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
, type
;
1086 type
= swp_type(entry
);
1087 p
= swap_type_to_swap_info(type
);
1090 if (!(p
->flags
& SWP_USED
))
1092 offset
= swp_offset(entry
);
1093 if (offset
>= p
->max
)
1098 pr_err("swap_info_get: %s%08lx\n", Bad_offset
, entry
.val
);
1101 pr_err("swap_info_get: %s%08lx\n", Unused_file
, entry
.val
);
1104 pr_err("swap_info_get: %s%08lx\n", Bad_file
, entry
.val
);
1109 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
1111 struct swap_info_struct
*p
;
1113 p
= __swap_info_get(entry
);
1116 if (!p
->swap_map
[swp_offset(entry
)])
1121 pr_err("swap_info_get: %s%08lx\n", Unused_offset
, entry
.val
);
1127 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
1129 struct swap_info_struct
*p
;
1131 p
= _swap_info_get(entry
);
1133 spin_lock(&p
->lock
);
1137 static struct swap_info_struct
*swap_info_get_cont(swp_entry_t entry
,
1138 struct swap_info_struct
*q
)
1140 struct swap_info_struct
*p
;
1142 p
= _swap_info_get(entry
);
1146 spin_unlock(&q
->lock
);
1148 spin_lock(&p
->lock
);
1153 static unsigned char __swap_entry_free_locked(struct swap_info_struct
*p
,
1154 unsigned long offset
,
1155 unsigned char usage
)
1157 unsigned char count
;
1158 unsigned char has_cache
;
1160 count
= p
->swap_map
[offset
];
1162 has_cache
= count
& SWAP_HAS_CACHE
;
1163 count
&= ~SWAP_HAS_CACHE
;
1165 if (usage
== SWAP_HAS_CACHE
) {
1166 VM_BUG_ON(!has_cache
);
1168 } else if (count
== SWAP_MAP_SHMEM
) {
1170 * Or we could insist on shmem.c using a special
1171 * swap_shmem_free() and free_shmem_swap_and_cache()...
1174 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
1175 if (count
== COUNT_CONTINUED
) {
1176 if (swap_count_continued(p
, offset
, count
))
1177 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
1179 count
= SWAP_MAP_MAX
;
1184 usage
= count
| has_cache
;
1185 p
->swap_map
[offset
] = usage
? : SWAP_HAS_CACHE
;
1190 static unsigned char __swap_entry_free(struct swap_info_struct
*p
,
1191 swp_entry_t entry
, unsigned char usage
)
1193 struct swap_cluster_info
*ci
;
1194 unsigned long offset
= swp_offset(entry
);
1196 ci
= lock_cluster_or_swap_info(p
, offset
);
1197 usage
= __swap_entry_free_locked(p
, offset
, usage
);
1198 unlock_cluster_or_swap_info(p
, ci
);
1200 free_swap_slot(entry
);
1205 static void swap_entry_free(struct swap_info_struct
*p
, swp_entry_t entry
)
1207 struct swap_cluster_info
*ci
;
1208 unsigned long offset
= swp_offset(entry
);
1209 unsigned char count
;
1211 ci
= lock_cluster(p
, offset
);
1212 count
= p
->swap_map
[offset
];
1213 VM_BUG_ON(count
!= SWAP_HAS_CACHE
);
1214 p
->swap_map
[offset
] = 0;
1215 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
1218 mem_cgroup_uncharge_swap(entry
, 1);
1219 swap_range_free(p
, offset
, 1);
1223 * Caller has made sure that the swap device corresponding to entry
1224 * is still around or has not been recycled.
1226 void swap_free(swp_entry_t entry
)
1228 struct swap_info_struct
*p
;
1230 p
= _swap_info_get(entry
);
1232 __swap_entry_free(p
, entry
, 1);
1236 * Called after dropping swapcache to decrease refcnt to swap entries.
1238 void put_swap_page(struct page
*page
, swp_entry_t entry
)
1240 unsigned long offset
= swp_offset(entry
);
1241 unsigned long idx
= offset
/ SWAPFILE_CLUSTER
;
1242 struct swap_cluster_info
*ci
;
1243 struct swap_info_struct
*si
;
1245 unsigned int i
, free_entries
= 0;
1247 int size
= swap_entry_size(hpage_nr_pages(page
));
1249 si
= _swap_info_get(entry
);
1253 ci
= lock_cluster_or_swap_info(si
, offset
);
1254 if (size
== SWAPFILE_CLUSTER
) {
1255 VM_BUG_ON(!cluster_is_huge(ci
));
1256 map
= si
->swap_map
+ offset
;
1257 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1259 VM_BUG_ON(!(val
& SWAP_HAS_CACHE
));
1260 if (val
== SWAP_HAS_CACHE
)
1263 cluster_clear_huge(ci
);
1264 if (free_entries
== SWAPFILE_CLUSTER
) {
1265 unlock_cluster_or_swap_info(si
, ci
);
1266 spin_lock(&si
->lock
);
1267 mem_cgroup_uncharge_swap(entry
, SWAPFILE_CLUSTER
);
1268 swap_free_cluster(si
, idx
);
1269 spin_unlock(&si
->lock
);
1273 for (i
= 0; i
< size
; i
++, entry
.val
++) {
1274 if (!__swap_entry_free_locked(si
, offset
+ i
, SWAP_HAS_CACHE
)) {
1275 unlock_cluster_or_swap_info(si
, ci
);
1276 free_swap_slot(entry
);
1279 lock_cluster_or_swap_info(si
, offset
);
1282 unlock_cluster_or_swap_info(si
, ci
);
1285 #ifdef CONFIG_THP_SWAP
1286 int split_swap_cluster(swp_entry_t entry
)
1288 struct swap_info_struct
*si
;
1289 struct swap_cluster_info
*ci
;
1290 unsigned long offset
= swp_offset(entry
);
1292 si
= _swap_info_get(entry
);
1295 ci
= lock_cluster(si
, offset
);
1296 cluster_clear_huge(ci
);
1302 static int swp_entry_cmp(const void *ent1
, const void *ent2
)
1304 const swp_entry_t
*e1
= ent1
, *e2
= ent2
;
1306 return (int)swp_type(*e1
) - (int)swp_type(*e2
);
1309 void swapcache_free_entries(swp_entry_t
*entries
, int n
)
1311 struct swap_info_struct
*p
, *prev
;
1321 * Sort swap entries by swap device, so each lock is only taken once.
1322 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1323 * so low that it isn't necessary to optimize further.
1325 if (nr_swapfiles
> 1)
1326 sort(entries
, n
, sizeof(entries
[0]), swp_entry_cmp
, NULL
);
1327 for (i
= 0; i
< n
; ++i
) {
1328 p
= swap_info_get_cont(entries
[i
], prev
);
1330 swap_entry_free(p
, entries
[i
]);
1334 spin_unlock(&p
->lock
);
1338 * How many references to page are currently swapped out?
1339 * This does not give an exact answer when swap count is continued,
1340 * but does include the high COUNT_CONTINUED flag to allow for that.
1342 int page_swapcount(struct page
*page
)
1345 struct swap_info_struct
*p
;
1346 struct swap_cluster_info
*ci
;
1348 unsigned long offset
;
1350 entry
.val
= page_private(page
);
1351 p
= _swap_info_get(entry
);
1353 offset
= swp_offset(entry
);
1354 ci
= lock_cluster_or_swap_info(p
, offset
);
1355 count
= swap_count(p
->swap_map
[offset
]);
1356 unlock_cluster_or_swap_info(p
, ci
);
1361 int __swap_count(struct swap_info_struct
*si
, swp_entry_t entry
)
1363 pgoff_t offset
= swp_offset(entry
);
1365 return swap_count(si
->swap_map
[offset
]);
1368 static int swap_swapcount(struct swap_info_struct
*si
, swp_entry_t entry
)
1371 pgoff_t offset
= swp_offset(entry
);
1372 struct swap_cluster_info
*ci
;
1374 ci
= lock_cluster_or_swap_info(si
, offset
);
1375 count
= swap_count(si
->swap_map
[offset
]);
1376 unlock_cluster_or_swap_info(si
, ci
);
1381 * How many references to @entry are currently swapped out?
