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>
45 #include <asm/tlbflush.h>
46 #include <linux/swapops.h>
47 #include <linux/swap_cgroup.h>
50 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
52 static void free_swap_count_continuations(struct swap_info_struct *);
54 static DEFINE_SPINLOCK(swap_lock);
55 static unsigned int nr_swapfiles;
56 atomic_long_t nr_swap_pages;
58 * Some modules use swappable objects and may try to swap them out under
59 * memory pressure (via the shrinker). Before doing so, they may wish to
60 * check to see if any swap space is available.
62 EXPORT_SYMBOL_GPL(nr_swap_pages);
63 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
64 long total_swap_pages;
65 static int least_priority = -1;
66 unsigned long swapfile_maximum_size;
67 #ifdef CONFIG_MIGRATION
68 bool swap_migration_ad_supported;
69 #endif /* CONFIG_MIGRATION */
71 static const char Bad_file[] = "Bad swap file entry ";
72 static const char Unused_file[] = "Unused swap file entry ";
73 static const char Bad_offset[] = "Bad swap offset entry ";
74 static const char Unused_offset[] = "Unused swap offset entry ";
77 * all active swap_info_structs
78 * protected with swap_lock, and ordered by priority.
80 static PLIST_HEAD(swap_active_head);
83 * all available (active, not full) swap_info_structs
84 * protected with swap_avail_lock, ordered by priority.
85 * This is used by folio_alloc_swap() instead of swap_active_head
86 * because swap_active_head includes all swap_info_structs,
87 * but folio_alloc_swap() doesn't need to look at full ones.
88 * This uses its own lock instead of swap_lock because when a
89 * swap_info_struct changes between not-full/full, it needs to
90 * add/remove itself to/from this list, but the swap_info_struct->lock
91 * is held and the locking order requires swap_lock to be taken
92 * before any swap_info_struct->lock.
94 static struct plist_head *swap_avail_heads;
95 static DEFINE_SPINLOCK(swap_avail_lock);
97 struct swap_info_struct *swap_info[MAX_SWAPFILES];
99 static DEFINE_MUTEX(swapon_mutex);
101 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
102 /* Activity counter to indicate that a swapon or swapoff has occurred */
103 static atomic_t proc_poll_event = ATOMIC_INIT(0);
105 atomic_t nr_rotate_swap = ATOMIC_INIT(0);
107 static struct swap_info_struct *swap_type_to_swap_info(int type)
109 if (type >= MAX_SWAPFILES)
112 return READ_ONCE(swap_info[type]); /* rcu_dereference() */
115 static inline unsigned char swap_count(unsigned char ent)
117 return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */
120 /* Reclaim the swap entry anyway if possible */
121 #define TTRS_ANYWAY 0x1
123 * Reclaim the swap entry if there are no more mappings of the
126 #define TTRS_UNMAPPED 0x2
127 /* Reclaim the swap entry if swap is getting full*/
128 #define TTRS_FULL 0x4
130 /* returns 1 if swap entry is freed */
131 static int __try_to_reclaim_swap(struct swap_info_struct *si,
132 unsigned long offset, unsigned long flags)
134 swp_entry_t entry = swp_entry(si->type, offset);
138 folio = filemap_get_folio(swap_address_space(entry), offset);
142 * When this function is called from scan_swap_map_slots() and it's
143 * called by vmscan.c at reclaiming folios. So we hold a folio lock
144 * here. We have to use trylock for avoiding deadlock. This is a special
145 * case and you should use folio_free_swap() with explicit folio_lock()
146 * in usual operations.
148 if (folio_trylock(folio)) {
149 if ((flags & TTRS_ANYWAY) ||
150 ((flags & TTRS_UNMAPPED) && !folio_mapped(folio)) ||
151 ((flags & TTRS_FULL) && mem_cgroup_swap_full(folio)))
152 ret = folio_free_swap(folio);
159 static inline struct swap_extent *first_se(struct swap_info_struct *sis)
161 struct rb_node *rb = rb_first(&sis->swap_extent_root);
162 return rb_entry(rb, struct swap_extent, rb_node);
165 static inline struct swap_extent *next_se(struct swap_extent *se)
167 struct rb_node *rb = rb_next(&se->rb_node);
168 return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
172 * swapon tell device that all the old swap contents can be discarded,
173 * to allow the swap device to optimize its wear-levelling.
175 static int discard_swap(struct swap_info_struct *si)
177 struct swap_extent *se;
178 sector_t start_block;
182 /* Do not discard the swap header page! */
184 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
185 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
187 err = blkdev_issue_discard(si->bdev, start_block,
188 nr_blocks, GFP_KERNEL);
194 for (se = next_se(se); se; se = next_se(se)) {
195 start_block = se->start_block << (PAGE_SHIFT - 9);
196 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
198 err = blkdev_issue_discard(si->bdev, start_block,
199 nr_blocks, GFP_KERNEL);
205 return err; /* That will often be -EOPNOTSUPP */
208 static struct swap_extent *
209 offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
211 struct swap_extent *se;
214 rb = sis->swap_extent_root.rb_node;
216 se = rb_entry(rb, struct swap_extent, rb_node);
217 if (offset < se->start_page)
219 else if (offset >= se->start_page + se->nr_pages)
224 /* It *must* be present */
228 sector_t swap_page_sector(struct page *page)
230 struct swap_info_struct *sis = page_swap_info(page);
231 struct swap_extent *se;
235 offset = __page_file_index(page);
236 se = offset_to_swap_extent(sis, offset);
237 sector = se->start_block + (offset - se->start_page);
238 return sector << (PAGE_SHIFT - 9);
242 * swap allocation tell device that a cluster of swap can now be discarded,
243 * to allow the swap device to optimize its wear-levelling.
245 static void discard_swap_cluster(struct swap_info_struct *si,
246 pgoff_t start_page, pgoff_t nr_pages)
248 struct swap_extent *se = offset_to_swap_extent(si, start_page);
251 pgoff_t offset = start_page - se->start_page;
252 sector_t start_block = se->start_block + offset;
253 sector_t nr_blocks = se->nr_pages - offset;
255 if (nr_blocks > nr_pages)
256 nr_blocks = nr_pages;
257 start_page += nr_blocks;
258 nr_pages -= nr_blocks;
260 start_block <<= PAGE_SHIFT - 9;
261 nr_blocks <<= PAGE_SHIFT - 9;
262 if (blkdev_issue_discard(si->bdev, start_block,
263 nr_blocks, GFP_NOIO))
270 #ifdef CONFIG_THP_SWAP
271 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
273 #define swap_entry_size(size) (size)
275 #define SWAPFILE_CLUSTER 256
278 * Define swap_entry_size() as constant to let compiler to optimize
279 * out some code if !CONFIG_THP_SWAP
281 #define swap_entry_size(size) 1
283 #define LATENCY_LIMIT 256
285 static inline void cluster_set_flag(struct swap_cluster_info *info,
291 static inline unsigned int cluster_count(struct swap_cluster_info *info)
296 static inline void cluster_set_count(struct swap_cluster_info *info,
302 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
303 unsigned int c, unsigned int f)
309 static inline unsigned int cluster_next(struct swap_cluster_info *info)
314 static inline void cluster_set_next(struct swap_cluster_info *info,
320 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
321 unsigned int n, unsigned int f)
327 static inline bool cluster_is_free(struct swap_cluster_info *info)
329 return info->flags & CLUSTER_FLAG_FREE;
332 static inline bool cluster_is_null(struct swap_cluster_info *info)
334 return info->flags & CLUSTER_FLAG_NEXT_NULL;
337 static inline void cluster_set_null(struct swap_cluster_info *info)
339 info->flags = CLUSTER_FLAG_NEXT_NULL;
343 static inline bool cluster_is_huge(struct swap_cluster_info *info)
345 if (IS_ENABLED(CONFIG_THP_SWAP))
346 return info->flags & CLUSTER_FLAG_HUGE;
350 static inline void cluster_clear_huge(struct swap_cluster_info *info)
352 info->flags &= ~CLUSTER_FLAG_HUGE;
355 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
356 unsigned long offset)
358 struct swap_cluster_info *ci;
360 ci = si->cluster_info;
362 ci += offset / SWAPFILE_CLUSTER;
363 spin_lock(&ci->lock);
368 static inline void unlock_cluster(struct swap_cluster_info *ci)
371 spin_unlock(&ci->lock);
375 * Determine the locking method in use for this device. Return
376 * swap_cluster_info if SSD-style cluster-based locking is in place.
378 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
379 struct swap_info_struct *si, unsigned long offset)
381 struct swap_cluster_info *ci;
383 /* Try to use fine-grained SSD-style locking if available: */
384 ci = lock_cluster(si, offset);
385 /* Otherwise, fall back to traditional, coarse locking: */
387 spin_lock(&si->lock);
392 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
393 struct swap_cluster_info *ci)
398 spin_unlock(&si->lock);
401 static inline bool cluster_list_empty(struct swap_cluster_list *list)
403 return cluster_is_null(&list->head);
406 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
408 return cluster_next(&list->head);
411 static void cluster_list_init(struct swap_cluster_list *list)
413 cluster_set_null(&list->head);
414 cluster_set_null(&list->tail);
417 static void cluster_list_add_tail(struct swap_cluster_list *list,
418 struct swap_cluster_info *ci,
421 if (cluster_list_empty(list)) {
422 cluster_set_next_flag(&list->head, idx, 0);
423 cluster_set_next_flag(&list->tail, idx, 0);
425 struct swap_cluster_info *ci_tail;
426 unsigned int tail = cluster_next(&list->tail);
429 * Nested cluster lock, but both cluster locks are
430 * only acquired when we held swap_info_struct->lock
433 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
434 cluster_set_next(ci_tail, idx);
435 spin_unlock(&ci_tail->lock);
436 cluster_set_next_flag(&list->tail, idx, 0);
440 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
441 struct swap_cluster_info *ci)
445 idx = cluster_next(&list->head);
446 if (cluster_next(&list->tail) == idx) {
447 cluster_set_null(&list->head);
448 cluster_set_null(&list->tail);
450 cluster_set_next_flag(&list->head,
451 cluster_next(&ci[idx]), 0);
456 /* Add a cluster to discard list and schedule it to do discard */
457 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
461 * If scan_swap_map_slots() can't find a free cluster, it will check
462 * si->swap_map directly. To make sure the discarding cluster isn't
463 * taken by scan_swap_map_slots(), mark the swap entries bad (occupied).
