4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
9 #include <linux/hugetlb.h>
10 #include <linux/mman.h>
11 #include <linux/slab.h>
12 #include <linux/kernel_stat.h>
13 #include <linux/swap.h>
14 #include <linux/vmalloc.h>
15 #include <linux/pagemap.h>
16 #include <linux/namei.h>
17 #include <linux/shmem_fs.h>
18 #include <linux/blkdev.h>
19 #include <linux/random.h>
20 #include <linux/writeback.h>
21 #include <linux/proc_fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/init.h>
24 #include <linux/ksm.h>
25 #include <linux/rmap.h>
26 #include <linux/security.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mutex.h>
29 #include <linux/capability.h>
30 #include <linux/syscalls.h>
31 #include <linux/memcontrol.h>
32 #include <linux/poll.h>
33 #include <linux/oom.h>
34 #include <linux/frontswap.h>
35 #include <linux/swapfile.h>
36 #include <linux/export.h>
38 #include <asm/pgtable.h>
39 #include <asm/tlbflush.h>
40 #include <linux/swapops.h>
41 #include <linux/swap_cgroup.h>
43 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
45 static void free_swap_count_continuations(struct swap_info_struct *);
46 static sector_t map_swap_entry(swp_entry_t, struct block_device**);
48 DEFINE_SPINLOCK(swap_lock);
49 static unsigned int nr_swapfiles;
50 atomic_long_t nr_swap_pages;
51 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
52 long total_swap_pages;
53 static int least_priority;
55 static const char Bad_file[] = "Bad swap file entry ";
56 static const char Unused_file[] = "Unused swap file entry ";
57 static const char Bad_offset[] = "Bad swap offset entry ";
58 static const char Unused_offset[] = "Unused swap offset entry ";
61 * all active swap_info_structs
62 * protected with swap_lock, and ordered by priority.
64 PLIST_HEAD(swap_active_head);
67 * all available (active, not full) swap_info_structs
68 * protected with swap_avail_lock, ordered by priority.
69 * This is used by get_swap_page() instead of swap_active_head
70 * because swap_active_head includes all swap_info_structs,
71 * but get_swap_page() doesn't need to look at full ones.
72 * This uses its own lock instead of swap_lock because when a
73 * swap_info_struct changes between not-full/full, it needs to
74 * add/remove itself to/from this list, but the swap_info_struct->lock
75 * is held and the locking order requires swap_lock to be taken
76 * before any swap_info_struct->lock.
78 static PLIST_HEAD(swap_avail_head);
79 static DEFINE_SPINLOCK(swap_avail_lock);
81 struct swap_info_struct *swap_info[MAX_SWAPFILES];
83 static DEFINE_MUTEX(swapon_mutex);
85 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
86 /* Activity counter to indicate that a swapon or swapoff has occurred */
87 static atomic_t proc_poll_event = ATOMIC_INIT(0);
89 static inline unsigned char swap_count(unsigned char ent)
91 return ent & ~SWAP_HAS_CACHE; /* may include SWAP_HAS_CONT flag */
94 /* returns 1 if swap entry is freed */
96 __try_to_reclaim_swap(struct swap_info_struct *si, unsigned long offset)
98 swp_entry_t entry = swp_entry(si->type, offset);
102 page = find_get_page(swap_address_space(entry), entry.val);
106 * This function is called from scan_swap_map() and it's called
107 * by vmscan.c at reclaiming pages. So, we hold a lock on a page, here.
108 * We have to use trylock for avoiding deadlock. This is a special
109 * case and you should use try_to_free_swap() with explicit lock_page()
110 * in usual operations.
112 if (trylock_page(page)) {
113 ret = try_to_free_swap(page);
116 page_cache_release(page);
121 * swapon tell device that all the old swap contents can be discarded,
122 * to allow the swap device to optimize its wear-levelling.
124 static int discard_swap(struct swap_info_struct *si)
126 struct swap_extent *se;
127 sector_t start_block;
131 /* Do not discard the swap header page! */
132 se = &si->first_swap_extent;
133 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
134 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
136 err = blkdev_issue_discard(si->bdev, start_block,
137 nr_blocks, GFP_KERNEL, 0);
143 list_for_each_entry(se, &si->first_swap_extent.list, list) {
144 start_block = se->start_block << (PAGE_SHIFT - 9);
145 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
147 err = blkdev_issue_discard(si->bdev, start_block,
148 nr_blocks, GFP_KERNEL, 0);
154 return err; /* That will often be -EOPNOTSUPP */
158 * swap allocation tell device that a cluster of swap can now be discarded,
159 * to allow the swap device to optimize its wear-levelling.
161 static void discard_swap_cluster(struct swap_info_struct *si,
162 pgoff_t start_page, pgoff_t nr_pages)
164 struct swap_extent *se = si->curr_swap_extent;
165 int found_extent = 0;
168 struct list_head *lh;
170 if (se->start_page <= start_page &&
171 start_page < se->start_page + se->nr_pages) {
172 pgoff_t offset = start_page - se->start_page;
173 sector_t start_block = se->start_block + offset;
174 sector_t nr_blocks = se->nr_pages - offset;
176 if (nr_blocks > nr_pages)
177 nr_blocks = nr_pages;
178 start_page += nr_blocks;
179 nr_pages -= nr_blocks;
182 si->curr_swap_extent = se;
184 start_block <<= PAGE_SHIFT - 9;
185 nr_blocks <<= PAGE_SHIFT - 9;
186 if (blkdev_issue_discard(si->bdev, start_block,
187 nr_blocks, GFP_NOIO, 0))
192 se = list_entry(lh, struct swap_extent, list);
196 #define SWAPFILE_CLUSTER 256
197 #define LATENCY_LIMIT 256
199 static inline void cluster_set_flag(struct swap_cluster_info *info,
205 static inline unsigned int cluster_count(struct swap_cluster_info *info)
210 static inline void cluster_set_count(struct swap_cluster_info *info,
216 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
217 unsigned int c, unsigned int f)
223 static inline unsigned int cluster_next(struct swap_cluster_info *info)
228 static inline void cluster_set_next(struct swap_cluster_info *info,
234 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
235 unsigned int n, unsigned int f)
241 static inline bool cluster_is_free(struct swap_cluster_info *info)
243 return info->flags & CLUSTER_FLAG_FREE;
246 static inline bool cluster_is_null(struct swap_cluster_info *info)
248 return info->flags & CLUSTER_FLAG_NEXT_NULL;
251 static inline void cluster_set_null(struct swap_cluster_info *info)
253 info->flags = CLUSTER_FLAG_NEXT_NULL;
257 /* Add a cluster to discard list and schedule it to do discard */
258 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
262 * If scan_swap_map() can't find a free cluster, it will check
263 * si->swap_map directly. To make sure the discarding cluster isn't
264 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
265 * will be cleared after discard
267 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
268 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
270 if (cluster_is_null(&si->discard_cluster_head)) {
271 cluster_set_next_flag(&si->discard_cluster_head,
273 cluster_set_next_flag(&si->discard_cluster_tail,
276 unsigned int tail = cluster_next(&si->discard_cluster_tail);
277 cluster_set_next(&si->cluster_info[tail], idx);
278 cluster_set_next_flag(&si->discard_cluster_tail,
282 schedule_work(&si->discard_work);
286 * Doing discard actually. After a cluster discard is finished, the cluster
287 * will be added to free cluster list. caller should hold si->lock.
289 static void swap_do_scheduled_discard(struct swap_info_struct *si)
291 struct swap_cluster_info *info;
294 info = si->cluster_info;
296 while (!cluster_is_null(&si->discard_cluster_head)) {
297 idx = cluster_next(&si->discard_cluster_head);
299 cluster_set_next_flag(&si->discard_cluster_head,
300 cluster_next(&info[idx]), 0);
301 if (cluster_next(&si->discard_cluster_tail) == idx) {
302 cluster_set_null(&si->discard_cluster_head);
303 cluster_set_null(&si->discard_cluster_tail);
305 spin_unlock(&si->lock);
307 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
310 spin_lock(&si->lock);
311 cluster_set_flag(&info[idx], CLUSTER_FLAG_FREE);
312 if (cluster_is_null(&si->free_cluster_head)) {
313 cluster_set_next_flag(&si->free_cluster_head,
315 cluster_set_next_flag(&si->free_cluster_tail,
320 tail = cluster_next(&si->free_cluster_tail);
321 cluster_set_next(&info[tail], idx);
322 cluster_set_next_flag(&si->free_cluster_tail,
325 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
326 0, SWAPFILE_CLUSTER);
330 static void swap_discard_work(struct work_struct *work)
332 struct swap_info_struct *si;
334 si = container_of(work, struct swap_info_struct, discard_work);
336 spin_lock(&si->lock);
337 swap_do_scheduled_discard(si);
338 spin_unlock(&si->lock);
342 * The cluster corresponding to page_nr will be used. The cluster will be
343 * removed from free cluster list and its usage counter will be increased.