1382 * This does not give an exact answer when swap count is continued,
1383 * but does include the high COUNT_CONTINUED flag to allow for that.
1385 int __swp_swapcount(swp_entry_t entry
)
1388 struct swap_info_struct
*si
;
1390 si
= __swap_info_get(entry
);
1392 count
= swap_swapcount(si
, entry
);
1397 * How many references to @entry are currently swapped out?
1398 * This considers COUNT_CONTINUED so it returns exact answer.
1400 int swp_swapcount(swp_entry_t entry
)
1402 int count
, tmp_count
, n
;
1403 struct swap_info_struct
*p
;
1404 struct swap_cluster_info
*ci
;
1409 p
= _swap_info_get(entry
);
1413 offset
= swp_offset(entry
);
1415 ci
= lock_cluster_or_swap_info(p
, offset
);
1417 count
= swap_count(p
->swap_map
[offset
]);
1418 if (!(count
& COUNT_CONTINUED
))
1421 count
&= ~COUNT_CONTINUED
;
1422 n
= SWAP_MAP_MAX
+ 1;
1424 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1425 offset
&= ~PAGE_MASK
;
1426 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1429 page
= list_next_entry(page
, lru
);
1430 map
= kmap_atomic(page
);
1431 tmp_count
= map
[offset
];
1434 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1435 n
*= (SWAP_CONT_MAX
+ 1);
1436 } while (tmp_count
& COUNT_CONTINUED
);
1438 unlock_cluster_or_swap_info(p
, ci
);
1442 static bool swap_page_trans_huge_swapped(struct swap_info_struct
*si
,
1445 struct swap_cluster_info
*ci
;
1446 unsigned char *map
= si
->swap_map
;
1447 unsigned long roffset
= swp_offset(entry
);
1448 unsigned long offset
= round_down(roffset
, SWAPFILE_CLUSTER
);
1452 ci
= lock_cluster_or_swap_info(si
, offset
);
1453 if (!ci
|| !cluster_is_huge(ci
)) {
1454 if (swap_count(map
[roffset
]))
1458 for (i
= 0; i
< SWAPFILE_CLUSTER
; i
++) {
1459 if (swap_count(map
[offset
+ i
])) {
1465 unlock_cluster_or_swap_info(si
, ci
);
1469 static bool page_swapped(struct page
*page
)
1472 struct swap_info_struct
*si
;
1474 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
)))
1475 return page_swapcount(page
) != 0;
1477 page
= compound_head(page
);
1478 entry
.val
= page_private(page
);
1479 si
= _swap_info_get(entry
);
1481 return swap_page_trans_huge_swapped(si
, entry
);
1485 static int page_trans_huge_map_swapcount(struct page
*page
, int *total_mapcount
,
1486 int *total_swapcount
)
1488 int i
, map_swapcount
, _total_mapcount
, _total_swapcount
;
1489 unsigned long offset
= 0;
1490 struct swap_info_struct
*si
;
1491 struct swap_cluster_info
*ci
= NULL
;
1492 unsigned char *map
= NULL
;
1493 int mapcount
, swapcount
= 0;
1495 /* hugetlbfs shouldn't call it */
1496 VM_BUG_ON_PAGE(PageHuge(page
), page
);
1498 if (!IS_ENABLED(CONFIG_THP_SWAP
) || likely(!PageTransCompound(page
))) {
1499 mapcount
= page_trans_huge_mapcount(page
, total_mapcount
);
1500 if (PageSwapCache(page
))
1501 swapcount
= page_swapcount(page
);
1502 if (total_swapcount
)
1503 *total_swapcount
= swapcount
;
1504 return mapcount
+ swapcount
;
1507 page
= compound_head(page
);
1509 _total_mapcount
= _total_swapcount
= map_swapcount
= 0;
1510 if (PageSwapCache(page
)) {
1513 entry
.val
= page_private(page
);
1514 si
= _swap_info_get(entry
);
1517 offset
= swp_offset(entry
);
1521 ci
= lock_cluster(si
, offset
);
1522 for (i
= 0; i
< HPAGE_PMD_NR
; i
++) {
1523 mapcount
= atomic_read(&page
[i
]._mapcount
) + 1;
1524 _total_mapcount
+= mapcount
;
1526 swapcount
= swap_count(map
[offset
+ i
]);
1527 _total_swapcount
+= swapcount
;
1529 map_swapcount
= max(map_swapcount
, mapcount
+ swapcount
);
1532 if (PageDoubleMap(page
)) {
1534 _total_mapcount
-= HPAGE_PMD_NR
;
1536 mapcount
= compound_mapcount(page
);
1537 map_swapcount
+= mapcount
;
1538 _total_mapcount
+= mapcount
;
1540 *total_mapcount
= _total_mapcount
;
1541 if (total_swapcount
)
1542 *total_swapcount
= _total_swapcount
;
1544 return map_swapcount
;
1548 * We can write to an anon page without COW if there are no other references
1549 * to it. And as a side-effect, free up its swap: because the old content
1550 * on disk will never be read, and seeking back there to write new content
1551 * later would only waste time away from clustering.
1553 * NOTE: total_map_swapcount should not be relied upon by the caller if
1554 * reuse_swap_page() returns false, but it may be always overwritten
1555 * (see the other implementation for CONFIG_SWAP=n).
1557 bool reuse_swap_page(struct page
*page
, int *total_map_swapcount
)
1559 int count
, total_mapcount
, total_swapcount
;
1561 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1562 if (unlikely(PageKsm(page
)))
1564 count
= page_trans_huge_map_swapcount(page
, &total_mapcount
,
1566 if (total_map_swapcount
)
1567 *total_map_swapcount
= total_mapcount
+ total_swapcount
;
1568 if (count
== 1 && PageSwapCache(page
) &&
1569 (likely(!PageTransCompound(page
)) ||
1570 /* The remaining swap count will be freed soon */
1571 total_swapcount
== page_swapcount(page
))) {
1572 if (!PageWriteback(page
)) {
1573 page
= compound_head(page
);
1574 delete_from_swap_cache(page
);
1578 struct swap_info_struct
*p
;
1580 entry
.val
= page_private(page
);
1581 p
= swap_info_get(entry
);
1582 if (p
->flags
& SWP_STABLE_WRITES
) {
1583 spin_unlock(&p
->lock
);
1586 spin_unlock(&p
->lock
);
1594 * If swap is getting full, or if there are no more mappings of this page,
1595 * then try_to_free_swap is called to free its swap space.
1597 int try_to_free_swap(struct page
*page
)
1599 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1601 if (!PageSwapCache(page
))
1603 if (PageWriteback(page
))
1605 if (page_swapped(page
))
1609 * Once hibernation has begun to create its image of memory,
1610 * there's a danger that one of the calls to try_to_free_swap()
1611 * - most probably a call from __try_to_reclaim_swap() while
1612 * hibernation is allocating its own swap pages for the image,
1613 * but conceivably even a call from memory reclaim - will free
1614 * the swap from a page which has already been recorded in the
1615 * image as a clean swapcache page, and then reuse its swap for
1616 * another page of the image. On waking from hibernation, the
1617 * original page might be freed under memory pressure, then
1618 * later read back in from swap, now with the wrong data.
1620 * Hibernation suspends storage while it is writing the image
1621 * to disk so check that here.
1623 if (pm_suspended_storage())
1626 page
= compound_head(page
);
1627 delete_from_swap_cache(page
);
1633 * Free the swap entry like above, but also try to
1634 * free the page cache entry if it is the last user.
1636 int free_swap_and_cache(swp_entry_t entry
)
1638 struct swap_info_struct
*p
;
1639 unsigned char count
;
1641 if (non_swap_entry(entry
))
1644 p
= _swap_info_get(entry
);
1646 count
= __swap_entry_free(p
, entry
, 1);
1647 if (count
== SWAP_HAS_CACHE
&&
1648 !swap_page_trans_huge_swapped(p
, entry
))
1649 __try_to_reclaim_swap(p
, swp_offset(entry
),
1650 TTRS_UNMAPPED
| TTRS_FULL
);
1655 #ifdef CONFIG_HIBERNATION
1657 * Find the swap type that corresponds to given device (if any).
