1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie
10 #include <linux/sched/mm.h>
11 #include <linux/sched/task.h>
12 #include <linux/hugetlb.h>
13 #include <linux/mman.h>
14 #include <linux/slab.h>
15 #include <linux/kernel_stat.h>
16 #include <linux/swap.h>
17 #include <linux/vmalloc.h>
18 #include <linux/pagemap.h>
19 #include <linux/namei.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/writeback.h>
24 #include <linux/proc_fs.h>
25 #include <linux/seq_file.h>
26 #include <linux/init.h>
27 #include <linux/ksm.h>
28 #include <linux/rmap.h>
29 #include <linux/security.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mutex.h>
32 #include <linux/capability.h>
33 #include <linux/syscalls.h>
34 #include <linux/memcontrol.h>
35 #include <linux/poll.h>
36 #include <linux/oom.h>
37 #include <linux/frontswap.h>
38 #include <linux/swapfile.h>
39 #include <linux/export.h>
40 #include <linux/swap_slots.h>
41 #include <linux/sort.h>
43 #include <asm/pgtable.h>
44 #include <asm/tlbflush.h>
45 #include <linux/swapops.h>
46 #include <linux/swap_cgroup.h>
48 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
50 static void free_swap_count_continuations(struct swap_info_struct
*);
51 static sector_t
map_swap_entry(swp_entry_t
, struct block_device
**);
53 DEFINE_SPINLOCK(swap_lock
);
54 static unsigned int nr_swapfiles
;
55 atomic_long_t nr_swap_pages
;
57 * Some modules use swappable objects and may try to swap them out under
58 * memory pressure (via the shrinker). Before doing so, they may wish to
59 * check to see if any swap space is available.
61 EXPORT_SYMBOL_GPL(nr_swap_pages
);
62 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
63 long total_swap_pages
;
64 static int least_priority
= -1;
66 static const char Bad_file
[] = "Bad swap file entry ";
67 static const char Unused_file
[] = "Unused swap file entry ";
68 static const char Bad_offset
[] = "Bad swap offset entry ";
69 static const char Unused_offset
[] = "Unused swap offset entry ";
72 * all active swap_info_structs
73 * protected with swap_lock, and ordered by priority.
75 PLIST_HEAD(swap_active_head
);
78 * all available (active, not full) swap_info_structs
79 * protected with swap_avail_lock, ordered by priority.
80 * This is used by get_swap_page() instead of swap_active_head
81 * because swap_active_head includes all swap_info_structs,
82 * but get_swap_page() doesn't need to look at full ones.
83 * This uses its own lock instead of swap_lock because when a
84 * swap_info_struct changes between not-full/full, it needs to
85 * add/remove itself to/from this list, but the swap_info_struct->lock
86 * is held and the locking order requires swap_lock to be taken
87 * before any swap_info_struct->lock.
89 static struct plist_head
*swap_avail_heads
;
90 static DEFINE_SPINLOCK(swap_avail_lock
);
92 struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
94 static DEFINE_MUTEX(swapon_mutex
);
96 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
97 /* Activity counter to indicate that a swapon or swapoff has occurred */
98 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
100 atomic_t nr_rotate_swap
= ATOMIC_INIT(0);
102 static struct swap_info_struct
*swap_type_to_swap_info(int type
)
104 if (type
>= READ_ONCE(nr_swapfiles
))
107 smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */
108 return READ_ONCE(swap_info
[type
]);
111 static inline unsigned char swap_count(unsigned char ent
)
113 return ent
& ~SWAP_HAS_CACHE
; /* may include COUNT_CONTINUED flag */
116 /* Reclaim the swap entry anyway if possible */
117 #define TTRS_ANYWAY 0x1
119 * Reclaim the swap entry if there are no more mappings of the
122 #define TTRS_UNMAPPED 0x2
123 /* Reclaim the swap entry if swap is getting full*/
124 #define TTRS_FULL 0x4
126 /* returns 1 if swap entry is freed */
127 static int __try_to_reclaim_swap(struct swap_info_struct
*si
,
128 unsigned long offset
, unsigned long flags
)
130 swp_entry_t entry
= swp_entry(si
->type
, offset
);
134 page
= find_get_page(swap_address_space(entry
), offset
);
138 * When this function is called from scan_swap_map_slots() and it's
139 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
140 * here. We have to use trylock for avoiding deadlock. This is a special
141 * case and you should use try_to_free_swap() with explicit lock_page()
142 * in usual operations.
144 if (trylock_page(page
)) {
145 if ((flags
& TTRS_ANYWAY
) ||
146 ((flags
& TTRS_UNMAPPED
) && !page_mapped(page
)) ||
147 ((flags
& TTRS_FULL
) && mem_cgroup_swap_full(page
)))
148 ret
= try_to_free_swap(page
);
156 * swapon tell device that all the old swap contents can be discarded,
157 * to allow the swap device to optimize its wear-levelling.
159 static int discard_swap(struct swap_info_struct
*si
)
161 struct swap_extent
*se
;
162 sector_t start_block
;
166 /* Do not discard the swap header page! */
167 se
= &si
->first_swap_extent
;
168 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
169 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
171 err
= blkdev_issue_discard(si
->bdev
, start_block
,
172 nr_blocks
, GFP_KERNEL
, 0);
178 list_for_each_entry(se
, &si
->first_swap_extent
.list
, list
) {
179 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
180 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
182 err
= blkdev_issue_discard(si
->bdev
, start_block
,
183 nr_blocks
, GFP_KERNEL
, 0);
189 return err
; /* That will often be -EOPNOTSUPP */
193 * swap allocation tell device that a cluster of swap can now be discarded,
194 * to allow the swap device to optimize its wear-levelling.
196 static void discard_swap_cluster(struct swap_info_struct
*si
,
197 pgoff_t start_page
, pgoff_t nr_pages
)
199 struct swap_extent
*se
= si
->curr_swap_extent
;
200 int found_extent
= 0;
203 if (se
->start_page
<= start_page
&&
204 start_page
< se
->start_page
+ se
->nr_pages
) {
205 pgoff_t offset
= start_page
- se
->start_page
;
206 sector_t start_block
= se
->start_block
+ offset
;
207 sector_t nr_blocks
= se
->nr_pages
- offset
;
209 if (nr_blocks
> nr_pages
)
210 nr_blocks
= nr_pages
;
211 start_page
+= nr_blocks
;
212 nr_pages
-= nr_blocks
;
215 si
->curr_swap_extent
= se
;
217 start_block
<<= PAGE_SHIFT
- 9;
218 nr_blocks
<<= PAGE_SHIFT
- 9;
219 if (blkdev_issue_discard(si
->bdev
, start_block
,
220 nr_blocks
, GFP_NOIO
, 0))
224 se
= list_next_entry(se
, list
);
228 #ifdef CONFIG_THP_SWAP
229 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
231 #define swap_entry_size(size) (size)
233 #define SWAPFILE_CLUSTER 256
236 * Define swap_entry_size() as constant to let compiler to optimize
237 * out some code if !CONFIG_THP_SWAP
239 #define swap_entry_size(size) 1
241 #define LATENCY_LIMIT 256
243 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
249 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
254 static inline void cluster_set_count(struct swap_cluster_info
*info
,
260 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
261 unsigned int c
, unsigned int f
)
267 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
272 static inline void cluster_set_next(struct swap_cluster_info
*info
,
278 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
279 unsigned int n
, unsigned int f
)
285 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
287 return info
->flags
& CLUSTER_FLAG_FREE
;
290 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
292 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
295 static inline void cluster_set_null(struct swap_cluster_info
*info
)
297 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
301 static inline bool cluster_is_huge(struct swap_cluster_info
*info
)
303 if (IS_ENABLED(CONFIG_THP_SWAP
))
304 return info
->flags
& CLUSTER_FLAG_HUGE
;
308 static inline void cluster_clear_huge(struct swap_cluster_info
*info
)
310 info
->flags
&= ~CLUSTER_FLAG_HUGE
;
313 static inline struct swap_cluster_info
*lock_cluster(struct swap_info_struct
*si
,
314 unsigned long offset
)
316 struct swap_cluster_info
*ci
;
318 ci
= si
->cluster_info
;
320 ci
+= offset
/ SWAPFILE_CLUSTER
;
321 spin_lock(&ci
->lock
);
326 static inline void unlock_cluster(struct swap_cluster_info
*ci
)
329 spin_unlock(&ci
->lock
);
333 * Determine the locking method in use for this device. Return
334 * swap_cluster_info if SSD-style cluster-based locking is in place.
336 static inline struct swap_cluster_info
*lock_cluster_or_swap_info(
337 struct swap_info_struct
*si
, unsigned long offset
)
339 struct swap_cluster_info
*ci
;
341 /* Try to use fine-grained SSD-style locking if available: */
342 ci
= lock_cluster(si
, offset
);
343 /* Otherwise, fall back to traditional, coarse locking: */
345 spin_lock(&si
->lock
);
350 static inline void unlock_cluster_or_swap_info(struct swap_info_struct
*si
,
351 struct swap_cluster_info
*ci
)
356 spin_unlock(&si
->lock
);
359 static inline bool cluster_list_empty(struct swap_cluster_list
*list
)
361 return cluster_is_null(&list
->head
);
364 static inline unsigned int cluster_list_first(struct swap_cluster_list
*list
)
366 return cluster_next(&list
->head
);
369 static void cluster_list_init(struct swap_cluster_list
*list
)
371 cluster_set_null(&list
->head
);
372 cluster_set_null(&list
->tail
);
375 static void cluster_list_add_tail(struct swap_cluster_list
*list
,
376 struct swap_cluster_info
*ci
,
379 if (cluster_list_empty(list
)) {
380 cluster_set_next_flag(&list
->head
, idx
, 0);
381 cluster_set_next_flag(&list
->tail
, idx
, 0);
383 struct swap_cluster_info
*ci_tail
;
384 unsigned int tail
= cluster_next(&list
->tail
);
387 * Nested cluster lock, but both cluster locks are
388 * only acquired when we held swap_info_struct->lock
391 spin_lock_nested(&ci_tail
->lock
, SINGLE_DEPTH_NESTING
);
392 cluster_set_next(ci_tail
, idx
);
393 spin_unlock(&ci_tail
->lock
);
394 cluster_set_next_flag(&list
->tail
, idx
, 0);
398 static unsigned int cluster_list_del_first(struct swap_cluster_list
*list
,
399 struct swap_cluster_info
*ci
)
403 idx
= cluster_next(&list
->head
);
404 if (cluster_next(&list
->tail
) == idx
) {
405 cluster_set_null(&list
->head
);
406 cluster_set_null(&list
->tail
);
408 cluster_set_next_flag(&list
->head
,
409 cluster_next(&ci
[idx
]), 0);
414 /* Add a cluster to discard list and schedule it to do discard */
415 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
419 * If scan_swap_map() can't find a free cluster, it will check
420 * si->swap_map directly. To make sure the discarding cluster isn't
421 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
422 * will be cleared after discard
424 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
425 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
427 cluster_list_add_tail(&si
->discard_clusters
, si
->cluster_info
, idx
);
429 schedule_work(&si
->discard_work
);
432 static void __free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
434 struct swap_cluster_info
*ci
= si
->cluster_info
;
436 cluster_set_flag(ci
+ idx
, CLUSTER_FLAG_FREE
);
437 cluster_list_add_tail(&si
->free_clusters
, ci
, idx
);
441 * Doing discard actually. After a cluster discard is finished, the cluster
442 * will be added to free cluster list. caller should hold si->lock.