345 static void inc_cluster_info_page(struct swap_info_struct *p,
346 struct swap_cluster_info *cluster_info, unsigned long page_nr)
348 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
352 if (cluster_is_free(&cluster_info[idx])) {
353 VM_BUG_ON(cluster_next(&p->free_cluster_head) != idx);
354 cluster_set_next_flag(&p->free_cluster_head,
355 cluster_next(&cluster_info[idx]), 0);
356 if (cluster_next(&p->free_cluster_tail) == idx) {
357 cluster_set_null(&p->free_cluster_tail);
358 cluster_set_null(&p->free_cluster_head);
360 cluster_set_count_flag(&cluster_info[idx], 0, 0);
363 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
364 cluster_set_count(&cluster_info[idx],
365 cluster_count(&cluster_info[idx]) + 1);
369 * The cluster corresponding to page_nr decreases one usage. If the usage
370 * counter becomes 0, which means no page in the cluster is in using, we can
371 * optionally discard the cluster and add it to free cluster list.
373 static void dec_cluster_info_page(struct swap_info_struct *p,
374 struct swap_cluster_info *cluster_info, unsigned long page_nr)
376 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
381 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
382 cluster_set_count(&cluster_info[idx],
383 cluster_count(&cluster_info[idx]) - 1);
385 if (cluster_count(&cluster_info[idx]) == 0) {
387 * If the swap is discardable, prepare discard the cluster
388 * instead of free it immediately. The cluster will be freed
391 if ((p->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
392 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
393 swap_cluster_schedule_discard(p, idx);
397 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
398 if (cluster_is_null(&p->free_cluster_head)) {
399 cluster_set_next_flag(&p->free_cluster_head, idx, 0);
400 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
402 unsigned int tail = cluster_next(&p->free_cluster_tail);
403 cluster_set_next(&cluster_info[tail], idx);
404 cluster_set_next_flag(&p->free_cluster_tail, idx, 0);
410 * It's possible scan_swap_map() uses a free cluster in the middle of free
411 * cluster list. Avoiding such abuse to avoid list corruption.
414 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
415 unsigned long offset)
417 struct percpu_cluster *percpu_cluster;
420 offset /= SWAPFILE_CLUSTER;
421 conflict = !cluster_is_null(&si->free_cluster_head) &&
422 offset != cluster_next(&si->free_cluster_head) &&
423 cluster_is_free(&si->cluster_info[offset]);
428 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
429 cluster_set_null(&percpu_cluster->index);
434 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
435 * might involve allocating a new cluster for current CPU too.
437 static void scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
438 unsigned long *offset, unsigned long *scan_base)
440 struct percpu_cluster *cluster;
445 cluster = this_cpu_ptr(si->percpu_cluster);
446 if (cluster_is_null(&cluster->index)) {
447 if (!cluster_is_null(&si->free_cluster_head)) {
448 cluster->index = si->free_cluster_head;
449 cluster->next = cluster_next(&cluster->index) *
451 } else if (!cluster_is_null(&si->discard_cluster_head)) {
453 * we don't have free cluster but have some clusters in
454 * discarding, do discard now and reclaim them
456 swap_do_scheduled_discard(si);
457 *scan_base = *offset = si->cluster_next;
466 * Other CPUs can use our cluster if they can't find a free cluster,
467 * check if there is still free entry in the cluster
470 while (tmp < si->max && tmp < (cluster_next(&cluster->index) + 1) *
472 if (!si->swap_map[tmp]) {
479 cluster_set_null(&cluster->index);
482 cluster->next = tmp + 1;
487 static unsigned long scan_swap_map(struct swap_info_struct *si,
490 unsigned long offset;
491 unsigned long scan_base;
492 unsigned long last_in_cluster = 0;
493 int latency_ration = LATENCY_LIMIT;
496 * We try to cluster swap pages by allocating them sequentially
497 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
498 * way, however, we resort to first-free allocation, starting
499 * a new cluster. This prevents us from scattering swap pages
500 * all over the entire swap partition, so that we reduce
501 * overall disk seek times between swap pages. -- sct
502 * But we do now try to find an empty cluster. -Andrea
503 * And we let swap pages go all over an SSD partition. Hugh
506 si->flags += SWP_SCANNING;
507 scan_base = offset = si->cluster_next;
510 if (si->cluster_info) {
511 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
515 if (unlikely(!si->cluster_nr--)) {
516 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
517 si->cluster_nr = SWAPFILE_CLUSTER - 1;
521 spin_unlock(&si->lock);
524 * If seek is expensive, start searching for new cluster from
525 * start of partition, to minimize the span of allocated swap.
526 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
527 * case, just handled by scan_swap_map_try_ssd_cluster() above.
529 scan_base = offset = si->lowest_bit;
530 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
532 /* Locate the first empty (unaligned) cluster */
533 for (; last_in_cluster <= si->highest_bit; offset++) {
534 if (si->swap_map[offset])
535 last_in_cluster = offset + SWAPFILE_CLUSTER;
536 else if (offset == last_in_cluster) {
537 spin_lock(&si->lock);
538 offset -= SWAPFILE_CLUSTER - 1;
539 si->cluster_next = offset;
540 si->cluster_nr = SWAPFILE_CLUSTER - 1;
543 if (unlikely(--latency_ration < 0)) {
545 latency_ration = LATENCY_LIMIT;
550 spin_lock(&si->lock);
551 si->cluster_nr = SWAPFILE_CLUSTER - 1;
555 if (si->cluster_info) {
556 while (scan_swap_map_ssd_cluster_conflict(si, offset))
557 scan_swap_map_try_ssd_cluster(si, &offset, &scan_base);
559 if (!(si->flags & SWP_WRITEOK))
561 if (!si->highest_bit)
563 if (offset > si->highest_bit)
564 scan_base = offset = si->lowest_bit;
566 /* reuse swap entry of cache-only swap if not busy. */
567 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
569 spin_unlock(&si->lock);
570 swap_was_freed = __try_to_reclaim_swap(si, offset);
571 spin_lock(&si->lock);
572 /* entry was freed successfully, try to use this again */
575 goto scan; /* check next one */
578 if (si->swap_map[offset])
581 if (offset == si->lowest_bit)
583 if (offset == si->highest_bit)
586 if (si->inuse_pages == si->pages) {
587 si->lowest_bit = si->max;
589 spin_lock(&swap_avail_lock);
590 plist_del(&si->avail_list, &swap_avail_head);
591 spin_unlock(&swap_avail_lock);
593 si->swap_map[offset] = usage;
594 inc_cluster_info_page(si, si->cluster_info, offset);
595 si->cluster_next = offset + 1;
596 si->flags -= SWP_SCANNING;
601 spin_unlock(&si->lock);
602 while (++offset <= si->highest_bit) {
603 if (!si->swap_map[offset]) {
604 spin_lock(&si->lock);
607 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
608 spin_lock(&si->lock);
611 if (unlikely(--latency_ration < 0)) {
613 latency_ration = LATENCY_LIMIT;
616 offset = si->lowest_bit;
617 while (offset < scan_base) {
618 if (!si->swap_map[offset]) {
619 spin_lock(&si->lock);
622 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
623 spin_lock(&si->lock);
626 if (unlikely(--latency_ration < 0)) {
628 latency_ration = LATENCY_LIMIT;
632 spin_lock(&si->lock);
635 si->flags -= SWP_SCANNING;
639 swp_entry_t get_swap_page(void)
641 struct swap_info_struct *si, *next;
644 if (atomic_long_read(&nr_swap_pages) <= 0)
646 atomic_long_dec(&nr_swap_pages);
648 spin_lock(&swap_avail_lock);
651 plist_for_each_entry_safe(si, next, &swap_avail_head, avail_list) {
652 /* requeue si to after same-priority siblings */
653 plist_requeue(&si->avail_list, &swap_avail_head);
654 spin_unlock(&swap_avail_lock);
655 spin_lock(&si->lock);
656 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
657 spin_lock(&swap_avail_lock);
658 if (plist_node_empty(&si->avail_list)) {
659 spin_unlock(&si->lock);
662 WARN(!si->highest_bit,
663 "swap_info %d in list but !highest_bit\n",
665 WARN(!(si->flags & SWP_WRITEOK),
666 "swap_info %d in list but !SWP_WRITEOK\n",
668 plist_del(&si->avail_list, &swap_avail_head);
669 spin_unlock(&si->lock);
673 /* This is called for allocating swap entry for cache */
674 offset = scan_swap_map(si, SWAP_HAS_CACHE);
675 spin_unlock(&si->lock);
677 return swp_entry(si->type, offset);
678 pr_debug("scan_swap_map of si %d failed to find offset\n",
680 spin_lock(&swap_avail_lock);
683 * if we got here, it's likely that si was almost full before,
684 * and since scan_swap_map() can drop the si->lock, multiple
685 * callers probably all tried to get a page from the same si
686 * and it filled up before we could get one; or, the si filled
687 * up between us dropping swap_avail_lock and taking si->lock.
688 * Since we dropped the swap_avail_lock, the swap_avail_head
689 * list may have been modified; so if next is still in the
690 * swap_avail_head list then try it, otherwise start over.
