1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
6 * Swap reorganised 29.12.95, Stephen Tweedie
10 #include <linux/sched/mm.h>
11 #include <linux/sched/task.h>
12 #include <linux/hugetlb.h>
13 #include <linux/mman.h>
14 #include <linux/slab.h>
15 #include <linux/kernel_stat.h>
16 #include <linux/swap.h>
17 #include <linux/vmalloc.h>
18 #include <linux/pagemap.h>
19 #include <linux/namei.h>
20 #include <linux/shmem_fs.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/writeback.h>
24 #include <linux/proc_fs.h>
25 #include <linux/seq_file.h>
26 #include <linux/init.h>
27 #include <linux/ksm.h>
28 #include <linux/rmap.h>
29 #include <linux/security.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mutex.h>
32 #include <linux/capability.h>
33 #include <linux/syscalls.h>
34 #include <linux/memcontrol.h>
35 #include <linux/poll.h>
36 #include <linux/oom.h>
37 #include <linux/frontswap.h>
38 #include <linux/swapfile.h>
39 #include <linux/export.h>
40 #include <linux/swap_slots.h>
41 #include <linux/sort.h>
42 #include <linux/completion.h>
44 #include <asm/tlbflush.h>
45 #include <linux/swapops.h>
46 #include <linux/swap_cgroup.h>
48 static bool swap_count_continued(struct swap_info_struct *, pgoff_t,
50 static void free_swap_count_continuations(struct swap_info_struct *);
52 DEFINE_SPINLOCK(swap_lock);
53 static unsigned int nr_swapfiles;
54 atomic_long_t nr_swap_pages;
56 * Some modules use swappable objects and may try to swap them out under
57 * memory pressure (via the shrinker). Before doing so, they may wish to
58 * check to see if any swap space is available.
60 EXPORT_SYMBOL_GPL(nr_swap_pages);
61 /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */
62 long total_swap_pages;
63 static int least_priority = -1;
65 static const char Bad_file[] = "Bad swap file entry ";
66 static const char Unused_file[] = "Unused swap file entry ";
67 static const char Bad_offset[] = "Bad swap offset entry ";
68 static const char Unused_offset[] = "Unused swap offset entry ";
71 * all active swap_info_structs
72 * protected with swap_lock, and ordered by priority.
74 PLIST_HEAD(swap_active_head);
77 * all available (active, not full) swap_info_structs
78 * protected with swap_avail_lock, ordered by priority.
79 * This is used by get_swap_page() instead of swap_active_head
80 * because swap_active_head includes all swap_info_structs,
81 * but get_swap_page() doesn't need to look at full ones.
82 * This uses its own lock instead of swap_lock because when a
83 * swap_info_struct changes between not-full/full, it needs to
84 * add/remove itself to/from this list, but the swap_info_struct->lock
85 * is held and the locking order requires swap_lock to be taken
86 * before any swap_info_struct->lock.
88 static struct plist_head *swap_avail_heads;
89 static DEFINE_SPINLOCK(swap_avail_lock);
91 struct swap_info_struct *swap_info[MAX_SWAPFILES];
93 static DEFINE_MUTEX(swapon_mutex);
95 static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait);
96 /* Activity counter to indicate that a swapon or swapoff has occurred */
97 static atomic_t proc_poll_event = ATOMIC_INIT(0);
99 atomic_t nr_rotate_swap = ATOMIC_INIT(0);
101 static struct swap_info_struct *swap_type_to_swap_info(int type)
103 if (type >= READ_ONCE(nr_swapfiles))
106 smp_rmb(); /* Pairs with smp_wmb in alloc_swap_info. */
107 return READ_ONCE(swap_info[type]);
110 static inline unsigned char swap_count(unsigned char ent)
112 return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */
115 /* Reclaim the swap entry anyway if possible */
116 #define TTRS_ANYWAY 0x1
118 * Reclaim the swap entry if there are no more mappings of the
121 #define TTRS_UNMAPPED 0x2
122 /* Reclaim the swap entry if swap is getting full*/
123 #define TTRS_FULL 0x4
125 /* returns 1 if swap entry is freed */
126 static int __try_to_reclaim_swap(struct swap_info_struct *si,
127 unsigned long offset, unsigned long flags)
129 swp_entry_t entry = swp_entry(si->type, offset);
133 page = find_get_page(swap_address_space(entry), offset);
137 * When this function is called from scan_swap_map_slots() and it's
138 * called by vmscan.c at reclaiming pages. So, we hold a lock on a page,
139 * here. We have to use trylock for avoiding deadlock. This is a special
140 * case and you should use try_to_free_swap() with explicit lock_page()
141 * in usual operations.
143 if (trylock_page(page)) {
144 if ((flags & TTRS_ANYWAY) ||
145 ((flags & TTRS_UNMAPPED) && !page_mapped(page)) ||
146 ((flags & TTRS_FULL) && mem_cgroup_swap_full(page)))
147 ret = try_to_free_swap(page);
154 static inline struct swap_extent *first_se(struct swap_info_struct *sis)
156 struct rb_node *rb = rb_first(&sis->swap_extent_root);
157 return rb_entry(rb, struct swap_extent, rb_node);
160 static inline struct swap_extent *next_se(struct swap_extent *se)
162 struct rb_node *rb = rb_next(&se->rb_node);
163 return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL;
167 * swapon tell device that all the old swap contents can be discarded,
168 * to allow the swap device to optimize its wear-levelling.
170 static int discard_swap(struct swap_info_struct *si)
172 struct swap_extent *se;
173 sector_t start_block;
177 /* Do not discard the swap header page! */
179 start_block = (se->start_block + 1) << (PAGE_SHIFT - 9);
180 nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9);
182 err = blkdev_issue_discard(si->bdev, start_block,
183 nr_blocks, GFP_KERNEL, 0);
189 for (se = next_se(se); se; se = next_se(se)) {
190 start_block = se->start_block << (PAGE_SHIFT - 9);
191 nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9);
193 err = blkdev_issue_discard(si->bdev, start_block,
194 nr_blocks, GFP_KERNEL, 0);
200 return err; /* That will often be -EOPNOTSUPP */
203 static struct swap_extent *
204 offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset)
206 struct swap_extent *se;
209 rb = sis->swap_extent_root.rb_node;
211 se = rb_entry(rb, struct swap_extent, rb_node);
212 if (offset < se->start_page)
214 else if (offset >= se->start_page + se->nr_pages)
219 /* It *must* be present */
223 sector_t swap_page_sector(struct page *page)
225 struct swap_info_struct *sis = page_swap_info(page);
226 struct swap_extent *se;
230 offset = __page_file_index(page);
231 se = offset_to_swap_extent(sis, offset);
232 sector = se->start_block + (offset - se->start_page);
233 return sector << (PAGE_SHIFT - 9);
237 * swap allocation tell device that a cluster of swap can now be discarded,
238 * to allow the swap device to optimize its wear-levelling.
240 static void discard_swap_cluster(struct swap_info_struct *si,
241 pgoff_t start_page, pgoff_t nr_pages)
243 struct swap_extent *se = offset_to_swap_extent(si, start_page);
246 pgoff_t offset = start_page - se->start_page;
247 sector_t start_block = se->start_block + offset;
248 sector_t nr_blocks = se->nr_pages - offset;
250 if (nr_blocks > nr_pages)
251 nr_blocks = nr_pages;
252 start_page += nr_blocks;
253 nr_pages -= nr_blocks;
255 start_block <<= PAGE_SHIFT - 9;
256 nr_blocks <<= PAGE_SHIFT - 9;
257 if (blkdev_issue_discard(si->bdev, start_block,
258 nr_blocks, GFP_NOIO, 0))
265 #ifdef CONFIG_THP_SWAP
266 #define SWAPFILE_CLUSTER HPAGE_PMD_NR
268 #define swap_entry_size(size) (size)
270 #define SWAPFILE_CLUSTER 256
273 * Define swap_entry_size() as constant to let compiler to optimize
274 * out some code if !CONFIG_THP_SWAP
276 #define swap_entry_size(size) 1
278 #define LATENCY_LIMIT 256
280 static inline void cluster_set_flag(struct swap_cluster_info *info,
286 static inline unsigned int cluster_count(struct swap_cluster_info *info)
291 static inline void cluster_set_count(struct swap_cluster_info *info,
297 static inline void cluster_set_count_flag(struct swap_cluster_info *info,
298 unsigned int c, unsigned int f)
304 static inline unsigned int cluster_next(struct swap_cluster_info *info)
309 static inline void cluster_set_next(struct swap_cluster_info *info,
315 static inline void cluster_set_next_flag(struct swap_cluster_info *info,
316 unsigned int n, unsigned int f)
322 static inline bool cluster_is_free(struct swap_cluster_info *info)
324 return info->flags & CLUSTER_FLAG_FREE;
327 static inline bool cluster_is_null(struct swap_cluster_info *info)
329 return info->flags & CLUSTER_FLAG_NEXT_NULL;
332 static inline void cluster_set_null(struct swap_cluster_info *info)
334 info->flags = CLUSTER_FLAG_NEXT_NULL;
338 static inline bool cluster_is_huge(struct swap_cluster_info *info)
340 if (IS_ENABLED(CONFIG_THP_SWAP))
341 return info->flags & CLUSTER_FLAG_HUGE;
345 static inline void cluster_clear_huge(struct swap_cluster_info *info)
347 info->flags &= ~CLUSTER_FLAG_HUGE;
350 static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si,
351 unsigned long offset)
353 struct swap_cluster_info *ci;
355 ci = si->cluster_info;
357 ci += offset / SWAPFILE_CLUSTER;
358 spin_lock(&ci->lock);
363 static inline void unlock_cluster(struct swap_cluster_info *ci)
366 spin_unlock(&ci->lock);
370 * Determine the locking method in use for this device. Return
371 * swap_cluster_info if SSD-style cluster-based locking is in place.
373 static inline struct swap_cluster_info *lock_cluster_or_swap_info(
374 struct swap_info_struct *si, unsigned long offset)
376 struct swap_cluster_info *ci;
378 /* Try to use fine-grained SSD-style locking if available: */
379 ci = lock_cluster(si, offset);
380 /* Otherwise, fall back to traditional, coarse locking: */
382 spin_lock(&si->lock);
387 static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si,
388 struct swap_cluster_info *ci)
393 spin_unlock(&si->lock);
396 static inline bool cluster_list_empty(struct swap_cluster_list *list)
398 return cluster_is_null(&list->head);
401 static inline unsigned int cluster_list_first(struct swap_cluster_list *list)
403 return cluster_next(&list->head);
406 static void cluster_list_init(struct swap_cluster_list *list)
408 cluster_set_null(&list->head);
409 cluster_set_null(&list->tail);
412 static void cluster_list_add_tail(struct swap_cluster_list *list,
413 struct swap_cluster_info *ci,
416 if (cluster_list_empty(list)) {
417 cluster_set_next_flag(&list->head, idx, 0);
418 cluster_set_next_flag(&list->tail, idx, 0);
420 struct swap_cluster_info *ci_tail;
421 unsigned int tail = cluster_next(&list->tail);
424 * Nested cluster lock, but both cluster locks are
425 * only acquired when we held swap_info_struct->lock
428 spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING);
429 cluster_set_next(ci_tail, idx);
430 spin_unlock(&ci_tail->lock);
431 cluster_set_next_flag(&list->tail, idx, 0);
435 static unsigned int cluster_list_del_first(struct swap_cluster_list *list,
436 struct swap_cluster_info *ci)
440 idx = cluster_next(&list->head);
441 if (cluster_next(&list->tail) == idx) {
442 cluster_set_null(&list->head);
443 cluster_set_null(&list->tail);
445 cluster_set_next_flag(&list->head,
446 cluster_next(&ci[idx]), 0);
451 /* Add a cluster to discard list and schedule it to do discard */
452 static void swap_cluster_schedule_discard(struct swap_info_struct *si,
456 * If scan_swap_map_slots() can't find a free cluster, it will check
457 * si->swap_map directly. To make sure the discarding cluster isn't
458 * taken by scan_swap_map_slots(), mark the swap entries bad (occupied).
