLK(inode
->i_mode
)) {
2694 struct block_device
*bdev
= I_BDEV(inode
);
2696 set_blocksize(bdev
, old_block_size
);
2697 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2701 inode
->i_flags
&= ~S_SWAPFILE
;
2702 inode_unlock(inode
);
2703 filp_close(swap_file
, NULL
);
2706 * Clear the SWP_USED flag after all resources are freed so that swapon
2707 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2708 * not hold p->lock after we cleared its SWP_WRITEOK.
2710 spin_lock(&swap_lock
);
2712 spin_unlock(&swap_lock
);
2715 atomic_inc(&proc_poll_event
);
2716 wake_up_interruptible(&proc_poll_wait
);
2719 filp_close(victim
, NULL
);
2725 #ifdef CONFIG_PROC_FS
2726 static __poll_t
swaps_poll(struct file
*file
, poll_table
*wait
)
2728 struct seq_file
*seq
= file
->private_data
;
2730 poll_wait(file
, &proc_poll_wait
, wait
);
2732 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2733 seq
->poll_event
= atomic_read(&proc_poll_event
);
2734 return EPOLLIN
| EPOLLRDNORM
| EPOLLERR
| EPOLLPRI
;
2737 return EPOLLIN
| EPOLLRDNORM
;
2741 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2743 struct swap_info_struct
*si
;
2747 mutex_lock(&swapon_mutex
);
2750 return SEQ_START_TOKEN
;
2752 for (type
= 0; (si
= swap_type_to_swap_info(type
)); type
++) {
2753 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2762 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2764 struct swap_info_struct
*si
= v
;
2767 if (v
== SEQ_START_TOKEN
)
2770 type
= si
->type
+ 1;
2773 for (; (si
= swap_type_to_swap_info(type
)); type
++) {
2774 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2782 static void swap_stop(struct seq_file
*swap
, void *v
)
2784 mutex_unlock(&swapon_mutex
);
2787 static int swap_show(struct seq_file
*swap
, void *v
)
2789 struct swap_info_struct
*si
= v
;
2792 unsigned int bytes
, inuse
;
2794 if (si
== SEQ_START_TOKEN
) {
2795 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2799 bytes
= si
->pages
<< (PAGE_SHIFT
- 10);
2800 inuse
= si
->inuse_pages
<< (PAGE_SHIFT
- 10);
2802 file
= si
->swap_file
;
2803 len
= seq_file_path(swap
, file
, " \t\n\\");
2804 seq_printf(swap
, "%*s%s\t%u\t%s%u\t%s%d\n",
2805 len
< 40 ? 40 - len
: 1, " ",
2806 S_ISBLK(file_inode(file
)->i_mode
) ?
2807 "partition" : "file\t",
2808 bytes
, bytes
< 10000000 ? "\t" : "",
2809 inuse
, inuse
< 10000000 ? "\t" : "",
2814 static const struct seq_operations swaps_op
= {
2815 .start
= swap_start
,
2821 static int swaps_open(struct inode
*inode
, struct file
*file
)
2823 struct seq_file
*seq
;
2826 ret
= seq_open(file
, &swaps_op
);
2830 seq
= file
->private_data
;
2831 seq
->poll_event
= atomic_read(&proc_poll_event
);
2835 static const struct proc_ops swaps_proc_ops
= {
2836 .proc_flags
= PROC_ENTRY_PERMANENT
,
2837 .proc_open
= swaps_open
,
2838 .proc_read
= seq_read
,
2839 .proc_lseek
= seq_lseek
,
2840 .proc_release
= seq_release
,
2841 .proc_poll
= swaps_poll
,
2844 static int __init
procswaps_init(void)
2846 proc_create("swaps", 0, NULL
, &swaps_proc_ops
);
2849 __initcall(procswaps_init
);
2850 #endif /* CONFIG_PROC_FS */
2852 #ifdef MAX_SWAPFILES_CHECK
2853 static int __init
max_swapfiles_check(void)
2855 MAX_SWAPFILES_CHECK();
2858 late_initcall(max_swapfiles_check
);
2861 static struct swap_info_struct
*alloc_swap_info(void)
2863 struct swap_info_struct
*p
;
2867 p
= kvzalloc(struct_size(p
, avail_lists
, nr_node_ids
), GFP_KERNEL
);
2869 return ERR_PTR(-ENOMEM
);
2871 spin_lock(&swap_lock
);
2872 for (type
= 0; type
< nr_swapfiles
; type
++) {
2873 if (!(swap_info
[type
]->flags
& SWP_USED
))
2876 if (type
>= MAX_SWAPFILES
) {
2877 spin_unlock(&swap_lock
);
2879 return ERR_PTR(-EPERM
);
2881 if (type
>= nr_swapfiles
) {
2883 WRITE_ONCE(swap_info
[type
], p
);
2885 * Write swap_info[type] before nr_swapfiles, in case a
2886 * racing procfs swap_start() or swap_next() is reading them.