692 if (plist_node_empty(&next->avail_list))
696 spin_unlock(&swap_avail_lock);
698 atomic_long_inc(&nr_swap_pages);
700 return (swp_entry_t) {0};
703 /* The only caller of this function is now suspend routine */
704 swp_entry_t get_swap_page_of_type(int type)
706 struct swap_info_struct *si;
709 si = swap_info[type];
710 spin_lock(&si->lock);
711 if (si && (si->flags & SWP_WRITEOK)) {
712 atomic_long_dec(&nr_swap_pages);
713 /* This is called for allocating swap entry, not cache */
714 offset = scan_swap_map(si, 1);
716 spin_unlock(&si->lock);
717 return swp_entry(type, offset);
719 atomic_long_inc(&nr_swap_pages);
721 spin_unlock(&si->lock);
722 return (swp_entry_t) {0};
725 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
727 struct swap_info_struct *p;
728 unsigned long offset, type;
732 type = swp_type(entry);
733 if (type >= nr_swapfiles)
736 if (!(p->flags & SWP_USED))
738 offset = swp_offset(entry);
739 if (offset >= p->max)
741 if (!p->swap_map[offset])
747 pr_err("swap_free: %s%08lx\n", Unused_offset, entry.val);
750 pr_err("swap_free: %s%08lx\n", Bad_offset, entry.val);
753 pr_err("swap_free: %s%08lx\n", Unused_file, entry.val);
756 pr_err("swap_free: %s%08lx\n", Bad_file, entry.val);
761 static unsigned char swap_entry_free(struct swap_info_struct *p,
762 swp_entry_t entry, unsigned char usage)
764 unsigned long offset = swp_offset(entry);
766 unsigned char has_cache;
768 count = p->swap_map[offset];
769 has_cache = count & SWAP_HAS_CACHE;
770 count &= ~SWAP_HAS_CACHE;
772 if (usage == SWAP_HAS_CACHE) {
773 VM_BUG_ON(!has_cache);
775 } else if (count == SWAP_MAP_SHMEM) {
777 * Or we could insist on shmem.c using a special
778 * swap_shmem_free() and free_shmem_swap_and_cache()...
781 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
782 if (count == COUNT_CONTINUED) {
783 if (swap_count_continued(p, offset, count))
784 count = SWAP_MAP_MAX | COUNT_CONTINUED;
786 count = SWAP_MAP_MAX;
792 mem_cgroup_uncharge_swap(entry);
794 usage = count | has_cache;
795 p->swap_map[offset] = usage;
797 /* free if no reference */
799 dec_cluster_info_page(p, p->cluster_info, offset);
800 if (offset < p->lowest_bit)
801 p->lowest_bit = offset;
802 if (offset > p->highest_bit) {
803 bool was_full = !p->highest_bit;
804 p->highest_bit = offset;
805 if (was_full && (p->flags & SWP_WRITEOK)) {
806 spin_lock(&swap_avail_lock);
807 WARN_ON(!plist_node_empty(&p->avail_list));
808 if (plist_node_empty(&p->avail_list))
809 plist_add(&p->avail_list,
811 spin_unlock(&swap_avail_lock);
814 atomic_long_inc(&nr_swap_pages);
816 frontswap_invalidate_page(p->type, offset);
817 if (p->flags & SWP_BLKDEV) {
818 struct gendisk *disk = p->bdev->bd_disk;
819 if (disk->fops->swap_slot_free_notify)
820 disk->fops->swap_slot_free_notify(p->bdev,
829 * Caller has made sure that the swap device corresponding to entry
830 * is still around or has not been recycled.
832 void swap_free(swp_entry_t entry)
834 struct swap_info_struct *p;
836 p = swap_info_get(entry);
838 swap_entry_free(p, entry, 1);
839 spin_unlock(&p->lock);
844 * Called after dropping swapcache to decrease refcnt to swap entries.
846 void swapcache_free(swp_entry_t entry)
848 struct swap_info_struct *p;
850 p = swap_info_get(entry);
852 swap_entry_free(p, entry, SWAP_HAS_CACHE);
853 spin_unlock(&p->lock);
858 * How many references to page are currently swapped out?
859 * This does not give an exact answer when swap count is continued,
860 * but does include the high COUNT_CONTINUED flag to allow for that.
862 int page_swapcount(struct page *page)
865 struct swap_info_struct *p;
868 entry.val = page_private(page);
869 p = swap_info_get(entry);
871 count = swap_count(p->swap_map[swp_offset(entry)]);
872 spin_unlock(&p->lock);
878 * How many references to @entry are currently swapped out?
879 * This considers COUNT_CONTINUED so it returns exact answer.
881 int swp_swapcount(swp_entry_t entry)
883 int count, tmp_count, n;
884 struct swap_info_struct *p;
889 p = swap_info_get(entry);
893 count = swap_count(p->swap_map[swp_offset(entry)]);
894 if (!(count & COUNT_CONTINUED))
897 count &= ~COUNT_CONTINUED;
898 n = SWAP_MAP_MAX + 1;
900 offset = swp_offset(entry);
901 page = vmalloc_to_page(p->swap_map + offset);
902 offset &= ~PAGE_MASK;
903 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
906 page = list_entry(page->lru.next, struct page, lru);
907 map = kmap_atomic(page);
908 tmp_count = map[offset];
911 count += (tmp_count & ~COUNT_CONTINUED) * n;
912 n *= (SWAP_CONT_MAX + 1);
913 } while (tmp_count & COUNT_CONTINUED);
915 spin_unlock(&p->lock);
920 * We can write to an anon page without COW if there are no other references
921 * to it. And as a side-effect, free up its swap: because the old content
922 * on disk will never be read, and seeking back there to write new content
923 * later would only waste time away from clustering.
925 int reuse_swap_page(struct page *page)
929 VM_BUG_ON_PAGE(!PageLocked(page), page);
930 if (unlikely(PageKsm(page)))
932 /* The page is part of THP and cannot be reused */
933 if (PageTransCompound(page))
935 count = page_mapcount(page);
936 if (count <= 1 && PageSwapCache(page)) {
937 count += page_swapcount(page);
938 if (count == 1 && !PageWriteback(page)) {
939 delete_from_swap_cache(page);
947 * If swap is getting full, or if there are no more mappings of this page,
948 * then try_to_free_swap is called to free its swap space.
950 int try_to_free_swap(struct page *page)
952 VM_BUG_ON_PAGE(!PageLocked(page), page);
954 if (!PageSwapCache(page))
956 if (PageWriteback(page))
958 if (page_swapcount(page))
962 * Once hibernation has begun to create its image of memory,
963 * there's a danger that one of the calls to try_to_free_swap()
964 * - most probably a call from __try_to_reclaim_swap() while
965 * hibernation is allocating its own swap pages for the image,
966 * but conceivably even a call from memory reclaim - will free
967 * the swap from a page which has already been recorded in the
968 * image as a clean swapcache page, and then reuse its swap for
969 * another page of the image. On waking from hibernation, the
970 * original page might be freed under memory pressure, then
971 * later read back in from swap, now with the wrong data.
973 * Hibernation suspends storage while it is writing the image
974 * to disk so check that here.
976 if (pm_suspended_storage())
979 delete_from_swap_cache(page);
985 * Free the swap entry like above, but also try to
986 * free the page cache entry if it is the last user.
988 int free_swap_and_cache(swp_entry_t entry)
990 struct swap_info_struct *p;
991 struct page *page = NULL;
993 if (non_swap_entry(entry))
996 p = swap_info_get(entry);
998 if (swap_entry_free(p, entry, 1) == SWAP_HAS_CACHE) {
999 page = find_get_page(swap_address_space(entry),
1001 if (page && !trylock_page(page)) {
1002 page_cache_release(page);
1006 spin_unlock(&p->lock);
1010 * Not mapped elsewhere, or swap space full? Free it!
1011 * Also recheck PageSwapCache now page is locked (above).
1013 if (PageSwapCache(page) && !PageWriteback(page) &&
1014 (!page_mapped(page) || vm_swap_full())) {
1015 delete_from_swap_cache(page);
1019 page_cache_release(page);
1024 #ifdef CONFIG_HIBERNATION
1026 * Find the swap type that corresponds to given device (if any).
1028 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1029 * from 0, in which the swap header is expected to be located.
1031 * This is needed for the suspend to disk (aka swsusp).
1033 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1035 struct block_device *bdev = NULL;
1039 bdev = bdget(device);
1041 spin_lock(&swap_lock);
1042 for (type = 0; type < nr_swapfiles; type++) {
1043 struct swap_info_struct *sis = swap_info[type];
1045 if (!(sis->flags & SWP_WRITEOK))
1050 *bdev_p = bdgrab(sis->bdev);
1052 spin_unlock(&swap_lock);
1055 if (bdev == sis->bdev) {
1056 struct swap_extent *se = &sis->first_swap_extent;
1058 if (se->start_block == offset) {
1060 *bdev_p = bdgrab(sis->bdev);
1062 spin_unlock(&swap_lock);
1068 spin_unlock(&swap_lock);
1076 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1077 * corresponding to given index in swap_info (swap type).
1079 sector_t swapdev_block(int type, pgoff_t offset)
1081 struct block_device *bdev;
1083 if ((unsigned int)type >= nr_swapfiles)
1085 if (!(swap_info[type]->flags & SWP_WRITEOK))
1087 return map_swap_entry(swp_entry(type, offset), &bdev);
1091 * Return either the total number of swap pages of given type, or the number
1092 * of free pages of that type (depending on @free)
1094 * This is needed for software suspend
1096 unsigned int count_swap_pages(int type, int free)
1100 spin_lock(&swap_lock);
1101 if ((unsigned int)type < nr_swapfiles) {
1102 struct swap_info_struct *sis = swap_info[type];
1104 spin_lock(&sis->lock);
1105 if (sis->flags & SWP_WRITEOK) {
1108 n -= sis->inuse_pages;
1110 spin_unlock(&sis->lock);
1112 spin_unlock(&swap_lock);
1115 #endif /* CONFIG_HIBERNATION */
1117 static inline int maybe_same_pte(pte_t pte, pte_t swp_pte)
1119 #ifdef CONFIG_MEM_SOFT_DIRTY
1121 * When pte keeps soft dirty bit the pte generated
1122 * from swap entry does not has it, still it's same
1123 * pte from logical point of view.
