1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie
10 #include <linux/sched/mm.h>
11 #include <linux/sched/task.h>
12 #include <linux/hugetlb.h>
13 #include <linux/mman.h>
14 #include <linux/slab.h>
15 #include <linux/kernel_stat.h>
16 #include <linux/swap.h>
17 #include <linux/vmalloc.h>
18 #include <linux/pagemap.h>
19 #include <linux/namei.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/writeback.h>
24 #include <linux/proc_fs.h>
25 #include <linux/seq_file.h>
26 #include <linux/init.h>
27 #include <linux/ksm.h>
28 #include <linux/rmap.h>
29 #include <linux/security.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mutex.h>
32 #include <linux/capability.h>
33 #include <linux/syscalls.h>
34 #include <linux/memcontrol.h>
35 #include <linux/poll.h>
36 #include <linux/oom.h>
37 #include <linux/frontswap.h>
38 #include <linux/swapfile.h>
39 #include <linux/export.h>
40 #include <linux/swap_slots.h>
41 #include <linux/sort.h>
43 #include <asm/pgtable.h>
44 #include <asm/tlbflush.h>
45 #include <linux/swapops.h>
46 #include <linux/swap_cgroup.h>
48 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
50 static void free_swap_count_continuations(struct swap_info_struct *);
51 static sector_t map_swap_entry(swp_entry_t, struct block_device**);
53 DEFINE_SPINLOCK(swap_lock);
54 static unsigned int nr_swapfiles;
55 atomic_long_t nr_swap_pages;
57 * Some modules use swappable objects and may try to swap them out under
58 * memory pressure (via the shrinker). Before doing so, they may wish to
59 * check to see if any swap space is available.
61 EXPORT_SYMBOL_GPL(nr_swap_pages);
62 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
63 long total_swap_pages;
64 static int least_priority = -1;
66 static const char Bad_file[] = "Bad swap file entry ";
67 static const char Unused_file[] = "Unused swap file entry ";
68 static const char Bad_offset[] = "Bad swap offset entry ";
69 static const char Unused_offset[] = "Unused swap offset entry ";
72 * all active swap_info_structs
73 * protected with swap_lock, and ordered by priority.
75 PLIST_HEAD(swap_active_head);
78 * all available (active, not full) swap_info_structs
79 * protected with swap_avail_lock, ordered by priority.
80 * This is used by get_swap_page() instead of swap_active_head
81 * because swap_active_head includes all swap_info_structs,
82 * but get_swap_page() doesn't need to look at full ones.
83 * This uses its own lock instead of swap_lock because when a
84 * swap_info_struct changes between not-full/full, it needs to
85 * add/remove itself to/from this list, but the swap_info_struct->lock
86 * is held and the locking order requires swap_lock to be taken
87 * before any swap_info_struct->lock.
89 static struct plist_head *swap_avail_heads;
90 static DEFINE_SPINLOCK(swap_avail_lock);
92 struct swap_info_struct *swap_info[MAX_SWAPFILES];
94 static DEFINE_MUTEX(swapon_mutex);
96 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
97 /* Activity counter to indicate that a swapon or swapoff has occurred */
98 static atomic_t proc_poll_event = ATOMIC_INIT(0);
100 atomic_t nr_rotate_swap = ATOMIC_INIT(0);
102 static struct swap_info_struct *swap_type_to_swap_info(int type)
104 if (type >= READ_ONCE(nr_swapfiles))
107 smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */
108 return READ_ONCE(swap_info[type]);
111 static inline unsigned char swap_count(unsigned char ent)
113 return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */
116 /* Reclaim the swap entry anyway if possible */
117 #define TTRS_ANYWAY 0x1
119 * Reclaim the swap entry if there are no more mappings of the
122 #define TTRS_UNMAPPED 0x2
123 /* Reclaim the swap entry if swap is getting full*/
124 #define TTRS_FULL 0x4
126 /* returns 1 if swap entry is freed */
127 static int __try_to_reclaim_swap(struct swap_info_struct *si,
128 unsigned long offset, unsigned long flags)
130 swp_entry_t entry = swp_entry(si->type, offset);
134 page = find_get_page(swap_address_space(entry), offset);
138 * When this function is called from scan_swap_map_slots() and it's
139 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
140 * here. We have to use trylock for avoiding deadlock. This is a special
141 * case and you should use try_to_free_swap() with explicit lock_page()
142 * in usual operations.
144 if (trylock_page(page)) {
145 if ((flags & TTRS_ANYWAY) ||
146 ((flags & TTRS_UNMAPPED) && !page_mapped(page)) ||
147 ((flags & TTRS_FULL) && mem_cgroup_swap_full(page)))
148 ret = try_to_free_swap(page);
155 static inline struct swap_extent *first_se(struct swap_info_struct *sis)
157 struct rb_node *rb = rb_first(&sis->swap_extent_root);
158 return rb_entry(rb, struct swap_extent, rb_node);
161 static inline struct swap_extent *next_se(struct swap_extent *se)
163 struct rb_node *rb = rb_next(&se->rb_node);
164 return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
168 * swapon tell device that all the old swap contents can be discarded,
169 * to allow the swap device to optimize its wear-levelling.
171 static int discard_swap(struct swap_info_struct *si)
173 struct swap_extent *se;
174 sector_t start_block;
178 /* Do not discard the swap header page! */
180 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
181 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
183 err = blkdev_issue_discard(si->bdev, start_block,
184 nr_blocks, GFP_KERNEL, 0);
190 for (se = next_se(se); se; se = next_se(se)) {
191 start_block = se->start_block << (PAGE_SHIFT - 9);
192 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
194 err = blkdev_issue_discard(si->bdev, start_block,
195 nr_blocks, GFP_KERNEL, 0);
201 return err; /* That will often be -EOPNOTSUPP */
204 static struct swap_extent *
205 offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
207 struct swap_extent *se;
210 rb = sis->swap_extent_root.rb_node;
212 se = rb_entry(rb, struct swap_extent, rb_node);
213 if (offset < se->start_page)
215 else if (offset >= se->start_page + se->nr_pages)
220 /* It *must* be present */
224 sector_t swap_page_sector(struct page *page)
226 struct swap_info_struct *sis = page_swap_info(page);
227 struct swap_extent *se;
231 offset = __page_file_index(page);
232 se = offset_to_swap_extent(sis, offset);
233 sector = se->start_block + (offset - se->start_page);
234 return sector << (PAGE_SHIFT - 9);
238 * swap allocation tell device that a cluster of swap can now be discarded,
239 * to allow the swap device to optimize its wear-levelling.
241 static void discard_swap_cluster(struct swap_info_struct *si,
242 pgoff_t start_page, pgoff_t nr_pages)
244 struct swap_extent *se = offset_to_swap_extent(si, start_page);
247 pgoff_t offset = start_page - se->start_page;
248 sector_t start_block = se->start_block + offset;
249 sector_t nr_blocks = se->nr_pages - offset;
251 if (nr_blocks > nr_pages)
252 nr_blocks = nr_pages;
253 start_page += nr_blocks;
254 nr_pages -= nr_blocks;
256 start_block <<= PAGE_SHIFT - 9;
257 nr_blocks <<= PAGE_SHIFT - 9;
258 if (blkdev_issue_discard(si->bdev, start_block,
259 nr_blocks, GFP_NOIO, 0))
266 #ifdef CONFIG_THP_SWAP
267 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
269 #define swap_entry_size(size) (size)
271 #define SWAPFILE_CLUSTER 256
274 * Define swap_entry_size() as constant to let compiler to optimize
275 * out some code if !CONFIG_THP_SWAP
277 #define swap_entry_size(size) 1
279 #define LATENCY_LIMIT 256
281 static inline void cluster_set_flag(struct swap_cluster_info *info,
287 static inline unsigned int cluster_count(struct swap_cluster_info *info)
292 static inline void cluster_set_count(struct swap_cluster_info *info,
298 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
299 unsigned int c, unsigned int f)
305 static inline unsigned int cluster_next(struct swap_cluster_info *info)
310 static inline void cluster_set_next(struct swap_cluster_info *info,
316 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
317 unsigned int n, unsigned int f)
323 static inline bool cluster_is_free(struct swap_cluster_info *info)
325 return info->flags & CLUSTER_FLAG_FREE;
328 static inline bool cluster_is_null(struct swap_cluster_info *info)
330 return info->flags & CLUSTER_FLAG_NEXT_NULL;
333 static inline void cluster_set_null(struct swap_cluster_info *info)
335 info->flags = CLUSTER_FLAG_NEXT_NULL;
339 static inline bool cluster_is_huge(struct swap_cluster_info *info)
341 if (IS_ENABLED(CONFIG_THP_SWAP))
342 return info->flags & CLUSTER_FLAG_HUGE;
346 static inline void cluster_clear_huge(struct swap_cluster_info *info)
348 info->flags &= ~CLUSTER_FLAG_HUGE;
351 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
352 unsigned long offset)
354 struct swap_cluster_info *ci;
356 ci = si->cluster_info;
358 ci += offset / SWAPFILE_CLUSTER;
359 spin_lock(&ci->lock);
364 static inline void unlock_cluster(struct swap_cluster_info *ci)
367 spin_unlock(&ci->lock);
371 * Determine the locking method in use for this device. Return
372 * swap_cluster_info if SSD-style cluster-based locking is in place.
374 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
375 struct swap_info_struct *si, unsigned long offset)
377 struct swap_cluster_info *ci;
379 /* Try to use fine-grained SSD-style locking if available: */
380 ci = lock_cluster(si, offset);
381 /* Otherwise, fall back to traditional, coarse locking: */
383 spin_lock(&si->lock);
388 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
389 struct swap_cluster_info *ci)
394 spin_unlock(&si->lock);
397 static inline bool cluster_list_empty(struct swap_cluster_list *list)
399 return cluster_is_null(&list->head);
402 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
404 return cluster_next(&list->head);
407 static void cluster_list_init(struct swap_cluster_list *list)
409 cluster_set_null(&list->head);
410 cluster_set_null(&list->tail);
413 static void cluster_list_add_tail(struct swap_cluster_list *list,
414 struct swap_cluster_info *ci,
417 if (cluster_list_empty(list)) {
418 cluster_set_next_flag(&list->head, idx, 0);
419 cluster_set_next_flag(&list->tail, idx, 0);
421 struct swap_cluster_info *ci_tail;
422 unsigned int tail = cluster_next(&list->tail);
425 * Nested cluster lock, but both cluster locks are
426 * only acquired when we held swap_info_struct->lock
429 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
430 cluster_set_next(ci_tail, idx);
431 spin_unlock(&ci_tail->lock);
432 cluster_set_next_flag(&list->tail, idx, 0);
436 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
437 struct swap_cluster_info *ci)
441 idx = cluster_next(&list->head);
442 if (cluster_next(&list->tail) == idx) {
443 cluster_set_null(&list->head);
444 cluster_set_null(&list->tail);
446 cluster_set_next_flag(&list->head,
447 cluster_next(&ci[idx]), 0);
452 /* Add a cluster to discard list and schedule it to do discard */
453 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
457 * If scan_swap_map() can't find a free cluster, it will check
458 * si->swap_map directly. To make sure the discarding cluster isn't
459 * taken by scan_swap_map(), mark the swap entries bad (occupied). It
460 * will be cleared after discard
462 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
463 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
465 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
467 schedule_work(&si->discard_work);
470 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
472 struct swap_cluster_info *ci = si->cluster_info;
474 cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
475 cluster_list_add_tail(&si->free_clusters, ci, idx);
479 * Doing discard actually. After a cluster discard is finished, the cluster
480 * will be added to free cluster list. caller should hold si->lock.