459 * It will be cleared after discard
461 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
462 SWAP_MAP_BAD, SWAPFILE_CLUSTER);
464 cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx);
466 schedule_work(&si->discard_work);
469 static void __free_cluster(struct swap_info_struct *si, unsigned long idx)
471 struct swap_cluster_info *ci = si->cluster_info;
473 cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE);
474 cluster_list_add_tail(&si->free_clusters, ci, idx);
478 * Doing discard actually. After a cluster discard is finished, the cluster
479 * will be added to free cluster list. caller should hold si->lock.
481 static void swap_do_scheduled_discard(struct swap_info_struct *si)
483 struct swap_cluster_info *info, *ci;
486 info = si->cluster_info;
488 while (!cluster_list_empty(&si->discard_clusters)) {
489 idx = cluster_list_del_first(&si->discard_clusters, info);
490 spin_unlock(&si->lock);
492 discard_swap_cluster(si, idx * SWAPFILE_CLUSTER,
495 spin_lock(&si->lock);
496 ci = lock_cluster(si, idx * SWAPFILE_CLUSTER);
497 __free_cluster(si, idx);
498 memset(si->swap_map + idx * SWAPFILE_CLUSTER,
499 0, SWAPFILE_CLUSTER);
504 static void swap_discard_work(struct work_struct *work)
506 struct swap_info_struct *si;
508 si = container_of(work, struct swap_info_struct, discard_work);
510 spin_lock(&si->lock);
511 swap_do_scheduled_discard(si);
512 spin_unlock(&si->lock);
515 static void swap_users_ref_free(struct percpu_ref *ref)
517 struct swap_info_struct *si;
519 si = container_of(ref, struct swap_info_struct, users);
523 static void alloc_cluster(struct swap_info_struct *si, unsigned long idx)
525 struct swap_cluster_info *ci = si->cluster_info;
527 VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx);
528 cluster_list_del_first(&si->free_clusters, ci);
529 cluster_set_count_flag(ci + idx, 0, 0);
532 static void free_cluster(struct swap_info_struct *si, unsigned long idx)
534 struct swap_cluster_info *ci = si->cluster_info + idx;
536 VM_BUG_ON(cluster_count(ci) != 0);
538 * If the swap is discardable, prepare discard the cluster
539 * instead of free it immediately. The cluster will be freed
542 if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) ==
543 (SWP_WRITEOK | SWP_PAGE_DISCARD)) {
544 swap_cluster_schedule_discard(si, idx);
548 __free_cluster(si, idx);
552 * The cluster corresponding to page_nr will be used. The cluster will be
553 * removed from free cluster list and its usage counter will be increased.
555 static void inc_cluster_info_page(struct swap_info_struct *p,
556 struct swap_cluster_info *cluster_info, unsigned long page_nr)
558 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
562 if (cluster_is_free(&cluster_info[idx]))
563 alloc_cluster(p, idx);
565 VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER);
566 cluster_set_count(&cluster_info[idx],
567 cluster_count(&cluster_info[idx]) + 1);
571 * The cluster corresponding to page_nr decreases one usage. If the usage
572 * counter becomes 0, which means no page in the cluster is in using, we can
573 * optionally discard the cluster and add it to free cluster list.
575 static void dec_cluster_info_page(struct swap_info_struct *p,
576 struct swap_cluster_info *cluster_info, unsigned long page_nr)
578 unsigned long idx = page_nr / SWAPFILE_CLUSTER;
583 VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0);
584 cluster_set_count(&cluster_info[idx],
585 cluster_count(&cluster_info[idx]) - 1);
587 if (cluster_count(&cluster_info[idx]) == 0)
588 free_cluster(p, idx);
592 * It's possible scan_swap_map_slots() uses a free cluster in the middle of free
593 * cluster list. Avoiding such abuse to avoid list corruption.
596 scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si,
597 unsigned long offset)
599 struct percpu_cluster *percpu_cluster;
602 offset /= SWAPFILE_CLUSTER;
603 conflict = !cluster_list_empty(&si->free_clusters) &&
604 offset != cluster_list_first(&si->free_clusters) &&
605 cluster_is_free(&si->cluster_info[offset]);
610 percpu_cluster = this_cpu_ptr(si->percpu_cluster);
611 cluster_set_null(&percpu_cluster->index);
616 * Try to get a swap entry from current cpu's swap entry pool (a cluster). This
617 * might involve allocating a new cluster for current CPU too.
619 static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si,
620 unsigned long *offset, unsigned long *scan_base)
622 struct percpu_cluster *cluster;
623 struct swap_cluster_info *ci;
624 unsigned long tmp, max;
627 cluster = this_cpu_ptr(si->percpu_cluster);
628 if (cluster_is_null(&cluster->index)) {
629 if (!cluster_list_empty(&si->free_clusters)) {
630 cluster->index = si->free_clusters.head;
631 cluster->next = cluster_next(&cluster->index) *
633 } else if (!cluster_list_empty(&si->discard_clusters)) {
635 * we don't have free cluster but have some clusters in
636 * discarding, do discard now and reclaim them, then
637 * reread cluster_next_cpu since we dropped si->lock
639 swap_do_scheduled_discard(si);
640 *scan_base = this_cpu_read(*si->cluster_next_cpu);
641 *offset = *scan_base;
648 * Other CPUs can use our cluster if they can't find a free cluster,
649 * check if there is still free entry in the cluster
652 max = min_t(unsigned long, si->max,
653 (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER);
655 ci = lock_cluster(si, tmp);
657 if (!si->swap_map[tmp])
664 cluster_set_null(&cluster->index);
667 cluster->next = tmp + 1;
673 static void __del_from_avail_list(struct swap_info_struct *p)
678 plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]);
681 static void del_from_avail_list(struct swap_info_struct *p)
683 spin_lock(&swap_avail_lock);
684 __del_from_avail_list(p);
685 spin_unlock(&swap_avail_lock);
688 static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset,
689 unsigned int nr_entries)
691 unsigned int end = offset + nr_entries - 1;
693 if (offset == si->lowest_bit)
694 si->lowest_bit += nr_entries;
695 if (end == si->highest_bit)
696 WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries);
697 si->inuse_pages += nr_entries;
698 if (si->inuse_pages == si->pages) {
699 si->lowest_bit = si->max;
701 del_from_avail_list(si);
705 static void add_to_avail_list(struct swap_info_struct *p)
709 spin_lock(&swap_avail_lock);
711 WARN_ON(!plist_node_empty(&p->avail_lists[nid]));
712 plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]);
714 spin_unlock(&swap_avail_lock);
717 static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
718 unsigned int nr_entries)
720 unsigned long begin = offset;
721 unsigned long end = offset + nr_entries - 1;
722 void (*swap_slot_free_notify)(struct block_device *, unsigned long);
724 if (offset < si->lowest_bit)
725 si->lowest_bit = offset;
726 if (end > si->highest_bit) {
727 bool was_full = !si->highest_bit;
729 WRITE_ONCE(si->highest_bit, end);
730 if (was_full && (si->flags & SWP_WRITEOK))
731 add_to_avail_list(si);
733 atomic_long_add(nr_entries, &nr_swap_pages);
734 si->inuse_pages -= nr_entries;
735 if (si->flags & SWP_BLKDEV)
736 swap_slot_free_notify =
737 si->bdev->bd_disk->fops->swap_slot_free_notify;
739 swap_slot_free_notify = NULL;
740 while (offset <= end) {
741 arch_swap_invalidate_page(si->type, offset);
742 frontswap_invalidate_page(si->type, offset);
743 if (swap_slot_free_notify)
744 swap_slot_free_notify(si->bdev, offset);
747 clear_shadow_from_swap_cache(si->type, begin, end);
750 static void set_cluster_next(struct swap_info_struct *si, unsigned long next)
754 if (!(si->flags & SWP_SOLIDSTATE)) {
755 si->cluster_next = next;
759 prev = this_cpu_read(*si->cluster_next_cpu);
761 * Cross the swap address space size aligned trunk, choose
762 * another trunk randomly to avoid lock contention on swap
763 * address space if possible.
765 if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) !=
766 (next >> SWAP_ADDRESS_SPACE_SHIFT)) {
767 /* No free swap slots available */
768 if (si->highest_bit <= si->lowest_bit)
770 next = si->lowest_bit +
771 prandom_u32_max(si->highest_bit - si->lowest_bit + 1);
772 next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES);
773 next = max_t(unsigned int, next, si->lowest_bit);
775 this_cpu_write(*si->cluster_next_cpu, next);
778 static int scan_swap_map_slots(struct swap_info_struct *si,
779 unsigned char usage, int nr,
782 struct swap_cluster_info *ci;
783 unsigned long offset;
784 unsigned long scan_base;
785 unsigned long last_in_cluster = 0;
786 int latency_ration = LATENCY_LIMIT;
788 bool scanned_many = false;
791 * We try to cluster swap pages by allocating them sequentially
792 * in swap. Once we've allocated SWAPFILE_CLUSTER pages this
793 * way, however, we resort to first-free allocation, starting
794 * a new cluster. This prevents us from scattering swap pages
795 * all over the entire swap partition, so that we reduce
796 * overall disk seek times between swap pages. -- sct
797 * But we do now try to find an empty cluster. -Andrea
798 * And we let swap pages go all over an SSD partition. Hugh
801 si->flags += SWP_SCANNING;
803 * Use percpu scan base for SSD to reduce lock contention on
804 * cluster and swap cache. For HDD, sequential access is more
807 if (si->flags & SWP_SOLIDSTATE)
808 scan_base = this_cpu_read(*si->cluster_next_cpu);
810 scan_base = si->cluster_next;
814 if (si->cluster_info) {
815 if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
817 } else if (unlikely(!si->cluster_nr--)) {
818 if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) {
819 si->cluster_nr = SWAPFILE_CLUSTER - 1;
823 spin_unlock(&si->lock);
826 * If seek is expensive, start searching for new cluster from
827 * start of partition, to minimize the span of allocated swap.
828 * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info
829 * case, just handled by scan_swap_map_try_ssd_cluster() above.
831 scan_base = offset = si->lowest_bit;
832 last_in_cluster = offset + SWAPFILE_CLUSTER - 1;
834 /* Locate the first empty (unaligned) cluster */
835 for (; last_in_cluster <= si->highest_bit; offset++) {
836 if (si->swap_map[offset])
837 last_in_cluster = offset + SWAPFILE_CLUSTER;
838 else if (offset == last_in_cluster) {
839 spin_lock(&si->lock);
840 offset -= SWAPFILE_CLUSTER - 1;
841 si->cluster_next = offset;
842 si->cluster_nr = SWAPFILE_CLUSTER - 1;
845 if (unlikely(--latency_ration < 0)) {
847 latency_ration = LATENCY_LIMIT;
852 spin_lock(&si->lock);
853 si->cluster_nr = SWAPFILE_CLUSTER - 1;
857 if (si->cluster_info) {
858 while (scan_swap_map_ssd_cluster_conflict(si, offset)) {
859 /* take a break if we already got some slots */
862 if (!scan_swap_map_try_ssd_cluster(si, &offset,
867 if (!(si->flags & SWP_WRITEOK))
869 if (!si->highest_bit)
871 if (offset > si->highest_bit)
872 scan_base = offset = si->lowest_bit;
874 ci = lock_cluster(si, offset);
875 /* reuse swap entry of cache-only swap if not busy. */
876 if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) {
879 spin_unlock(&si->lock);
880 swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY);
881 spin_lock(&si->lock);
882 /* entry was freed successfully, try to use this again */
885 goto scan; /* check next one */
888 if (si->swap_map[offset]) {
895 WRITE_ONCE(si->swap_map[offset], usage);
896 inc_cluster_info_page(si, si->cluster_info, offset);
899 swap_range_alloc(si, offset, 1);
900 slots[n_ret++] = swp_entry(si->type, offset);
902 /* got enough slots or reach max slots? */
903 if ((n_ret == nr) || (offset >= si->highest_bit))
906 /* search for next available slot */
908 /* time to take a break? */
909 if (unlikely(--latency_ration < 0)) {
912 spin_unlock(&si->lock);
914 spin_lock(&si->lock);
915 latency_ration = LATENCY_LIMIT;
918 /* try to get more slots in cluster */
919 if (si->cluster_info) {
920 if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base))
922 } else if (si->cluster_nr && !si->swap_map[++offset]) {
923 /* non-ssd case, still more slots in cluster? */
929 * Even if there's no free clusters available (fragmented),
930 * try to scan a little more quickly with lock held unless we
931 * have scanned too many slots already.