1125 pte_t swp_pte_dirty = pte_swp_mksoft_dirty(swp_pte);
1126 return pte_same(pte, swp_pte) || pte_same(pte, swp_pte_dirty);
1128 return pte_same(pte, swp_pte);
1133 * No need to decide whether this PTE shares the swap entry with others,
1134 * just let do_wp_page work it out if a write is requested later - to
1135 * force COW, vm_page_prot omits write permission from any private vma.
1137 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1138 unsigned long addr, swp_entry_t entry, struct page *page)
1140 struct page *swapcache;
1141 struct mem_cgroup *memcg;
1147 page = ksm_might_need_to_copy(page, vma, addr);
1148 if (unlikely(!page))
1151 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1157 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1158 if (unlikely(!maybe_same_pte(*pte, swp_entry_to_pte(entry)))) {
1159 mem_cgroup_cancel_charge(page, memcg, false);
1164 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1165 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1167 set_pte_at(vma->vm_mm, addr, pte,
1168 pte_mkold(mk_pte(page, vma->vm_page_prot)));
1169 if (page == swapcache) {
1170 page_add_anon_rmap(page, vma, addr, false);
1171 mem_cgroup_commit_charge(page, memcg, true, false);
1172 } else { /* ksm created a completely new copy */
1173 page_add_new_anon_rmap(page, vma, addr, false);
1174 mem_cgroup_commit_charge(page, memcg, false, false);
1175 lru_cache_add_active_or_unevictable(page, vma);
1179 * Move the page to the active list so it is not
1180 * immediately swapped out again after swapon.
1182 activate_page(page);
1184 pte_unmap_unlock(pte, ptl);
1186 if (page != swapcache) {
1193 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1194 unsigned long addr, unsigned long end,
1195 swp_entry_t entry, struct page *page)
1197 pte_t swp_pte = swp_entry_to_pte(entry);
1202 * We don't actually need pte lock while scanning for swp_pte: since
1203 * we hold page lock and mmap_sem, swp_pte cannot be inserted into the
1204 * page table while we're scanning; though it could get zapped, and on
1205 * some architectures (e.g. x86_32 with PAE) we might catch a glimpse
1206 * of unmatched parts which look like swp_pte, so unuse_pte must
1207 * recheck under pte lock. Scanning without pte lock lets it be
1208 * preemptable whenever CONFIG_PREEMPT but not CONFIG_HIGHPTE.
1210 pte = pte_offset_map(pmd, addr);
1213 * swapoff spends a _lot_ of time in this loop!
1214 * Test inline before going to call unuse_pte.
1216 if (unlikely(maybe_same_pte(*pte, swp_pte))) {
1218 ret = unuse_pte(vma, pmd, addr, entry, page);
1221 pte = pte_offset_map(pmd, addr);
1223 } while (pte++, addr += PAGE_SIZE, addr != end);
1229 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1230 unsigned long addr, unsigned long end,
1231 swp_entry_t entry, struct page *page)
1237 pmd = pmd_offset(pud, addr);
1239 next = pmd_addr_end(addr, end);
1240 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1242 ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
1245 } while (pmd++, addr = next, addr != end);
1249 static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
1250 unsigned long addr, unsigned long end,
1251 swp_entry_t entry, struct page *page)
1257 pud = pud_offset(pgd, addr);
1259 next = pud_addr_end(addr, end);
1260 if (pud_none_or_clear_bad(pud))
1262 ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
1265 } while (pud++, addr = next, addr != end);
1269 static int unuse_vma(struct vm_area_struct *vma,
1270 swp_entry_t entry, struct page *page)
1273 unsigned long addr, end, next;
1276 if (page_anon_vma(page)) {
1277 addr = page_address_in_vma(page, vma);
1278 if (addr == -EFAULT)
1281 end = addr + PAGE_SIZE;
1283 addr = vma->vm_start;
1287 pgd = pgd_offset(vma->vm_mm, addr);
1289 next = pgd_addr_end(addr, end);
1290 if (pgd_none_or_clear_bad(pgd))
1292 ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
1295 } while (pgd++, addr = next, addr != end);
1299 static int unuse_mm(struct mm_struct *mm,
1300 swp_entry_t entry, struct page *page)
1302 struct vm_area_struct *vma;
1305 if (!down_read_trylock(&mm->mmap_sem)) {
1307 * Activate page so shrink_inactive_list is unlikely to unmap
1308 * its ptes while lock is dropped, so swapoff can make progress.
1310 activate_page(page);
1312 down_read(&mm->mmap_sem);
1315 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1316 if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
1319 up_read(&mm->mmap_sem);
1320 return (ret < 0)? ret: 0;
1324 * Scan swap_map (or frontswap_map if frontswap parameter is true)
1325 * from current position to next entry still in use.
1326 * Recycle to start on reaching the end, returning 0 when empty.
1328 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
1329 unsigned int prev, bool frontswap)
1331 unsigned int max = si->max;
1332 unsigned int i = prev;
1333 unsigned char count;
1336 * No need for swap_lock here: we're just looking
1337 * for whether an entry is in use, not modifying it; false
1338 * hits are okay, and sys_swapoff() has already prevented new
1339 * allocations from this area (while holding swap_lock).
1348 * No entries in use at top of swap_map,
1349 * loop back to start and recheck there.
1356 if (frontswap_test(si, i))
1361 count = READ_ONCE(si->swap_map[i]);
1362 if (count && swap_count(count) != SWAP_MAP_BAD)
1369 * We completely avoid races by reading each swap page in advance,
1370 * and then search for the process using it. All the necessary
1371 * page table adjustments can then be made atomically.
1373 * if the boolean frontswap is true, only unuse pages_to_unuse pages;
1374 * pages_to_unuse==0 means all pages; ignored if frontswap is false
1376 int try_to_unuse(unsigned int type, bool frontswap,
1377 unsigned long pages_to_unuse)
1379 struct swap_info_struct *si = swap_info[type];
1380 struct mm_struct *start_mm;
1381 volatile unsigned char *swap_map; /* swap_map is accessed without
1382 * locking. Mark it as volatile
1383 * to prevent compiler doing
1386 unsigned char swcount;
1393 * When searching mms for an entry, a good strategy is to
1394 * start at the first mm we freed the previous entry from
1395 * (though actually we don't notice whether we or coincidence
1396 * freed the entry). Initialize this start_mm with a hold.
1398 * A simpler strategy would be to start at the last mm we
1399 * freed the previous entry from; but that would take less
1400 * advantage of mmlist ordering, which clusters forked mms
1401 * together, child after parent. If we race with dup_mmap(), we
1402 * prefer to resolve parent before child, lest we miss entries
1403 * duplicated after we scanned child: using last mm would invert
1406 start_mm = &init_mm;
1407 atomic_inc(&init_mm.mm_users);
1410 * Keep on scanning until all entries have gone. Usually,
1411 * one pass through swap_map is enough, but not necessarily:
1412 * there are races when an instance of an entry might be missed.
1414 while ((i = find_next_to_unuse(si, i, frontswap)) != 0) {
1415 if (signal_pending(current)) {
1421 * Get a page for the entry, using the existing swap
1422 * cache page if there is one. Otherwise, get a clean
1423 * page and read the swap into it.
1425 swap_map = &si->swap_map[i];
1426 entry = swp_entry(type, i);
1427 page = read_swap_cache_async(entry,
1428 GFP_HIGHUSER_MOVABLE, NULL, 0);
1431 * Either swap_duplicate() failed because entry
1432 * has been freed independently, and will not be
1433 * reused since sys_swapoff() already disabled
1434 * allocation from here, or alloc_page() failed.
1436 swcount = *swap_map;
1438 * We don't hold lock here, so the swap entry could be
1439 * SWAP_MAP_BAD (when the cluster is discarding).
1440 * Instead of fail out, We can just skip the swap
1441 * entry because swapoff will wait for discarding
1444 if (!swcount || swcount == SWAP_MAP_BAD)
1451 * Don't hold on to start_mm if it looks like exiting.
1453 if (atomic_read(&start_mm->mm_users) == 1) {
1455 start_mm = &init_mm;
1456 atomic_inc(&init_mm.mm_users);
1460 * Wait for and lock page. When do_swap_page races with
1461 * try_to_unuse, do_swap_page can handle the fault much
1462 * faster than try_to_unuse can locate the entry. This
1463 * apparently redundant "wait_on_page_locked" lets try_to_unuse
1464 * defer to do_swap_page in such a case - in some tests,
1465 * do_swap_page and try_to_unuse repeatedly compete.
1467 wait_on_page_locked(page);
1468 wait_on_page_writeback(page);
1470 wait_on_page_writeback(page);
1473 * Remove all references to entry.