444 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
446 struct swap_cluster_info
*info
, *ci
;
449 info
= si
->cluster_info
;
451 while (!cluster_list_empty(&si
->discard_clusters
)) {
452 idx
= cluster_list_del_first(&si
->discard_clusters
, info
);
453 spin_unlock(&si
->lock
);
455 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
458 spin_lock(&si
->lock
);
459 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
460 __free_cluster(si
, idx
);
461 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
462 0, SWAPFILE_CLUSTER
);
467 static void swap_discard_work(struct work_struct
*work
)
469 struct swap_info_struct
*si
;
471 si
= container_of(work
, struct swap_info_struct
, discard_work
);
473 spin_lock(&si
->lock
);
474 swap_do_scheduled_discard(si
);
475 spin_unlock(&si
->lock
);
478 static void alloc_cluster(struct swap_info_struct
*si
, unsigned long idx
)
480 struct swap_cluster_info
*ci
= si
->cluster_info
;
482 VM_BUG_ON(cluster_list_first(&si
->free_clusters
) != idx
);
483 cluster_list_del_first(&si
->free_clusters
, ci
);
484 cluster_set_count_flag(ci
+ idx
, 0, 0);
487 static void free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
489 struct swap_cluster_info
*ci
= si
->cluster_info
+ idx
;
491 VM_BUG_ON(cluster_count(ci
) != 0);
493 * If the swap is discardable, prepare discard the cluster
494 * instead of free it immediately. The cluster will be freed
497 if ((si
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
498 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
499 swap_cluster_schedule_discard(si
, idx
);
503 __free_cluster(si
, idx
);
507 * The cluster corresponding to page_nr will be used. The cluster will be
508 * removed from free cluster list and its usage counter will be increased.
510 static void inc_cluster_info_page(struct swap_info_struct
*p
,
511 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
513 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
517 if (cluster_is_free(&cluster_info
[idx
]))
518 alloc_cluster(p
, idx
);
520 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
521 cluster_set_count(&cluster_info
[idx
],
522 cluster_count(&cluster_info
[idx
]) + 1);
526 * The cluster corresponding to page_nr decreases one usage. If the usage
527 * counter becomes 0, which means no page in the cluster is in using, we can
528 * optionally discard the cluster and add it to free cluster list.
530 static void dec_cluster_info_page(struct swap_info_struct
*p
,
531 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
533 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
538 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
539 cluster_set_count(&cluster_info
[idx
],
540 cluster_count(&cluster_info
[idx
]) - 1);
542 if (cluster_count(&cluster_info
[idx
]) == 0)
543 free_cluster(p
, idx
);
547 * It's possible scan_swap_map() uses a free cluster in the middle of free
548 * cluster list. Avoiding such abuse to avoid list corruption.
551 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
552 unsigned long offset
)
554 struct percpu_cluster
*percpu_cluster
;
557 offset
/= SWAPFILE_CLUSTER
;
558 conflict
= !cluster_list_empty(&si
->free_clusters
) &&
559 offset
!= cluster_list_first(&si
->free_clusters
) &&
560 cluster_is_free(&si
->cluster_info
[offset
]);
565 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
566 cluster_set_null(&percpu_cluster
->index
);
571 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
572 * might involve allocating a new cluster for current CPU too.
574 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
575 unsigned long *offset
, unsigned long *scan_base
)
577 struct percpu_cluster
*cluster
;
578 struct swap_cluster_info
*ci
;
580 unsigned long tmp
, max
;
583 cluster
= this_cpu_ptr(si
->percpu_cluster
);
584 if (cluster_is_null(&cluster
->index
)) {
585 if (!cluster_list_empty(&si
->free_clusters
)) {
586 cluster
->index
= si
->free_clusters
.head
;
587 cluster
->next
= cluster_next(&cluster
->index
) *
589 } else if (!cluster_list_empty(&si
->discard_clusters
)) {
591 * we don't have free cluster but have some clusters in
592 * discarding, do discard now and reclaim them
594 swap_do_scheduled_discard(si
);
595 *scan_base
= *offset
= si
->cluster_next
;
604 * Other CPUs can use our cluster if they can't find a free cluster,
605 * check if there is still free entry in the cluster
608 max
= min_t(unsigned long, si
->max
,
609 (cluster_next(&cluster
->index
) + 1) * SWAPFILE_CLUSTER
);
611 cluster_set_null(&cluster
->index
);
614 ci
= lock_cluster(si
, tmp
);
616 if (!si
->swap_map
[tmp
]) {
624 cluster_set_null(&cluster
->index
);
627 cluster
->next
= tmp
+ 1;
633 static void __del_from_avail_list(struct swap_info_struct
*p
)
638 plist_del(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
641 static void del_from_avail_list(struct swap_info_struct
*p
)
643 spin_lock(&swap_avail_lock
);
644 __del_from_avail_list(p
);
645 spin_unlock(&swap_avail_lock
);
648 static void swap_range_alloc(struct swap_info_struct
*si
, unsigned long offset
,
649 unsigned int nr_entries
)
651 unsigned int end
= offset
+ nr_entries
- 1;
653 if (offset
== si
->lowest_bit
)
654 si
->lowest_bit
+= nr_entries
;
655 if (end
== si
->highest_bit
)
656 si
->highest_bit
-= nr_entries
;
657 si
->inuse_pages
+= nr_entries
;
658 if (si
->inuse_pages
== si
->pages
) {
659 si
->lowest_bit
= si
->max
;
661 del_from_avail_list(si
);
665 static void add_to_avail_list(struct swap_info_struct
*p
)
669 spin_lock(&swap_avail_lock
);
671 WARN_ON(!plist_node_empty(&p
->avail_lists
[nid
]));
672 plist_add(&p
->avail_lists
[nid
], &swap_avail_heads
[nid
]);
674 spin_unlock(&swap_avail_lock
);
677 static void swap_range_free(struct swap_info_struct
*si
, unsigned long offset
,
678 unsigned int nr_entries
)
680 unsigned long end
= offset
+ nr_entries
- 1;
681 void (*swap_slot_free_notify
)(struct block_device
*, unsigned long);
683 if (offset
< si
->lowest_bit
)
684 si
->lowest_bit
= offset
;
685 if (end
> si
->highest_bit
) {
686 bool was_full
= !si
->highest_bit
;
688 si
->highest_bit
= end
;
689 if (was_full
&& (si
->flags
& SWP_WRITEOK
))
690 add_to_avail_list(si
);
692 atomic_long_add(nr_entries
, &nr_swap_pages
);
693 si
->inuse_pages
-= nr_entries
;
694 if (si
->flags
& SWP_BLKDEV
)
695 swap_slot_free_notify
=
696 si
->bdev
->bd_disk
->fops
->swap_slot_free_notify
;
698 swap_slot_free_notify
= NULL
;
699 while (offset
<= end
) {
700 frontswap_invalidate_page(si
->type
, offset
);
701 if (swap_slot_free_notify
)
702 swap_slot_free_notify(si
->bdev
, offset
);
707 static int scan_swap_map_slots(struct swap_info_struct
*si
,
708 unsigned char usage
, int nr
,
711 struct swap_cluster_info
*ci
;
712 unsigned long offset
;
713 unsigned long scan_base
;
714 unsigned long last_in_cluster
= 0;
715 int latency_ration
= LATENCY_LIMIT
;
722 * We try to cluster swap pages by allocating them sequentially
723 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
724 * way, however, we resort to first-free allocation, starting
725 * a new cluster. This prevents us from scattering swap pages
726 * all over the entire swap partition, so that we reduce
727 * overall disk seek times between swap pages. -- sct
728 * But we do now try to find an empty cluster. -Andrea
729 * And we let swap pages go all over an SSD partition. Hugh
732 si
->flags
+= SWP_SCANNING
;
733 scan_base
= offset
= si
->cluster_next
;
736 if (si
->cluster_info
) {
737 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
743 if (unlikely(!si
->cluster_nr
--)) {
744 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
745 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
749 spin_unlock(&si
->lock
);
752 * If seek is expensive, start searching for new cluster from
753 * start of partition, to minimize the span of allocated swap.
754 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
755 * case, just handled by scan_swap_map_try_ssd_cluster() above.
757 scan_base
= offset
= si
->lowest_bit
;
758 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
760 /* Locate the first empty (unaligned) cluster */
761 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
762 if (si
->swap_map
[offset
])
763 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
764 else if (offset
== last_in_cluster
) {
765 spin_lock(&si
->lock
);
766 offset
-= SWAPFILE_CLUSTER
- 1;
767 si
->cluster_next
= offset
;
768 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
771 if (unlikely(--latency_ration
< 0)) {
773 latency_ration
= LATENCY_LIMIT
;
778 spin_lock(&si
->lock
);
779 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
783 if (si
->cluster_info
) {
784 while (scan_swap_map_ssd_cluster_conflict(si
, offset
)) {
785 /* take a break if we already got some slots */
788 if (!scan_swap_map_try_ssd_cluster(si
, &offset
,
793 if (!(si
->flags
& SWP_WRITEOK
))
795 if (!si
->highest_bit
)
797 if (offset
> si
->highest_bit
)
798 scan_base
= offset
= si
->lowest_bit
;
800 ci
= lock_cluster(si
, offset
);
801 /* reuse swap entry of cache-only swap if not busy. */
802 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
805 spin_unlock(&si
->lock
);
806 swap_was_freed
= __try_to_reclaim_swap(si
, offset
, TTRS_ANYWAY
);
807 spin_lock(&si
->lock
);
808 /* entry was freed successfully, try to use this again */
811 goto scan
; /* check next one */
814 if (si
->swap_map
[offset
]) {
821 si
->swap_map
[offset
] = usage
;
822 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
825 swap_range_alloc(si
, offset
, 1);
826 si
->cluster_next
= offset
+ 1;
827 slots
[n_ret
++] = swp_entry(si
->type
, offset
);
829 /* got enough slots or reach max slots? */
830 if ((n_ret
== nr
) || (offset
>= si
->highest_bit
))
833 /* search for next available slot */
835 /* time to take a break? */
836 if (unlikely(--latency_ration
< 0)) {
839 spin_unlock(&si
->lock
);
841 spin_lock(&si
->lock
);
842 latency_ration
= LATENCY_LIMIT
;
845 /* try to get more slots in cluster */
846 if (si
->cluster_info
) {
847 if (scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
))
855 /* non-ssd case, still more slots in cluster? */
856 if (si
->cluster_nr
&& !si
->swap_map
[offset
]) {
862 si
->flags
-= SWP_SCANNING
;
866 spin_unlock(&si
->lock
);
867 while (++offset
<= si
->highest_bit
) {
868 if (!si
->swap_map
[offset
]) {
869 spin_lock(&si
->lock
);
872 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
873 spin_lock(&si
->lock
);
876 if (unlikely(--latency_ration
< 0)) {
878 latency_ration
= LATENCY_LIMIT
;
881 offset
= si
->lowest_bit
;
882 while (offset
< scan_base
) {
883 if (!si
->swap_map
[offset
]) {
884 spin_lock(&si
->lock
);
887 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
888 spin_lock(&si
->lock
);
891 if (unlikely(--latency_ration
< 0)) {
893 latency_ration
= LATENCY_LIMIT
;
897 spin_lock(&si
->lock
);
900 si
->flags
-= SWP_SCANNING
;
904 static int swap_alloc_cluster(struct swap_info_struct
*si
, swp_entry_t
*slot
)
907 struct swap_cluster_info
*ci
;
908 unsigned long offset
, i
;
912 * Should not even be attempting cluster allocations when huge
913 * page swap is disabled. Warn and fail the allocation.