482 static void swap_do_scheduled_discard(struct swap_info_struct *si)
484 struct swap_cluster_info *info, *ci;
487 info = si->cluster_info;
489 while (!cluster_list_empty(&si->discard_clusters)) {
490 idx = cluster_list_del_first(&si->discard_clusters, info);
491 spin_unlock(&si->lock);
493 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
496 spin_lock(&si->lock);
497 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
498 __free_cluster(si, idx);
499 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
500 0, SWAPFILE_CLUSTER);
505 static void swap_discard_work(struct work_struct *work)
507 struct swap_info_struct *si;
509 si = container_of(work, struct swap_info_struct, discard_work);
511 spin_lock(&si->lock);
512 swap_do_scheduled_discard(si);
513 spin_unlock(&si->lock);
516 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
518 struct swap_cluster_info *ci = si->cluster_info;
520 VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
521 cluster_list_del_first(&si->free_clusters, ci);
522 cluster_set_count_flag(ci + idx, 0, 0);
525 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
527 struct swap_cluster_info *ci = si->cluster_info + idx;
529 VM_BUG_ON(cluster_count(ci) != 0);
531 * If the swap is discardable, prepare discard the cluster
532 * instead of free it immediately. The cluster will be freed
535 if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
536 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
537 swap_cluster_schedule_discard(si, idx);
541 __free_cluster(si, idx);
545 * The cluster corresponding to page_nr will be used. The cluster will be
546 * removed from free cluster list and its usage counter will be increased.
548 static void inc_cluster_info_page(struct swap_info_struct *p,
549 struct swap_cluster_info *cluster_info, unsigned long page_nr)
551 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
555 if (cluster_is_free(&cluster_info[idx]))
556 alloc_cluster(p, idx);
558 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
559 cluster_set_count(&cluster_info[idx],
560 cluster_count(&cluster_info[idx]) + 1);
564 * The cluster corresponding to page_nr decreases one usage. If the usage
565 * counter becomes 0, which means no page in the cluster is in using, we can
566 * optionally discard the cluster and add it to free cluster list.
568 static void dec_cluster_info_page(struct swap_info_struct *p,
569 struct swap_cluster_info *cluster_info, unsigned long page_nr)
571 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
576 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
577 cluster_set_count(&cluster_info[idx],
578 cluster_count(&cluster_info[idx]) - 1);
580 if (cluster_count(&cluster_info[idx]) == 0)
581 free_cluster(p, idx);
585 * It's possible scan_swap_map() uses a free cluster in the middle of free
586 * cluster list. Avoiding such abuse to avoid list corruption.
589 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
590 unsigned long offset)
592 struct percpu_cluster *percpu_cluster;
595 offset /= SWAPFILE_CLUSTER;
596 conflict = !cluster_list_empty(&si->free_clusters) &&
597 offset != cluster_list_first(&si->free_clusters) &&
598 cluster_is_free(&si->cluster_info[offset]);
603 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
604 cluster_set_null(&percpu_cluster->index);
609 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
610 * might involve allocating a new cluster for current CPU too.
612 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
613 unsigned long *offset, unsigned long *scan_base)
615 struct percpu_cluster *cluster;
616 struct swap_cluster_info *ci;
618 unsigned long tmp, max;
621 cluster = this_cpu_ptr(si->percpu_cluster);
622 if (cluster_is_null(&cluster->index)) {
623 if (!cluster_list_empty(&si->free_clusters)) {
624 cluster->index = si->free_clusters.head;
625 cluster->next = cluster_next(&cluster->index) *
627 } else if (!cluster_list_empty(&si->discard_clusters)) {
629 * we don't have free cluster but have some clusters in
630 * discarding, do discard now and reclaim them
632 swap_do_scheduled_discard(si);
633 *scan_base = *offset = si->cluster_next;
642 * Other CPUs can use our cluster if they can't find a free cluster,
643 * check if there is still free entry in the cluster
646 max = min_t(unsigned long, si->max,
647 (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
649 cluster_set_null(&cluster->index);
652 ci = lock_cluster(si, tmp);
654 if (!si->swap_map[tmp]) {
662 cluster_set_null(&cluster->index);
665 cluster->next = tmp + 1;
671 static void __del_from_avail_list(struct swap_info_struct *p)
676 plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
679 static void del_from_avail_list(struct swap_info_struct *p)
681 spin_lock(&swap_avail_lock);
682 __del_from_avail_list(p);
683 spin_unlock(&swap_avail_lock);
686 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
687 unsigned int nr_entries)
689 unsigned int end = offset + nr_entries - 1;
691 if (offset == si->lowest_bit)
692 si->lowest_bit += nr_entries;
693 if (end == si->highest_bit)
694 si->highest_bit -= nr_entries;
695 si->inuse_pages += nr_entries;
696 if (si->inuse_pages == si->pages) {
697 si->lowest_bit = si->max;
699 del_from_avail_list(si);
703 static void add_to_avail_list(struct swap_info_struct *p)
707 spin_lock(&swap_avail_lock);
709 WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
710 plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
712 spin_unlock(&swap_avail_lock);
715 static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
716 unsigned int nr_entries)
718 unsigned long end = offset + nr_entries - 1;
719 void (*swap_slot_free_notify)(struct block_device *, unsigned long);
721 if (offset < si->lowest_bit)
722 si->lowest_bit = offset;
723 if (end > si->highest_bit) {
724 bool was_full = !si->highest_bit;
726 si->highest_bit = end;
727 if (was_full && (si->flags & SWP_WRITEOK))
728 add_to_avail_list(si);
730 atomic_long_add(nr_entries, &nr_swap_pages);
731 si->inuse_pages -= nr_entries;
732 if (si->flags & SWP_BLKDEV)
733 swap_slot_free_notify =
734 si->bdev->bd_disk->fops->swap_slot_free_notify;
736 swap_slot_free_notify = NULL;
737 while (offset <= end) {
738 frontswap_invalidate_page(si->type, offset);
739 if (swap_slot_free_notify)
740 swap_slot_free_notify(si->bdev, offset);
745 static int scan_swap_map_slots(struct swap_info_struct *si,
746 unsigned char usage, int nr,
749 struct swap_cluster_info *ci;
750 unsigned long offset;
751 unsigned long scan_base;
752 unsigned long last_in_cluster = 0;
753 int latency_ration = LATENCY_LIMIT;
760 * We try to cluster swap pages by allocating them sequentially
761 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
762 * way, however, we resort to first-free allocation, starting
763 * a new cluster. This prevents us from scattering swap pages
764 * all over the entire swap partition, so that we reduce
765 * overall disk seek times between swap pages. -- sct
766 * But we do now try to find an empty cluster. -Andrea
767 * And we let swap pages go all over an SSD partition. Hugh
770 si->flags += SWP_SCANNING;
771 scan_base = offset = si->cluster_next;
774 if (si->cluster_info) {
775 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
781 if (unlikely(!si->cluster_nr--)) {
782 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
783 si->cluster_nr = SWAPFILE_CLUSTER - 1;
787 spin_unlock(&si->lock);
790 * If seek is expensive, start searching for new cluster from
791 * start of partition, to minimize the span of allocated swap.
792 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
793 * case, just handled by scan_swap_map_try_ssd_cluster() above.
795 scan_base = offset = si->lowest_bit;
796 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
798 /* Locate the first empty (unaligned) cluster */
799 for (; last_in_cluster <= si->highest_bit; offset++) {
800 if (si->swap_map[offset])
801 last_in_cluster = offset + SWAPFILE_CLUSTER;
802 else if (offset == last_in_cluster) {
803 spin_lock(&si->lock);
804 offset -= SWAPFILE_CLUSTER - 1;
805 si->cluster_next = offset;
806 si->cluster_nr = SWAPFILE_CLUSTER - 1;
809 if (unlikely(--latency_ration < 0)) {
811 latency_ration = LATENCY_LIMIT;
816 spin_lock(&si->lock);
817 si->cluster_nr = SWAPFILE_CLUSTER - 1;
821 if (si->cluster_info) {
822 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
823 /* take a break if we already got some slots */
826 if (!scan_swap_map_try_ssd_cluster(si, &offset,
831 if (!(si->flags & SWP_WRITEOK))
833 if (!si->highest_bit)
835 if (offset > si->highest_bit)
836 scan_base = offset = si->lowest_bit;
838 ci = lock_cluster(si, offset);
839 /* reuse swap entry of cache-only swap if not busy. */
840 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
843 spin_unlock(&si->lock);
844 swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
845 spin_lock(&si->lock);
846 /* entry was freed successfully, try to use this again */
849 goto scan; /* check next one */
852 if (si->swap_map[offset]) {
859 si->swap_map[offset] = usage;
860 inc_cluster_info_page(si, si->cluster_info, offset);
863 swap_range_alloc(si, offset, 1);
864 si->cluster_next = offset + 1;
865 slots[n_ret++] = swp_entry(si->type, offset);
867 /* got enough slots or reach max slots? */
868 if ((n_ret == nr) || (offset >= si->highest_bit))
871 /* search for next available slot */
873 /* time to take a break? */
874 if (unlikely(--latency_ration < 0)) {
877 spin_unlock(&si->lock);
879 spin_lock(&si->lock);
880 latency_ration = LATENCY_LIMIT;
883 /* try to get more slots in cluster */
884 if (si->cluster_info) {
885 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
893 /* non-ssd case, still more slots in cluster? */
894 if (si->cluster_nr && !si->swap_map[offset]) {
900 si->flags -= SWP_SCANNING;
904 spin_unlock(&si->lock);
905 while (++offset <= si->highest_bit) {
906 if (!si->swap_map[offset]) {
907 spin_lock(&si->lock);
910 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
911 spin_lock(&si->lock);
914 if (unlikely(--latency_ration < 0)) {
916 latency_ration = LATENCY_LIMIT;
919 offset = si->lowest_bit;
920 while (offset < scan_base) {
921 if (!si->swap_map[offset]) {
922 spin_lock(&si->lock);
925 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
926 spin_lock(&si->lock);
929 if (unlikely(--latency_ration < 0)) {
931 latency_ration = LATENCY_LIMIT;
935 spin_lock(&si->lock);
938 si->flags -= SWP_SCANNING;
942 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
945 struct swap_cluster_info *ci;
946 unsigned long offset, i;
950 * Should not even be attempting cluster allocations when huge
951 * page swap is disabled. Warn and fail the allocation.
