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
9 #include <linux/blkdev.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/task.h>
13 #include <linux/hugetlb.h>
14 #include <linux/mman.h>
15 #include <linux/slab.h>
16 #include <linux/kernel_stat.h>
17 #include <linux/swap.h>
18 #include <linux/vmalloc.h>
19 #include <linux/pagemap.h>
20 #include <linux/namei.h>
21 #include <linux/shmem_fs.h>
22 #include <linux/blk-cgroup.h>
23 #include <linux/random.h>
24 #include <linux/writeback.h>
25 #include <linux/proc_fs.h>
26 #include <linux/seq_file.h>
27 #include <linux/init.h>
28 #include <linux/ksm.h>
29 #include <linux/rmap.h>
30 #include <linux/security.h>
31 #include <linux/backing-dev.h>
32 #include <linux/mutex.h>
33 #include <linux/capability.h>
34 #include <linux/syscalls.h>
35 #include <linux/memcontrol.h>
36 #include <linux/poll.h>
37 #include <linux/oom.h>
38 #include <linux/swapfile.h>
39 #include <linux/export.h>
40 #include <linux/swap_slots.h>
41 #include <linux/sort.h>
42 #include <linux/completion.h>
43 #include <linux/suspend.h>
44 #include <linux/zswap.h>
45 #include <linux/plist.h>
47 #include <asm/tlbflush.h>
48 #include <linux/swapops.h>
49 #include <linux/swap_cgroup.h>
53 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
55 static void free_swap_count_continuations(struct swap_info_struct *);
57 static DEFINE_SPINLOCK(swap_lock);
58 static unsigned int nr_swapfiles;
59 atomic_long_t nr_swap_pages;
61 * Some modules use swappable objects and may try to swap them out under
62 * memory pressure (via the shrinker). Before doing so, they may wish to
63 * check to see if any swap space is available.
65 EXPORT_SYMBOL_GPL(nr_swap_pages);
66 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
67 long total_swap_pages;
68 static int least_priority = -1;
69 unsigned long swapfile_maximum_size;
70 #ifdef CONFIG_MIGRATION
71 bool swap_migration_ad_supported;
72 #endif /* CONFIG_MIGRATION */
74 static const char Bad_file[] = "Bad swap file entry ";
75 static const char Unused_file[] = "Unused swap file entry ";
76 static const char Bad_offset[] = "Bad swap offset entry ";
77 static const char Unused_offset[] = "Unused swap offset entry ";
80 * all active swap_info_structs
81 * protected with swap_lock, and ordered by priority.
83 static PLIST_HEAD(swap_active_head);
86 * all available (active, not full) swap_info_structs
87 * protected with swap_avail_lock, ordered by priority.
88 * This is used by folio_alloc_swap() instead of swap_active_head
89 * because swap_active_head includes all swap_info_structs,
90 * but folio_alloc_swap() doesn't need to look at full ones.
91 * This uses its own lock instead of swap_lock because when a
92 * swap_info_struct changes between not-full/full, it needs to
93 * add/remove itself to/from this list, but the swap_info_struct->lock
94 * is held and the locking order requires swap_lock to be taken
95 * before any swap_info_struct->lock.
97 static struct plist_head *swap_avail_heads;
98 static DEFINE_SPINLOCK(swap_avail_lock);
100 static struct swap_info_struct *swap_info[MAX_SWAPFILES];
102 static DEFINE_MUTEX(swapon_mutex);
104 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
105 /* Activity counter to indicate that a swapon or swapoff has occurred */
106 static atomic_t proc_poll_event = ATOMIC_INIT(0);
108 atomic_t nr_rotate_swap = ATOMIC_INIT(0);
110 static struct swap_info_struct *swap_type_to_swap_info(int type)
112 if (type >= MAX_SWAPFILES)
115 return READ_ONCE(swap_info[type]); /* rcu_dereference() */
118 static inline unsigned char swap_count(unsigned char ent)
120 return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */
123 /* Reclaim the swap entry anyway if possible */
124 #define TTRS_ANYWAY 0x1
126 * Reclaim the swap entry if there are no more mappings of the
129 #define TTRS_UNMAPPED 0x2
130 /* Reclaim the swap entry if swap is getting full*/
131 #define TTRS_FULL 0x4
133 /* returns 1 if swap entry is freed */
134 static int __try_to_reclaim_swap(struct swap_info_struct *si,
135 unsigned long offset, unsigned long flags)
137 swp_entry_t entry = swp_entry(si->type, offset);
141 folio = filemap_get_folio(swap_address_space(entry), offset);
145 * When this function is called from scan_swap_map_slots() and it's
146 * called by vmscan.c at reclaiming folios. So we hold a folio lock
147 * here. We have to use trylock for avoiding deadlock. This is a special
148 * case and you should use folio_free_swap() with explicit folio_lock()
149 * in usual operations.
151 if (folio_trylock(folio)) {
152 if ((flags & TTRS_ANYWAY) ||
153 ((flags & TTRS_UNMAPPED) && !folio_mapped(folio)) ||
154 ((flags & TTRS_FULL) && mem_cgroup_swap_full(folio)))
155 ret = folio_free_swap(folio);
162 static inline struct swap_extent *first_se(struct swap_info_struct *sis)
164 struct rb_node *rb = rb_first(&sis->swap_extent_root);
165 return rb_entry(rb, struct swap_extent, rb_node);
168 static inline struct swap_extent *next_se(struct swap_extent *se)
170 struct rb_node *rb = rb_next(&se->rb_node);
171 return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
175 * swapon tell device that all the old swap contents can be discarded,
176 * to allow the swap device to optimize its wear-levelling.
178 static int discard_swap(struct swap_info_struct *si)
180 struct swap_extent *se;
181 sector_t start_block;
185 /* Do not discard the swap header page! */
187 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
188 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
190 err = blkdev_issue_discard(si->bdev, start_block,
191 nr_blocks, GFP_KERNEL);
197 for (se = next_se(se); se; se = next_se(se)) {
198 start_block = se->start_block << (PAGE_SHIFT - 9);
199 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
201 err = blkdev_issue_discard(si->bdev, start_block,
202 nr_blocks, GFP_KERNEL);
208 return err; /* That will often be -EOPNOTSUPP */
211 static struct swap_extent *
212 offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
214 struct swap_extent *se;
217 rb = sis->swap_extent_root.rb_node;
219 se = rb_entry(rb, struct swap_extent, rb_node);
220 if (offset < se->start_page)
222 else if (offset >= se->start_page + se->nr_pages)
227 /* It *must* be present */
231 sector_t swap_folio_sector(struct folio *folio)
233 struct swap_info_struct *sis = swp_swap_info(folio->swap);
234 struct swap_extent *se;
238 offset = swp_offset(folio->swap);
239 se = offset_to_swap_extent(sis, offset);
240 sector = se->start_block + (offset - se->start_page);
241 return sector << (PAGE_SHIFT - 9);
245 * swap allocation tell device that a cluster of swap can now be discarded,
246 * to allow the swap device to optimize its wear-levelling.
248 static void discard_swap_cluster(struct swap_info_struct *si,
249 pgoff_t start_page, pgoff_t nr_pages)
251 struct swap_extent *se = offset_to_swap_extent(si, start_page);
254 pgoff_t offset = start_page - se->start_page;
255 sector_t start_block = se->start_block + offset;
256 sector_t nr_blocks = se->nr_pages - offset;
258 if (nr_blocks > nr_pages)
259 nr_blocks = nr_pages;
260 start_page += nr_blocks;
261 nr_pages -= nr_blocks;
263 start_block <<= PAGE_SHIFT - 9;
264 nr_blocks <<= PAGE_SHIFT - 9;
265 if (blkdev_issue_discard(si->bdev, start_block,
266 nr_blocks, GFP_NOIO))
273 #ifdef CONFIG_THP_SWAP
274 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
276 #define swap_entry_size(size) (size)
278 #define SWAPFILE_CLUSTER 256
281 * Define swap_entry_size() as constant to let compiler to optimize
282 * out some code if !CONFIG_THP_SWAP
284 #define swap_entry_size(size) 1
286 #define LATENCY_LIMIT 256
288 static inline void cluster_set_flag(struct swap_cluster_info *info,
294 static inline unsigned int cluster_count(struct swap_cluster_info *info)
299 static inline void cluster_set_count(struct swap_cluster_info *info,
305 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
306 unsigned int c, unsigned int f)
312 static inline unsigned int cluster_next(struct swap_cluster_info *info)
317 static inline void cluster_set_next(struct swap_cluster_info *info,
323 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
324 unsigned int n, unsigned int f)
330 static inline bool cluster_is_free(struct swap_cluster_info *info)
332 return info->flags & CLUSTER_FLAG_FREE;
335 static inline bool cluster_is_null(struct swap_cluster_info *info)
337 return info->flags & CLUSTER_FLAG_NEXT_NULL;
340 static inline void cluster_set_null(struct swap_cluster_info *info)
342 info->flags = CLUSTER_FLAG_NEXT_NULL;
346 static inline bool cluster_is_huge(struct swap_cluster_info *info)
348 if (IS_ENABLED(CONFIG_THP_SWAP))
349 return info->flags & CLUSTER_FLAG_HUGE;
353 static inline void cluster_clear_huge(struct swap_cluster_info *info)
355 info->flags &= ~CLUSTER_FLAG_HUGE;
358 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
359 unsigned long offset)
361 struct swap_cluster_info *ci;
363 ci = si->cluster_info;
365 ci += offset / SWAPFILE_CLUSTER;
366 spin_lock(&ci->lock);
371 static inline void unlock_cluster(struct swap_cluster_info *ci)
374 spin_unlock(&ci->lock);
378 * Determine the locking method in use for this device. Return
379 * swap_cluster_info if SSD-style cluster-based locking is in place.
381 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
382 struct swap_info_struct *si, unsigned long offset)
384 struct swap_cluster_info *ci;
386 /* Try to use fine-grained SSD-style locking if available: */
387 ci = lock_cluster(si, offset);
388 /* Otherwise, fall back to traditional, coarse locking: */
390 spin_lock(&si->lock);
395 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
396 struct swap_cluster_info *ci)
401 spin_unlock(&si->lock);
404 static inline bool cluster_list_empty(struct swap_cluster_list *list)
406 return cluster_is_null(&list->head);
409 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
411 return cluster_next(&list->head);
414 static void cluster_list_init(struct swap_cluster_list *list)
416 cluster_set_null(&list->head);
417 cluster_set_null(&list->tail);
420 static void cluster_list_add_tail(struct swap_cluster_list *list,
421 struct swap_cluster_info *ci,
424 if (cluster_list_empty(list)) {
425 cluster_set_next_flag(&list->head, idx, 0);
426 cluster_set_next_flag(&list->tail, idx, 0);
428 struct swap_cluster_info *ci_tail;
429 unsigned int tail = cluster_next(&list->tail);
432 * Nested cluster lock, but both cluster locks are
433 * only acquired when we held swap_info_struct->lock
436 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
437 cluster_set_next(ci_tail, idx);
438 spin_unlock(&ci_tail->lock);
439 cluster_set_next_flag(&list->tail, idx, 0);
443 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
444 struct swap_cluster_info *ci)
448 idx = cluster_next(&list->head);
449 if (cluster_next(&list->tail) == idx) {
450 cluster_set_null(&list->head);
451 cluster_set_null(&list->tail);
453 cluster_set_next_flag(&list->head,
454 cluster_next(&ci[idx]), 0);
459 /* Add a cluster to discard list and schedule it to do discard */
460 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
464 * If scan_swap_map_slots() can't find a free cluster, it will check
465 * si->swap_map directly. To make sure the discarding cluster isn't
466 * taken by scan_swap_map_slots(), mark the swap entries bad (occupied).
