4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
9 #include <linux/hugetlb.h>
10 #include <linux/mman.h>
11 #include <linux/slab.h>
12 #include <linux/kernel_stat.h>
13 #include <linux/swap.h>
14 #include <linux/vmalloc.h>
15 #include <linux/pagemap.h>
16 #include <linux/namei.h>
17 #include <linux/shmem_fs.h>
18 #include <linux/blkdev.h>
19 #include <linux/random.h>
20 #include <linux/writeback.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/init.h>
24 #include <linux/ksm.h>
25 #include <linux/rmap.h>
26 #include <linux/security.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mutex.h>
29 #include <linux/capability.h>
30 #include <linux/syscalls.h>
31 #include <linux/memcontrol.h>
32 #include <linux/poll.h>
33 #include <linux/oom.h>
34 #include <linux/frontswap.h>
35 #include <linux/swapfile.h>
36 #include <linux/export.h>
38 #include <asm/pgtable.h>
39 #include <asm/tlbflush.h>
40 #include <linux/swapops.h>
41 #include <linux/swap_cgroup.h>
43 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
45 static void free_swap_count_continuations(struct swap_info_struct
*);
46 static sector_t
map_swap_entry(swp_entry_t
, struct block_device
**);
48 DEFINE_SPINLOCK(swap_lock
);
49 static unsigned int nr_swapfiles
;
50 atomic_long_t nr_swap_pages
;
52 * Some modules use swappable objects and may try to swap them out under
53 * memory pressure (via the shrinker). Before doing so, they may wish to
54 * check to see if any swap space is available.
56 EXPORT_SYMBOL_GPL(nr_swap_pages
);
57 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
58 long total_swap_pages
;
59 static int least_priority
;
61 static const char Bad_file
[] = "Bad swap file entry ";
62 static const char Unused_file
[] = "Unused swap file entry ";
63 static const char Bad_offset
[] = "Bad swap offset entry ";
64 static const char Unused_offset
[] = "Unused swap offset entry ";
67 * all active swap_info_structs
68 * protected with swap_lock, and ordered by priority.
70 PLIST_HEAD(swap_active_head
);
73 * all available (active, not full) swap_info_structs
74 * protected with swap_avail_lock, ordered by priority.
75 * This is used by get_swap_page() instead of swap_active_head
76 * because swap_active_head includes all swap_info_structs,
77 * but get_swap_page() doesn't need to look at full ones.
78 * This uses its own lock instead of swap_lock because when a
79 * swap_info_struct changes between not-full/full, it needs to
80 * add/remove itself to/from this list, but the swap_info_struct->lock
81 * is held and the locking order requires swap_lock to be taken
82 * before any swap_info_struct->lock.
84 static PLIST_HEAD(swap_avail_head
);
85 static DEFINE_SPINLOCK(swap_avail_lock
);
87 struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
89 static DEFINE_MUTEX(swapon_mutex
);
91 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait
);
92 /* Activity counter to indicate that a swapon or swapoff has occurred */
93 static atomic_t proc_poll_event
= ATOMIC_INIT(0);
95 static inline unsigned char swap_count(unsigned char ent
)
97 return ent
& ~SWAP_HAS_CACHE
; /* may include SWAP_HAS_CONT flag */
100 /* returns 1 if swap entry is freed */
102 __try_to_reclaim_swap(struct swap_info_struct
*si
, unsigned long offset
)
104 swp_entry_t entry
= swp_entry(si
->type
, offset
);
108 page
= find_get_page(swap_address_space(entry
), swp_offset(entry
));
112 * This function is called from scan_swap_map() and it's called
113 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
114 * We have to use trylock for avoiding deadlock. This is a special
115 * case and you should use try_to_free_swap() with explicit lock_page()
116 * in usual operations.
118 if (trylock_page(page
)) {
119 ret
= try_to_free_swap(page
);
127 * swapon tell device that all the old swap contents can be discarded,
128 * to allow the swap device to optimize its wear-levelling.
130 static int discard_swap(struct swap_info_struct
*si
)
132 struct swap_extent
*se
;
133 sector_t start_block
;
137 /* Do not discard the swap header page! */
138 se
= &si
->first_swap_extent
;
139 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
140 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
142 err
= blkdev_issue_discard(si
->bdev
, start_block
,
143 nr_blocks
, GFP_KERNEL
, 0);
149 list_for_each_entry(se
, &si
->first_swap_extent
.list
, list
) {
150 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
151 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
153 err
= blkdev_issue_discard(si
->bdev
, start_block
,
154 nr_blocks
, GFP_KERNEL
, 0);
160 return err
; /* That will often be -EOPNOTSUPP */
164 * swap allocation tell device that a cluster of swap can now be discarded,
165 * to allow the swap device to optimize its wear-levelling.
167 static void discard_swap_cluster(struct swap_info_struct
*si
,
168 pgoff_t start_page
, pgoff_t nr_pages
)
170 struct swap_extent
*se
= si
->curr_swap_extent
;
171 int found_extent
= 0;
174 if (se
->start_page
<= start_page
&&
175 start_page
< se
->start_page
+ se
->nr_pages
) {
176 pgoff_t offset
= start_page
- se
->start_page
;
177 sector_t start_block
= se
->start_block
+ offset
;
178 sector_t nr_blocks
= se
->nr_pages
- offset
;
180 if (nr_blocks
> nr_pages
)
181 nr_blocks
= nr_pages
;
182 start_page
+= nr_blocks
;
183 nr_pages
-= nr_blocks
;
186 si
->curr_swap_extent
= se
;
188 start_block
<<= PAGE_SHIFT
- 9;
189 nr_blocks
<<= PAGE_SHIFT
- 9;
190 if (blkdev_issue_discard(si
->bdev
, start_block
,
191 nr_blocks
, GFP_NOIO
, 0))
195 se
= list_next_entry(se
, list
);
199 #define SWAPFILE_CLUSTER 256
200 #define LATENCY_LIMIT 256
202 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
208 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
213 static inline void cluster_set_count(struct swap_cluster_info
*info
,
219 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
220 unsigned int c
, unsigned int f
)
226 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
231 static inline void cluster_set_next(struct swap_cluster_info
*info
,
237 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
238 unsigned int n
, unsigned int f
)
244 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
246 return info
->flags
& CLUSTER_FLAG_FREE
;
249 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
251 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
254 static inline void cluster_set_null(struct swap_cluster_info
*info
)
256 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
260 static inline struct swap_cluster_info
*lock_cluster(struct swap_info_struct
*si
,
261 unsigned long offset
)
263 struct swap_cluster_info
*ci
;
265 ci
= si
->cluster_info
;
267 ci
+= offset
/ SWAPFILE_CLUSTER
;
268 spin_lock(&ci
->lock
);
273 static inline void unlock_cluster(struct swap_cluster_info
*ci
)
276 spin_unlock(&ci
->lock
);
279 static inline struct swap_cluster_info
*lock_cluster_or_swap_info(
280 struct swap_info_struct
*si
,
281 unsigned long offset
)
283 struct swap_cluster_info
*ci
;
285 ci
= lock_cluster(si
, offset
);
287 spin_lock(&si
->lock
);
292 static inline void unlock_cluster_or_swap_info(struct swap_info_struct
*si
,
293 struct swap_cluster_info
*ci
)
298 spin_unlock(&si
->lock
);
301 static inline bool cluster_list_empty(struct swap_cluster_list
*list
)
303 return cluster_is_null(&list
->head
);
306 static inline unsigned int cluster_list_first(struct swap_cluster_list
*list
)
308 return cluster_next(&list
->head
);
311 static void cluster_list_init(struct swap_cluster_list
*list
)
313 cluster_set_null(&list
->head
);
314 cluster_set_null(&list
->tail
);
317 static void cluster_list_add_tail(struct swap_cluster_list
*list
,
318 struct swap_cluster_info
*ci
,
321 if (cluster_list_empty(list
)) {
322 cluster_set_next_flag(&list
->head
, idx
, 0);
323 cluster_set_next_flag(&list
->tail
, idx
, 0);
325 struct swap_cluster_info
*ci_tail
;
326 unsigned int tail
= cluster_next(&list
->tail
);
329 * Nested cluster lock, but both cluster locks are
330 * only acquired when we held swap_info_struct->lock
333 spin_lock_nested(&ci_tail
->lock
, SINGLE_DEPTH_NESTING
);
334 cluster_set_next(ci_tail
, idx
);
335 unlock_cluster(ci_tail
);
336 cluster_set_next_flag(&list
->tail
, idx
, 0);
340 static unsigned int cluster_list_del_first(struct swap_cluster_list
*list
,
341 struct swap_cluster_info
*ci
)
345 idx
= cluster_next(&list
->head
);
346 if (cluster_next(&list
->tail
) == idx
) {
347 cluster_set_null(&list
->head
);
348 cluster_set_null(&list
->tail
);
350 cluster_set_next_flag(&list
->head
,
351 cluster_next(&ci
[idx
]), 0);
356 /* Add a cluster to discard list and schedule it to do discard */
357 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
361 * If scan_swap_map() can't find a free cluster, it will check
362 * si->swap_map directly. To make sure the discarding cluster isn't
363 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
364 * will be cleared after discard
366 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
367 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
369 cluster_list_add_tail(&si
->discard_clusters
, si
->cluster_info
, idx
);
371 schedule_work(&si
->discard_work
);
375 * Doing discard actually. After a cluster discard is finished, the cluster
376 * will be added to free cluster list. caller should hold si->lock.