915 if (!IS_ENABLED(CONFIG_THP_SWAP
)) {
920 if (cluster_list_empty(&si
->free_clusters
))
923 idx
= cluster_list_first(&si
->free_clusters
);
924 offset
= idx
* SWAPFILE_CLUSTER
;
925 ci
= lock_cluster(si
, offset
);
926 alloc_cluster(si
, idx
);
927 cluster_set_count_flag(ci
, SWAPFILE_CLUSTER
, CLUSTER_FLAG_HUGE
);
929 map
= si
->swap_map
+ offset
;
930 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++)
931 map
[i
] = SWAP_HAS_CACHE
;
933 swap_range_alloc(si
, offset
, SWAPFILE_CLUSTER
);
934 *slot
= swp_entry(si
->type
, offset
);
939 static void swap_free_cluster(struct swap_info_struct
*si
, unsigned long idx
)
941 unsigned long offset
= idx
* SWAPFILE_CLUSTER
;
942 struct swap_cluster_info
*ci
;
944 ci
= lock_cluster(si
, offset
);
945 memset(si
->swap_map
+ offset
, 0, SWAPFILE_CLUSTER
);
946 cluster_set_count_flag(ci
, 0, 0);
947 free_cluster(si
, idx
);
949 swap_range_free(si
, offset
, SWAPFILE_CLUSTER
);
952 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
958 n_ret
= scan_swap_map_slots(si
, usage
, 1, &entry
);
961 return swp_offset(entry
);
967 int get_swap_pages(int n_goal
, swp_entry_t swp_entries
[], int entry_size
)
969 unsigned long size
= swap_entry_size(entry_size
);
970 struct swap_info_struct
*si
, *next
;
975 /* Only single cluster request supported */
976 WARN_ON_ONCE(n_goal
> 1 && size
== SWAPFILE_CLUSTER
);
978 avail_pgs
= atomic_long_read(&nr_swap_pages
) / size
;
982 if (n_goal
> SWAP_BATCH
)
985 if (n_goal
> avail_pgs
)
988 atomic_long_sub(n_goal
* size
, &nr_swap_pages
);
990 spin_lock(&swap_avail_lock
);
993 node
= numa_node_id();
994 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[node
], avail_lists
[node
]) {
995 /* requeue si to after same-priority siblings */
996 plist_requeue(&si
->avail_lists
[node
], &swap_avail_heads
[node
]);
997 spin_unlock(&swap_avail_lock
);
998 spin_lock(&si
->lock
);
999 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
1000 spin_lock(&swap_avail_lock
);
1001 if (plist_node_empty(&si
->avail_lists
[node
])) {
1002 spin_unlock(&si
->lock
);
1005 WARN(!si
->highest_bit
,
1006 "swap_info %d in list but !highest_bit\n",
1008 WARN(!(si
->flags
& SWP_WRITEOK
),
1009 "swap_info %d in list but !SWP_WRITEOK\n",
1011 __del_from_avail_list(si
);
1012 spin_unlock(&si
->lock
);
1015 if (size
== SWAPFILE_CLUSTER
) {
1016 if (!(si
->flags
& SWP_FS
))
1017 n_ret
= swap_alloc_cluster(si
, swp_entries
);
1019 n_ret
= scan_swap_map_slots(si
, SWAP_HAS_CACHE
,
1020 n_goal
, swp_entries
);
1021 spin_unlock(&si
->lock
);
1022 if (n_ret
|| size
== SWAPFILE_CLUSTER
)
1024 pr_debug("scan_swap_map of si %d failed to find offset\n",
1027 spin_lock(&swap_avail_lock
);
1030 * if we got here, it's likely that si was almost full before,
1031 * and since scan_swap_map() can drop the si->lock, multiple
1032 * callers probably all tried to get a page from the same si
1033 * and it filled up before we could get one; or, the si filled
1034 * up between us dropping swap_avail_lock and taking si->lock.
1035 * Since we dropped the swap_avail_lock, the swap_avail_head
1036 * list may have been modified; so if next is still in the
1037 * swap_avail_head list then try it, otherwise start over
1038 * if we have not gotten any slots.
1040 if (plist_node_empty(&next
->avail_lists
[node
]))
1044 spin_unlock(&swap_avail_lock
);
1048 atomic_long_add((long)(n_goal
- n_ret
) * size
,
1054 /* The only caller of this function is now suspend routine */
1055 swp_entry_t
get_swap_page_of_type(int type
)
1057 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1063 spin_lock(&si
->lock
);
1064 if (si
->flags
& SWP_WRITEOK
) {
1065 atomic_long_dec(&nr_swap_pages
);
1066 /* This is called for allocating swap entry, not cache */
1067 offset
= scan_swap_map(si
, 1);
1069 spin_unlock(&si
->lock
);
1070 return swp_entry(type
, offset
);
1072 atomic_long_inc(&nr_swap_pages
);
1074 spin_unlock(&si
->lock
);
1076 return (swp_entry_t
) {0};
1079 static struct swap_info_struct
*__swap_info_get(swp_entry_t entry
)
1081 struct swap_info_struct
*p
;
1082 unsigned long offset
;
1086 p
= swp_swap_info(entry
);
1089 if (!(p
->flags
& SWP_USED
))
1091 offset
= swp_offset(entry
);
1092 if (offset
>= p
->max
)
1097 pr_err("swap_info_get: %s%08lx\n", Bad_offset
, entry
.val
);
1100 pr_err("swap_info_get: %s%08lx\n", Unused_file
, entry
.val
);
1103 pr_err("swap_info_get: %s%08lx\n", Bad_file
, entry
.val
);
1108 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
1110 struct swap_info_struct
*p
;
1112 p
= __swap_info_get(entry
);
1115 if (!p
->swap_map
[swp_offset(entry
)])
1120 pr_err("swap_info_get: %s%08lx\n", Unused_offset
, entry
.val
);
1126 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
1128 struct swap_info_struct
*p
;
1130 p
= _swap_info_get(entry
);
1132 spin_lock(&p
->lock
);
1136 static struct swap_info_struct
*swap_info_get_cont(swp_entry_t entry
,
1137 struct swap_info_struct
*q
)
1139 struct swap_info_struct
*p
;
1141 p
= _swap_info_get(entry
);
1145 spin_unlock(&q
->lock
);
1147 spin_lock(&p
->lock
);
1152 static unsigned char __swap_entry_free_locked(struct swap_info_struct
*p
,
1153 unsigned long offset
,
1154 unsigned char usage
)
1156 unsigned char count
;
1157 unsigned char has_cache
;
1159 count
= p
->swap_map
[offset
];
1161 has_cache
= count
& SWAP_HAS_CACHE
;
1162 count
&= ~SWAP_HAS_CACHE
;
1164 if (usage
== SWAP_HAS_CACHE
) {
1165 VM_BUG_ON(!has_cache
);
1167 } else if (count
== SWAP_MAP_SHMEM
) {
1169 * Or we could insist on shmem.c using a special
1170 * swap_shmem_free() and free_shmem_swap_and_cache()...
1173 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
1174 if (count
== COUNT_CONTINUED
) {
1175 if (swap_count_continued(p
, offset
, count
))
1176 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
1178 count
= SWAP_MAP_MAX
;
1183 usage
= count
| has_cache
;
1184 p
->swap_map
[offset
] = usage
? : SWAP_HAS_CACHE
;
1190 * Check whether swap entry is valid in the swap device. If so,
1191 * return pointer to swap_info_struct, and keep the swap entry valid
1192 * via preventing the swap device from being swapoff, until
1193 * put_swap_device() is called. Otherwise return NULL.
1195 * The entirety of the RCU read critical section must come before the
1196 * return from or after the call to synchronize_rcu() in
1197 * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is
1198 * true, the si->map, si->cluster_info, etc. must be valid in the
1201 * Notice that swapoff or swapoff+swapon can still happen before the
1202 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1203 * in put_swap_device() if there isn't any other way to prevent
1204 * swapoff, such as page lock, page table lock, etc. The caller must
1205 * be prepared for that. For example, the following situation is
1210 * ... swapoff+swapon
1211 * __read_swap_cache_async()
1212 * swapcache_prepare()
1213 * __swap_duplicate()
1215 * // verify PTE not changed
1217 * In __swap_duplicate(), the swap_map need to be checked before
1218 * changing partly because the specified swap entry may be for another
1219 * swap device which has been swapoff. And in do_swap_page(), after
1220 * the page is read from the swap device, the PTE is verified not
1221 * changed with the page table locked to check whether the swap device
1222 * has been swapoff or swapoff+swapon.
1224 struct swap_info_struct
*get_swap_device(swp_entry_t entry
)
1226 struct swap_info_struct
*si
;
1227 unsigned long offset
;
1231 si
= swp_swap_info(entry
);
1236 if (!(si
->flags
& SWP_VALID
))
1238 offset
= swp_offset(entry
);
1239 if (offset
>= si
->max
)
1244 pr_err("%s: %s%08lx\n", __func__
, Bad_file
, entry
.val
);
1252 static unsigned char __swap_entry_free(struct swap_info_struct
*p
,
1253 swp_entry_t entry
, unsigned char usage
)
1255 struct swap_cluster_info
*ci
;
1256 unsigned long offset
= swp_offset(entry
);
1258 ci
= lock_cluster_or_swap_info(p
, offset
);
1259 usage
= __swap_entry_free_locked(p
, offset
, usage
);
1260 unlock_cluster_or_swap_info(p
, ci
);
1262 free_swap_slot(entry
);
1267 static void swap_entry_free(struct swap_info_struct
*p
, swp_entry_t entry
)
1269 struct swap_cluster_info
*ci
;
1270 unsigned long offset
= swp_offset(entry
);
1271 unsigned char count
;
1273 ci
= lock_cluster(p
, offset
);
1274 count
= p
->swap_map
[offset
];
1275 VM_BUG_ON(count
!= SWAP_HAS_CACHE
);
1276 p
->swap_map
[offset
] = 0;
1277 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
1280 mem_cgroup_uncharge_swap(entry
, 1);
1281 swap_range_free(p
, offset
, 1);
1285 * Caller has made sure that the swap device corresponding to entry
1286 * is still around or has not been recycled.
1288 void swap_free(swp_entry_t entry
)
1290 struct swap_info_struct
*p
;
1292 p
= _swap_info_get(entry
);
1294 __swap_entry_free(p
, entry
, 1);
1298 * Called after dropping swapcache to decrease refcnt to swap entries.
1300 void put_swap_page(struct page
*page
, swp_entry_t entry
)
1302 unsigned long offset
= swp_offset(entry
);
1303 unsigned long idx
= offset
/ SWAPFILE_CLUSTER
;
1304 struct swap_cluster_info
*ci
;
1305 struct swap_info_struct
*si
;
1307 unsigned int i
, free_entries
= 0;
1309 int size
= swap_entry_size(hpage_nr_pages(page
));
1311 si
= _swap_info_get(entry
);
1315 ci
= lock_cluster_or_swap_info(si
, offset
);
1316 if (size
== SWAPFILE_CLUSTER
) {
1317 VM_BUG_ON(!cluster_is_huge(ci
));
1318 map
= si
->swap_map
+ offset
;
1319 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1321 VM_BUG_ON(!(val
& SWAP_HAS_CACHE
));
1322 if (val
== SWAP_HAS_CACHE
)
1325 cluster_clear_huge(ci
);
1326 if (free_entries
== SWAPFILE_CLUSTER
) {
1327 unlock_cluster_or_swap_info(si
, ci
);
1328 spin_lock(&si
->lock
);
1329 mem_cgroup_uncharge_swap(entry
, SWAPFILE_CLUSTER
);
1330 swap_free_cluster(si
, idx
);
1331 spin_unlock(&si
->lock
);
1335 for (i
= 0; i
< size
; i
++, entry
.val
++) {
1336 if (!__swap_entry_free_locked(si
, offset
+ i
, SWAP_HAS_CACHE
)) {
1337 unlock_cluster_or_swap_info(si
, ci
);
1338 free_swap_slot(entry
);
1341 lock_cluster_or_swap_info(si
, offset
);
1344 unlock_cluster_or_swap_info(si
, ci
);
1347 #ifdef CONFIG_THP_SWAP
1348 int split_swap_cluster(swp_entry_t entry
)
1350 struct swap_info_struct
*si
;
1351 struct swap_cluster_info
*ci
;
1352 unsigned long offset
= swp_offset(entry
);
1354 si
= _swap_info_get(entry
);
1357 ci
= lock_cluster(si
, offset
);
1358 cluster_clear_huge(ci
);
1364 static int swp_entry_cmp(const void *ent1
, const void *ent2
)
1366 const swp_entry_t
*e1
= ent1
, *e2
= ent2
;
1368 return (int)swp_type(*e1
) - (int)swp_type(*e2
);
1371 void swapcache_free_entries(swp_entry_t
*entries
, int n
)
1373 struct swap_info_struct
*p
, *prev
;
1383 * Sort swap entries by swap device, so each lock is only taken once.
