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/frontswap.h>
39 #include <linux/swapfile.h>
40 #include <linux/export.h>
41 #include <linux/swap_slots.h>
42 #include <linux/sort.h>
43 #include <linux/completion.h>
44 #include <linux/suspend.h>
46 #include <asm/tlbflush.h>
47 #include <linux/swapops.h>
48 #include <linux/swap_cgroup.h>
51 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
53 static void free_swap_count_continuations(struct swap_info_struct *);
55 static DEFINE_SPINLOCK(swap_lock);
56 static unsigned int nr_swapfiles;
57 atomic_long_t nr_swap_pages;
59 * Some modules use swappable objects and may try to swap them out under
60 * memory pressure (via the shrinker). Before doing so, they may wish to
61 * check to see if any swap space is available.
63 EXPORT_SYMBOL_GPL(nr_swap_pages);
64 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
65 long total_swap_pages;
66 static int least_priority = -1;
67 unsigned long swapfile_maximum_size;
68 #ifdef CONFIG_MIGRATION
69 bool swap_migration_ad_supported;
70 #endif /* CONFIG_MIGRATION */
72 static const char Bad_file[] = "Bad swap file entry ";
73 static const char Unused_file[] = "Unused swap file entry ";
74 static const char Bad_offset[] = "Bad swap offset entry ";
75 static const char Unused_offset[] = "Unused swap offset entry ";
78 * all active swap_info_structs
79 * protected with swap_lock, and ordered by priority.
81 static PLIST_HEAD(swap_active_head);
84 * all available (active, not full) swap_info_structs
85 * protected with swap_avail_lock, ordered by priority.
86 * This is used by folio_alloc_swap() instead of swap_active_head
87 * because swap_active_head includes all swap_info_structs,
88 * but folio_alloc_swap() doesn't need to look at full ones.
89 * This uses its own lock instead of swap_lock because when a
90 * swap_info_struct changes between not-full/full, it needs to
91 * add/remove itself to/from this list, but the swap_info_struct->lock
92 * is held and the locking order requires swap_lock to be taken
93 * before any swap_info_struct->lock.
95 static struct plist_head *swap_avail_heads;
96 static DEFINE_SPINLOCK(swap_avail_lock);
98 struct swap_info_struct *swap_info[MAX_SWAPFILES];
100 static DEFINE_MUTEX(swapon_mutex);
102 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
103 /* Activity counter to indicate that a swapon or swapoff has occurred */
104 static atomic_t proc_poll_event = ATOMIC_INIT(0);
106 atomic_t nr_rotate_swap = ATOMIC_INIT(0);
108 static struct swap_info_struct *swap_type_to_swap_info(int type)
110 if (type >= MAX_SWAPFILES)
113 return READ_ONCE(swap_info[type]); /* rcu_dereference() */
116 static inline unsigned char swap_count(unsigned char ent)
118 return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */
121 /* Reclaim the swap entry anyway if possible */
122 #define TTRS_ANYWAY 0x1
124 * Reclaim the swap entry if there are no more mappings of the
127 #define TTRS_UNMAPPED 0x2
128 /* Reclaim the swap entry if swap is getting full*/
129 #define TTRS_FULL 0x4
131 /* returns 1 if swap entry is freed */
132 static int __try_to_reclaim_swap(struct swap_info_struct *si,
133 unsigned long offset, unsigned long flags)
135 swp_entry_t entry = swp_entry(si->type, offset);
139 folio = filemap_get_folio(swap_address_space(entry), offset);
143 * When this function is called from scan_swap_map_slots() and it's
144 * called by vmscan.c at reclaiming folios. So we hold a folio lock
145 * here. We have to use trylock for avoiding deadlock. This is a special
146 * case and you should use folio_free_swap() with explicit folio_lock()
147 * in usual operations.
149 if (folio_trylock(folio)) {
150 if ((flags & TTRS_ANYWAY) ||
151 ((flags & TTRS_UNMAPPED) && !folio_mapped(folio)) ||
152 ((flags & TTRS_FULL) && mem_cgroup_swap_full(folio)))
153 ret = folio_free_swap(folio);
160 static inline struct swap_extent *first_se(struct swap_info_struct *sis)
162 struct rb_node *rb = rb_first(&sis->swap_extent_root);
163 return rb_entry(rb, struct swap_extent, rb_node);
166 static inline struct swap_extent *next_se(struct swap_extent *se)
168 struct rb_node *rb = rb_next(&se->rb_node);
169 return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
173 * swapon tell device that all the old swap contents can be discarded,
174 * to allow the swap device to optimize its wear-levelling.
176 static int discard_swap(struct swap_info_struct *si)
178 struct swap_extent *se;
179 sector_t start_block;
183 /* Do not discard the swap header page! */
185 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
186 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
188 err = blkdev_issue_discard(si->bdev, start_block,
189 nr_blocks, GFP_KERNEL);
195 for (se = next_se(se); se; se = next_se(se)) {
196 start_block = se->start_block << (PAGE_SHIFT - 9);
197 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
199 err = blkdev_issue_discard(si->bdev, start_block,
200 nr_blocks, GFP_KERNEL);
206 return err; /* That will often be -EOPNOTSUPP */
209 static struct swap_extent *
210 offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
212 struct swap_extent *se;
215 rb = sis->swap_extent_root.rb_node;
217 se = rb_entry(rb, struct swap_extent, rb_node);
218 if (offset < se->start_page)
220 else if (offset >= se->start_page + se->nr_pages)
225 /* It *must* be present */
229 sector_t swap_page_sector(struct page *page)
231 struct swap_info_struct *sis = page_swap_info(page);
232 struct swap_extent *se;
236 offset = __page_file_index(page);
237 se = offset_to_swap_extent(sis, offset);
238 sector = se->start_block + (offset - se->start_page);
239 return sector << (PAGE_SHIFT - 9);
243 * swap allocation tell device that a cluster of swap can now be discarded,
244 * to allow the swap device to optimize its wear-levelling.
246 static void discard_swap_cluster(struct swap_info_struct *si,
247 pgoff_t start_page, pgoff_t nr_pages)
249 struct swap_extent *se = offset_to_swap_extent(si, start_page);
252 pgoff_t offset = start_page - se->start_page;
253 sector_t start_block = se->start_block + offset;
254 sector_t nr_blocks = se->nr_pages - offset;
256 if (nr_blocks > nr_pages)
257 nr_blocks = nr_pages;
258 start_page += nr_blocks;
259 nr_pages -= nr_blocks;
261 start_block <<= PAGE_SHIFT - 9;
262 nr_blocks <<= PAGE_SHIFT - 9;
263 if (blkdev_issue_discard(si->bdev, start_block,
264 nr_blocks, GFP_NOIO))
271 #ifdef CONFIG_THP_SWAP
272 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
274 #define swap_entry_size(size) (size)
276 #define SWAPFILE_CLUSTER 256
279 * Define swap_entry_size() as constant to let compiler to optimize
280 * out some code if !CONFIG_THP_SWAP
282 #define swap_entry_size(size) 1
284 #define LATENCY_LIMIT 256
286 static inline void cluster_set_flag(struct swap_cluster_info *info,
292 static inline unsigned int cluster_count(struct swap_cluster_info *info)
297 static inline void cluster_set_count(struct swap_cluster_info *info,
303 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
304 unsigned int c, unsigned int f)
310 static inline unsigned int cluster_next(struct swap_cluster_info *info)
315 static inline void cluster_set_next(struct swap_cluster_info *info,
321 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
322 unsigned int n, unsigned int f)
328 static inline bool cluster_is_free(struct swap_cluster_info *info)
330 return info->flags & CLUSTER_FLAG_FREE;
333 static inline bool cluster_is_null(struct swap_cluster_info *info)
335 return info->flags & CLUSTER_FLAG_NEXT_NULL;
338 static inline void cluster_set_null(struct swap_cluster_info *info)
340 info->flags = CLUSTER_FLAG_NEXT_NULL;
344 static inline bool cluster_is_huge(struct swap_cluster_info *info)
346 if (IS_ENABLED(CONFIG_THP_SWAP))
347 return info->flags & CLUSTER_FLAG_HUGE;
351 static inline void cluster_clear_huge(struct swap_cluster_info *info)
353 info->flags &= ~CLUSTER_FLAG_HUGE;
356 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
357 unsigned long offset)
359 struct swap_cluster_info *ci;
361 ci = si->cluster_info;
363 ci += offset / SWAPFILE_CLUSTER;
364 spin_lock(&ci->lock);
369 static inline void unlock_cluster(struct swap_cluster_info *ci)
372 spin_unlock(&ci->lock);
376 * Determine the locking method in use for this device. Return
377 * swap_cluster_info if SSD-style cluster-based locking is in place.
379 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
380 struct swap_info_struct *si, unsigned long offset)
382 struct swap_cluster_info *ci;
384 /* Try to use fine-grained SSD-style locking if available: */
385 ci = lock_cluster(si, offset);
386 /* Otherwise, fall back to traditional, coarse locking: */
388 spin_lock(&si->lock);
393 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
394 struct swap_cluster_info *ci)
399 spin_unlock(&si->lock);
402 static inline bool cluster_list_empty(struct swap_cluster_list *list)
404 return cluster_is_null(&list->head);
407 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
409 return cluster_next(&list->head);
412 static void cluster_list_init(struct swap_cluster_list *list)
414 cluster_set_null(&list->head);
415 cluster_set_null(&list->tail);
418 static void cluster_list_add_tail(struct swap_cluster_list *list,
419 struct swap_cluster_info *ci,
422 if (cluster_list_empty(list)) {
423 cluster_set_next_flag(&list->head, idx, 0);
424 cluster_set_next_flag(&list->tail, idx, 0);
426 struct swap_cluster_info *ci_tail;
427 unsigned int tail = cluster_next(&list->tail);
430 * Nested cluster lock, but both cluster locks are
431 * only acquired when we held swap_info_struct->lock
434 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
435 cluster_set_next(ci_tail, idx);
436 spin_unlock(&ci_tail->lock);
437 cluster_set_next_flag(&list->tail, idx, 0);
441 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
442 struct swap_cluster_info *ci)
446 idx = cluster_next(&list->head);
447 if (cluster_next(&list->tail) == idx) {
448 cluster_set_null(&list->head);
449 cluster_set_null(&list->tail);
451 cluster_set_next_flag(&list->head,
452 cluster_next(&ci[idx]), 0);
457 /* Add a cluster to discard list and schedule it to do discard */
458 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
462 * If scan_swap_map_slots() can't find a free cluster, it will check
463 * si->swap_map directly. To make sure the discarding cluster isn't
464 * taken by scan_swap_map_slots(), mark the swap entries bad (occupied).
