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/page_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;
51 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
52 long total_swap_pages;
53 static int least_priority;
54 static atomic_t highest_priority_index = ATOMIC_INIT(-1);
56 static const char Bad_file[] = "Bad swap file entry ";
57 static const char Unused_file[] = "Unused swap file entry ";
58 static const char Bad_offset[] = "Bad swap offset entry ";
59 static const char Unused_offset[] = "Unused swap offset entry ";
61 struct swap_list_t swap_list = {-1, -1};
63 struct swap_info_struct *swap_info[MAX_SWAPFILES];
65 static DEFINE_MUTEX(swapon_mutex);
67 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
68 /* Activity counter to indicate that a swapon or swapoff has occurred */
69 static atomic_t proc_poll_event = ATOMIC_INIT(0);
71 static inline unsigned char swap_count(unsigned char ent)
73 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
76 /* returns 1 if swap entry is freed */
78 __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
80 swp_entry_t entry = swp_entry(si->type, offset);
84 page = find_get_page(swap_address_space(entry), entry.val);
88 * This function is called from scan_swap_map() and it's called
89 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
90 * We have to use trylock for avoiding deadlock. This is a special
91 * case and you should use try_to_free_swap() with explicit lock_page()
92 * in usual operations.
94 if (trylock_page(page)) {
95 ret = try_to_free_swap(page);
98 page_cache_release(page);
103 * swapon tell device that all the old swap contents can be discarded,
104 * to allow the swap device to optimize its wear-levelling.
106 static int discard_swap(struct swap_info_struct *si)
108 struct swap_extent *se;
109 sector_t start_block;
113 /* Do not discard the swap header page! */
114 se = &si->first_swap_extent;
115 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
116 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
118 err = blkdev_issue_discard(si->bdev, start_block,
119 nr_blocks, GFP_KERNEL, 0);
125 list_for_each_entry(se, &si->first_swap_extent.list, list) {
126 start_block = se->start_block << (PAGE_SHIFT - 9);
127 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
129 err = blkdev_issue_discard(si->bdev, start_block,
130 nr_blocks, GFP_KERNEL, 0);
136 return err; /* That will often be -EOPNOTSUPP */
140 * swap allocation tell device that a cluster of swap can now be discarded,
141 * to allow the swap device to optimize its wear-levelling.
143 static void discard_swap_cluster(struct swap_info_struct *si,
144 pgoff_t start_page, pgoff_t nr_pages)
146 struct swap_extent *se = si->curr_swap_extent;
147 int found_extent = 0;
150 struct list_head *lh;
152 if (se->start_page <= start_page &&
153 start_page < se->start_page + se->nr_pages) {
154 pgoff_t offset = start_page - se->start_page;
155 sector_t start_block = se->start_block + offset;
156 sector_t nr_blocks = se->nr_pages - offset;
158 if (nr_blocks > nr_pages)
159 nr_blocks = nr_pages;
160 start_page += nr_blocks;
161 nr_pages -= nr_blocks;
164 si->curr_swap_extent = se;
166 start_block <<= PAGE_SHIFT - 9;
167 nr_blocks <<= PAGE_SHIFT - 9;
168 if (blkdev_issue_discard(si->bdev, start_block,
169 nr_blocks, GFP_NOIO, 0))
174 se = list_entry(lh, struct swap_extent, list);
178 #define SWAPFILE_CLUSTER 256
179 #define LATENCY_LIMIT 256
181 static inline void cluster_set_flag(struct swap_cluster_info *info,
187 static inline unsigned int cluster_count(struct swap_cluster_info *info)
192 static inline void cluster_set_count(struct swap_cluster_info *info,
198 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
199 unsigned int c, unsigned int f)
205 static inline unsigned int cluster_next(struct swap_cluster_info *info)
210 static inline void cluster_set_next(struct swap_cluster_info *info,
216 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
217 unsigned int n, unsigned int f)
223 static inline bool cluster_is_free(struct swap_cluster_info *info)
225 return info->flags & CLUSTER_FLAG_FREE;
228 static inline bool cluster_is_null(struct swap_cluster_info *info)
230 return info->flags & CLUSTER_FLAG_NEXT_NULL;
233 static inline void cluster_set_null(struct swap_cluster_info *info)
235 info->flags = CLUSTER_FLAG_NEXT_NULL;
239 /* Add a cluster to discard list and schedule it to do discard */
240 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
244 * If scan_swap_map() can't find a free cluster, it will check
245 * si->swap_map directly. To make sure the discarding cluster isn't
246 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
247 * will be cleared after discard
249 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
250 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
252 if (cluster_is_null(&si->discard_cluster_head)) {
253 cluster_set_next_flag(&si->discard_cluster_head,
255 cluster_set_next_flag(&si->discard_cluster_tail,
258 unsigned int tail = cluster_next(&si->discard_cluster_tail);
259 cluster_set_next(&si->cluster_info[tail], idx);
260 cluster_set_next_flag(&si->discard_cluster_tail,
264 schedule_work(&si->discard_work);
268 * Doing discard actually. After a cluster discard is finished, the cluster
269 * will be added to free cluster list. caller should hold si->lock.
271 static void swap_do_scheduled_discard(struct swap_info_struct *si)
273 struct swap_cluster_info *info;
276 info = si->cluster_info;
278 while (!cluster_is_null(&si->discard_cluster_head)) {
279 idx = cluster_next(&si->discard_cluster_head);
281 cluster_set_next_flag(&si->discard_cluster_head,
282 cluster_next(&info[idx]), 0);
283 if (cluster_next(&si->discard_cluster_tail) == idx) {
284 cluster_set_null(&si->discard_cluster_head);
285 cluster_set_null(&si->discard_cluster_tail);
287 spin_unlock(&si->lock);
289 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
292 spin_lock(&si->lock);
293 cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
294 if (cluster_is_null(&si->free_cluster_head)) {
295 cluster_set_next_flag(&si->free_cluster_head,
297 cluster_set_next_flag(&si->free_cluster_tail,
302 tail = cluster_next(&si->free_cluster_tail);
303 cluster_set_next(&info[tail], idx);
304 cluster_set_next_flag(&si->free_cluster_tail,
307 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
308 0, SWAPFILE_CLUSTER);
312 static void swap_discard_work(struct work_struct *work)
314 struct swap_info_struct *si;
316 si = container_of(work, struct swap_info_struct, discard_work);
318 spin_lock(&si->lock);
319 swap_do_scheduled_discard(si);
320 spin_unlock(&si->lock);
324 * The cluster corresponding to page_nr will be used. The cluster will be
325 * removed from free cluster list and its usage counter will be increased.
327 static void inc_cluster_info_page(struct swap_info_struct *p,
328 struct swap_cluster_info *cluster_info, unsigned long page_nr)
330 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
334 if (cluster_is_free(&cluster_info[idx])) {
335 VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx);
336 cluster_set_next_flag(&p->free_cluster_head,
337 cluster_next(&cluster_info[idx]), 0);
338 if (cluster_next(&p->free_cluster_tail) == idx) {
339 cluster_set_null(&p->free_cluster_tail);
340 cluster_set_null(&p->free_cluster_head);
342 cluster_set_count_flag(&cluster_info[idx], 0, 0);
345 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
346 cluster_set_count(&cluster_info[idx],
347 cluster_count(&cluster_info[idx]) + 1);
351 * The cluster corresponding to page_nr decreases one usage. If the usage
352 * counter becomes 0, which means no page in the cluster is in using, we can
353 * optionally discard the cluster and add it to free cluster list.
355 static void dec_cluster_info_page(struct swap_info_struct *p,
356 struct swap_cluster_info *cluster_info, unsigned long page_nr)
358 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
363 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
364 cluster_set_count(&cluster_info[idx],
365 cluster_count(&cluster_info[idx]) - 1);
367 if (cluster_count(&cluster_info[idx]) == 0) {
369 * If the swap is discardable, prepare discard the cluster
370 * instead of free it immediately. The cluster will be freed
373 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
374 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
375 swap_cluster_schedule_discard(p, idx);
379 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
380 if (cluster_is_null(&p->free_cluster_head)) {
381 cluster_set_next_flag(&p->free_cluster_head, idx, 0);
382 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
384 unsigned int tail = cluster_next(&p->free_cluster_tail);
385 cluster_set_next(&cluster_info[tail], idx);
386 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
392 * It's possible scan_swap_map() uses a free cluster in the middle of free
393 * cluster list. Avoiding such abuse to avoid list corruption.