467 * It will be cleared after discard
469 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
470 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
472 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
474 schedule_work(&si->discard_work);
477 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
479 struct swap_cluster_info *ci = si->cluster_info;
481 cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
482 cluster_list_add_tail(&si->free_clusters, ci, idx);
486 * Doing discard actually. After a cluster discard is finished, the cluster
487 * will be added to free cluster list. caller should hold si->lock.
489 static void swap_do_scheduled_discard(struct swap_info_struct *si)
491 struct swap_cluster_info *info, *ci;
494 info = si->cluster_info;
496 while (!cluster_list_empty(&si->discard_clusters)) {
497 idx = cluster_list_del_first(&si->discard_clusters, info);
498 spin_unlock(&si->lock);
500 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
503 spin_lock(&si->lock);
504 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
505 __free_cluster(si, idx);
506 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
507 0, SWAPFILE_CLUSTER);
512 static void swap_discard_work(struct work_struct *work)
514 struct swap_info_struct *si;
516 si = container_of(work, struct swap_info_struct, discard_work);
518 spin_lock(&si->lock);
519 swap_do_scheduled_discard(si);
520 spin_unlock(&si->lock);
523 static void swap_users_ref_free(struct percpu_ref *ref)
525 struct swap_info_struct *si;
527 si = container_of(ref, struct swap_info_struct, users);
531 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
533 struct swap_cluster_info *ci = si->cluster_info;
535 VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
536 cluster_list_del_first(&si->free_clusters, ci);
537 cluster_set_count_flag(ci + idx, 0, 0);
540 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
542 struct swap_cluster_info *ci = si->cluster_info + idx;
544 VM_BUG_ON(cluster_count(ci) != 0);
546 * If the swap is discardable, prepare discard the cluster
547 * instead of free it immediately. The cluster will be freed
550 if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
551 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
552 swap_cluster_schedule_discard(si, idx);
556 __free_cluster(si, idx);
560 * The cluster corresponding to page_nr will be used. The cluster will be
561 * removed from free cluster list and its usage counter will be increased.
563 static void inc_cluster_info_page(struct swap_info_struct *p,
564 struct swap_cluster_info *cluster_info, unsigned long page_nr)
566 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
570 if (cluster_is_free(&cluster_info[idx]))
571 alloc_cluster(p, idx);
573 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
574 cluster_set_count(&cluster_info[idx],
575 cluster_count(&cluster_info[idx]) + 1);
579 * The cluster corresponding to page_nr decreases one usage. If the usage
580 * counter becomes 0, which means no page in the cluster is in using, we can
581 * optionally discard the cluster and add it to free cluster list.
583 static void dec_cluster_info_page(struct swap_info_struct *p,
584 struct swap_cluster_info *cluster_info, unsigned long page_nr)
586 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
591 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
592 cluster_set_count(&cluster_info[idx],
593 cluster_count(&cluster_info[idx]) - 1);
595 if (cluster_count(&cluster_info[idx]) == 0)
596 free_cluster(p, idx);
600 * It's possible scan_swap_map_slots() uses a free cluster in the middle of free
601 * cluster list. Avoiding such abuse to avoid list corruption.
604 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
605 unsigned long offset)
607 struct percpu_cluster *percpu_cluster;
610 offset /= SWAPFILE_CLUSTER;
611 conflict = !cluster_list_empty(&si->free_clusters) &&
612 offset != cluster_list_first(&si->free_clusters) &&
613 cluster_is_free(&si->cluster_info[offset]);
618 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
619 cluster_set_null(&percpu_cluster->index);
624 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
625 * might involve allocating a new cluster for current CPU too.
627 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
628 unsigned long *offset, unsigned long *scan_base)
630 struct percpu_cluster *cluster;
631 struct swap_cluster_info *ci;
632 unsigned long tmp, max;
635 cluster = this_cpu_ptr(si->percpu_cluster);
636 if (cluster_is_null(&cluster->index)) {
637 if (!cluster_list_empty(&si->free_clusters)) {
638 cluster->index = si->free_clusters.head;
639 cluster->next = cluster_next(&cluster->index) *
641 } else if (!cluster_list_empty(&si->discard_clusters)) {
643 * we don't have free cluster but have some clusters in
644 * discarding, do discard now and reclaim them, then
645 * reread cluster_next_cpu since we dropped si->lock
647 swap_do_scheduled_discard(si);
648 *scan_base = this_cpu_read(*si->cluster_next_cpu);
649 *offset = *scan_base;
656 * Other CPUs can use our cluster if they can't find a free cluster,
657 * check if there is still free entry in the cluster
660 max = min_t(unsigned long, si->max,
661 (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
663 ci = lock_cluster(si, tmp);
665 if (!si->swap_map[tmp])
672 cluster_set_null(&cluster->index);
675 cluster->next = tmp + 1;
681 static void __del_from_avail_list(struct swap_info_struct *p)
685 assert_spin_locked(&p->lock);
687 plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
690 static void del_from_avail_list(struct swap_info_struct *p)
692 spin_lock(&swap_avail_lock);
693 __del_from_avail_list(p);
694 spin_unlock(&swap_avail_lock);
697 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
698 unsigned int nr_entries)
700 unsigned int end = offset + nr_entries - 1;
702 if (offset == si->lowest_bit)
703 si->lowest_bit += nr_entries;
704 if (end == si->highest_bit)
705 WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries);
706 WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries);
707 if (si->inuse_pages == si->pages) {
708 si->lowest_bit = si->max;
710 del_from_avail_list(si);
714 static void add_to_avail_list(struct swap_info_struct *p)
718 spin_lock(&swap_avail_lock);
720 plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
721 spin_unlock(&swap_avail_lock);
724 static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
725 unsigned int nr_entries)
727 unsigned long begin = offset;
728 unsigned long end = offset + nr_entries - 1;
729 void (*swap_slot_free_notify)(struct block_device *, unsigned long);
731 if (offset < si->lowest_bit)
732 si->lowest_bit = offset;
733 if (end > si->highest_bit) {
734 bool was_full = !si->highest_bit;
736 WRITE_ONCE(si->highest_bit, end);
737 if (was_full && (si->flags & SWP_WRITEOK))
738 add_to_avail_list(si);
740 if (si->flags & SWP_BLKDEV)
741 swap_slot_free_notify =
742 si->bdev->bd_disk->fops->swap_slot_free_notify;
744 swap_slot_free_notify = NULL;
745 while (offset <= end) {
746 arch_swap_invalidate_page(si->type, offset);
747 if (swap_slot_free_notify)
748 swap_slot_free_notify(si->bdev, offset);
751 clear_shadow_from_swap_cache(si->type, begin, end);
754 * Make sure that try_to_unuse() observes si->inuse_pages reaching 0
755 * only after the above cleanups are done.
758 atomic_long_add(nr_entries, &nr_swap_pages);
759 WRITE_ONCE(si->inuse_pages, si->inuse_pages - nr_entries);
762 static void set_cluster_next(struct swap_info_struct *si, unsigned long next)
766 if (!(si->flags & SWP_SOLIDSTATE)) {
767 si->cluster_next = next;
771 prev = this_cpu_read(*si->cluster_next_cpu);
773 * Cross the swap address space size aligned trunk, choose
774 * another trunk randomly to avoid lock contention on swap
775 * address space if possible.
777 if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) !=
778 (next >> SWAP_ADDRESS_SPACE_SHIFT)) {
779 /* No free swap slots available */
780 if (si->highest_bit <= si->lowest_bit)
782 next = get_random_u32_inclusive(si->lowest_bit, si->highest_bit);
783 next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES);
784 next = max_t(unsigned int, next, si->lowest_bit);
786 this_cpu_write(*si->cluster_next_cpu, next);
789 static bool swap_offset_available_and_locked(struct swap_info_struct *si,
790 unsigned long offset)
792 if (data_race(!si->swap_map[offset])) {
793 spin_lock(&si->lock);
797 if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
798 spin_lock(&si->lock);
805 static int scan_swap_map_slots(struct swap_info_struct *si,
806 unsigned char usage, int nr,
809 struct swap_cluster_info *ci;
810 unsigned long offset;
811 unsigned long scan_base;
812 unsigned long last_in_cluster = 0;
813 int latency_ration = LATENCY_LIMIT;
815 bool scanned_many = false;
818 * We try to cluster swap pages by allocating them sequentially
819 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
820 * way, however, we resort to first-free allocation, starting
821 * a new cluster. This prevents us from scattering swap pages
822 * all over the entire swap partition, so that we reduce
823 * overall disk seek times between swap pages. -- sct
824 * But we do now try to find an empty cluster. -Andrea
825 * And we let swap pages go all over an SSD partition. Hugh
828 si->flags += SWP_SCANNING;
830 * Use percpu scan base for SSD to reduce lock contention on
831 * cluster and swap cache. For HDD, sequential access is more
834 if (si->flags & SWP_SOLIDSTATE)
835 scan_base = this_cpu_read(*si->cluster_next_cpu);
837 scan_base = si->cluster_next;
841 if (si->cluster_info) {
842 if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
844 } else if (unlikely(!si->cluster_nr--)) {
845 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
846 si->cluster_nr = SWAPFILE_CLUSTER - 1;
850 spin_unlock(&si->lock);
853 * If seek is expensive, start searching for new cluster from
854 * start of partition, to minimize the span of allocated swap.
855 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
856 * case, just handled by scan_swap_map_try_ssd_cluster() above.