378 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
380 struct swap_cluster_info
*info
, *ci
;
383 info
= si
->cluster_info
;
385 while (!cluster_list_empty(&si
->discard_clusters
)) {
386 idx
= cluster_list_del_first(&si
->discard_clusters
, info
);
387 spin_unlock(&si
->lock
);
389 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
392 spin_lock(&si
->lock
);
393 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
394 cluster_set_flag(ci
, CLUSTER_FLAG_FREE
);
396 cluster_list_add_tail(&si
->free_clusters
, info
, idx
);
397 ci
= lock_cluster(si
, idx
* SWAPFILE_CLUSTER
);
398 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
399 0, SWAPFILE_CLUSTER
);
404 static void swap_discard_work(struct work_struct
*work
)
406 struct swap_info_struct
*si
;
408 si
= container_of(work
, struct swap_info_struct
, discard_work
);
410 spin_lock(&si
->lock
);
411 swap_do_scheduled_discard(si
);
412 spin_unlock(&si
->lock
);
416 * The cluster corresponding to page_nr will be used. The cluster will be
417 * removed from free cluster list and its usage counter will be increased.
419 static void inc_cluster_info_page(struct swap_info_struct
*p
,
420 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
422 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
426 if (cluster_is_free(&cluster_info
[idx
])) {
427 VM_BUG_ON(cluster_list_first(&p
->free_clusters
) != idx
);
428 cluster_list_del_first(&p
->free_clusters
, cluster_info
);
429 cluster_set_count_flag(&cluster_info
[idx
], 0, 0);
432 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
433 cluster_set_count(&cluster_info
[idx
],
434 cluster_count(&cluster_info
[idx
]) + 1);
438 * The cluster corresponding to page_nr decreases one usage. If the usage
439 * counter becomes 0, which means no page in the cluster is in using, we can
440 * optionally discard the cluster and add it to free cluster list.
442 static void dec_cluster_info_page(struct swap_info_struct
*p
,
443 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
445 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
450 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
451 cluster_set_count(&cluster_info
[idx
],
452 cluster_count(&cluster_info
[idx
]) - 1);
454 if (cluster_count(&cluster_info
[idx
]) == 0) {
456 * If the swap is discardable, prepare discard the cluster
457 * instead of free it immediately. The cluster will be freed
460 if ((p
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
461 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
462 swap_cluster_schedule_discard(p
, idx
);
466 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
467 cluster_list_add_tail(&p
->free_clusters
, cluster_info
, idx
);
472 * It's possible scan_swap_map() uses a free cluster in the middle of free
473 * cluster list. Avoiding such abuse to avoid list corruption.
476 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
477 unsigned long offset
)
479 struct percpu_cluster
*percpu_cluster
;
482 offset
/= SWAPFILE_CLUSTER
;
483 conflict
= !cluster_list_empty(&si
->free_clusters
) &&
484 offset
!= cluster_list_first(&si
->free_clusters
) &&
485 cluster_is_free(&si
->cluster_info
[offset
]);
490 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
491 cluster_set_null(&percpu_cluster
->index
);
496 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
497 * might involve allocating a new cluster for current CPU too.
499 static void scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
500 unsigned long *offset
, unsigned long *scan_base
)
502 struct percpu_cluster
*cluster
;
503 struct swap_cluster_info
*ci
;
505 unsigned long tmp
, max
;
508 cluster
= this_cpu_ptr(si
->percpu_cluster
);
509 if (cluster_is_null(&cluster
->index
)) {
510 if (!cluster_list_empty(&si
->free_clusters
)) {
511 cluster
->index
= si
->free_clusters
.head
;
512 cluster
->next
= cluster_next(&cluster
->index
) *
514 } else if (!cluster_list_empty(&si
->discard_clusters
)) {
516 * we don't have free cluster but have some clusters in
517 * discarding, do discard now and reclaim them
519 swap_do_scheduled_discard(si
);
520 *scan_base
= *offset
= si
->cluster_next
;
529 * Other CPUs can use our cluster if they can't find a free cluster,
530 * check if there is still free entry in the cluster
533 max
= min_t(unsigned long, si
->max
,
534 (cluster_next(&cluster
->index
) + 1) * SWAPFILE_CLUSTER
);
536 cluster_set_null(&cluster
->index
);
539 ci
= lock_cluster(si
, tmp
);
541 if (!si
->swap_map
[tmp
]) {
549 cluster_set_null(&cluster
->index
);
552 cluster
->next
= tmp
+ 1;
557 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
560 struct swap_cluster_info
*ci
;
561 unsigned long offset
;
562 unsigned long scan_base
;
563 unsigned long last_in_cluster
= 0;
564 int latency_ration
= LATENCY_LIMIT
;
567 * We try to cluster swap pages by allocating them sequentially
568 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
569 * way, however, we resort to first-free allocation, starting
570 * a new cluster. This prevents us from scattering swap pages
571 * all over the entire swap partition, so that we reduce
572 * overall disk seek times between swap pages. -- sct
573 * But we do now try to find an empty cluster. -Andrea
574 * And we let swap pages go all over an SSD partition. Hugh
577 si
->flags
+= SWP_SCANNING
;
578 scan_base
= offset
= si
->cluster_next
;
581 if (si
->cluster_info
) {
582 scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
);
586 if (unlikely(!si
->cluster_nr
--)) {
587 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
588 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
592 spin_unlock(&si
->lock
);
595 * If seek is expensive, start searching for new cluster from
596 * start of partition, to minimize the span of allocated swap.
597 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
598 * case, just handled by scan_swap_map_try_ssd_cluster() above.
600 scan_base
= offset
= si
->lowest_bit
;
601 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
603 /* Locate the first empty (unaligned) cluster */
604 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
605 if (si
->swap_map
[offset
])
606 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
607 else if (offset
== last_in_cluster
) {
608 spin_lock(&si
->lock
);
609 offset
-= SWAPFILE_CLUSTER
- 1;
610 si
->cluster_next
= offset
;
611 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
614 if (unlikely(--latency_ration
< 0)) {
616 latency_ration
= LATENCY_LIMIT
;
621 spin_lock(&si
->lock
);
622 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
626 if (si
->cluster_info
) {
627 while (scan_swap_map_ssd_cluster_conflict(si
, offset
))
628 scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
);
630 if (!(si
->flags
& SWP_WRITEOK
))
632 if (!si
->highest_bit
)
634 if (offset
> si
->highest_bit
)
635 scan_base
= offset
= si
->lowest_bit
;
637 ci
= lock_cluster(si
, offset
);
638 /* reuse swap entry of cache-only swap if not busy. */
639 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
642 spin_unlock(&si
->lock
);
643 swap_was_freed
= __try_to_reclaim_swap(si
, offset
);
644 spin_lock(&si
->lock
);
645 /* entry was freed successfully, try to use this again */
648 goto scan
; /* check next one */
651 if (si
->swap_map
[offset
]) {
656 if (offset
== si
->lowest_bit
)
658 if (offset
== si
->highest_bit
)
661 if (si
->inuse_pages
== si
->pages
) {
662 si
->lowest_bit
= si
->max
;
664 spin_lock(&swap_avail_lock
);
665 plist_del(&si
->avail_list
, &swap_avail_head
);
666 spin_unlock(&swap_avail_lock
);
668 si
->swap_map
[offset
] = usage
;
669 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
671 si
->cluster_next
= offset
+ 1;
672 si
->flags
-= SWP_SCANNING
;
677 spin_unlock(&si
->lock
);
678 while (++offset
<= si
->highest_bit
) {
679 if (!si
->swap_map
[offset
]) {
680 spin_lock(&si
->lock
);
683 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
684 spin_lock(&si
->lock
);
687 if (unlikely(--latency_ration
< 0)) {
689 latency_ration
= LATENCY_LIMIT
;
692 offset
= si
->lowest_bit
;
693 while (offset
< scan_base
) {
694 if (!si
->swap_map
[offset
]) {
695 spin_lock(&si
->lock
);
698 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
699 spin_lock(&si
->lock
);
702 if (unlikely(--latency_ration
< 0)) {
704 latency_ration
= LATENCY_LIMIT
;
708 spin_lock(&si
->lock
);
711 si
->flags
-= SWP_SCANNING
;
715 swp_entry_t
get_swap_page(void)
717 struct swap_info_struct
*si
, *next
;
720 if (atomic_long_read(&nr_swap_pages
) <= 0)
722 atomic_long_dec(&nr_swap_pages
);
724 spin_lock(&swap_avail_lock
);
727 plist_for_each_entry_safe(si
, next
, &swap_avail_head
, avail_list
) {
728 /* requeue si to after same-priority siblings */
729 plist_requeue(&si
->avail_list
, &swap_avail_head
);
730 spin_unlock(&swap_avail_lock
);
731 spin_lock(&si
->lock
);
732 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
733 spin_lock(&swap_avail_lock
);
734 if (plist_node_empty(&si
->avail_list
)) {
735 spin_unlock(&si
->lock
);
738 WARN(!si
->highest_bit
,
739 "swap_info %d in list but !highest_bit\n",
741 WARN(!(si
->flags
& SWP_WRITEOK
),
742 "swap_info %d in list but !SWP_WRITEOK\n",
744 plist_del(&si
->avail_list
, &swap_avail_head
);
745 spin_unlock(&si
->lock
);
749 /* This is called for allocating swap entry for cache */
750 offset
= scan_swap_map(si
, SWAP_HAS_CACHE
);
751 spin_unlock(&si
->lock
);
753 return swp_entry(si
->type
, offset
);
754 pr_debug("scan_swap_map of si %d failed to find offset\n",
756 spin_lock(&swap_avail_lock
);
759 * if we got here, it's likely that si was almost full before,
760 * and since scan_swap_map() can drop the si->lock, multiple
761 * callers probably all tried to get a page from the same si
762 * and it filled up before we could get one; or, the si filled
763 * up between us dropping swap_avail_lock and taking si->lock.
764 * Since we dropped the swap_avail_lock, the swap_avail_head
765 * list may have been modified; so if next is still in the
766 * swap_avail_head list then try it, otherwise start over.