464 * It will be cleared after discard
466 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
467 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
469 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
471 schedule_work(&si->discard_work);
474 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
476 struct swap_cluster_info *ci = si->cluster_info;
478 cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
479 cluster_list_add_tail(&si->free_clusters, ci, idx);
483 * Doing discard actually. After a cluster discard is finished, the cluster
484 * will be added to free cluster list. caller should hold si->lock.
486 static void swap_do_scheduled_discard(struct swap_info_struct *si)
488 struct swap_cluster_info *info, *ci;
491 info = si->cluster_info;
493 while (!cluster_list_empty(&si->discard_clusters)) {
494 idx = cluster_list_del_first(&si->discard_clusters, info);
495 spin_unlock(&si->lock);
497 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
500 spin_lock(&si->lock);
501 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
502 __free_cluster(si, idx);
503 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
504 0, SWAPFILE_CLUSTER);
509 static void swap_discard_work(struct work_struct *work)
511 struct swap_info_struct *si;
513 si = container_of(work, struct swap_info_struct, discard_work);
515 spin_lock(&si->lock);
516 swap_do_scheduled_discard(si);
517 spin_unlock(&si->lock);
520 static void swap_users_ref_free(struct percpu_ref *ref)
522 struct swap_info_struct *si;
524 si = container_of(ref, struct swap_info_struct, users);
528 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
530 struct swap_cluster_info *ci = si->cluster_info;
532 VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
533 cluster_list_del_first(&si->free_clusters, ci);
534 cluster_set_count_flag(ci + idx, 0, 0);
537 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
539 struct swap_cluster_info *ci = si->cluster_info + idx;
541 VM_BUG_ON(cluster_count(ci) != 0);
543 * If the swap is discardable, prepare discard the cluster
544 * instead of free it immediately. The cluster will be freed
547 if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
548 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
549 swap_cluster_schedule_discard(si, idx);
553 __free_cluster(si, idx);
557 * The cluster corresponding to page_nr will be used. The cluster will be
558 * removed from free cluster list and its usage counter will be increased.
560 static void inc_cluster_info_page(struct swap_info_struct *p,
561 struct swap_cluster_info *cluster_info, unsigned long page_nr)
563 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
567 if (cluster_is_free(&cluster_info[idx]))
568 alloc_cluster(p, idx);
570 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
571 cluster_set_count(&cluster_info[idx],
572 cluster_count(&cluster_info[idx]) + 1);
576 * The cluster corresponding to page_nr decreases one usage. If the usage
577 * counter becomes 0, which means no page in the cluster is in using, we can
578 * optionally discard the cluster and add it to free cluster list.
580 static void dec_cluster_info_page(struct swap_info_struct *p,
581 struct swap_cluster_info *cluster_info, unsigned long page_nr)
583 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
588 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
589 cluster_set_count(&cluster_info[idx],
590 cluster_count(&cluster_info[idx]) - 1);
592 if (cluster_count(&cluster_info[idx]) == 0)
593 free_cluster(p, idx);
597 * It's possible scan_swap_map_slots() uses a free cluster in the middle of free
598 * cluster list. Avoiding such abuse to avoid list corruption.
601 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
602 unsigned long offset)
604 struct percpu_cluster *percpu_cluster;
607 offset /= SWAPFILE_CLUSTER;
608 conflict = !cluster_list_empty(&si->free_clusters) &&
609 offset != cluster_list_first(&si->free_clusters) &&
610 cluster_is_free(&si->cluster_info[offset]);
615 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
616 cluster_set_null(&percpu_cluster->index);
621 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
622 * might involve allocating a new cluster for current CPU too.
624 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
625 unsigned long *offset, unsigned long *scan_base)
627 struct percpu_cluster *cluster;
628 struct swap_cluster_info *ci;
629 unsigned long tmp, max;
632 cluster = this_cpu_ptr(si->percpu_cluster);
633 if (cluster_is_null(&cluster->index)) {
634 if (!cluster_list_empty(&si->free_clusters)) {
635 cluster->index = si->free_clusters.head;
636 cluster->next = cluster_next(&cluster->index) *
638 } else if (!cluster_list_empty(&si->discard_clusters)) {
640 * we don't have free cluster but have some clusters in
641 * discarding, do discard now and reclaim them, then
642 * reread cluster_next_cpu since we dropped si->lock
644 swap_do_scheduled_discard(si);
645 *scan_base = this_cpu_read(*si->cluster_next_cpu);
646 *offset = *scan_base;
653 * Other CPUs can use our cluster if they can't find a free cluster,
654 * check if there is still free entry in the cluster
657 max = min_t(unsigned long, si->max,
658 (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
660 ci = lock_cluster(si, tmp);
662 if (!si->swap_map[tmp])
669 cluster_set_null(&cluster->index);
672 cluster->next = tmp + 1;
678 static void __del_from_avail_list(struct swap_info_struct *p)
683 plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
686 static void del_from_avail_list(struct swap_info_struct *p)
688 spin_lock(&swap_avail_lock);
689 __del_from_avail_list(p);
690 spin_unlock(&swap_avail_lock);
693 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
694 unsigned int nr_entries)
696 unsigned int end = offset + nr_entries - 1;
698 if (offset == si->lowest_bit)
699 si->lowest_bit += nr_entries;
700 if (end == si->highest_bit)
701 WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries);
702 WRITE_ONCE(si->inuse_pages, si->inuse_pages + nr_entries);
703 if (si->inuse_pages == si->pages) {
704 si->lowest_bit = si->max;
706 del_from_avail_list(si);
710 static void add_to_avail_list(struct swap_info_struct *p)
714 spin_lock(&swap_avail_lock);
716 WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
717 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 = si->lowest_bit +
776 prandom_u32_max(si->highest_bit - si->lowest_bit + 1);
777 next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES);
778 next = max_t(unsigned int, next, si->lowest_bit);
780 this_cpu_write(*si->cluster_next_cpu, next);
783 static bool swap_offset_available_and_locked(struct swap_info_struct *si,
784 unsigned long offset)
786 if (data_race(!si->swap_map[offset])) {
787 spin_lock(&si->lock);
791 if (vm_swap_full() && READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
792 spin_lock(&si->lock);
799 static int scan_swap_map_slots(struct swap_info_struct *si,
800 unsigned char usage, int nr,
803 struct swap_cluster_info *ci;
804 unsigned long offset;
805 unsigned long scan_base;
806 unsigned long last_in_cluster = 0;
807 int latency_ration = LATENCY_LIMIT;
809 bool scanned_many = false;
812 * We try to cluster swap pages by allocating them sequentially
813 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
814 * way, however, we resort to first-free allocation, starting
815 * a new cluster. This prevents us from scattering swap pages
816 * all over the entire swap partition, so that we reduce
817 * overall disk seek times between swap pages. -- sct
818 * But we do now try to find an empty cluster. -Andrea
819 * And we let swap pages go all over an SSD partition. Hugh
822 si->flags += SWP_SCANNING;
824 * Use percpu scan base for SSD to reduce lock contention on
825 * cluster and swap cache. For HDD, sequential access is more
828 if (si->flags & SWP_SOLIDSTATE)
829 scan_base = this_cpu_read(*si->cluster_next_cpu);
831 scan_base = si->cluster_next;
835 if (si->cluster_info) {
836 if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
838 } else if (unlikely(!si->cluster_nr--)) {
839 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
840 si->cluster_nr = SWAPFILE_CLUSTER - 1;
844 spin_unlock(&si->lock);
847 * If seek is expensive, start searching for new cluster from
848 * start of partition, to minimize the span of allocated swap.
849 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
850 * case, just handled by scan_swap_map_try_ssd_cluster() above.
852 scan_base = offset = si->lowest_bit;
853 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
855 /* Locate the first empty (unaligned) cluster */
856 for (; last_in_cluster <= si->highest_bit; offset++) {
857 if (si->swap_map[offset])
858 last_in_cluster = offset + SWAPFILE_CLUSTER;
859 else if (offset == last_in_cluster) {
860 spin_lock(&si->lock);
861 offset -= SWAPFILE_CLUSTER - 1;
862 si->cluster_next = offset;
863 si->cluster_nr = SWAPFILE_CLUSTER - 1;
866 if (unlikely(--latency_ration < 0)) {
868 latency_ration = LATENCY_LIMIT;
873 spin_lock(&si->lock);
874 si->cluster_nr = SWAPFILE_CLUSTER - 1;
878 if (si->cluster_info) {
879 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
880 /* take a break if we already got some slots */
883 if (!scan_swap_map_try_ssd_cluster(si, &offset,
888 if (!(si->flags & SWP_WRITEOK))
890 if (!si->highest_bit)
892 if (offset > si->highest_bit)
893 scan_base = offset = si->lowest_bit;
895 ci = lock_cluster(si, offset);
896 /* reuse swap entry of cache-only swap if not busy. */
897 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
900 spin_unlock(&si->lock);
901 swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
902 spin_lock(&si->lock);
903 /* entry was freed successfully, try to use this again */
906 goto scan; /* check next one */
909 if (si->swap_map[offset]) {
916 WRITE_ONCE(si->swap_map[offset], usage);
917 inc_cluster_info_page(si, si->cluster_info, offset);
920 swap_range_alloc(si, offset, 1);
921 slots[n_ret++] = swp_entry(si->type, offset);
923 /* got enough slots or reach max slots? */
924 if ((n_ret == nr) || (offset >= si->highest_bit))
927 /* search for next available slot */
929 /* time to take a break? */
930 if (unlikely(--latency_ration < 0)) {
933 spin_unlock(&si->lock);
935 spin_lock(&si->lock);
936 latency_ration = LATENCY_LIMIT;
939 /* try to get more slots in cluster */
940 if (si->cluster_info) {
941 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
943 } else if (si->cluster_nr && !si->swap_map[++offset]) {
944 /* non-ssd case, still more slots in cluster? */
950 * Even if there's no free clusters available (fragmented),
951 * try to scan a little more quickly with lock held unless we
952 * have scanned too many slots already.
955 unsigned long scan_limit;
957 if (offset < scan_base)
958 scan_limit = scan_base;
960 scan_limit = si->highest_bit;
961 for (; offset <= scan_limit && --latency_ration > 0;
963 if (!si->swap_map[offset])
969 set_cluster_next(si, offset + 1);
970 si->flags -= SWP_SCANNING;
974 spin_unlock(&si->lock);
975 while (++offset <= READ_ONCE(si->highest_bit)) {
976 if (unlikely(--latency_ration < 0)) {
978 latency_ration = LATENCY_LIMIT;
981 if (swap_offset_available_and_locked(si, offset))
984 offset = si->lowest_bit;
985 while (offset < scan_base) {
986 if (unlikely(--latency_ration < 0)) {
988 latency_ration = LATENCY_LIMIT;
991 if (swap_offset_available_and_locked(si, offset))
995 spin_lock(&si->lock);
998 si->flags -= SWP_SCANNING;
1002 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
1005 struct swap_cluster_info *ci;
1006 unsigned long offset;
1009 * Should not even be attempting cluster allocations when huge
1010 * page swap is disabled. Warn and fail the allocation.