934 unsigned long scan_limit;
936 if (offset < scan_base)
937 scan_limit = scan_base;
939 scan_limit = si->highest_bit;
940 for (; offset <= scan_limit && --latency_ration > 0;
942 if (!si->swap_map[offset])
948 set_cluster_next(si, offset + 1);
949 si->flags -= SWP_SCANNING;
953 spin_unlock(&si->lock);
954 while (++offset <= READ_ONCE(si->highest_bit)) {
955 if (data_race(!si->swap_map[offset])) {
956 spin_lock(&si->lock);
959 if (vm_swap_full() &&
960 READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
961 spin_lock(&si->lock);
964 if (unlikely(--latency_ration < 0)) {
966 latency_ration = LATENCY_LIMIT;
970 offset = si->lowest_bit;
971 while (offset < scan_base) {
972 if (data_race(!si->swap_map[offset])) {
973 spin_lock(&si->lock);
976 if (vm_swap_full() &&
977 READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) {
978 spin_lock(&si->lock);
981 if (unlikely(--latency_ration < 0)) {
983 latency_ration = LATENCY_LIMIT;
988 spin_lock(&si->lock);
991 si->flags -= SWP_SCANNING;
995 static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot)
998 struct swap_cluster_info *ci;
999 unsigned long offset;
1002 * Should not even be attempting cluster allocations when huge
1003 * page swap is disabled. Warn and fail the allocation.
1005 if (!IS_ENABLED(CONFIG_THP_SWAP)) {
1010 if (cluster_list_empty(&si->free_clusters))
1013 idx = cluster_list_first(&si->free_clusters);
1014 offset = idx * SWAPFILE_CLUSTER;
1015 ci = lock_cluster(si, offset);
1016 alloc_cluster(si, idx);
1017 cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE);
1019 memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER);
1021 swap_range_alloc(si, offset, SWAPFILE_CLUSTER);
1022 *slot = swp_entry(si->type, offset);
1027 static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx)
1029 unsigned long offset = idx * SWAPFILE_CLUSTER;
1030 struct swap_cluster_info *ci;
1032 ci = lock_cluster(si, offset);
1033 memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER);
1034 cluster_set_count_flag(ci, 0, 0);
1035 free_cluster(si, idx);
1037 swap_range_free(si, offset, SWAPFILE_CLUSTER);
1040 int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
1042 unsigned long size = swap_entry_size(entry_size);
1043 struct swap_info_struct *si, *next;
1048 /* Only single cluster request supported */
1049 WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
1051 spin_lock(&swap_avail_lock);
1053 avail_pgs = atomic_long_read(&nr_swap_pages) / size;
1054 if (avail_pgs <= 0) {
1055 spin_unlock(&swap_avail_lock);
1059 n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
1061 atomic_long_sub(n_goal * size, &nr_swap_pages);
1064 node = numa_node_id();
1065 plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
1066 /* requeue si to after same-priority siblings */
1067 plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]);
1068 spin_unlock(&swap_avail_lock);
1069 spin_lock(&si->lock);
1070 if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) {
1071 spin_lock(&swap_avail_lock);
1072 if (plist_node_empty(&si->avail_lists[node])) {
1073 spin_unlock(&si->lock);
1076 WARN(!si->highest_bit,
1077 "swap_info %d in list but !highest_bit\n",
1079 WARN(!(si->flags & SWP_WRITEOK),
1080 "swap_info %d in list but !SWP_WRITEOK\n",
1082 __del_from_avail_list(si);
1083 spin_unlock(&si->lock);
1086 if (size == SWAPFILE_CLUSTER) {
1087 if (si->flags & SWP_BLKDEV)
1088 n_ret = swap_alloc_cluster(si, swp_entries);
1090 n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE,
1091 n_goal, swp_entries);
1092 spin_unlock(&si->lock);
1093 if (n_ret || size == SWAPFILE_CLUSTER)
1095 pr_debug("scan_swap_map of si %d failed to find offset\n",
1098 spin_lock(&swap_avail_lock);
1101 * if we got here, it's likely that si was almost full before,
1102 * and since scan_swap_map_slots() can drop the si->lock,
1103 * multiple callers probably all tried to get a page from the
1104 * same si and it filled up before we could get one; or, the si
1105 * filled up between us dropping swap_avail_lock and taking
1106 * si->lock. Since we dropped the swap_avail_lock, the
1107 * swap_avail_head list may have been modified; so if next is
1108 * still in the swap_avail_head list then try it, otherwise
1109 * start over if we have not gotten any slots.
1111 if (plist_node_empty(&next->avail_lists[node]))
1115 spin_unlock(&swap_avail_lock);
1119 atomic_long_add((long)(n_goal - n_ret) * size,
1125 static struct swap_info_struct *__swap_info_get(swp_entry_t entry)
1127 struct swap_info_struct *p;
1128 unsigned long offset;
1132 p = swp_swap_info(entry);
1135 if (data_race(!(p->flags & SWP_USED)))
1137 offset = swp_offset(entry);
1138 if (offset >= p->max)
1143 pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val);
1146 pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val);
1149 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1154 static struct swap_info_struct *_swap_info_get(swp_entry_t entry)
1156 struct swap_info_struct *p;
1158 p = __swap_info_get(entry);
1161 if (data_race(!p->swap_map[swp_offset(entry)]))
1166 pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val);
1171 static struct swap_info_struct *swap_info_get(swp_entry_t entry)
1173 struct swap_info_struct *p;
1175 p = _swap_info_get(entry);
1177 spin_lock(&p->lock);
1181 static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry,
1182 struct swap_info_struct *q)
1184 struct swap_info_struct *p;
1186 p = _swap_info_get(entry);
1190 spin_unlock(&q->lock);
1192 spin_lock(&p->lock);
1197 static unsigned char __swap_entry_free_locked(struct swap_info_struct *p,
1198 unsigned long offset,
1199 unsigned char usage)
1201 unsigned char count;
1202 unsigned char has_cache;
1204 count = p->swap_map[offset];
1206 has_cache = count & SWAP_HAS_CACHE;
1207 count &= ~SWAP_HAS_CACHE;
1209 if (usage == SWAP_HAS_CACHE) {
1210 VM_BUG_ON(!has_cache);
1212 } else if (count == SWAP_MAP_SHMEM) {
1214 * Or we could insist on shmem.c using a special
1215 * swap_shmem_free() and free_shmem_swap_and_cache()...
1218 } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) {
1219 if (count == COUNT_CONTINUED) {
1220 if (swap_count_continued(p, offset, count))
1221 count = SWAP_MAP_MAX | COUNT_CONTINUED;
1223 count = SWAP_MAP_MAX;
1228 usage = count | has_cache;
1230 WRITE_ONCE(p->swap_map[offset], usage);
1232 WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE);
1238 * Check whether swap entry is valid in the swap device. If so,
1239 * return pointer to swap_info_struct, and keep the swap entry valid
1240 * via preventing the swap device from being swapoff, until
1241 * put_swap_device() is called. Otherwise return NULL.
1243 * Notice that swapoff or swapoff+swapon can still happen before the
1244 * percpu_ref_tryget_live() in get_swap_device() or after the
1245 * percpu_ref_put() in put_swap_device() if there isn't any other way
1246 * to prevent swapoff, such as page lock, page table lock, etc. The
1247 * caller must be prepared for that. For example, the following
1248 * situation is possible.
1252 * ... swapoff+swapon
1253 * __read_swap_cache_async()
1254 * swapcache_prepare()
1255 * __swap_duplicate()
1257 * // verify PTE not changed
1259 * In __swap_duplicate(), the swap_map need to be checked before
1260 * changing partly because the specified swap entry may be for another
1261 * swap device which has been swapoff. And in do_swap_page(), after
1262 * the page is read from the swap device, the PTE is verified not
1263 * changed with the page table locked to check whether the swap device
1264 * has been swapoff or swapoff+swapon.
1266 struct swap_info_struct *get_swap_device(swp_entry_t entry)
1268 struct swap_info_struct *si;
1269 unsigned long offset;
1273 si = swp_swap_info(entry);
1276 if (!percpu_ref_tryget_live(&si->users))
1279 * Guarantee the si->users are checked before accessing other
1280 * fields of swap_info_struct.
1282 * Paired with the spin_unlock() after setup_swap_info() in
1283 * enable_swap_info().
1286 offset = swp_offset(entry);
1287 if (offset >= si->max)
1292 pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val);
1296 percpu_ref_put(&si->users);
1300 static unsigned char __swap_entry_free(struct swap_info_struct *p,
1303 struct swap_cluster_info *ci;
1304 unsigned long offset = swp_offset(entry);
1305 unsigned char usage;
1307 ci = lock_cluster_or_swap_info(p, offset);
1308 usage = __swap_entry_free_locked(p, offset, 1);
1309 unlock_cluster_or_swap_info(p, ci);
1311 free_swap_slot(entry);
1316 static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry)
1318 struct swap_cluster_info *ci;
1319 unsigned long offset = swp_offset(entry);
1320 unsigned char count;
1322 ci = lock_cluster(p, offset);
1323 count = p->swap_map[offset];
1324 VM_BUG_ON(count != SWAP_HAS_CACHE);
1325 p->swap_map[offset] = 0;
1326 dec_cluster_info_page(p, p->cluster_info, offset);
1329 mem_cgroup_uncharge_swap(entry, 1);
1330 swap_range_free(p, offset, 1);
1334 * Caller has made sure that the swap device corresponding to entry
1335 * is still around or has not been recycled.
1337 void swap_free(swp_entry_t entry)
1339 struct swap_info_struct *p;
1341 p = _swap_info_get(entry);
1343 __swap_entry_free(p, entry);
1347 * Called after dropping swapcache to decrease refcnt to swap entries.
1349 void put_swap_page(struct page *page, swp_entry_t entry)
1351 unsigned long offset = swp_offset(entry);
1352 unsigned long idx = offset / SWAPFILE_CLUSTER;
1353 struct swap_cluster_info *ci;
1354 struct swap_info_struct *si;
1356 unsigned int i, free_entries = 0;
1358 int size = swap_entry_size(thp_nr_pages(page));
1360 si = _swap_info_get(entry);
1364 ci = lock_cluster_or_swap_info(si, offset);
1365 if (size == SWAPFILE_CLUSTER) {
1366 VM_BUG_ON(!cluster_is_huge(ci));
1367 map = si->swap_map + offset;
1368 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1370 VM_BUG_ON(!(val & SWAP_HAS_CACHE));
1371 if (val == SWAP_HAS_CACHE)
1374 cluster_clear_huge(ci);
1375 if (free_entries == SWAPFILE_CLUSTER) {
1376 unlock_cluster_or_swap_info(si, ci);
1377 spin_lock(&si->lock);
1378 mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER);
1379 swap_free_cluster(si, idx);
1380 spin_unlock(&si->lock);
1384 for (i = 0; i < size; i++, entry.val++) {
1385 if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) {
1386 unlock_cluster_or_swap_info(si, ci);
1387 free_swap_slot(entry);
1390 lock_cluster_or_swap_info(si, offset);
1393 unlock_cluster_or_swap_info(si, ci);
1396 #ifdef CONFIG_THP_SWAP
1397 int split_swap_cluster(swp_entry_t entry)
1399 struct swap_info_struct *si;
1400 struct swap_cluster_info *ci;
1401 unsigned long offset = swp_offset(entry);
1403 si = _swap_info_get(entry);
1406 ci = lock_cluster(si, offset);
1407 cluster_clear_huge(ci);
1413 static int swp_entry_cmp(const void *ent1, const void *ent2)
1415 const swp_entry_t *e1 = ent1, *e2 = ent2;
1417 return (int)swp_type(*e1) - (int)swp_type(*e2);
1420 void swapcache_free_entries(swp_entry_t *entries, int n)
1422 struct swap_info_struct *p, *prev;
1432 * Sort swap entries by swap device, so each lock is only taken once.
1433 * nr_swapfiles isn't absolutely correct, but the overhead of sort() is
1434 * so low that it isn't necessary to optimize further.
1436 if (nr_swapfiles > 1)
1437 sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL);
1438 for (i = 0; i < n; ++i) {
1439 p = swap_info_get_cont(entries[i], prev);
1441 swap_entry_free(p, entries[i]);
1445 spin_unlock(&p->lock);
1449 * How many references to page are currently swapped out?