768 if (plist_node_empty(&next
->avail_list
))
772 spin_unlock(&swap_avail_lock
);
774 atomic_long_inc(&nr_swap_pages
);
776 return (swp_entry_t
) {0};
779 /* The only caller of this function is now suspend routine */
780 swp_entry_t
get_swap_page_of_type(int type
)
782 struct swap_info_struct
*si
;
785 si
= swap_info
[type
];
786 spin_lock(&si
->lock
);
787 if (si
&& (si
->flags
& SWP_WRITEOK
)) {
788 atomic_long_dec(&nr_swap_pages
);
789 /* This is called for allocating swap entry, not cache */
790 offset
= scan_swap_map(si
, 1);
792 spin_unlock(&si
->lock
);
793 return swp_entry(type
, offset
);
795 atomic_long_inc(&nr_swap_pages
);
797 spin_unlock(&si
->lock
);
798 return (swp_entry_t
) {0};
801 static struct swap_info_struct
*_swap_info_get(swp_entry_t entry
)
803 struct swap_info_struct
*p
;
804 unsigned long offset
, type
;
808 type
= swp_type(entry
);
809 if (type
>= nr_swapfiles
)
812 if (!(p
->flags
& SWP_USED
))
814 offset
= swp_offset(entry
);
815 if (offset
>= p
->max
)
817 if (!p
->swap_map
[offset
])
822 pr_err("swap_info_get: %s%08lx\n", Unused_offset
, entry
.val
);
825 pr_err("swap_info_get: %s%08lx\n", Bad_offset
, entry
.val
);
828 pr_err("swap_info_get: %s%08lx\n", Unused_file
, entry
.val
);
831 pr_err("swap_info_get: %s%08lx\n", Bad_file
, entry
.val
);
836 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
838 struct swap_info_struct
*p
;
840 p
= _swap_info_get(entry
);
846 static unsigned char swap_entry_free(struct swap_info_struct
*p
,
847 swp_entry_t entry
, unsigned char usage
,
848 bool swap_info_locked
)
850 struct swap_cluster_info
*ci
;
851 unsigned long offset
= swp_offset(entry
);
853 unsigned char has_cache
;
854 bool lock_swap_info
= false;
856 if (!swap_info_locked
) {
857 count
= p
->swap_map
[offset
];
858 if (!p
->cluster_info
|| count
== usage
|| count
== SWAP_MAP_SHMEM
) {
860 swap_info_locked
= true;
861 lock_swap_info
= true;
866 ci
= lock_cluster(p
, offset
);
868 count
= p
->swap_map
[offset
];
870 if (!swap_info_locked
&& (count
== usage
|| count
== SWAP_MAP_SHMEM
)) {
875 has_cache
= count
& SWAP_HAS_CACHE
;
876 count
&= ~SWAP_HAS_CACHE
;
878 if (usage
== SWAP_HAS_CACHE
) {
879 VM_BUG_ON(!has_cache
);
881 } else if (count
== SWAP_MAP_SHMEM
) {
883 * Or we could insist on shmem.c using a special
884 * swap_shmem_free() and free_shmem_swap_and_cache()...
887 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
888 if (count
== COUNT_CONTINUED
) {
889 if (swap_count_continued(p
, offset
, count
))
890 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
892 count
= SWAP_MAP_MAX
;
897 usage
= count
| has_cache
;
898 p
->swap_map
[offset
] = usage
;
902 /* free if no reference */
904 VM_BUG_ON(!swap_info_locked
);
905 mem_cgroup_uncharge_swap(entry
);
906 ci
= lock_cluster(p
, offset
);
907 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
909 if (offset
< p
->lowest_bit
)
910 p
->lowest_bit
= offset
;
911 if (offset
> p
->highest_bit
) {
912 bool was_full
= !p
->highest_bit
;
913 p
->highest_bit
= offset
;
914 if (was_full
&& (p
->flags
& SWP_WRITEOK
)) {
915 spin_lock(&swap_avail_lock
);
916 WARN_ON(!plist_node_empty(&p
->avail_list
));
917 if (plist_node_empty(&p
->avail_list
))
918 plist_add(&p
->avail_list
,
920 spin_unlock(&swap_avail_lock
);
923 atomic_long_inc(&nr_swap_pages
);
925 frontswap_invalidate_page(p
->type
, offset
);
926 if (p
->flags
& SWP_BLKDEV
) {
927 struct gendisk
*disk
= p
->bdev
->bd_disk
;
928 if (disk
->fops
->swap_slot_free_notify
)
929 disk
->fops
->swap_slot_free_notify(p
->bdev
,
935 spin_unlock(&p
->lock
);
941 * Caller has made sure that the swap device corresponding to entry
942 * is still around or has not been recycled.
944 void swap_free(swp_entry_t entry
)
946 struct swap_info_struct
*p
;
948 p
= _swap_info_get(entry
);
950 swap_entry_free(p
, entry
, 1, false);
954 * Called after dropping swapcache to decrease refcnt to swap entries.
956 void swapcache_free(swp_entry_t entry
)
958 struct swap_info_struct
*p
;
960 p
= _swap_info_get(entry
);
962 swap_entry_free(p
, entry
, SWAP_HAS_CACHE
, false);
966 * How many references to page are currently swapped out?
967 * This does not give an exact answer when swap count is continued,
968 * but does include the high COUNT_CONTINUED flag to allow for that.
970 int page_swapcount(struct page
*page
)
973 struct swap_info_struct
*p
;
974 struct swap_cluster_info
*ci
;
976 unsigned long offset
;
978 entry
.val
= page_private(page
);
979 p
= _swap_info_get(entry
);
981 offset
= swp_offset(entry
);
982 ci
= lock_cluster_or_swap_info(p
, offset
);
983 count
= swap_count(p
->swap_map
[offset
]);
984 unlock_cluster_or_swap_info(p
, ci
);
990 * How many references to @entry are currently swapped out?
991 * This considers COUNT_CONTINUED so it returns exact answer.
993 int swp_swapcount(swp_entry_t entry
)
995 int count
, tmp_count
, n
;
996 struct swap_info_struct
*p
;
997 struct swap_cluster_info
*ci
;
1002 p
= _swap_info_get(entry
);
1006 offset
= swp_offset(entry
);
1008 ci
= lock_cluster_or_swap_info(p
, offset
);
1010 count
= swap_count(p
->swap_map
[offset
]);
1011 if (!(count
& COUNT_CONTINUED
))
1014 count
&= ~COUNT_CONTINUED
;
1015 n
= SWAP_MAP_MAX
+ 1;
1017 page
= vmalloc_to_page(p
->swap_map
+ offset
);
1018 offset
&= ~PAGE_MASK
;
1019 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
1022 page
= list_next_entry(page
, lru
);
1023 map
= kmap_atomic(page
);
1024 tmp_count
= map
[offset
];
1027 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
1028 n
*= (SWAP_CONT_MAX
+ 1);
1029 } while (tmp_count
& COUNT_CONTINUED
);
1031 unlock_cluster_or_swap_info(p
, ci
);
1036 * We can write to an anon page without COW if there are no other references
1037 * to it. And as a side-effect, free up its swap: because the old content
1038 * on disk will never be read, and seeking back there to write new content
1039 * later would only waste time away from clustering.
1041 * NOTE: total_mapcount should not be relied upon by the caller if
1042 * reuse_swap_page() returns false, but it may be always overwritten
1043 * (see the other implementation for CONFIG_SWAP=n).
1045 bool reuse_swap_page(struct page
*page
, int *total_mapcount
)
1049 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1050 if (unlikely(PageKsm(page
)))
1052 count
= page_trans_huge_mapcount(page
, total_mapcount
);
1053 if (count
<= 1 && PageSwapCache(page
)) {
1054 count
+= page_swapcount(page
);
1057 if (!PageWriteback(page
)) {
1058 delete_from_swap_cache(page
);
1062 struct swap_info_struct
*p
;
1064 entry
.val
= page_private(page
);
1065 p
= swap_info_get(entry
);
1066 if (p
->flags
& SWP_STABLE_WRITES
) {
1067 spin_unlock(&p
->lock
);
1070 spin_unlock(&p
->lock
);
1078 * If swap is getting full, or if there are no more mappings of this page,
1079 * then try_to_free_swap is called to free its swap space.
1081 int try_to_free_swap(struct page
*page
)
1083 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1085 if (!PageSwapCache(page
))
1087 if (PageWriteback(page
))
1089 if (page_swapcount(page
))
1093 * Once hibernation has begun to create its image of memory,
1094 * there's a danger that one of the calls to try_to_free_swap()
1095 * - most probably a call from __try_to_reclaim_swap() while
1096 * hibernation is allocating its own swap pages for the image,
1097 * but conceivably even a call from memory reclaim - will free
1098 * the swap from a page which has already been recorded in the
1099 * image as a clean swapcache page, and then reuse its swap for
1100 * another page of the image. On waking from hibernation, the
1101 * original page might be freed under memory pressure, then
1102 * later read back in from swap, now with the wrong data.
1104 * Hibernation suspends storage while it is writing the image
1105 * to disk so check that here.
1107 if (pm_suspended_storage())
1110 delete_from_swap_cache(page
);
1116 * Free the swap entry like above, but also try to
1117 * free the page cache entry if it is the last user.
1119 int free_swap_and_cache(swp_entry_t entry
)
1121 struct swap_info_struct
*p
;
1122 struct page
*page
= NULL
;
1124 if (non_swap_entry(entry
))
1127 p
= swap_info_get(entry
);
1129 if (swap_entry_free(p
, entry
, 1, true) == SWAP_HAS_CACHE
) {
1130 page
= find_get_page(swap_address_space(entry
),
1132 if (page
&& !trylock_page(page
)) {
1137 spin_unlock(&p
->lock
);
1141 * Not mapped elsewhere, or swap space full? Free it!