953 if (!IS_ENABLED(CONFIG_THP_SWAP)) {
958 if (cluster_list_empty(&si->free_clusters))
961 idx = cluster_list_first(&si->free_clusters);
962 offset = idx * SWAPFILE_CLUSTER;
963 ci = lock_cluster(si, offset);
964 alloc_cluster(si, idx);
965 cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
967 map = si->swap_map + offset;
968 for (i = 0; i < SWAPFILE_CLUSTER; i++)
969 map[i] = SWAP_HAS_CACHE;
971 swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
972 *slot = swp_entry(si->type, offset);
977 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
979 unsigned long offset = idx * SWAPFILE_CLUSTER;
980 struct swap_cluster_info *ci;
982 ci = lock_cluster(si, offset);
983 memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
984 cluster_set_count_flag(ci, 0, 0);
985 free_cluster(si, idx);
987 swap_range_free(si, offset, SWAPFILE_CLUSTER);
990 static unsigned long scan_swap_map(struct swap_info_struct *si,
996 n_ret = scan_swap_map_slots(si, usage, 1, &entry);
999 return swp_offset(entry);
1005 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
1007 unsigned long size = swap_entry_size(entry_size);
1008 struct swap_info_struct *si, *next;
1013 /* Only single cluster request supported */
1014 WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1016 avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1020 if (n_goal > SWAP_BATCH)
1021 n_goal = SWAP_BATCH;
1023 if (n_goal > avail_pgs)
1026 atomic_long_sub(n_goal * size, &nr_swap_pages);
1028 spin_lock(&swap_avail_lock);
1031 node = numa_node_id();
1032 plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1033 /* requeue si to after same-priority siblings */
1034 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1035 spin_unlock(&swap_avail_lock);
1036 spin_lock(&si->lock);
1037 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1038 spin_lock(&swap_avail_lock);
1039 if (plist_node_empty(&si->avail_lists[node])) {
1040 spin_unlock(&si->lock);
1043 WARN(!si->highest_bit,
1044 "swap_info %d in list but !highest_bit\n",
1046 WARN(!(si->flags & SWP_WRITEOK),
1047 "swap_info %d in list but !SWP_WRITEOK\n",
1049 __del_from_avail_list(si);
1050 spin_unlock(&si->lock);
1053 if (size == SWAPFILE_CLUSTER) {
1054 if (si->flags & SWP_BLKDEV)
1055 n_ret = swap_alloc_cluster(si, swp_entries);
1057 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1058 n_goal, swp_entries);
1059 spin_unlock(&si->lock);
1060 if (n_ret || size == SWAPFILE_CLUSTER)
1062 pr_debug("scan_swap_map of si %d failed to find offset\n",
1065 spin_lock(&swap_avail_lock);
1068 * if we got here, it's likely that si was almost full before,
1069 * and since scan_swap_map() can drop the si->lock, multiple
1070 * callers probably all tried to get a page from the same si
1071 * and it filled up before we could get one; or, the si filled
1072 * up between us dropping swap_avail_lock and taking si->lock.
1073 * Since we dropped the swap_avail_lock, the swap_avail_head
1074 * list may have been modified; so if next is still in the
1075 * swap_avail_head list then try it, otherwise start over
1076 * if we have not gotten any slots.
1078 if (plist_node_empty(&next->avail_lists[node]))
1082 spin_unlock(&swap_avail_lock);
1086 atomic_long_add((long)(n_goal - n_ret) * size,
1092 /* The only caller of this function is now suspend routine */
1093 swp_entry_t get_swap_page_of_type(int type)
1095 struct swap_info_struct *si = swap_type_to_swap_info(type);
1101 spin_lock(&si->lock);
1102 if (si->flags & SWP_WRITEOK) {
1103 atomic_long_dec(&nr_swap_pages);
1104 /* This is called for allocating swap entry, not cache */
1105 offset = scan_swap_map(si, 1);
1107 spin_unlock(&si->lock);
1108 return swp_entry(type, offset);
1110 atomic_long_inc(&nr_swap_pages);
1112 spin_unlock(&si->lock);
1114 return (swp_entry_t) {0};
1117 static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
1119 struct swap_info_struct *p;
1120 unsigned long offset;
1124 p = swp_swap_info(entry);
1127 if (!(p->flags & SWP_USED))
1129 offset = swp_offset(entry);
1130 if (offset >= p->max)
1135 pr_err("swap_info_get: %s%08lx\n", Bad_offset, entry.val);
1138 pr_err("swap_info_get: %s%08lx\n", Unused_file, entry.val);
1141 pr_err("swap_info_get: %s%08lx\n", Bad_file, entry.val);
1146 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1148 struct swap_info_struct *p;
1150 p = __swap_info_get(entry);
1153 if (!p->swap_map[swp_offset(entry)])
1158 pr_err("swap_info_get: %s%08lx\n", Unused_offset, entry.val);
1164 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
1166 struct swap_info_struct *p;
1168 p = _swap_info_get(entry);
1170 spin_lock(&p->lock);
1174 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1175 struct swap_info_struct *q)
1177 struct swap_info_struct *p;
1179 p = _swap_info_get(entry);
1183 spin_unlock(&q->lock);
1185 spin_lock(&p->lock);
1190 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1191 unsigned long offset,
1192 unsigned char usage)
1194 unsigned char count;
1195 unsigned char has_cache;
1197 count = p->swap_map[offset];
1199 has_cache = count & SWAP_HAS_CACHE;
1200 count &= ~SWAP_HAS_CACHE;
1202 if (usage == SWAP_HAS_CACHE) {
1203 VM_BUG_ON(!has_cache);
1205 } else if (count == SWAP_MAP_SHMEM) {
1207 * Or we could insist on shmem.c using a special
1208 * swap_shmem_free() and free_shmem_swap_and_cache()...
1211 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1212 if (count == COUNT_CONTINUED) {
1213 if (swap_count_continued(p, offset, count))
1214 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1216 count = SWAP_MAP_MAX;
1221 usage = count | has_cache;
1222 p->swap_map[offset] = usage ? : SWAP_HAS_CACHE;
1228 * Check whether swap entry is valid in the swap device. If so,
1229 * return pointer to swap_info_struct, and keep the swap entry valid
1230 * via preventing the swap device from being swapoff, until
1231 * put_swap_device() is called. Otherwise return NULL.
1233 * The entirety of the RCU read critical section must come before the
1234 * return from or after the call to synchronize_rcu() in
1235 * enable_swap_info() or swapoff(). So if "si->flags & SWP_VALID" is
1236 * true, the si->map, si->cluster_info, etc. must be valid in the
1239 * Notice that swapoff or swapoff+swapon can still happen before the
1240 * rcu_read_lock() in get_swap_device() or after the rcu_read_unlock()
1241 * in put_swap_device() if there isn't any other way to prevent
1242 * swapoff, such as page lock, page table lock, etc. The caller must
1243 * be prepared for that. For example, the following situation is
1248 * ... swapoff+swapon
1249 * __read_swap_cache_async()
1250 * swapcache_prepare()
1251 * __swap_duplicate()
1253 * // verify PTE not changed
1255 * In __swap_duplicate(), the swap_map need to be checked before
1256 * changing partly because the specified swap entry may be for another
1257 * swap device which has been swapoff. And in do_swap_page(), after
1258 * the page is read from the swap device, the PTE is verified not
1259 * changed with the page table locked to check whether the swap device
1260 * has been swapoff or swapoff+swapon.
1262 struct swap_info_struct *get_swap_device(swp_entry_t entry)
1264 struct swap_info_struct *si;
1265 unsigned long offset;
1269 si = swp_swap_info(entry);
1274 if (!(si->flags & SWP_VALID))
1276 offset = swp_offset(entry);
1277 if (offset >= si->max)
1282 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1290 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1291 swp_entry_t entry, unsigned char usage)
1293 struct swap_cluster_info *ci;
1294 unsigned long offset = swp_offset(entry);
1296 ci = lock_cluster_or_swap_info(p, offset);
1297 usage = __swap_entry_free_locked(p, offset, usage);
1298 unlock_cluster_or_swap_info(p, ci);
1300 free_swap_slot(entry);
1305 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1307 struct swap_cluster_info *ci;
1308 unsigned long offset = swp_offset(entry);
1309 unsigned char count;
1311 ci = lock_cluster(p, offset);
1312 count = p->swap_map[offset];
1313 VM_BUG_ON(count != SWAP_HAS_CACHE);
1314 p->swap_map[offset] = 0;
1315 dec_cluster_info_page(p, p->cluster_info, offset);
1318 mem_cgroup_uncharge_swap(entry, 1);
1319 swap_range_free(p, offset, 1);
1323 * Caller has made sure that the swap device corresponding to entry
1324 * is still around or has not been recycled.
1326 void swap_free(swp_entry_t entry)
1328 struct swap_info_struct *p;
1330 p = _swap_info_get(entry);
1332 __swap_entry_free(p, entry, 1);
1336 * Called after dropping swapcache to decrease refcnt to swap entries.
1338 void put_swap_page(struct page *page, swp_entry_t entry)
1340 unsigned long offset = swp_offset(entry);
1341 unsigned long idx = offset / SWAPFILE_CLUSTER;
1342 struct swap_cluster_info *ci;
1343 struct swap_info_struct *si;
1345 unsigned int i, free_entries = 0;
1347 int size = swap_entry_size(hpage_nr_pages(page));
1349 si = _swap_info_get(entry);
1353 ci = lock_cluster_or_swap_info(si, offset);
1354 if (size == SWAPFILE_CLUSTER) {
1355 VM_BUG_ON(!cluster_is_huge(ci));
1356 map = si->swap_map + offset;
1357 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1359 VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1360 if (val == SWAP_HAS_CACHE)
1363 cluster_clear_huge(ci);
1364 if (free_entries == SWAPFILE_CLUSTER) {
1365 unlock_cluster_or_swap_info(si, ci);
1366 spin_lock(&si->lock);
1367 mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1368 swap_free_cluster(si, idx);
1369 spin_unlock(&si->lock);
1373 for (i = 0; i < size; i++, entry.val++) {
1374 if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1375 unlock_cluster_or_swap_info(si, ci);
1376 free_swap_slot(entry);
1379 lock_cluster_or_swap_info(si, offset);
1382 unlock_cluster_or_swap_info(si, ci);
1385 #ifdef CONFIG_THP_SWAP
1386 int split_swap_cluster(swp_entry_t entry)
1388 struct swap_info_struct *si;
1389 struct swap_cluster_info *ci;
1390 unsigned long offset = swp_offset(entry);
1392 si = _swap_info_get(entry);
1395 ci = lock_cluster(si, offset);
1396 cluster_clear_huge(ci);
1402 static int swp_entry_cmp(const void *ent1, const void *ent2)
1404 const swp_entry_t *e1 = ent1, *e2 = ent2;
1406 return (int)swp_type(*e1) - (int)swp_type(*e2);
1409 void swapcache_free_entries(swp_entry_t *entries, int n)
1411 struct swap_info_struct *p, *prev;
1421 * Sort swap entries by swap device, so each lock is only taken once.
1422 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1423 * so low that it isn't necessary to optimize further.