396 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
397 unsigned long offset)
399 struct percpu_cluster *percpu_cluster;
402 offset /= SWAPFILE_CLUSTER;
403 conflict = !cluster_is_null(&si->free_cluster_head) &&
404 offset != cluster_next(&si->free_cluster_head) &&
405 cluster_is_free(&si->cluster_info[offset]);
410 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
411 cluster_set_null(&percpu_cluster->index);
416 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
417 * might involve allocating a new cluster for current CPU too.
419 static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
420 unsigned long *offset, unsigned long *scan_base)
422 struct percpu_cluster *cluster;
427 cluster = this_cpu_ptr(si->percpu_cluster);
428 if (cluster_is_null(&cluster->index)) {
429 if (!cluster_is_null(&si->free_cluster_head)) {
430 cluster->index = si->free_cluster_head;
431 cluster->next = cluster_next(&cluster->index) *
433 } else if (!cluster_is_null(&si->discard_cluster_head)) {
435 * we don't have free cluster but have some clusters in
436 * discarding, do discard now and reclaim them
438 swap_do_scheduled_discard(si);
439 *scan_base = *offset = si->cluster_next;
448 * Other CPUs can use our cluster if they can't find a free cluster,
449 * check if there is still free entry in the cluster
452 while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
454 if (!si->swap_map[tmp]) {
461 cluster_set_null(&cluster->index);
464 cluster->next = tmp + 1;
469 static unsigned long scan_swap_map(struct swap_info_struct *si,
472 unsigned long offset;
473 unsigned long scan_base;
474 unsigned long last_in_cluster = 0;
475 int latency_ration = LATENCY_LIMIT;
478 * We try to cluster swap pages by allocating them sequentially
479 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
480 * way, however, we resort to first-free allocation, starting
481 * a new cluster. This prevents us from scattering swap pages
482 * all over the entire swap partition, so that we reduce
483 * overall disk seek times between swap pages. -- sct
484 * But we do now try to find an empty cluster. -Andrea
485 * And we let swap pages go all over an SSD partition. Hugh
488 si->flags += SWP_SCANNING;
489 scan_base = offset = si->cluster_next;
492 if (si->cluster_info) {
493 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
497 if (unlikely(!si->cluster_nr--)) {
498 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
499 si->cluster_nr = SWAPFILE_CLUSTER - 1;
503 spin_unlock(&si->lock);
506 * If seek is expensive, start searching for new cluster from
507 * start of partition, to minimize the span of allocated swap.
508 * But if seek is cheap, search from our current position, so
509 * that swap is allocated from all over the partition: if the
510 * Flash Translation Layer only remaps within limited zones,
511 * we don't want to wear out the first zone too quickly.
513 if (!(si->flags & SWP_SOLIDSTATE))
514 scan_base = offset = si->lowest_bit;
515 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
517 /* Locate the first empty (unaligned) cluster */
518 for (; last_in_cluster <= si->highest_bit; offset++) {
519 if (si->swap_map[offset])
520 last_in_cluster = offset + SWAPFILE_CLUSTER;
521 else if (offset == last_in_cluster) {
522 spin_lock(&si->lock);
523 offset -= SWAPFILE_CLUSTER - 1;
524 si->cluster_next = offset;
525 si->cluster_nr = SWAPFILE_CLUSTER - 1;
528 if (unlikely(--latency_ration < 0)) {
530 latency_ration = LATENCY_LIMIT;
534 offset = si->lowest_bit;
535 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
537 /* Locate the first empty (unaligned) cluster */
538 for (; last_in_cluster < scan_base; offset++) {
539 if (si->swap_map[offset])
540 last_in_cluster = offset + SWAPFILE_CLUSTER;
541 else if (offset == last_in_cluster) {
542 spin_lock(&si->lock);
543 offset -= SWAPFILE_CLUSTER - 1;
544 si->cluster_next = offset;
545 si->cluster_nr = SWAPFILE_CLUSTER - 1;
548 if (unlikely(--latency_ration < 0)) {
550 latency_ration = LATENCY_LIMIT;
555 spin_lock(&si->lock);
556 si->cluster_nr = SWAPFILE_CLUSTER - 1;
560 if (si->cluster_info) {
561 while (scan_swap_map_ssd_cluster_conflict(si, offset))
562 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
564 if (!(si->flags & SWP_WRITEOK))
566 if (!si->highest_bit)
568 if (offset > si->highest_bit)
569 scan_base = offset = si->lowest_bit;
571 /* reuse swap entry of cache-only swap if not busy. */
572 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
574 spin_unlock(&si->lock);
575 swap_was_freed = __try_to_reclaim_swap(si, offset);
576 spin_lock(&si->lock);
577 /* entry was freed successfully, try to use this again */
580 goto scan; /* check next one */
583 if (si->swap_map[offset])
586 if (offset == si->lowest_bit)
588 if (offset == si->highest_bit)
591 if (si->inuse_pages == si->pages) {
592 si->lowest_bit = si->max;
595 si->swap_map[offset] = usage;
596 inc_cluster_info_page(si, si->cluster_info, offset);
597 si->cluster_next = offset + 1;
598 si->flags -= SWP_SCANNING;
603 spin_unlock(&si->lock);
604 while (++offset <= si->highest_bit) {
605 if (!si->swap_map[offset]) {
606 spin_lock(&si->lock);
609 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
610 spin_lock(&si->lock);
613 if (unlikely(--latency_ration < 0)) {
615 latency_ration = LATENCY_LIMIT;
618 offset = si->lowest_bit;
619 while (offset < scan_base) {
620 if (!si->swap_map[offset]) {
621 spin_lock(&si->lock);
624 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
625 spin_lock(&si->lock);
628 if (unlikely(--latency_ration < 0)) {
630 latency_ration = LATENCY_LIMIT;
634 spin_lock(&si->lock);
637 si->flags -= SWP_SCANNING;
641 swp_entry_t get_swap_page(void)
643 struct swap_info_struct *si;
649 spin_lock(&swap_lock);
650 if (atomic_long_read(&nr_swap_pages) <= 0)
652 atomic_long_dec(&nr_swap_pages);
654 for (type = swap_list.next; type >= 0 && wrapped < 2; type = next) {
655 hp_index = atomic_xchg(&highest_priority_index, -1);
657 * highest_priority_index records current highest priority swap
658 * type which just frees swap entries. If its priority is
659 * higher than that of swap_list.next swap type, we use it. It
660 * isn't protected by swap_lock, so it can be an invalid value
661 * if the corresponding swap type is swapoff. We double check
662 * the flags here. It's even possible the swap type is swapoff
663 * and swapon again and its priority is changed. In such rare
664 * case, low prority swap type might be used, but eventually
665 * high priority swap will be used after several rounds of
668 if (hp_index != -1 && hp_index != type &&
669 swap_info[type]->prio < swap_info[hp_index]->prio &&
670 (swap_info[hp_index]->flags & SWP_WRITEOK)) {
672 swap_list.next = type;
675 si = swap_info[type];
678 (!wrapped && si->prio != swap_info[next]->prio)) {
679 next = swap_list.head;
683 spin_lock(&si->lock);
684 if (!si->highest_bit) {
685 spin_unlock(&si->lock);
688 if (!(si->flags & SWP_WRITEOK)) {
689 spin_unlock(&si->lock);
693 swap_list.next = next;
695 spin_unlock(&swap_lock);
696 /* This is called for allocating swap entry for cache */
697 offset = scan_swap_map(si, SWAP_HAS_CACHE);
698 spin_unlock(&si->lock);
700 return swp_entry(type, offset);
701 spin_lock(&swap_lock);
702 next = swap_list.next;
705 atomic_long_inc(&nr_swap_pages);
707 spin_unlock(&swap_lock);
708 return (swp_entry_t) {0};
711 /* The only caller of this function is now suspend routine */
712 swp_entry_t get_swap_page_of_type(int type)
714 struct swap_info_struct *si;
717 si = swap_info[type];
718 spin_lock(&si->lock);
719 if (si && (si->flags & SWP_WRITEOK)) {
720 atomic_long_dec(&nr_swap_pages);
721 /* This is called for allocating swap entry, not cache */
722 offset = scan_swap_map(si, 1);
724 spin_unlock(&si->lock);
725 return swp_entry(type, offset);
727 atomic_long_inc(&nr_swap_pages);
729 spin_unlock(&si->lock);
730 return (swp_entry_t) {0};
733 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
735 struct swap_info_struct *p;
736 unsigned long offset, type;
740 type = swp_type(entry);
741 if (type >= nr_swapfiles)
744 if (!(p->flags & SWP_USED))
746 offset = swp_offset(entry);
747 if (offset >= p->max)
749 if (!p->swap_map[offset])
755 pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
758 pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
761 pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
764 pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
770 * This swap type frees swap entry, check if it is the highest priority swap
771 * type which just frees swap entry. get_swap_page() uses
772 * highest_priority_index to search highest priority swap type. The
773 * swap_info_struct.lock can't protect us if there are multiple swap types
774 * active, so we use atomic_cmpxchg.