1659 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1660 * from 0, in which the swap header is expected to be located.
1662 * This is needed for the suspend to disk (aka swsusp).
1664 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
1666 struct block_device
*bdev
= NULL
;
1670 bdev
= bdget(device
);
1672 spin_lock(&swap_lock
);
1673 for (type
= 0; type
< nr_swapfiles
; type
++) {
1674 struct swap_info_struct
*sis
= swap_info
[type
];
1676 if (!(sis
->flags
& SWP_WRITEOK
))
1681 *bdev_p
= bdgrab(sis
->bdev
);
1683 spin_unlock(&swap_lock
);
1686 if (bdev
== sis
->bdev
) {
1687 struct swap_extent
*se
= &sis
->first_swap_extent
;
1689 if (se
->start_block
== offset
) {
1691 *bdev_p
= bdgrab(sis
->bdev
);
1693 spin_unlock(&swap_lock
);
1699 spin_unlock(&swap_lock
);
1707 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1708 * corresponding to given index in swap_info (swap type).
1710 sector_t
swapdev_block(int type
, pgoff_t offset
)
1712 struct block_device
*bdev
;
1713 struct swap_info_struct
*si
= swap_type_to_swap_info(type
);
1715 if (!si
|| !(si
->flags
& SWP_WRITEOK
))
1717 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
1721 * Return either the total number of swap pages of given type, or the number
1722 * of free pages of that type (depending on @free)
1724 * This is needed for software suspend
1726 unsigned int count_swap_pages(int type
, int free
)
1730 spin_lock(&swap_lock
);
1731 if ((unsigned int)type
< nr_swapfiles
) {
1732 struct swap_info_struct
*sis
= swap_info
[type
];
1734 spin_lock(&sis
->lock
);
1735 if (sis
->flags
& SWP_WRITEOK
) {
1738 n
-= sis
->inuse_pages
;
1740 spin_unlock(&sis
->lock
);
1742 spin_unlock(&swap_lock
);
1745 #endif /* CONFIG_HIBERNATION */
1747 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1749 return pte_same(pte_swp_clear_soft_dirty(pte
), swp_pte
);
1753 * No need to decide whether this PTE shares the swap entry with others,
1754 * just let do_wp_page work it out if a write is requested later - to
1755 * force COW, vm_page_prot omits write permission from any private vma.
1757 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1758 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1760 struct page
*swapcache
;
1761 struct mem_cgroup
*memcg
;
1767 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1768 if (unlikely(!page
))
1771 if (mem_cgroup_try_charge(page
, vma
->vm_mm
, GFP_KERNEL
,
1777 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1778 if (unlikely(!pte_same_as_swp(*pte
, swp_entry_to_pte(entry
)))) {
1779 mem_cgroup_cancel_charge(page
, memcg
, false);
1784 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1785 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1787 set_pte_at(vma
->vm_mm
, addr
, pte
,
1788 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1789 if (page
== swapcache
) {
1790 page_add_anon_rmap(page
, vma
, addr
, false);
1791 mem_cgroup_commit_charge(page
, memcg
, true, false);
1792 } else { /* ksm created a completely new copy */
1793 page_add_new_anon_rmap(page
, vma
, addr
, false);
1794 mem_cgroup_commit_charge(page
, memcg
, false, false);
1795 lru_cache_add_active_or_unevictable(page
, vma
);
1799 * Move the page to the active list so it is not
1800 * immediately swapped out again after swapon.
1802 activate_page(page
);
1804 pte_unmap_unlock(pte
, ptl
);
1806 if (page
!= swapcache
) {
1813 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1814 unsigned long addr
, unsigned long end
,
1815 unsigned int type
, bool frontswap
,
1816 unsigned long *fs_pages_to_unuse
)
1821 struct swap_info_struct
*si
;
1822 unsigned long offset
;
1824 volatile unsigned char *swap_map
;
1826 si
= swap_info
[type
];
1827 pte
= pte_offset_map(pmd
, addr
);
1829 struct vm_fault vmf
;
1831 if (!is_swap_pte(*pte
))
1834 entry
= pte_to_swp_entry(*pte
);
1835 if (swp_type(entry
) != type
)
1838 offset
= swp_offset(entry
);
1839 if (frontswap
&& !frontswap_test(si
, offset
))
1843 swap_map
= &si
->swap_map
[offset
];
1847 page
= swapin_readahead(entry
, GFP_HIGHUSER_MOVABLE
, &vmf
);
1849 if (*swap_map
== 0 || *swap_map
== SWAP_MAP_BAD
)
1855 wait_on_page_writeback(page
);
1856 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1863 try_to_free_swap(page
);
1867 if (*fs_pages_to_unuse
&& !--(*fs_pages_to_unuse
)) {
1868 ret
= FRONTSWAP_PAGES_UNUSED
;
1872 pte
= pte_offset_map(pmd
, addr
);
1873 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1881 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
1882 unsigned long addr
, unsigned long end
,
1883 unsigned int type
, bool frontswap
,
1884 unsigned long *fs_pages_to_unuse
)
1890 pmd
= pmd_offset(pud
, addr
);
1893 next
= pmd_addr_end(addr
, end
);
1894 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1896 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, type
,
1897 frontswap
, fs_pages_to_unuse
);
1900 } while (pmd
++, addr
= next
, addr
!= end
);
1904 static inline int unuse_pud_range(struct vm_area_struct
*vma
, p4d_t
*p4d
,
1905 unsigned long addr
, unsigned long end
,
1906 unsigned int type
, bool frontswap
,
1907 unsigned long *fs_pages_to_unuse
)
1913 pud
= pud_offset(p4d
, addr
);
1915 next
= pud_addr_end(addr
, end
);
1916 if (pud_none_or_clear_bad(pud
))
1918 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, type
,
1919 frontswap
, fs_pages_to_unuse
);
1922 } while (pud
++, addr
= next
, addr
!= end
);
1926 static inline int unuse_p4d_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
1927 unsigned long addr
, unsigned long end
,
1928 unsigned int type
, bool frontswap
,
1929 unsigned long *fs_pages_to_unuse
)
1935 p4d
= p4d_offset(pgd
, addr
);
1937 next
= p4d_addr_end(addr
, end
);
1938 if (p4d_none_or_clear_bad(p4d
))
1940 ret
= unuse_pud_range(vma
, p4d
, addr
, next
, type
,
1941 frontswap
, fs_pages_to_unuse
);
1944 } while (p4d
++, addr
= next
, addr
!= end
);
1948 static int unuse_vma(struct vm_area_struct
*vma
, unsigned int type
,
1949 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1952 unsigned long addr
, end
, next
;
1955 addr
= vma
->vm_start
;
1958 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1960 next
= pgd_addr_end(addr
, end
);
1961 if (pgd_none_or_clear_bad(pgd
))
1963 ret
= unuse_p4d_range(vma
, pgd
, addr
, next
, type
,
1964 frontswap
, fs_pages_to_unuse
);
1967 } while (pgd
++, addr
= next
, addr
!= end
);
1971 static int unuse_mm(struct mm_struct
*mm
, unsigned int type
,
1972 bool frontswap
, unsigned long *fs_pages_to_unuse
)
1974 struct vm_area_struct
*vma
;
1977 down_read(&mm
->mmap_sem
);
1978 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1979 if (vma
->anon_vma
) {
1980 ret
= unuse_vma(vma
, type
, frontswap
,
1987 up_read(&mm
->mmap_sem
);
1992 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1993 * from current position to next entry still in use. Return 0
1994 * if there are no inuse entries after prev till end of the map.
1996 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
1997 unsigned int prev
, bool frontswap
)
2000 unsigned char count
;
2003 * No need for swap_lock here: we're just looking
2004 * for whether an entry is in use, not modifying it; false
2005 * hits are okay, and sys_swapoff() has already prevented new
2006 * allocations from this area (while holding swap_lock).