1142 * Also recheck PageSwapCache now page is locked (above).
1144 if (PageSwapCache(page
) && !PageWriteback(page
) &&
1145 (!page_mapped(page
) || mem_cgroup_swap_full(page
))) {
1146 delete_from_swap_cache(page
);
1155 #ifdef CONFIG_HIBERNATION
1157 * Find the swap type that corresponds to given device (if any).
1159 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1160 * from 0, in which the swap header is expected to be located.
1162 * This is needed for the suspend to disk (aka swsusp).
1164 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
1166 struct block_device
*bdev
= NULL
;
1170 bdev
= bdget(device
);
1172 spin_lock(&swap_lock
);
1173 for (type
= 0; type
< nr_swapfiles
; type
++) {
1174 struct swap_info_struct
*sis
= swap_info
[type
];
1176 if (!(sis
->flags
& SWP_WRITEOK
))
1181 *bdev_p
= bdgrab(sis
->bdev
);
1183 spin_unlock(&swap_lock
);
1186 if (bdev
== sis
->bdev
) {
1187 struct swap_extent
*se
= &sis
->first_swap_extent
;
1189 if (se
->start_block
== offset
) {
1191 *bdev_p
= bdgrab(sis
->bdev
);
1193 spin_unlock(&swap_lock
);
1199 spin_unlock(&swap_lock
);
1207 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1208 * corresponding to given index in swap_info (swap type).
1210 sector_t
swapdev_block(int type
, pgoff_t offset
)
1212 struct block_device
*bdev
;
1214 if ((unsigned int)type
>= nr_swapfiles
)
1216 if (!(swap_info
[type
]->flags
& SWP_WRITEOK
))
1218 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
1222 * Return either the total number of swap pages of given type, or the number
1223 * of free pages of that type (depending on @free)
1225 * This is needed for software suspend
1227 unsigned int count_swap_pages(int type
, int free
)
1231 spin_lock(&swap_lock
);
1232 if ((unsigned int)type
< nr_swapfiles
) {
1233 struct swap_info_struct
*sis
= swap_info
[type
];
1235 spin_lock(&sis
->lock
);
1236 if (sis
->flags
& SWP_WRITEOK
) {
1239 n
-= sis
->inuse_pages
;
1241 spin_unlock(&sis
->lock
);
1243 spin_unlock(&swap_lock
);
1246 #endif /* CONFIG_HIBERNATION */
1248 static inline int pte_same_as_swp(pte_t pte
, pte_t swp_pte
)
1250 return pte_same(pte_swp_clear_soft_dirty(pte
), swp_pte
);
1254 * No need to decide whether this PTE shares the swap entry with others,
1255 * just let do_wp_page work it out if a write is requested later - to
1256 * force COW, vm_page_prot omits write permission from any private vma.
1258 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1259 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1261 struct page
*swapcache
;
1262 struct mem_cgroup
*memcg
;
1268 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1269 if (unlikely(!page
))
1272 if (mem_cgroup_try_charge(page
, vma
->vm_mm
, GFP_KERNEL
,
1278 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1279 if (unlikely(!pte_same_as_swp(*pte
, swp_entry_to_pte(entry
)))) {
1280 mem_cgroup_cancel_charge(page
, memcg
, false);
1285 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1286 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1288 set_pte_at(vma
->vm_mm
, addr
, pte
,
1289 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1290 if (page
== swapcache
) {
1291 page_add_anon_rmap(page
, vma
, addr
, false);
1292 mem_cgroup_commit_charge(page
, memcg
, true, false);
1293 } else { /* ksm created a completely new copy */
1294 page_add_new_anon_rmap(page
, vma
, addr
, false);
1295 mem_cgroup_commit_charge(page
, memcg
, false, false);
1296 lru_cache_add_active_or_unevictable(page
, vma
);
1300 * Move the page to the active list so it is not
1301 * immediately swapped out again after swapon.
1303 activate_page(page
);
1305 pte_unmap_unlock(pte
, ptl
);
1307 if (page
!= swapcache
) {
1314 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1315 unsigned long addr
, unsigned long end
,
1316 swp_entry_t entry
, struct page
*page
)
1318 pte_t swp_pte
= swp_entry_to_pte(entry
);
1323 * We don't actually need pte lock while scanning for swp_pte: since
1324 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1325 * page table while we're scanning; though it could get zapped, and on
1326 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1327 * of unmatched parts which look like swp_pte, so unuse_pte must
1328 * recheck under pte lock. Scanning without pte lock lets it be
1329 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1331 pte
= pte_offset_map(pmd
, addr
);
1334 * swapoff spends a _lot_ of time in this loop!
1335 * Test inline before going to call unuse_pte.
1337 if (unlikely(pte_same_as_swp(*pte
, swp_pte
))) {
1339 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1342 pte
= pte_offset_map(pmd
, addr
);
1344 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1350 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
1351 unsigned long addr
, unsigned long end
,
1352 swp_entry_t entry
, struct page
*page
)
1358 pmd
= pmd_offset(pud
, addr
);
1361 next
= pmd_addr_end(addr
, end
);
1362 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1364 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
);
1367 } while (pmd
++, addr
= next
, addr
!= end
);
1371 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
1372 unsigned long addr
, unsigned long end
,
1373 swp_entry_t entry
, struct page
*page
)
1379 pud
= pud_offset(pgd
, addr
);
1381 next
= pud_addr_end(addr
, end
);
1382 if (pud_none_or_clear_bad(pud
))
1384 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
);
1387 } while (pud
++, addr
= next
, addr
!= end
);
1391 static int unuse_vma(struct vm_area_struct
*vma
,
1392 swp_entry_t entry
, struct page
*page
)
1395 unsigned long addr
, end
, next
;
1398 if (page_anon_vma(page
)) {
1399 addr
= page_address_in_vma(page
, vma
);
1400 if (addr
== -EFAULT
)
1403 end
= addr
+ PAGE_SIZE
;
1405 addr
= vma
->vm_start
;
1409 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1411 next
= pgd_addr_end(addr
, end
);
1412 if (pgd_none_or_clear_bad(pgd
))
1414 ret
= unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
);
1417 } while (pgd
++, addr
= next
, addr
!= end
);
1421 static int unuse_mm(struct mm_struct
*mm
,
1422 swp_entry_t entry
, struct page
*page
)
1424 struct vm_area_struct
*vma
;
1427 if (!down_read_trylock(&mm
->mmap_sem
)) {
1429 * Activate page so shrink_inactive_list is unlikely to unmap
1430 * its ptes while lock is dropped, so swapoff can make progress.
1432 activate_page(page
);
1434 down_read(&mm
->mmap_sem
);
1437 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1438 if (vma
->anon_vma
&& (ret
= unuse_vma(vma
, entry
, page
)))
1442 up_read(&mm
->mmap_sem
);
1443 return (ret
< 0)? ret
: 0;
1447 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1448 * from current position to next entry still in use.
1449 * Recycle to start on reaching the end, returning 0 when empty.
1451 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
1452 unsigned int prev
, bool frontswap
)
1454 unsigned int max
= si
->max
;
1455 unsigned int i
= prev
;
1456 unsigned char count
;
1459 * No need for swap_lock here: we're just looking
1460 * for whether an entry is in use, not modifying it; false
1461 * hits are okay, and sys_swapoff() has already prevented new
1462 * allocations from this area (while holding swap_lock).
1471 * No entries in use at top of swap_map,
1472 * loop back to start and recheck there.
1478 count
= READ_ONCE(si
->swap_map
[i
]);
1479 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
1480 if (!frontswap
|| frontswap_test(si
, i
))
1482 if ((i
% LATENCY_LIMIT
) == 0)
1489 * We completely avoid races by reading each swap page in advance,
1490 * and then search for the process using it. All the necessary
1491 * page table adjustments can then be made atomically.
1493 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1494 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1496 int try_to_unuse(unsigned int type
, bool frontswap
,
1497 unsigned long pages_to_unuse
)
1499 struct swap_info_struct
*si
= swap_info
[type
];
1500 struct mm_struct
*start_mm
;
1501 volatile unsigned char *swap_map
; /* swap_map is accessed without
1502 * locking. Mark it as volatile
1503 * to prevent compiler doing
1506 unsigned char swcount
;
1513 * When searching mms for an entry, a good strategy is to
1514 * start at the first mm we freed the previous entry from
1515 * (though actually we don't notice whether we or coincidence
1516 * freed the entry). Initialize this start_mm with a hold.
1518 * A simpler strategy would be to start at the last mm we
1519 * freed the previous entry from; but that would take less
1520 * advantage of mmlist ordering, which clusters forked mms
1521 * together, child after parent. If we race with dup_mmap(), we
1522 * prefer to resolve parent before child, lest we miss entries
1523 * duplicated after we scanned child: using last mm would invert
1526 start_mm
= &init_mm
;
1527 atomic_inc(&init_mm
.mm_users
);
1530 * Keep on scanning until all entries have gone. Usually,
1531 * one pass through swap_map is enough, but not necessarily:
1532 * there are races when an instance of an entry might be missed.
1534 while ((i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
1535 if (signal_pending(current
)) {
1541 * Get a page for the entry, using the existing swap
1542 * cache page if there is one. Otherwise, get a clean
1543 * page and read the swap into it.
1545 swap_map
= &si
->swap_map
[i
];
1546 entry
= swp_entry(type
, i
);
1547 page
= read_swap_cache_async(entry
,
1548 GFP_HIGHUSER_MOVABLE
, NULL
, 0);
1551 * Either swap_duplicate() failed because entry
1552 * has been freed independently, and will not be
1553 * reused since sys_swapoff() already disabled
1554 * allocation from here, or alloc_page() failed.