776 static void set_highest_priority_index(int type)
778 int old_hp_index, new_hp_index;
781 old_hp_index = atomic_read(&highest_priority_index);
782 if (old_hp_index != -1 &&
783 swap_info[old_hp_index]->prio >= swap_info[type]->prio)
786 } while (atomic_cmpxchg(&highest_priority_index,
787 old_hp_index, new_hp_index) != old_hp_index);
790 static unsigned char swap_entry_free(struct swap_info_struct *p,
791 swp_entry_t entry, unsigned char usage)
793 unsigned long offset = swp_offset(entry);
795 unsigned char has_cache;
797 count = p->swap_map[offset];
798 has_cache = count & SWAP_HAS_CACHE;
799 count &= ~SWAP_HAS_CACHE;
801 if (usage == SWAP_HAS_CACHE) {
802 VM_BUG_ON(!has_cache);
804 } else if (count == SWAP_MAP_SHMEM) {
806 * Or we could insist on shmem.c using a special
807 * swap_shmem_free() and free_shmem_swap_and_cache()...
810 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
811 if (count == COUNT_CONTINUED) {
812 if (swap_count_continued(p, offset, count))
813 count = SWAP_MAP_MAX | COUNT_CONTINUED;
815 count = SWAP_MAP_MAX;
821 mem_cgroup_uncharge_swap(entry);
823 usage = count | has_cache;
824 p->swap_map[offset] = usage;
826 /* free if no reference */
828 dec_cluster_info_page(p, p->cluster_info, offset);
829 if (offset < p->lowest_bit)
830 p->lowest_bit = offset;
831 if (offset > p->highest_bit)
832 p->highest_bit = offset;
833 set_highest_priority_index(p->type);
834 atomic_long_inc(&nr_swap_pages);
836 frontswap_invalidate_page(p->type, offset);
837 if (p->flags & SWP_BLKDEV) {
838 struct gendisk *disk = p->bdev->bd_disk;
839 if (disk->fops->swap_slot_free_notify)
840 disk->fops->swap_slot_free_notify(p->bdev,
849 * Caller has made sure that the swap device corresponding to entry
850 * is still around or has not been recycled.
852 void swap_free(swp_entry_t entry)
854 struct swap_info_struct *p;
856 p = swap_info_get(entry);
858 swap_entry_free(p, entry, 1);
859 spin_unlock(&p->lock);
864 * Called after dropping swapcache to decrease refcnt to swap entries.
866 void swapcache_free(swp_entry_t entry, struct page *page)
868 struct swap_info_struct *p;
871 p = swap_info_get(entry);
873 count = swap_entry_free(p, entry, SWAP_HAS_CACHE);
875 mem_cgroup_uncharge_swapcache(page, entry, count != 0);
876 spin_unlock(&p->lock);
881 * How many references to page are currently swapped out?
882 * This does not give an exact answer when swap count is continued,
883 * but does include the high COUNT_CONTINUED flag to allow for that.
885 int page_swapcount(struct page *page)
888 struct swap_info_struct *p;
891 entry.val = page_private(page);
892 p = swap_info_get(entry);
894 count = swap_count(p->swap_map[swp_offset(entry)]);
895 spin_unlock(&p->lock);
901 * We can write to an anon page without COW if there are no other references
902 * to it. And as a side-effect, free up its swap: because the old content
903 * on disk will never be read, and seeking back there to write new content
904 * later would only waste time away from clustering.
906 int reuse_swap_page(struct page *page)
910 VM_BUG_ON_PAGE(!PageLocked(page), page);
911 if (unlikely(PageKsm(page)))
913 count = page_mapcount(page);
914 if (count <= 1 && PageSwapCache(page)) {
915 count += page_swapcount(page);
916 if (count == 1 && !PageWriteback(page)) {
917 delete_from_swap_cache(page);
925 * If swap is getting full, or if there are no more mappings of this page,
926 * then try_to_free_swap is called to free its swap space.
928 int try_to_free_swap(struct page *page)
930 VM_BUG_ON_PAGE(!PageLocked(page), page);
932 if (!PageSwapCache(page))
934 if (PageWriteback(page))
936 if (page_swapcount(page))
940 * Once hibernation has begun to create its image of memory,
941 * there's a danger that one of the calls to try_to_free_swap()
942 * - most probably a call from __try_to_reclaim_swap() while
943 * hibernation is allocating its own swap pages for the image,
944 * but conceivably even a call from memory reclaim - will free
945 * the swap from a page which has already been recorded in the
946 * image as a clean swapcache page, and then reuse its swap for
947 * another page of the image. On waking from hibernation, the
948 * original page might be freed under memory pressure, then
949 * later read back in from swap, now with the wrong data.
951 * Hibernation suspends storage while it is writing the image
952 * to disk so check that here.
954 if (pm_suspended_storage())
957 delete_from_swap_cache(page);
963 * Free the swap entry like above, but also try to
964 * free the page cache entry if it is the last user.
966 int free_swap_and_cache(swp_entry_t entry)
968 struct swap_info_struct *p;
969 struct page *page = NULL;
971 if (non_swap_entry(entry))
974 p = swap_info_get(entry);
976 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
977 page = find_get_page(swap_address_space(entry),
979 if (page && !trylock_page(page)) {
980 page_cache_release(page);
984 spin_unlock(&p->lock);
988 * Not mapped elsewhere, or swap space full? Free it!
989 * Also recheck PageSwapCache now page is locked (above).
991 if (PageSwapCache(page) && !PageWriteback(page) &&
992 (!page_mapped(page) || vm_swap_full())) {
993 delete_from_swap_cache(page);
997 page_cache_release(page);
1002 #ifdef CONFIG_HIBERNATION
1004 * Find the swap type that corresponds to given device (if any).
1006 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1007 * from 0, in which the swap header is expected to be located.
1009 * This is needed for the suspend to disk (aka swsusp).
1011 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1013 struct block_device *bdev = NULL;
1017 bdev = bdget(device);
1019 spin_lock(&swap_lock);
1020 for (type = 0; type < nr_swapfiles; type++) {
1021 struct swap_info_struct *sis = swap_info[type];
1023 if (!(sis->flags & SWP_WRITEOK))
1028 *bdev_p = bdgrab(sis->bdev);
1030 spin_unlock(&swap_lock);
1033 if (bdev == sis->bdev) {
1034 struct swap_extent *se = &sis->first_swap_extent;
1036 if (se->start_block == offset) {
1038 *bdev_p = bdgrab(sis->bdev);
1040 spin_unlock(&swap_lock);
1046 spin_unlock(&swap_lock);
1054 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1055 * corresponding to given index in swap_info (swap type).
1057 sector_t swapdev_block(int type, pgoff_t offset)
1059 struct block_device *bdev;
1061 if ((unsigned int)type >= nr_swapfiles)
1063 if (!(swap_info[type]->flags & SWP_WRITEOK))
1065 return map_swap_entry(swp_entry(type, offset), &bdev);
1069 * Return either the total number of swap pages of given type, or the number
1070 * of free pages of that type (depending on @free)
1072 * This is needed for software suspend
1074 unsigned int count_swap_pages(int type, int free)
1078 spin_lock(&swap_lock);
1079 if ((unsigned int)type < nr_swapfiles) {
1080 struct swap_info_struct *sis = swap_info[type];
1082 spin_lock(&sis->lock);
1083 if (sis->flags & SWP_WRITEOK) {
1086 n -= sis->inuse_pages;
1088 spin_unlock(&sis->lock);
1090 spin_unlock(&swap_lock);
1093 #endif /* CONFIG_HIBERNATION */
1095 static inline int maybe_same_pte(pte_t pte, pte_t swp_pte)
1097 #ifdef CONFIG_MEM_SOFT_DIRTY
1099 * When pte keeps soft dirty bit the pte generated
1100 * from swap entry does not has it, still it's same
1101 * pte from logical point of view.
1103 pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte);
1104 return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty);
1106 return pte_same(pte, swp_pte);
1111 * No need to decide whether this PTE shares the swap entry with others,
1112 * just let do_wp_page work it out if a write is requested later - to
1113 * force COW, vm_page_prot omits write permission from any private vma.