465 * It will be cleared after discard
467 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
468 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
470 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
472 schedule_work(&si->discard_work);
475 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
477 struct swap_cluster_info *ci = si->cluster_info;
479 cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
480 cluster_list_add_tail(&si->free_clusters, ci, idx);
484 * Doing discard actually. After a cluster discard is finished, the cluster
485 * will be added to free cluster list. caller should hold si->lock.
487 static void swap_do_scheduled_discard(struct swap_info_struct *si)
489 struct swap_cluster_info *info, *ci;
492 info = si->cluster_info;
494 while (!cluster_list_empty(&si->discard_clusters)) {
495 idx = cluster_list_del_first(&si->discard_clusters, info);
496 spin_unlock(&si->lock);
498 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
501 spin_lock(&si->lock);
502 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
503 __free_cluster(si, idx);
504 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
505 0, SWAPFILE_CLUSTER);
510 static void swap_discard_work(struct work_struct *work)
512 struct swap_info_struct *si;
514 si = container_of(work, struct swap_info_struct, discard_work);
516 spin_lock(&si->lock);
517 swap_do_scheduled_discard(si);
518 spin_unlock(&si->lock);
521 static void swap_users_ref_free(struct percpu_ref *ref)
523 struct swap_info_struct *si;
525 si = container_of(ref, struct swap_info_struct, users);
529 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
531 struct swap_cluster_info *ci = si->cluster_info;
533 VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
534 cluster_list_del_first(&si->free_clusters, ci);
535 cluster_set_count_flag(ci + idx, 0, 0);
538 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
540 struct swap_cluster_info *ci = si->cluster_info + idx;
542 VM_BUG_ON(cluster_count(ci) != 0);
544 * If the swap is discardable, prepare discard the cluster
545 * instead of free it immediately. The cluster will be freed
548 if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
549 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
550 swap_cluster_schedule_discard(si, idx);
554 __free_cluster(si, idx);
558 * The cluster corresponding to page_nr will be used. The cluster will be
559 * removed from free cluster list and its usage counter will be increased.
561 static void inc_cluster_info_page(struct swap_info_struct *p,
562 struct swap_cluster_info *cluster_info, unsigned long page_nr)
564 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
568 if (cluster_is_free(&cluster_info[idx]))
569 alloc_cluster(p, idx);
571 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
572 cluster_set_count(&cluster_info[idx],
573 cluster_count(&cluster_info[idx]) + 1);
577 * The cluster corresponding to page_nr decreases one usage. If the usage
578 * counter becomes 0, which means no page in the cluster is in using, we can
579 * optionally discard the cluster and add it to free cluster list.
581 static void dec_cluster_info_page(struct swap_info_struct *p,
582 struct swap_cluster_info *cluster_info, unsigned long page_nr)
584 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
589 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
590 cluster_set_count(&cluster_info[idx],
591 cluster_count(&cluster_info[idx]) - 1);
593 if (cluster_count(&cluster_info[idx]) == 0)
594 free_cluster(p, idx);
598 * It's possible scan_swap_map_slots() uses a free cluster in the middle of free
599 * cluster list. Avoiding such abuse to avoid list corruption.
602 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
603 unsigned long offset)
605 struct percpu_cluster *percpu_cluster;
608 offset /= SWAPFILE_CLUSTER;
609 conflict = !cluster_list_empty(&si->free_clusters) &&
610 offset != cluster_list_first(&si->free_clusters) &&
611 cluster_is_free(&si->cluster_info[offset]);
616 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
617 cluster_set_null(&percpu_cluster->index);
622 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
623 * might involve allocating a new cluster for current CPU too.
625 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
626 unsigned long *offset, unsigned long *scan_base)
628 struct percpu_cluster *cluster;
629 struct swap_cluster_info *ci;
630 unsigned long tmp, max;
633 cluster = this_cpu_ptr(si->percpu_cluster);
634 if (cluster_is_null(&cluster->index)) {
635 if (!cluster_list_empty(&si->free_clusters)) {
636 cluster->index = si->free_clusters.head;
637 cluster->next = cluster_next(&cluster->index) *
639 } else if (!cluster_list_empty(&si->discard_clusters)) {
641 * we don't have free cluster but have some clusters in
642 * discarding, do discard now and reclaim them, then
643 * reread cluster_next_cpu since we dropped si->lock
645 swap_do_scheduled_discard(si);
646 *scan_base = this_cpu_read(*si->cluster_next_cpu);
647 *offset = *scan_base;
654 * Other CPUs can use our cluster if they can't find a free cluster,
655 * check if there is still free entry in the cluster
658 max = min_t(unsigned long, si->max,
659 (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
661 ci = lock_cluster(si, tmp);
663 if (!si->swap_map[tmp])
670 cluster_set_null(&cluster->index);
673 cluster->next = tmp + 1;
679 static void __del_from_avail_list(struct swap_info_struct *p)
683 assert_spin_locked(&p->lock);
685 plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
688 static void del_from_avail_list(struct swap_info_struct *p)
690 spin_lock(&swap_avail_lock);
691 __del_from_avail_list(p);
692 spin_unlock(&swap_avail_lock);
695 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
696 unsigned int nr_entries)
698 unsigned int end = offset + nr_entries - 1;
700 if (offset == si->lowest_bit)
701 si->lowest_bit += nr_entries;
702 if (end == si->highest_bit)
703 WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries);
704 WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries);
705 if (si->inuse_pages == si->pages) {
706 si->lowest_bit = si->max;
708 del_from_avail_list(si);
712 static void add_to_avail_list(struct swap_info_struct *p)
716 spin_lock(&swap_avail_lock);
718 plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
719 spin_unlock(&swap_avail_lock);
722 static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
723 unsigned int nr_entries)
725 unsigned long begin = offset;
726 unsigned long end = offset + nr_entries - 1;
727 void (*swap_slot_free_notify)(struct block_device *, unsigned long);
729 if (offset < si->lowest_bit)
730 si->lowest_bit = offset;
731 if (end > si->highest_bit) {
732 bool was_full = !si->highest_bit;
734 WRITE_ONCE(si->highest_bit, end);
735 if (was_full && (si->flags & SWP_WRITEOK))
736 add_to_avail_list(si);
738 atomic_long_add(nr_entries, &nr_swap_pages);
739 WRITE_ONCE(si->inuse_pages, si->inuse_pages - nr_entries);
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 frontswap_invalidate_page(si->type, offset);
748 if (swap_slot_free_notify)
749 swap_slot_free_notify(si->bdev, offset);
752 clear_shadow_from_swap_cache(si->type, begin, end);
755 static void set_cluster_next(struct swap_info_struct *si, unsigned long next)
759 if (!(si->flags & SWP_SOLIDSTATE)) {
760 si->cluster_next = next;
764 prev = this_cpu_read(*si->cluster_next_cpu);
766 * Cross the swap address space size aligned trunk, choose
767 * another trunk randomly to avoid lock contention on swap
768 * address space if possible.
770 if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) !=
771 (next >> SWAP_ADDRESS_SPACE_SHIFT)) {
772 /* No free swap slots available */
773 if (si->highest_bit <= si->lowest_bit)
775 next = get_random_u32_inclusive(si->lowest_bit, si->highest_bit);
776 next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES);
777 next = max_t(unsigned int, next, si->lowest_bit);
779 this_cpu_write(*si->cluster_next_cpu, next);
782 static bool swap_offset_available_and_locked(struct swap_info_struct *si,
783 unsigned long offset)
785 if (data_race(!si->swap_map[offset])) {
786 spin_lock(&si->lock);
790 if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
791 spin_lock(&si->lock);
798 static int scan_swap_map_slots(struct swap_info_struct *si,
799 unsigned char usage, int nr,
802 struct swap_cluster_info *ci;
803 unsigned long offset;
804 unsigned long scan_base;
805 unsigned long last_in_cluster = 0;
806 int latency_ration = LATENCY_LIMIT;
808 bool scanned_many = false;
811 * We try to cluster swap pages by allocating them sequentially
812 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
813 * way, however, we resort to first-free allocation, starting
814 * a new cluster. This prevents us from scattering swap pages
815 * all over the entire swap partition, so that we reduce
816 * overall disk seek times between swap pages. -- sct
817 * But we do now try to find an empty cluster. -Andrea
818 * And we let swap pages go all over an SSD partition. Hugh
821 si->flags += SWP_SCANNING;
823 * Use percpu scan base for SSD to reduce lock contention on
824 * cluster and swap cache. For HDD, sequential access is more
827 if (si->flags & SWP_SOLIDSTATE)
828 scan_base = this_cpu_read(*si->cluster_next_cpu);
830 scan_base = si->cluster_next;
834 if (si->cluster_info) {
835 if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
837 } else if (unlikely(!si->cluster_nr--)) {
838 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
839 si->cluster_nr = SWAPFILE_CLUSTER - 1;
843 spin_unlock(&si->lock);
846 * If seek is expensive, start searching for new cluster from
847 * start of partition, to minimize the span of allocated swap.
848 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
849 * case, just handled by scan_swap_map_try_ssd_cluster() above.
851 scan_base = offset = si->lowest_bit;
852 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
854 /* Locate the first empty (unaligned) cluster */
855 for (; last_in_cluster <= si->highest_bit; offset++) {
856 if (si->swap_map[offset])
857 last_in_cluster = offset + SWAPFILE_CLUSTER;
858 else if (offset == last_in_cluster) {
859 spin_lock(&si->lock);
860 offset -= SWAPFILE_CLUSTER - 1;
861 si->cluster_next = offset;
862 si->cluster_nr = SWAPFILE_CLUSTER - 1;
865 if (unlikely(--latency_ration < 0)) {
867 latency_ration = LATENCY_LIMIT;
872 spin_lock(&si->lock);
873 si->cluster_nr = SWAPFILE_CLUSTER - 1;
877 if (si->cluster_info) {
878 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
879 /* take a break if we already got some slots */
882 if (!scan_swap_map_try_ssd_cluster(si, &offset,
887 if (!(si->flags & SWP_WRITEOK))
889 if (!si->highest_bit)
891 if (offset > si->highest_bit)
892 scan_base = offset = si->lowest_bit;
894 ci = lock_cluster(si, offset);
895 /* reuse swap entry of cache-only swap if not busy. */
896 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
899 spin_unlock(&si->lock);
900 swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
901 spin_lock(&si->lock);
902 /* entry was freed successfully, try to use this again */
905 goto scan; /* check next one */
908 if (si->swap_map[offset]) {
915 WRITE_ONCE(si->swap_map[offset], usage);
916 inc_cluster_info_page(si, si->cluster_info, offset);
919 swap_range_alloc(si, offset, 1);
920 slots[n_ret++] = swp_entry(si->type, offset);
922 /* got enough slots or reach max slots? */
923 if ((n_ret == nr) || (offset >= si->highest_bit))
926 /* search for next available slot */
928 /* time to take a break? */
929 if (unlikely(--latency_ration < 0)) {
932 spin_unlock(&si->lock);
934 spin_lock(&si->lock);
935 latency_ration = LATENCY_LIMIT;
938 /* try to get more slots in cluster */
939 if (si->cluster_info) {
940 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
942 } else if (si->cluster_nr && !si->swap_map[++offset]) {
943 /* non-ssd case, still more slots in cluster? */
949 * Even if there's no free clusters available (fragmented),
950 * try to scan a little more quickly with lock held unless we
951 * have scanned too many slots already.