2887 * (We never shrink nr_swapfiles, we never free this entry.)
2890 WRITE_ONCE(nr_swapfiles
, nr_swapfiles
+ 1);
2893 p
= swap_info
[type
];
2895 * Do not memset this entry: a racing procfs swap_next()
2896 * would be relying on p->type to remain valid.
2899 p
->swap_extent_root
= RB_ROOT
;
2900 plist_node_init(&p
->list
, 0);
2902 plist_node_init(&p
->avail_lists
[i
], 0);
2903 p
->flags
= SWP_USED
;
2904 spin_unlock(&swap_lock
);
2905 spin_lock_init(&p
->lock
);
2906 spin_lock_init(&p
->cont_lock
);
2911 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2915 if (S_ISBLK(inode
->i_mode
)) {
2916 p
->bdev
= bdgrab(I_BDEV(inode
));
2917 error
= blkdev_get(p
->bdev
,
2918 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, p
);
2923 p
->old_block_size
= block_size(p
->bdev
);
2924 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2928 * Zoned block devices contain zones that have a sequential
2929 * write only restriction. Hence zoned block devices are not
2930 * suitable for swapping. Disallow them here.
2932 if (blk_queue_is_zoned(p
->bdev
->bd_queue
))
2934 p
->flags
|= SWP_BLKDEV
;
2935 } else if (S_ISREG(inode
->i_mode
)) {
2936 p
->bdev
= inode
->i_sb
->s_bdev
;
2944 * Find out how many pages are allowed for a single swap device. There
2945 * are two limiting factors:
2946 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2947 * 2) the number of bits in the swap pte, as defined by the different
2950 * In order to find the largest possible bit mask, a swap entry with
2951 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2952 * decoded to a swp_entry_t again, and finally the swap offset is
2955 * This will mask all the bits from the initial ~0UL mask that can't
2956 * be encoded in either the swp_entry_t or the architecture definition
2959 unsigned long generic_max_swapfile_size(void)
2961 return swp_offset(pte_to_swp_entry(
2962 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2965 /* Can be overridden by an architecture for additional checks. */
2966 __weak
unsigned long max_swapfile_size(void)
2968 return generic_max_swapfile_size();
2971 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2972 union swap_header
*swap_header
,
2973 struct inode
*inode
)
2976 unsigned long maxpages
;
2977 unsigned long swapfilepages
;
2978 unsigned long last_page
;
2980 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2981 pr_err("Unable to find swap-space signature\n");
2985 /* swap partition endianess hack... */
2986 if (swab32(swap_header
->info
.version
) == 1) {
2987 swab32s(&swap_header
->info
.version
);
2988 swab32s(&swap_header
->info
.last_page
);
2989 swab32s(&swap_header
->info
.nr_badpages
);
2990 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2992 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2993 swab32s(&swap_header
->info
.badpages
[i
]);
2995 /* Check the swap header's sub-version */
2996 if (swap_header
->info
.version
!= 1) {
2997 pr_warn("Unable to handle swap header version %d\n",
2998 swap_header
->info
.version
);
3003 p
->cluster_next
= 1;
3006 maxpages
= max_swapfile_size();
3007 last_page
= swap_header
->info
.last_page
;
3009 pr_warn("Empty swap-file\n");
3012 if (last_page
> maxpages
) {
3013 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
3014 maxpages
<< (PAGE_SHIFT
- 10),
3015 last_page
<< (PAGE_SHIFT
- 10));
3017 if (maxpages
> last_page
) {
3018 maxpages
= last_page
+ 1;
3019 /* p->max is an unsigned int: don't overflow it */
3020 if ((unsigned int)maxpages
== 0)
3021 maxpages
= UINT_MAX
;
3023 p
->highest_bit
= maxpages
- 1;
3027 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
3028 if (swapfilepages
&& maxpages
> swapfilepages
) {
3029 pr_warn("Swap area shorter than signature indicates\n");
3032 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