1384 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1385 * so low that it isn't necessary to optimize further.
1387 if (nr_swapfiles
> 1)
1388 sort(entries
, n
, sizeof(entries
[0]), swp_entry_cmp
, NULL
);
1389 for (i
= 0; i
< n
; ++i
) {
1390 p
= swap_info_get_cont(entries
[i
], prev
);
1392 swap_entry_free(p
, entries
[i
]);
1396 spin_unlock(&p
->lock
);
1400 * How many references to page are currently swapped out?
1401 * This does not give an exact answer when swap count is continued,
1402 * but does include the high COUNT_CONTINUED flag to allow for that.
1404 int page_swapcount(struct page
*page
)
1407 struct swap_info_struct
*p
;
1408 struct swap_cluster_info
*ci
;
1410 unsigned long offset
;
1412 entry
.val
= page_private(page
);
1413 p
= _swap_info_get(entry
);
1415 offset
= swp_offset(entry
);
1416 ci
= lock_cluster_or_swap_info(p
, offset
);
1417 count
= swap_count(p
->swap_map
[offset
]);
1418 unlock_cluster_or_swap_info(p
, ci
);
1423 int __swap_count(swp_entry_t entry
)
1425 struct swap_info_struct
*si
;
1426 pgoff_t offset
= swp_offset(entry
);
1429 si
= get_swap_device(entry
);
1431 count
= swap_count(si
->swap_map
[offset
]);
1432 put_swap_device(si
);
1437 static int swap_swapcount(struct swap_info_struct
*si
, swp_entry_t entry
)
1440 pgoff_t offset
= swp_offset(entry
);
1441 struct swap_cluster_info
*ci
;
1443 ci
= lock_cluster_or_swap_info(si
, offset
);
1444 count
= swap_count(si
->swap_map
[offset
]);
1445 unlock_cluster_or_swap_info(si
, ci
);
1450 * How many references to @entry are currently swapped out?
1451 * This does not give an exact answer when swap count is continued,
1452 * but does include the high COUNT_CONTINUED flag to allow for that.
1454 int __swp_swapcount(swp_entry_t entry
)
1457 struct swap_info_struct
*si
;
1459 si
= get_swap_device(entry
);
1461 count
= swap_swapcount(si
, entry
);
1462 put_swap_device(si
);
1468 * How many references to @entry are currently swapped out?
1469 * This considers COUNT_CONTINUED so it returns exact answer.
1471 int swp_swapcount(swp_entry_t entry
)
1473 int count
, tmp_count
, n
;
1474 struct swap_info_struct
*p
;
1475 struct swap_cluster_info
*ci
;
1480 p
= _swap_info_get(entry
);
1484 offset
= swp_offset(entry
);
1486 ci
= lock_cluster_or_swap_info(p
, offset
);
1488 count
= swap_count(p
->swap_map
[offset
]);
1489 if (!(count
& COUNT_CONTINUED
))
1492 count
&= ~COUNT_CONTINUED
;
1493 n
= SWAP_MAP_MAX
+ 1;
1495 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1496 offset
&= ~PAGE_MASK
;
1497 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1500 page
= list_next_entry(page
, lru
);
1501 map
= kmap_atomic(page
);
1502 tmp_count
= map
[offset
];
1505 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1506 n
*= (SWAP_CONT_MAX
+ 1);
1507 } while (tmp_count
& COUNT_CONTINUED
);
1509 unlock_cluster_or_swap_info(p
, ci
);
1513 static bool swap_page_trans_huge_swapped(struct swap_info_struct
*si
,
1516 struct swap_cluster_info
*ci
;
1517 unsigned char *map
= si
->swap_map
;
1518 unsigned long roffset
= swp_offset(entry
);
1519 unsigned long offset
= round_down(roffset
, SWAPFILE_CLUSTER
);
1523 ci
= lock_cluster_or_swap_info(si
, offset
);
1524 if (!ci
|| !cluster_is_huge(ci
)) {
1525 if (swap_count(map
[roffset
]))
1529 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1530 if (swap_count(map
[offset
+ i
])) {
1536 unlock_cluster_or_swap_info(si
, ci
);
1540 static bool page_swapped(struct page
*page
)
1543 struct swap_info_struct
*si
;
1545 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
)))
1546 return page_swapcount(page
) != 0;
1548 page
= compound_head(page
);
1549 entry
.val
= page_private(page
);
1550 si
= _swap_info_get(entry
);
1552 return swap_page_trans_huge_swapped(si
, entry
);
1556 static int page_trans_huge_map_swapcount(struct page
*page
, int *total_mapcount
,
1557 int *total_swapcount
)
1559 int i
, map_swapcount
, _total_mapcount
, _total_swapcount
;
1560 unsigned long offset
= 0;
1561 struct swap_info_struct
*si
;
1562 struct swap_cluster_info
*ci
= NULL
;
1563 unsigned char *map
= NULL
;
1564 int mapcount
, swapcount
= 0;
1566 /* hugetlbfs shouldn't call it */
1567 VM_BUG_ON_PAGE(PageHuge(page
), page
);
1569 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
))) {
1570 mapcount
= page_trans_huge_mapcount(page
, total_mapcount
);
1571 if (PageSwapCache(page
))
1572 swapcount
= page_swapcount(page
);
1573 if (total_swapcount
)
1574 *total_swapcount
= swapcount
;
1575 return mapcount
+ swapcount
;
1578 page
= compound_head(page
);
1580 _total_mapcount
= _total_swapcount
= map_swapcount
= 0;
1581 if (PageSwapCache(page
)) {
1584 entry
.val
= page_private(page
);
1585 si
= _swap_info_get(entry
);
1588 offset
= swp_offset(entry
);
1592 ci
= lock_cluster(si
, offset
);
1593 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
1594 mapcount
= atomic_read(&page
[i
]._mapcount
) + 1;
1595 _total_mapcount
+= mapcount
;
1597 swapcount
= swap_count(map
[offset
+ i
]);
1598 _total_swapcount
+= swapcount
;
1600 map_swapcount
= max(map_swapcount
, mapcount
+ swapcount
);
1603 if (PageDoubleMap(page
)) {
1605 _total_mapcount
-= HPAGE_PMD_NR
;
1607 mapcount
= compound_mapcount(page
);
1608 map_swapcount
+= mapcount
;
1609 _total_mapcount
+= mapcount
;
1611 *total_mapcount
= _total_mapcount
;
1612 if (total_swapcount
)
1613 *total_swapcount
= _total_swapcount
;
1615 return map_swapcount
;
1619 * We can write to an anon page without COW if there are no other references
1620 * to it. And as a side-effect, free up its swap: because the old content
1621 * on disk will never be read, and seeking back there to write new content
1622 * later would only waste time away from clustering.
1624 * NOTE: total_map_swapcount should not be relied upon by the caller if
1625 * reuse_swap_page() returns false, but it may be always overwritten
1626 * (see the other implementation for CONFIG_SWAP=n).
1628 bool reuse_swap_page(struct page
*page
, int *total_map_swapcount
)
1630 int count
, total_mapcount
, total_swapcount
;
1632 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1633 if (unlikely(PageKsm(page
)))
1635 count
= page_trans_huge_map_swapcount(page
, &total_mapcount
,
1637 if (total_map_swapcount
)
1638 *total_map_swapcount
= total_mapcount
+ total_swapcount
;
1639 if (count
== 1 && PageSwapCache(page
) &&
1640 (likely(!PageTransCompound(page
)) ||
1641 /* The remaining swap count will be freed soon */
1642 total_swapcount
== page_swapcount(page
))) {
1643 if (!PageWriteback(page
)) {
1644 page
= compound_head(page
);
1645 delete_from_swap_cache(page
);
1649 struct swap_info_struct
*p
;
1651 entry
.val
= page_private(page
);
1652 p
= swap_info_get(entry
);
1653 if (p
->flags
& SWP_STABLE_WRITES
) {
1654 spin_unlock(&p
->lock
);
1657 spin_unlock(&p
->lock
);
1665 * If swap is getting full, or if there are no more mappings of this page,
1666 * then try_to_free_swap is called to free its swap space.
1668 int try_to_free_swap(struct page
*page
)
1670 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1672 if (!PageSwapCache(page
))
1674 if (PageWriteback(page
))
1676 if (page_swapped(page
))
1680 * Once hibernation has begun to create its image of memory,
1681 * there's a danger that one of the calls to try_to_free_swap()
1682 * - most probably a call from __try_to_reclaim_swap() while
1683 * hibernation is allocating its own swap pages for the image,
1684 * but conceivably even a call from memory reclaim - will free
1685 * the swap from a page which has already been recorded in the
1686 * image as a clean swapcache page, and then reuse its swap for
1687 * another page of the image. On waking from hibernation, the
1688 * original page might be freed under memory pressure, then
1689 * later read back in from swap, now with the wrong data.
1691 * Hibernation suspends storage while it is writing the image
1692 * to disk so check that here.
1694 if (pm_suspended_storage())
1697 page
= compound_head(page
);
1698 delete_from_swap_cache(page
);
1704 * Free the swap entry like above, but also try to
1705 * free the page cache entry if it is the last user.
1707 int free_swap_and_cache(swp_entry_t entry
)
1709 struct swap_info_struct
*p
;
1710 unsigned char count
;
1712 if (non_swap_entry(entry
))
1715 p
= _swap_info_get(entry
);
1717 count
= __swap_entry_free(p
, entry
, 1);
1718 if (count
== SWAP_HAS_CACHE
&&
1719 !swap_page_trans_huge_swapped(p
, entry
))
1720 __try_to_reclaim_swap(p
, swp_offset(entry
),
1721 TTRS_UNMAPPED
| TTRS_FULL
);
1726 #ifdef CONFIG_HIBERNATION
1728 * Find the swap type that corresponds to given device (if any).
1730 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1731 * from 0, in which the swap header is expected to be located.
1733 * This is needed for the suspend to disk (aka swsusp).
1735 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
1737 struct block_device
*bdev
= NULL
;
1741 bdev
= bdget(device
);
1743 spin_lock(&swap_lock
);
1744 for (type
= 0; type
< nr_swapfiles
; type
++) {
1745 struct swap_info_struct
*sis
= swap_info
[type
];
1747 if (!(sis
->flags
& SWP_WRITEOK
))
1752 *bdev_p
= bdgrab(sis
->bdev
);
1754 spin_unlock(&swap_lock
);
1757 if (bdev
== sis
->bdev
) {
1758 struct swap_extent
*se
= &sis
->first_swap_extent
;
1760 if (se
->start_block
== offset
) {
1762 *bdev_p
= bdgrab(sis
->bdev
);
1764 spin_unlock(&swap_lock
);
1770 spin_unlock(&swap_lock
);
1778 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1779 * corresponding to given index in swap_info (swap type).