858 scan_base = offset = si->lowest_bit;
859 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
861 /* Locate the first empty (unaligned) cluster */
862 for (; last_in_cluster <= si->highest_bit; offset++) {
863 if (si->swap_map[offset])
864 last_in_cluster = offset + SWAPFILE_CLUSTER;
865 else if (offset == last_in_cluster) {
866 spin_lock(&si->lock);
867 offset -= SWAPFILE_CLUSTER - 1;
868 si->cluster_next = offset;
869 si->cluster_nr = SWAPFILE_CLUSTER - 1;
872 if (unlikely(--latency_ration < 0)) {
874 latency_ration = LATENCY_LIMIT;
879 spin_lock(&si->lock);
880 si->cluster_nr = SWAPFILE_CLUSTER - 1;
884 if (si->cluster_info) {
885 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
886 /* take a break if we already got some slots */
889 if (!scan_swap_map_try_ssd_cluster(si, &offset,
894 if (!(si->flags & SWP_WRITEOK))
896 if (!si->highest_bit)
898 if (offset > si->highest_bit)
899 scan_base = offset = si->lowest_bit;
901 ci = lock_cluster(si, offset);
902 /* reuse swap entry of cache-only swap if not busy. */
903 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
906 spin_unlock(&si->lock);
907 swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
908 spin_lock(&si->lock);
909 /* entry was freed successfully, try to use this again */
912 goto scan; /* check next one */
915 if (si->swap_map[offset]) {
922 WRITE_ONCE(si->swap_map[offset], usage);
923 inc_cluster_info_page(si, si->cluster_info, offset);
926 swap_range_alloc(si, offset, 1);
927 slots[n_ret++] = swp_entry(si->type, offset);
929 /* got enough slots or reach max slots? */
930 if ((n_ret == nr) || (offset >= si->highest_bit))
933 /* search for next available slot */
935 /* time to take a break? */
936 if (unlikely(--latency_ration < 0)) {
939 spin_unlock(&si->lock);
941 spin_lock(&si->lock);
942 latency_ration = LATENCY_LIMIT;
945 /* try to get more slots in cluster */
946 if (si->cluster_info) {
947 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
949 } else if (si->cluster_nr && !si->swap_map[++offset]) {
950 /* non-ssd case, still more slots in cluster? */
956 * Even if there's no free clusters available (fragmented),
957 * try to scan a little more quickly with lock held unless we
958 * have scanned too many slots already.
961 unsigned long scan_limit;
963 if (offset < scan_base)
964 scan_limit = scan_base;
966 scan_limit = si->highest_bit;
967 for (; offset <= scan_limit && --latency_ration > 0;
969 if (!si->swap_map[offset])
975 set_cluster_next(si, offset + 1);
976 si->flags -= SWP_SCANNING;
980 spin_unlock(&si->lock);
981 while (++offset <= READ_ONCE(si->highest_bit)) {
982 if (unlikely(--latency_ration < 0)) {
984 latency_ration = LATENCY_LIMIT;
987 if (swap_offset_available_and_locked(si, offset))
990 offset = si->lowest_bit;
991 while (offset < scan_base) {
992 if (unlikely(--latency_ration < 0)) {
994 latency_ration = LATENCY_LIMIT;
997 if (swap_offset_available_and_locked(si, offset))
1001 spin_lock(&si->lock);
1004 si->flags -= SWP_SCANNING;
1008 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
1011 struct swap_cluster_info *ci;
1012 unsigned long offset;
1015 * Should not even be attempting cluster allocations when huge
1016 * page swap is disabled. Warn and fail the allocation.
1018 if (!IS_ENABLED(CONFIG_THP_SWAP)) {
1023 if (cluster_list_empty(&si->free_clusters))
1026 idx = cluster_list_first(&si->free_clusters);
1027 offset = idx * SWAPFILE_CLUSTER;
1028 ci = lock_cluster(si, offset);
1029 alloc_cluster(si, idx);
1030 cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
1032 memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER);
1034 swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
1035 *slot = swp_entry(si->type, offset);
1040 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
1042 unsigned long offset = idx * SWAPFILE_CLUSTER;
1043 struct swap_cluster_info *ci;
1045 ci = lock_cluster(si, offset);
1046 memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
1047 cluster_set_count_flag(ci, 0, 0);
1048 free_cluster(si, idx);
1050 swap_range_free(si, offset, SWAPFILE_CLUSTER);
1053 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
1055 unsigned long size = swap_entry_size(entry_size);
1056 struct swap_info_struct *si, *next;
1061 /* Only single cluster request supported */
1062 WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1064 spin_lock(&swap_avail_lock);
1066 avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1067 if (avail_pgs <= 0) {
1068 spin_unlock(&swap_avail_lock);
1072 n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
1074 atomic_long_sub(n_goal * size, &nr_swap_pages);
1077 node = numa_node_id();
1078 plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1079 /* requeue si to after same-priority siblings */
1080 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1081 spin_unlock(&swap_avail_lock);
1082 spin_lock(&si->lock);
1083 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1084 spin_lock(&swap_avail_lock);
1085 if (plist_node_empty(&si->avail_lists[node])) {
1086 spin_unlock(&si->lock);
1089 WARN(!si->highest_bit,
1090 "swap_info %d in list but !highest_bit\n",
1092 WARN(!(si->flags & SWP_WRITEOK),
1093 "swap_info %d in list but !SWP_WRITEOK\n",
1095 __del_from_avail_list(si);
1096 spin_unlock(&si->lock);
1099 if (size == SWAPFILE_CLUSTER) {
1100 if (si->flags & SWP_BLKDEV)
1101 n_ret = swap_alloc_cluster(si, swp_entries);
1103 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1104 n_goal, swp_entries);
1105 spin_unlock(&si->lock);
1106 if (n_ret || size == SWAPFILE_CLUSTER)
1110 spin_lock(&swap_avail_lock);
1113 * if we got here, it's likely that si was almost full before,
1114 * and since scan_swap_map_slots() can drop the si->lock,
1115 * multiple callers probably all tried to get a page from the
1116 * same si and it filled up before we could get one; or, the si
1117 * filled up between us dropping swap_avail_lock and taking
1118 * si->lock. Since we dropped the swap_avail_lock, the
1119 * swap_avail_head list may have been modified; so if next is
1120 * still in the swap_avail_head list then try it, otherwise
1121 * start over if we have not gotten any slots.
1123 if (plist_node_empty(&next->avail_lists[node]))
1127 spin_unlock(&swap_avail_lock);
1131 atomic_long_add((long)(n_goal - n_ret) * size,
1137 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1139 struct swap_info_struct *p;
1140 unsigned long offset;
1144 p = swp_swap_info(entry);
1147 if (data_race(!(p->flags & SWP_USED)))
1149 offset = swp_offset(entry);
1150 if (offset >= p->max)
1152 if (data_race(!p->swap_map[swp_offset(entry)]))
1157 pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val);
1160 pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1163 pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val);
1166 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1171 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1172 struct swap_info_struct *q)
1174 struct swap_info_struct *p;
1176 p = _swap_info_get(entry);
1180 spin_unlock(&q->lock);
1182 spin_lock(&p->lock);
1187 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1188 unsigned long offset,
1189 unsigned char usage)
1191 unsigned char count;
1192 unsigned char has_cache;
1194 count = p->swap_map[offset];
1196 has_cache = count & SWAP_HAS_CACHE;
1197 count &= ~SWAP_HAS_CACHE;
1199 if (usage == SWAP_HAS_CACHE) {
1200 VM_BUG_ON(!has_cache);
1202 } else if (count == SWAP_MAP_SHMEM) {
1204 * Or we could insist on shmem.c using a special
1205 * swap_shmem_free() and free_shmem_swap_and_cache()...
1208 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1209 if (count == COUNT_CONTINUED) {
1210 if (swap_count_continued(p, offset, count))
1211 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1213 count = SWAP_MAP_MAX;
1218 usage = count | has_cache;
1220 WRITE_ONCE(p->swap_map[offset], usage);
1222 WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE);
1228 * When we get a swap entry, if there aren't some other ways to
1229 * prevent swapoff, such as the folio in swap cache is locked, page
1230 * table lock is held, etc., the swap entry may become invalid because
1231 * of swapoff. Then, we need to enclose all swap related functions
1232 * with get_swap_device() and put_swap_device(), unless the swap
1233 * functions call get/put_swap_device() by themselves.
1235 * Note that when only holding the PTL, swapoff might succeed immediately
1236 * after freeing a swap entry. Therefore, immediately after
1237 * __swap_entry_free(), the swap info might become stale and should not
1238 * be touched without a prior get_swap_device().
1240 * Check whether swap entry is valid in the swap device. If so,
1241 * return pointer to swap_info_struct, and keep the swap entry valid
1242 * via preventing the swap device from being swapoff, until
1243 * put_swap_device() is called. Otherwise return NULL.
1245 * Notice that swapoff or swapoff+swapon can still happen before the
1246 * percpu_ref_tryget_live() in get_swap_device() or after the
1247 * percpu_ref_put() in put_swap_device() if there isn't any other way
1248 * to prevent swapoff. The caller must be prepared for that. For
1249 * example, the following situation is possible.
1253 * ... swapoff+swapon
1254 * __read_swap_cache_async()
1255 * swapcache_prepare()
1256 * __swap_duplicate()
1258 * // verify PTE not changed
1260 * In __swap_duplicate(), the swap_map need to be checked before
1261 * changing partly because the specified swap entry may be for another
1262 * swap device which has been swapoff. And in do_swap_page(), after
1263 * the page is read from the swap device, the PTE is verified not
1264 * changed with the page table locked to check whether the swap device
1265 * has been swapoff or swapoff+swapon.
1267 struct swap_info_struct *get_swap_device(swp_entry_t entry)
1269 struct swap_info_struct *si;
1270 unsigned long offset;
1274 si = swp_swap_info(entry);
1277 if (!percpu_ref_tryget_live(&si->users))
1280 * Guarantee the si->users are checked before accessing other
1281 * fields of swap_info_struct.
1283 * Paired with the spin_unlock() after setup_swap_info() in
1284 * enable_swap_info().
1287 offset = swp_offset(entry);
1288 if (offset >= si->max)
1293 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1297 pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1298 percpu_ref_put(&si->users);
1302 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1305 struct swap_cluster_info *ci;
1306 unsigned long offset = swp_offset(entry);
1307 unsigned char usage;
1309 ci = lock_cluster_or_swap_info(p, offset);
1310 usage = __swap_entry_free_locked(p, offset, 1);
1311 unlock_cluster_or_swap_info(p, ci);
1313 free_swap_slot(entry);
1318 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1320 struct swap_cluster_info *ci;
1321 unsigned long offset = swp_offset(entry);
1322 unsigned char count;
1324 ci = lock_cluster(p, offset);
1325 count = p->swap_map[offset];
1326 VM_BUG_ON(count != SWAP_HAS_CACHE);
1327 p->swap_map[offset] = 0;
1328 dec_cluster_info_page(p, p->cluster_info, offset);
1331 mem_cgroup_uncharge_swap(entry, 1);
1332 swap_range_free(p, offset, 1);
1336 * Caller has made sure that the swap device corresponding to entry
1337 * is still around or has not been recycled.
1339 void swap_free(swp_entry_t entry)
1341 struct swap_info_struct *p;
1343 p = _swap_info_get(entry);
1345 __swap_entry_free(p, entry);
1349 * Called after dropping swapcache to decrease refcnt to swap entries.
