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/shm.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/module.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>
33 #include <asm/pgtable.h>
34 #include <asm/tlbflush.h>
35 #include <linux/swapops.h>
36 #include <linux/page_cgroup.h>
38 static bool swap_count_continued(struct swap_info_struct
*, pgoff_t
,
40 static void free_swap_count_continuations(struct swap_info_struct
*);
42 static DEFINE_SPINLOCK(swap_lock
);
43 static unsigned int nr_swapfiles
;
45 long total_swap_pages
;
46 static int least_priority
;
48 static const char Bad_file
[] = "Bad swap file entry ";
49 static const char Unused_file
[] = "Unused swap file entry ";
50 static const char Bad_offset
[] = "Bad swap offset entry ";
51 static const char Unused_offset
[] = "Unused swap offset entry ";
53 static struct swap_list_t swap_list
= {-1, -1};
55 static struct swap_info_struct
*swap_info
[MAX_SWAPFILES
];
57 static DEFINE_MUTEX(swapon_mutex
);
59 static inline unsigned char swap_count(unsigned char ent
)
61 return ent
& ~SWAP_HAS_CACHE
; /* may include SWAP_HAS_CONT flag */
64 /* returns 1 if swap entry is freed */
66 __try_to_reclaim_swap(struct swap_info_struct
*si
, unsigned long offset
)
68 swp_entry_t entry
= swp_entry(si
->type
, offset
);
72 page
= find_get_page(&swapper_space
, entry
.val
);
76 * This function is called from scan_swap_map() and it's called
77 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
78 * We have to use trylock for avoiding deadlock. This is a special
79 * case and you should use try_to_free_swap() with explicit lock_page()
80 * in usual operations.
82 if (trylock_page(page
)) {
83 ret
= try_to_free_swap(page
);
86 page_cache_release(page
);
91 * We need this because the bdev->unplug_fn can sleep and we cannot
92 * hold swap_lock while calling the unplug_fn. And swap_lock
93 * cannot be turned into a mutex.
95 static DECLARE_RWSEM(swap_unplug_sem
);
97 void swap_unplug_io_fn(struct backing_dev_info
*unused_bdi
, struct page
*page
)
101 down_read(&swap_unplug_sem
);
102 entry
.val
= page_private(page
);
103 if (PageSwapCache(page
)) {
104 struct block_device
*bdev
= swap_info
[swp_type(entry
)]->bdev
;
105 struct backing_dev_info
*bdi
;
108 * If the page is removed from swapcache from under us (with a
109 * racy try_to_unuse/swapoff) we need an additional reference
110 * count to avoid reading garbage from page_private(page) above.
111 * If the WARN_ON triggers during a swapoff it maybe the race
112 * condition and it's harmless. However if it triggers without
113 * swapoff it signals a problem.
115 WARN_ON(page_count(page
) <= 1);
117 bdi
= bdev
->bd_inode
->i_mapping
->backing_dev_info
;
118 blk_run_backing_dev(bdi
, page
);
120 up_read(&swap_unplug_sem
);
124 * swapon tell device that all the old swap contents can be discarded,
125 * to allow the swap device to optimize its wear-levelling.
127 static int discard_swap(struct swap_info_struct
*si
)
129 struct swap_extent
*se
;
130 sector_t start_block
;
134 /* Do not discard the swap header page! */
135 se
= &si
->first_swap_extent
;
136 start_block
= (se
->start_block
+ 1) << (PAGE_SHIFT
- 9);
137 nr_blocks
= ((sector_t
)se
->nr_pages
- 1) << (PAGE_SHIFT
- 9);
139 err
= blkdev_issue_discard(si
->bdev
, start_block
,
140 nr_blocks
, GFP_KERNEL
, DISCARD_FL_BARRIER
);
146 list_for_each_entry(se
, &si
->first_swap_extent
.list
, list
) {
147 start_block
= se
->start_block
<< (PAGE_SHIFT
- 9);
148 nr_blocks
= (sector_t
)se
->nr_pages
<< (PAGE_SHIFT
- 9);
150 err
= blkdev_issue_discard(si
->bdev
, start_block
,
151 nr_blocks
, GFP_KERNEL
, DISCARD_FL_BARRIER
);
157 return err
; /* That will often be -EOPNOTSUPP */
161 * swap allocation tell device that a cluster of swap can now be discarded,
162 * to allow the swap device to optimize its wear-levelling.
164 static void discard_swap_cluster(struct swap_info_struct
*si
,
165 pgoff_t start_page
, pgoff_t nr_pages
)
167 struct swap_extent
*se
= si
->curr_swap_extent
;
168 int found_extent
= 0;
171 struct list_head
*lh
;
173 if (se
->start_page
<= start_page
&&
174 start_page
< se
->start_page
+ se
->nr_pages
) {
175 pgoff_t offset
= start_page
- se
->start_page
;
176 sector_t start_block
= se
->start_block
+ offset
;
177 sector_t nr_blocks
= se
->nr_pages
- offset
;
179 if (nr_blocks
> nr_pages
)
180 nr_blocks
= nr_pages
;
181 start_page
+= nr_blocks
;
182 nr_pages
-= nr_blocks
;
185 si
->curr_swap_extent
= se
;
187 start_block
<<= PAGE_SHIFT
- 9;
188 nr_blocks
<<= PAGE_SHIFT
- 9;
189 if (blkdev_issue_discard(si
->bdev
, start_block
,
190 nr_blocks
, GFP_NOIO
, DISCARD_FL_BARRIER
))
195 se
= list_entry(lh
, struct swap_extent
, list
);
199 static int wait_for_discard(void *word
)
205 #define SWAPFILE_CLUSTER 256
206 #define LATENCY_LIMIT 256
208 static inline unsigned long scan_swap_map(struct swap_info_struct
*si
,
211 unsigned long offset
;
212 unsigned long scan_base
;
213 unsigned long last_in_cluster
= 0;
214 int latency_ration
= LATENCY_LIMIT
;
215 int found_free_cluster
= 0;
218 * We try to cluster swap pages by allocating them sequentially
219 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
220 * way, however, we resort to first-free allocation, starting
221 * a new cluster. This prevents us from scattering swap pages
222 * all over the entire swap partition, so that we reduce
223 * overall disk seek times between swap pages. -- sct
224 * But we do now try to find an empty cluster. -Andrea
225 * And we let swap pages go all over an SSD partition. Hugh
228 si
->flags
+= SWP_SCANNING
;
229 scan_base
= offset
= si
->cluster_next
;
231 if (unlikely(!si
->cluster_nr
--)) {
232 if (si
->pages
- si
->inuse_pages
< SWAPFILE_CLUSTER
) {
233 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
236 if (si
->flags
& SWP_DISCARDABLE
) {
238 * Start range check on racing allocations, in case
239 * they overlap the cluster we eventually decide on
240 * (we scan without swap_lock to allow preemption).
241 * It's hardly conceivable that cluster_nr could be
242 * wrapped during our scan, but don't depend on it.
244 if (si
->lowest_alloc
)
246 si
->lowest_alloc
= si
->max
;
247 si
->highest_alloc
= 0;
249 spin_unlock(&swap_lock
);
252 * If seek is expensive, start searching for new cluster from
253 * start of partition, to minimize the span of allocated swap.
254 * But if seek is cheap, search from our current position, so
255 * that swap is allocated from all over the partition: if the
256 * Flash Translation Layer only remaps within limited zones,
257 * we don't want to wear out the first zone too quickly.