1012 if (!IS_ENABLED(CONFIG_THP_SWAP)) {
1017 if (cluster_list_empty(&si->free_clusters))
1020 idx = cluster_list_first(&si->free_clusters);
1021 offset = idx * SWAPFILE_CLUSTER;
1022 ci = lock_cluster(si, offset);
1023 alloc_cluster(si, idx);
1024 cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
1026 memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER);
1028 swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
1029 *slot = swp_entry(si->type, offset);
1034 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
1036 unsigned long offset = idx * SWAPFILE_CLUSTER;
1037 struct swap_cluster_info *ci;
1039 ci = lock_cluster(si, offset);
1040 memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
1041 cluster_set_count_flag(ci, 0, 0);
1042 free_cluster(si, idx);
1044 swap_range_free(si, offset, SWAPFILE_CLUSTER);
1047 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
1049 unsigned long size = swap_entry_size(entry_size);
1050 struct swap_info_struct *si, *next;
1055 /* Only single cluster request supported */
1056 WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1058 spin_lock(&swap_avail_lock);
1060 avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1061 if (avail_pgs <= 0) {
1062 spin_unlock(&swap_avail_lock);
1066 n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
1068 atomic_long_sub(n_goal * size, &nr_swap_pages);
1071 node = numa_node_id();
1072 plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1073 /* requeue si to after same-priority siblings */
1074 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1075 spin_unlock(&swap_avail_lock);
1076 spin_lock(&si->lock);
1077 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1078 spin_lock(&swap_avail_lock);
1079 if (plist_node_empty(&si->avail_lists[node])) {
1080 spin_unlock(&si->lock);
1083 WARN(!si->highest_bit,
1084 "swap_info %d in list but !highest_bit\n",
1086 WARN(!(si->flags & SWP_WRITEOK),
1087 "swap_info %d in list but !SWP_WRITEOK\n",
1089 __del_from_avail_list(si);
1090 spin_unlock(&si->lock);
1093 if (size == SWAPFILE_CLUSTER) {
1094 if (si->flags & SWP_BLKDEV)
1095 n_ret = swap_alloc_cluster(si, swp_entries);
1097 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1098 n_goal, swp_entries);
1099 spin_unlock(&si->lock);
1100 if (n_ret || size == SWAPFILE_CLUSTER)
1102 pr_debug("scan_swap_map of si %d failed to find offset\n",
1106 spin_lock(&swap_avail_lock);
1109 * if we got here, it's likely that si was almost full before,
1110 * and since scan_swap_map_slots() can drop the si->lock,
1111 * multiple callers probably all tried to get a page from the
1112 * same si and it filled up before we could get one; or, the si
1113 * filled up between us dropping swap_avail_lock and taking
1114 * si->lock. Since we dropped the swap_avail_lock, the
1115 * swap_avail_head list may have been modified; so if next is
1116 * still in the swap_avail_head list then try it, otherwise
1117 * start over if we have not gotten any slots.
1119 if (plist_node_empty(&next->avail_lists[node]))
1123 spin_unlock(&swap_avail_lock);
1127 atomic_long_add((long)(n_goal - n_ret) * size,
1133 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1135 struct swap_info_struct *p;
1136 unsigned long offset;
1140 p = swp_swap_info(entry);
1143 if (data_race(!(p->flags & SWP_USED)))
1145 offset = swp_offset(entry);
1146 if (offset >= p->max)
1148 if (data_race(!p->swap_map[swp_offset(entry)]))
1153 pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val);
1156 pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1159 pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val);
1162 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1167 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1168 struct swap_info_struct *q)
1170 struct swap_info_struct *p;
1172 p = _swap_info_get(entry);
1176 spin_unlock(&q->lock);
1178 spin_lock(&p->lock);
1183 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1184 unsigned long offset,
1185 unsigned char usage)
1187 unsigned char count;
1188 unsigned char has_cache;
1190 count = p->swap_map[offset];
1192 has_cache = count & SWAP_HAS_CACHE;
1193 count &= ~SWAP_HAS_CACHE;
1195 if (usage == SWAP_HAS_CACHE) {
1196 VM_BUG_ON(!has_cache);
1198 } else if (count == SWAP_MAP_SHMEM) {
1200 * Or we could insist on shmem.c using a special
1201 * swap_shmem_free() and free_shmem_swap_and_cache()...
1204 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1205 if (count == COUNT_CONTINUED) {
1206 if (swap_count_continued(p, offset, count))
1207 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1209 count = SWAP_MAP_MAX;
1214 usage = count | has_cache;
1216 WRITE_ONCE(p->swap_map[offset], usage);
1218 WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE);
1224 * Check whether swap entry is valid in the swap device. If so,
1225 * return pointer to swap_info_struct, and keep the swap entry valid
1226 * via preventing the swap device from being swapoff, until
1227 * put_swap_device() is called. Otherwise return NULL.
1229 * Notice that swapoff or swapoff+swapon can still happen before the
1230 * percpu_ref_tryget_live() in get_swap_device() or after the
1231 * percpu_ref_put() in put_swap_device() if there isn't any other way
1232 * to prevent swapoff, such as page lock, page table lock, etc. The
1233 * caller must be prepared for that. For example, the following
1234 * situation is possible.
1238 * ... swapoff+swapon
1239 * __read_swap_cache_async()
1240 * swapcache_prepare()
1241 * __swap_duplicate()
1243 * // verify PTE not changed
1245 * In __swap_duplicate(), the swap_map need to be checked before
1246 * changing partly because the specified swap entry may be for another
1247 * swap device which has been swapoff. And in do_swap_page(), after
1248 * the page is read from the swap device, the PTE is verified not
1249 * changed with the page table locked to check whether the swap device
1250 * has been swapoff or swapoff+swapon.
1252 struct swap_info_struct *get_swap_device(swp_entry_t entry)
1254 struct swap_info_struct *si;
1255 unsigned long offset;
1259 si = swp_swap_info(entry);
1262 if (!percpu_ref_tryget_live(&si->users))
1265 * Guarantee the si->users are checked before accessing other
1266 * fields of swap_info_struct.
1268 * Paired with the spin_unlock() after setup_swap_info() in
1269 * enable_swap_info().
1272 offset = swp_offset(entry);
1273 if (offset >= si->max)
1278 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1282 pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1283 percpu_ref_put(&si->users);
1287 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1290 struct swap_cluster_info *ci;
1291 unsigned long offset = swp_offset(entry);
1292 unsigned char usage;
1294 ci = lock_cluster_or_swap_info(p, offset);
1295 usage = __swap_entry_free_locked(p, offset, 1);
1296 unlock_cluster_or_swap_info(p, ci);
1298 free_swap_slot(entry);
1303 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1305 struct swap_cluster_info *ci;
1306 unsigned long offset = swp_offset(entry);
1307 unsigned char count;
1309 ci = lock_cluster(p, offset);
1310 count = p->swap_map[offset];
1311 VM_BUG_ON(count != SWAP_HAS_CACHE);
1312 p->swap_map[offset] = 0;
1313 dec_cluster_info_page(p, p->cluster_info, offset);
1316 mem_cgroup_uncharge_swap(entry, 1);
1317 swap_range_free(p, offset, 1);
1321 * Caller has made sure that the swap device corresponding to entry
1322 * is still around or has not been recycled.
1324 void swap_free(swp_entry_t entry)
1326 struct swap_info_struct *p;
1328 p = _swap_info_get(entry);
1330 __swap_entry_free(p, entry);
1334 * Called after dropping swapcache to decrease refcnt to swap entries.
1336 void put_swap_folio(struct folio *folio, swp_entry_t entry)
1338 unsigned long offset = swp_offset(entry);
1339 unsigned long idx = offset / SWAPFILE_CLUSTER;
1340 struct swap_cluster_info *ci;
1341 struct swap_info_struct *si;
1343 unsigned int i, free_entries = 0;
1345 int size = swap_entry_size(folio_nr_pages(folio));
1347 si = _swap_info_get(entry);
1351 ci = lock_cluster_or_swap_info(si, offset);
1352 if (size == SWAPFILE_CLUSTER) {
1353 VM_BUG_ON(!cluster_is_huge(ci));
1354 map = si->swap_map + offset;
1355 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1357 VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1358 if (val == SWAP_HAS_CACHE)
1361 cluster_clear_huge(ci);
1362 if (free_entries == SWAPFILE_CLUSTER) {
1363 unlock_cluster_or_swap_info(si, ci);
1364 spin_lock(&si->lock);
1365 mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1366 swap_free_cluster(si, idx);
1367 spin_unlock(&si->lock);
1371 for (i = 0; i < size; i++, entry.val++) {
1372 if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1373 unlock_cluster_or_swap_info(si, ci);
1374 free_swap_slot(entry);
1377 lock_cluster_or_swap_info(si, offset);
1380 unlock_cluster_or_swap_info(si, ci);
1383 #ifdef CONFIG_THP_SWAP
1384 int split_swap_cluster(swp_entry_t entry)
1386 struct swap_info_struct *si;
1387 struct swap_cluster_info *ci;
1388 unsigned long offset = swp_offset(entry);
1390 si = _swap_info_get(entry);
1393 ci = lock_cluster(si, offset);
1394 cluster_clear_huge(ci);
1400 static int swp_entry_cmp(const void *ent1, const void *ent2)
1402 const swp_entry_t *e1 = ent1, *e2 = ent2;
1404 return (int)swp_type(*e1) - (int)swp_type(*e2);
1407 void swapcache_free_entries(swp_entry_t *entries, int n)
1409 struct swap_info_struct *p, *prev;
1419 * Sort swap entries by swap device, so each lock is only taken once.
1420 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1421 * so low that it isn't necessary to optimize further.
1423 if (nr_swapfiles > 1)
1424 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1425 for (i = 0; i < n; ++i) {
1426 p = swap_info_get_cont(entries[i], prev);
1428 swap_entry_free(p, entries[i]);
1432 spin_unlock(&p->lock);
1435 int __swap_count(swp_entry_t entry)
1437 struct swap_info_struct *si;
1438 pgoff_t offset = swp_offset(entry);
1441 si = get_swap_device(entry);
1443 count = swap_count(si->swap_map[offset]);
1444 put_swap_device(si);
1450 * How many references to @entry are currently swapped out?