1351 void put_swap_folio(struct folio *folio, swp_entry_t entry)
1353 unsigned long offset = swp_offset(entry);
1354 unsigned long idx = offset / SWAPFILE_CLUSTER;
1355 struct swap_cluster_info *ci;
1356 struct swap_info_struct *si;
1358 unsigned int i, free_entries = 0;
1360 int size = swap_entry_size(folio_nr_pages(folio));
1362 si = _swap_info_get(entry);
1366 ci = lock_cluster_or_swap_info(si, offset);
1367 if (size == SWAPFILE_CLUSTER) {
1368 VM_BUG_ON(!cluster_is_huge(ci));
1369 map = si->swap_map + offset;
1370 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1372 VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1373 if (val == SWAP_HAS_CACHE)
1376 cluster_clear_huge(ci);
1377 if (free_entries == SWAPFILE_CLUSTER) {
1378 unlock_cluster_or_swap_info(si, ci);
1379 spin_lock(&si->lock);
1380 mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1381 swap_free_cluster(si, idx);
1382 spin_unlock(&si->lock);
1386 for (i = 0; i < size; i++, entry.val++) {
1387 if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1388 unlock_cluster_or_swap_info(si, ci);
1389 free_swap_slot(entry);
1392 lock_cluster_or_swap_info(si, offset);
1395 unlock_cluster_or_swap_info(si, ci);
1398 #ifdef CONFIG_THP_SWAP
1399 int split_swap_cluster(swp_entry_t entry)
1401 struct swap_info_struct *si;
1402 struct swap_cluster_info *ci;
1403 unsigned long offset = swp_offset(entry);
1405 si = _swap_info_get(entry);
1408 ci = lock_cluster(si, offset);
1409 cluster_clear_huge(ci);
1415 static int swp_entry_cmp(const void *ent1, const void *ent2)
1417 const swp_entry_t *e1 = ent1, *e2 = ent2;
1419 return (int)swp_type(*e1) - (int)swp_type(*e2);
1422 void swapcache_free_entries(swp_entry_t *entries, int n)
1424 struct swap_info_struct *p, *prev;
1434 * Sort swap entries by swap device, so each lock is only taken once.
1435 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1436 * so low that it isn't necessary to optimize further.
1438 if (nr_swapfiles > 1)
1439 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1440 for (i = 0; i < n; ++i) {
1441 p = swap_info_get_cont(entries[i], prev);
1443 swap_entry_free(p, entries[i]);
1447 spin_unlock(&p->lock);
1450 int __swap_count(swp_entry_t entry)
1452 struct swap_info_struct *si = swp_swap_info(entry);
1453 pgoff_t offset = swp_offset(entry);
1455 return swap_count(si->swap_map[offset]);
1459 * How many references to @entry are currently swapped out?
1460 * This does not give an exact answer when swap count is continued,
1461 * but does include the high COUNT_CONTINUED flag to allow for that.
1463 int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1465 pgoff_t offset = swp_offset(entry);
1466 struct swap_cluster_info *ci;
1469 ci = lock_cluster_or_swap_info(si, offset);
1470 count = swap_count(si->swap_map[offset]);
1471 unlock_cluster_or_swap_info(si, ci);
1476 * How many references to @entry are currently swapped out?
1477 * This considers COUNT_CONTINUED so it returns exact answer.
1479 int swp_swapcount(swp_entry_t entry)
1481 int count, tmp_count, n;
1482 struct swap_info_struct *p;
1483 struct swap_cluster_info *ci;
1488 p = _swap_info_get(entry);
1492 offset = swp_offset(entry);
1494 ci = lock_cluster_or_swap_info(p, offset);
1496 count = swap_count(p->swap_map[offset]);
1497 if (!(count & COUNT_CONTINUED))
1500 count &= ~COUNT_CONTINUED;
1501 n = SWAP_MAP_MAX + 1;
1503 page = vmalloc_to_page(p->swap_map + offset);
1504 offset &= ~PAGE_MASK;
1505 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1508 page = list_next_entry(page, lru);
1509 map = kmap_local_page(page);
1510 tmp_count = map[offset];
1513 count += (tmp_count & ~COUNT_CONTINUED) * n;
1514 n *= (SWAP_CONT_MAX + 1);
1515 } while (tmp_count & COUNT_CONTINUED);
1517 unlock_cluster_or_swap_info(p, ci);
1521 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1524 struct swap_cluster_info *ci;
1525 unsigned char *map = si->swap_map;
1526 unsigned long roffset = swp_offset(entry);
1527 unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1531 ci = lock_cluster_or_swap_info(si, offset);
1532 if (!ci || !cluster_is_huge(ci)) {
1533 if (swap_count(map[roffset]))
1537 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1538 if (swap_count(map[offset + i])) {
1544 unlock_cluster_or_swap_info(si, ci);
1548 static bool folio_swapped(struct folio *folio)
1550 swp_entry_t entry = folio->swap;
1551 struct swap_info_struct *si = _swap_info_get(entry);
1556 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio)))
1557 return swap_swapcount(si, entry) != 0;
1559 return swap_page_trans_huge_swapped(si, entry);
1563 * folio_free_swap() - Free the swap space used for this folio.
1564 * @folio: The folio to remove.
1566 * If swap is getting full, or if there are no more mappings of this folio,
1567 * then call folio_free_swap to free its swap space.
1569 * Return: true if we were able to release the swap space.
1571 bool folio_free_swap(struct folio *folio)
1573 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1575 if (!folio_test_swapcache(folio))
1577 if (folio_test_writeback(folio))
1579 if (folio_swapped(folio))
1583 * Once hibernation has begun to create its image of memory,
1584 * there's a danger that one of the calls to folio_free_swap()
1585 * - most probably a call from __try_to_reclaim_swap() while
1586 * hibernation is allocating its own swap pages for the image,
1587 * but conceivably even a call from memory reclaim - will free
1588 * the swap from a folio which has already been recorded in the
1589 * image as a clean swapcache folio, and then reuse its swap for
1590 * another page of the image. On waking from hibernation, the
1591 * original folio might be freed under memory pressure, then
1592 * later read back in from swap, now with the wrong data.
1594 * Hibernation suspends storage while it is writing the image
1595 * to disk so check that here.
1597 if (pm_suspended_storage())
1600 delete_from_swap_cache(folio);
1601 folio_set_dirty(folio);
1606 * Free the swap entry like above, but also try to
1607 * free the page cache entry if it is the last user.
1609 int free_swap_and_cache(swp_entry_t entry)
1611 struct swap_info_struct *p;
1612 unsigned char count;
1614 if (non_swap_entry(entry))
1617 p = get_swap_device(entry);
1619 if (WARN_ON(data_race(!p->swap_map[swp_offset(entry)]))) {
1624 count = __swap_entry_free(p, entry);
1625 if (count == SWAP_HAS_CACHE &&
1626 !swap_page_trans_huge_swapped(p, entry))
1627 __try_to_reclaim_swap(p, swp_offset(entry),
1628 TTRS_UNMAPPED | TTRS_FULL);
1634 #ifdef CONFIG_HIBERNATION
1636 swp_entry_t get_swap_page_of_type(int type)
1638 struct swap_info_struct *si = swap_type_to_swap_info(type);
1639 swp_entry_t entry = {0};
1644 /* This is called for allocating swap entry, not cache */
1645 spin_lock(&si->lock);
1646 if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry))
1647 atomic_long_dec(&nr_swap_pages);
1648 spin_unlock(&si->lock);
1654 * Find the swap type that corresponds to given device (if any).
1656 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1657 * from 0, in which the swap header is expected to be located.
1659 * This is needed for the suspend to disk (aka swsusp).
1661 int swap_type_of(dev_t device, sector_t offset)
1668 spin_lock(&swap_lock);
1669 for (type = 0; type < nr_swapfiles; type++) {
1670 struct swap_info_struct *sis = swap_info[type];
1672 if (!(sis->flags & SWP_WRITEOK))
1675 if (device == sis->bdev->bd_dev) {
1676 struct swap_extent *se = first_se(sis);
1678 if (se->start_block == offset) {
1679 spin_unlock(&swap_lock);
1684 spin_unlock(&swap_lock);
1688 int find_first_swap(dev_t *device)
1692 spin_lock(&swap_lock);
1693 for (type = 0; type < nr_swapfiles; type++) {
1694 struct swap_info_struct *sis = swap_info[type];
1696 if (!(sis->flags & SWP_WRITEOK))
1698 *device = sis->bdev->bd_dev;
1699 spin_unlock(&swap_lock);
1702 spin_unlock(&swap_lock);
1707 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1708 * corresponding to given index in swap_info (swap type).
1710 sector_t swapdev_block(int type, pgoff_t offset)
1712 struct swap_info_struct *si = swap_type_to_swap_info(type);
1713 struct swap_extent *se;
1715 if (!si || !(si->flags & SWP_WRITEOK))
1717 se = offset_to_swap_extent(si, offset);
1718 return se->start_block + (offset - se->start_page);
1722 * Return either the total number of swap pages of given type, or the number
1723 * of free pages of that type (depending on @free)
1725 * This is needed for software suspend
1727 unsigned int count_swap_pages(int type, int free)
1731 spin_lock(&swap_lock);
1732 if ((unsigned int)type < nr_swapfiles) {
1733 struct swap_info_struct *sis = swap_info[type];
1735 spin_lock(&sis->lock);
1736 if (sis->flags & SWP_WRITEOK) {
1739 n -= sis->inuse_pages;
1741 spin_unlock(&sis->lock);
1743 spin_unlock(&swap_lock);
1746 #endif /* CONFIG_HIBERNATION */
1748 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1750 return pte_same(pte_swp_clear_flags(pte), swp_pte);
1754 * No need to decide whether this PTE shares the swap entry with others,
1755 * just let do_wp_page work it out if a write is requested later - to
1756 * force COW, vm_page_prot omits write permission from any private vma.
1758 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1759 unsigned long addr, swp_entry_t entry, struct folio *folio)
1762 struct folio *swapcache;
1764 pte_t *pte, new_pte, old_pte;
1765 bool hwpoisoned = false;
1769 folio = ksm_might_need_to_copy(folio, vma, addr);
1770 if (unlikely(!folio))
1772 else if (unlikely(folio == ERR_PTR(-EHWPOISON))) {
1777 page = folio_file_page(folio, swp_offset(entry));
1778 if (PageHWPoison(page))
1781 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1782 if (unlikely(!pte || !pte_same_as_swp(ptep_get(pte),
1783 swp_entry_to_pte(entry)))) {
1788 old_pte = ptep_get(pte);
1790 if (unlikely(hwpoisoned || !folio_test_uptodate(folio))) {
1791 swp_entry_t swp_entry;
1793 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1795 swp_entry = make_hwpoison_entry(page);
1797 swp_entry = make_poisoned_swp_entry();
1799 new_pte = swp_entry_to_pte(swp_entry);
1805 * Some architectures may have to restore extra metadata to the page
1806 * when reading from swap. This metadata may be indexed by swap entry
1807 * so this must be called before swap_free().
1809 arch_swap_restore(entry, folio);
1811 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1812 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1814 if (folio == swapcache) {
1815 rmap_t rmap_flags = RMAP_NONE;
1818 * See do_swap_page(): writeback would be problematic.
