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
), entry
.val
);
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
);
122 page_cache_release(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 struct list_head
*lh
;
176 if (se
->start_page
<= start_page
&&
177 start_page
< se
->start_page
+ se
->nr_pages
) {
178 pgoff_t offset
= start_page
- se
->start_page
;
179 sector_t start_block
= se
->start_block
+ offset
;
180 sector_t nr_blocks
= se
->nr_pages
- offset
;
182 if (nr_blocks
> nr_pages
)
183 nr_blocks
= nr_pages
;
184 start_page
+= nr_blocks
;
185 nr_pages
-= nr_blocks
;
188 si
->curr_swap_extent
= se
;
190 start_block
<<= PAGE_SHIFT
- 9;
191 nr_blocks
<<= PAGE_SHIFT
- 9;
192 if (blkdev_issue_discard(si
->bdev
, start_block
,
193 nr_blocks
, GFP_NOIO
, 0))
198 se
= list_entry(lh
, struct swap_extent
, list
);
202 #define SWAPFILE_CLUSTER 256
203 #define LATENCY_LIMIT 256
205 static inline void cluster_set_flag(struct swap_cluster_info
*info
,
211 static inline unsigned int cluster_count(struct swap_cluster_info
*info
)
216 static inline void cluster_set_count(struct swap_cluster_info
*info
,
222 static inline void cluster_set_count_flag(struct swap_cluster_info
*info
,
223 unsigned int c
, unsigned int f
)
229 static inline unsigned int cluster_next(struct swap_cluster_info
*info
)
234 static inline void cluster_set_next(struct swap_cluster_info
*info
,
240 static inline void cluster_set_next_flag(struct swap_cluster_info
*info
,
241 unsigned int n
, unsigned int f
)
247 static inline bool cluster_is_free(struct swap_cluster_info
*info
)
249 return info
->flags
& CLUSTER_FLAG_FREE
;
252 static inline bool cluster_is_null(struct swap_cluster_info
*info
)
254 return info
->flags
& CLUSTER_FLAG_NEXT_NULL
;
257 static inline void cluster_set_null(struct swap_cluster_info
*info
)
259 info
->flags
= CLUSTER_FLAG_NEXT_NULL
;
263 /* Add a cluster to discard list and schedule it to do discard */
264 static void swap_cluster_schedule_discard(struct swap_info_struct
*si
,
268 * If scan_swap_map() can't find a free cluster, it will check
269 * si->swap_map directly. To make sure the discarding cluster isn't
270 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
271 * will be cleared after discard
273 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
274 SWAP_MAP_BAD
, SWAPFILE_CLUSTER
);
276 if (cluster_is_null(&si
->discard_cluster_head
)) {
277 cluster_set_next_flag(&si
->discard_cluster_head
,
279 cluster_set_next_flag(&si
->discard_cluster_tail
,
282 unsigned int tail
= cluster_next(&si
->discard_cluster_tail
);
283 cluster_set_next(&si
->cluster_info
[tail
], idx
);
284 cluster_set_next_flag(&si
->discard_cluster_tail
,
288 schedule_work(&si
->discard_work
);
292 * Doing discard actually. After a cluster discard is finished, the cluster
293 * will be added to free cluster list. caller should hold si->lock.
295 static void swap_do_scheduled_discard(struct swap_info_struct
*si
)
297 struct swap_cluster_info
*info
;
300 info
= si
->cluster_info
;
302 while (!cluster_is_null(&si
->discard_cluster_head
)) {
303 idx
= cluster_next(&si
->discard_cluster_head
);
305 cluster_set_next_flag(&si
->discard_cluster_head
,
306 cluster_next(&info
[idx
]), 0);
307 if (cluster_next(&si
->discard_cluster_tail
) == idx
) {
308 cluster_set_null(&si
->discard_cluster_head
);
309 cluster_set_null(&si
->discard_cluster_tail
);
311 spin_unlock(&si
->lock
);
313 discard_swap_cluster(si
, idx
* SWAPFILE_CLUSTER
,
316 spin_lock(&si
->lock
);
317 cluster_set_flag(&info
[idx
], CLUSTER_FLAG_FREE
);
318 if (cluster_is_null(&si
->free_cluster_head
)) {
319 cluster_set_next_flag(&si
->free_cluster_head
,
321 cluster_set_next_flag(&si
->free_cluster_tail
,
326 tail
= cluster_next(&si
->free_cluster_tail
);
327 cluster_set_next(&info
[tail
], idx
);
328 cluster_set_next_flag(&si
->free_cluster_tail
,
331 memset(si
->swap_map
+ idx
* SWAPFILE_CLUSTER
,
332 0, SWAPFILE_CLUSTER
);
336 static void swap_discard_work(struct work_struct
*work
)
338 struct swap_info_struct
*si
;
340 si
= container_of(work
, struct swap_info_struct
, discard_work
);
342 spin_lock(&si
->lock
);
343 swap_do_scheduled_discard(si
);
344 spin_unlock(&si
->lock
);
348 * The cluster corresponding to page_nr will be used. The cluster will be
349 * removed from free cluster list and its usage counter will be increased.
351 static void inc_cluster_info_page(struct swap_info_struct
*p
,
352 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
354 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
358 if (cluster_is_free(&cluster_info
[idx
])) {
359 VM_BUG_ON(cluster_next(&p
->free_cluster_head
) != idx
);
360 cluster_set_next_flag(&p
->free_cluster_head
,
361 cluster_next(&cluster_info
[idx
]), 0);
362 if (cluster_next(&p
->free_cluster_tail
) == idx
) {
363 cluster_set_null(&p
->free_cluster_tail
);
364 cluster_set_null(&p
->free_cluster_head
);
366 cluster_set_count_flag(&cluster_info
[idx
], 0, 0);
369 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) >= SWAPFILE_CLUSTER
);
370 cluster_set_count(&cluster_info
[idx
],
371 cluster_count(&cluster_info
[idx
]) + 1);
375 * The cluster corresponding to page_nr decreases one usage. If the usage
376 * counter becomes 0, which means no page in the cluster is in using, we can
377 * optionally discard the cluster and add it to free cluster list.
379 static void dec_cluster_info_page(struct swap_info_struct
*p
,
380 struct swap_cluster_info
*cluster_info
, unsigned long page_nr
)
382 unsigned long idx
= page_nr
/ SWAPFILE_CLUSTER
;
387 VM_BUG_ON(cluster_count(&cluster_info
[idx
]) == 0);
388 cluster_set_count(&cluster_info
[idx
],
389 cluster_count(&cluster_info
[idx
]) - 1);
391 if (cluster_count(&cluster_info
[idx
]) == 0) {
393 * If the swap is discardable, prepare discard the cluster
394 * instead of free it immediately. The cluster will be freed
397 if ((p
->flags
& (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) ==
398 (SWP_WRITEOK
| SWP_PAGE_DISCARD
)) {
399 swap_cluster_schedule_discard(p
, idx
);
403 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
404 if (cluster_is_null(&p
->free_cluster_head
)) {
405 cluster_set_next_flag(&p
->free_cluster_head
, idx
, 0);
406 cluster_set_next_flag(&p
->free_cluster_tail
, idx
, 0);
408 unsigned int tail
= cluster_next(&p
->free_cluster_tail
);
409 cluster_set_next(&cluster_info
[tail
], idx
);
410 cluster_set_next_flag(&p
->free_cluster_tail
, idx
, 0);
416 * It's possible scan_swap_map() uses a free cluster in the middle of free
417 * cluster list. Avoiding such abuse to avoid list corruption.
420 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct
*si
,
421 unsigned long offset
)
423 struct percpu_cluster
*percpu_cluster
;
426 offset
/= SWAPFILE_CLUSTER
;
427 conflict
= !cluster_is_null(&si
->free_cluster_head
) &&
428 offset
!= cluster_next(&si
->free_cluster_head
) &&
429 cluster_is_free(&si
->cluster_info
[offset
]);
434 percpu_cluster
= this_cpu_ptr(si
->percpu_cluster
);
435 cluster_set_null(&percpu_cluster
->index
);
440 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
441 * might involve allocating a new cluster for current CPU too.
443 static void scan_swap_map_try_ssd_cluster(struct swap_info_struct
*si
,
444 unsigned long *offset
, unsigned long *scan_base
)
446 struct percpu_cluster
*cluster
;
451 cluster
= this_cpu_ptr(si
->percpu_cluster
);
452 if (cluster_is_null(&cluster
->index
)) {
453 if (!cluster_is_null(&si
->free_cluster_head
)) {
454 cluster
->index
= si
->free_cluster_head
;
455 cluster
->next
= cluster_next(&cluster
->index
) *
457 } else if (!cluster_is_null(&si
->discard_cluster_head
)) {
459 * we don't have free cluster but have some clusters in
460 * discarding, do discard now and reclaim them
462 swap_do_scheduled_discard(si
);
463 *scan_base
= *offset
= si
->cluster_next
;
472 * Other CPUs can use our cluster if they can't find a free cluster,
473 * check if there is still free entry in the cluster
476 while (tmp
< si
->max
&& tmp
< (cluster_next(&cluster
->index
) + 1) *
478 if (!si
->swap_map
[tmp
]) {
485 cluster_set_null(&cluster
->index
);
488 cluster
->next
= tmp
+ 1;
493 static unsigned long scan_swap_map(struct swap_info_struct
*si
,
496 unsigned long offset
;
497 unsigned long scan_base
;
498 unsigned long last_in_cluster
= 0;
499 int latency_ration
= LATENCY_LIMIT
;
502 * We try to cluster swap pages by allocating them sequentially
503 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
504 * way, however, we resort to first-free allocation, starting
505 * a new cluster. This prevents us from scattering swap pages
506 * all over the entire swap partition, so that we reduce
507 * overall disk seek times between swap pages. -- sct
508 * But we do now try to find an empty cluster. -Andrea
509 * And we let swap pages go all over an SSD partition. Hugh
512 si
->flags
+= SWP_SCANNING
;
513 scan_base
= offset
= si
->cluster_next
;
516 if (si
->cluster_info
) {
517 scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
);
521 if (unlikely(!si
->cluster_nr
--)) {
522 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
523 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
527 spin_unlock(&si
->lock
);
530 * If seek is expensive, start searching for new cluster from
531 * start of partition, to minimize the span of allocated swap.