1781 sector_t
swapdev_block(int type
, pgoff_t offset
)
1783 struct block_device
*bdev
;
1784 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1786 if (!si
|| !(si
->flags
& SWP_WRITEOK
))
1788 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
1792 * Return either the total number of swap pages of given type, or the number
1793 * of free pages of that type (depending on @free)
1795 * This is needed for software suspend
1797 unsigned int count_swap_pages(int type
, int free
)
1801 spin_lock(&swap_lock
);
1802 if ((unsigned int)type
< nr_swapfiles
) {
1803 struct swap_info_struct
*sis
= swap_info
[type
];
1805 spin_lock(&sis
->lock
);
1806 if (sis
->flags
& SWP_WRITEOK
) {
1809 n
-= sis
->inuse_pages
;
1811 spin_unlock(&sis
->lock
);
1813 spin_unlock(&swap_lock
);
1816 #endif /* CONFIG_HIBERNATION */
1818 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1820 return pte_same(pte_swp_clear_soft_dirty(pte
), swp_pte
);
1824 * No need to decide whether this PTE shares the swap entry with others,
1825 * just let do_wp_page work it out if a write is requested later - to
1826 * force COW, vm_page_prot omits write permission from any private vma.
1828 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1829 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1831 struct page
*swapcache
;
1832 struct mem_cgroup
*memcg
;
1838 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1839 if (unlikely(!page
))
1842 if (mem_cgroup_try_charge(page
, vma
->vm_mm
, GFP_KERNEL
,
1848 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1849 if (unlikely(!pte_same_as_swp(*pte
, swp_entry_to_pte(entry
)))) {
1850 mem_cgroup_cancel_charge(page
, memcg
, false);
1855 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1856 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1858 set_pte_at(vma
->vm_mm
, addr
, pte
,
1859 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1860 if (page
== swapcache
) {
1861 page_add_anon_rmap(page
, vma
, addr
, false);
1862 mem_cgroup_commit_charge(page
, memcg
, true, false);
1863 } else { /* ksm created a completely new copy */
1864 page_add_new_anon_rmap(page
, vma
, addr
, false);
1865 mem_cgroup_commit_charge(page
, memcg
, false, false);
1866 lru_cache_add_active_or_unevictable(page
, vma
);
1870 * Move the page to the active list so it is not
1871 * immediately swapped out again after swapon.
1873 activate_page(page
);
1875 pte_unmap_unlock(pte
, ptl
);
1877 if (page
!= swapcache
) {
1884 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1885 unsigned long addr
, unsigned long end
,
1886 unsigned int type
, bool frontswap
,
1887 unsigned long *fs_pages_to_unuse
)
1892 struct swap_info_struct
*si
;
1893 unsigned long offset
;
1895 volatile unsigned char *swap_map
;
1897 si
= swap_info
[type
];
1898 pte
= pte_offset_map(pmd
, addr
);
1900 struct vm_fault vmf
;
1902 if (!is_swap_pte(*pte
))
1905 entry
= pte_to_swp_entry(*pte
);
1906 if (swp_type(entry
) != type
)
1909 offset
= swp_offset(entry
);
1910 if (frontswap
&& !frontswap_test(si
, offset
))
1914 swap_map
= &si
->swap_map
[offset
];
1918 page
= swapin_readahead(entry
, GFP_HIGHUSER_MOVABLE
, &vmf
);
1920 if (*swap_map
== 0 || *swap_map
== SWAP_MAP_BAD
)
1926 wait_on_page_writeback(page
);
1927 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1934 try_to_free_swap(page
);
1938 if (*fs_pages_to_unuse
&& !--(*fs_pages_to_unuse
)) {
1939 ret
= FRONTSWAP_PAGES_UNUSED
;
1943 pte
= pte_offset_map(pmd
, addr
);
1944 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1952 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
1953 unsigned long addr
, unsigned long end
,
1954 unsigned int type
, bool frontswap
,
1955 unsigned long *fs_pages_to_unuse
)
1961 pmd
= pmd_offset(pud
, addr
);
1964 next
= pmd_addr_end(addr
, end
);
1965 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1967 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, type
,
1968 frontswap
, fs_pages_to_unuse
);
1971 } while (pmd
++, addr
= next
, addr
!= end
);
1975 static inline int unuse_pud_range(struct vm_area_struct
*vma
, p4d_t
*p4d
,
1976 unsigned long addr
, unsigned long end
,
1977 unsigned int type
, bool frontswap
,
1978 unsigned long *fs_pages_to_unuse
)
1984 pud
= pud_offset(p4d
, addr
);
1986 next
= pud_addr_end(addr
, end
);
1987 if (pud_none_or_clear_bad(pud
))
1989 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, type
,
1990 frontswap
, fs_pages_to_unuse
);
1993 } while (pud
++, addr
= next
, addr
!= end
);
1997 static inline int unuse_p4d_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
1998 unsigned long addr
, unsigned long end
,
1999 unsigned int type
, bool frontswap
,
2000 unsigned long *fs_pages_to_unuse
)
2006 p4d
= p4d_offset(pgd
, addr
);
2008 next
= p4d_addr_end(addr
, end
);
2009 if (p4d_none_or_clear_bad(p4d
))
2011 ret
= unuse_pud_range(vma
, p4d
, addr
, next
, type
,
2012 frontswap
, fs_pages_to_unuse
);
2015 } while (p4d
++, addr
= next
, addr
!= end
);
2019 static int unuse_vma(struct vm_area_struct
*vma
, unsigned int type
,
2020 bool frontswap
, unsigned long *fs_pages_to_unuse
)
2023 unsigned long addr
, end
, next
;
2026 addr
= vma
->vm_start
;
2029 pgd
= pgd_offset(vma
->vm_mm
, addr
);
2031 next
= pgd_addr_end(addr
, end
);
2032 if (pgd_none_or_clear_bad(pgd
))
2034 ret
= unuse_p4d_range(vma
, pgd
, addr
, next
, type
,
2035 frontswap
, fs_pages_to_unuse
);
2038 } while (pgd
++, addr
= next
, addr
!= end
);
2042 static int unuse_mm(struct mm_struct
*mm
, unsigned int type
,
2043 bool frontswap
, unsigned long *fs_pages_to_unuse
)
2045 struct vm_area_struct
*vma
;
2048 down_read(&mm
->mmap_sem
);
2049 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
2050 if (vma
->anon_vma
) {
2051 ret
= unuse_vma(vma
, type
, frontswap
,
2058 up_read(&mm
->mmap_sem
);
2063 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2064 * from current position to next entry still in use. Return 0
2065 * if there are no inuse entries after prev till end of the map.
2067 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
2068 unsigned int prev
, bool frontswap
)
2071 unsigned char count
;
2074 * No need for swap_lock here: we're just looking
2075 * for whether an entry is in use, not modifying it; false
2076 * hits are okay, and sys_swapoff() has already prevented new
2077 * allocations from this area (while holding swap_lock).
2079 for (i
= prev
+ 1; i
< si
->max
; i
++) {
2080 count
= READ_ONCE(si
->swap_map
[i
]);
2081 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
2082 if (!frontswap
|| frontswap_test(si
, i
))
2084 if ((i
% LATENCY_LIMIT
) == 0)
2095 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2096 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2098 int try_to_unuse(unsigned int type
, bool frontswap
,
2099 unsigned long pages_to_unuse
)
2101 struct mm_struct
*prev_mm
;
2102 struct mm_struct
*mm
;
2103 struct list_head
*p
;
2105 struct swap_info_struct
*si
= swap_info
[type
];
2110 if (!si
->inuse_pages
)
2117 retval
= shmem_unuse(type
, frontswap
, &pages_to_unuse
);
2124 spin_lock(&mmlist_lock
);
2125 p
= &init_mm
.mmlist
;
2126 while (si
->inuse_pages
&&
2127 !signal_pending(current
) &&
2128 (p
= p
->next
) != &init_mm
.mmlist
) {
2130 mm
= list_entry(p
, struct mm_struct
, mmlist
);
2131 if (!mmget_not_zero(mm
))
2133 spin_unlock(&mmlist_lock
);
2136 retval
= unuse_mm(mm
, type
, frontswap
, &pages_to_unuse
);
2144 * Make sure that we aren't completely killing
2145 * interactive performance.
2148 spin_lock(&mmlist_lock
);
2150 spin_unlock(&mmlist_lock
);
2155 while (si
->inuse_pages
&&
2156 !signal_pending(current
) &&
2157 (i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
2159 entry
= swp_entry(type
, i
);
2160 page
= find_get_page(swap_address_space(entry
), i
);
2165 * It is conceivable that a racing task removed this page from
2166 * swap cache just before we acquired the page lock. The page
2167 * might even be back in swap cache on another swap area. But
2168 * that is okay, try_to_free_swap() only removes stale pages.
2171 wait_on_page_writeback(page
);
2172 try_to_free_swap(page
);
2177 * For frontswap, we just need to unuse pages_to_unuse, if
2178 * it was specified. Need not check frontswap again here as
2179 * we already zeroed out pages_to_unuse if not frontswap.
2181 if (pages_to_unuse
&& --pages_to_unuse
== 0)
2186 * Lets check again to see if there are still swap entries in the map.
2187 * If yes, we would need to do retry the unuse logic again.
2188 * Under global memory pressure, swap entries can be reinserted back
2189 * into process space after the mmlist loop above passes over them.
2191 * Limit the number of retries? No: when mmget_not_zero() above fails,
2192 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2193 * at its own independent pace; and even shmem_writepage() could have
2194 * been preempted after get_swap_page(), temporarily hiding that swap.
2195 * It's easy and robust (though cpu-intensive) just to keep retrying.
2197 if (si
->inuse_pages
) {
2198 if (!signal_pending(current
))
2203 return (retval
== FRONTSWAP_PAGES_UNUSED
) ? 0 : retval
;
2207 * After a successful try_to_unuse, if no swap is now in use, we know
2208 * we can empty the mmlist. swap_lock must be held on entry and exit.
2209 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2210 * added to the mmlist just after page_duplicate - before would be racy.
2212 static void drain_mmlist(void)
2214 struct list_head
*p
, *next
;
2217 for (type
= 0; type
< nr_swapfiles
; type
++)
2218 if (swap_info
[type
]->inuse_pages
)
2220 spin_lock(&mmlist_lock
);
2221 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
2223 spin_unlock(&mmlist_lock
);
2227 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2228 * corresponds to page offset for the specified swap entry.
2229 * Note that the type of this function is sector_t, but it returns page offset
2230 * into the bdev, not sector offset.
2232 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
2234 struct swap_info_struct
*sis
;
2235 struct swap_extent
*start_se
;
2236 struct swap_extent
*se
;
2239 sis
= swp_swap_info(entry
);
2242 offset
= swp_offset(entry
);
2243 start_se
= sis
->curr_swap_extent
;
2247 if (se
->start_page
<= offset
&&
2248 offset
< (se
->start_page
+ se
->nr_pages
)) {
2249 return se
->start_block
+ (offset
- se
->start_page
);
2251 se
= list_next_entry(se
, list
);
2252 sis
->curr_swap_extent
= se
;
2253 BUG_ON(se
== start_se
); /* It *must* be present */
2258 * Returns the page offset into bdev for the specified page's swap entry.
2260 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
2263 entry
.val
= page_private(page
);
2264 return map_swap_entry(entry
, bdev
);
2268 * Free all of a swapdev's extent information
2270 static void destroy_swap_extents(struct swap_info_struct
*sis
)
2272 while (!list_empty(&sis
->first_swap_extent
.list
)) {
2273 struct swap_extent
*se
;
2275 se
= list_first_entry(&sis
->first_swap_extent
.list
,
2276 struct swap_extent
, list
);
2277 list_del(&se
->list
);
2281 if (sis
->flags
& SWP_ACTIVATED
) {
2282 struct file
*swap_file
= sis
->swap_file
;
2283 struct address_space
*mapping
= swap_file
->f_mapping
;
2285 sis
->flags
&= ~SWP_ACTIVATED
;
2286 if (mapping
->a_ops
->swap_deactivate
)
2287 mapping
->a_ops
->swap_deactivate(swap_file
);
2292 * Add a block range (and the corresponding page range) into this swapdev's
2293 * extent list. The extent list is kept sorted in page order.