1819 * However, we do a folio_wait_writeback() just before this
1820 * call and have the folio locked.
1822 VM_BUG_ON_FOLIO(folio_test_writeback(folio), folio);
1823 if (pte_swp_exclusive(old_pte))
1824 rmap_flags |= RMAP_EXCLUSIVE;
1826 folio_add_anon_rmap_pte(folio, page, vma, addr, rmap_flags);
1827 } else { /* ksm created a completely new copy */
1828 folio_add_new_anon_rmap(folio, vma, addr);
1829 folio_add_lru_vma(folio, vma);
1831 new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot));
1832 if (pte_swp_soft_dirty(old_pte))
1833 new_pte = pte_mksoft_dirty(new_pte);
1834 if (pte_swp_uffd_wp(old_pte))
1835 new_pte = pte_mkuffd_wp(new_pte);
1837 set_pte_at(vma->vm_mm, addr, pte, new_pte);
1841 pte_unmap_unlock(pte, ptl);
1842 if (folio != swapcache) {
1843 folio_unlock(folio);
1849 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1850 unsigned long addr, unsigned long end,
1854 struct swap_info_struct *si;
1856 si = swap_info[type];
1858 struct folio *folio;
1859 unsigned long offset;
1860 unsigned char swp_count;
1866 pte = pte_offset_map(pmd, addr);
1871 ptent = ptep_get_lockless(pte);
1873 if (!is_swap_pte(ptent))
1876 entry = pte_to_swp_entry(ptent);
1877 if (swp_type(entry) != type)
1880 offset = swp_offset(entry);
1884 folio = swap_cache_get_folio(entry, vma, addr);
1887 struct vm_fault vmf = {
1890 .real_address = addr,
1894 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1897 folio = page_folio(page);
1900 swp_count = READ_ONCE(si->swap_map[offset]);
1901 if (swp_count == 0 || swp_count == SWAP_MAP_BAD)
1907 folio_wait_writeback(folio);
1908 ret = unuse_pte(vma, pmd, addr, entry, folio);
1910 folio_unlock(folio);
1915 folio_free_swap(folio);
1916 folio_unlock(folio);
1918 } while (addr += PAGE_SIZE, addr != end);
1925 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1926 unsigned long addr, unsigned long end,
1933 pmd = pmd_offset(pud, addr);
1936 next = pmd_addr_end(addr, end);
1937 ret = unuse_pte_range(vma, pmd, addr, next, type);
1940 } while (pmd++, addr = next, addr != end);
1944 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1945 unsigned long addr, unsigned long end,
1952 pud = pud_offset(p4d, addr);
1954 next = pud_addr_end(addr, end);
1955 if (pud_none_or_clear_bad(pud))
1957 ret = unuse_pmd_range(vma, pud, addr, next, type);
1960 } while (pud++, addr = next, addr != end);
1964 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1965 unsigned long addr, unsigned long end,
1972 p4d = p4d_offset(pgd, addr);
1974 next = p4d_addr_end(addr, end);
1975 if (p4d_none_or_clear_bad(p4d))
1977 ret = unuse_pud_range(vma, p4d, addr, next, type);
1980 } while (p4d++, addr = next, addr != end);
1984 static int unuse_vma(struct vm_area_struct *vma, unsigned int type)
1987 unsigned long addr, end, next;
1990 addr = vma->vm_start;
1993 pgd = pgd_offset(vma->vm_mm, addr);
1995 next = pgd_addr_end(addr, end);
1996 if (pgd_none_or_clear_bad(pgd))
1998 ret = unuse_p4d_range(vma, pgd, addr, next, type);
2001 } while (pgd++, addr = next, addr != end);
2005 static int unuse_mm(struct mm_struct *mm, unsigned int type)
2007 struct vm_area_struct *vma;
2009 VMA_ITERATOR(vmi, mm, 0);
2012 for_each_vma(vmi, vma) {
2013 if (vma->anon_vma) {
2014 ret = unuse_vma(vma, type);
2021 mmap_read_unlock(mm);
2026 * Scan swap_map from current position to next entry still in use.
2027 * Return 0 if there are no inuse entries after prev till end of
2030 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2034 unsigned char count;
2037 * No need for swap_lock here: we're just looking
2038 * for whether an entry is in use, not modifying it; false
2039 * hits are okay, and sys_swapoff() has already prevented new
2040 * allocations from this area (while holding swap_lock).
2042 for (i = prev + 1; i < si->max; i++) {
2043 count = READ_ONCE(si->swap_map[i]);
2044 if (count && swap_count(count) != SWAP_MAP_BAD)
2046 if ((i % LATENCY_LIMIT) == 0)
2056 static int try_to_unuse(unsigned int type)
2058 struct mm_struct *prev_mm;
2059 struct mm_struct *mm;
2060 struct list_head *p;
2062 struct swap_info_struct *si = swap_info[type];
2063 struct folio *folio;
2067 if (!READ_ONCE(si->inuse_pages))
2071 retval = shmem_unuse(type);
2078 spin_lock(&mmlist_lock);
2079 p = &init_mm.mmlist;
2080 while (READ_ONCE(si->inuse_pages) &&
2081 !signal_pending(current) &&
2082 (p = p->next) != &init_mm.mmlist) {
2084 mm = list_entry(p, struct mm_struct, mmlist);
2085 if (!mmget_not_zero(mm))
2087 spin_unlock(&mmlist_lock);
2090 retval = unuse_mm(mm, type);
2097 * Make sure that we aren't completely killing
2098 * interactive performance.
2101 spin_lock(&mmlist_lock);
2103 spin_unlock(&mmlist_lock);
2108 while (READ_ONCE(si->inuse_pages) &&
2109 !signal_pending(current) &&
2110 (i = find_next_to_unuse(si, i)) != 0) {
2112 entry = swp_entry(type, i);
2113 folio = filemap_get_folio(swap_address_space(entry), i);
2118 * It is conceivable that a racing task removed this folio from
2119 * swap cache just before we acquired the page lock. The folio
2120 * might even be back in swap cache on another swap area. But
2121 * that is okay, folio_free_swap() only removes stale folios.
2124 folio_wait_writeback(folio);
2125 folio_free_swap(folio);
2126 folio_unlock(folio);
2131 * Lets check again to see if there are still swap entries in the map.
2132 * If yes, we would need to do retry the unuse logic again.
2133 * Under global memory pressure, swap entries can be reinserted back
2134 * into process space after the mmlist loop above passes over them.
2136 * Limit the number of retries? No: when mmget_not_zero()
2137 * above fails, that mm is likely to be freeing swap from
2138 * exit_mmap(), which proceeds at its own independent pace;
2139 * and even shmem_writepage() could have been preempted after
2140 * folio_alloc_swap(), temporarily hiding that swap. It's easy
2141 * and robust (though cpu-intensive) just to keep retrying.
2143 if (READ_ONCE(si->inuse_pages)) {
2144 if (!signal_pending(current))
2151 * Make sure that further cleanups after try_to_unuse() returns happen
2152 * after swap_range_free() reduces si->inuse_pages to 0.
2159 * After a successful try_to_unuse, if no swap is now in use, we know
2160 * we can empty the mmlist. swap_lock must be held on entry and exit.
2161 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2162 * added to the mmlist just after page_duplicate - before would be racy.
2164 static void drain_mmlist(void)
2166 struct list_head *p, *next;
2169 for (type = 0; type < nr_swapfiles; type++)
2170 if (swap_info[type]->inuse_pages)
2172 spin_lock(&mmlist_lock);
2173 list_for_each_safe(p, next, &init_mm.mmlist)
2175 spin_unlock(&mmlist_lock);
2179 * Free all of a swapdev's extent information
2181 static void destroy_swap_extents(struct swap_info_struct *sis)
2183 while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2184 struct rb_node *rb = sis->swap_extent_root.rb_node;
2185 struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2187 rb_erase(rb, &sis->swap_extent_root);
2191 if (sis->flags & SWP_ACTIVATED) {
2192 struct file *swap_file = sis->swap_file;
2193 struct address_space *mapping = swap_file->f_mapping;
2195 sis->flags &= ~SWP_ACTIVATED;
2196 if (mapping->a_ops->swap_deactivate)
2197 mapping->a_ops->swap_deactivate(swap_file);
2202 * Add a block range (and the corresponding page range) into this swapdev's
2205 * This function rather assumes that it is called in ascending page order.
2208 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2209 unsigned long nr_pages, sector_t start_block)
2211 struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2212 struct swap_extent *se;
2213 struct swap_extent *new_se;
2216 * place the new node at the right most since the
2217 * function is called in ascending page order.
2221 link = &parent->rb_right;
2225 se = rb_entry(parent, struct swap_extent, rb_node);
2226 BUG_ON(se->start_page + se->nr_pages != start_page);
2227 if (se->start_block + se->nr_pages == start_block) {
2229 se->nr_pages += nr_pages;
2234 /* No merge, insert a new extent. */
2235 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2238 new_se->start_page = start_page;
2239 new_se->nr_pages = nr_pages;
2240 new_se->start_block = start_block;
2242 rb_link_node(&new_se->rb_node, parent, link);
2243 rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2246 EXPORT_SYMBOL_GPL(add_swap_extent);
2249 * A `swap extent' is a simple thing which maps a contiguous range of pages
2250 * onto a contiguous range of disk blocks. A rbtree of swap extents is
2251 * built at swapon time and is then used at swap_writepage/swap_read_folio
2252 * time for locating where on disk a page belongs.
2254 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2255 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2256 * swap files identically.
2258 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2259 * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2260 * swapfiles are handled *identically* after swapon time.
2262 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2263 * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray
2264 * blocks are found which do not fall within the PAGE_SIZE alignment
2265 * requirements, they are simply tossed out - we will never use those blocks
2268 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2269 * prevents users from writing to the swap device, which will corrupt memory.
2271 * The amount of disk space which a single swap extent represents varies.
2272 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2273 * extents in the rbtree. - akpm.
2275 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2277 struct file *swap_file = sis->swap_file;
2278 struct address_space *mapping = swap_file->f_mapping;
2279 struct inode *inode = mapping->host;
2282 if (S_ISBLK(inode->i_mode)) {
2283 ret = add_swap_extent(sis, 0, sis->max, 0);
2288 if (mapping->a_ops->swap_activate) {
2289 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2292 sis->flags |= SWP_ACTIVATED;
2293 if ((sis->flags & SWP_FS_OPS) &&
2294 sio_pool_init() != 0) {
2295 destroy_swap_extents(sis);
2301 return generic_swapfile_activate(sis, swap_file, span);
2304 static int swap_node(struct swap_info_struct *p)
2306 struct block_device *bdev;
2311 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2313 return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2316 static void setup_swap_info(struct swap_info_struct *p, int prio,
2317 unsigned char *swap_map,
2318 struct swap_cluster_info *cluster_info)
2325 p->prio = --least_priority;
2327 * the plist prio is negated because plist ordering is
2328 * low-to-high, while swap ordering is high-to-low
2330 p->list.prio = -p->prio;
2333 p->avail_lists[i].prio = -p->prio;
2335 if (swap_node(p) == i)
2336 p->avail_lists[i].prio = 1;
2338 p->avail_lists[i].prio = -p->prio;
2341 p->swap_map = swap_map;
2342 p->cluster_info = cluster_info;
2345 static void _enable_swap_info(struct swap_info_struct *p)
2347 p->flags |= SWP_WRITEOK;
2348 atomic_long_add(p->pages, &nr_swap_pages);
2349 total_swap_pages += p->pages;
2351 assert_spin_locked(&swap_lock);
2353 * both lists are plists, and thus priority ordered.