1450 * This does not give an exact answer when swap count is continued,
1451 * but does include the high COUNT_CONTINUED flag to allow for that.
1453 int page_swapcount(struct page *page)
1456 struct swap_info_struct *p;
1457 struct swap_cluster_info *ci;
1459 unsigned long offset;
1461 entry.val = page_private(page);
1462 p = _swap_info_get(entry);
1464 offset = swp_offset(entry);
1465 ci = lock_cluster_or_swap_info(p, offset);
1466 count = swap_count(p->swap_map[offset]);
1467 unlock_cluster_or_swap_info(p, ci);
1472 int __swap_count(swp_entry_t entry)
1474 struct swap_info_struct *si;
1475 pgoff_t offset = swp_offset(entry);
1478 si = get_swap_device(entry);
1480 count = swap_count(si->swap_map[offset]);
1481 put_swap_device(si);
1486 static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry)
1489 pgoff_t offset = swp_offset(entry);
1490 struct swap_cluster_info *ci;
1492 ci = lock_cluster_or_swap_info(si, offset);
1493 count = swap_count(si->swap_map[offset]);
1494 unlock_cluster_or_swap_info(si, ci);
1499 * How many references to @entry are currently swapped out?
1500 * This does not give an exact answer when swap count is continued,
1501 * but does include the high COUNT_CONTINUED flag to allow for that.
1503 int __swp_swapcount(swp_entry_t entry)
1506 struct swap_info_struct *si;
1508 si = get_swap_device(entry);
1510 count = swap_swapcount(si, entry);
1511 put_swap_device(si);
1517 * How many references to @entry are currently swapped out?
1518 * This considers COUNT_CONTINUED so it returns exact answer.
1520 int swp_swapcount(swp_entry_t entry)
1522 int count, tmp_count, n;
1523 struct swap_info_struct *p;
1524 struct swap_cluster_info *ci;
1529 p = _swap_info_get(entry);
1533 offset = swp_offset(entry);
1535 ci = lock_cluster_or_swap_info(p, offset);
1537 count = swap_count(p->swap_map[offset]);
1538 if (!(count & COUNT_CONTINUED))
1541 count &= ~COUNT_CONTINUED;
1542 n = SWAP_MAP_MAX + 1;
1544 page = vmalloc_to_page(p->swap_map + offset);
1545 offset &= ~PAGE_MASK;
1546 VM_BUG_ON(page_private(page) != SWP_CONTINUED);
1549 page = list_next_entry(page, lru);
1550 map = kmap_atomic(page);
1551 tmp_count = map[offset];
1554 count += (tmp_count & ~COUNT_CONTINUED) * n;
1555 n *= (SWAP_CONT_MAX + 1);
1556 } while (tmp_count & COUNT_CONTINUED);
1558 unlock_cluster_or_swap_info(p, ci);
1562 static bool swap_page_trans_huge_swapped(struct swap_info_struct *si,
1565 struct swap_cluster_info *ci;
1566 unsigned char *map = si->swap_map;
1567 unsigned long roffset = swp_offset(entry);
1568 unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER);
1572 ci = lock_cluster_or_swap_info(si, offset);
1573 if (!ci || !cluster_is_huge(ci)) {
1574 if (swap_count(map[roffset]))
1578 for (i = 0; i < SWAPFILE_CLUSTER; i++) {
1579 if (swap_count(map[offset + i])) {
1585 unlock_cluster_or_swap_info(si, ci);
1589 static bool page_swapped(struct page *page)
1592 struct swap_info_struct *si;
1594 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page)))
1595 return page_swapcount(page) != 0;
1597 page = compound_head(page);
1598 entry.val = page_private(page);
1599 si = _swap_info_get(entry);
1601 return swap_page_trans_huge_swapped(si, entry);
1605 static int page_trans_huge_map_swapcount(struct page *page, int *total_mapcount,
1606 int *total_swapcount)
1608 int i, map_swapcount, _total_mapcount, _total_swapcount;
1609 unsigned long offset = 0;
1610 struct swap_info_struct *si;
1611 struct swap_cluster_info *ci = NULL;
1612 unsigned char *map = NULL;
1613 int mapcount, swapcount = 0;
1615 /* hugetlbfs shouldn't call it */
1616 VM_BUG_ON_PAGE(PageHuge(page), page);
1618 if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) {
1619 mapcount = page_trans_huge_mapcount(page, total_mapcount);
1620 if (PageSwapCache(page))
1621 swapcount = page_swapcount(page);
1622 if (total_swapcount)
1623 *total_swapcount = swapcount;
1624 return mapcount + swapcount;
1627 page = compound_head(page);
1629 _total_mapcount = _total_swapcount = map_swapcount = 0;
1630 if (PageSwapCache(page)) {
1633 entry.val = page_private(page);
1634 si = _swap_info_get(entry);
1637 offset = swp_offset(entry);
1641 ci = lock_cluster(si, offset);
1642 for (i = 0; i < HPAGE_PMD_NR; i++) {
1643 mapcount = atomic_read(&page[i]._mapcount) + 1;
1644 _total_mapcount += mapcount;
1646 swapcount = swap_count(map[offset + i]);
1647 _total_swapcount += swapcount;
1649 map_swapcount = max(map_swapcount, mapcount + swapcount);
1652 if (PageDoubleMap(page)) {
1654 _total_mapcount -= HPAGE_PMD_NR;
1656 mapcount = compound_mapcount(page);
1657 map_swapcount += mapcount;
1658 _total_mapcount += mapcount;
1660 *total_mapcount = _total_mapcount;
1661 if (total_swapcount)
1662 *total_swapcount = _total_swapcount;
1664 return map_swapcount;
1668 * We can write to an anon page without COW if there are no other references
1669 * to it. And as a side-effect, free up its swap: because the old content
1670 * on disk will never be read, and seeking back there to write new content
1671 * later would only waste time away from clustering.
1673 * NOTE: total_map_swapcount should not be relied upon by the caller if
1674 * reuse_swap_page() returns false, but it may be always overwritten
1675 * (see the other implementation for CONFIG_SWAP=n).
1677 bool reuse_swap_page(struct page *page, int *total_map_swapcount)
1679 int count, total_mapcount, total_swapcount;
1681 VM_BUG_ON_PAGE(!PageLocked(page), page);
1682 if (unlikely(PageKsm(page)))
1684 count = page_trans_huge_map_swapcount(page, &total_mapcount,
1686 if (total_map_swapcount)
1687 *total_map_swapcount = total_mapcount + total_swapcount;
1688 if (count == 1 && PageSwapCache(page) &&
1689 (likely(!PageTransCompound(page)) ||
1690 /* The remaining swap count will be freed soon */
1691 total_swapcount == page_swapcount(page))) {
1692 if (!PageWriteback(page)) {
1693 page = compound_head(page);
1694 delete_from_swap_cache(page);
1698 struct swap_info_struct *p;
1700 entry.val = page_private(page);
1701 p = swap_info_get(entry);
1702 if (p->flags & SWP_STABLE_WRITES) {
1703 spin_unlock(&p->lock);
1706 spin_unlock(&p->lock);
1714 * If swap is getting full, or if there are no more mappings of this page,
1715 * then try_to_free_swap is called to free its swap space.
1717 int try_to_free_swap(struct page *page)
1719 VM_BUG_ON_PAGE(!PageLocked(page), page);
1721 if (!PageSwapCache(page))
1723 if (PageWriteback(page))
1725 if (page_swapped(page))
1729 * Once hibernation has begun to create its image of memory,
1730 * there's a danger that one of the calls to try_to_free_swap()
1731 * - most probably a call from __try_to_reclaim_swap() while
1732 * hibernation is allocating its own swap pages for the image,
1733 * but conceivably even a call from memory reclaim - will free
1734 * the swap from a page which has already been recorded in the
1735 * image as a clean swapcache page, and then reuse its swap for
1736 * another page of the image. On waking from hibernation, the
1737 * original page might be freed under memory pressure, then
1738 * later read back in from swap, now with the wrong data.
1740 * Hibernation suspends storage while it is writing the image
1741 * to disk so check that here.
1743 if (pm_suspended_storage())
1746 page = compound_head(page);
1747 delete_from_swap_cache(page);
1753 * Free the swap entry like above, but also try to
1754 * free the page cache entry if it is the last user.
1756 int free_swap_and_cache(swp_entry_t entry)
1758 struct swap_info_struct *p;
1759 unsigned char count;
1761 if (non_swap_entry(entry))
1764 p = _swap_info_get(entry);
1766 count = __swap_entry_free(p, entry);
1767 if (count == SWAP_HAS_CACHE &&
1768 !swap_page_trans_huge_swapped(p, entry))
1769 __try_to_reclaim_swap(p, swp_offset(entry),
1770 TTRS_UNMAPPED | TTRS_FULL);
1775 #ifdef CONFIG_HIBERNATION
1777 swp_entry_t get_swap_page_of_type(int type)
1779 struct swap_info_struct *si = swap_type_to_swap_info(type);
1780 swp_entry_t entry = {0};
1785 /* This is called for allocating swap entry, not cache */
1786 spin_lock(&si->lock);
1787 if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry))
1788 atomic_long_dec(&nr_swap_pages);
1789 spin_unlock(&si->lock);
1795 * Find the swap type that corresponds to given device (if any).
1797 * @offset - number of the PAGE_SIZE-sized block of the device, starting
1798 * from 0, in which the swap header is expected to be located.
1800 * This is needed for the suspend to disk (aka swsusp).
1802 int swap_type_of(dev_t device, sector_t offset)
1809 spin_lock(&swap_lock);
1810 for (type = 0; type < nr_swapfiles; type++) {
1811 struct swap_info_struct *sis = swap_info[type];
1813 if (!(sis->flags & SWP_WRITEOK))
1816 if (device == sis->bdev->bd_dev) {
1817 struct swap_extent *se = first_se(sis);
1819 if (se->start_block == offset) {
1820 spin_unlock(&swap_lock);
1825 spin_unlock(&swap_lock);
1829 int find_first_swap(dev_t *device)
1833 spin_lock(&swap_lock);
1834 for (type = 0; type < nr_swapfiles; type++) {
1835 struct swap_info_struct *sis = swap_info[type];
1837 if (!(sis->flags & SWP_WRITEOK))
1839 *device = sis->bdev->bd_dev;
1840 spin_unlock(&swap_lock);
1843 spin_unlock(&swap_lock);
1848 * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev
1849 * corresponding to given index in swap_info (swap type).
1851 sector_t swapdev_block(int type, pgoff_t offset)
1853 struct swap_info_struct *si = swap_type_to_swap_info(type);
1854 struct swap_extent *se;
1856 if (!si || !(si->flags & SWP_WRITEOK))
1858 se = offset_to_swap_extent(si, offset);
1859 return se->start_block + (offset - se->start_page);
1863 * Return either the total number of swap pages of given type, or the number
1864 * of free pages of that type (depending on @free)
1866 * This is needed for software suspend
1868 unsigned int count_swap_pages(int type, int free)
1872 spin_lock(&swap_lock);
1873 if ((unsigned int)type < nr_swapfiles) {
1874 struct swap_info_struct *sis = swap_info[type];
1876 spin_lock(&sis->lock);
1877 if (sis->flags & SWP_WRITEOK) {
1880 n -= sis->inuse_pages;
1882 spin_unlock(&sis->lock);
1884 spin_unlock(&swap_lock);
1887 #endif /* CONFIG_HIBERNATION */
1889 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
1891 return pte_same(pte_swp_clear_flags(pte), swp_pte);
1895 * No need to decide whether this PTE shares the swap entry with others,
1896 * just let do_wp_page work it out if a write is requested later - to
1897 * force COW, vm_page_prot omits write permission from any private vma.