1425 if (nr_swapfiles > 1)
1426 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1427 for (i = 0; i < n; ++i) {
1428 p = swap_info_get_cont(entries[i], prev);
1430 swap_entry_free(p, entries[i]);
1434 spin_unlock(&p->lock);
1438 * How many references to page are currently swapped out?
1439 * This does not give an exact answer when swap count is continued,
1440 * but does include the high COUNT_CONTINUED flag to allow for that.
1442 int page_swapcount(struct page *page)
1445 struct swap_info_struct *p;
1446 struct swap_cluster_info *ci;
1448 unsigned long offset;
1450 entry.val = page_private(page);
1451 p = _swap_info_get(entry);
1453 offset = swp_offset(entry);
1454 ci = lock_cluster_or_swap_info(p, offset);
1455 count = swap_count(p->swap_map[offset]);
1456 unlock_cluster_or_swap_info(p, ci);
1461 int __swap_count(swp_entry_t entry)
1463 struct swap_info_struct *si;
1464 pgoff_t offset = swp_offset(entry);
1467 si = get_swap_device(entry);
1469 count = swap_count(si->swap_map[offset]);
1470 put_swap_device(si);
1475 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1478 pgoff_t offset = swp_offset(entry);
1479 struct swap_cluster_info *ci;
1481 ci = lock_cluster_or_swap_info(si, offset);
1482 count = swap_count(si->swap_map[offset]);
1483 unlock_cluster_or_swap_info(si, ci);
1488 * How many references to @entry are currently swapped out?
1489 * This does not give an exact answer when swap count is continued,
1490 * but does include the high COUNT_CONTINUED flag to allow for that.
1492 int __swp_swapcount(swp_entry_t entry)
1495 struct swap_info_struct *si;
1497 si = get_swap_device(entry);
1499 count = swap_swapcount(si, entry);
1500 put_swap_device(si);
1506 * How many references to @entry are currently swapped out?
1507 * This considers COUNT_CONTINUED so it returns exact answer.
1509 int swp_swapcount(swp_entry_t entry)
1511 int count, tmp_count, n;
1512 struct swap_info_struct *p;
1513 struct swap_cluster_info *ci;
1518 p = _swap_info_get(entry);
1522 offset = swp_offset(entry);
1524 ci = lock_cluster_or_swap_info(p, offset);
1526 count = swap_count(p->swap_map[offset]);
1527 if (!(count & COUNT_CONTINUED))
1530 count &= ~COUNT_CONTINUED;
1531 n = SWAP_MAP_MAX + 1;
1533 page = vmalloc_to_page(p->swap_map + offset);
1534 offset &= ~PAGE_MASK;
1535 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1538 page = list_next_entry(page, lru);
1539 map = kmap_atomic(page);
1540 tmp_count = map[offset];
1543 count += (tmp_count & ~COUNT_CONTINUED) * n;
1544 n *= (SWAP_CONT_MAX + 1);
1545 } while (tmp_count & COUNT_CONTINUED);
1547 unlock_cluster_or_swap_info(p, ci);
1551 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1554 struct swap_cluster_info *ci;
1555 unsigned char *map = si->swap_map;
1556 unsigned long roffset = swp_offset(entry);
1557 unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1561 ci = lock_cluster_or_swap_info(si, offset);
1562 if (!ci || !cluster_is_huge(ci)) {
1563 if (swap_count(map[roffset]))
1567 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1568 if (swap_count(map[offset + i])) {
1574 unlock_cluster_or_swap_info(si, ci);
1578 static bool page_swapped(struct page *page)
1581 struct swap_info_struct *si;
1583 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page)))
1584 return page_swapcount(page) != 0;
1586 page = compound_head(page);
1587 entry.val = page_private(page);
1588 si = _swap_info_get(entry);
1590 return swap_page_trans_huge_swapped(si, entry);
1594 static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1595 int *total_swapcount)
1597 int i, map_swapcount, _total_mapcount, _total_swapcount;
1598 unsigned long offset = 0;
1599 struct swap_info_struct *si;
1600 struct swap_cluster_info *ci = NULL;
1601 unsigned char *map = NULL;
1602 int mapcount, swapcount = 0;
1604 /* hugetlbfs shouldn't call it */
1605 VM_BUG_ON_PAGE(PageHuge(page), page);
1607 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) {
1608 mapcount = page_trans_huge_mapcount(page, total_mapcount);
1609 if (PageSwapCache(page))
1610 swapcount = page_swapcount(page);
1611 if (total_swapcount)
1612 *total_swapcount = swapcount;
1613 return mapcount + swapcount;
1616 page = compound_head(page);
1618 _total_mapcount = _total_swapcount = map_swapcount = 0;
1619 if (PageSwapCache(page)) {
1622 entry.val = page_private(page);
1623 si = _swap_info_get(entry);
1626 offset = swp_offset(entry);
1630 ci = lock_cluster(si, offset);
1631 for (i = 0; i < HPAGE_PMD_NR; i++) {
1632 mapcount = atomic_read(&page[i]._mapcount) + 1;
1633 _total_mapcount += mapcount;
1635 swapcount = swap_count(map[offset + i]);
1636 _total_swapcount += swapcount;
1638 map_swapcount = max(map_swapcount, mapcount + swapcount);
1641 if (PageDoubleMap(page)) {
1643 _total_mapcount -= HPAGE_PMD_NR;
1645 mapcount = compound_mapcount(page);
1646 map_swapcount += mapcount;
1647 _total_mapcount += mapcount;
1649 *total_mapcount = _total_mapcount;
1650 if (total_swapcount)
1651 *total_swapcount = _total_swapcount;
1653 return map_swapcount;
1657 * We can write to an anon page without COW if there are no other references
1658 * to it. And as a side-effect, free up its swap: because the old content
1659 * on disk will never be read, and seeking back there to write new content
1660 * later would only waste time away from clustering.
1662 * NOTE: total_map_swapcount should not be relied upon by the caller if
1663 * reuse_swap_page() returns false, but it may be always overwritten
1664 * (see the other implementation for CONFIG_SWAP=n).
1666 bool reuse_swap_page(struct page *page, int *total_map_swapcount)
1668 int count, total_mapcount, total_swapcount;
1670 VM_BUG_ON_PAGE(!PageLocked(page), page);
1671 if (unlikely(PageKsm(page)))
1673 count = page_trans_huge_map_swapcount(page, &total_mapcount,
1675 if (total_map_swapcount)
1676 *total_map_swapcount = total_mapcount + total_swapcount;
1677 if (count == 1 && PageSwapCache(page) &&
1678 (likely(!PageTransCompound(page)) ||
1679 /* The remaining swap count will be freed soon */
1680 total_swapcount == page_swapcount(page))) {
1681 if (!PageWriteback(page)) {
1682 page = compound_head(page);
1683 delete_from_swap_cache(page);
1687 struct swap_info_struct *p;
1689 entry.val = page_private(page);
1690 p = swap_info_get(entry);
1691 if (p->flags & SWP_STABLE_WRITES) {
1692 spin_unlock(&p->lock);
1695 spin_unlock(&p->lock);
1703 * If swap is getting full, or if there are no more mappings of this page,
1704 * then try_to_free_swap is called to free its swap space.
1706 int try_to_free_swap(struct page *page)
1708 VM_BUG_ON_PAGE(!PageLocked(page), page);
1710 if (!PageSwapCache(page))
1712 if (PageWriteback(page))
1714 if (page_swapped(page))
1718 * Once hibernation has begun to create its image of memory,
1719 * there's a danger that one of the calls to try_to_free_swap()
1720 * - most probably a call from __try_to_reclaim_swap() while
1721 * hibernation is allocating its own swap pages for the image,
1722 * but conceivably even a call from memory reclaim - will free
1723 * the swap from a page which has already been recorded in the
1724 * image as a clean swapcache page, and then reuse its swap for
1725 * another page of the image. On waking from hibernation, the
1726 * original page might be freed under memory pressure, then
1727 * later read back in from swap, now with the wrong data.
1729 * Hibernation suspends storage while it is writing the image
1730 * to disk so check that here.
1732 if (pm_suspended_storage())
1735 page = compound_head(page);
1736 delete_from_swap_cache(page);
1742 * Free the swap entry like above, but also try to
1743 * free the page cache entry if it is the last user.
1745 int free_swap_and_cache(swp_entry_t entry)
1747 struct swap_info_struct *p;
1748 unsigned char count;
1750 if (non_swap_entry(entry))
1753 p = _swap_info_get(entry);
1755 count = __swap_entry_free(p, entry, 1);
1756 if (count == SWAP_HAS_CACHE &&
1757 !swap_page_trans_huge_swapped(p, entry))
1758 __try_to_reclaim_swap(p, swp_offset(entry),
1759 TTRS_UNMAPPED | TTRS_FULL);
1764 #ifdef CONFIG_HIBERNATION
1766 * Find the swap type that corresponds to given device (if any).
1768 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1769 * from 0, in which the swap header is expected to be located.
1771 * This is needed for the suspend to disk (aka swsusp).
1773 int swap_type_of(dev_t device, sector_t offset, struct block_device **bdev_p)
1775 struct block_device *bdev = NULL;
1779 bdev = bdget(device);
1781 spin_lock(&swap_lock);
1782 for (type = 0; type < nr_swapfiles; type++) {
1783 struct swap_info_struct *sis = swap_info[type];
1785 if (!(sis->flags & SWP_WRITEOK))
1790 *bdev_p = bdgrab(sis->bdev);
1792 spin_unlock(&swap_lock);
1795 if (bdev == sis->bdev) {
1796 struct swap_extent *se = first_se(sis);
1798 if (se->start_block == offset) {
1800 *bdev_p = bdgrab(sis->bdev);
1802 spin_unlock(&swap_lock);
1808 spin_unlock(&swap_lock);
1816 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1817 * corresponding to given index in swap_info (swap type).
1819 sector_t swapdev_block(int type, pgoff_t offset)
1821 struct block_device *bdev;
1822 struct swap_info_struct *si = swap_type_to_swap_info(type);
1824 if (!si || !(si->flags & SWP_WRITEOK))
1826 return map_swap_entry(swp_entry(type, offset), &bdev);
1830 * Return either the total number of swap pages of given type, or the number
1831 * of free pages of that type (depending on @free)
1833 * This is needed for software suspend
1835 unsigned int count_swap_pages(int type, int free)
1839 spin_lock(&swap_lock);
1840 if ((unsigned int)type < nr_swapfiles) {
1841 struct swap_info_struct *sis = swap_info[type];
1843 spin_lock(&sis->lock);
1844 if (sis->flags & SWP_WRITEOK) {
1847 n -= sis->inuse_pages;
1849 spin_unlock(&sis->lock);
1851 spin_unlock(&swap_lock);
1854 #endif /* CONFIG_HIBERNATION */
1856 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1858 return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
1862 * No need to decide whether this PTE shares the swap entry with others,
1863 * just let do_wp_page work it out if a write is requested later - to
1864 * force COW, vm_page_prot omits write permission from any private vma.