2354 * swap_active_head needs to be priority ordered for swapoff(),
2355 * which on removal of any swap_info_struct with an auto-assigned
2356 * (i.e. negative) priority increments the auto-assigned priority
2357 * of any lower-priority swap_info_structs.
2358 * swap_avail_head needs to be priority ordered for folio_alloc_swap(),
2359 * which allocates swap pages from the highest available priority
2362 plist_add(&p->list, &swap_active_head);
2364 /* add to available list iff swap device is not full */
2366 add_to_avail_list(p);
2369 static void enable_swap_info(struct swap_info_struct *p, int prio,
2370 unsigned char *swap_map,
2371 struct swap_cluster_info *cluster_info)
2373 spin_lock(&swap_lock);
2374 spin_lock(&p->lock);
2375 setup_swap_info(p, prio, swap_map, cluster_info);
2376 spin_unlock(&p->lock);
2377 spin_unlock(&swap_lock);
2379 * Finished initializing swap device, now it's safe to reference it.
2381 percpu_ref_resurrect(&p->users);
2382 spin_lock(&swap_lock);
2383 spin_lock(&p->lock);
2384 _enable_swap_info(p);
2385 spin_unlock(&p->lock);
2386 spin_unlock(&swap_lock);
2389 static void reinsert_swap_info(struct swap_info_struct *p)
2391 spin_lock(&swap_lock);
2392 spin_lock(&p->lock);
2393 setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2394 _enable_swap_info(p);
2395 spin_unlock(&p->lock);
2396 spin_unlock(&swap_lock);
2399 bool has_usable_swap(void)
2403 spin_lock(&swap_lock);
2404 if (plist_head_empty(&swap_active_head))
2406 spin_unlock(&swap_lock);
2410 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2412 struct swap_info_struct *p = NULL;
2413 unsigned char *swap_map;
2414 struct swap_cluster_info *cluster_info;
2415 struct file *swap_file, *victim;
2416 struct address_space *mapping;
2417 struct inode *inode;
2418 struct filename *pathname;
2420 unsigned int old_block_size;
2422 if (!capable(CAP_SYS_ADMIN))
2425 BUG_ON(!current->mm);
2427 pathname = getname(specialfile);
2428 if (IS_ERR(pathname))
2429 return PTR_ERR(pathname);
2431 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2432 err = PTR_ERR(victim);
2436 mapping = victim->f_mapping;
2437 spin_lock(&swap_lock);
2438 plist_for_each_entry(p, &swap_active_head, list) {
2439 if (p->flags & SWP_WRITEOK) {
2440 if (p->swap_file->f_mapping == mapping) {
2448 spin_unlock(&swap_lock);
2451 if (!security_vm_enough_memory_mm(current->mm, p->pages))
2452 vm_unacct_memory(p->pages);
2455 spin_unlock(&swap_lock);
2458 spin_lock(&p->lock);
2459 del_from_avail_list(p);
2461 struct swap_info_struct *si = p;
2464 plist_for_each_entry_continue(si, &swap_active_head, list) {
2467 for_each_node(nid) {
2468 if (si->avail_lists[nid].prio != 1)
2469 si->avail_lists[nid].prio--;
2474 plist_del(&p->list, &swap_active_head);
2475 atomic_long_sub(p->pages, &nr_swap_pages);
2476 total_swap_pages -= p->pages;
2477 p->flags &= ~SWP_WRITEOK;
2478 spin_unlock(&p->lock);
2479 spin_unlock(&swap_lock);
2481 disable_swap_slots_cache_lock();
2483 set_current_oom_origin();
2484 err = try_to_unuse(p->type);
2485 clear_current_oom_origin();
2488 /* re-insert swap space back into swap_list */
2489 reinsert_swap_info(p);
2490 reenable_swap_slots_cache_unlock();
2494 reenable_swap_slots_cache_unlock();
2497 * Wait for swap operations protected by get/put_swap_device()
2500 * We need synchronize_rcu() here to protect the accessing to
2501 * the swap cache data structure.
2503 percpu_ref_kill(&p->users);
2505 wait_for_completion(&p->comp);
2507 flush_work(&p->discard_work);
2509 destroy_swap_extents(p);
2510 if (p->flags & SWP_CONTINUED)
2511 free_swap_count_continuations(p);
2513 if (!p->bdev || !bdev_nonrot(p->bdev))
2514 atomic_dec(&nr_rotate_swap);
2516 mutex_lock(&swapon_mutex);
2517 spin_lock(&swap_lock);
2518 spin_lock(&p->lock);
2521 /* wait for anyone still in scan_swap_map_slots */
2522 p->highest_bit = 0; /* cuts scans short */
2523 while (p->flags >= SWP_SCANNING) {
2524 spin_unlock(&p->lock);
2525 spin_unlock(&swap_lock);
2526 schedule_timeout_uninterruptible(1);
2527 spin_lock(&swap_lock);
2528 spin_lock(&p->lock);
2531 swap_file = p->swap_file;
2532 old_block_size = p->old_block_size;
2533 p->swap_file = NULL;
2535 swap_map = p->swap_map;
2537 cluster_info = p->cluster_info;
2538 p->cluster_info = NULL;
2539 spin_unlock(&p->lock);
2540 spin_unlock(&swap_lock);
2541 arch_swap_invalidate_area(p->type);
2542 zswap_swapoff(p->type);
2543 mutex_unlock(&swapon_mutex);
2544 free_percpu(p->percpu_cluster);
2545 p->percpu_cluster = NULL;
2546 free_percpu(p->cluster_next_cpu);
2547 p->cluster_next_cpu = NULL;
2549 kvfree(cluster_info);
2550 /* Destroy swap account information */
2551 swap_cgroup_swapoff(p->type);
2552 exit_swap_address_space(p->type);
2554 inode = mapping->host;
2556 set_blocksize(p->bdev, old_block_size);
2558 p->bdev_file = NULL;
2562 inode->i_flags &= ~S_SWAPFILE;
2563 inode_unlock(inode);
2564 filp_close(swap_file, NULL);
2567 * Clear the SWP_USED flag after all resources are freed so that swapon
2568 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2569 * not hold p->lock after we cleared its SWP_WRITEOK.
2571 spin_lock(&swap_lock);
2573 spin_unlock(&swap_lock);
2576 atomic_inc(&proc_poll_event);
2577 wake_up_interruptible(&proc_poll_wait);
2580 filp_close(victim, NULL);
2586 #ifdef CONFIG_PROC_FS
2587 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2589 struct seq_file *seq = file->private_data;
2591 poll_wait(file, &proc_poll_wait, wait);
2593 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2594 seq->poll_event = atomic_read(&proc_poll_event);
2595 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2598 return EPOLLIN | EPOLLRDNORM;
2602 static void *swap_start(struct seq_file *swap, loff_t *pos)
2604 struct swap_info_struct *si;
2608 mutex_lock(&swapon_mutex);
2611 return SEQ_START_TOKEN;
2613 for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2614 if (!(si->flags & SWP_USED) || !si->swap_map)
2623 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2625 struct swap_info_struct *si = v;
2628 if (v == SEQ_START_TOKEN)
2631 type = si->type + 1;
2634 for (; (si = swap_type_to_swap_info(type)); type++) {
2635 if (!(si->flags & SWP_USED) || !si->swap_map)
2643 static void swap_stop(struct seq_file *swap, void *v)
2645 mutex_unlock(&swapon_mutex);
2648 static int swap_show(struct seq_file *swap, void *v)
2650 struct swap_info_struct *si = v;
2653 unsigned long bytes, inuse;
2655 if (si == SEQ_START_TOKEN) {
2656 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2660 bytes = K(si->pages);
2661 inuse = K(READ_ONCE(si->inuse_pages));
2663 file = si->swap_file;
2664 len = seq_file_path(swap, file, " \t\n\\");
2665 seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n",
2666 len < 40 ? 40 - len : 1, " ",
2667 S_ISBLK(file_inode(file)->i_mode) ?
2668 "partition" : "file\t",
2669 bytes, bytes < 10000000 ? "\t" : "",
2670 inuse, inuse < 10000000 ? "\t" : "",
2675 static const struct seq_operations swaps_op = {
2676 .start = swap_start,
2682 static int swaps_open(struct inode *inode, struct file *file)
2684 struct seq_file *seq;
2687 ret = seq_open(file, &swaps_op);
2691 seq = file->private_data;
2692 seq->poll_event = atomic_read(&proc_poll_event);
2696 static const struct proc_ops swaps_proc_ops = {
2697 .proc_flags = PROC_ENTRY_PERMANENT,
2698 .proc_open = swaps_open,
2699 .proc_read = seq_read,
2700 .proc_lseek = seq_lseek,
2701 .proc_release = seq_release,
2702 .proc_poll = swaps_poll,
2705 static int __init procswaps_init(void)
2707 proc_create("swaps", 0, NULL, &swaps_proc_ops);
2710 __initcall(procswaps_init);
2711 #endif /* CONFIG_PROC_FS */
2713 #ifdef MAX_SWAPFILES_CHECK
2714 static int __init max_swapfiles_check(void)
2716 MAX_SWAPFILES_CHECK();
2719 late_initcall(max_swapfiles_check);
2722 static struct swap_info_struct *alloc_swap_info(void)
2724 struct swap_info_struct *p;
2725 struct swap_info_struct *defer = NULL;
2729 p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2731 return ERR_PTR(-ENOMEM);
2733 if (percpu_ref_init(&p->users, swap_users_ref_free,
2734 PERCPU_REF_INIT_DEAD, GFP_KERNEL)) {
2736 return ERR_PTR(-ENOMEM);
2739 spin_lock(&swap_lock);
2740 for (type = 0; type < nr_swapfiles; type++) {
2741 if (!(swap_info[type]->flags & SWP_USED))
2744 if (type >= MAX_SWAPFILES) {
2745 spin_unlock(&swap_lock);
2746 percpu_ref_exit(&p->users);
2748 return ERR_PTR(-EPERM);
2750 if (type >= nr_swapfiles) {
2753 * Publish the swap_info_struct after initializing it.
2754 * Note that kvzalloc() above zeroes all its fields.
2756 smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */
2760 p = swap_info[type];
2762 * Do not memset this entry: a racing procfs swap_next()
2763 * would be relying on p->type to remain valid.