532 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
533 * case, just handled by scan_swap_map_try_ssd_cluster() above.
535 scan_base
= offset
= si
->lowest_bit
;
536 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
538 /* Locate the first empty (unaligned) cluster */
539 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
540 if (si
->swap_map
[offset
])
541 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
542 else if (offset
== last_in_cluster
) {
543 spin_lock(&si
->lock
);
544 offset
-= SWAPFILE_CLUSTER
- 1;
545 si
->cluster_next
= offset
;
546 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
549 if (unlikely(--latency_ration
< 0)) {
551 latency_ration
= LATENCY_LIMIT
;
556 spin_lock(&si
->lock
);
557 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
561 if (si
->cluster_info
) {
562 while (scan_swap_map_ssd_cluster_conflict(si
, offset
))
563 scan_swap_map_try_ssd_cluster(si
, &offset
, &scan_base
);
565 if (!(si
->flags
& SWP_WRITEOK
))
567 if (!si
->highest_bit
)
569 if (offset
> si
->highest_bit
)
570 scan_base
= offset
= si
->lowest_bit
;
572 /* reuse swap entry of cache-only swap if not busy. */
573 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
575 spin_unlock(&si
->lock
);
576 swap_was_freed
= __try_to_reclaim_swap(si
, offset
);
577 spin_lock(&si
->lock
);
578 /* entry was freed successfully, try to use this again */
581 goto scan
; /* check next one */
584 if (si
->swap_map
[offset
])
587 if (offset
== si
->lowest_bit
)
589 if (offset
== si
->highest_bit
)
592 if (si
->inuse_pages
== si
->pages
) {
593 si
->lowest_bit
= si
->max
;
595 spin_lock(&swap_avail_lock
);
596 plist_del(&si
->avail_list
, &swap_avail_head
);
597 spin_unlock(&swap_avail_lock
);
599 si
->swap_map
[offset
] = usage
;
600 inc_cluster_info_page(si
, si
->cluster_info
, offset
);
601 si
->cluster_next
= offset
+ 1;
602 si
->flags
-= SWP_SCANNING
;
607 spin_unlock(&si
->lock
);
608 while (++offset
<= si
->highest_bit
) {
609 if (!si
->swap_map
[offset
]) {
610 spin_lock(&si
->lock
);
613 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
614 spin_lock(&si
->lock
);
617 if (unlikely(--latency_ration
< 0)) {
619 latency_ration
= LATENCY_LIMIT
;
622 offset
= si
->lowest_bit
;
623 while (offset
< scan_base
) {
624 if (!si
->swap_map
[offset
]) {
625 spin_lock(&si
->lock
);
628 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
629 spin_lock(&si
->lock
);
632 if (unlikely(--latency_ration
< 0)) {
634 latency_ration
= LATENCY_LIMIT
;
638 spin_lock(&si
->lock
);
641 si
->flags
-= SWP_SCANNING
;
645 swp_entry_t
get_swap_page(void)
647 struct swap_info_struct
*si
, *next
;
650 if (atomic_long_read(&nr_swap_pages
) <= 0)
652 atomic_long_dec(&nr_swap_pages
);
654 spin_lock(&swap_avail_lock
);
657 plist_for_each_entry_safe(si
, next
, &swap_avail_head
, avail_list
) {
658 /* requeue si to after same-priority siblings */
659 plist_requeue(&si
->avail_list
, &swap_avail_head
);
660 spin_unlock(&swap_avail_lock
);
661 spin_lock(&si
->lock
);
662 if (!si
->highest_bit
|| !(si
->flags
& SWP_WRITEOK
)) {
663 spin_lock(&swap_avail_lock
);
664 if (plist_node_empty(&si
->avail_list
)) {
665 spin_unlock(&si
->lock
);
668 WARN(!si
->highest_bit
,
669 "swap_info %d in list but !highest_bit\n",
671 WARN(!(si
->flags
& SWP_WRITEOK
),
672 "swap_info %d in list but !SWP_WRITEOK\n",
674 plist_del(&si
->avail_list
, &swap_avail_head
);
675 spin_unlock(&si
->lock
);
679 /* This is called for allocating swap entry for cache */
680 offset
= scan_swap_map(si
, SWAP_HAS_CACHE
);
681 spin_unlock(&si
->lock
);
683 return swp_entry(si
->type
, offset
);
684 pr_debug("scan_swap_map of si %d failed to find offset\n",
686 spin_lock(&swap_avail_lock
);
689 * if we got here, it's likely that si was almost full before,
690 * and since scan_swap_map() can drop the si->lock, multiple
691 * callers probably all tried to get a page from the same si
692 * and it filled up before we could get one; or, the si filled
693 * up between us dropping swap_avail_lock and taking si->lock.
694 * Since we dropped the swap_avail_lock, the swap_avail_head
695 * list may have been modified; so if next is still in the
696 * swap_avail_head list then try it, otherwise start over.
698 if (plist_node_empty(&next
->avail_list
))
702 spin_unlock(&swap_avail_lock
);
704 atomic_long_inc(&nr_swap_pages
);
706 return (swp_entry_t
) {0};
709 /* The only caller of this function is now suspend routine */
710 swp_entry_t
get_swap_page_of_type(int type
)
712 struct swap_info_struct
*si
;
715 si
= swap_info
[type
];
716 spin_lock(&si
->lock
);
717 if (si
&& (si
->flags
& SWP_WRITEOK
)) {
718 atomic_long_dec(&nr_swap_pages
);
719 /* This is called for allocating swap entry, not cache */
720 offset
= scan_swap_map(si
, 1);
722 spin_unlock(&si
->lock
);
723 return swp_entry(type
, offset
);
725 atomic_long_inc(&nr_swap_pages
);
727 spin_unlock(&si
->lock
);
728 return (swp_entry_t
) {0};
731 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
733 struct swap_info_struct
*p
;
734 unsigned long offset
, type
;
738 type
= swp_type(entry
);
739 if (type
>= nr_swapfiles
)
742 if (!(p
->flags
& SWP_USED
))
744 offset
= swp_offset(entry
);
745 if (offset
>= p
->max
)
747 if (!p
->swap_map
[offset
])
753 pr_err("swap_free: %s%08lx\n", Unused_offset
, entry
.val
);
756 pr_err("swap_free: %s%08lx\n", Bad_offset
, entry
.val
);
759 pr_err("swap_free: %s%08lx\n", Unused_file
, entry
.val
);
762 pr_err("swap_free: %s%08lx\n", Bad_file
, entry
.val
);
767 static unsigned char swap_entry_free(struct swap_info_struct
*p
,
768 swp_entry_t entry
, unsigned char usage
)
770 unsigned long offset
= swp_offset(entry
);
772 unsigned char has_cache
;
774 count
= p
->swap_map
[offset
];
775 has_cache
= count
& SWAP_HAS_CACHE
;
776 count
&= ~SWAP_HAS_CACHE
;
778 if (usage
== SWAP_HAS_CACHE
) {
779 VM_BUG_ON(!has_cache
);
781 } else if (count
== SWAP_MAP_SHMEM
) {
783 * Or we could insist on shmem.c using a special
784 * swap_shmem_free() and free_shmem_swap_and_cache()...
787 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
788 if (count
== COUNT_CONTINUED
) {
789 if (swap_count_continued(p
, offset
, count
))
790 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
792 count
= SWAP_MAP_MAX
;
798 mem_cgroup_uncharge_swap(entry
);
800 usage
= count
| has_cache
;
801 p
->swap_map
[offset
] = usage
;
803 /* free if no reference */
805 dec_cluster_info_page(p
, p
->cluster_info
, offset
);
806 if (offset
< p
->lowest_bit
)
807 p
->lowest_bit
= offset
;
808 if (offset
> p
->highest_bit
) {
809 bool was_full
= !p
->highest_bit
;
810 p
->highest_bit
= offset
;
811 if (was_full
&& (p
->flags
& SWP_WRITEOK
)) {
812 spin_lock(&swap_avail_lock
);
813 WARN_ON(!plist_node_empty(&p
->avail_list
));
814 if (plist_node_empty(&p
->avail_list
))
815 plist_add(&p
->avail_list
,
817 spin_unlock(&swap_avail_lock
);
820 atomic_long_inc(&nr_swap_pages
);
822 frontswap_invalidate_page(p
->type
, offset
);
823 if (p
->flags
& SWP_BLKDEV
) {
824 struct gendisk
*disk
= p
->bdev
->bd_disk
;
825 if (disk
->fops
->swap_slot_free_notify
)
826 disk
->fops
->swap_slot_free_notify(p
->bdev
,
835 * Caller has made sure that the swap device corresponding to entry
836 * is still around or has not been recycled.