1556 swcount
= *swap_map
;
1558 * We don't hold lock here, so the swap entry could be
1559 * SWAP_MAP_BAD (when the cluster is discarding).
1560 * Instead of fail out, We can just skip the swap
1561 * entry because swapoff will wait for discarding
1564 if (!swcount
|| swcount
== SWAP_MAP_BAD
)
1571 * Don't hold on to start_mm if it looks like exiting.
1573 if (atomic_read(&start_mm
->mm_users
) == 1) {
1575 start_mm
= &init_mm
;
1576 atomic_inc(&init_mm
.mm_users
);
1580 * Wait for and lock page. When do_swap_page races with
1581 * try_to_unuse, do_swap_page can handle the fault much
1582 * faster than try_to_unuse can locate the entry. This
1583 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1584 * defer to do_swap_page in such a case - in some tests,
1585 * do_swap_page and try_to_unuse repeatedly compete.
1587 wait_on_page_locked(page
);
1588 wait_on_page_writeback(page
);
1590 wait_on_page_writeback(page
);
1593 * Remove all references to entry.
1595 swcount
= *swap_map
;
1596 if (swap_count(swcount
) == SWAP_MAP_SHMEM
) {
1597 retval
= shmem_unuse(entry
, page
);
1598 /* page has already been unlocked and released */
1603 if (swap_count(swcount
) && start_mm
!= &init_mm
)
1604 retval
= unuse_mm(start_mm
, entry
, page
);
1606 if (swap_count(*swap_map
)) {
1607 int set_start_mm
= (*swap_map
>= swcount
);
1608 struct list_head
*p
= &start_mm
->mmlist
;
1609 struct mm_struct
*new_start_mm
= start_mm
;
1610 struct mm_struct
*prev_mm
= start_mm
;
1611 struct mm_struct
*mm
;
1613 atomic_inc(&new_start_mm
->mm_users
);
1614 atomic_inc(&prev_mm
->mm_users
);
1615 spin_lock(&mmlist_lock
);
1616 while (swap_count(*swap_map
) && !retval
&&
1617 (p
= p
->next
) != &start_mm
->mmlist
) {
1618 mm
= list_entry(p
, struct mm_struct
, mmlist
);
1619 if (!atomic_inc_not_zero(&mm
->mm_users
))
1621 spin_unlock(&mmlist_lock
);
1627 swcount
= *swap_map
;
1628 if (!swap_count(swcount
)) /* any usage ? */
1630 else if (mm
== &init_mm
)
1633 retval
= unuse_mm(mm
, entry
, page
);
1635 if (set_start_mm
&& *swap_map
< swcount
) {
1636 mmput(new_start_mm
);
1637 atomic_inc(&mm
->mm_users
);
1641 spin_lock(&mmlist_lock
);
1643 spin_unlock(&mmlist_lock
);
1646 start_mm
= new_start_mm
;
1655 * If a reference remains (rare), we would like to leave
1656 * the page in the swap cache; but try_to_unmap could
1657 * then re-duplicate the entry once we drop page lock,
1658 * so we might loop indefinitely; also, that page could
1659 * not be swapped out to other storage meanwhile. So:
1660 * delete from cache even if there's another reference,
1661 * after ensuring that the data has been saved to disk -
1662 * since if the reference remains (rarer), it will be
1663 * read from disk into another page. Splitting into two
1664 * pages would be incorrect if swap supported "shared
1665 * private" pages, but they are handled by tmpfs files.
1667 * Given how unuse_vma() targets one particular offset
1668 * in an anon_vma, once the anon_vma has been determined,
1669 * this splitting happens to be just what is needed to
1670 * handle where KSM pages have been swapped out: re-reading
1671 * is unnecessarily slow, but we can fix that later on.
1673 if (swap_count(*swap_map
) &&
1674 PageDirty(page
) && PageSwapCache(page
)) {
1675 struct writeback_control wbc
= {
1676 .sync_mode
= WB_SYNC_NONE
,
1679 swap_writepage(page
, &wbc
);
1681 wait_on_page_writeback(page
);
1685 * It is conceivable that a racing task removed this page from
1686 * swap cache just before we acquired the page lock at the top,
1687 * or while we dropped it in unuse_mm(). The page might even
1688 * be back in swap cache on another swap area: that we must not
1689 * delete, since it may not have been written out to swap yet.
1691 if (PageSwapCache(page
) &&
1692 likely(page_private(page
) == entry
.val
))
1693 delete_from_swap_cache(page
);
1696 * So we could skip searching mms once swap count went
1697 * to 1, we did not mark any present ptes as dirty: must
1698 * mark page dirty so shrink_page_list will preserve it.
1705 * Make sure that we aren't completely killing
1706 * interactive performance.
1709 if (frontswap
&& pages_to_unuse
> 0) {
1710 if (!--pages_to_unuse
)
1720 * After a successful try_to_unuse, if no swap is now in use, we know
1721 * we can empty the mmlist. swap_lock must be held on entry and exit.
1722 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1723 * added to the mmlist just after page_duplicate - before would be racy.
1725 static void drain_mmlist(void)
1727 struct list_head
*p
, *next
;
1730 for (type
= 0; type
< nr_swapfiles
; type
++)
1731 if (swap_info
[type
]->inuse_pages
)
1733 spin_lock(&mmlist_lock
);
1734 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
1736 spin_unlock(&mmlist_lock
);
1740 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1741 * corresponds to page offset for the specified swap entry.
1742 * Note that the type of this function is sector_t, but it returns page offset
1743 * into the bdev, not sector offset.
1745 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
1747 struct swap_info_struct
*sis
;
1748 struct swap_extent
*start_se
;
1749 struct swap_extent
*se
;
1752 sis
= swap_info
[swp_type(entry
)];
1755 offset
= swp_offset(entry
);
1756 start_se
= sis
->curr_swap_extent
;
1760 if (se
->start_page
<= offset
&&
1761 offset
< (se
->start_page
+ se
->nr_pages
)) {
1762 return se
->start_block
+ (offset
- se
->start_page
);
1764 se
= list_next_entry(se
, list
);
1765 sis
->curr_swap_extent
= se
;
1766 BUG_ON(se
== start_se
); /* It *must* be present */
1771 * Returns the page offset into bdev for the specified page's swap entry.
1773 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
1776 entry
.val
= page_private(page
);
1777 return map_swap_entry(entry
, bdev
);
1781 * Free all of a swapdev's extent information
1783 static void destroy_swap_extents(struct swap_info_struct
*sis
)
1785 while (!list_empty(&sis
->first_swap_extent
.list
)) {
1786 struct swap_extent
*se
;
1788 se
= list_first_entry(&sis
->first_swap_extent
.list
,
1789 struct swap_extent
, list
);
1790 list_del(&se
->list
);
1794 if (sis
->flags
& SWP_FILE
) {
1795 struct file
*swap_file
= sis
->swap_file
;
1796 struct address_space
*mapping
= swap_file
->f_mapping
;
1798 sis
->flags
&= ~SWP_FILE
;
1799 mapping
->a_ops
->swap_deactivate(swap_file
);
1804 * Add a block range (and the corresponding page range) into this swapdev's
1805 * extent list. The extent list is kept sorted in page order.
1807 * This function rather assumes that it is called in ascending page order.
1810 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
1811 unsigned long nr_pages
, sector_t start_block
)
1813 struct swap_extent
*se
;
1814 struct swap_extent
*new_se
;
1815 struct list_head
*lh
;
1817 if (start_page
== 0) {
1818 se
= &sis
->first_swap_extent
;
1819 sis
->curr_swap_extent
= se
;
1821 se
->nr_pages
= nr_pages
;
1822 se
->start_block
= start_block
;
1825 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
1826 se
= list_entry(lh
, struct swap_extent
, list
);
1827 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
1828 if (se
->start_block
+ se
->nr_pages
== start_block
) {
1830 se
->nr_pages
+= nr_pages
;
1836 * No merge. Insert a new extent, preserving ordering.
1838 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
1841 new_se
->start_page
= start_page
;
1842 new_se
->nr_pages
= nr_pages
;
1843 new_se
->start_block
= start_block
;
1845 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
1850 * A `swap extent' is a simple thing which maps a contiguous range of pages
1851 * onto a contiguous range of disk blocks. An ordered list of swap extents
1852 * is built at swapon time and is then used at swap_writepage/swap_readpage
1853 * time for locating where on disk a page belongs.
1855 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1856 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1857 * swap files identically.
1859 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1860 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1861 * swapfiles are handled *identically* after swapon time.
1863 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1864 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1865 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1866 * requirements, they are simply tossed out - we will never use those blocks
1869 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1870 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1871 * which will scribble on the fs.
1873 * The amount of disk space which a single swap extent represents varies.
1874 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1875 * extents in the list. To avoid much list walking, we cache the previous
1876 * search location in `curr_swap_extent', and start new searches from there.
1877 * This is extremely effective. The average number of iterations in
1878 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1880 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
1882 struct file
*swap_file
= sis
->swap_file
;
1883 struct address_space
*mapping
= swap_file
->f_mapping
;
1884 struct inode
*inode
= mapping
->host
;
1887 if (S_ISBLK(inode
->i_mode
)) {
1888 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1893 if (mapping
->a_ops
->swap_activate
) {
1894 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
1896 sis
->flags
|= SWP_FILE
;
1897 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1903 return generic_swapfile_activate(sis
, swap_file
, span
);
1906 static void _enable_swap_info(struct swap_info_struct
*p
, int prio
,
1907 unsigned char *swap_map
,
1908 struct swap_cluster_info
*cluster_info
)
1913 p
->prio
= --least_priority
;
1915 * the plist prio is negated because plist ordering is
1916 * low-to-high, while swap ordering is high-to-low
1918 p
->list
.prio
= -p
->prio
;
1919 p
->avail_list
.prio
= -p
->prio
;
1920 p
->swap_map
= swap_map
;
1921 p
->cluster_info
= cluster_info
;
1922 p
->flags
|= SWP_WRITEOK
;
1923 atomic_long_add(p
->pages
, &nr_swap_pages
);
1924 total_swap_pages
+= p
->pages
;
1926 assert_spin_locked(&swap_lock
);
1928 * both lists are plists, and thus priority ordered.