2766 p->swap_extent_root = RB_ROOT;
2767 plist_node_init(&p->list, 0);
2769 plist_node_init(&p->avail_lists[i], 0);
2770 p->flags = SWP_USED;
2771 spin_unlock(&swap_lock);
2773 percpu_ref_exit(&defer->users);
2776 spin_lock_init(&p->lock);
2777 spin_lock_init(&p->cont_lock);
2778 init_completion(&p->comp);
2783 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2787 if (S_ISBLK(inode->i_mode)) {
2788 p->bdev_file = bdev_file_open_by_dev(inode->i_rdev,
2789 BLK_OPEN_READ | BLK_OPEN_WRITE, p, NULL);
2790 if (IS_ERR(p->bdev_file)) {
2791 error = PTR_ERR(p->bdev_file);
2792 p->bdev_file = NULL;
2795 p->bdev = file_bdev(p->bdev_file);
2796 p->old_block_size = block_size(p->bdev);
2797 error = set_blocksize(p->bdev, PAGE_SIZE);
2801 * Zoned block devices contain zones that have a sequential
2802 * write only restriction. Hence zoned block devices are not
2803 * suitable for swapping. Disallow them here.
2805 if (bdev_is_zoned(p->bdev))
2807 p->flags |= SWP_BLKDEV;
2808 } else if (S_ISREG(inode->i_mode)) {
2809 p->bdev = inode->i_sb->s_bdev;
2817 * Find out how many pages are allowed for a single swap device. There
2818 * are two limiting factors:
2819 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2820 * 2) the number of bits in the swap pte, as defined by the different
2823 * In order to find the largest possible bit mask, a swap entry with
2824 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2825 * decoded to a swp_entry_t again, and finally the swap offset is
2828 * This will mask all the bits from the initial ~0UL mask that can't
2829 * be encoded in either the swp_entry_t or the architecture definition
2832 unsigned long generic_max_swapfile_size(void)
2834 return swp_offset(pte_to_swp_entry(
2835 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2838 /* Can be overridden by an architecture for additional checks. */
2839 __weak unsigned long arch_max_swapfile_size(void)
2841 return generic_max_swapfile_size();
2844 static unsigned long read_swap_header(struct swap_info_struct *p,
2845 union swap_header *swap_header,
2846 struct inode *inode)
2849 unsigned long maxpages;
2850 unsigned long swapfilepages;
2851 unsigned long last_page;
2853 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2854 pr_err("Unable to find swap-space signature\n");
2858 /* swap partition endianness hack... */
2859 if (swab32(swap_header->info.version) == 1) {
2860 swab32s(&swap_header->info.version);
2861 swab32s(&swap_header->info.last_page);
2862 swab32s(&swap_header->info.nr_badpages);
2863 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2865 for (i = 0; i < swap_header->info.nr_badpages; i++)
2866 swab32s(&swap_header->info.badpages[i]);
2868 /* Check the swap header's sub-version */
2869 if (swap_header->info.version != 1) {
2870 pr_warn("Unable to handle swap header version %d\n",
2871 swap_header->info.version);
2876 p->cluster_next = 1;
2879 maxpages = swapfile_maximum_size;
2880 last_page = swap_header->info.last_page;
2882 pr_warn("Empty swap-file\n");
2885 if (last_page > maxpages) {
2886 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2887 K(maxpages), K(last_page));
2889 if (maxpages > last_page) {
2890 maxpages = last_page + 1;
2891 /* p->max is an unsigned int: don't overflow it */
2892 if ((unsigned int)maxpages == 0)
2893 maxpages = UINT_MAX;
2895 p->highest_bit = maxpages - 1;
2899 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2900 if (swapfilepages && maxpages > swapfilepages) {
2901 pr_warn("Swap area shorter than signature indicates\n");
2904 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2906 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2912 #define SWAP_CLUSTER_INFO_COLS \
2913 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2914 #define SWAP_CLUSTER_SPACE_COLS \
2915 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2916 #define SWAP_CLUSTER_COLS \
2917 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2919 static int setup_swap_map_and_extents(struct swap_info_struct *p,
2920 union swap_header *swap_header,
2921 unsigned char *swap_map,
2922 struct swap_cluster_info *cluster_info,
2923 unsigned long maxpages,
2927 unsigned int nr_good_pages;
2929 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2930 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2931 unsigned long i, idx;
2933 nr_good_pages = maxpages - 1; /* omit header page */
2935 cluster_list_init(&p->free_clusters);
2936 cluster_list_init(&p->discard_clusters);
2938 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2939 unsigned int page_nr = swap_header->info.badpages[i];
2940 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2942 if (page_nr < maxpages) {
2943 swap_map[page_nr] = SWAP_MAP_BAD;
2946 * Haven't marked the cluster free yet, no list
2947 * operation involved
2949 inc_cluster_info_page(p, cluster_info, page_nr);
2953 /* Haven't marked the cluster free yet, no list operation involved */
2954 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2955 inc_cluster_info_page(p, cluster_info, i);
2957 if (nr_good_pages) {
2958 swap_map[0] = SWAP_MAP_BAD;
2960 * Not mark the cluster free yet, no list
2961 * operation involved
2963 inc_cluster_info_page(p, cluster_info, 0);
2965 p->pages = nr_good_pages;
2966 nr_extents = setup_swap_extents(p, span);
2969 nr_good_pages = p->pages;
2971 if (!nr_good_pages) {
2972 pr_warn("Empty swap-file\n");
2981 * Reduce false cache line sharing between cluster_info and
2982 * sharing same address space.
2984 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2985 j = (k + col) % SWAP_CLUSTER_COLS;
2986 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2987 idx = i * SWAP_CLUSTER_COLS + j;
2988 if (idx >= nr_clusters)
2990 if (cluster_count(&cluster_info[idx]))
2992 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2993 cluster_list_add_tail(&p->free_clusters, cluster_info,
3000 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
3002 struct swap_info_struct *p;
3003 struct filename *name;
3004 struct file *swap_file = NULL;
3005 struct address_space *mapping;
3006 struct dentry *dentry;
3009 union swap_header *swap_header;
3012 unsigned long maxpages;
3013 unsigned char *swap_map = NULL;
3014 struct swap_cluster_info *cluster_info = NULL;
3015 struct page *page = NULL;
3016 struct inode *inode = NULL;
3017 bool inced_nr_rotate_swap = false;
3019 if (swap_flags & ~SWAP_FLAGS_VALID)
3022 if (!capable(CAP_SYS_ADMIN))
3025 if (!swap_avail_heads)
3028 p = alloc_swap_info();
3032 INIT_WORK(&p->discard_work, swap_discard_work);
3034 name = getname(specialfile);
3036 error = PTR_ERR(name);
3040 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3041 if (IS_ERR(swap_file)) {
3042 error = PTR_ERR(swap_file);
3047 p->swap_file = swap_file;
3048 mapping = swap_file->f_mapping;
3049 dentry = swap_file->f_path.dentry;
3050 inode = mapping->host;
3052 error = claim_swapfile(p, inode);
3053 if (unlikely(error))
3057 if (d_unlinked(dentry) || cant_mount(dentry)) {
3059 goto bad_swap_unlock_inode;
3061 if (IS_SWAPFILE(inode)) {
3063 goto bad_swap_unlock_inode;
3067 * Read the swap header.
3069 if (!mapping->a_ops->read_folio) {
3071 goto bad_swap_unlock_inode;
3073 page = read_mapping_page(mapping, 0, swap_file);
3075 error = PTR_ERR(page);
3076 goto bad_swap_unlock_inode;
3078 swap_header = kmap(page);
3080 maxpages = read_swap_header(p, swap_header, inode);
3081 if (unlikely(!maxpages)) {
3083 goto bad_swap_unlock_inode;
3086 /* OK, set up the swap map and apply the bad block list */
3087 swap_map = vzalloc(maxpages);
3090 goto bad_swap_unlock_inode;
3093 if (p->bdev && bdev_stable_writes(p->bdev))
3094 p->flags |= SWP_STABLE_WRITES;
3096 if (p->bdev && bdev_synchronous(p->bdev))
3097 p->flags |= SWP_SYNCHRONOUS_IO;
3099 if (p->bdev && bdev_nonrot(p->bdev)) {
3101 unsigned long ci, nr_cluster;
3103 p->flags |= SWP_SOLIDSTATE;
3104 p->cluster_next_cpu = alloc_percpu(unsigned int);
3105 if (!p->cluster_next_cpu) {
3107 goto bad_swap_unlock_inode;
3110 * select a random position to start with to help wear leveling
3113 for_each_possible_cpu(cpu) {
3114 per_cpu(*p->cluster_next_cpu, cpu) =
3115 get_random_u32_inclusive(1, p->highest_bit);
3117 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3119 cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3121 if (!cluster_info) {
3123 goto bad_swap_unlock_inode;
3126 for (ci = 0; ci < nr_cluster; ci++)
3127 spin_lock_init(&((cluster_info + ci)->lock));
3129 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3130 if (!p->percpu_cluster) {
3132 goto bad_swap_unlock_inode;
3134 for_each_possible_cpu(cpu) {
3135 struct percpu_cluster *cluster;
3136 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3137 cluster_set_null(&cluster->index);
3140 atomic_inc(&nr_rotate_swap);
3141 inced_nr_rotate_swap = true;
3144 error = swap_cgroup_swapon(p->type, maxpages);
3146 goto bad_swap_unlock_inode;
3148 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3149 cluster_info, maxpages, &span);
3150 if (unlikely(nr_extents < 0)) {
3152 goto bad_swap_unlock_inode;
3155 if ((swap_flags & SWAP_FLAG_DISCARD) &&
3156 p->bdev && bdev_max_discard_sectors(p->bdev)) {
3158 * When discard is enabled for swap with no particular
3159 * policy flagged, we set all swap discard flags here in
3160 * order to sustain backward compatibility with older
3161 * swapon(8) releases.
3163 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3167 * By flagging sys_swapon, a sysadmin can tell us to
3168 * either do single-time area discards only, or to just
3169 * perform discards for released swap page-clusters.
3170 * Now it's time to adjust the p->flags accordingly.