838 void swap_free(swp_entry_t entry
)
840 struct swap_info_struct
*p
;
842 p
= swap_info_get(entry
);
844 swap_entry_free(p
, entry
, 1);
845 spin_unlock(&p
->lock
);
850 * Called after dropping swapcache to decrease refcnt to swap entries.
852 void swapcache_free(swp_entry_t entry
)
854 struct swap_info_struct
*p
;
856 p
= swap_info_get(entry
);
858 swap_entry_free(p
, entry
, SWAP_HAS_CACHE
);
859 spin_unlock(&p
->lock
);
864 * How many references to page are currently swapped out?
865 * This does not give an exact answer when swap count is continued,
866 * but does include the high COUNT_CONTINUED flag to allow for that.
868 int page_swapcount(struct page
*page
)
871 struct swap_info_struct
*p
;
874 entry
.val
= page_private(page
);
875 p
= swap_info_get(entry
);
877 count
= swap_count(p
->swap_map
[swp_offset(entry
)]);
878 spin_unlock(&p
->lock
);
884 * How many references to @entry are currently swapped out?
885 * This considers COUNT_CONTINUED so it returns exact answer.
887 int swp_swapcount(swp_entry_t entry
)
889 int count
, tmp_count
, n
;
890 struct swap_info_struct
*p
;
895 p
= swap_info_get(entry
);
899 count
= swap_count(p
->swap_map
[swp_offset(entry
)]);
900 if (!(count
& COUNT_CONTINUED
))
903 count
&= ~COUNT_CONTINUED
;
904 n
= SWAP_MAP_MAX
+ 1;
906 offset
= swp_offset(entry
);
907 page
= vmalloc_to_page(p
->swap_map
+ offset
);
908 offset
&= ~PAGE_MASK
;
909 VM_BUG_ON(page_private(page
) != SWP_CONTINUED
);
912 page
= list_entry(page
->lru
.next
, struct page
, lru
);
913 map
= kmap_atomic(page
);
914 tmp_count
= map
[offset
];
917 count
+= (tmp_count
& ~COUNT_CONTINUED
) * n
;
918 n
*= (SWAP_CONT_MAX
+ 1);
919 } while (tmp_count
& COUNT_CONTINUED
);
921 spin_unlock(&p
->lock
);
926 * We can write to an anon page without COW if there are no other references
927 * to it. And as a side-effect, free up its swap: because the old content
928 * on disk will never be read, and seeking back there to write new content
929 * later would only waste time away from clustering.
931 int reuse_swap_page(struct page
*page
)
935 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
936 if (unlikely(PageKsm(page
)))
938 count
= page_mapcount(page
);
939 if (count
<= 1 && PageSwapCache(page
)) {
940 count
+= page_swapcount(page
);
941 if (count
== 1 && !PageWriteback(page
)) {
942 delete_from_swap_cache(page
);
950 * If swap is getting full, or if there are no more mappings of this page,
951 * then try_to_free_swap is called to free its swap space.
953 int try_to_free_swap(struct page
*page
)
955 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
957 if (!PageSwapCache(page
))
959 if (PageWriteback(page
))
961 if (page_swapcount(page
))
965 * Once hibernation has begun to create its image of memory,
966 * there's a danger that one of the calls to try_to_free_swap()
967 * - most probably a call from __try_to_reclaim_swap() while
968 * hibernation is allocating its own swap pages for the image,
969 * but conceivably even a call from memory reclaim - will free
970 * the swap from a page which has already been recorded in the
971 * image as a clean swapcache page, and then reuse its swap for
972 * another page of the image. On waking from hibernation, the
973 * original page might be freed under memory pressure, then
974 * later read back in from swap, now with the wrong data.
976 * Hibernation suspends storage while it is writing the image
977 * to disk so check that here.
979 if (pm_suspended_storage())
982 delete_from_swap_cache(page
);
988 * Free the swap entry like above, but also try to
989 * free the page cache entry if it is the last user.
991 int free_swap_and_cache(swp_entry_t entry
)
993 struct swap_info_struct
*p
;
994 struct page
*page
= NULL
;
996 if (non_swap_entry(entry
))
999 p
= swap_info_get(entry
);
1001 if (swap_entry_free(p
, entry
, 1) == SWAP_HAS_CACHE
) {
1002 page
= find_get_page(swap_address_space(entry
),
1004 if (page
&& !trylock_page(page
)) {
1005 page_cache_release(page
);
1009 spin_unlock(&p
->lock
);
1013 * Not mapped elsewhere, or swap space full? Free it!
1014 * Also recheck PageSwapCache now page is locked (above).
1016 if (PageSwapCache(page
) && !PageWriteback(page
) &&
1017 (!page_mapped(page
) || vm_swap_full())) {
1018 delete_from_swap_cache(page
);
1022 page_cache_release(page
);
1027 #ifdef CONFIG_HIBERNATION
1029 * Find the swap type that corresponds to given device (if any).
1031 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1032 * from 0, in which the swap header is expected to be located.
1034 * This is needed for the suspend to disk (aka swsusp).
1036 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
1038 struct block_device
*bdev
= NULL
;
1042 bdev
= bdget(device
);
1044 spin_lock(&swap_lock
);
1045 for (type
= 0; type
< nr_swapfiles
; type
++) {
1046 struct swap_info_struct
*sis
= swap_info
[type
];
1048 if (!(sis
->flags
& SWP_WRITEOK
))
1053 *bdev_p
= bdgrab(sis
->bdev
);
1055 spin_unlock(&swap_lock
);
1058 if (bdev
== sis
->bdev
) {
1059 struct swap_extent
*se
= &sis
->first_swap_extent
;
1061 if (se
->start_block
== offset
) {
1063 *bdev_p
= bdgrab(sis
->bdev
);
1065 spin_unlock(&swap_lock
);
1071 spin_unlock(&swap_lock
);
1079 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1080 * corresponding to given index in swap_info (swap type).
1082 sector_t
swapdev_block(int type
, pgoff_t offset
)
1084 struct block_device
*bdev
;
1086 if ((unsigned int)type
>= nr_swapfiles
)
1088 if (!(swap_info
[type
]->flags
& SWP_WRITEOK
))
1090 return map_swap_entry(swp_entry(type
, offset
), &bdev
);
1094 * Return either the total number of swap pages of given type, or the number
1095 * of free pages of that type (depending on @free)
1097 * This is needed for software suspend
1099 unsigned int count_swap_pages(int type
, int free
)
1103 spin_lock(&swap_lock
);
1104 if ((unsigned int)type
< nr_swapfiles
) {
1105 struct swap_info_struct
*sis
= swap_info
[type
];
1107 spin_lock(&sis
->lock
);
1108 if (sis
->flags
& SWP_WRITEOK
) {
1111 n
-= sis
->inuse_pages
;
1113 spin_unlock(&sis
->lock
);
1115 spin_unlock(&swap_lock
);
1118 #endif /* CONFIG_HIBERNATION */
1120 static inline int maybe_same_pte(pte_t pte
, pte_t swp_pte
)
1122 #ifdef CONFIG_MEM_SOFT_DIRTY
1124 * When pte keeps soft dirty bit the pte generated
1125 * from swap entry does not has it, still it's same
1126 * pte from logical point of view.
1128 pte_t swp_pte_dirty
= pte_swp_mksoft_dirty(swp_pte
);
1129 return pte_same(pte
, swp_pte
) || pte_same(pte
, swp_pte_dirty
);
1131 return pte_same(pte
, swp_pte
);
1136 * No need to decide whether this PTE shares the swap entry with others,
1137 * just let do_wp_page work it out if a write is requested later - to
1138 * force COW, vm_page_prot omits write permission from any private vma.
1140 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1141 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
1143 struct page
*swapcache
;
1144 struct mem_cgroup
*memcg
;
1150 page
= ksm_might_need_to_copy(page
, vma
, addr
);
1151 if (unlikely(!page
))
1154 if (mem_cgroup_try_charge(page
, vma
->vm_mm
, GFP_KERNEL
, &memcg
)) {
1159 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
1160 if (unlikely(!maybe_same_pte(*pte
, swp_entry_to_pte(entry
)))) {
1161 mem_cgroup_cancel_charge(page
, memcg
);
1166 dec_mm_counter(vma
->vm_mm
, MM_SWAPENTS
);
1167 inc_mm_counter(vma
->vm_mm
, MM_ANONPAGES
);
1169 set_pte_at(vma
->vm_mm
, addr
, pte
,
1170 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
1171 if (page
== swapcache
) {
1172 page_add_anon_rmap(page
, vma
, addr
);
1173 mem_cgroup_commit_charge(page
, memcg
, true);
1174 } else { /* ksm created a completely new copy */
1175 page_add_new_anon_rmap(page
, vma
, addr
);
1176 mem_cgroup_commit_charge(page
, memcg
, false);
1177 lru_cache_add_active_or_unevictable(page
, vma
);
1181 * Move the page to the active list so it is not
1182 * immediately swapped out again after swapon.
1184 activate_page(page
);
1186 pte_unmap_unlock(pte
, ptl
);
1188 if (page
!= swapcache
) {
1195 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
1196 unsigned long addr
, unsigned long end
,
1197 swp_entry_t entry
, struct page
*page
)
1199 pte_t swp_pte
= swp_entry_to_pte(entry
);
1204 * We don't actually need pte lock while scanning for swp_pte: since
1205 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1206 * page table while we're scanning; though it could get zapped, and on
1207 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1208 * of unmatched parts which look like swp_pte, so unuse_pte must
1209 * recheck under pte lock. Scanning without pte lock lets it be
1210 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1212 pte
= pte_offset_map(pmd
, addr
);
1215 * swapoff spends a _lot_ of time in this loop!