1899 static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd,
1900 unsigned long addr, swp_entry_t entry, struct page *page)
1902 struct page *swapcache;
1908 page = ksm_might_need_to_copy(page, vma, addr);
1909 if (unlikely(!page))
1912 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1913 if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) {
1918 dec_mm_counter(vma->vm_mm, MM_SWAPENTS);
1919 inc_mm_counter(vma->vm_mm, MM_ANONPAGES);
1921 set_pte_at(vma->vm_mm, addr, pte,
1922 pte_mkold(mk_pte(page, vma->vm_page_prot)));
1923 if (page == swapcache) {
1924 page_add_anon_rmap(page, vma, addr, false);
1925 } else { /* ksm created a completely new copy */
1926 page_add_new_anon_rmap(page, vma, addr, false);
1927 lru_cache_add_inactive_or_unevictable(page, vma);
1931 pte_unmap_unlock(pte, ptl);
1932 if (page != swapcache) {
1939 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
1940 unsigned long addr, unsigned long end,
1941 unsigned int type, bool frontswap,
1942 unsigned long *fs_pages_to_unuse)
1947 struct swap_info_struct *si;
1948 unsigned long offset;
1950 volatile unsigned char *swap_map;
1952 si = swap_info[type];
1953 pte = pte_offset_map(pmd, addr);
1955 if (!is_swap_pte(*pte))
1958 entry = pte_to_swp_entry(*pte);
1959 if (swp_type(entry) != type)
1962 offset = swp_offset(entry);
1963 if (frontswap && !frontswap_test(si, offset))
1967 swap_map = &si->swap_map[offset];
1968 page = lookup_swap_cache(entry, vma, addr);
1970 struct vm_fault vmf = {
1976 page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
1980 if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD)
1986 wait_on_page_writeback(page);
1987 ret = unuse_pte(vma, pmd, addr, entry, page);
1994 try_to_free_swap(page);
1998 if (*fs_pages_to_unuse && !--(*fs_pages_to_unuse)) {
1999 ret = FRONTSWAP_PAGES_UNUSED;
2003 pte = pte_offset_map(pmd, addr);
2004 } while (pte++, addr += PAGE_SIZE, addr != end);
2012 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
2013 unsigned long addr, unsigned long end,
2014 unsigned int type, bool frontswap,
2015 unsigned long *fs_pages_to_unuse)
2021 pmd = pmd_offset(pud, addr);
2024 next = pmd_addr_end(addr, end);
2025 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
2027 ret = unuse_pte_range(vma, pmd, addr, next, type,
2028 frontswap, fs_pages_to_unuse);
2031 } while (pmd++, addr = next, addr != end);
2035 static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d,
2036 unsigned long addr, unsigned long end,
2037 unsigned int type, bool frontswap,
2038 unsigned long *fs_pages_to_unuse)
2044 pud = pud_offset(p4d, addr);
2046 next = pud_addr_end(addr, end);
2047 if (pud_none_or_clear_bad(pud))
2049 ret = unuse_pmd_range(vma, pud, addr, next, type,
2050 frontswap, fs_pages_to_unuse);
2053 } while (pud++, addr = next, addr != end);
2057 static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd,
2058 unsigned long addr, unsigned long end,
2059 unsigned int type, bool frontswap,
2060 unsigned long *fs_pages_to_unuse)
2066 p4d = p4d_offset(pgd, addr);
2068 next = p4d_addr_end(addr, end);
2069 if (p4d_none_or_clear_bad(p4d))
2071 ret = unuse_pud_range(vma, p4d, addr, next, type,
2072 frontswap, fs_pages_to_unuse);
2075 } while (p4d++, addr = next, addr != end);
2079 static int unuse_vma(struct vm_area_struct *vma, unsigned int type,
2080 bool frontswap, unsigned long *fs_pages_to_unuse)
2083 unsigned long addr, end, next;
2086 addr = vma->vm_start;
2089 pgd = pgd_offset(vma->vm_mm, addr);
2091 next = pgd_addr_end(addr, end);
2092 if (pgd_none_or_clear_bad(pgd))
2094 ret = unuse_p4d_range(vma, pgd, addr, next, type,
2095 frontswap, fs_pages_to_unuse);
2098 } while (pgd++, addr = next, addr != end);
2102 static int unuse_mm(struct mm_struct *mm, unsigned int type,
2103 bool frontswap, unsigned long *fs_pages_to_unuse)
2105 struct vm_area_struct *vma;
2109 for (vma = mm->mmap; vma; vma = vma->vm_next) {
2110 if (vma->anon_vma) {
2111 ret = unuse_vma(vma, type, frontswap,
2118 mmap_read_unlock(mm);
2123 * Scan swap_map (or frontswap_map if frontswap parameter is true)
2124 * from current position to next entry still in use. Return 0
2125 * if there are no inuse entries after prev till end of the map.
2127 static unsigned int find_next_to_unuse(struct swap_info_struct *si,
2128 unsigned int prev, bool frontswap)
2131 unsigned char count;
2134 * No need for swap_lock here: we're just looking
2135 * for whether an entry is in use, not modifying it; false
2136 * hits are okay, and sys_swapoff() has already prevented new
2137 * allocations from this area (while holding swap_lock).
2139 for (i = prev + 1; i < si->max; i++) {
2140 count = READ_ONCE(si->swap_map[i]);
2141 if (count && swap_count(count) != SWAP_MAP_BAD)
2142 if (!frontswap || frontswap_test(si, i))
2144 if ((i % LATENCY_LIMIT) == 0)
2155 * If the boolean frontswap is true, only unuse pages_to_unuse pages;
2156 * pages_to_unuse==0 means all pages; ignored if frontswap is false
2158 int try_to_unuse(unsigned int type, bool frontswap,
2159 unsigned long pages_to_unuse)
2161 struct mm_struct *prev_mm;
2162 struct mm_struct *mm;
2163 struct list_head *p;
2165 struct swap_info_struct *si = swap_info[type];
2170 if (!READ_ONCE(si->inuse_pages))
2177 retval = shmem_unuse(type, frontswap, &pages_to_unuse);
2184 spin_lock(&mmlist_lock);
2185 p = &init_mm.mmlist;
2186 while (READ_ONCE(si->inuse_pages) &&
2187 !signal_pending(current) &&
2188 (p = p->next) != &init_mm.mmlist) {
2190 mm = list_entry(p, struct mm_struct, mmlist);
2191 if (!mmget_not_zero(mm))
2193 spin_unlock(&mmlist_lock);
2196 retval = unuse_mm(mm, type, frontswap, &pages_to_unuse);
2204 * Make sure that we aren't completely killing
2205 * interactive performance.
2208 spin_lock(&mmlist_lock);
2210 spin_unlock(&mmlist_lock);
2215 while (READ_ONCE(si->inuse_pages) &&
2216 !signal_pending(current) &&
2217 (i = find_next_to_unuse(si, i, frontswap)) != 0) {
2219 entry = swp_entry(type, i);
2220 page = find_get_page(swap_address_space(entry), i);
2225 * It is conceivable that a racing task removed this page from
2226 * swap cache just before we acquired the page lock. The page
2227 * might even be back in swap cache on another swap area. But
2228 * that is okay, try_to_free_swap() only removes stale pages.
2231 wait_on_page_writeback(page);
2232 try_to_free_swap(page);
2237 * For frontswap, we just need to unuse pages_to_unuse, if
2238 * it was specified. Need not check frontswap again here as
2239 * we already zeroed out pages_to_unuse if not frontswap.
2241 if (pages_to_unuse && --pages_to_unuse == 0)
2246 * Lets check again to see if there are still swap entries in the map.
2247 * If yes, we would need to do retry the unuse logic again.
2248 * Under global memory pressure, swap entries can be reinserted back
2249 * into process space after the mmlist loop above passes over them.
2251 * Limit the number of retries? No: when mmget_not_zero() above fails,
2252 * that mm is likely to be freeing swap from exit_mmap(), which proceeds
2253 * at its own independent pace; and even shmem_writepage() could have
2254 * been preempted after get_swap_page(), temporarily hiding that swap.
2255 * It's easy and robust (though cpu-intensive) just to keep retrying.
2257 if (READ_ONCE(si->inuse_pages)) {
2258 if (!signal_pending(current))
2263 return (retval == FRONTSWAP_PAGES_UNUSED) ? 0 : retval;
2267 * After a successful try_to_unuse, if no swap is now in use, we know
2268 * we can empty the mmlist. swap_lock must be held on entry and exit.
2269 * Note that mmlist_lock nests inside swap_lock, and an mm must be
2270 * added to the mmlist just after page_duplicate - before would be racy.
2272 static void drain_mmlist(void)
2274 struct list_head *p, *next;
2277 for (type = 0; type < nr_swapfiles; type++)
2278 if (swap_info[type]->inuse_pages)
2280 spin_lock(&mmlist_lock);
2281 list_for_each_safe(p, next, &init_mm.mmlist)
2283 spin_unlock(&mmlist_lock);
2287 * Free all of a swapdev's extent information
2289 static void destroy_swap_extents(struct swap_info_struct *sis)
2291 while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) {
2292 struct rb_node *rb = sis->swap_extent_root.rb_node;
2293 struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node);
2295 rb_erase(rb, &sis->swap_extent_root);
2299 if (sis->flags & SWP_ACTIVATED) {
2300 struct file *swap_file = sis->swap_file;
2301 struct address_space *mapping = swap_file->f_mapping;
2303 sis->flags &= ~SWP_ACTIVATED;
2304 if (mapping->a_ops->swap_deactivate)
2305 mapping->a_ops->swap_deactivate(swap_file);
2310 * Add a block range (and the corresponding page range) into this swapdev's
2313 * This function rather assumes that it is called in ascending page order.
2316 add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
2317 unsigned long nr_pages, sector_t start_block)
2319 struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL;
2320 struct swap_extent *se;
2321 struct swap_extent *new_se;
2324 * place the new node at the right most since the
2325 * function is called in ascending page order.
2329 link = &parent->rb_right;
2333 se = rb_entry(parent, struct swap_extent, rb_node);
2334 BUG_ON(se->start_page + se->nr_pages != start_page);
2335 if (se->start_block + se->nr_pages == start_block) {
2337 se->nr_pages += nr_pages;
2342 /* No merge, insert a new extent. */
2343 new_se = kmalloc(sizeof(*se), GFP_KERNEL);
2346 new_se->start_page = start_page;
2347 new_se->nr_pages = nr_pages;
2348 new_se->start_block = start_block;
2350 rb_link_node(&new_se->rb_node, parent, link);
2351 rb_insert_color(&new_se->rb_node, &sis->swap_extent_root);
2354 EXPORT_SYMBOL_GPL(add_swap_extent);
2357 * A `swap extent' is a simple thing which maps a contiguous range of pages
2358 * onto a contiguous range of disk blocks. An ordered list of swap extents
2359 * is built at swapon time and is then used at swap_writepage/swap_readpage
2360 * time for locating where on disk a page belongs.
2362 * If the swapfile is an S_ISBLK block device, a single extent is installed.
2363 * This is done so that the main operating code can treat S_ISBLK and S_ISREG
2364 * swap files identically.
2366 * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap
2367 * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK
2368 * swapfiles are handled *identically* after swapon time.
2370 * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks
2371 * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If
2372 * some stray blocks are found which do not fall within the PAGE_SIZE alignment
2373 * requirements, they are simply tossed out - we will never use those blocks
2376 * For all swap devices we set S_SWAPFILE across the life of the swapon. This
2377 * prevents users from writing to the swap device, which will corrupt memory.
2379 * The amount of disk space which a single swap extent represents varies.
2380 * Typically it is in the 1-4 megabyte range. So we can have hundreds of
2381 * extents in the list. To avoid much list walking, we cache the previous
2382 * search location in `curr_swap_extent', and start new searches from there.
2383 * This is extremely effective. The average number of iterations in
2384 * map_swap_page() has been measured at about 0.3 per page. - akpm.