2295 * This function rather assumes that it is called in ascending page order.
2298 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
2299 unsigned long nr_pages
, sector_t start_block
)
2301 struct swap_extent
*se
;
2302 struct swap_extent
*new_se
;
2303 struct list_head
*lh
;
2305 if (start_page
== 0) {
2306 se
= &sis
->first_swap_extent
;
2307 sis
->curr_swap_extent
= se
;
2309 se
->nr_pages
= nr_pages
;
2310 se
->start_block
= start_block
;
2313 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
2314 se
= list_entry(lh
, struct swap_extent
, list
);
2315 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
2316 if (se
->start_block
+ se
->nr_pages
== start_block
) {
2318 se
->nr_pages
+= nr_pages
;
2324 * No merge. Insert a new extent, preserving ordering.
2326 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
2329 new_se
->start_page
= start_page
;
2330 new_se
->nr_pages
= nr_pages
;
2331 new_se
->start_block
= start_block
;
2333 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
2336 EXPORT_SYMBOL_GPL(add_swap_extent
);
2339 * A `swap extent' is a simple thing which maps a contiguous range of pages
2340 * onto a contiguous range of disk blocks. An ordered list of swap extents
2341 * is built at swapon time and is then used at swap_writepage/swap_readpage
2342 * time for locating where on disk a page belongs.
2344 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2345 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2346 * swap files identically.
2348 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2349 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2350 * swapfiles are handled *identically* after swapon time.
2352 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2353 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2354 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2355 * requirements, they are simply tossed out - we will never use those blocks
2358 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
2359 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
2360 * which will scribble on the fs.
2362 * The amount of disk space which a single swap extent represents varies.
2363 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2364 * extents in the list. To avoid much list walking, we cache the previous
2365 * search location in `curr_swap_extent', and start new searches from there.
2366 * This is extremely effective. The average number of iterations in
2367 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2369 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
2371 struct file
*swap_file
= sis
->swap_file
;
2372 struct address_space
*mapping
= swap_file
->f_mapping
;
2373 struct inode
*inode
= mapping
->host
;
2376 if (S_ISBLK(inode
->i_mode
)) {
2377 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2382 if (mapping
->a_ops
->swap_activate
) {
2383 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
2385 sis
->flags
|= SWP_ACTIVATED
;
2387 sis
->flags
|= SWP_FS
;
2388 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2394 return generic_swapfile_activate(sis
, swap_file
, span
);
2397 static int swap_node(struct swap_info_struct
*p
)
2399 struct block_device
*bdev
;
2404 bdev
= p
->swap_file
->f_inode
->i_sb
->s_bdev
;
2406 return bdev
? bdev
->bd_disk
->node_id
: NUMA_NO_NODE
;
2409 static void setup_swap_info(struct swap_info_struct
*p
, int prio
,
2410 unsigned char *swap_map
,
2411 struct swap_cluster_info
*cluster_info
)
2418 p
->prio
= --least_priority
;
2420 * the plist prio is negated because plist ordering is
2421 * low-to-high, while swap ordering is high-to-low
2423 p
->list
.prio
= -p
->prio
;
2426 p
->avail_lists
[i
].prio
= -p
->prio
;
2428 if (swap_node(p
) == i
)
2429 p
->avail_lists
[i
].prio
= 1;
2431 p
->avail_lists
[i
].prio
= -p
->prio
;
2434 p
->swap_map
= swap_map
;
2435 p
->cluster_info
= cluster_info
;
2438 static void _enable_swap_info(struct swap_info_struct
*p
)
2440 p
->flags
|= SWP_WRITEOK
| SWP_VALID
;
2441 atomic_long_add(p
->pages
, &nr_swap_pages
);
2442 total_swap_pages
+= p
->pages
;
2444 assert_spin_locked(&swap_lock
);
2446 * both lists are plists, and thus priority ordered.
2447 * swap_active_head needs to be priority ordered for swapoff(),
2448 * which on removal of any swap_info_struct with an auto-assigned
2449 * (i.e. negative) priority increments the auto-assigned priority
2450 * of any lower-priority swap_info_structs.
2451 * swap_avail_head needs to be priority ordered for get_swap_page(),
2452 * which allocates swap pages from the highest available priority
2455 plist_add(&p
->list
, &swap_active_head
);
2456 add_to_avail_list(p
);
2459 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
2460 unsigned char *swap_map
,
2461 struct swap_cluster_info
*cluster_info
,
2462 unsigned long *frontswap_map
)
2464 frontswap_init(p
->type
, frontswap_map
);
2465 spin_lock(&swap_lock
);
2466 spin_lock(&p
->lock
);
2467 setup_swap_info(p
, prio
, swap_map
, cluster_info
);
2468 spin_unlock(&p
->lock
);
2469 spin_unlock(&swap_lock
);
2471 * Guarantee swap_map, cluster_info, etc. fields are valid
2472 * between get/put_swap_device() if SWP_VALID bit is set
2475 spin_lock(&swap_lock
);
2476 spin_lock(&p
->lock
);
2477 _enable_swap_info(p
);
2478 spin_unlock(&p
->lock
);
2479 spin_unlock(&swap_lock
);
2482 static void reinsert_swap_info(struct swap_info_struct
*p
)
2484 spin_lock(&swap_lock
);
2485 spin_lock(&p
->lock
);
2486 setup_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
2487 _enable_swap_info(p
);
2488 spin_unlock(&p
->lock
);
2489 spin_unlock(&swap_lock
);
2492 bool has_usable_swap(void)
2496 spin_lock(&swap_lock
);
2497 if (plist_head_empty(&swap_active_head
))
2499 spin_unlock(&swap_lock
);
2503 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
2505 struct swap_info_struct
*p
= NULL
;
2506 unsigned char *swap_map
;
2507 struct swap_cluster_info
*cluster_info
;
2508 unsigned long *frontswap_map
;
2509 struct file
*swap_file
, *victim
;
2510 struct address_space
*mapping
;
2511 struct inode
*inode
;
2512 struct filename
*pathname
;
2514 unsigned int old_block_size
;
2516 if (!capable(CAP_SYS_ADMIN
))
2519 BUG_ON(!current
->mm
);
2521 pathname
= getname(specialfile
);
2522 if (IS_ERR(pathname
))
2523 return PTR_ERR(pathname
);
2525 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
2526 err
= PTR_ERR(victim
);
2530 mapping
= victim
->f_mapping
;
2531 spin_lock(&swap_lock
);
2532 plist_for_each_entry(p
, &swap_active_head
, list
) {
2533 if (p
->flags
& SWP_WRITEOK
) {
2534 if (p
->swap_file
->f_mapping
== mapping
) {
2542 spin_unlock(&swap_lock
);
2545 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2546 vm_unacct_memory(p
->pages
);
2549 spin_unlock(&swap_lock
);
2552 del_from_avail_list(p
);
2553 spin_lock(&p
->lock
);
2555 struct swap_info_struct
*si
= p
;
2558 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2561 for_each_node(nid
) {
2562 if (si
->avail_lists
[nid
].prio
!= 1)
2563 si
->avail_lists
[nid
].prio
--;
2568 plist_del(&p
->list
, &swap_active_head
);
2569 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2570 total_swap_pages
-= p
->pages
;
2571 p
->flags
&= ~SWP_WRITEOK
;
2572 spin_unlock(&p
->lock
);
2573 spin_unlock(&swap_lock
);
2575 disable_swap_slots_cache_lock();
2577 set_current_oom_origin();
2578 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
2579 clear_current_oom_origin();
2582 /* re-insert swap space back into swap_list */
2583 reinsert_swap_info(p
);
2584 reenable_swap_slots_cache_unlock();
2588 reenable_swap_slots_cache_unlock();
2590 spin_lock(&swap_lock
);
2591 spin_lock(&p
->lock
);
2592 p
->flags
&= ~SWP_VALID
; /* mark swap device as invalid */
2593 spin_unlock(&p
->lock
);
2594 spin_unlock(&swap_lock
);
2596 * wait for swap operations protected by get/put_swap_device()
2601 flush_work(&p
->discard_work
);
2603 destroy_swap_extents(p
);
2604 if (p
->flags
& SWP_CONTINUED
)
2605 free_swap_count_continuations(p
);
2607 if (!p
->bdev
|| !blk_queue_nonrot(bdev_get_queue(p
->bdev
)))
2608 atomic_dec(&nr_rotate_swap
);
2610 mutex_lock(&swapon_mutex
);
2611 spin_lock(&swap_lock
);
2612 spin_lock(&p
->lock
);
2615 /* wait for anyone still in scan_swap_map */
2616 p
->highest_bit
= 0; /* cuts scans short */
2617 while (p
->flags
>= SWP_SCANNING
) {
2618 spin_unlock(&p
->lock
);
2619 spin_unlock(&swap_lock
);
2620 schedule_timeout_uninterruptible(1);
2621 spin_lock(&swap_lock
);
2622 spin_lock(&p
->lock
);
2625 swap_file
= p
->swap_file
;
2626 old_block_size
= p
->old_block_size
;
2627 p
->swap_file
= NULL
;
2629 swap_map
= p
->swap_map
;
2631 cluster_info
= p
->cluster_info
;
2632 p
->cluster_info
= NULL
;
2633 frontswap_map
= frontswap_map_get(p
);
2634 spin_unlock(&p
->lock
);
2635 spin_unlock(&swap_lock
);
2636 frontswap_invalidate_area(p
->type
);
2637 frontswap_map_set(p
, NULL
);
2638 mutex_unlock(&swapon_mutex
);
2639 free_percpu(p
->percpu_cluster
);
2640 p
->percpu_cluster
= NULL
;
2642 kvfree(cluster_info
);
2643 kvfree(frontswap_map
);
2644 /* Destroy swap account information */
2645 swap_cgroup_swapoff(p
->type
);
2646 exit_swap_address_space(p
->type
);
2648 inode
= mapping
->host
;
2649 if (S_ISBLK(inode
->i_mode
)) {
2650 struct block_device
*bdev
= I_BDEV(inode
);
2651 set_blocksize(bdev
, old_block_size
);
2652 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2655 inode
->i_flags
&= ~S_SWAPFILE
;
2656 inode_unlock(inode
);
2658 filp_close(swap_file
, NULL
);
2661 * Clear the SWP_USED flag after all resources are freed so that swapon
2662 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2663 * not hold p->lock after we cleared its SWP_WRITEOK.
2665 spin_lock(&swap_lock
);
2667 spin_unlock(&swap_lock
);
2670 atomic_inc(&proc_poll_event
);
2671 wake_up_interruptible(&proc_poll_wait
);
2674 filp_close(victim
, NULL
);
2680 #ifdef CONFIG_PROC_FS
2681 static __poll_t
swaps_poll(struct file
*file
, poll_table
*wait
)
2683 struct seq_file
*seq
= file
->private_data
;
2685 poll_wait(file
, &proc_poll_wait
, wait
);
2687 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2688 seq
->poll_event
= atomic_read(&proc_poll_event
);
2689 return EPOLLIN
| EPOLLRDNORM
| EPOLLERR
| EPOLLPRI
;
2692 return EPOLLIN
| EPOLLRDNORM
;
2696 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2698 struct swap_info_struct
*si
;
2702 mutex_lock(&swapon_mutex
);
2705 return SEQ_START_TOKEN
;
2707 for (type
= 0; (si
= swap_type_to_swap_info(type
)); type
++) {
2708 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2717 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2719 struct swap_info_struct
*si
= v
;
2722 if (v
== SEQ_START_TOKEN
)
2725 type
= si
->type
+ 1;
2727 for (; (si
= swap_type_to_swap_info(type
)); type
++) {
2728 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2737 static void swap_stop(struct seq_file
*swap
, void *v
)
2739 mutex_unlock(&swapon_mutex
);
2742 static int swap_show(struct seq_file
*swap
, void *v
)
2744 struct swap_info_struct
*si
= v
;
2748 if (si
== SEQ_START_TOKEN
) {
2749 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2753 file
= si
->swap_file
;
2754 len
= seq_file_path(swap
, file
, " \t\n\\");
2755 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
2756 len
< 40 ? 40 - len
: 1, " ",
2757 S_ISBLK(file_inode(file
)->i_mode
) ?