954 unsigned long scan_limit;
956 if (offset < scan_base)
957 scan_limit = scan_base;
959 scan_limit = si->highest_bit;
960 for (; offset <= scan_limit && --latency_ration > 0;
962 if (!si->swap_map[offset])
968 set_cluster_next(si, offset + 1);
969 si->flags -= SWP_SCANNING;
973 spin_unlock(&si->lock);
974 while (++offset <= READ_ONCE(si->highest_bit)) {
975 if (unlikely(--latency_ration < 0)) {
977 latency_ration = LATENCY_LIMIT;
980 if (swap_offset_available_and_locked(si, offset))
983 offset = si->lowest_bit;
984 while (offset < scan_base) {
985 if (unlikely(--latency_ration < 0)) {
987 latency_ration = LATENCY_LIMIT;
990 if (swap_offset_available_and_locked(si, offset))
994 spin_lock(&si->lock);
997 si->flags -= SWP_SCANNING;
1001 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
1004 struct swap_cluster_info *ci;
1005 unsigned long offset;
1008 * Should not even be attempting cluster allocations when huge
1009 * page swap is disabled. Warn and fail the allocation.
1011 if (!IS_ENABLED(CONFIG_THP_SWAP)) {
1016 if (cluster_list_empty(&si->free_clusters))
1019 idx = cluster_list_first(&si->free_clusters);
1020 offset = idx * SWAPFILE_CLUSTER;
1021 ci = lock_cluster(si, offset);
1022 alloc_cluster(si, idx);
1023 cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
1025 memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER);
1027 swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
1028 *slot = swp_entry(si->type, offset);
1033 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
1035 unsigned long offset = idx * SWAPFILE_CLUSTER;
1036 struct swap_cluster_info *ci;
1038 ci = lock_cluster(si, offset);
1039 memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
1040 cluster_set_count_flag(ci, 0, 0);
1041 free_cluster(si, idx);
1043 swap_range_free(si, offset, SWAPFILE_CLUSTER);
1046 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
1048 unsigned long size = swap_entry_size(entry_size);
1049 struct swap_info_struct *si, *next;
1054 /* Only single cluster request supported */
1055 WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1057 spin_lock(&swap_avail_lock);
1059 avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1060 if (avail_pgs <= 0) {
1061 spin_unlock(&swap_avail_lock);
1065 n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
1067 atomic_long_sub(n_goal * size, &nr_swap_pages);
1070 node = numa_node_id();
1071 plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1072 /* requeue si to after same-priority siblings */
1073 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1074 spin_unlock(&swap_avail_lock);
1075 spin_lock(&si->lock);
1076 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1077 spin_lock(&swap_avail_lock);
1078 if (plist_node_empty(&si->avail_lists[node])) {
1079 spin_unlock(&si->lock);
1082 WARN(!si->highest_bit,
1083 "swap_info %d in list but !highest_bit\n",
1085 WARN(!(si->flags & SWP_WRITEOK),
1086 "swap_info %d in list but !SWP_WRITEOK\n",
1088 __del_from_avail_list(si);
1089 spin_unlock(&si->lock);
1092 if (size == SWAPFILE_CLUSTER) {
1093 if (si->flags & SWP_BLKDEV)
1094 n_ret = swap_alloc_cluster(si, swp_entries);
1096 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1097 n_goal, swp_entries);
1098 spin_unlock(&si->lock);
1099 if (n_ret || size == SWAPFILE_CLUSTER)
1103 spin_lock(&swap_avail_lock);
1106 * if we got here, it's likely that si was almost full before,
1107 * and since scan_swap_map_slots() can drop the si->lock,
1108 * multiple callers probably all tried to get a page from the
1109 * same si and it filled up before we could get one; or, the si
1110 * filled up between us dropping swap_avail_lock and taking
1111 * si->lock. Since we dropped the swap_avail_lock, the
1112 * swap_avail_head list may have been modified; so if next is
1113 * still in the swap_avail_head list then try it, otherwise
1114 * start over if we have not gotten any slots.
1116 if (plist_node_empty(&next->avail_lists[node]))
1120 spin_unlock(&swap_avail_lock);
1124 atomic_long_add((long)(n_goal - n_ret) * size,
1130 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1132 struct swap_info_struct *p;
1133 unsigned long offset;
1137 p = swp_swap_info(entry);
1140 if (data_race(!(p->flags & SWP_USED)))
1142 offset = swp_offset(entry);
1143 if (offset >= p->max)
1145 if (data_race(!p->swap_map[swp_offset(entry)]))
1150 pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val);
1153 pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1156 pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val);
1159 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1164 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1165 struct swap_info_struct *q)
1167 struct swap_info_struct *p;
1169 p = _swap_info_get(entry);
1173 spin_unlock(&q->lock);
1175 spin_lock(&p->lock);
1180 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1181 unsigned long offset,
1182 unsigned char usage)
1184 unsigned char count;
1185 unsigned char has_cache;
1187 count = p->swap_map[offset];
1189 has_cache = count & SWAP_HAS_CACHE;
1190 count &= ~SWAP_HAS_CACHE;
1192 if (usage == SWAP_HAS_CACHE) {
1193 VM_BUG_ON(!has_cache);
1195 } else if (count == SWAP_MAP_SHMEM) {
1197 * Or we could insist on shmem.c using a special
1198 * swap_shmem_free() and free_shmem_swap_and_cache()...
1201 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1202 if (count == COUNT_CONTINUED) {
1203 if (swap_count_continued(p, offset, count))
1204 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1206 count = SWAP_MAP_MAX;
1211 usage = count | has_cache;
1213 WRITE_ONCE(p->swap_map[offset], usage);
1215 WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE);
1221 * When we get a swap entry, if there aren't some other ways to
1222 * prevent swapoff, such as the folio in swap cache is locked, page
1223 * table lock is held, etc., the swap entry may become invalid because
1224 * of swapoff. Then, we need to enclose all swap related functions
1225 * with get_swap_device() and put_swap_device(), unless the swap
1226 * functions call get/put_swap_device() by themselves.
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 * Notice that swapoff or swapoff+swapon can still happen before the
1234 * percpu_ref_tryget_live() in get_swap_device() or after the
1235 * percpu_ref_put() in put_swap_device() if there isn't any other way
1236 * to prevent swapoff. The caller must be prepared for that. For
1237 * example, the following situation is possible.
1241 * ... swapoff+swapon
1242 * __read_swap_cache_async()
1243 * swapcache_prepare()
1244 * __swap_duplicate()
1246 * // verify PTE not changed
1248 * In __swap_duplicate(), the swap_map need to be checked before
1249 * changing partly because the specified swap entry may be for another
1250 * swap device which has been swapoff. And in do_swap_page(), after
1251 * the page is read from the swap device, the PTE is verified not
1252 * changed with the page table locked to check whether the swap device
1253 * has been swapoff or swapoff+swapon.
1255 struct swap_info_struct *get_swap_device(swp_entry_t entry)
1257 struct swap_info_struct *si;
1258 unsigned long offset;
1262 si = swp_swap_info(entry);
1265 if (!percpu_ref_tryget_live(&si->users))
1268 * Guarantee the si->users are checked before accessing other
1269 * fields of swap_info_struct.
1271 * Paired with the spin_unlock() after setup_swap_info() in
1272 * enable_swap_info().
1275 offset = swp_offset(entry);
1276 if (offset >= si->max)
1281 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1285 pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1286 percpu_ref_put(&si->users);
1290 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1293 struct swap_cluster_info *ci;
1294 unsigned long offset = swp_offset(entry);
1295 unsigned char usage;
1297 ci = lock_cluster_or_swap_info(p, offset);
1298 usage = __swap_entry_free_locked(p, offset, 1);
1299 unlock_cluster_or_swap_info(p, ci);
1301 free_swap_slot(entry);
1306 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1308 struct swap_cluster_info *ci;
1309 unsigned long offset = swp_offset(entry);
1310 unsigned char count;
1312 ci = lock_cluster(p, offset);
1313 count = p->swap_map[offset];
1314 VM_BUG_ON(count != SWAP_HAS_CACHE);
1315 p->swap_map[offset] = 0;
1316 dec_cluster_info_page(p, p->cluster_info, offset);
1319 mem_cgroup_uncharge_swap(entry, 1);
1320 swap_range_free(p, offset, 1);
1324 * Caller has made sure that the swap device corresponding to entry
1325 * is still around or has not been recycled.
1327 void swap_free(swp_entry_t entry)
1329 struct swap_info_struct *p;
1331 p = _swap_info_get(entry);
1333 __swap_entry_free(p, entry);
1337 * Called after dropping swapcache to decrease refcnt to swap entries.
1339 void put_swap_folio(struct folio *folio, swp_entry_t entry)
1341 unsigned long offset = swp_offset(entry);
1342 unsigned long idx = offset / SWAPFILE_CLUSTER;
1343 struct swap_cluster_info *ci;
1344 struct swap_info_struct *si;
1346 unsigned int i, free_entries = 0;
1348 int size = swap_entry_size(folio_nr_pages(folio));
1350 si = _swap_info_get(entry);
1354 ci = lock_cluster_or_swap_info(si, offset);
1355 if (size == SWAPFILE_CLUSTER) {
1356 VM_BUG_ON(!cluster_is_huge(ci));
1357 map = si->swap_map + offset;
1358 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1360 VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1361 if (val == SWAP_HAS_CACHE)
1364 cluster_clear_huge(ci);
1365 if (free_entries == SWAPFILE_CLUSTER) {
1366 unlock_cluster_or_swap_info(si, ci);
1367 spin_lock(&si->lock);
1368 mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1369 swap_free_cluster(si, idx);
1370 spin_unlock(&si->lock);
1374 for (i = 0; i < size; i++, entry.val++) {
1375 if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1376 unlock_cluster_or_swap_info(si, ci);
1377 free_swap_slot(entry);
1380 lock_cluster_or_swap_info(si, offset);
1383 unlock_cluster_or_swap_info(si, ci);
1386 #ifdef CONFIG_THP_SWAP
1387 int split_swap_cluster(swp_entry_t entry)
1389 struct swap_info_struct *si;
1390 struct swap_cluster_info *ci;
1391 unsigned long offset = swp_offset(entry);
1393 si = _swap_info_get(entry);
1396 ci = lock_cluster(si, offset);
1397 cluster_clear_huge(ci);
1403 static int swp_entry_cmp(const void *ent1, const void *ent2)
1405 const swp_entry_t *e1 = ent1, *e2 = ent2;
1407 return (int)swp_type(*e1) - (int)swp_type(*e2);
1410 void swapcache_free_entries(swp_entry_t *entries, int n)
1412 struct swap_info_struct *p, *prev;
1422 * Sort swap entries by swap device, so each lock is only taken once.