1216 * Test inline before going to call unuse_pte.
1218 if (unlikely(maybe_same_pte(*pte
, swp_pte
))) {
1220 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
1223 pte
= pte_offset_map(pmd
, addr
);
1225 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1231 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
1232 unsigned long addr
, unsigned long end
,
1233 swp_entry_t entry
, struct page
*page
)
1239 pmd
= pmd_offset(pud
, addr
);
1241 next
= pmd_addr_end(addr
, end
);
1242 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1244 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
);
1247 } while (pmd
++, addr
= next
, addr
!= end
);
1251 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
1252 unsigned long addr
, unsigned long end
,
1253 swp_entry_t entry
, struct page
*page
)
1259 pud
= pud_offset(pgd
, addr
);
1261 next
= pud_addr_end(addr
, end
);
1262 if (pud_none_or_clear_bad(pud
))
1264 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
);
1267 } while (pud
++, addr
= next
, addr
!= end
);
1271 static int unuse_vma(struct vm_area_struct
*vma
,
1272 swp_entry_t entry
, struct page
*page
)
1275 unsigned long addr
, end
, next
;
1278 if (page_anon_vma(page
)) {
1279 addr
= page_address_in_vma(page
, vma
);
1280 if (addr
== -EFAULT
)
1283 end
= addr
+ PAGE_SIZE
;
1285 addr
= vma
->vm_start
;
1289 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1291 next
= pgd_addr_end(addr
, end
);
1292 if (pgd_none_or_clear_bad(pgd
))
1294 ret
= unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
);
1297 } while (pgd
++, addr
= next
, addr
!= end
);
1301 static int unuse_mm(struct mm_struct
*mm
,
1302 swp_entry_t entry
, struct page
*page
)
1304 struct vm_area_struct
*vma
;
1307 if (!down_read_trylock(&mm
->mmap_sem
)) {
1309 * Activate page so shrink_inactive_list is unlikely to unmap
1310 * its ptes while lock is dropped, so swapoff can make progress.
1312 activate_page(page
);
1314 down_read(&mm
->mmap_sem
);
1317 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1318 if (vma
->anon_vma
&& (ret
= unuse_vma(vma
, entry
, page
)))
1321 up_read(&mm
->mmap_sem
);
1322 return (ret
< 0)? ret
: 0;
1326 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1327 * from current position to next entry still in use.
1328 * Recycle to start on reaching the end, returning 0 when empty.
1330 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
1331 unsigned int prev
, bool frontswap
)
1333 unsigned int max
= si
->max
;
1334 unsigned int i
= prev
;
1335 unsigned char count
;
1338 * No need for swap_lock here: we're just looking
1339 * for whether an entry is in use, not modifying it; false
1340 * hits are okay, and sys_swapoff() has already prevented new
1341 * allocations from this area (while holding swap_lock).
1350 * No entries in use at top of swap_map,
1351 * loop back to start and recheck there.
1358 if (frontswap_test(si
, i
))
1363 count
= READ_ONCE(si
->swap_map
[i
]);
1364 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
1371 * We completely avoid races by reading each swap page in advance,
1372 * and then search for the process using it. All the necessary
1373 * page table adjustments can then be made atomically.
1375 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1376 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1378 int try_to_unuse(unsigned int type
, bool frontswap
,
1379 unsigned long pages_to_unuse
)
1381 struct swap_info_struct
*si
= swap_info
[type
];
1382 struct mm_struct
*start_mm
;
1383 volatile unsigned char *swap_map
; /* swap_map is accessed without
1384 * locking. Mark it as volatile
1385 * to prevent compiler doing
1388 unsigned char swcount
;
1395 * When searching mms for an entry, a good strategy is to
1396 * start at the first mm we freed the previous entry from
1397 * (though actually we don't notice whether we or coincidence
1398 * freed the entry). Initialize this start_mm with a hold.
1400 * A simpler strategy would be to start at the last mm we
1401 * freed the previous entry from; but that would take less
1402 * advantage of mmlist ordering, which clusters forked mms
1403 * together, child after parent. If we race with dup_mmap(), we
1404 * prefer to resolve parent before child, lest we miss entries
1405 * duplicated after we scanned child: using last mm would invert
1408 start_mm
= &init_mm
;
1409 atomic_inc(&init_mm
.mm_users
);
1412 * Keep on scanning until all entries have gone. Usually,
1413 * one pass through swap_map is enough, but not necessarily:
1414 * there are races when an instance of an entry might be missed.
1416 while ((i
= find_next_to_unuse(si
, i
, frontswap
)) != 0) {
1417 if (signal_pending(current
)) {
1423 * Get a page for the entry, using the existing swap
1424 * cache page if there is one. Otherwise, get a clean
1425 * page and read the swap into it.
1427 swap_map
= &si
->swap_map
[i
];
1428 entry
= swp_entry(type
, i
);
1429 page
= read_swap_cache_async(entry
,
1430 GFP_HIGHUSER_MOVABLE
, NULL
, 0);
1433 * Either swap_duplicate() failed because entry
1434 * has been freed independently, and will not be
1435 * reused since sys_swapoff() already disabled
1436 * allocation from here, or alloc_page() failed.
1438 swcount
= *swap_map
;
1440 * We don't hold lock here, so the swap entry could be
1441 * SWAP_MAP_BAD (when the cluster is discarding).
1442 * Instead of fail out, We can just skip the swap
1443 * entry because swapoff will wait for discarding
1446 if (!swcount
|| swcount
== SWAP_MAP_BAD
)
1453 * Don't hold on to start_mm if it looks like exiting.
1455 if (atomic_read(&start_mm
->mm_users
) == 1) {
1457 start_mm
= &init_mm
;
1458 atomic_inc(&init_mm
.mm_users
);
1462 * Wait for and lock page. When do_swap_page races with
1463 * try_to_unuse, do_swap_page can handle the fault much
1464 * faster than try_to_unuse can locate the entry. This
1465 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1466 * defer to do_swap_page in such a case - in some tests,
1467 * do_swap_page and try_to_unuse repeatedly compete.
1469 wait_on_page_locked(page
);
1470 wait_on_page_writeback(page
);
1472 wait_on_page_writeback(page
);
1475 * Remove all references to entry.
1477 swcount
= *swap_map
;
1478 if (swap_count(swcount
) == SWAP_MAP_SHMEM
) {
1479 retval
= shmem_unuse(entry
, page
);
1480 /* page has already been unlocked and released */
1485 if (swap_count(swcount
) && start_mm
!= &init_mm
)
1486 retval
= unuse_mm(start_mm
, entry
, page
);
1488 if (swap_count(*swap_map
)) {
1489 int set_start_mm
= (*swap_map
>= swcount
);
1490 struct list_head
*p
= &start_mm
->mmlist
;
1491 struct mm_struct
*new_start_mm
= start_mm
;
1492 struct mm_struct
*prev_mm
= start_mm
;
1493 struct mm_struct
*mm
;
1495 atomic_inc(&new_start_mm
->mm_users
);
1496 atomic_inc(&prev_mm
->mm_users
);
1497 spin_lock(&mmlist_lock
);
1498 while (swap_count(*swap_map
) && !retval
&&
1499 (p
= p
->next
) != &start_mm
->mmlist
) {
1500 mm
= list_entry(p
, struct mm_struct
, mmlist
);
1501 if (!atomic_inc_not_zero(&mm
->mm_users
))
1503 spin_unlock(&mmlist_lock
);
1509 swcount
= *swap_map
;
1510 if (!swap_count(swcount
)) /* any usage ? */
1512 else if (mm
== &init_mm
)
1515 retval
= unuse_mm(mm
, entry
, page
);
1517 if (set_start_mm
&& *swap_map
< swcount
) {
1518 mmput(new_start_mm
);
1519 atomic_inc(&mm
->mm_users
);
1523 spin_lock(&mmlist_lock
);
1525 spin_unlock(&mmlist_lock
);
1528 start_mm
= new_start_mm
;
1532 page_cache_release(page
);
1537 * If a reference remains (rare), we would like to leave
1538 * the page in the swap cache; but try_to_unmap could
1539 * then re-duplicate the entry once we drop page lock,
1540 * so we might loop indefinitely; also, that page could
1541 * not be swapped out to other storage meanwhile. So:
1542 * delete from cache even if there's another reference,
1543 * after ensuring that the data has been saved to disk -
1544 * since if the reference remains (rarer), it will be
1545 * read from disk into another page. Splitting into two
1546 * pages would be incorrect if swap supported "shared
1547 * private" pages, but they are handled by tmpfs files.
1549 * Given how unuse_vma() targets one particular offset
1550 * in an anon_vma, once the anon_vma has been determined,
1551 * this splitting happens to be just what is needed to
1552 * handle where KSM pages have been swapped out: re-reading
1553 * is unnecessarily slow, but we can fix that later on.
1555 if (swap_count(*swap_map
) &&
1556 PageDirty(page
) && PageSwapCache(page
)) {
1557 struct writeback_control wbc
= {
1558 .sync_mode
= WB_SYNC_NONE
,
1561 swap_writepage(page
, &wbc
);
1563 wait_on_page_writeback(page
);
1567 * It is conceivable that a racing task removed this page from
1568 * swap cache just before we acquired the page lock at the top,
1569 * or while we dropped it in unuse_mm(). The page might even
1570 * be back in swap cache on another swap area: that we must not
1571 * delete, since it may not have been written out to swap yet.