3034 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
3040 #define SWAP_CLUSTER_INFO_COLS \
3041 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3042 #define SWAP_CLUSTER_SPACE_COLS \
3043 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3044 #define SWAP_CLUSTER_COLS \
3045 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3047 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
3048 union swap_header
*swap_header
,
3049 unsigned char *swap_map
,
3050 struct swap_cluster_info
*cluster_info
,
3051 unsigned long maxpages
,
3055 unsigned int nr_good_pages
;
3057 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3058 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
3059 unsigned long i
, idx
;
3061 nr_good_pages
= maxpages
- 1; /* omit header page */
3063 cluster_list_init(&p
->free_clusters
);
3064 cluster_list_init(&p
->discard_clusters
);
3066 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
3067 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
3068 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
3070 if (page_nr
< maxpages
) {
3071 swap_map
[page_nr
] = SWAP_MAP_BAD
;
3074 * Haven't marked the cluster free yet, no list
3075 * operation involved
3077 inc_cluster_info_page(p
, cluster_info
, page_nr
);
3081 /* Haven't marked the cluster free yet, no list operation involved */
3082 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
3083 inc_cluster_info_page(p
, cluster_info
, i
);
3085 if (nr_good_pages
) {
3086 swap_map
[0] = SWAP_MAP_BAD
;
3088 * Not mark the cluster free yet, no list
3089 * operation involved
3091 inc_cluster_info_page(p
, cluster_info
, 0);
3093 p
->pages
= nr_good_pages
;
3094 nr_extents
= setup_swap_extents(p
, span
);
3097 nr_good_pages
= p
->pages
;
3099 if (!nr_good_pages
) {
3100 pr_warn("Empty swap-file\n");
3109 * Reduce false cache line sharing between cluster_info and
3110 * sharing same address space.
3112 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
3113 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
3114 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
3115 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
3116 if (idx
>= nr_clusters
)
3118 if (cluster_count(&cluster_info
[idx
]))
3120 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
3121 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
3129 * Helper to sys_swapon determining if a given swap
3130 * backing device queue supports DISCARD operations.
3132 static bool swap_discardable(struct swap_info_struct
*si
)
3134 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
3136 if (!q
|| !blk_queue_discard(q
))
3142 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
3144 struct swap_info_struct
*p
;
3145 struct filename
*name
;
3146 struct file
*swap_file
= NULL
;
3147 struct address_space
*mapping
;
3150 union swap_header
*swap_header
;
3153 unsigned long maxpages
;
3154 unsigned char *swap_map
= NULL
;
3155 struct swap_cluster_info
*cluster_info
= NULL
;
3156 unsigned long *frontswap_map
= NULL
;
3157 struct page
*page
= NULL
;
3158 struct inode
*inode
= NULL
;
3159 bool inced_nr_rotate_swap
= false;
3161 if (swap_flags
& ~SWAP_FLAGS_VALID
)
3164 if (!capable(CAP_SYS_ADMIN
))
3167 if (!swap_avail_heads
)
3170 p
= alloc_swap_info();
3174 INIT_WORK(&p
->discard_work
, swap_discard_work
);
3176 name
= getname(specialfile
);
3178 error
= PTR_ERR(name
);
3182 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
3183 if (IS_ERR(swap_file
)) {
3184 error
= PTR_ERR(swap_file
);
3189 p
->swap_file
= swap_file
;
3190 mapping
= swap_file
->f_mapping
;
3191 inode
= mapping
->host
;
3193 error
= claim_swapfile(p
, inode
);
3194 if (unlikely(error
))
3198 if (IS_SWAPFILE(inode
)) {
3200 goto bad_swap_unlock_inode
;
3204 * Read the swap header.