2008 for (i
= prev
+ 1; i
< si
->max
; i
++) {
2009 count
= READ_ONCE(si
->swap_map
[i
]);
2010 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
2011 if (!frontswap
|| frontswap_test(si
, i
))
2013 if ((i
% LATENCY_LIMIT
) == 0)
2024 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2025 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2027 int try_to_unuse(unsigned int type
, bool frontswap
,
2028 unsigned long pages_to_unuse
)
2030 struct mm_struct
*prev_mm
;
2031 struct mm_struct
*mm
;
2032 struct list_head
*p
;
2034 struct swap_info_struct
*si
= swap_info
[type
];
2039 if (!si
->inuse_pages
)
2046 retval
= shmem_unuse(type
, frontswap
, &pages_to_unuse
);
2053 spin_lock(&mmlist_lock
);
2054 p
= &init_mm
.mmlist
;
2055 while (si
->inuse_pages
&&
2056 !signal_pending(current
) &&
2057 (p
= p
->next
) != &init_mm
.mmlist
) {
2059 mm
= list_entry(p
, struct mm_struct
, mmlist
);
2060 if (!mmget_not_zero(mm
))
2062 spin_unlock(&mmlist_lock
);
2065 retval
= unuse_mm(mm
, type
, frontswap
, &pages_to_unuse
);
2073 * Make sure that we aren't completely killing
2074 * interactive performance.
2077 spin_lock(&mmlist_lock
);
2079 spin_unlock(&mmlist_lock
);
2084 while (si
->inuse_pages
&&
2085 !signal_pending(current
) &&
2086 (i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
2088 entry
= swp_entry(type
, i
);
2089 page
= find_get_page(swap_address_space(entry
), i
);
2094 * It is conceivable that a racing task removed this page from
2095 * swap cache just before we acquired the page lock. The page
2096 * might even be back in swap cache on another swap area. But
2097 * that is okay, try_to_free_swap() only removes stale pages.
2100 wait_on_page_writeback(page
);
2101 try_to_free_swap(page
);
2106 * For frontswap, we just need to unuse pages_to_unuse, if
2107 * it was specified. Need not check frontswap again here as
2108 * we already zeroed out pages_to_unuse if not frontswap.
2110 if (pages_to_unuse
&& --pages_to_unuse
== 0)
2115 * Lets check again to see if there are still swap entries in the map.
2116 * If yes, we would need to do retry the unuse logic again.
2117 * Under global memory pressure, swap entries can be reinserted back
2118 * into process space after the mmlist loop above passes over them.
2120 * Limit the number of retries? No: when mmget_not_zero() above fails,
2121 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2122 * at its own independent pace; and even shmem_writepage() could have
2123 * been preempted after get_swap_page(), temporarily hiding that swap.
2124 * It's easy and robust (though cpu-intensive) just to keep retrying.
2126 if (si
->inuse_pages
) {
2127 if (!signal_pending(current
))
2132 return (retval
== FRONTSWAP_PAGES_UNUSED
) ? 0 : retval
;
2136 * After a successful try_to_unuse, if no swap is now in use, we know
2137 * we can empty the mmlist. swap_lock must be held on entry and exit.
2138 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2139 * added to the mmlist just after page_duplicate - before would be racy.
2141 static void drain_mmlist(void)
2143 struct list_head
*p
, *next
;
2146 for (type
= 0; type
< nr_swapfiles
; type
++)
2147 if (swap_info
[type
]->inuse_pages
)
2149 spin_lock(&mmlist_lock
);
2150 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
2152 spin_unlock(&mmlist_lock
);
2156 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2157 * corresponds to page offset for the specified swap entry.
2158 * Note that the type of this function is sector_t, but it returns page offset
2159 * into the bdev, not sector offset.
2161 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
2163 struct swap_info_struct
*sis
;
2164 struct swap_extent
*start_se
;
2165 struct swap_extent
*se
;
2168 sis
= swp_swap_info(entry
);
2171 offset
= swp_offset(entry
);
2172 start_se
= sis
->curr_swap_extent
;
2176 if (se
->start_page
<= offset
&&
2177 offset
< (se
->start_page
+ se
->nr_pages
)) {
2178 return se
->start_block
+ (offset
- se
->start_page
);
2180 se
= list_next_entry(se
, list
);
2181 sis
->curr_swap_extent
= se
;
2182 BUG_ON(se
== start_se
); /* It *must* be present */
2187 * Returns the page offset into bdev for the specified page's swap entry.
2189 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
2192 entry
.val
= page_private(page
);
2193 return map_swap_entry(entry
, bdev
);
2197 * Free all of a swapdev's extent information
2199 static void destroy_swap_extents(struct swap_info_struct
*sis
)
2201 while (!list_empty(&sis
->first_swap_extent
.list
)) {
2202 struct swap_extent
*se
;
2204 se
= list_first_entry(&sis
->first_swap_extent
.list
,
2205 struct swap_extent
, list
);
2206 list_del(&se
->list
);
2210 if (sis
->flags
& SWP_ACTIVATED
) {
2211 struct file
*swap_file
= sis
->swap_file
;
2212 struct address_space
*mapping
= swap_file
->f_mapping
;
2214 sis
->flags
&= ~SWP_ACTIVATED
;
2215 if (mapping
->a_ops
->swap_deactivate
)
2216 mapping
->a_ops
->swap_deactivate(swap_file
);
2221 * Add a block range (and the corresponding page range) into this swapdev's
2222 * extent list. The extent list is kept sorted in page order.
2224 * This function rather assumes that it is called in ascending page order.
2227 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
2228 unsigned long nr_pages
, sector_t start_block
)
2230 struct swap_extent
*se
;
2231 struct swap_extent
*new_se
;
2232 struct list_head
*lh
;
2234 if (start_page
== 0) {
2235 se
= &sis
->first_swap_extent
;
2236 sis
->curr_swap_extent
= se
;
2238 se
->nr_pages
= nr_pages
;
2239 se
->start_block
= start_block
;
2242 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
2243 se
= list_entry(lh
, struct swap_extent
, list
);
2244 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
2245 if (se
->start_block
+ se
->nr_pages
== start_block
) {
2247 se
->nr_pages
+= nr_pages
;
2253 * No merge. Insert a new extent, preserving ordering.
2255 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
2258 new_se
->start_page
= start_page
;
2259 new_se
->nr_pages
= nr_pages
;
2260 new_se
->start_block
= start_block
;
2262 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
2265 EXPORT_SYMBOL_GPL(add_swap_extent
);
2268 * A `swap extent' is a simple thing which maps a contiguous range of pages
2269 * onto a contiguous range of disk blocks. An ordered list of swap extents
2270 * is built at swapon time and is then used at swap_writepage/swap_readpage
2271 * time for locating where on disk a page belongs.
2273 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2274 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2275 * swap files identically.
2277 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2278 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2279 * swapfiles are handled *identically* after swapon time.
2281 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2282 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2283 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2284 * requirements, they are simply tossed out - we will never use those blocks
2287 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
2288 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
2289 * which will scribble on the fs.
2291 * The amount of disk space which a single swap extent represents varies.
2292 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2293 * extents in the list. To avoid much list walking, we cache the previous
2294 * search location in `curr_swap_extent', and start new searches from there.