1573 if (PageSwapCache(page
) &&
1574 likely(page_private(page
) == entry
.val
))
1575 delete_from_swap_cache(page
);
1578 * So we could skip searching mms once swap count went
1579 * to 1, we did not mark any present ptes as dirty: must
1580 * mark page dirty so shrink_page_list will preserve it.
1584 page_cache_release(page
);
1587 * Make sure that we aren't completely killing
1588 * interactive performance.
1591 if (frontswap
&& pages_to_unuse
> 0) {
1592 if (!--pages_to_unuse
)
1602 * After a successful try_to_unuse, if no swap is now in use, we know
1603 * we can empty the mmlist. swap_lock must be held on entry and exit.
1604 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1605 * added to the mmlist just after page_duplicate - before would be racy.
1607 static void drain_mmlist(void)
1609 struct list_head
*p
, *next
;
1612 for (type
= 0; type
< nr_swapfiles
; type
++)
1613 if (swap_info
[type
]->inuse_pages
)
1615 spin_lock(&mmlist_lock
);
1616 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
1618 spin_unlock(&mmlist_lock
);
1622 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1623 * corresponds to page offset for the specified swap entry.
1624 * Note that the type of this function is sector_t, but it returns page offset
1625 * into the bdev, not sector offset.
1627 static sector_t
map_swap_entry(swp_entry_t entry
, struct block_device
**bdev
)
1629 struct swap_info_struct
*sis
;
1630 struct swap_extent
*start_se
;
1631 struct swap_extent
*se
;
1634 sis
= swap_info
[swp_type(entry
)];
1637 offset
= swp_offset(entry
);
1638 start_se
= sis
->curr_swap_extent
;
1642 struct list_head
*lh
;
1644 if (se
->start_page
<= offset
&&
1645 offset
< (se
->start_page
+ se
->nr_pages
)) {
1646 return se
->start_block
+ (offset
- se
->start_page
);
1649 se
= list_entry(lh
, struct swap_extent
, list
);
1650 sis
->curr_swap_extent
= se
;
1651 BUG_ON(se
== start_se
); /* It *must* be present */
1656 * Returns the page offset into bdev for the specified page's swap entry.
1658 sector_t
map_swap_page(struct page
*page
, struct block_device
**bdev
)
1661 entry
.val
= page_private(page
);
1662 return map_swap_entry(entry
, bdev
);
1666 * Free all of a swapdev's extent information
1668 static void destroy_swap_extents(struct swap_info_struct
*sis
)
1670 while (!list_empty(&sis
->first_swap_extent
.list
)) {
1671 struct swap_extent
*se
;
1673 se
= list_entry(sis
->first_swap_extent
.list
.next
,
1674 struct swap_extent
, list
);
1675 list_del(&se
->list
);
1679 if (sis
->flags
& SWP_FILE
) {
1680 struct file
*swap_file
= sis
->swap_file
;
1681 struct address_space
*mapping
= swap_file
->f_mapping
;
1683 sis
->flags
&= ~SWP_FILE
;
1684 mapping
->a_ops
->swap_deactivate(swap_file
);
1689 * Add a block range (and the corresponding page range) into this swapdev's
1690 * extent list. The extent list is kept sorted in page order.
1692 * This function rather assumes that it is called in ascending page order.
1695 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
1696 unsigned long nr_pages
, sector_t start_block
)
1698 struct swap_extent
*se
;
1699 struct swap_extent
*new_se
;
1700 struct list_head
*lh
;
1702 if (start_page
== 0) {
1703 se
= &sis
->first_swap_extent
;
1704 sis
->curr_swap_extent
= se
;
1706 se
->nr_pages
= nr_pages
;
1707 se
->start_block
= start_block
;
1710 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
1711 se
= list_entry(lh
, struct swap_extent
, list
);
1712 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
1713 if (se
->start_block
+ se
->nr_pages
== start_block
) {
1715 se
->nr_pages
+= nr_pages
;
1721 * No merge. Insert a new extent, preserving ordering.
1723 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
1726 new_se
->start_page
= start_page
;
1727 new_se
->nr_pages
= nr_pages
;
1728 new_se
->start_block
= start_block
;
1730 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
1735 * A `swap extent' is a simple thing which maps a contiguous range of pages
1736 * onto a contiguous range of disk blocks. An ordered list of swap extents
1737 * is built at swapon time and is then used at swap_writepage/swap_readpage
1738 * time for locating where on disk a page belongs.
1740 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1741 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1742 * swap files identically.
1744 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1745 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1746 * swapfiles are handled *identically* after swapon time.
1748 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1749 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1750 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1751 * requirements, they are simply tossed out - we will never use those blocks
1754 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1755 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1756 * which will scribble on the fs.
1758 * The amount of disk space which a single swap extent represents varies.
1759 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1760 * extents in the list. To avoid much list walking, we cache the previous
1761 * search location in `curr_swap_extent', and start new searches from there.
1762 * This is extremely effective. The average number of iterations in
1763 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1765 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
1767 struct file
*swap_file
= sis
->swap_file
;
1768 struct address_space
*mapping
= swap_file
->f_mapping
;
1769 struct inode
*inode
= mapping
->host
;
1772 if (S_ISBLK(inode
->i_mode
)) {
1773 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1778 if (mapping
->a_ops
->swap_activate
) {
1779 ret
= mapping
->a_ops
->swap_activate(sis
, swap_file
, span
);
1781 sis
->flags
|= SWP_FILE
;
1782 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1788 return generic_swapfile_activate(sis
, swap_file
, span
);
1791 static void _enable_swap_info(struct swap_info_struct
*p
, int prio
,
1792 unsigned char *swap_map
,
1793 struct swap_cluster_info
*cluster_info
)
1798 p
->prio
= --least_priority
;
1800 * the plist prio is negated because plist ordering is
1801 * low-to-high, while swap ordering is high-to-low
1803 p
->list
.prio
= -p
->prio
;
1804 p
->avail_list
.prio
= -p
->prio
;
1805 p
->swap_map
= swap_map
;
1806 p
->cluster_info
= cluster_info
;
1807 p
->flags
|= SWP_WRITEOK
;
1808 atomic_long_add(p
->pages
, &nr_swap_pages
);
1809 total_swap_pages
+= p
->pages
;
1811 assert_spin_locked(&swap_lock
);
1813 * both lists are plists, and thus priority ordered.
1814 * swap_active_head needs to be priority ordered for swapoff(),
1815 * which on removal of any swap_info_struct with an auto-assigned
1816 * (i.e. negative) priority increments the auto-assigned priority
1817 * of any lower-priority swap_info_structs.