3206 if (!mapping
->a_ops
->readpage
) {
3208 goto bad_swap_unlock_inode
;
3210 page
= read_mapping_page(mapping
, 0, swap_file
);
3212 error
= PTR_ERR(page
);
3213 goto bad_swap_unlock_inode
;
3215 swap_header
= kmap(page
);
3217 maxpages
= read_swap_header(p
, swap_header
, inode
);
3218 if (unlikely(!maxpages
)) {
3220 goto bad_swap_unlock_inode
;
3223 /* OK, set up the swap map and apply the bad block list */
3224 swap_map
= vzalloc(maxpages
);
3227 goto bad_swap_unlock_inode
;
3230 if (bdi_cap_stable_pages_required(inode_to_bdi(inode
)))
3231 p
->flags
|= SWP_STABLE_WRITES
;
3233 if (bdi_cap_synchronous_io(inode_to_bdi(inode
)))
3234 p
->flags
|= SWP_SYNCHRONOUS_IO
;
3236 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
3238 unsigned long ci
, nr_cluster
;
3240 p
->flags
|= SWP_SOLIDSTATE
;
3241 p
->cluster_next_cpu
= alloc_percpu(unsigned int);
3242 if (!p
->cluster_next_cpu
) {
3244 goto bad_swap_unlock_inode
;
3247 * select a random position to start with to help wear leveling
3250 for_each_possible_cpu(cpu
) {
3251 per_cpu(*p
->cluster_next_cpu
, cpu
) =
3252 1 + prandom_u32_max(p
->highest_bit
);
3254 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3256 cluster_info
= kvcalloc(nr_cluster
, sizeof(*cluster_info
),
3258 if (!cluster_info
) {
3260 goto bad_swap_unlock_inode
;
3263 for (ci
= 0; ci
< nr_cluster
; ci
++)
3264 spin_lock_init(&((cluster_info
+ ci
)->lock
));
3266 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
3267 if (!p
->percpu_cluster
) {
3269 goto bad_swap_unlock_inode
;
3271 for_each_possible_cpu(cpu
) {
3272 struct percpu_cluster
*cluster
;
3273 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
3274 cluster_set_null(&cluster
->index
);
3277 atomic_inc(&nr_rotate_swap
);
3278 inced_nr_rotate_swap
= true;
3281 error
= swap_cgroup_swapon(p
->type
, maxpages
);
3283 goto bad_swap_unlock_inode
;
3285 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
3286 cluster_info
, maxpages
, &span
);
3287 if (unlikely(nr_extents
< 0)) {
3289 goto bad_swap_unlock_inode
;
3291 /* frontswap enabled? set up bit-per-page map for frontswap */
3292 if (IS_ENABLED(CONFIG_FRONTSWAP
))
3293 frontswap_map
= kvcalloc(BITS_TO_LONGS(maxpages
),
3297 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
3299 * When discard is enabled for swap with no particular
3300 * policy flagged, we set all swap discard flags here in
3301 * order to sustain backward compatibility with older
3302 * swapon(8) releases.
3304 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
3308 * By flagging sys_swapon, a sysadmin can tell us to
3309 * either do single-time area discards only, or to just
3310 * perform discards for released swap page-clusters.