2386 static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span)
2388 struct file *swap_file = sis->swap_file;
2389 struct address_space *mapping = swap_file->f_mapping;
2390 struct inode *inode = mapping->host;
2393 if (S_ISBLK(inode->i_mode)) {
2394 ret = add_swap_extent(sis, 0, sis->max, 0);
2399 if (mapping->a_ops->swap_activate) {
2400 ret = mapping->a_ops->swap_activate(sis, swap_file, span);
2402 sis->flags |= SWP_ACTIVATED;
2404 sis->flags |= SWP_FS_OPS;
2405 ret = add_swap_extent(sis, 0, sis->max, 0);
2411 return generic_swapfile_activate(sis, swap_file, span);
2414 static int swap_node(struct swap_info_struct *p)
2416 struct block_device *bdev;
2421 bdev = p->swap_file->f_inode->i_sb->s_bdev;
2423 return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE;
2426 static void setup_swap_info(struct swap_info_struct *p, int prio,
2427 unsigned char *swap_map,
2428 struct swap_cluster_info *cluster_info)
2435 p->prio = --least_priority;
2437 * the plist prio is negated because plist ordering is
2438 * low-to-high, while swap ordering is high-to-low
2440 p->list.prio = -p->prio;
2443 p->avail_lists[i].prio = -p->prio;
2445 if (swap_node(p) == i)
2446 p->avail_lists[i].prio = 1;
2448 p->avail_lists[i].prio = -p->prio;
2451 p->swap_map = swap_map;
2452 p->cluster_info = cluster_info;
2455 static void _enable_swap_info(struct swap_info_struct *p)
2457 p->flags |= SWP_WRITEOK;
2458 atomic_long_add(p->pages, &nr_swap_pages);
2459 total_swap_pages += p->pages;
2461 assert_spin_locked(&swap_lock);
2463 * both lists are plists, and thus priority ordered.
2464 * swap_active_head needs to be priority ordered for swapoff(),
2465 * which on removal of any swap_info_struct with an auto-assigned
2466 * (i.e. negative) priority increments the auto-assigned priority
2467 * of any lower-priority swap_info_structs.
2468 * swap_avail_head needs to be priority ordered for get_swap_page(),
2469 * which allocates swap pages from the highest available priority
2472 plist_add(&p->list, &swap_active_head);
2473 add_to_avail_list(p);
2476 static void enable_swap_info(struct swap_info_struct *p, int prio,
2477 unsigned char *swap_map,
2478 struct swap_cluster_info *cluster_info,
2479 unsigned long *frontswap_map)
2481 frontswap_init(p->type, frontswap_map);
2482 spin_lock(&swap_lock);
2483 spin_lock(&p->lock);
2484 setup_swap_info(p, prio, swap_map, cluster_info);
2485 spin_unlock(&p->lock);
2486 spin_unlock(&swap_lock);
2488 * Finished initializing swap device, now it's safe to reference it.
2490 percpu_ref_resurrect(&p->users);
2491 spin_lock(&swap_lock);
2492 spin_lock(&p->lock);
2493 _enable_swap_info(p);
2494 spin_unlock(&p->lock);
2495 spin_unlock(&swap_lock);
2498 static void reinsert_swap_info(struct swap_info_struct *p)
2500 spin_lock(&swap_lock);
2501 spin_lock(&p->lock);
2502 setup_swap_info(p, p->prio, p->swap_map, p->cluster_info);
2503 _enable_swap_info(p);
2504 spin_unlock(&p->lock);
2505 spin_unlock(&swap_lock);
2508 bool has_usable_swap(void)
2512 spin_lock(&swap_lock);
2513 if (plist_head_empty(&swap_active_head))
2515 spin_unlock(&swap_lock);
2519 SYSCALL_DEFINE1(swapoff, const char __user *, specialfile)
2521 struct swap_info_struct *p = NULL;
2522 unsigned char *swap_map;
2523 struct swap_cluster_info *cluster_info;
2524 unsigned long *frontswap_map;
2525 struct file *swap_file, *victim;
2526 struct address_space *mapping;
2527 struct inode *inode;
2528 struct filename *pathname;
2530 unsigned int old_block_size;
2532 if (!capable(CAP_SYS_ADMIN))
2535 BUG_ON(!current->mm);
2537 pathname = getname(specialfile);
2538 if (IS_ERR(pathname))
2539 return PTR_ERR(pathname);
2541 victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0);
2542 err = PTR_ERR(victim);
2546 mapping = victim->f_mapping;
2547 spin_lock(&swap_lock);
2548 plist_for_each_entry(p, &swap_active_head, list) {
2549 if (p->flags & SWP_WRITEOK) {
2550 if (p->swap_file->f_mapping == mapping) {
2558 spin_unlock(&swap_lock);
2561 if (!security_vm_enough_memory_mm(current->mm, p->pages))
2562 vm_unacct_memory(p->pages);
2565 spin_unlock(&swap_lock);
2568 del_from_avail_list(p);
2569 spin_lock(&p->lock);
2571 struct swap_info_struct *si = p;
2574 plist_for_each_entry_continue(si, &swap_active_head, list) {
2577 for_each_node(nid) {
2578 if (si->avail_lists[nid].prio != 1)
2579 si->avail_lists[nid].prio--;
2584 plist_del(&p->list, &swap_active_head);
2585 atomic_long_sub(p->pages, &nr_swap_pages);
2586 total_swap_pages -= p->pages;
2587 p->flags &= ~SWP_WRITEOK;
2588 spin_unlock(&p->lock);
2589 spin_unlock(&swap_lock);
2591 disable_swap_slots_cache_lock();
2593 set_current_oom_origin();
2594 err = try_to_unuse(p->type, false, 0); /* force unuse all pages */
2595 clear_current_oom_origin();
2598 /* re-insert swap space back into swap_list */
2599 reinsert_swap_info(p);
2600 reenable_swap_slots_cache_unlock();
2604 reenable_swap_slots_cache_unlock();
2607 * Wait for swap operations protected by get/put_swap_device()
2610 * We need synchronize_rcu() here to protect the accessing to
2611 * the swap cache data structure.
2613 percpu_ref_kill(&p->users);
2615 wait_for_completion(&p->comp);
2617 flush_work(&p->discard_work);
2619 destroy_swap_extents(p);
2620 if (p->flags & SWP_CONTINUED)
2621 free_swap_count_continuations(p);
2623 if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev)))
2624 atomic_dec(&nr_rotate_swap);
2626 mutex_lock(&swapon_mutex);
2627 spin_lock(&swap_lock);
2628 spin_lock(&p->lock);
2631 /* wait for anyone still in scan_swap_map_slots */
2632 p->highest_bit = 0; /* cuts scans short */
2633 while (p->flags >= SWP_SCANNING) {
2634 spin_unlock(&p->lock);
2635 spin_unlock(&swap_lock);
2636 schedule_timeout_uninterruptible(1);
2637 spin_lock(&swap_lock);
2638 spin_lock(&p->lock);
2641 swap_file = p->swap_file;
2642 old_block_size = p->old_block_size;
2643 p->swap_file = NULL;
2645 swap_map = p->swap_map;
2647 cluster_info = p->cluster_info;
2648 p->cluster_info = NULL;
2649 frontswap_map = frontswap_map_get(p);
2650 spin_unlock(&p->lock);
2651 spin_unlock(&swap_lock);
2652 arch_swap_invalidate_area(p->type);
2653 frontswap_invalidate_area(p->type);
2654 frontswap_map_set(p, NULL);
2655 mutex_unlock(&swapon_mutex);
2656 free_percpu(p->percpu_cluster);
2657 p->percpu_cluster = NULL;
2658 free_percpu(p->cluster_next_cpu);
2659 p->cluster_next_cpu = NULL;
2661 kvfree(cluster_info);
2662 kvfree(frontswap_map);
2663 /* Destroy swap account information */
2664 swap_cgroup_swapoff(p->type);
2665 exit_swap_address_space(p->type);
2667 inode = mapping->host;
2668 if (S_ISBLK(inode->i_mode)) {
2669 struct block_device *bdev = I_BDEV(inode);
2671 set_blocksize(bdev, old_block_size);
2672 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
2676 inode->i_flags &= ~S_SWAPFILE;
2677 inode_unlock(inode);
2678 filp_close(swap_file, NULL);
2681 * Clear the SWP_USED flag after all resources are freed so that swapon
2682 * can reuse this swap_info in alloc_swap_info() safely. It is ok to
2683 * not hold p->lock after we cleared its SWP_WRITEOK.
2685 spin_lock(&swap_lock);
2687 spin_unlock(&swap_lock);
2690 atomic_inc(&proc_poll_event);
2691 wake_up_interruptible(&proc_poll_wait);
2694 filp_close(victim, NULL);
2700 #ifdef CONFIG_PROC_FS
2701 static __poll_t swaps_poll(struct file *file, poll_table *wait)
2703 struct seq_file *seq = file->private_data;
2705 poll_wait(file, &proc_poll_wait, wait);
2707 if (seq->poll_event != atomic_read(&proc_poll_event)) {
2708 seq->poll_event = atomic_read(&proc_poll_event);
2709 return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI;
2712 return EPOLLIN | EPOLLRDNORM;
2716 static void *swap_start(struct seq_file *swap, loff_t *pos)
2718 struct swap_info_struct *si;
2722 mutex_lock(&swapon_mutex);
2725 return SEQ_START_TOKEN;
2727 for (type = 0; (si = swap_type_to_swap_info(type)); type++) {
2728 if (!(si->flags & SWP_USED) || !si->swap_map)
2737 static void *swap_next(struct seq_file *swap, void *v, loff_t *pos)
2739 struct swap_info_struct *si = v;
2742 if (v == SEQ_START_TOKEN)
2745 type = si->type + 1;
2748 for (; (si = swap_type_to_swap_info(type)); type++) {
2749 if (!(si->flags & SWP_USED) || !si->swap_map)
2757 static void swap_stop(struct seq_file *swap, void *v)
2759 mutex_unlock(&swapon_mutex);
2762 static int swap_show(struct seq_file *swap, void *v)
2764 struct swap_info_struct *si = v;
2767 unsigned int bytes, inuse;
2769 if (si == SEQ_START_TOKEN) {
2770 seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n");
2774 bytes = si->pages << (PAGE_SHIFT - 10);
2775 inuse = si->inuse_pages << (PAGE_SHIFT - 10);
2777 file = si->swap_file;
2778 len = seq_file_path(swap, file, " \t\n\\");
2779 seq_printf(swap, "%*s%s\t%u\t%s%u\t%s%d\n",
2780 len < 40 ? 40 - len : 1, " ",
2781 S_ISBLK(file_inode(file)->i_mode) ?
2782 "partition" : "file\t",
2783 bytes, bytes < 10000000 ? "\t" : "",
2784 inuse, inuse < 10000000 ? "\t" : "",
2789 static const struct seq_operations swaps_op = {
2790 .start = swap_start,
2796 static int swaps_open(struct inode *inode, struct file *file)
2798 struct seq_file *seq;
2801 ret = seq_open(file, &swaps_op);
2805 seq = file->private_data;
2806 seq->poll_event = atomic_read(&proc_poll_event);
2810 static const struct proc_ops swaps_proc_ops = {
2811 .proc_flags = PROC_ENTRY_PERMANENT,
2812 .proc_open = swaps_open,
2813 .proc_read = seq_read,
2814 .proc_lseek = seq_lseek,
2815 .proc_release = seq_release,
2816 .proc_poll = swaps_poll,
2819 static int __init procswaps_init(void)
2821 proc_create("swaps", 0, NULL, &swaps_proc_ops);
2824 __initcall(procswaps_init);
2825 #endif /* CONFIG_PROC_FS */
2827 #ifdef MAX_SWAPFILES_CHECK
2828 static int __init max_swapfiles_check(void)
2830 MAX_SWAPFILES_CHECK();
2833 late_initcall(max_swapfiles_check);
2836 static struct swap_info_struct *alloc_swap_info(void)
2838 struct swap_info_struct *p;
2839 struct swap_info_struct *defer = NULL;
2843 p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL);
2845 return ERR_PTR(-ENOMEM);
2847 if (percpu_ref_init(&p->users, swap_users_ref_free,
2848 PERCPU_REF_INIT_DEAD, GFP_KERNEL)) {
2850 return ERR_PTR(-ENOMEM);
2853 spin_lock(&swap_lock);
2854 for (type = 0; type < nr_swapfiles; type++) {
2855 if (!(swap_info[type]->flags & SWP_USED))
2858 if (type >= MAX_SWAPFILES) {
2859 spin_unlock(&swap_lock);
2860 percpu_ref_exit(&p->users);
2862 return ERR_PTR(-EPERM);
2864 if (type >= nr_swapfiles) {
2866 WRITE_ONCE(swap_info[type], p);
2868 * Write swap_info[type] before nr_swapfiles, in case a
2869 * racing procfs swap_start() or swap_next() is reading them.
2870 * (We never shrink nr_swapfiles, we never free this entry.)
2873 WRITE_ONCE(nr_swapfiles, nr_swapfiles + 1);
2876 p = swap_info[type];
2878 * Do not memset this entry: a racing procfs swap_next()
2879 * would be relying on p->type to remain valid.