1451 * This does not give an exact answer when swap count is continued,
1452 * but does include the high COUNT_CONTINUED flag to allow for that.
1454 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1456 pgoff_t offset = swp_offset(entry);
1457 struct swap_cluster_info *ci;
1460 ci = lock_cluster_or_swap_info(si, offset);
1461 count = swap_count(si->swap_map[offset]);
1462 unlock_cluster_or_swap_info(si, ci);
1467 * How many references to @entry are currently swapped out?
1468 * This does not give an exact answer when swap count is continued,
1469 * but does include the high COUNT_CONTINUED flag to allow for that.
1471 int __swp_swapcount(swp_entry_t entry)
1474 struct swap_info_struct *si;
1476 si = get_swap_device(entry);
1478 count = swap_swapcount(si, entry);
1479 put_swap_device(si);
1485 * How many references to @entry are currently swapped out?
1486 * This considers COUNT_CONTINUED so it returns exact answer.
1488 int swp_swapcount(swp_entry_t entry)
1490 int count, tmp_count, n;
1491 struct swap_info_struct *p;
1492 struct swap_cluster_info *ci;
1497 p = _swap_info_get(entry);
1501 offset = swp_offset(entry);
1503 ci = lock_cluster_or_swap_info(p, offset);
1505 count = swap_count(p->swap_map[offset]);
1506 if (!(count & COUNT_CONTINUED))
1509 count &= ~COUNT_CONTINUED;
1510 n = SWAP_MAP_MAX + 1;
1512 page = vmalloc_to_page(p->swap_map + offset);
1513 offset &= ~PAGE_MASK;
1514 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1517 page = list_next_entry(page, lru);
1518 map = kmap_atomic(page);
1519 tmp_count = map[offset];
1522 count += (tmp_count & ~COUNT_CONTINUED) * n;
1523 n *= (SWAP_CONT_MAX + 1);
1524 } while (tmp_count & COUNT_CONTINUED);
1526 unlock_cluster_or_swap_info(p, ci);
1530 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1533 struct swap_cluster_info *ci;
1534 unsigned char *map = si->swap_map;
1535 unsigned long roffset = swp_offset(entry);
1536 unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1540 ci = lock_cluster_or_swap_info(si, offset);
1541 if (!ci || !cluster_is_huge(ci)) {
1542 if (swap_count(map[roffset]))
1546 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1547 if (swap_count(map[offset + i])) {
1553 unlock_cluster_or_swap_info(si, ci);
1557 static bool folio_swapped(struct folio *folio)
1559 swp_entry_t entry = folio_swap_entry(folio);
1560 struct swap_info_struct *si = _swap_info_get(entry);
1565 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!folio_test_large(folio)))
1566 return swap_swapcount(si, entry) != 0;
1568 return swap_page_trans_huge_swapped(si, entry);
1572 * folio_free_swap() - Free the swap space used for this folio.
1573 * @folio: The folio to remove.
1575 * If swap is getting full, or if there are no more mappings of this folio,
1576 * then call folio_free_swap to free its swap space.
1578 * Return: true if we were able to release the swap space.
1580 bool folio_free_swap(struct folio *folio)
1582 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1584 if (!folio_test_swapcache(folio))
1586 if (folio_test_writeback(folio))
1588 if (folio_swapped(folio))
1592 * Once hibernation has begun to create its image of memory,
1593 * there's a danger that one of the calls to folio_free_swap()
1594 * - most probably a call from __try_to_reclaim_swap() while
1595 * hibernation is allocating its own swap pages for the image,
1596 * but conceivably even a call from memory reclaim - will free
1597 * the swap from a folio which has already been recorded in the
1598 * image as a clean swapcache folio, and then reuse its swap for
1599 * another page of the image. On waking from hibernation, the
1600 * original folio might be freed under memory pressure, then
1601 * later read back in from swap, now with the wrong data.
1603 * Hibernation suspends storage while it is writing the image
1604 * to disk so check that here.
1606 if (pm_suspended_storage())
1609 delete_from_swap_cache(folio);
1610 folio_set_dirty(folio);
1615 * Free the swap entry like above, but also try to
1616 * free the page cache entry if it is the last user.
1618 int free_swap_and_cache(swp_entry_t entry)
1620 struct swap_info_struct *p;
1621 unsigned char count;
1623 if (non_swap_entry(entry))
1626 p = _swap_info_get(entry);
1628 count = __swap_entry_free(p, entry);
1629 if (count == SWAP_HAS_CACHE &&
1630 !swap_page_trans_huge_swapped(p, entry))
1631 __try_to_reclaim_swap(p, swp_offset(entry),
1632 TTRS_UNMAPPED | TTRS_FULL);
1637 #ifdef CONFIG_HIBERNATION
1639 swp_entry_t get_swap_page_of_type(int type)
1641 struct swap_info_struct *si = swap_type_to_swap_info(type);
1642 swp_entry_t entry = {0};
1647 /* This is called for allocating swap entry, not cache */
1648 spin_lock(&si->lock);
1649 if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry))
1650 atomic_long_dec(&nr_swap_pages);
1651 spin_unlock(&si->lock);
1657 * Find the swap type that corresponds to given device (if any).
1659 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1660 * from 0, in which the swap header is expected to be located.
1662 * This is needed for the suspend to disk (aka swsusp).
1664 int swap_type_of(dev_t device, sector_t offset)
1671 spin_lock(&swap_lock);
1672 for (type = 0; type < nr_swapfiles; type++) {
1673 struct swap_info_struct *sis = swap_info[type];
1675 if (!(sis->flags & SWP_WRITEOK))
1678 if (device == sis->bdev->bd_dev) {
1679 struct swap_extent *se = first_se(sis);
1681 if (se->start_block == offset) {
1682 spin_unlock(&swap_lock);
1687 spin_unlock(&swap_lock);
1691 int find_first_swap(dev_t *device)
1695 spin_lock(&swap_lock);
1696 for (type = 0; type < nr_swapfiles; type++) {
1697 struct swap_info_struct *sis = swap_info[type];
1699 if (!(sis->flags & SWP_WRITEOK))
1701 *device = sis->bdev->bd_dev;
1702 spin_unlock(&swap_lock);
1705 spin_unlock(&swap_lock);
1710 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1711 * corresponding to given index in swap_info (swap type).
1713 sector_t swapdev_block(int type, pgoff_t offset)
1715 struct swap_info_struct *si = swap_type_to_swap_info(type);
1716 struct swap_extent *se;
1718 if (!si || !(si->flags & SWP_WRITEOK))
1720 se = offset_to_swap_extent(si, offset);
1721 return se->start_block + (offset - se->start_page);
1725 * Return either the total number of swap pages of given type, or the number
1726 * of free pages of that type (depending on @free)
1728 * This is needed for software suspend
1730 unsigned int count_swap_pages(int type, int free)
1734 spin_lock(&swap_lock);
1735 if ((unsigned int)type < nr_swapfiles) {
1736 struct swap_info_struct *sis = swap_info[type];
1738 spin_lock(&sis->lock);
1739 if (sis->flags & SWP_WRITEOK) {
1742 n -= sis->inuse_pages;
1744 spin_unlock(&sis->lock);
1746 spin_unlock(&swap_lock);
1749 #endif /* CONFIG_HIBERNATION */
1751 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1753 return pte_same(pte_swp_clear_flags(pte), swp_pte);
1757 * No need to decide whether this PTE shares the swap entry with others,
1758 * just let do_wp_page work it out if a write is requested later - to
1759 * force COW, vm_page_prot omits write permission from any private vma.
1761 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1762 unsigned long addr, swp_entry_t entry, struct folio *folio)
1764 struct page *page = folio_file_page(folio, swp_offset(entry));
1765 struct page *swapcache;
1767 pte_t *pte, new_pte;
1771 page = ksm_might_need_to_copy(page, vma, addr);
1772 if (unlikely(!page))
1775 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1776 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1781 if (unlikely(!PageUptodate(page))) {
1784 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1785 pteval = swp_entry_to_pte(make_swapin_error_entry(page));
1786 set_pte_at(vma->vm_mm, addr, pte, pteval);
1792 /* See do_swap_page() */
1793 BUG_ON(!PageAnon(page) && PageMappedToDisk(page));
1794 BUG_ON(PageAnon(page) && PageAnonExclusive(page));
1796 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1797 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1799 if (page == swapcache) {
1800 rmap_t rmap_flags = RMAP_NONE;
1803 * See do_swap_page(): PageWriteback() would be problematic.
1804 * However, we do a wait_on_page_writeback() just before this
1805 * call and have the page locked.