259 if (!(si
->flags
& SWP_SOLIDSTATE
))
260 scan_base
= offset
= si
->lowest_bit
;
261 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
263 /* Locate the first empty (unaligned) cluster */
264 for (; last_in_cluster
<= si
->highest_bit
; offset
++) {
265 if (si
->swap_map
[offset
])
266 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
267 else if (offset
== last_in_cluster
) {
268 spin_lock(&swap_lock
);
269 offset
-= SWAPFILE_CLUSTER
- 1;
270 si
->cluster_next
= offset
;
271 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
272 found_free_cluster
= 1;
275 if (unlikely(--latency_ration
< 0)) {
277 latency_ration
= LATENCY_LIMIT
;
281 offset
= si
->lowest_bit
;
282 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
- 1;
284 /* Locate the first empty (unaligned) cluster */
285 for (; last_in_cluster
< scan_base
; offset
++) {
286 if (si
->swap_map
[offset
])
287 last_in_cluster
= offset
+ SWAPFILE_CLUSTER
;
288 else if (offset
== last_in_cluster
) {
289 spin_lock(&swap_lock
);
290 offset
-= SWAPFILE_CLUSTER
- 1;
291 si
->cluster_next
= offset
;
292 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
293 found_free_cluster
= 1;
296 if (unlikely(--latency_ration
< 0)) {
298 latency_ration
= LATENCY_LIMIT
;
303 spin_lock(&swap_lock
);
304 si
->cluster_nr
= SWAPFILE_CLUSTER
- 1;
305 si
->lowest_alloc
= 0;
309 if (!(si
->flags
& SWP_WRITEOK
))
311 if (!si
->highest_bit
)
313 if (offset
> si
->highest_bit
)
314 scan_base
= offset
= si
->lowest_bit
;
316 /* reuse swap entry of cache-only swap if not busy. */
317 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
319 spin_unlock(&swap_lock
);
320 swap_was_freed
= __try_to_reclaim_swap(si
, offset
);
321 spin_lock(&swap_lock
);
322 /* entry was freed successfully, try to use this again */
325 goto scan
; /* check next one */
328 if (si
->swap_map
[offset
])
331 if (offset
== si
->lowest_bit
)
333 if (offset
== si
->highest_bit
)
336 if (si
->inuse_pages
== si
->pages
) {
337 si
->lowest_bit
= si
->max
;
340 si
->swap_map
[offset
] = usage
;
341 si
->cluster_next
= offset
+ 1;
342 si
->flags
-= SWP_SCANNING
;
344 if (si
->lowest_alloc
) {
346 * Only set when SWP_DISCARDABLE, and there's a scan
347 * for a free cluster in progress or just completed.
349 if (found_free_cluster
) {
351 * To optimize wear-levelling, discard the
352 * old data of the cluster, taking care not to
353 * discard any of its pages that have already
354 * been allocated by racing tasks (offset has
355 * already stepped over any at the beginning).
357 if (offset
< si
->highest_alloc
&&
358 si
->lowest_alloc
<= last_in_cluster
)
359 last_in_cluster
= si
->lowest_alloc
- 1;
360 si
->flags
|= SWP_DISCARDING
;
361 spin_unlock(&swap_lock
);
363 if (offset
< last_in_cluster
)
364 discard_swap_cluster(si
, offset
,
365 last_in_cluster
- offset
+ 1);
367 spin_lock(&swap_lock
);
368 si
->lowest_alloc
= 0;
369 si
->flags
&= ~SWP_DISCARDING
;
371 smp_mb(); /* wake_up_bit advises this */
372 wake_up_bit(&si
->flags
, ilog2(SWP_DISCARDING
));
374 } else if (si
->flags
& SWP_DISCARDING
) {
376 * Delay using pages allocated by racing tasks
377 * until the whole discard has been issued. We
378 * could defer that delay until swap_writepage,
379 * but it's easier to keep this self-contained.
381 spin_unlock(&swap_lock
);
382 wait_on_bit(&si
->flags
, ilog2(SWP_DISCARDING
),
383 wait_for_discard
, TASK_UNINTERRUPTIBLE
);
384 spin_lock(&swap_lock
);
387 * Note pages allocated by racing tasks while
388 * scan for a free cluster is in progress, so
389 * that its final discard can exclude them.
391 if (offset
< si
->lowest_alloc
)
392 si
->lowest_alloc
= offset
;
393 if (offset
> si
->highest_alloc
)
394 si
->highest_alloc
= offset
;
400 spin_unlock(&swap_lock
);
401 while (++offset
<= si
->highest_bit
) {
402 if (!si
->swap_map
[offset
]) {
403 spin_lock(&swap_lock
);
406 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
407 spin_lock(&swap_lock
);
410 if (unlikely(--latency_ration
< 0)) {
412 latency_ration
= LATENCY_LIMIT
;
415 offset
= si
->lowest_bit
;
416 while (++offset
< scan_base
) {
417 if (!si
->swap_map
[offset
]) {
418 spin_lock(&swap_lock
);
421 if (vm_swap_full() && si
->swap_map
[offset
] == SWAP_HAS_CACHE
) {
422 spin_lock(&swap_lock
);
425 if (unlikely(--latency_ration
< 0)) {
427 latency_ration
= LATENCY_LIMIT
;
430 spin_lock(&swap_lock
);
433 si
->flags
-= SWP_SCANNING
;
437 swp_entry_t
get_swap_page(void)
439 struct swap_info_struct
*si
;
444 spin_lock(&swap_lock
);
445 if (nr_swap_pages
<= 0)
449 for (type
= swap_list
.next
; type
>= 0 && wrapped
< 2; type
= next
) {
450 si
= swap_info
[type
];
453 (!wrapped
&& si
->prio
!= swap_info
[next
]->prio
)) {
454 next
= swap_list
.head
;
458 if (!si
->highest_bit
)
460 if (!(si
->flags
& SWP_WRITEOK
))
463 swap_list
.next
= next
;
464 /* This is called for allocating swap entry for cache */
465 offset
= scan_swap_map(si
, SWAP_HAS_CACHE
);
467 spin_unlock(&swap_lock
);
468 return swp_entry(type
, offset
);
470 next
= swap_list
.next
;
475 spin_unlock(&swap_lock
);
476 return (swp_entry_t
) {0};
479 /* The only caller of this function is now susupend routine */
480 swp_entry_t
get_swap_page_of_type(int type
)
482 struct swap_info_struct
*si
;
485 spin_lock(&swap_lock
);
486 si
= swap_info
[type
];
487 if (si
&& (si
->flags
& SWP_WRITEOK
)) {
489 /* This is called for allocating swap entry, not cache */
490 offset
= scan_swap_map(si
, 1);
492 spin_unlock(&swap_lock
);
493 return swp_entry(type
, offset
);
497 spin_unlock(&swap_lock
);
498 return (swp_entry_t
) {0};
501 static struct swap_info_struct
*swap_info_get(swp_entry_t entry
)
503 struct swap_info_struct
*p
;
504 unsigned long offset
, type
;
508 type
= swp_type(entry
);
509 if (type
>= nr_swapfiles
)
512 if (!(p
->flags
& SWP_USED
))
514 offset
= swp_offset(entry
);
515 if (offset
>= p
->max
)
517 if (!p
->swap_map
[offset
])
519 spin_lock(&swap_lock
);
523 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_offset
, entry
.val
);
526 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_offset
, entry
.val
);
529 printk(KERN_ERR
"swap_free: %s%08lx\n", Unused_file
, entry
.val
);
532 printk(KERN_ERR
"swap_free: %s%08lx\n", Bad_file
, entry
.val
);
537 static unsigned char swap_entry_free(struct swap_info_struct
*p
,
538 swp_entry_t entry
, unsigned char usage
)
540 unsigned long offset
= swp_offset(entry
);
542 unsigned char has_cache
;
544 count
= p
->swap_map
[offset
];
545 has_cache
= count
& SWAP_HAS_CACHE
;
546 count
&= ~SWAP_HAS_CACHE
;
548 if (usage
== SWAP_HAS_CACHE
) {
549 VM_BUG_ON(!has_cache
);
551 } else if ((count
& ~COUNT_CONTINUED
) <= SWAP_MAP_MAX
) {
552 if (count
== COUNT_CONTINUED
) {
553 if (swap_count_continued(p
, offset
, count
))
554 count
= SWAP_MAP_MAX
| COUNT_CONTINUED
;
556 count
= SWAP_MAP_MAX
;
562 mem_cgroup_uncharge_swap(entry
);
564 usage
= count
| has_cache
;
565 p
->swap_map
[offset
] = usage
;
567 /* free if no reference */
569 if (offset
< p
->lowest_bit
)
570 p
->lowest_bit
= offset
;
571 if (offset
> p
->highest_bit
)
572 p
->highest_bit
= offset
;
573 if (swap_list
.next
>= 0 &&
574 p
->prio
> swap_info
[swap_list
.next
]->prio
)
575 swap_list
.next
= p
->type
;
584 * Caller has made sure that the swapdevice corresponding to entry
585 * is still around or has not been recycled.