2295 * This is extremely effective. The average number of iterations in
2296 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2298 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
2300 struct file
*swap_file
= sis
->swap_file
;
2301 struct address_space
*mapping
= swap_file
->f_mapping
;
2302 struct inode
*inode
= mapping
->host
;
2305 if (S_ISBLK(inode
->i_mode
)) {
2306 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2311 if (mapping
->a_ops
->swap_activate
) {
2312 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
2314 sis
->flags
|= SWP_ACTIVATED
;
2316 sis
->flags
|= SWP_FS
;
2317 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
2323 return generic_swapfile_activate(sis
, swap_file
, span
);
2326 static int swap_node(struct swap_info_struct
*p
)
2328 struct block_device
*bdev
;
2333 bdev
= p
->swap_file
->f_inode
->i_sb
->s_bdev
;
2335 return bdev
? bdev
->bd_disk
->node_id
: NUMA_NO_NODE
;
2338 static void _enable_swap_info(struct swap_info_struct
*p
, int prio
,
2339 unsigned char *swap_map
,
2340 struct swap_cluster_info
*cluster_info
)
2347 p
->prio
= --least_priority
;
2349 * the plist prio is negated because plist ordering is
2350 * low-to-high, while swap ordering is high-to-low
2352 p
->list
.prio
= -p
->prio
;
2355 p
->avail_lists
[i
].prio
= -p
->prio
;
2357 if (swap_node(p
) == i
)
2358 p
->avail_lists
[i
].prio
= 1;
2360 p
->avail_lists
[i
].prio
= -p
->prio
;
2363 p
->swap_map
= swap_map
;
2364 p
->cluster_info
= cluster_info
;
2365 p
->flags
|= SWP_WRITEOK
;
2366 atomic_long_add(p
->pages
, &nr_swap_pages
);
2367 total_swap_pages
+= p
->pages
;
2369 assert_spin_locked(&swap_lock
);
2371 * both lists are plists, and thus priority ordered.
2372 * swap_active_head needs to be priority ordered for swapoff(),
2373 * which on removal of any swap_info_struct with an auto-assigned
2374 * (i.e. negative) priority increments the auto-assigned priority
2375 * of any lower-priority swap_info_structs.
2376 * swap_avail_head needs to be priority ordered for get_swap_page(),
2377 * which allocates swap pages from the highest available priority
2380 plist_add(&p
->list
, &swap_active_head
);
2381 add_to_avail_list(p
);
2384 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
2385 unsigned char *swap_map
,
2386 struct swap_cluster_info
*cluster_info
,
2387 unsigned long *frontswap_map
)
2389 frontswap_init(p
->type
, frontswap_map
);
2390 spin_lock(&swap_lock
);
2391 spin_lock(&p
->lock
);
2392 _enable_swap_info(p
, prio
, swap_map
, cluster_info
);
2393 spin_unlock(&p
->lock
);
2394 spin_unlock(&swap_lock
);
2397 static void reinsert_swap_info(struct swap_info_struct
*p
)
2399 spin_lock(&swap_lock
);
2400 spin_lock(&p
->lock
);
2401 _enable_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
2402 spin_unlock(&p
->lock
);
2403 spin_unlock(&swap_lock
);
2406 bool has_usable_swap(void)
2410 spin_lock(&swap_lock
);
2411 if (plist_head_empty(&swap_active_head
))
2413 spin_unlock(&swap_lock
);
2417 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
2419 struct swap_info_struct
*p
= NULL
;
2420 unsigned char *swap_map
;
2421 struct swap_cluster_info
*cluster_info
;
2422 unsigned long *frontswap_map
;
2423 struct file
*swap_file
, *victim
;
2424 struct address_space
*mapping
;
2425 struct inode
*inode
;
2426 struct filename
*pathname
;
2428 unsigned int old_block_size
;
2430 if (!capable(CAP_SYS_ADMIN
))
2433 BUG_ON(!current
->mm
);
2435 pathname
= getname(specialfile
);
2436 if (IS_ERR(pathname
))
2437 return PTR_ERR(pathname
);
2439 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
2440 err
= PTR_ERR(victim
);
2444 mapping
= victim
->f_mapping
;
2445 spin_lock(&swap_lock
);
2446 plist_for_each_entry(p
, &swap_active_head
, list
) {
2447 if (p
->flags
& SWP_WRITEOK
) {
2448 if (p
->swap_file
->f_mapping
== mapping
) {
2456 spin_unlock(&swap_lock
);
2459 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2460 vm_unacct_memory(p
->pages
);
2463 spin_unlock(&swap_lock
);
2466 del_from_avail_list(p
);
2467 spin_lock(&p
->lock
);
2469 struct swap_info_struct
*si
= p
;
2472 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2475 for_each_node(nid
) {
2476 if (si
->avail_lists
[nid
].prio
!= 1)
2477 si
->avail_lists
[nid
].prio
--;
2482 plist_del(&p
->list
, &swap_active_head
);
2483 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2484 total_swap_pages
-= p
->pages
;
2485 p
->flags
&= ~SWP_WRITEOK
;
2486 spin_unlock(&p
->lock
);
2487 spin_unlock(&swap_lock
);
2489 disable_swap_slots_cache_lock();
2491 set_current_oom_origin();
2492 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
2493 clear_current_oom_origin();
2496 /* re-insert swap space back into swap_list */
2497 reinsert_swap_info(p
);
2498 reenable_swap_slots_cache_unlock();
2502 reenable_swap_slots_cache_unlock();
2504 flush_work(&p
->discard_work
);
2506 destroy_swap_extents(p
);
2507 if (p
->flags
& SWP_CONTINUED
)
2508 free_swap_count_continuations(p
);
2510 if (!p
->bdev
|| !blk_queue_nonrot(bdev_get_queue(p
->bdev
)))
2511 atomic_dec(&nr_rotate_swap
);
2513 mutex_lock(&swapon_mutex
);
2514 spin_lock(&swap_lock
);
2515 spin_lock(&p
->lock
);
2518 /* wait for anyone still in scan_swap_map */
2519 p
->highest_bit
= 0; /* cuts scans short */
2520 while (p
->flags
>= SWP_SCANNING
) {
2521 spin_unlock(&p
->lock
);
2522 spin_unlock(&swap_lock
);
2523 schedule_timeout_uninterruptible(1);
2524 spin_lock(&swap_lock
);
2525 spin_lock(&p
->lock
);
2528 swap_file
= p
->swap_file
;
2529 old_block_size
= p
->old_block_size
;
2530 p
->swap_file
= NULL
;
2532 swap_map
= p
->swap_map
;
2534 cluster_info
= p
->cluster_info
;
2535 p
->cluster_info
= NULL
;
2536 frontswap_map
= frontswap_map_get(p
);
2537 spin_unlock(&p
->lock
);
2538 spin_unlock(&swap_lock
);
2539 frontswap_invalidate_area(p
->type
);
2540 frontswap_map_set(p
, NULL
);
2541 mutex_unlock(&swapon_mutex
);
2542 free_percpu(p
->percpu_cluster
);
2543 p
->percpu_cluster
= NULL
;
2545 kvfree(cluster_info
);
2546 kvfree(frontswap_map
);
2547 /* Destroy swap account information */
2548 swap_cgroup_swapoff(p
->type
);
2549 exit_swap_address_space(p
->type
);
2551 inode
= mapping
->host
;
2552 if (S_ISBLK(inode
->i_mode
)) {
2553 struct block_device
*bdev
= I_BDEV(inode
);
2554 set_blocksize(bdev
, old_block_size
);
2555 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2558 inode
->i_flags
&= ~S_SWAPFILE
;
2559 inode_unlock(inode
);
2561 filp_close(swap_file
, NULL
);
2564 * Clear the SWP_USED flag after all resources are freed so that swapon
2565 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2566 * not hold p->lock after we cleared its SWP_WRITEOK.