1807 VM_BUG_ON_PAGE(PageWriteback(page), page);
1808 if (pte_swp_exclusive(*pte))
1809 rmap_flags |= RMAP_EXCLUSIVE;
1811 page_add_anon_rmap(page, vma, addr, rmap_flags);
1812 } else { /* ksm created a completely new copy */
1813 page_add_new_anon_rmap(page, vma, addr);
1814 lru_cache_add_inactive_or_unevictable(page, vma);
1816 new_pte = pte_mkold(mk_pte(page, vma->vm_page_prot));
1817 if (pte_swp_soft_dirty(*pte))
1818 new_pte = pte_mksoft_dirty(new_pte);
1819 if (pte_swp_uffd_wp(*pte))
1820 new_pte = pte_mkuffd_wp(new_pte);
1821 set_pte_at(vma->vm_mm, addr, pte, new_pte);
1824 pte_unmap_unlock(pte, ptl);
1825 if (page != swapcache) {
1832 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1833 unsigned long addr, unsigned long end,
1838 struct swap_info_struct *si;
1840 volatile unsigned char *swap_map;
1842 si = swap_info[type];
1843 pte = pte_offset_map(pmd, addr);
1845 struct folio *folio;
1846 unsigned long offset;
1848 if (!is_swap_pte(*pte))
1851 entry = pte_to_swp_entry(*pte);
1852 if (swp_type(entry) != type)
1855 offset = swp_offset(entry);
1857 swap_map = &si->swap_map[offset];
1858 folio = swap_cache_get_folio(entry, vma, addr);
1861 struct vm_fault vmf = {
1864 .real_address = addr,
1868 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1871 folio = page_folio(page);
1874 if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1880 folio_wait_writeback(folio);
1881 ret = unuse_pte(vma, pmd, addr, entry, folio);
1883 folio_unlock(folio);
1888 folio_free_swap(folio);
1889 folio_unlock(folio);
1892 pte = pte_offset_map(pmd, addr);
1893 } while (pte++, addr += PAGE_SIZE, addr != end);
1901 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
1902 unsigned long addr, unsigned long end,
1909 pmd = pmd_offset(pud, addr);
1912 next = pmd_addr_end(addr, end);
1913 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
1915 ret = unuse_pte_range(vma, pmd, addr, next, type);
1918 } while (pmd++, addr = next, addr != end);
1922 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
1923 unsigned long addr, unsigned long end,
1930 pud = pud_offset(p4d, addr);
1932 next = pud_addr_end(addr, end);
1933 if (pud_none_or_clear_bad(pud))
1935 ret = unuse_pmd_range(vma, pud, addr, next, type);
1938 } while (pud++, addr = next, addr != end);
1942 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
1943 unsigned long addr, unsigned long end,
1950 p4d = p4d_offset(pgd, addr);
1952 next = p4d_addr_end(addr, end);
1953 if (p4d_none_or_clear_bad(p4d))
1955 ret = unuse_pud_range(vma, p4d, addr, next, type);
1958 } while (p4d++, addr = next, addr != end);
1962 static int unuse_vma(struct vm_area_struct *vma, unsigned int type)
1965 unsigned long addr, end, next;
1968 addr = vma->vm_start;
1971 pgd = pgd_offset(vma->vm_mm, addr);
1973 next = pgd_addr_end(addr, end);
1974 if (pgd_none_or_clear_bad(pgd))
1976 ret = unuse_p4d_range(vma, pgd, addr, next, type);
1979 } while (pgd++, addr = next, addr != end);
1983 static int unuse_mm(struct mm_struct *mm, unsigned int type)
1985 struct vm_area_struct *vma;
1987 VMA_ITERATOR(vmi, mm, 0);
1990 for_each_vma(vmi, vma) {
1991 if (vma->anon_vma) {
1992 ret = unuse_vma(vma, type);
1999 mmap_read_unlock(mm);
2004 * Scan swap_map from current position to next entry still in use.
2005 * Return 0 if there are no inuse entries after prev till end of
2008 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2012 unsigned char count;
2015 * No need for swap_lock here: we're just looking
2016 * for whether an entry is in use, not modifying it; false
2017 * hits are okay, and sys_swapoff() has already prevented new
2018 * allocations from this area (while holding swap_lock).
2020 for (i = prev + 1; i < si->max; i++) {
2021 count = READ_ONCE(si->swap_map[i]);
2022 if (count && swap_count(count) != SWAP_MAP_BAD)
2024 if ((i % LATENCY_LIMIT) == 0)
2034 static int try_to_unuse(unsigned int type)
2036 struct mm_struct *prev_mm;
2037 struct mm_struct *mm;
2038 struct list_head *p;
2040 struct swap_info_struct *si = swap_info[type];
2041 struct folio *folio;
2045 if (!READ_ONCE(si->inuse_pages))
2049 retval = shmem_unuse(type);
2056 spin_lock(&mmlist_lock);
2057 p = &init_mm.mmlist;
2058 while (READ_ONCE(si->inuse_pages) &&
2059 !signal_pending(current) &&
2060 (p = p->next) != &init_mm.mmlist) {
2062 mm = list_entry(p, struct mm_struct, mmlist);
2063 if (!mmget_not_zero(mm))
2065 spin_unlock(&mmlist_lock);
2068 retval = unuse_mm(mm, type);
2075 * Make sure that we aren't completely killing
2076 * interactive performance.
2079 spin_lock(&mmlist_lock);
2081 spin_unlock(&mmlist_lock);
2086 while (READ_ONCE(si->inuse_pages) &&
2087 !signal_pending(current) &&
2088 (i = find_next_to_unuse(si, i)) != 0) {
2090 entry = swp_entry(type, i);
2091 folio = filemap_get_folio(swap_address_space(entry), i);
2096 * It is conceivable that a racing task removed this folio from
2097 * swap cache just before we acquired the page lock. The folio
2098 * might even be back in swap cache on another swap area. But
2099 * that is okay, folio_free_swap() only removes stale folios.
2102 folio_wait_writeback(folio);
2103 folio_free_swap(folio);
2104 folio_unlock(folio);
2109 * Lets check again to see if there are still swap entries in the map.
2110 * If yes, we would need to do retry the unuse logic again.
2111 * Under global memory pressure, swap entries can be reinserted back
2112 * into process space after the mmlist loop above passes over them.
2114 * Limit the number of retries? No: when mmget_not_zero()
2115 * above fails, that mm is likely to be freeing swap from
2116 * exit_mmap(), which proceeds at its own independent pace;
2117 * and even shmem_writepage() could have been preempted after
2118 * folio_alloc_swap(), temporarily hiding that swap. It's easy
2119 * and robust (though cpu-intensive) just to keep retrying.
2121 if (READ_ONCE(si->inuse_pages)) {
2122 if (!signal_pending(current))
2131 * After a successful try_to_unuse, if no swap is now in use, we know
2132 * we can empty the mmlist. swap_lock must be held on entry and exit.
2133 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2134 * added to the mmlist just after page_duplicate - before would be racy.
2136 static void drain_mmlist(void)
2138 struct list_head *p, *next;
2141 for (type = 0; type < nr_swapfiles; type++)
2142 if (swap_info[type]->inuse_pages)
2144 spin_lock(&mmlist_lock);
2145 list_for_each_safe(p, next, &init_mm.mmlist)
2147 spin_unlock(&mmlist_lock);
2151 * Free all of a swapdev's extent information
2153 static void destroy_swap_extents(struct swap_info_struct *sis)
2155 while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2156 struct rb_node *rb = sis->swap_extent_root.rb_node;
2157 struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2159 rb_erase(rb, &sis->swap_extent_root);
2163 if (sis->flags & SWP_ACTIVATED) {
2164 struct file *swap_file = sis->swap_file;
2165 struct address_space *mapping = swap_file->f_mapping;
2167 sis->flags &= ~SWP_ACTIVATED;
2168 if (mapping->a_ops->swap_deactivate)
2169 mapping->a_ops->swap_deactivate(swap_file);
2174 * Add a block range (and the corresponding page range) into this swapdev's
2177 * This function rather assumes that it is called in ascending page order.
2180 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2181 unsigned long nr_pages, sector_t start_block)
2183 struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2184 struct swap_extent *se;
2185 struct swap_extent *new_se;
2188 * place the new node at the right most since the
2189 * function is called in ascending page order.
2193 link = &parent->rb_right;
2197 se = rb_entry(parent, struct swap_extent, rb_node);
2198 BUG_ON(se->start_page + se->nr_pages != start_page);
2199 if (se->start_block + se->nr_pages == start_block) {
2201 se->nr_pages += nr_pages;
2206 /* No merge, insert a new extent. */
2207 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2210 new_se->start_page = start_page;
2211 new_se->nr_pages = nr_pages;
2212 new_se->start_block = start_block;
2214 rb_link_node(&new_se->rb_node, parent, link);
2215 rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2218 EXPORT_SYMBOL_GPL(add_swap_extent);
2221 * A `swap extent' is a simple thing which maps a contiguous range of pages
2222 * onto a contiguous range of disk blocks. A rbtree of swap extents is
2223 * built at swapon time and is then used at swap_writepage/swap_readpage
2224 * time for locating where on disk a page belongs.
2226 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2227 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2228 * swap files identically.
2230 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2231 * extent rbtree operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2232 * swapfiles are handled *identically* after swapon time.
2234 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2235 * and will parse them into a rbtree, in PAGE_SIZE chunks. If some stray
2236 * blocks are found which do not fall within the PAGE_SIZE alignment
2237 * requirements, they are simply tossed out - we will never use those blocks
2240 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2241 * prevents users from writing to the swap device, which will corrupt memory.
2243 * The amount of disk space which a single swap extent represents varies.
2244 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2245 * extents in the rbtree. - akpm.
2247 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2249 struct file *swap_file = sis->swap_file;
2250 struct address_space *mapping = swap_file->f_mapping;
2251 struct inode *inode = mapping->host;
2254 if (S_ISBLK(inode->i_mode)) {
2255 ret = add_swap_extent(sis, 0, sis->max, 0);
2260 if (mapping->a_ops->swap_activate) {
2261 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2264 sis->flags |= SWP_ACTIVATED;
2265 if ((sis->flags & SWP_FS_OPS) &&
2266 sio_pool_init() != 0) {
2267 destroy_swap_extents(sis);
2273 return generic_swapfile_activate(sis, swap_file, span);
2276 static int swap_node(struct swap_info_struct *p)
2278 struct block_device *bdev;
2283 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2285 return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2288 static void setup_swap_info(struct swap_info_struct *p, int prio,
2289 unsigned char *swap_map,
2290 struct swap_cluster_info *cluster_info)
2297 p->prio = --least_priority;
2299 * the plist prio is negated because plist ordering is
2300 * low-to-high, while swap ordering is high-to-low
2302 p->list.prio = -p->prio;
2305 p->avail_lists[i].prio = -p->prio;
2307 if (swap_node(p) == i)
2308 p->avail_lists[i].prio = 1;
2310 p->avail_lists[i].prio = -p->prio;
2313 p->swap_map = swap_map;
2314 p->cluster_info = cluster_info;
2317 static void _enable_swap_info(struct swap_info_struct *p)
2319 p->flags |= SWP_WRITEOK;
2320 atomic_long_add(p->pages, &nr_swap_pages);
2321 total_swap_pages += p->pages;
2323 assert_spin_locked(&swap_lock);
2325 * both lists are plists, and thus priority ordered.
2326 * swap_active_head needs to be priority ordered for swapoff(),
2327 * which on removal of any swap_info_struct with an auto-assigned
2328 * (i.e. negative) priority increments the auto-assigned priority
2329 * of any lower-priority swap_info_structs.
2330 * swap_avail_head needs to be priority ordered for folio_alloc_swap(),
2331 * which allocates swap pages from the highest available priority
2334 plist_add(&p->list, &swap_active_head);
2335 add_to_avail_list(p);
2338 static void enable_swap_info(struct swap_info_struct *p, int prio,
2339 unsigned char *swap_map,
2340 struct swap_cluster_info *cluster_info,
2341 unsigned long *frontswap_map)
2343 if (IS_ENABLED(CONFIG_FRONTSWAP))
2344 frontswap_init(p->type, frontswap_map);
2345 spin_lock(&swap_lock);
2346 spin_lock(&p->lock);
2347 setup_swap_info(p, prio, swap_map, cluster_info);
2348 spin_unlock(&p->lock);
2349 spin_unlock(&swap_lock);
2351 * Finished initializing swap device, now it's safe to reference it.