1475 swcount = *swap_map;
1476 if (swap_count(swcount) == SWAP_MAP_SHMEM) {
1477 retval = shmem_unuse(entry, page);
1478 /* page has already been unlocked and released */
1483 if (swap_count(swcount) && start_mm != &init_mm)
1484 retval = unuse_mm(start_mm, entry, page);
1486 if (swap_count(*swap_map)) {
1487 int set_start_mm = (*swap_map >= swcount);
1488 struct list_head *p = &start_mm->mmlist;
1489 struct mm_struct *new_start_mm = start_mm;
1490 struct mm_struct *prev_mm = start_mm;
1491 struct mm_struct *mm;
1493 atomic_inc(&new_start_mm->mm_users);
1494 atomic_inc(&prev_mm->mm_users);
1495 spin_lock(&mmlist_lock);
1496 while (swap_count(*swap_map) && !retval &&
1497 (p = p->next) != &start_mm->mmlist) {
1498 mm = list_entry(p, struct mm_struct, mmlist);
1499 if (!atomic_inc_not_zero(&mm->mm_users))
1501 spin_unlock(&mmlist_lock);
1507 swcount = *swap_map;
1508 if (!swap_count(swcount)) /* any usage ? */
1510 else if (mm == &init_mm)
1513 retval = unuse_mm(mm, entry, page);
1515 if (set_start_mm && *swap_map < swcount) {
1516 mmput(new_start_mm);
1517 atomic_inc(&mm->mm_users);
1521 spin_lock(&mmlist_lock);
1523 spin_unlock(&mmlist_lock);
1526 start_mm = new_start_mm;
1530 page_cache_release(page);
1535 * If a reference remains (rare), we would like to leave
1536 * the page in the swap cache; but try_to_unmap could
1537 * then re-duplicate the entry once we drop page lock,
1538 * so we might loop indefinitely; also, that page could
1539 * not be swapped out to other storage meanwhile. So:
1540 * delete from cache even if there's another reference,
1541 * after ensuring that the data has been saved to disk -
1542 * since if the reference remains (rarer), it will be
1543 * read from disk into another page. Splitting into two
1544 * pages would be incorrect if swap supported "shared
1545 * private" pages, but they are handled by tmpfs files.
1547 * Given how unuse_vma() targets one particular offset
1548 * in an anon_vma, once the anon_vma has been determined,
1549 * this splitting happens to be just what is needed to
1550 * handle where KSM pages have been swapped out: re-reading
1551 * is unnecessarily slow, but we can fix that later on.
1553 if (swap_count(*swap_map) &&
1554 PageDirty(page) && PageSwapCache(page)) {
1555 struct writeback_control wbc = {
1556 .sync_mode = WB_SYNC_NONE,
1559 swap_writepage(page, &wbc);
1561 wait_on_page_writeback(page);
1565 * It is conceivable that a racing task removed this page from
1566 * swap cache just before we acquired the page lock at the top,
1567 * or while we dropped it in unuse_mm(). The page might even
1568 * be back in swap cache on another swap area: that we must not
1569 * delete, since it may not have been written out to swap yet.
1571 if (PageSwapCache(page) &&
1572 likely(page_private(page) == entry.val))
1573 delete_from_swap_cache(page);
1576 * So we could skip searching mms once swap count went
1577 * to 1, we did not mark any present ptes as dirty: must
1578 * mark page dirty so shrink_page_list will preserve it.
1582 page_cache_release(page);
1585 * Make sure that we aren't completely killing
1586 * interactive performance.
1589 if (frontswap && pages_to_unuse > 0) {
1590 if (!--pages_to_unuse)
1600 * After a successful try_to_unuse, if no swap is now in use, we know
1601 * we can empty the mmlist. swap_lock must be held on entry and exit.
1602 * Note that mmlist_lock nests inside swap_lock, and an mm must be
1603 * added to the mmlist just after page_duplicate - before would be racy.
1605 static void drain_mmlist(void)
1607 struct list_head *p, *next;
1610 for (type = 0; type < nr_swapfiles; type++)
1611 if (swap_info[type]->inuse_pages)
1613 spin_lock(&mmlist_lock);
1614 list_for_each_safe(p, next, &init_mm.mmlist)
1616 spin_unlock(&mmlist_lock);
1620 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
1621 * corresponds to page offset for the specified swap entry.
1622 * Note that the type of this function is sector_t, but it returns page offset
1623 * into the bdev, not sector offset.
1625 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
1627 struct swap_info_struct *sis;
1628 struct swap_extent *start_se;
1629 struct swap_extent *se;
1632 sis = swap_info[swp_type(entry)];
1635 offset = swp_offset(entry);
1636 start_se = sis->curr_swap_extent;
1640 struct list_head *lh;
1642 if (se->start_page <= offset &&
1643 offset < (se->start_page + se->nr_pages)) {
1644 return se->start_block + (offset - se->start_page);
1647 se = list_entry(lh, struct swap_extent, list);
1648 sis->curr_swap_extent = se;
1649 BUG_ON(se == start_se); /* It *must* be present */
1654 * Returns the page offset into bdev for the specified page's swap entry.
1656 sector_t map_swap_page(struct page *page, struct block_device **bdev)
1659 entry.val = page_private(page);
1660 return map_swap_entry(entry, bdev);
1664 * Free all of a swapdev's extent information
1666 static void destroy_swap_extents(struct swap_info_struct *sis)
1668 while (!list_empty(&sis->first_swap_extent.list)) {
1669 struct swap_extent *se;
1671 se = list_entry(sis->first_swap_extent.list.next,
1672 struct swap_extent, list);
1673 list_del(&se->list);
1677 if (sis->flags & SWP_FILE) {
1678 struct file *swap_file = sis->swap_file;
1679 struct address_space *mapping = swap_file->f_mapping;
1681 sis->flags &= ~SWP_FILE;
1682 mapping->a_ops->swap_deactivate(swap_file);
1687 * Add a block range (and the corresponding page range) into this swapdev's
1688 * extent list. The extent list is kept sorted in page order.
1690 * This function rather assumes that it is called in ascending page order.
1693 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
1694 unsigned long nr_pages, sector_t start_block)
1696 struct swap_extent *se;
1697 struct swap_extent *new_se;
1698 struct list_head *lh;
1700 if (start_page == 0) {
1701 se = &sis->first_swap_extent;
1702 sis->curr_swap_extent = se;
1704 se->nr_pages = nr_pages;
1705 se->start_block = start_block;
1708 lh = sis->first_swap_extent.list.prev; /* Highest extent */
1709 se = list_entry(lh, struct swap_extent, list);
1710 BUG_ON(se->start_page + se->nr_pages != start_page);
1711 if (se->start_block + se->nr_pages == start_block) {
1713 se->nr_pages += nr_pages;
1719 * No merge. Insert a new extent, preserving ordering.
1721 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
1724 new_se->start_page = start_page;
1725 new_se->nr_pages = nr_pages;
1726 new_se->start_block = start_block;
1728 list_add_tail(&new_se->list, &sis->first_swap_extent.list);
1733 * A `swap extent' is a simple thing which maps a contiguous range of pages
1734 * onto a contiguous range of disk blocks. An ordered list of swap extents
1735 * is built at swapon time and is then used at swap_writepage/swap_readpage
1736 * time for locating where on disk a page belongs.
1738 * If the swapfile is an S_ISBLK block device, a single extent is installed.
1739 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
1740 * swap files identically.
1742 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
1743 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
1744 * swapfiles are handled *identically* after swapon time.
1746 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
1747 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
1748 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
1749 * requirements, they are simply tossed out - we will never use those blocks
1752 * For S_ISREG swapfiles we set S_SWAPFILE across the life of the swapon. This
1753 * prevents root from shooting her foot off by ftruncating an in-use swapfile,
1754 * which will scribble on the fs.
1756 * The amount of disk space which a single swap extent represents varies.
1757 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
1758 * extents in the list. To avoid much list walking, we cache the previous
1759 * search location in `curr_swap_extent', and start new searches from there.
1760 * This is extremely effective. The average number of iterations in
1761 * map_swap_page() has been measured at about 0.3 per page. - akpm.
1763 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
1765 struct file *swap_file = sis->swap_file;
1766 struct address_space *mapping = swap_file->f_mapping;
1767 struct inode *inode = mapping->host;
1770 if (S_ISBLK(inode->i_mode)) {
1771 ret = add_swap_extent(sis, 0, sis->max, 0);
1776 if (mapping->a_ops->swap_activate) {
1777 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
1779 sis->flags |= SWP_FILE;
1780 ret = add_swap_extent(sis, 0, sis->max, 0);
1786 return generic_swapfile_activate(sis, swap_file, span);
1789 static void _enable_swap_info(struct swap_info_struct *p, int prio,
1790 unsigned char *swap_map,
1791 struct swap_cluster_info *cluster_info)
1796 p->prio = --least_priority;
1798 * the plist prio is negated because plist ordering is
1799 * low-to-high, while swap ordering is high-to-low
1801 p->list.prio = -p->prio;
1802 p->avail_list.prio = -p->prio;
1803 p->swap_map = swap_map;
1804 p->cluster_info = cluster_info;
1805 p->flags |= SWP_WRITEOK;
1806 atomic_long_add(p->pages, &nr_swap_pages);
1807 total_swap_pages += p->pages;
1809 assert_spin_locked(&swap_lock);
1811 * both lists are plists, and thus priority ordered.
1812 * swap_active_head needs to be priority ordered for swapoff(),
1813 * which on removal of any swap_info_struct with an auto-assigned
1814 * (i.e. negative) priority increments the auto-assigned priority
1815 * of any lower-priority swap_info_structs.