1423 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1424 * so low that it isn't necessary to optimize further.
1426 if (nr_swapfiles > 1)
1427 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1428 for (i = 0; i < n; ++i) {
1429 p = swap_info_get_cont(entries[i], prev);
1431 swap_entry_free(p, entries[i]);
1435 spin_unlock(&p->lock);
1438 int __swap_count(swp_entry_t entry)
1440 struct swap_info_struct *si = swp_swap_info(entry);
1441 pgoff_t offset = swp_offset(entry);
1443 return swap_count(si->swap_map[offset]);
1447 * How many references to @entry are currently swapped out?
1448 * This does not give an exact answer when swap count is continued,
1449 * but does include the high COUNT_CONTINUED flag to allow for that.
1451 int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1453 pgoff_t offset = swp_offset(entry);
1454 struct swap_cluster_info *ci;
1457 ci = lock_cluster_or_swap_info(si, offset);
1458 count = swap_count(si->swap_map[offset]);
1459 unlock_cluster_or_swap_info(si, ci);
1464 * How many references to @entry are currently swapped out?
1465 * This considers COUNT_CONTINUED so it returns exact answer.
1467 int swp_swapcount(swp_entry_t entry)
1469 int count, tmp_count, n;
1470 struct swap_info_struct *p;
1471 struct swap_cluster_info *ci;
1476 p = _swap_info_get(entry);
1480 offset = swp_offset(entry);
1482 ci = lock_cluster_or_swap_info(p, offset);
1484 count = swap_count(p->swap_map[offset]);
1485 if (!(count & COUNT_CONTINUED))
1488 count &= ~COUNT_CONTINUED;
1489 n = SWAP_MAP_MAX + 1;
1491 page = vmalloc_to_page(p->swap_map + offset);
1492 offset &= ~PAGE_MASK;
1493 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1496 page = list_next_entry(page, lru);
1497 map = kmap_atomic(page);
1498 tmp_count = map[offset];
1501 count += (tmp_count & ~COUNT_CONTINUED) * n;
1502 n *= (SWAP_CONT_MAX + 1);
1503 } while (tmp_count & COUNT_CONTINUED);
1505 unlock_cluster_or_swap_info(p, ci);
1509 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1512 struct swap_cluster_info *ci;
1513 unsigned char *map = si->swap_map;
1514 unsigned long roffset = swp_offset(entry);
1515 unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1519 ci = lock_cluster_or_swap_info(si, offset);
1520 if (!ci || !cluster_is_huge(ci)) {
1521 if (swap_count(map[roffset]))
1525 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1526 if (swap_count(map[offset + i])) {
1532 unlock_cluster_or_swap_info(si, ci);
1536 static bool folio_swapped(struct folio *folio)
1538 swp_entry_t entry = folio_swap_entry(folio);
1539 struct swap_info_struct *si = _swap_info_get(entry);
1544 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio)))
1545 return swap_swapcount(si, entry) != 0;
1547 return swap_page_trans_huge_swapped(si, entry);
1551 * folio_free_swap() - Free the swap space used for this folio.
1552 * @folio: The folio to remove.
1554 * If swap is getting full, or if there are no more mappings of this folio,
1555 * then call folio_free_swap to free its swap space.
1557 * Return: true if we were able to release the swap space.
1559 bool folio_free_swap(struct folio *folio)
1561 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1563 if (!folio_test_swapcache(folio))
1565 if (folio_test_writeback(folio))
1567 if (folio_swapped(folio))
1571 * Once hibernation has begun to create its image of memory,
1572 * there's a danger that one of the calls to folio_free_swap()
1573 * - most probably a call from __try_to_reclaim_swap() while
1574 * hibernation is allocating its own swap pages for the image,
1575 * but conceivably even a call from memory reclaim - will free
1576 * the swap from a folio which has already been recorded in the
1577 * image as a clean swapcache folio, and then reuse its swap for
1578 * another page of the image. On waking from hibernation, the
1579 * original folio might be freed under memory pressure, then
1580 * later read back in from swap, now with the wrong data.
1582 * Hibernation suspends storage while it is writing the image
1583 * to disk so check that here.
1585 if (pm_suspended_storage())
1588 delete_from_swap_cache(folio);
1589 folio_set_dirty(folio);
1594 * Free the swap entry like above, but also try to
1595 * free the page cache entry if it is the last user.
1597 int free_swap_and_cache(swp_entry_t entry)
1599 struct swap_info_struct *p;
1600 unsigned char count;
1602 if (non_swap_entry(entry))
1605 p = _swap_info_get(entry);
1607 count = __swap_entry_free(p, entry);
1608 if (count == SWAP_HAS_CACHE &&
1609 !swap_page_trans_huge_swapped(p, entry))
1610 __try_to_reclaim_swap(p, swp_offset(entry),
1611 TTRS_UNMAPPED | TTRS_FULL);
1616 #ifdef CONFIG_HIBERNATION
1618 swp_entry_t get_swap_page_of_type(int type)
1620 struct swap_info_struct *si = swap_type_to_swap_info(type);
1621 swp_entry_t entry = {0};
1626 /* This is called for allocating swap entry, not cache */
1627 spin_lock(&si->lock);
1628 if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry))
1629 atomic_long_dec(&nr_swap_pages);
1630 spin_unlock(&si->lock);
1636 * Find the swap type that corresponds to given device (if any).
1638 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1639 * from 0, in which the swap header is expected to be located.
1641 * This is needed for the suspend to disk (aka swsusp).
1643 int swap_type_of(dev_t device, sector_t offset)
1650 spin_lock(&swap_lock);
1651 for (type = 0; type < nr_swapfiles; type++) {
1652 struct swap_info_struct *sis = swap_info[type];
1654 if (!(sis->flags & SWP_WRITEOK))
1657 if (device == sis->bdev->bd_dev) {
1658 struct swap_extent *se = first_se(sis);
1660 if (se->start_block == offset) {
1661 spin_unlock(&swap_lock);
1666 spin_unlock(&swap_lock);
1670 int find_first_swap(dev_t *device)
1674 spin_lock(&swap_lock);
1675 for (type = 0; type < nr_swapfiles; type++) {
1676 struct swap_info_struct *sis = swap_info[type];
1678 if (!(sis->flags & SWP_WRITEOK))
1680 *device = sis->bdev->bd_dev;
1681 spin_unlock(&swap_lock);
1684 spin_unlock(&swap_lock);
1689 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1690 * corresponding to given index in swap_info (swap type).
1692 sector_t swapdev_block(int type, pgoff_t offset)
1694 struct swap_info_struct *si = swap_type_to_swap_info(type);
1695 struct swap_extent *se;
1697 if (!si || !(si->flags & SWP_WRITEOK))
1699 se = offset_to_swap_extent(si, offset);
1700 return se->start_block + (offset - se->start_page);
1704 * Return either the total number of swap pages of given type, or the number
1705 * of free pages of that type (depending on @free)
1707 * This is needed for software suspend
1709 unsigned int count_swap_pages(int type, int free)
1713 spin_lock(&swap_lock);
1714 if ((unsigned int)type < nr_swapfiles) {
1715 struct swap_info_struct *sis = swap_info[type];
1717 spin_lock(&sis->lock);
1718 if (sis->flags & SWP_WRITEOK) {
1721 n -= sis->inuse_pages;
1723 spin_unlock(&sis->lock);
1725 spin_unlock(&swap_lock);
1728 #endif /* CONFIG_HIBERNATION */
1730 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1732 return pte_same(pte_swp_clear_flags(pte), swp_pte);
1736 * No need to decide whether this PTE shares the swap entry with others,
1737 * just let do_wp_page work it out if a write is requested later - to
1738 * force COW, vm_page_prot omits write permission from any private vma.
1740 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1741 unsigned long addr, swp_entry_t entry, struct folio *folio)
1743 struct page *page = folio_file_page(folio, swp_offset(entry));
1744 struct page *swapcache;
1746 pte_t *pte, new_pte, old_pte;
1747 bool hwposioned = false;
1751 page = ksm_might_need_to_copy(page, vma, addr);
1752 if (unlikely(!page))
1754 else if (unlikely(PTR_ERR(page) == -EHWPOISON))
1757 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1758 if (unlikely(!pte || !pte_same_as_swp(ptep_get(pte),
1759 swp_entry_to_pte(entry)))) {
1764 old_pte = ptep_get(pte);
1766 if (unlikely(hwposioned || !PageUptodate(page))) {
1767 swp_entry_t swp_entry;
1769 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1771 swp_entry = make_hwpoison_entry(swapcache);
1774 swp_entry = make_poisoned_swp_entry();
1776 new_pte = swp_entry_to_pte(swp_entry);
1782 * Some architectures may have to restore extra metadata to the page
1783 * when reading from swap. This metadata may be indexed by swap entry
1784 * so this must be called before swap_free().
1786 arch_swap_restore(entry, page_folio(page));
1788 /* See do_swap_page() */
1789 BUG_ON(!PageAnon(page) && PageMappedToDisk(page));
1790 BUG_ON(PageAnon(page) && PageAnonExclusive(page));
1792 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1793 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1795 if (page == swapcache) {
1796 rmap_t rmap_flags = RMAP_NONE;
1799 * See do_swap_page(): PageWriteback() would be problematic.
1800 * However, we do a wait_on_page_writeback() just before this
1801 * call and have the page locked.