2882 p->swap_extent_root = RB_ROOT;
2883 plist_node_init(&p->list, 0);
2885 plist_node_init(&p->avail_lists[i], 0);
2886 p->flags = SWP_USED;
2887 spin_unlock(&swap_lock);
2889 percpu_ref_exit(&defer->users);
2892 spin_lock_init(&p->lock);
2893 spin_lock_init(&p->cont_lock);
2894 init_completion(&p->comp);
2899 static int claim_swapfile(struct swap_info_struct *p, struct inode *inode)
2903 if (S_ISBLK(inode->i_mode)) {
2904 p->bdev = blkdev_get_by_dev(inode->i_rdev,
2905 FMODE_READ | FMODE_WRITE | FMODE_EXCL, p);
2906 if (IS_ERR(p->bdev)) {
2907 error = PTR_ERR(p->bdev);
2911 p->old_block_size = block_size(p->bdev);
2912 error = set_blocksize(p->bdev, PAGE_SIZE);
2916 * Zoned block devices contain zones that have a sequential
2917 * write only restriction. Hence zoned block devices are not
2918 * suitable for swapping. Disallow them here.
2920 if (blk_queue_is_zoned(p->bdev->bd_disk->queue))
2922 p->flags |= SWP_BLKDEV;
2923 } else if (S_ISREG(inode->i_mode)) {
2924 p->bdev = inode->i_sb->s_bdev;
2932 * Find out how many pages are allowed for a single swap device. There
2933 * are two limiting factors:
2934 * 1) the number of bits for the swap offset in the swp_entry_t type, and
2935 * 2) the number of bits in the swap pte, as defined by the different
2938 * In order to find the largest possible bit mask, a swap entry with
2939 * swap type 0 and swap offset ~0UL is created, encoded to a swap pte,
2940 * decoded to a swp_entry_t again, and finally the swap offset is
2943 * This will mask all the bits from the initial ~0UL mask that can't
2944 * be encoded in either the swp_entry_t or the architecture definition
2947 unsigned long generic_max_swapfile_size(void)
2949 return swp_offset(pte_to_swp_entry(
2950 swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1;
2953 /* Can be overridden by an architecture for additional checks. */
2954 __weak unsigned long max_swapfile_size(void)
2956 return generic_max_swapfile_size();
2959 static unsigned long read_swap_header(struct swap_info_struct *p,
2960 union swap_header *swap_header,
2961 struct inode *inode)
2964 unsigned long maxpages;
2965 unsigned long swapfilepages;
2966 unsigned long last_page;
2968 if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) {
2969 pr_err("Unable to find swap-space signature\n");
2973 /* swap partition endianess hack... */
2974 if (swab32(swap_header->info.version) == 1) {
2975 swab32s(&swap_header->info.version);
2976 swab32s(&swap_header->info.last_page);
2977 swab32s(&swap_header->info.nr_badpages);
2978 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
2980 for (i = 0; i < swap_header->info.nr_badpages; i++)
2981 swab32s(&swap_header->info.badpages[i]);
2983 /* Check the swap header's sub-version */
2984 if (swap_header->info.version != 1) {
2985 pr_warn("Unable to handle swap header version %d\n",
2986 swap_header->info.version);
2991 p->cluster_next = 1;
2994 maxpages = max_swapfile_size();
2995 last_page = swap_header->info.last_page;
2997 pr_warn("Empty swap-file\n");
3000 if (last_page > maxpages) {
3001 pr_warn("Truncating oversized swap area, only using %luk out of %luk\n",
3002 maxpages << (PAGE_SHIFT - 10),
3003 last_page << (PAGE_SHIFT - 10));
3005 if (maxpages > last_page) {
3006 maxpages = last_page + 1;
3007 /* p->max is an unsigned int: don't overflow it */
3008 if ((unsigned int)maxpages == 0)
3009 maxpages = UINT_MAX;
3011 p->highest_bit = maxpages - 1;
3015 swapfilepages = i_size_read(inode) >> PAGE_SHIFT;
3016 if (swapfilepages && maxpages > swapfilepages) {
3017 pr_warn("Swap area shorter than signature indicates\n");
3020 if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode))
3022 if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES)
3028 #define SWAP_CLUSTER_INFO_COLS \
3029 DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info))
3030 #define SWAP_CLUSTER_SPACE_COLS \
3031 DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER)
3032 #define SWAP_CLUSTER_COLS \
3033 max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS)
3035 static int setup_swap_map_and_extents(struct swap_info_struct *p,
3036 union swap_header *swap_header,
3037 unsigned char *swap_map,
3038 struct swap_cluster_info *cluster_info,
3039 unsigned long maxpages,
3043 unsigned int nr_good_pages;
3045 unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3046 unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS;
3047 unsigned long i, idx;
3049 nr_good_pages = maxpages - 1; /* omit header page */
3051 cluster_list_init(&p->free_clusters);
3052 cluster_list_init(&p->discard_clusters);
3054 for (i = 0; i < swap_header->info.nr_badpages; i++) {
3055 unsigned int page_nr = swap_header->info.badpages[i];
3056 if (page_nr == 0 || page_nr > swap_header->info.last_page)
3058 if (page_nr < maxpages) {
3059 swap_map[page_nr] = SWAP_MAP_BAD;
3062 * Haven't marked the cluster free yet, no list
3063 * operation involved
3065 inc_cluster_info_page(p, cluster_info, page_nr);
3069 /* Haven't marked the cluster free yet, no list operation involved */
3070 for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++)
3071 inc_cluster_info_page(p, cluster_info, i);
3073 if (nr_good_pages) {
3074 swap_map[0] = SWAP_MAP_BAD;
3076 * Not mark the cluster free yet, no list
3077 * operation involved
3079 inc_cluster_info_page(p, cluster_info, 0);
3081 p->pages = nr_good_pages;
3082 nr_extents = setup_swap_extents(p, span);
3085 nr_good_pages = p->pages;
3087 if (!nr_good_pages) {
3088 pr_warn("Empty swap-file\n");
3097 * Reduce false cache line sharing between cluster_info and
3098 * sharing same address space.
3100 for (k = 0; k < SWAP_CLUSTER_COLS; k++) {
3101 j = (k + col) % SWAP_CLUSTER_COLS;
3102 for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) {
3103 idx = i * SWAP_CLUSTER_COLS + j;
3104 if (idx >= nr_clusters)
3106 if (cluster_count(&cluster_info[idx]))
3108 cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE);
3109 cluster_list_add_tail(&p->free_clusters, cluster_info,
3117 * Helper to sys_swapon determining if a given swap
3118 * backing device queue supports DISCARD operations.
3120 static bool swap_discardable(struct swap_info_struct *si)
3122 struct request_queue *q = bdev_get_queue(si->bdev);
3124 if (!q || !blk_queue_discard(q))
3130 SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
3132 struct swap_info_struct *p;
3133 struct filename *name;
3134 struct file *swap_file = NULL;
3135 struct address_space *mapping;
3138 union swap_header *swap_header;
3141 unsigned long maxpages;
3142 unsigned char *swap_map = NULL;
3143 struct swap_cluster_info *cluster_info = NULL;
3144 unsigned long *frontswap_map = NULL;
3145 struct page *page = NULL;
3146 struct inode *inode = NULL;
3147 bool inced_nr_rotate_swap = false;
3149 if (swap_flags & ~SWAP_FLAGS_VALID)
3152 if (!capable(CAP_SYS_ADMIN))
3155 if (!swap_avail_heads)
3158 p = alloc_swap_info();
3162 INIT_WORK(&p->discard_work, swap_discard_work);
3164 name = getname(specialfile);
3166 error = PTR_ERR(name);
3170 swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0);
3171 if (IS_ERR(swap_file)) {
3172 error = PTR_ERR(swap_file);
3177 p->swap_file = swap_file;
3178 mapping = swap_file->f_mapping;
3179 inode = mapping->host;
3181 error = claim_swapfile(p, inode);
3182 if (unlikely(error))
3186 if (IS_SWAPFILE(inode)) {
3188 goto bad_swap_unlock_inode;
3192 * Read the swap header.
3194 if (!mapping->a_ops->readpage) {
3196 goto bad_swap_unlock_inode;
3198 page = read_mapping_page(mapping, 0, swap_file);
3200 error = PTR_ERR(page);
3201 goto bad_swap_unlock_inode;
3203 swap_header = kmap(page);
3205 maxpages = read_swap_header(p, swap_header, inode);
3206 if (unlikely(!maxpages)) {
3208 goto bad_swap_unlock_inode;
3211 /* OK, set up the swap map and apply the bad block list */
3212 swap_map = vzalloc(maxpages);
3215 goto bad_swap_unlock_inode;
3218 if (p->bdev && blk_queue_stable_writes(p->bdev->bd_disk->queue))
3219 p->flags |= SWP_STABLE_WRITES;
3221 if (p->bdev && p->bdev->bd_disk->fops->rw_page)
3222 p->flags |= SWP_SYNCHRONOUS_IO;
3224 if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) {
3226 unsigned long ci, nr_cluster;
3228 p->flags |= SWP_SOLIDSTATE;
3229 p->cluster_next_cpu = alloc_percpu(unsigned int);
3230 if (!p->cluster_next_cpu) {
3232 goto bad_swap_unlock_inode;
3235 * select a random position to start with to help wear leveling
3238 for_each_possible_cpu(cpu) {
3239 per_cpu(*p->cluster_next_cpu, cpu) =
3240 1 + prandom_u32_max(p->highest_bit);
3242 nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER);
3244 cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info),
3246 if (!cluster_info) {
3248 goto bad_swap_unlock_inode;
3251 for (ci = 0; ci < nr_cluster; ci++)
3252 spin_lock_init(&((cluster_info + ci)->lock));
3254 p->percpu_cluster = alloc_percpu(struct percpu_cluster);
3255 if (!p->percpu_cluster) {
3257 goto bad_swap_unlock_inode;
3259 for_each_possible_cpu(cpu) {
3260 struct percpu_cluster *cluster;
3261 cluster = per_cpu_ptr(p->percpu_cluster, cpu);
3262 cluster_set_null(&cluster->index);
3265 atomic_inc(&nr_rotate_swap);
3266 inced_nr_rotate_swap = true;
3269 error = swap_cgroup_swapon(p->type, maxpages);
3271 goto bad_swap_unlock_inode;
3273 nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map,
3274 cluster_info, maxpages, &span);
3275 if (unlikely(nr_extents < 0)) {
3277 goto bad_swap_unlock_inode;
3279 /* frontswap enabled? set up bit-per-page map for frontswap */
3280 if (IS_ENABLED(CONFIG_FRONTSWAP))
3281 frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages),
3285 if (p->bdev && (swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) {
3287 * When discard is enabled for swap with no particular
3288 * policy flagged, we set all swap discard flags here in
3289 * order to sustain backward compatibility with older
3290 * swapon(8) releases.
3292 p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD |
3296 * By flagging sys_swapon, a sysadmin can tell us to
3297 * either do single-time area discards only, or to just
3298 * perform discards for released swap page-clusters.
3299 * Now it's time to adjust the p->flags accordingly.