3311 * Now it's time to adjust the p->flags accordingly.
3313 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
3314 p
->flags
&= ~SWP_PAGE_DISCARD
;
3315 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
3316 p
->flags
&= ~SWP_AREA_DISCARD
;
3318 /* issue a swapon-time discard if it's still required */
3319 if (p
->flags
& SWP_AREA_DISCARD
) {
3320 int err
= discard_swap(p
);
3322 pr_err("swapon: discard_swap(%p): %d\n",
3327 error
= init_swap_address_space(p
->type
, maxpages
);
3329 goto bad_swap_unlock_inode
;
3332 * Flush any pending IO and dirty mappings before we start using this
3335 inode
->i_flags
|= S_SWAPFILE
;
3336 error
= inode_drain_writes(inode
);
3338 inode
->i_flags
&= ~S_SWAPFILE
;
3339 goto bad_swap_unlock_inode
;
3342 mutex_lock(&swapon_mutex
);
3344 if (swap_flags
& SWAP_FLAG_PREFER
)
3346 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
3347 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
3349 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3350 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
3351 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
3352 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
3353 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
3354 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
3355 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
3356 (frontswap_map
) ? "FS" : "");
3358 mutex_unlock(&swapon_mutex
);
3359 atomic_inc(&proc_poll_event
);
3360 wake_up_interruptible(&proc_poll_wait
);
3364 bad_swap_unlock_inode
:
3365 inode_unlock(inode
);
3367 free_percpu(p
->percpu_cluster
);
3368 p
->percpu_cluster
= NULL
;
3369 free_percpu(p
->cluster_next_cpu
);
3370 p
->cluster_next_cpu
= NULL
;
3371 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
3372 set_blocksize(p
->bdev
, p
->old_block_size
);
3373 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
3376 destroy_swap_extents(p
);
3377 swap_cgroup_swapoff(p
->type
);
3378 spin_lock(&swap_lock
);
3379 p
->swap_file
= NULL
;
3381 spin_unlock(&swap_lock
);
3383 kvfree(cluster_info
);
3384 kvfree(frontswap_map
);
3385 if (inced_nr_rotate_swap
)
3386 atomic_dec(&nr_rotate_swap
);
3388 filp_close(swap_file
, NULL
);
3390 if (page
&& !IS_ERR(page
)) {
3397 inode_unlock(inode
);
3399 enable_swap_slots_cache();
3403 void si_swapinfo(struct sysinfo
*val
)
3406 unsigned long nr_to_be_unused
= 0;
3408 spin_lock(&swap_lock
);
3409 for (type
= 0; type
< nr_swapfiles
; type
++) {
3410 struct swap_info_struct
*si
= swap_info
[type
];
3412 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
3413 nr_to_be_unused
+= si
->inuse_pages
;
3415 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
3416 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
3417 spin_unlock(&swap_lock
);
3421 * Verify that a swap entry is valid and increment its swap map count.
3423 * Returns error code in following case.
3425 * - swp_entry is invalid -> EINVAL
3426 * - swp_entry is migration entry -> EINVAL
3427 * - swap-cache reference is requested but there is already one. -> EEXIST
3428 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3429 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3431 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
3433 struct swap_info_struct
*p
;
3434 struct swap_cluster_info
*ci
;
3435 unsigned long offset
;
3436 unsigned char count
;
3437 unsigned char has_cache
;
3440 p
= get_swap_device(entry
);
3444 offset
= swp_offset(entry
);
3445 ci
= lock_cluster_or_swap_info(p
, offset
);
3447 count
= p
->swap_map
[offset
];
3450 * swapin_readahead() doesn't check if a swap entry is valid, so the
3451 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3453 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
3458 has_cache
= count
& SWAP_HAS_CACHE
;
3459 count
&= ~SWAP_HAS_CACHE
;
3462 if (usage
== SWAP_HAS_CACHE
) {
3464 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3465 if (!has_cache
&& count
)
3466 has_cache
= SWAP_HAS_CACHE
;
3467 else if (has_cache
) /* someone else added cache */
3469 else /* no users remaining */
3472 } else if (count
|| has_cache
) {
3474 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
3476 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
3478 else if (swap_count_continued(p
, offset
, count
))
3479 count
= COUNT_CONTINUED
;
3483 err
= -ENOENT
; /* unused swap entry */
3485 p
->swap_map
[offset
] = count
| has_cache
;
3488 unlock_cluster_or_swap_info(p
, ci
);
3496 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3497 * (in which case its reference count is never incremented).