1803 VM_BUG_ON_PAGE(PageWriteback(page), page);
1804 if (pte_swp_exclusive(old_pte))
1805 rmap_flags |= RMAP_EXCLUSIVE;
1807 page_add_anon_rmap(page, vma, addr, rmap_flags);
1808 } else { /* ksm created a completely new copy */
1809 page_add_new_anon_rmap(page, vma, addr);
1810 lru_cache_add_inactive_or_unevictable(page, vma);
1812 new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot));
1813 if (pte_swp_soft_dirty(old_pte))
1814 new_pte = pte_mksoft_dirty(new_pte);
1815 if (pte_swp_uffd_wp(old_pte))
1816 new_pte = pte_mkuffd_wp(new_pte);
1818 set_pte_at(vma->vm_mm, addr, pte, new_pte);
1822 pte_unmap_unlock(pte, ptl);
1823 if (page != swapcache) {
1830 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1831 unsigned long addr, unsigned long end,
1835 struct swap_info_struct *si;
1837 si = swap_info[type];
1839 struct folio *folio;
1840 unsigned long offset;
1841 unsigned char swp_count;
1847 pte = pte_offset_map(pmd, addr);
1852 ptent = ptep_get_lockless(pte);
1854 if (!is_swap_pte(ptent))
1857 entry = pte_to_swp_entry(ptent);
1858 if (swp_type(entry) != type)
1861 offset = swp_offset(entry);
1865 folio = swap_cache_get_folio(entry, vma, addr);
1868 struct vm_fault vmf = {
1871 .real_address = addr,
1875 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1878 folio = page_folio(page);
1881 swp_count = READ_ONCE(si->swap_map[offset]);
1882 if (swp_count == 0 || swp_count == SWAP_MAP_BAD)
1888 folio_wait_writeback(folio);
1889 ret = unuse_pte(vma, pmd, addr, entry, folio);
1891 folio_unlock(folio);
1896 folio_free_swap(folio);
1897 folio_unlock(folio);
1899 } while (addr += PAGE_SIZE, addr != end);
1906 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1907 unsigned long addr, unsigned long end,
1914 pmd = pmd_offset(pud, addr);
1917 next = pmd_addr_end(addr, end);
1918 ret = unuse_pte_range(vma, pmd, addr, next, type);
1921 } while (pmd++, addr = next, addr != end);
1925 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1926 unsigned long addr, unsigned long end,
1933 pud = pud_offset(p4d, addr);
1935 next = pud_addr_end(addr, end);
1936 if (pud_none_or_clear_bad(pud))
1938 ret = unuse_pmd_range(vma, pud, addr, next, type);
1941 } while (pud++, addr = next, addr != end);
1945 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1946 unsigned long addr, unsigned long end,
1953 p4d = p4d_offset(pgd, addr);
1955 next = p4d_addr_end(addr, end);
1956 if (p4d_none_or_clear_bad(p4d))
1958 ret = unuse_pud_range(vma, p4d, addr, next, type);
1961 } while (p4d++, addr = next, addr != end);
1965 static int unuse_vma(struct vm_area_struct *vma, unsigned int type)
1968 unsigned long addr, end, next;
1971 addr = vma->vm_start;
1974 pgd = pgd_offset(vma->vm_mm, addr);
1976 next = pgd_addr_end(addr, end);
1977 if (pgd_none_or_clear_bad(pgd))
1979 ret = unuse_p4d_range(vma, pgd, addr, next, type);
1982 } while (pgd++, addr = next, addr != end);
1986 static int unuse_mm(struct mm_struct *mm, unsigned int type)
1988 struct vm_area_struct *vma;
1990 VMA_ITERATOR(vmi, mm, 0);
1993 for_each_vma(vmi, vma) {
1994 if (vma->anon_vma) {
1995 ret = unuse_vma(vma, type);
2002 mmap_read_unlock(mm);
2007 * Scan swap_map from current position to next entry still in use.
2008 * Return 0 if there are no inuse entries after prev till end of
2011 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2015 unsigned char count;
2018 * No need for swap_lock here: we're just looking
2019 * for whether an entry is in use, not modifying it; false
2020 * hits are okay, and sys_swapoff() has already prevented new
2021 * allocations from this area (while holding swap_lock).
2023 for (i = prev + 1; i < si->max; i++) {
2024 count = READ_ONCE(si->swap_map[i]);
2025 if (count && swap_count(count) != SWAP_MAP_BAD)
2027 if ((i % LATENCY_LIMIT) == 0)
2037 static int try_to_unuse(unsigned int type)
2039 struct mm_struct *prev_mm;
2040 struct mm_struct *mm;
2041 struct list_head *p;
2043 struct swap_info_struct *si = swap_info[type];
2044 struct folio *folio;
2048 if (!READ_ONCE(si->inuse_pages))
2052 retval = shmem_unuse(type);
2059 spin_lock(&mmlist_lock);
2060 p = &init_mm.mmlist;
2061 while (READ_ONCE(si->inuse_pages) &&
2062 !signal_pending(current) &&
2063 (p = p->next) != &init_mm.mmlist) {
2065 mm = list_entry(p, struct mm_struct, mmlist);
2066 if (!mmget_not_zero(mm))
2068 spin_unlock(&mmlist_lock);
2071 retval = unuse_mm(mm, type);
2078 * Make sure that we aren't completely killing
2079 * interactive performance.
2082 spin_lock(&mmlist_lock);
2084 spin_unlock(&mmlist_lock);
2089 while (READ_ONCE(si->inuse_pages) &&
2090 !signal_pending(current) &&
2091 (i = find_next_to_unuse(si, i)) != 0) {
2093 entry = swp_entry(type, i);
2094 folio = filemap_get_folio(swap_address_space(entry), i);
2099 * It is conceivable that a racing task removed this folio from
2100 * swap cache just before we acquired the page lock. The folio
2101 * might even be back in swap cache on another swap area. But
2102 * that is okay, folio_free_swap() only removes stale folios.
2105 folio_wait_writeback(folio);
2106 folio_free_swap(folio);
2107 folio_unlock(folio);
2112 * Lets check again to see if there are still swap entries in the map.
2113 * If yes, we would need to do retry the unuse logic again.
2114 * Under global memory pressure, swap entries can be reinserted back
2115 * into process space after the mmlist loop above passes over them.
2117 * Limit the number of retries? No: when mmget_not_zero()
2118 * above fails, that mm is likely to be freeing swap from
2119 * exit_mmap(), which proceeds at its own independent pace;
2120 * and even shmem_writepage() could have been preempted after
2121 * folio_alloc_swap(), temporarily hiding that swap. It's easy
2122 * and robust (though cpu-intensive) just to keep retrying.
2124 if (READ_ONCE(si->inuse_pages)) {
2125 if (!signal_pending(current))
2134 * After a successful try_to_unuse, if no swap is now in use, we know
2135 * we can empty the mmlist. swap_lock must be held on entry and exit.
2136 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2137 * added to the mmlist just after page_duplicate - before would be racy.
2139 static void drain_mmlist(void)
2141 struct list_head *p, *next;
2144 for (type = 0; type < nr_swapfiles; type++)
2145 if (swap_info[type]->inuse_pages)
2147 spin_lock(&mmlist_lock);
2148 list_for_each_safe(p, next, &init_mm.mmlist)
2150 spin_unlock(&mmlist_lock);
2154 * Free all of a swapdev's extent information
2156 static void destroy_swap_extents(struct swap_info_struct *sis)
2158 while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2159 struct rb_node *rb = sis->swap_extent_root.rb_node;
2160 struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2162 rb_erase(rb, &sis->swap_extent_root);
2166 if (sis->flags & SWP_ACTIVATED) {
2167 struct file *swap_file = sis->swap_file;
2168 struct address_space *mapping = swap_file->f_mapping;
2170 sis->flags &= ~SWP_ACTIVATED;
2171 if (mapping->a_ops->swap_deactivate)
2172 mapping->a_ops->swap_deactivate(swap_file);
2177 * Add a block range (and the corresponding page range) into this swapdev's
2180 * This function rather assumes that it is called in ascending page order.
2183 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2184 unsigned long nr_pages, sector_t start_block)
2186 struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2187 struct swap_extent *se;
2188 struct swap_extent *new_se;
2191 * place the new node at the right most since the
2192 * function is called in ascending page order.
2196 link = &parent->rb_right;
2200 se = rb_entry(parent, struct swap_extent, rb_node);
2201 BUG_ON(se->start_page + se->nr_pages != start_page);
2202 if (se->start_block + se->nr_pages == start_block) {
2204 se->nr_pages += nr_pages;
2209 /* No merge, insert a new extent. */
2210 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2213 new_se->start_page = start_page;
2214 new_se->nr_pages = nr_pages;
2215 new_se->start_block = start_block;
2217 rb_link_node(&new_se->rb_node, parent, link);
2218 rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2221 EXPORT_SYMBOL_GPL(add_swap_extent);
2224 * A `swap extent' is a simple thing which maps a contiguous range of pages
2225 * onto a contiguous range of disk blocks. A rbtree of swap extents is
2226 * built at swapon time and is then used at swap_writepage/swap_readpage
2227 * time for locating where on disk a page belongs.
2229 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2230 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2231 * swap files identically.
2233 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2234 * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2235 * swapfiles are handled *identically* after swapon time.
2237 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2238 * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray
2239 * blocks are found which do not fall within the PAGE_SIZE alignment
2240 * requirements, they are simply tossed out - we will never use those blocks
2243 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2244 * prevents users from writing to the swap device, which will corrupt memory.
2246 * The amount of disk space which a single swap extent represents varies.
2247 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2248 * extents in the rbtree. - akpm.
2250 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2252 struct file *swap_file = sis->swap_file;
2253 struct address_space *mapping = swap_file->f_mapping;
2254 struct inode *inode = mapping->host;
2257 if (S_ISBLK(inode->i_mode)) {
2258 ret = add_swap_extent(sis, 0, sis->max, 0);
2263 if (mapping->a_ops->swap_activate) {
2264 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2267 sis->flags |= SWP_ACTIVATED;
2268 if ((sis->flags & SWP_FS_OPS) &&
2269 sio_pool_init() != 0) {
2270 destroy_swap_extents(sis);
2276 return generic_swapfile_activate(sis, swap_file, span);
2279 static int swap_node(struct swap_info_struct *p)
2281 struct block_device *bdev;
2286 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2288 return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2291 static void setup_swap_info(struct swap_info_struct *p, int prio,
2292 unsigned char *swap_map,
2293 struct swap_cluster_info *cluster_info)
2300 p->prio = --least_priority;
2302 * the plist prio is negated because plist ordering is
2303 * low-to-high, while swap ordering is high-to-low
2305 p->list.prio = -p->prio;
2308 p->avail_lists[i].prio = -p->prio;
2310 if (swap_node(p) == i)
2311 p->avail_lists[i].prio = 1;
2313 p->avail_lists[i].prio = -p->prio;
2316 p->swap_map = swap_map;
2317 p->cluster_info = cluster_info;
2320 static void _enable_swap_info(struct swap_info_struct *p)
2322 p->flags |= SWP_WRITEOK;
2323 atomic_long_add(p->pages, &nr_swap_pages);
2324 total_swap_pages += p->pages;
2326 assert_spin_locked(&swap_lock);
2328 * both lists are plists, and thus priority ordered.
2329 * swap_active_head needs to be priority ordered for swapoff(),
2330 * which on removal of any swap_info_struct with an auto-assigned
2331 * (i.e. negative) priority increments the auto-assigned priority
2332 * of any lower-priority swap_info_structs.