2568 spin_lock(&swap_lock
);
2570 spin_unlock(&swap_lock
);
2573 atomic_inc(&proc_poll_event
);
2574 wake_up_interruptible(&proc_poll_wait
);
2577 filp_close(victim
, NULL
);
2583 #ifdef CONFIG_PROC_FS
2584 static __poll_t
swaps_poll(struct file
*file
, poll_table
*wait
)
2586 struct seq_file
*seq
= file
->private_data
;
2588 poll_wait(file
, &proc_poll_wait
, wait
);
2590 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2591 seq
->poll_event
= atomic_read(&proc_poll_event
);
2592 return EPOLLIN
| EPOLLRDNORM
| EPOLLERR
| EPOLLPRI
;
2595 return EPOLLIN
| EPOLLRDNORM
;
2599 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2601 struct swap_info_struct
*si
;
2605 mutex_lock(&swapon_mutex
);
2608 return SEQ_START_TOKEN
;
2610 for (type
= 0; (si
= swap_type_to_swap_info(type
)); type
++) {
2611 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2620 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2622 struct swap_info_struct
*si
= v
;
2625 if (v
== SEQ_START_TOKEN
)
2628 type
= si
->type
+ 1;
2630 for (; (si
= swap_type_to_swap_info(type
)); type
++) {
2631 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2640 static void swap_stop(struct seq_file
*swap
, void *v
)
2642 mutex_unlock(&swapon_mutex
);
2645 static int swap_show(struct seq_file
*swap
, void *v
)
2647 struct swap_info_struct
*si
= v
;
2651 if (si
== SEQ_START_TOKEN
) {
2652 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2656 file
= si
->swap_file
;
2657 len
= seq_file_path(swap
, file
, " \t\n\\");
2658 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
2659 len
< 40 ? 40 - len
: 1, " ",
2660 S_ISBLK(file_inode(file
)->i_mode
) ?
2661 "partition" : "file\t",
2662 si
->pages
<< (PAGE_SHIFT
- 10),
2663 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
2668 static const struct seq_operations swaps_op
= {
2669 .start
= swap_start
,
2675 static int swaps_open(struct inode
*inode
, struct file
*file
)
2677 struct seq_file
*seq
;
2680 ret
= seq_open(file
, &swaps_op
);
2684 seq
= file
->private_data
;
2685 seq
->poll_event
= atomic_read(&proc_poll_event
);
2689 static const struct file_operations proc_swaps_operations
= {
2692 .llseek
= seq_lseek
,
2693 .release
= seq_release
,
2697 static int __init
procswaps_init(void)
2699 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
2702 __initcall(procswaps_init
);
2703 #endif /* CONFIG_PROC_FS */
2705 #ifdef MAX_SWAPFILES_CHECK
2706 static int __init
max_swapfiles_check(void)
2708 MAX_SWAPFILES_CHECK();
2711 late_initcall(max_swapfiles_check
);
2714 static struct swap_info_struct
*alloc_swap_info(void)
2716 struct swap_info_struct
*p
;
2720 p
= kvzalloc(struct_size(p
, avail_lists
, nr_node_ids
), GFP_KERNEL
);
2722 return ERR_PTR(-ENOMEM
);
2724 spin_lock(&swap_lock
);
2725 for (type
= 0; type
< nr_swapfiles
; type
++) {
2726 if (!(swap_info
[type
]->flags
& SWP_USED
))
2729 if (type
>= MAX_SWAPFILES
) {
2730 spin_unlock(&swap_lock
);
2732 return ERR_PTR(-EPERM
);
2734 if (type
>= nr_swapfiles
) {
2736 WRITE_ONCE(swap_info
[type
], p
);
2738 * Write swap_info[type] before nr_swapfiles, in case a
2739 * racing procfs swap_start() or swap_next() is reading them.
2740 * (We never shrink nr_swapfiles, we never free this entry.)
2743 WRITE_ONCE(nr_swapfiles
, nr_swapfiles
+ 1);
2746 p
= swap_info
[type
];
2748 * Do not memset this entry: a racing procfs swap_next()
2749 * would be relying on p->type to remain valid.
2752 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
2753 plist_node_init(&p
->list
, 0);
2755 plist_node_init(&p
->avail_lists
[i
], 0);
2756 p
->flags
= SWP_USED
;
2757 spin_unlock(&swap_lock
);
2758 spin_lock_init(&p
->lock
);
2759 spin_lock_init(&p
->cont_lock
);
2764 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2768 if (S_ISBLK(inode
->i_mode
)) {
2769 p
->bdev
= bdgrab(I_BDEV(inode
));
2770 error
= blkdev_get(p
->bdev
,
2771 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, p
);
2776 p
->old_block_size
= block_size(p
->bdev
);
2777 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2780 p
->flags
|= SWP_BLKDEV
;
2781 } else if (S_ISREG(inode
->i_mode
)) {
2782 p
->bdev
= inode
->i_sb
->s_bdev
;
2784 if (IS_SWAPFILE(inode
))
2794 * Find out how many pages are allowed for a single swap device. There
2795 * are two limiting factors:
2796 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2797 * 2) the number of bits in the swap pte, as defined by the different
2800 * In order to find the largest possible bit mask, a swap entry with
2801 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2802 * decoded to a swp_entry_t again, and finally the swap offset is
2805 * This will mask all the bits from the initial ~0UL mask that can't
2806 * be encoded in either the swp_entry_t or the architecture definition
2809 unsigned long generic_max_swapfile_size(void)
2811 return swp_offset(pte_to_swp_entry(
2812 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2815 /* Can be overridden by an architecture for additional checks. */
2816 __weak
unsigned long max_swapfile_size(void)
2818 return generic_max_swapfile_size();
2821 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2822 union swap_header
*swap_header
,
2823 struct inode
*inode
)
2826 unsigned long maxpages
;
2827 unsigned long swapfilepages
;
2828 unsigned long last_page
;
2830 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2831 pr_err("Unable to find swap-space signature\n");
2835 /* swap partition endianess hack... */
2836 if (swab32(swap_header
->info
.version
) == 1) {
2837 swab32s(&swap_header
->info
.version
);
2838 swab32s(&swap_header
->info
.last_page
);
2839 swab32s(&swap_header
->info
.nr_badpages
);
2840 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2842 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2843 swab32s(&swap_header
->info
.badpages
[i
]);
2845 /* Check the swap header's sub-version */
2846 if (swap_header
->info
.version
!= 1) {
2847 pr_warn("Unable to handle swap header version %d\n",
2848 swap_header
->info
.version
);
2853 p
->cluster_next
= 1;
2856 maxpages
= max_swapfile_size();
2857 last_page
= swap_header
->info
.last_page
;
2859 pr_warn("Empty swap-file\n");
2862 if (last_page
> maxpages
) {
2863 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2864 maxpages
<< (PAGE_SHIFT
- 10),
2865 last_page
<< (PAGE_SHIFT
- 10));
2867 if (maxpages
> last_page
) {
2868 maxpages
= last_page
+ 1;
2869 /* p->max is an unsigned int: don't overflow it */
2870 if ((unsigned int)maxpages
== 0)
2871 maxpages
= UINT_MAX
;
2873 p
->highest_bit
= maxpages
- 1;
2877 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
2878 if (swapfilepages
&& maxpages
> swapfilepages
) {
2879 pr_warn("Swap area shorter than signature indicates\n");
2882 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
2884 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2890 #define SWAP_CLUSTER_INFO_COLS \
2891 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2892 #define SWAP_CLUSTER_SPACE_COLS \
2893 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2894 #define SWAP_CLUSTER_COLS \
2895 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2897 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
2898 union swap_header
*swap_header
,
2899 unsigned char *swap_map
,
2900 struct swap_cluster_info
*cluster_info
,
2901 unsigned long maxpages
,
2905 unsigned int nr_good_pages
;
2907 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2908 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
2909 unsigned long i
, idx
;
2911 nr_good_pages
= maxpages
- 1; /* omit header page */
2913 cluster_list_init(&p
->free_clusters
);
2914 cluster_list_init(&p
->discard_clusters
);
2916 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
2917 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
2918 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
2920 if (page_nr
< maxpages
) {
2921 swap_map
[page_nr
] = SWAP_MAP_BAD
;
2924 * Haven't marked the cluster free yet, no list
2925 * operation involved
2927 inc_cluster_info_page(p
, cluster_info
, page_nr
);
2931 /* Haven't marked the cluster free yet, no list operation involved */
2932 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
2933 inc_cluster_info_page(p
, cluster_info
, i
);
2935 if (nr_good_pages
) {
2936 swap_map
[0] = SWAP_MAP_BAD
;
2938 * Not mark the cluster free yet, no list
2939 * operation involved
2941 inc_cluster_info_page(p
, cluster_info
, 0);
2943 p
->pages
= nr_good_pages
;
2944 nr_extents
= setup_swap_extents(p
, span
);
2947 nr_good_pages
= p
->pages
;
2949 if (!nr_good_pages
) {
2950 pr_warn("Empty swap-file\n");
2959 * Reduce false cache line sharing between cluster_info and
2960 * sharing same address space.