1866 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1867 unsigned long addr, swp_entry_t entry, struct page *page)
1869 struct page *swapcache;
1870 struct mem_cgroup *memcg;
1876 page = ksm_might_need_to_copy(page, vma, addr);
1877 if (unlikely(!page))
1880 if (mem_cgroup_try_charge(page, vma->vm_mm, GFP_KERNEL,
1886 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1887 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1888 mem_cgroup_cancel_charge(page, memcg, false);
1893 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1894 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1896 set_pte_at(vma->vm_mm, addr, pte,
1897 pte_mkold(mk_pte(page, vma->vm_page_prot)));
1898 if (page == swapcache) {
1899 page_add_anon_rmap(page, vma, addr, false);
1900 mem_cgroup_commit_charge(page, memcg, true, false);
1901 } else { /* ksm created a completely new copy */
1902 page_add_new_anon_rmap(page, vma, addr, false);
1903 mem_cgroup_commit_charge(page, memcg, false, false);
1904 lru_cache_add_active_or_unevictable(page, vma);
1908 * Move the page to the active list so it is not
1909 * immediately swapped out again after swapon.
1911 activate_page(page);
1913 pte_unmap_unlock(pte, ptl);
1915 if (page != swapcache) {
1922 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1923 unsigned long addr, unsigned long end,
1924 unsigned int type, bool frontswap,
1925 unsigned long *fs_pages_to_unuse)
1930 struct swap_info_struct *si;
1931 unsigned long offset;
1933 volatile unsigned char *swap_map;
1935 si = swap_info[type];
1936 pte = pte_offset_map(pmd, addr);
1938 struct vm_fault vmf;
1940 if (!is_swap_pte(*pte))
1943 entry = pte_to_swp_entry(*pte);
1944 if (swp_type(entry) != type)
1947 offset = swp_offset(entry);
1948 if (frontswap && !frontswap_test(si, offset))
1952 swap_map = &si->swap_map[offset];
1956 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, &vmf);
1958 if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1964 wait_on_page_writeback(page);
1965 ret = unuse_pte(vma, pmd, addr, entry, page);
1972 try_to_free_swap(page);
1976 if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) {
1977 ret = FRONTSWAP_PAGES_UNUSED;
1981 pte = pte_offset_map(pmd, addr);
1982 } while (pte++, addr += PAGE_SIZE, addr != end);
1990 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1991 unsigned long addr, unsigned long end,
1992 unsigned int type, bool frontswap,
1993 unsigned long *fs_pages_to_unuse)
1999 pmd = pmd_offset(pud, addr);
2002 next = pmd_addr_end(addr, end);
2003 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
2005 ret = unuse_pte_range(vma, pmd, addr, next, type,
2006 frontswap, fs_pages_to_unuse);
2009 } while (pmd++, addr = next, addr != end);
2013 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
2014 unsigned long addr, unsigned long end,
2015 unsigned int type, bool frontswap,
2016 unsigned long *fs_pages_to_unuse)
2022 pud = pud_offset(p4d, addr);
2024 next = pud_addr_end(addr, end);
2025 if (pud_none_or_clear_bad(pud))
2027 ret = unuse_pmd_range(vma, pud, addr, next, type,
2028 frontswap, fs_pages_to_unuse);
2031 } while (pud++, addr = next, addr != end);
2035 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
2036 unsigned long addr, unsigned long end,
2037 unsigned int type, bool frontswap,
2038 unsigned long *fs_pages_to_unuse)
2044 p4d = p4d_offset(pgd, addr);
2046 next = p4d_addr_end(addr, end);
2047 if (p4d_none_or_clear_bad(p4d))
2049 ret = unuse_pud_range(vma, p4d, addr, next, type,
2050 frontswap, fs_pages_to_unuse);
2053 } while (p4d++, addr = next, addr != end);
2057 static int unuse_vma(struct vm_area_struct *vma, unsigned int type,
2058 bool frontswap, unsigned long *fs_pages_to_unuse)
2061 unsigned long addr, end, next;
2064 addr = vma->vm_start;
2067 pgd = pgd_offset(vma->vm_mm, addr);
2069 next = pgd_addr_end(addr, end);
2070 if (pgd_none_or_clear_bad(pgd))
2072 ret = unuse_p4d_range(vma, pgd, addr, next, type,
2073 frontswap, fs_pages_to_unuse);
2076 } while (pgd++, addr = next, addr != end);
2080 static int unuse_mm(struct mm_struct *mm, unsigned int type,
2081 bool frontswap, unsigned long *fs_pages_to_unuse)
2083 struct vm_area_struct *vma;
2086 down_read(&mm->mmap_sem);
2087 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2088 if (vma->anon_vma) {
2089 ret = unuse_vma(vma, type, frontswap,
2096 up_read(&mm->mmap_sem);
2101 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2102 * from current position to next entry still in use. Return 0
2103 * if there are no inuse entries after prev till end of the map.
2105 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2106 unsigned int prev, bool frontswap)
2109 unsigned char count;
2112 * No need for swap_lock here: we're just looking
2113 * for whether an entry is in use, not modifying it; false
2114 * hits are okay, and sys_swapoff() has already prevented new
2115 * allocations from this area (while holding swap_lock).
2117 for (i = prev + 1; i < si->max; i++) {
2118 count = READ_ONCE(si->swap_map[i]);
2119 if (count && swap_count(count) != SWAP_MAP_BAD)
2120 if (!frontswap || frontswap_test(si, i))
2122 if ((i % LATENCY_LIMIT) == 0)
2133 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2134 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2136 int try_to_unuse(unsigned int type, bool frontswap,
2137 unsigned long pages_to_unuse)
2139 struct mm_struct *prev_mm;
2140 struct mm_struct *mm;
2141 struct list_head *p;
2143 struct swap_info_struct *si = swap_info[type];
2148 if (!READ_ONCE(si->inuse_pages))
2155 retval = shmem_unuse(type, frontswap, &pages_to_unuse);
2162 spin_lock(&mmlist_lock);
2163 p = &init_mm.mmlist;
2164 while (READ_ONCE(si->inuse_pages) &&
2165 !signal_pending(current) &&
2166 (p = p->next) != &init_mm.mmlist) {
2168 mm = list_entry(p, struct mm_struct, mmlist);
2169 if (!mmget_not_zero(mm))
2171 spin_unlock(&mmlist_lock);
2174 retval = unuse_mm(mm, type, frontswap, &pages_to_unuse);
2182 * Make sure that we aren't completely killing
2183 * interactive performance.
2186 spin_lock(&mmlist_lock);
2188 spin_unlock(&mmlist_lock);
2193 while (READ_ONCE(si->inuse_pages) &&
2194 !signal_pending(current) &&
2195 (i = find_next_to_unuse(si, i, frontswap)) != 0) {
2197 entry = swp_entry(type, i);
2198 page = find_get_page(swap_address_space(entry), i);
2203 * It is conceivable that a racing task removed this page from
2204 * swap cache just before we acquired the page lock. The page
2205 * might even be back in swap cache on another swap area. But
2206 * that is okay, try_to_free_swap() only removes stale pages.
2209 wait_on_page_writeback(page);
2210 try_to_free_swap(page);
2215 * For frontswap, we just need to unuse pages_to_unuse, if
2216 * it was specified. Need not check frontswap again here as
2217 * we already zeroed out pages_to_unuse if not frontswap.
2219 if (pages_to_unuse && --pages_to_unuse == 0)
2224 * Lets check again to see if there are still swap entries in the map.
2225 * If yes, we would need to do retry the unuse logic again.
2226 * Under global memory pressure, swap entries can be reinserted back
2227 * into process space after the mmlist loop above passes over them.
2229 * Limit the number of retries? No: when mmget_not_zero() above fails,
2230 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2231 * at its own independent pace; and even shmem_writepage() could have
2232 * been preempted after get_swap_page(), temporarily hiding that swap.
2233 * It's easy and robust (though cpu-intensive) just to keep retrying.
2235 if (READ_ONCE(si->inuse_pages)) {
2236 if (!signal_pending(current))
2241 return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval;
2245 * After a successful try_to_unuse, if no swap is now in use, we know
2246 * we can empty the mmlist. swap_lock must be held on entry and exit.
2247 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2248 * added to the mmlist just after page_duplicate - before would be racy.
2250 static void drain_mmlist(void)
2252 struct list_head *p, *next;
2255 for (type = 0; type < nr_swapfiles; type++)
2256 if (swap_info[type]->inuse_pages)
2258 spin_lock(&mmlist_lock);
2259 list_for_each_safe(p, next, &init_mm.mmlist)
2261 spin_unlock(&mmlist_lock);
2265 * Use this swapdev's extent info to locate the (PAGE_SIZE) block which
2266 * corresponds to page offset for the specified swap entry.
2267 * Note that the type of this function is sector_t, but it returns page offset
2268 * into the bdev, not sector offset.
2270 static sector_t map_swap_entry(swp_entry_t entry, struct block_device **bdev)
2272 struct swap_info_struct *sis;
2273 struct swap_extent *se;
2276 sis = swp_swap_info(entry);
2279 offset = swp_offset(entry);
2280 se = offset_to_swap_extent(sis, offset);
2281 return se->start_block + (offset - se->start_page);
2285 * Returns the page offset into bdev for the specified page's swap entry.
2287 sector_t map_swap_page(struct page *page, struct block_device **bdev)
2290 entry.val = page_private(page);
2291 return map_swap_entry(entry, bdev);
2295 * Free all of a swapdev's extent information
2297 static void destroy_swap_extents(struct swap_info_struct *sis)
2299 while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2300 struct rb_node *rb = sis->swap_extent_root.rb_node;
2301 struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2303 rb_erase(rb, &sis->swap_extent_root);
2307 if (sis->flags & SWP_ACTIVATED) {
2308 struct file *swap_file = sis->swap_file;
2309 struct address_space *mapping = swap_file->f_mapping;
2311 sis->flags &= ~SWP_ACTIVATED;
2312 if (mapping->a_ops->swap_deactivate)
2313 mapping->a_ops->swap_deactivate(swap_file);
2318 * Add a block range (and the corresponding page range) into this swapdev's
2321 * This function rather assumes that it is called in ascending page order.
2324 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2325 unsigned long nr_pages, sector_t start_block)
2327 struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2328 struct swap_extent *se;
2329 struct swap_extent *new_se;
2332 * place the new node at the right most since the
2333 * function is called in ascending page order.
2337 link = &parent->rb_right;
2341 se = rb_entry(parent, struct swap_extent, rb_node);
2342 BUG_ON(se->start_page + se->nr_pages != start_page);
2343 if (se->start_block + se->nr_pages == start_block) {
2345 se->nr_pages += nr_pages;
2350 /* No merge, insert a new extent. */
2351 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2354 new_se->start_page = start_page;
2355 new_se->nr_pages = nr_pages;
2356 new_se->start_block = start_block;
2358 rb_link_node(&new_se->rb_node, parent, link);
2359 rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2362 EXPORT_SYMBOL_GPL(add_swap_extent);
2365 * A `swap extent' is a simple thing which maps a contiguous range of pages
2366 * onto a contiguous range of disk blocks. An ordered list of swap extents
2367 * is built at swapon time and is then used at swap_writepage/swap_readpage
2368 * time for locating where on disk a page belongs.