1816 * swap_avail_head needs to be priority ordered for get_swap_page(),
1817 * which allocates swap pages from the highest available priority
1820 plist_add(&p->list, &swap_active_head);
1821 spin_lock(&swap_avail_lock);
1822 plist_add(&p->avail_list, &swap_avail_head);
1823 spin_unlock(&swap_avail_lock);
1826 static void enable_swap_info(struct swap_info_struct *p, int prio,
1827 unsigned char *swap_map,
1828 struct swap_cluster_info *cluster_info,
1829 unsigned long *frontswap_map)
1831 frontswap_init(p->type, frontswap_map);
1832 spin_lock(&swap_lock);
1833 spin_lock(&p->lock);
1834 _enable_swap_info(p, prio, swap_map, cluster_info);
1835 spin_unlock(&p->lock);
1836 spin_unlock(&swap_lock);
1839 static void reinsert_swap_info(struct swap_info_struct *p)
1841 spin_lock(&swap_lock);
1842 spin_lock(&p->lock);
1843 _enable_swap_info(p, p->prio, p->swap_map, p->cluster_info);
1844 spin_unlock(&p->lock);
1845 spin_unlock(&swap_lock);
1848 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
1850 struct swap_info_struct *p = NULL;
1851 unsigned char *swap_map;
1852 struct swap_cluster_info *cluster_info;
1853 unsigned long *frontswap_map;
1854 struct file *swap_file, *victim;
1855 struct address_space *mapping;
1856 struct inode *inode;
1857 struct filename *pathname;
1859 unsigned int old_block_size;
1861 if (!capable(CAP_SYS_ADMIN))
1864 BUG_ON(!current->mm);
1866 pathname = getname(specialfile);
1867 if (IS_ERR(pathname))
1868 return PTR_ERR(pathname);
1870 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
1871 err = PTR_ERR(victim);
1875 mapping = victim->f_mapping;
1876 spin_lock(&swap_lock);
1877 plist_for_each_entry(p, &swap_active_head, list) {
1878 if (p->flags & SWP_WRITEOK) {
1879 if (p->swap_file->f_mapping == mapping) {
1887 spin_unlock(&swap_lock);
1890 if (!security_vm_enough_memory_mm(current->mm, p->pages))
1891 vm_unacct_memory(p->pages);
1894 spin_unlock(&swap_lock);
1897 spin_lock(&swap_avail_lock);
1898 plist_del(&p->avail_list, &swap_avail_head);
1899 spin_unlock(&swap_avail_lock);
1900 spin_lock(&p->lock);
1902 struct swap_info_struct *si = p;
1904 plist_for_each_entry_continue(si, &swap_active_head, list) {
1907 si->avail_list.prio--;
1911 plist_del(&p->list, &swap_active_head);
1912 atomic_long_sub(p->pages, &nr_swap_pages);
1913 total_swap_pages -= p->pages;
1914 p->flags &= ~SWP_WRITEOK;
1915 spin_unlock(&p->lock);
1916 spin_unlock(&swap_lock);
1918 set_current_oom_origin();
1919 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
1920 clear_current_oom_origin();
1923 /* re-insert swap space back into swap_list */
1924 reinsert_swap_info(p);
1928 flush_work(&p->discard_work);
1930 destroy_swap_extents(p);
1931 if (p->flags & SWP_CONTINUED)
1932 free_swap_count_continuations(p);
1934 mutex_lock(&swapon_mutex);
1935 spin_lock(&swap_lock);
1936 spin_lock(&p->lock);
1939 /* wait for anyone still in scan_swap_map */
1940 p->highest_bit = 0; /* cuts scans short */
1941 while (p->flags >= SWP_SCANNING) {
1942 spin_unlock(&p->lock);
1943 spin_unlock(&swap_lock);
1944 schedule_timeout_uninterruptible(1);
1945 spin_lock(&swap_lock);
1946 spin_lock(&p->lock);
1949 swap_file = p->swap_file;
1950 old_block_size = p->old_block_size;
1951 p->swap_file = NULL;
1953 swap_map = p->swap_map;
1955 cluster_info = p->cluster_info;
1956 p->cluster_info = NULL;
1957 frontswap_map = frontswap_map_get(p);
1958 spin_unlock(&p->lock);
1959 spin_unlock(&swap_lock);
1960 frontswap_invalidate_area(p->type);
1961 frontswap_map_set(p, NULL);
1962 mutex_unlock(&swapon_mutex);
1963 free_percpu(p->percpu_cluster);
1964 p->percpu_cluster = NULL;
1966 vfree(cluster_info);
1967 vfree(frontswap_map);
1968 /* Destroy swap account information */
1969 swap_cgroup_swapoff(p->type);
1971 inode = mapping->host;
1972 if (S_ISBLK(inode->i_mode)) {
1973 struct block_device *bdev = I_BDEV(inode);
1974 set_blocksize(bdev, old_block_size);
1975 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
1977 mutex_lock(&inode->i_mutex);
1978 inode->i_flags &= ~S_SWAPFILE;
1979 mutex_unlock(&inode->i_mutex);
1981 filp_close(swap_file, NULL);
1984 * Clear the SWP_USED flag after all resources are freed so that swapon
1985 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
1986 * not hold p->lock after we cleared its SWP_WRITEOK.
1988 spin_lock(&swap_lock);
1990 spin_unlock(&swap_lock);
1993 atomic_inc(&proc_poll_event);
1994 wake_up_interruptible(&proc_poll_wait);
1997 filp_close(victim, NULL);
2003 #ifdef CONFIG_PROC_FS
2004 static unsigned swaps_poll(struct file *file, poll_table *wait)
2006 struct seq_file *seq = file->private_data;
2008 poll_wait(file, &proc_poll_wait, wait);
2010 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2011 seq->poll_event = atomic_read(&proc_poll_event);
2012 return POLLIN | POLLRDNORM | POLLERR | POLLPRI;
2015 return POLLIN | POLLRDNORM;
2019 static void *swap_start(struct seq_file *swap, loff_t *pos)
2021 struct swap_info_struct *si;
2025 mutex_lock(&swapon_mutex);
2028 return SEQ_START_TOKEN;
2030 for (type = 0; type < nr_swapfiles; type++) {
2031 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2032 si = swap_info[type];
2033 if (!(si->flags & SWP_USED) || !si->swap_map)
2042 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2044 struct swap_info_struct *si = v;
2047 if (v == SEQ_START_TOKEN)
2050 type = si->type + 1;
2052 for (; type < nr_swapfiles; type++) {
2053 smp_rmb(); /* read nr_swapfiles before swap_info[type] */
2054 si = swap_info[type];
2055 if (!(si->flags & SWP_USED) || !si->swap_map)
2064 static void swap_stop(struct seq_file *swap, void *v)
2066 mutex_unlock(&swapon_mutex);
2069 static int swap_show(struct seq_file *swap, void *v)
2071 struct swap_info_struct *si = v;
2075 if (si == SEQ_START_TOKEN) {
2076 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2080 file = si->swap_file;
2081 len = seq_file_path(swap, file, " \t\n\\");
2082 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2083 len < 40 ? 40 - len : 1, " ",
2084 S_ISBLK(file_inode(file)->i_mode) ?
2085 "partition" : "file\t",
2086 si->pages << (PAGE_SHIFT - 10),
2087 si->inuse_pages << (PAGE_SHIFT - 10),
2092 static const struct seq_operations swaps_op = {
2093 .start = swap_start,
2099 static int swaps_open(struct inode *inode, struct file *file)
2101 struct seq_file *seq;
2104 ret = seq_open(file, &swaps_op);
2108 seq = file->private_data;
2109 seq->poll_event = atomic_read(&proc_poll_event);
2113 static const struct file_operations proc_swaps_operations = {
2116 .llseek = seq_lseek,
2117 .release = seq_release,
2121 static int __init procswaps_init(void)
2123 proc_create("swaps", 0, NULL, &proc_swaps_operations);
2126 __initcall(procswaps_init);
2127 #endif /* CONFIG_PROC_FS */
2129 #ifdef MAX_SWAPFILES_CHECK
2130 static int __init max_swapfiles_check(void)
2132 MAX_SWAPFILES_CHECK();
2135 late_initcall(max_swapfiles_check);
2138 static struct swap_info_struct *alloc_swap_info(void)
2140 struct swap_info_struct *p;
2143 p = kzalloc(sizeof(*p), GFP_KERNEL);
2145 return ERR_PTR(-ENOMEM);
2147 spin_lock(&swap_lock);
2148 for (type = 0; type < nr_swapfiles; type++) {
2149 if (!(swap_info[type]->flags & SWP_USED))
2152 if (type >= MAX_SWAPFILES) {
2153 spin_unlock(&swap_lock);
2155 return ERR_PTR(-EPERM);
2157 if (type >= nr_swapfiles) {
2159 swap_info[type] = p;
2161 * Write swap_info[type] before nr_swapfiles, in case a
2162 * racing procfs swap_start() or swap_next() is reading them.
2163 * (We never shrink nr_swapfiles, we never free this entry.)
2169 p = swap_info[type];
2171 * Do not memset this entry: a racing procfs swap_next()
2172 * would be relying on p->type to remain valid.