2962 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
2963 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
2964 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
2965 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
2966 if (idx
>= nr_clusters
)
2968 if (cluster_count(&cluster_info
[idx
]))
2970 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
2971 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
2979 * Helper to sys_swapon determining if a given swap
2980 * backing device queue supports DISCARD operations.
2982 static bool swap_discardable(struct swap_info_struct
*si
)
2984 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
2986 if (!q
|| !blk_queue_discard(q
))
2992 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
2994 struct swap_info_struct
*p
;
2995 struct filename
*name
;
2996 struct file
*swap_file
= NULL
;
2997 struct address_space
*mapping
;
3000 union swap_header
*swap_header
;
3003 unsigned long maxpages
;
3004 unsigned char *swap_map
= NULL
;
3005 struct swap_cluster_info
*cluster_info
= NULL
;
3006 unsigned long *frontswap_map
= NULL
;
3007 struct page
*page
= NULL
;
3008 struct inode
*inode
= NULL
;
3009 bool inced_nr_rotate_swap
= false;
3011 if (swap_flags
& ~SWAP_FLAGS_VALID
)
3014 if (!capable(CAP_SYS_ADMIN
))
3017 if (!swap_avail_heads
)
3020 p
= alloc_swap_info();
3024 INIT_WORK(&p
->discard_work
, swap_discard_work
);
3026 name
= getname(specialfile
);
3028 error
= PTR_ERR(name
);
3032 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
3033 if (IS_ERR(swap_file
)) {
3034 error
= PTR_ERR(swap_file
);
3039 p
->swap_file
= swap_file
;
3040 mapping
= swap_file
->f_mapping
;
3041 inode
= mapping
->host
;
3043 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
3044 error
= claim_swapfile(p
, inode
);
3045 if (unlikely(error
))
3049 * Read the swap header.
3051 if (!mapping
->a_ops
->readpage
) {
3055 page
= read_mapping_page(mapping
, 0, swap_file
);
3057 error
= PTR_ERR(page
);
3060 swap_header
= kmap(page
);
3062 maxpages
= read_swap_header(p
, swap_header
, inode
);
3063 if (unlikely(!maxpages
)) {
3068 /* OK, set up the swap map and apply the bad block list */
3069 swap_map
= vzalloc(maxpages
);
3075 if (bdi_cap_stable_pages_required(inode_to_bdi(inode
)))
3076 p
->flags
|= SWP_STABLE_WRITES
;
3078 if (bdi_cap_synchronous_io(inode_to_bdi(inode
)))
3079 p
->flags
|= SWP_SYNCHRONOUS_IO
;
3081 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
3083 unsigned long ci
, nr_cluster
;
3085 p
->flags
|= SWP_SOLIDSTATE
;
3087 * select a random position to start with to help wear leveling
3090 p
->cluster_next
= 1 + (prandom_u32() % p
->highest_bit
);
3091 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
3093 cluster_info
= kvcalloc(nr_cluster
, sizeof(*cluster_info
),
3095 if (!cluster_info
) {
3100 for (ci
= 0; ci
< nr_cluster
; ci
++)
3101 spin_lock_init(&((cluster_info
+ ci
)->lock
));
3103 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
3104 if (!p
->percpu_cluster
) {
3108 for_each_possible_cpu(cpu
) {
3109 struct percpu_cluster
*cluster
;
3110 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
3111 cluster_set_null(&cluster
->index
);
3114 atomic_inc(&nr_rotate_swap
);
3115 inced_nr_rotate_swap
= true;
3118 error
= swap_cgroup_swapon(p
->type
, maxpages
);
3122 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
3123 cluster_info
, maxpages
, &span
);
3124 if (unlikely(nr_extents
< 0)) {
3128 /* frontswap enabled? set up bit-per-page map for frontswap */
3129 if (IS_ENABLED(CONFIG_FRONTSWAP
))
3130 frontswap_map
= kvcalloc(BITS_TO_LONGS(maxpages
),
3134 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
3136 * When discard is enabled for swap with no particular
3137 * policy flagged, we set all swap discard flags here in
3138 * order to sustain backward compatibility with older
3139 * swapon(8) releases.
3141 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
3145 * By flagging sys_swapon, a sysadmin can tell us to
3146 * either do single-time area discards only, or to just
3147 * perform discards for released swap page-clusters.
3148 * Now it's time to adjust the p->flags accordingly.
3150 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
3151 p
->flags
&= ~SWP_PAGE_DISCARD
;
3152 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
3153 p
->flags
&= ~SWP_AREA_DISCARD
;
3155 /* issue a swapon-time discard if it's still required */
3156 if (p
->flags
& SWP_AREA_DISCARD
) {
3157 int err
= discard_swap(p
);
3159 pr_err("swapon: discard_swap(%p): %d\n",
3164 error
= init_swap_address_space(p
->type
, maxpages
);
3168 mutex_lock(&swapon_mutex
);
3170 if (swap_flags
& SWAP_FLAG_PREFER
)
3172 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
3173 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
3175 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3176 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
3177 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
3178 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
3179 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
3180 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
3181 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
3182 (frontswap_map
) ? "FS" : "");
3184 mutex_unlock(&swapon_mutex
);
3185 atomic_inc(&proc_poll_event
);
3186 wake_up_interruptible(&proc_poll_wait
);
3188 if (S_ISREG(inode
->i_mode
))
3189 inode
->i_flags
|= S_SWAPFILE
;
3193 free_percpu(p
->percpu_cluster
);
3194 p
->percpu_cluster
= NULL
;
3195 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
3196 set_blocksize(p
->bdev
, p
->old_block_size
);
3197 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
3199 destroy_swap_extents(p
);
3200 swap_cgroup_swapoff(p
->type
);
3201 spin_lock(&swap_lock
);
3202 p
->swap_file
= NULL
;
3204 spin_unlock(&swap_lock
);
3206 kvfree(cluster_info
);
3207 kvfree(frontswap_map
);
3208 if (inced_nr_rotate_swap
)
3209 atomic_dec(&nr_rotate_swap
);
3211 if (inode
&& S_ISREG(inode
->i_mode
)) {
3212 inode_unlock(inode
);
3215 filp_close(swap_file
, NULL
);
3218 if (page
&& !IS_ERR(page
)) {
3224 if (inode
&& S_ISREG(inode
->i_mode
))
3225 inode_unlock(inode
);
3227 enable_swap_slots_cache();
3231 void si_swapinfo(struct sysinfo
*val
)
3234 unsigned long nr_to_be_unused
= 0;
3236 spin_lock(&swap_lock
);
3237 for (type
= 0; type
< nr_swapfiles
; type
++) {
3238 struct swap_info_struct
*si
= swap_info
[type
];
3240 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
3241 nr_to_be_unused
+= si
->inuse_pages
;
3243 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
3244 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
3245 spin_unlock(&swap_lock
);
3249 * Verify that a swap entry is valid and increment its swap map count.
3251 * Returns error code in following case.
3253 * - swp_entry is invalid -> EINVAL
3254 * - swp_entry is migration entry -> EINVAL
3255 * - swap-cache reference is requested but there is already one. -> EEXIST
3256 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3257 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3259 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
3261 struct swap_info_struct
*p
;
3262 struct swap_cluster_info
*ci
;
3263 unsigned long offset
;
3264 unsigned char count
;
3265 unsigned char has_cache
;
3268 if (non_swap_entry(entry
))
3271 p
= swp_swap_info(entry
);
3275 offset
= swp_offset(entry
);
3276 if (unlikely(offset
>= p
->max
))
3279 ci
= lock_cluster_or_swap_info(p
, offset
);
3281 count
= p
->swap_map
[offset
];
3284 * swapin_readahead() doesn't check if a swap entry is valid, so the
3285 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3287 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
3292 has_cache
= count
& SWAP_HAS_CACHE
;
3293 count
&= ~SWAP_HAS_CACHE
;
3296 if (usage
== SWAP_HAS_CACHE
) {
3298 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3299 if (!has_cache
&& count
)
3300 has_cache
= SWAP_HAS_CACHE
;
3301 else if (has_cache
) /* someone else added cache */
3303 else /* no users remaining */
3306 } else if (count
|| has_cache
) {
3308 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
3310 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
3312 else if (swap_count_continued(p
, offset
, count
))
3313 count
= COUNT_CONTINUED
;
3317 err
= -ENOENT
; /* unused swap entry */
3319 p
->swap_map
[offset
] = count
| has_cache
;
3322 unlock_cluster_or_swap_info(p
, ci
);
3327 pr_err("swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
3332 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3333 * (in which case its reference count is never incremented).