2370 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2371 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2372 * swap files identically.
2374 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2375 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2376 * swapfiles are handled *identically* after swapon time.
2378 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2379 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2380 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2381 * requirements, they are simply tossed out - we will never use those blocks
2384 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2385 * prevents users from writing to the swap device, which will corrupt memory.
2387 * The amount of disk space which a single swap extent represents varies.
2388 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2389 * extents in the list. To avoid much list walking, we cache the previous
2390 * search location in `curr_swap_extent', and start new searches from there.
2391 * This is extremely effective. The average number of iterations in
2392 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2394 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2396 struct file *swap_file = sis->swap_file;
2397 struct address_space *mapping = swap_file->f_mapping;
2398 struct inode *inode = mapping->host;
2401 if (S_ISBLK(inode->i_mode)) {
2402 ret = add_swap_extent(sis, 0, sis->max, 0);
2407 if (mapping->a_ops->swap_activate) {
2408 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2410 sis->flags |= SWP_ACTIVATED;
2412 sis->flags |= SWP_FS;
2413 ret = add_swap_extent(sis, 0, sis->max, 0);
2419 return generic_swapfile_activate(sis, swap_file, span);
2422 static int swap_node(struct swap_info_struct *p)
2424 struct block_device *bdev;
2429 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2431 return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2434 static void setup_swap_info(struct swap_info_struct *p, int prio,
2435 unsigned char *swap_map,
2436 struct swap_cluster_info *cluster_info)
2443 p->prio = --least_priority;
2445 * the plist prio is negated because plist ordering is
2446 * low-to-high, while swap ordering is high-to-low
2448 p->list.prio = -p->prio;
2451 p->avail_lists[i].prio = -p->prio;
2453 if (swap_node(p) == i)
2454 p->avail_lists[i].prio = 1;
2456 p->avail_lists[i].prio = -p->prio;
2459 p->swap_map = swap_map;
2460 p->cluster_info = cluster_info;
2463 static void _enable_swap_info(struct swap_info_struct *p)
2465 p->flags |= SWP_WRITEOK | SWP_VALID;
2466 atomic_long_add(p->pages, &nr_swap_pages);
2467 total_swap_pages += p->pages;
2469 assert_spin_locked(&swap_lock);
2471 * both lists are plists, and thus priority ordered.
2472 * swap_active_head needs to be priority ordered for swapoff(),
2473 * which on removal of any swap_info_struct with an auto-assigned
2474 * (i.e. negative) priority increments the auto-assigned priority
2475 * of any lower-priority swap_info_structs.
2476 * swap_avail_head needs to be priority ordered for get_swap_page(),
2477 * which allocates swap pages from the highest available priority
2480 plist_add(&p->list, &swap_active_head);
2481 add_to_avail_list(p);
2484 static void enable_swap_info(struct swap_info_struct *p, int prio,
2485 unsigned char *swap_map,
2486 struct swap_cluster_info *cluster_info,
2487 unsigned long *frontswap_map)
2489 frontswap_init(p->type, frontswap_map);
2490 spin_lock(&swap_lock);
2491 spin_lock(&p->lock);
2492 setup_swap_info(p, prio, swap_map, cluster_info);
2493 spin_unlock(&p->lock);
2494 spin_unlock(&swap_lock);
2496 * Guarantee swap_map, cluster_info, etc. fields are valid
2497 * between get/put_swap_device() if SWP_VALID bit is set
2500 spin_lock(&swap_lock);
2501 spin_lock(&p->lock);
2502 _enable_swap_info(p);
2503 spin_unlock(&p->lock);
2504 spin_unlock(&swap_lock);
2507 static void reinsert_swap_info(struct swap_info_struct *p)
2509 spin_lock(&swap_lock);
2510 spin_lock(&p->lock);
2511 setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2512 _enable_swap_info(p);
2513 spin_unlock(&p->lock);
2514 spin_unlock(&swap_lock);
2517 bool has_usable_swap(void)
2521 spin_lock(&swap_lock);
2522 if (plist_head_empty(&swap_active_head))
2524 spin_unlock(&swap_lock);
2528 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2530 struct swap_info_struct *p = NULL;
2531 unsigned char *swap_map;
2532 struct swap_cluster_info *cluster_info;
2533 unsigned long *frontswap_map;
2534 struct file *swap_file, *victim;
2535 struct address_space *mapping;
2536 struct inode *inode;
2537 struct filename *pathname;
2539 unsigned int old_block_size;
2541 if (!capable(CAP_SYS_ADMIN))
2544 BUG_ON(!current->mm);
2546 pathname = getname(specialfile);
2547 if (IS_ERR(pathname))
2548 return PTR_ERR(pathname);
2550 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2551 err = PTR_ERR(victim);
2555 mapping = victim->f_mapping;
2556 spin_lock(&swap_lock);
2557 plist_for_each_entry(p, &swap_active_head, list) {
2558 if (p->flags & SWP_WRITEOK) {
2559 if (p->swap_file->f_mapping == mapping) {
2567 spin_unlock(&swap_lock);
2570 if (!security_vm_enough_memory_mm(current->mm, p->pages))
2571 vm_unacct_memory(p->pages);
2574 spin_unlock(&swap_lock);
2577 del_from_avail_list(p);
2578 spin_lock(&p->lock);
2580 struct swap_info_struct *si = p;
2583 plist_for_each_entry_continue(si, &swap_active_head, list) {
2586 for_each_node(nid) {
2587 if (si->avail_lists[nid].prio != 1)
2588 si->avail_lists[nid].prio--;
2593 plist_del(&p->list, &swap_active_head);
2594 atomic_long_sub(p->pages, &nr_swap_pages);
2595 total_swap_pages -= p->pages;
2596 p->flags &= ~SWP_WRITEOK;
2597 spin_unlock(&p->lock);
2598 spin_unlock(&swap_lock);
2600 disable_swap_slots_cache_lock();
2602 set_current_oom_origin();
2603 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2604 clear_current_oom_origin();
2607 /* re-insert swap space back into swap_list */
2608 reinsert_swap_info(p);
2609 reenable_swap_slots_cache_unlock();
2613 reenable_swap_slots_cache_unlock();
2615 spin_lock(&swap_lock);
2616 spin_lock(&p->lock);
2617 p->flags &= ~SWP_VALID; /* mark swap device as invalid */
2618 spin_unlock(&p->lock);
2619 spin_unlock(&swap_lock);
2621 * wait for swap operations protected by get/put_swap_device()
2626 flush_work(&p->discard_work);
2628 destroy_swap_extents(p);
2629 if (p->flags & SWP_CONTINUED)
2630 free_swap_count_continuations(p);
2632 if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
2633 atomic_dec(&nr_rotate_swap);
2635 mutex_lock(&swapon_mutex);
2636 spin_lock(&swap_lock);
2637 spin_lock(&p->lock);
2640 /* wait for anyone still in scan_swap_map */
2641 p->highest_bit = 0; /* cuts scans short */
2642 while (p->flags >= SWP_SCANNING) {
2643 spin_unlock(&p->lock);
2644 spin_unlock(&swap_lock);
2645 schedule_timeout_uninterruptible(1);
2646 spin_lock(&swap_lock);
2647 spin_lock(&p->lock);
2650 swap_file = p->swap_file;
2651 old_block_size = p->old_block_size;
2652 p->swap_file = NULL;
2654 swap_map = p->swap_map;
2656 cluster_info = p->cluster_info;
2657 p->cluster_info = NULL;
2658 frontswap_map = frontswap_map_get(p);
2659 spin_unlock(&p->lock);
2660 spin_unlock(&swap_lock);
2661 frontswap_invalidate_area(p->type);
2662 frontswap_map_set(p, NULL);
2663 mutex_unlock(&swapon_mutex);
2664 free_percpu(p->percpu_cluster);
2665 p->percpu_cluster = NULL;
2667 kvfree(cluster_info);
2668 kvfree(frontswap_map);
2669 /* Destroy swap account information */
2670 swap_cgroup_swapoff(p->type);
2671 exit_swap_address_space(p->type);
2673 inode = mapping->host;
2674 if (S_ISBLK(inode->i_mode)) {
2675 struct block_device *bdev = I_BDEV(inode);
2677 set_blocksize(bdev, old_block_size);
2678 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2682 inode->i_flags &= ~S_SWAPFILE;
2683 inode_unlock(inode);
2684 filp_close(swap_file, NULL);
2687 * Clear the SWP_USED flag after all resources are freed so that swapon
2688 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2689 * not hold p->lock after we cleared its SWP_WRITEOK.
2691 spin_lock(&swap_lock);
2693 spin_unlock(&swap_lock);
2696 atomic_inc(&proc_poll_event);
2697 wake_up_interruptible(&proc_poll_wait);
2700 filp_close(victim, NULL);
2706 #ifdef CONFIG_PROC_FS
2707 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2709 struct seq_file *seq = file->private_data;
2711 poll_wait(file, &proc_poll_wait, wait);
2713 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2714 seq->poll_event = atomic_read(&proc_poll_event);
2715 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2718 return EPOLLIN | EPOLLRDNORM;
2722 static void *swap_start(struct seq_file *swap, loff_t *pos)
2724 struct swap_info_struct *si;
2728 mutex_lock(&swapon_mutex);
2731 return SEQ_START_TOKEN;
2733 for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2734 if (!(si->flags & SWP_USED) || !si->swap_map)
2743 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2745 struct swap_info_struct *si = v;
2748 if (v == SEQ_START_TOKEN)
2751 type = si->type + 1;
2754 for (; (si = swap_type_to_swap_info(type)); type++) {
2755 if (!(si->flags & SWP_USED) || !si->swap_map)
2763 static void swap_stop(struct seq_file *swap, void *v)
2765 mutex_unlock(&swapon_mutex);
2768 static int swap_show(struct seq_file *swap, void *v)
2770 struct swap_info_struct *si = v;
2774 if (si == SEQ_START_TOKEN) {
2775 seq_puts(swap,"Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
2779 file = si->swap_file;
2780 len = seq_file_path(swap, file, " \t\n\\");
2781 seq_printf(swap, "%*s%s\t%u\t%u\t%d\n",
2782 len < 40 ? 40 - len : 1, " ",
2783 S_ISBLK(file_inode(file)->i_mode) ?
2784 "partition" : "file\t",
2785 si->pages << (PAGE_SHIFT - 10),
2786 si->inuse_pages << (PAGE_SHIFT - 10),
2791 static const struct seq_operations swaps_op = {
2792 .start = swap_start,
2798 static int swaps_open(struct inode *inode, struct file *file)
2800 struct seq_file *seq;
2803 ret = seq_open(file, &swaps_op);
2807 seq = file->private_data;
2808 seq->poll_event = atomic_read(&proc_poll_event);
2812 static const struct file_operations proc_swaps_operations = {
2815 .llseek = seq_lseek,
2816 .release = seq_release,
2820 static int __init procswaps_init(void)
2822 proc_create("swaps", 0, NULL, &proc_swaps_operations);
2825 __initcall(procswaps_init);
2826 #endif /* CONFIG_PROC_FS */
2828 #ifdef MAX_SWAPFILES_CHECK
2829 static int __init max_swapfiles_check(void)
2831 MAX_SWAPFILES_CHECK();
2834 late_initcall(max_swapfiles_check);
2837 static struct swap_info_struct *alloc_swap_info(void)
2839 struct swap_info_struct *p;
2840 struct swap_info_struct *defer = NULL;
2844 p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2846 return ERR_PTR(-ENOMEM);
2848 spin_lock(&swap_lock);
2849 for (type = 0; type < nr_swapfiles; type++) {
2850 if (!(swap_info[type]->flags & SWP_USED))
2853 if (type >= MAX_SWAPFILES) {
2854 spin_unlock(&swap_lock);
2856 return ERR_PTR(-EPERM);
2858 if (type >= nr_swapfiles) {
2860 WRITE_ONCE(swap_info[type], p);
2862 * Write swap_info[type] before nr_swapfiles, in case a
2863 * racing procfs swap_start() or swap_next() is reading them.