2175 INIT_LIST_HEAD(&p->first_swap_extent.list);
2176 plist_node_init(&p->list, 0);
2177 plist_node_init(&p->avail_list, 0);
2178 p->flags = SWP_USED;
2179 spin_unlock(&swap_lock);
2180 spin_lock_init(&p->lock);
2185 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2189 if (S_ISBLK(inode->i_mode)) {
2190 p->bdev = bdgrab(I_BDEV(inode));
2191 error = blkdev_get(p->bdev,
2192 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2197 p->old_block_size = block_size(p->bdev);
2198 error = set_blocksize(p->bdev, PAGE_SIZE);
2201 p->flags |= SWP_BLKDEV;
2202 } else if (S_ISREG(inode->i_mode)) {
2203 p->bdev = inode->i_sb->s_bdev;
2204 mutex_lock(&inode->i_mutex);
2205 if (IS_SWAPFILE(inode))
2213 static unsigned long read_swap_header(struct swap_info_struct *p,
2214 union swap_header *swap_header,
2215 struct inode *inode)
2218 unsigned long maxpages;
2219 unsigned long swapfilepages;
2220 unsigned long last_page;
2222 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2223 pr_err("Unable to find swap-space signature\n");
2227 /* swap partition endianess hack... */
2228 if (swab32(swap_header->info.version) == 1) {
2229 swab32s(&swap_header->info.version);
2230 swab32s(&swap_header->info.last_page);
2231 swab32s(&swap_header->info.nr_badpages);
2232 for (i = 0; i < swap_header->info.nr_badpages; i++)
2233 swab32s(&swap_header->info.badpages[i]);
2235 /* Check the swap header's sub-version */
2236 if (swap_header->info.version != 1) {
2237 pr_warn("Unable to handle swap header version %d\n",
2238 swap_header->info.version);
2243 p->cluster_next = 1;
2247 * Find out how many pages are allowed for a single swap
2248 * device. There are two limiting factors: 1) the number
2249 * of bits for the swap offset in the swp_entry_t type, and
2250 * 2) the number of bits in the swap pte as defined by the
2251 * different architectures. In order to find the
2252 * largest possible bit mask, a swap entry with swap type 0
2253 * and swap offset ~0UL is created, encoded to a swap pte,
2254 * decoded to a swp_entry_t again, and finally the swap
2255 * offset is extracted. This will mask all the bits from
2256 * the initial ~0UL mask that can't be encoded in either
2257 * the swp_entry_t or the architecture definition of a
2260 maxpages = swp_offset(pte_to_swp_entry(
2261 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2262 last_page = swap_header->info.last_page;
2263 if (last_page > maxpages) {
2264 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2265 maxpages << (PAGE_SHIFT - 10),
2266 last_page << (PAGE_SHIFT - 10));
2268 if (maxpages > last_page) {
2269 maxpages = last_page + 1;
2270 /* p->max is an unsigned int: don't overflow it */
2271 if ((unsigned int)maxpages == 0)
2272 maxpages = UINT_MAX;
2274 p->highest_bit = maxpages - 1;
2278 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2279 if (swapfilepages && maxpages > swapfilepages) {
2280 pr_warn("Swap area shorter than signature indicates\n");
2283 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2285 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2291 static int setup_swap_map_and_extents(struct swap_info_struct *p,
2292 union swap_header *swap_header,
2293 unsigned char *swap_map,
2294 struct swap_cluster_info *cluster_info,
2295 unsigned long maxpages,
2299 unsigned int nr_good_pages;
2301 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2302 unsigned long idx = p->cluster_next / SWAPFILE_CLUSTER;
2304 nr_good_pages = maxpages - 1; /* omit header page */
2306 cluster_set_null(&p->free_cluster_head);
2307 cluster_set_null(&p->free_cluster_tail);
2308 cluster_set_null(&p->discard_cluster_head);
2309 cluster_set_null(&p->discard_cluster_tail);
2311 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2312 unsigned int page_nr = swap_header->info.badpages[i];
2313 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2315 if (page_nr < maxpages) {
2316 swap_map[page_nr] = SWAP_MAP_BAD;
2319 * Haven't marked the cluster free yet, no list
2320 * operation involved
2322 inc_cluster_info_page(p, cluster_info, page_nr);
2326 /* Haven't marked the cluster free yet, no list operation involved */
2327 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2328 inc_cluster_info_page(p, cluster_info, i);
2330 if (nr_good_pages) {
2331 swap_map[0] = SWAP_MAP_BAD;
2333 * Not mark the cluster free yet, no list
2334 * operation involved
2336 inc_cluster_info_page(p, cluster_info, 0);
2338 p->pages = nr_good_pages;
2339 nr_extents = setup_swap_extents(p, span);
2342 nr_good_pages = p->pages;
2344 if (!nr_good_pages) {
2345 pr_warn("Empty swap-file\n");
2352 for (i = 0; i < nr_clusters; i++) {
2353 if (!cluster_count(&cluster_info[idx])) {
2354 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2355 if (cluster_is_null(&p->free_cluster_head)) {
2356 cluster_set_next_flag(&p->free_cluster_head,
2358 cluster_set_next_flag(&p->free_cluster_tail,
2363 tail = cluster_next(&p->free_cluster_tail);
2364 cluster_set_next(&cluster_info[tail], idx);
2365 cluster_set_next_flag(&p->free_cluster_tail,
2370 if (idx == nr_clusters)
2377 * Helper to sys_swapon determining if a given swap
2378 * backing device queue supports DISCARD operations.
2380 static bool swap_discardable(struct swap_info_struct *si)
2382 struct request_queue *q = bdev_get_queue(si->bdev);
2384 if (!q || !blk_queue_discard(q))
2390 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2392 struct swap_info_struct *p;
2393 struct filename *name;
2394 struct file *swap_file = NULL;
2395 struct address_space *mapping;
2398 union swap_header *swap_header;
2401 unsigned long maxpages;
2402 unsigned char *swap_map = NULL;
2403 struct swap_cluster_info *cluster_info = NULL;
2404 unsigned long *frontswap_map = NULL;
2405 struct page *page = NULL;
2406 struct inode *inode = NULL;
2408 if (swap_flags & ~SWAP_FLAGS_VALID)
2411 if (!capable(CAP_SYS_ADMIN))
2414 p = alloc_swap_info();
2418 INIT_WORK(&p->discard_work, swap_discard_work);
2420 name = getname(specialfile);
2422 error = PTR_ERR(name);
2426 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
2427 if (IS_ERR(swap_file)) {
2428 error = PTR_ERR(swap_file);
2433 p->swap_file = swap_file;
2434 mapping = swap_file->f_mapping;
2435 inode = mapping->host;
2437 /* If S_ISREG(inode->i_mode) will do mutex_lock(&inode->i_mutex); */
2438 error = claim_swapfile(p, inode);
2439 if (unlikely(error))
2443 * Read the swap header.
2445 if (!mapping->a_ops->readpage) {
2449 page = read_mapping_page(mapping, 0, swap_file);
2451 error = PTR_ERR(page);
2454 swap_header = kmap(page);
2456 maxpages = read_swap_header(p, swap_header, inode);
2457 if (unlikely(!maxpages)) {
2462 /* OK, set up the swap map and apply the bad block list */
2463 swap_map = vzalloc(maxpages);
2468 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
2471 p->flags |= SWP_SOLIDSTATE;
2473 * select a random position to start with to help wear leveling
2476 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
2478 cluster_info = vzalloc(DIV_ROUND_UP(maxpages,
2479 SWAPFILE_CLUSTER) * sizeof(*cluster_info));
2480 if (!cluster_info) {
2484 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
2485 if (!p->percpu_cluster) {
2489 for_each_possible_cpu(cpu) {
2490 struct percpu_cluster *cluster;
2491 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
2492 cluster_set_null(&cluster->index);
2496 error = swap_cgroup_swapon(p->type, maxpages);
2500 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
2501 cluster_info, maxpages, &span);
2502 if (unlikely(nr_extents < 0)) {
2506 /* frontswap enabled? set up bit-per-page map for frontswap */
2507 if (frontswap_enabled)
2508 frontswap_map = vzalloc(BITS_TO_LONGS(maxpages) * sizeof(long));
2510 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
2512 * When discard is enabled for swap with no particular
2513 * policy flagged, we set all swap discard flags here in
2514 * order to sustain backward compatibility with older
2515 * swapon(8) releases.
2517 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
2521 * By flagging sys_swapon, a sysadmin can tell us to
2522 * either do single-time area discards only, or to just
2523 * perform discards for released swap page-clusters.
2524 * Now it's time to adjust the p->flags accordingly.
2526 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
2527 p->flags &= ~SWP_PAGE_DISCARD;
2528 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
2529 p->flags &= ~SWP_AREA_DISCARD;
2531 /* issue a swapon-time discard if it's still required */
2532 if (p->flags & SWP_AREA_DISCARD) {
2533 int err = discard_swap(p);
2535 pr_err("swapon: discard_swap(%p): %d\n",
2540 mutex_lock(&swapon_mutex);
2542 if (swap_flags & SWAP_FLAG_PREFER)
2544 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
2545 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
2547 pr_info("Adding %uk swap on %s. "
2548 "Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
2549 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
2550 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
2551 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
2552 (p->flags & SWP_DISCARDABLE) ? "D" : "",
2553 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
2554 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
2555 (frontswap_map) ? "FS" : "");
2557 mutex_unlock(&swapon_mutex);
2558 atomic_inc(&proc_poll_event);
2559 wake_up_interruptible(&proc_poll_wait);
2561 if (S_ISREG(inode->i_mode))
2562 inode->i_flags |= S_SWAPFILE;
2566 free_percpu(p->percpu_cluster);
2567 p->percpu_cluster = NULL;
2568 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
2569 set_blocksize(p->bdev, p->old_block_size);
2570 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2572 destroy_swap_extents(p);
2573 swap_cgroup_swapoff(p->type);
2574 spin_lock(&swap_lock);
2575 p->swap_file = NULL;
2577 spin_unlock(&swap_lock);
2579 vfree(cluster_info);
2581 if (inode && S_ISREG(inode->i_mode)) {
2582 mutex_unlock(&inode->i_mutex);
2585 filp_close(swap_file, NULL);
2588 if (page && !IS_ERR(page)) {
2590 page_cache_release(page);
2594 if (inode && S_ISREG(inode->i_mode))
2595 mutex_unlock(&inode->i_mutex);
2599 void si_swapinfo(struct sysinfo *val)
2602 unsigned long nr_to_be_unused = 0;
2604 spin_lock(&swap_lock);
2605 for (type = 0; type < nr_swapfiles; type++) {
2606 struct swap_info_struct *si = swap_info[type];
2608 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
2609 nr_to_be_unused += si->inuse_pages;
2611 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
2612 val->totalswap = total_swap_pages + nr_to_be_unused;
2613 spin_unlock(&swap_lock);
2617 * Verify that a swap entry is valid and increment its swap map count.