1115 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1116 unsigned long addr, swp_entry_t entry, struct page *page)
1118 struct page *swapcache;
1119 struct mem_cgroup *memcg;
1125 page = ksm_might_need_to_copy(page, vma, addr);
1126 if (unlikely(!page))
1129 if (mem_cgroup_try_charge_swapin(vma->vm_mm, page,
1130 GFP_KERNEL, &memcg)) {
1135 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1136 if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) {
1137 mem_cgroup_cancel_charge_swapin(memcg);
1142 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1143 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1145 set_pte_at(vma->vm_mm, addr, pte,
1146 pte_mkold(mk_pte(page, vma->vm_page_prot)));
1147 if (page == swapcache)
1148 page_add_anon_rmap(page, vma, addr);
1149 else /* ksm created a completely new copy */
1150 page_add_new_anon_rmap(page, vma, addr);
1151 mem_cgroup_commit_charge_swapin(page, memcg);
1154 * Move the page to the active list so it is not
1155 * immediately swapped out again after swapon.
1157 activate_page(page);
1159 pte_unmap_unlock(pte, ptl);
1161 if (page != swapcache) {
1168 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1169 unsigned long addr, unsigned long end,
1170 swp_entry_t entry, struct page *page)
1172 pte_t swp_pte = swp_entry_to_pte(entry);
1177 * We don't actually need pte lock while scanning for swp_pte: since
1178 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1179 * page table while we're scanning; though it could get zapped, and on
1180 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1181 * of unmatched parts which look like swp_pte, so unuse_pte must
1182 * recheck under pte lock. Scanning without pte lock lets it be
1183 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1185 pte = pte_offset_map(pmd, addr);
1188 * swapoff spends a _lot_ of time in this loop!
1189 * Test inline before going to call unuse_pte.
1191 if (unlikely(maybe_same_pte(*pte, swp_pte))) {
1193 ret = unuse_pte(vma, pmd, addr, entry, page);
1196 pte = pte_offset_map(pmd, addr);
1198 } while (pte++, addr += PAGE_SIZE, addr != end);
1204 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1205 unsigned long addr, unsigned long end,
1206 swp_entry_t entry, struct page *page)
1212 pmd = pmd_offset(pud, addr);
1214 next = pmd_addr_end(addr, end);
1215 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1217 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1220 } while (pmd++, addr = next, addr != end);
1224 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1225 unsigned long addr, unsigned long end,
1226 swp_entry_t entry, struct page *page)
1232 pud = pud_offset(pgd, addr);
1234 next = pud_addr_end(addr, end);
1235 if (pud_none_or_clear_bad(pud))
1237 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1240 } while (pud++, addr = next, addr != end);
1244 static int unuse_vma(struct vm_area_struct *vma,
1245 swp_entry_t entry, struct page *page)
1248 unsigned long addr, end, next;
1251 if (page_anon_vma(page)) {
1252 addr = page_address_in_vma(page, vma);
1253 if (addr == -EFAULT)
1256 end = addr + PAGE_SIZE;
1258 addr = vma->vm_start;
1262 pgd = pgd_offset(vma->vm_mm, addr);
1264 next = pgd_addr_end(addr, end);
1265 if (pgd_none_or_clear_bad(pgd))
1267 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1270 } while (pgd++, addr = next, addr != end);
1274 static int unuse_mm(struct mm_struct *mm,
1275 swp_entry_t entry, struct page *page)
1277 struct vm_area_struct *vma;
1280 if (!down_read_trylock(&mm->mmap_sem)) {
1282 * Activate page so shrink_inactive_list is unlikely to unmap
1283 * its ptes while lock is dropped, so swapoff can make progress.
1285 activate_page(page);
1287 down_read(&mm->mmap_sem);
1290 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1291 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
1294 up_read(&mm->mmap_sem);
1295 return (ret < 0)? ret: 0;
1299 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1300 * from current position to next entry still in use.
1301 * Recycle to start on reaching the end, returning 0 when empty.
1303 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1304 unsigned int prev, bool frontswap)
1306 unsigned int max = si->max;
1307 unsigned int i = prev;
1308 unsigned char count;
1311 * No need for swap_lock here: we're just looking
1312 * for whether an entry is in use, not modifying it; false
1313 * hits are okay, and sys_swapoff() has already prevented new
1314 * allocations from this area (while holding swap_lock).
1323 * No entries in use at top of swap_map,
1324 * loop back to start and recheck there.
1331 if (frontswap_test(si, i))
1336 count = ACCESS_ONCE(si->swap_map[i]);
1337 if (count && swap_count(count) != SWAP_MAP_BAD)
1344 * We completely avoid races by reading each swap page in advance,
1345 * and then search for the process using it. All the necessary
1346 * page table adjustments can then be made atomically.
1348 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1349 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1351 int try_to_unuse(unsigned int type, bool frontswap,
1352 unsigned long pages_to_unuse)
1354 struct swap_info_struct *si = swap_info[type];
1355 struct mm_struct *start_mm;
1356 volatile unsigned char *swap_map; /* swap_map is accessed without
1357 * locking. Mark it as volatile
1358 * to prevent compiler doing
1361 unsigned char swcount;
1368 * When searching mms for an entry, a good strategy is to
1369 * start at the first mm we freed the previous entry from
1370 * (though actually we don't notice whether we or coincidence
1371 * freed the entry). Initialize this start_mm with a hold.
1373 * A simpler strategy would be to start at the last mm we
1374 * freed the previous entry from; but that would take less
1375 * advantage of mmlist ordering, which clusters forked mms
1376 * together, child after parent. If we race with dup_mmap(), we
1377 * prefer to resolve parent before child, lest we miss entries
1378 * duplicated after we scanned child: using last mm would invert
1381 start_mm = &init_mm;
1382 atomic_inc(&init_mm.mm_users);
1385 * Keep on scanning until all entries have gone. Usually,
1386 * one pass through swap_map is enough, but not necessarily:
1387 * there are races when an instance of an entry might be missed.
1389 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
1390 if (signal_pending(current)) {
1396 * Get a page for the entry, using the existing swap
1397 * cache page if there is one. Otherwise, get a clean
1398 * page and read the swap into it.
1400 swap_map = &si->swap_map[i];
1401 entry = swp_entry(type, i);
1402 page = read_swap_cache_async(entry,
1403 GFP_HIGHUSER_MOVABLE, NULL, 0);
1406 * Either swap_duplicate() failed because entry
1407 * has been freed independently, and will not be
1408 * reused since sys_swapoff() already disabled
1409 * allocation from here, or alloc_page() failed.
1411 swcount = *swap_map;
1413 * We don't hold lock here, so the swap entry could be
1414 * SWAP_MAP_BAD (when the cluster is discarding).
1415 * Instead of fail out, We can just skip the swap
1416 * entry because swapoff will wait for discarding
1419 if (!swcount || swcount == SWAP_MAP_BAD)
1426 * Don't hold on to start_mm if it looks like exiting.
1428 if (atomic_read(&start_mm->mm_users) == 1) {
1430 start_mm = &init_mm;
1431 atomic_inc(&init_mm.mm_users);
1435 * Wait for and lock page. When do_swap_page races with
1436 * try_to_unuse, do_swap_page can handle the fault much
1437 * faster than try_to_unuse can locate the entry. This
1438 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1439 * defer to do_swap_page in such a case - in some tests,
1440 * do_swap_page and try_to_unuse repeatedly compete.
1442 wait_on_page_locked(page);
1443 wait_on_page_writeback(page);
1445 wait_on_page_writeback(page);
1448 * Remove all references to entry.
1450 swcount = *swap_map;
1451 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1452 retval = shmem_unuse(entry, page);
1453 /* page has already been unlocked and released */
1458 if (swap_count(swcount) && start_mm != &init_mm)
1459 retval = unuse_mm(start_mm, entry, page);
1461 if (swap_count(*swap_map)) {
1462 int set_start_mm = (*swap_map >= swcount);
1463 struct list_head *p = &start_mm->mmlist;
1464 struct mm_struct *new_start_mm = start_mm;
1465 struct mm_struct *prev_mm = start_mm;
1466 struct mm_struct *mm;
1468 atomic_inc(&new_start_mm->mm_users);
1469 atomic_inc(&prev_mm->mm_users);
1470 spin_lock(&mmlist_lock);
1471 while (swap_count(*swap_map) && !retval &&
1472 (p = p->next) != &start_mm->mmlist) {
1473 mm = list_entry(p, struct mm_struct, mmlist);
1474 if (!atomic_inc_not_zero(&mm->mm_users))
1476 spin_unlock(&mmlist_lock);
1482 swcount = *swap_map;
1483 if (!swap_count(swcount)) /* any usage ? */
1485 else if (mm == &init_mm)
1488 retval = unuse_mm(mm, entry, page);
1490 if (set_start_mm && *swap_map < swcount) {
1491 mmput(new_start_mm);
1492 atomic_inc(&mm->mm_users);
1496 spin_lock(&mmlist_lock);
1498 spin_unlock(&mmlist_lock);
1501 start_mm = new_start_mm;
1505 page_cache_release(page);
1510 * If a reference remains (rare), we would like to leave
1511 * the page in the swap cache; but try_to_unmap could
1512 * then re-duplicate the entry once we drop page lock,
1513 * so we might loop indefinitely; also, that page could
1514 * not be swapped out to other storage meanwhile. So:
1515 * delete from cache even if there's another reference,
1516 * after ensuring that the data has been saved to disk -
1517 * since if the reference remains (rarer), it will be
1518 * read from disk into another page. Splitting into two
1519 * pages would be incorrect if swap supported "shared
1520 * private" pages, but they are handled by tmpfs files.