1818 * swap_avail_head needs to be priority ordered for get_swap_page(),
1819 * which allocates swap pages from the highest available priority
1822 plist_add(&p
->list
, &swap_active_head
);
1823 spin_lock(&swap_avail_lock
);
1824 plist_add(&p
->avail_list
, &swap_avail_head
);
1825 spin_unlock(&swap_avail_lock
);
1828 static void enable_swap_info(struct swap_info_struct
*p
, int prio
,
1829 unsigned char *swap_map
,
1830 struct swap_cluster_info
*cluster_info
,
1831 unsigned long *frontswap_map
)
1833 frontswap_init(p
->type
, frontswap_map
);
1834 spin_lock(&swap_lock
);
1835 spin_lock(&p
->lock
);
1836 _enable_swap_info(p
, prio
, swap_map
, cluster_info
);
1837 spin_unlock(&p
->lock
);
1838 spin_unlock(&swap_lock
);
1841 static void reinsert_swap_info(struct swap_info_struct
*p
)
1843 spin_lock(&swap_lock
);
1844 spin_lock(&p
->lock
);
1845 _enable_swap_info(p
, p
->prio
, p
->swap_map
, p
->cluster_info
);
1846 spin_unlock(&p
->lock
);
1847 spin_unlock(&swap_lock
);
1850 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
1852 struct swap_info_struct
*p
= NULL
;
1853 unsigned char *swap_map
;
1854 struct swap_cluster_info
*cluster_info
;
1855 unsigned long *frontswap_map
;
1856 struct file
*swap_file
, *victim
;
1857 struct address_space
*mapping
;
1858 struct inode
*inode
;
1859 struct filename
*pathname
;
1861 unsigned int old_block_size
;
1863 if (!capable(CAP_SYS_ADMIN
))
1866 BUG_ON(!current
->mm
);
1868 pathname
= getname(specialfile
);
1869 if (IS_ERR(pathname
))
1870 return PTR_ERR(pathname
);
1872 victim
= file_open_name(pathname
, O_RDWR
|O_LARGEFILE
, 0);
1873 err
= PTR_ERR(victim
);
1877 mapping
= victim
->f_mapping
;
1878 spin_lock(&swap_lock
);
1879 plist_for_each_entry(p
, &swap_active_head
, list
) {
1880 if (p
->flags
& SWP_WRITEOK
) {
1881 if (p
->swap_file
->f_mapping
== mapping
) {
1889 spin_unlock(&swap_lock
);
1892 if (!security_vm_enough_memory_mm(current
->mm
, p
->pages
))
1893 vm_unacct_memory(p
->pages
);
1896 spin_unlock(&swap_lock
);
1899 spin_lock(&swap_avail_lock
);
1900 plist_del(&p
->avail_list
, &swap_avail_head
);
1901 spin_unlock(&swap_avail_lock
);
1902 spin_lock(&p
->lock
);
1904 struct swap_info_struct
*si
= p
;
1906 plist_for_each_entry_continue(si
, &swap_active_head
, list
) {
1909 si
->avail_list
.prio
--;
1913 plist_del(&p
->list
, &swap_active_head
);
1914 atomic_long_sub(p
->pages
, &nr_swap_pages
);
1915 total_swap_pages
-= p
->pages
;
1916 p
->flags
&= ~SWP_WRITEOK
;
1917 spin_unlock(&p
->lock
);
1918 spin_unlock(&swap_lock
);
1920 set_current_oom_origin();
1921 err
= try_to_unuse(p
->type
, false, 0); /* force unuse all pages */
1922 clear_current_oom_origin();
1925 /* re-insert swap space back into swap_list */
1926 reinsert_swap_info(p
);
1930 flush_work(&p
->discard_work
);
1932 destroy_swap_extents(p
);
1933 if (p
->flags
& SWP_CONTINUED
)
1934 free_swap_count_continuations(p
);
1936 mutex_lock(&swapon_mutex
);
1937 spin_lock(&swap_lock
);
1938 spin_lock(&p
->lock
);
1941 /* wait for anyone still in scan_swap_map */
1942 p
->highest_bit
= 0; /* cuts scans short */
1943 while (p
->flags
>= SWP_SCANNING
) {
1944 spin_unlock(&p
->lock
);
1945 spin_unlock(&swap_lock
);
1946 schedule_timeout_uninterruptible(1);
1947 spin_lock(&swap_lock
);
1948 spin_lock(&p
->lock
);
1951 swap_file
= p
->swap_file
;
1952 old_block_size
= p
->old_block_size
;
1953 p
->swap_file
= NULL
;
1955 swap_map
= p
->swap_map
;
1957 cluster_info
= p
->cluster_info
;
1958 p
->cluster_info
= NULL
;
1959 frontswap_map
= frontswap_map_get(p
);
1960 spin_unlock(&p
->lock
);
1961 spin_unlock(&swap_lock
);
1962 frontswap_invalidate_area(p
->type
);
1963 frontswap_map_set(p
, NULL
);
1964 mutex_unlock(&swapon_mutex
);
1965 free_percpu(p
->percpu_cluster
);
1966 p
->percpu_cluster
= NULL
;
1968 vfree(cluster_info
);
1969 vfree(frontswap_map
);
1970 /* Destroy swap account information */
1971 swap_cgroup_swapoff(p
->type
);
1973 inode
= mapping
->host
;
1974 if (S_ISBLK(inode
->i_mode
)) {
1975 struct block_device
*bdev
= I_BDEV(inode
);
1976 set_blocksize(bdev
, old_block_size
);
1977 blkdev_put(bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
1979 mutex_lock(&inode
->i_mutex
);
1980 inode
->i_flags
&= ~S_SWAPFILE
;
1981 mutex_unlock(&inode
->i_mutex
);
1983 filp_close(swap_file
, NULL
);
1986 * Clear the SWP_USED flag after all resources are freed so that swapon
1987 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1988 * not hold p->lock after we cleared its SWP_WRITEOK.
1990 spin_lock(&swap_lock
);
1992 spin_unlock(&swap_lock
);
1995 atomic_inc(&proc_poll_event
);
1996 wake_up_interruptible(&proc_poll_wait
);
1999 filp_close(victim
, NULL
);
2005 #ifdef CONFIG_PROC_FS
2006 static unsigned swaps_poll(struct file
*file
, poll_table
*wait
)
2008 struct seq_file
*seq
= file
->private_data
;
2010 poll_wait(file
, &proc_poll_wait
, wait
);
2012 if (seq
->poll_event
!= atomic_read(&proc_poll_event
)) {
2013 seq
->poll_event
= atomic_read(&proc_poll_event
);
2014 return POLLIN
| POLLRDNORM
| POLLERR
| POLLPRI
;
2017 return POLLIN
| POLLRDNORM
;
2021 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
2023 struct swap_info_struct
*si
;
2027 mutex_lock(&swapon_mutex
);
2030 return SEQ_START_TOKEN
;
2032 for (type
= 0; type
< nr_swapfiles
; type
++) {
2033 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2034 si
= swap_info
[type
];
2035 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2044 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
2046 struct swap_info_struct
*si
= v
;
2049 if (v
== SEQ_START_TOKEN
)
2052 type
= si
->type
+ 1;
2054 for (; type
< nr_swapfiles
; type
++) {
2055 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2056 si
= swap_info
[type
];
2057 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
2066 static void swap_stop(struct seq_file
*swap
, void *v
)
2068 mutex_unlock(&swapon_mutex
);
2071 static int swap_show(struct seq_file
*swap
, void *v
)
2073 struct swap_info_struct
*si
= v
;
2077 if (si
== SEQ_START_TOKEN
) {
2078 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2082 file
= si
->swap_file
;
2083 len
= seq_file_path(swap
, file
, " \t\n\\");
2084 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
2085 len
< 40 ? 40 - len
: 1, " ",
2086 S_ISBLK(file_inode(file
)->i_mode
) ?
2087 "partition" : "file\t",
2088 si
->pages
<< (PAGE_SHIFT
- 10),
2089 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
2094 static const struct seq_operations swaps_op
= {
2095 .start
= swap_start
,
2101 static int swaps_open(struct inode
*inode
, struct file
*file
)
2103 struct seq_file
*seq
;
2106 ret
= seq_open(file
, &swaps_op
);
2110 seq
= file
->private_data
;
2111 seq
->poll_event
= atomic_read(&proc_poll_event
);
2115 static const struct file_operations proc_swaps_operations
= {
2118 .llseek
= seq_lseek
,
2119 .release
= seq_release
,
2123 static int __init
procswaps_init(void)
2125 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
2128 __initcall(procswaps_init
);
2129 #endif /* CONFIG_PROC_FS */
2131 #ifdef MAX_SWAPFILES_CHECK
2132 static int __init
max_swapfiles_check(void)
2134 MAX_SWAPFILES_CHECK();
2137 late_initcall(max_swapfiles_check
);
2140 static struct swap_info_struct
*alloc_swap_info(void)
2142 struct swap_info_struct
*p
;
2145 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
2147 return ERR_PTR(-ENOMEM
);
2149 spin_lock(&swap_lock
);
2150 for (type
= 0; type
< nr_swapfiles
; type
++) {
2151 if (!(swap_info
[type
]->flags
& SWP_USED
))
2154 if (type
>= MAX_SWAPFILES
) {
2155 spin_unlock(&swap_lock
);
2157 return ERR_PTR(-EPERM
);
2159 if (type
>= nr_swapfiles
) {
2161 swap_info
[type
] = p
;
2163 * Write swap_info[type] before nr_swapfiles, in case a
2164 * racing procfs swap_start() or swap_next() is reading them.
2165 * (We never shrink nr_swapfiles, we never free this entry.)
2171 p
= swap_info
[type
];
2173 * Do not memset this entry: a racing procfs swap_next()
2174 * would be relying on p->type to remain valid.