3335 void swap_shmem_alloc(swp_entry_t entry
)
3337 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
3341 * Increase reference count of swap entry by 1.
3342 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3343 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3344 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3345 * might occur if a page table entry has got corrupted.
3347 int swap_duplicate(swp_entry_t entry
)
3351 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
3352 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
3357 * @entry: swap entry for which we allocate swap cache.
3359 * Called when allocating swap cache for existing swap entry,
3360 * This can return error codes. Returns 0 at success.
3361 * -EBUSY means there is a swap cache.
3362 * Note: return code is different from swap_duplicate().
3364 int swapcache_prepare(swp_entry_t entry
)
3366 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
3369 struct swap_info_struct
*swp_swap_info(swp_entry_t entry
)
3371 return swap_type_to_swap_info(swp_type(entry
));
3374 struct swap_info_struct
*page_swap_info(struct page
*page
)
3376 swp_entry_t entry
= { .val
= page_private(page
) };
3377 return swp_swap_info(entry
);
3381 * out-of-line __page_file_ methods to avoid include hell.
3383 struct address_space
*__page_file_mapping(struct page
*page
)
3385 return page_swap_info(page
)->swap_file
->f_mapping
;
3387 EXPORT_SYMBOL_GPL(__page_file_mapping
);
3389 pgoff_t
__page_file_index(struct page
*page
)
3391 swp_entry_t swap
= { .val
= page_private(page
) };
3392 return swp_offset(swap
);
3394 EXPORT_SYMBOL_GPL(__page_file_index
);
3397 * add_swap_count_continuation - called when a swap count is duplicated
3398 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3399 * page of the original vmalloc'ed swap_map, to hold the continuation count
3400 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3401 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3403 * These continuation pages are seldom referenced: the common paths all work
3404 * on the original swap_map, only referring to a continuation page when the
3405 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3407 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3408 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3409 * can be called after dropping locks.
3411 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
3413 struct swap_info_struct
*si
;
3414 struct swap_cluster_info
*ci
;
3417 struct page
*list_page
;
3419 unsigned char count
;
3422 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3423 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3425 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
3427 si
= swap_info_get(entry
);
3430 * An acceptable race has occurred since the failing
3431 * __swap_duplicate(): the swap entry has been freed,
3432 * perhaps even the whole swap_map cleared for swapoff.
3437 offset
= swp_offset(entry
);
3439 ci
= lock_cluster(si
, offset
);
3441 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
3443 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
3445 * The higher the swap count, the more likely it is that tasks
3446 * will race to add swap count continuation: we need to avoid
3447 * over-provisioning.
3454 spin_unlock(&si
->lock
);
3459 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3460 * no architecture is using highmem pages for kernel page tables: so it
3461 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3463 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3464 offset
&= ~PAGE_MASK
;
3466 spin_lock(&si
->cont_lock
);
3468 * Page allocation does not initialize the page's lru field,
3469 * but it does always reset its private field.
3471 if (!page_private(head
)) {
3472 BUG_ON(count
& COUNT_CONTINUED
);
3473 INIT_LIST_HEAD(&head
->lru
);
3474 set_page_private(head
, SWP_CONTINUED
);
3475 si
->flags
|= SWP_CONTINUED
;
3478 list_for_each_entry(list_page
, &head
->lru
, lru
) {
3482 * If the previous map said no continuation, but we've found
3483 * a continuation page, free our allocation and use this one.
3485 if (!(count
& COUNT_CONTINUED
))
3486 goto out_unlock_cont
;
3488 map
= kmap_atomic(list_page
) + offset
;
3493 * If this continuation count now has some space in it,
3494 * free our allocation and use this one.
3496 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
3497 goto out_unlock_cont
;
3500 list_add_tail(&page
->lru
, &head
->lru
);
3501 page
= NULL
; /* now it's attached, don't free it */
3503 spin_unlock(&si
->cont_lock
);
3506 spin_unlock(&si
->lock
);
3514 * swap_count_continued - when the original swap_map count is incremented
3515 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3516 * into, carry if so, or else fail until a new continuation page is allocated;
3517 * when the original swap_map count is decremented from 0 with continuation,
3518 * borrow from the continuation and report whether it still holds more.
3519 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3522 static bool swap_count_continued(struct swap_info_struct
*si
,
3523 pgoff_t offset
, unsigned char count
)
3530 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3531 if (page_private(head
) != SWP_CONTINUED
) {
3532 BUG_ON(count
& COUNT_CONTINUED
);
3533 return false; /* need to add count continuation */
3536 spin_lock(&si
->cont_lock
);
3537 offset
&= ~PAGE_MASK
;
3538 page
= list_entry(head
->lru
.next
, struct page
, lru
);
3539 map
= kmap_atomic(page
) + offset
;
3541 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3542 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3544 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3546 * Think of how you add 1 to 999
3548 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3550 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3551 BUG_ON(page
== head
);
3552 map
= kmap_atomic(page
) + offset
;
3554 if (*map
== SWAP_CONT_MAX
) {
3556 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3558 ret
= false; /* add count continuation */
3561 map
= kmap_atomic(page
) + offset
;
3562 init_map
: *map
= 0; /* we didn't zero the page */
3566 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3567 while (page
!= head
) {
3568 map
= kmap_atomic(page
) + offset
;
3569 *map
= COUNT_CONTINUED
;
3571 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3573 ret
= true; /* incremented */
3575 } else { /* decrementing */
3577 * Think of how you subtract 1 from 1000
3579 BUG_ON(count
!= COUNT_CONTINUED
);
3580 while (*map
== COUNT_CONTINUED
) {
3582 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3583 BUG_ON(page
== head
);
3584 map
= kmap_atomic(page
) + offset
;
3591 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3592 while (page
!= head
) {
3593 map
= kmap_atomic(page
) + offset
;
3594 *map
= SWAP_CONT_MAX
| count
;
3595 count
= COUNT_CONTINUED
;
3597 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3599 ret
= count
== COUNT_CONTINUED
;
3602 spin_unlock(&si
->cont_lock
);
3607 * free_swap_count_continuations - swapoff free all the continuation pages
3608 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3610 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3614 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3616 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3617 if (page_private(head
)) {
3618 struct page
*page
, *next
;
3620 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3621 list_del(&page
->lru
);
3628 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3629 void mem_cgroup_throttle_swaprate(struct mem_cgroup
*memcg
, int node
,
3632 struct swap_info_struct
*si
, *next
;
3633 if (!(gfp_mask
& __GFP_IO
) || !memcg
)
3636 if (!blk_cgroup_congested())
3640 * We've already scheduled a throttle, avoid taking the global swap
3643 if (current
->throttle_queue
)
3646 spin_lock(&swap_avail_lock
);
3647 plist_for_each_entry_safe(si
, next
, &swap_avail_heads
[node
],
3648 avail_lists
[node
]) {
3650 blkcg_schedule_throttle(bdev_get_queue(si
->bdev
),
3655 spin_unlock(&swap_avail_lock
);
3659 static int __init
swapfile_init(void)
3663 swap_avail_heads
= kmalloc_array(nr_node_ids
, sizeof(struct plist_head
),
3665 if (!swap_avail_heads
) {
3666 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3671 plist_head_init(&swap_avail_heads
[nid
]);
3675 subsys_initcall(swapfile_init
);