3499 void swap_shmem_alloc(swp_entry_t entry
)
3501 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
3505 * Increase reference count of swap entry by 1.
3506 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3507 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3508 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3509 * might occur if a page table entry has got corrupted.
3511 int swap_duplicate(swp_entry_t entry
)
3515 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
3516 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
3521 * @entry: swap entry for which we allocate swap cache.
3523 * Called when allocating swap cache for existing swap entry,
3524 * This can return error codes. Returns 0 at success.
3525 * -EEXIST means there is a swap cache.
3526 * Note: return code is different from swap_duplicate().
3528 int swapcache_prepare(swp_entry_t entry
)
3530 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
3533 struct swap_info_struct
*swp_swap_info(swp_entry_t entry
)
3535 return swap_type_to_swap_info(swp_type(entry
));
3538 struct swap_info_struct
*page_swap_info(struct page
*page
)
3540 swp_entry_t entry
= { .val
= page_private(page
) };
3541 return swp_swap_info(entry
);
3545 * out-of-line __page_file_ methods to avoid include hell.
3547 struct address_space
*__page_file_mapping(struct page
*page
)
3549 return page_swap_info(page
)->swap_file
->f_mapping
;
3551 EXPORT_SYMBOL_GPL(__page_file_mapping
);
3553 pgoff_t
__page_file_index(struct page
*page
)
3555 swp_entry_t swap
= { .val
= page_private(page
) };
3556 return swp_offset(swap
);
3558 EXPORT_SYMBOL_GPL(__page_file_index
);
3561 * add_swap_count_continuation - called when a swap count is duplicated
3562 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3563 * page of the original vmalloc'ed swap_map, to hold the continuation count
3564 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3565 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3567 * These continuation pages are seldom referenced: the common paths all work
3568 * on the original swap_map, only referring to a continuation page when the
3569 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3571 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3572 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3573 * can be called after dropping locks.
3575 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
3577 struct swap_info_struct
*si
;
3578 struct swap_cluster_info
*ci
;
3581 struct page
*list_page
;
3583 unsigned char count
;
3587 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3588 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3590 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
3592 si
= get_swap_device(entry
);
3595 * An acceptable race has occurred since the failing
3596 * __swap_duplicate(): the swap device may be swapoff
3600 spin_lock(&si
->lock
);
3602 offset
= swp_offset(entry
);
3604 ci
= lock_cluster(si
, offset
);
3606 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
3608 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
3610 * The higher the swap count, the more likely it is that tasks
3611 * will race to add swap count continuation: we need to avoid
3612 * over-provisioning.
3623 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3624 * no architecture is using highmem pages for kernel page tables: so it
3625 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3627 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3628 offset
&= ~PAGE_MASK
;
3630 spin_lock(&si
->cont_lock
);
3632 * Page allocation does not initialize the page's lru field,
3633 * but it does always reset its private field.
3635 if (!page_private(head
)) {
3636 BUG_ON(count
& COUNT_CONTINUED
);
3637 INIT_LIST_HEAD(&head
->lru
);
3638 set_page_private(head
, SWP_CONTINUED
);
3639 si
->flags
|= SWP_CONTINUED
;
3642 list_for_each_entry(list_page
, &head
->lru
, lru
) {
3646 * If the previous map said no continuation, but we've found
3647 * a continuation page, free our allocation and use this one.