1929 * swap_active_head needs to be priority ordered for swapoff(),
1930 * which on removal of any swap_info_struct with an auto-assigned
1931 * (i.e. negative) priority increments the auto-assigned priority
1932 * of any lower-priority swap_info_structs.
1933 * swap_avail_head needs to be priority ordered for get_swap_page(),
1934 * which allocates swap pages from the highest available priority
1937 plist_add(&p
->list
, &swap_active_head
);
1938 spin_lock(&swap_avail_lock
);
1939 plist_add(&p
->avail_list
, &swap_avail_head
);
1940 spin_unlock(&swap_avail_lock
);
1943 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
1944 unsigned char *swap_map
,
1945 struct swap_cluster_info
*cluster_info
,
1946 unsigned long *frontswap_map
)
1948 frontswap_init(p
->type
, frontswap_map
);
1949 spin_lock(&swap_lock
);
1950 spin_lock(&p
->lock
);
1951 _enable_swap_info(p
, prio
, swap_map
, cluster_info
);
1952 spin_unlock(&p
->lock
);
1953 spin_unlock(&swap_lock
);
1956 static void reinsert_swap_info(struct swap_info_struct
*p
)
1958 spin_lock(&swap_lock
);
1959 spin_lock(&p
->lock
);
1960 _enable_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
1961 spin_unlock(&p
->lock
);
1962 spin_unlock(&swap_lock
);
1965 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
1967 struct swap_info_struct
*p
= NULL
;
1968 unsigned char *swap_map
;
1969 struct swap_cluster_info
*cluster_info
;
1970 unsigned long *frontswap_map
;
1971 struct file
*swap_file
, *victim
;
1972 struct address_space
*mapping
;
1973 struct inode
*inode
;
1974 struct filename
*pathname
;
1976 unsigned int old_block_size
;
1978 if (!capable(CAP_SYS_ADMIN
))
1981 BUG_ON(!current
->mm
);
1983 pathname
= getname(specialfile
);
1984 if (IS_ERR(pathname
))
1985 return PTR_ERR(pathname
);
1987 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
1988 err
= PTR_ERR(victim
);
1992 mapping
= victim
->f_mapping
;
1993 spin_lock(&swap_lock
);
1994 plist_for_each_entry(p
, &swap_active_head
, list
) {
1995 if (p
->flags
& SWP_WRITEOK
) {
1996 if (p
->swap_file
->f_mapping
== mapping
) {
2004 spin_unlock(&swap_lock
);
2007 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
2008 vm_unacct_memory(p
->pages
);
2011 spin_unlock(&swap_lock
);
2014 spin_lock(&swap_avail_lock
);
2015 plist_del(&p
->avail_list
, &swap_avail_head
);
2016 spin_unlock(&swap_avail_lock
);
2017 spin_lock(&p
->lock
);
2019 struct swap_info_struct
*si
= p
;
2021 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
2024 si
->avail_list
.prio
--;
2028 plist_del(&p
->list
, &swap_active_head
);
2029 atomic_long_sub(p
->pages
, &nr_swap_pages
);
2030 total_swap_pages
-= p
->pages
;
2031 p
->flags
&= ~SWP_WRITEOK
;
2032 spin_unlock(&p
->lock
);
2033 spin_unlock(&swap_lock
);
2035 set_current_oom_origin();
2036 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
2037 clear_current_oom_origin();
2040 /* re-insert swap space back into swap_list */
2041 reinsert_swap_info(p
);
2045 flush_work(&p
->discard_work
);
2047 destroy_swap_extents(p
);
2048 if (p
->flags
& SWP_CONTINUED
)
2049 free_swap_count_continuations(p
);
2051 mutex_lock(&swapon_mutex
);
2052 spin_lock(&swap_lock
);
2053 spin_lock(&p
->lock
);
2056 /* wait for anyone still in scan_swap_map */
2057 p
->highest_bit
= 0; /* cuts scans short */
2058 while (p
->flags
>= SWP_SCANNING
) {
2059 spin_unlock(&p
->lock
);
2060 spin_unlock(&swap_lock
);
2061 schedule_timeout_uninterruptible(1);
2062 spin_lock(&swap_lock
);
2063 spin_lock(&p
->lock
);
2066 swap_file
= p
->swap_file
;
2067 old_block_size
= p
->old_block_size
;
2068 p
->swap_file
= NULL
;
2070 swap_map
= p
->swap_map
;
2072 cluster_info
= p
->cluster_info
;
2073 p
->cluster_info
= NULL
;
2074 frontswap_map
= frontswap_map_get(p
);
2075 spin_unlock(&p
->lock
);
2076 spin_unlock(&swap_lock
);
2077 frontswap_invalidate_area(p
->type
);
2078 frontswap_map_set(p
, NULL
);
2079 mutex_unlock(&swapon_mutex
);
2080 free_percpu(p
->percpu_cluster
);
2081 p
->percpu_cluster
= NULL
;
2083 vfree(cluster_info
);
2084 vfree(frontswap_map
);
2085 /* Destroy swap account information */
2086 swap_cgroup_swapoff(p
->type
);
2088 inode
= mapping
->host
;
2089 if (S_ISBLK(inode
->i_mode
)) {
2090 struct block_device
*bdev
= I_BDEV(inode
);
2091 set_blocksize(bdev
, old_block_size
);
2092 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2095 inode
->i_flags
&= ~S_SWAPFILE
;
2096 inode_unlock(inode
);
2098 filp_close(swap_file
, NULL
);
2101 * Clear the SWP_USED flag after all resources are freed so that swapon
2102 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2103 * not hold p->lock after we cleared its SWP_WRITEOK.
2105 spin_lock(&swap_lock
);
2107 spin_unlock(&swap_lock
);
2110 atomic_inc(&proc_poll_event
);
2111 wake_up_interruptible(&proc_poll_wait
);
2114 filp_close(victim
, NULL
);
2120 #ifdef CONFIG_PROC_FS
2121 static unsigned swaps_poll(struct file
*file
, poll_table
*wait
)
2123 struct seq_file
*seq
= file
->private_data
;
2125 poll_wait(file
, &proc_poll_wait
, wait
);
2127 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2128 seq
->poll_event
= atomic_read(&proc_poll_event
);
2129 return POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
2132 return POLLIN
| POLLRDNORM
;
2136 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2138 struct swap_info_struct
*si
;
2142 mutex_lock(&swapon_mutex
);
2145 return SEQ_START_TOKEN
;
2147 for (type
= 0; type
< nr_swapfiles
; type
++) {
2148 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2149 si
= swap_info
[type
];
2150 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2159 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2161 struct swap_info_struct
*si
= v
;
2164 if (v
== SEQ_START_TOKEN
)
2167 type
= si
->type
+ 1;
2169 for (; type
< nr_swapfiles
; type
++) {
2170 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2171 si
= swap_info
[type
];
2172 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2181 static void swap_stop(struct seq_file
*swap
, void *v
)
2183 mutex_unlock(&swapon_mutex
);
2186 static int swap_show(struct seq_file
*swap
, void *v
)
2188 struct swap_info_struct
*si
= v
;
2192 if (si
== SEQ_START_TOKEN
) {
2193 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2197 file
= si
->swap_file
;
2198 len
= seq_file_path(swap
, file
, " \t\n\\");
2199 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
2200 len
< 40 ? 40 - len
: 1, " ",
2201 S_ISBLK(file_inode(file
)->i_mode
) ?
2202 "partition" : "file\t",
2203 si
->pages
<< (PAGE_SHIFT
- 10),
2204 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
2209 static const struct seq_operations swaps_op
= {
2210 .start
= swap_start
,
2216 static int swaps_open(struct inode
*inode
, struct file
*file
)
2218 struct seq_file
*seq
;
2221 ret
= seq_open(file
, &swaps_op
);
2225 seq
= file
->private_data
;
2226 seq
->poll_event
= atomic_read(&proc_poll_event
);
2230 static const struct file_operations proc_swaps_operations
= {
2233 .llseek
= seq_lseek
,
2234 .release
= seq_release
,
2238 static int __init
procswaps_init(void)
2240 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
2243 __initcall(procswaps_init
);
2244 #endif /* CONFIG_PROC_FS */
2246 #ifdef MAX_SWAPFILES_CHECK
2247 static int __init
max_swapfiles_check(void)
2249 MAX_SWAPFILES_CHECK();
2252 late_initcall(max_swapfiles_check
);
2255 static struct swap_info_struct
*alloc_swap_info(void)
2257 struct swap_info_struct
*p
;
2260 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
2262 return ERR_PTR(-ENOMEM
);
2264 spin_lock(&swap_lock
);
2265 for (type
= 0; type
< nr_swapfiles
; type
++) {
2266 if (!(swap_info
[type
]->flags
& SWP_USED
))
2269 if (type
>= MAX_SWAPFILES
) {
2270 spin_unlock(&swap_lock
);
2272 return ERR_PTR(-EPERM
);
2274 if (type
>= nr_swapfiles
) {
2276 swap_info
[type
] = p
;
2278 * Write swap_info[type] before nr_swapfiles, in case a
2279 * racing procfs swap_start() or swap_next() is reading them.
2280 * (We never shrink nr_swapfiles, we never free this entry.)
2286 p
= swap_info
[type
];
2288 * Do not memset this entry: a racing procfs swap_next()
2289 * would be relying on p->type to remain valid.