1522 * Given how unuse_vma() targets one particular offset
1523 * in an anon_vma, once the anon_vma has been determined,
1524 * this splitting happens to be just what is needed to
1525 * handle where KSM pages have been swapped out: re-reading
1526 * is unnecessarily slow, but we can fix that later on.
1528 if (swap_count(*swap_map) &&
1529 PageDirty(page) && PageSwapCache(page)) {
1530 struct writeback_control wbc = {
1531 .sync_mode = WB_SYNC_NONE,
1534 swap_writepage(page, &wbc);
1536 wait_on_page_writeback(page);
1540 * It is conceivable that a racing task removed this page from
1541 * swap cache just before we acquired the page lock at the top,
1542 * or while we dropped it in unuse_mm(). The page might even
1543 * be back in swap cache on another swap area: that we must not
1544 * delete, since it may not have been written out to swap yet.
1546 if (PageSwapCache(page) &&
1547 likely(page_private(page) == entry.val))
1548 delete_from_swap_cache(page);
1551 * So we could skip searching mms once swap count went
1552 * to 1, we did not mark any present ptes as dirty: must
1553 * mark page dirty so shrink_page_list will preserve it.
1557 page_cache_release(page);
1560 * Make sure that we aren't completely killing
1561 * interactive performance.
1564 if (frontswap && pages_to_unuse > 0) {
1565 if (!--pages_to_unuse)
1575 * After a successful try_to_unuse, if no swap is now in use, we know
1576 * we can empty the mmlist. swap_lock must be held on entry and exit.
1577 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1578 * added to the mmlist just after page_duplicate - before would be racy.
1580 static void drain_mmlist(void)
1582 struct list_head *p, *next;
1585 for (type = 0; type < nr_swapfiles; type++)
1586 if (swap_info[type]->inuse_pages)
1588 spin_lock(&mmlist_lock);
1589 list_for_each_safe(p, next, &init_mm.mmlist)
1591 spin_unlock(&mmlist_lock);
1595 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1596 * corresponds to page offset for the specified swap entry.
1597 * Note that the type of this function is sector_t, but it returns page offset
1598 * into the bdev, not sector offset.
1600 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
1602 struct swap_info_struct *sis;
1603 struct swap_extent *start_se;
1604 struct swap_extent *se;
1607 sis = swap_info[swp_type(entry)];
1610 offset = swp_offset(entry);
1611 start_se = sis->curr_swap_extent;
1615 struct list_head *lh;
1617 if (se->start_page <= offset &&
1618 offset < (se->start_page + se->nr_pages)) {
1619 return se->start_block + (offset - se->start_page);
1622 se = list_entry(lh, struct swap_extent, list);
1623 sis->curr_swap_extent = se;
1624 BUG_ON(se == start_se); /* It *must* be present */
1629 * Returns the page offset into bdev for the specified page's swap entry.
1631 sector_t map_swap_page(struct page *page, struct block_device **bdev)
1634 entry.val = page_private(page);
1635 return map_swap_entry(entry, bdev);
1639 * Free all of a swapdev's extent information
1641 static void destroy_swap_extents(struct swap_info_struct *sis)
1643 while (!list_empty(&sis->first_swap_extent.list)) {
1644 struct swap_extent *se;
1646 se = list_entry(sis->first_swap_extent.list.next,
1647 struct swap_extent, list);
1648 list_del(&se->list);
1652 if (sis->flags & SWP_FILE) {
1653 struct file *swap_file = sis->swap_file;
1654 struct address_space *mapping = swap_file->f_mapping;
1656 sis->flags &= ~SWP_FILE;
1657 mapping->a_ops->swap_deactivate(swap_file);
1662 * Add a block range (and the corresponding page range) into this swapdev's
1663 * extent list. The extent list is kept sorted in page order.
1665 * This function rather assumes that it is called in ascending page order.
1668 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1669 unsigned long nr_pages, sector_t start_block)
1671 struct swap_extent *se;
1672 struct swap_extent *new_se;
1673 struct list_head *lh;
1675 if (start_page == 0) {
1676 se = &sis->first_swap_extent;
1677 sis->curr_swap_extent = se;
1679 se->nr_pages = nr_pages;
1680 se->start_block = start_block;
1683 lh = sis->first_swap_extent.list.prev; /* Highest extent */
1684 se = list_entry(lh, struct swap_extent, list);
1685 BUG_ON(se->start_page + se->nr_pages != start_page);
1686 if (se->start_block + se->nr_pages == start_block) {
1688 se->nr_pages += nr_pages;
1694 * No merge. Insert a new extent, preserving ordering.
1696 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1699 new_se->start_page = start_page;
1700 new_se->nr_pages = nr_pages;
1701 new_se->start_block = start_block;
1703 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
1708 * A `swap extent' is a simple thing which maps a contiguous range of pages
1709 * onto a contiguous range of disk blocks. An ordered list of swap extents
1710 * is built at swapon time and is then used at swap_writepage/swap_readpage
1711 * time for locating where on disk a page belongs.
1713 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1714 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1715 * swap files identically.
1717 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1718 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1719 * swapfiles are handled *identically* after swapon time.
1721 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1722 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1723 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1724 * requirements, they are simply tossed out - we will never use those blocks
1727 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1728 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1729 * which will scribble on the fs.
1731 * The amount of disk space which a single swap extent represents varies.
1732 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1733 * extents in the list. To avoid much list walking, we cache the previous
1734 * search location in `curr_swap_extent', and start new searches from there.
1735 * This is extremely effective. The average number of iterations in
1736 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1738 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1740 struct file *swap_file = sis->swap_file;
1741 struct address_space *mapping = swap_file->f_mapping;
1742 struct inode *inode = mapping->host;
1745 if (S_ISBLK(inode->i_mode)) {
1746 ret = add_swap_extent(sis, 0, sis->max, 0);
1751 if (mapping->a_ops->swap_activate) {
1752 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
1754 sis->flags |= SWP_FILE;
1755 ret = add_swap_extent(sis, 0, sis->max, 0);
1761 return generic_swapfile_activate(sis, swap_file, span);
1764 static void _enable_swap_info(struct swap_info_struct *p, int prio,
1765 unsigned char *swap_map,
1766 struct swap_cluster_info *cluster_info)
1773 p->prio = --least_priority;
1774 p->swap_map = swap_map;
1775 p->cluster_info = cluster_info;
1776 p->flags |= SWP_WRITEOK;
1777 atomic_long_add(p->pages, &nr_swap_pages);
1778 total_swap_pages += p->pages;
1780 /* insert swap space into swap_list: */
1782 for (i = swap_list.head; i >= 0; i = swap_info[i]->next) {
1783 if (p->prio >= swap_info[i]->prio)
1789 swap_list.head = swap_list.next = p->type;
1791 swap_info[prev]->next = p->type;
1794 static void enable_swap_info(struct swap_info_struct *p, int prio,