2353 percpu_ref_resurrect(&p->users);
2354 spin_lock(&swap_lock);
2355 spin_lock(&p->lock);
2356 _enable_swap_info(p);
2357 spin_unlock(&p->lock);
2358 spin_unlock(&swap_lock);
2361 static void reinsert_swap_info(struct swap_info_struct *p)
2363 spin_lock(&swap_lock);
2364 spin_lock(&p->lock);
2365 setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2366 _enable_swap_info(p);
2367 spin_unlock(&p->lock);
2368 spin_unlock(&swap_lock);
2371 bool has_usable_swap(void)
2375 spin_lock(&swap_lock);
2376 if (plist_head_empty(&swap_active_head))
2378 spin_unlock(&swap_lock);
2382 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2384 struct swap_info_struct *p = NULL;
2385 unsigned char *swap_map;
2386 struct swap_cluster_info *cluster_info;
2387 unsigned long *frontswap_map;
2388 struct file *swap_file, *victim;
2389 struct address_space *mapping;
2390 struct inode *inode;
2391 struct filename *pathname;
2393 unsigned int old_block_size;
2395 if (!capable(CAP_SYS_ADMIN))
2398 BUG_ON(!current->mm);
2400 pathname = getname(specialfile);
2401 if (IS_ERR(pathname))
2402 return PTR_ERR(pathname);
2404 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2405 err = PTR_ERR(victim);
2409 mapping = victim->f_mapping;
2410 spin_lock(&swap_lock);
2411 plist_for_each_entry(p, &swap_active_head, list) {
2412 if (p->flags & SWP_WRITEOK) {
2413 if (p->swap_file->f_mapping == mapping) {
2421 spin_unlock(&swap_lock);
2424 if (!security_vm_enough_memory_mm(current->mm, p->pages))
2425 vm_unacct_memory(p->pages);
2428 spin_unlock(&swap_lock);
2431 del_from_avail_list(p);
2432 spin_lock(&p->lock);
2434 struct swap_info_struct *si = p;
2437 plist_for_each_entry_continue(si, &swap_active_head, list) {
2440 for_each_node(nid) {
2441 if (si->avail_lists[nid].prio != 1)
2442 si->avail_lists[nid].prio--;
2447 plist_del(&p->list, &swap_active_head);
2448 atomic_long_sub(p->pages, &nr_swap_pages);
2449 total_swap_pages -= p->pages;
2450 p->flags &= ~SWP_WRITEOK;
2451 spin_unlock(&p->lock);
2452 spin_unlock(&swap_lock);
2454 disable_swap_slots_cache_lock();
2456 set_current_oom_origin();
2457 err = try_to_unuse(p->type);
2458 clear_current_oom_origin();
2461 /* re-insert swap space back into swap_list */
2462 reinsert_swap_info(p);
2463 reenable_swap_slots_cache_unlock();
2467 reenable_swap_slots_cache_unlock();
2470 * Wait for swap operations protected by get/put_swap_device()
2473 * We need synchronize_rcu() here to protect the accessing to
2474 * the swap cache data structure.
2476 percpu_ref_kill(&p->users);
2478 wait_for_completion(&p->comp);
2480 flush_work(&p->discard_work);
2482 destroy_swap_extents(p);
2483 if (p->flags & SWP_CONTINUED)
2484 free_swap_count_continuations(p);
2486 if (!p->bdev || !bdev_nonrot(p->bdev))
2487 atomic_dec(&nr_rotate_swap);
2489 mutex_lock(&swapon_mutex);
2490 spin_lock(&swap_lock);
2491 spin_lock(&p->lock);
2494 /* wait for anyone still in scan_swap_map_slots */
2495 p->highest_bit = 0; /* cuts scans short */
2496 while (p->flags >= SWP_SCANNING) {
2497 spin_unlock(&p->lock);
2498 spin_unlock(&swap_lock);
2499 schedule_timeout_uninterruptible(1);
2500 spin_lock(&swap_lock);
2501 spin_lock(&p->lock);
2504 swap_file = p->swap_file;
2505 old_block_size = p->old_block_size;
2506 p->swap_file = NULL;
2508 swap_map = p->swap_map;
2510 cluster_info = p->cluster_info;
2511 p->cluster_info = NULL;
2512 frontswap_map = frontswap_map_get(p);
2513 spin_unlock(&p->lock);
2514 spin_unlock(&swap_lock);
2515 arch_swap_invalidate_area(p->type);
2516 frontswap_invalidate_area(p->type);
2517 frontswap_map_set(p, NULL);
2518 mutex_unlock(&swapon_mutex);
2519 free_percpu(p->percpu_cluster);
2520 p->percpu_cluster = NULL;
2521 free_percpu(p->cluster_next_cpu);
2522 p->cluster_next_cpu = NULL;
2524 kvfree(cluster_info);
2525 kvfree(frontswap_map);
2526 /* Destroy swap account information */
2527 swap_cgroup_swapoff(p->type);
2528 exit_swap_address_space(p->type);
2530 inode = mapping->host;
2531 if (S_ISBLK(inode->i_mode)) {
2532 struct block_device *bdev = I_BDEV(inode);
2534 set_blocksize(bdev, old_block_size);
2535 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2539 inode->i_flags &= ~S_SWAPFILE;
2540 inode_unlock(inode);
2541 filp_close(swap_file, NULL);
2544 * Clear the SWP_USED flag after all resources are freed so that swapon
2545 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2546 * not hold p->lock after we cleared its SWP_WRITEOK.
2548 spin_lock(&swap_lock);
2550 spin_unlock(&swap_lock);
2553 atomic_inc(&proc_poll_event);
2554 wake_up_interruptible(&proc_poll_wait);
2557 filp_close(victim, NULL);
2563 #ifdef CONFIG_PROC_FS
2564 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2566 struct seq_file *seq = file->private_data;
2568 poll_wait(file, &proc_poll_wait, wait);
2570 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2571 seq->poll_event = atomic_read(&proc_poll_event);
2572 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2575 return EPOLLIN | EPOLLRDNORM;
2579 static void *swap_start(struct seq_file *swap, loff_t *pos)
2581 struct swap_info_struct *si;
2585 mutex_lock(&swapon_mutex);
2588 return SEQ_START_TOKEN;
2590 for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2591 if (!(si->flags & SWP_USED) || !si->swap_map)
2600 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2602 struct swap_info_struct *si = v;
2605 if (v == SEQ_START_TOKEN)
2608 type = si->type + 1;
2611 for (; (si = swap_type_to_swap_info(type)); type++) {
2612 if (!(si->flags & SWP_USED) || !si->swap_map)
2620 static void swap_stop(struct seq_file *swap, void *v)
2622 mutex_unlock(&swapon_mutex);
2625 static int swap_show(struct seq_file *swap, void *v)
2627 struct swap_info_struct *si = v;
2630 unsigned long bytes, inuse;
2632 if (si == SEQ_START_TOKEN) {
2633 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2637 bytes = si->pages << (PAGE_SHIFT - 10);
2638 inuse = READ_ONCE(si->inuse_pages) << (PAGE_SHIFT - 10);
2640 file = si->swap_file;
2641 len = seq_file_path(swap, file, " \t\n\\");
2642 seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n",
2643 len < 40 ? 40 - len : 1, " ",
2644 S_ISBLK(file_inode(file)->i_mode) ?
2645 "partition" : "file\t",
2646 bytes, bytes < 10000000 ? "\t" : "",
2647 inuse, inuse < 10000000 ? "\t" : "",
2652 static const struct seq_operations swaps_op = {
2653 .start = swap_start,
2659 static int swaps_open(struct inode *inode, struct file *file)
2661 struct seq_file *seq;
2664 ret = seq_open(file, &swaps_op);
2668 seq = file->private_data;
2669 seq->poll_event = atomic_read(&proc_poll_event);
2673 static const struct proc_ops swaps_proc_ops = {
2674 .proc_flags = PROC_ENTRY_PERMANENT,
2675 .proc_open = swaps_open,
2676 .proc_read = seq_read,
2677 .proc_lseek = seq_lseek,
2678 .proc_release = seq_release,
2679 .proc_poll = swaps_poll,
2682 static int __init procswaps_init(void)
2684 proc_create("swaps", 0, NULL, &swaps_proc_ops);
2687 __initcall(procswaps_init);
2688 #endif /* CONFIG_PROC_FS */
2690 #ifdef MAX_SWAPFILES_CHECK
2691 static int __init max_swapfiles_check(void)
2693 MAX_SWAPFILES_CHECK();
2696 late_initcall(max_swapfiles_check);
2699 static struct swap_info_struct *alloc_swap_info(void)
2701 struct swap_info_struct *p;
2702 struct swap_info_struct *defer = NULL;
2706 p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2708 return ERR_PTR(-ENOMEM);
2710 if (percpu_ref_init(&p->users, swap_users_ref_free,
2711 PERCPU_REF_INIT_DEAD, GFP_KERNEL)) {
2713 return ERR_PTR(-ENOMEM);
2716 spin_lock(&swap_lock);
2717 for (type = 0; type < nr_swapfiles; type++) {
2718 if (!(swap_info[type]->flags & SWP_USED))
2721 if (type >= MAX_SWAPFILES) {
2722 spin_unlock(&swap_lock);
2723 percpu_ref_exit(&p->users);
2725 return ERR_PTR(-EPERM);
2727 if (type >= nr_swapfiles) {
2730 * Publish the swap_info_struct after initializing it.
2731 * Note that kvzalloc() above zeroes all its fields.
2733 smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */
2737 p = swap_info[type];
2739 * Do not memset this entry: a racing procfs swap_next()
2740 * would be relying on p->type to remain valid.
2743 p->swap_extent_root = RB_ROOT;
2744 plist_node_init(&p->list, 0);
2746 plist_node_init(&p->avail_lists[i], 0);
2747 p->flags = SWP_USED;
2748 spin_unlock(&swap_lock);
2750 percpu_ref_exit(&defer->users);
2753 spin_lock_init(&p->lock);
2754 spin_lock_init(&p->cont_lock);
2755 init_completion(&p->comp);
2760 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2764 if (S_ISBLK(inode->i_mode)) {
2765 p->bdev = blkdev_get_by_dev(inode->i_rdev,
2766 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2767 if (IS_ERR(p->bdev)) {
2768 error = PTR_ERR(p->bdev);
2772 p->old_block_size = block_size(p->bdev);
2773 error = set_blocksize(p->bdev, PAGE_SIZE);
2777 * Zoned block devices contain zones that have a sequential
2778 * write only restriction. Hence zoned block devices are not
2779 * suitable for swapping. Disallow them here.