2333 * swap_avail_head needs to be priority ordered for folio_alloc_swap(),
2334 * which allocates swap pages from the highest available priority
2337 plist_add(&p->list, &swap_active_head);
2339 /* add to available list iff swap device is not full */
2341 add_to_avail_list(p);
2344 static void enable_swap_info(struct swap_info_struct *p, int prio,
2345 unsigned char *swap_map,
2346 struct swap_cluster_info *cluster_info,
2347 unsigned long *frontswap_map)
2349 if (IS_ENABLED(CONFIG_FRONTSWAP))
2350 frontswap_init(p->type, frontswap_map);
2351 spin_lock(&swap_lock);
2352 spin_lock(&p->lock);
2353 setup_swap_info(p, prio, swap_map, cluster_info);
2354 spin_unlock(&p->lock);
2355 spin_unlock(&swap_lock);
2357 * Finished initializing swap device, now it's safe to reference it.
2359 percpu_ref_resurrect(&p->users);
2360 spin_lock(&swap_lock);
2361 spin_lock(&p->lock);
2362 _enable_swap_info(p);
2363 spin_unlock(&p->lock);
2364 spin_unlock(&swap_lock);
2367 static void reinsert_swap_info(struct swap_info_struct *p)
2369 spin_lock(&swap_lock);
2370 spin_lock(&p->lock);
2371 setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2372 _enable_swap_info(p);
2373 spin_unlock(&p->lock);
2374 spin_unlock(&swap_lock);
2377 bool has_usable_swap(void)
2381 spin_lock(&swap_lock);
2382 if (plist_head_empty(&swap_active_head))
2384 spin_unlock(&swap_lock);
2388 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2390 struct swap_info_struct *p = NULL;
2391 unsigned char *swap_map;
2392 struct swap_cluster_info *cluster_info;
2393 unsigned long *frontswap_map;
2394 struct file *swap_file, *victim;
2395 struct address_space *mapping;
2396 struct inode *inode;
2397 struct filename *pathname;
2399 unsigned int old_block_size;
2401 if (!capable(CAP_SYS_ADMIN))
2404 BUG_ON(!current->mm);
2406 pathname = getname(specialfile);
2407 if (IS_ERR(pathname))
2408 return PTR_ERR(pathname);
2410 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2411 err = PTR_ERR(victim);
2415 mapping = victim->f_mapping;
2416 spin_lock(&swap_lock);
2417 plist_for_each_entry(p, &swap_active_head, list) {
2418 if (p->flags & SWP_WRITEOK) {
2419 if (p->swap_file->f_mapping == mapping) {
2427 spin_unlock(&swap_lock);
2430 if (!security_vm_enough_memory_mm(current->mm, p->pages))
2431 vm_unacct_memory(p->pages);
2434 spin_unlock(&swap_lock);
2437 spin_lock(&p->lock);
2438 del_from_avail_list(p);
2440 struct swap_info_struct *si = p;
2443 plist_for_each_entry_continue(si, &swap_active_head, list) {
2446 for_each_node(nid) {
2447 if (si->avail_lists[nid].prio != 1)
2448 si->avail_lists[nid].prio--;
2453 plist_del(&p->list, &swap_active_head);
2454 atomic_long_sub(p->pages, &nr_swap_pages);
2455 total_swap_pages -= p->pages;
2456 p->flags &= ~SWP_WRITEOK;
2457 spin_unlock(&p->lock);
2458 spin_unlock(&swap_lock);
2460 disable_swap_slots_cache_lock();
2462 set_current_oom_origin();
2463 err = try_to_unuse(p->type);
2464 clear_current_oom_origin();
2467 /* re-insert swap space back into swap_list */
2468 reinsert_swap_info(p);
2469 reenable_swap_slots_cache_unlock();
2473 reenable_swap_slots_cache_unlock();
2476 * Wait for swap operations protected by get/put_swap_device()
2479 * We need synchronize_rcu() here to protect the accessing to
2480 * the swap cache data structure.
2482 percpu_ref_kill(&p->users);
2484 wait_for_completion(&p->comp);
2486 flush_work(&p->discard_work);
2488 destroy_swap_extents(p);
2489 if (p->flags & SWP_CONTINUED)
2490 free_swap_count_continuations(p);
2492 if (!p->bdev || !bdev_nonrot(p->bdev))
2493 atomic_dec(&nr_rotate_swap);
2495 mutex_lock(&swapon_mutex);
2496 spin_lock(&swap_lock);
2497 spin_lock(&p->lock);
2500 /* wait for anyone still in scan_swap_map_slots */
2501 p->highest_bit = 0; /* cuts scans short */
2502 while (p->flags >= SWP_SCANNING) {
2503 spin_unlock(&p->lock);
2504 spin_unlock(&swap_lock);
2505 schedule_timeout_uninterruptible(1);
2506 spin_lock(&swap_lock);
2507 spin_lock(&p->lock);
2510 swap_file = p->swap_file;
2511 old_block_size = p->old_block_size;
2512 p->swap_file = NULL;
2514 swap_map = p->swap_map;
2516 cluster_info = p->cluster_info;
2517 p->cluster_info = NULL;
2518 frontswap_map = frontswap_map_get(p);
2519 spin_unlock(&p->lock);
2520 spin_unlock(&swap_lock);
2521 arch_swap_invalidate_area(p->type);
2522 frontswap_invalidate_area(p->type);
2523 frontswap_map_set(p, NULL);
2524 mutex_unlock(&swapon_mutex);
2525 free_percpu(p->percpu_cluster);
2526 p->percpu_cluster = NULL;
2527 free_percpu(p->cluster_next_cpu);
2528 p->cluster_next_cpu = NULL;
2530 kvfree(cluster_info);
2531 kvfree(frontswap_map);
2532 /* Destroy swap account information */
2533 swap_cgroup_swapoff(p->type);
2534 exit_swap_address_space(p->type);
2536 inode = mapping->host;
2537 if (S_ISBLK(inode->i_mode)) {
2538 struct block_device *bdev = I_BDEV(inode);
2540 set_blocksize(bdev, old_block_size);
2541 blkdev_put(bdev, p);
2545 inode->i_flags &= ~S_SWAPFILE;
2546 inode_unlock(inode);
2547 filp_close(swap_file, NULL);
2550 * Clear the SWP_USED flag after all resources are freed so that swapon
2551 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2552 * not hold p->lock after we cleared its SWP_WRITEOK.
2554 spin_lock(&swap_lock);
2556 spin_unlock(&swap_lock);
2559 atomic_inc(&proc_poll_event);
2560 wake_up_interruptible(&proc_poll_wait);
2563 filp_close(victim, NULL);
2569 #ifdef CONFIG_PROC_FS
2570 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2572 struct seq_file *seq = file->private_data;
2574 poll_wait(file, &proc_poll_wait, wait);
2576 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2577 seq->poll_event = atomic_read(&proc_poll_event);
2578 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2581 return EPOLLIN | EPOLLRDNORM;
2585 static void *swap_start(struct seq_file *swap, loff_t *pos)
2587 struct swap_info_struct *si;
2591 mutex_lock(&swapon_mutex);
2594 return SEQ_START_TOKEN;
2596 for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2597 if (!(si->flags & SWP_USED) || !si->swap_map)
2606 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2608 struct swap_info_struct *si = v;
2611 if (v == SEQ_START_TOKEN)
2614 type = si->type + 1;
2617 for (; (si = swap_type_to_swap_info(type)); type++) {
2618 if (!(si->flags & SWP_USED) || !si->swap_map)
2626 static void swap_stop(struct seq_file *swap, void *v)
2628 mutex_unlock(&swapon_mutex);
2631 static int swap_show(struct seq_file *swap, void *v)
2633 struct swap_info_struct *si = v;
2636 unsigned long bytes, inuse;
2638 if (si == SEQ_START_TOKEN) {
2639 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2643 bytes = si->pages << (PAGE_SHIFT - 10);
2644 inuse = READ_ONCE(si->inuse_pages) << (PAGE_SHIFT - 10);
2646 file = si->swap_file;
2647 len = seq_file_path(swap, file, " \t\n\\");
2648 seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n",
2649 len < 40 ? 40 - len : 1, " ",
2650 S_ISBLK(file_inode(file)->i_mode) ?
2651 "partition" : "file\t",
2652 bytes, bytes < 10000000 ? "\t" : "",
2653 inuse, inuse < 10000000 ? "\t" : "",
2658 static const struct seq_operations swaps_op = {
2659 .start = swap_start,
2665 static int swaps_open(struct inode *inode, struct file *file)
2667 struct seq_file *seq;
2670 ret = seq_open(file, &swaps_op);
2674 seq = file->private_data;
2675 seq->poll_event = atomic_read(&proc_poll_event);
2679 static const struct proc_ops swaps_proc_ops = {
2680 .proc_flags = PROC_ENTRY_PERMANENT,
2681 .proc_open = swaps_open,
2682 .proc_read = seq_read,
2683 .proc_lseek = seq_lseek,
2684 .proc_release = seq_release,
2685 .proc_poll = swaps_poll,
2688 static int __init procswaps_init(void)
2690 proc_create("swaps", 0, NULL, &swaps_proc_ops);
2693 __initcall(procswaps_init);
2694 #endif /* CONFIG_PROC_FS */
2696 #ifdef MAX_SWAPFILES_CHECK
2697 static int __init max_swapfiles_check(void)
2699 MAX_SWAPFILES_CHECK();
2702 late_initcall(max_swapfiles_check);
2705 static struct swap_info_struct *alloc_swap_info(void)
2707 struct swap_info_struct *p;
2708 struct swap_info_struct *defer = NULL;
2712 p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2714 return ERR_PTR(-ENOMEM);
2716 if (percpu_ref_init(&p->users, swap_users_ref_free,
2717 PERCPU_REF_INIT_DEAD, GFP_KERNEL)) {
2719 return ERR_PTR(-ENOMEM);
2722 spin_lock(&swap_lock);
2723 for (type = 0; type < nr_swapfiles; type++) {
2724 if (!(swap_info[type]->flags & SWP_USED))
2727 if (type >= MAX_SWAPFILES) {
2728 spin_unlock(&swap_lock);
2729 percpu_ref_exit(&p->users);
2731 return ERR_PTR(-EPERM);
2733 if (type >= nr_swapfiles) {
2736 * Publish the swap_info_struct after initializing it.
2737 * Note that kvzalloc() above zeroes all its fields.
2739 smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */
2743 p = swap_info[type];
2745 * Do not memset this entry: a racing procfs swap_next()
2746 * would be relying on p->type to remain valid.