1795 unsigned char *swap_map,
1796 struct swap_cluster_info *cluster_info,
1797 unsigned long *frontswap_map)
1799 frontswap_init(p->type, frontswap_map);
1800 spin_lock(&swap_lock);
1801 spin_lock(&p->lock);
1802 _enable_swap_info(p, prio, swap_map, cluster_info);
1803 spin_unlock(&p->lock);
1804 spin_unlock(&swap_lock);
1807 static void reinsert_swap_info(struct swap_info_struct *p)
1809 spin_lock(&swap_lock);
1810 spin_lock(&p->lock);
1811 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
1812 spin_unlock(&p->lock);
1813 spin_unlock(&swap_lock);
1816 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
1818 struct swap_info_struct *p = NULL;
1819 unsigned char *swap_map;
1820 struct swap_cluster_info *cluster_info;
1821 unsigned long *frontswap_map;
1822 struct file *swap_file, *victim;
1823 struct address_space *mapping;
1824 struct inode *inode;
1825 struct filename *pathname;
1828 unsigned int old_block_size;
1830 if (!capable(CAP_SYS_ADMIN))
1833 BUG_ON(!current->mm);
1835 pathname = getname(specialfile);
1836 if (IS_ERR(pathname))
1837 return PTR_ERR(pathname);
1839 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
1840 err = PTR_ERR(victim);
1844 mapping = victim->f_mapping;
1846 spin_lock(&swap_lock);
1847 for (type = swap_list.head; type >= 0; type = swap_info[type]->next) {
1848 p = swap_info[type];
1849 if (p->flags & SWP_WRITEOK) {
1850 if (p->swap_file->f_mapping == mapping)
1857 spin_unlock(&swap_lock);
1860 if (!security_vm_enough_memory_mm(current->mm, p->pages))
1861 vm_unacct_memory(p->pages);
1864 spin_unlock(&swap_lock);
1868 swap_list.head = p->next;
1870 swap_info[prev]->next = p->next;
1871 if (type == swap_list.next) {
1872 /* just pick something that's safe... */
1873 swap_list.next = swap_list.head;
1875 spin_lock(&p->lock);
1877 for (i = p->next; i >= 0; i = swap_info[i]->next)
1878 swap_info[i]->prio = p->prio--;
1881 atomic_long_sub(p->pages, &nr_swap_pages);
1882 total_swap_pages -= p->pages;
1883 p->flags &= ~SWP_WRITEOK;
1884 spin_unlock(&p->lock);
1885 spin_unlock(&swap_lock);
1887 set_current_oom_origin();
1888 err = try_to_unuse(type, false, 0); /* force all pages to be unused */
1889 clear_current_oom_origin();
1892 /* re-insert swap space back into swap_list */
1893 reinsert_swap_info(p);
1897 flush_work(&p->discard_work);
1899 destroy_swap_extents(p);
1900 if (p->flags & SWP_CONTINUED)
1901 free_swap_count_continuations(p);
1903 mutex_lock(&swapon_mutex);
1904 spin_lock(&swap_lock);
1905 spin_lock(&p->lock);
1908 /* wait for anyone still in scan_swap_map */
1909 p->highest_bit = 0; /* cuts scans short */
1910 while (p->flags >= SWP_SCANNING) {
1911 spin_unlock(&p->lock);
1912 spin_unlock(&swap_lock);
1913 schedule_timeout_uninterruptible(1);
1914 spin_lock(&swap_lock);
1915 spin_lock(&p->lock);
1918 swap_file = p->swap_file;
1919 old_block_size = p->old_block_size;
1920 p->swap_file = NULL;
1922 swap_map = p->swap_map;
1924 cluster_info = p->cluster_info;
1925 p->cluster_info = NULL;
1926 frontswap_map = frontswap_map_get(p);
1927 spin_unlock(&p->lock);
1928 spin_unlock(&swap_lock);
1929 frontswap_invalidate_area(type);
1930 frontswap_map_set(p, NULL);
1931 mutex_unlock(&swapon_mutex);
1932 free_percpu(p->percpu_cluster);
1933 p->percpu_cluster = NULL;
1935 vfree(cluster_info);
1936 vfree(frontswap_map);
1937 /* Destroy swap account information */
1938 swap_cgroup_swapoff(type);
1940 inode = mapping->host;
1941 if (S_ISBLK(inode->i_mode)) {
1942 struct block_device *bdev = I_BDEV(inode);
1943 set_blocksize(bdev, old_block_size);
1944 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1946 mutex_lock(&inode->i_mutex);
1947 inode->i_flags &= ~S_SWAPFILE;
1948 mutex_unlock(&inode->i_mutex);
1950 filp_close(swap_file, NULL);
1953 * Clear the SWP_USED flag after all resources are freed so that swapon
1954 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1955 * not hold p->lock after we cleared its SWP_WRITEOK.
1957 spin_lock(&swap_lock);
1959 spin_unlock(&swap_lock);
1962 atomic_inc(&proc_poll_event);
1963 wake_up_interruptible(&proc_poll_wait);
1966 filp_close(victim, NULL);
1972 #ifdef CONFIG_PROC_FS
1973 static unsigned swaps_poll(struct file *file, poll_table *wait)
1975 struct seq_file *seq = file->private_data;
1977 poll_wait(file, &proc_poll_wait, wait);
1979 if (seq->poll_event != atomic_read(&proc_poll_event)) {
1980 seq->poll_event = atomic_read(&proc_poll_event);
1981 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
1984 return POLLIN | POLLRDNORM;
1988 static void *swap_start(struct seq_file *swap, loff_t *pos)
1990 struct swap_info_struct *si;
1994 mutex_lock(&swapon_mutex);
1997 return SEQ_START_TOKEN;
1999 for (type = 0; type < nr_swapfiles; type++) {
2000 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2001 si = swap_info[type];
2002 if (!(si->flags & SWP_USED) || !si->swap_map)
2011 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2013 struct swap_info_struct *si = v;
2016 if (v == SEQ_START_TOKEN)
2019 type = si->type + 1;
2021 for (; type < nr_swapfiles; type++) {
2022 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2023 si = swap_info[type];
2024 if (!(si->flags & SWP_USED) || !si->swap_map)
2033 static void swap_stop(struct seq_file *swap, void *v)
2035 mutex_unlock(&swapon_mutex);
2038 static int swap_show(struct seq_file *swap, void *v)
2040 struct swap_info_struct *si = v;
2044 if (si == SEQ_START_TOKEN) {
2045 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2049 file = si->swap_file;
2050 len = seq_path(swap, &file->f_path, " \t\n\\");
2051 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2052 len < 40 ? 40 - len : 1, " ",
2053 S_ISBLK(file_inode(file)->i_mode) ?
2054 "partition" : "file\t",
2055 si->pages << (PAGE_SHIFT - 10),
2056 si->inuse_pages << (PAGE_SHIFT - 10),
2061 static const struct seq_operations swaps_op = {
2062 .start = swap_start,
2068 static int swaps_open(struct inode *inode, struct file *file)
2070 struct seq_file *seq;
2073 ret = seq_open(file, &swaps_op);
2077 seq = file->private_data;
2078 seq->poll_event = atomic_read(&proc_poll_event);
2082 static const struct file_operations proc_swaps_operations = {
2085 .llseek = seq_lseek,
2086 .release = seq_release,
2090 static int __init procswaps_init(void)
2092 proc_create("swaps", 0, NULL, &proc_swaps_operations);
2095 __initcall(procswaps_init);
2096 #endif /* CONFIG_PROC_FS */
2098 #ifdef MAX_SWAPFILES_CHECK
2099 static int __init max_swapfiles_check(void)
2101 MAX_SWAPFILES_CHECK();
2104 late_initcall(max_swapfiles_check);
2107 static struct swap_info_struct *alloc_swap_info(void)
2109 struct swap_info_struct *p;
2112 p = kzalloc(sizeof(*p), GFP_KERNEL);
2114 return ERR_PTR(-ENOMEM);
2116 spin_lock(&swap_lock);
2117 for (type = 0; type < nr_swapfiles; type++) {
2118 if (!(swap_info[type]->flags & SWP_USED))
2121 if (type >= MAX_SWAPFILES) {
2122 spin_unlock(&swap_lock);
2124 return ERR_PTR(-EPERM);
2126 if (type >= nr_swapfiles) {
2128 swap_info[type] = p;
2130 * Write swap_info[type] before nr_swapfiles, in case a
2131 * racing procfs swap_start() or swap_next() is reading them.
2132 * (We never shrink nr_swapfiles, we never free this entry.)
2138 p = swap_info[type];
2140 * Do not memset this entry: a racing procfs swap_next()
2141 * would be relying on p->type to remain valid.