3301 if (swap_flags & SWAP_FLAG_DISCARD_ONCE)
3302 p->flags &= ~SWP_PAGE_DISCARD;
3303 else if (swap_flags & SWAP_FLAG_DISCARD_PAGES)
3304 p->flags &= ~SWP_AREA_DISCARD;
3306 /* issue a swapon-time discard if it's still required */
3307 if (p->flags & SWP_AREA_DISCARD) {
3308 int err = discard_swap(p);
3310 pr_err("swapon: discard_swap(%p): %d\n",
3315 error = init_swap_address_space(p->type, maxpages);
3317 goto bad_swap_unlock_inode;
3320 * Flush any pending IO and dirty mappings before we start using this
3323 inode->i_flags |= S_SWAPFILE;
3324 error = inode_drain_writes(inode);
3326 inode->i_flags &= ~S_SWAPFILE;
3327 goto free_swap_address_space;
3330 mutex_lock(&swapon_mutex);
3332 if (swap_flags & SWAP_FLAG_PREFER)
3334 (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT;
3335 enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map);
3337 pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n",
3338 p->pages<<(PAGE_SHIFT-10), name->name, p->prio,
3339 nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10),
3340 (p->flags & SWP_SOLIDSTATE) ? "SS" : "",
3341 (p->flags & SWP_DISCARDABLE) ? "D" : "",
3342 (p->flags & SWP_AREA_DISCARD) ? "s" : "",
3343 (p->flags & SWP_PAGE_DISCARD) ? "c" : "",
3344 (frontswap_map) ? "FS" : "");
3346 mutex_unlock(&swapon_mutex);
3347 atomic_inc(&proc_poll_event);
3348 wake_up_interruptible(&proc_poll_wait);
3352 free_swap_address_space:
3353 exit_swap_address_space(p->type);
3354 bad_swap_unlock_inode:
3355 inode_unlock(inode);
3357 free_percpu(p->percpu_cluster);
3358 p->percpu_cluster = NULL;
3359 free_percpu(p->cluster_next_cpu);
3360 p->cluster_next_cpu = NULL;
3361 if (inode && S_ISBLK(inode->i_mode) && p->bdev) {
3362 set_blocksize(p->bdev, p->old_block_size);
3363 blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
3366 destroy_swap_extents(p);
3367 swap_cgroup_swapoff(p->type);
3368 spin_lock(&swap_lock);
3369 p->swap_file = NULL;
3371 spin_unlock(&swap_lock);
3373 kvfree(cluster_info);
3374 kvfree(frontswap_map);
3375 if (inced_nr_rotate_swap)
3376 atomic_dec(&nr_rotate_swap);
3378 filp_close(swap_file, NULL);
3380 if (page && !IS_ERR(page)) {
3387 inode_unlock(inode);
3389 enable_swap_slots_cache();
3393 void si_swapinfo(struct sysinfo *val)
3396 unsigned long nr_to_be_unused = 0;
3398 spin_lock(&swap_lock);
3399 for (type = 0; type < nr_swapfiles; type++) {
3400 struct swap_info_struct *si = swap_info[type];
3402 if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK))
3403 nr_to_be_unused += si->inuse_pages;
3405 val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused;
3406 val->totalswap = total_swap_pages + nr_to_be_unused;
3407 spin_unlock(&swap_lock);
3411 * Verify that a swap entry is valid and increment its swap map count.
3413 * Returns error code in following case.
3415 * - swp_entry is invalid -> EINVAL
3416 * - swp_entry is migration entry -> EINVAL
3417 * - swap-cache reference is requested but there is already one. -> EEXIST
3418 * - swap-cache reference is requested but the entry is not used. -> ENOENT
3419 * - swap-mapped reference requested but needs continued swap count. -> ENOMEM
3421 static int __swap_duplicate(swp_entry_t entry, unsigned char usage)
3423 struct swap_info_struct *p;
3424 struct swap_cluster_info *ci;
3425 unsigned long offset;
3426 unsigned char count;
3427 unsigned char has_cache;
3430 p = get_swap_device(entry);
3434 offset = swp_offset(entry);
3435 ci = lock_cluster_or_swap_info(p, offset);
3437 count = p->swap_map[offset];
3440 * swapin_readahead() doesn't check if a swap entry is valid, so the
3441 * swap entry could be SWAP_MAP_BAD. Check here with lock held.
3443 if (unlikely(swap_count(count) == SWAP_MAP_BAD)) {
3448 has_cache = count & SWAP_HAS_CACHE;
3449 count &= ~SWAP_HAS_CACHE;
3452 if (usage == SWAP_HAS_CACHE) {
3454 /* set SWAP_HAS_CACHE if there is no cache and entry is used */
3455 if (!has_cache && count)
3456 has_cache = SWAP_HAS_CACHE;
3457 else if (has_cache) /* someone else added cache */
3459 else /* no users remaining */
3462 } else if (count || has_cache) {
3464 if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX)
3466 else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX)
3468 else if (swap_count_continued(p, offset, count))
3469 count = COUNT_CONTINUED;
3473 err = -ENOENT; /* unused swap entry */
3475 WRITE_ONCE(p->swap_map[offset], count | has_cache);
3478 unlock_cluster_or_swap_info(p, ci);
3485 * Help swapoff by noting that swap entry belongs to shmem/tmpfs
3486 * (in which case its reference count is never incremented).
3488 void swap_shmem_alloc(swp_entry_t entry)
3490 __swap_duplicate(entry, SWAP_MAP_SHMEM);
3494 * Increase reference count of swap entry by 1.
3495 * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required
3496 * but could not be atomically allocated. Returns 0, just as if it succeeded,
3497 * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which
3498 * might occur if a page table entry has got corrupted.
3500 int swap_duplicate(swp_entry_t entry)
3504 while (!err && __swap_duplicate(entry, 1) == -ENOMEM)
3505 err = add_swap_count_continuation(entry, GFP_ATOMIC);
3510 * @entry: swap entry for which we allocate swap cache.
3512 * Called when allocating swap cache for existing swap entry,
3513 * This can return error codes. Returns 0 at success.
3514 * -EEXIST means there is a swap cache.
3515 * Note: return code is different from swap_duplicate().
3517 int swapcache_prepare(swp_entry_t entry)
3519 return __swap_duplicate(entry, SWAP_HAS_CACHE);
3522 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
3524 return swap_type_to_swap_info(swp_type(entry));
3527 struct swap_info_struct *page_swap_info(struct page *page)
3529 swp_entry_t entry = { .val = page_private(page) };
3530 return swp_swap_info(entry);
3534 * out-of-line __page_file_ methods to avoid include hell.
3536 struct address_space *__page_file_mapping(struct page *page)
3538 return page_swap_info(page)->swap_file->f_mapping;
3540 EXPORT_SYMBOL_GPL(__page_file_mapping);
3542 pgoff_t __page_file_index(struct page *page)
3544 swp_entry_t swap = { .val = page_private(page) };
3545 return swp_offset(swap);
3547 EXPORT_SYMBOL_GPL(__page_file_index);
3550 * add_swap_count_continuation - called when a swap count is duplicated
3551 * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's
3552 * page of the original vmalloc'ed swap_map, to hold the continuation count
3553 * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called
3554 * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc.
3556 * These continuation pages are seldom referenced: the common paths all work
3557 * on the original swap_map, only referring to a continuation page when the
3558 * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX.
3560 * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding
3561 * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL)
3562 * can be called after dropping locks.
3564 int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask)
3566 struct swap_info_struct *si;
3567 struct swap_cluster_info *ci;
3570 struct page *list_page;
3572 unsigned char count;
3576 * When debugging, it's easier to use __GFP_ZERO here; but it's better
3577 * for latency not to zero a page while GFP_ATOMIC and holding locks.
3579 page = alloc_page(gfp_mask | __GFP_HIGHMEM);
3581 si = get_swap_device(entry);
3584 * An acceptable race has occurred since the failing
3585 * __swap_duplicate(): the swap device may be swapoff
3589 spin_lock(&si->lock);
3591 offset = swp_offset(entry);
3593 ci = lock_cluster(si, offset);
3595 count = swap_count(si->swap_map[offset]);
3597 if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) {
3599 * The higher the swap count, the more likely it is that tasks
3600 * will race to add swap count continuation: we need to avoid
3601 * over-provisioning.
3612 * We are fortunate that although vmalloc_to_page uses pte_offset_map,
3613 * no architecture is using highmem pages for kernel page tables: so it
3614 * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps.
3616 head = vmalloc_to_page(si->swap_map + offset);
3617 offset &= ~PAGE_MASK;
3619 spin_lock(&si->cont_lock);
3621 * Page allocation does not initialize the page's lru field,
3622 * but it does always reset its private field.
3624 if (!page_private(head)) {
3625 BUG_ON(count & COUNT_CONTINUED);
3626 INIT_LIST_HEAD(&head->lru);
3627 set_page_private(head, SWP_CONTINUED);
3628 si->flags |= SWP_CONTINUED;
3631 list_for_each_entry(list_page, &head->lru, lru) {
3635 * If the previous map said no continuation, but we've found
3636 * a continuation page, free our allocation and use this one.
3638 if (!(count & COUNT_CONTINUED))
3639 goto out_unlock_cont;
3641 map = kmap_atomic(list_page) + offset;
3646 * If this continuation count now has some space in it,
3647 * free our allocation and use this one.
3649 if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX)
3650 goto out_unlock_cont;
3653 list_add_tail(&page->lru, &head->lru);
3654 page = NULL; /* now it's attached, don't free it */
3656 spin_unlock(&si->cont_lock);
3659 spin_unlock(&si->lock);
3660 put_swap_device(si);
3668 * swap_count_continued - when the original swap_map count is incremented
3669 * from SWAP_MAP_MAX, check if there is already a continuation page to carry
3670 * into, carry if so, or else fail until a new continuation page is allocated;
3671 * when the original swap_map count is decremented from 0 with continuation,
3672 * borrow from the continuation and report whether it still holds more.
3673 * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster
3676 static bool swap_count_continued(struct swap_info_struct *si,
3677 pgoff_t offset, unsigned char count)
3684 head = vmalloc_to_page(si->swap_map + offset);
3685 if (page_private(head) != SWP_CONTINUED) {
3686 BUG_ON(count & COUNT_CONTINUED);
3687 return false; /* need to add count continuation */
3690 spin_lock(&si->cont_lock);
3691 offset &= ~PAGE_MASK;
3692 page = list_next_entry(head, lru);
3693 map = kmap_atomic(page) + offset;
3695 if (count == SWAP_MAP_MAX) /* initial increment from swap_map */
3696 goto init_map; /* jump over SWAP_CONT_MAX checks */
3698 if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */
3700 * Think of how you add 1 to 999
3702 while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) {
3704 page = list_next_entry(page, lru);
3705 BUG_ON(page == head);
3706 map = kmap_atomic(page) + offset;
3708 if (*map == SWAP_CONT_MAX) {
3710 page = list_next_entry(page, lru);
3712 ret = false; /* add count continuation */
3715 map = kmap_atomic(page) + offset;
3716 init_map: *map = 0; /* we didn't zero the page */
3720 while ((page = list_prev_entry(page, lru)) != head) {
3721 map = kmap_atomic(page) + offset;
3722 *map = COUNT_CONTINUED;
3725 ret = true; /* incremented */
3727 } else { /* decrementing */
3729 * Think of how you subtract 1 from 1000
3731 BUG_ON(count != COUNT_CONTINUED);
3732 while (*map == COUNT_CONTINUED) {
3734 page = list_next_entry(page, lru);
3735 BUG_ON(page == head);
3736 map = kmap_atomic(page) + offset;
3743 while ((page = list_prev_entry(page, lru)) != head) {
3744 map = kmap_atomic(page) + offset;
3745 *map = SWAP_CONT_MAX | count;
3746 count = COUNT_CONTINUED;
3749 ret = count == COUNT_CONTINUED;
3752 spin_unlock(&si->cont_lock);
3757 * free_swap_count_continuations - swapoff free all the continuation pages
3758 * appended to the swap_map, after swap_map is quiesced, before vfree'ing it.
3760 static void free_swap_count_continuations(struct swap_info_struct *si)
3764 for (offset = 0; offset < si->max; offset += PAGE_SIZE) {
3766 head = vmalloc_to_page(si->swap_map + offset);
3767 if (page_private(head)) {
3768 struct page *page, *next;
3770 list_for_each_entry_safe(page, next, &head->lru, lru) {
3771 list_del(&page->lru);
3778 #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP)
3779 void cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask)
3781 struct swap_info_struct *si, *next;
3782 int nid = page_to_nid(page);
3784 if (!(gfp_mask & __GFP_IO))
3787 if (!blk_cgroup_congested())
3791 * We've already scheduled a throttle, avoid taking the global swap
3794 if (current->throttle_queue)
3797 spin_lock(&swap_avail_lock);
3798 plist_for_each_entry_safe(si, next, &swap_avail_heads[nid],
3801 blkcg_schedule_throttle(bdev_get_queue(si->bdev), true);
3805 spin_unlock(&swap_avail_lock);
3809 static int __init swapfile_init(void)
3813 swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head),
3815 if (!swap_avail_heads) {
3816 pr_emerg("Not enough memory for swap heads, swap is disabled\n");
3821 plist_head_init(&swap_avail_heads[nid]);
3825 subsys_initcall(swapfile_init);