2864 * (We never shrink nr_swapfiles, we never free this entry.)
2867 WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1);
2870 p = swap_info[type];
2872 * Do not memset this entry: a racing procfs swap_next()
2873 * would be relying on p->type to remain valid.
2876 p->swap_extent_root = RB_ROOT;
2877 plist_node_init(&p->list, 0);
2879 plist_node_init(&p->avail_lists[i], 0);
2880 p->flags = SWP_USED;
2881 spin_unlock(&swap_lock);
2883 spin_lock_init(&p->lock);
2884 spin_lock_init(&p->cont_lock);
2889 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2893 if (S_ISBLK(inode->i_mode)) {
2894 p->bdev = bdgrab(I_BDEV(inode));
2895 error = blkdev_get(p->bdev,
2896 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2901 p->old_block_size = block_size(p->bdev);
2902 error = set_blocksize(p->bdev, PAGE_SIZE);
2905 p->flags |= SWP_BLKDEV;
2906 } else if (S_ISREG(inode->i_mode)) {
2907 p->bdev = inode->i_sb->s_bdev;
2915 * Find out how many pages are allowed for a single swap device. There
2916 * are two limiting factors:
2917 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2918 * 2) the number of bits in the swap pte, as defined by the different
2921 * In order to find the largest possible bit mask, a swap entry with
2922 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2923 * decoded to a swp_entry_t again, and finally the swap offset is
2926 * This will mask all the bits from the initial ~0UL mask that can't
2927 * be encoded in either the swp_entry_t or the architecture definition
2930 unsigned long generic_max_swapfile_size(void)
2932 return swp_offset(pte_to_swp_entry(
2933 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2936 /* Can be overridden by an architecture for additional checks. */
2937 __weak unsigned long max_swapfile_size(void)
2939 return generic_max_swapfile_size();
2942 static unsigned long read_swap_header(struct swap_info_struct *p,
2943 union swap_header *swap_header,
2944 struct inode *inode)
2947 unsigned long maxpages;
2948 unsigned long swapfilepages;
2949 unsigned long last_page;
2951 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2952 pr_err("Unable to find swap-space signature\n");
2956 /* swap partition endianess hack... */
2957 if (swab32(swap_header->info.version) == 1) {
2958 swab32s(&swap_header->info.version);
2959 swab32s(&swap_header->info.last_page);
2960 swab32s(&swap_header->info.nr_badpages);
2961 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2963 for (i = 0; i < swap_header->info.nr_badpages; i++)
2964 swab32s(&swap_header->info.badpages[i]);
2966 /* Check the swap header's sub-version */
2967 if (swap_header->info.version != 1) {
2968 pr_warn("Unable to handle swap header version %d\n",
2969 swap_header->info.version);
2974 p->cluster_next = 1;
2977 maxpages = max_swapfile_size();
2978 last_page = swap_header->info.last_page;
2980 pr_warn("Empty swap-file\n");
2983 if (last_page > maxpages) {
2984 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2985 maxpages << (PAGE_SHIFT - 10),
2986 last_page << (PAGE_SHIFT - 10));
2988 if (maxpages > last_page) {
2989 maxpages = last_page + 1;
2990 /* p->max is an unsigned int: don't overflow it */
2991 if ((unsigned int)maxpages == 0)
2992 maxpages = UINT_MAX;
2994 p->highest_bit = maxpages - 1;
2998 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2999 if (swapfilepages && maxpages > swapfilepages) {
3000 pr_warn("Swap area shorter than signature indicates\n");
3003 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
3005 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
3011 #define SWAP_CLUSTER_INFO_COLS \
3012 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3013 #define SWAP_CLUSTER_SPACE_COLS \
3014 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3015 #define SWAP_CLUSTER_COLS \
3016 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3018 static int setup_swap_map_and_extents(struct swap_info_struct *p,
3019 union swap_header *swap_header,
3020 unsigned char *swap_map,
3021 struct swap_cluster_info *cluster_info,
3022 unsigned long maxpages,
3026 unsigned int nr_good_pages;
3028 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3029 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
3030 unsigned long i, idx;
3032 nr_good_pages = maxpages - 1; /* omit header page */
3034 cluster_list_init(&p->free_clusters);
3035 cluster_list_init(&p->discard_clusters);
3037 for (i = 0; i < swap_header->info.nr_badpages; i++) {
3038 unsigned int page_nr = swap_header->info.badpages[i];
3039 if (page_nr == 0 || page_nr > swap_header->info.last_page)
3041 if (page_nr < maxpages) {
3042 swap_map[page_nr] = SWAP_MAP_BAD;
3045 * Haven't marked the cluster free yet, no list
3046 * operation involved
3048 inc_cluster_info_page(p, cluster_info, page_nr);
3052 /* Haven't marked the cluster free yet, no list operation involved */
3053 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
3054 inc_cluster_info_page(p, cluster_info, i);
3056 if (nr_good_pages) {
3057 swap_map[0] = SWAP_MAP_BAD;
3059 * Not mark the cluster free yet, no list
3060 * operation involved
3062 inc_cluster_info_page(p, cluster_info, 0);
3064 p->pages = nr_good_pages;
3065 nr_extents = setup_swap_extents(p, span);
3068 nr_good_pages = p->pages;
3070 if (!nr_good_pages) {
3071 pr_warn("Empty swap-file\n");
3080 * Reduce false cache line sharing between cluster_info and
3081 * sharing same address space.
3083 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
3084 j = (k + col) % SWAP_CLUSTER_COLS;
3085 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
3086 idx = i * SWAP_CLUSTER_COLS + j;
3087 if (idx >= nr_clusters)
3089 if (cluster_count(&cluster_info[idx]))
3091 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
3092 cluster_list_add_tail(&p->free_clusters, cluster_info,
3100 * Helper to sys_swapon determining if a given swap
3101 * backing device queue supports DISCARD operations.
3103 static bool swap_discardable(struct swap_info_struct *si)
3105 struct request_queue *q = bdev_get_queue(si->bdev);
3107 if (!q || !blk_queue_discard(q))
3113 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
3115 struct swap_info_struct *p;
3116 struct filename *name;
3117 struct file *swap_file = NULL;
3118 struct address_space *mapping;
3121 union swap_header *swap_header;
3124 unsigned long maxpages;
3125 unsigned char *swap_map = NULL;
3126 struct swap_cluster_info *cluster_info = NULL;
3127 unsigned long *frontswap_map = NULL;
3128 struct page *page = NULL;
3129 struct inode *inode = NULL;
3130 bool inced_nr_rotate_swap = false;
3132 if (swap_flags & ~SWAP_FLAGS_VALID)
3135 if (!capable(CAP_SYS_ADMIN))
3138 if (!swap_avail_heads)
3141 p = alloc_swap_info();
3145 INIT_WORK(&p->discard_work, swap_discard_work);
3147 name = getname(specialfile);
3149 error = PTR_ERR(name);
3153 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3154 if (IS_ERR(swap_file)) {
3155 error = PTR_ERR(swap_file);
3160 p->swap_file = swap_file;
3161 mapping = swap_file->f_mapping;
3162 inode = mapping->host;
3164 error = claim_swapfile(p, inode);
3165 if (unlikely(error))
3169 if (IS_SWAPFILE(inode)) {
3171 goto bad_swap_unlock_inode;
3175 * Read the swap header.
3177 if (!mapping->a_ops->readpage) {
3179 goto bad_swap_unlock_inode;
3181 page = read_mapping_page(mapping, 0, swap_file);
3183 error = PTR_ERR(page);
3186 swap_header = kmap(page);
3188 maxpages = read_swap_header(p, swap_header, inode);
3189 if (unlikely(!maxpages)) {
3191 goto bad_swap_unlock_inode;
3194 /* OK, set up the swap map and apply the bad block list */
3195 swap_map = vzalloc(maxpages);
3198 goto bad_swap_unlock_inode;
3201 if (bdi_cap_stable_pages_required(inode_to_bdi(inode)))
3202 p->flags |= SWP_STABLE_WRITES;
3204 if (bdi_cap_synchronous_io(inode_to_bdi(inode)))
3205 p->flags |= SWP_SYNCHRONOUS_IO;
3207 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
3209 unsigned long ci, nr_cluster;
3211 p->flags |= SWP_SOLIDSTATE;
3213 * select a random position to start with to help wear leveling
3216 p->cluster_next = 1 + (prandom_u32() % p->highest_bit);
3217 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3219 cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3221 if (!cluster_info) {
3223 goto bad_swap_unlock_inode;
3226 for (ci = 0; ci < nr_cluster; ci++)
3227 spin_lock_init(&((cluster_info + ci)->lock));
3229 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3230 if (!p->percpu_cluster) {
3232 goto bad_swap_unlock_inode;
3234 for_each_possible_cpu(cpu) {
3235 struct percpu_cluster *cluster;
3236 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3237 cluster_set_null(&cluster->index);
3240 atomic_inc(&nr_rotate_swap);
3241 inced_nr_rotate_swap = true;
3244 error = swap_cgroup_swapon(p->type, maxpages);
3246 goto bad_swap_unlock_inode;
3248 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3249 cluster_info, maxpages, &span);
3250 if (unlikely(nr_extents < 0)) {
3252 goto bad_swap_unlock_inode;
3254 /* frontswap enabled? set up bit-per-page map for frontswap */
3255 if (IS_ENABLED(CONFIG_FRONTSWAP))
3256 frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3260 if (p->bdev &&(swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
3262 * When discard is enabled for swap with no particular
3263 * policy flagged, we set all swap discard flags here in
3264 * order to sustain backward compatibility with older
3265 * swapon(8) releases.
3267 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3271 * By flagging sys_swapon, a sysadmin can tell us to
3272 * either do single-time area discards only, or to just
3273 * perform discards for released swap page-clusters.
3274 * Now it's time to adjust the p->flags accordingly.