2781 if (bdev_is_zoned(p->bdev))
2783 p->flags |= SWP_BLKDEV;
2784 } else if (S_ISREG(inode->i_mode)) {
2785 p->bdev = inode->i_sb->s_bdev;
2793 * Find out how many pages are allowed for a single swap device. There
2794 * are two limiting factors:
2795 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2796 * 2) the number of bits in the swap pte, as defined by the different
2799 * In order to find the largest possible bit mask, a swap entry with
2800 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2801 * decoded to a swp_entry_t again, and finally the swap offset is
2804 * This will mask all the bits from the initial ~0UL mask that can't
2805 * be encoded in either the swp_entry_t or the architecture definition
2808 unsigned long generic_max_swapfile_size(void)
2810 return swp_offset(pte_to_swp_entry(
2811 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2814 /* Can be overridden by an architecture for additional checks. */
2815 __weak unsigned long arch_max_swapfile_size(void)
2817 return generic_max_swapfile_size();
2820 static unsigned long read_swap_header(struct swap_info_struct *p,
2821 union swap_header *swap_header,
2822 struct inode *inode)
2825 unsigned long maxpages;
2826 unsigned long swapfilepages;
2827 unsigned long last_page;
2829 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2830 pr_err("Unable to find swap-space signature\n");
2834 /* swap partition endianness hack... */
2835 if (swab32(swap_header->info.version) == 1) {
2836 swab32s(&swap_header->info.version);
2837 swab32s(&swap_header->info.last_page);
2838 swab32s(&swap_header->info.nr_badpages);
2839 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2841 for (i = 0; i < swap_header->info.nr_badpages; i++)
2842 swab32s(&swap_header->info.badpages[i]);
2844 /* Check the swap header's sub-version */
2845 if (swap_header->info.version != 1) {
2846 pr_warn("Unable to handle swap header version %d\n",
2847 swap_header->info.version);
2852 p->cluster_next = 1;
2855 maxpages = swapfile_maximum_size;
2856 last_page = swap_header->info.last_page;
2858 pr_warn("Empty swap-file\n");
2861 if (last_page > maxpages) {
2862 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
2863 maxpages << (PAGE_SHIFT - 10),
2864 last_page << (PAGE_SHIFT - 10));
2866 if (maxpages > last_page) {
2867 maxpages = last_page + 1;
2868 /* p->max is an unsigned int: don't overflow it */
2869 if ((unsigned int)maxpages == 0)
2870 maxpages = UINT_MAX;
2872 p->highest_bit = maxpages - 1;
2876 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
2877 if (swapfilepages && maxpages > swapfilepages) {
2878 pr_warn("Swap area shorter than signature indicates\n");
2881 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
2883 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2889 #define SWAP_CLUSTER_INFO_COLS \
2890 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
2891 #define SWAP_CLUSTER_SPACE_COLS \
2892 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
2893 #define SWAP_CLUSTER_COLS \
2894 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
2896 static int setup_swap_map_and_extents(struct swap_info_struct *p,
2897 union swap_header *swap_header,
2898 unsigned char *swap_map,
2899 struct swap_cluster_info *cluster_info,
2900 unsigned long maxpages,
2904 unsigned int nr_good_pages;
2906 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
2907 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
2908 unsigned long i, idx;
2910 nr_good_pages = maxpages - 1; /* omit header page */
2912 cluster_list_init(&p->free_clusters);
2913 cluster_list_init(&p->discard_clusters);
2915 for (i = 0; i < swap_header->info.nr_badpages; i++) {
2916 unsigned int page_nr = swap_header->info.badpages[i];
2917 if (page_nr == 0 || page_nr > swap_header->info.last_page)
2919 if (page_nr < maxpages) {
2920 swap_map[page_nr] = SWAP_MAP_BAD;
2923 * Haven't marked the cluster free yet, no list
2924 * operation involved
2926 inc_cluster_info_page(p, cluster_info, page_nr);
2930 /* Haven't marked the cluster free yet, no list operation involved */
2931 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
2932 inc_cluster_info_page(p, cluster_info, i);
2934 if (nr_good_pages) {
2935 swap_map[0] = SWAP_MAP_BAD;
2937 * Not mark the cluster free yet, no list
2938 * operation involved
2940 inc_cluster_info_page(p, cluster_info, 0);
2942 p->pages = nr_good_pages;
2943 nr_extents = setup_swap_extents(p, span);
2946 nr_good_pages = p->pages;
2948 if (!nr_good_pages) {
2949 pr_warn("Empty swap-file\n");
2958 * Reduce false cache line sharing between cluster_info and
2959 * sharing same address space.
2961 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
2962 j = (k + col) % SWAP_CLUSTER_COLS;
2963 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
2964 idx = i * SWAP_CLUSTER_COLS + j;
2965 if (idx >= nr_clusters)
2967 if (cluster_count(&cluster_info[idx]))
2969 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
2970 cluster_list_add_tail(&p->free_clusters, cluster_info,
2977 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
2979 struct swap_info_struct *p;
2980 struct filename *name;
2981 struct file *swap_file = NULL;
2982 struct address_space *mapping;
2983 struct dentry *dentry;
2986 union swap_header *swap_header;
2989 unsigned long maxpages;
2990 unsigned char *swap_map = NULL;
2991 struct swap_cluster_info *cluster_info = NULL;
2992 unsigned long *frontswap_map = NULL;
2993 struct page *page = NULL;
2994 struct inode *inode = NULL;
2995 bool inced_nr_rotate_swap = false;
2997 if (swap_flags & ~SWAP_FLAGS_VALID)
3000 if (!capable(CAP_SYS_ADMIN))
3003 if (!swap_avail_heads)
3006 p = alloc_swap_info();
3010 INIT_WORK(&p->discard_work, swap_discard_work);
3012 name = getname(specialfile);
3014 error = PTR_ERR(name);
3018 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3019 if (IS_ERR(swap_file)) {
3020 error = PTR_ERR(swap_file);
3025 p->swap_file = swap_file;
3026 mapping = swap_file->f_mapping;
3027 dentry = swap_file->f_path.dentry;
3028 inode = mapping->host;
3030 error = claim_swapfile(p, inode);
3031 if (unlikely(error))
3035 if (d_unlinked(dentry) || cant_mount(dentry)) {
3037 goto bad_swap_unlock_inode;
3039 if (IS_SWAPFILE(inode)) {
3041 goto bad_swap_unlock_inode;
3045 * Read the swap header.
3047 if (!mapping->a_ops->read_folio) {
3049 goto bad_swap_unlock_inode;
3051 page = read_mapping_page(mapping, 0, swap_file);
3053 error = PTR_ERR(page);
3054 goto bad_swap_unlock_inode;
3056 swap_header = kmap(page);
3058 maxpages = read_swap_header(p, swap_header, inode);
3059 if (unlikely(!maxpages)) {
3061 goto bad_swap_unlock_inode;
3064 /* OK, set up the swap map and apply the bad block list */
3065 swap_map = vzalloc(maxpages);
3068 goto bad_swap_unlock_inode;
3071 if (p->bdev && bdev_stable_writes(p->bdev))
3072 p->flags |= SWP_STABLE_WRITES;
3074 if (p->bdev && p->bdev->bd_disk->fops->rw_page)
3075 p->flags |= SWP_SYNCHRONOUS_IO;
3077 if (p->bdev && bdev_nonrot(p->bdev)) {
3079 unsigned long ci, nr_cluster;
3081 p->flags |= SWP_SOLIDSTATE;
3082 p->cluster_next_cpu = alloc_percpu(unsigned int);
3083 if (!p->cluster_next_cpu) {
3085 goto bad_swap_unlock_inode;
3088 * select a random position to start with to help wear leveling
3091 for_each_possible_cpu(cpu) {
3092 per_cpu(*p->cluster_next_cpu, cpu) =
3093 1 + prandom_u32_max(p->highest_bit);
3095 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3097 cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3099 if (!cluster_info) {
3101 goto bad_swap_unlock_inode;
3104 for (ci = 0; ci < nr_cluster; ci++)
3105 spin_lock_init(&((cluster_info + ci)->lock));
3107 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3108 if (!p->percpu_cluster) {
3110 goto bad_swap_unlock_inode;
3112 for_each_possible_cpu(cpu) {
3113 struct percpu_cluster *cluster;
3114 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3115 cluster_set_null(&cluster->index);
3118 atomic_inc(&nr_rotate_swap);
3119 inced_nr_rotate_swap = true;
3122 error = swap_cgroup_swapon(p->type, maxpages);
3124 goto bad_swap_unlock_inode;
3126 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3127 cluster_info, maxpages, &span);
3128 if (unlikely(nr_extents < 0)) {
3130 goto bad_swap_unlock_inode;
3132 /* frontswap enabled? set up bit-per-page map for frontswap */
3133 if (IS_ENABLED(CONFIG_FRONTSWAP))
3134 frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3138 if ((swap_flags & SWAP_FLAG_DISCARD) &&
3139 p->bdev && bdev_max_discard_sectors(p->bdev)) {
3141 * When discard is enabled for swap with no particular
3142 * policy flagged, we set all swap discard flags here in
3143 * order to sustain backward compatibility with older
3144 * swapon(8) releases.
3146 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3150 * By flagging sys_swapon, a sysadmin can tell us to
3151 * either do single-time area discards only, or to just
3152 * perform discards for released swap page-clusters.
3153 * Now it's time to adjust the p->flags accordingly.