2144 INIT_LIST_HEAD(&p->first_swap_extent.list);
2145 p->flags = SWP_USED;
2147 spin_unlock(&swap_lock);
2148 spin_lock_init(&p->lock);
2153 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2157 if (S_ISBLK(inode->i_mode)) {
2158 p->bdev = bdgrab(I_BDEV(inode));
2159 error = blkdev_get(p->bdev,
2160 FMODE_READ | FMODE_WRITE | FMODE_EXCL,
2166 p->old_block_size = block_size(p->bdev);
2167 error = set_blocksize(p->bdev, PAGE_SIZE);
2170 p->flags |= SWP_BLKDEV;
2171 } else if (S_ISREG(inode->i_mode)) {
2172 p->bdev = inode->i_sb->s_bdev;
2173 mutex_lock(&inode->i_mutex);
2174 if (IS_SWAPFILE(inode))
2182 static unsigned long read_swap_header(struct swap_info_struct *p,
2183 union swap_header *swap_header,
2184 struct inode *inode)
2187 unsigned long maxpages;
2188 unsigned long swapfilepages;
2189 unsigned long last_page;
2191 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2192 pr_err("Unable to find swap-space signature\n");
2196 /* swap partition endianess hack... */
2197 if (swab32(swap_header->info.version) == 1) {
2198 swab32s(&swap_header->info.version);
2199 swab32s(&swap_header->info.last_page);
2200 swab32s(&swap_header->info.nr_badpages);
2201 for (i = 0; i < swap_header->info.nr_badpages; i++)
2202 swab32s(&swap_header->info.badpages[i]);
2204 /* Check the swap header's sub-version */
2205 if (swap_header->info.version != 1) {
2206 pr_warn("Unable to handle swap header version %d\n",
2207 swap_header->info.version);
2212 p->cluster_next = 1;
2216 * Find out how many pages are allowed for a single swap
2217 * device. There are two limiting factors: 1) the number
2218 * of bits for the swap offset in the swp_entry_t type, and
2219 * 2) the number of bits in the swap pte as defined by the
2220 * different architectures. In order to find the
2221 * largest possible bit mask, a swap entry with swap type 0
2222 * and swap offset ~0UL is created, encoded to a swap pte,
2223 * decoded to a swp_entry_t again, and finally the swap
2224 * offset is extracted. This will mask all the bits from
2225 * the initial ~0UL mask that can't be encoded in either
2226 * the swp_entry_t or the architecture definition of a
2229 maxpages = swp_offset(pte_to_swp_entry(
2230 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2231 last_page = swap_header->info.last_page;
2232 if (last_page > maxpages) {
2233 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2234 maxpages << (PAGE_SHIFT - 10),
2235 last_page << (PAGE_SHIFT - 10));
2237 if (maxpages > last_page) {
2238 maxpages = last_page + 1;
2239 /* p->max is an unsigned int: don't overflow it */
2240 if ((unsigned int)maxpages == 0)
2241 maxpages = UINT_MAX;
2243 p->highest_bit = maxpages - 1;
2247 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2248 if (swapfilepages && maxpages > swapfilepages) {
2249 pr_warn("Swap area shorter than signature indicates\n");
2252 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2254 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2260 static int setup_swap_map_and_extents(struct swap_info_struct *p,
2261 union swap_header *swap_header,
2262 unsigned char *swap_map,
2263 struct swap_cluster_info *cluster_info,
2264 unsigned long maxpages,
2268 unsigned int nr_good_pages;
2270 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2271 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
2273 nr_good_pages = maxpages - 1; /* omit header page */
2275 cluster_set_null(&p->free_cluster_head);
2276 cluster_set_null(&p->free_cluster_tail);
2277 cluster_set_null(&p->discard_cluster_head);
2278 cluster_set_null(&p->discard_cluster_tail);
2280 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2281 unsigned int page_nr = swap_header->info.badpages[i];
2282 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2284 if (page_nr < maxpages) {
2285 swap_map[page_nr] = SWAP_MAP_BAD;
2288 * Haven't marked the cluster free yet, no list
2289 * operation involved
2291 inc_cluster_info_page(p, cluster_info, page_nr);
2295 /* Haven't marked the cluster free yet, no list operation involved */
2296 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2297 inc_cluster_info_page(p, cluster_info, i);
2299 if (nr_good_pages) {
2300 swap_map[0] = SWAP_MAP_BAD;
2302 * Not mark the cluster free yet, no list
2303 * operation involved
2305 inc_cluster_info_page(p, cluster_info, 0);
2307 p->pages = nr_good_pages;
2308 nr_extents = setup_swap_extents(p, span);
2311 nr_good_pages = p->pages;
2313 if (!nr_good_pages) {
2314 pr_warn("Empty swap-file\n");
2321 for (i = 0; i < nr_clusters; i++) {
2322 if (!cluster_count(&cluster_info[idx])) {
2323 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2324 if (cluster_is_null(&p->free_cluster_head)) {
2325 cluster_set_next_flag(&p->free_cluster_head,
2327 cluster_set_next_flag(&p->free_cluster_tail,
2332 tail = cluster_next(&p->free_cluster_tail);
2333 cluster_set_next(&cluster_info[tail], idx);
2334 cluster_set_next_flag(&p->free_cluster_tail,
2339 if (idx == nr_clusters)
2346 * Helper to sys_swapon determining if a given swap
2347 * backing device queue supports DISCARD operations.
2349 static bool swap_discardable(struct swap_info_struct *si)
2351 struct request_queue *q = bdev_get_queue(si->bdev);
2353 if (!q || !blk_queue_discard(q))
2359 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2361 struct swap_info_struct *p;
2362 struct filename *name;
2363 struct file *swap_file = NULL;
2364 struct address_space *mapping;
2368 union swap_header *swap_header;
2371 unsigned long maxpages;
2372 unsigned char *swap_map = NULL;
2373 struct swap_cluster_info *cluster_info = NULL;
2374 unsigned long *frontswap_map = NULL;
2375 struct page *page = NULL;
2376 struct inode *inode = NULL;
2378 if (swap_flags & ~SWAP_FLAGS_VALID)
2381 if (!capable(CAP_SYS_ADMIN))
2384 p = alloc_swap_info();
2388 INIT_WORK(&p->discard_work, swap_discard_work);
2390 name = getname(specialfile);
2392 error = PTR_ERR(name);
2396 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
2397 if (IS_ERR(swap_file)) {
2398 error = PTR_ERR(swap_file);
2403 p->swap_file = swap_file;
2404 mapping = swap_file->f_mapping;
2406 for (i = 0; i < nr_swapfiles; i++) {
2407 struct swap_info_struct *q = swap_info[i];
2409 if (q == p || !q->swap_file)
2411 if (mapping == q->swap_file->f_mapping) {
2417 inode = mapping->host;
2418 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
2419 error = claim_swapfile(p, inode);
2420 if (unlikely(error))
2424 * Read the swap header.
2426 if (!mapping->a_ops->readpage) {
2430 page = read_mapping_page(mapping, 0, swap_file);
2432 error = PTR_ERR(page);
2435 swap_header = kmap(page);
2437 maxpages = read_swap_header(p, swap_header, inode);
2438 if (unlikely(!maxpages)) {
2443 /* OK, set up the swap map and apply the bad block list */
2444 swap_map = vzalloc(maxpages);
2449 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2450 p->flags |= SWP_SOLIDSTATE;
2452 * select a random position to start with to help wear leveling
2455 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2457 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2458 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2459 if (!cluster_info) {
2463 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2464 if (!p->percpu_cluster) {
2468 for_each_possible_cpu(i) {
2469 struct percpu_cluster *cluster;
2470 cluster = per_cpu_ptr(p->percpu_cluster, i);
2471 cluster_set_null(&cluster->index);
2475 error = swap_cgroup_swapon(p->type, maxpages);
2479 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2480 cluster_info, maxpages, &span);
2481 if (unlikely(nr_extents < 0)) {
2485 /* frontswap enabled? set up bit-per-page map for frontswap */
2486 if (frontswap_enabled)
2487 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
2489 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2491 * When discard is enabled for swap with no particular
2492 * policy flagged, we set all swap discard flags here in
2493 * order to sustain backward compatibility with older
2494 * swapon(8) releases.
2496 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2500 * By flagging sys_swapon, a sysadmin can tell us to
2501 * either do single-time area discards only, or to just
2502 * perform discards for released swap page-clusters.
2503 * Now it's time to adjust the p->flags accordingly.
2505 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2506 p->flags &= ~SWP_PAGE_DISCARD;
2507 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2508 p->flags &= ~SWP_AREA_DISCARD;
2510 /* issue a swapon-time discard if it's still required */
2511 if (p->flags & SWP_AREA_DISCARD) {
2512 int err = discard_swap(p);
2514 pr_err("swapon: discard_swap(%p): %d\n",
2519 mutex_lock(&swapon_mutex);
2521 if (swap_flags & SWAP_FLAG_PREFER)
2523 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2524 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
2526 pr_info("Adding %uk swap on %s. "
2527 "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2528 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
2529 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2530 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2531 (p->flags & SWP_DISCARDABLE) ? "D" : "",
2532 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2533 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
2534 (frontswap_map) ? "FS" : "");
2536 mutex_unlock(&swapon_mutex);
2537 atomic_inc(&proc_poll_event);
2538 wake_up_interruptible(&proc_poll_wait);
2540 if (S_ISREG(inode->i_mode))
2541 inode->i_flags |= S_SWAPFILE;
2545 free_percpu(p->percpu_cluster);
2546 p->percpu_cluster = NULL;
2547 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
2548 set_blocksize(p->bdev, p->old_block_size);
2549 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2551 destroy_swap_extents(p);
2552 swap_cgroup_swapoff(p->type);
2553 spin_lock(&swap_lock);
2554 p->swap_file = NULL;
2556 spin_unlock(&swap_lock);
2558 vfree(cluster_info);
2560 if (inode && S_ISREG(inode->i_mode)) {
2561 mutex_unlock(&inode->i_mutex);
2564 filp_close(swap_file, NULL);
2567 if (page && !IS_ERR(page)) {
2569 page_cache_release(page);
2573 if (inode && S_ISREG(inode->i_mode))
2574 mutex_unlock(&inode->i_mutex);
2578 void si_swapinfo(struct sysinfo *val)
2581 unsigned long nr_to_be_unused = 0;
2583 spin_lock(&swap_lock);
2584 for (type = 0; type < nr_swapfiles; type++) {
2585 struct swap_info_struct *si = swap_info[type];
2587 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2588 nr_to_be_unused += si->inuse_pages;
2590 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
2591 val->totalswap = total_swap_pages + nr_to_be_unused;
2592 spin_unlock(&swap_lock);
2596 * Verify that a swap entry is valid and increment its swap map count.