2177 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
2178 plist_node_init(&p
->list
, 0);
2179 plist_node_init(&p
->avail_list
, 0);
2180 p
->flags
= SWP_USED
;
2181 spin_unlock(&swap_lock
);
2182 spin_lock_init(&p
->lock
);
2187 static int claim_swapfile(struct swap_info_struct
*p
, struct inode
*inode
)
2191 if (S_ISBLK(inode
->i_mode
)) {
2192 p
->bdev
= bdgrab(I_BDEV(inode
));
2193 error
= blkdev_get(p
->bdev
,
2194 FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
, p
);
2199 p
->old_block_size
= block_size(p
->bdev
);
2200 error
= set_blocksize(p
->bdev
, PAGE_SIZE
);
2203 p
->flags
|= SWP_BLKDEV
;
2204 } else if (S_ISREG(inode
->i_mode
)) {
2205 p
->bdev
= inode
->i_sb
->s_bdev
;
2206 mutex_lock(&inode
->i_mutex
);
2207 if (IS_SWAPFILE(inode
))
2215 static unsigned long read_swap_header(struct swap_info_struct
*p
,
2216 union swap_header
*swap_header
,
2217 struct inode
*inode
)
2220 unsigned long maxpages
;
2221 unsigned long swapfilepages
;
2222 unsigned long last_page
;
2224 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
2225 pr_err("Unable to find swap-space signature\n");
2229 /* swap partition endianess hack... */
2230 if (swab32(swap_header
->info
.version
) == 1) {
2231 swab32s(&swap_header
->info
.version
);
2232 swab32s(&swap_header
->info
.last_page
);
2233 swab32s(&swap_header
->info
.nr_badpages
);
2234 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2236 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
2237 swab32s(&swap_header
->info
.badpages
[i
]);
2239 /* Check the swap header's sub-version */
2240 if (swap_header
->info
.version
!= 1) {
2241 pr_warn("Unable to handle swap header version %d\n",
2242 swap_header
->info
.version
);
2247 p
->cluster_next
= 1;
2251 * Find out how many pages are allowed for a single swap
2252 * device. There are two limiting factors: 1) the number
2253 * of bits for the swap offset in the swp_entry_t type, and
2254 * 2) the number of bits in the swap pte as defined by the
2255 * different architectures. In order to find the
2256 * largest possible bit mask, a swap entry with swap type 0
2257 * and swap offset ~0UL is created, encoded to a swap pte,
2258 * decoded to a swp_entry_t again, and finally the swap
2259 * offset is extracted. This will mask all the bits from
2260 * the initial ~0UL mask that can't be encoded in either
2261 * the swp_entry_t or the architecture definition of a
2264 maxpages
= swp_offset(pte_to_swp_entry(
2265 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2266 last_page
= swap_header
->info
.last_page
;
2267 if (last_page
> maxpages
) {
2268 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2269 maxpages
<< (PAGE_SHIFT
- 10),
2270 last_page
<< (PAGE_SHIFT
- 10));
2272 if (maxpages
> last_page
) {
2273 maxpages
= last_page
+ 1;
2274 /* p->max is an unsigned int: don't overflow it */
2275 if ((unsigned int)maxpages
== 0)
2276 maxpages
= UINT_MAX
;
2278 p
->highest_bit
= maxpages
- 1;
2282 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
2283 if (swapfilepages
&& maxpages
> swapfilepages
) {
2284 pr_warn("Swap area shorter than signature indicates\n");
2287 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
2289 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
2295 static int setup_swap_map_and_extents(struct swap_info_struct
*p
,
2296 union swap_header
*swap_header
,
2297 unsigned char *swap_map
,
2298 struct swap_cluster_info
*cluster_info
,
2299 unsigned long maxpages
,
2303 unsigned int nr_good_pages
;
2305 unsigned long nr_clusters
= DIV_ROUND_UP(maxpages
, SWAPFILE_CLUSTER
);
2306 unsigned long idx
= p
->cluster_next
/ SWAPFILE_CLUSTER
;
2308 nr_good_pages
= maxpages
- 1; /* omit header page */
2310 cluster_set_null(&p
->free_cluster_head
);
2311 cluster_set_null(&p
->free_cluster_tail
);
2312 cluster_set_null(&p
->discard_cluster_head
);
2313 cluster_set_null(&p
->discard_cluster_tail
);
2315 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
2316 unsigned int page_nr
= swap_header
->info
.badpages
[i
];
2317 if (page_nr
== 0 || page_nr
> swap_header
->info
.last_page
)
2319 if (page_nr
< maxpages
) {
2320 swap_map
[page_nr
] = SWAP_MAP_BAD
;
2323 * Haven't marked the cluster free yet, no list
2324 * operation involved
2326 inc_cluster_info_page(p
, cluster_info
, page_nr
);
2330 /* Haven't marked the cluster free yet, no list operation involved */
2331 for (i
= maxpages
; i
< round_up(maxpages
, SWAPFILE_CLUSTER
); i
++)
2332 inc_cluster_info_page(p
, cluster_info
, i
);
2334 if (nr_good_pages
) {
2335 swap_map
[0] = SWAP_MAP_BAD
;
2337 * Not mark the cluster free yet, no list
2338 * operation involved
2340 inc_cluster_info_page(p
, cluster_info
, 0);
2342 p
->pages
= nr_good_pages
;
2343 nr_extents
= setup_swap_extents(p
, span
);
2346 nr_good_pages
= p
->pages
;
2348 if (!nr_good_pages
) {
2349 pr_warn("Empty swap-file\n");
2356 for (i
= 0; i
< nr_clusters
; i
++) {
2357 if (!cluster_count(&cluster_info
[idx
])) {
2358 cluster_set_flag(&cluster_info
[idx
], CLUSTER_FLAG_FREE
);
2359 if (cluster_is_null(&p
->free_cluster_head
)) {
2360 cluster_set_next_flag(&p
->free_cluster_head
,
2362 cluster_set_next_flag(&p
->free_cluster_tail
,
2367 tail
= cluster_next(&p
->free_cluster_tail
);
2368 cluster_set_next(&cluster_info
[tail
], idx
);
2369 cluster_set_next_flag(&p
->free_cluster_tail
,
2374 if (idx
== nr_clusters
)
2381 * Helper to sys_swapon determining if a given swap
2382 * backing device queue supports DISCARD operations.
2384 static bool swap_discardable(struct swap_info_struct
*si
)
2386 struct request_queue
*q
= bdev_get_queue(si
->bdev
);
2388 if (!q
|| !blk_queue_discard(q
))
2394 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
2396 struct swap_info_struct
*p
;
2397 struct filename
*name
;
2398 struct file
*swap_file
= NULL
;
2399 struct address_space
*mapping
;
2402 union swap_header
*swap_header
;
2405 unsigned long maxpages
;
2406 unsigned char *swap_map
= NULL
;
2407 struct swap_cluster_info
*cluster_info
= NULL
;
2408 unsigned long *frontswap_map
= NULL
;
2409 struct page
*page
= NULL
;
2410 struct inode
*inode
= NULL
;
2412 if (swap_flags
& ~SWAP_FLAGS_VALID
)
2415 if (!capable(CAP_SYS_ADMIN
))
2418 p
= alloc_swap_info();
2422 INIT_WORK(&p
->discard_work
, swap_discard_work
);
2424 name
= getname(specialfile
);
2426 error
= PTR_ERR(name
);
2430 swap_file
= file_open_name(name
, O_RDWR
|O_LARGEFILE
, 0);
2431 if (IS_ERR(swap_file
)) {
2432 error
= PTR_ERR(swap_file
);
2437 p
->swap_file
= swap_file
;
2438 mapping
= swap_file
->f_mapping
;
2439 inode
= mapping
->host
;
2441 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
2442 error
= claim_swapfile(p
, inode
);
2443 if (unlikely(error
))
2447 * Read the swap header.
2449 if (!mapping
->a_ops
->readpage
) {
2453 page
= read_mapping_page(mapping
, 0, swap_file
);
2455 error
= PTR_ERR(page
);
2458 swap_header
= kmap(page
);
2460 maxpages
= read_swap_header(p
, swap_header
, inode
);
2461 if (unlikely(!maxpages
)) {
2466 /* OK, set up the swap map and apply the bad block list */
2467 swap_map
= vzalloc(maxpages
);
2472 if (p
->bdev
&& blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
2475 p
->flags
|= SWP_SOLIDSTATE
;
2477 * select a random position to start with to help wear leveling
2480 p
->cluster_next
= 1 + (prandom_u32() % p
->highest_bit
);
2482 cluster_info
= vzalloc(DIV_ROUND_UP(maxpages
,
2483 SWAPFILE_CLUSTER
) * sizeof(*cluster_info
));
2484 if (!cluster_info
) {
2488 p
->percpu_cluster
= alloc_percpu(struct percpu_cluster
);
2489 if (!p
->percpu_cluster
) {
2493 for_each_possible_cpu(cpu
) {
2494 struct percpu_cluster
*cluster
;
2495 cluster
= per_cpu_ptr(p
->percpu_cluster
, cpu
);
2496 cluster_set_null(&cluster
->index
);
2500 error
= swap_cgroup_swapon(p
->type
, maxpages
);
2504 nr_extents
= setup_swap_map_and_extents(p
, swap_header
, swap_map
,
2505 cluster_info
, maxpages
, &span
);
2506 if (unlikely(nr_extents
< 0)) {
2510 /* frontswap enabled? set up bit-per-page map for frontswap */
2511 if (frontswap_enabled
)
2512 frontswap_map
= vzalloc(BITS_TO_LONGS(maxpages
) * sizeof(long));
2514 if (p
->bdev
&&(swap_flags
& SWAP_FLAG_DISCARD
) && swap_discardable(p
)) {
2516 * When discard is enabled for swap with no particular
2517 * policy flagged, we set all swap discard flags here in
2518 * order to sustain backward compatibility with older
2519 * swapon(8) releases.
2521 p
->flags
|= (SWP_DISCARDABLE
| SWP_AREA_DISCARD
|
2525 * By flagging sys_swapon, a sysadmin can tell us to
2526 * either do single-time area discards only, or to just
2527 * perform discards for released swap page-clusters.
2528 * Now it's time to adjust the p->flags accordingly.