2292 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
2293 plist_node_init(&p
->list
, 0);
2294 plist_node_init(&p
->avail_list
, 0);
2295 p
->flags
= SWP_USED
;
2296 spin_unlock(&swap_lock
);
2297 spin_lock_init(&p
->lock
);
2302 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2306 if (S_ISBLK(inode
->i_mode
)) {
2307 p
->bdev
= bdgrab(I_BDEV(inode
));
2308 error
= blkdev_get(p
->bdev
,
2309 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, p
);
2314 p
->old_block_size
= block_size(p
->bdev
);
2315 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2318 p
->flags
|= SWP_BLKDEV
;
2319 } else if (S_ISREG(inode
->i_mode
)) {
2320 p
->bdev
= inode
->i_sb
->s_bdev
;
2322 if (IS_SWAPFILE(inode
))
2330 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2331 union swap_header
*swap_header
,
2332 struct inode
*inode
)
2335 unsigned long maxpages
;
2336 unsigned long swapfilepages
;
2337 unsigned long last_page
;
2339 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2340 pr_err("Unable to find swap-space signature\n");
2344 /* swap partition endianess hack... */
2345 if (swab32(swap_header
->info
.version
) == 1) {
2346 swab32s(&swap_header
->info
.version
);
2347 swab32s(&swap_header
->info
.last_page
);
2348 swab32s(&swap_header
->info
.nr_badpages
);
2349 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2351 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2352 swab32s(&swap_header
->info
.badpages
[i
]);
2354 /* Check the swap header's sub-version */
2355 if (swap_header
->info
.version
!= 1) {
2356 pr_warn("Unable to handle swap header version %d\n",
2357 swap_header
->info
.version
);
2362 p
->cluster_next
= 1;
2366 * Find out how many pages are allowed for a single swap
2367 * device. There are two limiting factors: 1) the number
2368 * of bits for the swap offset in the swp_entry_t type, and
2369 * 2) the number of bits in the swap pte as defined by the
2370 * different architectures. In order to find the
2371 * largest possible bit mask, a swap entry with swap type 0
2372 * and swap offset ~0UL is created, encoded to a swap pte,
2373 * decoded to a swp_entry_t again, and finally the swap
2374 * offset is extracted. This will mask all the bits from
2375 * the initial ~0UL mask that can't be encoded in either
2376 * the swp_entry_t or the architecture definition of a
2379 maxpages
= swp_offset(pte_to_swp_entry(
2380 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2381 last_page
= swap_header
->info
.last_page
;
2382 if (last_page
> maxpages
) {
2383 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2384 maxpages
<< (PAGE_SHIFT
- 10),
2385 last_page
<< (PAGE_SHIFT
- 10));
2387 if (maxpages
> last_page
) {
2388 maxpages
= last_page
+ 1;
2389 /* p->max is an unsigned int: don't overflow it */
2390 if ((unsigned int)maxpages
== 0)
2391 maxpages
= UINT_MAX
;
2393 p
->highest_bit
= maxpages
- 1;
2397 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
2398 if (swapfilepages
&& maxpages
> swapfilepages
) {
2399 pr_warn("Swap area shorter than signature indicates\n");
2402 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
2404 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2410 #define SWAP_CLUSTER_COLS \
2411 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2413 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
2414 union swap_header
*swap_header
,
2415 unsigned char *swap_map
,
2416 struct swap_cluster_info
*cluster_info
,
2417 unsigned long maxpages
,
2421 unsigned int nr_good_pages
;
2423 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2424 unsigned long col
= p
->cluster_next
/ SWAPFILE_CLUSTER
% SWAP_CLUSTER_COLS
;
2425 unsigned long i
, idx
;
2427 nr_good_pages
= maxpages
- 1; /* omit header page */
2429 cluster_list_init(&p
->free_clusters
);
2430 cluster_list_init(&p
->discard_clusters
);
2432 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
2433 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
2434 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
2436 if (page_nr
< maxpages
) {
2437 swap_map
[page_nr
] = SWAP_MAP_BAD
;
2440 * Haven't marked the cluster free yet, no list
2441 * operation involved
2443 inc_cluster_info_page(p
, cluster_info
, page_nr
);
2447 /* Haven't marked the cluster free yet, no list operation involved */
2448 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
2449 inc_cluster_info_page(p
, cluster_info
, i
);
2451 if (nr_good_pages
) {
2452 swap_map
[0] = SWAP_MAP_BAD
;
2454 * Not mark the cluster free yet, no list
2455 * operation involved
2457 inc_cluster_info_page(p
, cluster_info
, 0);
2459 p
->pages
= nr_good_pages
;
2460 nr_extents
= setup_swap_extents(p
, span
);
2463 nr_good_pages
= p
->pages
;
2465 if (!nr_good_pages
) {
2466 pr_warn("Empty swap-file\n");
2474 /* Reduce false cache line sharing between cluster_info */
2475 for (k
= 0; k
< SWAP_CLUSTER_COLS
; k
++) {
2476 j
= (k
+ col
) % SWAP_CLUSTER_COLS
;
2477 for (i
= 0; i
< DIV_ROUND_UP(nr_clusters
, SWAP_CLUSTER_COLS
); i
++) {
2478 idx
= i
* SWAP_CLUSTER_COLS
+ j
;
2479 if (idx
>= nr_clusters
)
2481 if (cluster_count(&cluster_info
[idx
]))
2483 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
2484 cluster_list_add_tail(&p
->free_clusters
, cluster_info
,
2492 * Helper to sys_swapon determining if a given swap
2493 * backing device queue supports DISCARD operations.
2495 static bool swap_discardable(struct swap_info_struct
*si
)
2497 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
2499 if (!q
|| !blk_queue_discard(q
))
2505 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
2507 struct swap_info_struct
*p
;
2508 struct filename
*name
;
2509 struct file
*swap_file
= NULL
;
2510 struct address_space
*mapping
;
2513 union swap_header
*swap_header
;
2516 unsigned long maxpages
;
2517 unsigned char *swap_map
= NULL
;
2518 struct swap_cluster_info
*cluster_info
= NULL
;
2519 unsigned long *frontswap_map
= NULL
;
2520 struct page
*page
= NULL
;
2521 struct inode
*inode
= NULL
;
2523 if (swap_flags
& ~SWAP_FLAGS_VALID
)
2526 if (!capable(CAP_SYS_ADMIN
))
2529 p
= alloc_swap_info();
2533 INIT_WORK(&p
->discard_work
, swap_discard_work
);
2535 name
= getname(specialfile
);
2537 error
= PTR_ERR(name
);
2541 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
2542 if (IS_ERR(swap_file
)) {
2543 error
= PTR_ERR(swap_file
);
2548 p
->swap_file
= swap_file
;
2549 mapping
= swap_file
->f_mapping
;
2550 inode
= mapping
->host
;
2552 /* If S_ISREG(inode->i_mode) will do inode_lock(inode); */
2553 error
= claim_swapfile(p
, inode
);
2554 if (unlikely(error
))
2558 * Read the swap header.
2560 if (!mapping
->a_ops
->readpage
) {
2564 page
= read_mapping_page(mapping
, 0, swap_file
);
2566 error
= PTR_ERR(page
);
2569 swap_header
= kmap(page
);
2571 maxpages
= read_swap_header(p
, swap_header
, inode
);
2572 if (unlikely(!maxpages
)) {
2577 /* OK, set up the swap map and apply the bad block list */
2578 swap_map
= vzalloc(maxpages
);
2584 if (bdi_cap_stable_pages_required(inode_to_bdi(inode
)))
2585 p
->flags
|= SWP_STABLE_WRITES
;
2587 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
2589 unsigned long ci
, nr_cluster
;
2591 p
->flags
|= SWP_SOLIDSTATE
;
2593 * select a random position to start with to help wear leveling
2596 p
->cluster_next
= 1 + (prandom_u32() % p
->highest_bit
);
2597 nr_cluster
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2599 cluster_info
= vzalloc(nr_cluster
* sizeof(*cluster_info
));
2600 if (!cluster_info
) {
2605 for (ci
= 0; ci
< nr_cluster
; ci
++)
2606 spin_lock_init(&((cluster_info
+ ci
)->lock
));
2608 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
2609 if (!p
->percpu_cluster
) {
2613 for_each_possible_cpu(cpu
) {
2614 struct percpu_cluster
*cluster
;
2615 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
2616 cluster_set_null(&cluster
->index
);
2620 error
= swap_cgroup_swapon(p
->type
, maxpages
);
2624 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
2625 cluster_info
, maxpages
, &span
);
2626 if (unlikely(nr_extents
< 0)) {
2630 /* frontswap enabled? set up bit-per-page map for frontswap */
2631 if (IS_ENABLED(CONFIG_FRONTSWAP
))
2632 frontswap_map
= vzalloc(BITS_TO_LONGS(maxpages
) * sizeof(long));
2634 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
2636 * When discard is enabled for swap with no particular
2637 * policy flagged, we set all swap discard flags here in
2638 * order to sustain backward compatibility with older
2639 * swapon(8) releases.
2641 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
2645 * By flagging sys_swapon, a sysadmin can tell us to
2646 * either do single-time area discards only, or to just
2647 * perform discards for released swap page-clusters.
2648 * Now it's time to adjust the p->flags accordingly.