2598 * Returns error code in following case.
2600 * - swp_entry is invalid -> EINVAL
2601 * - swp_entry is migration entry -> EINVAL
2602 * - swap-cache reference is requested but there is already one. -> EEXIST
2603 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2604 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2606 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
2608 struct swap_info_struct *p;
2609 unsigned long offset, type;
2610 unsigned char count;
2611 unsigned char has_cache;
2614 if (non_swap_entry(entry))
2617 type = swp_type(entry);
2618 if (type >= nr_swapfiles)
2620 p = swap_info[type];
2621 offset = swp_offset(entry);
2623 spin_lock(&p->lock);
2624 if (unlikely(offset >= p->max))
2627 count = p->swap_map[offset];
2630 * swapin_readahead() doesn't check if a swap entry is valid, so the
2631 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2633 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2638 has_cache = count & SWAP_HAS_CACHE;
2639 count &= ~SWAP_HAS_CACHE;
2642 if (usage == SWAP_HAS_CACHE) {
2644 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2645 if (!has_cache && count)
2646 has_cache = SWAP_HAS_CACHE;
2647 else if (has_cache) /* someone else added cache */
2649 else /* no users remaining */
2652 } else if (count || has_cache) {
2654 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2656 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
2658 else if (swap_count_continued(p, offset, count))
2659 count = COUNT_CONTINUED;
2663 err = -ENOENT; /* unused swap entry */
2665 p->swap_map[offset] = count | has_cache;
2668 spin_unlock(&p->lock);
2673 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
2678 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2679 * (in which case its reference count is never incremented).
2681 void swap_shmem_alloc(swp_entry_t entry)
2683 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2687 * Increase reference count of swap entry by 1.
2688 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2689 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2690 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2691 * might occur if a page table entry has got corrupted.
2693 int swap_duplicate(swp_entry_t entry)
2697 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2698 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2703 * @entry: swap entry for which we allocate swap cache.
2705 * Called when allocating swap cache for existing swap entry,
2706 * This can return error codes. Returns 0 at success.
2707 * -EBUSY means there is a swap cache.
2708 * Note: return code is different from swap_duplicate().
2710 int swapcache_prepare(swp_entry_t entry)
2712 return __swap_duplicate(entry, SWAP_HAS_CACHE);
2715 struct swap_info_struct *page_swap_info(struct page *page)
2717 swp_entry_t swap = { .val = page_private(page) };
2718 BUG_ON(!PageSwapCache(page));
2719 return swap_info[swp_type(swap)];
2723 * out-of-line __page_file_ methods to avoid include hell.
2725 struct address_space *__page_file_mapping(struct page *page)
2727 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
2728 return page_swap_info(page)->swap_file->f_mapping;
2730 EXPORT_SYMBOL_GPL(__page_file_mapping);
2732 pgoff_t __page_file_index(struct page *page)
2734 swp_entry_t swap = { .val = page_private(page) };
2735 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
2736 return swp_offset(swap);
2738 EXPORT_SYMBOL_GPL(__page_file_index);
2741 * add_swap_count_continuation - called when a swap count is duplicated
2742 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2743 * page of the original vmalloc'ed swap_map, to hold the continuation count
2744 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2745 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2747 * These continuation pages are seldom referenced: the common paths all work
2748 * on the original swap_map, only referring to a continuation page when the
2749 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2751 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2752 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2753 * can be called after dropping locks.
2755 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2757 struct swap_info_struct *si;
2760 struct page *list_page;
2762 unsigned char count;
2765 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2766 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2768 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2770 si = swap_info_get(entry);
2773 * An acceptable race has occurred since the failing
2774 * __swap_duplicate(): the swap entry has been freed,
2775 * perhaps even the whole swap_map cleared for swapoff.
2780 offset = swp_offset(entry);
2781 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2783 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2785 * The higher the swap count, the more likely it is that tasks
2786 * will race to add swap count continuation: we need to avoid
2787 * over-provisioning.
2793 spin_unlock(&si->lock);
2798 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2799 * no architecture is using highmem pages for kernel page tables: so it
2800 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
2802 head = vmalloc_to_page(si->swap_map + offset);
2803 offset &= ~PAGE_MASK;
2806 * Page allocation does not initialize the page's lru field,
2807 * but it does always reset its private field.
2809 if (!page_private(head)) {
2810 BUG_ON(count & COUNT_CONTINUED);
2811 INIT_LIST_HEAD(&head->lru);
2812 set_page_private(head, SWP_CONTINUED);
2813 si->flags |= SWP_CONTINUED;
2816 list_for_each_entry(list_page, &head->lru, lru) {
2820 * If the previous map said no continuation, but we've found
2821 * a continuation page, free our allocation and use this one.
2823 if (!(count & COUNT_CONTINUED))
2826 map = kmap_atomic(list_page) + offset;
2831 * If this continuation count now has some space in it,
2832 * free our allocation and use this one.
2834 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2838 list_add_tail(&page->lru, &head->lru);
2839 page = NULL; /* now it's attached, don't free it */
2841 spin_unlock(&si->lock);
2849 * swap_count_continued - when the original swap_map count is incremented
2850 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2851 * into, carry if so, or else fail until a new continuation page is allocated;
2852 * when the original swap_map count is decremented from 0 with continuation,
2853 * borrow from the continuation and report whether it still holds more.
2854 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2856 static bool swap_count_continued(struct swap_info_struct *si,
2857 pgoff_t offset, unsigned char count)
2863 head = vmalloc_to_page(si->swap_map + offset);
2864 if (page_private(head) != SWP_CONTINUED) {
2865 BUG_ON(count & COUNT_CONTINUED);
2866 return false; /* need to add count continuation */
2869 offset &= ~PAGE_MASK;
2870 page = list_entry(head->lru.next, struct page, lru);
2871 map = kmap_atomic(page) + offset;
2873 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2874 goto init_map; /* jump over SWAP_CONT_MAX checks */
2876 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2878 * Think of how you add 1 to 999
2880 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
2882 page = list_entry(page->lru.next, struct page, lru);
2883 BUG_ON(page == head);
2884 map = kmap_atomic(page) + offset;
2886 if (*map == SWAP_CONT_MAX) {
2888 page = list_entry(page->lru.next, struct page, lru);
2890 return false; /* add count continuation */
2891 map = kmap_atomic(page) + offset;
2892 init_map: *map = 0; /* we didn't zero the page */
2896 page = list_entry(page->lru.prev, struct page, lru);
2897 while (page != head) {
2898 map = kmap_atomic(page) + offset;
2899 *map = COUNT_CONTINUED;
2901 page = list_entry(page->lru.prev, struct page, lru);
2903 return true; /* incremented */
2905 } else { /* decrementing */
2907 * Think of how you subtract 1 from 1000
2909 BUG_ON(count != COUNT_CONTINUED);
2910 while (*map == COUNT_CONTINUED) {
2912 page = list_entry(page->lru.next, struct page, lru);
2913 BUG_ON(page == head);
2914 map = kmap_atomic(page) + offset;
2921 page = list_entry(page->lru.prev, struct page, lru);
2922 while (page != head) {
2923 map = kmap_atomic(page) + offset;
2924 *map = SWAP_CONT_MAX | count;
2925 count = COUNT_CONTINUED;
2927 page = list_entry(page->lru.prev, struct page, lru);
2929 return count == COUNT_CONTINUED;
2934 * free_swap_count_continuations - swapoff free all the continuation pages
2935 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2937 static void free_swap_count_continuations(struct swap_info_struct *si)
2941 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2943 head = vmalloc_to_page(si->swap_map + offset);
2944 if (page_private(head)) {
2945 struct list_head *this, *next;
2946 list_for_each_safe(this, next, &head->lru) {
2948 page = list_entry(this, struct page, lru);