587 void swap_free(swp_entry_t entry
)
589 struct swap_info_struct
*p
;
591 p
= swap_info_get(entry
);
593 swap_entry_free(p
, entry
, 1);
594 spin_unlock(&swap_lock
);
599 * Called after dropping swapcache to decrease refcnt to swap entries.
601 void swapcache_free(swp_entry_t entry
, struct page
*page
)
603 struct swap_info_struct
*p
;
606 p
= swap_info_get(entry
);
608 count
= swap_entry_free(p
, entry
, SWAP_HAS_CACHE
);
610 mem_cgroup_uncharge_swapcache(page
, entry
, count
!= 0);
611 spin_unlock(&swap_lock
);
616 * How many references to page are currently swapped out?
617 * This does not give an exact answer when swap count is continued,
618 * but does include the high COUNT_CONTINUED flag to allow for that.
620 static inline int page_swapcount(struct page
*page
)
623 struct swap_info_struct
*p
;
626 entry
.val
= page_private(page
);
627 p
= swap_info_get(entry
);
629 count
= swap_count(p
->swap_map
[swp_offset(entry
)]);
630 spin_unlock(&swap_lock
);
636 * We can write to an anon page without COW if there are no other references
637 * to it. And as a side-effect, free up its swap: because the old content
638 * on disk will never be read, and seeking back there to write new content
639 * later would only waste time away from clustering.
641 int reuse_swap_page(struct page
*page
)
645 VM_BUG_ON(!PageLocked(page
));
646 count
= page_mapcount(page
);
647 if (count
<= 1 && PageSwapCache(page
)) {
648 count
+= page_swapcount(page
);
649 if (count
== 1 && !PageWriteback(page
)) {
650 delete_from_swap_cache(page
);
658 * If swap is getting full, or if there are no more mappings of this page,
659 * then try_to_free_swap is called to free its swap space.
661 int try_to_free_swap(struct page
*page
)
663 VM_BUG_ON(!PageLocked(page
));
665 if (!PageSwapCache(page
))
667 if (PageWriteback(page
))
669 if (page_swapcount(page
))
672 delete_from_swap_cache(page
);
678 * Free the swap entry like above, but also try to
679 * free the page cache entry if it is the last user.
681 int free_swap_and_cache(swp_entry_t entry
)
683 struct swap_info_struct
*p
;
684 struct page
*page
= NULL
;
686 if (non_swap_entry(entry
))
689 p
= swap_info_get(entry
);
691 if (swap_entry_free(p
, entry
, 1) == SWAP_HAS_CACHE
) {
692 page
= find_get_page(&swapper_space
, entry
.val
);
693 if (page
&& !trylock_page(page
)) {
694 page_cache_release(page
);
698 spin_unlock(&swap_lock
);
702 * Not mapped elsewhere, or swap space full? Free it!
703 * Also recheck PageSwapCache now page is locked (above).
705 if (PageSwapCache(page
) && !PageWriteback(page
) &&
706 (!page_mapped(page
) || vm_swap_full())) {
707 delete_from_swap_cache(page
);
711 page_cache_release(page
);
716 #ifdef CONFIG_HIBERNATION
718 * Find the swap type that corresponds to given device (if any).
720 * @offset - number of the PAGE_SIZE-sized block of the device, starting
721 * from 0, in which the swap header is expected to be located.
723 * This is needed for the suspend to disk (aka swsusp).
725 int swap_type_of(dev_t device
, sector_t offset
, struct block_device
**bdev_p
)
727 struct block_device
*bdev
= NULL
;
731 bdev
= bdget(device
);
733 spin_lock(&swap_lock
);
734 for (type
= 0; type
< nr_swapfiles
; type
++) {
735 struct swap_info_struct
*sis
= swap_info
[type
];
737 if (!(sis
->flags
& SWP_WRITEOK
))
742 *bdev_p
= bdgrab(sis
->bdev
);
744 spin_unlock(&swap_lock
);
747 if (bdev
== sis
->bdev
) {
748 struct swap_extent
*se
= &sis
->first_swap_extent
;
750 if (se
->start_block
== offset
) {
752 *bdev_p
= bdgrab(sis
->bdev
);
754 spin_unlock(&swap_lock
);
760 spin_unlock(&swap_lock
);
768 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
769 * corresponding to given index in swap_info (swap type).
771 sector_t
swapdev_block(int type
, pgoff_t offset
)
773 struct block_device
*bdev
;
775 if ((unsigned int)type
>= nr_swapfiles
)
777 if (!(swap_info
[type
]->flags
& SWP_WRITEOK
))
779 return map_swap_page(swp_entry(type
, offset
), &bdev
);
783 * Return either the total number of swap pages of given type, or the number
784 * of free pages of that type (depending on @free)
786 * This is needed for software suspend
788 unsigned int count_swap_pages(int type
, int free
)
792 spin_lock(&swap_lock
);
793 if ((unsigned int)type
< nr_swapfiles
) {
794 struct swap_info_struct
*sis
= swap_info
[type
];
796 if (sis
->flags
& SWP_WRITEOK
) {
799 n
-= sis
->inuse_pages
;
802 spin_unlock(&swap_lock
);
805 #endif /* CONFIG_HIBERNATION */
808 * No need to decide whether this PTE shares the swap entry with others,
809 * just let do_wp_page work it out if a write is requested later - to
810 * force COW, vm_page_prot omits write permission from any private vma.
812 static int unuse_pte(struct vm_area_struct
*vma
, pmd_t
*pmd
,
813 unsigned long addr
, swp_entry_t entry
, struct page
*page
)
815 struct mem_cgroup
*ptr
= NULL
;
820 if (mem_cgroup_try_charge_swapin(vma
->vm_mm
, page
, GFP_KERNEL
, &ptr
)) {
825 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
826 if (unlikely(!pte_same(*pte
, swp_entry_to_pte(entry
)))) {
828 mem_cgroup_cancel_charge_swapin(ptr
);
833 inc_mm_counter(vma
->vm_mm
, anon_rss
);
835 set_pte_at(vma
->vm_mm
, addr
, pte
,
836 pte_mkold(mk_pte(page
, vma
->vm_page_prot
)));
837 page_add_anon_rmap(page
, vma
, addr
);
838 mem_cgroup_commit_charge_swapin(page
, ptr
);
841 * Move the page to the active list so it is not
842 * immediately swapped out again after swapon.
846 pte_unmap_unlock(pte
, ptl
);
851 static int unuse_pte_range(struct vm_area_struct
*vma
, pmd_t
*pmd
,
852 unsigned long addr
, unsigned long end
,
853 swp_entry_t entry
, struct page
*page
)
855 pte_t swp_pte
= swp_entry_to_pte(entry
);
860 * We don't actually need pte lock while scanning for swp_pte: since
861 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
862 * page table while we're scanning; though it could get zapped, and on
863 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
864 * of unmatched parts which look like swp_pte, so unuse_pte must
865 * recheck under pte lock. Scanning without pte lock lets it be
866 * preemptible whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
868 pte
= pte_offset_map(pmd
, addr
);
871 * swapoff spends a _lot_ of time in this loop!
872 * Test inline before going to call unuse_pte.