2650 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
2651 p
->flags
&= ~SWP_PAGE_DISCARD
;
2652 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
2653 p
->flags
&= ~SWP_AREA_DISCARD
;
2655 /* issue a swapon-time discard if it's still required */
2656 if (p
->flags
& SWP_AREA_DISCARD
) {
2657 int err
= discard_swap(p
);
2659 pr_err("swapon: discard_swap(%p): %d\n",
2664 mutex_lock(&swapon_mutex
);
2666 if (swap_flags
& SWAP_FLAG_PREFER
)
2668 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
2669 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
2671 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2672 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
2673 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
2674 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
2675 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
2676 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
2677 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
2678 (frontswap_map
) ? "FS" : "");
2680 mutex_unlock(&swapon_mutex
);
2681 atomic_inc(&proc_poll_event
);
2682 wake_up_interruptible(&proc_poll_wait
);
2684 if (S_ISREG(inode
->i_mode
))
2685 inode
->i_flags
|= S_SWAPFILE
;
2689 free_percpu(p
->percpu_cluster
);
2690 p
->percpu_cluster
= NULL
;
2691 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
2692 set_blocksize(p
->bdev
, p
->old_block_size
);
2693 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2695 destroy_swap_extents(p
);
2696 swap_cgroup_swapoff(p
->type
);
2697 spin_lock(&swap_lock
);
2698 p
->swap_file
= NULL
;
2700 spin_unlock(&swap_lock
);
2702 vfree(cluster_info
);
2704 if (inode
&& S_ISREG(inode
->i_mode
)) {
2705 inode_unlock(inode
);
2708 filp_close(swap_file
, NULL
);
2711 if (page
&& !IS_ERR(page
)) {
2717 if (inode
&& S_ISREG(inode
->i_mode
))
2718 inode_unlock(inode
);
2722 void si_swapinfo(struct sysinfo
*val
)
2725 unsigned long nr_to_be_unused
= 0;
2727 spin_lock(&swap_lock
);
2728 for (type
= 0; type
< nr_swapfiles
; type
++) {
2729 struct swap_info_struct
*si
= swap_info
[type
];
2731 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
2732 nr_to_be_unused
+= si
->inuse_pages
;
2734 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
2735 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
2736 spin_unlock(&swap_lock
);
2740 * Verify that a swap entry is valid and increment its swap map count.
2742 * Returns error code in following case.
2744 * - swp_entry is invalid -> EINVAL
2745 * - swp_entry is migration entry -> EINVAL
2746 * - swap-cache reference is requested but there is already one. -> EEXIST
2747 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2748 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2750 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
2752 struct swap_info_struct
*p
;
2753 struct swap_cluster_info
*ci
;
2754 unsigned long offset
, type
;
2755 unsigned char count
;
2756 unsigned char has_cache
;
2759 if (non_swap_entry(entry
))
2762 type
= swp_type(entry
);
2763 if (type
>= nr_swapfiles
)
2765 p
= swap_info
[type
];
2766 offset
= swp_offset(entry
);
2767 if (unlikely(offset
>= p
->max
))
2770 ci
= lock_cluster_or_swap_info(p
, offset
);
2772 count
= p
->swap_map
[offset
];
2775 * swapin_readahead() doesn't check if a swap entry is valid, so the
2776 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2778 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
2783 has_cache
= count
& SWAP_HAS_CACHE
;
2784 count
&= ~SWAP_HAS_CACHE
;
2787 if (usage
== SWAP_HAS_CACHE
) {
2789 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2790 if (!has_cache
&& count
)
2791 has_cache
= SWAP_HAS_CACHE
;
2792 else if (has_cache
) /* someone else added cache */
2794 else /* no users remaining */
2797 } else if (count
|| has_cache
) {
2799 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
2801 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
2803 else if (swap_count_continued(p
, offset
, count
))
2804 count
= COUNT_CONTINUED
;
2808 err
= -ENOENT
; /* unused swap entry */
2810 p
->swap_map
[offset
] = count
| has_cache
;
2813 unlock_cluster_or_swap_info(p
, ci
);
2818 pr_err("swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
2823 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2824 * (in which case its reference count is never incremented).
2826 void swap_shmem_alloc(swp_entry_t entry
)
2828 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
2832 * Increase reference count of swap entry by 1.
2833 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2834 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2835 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2836 * might occur if a page table entry has got corrupted.
2838 int swap_duplicate(swp_entry_t entry
)
2842 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
2843 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
2848 * @entry: swap entry for which we allocate swap cache.
2850 * Called when allocating swap cache for existing swap entry,
2851 * This can return error codes. Returns 0 at success.
2852 * -EBUSY means there is a swap cache.
2853 * Note: return code is different from swap_duplicate().
2855 int swapcache_prepare(swp_entry_t entry
)
2857 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
2860 struct swap_info_struct
*page_swap_info(struct page
*page
)
2862 swp_entry_t swap
= { .val
= page_private(page
) };
2863 return swap_info
[swp_type(swap
)];
2867 * out-of-line __page_file_ methods to avoid include hell.
2869 struct address_space
*__page_file_mapping(struct page
*page
)
2871 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
2872 return page_swap_info(page
)->swap_file
->f_mapping
;
2874 EXPORT_SYMBOL_GPL(__page_file_mapping
);
2876 pgoff_t
__page_file_index(struct page
*page
)
2878 swp_entry_t swap
= { .val
= page_private(page
) };
2879 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
2880 return swp_offset(swap
);
2882 EXPORT_SYMBOL_GPL(__page_file_index
);
2885 * add_swap_count_continuation - called when a swap count is duplicated
2886 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2887 * page of the original vmalloc'ed swap_map, to hold the continuation count
2888 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2889 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2891 * These continuation pages are seldom referenced: the common paths all work
2892 * on the original swap_map, only referring to a continuation page when the
2893 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2895 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2896 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2897 * can be called after dropping locks.
2899 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
2901 struct swap_info_struct
*si
;
2902 struct swap_cluster_info
*ci
;
2905 struct page
*list_page
;
2907 unsigned char count
;
2910 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2911 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2913 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
2915 si
= swap_info_get(entry
);
2918 * An acceptable race has occurred since the failing
2919 * __swap_duplicate(): the swap entry has been freed,
2920 * perhaps even the whole swap_map cleared for swapoff.
2925 offset
= swp_offset(entry
);
2927 ci
= lock_cluster(si
, offset
);
2929 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
2931 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
2933 * The higher the swap count, the more likely it is that tasks
2934 * will race to add swap count continuation: we need to avoid
2935 * over-provisioning.
2942 spin_unlock(&si
->lock
);
2947 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2948 * no architecture is using highmem pages for kernel page tables: so it
2949 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
2951 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2952 offset
&= ~PAGE_MASK
;
2955 * Page allocation does not initialize the page's lru field,
2956 * but it does always reset its private field.
2958 if (!page_private(head
)) {
2959 BUG_ON(count
& COUNT_CONTINUED
);
2960 INIT_LIST_HEAD(&head
->lru
);
2961 set_page_private(head
, SWP_CONTINUED
);
2962 si
->flags
|= SWP_CONTINUED
;
2965 list_for_each_entry(list_page
, &head
->lru
, lru
) {
2969 * If the previous map said no continuation, but we've found
2970 * a continuation page, free our allocation and use this one.
2972 if (!(count
& COUNT_CONTINUED
))
2975 map
= kmap_atomic(list_page
) + offset
;
2980 * If this continuation count now has some space in it,
2981 * free our allocation and use this one.
2983 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
2987 list_add_tail(&page
->lru
, &head
->lru
);
2988 page
= NULL
; /* now it's attached, don't free it */
2991 spin_unlock(&si
->lock
);
2999 * swap_count_continued - when the original swap_map count is incremented
3000 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3001 * into, carry if so, or else fail until a new continuation page is allocated;
3002 * when the original swap_map count is decremented from 0 with continuation,
3003 * borrow from the continuation and report whether it still holds more.
3004 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3007 static bool swap_count_continued(struct swap_info_struct
*si
,
3008 pgoff_t offset
, unsigned char count
)
3014 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3015 if (page_private(head
) != SWP_CONTINUED
) {
3016 BUG_ON(count
& COUNT_CONTINUED
);
3017 return false; /* need to add count continuation */
3020 offset
&= ~PAGE_MASK
;
3021 page
= list_entry(head
->lru
.next
, struct page
, lru
);
3022 map
= kmap_atomic(page
) + offset
;
3024 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
3025 goto init_map
; /* jump over SWAP_CONT_MAX checks */
3027 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
3029 * Think of how you add 1 to 999
3031 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
3033 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3034 BUG_ON(page
== head
);
3035 map
= kmap_atomic(page
) + offset
;
3037 if (*map
== SWAP_CONT_MAX
) {
3039 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3041 return false; /* add count continuation */
3042 map
= kmap_atomic(page
) + offset
;
3043 init_map
: *map
= 0; /* we didn't zero the page */
3047 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3048 while (page
!= head
) {
3049 map
= kmap_atomic(page
) + offset
;
3050 *map
= COUNT_CONTINUED
;
3052 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3054 return true; /* incremented */
3056 } else { /* decrementing */
3058 * Think of how you subtract 1 from 1000
3060 BUG_ON(count
!= COUNT_CONTINUED
);
3061 while (*map
== COUNT_CONTINUED
) {
3063 page
= list_entry(page
->lru
.next
, struct page
, lru
);
3064 BUG_ON(page
== head
);
3065 map
= kmap_atomic(page
) + offset
;
3072 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3073 while (page
!= head
) {
3074 map
= kmap_atomic(page
) + offset
;
3075 *map
= SWAP_CONT_MAX
| count
;
3076 count
= COUNT_CONTINUED
;
3078 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
3080 return count
== COUNT_CONTINUED
;
3085 * free_swap_count_continuations - swapoff free all the continuation pages
3086 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3088 static void free_swap_count_continuations(struct swap_info_struct
*si
)
3092 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
3094 head
= vmalloc_to_page(si
->swap_map
+ offset
);
3095 if (page_private(head
)) {
3096 struct page
*page
, *next
;
3098 list_for_each_entry_safe(page
, next
, &head
->lru
, lru
) {
3099 list_del(&page
->lru
);