2749 p->swap_extent_root = RB_ROOT;
2750 plist_node_init(&p->list, 0);
2752 plist_node_init(&p->avail_lists[i], 0);
2753 p->flags = SWP_USED;
2754 spin_unlock(&swap_lock);
2756 percpu_ref_exit(&defer->users);
2759 spin_lock_init(&p->lock);
2760 spin_lock_init(&p->cont_lock);
2761 init_completion(&p->comp);
2766 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2770 if (S_ISBLK(inode->i_mode)) {
2771 p->bdev = blkdev_get_by_dev(inode->i_rdev,
2772 BLK_OPEN_READ | BLK_OPEN_WRITE, p, NULL);
2773 if (IS_ERR(p->bdev)) {
2774 error = PTR_ERR(p->bdev);
2778 p->old_block_size = block_size(p->bdev);
2779 error = set_blocksize(p->bdev, PAGE_SIZE);
2783 * Zoned block devices contain zones that have a sequential
2784 * write only restriction. Hence zoned block devices are not
2785 * suitable for swapping. Disallow them here.
2787 if (bdev_is_zoned(p->bdev))
2789 p->flags |= SWP_BLKDEV;
2790 } else if (S_ISREG(inode->i_mode)) {
2791 p->bdev = inode->i_sb->s_bdev;
2799 * Find out how many pages are allowed for a single swap device. There
2800 * are two limiting factors:
2801 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2802 * 2) the number of bits in the swap pte, as defined by the different
2805 * In order to find the largest possible bit mask, a swap entry with
2806 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2807 * decoded to a swp_entry_t again, and finally the swap offset is
2810 * This will mask all the bits from the initial ~0UL mask that can't
2811 * be encoded in either the swp_entry_t or the architecture definition
2814 unsigned long generic_max_swapfile_size(void)
2816 return swp_offset(pte_to_swp_entry(
2817 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2820 /* Can be overridden by an architecture for additional checks. */
2821 __weak unsigned long arch_max_swapfile_size(void)
2823 return generic_max_swapfile_size();
2826 static unsigned long read_swap_header(struct swap_info_struct *p,
2827 union swap_header *swap_header,
2828 struct inode *inode)
2831 unsigned long maxpages;
2832 unsigned long swapfilepages;
2833 unsigned long last_page;
2835 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2836 pr_err("Unable to find swap-space signature\n");
2840 /* swap partition endianness hack... */
2841 if (swab32(swap_header->info.version) == 1) {
2842 swab32s(&swap_header->info.version);
2843 swab32s(&swap_header->info.last_page);
2844 swab32s(&swap_header->info.nr_badpages);
2845 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2847 for (i = 0; i < swap_header->info.nr_badpages; i++)
2848 swab32s(&swap_header->info.badpages[i]);
2850 /* Check the swap header's sub-version */
2851 if (swap_header->info.version != 1) {
2852 pr_warn("Unable to handle swap header version %d\n",
2853 swap_header->info.version);
2858 p->cluster_next = 1;
2861 maxpages = swapfile_maximum_size;
2862 last_page = swap_header->info.last_page;
2864 pr_warn("Empty swap-file\n");
2867 if (last_page > maxpages) {
2868 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2869 maxpages << (PAGE_SHIFT - 10),
2870 last_page << (PAGE_SHIFT - 10));
2872 if (maxpages > last_page) {
2873 maxpages = last_page + 1;
2874 /* p->max is an unsigned int: don't overflow it */
2875 if ((unsigned int)maxpages == 0)
2876 maxpages = UINT_MAX;
2878 p->highest_bit = maxpages - 1;
2882 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2883 if (swapfilepages && maxpages > swapfilepages) {
2884 pr_warn("Swap area shorter than signature indicates\n");
2887 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2889 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2895 #define SWAP_CLUSTER_INFO_COLS \
2896 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2897 #define SWAP_CLUSTER_SPACE_COLS \
2898 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2899 #define SWAP_CLUSTER_COLS \
2900 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2902 static int setup_swap_map_and_extents(struct swap_info_struct *p,
2903 union swap_header *swap_header,
2904 unsigned char *swap_map,
2905 struct swap_cluster_info *cluster_info,
2906 unsigned long maxpages,
2910 unsigned int nr_good_pages;
2912 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2913 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2914 unsigned long i, idx;
2916 nr_good_pages = maxpages - 1; /* omit header page */
2918 cluster_list_init(&p->free_clusters);
2919 cluster_list_init(&p->discard_clusters);
2921 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2922 unsigned int page_nr = swap_header->info.badpages[i];
2923 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2925 if (page_nr < maxpages) {
2926 swap_map[page_nr] = SWAP_MAP_BAD;
2929 * Haven't marked the cluster free yet, no list
2930 * operation involved
2932 inc_cluster_info_page(p, cluster_info, page_nr);
2936 /* Haven't marked the cluster free yet, no list operation involved */
2937 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2938 inc_cluster_info_page(p, cluster_info, i);
2940 if (nr_good_pages) {
2941 swap_map[0] = SWAP_MAP_BAD;
2943 * Not mark the cluster free yet, no list
2944 * operation involved
2946 inc_cluster_info_page(p, cluster_info, 0);
2948 p->pages = nr_good_pages;
2949 nr_extents = setup_swap_extents(p, span);
2952 nr_good_pages = p->pages;
2954 if (!nr_good_pages) {
2955 pr_warn("Empty swap-file\n");
2964 * Reduce false cache line sharing between cluster_info and
2965 * sharing same address space.
2967 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2968 j = (k + col) % SWAP_CLUSTER_COLS;
2969 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2970 idx = i * SWAP_CLUSTER_COLS + j;
2971 if (idx >= nr_clusters)
2973 if (cluster_count(&cluster_info[idx]))
2975 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2976 cluster_list_add_tail(&p->free_clusters, cluster_info,
2983 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2985 struct swap_info_struct *p;
2986 struct filename *name;
2987 struct file *swap_file = NULL;
2988 struct address_space *mapping;
2989 struct dentry *dentry;
2992 union swap_header *swap_header;
2995 unsigned long maxpages;
2996 unsigned char *swap_map = NULL;
2997 struct swap_cluster_info *cluster_info = NULL;
2998 unsigned long *frontswap_map = NULL;
2999 struct page *page = NULL;
3000 struct inode *inode = NULL;
3001 bool inced_nr_rotate_swap = false;
3003 if (swap_flags & ~SWAP_FLAGS_VALID)
3006 if (!capable(CAP_SYS_ADMIN))
3009 if (!swap_avail_heads)
3012 p = alloc_swap_info();
3016 INIT_WORK(&p->discard_work, swap_discard_work);
3018 name = getname(specialfile);
3020 error = PTR_ERR(name);
3024 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3025 if (IS_ERR(swap_file)) {
3026 error = PTR_ERR(swap_file);
3031 p->swap_file = swap_file;
3032 mapping = swap_file->f_mapping;
3033 dentry = swap_file->f_path.dentry;
3034 inode = mapping->host;
3036 error = claim_swapfile(p, inode);
3037 if (unlikely(error))
3041 if (d_unlinked(dentry) || cant_mount(dentry)) {
3043 goto bad_swap_unlock_inode;
3045 if (IS_SWAPFILE(inode)) {
3047 goto bad_swap_unlock_inode;
3051 * Read the swap header.
3053 if (!mapping->a_ops->read_folio) {
3055 goto bad_swap_unlock_inode;
3057 page = read_mapping_page(mapping, 0, swap_file);
3059 error = PTR_ERR(page);
3060 goto bad_swap_unlock_inode;
3062 swap_header = kmap(page);
3064 maxpages = read_swap_header(p, swap_header, inode);
3065 if (unlikely(!maxpages)) {
3067 goto bad_swap_unlock_inode;
3070 /* OK, set up the swap map and apply the bad block list */
3071 swap_map = vzalloc(maxpages);
3074 goto bad_swap_unlock_inode;
3077 if (p->bdev && bdev_stable_writes(p->bdev))
3078 p->flags |= SWP_STABLE_WRITES;
3080 if (p->bdev && bdev_synchronous(p->bdev))
3081 p->flags |= SWP_SYNCHRONOUS_IO;
3083 if (p->bdev && bdev_nonrot(p->bdev)) {
3085 unsigned long ci, nr_cluster;
3087 p->flags |= SWP_SOLIDSTATE;
3088 p->cluster_next_cpu = alloc_percpu(unsigned int);
3089 if (!p->cluster_next_cpu) {
3091 goto bad_swap_unlock_inode;
3094 * select a random position to start with to help wear leveling
3097 for_each_possible_cpu(cpu) {
3098 per_cpu(*p->cluster_next_cpu, cpu) =
3099 get_random_u32_inclusive(1, p->highest_bit);
3101 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3103 cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3105 if (!cluster_info) {
3107 goto bad_swap_unlock_inode;
3110 for (ci = 0; ci < nr_cluster; ci++)
3111 spin_lock_init(&((cluster_info + ci)->lock));
3113 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3114 if (!p->percpu_cluster) {
3116 goto bad_swap_unlock_inode;
3118 for_each_possible_cpu(cpu) {
3119 struct percpu_cluster *cluster;
3120 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3121 cluster_set_null(&cluster->index);
3124 atomic_inc(&nr_rotate_swap);
3125 inced_nr_rotate_swap = true;
3128 error = swap_cgroup_swapon(p->type, maxpages);
3130 goto bad_swap_unlock_inode;
3132 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3133 cluster_info, maxpages, &span);
3134 if (unlikely(nr_extents < 0)) {
3136 goto bad_swap_unlock_inode;
3138 /* frontswap enabled? set up bit-per-page map for frontswap */
3139 if (IS_ENABLED(CONFIG_FRONTSWAP))
3140 frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3144 if ((swap_flags & SWAP_FLAG_DISCARD) &&
3145 p->bdev && bdev_max_discard_sectors(p->bdev)) {
3147 * When discard is enabled for swap with no particular
3148 * policy flagged, we set all swap discard flags here in
3149 * order to sustain backward compatibility with older
3150 * swapon(8) releases.
3152 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3156 * By flagging sys_swapon, a sysadmin can tell us to
3157 * either do single-time area discards only, or to just
3158 * perform discards for released swap page-clusters.
3159 * Now it's time to adjust the p->flags accordingly.