874 if (unlikely(pte_same(*pte
, swp_pte
))) {
876 ret
= unuse_pte(vma
, pmd
, addr
, entry
, page
);
879 pte
= pte_offset_map(pmd
, addr
);
881 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
887 static inline int unuse_pmd_range(struct vm_area_struct
*vma
, pud_t
*pud
,
888 unsigned long addr
, unsigned long end
,
889 swp_entry_t entry
, struct page
*page
)
895 pmd
= pmd_offset(pud
, addr
);
897 next
= pmd_addr_end(addr
, end
);
898 if (pmd_none_or_clear_bad(pmd
))
900 ret
= unuse_pte_range(vma
, pmd
, addr
, next
, entry
, page
);
903 } while (pmd
++, addr
= next
, addr
!= end
);
907 static inline int unuse_pud_range(struct vm_area_struct
*vma
, pgd_t
*pgd
,
908 unsigned long addr
, unsigned long end
,
909 swp_entry_t entry
, struct page
*page
)
915 pud
= pud_offset(pgd
, addr
);
917 next
= pud_addr_end(addr
, end
);
918 if (pud_none_or_clear_bad(pud
))
920 ret
= unuse_pmd_range(vma
, pud
, addr
, next
, entry
, page
);
923 } while (pud
++, addr
= next
, addr
!= end
);
927 static int unuse_vma(struct vm_area_struct
*vma
,
928 swp_entry_t entry
, struct page
*page
)
931 unsigned long addr
, end
, next
;
935 addr
= page_address_in_vma(page
, vma
);
939 end
= addr
+ PAGE_SIZE
;
941 addr
= vma
->vm_start
;
945 pgd
= pgd_offset(vma
->vm_mm
, addr
);
947 next
= pgd_addr_end(addr
, end
);
948 if (pgd_none_or_clear_bad(pgd
))
950 ret
= unuse_pud_range(vma
, pgd
, addr
, next
, entry
, page
);
953 } while (pgd
++, addr
= next
, addr
!= end
);
957 static int unuse_mm(struct mm_struct
*mm
,
958 swp_entry_t entry
, struct page
*page
)
960 struct vm_area_struct
*vma
;
963 if (!down_read_trylock(&mm
->mmap_sem
)) {
965 * Activate page so shrink_inactive_list is unlikely to unmap
966 * its ptes while lock is dropped, so swapoff can make progress.
970 down_read(&mm
->mmap_sem
);
973 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
974 if (vma
->anon_vma
&& (ret
= unuse_vma(vma
, entry
, page
)))
977 up_read(&mm
->mmap_sem
);
978 return (ret
< 0)? ret
: 0;
982 * Scan swap_map from current position to next entry still in use.
983 * Recycle to start on reaching the end, returning 0 when empty.
985 static unsigned int find_next_to_unuse(struct swap_info_struct
*si
,
988 unsigned int max
= si
->max
;
989 unsigned int i
= prev
;
993 * No need for swap_lock here: we're just looking
994 * for whether an entry is in use, not modifying it; false
995 * hits are okay, and sys_swapoff() has already prevented new
996 * allocations from this area (while holding swap_lock).
1005 * No entries in use at top of swap_map,
1006 * loop back to start and recheck there.
1012 count
= si
->swap_map
[i
];
1013 if (count
&& swap_count(count
) != SWAP_MAP_BAD
)
1020 * We completely avoid races by reading each swap page in advance,
1021 * and then search for the process using it. All the necessary
1022 * page table adjustments can then be made atomically.
1024 static int try_to_unuse(unsigned int type
)
1026 struct swap_info_struct
*si
= swap_info
[type
];
1027 struct mm_struct
*start_mm
;
1028 unsigned char *swap_map
;
1029 unsigned char swcount
;
1037 * When searching mms for an entry, a good strategy is to
1038 * start at the first mm we freed the previous entry from
1039 * (though actually we don't notice whether we or coincidence
1040 * freed the entry). Initialize this start_mm with a hold.
1042 * A simpler strategy would be to start at the last mm we
1043 * freed the previous entry from; but that would take less
1044 * advantage of mmlist ordering, which clusters forked mms
1045 * together, child after parent. If we race with dup_mmap(), we
1046 * prefer to resolve parent before child, lest we miss entries
1047 * duplicated after we scanned child: using last mm would invert
1050 start_mm
= &init_mm
;
1051 atomic_inc(&init_mm
.mm_users
);
1054 * Keep on scanning until all entries have gone. Usually,
1055 * one pass through swap_map is enough, but not necessarily:
1056 * there are races when an instance of an entry might be missed.
1058 while ((i
= find_next_to_unuse(si
, i
)) != 0) {
1059 if (signal_pending(current
)) {
1065 * Get a page for the entry, using the existing swap
1066 * cache page if there is one. Otherwise, get a clean
1067 * page and read the swap into it.
1069 swap_map
= &si
->swap_map
[i
];
1070 entry
= swp_entry(type
, i
);
1071 page
= read_swap_cache_async(entry
,
1072 GFP_HIGHUSER_MOVABLE
, NULL
, 0);
1075 * Either swap_duplicate() failed because entry
1076 * has been freed independently, and will not be
1077 * reused since sys_swapoff() already disabled
1078 * allocation from here, or alloc_page() failed.
1087 * Don't hold on to start_mm if it looks like exiting.
1089 if (atomic_read(&start_mm
->mm_users
) == 1) {
1091 start_mm
= &init_mm
;
1092 atomic_inc(&init_mm
.mm_users
);
1096 * Wait for and lock page. When do_swap_page races with
1097 * try_to_unuse, do_swap_page can handle the fault much
1098 * faster than try_to_unuse can locate the entry. This
1099 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1100 * defer to do_swap_page in such a case - in some tests,
1101 * do_swap_page and try_to_unuse repeatedly compete.
1103 wait_on_page_locked(page
);
1104 wait_on_page_writeback(page
);
1106 wait_on_page_writeback(page
);
1109 * Remove all references to entry.
1110 * Whenever we reach init_mm, there's no address space
1111 * to search, but use it as a reminder to search shmem.
1114 swcount
= *swap_map
;
1115 if (swap_count(swcount
)) {
1116 if (start_mm
== &init_mm
)
1117 shmem
= shmem_unuse(entry
, page
);
1119 retval
= unuse_mm(start_mm
, entry
, page
);
1121 if (swap_count(*swap_map
)) {
1122 int set_start_mm
= (*swap_map
>= swcount
);
1123 struct list_head
*p
= &start_mm
->mmlist
;
1124 struct mm_struct
*new_start_mm
= start_mm
;
1125 struct mm_struct
*prev_mm
= start_mm
;
1126 struct mm_struct
*mm
;
1128 atomic_inc(&new_start_mm
->mm_users
);
1129 atomic_inc(&prev_mm
->mm_users
);
1130 spin_lock(&mmlist_lock
);
1131 while (swap_count(*swap_map
) && !retval
&& !shmem
&&
1132 (p
= p
->next
) != &start_mm
->mmlist
) {
1133 mm
= list_entry(p
, struct mm_struct
, mmlist
);
1134 if (!atomic_inc_not_zero(&mm
->mm_users
))
1136 spin_unlock(&mmlist_lock
);
1142 swcount
= *swap_map
;
1143 if (!swap_count(swcount
)) /* any usage ? */
1145 else if (mm
== &init_mm
) {
1147 shmem
= shmem_unuse(entry
, page
);
1149 retval
= unuse_mm(mm
, entry
, page
);
1151 if (set_start_mm
&& *swap_map
< swcount
) {
1152 mmput(new_start_mm
);
1153 atomic_inc(&mm
->mm_users
);
1157 spin_lock(&mmlist_lock
);
1159 spin_unlock(&mmlist_lock
);
1162 start_mm
= new_start_mm
;
1165 /* page has already been unlocked and released */
1173 page_cache_release(page
);
1178 * If a reference remains (rare), we would like to leave
1179 * the page in the swap cache; but try_to_unmap could
1180 * then re-duplicate the entry once we drop page lock,
1181 * so we might loop indefinitely; also, that page could
1182 * not be swapped out to other storage meanwhile. So:
1183 * delete from cache even if there's another reference,
1184 * after ensuring that the data has been saved to disk -
1185 * since if the reference remains (rarer), it will be
1186 * read from disk into another page. Splitting into two
1187 * pages would be incorrect if swap supported "shared
1188 * private" pages, but they are handled by tmpfs files.
1190 if (swap_count(*swap_map
) &&
1191 PageDirty(page
) && PageSwapCache(page
)) {
1192 struct writeback_control wbc
= {
1193 .sync_mode
= WB_SYNC_NONE
,
1196 swap_writepage(page
, &wbc
);
1198 wait_on_page_writeback(page
);
1202 * It is conceivable that a racing task removed this page from
1203 * swap cache just before we acquired the page lock at the top,
1204 * or while we dropped it in unuse_mm(). The page might even
1205 * be back in swap cache on another swap area: that we must not
1206 * delete, since it may not have been written out to swap yet.
1208 if (PageSwapCache(page
) &&
1209 likely(page_private(page
) == entry
.val
))
1210 delete_from_swap_cache(page
);
1213 * So we could skip searching mms once swap count went
1214 * to 1, we did not mark any present ptes as dirty: must
1215 * mark page dirty so shrink_page_list will preserve it.
1219 page_cache_release(page
);
1222 * Make sure that we aren't completely killing
1223 * interactive performance.