3172 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3173 p->flags &= ~SWP_PAGE_DISCARD;
3174 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3175 p->flags &= ~SWP_AREA_DISCARD;
3177 /* issue a swapon-time discard if it's still required */
3178 if (p->flags & SWP_AREA_DISCARD) {
3179 int err = discard_swap(p);
3181 pr_err("swapon: discard_swap(%p): %d\n",
3186 error = init_swap_address_space(p->type, maxpages);
3188 goto bad_swap_unlock_inode;
3190 error = zswap_swapon(p->type, maxpages);
3192 goto free_swap_address_space;
3195 * Flush any pending IO and dirty mappings before we start using this
3198 inode->i_flags |= S_SWAPFILE;
3199 error = inode_drain_writes(inode);
3201 inode->i_flags &= ~S_SWAPFILE;
3202 goto free_swap_zswap;
3205 mutex_lock(&swapon_mutex);
3207 if (swap_flags & SWAP_FLAG_PREFER)
3209 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3210 enable_swap_info(p, prio, swap_map, cluster_info);
3212 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s\n",
3213 K(p->pages), name->name, p->prio, nr_extents,
3214 K((unsigned long long)span),
3215 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3216 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3217 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3218 (p->flags & SWP_PAGE_DISCARD) ? "c" : "");
3220 mutex_unlock(&swapon_mutex);
3221 atomic_inc(&proc_poll_event);
3222 wake_up_interruptible(&proc_poll_wait);
3227 zswap_swapoff(p->type);
3228 free_swap_address_space:
3229 exit_swap_address_space(p->type);
3230 bad_swap_unlock_inode:
3231 inode_unlock(inode);
3233 free_percpu(p->percpu_cluster);
3234 p->percpu_cluster = NULL;
3235 free_percpu(p->cluster_next_cpu);
3236 p->cluster_next_cpu = NULL;
3238 set_blocksize(p->bdev, p->old_block_size);
3240 p->bdev_file = NULL;
3243 destroy_swap_extents(p);
3244 swap_cgroup_swapoff(p->type);
3245 spin_lock(&swap_lock);
3246 p->swap_file = NULL;
3248 spin_unlock(&swap_lock);
3250 kvfree(cluster_info);
3251 if (inced_nr_rotate_swap)
3252 atomic_dec(&nr_rotate_swap);
3254 filp_close(swap_file, NULL);
3256 if (page && !IS_ERR(page)) {
3263 inode_unlock(inode);
3265 enable_swap_slots_cache();
3269 void si_swapinfo(struct sysinfo *val)
3272 unsigned long nr_to_be_unused = 0;
3274 spin_lock(&swap_lock);
3275 for (type = 0; type < nr_swapfiles; type++) {
3276 struct swap_info_struct *si = swap_info[type];
3278 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3279 nr_to_be_unused += READ_ONCE(si->inuse_pages);
3281 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3282 val->totalswap = total_swap_pages + nr_to_be_unused;
3283 spin_unlock(&swap_lock);
3287 * Verify that a swap entry is valid and increment its swap map count.
3289 * Returns error code in following case.
3291 * - swp_entry is invalid -> EINVAL
3292 * - swp_entry is migration entry -> EINVAL
3293 * - swap-cache reference is requested but there is already one. -> EEXIST
3294 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3295 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3297 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3299 struct swap_info_struct *p;
3300 struct swap_cluster_info *ci;
3301 unsigned long offset;
3302 unsigned char count;
3303 unsigned char has_cache;
3306 p = swp_swap_info(entry);
3308 offset = swp_offset(entry);
3309 ci = lock_cluster_or_swap_info(p, offset);
3311 count = p->swap_map[offset];
3314 * swapin_readahead() doesn't check if a swap entry is valid, so the
3315 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3317 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3322 has_cache = count & SWAP_HAS_CACHE;
3323 count &= ~SWAP_HAS_CACHE;
3326 if (usage == SWAP_HAS_CACHE) {
3328 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3329 if (!has_cache && count)
3330 has_cache = SWAP_HAS_CACHE;
3331 else if (has_cache) /* someone else added cache */
3333 else /* no users remaining */
3336 } else if (count || has_cache) {
3338 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3340 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3342 else if (swap_count_continued(p, offset, count))
3343 count = COUNT_CONTINUED;
3347 err = -ENOENT; /* unused swap entry */
3350 WRITE_ONCE(p->swap_map[offset], count | has_cache);
3353 unlock_cluster_or_swap_info(p, ci);
3358 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3359 * (in which case its reference count is never incremented).
3361 void swap_shmem_alloc(swp_entry_t entry)
3363 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3367 * Increase reference count of swap entry by 1.
3368 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3369 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3370 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3371 * might occur if a page table entry has got corrupted.
3373 int swap_duplicate(swp_entry_t entry)
3377 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3378 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3383 * @entry: swap entry for which we allocate swap cache.
3385 * Called when allocating swap cache for existing swap entry,
3386 * This can return error codes. Returns 0 at success.
3387 * -EEXIST means there is a swap cache.
3388 * Note: return code is different from swap_duplicate().
3390 int swapcache_prepare(swp_entry_t entry)
3392 return __swap_duplicate(entry, SWAP_HAS_CACHE);
3395 void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry)
3397 struct swap_cluster_info *ci;
3398 unsigned long offset = swp_offset(entry);
3399 unsigned char usage;
3401 ci = lock_cluster_or_swap_info(si, offset);
3402 usage = __swap_entry_free_locked(si, offset, SWAP_HAS_CACHE);
3403 unlock_cluster_or_swap_info(si, ci);
3405 free_swap_slot(entry);
3408 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3410 return swap_type_to_swap_info(swp_type(entry));
3414 * out-of-line methods to avoid include hell.
3416 struct address_space *swapcache_mapping(struct folio *folio)
3418 return swp_swap_info(folio->swap)->swap_file->f_mapping;
3420 EXPORT_SYMBOL_GPL(swapcache_mapping);
3422 pgoff_t __page_file_index(struct page *page)
3424 swp_entry_t swap = page_swap_entry(page);
3425 return swp_offset(swap);
3427 EXPORT_SYMBOL_GPL(__page_file_index);
3430 * add_swap_count_continuation - called when a swap count is duplicated
3431 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3432 * page of the original vmalloc'ed swap_map, to hold the continuation count
3433 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3434 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3436 * These continuation pages are seldom referenced: the common paths all work
3437 * on the original swap_map, only referring to a continuation page when the
3438 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3440 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3441 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3442 * can be called after dropping locks.
3444 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3446 struct swap_info_struct *si;
3447 struct swap_cluster_info *ci;
3450 struct page *list_page;
3452 unsigned char count;
3456 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3457 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3459 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3461 si = get_swap_device(entry);
3464 * An acceptable race has occurred since the failing
3465 * __swap_duplicate(): the swap device may be swapoff
3469 spin_lock(&si->lock);
3471 offset = swp_offset(entry);
3473 ci = lock_cluster(si, offset);
3475 count = swap_count(si->swap_map[offset]);
3477 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3479 * The higher the swap count, the more likely it is that tasks
3480 * will race to add swap count continuation: we need to avoid
3481 * over-provisioning.
3491 head = vmalloc_to_page(si->swap_map + offset);
3492 offset &= ~PAGE_MASK;
3494 spin_lock(&si->cont_lock);
3496 * Page allocation does not initialize the page's lru field,
3497 * but it does always reset its private field.
3499 if (!page_private(head)) {
3500 BUG_ON(count & COUNT_CONTINUED);
3501 INIT_LIST_HEAD(&head->lru);
3502 set_page_private(head, SWP_CONTINUED);
3503 si->flags |= SWP_CONTINUED;
3506 list_for_each_entry(list_page, &head->lru, lru) {
3510 * If the previous map said no continuation, but we've found
3511 * a continuation page, free our allocation and use this one.
3513 if (!(count & COUNT_CONTINUED))
3514 goto out_unlock_cont;
3516 map = kmap_local_page(list_page) + offset;
3521 * If this continuation count now has some space in it,
3522 * free our allocation and use this one.
3524 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3525 goto out_unlock_cont;
3528 list_add_tail(&page->lru, &head->lru);
3529 page = NULL; /* now it's attached, don't free it */
3531 spin_unlock(&si->cont_lock);
3534 spin_unlock(&si->lock);
3535 put_swap_device(si);
3543 * swap_count_continued - when the original swap_map count is incremented
3544 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3545 * into, carry if so, or else fail until a new continuation page is allocated;
3546 * when the original swap_map count is decremented from 0 with continuation,
3547 * borrow from the continuation and report whether it still holds more.
3548 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3551 static bool swap_count_continued(struct swap_info_struct *si,
3552 pgoff_t offset, unsigned char count)
3559 head = vmalloc_to_page(si->swap_map + offset);
3560 if (page_private(head) != SWP_CONTINUED) {
3561 BUG_ON(count & COUNT_CONTINUED);
3562 return false; /* need to add count continuation */
3565 spin_lock(&si->cont_lock);
3566 offset &= ~PAGE_MASK;
3567 page = list_next_entry(head, lru);
3568 map = kmap_local_page(page) + offset;
3570 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3571 goto init_map; /* jump over SWAP_CONT_MAX checks */
3573 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3575 * Think of how you add 1 to 999
3577 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3579 page = list_next_entry(page, lru);
3580 BUG_ON(page == head);
3581 map = kmap_local_page(page) + offset;
3583 if (*map == SWAP_CONT_MAX) {
3585 page = list_next_entry(page, lru);
3587 ret = false; /* add count continuation */
3590 map = kmap_local_page(page) + offset;
3591 init_map: *map = 0; /* we didn't zero the page */
3595 while ((page = list_prev_entry(page, lru)) != head) {
3596 map = kmap_local_page(page) + offset;
3597 *map = COUNT_CONTINUED;
3600 ret = true; /* incremented */
3602 } else { /* decrementing */
3604 * Think of how you subtract 1 from 1000
3606 BUG_ON(count != COUNT_CONTINUED);
3607 while (*map == COUNT_CONTINUED) {
3609 page = list_next_entry(page, lru);
3610 BUG_ON(page == head);
3611 map = kmap_local_page(page) + offset;
3618 while ((page = list_prev_entry(page, lru)) != head) {
3619 map = kmap_local_page(page) + offset;
3620 *map = SWAP_CONT_MAX | count;
3621 count = COUNT_CONTINUED;
3624 ret = count == COUNT_CONTINUED;
3627 spin_unlock(&si->cont_lock);
3632 * free_swap_count_continuations - swapoff free all the continuation pages
3633 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3635 static void free_swap_count_continuations(struct swap_info_struct *si)
3639 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3641 head = vmalloc_to_page(si->swap_map + offset);
3642 if (page_private(head)) {
3643 struct page *page, *next;
3645 list_for_each_entry_safe(page, next, &head->lru, lru) {
3646 list_del(&page->lru);
3653 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3654 void __folio_throttle_swaprate(struct folio *folio, gfp_t gfp)
3656 struct swap_info_struct *si, *next;
3657 int nid = folio_nid(folio);
3659 if (!(gfp & __GFP_IO))
3662 if (!blk_cgroup_congested())
3666 * We've already scheduled a throttle, avoid taking the global swap
3669 if (current->throttle_disk)
3672 spin_lock(&swap_avail_lock);
3673 plist_for_each_entry_safe(si, next, &swap_avail_heads[nid],
3676 blkcg_schedule_throttle(si->bdev->bd_disk, true);
3680 spin_unlock(&swap_avail_lock);
3684 static int __init swapfile_init(void)
3688 swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3690 if (!swap_avail_heads) {
3691 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3696 plist_head_init(&swap_avail_heads[nid]);
3698 swapfile_maximum_size = arch_max_swapfile_size();
3700 #ifdef CONFIG_MIGRATION
3701 if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS))
3702 swap_migration_ad_supported = true;
3703 #endif /* CONFIG_MIGRATION */
3707 subsys_initcall(swapfile_init);