2758 "partition" : "file\t",
2759 si
->pages
<< (PAGE_SHIFT
- 10),
2760 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
2765 static const struct seq_operations swaps_op
= {
2766 .start
= swap_start
,
2772 static int swaps_open(struct inode
*inode
, struct file
*file
)
2774 struct seq_file
*seq
;
2777 ret
= seq_open(file
, &swaps_op
);
2781 seq
= file
->private_data
;
2782 seq
->poll_event
= atomic_read(&proc_poll_event
);
2786 static const struct file_operations proc_swaps_operations
= {
2789 .llseek
= seq_lseek
,
2790 .release
= seq_release
,
2794 static int __init
procswaps_init(void)
2796 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
2799 __initcall(procswaps_init
);
2800 #endif /* CONFIG_PROC_FS */
2802 #ifdef MAX_SWAPFILES_CHECK
2803 static int __init
max_swapfiles_check(void)
2805 MAX_SWAPFILES_CHECK();
2808 late_initcall(max_swapfiles_check
);
2811 static struct swap_info_struct
*alloc_swap_info(void)
2813 struct swap_info_struct
*p
;
2817 p
= kvzalloc(struct_size(p
, avail_lists
, nr_node_ids
), GFP_KERNEL
);
2819 return ERR_PTR(-ENOMEM
);
2821 spin_lock(&swap_lock
);
2822 for (type
= 0; type
< nr_swapfiles
; type
++) {
2823 if (!(swap_info
[type
]->flags
& SWP_USED
))
2826 if (type
>= MAX_SWAPFILES
) {
2827 spin_unlock(&swap_lock
);
2829 return ERR_PTR(-EPERM
);
2831 if (type
>= nr_swapfiles
) {
2833 WRITE_ONCE(swap_info
[type
], p
);
2835 * Write swap_info[type] before nr_swapfiles, in case a
2836 * racing procfs swap_start() or swap_next() is reading them.
2837 * (We never shrink nr_swapfiles, we never free this entry.)
2840 WRITE_ONCE(nr_swapfiles
, nr_swapfiles
+ 1);
2843 p
= swap_info
[type
];
2845 * Do not memset this entry: a racing procfs swap_next()
2846 * would be relying on p->type to remain valid.
2849 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
2850 plist_node_init(&p
->list
, 0);
2852 plist_node_init(&p
->avail_lists
[i
], 0);
2853 p
->flags
= SWP_USED
;
2854 spin_unlock(&swap_lock
);
2855 spin_lock_init(&p
->lock
);
2856 spin_lock_init(&p
->cont_lock
);
2861 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2865 if (S_ISBLK(inode
->i_mode
)) {
2866 p
->bdev
= bdgrab(I_BDEV(inode
));
2867 error
= blkdev_get(p
->bdev
,
2868 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, p
);
2873 p
->old_block_size
= block_size(p
->bdev
);
2874 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2877 p
->flags
|= SWP_BLKDEV
;
2878 } else if (S_ISREG(inode
->i_mode
)) {
2879 p
->bdev
= inode
->i_sb
->s_bdev
;
2881 if (IS_SWAPFILE(inode
))
2891 * Find out how many pages are allowed for a single swap device. There
2892 * are two limiting factors:
2893 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2894 * 2) the number of bits in the swap pte, as defined by the different
2897 * In order to find the largest possible bit mask, a swap entry with
2898 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2899 * decoded to a swp_entry_t again, and finally the swap offset is
2902 * This will mask all the bits from the initial ~0UL mask that can't
2903 * be encoded in either the swp_entry_t or the architecture definition
2906 unsigned long generic_max_swapfile_size(void)
2908 return swp_offset(pte_to_swp_entry(
2909 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2912 /* Can be overridden by an architecture for additional checks. */
2913 __weak
unsigned long max_swapfile_size(void)
2915 return generic_max_swapfile_size();
2918 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2919 union swap_header
*swap_header
,
2920 struct inode
*inode
)
2923 unsigned long maxpages
;
2924 unsigned long swapfilepages
;
2925 unsigned long last_page
;
2927 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2928 pr_err("Unable to find swap-space signature\n");
2932 /* swap partition endianess hack... */
2933 if (swab32(swap_header
->info
.version
) == 1) {
2934 swab32s(&swap_header
->info
.version
);
2935 swab32s(&swap_header
->info
.last_page
);
2936 swab32s(&swap_header
->info
.nr_badpages
);
2937 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2939 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2940 swab32s(&swap_header
->info
.badpages
[i
]);
2942 /* Check the swap header's sub-version */
2943 if (swap_header
->info
.version
!= 1) {
2944 pr_warn("Unable to handle swap header version %d\n",
2945 swap_header
->info
.version
);
2950 p
->cluster_next
= 1;
2953 maxpages
= max_swapfile_size();
2954 last_page
= swap_header
->info
.last_page
;
2956 pr_warn("Empty swap-file\n");
2959 if (last_page
> maxpages
) {
2960 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2961 maxpages
<< (PAGE_SHIFT
- 10),
2962 last_page
<< (PAGE_SHIFT
- 10));
2964 if (maxpages
> last_page
) {
2965 maxpages
= last_page
+ 1;
2966 /* p->max is an unsigned int: don't overflow it */
2967 if ((unsigned int)maxpages
== 0)
2968 maxpages
= UINT_MAX
;
2970 p
->highest_bit
= maxpages
- 1;
2974 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
2975 if (swapfilepages
&& maxpages
> swapfilepages
) {
2976 pr_warn("Swap area shorter than signature indicates\n");
2979 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
2981 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2987 #define SWAP_CLUSTER_INFO_COLS \
2988 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2989 #define SWAP_CLUSTER_SPACE_COLS \
2990 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2991 #define SWAP_CLUSTER_COLS \
2992 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2994 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
2995 union swap_header
*swap_header
,
2996 unsigned char *swap_map
,
2997 struct swap_cluster_info
*cluster_info
,
2998 unsigned long maxpages
,
3002 unsigned int nr_good_pages
;
3004 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3005 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
3006 unsigned long i
, idx
;
3008 nr_good_pages
= maxpages
- 1; /* omit header page */
3010 cluster_list_init(&p
->free_clusters
);
3011 cluster_list_init(&p
->discard_clusters
);
3013 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
3014 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
3015 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
3017 if (page_nr
< maxpages
) {
3018 swap_map
[page_nr
] = SWAP_MAP_BAD
;
3021 * Haven't marked the cluster free yet, no list
3022 * operation involved
3024 inc_cluster_info_page(p
, cluster_info
, page_nr
);
3028 /* Haven't marked the cluster free yet, no list operation involved */
3029 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
3030 inc_cluster_info_page(p
, cluster_info
, i
);
3032 if (nr_good_pages
) {
3033 swap_map
[0] = SWAP_MAP_BAD
;
3035 * Not mark the cluster free yet, no list
3036 * operation involved
3038 inc_cluster_info_page(p
, cluster_info
, 0);
3040 p
->pages
= nr_good_pages
;
3041 nr_extents
= setup_swap_extents(p
, span
);
3044 nr_good_pages
= p
->pages
;
3046 if (!nr_good_pages
) {
3047 pr_warn("Empty swap-file\n");
3056 * Reduce false cache line sharing between cluster_info and
3057 * sharing same address space.
3059 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
3060 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
3061 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
3062 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
3063 if (idx
>= nr_clusters
)
3065 if (cluster_count(&cluster_info
[idx
]))
3067 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
3068 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
3076 * Helper to sys_swapon determining if a given swap
3077 * backing device queue supports DISCARD operations.
3079 static bool swap_discardable(struct swap_info_struct
*si
)
3081 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
3083 if (!q
|| !blk_queue_discard(q
))
3089 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
3091 struct swap_info_struct
*p
;
3092 struct filename
*name
;
3093 struct file
*swap_file
= NULL
;
3094 struct address_space
*mapping
;
3097 union swap_header
*swap_header
;
3100 unsigned long maxpages
;
3101 unsigned char *swap_map
= NULL
;
3102 struct swap_cluster_info
*cluster_info
= NULL
;
3103 unsigned long *frontswap_map
= NULL
;
3104 struct page
*page
= NULL
;
3105 struct inode
*inode
= NULL
;
3106 bool inced_nr_rotate_swap
= false;
3108 if (swap_flags
& ~SWAP_FLAGS_VALID
)
3111 if (!capable(CAP_SYS_ADMIN
))
3114 if (!swap_avail_heads
)
3117 p
= alloc_swap_info();
3121 INIT_WORK(&p
->discard_work
, swap_discard_work
);
3123 name
= getname(specialfile
);
3125 error
= PTR_ERR(name
);
3129 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
3130 if (IS_ERR(swap_file
)) {
3131 error
= PTR_ERR(swap_file
);
3136 p
->swap_file
= swap_file
;
3137 mapping
= swap_file
->f_mapping
;
3138 inode
= mapping
->host
;
3140 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
3141 error
= claim_swapfile(p
, inode
);
3142 if (unlikely(error
))
3146 * Read the swap header.
3148 if (!mapping
->a_ops
->readpage
) {
3152 page
= read_mapping_page(mapping
, 0, swap_file
);
3154 error
= PTR_ERR(page
);
3157 swap_header
= kmap(page
);
3159 maxpages
= read_swap_header(p
, swap_header
, inode
);
3160 if (unlikely(!maxpages
)) {
3165 /* OK, set up the swap map and apply the bad block list */
3166 swap_map
= vzalloc(maxpages
);
3172 if (bdi_cap_stable_pages_required(inode_to_bdi(inode
)))
3173 p
->flags
|= SWP_STABLE_WRITES
;
3175 if (bdi_cap_synchronous_io(inode_to_bdi(inode
)))
3176 p
->flags
|= SWP_SYNCHRONOUS_IO
;
3178 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
3180 unsigned long ci
, nr_cluster
;
3182 p
->flags
|= SWP_SOLIDSTATE
;
3184 * select a random position to start with to help wear leveling
3187 p
->cluster_next
= 1 + (prandom_u32() % p
->highest_bit
);
3188 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3190 cluster_info
= kvcalloc(nr_cluster
, sizeof(*cluster_info
),
3192 if (!cluster_info
) {
3197 for (ci
= 0; ci
< nr_cluster
; ci
++)
3198 spin_lock_init(&((cluster_info
+ ci
)->lock
));
3200 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
3201 if (!p
->percpu_cluster
) {
3205 for_each_possible_cpu(cpu
) {
3206 struct percpu_cluster
*cluster
;
3207 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
3208 cluster_set_null(&cluster
->index
);
3211 atomic_inc(&nr_rotate_swap
);
3212 inced_nr_rotate_swap
= true;
3215 error
= swap_cgroup_swapon(p
->type
, maxpages
);
3219 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
3220 cluster_info
, maxpages
, &span
);
3221 if (unlikely(nr_extents
< 0)) {
3225 /* frontswap enabled? set up bit-per-page map for frontswap */
3226 if (IS_ENABLED(CONFIG_FRONTSWAP
))
3227 frontswap_map
= kvcalloc(BITS_TO_LONGS(maxpages
),
3231 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
3233 * When discard is enabled for swap with no particular
3234 * policy flagged, we set all swap discard flags here in
3235 * order to sustain backward compatibility with older
3236 * swapon(8) releases.