1233 * After a successful try_to_unuse, if no swap is now in use, we know
1234 * we can empty the mmlist. swap_lock must be held on entry and exit.
1235 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1236 * added to the mmlist just after page_duplicate - before would be racy.
1238 static void drain_mmlist(void)
1240 struct list_head
*p
, *next
;
1243 for (type
= 0; type
< nr_swapfiles
; type
++)
1244 if (swap_info
[type
]->inuse_pages
)
1246 spin_lock(&mmlist_lock
);
1247 list_for_each_safe(p
, next
, &init_mm
.mmlist
)
1249 spin_unlock(&mmlist_lock
);
1253 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1254 * corresponds to page offset `offset'. Note that the type of this function
1255 * is sector_t, but it returns page offset into the bdev, not sector offset.
1257 sector_t
map_swap_page(swp_entry_t entry
, struct block_device
**bdev
)
1259 struct swap_info_struct
*sis
;
1260 struct swap_extent
*start_se
;
1261 struct swap_extent
*se
;
1264 sis
= swap_info
[swp_type(entry
)];
1267 offset
= swp_offset(entry
);
1268 start_se
= sis
->curr_swap_extent
;
1272 struct list_head
*lh
;
1274 if (se
->start_page
<= offset
&&
1275 offset
< (se
->start_page
+ se
->nr_pages
)) {
1276 return se
->start_block
+ (offset
- se
->start_page
);
1279 se
= list_entry(lh
, struct swap_extent
, list
);
1280 sis
->curr_swap_extent
= se
;
1281 BUG_ON(se
== start_se
); /* It *must* be present */
1286 * Free all of a swapdev's extent information
1288 static void destroy_swap_extents(struct swap_info_struct
*sis
)
1290 while (!list_empty(&sis
->first_swap_extent
.list
)) {
1291 struct swap_extent
*se
;
1293 se
= list_entry(sis
->first_swap_extent
.list
.next
,
1294 struct swap_extent
, list
);
1295 list_del(&se
->list
);
1301 * Add a block range (and the corresponding page range) into this swapdev's
1302 * extent list. The extent list is kept sorted in page order.
1304 * This function rather assumes that it is called in ascending page order.
1307 add_swap_extent(struct swap_info_struct
*sis
, unsigned long start_page
,
1308 unsigned long nr_pages
, sector_t start_block
)
1310 struct swap_extent
*se
;
1311 struct swap_extent
*new_se
;
1312 struct list_head
*lh
;
1314 if (start_page
== 0) {
1315 se
= &sis
->first_swap_extent
;
1316 sis
->curr_swap_extent
= se
;
1318 se
->nr_pages
= nr_pages
;
1319 se
->start_block
= start_block
;
1322 lh
= sis
->first_swap_extent
.list
.prev
; /* Highest extent */
1323 se
= list_entry(lh
, struct swap_extent
, list
);
1324 BUG_ON(se
->start_page
+ se
->nr_pages
!= start_page
);
1325 if (se
->start_block
+ se
->nr_pages
== start_block
) {
1327 se
->nr_pages
+= nr_pages
;
1333 * No merge. Insert a new extent, preserving ordering.
1335 new_se
= kmalloc(sizeof(*se
), GFP_KERNEL
);
1338 new_se
->start_page
= start_page
;
1339 new_se
->nr_pages
= nr_pages
;
1340 new_se
->start_block
= start_block
;
1342 list_add_tail(&new_se
->list
, &sis
->first_swap_extent
.list
);
1347 * A `swap extent' is a simple thing which maps a contiguous range of pages
1348 * onto a contiguous range of disk blocks. An ordered list of swap extents
1349 * is built at swapon time and is then used at swap_writepage/swap_readpage
1350 * time for locating where on disk a page belongs.
1352 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1353 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1354 * swap files identically.
1356 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1357 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1358 * swapfiles are handled *identically* after swapon time.
1360 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1361 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1362 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1363 * requirements, they are simply tossed out - we will never use those blocks
1366 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1367 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1368 * which will scribble on the fs.
1370 * The amount of disk space which a single swap extent represents varies.
1371 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1372 * extents in the list. To avoid much list walking, we cache the previous
1373 * search location in `curr_swap_extent', and start new searches from there.
1374 * This is extremely effective. The average number of iterations in
1375 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1377 static int setup_swap_extents(struct swap_info_struct
*sis
, sector_t
*span
)
1379 struct inode
*inode
;
1380 unsigned blocks_per_page
;
1381 unsigned long page_no
;
1383 sector_t probe_block
;
1384 sector_t last_block
;
1385 sector_t lowest_block
= -1;
1386 sector_t highest_block
= 0;
1390 inode
= sis
->swap_file
->f_mapping
->host
;
1391 if (S_ISBLK(inode
->i_mode
)) {
1392 ret
= add_swap_extent(sis
, 0, sis
->max
, 0);
1397 blkbits
= inode
->i_blkbits
;
1398 blocks_per_page
= PAGE_SIZE
>> blkbits
;
1401 * Map all the blocks into the extent list. This code doesn't try
1406 last_block
= i_size_read(inode
) >> blkbits
;
1407 while ((probe_block
+ blocks_per_page
) <= last_block
&&
1408 page_no
< sis
->max
) {
1409 unsigned block_in_page
;
1410 sector_t first_block
;
1412 first_block
= bmap(inode
, probe_block
);
1413 if (first_block
== 0)
1417 * It must be PAGE_SIZE aligned on-disk
1419 if (first_block
& (blocks_per_page
- 1)) {
1424 for (block_in_page
= 1; block_in_page
< blocks_per_page
;
1428 block
= bmap(inode
, probe_block
+ block_in_page
);
1431 if (block
!= first_block
+ block_in_page
) {
1438 first_block
>>= (PAGE_SHIFT
- blkbits
);
1439 if (page_no
) { /* exclude the header page */
1440 if (first_block
< lowest_block
)
1441 lowest_block
= first_block
;
1442 if (first_block
> highest_block
)
1443 highest_block
= first_block
;
1447 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
1449 ret
= add_swap_extent(sis
, page_no
, 1, first_block
);
1454 probe_block
+= blocks_per_page
;
1459 *span
= 1 + highest_block
- lowest_block
;
1461 page_no
= 1; /* force Empty message */
1463 sis
->pages
= page_no
- 1;
1464 sis
->highest_bit
= page_no
- 1;
1468 printk(KERN_ERR
"swapon: swapfile has holes\n");
1473 SYSCALL_DEFINE1(swapoff
, const char __user
*, specialfile
)
1475 struct swap_info_struct
*p
= NULL
;
1476 unsigned char *swap_map
;
1477 struct file
*swap_file
, *victim
;
1478 struct address_space
*mapping
;
1479 struct inode
*inode
;
1484 if (!capable(CAP_SYS_ADMIN
))
1487 pathname
= getname(specialfile
);
1488 err
= PTR_ERR(pathname
);
1489 if (IS_ERR(pathname
))
1492 victim
= filp_open(pathname
, O_RDWR
|O_LARGEFILE
, 0);
1494 err
= PTR_ERR(victim
);
1498 mapping
= victim
->f_mapping
;
1500 spin_lock(&swap_lock
);
1501 for (type
= swap_list
.head
; type
>= 0; type