2619 * Returns error code in following case.
2621 * - swp_entry is invalid -> EINVAL
2622 * - swp_entry is migration entry -> EINVAL
2623 * - swap-cache reference is requested but there is already one. -> EEXIST
2624 * - swap-cache reference is requested but the entry is not used. -> ENOENT
2625 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
2627 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
2629 struct swap_info_struct *p;
2630 unsigned long offset, type;
2631 unsigned char count;
2632 unsigned char has_cache;
2635 if (non_swap_entry(entry))
2638 type = swp_type(entry);
2639 if (type >= nr_swapfiles)
2641 p = swap_info[type];
2642 offset = swp_offset(entry);
2644 spin_lock(&p->lock);
2645 if (unlikely(offset >= p->max))
2648 count = p->swap_map[offset];
2651 * swapin_readahead() doesn't check if a swap entry is valid, so the
2652 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
2654 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
2659 has_cache = count & SWAP_HAS_CACHE;
2660 count &= ~SWAP_HAS_CACHE;
2663 if (usage == SWAP_HAS_CACHE) {
2665 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
2666 if (!has_cache && count)
2667 has_cache = SWAP_HAS_CACHE;
2668 else if (has_cache) /* someone else added cache */
2670 else /* no users remaining */
2673 } else if (count || has_cache) {
2675 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
2677 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
2679 else if (swap_count_continued(p, offset, count))
2680 count = COUNT_CONTINUED;
2684 err = -ENOENT; /* unused swap entry */
2686 p->swap_map[offset] = count | has_cache;
2689 spin_unlock(&p->lock);
2694 pr_err("swap_dup: %s%08lx\n", Bad_file, entry.val);
2699 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
2700 * (in which case its reference count is never incremented).
2702 void swap_shmem_alloc(swp_entry_t entry)
2704 __swap_duplicate(entry, SWAP_MAP_SHMEM);
2708 * Increase reference count of swap entry by 1.
2709 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
2710 * but could not be atomically allocated. Returns 0, just as if it succeeded,
2711 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
2712 * might occur if a page table entry has got corrupted.
2714 int swap_duplicate(swp_entry_t entry)
2718 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
2719 err = add_swap_count_continuation(entry, GFP_ATOMIC);
2724 * @entry: swap entry for which we allocate swap cache.
2726 * Called when allocating swap cache for existing swap entry,
2727 * This can return error codes. Returns 0 at success.
2728 * -EBUSY means there is a swap cache.
2729 * Note: return code is different from swap_duplicate().
2731 int swapcache_prepare(swp_entry_t entry)
2733 return __swap_duplicate(entry, SWAP_HAS_CACHE);
2736 struct swap_info_struct *page_swap_info(struct page *page)
2738 swp_entry_t swap = { .val = page_private(page) };
2739 BUG_ON(!PageSwapCache(page));
2740 return swap_info[swp_type(swap)];
2744 * out-of-line __page_file_ methods to avoid include hell.
2746 struct address_space *__page_file_mapping(struct page *page)
2748 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
2749 return page_swap_info(page)->swap_file->f_mapping;
2751 EXPORT_SYMBOL_GPL(__page_file_mapping);
2753 pgoff_t __page_file_index(struct page *page)
2755 swp_entry_t swap = { .val = page_private(page) };
2756 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
2757 return swp_offset(swap);
2759 EXPORT_SYMBOL_GPL(__page_file_index);
2762 * add_swap_count_continuation - called when a swap count is duplicated
2763 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
2764 * page of the original vmalloc'ed swap_map, to hold the continuation count
2765 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
2766 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
2768 * These continuation pages are seldom referenced: the common paths all work
2769 * on the original swap_map, only referring to a continuation page when the
2770 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
2772 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
2773 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
2774 * can be called after dropping locks.
2776 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
2778 struct swap_info_struct *si;
2781 struct page *list_page;
2783 unsigned char count;
2786 * When debugging, it's easier to use __GFP_ZERO here; but it's better
2787 * for latency not to zero a page while GFP_ATOMIC and holding locks.
2789 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
2791 si = swap_info_get(entry);
2794 * An acceptable race has occurred since the failing
2795 * __swap_duplicate(): the swap entry has been freed,
2796 * perhaps even the whole swap_map cleared for swapoff.
2801 offset = swp_offset(entry);
2802 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
2804 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
2806 * The higher the swap count, the more likely it is that tasks
2807 * will race to add swap count continuation: we need to avoid
2808 * over-provisioning.
2814 spin_unlock(&si->lock);
2819 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
2820 * no architecture is using highmem pages for kernel page tables: so it
2821 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
2823 head = vmalloc_to_page(si->swap_map + offset);
2824 offset &= ~PAGE_MASK;
2827 * Page allocation does not initialize the page's lru field,
2828 * but it does always reset its private field.
2830 if (!page_private(head)) {
2831 BUG_ON(count & COUNT_CONTINUED);
2832 INIT_LIST_HEAD(&head->lru);
2833 set_page_private(head, SWP_CONTINUED);
2834 si->flags |= SWP_CONTINUED;
2837 list_for_each_entry(list_page, &head->lru, lru) {
2841 * If the previous map said no continuation, but we've found
2842 * a continuation page, free our allocation and use this one.
2844 if (!(count & COUNT_CONTINUED))
2847 map = kmap_atomic(list_page) + offset;
2852 * If this continuation count now has some space in it,
2853 * free our allocation and use this one.
2855 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
2859 list_add_tail(&page->lru, &head->lru);
2860 page = NULL; /* now it's attached, don't free it */
2862 spin_unlock(&si->lock);
2870 * swap_count_continued - when the original swap_map count is incremented
2871 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
2872 * into, carry if so, or else fail until a new continuation page is allocated;
2873 * when the original swap_map count is decremented from 0 with continuation,
2874 * borrow from the continuation and report whether it still holds more.
2875 * Called while __swap_duplicate() or swap_entry_free() holds swap_lock.
2877 static bool swap_count_continued(struct swap_info_struct *si,
2878 pgoff_t offset, unsigned char count)
2884 head = vmalloc_to_page(si->swap_map + offset);
2885 if (page_private(head) != SWP_CONTINUED) {
2886 BUG_ON(count & COUNT_CONTINUED);
2887 return false; /* need to add count continuation */
2890 offset &= ~PAGE_MASK;
2891 page = list_entry(head->lru.next, struct page, lru);
2892 map = kmap_atomic(page) + offset;
2894 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
2895 goto init_map; /* jump over SWAP_CONT_MAX checks */
2897 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
2899 * Think of how you add 1 to 999
2901 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
2903 page = list_entry(page->lru.next, struct page, lru);
2904 BUG_ON(page == head);
2905 map = kmap_atomic(page) + offset;
2907 if (*map == SWAP_CONT_MAX) {
2909 page = list_entry(page->lru.next, struct page, lru);
2911 return false; /* add count continuation */
2912 map = kmap_atomic(page) + offset;
2913 init_map: *map = 0; /* we didn't zero the page */
2917 page = list_entry(page->lru.prev, struct page, lru);
2918 while (page != head) {
2919 map = kmap_atomic(page) + offset;
2920 *map = COUNT_CONTINUED;
2922 page = list_entry(page->lru.prev, struct page, lru);
2924 return true; /* incremented */
2926 } else { /* decrementing */
2928 * Think of how you subtract 1 from 1000
2930 BUG_ON(count != COUNT_CONTINUED);
2931 while (*map == COUNT_CONTINUED) {
2933 page = list_entry(page->lru.next, struct page, lru);
2934 BUG_ON(page == head);
2935 map = kmap_atomic(page) + offset;
2942 page = list_entry(page->lru.prev, struct page, lru);
2943 while (page != head) {
2944 map = kmap_atomic(page) + offset;
2945 *map = SWAP_CONT_MAX | count;
2946 count = COUNT_CONTINUED;
2948 page = list_entry(page->lru.prev, struct page, lru);
2950 return count == COUNT_CONTINUED;
2955 * free_swap_count_continuations - swapoff free all the continuation pages
2956 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
2958 static void free_swap_count_continuations(struct swap_info_struct *si)
2962 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
2964 head = vmalloc_to_page(si->swap_map + offset);
2965 if (page_private(head)) {
2966 struct list_head *this, *next;
2967 list_for_each_safe(this, next, &head->lru) {
2969 page = list_entry(this, struct page, lru);