3161 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3162 p->flags &= ~SWP_PAGE_DISCARD;
3163 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3164 p->flags &= ~SWP_AREA_DISCARD;
3166 /* issue a swapon-time discard if it's still required */
3167 if (p->flags & SWP_AREA_DISCARD) {
3168 int err = discard_swap(p);
3170 pr_err("swapon: discard_swap(%p): %d\n",
3175 error = init_swap_address_space(p->type, maxpages);
3177 goto bad_swap_unlock_inode;
3180 * Flush any pending IO and dirty mappings before we start using this
3183 inode->i_flags |= S_SWAPFILE;
3184 error = inode_drain_writes(inode);
3186 inode->i_flags &= ~S_SWAPFILE;
3187 goto free_swap_address_space;
3190 mutex_lock(&swapon_mutex);
3192 if (swap_flags & SWAP_FLAG_PREFER)
3194 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3195 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3197 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3198 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3199 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3200 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3201 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3202 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3203 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3204 (frontswap_map) ? "FS" : "");
3206 mutex_unlock(&swapon_mutex);
3207 atomic_inc(&proc_poll_event);
3208 wake_up_interruptible(&proc_poll_wait);
3212 free_swap_address_space:
3213 exit_swap_address_space(p->type);
3214 bad_swap_unlock_inode:
3215 inode_unlock(inode);
3217 free_percpu(p->percpu_cluster);
3218 p->percpu_cluster = NULL;
3219 free_percpu(p->cluster_next_cpu);
3220 p->cluster_next_cpu = NULL;
3221 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3222 set_blocksize(p->bdev, p->old_block_size);
3223 blkdev_put(p->bdev, p);
3226 destroy_swap_extents(p);
3227 swap_cgroup_swapoff(p->type);
3228 spin_lock(&swap_lock);
3229 p->swap_file = NULL;
3231 spin_unlock(&swap_lock);
3233 kvfree(cluster_info);
3234 kvfree(frontswap_map);
3235 if (inced_nr_rotate_swap)
3236 atomic_dec(&nr_rotate_swap);
3238 filp_close(swap_file, NULL);
3240 if (page && !IS_ERR(page)) {
3247 inode_unlock(inode);
3249 enable_swap_slots_cache();
3253 void si_swapinfo(struct sysinfo *val)
3256 unsigned long nr_to_be_unused = 0;
3258 spin_lock(&swap_lock);
3259 for (type = 0; type < nr_swapfiles; type++) {
3260 struct swap_info_struct *si = swap_info[type];
3262 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3263 nr_to_be_unused += READ_ONCE(si->inuse_pages);
3265 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3266 val->totalswap = total_swap_pages + nr_to_be_unused;
3267 spin_unlock(&swap_lock);
3271 * Verify that a swap entry is valid and increment its swap map count.
3273 * Returns error code in following case.
3275 * - swp_entry is invalid -> EINVAL
3276 * - swp_entry is migration entry -> EINVAL
3277 * - swap-cache reference is requested but there is already one. -> EEXIST
3278 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3279 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3281 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3283 struct swap_info_struct *p;
3284 struct swap_cluster_info *ci;
3285 unsigned long offset;
3286 unsigned char count;
3287 unsigned char has_cache;
3290 p = swp_swap_info(entry);
3292 offset = swp_offset(entry);
3293 ci = lock_cluster_or_swap_info(p, offset);
3295 count = p->swap_map[offset];
3298 * swapin_readahead() doesn't check if a swap entry is valid, so the
3299 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3301 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3306 has_cache = count & SWAP_HAS_CACHE;
3307 count &= ~SWAP_HAS_CACHE;
3310 if (usage == SWAP_HAS_CACHE) {
3312 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3313 if (!has_cache && count)
3314 has_cache = SWAP_HAS_CACHE;
3315 else if (has_cache) /* someone else added cache */
3317 else /* no users remaining */
3320 } else if (count || has_cache) {
3322 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3324 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3326 else if (swap_count_continued(p, offset, count))
3327 count = COUNT_CONTINUED;
3331 err = -ENOENT; /* unused swap entry */
3333 WRITE_ONCE(p->swap_map[offset], count | has_cache);
3336 unlock_cluster_or_swap_info(p, ci);
3341 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3342 * (in which case its reference count is never incremented).
3344 void swap_shmem_alloc(swp_entry_t entry)
3346 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3350 * Increase reference count of swap entry by 1.
3351 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3352 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3353 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3354 * might occur if a page table entry has got corrupted.
3356 int swap_duplicate(swp_entry_t entry)
3360 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3361 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3366 * @entry: swap entry for which we allocate swap cache.
3368 * Called when allocating swap cache for existing swap entry,
3369 * This can return error codes. Returns 0 at success.
3370 * -EEXIST means there is a swap cache.
3371 * Note: return code is different from swap_duplicate().
3373 int swapcache_prepare(swp_entry_t entry)
3375 return __swap_duplicate(entry, SWAP_HAS_CACHE);
3378 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3380 return swap_type_to_swap_info(swp_type(entry));
3383 struct swap_info_struct *page_swap_info(struct page *page)
3385 swp_entry_t entry = { .val = page_private(page) };
3386 return swp_swap_info(entry);
3390 * out-of-line methods to avoid include hell.
3392 struct address_space *swapcache_mapping(struct folio *folio)
3394 return page_swap_info(&folio->page)->swap_file->f_mapping;
3396 EXPORT_SYMBOL_GPL(swapcache_mapping);
3398 pgoff_t __page_file_index(struct page *page)
3400 swp_entry_t swap = { .val = page_private(page) };
3401 return swp_offset(swap);
3403 EXPORT_SYMBOL_GPL(__page_file_index);
3406 * add_swap_count_continuation - called when a swap count is duplicated
3407 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3408 * page of the original vmalloc'ed swap_map, to hold the continuation count
3409 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3410 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3412 * These continuation pages are seldom referenced: the common paths all work
3413 * on the original swap_map, only referring to a continuation page when the
3414 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3416 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3417 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3418 * can be called after dropping locks.
3420 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3422 struct swap_info_struct *si;
3423 struct swap_cluster_info *ci;
3426 struct page *list_page;
3428 unsigned char count;
3432 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3433 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3435 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3437 si = get_swap_device(entry);
3440 * An acceptable race has occurred since the failing
3441 * __swap_duplicate(): the swap device may be swapoff
3445 spin_lock(&si->lock);
3447 offset = swp_offset(entry);
3449 ci = lock_cluster(si, offset);
3451 count = swap_count(si->swap_map[offset]);
3453 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3455 * The higher the swap count, the more likely it is that tasks
3456 * will race to add swap count continuation: we need to avoid
3457 * over-provisioning.
3467 head = vmalloc_to_page(si->swap_map + offset);
3468 offset &= ~PAGE_MASK;
3470 spin_lock(&si->cont_lock);
3472 * Page allocation does not initialize the page's lru field,
3473 * but it does always reset its private field.
3475 if (!page_private(head)) {
3476 BUG_ON(count & COUNT_CONTINUED);
3477 INIT_LIST_HEAD(&head->lru);
3478 set_page_private(head, SWP_CONTINUED);
3479 si->flags |= SWP_CONTINUED;
3482 list_for_each_entry(list_page, &head->lru, lru) {
3486 * If the previous map said no continuation, but we've found
3487 * a continuation page, free our allocation and use this one.
3489 if (!(count & COUNT_CONTINUED))
3490 goto out_unlock_cont;
3492 map = kmap_atomic(list_page) + offset;
3497 * If this continuation count now has some space in it,
3498 * free our allocation and use this one.
3500 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3501 goto out_unlock_cont;
3504 list_add_tail(&page->lru, &head->lru);
3505 page = NULL; /* now it's attached, don't free it */
3507 spin_unlock(&si->cont_lock);
3510 spin_unlock(&si->lock);
3511 put_swap_device(si);
3519 * swap_count_continued - when the original swap_map count is incremented
3520 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3521 * into, carry if so, or else fail until a new continuation page is allocated;
3522 * when the original swap_map count is decremented from 0 with continuation,
3523 * borrow from the continuation and report whether it still holds more.
3524 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3527 static bool swap_count_continued(struct swap_info_struct *si,
3528 pgoff_t offset, unsigned char count)
3535 head = vmalloc_to_page(si->swap_map + offset);
3536 if (page_private(head) != SWP_CONTINUED) {
3537 BUG_ON(count & COUNT_CONTINUED);
3538 return false; /* need to add count continuation */
3541 spin_lock(&si->cont_lock);
3542 offset &= ~PAGE_MASK;
3543 page = list_next_entry(head, lru);
3544 map = kmap_atomic(page) + offset;
3546 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3547 goto init_map; /* jump over SWAP_CONT_MAX checks */
3549 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3551 * Think of how you add 1 to 999
3553 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3555 page = list_next_entry(page, lru);
3556 BUG_ON(page == head);
3557 map = kmap_atomic(page) + offset;
3559 if (*map == SWAP_CONT_MAX) {
3561 page = list_next_entry(page, lru);
3563 ret = false; /* add count continuation */
3566 map = kmap_atomic(page) + offset;
3567 init_map: *map = 0; /* we didn't zero the page */
3571 while ((page = list_prev_entry(page, lru)) != head) {
3572 map = kmap_atomic(page) + offset;
3573 *map = COUNT_CONTINUED;
3576 ret = true; /* incremented */
3578 } else { /* decrementing */
3580 * Think of how you subtract 1 from 1000
3582 BUG_ON(count != COUNT_CONTINUED);
3583 while (*map == COUNT_CONTINUED) {
3585 page = list_next_entry(page, lru);
3586 BUG_ON(page == head);
3587 map = kmap_atomic(page) + offset;
3594 while ((page = list_prev_entry(page, lru)) != head) {
3595 map = kmap_atomic(page) + offset;
3596 *map = SWAP_CONT_MAX | count;
3597 count = COUNT_CONTINUED;
3600 ret = count == COUNT_CONTINUED;
3603 spin_unlock(&si->cont_lock);
3608 * free_swap_count_continuations - swapoff free all the continuation pages
3609 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3611 static void free_swap_count_continuations(struct swap_info_struct *si)
3615 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3617 head = vmalloc_to_page(si->swap_map + offset);
3618 if (page_private(head)) {
3619 struct page *page, *next;
3621 list_for_each_entry_safe(page, next, &head->lru, lru) {
3622 list_del(&page->lru);
3629 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3630 void __folio_throttle_swaprate(struct folio *folio, gfp_t gfp)
3632 struct swap_info_struct *si, *next;
3633 int nid = folio_nid(folio);
3635 if (!(gfp & __GFP_IO))
3638 if (!blk_cgroup_congested())
3642 * We've already scheduled a throttle, avoid taking the global swap
3645 if (current->throttle_disk)
3648 spin_lock(&swap_avail_lock);
3649 plist_for_each_entry_safe(si, next, &swap_avail_heads[nid],
3652 blkcg_schedule_throttle(si->bdev->bd_disk, true);
3656 spin_unlock(&swap_avail_lock);
3660 static int __init swapfile_init(void)
3664 swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3666 if (!swap_avail_heads) {
3667 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3672 plist_head_init(&swap_avail_heads[nid]);
3674 swapfile_maximum_size = arch_max_swapfile_size();
3676 #ifdef CONFIG_MIGRATION
3677 if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS))
3678 swap_migration_ad_supported = true;
3679 #endif /* CONFIG_MIGRATION */
3683 subsys_initcall(swapfile_init);