3649 if (!(count
& COUNT_CONTINUED
))
3650 goto out_unlock_cont
;
3652 map
= kmap_atomic(list_page
) + offset
;
3657 * If this continuation count now has some space in it,
3658 * free our allocation and use this one.
3660 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
3661 goto out_unlock_cont
;
3664 list_add_tail(&page
->lru
, &head
->lru
);
3665 page
= NULL
; /* now it's attached, don't free it */
3667 spin_unlock(&si
->cont_lock
);
3670 spin_unlock(&si
->lock
);
3671 put_swap_device(si
);
3679 * swap_count_continued - when the original swap_map count is incremented
3680 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3681 * into, carry if so, or else fail until a new continuation page is allocated;
3682 * when the original swap_map count is decremented from 0 with continuation,
3683 * borrow from the continuation and report whether it still holds more.
3684 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3687 static bool swap_count_continued(struct swap_info_struct
*si
,
3688 pgoff_t offset
, unsigned char count
)
3695 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3696 if (page_private(head
) != SWP_CONTINUED
) {
3697 BUG_ON(count
& COUNT_CONTINUED
);
3698 return false; /* need to add count continuation */
3701 spin_lock(&si
->cont_lock
);
3702 offset
&= ~PAGE_MASK
;
3703 page
= list_next_entry(head
, lru
);
3704 map
= kmap_atomic(page
) + offset
;
3706 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3707 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3709 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3711 * Think of how you add 1 to 999
3713 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3715 page
= list_next_entry(page
, lru
);
3716 BUG_ON(page
== head
);
3717 map
= kmap_atomic(page
) + offset
;
3719 if (*map
== SWAP_CONT_MAX
) {
3721 page
= list_next_entry(page
, lru
);
3723 ret
= false; /* add count continuation */
3726 map
= kmap_atomic(page
) + offset
;
3727 init_map
: *map
= 0; /* we didn't zero the page */
3731 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3732 map
= kmap_atomic(page
) + offset
;
3733 *map
= COUNT_CONTINUED
;
3736 ret
= true; /* incremented */
3738 } else { /* decrementing */
3740 * Think of how you subtract 1 from 1000
3742 BUG_ON(count
!= COUNT_CONTINUED
);
3743 while (*map
== COUNT_CONTINUED
) {
3745 page
= list_next_entry(page
, lru
);
3746 BUG_ON(page
== head
);
3747 map
= kmap_atomic(page
) + offset
;
3754 while ((page
= list_prev_entry(page
, lru
)) != head
) {
3755 map
= kmap_atomic(page
) + offset
;
3756 *map
= SWAP_CONT_MAX
| count
;
3757 count
= COUNT_CONTINUED
;
3760 ret
= count
== COUNT_CONTINUED
;
3763 spin_unlock(&si
->cont_lock
);
3768 * free_swap_count_continuations - swapoff free all the continuation pages
3769 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3771 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3775 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3777 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3778 if (page_private(head
)) {
3779 struct page
*page
, *next
;
3781 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3782 list_del(&page
->lru
);
3789 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3790 void cgroup_throttle_swaprate(struct page
*page
, gfp_t gfp_mask
)
3792 struct swap_info_struct
*si
, *next
;
3793 int nid
= page_to_nid(page
);
3795 if (!(gfp_mask
& __GFP_IO
))
3798 if (!blk_cgroup_congested())
3802 * We've already scheduled a throttle, avoid taking the global swap
3805 if (current
->throttle_queue
)
3808 spin_lock(&swap_avail_lock
);
3809 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[nid
],
3812 blkcg_schedule_throttle(bdev_get_queue(si
->bdev
), true);
3816 spin_unlock(&swap_avail_lock
);
3820 static int __init
swapfile_init(void)
3824 swap_avail_heads
= kmalloc_array(nr_node_ids
, sizeof(struct plist_head
),
3826 if (!swap_avail_heads
) {
3827 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3832 plist_head_init(&swap_avail_heads
[nid
]);
3836 subsys_initcall(swapfile_init
);