= swap_info
[type
]->next
) {
1502 p
= swap_info
[type
];
1503 if (p
->flags
& SWP_WRITEOK
) {
1504 if (p
->swap_file
->f_mapping
== mapping
)
1511 spin_unlock(&swap_lock
);
1514 if (!security_vm_enough_memory(p
->pages
))
1515 vm_unacct_memory(p
->pages
);
1518 spin_unlock(&swap_lock
);
1522 swap_list
.head
= p
->next
;
1524 swap_info
[prev
]->next
= p
->next
;
1525 if (type
== swap_list
.next
) {
1526 /* just pick something that's safe... */
1527 swap_list
.next
= swap_list
.head
;
1530 for (i
= p
->next
; i
>= 0; i
= swap_info
[i
]->next
)
1531 swap_info
[i
]->prio
= p
->prio
--;
1534 nr_swap_pages
-= p
->pages
;
1535 total_swap_pages
-= p
->pages
;
1536 p
->flags
&= ~SWP_WRITEOK
;
1537 spin_unlock(&swap_lock
);
1539 current
->flags
|= PF_OOM_ORIGIN
;
1540 err
= try_to_unuse(type
);
1541 current
->flags
&= ~PF_OOM_ORIGIN
;
1544 /* re-insert swap space back into swap_list */
1545 spin_lock(&swap_lock
);
1547 p
->prio
= --least_priority
;
1549 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
]->next
) {
1550 if (p
->prio
>= swap_info
[i
]->prio
)
1556 swap_list
.head
= swap_list
.next
= type
;
1558 swap_info
[prev
]->next
= type
;
1559 nr_swap_pages
+= p
->pages
;
1560 total_swap_pages
+= p
->pages
;
1561 p
->flags
|= SWP_WRITEOK
;
1562 spin_unlock(&swap_lock
);
1566 /* wait for any unplug function to finish */
1567 down_write(&swap_unplug_sem
);
1568 up_write(&swap_unplug_sem
);
1570 destroy_swap_extents(p
);
1571 if (p
->flags
& SWP_CONTINUED
)
1572 free_swap_count_continuations(p
);
1574 mutex_lock(&swapon_mutex
);
1575 spin_lock(&swap_lock
);
1578 /* wait for anyone still in scan_swap_map */
1579 p
->highest_bit
= 0; /* cuts scans short */
1580 while (p
->flags
>= SWP_SCANNING
) {
1581 spin_unlock(&swap_lock
);
1582 schedule_timeout_uninterruptible(1);
1583 spin_lock(&swap_lock
);
1586 swap_file
= p
->swap_file
;
1587 p
->swap_file
= NULL
;
1589 swap_map
= p
->swap_map
;
1592 spin_unlock(&swap_lock
);
1593 mutex_unlock(&swapon_mutex
);
1595 /* Destroy swap account informatin */
1596 swap_cgroup_swapoff(type
);
1598 inode
= mapping
->host
;
1599 if (S_ISBLK(inode
->i_mode
)) {
1600 struct block_device
*bdev
= I_BDEV(inode
);
1601 set_blocksize(bdev
, p
->old_block_size
);
1604 mutex_lock(&inode
->i_mutex
);
1605 inode
->i_flags
&= ~S_SWAPFILE
;
1606 mutex_unlock(&inode
->i_mutex
);
1608 filp_close(swap_file
, NULL
);
1612 filp_close(victim
, NULL
);
1617 #ifdef CONFIG_PROC_FS
1619 static void *swap_start(struct seq_file
*swap
, loff_t
*pos
)
1621 struct swap_info_struct
*si
;
1625 mutex_lock(&swapon_mutex
);
1628 return SEQ_START_TOKEN
;
1630 for (type
= 0; type
< nr_swapfiles
; type
++) {
1631 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1632 si
= swap_info
[type
];
1633 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
1642 static void *swap_next(struct seq_file
*swap
, void *v
, loff_t
*pos
)
1644 struct swap_info_struct
*si
= v
;
1647 if (v
== SEQ_START_TOKEN
)
1650 type
= si
->type
+ 1;
1652 for (; type
< nr_swapfiles
; type
++) {
1653 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
1654 si
= swap_info
[type
];
1655 if (!(si
->flags
& SWP_USED
) || !si
->swap_map
)
1664 static void swap_stop(struct seq_file
*swap
, void *v
)
1666 mutex_unlock(&swapon_mutex
);
1669 static int swap_show(struct seq_file
*swap
, void *v
)
1671 struct swap_info_struct
*si
= v
;
1675 if (si
== SEQ_START_TOKEN
) {
1676 seq_puts(swap
,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
1680 file
= si
->swap_file
;
1681 len
= seq_path(swap
, &file
->f_path
, " \t\n\\");
1682 seq_printf(swap
, "%*s%s\t%u\t%u\t%d\n",
1683 len
< 40 ? 40 - len
: 1, " ",
1684 S_ISBLK(file
->f_path
.dentry
->d_inode
->i_mode
) ?
1685 "partition" : "file\t",
1686 si
->pages
<< (PAGE_SHIFT
- 10),
1687 si
->inuse_pages
<< (PAGE_SHIFT
- 10),
1692 static const struct seq_operations swaps_op
= {
1693 .start
= swap_start
,
1699 static int swaps_open(struct inode
*inode
, struct file
*file
)
1701 return seq_open(file
, &swaps_op
);
1704 static const struct file_operations proc_swaps_operations
= {
1707 .llseek
= seq_lseek
,
1708 .release
= seq_release
,
1711 static int __init
procswaps_init(void)
1713 proc_create("swaps", 0, NULL
, &proc_swaps_operations
);
1716 __initcall(procswaps_init
);
1717 #endif /* CONFIG_PROC_FS */
1719 #ifdef MAX_SWAPFILES_CHECK
1720 static int __init
max_swapfiles_check(void)
1722 MAX_SWAPFILES_CHECK();
1725 late_initcall(max_swapfiles_check
);
1729 * Written 01/25/92 by Simmule Turner, heavily changed by Linus.
1731 * The swapon system call
1733 SYSCALL_DEFINE2(swapon
, const char __user
*, specialfile
, int, swap_flags
)
1735 struct swap_info_struct
*p
;
1737 struct block_device
*bdev
= NULL
;
1738 struct file
*swap_file
= NULL
;
1739 struct address_space
*mapping
;
1743 union swap_header
*swap_header
= NULL
;
1744 unsigned int nr_good_pages
= 0;
1747 unsigned long maxpages
= 1;
1748 unsigned long swapfilepages
;
1749 unsigned char *swap_map
= NULL
;
1750 struct page
*page
= NULL
;
1751 struct inode
*inode
= NULL
;
1754 if (!capable(CAP_SYS_ADMIN
))
1757 p
= kzalloc(sizeof(*p
), GFP_KERNEL
);
1761 spin_lock(&swap_lock
);
1762 for (type
= 0; type
< nr_swapfiles
; type
++) {
1763 if (!(swap_info
[type
]->flags
& SWP_USED
))
1767 if (type
>= MAX_SWAPFILES
) {
1768 spin_unlock(&swap_lock
);
1772 if (type
>= nr_swapfiles
) {
1774 swap_info
[type
] = p
;
1776 * Write swap_info[type] before nr_swapfiles, in case a
1777 * racing procfs swap_start() or swap_next() is reading them.
1778 * (We never shrink nr_swapfiles, we never free this entry.)
1784 p
= swap_info
[type
];
1786 * Do not memset this entry: a racing procfs swap_next()
1787 * would be relying on p->type to remain valid.
1790 INIT_LIST_HEAD(&p
->first_swap_extent
.list
);
1791 p
->flags
= SWP_USED
;
1793 spin_unlock(&swap_lock
);
1795 name
= getname(specialfile
);
1796 error
= PTR_ERR(name
);
1801 swap_file
= filp_open(name
, O_RDWR
|O_LARGEFILE
, 0);
1802 error
= PTR_ERR(swap_file
);
1803 if (IS_ERR(swap_file
)) {
1808 p
->swap_file
= swap_file
;
1809 mapping
= swap_file
->f_mapping
;
1810 inode
= mapping
->host
;
1813 for (i
= 0; i
< nr_swapfiles
; i
++) {
1814 struct swap_info_struct
*q
= swap_info
[i
];
1816 if (i
== type
|| !q
->swap_file
)
1818 if (mapping
== q
->swap_file
->f_mapping
)
1823 if (S_ISBLK(inode
->i_mode
)) {
1824 bdev
= I_BDEV(inode
);
1825 error
= bd_claim(bdev
, sys_swapon
);
1831 p
->old_block_size
= block_size(bdev
);
1832 error
= set_blocksize(bdev
, PAGE_SIZE
);
1836 } else if (S_ISREG(inode
->i_mode
)) {
1837 p
->bdev
= inode
->i_sb
->s_bdev
;
1838 mutex_lock(&inode
->i_mutex
);
1840 if (IS_SWAPFILE(inode
)) {
1848 swapfilepages
= i_size_read(inode
) >> PAGE_SHIFT
;
1851 * Read the swap header.