2530 if (swap_flags
& SWAP_FLAG_DISCARD_ONCE
)
2531 p
->flags
&= ~SWP_PAGE_DISCARD
;
2532 else if (swap_flags
& SWAP_FLAG_DISCARD_PAGES
)
2533 p
->flags
&= ~SWP_AREA_DISCARD
;
2535 /* issue a swapon-time discard if it's still required */
2536 if (p
->flags
& SWP_AREA_DISCARD
) {
2537 int err
= discard_swap(p
);
2539 pr_err("swapon: discard_swap(%p): %d\n",
2544 mutex_lock(&swapon_mutex
);
2546 if (swap_flags
& SWAP_FLAG_PREFER
)
2548 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
2549 enable_swap_info(p
, prio
, swap_map
, cluster_info
, frontswap_map
);
2551 pr_info("Adding %uk swap on %s. "
2552 "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2553 p
->pages
<<(PAGE_SHIFT
-10), name
->name
, p
->prio
,
2554 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
2555 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
2556 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "",
2557 (p
->flags
& SWP_AREA_DISCARD
) ? "s" : "",
2558 (p
->flags
& SWP_PAGE_DISCARD
) ? "c" : "",
2559 (frontswap_map
) ? "FS" : "");
2561 mutex_unlock(&swapon_mutex
);
2562 atomic_inc(&proc_poll_event
);
2563 wake_up_interruptible(&proc_poll_wait
);
2565 if (S_ISREG(inode
->i_mode
))
2566 inode
->i_flags
|= S_SWAPFILE
;
2570 free_percpu(p
->percpu_cluster
);
2571 p
->percpu_cluster
= NULL
;
2572 if (inode
&& S_ISBLK(inode
->i_mode
) && p
->bdev
) {
2573 set_blocksize(p
->bdev
, p
->old_block_size
);
2574 blkdev_put(p
->bdev
, FMODE_READ
| FMODE_WRITE
| FMODE_EXCL
);
2576 destroy_swap_extents(p
);
2577 swap_cgroup_swapoff(p
->type
);
2578 spin_lock(&swap_lock
);
2579 p
->swap_file
= NULL
;
2581 spin_unlock(&swap_lock
);
2583 vfree(cluster_info
);
2585 if (inode
&& S_ISREG(inode
->i_mode
)) {
2586 mutex_unlock(&inode
->i_mutex
);
2589 filp_close(swap_file
, NULL
);
2592 if (page
&& !IS_ERR(page
)) {
2594 page_cache_release(page
);
2598 if (inode
&& S_ISREG(inode
->i_mode
))
2599 mutex_unlock(&inode
->i_mutex
);
2603 void si_swapinfo(struct sysinfo
*val
)
2606 unsigned long nr_to_be_unused
= 0;
2608 spin_lock(&swap_lock
);
2609 for (type
= 0; type
< nr_swapfiles
; type
++) {
2610 struct swap_info_struct
*si
= swap_info
[type
];
2612 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
2613 nr_to_be_unused
+= si
->inuse_pages
;
2615 val
->freeswap
= atomic_long_read(&nr_swap_pages
) + nr_to_be_unused
;
2616 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
2617 spin_unlock(&swap_lock
);
2621 * Verify that a swap entry is valid and increment its swap map count.
2623 * Returns error code in following case.
2625 * - swp_entry is invalid -> EINVAL
2626 * - swp_entry is migration entry -> EINVAL
2627 * - swap-cache reference is requested but there is already one. -> EEXIST
2628 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2629 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2631 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
2633 struct swap_info_struct
*p
;
2634 unsigned long offset
, type
;
2635 unsigned char count
;
2636 unsigned char has_cache
;
2639 if (non_swap_entry(entry
))
2642 type
= swp_type(entry
);
2643 if (type
>= nr_swapfiles
)
2645 p
= swap_info
[type
];
2646 offset
= swp_offset(entry
);
2648 spin_lock(&p
->lock
);
2649 if (unlikely(offset
>= p
->max
))
2652 count
= p
->swap_map
[offset
];
2655 * swapin_readahead() doesn't check if a swap entry is valid, so the
2656 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2658 if (unlikely(swap_count(count
) == SWAP_MAP_BAD
)) {
2663 has_cache
= count
& SWAP_HAS_CACHE
;
2664 count
&= ~SWAP_HAS_CACHE
;
2667 if (usage
== SWAP_HAS_CACHE
) {
2669 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2670 if (!has_cache
&& count
)
2671 has_cache
= SWAP_HAS_CACHE
;
2672 else if (has_cache
) /* someone else added cache */
2674 else /* no users remaining */
2677 } else if (count
|| has_cache
) {
2679 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
2681 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
2683 else if (swap_count_continued(p
, offset
, count
))
2684 count
= COUNT_CONTINUED
;
2688 err
= -ENOENT
; /* unused swap entry */
2690 p
->swap_map
[offset
] = count
| has_cache
;
2693 spin_unlock(&p
->lock
);
2698 pr_err("swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
2703 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2704 * (in which case its reference count is never incremented).
2706 void swap_shmem_alloc(swp_entry_t entry
)
2708 __swap_duplicate(entry
, SWAP_MAP_SHMEM
);
2712 * Increase reference count of swap entry by 1.
2713 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2714 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2715 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2716 * might occur if a page table entry has got corrupted.
2718 int swap_duplicate(swp_entry_t entry
)
2722 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
2723 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
2728 * @entry: swap entry for which we allocate swap cache.
2730 * Called when allocating swap cache for existing swap entry,
2731 * This can return error codes. Returns 0 at success.
2732 * -EBUSY means there is a swap cache.
2733 * Note: return code is different from swap_duplicate().
2735 int swapcache_prepare(swp_entry_t entry
)
2737 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
2740 struct swap_info_struct
*page_swap_info(struct page
*page
)
2742 swp_entry_t swap
= { .val
= page_private(page
) };
2743 BUG_ON(!PageSwapCache(page
));
2744 return swap_info
[swp_type(swap
)];
2748 * out-of-line __page_file_ methods to avoid include hell.
2750 struct address_space
*__page_file_mapping(struct page
*page
)
2752 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
2753 return page_swap_info(page
)->swap_file
->f_mapping
;
2755 EXPORT_SYMBOL_GPL(__page_file_mapping
);
2757 pgoff_t
__page_file_index(struct page
*page
)
2759 swp_entry_t swap
= { .val
= page_private(page
) };
2760 VM_BUG_ON_PAGE(!PageSwapCache(page
), page
);
2761 return swp_offset(swap
);
2763 EXPORT_SYMBOL_GPL(__page_file_index
);
2766 * add_swap_count_continuation - called when a swap count is duplicated
2767 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2768 * page of the original vmalloc'ed swap_map, to hold the continuation count
2769 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2770 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2772 * These continuation pages are seldom referenced: the common paths all work
2773 * on the original swap_map, only referring to a continuation page when the
2774 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2776 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2777 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2778 * can be called after dropping locks.
2780 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
2782 struct swap_info_struct
*si
;
2785 struct page
*list_page
;
2787 unsigned char count
;
2790 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2791 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2793 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
2795 si
= swap_info_get(entry
);
2798 * An acceptable race has occurred since the failing
2799 * __swap_duplicate(): the swap entry has been freed,
2800 * perhaps even the whole swap_map cleared for swapoff.
2805 offset
= swp_offset(entry
);
2806 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
2808 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
2810 * The higher the swap count, the more likely it is that tasks
2811 * will race to add swap count continuation: we need to avoid
2812 * over-provisioning.
2818 spin_unlock(&si
->lock
);
2823 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2824 * no architecture is using highmem pages for kernel page tables: so it
2825 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
2827 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2828 offset
&= ~PAGE_MASK
;
2831 * Page allocation does not initialize the page's lru field,
2832 * but it does always reset its private field.
2834 if (!page_private(head
)) {
2835 BUG_ON(count
& COUNT_CONTINUED
);
2836 INIT_LIST_HEAD(&head
->lru
);
2837 set_page_private(head
, SWP_CONTINUED
);
2838 si
->flags
|= SWP_CONTINUED
;
2841 list_for_each_entry(list_page
, &head
->lru
, lru
) {
2845 * If the previous map said no continuation, but we've found
2846 * a continuation page, free our allocation and use this one.
2848 if (!(count
& COUNT_CONTINUED
))
2851 map
= kmap_atomic(list_page
) + offset
;
2856 * If this continuation count now has some space in it,
2857 * free our allocation and use this one.
2859 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
2863 list_add_tail(&page
->lru
, &head
->lru
);
2864 page
= NULL
; /* now it's attached, don't free it */
2866 spin_unlock(&si
->lock
);
2874 * swap_count_continued - when the original swap_map count is incremented
2875 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2876 * into, carry if so, or else fail until a new continuation page is allocated;
2877 * when the original swap_map count is decremented from 0 with continuation,
2878 * borrow from the continuation and report whether it still holds more.
2879 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2881 static bool swap_count_continued(struct swap_info_struct
*si
,
2882 pgoff_t offset
, unsigned char count
)
2888 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2889 if (page_private(head
) != SWP_CONTINUED
) {
2890 BUG_ON(count
& COUNT_CONTINUED
);
2891 return false; /* need to add count continuation */
2894 offset
&= ~PAGE_MASK
;
2895 page
= list_entry(head
->lru
.next
, struct page
, lru
);
2896 map
= kmap_atomic(page
) + offset
;
2898 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
2899 goto init_map
; /* jump over SWAP_CONT_MAX checks */
2901 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
2903 * Think of how you add 1 to 999
2905 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
2907 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2908 BUG_ON(page
== head
);
2909 map
= kmap_atomic(page
) + offset
;
2911 if (*map
== SWAP_CONT_MAX
) {
2913 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2915 return false; /* add count continuation */
2916 map
= kmap_atomic(page
) + offset
;
2917 init_map
: *map
= 0; /* we didn't zero the page */
2921 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2922 while (page
!= head
) {
2923 map
= kmap_atomic(page
) + offset
;
2924 *map
= COUNT_CONTINUED
;
2926 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2928 return true; /* incremented */
2930 } else { /* decrementing */
2932 * Think of how you subtract 1 from 1000
2934 BUG_ON(count
!= COUNT_CONTINUED
);
2935 while (*map
== COUNT_CONTINUED
) {
2937 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2938 BUG_ON(page
== head
);
2939 map
= kmap_atomic(page
) + offset
;
2946 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2947 while (page
!= head
) {
2948 map
= kmap_atomic(page
) + offset
;
2949 *map
= SWAP_CONT_MAX
| count
;
2950 count
= COUNT_CONTINUED
;
2952 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2954 return count
== COUNT_CONTINUED
;
2959 * free_swap_count_continuations - swapoff free all the continuation pages
2960 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2962 static void free_swap_count_continuations(struct swap_info_struct
*si
)
2966 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
2968 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2969 if (page_private(head
)) {
2970 struct list_head
*this, *next
;
2971 list_for_each_safe(this, next
, &head
->lru
) {
2973 page
= list_entry(this, struct page
, lru
);