3276 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3277 p->flags &= ~SWP_PAGE_DISCARD;
3278 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3279 p->flags &= ~SWP_AREA_DISCARD;
3281 /* issue a swapon-time discard if it's still required */
3282 if (p->flags & SWP_AREA_DISCARD) {
3283 int err = discard_swap(p);
3285 pr_err("swapon: discard_swap(%p): %d\n",
3290 error = init_swap_address_space(p->type, maxpages);
3292 goto bad_swap_unlock_inode;
3295 * Flush any pending IO and dirty mappings before we start using this
3298 inode->i_flags |= S_SWAPFILE;
3299 error = inode_drain_writes(inode);
3301 inode->i_flags &= ~S_SWAPFILE;
3302 goto free_swap_address_space;
3305 mutex_lock(&swapon_mutex);
3307 if (swap_flags & SWAP_FLAG_PREFER)
3309 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3310 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3312 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3313 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3314 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3315 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3316 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3317 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3318 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3319 (frontswap_map) ? "FS" : "");
3321 mutex_unlock(&swapon_mutex);
3322 atomic_inc(&proc_poll_event);
3323 wake_up_interruptible(&proc_poll_wait);
3327 free_swap_address_space:
3328 exit_swap_address_space(p->type);
3329 bad_swap_unlock_inode:
3330 inode_unlock(inode);
3332 free_percpu(p->percpu_cluster);
3333 p->percpu_cluster = NULL;
3334 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3335 set_blocksize(p->bdev, p->old_block_size);
3336 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3339 destroy_swap_extents(p);
3340 swap_cgroup_swapoff(p->type);
3341 spin_lock(&swap_lock);
3342 p->swap_file = NULL;
3344 spin_unlock(&swap_lock);
3346 kvfree(cluster_info);
3347 kvfree(frontswap_map);
3348 if (inced_nr_rotate_swap)
3349 atomic_dec(&nr_rotate_swap);
3351 filp_close(swap_file, NULL);
3353 if (page && !IS_ERR(page)) {
3360 inode_unlock(inode);
3362 enable_swap_slots_cache();
3366 void si_swapinfo(struct sysinfo *val)
3369 unsigned long nr_to_be_unused = 0;
3371 spin_lock(&swap_lock);
3372 for (type = 0; type < nr_swapfiles; type++) {
3373 struct swap_info_struct *si = swap_info[type];
3375 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3376 nr_to_be_unused += si->inuse_pages;
3378 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3379 val->totalswap = total_swap_pages + nr_to_be_unused;
3380 spin_unlock(&swap_lock);
3384 * Verify that a swap entry is valid and increment its swap map count.
3386 * Returns error code in following case.
3388 * - swp_entry is invalid -> EINVAL
3389 * - swp_entry is migration entry -> EINVAL
3390 * - swap-cache reference is requested but there is already one. -> EEXIST
3391 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3392 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3394 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3396 struct swap_info_struct *p;
3397 struct swap_cluster_info *ci;
3398 unsigned long offset;
3399 unsigned char count;
3400 unsigned char has_cache;
3403 p = get_swap_device(entry);
3407 offset = swp_offset(entry);
3408 ci = lock_cluster_or_swap_info(p, offset);
3410 count = p->swap_map[offset];
3413 * swapin_readahead() doesn't check if a swap entry is valid, so the
3414 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3416 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3421 has_cache = count & SWAP_HAS_CACHE;
3422 count &= ~SWAP_HAS_CACHE;
3425 if (usage == SWAP_HAS_CACHE) {
3427 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3428 if (!has_cache && count)
3429 has_cache = SWAP_HAS_CACHE;
3430 else if (has_cache) /* someone else added cache */
3432 else /* no users remaining */
3435 } else if (count || has_cache) {
3437 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3439 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3441 else if (swap_count_continued(p, offset, count))
3442 count = COUNT_CONTINUED;
3446 err = -ENOENT; /* unused swap entry */
3448 p->swap_map[offset] = count | has_cache;
3451 unlock_cluster_or_swap_info(p, ci);
3459 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3460 * (in which case its reference count is never incremented).
3462 void swap_shmem_alloc(swp_entry_t entry)
3464 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3468 * Increase reference count of swap entry by 1.
3469 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3470 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3471 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3472 * might occur if a page table entry has got corrupted.
3474 int swap_duplicate(swp_entry_t entry)
3478 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3479 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3484 * @entry: swap entry for which we allocate swap cache.
3486 * Called when allocating swap cache for existing swap entry,
3487 * This can return error codes. Returns 0 at success.
3488 * -EBUSY means there is a swap cache.
3489 * Note: return code is different from swap_duplicate().
3491 int swapcache_prepare(swp_entry_t entry)
3493 return __swap_duplicate(entry, SWAP_HAS_CACHE);
3496 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3498 return swap_type_to_swap_info(swp_type(entry));
3501 struct swap_info_struct *page_swap_info(struct page *page)
3503 swp_entry_t entry = { .val = page_private(page) };
3504 return swp_swap_info(entry);
3508 * out-of-line __page_file_ methods to avoid include hell.
3510 struct address_space *__page_file_mapping(struct page *page)
3512 return page_swap_info(page)->swap_file->f_mapping;
3514 EXPORT_SYMBOL_GPL(__page_file_mapping);
3516 pgoff_t __page_file_index(struct page *page)
3518 swp_entry_t swap = { .val = page_private(page) };
3519 return swp_offset(swap);
3521 EXPORT_SYMBOL_GPL(__page_file_index);
3524 * add_swap_count_continuation - called when a swap count is duplicated
3525 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3526 * page of the original vmalloc'ed swap_map, to hold the continuation count
3527 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3528 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3530 * These continuation pages are seldom referenced: the common paths all work
3531 * on the original swap_map, only referring to a continuation page when the
3532 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3534 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3535 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3536 * can be called after dropping locks.
3538 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3540 struct swap_info_struct *si;
3541 struct swap_cluster_info *ci;
3544 struct page *list_page;
3546 unsigned char count;
3550 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3551 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3553 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3555 si = get_swap_device(entry);
3558 * An acceptable race has occurred since the failing
3559 * __swap_duplicate(): the swap device may be swapoff
3563 spin_lock(&si->lock);
3565 offset = swp_offset(entry);
3567 ci = lock_cluster(si, offset);
3569 count = si->swap_map[offset] & ~SWAP_HAS_CACHE;
3571 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3573 * The higher the swap count, the more likely it is that tasks
3574 * will race to add swap count continuation: we need to avoid
3575 * over-provisioning.
3586 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3587 * no architecture is using highmem pages for kernel page tables: so it
3588 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3590 head = vmalloc_to_page(si->swap_map + offset);
3591 offset &= ~PAGE_MASK;
3593 spin_lock(&si->cont_lock);
3595 * Page allocation does not initialize the page's lru field,
3596 * but it does always reset its private field.
3598 if (!page_private(head)) {
3599 BUG_ON(count & COUNT_CONTINUED);
3600 INIT_LIST_HEAD(&head->lru);
3601 set_page_private(head, SWP_CONTINUED);
3602 si->flags |= SWP_CONTINUED;
3605 list_for_each_entry(list_page, &head->lru, lru) {
3609 * If the previous map said no continuation, but we've found
3610 * a continuation page, free our allocation and use this one.
3612 if (!(count & COUNT_CONTINUED))
3613 goto out_unlock_cont;
3615 map = kmap_atomic(list_page) + offset;
3620 * If this continuation count now has some space in it,
3621 * free our allocation and use this one.
3623 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3624 goto out_unlock_cont;
3627 list_add_tail(&page->lru, &head->lru);
3628 page = NULL; /* now it's attached, don't free it */
3630 spin_unlock(&si->cont_lock);
3633 spin_unlock(&si->lock);
3634 put_swap_device(si);
3642 * swap_count_continued - when the original swap_map count is incremented
3643 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3644 * into, carry if so, or else fail until a new continuation page is allocated;
3645 * when the original swap_map count is decremented from 0 with continuation,
3646 * borrow from the continuation and report whether it still holds more.
3647 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3650 static bool swap_count_continued(struct swap_info_struct *si,
3651 pgoff_t offset, unsigned char count)
3658 head = vmalloc_to_page(si->swap_map + offset);
3659 if (page_private(head) != SWP_CONTINUED) {
3660 BUG_ON(count & COUNT_CONTINUED);
3661 return false; /* need to add count continuation */
3664 spin_lock(&si->cont_lock);
3665 offset &= ~PAGE_MASK;
3666 page = list_entry(head->lru.next, struct page, lru);
3667 map = kmap_atomic(page) + offset;
3669 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3670 goto init_map; /* jump over SWAP_CONT_MAX checks */
3672 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3674 * Think of how you add 1 to 999
3676 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3678 page = list_entry(page->lru.next, struct page, lru);
3679 BUG_ON(page == head);
3680 map = kmap_atomic(page) + offset;
3682 if (*map == SWAP_CONT_MAX) {
3684 page = list_entry(page->lru.next, struct page, lru);
3686 ret = false; /* add count continuation */
3689 map = kmap_atomic(page) + offset;
3690 init_map: *map = 0; /* we didn't zero the page */
3694 page = list_entry(page->lru.prev, struct page, lru);
3695 while (page != head) {
3696 map = kmap_atomic(page) + offset;
3697 *map = COUNT_CONTINUED;
3699 page = list_entry(page->lru.prev, struct page, lru);
3701 ret = true; /* incremented */
3703 } else { /* decrementing */
3705 * Think of how you subtract 1 from 1000
3707 BUG_ON(count != COUNT_CONTINUED);
3708 while (*map == COUNT_CONTINUED) {
3710 page = list_entry(page->lru.next, struct page, lru);
3711 BUG_ON(page == head);
3712 map = kmap_atomic(page) + offset;
3719 page = list_entry(page->lru.prev, struct page, lru);
3720 while (page != head) {
3721 map = kmap_atomic(page) + offset;
3722 *map = SWAP_CONT_MAX | count;
3723 count = COUNT_CONTINUED;
3725 page = list_entry(page->lru.prev, struct page, lru);
3727 ret = count == COUNT_CONTINUED;
3730 spin_unlock(&si->cont_lock);
3735 * free_swap_count_continuations - swapoff free all the continuation pages
3736 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3738 static void free_swap_count_continuations(struct swap_info_struct *si)
3742 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3744 head = vmalloc_to_page(si->swap_map + offset);
3745 if (page_private(head)) {
3746 struct page *page, *next;
3748 list_for_each_entry_safe(page, next, &head->lru, lru) {
3749 list_del(&page->lru);
3756 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3757 void mem_cgroup_throttle_swaprate(struct mem_cgroup *memcg, int node,
3760 struct swap_info_struct *si, *next;
3761 if (!(gfp_mask & __GFP_IO) || !memcg)
3764 if (!blk_cgroup_congested())
3768 * We've already scheduled a throttle, avoid taking the global swap
3771 if (current->throttle_queue)
3774 spin_lock(&swap_avail_lock);
3775 plist_for_each_entry_safe(si, next, &swap_avail_heads[node],
3776 avail_lists[node]) {
3778 blkcg_schedule_throttle(bdev_get_queue(si->bdev),
3783 spin_unlock(&swap_avail_lock);
3787 static int __init swapfile_init(void)
3791 swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3793 if (!swap_avail_heads) {
3794 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3799 plist_head_init(&swap_avail_heads[nid]);
3803 subsys_initcall(swapfile_init);