1853 if (!mapping
->a_ops
->readpage
) {
1857 page
= read_mapping_page(mapping
, 0, swap_file
);
1859 error
= PTR_ERR(page
);
1862 swap_header
= kmap(page
);
1864 if (memcmp("SWAPSPACE2", swap_header
->magic
.magic
, 10)) {
1865 printk(KERN_ERR
"Unable to find swap-space signature\n");
1870 /* swap partition endianess hack... */
1871 if (swab32(swap_header
->info
.version
) == 1) {
1872 swab32s(&swap_header
->info
.version
);
1873 swab32s(&swap_header
->info
.last_page
);
1874 swab32s(&swap_header
->info
.nr_badpages
);
1875 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++)
1876 swab32s(&swap_header
->info
.badpages
[i
]);
1878 /* Check the swap header's sub-version */
1879 if (swap_header
->info
.version
!= 1) {
1881 "Unable to handle swap header version %d\n",
1882 swap_header
->info
.version
);
1888 p
->cluster_next
= 1;
1892 * Find out how many pages are allowed for a single swap
1893 * device. There are two limiting factors: 1) the number of
1894 * bits for the swap offset in the swp_entry_t type and
1895 * 2) the number of bits in the a swap pte as defined by
1896 * the different architectures. In order to find the
1897 * largest possible bit mask a swap entry with swap type 0
1898 * and swap offset ~0UL is created, encoded to a swap pte,
1899 * decoded to a swp_entry_t again and finally the swap
1900 * offset is extracted. This will mask all the bits from
1901 * the initial ~0UL mask that can't be encoded in either
1902 * the swp_entry_t or the architecture definition of a
1905 maxpages
= swp_offset(pte_to_swp_entry(
1906 swp_entry_to_pte(swp_entry(0, ~0UL)))) - 1;
1907 if (maxpages
> swap_header
->info
.last_page
)
1908 maxpages
= swap_header
->info
.last_page
;
1909 p
->highest_bit
= maxpages
- 1;
1914 if (swapfilepages
&& maxpages
> swapfilepages
) {
1916 "Swap area shorter than signature indicates\n");
1919 if (swap_header
->info
.nr_badpages
&& S_ISREG(inode
->i_mode
))
1921 if (swap_header
->info
.nr_badpages
> MAX_SWAP_BADPAGES
)
1924 /* OK, set up the swap map and apply the bad block list */
1925 swap_map
= vmalloc(maxpages
);
1931 memset(swap_map
, 0, maxpages
);
1932 for (i
= 0; i
< swap_header
->info
.nr_badpages
; i
++) {
1933 int page_nr
= swap_header
->info
.badpages
[i
];
1934 if (page_nr
<= 0 || page_nr
>= swap_header
->info
.last_page
) {
1938 swap_map
[page_nr
] = SWAP_MAP_BAD
;
1941 error
= swap_cgroup_swapon(type
, maxpages
);
1945 nr_good_pages
= swap_header
->info
.last_page
-
1946 swap_header
->info
.nr_badpages
-
1947 1 /* header page */;
1949 if (nr_good_pages
) {
1950 swap_map
[0] = SWAP_MAP_BAD
;
1952 p
->pages
= nr_good_pages
;
1953 nr_extents
= setup_swap_extents(p
, &span
);
1954 if (nr_extents
< 0) {
1958 nr_good_pages
= p
->pages
;
1960 if (!nr_good_pages
) {
1961 printk(KERN_WARNING
"Empty swap-file\n");
1967 if (blk_queue_nonrot(bdev_get_queue(p
->bdev
))) {
1968 p
->flags
|= SWP_SOLIDSTATE
;
1969 p
->cluster_next
= 1 + (random32() % p
->highest_bit
);
1971 if (discard_swap(p
) == 0)
1972 p
->flags
|= SWP_DISCARDABLE
;
1975 mutex_lock(&swapon_mutex
);
1976 spin_lock(&swap_lock
);
1977 if (swap_flags
& SWAP_FLAG_PREFER
)
1979 (swap_flags
& SWAP_FLAG_PRIO_MASK
) >> SWAP_FLAG_PRIO_SHIFT
;
1981 p
->prio
= --least_priority
;
1982 p
->swap_map
= swap_map
;
1983 p
->flags
|= SWP_WRITEOK
;
1984 nr_swap_pages
+= nr_good_pages
;
1985 total_swap_pages
+= nr_good_pages
;
1987 printk(KERN_INFO
"Adding %uk swap on %s. "
1988 "Priority:%d extents:%d across:%lluk %s%s\n",
1989 nr_good_pages
<<(PAGE_SHIFT
-10), name
, p
->prio
,
1990 nr_extents
, (unsigned long long)span
<<(PAGE_SHIFT
-10),
1991 (p
->flags
& SWP_SOLIDSTATE
) ? "SS" : "",
1992 (p
->flags
& SWP_DISCARDABLE
) ? "D" : "");
1994 /* insert swap space into swap_list: */
1996 for (i
= swap_list
.head
; i
>= 0; i
= swap_info
[i
]->next
) {
1997 if (p
->prio
>= swap_info
[i
]->prio
)
2003 swap_list
.head
= swap_list
.next
= type
;
2005 swap_info
[prev
]->next
= type
;
2006 spin_unlock(&swap_lock
);
2007 mutex_unlock(&swapon_mutex
);
2012 set_blocksize(bdev
, p
->old_block_size
);
2015 destroy_swap_extents(p
);
2016 swap_cgroup_swapoff(type
);
2018 spin_lock(&swap_lock
);
2019 p
->swap_file
= NULL
;
2021 spin_unlock(&swap_lock
);
2024 filp_close(swap_file
, NULL
);
2026 if (page
&& !IS_ERR(page
)) {
2028 page_cache_release(page
);
2034 inode
->i_flags
|= S_SWAPFILE
;
2035 mutex_unlock(&inode
->i_mutex
);
2040 void si_swapinfo(struct sysinfo
*val
)
2043 unsigned long nr_to_be_unused
= 0;
2045 spin_lock(&swap_lock
);
2046 for (type
= 0; type
< nr_swapfiles
; type
++) {
2047 struct swap_info_struct
*si
= swap_info
[type
];
2049 if ((si
->flags
& SWP_USED
) && !(si
->flags
& SWP_WRITEOK
))
2050 nr_to_be_unused
+= si
->inuse_pages
;
2052 val
->freeswap
= nr_swap_pages
+ nr_to_be_unused
;
2053 val
->totalswap
= total_swap_pages
+ nr_to_be_unused
;
2054 spin_unlock(&swap_lock
);
2058 * Verify that a swap entry is valid and increment its swap map count.
2060 * Returns error code in following case.
2062 * - swp_entry is invalid -> EINVAL
2063 * - swp_entry is migration entry -> EINVAL
2064 * - swap-cache reference is requested but there is already one. -> EEXIST
2065 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2066 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2068 static int __swap_duplicate(swp_entry_t entry
, unsigned char usage
)
2070 struct swap_info_struct
*p
;
2071 unsigned long offset
, type
;
2072 unsigned char count
;
2073 unsigned char has_cache
;
2076 if (non_swap_entry(entry
))
2079 type
= swp_type(entry
);
2080 if (type
>= nr_swapfiles
)
2082 p
= swap_info
[type
];
2083 offset
= swp_offset(entry
);
2085 spin_lock(&swap_lock
);
2086 if (unlikely(offset
>= p
->max
))
2089 count
= p
->swap_map
[offset
];
2090 has_cache
= count
& SWAP_HAS_CACHE
;
2091 count
&= ~SWAP_HAS_CACHE
;
2094 if (usage
== SWAP_HAS_CACHE
) {
2096 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2097 if (!has_cache
&& count
)
2098 has_cache
= SWAP_HAS_CACHE
;
2099 else if (has_cache
) /* someone else added cache */
2101 else /* no users remaining */
2104 } else if (count
|| has_cache
) {
2106 if ((count
& ~COUNT_CONTINUED
) < SWAP_MAP_MAX
)
2108 else if ((count
& ~COUNT_CONTINUED
) > SWAP_MAP_MAX
)
2110 else if (swap_count_continued(p
, offset
, count
))
2111 count
= COUNT_CONTINUED
;
2115 err
= -ENOENT
; /* unused swap entry */
2117 p
->swap_map
[offset
] = count
| has_cache
;
2120 spin_unlock(&swap_lock
);
2125 printk(KERN_ERR
"swap_dup: %s%08lx\n", Bad_file
, entry
.val
);
2130 * increase reference count of swap entry by 1.