3155 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3156 p->flags &= ~SWP_PAGE_DISCARD;
3157 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3158 p->flags &= ~SWP_AREA_DISCARD;
3160 /* issue a swapon-time discard if it's still required */
3161 if (p->flags & SWP_AREA_DISCARD) {
3162 int err = discard_swap(p);
3164 pr_err("swapon: discard_swap(%p): %d\n",
3169 error = init_swap_address_space(p->type, maxpages);
3171 goto bad_swap_unlock_inode;
3174 * Flush any pending IO and dirty mappings before we start using this
3177 inode->i_flags |= S_SWAPFILE;
3178 error = inode_drain_writes(inode);
3180 inode->i_flags &= ~S_SWAPFILE;
3181 goto free_swap_address_space;
3184 mutex_lock(&swapon_mutex);
3186 if (swap_flags & SWAP_FLAG_PREFER)
3188 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3189 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3191 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3192 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3193 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3194 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3195 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3196 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3197 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3198 (frontswap_map) ? "FS" : "");
3200 mutex_unlock(&swapon_mutex);
3201 atomic_inc(&proc_poll_event);
3202 wake_up_interruptible(&proc_poll_wait);
3206 free_swap_address_space:
3207 exit_swap_address_space(p->type);
3208 bad_swap_unlock_inode:
3209 inode_unlock(inode);
3211 free_percpu(p->percpu_cluster);
3212 p->percpu_cluster = NULL;
3213 free_percpu(p->cluster_next_cpu);
3214 p->cluster_next_cpu = NULL;
3215 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3216 set_blocksize(p->bdev, p->old_block_size);
3217 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3220 destroy_swap_extents(p);
3221 swap_cgroup_swapoff(p->type);
3222 spin_lock(&swap_lock);
3223 p->swap_file = NULL;
3225 spin_unlock(&swap_lock);
3227 kvfree(cluster_info);
3228 kvfree(frontswap_map);
3229 if (inced_nr_rotate_swap)
3230 atomic_dec(&nr_rotate_swap);
3232 filp_close(swap_file, NULL);
3234 if (page && !IS_ERR(page)) {
3241 inode_unlock(inode);
3243 enable_swap_slots_cache();
3247 void si_swapinfo(struct sysinfo *val)
3250 unsigned long nr_to_be_unused = 0;
3252 spin_lock(&swap_lock);
3253 for (type = 0; type < nr_swapfiles; type++) {
3254 struct swap_info_struct *si = swap_info[type];
3256 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3257 nr_to_be_unused += READ_ONCE(si->inuse_pages);
3259 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3260 val->totalswap = total_swap_pages + nr_to_be_unused;
3261 spin_unlock(&swap_lock);
3265 * Verify that a swap entry is valid and increment its swap map count.
3267 * Returns error code in following case.
3269 * - swp_entry is invalid -> EINVAL
3270 * - swp_entry is migration entry -> EINVAL
3271 * - swap-cache reference is requested but there is already one. -> EEXIST
3272 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3273 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3275 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3277 struct swap_info_struct *p;
3278 struct swap_cluster_info *ci;
3279 unsigned long offset;
3280 unsigned char count;
3281 unsigned char has_cache;
3284 p = get_swap_device(entry);
3288 offset = swp_offset(entry);
3289 ci = lock_cluster_or_swap_info(p, offset);
3291 count = p->swap_map[offset];
3294 * swapin_readahead() doesn't check if a swap entry is valid, so the
3295 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3297 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3302 has_cache = count & SWAP_HAS_CACHE;
3303 count &= ~SWAP_HAS_CACHE;
3306 if (usage == SWAP_HAS_CACHE) {
3308 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3309 if (!has_cache && count)
3310 has_cache = SWAP_HAS_CACHE;
3311 else if (has_cache) /* someone else added cache */
3313 else /* no users remaining */
3316 } else if (count || has_cache) {
3318 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3320 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3322 else if (swap_count_continued(p, offset, count))
3323 count = COUNT_CONTINUED;
3327 err = -ENOENT; /* unused swap entry */
3329 WRITE_ONCE(p->swap_map[offset], count | has_cache);
3332 unlock_cluster_or_swap_info(p, ci);
3338 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3339 * (in which case its reference count is never incremented).
3341 void swap_shmem_alloc(swp_entry_t entry)
3343 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3347 * Increase reference count of swap entry by 1.
3348 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3349 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3350 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3351 * might occur if a page table entry has got corrupted.
3353 int swap_duplicate(swp_entry_t entry)
3357 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3358 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3363 * @entry: swap entry for which we allocate swap cache.
3365 * Called when allocating swap cache for existing swap entry,
3366 * This can return error codes. Returns 0 at success.
3367 * -EEXIST means there is a swap cache.
3368 * Note: return code is different from swap_duplicate().
3370 int swapcache_prepare(swp_entry_t entry)
3372 return __swap_duplicate(entry, SWAP_HAS_CACHE);
3375 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3377 return swap_type_to_swap_info(swp_type(entry));
3380 struct swap_info_struct *page_swap_info(struct page *page)
3382 swp_entry_t entry = { .val = page_private(page) };
3383 return swp_swap_info(entry);
3387 * out-of-line methods to avoid include hell.
3389 struct address_space *swapcache_mapping(struct folio *folio)
3391 return page_swap_info(&folio->page)->swap_file->f_mapping;
3393 EXPORT_SYMBOL_GPL(swapcache_mapping);
3395 pgoff_t __page_file_index(struct page *page)
3397 swp_entry_t swap = { .val = page_private(page) };
3398 return swp_offset(swap);
3400 EXPORT_SYMBOL_GPL(__page_file_index);
3403 * add_swap_count_continuation - called when a swap count is duplicated
3404 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3405 * page of the original vmalloc'ed swap_map, to hold the continuation count
3406 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3407 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3409 * These continuation pages are seldom referenced: the common paths all work
3410 * on the original swap_map, only referring to a continuation page when the
3411 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3413 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3414 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3415 * can be called after dropping locks.
3417 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3419 struct swap_info_struct *si;
3420 struct swap_cluster_info *ci;
3423 struct page *list_page;
3425 unsigned char count;
3429 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3430 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3432 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3434 si = get_swap_device(entry);
3437 * An acceptable race has occurred since the failing
3438 * __swap_duplicate(): the swap device may be swapoff
3442 spin_lock(&si->lock);
3444 offset = swp_offset(entry);
3446 ci = lock_cluster(si, offset);
3448 count = swap_count(si->swap_map[offset]);
3450 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3452 * The higher the swap count, the more likely it is that tasks
3453 * will race to add swap count continuation: we need to avoid
3454 * over-provisioning.
3465 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3466 * no architecture is using highmem pages for kernel page tables: so it
3467 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3469 head = vmalloc_to_page(si->swap_map + offset);
3470 offset &= ~PAGE_MASK;
3472 spin_lock(&si->cont_lock);
3474 * Page allocation does not initialize the page's lru field,
3475 * but it does always reset its private field.
3477 if (!page_private(head)) {
3478 BUG_ON(count & COUNT_CONTINUED);
3479 INIT_LIST_HEAD(&head->lru);
3480 set_page_private(head, SWP_CONTINUED);
3481 si->flags |= SWP_CONTINUED;
3484 list_for_each_entry(list_page, &head->lru, lru) {
3488 * If the previous map said no continuation, but we've found
3489 * a continuation page, free our allocation and use this one.
3491 if (!(count & COUNT_CONTINUED))
3492 goto out_unlock_cont;
3494 map = kmap_atomic(list_page) + offset;
3499 * If this continuation count now has some space in it,
3500 * free our allocation and use this one.
3502 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3503 goto out_unlock_cont;
3506 list_add_tail(&page->lru, &head->lru);
3507 page = NULL; /* now it's attached, don't free it */
3509 spin_unlock(&si->cont_lock);
3512 spin_unlock(&si->lock);
3513 put_swap_device(si);
3521 * swap_count_continued - when the original swap_map count is incremented
3522 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3523 * into, carry if so, or else fail until a new continuation page is allocated;
3524 * when the original swap_map count is decremented from 0 with continuation,
3525 * borrow from the continuation and report whether it still holds more.
3526 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3529 static bool swap_count_continued(struct swap_info_struct *si,
3530 pgoff_t offset, unsigned char count)
3537 head = vmalloc_to_page(si->swap_map + offset);
3538 if (page_private(head) != SWP_CONTINUED) {
3539 BUG_ON(count & COUNT_CONTINUED);
3540 return false; /* need to add count continuation */
3543 spin_lock(&si->cont_lock);
3544 offset &= ~PAGE_MASK;
3545 page = list_next_entry(head, lru);
3546 map = kmap_atomic(page) + offset;
3548 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3549 goto init_map; /* jump over SWAP_CONT_MAX checks */
3551 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3553 * Think of how you add 1 to 999
3555 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3557 page = list_next_entry(page, lru);
3558 BUG_ON(page == head);
3559 map = kmap_atomic(page) + offset;
3561 if (*map == SWAP_CONT_MAX) {
3563 page = list_next_entry(page, lru);
3565 ret = false; /* add count continuation */
3568 map = kmap_atomic(page) + offset;
3569 init_map: *map = 0; /* we didn't zero the page */
3573 while ((page = list_prev_entry(page, lru)) != head) {
3574 map = kmap_atomic(page) + offset;
3575 *map = COUNT_CONTINUED;
3578 ret = true; /* incremented */
3580 } else { /* decrementing */
3582 * Think of how you subtract 1 from 1000
3584 BUG_ON(count != COUNT_CONTINUED);
3585 while (*map == COUNT_CONTINUED) {
3587 page = list_next_entry(page, lru);
3588 BUG_ON(page == head);
3589 map = kmap_atomic(page) + offset;
3596 while ((page = list_prev_entry(page, lru)) != head) {
3597 map = kmap_atomic(page) + offset;
3598 *map = SWAP_CONT_MAX | count;
3599 count = COUNT_CONTINUED;
3602 ret = count == COUNT_CONTINUED;
3605 spin_unlock(&si->cont_lock);
3610 * free_swap_count_continuations - swapoff free all the continuation pages
3611 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3613 static void free_swap_count_continuations(struct swap_info_struct *si)
3617 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3619 head = vmalloc_to_page(si->swap_map + offset);
3620 if (page_private(head)) {
3621 struct page *page, *next;
3623 list_for_each_entry_safe(page, next, &head->lru, lru) {
3624 list_del(&page->lru);
3631 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3632 void __cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask)
3634 struct swap_info_struct *si, *next;
3635 int nid = page_to_nid(page);
3637 if (!(gfp_mask & __GFP_IO))
3640 if (!blk_cgroup_congested())
3644 * We've already scheduled a throttle, avoid taking the global swap
3647 if (current->throttle_queue)
3650 spin_lock(&swap_avail_lock);
3651 plist_for_each_entry_safe(si, next, &swap_avail_heads[nid],
3654 blkcg_schedule_throttle(si->bdev->bd_disk, true);
3658 spin_unlock(&swap_avail_lock);
3662 static int __init swapfile_init(void)
3666 swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3668 if (!swap_avail_heads) {
3669 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3674 plist_head_init(&swap_avail_heads[nid]);
3676 swapfile_maximum_size = arch_max_swapfile_size();
3678 #ifdef CONFIG_MIGRATION
3679 if (swapfile_maximum_size >= (1UL << SWP_MIG_TOTAL_BITS))
3680 swap_migration_ad_supported = true;
3681 #endif /* CONFIG_MIGRATION */
3685 subsys_initcall(swapfile_init);