3238 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
3242 * By flagging sys_swapon, a sysadmin can tell us to
3243 * either do single-time area discards only, or to just
3244 * perform discards for released swap page-clusters.
3245 * Now it's time to adjust the p->flags accordingly.
3247 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
3248 p
->flags
&= ~SWP_PAGE_DISCARD
;
3249 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
3250 p
->flags
&= ~SWP_AREA_DISCARD
;
3252 /* issue a swapon-time discard if it's still required */
3253 if (p
->flags
& SWP_AREA_DISCARD
) {
3254 int err
= discard_swap(p
);
3256 pr_err("swapon: discard_swap(%p): %d\n",
3261 error
= init_swap_address_space(p
->type
, maxpages
);
3265 mutex_lock(&swapon_mutex
);
3267 if (swap_flags
& SWAP_FLAG_PREFER
)
3269 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
3270 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
3272 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3273 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
3274 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
3275 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
3276 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
3277 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
3278 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
3279 (frontswap_map
) ? "FS" : "");
3281 mutex_unlock(&swapon_mutex
);
3282 atomic_inc(&proc_poll_event
);
3283 wake_up_interruptible(&proc_poll_wait
);
3285 if (S_ISREG(inode
->i_mode
))
3286 inode
->i_flags
|= S_SWAPFILE
;
3290 free_percpu(p
->percpu_cluster
);
3291 p
->percpu_cluster
= NULL
;
3292 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
3293 set_blocksize(p
->bdev
, p
->old_block_size
);
3294 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
3296 destroy_swap_extents(p
);
3297 swap_cgroup_swapoff(p
->type
);
3298 spin_lock(&swap_lock
);
3299 p
->swap_file
= NULL
;
3301 spin_unlock(&swap_lock
);
3303 kvfree(cluster_info
);
3304 kvfree(frontswap_map
);
3305 if (inced_nr_rotate_swap
)
3306 atomic_dec(&nr_rotate_swap
);
3308 if (inode
&& S_ISREG(inode
->i_mode
)) {
3309 inode_unlock(inode
);
3312 filp_close(swap_file
, NULL
);
3315 if (page
&& !IS_ERR(page
)) {
3321 if (inode
&& S_ISREG(inode
->i_mode
))
3322 inode_unlock(inode
);
3324 enable_swap_slots_cache();
3328 void si_swapinfo(struct sysinfo
*val
)
3331 unsigned long nr_to_be_unused
= 0;
3333 spin_lock(&swap_lock
);
3334 for (type
= 0; type
< nr_swapfiles
; type
++) {
3335 struct swap_info_struct
*si
= swap_info
[type
];
3337 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
3338 nr_to_be_unused
+= si
->inuse_pages
;
3340 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
3341 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
3342 spin_unlock(&swap_lock
);
3346 * Verify that a swap entry is valid and increment its swap map count.
3348 * Returns error code in following case.
3350 * - swp_entry is invalid -> EINVAL
3351 * - swp_entry is migration entry -> EINVAL
3352 * - swap-cache reference is requested but there is already one. -> EEXIST
3353 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3354 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3356 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
3358 struct swap_info_struct
*p
;
3359 struct swap_cluster_info
*ci
;
3360 unsigned long offset
;
3361 unsigned char count
;
3362 unsigned char has_cache
;
3365 p
= get_swap_device(entry
);
3369 offset
= swp_offset(entry
);
3370 ci
= lock_cluster_or_swap_info(p
, offset
);
3372 count
= p
->swap_map
[offset
];
3375 * swapin_readahead() doesn't check if a swap entry is valid, so the
3376 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3378 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
3383 has_cache
= count
& SWAP_HAS_CACHE
;
3384 count
&= ~SWAP_HAS_CACHE
;
3387 if (usage
== SWAP_HAS_CACHE
) {
3389 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3390 if (!has_cache
&& count
)
3391 has_cache
= SWAP_HAS_CACHE
;
3392 else if (has_cache
) /* someone else added cache */
3394 else /* no users remaining */
3397 } else if (count
|| has_cache
) {
3399 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
3401 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
3403 else if (swap_count_continued(p
, offset
, count
))
3404 count
= COUNT_CONTINUED
;
3408 err
= -ENOENT
; /* unused swap entry */
3410 p
->swap_map
[offset
] = count
| has_cache
;
3413 unlock_cluster_or_swap_info(p
, ci
);
3421 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3422 * (in which case its reference count is never incremented).
3424 void swap_shmem_alloc(swp_entry_t entry
)
3426 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
3430 * Increase reference count of swap entry by 1.
3431 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3432 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3433 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3434 * might occur if a page table entry has got corrupted.
3436 int swap_duplicate(swp_entry_t entry
)
3440 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
3441 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
3446 * @entry: swap entry for which we allocate swap cache.
3448 * Called when allocating swap cache for existing swap entry,
3449 * This can return error codes. Returns 0 at success.
3450 * -EBUSY means there is a swap cache.
3451 * Note: return code is different from swap_duplicate().
3453 int swapcache_prepare(swp_entry_t entry
)
3455 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
3458 struct swap_info_struct
*swp_swap_info(swp_entry_t entry
)
3460 return swap_type_to_swap_info(swp_type(entry
));
3463 struct swap_info_struct
*page_swap_info(struct page
*page
)
3465 swp_entry_t entry
= { .val
= page_private(page
) };
3466 return swp_swap_info(entry
);
3470 * out-of-line __page_file_ methods to avoid include hell.
3472 struct address_space
*__page_file_mapping(struct page
*page
)
3474 return page_swap_info(page
)->swap_file
->f_mapping
;
3476 EXPORT_SYMBOL_GPL(__page_file_mapping
);
3478 pgoff_t
__page_file_index(struct page
*page
)
3480 swp_entry_t swap
= { .val
= page_private(page
) };
3481 return swp_offset(swap
);
3483 EXPORT_SYMBOL_GPL(__page_file_index
);
3486 * add_swap_count_continuation - called when a swap count is duplicated
3487 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3488 * page of the original vmalloc'ed swap_map, to hold the continuation count
3489 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3490 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3492 * These continuation pages are seldom referenced: the common paths all work
3493 * on the original swap_map, only referring to a continuation page when the
3494 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3496 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3497 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3498 * can be called after dropping locks.
3500 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
3502 struct swap_info_struct
*si
;
3503 struct swap_cluster_info
*ci
;
3506 struct page
*list_page
;
3508 unsigned char count
;
3512 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3513 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3515 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
3517 si
= get_swap_device(entry
);
3520 * An acceptable race has occurred since the failing
3521 * __swap_duplicate(): the swap device may be swapoff
3525 spin_lock(&si
->lock
);
3527 offset
= swp_offset(entry
);
3529 ci
= lock_cluster(si
, offset
);
3531 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
3533 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
3535 * The higher the swap count, the more likely it is that tasks
3536 * will race to add swap count continuation: we need to avoid
3537 * over-provisioning.
3548 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3549 * no architecture is using highmem pages for kernel page tables: so it
3550 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3552 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3553 offset
&= ~PAGE_MASK
;
3555 spin_lock(&si
->cont_lock
);
3557 * Page allocation does not initialize the page's lru field,
3558 * but it does always reset its private field.
3560 if (!page_private(head
)) {
3561 BUG_ON(count
& COUNT_CONTINUED
);
3562 INIT_LIST_HEAD(&head
->lru
);
3563 set_page_private(head
, SWP_CONTINUED
);
3564 si
->flags
|= SWP_CONTINUED
;
3567 list_for_each_entry(list_page
, &head
->lru
, lru
) {
3571 * If the previous map said no continuation, but we've found
3572 * a continuation page, free our allocation and use this one.
3574 if (!(count
& COUNT_CONTINUED
))
3575 goto out_unlock_cont
;
3577 map
= kmap_atomic(list_page
) + offset
;
3582 * If this continuation count now has some space in it,
3583 * free our allocation and use this one.
3585 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
3586 goto out_unlock_cont
;
3589 list_add_tail(&page
->lru
, &head
->lru
);
3590 page
= NULL
; /* now it's attached, don't free it */
3592 spin_unlock(&si
->cont_lock
);
3595 spin_unlock(&si
->lock
);
3596 put_swap_device(si
);
3604 * swap_count_continued - when the original swap_map count is incremented
3605 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3606 * into, carry if so, or else fail until a new continuation page is allocated;
3607 * when the original swap_map count is decremented from 0 with continuation,
3608 * borrow from the continuation and report whether it still holds more.
3609 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3612 static bool swap_count_continued(struct swap_info_struct
*si
,
3613 pgoff_t offset
, unsigned char count
)
3620 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3621 if (page_private(head
) != SWP_CONTINUED
) {
3622 BUG_ON(count
& COUNT_CONTINUED
);
3623 return false; /* need to add count continuation */
3626 spin_lock(&si
->cont_lock
);
3627 offset
&= ~PAGE_MASK
;
3628 page
= list_entry(head
->lru
.next
, struct page
, lru
);
3629 map
= kmap_atomic(page
) + offset
;
3631 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3632 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3634 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3636 * Think of how you add 1 to 999
3638 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3640 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3641 BUG_ON(page
== head
);
3642 map
= kmap_atomic(page
) + offset
;
3644 if (*map
== SWAP_CONT_MAX
) {
3646 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3648 ret
= false; /* add count continuation */
3651 map
= kmap_atomic(page
) + offset
;
3652 init_map
: *map
= 0; /* we didn't zero the page */
3656 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3657 while (page
!= head
) {
3658 map
= kmap_atomic(page
) + offset
;
3659 *map
= COUNT_CONTINUED
;
3661 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3663 ret
= true; /* incremented */
3665 } else { /* decrementing */
3667 * Think of how you subtract 1 from 1000
3669 BUG_ON(count
!= COUNT_CONTINUED
);
3670 while (*map
== COUNT_CONTINUED
) {
3672 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3673 BUG_ON(page
== head
);
3674 map
= kmap_atomic(page
) + offset
;
3681 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3682 while (page
!= head
) {
3683 map
= kmap_atomic(page
) + offset
;
3684 *map
= SWAP_CONT_MAX
| count
;
3685 count
= COUNT_CONTINUED
;
3687 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3689 ret
= count
== COUNT_CONTINUED
;
3692 spin_unlock(&si
->cont_lock
);
3697 * free_swap_count_continuations - swapoff free all the continuation pages
3698 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3700 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3704 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3706 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3707 if (page_private(head
)) {
3708 struct page
*page
, *next
;
3710 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3711 list_del(&page
->lru
);
3718 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3719 void mem_cgroup_throttle_swaprate(struct mem_cgroup
*memcg
, int node
,
3722 struct swap_info_struct
*si
, *next
;
3723 if (!(gfp_mask
& __GFP_IO
) || !memcg
)
3726 if (!blk_cgroup_congested())
3730 * We've already scheduled a throttle, avoid taking the global swap
3733 if (current
->throttle_queue
)
3736 spin_lock(&swap_avail_lock
);
3737 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[node
],
3738 avail_lists
[node
]) {
3740 blkcg_schedule_throttle(bdev_get_queue(si
->bdev
),
3745 spin_unlock(&swap_avail_lock
);
3749 static int __init
swapfile_init(void)
3753 swap_avail_heads
= kmalloc_array(nr_node_ids
, sizeof(struct plist_head
),
3755 if (!swap_avail_heads
) {
3756 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3761 plist_head_init(&swap_avail_heads
[nid
]);
3765 subsys_initcall(swapfile_init
);