2132 int swap_duplicate(swp_entry_t entry
)
2136 while (!err
&& __swap_duplicate(entry
, 1) == -ENOMEM
)
2137 err
= add_swap_count_continuation(entry
, GFP_ATOMIC
);
2142 * @entry: swap entry for which we allocate swap cache.
2144 * Called when allocating swap cache for existing swap entry,
2145 * This can return error codes. Returns 0 at success.
2146 * -EBUSY means there is a swap cache.
2147 * Note: return code is different from swap_duplicate().
2149 int swapcache_prepare(swp_entry_t entry
)
2151 return __swap_duplicate(entry
, SWAP_HAS_CACHE
);
2155 * swap_lock prevents swap_map being freed. Don't grab an extra
2156 * reference on the swaphandle, it doesn't matter if it becomes unused.
2158 int valid_swaphandles(swp_entry_t entry
, unsigned long *offset
)
2160 struct swap_info_struct
*si
;
2161 int our_page_cluster
= page_cluster
;
2162 pgoff_t target
, toff
;
2166 if (!our_page_cluster
) /* no readahead */
2169 si
= swap_info
[swp_type(entry
)];
2170 target
= swp_offset(entry
);
2171 base
= (target
>> our_page_cluster
) << our_page_cluster
;
2172 end
= base
+ (1 << our_page_cluster
);
2173 if (!base
) /* first page is swap header */
2176 spin_lock(&swap_lock
);
2177 if (end
> si
->max
) /* don't go beyond end of map */
2180 /* Count contiguous allocated slots above our target */
2181 for (toff
= target
; ++toff
< end
; nr_pages
++) {
2182 /* Don't read in free or bad pages */
2183 if (!si
->swap_map
[toff
])
2185 if (swap_count(si
->swap_map
[toff
]) == SWAP_MAP_BAD
)
2188 /* Count contiguous allocated slots below our target */
2189 for (toff
= target
; --toff
>= base
; nr_pages
++) {
2190 /* Don't read in free or bad pages */
2191 if (!si
->swap_map
[toff
])
2193 if (swap_count(si
->swap_map
[toff
]) == SWAP_MAP_BAD
)
2196 spin_unlock(&swap_lock
);
2199 * Indicate starting offset, and return number of pages to get:
2200 * if only 1, say 0, since there's then no readahead to be done.
2203 return nr_pages
? ++nr_pages
: 0;
2207 * add_swap_count_continuation - called when a swap count is duplicated
2208 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2209 * page of the original vmalloc'ed swap_map, to hold the continuation count
2210 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2211 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2213 * These continuation pages are seldom referenced: the common paths all work
2214 * on the original swap_map, only referring to a continuation page when the
2215 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2217 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2218 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2219 * can be called after dropping locks.
2221 int add_swap_count_continuation(swp_entry_t entry
, gfp_t gfp_mask
)
2223 struct swap_info_struct
*si
;
2226 struct page
*list_page
;
2228 unsigned char count
;
2231 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2232 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2234 page
= alloc_page(gfp_mask
| __GFP_HIGHMEM
);
2236 si
= swap_info_get(entry
);
2239 * An acceptable race has occurred since the failing
2240 * __swap_duplicate(): the swap entry has been freed,
2241 * perhaps even the whole swap_map cleared for swapoff.
2246 offset
= swp_offset(entry
);
2247 count
= si
->swap_map
[offset
] & ~SWAP_HAS_CACHE
;
2249 if ((count
& ~COUNT_CONTINUED
) != SWAP_MAP_MAX
) {
2251 * The higher the swap count, the more likely it is that tasks
2252 * will race to add swap count continuation: we need to avoid
2253 * over-provisioning.
2259 spin_unlock(&swap_lock
);
2264 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2265 * no architecture is using highmem pages for kernel pagetables: so it
2266 * will not corrupt the GFP_ATOMIC caller's atomic pagetable kmaps.
2268 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2269 offset
&= ~PAGE_MASK
;
2272 * Page allocation does not initialize the page's lru field,
2273 * but it does always reset its private field.
2275 if (!page_private(head
)) {
2276 BUG_ON(count
& COUNT_CONTINUED
);
2277 INIT_LIST_HEAD(&head
->lru
);
2278 set_page_private(head
, SWP_CONTINUED
);
2279 si
->flags
|= SWP_CONTINUED
;
2282 list_for_each_entry(list_page
, &head
->lru
, lru
) {
2286 * If the previous map said no continuation, but we've found
2287 * a continuation page, free our allocation and use this one.
2289 if (!(count
& COUNT_CONTINUED
))
2292 map
= kmap_atomic(list_page
, KM_USER0
) + offset
;
2294 kunmap_atomic(map
, KM_USER0
);
2297 * If this continuation count now has some space in it,
2298 * free our allocation and use this one.
2300 if ((count
& ~COUNT_CONTINUED
) != SWAP_CONT_MAX
)
2304 list_add_tail(&page
->lru
, &head
->lru
);
2305 page
= NULL
; /* now it's attached, don't free it */
2307 spin_unlock(&swap_lock
);
2315 * swap_count_continued - when the original swap_map count is incremented
2316 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2317 * into, carry if so, or else fail until a new continuation page is allocated;
2318 * when the original swap_map count is decremented from 0 with continuation,
2319 * borrow from the continuation and report whether it still holds more.
2320 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2322 static bool swap_count_continued(struct swap_info_struct
*si
,
2323 pgoff_t offset
, unsigned char count
)
2329 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2330 if (page_private(head
) != SWP_CONTINUED
) {
2331 BUG_ON(count
& COUNT_CONTINUED
);
2332 return false; /* need to add count continuation */
2335 offset
&= ~PAGE_MASK
;
2336 page
= list_entry(head
->lru
.next
, struct page
, lru
);
2337 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2339 if (count
== SWAP_MAP_MAX
) /* initial increment from swap_map */
2340 goto init_map
; /* jump over SWAP_CONT_MAX checks */
2342 if (count
== (SWAP_MAP_MAX
| COUNT_CONTINUED
)) { /* incrementing */
2344 * Think of how you add 1 to 999
2346 while (*map
== (SWAP_CONT_MAX
| COUNT_CONTINUED
)) {
2347 kunmap_atomic(map
, KM_USER0
);
2348 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2349 BUG_ON(page
== head
);
2350 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2352 if (*map
== SWAP_CONT_MAX
) {
2353 kunmap_atomic(map
, KM_USER0
);
2354 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2356 return false; /* add count continuation */
2357 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2358 init_map
: *map
= 0; /* we didn't zero the page */
2361 kunmap_atomic(map
, KM_USER0
);
2362 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2363 while (page
!= head
) {
2364 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2365 *map
= COUNT_CONTINUED
;
2366 kunmap_atomic(map
, KM_USER0
);
2367 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2369 return true; /* incremented */
2371 } else { /* decrementing */
2373 * Think of how you subtract 1 from 1000
2375 BUG_ON(count
!= COUNT_CONTINUED
);
2376 while (*map
== COUNT_CONTINUED
) {
2377 kunmap_atomic(map
, KM_USER0
);
2378 page
= list_entry(page
->lru
.next
, struct page
, lru
);
2379 BUG_ON(page
== head
);
2380 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2386 kunmap_atomic(map
, KM_USER0
);
2387 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2388 while (page
!= head
) {
2389 map
= kmap_atomic(page
, KM_USER0
) + offset
;
2390 *map
= SWAP_CONT_MAX
| count
;
2391 count
= COUNT_CONTINUED
;
2392 kunmap_atomic(map
, KM_USER0
);
2393 page
= list_entry(page
->lru
.prev
, struct page
, lru
);
2395 return count
== COUNT_CONTINUED
;
2400 * free_swap_count_continuations - swapoff free all the continuation pages
2401 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2403 static void free_swap_count_continuations(struct swap_info_struct
*si
)
2407 for (offset
= 0; offset
< si
->max
; offset
+= PAGE_SIZE
) {
2409 head
= vmalloc_to_page(si
->swap_map
+ offset
);
2410 if (page_private(head
)) {
2411 struct list_head
*this, *next
;
2412 list_for_each_safe(this, next
, &head
->lru
) {
2414 page
= list_entry(this, struct page
, lru
);